Log Attribute Replay
v24 cover Log Attribute Replay
Hi all,
This set is a subset of a larger series parent pointers. Delayed attributes allow attribute operations (set and remove) to be logged and committed in the same way that other delayed operations do. This allows more complex operations (like parent pointers) to be broken up into multiple smaller transactions. To do this, the existing attr operations must be modified to operate as a delayed operation. This means that they cannot roll, commit, or finish transactions. Instead, they return -EAGAIN to allow the calling function to handle the transaction. In this series, we focus on only the delayed attribute portion. We will introduce parent pointers in a later set.
The set as a whole is a bit much to digest at once, so I usually send out the smaller sub series to reduce reviewer burn out. But the entire extended series is visible through the included github links.
Updates since v23:
xfs: Return from xfs_attr_set_iter if there are no more rmtblks to process
Fixed xfs/125 fail
xfs: Capture buffers for delayed ops
NEW
xfs: Set up infrastructure for deferred attribute operations
Commit message updated with new name
Removed hasdelattr() from xfs_attri_validate
xfs: Implement attr logging and replay
Fixed xfs_sb_version_hasdelattr to use sb_features_log_incompat Removed redundant flag set in xfs_trans_attr_finish_update Renamed XFS_SB_FEAT_INCOMPAT_LOG_DELATTR to XFS_SB_FEAT_INCOMPAT_LOG_XATTRS Renamed xfs_sb_version_hasdelattr sb_version_haslogxattrs Removed unneeded xfs_qm_dqattach_locked in xfs_trans_attr_finish_update Changed xfs_hasdelattr to sb_version_haslogxattrs in xfs_attr_create_intent Simplified args pointer in xfs_attri_item_recover
Removed uneeded buffer rejoin logic
RFC xfs: Skip flip flags for delayed attrs
Fixed xfs/125 fail
xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
added delayed sample to avoid potential race when enabling feature moved warn_once to after error check
xfs: Remove unused xfs_attr_*_args
Removed local variable leaf_bp
xfs: Add log attribute error tag
Renamed error tag to ERRTAG_LARP
xfs: Add larp debug option
NEW
xfs: Merge xfs_delattr_context into xfs_attr_item
Rebase adjustments
xfs: Add helper function xfs_attr_leaf_addname
Rebase adjustments
This series can be viewed on github here:
https://github.com/allisonhenderson/xfs_work/tree/delayed_attrs_v24
As well as the extended delayed attribute and parent pointer series:
https://github.com/allisonhenderson/xfs_work/tree/delayed_attrs_v24_extended
And the test cases:
https://github.com/allisonhenderson/xfs_work/tree/pptr_xfstestsv4
In order to run the test cases, you will need have the corresponding xfsprogs changes as well. Which can be found here:
https://github.com/allisonhenderson/xfs_work/tree/delayed_attrs_xfsprogs_v24
https://github.com/allisonhenderson/xfs_work/tree/delayed_attrs_xfsprogs_v24_extended
To run the xfs attributes tests run:
check -g attr
To run as delayed attributes run:
echo 1 > /sys/fs/xfs/debug/larp;
check -g attr
To run parent pointer tests:
check -g parent
I've also made the corresponding updates to the user space side as well, and ported anything they need to seat correctly.
Questions, comment and feedback appreciated!
Allison
Allison Collins (1):
xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
Allison Henderson (10):
xfs: Return from xfs_attr_set_iter if there are no more rmtblks to
process
xfs: Capture buffers for delayed ops
xfs: Set up infrastructure for log atrribute replay
xfs: Implement attr logging and replay
RFC xfs: Skip flip flags for delayed attrs
xfs: Remove unused xfs_attr_*_args
xfs: Add log attribute error tag
xfs: Add larp debug option
xfs: Merge xfs_delattr_context into xfs_attr_item
xfs: Add helper function xfs_attr_leaf_addname
fs/xfs/Makefile | 1 + fs/xfs/libxfs/xfs_attr.c | 454 +++++++++--------- fs/xfs/libxfs/xfs_attr.h | 57 ++- fs/xfs/libxfs/xfs_attr_leaf.c | 3 +- fs/xfs/libxfs/xfs_attr_remote.c | 37 +- fs/xfs/libxfs/xfs_attr_remote.h | 6 +- fs/xfs/libxfs/xfs_defer.c | 8 +- fs/xfs/libxfs/xfs_defer.h | 7 +- fs/xfs/libxfs/xfs_errortag.h | 4 +- fs/xfs/libxfs/xfs_format.h | 10 +- fs/xfs/libxfs/xfs_log_format.h | 44 +- fs/xfs/libxfs/xfs_log_recover.h | 2 + fs/xfs/scrub/common.c | 2 + fs/xfs/xfs_attr_item.c | 814 ++++++++++++++++++++++++++++++++ fs/xfs/xfs_attr_item.h | 52 ++ fs/xfs/xfs_attr_list.c | 1 + fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_error.c | 3 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_globals.c | 1 + fs/xfs/xfs_ioctl32.c | 2 + fs/xfs/xfs_iops.c | 2 + fs/xfs/xfs_log.c | 45 ++ fs/xfs/xfs_log.h | 1 + fs/xfs/xfs_log_recover.c | 9 + fs/xfs/xfs_ondisk.h | 2 + fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- fs/xfs/xfs_sysctl.h | 1 + fs/xfs/xfs_sysfs.c | 24 + fs/xfs/xfs_trace.h | 1 + 33 files changed, 1333 insertions(+), 270 deletions(-) create mode 100644 fs/xfs/xfs_attr_item.c create mode 100644 fs/xfs/xfs_attr_item.h
Signature · 1 line
2.25.1
PATCH v24 1/11 xfs: Return from xfs_attr_set_iter if there are no more rmtblks to process
During an attr rename operation, blocks are saved for later removal as rmtblkno2. The rmtblkno is used in the case of needing to alloc more blocks if not enough were available. However, in the case that no further blocks need to be added or removed, we can return as soon as xfs_attr_node_addname completes, rather than rolling the transaction with an -EAGAIN return. This extra loop does not hurt anything right now, but it will be a problem later when we get into log items because we end up with an empty log transaction. So, add a simple check to cut out the unneeded iteration.
--- fs/xfs/libxfs/xfs_attr.c | 8 ++++++++ 1 file changed, 8 insertions(+)
fs/xfs/libxfs/xfs_attr.c +8 -0
@@ -412,6 +412,14 @@ xfs_attr_set_iter(if (error)return error; +/* +* If addname was successful, and we dont need to alloc +* or remove anymore blks, we're done. +*/ +if (!args->rmtblkno && + !(args->op_flags &XFS_DA_OP_RENAME )) +return 0 ; + dac->dela_state =XFS_DAS_FOUND_NBLK ; }trace_xfs_attr_set_iter_return (dac->dela_state, args->dp);
Signature · 1 line
2.25.1
PATCH v24 2/11 xfs: Capture buffers for delayed ops
In reply to Allison HendersonThis patch enables delayed operations to capture held buffers with in the xfs_defer_capture. Buffers are then rejoined to the new transaction in xlog_finish_defer_ops
--- fs/xfs/libxfs/xfs_defer.c | 7 ++++++- fs/xfs/libxfs/xfs_defer.h | 4 +++- fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_log_recover.c | 7 +++++++ fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- 9 files changed, 22 insertions(+), 6 deletions(-)
fs/xfs/libxfs/xfs_defer.c +6 -1
@@ -639,6 +639,7 @@ xfs_defer_ops_capture( dfc =kmem_zalloc (sizeof (*dfc),KM_NOFS );INIT_LIST_HEAD (&dfc->dfc_list);INIT_LIST_HEAD (&dfc->dfc_dfops); +INIT_LIST_HEAD (&dfc->dfc_buffers);xfs_defer_create_intents (tp);
@@ -690,7 +691,8 @@ intxfs_defer_ops_capture_and_commit ( struct xfs_trans *tp, struct xfs_inode *capture_ip, - struct list_head *capture_list) + struct list_head *capture_list, + struct xfs_buf *bp) { struct xfs_mount *mp = tp->t_mountp;struct xfs_defer_capture *dfc;
@@ -703,6 +705,9 @@ xfs_defer_ops_capture_and_commit(if (!dfc)return xfs_trans_commit (tp); +if (bp && bp->b_transp == tp) +list_add_tail (&bp->b_delay, &dfc->dfc_buffers); +/* Commit the transaction and add the capture structure to the list. */ error =xfs_trans_commit (tp);if (error) {
fs/xfs/libxfs/xfs_defer.h +3 -1
@@ -74,6 +74,7 @@ struct xfs_defer_capture {/* Deferred ops state saved from the transaction. */ struct list_head dfc_dfops; +struct list_head dfc_buffers;unsigned int dfc_tpflags;/* Block reservations for the data and rt devices. */
@@ -95,7 +96,8 @@ struct xfs_defer_capture {* This doesn't normally happen except log recovery. */ int xfs_defer_ops_capture_and_commit (struct xfs_trans *tp, -struct xfs_inode *capture_ip,struct list_head *capture_list); +struct xfs_inode *capture_ip,struct list_head *capture_list, +struct xfs_buf *bp);void xfs_defer_ops_continue (struct xfs_defer_capture *d,struct xfs_trans *tp,struct xfs_inode **captured_ipp);void xfs_defer_ops_release (struct xfs_mount *mp,struct xfs_defer_capture *d);
fs/xfs/xfs_bmap_item.c +1 -1
@@ -532,7 +532,7 @@ xfs_bui_item_recover(* Commit transaction, which frees the transaction and saves the inode * for later replay activities. */ - error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list); + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list,NULL );if (error) goto err_unlock;
fs/xfs/xfs_buf.c +1 -0
@@ -233,6 +233,7 @@ _xfs_buf_alloc(init_completion (&bp->b_iowait);INIT_LIST_HEAD (&bp->b_lru);INIT_LIST_HEAD (&bp->b_list); +INIT_LIST_HEAD (&bp->b_delay);INIT_LIST_HEAD (&bp->b_li_list);sema_init (&bp->b_sema,0 );/* held, no waiters */ spin_lock_init (&bp->b_lock);
fs/xfs/xfs_buf.h +1 -0
@@ -151,6 +151,7 @@ struct xfs_buf {int b_io_error;/* internal IO error state */ wait_queue_head_t b_waiters;/* unpin waiters */ struct list_head b_list; +struct list_head b_delay;/* delayed operations list */ struct xfs_perag *b_pag;/* contains rbtree root */ struct xfs_mount *b_mount;struct xfs_buftarg *b_target;/* buffer target (device) */
fs/xfs/xfs_extfree_item.c +1 -1
@@ -637,7 +637,7 @@ xfs_efi_item_recover( } -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_trans_cancel (tp);
fs/xfs/xfs_log_recover.c +7 -0
@@ -2465,6 +2465,7 @@ xlog_finish_defer_ops( struct list_head *capture_list) {struct xfs_defer_capture *dfc, *next; +struct xfs_buf *bp, *bnext;struct xfs_trans *tp;struct xfs_inode *ip;int error =0 ;
@@ -2489,6 +2490,12 @@ xlog_finish_defer_ops(return error; } +list_for_each_entry_safe (bp, bnext, &dfc->dfc_buffers, b_delay) { +xfs_trans_bjoin (tp, bp); +xfs_trans_bhold (tp, bp); +list_del_init (&bp->b_delay); + } +/* * Transfer to this new transaction all the dfops we captured * from recovering a single intent item.
fs/xfs/xfs_refcount_item.c +1 -1
@@ -557,7 +557,7 @@ xfs_cui_item_recover( }xfs_refcount_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_refcount_finish_one_cleanup (tp, rcur, error);
fs/xfs/xfs_rmap_item.c +1 -1
@@ -587,7 +587,7 @@ xfs_rui_item_recover( }xfs_rmap_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_rmap_finish_one_cleanup (tp, rcur, error);
Signature · 1 line
2.25.1
Re: [PATCH v24 02/11] xfs: Capture buffers for delayed ops
Quoted 4 lines · Allison Henderson
This patch enables delayed operations to capture held buffers with in the
xfs_defer_capture. Buffers are then rejoined to the new transaction inxlog_finish_defer_ops
Looks good to me.
Quoted 168 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 7 ++++++- fs/xfs/libxfs/xfs_defer.h | 4 +++- fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_log_recover.c | 7 +++++++ fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- 9 files changed, 22 insertions(+), 6 deletions(-)fs/xfs/libxfs/xfs_defer.c +6
-1@@ -639,6 +639,7 @@ xfs_defer_ops_capture( dfc =kmem_zalloc (sizeof (*dfc),KM_NOFS );INIT_LIST_HEAD (&dfc->dfc_list);INIT_LIST_HEAD (&dfc->dfc_dfops); +INIT_LIST_HEAD (&dfc->dfc_buffers);xfs_defer_create_intents (tp);@@ -690,7 +691,8 @@ intxfs_defer_ops_capture_and_commit ( struct xfs_trans *tp, struct xfs_inode *capture_ip, - struct list_head *capture_list) + struct list_head *capture_list, + struct xfs_buf *bp) { struct xfs_mount *mp = tp->t_mountp;struct xfs_defer_capture *dfc;@@ -703,6 +705,9 @@ xfs_defer_ops_capture_and_commit(if (!dfc)return xfs_trans_commit (tp); +if (bp && bp->b_transp == tp) +list_add_tail (&bp->b_delay, &dfc->dfc_buffers); +/* Commit the transaction and add the capture structure to the list. */ error =xfs_trans_commit (tp);if (error) {fs/xfs/libxfs/xfs_defer.h +3
-1@@ -74,6 +74,7 @@ struct xfs_defer_capture {/* Deferred ops state saved from the transaction. */ struct list_head dfc_dfops; +struct list_head dfc_buffers;unsigned int dfc_tpflags;/* Block reservations for the data and rt devices. */ @@ -95,7 +96,8 @@ struct xfs_defer_capture {* This doesn't normally happen except log recovery. */ int xfs_defer_ops_capture_and_commit (struct xfs_trans *tp, -struct xfs_inode *capture_ip,struct list_head *capture_list); +struct xfs_inode *capture_ip,struct list_head *capture_list, +struct xfs_buf *bp);void xfs_defer_ops_continue (struct xfs_defer_capture *d,struct xfs_trans *tp,struct xfs_inode **captured_ipp);void xfs_defer_ops_release (struct xfs_mount *mp,struct xfs_defer_capture *d);fs/xfs/xfs_bmap_item.c +1
-1@@ -532,7 +532,7 @@ xfs_bui_item_recover(* Commit transaction, which frees the transaction and saves the inode * for later replay activities. */ - error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list); + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list,NULL );if (error) goto err_unlock;fs/xfs/xfs_buf.c +1
-0@@ -233,6 +233,7 @@ _xfs_buf_alloc(init_completion (&bp->b_iowait);INIT_LIST_HEAD (&bp->b_lru);INIT_LIST_HEAD (&bp->b_list); +INIT_LIST_HEAD (&bp->b_delay);INIT_LIST_HEAD (&bp->b_li_list);sema_init (&bp->b_sema,0 );/* held, no waiters */ spin_lock_init (&bp->b_lock);fs/xfs/xfs_buf.h +1
-0@@ -151,6 +151,7 @@ struct xfs_buf {int b_io_error;/* internal IO error state */ wait_queue_head_t b_waiters;/* unpin waiters */ struct list_head b_list; +struct list_head b_delay;/* delayed operations list */ struct xfs_perag *b_pag;/* contains rbtree root */ struct xfs_mount *b_mount;struct xfs_buftarg *b_target;/* buffer target (device) */ fs/xfs/xfs_extfree_item.c +1
-1@@ -637,7 +637,7 @@ xfs_efi_item_recover( } -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_trans_cancel (tp);fs/xfs/xfs_log_recover.c +7
-0@@ -2465,6 +2465,7 @@ xlog_finish_defer_ops( struct list_head *capture_list) {struct xfs_defer_capture *dfc, *next; +struct xfs_buf *bp, *bnext;struct xfs_trans *tp;struct xfs_inode *ip;int error =0 ;@@ -2489,6 +2490,12 @@ xlog_finish_defer_ops(return error; } +list_for_each_entry_safe (bp, bnext, &dfc->dfc_buffers, b_delay) { +xfs_trans_bjoin (tp, bp); +xfs_trans_bhold (tp, bp); +list_del_init (&bp->b_delay); + } +/* * Transfer to this new transaction all the dfops we captured * from recovering a single intent item. fs/xfs/xfs_refcount_item.c +1
-1@@ -557,7 +557,7 @@ xfs_cui_item_recover( }xfs_refcount_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_refcount_finish_one_cleanup (tp, rcur, error);fs/xfs/xfs_rmap_item.c +1
-1@@ -587,7 +587,7 @@ xfs_rui_item_recover( }xfs_rmap_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_rmap_finish_one_cleanup (tp, rcur, error);
Signature · 1 line
chandan
Re: [PATCH v24 02/11] xfs: Capture buffers for delayed ops
Quoted 9 lines · Chandan Babu R
Quoted 4 lines · Allison Henderson
This patch enables delayed operations to capture held buffers with in the
xfs_defer_capture. Buffers are then rejoined to the new transaction inxlog_finish_defer_opsLooks good to me.
Reviewed-by Chandan Babu R <[email protected]>
Thank you!
Allison
Quoted 171 lines
Quoted 169 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 7 ++++++- fs/xfs/libxfs/xfs_defer.h | 4 +++- fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_log_recover.c | 7 +++++++ fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- 9 files changed, 22 insertions(+), 6 deletions(-)fs/xfs/libxfs/xfs_defer.c +6
-1@@ -639,6 +639,7 @@ xfs_defer_ops_capture( dfc =kmem_zalloc (sizeof (*dfc),KM_NOFS );INIT_LIST_HEAD (&dfc->dfc_list);INIT_LIST_HEAD (&dfc->dfc_dfops); +INIT_LIST_HEAD (&dfc->dfc_buffers);xfs_defer_create_intents (tp);@@ -690,7 +691,8 @@ intxfs_defer_ops_capture_and_commit ( struct xfs_trans *tp, struct xfs_inode *capture_ip, - struct list_head *capture_list) + struct list_head *capture_list, + struct xfs_buf *bp) { struct xfs_mount *mp = tp->t_mountp;struct xfs_defer_capture *dfc;@@ -703,6 +705,9 @@ xfs_defer_ops_capture_and_commit(if (!dfc)return xfs_trans_commit (tp); +if (bp && bp->b_transp == tp) +list_add_tail (&bp->b_delay, &dfc->dfc_buffers); +/* Commit the transaction and add the capture structure to the list. */ error =xfs_trans_commit (tp);if (error) {fs/xfs/libxfs/xfs_defer.h +3
-1@@ -74,6 +74,7 @@ struct xfs_defer_capture {/* Deferred ops state saved from the transaction. */ struct list_head dfc_dfops; +struct list_head dfc_buffers;unsigned int dfc_tpflags;/* Block reservations for the data and rt devices. */ @@ -95,7 +96,8 @@ struct xfs_defer_capture {* This doesn't normally happen except log recovery. */ int xfs_defer_ops_capture_and_commit (struct xfs_trans *tp, -struct xfs_inode *capture_ip,struct list_head *capture_list); +struct xfs_inode *capture_ip,struct list_head *capture_list, +struct xfs_buf *bp);void xfs_defer_ops_continue (struct xfs_defer_capture *d,struct xfs_trans *tp,struct xfs_inode **captured_ipp);void xfs_defer_ops_release (struct xfs_mount *mp,struct xfs_defer_capture *d);fs/xfs/xfs_bmap_item.c +1
-1@@ -532,7 +532,7 @@ xfs_bui_item_recover(* Commit transaction, which frees the transaction and saves the inode * for later replay activities. */ - error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list); + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list,NULL );if (error) goto err_unlock;fs/xfs/xfs_buf.c +1
-0@@ -233,6 +233,7 @@ _xfs_buf_alloc(init_completion (&bp->b_iowait);INIT_LIST_HEAD (&bp->b_lru);INIT_LIST_HEAD (&bp->b_list); +INIT_LIST_HEAD (&bp->b_delay);INIT_LIST_HEAD (&bp->b_li_list);sema_init (&bp->b_sema,0 );/* held, no waiters */ spin_lock_init (&bp->b_lock);fs/xfs/xfs_buf.h +1
-0@@ -151,6 +151,7 @@ struct xfs_buf {int b_io_error;/* internal IO error state */ wait_queue_head_t b_waiters;/* unpin waiters */ struct list_head b_list; +struct list_head b_delay;/* delayed operations list */ struct xfs_perag *b_pag;/* contains rbtree root */ struct xfs_mount *b_mount;struct xfs_buftarg *b_target;/* buffer target (device) */ fs/xfs/xfs_extfree_item.c +1
-1@@ -637,7 +637,7 @@ xfs_efi_item_recover( } -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_trans_cancel (tp);fs/xfs/xfs_log_recover.c +7
-0@@ -2465,6 +2465,7 @@ xlog_finish_defer_ops( struct list_head *capture_list) {struct xfs_defer_capture *dfc, *next; +struct xfs_buf *bp, *bnext;struct xfs_trans *tp;struct xfs_inode *ip;int error =0 ;@@ -2489,6 +2490,12 @@ xlog_finish_defer_ops(return error; } +list_for_each_entry_safe (bp, bnext, &dfc->dfc_buffers, b_delay) { +xfs_trans_bjoin (tp, bp); +xfs_trans_bhold (tp, bp); +list_del_init (&bp->b_delay); + } +/* * Transfer to this new transaction all the dfops we captured * from recovering a single intent item. fs/xfs/xfs_refcount_item.c +1
-1@@ -557,7 +557,7 @@ xfs_cui_item_recover( }xfs_refcount_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_refcount_finish_one_cleanup (tp, rcur, error);fs/xfs/xfs_rmap_item.c +1
-1@@ -587,7 +587,7 @@ xfs_rui_item_recover( }xfs_rmap_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_rmap_finish_one_cleanup (tp, rcur, error);
Re: [PATCH v24 02/11] xfs: Capture buffers for delayed ops
In reply to Allison HendersonQuoted 30 lines · Allison Henderson
This patch enables delayed operations to capture held buffers with in the
xfs_defer_capture. Buffers are then rejoined to the new transaction inxlog_finish_defer_opsSigned-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 7 ++++++- fs/xfs/libxfs/xfs_defer.h | 4 +++- fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_log_recover.c | 7 +++++++ fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- 9 files changed, 22 insertions(+), 6 deletions(-)fs/xfs/libxfs/xfs_defer.c +1
-0@@ -639,6 +639,7 @@ xfs_defer_ops_capture( dfc =kmem_zalloc (sizeof (*dfc),KM_NOFS );INIT_LIST_HEAD (&dfc->dfc_list);INIT_LIST_HEAD (&dfc->dfc_dfops); +INIT_LIST_HEAD (&dfc->dfc_buffers);xfs_defer_create_intents (tp);
fs/xfs/libxfs/xfs_defer.c int · line 695
Quoted 6 lines
fs/xfs/libxfs/xfs_defer.c +2
-1@@ -690,7 +691,8 @@ intxfs_defer_ops_capture_and_commit ( struct xfs_trans *tp, struct xfs_inode *capture_ip, - struct list_head *capture_list) + struct list_head *capture_list, + struct xfs_buf *bp)
I wonder if xfs_defer_ops_capture should learn to pick up the inodes and buffers to hold automatically from the transaction that's being committed? Seeing as xfs_defer_trans_roll already knows how to do that across transaction rolls, and that's more or less the same thing we're doing here, but in a much more roundabout way.
Quoted 65 lines
fs/xfs/libxfs/xfs_defer.c +3
-0{struct xfs_mount *mp = tp->t_mountp;struct xfs_defer_capture *dfc;@@ -703,6 +705,9 @@ xfs_defer_ops_capture_and_commit(if (!dfc)return xfs_trans_commit (tp); +if (bp && bp->b_transp == tp) +list_add_tail (&bp->b_delay, &dfc->dfc_buffers); +/* Commit the transaction and add the capture structure to the list. */ error =xfs_trans_commit (tp);if (error) {fs/xfs/libxfs/xfs_defer.h +3
-1@@ -74,6 +74,7 @@ struct xfs_defer_capture {/* Deferred ops state saved from the transaction. */ struct list_head dfc_dfops; +struct list_head dfc_buffers;unsigned int dfc_tpflags;/* Block reservations for the data and rt devices. */ @@ -95,7 +96,8 @@ struct xfs_defer_capture {* This doesn't normally happen except log recovery. */ int xfs_defer_ops_capture_and_commit (struct xfs_trans *tp, -struct xfs_inode *capture_ip,struct list_head *capture_list); +struct xfs_inode *capture_ip,struct list_head *capture_list, +struct xfs_buf *bp);void xfs_defer_ops_continue (struct xfs_defer_capture *d,struct xfs_trans *tp,struct xfs_inode **captured_ipp);void xfs_defer_ops_release (struct xfs_mount *mp,struct xfs_defer_capture *d);fs/xfs/xfs_bmap_item.c +1
-1@@ -532,7 +532,7 @@ xfs_bui_item_recover(* Commit transaction, which frees the transaction and saves the inode * for later replay activities. */ - error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list); + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list,NULL );if (error) goto err_unlock;fs/xfs/xfs_buf.c +1
-0@@ -233,6 +233,7 @@ _xfs_buf_alloc(init_completion (&bp->b_iowait);INIT_LIST_HEAD (&bp->b_lru);INIT_LIST_HEAD (&bp->b_list); +INIT_LIST_HEAD (&bp->b_delay);INIT_LIST_HEAD (&bp->b_li_list);sema_init (&bp->b_sema,0 );/* held, no waiters */ spin_lock_init (&bp->b_lock);fs/xfs/xfs_buf.h +0
-0
fs/xfs/xfs_buf.h struct xfs_buf { · line 157
Quoted 7 lines
fs/xfs/xfs_buf.h +1
-0@@ -151,6 +151,7 @@ struct xfs_buf {int b_io_error;/* internal IO error state */ wait_queue_head_t b_waiters;/* unpin waiters */ struct list_head b_list; +struct list_head b_delay;/* delayed operations list */ struct xfs_perag *b_pag;/* contains rbtree root */ struct xfs_mount *b_mount;struct xfs_buftarg *b_target;/* buffer target (device) */
The bare list-conveyance machinery looks fine to me, but adding 16 bytes to struct xfs_buf for something that only happens during log recovery is rather expensive. Can you reuse b_list for this purpose? I think the only user of b_list are the buffer delwri functions, which shouldn't be active here since the xattr recovery mechanism (a) holds the buffer lock and (b) doesn't itself use delwri buffer lists for xattr leaf blocks.
(The AIL uses delwri lists, but it won't touch a locked buffer.)
Quoted 26 lines
fs/xfs/xfs_extfree_item.c +1
-1@@ -637,7 +637,7 @@ xfs_efi_item_recover( } -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_trans_cancel (tp);fs/xfs/xfs_log_recover.c +1
-0@@ -2465,6 +2465,7 @@ xlog_finish_defer_ops( struct list_head *capture_list) {struct xfs_defer_capture *dfc, *next; +struct xfs_buf *bp, *bnext;struct xfs_trans *tp;struct xfs_inode *ip;int error =0 ;
fs/xfs/xfs_log_recover.c xlog_finish_defer_ops( · line 2497
Quoted 8 lines
fs/xfs/xfs_log_recover.c +5
-0@@ -2489,6 +2490,12 @@ xlog_finish_defer_ops(return error; } +list_for_each_entry_safe (bp, bnext, &dfc->dfc_buffers, b_delay) { +xfs_trans_bjoin (tp, bp); +xfs_trans_bhold (tp, bp); +list_del_init (&bp->b_delay); + }
Why isn't this in xfs_defer_ops_continue, like the code that extracts the inodes from the capture struct and hands them back to the caller?
Quoted 33 lines
fs/xfs/xfs_log_recover.c +1
-0+/* * Transfer to this new transaction all the dfops we captured * from recovering a single intent item. fs/xfs/xfs_refcount_item.c +1
-1@@ -557,7 +557,7 @@ xfs_cui_item_recover( }xfs_refcount_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_refcount_finish_one_cleanup (tp, rcur, error);fs/xfs/xfs_rmap_item.c +1
-1@@ -587,7 +587,7 @@ xfs_rui_item_recover( }xfs_rmap_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_rmap_finish_one_cleanup (tp, rcur, error);Signature · 1 line
2.25.1
Re: [PATCH v24 02/11] xfs: Capture buffers for delayed ops
Quoted 30 lines · Darrick J. Wong
Quoted 29 lines · Allison Henderson
This patch enables delayed operations to capture held buffers with in the
xfs_defer_capture. Buffers are then rejoined to the new transaction inxlog_finish_defer_opsSigned-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 7 ++++++- fs/xfs/libxfs/xfs_defer.h | 4 +++- fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_log_recover.c | 7 +++++++ fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- 9 files changed, 22 insertions(+), 6 deletions(-)fs/xfs/libxfs/xfs_defer.c +1
-0@@ -639,6 +639,7 @@ xfs_defer_ops_capture( dfc =kmem_zalloc (sizeof (*dfc),KM_NOFS );INIT_LIST_HEAD (&dfc->dfc_list);INIT_LIST_HEAD (&dfc->dfc_dfops); +INIT_LIST_HEAD (&dfc->dfc_buffers);xfs_defer_create_intents (tp);
Quoted 13 lines
fs/xfs/libxfs/xfs_defer.c int · line 695
Quoted 7 lines
fs/xfs/libxfs/xfs_defer.c +2
-1@@ -690,7 +691,8 @@ intxfs_defer_ops_capture_and_commit ( struct xfs_trans *tp, struct xfs_inode *capture_ip, - struct list_head *capture_list) + struct list_head *capture_list, + struct xfs_buf *bp)I wonder if
xfs_defer_ops_captureshould learn to pick up the inodes and buffers to hold automatically from the transaction that's being committed? Seeing asxfs_defer_trans_rollalready knows how to do that across transaction rolls, and that's more or less the same thing we're doing here, but in a much more roundabout way.
I see, I suppose it could? But it wouldnt be used in this case though, at least not yet. I sort of got the impression that people like to see things added as they are needed, and then unused code culled where it can be. I would think that if the need does arise though, b_delay would be easy to expand into list of xfs_delay_items or something similar to what xfs_defer_trans_roll has.
Quoted 81 lines
Quoted 66 lines
fs/xfs/libxfs/xfs_defer.c +3
-0{struct xfs_mount *mp = tp->t_mountp;struct xfs_defer_capture *dfc;@@ -703,6 +705,9 @@ xfs_defer_ops_capture_and_commit(if (!dfc)return xfs_trans_commit (tp); +if (bp && bp->b_transp == tp) +list_add_tail (&bp->b_delay, &dfc->dfc_buffers); +/* Commit the transaction and add the capture structure to the list. */ error =xfs_trans_commit (tp);if (error) {fs/xfs/libxfs/xfs_defer.h +3
-1@@ -74,6 +74,7 @@ struct xfs_defer_capture {/* Deferred ops state saved from the transaction. */ struct list_head dfc_dfops; +struct list_head dfc_buffers;unsigned int dfc_tpflags;/* Block reservations for the data and rt devices. */ @@ -95,7 +96,8 @@ struct xfs_defer_capture {* This doesn't normally happen except log recovery. */ int xfs_defer_ops_capture_and_commit (struct xfs_trans *tp, -struct xfs_inode *capture_ip,struct list_head *capture_list); +struct xfs_inode *capture_ip,struct list_head *capture_list, +struct xfs_buf *bp);void xfs_defer_ops_continue (struct xfs_defer_capture *d,struct xfs_trans *tp,struct xfs_inode **captured_ipp);void xfs_defer_ops_release (struct xfs_mount *mp,struct xfs_defer_capture *d);fs/xfs/xfs_bmap_item.c +1
-1@@ -532,7 +532,7 @@ xfs_bui_item_recover(* Commit transaction, which frees the transaction and saves the inode * for later replay activities. */ - error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list); + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list,NULL );if (error) goto err_unlock;fs/xfs/xfs_buf.c +1
-0@@ -233,6 +233,7 @@ _xfs_buf_alloc(init_completion (&bp->b_iowait);INIT_LIST_HEAD (&bp->b_lru);INIT_LIST_HEAD (&bp->b_list); +INIT_LIST_HEAD (&bp->b_delay);INIT_LIST_HEAD (&bp->b_li_list);sema_init (&bp->b_sema,0 );/* held, no waiters */ spin_lock_init (&bp->b_lock);fs/xfs/xfs_buf.h +0
-0fs/xfs/xfs_buf.h struct xfs_buf { · line 157
Quoted 7 lines
fs/xfs/xfs_buf.h +1
-0@@ -151,6 +151,7 @@ struct xfs_buf {int b_io_error;/* internal IO error state */ wait_queue_head_t b_waiters;/* unpin waiters */ struct list_head b_list; +struct list_head b_delay;/* delayed operations list */ struct xfs_perag *b_pag;/* contains rbtree root */ struct xfs_mount *b_mount;struct xfs_buftarg *b_target;/* buffer target (device) */ The bare list-conveyance machinery looks fine to me, but adding 16 bytes to struct
xfs_buffor something that only happens during log recovery is rather expensive. Can you reuseb_listfor this purpose? I think the only user ofb_listare the buffer delwri functions, which shouldn't be active here since the xattr recovery mechanism (a) holds the buffer lock and (b) doesn't itself use delwri buffer lists for xattr leaf blocks.
Quoted 2 lines
(The AIL uses delwri lists, but it won't touch a locked buffer.)
Sure, it sounds like it would work, will try it out.
Quoted 29 lines
Quoted 29 lines
fs/xfs/xfs_extfree_item.c +1
-1@@ -637,7 +637,7 @@ xfs_efi_item_recover( } -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_trans_cancel (tp);fs/xfs/xfs_log_recover.c +1
-0@@ -2465,6 +2465,7 @@ xlog_finish_defer_ops( struct list_head *capture_list) {struct xfs_defer_capture *dfc, *next; +struct xfs_buf *bp, *bnext;struct xfs_trans *tp;struct xfs_inode *ip;int error =0 ;@@ -2489,6 +2490,12 @@ xlog_finish_defer_ops(return error; }
Quoted 8 lines
fs/xfs/xfs_log_recover.c xlog_finish_defer_ops( · line 2497
Quoted 5 lines
fs/xfs/xfs_log_recover.c +5
-0+list_for_each_entry_safe (bp, bnext, &dfc->dfc_buffers, b_delay) { +xfs_trans_bjoin (tp, bp); +xfs_trans_bhold (tp, bp); +list_del_init (&bp->b_delay); + }Why isn't this in
xfs_defer_ops_continue, like the code that extracts the inodes from the capture struct and hands them back to the caller?
Its just what was discussed in the last review is all. That does look like a better place for it though. Will move there.
Allison
Quoted 34 lines
Quoted 34 lines
fs/xfs/xfs_log_recover.c +1
-0+/* * Transfer to this new transaction all the dfops we captured * from recovering a single intent item. fs/xfs/xfs_refcount_item.c +1
-1@@ -557,7 +557,7 @@ xfs_cui_item_recover( }xfs_refcount_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_refcount_finish_one_cleanup (tp, rcur, error);fs/xfs/xfs_rmap_item.c +1
-1@@ -587,7 +587,7 @@ xfs_rui_item_recover( }xfs_rmap_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_rmap_finish_one_cleanup (tp, rcur, error);Signature · 1 line
2.25.1
Re: [PATCH v24 02/11] xfs: Capture buffers for delayed ops
Quoted 44 lines · Allison Henderson
Quoted 33 lines · Darrick J. Wong
Quoted 32 lines · Allison Henderson
This patch enables delayed operations to capture held buffers with in the
xfs_defer_capture. Buffers are then rejoined to the new transaction inxlog_finish_defer_opsSigned-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 7 ++++++- fs/xfs/libxfs/xfs_defer.h | 4 +++- fs/xfs/xfs_bmap_item.c | 2 +- fs/xfs/xfs_buf.c | 1 + fs/xfs/xfs_buf.h | 1 + fs/xfs/xfs_extfree_item.c | 2 +- fs/xfs/xfs_log_recover.c | 7 +++++++ fs/xfs/xfs_refcount_item.c | 2 +- fs/xfs/xfs_rmap_item.c | 2 +- 9 files changed, 22 insertions(+), 6 deletions(-)fs/xfs/libxfs/xfs_defer.c +1
-1@@ -639,6 +639,7 @@ xfs_defer_ops_capture( dfc =kmem_zalloc (sizeof (*dfc),KM_NOFS );INIT_LIST_HEAD (&dfc->dfc_list);INIT_LIST_HEAD (&dfc->dfc_dfops); +INIT_LIST_HEAD (&dfc->dfc_buffers);xfs_defer_create_intents (tp);@@ -690,7 +691,8 @@ intxfs_defer_ops_capture_and_commit ( struct xfs_trans *tp, struct xfs_inode *capture_ip, - struct list_head *capture_list)Quoted 8 lines
fs/xfs/libxfs/xfs_defer.c int · line 695
Quoted 2 lines
fs/xfs/libxfs/xfs_defer.c +2
-0+struct list_head *capture_list, + struct xfs_buf *bp)I wonder if
xfs_defer_ops_captureshould learn to pick up the inodes and buffers to hold automatically from the transaction that's being committed? Seeing asxfs_defer_trans_rollalready knows how to do that across transaction rolls, and that's more or less the same thing we're doing here, but in a much more roundabout way.
Quoted 6 lines
I see, I suppose it could? But it wouldnt be used in this case though, at least not yet. I sort of got the impression that people like to see things added as they are needed, and then unused code culled where it can be. I would think that if the need does arise though,
b_delaywould be easy to expand into list ofxfs_delay_itemsor something similar to whatxfs_defer_trans_rollhas.
On further thought, I decided that log recovery of intent items really ought to be able to hold the same number and type of resources across a transaction "roll" that we can do at runtime, since there really ought to be no difference.
