| Message ID | 20210226140011.2883498-1-willy@infradead.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mm/filemap: Use filemap_read_page in filemap_fault | expand |
On Fri, 26 Feb 2021 14:00:11 +0000 "Matthew Wilcox (Oracle)" <willy@infradead.org> wrote: > After splitting generic_file_buffered_read() into smaller parts, it > turns out we can reuse one of the parts in filemap_fault(). This fixes > an oversight -- waiting for the I/O to complete is now interruptible > by a fatal signal. And it saves us a few bytes of text in an unlikely > path. We also handle AOP_TRUNCATED_PAGE which the present code fails to do. Should this be in the changelog? Did we handle AOP_TRUNCATED_PAGE in the pre-splitup code, or is this new?
On Tue, Mar 02, 2021 at 05:30:39PM -0800, Andrew Morton wrote: > On Fri, 26 Feb 2021 14:00:11 +0000 "Matthew Wilcox (Oracle)" <willy@infradead.org> wrote: > > > After splitting generic_file_buffered_read() into smaller parts, it > > turns out we can reuse one of the parts in filemap_fault(). This fixes > > an oversight -- waiting for the I/O to complete is now interruptible > > by a fatal signal. And it saves us a few bytes of text in an unlikely > > path. > > We also handle AOP_TRUNCATED_PAGE which the present code fails to do. > Should this be in the changelog? No, the present code does handle AOP_TRUNCATED_PAGE. It's perhaps not the clearest in the diff, but it's there. Here's git show -U5: - ClearPageError(page); fpin = maybe_unlock_mmap_for_io(vmf, fpin); - error = mapping->a_ops->readpage(file, page); - if (!error) { - wait_on_page_locked(page); - if (!PageUptodate(page)) - error = -EIO; - } + error = filemap_read_page(file, mapping, page); if (fpin) goto out_retry; put_page(page); if (!error || error == AOP_TRUNCATED_PAGE) goto retry_find; - shrink_readahead_size_eio(ra); return VM_FAULT_SIGBUS;
On Wed, 3 Mar 2021 01:33:13 +0000 Matthew Wilcox <willy@infradead.org> wrote: > On Tue, Mar 02, 2021 at 05:30:39PM -0800, Andrew Morton wrote: > > On Fri, 26 Feb 2021 14:00:11 +0000 "Matthew Wilcox (Oracle)" <willy@infradead.org> wrote: > > > > > After splitting generic_file_buffered_read() into smaller parts, it > > > turns out we can reuse one of the parts in filemap_fault(). This fixes > > > an oversight -- waiting for the I/O to complete is now interruptible > > > by a fatal signal. And it saves us a few bytes of text in an unlikely > > > path. > > > > We also handle AOP_TRUNCATED_PAGE which the present code fails to do. > > Should this be in the changelog? > > No, the present code does handle AOP_TRUNCATED_PAGE. It's perhaps not > the clearest in the diff, but it's there. Here's git show -U5: > > - ClearPageError(page); > fpin = maybe_unlock_mmap_for_io(vmf, fpin); > - error = mapping->a_ops->readpage(file, page); > - if (!error) { > - wait_on_page_locked(page); > - if (!PageUptodate(page)) > - error = -EIO; > - } > + error = filemap_read_page(file, mapping, page); > if (fpin) > goto out_retry; > put_page(page); > > if (!error || error == AOP_TRUNCATED_PAGE) > goto retry_find; > > - shrink_readahead_size_eio(ra); > return VM_FAULT_SIGBUS; But ->readpage() doesn't check for !mapping (does it?). So the ->readpage() cannot return AOP_TRUNCATED_PAGE. However filemap_read_page() does check for !mapping. So current -linus doesn't check for !mapping, and post-willy does do this?
