[PATCH 0/2] generic_file_buffered_read() refactoring, perf improvements

[PATCH 0/2] generic_file_buffered_read() refactoring, perf improvements

[Kent Overstreet]

Rebased this patchset onto 5.9, I'd like to finally get this because
generic_file_buffered_read() has turned into a real monstrosity to work with.
And it's a major performance improvement, for both small random and large
sequential reads. On my test box, 4k buffered random reads go from ~150k to
~250k iops, and the improvements to big sequential reads are even bigger.

This incorporates the fix for IOCB_WAITQ handling that Jens just posted as well,
also factors out lock_page_for_iocb() to improve handling of the various iocb
flags.

Kent Overstreet (2):
  fs: Break generic_file_buffered_read up into multiple functions
  fs: generic_file_buffered_read() now uses find_get_pages_contig

 mm/filemap.c | 563 ++++++++++++++++++++++++++++++---------------------
 1 file changed, 328 insertions(+), 235 deletions(-)

-- 
2.28.0

[PATCH 1/2] fs: Break generic_file_buffered_read up into multiple functions

[Kent Overstreet]

This is prep work for changing generic_file_buffered_read() to use
find_get_pages_contig() to batch up all the pagecache lookups.

This patch should be functionally identical to the existing code and
changes as little as of the flow control as possible. More refactoring
could be done, this patch is intended to be relatively minimal.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
---
 mm/filemap.c | 473 ++++++++++++++++++++++++++++-----------------------
 1 file changed, 261 insertions(+), 212 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index 99c49eeae7..482fd75d66 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2138,6 +2138,234 @@ static void shrink_readahead_size_eio(struct file_ra_state *ra)
 	ra->ra_pages /= 4;
 }
 
+static int lock_page_for_iocb(struct kiocb *iocb, struct page *page)
+{
+	if (iocb->ki_flags & IOCB_WAITQ)
+		return lock_page_async(page, iocb->ki_waitq);
+	else if (iocb->ki_flags & IOCB_NOWAIT)
+		return trylock_page(page) ? 0 : -EAGAIN;
+	else
+		return lock_page_killable(page);
+}
+
+static int generic_file_buffered_read_page_ok(struct kiocb *iocb,
+			struct iov_iter *iter,
+			struct page *page)
+{
+	struct address_space *mapping = iocb->ki_filp->f_mapping;
+	struct inode *inode = mapping->host;
+	struct file_ra_state *ra = &iocb->ki_filp->f_ra;
+	unsigned int offset = iocb->ki_pos & ~PAGE_MASK;
+	unsigned int bytes, copied;
+	loff_t isize, end_offset;
+
+	BUG_ON(iocb->ki_pos >> PAGE_SHIFT != page->index);
+
+	/*
+	 * i_size must be checked after we know the page is Uptodate.
+	 *
+	 * Checking i_size after the check allows us to calculate
+	 * the correct value for "bytes", which means the zero-filled
+	 * part of the page is not copied back to userspace (unless
+	 * another truncate extends the file - this is desired though).
+	 */
+
+	isize = i_size_read(inode);
+	if (unlikely(iocb->ki_pos >= isize))
+		return 1;
+
+	end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
+
+	bytes = min_t(loff_t, end_offset - iocb->ki_pos, PAGE_SIZE - offset);
+
+	/* If users can be writing to this page using arbitrary
+	 * virtual addresses, take care about potential aliasing
+	 * before reading the page on the kernel side.
+	 */
+	if (mapping_writably_mapped(mapping))
+		flush_dcache_page(page);
+
+	/*
+	 * Ok, we have the page, and it's up-to-date, so
+	 * now we can copy it to user space...
+	 */
+
+	copied = copy_page_to_iter(page, offset, bytes, iter);
+
+	iocb->ki_pos += copied;
+
+	/*
+	 * When a sequential read accesses a page several times,
+	 * only mark it as accessed the first time.
+	 */
+	if (iocb->ki_pos >> PAGE_SHIFT != ra->prev_pos >> PAGE_SHIFT)
+		mark_page_accessed(page);
+
+	ra->prev_pos = iocb->ki_pos;
+
+	if (copied < bytes)
+		return -EFAULT;
+
+	return !iov_iter_count(iter) || iocb->ki_pos == isize;
+}
+
+static struct page *
+generic_file_buffered_read_readpage(struct kiocb *iocb,
+				    struct file *filp,
+				    struct address_space *mapping,
+				    struct page *page)
+{
+	struct file_ra_state *ra = &filp->f_ra;
+	int error;
+
+	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) {
+		unlock_page(page);
+		put_page(page);
+		return ERR_PTR(-EAGAIN);
+	}
+
+	/*
+	 * A previous I/O error may have been due to temporary
+	 * failures, eg. multipath errors.
+	 * PG_error will be set again if readpage fails.
+	 */
+	ClearPageError(page);
+	/* Start the actual read. The read will unlock the page. */
+	error = mapping->a_ops->readpage(filp, page);
+
+	if (unlikely(error)) {
+		put_page(page);
+		return error != AOP_TRUNCATED_PAGE ? ERR_PTR(error) : NULL;
+	}
+
+	if (!PageUptodate(page)) {
+		error = lock_page_for_iocb(iocb, page);
+		if (unlikely(error)) {
+			put_page(page);
+			return ERR_PTR(error);
+		}
+		if (!PageUptodate(page)) {
+			if (page->mapping == NULL) {
+				/*
+				 * invalidate_mapping_pages got it
+				 */
+				unlock_page(page);
+				put_page(page);
+				return NULL;
+			}
+			unlock_page(page);
+			shrink_readahead_size_eio(ra);
+			put_page(page);
+			return ERR_PTR(-EIO);
+		}
+		unlock_page(page);
+	}
+
+	return page;
+}
+
+static struct page *
+generic_file_buffered_read_pagenotuptodate(struct kiocb *iocb,
+					   struct file *filp,
+					   struct iov_iter *iter,
+					   struct page *page,
+					   loff_t pos, loff_t count)
+{
+	struct address_space *mapping = filp->f_mapping;
+	struct inode *inode = mapping->host;
+	int error;
+
+	/*
+	 * See comment in do_read_cache_page on why
+	 * wait_on_page_locked is used to avoid unnecessarily
+	 * serialisations and why it's safe.
+	 */
+	if (iocb->ki_flags & IOCB_WAITQ) {
+		error = wait_on_page_locked_async(page,
+						iocb->ki_waitq);
+	} else {
+		error = wait_on_page_locked_killable(page);
+	}
+	if (unlikely(error)) {
+		put_page(page);
+		return ERR_PTR(error);
+	}
+	if (PageUptodate(page))
+		return page;
+
+	if (inode->i_blkbits == PAGE_SHIFT ||
+			!mapping->a_ops->is_partially_uptodate)
+		goto page_not_up_to_date;
+	/* pipes can't handle partially uptodate pages */
+	if (unlikely(iov_iter_is_pipe(iter)))
+		goto page_not_up_to_date;
+	if (!trylock_page(page))
+		goto page_not_up_to_date;
+	/* Did it get truncated before we got the lock? */
+	if (!page->mapping)
+		goto page_not_up_to_date_locked;
+	if (!mapping->a_ops->is_partially_uptodate(page,
+				pos & ~PAGE_MASK, count))
+		goto page_not_up_to_date_locked;
+	unlock_page(page);
+	return page;
+
+page_not_up_to_date:
+	/* Get exclusive access to the page ... */
+	error = lock_page_for_iocb(iocb, page);
+	if (unlikely(error)) {
+		put_page(page);
+		return ERR_PTR(error);
+	}
+
+page_not_up_to_date_locked:
+	/* Did it get truncated before we got the lock? */
+	if (!page->mapping) {
+		unlock_page(page);
+		put_page(page);
+		return NULL;
+	}
+
+	/* Did somebody else fill it already? */
+	if (PageUptodate(page)) {
+		unlock_page(page);
+		return page;
+	}
+
+	return generic_file_buffered_read_readpage(iocb, filp, mapping, page);
+}
+
+static struct page *
+generic_file_buffered_read_no_cached_page(struct kiocb *iocb,
+					  struct iov_iter *iter)
+{
+	struct file *filp = iocb->ki_filp;
+	struct address_space *mapping = filp->f_mapping;
+	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
+	struct page *page;
+	int error;
+
+	if (iocb->ki_flags & IOCB_NOIO)
+		return ERR_PTR(-EAGAIN);
+
+	/*
+	 * Ok, it wasn't cached, so we need to create a new
+	 * page..
+	 */
+	page = page_cache_alloc(mapping);
+	if (!page)
+		return ERR_PTR(-ENOMEM);
+
+	error = add_to_page_cache_lru(page, mapping, index,
+				      mapping_gfp_constraint(mapping, GFP_KERNEL));
+	if (error) {
+		put_page(page);
+		return error != -EEXIST ? ERR_PTR(error) : NULL;
+	}
+
+	return generic_file_buffered_read_readpage(iocb, filp, mapping, page);
+}
+
 /**
  * generic_file_buffered_read - generic file read routine
  * @iocb:	the iocb to read
@@ -2161,29 +2389,19 @@ ssize_t generic_file_buffered_read(struct kiocb *iocb,
 	struct address_space *mapping = filp->f_mapping;
 	struct inode *inode = mapping->host;
 	struct file_ra_state *ra = &filp->f_ra;
-	loff_t *ppos = &iocb->ki_pos;
-	pgoff_t index;
+	size_t orig_count = iov_iter_count(iter);
 	pgoff_t last_index;
-	pgoff_t prev_index;
-	unsigned long offset;      /* offset into pagecache page */
-	unsigned int prev_offset;
 	int error = 0;
 
