nfsd4: typo logical vs bitwise negate for want_mask
[linux-btrfs-devel.git] / fs / mpage.c
blobfdfae9fa98cda52f26b730ab6786db084124d279
1 /*
2 * fs/mpage.c
4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains functions related to preparing and submitting BIOs which contain
7 * multiple pagecache pages.
9 * 15May2002 Andrew Morton
10 * Initial version
11 * 27Jun2002 axboe@suse.de
12 * use bio_add_page() to build bio's just the right size
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/mm.h>
18 #include <linux/kdev_t.h>
19 #include <linux/gfp.h>
20 #include <linux/bio.h>
21 #include <linux/fs.h>
22 #include <linux/buffer_head.h>
23 #include <linux/blkdev.h>
24 #include <linux/highmem.h>
25 #include <linux/prefetch.h>
26 #include <linux/mpage.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/cleancache.h>
33 * I/O completion handler for multipage BIOs.
35 * The mpage code never puts partial pages into a BIO (except for end-of-file).
36 * If a page does not map to a contiguous run of blocks then it simply falls
37 * back to block_read_full_page().
39 * Why is this? If a page's completion depends on a number of different BIOs
40 * which can complete in any order (or at the same time) then determining the
41 * status of that page is hard. See end_buffer_async_read() for the details.
42 * There is no point in duplicating all that complexity.
44 static void mpage_end_io(struct bio *bio, int err)
46 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
47 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
49 do {
50 struct page *page = bvec->bv_page;
52 if (--bvec >= bio->bi_io_vec)
53 prefetchw(&bvec->bv_page->flags);
54 if (bio_data_dir(bio) == READ) {
55 if (uptodate) {
56 SetPageUptodate(page);
57 } else {
58 ClearPageUptodate(page);
59 SetPageError(page);
61 unlock_page(page);
62 } else { /* bio_data_dir(bio) == WRITE */
63 if (!uptodate) {
64 SetPageError(page);
65 if (page->mapping)
66 set_bit(AS_EIO, &page->mapping->flags);
68 end_page_writeback(page);
70 } while (bvec >= bio->bi_io_vec);
71 bio_put(bio);
74 static struct bio *mpage_bio_submit(int rw, struct bio *bio)
76 bio->bi_end_io = mpage_end_io;
77 submit_bio(rw, bio);
78 return NULL;
81 static struct bio *
82 mpage_alloc(struct block_device *bdev,
83 sector_t first_sector, int nr_vecs,
84 gfp_t gfp_flags)
86 struct bio *bio;
88 bio = bio_alloc(gfp_flags, nr_vecs);
90 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
91 while (!bio && (nr_vecs /= 2))
92 bio = bio_alloc(gfp_flags, nr_vecs);
95 if (bio) {
96 bio->bi_bdev = bdev;
97 bio->bi_sector = first_sector;
99 return bio;
103 * support function for mpage_readpages. The fs supplied get_block might
104 * return an up to date buffer. This is used to map that buffer into
105 * the page, which allows readpage to avoid triggering a duplicate call
106 * to get_block.
108 * The idea is to avoid adding buffers to pages that don't already have
109 * them. So when the buffer is up to date and the page size == block size,
110 * this marks the page up to date instead of adding new buffers.
112 static void
113 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
115 struct inode *inode = page->mapping->host;
116 struct buffer_head *page_bh, *head;
117 int block = 0;
119 if (!page_has_buffers(page)) {
121 * don't make any buffers if there is only one buffer on
122 * the page and the page just needs to be set up to date
124 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
125 buffer_uptodate(bh)) {
126 SetPageUptodate(page);
127 return;
129 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
131 head = page_buffers(page);
132 page_bh = head;
133 do {
134 if (block == page_block) {
135 page_bh->b_state = bh->b_state;
136 page_bh->b_bdev = bh->b_bdev;
137 page_bh->b_blocknr = bh->b_blocknr;
138 break;
140 page_bh = page_bh->b_this_page;
141 block++;
142 } while (page_bh != head);
146 * This is the worker routine which does all the work of mapping the disk
147 * blocks and constructs largest possible bios, submits them for IO if the
148 * blocks are not contiguous on the disk.
