fs: new inode i_state corruption fix
Subsystems (3)
PATCH 1/1 fs: new inode i_state corruption fix
There was a report of a data corruption http://lkml.org/lkml/2008/11/14/121.
There is a script included to reproduce the problem.
During testing, I encountered a number of strange things with ext3, so I tried ext2 to attempt to reduce complexity of the problem. I found that fsstress would quickly hang in wait_on_inode, waiting for I_LOCK to be cleared, even though instrumentation showed that unlock_new_inode had already been called for that inode. This points to memory scribble, or synchronisation problme.
i_state of I_NEW inodes is not protected by inode_lock because other processes are not supposed to touch them until I_LOCK (and I_NEW) is cleared. Adding WARN_ON(inode->i_state & I_NEW) to sites where we modify i_state revealed that generic_sync_sb_inodes is picking up new inodes from the inode lists and passing them to __writeback_single_inode without waiting for I_NEW. Subsequently modifying i_state causes corruption. In my case it would look like this:
CPU0 CPU1unlock_new_inode() __sync_single_inode()
reg <- inode->i_state
reg -> reg & ~(I_LOCK|I_NEW) reg <- inode->i_state
reg -> inode->i_state reg -> reg | I_SYNC
reg -> inode->i_state
Non-atomic RMW on CPU1 overwrites CPU0 store and sets I_LOCK|I_NEW again.
Fix for this is rather than wait for I_NEW inodes, just skip over them:
inodes concurrently being created are not subject to data integrity operations, and should not significantly contribute to dirty memory either.
After this change, I'm unable to reproduce any of the added warnings or hangs after ~1hour of running. Previously, the new warnings would start immediately and hang would happen in under 5 minutes.
I'm also testing on ext3 now, and so far no problems there either. I don't know whether this fixes the problem reported above, but it fixes a real problem for me.
fs/inode.c +7 -0
@@ -359,6 +359,7 @@ static int invalidate_list(struct list_hinvalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; count++;continue ;
@@ -460,6 +461,7 @@ static void prune_icache(int nr_to_scan)continue ; }list_move (&inode->i_list, &freeable); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; nr_pruned++; }
@@ -656,6 +658,7 @@ void unlock_new_inode(struct inode *inod* just created it (so there can be no old holders * that haven't tested I_LOCK). */ +WARN_ON ((inode->i_state & (I_LOCK |I_NEW )) != (I_LOCK |I_NEW )); inode->i_state &= ~(I_LOCK |I_NEW );wake_up_inode (inode); }
@@ -1145,6 +1148,7 @@ void generic_delete_inode(struct inode *list_del_init (&inode->i_list);list_del_init (&inode->i_sb_list); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; inodes_stat.nr_inodes--;spin_unlock (&inode_lock);
@@ -1186,16 +1190,19 @@ static void generic_forget_inode(structspin_unlock (&inode_lock);return ; } +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_WILL_FREE ;spin_unlock (&inode_lock);write_inode_now (inode,1 );spin_lock (&inode_lock); +WARN_ON (inode->i_state &I_NEW ); inode->i_state &= ~I_WILL_FREE ; inodes_stat.nr_unused--;hlist_del_init (&inode->i_hash); }list_del_init (&inode->i_list);list_del_init (&inode->i_sb_list); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; inodes_stat.nr_inodes--;spin_unlock (&inode_lock);
fs/fs-writeback.c +8 -1
@@ -274,6 +274,7 @@ __sync_single_inode(struct inode *inode,int ret;BUG_ON (inode->i_state &I_SYNC ); +WARN_ON (inode->i_state &I_NEW );/* Set I_SYNC, reset I_DIRTY */ dirty = inode->i_state &I_DIRTY ;
@@ -298,6 +299,7 @@ __sync_single_inode(struct inode *inode, }spin_lock (&inode_lock); +WARN_ON (inode->i_state &I_NEW ); inode->i_state &= ~I_SYNC ;if (!(inode->i_state &I_FREEING )) { if (!(inode->i_state & I_DIRTY) &&
@@ -470,6 +472,11 @@ void generic_sync_sb_inodes(struct superbreak ; } +if (inode->i_state &I_NEW ) { +requeue_io (inode); +continue ; + } +if (wbc->nonblocking &&bdi_write_congested (bdi)) { wbc->encountered_congestion =1 ;if (!sb_is_blkdev_sb (sb))
@@ -531,7 +538,7 @@ void generic_sync_sb_inodes(struct superlist_for_each_entry (inode, &sb->s_inodes, i_sb_list) {struct address_space *mapping; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ; mapping = inode->i_mapping;if (mapping->nrpages ==0 )
Jan Kara
maintainerQuoted 11 lines · Nick Piggin
There was a report of a data corruption http://lkml.org/lkml/2008/11/14/121.