-	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
+	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
 		return 0;
 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
 
-	index = *ppos >> PAGE_SHIFT;
-	prev_index = ra->prev_pos >> PAGE_SHIFT;
-	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
-	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
-	offset = *ppos & ~PAGE_MASK;
+	last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
 
 	for (;;) {
+		pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
 		struct page *page;
-		pgoff_t end_index;
-		loff_t isize;
-		unsigned long nr, ret;
 
 		cond_resched();
 find_page:
@@ -2192,6 +2410,14 @@ ssize_t generic_file_buffered_read(struct kiocb *iocb,
 			goto out;
 		}
 
+		/*
+		 * We can't return -EIOCBQUEUED once we've done some work, so
+		 * ensure we don't block:
+		 */
+		if ((iocb->ki_flags & IOCB_WAITQ) &&
+		    (written + orig_count - iov_iter_count(iter)))
+			iocb->ki_flags |= IOCB_NOWAIT;
+
 		page = find_get_page(mapping, index);
 		if (!page) {
 			if (iocb->ki_flags & IOCB_NOIO)
@@ -2200,8 +2426,15 @@ ssize_t generic_file_buffered_read(struct kiocb *iocb,
 					ra, filp,
 					index, last_index - index);
 			page = find_get_page(mapping, index);
-			if (unlikely(page == NULL))
-				goto no_cached_page;
+			if (unlikely(page == NULL)) {
+				page = generic_file_buffered_read_no_cached_page(iocb, iter);
+				if (!page)
+					goto find_page;
+				if (IS_ERR(page)) {
+					error = PTR_ERR(page);
+					goto out;
+				}
+			}
 		}
 		if (PageReadahead(page)) {
 			if (iocb->ki_flags & IOCB_NOIO) {
@@ -2213,221 +2446,37 @@ ssize_t generic_file_buffered_read(struct kiocb *iocb,
 					index, last_index - index);
 		}
 		if (!PageUptodate(page)) {
-			/*
-			 * See comment in do_read_cache_page on why
-			 * wait_on_page_locked is used to avoid unnecessarily
-			 * serialisations and why it's safe.
-			 */
-			if (iocb->ki_flags & IOCB_WAITQ) {
-				if (written) {
-					put_page(page);
-					goto out;
-				}
-				error = wait_on_page_locked_async(page,
-								iocb->ki_waitq);
-			} else {
-				if (iocb->ki_flags & IOCB_NOWAIT) {
-					put_page(page);
-					goto would_block;
-				}
-				error = wait_on_page_locked_killable(page);
-			}
-			if (unlikely(error))
-				goto readpage_error;
-			if (PageUptodate(page))
-				goto page_ok;
-
-			if (inode->i_blkbits == PAGE_SHIFT ||
-					!mapping->a_ops->is_partially_uptodate)
-				goto page_not_up_to_date;
-			/* pipes can't handle partially uptodate pages */
-			if (unlikely(iov_iter_is_pipe(iter)))
-				goto page_not_up_to_date;
-			if (!trylock_page(page))
-				goto page_not_up_to_date;
-			/* Did it get truncated before we got the lock? */
-			if (!page->mapping)
-				goto page_not_up_to_date_locked;
-			if (!mapping->a_ops->is_partially_uptodate(page,
-							offset, iter->count))
-				goto page_not_up_to_date_locked;
-			unlock_page(page);
-		}
-page_ok:
-		/*
-		 * i_size must be checked after we know the page is Uptodate.
-		 *
-		 * Checking i_size after the check allows us to calculate
-		 * the correct value for "nr", which means the zero-filled
-		 * part of the page is not copied back to userspace (unless
-		 * another truncate extends the file - this is desired though).
-		 */
-
-		isize = i_size_read(inode);
-		end_index = (isize - 1) >> PAGE_SHIFT;
-		if (unlikely(!isize || index > end_index)) {
-			put_page(page);
-			goto out;
-		}
-
-		/* nr is the maximum number of bytes to copy from this page */
-		nr = PAGE_SIZE;
-		if (index == end_index) {
-			nr = ((isize - 1) & ~PAGE_MASK) + 1;
-			if (nr <= offset) {
+			if (iocb->ki_flags & IOCB_NOWAIT) {
 				put_page(page);
+				error = -EAGAIN;
 				goto out;
 			}
-		}
-		nr = nr - offset;
-
-		/* If users can be writing to this page using arbitrary
-		 * virtual addresses, take care about potential aliasing
-		 * before reading the page on the kernel side.
-		 */
-		if (mapping_writably_mapped(mapping))
-			flush_dcache_page(page);
-
-		/*
-		 * When a sequential read accesses a page several times,
-		 * only mark it as accessed the first time.
-		 */
-		if (prev_index != index || offset != prev_offset)
-			mark_page_accessed(page);
-		prev_index = index;
-
-		/*
-		 * Ok, we have the page, and it's up-to-date, so
-		 * now we can copy it to user space...
-		 */
-
-		ret = copy_page_to_iter(page, offset, nr, iter);
-		offset += ret;
-		index += offset >> PAGE_SHIFT;
-		offset &= ~PAGE_MASK;
-		prev_offset = offset;
-
-		put_page(page);
-		written += ret;
-		if (!iov_iter_count(iter))
-			goto out;
-		if (ret < nr) {
-			error = -EFAULT;
-			goto out;
-		}
-		continue;
-
-page_not_up_to_date:
-		/* Get exclusive access to the page ... */
-		if (iocb->ki_flags & IOCB_WAITQ)
-			error = lock_page_async(page, iocb->ki_waitq);
-		else
-			error = lock_page_killable(page);
-		if (unlikely(error))
-			goto readpage_error;
-
-page_not_up_to_date_locked:
-		/* Did it get truncated before we got the lock? */
-		if (!page->mapping) {
-			unlock_page(page);
-			put_page(page);
-			continue;
-		}
-
-		/* Did somebody else fill it already? */
-		if (PageUptodate(page)) {
-			unlock_page(page);
-			goto page_ok;
-		}
-
-readpage:
-		if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) {
-			unlock_page(page);
-			put_page(page);
-			goto would_block;
-		}
-		/*
-		 * A previous I/O error may have been due to temporary
-		 * failures, eg. multipath errors.
-		 * PG_error will be set again if readpage fails.
-		 */
-		ClearPageError(page);
-		/* Start the actual read. The read will unlock the page. */
-		error = mapping->a_ops->readpage(filp, page);
-
-		if (unlikely(error)) {
-			if (error == AOP_TRUNCATED_PAGE) {
-				put_page(page);
-				error = 0;
+			page = generic_file_buffered_read_pagenotuptodate(iocb,
+					filp, iter, page, iocb->ki_pos, iter->count);
+			if (!page)
 				goto find_page;
+			if (IS_ERR(page)) {
+				error = PTR_ERR(page);
+				goto out;
 			}
-			goto readpage_error;
-		}
-
-		if (!PageUptodate(page)) {
-			if (iocb->ki_flags & IOCB_WAITQ)
-				error = lock_page_async(page, iocb->ki_waitq);
-			else
-				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);
-					goto find_page;
-				}
-				unlock_page(page);
-				shrink_readahead_size_eio(ra);
-				error = -EIO;
-				goto readpage_error;
-			}
-			unlock_page(page);
 		}
 
