mm: close page_mkwrite races (try 2)
Subsystems (6)
PATCH 1/1 mm: close page_mkwrite races (try 2)
Let's try this again with a better changelog.
--
Change page_mkwrite to allow callers to return with the page locked, and change it's page fault callers to hold the lock until the page is marked dirty. This allows the filesystem to have full control of metadata associated with a dirty page. We'd like to call page_mkwrite with the page unlocked and return with it locked, so that filesystems can avoid LOR conditions with page lock.
A filesystem that wants to set some metadata to a page while it is dirty, will manipulate the metadata in its ->page_mkwrite. At this point, even if it does a set_page_dirty in its page_mkwrite handler, it must return with the page unlocked (according to the page_mkwrite convention).
In this window, the VM could write out the page, clearing page-dirty. The filesystem has no way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the dirty-page metadata might be freed.
It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail.
The VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail (and anyway, it would probably be hard to avoid races where the pte gets cleaned along the way).
Holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, because the page must be locked to clean it for writeout. It provides the filesystem with a strong synchronisation against VM.
- Sage needs this race closed for ceph filesystem.
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).
- I need it for fsblock.
- I suspect other filesystems may need it too (eg. btrfs).
- I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch).
- Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves.
[ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ]
--- Documentation/filesystems/Locking | 24 +++++++--- fs/buffer.c | 10 ++-- mm/memory.c | 83 ++++++++++++++++++++++++++------------ 3 files changed, 79 insertions(+), 38 deletions(-)
fs/buffer.c +6 -4
@@ -2383,7 +2383,8 @@ block_page_mkwrite(struct vm_area_structif ((page->mapping != inode->i_mapping) || (page_offset (page) > size)) {/* page got truncated out from underneath us */ - goto out_unlock; +unlock_page (page); + goto out; }/* page is wholly or partially inside EOF */
@@ -2397,14 +2398,15 @@ block_page_mkwrite(struct vm_area_struct ret =block_commit_write (page,0 , end);if (unlikely (ret)) { +unlock_page (page);if (ret == -ENOMEM ) ret =VM_FAULT_OOM ;else /* -ENOSPC, -EIO, etc */ ret =VM_FAULT_SIGBUS ; - } + }else + ret =VM_FAULT_LOCKED ; -out_unlock: -unlock_page (page); +out:return ret; }
mm/memory.c +57 -26
134 lines not on this page · raw message
Documentation/filesystems/Locking +16 -8
@@ -512,16 +512,24 @@ locking rules: BKL mmap_sem PageLocked(page) open: no yes close: no yes -fault: no yes -page_mkwrite: no yes no +fault: no yes can return with page locked +page_mkwrite: no yes can return with page locked access: no yes - ->page_mkwrite() is called when a previously read-only page is -about to become writeable. The file system is responsible for -protecting against truncate races. Once appropriate action has been -taking to lock out truncate, the page range should be verified to be -within i_size. The page mapping should also be checked that it is not -NULL. + ->fault() is called when a previously not present pte is about +to be faulted in. The filesystem must find and return the page associated +with the passed in "pgoff" in the vm_fault structure. If it is possible that +the page may be truncated and/or invalidated, then the filesystem must lock +the page, then ensure it is not already truncated (the page lock will block +subsequent truncate), and then return with VM_FAULT_LOCKED, and the page +locked. The VM will unlock the page. + + ->page_mkwrite() is called when a previously read-only pte is +about to become writeable. The filesystem again must ensure that there are +no truncate/invalidate races, and then return with the page locked. If +the page has been truncated, the filesystem should not look up a new page +like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which +will cause the VM to retry the fault. ->access() is called when get_user_pages() fails in acces_process_vm(), typically used to debug a process through
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
PATCH 1/1 mm: close page_mkwrite races (try 3)
OK, that one had some rough patches (in the changelog and the patch itself). One more try... if it's still misunderstandable, I give up :)
--
Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM.
Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock.
The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention).
In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty.
It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example.
And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail.
This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here.
- Sage needs this race closed for ceph filesystem.
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).
- I need it for fsblock.
- I suspect other filesystems may need it too (eg. btrfs).
- I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch).
- Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves.
