Bail out if transaction can cause extent count to overflow
v11 cover Bail out if transaction can cause extent count to overflow
XFS does not check for possible overflow of per-inode extent counter fields when adding extents to either data or attr fork.
For e.g.
1. Insert 5 million xattrs (each having a value size of 255 bytes) and then delete 50% of them in an alternating manner.
2. On a 4k block sized XFS filesystem instance, the above causes 98511 extents to be created in the attr fork of the inode.
xfsaild/loop0 2008 [003] 1475.127209: probe:xfs_inode_to_disk: (ffffffffa43fb6b0) if_nextents=98511 i_ino=131
3. The incore inode fork extent counter is a signed 32-bit quantity. However, the on-disk extent counter is an unsigned 16-bit quantity and hence cannot hold 98511 extents.
4. The following incorrect value is stored in the xattr extent counter,
# xfs_db -f -c 'inode 131' -c 'print core.naextents' /dev/loop0 core.naextents = -32561
This patchset adds a new helper function
(i.e. xfs_iext_count_may_overflow()) to check for overflow of the per-inode data and xattr extent counters and invokes it before starting an fs operation (e.g. creating a new directory entry). With this patchset applied, XFS detects counter overflows and returns with an error rather than causing a silent corruption.
The patchset has been tested by executing xfstests with the following mkfs.xfs options,
1. -m crc=0 -b size=1k
2. -m crc=0 -b size=4k
3. -m crc=0 -b size=512
4. -m rmapbt=1,reflink=1 -b size=1k
5. -m rmapbt=1,reflink=1 -b size=4k
The patches can also be obtained from
https://github.com/chandanr/linux.git at branch xfs-reserve-extent-count-v11.
I have two patches that define the newly introduced error injection tags in xfsprogs
(https://lore.kernel.org/linux-xfs/[email protected]/).
I have also written tests
(https://github.com/chandanr/xfstests/commits/extent-overflow-tests)
for verifying the checks introduced in the kernel.
Changelog:
V10 -> V11:
1. For directory/xattr insert operations we now reserve sufficient number of "extent count" so as to guarantee a future directory/xattr remove operation.
2. The pseudo max extent count value has been increased to 35.
V9 -> V10:
1. Pull back changes which cause xfs_bmap_compute_alignments() to return "stripe alignment" into 12th patch i.e. "xfs: Compute bmap extent alignments in a separate function".
V8 -> V9:
1. Enabling XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT error tag will always allocate single block sized free extents (if available).
2. xfs_bmap_compute_alignments() now returns stripe alignment as its return value.
3. Dropped Allison's RVB tag for "xfs: Compute bmap extent alignments in a separate function" and "xfs: Introduce error injection to allocate only minlen size extents for files".
V7 -> V8:
1. Rename local variable in xfs_alloc_fix_freelist() from "i" to "stat".
V6 -> V7:
1. Create new function xfs_bmap_exact_minlen_extent_alloc() (enabled only when CONFIG_XFS_DEBUG is set to y) which issues allocation requests for minlen sized extents only. In order to achieve this, common code from xfs_bmap_btalloc() have been refactored into new functions.
2. All major functions implementing logic associated with XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT error tag are compiled only when CONFIG_XFS_DEBUG is set to y.
3. Remove XFS_IEXT_REFLINK_REMAP_CNT macro and replace it with an integer which holds the number of new extents to be added to the data fork.
V5 -> V6:
1. Rebased the patchset on xfs-linux/for-next branch.
2. Drop "xfs: Set tp->t_firstblock only once during a transaction's lifetime" patch from the patchset.
3. Add a comment to xfs_bmap_btalloc() describing why it was chosen to start "free space extent search" from AG 0 when XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is enabled and when the transaction is allocating its first extent.
4. Fix review comments associated with coding style.
V4 -> V5:
1. Introduce new error tag XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT to let user space programs to be able to guarantee that free space requests for files are satisfied by allocating minlen sized extents.
2. Change xfs_bmap_btalloc() and xfs_alloc_vextent() to allocate minlen sized extents when XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is enabled.
3. Introduce a new patch that causes tp->t_firstblock to be assigned to a value only when its previous value is NULLFSBLOCK.
4. Replace the previously introduced MAXERRTAGEXTNUM (maximum inode fork extent count) with the hardcoded value of 10.
5. xfs_bui_item_recover(): Use XFS_IEXT_ADD_NOSPLIT_CNT when mapping an extent.
6. xfs_swap_extent_rmap(): Use xfs_bmap_is_real_extent() instead of xfs_bmap_is_update_needed() to assess if the extent really needs to be swapped.
V3 -> V4:
1. Introduce new patch which lets userspace programs to test "extent count overflow detection" by injecting an error tag. The new error tag reduces the maximum allowed extent count to 10.
2. Injecting the newly defined error tag prevents xfs_bmap_add_extent_hole_real() from merging a new extent with its neighbours to allow writing deterministic tests for testing extent count overflow for Directories, Xattr and growing realtime devices. This is required because the new extent being allocated can be contiguous with its neighbours (w.r.t both file and disk offsets).
3. Injecting the newly defined error tag forces block sized extents to be allocated for summary/bitmap files when growing a realtime device. This is required because xfs_growfs_rt_alloc() allocates as large an extent as possible for summary/bitmap files and hence it would be impossible to write deterministic tests.
4. Rename XFS_IEXT_REMOVE_CNT to XFS_IEXT_PUNCH_HOLE_CNT to reflect the actual meaning of the fs operation.
5. Fold XFS_IEXT_INSERT_HOLE_CNT code into that associated with XFS_IEXT_PUNCH_HOLE_CNT since both perform the same job.
6. xfs_swap_extent_rmap(): Check for extent overflow should be made on the source file only if the donor file extent has a valid on-disk mapping and vice versa.
V2 -> V3:
1. Move the definition of xfs_iext_count_may_overflow() from libxfs/xfs_trans_resv.c to libxfs/xfs_inode_fork.c. Also, I tried to make xfs_iext_count_may_overflow() an inline function by placing the definition in libxfs/xfs_inode_fork.h. However this required that the definition of 'struct xfs_inode' be available, since xfs_iext_count_may_overflow() uses a 'struct xfs_inode *' type variable.
2. Handle XFS_COW_FORK within xfs_iext_count_may_overflow() by returning a success value.
3. Rename XFS_IEXT_ADD_CNT to XFS_IEXT_ADD_NOSPLIT_CNT. Thanks to Darrick for the suggesting the new name.
4. Expand comments to make use of 80 columns.
V1 -> V2:
1. Rename helper function from xfs_trans_resv_ext_cnt() to xfs_iext_count_may_overflow().
2. Define and use macros to represent fs operations and the corresponding increase in extent count.
3. Split the patches based on the fs operation being performed.
Chandan Babu R (14):
xfs: Add helper for checking per-inode extent count overflow
xfs: Check for extent overflow when trivally adding a new extent
xfs: Check for extent overflow when punching a hole
xfs: Check for extent overflow when adding/removing xattrs
xfs: Check for extent overflow when adding/removing dir entries
xfs: Check for extent overflow when writing to unwritten extent
xfs: Check for extent overflow when moving extent from cow to data
fork
xfs: Check for extent overflow when remapping an extent
xfs: Check for extent overflow when swapping extents
xfs: Introduce error injection to reduce maximum inode fork extent
count
xfs: Remove duplicate assert statement in xfs_bmap_btalloc()
xfs: Compute bmap extent alignments in a separate function
xfs: Process allocated extent in a separate function
xfs: Introduce error injection to allocate only minlen size extents
for files
fs/xfs/libxfs/xfs_alloc.c | 50 +++++++ fs/xfs/libxfs/xfs_alloc.h | 3 + fs/xfs/libxfs/xfs_attr.c | 20 +++ fs/xfs/libxfs/xfs_bmap.c | 264 +++++++++++++++++++++++---------- fs/xfs/libxfs/xfs_errortag.h | 6 +- fs/xfs/libxfs/xfs_inode_fork.c | 27 ++++ fs/xfs/libxfs/xfs_inode_fork.h | 62 ++++++++ fs/xfs/xfs_bmap_item.c | 10 ++ fs/xfs/xfs_bmap_util.c | 31 ++++ fs/xfs/xfs_dquot.c | 8 +- fs/xfs/xfs_error.c | 6 + fs/xfs/xfs_inode.c | 27 ++++ fs/xfs/xfs_iomap.c | 10 ++ fs/xfs/xfs_reflink.c | 16 ++ fs/xfs/xfs_rtalloc.c | 5 + fs/xfs/xfs_symlink.c | 5 + 16 files changed, 472 insertions(+), 78 deletions(-)
Signature · 1 line
2.28.0
PATCH v11 1/14 xfs: Add helper for checking per-inode extent count overflow
XFS does not check for possible overflow of per-inode extent counter fields when adding extents to either data or attr fork.
For e.g.
1. Insert 5 million xattrs (each having a value size of 255 bytes) and then delete 50% of them in an alternating manner.
2. On a 4k block sized XFS filesystem instance, the above causes 98511 extents to be created in the attr fork of the inode.
xfsaild/loop0 2008 [003] 1475.127209: probe:xfs_inode_to_disk: (ffffffffa43fb6b0) if_nextents=98511 i_ino=131
3. The incore inode fork extent counter is a signed 32-bit quantity. However the on-disk extent counter is an unsigned 16-bit quantity and hence cannot hold 98511 extents.
4. The following incorrect value is stored in the attr extent counter,
# xfs_db -f -c 'inode 131' -c 'print core.naextents' /dev/loop0 core.naextents = -32561
This commit adds a new helper function (i.e. xfs_iext_count_may_overflow()) to check for overflow of the per-inode data and xattr extent counters. Future patches will use this function to make sure that an FS operation won't cause the extent counter to overflow.
--- fs/xfs/libxfs/xfs_inode_fork.c | 23 +++++++++++++++++++++++ fs/xfs/libxfs/xfs_inode_fork.h | 2 ++ 2 files changed, 25 insertions(+)
fs/xfs/libxfs/xfs_inode_fork.c +23 -0
@@ -23,6 +23,7 @@#include "xfs_da_btree.h" #include "xfs_dir2_priv.h" #include "xfs_attr_leaf.h" +#include "xfs_types.h" kmem_zone_t *xfs_ifork_zone;
@@ -728,3 +729,25 @@ xfs_ifork_verify_local_attr(return 0 ; } + +int +xfs_iext_count_may_overflow ( +struct xfs_inode *ip, +int whichfork, +int nr_to_add) +{ +struct xfs_ifork *ifp =XFS_IFORK_PTR (ip, whichfork); +uint64_t max_exts; +uint64_t nr_exts; + +if (whichfork ==XFS_COW_FORK ) +return 0 ; + + max_exts = (whichfork ==XFS_ATTR_FORK ) ?MAXAEXTNUM :MAXEXTNUM ; + + nr_exts = ifp->if_nextents + nr_to_add; +if (nr_exts < ifp->if_nextents || nr_exts > max_exts) +return -EFBIG ; + +return 0 ; +}
fs/xfs/libxfs/xfs_inode_fork.h +2 -0
@@ -172,5 +172,7 @@ extern void xfs_ifork_init_cow(struct xfs_inode *ip);int xfs_ifork_verify_local_data (struct xfs_inode *ip);int xfs_ifork_verify_local_attr (struct xfs_inode *ip); +int xfs_iext_count_may_overflow (struct xfs_inode *ip,int whichfork, +int nr_to_add);#endif /* __XFS_INODE_FORK_H__ */
Signature · 1 line
2.28.0
PATCH v11 2/14 xfs: Check for extent overflow when trivally adding a new extent
In reply to Chandan Babu RWhen adding a new data extent (without modifying an inode's existing extents) the extent count increases only by 1. This commit checks for extent count overflow in such cases.
--- fs/xfs/libxfs/xfs_bmap.c | 6 ++++++ fs/xfs/libxfs/xfs_inode_fork.h | 6 ++++++ fs/xfs/xfs_bmap_item.c | 7 +++++++ fs/xfs/xfs_bmap_util.c | 5 +++++ fs/xfs/xfs_dquot.c | 8 +++++++- fs/xfs/xfs_iomap.c | 5 +++++ fs/xfs/xfs_rtalloc.c | 5 +++++ 7 files changed, 41 insertions(+), 1 deletion(-)
fs/xfs/libxfs/xfs_bmap.c +6 -0
@@ -4527,6 +4527,12 @@ xfs_bmapi_convert_delalloc(return error;xfs_ilock (ip,XFS_ILOCK_EXCL ); + + error =xfs_iext_count_may_overflow (ip, whichfork, +XFS_IEXT_ADD_NOSPLIT_CNT ); +if (error) + goto out_trans_cancel; +xfs_trans_ijoin (tp, ip,0 ); if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
fs/xfs/libxfs/xfs_inode_fork.h +6 -0
@@ -34,6 +34,12 @@ struct xfs_ifork {#define XFS_IFEXTENTS 0x02/* All extent pointers are read in */ #define XFS_IFBROOT 0x04/* i_broot points to the bmap b-tree root */ +/* +* Worst-case increase in the fork extent count when we're adding a single +* extent to a fork and there's no possibility of splitting an existing mapping. +*/ +#define XFS_IEXT_ADD_NOSPLIT_CNT (1) +/* * Fork handling. */
fs/xfs/xfs_bmap_item.c +7 -0
@@ -497,6 +497,13 @@ xfs_bui_item_recover(xfs_ilock (ip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,0 ); +if (bui_type ==XFS_BMAP_MAP ) { + error =xfs_iext_count_may_overflow (ip, whichfork, +XFS_IEXT_ADD_NOSPLIT_CNT ); +if (error) + goto err_cancel; + } + count = bmap->me_len; error = xfs_trans_log_finish_bmap_update(tp, budp, bui_type, ip, whichfork, bmap->me_startoff, bmap->me_startblock,
fs/xfs/xfs_bmap_util.c +5 -0
@@ -822,6 +822,11 @@ xfs_alloc_file_space(if (error) goto error1; + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_ADD_NOSPLIT_CNT ); +if (error) + goto error0; +xfs_trans_ijoin (tp, ip,0 ); error = xfs_bmapi_write(tp, ip, startoffset_fsb,
fs/xfs/xfs_dquot.c +7 -1
@@ -314,8 +314,14 @@ xfs_dquot_disk_alloc(return -ESRCH ; } -/* Create the block mapping. */ xfs_trans_ijoin (tp, quotip,XFS_ILOCK_EXCL ); + + error =xfs_iext_count_may_overflow (quotip,XFS_DATA_FORK , +XFS_IEXT_ADD_NOSPLIT_CNT ); +if (error) +return error; + +/* Create the block mapping. */ error =xfs_bmapi_write (tp, quotip, dqp->q_fileoffset,XFS_DQUOT_CLUSTER_SIZE_FSB ,XFS_BMAPI_METADATA ,0 , &map, &nmaps);
fs/xfs/xfs_iomap.c +5 -0
@@ -250,6 +250,11 @@ xfs_iomap_write_direct(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_ADD_NOSPLIT_CNT ); +if (error) + goto out_trans_cancel; +xfs_trans_ijoin (tp, ip,0 );/*
fs/xfs/xfs_rtalloc.c +5 -0
@@ -804,6 +804,11 @@ xfs_growfs_rt_alloc(xfs_ilock (ip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_ADD_NOSPLIT_CNT ); +if (error) + goto out_trans_cancel; +/* * Allocate blocks to the bitmap file. */
Signature · 1 line
2.28.0
PATCH v11 3/14 xfs: Check for extent overflow when punching a hole
In reply to Chandan Babu RThe extent mapping the file offset at which a hole has to be inserted will be split into two extents causing extent count to increase by 1.