150 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
151 * represent the validity of its disk mapping and to decide when to do the next
152 * get_block() call.
154 static struct bio *
155 do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
156 sector_t *last_block_in_bio, struct buffer_head *map_bh,
157 unsigned long *first_logical_block, get_block_t get_block)
159 struct inode *inode = page->mapping->host;
160 const unsigned blkbits = inode->i_blkbits;
161 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
162 const unsigned blocksize = 1 << blkbits;
163 sector_t block_in_file;
164 sector_t last_block;
165 sector_t last_block_in_file;
166 sector_t blocks[MAX_BUF_PER_PAGE];
167 unsigned page_block;
168 unsigned first_hole = blocks_per_page;
169 struct block_device *bdev = NULL;
170 int length;
171 int fully_mapped = 1;
172 unsigned nblocks;
173 unsigned relative_block;
175 if (page_has_buffers(page))
176 goto confused;
178 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
179 last_block = block_in_file + nr_pages * blocks_per_page;
180 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
181 if (last_block > last_block_in_file)
182 last_block = last_block_in_file;
183 page_block = 0;
186 * Map blocks using the result from the previous get_blocks call first.
188 nblocks = map_bh->b_size >> blkbits;
189 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
190 block_in_file < (*first_logical_block + nblocks)) {
191 unsigned map_offset = block_in_file - *first_logical_block;
192 unsigned last = nblocks - map_offset;
194 for (relative_block = 0; ; relative_block++) {
195 if (relative_block == last) {
196 clear_buffer_mapped(map_bh);
197 break;
199 if (page_block == blocks_per_page)
200 break;
201 blocks[page_block] = map_bh->b_blocknr + map_offset +
202 relative_block;
203 page_block++;
204 block_in_file++;
206 bdev = map_bh->b_bdev;
210 * Then do more get_blocks calls until we are done with this page.
212 map_bh->b_page = page;
213 while (page_block < blocks_per_page) {
214 map_bh->b_state = 0;
215 map_bh->b_size = 0;
217 if (block_in_file < last_block) {
218 map_bh->b_size = (last_block-block_in_file) << blkbits;
219 if (get_block(inode, block_in_file, map_bh, 0))
220 goto confused;
221 *first_logical_block = block_in_file;
224 if (!buffer_mapped(map_bh)) {
225 fully_mapped = 0;
226 if (first_hole == blocks_per_page)
227 first_hole = page_block;
228 page_block++;
229 block_in_file++;
230 continue;
233 /* some filesystems will copy data into the page during
234 * the get_block call, in which case we don't want to
235 * read it again. map_buffer_to_page copies the data
236 * we just collected from get_block into the page's buffers
237 * so readpage doesn't have to repeat the get_block call
239 if (buffer_uptodate(map_bh)) {
240 map_buffer_to_page(page, map_bh, page_block);
241 goto confused;
244 if (first_hole != blocks_per_page)
245 goto confused; /* hole -> non-hole */
247 /* Contiguous blocks? */
248 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
249 goto confused;
250 nblocks = map_bh->b_size >> blkbits;
251 for (relative_block = 0; ; relative_block++) {
252 if (relative_block == nblocks) {
253 clear_buffer_mapped(map_bh);
254 break;
255 } else if (page_block == blocks_per_page)
256 break;
257 blocks[page_block] = map_bh->b_blocknr+relative_block;
258 page_block++;
259 block_in_file++;
261 bdev = map_bh->b_bdev;
264 if (first_hole != blocks_per_page) {
265 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
266 if (first_hole == 0) {
267 SetPageUptodate(page);
268 unlock_page(page);
269 goto out;
271 } else if (fully_mapped) {
272 SetPageMappedToDisk(page);
275 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
276 cleancache_get_page(page) == 0) {
277 SetPageUptodate(page);
278 goto confused;
282 * This page will go to BIO. Do we need to send this BIO off first?
