| Message ID | 20200219210103.32400-22-willy@infradead.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Change readahead API | expand |
On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > By putting the 'have we reached the end of the page' condition at the end > of the loop instead of the beginning, we can remove the 'submit the last > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > didn't return 0, which would lead to an endless loop. I'm obviously biassed a I wrote the original code, but I find the new very much harder to understand (not that the previous one was easy, this is tricky code..).
On Thu, Feb 20, 2020 at 07:47:41AM -0800, Christoph Hellwig wrote: > On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > > > By putting the 'have we reached the end of the page' condition at the end > > of the loop instead of the beginning, we can remove the 'submit the last > > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > > didn't return 0, which would lead to an endless loop. > > I'm obviously biassed a I wrote the original code, but I find the new > very much harder to understand (not that the previous one was easy, this > is tricky code..). Agreed, I found the original code hard to understand. I think this is easier because now cur_page doesn't leak outside this loop, so it has an obvious lifecycle. I'm kind of optimistic for Dave Howells' iov_iter addition of an ITER_MAPPING. That might simplify all of this code.
On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > By putting the 'have we reached the end of the page' condition at the end > of the loop instead of the beginning, we can remove the 'submit the last > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > didn't return 0, which would lead to an endless loop. > > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> > --- > fs/iomap/buffered-io.c | 32 ++++++++++++++++++-------------- > 1 file changed, 18 insertions(+), 14 deletions(-) > > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c > index cb3511eb152a..31899e6cb0f8 100644 > --- a/fs/iomap/buffered-io.c > +++ b/fs/iomap/buffered-io.c > @@ -400,15 +400,9 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, > void *data, struct iomap *iomap, struct iomap *srcmap) > { > struct iomap_readpage_ctx *ctx = data; > - loff_t done, ret; > - > - for (done = 0; done < length; done += ret) { > - if (ctx->cur_page && offset_in_page(pos + done) == 0) { > - if (!ctx->cur_page_in_bio) > - unlock_page(ctx->cur_page); > - put_page(ctx->cur_page); > - ctx->cur_page = NULL; > - } > + loff_t ret, done = 0; > + > + while (done < length) { > if (!ctx->cur_page) { > ctx->cur_page = iomap_next_page(inode, ctx->pages, > pos, length, &done); > @@ -418,6 +412,20 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, > } > ret = iomap_readpage_actor(inode, pos + done, length - done, > ctx, iomap, srcmap); > + done += ret; > + > + /* Keep working on a partial page */ > + if (ret && offset_in_page(pos + done)) > + continue; > + > + if (!ctx->cur_page_in_bio) > + unlock_page(ctx->cur_page); > + put_page(ctx->cur_page); > + ctx->cur_page = NULL; > + > + /* Don't loop forever if we made no progress */ > + if (WARN_ON(!ret)) > + break; > } > > return done; > @@ -451,11 +459,7 @@ iomap_readpages(struct address_space *mapping, struct list_head *pages, > done: > if (ctx.bio) > submit_bio(ctx.bio); > - if (ctx.cur_page) { > - if (!ctx.cur_page_in_bio) > - unlock_page(ctx.cur_page); > - put_page(ctx.cur_page); > - } > + BUG_ON(ctx.cur_page); Whoah, is the system totally unrecoverably hosed at this point? I get that this /shouldn't/ happen, but should we somehow end up with a page here, are we unable either to release it or even just leak it? I'd have thought a WARN_ON would be just fine here. --D > > /* > * Check that we didn't lose a page due to the arcance calling > -- > 2.25.0 >
On Fri, Feb 21, 2020 at 04:44:25PM -0800, Darrick J. Wong wrote: > On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > > > By putting the 'have we reached the end of the page' condition at the end > > of the loop instead of the beginning, we can remove the 'submit the last > > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > > didn't return 0, which would lead to an endless loop. > > > > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> > > --- > > fs/iomap/buffered-io.c | 32 ++++++++++++++++++-------------- > > 1 file changed, 18 insertions(+), 14 deletions(-) > > > > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c > > index cb3511eb152a..31899e6cb0f8 100644 > > --- a/fs/iomap/buffered-io.c > > +++ b/fs/iomap/buffered-io.c > > @@ -400,15 +400,9 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, > > void *data, struct iomap *iomap, struct iomap *srcmap) > > { > > struct iomap_readpage_ctx *ctx = data; > > - loff_t done, ret; > > - > > - for (done = 0; done < length; done += ret) { > > - if (ctx->cur_page && offset_in_page(pos + done) == 0) { > > - if (!ctx->cur_page_in_bio) > > - unlock_page(ctx->cur_page); > > - put_page(ctx->cur_page); > > - ctx->cur_page = NULL; > > - } > > + loff_t ret, done = 0; > > + > > + while (done < length) { > > if (!ctx->cur_page) { > > ctx->cur_page = iomap_next_page(inode, ctx->pages, > > pos, length, &done); > > @@ -418,6 +412,20 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, > > } > > ret = iomap_readpage_actor(inode, pos + done, length - done, > > ctx, iomap, srcmap); > > + done += ret; > > + > > + /* Keep working on a partial page */ > > + if (ret && offset_in_page(pos + done)) > > + continue; > > + > > + if (!ctx->cur_page_in_bio) > > + unlock_page(ctx->cur_page); > > + put_page(ctx->cur_page); > > + ctx->cur_page = NULL; > > + > > + /* Don't loop forever if we made no progress */ > > + if (WARN_ON(!ret)) > > + break; > > } > > > > return done; > > @@ -451,11 +459,7 @@ iomap_readpages(struct address_space *mapping, struct list_head *pages, > > done: > > if (ctx.bio) > > submit_bio(ctx.bio); > > - if (ctx.cur_page) { > > - if (!ctx.cur_page_in_bio) > > - unlock_page(ctx.cur_page); > > - put_page(ctx.cur_page); > > - } > > + BUG_ON(ctx.cur_page); > > Whoah, is the system totally unrecoverably hosed at this point? > > I get that this /shouldn't/ happen, but should we somehow end up with a > page here, are we unable either to release it or even just leak it? I'd > have thought a WARN_ON would be just fine here. If we do find a page here, we don't actually know what to do with it. It might be (currently) locked, it might have the wrong refcount. Whatever is going on, it's probably better that we stop everything right here rather than allow things to go further and possibly present bad data to the application. I mean, we could even be leaking the previous contents of this page to userspace. Or maybe the future contents of a page which shouldn't be in the page cache any more, but userspace gets a mapping to it. I'm not enthusiastic about putting in some code here to try to handle a "can't happen" case, since it's never going to be tested, and might end up causing more problems than it tries to solve. Let's just stop.
On Fri, Feb 21, 2020 at 05:54:35PM -0800, Matthew Wilcox wrote: > On Fri, Feb 21, 2020 at 04:44:25PM -0800, Darrick J. Wong wrote: > > On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > > > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > > > > > By putting the 'have we reached the end of the page' condition at the end > > > of the loop instead of the beginning, we can remove the 'submit the last > > > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > > > didn't return 0, which would lead to an endless loop. > > > > > > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> > > > --- > > > fs/iomap/buffered-io.c | 32 ++++++++++++++++++-------------- > > > 1 file changed, 18 insertions(+), 14 deletions(-) > > > > > > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c > > > index cb3511eb152a..31899e6cb0f8 100644 > > > --- a/fs/iomap/buffered-io.c > > > +++ b/fs/iomap/buffered-io.c > > > @@ -400,15 +400,9 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, > > > void *data, struct iomap *iomap, struct iomap *srcmap) > > > { > > > struct iomap_readpage_ctx *ctx = data; > > > - loff_t done, ret; > > > - > > > - for (done = 0; done < length; done += ret) { > > > - if (ctx->cur_page && offset_in_page(pos + done) == 0) { > > > - if (!ctx->cur_page_in_bio) > > > - unlock_page(ctx->cur_page); > > > - put_page(ctx->cur_page); > > > - ctx->cur_page = NULL; > > > - } > > > + loff_t ret, done = 0; > > > + > > > + while (done < length) { > > > if (!ctx->cur_page) { > > > ctx->cur_page = iomap_next_page(inode, ctx->pages, > > > pos, length, &done); > > > @@ -418,6 +412,20 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, > > > } > > > ret = iomap_readpage_actor(inode, pos + done, length - done, > > > ctx, iomap, srcmap); > > > + done += ret; > > > + > > > + /* Keep working on a partial page */ > > > + if (ret && offset_in_page(pos + done)) > > > + continue; > > > + > > > + if (!ctx->cur_page_in_bio) > > > + unlock_page(ctx->cur_page); > > > + put_page(ctx->cur_page); > > > + ctx->cur_page = NULL; > > > + > > > + /* Don't loop forever if we made no progress */ > > > + if (WARN_ON(!ret)) > > > + break; > > > } > > > > > > return done; > > > @@ -451,11 +459,7 @@ iomap_readpages(struct address_space *mapping, struct list_head *pages, > > > done: > > > if (ctx.bio) > > > submit_bio(ctx.bio); > > > - if (ctx.cur_page) { > > > - if (!ctx.cur_page_in_bio) > > > - unlock_page(ctx.cur_page); > > > - put_page(ctx.cur_page); > > > - } > > > + BUG_ON(ctx.cur_page); > > > > Whoah, is the system totally unrecoverably hosed at this point? > > > > I get that this /shouldn't/ happen, but should we somehow end up with a > > page