--- fs/xfs/libxfs/xfs_inode_fork.h | 7 +++++++ fs/xfs/xfs_bmap_item.c | 15 +++++++++------ fs/xfs/xfs_bmap_util.c | 10 ++++++++++ 3 files changed, 26 insertions(+), 6 deletions(-)
fs/xfs/libxfs/xfs_inode_fork.h +7 -0
@@ -40,6 +40,13 @@ struct xfs_ifork {*/ #define XFS_IEXT_ADD_NOSPLIT_CNT (1) +/* +* Punching out an extent from the middle of an existing extent can cause the +* extent count to increase by 1. +* i.e. | Old extent | Hole | Old extent | +*/ +#define XFS_IEXT_PUNCH_HOLE_CNT (1) +/* * Fork handling. */
fs/xfs/xfs_bmap_item.c +9 -6
@@ -439,6 +439,7 @@ xfs_bui_item_recover(xfs_exntst_t state;unsigned int bui_type;int whichfork; +int iext_delta;int error =0 ;/* Only one mapping operation per BUI... */
@@ -497,12 +498,14 @@ xfs_bui_item_recover(xfs_ilock (ip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,0 ); -if (bui_type ==XFS_BMAP_MAP ) { - error =xfs_iext_count_may_overflow (ip, whichfork, -XFS_IEXT_ADD_NOSPLIT_CNT ); -if (error) - goto err_cancel; - } +if (bui_type ==XFS_BMAP_MAP ) + iext_delta =XFS_IEXT_ADD_NOSPLIT_CNT ; +else + iext_delta =XFS_IEXT_PUNCH_HOLE_CNT ; + + error =xfs_iext_count_may_overflow (ip, whichfork, iext_delta); +if (error) + goto err_cancel; count = bmap->me_len; error = xfs_trans_log_finish_bmap_update(tp, budp, bui_type, ip,
fs/xfs/xfs_bmap_util.c +10 -0
@@ -891,6 +891,11 @@ xfs_unmap_extent(xfs_trans_ijoin (tp, ip,0 ); + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_PUNCH_HOLE_CNT ); +if (error) + goto out_trans_cancel; + error =xfs_bunmapi (tp, ip, startoffset_fsb, len_fsb,0 ,2 , done);if (error) goto out_trans_cancel;
@@ -1176,6 +1181,11 @@ xfs_insert_file_space(xfs_ilock (ip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,0 ); + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_PUNCH_HOLE_CNT ); +if (error) + goto out_trans_cancel; +/* * The extent shifting code works on extent granularity. So, if stop_fsb * is not the starting block of extent, we need to split the extent at
Signature · 1 line
2.28.0
PATCH v11 4/14 xfs: Check for extent overflow when adding/removing xattrs
In reply to Chandan Babu RAdding/removing an xattr can cause XFS_DA_NODE_MAXDEPTH extents to be added. One extra extent for dabtree in case a local attr is large enough to cause a double split. It can also cause extent count to increase proportional to the size of a remote xattr's value.
To be able to always remove an existing xattr, when adding an xattr we make sure to reserve inode fork extent count required for removing max sized xattr in addition to that required by the xattr add operation.
--- fs/xfs/libxfs/xfs_attr.c | 20 ++++++++++++++++++++ fs/xfs/libxfs/xfs_inode_fork.h | 10 ++++++++++ 2 files changed, 30 insertions(+)
fs/xfs/libxfs/xfs_attr.c +20 -0
@@ -396,6 +396,8 @@ xfs_attr_set(struct xfs_trans_res tres;bool rsvd = (args->attr_filter &XFS_ATTR_ROOT );int error, local; +int iext_cnt; +int rmt_blks;unsigned int total;if (XFS_FORCED_SHUTDOWN (dp->i_mount))
@@ -416,6 +418,9 @@ xfs_attr_set(*/ args->op_flags =XFS_DA_OP_OKNOENT ; + rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); + iext_cnt =XFS_IEXT_ATTR_MANIP_CNT (rmt_blks); +if (args->value) {XFS_STATS_INC (mp, xs_attr_set);
@@ -442,6 +447,13 @@ xfs_attr_set( tres.tr_logcount =XFS_ATTRSET_LOG_COUNT ; tres.tr_logflags =XFS_TRANS_PERM_LOG_RES ; total = args->total; + +if (local) + rmt_blks =0 ; +else + rmt_blks =xfs_attr3_rmt_blocks (mp, args->valuelen); + + iext_cnt +=XFS_IEXT_ATTR_MANIP_CNT (rmt_blks); }else {XFS_STATS_INC (mp, xs_attr_remove);
@@ -460,6 +472,14 @@ xfs_attr_set(xfs_ilock (dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (args->trans, dp,0 ); + +if (args->value ||xfs_inode_hasattr (dp)) { + error =xfs_iext_count_may_overflow (dp,XFS_ATTR_FORK , + iext_cnt); +if (error) + goto out_trans_cancel; + } +if (args->value) {unsigned int quota_flags =XFS_QMOPT_RES_REGBLKS ;
fs/xfs/libxfs/xfs_inode_fork.h +10 -0
@@ -47,6 +47,16 @@ struct xfs_ifork {*/ #define XFS_IEXT_PUNCH_HOLE_CNT (1) +/* +* Adding/removing an xattr can cause XFS_DA_NODE_MAXDEPTH extents to +* be added. One extra extent for dabtree in case a local attr is +* large enough to cause a double split. It can also cause extent +* count to increase proportional to the size of a remote xattr's +* value. +*/ +#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ + (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* * Fork handling. */
Signature · 1 line
2.28.0
Re: [PATCH V11 04/14] xfs: Check for extent overflow when adding/removing xattrs
Quoted 29 lines · Chandan Babu R
Adding/removing an xattr can cause
XFS_DA_NODE_MAXDEPTHextents to be added. One extra extent for dabtree in case a local attr is large enough to cause a double split. It can also cause extent count to increase proportional to the size of a remote xattr's value.To be able to always remove an existing xattr, when adding an xattr we make sure to reserve inode fork extent count required for removing max sized xattr in addition to that required by the xattr add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 20 ++++++++++++++++++++ fs/xfs/libxfs/xfs_inode_fork.h | 10 ++++++++++ 2 files changed, 30 insertions(+)fs/xfs/libxfs/xfs_attr.c +2
-0@@ -396,6 +396,8 @@ xfs_attr_set(struct xfs_trans_res tres;bool rsvd = (args->attr_filter &XFS_ATTR_ROOT );int error, local; +int iext_cnt; +int rmt_blks;unsigned int total;if (XFS_FORCED_SHUTDOWN (dp->i_mount))
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 422
Quoted 5 lines
fs/xfs/libxfs/xfs_attr.c +2
-0@@ -416,6 +418,9 @@ xfs_attr_set(*/ args->op_flags =XFS_DA_OP_OKNOENT ; + rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); + iext_cnt =XFS_IEXT_ATTR_MANIP_CNT (rmt_blks);
These values are only relevant for the xattr removal case, right? AFAICT the args->value != NULL case immediately after will set new values, so why not just move this into...
Quoted 7 lines
fs/xfs/libxfs/xfs_attr.c +1
-0+if (args->value) {XFS_STATS_INC (mp, xs_attr_set);@@ -442,6 +447,13 @@ xfs_attr_set( tres.tr_logcount =XFS_ATTRSET_LOG_COUNT ; tres.tr_logflags =XFS_TRANS_PERM_LOG_RES ;
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 458
Quoted 10 lines
fs/xfs/libxfs/xfs_attr.c +7
-0total = args->total; + +if (local) + rmt_blks =0 ; +else + rmt_blks =xfs_attr3_rmt_blocks (mp, args->valuelen); + + iext_cnt +=XFS_IEXT_ATTR_MANIP_CNT (rmt_blks); }else {XFS_STATS_INC (mp, xs_attr_remove);
...the bottom of this clause here.
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 476
Quoted 7 lines
fs/xfs/libxfs/xfs_attr.c +2
-0@@ -460,6 +472,14 @@ xfs_attr_set(xfs_ilock (dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (args->trans, dp,0 ); + +if (args->value ||xfs_inode_hasattr (dp)) {
Can this simply be "if (iext_cnt != 0)" ?
--D
Quoted 33 lines
fs/xfs/libxfs/xfs_attr.c +6
-0+ error =xfs_iext_count_may_overflow (dp,XFS_ATTR_FORK , + iext_cnt); +if (error) + goto out_trans_cancel; + } +if (args->value) {unsigned int quota_flags =XFS_QMOPT_RES_REGBLKS ;fs/xfs/libxfs/xfs_inode_fork.h +10
-0@@ -47,6 +47,16 @@ struct xfs_ifork {*/ #define XFS_IEXT_PUNCH_HOLE_CNT (1) +/* +* Adding/removing an xattr can cause XFS_DA_NODE_MAXDEPTH extents to +* be added. One extra extent for dabtree in case a local attr is +* large enough to cause a double split. It can also cause extent +* count to increase proportional to the size of a remote xattr's +* value. +*/ +#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ + (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* * Fork handling. */ Signature · 1 line
2.28.0
Re: [PATCH V11 04/14] xfs: Check for extent overflow when adding/removing xattrs
Quoted 33 lines · Darrick J. Wong
Quoted 32 lines · Chandan Babu R
Adding/removing an xattr can cause
XFS_DA_NODE_MAXDEPTHextents to be added. One extra extent for dabtree in case a local attr is large enough to cause a double split. It can also cause extent count to increase proportional to the size of a remote xattr's value.To be able to always remove an existing xattr, when adding an xattr we make sure to reserve inode fork extent count required for removing max sized xattr in addition to that required by the xattr add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 20 ++++++++++++++++++++ fs/xfs/libxfs/xfs_inode_fork.h | 10 ++++++++++ 2 files changed, 30 insertions(+)fs/xfs/libxfs/xfs_attr.c +2
-0@@ -396,6 +396,8 @@ xfs_attr_set(struct xfs_trans_res tres;bool rsvd = (args->attr_filter &XFS_ATTR_ROOT );int error, local; +int iext_cnt; +int rmt_blks;unsigned int total;if (XFS_FORCED_SHUTDOWN (dp->i_mount))@@ -416,6 +418,9 @@ xfs_attr_set(*/ args->op_flags =XFS_DA_OP_OKNOENT ;
Quoted 6 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 422
Quoted 2 lines
fs/xfs/libxfs/xfs_attr.c +2
-0+ rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); + iext_cnt =XFS_IEXT_ATTR_MANIP_CNT (rmt_blks);These values are only relevant for the xattr removal case, right? AFAICT the
args->value!= NULL case immediately after will set new values, so why not just move this into...
The above statements compute the extent count required to remove a maximum sized remote xattr.
To guarantee that a user can always remove an xattr, the "args->value != NULL" case adds to the value of iext_cnt that has been computed above. I had extracted and placed the above set of statements since they were now common to both "insert" and "remove" xattr operations.
Quoted 22 lines
Quoted 8 lines
fs/xfs/libxfs/xfs_attr.c +1
-0+if (args->value) {XFS_STATS_INC (mp, xs_attr_set);@@ -442,6 +447,13 @@ xfs_attr_set( tres.tr_logcount =XFS_ATTRSET_LOG_COUNT ; tres.tr_logflags =XFS_TRANS_PERM_LOG_RES ;fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 458
Quoted 10 lines
fs/xfs/libxfs/xfs_attr.c +7
-0total = args->total; + +if (local) + rmt_blks =0 ; +else + rmt_blks =xfs_attr3_rmt_blocks (mp, args->valuelen); + + iext_cnt +=XFS_IEXT_ATTR_MANIP_CNT (rmt_blks); }else {XFS_STATS_INC (mp, xs_attr_remove);...the bottom of this clause here.
Quoted 1 line
Quoted 8 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 476
Quoted 6 lines
fs/xfs/libxfs/xfs_attr.c +2
-0@@ -460,6 +472,14 @@ xfs_attr_set(xfs_ilock (dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (args->trans, dp,0 ); + +if (args->value ||xfs_inode_hasattr (dp)) {Can this simply be "if (
iext_cnt!= 0)" ?
That would lead to a bug since iext_cnt is computed unconditionally at the beginning of the function. An extent count reservation will be attempted when xattr delete operation is executed against an inode which does not have an associated attr fork. This will cause xfs_iext_count_may_overflow() to dereference the NULL pointer at xfs_inode->i_afp->if_nextents.
