• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/mpage.c
4  *
5  * Copyright (C) 2002, Linus Torvalds.
6  *
7  * Contains functions related to preparing and submitting BIOs which contain
8  * multiple pagecache pages.
9  *
10  * 15May2002	Andrew Morton
11  *		Initial version
12  * 27Jun2002	axboe@suse.de
13  *		use bio_add_page() to build bio's just the right size
14  */
15 
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/mm.h>
19 #include <linux/kdev_t.h>
20 #include <linux/gfp.h>
21 #include <linux/bio.h>
22 #include <linux/fs.h>
23 #include <linux/buffer_head.h>
24 #include <linux/blkdev.h>
25 #include <linux/highmem.h>
26 #include <linux/prefetch.h>
27 #include <linux/mpage.h>
28 #include <linux/mm_inline.h>
29 #include <linux/writeback.h>
30 #include <linux/backing-dev.h>
31 #include <linux/pagevec.h>
32 #include <linux/cleancache.h>
33 #include "internal.h"
34 
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/android_fs.h>
37 
38 EXPORT_TRACEPOINT_SYMBOL(android_fs_datawrite_start);
39 EXPORT_TRACEPOINT_SYMBOL(android_fs_datawrite_end);
40 EXPORT_TRACEPOINT_SYMBOL(android_fs_dataread_start);
41 EXPORT_TRACEPOINT_SYMBOL(android_fs_dataread_end);
42 
43 /*
44  * I/O completion handler for multipage BIOs.
45  *
46  * The mpage code never puts partial pages into a BIO (except for end-of-file).
47  * If a page does not map to a contiguous run of blocks then it simply falls
48  * back to block_read_full_page().
49  *
50  * Why is this?  If a page's completion depends on a number of different BIOs
51  * which can complete in any order (or at the same time) then determining the
52  * status of that page is hard.  See end_buffer_async_read() for the details.
53  * There is no point in duplicating all that complexity.
54  */
mpage_end_io(struct bio * bio)55 static void mpage_end_io(struct bio *bio)
56 {
57 	struct bio_vec *bv;
58 	struct bvec_iter_all iter_all;
59 
60 	if (trace_android_fs_dataread_end_enabled() &&
61 	    (bio_data_dir(bio) == READ)) {
62 		struct page *first_page = bio->bi_io_vec[0].bv_page;
63 
64 		if (first_page != NULL)
65 			trace_android_fs_dataread_end(first_page->mapping->host,
66 						      page_offset(first_page),
67 						      bio->bi_iter.bi_size);
68 	}
69 
70 	bio_for_each_segment_all(bv, bio, iter_all) {
71 		struct page *page = bv->bv_page;
72 		page_endio(page, bio_op(bio),
73 			   blk_status_to_errno(bio->bi_status));
74 	}
75 
76 	bio_put(bio);
77 }
78 
mpage_bio_submit(int op,int op_flags,struct bio * bio)79 static struct bio *mpage_bio_submit(int op, int op_flags, struct bio *bio)
80 {
81 	if (trace_android_fs_dataread_start_enabled() && (op == REQ_OP_READ)) {
82 		struct page *first_page = bio->bi_io_vec[0].bv_page;
83 
84 		if (first_page != NULL) {
85 			char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
86 
87 			path = android_fstrace_get_pathname(pathbuf,
88 						    MAX_TRACE_PATHBUF_LEN,
89 						    first_page->mapping->host);
90 			trace_android_fs_dataread_start(
91 				first_page->mapping->host,
92 				page_offset(first_page),
93 				bio->bi_iter.bi_size,
94 				current->pid,
95 				path,
96 				current->comm);
97 		}
98 	}
99 	bio->bi_end_io = mpage_end_io;
100 	bio_set_op_attrs(bio, op, op_flags);
101 	guard_bio_eod(bio);
102 	submit_bio(bio);
103 	return NULL;
104 }
105 
106 static struct bio *
mpage_alloc(struct block_device * bdev,sector_t first_sector,int nr_vecs,gfp_t gfp_flags)107 mpage_alloc(struct block_device *bdev,
108 		sector_t first_sector, int nr_vecs,
109 		gfp_t gfp_flags)
110 {
111 	struct bio *bio;
112 
113 	/* Restrict the given (page cache) mask for slab allocations */
114 	gfp_flags &= GFP_KERNEL;
115 	bio = bio_alloc(gfp_flags, nr_vecs);
116 
117 	if (bio == NULL && (current->flags & PF_MEMALLOC)) {
118 		while (!bio && (nr_vecs /= 2))
119 			bio = bio_alloc(gfp_flags, nr_vecs);
120 	}
121 
122 	if (bio) {
123 		bio_set_dev(bio, bdev);
124 		bio->bi_iter.bi_sector = first_sector;
125 	}
126 	return bio;
127 }
128 
129 /*
130  * support function for mpage_readahead.  The fs supplied get_block might
131  * return an up to date buffer.  This is used to map that buffer into
132  * the page, which allows readpage to avoid triggering a duplicate call
133  * to get_block.
