Problems with mmap consistency]
Subsystems (2)
Hello,
Just a reminder, the below patch should be applied to mainline (it's still missing according to my log).
Thanks.
----- Forwarded message from Andrea Arcangeli <[email protected]> -----
Date: Fri, 24 Feb 2006 05:01:42 +0100
From: Andrea Arcangeli <[email protected]>
To: Trond Myklebust <[email protected]>
Cc: Olaf Kirch <[email protected]>, [email protected]
Subject: Re: [NFS] Problems with mmap consistency
Hello,
Quoted 8 lines · Trond Myklebust
Quoted 7 lines · Olaf Kirch
It seems the problem is that there is still a fairly big race window in this code - after
unmap_mapping_rangereturns, we're waiting for all dirty pages to be flushed to the server. A lot can happen while we wait - for instance, some other process may modify a page that has already been written. This modification will be lost during the subsequentinvalidate_inode_pages2call.
Quoted 5 lines
Quoted 3 lines
Is there a simple VM mechanism that could be used to prevent this, or does it need some extra logic to block a process in
nfs_readpagewhile we're revalidating?No. I believe I have discussed this with Andrea and others before.
Yep.
Quoted 2 lines
We need to add a VM helper that will invalidate the page cache and flush dirty pages atomically (such a thing does not yet exist).
Such thing will hardly exist if page faults can mark pages dirty at any time mostly lockless. Things may change for the better if we start taking the page lock in do_no_page but I doubt we'll serialize atomically against the whole disk-bound fdatawriteandwait.
I think it's enough to make invalidate_inode_pages2 non destructive. Then we can wonder if we should join unmap_mapping*+fdatasyncandwait+invalidate_inode_pages2 in a single function (so they look more atomic even if they're not internally), or something like that, but currently there are reasons why they're split (notably O_DIRECT needs the invalidate only for writes, for o_direct reads not).
If a buffered write happens at the same time of a direct write and it's making pages dirty again after our fdatasyncandwait, there's no guarantee anyway because the two syscalls are parallel. So I guess we could just change invalidate_inode_pages2 not to be destructive, then the race will go away, and an nfs invalidate won't nuke the newly written data anymore (even if the newly written data is written between the fdatasyncwait and invalidate_inode_pages2).
Basically this way dirty pages gets higher prio, and they can't be dropped anymore. They can only be flushed safely to disk. That's enough for O_DIRECT coherency protocol and it should fix nfs.
I seem to remember I asked explanations on the "was_dirty" code in my first email on the subject (before noticing you also worked on the nfs part about at the same time ;), but I didn't get any answer.
I didn't check if this fixes the testcase, patch still untested sorry.
Thanks.
Index: sp3/mm/truncate.c
mm/truncate.c renamed from mm/truncate.c.~1~ +11 -10
@@ -246,9 +246,14 @@ EXPORT_SYMBOL(invalidate_inode_pages);* @start: the page offset 'from' which to invalidate * @end: the page offset 'to' which to invalidate (inclusive) * Any pages which are found to be mapped into pagetables are unmapped prior to -* invalidation. +* invalidation. invalidate_inode_pages2_range is non destructive and it can't +* lose dirty data (if dirty data exists -EIO will be returned). It's up to +* the caller to call unmap_mapping_range and filemap_write_and_wait before +* invalidate_inode_pages2 if needed. * -* Returns -EIO if any pages could not be invalidated. +* Returns -EIO if any pages could not be invalidated. Before returning -EIO +* it tries invalidating all pages in the range, it doesn't stop at the first +* page invalidation failure. */ int invalidate_inode_pages2_range (struct address_space *mapping,pgoff_t start,pgoff_t end)
@@ -262,13 +267,12 @@ int invalidate_inode_pages2_range(structpagevec_init (&pvec,0 ); next = start; -while (next <= end && !ret && !wrapped && +while (next <= end && !wrapped &&pagevec_lookup (&pvec, mapping, next,min (end - next, (pgoff_t )PAGEVEC_SIZE -1 ) +1 )) { -for (i =0 ; !ret && i <pagevec_count (&pvec); i++) { +for (i =0 ; i <pagevec_count (&pvec); i++) {struct page *page = pvec.pages[i];pgoff_t page_index; -int was_dirty;lock_page (page);if (page->mapping != mapping) {
@@ -304,12 +308,9 @@ int invalidate_inode_pages2_range(struct PAGE_CACHE_SIZE, 0); } } - was_dirty =test_clear_page_dirty (page); -if (!invalidate_complete_page (mapping, page)) { -if (was_dirty) -set_page_dirty (page); + +if (!invalidate_complete_page (mapping, page)) ret = -EIO ; - }unlock_page (page); }pagevec_release (&pvec);
----- End forwarded message -----
Quoted 6 lines · Andrea Arcangeli
Hello,
Just a reminder, the below patch should be applied to mainline (it's still missing according to my log).