Quoted 34 lines
--D
Quoted 30 lines
fs/xfs/libxfs/xfs_attr.c +6
-0+ error =xfs_iext_count_may_overflow (dp,XFS_ATTR_FORK , + iext_cnt); +if (error) + goto out_trans_cancel; + } +if (args->value) {unsigned int quota_flags =XFS_QMOPT_RES_REGBLKS ;fs/xfs/libxfs/xfs_inode_fork.h +10
-0@@ -47,6 +47,16 @@ struct xfs_ifork {*/ #define XFS_IEXT_PUNCH_HOLE_CNT (1) +/* +* Adding/removing an xattr can cause XFS_DA_NODE_MAXDEPTH extents to +* be added. One extra extent for dabtree in case a local attr is +* large enough to cause a double split. It can also cause extent +* count to increase proportional to the size of a remote xattr's +* value. +*/ +#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ + (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* * Fork handling. */
Signature · 1 line
chandan
Darrick J. Wong
Reviewed-byRe: [PATCH V11 04/14] xfs: Check for extent overflow when adding/removing xattrs
Quoted 41 lines · Chandan Babu R
Quoted 33 lines · Darrick J. Wong
Quoted 32 lines · Chandan Babu R
Adding/removing an xattr can cause
XFS_DA_NODE_MAXDEPTHextents to be added. One extra extent for dabtree in case a local attr is large enough to cause a double split. It can also cause extent count to increase proportional to the size of a remote xattr's value.To be able to always remove an existing xattr, when adding an xattr we make sure to reserve inode fork extent count required for removing max sized xattr in addition to that required by the xattr add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_attr.c | 20 ++++++++++++++++++++ fs/xfs/libxfs/xfs_inode_fork.h | 10 ++++++++++ 2 files changed, 30 insertions(+)fs/xfs/libxfs/xfs_attr.c +2
-0@@ -396,6 +396,8 @@ xfs_attr_set(struct xfs_trans_res tres;bool rsvd = (args->attr_filter &XFS_ATTR_ROOT );int error, local; +int iext_cnt; +int rmt_blks;unsigned int total;if (XFS_FORCED_SHUTDOWN (dp->i_mount))@@ -416,6 +418,9 @@ xfs_attr_set(*/ args->op_flags =XFS_DA_OP_OKNOENT ;Quoted 6 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 422
Quoted 2 lines
fs/xfs/libxfs/xfs_attr.c +2
-0+ rmt_blks =xfs_attr3_rmt_blocks (mp,XFS_XATTR_SIZE_MAX ); + iext_cnt =XFS_IEXT_ATTR_MANIP_CNT (rmt_blks);These values are only relevant for the xattr removal case, right? AFAICT the
args->value!= NULL case immediately after will set new values, so why not just move this into...
Quoted 7 lines
The above statements compute the extent count required to remove a maximum sized remote xattr.
To guarantee that a user can always remove an xattr, the "
args->value!= NULL" case adds to the value ofiext_cntthat has been computed above. I had extracted and placed the above set of statements since they were now common to both "insert" and "remove" xattr operations.
D'oh, you're right.
Quoted 31 lines
Quoted 22 lines
Quoted 8 lines
fs/xfs/libxfs/xfs_attr.c +1
-0+if (args->value) {XFS_STATS_INC (mp, xs_attr_set);@@ -442,6 +447,13 @@ xfs_attr_set( tres.tr_logcount =XFS_ATTRSET_LOG_COUNT ; tres.tr_logflags =XFS_TRANS_PERM_LOG_RES ;fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 458
Quoted 10 lines
fs/xfs/libxfs/xfs_attr.c +7
-0total = args->total; + +if (local) + rmt_blks =0 ; +else + rmt_blks =xfs_attr3_rmt_blocks (mp, args->valuelen); + + iext_cnt +=XFS_IEXT_ATTR_MANIP_CNT (rmt_blks); }else {XFS_STATS_INC (mp, xs_attr_remove);...the bottom of this clause here.
Quoted 1 line
Quoted 8 lines
fs/xfs/libxfs/xfs_attr.c xfs_attr_set( · line 476
Quoted 6 lines
fs/xfs/libxfs/xfs_attr.c +2
-0@@ -460,6 +472,14 @@ xfs_attr_set(xfs_ilock (dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (args->trans, dp,0 ); + +if (args->value ||xfs_inode_hasattr (dp)) {Can this simply be "if (
iext_cnt!= 0)" ?
Quoted 5 lines
That would lead to a bug since
iext_cntis computed unconditionally at the beginning of the function. An extent count reservation will be attempted when xattr delete operation is executed against an inode which does not have an associated attr fork. This will causexfs_iext_count_may_overflow()to dereference the NULL pointer atxfs_inode->i_afp->if_nextents.
Ah, right, got it. This looks fine to me then...
--D
Quoted 41 lines
Quoted 34 lines
--D
Quoted 30 lines
fs/xfs/libxfs/xfs_attr.c +6
-0+ error =xfs_iext_count_may_overflow (dp,XFS_ATTR_FORK , + iext_cnt); +if (error) + goto out_trans_cancel; + } +if (args->value) {unsigned int quota_flags =XFS_QMOPT_RES_REGBLKS ;fs/xfs/libxfs/xfs_inode_fork.h +10
-0@@ -47,6 +47,16 @@ struct xfs_ifork {*/ #define XFS_IEXT_PUNCH_HOLE_CNT (1) +/* +* Adding/removing an xattr can cause XFS_DA_NODE_MAXDEPTH extents to +* be added. One extra extent for dabtree in case a local attr is +* large enough to cause a double split. It can also cause extent +* count to increase proportional to the size of a remote xattr's +* value. +*/ +#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ + (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* * Fork handling. */ Signature · 1 line
chandan
PATCH v11 5/14 xfs: Check for extent overflow when adding/removing dir entries
In reply to Chandan Babu RDirectory entry addition/removal can cause the following,
1. Data block can be added/removed.
A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
Same behaviour as described above for Data block.
3. Dabtree blocks.
XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH.
To be able to always remove an existing directory entry, when adding a new directory entry we make sure to reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
--- fs/xfs/libxfs/xfs_inode_fork.h | 13 +++++++++++++ fs/xfs/xfs_inode.c | 27 +++++++++++++++++++++++++++ fs/xfs/xfs_symlink.c | 5 +++++ 3 files changed, 45 insertions(+)
fs/xfs/libxfs/xfs_inode_fork.h +13 -0
@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */
fs/xfs/xfs_inode.c +27 -0
@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 ); +if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry
@@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 ); +if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else
@@ -2861,6 +2871,11 @@ xfs_remove(xfs_trans_ijoin (tp, dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * If we're removing a directory perform some additional validation. */
@@ -3221,6 +3236,18 @@ xfs_rename(if (wip)xfs_trans_ijoin (tp, wip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (src_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + +if (target_ip ==NULL ) { + error =xfs_iext_count_may_overflow (target_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 ); +if (error) + goto out_trans_cancel; + } +/* * If we are using project inheritance, we only allow renames * into our tree when the project IDs are the same; else the
fs/xfs/xfs_symlink.c +5 -0
@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 ); +if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */
Signature · 1 line
2.28.0
Re: [PATCH V11 05/14] xfs: Check for extent overflow when adding/removing dir entries
Quoted 51 lines · Chandan Babu R
Directory entry addition/removal can cause the following,
1. Data block can be added/removed.
A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
Same behaviour as described above for Data block.
3. Dabtree blocks.
XFS_DA_NODE_MAXDEPTHblocks can be added. Each of these can be new extents. Hence extent count can increase byXFS_DA_NODE_MAXDEPTH.To be able to always remove an existing directory entry, when adding a new directory entry we make sure to reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_inode_fork.h | 13 +++++++++++++ fs/xfs/xfs_inode.c | 27 +++++++++++++++++++++++++++ fs/xfs/xfs_symlink.c | 5 +++++ 3 files changed, 45 insertions(+)fs/xfs/libxfs/xfs_inode_fork.h +13
-0@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */ fs/xfs/xfs_inode.c +0
-0
fs/xfs/xfs_inode.c xfs_create( · line 1181
Quoted 5 lines
fs/xfs/xfs_inode.c +2
-0@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );
Er, why did these double since V10? We're only adding one entry, right?
fs/xfs/xfs_inode.c xfs_link( · line 1402
Quoted 12 lines
fs/xfs/xfs_inode.c +5
-0+if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry @@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );
Same question here.
Quoted 20 lines
fs/xfs/xfs_inode.c +8
-0+if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else @@ -2861,6 +2871,11 @@ xfs_remove(xfs_trans_ijoin (tp, dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * If we're removing a directory perform some additional validation. */ @@ -3221,6 +3236,18 @@ xfs_rename(if (wip)
fs/xfs/xfs_inode.c xfs_rename( · line 3246
Quoted 10 lines
fs/xfs/xfs_inode.c +8
-0xfs_trans_ijoin (tp, wip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (src_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + +if (target_ip ==NULL ) { + error =xfs_iext_count_may_overflow (target_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );
Why did this change to "<< 1" since V10?
I'm sorry, but I've lost my recollection on how the accounting works here. This seems (to me anyway ;)) a good candidate for a comment:
For a rename within the same dir where target_name doesn't yet exist, we are removing a name and then adding a name. We therefore check for iext overflow with (DIR_MANIP_CNT * 2), right? And I think that "target name does not exist" is synonymous with target_ip == NULL?
For a rename between dirs where the target name doesn't exist, we're removing src_name from src_dp and adding target_name to target_dp. Therefore we have to check for DIR_MANIP_CNT overflow on each of src_dp and target_dp, right?
For a rename where target_name /does/ exist, we're only removing the src_name, so we have to check for DIR_MANIP_CNT on src_dp, right?
For a RENAME_EXCHANGE we're not removing either name, so we don't need to check for iext overflow of src_dp or target_dp, right?
Quoted 12 lines
fs/xfs/xfs_inode.c +4
-0+if (error) + goto out_trans_cancel; + } +/* * If we are using project inheritance, we only allow renames * into our tree when the project IDs are the same; else the fs/xfs/xfs_symlink.c +0
-0
fs/xfs/xfs_symlink.c xfs_symlink( · line 224
Quoted 5 lines
fs/xfs/xfs_symlink.c +2
-0@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );
Same question as xfs_create.
--D
Quoted 9 lines
fs/xfs/xfs_symlink.c +3
-0+if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */ Signature · 1 line
2.28.0
Re: [PATCH V11 05/14] xfs: Check for extent overflow when adding/removing dir entries
Quoted 55 lines · Darrick J. Wong
Quoted 54 lines · Chandan Babu R
Directory entry addition/removal can cause the following,
1. Data block can be added/removed.
A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
Same behaviour as described above for Data block.
3. Dabtree blocks.
XFS_DA_NODE_MAXDEPTHblocks can be added. Each of these can be new extents. Hence extent count can increase byXFS_DA_NODE_MAXDEPTH.To be able to always remove an existing directory entry, when adding a new directory entry we make sure to reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_inode_fork.h | 13 +++++++++++++ fs/xfs/xfs_inode.c | 27 +++++++++++++++++++++++++++ fs/xfs/xfs_symlink.c | 5 +++++ 3 files changed, 45 insertions(+)fs/xfs/libxfs/xfs_inode_fork.h +13
-0@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */ fs/xfs/xfs_inode.c +0
-0@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel;
Quoted 4 lines
fs/xfs/xfs_inode.c xfs_create( · line 1181
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Er, why did these double since V10? We're only adding one entry, right?
To be able to always guarantee the removal of an existing directory entry, we reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
A bug was discovered when executing the following sequence of operations,
1. Keep inserting directory entries until the pseudo max extent count limit is reached.
2. At this stage, a directory entry remove operation will fail because it tries to reserve XFS_IEXT_DIR_MANIP_CNT(mp) worth of extent count. This reservation fails since the extent count would go over the pseudo max extent count limit as it did in step 1.
We would end up with a directory which can never be deleted.
Hence V11 doubles the extent count reservation for "directory entry insert" operations. The first XFS_IEXT_DIR_MANIP_CNT(mp) instance is for "insert" operation while the second XFS_IEXT_DIR_MANIP_CNT(mp) instance is for guaranteeing a possible future "remove" operation to succeed.
Quoted 11 lines
Quoted 11 lines
fs/xfs/xfs_inode.c +3
-0+if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry @@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL );
Quoted 4 lines
fs/xfs/xfs_inode.c xfs_link( · line 1402
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question here.
Creating a new hard link involves adding a new directory entry. Hence apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
Quoted 23 lines
Quoted 23 lines
fs/xfs/xfs_inode.c +8
-0+if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else @@ -2861,6 +2871,11 @@ xfs_remove(xfs_trans_ijoin (tp, dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * If we're removing a directory perform some additional validation. */ @@ -3221,6 +3236,18 @@ xfs_rename(if (wip)xfs_trans_ijoin (tp, wip,XFS_ILOCK_EXCL );
Quoted 10 lines
fs/xfs/xfs_inode.c xfs_rename( · line 3246
Quoted 8 lines
fs/xfs/xfs_inode.c +8
-0+ error =xfs_iext_count_may_overflow (src_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + +if (target_ip ==NULL ) { + error =xfs_iext_count_may_overflow (target_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Why did this change to "<< 1" since V10?
Extent count is doubled since this is essentially a directory insert operation w.r.t target_dp directory. One instance of XFS_IEXT_DIR_MANIP_CNT(mp) is for the directory entry being added to target_dp directory and another instance of XFS_IEXT_DIR_MANIP_CNT(mp) is for guaranteeing a future directory entry removal from target_dp directory to succeed.
Quoted 8 lines
I'm sorry, but I've lost my recollection on how the accounting works here. This seems (to me anyway ;)) a good candidate for a comment:
For a rename between dirs where the target name doesn't exist, we're removing
src_namefromsrc_dpand addingtarget_nametotarget_dp. Therefore we have to check forDIR_MANIP_CNToverflow on each ofsrc_dpandtarget_dp, right?
Extent count check is doubled since this is a directory insert operation w.r.t target_dp directory ... One instance of XFS_IEXT_DIR_MANIP_CNT(mp) is for the directory entry being added to target_dp directory and another instance of XFS_IEXT_DIR_MANIP_CNT(mp) is for guaranteeing a future directory entry removal from target_dp directory to succeed.
Since a directory entry is being removed from src_dp, reserving only a single instance of XFS_IEXT_DIR_MANIP_CNT(mp) would suffice.
Quoted 5 lines
For a rename within the same dir where
target_namedoesn't yet exist, we are removing a name and then adding a name. We therefore check for iext overflow with (DIR_MANIP_CNT* 2), right? And I think that "target name does not exist" is synonymous withtarget_ip== NULL?