134  *
135  * The idea is to avoid adding buffers to pages that don't already have
136  * them.  So when the buffer is up to date and the page size == block size,
137  * this marks the page up to date instead of adding new buffers.
138  */
139 static void
map_buffer_to_page(struct page * page,struct buffer_head * bh,int page_block)140 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
141 {
142 	struct inode *inode = page->mapping->host;
143 	struct buffer_head *page_bh, *head;
144 	int block = 0;
145 
146 	if (!page_has_buffers(page)) {
147 		/*
148 		 * don't make any buffers if there is only one buffer on
149 		 * the page and the page just needs to be set up to date
150 		 */
151 		if (inode->i_blkbits == PAGE_SHIFT &&
152 		    buffer_uptodate(bh)) {
153 			SetPageUptodate(page);
154 			return;
155 		}
156 		create_empty_buffers(page, i_blocksize(inode), 0);
157 	}
158 	head = page_buffers(page);
159 	page_bh = head;
160 	do {
161 		if (block == page_block) {
162 			page_bh->b_state = bh->b_state;
163 			page_bh->b_bdev = bh->b_bdev;
164 			page_bh->b_blocknr = bh->b_blocknr;
165 			break;
166 		}
167 		page_bh = page_bh->b_this_page;
168 		block++;
169 	} while (page_bh != head);
170 }
171 
172 struct mpage_readpage_args {
173 	struct bio *bio;
174 	struct page *page;
175 	unsigned int nr_pages;
176 	bool is_readahead;
177 	sector_t last_block_in_bio;
178 	struct buffer_head map_bh;
179 	unsigned long first_logical_block;
180 	get_block_t *get_block;
181 };
182 
183 /*
184  * This is the worker routine which does all the work of mapping the disk
185  * blocks and constructs largest possible bios, submits them for IO if the
186  * blocks are not contiguous on the disk.
187  *
188  * We pass a buffer_head back and forth and use its buffer_mapped() flag to
189  * represent the validity of its disk mapping and to decide when to do the next
190  * get_block() call.
191  */
do_mpage_readpage(struct mpage_readpage_args * args)192 static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
193 {
194 	struct page *page = args->page;
195 	struct inode *inode = page->mapping->host;
196 	const unsigned blkbits = inode->i_blkbits;
197 	const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
198 	const unsigned blocksize = 1 << blkbits;
199 	struct buffer_head *map_bh = &args->map_bh;
200 	sector_t block_in_file;
201 	sector_t last_block;
202 	sector_t last_block_in_file;
203 	sector_t blocks[MAX_BUF_PER_PAGE];
204 	unsigned page_block;
205 	unsigned first_hole = blocks_per_page;
206 	struct block_device *bdev = NULL;
207 	int length;
208 	int fully_mapped = 1;
209 	int op_flags;
210 	unsigned nblocks;
211 	unsigned relative_block;
212 	gfp_t gfp;
213 
214 	if (args->is_readahead) {
215 		op_flags = REQ_RAHEAD;
216 		gfp = readahead_gfp_mask(page->mapping);
217 	} else {
218 		op_flags = 0;
219 		gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
220 	}
221 
222 	if (page_has_buffers(page))
223 		goto confused;
224 
225 	block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
226 	last_block = block_in_file + args->nr_pages * blocks_per_page;
227 	last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
228 	if (last_block > last_block_in_file)
229 		last_block = last_block_in_file;
230 	page_block = 0;
231 
232 	/*
233 	 * Map blocks using the result from the previous get_blocks call first.