I wrote a quick patch series to hoist the runtime code path to store the held buffers and inodes during xfs_defer_trans_roll in an explicitly named structure with separate save and restore helpers, then refactored log recovery to use it.
This cleans up the API a bit so we that the defer capture system picks up the held resources from the log item state just like a regular defer roll. It occurs to me that another way to handle intent item recovery would be to create workqueue items for each recovered intent and then kick the transactions off in background threads so that the end of all the _recover functions would be xfs_trans_commit. But that's a lot more surgery on the recovery code and I don't want to go there right now.
Anyway, I'll send the RFC series shortly and cc you on it, in case you want to incorporate it at the head of your series and nuke patch #2.
Note: the end of the xattri recovery function then becomes:
error = xfs_defer_ops_capture_and_commit(tp, capture_list); if (error) /* pretend I handled the error */ if (attr->xattri_dac.leaf_bp) xfs_buf_relse(attr->xattri_dac.leaf_bp); xfs_iunlock(ip, XFS_ILOCK_EXCL); xfs_irele(ip); return 0;
I think it's a bug that patch 4 doesn't release leaf_bp, since we require code that holds a buffer across a commit to relse it explicitly.
--D
Quoted 150 lines
Quoted 78 lines
Quoted 63 lines
fs/xfs/libxfs/xfs_defer.c +3
-0{struct xfs_mount *mp = tp->t_mountp;struct xfs_defer_capture *dfc;@@ -703,6 +705,9 @@ xfs_defer_ops_capture_and_commit(if (!dfc)return xfs_trans_commit (tp); +if (bp && bp->b_transp == tp) +list_add_tail (&bp->b_delay, &dfc->dfc_buffers); +/* Commit the transaction and add the capture structure to the list. */ error =xfs_trans_commit (tp);if (error) {fs/xfs/libxfs/xfs_defer.h +3
-1@@ -74,6 +74,7 @@ struct xfs_defer_capture {/* Deferred ops state saved from the transaction. */ struct list_head dfc_dfops; +struct list_head dfc_buffers;unsigned int dfc_tpflags;/* Block reservations for the data and rt devices. */ @@ -95,7 +96,8 @@ struct xfs_defer_capture {* This doesn't normally happen except log recovery. */ int xfs_defer_ops_capture_and_commit (struct xfs_trans *tp, -struct xfs_inode *capture_ip,struct list_head *capture_list); +struct xfs_inode *capture_ip,struct list_head *capture_list, +struct xfs_buf *bp);void xfs_defer_ops_continue (struct xfs_defer_capture *d,struct xfs_trans *tp,struct xfs_inode **captured_ipp);void xfs_defer_ops_release (struct xfs_mount *mp,struct xfs_defer_capture *d);fs/xfs/xfs_bmap_item.c +1
-1@@ -532,7 +532,7 @@ xfs_bui_item_recover(* Commit transaction, which frees the transaction and saves the inode * for later replay activities. */ - error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list); + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list,NULL );if (error) goto err_unlock;fs/xfs/xfs_buf.c +1
-0@@ -233,6 +233,7 @@ _xfs_buf_alloc(init_completion (&bp->b_iowait);INIT_LIST_HEAD (&bp->b_lru);INIT_LIST_HEAD (&bp->b_list); +INIT_LIST_HEAD (&bp->b_delay);INIT_LIST_HEAD (&bp->b_li_list);sema_init (&bp->b_sema,0 );/* held, no waiters */ spin_lock_init (&bp->b_lock);fs/xfs/xfs_buf.h +0
-0fs/xfs/xfs_buf.h struct xfs_buf { · line 157
Quoted 7 lines
fs/xfs/xfs_buf.h +1
-0@@ -151,6 +151,7 @@ struct xfs_buf {int b_io_error;/* internal IO error state */ wait_queue_head_t b_waiters;/* unpin waiters */ struct list_head b_list; +struct list_head b_delay;/* delayed operations list */ struct xfs_perag *b_pag;/* contains rbtree root */ struct xfs_mount *b_mount;struct xfs_buftarg *b_target;/* buffer target (device) */ The bare list-conveyance machinery looks fine to me, but adding 16 bytes to struct
xfs_buffor something that only happens during log recovery is rather expensive. Can you reuseb_listfor this purpose? I think the only user ofb_listare the buffer delwri functions, which shouldn't be active here since the xattr recovery mechanism (a) holds the buffer lock and (b) doesn't itself use delwri buffer lists for xattr leaf blocks.Quoted 2 lines
(The AIL uses delwri lists, but it won't touch a locked buffer.)
Sure, it sounds like it would work, will try it out.
Quoted 25 lines
Quoted 25 lines
fs/xfs/xfs_extfree_item.c +1
-1@@ -637,7 +637,7 @@ xfs_efi_item_recover( } -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_trans_cancel (tp);fs/xfs/xfs_log_recover.c +1
-0@@ -2465,6 +2465,7 @@ xlog_finish_defer_ops( struct list_head *capture_list) {struct xfs_defer_capture *dfc, *next; +struct xfs_buf *bp, *bnext;struct xfs_trans *tp;struct xfs_inode *ip;int error =0 ;@@ -2489,6 +2490,12 @@ xlog_finish_defer_ops(return error; }Quoted 8 lines
fs/xfs/xfs_log_recover.c xlog_finish_defer_ops( · line 2497
Quoted 5 lines
fs/xfs/xfs_log_recover.c +5
-0+list_for_each_entry_safe (bp, bnext, &dfc->dfc_buffers, b_delay) { +xfs_trans_bjoin (tp, bp); +xfs_trans_bhold (tp, bp); +list_del_init (&bp->b_delay); + }Why isn't this in
xfs_defer_ops_continue, like the code that extracts the inodes from the capture struct and hands them back to the caller?Its just what was discussed in the last review is all. That does look like a better place for it though. Will move there.
Allison
Quoted 30 lines
Quoted 30 lines
fs/xfs/xfs_log_recover.c +1
-0+/* * Transfer to this new transaction all the dfops we captured * from recovering a single intent item. fs/xfs/xfs_refcount_item.c +1
-1@@ -557,7 +557,7 @@ xfs_cui_item_recover( }xfs_refcount_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_refcount_finish_one_cleanup (tp, rcur, error);fs/xfs/xfs_rmap_item.c +1
-1@@ -587,7 +587,7 @@ xfs_rui_item_recover( }xfs_rmap_finish_one_cleanup (tp, rcur, error); -return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list); +return xfs_defer_ops_capture_and_commit (tp,NULL , capture_list,NULL ); abort_error:xfs_rmap_finish_one_cleanup (tp, rcur, error);Signature · 1 line
2.25.1
PATCH v24 3/11 xfs: Set up infrastructure for log atrribute replay
In reply to Allison HendersonCurrently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this.
This patch adds two new log item types for setting or removing attributes as deferred operations. The xfs_attri_log_item will log an intent to set or remove an attribute. The corresponding xfs_attrd_log_item holds a reference to the xfs_attri_log_item and is freed once the transaction is done. Both log items use a generic xfs_attr_log_format structure that contains the attribute name, value, flags, inode, and an op_flag that indicates if the operations is a set or remove.
--- fs/xfs/Makefile | 1 + fs/xfs/libxfs/xfs_attr.c | 5 +- fs/xfs/libxfs/xfs_attr.h | 31 +++ fs/xfs/libxfs/xfs_defer.h | 2 + fs/xfs/libxfs/xfs_log_format.h | 44 +++- fs/xfs/libxfs/xfs_log_recover.h | 2 + fs/xfs/scrub/common.c | 2 + fs/xfs/xfs_attr_item.c | 453 ++++++++++++++++++++++++++++++++ fs/xfs/xfs_attr_item.h | 52 ++++ fs/xfs/xfs_attr_list.c | 1 + fs/xfs/xfs_ioctl32.c | 2 + fs/xfs/xfs_iops.c | 2 + fs/xfs/xfs_log.c | 4 + fs/xfs/xfs_log_recover.c | 2 + fs/xfs/xfs_ondisk.h | 2 + 15 files changed, 600 insertions(+), 5 deletions(-)
fs/xfs/Makefile +1 -0
@@ -102,6 +102,7 @@ xfs-y += xfs_log.o \ xfs_buf_item_recover.o \ xfs_dquot_item_recover.o \ xfs_extfree_item.o \ + xfs_attr_item.o \ xfs_icreate_item.o \ xfs_inode_item.o \ xfs_inode_item_recover.o \
fs/xfs/libxfs/xfs_attr.c +2 -3
@@ -24,6 +24,7 @@#include "xfs_quota.h" #include "xfs_trans_space.h" #include "xfs_trace.h" +#include "xfs_attr_item.h" /* * xfs_attr.c
@@ -61,8 +62,6 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args, struct xfs_da_state **state); STATIC int xfs_attr_fillstate(xfs_da_state_t *state); STATIC int xfs_attr_refillstate(xfs_da_state_t *state); -STATIC int xfs_attr_set_iter(struct xfs_delattr_context *dac, -struct xfs_buf **leaf_bp); STATIC int xfs_attr_node_removename(struct xfs_da_args *args,struct xfs_da_state *state);
@@ -166,7 +165,7 @@ xfs_attr_get(/* * Calculate how many blocks we need for the new attribute, */ -STATIC int +int xfs_attr_calc_size (struct xfs_da_args *args,int *local)
fs/xfs/libxfs/xfs_attr.h +31 -0
@@ -28,6 +28,11 @@ struct xfs_attr_list_context;*/ #define ATTR_MAX_VALUELEN (64*1024)/* max length of a value */ +static inline bool xfs_has_larp (struct xfs_mount *mp) +{ +return false; +} +/* * Kernel-internal version of the attrlist cursor. */
@@ -454,6 +459,7 @@ enum xfs_delattr_state {*/ #define XFS_DAC_DEFER_FINISH 0x01/* finish the transaction */ #define XFS_DAC_LEAF_ADDNAME_INIT 0x02/* xfs_attr_leaf_addname init*/ +#define XFS_DAC_DELAYED_OP_INIT 0x04/* delayed operations init*/ /* * Context used for keeping track of delayed attribute operations
@@ -461,6 +467,11 @@ enum xfs_delattr_state {struct xfs_delattr_context {struct xfs_da_args *da_args; +/* +* Used by xfs_attr_set to hold a leaf buffer across a transaction roll +*/ +struct xfs_buf *leaf_bp; +/* Used in xfs_attr_rmtval_set_blk to roll through allocating blocks */ struct xfs_bmbt_irec map;xfs_dablk_t lblkno;
@@ -474,6 +485,23 @@ struct xfs_delattr_context {enum xfs_delattr_state dela_state; }; +/* +* List of attrs to commit later. +*/ +struct xfs_attr_item { +struct xfs_delattr_context xattri_dac; + +/* +* Indicates if the attr operation is a set or a remove +* XFS_ATTR_OP_FLAGS_{SET,REMOVE} +*/ +unsigned int xattri_op_flags; + +/* used to log this item to an intent */ +struct list_head xattri_list; +}; + +/*======================================================================== * Function prototypes for the kernel. *========================================================================*/
@@ -490,10 +518,13 @@ int xfs_attr_get_ilocked(struct xfs_da_args *args);int xfs_attr_get (struct xfs_da_args *args);int xfs_attr_set (struct xfs_da_args *args);int xfs_attr_set_args (struct xfs_da_args *args); +int xfs_attr_set_iter (struct xfs_delattr_context *dac, +struct xfs_buf **leaf_bp);int xfs_attr_remove_args (struct xfs_da_args *args);int xfs_attr_remove_iter (struct xfs_delattr_context *dac);bool xfs_attr_namecheck (const void *name,size_t length);void xfs_delattr_context_init (struct xfs_delattr_context *dac,struct xfs_da_args *args); +int xfs_attr_calc_size (struct xfs_da_args *args,int *local);#endif /* __XFS_ATTR_H__ */
fs/xfs/libxfs/xfs_defer.h +2 -0
@@ -63,6 +63,8 @@ extern const struct xfs_defer_op_type xfs_refcount_update_defer_type;extern const struct xfs_defer_op_type xfs_rmap_update_defer_type;extern const struct xfs_defer_op_type xfs_extent_free_defer_type;extern const struct xfs_defer_op_type xfs_agfl_free_defer_type; +extern const struct xfs_defer_op_type xfs_attr_defer_type; +/* * This structure enables a dfops user to detach the chain of deferred
fs/xfs/libxfs/xfs_log_format.h +42 -2
66 lines not on this page · raw message
fs/xfs/libxfs/xfs_log_recover.h +2 -0
@@ -72,6 +72,8 @@ extern const struct xlog_recover_item_ops xlog_rui_item_ops;extern const struct xlog_recover_item_ops xlog_rud_item_ops;extern const struct xlog_recover_item_ops xlog_cui_item_ops;extern const struct xlog_recover_item_ops xlog_cud_item_ops; +extern const struct xlog_recover_item_ops xlog_attri_item_ops; +extern const struct xlog_recover_item_ops xlog_attrd_item_ops;/* * Macros, structures, prototypes for internal log manager use.
fs/xfs/scrub/common.c +2 -0
@@ -23,6 +23,8 @@#include "xfs_rmap_btree.h" #include "xfs_log.h" #include "xfs_trans_priv.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_reflink.h" #include "xfs_ag.h"
fs/xfs/xfs_attr_item.c new file +453 -0
453 lines not on this page · raw message
fs/xfs/xfs_attr_item.h new file +52 -0
52 lines not on this page · raw message
fs/xfs/xfs_attr_list.c +1 -0
@@ -15,6 +15,7 @@#include "xfs_inode.h" #include "xfs_trans.h" #include "xfs_bmap.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_attr_sf.h" #include "xfs_attr_leaf.h"
fs/xfs/xfs_ioctl32.c +2 -0
@@ -17,6 +17,8 @@#include "xfs_itable.h" #include "xfs_fsops.h" #include "xfs_rtalloc.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_ioctl.h" #include "xfs_ioctl32.h"
fs/xfs/xfs_iops.c +2 -0
@@ -13,6 +13,8 @@#include "xfs_inode.h" #include "xfs_acl.h" #include "xfs_quota.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_trans.h" #include "xfs_trace.h"
fs/xfs/xfs_log.c +4 -0
@@ -2157,6 +2157,10 @@ xlog_print_tic_res(REG_TYPE_STR (CUD_FORMAT ,"cud_format" ),REG_TYPE_STR (BUI_FORMAT ,"bui_format" ),REG_TYPE_STR (BUD_FORMAT ,"bud_format" ), +REG_TYPE_STR (ATTRI_FORMAT ,"attri_format" ), +REG_TYPE_STR (ATTRD_FORMAT ,"attrd_format" ), +REG_TYPE_STR (ATTR_NAME ,"attr name" ), +REG_TYPE_STR (ATTR_VALUE ,"attr value" ), };BUILD_BUG_ON (ARRAY_SIZE (res_type_str) !=XLOG_REG_TYPE_MAX +1 );#undef REG_TYPE_STR
fs/xfs/xfs_log_recover.c +2 -0
@@ -1800,6 +1800,8 @@ static const struct xlog_recover_item_ops *xlog_recover_item_ops[] = { &xlog_cud_item_ops, &xlog_bui_item_ops, &xlog_bud_item_ops, + &xlog_attri_item_ops, + &xlog_attrd_item_ops, }; static const struct xlog_recover_item_ops *
fs/xfs/xfs_ondisk.h +2 -0
@@ -132,6 +132,8 @@ xfs_check_ondisk_structs(void) XFS_CHECK_STRUCT_SIZE(struct xfs_inode_log_format, 56); XFS_CHECK_STRUCT_SIZE(struct xfs_qoff_logformat, 20); XFS_CHECK_STRUCT_SIZE(struct xfs_trans_header, 16); + XFS_CHECK_STRUCT_SIZE(struct xfs_attri_log_format, 40); + XFS_CHECK_STRUCT_SIZE(struct xfs_attrd_log_format, 16);/* * The v5 superblock format extended several v4 header structures with
Signature · 1 line
2.25.1
Re: [PATCH v24 03/11] xfs: Set up infrastructure for log atrribute replay
Quoted 13 lines · Allison Henderson
Currently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this.
Quoted 8 lines
This patch adds two new log item types for setting or removing attributes as deferred operations. The
xfs_attri_log_itemwill log an intent to set or remove an attribute. The correspondingxfs_attrd_log_itemholds a reference to thexfs_attri_log_itemand is freed once the transaction is done. Both log items use a genericxfs_attr_log_formatstructure that contains the attribute name, value, flags, inode, and anop_flagthat indicates if the operations is a set or remove.
Apart from the minor nits mentioned below, the remaining changes look good to me.
Quoted 79 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/Makefile | 1 + fs/xfs/libxfs/xfs_attr.c | 5 +- fs/xfs/libxfs/xfs_attr.h | 31 +++ fs/xfs/libxfs/xfs_defer.h | 2 + fs/xfs/libxfs/xfs_log_format.h | 44 +++- fs/xfs/libxfs/xfs_log_recover.h | 2 + fs/xfs/scrub/common.c | 2 + fs/xfs/xfs_attr_item.c | 453 ++++++++++++++++++++++++++++++++ fs/xfs/xfs_attr_item.h | 52 ++++ fs/xfs/xfs_attr_list.c | 1 + fs/xfs/xfs_ioctl32.c | 2 + fs/xfs/xfs_iops.c | 2 + fs/xfs/xfs_log.c | 4 + fs/xfs/xfs_log_recover.c | 2 + fs/xfs/xfs_ondisk.h | 2 + 15 files changed, 600 insertions(+), 5 deletions(-)fs/xfs/Makefile +1
-0@@ -102,6 +102,7 @@ xfs-y += xfs_log.o \ xfs_buf_item_recover.o \ xfs_dquot_item_recover.o \ xfs_extfree_item.o \ + xfs_attr_item.o \ xfs_icreate_item.o \ xfs_inode_item.o \ xfs_inode_item_recover.o \fs/xfs/libxfs/xfs_attr.c +2
-3@@ -24,6 +24,7 @@#include "xfs_quota.h" #include "xfs_trans_space.h" #include "xfs_trace.h" +#include "xfs_attr_item.h" /* * xfs_attr.c @@ -61,8 +62,6 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args, struct xfs_da_state **state); STATIC int xfs_attr_fillstate(xfs_da_state_t *state); STATIC int xfs_attr_refillstate(xfs_da_state_t *state); -STATIC int xfs_attr_set_iter(struct xfs_delattr_context *dac, -struct xfs_buf **leaf_bp); STATIC int xfs_attr_node_removename(struct xfs_da_args *args,struct xfs_da_state *state);@@ -166,7 +165,7 @@ xfs_attr_get(/* * Calculate how many blocks we need for the new attribute, */ -STATIC int +int xfs_attr_calc_size (struct xfs_da_args *args,int *local)fs/xfs/libxfs/xfs_attr.h +5
-0@@ -28,6 +28,11 @@ struct xfs_attr_list_context;*/ #define ATTR_MAX_VALUELEN (64*1024)/* max length of a value */ +static inline bool xfs_has_larp (struct xfs_mount *mp) +{ +return false; +} +/* * Kernel-internal version of the attrlist cursor. */
fs/xfs/libxfs/xfs_attr.h enum xfs_delattr_state { · line 462
Quoted 4 lines
fs/xfs/libxfs/xfs_attr.h +1
-0@@ -454,6 +459,7 @@ enum xfs_delattr_state {*/ #define XFS_DAC_DEFER_FINISH 0x01/* finish the transaction */ #define XFS_DAC_LEAF_ADDNAME_INIT 0x02/* xfs_attr_leaf_addname init*/ +#define XFS_DAC_DELAYED_OP_INIT 0x04/* delayed operations init*/
A future patch in the series assigns a value of 0x02 to XFS_DAC_DELAYED_OP_INIT. Also, this macro is not used by any of the patches in this series.
Quoted 172 lines
fs/xfs/libxfs/xfs_attr.h +25
-0/* * Context used for keeping track of delayed attribute operations @@ -461,6 +467,11 @@ enum xfs_delattr_state {struct xfs_delattr_context {struct xfs_da_args *da_args; +/* +* Used by xfs_attr_set to hold a leaf buffer across a transaction roll +*/ +struct xfs_buf *leaf_bp; +/* Used in xfs_attr_rmtval_set_blk to roll through allocating blocks */ struct xfs_bmbt_irec map;xfs_dablk_t lblkno;@@ -474,6 +485,23 @@ struct xfs_delattr_context {enum xfs_delattr_state dela_state; }; +/* +* List of attrs to commit later. +*/ +struct xfs_attr_item { +struct xfs_delattr_context xattri_dac; + +/* +* Indicates if the attr operation is a set or a remove +* XFS_ATTR_OP_FLAGS_{SET,REMOVE} +*/ +unsigned int xattri_op_flags; + +/* used to log this item to an intent */ +struct list_head xattri_list; +}; + +/*======================================================================== * Function prototypes for the kernel. *========================================================================*/ @@ -490,10 +518,13 @@ int xfs_attr_get_ilocked(struct xfs_da_args *args);int xfs_attr_get (struct xfs_da_args *args);int xfs_attr_set (struct xfs_da_args *args);int xfs_attr_set_args (struct xfs_da_args *args); +int xfs_attr_set_iter (struct xfs_delattr_context *dac, +struct xfs_buf **leaf_bp);int xfs_attr_remove_args (struct xfs_da_args *args);int xfs_attr_remove_iter (struct xfs_delattr_context *dac);bool xfs_attr_namecheck (const void *name,size_t length);void xfs_delattr_context_init (struct xfs_delattr_context *dac,struct xfs_da_args *args); +int xfs_attr_calc_size (struct xfs_da_args *args,int *local);#endif /* __XFS_ATTR_H__ */ fs/xfs/libxfs/xfs_defer.h +2
-0@@ -63,6 +63,8 @@ extern const struct xfs_defer_op_type xfs_refcount_update_defer_type;extern const struct xfs_defer_op_type xfs_rmap_update_defer_type;extern const struct xfs_defer_op_type xfs_extent_free_defer_type;extern const struct xfs_defer_op_type xfs_agfl_free_defer_type; +extern const struct xfs_defer_op_type xfs_attr_defer_type; +/* * This structure enables a dfops user to detach the chain of deferred fs/xfs/libxfs/xfs_log_format.h +42
-2@@ -114,7 +114,12 @@ struct xfs_unmount_log_format {#define XLOG_REG_TYPE_CUD_FORMAT 24#define XLOG_REG_TYPE_BUI_FORMAT 25#define XLOG_REG_TYPE_BUD_FORMAT 26 -#define XLOG_REG_TYPE_MAX 26 +#define XLOG_REG_TYPE_ATTRI_FORMAT 27 +#define XLOG_REG_TYPE_ATTRD_FORMAT 28 +#define XLOG_REG_TYPE_ATTR_NAME 29 +#define XLOG_REG_TYPE_ATTR_VALUE 30 +#define XLOG_REG_TYPE_MAX 30 +/* * Flags to log operation header @@ -237,6 +242,8 @@ typedef struct xfs_trans_header {#define XFS_LI_CUD 0x1243#define XFS_LI_BUI 0x1244/* bmbt update intent */ #define XFS_LI_BUD 0x1245 +#define XFS_LI_ATTRI 0x1246/* attr set/remove intent*/ +#define XFS_LI_ATTRD 0x1247/* attr set/remove done */ #define XFS_LI_TYPE_DESC \ { XFS_LI_EFI, "XFS_LI_EFI" }, \@@ -252,7 +259,9 @@ typedef struct xfs_trans_header { { XFS_LI_CUI, "XFS_LI_CUI" }, \ { XFS_LI_CUD, "XFS_LI_CUD" }, \ { XFS_LI_BUI, "XFS_LI_BUI" }, \ - { XFS_LI_BUD, "XFS_LI_BUD" } + { XFS_LI_BUD, "XFS_LI_BUD" }, \ + { XFS_LI_ATTRI, "XFS_LI_ATTRI" }, \ + { XFS_LI_ATTRD, "XFS_LI_ATTRD" }/* * Inode Log Item Format definitions. @@ -869,4 +878,35 @@ struct xfs_icreate_log { __be32 icl_gen; /* inode generation number to use */ }; +/* +* Flags for deferred attribute operations. +* Upper bits are flags, lower byte is type code +*/ +#define XFS_ATTR_OP_FLAGS_SET 1/* Set the attribute */ +#define XFS_ATTR_OP_FLAGS_REMOVE 2/* Remove the attribute */ +#define XFS_ATTR_OP_FLAGS_TYPE_MASK 0xFF/* Flags type mask */ + +/* +* This is the structure used to lay out an attr log item in the +* log. +*/ +struct xfs_attri_log_format { +uint16_t alfi_type;/* attri log item type */ +uint16_t alfi_size;/* size of this item */ +uint32_t __pad;/* pad to 64 bit aligned */ +uint64_t alfi_id;/* attri identifier */ +uint64_t alfi_ino;/* the inode for this attr operation */ +uint32_t alfi_op_flags;/* marks the op as a set or remove */ +uint32_t alfi_name_len;/* attr name length */ +uint32_t alfi_value_len;/* attr value length */ +uint32_t alfi_attr_flags;/* attr flags */ +}; + +struct xfs_attrd_log_format { +uint16_t alfd_type;/* attrd log item type */ +uint16_t alfd_size;/* size of this item */ +uint32_t __pad;/* pad to 64 bit aligned */ +uint64_t alfd_alf_id;/* id of corresponding attri */ +}; +#endif /* __XFS_LOG_FORMAT_H__ */ fs/xfs/libxfs/xfs_log_recover.h +2
-0@@ -72,6 +72,8 @@ extern const struct xlog_recover_item_ops xlog_rui_item_ops;extern const struct xlog_recover_item_ops xlog_rud_item_ops;extern const struct xlog_recover_item_ops xlog_cui_item_ops;extern const struct xlog_recover_item_ops xlog_cud_item_ops; +extern const struct xlog_recover_item_ops xlog_attri_item_ops; +extern const struct xlog_recover_item_ops xlog_attrd_item_ops;/* * Macros, structures, prototypes for internal log manager use. fs/xfs/scrub/common.c +2
-0@@ -23,6 +23,8 @@#include "xfs_rmap_btree.h" #include "xfs_log.h" #include "xfs_trans_priv.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_reflink.h" #include "xfs_ag.h" fs/xfs/xfs_attr_item.c new file +0
-0
fs/xfs/xfs_attr_item.c line 6
Quoted 6 lines
fs/xfs/xfs_attr_item.c new file +6
-0@@ -0,0 +1,453 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Copyright (C) 2021 Oracle. All Rights Reserved. +* Author: Allison Collins <[email protected]> +*/ +
Some of the header files included below are not required. I was able to compile successfully without including xfs_bit.h, xfs_bmap_btree.h, xfs_buf_item.h, xfs_btree.h, xfs_rmap.h, xfs_icache.h, xfs_alloc.h, xfs_bmap.h and xfs_quota.h. Also xfs_shared.h is included twice.
Quoted 460 lines
fs/xfs/xfs_attr_item.c new file +447
-0+#include "xfs.h" +#include "xfs_fs.h" +#include "xfs_format.h" +#include "xfs_trans_resv.h" +#include "xfs_bit.h" +#include "xfs_shared.h" +#include "xfs_mount.h" +#include "xfs_defer.h" +#include "xfs_log_format.h" +#include "xfs_trans.h" +#include "xfs_bmap_btree.h" +#include "xfs_trans_priv.h" +#include "xfs_buf_item.h" +#include "xfs_log.h" +#include "xfs_btree.h" +#include "xfs_rmap.h" +#include "xfs_inode.h" +#include "xfs_icache.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" +#include "xfs_attr.h" +#include "xfs_shared.h" +#include "xfs_attr_item.h" +#include "xfs_alloc.h" +#include "xfs_bmap.h" +#include "xfs_trace.h" +#include "libxfs/xfs_da_format.h" +#include "xfs_inode.h" +#include "xfs_quota.h" +#include "xfs_trans_space.h" +#include "xfs_error.h" +#include "xfs_log_priv.h" +#include "xfs_log_recover.h" + +static const struct xfs_item_ops xfs_attri_item_ops; +static const struct xfs_item_ops xfs_attrd_item_ops; + +/* iovec length must be 32-bit aligned */ +static inline size_t ATTR_NVEC_SIZE (size_t size) +{ +return size ==sizeof (int32_t ) ? size : +sizeof (int32_t ) +round_up (size,sizeof (int32_t )); +} + +static inline struct xfs_attri_log_item *ATTRI_ITEM (struct xfs_log_item *lip) +{ + return container_of(lip, struct xfs_attri_log_item, attri_item); +} + +STATIC void +xfs_attri_item_free ( +struct xfs_attri_log_item *attrip) +{ +kmem_free (attrip->attri_item.li_lv_shadow); +kmem_free (attrip); +} + +/* +* Freeing the attrip requires that we remove it from the AIL if it has already +* been placed there. However, the ATTRI may not yet have been placed in the +* AIL when called by xfs_attri_release() from ATTRD processing due to the +* ordering of committed vs unpin operations in bulk insert operations. Hence +* the reference count to ensure only the last caller frees the ATTRI. +*/ +STATIC void +xfs_attri_release ( +struct xfs_attri_log_item *attrip) +{ +ASSERT (atomic_read (&attrip->attri_refcount) >0 ); +if (atomic_dec_and_test (&attrip->attri_refcount)) { +xfs_trans_ail_delete (&attrip->attri_item, +SHUTDOWN_LOG_IO_ERROR ); +xfs_attri_item_free (attrip); + } +} + +STATIC void +xfs_attri_item_size ( +struct xfs_log_item *lip, +int *nvecs, +int *nbytes) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); + + *nvecs +=1 ; + *nbytes +=sizeof (struct xfs_attri_log_format ); + +/* Attr set and remove operations require a name */ +ASSERT (attrip->attri_name_len >0 ); + + *nvecs +=1 ; + *nbytes +=ATTR_NVEC_SIZE (attrip->attri_name_len); + +if (attrip->attri_value_len >0 ) { + *nvecs +=1 ; + *nbytes +=ATTR_NVEC_SIZE (attrip->attri_value_len); + } +} + +/* +* This is called to fill in the log iovecs for the given attri log +* item. We use 1 iovec for the attri_format_item, 1 for the name, and +* another for the value if it is present +*/ +STATIC void +xfs_attri_item_format ( +struct xfs_log_item *lip, +struct xfs_log_vec *lv) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); +struct xfs_log_iovec *vecp =NULL ; + + attrip->attri_format.alfi_type =XFS_LI_ATTRI ; + attrip->attri_format.alfi_size =1 ; + +/* +* This size accounting must be done before copying the attrip into the +* iovec. If we do it after, the wrong size will be recorded to the log +* and we trip across assertion checks for bad region sizes later during +* the log recovery. +*/ + +ASSERT (attrip->attri_name_len >0 ); + attrip->attri_format.alfi_size++; + +if (attrip->attri_value_len >0 ) + attrip->attri_format.alfi_size++; + +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTRI_FORMAT , + &attrip->attri_format, +sizeof (struct xfs_attri_log_format )); +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTR_NAME , + attrip->attri_name, +ATTR_NVEC_SIZE (attrip->attri_name_len)); +if (attrip->attri_value_len >0 ) +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTR_VALUE , + attrip->attri_value, +ATTR_NVEC_SIZE (attrip->attri_value_len)); +} + +/* +* The unpin operation is the last place an ATTRI is manipulated in the log. It +* is either inserted in the AIL or aborted in the event of a log I/O error. In +* either case, the ATTRI transaction has been successfully committed to make +* it this far. Therefore, we expect whoever committed the ATTRI to either +* construct and commit the ATTRD or drop the ATTRD's reference in the event of +* error. Simply drop the log's ATTRI reference now that the log is done with +* it. +*/ +STATIC void +xfs_attri_item_unpin ( +struct xfs_log_item *lip, +int remove) +{ +xfs_attri_release (ATTRI_ITEM (lip)); +} + + +STATIC void +xfs_attri_item_release ( +struct xfs_log_item *lip) +{ +xfs_attri_release (ATTRI_ITEM (lip)); +} + +/* +* Allocate and initialize an attri item. Caller may allocate an additional +* trailing buffer of the specified size +*/ +STATIC struct xfs_attri_log_item * +xfs_attri_init ( +struct xfs_mount *mp, +int buffer_size) + +{ +struct xfs_attri_log_item *attrip; +uint size; + + size =sizeof (struct xfs_attri_log_item ) + buffer_size; + attrip =kvmalloc (size,KM_ZERO ); +if (attrip ==NULL ) +return NULL ; + +xfs_log_item_init (mp, &attrip->attri_item,XFS_LI_ATTRI , + &xfs_attri_item_ops); + attrip->attri_format.alfi_id = (uintptr_t )(void *)attrip; +atomic_set (&attrip->attri_refcount,2 ); + +return attrip; +} + +/* +* Copy an attr format buffer from the given buf, and into the destination attr +* format structure. +*/ +STATIC int +xfs_attri_copy_format ( +struct xfs_log_iovec *buf, +struct xfs_attri_log_format *dst_attr_fmt) +{ +struct xfs_attri_log_format *src_attr_fmt = buf->i_addr; +uint len; + + len =sizeof (struct xfs_attri_log_format ); +if (buf->i_len != len) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW ,NULL ); +return -EFSCORRUPTED ; + } + +memcpy ((char *)dst_attr_fmt, (char *)src_attr_fmt, len); +return 0 ; +} + +static inline struct xfs_attrd_log_item *ATTRD_ITEM (struct xfs_log_item *lip) +{ + return container_of(lip, struct xfs_attrd_log_item, attrd_item); +} + +STATIC void +xfs_attrd_item_free (struct xfs_attrd_log_item *attrdp) +{ +kmem_free (attrdp->attrd_item.li_lv_shadow); +kmem_free (attrdp); +} + +STATIC void +xfs_attrd_item_size ( +struct xfs_log_item *lip, +int *nvecs, +int *nbytes) +{ + *nvecs +=1 ; + *nbytes +=sizeof (struct xfs_attrd_log_format ); +} + +/* +* This is called to fill in the log iovecs for the given attrd log item. We use +* only 1 iovec for the attrd_format, and we point that at the attr_log_format +* structure embedded in the attrd item. +*/ +STATIC void +xfs_attrd_item_format ( +struct xfs_log_item *lip, +struct xfs_log_vec *lv) +{ +struct xfs_attrd_log_item *attrdp =ATTRD_ITEM (lip); +struct xfs_log_iovec *vecp =NULL ; + + attrdp->attrd_format.alfd_type =XFS_LI_ATTRD ; + attrdp->attrd_format.alfd_size =1 ; + +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTRD_FORMAT , + &attrdp->attrd_format, +sizeof (struct xfs_attrd_log_format )); +} + +/* +* The ATTRD is either committed or aborted if the transaction is canceled. If +* the transaction is canceled, drop our reference to the ATTRI and free the +* ATTRD. +*/ +STATIC void +xfs_attrd_item_release ( +struct xfs_log_item *lip) +{ +struct xfs_attrd_log_item *attrdp =ATTRD_ITEM (lip); + +xfs_attri_release (attrdp->attrd_attrip); +xfs_attrd_item_free (attrdp); +} + +STATIC xfs_lsn_t +xfs_attri_item_committed ( +struct xfs_log_item *lip, +xfs_lsn_t lsn) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); + +/* +* The attrip refers to xfs_attr_item memory to log the name and value +* with the intent item. This already occurred when the intent was +* committed so these fields are no longer accessed. Clear them out of +* caution since we're about to free the xfs_attr_item. +*/ + attrip->attri_name =NULL ; + attrip->attri_value =NULL ; + +/* +* The ATTRI is logged only once and cannot be moved in the log, so +* simply return the lsn at which it's been logged. +*/ +return lsn; +} + +STATIC bool +xfs_attri_item_match ( +struct xfs_log_item *lip, +uint64_t intent_id) +{ +return ATTRI_ITEM (lip)->attri_format.alfi_id == intent_id; +} + +static const struct xfs_item_ops xfs_attrd_item_ops = { + .flags =XFS_ITEM_RELEASE_WHEN_COMMITTED , + .iop_size = xfs_attrd_item_size, + .iop_format = xfs_attrd_item_format, + .iop_release = xfs_attrd_item_release, +}; + +/* Is this recovered ATTRI ok? */ +static inline bool +xfs_attri_validate ( +struct xfs_mount *mp, +struct xfs_attri_log_item *attrip) +{ +struct xfs_attri_log_format *attrp = &attrip->attri_format; +unsigned int op = attrp->alfi_op_flags & +XFS_ATTR_OP_FLAGS_TYPE_MASK ; + +/* alfi_op_flags should be either a set or remove */ +if (op !=XFS_ATTR_OP_FLAGS_SET && op !=XFS_ATTR_OP_FLAGS_REMOVE ) +return false; + +if (attrp->alfi_value_len >XATTR_SIZE_MAX ) +return false; + +if ((attrp->alfi_name_len >XATTR_NAME_MAX ) || + (attrp->alfi_name_len ==0 )) +return false; + +return xfs_verify_ino (mp, attrp->alfi_ino); +} + +static const struct xfs_item_ops xfs_attri_item_ops = { + .iop_size = xfs_attri_item_size, + .iop_format = xfs_attri_item_format, + .iop_unpin = xfs_attri_item_unpin, + .iop_committed = xfs_attri_item_committed, + .iop_release = xfs_attri_item_release, + .iop_match = xfs_attri_item_match, +}; + + + +STATIC int +xlog_recover_attri_commit_pass2 ( +struct xlog *log, +struct list_head *buffer_list, +struct xlog_recover_item *item, +xfs_lsn_t lsn) +{ +int error; +struct xfs_mount *mp = log->l_mp; +struct xfs_attri_log_item *attrip; +struct xfs_attri_log_format *attri_formatp; +char *name =NULL ; +char *value =NULL ; +int region =0 ; +int buffer_size; + + attri_formatp = item->ri_buf[region].i_addr; + +/* Validate xfs_attri_log_format */ +if (attri_formatp->__pad !=0 || attri_formatp->alfi_name_len ==0 || + (attri_formatp->alfi_op_flags ==XFS_ATTR_OP_FLAGS_REMOVE && + attri_formatp->alfi_value_len !=0 )) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW , mp); +return -EFSCORRUPTED ; + } + + buffer_size = attri_formatp->alfi_name_len + + attri_formatp->alfi_value_len; + + attrip =xfs_attri_init (mp, buffer_size); +if (attrip ==NULL ) +return -ENOMEM ; + + error =xfs_attri_copy_format (&item->ri_buf[region], + &attrip->attri_format); +if (error) { +xfs_attri_item_free (attrip); +return error; + } + + attrip->attri_name_len = attri_formatp->alfi_name_len; + attrip->attri_value_len = attri_formatp->alfi_value_len; + region++; + name = ((char *)attrip) +sizeof (struct xfs_attri_log_item ); +memcpy (name, item->ri_buf[region].i_addr, attrip->attri_name_len); + attrip->attri_name = name; + +if (attrip->attri_value_len >0 ) { + region++; + value = ((char *)attrip) +sizeof (struct xfs_attri_log_item ) + + attrip->attri_name_len; +memcpy (value, item->ri_buf[region].i_addr, + attrip->attri_value_len); + attrip->attri_value = value; + } + +/* +* The ATTRI has two references. One for the ATTRD and one for ATTRI to +* ensure it makes it into the AIL. Insert the ATTRI into the AIL +* directly and drop the ATTRI reference. Note that +* xfs_trans_ail_update() drops the AIL lock. +*/ +xfs_trans_ail_insert (log->l_ailp, &attrip->attri_item, lsn); +xfs_attri_release (attrip); +return 0 ; +} + +const struct xlog_recover_item_ops xlog_attri_item_ops = { + .item_type =XFS_LI_ATTRI , + .commit_pass2 = xlog_recover_attri_commit_pass2, +}; + +/* +* This routine is called when an ATTRD format structure is found in a committed +* transaction in the log. Its purpose is to cancel the corresponding ATTRI if +* it was still in the log. To do this it searches the AIL for the ATTRI with +* an id equal to that in the ATTRD format structure. If we find it we drop +* the ATTRD reference, which removes the ATTRI from the AIL and frees it. +*/ +STATIC int +xlog_recover_attrd_commit_pass2 ( +struct xlog *log, +struct list_head *buffer_list, +struct xlog_recover_item *item, +xfs_lsn_t lsn) +{ +struct xfs_attrd_log_format *attrd_formatp; + + attrd_formatp = item->ri_buf[0 ].i_addr; +if (item->ri_buf[0 ].i_len !=sizeof (struct xfs_attrd_log_format )) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW ,NULL ); +return -EFSCORRUPTED ; + } + +xlog_recover_release_intent (log,XFS_LI_ATTRI , + attrd_formatp->alfd_alf_id); +return 0 ; +} + +const struct xlog_recover_item_ops xlog_attrd_item_ops = { + .item_type =XFS_LI_ATTRD , + .commit_pass2 = xlog_recover_attrd_commit_pass2, +};fs/xfs/xfs_attr_item.h new file +7
-0@@ -0,0 +1,52 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later +* +* Copyright (C) 2021 Oracle. All Rights Reserved. +* Author: Allison Collins <[email protected]> +*/ +#ifndef __XFS_ATTR_ITEM_H__ +#define __XFS_ATTR_ITEM_H__
fs/xfs/xfs_attr_item.h line 17
Quoted 10 lines
fs/xfs/xfs_attr_item.h new file +10
-0+ +/* kernel only ATTRI/ATTRD definitions */ + +struct xfs_mount ; +struct kmem_zone ; + +/* +* Define ATTR flag bits. Manipulated by set/clear/test_bit operators. +*/ +#define XFS_ATTRI_RECOVERED 1
XFS_ATTRI_RECOVERED isn't used in this patch nor in any of the future patches in this series.