Here again we have to reserve two instances of XFS_IEXT_DIR_MANIP_CNT(mp) for target_name insertion and one instance of XFS_IEXT_DIR_MANIP_CNT(mp) for src_name removal. This is because insertion and removal of src_name may each end up consuming XFS_IEXT_DIR_MANIP_CNT(mp) extent counts in the worst case. A future directory entry remove operation will require XFS_IEXT_DIR_MANIP_CNT(mp) extent counts to be reserved.
Also, You are right about "target name does not exist" being synonymous with target_ip == NULL.
Quoted 3 lines
For a rename where
target_name/does/ exist, we're only removing thesrc_name, so we have to check forDIR_MANIP_CNTonsrc_dp, right?
Yes, you are right.
Quoted 3 lines
For a
RENAME_EXCHANGEwe're not removing either name, so we don't need to check for iext overflow ofsrc_dportarget_dp, right?
You are right. Sorry, I missed this. I will move the extent count reservation logic to come after the invocation of xfs_cross_rename().
I will also add appropriate comments into xfs_rename() describing the scenarios that have been discussed above.
PS: I have swapped the order of two comments from your original reply since I think it is easier to explain the scenarios with the order of comments/questions swapped.
Quoted 16 lines
Quoted 16 lines
fs/xfs/xfs_inode.c +4
-0+if (error) + goto out_trans_cancel; + } +/* * If we are using project inheritance, we only allow renames * into our tree when the project IDs are the same; else the fs/xfs/xfs_symlink.c +0
-0@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel;
Quoted 4 lines
fs/xfs/xfs_symlink.c xfs_symlink( · line 224
Quoted 2 lines
fs/xfs/xfs_symlink.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question as
xfs_create.
This is again similar to adding a new directory entry. Hence, apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
Quoted 10 lines
--D
Quoted 6 lines
fs/xfs/xfs_symlink.c +3
-0+if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */
Signature · 1 line
chandan
Re: [PATCH V11 05/14] xfs: Check for extent overflow when adding/removing dir entries
Quoted 74 lines · Chandan Babu R
Quoted 55 lines · Darrick J. Wong
Quoted 54 lines · Chandan Babu R
Directory entry addition/removal can cause the following,
1. Data block can be added/removed.
A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
Same behaviour as described above for Data block.
3. Dabtree blocks.
XFS_DA_NODE_MAXDEPTHblocks can be added. Each of these can be new extents. Hence extent count can increase byXFS_DA_NODE_MAXDEPTH.To be able to always remove an existing directory entry, when adding a new directory entry we make sure to reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_inode_fork.h | 13 +++++++++++++ fs/xfs/xfs_inode.c | 27 +++++++++++++++++++++++++++ fs/xfs/xfs_symlink.c | 5 +++++ 3 files changed, 45 insertions(+)fs/xfs/libxfs/xfs_inode_fork.h +13
-0@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */ fs/xfs/xfs_inode.c +0
-0@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel;Quoted 4 lines
fs/xfs/xfs_inode.c xfs_create( · line 1181
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Er, why did these double since V10? We're only adding one entry, right?
To be able to always guarantee the removal of an existing directory entry, we reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
A bug was discovered when executing the following sequence of operations,
1. Keep inserting directory entries until the pseudo max extent count limit is reached.
2. At this stage, a directory entry remove operation will fail because it tries to reserveXFS_IEXT_DIR_MANIP_CNT(mp)worth of extent count. This reservation fails since the extent count would go over the pseudo max extent count limit as it did in step 1.
Quoted 1 line
We would end up with a directory which can never be deleted.
I just found that reserving an extra XFS_IEXT_DIR_MANIP_CNT(mp) extent count, when performing a directory insert operation, would not prevent us from ending up with a directory which can never be deleted.
Let x be a directory's data fork extent count and lets assume its value to be,
x = MAX_EXT_COUNT - XFS_IEXT_DIR_MANIP_CNT(mp)
So in this case we do have sufficient "extent count" to be able to perform a directory entry remove operation. But the directory remove operation itself can cause extent count to increase by XFS_IEXT_DIR_MANIP_CNT(mp) units in the worst case. This happens when freeing 5 dabtree blocks, one data block and one free block causes file extents to be split for each of the above mentioned blocks.
If on the other hand, the current value of 'x' were,
x = MAX_EXT_COUNT - (2 * XFS_IEXT_DIR_MANIP_CNT(mp))
'x' can still reach MAX_EXT_COUNT if two consecutive directory remove operations can each cause extent count to increase by XFS_IEXT_DIR_MANIP_CNT(mp).
IMHO there is no way to prevent a directory from becoming un-deletable once its data fork extent count reaches close to MAX_EXT_COUNT. The other choice of not checking for extent overflow would mean silent data corruption. Hence maybe the former result is better one to go with.
W.r.t xattrs, not reserving an extra XFS_IEXT_ATTR_MANIP_CNT(mp) extent count units would prevent the user from removing xattrs when the inode's attr fork extent count value is close to MAX_EXT_COUNT. However, the file and the associated extents will be removed during file deletion operation.
Quoted 157 lines
Hence V11 doubles the extent count reservation for "directory entry insert" operations. The first
XFS_IEXT_DIR_MANIP_CNT(mp)instance is for "insert" operation while the secondXFS_IEXT_DIR_MANIP_CNT(mp)instance is for guaranteeing a possible future "remove" operation to succeed.Quoted 11 lines
Quoted 11 lines
fs/xfs/xfs_inode.c +3
-0+if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry @@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL );Quoted 4 lines
fs/xfs/xfs_inode.c xfs_link( · line 1402
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question here.
Creating a new hard link involves adding a new directory entry. Hence apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
Quoted 23 lines
Quoted 23 lines
fs/xfs/xfs_inode.c +8
-0+if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else @@ -2861,6 +2871,11 @@ xfs_remove(xfs_trans_ijoin (tp, dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * If we're removing a directory perform some additional validation. */ @@ -3221,6 +3236,18 @@ xfs_rename(if (wip)xfs_trans_ijoin (tp, wip,XFS_ILOCK_EXCL );Quoted 10 lines
fs/xfs/xfs_inode.c xfs_rename( · line 3246
Quoted 8 lines
fs/xfs/xfs_inode.c +8
-0+ error =xfs_iext_count_may_overflow (src_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + +if (target_ip ==NULL ) { + error =xfs_iext_count_may_overflow (target_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Why did this change to "<< 1" since V10?
Extent count is doubled since this is essentially a directory insert operation w.r.t
target_dpdirectory. One instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for the directory entry being added totarget_dpdirectory and another instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for guaranteeing a future directory entry removal fromtarget_dpdirectory to succeed.Quoted 8 lines
I'm sorry, but I've lost my recollection on how the accounting works here. This seems (to me anyway ;)) a good candidate for a comment:
For a rename between dirs where the target name doesn't exist, we're removing
src_namefromsrc_dpand addingtarget_nametotarget_dp. Therefore we have to check forDIR_MANIP_CNToverflow on each ofsrc_dpandtarget_dp, right?Extent count check is doubled since this is a directory insert operation w.r.t
target_dpdirectory ... One instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for the directory entry being added totarget_dpdirectory and another instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for guaranteeing a future directory entry removal fromtarget_dpdirectory to succeed.Since a directory entry is being removed from
src_dp, reserving only a single instance ofXFS_IEXT_DIR_MANIP_CNT(mp)would suffice.Quoted 5 lines
For a rename within the same dir where
target_namedoesn't yet exist, we are removing a name and then adding a name. We therefore check for iext overflow with (DIR_MANIP_CNT* 2), right? And I think that "target name does not exist" is synonymous withtarget_ip== NULL?Here again we have to reserve two instances of
XFS_IEXT_DIR_MANIP_CNT(mp)fortarget_nameinsertion and one instance ofXFS_IEXT_DIR_MANIP_CNT(mp)forsrc_nameremoval. This is because insertion and removal ofsrc_namemay each end up consumingXFS_IEXT_DIR_MANIP_CNT(mp)extent counts in the worst case. A future directory entry remove operation will requireXFS_IEXT_DIR_MANIP_CNT(mp)extent counts to be reserved.Also, You are right about "target name does not exist" being synonymous with
target_ip== NULL.Quoted 3 lines
For a rename where
target_name/does/ exist, we're only removing thesrc_name, so we have to check forDIR_MANIP_CNTonsrc_dp, right?Yes, you are right.
Quoted 3 lines
For a
RENAME_EXCHANGEwe're not removing either name, so we don't need to check for iext overflow ofsrc_dportarget_dp, right?You are right. Sorry, I missed this. I will move the extent count reservation logic to come after the invocation of
xfs_cross_rename().I will also add appropriate comments into
xfs_rename()describing the scenarios that have been discussed above.PS: I have swapped the order of two comments from your original reply since I think it is easier to explain the scenarios with the order of comments/questions swapped.
Quoted 16 lines
Quoted 16 lines
fs/xfs/xfs_inode.c +4
-0+if (error) + goto out_trans_cancel; + } +/* * If we are using project inheritance, we only allow renames * into our tree when the project IDs are the same; else the fs/xfs/xfs_symlink.c +0
-0@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel;Quoted 4 lines
fs/xfs/xfs_symlink.c xfs_symlink( · line 224
Quoted 2 lines
fs/xfs/xfs_symlink.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question as
xfs_create.This is again similar to adding a new directory entry. Hence, apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
Quoted 10 lines
--D
Quoted 6 lines
fs/xfs/xfs_symlink.c +3
-0+if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */
Signature · 1 line
chandan
Re: [PATCH V11 05/14] xfs: Check for extent overflow when adding/removing dir entries
Quoted 96 lines · Chandan Babu R
Quoted 74 lines · Chandan Babu R
Quoted 55 lines · Darrick J. Wong
Quoted 54 lines · Chandan Babu R
Directory entry addition/removal can cause the following,
1. Data block can be added/removed.
A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
Same behaviour as described above for Data block.
3. Dabtree blocks.
XFS_DA_NODE_MAXDEPTHblocks can be added. Each of these can be new extents. Hence extent count can increase byXFS_DA_NODE_MAXDEPTH.To be able to always remove an existing directory entry, when adding a new directory entry we make sure to reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_inode_fork.h | 13 +++++++++++++ fs/xfs/xfs_inode.c | 27 +++++++++++++++++++++++++++ fs/xfs/xfs_symlink.c | 5 +++++ 3 files changed, 45 insertions(+)fs/xfs/libxfs/xfs_inode_fork.h +13
-0@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */ fs/xfs/xfs_inode.c +0
-0@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel;Quoted 4 lines
fs/xfs/xfs_inode.c xfs_create( · line 1181
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Er, why did these double since V10? We're only adding one entry, right?
To be able to always guarantee the removal of an existing directory entry, we reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
A bug was discovered when executing the following sequence of operations,
1. Keep inserting directory entries until the pseudo max extent count limit is reached.
2. At this stage, a directory entry remove operation will fail because it tries to reserveXFS_IEXT_DIR_MANIP_CNT(mp)worth of extent count. This reservation fails since the extent count would go over the pseudo max extent count limit as it did in step 1.Quoted 1 line
We would end up with a directory which can never be deleted.
I just found that reserving an extra
XFS_IEXT_DIR_MANIP_CNT(mp)extent count, when performing a directory insert operation, would not prevent us from ending up with a directory which can never be deleted.Let x be a directory's data fork extent count and lets assume its value to be,
x =
MAX_EXT_COUNT-XFS_IEXT_DIR_MANIP_CNT(mp)So in this case we do have sufficient "extent count" to be able to perform a directory entry remove operation. But the directory remove operation itself can cause extent count to increase by
XFS_IEXT_DIR_MANIP_CNT(mp)units in the worst case. This happens when freeing 5 dabtree blocks, one data block and one free block causes file extents to be split for each of the above mentioned blocks.If on the other hand, the current value of 'x' were,
x =
MAX_EXT_COUNT- (2 *XFS_IEXT_DIR_MANIP_CNT(mp))
Quoted 8 lines
'x' can still reach
MAX_EXT_COUNTif two consecutive directory remove operations can each cause extent count to increase byXFS_IEXT_DIR_MANIP_CNT(mp).IMHO there is no way to prevent a directory from becoming un-deletable once its data fork extent count reaches close to
MAX_EXT_COUNT. The other choice of not checking for extent overflow would mean silent data corruption. Hence maybe the former result is better one to go with.
So in other words you're doubling the amount you pass into the overflow check so that we can guarantee that a future dirent removal will work.
In other words, the doubling is to preserve future functionality, and is not required by the create() call itself. This should be captured in a comment above the call to xfs_iext_count_may_overflow.
Or I guess you could create an XFS_IEXT_DIRENT_CREATE macro that wraps all that (along with that comment explaining why).
Quoted 4 lines
W.r.t xattrs, not reserving an extra
XFS_IEXT_ATTR_MANIP_CNT(mp)extent count units would prevent the user from removing xattrs when the inode's attr fork extent count value is close toMAX_EXT_COUNT. However, the file and the associated extents will be removed during file deletion operation.
<shrug> I doubt xattr trees often get close to 64k extents, so you might as well apply the same logic to them. Better to cut off the user early than to force them to delete the whole file just to wipe out the xattrs.
Quoted 17 lines
Quoted 17 lines
Hence V11 doubles the extent count reservation for "directory entry insert" operations. The first
XFS_IEXT_DIR_MANIP_CNT(mp)instance is for "insert" operation while the secondXFS_IEXT_DIR_MANIP_CNT(mp)instance is for guaranteeing a possible future "remove" operation to succeed.Quoted 11 lines
Quoted 11 lines
fs/xfs/xfs_inode.c +3
-0+if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry @@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL );
Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
fs/xfs/xfs_inode.c xfs_link( · line 1402
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question here.
Creating a new hard link involves adding a new directory entry. Hence apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
In other words, we also want XFS_IEXT_DIRENT_CREATE here?
Quoted 35 lines
Quoted 35 lines
Quoted 34 lines
Quoted 22 lines
fs/xfs/xfs_inode.c +8
-0+if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else @@ -2861,6 +2871,11 @@ xfs_remove(xfs_trans_ijoin (tp, dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * If we're removing a directory perform some additional validation. */ @@ -3221,6 +3236,18 @@ xfs_rename(if (wip)fs/xfs/xfs_inode.c xfs_rename( · line 3246
Quoted 10 lines
fs/xfs/xfs_inode.c +8
-0xfs_trans_ijoin (tp, wip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (src_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + +if (target_ip ==NULL ) { + error =xfs_iext_count_may_overflow (target_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Why did this change to "<< 1" since V10?