[ This also moves page_mkwrite another step closer to fault, which should
eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ]
--- Documentation/filesystems/Locking | 24 +++++++--- fs/buffer.c | 10 ++-- mm/memory.c | 85 +++++++++++++++++++++++++------------- 3 files changed, 80 insertions(+), 39 deletions(-)
fs/buffer.c +6 -4
@@ -2383,7 +2383,8 @@ block_page_mkwrite(struct vm_area_structif ((page->mapping != inode->i_mapping) || (page_offset (page) > size)) {/* page got truncated out from underneath us */ - goto out_unlock; +unlock_page (page); + goto out; }/* page is wholly or partially inside EOF */
@@ -2397,14 +2398,15 @@ block_page_mkwrite(struct vm_area_struct ret =block_commit_write (page,0 , end);if (unlikely (ret)) { +unlock_page (page);if (ret == -ENOMEM ) ret =VM_FAULT_OOM ;else /* -ENOSPC, -EIO, etc */ ret =VM_FAULT_SIGBUS ; - } + }else + ret =VM_FAULT_LOCKED ; -out_unlock: -unlock_page (page); +out:return ret; }
mm/memory.c +58 -27
135 lines not on this page · raw message
Documentation/filesystems/Locking +16 -8
@@ -512,16 +512,24 @@ locking rules: BKL mmap_sem PageLocked(page) open: no yes close: no yes -fault: no yes -page_mkwrite: no yes no +fault: no yes can return with page locked +page_mkwrite: no yes can return with page locked access: no yes - ->page_mkwrite() is called when a previously read-only page is -about to become writeable. The file system is responsible for -protecting against truncate races. Once appropriate action has been -taking to lock out truncate, the page range should be verified to be -within i_size. The page mapping should also be checked that it is not -NULL. + ->fault() is called when a previously not present pte is about +to be faulted in. The filesystem must find and return the page associated +with the passed in "pgoff" in the vm_fault structure. If it is possible that +the page may be truncated and/or invalidated, then the filesystem must lock +the page, then ensure it is not already truncated (the page lock will block +subsequent truncate), and then return with VM_FAULT_LOCKED, and the page +locked. The VM will unlock the page. + + ->page_mkwrite() is called when a previously read-only pte is +about to become writeable. The filesystem again must ensure that there are +no truncate/invalidate races, and then return with the page locked. If +the page has been truncated, the filesystem should not look up a new page +like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which +will cause the VM to retry the fault. ->access() is called when get_user_pages() fails in acces_process_vm(), typically used to debug a process through
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Andrew Morton
maintainerRe: [patch] mm: close page_mkwrite races (try 3)
Quoted 5 lines · On Wed, 15 Apr 2009 10:25:07 +0200 Nick Piggin <[email protected]> wrote:
OK, that one had some rough patches (in the changelog and the patch itself). One more try... if it's still misunderstandable, I give up :)
--
Quoted 8 lines
Change
page_mkwriteto allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM.Rather than simply hold the page locked over the
page_mkwritecall, we callpage_mkwritewith the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock.
All right, I give up. What's LOR?
Quoted 19 lines
The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing
page_mkwriteconvention).In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty.
It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example.
And the VM cannot mark the pte dirty before
page_mkwrite, becausepage_mkwriteis allowed to fail, so we must not allow any window where the page could be written to ifpage_mkwritedoes fail.
Quoted 8 lines
This solution of holding the page locked over the 3 critical operations (
page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here.- Sage needs this race closed for ceph filesystem.
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).
I wonder which kernel version(s) we should put this in.
Going BUG isn't nice, but that report is against 2.6.27. Is the BUG super-rare, or did we avoid it via other means, or what?
Quoted 38 lines
- I need it for fsblock.
- I suspect other filesystems may need it too (eg. btrfs).
- I have convertedbuffer.cto the new locking. Even simple block allocation under dirty pages might be susceptible toi_sizechanging under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch).