-		goto page_ok;
-
-readpage_error:
-		/* UHHUH! A synchronous read error occurred. Report it */
+		error = generic_file_buffered_read_page_ok(iocb, iter, page);
 		put_page(page);
-		goto out;
 
-no_cached_page:
-		/*
-		 * Ok, it wasn't cached, so we need to create a new
-		 * page..
-		 */
-		page = page_cache_alloc(mapping);
-		if (!page) {
-			error = -ENOMEM;
-			goto out;
-		}
-		error = add_to_page_cache_lru(page, mapping, index,
-				mapping_gfp_constraint(mapping, GFP_KERNEL));
 		if (error) {
-			put_page(page);
-			if (error == -EEXIST) {
+			if (error > 0)
 				error = 0;
-				goto find_page;
-			}
 			goto out;
 		}
-		goto readpage;
 	}
 
 would_block:
 	error = -EAGAIN;
 out:
-	ra->prev_pos = prev_index;
-	ra->prev_pos <<= PAGE_SHIFT;
-	ra->prev_pos |= prev_offset;
-
-	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
 	file_accessed(filp);
+	written += orig_count - iov_iter_count(iter);
+
 	return written ? written : error;
 }
 EXPORT_SYMBOL_GPL(generic_file_buffered_read);
-- 
2.28.0

[PATCH 2/2] fs: generic_file_buffered_read() now uses find_get_pages_contig

[Kent Overstreet]

Convert generic_file_buffered_read() to get pages to read from in
batches, and then copy data to userspace from many pages at once - in
particular, we now don't touch any cachelines that might be contended
while we're in the loop to copy data to userspace.

This is is a performance improvement on workloads that do buffered reads
with large blocksizes, and a very large performance improvement if that
file is also being accessed concurrently by different threads.

On smaller reads (512 bytes), there's a very small performance
improvement (1%, within the margin of error).

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
---
 mm/filemap.c | 300 +++++++++++++++++++++++++++++----------------------
 1 file changed, 172 insertions(+), 128 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index 482fd75d66..8cf5b5d9e6 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2148,67 +2148,6 @@ static int lock_page_for_iocb(struct kiocb *iocb, struct page *page)
 		return lock_page_killable(page);
 }
 
-static int generic_file_buffered_read_page_ok(struct kiocb *iocb,
-			struct iov_iter *iter,
-			struct page *page)
-{
-	struct address_space *mapping = iocb->ki_filp->f_mapping;
-	struct inode *inode = mapping->host;
-	struct file_ra_state *ra = &iocb->ki_filp->f_ra;
-	unsigned int offset = iocb->ki_pos & ~PAGE_MASK;
-	unsigned int bytes, copied;
-	loff_t isize, end_offset;
-
-	BUG_ON(iocb->ki_pos >> PAGE_SHIFT != page->index);
-
-	/*
-	 * i_size must be checked after we know the page is Uptodate.
-	 *
-	 * Checking i_size after the check allows us to calculate
-	 * the correct value for "bytes", which means the zero-filled
-	 * part of the page is not copied back to userspace (unless
-	 * another truncate extends the file - this is desired though).
-	 */
-
-	isize = i_size_read(inode);
-	if (unlikely(iocb->ki_pos >= isize))
-		return 1;
-
-	end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
-
-	bytes = min_t(loff_t, end_offset - iocb->ki_pos, PAGE_SIZE - offset);
-
-	/* If users can be writing to this page using arbitrary
-	 * virtual addresses, take care about potential aliasing
-	 * before reading the page on the kernel side.
-	 */
-	if (mapping_writably_mapped(mapping))
-		flush_dcache_page(page);
-
-	/*
-	 * Ok, we have the page, and it's up-to-date, so
-	 * now we can copy it to user space...
-	 */
-
-	copied = copy_page_to_iter(page, offset, bytes, iter);
-
-	iocb->ki_pos += copied;
-
-	/*
-	 * When a sequential read accesses a page several times,
-	 * only mark it as accessed the first time.
-	 */
-	if (iocb->ki_pos >> PAGE_SHIFT != ra->prev_pos >> PAGE_SHIFT)
-		mark_page_accessed(page);
-
-	ra->prev_pos = iocb->ki_pos;
-
-	if (copied < bytes)
-		return -EFAULT;
-
-	return !iov_iter_count(iter) || iocb->ki_pos == isize;
-}
-
 static struct page *
 generic_file_buffered_read_readpage(struct kiocb *iocb,
 				    struct file *filp,
@@ -2366,6 +2305,92 @@ generic_file_buffered_read_no_cached_page(struct kiocb *iocb,
 	return generic_file_buffered_read_readpage(iocb, filp, mapping, page);
 }
 