234 	 */
235 	nblocks = map_bh->b_size >> blkbits;
236 	if (buffer_mapped(map_bh) &&
237 			block_in_file > args->first_logical_block &&
238 			block_in_file < (args->first_logical_block + nblocks)) {
239 		unsigned map_offset = block_in_file - args->first_logical_block;
240 		unsigned last = nblocks - map_offset;
241 
242 		for (relative_block = 0; ; relative_block++) {
243 			if (relative_block == last) {
244 				clear_buffer_mapped(map_bh);
245 				break;
246 			}
247 			if (page_block == blocks_per_page)
248 				break;
249 			blocks[page_block] = map_bh->b_blocknr + map_offset +
250 						relative_block;
251 			page_block++;
252 			block_in_file++;
253 		}
254 		bdev = map_bh->b_bdev;
255 	}
256 
257 	/*
258 	 * Then do more get_blocks calls until we are done with this page.
259 	 */
260 	map_bh->b_page = page;
261 	while (page_block < blocks_per_page) {
262 		map_bh->b_state = 0;
263 		map_bh->b_size = 0;
264 
265 		if (block_in_file < last_block) {
266 			map_bh->b_size = (last_block-block_in_file) << blkbits;
267 			if (args->get_block(inode, block_in_file, map_bh, 0))
268 				goto confused;
269 			args->first_logical_block = block_in_file;
270 		}
271 
272 		if (!buffer_mapped(map_bh)) {
273 			fully_mapped = 0;
274 			if (first_hole == blocks_per_page)
275 				first_hole = page_block;
276 			page_block++;
277 			block_in_file++;
278 			continue;
279 		}
280 
281 		/* some filesystems will copy data into the page during
282 		 * the get_block call, in which case we don't want to
283 		 * read it again.  map_buffer_to_page copies the data
284 		 * we just collected from get_block into the page's buffers
285 		 * so readpage doesn't have to repeat the get_block call
286 		 */
287 		if (buffer_uptodate(map_bh)) {
288 			map_buffer_to_page(page, map_bh, page_block);
289 			goto confused;
290 		}
291 
292 		if (first_hole != blocks_per_page)
293 			goto confused;		/* hole -> non-hole */
294 
295 		/* Contiguous blocks? */
296 		if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
297 			goto confused;
298 		nblocks = map_bh->b_size >> blkbits;
299 		for (relative_block = 0; ; relative_block++) {
300 			if (relative_block == nblocks) {
301 				clear_buffer_mapped(map_bh);
302 				break;
303 			} else if (page_block == blocks_per_page)
304 				break;
305 			blocks[page_block] = map_bh->b_blocknr+relative_block;
306 			page_block++;
307 			block_in_file++;
308 		}
309 		bdev = map_bh->b_bdev;
310 	}
311 
312 	if (first_hole != blocks_per_page) {
313 		zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
314 		if (first_hole == 0) {
315 			SetPageUptodate(page);
316 			unlock_page(page);
317 			goto out;
318 		}
319 	} else if (fully_mapped) {
320 		SetPageMappedToDisk(page);
321 	}
322 
323 	if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
324 	    cleancache_get_page(page) == 0) {
325 		SetPageUptodate(page);
326 		goto confused;
327 	}
328 
329 	/*
330 	 * This page will go to BIO.  Do we need to send this BIO off first?