Quoted 114 lines
fs/xfs/xfs_attr_item.h new file +35
-0+ + +/* +* This is the "attr intention" log item. It is used to log the fact that some +* attribute operations need to be processed. An operation is currently either +* a set or remove. Set or remove operations are described by the xfs_attr_item +* which may be logged to this intent. +* +* During a normal attr operation, name and value point to the name and value +* fields of the calling functions xfs_da_args. During a recovery, the name +* and value buffers are copied from the log, and stored in a trailing buffer +* attached to the xfs_attr_item until they are committed. They are freed when +* the xfs_attr_item itself is freed when the work is done. +*/ +struct xfs_attri_log_item { +struct xfs_log_item attri_item; +atomic_t attri_refcount; +int attri_name_len; +int attri_value_len; +void *attri_name; +void *attri_value; +struct xfs_attri_log_format attri_format; +}; + +/* +* This is the "attr done" log item. It is used to log the fact that some attrs +* earlier mentioned in an attri item have been freed. +*/ +struct xfs_attrd_log_item { +struct xfs_attri_log_item *attrd_attrip; +struct xfs_log_item attrd_item; +struct xfs_attrd_log_format attrd_format; +}; + +#endif /* __XFS_ATTR_ITEM_H__ */ fs/xfs/xfs_attr_list.c +1
-0@@ -15,6 +15,7 @@#include "xfs_inode.h" #include "xfs_trans.h" #include "xfs_bmap.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_attr_sf.h" #include "xfs_attr_leaf.h" fs/xfs/xfs_ioctl32.c +2
-0@@ -17,6 +17,8 @@#include "xfs_itable.h" #include "xfs_fsops.h" #include "xfs_rtalloc.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_ioctl.h" #include "xfs_ioctl32.h" fs/xfs/xfs_iops.c +2
-0@@ -13,6 +13,8 @@#include "xfs_inode.h" #include "xfs_acl.h" #include "xfs_quota.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_trans.h" #include "xfs_trace.h" fs/xfs/xfs_log.c +4
-0@@ -2157,6 +2157,10 @@ xlog_print_tic_res(REG_TYPE_STR (CUD_FORMAT ,"cud_format" ),REG_TYPE_STR (BUI_FORMAT ,"bui_format" ),REG_TYPE_STR (BUD_FORMAT ,"bud_format" ), +REG_TYPE_STR (ATTRI_FORMAT ,"attri_format" ), +REG_TYPE_STR (ATTRD_FORMAT ,"attrd_format" ), +REG_TYPE_STR (ATTR_NAME ,"attr name" ), +REG_TYPE_STR (ATTR_VALUE ,"attr value" ), };BUILD_BUG_ON (ARRAY_SIZE (res_type_str) !=XLOG_REG_TYPE_MAX +1 );#undef REG_TYPE_STRfs/xfs/xfs_log_recover.c +2
-0@@ -1800,6 +1800,8 @@ static const struct xlog_recover_item_ops *xlog_recover_item_ops[] = { &xlog_cud_item_ops, &xlog_bui_item_ops, &xlog_bud_item_ops, + &xlog_attri_item_ops, + &xlog_attrd_item_ops, }; static const struct xlog_recover_item_ops *fs/xfs/xfs_ondisk.h +2
-0@@ -132,6 +132,8 @@ xfs_check_ondisk_structs(void) XFS_CHECK_STRUCT_SIZE(struct xfs_inode_log_format, 56); XFS_CHECK_STRUCT_SIZE(struct xfs_qoff_logformat, 20); XFS_CHECK_STRUCT_SIZE(struct xfs_trans_header, 16); + XFS_CHECK_STRUCT_SIZE(struct xfs_attri_log_format, 40); + XFS_CHECK_STRUCT_SIZE(struct xfs_attrd_log_format, 16);/* * The v5 superblock format extended several v4 header structures with
Signature · 1 line
chandan
Re: [PATCH v24 03/11] xfs: Set up infrastructure for log atrribute replay
Quoted 23 lines · Chandan Babu R
Quoted 13 lines · Allison Henderson
Currently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this.
Quoted 8 lines
This patch adds two new log item types for setting or removing attributes as deferred operations. The
xfs_attri_log_itemwill log an intent to set or remove an attribute. The correspondingxfs_attrd_log_itemholds a reference to thexfs_attri_log_itemand is freed once the transaction is done. Both log items use a genericxfs_attr_log_formatstructure that contains the attribute name, value, flags, inode, and anop_flagthat indicates if the operations is a set or remove.
Quoted 4 lines
Apart from the minor nits mentioned below, the remaining changes look good to me.
Reviewed-by Chandan Babu R <[email protected]>
Great, thank you!
Quoted 264 lines
Quoted 80 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/Makefile | 1 + fs/xfs/libxfs/xfs_attr.c | 5 +- fs/xfs/libxfs/xfs_attr.h | 31 +++ fs/xfs/libxfs/xfs_defer.h | 2 + fs/xfs/libxfs/xfs_log_format.h | 44 +++- fs/xfs/libxfs/xfs_log_recover.h | 2 + fs/xfs/scrub/common.c | 2 + fs/xfs/xfs_attr_item.c | 453 ++++++++++++++++++++++++++++++++ fs/xfs/xfs_attr_item.h | 52 ++++ fs/xfs/xfs_attr_list.c | 1 + fs/xfs/xfs_ioctl32.c | 2 + fs/xfs/xfs_iops.c | 2 + fs/xfs/xfs_log.c | 4 + fs/xfs/xfs_log_recover.c | 2 + fs/xfs/xfs_ondisk.h | 2 + 15 files changed, 600 insertions(+), 5 deletions(-)fs/xfs/Makefile +1
-0@@ -102,6 +102,7 @@ xfs-y += xfs_log.o \ xfs_buf_item_recover.o \ xfs_dquot_item_recover.o \ xfs_extfree_item.o \ + xfs_attr_item.o \ xfs_icreate_item.o \ xfs_inode_item.o \ xfs_inode_item_recover.o \fs/xfs/libxfs/xfs_attr.c +2
-3@@ -24,6 +24,7 @@#include "xfs_quota.h" #include "xfs_trans_space.h" #include "xfs_trace.h" +#include "xfs_attr_item.h" /* * xfs_attr.c @@ -61,8 +62,6 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args, struct xfs_da_state **state); STATIC int xfs_attr_fillstate(xfs_da_state_t *state); STATIC int xfs_attr_refillstate(xfs_da_state_t *state); -STATIC int xfs_attr_set_iter(struct xfs_delattr_context *dac, -struct xfs_buf **leaf_bp); STATIC int xfs_attr_node_removename(struct xfs_da_args *args,struct xfs_da_state *state);@@ -166,7 +165,7 @@ xfs_attr_get(/* * Calculate how many blocks we need for the new attribute, */ -STATIC int +int xfs_attr_calc_size (struct xfs_da_args *args,int *local)fs/xfs/libxfs/xfs_attr.h +5
-0@@ -28,6 +28,11 @@ struct xfs_attr_list_context;*/ #define ATTR_MAX_VALUELEN (64*1024)/* max length of a value */ +static inline bool xfs_has_larp (struct xfs_mount *mp) +{ +return false; +} +/* * Kernel-internal version of the attrlist cursor. */ fs/xfs/libxfs/xfs_attr.h enum xfs_delattr_state { · line 462
Quoted 4 lines
fs/xfs/libxfs/xfs_attr.h +1
-0@@ -454,6 +459,7 @@ enum xfs_delattr_state {*/ #define XFS_DAC_DEFER_FINISH 0x01/* finish the transaction */ #define XFS_DAC_LEAF_ADDNAME_INIT 0x02/* xfs_attr_leaf_addname init*/ +#define XFS_DAC_DELAYED_OP_INIT 0x04/* delayed operations init*/ A future patch in the series assigns a value of
0x02toXFS_DAC_DELAYED_OP_INIT. Also, this macro is not used by any of the patches in this series.Quoted 175 lines
fs/xfs/libxfs/xfs_attr.h +25
-0/* * Context used for keeping track of delayed attribute operations @@ -461,6 +467,11 @@ enum xfs_delattr_state {struct xfs_delattr_context {struct xfs_da_args *da_args; +/* +* Used by xfs_attr_set to hold a leaf buffer across a transaction roll +*/ +struct xfs_buf *leaf_bp; +/* Used in xfs_attr_rmtval_set_blk to roll through allocating blocks */ struct xfs_bmbt_irec map;xfs_dablk_t lblkno;@@ -474,6 +485,23 @@ struct xfs_delattr_context {enum xfs_delattr_state dela_state; }; +/* +* List of attrs to commit later. +*/ +struct xfs_attr_item { +struct xfs_delattr_context xattri_dac; + +/* +* Indicates if the attr operation is a set or a remove +* XFS_ATTR_OP_FLAGS_{SET,REMOVE} +*/ +unsigned int xattri_op_flags; + +/* used to log this item to an intent */ +struct list_head xattri_list; +}; + +/*======================================================================== * Function prototypes for the kernel. *========================================================================*/ @@ -490,10 +518,13 @@ int xfs_attr_get_ilocked(struct xfs_da_args *args);int xfs_attr_get (struct xfs_da_args *args);int xfs_attr_set (struct xfs_da_args *args);int xfs_attr_set_args (struct xfs_da_args *args); +int xfs_attr_set_iter (struct xfs_delattr_context *dac, +struct xfs_buf **leaf_bp);int xfs_attr_remove_args (struct xfs_da_args *args);int xfs_attr_remove_iter (struct xfs_delattr_context *dac);bool xfs_attr_namecheck (const void *name,size_t length);void xfs_delattr_context_init (struct xfs_delattr_context *dac,struct xfs_da_args *args); +int xfs_attr_calc_size (struct xfs_da_args *args,int *local);#endif /* __XFS_ATTR_H__ */ fs/xfs/libxfs/xfs_defer.h +2
-0@@ -63,6 +63,8 @@ extern const struct xfs_defer_op_type xfs_refcount_update_defer_type;extern const struct xfs_defer_op_type xfs_rmap_update_defer_type;extern const struct xfs_defer_op_type xfs_extent_free_defer_type;extern const struct xfs_defer_op_type xfs_agfl_free_defer_type; +extern const struct xfs_defer_op_type xfs_attr_defer_type; +/* * This structure enables a dfops user to detach the chain of deferred fs/xfs/libxfs/xfs_log_format.h +42
-2@@ -114,7 +114,12 @@ struct xfs_unmount_log_format {#define XLOG_REG_TYPE_CUD_FORMAT 24#define XLOG_REG_TYPE_BUI_FORMAT 25#define XLOG_REG_TYPE_BUD_FORMAT 26 -#define XLOG_REG_TYPE_MAX 26 +#define XLOG_REG_TYPE_ATTRI_FORMAT 27 +#define XLOG_REG_TYPE_ATTRD_FORMAT 28 +#define XLOG_REG_TYPE_ATTR_NAME 29 +#define XLOG_REG_TYPE_ATTR_VALUE 30 +#define XLOG_REG_TYPE_MAX 30 +/* * Flags to log operation header @@ -237,6 +242,8 @@ typedef struct xfs_trans_header {#define XFS_LI_CUD 0x1243#define XFS_LI_BUI 0x1244/* bmbt update intent */ #define XFS_LI_BUD 0x1245 +#define XFS_LI_ATTRI 0x1246/* attr set/remove intent*/ +#define XFS_LI_ATTRD 0x1247/* attr set/remove done */ #define XFS_LI_TYPE_DESC \ { XFS_LI_EFI, "XFS_LI_EFI" }, \@@ -252,7 +259,9 @@ typedef struct xfs_trans_header { { XFS_LI_CUI, "XFS_LI_CUI" }, \ { XFS_LI_CUD, "XFS_LI_CUD" }, \ { XFS_LI_BUI, "XFS_LI_BUI" }, \ - { XFS_LI_BUD, "XFS_LI_BUD" } + { XFS_LI_BUD, "XFS_LI_BUD" }, \ + { XFS_LI_ATTRI, "XFS_LI_ATTRI" }, \ + { XFS_LI_ATTRD, "XFS_LI_ATTRD" }/* * Inode Log Item Format definitions. @@ -869,4 +878,35 @@ struct xfs_icreate_log { __be32 icl_gen; /* inode generation number to use */ }; +/* +* Flags for deferred attribute operations. +* Upper bits are flags, lower byte is type code +*/ +#define XFS_ATTR_OP_FLAGS_SET 1/* Set the attribute */ +#define XFS_ATTR_OP_FLAGS_REMOVE 2/* Remove the attribute */ +#define XFS_ATTR_OP_FLAGS_TYPE_MASK 0xFF/* Flags type mask */ + +/* +* This is the structure used to lay out an attr log item in the +* log. +*/ +struct xfs_attri_log_format { +uint16_t alfi_type;/* attri log item type */ +uint16_t alfi_size;/* size of this item */ +uint32_t __pad;/* pad to 64 bit aligned */ +uint64_t alfi_id;/* attri identifier */ +uint64_t alfi_ino;/* the inode for this attr operation */ +uint32_t alfi_op_flags;/* marks the op as a set or remove */ +uint32_t alfi_name_len;/* attr name length */ +uint32_t alfi_value_len;/* attr value length */ +uint32_t alfi_attr_flags;/* attr flags */ +}; + +struct xfs_attrd_log_format { +uint16_t alfd_type;/* attrd log item type */ +uint16_t alfd_size;/* size of this item */ +uint32_t __pad;/* pad to 64 bit aligned */ +uint64_t alfd_alf_id;/* id of corresponding attri */ +}; +#endif /* __XFS_LOG_FORMAT_H__ */ fs/xfs/libxfs/xfs_log_recover.h +2
-0@@ -72,6 +72,8 @@ extern const struct xlog_recover_item_ops xlog_rui_item_ops;extern const struct xlog_recover_item_ops xlog_rud_item_ops;extern const struct xlog_recover_item_ops xlog_cui_item_ops;extern const struct xlog_recover_item_ops xlog_cud_item_ops; +extern const struct xlog_recover_item_ops xlog_attri_item_ops; +extern const struct xlog_recover_item_ops xlog_attrd_item_ops;/* * Macros, structures, prototypes for internal log manager use. fs/xfs/scrub/common.c +2
-0@@ -23,6 +23,8 @@#include "xfs_rmap_btree.h" #include "xfs_log.h" #include "xfs_trans_priv.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_reflink.h" #include "xfs_ag.h" fs/xfs/xfs_attr_item.c new file +3
-0@@ -0,0 +1,453 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Copyright (C) 2021 Oracle. All Rights Reserved.
Quoted 8 lines
fs/xfs/xfs_attr_item.c line 6
Quoted 3 lines
fs/xfs/xfs_attr_item.c new file +3
-0+* Author: Allison Collins <[email protected]> +*/ +Some of the header files included below are not required. I was able to compile successfully without including
xfs_bit.h,xfs_bmap_btree.h,xfs_buf_item.h,xfs_btree.h,xfs_rmap.h,xfs_icache.h,xfs_alloc.h,xfs_bmap.handxfs_quota.h. Alsoxfs_shared.his included twice.
Ok, I will see if I can clean out some includes here.
Quoted 466 lines
Quoted 466 lines
fs/xfs/xfs_attr_item.c new file +447
-0+#include "xfs.h" +#include "xfs_fs.h" +#include "xfs_format.h" +#include "xfs_trans_resv.h" +#include "xfs_bit.h" +#include "xfs_shared.h" +#include "xfs_mount.h" +#include "xfs_defer.h" +#include "xfs_log_format.h" +#include "xfs_trans.h" +#include "xfs_bmap_btree.h" +#include "xfs_trans_priv.h" +#include "xfs_buf_item.h" +#include "xfs_log.h" +#include "xfs_btree.h" +#include "xfs_rmap.h" +#include "xfs_inode.h" +#include "xfs_icache.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" +#include "xfs_attr.h" +#include "xfs_shared.h" +#include "xfs_attr_item.h" +#include "xfs_alloc.h" +#include "xfs_bmap.h" +#include "xfs_trace.h" +#include "libxfs/xfs_da_format.h" +#include "xfs_inode.h" +#include "xfs_quota.h" +#include "xfs_trans_space.h" +#include "xfs_error.h" +#include "xfs_log_priv.h" +#include "xfs_log_recover.h" + +static const struct xfs_item_ops xfs_attri_item_ops; +static const struct xfs_item_ops xfs_attrd_item_ops; + +/* iovec length must be 32-bit aligned */ +static inline size_t ATTR_NVEC_SIZE (size_t size) +{ +return size ==sizeof (int32_t ) ? size : +sizeof (int32_t ) +round_up (size,sizeof (int32_t )); +} + +static inline struct xfs_attri_log_item *ATTRI_ITEM (struct xfs_log_item *lip) +{ + return container_of(lip, struct xfs_attri_log_item, attri_item); +} + +STATIC void +xfs_attri_item_free ( +struct xfs_attri_log_item *attrip) +{ +kmem_free (attrip->attri_item.li_lv_shadow); +kmem_free (attrip); +} + +/* +* Freeing the attrip requires that we remove it from the AIL if it has already +* been placed there. However, the ATTRI may not yet have been placed in the +* AIL when called by xfs_attri_release() from ATTRD processing due to the +* ordering of committed vs unpin operations in bulk insert operations. Hence +* the reference count to ensure only the last caller frees the ATTRI. +*/ +STATIC void +xfs_attri_release ( +struct xfs_attri_log_item *attrip) +{ +ASSERT (atomic_read (&attrip->attri_refcount) >0 ); +if (atomic_dec_and_test (&attrip->attri_refcount)) { +xfs_trans_ail_delete (&attrip->attri_item, +SHUTDOWN_LOG_IO_ERROR ); +xfs_attri_item_free (attrip); + } +} + +STATIC void +xfs_attri_item_size ( +struct xfs_log_item *lip, +int *nvecs, +int *nbytes) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); + + *nvecs +=1 ; + *nbytes +=sizeof (struct xfs_attri_log_format ); + +/* Attr set and remove operations require a name */ +ASSERT (attrip->attri_name_len >0 ); + + *nvecs +=1 ; + *nbytes +=ATTR_NVEC_SIZE (attrip->attri_name_len); + +if (attrip->attri_value_len >0 ) { + *nvecs +=1 ; + *nbytes +=ATTR_NVEC_SIZE (attrip->attri_value_len); + } +} + +/* +* This is called to fill in the log iovecs for the given attri log +* item. We use 1 iovec for the attri_format_item, 1 for the name, and +* another for the value if it is present +*/ +STATIC void +xfs_attri_item_format ( +struct xfs_log_item *lip, +struct xfs_log_vec *lv) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); +struct xfs_log_iovec *vecp =NULL ; + + attrip->attri_format.alfi_type =XFS_LI_ATTRI ; + attrip->attri_format.alfi_size =1 ; + +/* +* This size accounting must be done before copying the attrip into the +* iovec. If we do it after, the wrong size will be recorded to the log +* and we trip across assertion checks for bad region sizes later during +* the log recovery. +*/ + +ASSERT (attrip->attri_name_len >0 ); + attrip->attri_format.alfi_size++; + +if (attrip->attri_value_len >0 ) + attrip->attri_format.alfi_size++; + +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTRI_FORMAT , + &attrip->attri_format, +sizeof (struct xfs_attri_log_format )); +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTR_NAME , + attrip->attri_name, +ATTR_NVEC_SIZE (attrip->attri_name_len)); +if (attrip->attri_value_len >0 ) +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTR_VALUE , + attrip->attri_value, +ATTR_NVEC_SIZE (attrip->attri_value_len)); +} + +/* +* The unpin operation is the last place an ATTRI is manipulated in the log. It +* is either inserted in the AIL or aborted in the event of a log I/O error. In +* either case, the ATTRI transaction has been successfully committed to make +* it this far. Therefore, we expect whoever committed the ATTRI to either +* construct and commit the ATTRD or drop the ATTRD's reference in the event of +* error. Simply drop the log's ATTRI reference now that the log is done with +* it. +*/ +STATIC void +xfs_attri_item_unpin ( +struct xfs_log_item *lip, +int remove) +{ +xfs_attri_release (ATTRI_ITEM (lip)); +} + + +STATIC void +xfs_attri_item_release ( +struct xfs_log_item *lip) +{ +xfs_attri_release (ATTRI_ITEM (lip)); +} + +/* +* Allocate and initialize an attri item. Caller may allocate an additional +* trailing buffer of the specified size +*/ +STATIC struct xfs_attri_log_item * +xfs_attri_init ( +struct xfs_mount *mp, +int buffer_size) + +{ +struct xfs_attri_log_item *attrip; +uint size; + + size =sizeof (struct xfs_attri_log_item ) + buffer_size; + attrip =kvmalloc (size,KM_ZERO ); +if (attrip ==NULL ) +return NULL ; + +xfs_log_item_init (mp, &attrip->attri_item,XFS_LI_ATTRI , + &xfs_attri_item_ops); + attrip->attri_format.alfi_id = (uintptr_t )(void *)attrip; +atomic_set (&attrip->attri_refcount,2 ); + +return attrip; +} + +/* +* Copy an attr format buffer from the given buf, and into the destination attr +* format structure. +*/ +STATIC int +xfs_attri_copy_format ( +struct xfs_log_iovec *buf, +struct xfs_attri_log_format *dst_attr_fmt) +{ +struct xfs_attri_log_format *src_attr_fmt = buf->i_addr; +uint len; + + len =sizeof (struct xfs_attri_log_format ); +if (buf->i_len != len) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW ,NULL ); +return -EFSCORRUPTED ; + } + +memcpy ((char *)dst_attr_fmt, (char *)src_attr_fmt, len); +return 0 ; +} + +static inline struct xfs_attrd_log_item *ATTRD_ITEM (struct xfs_log_item *lip) +{ + return container_of(lip, struct xfs_attrd_log_item, attrd_item); +} + +STATIC void +xfs_attrd_item_free (struct xfs_attrd_log_item *attrdp) +{ +kmem_free (attrdp->attrd_item.li_lv_shadow); +kmem_free (attrdp); +} + +STATIC void +xfs_attrd_item_size ( +struct xfs_log_item *lip, +int *nvecs, +int *nbytes) +{ + *nvecs +=1 ; + *nbytes +=sizeof (struct xfs_attrd_log_format ); +} + +/* +* This is called to fill in the log iovecs for the given attrd log item. We use +* only 1 iovec for the attrd_format, and we point that at the attr_log_format +* structure embedded in the attrd item. +*/ +STATIC void +xfs_attrd_item_format ( +struct xfs_log_item *lip, +struct xfs_log_vec *lv) +{ +struct xfs_attrd_log_item *attrdp =ATTRD_ITEM (lip); +struct xfs_log_iovec *vecp =NULL ; + + attrdp->attrd_format.alfd_type =XFS_LI_ATTRD ; + attrdp->attrd_format.alfd_size =1 ; + +xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTRD_FORMAT , + &attrdp->attrd_format, +sizeof (struct xfs_attrd_log_format )); +} + +/* +* The ATTRD is either committed or aborted if the transaction is canceled. If +* the transaction is canceled, drop our reference to the ATTRI and free the +* ATTRD. +*/ +STATIC void +xfs_attrd_item_release ( +struct xfs_log_item *lip) +{ +struct xfs_attrd_log_item *attrdp =ATTRD_ITEM (lip); + +xfs_attri_release (attrdp->attrd_attrip); +xfs_attrd_item_free (attrdp); +} + +STATIC xfs_lsn_t +xfs_attri_item_committed ( +struct xfs_log_item *lip, +xfs_lsn_t lsn) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); + +/* +* The attrip refers to xfs_attr_item memory to log the name and value +* with the intent item. This already occurred when the intent was +* committed so these fields are no longer accessed. Clear them out of +* caution since we're about to free the xfs_attr_item. +*/ + attrip->attri_name =NULL ; + attrip->attri_value =NULL ; + +/* +* The ATTRI is logged only once and cannot be moved in the log, so +* simply return the lsn at which it's been logged. +*/ +return lsn; +} + +STATIC bool +xfs_attri_item_match ( +struct xfs_log_item *lip, +uint64_t intent_id) +{ +return ATTRI_ITEM (lip)->attri_format.alfi_id == intent_id; +} + +static const struct xfs_item_ops xfs_attrd_item_ops = { + .flags =XFS_ITEM_RELEASE_WHEN_COMMITTED , + .iop_size = xfs_attrd_item_size, + .iop_format = xfs_attrd_item_format, + .iop_release = xfs_attrd_item_release, +}; + +/* Is this recovered ATTRI ok? */ +static inline bool +xfs_attri_validate ( +struct xfs_mount *mp, +struct xfs_attri_log_item *attrip) +{ +struct xfs_attri_log_format *attrp = &attrip->attri_format; +unsigned int op = attrp->alfi_op_flags & +XFS_ATTR_OP_FLAGS_TYPE_MASK ; + +/* alfi_op_flags should be either a set or remove */ +if (op !=XFS_ATTR_OP_FLAGS_SET && op !=XFS_ATTR_OP_FLAGS_REMOVE ) +return false; + +if (attrp->alfi_value_len >XATTR_SIZE_MAX ) +return false; + +if ((attrp->alfi_name_len >XATTR_NAME_MAX ) || + (attrp->alfi_name_len ==0 )) +return false; + +return xfs_verify_ino (mp, attrp->alfi_ino); +} + +static const struct xfs_item_ops xfs_attri_item_ops = { + .iop_size = xfs_attri_item_size, + .iop_format = xfs_attri_item_format, + .iop_unpin = xfs_attri_item_unpin, + .iop_committed = xfs_attri_item_committed, + .iop_release = xfs_attri_item_release, + .iop_match = xfs_attri_item_match, +}; + + + +STATIC int +xlog_recover_attri_commit_pass2 ( +struct xlog *log, +struct list_head *buffer_list, +struct xlog_recover_item *item, +xfs_lsn_t lsn) +{ +int error; +struct xfs_mount *mp = log->l_mp; +struct xfs_attri_log_item *attrip; +struct xfs_attri_log_format *attri_formatp; +char *name =NULL ; +char *value =NULL ; +int region =0 ; +int buffer_size; + + attri_formatp = item->ri_buf[region].i_addr; + +/* Validate xfs_attri_log_format */ +if (attri_formatp->__pad !=0 || attri_formatp->alfi_name_len ==0 || + (attri_formatp->alfi_op_flags ==XFS_ATTR_OP_FLAGS_REMOVE && + attri_formatp->alfi_value_len !=0 )) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW , mp); +return -EFSCORRUPTED ; + } + + buffer_size = attri_formatp->alfi_name_len + + attri_formatp->alfi_value_len; + + attrip =xfs_attri_init (mp, buffer_size); +if (attrip ==NULL ) +return -ENOMEM ; + + error =xfs_attri_copy_format (&item->ri_buf[region], + &attrip->attri_format); +if (error) { +xfs_attri_item_free (attrip); +return error; + } + + attrip->attri_name_len = attri_formatp->alfi_name_len; + attrip->attri_value_len = attri_formatp->alfi_value_len; + region++; + name = ((char *)attrip) +sizeof (struct xfs_attri_log_item ); +memcpy (name, item->ri_buf[region].i_addr, attrip->attri_name_len); + attrip->attri_name = name; + +if (attrip->attri_value_len >0 ) { + region++; + value = ((char *)attrip) +sizeof (struct xfs_attri_log_item ) + + attrip->attri_name_len; +memcpy (value, item->ri_buf[region].i_addr, + attrip->attri_value_len); + attrip->attri_value = value; + } + +/* +* The ATTRI has two references. One for the ATTRD and one for ATTRI to +* ensure it makes it into the AIL. Insert the ATTRI into the AIL +* directly and drop the ATTRI reference. Note that +* xfs_trans_ail_update() drops the AIL lock. +*/ +xfs_trans_ail_insert (log->l_ailp, &attrip->attri_item, lsn); +xfs_attri_release (attrip); +return 0 ; +} + +const struct xlog_recover_item_ops xlog_attri_item_ops = { + .item_type =XFS_LI_ATTRI , + .commit_pass2 = xlog_recover_attri_commit_pass2, +}; + +/* +* This routine is called when an ATTRD format structure is found in a committed +* transaction in the log. Its purpose is to cancel the corresponding ATTRI if +* it was still in the log. To do this it searches the AIL for the ATTRI with +* an id equal to that in the ATTRD format structure. If we find it we drop +* the ATTRD reference, which removes the ATTRI from the AIL and frees it. +*/ +STATIC int +xlog_recover_attrd_commit_pass2 ( +struct xlog *log, +struct list_head *buffer_list, +struct xlog_recover_item *item, +xfs_lsn_t lsn) +{ +struct xfs_attrd_log_format *attrd_formatp; + + attrd_formatp = item->ri_buf[0 ].i_addr; +if (item->ri_buf[0 ].i_len !=sizeof (struct xfs_attrd_log_format )) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW ,NULL ); +return -EFSCORRUPTED ; + } + +xlog_recover_release_intent (log,XFS_LI_ATTRI , + attrd_formatp->alfd_alf_id); +return 0 ; +} + +const struct xlog_recover_item_ops xlog_attrd_item_ops = { + .item_type =XFS_LI_ATTRD , + .commit_pass2 = xlog_recover_attrd_commit_pass2, +};fs/xfs/xfs_attr_item.h new file +12
-0@@ -0,0 +1,52 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later +* +* Copyright (C) 2021 Oracle. All Rights Reserved. +* Author: Allison Collins <[email protected]> +*/ +#ifndef __XFS_ATTR_ITEM_H__ +#define __XFS_ATTR_ITEM_H__ + +/* kernel only ATTRI/ATTRD definitions */ + +struct xfs_mount ; +struct kmem_zone ;
Quoted 8 lines
fs/xfs/xfs_attr_item.h line 17
Quoted 5 lines
fs/xfs/xfs_attr_item.h new file +5
-0+ +/* +* Define ATTR flag bits. Manipulated by set/clear/test_bit operators. +*/ +#define XFS_ATTRI_RECOVERED 1
XFS_ATTRI_RECOVEREDisn't used in this patch nor in any of the future patches in this series.
I think the code that used this was removed in one of the older reviews.
Will clean out.
Thanks!
Allison
Quoted 117 lines
Quoted 115 lines
fs/xfs/xfs_attr_item.h new file +35
-0+ + +/* +* This is the "attr intention" log item. It is used to log the fact that some +* attribute operations need to be processed. An operation is currently either +* a set or remove. Set or remove operations are described by the xfs_attr_item +* which may be logged to this intent. +* +* During a normal attr operation, name and value point to the name and value +* fields of the calling functions xfs_da_args. During a recovery, the name +* and value buffers are copied from the log, and stored in a trailing buffer +* attached to the xfs_attr_item until they are committed. They are freed when +* the xfs_attr_item itself is freed when the work is done. +*/ +struct xfs_attri_log_item { +struct xfs_log_item attri_item; +atomic_t attri_refcount; +int attri_name_len; +int attri_value_len; +void *attri_name; +void *attri_value; +struct xfs_attri_log_format attri_format; +}; + +/* +* This is the "attr done" log item. It is used to log the fact that some attrs +* earlier mentioned in an attri item have been freed. +*/ +struct xfs_attrd_log_item { +struct xfs_attri_log_item *attrd_attrip; +struct xfs_log_item attrd_item; +struct xfs_attrd_log_format attrd_format; +}; + +#endif /* __XFS_ATTR_ITEM_H__ */ fs/xfs/xfs_attr_list.c +1
-0@@ -15,6 +15,7 @@#include "xfs_inode.h" #include "xfs_trans.h" #include "xfs_bmap.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_attr_sf.h" #include "xfs_attr_leaf.h" fs/xfs/xfs_ioctl32.c +2
-0@@ -17,6 +17,8 @@#include "xfs_itable.h" #include "xfs_fsops.h" #include "xfs_rtalloc.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_ioctl.h" #include "xfs_ioctl32.h" fs/xfs/xfs_iops.c +2
-0@@ -13,6 +13,8 @@#include "xfs_inode.h" #include "xfs_acl.h" #include "xfs_quota.h" +#include "xfs_da_format.h" +#include "xfs_da_btree.h" #include "xfs_attr.h" #include "xfs_trans.h" #include "xfs_trace.h" fs/xfs/xfs_log.c +4
-0@@ -2157,6 +2157,10 @@ xlog_print_tic_res(REG_TYPE_STR (CUD_FORMAT ,"cud_format" ),REG_TYPE_STR (BUI_FORMAT ,"bui_format" ),REG_TYPE_STR (BUD_FORMAT ,"bud_format" ), +REG_TYPE_STR (ATTRI_FORMAT ,"attri_format" ), +REG_TYPE_STR (ATTRD_FORMAT ,"attrd_format" ), +REG_TYPE_STR (ATTR_NAME ,"attr name" ), +REG_TYPE_STR (ATTR_VALUE ,"attr value" ), };BUILD_BUG_ON (ARRAY_SIZE (res_type_str) !=XLOG_REG_TYPE_MAX +1 );#undef REG_TYPE_STRfs/xfs/xfs_log_recover.c +2
-0@@ -1800,6 +1800,8 @@ static const struct xlog_recover_item_ops *xlog_recover_item_ops[] = { &xlog_cud_item_ops, &xlog_bui_item_ops, &xlog_bud_item_ops, + &xlog_attri_item_ops, + &xlog_attrd_item_ops, }; static const struct xlog_recover_item_ops *fs/xfs/xfs_ondisk.h +2
-0@@ -132,6 +132,8 @@ xfs_check_ondisk_structs(void) XFS_CHECK_STRUCT_SIZE(struct xfs_inode_log_format, 56); XFS_CHECK_STRUCT_SIZE(struct xfs_qoff_logformat, 20); XFS_CHECK_STRUCT_SIZE(struct xfs_trans_header, 16); + XFS_CHECK_STRUCT_SIZE(struct xfs_attri_log_format, 40); + XFS_CHECK_STRUCT_SIZE(struct xfs_attrd_log_format, 16);/* * The v5 superblock format extended several v4 header structures with
Re: [PATCH v24 03/11] xfs: Set up infrastructure for log atrribute replay
In reply to Allison HendersonQuoted 7 lines · Allison Henderson
fs/xfs/xfs_attr_item.c new file +7
-0+/* +* Allocate and initialize an attri item. Caller may allocate an additional +* trailing buffer of the specified size +*/ +STATIC struct xfs_attri_log_item * +xfs_attri_init( + struct xfs_mount *mp,
fs/xfs/xfs_attr_item.c line 188
Quoted 10 lines
fs/xfs/xfs_attr_item.c new file +10
-0+ int buffer_size) + +{ +struct xfs_attri_log_item *attrip; +uint size; + + size =sizeof (struct xfs_attri_log_item ) + buffer_size; + attrip =kvmalloc (size,KM_ZERO ); +if (attrip ==NULL ) +return NULL ;
kvmalloc() takes GFP flags. I think you want GFP_KERNEL | __GFP_ZERO here.