+static int generic_file_buffered_read_get_pages(struct kiocb *iocb,
+						struct iov_iter *iter,
+						struct page **pages,
+						unsigned int nr)
+{
+	struct file *filp = iocb->ki_filp;
+	struct address_space *mapping = filp->f_mapping;
+	struct file_ra_state *ra = &filp->f_ra;
+	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
+	pgoff_t last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
+	int i, j, nr_got, err = 0;
+
+	nr = min_t(unsigned long, last_index - index, nr);
+find_page:
+	if (fatal_signal_pending(current))
+		return -EINTR;
+
+	nr_got = find_get_pages_contig(mapping, index, nr, pages);
+	if (nr_got)
+		goto got_pages;
+
+	if (iocb->ki_flags & IOCB_NOIO)
+		return -EAGAIN;
+
+	page_cache_sync_readahead(mapping, ra, filp, index, last_index - index);
+
+	nr_got = find_get_pages_contig(mapping, index, nr, pages);
+	if (nr_got)
+		goto got_pages;
+
+	pages[0] = generic_file_buffered_read_no_cached_page(iocb, iter);
+	err = PTR_ERR_OR_ZERO(pages[0]);
+	if (!IS_ERR_OR_NULL(pages[0]))
+		nr_got = 1;
+got_pages:
+	for (i = 0; i < nr_got; i++) {
+		struct page *page = pages[i];
+		pgoff_t pg_index = index + i;
+		loff_t pg_pos = max(iocb->ki_pos,
+				    (loff_t) pg_index << PAGE_SHIFT);
+		loff_t pg_count = iocb->ki_pos + iter->count - pg_pos;
+
+		if (PageReadahead(page)) {
+			if (iocb->ki_flags & IOCB_NOIO) {
+				for (j = i; j < nr_got; j++)
+					put_page(pages[j]);
+				nr_got = i;
+				err = -EAGAIN;
+				break;
+			}
+			page_cache_async_readahead(mapping, ra, filp, page,
+					pg_index, last_index - pg_index);
+		}
+
+		if (!PageUptodate(page)) {
+			if ((iocb->ki_flags & IOCB_NOWAIT) ||
+			    ((iocb->ki_flags & IOCB_WAITQ) && i)) {
+				for (j = i; j < nr_got; j++)
+					put_page(pages[j]);
+				nr_got = i;
+				err = -EAGAIN;
+				break;
+			}
+
+			page = generic_file_buffered_read_pagenotuptodate(iocb,
+					filp, iter, page, pg_pos, pg_count);
+			if (IS_ERR_OR_NULL(page)) {
+				for (j = i + 1; j < nr_got; j++)
+					put_page(pages[j]);
+				nr_got = i;
+				err = PTR_ERR_OR_ZERO(page);
+				break;
+			}
+		}
+	}
+
+	if (likely(nr_got))
+		return nr_got;
+	if (err)
+		return err;
+	/*
+	 * No pages and no error means we raced and should retry:
+	 */
+	goto find_page;
+}
+
 /**
  * generic_file_buffered_read - generic file read routine
  * @iocb:	the iocb to read
@@ -2386,96 +2411,115 @@ ssize_t generic_file_buffered_read(struct kiocb *iocb,
 		struct iov_iter *iter, ssize_t written)
 {
 	struct file *filp = iocb->ki_filp;
+	struct file_ra_state *ra = &filp->f_ra;
 	struct address_space *mapping = filp->f_mapping;
 	struct inode *inode = mapping->host;
-	struct file_ra_state *ra = &filp->f_ra;
-	size_t orig_count = iov_iter_count(iter);
-	pgoff_t last_index;
-	int error = 0;
+	struct page *pages_onstack[PAGEVEC_SIZE], **pages = NULL;
+	unsigned int nr_pages = min_t(unsigned int, 512,
+			((iocb->ki_pos + iter->count + PAGE_SIZE - 1) >> PAGE_SHIFT) -
+			(iocb->ki_pos >> PAGE_SHIFT));
+	int i, pg_nr, error = 0;
+	bool writably_mapped;
+	loff_t isize, end_offset;
 
 	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
 		return 0;
 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
 
-	last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
+	if (nr_pages > ARRAY_SIZE(pages_onstack))
+		pages = kmalloc_array(nr_pages, sizeof(void *), GFP_KERNEL);
 
-	for (;;) {
-		pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
-		struct page *page;
+	if (!pages) {
+		pages = pages_onstack;
+		nr_pages = min_t(unsigned int, nr_pages, ARRAY_SIZE(pages_onstack));
+	}
 
+	do {
 		cond_resched();
-find_page:
-		if (fatal_signal_pending(current)) {
-			error = -EINTR;
-			goto out;
-		}
 
 		/*
 		 * We can't return -EIOCBQUEUED once we've done some work, so
 		 * ensure we don't block:
 		 */
-		if ((iocb->ki_flags & IOCB_WAITQ) &&
-		    (written + orig_count - iov_iter_count(iter)))
+		if ((iocb->ki_flags & IOCB_WAITQ) && written)
 			iocb->ki_flags |= IOCB_NOWAIT;
 
-		page = find_get_page(mapping, index);
-		if (!page) {
-			if (iocb->ki_flags & IOCB_NOIO)
-				goto would_block;
-			page_cache_sync_readahead(mapping,
-					ra, filp,
-					index, last_index - index);
-			page = find_get_page(mapping, index);
-			if (unlikely(page == NULL)) {
-				page = generic_file_buffered_read_no_cached_page(iocb, iter);
-				if (!page)
-					goto find_page;
-				if (IS_ERR(page)) {
-					error = PTR_ERR(page);
-					goto out;
-				}
-			}
-		}
-		if (PageReadahead(page)) {
-			if (iocb->ki_flags & IOCB_NOIO) {
-				put_page(page);
-				goto out;
-			}
-			page_cache_async_readahead(mapping,
-					ra, filp, page,
-					index, last_index - index);
-		}
-		if (!PageUptodate(page)) {
-			if (iocb->ki_flags & IOCB_NOWAIT) {
-				put_page(page);
-				error = -EAGAIN;
-				goto out;
-			}
-			page = generic_file_buffered_read_pagenotuptodate(iocb,
-					filp, iter, page, iocb->ki_pos, iter->count);
-			if (!page)
-				goto find_page;
-			if (IS_ERR(page)) {
-				error = PTR_ERR(page);
-				goto out;
-			}
+		i = 0;
+		pg_nr = generic_file_buffered_read_get_pages(iocb, iter,
+							     pages, nr_pages);
+		if (pg_nr < 0) {
+			error = pg_nr;
+			break;
 		}
 
-		error = generic_file_buffered_read_page_ok(iocb, iter, page);
-		put_page(page);
+		/*
+		 * i_size must be checked after we know the pages are Uptodate.
+		 *
+		 * Checking i_size after the check allows us to calculate
+		 * the correct value for "nr", which means the zero-filled
+		 * part of the page is not copied back to userspace (unless
+		 * another truncate extends the file - this is desired though).
+		 */
+		isize = i_size_read(inode);
+		if (unlikely(iocb->ki_pos >= isize))
+			goto put_pages;
 
-		if (error) {
-			if (error > 0)
-				error = 0;
-			goto out;
+		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
+
+		while ((iocb->ki_pos >> PAGE_SHIFT) + pg_nr >
+		       (end_offset + PAGE_SIZE - 1) >> PAGE_SHIFT)
+			put_page(pages[--pg_nr]);
+
+		/*
+		 * Once we start copying data, we don't want to be touching any
+		 * cachelines that might be contended:
+		 */
+		writably_mapped = mapping_writably_mapped(mapping);
+
+		/*
+		 * When a sequential read accesses a page several times, only
+		 * mark it as accessed the first time.
+		 */
+		if (iocb->ki_pos >> PAGE_SHIFT !=
+		    ra->prev_pos >> PAGE_SHIFT)
+			mark_page_accessed(pages[0]);
+		for (i = 1; i < pg_nr; i++)
+			mark_page_accessed(pages[i]);
+
+		for (i = 0; i < pg_nr; i++) {
+			unsigned int offset = iocb->ki_pos & ~PAGE_MASK;
+			unsigned int bytes = min_t(loff_t, end_offset - iocb->ki_pos,
+						   PAGE_SIZE - offset);
+			unsigned int copied;
+
+			/*
+			 * If users can be writing to this page using arbitrary
+			 * virtual addresses, take care about potential aliasing
+			 * before reading the page on the kernel side.
+			 */
+			if (writably_mapped)
+				flush_dcache_page(pages[i]);
+
+			copied = copy_page_to_iter(pages[i], offset, bytes, iter);
+
+			written += copied;
+			iocb->ki_pos += copied;
+			ra->prev_pos = iocb->ki_pos;
+
+			if (copied < bytes) {
+				error = -EFAULT;
+				break;
+			}
 		}
-	}
+put_pages:
+		for (i = 0; i < pg_nr; i++)
+			put_page(pages[i]);
+	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
 
-would_block:
-	error = -EAGAIN;
-out:
 	file_accessed(filp);
-	written += orig_count - iov_iter_count(iter);
+
+	if (pages != pages_onstack)
+		kfree(pages);
 
 	return written ? written : error;
 }
-- 
2.28.0

Re: [PATCH 1/2] fs: Break generic_file_buffered_read up into multiple functions

[Jens Axboe]

On 10/17/20 2:10 PM, Kent Overstreet wrote:
> This is prep work for changing generic_file_buffered_read() to use
> find_get_pages_contig() to batch up all the pagecache lookups.
> 
> This patch should be functionally identical to the existing code and
> changes as little as of the flow control as possible. More refactoring
> could be done, this patch is intended to be relatively minimal.