On Tue, Mar 02, 2021 at 10:07:35PM -0800, Andrew Morton wrote: > > > We also handle AOP_TRUNCATED_PAGE which the present code fails to do. > > > Should this be in the changelog? > > > > No, the present code does handle AOP_TRUNCATED_PAGE. It's perhaps not > > the clearest in the diff, but it's there. Here's git show -U5: > > > > - ClearPageError(page); > > fpin = maybe_unlock_mmap_for_io(vmf, fpin); > > - error = mapping->a_ops->readpage(file, page); > > - if (!error) { > > - wait_on_page_locked(page); > > - if (!PageUptodate(page)) > > - error = -EIO; > > - } > > + error = filemap_read_page(file, mapping, page); > > if (fpin) > > goto out_retry; > > put_page(page); > > > > if (!error || error == AOP_TRUNCATED_PAGE) > > goto retry_find; > > > > - shrink_readahead_size_eio(ra); > > return VM_FAULT_SIGBUS; > > But ->readpage() doesn't check for !mapping (does it?). So the > ->readpage() cannot return AOP_TRUNCATED_PAGE. Filesystems can return AOP_TRUNCATED_PAGE from ->readpage. I think ocfs2 is the only one to do so currently. > However filemap_read_page() does check for !mapping. So current -linus > doesn't check for !mapping, and post-willy does do this? Oh! I see what you mean now. I can't come up with a sequence of events where this check is going to do anything useful. It may be left over from earlier times. Here's an earlier version of generic_file_buffered_read() before Kent refactored it: /* Start the actual read. The read will unlock the page. */ error = mapping->a_ops->readpage(filp, page); [...] if (!PageUptodate(page)) { error = lock_page_killable(page); if (unlikely(error)) goto readpage_error; if (!PageUptodate(page)) { if (page->mapping == NULL) { /* * invalidate_mapping_pages got it */ unlock_page(page); put_page(page); But here's the thing ... invalidate_mapping_pages() doesn't ClearPageUptodate. The only places where we ClearPageUptodate is on an I/O error. So ... as far as I can tell, the only way to hit this is: - Get an I/O error during the wait - Have another thread cause the page to be removed from the page cache (eg do direct I/O to the file) before this thread is run. and the consequence to this change is that we have another attempt to read the page instead of returning an error immediately. I'm OK with that unintentional change, although I think the previous behaviour was also perfectly acceptable (after all, there was an I/O error while trying to read this page). Delving into the linux-fullhistory tree, this code was introduced by ... commit 56f0d5fe6851037214a041a5cb4fc66199256544 Author: Andrew Morton <akpm@osdl.org> Date: Fri Jan 7 22:03:01 2005 -0800 [PATCH] readpage-vs-invalidate fix A while ago we merged a patch which tried to solve a problem wherein a concurrent read() and invalidate_inode_pages() would cause the read() to return -EIO because invalidate cleared PageUptodate() at the wrong time. We no longer clear PageUptodate, so I think this is stale code? Perhaps you could check with the original author ...
On Wed, 3 Mar 2021 13:26:40 +0000 Matthew Wilcox <willy@infradead.org> wrote: > But here's the thing ... invalidate_mapping_pages() doesn't > ClearPageUptodate. The only places where we ClearPageUptodate is on an > I/O error. yup. > So ... as far as I can tell, the only way to hit this is: > > - Get an I/O error during the wait > - Have another thread cause the page to be removed from the page cache > (eg do direct I/O to the file) before this thread is run. > > and the consequence to this change is that we have another attempt to > read the page instead of returning an error immediately. I'm OK with > that unintentional change, although I think the previous behaviour was > also perfectly acceptable (after all, there was an I/O error while trying > to read this page). > > Delving into the linux-fullhistory tree, this code was introduced by ... > > commit 56f0d5fe6851037214a041a5cb4fc66199256544 > Author: Andrew Morton <akpm@osdl.org> > Date: Fri Jan 7 22:03:01 2005 -0800 > > [PATCH] readpage-vs-invalidate fix > > A while ago we merged a patch which tried to solve a problem wherein a > concurrent read() and invalidate_inode_pages() would cause the read() to > return -EIO because invalidate cleared PageUptodate() at the wrong time. > > We no longer clear PageUptodate, so I think this is stale code? Perhaps > you could check with the original author ... Which code do you think might be stale? We need the !PageUptodate check to catch IO errors and we need the !page->mapping check to catch invalidates. Am a bit confused.