284 if (bio && (*last_block_in_bio != blocks[0] - 1))
285 bio = mpage_bio_submit(READ, bio);
287 alloc_new:
288 if (bio == NULL) {
289 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
290 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
291 GFP_KERNEL);
292 if (bio == NULL)
293 goto confused;
296 length = first_hole << blkbits;
297 if (bio_add_page(bio, page, length, 0) < length) {
298 bio = mpage_bio_submit(READ, bio);
299 goto alloc_new;
302 relative_block = block_in_file - *first_logical_block;
303 nblocks = map_bh->b_size >> blkbits;
304 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
305 (first_hole != blocks_per_page))
306 bio = mpage_bio_submit(READ, bio);
307 else
308 *last_block_in_bio = blocks[blocks_per_page - 1];
309 out:
310 return bio;
312 confused:
313 if (bio)
314 bio = mpage_bio_submit(READ, bio);
315 if (!PageUptodate(page))
316 block_read_full_page(page, get_block);
317 else
318 unlock_page(page);
319 goto out;
323 * mpage_readpages - populate an address space with some pages & start reads against them
324 * @mapping: the address_space
325 * @pages: The address of a list_head which contains the target pages. These
326 * pages have their ->index populated and are otherwise uninitialised.
327 * The page at @pages->prev has the lowest file offset, and reads should be
328 * issued in @pages->prev to @pages->next order.
329 * @nr_pages: The number of pages at *@pages
330 * @get_block: The filesystem's block mapper function.
332 * This function walks the pages and the blocks within each page, building and
333 * emitting large BIOs.
335 * If anything unusual happens, such as:
337 * - encountering a page which has buffers
338 * - encountering a page which has a non-hole after a hole
339 * - encountering a page with non-contiguous blocks
341 * then this code just gives up and calls the buffer_head-based read function.
342 * It does handle a page which has holes at the end - that is a common case:
343 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
345 * BH_Boundary explanation:
347 * There is a problem. The mpage read code assembles several pages, gets all
348 * their disk mappings, and then submits them all. That's fine, but obtaining
349 * the disk mappings may require I/O. Reads of indirect blocks, for example.
351 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
352 * submitted in the following order:
353 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
355 * because the indirect block has to be read to get the mappings of blocks
356 * 13,14,15,16. Obviously, this impacts performance.
358 * So what we do it to allow the filesystem's get_block() function to set
359 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
360 * after this one will require I/O against a block which is probably close to
361 * this one. So you should push what I/O you have currently accumulated.
363 * This all causes the disk requests to be issued in the correct order.
366 mpage_readpages(struct address_space *mapping, struct list_head *pages,
367 unsigned nr_pages, get_block_t get_block)
369 struct bio *bio = NULL;
370 unsigned page_idx;
371 sector_t last_block_in_bio = 0;
372 struct buffer_head map_bh;
373 unsigned long first_logical_block = 0;
374 struct blk_plug plug;
376 blk_start_plug(&plug);
378 map_bh.b_state = 0;
379 map_bh.b_size = 0;
380 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
381 struct page *page = list_entry(pages->prev, struct page, lru);
383 prefetchw(&page->flags);
384 list_del(&page->lru);
385 if (!add_to_page_cache_lru(page, mapping,
386 page->index, GFP_KERNEL)) {
387 bio = do_mpage_readpage(bio, page,
388 nr_pages - page_idx,
389 &last_block_in_bio, &map_bh,
390 &first_logical_block,
391 get_block);
393 page_cache_release(page);
395 BUG_ON(!list_empty(pages));
396 if (bio)
397 mpage_bio_submit(READ, bio);
398 blk_finish_plug(&plug);
399 return 0;
401 EXPORT_SYMBOL(mpage_readpages);
404 * This isn't called much at all
406 int mpage_readpage(struct page *page, get_block_t get_block)
408 struct bio *bio = NULL;
409 sector_t last_block_in_bio = 0;
410 struct buffer_head map_bh;
411 unsigned long first_logical_block = 0;
413 map_bh.b_state = 0;
414 map_bh.b_size = 0;
415 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
416 &map_bh, &first_logical_block, get_block);
417 if (bio)
418 mpage_bio_submit(READ, bio);
419 return 0;
421 EXPORT_SYMBOL(mpage_readpage);
424 * Writing is not so simple.
426 * If the page has buffers then they will be used for obtaining the disk
427 * mapping. We only support pages which are fully mapped-and-dirty, with a
428 * special case for pages which are unmapped at the end: end-of-file.
430 * If the page has no buffers (preferred) then the page is mapped here.