This is a sorely needed cleanup.

Reviewed-by: Jens Axboe <axboe@kernel.dk>

-- 
Jens Axboe

Re: [PATCH 2/2] fs: generic_file_buffered_read() now uses find_get_pages_contig

[Jens Axboe]

On Sat, Oct 17, 2020 at 2:11 PM Kent Overstreet <kent.overstreet@gmail.com> wrote:
>
> Convert generic_file_buffered_read() to get pages to read from in
> batches, and then copy data to userspace from many pages at once - in
> particular, we now don't touch any cachelines that might be contended
> while we're in the loop to copy data to userspace.
>
> This is is a performance improvement on workloads that do buffered reads
> with large blocksizes, and a very large performance improvement if that
> file is also being accessed concurrently by different threads.
>
> On smaller reads (512 bytes), there's a very small performance
> improvement (1%, within the margin of error).

I ran this through my buffered testing, and no ill effects observed. It
also provides a nice boost on the read side for a mixed read/write
verification workload I have.

Reviewed-by: Jens Axboe <axboe@kernel.dk>

-- 
Jens Axboe

[PATCH 1/2] fs: Break generic_file_buffered_read up into multiple functions

[Kent Overstreet]

This is prep work for changing generic_file_buffered_read() to use
find_get_pages_contig() to batch up all the pagecache lookups.

This patch should be functionally identical to the existing code and
changes as little as of the flow control as possible. More refactoring
could be done, this patch is intended to be relatively minimal.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
---
 mm/filemap.c | 418 ++++++++++++++++++++++++++++-----------------------
 1 file changed, 228 insertions(+), 190 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index e67fa8ab48..206d51a1c9 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2051,6 +2051,210 @@ static void shrink_readahead_size_eio(struct file_ra_state *ra)
 	ra->ra_pages /= 4;
 }
 
+static int generic_file_buffered_read_page_ok(struct kiocb *iocb,
+			struct iov_iter *iter,
+			struct page *page)
+{
+	struct address_space *mapping = iocb->ki_filp->f_mapping;
+	struct inode *inode = mapping->host;
+	struct file_ra_state *ra = &iocb->ki_filp->f_ra;
+	unsigned offset = iocb->ki_pos & ~PAGE_MASK;
+	unsigned bytes, copied;
+	loff_t isize, end_offset;
+
+	BUG_ON(iocb->ki_pos >> PAGE_SHIFT != page->index);
+
+	/*
+	 * i_size must be checked after we know the page is Uptodate.
+	 *
+	 * Checking i_size after the check allows us to calculate
+	 * the correct value for "bytes", which means the zero-filled
+	 * part of the page is not copied back to userspace (unless
+	 * another truncate extends the file - this is desired though).
+	 */
+
+	isize = i_size_read(inode);
+	if (unlikely(iocb->ki_pos >= isize))
+		return 1;
+
+	end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
+
+	bytes = min_t(loff_t, end_offset - iocb->ki_pos, PAGE_SIZE - offset);
+
+	/* If users can be writing to this page using arbitrary
+	 * virtual addresses, take care about potential aliasing
+	 * before reading the page on the kernel side.
+	 */
+	if (mapping_writably_mapped(mapping))
+		flush_dcache_page(page);
+
+	/*
+	 * Ok, we have the page, and it's up-to-date, so
+	 * now we can copy it to user space...
+	 */
+
+	copied = copy_page_to_iter(page, offset, bytes, iter);
+
+	iocb->ki_pos += copied;
+
+	/*
+	 * When a sequential read accesses a page several times,
+	 * only mark it as accessed the first time.
+	 */
+	if (iocb->ki_pos >> PAGE_SHIFT != ra->prev_pos >> PAGE_SHIFT)
+		mark_page_accessed(page);
+
+	ra->prev_pos = iocb->ki_pos;
+
+	if (copied < bytes)
+		return -EFAULT;
+
+	return !iov_iter_count(iter) || iocb->ki_pos == isize;
+}
+
+static struct page *
+generic_file_buffered_read_readpage(struct file *filp,
+				    struct address_space *mapping,
+				    struct page *page)
+{
+	struct file_ra_state *ra = &filp->f_ra;
+	int error;
+
+	/*
+	 * A previous I/O error may have been due to temporary
+	 * failures, eg. multipath errors.
+	 * PG_error will be set again if readpage fails.
+	 */
+	ClearPageError(page);
+	/* Start the actual read. The read will unlock the page. */
+	error = mapping->a_ops->readpage(filp, page);
+
+	if (unlikely(error)) {
+		put_page(page);
+		return error != AOP_TRUNCATED_PAGE ? ERR_PTR(error) : NULL;
+	}
+
+	if (!PageUptodate(page)) {
+		error = lock_page_killable(page);
+		if (unlikely(error)) {
+			put_page(page);
+			return ERR_PTR(error);
+		}
+		if (!PageUptodate(page)) {
+			if (page->mapping == NULL) {
+				/*
+				 * invalidate_mapping_pages got it
+				 */
+				unlock_page(page);
+				put_page(page);
+				return NULL;
+			}
+			unlock_page(page);
+			shrink_readahead_size_eio(ra);
+			put_page(page);
+			return ERR_PTR(-EIO);
+		}
+		unlock_page(page);
+	}
+
+	return page;
+}
+
+static struct page *
+generic_file_buffered_read_pagenotuptodate(struct file *filp,
+					   struct iov_iter *iter,
+					   struct page *page,
+					   loff_t pos, loff_t count)
+{
+	struct address_space *mapping = filp->f_mapping;
+	struct inode *inode = mapping->host;
+	int error;
+
+	/*
+	 * See comment in do_read_cache_page on why
+	 * wait_on_page_locked is used to avoid unnecessarily
+	 * serialisations and why it's safe.
+	 */
+	error = wait_on_page_locked_killable(page);
+	if (unlikely(error)) {
+		put_page(page);
+		return ERR_PTR(error);
+	}
+
+	if (PageUptodate(page))
+		return page;
+
+	if (inode->i_blkbits == PAGE_SHIFT ||
+	    !mapping->a_ops->is_partially_uptodate)
+		goto page_not_up_to_date;
+	/* pipes can't handle partially uptodate pages */
+	if (unlikely(iov_iter_is_pipe(iter)))
+		goto page_not_up_to_date;
+	if (!trylock_page(page))
+		goto page_not_up_to_date;
+	/* Did it get truncated before we got the lock? */
+	if (!page->mapping)
+		goto page_not_up_to_date_locked;
+
+	if (!mapping->a_ops->is_partially_uptodate(page,
+				pos & ~PAGE_MASK, count))
+		goto page_not_up_to_date_locked;
+	unlock_page(page);
+	return page;
+
+page_not_up_to_date:
+	/* Get exclusive access to the page ... */
+	error = lock_page_killable(page);
+	if (unlikely(error)) {
+		put_page(page);
+		return ERR_PTR(error);
+	}
+
+page_not_up_to_date_locked:
+	/* Did it get truncated before we got the lock? */
+	if (!page->mapping) {
+		unlock_page(page);
+		put_page(page);
+		return NULL;
+	}
+
+	/* Did somebody else fill it already? */
+	if (PageUptodate(page)) {
+		unlock_page(page);
+		return page;
+	}
+
+	return generic_file_buffered_read_readpage(filp, mapping, page);
+}
+
+static struct page *
+generic_file_buffered_read_no_cached_page(struct kiocb *iocb,
+					  struct iov_iter *iter)
+{
+	struct file *filp = iocb->ki_filp;
+	struct address_space *mapping = filp->f_mapping;
+	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
+	struct page *page;
+	int error;
+
+	/*
+	 * Ok, it wasn't cached, so we need to create a new
+	 * page..
+	 */
+	page = page_cache_alloc(mapping);
+	if (!page)
+		return ERR_PTR(-ENOMEM);
+
+	error = add_to_page_cache_lru(page, mapping, index,
+				      mapping_gfp_constraint(mapping, GFP_KERNEL));
+	if (error) {
+		put_page(page);
+		return error != -EEXIST ? ERR_PTR(error) : NULL;
+	}
+
+	return generic_file_buffered_read_readpage(filp, mapping, page);
+}
+
 /**
  * generic_file_buffered_read - generic file read routine
  * @iocb:	the iocb to read
@@ -2074,29 +2278,19 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
 	struct address_space *mapping = filp->f_mapping;
 	struct inode *inode = mapping->host;
 	struct file_ra_state *ra = &filp->f_ra;
-	loff_t *ppos = &iocb->ki_pos;
-	pgoff_t index;
+	size_t orig_count = iov_iter_count(iter);
 	pgoff_t last_index;
-	pgoff_t prev_index;
-	unsigned long offset;      /* offset into pagecache page */
-	unsigned int prev_offset;
 	int error = 0;
 