Quoted 5 lines
Quoted 5 lines
Extent count is doubled since this is essentially a directory insert operation w.r.t
target_dpdirectory. One instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for the directory entry being added totarget_dpdirectory and another instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for guaranteeing a future directory entry removal fromtarget_dpdirectory to succeed.
...and here too?
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
I'm sorry, but I've lost my recollection on how the accounting works here. This seems (to me anyway ;)) a good candidate for a comment:
For a rename between dirs where the target name doesn't exist, we're removing
src_namefromsrc_dpand addingtarget_nametotarget_dp. Therefore we have to check forDIR_MANIP_CNToverflow on each ofsrc_dpandtarget_dp, right?
Quoted 5 lines
Quoted 5 lines
Extent count check is doubled since this is a directory insert operation w.r.t
target_dpdirectory ... One instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for the directory entry being added totarget_dpdirectory and another instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for guaranteeing a future directory entry removal fromtarget_dpdirectory to succeed.
Or in other words, another place for XFS_IEXT_DIRENT_CREATE...
Quoted 2 lines
Quoted 2 lines
Since a directory entry is being removed from
src_dp, reserving only a single instance ofXFS_IEXT_DIR_MANIP_CNT(mp)would suffice.
<nod>
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
For a rename within the same dir where
target_namedoesn't yet exist, we are removing a name and then adding a name. We therefore check for iext overflow with (DIR_MANIP_CNT* 2), right? And I think that "target name does not exist" is synonymous withtarget_ip== NULL?
Quoted 6 lines
Quoted 6 lines
Here again we have to reserve two instances of
XFS_IEXT_DIR_MANIP_CNT(mp)fortarget_nameinsertion and one instance ofXFS_IEXT_DIR_MANIP_CNT(mp)forsrc_nameremoval. This is because insertion and removal ofsrc_namemay each end up consumingXFS_IEXT_DIR_MANIP_CNT(mp)extent counts in the worst case. A future directory entry remove operation will requireXFS_IEXT_DIR_MANIP_CNT(mp)extent counts to be reserved.
...and another place for DIRENT_CREATE...
Quoted 8 lines
Quoted 8 lines
Also, You are right about "target name does not exist" being synonymous with
target_ip== NULL.Quoted 3 lines
For a rename where
target_name/does/ exist, we're only removing thesrc_name, so we have to check forDIR_MANIP_CNTonsrc_dp, right?
Quoted 8 lines
Quoted 8 lines
Yes, you are right.
Quoted 3 lines
For a
RENAME_EXCHANGEwe're not removing either name, so we don't need to check for iext overflow ofsrc_dportarget_dp, right?You are right. Sorry, I missed this. I will move the extent count reservation logic to come after the invocation of
xfs_cross_rename().
Ok.
Quoted 2 lines
Quoted 2 lines
I will also add appropriate comments into
xfs_rename()describing the scenarios that have been discussed above.
Thanks.
Quoted 3 lines
Quoted 3 lines
PS: I have swapped the order of two comments from your original reply since I think it is easier to explain the scenarios with the order of comments/questions swapped.
Ok.
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
fs/xfs/xfs_inode.c +4
-0+if (error) + goto out_trans_cancel; + } +/* * If we are using project inheritance, we only allow renames * into our tree when the project IDs are the same; else the fs/xfs/xfs_symlink.c +0
-0@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel;
Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
fs/xfs/xfs_symlink.c xfs_symlink( · line 224
Quoted 2 lines
fs/xfs/xfs_symlink.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question as
xfs_create.This is again similar to adding a new directory entry. Hence, apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
...and here yet another place to use XFS_IEXT_DIRENT_CREATE?
--D
Quoted 20 lines
Quoted 13 lines
Quoted 11 lines
--D
Quoted 6 lines
fs/xfs/xfs_symlink.c +3
-0+if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */ Signature · 1 line
chandan
Re: [PATCH V11 05/14] xfs: Check for extent overflow when adding/removing dir entries
Quoted 109 lines · Darrick J. Wong
Quoted 96 lines · Chandan Babu R
Quoted 74 lines · Chandan Babu R
Quoted 55 lines · Darrick J. Wong
Quoted 54 lines · Chandan Babu R
Directory entry addition/removal can cause the following,
1. Data block can be added/removed.
A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
Same behaviour as described above for Data block.
3. Dabtree blocks.
XFS_DA_NODE_MAXDEPTHblocks can be added. Each of these can be new extents. Hence extent count can increase byXFS_DA_NODE_MAXDEPTH.To be able to always remove an existing directory entry, when adding a new directory entry we make sure to reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_inode_fork.h | 13 +++++++++++++ fs/xfs/xfs_inode.c | 27 +++++++++++++++++++++++++++ fs/xfs/xfs_symlink.c | 5 +++++ 3 files changed, 45 insertions(+)fs/xfs/libxfs/xfs_inode_fork.h +13
-0@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */ fs/xfs/xfs_inode.c +0
-0@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel;Quoted 4 lines
fs/xfs/xfs_inode.c xfs_create( · line 1181
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Er, why did these double since V10? We're only adding one entry, right?
To be able to always guarantee the removal of an existing directory entry, we reserve inode fork extent count required for removing a directory entry in addition to that required for the directory entry add operation.
A bug was discovered when executing the following sequence of operations,
1. Keep inserting directory entries until the pseudo max extent count limit is reached.
2. At this stage, a directory entry remove operation will fail because it tries to reserveXFS_IEXT_DIR_MANIP_CNT(mp)worth of extent count. This reservation fails since the extent count would go over the pseudo max extent count limit as it did in step 1.Quoted 1 line
We would end up with a directory which can never be deleted.
I just found that reserving an extra
XFS_IEXT_DIR_MANIP_CNT(mp)extent count, when performing a directory insert operation, would not prevent us from ending up with a directory which can never be deleted.Let x be a directory's data fork extent count and lets assume its value to be,
x =
MAX_EXT_COUNT-XFS_IEXT_DIR_MANIP_CNT(mp)So in this case we do have sufficient "extent count" to be able to perform a directory entry remove operation. But the directory remove operation itself can cause extent count to increase by
XFS_IEXT_DIR_MANIP_CNT(mp)units in the worst case. This happens when freeing 5 dabtree blocks, one data block and one free block causes file extents to be split for each of the above mentioned blocks.If on the other hand, the current value of 'x' were,
x =
MAX_EXT_COUNT- (2 *XFS_IEXT_DIR_MANIP_CNT(mp))Quoted 8 lines
'x' can still reach
MAX_EXT_COUNTif two consecutive directory remove operations can each cause extent count to increase byXFS_IEXT_DIR_MANIP_CNT(mp).IMHO there is no way to prevent a directory from becoming un-deletable once its data fork extent count reaches close to
MAX_EXT_COUNT. The other choice of not checking for extent overflow would mean silent data corruption. Hence maybe the former result is better one to go with.So in other words you're doubling the amount you pass into the overflow check so that we can guarantee that a future dirent removal will work.
Quoted 6 lines
In other words, the doubling is to preserve future functionality, and is not required by the
create()call itself. This should be captured in a comment above the call toxfs_iext_count_may_overflow.Or I guess you could create an
XFS_IEXT_DIRENT_CREATEmacro that wraps all that (along with that comment explaining why).
Sorry, the previous explaination I had provided earlier was not clear enough. I hope the following will add more clarity.
The bottom line is that extent count reservation cannot guarantee that a future "directory entry remove" operation will always have sufficient amount of "extent count" left in the inode's data fork extent count.
Doubling the extent count that gets reserved during directory entry insert operation can leave a minimum of XFS_IEXT_DIR_MANIP_CNT extent count units free after the operation is completed. But this extent count can be consumed away by a directory entry remove operation that ends up freeing directory blocks due to the fact that it can cause extent count of directory inode to increase by XFS_IEXT_DIR_MANIP_CNT units.
For example, Assume that a directory's data fork extent count is X and it has the following extent record in its bmbt,
| Y | Y+1 | Y+2 |
Here Y is the offset within the directory. The directory blocks are within the same extent record since they are contiguous in terms of both file offset and disk offset. Now, if a directory entry remove operation is executed, it can free the block Y+1. This causes an increase in the extent count.
So a single directory entry remove operation has the potential to increase extent count by XFS_IEXT_DIR_MANIP_CNT units in the worst case (one data block, one free disk block and 5 dabtree blocks). Hence reserving an extra XFS_IEXT_DIR_MANIP_CNT units of extent count during directory entry insertion would not help solve the problem.
The above mentioned scenario can be further extended:
If a directory has (2 * XFS_IEXT_DIR_MANIP_CNT) units of free extent count left, two directory entry remove operations can potentially increase extent count by XFS_IEXT_DIR_MANIP_CNT units each and hence the execution of a third consecutive directory entry remove operation fails due to lack of availability of free extent count.
A related problem has already been solved in XFS.
File deletion can fail in the case of low disk space scenarios. This happens because of failure to successfully reserve disk blocks for "directory entry remove" transaction. In such a case, xfs_remove() allocates a transaction with tp->t_blk_res set to 0. During the execution of the operation, if we end up having to remove a block (say a "directory data block") from the directory, the following events could occur,
1. xfs_bmap_del_extent_real(): Extent count increases because the block that is being unmapped from bmbt occurs in the middle of an extent record.
2. We truncate the length of the existing extent record and try to insert a new extent record which maps the blocks of the original extent record that occurs after the block being freed.
3. The following sequence of functions are invoked if inserting a new record requires a btree block to be split,
xfs_btree_insert() => xfs_btree_insrec() => xfs_btree_make_block_unfull() => __xfs_btree_split() => xfs_bmbt_alloc_block()
4. xfs_bmbt_alloc_block() returns -ENOSPC when it notices tp->t_blk_res having a value of 0.
5. Upon receiving -ENOSPC return value, xfs_bmap_del_extent_real() restores the original extent record.
6. Invokers of xfs_dir2_shrink_inode() (e.g. xfs_dir3_data_block_free()) would ignore the -ENOSPC error code and hence the corresponding directory block is not freed. It is left to a future user of the block to be able to free it.
I think we have to use a similar approach to solve the "undeletable directory" problem. To that end, I have written a patch which implements the following logic,
W.r.t directory entry remove operation, we check for extent count overflow in xfs_bmap_del_extent_real() only when the block being unmapped could cause an increase the extent count overflow. If unmapping can cause extent count to overflow, xfs_bmap_del_extent_real() would return -ENOSPC, causing the invokers of xfs_dir2_shrink_inode() to ignore -ENOSPC and leaving the responsibility of freeing the directory block for future.
For "directory rename operation", the explanation provided for "directory entry remove" operation holds for the "source directory entry". For rename's destination directory entry, we now check for extent overflow only when we have successfully reserved non-zero blocks for the transaction. This is because with zero block-sized reservation, the rename either
1. Fails due to non-availability of space in existing directory blocks for holding the new directory entry.
2. Succeeds since the directory has enough space in existing blocks to hold the new directory entry. In this case, XFS wouldn't add new blocks to the directory.
For the remaining directory operations (e.g. create, link and symlink) we continue to reserve XFS_IEXT_ATTR_MANIP_CNT units of extent count before the corresponding transaction starts dirtying metadata.
fs/xfs/libxfs/xfs_bmap.c +6 -0
@@ -5050,6 +5050,12 @@ xfs_bmap_del_extent_real( del->br_startoff > got.br_startoff && del_endoff < got_endoff)return -ENOSPC ; +if (S_ISDIR (VFS_I (ip)->i_mode) && + whichfork ==XFS_DATA_FORK && + del->br_startoff > got.br_startoff && del_endoff < got_endoff && +xfs_iext_count_may_overflow (ip, whichfork,1 )) +return -ENOSPC ; + flags =XFS_ILOG_CORE ;if (whichfork ==XFS_DATA_FORK &&XFS_IS_REALTIME_INODE (ip)) {xfs_filblks_t len;
fs/xfs/libxfs/xfs_inode_fork.h +13 -0
@@ -57,6 +57,19 @@ struct xfs_ifork {#define XFS_IEXT_ATTR_MANIP_CNT (rmt_blks) \ (XFS_DA_NODE_MAXDEPTH + max(1, rmt_blks)) +/* +* Directory entry addition/removal can cause the following, +* 1. Data block can be added/removed. +* A new extent can cause extent count to increase by 1. +* 2. Free disk block can be added/removed. +* Same behaviour as described above for Data block. +* 3. Dabtree blocks. +* XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these can be new +* extents. Hence extent count can increase by XFS_DA_NODE_MAXDEPTH. +*/ +#define XFS_IEXT_DIR_MANIP_CNT (mp) \ + ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* * Fork handling. */
fs/xfs/xfs_inode.c +25 -0
@@ -1177,6 +1177,11 @@ xfs_create(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry
@@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else
@@ -3246,12 +3256,27 @@ xfs_rename(/* * If there's no space reservation, check the entry will * fit before actually inserting it. +* +* If the entry does fit in, then there is no need to check for +* extent count overflow since no new extents will be added to +* the directory's data fork. */ if (!spaceres) { error =xfs_dir_canenter (tp, target_dp, target_name);if (error) goto out_trans_cancel; } +/* +* Otherwise, Check if inserting the new entry can cause extent +* count to overflow. +*/ +else { + error =xfs_iext_count_may_overflow (target_dp, +XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + } }else {/* * If target exists and it's a directory, check that whether
fs/xfs/xfs_symlink.c +5 -0
@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel; + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */
The ideal way to test this patch would be to have a directory whose extent count has reached the maximum limit and also has an extent record with atleast three blocks. Freeing directory entries occupying the middle block in such an extent record would trigger the -ENOSPC error code handling described above.
However, I don't think it is possible to deterministically create such a directory layout by executing commands from userspace. Hence my testing was limited to,
1. Fill a directory with entries until the maximum extent count limit is reached. Remove the corresponding directory. The remove operation should succeed.