Also, buffer size is taken directly from on-disk without bounds/length validation, meaning this could end up being an attacker controlled memory allocation, so .....
Quoted 22 lines
fs/xfs/xfs_attr_item.c new file +22
-0+STATIC int +xlog_recover_attri_commit_pass2 ( +struct xlog *log, +struct list_head *buffer_list, +struct xlog_recover_item *item, +xfs_lsn_t lsn) +{ +int error; +struct xfs_mount *mp = log->l_mp; +struct xfs_attri_log_item *attrip; +struct xfs_attri_log_format *attri_formatp; +char *name =NULL ; +char *value =NULL ; +int region =0 ; +int buffer_size; + + attri_formatp = item->ri_buf[region].i_addr; + +/* Validate xfs_attri_log_format */ +if (attri_formatp->__pad !=0 || attri_formatp->alfi_name_len ==0 || + (attri_formatp->alfi_op_flags ==XFS_ATTR_OP_FLAGS_REMOVE && + attri_formatp->alfi_value_len !=0 )) {
fs/xfs/xfs_attr_item.c line 382
Quoted 10 lines
fs/xfs/xfs_attr_item.c new file +10
-0+XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW , mp); +return -EFSCORRUPTED ; + } + + buffer_size = attri_formatp->alfi_name_len + + attri_formatp->alfi_value_len; + + attrip =xfs_attri_init (mp, buffer_size); +if (attrip ==NULL ) +return -ENOMEM ;
There needs to be a lot better validation of the attribute name/value lengths here. Also, memory allocation failure here will abort recovery, so it might be worth adding a comment here....
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH v24 03/11] xfs: Set up infrastructure for log atrribute replay
Quoted 22 lines · Dave Chinner
Quoted 7 lines · Allison Henderson
fs/xfs/xfs_attr_item.c new file +7
-0+/* +* Allocate and initialize an attri item. Caller may allocate an additional +* trailing buffer of the specified size +*/ +STATIC struct xfs_attri_log_item * +xfs_attri_init( + struct xfs_mount *mp,fs/xfs/xfs_attr_item.c line 188
Quoted 10 lines
fs/xfs/xfs_attr_item.c new file +10
-0+ int buffer_size) + +{ +struct xfs_attri_log_item *attrip; +uint size; + + size =sizeof (struct xfs_attri_log_item ) + buffer_size; + attrip =kvmalloc (size,KM_ZERO ); +if (attrip ==NULL ) +return NULL ;
kvmalloc()takes GFP flags. I think you wantGFP_KERNEL|__GFP_ZEROhere.
Quoted 4 lines
Also, buffer size is taken directly from on-disk without bounds/length validation, meaning this could end up being an attacker controlled memory allocation, so .....
Ok, will fix
Quoted 28 lines
Quoted 28 lines
fs/xfs/xfs_attr_item.c new file +28
-0+STATIC int +xlog_recover_attri_commit_pass2 ( +struct xlog *log, +struct list_head *buffer_list, +struct xlog_recover_item *item, +xfs_lsn_t lsn) +{ +int error; +struct xfs_mount *mp = log->l_mp; +struct xfs_attri_log_item *attrip; +struct xfs_attri_log_format *attri_formatp; +char *name =NULL ; +char *value =NULL ; +int region =0 ; +int buffer_size; + + attri_formatp = item->ri_buf[region].i_addr; + +/* Validate xfs_attri_log_format */ +if (attri_formatp->__pad !=0 || attri_formatp->alfi_name_len ==0 || + (attri_formatp->alfi_op_flags ==XFS_ATTR_OP_FLAGS_REMOVE && + attri_formatp->alfi_value_len !=0 )) { +XFS_ERROR_REPORT (__func__,XFS_ERRLEVEL_LOW , mp); +return -EFSCORRUPTED ; + } + + buffer_size = attri_formatp->alfi_name_len + + attri_formatp->alfi_value_len;
Quoted 8 lines
fs/xfs/xfs_attr_item.c line 382
Quoted 4 lines
fs/xfs/xfs_attr_item.c new file +4
-0+ + attrip =xfs_attri_init (mp, buffer_size); +if (attrip ==NULL ) +return -ENOMEM ;There needs to be a lot better validation of the attribute name/value lengths here. Also, memory allocation failure here will abort recovery, so it might be worth adding a comment here....
Maybe we can add a call to xfs_attri_validate here? I think we can just modify it to directly check the xfs_attri_log_format.
Thanks!
Allison
Quoted 5 lines
Cheers,
Dave.
Re: [PATCH v24 03/11] xfs: Set up infrastructure for log atrribute replay
In reply to Allison HendersonQuoted 23 lines · Allison Henderson
Currently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this.
This patch adds two new log item types for setting or removing attributes as deferred operations. The
xfs_attri_log_itemwill log an intent to set or remove an attribute. The correspondingxfs_attrd_log_itemholds a reference to thexfs_attri_log_itemand is freed once the transaction is done. Both log items use a genericxfs_attr_log_formatstructure that contains the attribute name, value, flags, inode, and anop_flagthat indicates if the operations is a set or remove.Signed-off-by Allison Henderson <[email protected]>
Quoted 9 lines
--- fs/xfs/Makefile | 1 + fs/xfs/libxfs/xfs_attr.c | 5 +- fs/xfs/libxfs/xfs_attr.h | 31 +++ fs/xfs/libxfs/xfs_defer.h | 2 + fs/xfs/libxfs/xfs_log_format.h | 44 +++- fs/xfs/libxfs/xfs_log_recover.h | 2 + fs/xfs/scrub/common.c | 2 + fs/xfs/xfs_attr_item.c | 453 ++++++++++++++++++++++++++++++++
Comment on the overall structure of this file now I've been trying to navigate through it for a little while. It is structured like:
<some attri stuff>
<some attrd stuff>
static const struct xfs_item_ops xfs_attrd_item_ops = {...} <some more attri stuff>
static const struct xfs_item_ops xfs_attri_item_ops = {...} <some attri log recovery stuff>
<some attrd log recovery stuff>
IOWs, the attri and attrd functions are interleaved non-obvious ways and that makes it hard to navigate around when trying to find related information. It would make more sense to me to structure this as:
<attri stuff>
<attri log recovery stuff>
<some attrd stuff>
<attrd log recovery stuff>
static const struct xfs_item_ops xfs_attri_item_ops = {...} const struct xlog_recover_item_ops xlog_attri_item_ops = {...} static const struct xfs_item_ops xfs_attrd_item_ops = {...} const struct xlog_recover_item_ops xlog_attrd_item_ops = {...}
because then all the related functionality is grouped together. It also puts all the ops structures together in the one place, so we don't have to jump around all over the file when just looking at what ops the items run at different times...
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH v24 03/11] xfs: Set up infrastructure for log atrribute replay
Quoted 59 lines · Dave Chinner
Quoted 23 lines · Allison Henderson
Currently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this.
This patch adds two new log item types for setting or removing attributes as deferred operations. The
xfs_attri_log_itemwill log an intent to set or remove an attribute. The correspondingxfs_attrd_log_itemholds a reference to thexfs_attri_log_itemand is freed once the transaction is done. Both log items use a genericxfs_attr_log_formatstructure that contains the attribute name, value, flags, inode, and anop_flagthat indicates if the operations is a set or remove.Signed-off-by Allison Henderson <[email protected]>Quoted 9 lines
--- fs/xfs/Makefile | 1 + fs/xfs/libxfs/xfs_attr.c | 5 +- fs/xfs/libxfs/xfs_attr.h | 31 +++ fs/xfs/libxfs/xfs_defer.h | 2 + fs/xfs/libxfs/xfs_log_format.h | 44 +++- fs/xfs/libxfs/xfs_log_recover.h | 2 + fs/xfs/scrub/common.c | 2 + fs/xfs/xfs_attr_item.c | 453 ++++++++++++++++++++++++++++++++Comment on the overall structure of this file now I've been trying to navigate through it for a little while. It is structured like:
<some attri stuff>
<some attrd stuff>
static const structxfs_item_opsxfs_attrd_item_ops= {...} <some more attri stuff>
static const structxfs_item_opsxfs_attri_item_ops= {...} <some attri log recovery stuff>
<some attrd log recovery stuff>IOWs, the attri and attrd functions are interleaved non-obvious ways and that makes it hard to navigate around when trying to find related information. It would make more sense to me to structure this as:
<attri stuff>
<attri log recovery stuff>
<some attrd stuff>
<attrd log recovery stuff>
static const structxfs_item_opsxfs_attri_item_ops= {...} const structxlog_recover_item_opsxlog_attri_item_ops= {...} static const structxfs_item_opsxfs_attrd_item_ops= {...} const structxlog_recover_item_opsxlog_attrd_item_ops= {...}
Quoted 5 lines
because then all the related functionality is grouped together. It also puts all the ops structures together in the one place, so we don't have to jump around all over the file when just looking at what ops the items run at different times...
Sure, will make a note to re-arrange some of these in the next version
Allison
Quoted 4 lines
Cheers,
Dave.
PATCH v24 4/11 xfs: Implement attr logging and replay
In reply to Allison HendersonThis patch adds the needed routines to create, log and recover logged extended attribute intents.
--- fs/xfs/libxfs/xfs_defer.c | 1 + fs/xfs/libxfs/xfs_defer.h | 1 + fs/xfs/libxfs/xfs_format.h | 10 +- fs/xfs/xfs_attr_item.c | 358 +++++++++++++++++++++++++++++++++++++ 4 files changed, 369 insertions(+), 1 deletion(-)
fs/xfs/libxfs/xfs_defer.c +1 -0
@@ -178,6 +178,7 @@ static const struct xfs_defer_op_type *defer_op_types[] = { [XFS_DEFER_OPS_TYPE_RMAP] = &xfs_rmap_update_defer_type, [XFS_DEFER_OPS_TYPE_FREE] = &xfs_extent_free_defer_type, [XFS_DEFER_OPS_TYPE_AGFL_FREE] = &xfs_agfl_free_defer_type, + [XFS_DEFER_OPS_TYPE_ATTR] = &xfs_attr_defer_type, }; static void
fs/xfs/libxfs/xfs_defer.h +1 -0
@@ -19,6 +19,7 @@ enum xfs_defer_ops_type {XFS_DEFER_OPS_TYPE_RMAP ,XFS_DEFER_OPS_TYPE_FREE ,XFS_DEFER_OPS_TYPE_AGFL_FREE , +XFS_DEFER_OPS_TYPE_ATTR , XFS_DEFER_OPS_TYPE_MAX, };
fs/xfs/libxfs/xfs_format.h +9 -1
@@ -388,7 +388,9 @@ xfs_sb_has_incompat_feature(return (sbp->sb_features_incompat & feature) !=0 ; } -#define XFS_SB_FEAT_INCOMPAT_LOG_ALL 0 +#define XFS_SB_FEAT_INCOMPAT_LOG_XATTRS (1 << 0)/* Delayed Attributes */ +#define XFS_SB_FEAT_INCOMPAT_LOG_ALL \ + (XFS_SB_FEAT_INCOMPAT_LOG_XATTRS)#define XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN ~XFS_SB_FEAT_INCOMPAT_LOG_ALLstatic inline bool xfs_sb_has_incompat_log_feature(
@@ -413,6 +415,12 @@ xfs_sb_add_incompat_log_features( sbp->sb_features_log_incompat |= features; } +static inline bool sb_version_haslogxattrs (struct xfs_sb *sbp) +{ +return XFS_SB_VERSION_NUM (sbp) ==XFS_SB_VERSION_5 && + (sbp->sb_features_log_incompat & +XFS_SB_FEAT_INCOMPAT_LOG_XATTRS ); +}static inline bool xfs_is_quota_inode (struct xfs_sb *sbp,xfs_ino_t ino)
fs/xfs/xfs_attr_item.c +358 -0
389 lines not on this page · raw message
Signature · 1 line
2.25.1
Re: [PATCH v24 04/11] xfs: Implement attr logging and replay
Quoted 3 lines · Allison Henderson
This patch adds the needed routines to create, log and recover logged extended attribute intents.
Apart from the two nits mentioned below, the remaining changes seem to correctly handle attr set/remove operations regardless of whether delayed logging is enabled or not.
Quoted 48 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 1 + fs/xfs/libxfs/xfs_defer.h | 1 + fs/xfs/libxfs/xfs_format.h | 10 +- fs/xfs/xfs_attr_item.c | 358 +++++++++++++++++++++++++++++++++++++ 4 files changed, 369 insertions(+), 1 deletion(-)fs/xfs/libxfs/xfs_defer.c +1
-0@@ -178,6 +178,7 @@ static const struct xfs_defer_op_type *defer_op_types[] = { [XFS_DEFER_OPS_TYPE_RMAP] = &xfs_rmap_update_defer_type, [XFS_DEFER_OPS_TYPE_FREE] = &xfs_extent_free_defer_type, [XFS_DEFER_OPS_TYPE_AGFL_FREE] = &xfs_agfl_free_defer_type, + [XFS_DEFER_OPS_TYPE_ATTR] = &xfs_attr_defer_type, }; static voidfs/xfs/libxfs/xfs_defer.h +1
-0@@ -19,6 +19,7 @@ enum xfs_defer_ops_type {XFS_DEFER_OPS_TYPE_RMAP ,XFS_DEFER_OPS_TYPE_FREE ,XFS_DEFER_OPS_TYPE_AGFL_FREE , +XFS_DEFER_OPS_TYPE_ATTR , XFS_DEFER_OPS_TYPE_MAX, };fs/xfs/libxfs/xfs_format.h +3
-1@@ -388,7 +388,9 @@ xfs_sb_has_incompat_feature(return (sbp->sb_features_incompat & feature) !=0 ; } -#define XFS_SB_FEAT_INCOMPAT_LOG_ALL 0 +#define XFS_SB_FEAT_INCOMPAT_LOG_XATTRS (1 << 0)/* Delayed Attributes */ +#define XFS_SB_FEAT_INCOMPAT_LOG_ALL \ + (XFS_SB_FEAT_INCOMPAT_LOG_XATTRS)#define XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN ~XFS_SB_FEAT_INCOMPAT_LOG_ALLstatic inline bool xfs_sb_has_incompat_log_feature(
fs/xfs/libxfs/xfs_format.h xfs_sb_add_incompat_log_features( · line 417
Quoted 3 lines
fs/xfs/libxfs/xfs_format.h +0
-0@@ -413,6 +415,12 @@ xfs_sb_add_incompat_log_features( sbp->sb_features_log_incompat |= features; }
To be consistent with the existing convention, may be the function below should be renamed as xfs_sb_version_haslogxattrs()?
fs/xfs/libxfs/xfs_format.h xfs_sb_add_incompat_log_features( · line 420
Quoted 3 lines
fs/xfs/libxfs/xfs_format.h +3
-0+static inline bool sb_version_haslogxattrs (struct xfs_sb *sbp) +{ + return XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
The above comparison can be replaced with a call to xfs_sb_is_v5().
Quoted 401 lines
fs/xfs/libxfs/xfs_format.h +3
-0+ (sbp->sb_features_log_incompat & +XFS_SB_FEAT_INCOMPAT_LOG_XATTRS ); +}static inline bool xfs_is_quota_inode (struct xfs_sb *sbp,xfs_ino_t ino)fs/xfs/xfs_attr_item.c +358
-0@@ -275,6 +275,163 @@ xfs_attrd_item_release(xfs_attrd_item_free (attrdp); } +/* +* Performs one step of an attribute update intent and marks the attrd item +* dirty.. An attr operation may be a set or a remove. Note that the +* transaction is marked dirty regardless of whether the operation succeeds or +* fails to support the ATTRI/ATTRD lifecycle rules. +*/ +STATIC int +xfs_trans_attr_finish_update ( +struct xfs_delattr_context *dac, +struct xfs_attrd_log_item *attrdp, +struct xfs_buf **leaf_bp, +uint32_t op_flags) +{ +struct xfs_da_args *args = dac->da_args; +unsigned int op = op_flags & +XFS_ATTR_OP_FLAGS_TYPE_MASK ; +int error; + +switch (op) { +case XFS_ATTR_OP_FLAGS_SET : + error =xfs_attr_set_iter (dac, leaf_bp); +break ; +case XFS_ATTR_OP_FLAGS_REMOVE : +ASSERT (XFS_IFORK_Q (args->dp)); + error =xfs_attr_remove_iter (dac); +break ; +default : + error = -EFSCORRUPTED ; +break ; + } + +/* +* Mark the transaction dirty, even on error. This ensures the +* transaction is aborted, which: +* +* 1.) releases the ATTRI and frees the ATTRD +* 2.) shuts down the filesystem +*/ + args->trans->t_flags |=XFS_TRANS_DIRTY ; + +/* +* attr intent/done items are null when delayed attributes are disabled +*/ +if (attrdp) +set_bit (XFS_LI_DIRTY , &attrdp->attrd_item.li_flags); + +return error; +} + +/* Log an attr to the intent item. */ +STATIC void +xfs_attr_log_item ( +struct xfs_trans *tp, +struct xfs_attri_log_item *attrip, +struct xfs_attr_item *attr) +{ +struct xfs_attri_log_format *attrp; + + tp->t_flags |=XFS_TRANS_DIRTY ; +set_bit (XFS_LI_DIRTY , &attrip->attri_item.li_flags); + +/* +* At this point the xfs_attr_item has been constructed, and we've +* created the log intent. Fill in the attri log item and log format +* structure with fields from this xfs_attr_item +*/ + attrp = &attrip->attri_format; + attrp->alfi_ino = attr->xattri_dac.da_args->dp->i_ino; + attrp->alfi_op_flags = attr->xattri_op_flags; + attrp->alfi_value_len = attr->xattri_dac.da_args->valuelen; + attrp->alfi_name_len = attr->xattri_dac.da_args->namelen; + attrp->alfi_attr_flags = attr->xattri_dac.da_args->attr_filter; + + attrip->attri_name = (void *)attr->xattri_dac.da_args->name; + attrip->attri_value = attr->xattri_dac.da_args->value; + attrip->attri_name_len = attr->xattri_dac.da_args->namelen; + attrip->attri_value_len = attr->xattri_dac.da_args->valuelen; +} + +/* Get an ATTRI. */ +static struct xfs_log_item * +xfs_attr_create_intent ( +struct xfs_trans *tp, +struct list_head *items, +unsigned int count, +bool sort) +{ +struct xfs_mount *mp = tp->t_mountp; +struct xfs_attri_log_item *attrip; +struct xfs_attr_item *attr; + +ASSERT (count ==1 ); + +if (!sb_version_haslogxattrs (&mp->m_sb)) +return NULL ; + + attrip =xfs_attri_init (mp,0 ); +if (attrip ==NULL ) +return NULL ; + +xfs_trans_add_item (tp, &attrip->attri_item); +list_for_each_entry (attr, items, xattri_list) +xfs_attr_log_item (tp, attrip, attr); +return &attrip->attri_item; +} + +/* Process an attr. */ +STATIC int +xfs_attr_finish_item ( +struct xfs_trans *tp, +struct xfs_log_item *done, +struct list_head *item, +struct xfs_btree_cur **state) +{ +struct xfs_attr_item *attr; +struct xfs_attrd_log_item *done_item =NULL ; +int error; +struct xfs_delattr_context *dac; + + attr = container_of(item, struct xfs_attr_item, xattri_list); + dac = &attr->xattri_dac; +if (done) + done_item =ATTRD_ITEM (done); + +/* +* Always reset trans after EAGAIN cycle +* since the transaction is new +*/ + dac->da_args->trans = tp; + + error =xfs_trans_attr_finish_update (dac, done_item, &dac->leaf_bp, + attr->xattri_op_flags); +if (error != -EAGAIN ) +kmem_free (attr); + +return error; +} + +/* Abort all pending ATTRs. */ +STATIC void +xfs_attr_abort_intent ( +struct xfs_log_item *intent) +{ +xfs_attri_release (ATTRI_ITEM (intent)); +} + +/* Cancel an attr */ +STATIC void +xfs_attr_cancel_item ( +struct list_head *item) +{ +struct xfs_attr_item *attr; + + attr = container_of(item, struct xfs_attr_item, xattri_list); +kmem_free (attr); +} + STATIC xfs_lsn_t xfs_attri_item_committed( struct xfs_log_item *lip,@@ -306,6 +463,30 @@ xfs_attri_item_match(return ATTRI_ITEM (lip)->attri_format.alfi_id == intent_id; } +/* +* This routine is called to allocate an "attr free done" log item. +*/ +static struct xfs_attrd_log_item * +xfs_trans_get_attrd (struct xfs_trans *tp, +struct xfs_attri_log_item *attrip) +{ +struct xfs_attrd_log_item *attrdp; +uint size; + +ASSERT (tp !=NULL ); + + size =sizeof (struct xfs_attrd_log_item ); + attrdp =kmem_zalloc (size,0 ); + +xfs_log_item_init (tp->t_mountp, &attrdp->attrd_item,XFS_LI_ATTRD , + &xfs_attrd_item_ops); + attrdp->attrd_attrip = attrip; + attrdp->attrd_format.alfd_alf_id = attrip->attri_format.alfi_id; + +xfs_trans_add_item (tp, &attrdp->attrd_item); +return attrdp; +} +static const struct xfs_item_ops xfs_attrd_item_ops = { .flags =XFS_ITEM_RELEASE_WHEN_COMMITTED , .iop_size = xfs_attrd_item_size,@@ -313,6 +494,29 @@ static const struct xfs_item_ops xfs_attrd_item_ops = { .iop_release = xfs_attrd_item_release, }; + +/* Get an ATTRD so we can process all the attrs. */ +static struct xfs_log_item * +xfs_attr_create_done ( +struct xfs_trans *tp, +struct xfs_log_item *intent, +unsigned int count) +{ +if (!intent) +return NULL ; + +return &xfs_trans_get_attrd (tp,ATTRI_ITEM (intent))->attrd_item; +} + +const struct xfs_defer_op_type xfs_attr_defer_type = { + .max_items =1 , + .create_intent = xfs_attr_create_intent, + .abort_intent = xfs_attr_abort_intent, + .create_done = xfs_attr_create_done, + .finish_item = xfs_attr_finish_item, + .cancel_item = xfs_attr_cancel_item, +}; +/* Is this recovered ATTRI ok? */ static inline bool xfs_attri_validate(@@ -337,13 +541,167 @@ xfs_attri_validate(return xfs_verify_ino (mp, attrp->alfi_ino); } +/* +* Process an attr intent item that was recovered from the log. We need to +* delete the attr that it describes. +*/ +STATIC int +xfs_attri_item_recover ( +struct xfs_log_item *lip, +struct list_head *capture_list) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); +struct xfs_attr_item *attr; +struct xfs_mount *mp = lip->li_mountp; +struct xfs_inode *ip; +struct xfs_da_args *args; +struct xfs_trans *tp; +struct xfs_trans_res tres; +struct xfs_attri_log_format *attrp; +int error, ret =0 ; +int total; +int local; +struct xfs_attrd_log_item *done_item =NULL ; + +/* +* First check the validity of the attr described by the ATTRI. If any +* are bad, then assume that all are bad and just toss the ATTRI. +*/ + attrp = &attrip->attri_format; +if (!xfs_attri_validate (mp, attrip)) +return -EFSCORRUPTED ; + + error =xlog_recover_iget (mp, attrp->alfi_ino, &ip); +if (error) +return error; + + attr =kmem_zalloc (sizeof (struct xfs_attr_item ) + +sizeof (struct xfs_da_args ),KM_NOFS ); + args = (struct xfs_da_args *)(attr +1 ); + + attr->xattri_dac.da_args = args; + attr->xattri_op_flags = attrp->alfi_op_flags; + + args->dp = ip; + args->geo = mp->m_attr_geo; + args->op_flags = attrp->alfi_op_flags; + args->whichfork =XFS_ATTR_FORK ; + args->name = attrip->attri_name; + args->namelen = attrp->alfi_name_len; + args->hashval =xfs_da_hashname (args->name, args->namelen); + args->attr_filter = attrp->alfi_attr_flags; + +if (attrp->alfi_op_flags ==XFS_ATTR_OP_FLAGS_SET ) { + args->value = attrip->attri_value; + args->valuelen = attrp->alfi_value_len; + args->total =xfs_attr_calc_size (args, &local); + + tres.tr_logres =M_RES (mp)->tr_attrsetm.tr_logres + +M_RES (mp)->tr_attrsetrt.tr_logres * + args->total; + tres.tr_logcount =XFS_ATTRSET_LOG_COUNT ; + tres.tr_logflags =XFS_TRANS_PERM_LOG_RES ; + total = args->total; + }else { + tres =M_RES (mp)->tr_attrrm; + total =XFS_ATTRRM_SPACE_RES (mp); + } + error =xfs_trans_alloc (mp, &tres, total,0 ,XFS_TRANS_RESERVE , &tp); +if (error) + goto out; + + args->trans = tp; + done_item =xfs_trans_get_attrd (tp, attrip); + +xfs_ilock (ip,XFS_ILOCK_EXCL ); +xfs_trans_ijoin (tp, ip,0 ); + + ret =xfs_trans_attr_finish_update (&attr->xattri_dac, done_item, + &attr->xattri_dac.leaf_bp, + attrp->alfi_op_flags); +if (ret == -EAGAIN ) { +/* There's more work to do, so add it to this transaction */ +xfs_defer_add (tp,XFS_DEFER_OPS_TYPE_ATTR , &attr->xattri_list); + }else + error = ret; + +if (error) { +xfs_trans_cancel (tp); + goto out_unlock; + } + + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list, + attr->xattri_dac.leaf_bp); + +out_unlock: +xfs_iunlock (ip,XFS_ILOCK_EXCL ); +xfs_irele (ip); +out: +if (ret != -EAGAIN ) +kmem_free (attr); +return error; +} + +/* Re-log an intent item to push the log tail forward. */ +static struct xfs_log_item * +xfs_attri_item_relog ( +struct xfs_log_item *intent, +struct xfs_trans *tp) +{ +struct xfs_attrd_log_item *attrdp; +struct xfs_attri_log_item *old_attrip; +struct xfs_attri_log_item *new_attrip; +struct xfs_attri_log_format *new_attrp; +struct xfs_attri_log_format *old_attrp; +int buffer_size; + + old_attrip =ATTRI_ITEM (intent); + old_attrp = &old_attrip->attri_format; + buffer_size = old_attrp->alfi_value_len + old_attrp->alfi_name_len; + + tp->t_flags |=XFS_TRANS_DIRTY ; + attrdp =xfs_trans_get_attrd (tp, old_attrip); +set_bit (XFS_LI_DIRTY , &attrdp->attrd_item.li_flags); + + new_attrip =xfs_attri_init (tp->t_mountp, buffer_size); + new_attrp = &new_attrip->attri_format; + + new_attrp->alfi_ino = old_attrp->alfi_ino; + new_attrp->alfi_op_flags = old_attrp->alfi_op_flags; + new_attrp->alfi_value_len = old_attrp->alfi_value_len; + new_attrp->alfi_name_len = old_attrp->alfi_name_len; + new_attrp->alfi_attr_flags = old_attrp->alfi_attr_flags; + + new_attrip->attri_name_len = old_attrip->attri_name_len; + new_attrip->attri_name = ((char *)new_attrip) + +sizeof (struct xfs_attri_log_item ); +memcpy (new_attrip->attri_name, old_attrip->attri_name, + new_attrip->attri_name_len); + + new_attrip->attri_value_len = old_attrip->attri_value_len; +if (new_attrip->attri_value_len >0 ) { + new_attrip->attri_value = new_attrip->attri_name + + new_attrip->attri_name_len; + +memcpy (new_attrip->attri_value, old_attrip->attri_value, + new_attrip->attri_value_len); + } + +xfs_trans_add_item (tp, &new_attrip->attri_item); +set_bit (XFS_LI_DIRTY , &new_attrip->attri_item.li_flags); + +return &new_attrip->attri_item; +} +static const struct xfs_item_ops xfs_attri_item_ops = { .iop_size = xfs_attri_item_size, .iop_format = xfs_attri_item_format, .iop_unpin = xfs_attri_item_unpin, .iop_committed = xfs_attri_item_committed, .iop_release = xfs_attri_item_release, + .iop_recover = xfs_attri_item_recover, .iop_match = xfs_attri_item_match, + .iop_relog = xfs_attri_item_relog, };
Signature · 1 line
chandan
Re: [PATCH v24 04/11] xfs: Implement attr logging and replay
Quoted 10 lines · Chandan Babu R
Quoted 3 lines · Allison Henderson
This patch adds the needed routines to create, log and recover logged extended attribute intents.
Apart from the two nits mentioned below, the remaining changes seem to correctly handle attr set/remove operations regardless of whether delayed logging is enabled or not.
Reviewed-by Chandan Babu R <[email protected]>
Ok, thank you!
Quoted 41 lines
Quoted 41 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_defer.c | 1 + fs/xfs/libxfs/xfs_defer.h | 1 + fs/xfs/libxfs/xfs_format.h | 10 +- fs/xfs/xfs_attr_item.c | 358 +++++++++++++++++++++++++++++++++++++ 4 files changed, 369 insertions(+), 1 deletion(-)fs/xfs/libxfs/xfs_defer.c +1
-0@@ -178,6 +178,7 @@ static const struct xfs_defer_op_type *defer_op_types[] = { [XFS_DEFER_OPS_TYPE_RMAP] = &xfs_rmap_update_defer_type, [XFS_DEFER_OPS_TYPE_FREE] = &xfs_extent_free_defer_type, [XFS_DEFER_OPS_TYPE_AGFL_FREE] = &xfs_agfl_free_defer_type, + [XFS_DEFER_OPS_TYPE_ATTR] = &xfs_attr_defer_type, }; static voidfs/xfs/libxfs/xfs_defer.h +1
-0@@ -19,6 +19,7 @@ enum xfs_defer_ops_type {XFS_DEFER_OPS_TYPE_RMAP ,XFS_DEFER_OPS_TYPE_FREE ,XFS_DEFER_OPS_TYPE_AGFL_FREE , +XFS_DEFER_OPS_TYPE_ATTR , XFS_DEFER_OPS_TYPE_MAX, };fs/xfs/libxfs/xfs_format.h +0
-0@@ -388,7 +388,9 @@ xfs_sb_has_incompat_feature(return (sbp->sb_features_incompat & feature) !=0 ; }
Quoted 14 lines
fs/xfs/libxfs/xfs_format.h xfs_sb_add_incompat_log_features( · line 417
Quoted 11 lines
fs/xfs/libxfs/xfs_format.h +3
-1-#define XFS_SB_FEAT_INCOMPAT_LOG_ALL 0 +#define XFS_SB_FEAT_INCOMPAT_LOG_XATTRS (1 << 0)/* Delayed Attributes */ +#define XFS_SB_FEAT_INCOMPAT_LOG_ALL \ + (XFS_SB_FEAT_INCOMPAT_LOG_XATTRS)#define XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN ~XFS_SB_FEAT_INCOMPAT_LOG_ALLstatic inline bool xfs_sb_has_incompat_log_feature(@@ -413,6 +415,12 @@ xfs_sb_add_incompat_log_features( sbp->sb_features_log_incompat |= features; }To be consistent with the existing convention, may be the function below should be renamed as
xfs_sb_version_haslogxattrs()?
Sure, will add xfs_ prefix
Quoted 6 lines
fs/xfs/libxfs/xfs_format.h xfs_sb_add_incompat_log_features( · line 420
Quoted 4 lines
fs/xfs/libxfs/xfs_format.h +3
-0+static inline bool sb_version_haslogxattrs (struct xfs_sb *sbp) +{ + return XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&The above comparison can be replaced with a call to
xfs_sb_is_v5().
Ok, will update. Thanks!