- Other filesystems (eg. NFS, maybe btrfs) will need to change theirpage_mkwritefunctions themselves.[ This also moves
page_mkwriteanother step closer to fault, which should
eventually allowpage_mkwriteto be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in__do_fault. ]...
mm/memory.c +6
-4@@ -1980,9 +1989,11 @@ static int do_wp_page(struct mm_struct **/ page_table =pte_offset_map_lock (mm, pmd, address, &ptl); -page_cache_release (old_page); -if (!pte_same (*page_table, orig_pte)) +if (!pte_same (*page_table, orig_pte)) { +unlock_page (old_page); +page_cache_release (old_page); goto unlock; + } page_mkwrite =1 ; }@@ -2105,16 +2116,31 @@ unlock:* * do_no_page is protected similarly. */ -wait_on_page_locked (dirty_page); -set_page_dirty_balance (dirty_page, page_mkwrite); +if (!page_mkwrite) { +wait_on_page_locked (dirty_page);
mm/memory.c unlock: · line 2130
Quoted 10 lines
mm/memory.c +9
-0+set_page_dirty_balance (dirty_page, page_mkwrite); + }put_page (dirty_page); +if (page_mkwrite) { +struct address_space *mapping = dirty_page->mapping; + +set_page_dirty (dirty_page); +unlock_page (dirty_page); +page_cache_release (dirty_page); +balance_dirty_pages_ratelimited (mapping);
hm. I wonder what prevents (prevented) *mapping from vanishing under our feet here.
Quoted 43 lines
mm/memory.c +17
-7+ } }return ret; oom_free_new:page_cache_release (new_page); oom: -if (old_page) +if (old_page) { +if (page_mkwrite) { +unlock_page (old_page); +page_cache_release (old_page); + }page_cache_release (old_page); + }return VM_FAULT_OOM ;@@ -2736,19 +2757,29 @@ static int __do_fault(struct mm_struct *pte_unmap_unlock (page_table, ptl); out: -unlock_page (vmf.page); -out_unlocked: -if (anon) -page_cache_release (vmf.page); -else if (dirty_page) { +if (dirty_page) { +struct address_space *mapping = page->mapping; +if (vma->vm_file)file_update_time (vma->vm_file); -set_page_dirty_balance (dirty_page, page_mkwrite); +if (set_page_dirty (dirty_page)) + page_mkwrite =1 ; +unlock_page (dirty_page);put_page (dirty_page); +if (page_mkwrite) +balance_dirty_pages_ratelimited (mapping); + }else { +unlock_page (vmf.page);
mm/memory.c static int __do_fault(struct mm_struct * · line 2783
Quoted 10 lines
mm/memory.c +6
-0+if (anon) +page_cache_release (vmf.page); }return ret; + +unwritable_page: +page_cache_release (page); +return ret; }
Whoa. Running file_update_time() under lock_page() opens a whole can of worms, doesn't it? That thing can do journal commits and all sorts of stuff. And I don't think this ordering is necessary here?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 5 lines · Andrew Morton
Quoted 1 line · On Wed, 15 Apr 2009 10:25:07 +0200 Nick Piggin <[email protected]> wrote:
so filesystems can avoid LOR conditions with page lock.
All right, I give up. What's LOR?
Lock order reversal.
Quoted 7 lines
Quoted 2 lines
- Sage needs this race closed for ceph filesystem.
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).I wonder which kernel version(s) we should put this in.
Going BUG isn't nice, but that report is against 2.6.27. Is the BUG super-rare, or did we avoid it via other means, or what?
Trond?
Quoted 13 lines
Quoted 13 lines
mm/memory.c +6
-2@@ -2105,16 +2116,31 @@ unlock:* * do_no_page is protected similarly. */ -wait_on_page_locked (dirty_page); -set_page_dirty_balance (dirty_page, page_mkwrite); +if (!page_mkwrite) { +wait_on_page_locked (dirty_page); +set_page_dirty_balance (dirty_page, page_mkwrite); + }put_page (dirty_page); +if (page_mkwrite) { +struct address_space *mapping = dirty_page->mapping;
Quoted 8 lines
mm/memory.c unlock: · line 2130
Quoted 5 lines
mm/memory.c +5
-0+ +set_page_dirty (dirty_page); +unlock_page (dirty_page); +page_cache_release (dirty_page); +balance_dirty_pages_ratelimited (mapping);hm. I wonder what prevents (prevented) *mapping from vanishing under our feet here.
down_read on mmap_sem should be keeping it pinned.