There is a script included to reproduce the problem.During testing, I encountered a number of strange things with ext3, so I tried ext2 to attempt to reduce complexity of the problem. I found that fsstress would quickly hang in
wait_on_inode, waiting forI_LOCKto be cleared, even though instrumentation showed thatunlock_new_inodehad already been called for that inode. This points to memory scribble, or synchronisation problme.
Quoted 7 lines
i_stateofI_NEWinodes is not protected byinode_lockbecause other processes are not supposed to touch them untilI_LOCK(andI_NEW) is cleared. AddingWARN_ON(inode->i_state & I_NEW)to sites where we modifyi_staterevealed thatgeneric_sync_sb_inodesis picking up new inodes from the inode lists and passing them to__writeback_single_inodewithout waiting forI_NEW. Subsequently modifyingi_statecauses corruption. In my case it would look like this:
Good catch.
Quoted 9 lines
CPU0 CPU1
unlock_new_inode()__sync_single_inode()
reg <-inode->i_state
reg -> reg & ~(I_LOCK|I_NEW) reg <-inode->i_state
reg ->inode->i_statereg -> reg |I_SYNCreg -> inode->i_stateNon-atomic RMW on CPU1 overwrites CPU0 store and sets I_LOCK|I_NEW again.
Quoted 7 lines
Fix for this is rather than wait for
I_NEWinodes, just skip over them:
inodes concurrently being created are not subject to data integrity operations, and should not significantly contribute to dirty memory either.After this change, I'm unable to reproduce any of the added warnings or hangs after ~1hour of running. Previously, the new warnings would start immediately and hang would happen in under 5 minutes.
A quick grep seems to indicate that you've still missed a few cases,
haven't you? I still see the same problem in drop_caches.c:drop_pagecache_sb() scanning, inode.c:invalidate_inodes() scanning, and dquot.c:add_dquot_ref() scanning.
Otherwise the patch looks fine.
Quoted 9 lines
I'm also testing on ext3 now, and so far no problems there either. I don't know whether this fixes the problem reported above, but it fixes a real problem for me.
Cc "Jorge Boncompte [DTI2]" <[email protected]>Cc Adrian Hunter <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Nick Piggin <[email protected]>
Honza
Quoted 98 lines
Index:
linux-2.6/fs/inode.c
===================================================================fs/inode.c +7
-0@@ -359,6 +359,7 @@ static int invalidate_list(struct list_hinvalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; count++;continue ;@@ -460,6 +461,7 @@ static void prune_icache(int nr_to_scan)continue ; }list_move (&inode->i_list, &freeable); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; nr_pruned++; }@@ -656,6 +658,7 @@ void unlock_new_inode(struct inode *inod* just created it (so there can be no old holders * that haven't tested I_LOCK). */ +WARN_ON ((inode->i_state & (I_LOCK |I_NEW )) != (I_LOCK |I_NEW )); inode->i_state &= ~(I_LOCK |I_NEW );wake_up_inode (inode); }@@ -1145,6 +1148,7 @@ void generic_delete_inode(struct inode *list_del_init (&inode->i_list);list_del_init (&inode->i_sb_list); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; inodes_stat.nr_inodes--;spin_unlock (&inode_lock);@@ -1186,16 +1190,19 @@ static void generic_forget_inode(structspin_unlock (&inode_lock);return ; } +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_WILL_FREE ;spin_unlock (&inode_lock);write_inode_now (inode,1 );spin_lock (&inode_lock); +WARN_ON (inode->i_state &I_NEW ); inode->i_state &= ~I_WILL_FREE ; inodes_stat.nr_unused--;hlist_del_init (&inode->i_hash); }list_del_init (&inode->i_list);list_del_init (&inode->i_sb_list); +WARN_ON (inode->i_state &I_NEW ); inode->i_state |=I_FREEING ; inodes_stat.nr_inodes--;spin_unlock (&inode_lock);Index:
linux-2.6/fs/fs-writeback.c
===================================================================fs/fs-writeback.c +8
-1@@ -274,6 +274,7 @@ __sync_single_inode(struct inode *inode,int ret;BUG_ON (inode->i_state &I_SYNC ); +WARN_ON (inode->i_state &I_NEW );/* Set I_SYNC, reset I_DIRTY */ dirty = inode->i_state &I_DIRTY ;@@ -298,6 +299,7 @@ __sync_single_inode(struct inode *inode, }spin_lock (&inode_lock); +WARN_ON (inode->i_state &I_NEW ); inode->i_state &= ~I_SYNC ;if (!(inode->i_state &I_FREEING )) { if (!(inode->i_state & I_DIRTY) &&@@ -470,6 +472,11 @@ void generic_sync_sb_inodes(struct superbreak ; } +if (inode->i_state &I_NEW ) { +requeue_io (inode); +continue ; + } +if (wbc->nonblocking &&bdi_write_congested (bdi)) { wbc->encountered_congestion =1 ;if (!sb_is_blkdev_sb (sb))@@ -531,7 +538,7 @@ void generic_sync_sb_inodes(struct superlist_for_each_entry (inode, &sb->s_inodes, i_sb_list) {struct address_space *mapping; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ; mapping = inode->i_mapping;if (mapping->nrpages ==0 )
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 8 lines · Jan Kara
Quoted 2 lines · Nick Piggin
after ~1hour of running. Previously, the new warnings would start immediately and hang would happen in under 5 minutes.