2. Regular run of fstests with various mount options.
Quoted 9 lines
Quoted 5 lines
W.r.t xattrs, not reserving an extra
XFS_IEXT_ATTR_MANIP_CNT(mp)extent count units would prevent the user from removing xattrs when the inode's attr fork extent count value is close toMAX_EXT_COUNT. However, the file and the associated extents will be removed during file deletion operation.<shrug> I doubt xattr trees often get close to 64k extents, so you might as well apply the same logic to them. Better to cut off the user early than to force them to delete the whole file just to wipe out the xattrs.
As described above reserving twice the amount of extent count units during xattr insertion would not guarantee a future xattr remove operation to obtain extent count reservation successfully. Hence to always allow xattr remove operation, we have to implement some of the logic associated with "directory remove operation". This includes adding the ability to swap dabtree block (whose removal can cause -ENOSPC to be returned from xfs_bunmapi()) with the last block of the xattr dabtree (i.e. the logic implemented by xfs_da3_swap_lastblock()). Please let me know if you prefer this approach to be implemented instead of file deletion.
Quoted 186 lines
Quoted 18 lines
Quoted 18 lines
Hence V11 doubles the extent count reservation for "directory entry insert" operations. The first
XFS_IEXT_DIR_MANIP_CNT(mp)instance is for "insert" operation while the secondXFS_IEXT_DIR_MANIP_CNT(mp)instance is for guaranteeing a possible future "remove" operation to succeed.Quoted 11 lines
Quoted 11 lines
fs/xfs/xfs_inode.c +3
-0+if (error) + goto out_trans_cancel; +/* * A newly created regular or special file just has one directory * entry pointing to them, but a directory also the "." entry @@ -1393,6 +1398,11 @@ xfs_link(xfs_trans_ijoin (tp, sip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, tdp,XFS_ILOCK_EXCL );Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
fs/xfs/xfs_inode.c xfs_link( · line 1402
Quoted 2 lines
fs/xfs/xfs_inode.c +2
-0+ error =xfs_iext_count_may_overflow (tdp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question here.
Creating a new hard link involves adding a new directory entry. Hence apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
In other words, we also want
XFS_IEXT_DIRENT_CREATEhere?Quoted 35 lines
Quoted 35 lines
Quoted 34 lines
Quoted 22 lines
fs/xfs/xfs_inode.c +8
-0+if (error) + goto error_return; +/* * If we are using project inheritance, we only allow hard link * creation in our tree when the project IDs are the same; else @@ -2861,6 +2871,11 @@ xfs_remove(xfs_trans_ijoin (tp, dp,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; +/* * If we're removing a directory perform some additional validation. */ @@ -3221,6 +3236,18 @@ xfs_rename(if (wip)fs/xfs/xfs_inode.c xfs_rename( · line 3246
Quoted 10 lines
fs/xfs/xfs_inode.c +8
-0xfs_trans_ijoin (tp, wip,XFS_ILOCK_EXCL ); + error =xfs_iext_count_may_overflow (src_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp)); +if (error) + goto out_trans_cancel; + +if (target_ip ==NULL ) { + error =xfs_iext_count_may_overflow (target_dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Why did this change to "<< 1" since V10?
Quoted 5 lines
Quoted 5 lines
Extent count is doubled since this is essentially a directory insert operation w.r.t
target_dpdirectory. One instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for the directory entry being added totarget_dpdirectory and another instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for guaranteeing a future directory entry removal fromtarget_dpdirectory to succeed....and here too?
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
I'm sorry, but I've lost my recollection on how the accounting works here. This seems (to me anyway ;)) a good candidate for a comment:
For a rename between dirs where the target name doesn't exist, we're removing
src_namefromsrc_dpand addingtarget_nametotarget_dp. Therefore we have to check forDIR_MANIP_CNToverflow on each ofsrc_dpandtarget_dp, right?Quoted 5 lines
Quoted 5 lines
Extent count check is doubled since this is a directory insert operation w.r.t
target_dpdirectory ... One instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for the directory entry being added totarget_dpdirectory and another instance ofXFS_IEXT_DIR_MANIP_CNT(mp)is for guaranteeing a future directory entry removal fromtarget_dpdirectory to succeed.Or in other words, another place for
XFS_IEXT_DIRENT_CREATE...Quoted 2 lines
Quoted 2 lines
Since a directory entry is being removed from
src_dp, reserving only a single instance ofXFS_IEXT_DIR_MANIP_CNT(mp)would suffice.<nod>
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
For a rename within the same dir where
target_namedoesn't yet exist, we are removing a name and then adding a name. We therefore check for iext overflow with (DIR_MANIP_CNT* 2), right? And I think that "target name does not exist" is synonymous withtarget_ip== NULL?Quoted 6 lines
Quoted 6 lines
Here again we have to reserve two instances of
XFS_IEXT_DIR_MANIP_CNT(mp)fortarget_nameinsertion and one instance ofXFS_IEXT_DIR_MANIP_CNT(mp)forsrc_nameremoval. This is because insertion and removal ofsrc_namemay each end up consumingXFS_IEXT_DIR_MANIP_CNT(mp)extent counts in the worst case. A future directory entry remove operation will requireXFS_IEXT_DIR_MANIP_CNT(mp)extent counts to be reserved....and another place for
DIRENT_CREATE...Quoted 8 lines
Quoted 8 lines
Also, You are right about "target name does not exist" being synonymous with
target_ip== NULL.Quoted 3 lines
For a rename where
target_name/does/ exist, we're only removing thesrc_name, so we have to check forDIR_MANIP_CNTonsrc_dp, right?Quoted 8 lines
Quoted 8 lines
Yes, you are right.
Quoted 3 lines
For a
RENAME_EXCHANGEwe're not removing either name, so we don't need to check for iext overflow ofsrc_dportarget_dp, right?You are right. Sorry, I missed this. I will move the extent count reservation logic to come after the invocation of
xfs_cross_rename().Ok.
Quoted 2 lines
Quoted 2 lines
I will also add appropriate comments into
xfs_rename()describing the scenarios that have been discussed above.Thanks.
Quoted 3 lines
Quoted 3 lines
PS: I have swapped the order of two comments from your original reply since I think it is easier to explain the scenarios with the order of comments/questions swapped.
Ok.
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
fs/xfs/xfs_inode.c +4
-0+if (error) + goto out_trans_cancel; + } +/* * If we are using project inheritance, we only allow renames * into our tree when the project IDs are the same; else the fs/xfs/xfs_symlink.c +0
-0@@ -220,6 +220,11 @@ xfs_symlink(if (error) goto out_trans_cancel;Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
fs/xfs/xfs_symlink.c xfs_symlink( · line 224
Quoted 2 lines
fs/xfs/xfs_symlink.c +2
-0+ error =xfs_iext_count_may_overflow (dp,XFS_DATA_FORK , +XFS_IEXT_DIR_MANIP_CNT (mp) <<1 );Same question as
xfs_create.This is again similar to adding a new directory entry. Hence, apart from reserving extent count for directory entry addition we will have to reserve extent count for a future directory entry removal as well.
...and here yet another place to use
XFS_IEXT_DIRENT_CREATE?--D
Quoted 15 lines
Quoted 13 lines
Quoted 11 lines
--D
Quoted 6 lines
fs/xfs/xfs_symlink.c +3
-0+if (error) + goto out_trans_cancel; +/* * Allocate an inode for the symlink. */
Signature · 1 line
chandan
PATCH v11 6/14 xfs: Check for extent overflow when writing to unwritten extent
In reply to Chandan Babu RA write to a sub-interval of an existing unwritten extent causes the original extent to be split into 3 extents
i.e. | Unwritten | Real | Unwritten |
Hence extent count can increase by 2.
--- fs/xfs/libxfs/xfs_inode_fork.h | 8 ++++++++ fs/xfs/xfs_iomap.c | 5 +++++ 2 files changed, 13 insertions(+)
fs/xfs/libxfs/xfs_inode_fork.h +8 -0
@@ -70,6 +70,14 @@ struct xfs_ifork {#define XFS_IEXT_DIR_MANIP_CNT (mp) \ ((XFS_DA_NODE_MAXDEPTH + 1 + 1) * (mp)->m_dir_geo->fsbcount) +/* +* A write to a sub-interval of an existing unwritten extent causes the original +* extent to be split into 3 extents +* i.e. | Unwritten | Real | Unwritten | +* Hence extent count can increase by 2. +*/ +#define XFS_IEXT_WRITE_UNWRITTEN_CNT (2) +/* * Fork handling. */
fs/xfs/xfs_iomap.c +5 -0
@@ -566,6 +566,11 @@ xfs_iomap_write_unwritten(if (error) goto error_on_bmapi_transaction; + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_WRITE_UNWRITTEN_CNT ); +if (error) + goto error_on_bmapi_transaction; +/* * Modify the unwritten extent state of the buffer. */
Signature · 1 line
2.28.0
PATCH v11 7/14 xfs: Check for extent overflow when moving extent from cow to data fork
In reply to Chandan Babu RMoving an extent to data fork can cause a sub-interval of an existing extent to be unmapped. This will increase extent count by 1. Mapping in the new extent can increase the extent count by 1 again i.e.
| Old extent | New extent | Old extent |
Hence number of extents increases by 2.
--- fs/xfs/libxfs/xfs_inode_fork.h | 9 +++++++++ fs/xfs/xfs_reflink.c | 5 +++++ 2 files changed, 14 insertions(+)
fs/xfs/libxfs/xfs_inode_fork.h +9 -0
@@ -78,6 +78,15 @@ struct xfs_ifork {*/ #define XFS_IEXT_WRITE_UNWRITTEN_CNT (2) +/* +* Moving an extent to data fork can cause a sub-interval of an existing extent +* to be unmapped. This will increase extent count by 1. Mapping in the new +* extent can increase the extent count by 1 again i.e. +* | Old extent | New extent | Old extent | +* Hence number of extents increases by 2. +*/ +#define XFS_IEXT_REFLINK_END_COW_CNT (2) +/* * Fork handling. */
fs/xfs/xfs_reflink.c +5 -0
@@ -628,6 +628,11 @@ xfs_reflink_end_cow_extent(xfs_ilock (ip,XFS_ILOCK_EXCL );xfs_trans_ijoin (tp, ip,0 ); + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , +XFS_IEXT_REFLINK_END_COW_CNT ); +if (error) + goto out_cancel; +/* * In case of racing, overlapping AIO writes no COW extents might be * left by the time I/O completes for the loser of the race. In that
Signature · 1 line
2.28.0
PATCH v11 8/14 xfs: Check for extent overflow when remapping an extent
In reply to Chandan Babu RRemapping an extent involves unmapping the existing extent and mapping in the new extent. When unmapping, an extent containing the entire unmap range can be split into two extents,
i.e. | Old extent | hole | Old extent |
Hence extent count increases by 1.
Mapping in the new extent into the destination file can increase the extent count by 1.
--- fs/xfs/xfs_reflink.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
fs/xfs/xfs_reflink.c +11 -0
@@ -1006,6 +1006,7 @@ xfs_reflink_remap_extent(unsigned int resblks;bool smap_real;bool dmap_written =xfs_bmap_is_written_extent (dmap); +int iext_delta =0 ;int nimaps;int error;
@@ -1099,6 +1100,16 @@ xfs_reflink_remap_extent( goto out_cancel; } +if (smap_real) + ++iext_delta; + +if (dmap_written) + ++iext_delta; + + error =xfs_iext_count_may_overflow (ip,XFS_DATA_FORK , iext_delta); +if (error) + goto out_cancel; +if (smap_real) {/* * If the extent we're unmapping is backed by storage (written
Signature · 1 line
2.28.0
PATCH v11 9/14 xfs: Check for extent overflow when swapping extents
In reply to Chandan Babu RRemoving an initial range of source/donor file's extent and adding a new extent (from donor/source file) in its place will cause extent count to increase by 1.
--- fs/xfs/libxfs/xfs_inode_fork.h | 7 +++++++ fs/xfs/xfs_bmap_util.c | 16 ++++++++++++++++ 2 files changed, 23 insertions(+)
fs/xfs/libxfs/xfs_inode_fork.h +7 -0
@@ -87,6 +87,13 @@ struct xfs_ifork {*/ #define XFS_IEXT_REFLINK_END_COW_CNT (2) +/* +* Removing an initial range of source/donor file's extent and adding a new +* extent (from donor/source file) in its place will cause extent count to +* increase by 1. +*/ +#define XFS_IEXT_SWAP_RMAP_CNT (1) +/* * Fork handling. */
fs/xfs/xfs_bmap_util.c +16 -0
@@ -1407,6 +1407,22 @@ xfs_swap_extent_rmap( irec.br_blockcount);trace_xfs_swap_extent_rmap_remap_piece (tip, &uirec); +if (xfs_bmap_is_real_extent (&uirec)) { + error =xfs_iext_count_may_overflow (ip, +XFS_DATA_FORK , +XFS_IEXT_SWAP_RMAP_CNT ); +if (error) + goto out; + } + +if (xfs_bmap_is_real_extent (&irec)) { + error =xfs_iext_count_may_overflow (tip, +XFS_DATA_FORK , +XFS_IEXT_SWAP_RMAP_CNT ); +if (error) + goto out; + } +/* Remove the mapping from the donor file. */ xfs_bmap_unmap_extent (tp, tip, &uirec);
Signature · 1 line
2.28.0
PATCH v11 10/14 xfs: Introduce error injection to reduce maximum inode fork extent count
In reply to Chandan Babu RThis commit adds XFS_ERRTAG_REDUCE_MAX_IEXTENTS error tag which enables userspace programs to test "Inode fork extent count overflow detection" by reducing maximum possible inode fork extent count to 35.
With block size of 4k, xattr (with local value) insert operation would require in the worst case "XFS_DA_NODE_MAXDEPTH + 1" plus "XFS_DA_NODE_MAXDEPTH + (64k / 4k)" (required for guaranteeing removal of a maximum sized xattr) number of extents. This evaluates to ~28 extents. To allow for additions of two or more xattrs during extent overflow testing, the pseudo max extent count is set to 35.