Quoted 19 lines
Quoted 19 lines
mm/memory.c +12
-1+if (dirty_page) { +struct address_space *mapping = page->mapping; +if (vma->vm_file)file_update_time (vma->vm_file); -set_page_dirty_balance (dirty_page, page_mkwrite); +if (set_page_dirty (dirty_page)) + page_mkwrite =1 ; +unlock_page (dirty_page);put_page (dirty_page); +if (page_mkwrite) +balance_dirty_pages_ratelimited (mapping); + }else { +unlock_page (vmf.page); +if (anon) +page_cache_release (vmf.page); }
Quoted 10 lines
mm/memory.c static int __do_fault(struct mm_struct * · line 2783
Quoted 6 lines
mm/memory.c +4
-0return ret; + +unwritable_page: +page_cache_release (page); +return ret; }Whoa. Running
file_update_time()underlock_page()opens a whole can of worms, doesn't it? That thing can do journal commits and all sorts of stuff. And I don't think this ordering is necessary here?
Oh good catch. Yes I think we can just move that out to after the put_page no problem.
Thanks,
Nick
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
In reply to Andrew MortonHere's an incremental patch.
--
Quoted 3 lines · Andrew Morton
Whoa. Running
file_update_time()underlock_page()opens a whole can of worms, doesn't it? That thing can do journal commits and all sorts of stuff. And I don't think this ordering is necessary here?
--- mm/memory.c | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-)
mm/memory.c +8 -6
@@ -2105,9 +2105,6 @@ gotten: unlock:pte_unmap_unlock (page_table, ptl);if (dirty_page) { -if (vma->vm_file) -file_update_time (vma->vm_file); -/* * Yes, Virginia, this is actually required to prevent a race * with clear_page_dirty_for_io() from clearing the page dirty
@@ -2129,6 +2126,10 @@ unlock:page_cache_release (dirty_page);balance_dirty_pages_ratelimited (mapping); } + +/* file_update_time outside page_lock */ +if (vma->vm_file) +file_update_time (vma->vm_file); }return ret; oom_free_new:
@@ -2760,15 +2761,16 @@ out:if (dirty_page) {struct address_space *mapping = page->mapping; -if (vma->vm_file) -file_update_time (vma->vm_file); -if (set_page_dirty (dirty_page)) page_mkwrite =1 ;unlock_page (dirty_page);put_page (dirty_page);if (page_mkwrite)balance_dirty_pages_ratelimited (mapping); + +/* file_update_time outside page_lock */ +if (vma->vm_file) +file_update_time (vma->vm_file); }else {unlock_page (vmf.page);if (anon)
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
In reply to Andrew MortonQuoted 9 lines · Andrew Morton
Quoted 1 line · On Wed, 15 Apr 2009 10:25:07 +0200 Nick Piggin <[email protected]> wrote:
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).
I wonder which kernel version(s) we should put this in.
Going BUG isn't nice, but that report is against 2.6.27. Is the BUG super-rare, or did we avoid it via other means, or what?
Jumping in late in after being bit by this bug many times
over with 2.6.27. The bug (http://bugzilla.kernel.org/show_bug.cgi?id=12913)
is not rare with the right workload at all.
I can easily make it happen on smp machines, when multiple processes are writing to the same NFS mounted file system. AFAICT this needs to be back ported to 27 stable and 29 stable as well.
Nick, are there 27 based patches already available someplace? Obviously, I have verified these patches + Trond's patch --
http://lkml.org/lkml/2009/4/25/64 fixes the issue with 2.6.30-rc3
Thanks,
Kiran
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 7 lines · Ravikiran G Thirumalai
Quoted 6 lines · Andrew Morton
Quoted 1 line · On Wed, 15 Apr 2009 10:25:07 +0200 Nick Piggin <[email protected]> wrote:
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).
I wonder which kernel version(s) we should put this in.
Quoted 7 lines
Quoted 3 lines
Going BUG isn't nice, but that report is against 2.6.27. Is the BUG super-rare, or did we avoid it via other means, or what?
Jumping in late in after being bit by this bug many times
over with 2.6.27. The bug (http://bugzilla.kernel.org/show_bug.cgi?id=12913)
is not rare with the right workload at all.
Good data point, thanks.
Quoted 7 lines
I can easily make it happen on smp machines, when multiple processes are writing to the same NFS mounted file system. AFAICT this needs to be back ported to 27 stable and 29 stable as well.