-	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
+	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
 		return 0;
 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
 
-	index = *ppos >> PAGE_SHIFT;
-	prev_index = ra->prev_pos >> PAGE_SHIFT;
-	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
-	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
-	offset = *ppos & ~PAGE_MASK;
+	last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
 
 	for (;;) {
+		pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
 		struct page *page;
-		pgoff_t end_index;
-		loff_t isize;
-		unsigned long nr, ret;
 
 		cond_resched();
 find_page:
@@ -2113,8 +2307,15 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
 					ra, filp,
 					index, last_index - index);
 			page = find_get_page(mapping, index);
-			if (unlikely(page == NULL))
-				goto no_cached_page;
+			if (unlikely(page == NULL)) {
+				page = generic_file_buffered_read_no_cached_page(iocb, iter);
+				if (!page)
+					goto find_page;
+				if (IS_ERR(page)) {
+					error = PTR_ERR(page);
+					goto out;
+				}
+			}
 		}
 		if (PageReadahead(page)) {
 			page_cache_async_readahead(mapping,
@@ -2124,199 +2325,36 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
 		if (!PageUptodate(page)) {
 			if (iocb->ki_flags & IOCB_NOWAIT) {
 				put_page(page);
-				goto would_block;
-			}
-
-			/*
-			 * See comment in do_read_cache_page on why
-			 * wait_on_page_locked is used to avoid unnecessarily
-			 * serialisations and why it's safe.
-			 */
-			error = wait_on_page_locked_killable(page);
-			if (unlikely(error))
-				goto readpage_error;
-			if (PageUptodate(page))
-				goto page_ok;
-
-			if (inode->i_blkbits == PAGE_SHIFT ||
-					!mapping->a_ops->is_partially_uptodate)
-				goto page_not_up_to_date;
-			/* pipes can't handle partially uptodate pages */
-			if (unlikely(iov_iter_is_pipe(iter)))
-				goto page_not_up_to_date;
-			if (!trylock_page(page))
-				goto page_not_up_to_date;
-			/* Did it get truncated before we got the lock? */
-			if (!page->mapping)
-				goto page_not_up_to_date_locked;
-			if (!mapping->a_ops->is_partially_uptodate(page,
-							offset, iter->count))
-				goto page_not_up_to_date_locked;
-			unlock_page(page);
-		}
-page_ok:
-		/*
-		 * i_size must be checked after we know the page is Uptodate.
-		 *
-		 * Checking i_size after the check allows us to calculate
-		 * the correct value for "nr", which means the zero-filled
-		 * part of the page is not copied back to userspace (unless
-		 * another truncate extends the file - this is desired though).
-		 */
-
-		isize = i_size_read(inode);
-		end_index = (isize - 1) >> PAGE_SHIFT;
-		if (unlikely(!isize || index > end_index)) {
-			put_page(page);
-			goto out;
-		}
-
-		/* nr is the maximum number of bytes to copy from this page */
-		nr = PAGE_SIZE;
-		if (index == end_index) {
-			nr = ((isize - 1) & ~PAGE_MASK) + 1;
-			if (nr <= offset) {
-				put_page(page);
+				error = -EAGAIN;
 				goto out;
 			}
-		}
-		nr = nr - offset;
-
-		/* If users can be writing to this page using arbitrary
-		 * virtual addresses, take care about potential aliasing
-		 * before reading the page on the kernel side.
-		 */
-		if (mapping_writably_mapped(mapping))
-			flush_dcache_page(page);
-
-		/*
-		 * When a sequential read accesses a page several times,
-		 * only mark it as accessed the first time.
-		 */
-		if (prev_index != index || offset != prev_offset)
-			mark_page_accessed(page);
-		prev_index = index;
-
-		/*
-		 * Ok, we have the page, and it's up-to-date, so
-		 * now we can copy it to user space...
-		 */
-
-		ret = copy_page_to_iter(page, offset, nr, iter);
-		offset += ret;
-		index += offset >> PAGE_SHIFT;
-		offset &= ~PAGE_MASK;
-		prev_offset = offset;
 
-		put_page(page);
-		written += ret;
-		if (!iov_iter_count(iter))
-			goto out;
-		if (ret < nr) {
-			error = -EFAULT;
-			goto out;
-		}
-		continue;
-
-page_not_up_to_date:
-		/* Get exclusive access to the page ... */
-		error = lock_page_killable(page);
-		if (unlikely(error))
-			goto readpage_error;
-
-page_not_up_to_date_locked:
-		/* Did it get truncated before we got the lock? */
-		if (!page->mapping) {
-			unlock_page(page);
-			put_page(page);
-			continue;
-		}
-
-		/* Did somebody else fill it already? */
-		if (PageUptodate(page)) {
-			unlock_page(page);
-			goto page_ok;
-		}
-
-readpage:
-		/*
-		 * A previous I/O error may have been due to temporary
-		 * failures, eg. multipath errors.
-		 * PG_error will be set again if readpage fails.
-		 */
-		ClearPageError(page);
-		/* Start the actual read. The read will unlock the page. */
-		error = mapping->a_ops->readpage(filp, page);
-
-		if (unlikely(error)) {
-			if (error == AOP_TRUNCATED_PAGE) {
-				put_page(page);
-				error = 0;
+			page = generic_file_buffered_read_pagenotuptodate(filp,
+					iter, page, iocb->ki_pos, iter->count);
+			if (!page)
 				goto find_page;
+			if (IS_ERR(page)) {
+				error = PTR_ERR(page);
+				goto out;
 			}
-			goto readpage_error;
-		}
-
-		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);
-					goto find_page;
-				}
-				unlock_page(page);
-				shrink_readahead_size_eio(ra);
-				error = -EIO;
-				goto readpage_error;
-			}
-			unlock_page(page);
 		}
 
-		goto page_ok;
-
-readpage_error:
-		/* UHHUH! A synchronous read error occurred. Report it */
+		error = generic_file_buffered_read_page_ok(iocb, iter, page);
 		put_page(page);
-		goto out;
 
-no_cached_page:
-		/*
-		 * Ok, it wasn't cached, so we need to create a new
-		 * page..
-		 */
-		page = page_cache_alloc(mapping);
-		if (!page) {
-			error = -ENOMEM;
-			goto out;
-		}
-		error = add_to_page_cache_lru(page, mapping, index,
-				mapping_gfp_constraint(mapping, GFP_KERNEL));
 		if (error) {
-			put_page(page);
-			if (error == -EEXIST) {
+			if (error > 0)
 				error = 0;
-				goto find_page;
-			}
 			goto out;
 		}
-		goto readpage;
 	}
 
 would_block:
 	error = -EAGAIN;
 out:
-	ra->prev_pos = prev_index;
-	ra->prev_pos <<= PAGE_SHIFT;
-	ra->prev_pos |= prev_offset;
-
-	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
 	file_accessed(filp);
+	written += orig_count - iov_iter_count(iter);
+
 	return written ? written : error;
 }
 
-- 
2.27.0

[PATCH 1/2] fs: Break generic_file_buffered_read up into multiple functions

[Kent Overstreet]

This is prep work for changing generic_file_buffered_read() to use
find_get_pages_contig() to batch up all the pagecache lookups.