Allison
Quoted 404 lines
Quoted 402 lines
fs/xfs/libxfs/xfs_format.h +3
-0+ (sbp->sb_features_log_incompat & +XFS_SB_FEAT_INCOMPAT_LOG_XATTRS ); +}static inline bool xfs_is_quota_inode (struct xfs_sb *sbp,xfs_ino_t ino)fs/xfs/xfs_attr_item.c +358
-0@@ -275,6 +275,163 @@ xfs_attrd_item_release(xfs_attrd_item_free (attrdp); } +/* +* Performs one step of an attribute update intent and marks the attrd item +* dirty.. An attr operation may be a set or a remove. Note that the +* transaction is marked dirty regardless of whether the operation succeeds or +* fails to support the ATTRI/ATTRD lifecycle rules. +*/ +STATIC int +xfs_trans_attr_finish_update ( +struct xfs_delattr_context *dac, +struct xfs_attrd_log_item *attrdp, +struct xfs_buf **leaf_bp, +uint32_t op_flags) +{ +struct xfs_da_args *args = dac->da_args; +unsigned int op = op_flags & +XFS_ATTR_OP_FLAGS_TYPE_MASK ; +int error; + +switch (op) { +case XFS_ATTR_OP_FLAGS_SET : + error =xfs_attr_set_iter (dac, leaf_bp); +break ; +case XFS_ATTR_OP_FLAGS_REMOVE : +ASSERT (XFS_IFORK_Q (args->dp)); + error =xfs_attr_remove_iter (dac); +break ; +default : + error = -EFSCORRUPTED ; +break ; + } + +/* +* Mark the transaction dirty, even on error. This ensures the +* transaction is aborted, which: +* +* 1.) releases the ATTRI and frees the ATTRD +* 2.) shuts down the filesystem +*/ + args->trans->t_flags |=XFS_TRANS_DIRTY ; + +/* +* attr intent/done items are null when delayed attributes are disabled +*/ +if (attrdp) +set_bit (XFS_LI_DIRTY , &attrdp->attrd_item.li_flags); + +return error; +} + +/* Log an attr to the intent item. */ +STATIC void +xfs_attr_log_item ( +struct xfs_trans *tp, +struct xfs_attri_log_item *attrip, +struct xfs_attr_item *attr) +{ +struct xfs_attri_log_format *attrp; + + tp->t_flags |=XFS_TRANS_DIRTY ; +set_bit (XFS_LI_DIRTY , &attrip->attri_item.li_flags); + +/* +* At this point the xfs_attr_item has been constructed, and we've +* created the log intent. Fill in the attri log item and log format +* structure with fields from this xfs_attr_item +*/ + attrp = &attrip->attri_format; + attrp->alfi_ino = attr->xattri_dac.da_args->dp->i_ino; + attrp->alfi_op_flags = attr->xattri_op_flags; + attrp->alfi_value_len = attr->xattri_dac.da_args->valuelen; + attrp->alfi_name_len = attr->xattri_dac.da_args->namelen; + attrp->alfi_attr_flags = attr->xattri_dac.da_args->attr_filter; + + attrip->attri_name = (void *)attr->xattri_dac.da_args->name; + attrip->attri_value = attr->xattri_dac.da_args->value; + attrip->attri_name_len = attr->xattri_dac.da_args->namelen; + attrip->attri_value_len = attr->xattri_dac.da_args->valuelen; +} + +/* Get an ATTRI. */ +static struct xfs_log_item * +xfs_attr_create_intent ( +struct xfs_trans *tp, +struct list_head *items, +unsigned int count, +bool sort) +{ +struct xfs_mount *mp = tp->t_mountp; +struct xfs_attri_log_item *attrip; +struct xfs_attr_item *attr; + +ASSERT (count ==1 ); + +if (!sb_version_haslogxattrs (&mp->m_sb)) +return NULL ; + + attrip =xfs_attri_init (mp,0 ); +if (attrip ==NULL ) +return NULL ; + +xfs_trans_add_item (tp, &attrip->attri_item); +list_for_each_entry (attr, items, xattri_list) +xfs_attr_log_item (tp, attrip, attr); +return &attrip->attri_item; +} + +/* Process an attr. */ +STATIC int +xfs_attr_finish_item ( +struct xfs_trans *tp, +struct xfs_log_item *done, +struct list_head *item, +struct xfs_btree_cur **state) +{ +struct xfs_attr_item *attr; +struct xfs_attrd_log_item *done_item =NULL ; +int error; +struct xfs_delattr_context *dac; + + attr = container_of(item, struct xfs_attr_item, xattri_list); + dac = &attr->xattri_dac; +if (done) + done_item =ATTRD_ITEM (done); + +/* +* Always reset trans after EAGAIN cycle +* since the transaction is new +*/ + dac->da_args->trans = tp; + + error =xfs_trans_attr_finish_update (dac, done_item, &dac->leaf_bp, + attr->xattri_op_flags); +if (error != -EAGAIN ) +kmem_free (attr); + +return error; +} + +/* Abort all pending ATTRs. */ +STATIC void +xfs_attr_abort_intent ( +struct xfs_log_item *intent) +{ +xfs_attri_release (ATTRI_ITEM (intent)); +} + +/* Cancel an attr */ +STATIC void +xfs_attr_cancel_item ( +struct list_head *item) +{ +struct xfs_attr_item *attr; + + attr = container_of(item, struct xfs_attr_item, xattri_list); +kmem_free (attr); +} + STATIC xfs_lsn_t xfs_attri_item_committed( struct xfs_log_item *lip,@@ -306,6 +463,30 @@ xfs_attri_item_match(return ATTRI_ITEM (lip)->attri_format.alfi_id == intent_id; } +/* +* This routine is called to allocate an "attr free done" log item. +*/ +static struct xfs_attrd_log_item * +xfs_trans_get_attrd (struct xfs_trans *tp, +struct xfs_attri_log_item *attrip) +{ +struct xfs_attrd_log_item *attrdp; +uint size; + +ASSERT (tp !=NULL ); + + size =sizeof (struct xfs_attrd_log_item ); + attrdp =kmem_zalloc (size,0 ); + +xfs_log_item_init (tp->t_mountp, &attrdp->attrd_item,XFS_LI_ATTRD , + &xfs_attrd_item_ops); + attrdp->attrd_attrip = attrip; + attrdp->attrd_format.alfd_alf_id = attrip->attri_format.alfi_id; + +xfs_trans_add_item (tp, &attrdp->attrd_item); +return attrdp; +} +static const struct xfs_item_ops xfs_attrd_item_ops = { .flags =XFS_ITEM_RELEASE_WHEN_COMMITTED , .iop_size = xfs_attrd_item_size,@@ -313,6 +494,29 @@ static const struct xfs_item_ops xfs_attrd_item_ops = { .iop_release = xfs_attrd_item_release, }; + +/* Get an ATTRD so we can process all the attrs. */ +static struct xfs_log_item * +xfs_attr_create_done ( +struct xfs_trans *tp, +struct xfs_log_item *intent, +unsigned int count) +{ +if (!intent) +return NULL ; + +return &xfs_trans_get_attrd (tp,ATTRI_ITEM (intent))->attrd_item; +} + +const struct xfs_defer_op_type xfs_attr_defer_type = { + .max_items =1 , + .create_intent = xfs_attr_create_intent, + .abort_intent = xfs_attr_abort_intent, + .create_done = xfs_attr_create_done, + .finish_item = xfs_attr_finish_item, + .cancel_item = xfs_attr_cancel_item, +}; +/* Is this recovered ATTRI ok? */ static inline bool xfs_attri_validate(@@ -337,13 +541,167 @@ xfs_attri_validate(return xfs_verify_ino (mp, attrp->alfi_ino); } +/* +* Process an attr intent item that was recovered from the log. We need to +* delete the attr that it describes. +*/ +STATIC int +xfs_attri_item_recover ( +struct xfs_log_item *lip, +struct list_head *capture_list) +{ +struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); +struct xfs_attr_item *attr; +struct xfs_mount *mp = lip->li_mountp; +struct xfs_inode *ip; +struct xfs_da_args *args; +struct xfs_trans *tp; +struct xfs_trans_res tres; +struct xfs_attri_log_format *attrp; +int error, ret =0 ; +int total; +int local; +struct xfs_attrd_log_item *done_item =NULL ; + +/* +* First check the validity of the attr described by the ATTRI. If any +* are bad, then assume that all are bad and just toss the ATTRI. +*/ + attrp = &attrip->attri_format; +if (!xfs_attri_validate (mp, attrip)) +return -EFSCORRUPTED ; + + error =xlog_recover_iget (mp, attrp->alfi_ino, &ip); +if (error) +return error; + + attr =kmem_zalloc (sizeof (struct xfs_attr_item ) + +sizeof (struct xfs_da_args ),KM_NOFS ); + args = (struct xfs_da_args *)(attr +1 ); + + attr->xattri_dac.da_args = args; + attr->xattri_op_flags = attrp->alfi_op_flags; + + args->dp = ip; + args->geo = mp->m_attr_geo; + args->op_flags = attrp->alfi_op_flags; + args->whichfork =XFS_ATTR_FORK ; + args->name = attrip->attri_name; + args->namelen = attrp->alfi_name_len; + args->hashval =xfs_da_hashname (args->name, args->namelen); + args->attr_filter = attrp->alfi_attr_flags; + +if (attrp->alfi_op_flags ==XFS_ATTR_OP_FLAGS_SET ) { + args->value = attrip->attri_value; + args->valuelen = attrp->alfi_value_len; + args->total =xfs_attr_calc_size (args, &local); + + tres.tr_logres =M_RES (mp)->tr_attrsetm.tr_logres + +M_RES (mp)->tr_attrsetrt.tr_logres * + args->total; + tres.tr_logcount =XFS_ATTRSET_LOG_COUNT ; + tres.tr_logflags =XFS_TRANS_PERM_LOG_RES ; + total = args->total; + }else { + tres =M_RES (mp)->tr_attrrm; + total =XFS_ATTRRM_SPACE_RES (mp); + } + error =xfs_trans_alloc (mp, &tres, total,0 ,XFS_TRANS_RESERVE , &tp); +if (error) + goto out; + + args->trans = tp; + done_item =xfs_trans_get_attrd (tp, attrip); + +xfs_ilock (ip,XFS_ILOCK_EXCL ); +xfs_trans_ijoin (tp, ip,0 ); + + ret =xfs_trans_attr_finish_update (&attr->xattri_dac, done_item, + &attr->xattri_dac.leaf_bp, + attrp->alfi_op_flags); +if (ret == -EAGAIN ) { +/* There's more work to do, so add it to this transaction */ +xfs_defer_add (tp,XFS_DEFER_OPS_TYPE_ATTR , &attr->xattri_list); + }else + error = ret; + +if (error) { +xfs_trans_cancel (tp); + goto out_unlock; + } + + error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list, + attr->xattri_dac.leaf_bp); + +out_unlock: +xfs_iunlock (ip,XFS_ILOCK_EXCL ); +xfs_irele (ip); +out: +if (ret != -EAGAIN ) +kmem_free (attr); +return error; +} + +/* Re-log an intent item to push the log tail forward. */ +static struct xfs_log_item * +xfs_attri_item_relog ( +struct xfs_log_item *intent, +struct xfs_trans *tp) +{ +struct xfs_attrd_log_item *attrdp; +struct xfs_attri_log_item *old_attrip; +struct xfs_attri_log_item *new_attrip; +struct xfs_attri_log_format *new_attrp; +struct xfs_attri_log_format *old_attrp; +int buffer_size; + + old_attrip =ATTRI_ITEM (intent); + old_attrp = &old_attrip->attri_format; + buffer_size = old_attrp->alfi_value_len + old_attrp->alfi_name_len; + + tp->t_flags |=XFS_TRANS_DIRTY ; + attrdp =xfs_trans_get_attrd (tp, old_attrip); +set_bit (XFS_LI_DIRTY , &attrdp->attrd_item.li_flags); + + new_attrip =xfs_attri_init (tp->t_mountp, buffer_size); + new_attrp = &new_attrip->attri_format; + + new_attrp->alfi_ino = old_attrp->alfi_ino; + new_attrp->alfi_op_flags = old_attrp->alfi_op_flags; + new_attrp->alfi_value_len = old_attrp->alfi_value_len; + new_attrp->alfi_name_len = old_attrp->alfi_name_len; + new_attrp->alfi_attr_flags = old_attrp->alfi_attr_flags; + + new_attrip->attri_name_len = old_attrip->attri_name_len; + new_attrip->attri_name = ((char *)new_attrip) + +sizeof (struct xfs_attri_log_item ); +memcpy (new_attrip->attri_name, old_attrip->attri_name, + new_attrip->attri_name_len); + + new_attrip->attri_value_len = old_attrip->attri_value_len; +if (new_attrip->attri_value_len >0 ) { + new_attrip->attri_value = new_attrip->attri_name + + new_attrip->attri_name_len; + +memcpy (new_attrip->attri_value, old_attrip->attri_value, + new_attrip->attri_value_len); + } + +xfs_trans_add_item (tp, &new_attrip->attri_item); +set_bit (XFS_LI_DIRTY , &new_attrip->attri_item.li_flags); + +return &new_attrip->attri_item; +} +static const struct xfs_item_ops xfs_attri_item_ops = { .iop_size = xfs_attri_item_size, .iop_format = xfs_attri_item_format, .iop_unpin = xfs_attri_item_unpin, .iop_committed = xfs_attri_item_committed, .iop_release = xfs_attri_item_release, + .iop_recover = xfs_attri_item_recover, .iop_match = xfs_attri_item_match, + .iop_relog = xfs_attri_item_relog, };
PATCH v24 5/11 RFC xfs: Skip flip flags for delayed attrs
In reply to Allison HendersonThis is a clean up patch that skips the flip flag logic for delayed attr renames. Since the log replay keeps the inode locked, we do not need to worry about race windows with attr lookups. So we can skip over flipping the flag and the extra transaction roll for it
RFC: In the last review, folks asked for some performance analysis, so I did a few perf captures with and with out this patch. What I found was that there wasnt very much difference at all between having the patch or not having it. Of the time we do spend in the affected code, the percentage is small. Most of the time we spend about %0.03 of the time in this function, with or with out the patch. Occasionally we get a 0.02%, though not often. So I think this starts to challenge needing this patch at all. This patch was requested some number of reviews ago, be perhaps in light of the findings, it may no longer be of interest.
0.03% 0.00% fsstress [xfs] [k] xfs_attr_set_iter
Keep it or drop it?
--- fs/xfs/libxfs/xfs_attr.c | 54 +++++++++++++++++++++-------------- fs/xfs/libxfs/xfs_attr_leaf.c | 3 +- 2 files changed, 35 insertions(+), 22 deletions(-)
fs/xfs/libxfs/xfs_attr.c +33 -21
@@ -355,6 +355,7 @@ xfs_attr_set_iter(struct xfs_inode *dp = args->dp;struct xfs_buf *bp =NULL ;int forkoff, error =0 ; +struct xfs_mount *mp = args->dp->i_mount;/* State machine switch */ switch (dac->dela_state) {
@@ -477,16 +478,21 @@ xfs_attr_set_iter(* In a separate transaction, set the incomplete flag on the * "old" attr and clear the incomplete flag on the "new" attr. */ - error =xfs_attr3_leaf_flipflags (args); -if (error) -return error; -/* -* Commit the flag value change and start the next trans in -* series. -*/ - dac->dela_state =XFS_DAS_FLIP_LFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); -return -EAGAIN ; +if (!xfs_has_larp (mp)) { + error =xfs_attr3_leaf_flipflags (args); +if (error) +return error; +/* +* Commit the flag value change and start the next trans +* in series. +*/ + dac->dela_state =XFS_DAS_FLIP_LFLAG ; +trace_xfs_attr_set_iter_return (dac->dela_state, + args->dp); +return -EAGAIN ; + } + +/* fallthrough */ case XFS_DAS_FLIP_LFLAG :/* * Dismantle the "old" attribute/value pair by removing a
@@ -589,17 +595,21 @@ xfs_attr_set_iter(* In a separate transaction, set the incomplete flag on the * "old" attr and clear the incomplete flag on the "new" attr. */ - error =xfs_attr3_leaf_flipflags (args); -if (error) - goto out; -/* -* Commit the flag value change and start the next trans in -* series -*/ - dac->dela_state =XFS_DAS_FLIP_NFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); -return -EAGAIN ; +if (!xfs_has_larp (mp)) { + error =xfs_attr3_leaf_flipflags (args); +if (error) + goto out; +/* +* Commit the flag value change and start the next trans +* in series +*/ + dac->dela_state =XFS_DAS_FLIP_NFLAG ; +trace_xfs_attr_set_iter_return (dac->dela_state, + args->dp); +return -EAGAIN ; + } +/* fallthrough */ case XFS_DAS_FLIP_NFLAG :/* * Dismantle the "old" attribute/value pair by removing a
@@ -1236,6 +1246,7 @@ xfs_attr_node_addname_clear_incomplete( {struct xfs_da_args *args = dac->da_args;struct xfs_da_state *state =NULL ; +struct xfs_mount *mp = args->dp->i_mount;int retval =0 ;int error =0 ;
@@ -1243,7 +1254,8 @@ xfs_attr_node_addname_clear_incomplete(* Re-find the "old" attribute entry after any split ops. The INCOMPLETE * flag means that we will find the "old" attr, not the "new" one. */ - args->attr_filter |=XFS_ATTR_INCOMPLETE ; +if (!xfs_has_larp (mp)) + args->attr_filter |=XFS_ATTR_INCOMPLETE ; state =xfs_da_state_alloc (args); state->inleaf =0 ; error =xfs_da3_node_lookup_int (state, &retval);
fs/xfs/libxfs/xfs_attr_leaf.c +2 -1
@@ -1487,7 +1487,8 @@ xfs_attr3_leaf_add_work(if (tmp) entry->flags |=XFS_ATTR_LOCAL ;if (args->op_flags &XFS_DA_OP_RENAME ) { - entry->flags |=XFS_ATTR_INCOMPLETE ; +if (!xfs_has_larp (mp)) + entry->flags |=XFS_ATTR_INCOMPLETE ;if ((args->blkno2 == args->blkno) && (args->index2 <= args->index)) { args->index2++;
Signature · 1 line
2.25.1
Re: [PATCH v24 05/11] RFC xfs: Skip flip flags for delayed attrs
Quoted 15 lines · Allison Henderson
This is a clean up patch that skips the flip flag logic for delayed attr renames. Since the log replay keeps the inode locked, we do not need to worry about race windows with attr lookups. So we can skip over flipping the flag and the extra transaction roll for it
RFC: In the last review, folks asked for some performance analysis, so I did a few perf captures with and with out this patch. What I found was that there wasnt very much difference at all between having the patch or not having it. Of the time we do spend in the affected code, the percentage is small. Most of the time we spend about %0.03 of the time in this function, with or with out the patch. Occasionally we get a 0.02%, though not often. So I think this starts to challenge needing this patch at all. This patch was requested some number of reviews ago, be perhaps in light of the findings, it may no longer be of interest.
Quoted 3 lines
0.03% 0.00% fsstress [xfs] [k]
xfs_attr_set_iterKeep it or drop it?
Looks good to me.
Quoted 116 lines
Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 54 +++++++++++++++++++++-------------- fs/xfs/libxfs/xfs_attr_leaf.c | 3 +- 2 files changed, 35 insertions(+), 22 deletions(-)fs/xfs/libxfs/xfs_attr.c +33
-21@@ -355,6 +355,7 @@ xfs_attr_set_iter(struct xfs_inode *dp = args->dp;struct xfs_buf *bp =NULL ;int forkoff, error =0 ; +struct xfs_mount *mp = args->dp->i_mount;/* State machine switch */ switch (dac->dela_state) {@@ -477,16 +478,21 @@ xfs_attr_set_iter(* In a separate transaction, set the incomplete flag on the * "old" attr and clear the incomplete flag on the "new" attr. */ - error =xfs_attr3_leaf_flipflags (args); -if (error) -return error; -/* -* Commit the flag value change and start the next trans in -* series. -*/ - dac->dela_state =XFS_DAS_FLIP_LFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); -return -EAGAIN ; +if (!xfs_has_larp (mp)) { + error =xfs_attr3_leaf_flipflags (args); +if (error) +return error; +/* +* Commit the flag value change and start the next trans +* in series. +*/ + dac->dela_state =XFS_DAS_FLIP_LFLAG ; +trace_xfs_attr_set_iter_return (dac->dela_state, + args->dp); +return -EAGAIN ; + } + +/* fallthrough */ case XFS_DAS_FLIP_LFLAG :/* * Dismantle the "old" attribute/value pair by removing a @@ -589,17 +595,21 @@ xfs_attr_set_iter(* In a separate transaction, set the incomplete flag on the * "old" attr and clear the incomplete flag on the "new" attr. */ - error =xfs_attr3_leaf_flipflags (args); -if (error) - goto out; -/* -* Commit the flag value change and start the next trans in -* series -*/ - dac->dela_state =XFS_DAS_FLIP_NFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); -return -EAGAIN ; +if (!xfs_has_larp (mp)) { + error =xfs_attr3_leaf_flipflags (args); +if (error) + goto out; +/* +* Commit the flag value change and start the next trans +* in series +*/ + dac->dela_state =XFS_DAS_FLIP_NFLAG ; +trace_xfs_attr_set_iter_return (dac->dela_state, + args->dp); +return -EAGAIN ; + } +/* fallthrough */ case XFS_DAS_FLIP_NFLAG :/* * Dismantle the "old" attribute/value pair by removing a @@ -1236,6 +1246,7 @@ xfs_attr_node_addname_clear_incomplete( {struct xfs_da_args *args = dac->da_args;struct xfs_da_state *state =NULL ; +struct xfs_mount *mp = args->dp->i_mount;int retval =0 ;int error =0 ;@@ -1243,7 +1254,8 @@ xfs_attr_node_addname_clear_incomplete(* Re-find the "old" attribute entry after any split ops. The INCOMPLETE * flag means that we will find the "old" attr, not the "new" one. */ - args->attr_filter |=XFS_ATTR_INCOMPLETE ; +if (!xfs_has_larp (mp)) + args->attr_filter |=XFS_ATTR_INCOMPLETE ; state =xfs_da_state_alloc (args); state->inleaf =0 ; error =xfs_da3_node_lookup_int (state, &retval);fs/xfs/libxfs/xfs_attr_leaf.c +2
-1@@ -1487,7 +1487,8 @@ xfs_attr3_leaf_add_work(if (tmp) entry->flags |=XFS_ATTR_LOCAL ;if (args->op_flags &XFS_DA_OP_RENAME ) { - entry->flags |=XFS_ATTR_INCOMPLETE ; +if (!xfs_has_larp (mp)) + entry->flags |=XFS_ATTR_INCOMPLETE ;if ((args->blkno2 == args->blkno) && (args->index2 <= args->index)) { args->index2++;
Signature · 1 line
chandan
Re: [PATCH v24 05/11] RFC xfs: Skip flip flags for delayed attrs
Quoted 16 lines · Chandan Babu R
Quoted 15 lines · Allison Henderson
This is a clean up patch that skips the flip flag logic for delayed attr renames. Since the log replay keeps the inode locked, we do not need to worry about race windows with attr lookups. So we can skip over flipping the flag and the extra transaction roll for it
RFC: In the last review, folks asked for some performance analysis, so I did a few perf captures with and with out this patch. What I found was that there wasnt very much difference at all between having the patch or not having it. Of the time we do spend in the affected code, the percentage is small. Most of the time we spend about %0.03 of the time in this function, with or with out the patch. Occasionally we get a 0.02%, though not often. So I think this starts to challenge needing this patch at all. This patch was requested some number of reviews ago, be perhaps in light of the findings, it may no longer be of interest.
Quoted 7 lines
Quoted 3 lines
0.03% 0.00% fsstress [xfs] [k]
xfs_attr_set_iterKeep it or drop it?
Looks good to me.
Reviewed-by Chandan Babu R <[email protected]>
Ok, thank you!
Allison
Quoted 119 lines
Quoted 117 lines
Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 54 +++++++++++++++++++++-------------- fs/xfs/libxfs/xfs_attr_leaf.c | 3 +- 2 files changed, 35 insertions(+), 22 deletions(-)fs/xfs/libxfs/xfs_attr.c +33
-21@@ -355,6 +355,7 @@ xfs_attr_set_iter(struct xfs_inode *dp = args->dp;struct xfs_buf *bp =NULL ;int forkoff, error =0 ; +struct xfs_mount *mp = args->dp->i_mount;/* State machine switch */ switch (dac->dela_state) {@@ -477,16 +478,21 @@ xfs_attr_set_iter(* In a separate transaction, set the incomplete flag on the * "old" attr and clear the incomplete flag on the "new" attr. */ - error =xfs_attr3_leaf_flipflags (args); -if (error) -return error; -/* -* Commit the flag value change and start the next trans in -* series. -*/ - dac->dela_state =XFS_DAS_FLIP_LFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); -return -EAGAIN ; +if (!xfs_has_larp (mp)) { + error =xfs_attr3_leaf_flipflags (args); +if (error) +return error; +/* +* Commit the flag value change and start the next trans +* in series. +*/ + dac->dela_state =XFS_DAS_FLIP_LFLAG ; +trace_xfs_attr_set_iter_return (dac->dela_state, + args->dp); +return -EAGAIN ; + } + +/* fallthrough */ case XFS_DAS_FLIP_LFLAG :/* * Dismantle the "old" attribute/value pair by removing a @@ -589,17 +595,21 @@ xfs_attr_set_iter(* In a separate transaction, set the incomplete flag on the * "old" attr and clear the incomplete flag on the "new" attr. */ - error =xfs_attr3_leaf_flipflags (args); -if (error) - goto out; -/* -* Commit the flag value change and start the next trans in -* series -*/ - dac->dela_state =XFS_DAS_FLIP_NFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); -return -EAGAIN ; +if (!xfs_has_larp (mp)) { + error =xfs_attr3_leaf_flipflags (args); +if (error) + goto out; +/* +* Commit the flag value change and start the next trans +* in series +*/ + dac->dela_state =XFS_DAS_FLIP_NFLAG ; +trace_xfs_attr_set_iter_return (dac->dela_state, + args->dp); +return -EAGAIN ; + } +/* fallthrough */ case XFS_DAS_FLIP_NFLAG :/* * Dismantle the "old" attribute/value pair by removing a @@ -1236,6 +1246,7 @@ xfs_attr_node_addname_clear_incomplete( {struct xfs_da_args *args = dac->da_args;struct xfs_da_state *state =NULL ; +struct xfs_mount *mp = args->dp->i_mount;int retval =0 ;int error =0 ;@@ -1243,7 +1254,8 @@ xfs_attr_node_addname_clear_incomplete(* Re-find the "old" attribute entry after any split ops. The INCOMPLETE * flag means that we will find the "old" attr, not the "new" one. */ - args->attr_filter |=XFS_ATTR_INCOMPLETE ; +if (!xfs_has_larp (mp)) + args->attr_filter |=XFS_ATTR_INCOMPLETE ; state =xfs_da_state_alloc (args); state->inleaf =0 ; error =xfs_da3_node_lookup_int (state, &retval);fs/xfs/libxfs/xfs_attr_leaf.c +2
-1@@ -1487,7 +1487,8 @@ xfs_attr3_leaf_add_work(if (tmp) entry->flags |=XFS_ATTR_LOCAL ;if (args->op_flags &XFS_DA_OP_RENAME ) { - entry->flags |=XFS_ATTR_INCOMPLETE ; +if (!xfs_has_larp (mp)) + entry->flags |=XFS_ATTR_INCOMPLETE ;if ((args->blkno2 == args->blkno) && (args->index2 <= args->index)) { args->index2++;
PATCH v24 6/11 xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
In reply to Allison HendersonFrom: Allison Collins <[email protected]>
These routines set up and queue a new deferred attribute operations. These functions are meant to be called by any routine needing to initiate a deferred attribute operation as opposed to the existing inline operations. New helper function xfs_attr_item_init also added.
Finally enable delayed attributes in xfs_attr_set and xfs_attr_remove.
--- fs/xfs/libxfs/xfs_attr.c | 71 ++++++++++++++++++++++++++++++++++++++-- fs/xfs/libxfs/xfs_attr.h | 2 ++ fs/xfs/xfs_log.c | 41 +++++++++++++++++++++++ fs/xfs/xfs_log.h | 1 + 4 files changed, 112 insertions(+), 3 deletions(-)
fs/xfs/libxfs/xfs_attr.c +68 -3
@@ -25,6 +25,8 @@#include "xfs_trans_space.h" #include "xfs_trace.h" #include "xfs_attr_item.h" +#include "xfs_attr.h" +#include "xfs_log.h" /* * xfs_attr.c
@@ -726,6 +728,7 @@ xfs_attr_set(int error, local;int rmt_blks =0 ;unsigned int total; +int delayed =xfs_has_larp (mp);if (xfs_is_shutdown (dp->i_mount))return -EIO ;
@@ -782,13 +785,19 @@ xfs_attr_set( rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); } +if (delayed) { + error =xfs_attr_use_log_assist (mp); +if (error) +return error; + } +/* * Root fork attributes can use reserved data blocks for this * operation if necessary */ error =xfs_trans_alloc_inode (dp, &tres, total,0 , rsvd, &args->trans);if (error) -return error; + goto drop_incompat;if (args->value ||xfs_inode_hasattr (dp)) { error = xfs_iext_count_may_overflow(dp, XFS_ATTR_FORK,
@@ -806,9 +815,10 @@ xfs_attr_set(if (error != -ENOATTR && error != -EEXIST ) goto out_trans_cancel; - error =xfs_attr_set_args (args); + error =xfs_attr_set_deferred (args);if (error) goto out_trans_cancel; +/* shortform attribute has already been committed */ if (!args->trans) goto out_unlock;
@@ -816,7 +826,7 @@ xfs_attr_set(if (error != -EEXIST ) goto out_trans_cancel; - error =xfs_attr_remove_args (args); + error =xfs_attr_remove_deferred (args);if (error) goto out_trans_cancel; }
@@ -838,6 +848,9 @@ xfs_attr_set( error =xfs_trans_commit (args->trans); out_unlock:xfs_iunlock (dp,XFS_ILOCK_EXCL ); +drop_incompat: +if (delayed) +xlog_drop_incompat_feat (mp->m_log);return error; out_trans_cancel:
@@ -846,6 +859,58 @@ xfs_attr_set( goto out_unlock; } +STATIC int +xfs_attr_item_init ( +struct xfs_da_args *args, +unsigned int op_flags,/* op flag (set or remove) */ +struct xfs_attr_item **attr)/* new xfs_attr_item */ +{ + +struct xfs_attr_item *new; + + new =kmem_zalloc (sizeof (struct xfs_attr_item ),KM_NOFS ); + new->xattri_op_flags = op_flags; + new->xattri_dac.da_args = args; + + *attr = new; +return 0 ; +} + +/* Sets an attribute for an inode as a deferred operation */ +int +xfs_attr_set_deferred ( +struct xfs_da_args *args) +{ +struct xfs_attr_item *new; +int error =0 ; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_SET , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} + +/* Removes an attribute for an inode as a deferred operation */ +int +xfs_attr_remove_deferred ( +struct xfs_da_args *args) +{ + +struct xfs_attr_item *new; +int error; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_REMOVE , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} +/*======================================================================== * External routines when attribute list is inside the inode *========================================================================*/
fs/xfs/libxfs/xfs_attr.h +2 -0
@@ -526,5 +526,7 @@ bool xfs_attr_namecheck(const void *name, size_t length);void xfs_delattr_context_init (struct xfs_delattr_context *dac,struct xfs_da_args *args);int xfs_attr_calc_size (struct xfs_da_args *args,int *local); +int xfs_attr_set_deferred (struct xfs_da_args *args); +int xfs_attr_remove_deferred (struct xfs_da_args *args);#endif /* __XFS_ATTR_H__ */
fs/xfs/xfs_log.c +41 -0
44 lines not on this page · raw message
fs/xfs/xfs_log.h +1 -0
@@ -144,5 +144,6 @@ bool xlog_force_shutdown(struct xlog *log, int shutdown_flags);void xlog_use_incompat_feat (struct xlog *log);void xlog_drop_incompat_feat (struct xlog *log); +int xfs_attr_use_log_assist (struct xfs_mount *mp);#endif /* __XFS_LOG_H__ */
Signature · 1 line
2.25.1
Re: [PATCH v24 06/11] xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
Quoted 9 lines · Allison Henderson
From: Allison Collins <[email protected]>
These routines set up and queue a new deferred attribute operations. These functions are meant to be called by any routine needing to initiate a deferred attribute operation as opposed to the existing inline operations. New helper function
xfs_attr_item_initalso added.Finally enable delayed attributes in
xfs_attr_setandxfs_attr_remove.
Looks good to me.
Quoted 213 lines
Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 71 ++++++++++++++++++++++++++++++++++++++-- fs/xfs/libxfs/xfs_attr.h | 2 ++ fs/xfs/xfs_log.c | 41 +++++++++++++++++++++++ fs/xfs/xfs_log.h | 1 + 4 files changed, 112 insertions(+), 3 deletions(-)fs/xfs/libxfs/xfs_attr.c +68
-3@@ -25,6 +25,8 @@#include "xfs_trans_space.h" #include "xfs_trace.h" #include "xfs_attr_item.h" +#include "xfs_attr.h" +#include "xfs_log.h" /* * xfs_attr.c @@ -726,6 +728,7 @@ xfs_attr_set(int error, local;int rmt_blks =0 ;unsigned int total; +int delayed =xfs_has_larp (mp);if (xfs_is_shutdown (dp->i_mount))return -EIO ;@@ -782,13 +785,19 @@ xfs_attr_set( rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); } +if (delayed) { + error =xfs_attr_use_log_assist (mp); +if (error) +return error; + } +/* * Root fork attributes can use reserved data blocks for this * operation if necessary */ error =xfs_trans_alloc_inode (dp, &tres, total,0 , rsvd, &args->trans);if (error) -return error; + goto drop_incompat;if (args->value ||xfs_inode_hasattr (dp)) { error = xfs_iext_count_may_overflow(dp, XFS_ATTR_FORK,@@ -806,9 +815,10 @@ xfs_attr_set(if (error != -ENOATTR && error != -EEXIST ) goto out_trans_cancel; - error =xfs_attr_set_args (args); + error =xfs_attr_set_deferred (args);if (error) goto out_trans_cancel; +/* shortform attribute has already been committed */ if (!args->trans) goto out_unlock;@@ -816,7 +826,7 @@ xfs_attr_set(if (error != -EEXIST ) goto out_trans_cancel; - error =xfs_attr_remove_args (args); + error =xfs_attr_remove_deferred (args);if (error) goto out_trans_cancel; }@@ -838,6 +848,9 @@ xfs_attr_set( error =xfs_trans_commit (args->trans); out_unlock:xfs_iunlock (dp,XFS_ILOCK_EXCL ); +drop_incompat: +if (delayed) +xlog_drop_incompat_feat (mp->m_log);return error; out_trans_cancel:@@ -846,6 +859,58 @@ xfs_attr_set( goto out_unlock; } +STATIC int +xfs_attr_item_init ( +struct xfs_da_args *args, +unsigned int op_flags,/* op flag (set or remove) */ +struct xfs_attr_item **attr)/* new xfs_attr_item */ +{ + +struct xfs_attr_item *new; + + new =kmem_zalloc (sizeof (struct xfs_attr_item ),KM_NOFS ); + new->xattri_op_flags = op_flags; + new->xattri_dac.da_args = args; + + *attr = new; +return 0 ; +} + +/* Sets an attribute for an inode as a deferred operation */ +int +xfs_attr_set_deferred ( +struct xfs_da_args *args) +{ +struct xfs_attr_item *new; +int error =0 ; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_SET , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} + +/* Removes an attribute for an inode as a deferred operation */ +int +xfs_attr_remove_deferred ( +struct xfs_da_args *args) +{ + +struct xfs_attr_item *new; +int error; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_REMOVE , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} +/*======================================================================== * External routines when attribute list is inside the inode *========================================================================*/ fs/xfs/libxfs/xfs_attr.h +2
-0@@ -526,5 +526,7 @@ bool xfs_attr_namecheck(const void *name, size_t length);void xfs_delattr_context_init (struct xfs_delattr_context *dac,struct xfs_da_args *args);int xfs_attr_calc_size (struct xfs_da_args *args,int *local); +int xfs_attr_set_deferred (struct xfs_da_args *args); +int xfs_attr_remove_deferred (struct xfs_da_args *args);#endif /* __XFS_ATTR_H__ */ fs/xfs/xfs_log.c +41
-0@@ -3993,3 +3993,44 @@ xlog_drop_incompat_feat( {up_read (&log->l_incompat_users); } + +/* +* Get permission to use log-assisted atomic exchange of file extents. +* +* Callers must not be running any transactions or hold any inode locks, and +* they must release the permission by calling xlog_drop_incompat_feat +* when they're done. +*/ +int +xfs_attr_use_log_assist ( +struct xfs_mount *mp) +{ +int error =0 ; + +/* +* Protect ourselves from an idle log clearing the logged xattrs log +* incompat feature bit. +*/ +xlog_use_incompat_feat (mp->m_log); + +/* +* If log-assisted xattrs are already enabled, the caller can use the +* log assisted swap functions with the log-incompat reference we got. +*/ +if (sb_version_haslogxattrs (&mp->m_sb)) +return 0 ; + +/* Enable log-assisted xattrs. */ + error =xfs_add_incompat_log_feature (mp, +XFS_SB_FEAT_INCOMPAT_LOG_XATTRS ); +if (error) + goto drop_incompat; + +xfs_warn_once (mp, +"EXPERIMENTAL logged extended attributes feature added. Use at your own risk!" ); + +return 0 ; +drop_incompat: +xlog_drop_incompat_feat (mp->m_log); +return error; +}fs/xfs/xfs_log.h +1
-0@@ -144,5 +144,6 @@ bool xlog_force_shutdown(struct xlog *log, int shutdown_flags);void xlog_use_incompat_feat (struct xlog *log);void xlog_drop_incompat_feat (struct xlog *log); +int xfs_attr_use_log_assist (struct xfs_mount *mp);#endif /* __XFS_LOG_H__ */
Signature · 1 line
chandan
Re: [PATCH v24 06/11] xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
Quoted 8 lines · Chandan Babu R
Quoted 7 lines · Allison Henderson
From: Allison Collins <[email protected]>
These routines set up and queue a new deferred attribute operations. These functions are meant to be called by any routine needing to initiate a deferred attribute operation as opposed to the existing inline operations. New helper function
xfs_attr_item_initalso added.
Quoted 6 lines
Quoted 2 lines
Finally enable delayed attributes in
xfs_attr_setandxfs_attr_remove.Looks good to me.
Reviewed-by Chandan Babu R <[email protected]>
Thank you!
Allison
Quoted 216 lines
Quoted 214 lines
Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 71 ++++++++++++++++++++++++++++++++++++++-- fs/xfs/libxfs/xfs_attr.h | 2 ++ fs/xfs/xfs_log.c | 41 +++++++++++++++++++++++ fs/xfs/xfs_log.h | 1 + 4 files changed, 112 insertions(+), 3 deletions(-)fs/xfs/libxfs/xfs_attr.c +68
-3@@ -25,6 +25,8 @@#include "xfs_trans_space.h" #include "xfs_trace.h" #include "xfs_attr_item.h" +#include "xfs_attr.h" +#include "xfs_log.h" /* * xfs_attr.c @@ -726,6 +728,7 @@ xfs_attr_set(int error, local;int rmt_blks =0 ;unsigned int total; +int delayed =xfs_has_larp (mp);if (xfs_is_shutdown (dp->i_mount))return -EIO ;@@ -782,13 +785,19 @@ xfs_attr_set( rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); } +if (delayed) { + error =xfs_attr_use_log_assist (mp); +if (error) +return error; + } +/* * Root fork attributes can use reserved data blocks for this * operation if necessary */ error =xfs_trans_alloc_inode (dp, &tres, total,0 , rsvd, &args->trans);if (error) -return error; + goto drop_incompat;if (args->value ||xfs_inode_hasattr (dp)) { error = xfs_iext_count_may_overflow(dp, XFS_ATTR_FORK,@@ -806,9 +815,10 @@ xfs_attr_set(if (error != -ENOATTR && error != -EEXIST ) goto out_trans_cancel; - error =xfs_attr_set_args (args); + error =xfs_attr_set_deferred (args);if (error) goto out_trans_cancel; +/* shortform attribute has already been committed */ if (!args->trans) goto out_unlock;@@ -816,7 +826,7 @@ xfs_attr_set(if (error != -EEXIST ) goto out_trans_cancel; - error =xfs_attr_remove_args (args); + error =xfs_attr_remove_deferred (args);if (error) goto out_trans_cancel; }@@ -838,6 +848,9 @@ xfs_attr_set( error =xfs_trans_commit (args->trans); out_unlock:xfs_iunlock (dp,XFS_ILOCK_EXCL ); +drop_incompat: +if (delayed) +xlog_drop_incompat_feat (mp->m_log);return error; out_trans_cancel:@@ -846,6 +859,58 @@ xfs_attr_set( goto out_unlock; } +STATIC int +xfs_attr_item_init ( +struct xfs_da_args *args, +unsigned int op_flags,/* op flag (set or remove) */ +struct xfs_attr_item **attr)/* new xfs_attr_item */ +{ + +struct xfs_attr_item *new; + + new =kmem_zalloc (sizeof (struct xfs_attr_item ),KM_NOFS ); + new->xattri_op_flags = op_flags; + new->xattri_dac.da_args = args; + + *attr = new; +return 0 ; +} + +/* Sets an attribute for an inode as a deferred operation */ +int +xfs_attr_set_deferred ( +struct xfs_da_args *args) +{ +struct xfs_attr_item *new; +int error =0 ; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_SET , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} + +/* Removes an attribute for an inode as a deferred operation */ +int +xfs_attr_remove_deferred ( +struct xfs_da_args *args) +{ + +struct xfs_attr_item *new; +int error; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_REMOVE , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} +/*======================================================================== * External routines when attribute list is inside the inode *========================================================================*/ fs/xfs/libxfs/xfs_attr.h +2
-0@@ -526,5 +526,7 @@ bool xfs_attr_namecheck(const void *name, size_t length);void xfs_delattr_context_init (struct xfs_delattr_context *dac,struct xfs_da_args *args);int xfs_attr_calc_size (struct xfs_da_args *args,int *local); +int xfs_attr_set_deferred (struct xfs_da_args *args); +int xfs_attr_remove_deferred (struct xfs_da_args *args);#endif /* __XFS_ATTR_H__ */ fs/xfs/xfs_log.c +41
-0@@ -3993,3 +3993,44 @@ xlog_drop_incompat_feat( {up_read (&log->l_incompat_users); } + +/* +* Get permission to use log-assisted atomic exchange of file extents. +* +* Callers must not be running any transactions or hold any inode locks, and +* they must release the permission by calling xlog_drop_incompat_feat +* when they're done. +*/ +int +xfs_attr_use_log_assist ( +struct xfs_mount *mp) +{ +int error =0 ; + +/* +* Protect ourselves from an idle log clearing the logged xattrs log +* incompat feature bit. +*/ +xlog_use_incompat_feat (mp->m_log); + +/* +* If log-assisted xattrs are already enabled, the caller can use the +* log assisted swap functions with the log-incompat reference we got. +*/ +if (sb_version_haslogxattrs (&mp->m_sb)) +return 0 ; + +/* Enable log-assisted xattrs. */ + error =xfs_add_incompat_log_feature (mp, +XFS_SB_FEAT_INCOMPAT_LOG_XATTRS ); +if (error) + goto drop_incompat; + +xfs_warn_once (mp, +"EXPERIMENTAL logged extended attributes feature added. Use at your own risk!" ); + +return 0 ; +drop_incompat: +xlog_drop_incompat_feat (mp->m_log); +return error; +}fs/xfs/xfs_log.h +1
-0@@ -144,5 +144,6 @@ bool xlog_force_shutdown(struct xlog *log, int shutdown_flags);void xlog_use_incompat_feat (struct xlog *log);void xlog_drop_incompat_feat (struct xlog *log); +int xfs_attr_use_log_assist (struct xfs_mount *mp);#endif /* __XFS_LOG_H__ */
Re: [PATCH v24 06/11] xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
In reply to Allison HendersonQuoted 11 lines · Allison Henderson
From: Allison Collins <[email protected]>
These routines set up and queue a new deferred attribute operations. These functions are meant to be called by any routine needing to initiate a deferred attribute operation as opposed to the existing inline operations. New helper function
xfs_attr_item_initalso added.Finally enable delayed attributes in
xfs_attr_setandxfs_attr_remove.Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>
.....