Nick, are there 27 based patches already available someplace? Obviously, I have verified these patches + Trond's patch --
http://lkml.org/lkml/2009/4/25/64 fixes the issue with 2.6.30-rc3
I haven't got any prepared, but they should be a pretty trivial backport, provided we also backport c2ec175c39f62949438354f603f4aa170846aabb (which is probably a good idea anyway).
However I will probably wait for a bit, given that the patch isn't upstream yet.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Andrew Morton
maintainerRe: [patch] mm: close page_mkwrite races (try 3)
Quoted 27 lines · On Wed, 29 Apr 2009 09:12:33 +0200 Nick Piggin <[email protected]> wrote:
Quoted 7 lines · Ravikiran G Thirumalai
Quoted 6 lines · Andrew Morton
Quoted 1 line · On Wed, 15 Apr 2009 10:25:07 +0200 Nick Piggin <[email protected]> wrote:
- Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913).
I wonder which kernel version(s) we should put this in.
Quoted 7 lines
Quoted 3 lines
Going BUG isn't nice, but that report is against 2.6.27. Is the BUG super-rare, or did we avoid it via other means, or what?
Jumping in late in after being bit by this bug many times
over with 2.6.27. The bug (http://bugzilla.kernel.org/show_bug.cgi?id=12913)
is not rare with the right workload at all.Good data point, thanks.
Quoted 7 lines
I can easily make it happen on smp machines, when multiple processes are writing to the same NFS mounted file system. AFAICT this needs to be back ported to 27 stable and 29 stable as well.
Nick, are there 27 based patches already available someplace? Obviously, I have verified these patches + Trond's patch --
http://lkml.org/lkml/2009/4/25/64 fixes the issue with 2.6.30-rc3
Quoted 6 lines
I haven't got any prepared, but they should be a pretty trivial backport, provided we also backport
c2ec175c39f62949438354f603f4aa170846aabb(which is probably a good idea anyway).However I will probably wait for a bit, given that the patch isn't upstream yet.
err, I'd marked it as for-2.6.31. It looks like that was wrong?
all this:
#mm-close-page_mkwrite-races-try-3.patch: akpm issues! #mm-close-page_mkwrite-races-try-3.patch: check akpm hack mm-close-page_mkwrite-races-try-3.patch mm-close-page_mkwrite-races-try-3-update.patch mm-close-page_mkwrite-races-try-3-fix.patch mm-close-page_mkwrite-races-try-3-fix-fix.patch
is a bit of a worry. But I guess we won't know until we merge it.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 10 lines · Andrew Morton
Quoted 6 lines · On Wed, 29 Apr 2009 09:12:33 +0200 Nick Piggin <[email protected]> wrote:
I haven't got any prepared, but they should be a pretty trivial backport, provided we also backport
c2ec175c39f62949438354f603f4aa170846aabb(which is probably a good idea anyway).However I will probably wait for a bit, given that the patch isn't upstream yet.
err, I'd marked it as for-2.6.31. It looks like that was wrong?
At the time I agreed because I didn't know the severity of the NFS bugs. So it is up to you and Trond / nfs guys I guess.
Quoted 10 lines
all this:
#mm-close-page_mkwrite-races-try-3.patch: akpm issues! #mm-close-page_mkwrite-races-try-3.patch: check akpm hack mm-close-page_mkwrite-races-try-3.patch mm-close-page_mkwrite-races-try-3-update.patch mm-close-page_mkwrite-races-try-3-fix.patch mm-close-page_mkwrite-races-try-3-fix-fix.patch
is a bit of a worry. But I guess we won't know until we merge it.
I have nothing against merging it now if you think it is needed. It's only adding synchronisation, so I doubt it will cause a problem that pushes the release out.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 7 lines · Nick Piggin
Quoted 6 lines · Andrew Morton
Quoted 4 lines · On Wed, 29 Apr 2009 09:12:33 +0200 Nick Piggin <[email protected]> wrote:
I haven't got any prepared, but they should be a pretty trivial backport, provided we also backport
c2ec175c39f62949438354f603f4aa170846aabb(which is probably a good idea anyway).
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
However I will probably wait for a bit, given that the patch isn't upstream yet.
err, I'd marked it as for-2.6.31. It looks like that was wrong?
At the time I agreed because I didn't know the severity of the NFS bugs. So it is up to you and Trond / nfs guys I guess.