331 	 */
332 	if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
333 		args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
334 
335 alloc_new:
336 	if (args->bio == NULL) {
337 		if (first_hole == blocks_per_page) {
338 			if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
339 								page))
340 				goto out;
341 		}
342 		args->bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
343 					min_t(int, args->nr_pages,
344 					      BIO_MAX_PAGES),
345 					gfp);
346 		if (args->bio == NULL)
347 			goto confused;
348 	}
349 
350 	length = first_hole << blkbits;
351 	if (bio_add_page(args->bio, page, length, 0) < length) {
352 		args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
353 		goto alloc_new;
354 	}
355 
356 	relative_block = block_in_file - args->first_logical_block;
357 	nblocks = map_bh->b_size >> blkbits;
358 	if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
359 	    (first_hole != blocks_per_page))
360 		args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
361 	else
362 		args->last_block_in_bio = blocks[blocks_per_page - 1];
363 out:
364 	return args->bio;
365 
366 confused:
367 	if (args->bio)
368 		args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
369 	if (!PageUptodate(page))
370 		block_read_full_page(page, args->get_block);
371 	else
372 		unlock_page(page);
373 	goto out;
374 }
375 
376 /**
377  * mpage_readahead - start reads against pages
378  * @rac: Describes which pages to read.
379  * @get_block: The filesystem's block mapper function.
380  *
381  * This function walks the pages and the blocks within each page, building and
382  * emitting large BIOs.
383  *
384  * If anything unusual happens, such as:
385  *
386  * - encountering a page which has buffers
387  * - encountering a page which has a non-hole after a hole
388  * - encountering a page with non-contiguous blocks
389  *
390  * then this code just gives up and calls the buffer_head-based read function.
391  * It does handle a page which has holes at the end - that is a common case:
392  * the end-of-file on blocksize < PAGE_SIZE setups.
393  *
394  * BH_Boundary explanation:
395  *
396  * There is a problem.  The mpage read code assembles several pages, gets all
397  * their disk mappings, and then submits them all.  That's fine, but obtaining
398  * the disk mappings may require I/O.  Reads of indirect blocks, for example.
399  *
400  * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
401  * submitted in the following order:
402  *
403  * 	12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
404  *
405  * because the indirect block has to be read to get the mappings of blocks
406  * 13,14,15,16.  Obviously, this impacts performance.
407  *
408  * So what we do it to allow the filesystem's get_block() function to set
409  * BH_Boundary when it maps block 11.  BH_Boundary says: mapping of the block
410  * after this one will require I/O against a block which is probably close to
411  * this one.  So you should push what I/O you have currently accumulated.
412  *
413  * This all causes the disk requests to be issued in the correct order.
414  */
mpage_readahead(struct readahead_control * rac,get_block_t get_block)415 void mpage_readahead(struct readahead_control *rac, get_block_t get_block)
416 {
417 	struct page *page;
418 	struct mpage_readpage_args args = {
419 		.get_block = get_block,
420 		.is_readahead = true,
421 	};
422 
423 	while ((page = readahead_page(rac))) {
424 		prefetchw(&page->flags);
425 		args.page = page;
426 		args.nr_pages = readahead_count(rac);
427 		args.bio = do_mpage_readpage(&args);
428 		put_page(page);
429 	}
430 	if (args.bio)
431 		mpage_bio_submit(REQ_OP_READ, REQ_RAHEAD, args.bio);
432 }
433 EXPORT_SYMBOL_NS(mpage_readahead, ANDROID_GKI_VFS_EXPORT_ONLY);
434 
435 /*
436  * This isn't called much at all
437  */
mpage_readpage(struct page * page,get_block_t get_block)438 int mpage_readpage(struct page *page, get_block_t get_block)
439 {
440 	struct mpage_readpage_args args = {
441 		.page = page,
442 		.nr_pages = 1,
443 		.get_block = get_block,
444 	};
445 
446 	args.bio = do_mpage_readpage(&args);
447 	if (args.bio)
448 		mpage_bio_submit(REQ_OP_READ, 0, args.bio);
449 	return 0;
450 }
451 EXPORT_SYMBOL_NS(mpage_readpage, ANDROID_GKI_VFS_EXPORT_ONLY);
452 
453 /*
454  * Writing is not so simple.