A quick grep seems to indicate that you've still missed a few cases,
haven't you? I still see the same problem in drop_caches.c:drop_pagecache_sb() scanning, inode.c:invalidate_inodes() scanning, and dquot.c:add_dquot_ref() scanning.
Otherwise the patch looks fine.
I thought they should be OK; drop_pagecache_sb doesn't play with flags, invalidate_inodes won't if refcount is elevated, and I think add_dquot_ref won't if writecount is not elevated...
But maybe that's abit fragile and it would be better policy to always skip I_NEW in these traverals?
Jan Kara
maintainerQuoted 10 lines · Nick Piggin
Quoted 8 lines · Jan Kara
Quoted 2 lines · Nick Piggin
after ~1hour of running. Previously, the new warnings would start immediately and hang would happen in under 5 minutes.
A quick grep seems to indicate that you've still missed a few cases,
haven't you? I still see the same problem in drop_caches.c:drop_pagecache_sb() scanning, inode.c:invalidate_inodes() scanning, and dquot.c:add_dquot_ref() scanning.
Otherwise the patch looks fine.
Quoted 3 lines
I thought they should be OK;
drop_pagecache_sbdoesn't play with flags,invalidate_inodeswon't if refcount is elevated, and I thinkadd_dquot_refwon't if writecount is not elevated...
Ah, ok, you are probably right.
Quoted 2 lines
But maybe that's abit fragile and it would be better policy to always skip
I_NEWin these traverals?
Yes, it seems too fragile to me. I'm not saying we have to forbid
everything for I_NEW inodes but I think we should set clear simple rules what is protected by I_NEW and then verify that all sites which can come across such inodes obey them.
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
PATCH 1/1 fs: avoid I_NEW inodes
Quoted 14 lines · Jan Kara
Quoted 10 lines · Nick Piggin
Quoted 8 lines · Jan Kara
Quoted 2 lines · Nick Piggin
after ~1hour of running. Previously, the new warnings would start immediately and hang would happen in under 5 minutes.
A quick grep seems to indicate that you've still missed a few cases,
haven't you? I still see the same problem in drop_caches.c:drop_pagecache_sb() scanning, inode.c:invalidate_inodes() scanning, and dquot.c:add_dquot_ref() scanning.
Otherwise the patch looks fine.Quoted 3 lines
I thought they should be OK;
drop_pagecache_sbdoesn't play with flags,invalidate_inodeswon't if refcount is elevated, and I thinkadd_dquot_refwon't if writecount is not elevated...
Quoted 8 lines
Ah, ok, you are probably right.
Quoted 2 lines
But maybe that's abit fragile and it would be better policy to always skip
I_NEWin these traverals?Yes, it seems too fragile to me. I'm not saying we have to forbid
everything forI_NEWinodes but I think we should set clear simple rules what is protected byI_NEWand then verify that all sites which can come across such inodes obey them.
OK, sorry for the delay, what do you think of the following patch on top of the last?
---
To be on the safe side, it should be less fragile to exclude I_NEW inodes from inode list scans by default (unless there is an important reason to have them).
Normally they will get excluded (eg. by zero refcount or writecount etc), however it is a bit fragile for list walkers to know exactly what parts of the inode state is set up and valid to test when in I_NEW. So along these lines, move I_NEW checks upward as well (sometimes taking I_FREEING etc checks with them too -- this shouldn't be a problem should it?)
