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