On Wed, Mar 03, 2021 at 12:12:53PM -0800, Andrew Morton wrote: > On Wed, 3 Mar 2021 13:26:40 +0000 Matthew Wilcox <willy@infradead.org> wrote: > > > But here's the thing ... invalidate_mapping_pages() doesn't > > ClearPageUptodate. The only places where we ClearPageUptodate is on an > > I/O error. > > yup. > > > So ... as far as I can tell, the only way to hit this is: > > > > - Get an I/O error during the wait > > - Have another thread cause the page to be removed from the page cache > > (eg do direct I/O to the file) before this thread is run. > > > > and the consequence to this change is that we have another attempt to > > read the page instead of returning an error immediately. I'm OK with > > that unintentional change, although I think the previous behaviour was > > also perfectly acceptable (after all, there was an I/O error while trying > > to read this page). > > > > Delving into the linux-fullhistory tree, this code was introduced by ... > > > > commit 56f0d5fe6851037214a041a5cb4fc66199256544 > > Author: Andrew Morton <akpm@osdl.org> > > Date: Fri Jan 7 22:03:01 2005 -0800 > > > > [PATCH] readpage-vs-invalidate fix > > > > A while ago we merged a patch which tried to solve a problem wherein a > > concurrent read() and invalidate_inode_pages() would cause the read() to > > return -EIO because invalidate cleared PageUptodate() at the wrong time. > > > > We no longer clear PageUptodate, so I think this is stale code? Perhaps > > you could check with the original author ... > > Which code do you think might be stale? We need the !PageUptodate > check to catch IO errors and we need the !page->mapping check to catch > invalidates. Am a bit confused. I think the check of !page->mapping here: if (PageUptodate(page)) return 0; if (!page->mapping) /* page truncated */ return AOP_TRUNCATED_PAGE; is no longer needed. If we didn't see an error, the page will be Uptodate, regardless of whether it's been removed from the page cache. If we did see an error, it's OK to return -EIO, even if the page has been removed from the page cache in the interim.
On Wed, 3 Mar 2021 20:57:36 +0000 Matthew Wilcox <willy@infradead.org> wrote: > On Wed, Mar 03, 2021 at 12:12:53PM -0800, Andrew Morton wrote: > > On Wed, 3 Mar 2021 13:26:40 +0000 Matthew Wilcox <willy@infradead.org> wrote: > > > > > But here's the thing ... invalidate_mapping_pages() doesn't > > > ClearPageUptodate. The only places where we ClearPageUptodate is on an > > > I/O error. > > > > yup. > > > > > So ... as far as I can tell, the only way to hit this is: > > > > > > - Get an I/O error during the wait > > > - Have another thread cause the page to be removed from the page cache > > > (eg do direct I/O to the file) before this thread is run. > > > > > > and the consequence to this change is that we have another attempt to > > > read the page instead of returning an error immediately. I'm OK with > > > that unintentional change, although I think the previous behaviour was > > > also perfectly acceptable (after all, there was an I/O error while trying > > > to read this page). > > > > > > Delving into the linux-fullhistory tree, this code was introduced by ... > > > > > > commit 56f0d5fe6851037214a041a5cb4fc66199256544 > > > Author: Andrew Morton <akpm@osdl.org> > > > Date: Fri Jan 7 22:03:01 2005 -0800 > > > > > > [PATCH] readpage-vs-invalidate fix > > > > > > A while ago we merged a patch which tried to solve a problem wherein a > > > concurrent read() and invalidate_inode_pages() would cause the read() to > > > return -EIO because invalidate cleared PageUptodate() at the wrong time. > > > > > > We no longer clear PageUptodate, so I think this is stale code? Perhaps > > > you could check with the original author ... > > > > Which code do you think might be stale? We need the !PageUptodate > > check to catch IO errors and we need the !page->mapping check to catch > > invalidates. Am a bit confused. > > I think the check of !page->mapping here: > > if (PageUptodate(page)) > return 0; > if (!page->mapping) /* page truncated */ > return AOP_TRUNCATED_PAGE; > > is no longer needed. If we didn't see an error, the page will be Uptodate, > regardless of whether it's been removed from the page cache. If we did > see an error, it's OK to return -EIO, even if the page has been removed > from the page cache in the interim. OK. Checking page->mapping of an unlocked page seems meaningless anyway - what's to prevent it from being truncated just after we checked?