--- fs/dquot.c | 6 ++++-- fs/drop_caches.c | 2 +- fs/inode.c | 2 ++ fs/notify/inotify/inotify.c | 16 ++++++++-------- 4 files changed, 15 insertions(+), 11 deletions(-)
fs/dquot.c +4 -2
@@ -789,12 +789,12 @@ static void add_dquot_ref(struct super_bspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW )) +continue ;if (!atomic_read (&inode->i_writecount))continue ;if (!dqinit_needed (inode, type))continue ; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) -continue ;__iget (inode);spin_unlock (&inode_lock);
@@ -870,6 +870,8 @@ static void remove_dquot_ref(struct supespin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state &I_NEW ) +continue ;if (!IS_NOQUOTA (inode))remove_inode_dquot_ref (inode, type, tofree_head); }
fs/drop_caches.c +1 -1
@@ -18,7 +18,7 @@ static void drop_pagecache_sb(struct supspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ;if (inode->i_mapping->nrpages ==0 )continue ;
fs/inode.c +2 -0
@@ -356,6 +356,8 @@ static int invalidate_list(struct list_hif (tmp == head)break ; inode = list_entry(tmp, struct inode, i_sb_list); +if (inode->i_state &I_NEW ) +continue ;invalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose);
fs/notify/inotify/inotify.c +8 -8
@@ -380,6 +380,14 @@ void inotify_unmount_inodes(struct list_struct list_head *watches;/* +* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or +* I_WILL_FREE which is fine because by that point the inode +* cannot have any associated watches. +*/ +if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE |I_NEW )) +continue ; + +/* * If i_count is zero, the inode cannot have any watches and * doing an __iget/iput with MS_ACTIVE clear would actually * evict all inodes with zero i_count from icache which is
@@ -388,14 +396,6 @@ void inotify_unmount_inodes(struct list_if (!atomic_read (&inode->i_count))continue ; -/* -* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or -* I_WILL_FREE which is fine because by that point the inode -* cannot have any associated watches. -*/ -if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE )) -continue ; - need_iput_tmp = need_iput; need_iput =NULL ;/* In case inotify_remove_watch_locked() drops a reference. */
Jan Kara
maintainerRe: [patch] fs: avoid I_NEW inodes
Hi,
Quoted 19 lines · Nick Piggin
Quoted 18 lines · Jan Kara
Quoted 10 lines · Nick Piggin
Quoted 8 lines · Jan Kara
Quoted 2 lines · Nick Piggin
after ~1hour of running. Previously, the new warnings would start immediately and hang would happen in under 5 minutes.
A quick grep seems to indicate that you've still missed a few cases,
haven't you? I still see the same problem in drop_caches.c:drop_pagecache_sb() scanning, inode.c:invalidate_inodes() scanning, and dquot.c:add_dquot_ref() scanning.
Otherwise the patch looks fine.Quoted 3 lines
I thought they should be OK;
drop_pagecache_sbdoesn't play with flags,invalidate_inodeswon't if refcount is elevated, and I thinkadd_dquot_refwon't if writecount is not elevated...Ah, ok, you are probably right.
Quoted 2 lines
But maybe that's abit fragile and it would be better policy to always skip
I_NEWin these traverals?
Quoted 7 lines
Quoted 4 lines
Yes, it seems too fragile to me. I'm not saying we have to forbid
everything forI_NEWinodes but I think we should set clear simple rules what is protected byI_NEWand then verify that all sites which can come across such inodes obey them.OK, sorry for the delay, what do you think of the following patch on top of the last?
Thanks for the patch. I have a few comments. See below.
Quoted 41 lines
---
To be on the safe side, it should be less fragile to exclude
I_NEWinodes from inode list scans by default (unless there is an important reason to have them).Normally they will get excluded (eg. by zero refcount or writecount etc), however it is a bit fragile for list walkers to know exactly what parts of the inode state is set up and valid to test when in
I_NEW. So along these lines, moveI_NEWchecks upward as well (sometimes takingI_FREEINGetc checks with them too -- this shouldn't be a problem should it?)Signed-off-by Nick Piggin <[email protected]>--- fs/dquot.c | 6 ++++-- fs/drop_caches.c | 2 +- fs/inode.c | 2 ++ fs/notify/inotify/inotify.c | 16 ++++++++-------- 4 files changed, 15 insertions(+), 11 deletions(-)Index:
linux-2.6/fs/dquot.c
===================================================================fs/dquot.c +2
-2@@ -789,12 +789,12 @@ static void add_dquot_ref(struct super_bspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW )) +continue ;if (!atomic_read (&inode->i_writecount))continue ;if (!dqinit_needed (inode, type))continue ; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) -continue ;__iget (inode);spin_unlock (&inode_lock);
fs/dquot.c static void remove_dquot_ref(struct supe · line 877
Quoted 8 lines
fs/dquot.c +2
-0@@ -870,6 +870,8 @@ static void remove_dquot_ref(struct supespin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state &I_NEW ) +continue ;if (!IS_NOQUOTA (inode))remove_inode_dquot_ref (inode, type, tofree_head); }
Hmm, in this scan, we have to scan also I_NEW inodes because they can
already have quota pointers initialized and so we could leave some dangling quota references if we skipped I_NEW inodes. Nasty. So just add a comment here like this one here:
/*
* We have to scan also I_NEW inodes because they can already have quota * pointer initialized. Luckily, we need to touch only quota pointers and * these have separate locking (dqptr_sem).