This patch should be functionally identical to the existing code and
changes as little as of the flow control as possible. More refactoring
could be done, this patch is intended to be relatively minimal.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
---
 mm/filemap.c | 418 ++++++++++++++++++++++++++++-----------------------
 1 file changed, 228 insertions(+), 190 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index c7723bbbbd..308bdd466f 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2110,6 +2110,210 @@ static void shrink_readahead_size_eio(struct file *filp,
 	ra->ra_pages /= 4;
 }
 
+static int generic_file_buffered_read_page_ok(struct kiocb *iocb,
+			struct iov_iter *iter,
+			struct page *page)
+{
+	struct address_space *mapping = iocb->ki_filp->f_mapping;
+	struct inode *inode = mapping->host;
+	struct file_ra_state *ra = &iocb->ki_filp->f_ra;
+	unsigned offset = iocb->ki_pos & ~PAGE_MASK;
+	unsigned bytes, copied;
+	loff_t isize, end_offset;
+
+	BUG_ON(iocb->ki_pos >> PAGE_SHIFT != page->index);
+
+	/*
+	 * i_size must be checked after we know the page is Uptodate.
+	 *
+	 * Checking i_size after the check allows us to calculate
+	 * the correct value for "bytes", which means the zero-filled
+	 * part of the page is not copied back to userspace (unless
+	 * another truncate extends the file - this is desired though).
+	 */
+
+	isize = i_size_read(inode);
+	if (unlikely(iocb->ki_pos >= isize))
+		return 1;
+
+	end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
+
+	bytes = min_t(loff_t, end_offset - iocb->ki_pos, PAGE_SIZE - offset);
+
+	/* If users can be writing to this page using arbitrary
+	 * virtual addresses, take care about potential aliasing
+	 * before reading the page on the kernel side.
+	 */
+	if (mapping_writably_mapped(mapping))
+		flush_dcache_page(page);
+
+	/*
+	 * Ok, we have the page, and it's up-to-date, so
+	 * now we can copy it to user space...
+	 */
+
+	copied = copy_page_to_iter(page, offset, bytes, iter);
+
+	iocb->ki_pos += copied;
+
+	/*
+	 * When a sequential read accesses a page several times,
+	 * only mark it as accessed the first time.
+	 */
+	if (iocb->ki_pos >> PAGE_SHIFT != ra->prev_pos >> PAGE_SHIFT)
+		mark_page_accessed(page);
+
+	ra->prev_pos = iocb->ki_pos;
+
+	if (copied < bytes)
+		return -EFAULT;
+
+	return !iov_iter_count(iter) || iocb->ki_pos == isize;
+}
+
+static struct page *
+generic_file_buffered_read_readpage(struct file *filp,
+				    struct address_space *mapping,
+				    struct page *page)
+{
+	struct file_ra_state *ra = &filp->f_ra;
+	int error;
+
+	/*
+	 * A previous I/O error may have been due to temporary
+	 * failures, eg. multipath errors.
+	 * PG_error will be set again if readpage fails.
+	 */
+	ClearPageError(page);
+	/* Start the actual read. The read will unlock the page. */
+	error = mapping->a_ops->readpage(filp, page);
+
+	if (unlikely(error)) {
+		put_page(page);
+		return error != AOP_TRUNCATED_PAGE ? ERR_PTR(error) : NULL;
+	}
+
+	if (!PageUptodate(page)) {
+		error = lock_page_killable(page);
+		if (unlikely(error)) {
+			put_page(page);
+			return ERR_PTR(error);
+		}
+		if (!PageUptodate(page)) {
+			if (page->mapping == NULL) {
+				/*
+				 * invalidate_mapping_pages got it
+				 */
+				unlock_page(page);
+				put_page(page);
+				return NULL;
+			}
+			unlock_page(page);
+			shrink_readahead_size_eio(filp, ra);
+			put_page(page);
+			return ERR_PTR(-EIO);
+		}
+		unlock_page(page);
+	}
+
+	return page;
+}
+
+static struct page *
+generic_file_buffered_read_pagenotuptodate(struct file *filp,
+					   struct iov_iter *iter,
+					   struct page *page,
+					   loff_t pos, loff_t count)
+{
+	struct address_space *mapping = filp->f_mapping;
+	struct inode *inode = mapping->host;
+	int error;
+
+	/*
+	 * See comment in do_read_cache_page on why
+	 * wait_on_page_locked is used to avoid unnecessarily
+	 * serialisations and why it's safe.
+	 */
+	error = wait_on_page_locked_killable(page);
+	if (unlikely(error)) {
+		put_page(page);
+		return ERR_PTR(error);
+	}
+
+	if (PageUptodate(page))
+		return page;
+
+	if (inode->i_blkbits == PAGE_SHIFT ||
+	    !mapping->a_ops->is_partially_uptodate)
+		goto page_not_up_to_date;
+	/* pipes can't handle partially uptodate pages */
+	if (unlikely(iter->type & ITER_PIPE))
+		goto page_not_up_to_date;
+	if (!trylock_page(page))
+		goto page_not_up_to_date;
+	/* Did it get truncated before we got the lock? */
+	if (!page->mapping)
+		goto page_not_up_to_date_locked;
+
+	if (!mapping->a_ops->is_partially_uptodate(page,
+				pos & ~PAGE_MASK, count))
+		goto page_not_up_to_date_locked;
+	unlock_page(page);
+	return page;
+
+page_not_up_to_date:
+	/* Get exclusive access to the page ... */
+	error = lock_page_killable(page);
+	if (unlikely(error)) {
+		put_page(page);
+		return ERR_PTR(error);
+	}
+
+page_not_up_to_date_locked:
+	/* Did it get truncated before we got the lock? */
+	if (!page->mapping) {
+		unlock_page(page);
+		put_page(page);
+		return NULL;
+	}
+
+	/* Did somebody else fill it already? */
+	if (PageUptodate(page)) {
+		unlock_page(page);
+		return page;
+	}
+
+	return generic_file_buffered_read_readpage(filp, mapping, page);
+}
+
+static struct page *
+generic_file_buffered_read_no_cached_page(struct kiocb *iocb,
+					  struct iov_iter *iter)
+{
+	struct file *filp = iocb->ki_filp;
+	struct address_space *mapping = filp->f_mapping;
+	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
+	struct page *page;
+	int error;
+
+	/*
+	 * Ok, it wasn't cached, so we need to create a new
+	 * page..
+	 */
+	page = page_cache_alloc(mapping);
+	if (!page)
+		return ERR_PTR(-ENOMEM);
+
+	error = add_to_page_cache_lru(page, mapping, index,
+				      mapping_gfp_constraint(mapping, GFP_KERNEL));
+	if (error) {
+		put_page(page);
+		return error != -EEXIST ? ERR_PTR(error) : NULL;
+	}
+
+	return generic_file_buffered_read_readpage(filp, mapping, page);
+}
+
 /**
  * generic_file_buffered_read - generic file read routine
  * @iocb:	the iocb to read
@@ -2129,29 +2333,19 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
 	struct address_space *mapping = filp->f_mapping;
 	struct inode *inode = mapping->host;
 	struct file_ra_state *ra = &filp->f_ra;
-	loff_t *ppos = &iocb->ki_pos;
-	pgoff_t index;
+	size_t orig_count = iov_iter_count(iter);
 	pgoff_t last_index;
-	pgoff_t prev_index;
-	unsigned long offset;      /* offset into pagecache page */
-	unsigned int prev_offset;
 	int error = 0;
 