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 871
Quoted 11 lines
fs/xfs/libxfs/xfs_attr.c +10
-0+STATIC int +xfs_attr_item_init ( +struct xfs_da_args *args, +unsigned int op_flags,/* op flag (set or remove) */ +struct xfs_attr_item **attr)/* new xfs_attr_item */ +{ + +struct xfs_attr_item *new; + + new =kmem_zalloc (sizeof (struct xfs_attr_item ),KM_NOFS );
In transaction context here so we don't need KM_NOFS.
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 877
Quoted 6 lines
fs/xfs/libxfs/xfs_attr.c +6
-0+ new->xattri_op_flags = op_flags; + new->xattri_dac.da_args = args; + + *attr = new; +return 0 ; +}
Why doesn't this just return the object or NULL on allocation failure? What other error could it ever return?
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 889
Quoted 12 lines
fs/xfs/libxfs/xfs_attr.c +12
-0+ +/* Sets an attribute for an inode as a deferred operation */ +int +xfs_attr_set_deferred ( +struct xfs_da_args *args) +{ +struct xfs_attr_item *new; +int error =0 ; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_SET , &new); +if (error) +return error;
i.e.
attri = xfs_attr_item_init(args, XFS_ATTR_OP_FLAGS_SET); if (!attri) return -ENOMEM;
Quoted 13 lines
fs/xfs/libxfs/xfs_attr.c +13
-0+ +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} + +/* Removes an attribute for an inode as a deferred operation */ +int +xfs_attr_remove_deferred ( +struct xfs_da_args *args) +{ + +struct xfs_attr_item *new;
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 912
Quoted 10 lines
fs/xfs/libxfs/xfs_attr.c +10
-0+int error; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_REMOVE , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +}
We really should not use "new" as a variable name. As a general rule, the common pattern set by this file is that xfs_attri_item objects in a function are named "attri". Just because it's newly allocated doesn't mean we should use a different convention for naming xfs_attri_item objects...
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
Re: [PATCH v24 06/11] xfs: Add xfs_attr_set_deferred and xfs_attr_remove_deferred
Quoted 14 lines · Dave Chinner
Quoted 11 lines · Allison Henderson
From: Allison Collins <[email protected]>
These routines set up and queue a new deferred attribute operations. These functions are meant to be called by any routine needing to initiate a deferred attribute operation as opposed to the existing inline operations. New helper function
xfs_attr_item_initalso added.Finally enable delayed attributes in
xfs_attr_setandxfs_attr_remove.Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>.....
Quoted 1 line
Quoted 12 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 871
Quoted 10 lines
fs/xfs/libxfs/xfs_attr.c +10
-0+STATIC int +xfs_attr_item_init ( +struct xfs_da_args *args, +unsigned int op_flags,/* op flag (set or remove) */ +struct xfs_attr_item **attr)/* new xfs_attr_item */ +{ + +struct xfs_attr_item *new; + + new =kmem_zalloc (sizeof (struct xfs_attr_item ),KM_NOFS );In transaction context here so we don't need
KM_NOFS.
ok, will remove
Quoted 10 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 877
Quoted 7 lines
fs/xfs/libxfs/xfs_attr.c +6
-0+ new->xattri_op_flags = op_flags; + new->xattri_dac.da_args = args; + + *attr = new; +return 0 ; +}Why doesn't this just return the object or NULL on allocation failure? What other error could it ever return?
I had adopted this function signature just to be consistent with other *_item_init routines at the time. Mostly just trying to be uniform, though they may have changed since. This patch spent some time on the back burner while we were more focused on the state machine refactoring.
Quoted 40 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 889
Quoted 13 lines
fs/xfs/libxfs/xfs_attr.c +12
-0+ +/* Sets an attribute for an inode as a deferred operation */ +int +xfs_attr_set_deferred ( +struct xfs_da_args *args) +{ +struct xfs_attr_item *new; +int error =0 ; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_SET , &new); +if (error) +return error;i.e.
attri = xfs_attr_item_init(args, XFS_ATTR_OP_FLAGS_SET); if (!attri) return -ENOMEM;Quoted 21 lines
fs/xfs/libxfs/xfs_attr.c +21
-0+ +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); + +return 0 ; +} + +/* Removes an attribute for an inode as a deferred operation */ +int +xfs_attr_remove_deferred ( +struct xfs_da_args *args) +{ + +struct xfs_attr_item *new; +int error; + + error =xfs_attr_item_init (args,XFS_ATTR_OP_FLAGS_REMOVE , &new); +if (error) +return error; + +xfs_defer_add (args->trans,XFS_DEFER_OPS_TYPE_ATTR , &new->xattri_list); +
Quoted 9 lines
Review
Quoted 2 lines
(quoted diff) +2
-0+ return 0; +}We really should not use "new" as a variable name. As a general rule, the common pattern set by this file is that
xfs_attri_itemobjects in a function are named "attri". Just because it's newly allocated doesn't mean we should use a different convention for namingxfs_attri_itemobjects...
Ok, I had seen the pattern around and reused it. Will change to attri
Allison
Quoted 4 lines
Cheers,
Dave.
PATCH v24 7/11 xfs: Remove unused xfs_attr_*_args
In reply to Allison HendersonRemove xfs_attr_set_args, xfs_attr_remove_args, and xfs_attr_trans_roll. These high level loops are now driven by the delayed operations code, and can be removed.
Additionally collapse in the leaf_bp parameter of xfs_attr_set_iter since we only have one caller that passes dac->leaf_bp
--- fs/xfs/libxfs/xfs_attr.c | 106 +++----------------------------- fs/xfs/libxfs/xfs_attr.h | 10 +-- fs/xfs/libxfs/xfs_attr_remote.c | 1 - fs/xfs/xfs_attr_item.c | 6 +- 4 files changed, 14 insertions(+), 109 deletions(-)
fs/xfs/libxfs/xfs_attr.c +9 -97
@@ -244,64 +244,9 @@ xfs_attr_is_shortform( ip->i_afp->if_nextents == 0); } -/* -* Checks to see if a delayed attribute transaction should be rolled. If so, -* transaction is finished or rolled as needed. -*/ -STATIC int -xfs_attr_trans_roll ( -struct xfs_delattr_context *dac) -{ -struct xfs_da_args *args = dac->da_args; -int error; - -if (dac->flags &XFS_DAC_DEFER_FINISH ) { -/* -* The caller wants us to finish all the deferred ops so that we -* avoid pinning the log tail with a large number of deferred -* ops. -*/ - dac->flags &= ~XFS_DAC_DEFER_FINISH ; - error =xfs_defer_finish (&args->trans); - }else - error =xfs_trans_roll_inode (&args->trans, args->dp); - -return error; -} - -/* -* Set the attribute specified in @args. -*/ -int -xfs_attr_set_args ( -struct xfs_da_args *args) -{ -struct xfs_buf *leaf_bp =NULL ; -int error =0 ; -struct xfs_delattr_context dac = { - .da_args = args, - }; - -do { - error =xfs_attr_set_iter (&dac, &leaf_bp); -if (error != -EAGAIN ) -break ; - - error =xfs_attr_trans_roll (&dac); -if (error) { -if (leaf_bp) -xfs_trans_brelse (args->trans, leaf_bp); -return error; - } - }while (true); - -return error; -} - STATIC intxfs_attr_sf_addname ( -struct xfs_delattr_context *dac, -struct xfs_buf **leaf_bp) +struct xfs_delattr_context *dac) {struct xfs_da_args *args = dac->da_args;struct xfs_inode *dp = args->dp;
@@ -320,7 +265,7 @@ xfs_attr_sf_addname(* It won't fit in the shortform, transform to a leaf block. GROT: * another possible req'mt for a double-split btree op. */ - error =xfs_attr_shortform_to_leaf (args, leaf_bp); + error =xfs_attr_shortform_to_leaf (args, &dac->leaf_bp);if (error)return error;
@@ -329,7 +274,7 @@ xfs_attr_sf_addname(* push cannot grab the half-baked leaf buffer and run into problems * with the write verifier. */ -xfs_trans_bhold (args->trans, *leaf_bp); +xfs_trans_bhold (args->trans, dac->leaf_bp);/* * We're still in XFS_DAS_UNINIT state here. We've converted
@@ -337,7 +282,6 @@ xfs_attr_sf_addname(* add. */ trace_xfs_attr_sf_addname_return (XFS_DAS_UNINIT , args->dp); - dac->flags |=XFS_DAC_DEFER_FINISH ;return -EAGAIN ; }
@@ -350,8 +294,7 @@ xfs_attr_sf_addname(*/ int xfs_attr_set_iter ( -struct xfs_delattr_context *dac, -struct xfs_buf **leaf_bp) +struct xfs_delattr_context *dac) {struct xfs_da_args *args = dac->da_args;struct xfs_inode *dp = args->dp;
@@ -370,14 +313,14 @@ xfs_attr_set_iter(* release the hold once we return with a clean transaction. */ if (xfs_attr_is_shortform (dp)) -return xfs_attr_sf_addname (dac, leaf_bp); -if (*leaf_bp !=NULL ) { -xfs_trans_bhold_release (args->trans, *leaf_bp); - *leaf_bp =NULL ; +return xfs_attr_sf_addname (dac); +if (dac->leaf_bp !=NULL ) { +xfs_trans_bhold_release (args->trans, dac->leaf_bp); + dac->leaf_bp =NULL ; }if (xfs_attr_is_leaf (dp)) { - error =xfs_attr_leaf_try_add (args, *leaf_bp); + error =xfs_attr_leaf_try_add (args, dac->leaf_bp);if (error == -ENOSPC ) { error =xfs_attr3_leaf_to_node (args);if (error)
@@ -396,7 +339,6 @@ xfs_attr_set_iter(* be a node, so we'll fall down into the node * handling code below */ - dac->flags |=XFS_DAC_DEFER_FINISH ;trace_xfs_attr_set_iter_return ( dac->dela_state, args->dp);return -EAGAIN ;
@@ -687,32 +629,6 @@ xfs_attr_lookup(return xfs_attr_node_hasname (args,NULL ); } -/* -* Remove the attribute specified in @args. -*/ -int -xfs_attr_remove_args ( -struct xfs_da_args *args) -{ -int error; -struct xfs_delattr_context dac = { - .da_args = args, - }; - -do { - error =xfs_attr_remove_iter (&dac); -if (error != -EAGAIN ) -break ; - - error =xfs_attr_trans_roll (&dac); -if (error) -return error; - - }while (true); - -return error; -} -/* * Note: If args->value is NULL the attribute will be removed, just like the * Linux ->setattr API.
@@ -1275,7 +1191,6 @@ xfs_attr_node_addname(* this. dela_state is still unset by this function at * this point. */ - dac->flags |=XFS_DAC_DEFER_FINISH ;trace_xfs_attr_node_addname_return ( dac->dela_state, args->dp);return -EAGAIN ;
@@ -1290,7 +1205,6 @@ xfs_attr_node_addname( error =xfs_da3_split (state);if (error) goto out; - dac->flags |=XFS_DAC_DEFER_FINISH ; }else {/* * Addition succeeded, update Btree hashvals.
@@ -1544,7 +1458,6 @@ xfs_attr_remove_iter(if (error) goto out; dac->dela_state =XFS_DAS_RM_NAME ; - dac->flags |=XFS_DAC_DEFER_FINISH ;trace_xfs_attr_remove_iter_return (dac->dela_state, args->dp);return -EAGAIN ; }
@@ -1572,7 +1485,6 @@ xfs_attr_remove_iter(if (error) goto out; - dac->flags |=XFS_DAC_DEFER_FINISH ; dac->dela_state =XFS_DAS_RM_SHRINK ;trace_xfs_attr_remove_iter_return ( dac->dela_state, args->dp);
fs/xfs/libxfs/xfs_attr.h +3 -7
@@ -457,9 +457,8 @@ enum xfs_delattr_state {/* * Defines for xfs_delattr_context.flags */ -#define XFS_DAC_DEFER_FINISH 0x01/* finish the transaction */ -#define XFS_DAC_LEAF_ADDNAME_INIT 0x02/* xfs_attr_leaf_addname init*/ -#define XFS_DAC_DELAYED_OP_INIT 0x04/* delayed operations init*/ +#define XFS_DAC_LEAF_ADDNAME_INIT 0x01/* xfs_attr_leaf_addname init*/ +#define XFS_DAC_DELAYED_OP_INIT 0x02/* delayed operations init*/ /* * Context used for keeping track of delayed attribute operations
@@ -517,10 +516,7 @@ bool xfs_attr_is_leaf(struct xfs_inode *ip);int xfs_attr_get_ilocked (struct xfs_da_args *args);int xfs_attr_get (struct xfs_da_args *args);int xfs_attr_set (struct xfs_da_args *args); -int xfs_attr_set_args (struct xfs_da_args *args); -int xfs_attr_set_iter (struct xfs_delattr_context *dac, -struct xfs_buf **leaf_bp); -int xfs_attr_remove_args (struct xfs_da_args *args); +int xfs_attr_set_iter (struct xfs_delattr_context *dac);int xfs_attr_remove_iter (struct xfs_delattr_context *dac);bool xfs_attr_namecheck (const void *name,size_t length); void xfs_delattr_context_init(struct xfs_delattr_context *dac,
fs/xfs/libxfs/xfs_attr_remote.c +0 -1
@@ -695,7 +695,6 @@ xfs_attr_rmtval_remove(* the parent */ if (!done) { - dac->flags |=XFS_DAC_DEFER_FINISH ;trace_xfs_attr_rmtval_remove_return (dac->dela_state, args->dp);return -EAGAIN ; }
fs/xfs/xfs_attr_item.c +2 -4
@@ -285,7 +285,6 @@ STATIC intxfs_trans_attr_finish_update ( struct xfs_delattr_context *dac, struct xfs_attrd_log_item *attrdp, - struct xfs_buf **leaf_bp, uint32_t op_flags) { struct xfs_da_args *args = dac->da_args;
@@ -295,7 +294,7 @@ xfs_trans_attr_finish_update(switch (op) {case XFS_ATTR_OP_FLAGS_SET : - error =xfs_attr_set_iter (dac, leaf_bp); + error =xfs_attr_set_iter (dac);break ;case XFS_ATTR_OP_FLAGS_REMOVE :ASSERT (XFS_IFORK_Q (args->dp));
@@ -405,7 +404,7 @@ xfs_attr_finish_item(*/ dac->da_args->trans = tp; - error =xfs_trans_attr_finish_update (dac, done_item, &dac->leaf_bp, + error =xfs_trans_attr_finish_update (dac, done_item, attr->xattri_op_flags);if (error != -EAGAIN )kmem_free (attr);
@@ -617,7 +616,6 @@ xfs_attri_item_recover(xfs_trans_ijoin (tp, ip,0 ); ret =xfs_trans_attr_finish_update (&attr->xattri_dac, done_item, - &attr->xattri_dac.leaf_bp, attrp->alfi_op_flags);if (ret == -EAGAIN ) {/* There's more work to do, so add it to this transaction */
Signature · 1 line
2.25.1
PATCH v24 8/11 xfs: Add log attribute error tag
In reply to Allison HendersonThis patch adds an error tag that we can use to test log attribute recovery and replay
--- fs/xfs/libxfs/xfs_errortag.h | 4 +++- fs/xfs/xfs_attr_item.c | 7 +++++++ fs/xfs/xfs_error.c | 3 +++ 3 files changed, 13 insertions(+), 1 deletion(-)
fs/xfs/libxfs/xfs_errortag.h +3 -1
@@ -59,7 +59,8 @@#define XFS_ERRTAG_REDUCE_MAX_IEXTENTS 36#define XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT 37#define XFS_ERRTAG_AG_RESV_FAIL 38 -#define XFS_ERRTAG_MAX 39 +#define XFS_ERRTAG_LARP 39 +#define XFS_ERRTAG_MAX 40/* * Random factors for above tags, 1 means always, 2 means 1/2 time, etc.
@@ -103,5 +104,6 @@#define XFS_RANDOM_REDUCE_MAX_IEXTENTS 1#define XFS_RANDOM_BMAP_ALLOC_MINLEN_EXTENT 1#define XFS_RANDOM_AG_RESV_FAIL 1 +#define XFS_RANDOM_LARP 1#endif /* __XFS_ERRORTAG_H_ */
fs/xfs/xfs_attr_item.c +7 -0
@@ -34,6 +34,7 @@#include "xfs_inode.h" #include "xfs_quota.h" #include "xfs_trans_space.h" +#include "xfs_errortag.h" #include "xfs_error.h" #include "xfs_log_priv.h" #include "xfs_log_recover.h"
@@ -292,6 +293,11 @@ xfs_trans_attr_finish_update(XFS_ATTR_OP_FLAGS_TYPE_MASK ;int error; +if (XFS_TEST_ERROR (false, args->dp->i_mount,XFS_ERRTAG_LARP )) { + error = -EIO ; + goto out; + } +switch (op) {case XFS_ATTR_OP_FLAGS_SET : error =xfs_attr_set_iter (dac);
@@ -305,6 +311,7 @@ xfs_trans_attr_finish_update(break ; } +out:/* * Mark the transaction dirty, even on error. This ensures the * transaction is aborted, which:
fs/xfs/xfs_error.c +3 -0
@@ -57,6 +57,7 @@ static unsigned int xfs_errortag_random_default[] = {XFS_RANDOM_REDUCE_MAX_IEXTENTS ,XFS_RANDOM_BMAP_ALLOC_MINLEN_EXTENT ,XFS_RANDOM_AG_RESV_FAIL , + XFS_RANDOM_LARP, }; struct xfs_errortag_attr {
@@ -170,6 +171,7 @@ XFS_ERRORTAG_ATTR_RW(buf_ioerror, XFS_ERRTAG_BUF_IOERROR);XFS_ERRORTAG_ATTR_RW (reduce_max_iextents,XFS_ERRTAG_REDUCE_MAX_IEXTENTS );XFS_ERRORTAG_ATTR_RW (bmap_alloc_minlen_extent,XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT );XFS_ERRORTAG_ATTR_RW (ag_resv_fail,XFS_ERRTAG_AG_RESV_FAIL ); +XFS_ERRORTAG_ATTR_RW (larp,XFS_ERRTAG_LARP );static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (noerror),
@@ -211,6 +213,7 @@ static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (reduce_max_iextents),XFS_ERRORTAG_ATTR_LIST (bmap_alloc_minlen_extent),XFS_ERRORTAG_ATTR_LIST (ag_resv_fail), +XFS_ERRORTAG_ATTR_LIST (larp), NULL, };
Signature · 1 line
2.25.1
PATCH v24 9/11 xfs: Add larp debug option
In reply to Allison HendersonThis patch adds a mount option to enable log attribute replay. Eventually this can be removed when delayed attrs becomes permanent.
--- fs/xfs/libxfs/xfs_attr.h | 2 +- fs/xfs/xfs_globals.c | 1 + fs/xfs/xfs_sysctl.h | 1 + fs/xfs/xfs_sysfs.c | 24 ++++++++++++++++++++++++ 4 files changed, 27 insertions(+), 1 deletion(-)
fs/xfs/libxfs/xfs_attr.h +1 -1
@@ -30,7 +30,7 @@ struct xfs_attr_list_context;static inline bool xfs_has_larp (struct xfs_mount *mp) { -return false; +return xfs_globals.larp; }/*
fs/xfs/xfs_globals.c +1 -0
@@ -41,5 +41,6 @@ struct xfs_globals xfs_globals = {#endif #ifdef DEBUG .pwork_threads = -1, /* automatic thread detection */ + .larp = false, /* log attribute replay */#endif };
fs/xfs/xfs_sysctl.h +1 -0
@@ -83,6 +83,7 @@ extern xfs_param_t xfs_params;struct xfs_globals {#ifdef DEBUG int pwork_threads;/* parallel workqueue threads */ +bool larp;/* log attribute replay */ #endif int log_recovery_delay;/* log recovery delay (secs) */ int mount_delay;/* mount setup delay (secs) */
fs/xfs/xfs_sysfs.c +24 -0
@@ -227,6 +227,29 @@ pwork_threads_show(return snprintf (buf,PAGE_SIZE ,"%d\n" , xfs_globals.pwork_threads); }XFS_SYSFS_ATTR_RW (pwork_threads); + +static ssize_t +larp_store ( +struct kobject *kobject, +const char *buf, +size_t count) +{ +ssize_t ret; + + ret =kstrtobool (buf, &xfs_globals.larp); +if (ret <0 ) +return ret; +return count; +} + +STATIC ssize_t +larp_show ( +struct kobject *kobject, +char *buf) +{ +return snprintf (buf,PAGE_SIZE ,"%d\n" , xfs_globals.larp); +} +XFS_SYSFS_ATTR_RW (larp);#endif /* DEBUG */ static struct attribute *xfs_dbg_attrs[] = {
@@ -236,6 +259,7 @@ static struct attribute *xfs_dbg_attrs[] = { ATTR_LIST(always_cow),#ifdef DEBUG ATTR_LIST (pwork_threads), +ATTR_LIST (larp), #endif NULL, };
Signature · 1 line
2.25.1
Re: [PATCH v24 09/11] xfs: Add larp debug option
Quoted 1 line · Allison Henderson
This patch adds a mount option to enable log attribute replay. Eventually
s/mount option/debug option//
Quoted 1 line
this can be removed when delayed attrs becomes permanent.
The rest looks good to me.
Quoted 87 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_attr.h | 2 +- fs/xfs/xfs_globals.c | 1 + fs/xfs/xfs_sysctl.h | 1 + fs/xfs/xfs_sysfs.c | 24 ++++++++++++++++++++++++ 4 files changed, 27 insertions(+), 1 deletion(-)fs/xfs/libxfs/xfs_attr.h +1
-1@@ -30,7 +30,7 @@ struct xfs_attr_list_context;static inline bool xfs_has_larp (struct xfs_mount *mp) { -return false; +return xfs_globals.larp; }/* fs/xfs/xfs_globals.c +1
-0@@ -41,5 +41,6 @@ struct xfs_globals xfs_globals = {#endif #ifdef DEBUG .pwork_threads = -1, /* automatic thread detection */ + .larp = false, /* log attribute replay */#endif };fs/xfs/xfs_sysctl.h +1
-0@@ -83,6 +83,7 @@ extern xfs_param_t xfs_params;struct xfs_globals {#ifdef DEBUG int pwork_threads;/* parallel workqueue threads */ +bool larp;/* log attribute replay */ #endif int log_recovery_delay;/* log recovery delay (secs) */ int mount_delay;/* mount setup delay (secs) */ fs/xfs/xfs_sysfs.c +24
-0@@ -227,6 +227,29 @@ pwork_threads_show(return snprintf (buf,PAGE_SIZE ,"%d\n" , xfs_globals.pwork_threads); }XFS_SYSFS_ATTR_RW (pwork_threads); + +static ssize_t +larp_store ( +struct kobject *kobject, +const char *buf, +size_t count) +{ +ssize_t ret; + + ret =kstrtobool (buf, &xfs_globals.larp); +if (ret <0 ) +return ret; +return count; +} + +STATIC ssize_t +larp_show ( +struct kobject *kobject, +char *buf) +{ +return snprintf (buf,PAGE_SIZE ,"%d\n" , xfs_globals.larp); +} +XFS_SYSFS_ATTR_RW (larp);#endif /* DEBUG */ static struct attribute *xfs_dbg_attrs[] = {@@ -236,6 +259,7 @@ static struct attribute *xfs_dbg_attrs[] = { ATTR_LIST(always_cow),#ifdef DEBUG ATTR_LIST (pwork_threads), +ATTR_LIST (larp), #endif NULL, };
Signature · 1 line
chandan
Re: [PATCH v24 09/11] xfs: Add larp debug option
Quoted 4 lines · Chandan Babu R
Quoted 1 line · Allison Henderson
This patch adds a mount option to enable log attribute replay. Eventually
s/mount option/debug option//
Ok, will fix.
Quoted 6 lines
Quoted 2 lines
this can be removed when delayed attrs becomes permanent.
The rest looks good to me.
Reviewed-by Chandan Babu R <[email protected]>
Thanks!
Allison
Quoted 90 lines
Quoted 88 lines
Signed-off-by Allison Henderson <[email protected]>--- fs/xfs/libxfs/xfs_attr.h | 2 +- fs/xfs/xfs_globals.c | 1 + fs/xfs/xfs_sysctl.h | 1 + fs/xfs/xfs_sysfs.c | 24 ++++++++++++++++++++++++ 4 files changed, 27 insertions(+), 1 deletion(-)fs/xfs/libxfs/xfs_attr.h +1
-1@@ -30,7 +30,7 @@ struct xfs_attr_list_context;static inline bool xfs_has_larp (struct xfs_mount *mp) { -return false; +return xfs_globals.larp; }/* fs/xfs/xfs_globals.c +1
-0@@ -41,5 +41,6 @@ struct xfs_globals xfs_globals = {#endif #ifdef DEBUG .pwork_threads = -1, /* automatic thread detection */ + .larp = false, /* log attribute replay */#endif };fs/xfs/xfs_sysctl.h +1
-0@@ -83,6 +83,7 @@ extern xfs_param_t xfs_params;struct xfs_globals {#ifdef DEBUG int pwork_threads;/* parallel workqueue threads */ +bool larp;/* log attribute replay */ #endif int log_recovery_delay;/* log recovery delay (secs) */ int mount_delay;/* mount setup delay (secs) */ fs/xfs/xfs_sysfs.c +24
-0@@ -227,6 +227,29 @@ pwork_threads_show(return snprintf (buf,PAGE_SIZE ,"%d\n" , xfs_globals.pwork_threads); }XFS_SYSFS_ATTR_RW (pwork_threads); + +static ssize_t +larp_store ( +struct kobject *kobject, +const char *buf, +size_t count) +{ +ssize_t ret; + + ret =kstrtobool (buf, &xfs_globals.larp); +if (ret <0 ) +return ret; +return count; +} + +STATIC ssize_t +larp_show ( +struct kobject *kobject, +char *buf) +{ +return snprintf (buf,PAGE_SIZE ,"%d\n" , xfs_globals.larp); +} +XFS_SYSFS_ATTR_RW (larp);#endif /* DEBUG */ static struct attribute *xfs_dbg_attrs[] = {@@ -236,6 +259,7 @@ static struct attribute *xfs_dbg_attrs[] = { ATTR_LIST(always_cow),#ifdef DEBUG ATTR_LIST (pwork_threads), +ATTR_LIST (larp), #endif NULL, };
PATCH v24 10/11 xfs: Merge xfs_delattr_context into xfs_attr_item
In reply to Allison HendersonThis is a clean up patch that merges xfs_delattr_context into xfs_attr_item. Now that the refactoring is complete and the delayed operation infrastructure is in place, we can combine these to eliminate the extra struct
--- fs/xfs/libxfs/xfs_attr.c | 162 +++++++++++++++++--------------- fs/xfs/libxfs/xfs_attr.h | 40 ++++---- fs/xfs/libxfs/xfs_attr_remote.c | 36 +++---- fs/xfs/libxfs/xfs_attr_remote.h | 6 +- fs/xfs/xfs_attr_item.c | 38 ++++---- 5 files changed, 141 insertions(+), 141 deletions(-)
fs/xfs/libxfs/xfs_attr.c +84 -78
@@ -56,10 +56,9 @@ STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);*/ STATIC int xfs_attr_node_get(xfs_da_args_t *args); STATIC void xfs_attr_restore_rmt_blk(struct xfs_da_args *args); -STATIC int xfs_attr_node_addname(struct xfs_delattr_context *dac); -STATIC int xfs_attr_node_addname_find_attr(struct xfs_delattr_context *dac); -STATIC int xfs_attr_node_addname_clear_incomplete( -struct xfs_delattr_context *dac); +STATIC int xfs_attr_node_addname(struct xfs_attr_item *attr); +STATIC int xfs_attr_node_addname_find_attr(struct xfs_attr_item *attr); +STATIC int xfs_attr_node_addname_clear_incomplete(struct xfs_attr_item *attr); STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,struct xfs_da_state **state); STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
@@ -246,9 +245,9 @@ xfs_attr_is_shortform( STATIC intxfs_attr_sf_addname ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;struct xfs_inode *dp = args->dp;int error =0 ;
@@ -265,7 +264,7 @@ xfs_attr_sf_addname(* It won't fit in the shortform, transform to a leaf block. GROT: * another possible req'mt for a double-split btree op. */ - error =xfs_attr_shortform_to_leaf (args, &dac->leaf_bp); + error =xfs_attr_shortform_to_leaf (args, &attr->xattri_leaf_bp);if (error)return error;
@@ -274,7 +273,7 @@ xfs_attr_sf_addname(* push cannot grab the half-baked leaf buffer and run into problems * with the write verifier. */ -xfs_trans_bhold (args->trans, dac->leaf_bp); +xfs_trans_bhold (args->trans, attr->xattri_leaf_bp);/* * We're still in XFS_DAS_UNINIT state here. We've converted
@@ -294,16 +293,16 @@ xfs_attr_sf_addname(*/ int xfs_attr_set_iter ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;struct xfs_inode *dp = args->dp;struct xfs_buf *bp =NULL ;int forkoff, error =0 ;struct xfs_mount *mp = args->dp->i_mount;/* State machine switch */ -switch (dac->dela_state) { +switch (attr->xattri_dela_state) {case XFS_DAS_UNINIT :/* * If the fork is shortform, attempt to add the attr. If there
@@ -313,14 +312,16 @@ xfs_attr_set_iter(* release the hold once we return with a clean transaction. */ if (xfs_attr_is_shortform (dp)) -return xfs_attr_sf_addname (dac); -if (dac->leaf_bp !=NULL ) { -xfs_trans_bhold_release (args->trans, dac->leaf_bp); - dac->leaf_bp =NULL ; +return xfs_attr_sf_addname (attr); +if (attr->xattri_leaf_bp !=NULL ) { +xfs_trans_bhold_release (args->trans, + attr->xattri_leaf_bp); + attr->xattri_leaf_bp =NULL ; }if (xfs_attr_is_leaf (dp)) { - error =xfs_attr_leaf_try_add (args, dac->leaf_bp); + error =xfs_attr_leaf_try_add (args, + attr->xattri_leaf_bp);if (error == -ENOSPC ) { error =xfs_attr3_leaf_to_node (args);if (error)
@@ -340,19 +341,19 @@ xfs_attr_set_iter(* handling code below */ trace_xfs_attr_set_iter_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);return -EAGAIN ; }else if (error) {return error; } - dac->dela_state =XFS_DAS_FOUND_LBLK ; + attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; }else { - error =xfs_attr_node_addname_find_attr (dac); + error =xfs_attr_node_addname_find_attr (attr);if (error)return error; - error =xfs_attr_node_addname (dac); + error =xfs_attr_node_addname (attr);if (error)return error;
@@ -364,9 +365,10 @@ xfs_attr_set_iter( !(args->op_flags & XFS_DA_OP_RENAME))return 0 ; - dac->dela_state =XFS_DAS_FOUND_NBLK ; + attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; } -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); +trace_xfs_attr_set_iter_return (attr->xattri_dela_state, + args->dp);return -EAGAIN ;case XFS_DAS_FOUND_LBLK :/*
@@ -377,10 +379,10 @@ xfs_attr_set_iter(*/ /* Open coded xfs_attr_rmtval_set without trans handling */ -if ((dac->flags &XFS_DAC_LEAF_ADDNAME_INIT ) ==0 ) { - dac->flags |=XFS_DAC_LEAF_ADDNAME_INIT ; +if ((attr->xattri_flags &XFS_DAC_LEAF_ADDNAME_INIT ) ==0 ) { + attr->xattri_flags |=XFS_DAC_LEAF_ADDNAME_INIT ;if (args->rmtblkno >0 ) { - error =xfs_attr_rmtval_find_space (dac); + error =xfs_attr_rmtval_find_space (attr);if (error)return error; }
@@ -390,11 +392,11 @@ xfs_attr_set_iter(* Repeat allocating remote blocks for the attr value until * blkcnt drops to zero. */ -if (dac->blkcnt >0 ) { - error =xfs_attr_rmtval_set_blk (dac); +if (attr->xattri_blkcnt >0 ) { + error =xfs_attr_rmtval_set_blk (attr);if (error)return error; -trace_xfs_attr_set_iter_return (dac->dela_state, +trace_xfs_attr_set_iter_return (attr->xattri_dela_state, args->dp);return -EAGAIN ; }
@@ -430,8 +432,8 @@ xfs_attr_set_iter(* Commit the flag value change and start the next trans * in series. */ - dac->dela_state =XFS_DAS_FLIP_LFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, + attr->xattri_dela_state =XFS_DAS_FLIP_LFLAG ; +trace_xfs_attr_set_iter_return (attr->xattri_dela_state, args->dp);return -EAGAIN ; }
@@ -450,17 +452,18 @@ xfs_attr_set_iter( fallthrough;case XFS_DAS_RM_LBLK :/* Set state in case xfs_attr_rmtval_remove returns -EAGAIN */ - dac->dela_state =XFS_DAS_RM_LBLK ; + attr->xattri_dela_state =XFS_DAS_RM_LBLK ;if (args->rmtblkno) { - error =xfs_attr_rmtval_remove (dac); + error =xfs_attr_rmtval_remove (attr);if (error == -EAGAIN )trace_xfs_attr_set_iter_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);if (error)return error; - dac->dela_state =XFS_DAS_RD_LEAF ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); + attr->xattri_dela_state =XFS_DAS_RD_LEAF ; +trace_xfs_attr_set_iter_return (attr->xattri_dela_state, + args->dp);return -EAGAIN ; }
@@ -491,7 +494,7 @@ xfs_attr_set_iter(* state. */ if (args->rmtblkno >0 ) { - error =xfs_attr_rmtval_find_space (dac); + error =xfs_attr_rmtval_find_space (attr);if (error)return error; }
@@ -504,14 +507,14 @@ xfs_attr_set_iter(* after we create the attribute so that we don't overflow the * maximum size of a transaction and/or hit a deadlock. */ - dac->dela_state =XFS_DAS_ALLOC_NODE ; + attr->xattri_dela_state =XFS_DAS_ALLOC_NODE ;if (args->rmtblkno >0 ) { -if (dac->blkcnt >0 ) { - error =xfs_attr_rmtval_set_blk (dac); +if (attr->xattri_blkcnt >0 ) { + error =xfs_attr_rmtval_set_blk (attr);if (error)return error;trace_xfs_attr_set_iter_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);return -EAGAIN ; }
@@ -547,8 +550,8 @@ xfs_attr_set_iter(* Commit the flag value change and start the next trans * in series */ - dac->dela_state =XFS_DAS_FLIP_NFLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, + attr->xattri_dela_state =XFS_DAS_FLIP_NFLAG ; +trace_xfs_attr_set_iter_return (attr->xattri_dela_state, args->dp);return -EAGAIN ; }
@@ -568,18 +571,19 @@ xfs_attr_set_iter( fallthrough;case XFS_DAS_RM_NBLK :/* Set state in case xfs_attr_rmtval_remove returns -EAGAIN */ - dac->dela_state =XFS_DAS_RM_NBLK ; + attr->xattri_dela_state =XFS_DAS_RM_NBLK ;if (args->rmtblkno) { - error =xfs_attr_rmtval_remove (dac); + error =xfs_attr_rmtval_remove (attr);if (error == -EAGAIN )trace_xfs_attr_set_iter_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);if (error)return error; - dac->dela_state =XFS_DAS_CLR_FLAG ; -trace_xfs_attr_set_iter_return (dac->dela_state, args->dp); + attr->xattri_dela_state =XFS_DAS_CLR_FLAG ; +trace_xfs_attr_set_iter_return (attr->xattri_dela_state, + args->dp);return -EAGAIN ; }
@@ -589,7 +593,7 @@ xfs_attr_set_iter(* The last state for node format. Look up the old attr and * remove it. */ - error =xfs_attr_node_addname_clear_incomplete (dac); + error =xfs_attr_node_addname_clear_incomplete (attr);break ;default :ASSERT (0 );
@@ -786,7 +790,7 @@ xfs_attr_item_init( new =kmem_zalloc (sizeof (struct xfs_attr_item ),KM_NOFS ); new->xattri_op_flags = op_flags; - new->xattri_dac.da_args = args; + new->xattri_da_args = args; *attr = new;return 0 ;
@@ -1099,16 +1103,16 @@ xfs_attr_node_hasname( STATIC intxfs_attr_node_addname_find_attr ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;int retval;/* * Search to see if name already exists, and get back a pointer * to where it should go. */ - retval =xfs_attr_node_hasname (args, &dac->da_state); + retval =xfs_attr_node_hasname (args, &attr->xattri_da_state);if (retval != -ENOATTR && retval != -EEXIST )return retval;
@@ -1136,8 +1140,8 @@ xfs_attr_node_addname_find_attr(return 0 ; error: -if (dac->da_state) -xfs_da_state_free (dac->da_state); +if (attr->xattri_da_state) +xfs_da_state_free (attr->xattri_da_state);return retval; }
@@ -1158,10 +1162,10 @@ xfs_attr_node_addname_find_attr(*/ STATIC intxfs_attr_node_addname ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; -struct xfs_da_state *state = dac->da_state; +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_da_state *state = attr->xattri_da_state;struct xfs_da_state_blk *blk;int error;
@@ -1192,7 +1196,7 @@ xfs_attr_node_addname(* this point. */ trace_xfs_attr_node_addname_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);return -EAGAIN ; }
@@ -1221,9 +1225,9 @@ xfs_attr_node_addname( STATIC intxfs_attr_node_addname_clear_incomplete ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;struct xfs_da_state *state =NULL ;struct xfs_mount *mp = args->dp->i_mount;int retval =0 ;
@@ -1327,10 +1331,10 @@ xfs_attr_leaf_mark_incomplete(*/ STATIC int xfs_attr_node_removename_setup( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; -struct xfs_da_state **state = &dac->da_state; +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_da_state **state = &attr->xattri_da_state;int error; error =xfs_attr_node_hasname (args, state);
@@ -1389,16 +1393,16 @@ xfs_attr_node_removename(*/ int xfs_attr_remove_iter ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; -struct xfs_da_state *state = dac->da_state; +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_da_state *state = attr->xattri_da_state;int retval, error =0 ;struct xfs_inode *dp = args->dp;trace_xfs_attr_node_removename (args); -switch (dac->dela_state) { +switch (attr->xattri_dela_state) {case XFS_DAS_UNINIT :if (!xfs_inode_hasattr (dp))return -ENOATTR ;
@@ -1417,16 +1421,16 @@ xfs_attr_remove_iter(* Node format may require transaction rolls. Set up the * state context and fall into the state machine. */ -if (!dac->da_state) { - error =xfs_attr_node_removename_setup (dac); +if (!attr->xattri_da_state) { + error =xfs_attr_node_removename_setup (attr);if (error)return error; - state = dac->da_state; + state = attr->xattri_da_state; } fallthrough;case XFS_DAS_RMTBLK : - dac->dela_state =XFS_DAS_RMTBLK ; + attr->xattri_dela_state =XFS_DAS_RMTBLK ;/* * If there is an out-of-line value, de-allocate the blocks.