*/
Quoted 18 lines
Index:
linux-2.6/fs/drop_caches.c
===================================================================fs/drop_caches.c +1
-1@@ -18,7 +18,7 @@ static void drop_pagecache_sb(struct supspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ;if (inode->i_mapping->nrpages ==0 )continue ;Index:
linux-2.6/fs/inode.c
===================================================================
fs/inode.c static int invalidate_list(struct list_h · line 360
Quoted 5 lines
fs/inode.c +2
-0@@ -356,6 +356,8 @@ static int invalidate_list(struct list_hif (tmp == head)break ; inode = list_entry(tmp, struct inode, i_sb_list); +if (inode->i_state &I_NEW ) +continue ;
If somebody is setting up inodes at this point, we are in serious
trouble I think. So WARN_ON would be more appropriate I think.
Quoted 8 lines
fs/inode.c +0
-0invalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose);Index:
linux-2.6/fs/notify/inotify/inotify.c
===================================================================
fs/notify/inotify/inotify.c void inotify_unmount_inodes(struct list_ · line 386
Quoted 7 lines
fs/notify/inotify/inotify.c +4
-0@@ -380,6 +380,14 @@ void inotify_unmount_inodes(struct list_struct list_head *watches;/* +* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or +* I_WILL_FREE which is fine because by that point the inode +* cannot have any associated watches. +*/
Update the comment?
Quoted 12 lines
fs/notify/inotify/inotify.c +4
-1+if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE |I_NEW )) +continue ; + +/* * If i_count is zero, the inode cannot have any watches and * doing an __iget/iput with MS_ACTIVE clear would actually * evict all inodes with zero i_count from icache which is @@ -388,14 +396,6 @@ void inotify_unmount_inodes(struct list_if (!atomic_read (&inode->i_count))continue ; -/*
fs/notify/inotify/inotify.c void inotify_unmount_inodes(struct list_ · line 401
Quoted 10 lines
fs/notify/inotify/inotify.c +0
-7-* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or -* I_WILL_FREE which is fine because by that point the inode -* cannot have any associated watches. -*/ -if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE )) -continue ; - need_iput_tmp = need_iput; need_iput =NULL ;/* In case inotify_remove_watch_locked() drops a reference. */
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [patch] fs: avoid I_NEW inodes
Quoted 58 lines · Jan Kara
Quoted 2 lines
OK, sorry for the delay, what do you think of the following patch on top of the last?
Thanks for the patch. I have a few comments. See below.
Quoted 41 lines
---
To be on the safe side, it should be less fragile to exclude
I_NEWinodes from inode list scans by default (unless there is an important reason to have them).Normally they will get excluded (eg. by zero refcount or writecount etc), however it is a bit fragile for list walkers to know exactly what parts of the inode state is set up and valid to test when in
I_NEW. So along these lines, moveI_NEWchecks upward as well (sometimes takingI_FREEINGetc checks with them too -- this shouldn't be a problem should it?)Signed-off-by Nick Piggin <[email protected]>--- fs/dquot.c | 6 ++++-- fs/drop_caches.c | 2 +- fs/inode.c | 2 ++ fs/notify/inotify/inotify.c | 16 ++++++++-------- 4 files changed, 15 insertions(+), 11 deletions(-)Index:
linux-2.6/fs/dquot.c
===================================================================fs/dquot.c +2
-2@@ -789,12 +789,12 @@ static void add_dquot_ref(struct super_bspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW )) +continue ;if (!atomic_read (&inode->i_writecount))continue ;if (!dqinit_needed (inode, type))continue ; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) -continue ;__iget (inode);spin_unlock (&inode_lock);fs/dquot.c static void remove_dquot_ref(struct supe · line 877
Quoted 8 lines
fs/dquot.c +2
-0@@ -870,6 +870,8 @@ static void remove_dquot_ref(struct supespin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state &I_NEW ) +continue ;if (!IS_NOQUOTA (inode))remove_inode_dquot_ref (inode, type, tofree_head); }Hmm, in this scan, we have to scan also
I_NEWinodes because they can
already have quota pointers initialized and so we could leave some dangling quota references if we skippedI_NEWinodes. Nasty. So just add a comment here like this one here:
/*
Quoted 4 lines
* We have to scan also I_NEW inodes because they can already have quota * pointer initialized. Luckily, we need to touch only quota pointers and * these have separate locking (dqptr_sem).*/
OK, thanks. This is what I was unsure of.