432 * If all blocks are found to be contiguous then the page can go into the
433 * BIO. Otherwise fall back to the mapping's writepage().
435 * FIXME: This code wants an estimate of how many pages are still to be
436 * written, so it can intelligently allocate a suitably-sized BIO. For now,
437 * just allocate full-size (16-page) BIOs.
440 struct mpage_data {
441 struct bio *bio;
442 sector_t last_block_in_bio;
443 get_block_t *get_block;
444 unsigned use_writepage;
447 static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
448 void *data)
450 struct mpage_data *mpd = data;
451 struct bio *bio = mpd->bio;
452 struct address_space *mapping = page->mapping;
453 struct inode *inode = page->mapping->host;
454 const unsigned blkbits = inode->i_blkbits;
455 unsigned long end_index;
456 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
457 sector_t last_block;
458 sector_t block_in_file;
459 sector_t blocks[MAX_BUF_PER_PAGE];
460 unsigned page_block;
461 unsigned first_unmapped = blocks_per_page;
462 struct block_device *bdev = NULL;
463 int boundary = 0;
464 sector_t boundary_block = 0;
465 struct block_device *boundary_bdev = NULL;
466 int length;
467 struct buffer_head map_bh;
468 loff_t i_size = i_size_read(inode);
469 int ret = 0;
471 if (page_has_buffers(page)) {
472 struct buffer_head *head = page_buffers(page);
473 struct buffer_head *bh = head;
475 /* If they're all mapped and dirty, do it */
476 page_block = 0;
477 do {
478 BUG_ON(buffer_locked(bh));
479 if (!buffer_mapped(bh)) {
481 * unmapped dirty buffers are created by
482 * __set_page_dirty_buffers -> mmapped data
484 if (buffer_dirty(bh))
485 goto confused;
486 if (first_unmapped == blocks_per_page)
487 first_unmapped = page_block;
488 continue;
491 if (first_unmapped != blocks_per_page)
492 goto confused; /* hole -> non-hole */
494 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
495 goto confused;
496 if (page_block) {
497 if (bh->b_blocknr != blocks[page_block-1] + 1)
498 goto confused;
500 blocks[page_block++] = bh->b_blocknr;
501 boundary = buffer_boundary(bh);
502 if (boundary) {
503 boundary_block = bh->b_blocknr;
504 boundary_bdev = bh->b_bdev;
506 bdev = bh->b_bdev;
507 } while ((bh = bh->b_this_page) != head);
509 if (first_unmapped)
510 goto page_is_mapped;
513 * Page has buffers, but they are all unmapped. The page was
514 * created by pagein or read over a hole which was handled by
515 * block_read_full_page(). If this address_space is also
516 * using mpage_readpages then this can rarely happen.
518 goto confused;
522 * The page has no buffers: map it to disk
524 BUG_ON(!PageUptodate(page));
525 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
526 last_block = (i_size - 1) >> blkbits;
527 map_bh.b_page = page;
528 for (page_block = 0; page_block < blocks_per_page; ) {
530 map_bh.b_state = 0;
531 map_bh.b_size = 1 << blkbits;
532 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
533 goto confused;
534 if (buffer_new(&map_bh))
535 unmap_underlying_metadata(map_bh.b_bdev,
536 map_bh.b_blocknr);
537 if (buffer_boundary(&map_bh)) {
538 boundary_block = map_bh.b_blocknr;
539 boundary_bdev = map_bh.b_bdev;
541 if (page_block) {
542 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
543 goto confused;
545 blocks[page_block++] = map_bh.b_blocknr;
546 boundary = buffer_boundary(&map_bh);
547 bdev = map_bh.b_bdev;
548 if (block_in_file == last_block)
549 break;
550 block_in_file++;
552 BUG_ON(page_block == 0);
554 first_unmapped = page_block;
556 page_is_mapped:
557 end_index = i_size >> PAGE_CACHE_SHIFT;
558 if (page->index >= end_index) {
560 * The page straddles i_size. It must be zeroed out on each
561 * and every writepage invocation because it may be mmapped.
562 * "A file is mapped in multiples of the page size. For a file
563 * that is not a multiple of the page size, the remaining memory
564 * is zeroed when mapped, and writes to that region are not
565 * written out to the file."