On Wed, Mar 03, 2021 at 01:51:42PM -0800, Andrew Morton wrote: > On Wed, 3 Mar 2021 20:57:36 +0000 Matthew Wilcox <willy@infradead.org> wrote: > > > On Wed, Mar 03, 2021 at 12:12:53PM -0800, Andrew Morton wrote: > > > On Wed, 3 Mar 2021 13:26:40 +0000 Matthew Wilcox <willy@infradead.org> wrote: > > > > > > > But here's the thing ... invalidate_mapping_pages() doesn't > > > > ClearPageUptodate. The only places where we ClearPageUptodate is on an > > > > I/O error. > > > > > > yup. > > > > > > > So ... as far as I can tell, the only way to hit this is: > > > > > > > > - Get an I/O error during the wait > > > > - Have another thread cause the page to be removed from the page cache > > > > (eg do direct I/O to the file) before this thread is run. > > > > > > > > and the consequence to this change is that we have another attempt to > > > > read the page instead of returning an error immediately. I'm OK with > > > > that unintentional change, although I think the previous behaviour was > > > > also perfectly acceptable (after all, there was an I/O error while trying > > > > to read this page). > > > > > > > > Delving into the linux-fullhistory tree, this code was introduced by ... > > > > > > > > commit 56f0d5fe6851037214a041a5cb4fc66199256544 > > > > Author: Andrew Morton <akpm@osdl.org> > > > > Date: Fri Jan 7 22:03:01 2005 -0800 > > > > > > > > [PATCH] readpage-vs-invalidate fix > > > > > > > > A while ago we merged a patch which tried to solve a problem wherein a > > > > concurrent read() and invalidate_inode_pages() would cause the read() to > > > > return -EIO because invalidate cleared PageUptodate() at the wrong time. > > > > > > > > We no longer clear PageUptodate, so I think this is stale code? Perhaps > > > > you could check with the original author ... > > > > > > Which code do you think might be stale? We need the !PageUptodate > > > check to catch IO errors and we need the !page->mapping check to catch > > > invalidates. Am a bit confused. > > > > I think the check of !page->mapping here: > > > > if (PageUptodate(page)) > > return 0; > > if (!page->mapping) /* page truncated */ > > return AOP_TRUNCATED_PAGE; > > > > is no longer needed. If we didn't see an error, the page will be Uptodate, > > regardless of whether it's been removed from the page cache. If we did > > see an error, it's OK to return -EIO, even if the page has been removed > > from the page cache in the interim. > > OK. > > Checking page->mapping of an unlocked page seems meaningless anyway - > what's to prevent it from being truncated just after we checked? Originally it prevented the opposite race from being misinterpreted; we didn't care if it was truncated after we saw it was Uptodate; we cared if it was !Uptodate because it had been truncated. Now, it's just pointless.
diff --git a/mm/filemap.c b/mm/filemap.c index 46a8b9e82434..f7ab86d13692 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -2748,7 +2748,6 @@ vm_fault_t filemap_fault(struct vm_fault *vmf) struct file *file = vmf->vma->vm_file; struct file *fpin = NULL; struct address_space *mapping = file->f_mapping; - struct file_ra_state *ra = &file->f_ra; struct inode *inode = mapping->host; pgoff_t offset = vmf->pgoff; pgoff_t max_off; @@ -2835,14 +2834,8 @@ vm_fault_t filemap_fault(struct vm_fault *vmf) * because there really aren't any performance issues here * and we need to check for errors. */ - ClearPageError(page); fpin = maybe_unlock_mmap_for_io(vmf, fpin); - error = mapping->a_ops->readpage(file, page); - if (!error) { - wait_on_page_locked(page); - if (!PageUptodate(page)) - error = -EIO; - } + error = filemap_read_page(file, mapping, page); if (fpin) goto out_retry; put_page(page); @@ -2850,7 +2843,6 @@ vm_fault_t filemap_fault(struct vm_fault *vmf) if (!error || error == AOP_TRUNCATED_PAGE) goto retry_find; - shrink_readahead_size_eio(ra); return VM_FAULT_SIGBUS; out_retry:
After splitting generic_file_buffered_read() into smaller parts, it turns out we can reuse one of the parts in filemap_fault(). This fixes an oversight -- waiting for the I/O to complete is now interruptible by a fatal signal. And it saves us a few bytes of text in an unlikely path. $ ./scripts/bloat-o-meter before.o after.o add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-207 (-207) Function old new delta filemap_fault 2187 1980 -207 Total: Before=37491, After=37284, chg -0.55% Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> --- mm/filemap.c | 10 +--------- 1 file changed, 1 insertion(+), 9 deletions(-)