Quoted 17 lines
Quoted 17 lines
Index:
linux-2.6/fs/drop_caches.c
===================================================================fs/drop_caches.c +1
-1@@ -18,7 +18,7 @@ static void drop_pagecache_sb(struct supspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ;if (inode->i_mapping->nrpages ==0 )continue ;Index:
linux-2.6/fs/inode.c
===================================================================fs/inode.c +0
-0
Quoted 8 lines
fs/inode.c static int invalidate_list(struct list_h · line 360
Quoted 6 lines
fs/inode.c +2
-0@@ -356,6 +356,8 @@ static int invalidate_list(struct list_hif (tmp == head)break ; inode = list_entry(tmp, struct inode, i_sb_list); +if (inode->i_state &I_NEW ) +continue ;If somebody is setting up inodes at this point, we are in serious
trouble I think. SoWARN_ONwould be more appropriate I think.
Really? Hmm, this is also called via flush_disk which seems like it can operate under a mounted filesystem?
Quoted 10 lines
Quoted 10 lines
fs/inode.c +0
-0invalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose);Index:
linux-2.6/fs/notify/inotify/inotify.c
===================================================================fs/notify/inotify/inotify.c +0
-0@@ -380,6 +380,14 @@ void inotify_unmount_inodes(struct list_struct list_head *watches;
Quoted 6 lines
fs/notify/inotify/inotify.c void inotify_unmount_inodes(struct list_ · line 386
Quoted 5 lines
fs/notify/inotify/inotify.c +4
-0/* +* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or +* I_WILL_FREE which is fine because by that point the inode +* cannot have any associated watches. +*/ Update the comment?
Will do.
Thanks,
Nick
Jan Kara
maintainerRe: [patch] fs: avoid I_NEW inodes
Quoted 27 lines · Nick Piggin
Quoted 17 lines · Jan Kara
Quoted 17 lines
Index:
linux-2.6/fs/drop_caches.c
===================================================================fs/drop_caches.c +1
-1@@ -18,7 +18,7 @@ static void drop_pagecache_sb(struct supspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ;if (inode->i_mapping->nrpages ==0 )continue ;Index:
linux-2.6/fs/inode.c
===================================================================fs/inode.c +0
-0Quoted 8 lines
fs/inode.c static int invalidate_list(struct list_h · line 360
Quoted 6 lines
fs/inode.c +2
-0@@ -356,6 +356,8 @@ static int invalidate_list(struct list_hif (tmp == head)break ; inode = list_entry(tmp, struct inode, i_sb_list); +if (inode->i_state &I_NEW ) +continue ;If somebody is setting up inodes at this point, we are in serious
trouble I think. SoWARN_ONwould be more appropriate I think.
Quoted 2 lines
Really? Hmm, this is also called via
flush_diskwhich seems like it can operate under a mounted filesystem?
Ah, I was not following calls far enough. You're right.
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [patch] fs: avoid I_NEW inodes
In reply to Jan KaraHow about this?
--
To be on the safe side, it should be less fragile to exclude I_NEW inodes from inode list scans by default (unless there is an important reason to have them).
Normally they will get excluded (eg. by zero refcount or writecount etc), however it is a bit fragile for list walkers to know exactly what parts of the inode state is set up and valid to test when in I_NEW. So along these lines, move I_NEW checks upward as well (sometimes taking I_FREEING etc checks with them too -- this shouldn't be a problem should it?)