@@ -1439,10 +1443,10 @@ xfs_attr_remove_iter(* May return -EAGAIN. Roll and repeat until all remote * blocks are removed. */ - error =xfs_attr_rmtval_remove (dac); + error =xfs_attr_rmtval_remove (attr);if (error == -EAGAIN ) {trace_xfs_attr_remove_iter_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);return error; }else if (error) { goto out;
@@ -1457,8 +1461,10 @@ xfs_attr_remove_iter( error =xfs_attr_refillstate (state);if (error) goto out; - dac->dela_state =XFS_DAS_RM_NAME ; -trace_xfs_attr_remove_iter_return (dac->dela_state, args->dp); + + attr->xattri_dela_state =XFS_DAS_RM_NAME ; +trace_xfs_attr_remove_iter_return ( + attr->xattri_dela_state, args->dp);return -EAGAIN ; }
@@ -1468,7 +1474,7 @@ xfs_attr_remove_iter(* If we came here fresh from a transaction roll, reattach all * the buffers to the current transaction. */ -if (dac->dela_state ==XFS_DAS_RM_NAME ) { +if (attr->xattri_dela_state ==XFS_DAS_RM_NAME ) { error =xfs_attr_refillstate (state);if (error) goto out;
@@ -1485,9 +1491,9 @@ xfs_attr_remove_iter(if (error) goto out; - dac->dela_state =XFS_DAS_RM_SHRINK ; + attr->xattri_dela_state =XFS_DAS_RM_SHRINK ;trace_xfs_attr_remove_iter_return ( - dac->dela_state, args->dp); + attr->xattri_dela_state, args->dp);return -EAGAIN ; }
fs/xfs/libxfs/xfs_attr.h +17 -23
@@ -430,7 +430,7 @@ struct xfs_attr_list_context {*/ /* -* Enum values for xfs_delattr_context.da_state +* Enum values for xfs_attr_item.xattri_da_state * * These values are used by delayed attribute operations to keep track of where * they were before they returned -EAGAIN. A return code of -EAGAIN signals the
@@ -455,7 +455,7 @@ enum xfs_delattr_state { };/* -* Defines for xfs_delattr_context.flags +* Defines for xfs_attr_item.xattri_flags */ #define XFS_DAC_LEAF_ADDNAME_INIT 0x01/* xfs_attr_leaf_addname init*/ #define XFS_DAC_DELAYED_OP_INIT 0x02/* delayed operations init*/
@@ -463,32 +463,25 @@ enum xfs_delattr_state {/* * Context used for keeping track of delayed attribute operations */ -struct xfs_delattr_context { -struct xfs_da_args *da_args; +struct xfs_attr_item { +struct xfs_da_args *xattri_da_args;/* * Used by xfs_attr_set to hold a leaf buffer across a transaction roll */ -struct xfs_buf *leaf_bp; +struct xfs_buf *xattri_leaf_bp;/* Used in xfs_attr_rmtval_set_blk to roll through allocating blocks */ -struct xfs_bmbt_irec map; -xfs_dablk_t lblkno; -int blkcnt; +struct xfs_bmbt_irec xattri_map; +xfs_dablk_t xattri_lblkno; +int xattri_blkcnt;/* Used in xfs_attr_node_removename to roll through removing blocks */ -struct xfs_da_state *da_state; +struct xfs_da_state *xattri_da_state;/* Used to keep track of current state of delayed operation */ -unsigned int flags; -enum xfs_delattr_state dela_state; -}; - -/* -* List of attrs to commit later. -*/ -struct xfs_attr_item { -struct xfs_delattr_context xattri_dac; +unsigned int xattri_flags; +enum xfs_delattr_state xattri_dela_state;/* * Indicates if the attr operation is a set or a remove
@@ -496,7 +489,10 @@ struct xfs_attr_item {*/ unsigned int xattri_op_flags; -/* used to log this item to an intent */ +/* +* used to log this item to an intent containing a list of attrs to +* commit later +*/ struct list_head xattri_list; };
@@ -516,11 +512,9 @@ bool xfs_attr_is_leaf(struct xfs_inode *ip);int xfs_attr_get_ilocked (struct xfs_da_args *args);int xfs_attr_get (struct xfs_da_args *args);int xfs_attr_set (struct xfs_da_args *args); -int xfs_attr_set_iter (struct xfs_delattr_context *dac); -int xfs_attr_remove_iter (struct xfs_delattr_context *dac); +int xfs_attr_set_iter (struct xfs_attr_item *attr); +int xfs_attr_remove_iter (struct xfs_attr_item *attr);bool xfs_attr_namecheck (const void *name,size_t length); -void xfs_delattr_context_init (struct xfs_delattr_context *dac, -struct xfs_da_args *args);int xfs_attr_calc_size (struct xfs_da_args *args,int *local);int xfs_attr_set_deferred (struct xfs_da_args *args);int xfs_attr_remove_deferred (struct xfs_da_args *args);
fs/xfs/libxfs/xfs_attr_remote.c +19 -17
@@ -568,14 +568,14 @@ xfs_attr_rmtval_stale(*/ int xfs_attr_rmtval_find_space ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; -struct xfs_bmbt_irec *map = &dac->map; +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_bmbt_irec *map = &attr->xattri_map;int error; - dac->lblkno =0 ; - dac->blkcnt =0 ; + attr->xattri_lblkno =0 ; + attr->xattri_blkcnt =0 ; args->rmtblkcnt =0 ; args->rmtblkno =0 ;memset (map,0 ,sizeof (struct xfs_bmbt_irec ));
@@ -584,8 +584,8 @@ xfs_attr_rmtval_find_space(if (error)return error; - dac->blkcnt = args->rmtblkcnt; - dac->lblkno = args->rmtblkno; + attr->xattri_blkcnt = args->rmtblkcnt; + attr->xattri_lblkno = args->rmtblkno;return 0 ; }
@@ -598,17 +598,18 @@ xfs_attr_rmtval_find_space(*/ int xfs_attr_rmtval_set_blk ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;struct xfs_inode *dp = args->dp; -struct xfs_bmbt_irec *map = &dac->map; +struct xfs_bmbt_irec *map = &attr->xattri_map;int nmap;int error; nmap =1 ; - error =xfs_bmapi_write (args->trans, dp, (xfs_fileoff_t )dac->lblkno, - dac->blkcnt,XFS_BMAPI_ATTRFORK , args->total, + error =xfs_bmapi_write (args->trans, dp, + (xfs_fileoff_t )attr->xattri_lblkno, + attr->xattri_blkcnt,XFS_BMAPI_ATTRFORK , args->total, map, &nmap);if (error)return error;
@@ -618,8 +619,8 @@ xfs_attr_rmtval_set_blk( (map->br_startblock != HOLESTARTBLOCK));/* roll attribute extent map forwards */ - dac->lblkno += map->br_blockcount; - dac->blkcnt -= map->br_blockcount; + attr->xattri_lblkno += map->br_blockcount; + attr->xattri_blkcnt -= map->br_blockcount;return 0 ; }
@@ -673,9 +674,9 @@ xfs_attr_rmtval_invalidate(*/ int xfs_attr_rmtval_remove ( -struct xfs_delattr_context *dac) +struct xfs_attr_item *attr) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;int error, done;/*
@@ -695,7 +696,8 @@ xfs_attr_rmtval_remove(* the parent */ if (!done) { -trace_xfs_attr_rmtval_remove_return (dac->dela_state, args->dp); +trace_xfs_attr_rmtval_remove_return (attr->xattri_dela_state, + args->dp);return -EAGAIN ; }
fs/xfs/libxfs/xfs_attr_remote.h +3 -3
@@ -12,9 +12,9 @@ int xfs_attr_rmtval_get(struct xfs_da_args *args);int xfs_attr_rmtval_stale (struct xfs_inode *ip,struct xfs_bmbt_irec *map,xfs_buf_flags_t incore_flags);int xfs_attr_rmtval_invalidate (struct xfs_da_args *args); -int xfs_attr_rmtval_remove (struct xfs_delattr_context *dac); +int xfs_attr_rmtval_remove (struct xfs_attr_item *attr);int xfs_attr_rmt_find_hole (struct xfs_da_args *args);int xfs_attr_rmtval_set_value (struct xfs_da_args *args); -int xfs_attr_rmtval_set_blk (struct xfs_delattr_context *dac); -int xfs_attr_rmtval_find_space (struct xfs_delattr_context *dac); +int xfs_attr_rmtval_set_blk (struct xfs_attr_item *attr); +int xfs_attr_rmtval_find_space (struct xfs_attr_item *attr);#endif /* __XFS_ATTR_REMOTE_H__ */
fs/xfs/xfs_attr_item.c +18 -20
@@ -284,11 +284,11 @@ xfs_attrd_item_release(*/ STATIC intxfs_trans_attr_finish_update ( -struct xfs_delattr_context *dac, +struct xfs_attr_item *attr,struct xfs_attrd_log_item *attrdp,uint32_t op_flags) { -struct xfs_da_args *args = dac->da_args; +struct xfs_da_args *args = attr->xattri_da_args;unsigned int op = op_flags &XFS_ATTR_OP_FLAGS_TYPE_MASK ;int error;
@@ -300,11 +300,11 @@ xfs_trans_attr_finish_update(switch (op) {case XFS_ATTR_OP_FLAGS_SET : - error =xfs_attr_set_iter (dac); + error =xfs_attr_set_iter (attr);break ;case XFS_ATTR_OP_FLAGS_REMOVE :ASSERT (XFS_IFORK_Q (args->dp)); - error =xfs_attr_remove_iter (dac); + error =xfs_attr_remove_iter (attr);break ;default : error = -EFSCORRUPTED ;
@@ -348,16 +348,16 @@ xfs_attr_log_item(* structure with fields from this xfs_attr_item */ attrp = &attrip->attri_format; - attrp->alfi_ino = attr->xattri_dac.da_args->dp->i_ino; + attrp->alfi_ino = attr->xattri_da_args->dp->i_ino; attrp->alfi_op_flags = attr->xattri_op_flags; - attrp->alfi_value_len = attr->xattri_dac.da_args->valuelen; - attrp->alfi_name_len = attr->xattri_dac.da_args->namelen; - attrp->alfi_attr_flags = attr->xattri_dac.da_args->attr_filter; - - attrip->attri_name = (void *)attr->xattri_dac.da_args->name; - attrip->attri_value = attr->xattri_dac.da_args->value; - attrip->attri_name_len = attr->xattri_dac.da_args->namelen; - attrip->attri_value_len = attr->xattri_dac.da_args->valuelen; + attrp->alfi_value_len = attr->xattri_da_args->valuelen; + attrp->alfi_name_len = attr->xattri_da_args->namelen; + attrp->alfi_attr_flags = attr->xattri_da_args->attr_filter; + + attrip->attri_name = (void *)attr->xattri_da_args->name; + attrip->attri_value = attr->xattri_da_args->value; + attrip->attri_name_len = attr->xattri_da_args->namelen; + attrip->attri_value_len = attr->xattri_da_args->valuelen; }/* Get an ATTRI. */
@@ -398,10 +398,8 @@ xfs_attr_finish_item(struct xfs_attr_item *attr;struct xfs_attrd_log_item *done_item =NULL ;int error; -struct xfs_delattr_context *dac; attr = container_of(item, struct xfs_attr_item, xattri_list); - dac = &attr->xattri_dac;if (done) done_item =ATTRD_ITEM (done);
@@ -409,9 +407,9 @@ xfs_attr_finish_item(* Always reset trans after EAGAIN cycle * since the transaction is new */ - dac->da_args->trans = tp; + attr->xattri_da_args->trans = tp; - error =xfs_trans_attr_finish_update (dac, done_item, + error =xfs_trans_attr_finish_update (attr, done_item, attr->xattri_op_flags);if (error != -EAGAIN )kmem_free (attr);
@@ -585,7 +583,7 @@ xfs_attri_item_recover( sizeof(struct xfs_da_args), KM_NOFS); args = (struct xfs_da_args *)(attr +1 ); - attr->xattri_dac.da_args = args; + attr->xattri_da_args = args; attr->xattri_op_flags = attrp->alfi_op_flags; args->dp = ip;
@@ -622,7 +620,7 @@ xfs_attri_item_recover(xfs_ilock (ip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,0 ); - ret =xfs_trans_attr_finish_update (&attr->xattri_dac, done_item, + ret =xfs_trans_attr_finish_update (attr, done_item, attrp->alfi_op_flags);if (ret == -EAGAIN ) {/* There's more work to do, so add it to this transaction */
@@ -636,7 +634,7 @@ xfs_attri_item_recover( } error =xfs_defer_ops_capture_and_commit (tp, ip, capture_list, - attr->xattri_dac.leaf_bp); + attr->xattri_leaf_bp); out_unlock:xfs_iunlock (ip,XFS_ILOCK_EXCL );
Signature · 1 line
2.25.1
PATCH v24 11/11 xfs: Add helper function xfs_attr_leaf_addname
In reply to Allison HendersonThis patch adds a helper function xfs_attr_leaf_addname. While this does help to break down xfs_attr_set_iter, it does also hoist out some of the state management. This patch has been moved to the end of the clean up series for further discussion.
--- fs/xfs/libxfs/xfs_attr.c | 110 +++++++++++++++++++++------------------ fs/xfs/xfs_trace.h | 1 + 2 files changed, 61 insertions(+), 50 deletions(-)
fs/xfs/libxfs/xfs_attr.c +60 -50
@@ -284,6 +284,65 @@ xfs_attr_sf_addname(return -EAGAIN ; } +STATIC int +xfs_attr_leaf_addname ( +struct xfs_attr_item *attr) +{ +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_inode *dp = args->dp; +int error; + +if (xfs_attr_is_leaf (dp)) { + error =xfs_attr_leaf_try_add (args, attr->xattri_leaf_bp); +if (error == -ENOSPC ) { + error =xfs_attr3_leaf_to_node (args); +if (error) +return error; + +/* +* Finish any deferred work items and roll the +* transaction once more. The goal here is to call +* node_addname with the inode and transaction in the +* same state (inode locked and joined, transaction +* clean) no matter how we got to this step. +* +* At this point, we are still in XFS_DAS_UNINIT, but +* when we come back, we'll be a node, so we'll fall +* down into the node handling code below +*/ +trace_xfs_attr_set_iter_return ( + attr->xattri_dela_state, args->dp); +return -EAGAIN ; + } + +if (error) +return error; + + attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; + }else { + error =xfs_attr_node_addname_find_attr (attr); +if (error) +return error; + + error =xfs_attr_node_addname (attr); +if (error) +return error; + +/* +* If addname was successful, and we dont need to alloc or +* remove anymore blks, we're done. +*/ +if (!args->rmtblkno && + !(args->op_flags &XFS_DA_OP_RENAME )) +return 0 ; + + attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; + } + +trace_xfs_attr_leaf_addname_return (attr->xattri_dela_state, args->dp); +return -EAGAIN ; +} +/* * Set the attribute specified in @args. * This routine is meant to function as a delayed operation, and may return
@@ -319,57 +378,8 @@ xfs_attr_set_iter( attr->xattri_leaf_bp =NULL ; } -if (xfs_attr_is_leaf (dp)) { - error =xfs_attr_leaf_try_add (args, - attr->xattri_leaf_bp); -if (error == -ENOSPC ) { - error =xfs_attr3_leaf_to_node (args); -if (error) -return error; - -/* -* Finish any deferred work items and roll the -* transaction once more. The goal here is to -* call node_addname with the inode and -* transaction in the same state (inode locked -* and joined, transaction clean) no matter how -* we got to this step. -* -* At this point, we are still in -* XFS_DAS_UNINIT, but when we come back, we'll -* be a node, so we'll fall down into the node -* handling code below -*/ -trace_xfs_attr_set_iter_return ( - attr->xattri_dela_state, args->dp); -return -EAGAIN ; - }else if (error) { -return error; - } - - attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; - }else { - error =xfs_attr_node_addname_find_attr (attr); -if (error) -return error; +return xfs_attr_leaf_addname (attr); - error =xfs_attr_node_addname (attr); -if (error) -return error; - -/* -* If addname was successful, and we dont need to alloc -* or remove anymore blks, we're done. -*/ -if (!args->rmtblkno && - !(args->op_flags &XFS_DA_OP_RENAME )) -return 0 ; - - attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; - } -trace_xfs_attr_set_iter_return (attr->xattri_dela_state, - args->dp); -return -EAGAIN ;case XFS_DAS_FOUND_LBLK :/* * If there was an out-of-line value, allocate the blocks we
fs/xfs/xfs_trace.h +1 -0
@@ -4132,6 +4132,7 @@ DEFINE_EVENT(xfs_das_state_class, name, \ TP_ARGS(das, ip))DEFINE_DAS_STATE_EVENT (xfs_attr_sf_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_set_iter_return); +DEFINE_DAS_STATE_EVENT (xfs_attr_leaf_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_node_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_remove_iter_return);DEFINE_DAS_STATE_EVENT (xfs_attr_rmtval_remove_return);
Signature · 1 line
2.25.1
Re: [PATCH v24 11/11] xfs: Add helper function xfs_attr_leaf_addname
Quoted 5 lines · Allison Henderson
This patch adds a helper function
xfs_attr_leaf_addname. While this does help to break downxfs_attr_set_iter, it does also hoist out some of the state management. This patch has been moved to the end of the clean up series for further discussion.
Looks good to me.
Quoted 149 lines
Suggested-by Darrick J. Wong <[email protected]>Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 110 +++++++++++++++++++++------------------ fs/xfs/xfs_trace.h | 1 + 2 files changed, 61 insertions(+), 50 deletions(-)fs/xfs/libxfs/xfs_attr.c +60
-50@@ -284,6 +284,65 @@ xfs_attr_sf_addname(return -EAGAIN ; } +STATIC int +xfs_attr_leaf_addname ( +struct xfs_attr_item *attr) +{ +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_inode *dp = args->dp; +int error; + +if (xfs_attr_is_leaf (dp)) { + error =xfs_attr_leaf_try_add (args, attr->xattri_leaf_bp); +if (error == -ENOSPC ) { + error =xfs_attr3_leaf_to_node (args); +if (error) +return error; + +/* +* Finish any deferred work items and roll the +* transaction once more. The goal here is to call +* node_addname with the inode and transaction in the +* same state (inode locked and joined, transaction +* clean) no matter how we got to this step. +* +* At this point, we are still in XFS_DAS_UNINIT, but +* when we come back, we'll be a node, so we'll fall +* down into the node handling code below +*/ +trace_xfs_attr_set_iter_return ( + attr->xattri_dela_state, args->dp); +return -EAGAIN ; + } + +if (error) +return error; + + attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; + }else { + error =xfs_attr_node_addname_find_attr (attr); +if (error) +return error; + + error =xfs_attr_node_addname (attr); +if (error) +return error; + +/* +* If addname was successful, and we dont need to alloc or +* remove anymore blks, we're done. +*/ +if (!args->rmtblkno && + !(args->op_flags &XFS_DA_OP_RENAME )) +return 0 ; + + attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; + } + +trace_xfs_attr_leaf_addname_return (attr->xattri_dela_state, args->dp); +return -EAGAIN ; +} +/* * Set the attribute specified in @args. * This routine is meant to function as a delayed operation, and may return @@ -319,57 +378,8 @@ xfs_attr_set_iter( attr->xattri_leaf_bp =NULL ; } -if (xfs_attr_is_leaf (dp)) { - error =xfs_attr_leaf_try_add (args, - attr->xattri_leaf_bp); -if (error == -ENOSPC ) { - error =xfs_attr3_leaf_to_node (args); -if (error) -return error; - -/* -* Finish any deferred work items and roll the -* transaction once more. The goal here is to -* call node_addname with the inode and -* transaction in the same state (inode locked -* and joined, transaction clean) no matter how -* we got to this step. -* -* At this point, we are still in -* XFS_DAS_UNINIT, but when we come back, we'll -* be a node, so we'll fall down into the node -* handling code below -*/ -trace_xfs_attr_set_iter_return ( - attr->xattri_dela_state, args->dp); -return -EAGAIN ; - }else if (error) { -return error; - } - - attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; - }else { - error =xfs_attr_node_addname_find_attr (attr); -if (error) -return error; +return xfs_attr_leaf_addname (attr); - error =xfs_attr_node_addname (attr); -if (error) -return error; - -/* -* If addname was successful, and we dont need to alloc -* or remove anymore blks, we're done. -*/ -if (!args->rmtblkno && - !(args->op_flags &XFS_DA_OP_RENAME )) -return 0 ; - - attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; - } -trace_xfs_attr_set_iter_return (attr->xattri_dela_state, - args->dp); -return -EAGAIN ;case XFS_DAS_FOUND_LBLK :/* * If there was an out-of-line value, allocate the blocks we fs/xfs/xfs_trace.h +1
-0@@ -4132,6 +4132,7 @@ DEFINE_EVENT(xfs_das_state_class, name, \ TP_ARGS(das, ip))DEFINE_DAS_STATE_EVENT (xfs_attr_sf_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_set_iter_return); +DEFINE_DAS_STATE_EVENT (xfs_attr_leaf_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_node_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_remove_iter_return);DEFINE_DAS_STATE_EVENT (xfs_attr_rmtval_remove_return);
Signature · 1 line
chandan
Re: [PATCH v24 11/11] xfs: Add helper function xfs_attr_leaf_addname
Quoted 10 lines · Chandan Babu R
Quoted 5 lines · Allison Henderson
This patch adds a helper function
xfs_attr_leaf_addname. While this does help to break downxfs_attr_set_iter, it does also hoist out some of the state management. This patch has been moved to the end of the clean up series for further discussion.Looks good to me.
Reviewed-by Chandan Babu R <[email protected]>
Great, thankyou!
Allison
Quoted 152 lines
Quoted 150 lines
Suggested-by Darrick J. Wong <[email protected]>Signed-off-by Allison Henderson <[email protected]>Reviewed-by Darrick J. Wong <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 110 +++++++++++++++++++++------------------ fs/xfs/xfs_trace.h | 1 + 2 files changed, 61 insertions(+), 50 deletions(-)fs/xfs/libxfs/xfs_attr.c +60
-50@@ -284,6 +284,65 @@ xfs_attr_sf_addname(return -EAGAIN ; } +STATIC int +xfs_attr_leaf_addname ( +struct xfs_attr_item *attr) +{ +struct xfs_da_args *args = attr->xattri_da_args; +struct xfs_inode *dp = args->dp; +int error; + +if (xfs_attr_is_leaf (dp)) { + error =xfs_attr_leaf_try_add (args, attr->xattri_leaf_bp); +if (error == -ENOSPC ) { + error =xfs_attr3_leaf_to_node (args); +if (error) +return error; + +/* +* Finish any deferred work items and roll the +* transaction once more. The goal here is to call +* node_addname with the inode and transaction in the +* same state (inode locked and joined, transaction +* clean) no matter how we got to this step. +* +* At this point, we are still in XFS_DAS_UNINIT, but +* when we come back, we'll be a node, so we'll fall +* down into the node handling code below +*/ +trace_xfs_attr_set_iter_return ( + attr->xattri_dela_state, args->dp); +return -EAGAIN ; + } + +if (error) +return error; + + attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; + }else { + error =xfs_attr_node_addname_find_attr (attr); +if (error) +return error; + + error =xfs_attr_node_addname (attr); +if (error) +return error; + +/* +* If addname was successful, and we dont need to alloc or +* remove anymore blks, we're done. +*/ +if (!args->rmtblkno && + !(args->op_flags &XFS_DA_OP_RENAME )) +return 0 ; + + attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; + } + +trace_xfs_attr_leaf_addname_return (attr->xattri_dela_state, args->dp); +return -EAGAIN ; +} +/* * Set the attribute specified in @args. * This routine is meant to function as a delayed operation, and may return @@ -319,57 +378,8 @@ xfs_attr_set_iter( attr->xattri_leaf_bp =NULL ; } -if (xfs_attr_is_leaf (dp)) { - error =xfs_attr_leaf_try_add (args, - attr->xattri_leaf_bp); -if (error == -ENOSPC ) { - error =xfs_attr3_leaf_to_node (args); -if (error) -return error; - -/* -* Finish any deferred work items and roll the -* transaction once more. The goal here is to -* call node_addname with the inode and -* transaction in the same state (inode locked -* and joined, transaction clean) no matter how -* we got to this step. -* -* At this point, we are still in -* XFS_DAS_UNINIT, but when we come back, we'll -* be a node, so we'll fall down into the node -* handling code below -*/ -trace_xfs_attr_set_iter_return ( - attr->xattri_dela_state, args->dp); -return -EAGAIN ; - }else if (error) { -return error; - } - - attr->xattri_dela_state =XFS_DAS_FOUND_LBLK ; - }else { - error =xfs_attr_node_addname_find_attr (attr); -if (error) -return error; +return xfs_attr_leaf_addname (attr); - error =xfs_attr_node_addname (attr); -if (error) -return error; - -/* -* If addname was successful, and we dont need to alloc -* or remove anymore blks, we're done. -*/ -if (!args->rmtblkno && - !(args->op_flags &XFS_DA_OP_RENAME )) -return 0 ; - - attr->xattri_dela_state =XFS_DAS_FOUND_NBLK ; - } -trace_xfs_attr_set_iter_return (attr->xattri_dela_state, - args->dp); -return -EAGAIN ;case XFS_DAS_FOUND_LBLK :/* * If there was an out-of-line value, allocate the blocks we fs/xfs/xfs_trace.h +1
-0@@ -4132,6 +4132,7 @@ DEFINE_EVENT(xfs_das_state_class, name, \ TP_ARGS(das, ip))DEFINE_DAS_STATE_EVENT (xfs_attr_sf_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_set_iter_return); +DEFINE_DAS_STATE_EVENT (xfs_attr_leaf_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_node_addname_return);DEFINE_DAS_STATE_EVENT (xfs_attr_remove_iter_return);DEFINE_DAS_STATE_EVENT (xfs_attr_rmtval_remove_return);
Quoted 12 lines · Allison Henderson
Hi all,
This set is a subset of a larger series parent pointers. Delayed attributes allow attribute operations (set and remove) to be logged and committed in the same way that other delayed operations do. This allows more complex operations (like parent pointers) to be broken up into multiple smaller transactions. To do this, the existing attr operations must be modified to operate as a delayed operation. This means that they cannot roll, commit, or finish transactions. Instead, they return
-EAGAINto allow the calling function to handle the transaction. In this series, we focus on only the delayed attribute portion. We will introduce parent pointers in a later set.
Quoted 3 lines
The set as a whole is a bit much to digest at once, so I usually send out the smaller sub series to reduce reviewer burn out. But the entire extended series is visible through the included github links.
Ok, so like I did with Darrick's deferred inactivation series, the first thing I'm doing here is throwing this patchset at scalability/performance worklaods and finding out what is different.
I've merged this series with 5.14 + xfs/for-next + xfs-cil-scale and then run some tests on it. First up is fsmark creating zero length files w/ 64 byte xattrs. This should stress only shortform attribute manipulations.
I have not enabled delayed attributes yet (i.e./sys/fs/xfs/debug/larp = 0)
First thing I notice is the transaction commit rate during create is up around 900k/s, so we are doing 3 transactions per inode - 1 for create, 2 for attributes. That looks like a regression - existing shortform attribute creation only takes a single transaction commit, so this workload prior to this patchset only ran at 600k commits/s.
Note that hte only reason I'm getting 900k transactions/s is the CIL scalability patchset - without that the system tops out at ~800k transactions/s and so this would be a significant performance regression (20%) vs the current xfs/for-next code.
Essentially, this looks like we are doing an extra transaction commit to defer the creation of the attribute, then doing another transaction to actually modify the attribute. i.e.:
- 11.04% xfs_attr_set
- 8.70% xfs_trans_commit
- 8.69% __xfs_trans_commit
- 5.10% xfs_defer_finish_noroll
- 3.74% xfs_defer_trans_roll
- 3.57% xfs_trans_roll
- 3.13% __xfs_trans_commit
- 3.01% xlog_cil_commit
0.66% down_read
0.63% xfs_log_ticket_regrant
- 1.16% xfs_attr_finish_item
- 1.06% xfs_trans_attr_finish_update
- 1.03% xfs_attr_set_iter
- 1.01% xfs_attr_sf_addname
- 0.99% xfs_attr_try_sf_addname
- 0.61% xfs_attr_shortform_addname 0.55% xfs_attr_shortform_add
AFAICT, for non-delayed attributes, this first transaction commit logs the inode but does not create intent or intent done items (returns NULL for both operations), so just rolls and runs the ->finish_item. So it would seem that the first transaction just changes the inode timestamps and does nothing else.
Firstly, this means the inode timestamp change is not atomic w.r.t. the attribute change the timestamp change relates to and it's essentially new overhead for the non-delayed path.
Looking at the unlink path, I see the same thing - there's an extra transaction for the attr remove path, the same as the attr set path. This drives the unlink path to 1.1 million transaction commits/sec instead of 800k/s, so it's likely that there's a substantial performance regression here on a kernel without the CIL scalability patchset.
IOWs, there's significant behavioural changes with the non-delayed logging version of this patchset, both in terms of performance and the atomicity of changes that appear in the journal and hence recovery behaviour.
At this point I have to ask: why are we trying to retain the "old" way of doing things (even for testing) if it is substantially changing behaviour and on-disk journal contents for attribute modifications?
So, lets turn on delayed logging:
$ sudo sh -c 'echo 1 > /sys/fs/xfs/debug/larp' $ ~/tests/fsmark-50-test-xfs.sh -t 16 -X 64 -d /dev/mapper/fast -- -l size=2000m -d agcount=67 QUOTA= MKFSOPTS= -l size=2000m -d agcount=67 DEV=/dev/mapper/fast THREADS=16 .....
Message from syslogd@test4 at Aug 31 09:12:55 ...
kernel:[ 2342.737931] XFS: Assertion failed: !test_bit(XFS_LI_DIRTY, &lip->li_flags), file: fs/xfs/xfs_trans.c, line: 652
Instant assert fail and the machine locks up hard.
Actually, now that I reproduce it with a full console trace (which is terribly interleaved and almost impossible to read) there's bad stuff all over the place. Null pointer dereferences in xlog_cil_commit, "sleeping in atomic" failures, and the assert failure above.
Ok, run a single thread, and...
[ 84.119162] BUG: kernel NULL pointer dereference, address: 000000000000000d [ 84.123541] #PF: supervisor write access in kernel mode [ 84.126028] #PF: error_code(0x0002) - not-present page [ 84.127312] PGD 0 P4D 0 [ 84.127966] Oops: 0002 [#1] PREEMPT SMP [ 84.128960] CPU: 8 PID: 5139 Comm: fs_mark Not tainted 5.14.0-dgc+ #552 [ 84.130632] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-1 04/01/2014 [ 84.132723] RIP: 0010:xlog_prepare_iovec+0x59/0xe0 [ 84.133951] Code: 4c 89 f8 4c 29 e0 48 c1 f8 04 48 39 d0 7d 76 4d 8d 67 10 8b 43 34 8d 50 0c 83 e2 07 74 0c 83 c0 0b 83 c8 07 83 e8 0b 89 43 34 <45> 89 74 24 0c 48 63 43 34 48 03 43 28 49 89 04 24 c7 40 08 69 00 [ 84.137980] RSP: 0018:ffffc900021d7838 EFLAGS: 00010202 [ 84.138968] RAX: 0000000000000004 RBX: ffff888140f3e100 RCX: 0000000000000006 [ 84.140321] RDX: 0000000000000004 RSI: ffffc900021d7880 RDI: ffff888140f3e100 [ 84.141671] RBP: ffffc900021d7868 R08: ffffffff82a8fb88 R09: 000000000000494e [ 84.143069] R10: ffff88823ffd5000 R11: 00000000000319c8 R12: 0000000000000001 [ 84.144447] R13: ffffc900021d7880 R14: 000000000000001b R15: 0000000000000000 [ 84.145825] FS: 00007f51fb33c740(0000) GS:ffff88823bc00000(0000) knlGS:0000000000000000 [ 84.147413] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 84.148535] CR2: 000000000000000d CR3: 00000001473b1003 CR4: 0000000000770ee0 [ 84.149927] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 84.151291] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 84.152331] PKRU: 55555554 [ 84.152732] Call Trace: [ 84.153106] xfs_attri_item_format+0x87/0x230 [ 84.153748] xlog_cil_commit+0x253/0xa00 [ 84.154329] ? kvmalloc_node+0x79/0x80 [ 84.154881] __xfs_trans_commit+0xc1/0x330 [ 84.155490] xfs_trans_roll+0x53/0xe0 [ 84.156027] xfs_defer_trans_roll+0x10d/0x2b0 [ 84.156659] xfs_defer_finish_noroll+0xb1/0x650 [ 84.157312] __xfs_trans_commit+0x143/0x330 [ 84.157918] xfs_trans_commit+0x10/0x20 [ 84.158498] xfs_attr_set+0x41a/0x4e0 [ 84.159030] xfs_xattr_set+0x8d/0xe0 [ 84.159554] __vfs_setxattr+0x6b/0x90 [ 84.160090] __vfs_setxattr_noperm+0x7d/0x1f0 [ 84.160718] __vfs_setxattr_locked+0xdf/0x100 [ 84.161346] vfs_setxattr+0x9b/0x170 [ 84.161862] setxattr+0x110/0x200 [ 84.162346] ? _raw_spin_unlock+0xe/0x20 [ 84.162914] ? __handle_mm_fault+0xc1b/0x16d0 [ 84.163556] ? __might_sleep+0x49/0x80 [ 84.164132] __x64_sys_fsetxattr+0xb1/0xe0 [ 84.164782] do_syscall_64+0x35/0x80
Ok, there's the first failure.
This looks like it's a problem with xfs_attri_item_{size,format} in calculating the number of bytes to log. They use ATTR_NVEC_SIZE() to calculate the number of bytes of copy from the attribute item which rounds up the length to copy to 4 byte aligned values. I'm not sure what this function is calculating:
/* iovec length must be 32-bit aligned */
static inline size_t ATTR_NVEC_SIZE(size_t size)
{
return size == sizeof(int32_t) ? size :
sizeof(int32_t) + round_up(size, sizeof(int32_t));
}It appears to be saying if the size == 4, then return 4, otherwise return 4 + roundup(size)... which leads me to struct xfs_attri_log_format:
struct xfs_attri_log_format {
uint16_t alfi_type; /* attri log item type */
uint16_t alfi_size; /* size of this item */
uint32_t __pad; /* pad to 64 bit aligned */
uint64_t alfi_id; /* attri identifier */
uint64_t alfi_ino; /* the inode for this attr operation */
uint32_t alfi_op_flags; /* marks the op as a set or remove */
uint32_t alfi_name_len; /* attr name length */
uint32_t alfi_value_len; /* attr value length */
uint32_t alfi_attr_flags;/* attr flags */};
I don't see where the extra 4 bytes for the attribute vector size comes from. It's not needed to store the length, so this could oversize the amount of data to be copied from the source buffer by up to 7 bytes?
I can see that it might need rounding with the existing log code (because the formatter is responsible for 32 bit alignment of log vectors), but that goes away with the CIL scalability patchset that always aligns iovecs to 4 byte alignment so the formatters do not need to do that.
Hiding it in a "macro" is not necessary, either - look at how xfs_inode_item_{data,attr}_fork_size handle the rounding up of the local format fork size. They round up the fork byte count to 4 directly, and the format code copies those bytes because xfs_idata_realloc() allocates those bytes.