455  *
456  * If the page has buffers then they will be used for obtaining the disk
457  * mapping.  We only support pages which are fully mapped-and-dirty, with a
458  * special case for pages which are unmapped at the end: end-of-file.
459  *
460  * If the page has no buffers (preferred) then the page is mapped here.
461  *
462  * If all blocks are found to be contiguous then the page can go into the
463  * BIO.  Otherwise fall back to the mapping's writepage().
464  *
465  * FIXME: This code wants an estimate of how many pages are still to be
466  * written, so it can intelligently allocate a suitably-sized BIO.  For now,
467  * just allocate full-size (16-page) BIOs.
468  */
469 
470 struct mpage_data {
471 	struct bio *bio;
472 	sector_t last_block_in_bio;
473 	get_block_t *get_block;
474 	unsigned use_writepage;
475 };
476 
477 /*
478  * We have our BIO, so we can now mark the buffers clean.  Make
479  * sure to only clean buffers which we know we'll be writing.
480  */
clean_buffers(struct page * page,unsigned first_unmapped)481 static void clean_buffers(struct page *page, unsigned first_unmapped)
482 {
483 	unsigned buffer_counter = 0;
484 	struct buffer_head *bh, *head;
485 	if (!page_has_buffers(page))
486 		return;
487 	head = page_buffers(page);
488 	bh = head;
489 
490 	do {
491 		if (buffer_counter++ == first_unmapped)
492 			break;
493 		clear_buffer_dirty(bh);
494 		bh = bh->b_this_page;
495 	} while (bh != head);
496 
497 	/*
498 	 * we cannot drop the bh if the page is not uptodate or a concurrent
499 	 * readpage would fail to serialize with the bh and it would read from
500 	 * disk before we reach the platter.
501 	 */
502 	if (buffer_heads_over_limit && PageUptodate(page))
503 		try_to_free_buffers(page);
504 }
505 
506 /*
507  * For situations where we want to clean all buffers attached to a page.
508  * We don't need to calculate how many buffers are attached to the page,
509  * we just need to specify a number larger than the maximum number of buffers.
510  */
clean_page_buffers(struct page * page)511 void clean_page_buffers(struct page *page)
512 {
513 	clean_buffers(page, ~0U);
514 }
515 
__mpage_writepage(struct page * page,struct writeback_control * wbc,void * data)516 static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
517 		      void *data)
518 {
519 	struct mpage_data *mpd = data;
520 	struct bio *bio = mpd->bio;
521 	struct address_space *mapping = page->mapping;
522 	struct inode *inode = page->mapping->host;
523 	const unsigned blkbits = inode->i_blkbits;
524 	unsigned long end_index;
525 	const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
526 	sector_t last_block;
527 	sector_t block_in_file;
528 	sector_t blocks[MAX_BUF_PER_PAGE];
529 	unsigned page_block;
530 	unsigned first_unmapped = blocks_per_page;
531 	struct block_device *bdev = NULL;
532 	int boundary = 0;
533 	sector_t boundary_block = 0;
534 	struct block_device *boundary_bdev = NULL;
535 	int length;
536 	struct buffer_head map_bh;
537 	loff_t i_size = i_size_read(inode);
538 	int ret = 0;
539 	int op_flags = wbc_to_write_flags(wbc);
540 
541 	if (page_has_buffers(page)) {
542 		struct buffer_head *head = page_buffers(page);