-	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
+	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
 		return 0;
 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
 
-	index = *ppos >> PAGE_SHIFT;
-	prev_index = ra->prev_pos >> PAGE_SHIFT;
-	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
-	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
-	offset = *ppos & ~PAGE_MASK;
+	last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
 
 	for (;;) {
+		pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
 		struct page *page;
-		pgoff_t end_index;
-		loff_t isize;
-		unsigned long nr, ret;
 
 		cond_resched();
 find_page:
@@ -2168,8 +2362,15 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
 					ra, filp,
 					index, last_index - index);
 			page = find_get_page(mapping, index);
-			if (unlikely(page == NULL))
-				goto no_cached_page;
+			if (unlikely(page == NULL)) {
+				page = generic_file_buffered_read_no_cached_page(iocb, iter);
+				if (!page)
+					goto find_page;
+				if (IS_ERR(page)) {
+					error = PTR_ERR(page);
+					goto out;
+				}
+			}
 		}
 		if (PageReadahead(page)) {
 			page_cache_async_readahead(mapping,
@@ -2179,199 +2380,36 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
 		if (!PageUptodate(page)) {
 			if (iocb->ki_flags & IOCB_NOWAIT) {
 				put_page(page);
-				goto would_block;
-			}
-
-			/*
-			 * See comment in do_read_cache_page on why
-			 * wait_on_page_locked is used to avoid unnecessarily
-			 * serialisations and why it's safe.
-			 */
-			error = wait_on_page_locked_killable(page);
-			if (unlikely(error))
-				goto readpage_error;
-			if (PageUptodate(page))
-				goto page_ok;
-
-			if (inode->i_blkbits == PAGE_SHIFT ||
-					!mapping->a_ops->is_partially_uptodate)
-				goto page_not_up_to_date;
-			/* pipes can't handle partially uptodate pages */
-			if (unlikely(iter->type & ITER_PIPE))
-				goto page_not_up_to_date;
-			if (!trylock_page(page))
-				goto page_not_up_to_date;
-			/* Did it get truncated before we got the lock? */
-			if (!page->mapping)
-				goto page_not_up_to_date_locked;
-			if (!mapping->a_ops->is_partially_uptodate(page,
-							offset, iter->count))
-				goto page_not_up_to_date_locked;
-			unlock_page(page);
-		}
-page_ok:
-		/*
-		 * i_size must be checked after we know the page is Uptodate.
-		 *
-		 * Checking i_size after the check allows us to calculate
-		 * the correct value for "nr", which means the zero-filled
-		 * part of the page is not copied back to userspace (unless
-		 * another truncate extends the file - this is desired though).
-		 */
-
-		isize = i_size_read(inode);
-		end_index = (isize - 1) >> PAGE_SHIFT;
-		if (unlikely(!isize || index > end_index)) {
-			put_page(page);
-			goto out;
-		}
-
-		/* nr is the maximum number of bytes to copy from this page */
-		nr = PAGE_SIZE;
-		if (index == end_index) {
-			nr = ((isize - 1) & ~PAGE_MASK) + 1;
-			if (nr <= offset) {
-				put_page(page);
+				error = -EAGAIN;
 				goto out;
 			}
-		}
-		nr = nr - offset;
-
-		/* If users can be writing to this page using arbitrary
-		 * virtual addresses, take care about potential aliasing
-		 * before reading the page on the kernel side.
-		 */
-		if (mapping_writably_mapped(mapping))
-			flush_dcache_page(page);
-
-		/*
-		 * When a sequential read accesses a page several times,
-		 * only mark it as accessed the first time.
-		 */
-		if (prev_index != index || offset != prev_offset)
-			mark_page_accessed(page);
-		prev_index = index;
-
-		/*
-		 * Ok, we have the page, and it's up-to-date, so
-		 * now we can copy it to user space...
-		 */
-
-		ret = copy_page_to_iter(page, offset, nr, iter);
-		offset += ret;
-		index += offset >> PAGE_SHIFT;
-		offset &= ~PAGE_MASK;
-		prev_offset = offset;
 
-		put_page(page);
-		written += ret;
-		if (!iov_iter_count(iter))
-			goto out;
-		if (ret < nr) {
-			error = -EFAULT;
-			goto out;
-		}
-		continue;
-
-page_not_up_to_date:
-		/* Get exclusive access to the page ... */
-		error = lock_page_killable(page);
-		if (unlikely(error))
-			goto readpage_error;
-
-page_not_up_to_date_locked:
-		/* Did it get truncated before we got the lock? */
-		if (!page->mapping) {
-			unlock_page(page);
-			put_page(page);
-			continue;
-		}
-
-		/* Did somebody else fill it already? */
-		if (PageUptodate(page)) {
-			unlock_page(page);
-			goto page_ok;
-		}
-
-readpage:
-		/*
-		 * A previous I/O error may have been due to temporary
-		 * failures, eg. multipath errors.
-		 * PG_error will be set again if readpage fails.
-		 */
-		ClearPageError(page);
-		/* Start the actual read. The read will unlock the page. */
-		error = mapping->a_ops->readpage(filp, page);
-
-		if (unlikely(error)) {
-			if (error == AOP_TRUNCATED_PAGE) {
-				put_page(page);
-				error = 0;
+			page = generic_file_buffered_read_pagenotuptodate(filp,
+					iter, page, iocb->ki_pos, iter->count);
+			if (!page)
 				goto find_page;
+			if (IS_ERR(page)) {
+				error = PTR_ERR(page);
+				goto out;
 			}
-			goto readpage_error;
-		}
-
-		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);
-					goto find_page;
-				}
-				unlock_page(page);
-				shrink_readahead_size_eio(filp, ra);
-				error = -EIO;
-				goto readpage_error;
-			}
-			unlock_page(page);
 		}
 
-		goto page_ok;
-
-readpage_error:
-		/* UHHUH! A synchronous read error occurred. Report it */
+		error = generic_file_buffered_read_page_ok(iocb, iter, page);
 		put_page(page);
-		goto out;
 
-no_cached_page:
-		/*
-		 * Ok, it wasn't cached, so we need to create a new
-		 * page..
-		 */
-		page = page_cache_alloc(mapping);
-		if (!page) {
-			error = -ENOMEM;
-			goto out;
-		}
-		error = add_to_page_cache_lru(page, mapping, index,
-				mapping_gfp_constraint(mapping, GFP_KERNEL));
 		if (error) {
-			put_page(page);
-			if (error == -EEXIST) {
+			if (error > 0)
 				error = 0;
-				goto find_page;
-			}
 			goto out;
 		}
-		goto readpage;
 	}
 
 would_block:
 	error = -EAGAIN;
 out:
-	ra->prev_pos = prev_index;
-	ra->prev_pos <<= PAGE_SHIFT;
-	ra->prev_pos |= prev_offset;
-
-	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
 	file_accessed(filp);
+	written += orig_count - iov_iter_count(iter);
+
 	return written ? written : error;
 }
 
-- 
2.18.0