However, for the attribute buffers, this isn't guaranteed. Look at xfs_xattr_set():
struct xfs_da_args args = {
.dp = XFS_I(inode),
.attr_filter = handler->flags,
.attr_flags = flags,
.name = name,
.namelen = strlen(name),
.value = (void *)value,
.valuelen = size,
};There is no rounding up of the name or value lengths, and these end up directly referenced by the deferred logging via xfs_attr_log_item() and attrip->da_args->...
attrip->attri_name = (void *)attr->xattri_da_args->name;
attrip->attri_value = attr->xattri_da_args->value;
attrip->attri_name_len = attr->xattri_da_args->namelen;
attrip->attri_value_len = attr->xattri_da_args->valuelen;We then pass those pointers directly to xlog_iovec_copy() but with a rounded up length that is longer than the source buffer:
xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ATTR_NAME,
attrip->attri_name,
ATTR_NVEC_SIZE(attrip->attri_name_len));
if (attrip->attri_value_len > 0)
xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ATTR_VALUE,
attrip->attri_value,
ATTR_NVEC_SIZE(attrip->attri_value_len));So while this might not be the source of the above crash, it's certainly a bug that needs fixing.
At this point, I'm just going to hack on the code to make it work, and we can go from there...
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
Quoted 13 lines · Dave Chinner
Quoted 12 lines · Allison Henderson
Hi all,
This set is a subset of a larger series parent pointers. Delayed attributes allow attribute operations (set and remove) to be logged and committed in the same way that other delayed operations do. This allows more complex operations (like parent pointers) to be broken up into multiple smaller transactions. To do this, the existing attr operations must be modified to operate as a delayed operation. This means that they cannot roll, commit, or finish transactions. Instead, they return
-EAGAINto allow the calling function to handle the transaction. In this series, we focus on only the delayed attribute portion. We will introduce parent pointers in a later set.
Quoted 3 lines
Quoted 3 lines
The set as a whole is a bit much to digest at once, so I usually send out the smaller sub series to reduce reviewer burn out. But the entire extended series is visible through the included github links.
Some of this we worked out in the chat last night, but I will echo it here for the archives
Quoted 69 lines
Ok, so like I did with Darrick's deferred inactivation series, the first thing I'm doing here is throwing this patchset at scalability/performance worklaods and finding out what is different.
I've merged this series with 5.14 + xfs/for-next + xfs-cil-scale and then run some tests on it. First up is fsmark creating zero length files w/ 64 byte xattrs. This should stress only shortform attribute manipulations.
I have not enabled delayed attributes yet (i.e.
/sys/fs/xfs/debug/larp= 0)First thing I notice is the transaction commit rate during create is up around 900k/s, so we are doing 3 transactions per inode - 1 for create, 2 for attributes. That looks like a regression - existing shortform attribute creation only takes a single transaction commit, so this workload prior to this patchset only ran at 600k commits/s.
Note that hte only reason I'm getting 900k transactions/s is the CIL scalability patchset - without that the system tops out at ~800k transactions/s and so this would be a significant performance regression (20%) vs the current xfs/for-next code.
Essentially, this looks like we are doing an extra transaction commit to defer the creation of the attribute, then doing another transaction to actually modify the attribute. i.e.:
- 11.04%
xfs_attr_set
- 8.70%xfs_trans_commit
- 8.69%__xfs_trans_commit
- 5.10%xfs_defer_finish_noroll
- 3.74%xfs_defer_trans_roll
- 3.57%xfs_trans_roll
- 3.13%__xfs_trans_commit
- 3.01%xlog_cil_commit
0.66%down_read
0.63%xfs_log_ticket_regrant
- 1.16%xfs_attr_finish_item
- 1.06%xfs_trans_attr_finish_update
- 1.03%xfs_attr_set_iter
- 1.01%xfs_attr_sf_addname
- 0.99%xfs_attr_try_sf_addname
- 0.61%xfs_attr_shortform_addname0.55%xfs_attr_shortform_addAFAICT, for non-delayed attributes, this first transaction commit logs the inode but does not create intent or intent done items (returns NULL for both operations), so just rolls and runs the ->finish_item. So it would seem that the first transaction just changes the inode timestamps and does nothing else.
Firstly, this means the inode timestamp change is not atomic w.r.t. the attribute change the timestamp change relates to and it's essentially new overhead for the non-delayed path.
Looking at the unlink path, I see the same thing - there's an extra transaction for the attr remove path, the same as the attr set path. This drives the unlink path to 1.1 million transaction commits/sec instead of 800k/s, so it's likely that there's a substantial performance regression here on a kernel without the CIL scalability patchset.
IOWs, there's significant behavioural changes with the non-delayed logging version of this patchset, both in terms of performance and the atomicity of changes that appear in the journal and hence recovery behaviour.
Quoted 4 lines
At this point I have to ask: why are we trying to retain the "old" way of doing things (even for testing) if it is substantially changing behaviour and on-disk journal contents for attribute modifications?
Per the chat discussion, we have to keep both methods since sb v4 would not use the new log entries.
Quoted 104 lines
So, lets turn on delayed logging:
$ sudo sh -c 'echo 1 > /sys/fs/xfs/debug/larp' $ ~/tests/fsmark-50-test-xfs.sh -t 16 -X 64 -d /dev/mapper/fast -- -l size=2000m -d agcount=67 QUOTA= MKFSOPTS= -l size=2000m -d agcount=67 DEV=/dev/mapper/fast THREADS=16 .....Message from syslogd@test4 at Aug 31 09:12:55 ...
kernel:[ 2342.737931] XFS: Assertion failed: !test_bit(XFS_LI_DIRTY,&lip->li_flags), file:fs/xfs/xfs_trans.c, line: 652Instant assert fail and the machine locks up hard.
Actually, now that I reproduce it with a full console trace (which is terribly interleaved and almost impossible to read) there's bad stuff all over the place. Null pointer dereferences in
xlog_cil_commit, "sleeping in atomic" failures, and the assert failure above.Ok, run a single thread, and...
[ 84.119162] BUG: kernel NULL pointer dereference, address: 000000000000000d [ 84.123541] #PF: supervisor write access in kernel mode [ 84.126028] #PF: error_code(0x0002) - not-present page [ 84.127312] PGD 0 P4D 0 [ 84.127966] Oops: 0002 [#1] PREEMPT SMP [ 84.128960] CPU: 8 PID: 5139 Comm: fs_mark Not tainted 5.14.0-dgc+ #552 [ 84.130632] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-1 04/01/2014 [ 84.132723] RIP: 0010:xlog_prepare_iovec+0x59/0xe0 [ 84.133951] Code: 4c 89 f8 4c 29 e0 48 c1 f8 04 48 39 d0 7d 76 4d 8d 67 10 8b 43 34 8d 50 0c 83 e2 07 74 0c 83 c0 0b 83 c8 07 83 e8 0b 89 43 34 <45> 89 74 24 0c 48 63 43 34 48 03 43 28 49 89 04 24 c7 40 08 69 00 [ 84.137980] RSP: 0018:ffffc900021d7838 EFLAGS: 00010202 [ 84.138968] RAX: 0000000000000004 RBX: ffff888140f3e100 RCX: 0000000000000006 [ 84.140321] RDX: 0000000000000004 RSI: ffffc900021d7880 RDI: ffff888140f3e100 [ 84.141671] RBP: ffffc900021d7868 R08: ffffffff82a8fb88 R09: 000000000000494e [ 84.143069] R10: ffff88823ffd5000 R11: 00000000000319c8 R12: 0000000000000001 [ 84.144447] R13: ffffc900021d7880 R14: 000000000000001b R15: 0000000000000000 [ 84.145825] FS: 00007f51fb33c740(0000) GS:ffff88823bc00000(0000) knlGS:0000000000000000 [ 84.147413] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 84.148535] CR2: 000000000000000d CR3: 00000001473b1003 CR4: 0000000000770ee0 [ 84.149927] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 84.151291] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 84.152331] PKRU: 55555554 [ 84.152732] Call Trace: [ 84.153106] xfs_attri_item_format+0x87/0x230 [ 84.153748] xlog_cil_commit+0x253/0xa00 [ 84.154329] ? kvmalloc_node+0x79/0x80 [ 84.154881] __xfs_trans_commit+0xc1/0x330 [ 84.155490] xfs_trans_roll+0x53/0xe0 [ 84.156027] xfs_defer_trans_roll+0x10d/0x2b0 [ 84.156659] xfs_defer_finish_noroll+0xb1/0x650 [ 84.157312] __xfs_trans_commit+0x143/0x330 [ 84.157918] xfs_trans_commit+0x10/0x20 [ 84.158498] xfs_attr_set+0x41a/0x4e0 [ 84.159030] xfs_xattr_set+0x8d/0xe0 [ 84.159554] __vfs_setxattr+0x6b/0x90 [ 84.160090] __vfs_setxattr_noperm+0x7d/0x1f0 [ 84.160718] __vfs_setxattr_locked+0xdf/0x100 [ 84.161346] vfs_setxattr+0x9b/0x170 [ 84.161862] setxattr+0x110/0x200 [ 84.162346] ? _raw_spin_unlock+0xe/0x20 [ 84.162914] ? __handle_mm_fault+0xc1b/0x16d0 [ 84.163556] ? __might_sleep+0x49/0x80 [ 84.164132] __x64_sys_fsetxattr+0xb1/0xe0 [ 84.164782] do_syscall_64+0x35/0x80Ok, there's the first failure.
This looks like it's a problem with xfs_attri_item_{size,format} in calculating the number of bytes to log. They use
ATTR_NVEC_SIZE()to calculate the number of bytes of copy from the attribute item which rounds up the length to copy to 4 byte aligned values. I'm not sure what this function is calculating:/* iovec length must be 32-bit aligned */
static inlinesize_tATTR_NVEC_SIZE(size_t size){ return size == sizeof(int32_t) ? size : sizeof(int32_t) + round_up(size, sizeof(int32_t)); }It appears to be saying if the size == 4, then return 4, otherwise return 4 + roundup(size)... which leads me to struct
xfs_attri_log_format:struct
xfs_attri_log_format{uint16_t alfi_type; /* attri log item type */ uint16_t alfi_size; /* size of this item */ uint32_t __pad; /* pad to 64 bit aligned */ uint64_t alfi_id; /* attri identifier */ uint64_t alfi_ino; /* the inode for this attr operation */ uint32_t alfi_op_flags; /* marks the op as a set or remove */ uint32_t alfi_name_len; /* attr name length */ uint32_t alfi_value_len; /* attr value length */ uint32_t alfi_attr_flags;/* attr flags */};
I don't see where the extra 4 bytes for the attribute vector size comes from. It's not needed to store the length, so this could oversize the amount of data to be copied from the source buffer by up to 7 bytes?
Quoted 5 lines
I can see that it might need rounding with the existing log code (because the formatter is responsible for 32 bit alignment of log vectors), but that goes away with the CIL scalability patchset that always aligns iovecs to 4 byte alignment so the formatters do not need to do that.
I think we figured this out last night, initially this was here for an assertion check in the log code, but I think just the round up will suffice for the check.
Quoted 40 lines
Hiding it in a "macro" is not necessary, either - look at how xfs_inode_item_{data,attr}_fork_size handle the rounding up of the local format fork size. They round up the fork byte count to 4 directly, and the format code copies those bytes because
xfs_idata_realloc()allocates those bytes.However, for the attribute buffers, this isn't guaranteed. Look at
xfs_xattr_set():struct xfs_da_args args = { .dp = XFS_I(inode), .attr_filter = handler->flags, .attr_flags = flags, .name = name, .namelen = strlen(name), .value = (void *)value, .valuelen = size, };There is no rounding up of the name or value lengths, and these end up directly referenced by the deferred logging via
xfs_attr_log_item()and attrip->da_args->...attrip->attri_name = (void *)attr->xattri_da_args->name; attrip->attri_value = attr->xattri_da_args->value; attrip->attri_name_len = attr->xattri_da_args->namelen; attrip->attri_value_len = attr->xattri_da_args->valuelen;We then pass those pointers directly to
xlog_iovec_copy()but with a rounded up length that is longer than the source buffer:xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ATTR_NAME, attrip->attri_name, ATTR_NVEC_SIZE(attrip->attri_name_len)); if (attrip->attri_value_len > 0) xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ATTR_VALUE, attrip->attri_value, ATTR_NVEC_SIZE(attrip->attri_value_len));
Quoted 5 lines
So while this might not be the source of the above crash, it's certainly a bug that needs fixing.
At this point, I'm just going to hack on the code to make it work, and we can go from there...
I think when we left off last night, we are more concerned with the extra transaction overhead in the perf captures. I will see if I can replicate what you are seeing with perf and maybe we can work out some short cuts. Thank for your help here!
Allison
Quoted 5 lines
Cheers,
Dave.
Quoted 92 lines · Allison Henderson
Quoted 13 lines · Dave Chinner
Quoted 12 lines · Allison Henderson
Hi all,
This set is a subset of a larger series parent pointers. Delayed attributes allow attribute operations (set and remove) to be logged and committed in the same way that other delayed operations do. This allows more complex operations (like parent pointers) to be broken up into multiple smaller transactions. To do this, the existing attr operations must be modified to operate as a delayed operation. This means that they cannot roll, commit, or finish transactions. Instead, they return
-EAGAINto allow the calling function to handle the transaction. In this series, we focus on only the delayed attribute portion. We will introduce parent pointers in a later set.Quoted 3 lines
Quoted 3 lines
The set as a whole is a bit much to digest at once, so I usually send out the smaller sub series to reduce reviewer burn out. But the entire extended series is visible through the included github links.
Some of this we worked out in the chat last night, but I will echo it here for the archives
Quoted 69 lines
Ok, so like I did with Darrick's deferred inactivation series, the first thing I'm doing here is throwing this patchset at scalability/performance worklaods and finding out what is different.
I've merged this series with 5.14 + xfs/for-next + xfs-cil-scale and then run some tests on it. First up is fsmark creating zero length files w/ 64 byte xattrs. This should stress only shortform attribute manipulations.
I have not enabled delayed attributes yet (i.e.
/sys/fs/xfs/debug/larp= 0)First thing I notice is the transaction commit rate during create is up around 900k/s, so we are doing 3 transactions per inode - 1 for create, 2 for attributes. That looks like a regression - existing shortform attribute creation only takes a single transaction commit, so this workload prior to this patchset only ran at 600k commits/s.
Note that hte only reason I'm getting 900k transactions/s is the CIL scalability patchset - without that the system tops out at ~800k transactions/s and so this would be a significant performance regression (20%) vs the current xfs/for-next code.
Essentially, this looks like we are doing an extra transaction commit to defer the creation of the attribute, then doing another transaction to actually modify the attribute. i.e.:
- 11.04%
xfs_attr_set
- 8.70%xfs_trans_commit
- 8.69%__xfs_trans_commit
- 5.10%xfs_defer_finish_noroll
- 3.74%xfs_defer_trans_roll
- 3.57%xfs_trans_roll
- 3.13%__xfs_trans_commit
- 3.01%xlog_cil_commit
0.66%down_read
0.63%xfs_log_ticket_regrant
- 1.16%xfs_attr_finish_item
- 1.06%xfs_trans_attr_finish_update
- 1.03%xfs_attr_set_iter
- 1.01%xfs_attr_sf_addname
- 0.99%xfs_attr_try_sf_addname
- 0.61%xfs_attr_shortform_addname0.55%xfs_attr_shortform_addAFAICT, for non-delayed attributes, this first transaction commit logs the inode but does not create intent or intent done items (returns NULL for both operations), so just rolls and runs the ->finish_item. So it would seem that the first transaction just changes the inode timestamps and does nothing else.
Firstly, this means the inode timestamp change is not atomic w.r.t. the attribute change the timestamp change relates to and it's essentially new overhead for the non-delayed path.
Looking at the unlink path, I see the same thing - there's an extra transaction for the attr remove path, the same as the attr set path. This drives the unlink path to 1.1 million transaction commits/sec instead of 800k/s, so it's likely that there's a substantial performance regression here on a kernel without the CIL scalability patchset.
IOWs, there's significant behavioural changes with the non-delayed logging version of this patchset, both in terms of performance and the atomicity of changes that appear in the journal and hence recovery behaviour.
Quoted 6 lines
Quoted 4 lines
At this point I have to ask: why are we trying to retain the "old" way of doing things (even for testing) if it is substantially changing behaviour and on-disk journal contents for attribute modifications?
Per the chat discussion, we have to keep both methods since sb v4 would not use the new log entries.
Yup, but we likely have better reasons than this. Tests I've run today have indicated that as the xattr size goes up, the performance regression of the delayed attrs gets worse. At max sized xattrs, performance is only ~25% of the existing attr implementation.
Largely the problem is the number of copies of the xattr value we end up in memory when we are using delalyed attrs. In the current case, we have this all in a single syscall context:
- xattr value allocated by VFS
- attached to xfs_da_args
- xfs_da_args passed to xfs_attr_rmtval_set_value() where we allocate buffer(s) to hold the encoded name.
- buffer is written to disk.
- xattr VFS value freed.
And the xfs_buf is freed when memory pressure occurs.
So, essentially we have two copies of the xattr value and two memcpy()s of it to get it to disk.
Now, with delayed attrs we have this in the syscall context:
- xattr value allocated by VFS
- attached to xfs_da_args
- attached to xfs_attri_item
- attri item intent gets created, allocates a 64k+ shadow buffer and memcpy()s the value into it.
- intent gets committed and attached to the CIL
- xfs_da_args passed to xfs_attr_rmtval_set_value() where we allocate buffer(s) to hold the encoded name.
- buffer is written to disk.
- xattr VFS value freed.
Now userspace runs more xattr creation syscalls, all queuing up shadow buffers on the CIL. memory footprint goes up massively.
Some time later, the CIL commits and we:
- shadow buffer attached to CIL commit
- memcpy() the value from the shadow buffer to the iclog
- commit the CIL checkpoint
- on CIL checkpoint completion, shadow buffer is freed.
And the xfs_bufs is freed when memory pressure occurs.
The result is that with delayed attrs, large attr creation hammers the page allocator really badly - more than 50% of the CPU time is spent trying to allocate pages and compact memory into contiguous regions and so performance goes way down. e.g.
- 51.90% xlog_cil_commit
- 36.96% kvmalloc_node
- 36.51% __kmalloc_node
- 36.32% kmalloc_large_node
- 36.24% __alloc_pages
- 34.53% __alloc_pages_slowpath.constprop.0
- 33.80% __alloc_pages_direct_compact
- 33.62% try_to_compact_pages
- compact_zone_order
- 30.58% compact_zone 14.72% PageHuge
I suspect we need to look at our use of kvmalloc() and the flags we pass it because I think if we can't get contiguous ranges immediately, we should go straight to vmalloc rather than burn a dozen CPUs trying to compact memory...
The amount we log also goes way up. At 16 threads and 4kB xattrs, the log alone is sustaining more than 1GB/s throughput at 50k xattrs creates/s. There is over 2GB/s going to disk.
With 64k xattrs, there is over 2.5GB/s being written to the log and just over 3GB/s being written to disk. Performance is about 9000 xattr creates/s.
In comparison, with delayed attrs turned off for 64k xattrs, log throughput is roughly 500MB/s, disk throughput is between 3.5-4GB/s and the create rate is around 45,000 xattrs/s. So you can see that logging large xattrs really hurts performance.
We want delayed attrs for modifying attributes atomically with other inode modifications (i.e. parent pointers, security labels, ACLs, etc) and none of these use cases use large xattrs - they are all relatively small and so the logging and memory overhead of delayed attrs isn't a huge deal for them. For anything else, small or large, a 10-15% regression is a deal breaker (think ceph, samba, etc).
So I think we're going to need to select attribute modification behaviour at the call site. e.g. from xattr syscalls we don't get any real integrity benefit from logging xattrs, so maybe this path always uses the unlogged path. The we can change the internal create path to use delayed attrs for parent pointers and security attrs. As Darrick said, this is likely what the current larp debug knob should turn into - caller selected behaviour.
Quoted 42 lines
Quoted 36 lines
Ok, there's the first failure.
This looks like it's a problem with xfs_attri_item_{size,format} in calculating the number of bytes to log. They use
ATTR_NVEC_SIZE()to calculate the number of bytes of copy from the attribute item which rounds up the length to copy to 4 byte aligned values. I'm not sure what this function is calculating:/* iovec length must be 32-bit aligned */
static inlinesize_tATTR_NVEC_SIZE(size_t size){ return size == sizeof(int32_t) ? size : sizeof(int32_t) + round_up(size, sizeof(int32_t)); }It appears to be saying if the size == 4, then return 4, otherwise return 4 + roundup(size)... which leads me to struct
xfs_attri_log_format:struct
xfs_attri_log_format{uint16_t alfi_type; /* attri log item type */ uint16_t alfi_size; /* size of this item */ uint32_t __pad; /* pad to 64 bit aligned */ uint64_t alfi_id; /* attri identifier */ uint64_t alfi_ino; /* the inode for this attr operation */ uint32_t alfi_op_flags; /* marks the op as a set or remove */ uint32_t alfi_name_len; /* attr name length */ uint32_t alfi_value_len; /* attr value length */ uint32_t alfi_attr_flags;/* attr flags */};
I don't see where the extra 4 bytes for the attribute vector size comes from. It's not needed to store the length, so this could oversize the amount of data to be copied from the source buffer by up to 7 bytes?
Quoted 5 lines
I can see that it might need rounding with the existing log code (because the formatter is responsible for 32 bit alignment of log vectors), but that goes away with the CIL scalability patchset that always aligns iovecs to 4 byte alignment so the formatters do not need to do that.
Quoted 3 lines
I think we figured this out last night, initially this was here for an assertion check in the log code, but I think just the round up will suffice for the check.
I've got a cleaner solution in the works - we should be hiding the alignment behind the log iovec API, not force callers to know about it and always get it right.
Quoted 42 lines
Quoted 42 lines
Hiding it in a "macro" is not necessary, either - look at how xfs_inode_item_{data,attr}_fork_size handle the rounding up of the local format fork size. They round up the fork byte count to 4 directly, and the format code copies those bytes because
xfs_idata_realloc()allocates those bytes.However, for the attribute buffers, this isn't guaranteed. Look at
xfs_xattr_set():struct xfs_da_args args = { .dp = XFS_I(inode), .attr_filter = handler->flags, .attr_flags = flags, .name = name, .namelen = strlen(name), .value = (void *)value, .valuelen = size, };There is no rounding up of the name or value lengths, and these end up directly referenced by the deferred logging via
xfs_attr_log_item()and attrip->da_args->...attrip->attri_name = (void *)attr->xattri_da_args->name; attrip->attri_value = attr->xattri_da_args->value; attrip->attri_name_len = attr->xattri_da_args->namelen; attrip->attri_value_len = attr->xattri_da_args->valuelen;We then pass those pointers directly to
xlog_iovec_copy()but with a rounded up length that is longer than the source buffer:xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ATTR_NAME, attrip->attri_name, ATTR_NVEC_SIZE(attrip->attri_name_len)); if (attrip->attri_value_len > 0) xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ATTR_VALUE, attrip->attri_value, ATTR_NVEC_SIZE(attrip->attri_value_len));So while this might not be the source of the above crash, it's certainly a bug that needs fixing.
Quoted 7 lines
Quoted 2 lines
At this point, I'm just going to hack on the code to make it work, and we can go from there...
I think when we left off last night, we are more concerned with the extra transaction overhead in the perf captures. I will see if I can replicate what you are seeing with perf and maybe we can work out some short cuts. Thank for your help here!
Yup, there's a few issues. I've hacked out the initial transaction for non-delayed attrs and that brings the commit count down as expected. it might be acceptible, but we'll see.
I've also added intent whiteouts to try to avoid having to log intents that are completed before the intent has been written to the log. With that in place, the above large xattr creation workloads have relatively consistent log throughput of around 600-700MB/s, and performance for 4k and 64kB xattrs roughly doubles. It's still way behind the current implementation, though, because it hammers the page allocator even harder now. However, intent whiteouts are generic, so I'll write this up into a separate patchset that addresses all the other types of intents we have as well.
There's some other things we can potentially do to minimise the memory footprint and the number of memcpy()s of the logged values, so I'll look into these tomorrow.
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
cover xfs: various logged attribute fixes
In reply to Allison HendersonHi Allison,
Here are the first set of fixups I've made while testing this series. The intent whiteouts are generic so I'm keeping that separate, even though it's the change that fixes most of the performance regressions for small xattr sizes that result from enabling logged attributes.
Cheers,
Dave.
PATCH 1/5 xfs: fix fallthrough annotations in xfs_attr_set_iter()
From: Dave Chinner <[email protected]>
--- fs/xfs/libxfs/xfs_attr.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
fs/xfs/libxfs/xfs_attr.c +2 -2
@@ -448,7 +448,7 @@ xfs_attr_set_iter(return -EAGAIN ; } -/* fallthrough */ + fallthrough;case XFS_DAS_FLIP_LFLAG :/* * Dismantle the "old" attribute/value pair by removing a
@@ -566,7 +566,7 @@ xfs_attr_set_iter(return -EAGAIN ; } -/* fallthrough */ + fallthrough;case XFS_DAS_FLIP_NFLAG :/* * Dismantle the "old" attribute/value pair by removing a
Signature · 1 line
2.31.1
PATCH 2/5 xfs: fix flags passed to kvmalloc() by xfs_attri_init()
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
--- fs/xfs/xfs_attr_item.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/xfs/xfs_attr_item.c +1 -1
@@ -184,7 +184,7 @@ xfs_attri_init(uint size; size =sizeof (struct xfs_attri_log_item ) + buffer_size; - attrip =kvmalloc (size,KM_ZERO ); + attrip =kvmalloc (size,GFP_KERNEL | __GFP_ZERO);if (attrip ==NULL )return NULL ;
Signature · 1 line
2.31.1
PATCH 3/5 xfs: hide log iovec alignment constraints
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Callers currently have to round out the size of buffers to match the aligment constraints of log iovecs and xlog_write(). They should not need to know this detail, so introduce a new function to calculate the iovec length (for use in ->iop_size implementations). Also modify xlog_finish_iovec() to round up the length to the correct alignment so the callers don't need to do this, either.
Convert the only user - inode forks - of this alignment rounding to use the new interface.
--- fs/xfs/libxfs/xfs_inode_fork.c | 10 ++-------- fs/xfs/xfs_inode_item.c | 25 +++++++------------------ fs/xfs/xfs_log.h | 20 ++++++++++++++++++++ 3 files changed, 29 insertions(+), 26 deletions(-)
fs/xfs/libxfs/xfs_inode_fork.c +2 -8
@@ -50,8 +50,7 @@ xfs_init_local_fork( mem_size++;if (size) { - real_size =roundup (mem_size,4 ); - ifp->if_u1.if_data =kmem_alloc (real_size,KM_NOFS ); + ifp->if_u1.if_data =kmem_alloc (mem_size,KM_NOFS );memcpy (ifp->if_u1.if_data, data, size);if (zero_terminate) ifp->if_u1.if_data[size] ='\0' ;
@@ -497,12 +496,7 @@ xfs_idata_realloc(return ; } -/* -* For inline data, the underlying buffer must be a multiple of 4 bytes -* in size so that it can be logged and stay on word boundaries. -* We enforce that here. -*/ - ifp->if_u1.if_data =krealloc (ifp->if_u1.if_data,roundup (new_size,4 ), + ifp->if_u1.if_data =krealloc (ifp->if_u1.if_data, new_size,GFP_NOFS | __GFP_NOFAIL); ifp->if_bytes = new_size; }
fs/xfs/xfs_inode_item.c +7 -18
@@ -70,7 +70,7 @@ xfs_inode_item_data_fork_size(case XFS_DINODE_FMT_LOCAL :if ((iip->ili_fields &XFS_ILOG_DDATA ) && ip->i_df.if_bytes >0 ) { - *nbytes +=roundup (ip->i_df.if_bytes,4 ); + *nbytes +=xlog_calc_iovec_len (ip->i_df.if_bytes); *nvecs +=1 ; }break ;
@@ -111,7 +111,7 @@ xfs_inode_item_attr_fork_size(case XFS_DINODE_FMT_LOCAL :if ((iip->ili_fields &XFS_ILOG_ADATA ) && ip->i_afp->if_bytes >0 ) { - *nbytes +=roundup (ip->i_afp->if_bytes,4 ); + *nbytes +=xlog_calc_iovec_len (ip->i_df.if_bytes); *nvecs +=1 ; }break ;
@@ -203,17 +203,12 @@ xfs_inode_item_format_data_fork( ~(XFS_ILOG_DEXT |XFS_ILOG_DBROOT |XFS_ILOG_DEV );if ((iip->ili_fields &XFS_ILOG_DDATA ) && ip->i_df.if_bytes >0 ) { -/* -* Round i_bytes up to a word boundary. -* The underlying memory is guaranteed -* to be there by xfs_idata_realloc(). -*/ - data_bytes =roundup (ip->i_df.if_bytes,4 );ASSERT (ip->i_df.if_u1.if_data !=NULL );ASSERT (ip->i_disk_size >0 );xlog_copy_iovec (lv, vecp,XLOG_REG_TYPE_ILOCAL , - ip->i_df.if_u1.if_data, data_bytes); - ilf->ilf_dsize = (unsigned )data_bytes; + ip->i_df.if_u1.if_data, + ip->i_df.if_bytes); + ilf->ilf_dsize = (unsigned )ip->i_df.if_bytes; ilf->ilf_size++; }else { iip->ili_fields &= ~XFS_ILOG_DDATA ;
@@ -287,17 +282,11 @@ xfs_inode_item_format_attr_fork(if ((iip->ili_fields &XFS_ILOG_ADATA ) && ip->i_afp->if_bytes >0 ) { -/* -* Round i_bytes up to a word boundary. -* The underlying memory is guaranteed -* to be there by xfs_idata_realloc(). -*/ - data_bytes =roundup (ip->i_afp->if_bytes,4 );ASSERT (ip->i_afp->if_u1.if_data !=NULL );xlog_copy_iovec (lv, vecp,XLOG_REG_TYPE_IATTR_LOCAL , ip->i_afp->if_u1.if_data, - data_bytes); - ilf->ilf_asize = (unsigned )data_bytes; + ip->i_afp->if_bytes); + ilf->ilf_asize = (unsigned )ip->i_afp->if_bytes; ilf->ilf_size++; }else { iip->ili_fields &= ~XFS_ILOG_ADATA ;
fs/xfs/xfs_log.h +20 -0
@@ -22,6 +22,17 @@ struct xfs_log_vec {#define XFS_LOG_VEC_ORDERED (-1) +/* +* Calculate the log iovec length for a given user buffer length. Intended to be +* used by ->iop_size implementations when sizing buffers of arbitrary +* alignments. +*/ +static inline int +xlog_calc_iovec_len (int len) +{ +return roundup (len,4 ); +} +void *xlog_prepare_iovec (struct xfs_log_vec *lv,struct xfs_log_iovec **vecp,uint type);
@@ -30,6 +41,12 @@ xlog_finish_iovec(struct xfs_log_vec *lv, struct xfs_log_iovec *vec, int len) {struct xlog_op_header *oph = vec->i_addr; +/* +* Always round up the length to the correct alignment so callers don't +* need to know anything about this log vec layout requirement. +*/ + len =xlog_calc_iovec_len (len); +/* opheader tracks payload length, logvec tracks region length */ oph->oh_len = len;
@@ -39,6 +56,9 @@ xlog_finish_iovec(struct xfs_log_vec *lv, struct xfs_log_iovec *vec, int len) vec->i_len = len; } +/* +* Copy the amount of data requested by the caller into a new log iovec. +*/ static inline void *xlog_copy_iovec (struct xfs_log_vec *lv,struct xfs_log_iovec **vecp,uint type,void *data,int len)
Signature · 1 line
2.31.1
PATCH 4/5 xfs: fix attribute log iovec sizing
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
The sizing of the attri name and value log iovecs is incorrect. This results in memory corruption and crashes on a kernel with the current CIL scalability patchset applied as it relies on the callers playing by slightly different alignment rules.
Convert the attri code to the new xlog_calc_iovec_size() API to fix these issues.
--- fs/xfs/xfs_attr_item.c | 29 ++++++++--------------------- 1 file changed, 8 insertions(+), 21 deletions(-)
fs/xfs/xfs_attr_item.c +8 -21
@@ -42,13 +42,6 @@static const struct xfs_item_ops xfs_attri_item_ops;static const struct xfs_item_ops xfs_attrd_item_ops; -/* iovec length must be 32-bit aligned */ -static inline size_t ATTR_NVEC_SIZE (size_t size) -{ -return size ==sizeof (int32_t ) ? size : -sizeof (int32_t ) +round_up (size,sizeof (int32_t )); -} -static inline struct xfs_attri_log_item *ATTRI_ITEM (struct xfs_log_item *lip) { return container_of(lip, struct xfs_attri_log_item, attri_item);
@@ -89,19 +82,15 @@ xfs_attri_item_size( {struct xfs_attri_log_item *attrip =ATTRI_ITEM (lip); - *nvecs +=1 ; - *nbytes +=sizeof (struct xfs_attri_log_format ); + *nvecs +=2 ; + *nbytes +=sizeof (struct xfs_attri_log_format ) + +xlog_calc_iovec_len (attrip->attri_name_len); -/* Attr set and remove operations require a name */ -ASSERT (attrip->attri_name_len >0 ); +if (!attrip->attri_value_len) +return ; *nvecs +=1 ; - *nbytes +=ATTR_NVEC_SIZE (attrip->attri_name_len); - -if (attrip->attri_value_len >0 ) { - *nvecs +=1 ; - *nbytes +=ATTR_NVEC_SIZE (attrip->attri_value_len); - } + *nbytes +=xlog_calc_iovec_len (attrip->attri_value_len); }/*
@@ -137,12 +126,10 @@ xfs_attri_item_format( &attrip->attri_format, sizeof(struct xfs_attri_log_format));xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTR_NAME , - attrip->attri_name, -ATTR_NVEC_SIZE (attrip->attri_name_len)); + attrip->attri_name, attrip->attri_name_len);if (attrip->attri_value_len >0 )xlog_copy_iovec (lv, &vecp,XLOG_REG_TYPE_ATTR_VALUE , - attrip->attri_value, -ATTR_NVEC_SIZE (attrip->attri_value_len)); + attrip->attri_value, attrip->attri_value_len); }/*
Signature · 1 line
2.31.1
RFC PATCH 5/5 xfs: don't commit the first deferred transaction without intents
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
If the first operation in a string of defer ops has no intents, then there is no reason to commit it before running the first call to xfs_defer_finish_one(). This allows the defer ops to be used effectively for non-intent based operations without requiring an unnecessary extra transaction commit when first called.
This fixes a regression in per-attribute modification transaction count when delayed attributes are not being used.
--- fs/xfs/libxfs/xfs_defer.c | 29 +++++++++++++++++------------ 1 file changed, 17 insertions(+), 12 deletions(-)
fs/xfs/libxfs/xfs_defer.c +17 -12
@@ -181,7 +181,7 @@ static const struct xfs_defer_op_type *defer_op_types[] = { [XFS_DEFER_OPS_TYPE_ATTR] = &xfs_attr_defer_type, }; -static void +static bool xfs_defer_create_intent( struct xfs_trans *tp, struct xfs_defer_pending *dfp,
@@ -192,6 +192,7 @@ xfs_defer_create_intent(if (!dfp->dfp_intent) dfp->dfp_intent = ops->create_intent (tp, &dfp->dfp_work, dfp->dfp_count, sort); +return dfp->dfp_intent; }/*
@@ -199,16 +200,18 @@ xfs_defer_create_intent(* associated extents, then add the entire intake list to the end of * the pending list. */ -STATIC void +STATIC boolxfs_defer_create_intents (struct xfs_trans *tp) {struct xfs_defer_pending *dfp; +bool ret = false;list_for_each_entry (dfp, &tp->t_dfops, dfp_list) {trace_xfs_defer_create_intent (tp->t_mountp, dfp); -xfs_defer_create_intent (tp, dfp, true); + ret |=xfs_defer_create_intent (tp, dfp, true); } +return ret; }/* Abort all the intents that were committed. */
@@ -459,7 +462,7 @@ intxfs_defer_finish_noroll ( struct xfs_trans **tp) { -struct xfs_defer_pending *dfp; +struct xfs_defer_pending *dfp =NULL ;int error =0 ;LIST_HEAD (dop_pending);
@@ -478,17 +481,19 @@ xfs_defer_finish_noroll(* of time that any one intent item can stick around in memory, * pinning the log tail. */ -xfs_defer_create_intents (*tp); +bool has_intents =xfs_defer_create_intents (*tp);list_splice_init (&(*tp)->t_dfops, &dop_pending); - error =xfs_defer_trans_roll (tp); -if (error) - goto out_shutdown; +if (has_intents || dfp) { + error =xfs_defer_trans_roll (tp); +if (error) + goto out_shutdown; -/* Possibly relog intent items to keep the log moving. */ - error =xfs_defer_relog (tp, &dop_pending); -if (error) - goto out_shutdown; +/* Possibly relog intent items to keep the log moving. */ + error =xfs_defer_relog (tp, &dop_pending); +if (error) + goto out_shutdown; + } dfp = list_first_entry(&dop_pending, struct xfs_defer_pending, dfp_list);
Signature · 1 line
2.31.1
Re: [PATCH 0/5] xfs: various logged attribute fixes
In reply to Dave ChinnerAlrighty, thanks for all your help on this. These changes make sense to me, I can add patches 3 and 5 to the set, and the rest looks like they can be amended into their respective patches in the series.
I will keep an eye out for the whiteout set you are working on then. Thank you!!
Allison
Quoted 12 lines · Dave Chinner
Hi Allison,
Here are the first set of fixups I've made while testing this series. The intent whiteouts are generic so I'm keeping that separate, even though it's the change that fixes most of the performance regressions for small xattr sizes that result from enabling logged attributes.
Cheers,
Dave.