--- fs/dquot.c | 10 ++++++++-- fs/drop_caches.c | 2 +- fs/inode.c | 2 ++ fs/notify/inotify/inotify.c | 16 ++++++++-------- 4 files changed, 19 insertions(+), 11 deletions(-)
fs/dquot.c +8 -2
@@ -789,12 +789,12 @@ static void add_dquot_ref(struct super_bspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW )) +continue ;if (!atomic_read (&inode->i_writecount))continue ;if (!dqinit_needed (inode, type))continue ; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) -continue ;__iget (inode);spin_unlock (&inode_lock);
@@ -870,6 +870,12 @@ static void remove_dquot_ref(struct supespin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +/* +* We have to scan also I_NEW inodes because they can already +* have quota pointer initialized. Luckily, we need to touch +* only quota pointers and these have separate locking +* (dqptr_sem). +*/ if (!IS_NOQUOTA (inode))remove_inode_dquot_ref (inode, type, tofree_head); }
fs/drop_caches.c +1 -1
@@ -18,7 +18,7 @@ static void drop_pagecache_sb(struct supspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ;if (inode->i_mapping->nrpages ==0 )continue ;
fs/inode.c +2 -0
@@ -356,6 +356,8 @@ static int invalidate_list(struct list_hif (tmp == head)break ; inode = list_entry(tmp, struct inode, i_sb_list); +if (inode->i_state &I_NEW ) +continue ;invalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose);
fs/notify/inotify/inotify.c +8 -8
@@ -380,6 +380,14 @@ void inotify_unmount_inodes(struct list_struct list_head *watches;/* +* We cannot __iget() an inode in state I_CLEAR, I_FREEING, +* I_WILL_FREE, or I_NEW which is fine because by that point +* the inode cannot have any associated watches. +*/ +if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE |I_NEW )) +continue ; + +/* * If i_count is zero, the inode cannot have any watches and * doing an __iget/iput with MS_ACTIVE clear would actually * evict all inodes with zero i_count from icache which is
@@ -388,14 +396,6 @@ void inotify_unmount_inodes(struct list_if (!atomic_read (&inode->i_count))continue ; -/* -* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or -* I_WILL_FREE which is fine because by that point the inode -* cannot have any associated watches. -*/ -if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE )) -continue ; - need_iput_tmp = need_iput; need_iput =NULL ;/* In case inotify_remove_watch_locked() drops a reference. */
Re: [patch] fs: avoid I_NEW inodes
Quoted 1 line · Nick Piggin
How about this?
Looks fine to me.
Quoted 5 lines
--
To be on the safe side, it should be less fragile to excludeI_NEWinodes from inode list scans by default (unless there is an important reason to have them).
Quoted 7 lines
Normally they will get excluded (eg. by zero refcount or writecount etc), however it is a bit fragile for list walkers to know exactly what parts of the inode state is set up and valid to test when in
I_NEW. So along these lines, moveI_NEWchecks upward as well (sometimes takingI_FREEINGetc checks with them too -- this shouldn't be a problem should it?)Signed-off-by Nick Piggin <[email protected]>
Honza
Quoted 99 lines
--- fs/dquot.c | 10 ++++++++-- fs/drop_caches.c | 2 +- fs/inode.c | 2 ++ fs/notify/inotify/inotify.c | 16 ++++++++-------- 4 files changed, 19 insertions(+), 11 deletions(-)Index:
linux-2.6/fs/dquot.c
===================================================================fs/dquot.c +8
-2@@ -789,12 +789,12 @@ static void add_dquot_ref(struct super_bspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW )) +continue ;if (!atomic_read (&inode->i_writecount))continue ;if (!dqinit_needed (inode, type))continue ; -if (inode->i_state & (I_FREEING |I_WILL_FREE )) -continue ;__iget (inode);spin_unlock (&inode_lock);@@ -870,6 +870,12 @@ static void remove_dquot_ref(struct supespin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { +/* +* We have to scan also I_NEW inodes because they can already +* have quota pointer initialized. Luckily, we need to touch +* only quota pointers and these have separate locking +* (dqptr_sem). +*/ if (!IS_NOQUOTA (inode))remove_inode_dquot_ref (inode, type, tofree_head); }Index:
linux-2.6/fs/drop_caches.c
===================================================================fs/drop_caches.c +1
-1@@ -18,7 +18,7 @@ static void drop_pagecache_sb(struct supspin_lock (&inode_lock);list_for_each_entry (inode, &sb->s_inodes, i_sb_list) { -if (inode->i_state & (I_FREEING |I_WILL_FREE )) +if (inode->i_state & (I_FREEING |I_WILL_FREE |I_NEW ))continue ;if (inode->i_mapping->nrpages ==0 )continue ;Index:
linux-2.6/fs/inode.c
===================================================================fs/inode.c +2
-0@@ -356,6 +356,8 @@ static int invalidate_list(struct list_hif (tmp == head)break ; inode = list_entry(tmp, struct inode, i_sb_list); +if (inode->i_state &I_NEW ) +continue ;invalidate_inode_buffers (inode);if (!atomic_read (&inode->i_count)) {list_move (&inode->i_list, dispose);Index:
linux-2.6/fs/notify/inotify/inotify.c
===================================================================fs/notify/inotify/inotify.c +8
-8@@ -380,6 +380,14 @@ void inotify_unmount_inodes(struct list_struct list_head *watches;/* +* We cannot __iget() an inode in state I_CLEAR, I_FREEING, +* I_WILL_FREE, or I_NEW which is fine because by that point +* the inode cannot have any associated watches. +*/ +if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE |I_NEW )) +continue ; + +/* * If i_count is zero, the inode cannot have any watches and * doing an __iget/iput with MS_ACTIVE clear would actually * evict all inodes with zero i_count from icache which is @@ -388,14 +396,6 @@ void inotify_unmount_inodes(struct list_if (!atomic_read (&inode->i_count))continue ; -/* -* We cannot __iget() an inode in state I_CLEAR, I_FREEING, or -* I_WILL_FREE which is fine because by that point the inode -* cannot have any associated watches. -*/ -if (inode->i_state & (I_CLEAR |I_FREEING |I_WILL_FREE )) -continue ; - need_iput_tmp = need_iput; need_iput =NULL ;/* In case inotify_remove_watch_locked() drops a reference. */
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [patch] fs: avoid I_NEW inodes
Quoted 3 lines · Jan Kara
Quoted 1 line · Nick Piggin
How about this?
Looks fine to me.
Thanks for the good review. Andrew, do you think you can apply this on top of the previous patch? I'm undecided as to whether they should go together or not. Probably the first one is a minimal fix that doesn't alter behaviour as much, but things seem more robust after this 2nd patch. I think both would probably be suitable for 2.6.29, being a nasty bug, but it isn't a recent regression AFAIKS.
Quoted 14 lines
Quoted 6 lines
--
To be on the safe side, it should be less fragile to excludeI_NEWinodes from inode list scans by default (unless there is an important reason to have them).Quoted 7 lines
Normally they will get excluded (eg. by zero refcount or writecount etc), however it is a bit fragile for list walkers to know exactly what parts of the inode state is set up and valid to test when in
I_NEW. So along these lines, moveI_NEWchecks upward as well (sometimes takingI_FREEINGetc checks with them too -- this shouldn't be a problem should it?)Signed-off-by Nick Piggin <[email protected]>Acked-by Jan Kara <[email protected]>
Re: [patch] fs: avoid I_NEW inodes
Quoted 12 lines · On Wed, 11 Mar 2009 13:57:48 +0100 Nick Piggin <[email protected]> wrote:
Quoted 3 lines · Jan Kara
Quoted 1 line · Nick Piggin
How about this?
Looks fine to me.
Thanks for the good review. Andrew, do you think you can apply this on top of the previous patch? I'm undecided as to whether they should go together or not. Probably the first one is a minimal fix that doesn't alter behaviour as much, but things seem more robust after this 2nd patch. I think both would probably be suitable for 2.6.29, being a nasty bug, but it isn't a recent regression AFAIKS.
How's about we do fs-new-inode-i_state-corruption-fix.patch in 2.6.29 and fs-avoid-i_new-inodes.patch in 2.6.30? We could backport fs-avoid-i_new-inodes.patch into 2.6.29.x if needed.
Re: [patch] fs: avoid I_NEW inodes
Quoted 16 lines · Andrew Morton
Quoted 12 lines · On Wed, 11 Mar 2009 13:57:48 +0100 Nick Piggin <[email protected]> wrote:
Quoted 3 lines · Jan Kara
Quoted 1 line · Nick Piggin
How about this?
Looks fine to me.
Thanks for the good review. Andrew, do you think you can apply this on top of the previous patch? I'm undecided as to whether they should go together or not. Probably the first one is a minimal fix that doesn't alter behaviour as much, but things seem more robust after this 2nd patch. I think both would probably be suitable for 2.6.29, being a nasty bug, but it isn't a recent regression AFAIKS.
Quoted 3 lines
How's about we do fs-new-inode-i_state-corruption-fix.patch in 2.6.29 and fs-avoid-i_new-inodes.patch in 2.6.30? We could backport fs-avoid-i_new-inodes.patch into 2.6.29.x if needed.
Yes that's probably best.