--- fs/xfs/libxfs/xfs_errortag.h | 4 +++- fs/xfs/libxfs/xfs_inode_fork.c | 4 ++++ fs/xfs/xfs_error.c | 3 +++ 3 files changed, 10 insertions(+), 1 deletion(-)
fs/xfs/libxfs/xfs_errortag.h +3 -1
@@ -56,7 +56,8 @@#define XFS_ERRTAG_FORCE_SUMMARY_RECALC 33#define XFS_ERRTAG_IUNLINK_FALLBACK 34#define XFS_ERRTAG_BUF_IOERROR 35 -#define XFS_ERRTAG_MAX 36 +#define XFS_ERRTAG_REDUCE_MAX_IEXTENTS 36 +#define XFS_ERRTAG_MAX 37/* * Random factors for above tags, 1 means always, 2 means 1/2 time, etc.
@@ -97,5 +98,6 @@#define XFS_RANDOM_FORCE_SUMMARY_RECALC 1#define XFS_RANDOM_IUNLINK_FALLBACK (XFS_RANDOM_DEFAULT/10)#define XFS_RANDOM_BUF_IOERROR XFS_RANDOM_DEFAULT +#define XFS_RANDOM_REDUCE_MAX_IEXTENTS 1#endif /* __XFS_ERRORTAG_H_ */
fs/xfs/libxfs/xfs_inode_fork.c +4 -0
@@ -24,6 +24,7 @@#include "xfs_dir2_priv.h" #include "xfs_attr_leaf.h" #include "xfs_types.h" +#include "xfs_errortag.h" kmem_zone_t *xfs_ifork_zone;
@@ -745,6 +746,9 @@ xfs_iext_count_may_overflow( max_exts = (whichfork ==XFS_ATTR_FORK ) ?MAXAEXTNUM :MAXEXTNUM ; +if (XFS_TEST_ERROR (false, ip->i_mount,XFS_ERRTAG_REDUCE_MAX_IEXTENTS )) + max_exts =35 ; + nr_exts = ifp->if_nextents + nr_to_add;if (nr_exts < ifp->if_nextents || nr_exts > max_exts)return -EFBIG ;
fs/xfs/xfs_error.c +3 -0
@@ -54,6 +54,7 @@ static unsigned int xfs_errortag_random_default[] = {XFS_RANDOM_FORCE_SUMMARY_RECALC ,XFS_RANDOM_IUNLINK_FALLBACK ,XFS_RANDOM_BUF_IOERROR , + XFS_RANDOM_REDUCE_MAX_IEXTENTS, }; struct xfs_errortag_attr {
@@ -164,6 +165,7 @@ XFS_ERRORTAG_ATTR_RW(force_repair, XFS_ERRTAG_FORCE_SCRUB_REPAIR);XFS_ERRORTAG_ATTR_RW (bad_summary,XFS_ERRTAG_FORCE_SUMMARY_RECALC );XFS_ERRORTAG_ATTR_RW (iunlink_fallback,XFS_ERRTAG_IUNLINK_FALLBACK );XFS_ERRORTAG_ATTR_RW (buf_ioerror,XFS_ERRTAG_BUF_IOERROR ); +XFS_ERRORTAG_ATTR_RW (reduce_max_iextents,XFS_ERRTAG_REDUCE_MAX_IEXTENTS );static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (noerror),
@@ -202,6 +204,7 @@ static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (bad_summary),XFS_ERRORTAG_ATTR_LIST (iunlink_fallback),XFS_ERRORTAG_ATTR_LIST (buf_ioerror), +XFS_ERRORTAG_ATTR_LIST (reduce_max_iextents), NULL, };
Signature · 1 line
2.28.0
Re: [PATCH V11 10/14] xfs: Introduce error injection to reduce maximum inode fork extent count
Quoted 52 lines · Chandan Babu R
This commit adds
XFS_ERRTAG_REDUCE_MAX_IEXTENTSerror tag which enables userspace programs to test "Inode fork extent count overflow detection" by reducing maximum possible inode fork extent count to 35.With block size of 4k, xattr (with local value) insert operation would require in the worst case "
XFS_DA_NODE_MAXDEPTH+ 1" plus "XFS_DA_NODE_MAXDEPTH+ (64k / 4k)" (required for guaranteeing removal of a maximum sized xattr) number of extents. This evaluates to ~28 extents. To allow for additions of two or more xattrs during extent overflow testing, the pseudo max extent count is set to 35.Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_errortag.h | 4 +++- fs/xfs/libxfs/xfs_inode_fork.c | 4 ++++ fs/xfs/xfs_error.c | 3 +++ 3 files changed, 10 insertions(+), 1 deletion(-)fs/xfs/libxfs/xfs_errortag.h +3
-1@@ -56,7 +56,8 @@#define XFS_ERRTAG_FORCE_SUMMARY_RECALC 33#define XFS_ERRTAG_IUNLINK_FALLBACK 34#define XFS_ERRTAG_BUF_IOERROR 35 -#define XFS_ERRTAG_MAX 36 +#define XFS_ERRTAG_REDUCE_MAX_IEXTENTS 36 +#define XFS_ERRTAG_MAX 37/* * Random factors for above tags, 1 means always, 2 means 1/2 time, etc. @@ -97,5 +98,6 @@#define XFS_RANDOM_FORCE_SUMMARY_RECALC 1#define XFS_RANDOM_IUNLINK_FALLBACK (XFS_RANDOM_DEFAULT/10)#define XFS_RANDOM_BUF_IOERROR XFS_RANDOM_DEFAULT +#define XFS_RANDOM_REDUCE_MAX_IEXTENTS 1#endif /* __XFS_ERRORTAG_H_ */ fs/xfs/libxfs/xfs_inode_fork.c +1
-0@@ -24,6 +24,7 @@#include "xfs_dir2_priv.h" #include "xfs_attr_leaf.h" #include "xfs_types.h" +#include "xfs_errortag.h" kmem_zone_t *xfs_ifork_zone;
fs/xfs/libxfs/xfs_inode_fork.c xfs_iext_count_may_overflow( · line 750
Quoted 5 lines
fs/xfs/libxfs/xfs_inode_fork.c +2
-0@@ -745,6 +746,9 @@ xfs_iext_count_may_overflow( max_exts = (whichfork ==XFS_ATTR_FORK ) ?MAXAEXTNUM :MAXEXTNUM ; +if (XFS_TEST_ERROR (false, ip->i_mount,XFS_ERRTAG_REDUCE_MAX_IEXTENTS )) + max_exts =35 ;
Please add a comment here explaining why 35.
Sorry about the longish review delay, last week was a US holiday and this week I have eye problems again. :(
--D
Quoted 35 lines
fs/xfs/libxfs/xfs_inode_fork.c +1
-0+ nr_exts = ifp->if_nextents + nr_to_add;if (nr_exts < ifp->if_nextents || nr_exts > max_exts)return -EFBIG ;fs/xfs/xfs_error.c +3
-0@@ -54,6 +54,7 @@ static unsigned int xfs_errortag_random_default[] = {XFS_RANDOM_FORCE_SUMMARY_RECALC ,XFS_RANDOM_IUNLINK_FALLBACK ,XFS_RANDOM_BUF_IOERROR , + XFS_RANDOM_REDUCE_MAX_IEXTENTS, }; struct xfs_errortag_attr {@@ -164,6 +165,7 @@ XFS_ERRORTAG_ATTR_RW(force_repair, XFS_ERRTAG_FORCE_SCRUB_REPAIR);XFS_ERRORTAG_ATTR_RW (bad_summary,XFS_ERRTAG_FORCE_SUMMARY_RECALC );XFS_ERRORTAG_ATTR_RW (iunlink_fallback,XFS_ERRTAG_IUNLINK_FALLBACK );XFS_ERRORTAG_ATTR_RW (buf_ioerror,XFS_ERRTAG_BUF_IOERROR ); +XFS_ERRORTAG_ATTR_RW (reduce_max_iextents,XFS_ERRTAG_REDUCE_MAX_IEXTENTS );static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (noerror),@@ -202,6 +204,7 @@ static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (bad_summary),XFS_ERRORTAG_ATTR_LIST (iunlink_fallback),XFS_ERRORTAG_ATTR_LIST (buf_ioerror), +XFS_ERRORTAG_ATTR_LIST (reduce_max_iextents), NULL, };Signature · 1 line
2.28.0
Re: [PATCH V11 10/14] xfs: Introduce error injection to reduce maximum inode fork extent count
Quoted 52 lines · Darrick J. Wong
Quoted 51 lines · Chandan Babu R
This commit adds
XFS_ERRTAG_REDUCE_MAX_IEXTENTSerror tag which enables userspace programs to test "Inode fork extent count overflow detection" by reducing maximum possible inode fork extent count to 35.With block size of 4k, xattr (with local value) insert operation would require in the worst case "
XFS_DA_NODE_MAXDEPTH+ 1" plus "XFS_DA_NODE_MAXDEPTH+ (64k / 4k)" (required for guaranteeing removal of a maximum sized xattr) number of extents. This evaluates to ~28 extents. To allow for additions of two or more xattrs during extent overflow testing, the pseudo max extent count is set to 35.Signed-off-by Chandan Babu R <[email protected]>--- fs/xfs/libxfs/xfs_errortag.h | 4 +++- fs/xfs/libxfs/xfs_inode_fork.c | 4 ++++ fs/xfs/xfs_error.c | 3 +++ 3 files changed, 10 insertions(+), 1 deletion(-)fs/xfs/libxfs/xfs_errortag.h +3
-1@@ -56,7 +56,8 @@#define XFS_ERRTAG_FORCE_SUMMARY_RECALC 33#define XFS_ERRTAG_IUNLINK_FALLBACK 34#define XFS_ERRTAG_BUF_IOERROR 35 -#define XFS_ERRTAG_MAX 36 +#define XFS_ERRTAG_REDUCE_MAX_IEXTENTS 36 +#define XFS_ERRTAG_MAX 37/* * Random factors for above tags, 1 means always, 2 means 1/2 time, etc. @@ -97,5 +98,6 @@#define XFS_RANDOM_FORCE_SUMMARY_RECALC 1#define XFS_RANDOM_IUNLINK_FALLBACK (XFS_RANDOM_DEFAULT/10)#define XFS_RANDOM_BUF_IOERROR XFS_RANDOM_DEFAULT +#define XFS_RANDOM_REDUCE_MAX_IEXTENTS 1#endif /* __XFS_ERRORTAG_H_ */ fs/xfs/libxfs/xfs_inode_fork.c +1
-0@@ -24,6 +24,7 @@#include "xfs_dir2_priv.h" #include "xfs_attr_leaf.h" #include "xfs_types.h" +#include "xfs_errortag.h" kmem_zone_t *xfs_ifork_zone;
Quoted 8 lines
fs/xfs/libxfs/xfs_inode_fork.c xfs_iext_count_may_overflow( · line 750
Quoted 6 lines
fs/xfs/libxfs/xfs_inode_fork.c +2
-0@@ -745,6 +746,9 @@ xfs_iext_count_may_overflow( max_exts = (whichfork ==XFS_ATTR_FORK ) ?MAXAEXTNUM :MAXEXTNUM ; +if (XFS_TEST_ERROR (false, ip->i_mount,XFS_ERRTAG_REDUCE_MAX_IEXTENTS )) + max_exts =35 ;Please add a comment here explaining why 35.
Sure. I will do that.
Quoted 3 lines
Sorry about the longish review delay, last week was a US holiday and this week I have eye problems again. :(
Np. Please take care.
Quoted 36 lines
--D
Quoted 32 lines
fs/xfs/libxfs/xfs_inode_fork.c +1
-0+ nr_exts = ifp->if_nextents + nr_to_add;if (nr_exts < ifp->if_nextents || nr_exts > max_exts)return -EFBIG ;fs/xfs/xfs_error.c +3
-0@@ -54,6 +54,7 @@ static unsigned int xfs_errortag_random_default[] = {XFS_RANDOM_FORCE_SUMMARY_RECALC ,XFS_RANDOM_IUNLINK_FALLBACK ,XFS_RANDOM_BUF_IOERROR , + XFS_RANDOM_REDUCE_MAX_IEXTENTS, }; struct xfs_errortag_attr {@@ -164,6 +165,7 @@ XFS_ERRORTAG_ATTR_RW(force_repair, XFS_ERRTAG_FORCE_SCRUB_REPAIR);XFS_ERRORTAG_ATTR_RW (bad_summary,XFS_ERRTAG_FORCE_SUMMARY_RECALC );XFS_ERRORTAG_ATTR_RW (iunlink_fallback,XFS_ERRTAG_IUNLINK_FALLBACK );XFS_ERRORTAG_ATTR_RW (buf_ioerror,XFS_ERRTAG_BUF_IOERROR ); +XFS_ERRORTAG_ATTR_RW (reduce_max_iextents,XFS_ERRTAG_REDUCE_MAX_IEXTENTS );static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (noerror),@@ -202,6 +204,7 @@ static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (bad_summary),XFS_ERRORTAG_ATTR_LIST (iunlink_fallback),XFS_ERRORTAG_ATTR_LIST (buf_ioerror), +XFS_ERRORTAG_ATTR_LIST (reduce_max_iextents), NULL, };
Signature · 1 line
chandan
PATCH v11 11/14 xfs: Remove duplicate assert statement in xfs_bmap_btalloc()
In reply to Chandan Babu RThe check for verifying if the allocated extent is from an AG whose index is greater than or equal to that of tp->t_firstblock is already done a couple of statements earlier in the same function. Hence this commit removes the redundant assert statement.
--- fs/xfs/libxfs/xfs_bmap.c | 1 - 1 file changed, 1 deletion(-)
fs/xfs/libxfs/xfs_bmap.c +0 -1
@@ -3699,7 +3699,6 @@ xfs_bmap_btalloc( ap->blkno = args.fsbno;if (ap->tp->t_firstblock ==NULLFSBLOCK ) ap->tp->t_firstblock = args.fsbno; -ASSERT (nullfb || fb_agno <= args.agno); ap->length = args.len;/* * If the extent size hint is active, we tried to round the
Signature · 1 line
2.28.0
PATCH v11 12/14 xfs: Compute bmap extent alignments in a separate function
In reply to Chandan Babu RThis commit moves over the code which computes stripe alignment and extent size hint alignment into a separate function. Apart from xfs_bmap_btalloc(), the new function will be used by another function introduced in a future commit.