here, are we unable either to release it or even just leak it? I'd > > have thought a WARN_ON would be just fine here. > > If we do find a page here, we don't actually know what to do with it. > It might be (currently) locked, it might have the wrong refcount. > Whatever is going on, it's probably better that we stop everything right > here rather than allow things to go further and possibly present bad > data to the application. I mean, we could even be leaking the previous > contents of this page to userspace. Or maybe the future contents of a > page which shouldn't be in the page cache any more, but userspace gets > a mapping to it. > > I'm not enthusiastic about putting in some code here to try to handle > a "can't happen" case, since it's never going to be tested, and might > end up causing more problems than it tries to solve. Let's just stop. Seeing how Linus (and others like myself) are a bit allergic to BUG these days, could you add the first paragraph of the above justification as a comment adjacent to the BUG_ON(), please? :) --D
On Thu, Feb 20, 2020 at 08:24:04AM -0800, Matthew Wilcox wrote: > On Thu, Feb 20, 2020 at 07:47:41AM -0800, Christoph Hellwig wrote: > > On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > > > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > > > > > By putting the 'have we reached the end of the page' condition at the end > > > of the loop instead of the beginning, we can remove the 'submit the last > > > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > > > didn't return 0, which would lead to an endless loop. > > > > I'm obviously biassed a I wrote the original code, but I find the new > > very much harder to understand (not that the previous one was easy, this > > is tricky code..). > > Agreed, I found the original code hard to understand. I think this is > easier because now cur_page doesn't leak outside this loop, so it has > an obvious lifecycle. I really don't like this patch, and would prefer if the series goes ahead without it, as the current sctructure works just fine even with the readahead changes.
On Mon, Feb 24, 2020 at 02:17:49PM -0800, Christoph Hellwig wrote: > On Thu, Feb 20, 2020 at 08:24:04AM -0800, Matthew Wilcox wrote: > > On Thu, Feb 20, 2020 at 07:47:41AM -0800, Christoph Hellwig wrote: > > > On Wed, Feb 19, 2020 at 01:01:00PM -0800, Matthew Wilcox wrote: > > > > From: "Matthew Wilcox (Oracle)" <willy@infradead.org> > > > > > > > > By putting the 'have we reached the end of the page' condition at the end > > > > of the loop instead of the beginning, we can remove the 'submit the last > > > > page' code from iomap_readpages(). Also check that iomap_readpage_actor() > > > > didn't return 0, which would lead to an endless loop. > > > > > > I'm obviously biassed a I wrote the original code, but I find the new > > > very much harder to understand (not that the previous one was easy, this > > > is tricky code..). > > > > Agreed, I found the original code hard to understand. I think this is > > easier because now cur_page doesn't leak outside this loop, so it has > > an obvious lifecycle. > > I really don't like this patch, and would prefer if the series goes > ahead without it, as the current sctructure works just fine even > with the readahead changes. Dave Chinner specifically asked me to do it this way, so please fight amongst yourselves.
diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c index cb3511eb152a..31899e6cb0f8 100644 --- a/fs/iomap/buffered-io.c +++ b/fs/iomap/buffered-io.c @@ -400,15 +400,9 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, void *data, struct iomap *iomap, struct iomap *srcmap) { struct iomap_readpage_ctx *ctx = data; - loff_t done, ret; - - for (done = 0; done < length; done += ret) { - if (ctx->cur_page && offset_in_page(pos + done) == 0) { - if (!ctx->cur_page_in_bio) - unlock_page(ctx->cur_page); - put_page(ctx->cur_page); - ctx->cur_page = NULL; - } + loff_t ret, done = 0; + + while (done < length) { if (!ctx->cur_page) { ctx->cur_page = iomap_next_page(inode, ctx->pages, pos, length, &done); @@ -418,6 +412,20 @@ iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, } ret = iomap_readpage_actor(inode, pos + done, length - done, ctx, iomap, srcmap); + done += ret; + + /* Keep working on a partial page */ + if (ret && offset_in_page(pos + done)) + continue; + + if (!ctx->cur_page_in_bio) + unlock_page(ctx->cur_page); + put_page(ctx->cur_page); + ctx->cur_page = NULL; + + /* Don't loop forever if we made no progress */ + if (WARN_ON(!ret)) + break; } return done; @@ -451,11 +459,7 @@ iomap_readpages(struct address_space *mapping, struct list_head *pages, done: if (ctx.bio) submit_bio(ctx.bio); - if (ctx.cur_page) { - if (!ctx.cur_page_in_bio) - unlock_page(ctx.cur_page); - put_page(ctx.cur_page); - } + BUG_ON(ctx.cur_page); /* * Check that we didn't lose a page due to the arcance calling