543 		struct buffer_head *bh = head;
544 
545 		/* If they're all mapped and dirty, do it */
546 		page_block = 0;
547 		do {
548 			BUG_ON(buffer_locked(bh));
549 			if (!buffer_mapped(bh)) {
550 				/*
551 				 * unmapped dirty buffers are created by
552 				 * __set_page_dirty_buffers -> mmapped data
553 				 */
554 				if (buffer_dirty(bh))
555 					goto confused;
556 				if (first_unmapped == blocks_per_page)
557 					first_unmapped = page_block;
558 				continue;
559 			}
560 
561 			if (first_unmapped != blocks_per_page)
562 				goto confused;	/* hole -> non-hole */
563 
564 			if (!buffer_dirty(bh) || !buffer_uptodate(bh))
565 				goto confused;
566 			if (page_block) {
567 				if (bh->b_blocknr != blocks[page_block-1] + 1)
568 					goto confused;
569 			}
570 			blocks[page_block++] = bh->b_blocknr;
571 			boundary = buffer_boundary(bh);
572 			if (boundary) {
573 				boundary_block = bh->b_blocknr;
574 				boundary_bdev = bh->b_bdev;
575 			}
576 			bdev = bh->b_bdev;
577 		} while ((bh = bh->b_this_page) != head);
578 
579 		if (first_unmapped)
580 			goto page_is_mapped;
581 
582 		/*
583 		 * Page has buffers, but they are all unmapped. The page was
584 		 * created by pagein or read over a hole which was handled by
585 		 * block_read_full_page().  If this address_space is also
586 		 * using mpage_readahead then this can rarely happen.
587 		 */
588 		goto confused;
589 	}
590 
591 	/*
592 	 * The page has no buffers: map it to disk
593 	 */
594 	BUG_ON(!PageUptodate(page));
595 	block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
596 	last_block = (i_size - 1) >> blkbits;
597 	map_bh.b_page = page;
598 	for (page_block = 0; page_block < blocks_per_page; ) {
599 
600 		map_bh.b_state = 0;
601 		map_bh.b_size = 1 << blkbits;
602 		if (mpd->get_block(inode, block_in_file, &map_bh, 1))
603 			goto confused;
604 		if (buffer_new(&map_bh))
605 			clean_bdev_bh_alias(&map_bh);
606 		if (buffer_boundary(&map_bh)) {
607 			boundary_block = map_bh.b_blocknr;
608 			boundary_bdev = map_bh.b_bdev;
609 		}
610 		if (page_block) {
611 			if (map_bh.b_blocknr != blocks[page_block-1] + 1)
612 				goto confused;
613 		}
614 		blocks[page_block++] = map_bh.b_blocknr;
615 		boundary = buffer_boundary(&map_bh);
616 		bdev = map_bh.b_bdev;
617 		if (block_in_file == last_block)
618 			break;
619 		block_in_file++;
620 	}
621 	BUG_ON(page_block == 0);
622 
623 	first_unmapped = page_block;
624 
625 page_is_mapped:
626 	end_index = i_size >> PAGE_SHIFT;
627 	if (page->index >= end_index) {
628 		/*
629 		 * The page straddles i_size.  It must be zeroed out on each
630 		 * and every writepage invocation because it may be mmapped.
631 		 * "A file is mapped in multiples of the page size.  For a file
632 		 * that is not a multiple of the page size, the remaining memory
633 		 * is zeroed when mapped, and writes to that region are not
634 		 * written out to the file."
635 		 */
636 		unsigned offset = i_size & (PAGE_SIZE - 1);
637 
638 		if (page->index > end_index || !offset)
639 			goto confused;
640 		zero_user_segment(page, offset, PAGE_SIZE);
641 	}
642 
643 	/*
644 	 * This page will go to BIO.  Do we need to send this BIO off first?