567 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
569 if (page->index > end_index || !offset)
570 goto confused;
571 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
575 * This page will go to BIO. Do we need to send this BIO off first?
577 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
578 bio = mpage_bio_submit(WRITE, bio);
580 alloc_new:
581 if (bio == NULL) {
582 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
583 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
584 if (bio == NULL)
585 goto confused;
589 * Must try to add the page before marking the buffer clean or
590 * the confused fail path above (OOM) will be very confused when
591 * it finds all bh marked clean (i.e. it will not write anything)
593 length = first_unmapped << blkbits;
594 if (bio_add_page(bio, page, length, 0) < length) {
595 bio = mpage_bio_submit(WRITE, bio);
596 goto alloc_new;
600 * OK, we have our BIO, so we can now mark the buffers clean. Make
601 * sure to only clean buffers which we know we'll be writing.
603 if (page_has_buffers(page)) {
604 struct buffer_head *head = page_buffers(page);
605 struct buffer_head *bh = head;
606 unsigned buffer_counter = 0;
608 do {
609 if (buffer_counter++ == first_unmapped)
610 break;
611 clear_buffer_dirty(bh);
612 bh = bh->b_this_page;
613 } while (bh != head);
616 * we cannot drop the bh if the page is not uptodate
617 * or a concurrent readpage would fail to serialize with the bh
618 * and it would read from disk before we reach the platter.
620 if (buffer_heads_over_limit && PageUptodate(page))
621 try_to_free_buffers(page);
624 BUG_ON(PageWriteback(page));
625 set_page_writeback(page);
626 unlock_page(page);
627 if (boundary || (first_unmapped != blocks_per_page)) {
628 bio = mpage_bio_submit(WRITE, bio);
629 if (boundary_block) {
630 write_boundary_block(boundary_bdev,
631 boundary_block, 1 << blkbits);
633 } else {
634 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
636 goto out;
638 confused:
639 if (bio)
640 bio = mpage_bio_submit(WRITE, bio);
642 if (mpd->use_writepage) {
643 ret = mapping->a_ops->writepage(page, wbc);
644 } else {
645 ret = -EAGAIN;
646 goto out;
649 * The caller has a ref on the inode, so *mapping is stable
651 mapping_set_error(mapping, ret);
652 out:
653 mpd->bio = bio;
654 return ret;
658 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
659 * @mapping: address space structure to write
660 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
661 * @get_block: the filesystem's block mapper function.
662 * If this is NULL then use a_ops->writepage. Otherwise, go
663 * direct-to-BIO.
665 * This is a library function, which implements the writepages()
666 * address_space_operation.
668 * If a page is already under I/O, generic_writepages() skips it, even
669 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
670 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
671 * and msync() need to guarantee that all the data which was dirty at the time
672 * the call was made get new I/O started against them. If wbc->sync_mode is
673 * WB_SYNC_ALL then we were called for data integrity and we must wait for
674 * existing IO to complete.
677 mpage_writepages(struct address_space *mapping,
678 struct writeback_control *wbc, get_block_t get_block)
680 struct blk_plug plug;
681 int ret;
683 blk_start_plug(&plug);
685 if (!get_block)
686 ret = generic_writepages(mapping, wbc);
687 else {
688 struct mpage_data mpd = {
689 .bio = NULL,
690 .last_block_in_bio = 0,
691 .get_block = get_block,
692 .use_writepage = 1,
695 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
696 if (mpd.bio)
697 mpage_bio_submit(WRITE, mpd.bio);
699 blk_finish_plug(&plug);
700 return ret;
702 EXPORT_SYMBOL(mpage_writepages);
704 int mpage_writepage(struct page *page, get_block_t get_block,
705 struct writeback_control *wbc)
707 struct mpage_data mpd = {
708 .bio = NULL,
709 .last_block_in_bio = 0,
710 .get_block = get_block,
711 .use_writepage = 0,
713 int ret = __mpage_writepage(page, wbc, &mpd);
714 if (mpd.bio)
715 mpage_bio_submit(WRITE, mpd.bio);
716 return ret;
718 EXPORT_SYMBOL(mpage_writepage);