The bug affects any serious use of shared mmap writes, so it's pretty urgent to get it merged together with the NFS 3 liner. As I said in an earlier email, I'm able to reproduce it at will within 1 minute or so of using iozone with the right options.
I also concur with the opinion that we should backport it to the stable kernels as soon as possible.
Cheers
Trond
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Andrew Morton
maintainerRe: [patch] mm: close page_mkwrite races (try 3)
Quoted 17 lines · On Wed, 29 Apr 2009 08:39:22 -0400 Trond Myklebust…
Quoted 7 lines · Nick Piggin
Quoted 6 lines · Andrew Morton
Quoted 4 lines · On Wed, 29 Apr 2009 09:12:33 +0200 Nick Piggin <[email protected]> wrote:
I haven't got any prepared, but they should be a pretty trivial backport, provided we also backport
c2ec175c39f62949438354f603f4aa170846aabb(which is probably a good idea anyway).Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
However I will probably wait for a bit, given that the patch isn't upstream yet.
err, I'd marked it as for-2.6.31. It looks like that was wrong?
At the time I agreed because I didn't know the severity of the NFS bugs. So it is up to you and Trond / nfs guys I guess.
Quoted 8 lines
The bug affects any serious use of shared mmap writes, so it's pretty urgent to get it merged together with the NFS 3 liner. As I said in an earlier email, I'm able to reproduce it at will within 1 minute or so of using iozone with the right options.
I also concur with the opinion that we should backport it to the stable kernels as soon as possible.
ok...
I don't know what the "NFS 3 liner" is, but I trust you'll take care of it.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 2 lines · Andrew Morton
I don't know what the "NFS 3 liner" is, but I trust you'll take care of it.
It's this one. I can send it on to Linus as soon as Nick's stuff is upstream unless you'd prefer to fold it in with his patch.
Cheers
Trond
---------------------------------------------------------------------
Quoted 1 line
From
f0258852dcb43c748854d2ee550c9c270bb25f21Mon Sep 17 00:00:00 2001
From: Trond Myklebust <[email protected]>
Date: Fri, 24 Apr 2009 17:32:22 -0400
Subject: [PATCH] NFS: Close page_mkwrite() races
Follow up to Nick Piggin's patches to ensure that nfs_vm_page_mkwrite returns with the page lock held, and sets the VM_FAULT_LOCKED flag.
--- fs/nfs/file.c | 6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-)
fs/nfs/file.c +3 -3
@@ -517,10 +517,10 @@ static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) ret =nfs_updatepage (filp, page,0 , pagelen); out_unlock: +if (!ret) +return VM_FAULT_LOCKED ;unlock_page (page); -if (ret) - ret =VM_FAULT_SIGBUS ; -return ret; +return VM_FAULT_SIGBUS ; }static struct vm_operations_struct nfs_file_vm_ops = {
Signature · 9 lines
1.6.0.6 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Bad news for everyone...
My kernel, with the four patches mentioned, left a steaming present on my desk this morning.
------------[ cut here ]------------ kernel BUG at fs/nfs/write.c:252!
invalid opcode: 0000 [#1] SMP
[...]
Fascinating, no?
- Rich
Signature · 7 lines
((lambda (foo) (bar foo)) (baz)) -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 1 line · Rince
four patches mentioned
Hm, somehow I missed that the NFS 3-liner above wasn't included in the four patches.
Let's see how this goes. Apologies for the misunderstanding. :)
- Rich
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Re: [patch] mm: close page_mkwrite races (try 3)
Well...that's different.