There's basically no description of what the patch does here. Patches which turn up in the middle of an email discussion tend to not get applied. Normally someone will send a new patch with a reworked description once the discussion has come to a conclusion.
Quoted 4 lines
I seem to remember I asked explanations on the "
was_dirty" code in my first email on the subject (before noticing you also worked on the nfs part about at the same time ;), but I didn't get any answer.
It's all there in bk somewhere ;)
ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.10-rc1/2.6.10-rc1-mm1/broken-out/invalidate_inode_pages-mmap-coherency-fix.patch
Quoted 2 lines
I didn't check if this fixes the testcase, patch still untested sorry.
Did it?
What was the testcase?
Quoted 43 lines
Index:
sp3/mm/truncate.cmm/truncate.c renamed from mm/truncate.c.~1~ +9
-6@@ -246,9 +246,14 @@ EXPORT_SYMBOL(invalidate_inode_pages);* @start: the page offset 'from' which to invalidate * @end: the page offset 'to' which to invalidate (inclusive) * Any pages which are found to be mapped into pagetables are unmapped prior to -* invalidation. +* invalidation. invalidate_inode_pages2_range is non destructive and it can't +* lose dirty data (if dirty data exists -EIO will be returned). It's up to +* the caller to call unmap_mapping_range and filemap_write_and_wait before +* invalidate_inode_pages2 if needed. * -* Returns -EIO if any pages could not be invalidated. +* Returns -EIO if any pages could not be invalidated. Before returning -EIO +* it tries invalidating all pages in the range, it doesn't stop at the first +* page invalidation failure. */ int invalidate_inode_pages2_range (struct address_space *mapping,pgoff_t start,pgoff_t end)@@ -262,13 +267,12 @@ int invalidate_inode_pages2_range(structpagevec_init (&pvec,0 ); next = start; -while (next <= end && !ret && !wrapped && +while (next <= end && !wrapped &&pagevec_lookup (&pvec, mapping, next,min (end - next, (pgoff_t )PAGEVEC_SIZE -1 ) +1 )) { -for (i =0 ; !ret && i <pagevec_count (&pvec); i++) { +for (i =0 ; i <pagevec_count (&pvec); i++) {struct page *page = pvec.pages[i];pgoff_t page_index; -int was_dirty;lock_page (page);if (page->mapping != mapping) {@@ -304,12 +308,9 @@ int invalidate_inode_pages2_range(struct PAGE_CACHE_SIZE, 0); } } - was_dirty =test_clear_page_dirty (page);
mm/truncate.c int invalidate_inode_pages2_range(struct · line 316
Quoted 10 lines
mm/truncate.c renamed from mm/truncate.c.~1~ +2
-4-if (!invalidate_complete_page (mapping, page)) { -if (was_dirty) -set_page_dirty (page); + +if (!invalidate_complete_page (mapping, page)) ret = -EIO ; - }unlock_page (page); }pagevec_release (&pvec);
Two separate things are happening here.
a) We go all the way to the end of the range even if something went wrong partway through. Why?
b) Remove the was_dirty logic.
That's there to address the unmap-dirtyied-the-page-and-buffers problem described in the above changelog. If that can happen then this patch will end up giving us a page which is marked clean but which in fact has dirty buffers.
Quoted 4 lines · On Saturday April 1, [email protected] wrote:
Quoted 3 lines · Andrea Arcangeli
Hello,
Quoted 8 lines
Quoted 3 lines
Just a reminder, the below patch should be applied to mainline (it's still missing according to my log).
There's basically no description of what the patch does here. Patches which turn up in the middle of an email discussion tend to not get applied. Normally someone will send a new patch with a reworked description once the discussion has come to a conclusion.
I'll give it a go. See below.
Quoted 6 lines
Quoted 5 lines
I seem to remember I asked explanations on the "
was_dirty" code in my first email on the subject (before noticing you also worked on the nfs part about at the same time ;), but I didn't get any answer.
Quoted 8 lines
It's all there in bk somewhere ;)
ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.10-rc1/2.6.10-rc1-mm1/broken-out/invalidate_inode_pages-mmap-coherency-fix.patch
Quoted 2 lines
I didn't check if this fixes the testcase, patch still untested sorry.
Did it?
Yep.
Quoted 2 lines
What was the testcase?
Olaf Kirch had something that we can dig up.
Quoted 5 lines
Two separate things are happening here.
a) We go all the way to the end of the range even if something went wrong partway through. Why?
"wrong" is not well defined in this context. More precisely, the goal of invalidate_inode_pages2 is to ensure that future reads are not satisfied by the page cache. The failure of invalidate_complete_page does not necessarily imply that gaol has failed. See changelog for more discussion.