--- fs/xfs/libxfs/xfs_bmap.c | 89 +++++++++++++++++++++++----------------- 1 file changed, 52 insertions(+), 37 deletions(-)
fs/xfs/libxfs/xfs_bmap.c +52 -37
@@ -3463,13 +3463,59 @@ xfs_bmap_btalloc_accounting( args->len); } +static int +xfs_bmap_compute_alignments ( +struct xfs_bmalloca *ap, +struct xfs_alloc_arg *args) +{ +struct xfs_mount *mp = args->mp; +xfs_extlen_t align =0 ;/* minimum allocation alignment */ +int stripe_align =0 ; +int error; + +/* stripe alignment for allocation is determined by mount parameters */ +if (mp->m_swidth && (mp->m_flags &XFS_MOUNT_SWALLOC )) + stripe_align = mp->m_swidth; +else if (mp->m_dalign) + stripe_align = mp->m_dalign; + +if (ap->flags &XFS_BMAPI_COWFORK ) + align =xfs_get_cowextsz_hint (ap->ip); +else if (ap->datatype &XFS_ALLOC_USERDATA ) + align =xfs_get_extsz_hint (ap->ip); +if (align) { + error =xfs_bmap_extsize_align (mp, &ap->got, &ap->prev, + align,0 , ap->eof,0 , ap->conv, + &ap->offset, &ap->length); +ASSERT (!error); +ASSERT (ap->length); + } + +/* apply extent size hints if obtained earlier */ +if (align) { + args->prod = align; +div_u64_rem (ap->offset, args->prod, &args->mod); +if (args->mod) + args->mod = args->prod - args->mod; + }else if (mp->m_sb.sb_blocksize >=PAGE_SIZE ) { + args->prod =1 ; + args->mod =0 ; + }else { + args->prod =PAGE_SIZE >> mp->m_sb.sb_blocklog; +div_u64_rem (ap->offset, args->prod, &args->mod); +if (args->mod) + args->mod = args->prod - args->mod; + } + +return stripe_align; +} + STATIC intxfs_bmap_btalloc (struct xfs_bmalloca *ap)/* bmap alloc argument struct */ {xfs_mount_t *mp;/* mount point structure */ xfs_alloctype_t atype =0 ;/* type for allocation routines */ -xfs_extlen_t align =0 ;/* minimum allocation alignment */ xfs_agnumber_t fb_agno;/* ag number of ap->firstblock */ xfs_agnumber_t ag;xfs_alloc_arg_t args;
@@ -3489,25 +3535,11 @@ xfs_bmap_btalloc( mp = ap->ip->i_mount; -/* stripe alignment for allocation is determined by mount parameters */ - stripe_align =0 ; -if (mp->m_swidth && (mp->m_flags &XFS_MOUNT_SWALLOC )) - stripe_align = mp->m_swidth; -else if (mp->m_dalign) - stripe_align = mp->m_dalign; - -if (ap->flags &XFS_BMAPI_COWFORK ) - align =xfs_get_cowextsz_hint (ap->ip); -else if (ap->datatype &XFS_ALLOC_USERDATA ) - align =xfs_get_extsz_hint (ap->ip); -if (align) { - error =xfs_bmap_extsize_align (mp, &ap->got, &ap->prev, - align,0 , ap->eof,0 , ap->conv, - &ap->offset, &ap->length); -ASSERT (!error); -ASSERT (ap->length); - } +memset (&args,0 ,sizeof (args)); + args.tp = ap->tp; + args.mp = mp; + stripe_align =xfs_bmap_compute_alignments (ap, &args); nullfb = ap->tp->t_firstblock ==NULLFSBLOCK ; fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
@@ -3538,9 +3570,6 @@ xfs_bmap_btalloc(* Normal allocation, done through xfs_alloc_vextent. */ tryagain = isaligned =0 ; -memset (&args,0 ,sizeof (args)); - args.tp = ap->tp; - args.mp = mp; args.fsbno = ap->blkno; args.oinfo =XFS_RMAP_OINFO_SKIP_UPDATE ;
@@ -3571,21 +3600,7 @@ xfs_bmap_btalloc( args.total = ap->total; args.minlen = ap->minlen; } -/* apply extent size hints if obtained earlier */ -if (align) { - args.prod = align; -div_u64_rem (ap->offset, args.prod, &args.mod); -if (args.mod) - args.mod = args.prod - args.mod; - }else if (mp->m_sb.sb_blocksize >=PAGE_SIZE ) { - args.prod =1 ; - args.mod =0 ; - }else { - args.prod =PAGE_SIZE >> mp->m_sb.sb_blocklog; -div_u64_rem (ap->offset, args.prod, &args.mod); -if (args.mod) - args.mod = args.prod - args.mod; - } +/* * If we are not low on available data blocks, and the underlying * logical volume manager is a stripe, and the file offset is zero then
Signature · 1 line
2.28.0
PATCH v11 13/14 xfs: Process allocated extent in a separate function
In reply to Chandan Babu RThis commit moves over the code in xfs_bmap_btalloc() which is responsible for processing an allocated extent to a new function. Apart from xfs_bmap_btalloc(), the new function will be invoked by another function introduced in a future commit.
--- fs/xfs/libxfs/xfs_bmap.c | 74 ++++++++++++++++++++++++---------------- 1 file changed, 45 insertions(+), 29 deletions(-)
fs/xfs/libxfs/xfs_bmap.c +45 -29
@@ -3510,6 +3510,48 @@ xfs_bmap_compute_alignments(return stripe_align; } +static void +xfs_bmap_process_allocated_extent ( +struct xfs_bmalloca *ap, +struct xfs_alloc_arg *args, +xfs_fileoff_t orig_offset, +xfs_extlen_t orig_length) +{ +int nullfb; + + nullfb = ap->tp->t_firstblock ==NULLFSBLOCK ; + +/* +* check the allocation happened at the same or higher AG than +* the first block that was allocated. +*/ +ASSERT (nullfb || +XFS_FSB_TO_AGNO (args->mp, ap->tp->t_firstblock) <= +XFS_FSB_TO_AGNO (args->mp, args->fsbno)); + + ap->blkno = args->fsbno; +if (nullfb) + ap->tp->t_firstblock = args->fsbno; + ap->length = args->len; +/* +* If the extent size hint is active, we tried to round the +* caller's allocation request offset down to extsz and the +* length up to another extsz boundary. If we found a free +* extent we mapped it in starting at this new offset. If the +* newly mapped space isn't long enough to cover any of the +* range of offsets that was originally requested, move the +* mapping up so that we can fill as much of the caller's +* original request as possible. Free space is apparently +* very fragmented so we're unlikely to be able to satisfy the +* hints anyway. +*/ +if (ap->length <= orig_length) + ap->offset = orig_offset; +else if (ap->offset + ap->length < orig_offset + orig_length) + ap->offset = orig_offset + orig_length - ap->length; +xfs_bmap_btalloc_accounting (ap, args); +} + STATIC intxfs_bmap_btalloc (struct xfs_bmalloca *ap)/* bmap alloc argument struct */
@@ -3702,36 +3744,10 @@ xfs_bmap_btalloc(return error; ap->tp->t_flags |=XFS_TRANS_LOWMODE ; } +if (args.fsbno !=NULLFSBLOCK ) { -/* -* check the allocation happened at the same or higher AG than -* the first block that was allocated. -*/ -ASSERT (ap->tp->t_firstblock ==NULLFSBLOCK || -XFS_FSB_TO_AGNO (mp, ap->tp->t_firstblock) <= -XFS_FSB_TO_AGNO (mp, args.fsbno)); - - ap->blkno = args.fsbno; -if (ap->tp->t_firstblock ==NULLFSBLOCK ) - ap->tp->t_firstblock = args.fsbno; - ap->length = args.len; -/* -* If the extent size hint is active, we tried to round the -* caller's allocation request offset down to extsz and the -* length up to another extsz boundary. If we found a free -* extent we mapped it in starting at this new offset. If the -* newly mapped space isn't long enough to cover any of the -* range of offsets that was originally requested, move the -* mapping up so that we can fill as much of the caller's -* original request as possible. Free space is apparently -* very fragmented so we're unlikely to be able to satisfy the -* hints anyway. -*/ -if (ap->length <= orig_length) - ap->offset = orig_offset; -else if (ap->offset + ap->length < orig_offset + orig_length) - ap->offset = orig_offset + orig_length - ap->length; -xfs_bmap_btalloc_accounting (ap, &args); +xfs_bmap_process_allocated_extent (ap, &args, orig_offset, + orig_length); }else { ap->blkno =NULLFSBLOCK ; ap->length =0 ;
Signature · 1 line
2.28.0
PATCH v11 14/14 xfs: Introduce error injection to allocate only minlen size extents for files
In reply to Chandan Babu RThis commit adds XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT error tag which helps userspace test programs to get xfs_bmap_btalloc() to always allocate minlen sized extents.
This is required for test programs which need a guarantee that minlen extents allocated for a file do not get merged with their existing neighbours in the inode's BMBT. "Inode fork extent overflow check" for Directories, Xattrs and extension of realtime inodes need this since the file offset at which the extents are being allocated cannot be explicitly controlled from userspace.
One way to use this error tag is to,
1. Consume all of the free space by sequentially writing to a file.
2. Punch alternate blocks of the file. This causes CNTBT to contain sufficient number of one block sized extent records.
3. Inject XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT error tag. After step 3, xfs_bmap_btalloc() will issue space allocation requests for minlen sized extents only.
ENOSPC error code is returned to userspace when there aren't any "one block sized" extents left in any of the AGs.
--- fs/xfs/libxfs/xfs_alloc.c | 50 ++++++++++++++ fs/xfs/libxfs/xfs_alloc.h | 3 + fs/xfs/libxfs/xfs_bmap.c | 124 ++++++++++++++++++++++++++++------- fs/xfs/libxfs/xfs_errortag.h | 4 +- fs/xfs/xfs_error.c | 3 + 5 files changed, 159 insertions(+), 25 deletions(-)
fs/xfs/libxfs/xfs_alloc.c +50 -0
@@ -2473,6 +2473,47 @@ xfs_defer_agfl_block(xfs_defer_add (tp,XFS_DEFER_OPS_TYPE_AGFL_FREE , &new->xefi_list); } +#ifdef DEBUG +/* +* Check if an AGF has a free extent record whose length is equal to +* args->minlen. +*/ +STATIC int +xfs_exact_minlen_extent_available ( +struct xfs_alloc_arg *args, +struct xfs_buf *agbp, +int *stat) +{ +struct xfs_btree_cur *cnt_cur; +xfs_agblock_t fbno; +xfs_extlen_t flen; +int error =0 ; + + cnt_cur =xfs_allocbt_init_cursor (args->mp, args->tp, agbp, + args->agno,XFS_BTNUM_CNT ); + error =xfs_alloc_lookup_ge (cnt_cur,0 , args->minlen, stat); +if (error) + goto out; + +if (*stat ==0 ) { + error = -EFSCORRUPTED ; + goto out; + } + + error =xfs_alloc_get_rec (cnt_cur, &fbno, &flen, stat); +if (error) + goto out; + +if (*stat ==1 && flen != args->minlen) + *stat =0 ; + +out: +xfs_btree_del_cursor (cnt_cur, error); + +return error; +} +#endif +/* * Decide whether to use this allocation group for this allocation. * If so, fix up the btree freelist's size.
@@ -2544,6 +2585,15 @@ xfs_alloc_fix_freelist(if (!xfs_alloc_space_available (args, need, flags)) goto out_agbp_relse; +#ifdef DEBUG +if (args->alloc_minlen_only) { +int stat; + + error =xfs_exact_minlen_extent_available (args, agbp, &stat); +if (error || !stat) + goto out_agbp_relse; + } +#endif /* * Make the freelist shorter if it's too long. *
fs/xfs/libxfs/xfs_alloc.h +3 -0
@@ -75,6 +75,9 @@ typedef struct xfs_alloc_arg {char wasfromfl;/* set if allocation is from freelist */ struct xfs_owner_info oinfo;/* owner of blocks being allocated */ enum xfs_ag_resv_type resv;/* block reservation to use */ +#ifdef DEBUG +bool alloc_minlen_only;/* allocate exact minlen extent */ +#endif } xfs_alloc_arg_t;/*
fs/xfs/libxfs/xfs_bmap.c +100 -24
150 lines not on this page · raw message
fs/xfs/libxfs/xfs_errortag.h +3 -1
@@ -57,7 +57,8 @@#define XFS_ERRTAG_IUNLINK_FALLBACK 34#define XFS_ERRTAG_BUF_IOERROR 35#define XFS_ERRTAG_REDUCE_MAX_IEXTENTS 36 -#define XFS_ERRTAG_MAX 37 +#define XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT 37 +#define XFS_ERRTAG_MAX 38/* * Random factors for above tags, 1 means always, 2 means 1/2 time, etc.
@@ -99,5 +100,6 @@#define XFS_RANDOM_IUNLINK_FALLBACK (XFS_RANDOM_DEFAULT/10)#define XFS_RANDOM_BUF_IOERROR XFS_RANDOM_DEFAULT#define XFS_RANDOM_REDUCE_MAX_IEXTENTS 1 +#define XFS_RANDOM_BMAP_ALLOC_MINLEN_EXTENT 1#endif /* __XFS_ERRORTAG_H_ */
fs/xfs/xfs_error.c +3 -0
@@ -55,6 +55,7 @@ static unsigned int xfs_errortag_random_default[] = {XFS_RANDOM_IUNLINK_FALLBACK ,XFS_RANDOM_BUF_IOERROR ,XFS_RANDOM_REDUCE_MAX_IEXTENTS , + XFS_RANDOM_BMAP_ALLOC_MINLEN_EXTENT, }; struct xfs_errortag_attr {
@@ -166,6 +167,7 @@ XFS_ERRORTAG_ATTR_RW(bad_summary, XFS_ERRTAG_FORCE_SUMMARY_RECALC);XFS_ERRORTAG_ATTR_RW (iunlink_fallback,XFS_ERRTAG_IUNLINK_FALLBACK );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 );static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (noerror),
@@ -205,6 +207,7 @@ static struct attribute *xfs_errortag_attrs[] = {XFS_ERRORTAG_ATTR_LIST (iunlink_fallback),XFS_ERRORTAG_ATTR_LIST (buf_ioerror),XFS_ERRORTAG_ATTR_LIST (reduce_max_iextents), +XFS_ERRORTAG_ATTR_LIST (bmap_alloc_minlen_extent), NULL, };
Signature · 1 line
2.28.0