BUG: unable to handle kernel NULL pointer dereference at 0000000000000010 IP: [<ffffffffa02c8ff2>] _nfs4_do_setlk+0xe3/0x289 [nfs] PGD 10e4f7067 PUD 109221067 PMD 0 Oops: 0000 [#1] SMP last sysfs file: /sys/devices/virtual/block/md0/md/metadata_version CPU 0 Modules linked in: autofs4 coretemp hwmon nfs lockd nfs_acl
auth_rpcgss sunrpc cachefiles fscache ipv6 cpufreq_ondemand acpi_cpufreq freq_table kvm_intel kvm snd_hda_codec_idt snd_hda_intel snd_hda_codec snd_hwdep snd_seq_dummy snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss snd_pcm cpia_usb e1000e snd_timer ppdev cpia snd ums_cypress parport_pc videodev firewire_ohci i82975x_edac usb_storage parport firewire_core i2c_i801 edac_core v4l1_compat soundcore snd_page_alloc iTCO_wdt v4l2_compat_ioctl32 pcspkr i2c_core crc_itu_t iTCO_vendor_support raid1 [last unloaded: scsi_wait_scan]
Pid: 29418, comm: 10.1.1.2-manage Not tainted 2.6.30-rc4 #1
RIP: 0010:[<ffffffffa02c8ff2>] [<ffffffffa02c8ff2>] _nfs4_do_setlk+0xe3/0x289 [nfs] RSP: 0018:ffff880102361d30 EFLAGS: 00010246 RAX: ffff8800ce865f00 RBX: ffff88010b4b74d8 RCX: 0000000000000000 RDX: 0000000000000000 RSI: 00000000000000d0 RDI: 0000000000000138 RBP: ffff880102361de0 R08: ffff880126557000 R09: ffff8800b38c3900 R10: ffffffffa02cbd1c R11: ffff880126553c00 R12: 00000000fffffff4 R13: 0000000000000000 R14: ffff88012d42f5c0 R15: ffff88012d42f5c0 FS: 0000000000000000(0000) GS:ffff880028023000(0000) knlGS:0000000000000000 CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b CR2: 0000000000000010 CR3: 0000000016eda000 CR4: 00000000000026e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Process 10.1.1.2-manage (pid: 29418, threadinfo ffff880102360000, task ffff8800015b5c00) Stack: ffff880102361d40 0000000600000001 ffff8800ce865f00 ffffffffa02cbc5c 0000000000000000 ffff880126555c00 ffff880102361d90 ffffffffa02db960 0000000000000000 ffff880128093000 0000000000000001 ffffffffa02b88aa Call Trace: [<ffffffffa02cbc5c>] ? nfs4_open_recover_helper+0x82/0x97 [nfs] [<ffffffffa02b88aa>] ? __put_nfs_open_context+0x31/0x98 [nfs] [<ffffffffa02c9646>] nfs4_lock_reclaim+0x60/0x8d [nfs] [<ffffffffa02d57a3>] nfs4_do_reclaim+0x13d/0x322 [nfs] [<ffffffffa02d5b21>] nfs4_run_state_manager+0x199/0x27f [nfs] [<ffffffffa02d5988>] ? nfs4_run_state_manager+0x0/0x27f [nfs] [<ffffffffa02d5988>] ? nfs4_run_state_manager+0x0/0x27f [nfs] [<ffffffff8105e7bf>] kthread+0x5b/0x88 [<ffffffff81011dba>] child_rip+0xa/0x20 [<ffffffff8101177d>] ? restore_args+0x0/0x30 [<ffffffff8105e764>] ? kthread+0x0/0x88 [<ffffffff81011db0>] ? child_rip+0x0/0x20 Code: 10 e1 49 8b 47 58 4d 8b af 90 00 00 00 be d0 00 00 00 bf 38 01
00 00 41 bc f4 ff ff ff 48 8b 80 a0 00 00 00 48 89 85 60 ff ff ff <49> 8b 45 10 4c 8b 70 38 49 8b 86 00 01 00 00 48 8b 80 a8 02 00 RIP [<ffffffffa02c8ff2>] _nfs4_do_setlk+0xe3/0x289 [nfs]
RSP <ffff880102361d30>
CR2: 0000000000000010
---[ end trace 205a6f9494aa30de ]---
It's unclear to me whether I should blame this on the patches applied, or that this is just something never triggered unless the aforementioned bug is fixed...
- Rich
Signature · 5 lines
Aquele que ri do destino conquistará a fortuna. -- Benjamin Disraeli -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [patch] mm: close page_mkwrite races (try 3)
Quoted 57 lines · Rince
Well...that's different.