Quoted 2 lines
b) Remove the
was_dirtylogic.
This was a big problem. Removing dirty data from the page_cache before it has been written out causes data corruption. The fact that nfs how has a ->releasepage which will fail appropriately means that this isn't such a problem for nfs. However clearing and then possibly setting the flag back doesn't seem like a race-free thing to do...
Quoted 5 lines
That's there to address the unmap-dirtyied-the-page-and-buffers problem described in the above changelog. If that can happen then this patch will end up giving us a page which is marked clean but which in fact has dirty buffers.
So don't mark the page clean until the buffers are clean?
You will note the my changelog entry isn't completely definitive: it leaves some questions unanswered. So maybe the discussion is still ongoing, and input is welcome.
NeilBrown
-------------------------------------
Fix problems with invalidate_inode_pages2{,_range}
invalidate_inode_page2 (or the _range version) is called to ensure that future read requests will not be satisfied from the page cache, but rather will cause data to be retrieved from whatever storage the filesystem (host?) uses.
There are two general cases where it is used.
1/ Network filesystem which determine that the cache is, or may be,
out of date. 9p calls it after a write. nfs calls it in
nfs_revalidate_mapping which is called when any other piece of code decides that the data in the page cache might be out of date.
2/ When directIO has performed a write bypassing the cache. In this case
the cache is clearly out of date and must be purged.
There are two ways the future reads can be disabled. One is by removing pages from the cache, and the other is by flagging them as not having valid data (clear PG_uptodate). The former is preferred.
There is a potential difficulty if a page is dirty. This can happen even if the mapping is flushed before the invalidate begins, as there is generally no exclusion between the invalidate and write calls. If a page is found to be dirty is cannot be removed as this would result in data loss. We could clear the PG_uptodate bit however instead of doing this in the generic code, it is left to the filesystem to do. i.e. every page is passed to ->releasepage (if the Private flag is set).
The filesystem should do anything that is necessary to ensure that a future read will not be served from the cache.
The current code is wrong in two respects.
- it clears the 'dirty' bit without actually saving the data. This can result in data loss (though if ->releasepage fails, which it probably would, the flag gets restored).
- it gives up after the first ->releasepage call that fails. This is inappropriate as a failure of ->releasepage doesn't mean that a subsequent read will return data from the cache, and so it doesn't mean that the purpose of invalidate_inode_page2 has failed.
Possibly invalidate_inode_pages2 should clear the PG_uptodate bit if invalidate_complete_page fails, however it might be safer to leave that to the filesystem. In this case, try_to_free_buffers should possibly clear PG_uptodate.
### Diffstat output
./mm/truncate.c | 21 +++++++++++---------- 1 file changed, 11 insertions(+), 10 deletions(-)
mm/truncate.c renamed from mm/truncate.c~current~ +11 -10
@@ -274,9 +274,14 @@ EXPORT_SYMBOL(invalidate_inode_pages);* @end: the page offset 'to' which to invalidate (inclusive) * * Any pages which are found to be mapped into pagetables are unmapped prior to -* invalidation. +* invalidation. invalidate_inode_pages2_range is non destructive and it can't +* lose dirty data (if dirty data exists -EIO will be returned). It's up to +* the caller to call unmap_mapping_range and filemap_write_and_wait before +* invalidate_inode_pages2 if needed. * -* Returns -EIO if any pages could not be invalidated. +* Returns -EIO if any pages could not be invalidated. Before returning -EIO +* it tries invalidating all pages in the range, it doesn't stop at the first +* page invalidation failure. */ int invalidate_inode_pages2_range (struct address_space *mapping,pgoff_t start,pgoff_t end)
@@ -290,13 +295,12 @@ int invalidate_inode_pages2_range(structpagevec_init (&pvec,0 ); next = start; -while (next <= end && !ret && !wrapped && +while (next <= end && !wrapped &&pagevec_lookup (&pvec, mapping, next,min (end - next, (pgoff_t )PAGEVEC_SIZE -1 ) +1 )) { -for (i =0 ; !ret && i <pagevec_count (&pvec); i++) { +for (i =0 ; i <pagevec_count (&pvec); i++) {struct page *page = pvec.pages[i];pgoff_t page_index; -int was_dirty;lock_page (page);if (page->mapping != mapping) {
@@ -332,12 +336,9 @@ int invalidate_inode_pages2_range(struct PAGE_CACHE_SIZE, 0); } } - was_dirty =test_clear_page_dirty (page); -if (!invalidate_complete_page (mapping, page)) { -if (was_dirty) -set_page_dirty (page); + +if (!invalidate_complete_page (mapping, page)) ret = -EIO ; - }unlock_page (page); }pagevec_release (&pvec);