645 	 */
646 	if (bio && mpd->last_block_in_bio != blocks[0] - 1)
647 		bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
648 
649 alloc_new:
650 	if (bio == NULL) {
651 		if (first_unmapped == blocks_per_page) {
652 			if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
653 								page, wbc))
654 				goto out;
655 		}
656 		bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
657 				BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
658 		if (bio == NULL)
659 			goto confused;
660 
661 		wbc_init_bio(wbc, bio);
662 		bio->bi_write_hint = inode->i_write_hint;
663 	}
664 
665 	/*
666 	 * Must try to add the page before marking the buffer clean or
667 	 * the confused fail path above (OOM) will be very confused when
668 	 * it finds all bh marked clean (i.e. it will not write anything)
669 	 */
670 	wbc_account_cgroup_owner(wbc, page, PAGE_SIZE);
671 	length = first_unmapped << blkbits;
672 	if (bio_add_page(bio, page, length, 0) < length) {
673 		bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
674 		goto alloc_new;
675 	}
676 
677 	clean_buffers(page, first_unmapped);
678 
679 	BUG_ON(PageWriteback(page));
680 	set_page_writeback(page);
681 	unlock_page(page);
682 	if (boundary || (first_unmapped != blocks_per_page)) {
683 		bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
684 		if (boundary_block) {
685 			write_boundary_block(boundary_bdev,
686 					boundary_block, 1 << blkbits);
687 		}
688 	} else {
689 		mpd->last_block_in_bio = blocks[blocks_per_page - 1];
690 	}
691 	goto out;
692 
693 confused:
694 	if (bio)
695 		bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
696 
697 	if (mpd->use_writepage) {
698 		ret = mapping->a_ops->writepage(page, wbc);
699 	} else {
700 		ret = -EAGAIN;
701 		goto out;
702 	}
703 	/*
704 	 * The caller has a ref on the inode, so *mapping is stable
705 	 */
706 	mapping_set_error(mapping, ret);
707 out:
708 	mpd->bio = bio;
709 	return ret;
710 }
711 
712 /**
713  * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
714  * @mapping: address space structure to write
715  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
716  * @get_block: the filesystem's block mapper function.
717  *             If this is NULL then use a_ops->writepage.  Otherwise, go
718  *             direct-to-BIO.
719  *
720  * This is a library function, which implements the writepages()
721  * address_space_operation.
722  *
723  * If a page is already under I/O, generic_writepages() skips it, even
724  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
725  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
726  * and msync() need to guarantee that all the data which was dirty at the time
727  * the call was made get new I/O started against them.  If wbc->sync_mode is
728  * WB_SYNC_ALL then we were called for data integrity and we must wait for
729  * existing IO to complete.
730  */
731 int
mpage_writepages(struct address_space * mapping,struct writeback_control * wbc,get_block_t get_block)732 mpage_writepages(struct address_space *mapping,
733 		struct writeback_control *wbc, get_block_t get_block)
734 {
735 	struct blk_plug plug;
736 	int ret;
737 
738 	blk_start_plug(&plug);
739 
740 	if (!get_block)
741 		ret = generic_writepages(mapping, wbc);
742 	else {
743 		struct mpage_data mpd = {
744 			.bio = NULL,
745 			.last_block_in_bio = 0,
746 			.get_block = get_block,
747 			.use_writepage = 1,
748 		};
749 
750 		ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
751 		if (mpd.bio) {
752 			int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
753 				  REQ_SYNC : 0);
754 			mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
755 		}
756 	}
757 	blk_finish_plug(&plug);
758 	return ret;
759 }
760 EXPORT_SYMBOL(mpage_writepages);
761 
mpage_writepage(struct page * page,get_block_t get_block,struct writeback_control * wbc)762 int mpage_writepage(struct page *page, get_block_t get_block,
763 	struct writeback_control *wbc)
764 {
765 	struct mpage_data mpd = {
766 		.bio = NULL,
767 		.last_block_in_bio = 0,
768 		.get_block = get_block,
769 		.use_writepage = 0,
770 	};
771 	int ret = __mpage_writepage(page, wbc, &mpd);
772 	if (mpd.bio) {
773 		int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
774 			  REQ_SYNC : 0);
775 		mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
776 	}
777 	return ret;
778 }
779 EXPORT_SYMBOL(mpage_writepage);
780