BUG: unable to handle kernel NULL pointer dereference at 0000000000000010 IP: [<ffffffffa02c8ff2>] _nfs4_do_setlk+0xe3/0x289 [nfs] PGD 10e4f7067 PUD 109221067 PMD 0 Oops: 0000 [#1] SMP last sysfs file: /sys/devices/virtual/block/md0/md/metadata_version CPU 0 Modules linked in: autofs4 coretemp hwmon nfs lockd nfs_acl
auth_rpcgsssunrpc cachefiles fscache ipv6cpufreq_ondemandacpi_cpufreqfreq_tablekvm_intelkvmsnd_hda_codec_idtsnd_hda_intelsnd_hda_codecsnd_hwdepsnd_seq_dummysnd_seq_osssnd_seq_midi_eventsnd_seqsnd_seq_devicesnd_pcm_osssnd_mixer_osssnd_pcmcpia_usbe1000esnd_timerppdev cpia sndums_cypressparport_pcvideodevfirewire_ohcii82975x_edacusb_storageparportfirewire_corei2c_i801edac_corev4l1_compatsoundcoresnd_page_allociTCO_wdtv4l2_compat_ioctl32pcspkri2c_corecrc_itu_tiTCO_vendor_supportraid1 [last unloaded:scsi_wait_scan]
Pid: 29418, comm: 10.1.1.2-manage Not tainted 2.6.30-rc4 #1RIP: 0010:[<ffffffffa02c8ff2>] [<ffffffffa02c8ff2>] _nfs4_do_setlk+0xe3/0x289 [nfs] RSP: 0018:ffff880102361d30 EFLAGS: 00010246 RAX: ffff8800ce865f00 RBX: ffff88010b4b74d8 RCX: 0000000000000000 RDX: 0000000000000000 RSI: 00000000000000d0 RDI: 0000000000000138 RBP: ffff880102361de0 R08: ffff880126557000 R09: ffff8800b38c3900 R10: ffffffffa02cbd1c R11: ffff880126553c00 R12: 00000000fffffff4 R13: 0000000000000000 R14: ffff88012d42f5c0 R15: ffff88012d42f5c0 FS: 0000000000000000(0000) GS:ffff880028023000(0000) knlGS:0000000000000000 CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b CR2: 0000000000000010 CR3: 0000000016eda000 CR4: 00000000000026e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Process 10.1.1.2-manage (pid: 29418, threadinfo ffff880102360000, task ffff8800015b5c00) Stack: ffff880102361d40 0000000600000001 ffff8800ce865f00 ffffffffa02cbc5c 0000000000000000 ffff880126555c00 ffff880102361d90 ffffffffa02db960 0000000000000000 ffff880128093000 0000000000000001 ffffffffa02b88aa Call Trace: [<ffffffffa02cbc5c>] ? nfs4_open_recover_helper+0x82/0x97 [nfs] [<ffffffffa02b88aa>] ? __put_nfs_open_context+0x31/0x98 [nfs] [<ffffffffa02c9646>] nfs4_lock_reclaim+0x60/0x8d [nfs] [<ffffffffa02d57a3>] nfs4_do_reclaim+0x13d/0x322 [nfs] [<ffffffffa02d5b21>] nfs4_run_state_manager+0x199/0x27f [nfs] [<ffffffffa02d5988>] ? nfs4_run_state_manager+0x0/0x27f [nfs] [<ffffffffa02d5988>] ? nfs4_run_state_manager+0x0/0x27f [nfs] [<ffffffff8105e7bf>] kthread+0x5b/0x88 [<ffffffff81011dba>] child_rip+0xa/0x20 [<ffffffff8101177d>] ? restore_args+0x0/0x30 [<ffffffff8105e764>] ? kthread+0x0/0x88 [<ffffffff81011db0>] ? child_rip+0x0/0x20 Code: 10 e1 49 8b 47 58 4d 8b af 90 00 00 00 be d0 00 00 00 bf 38 0100 00 41 bc f4 ff ff ff 48 8b 80 a0 00 00 00 48 89 85 60 ff ff ff <49> 8b 45 10 4c 8b 70 38 49 8b 86 00 01 00 00 48 8b 80 a8 02 00 RIP [<
ffffffffa02c8ff2>]_nfs4_do_setlk+0xe3/0x289[nfs]
RSP <ffff880102361d30>
CR2: 0000000000000010
Quoted 7 lines
---[ end trace 205a6f9494aa30de ]---It's unclear to me whether I should blame this on the patches applied, or that this is just something never triggered unless the aforementioned bug is fixed...
- Rich
It is completely unrelated. The above appears to be due to a lock recovery error after a server reboot. It has nothing to do with the mmap writeback code.
Trond