xref: /openbmc/linux/fs/ext4/page-io.c (revision 4f205687) 
1 /*
2  * linux/fs/ext4/page-io.c
3  *
4  * This contains the new page_io functions for ext4
5  *
6  * Written by Theodore Ts'o, 2010.
7  */
8 
9 #include <linux/fs.h>
10 #include <linux/time.h>
11 #include <linux/highuid.h>
12 #include <linux/pagemap.h>
13 #include <linux/quotaops.h>
14 #include <linux/string.h>
15 #include <linux/buffer_head.h>
16 #include <linux/writeback.h>
17 #include <linux/pagevec.h>
18 #include <linux/mpage.h>
19 #include <linux/namei.h>
20 #include <linux/uio.h>
21 #include <linux/bio.h>
22 #include <linux/workqueue.h>
23 #include <linux/kernel.h>
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/backing-dev.h>
27 
28 #include "ext4_jbd2.h"
29 #include "xattr.h"
30 #include "acl.h"
31 
32 static struct kmem_cache *io_end_cachep;
33 
34 int __init ext4_init_pageio(void)
35 {
36 	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
37 	if (io_end_cachep == NULL)
38 		return -ENOMEM;
39 	return 0;
40 }
41 
42 void ext4_exit_pageio(void)
43 {
44 	kmem_cache_destroy(io_end_cachep);
45 }
46 
47 /*
48  * Print an buffer I/O error compatible with the fs/buffer.c.  This
49  * provides compatibility with dmesg scrapers that look for a specific
50  * buffer I/O error message.  We really need a unified error reporting
51  * structure to userspace ala Digital Unix's uerf system, but it's
52  * probably not going to happen in my lifetime, due to LKML politics...
53  */
54 static void buffer_io_error(struct buffer_head *bh)
55 {
56 	printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
57 		       bh->b_bdev,
58 			(unsigned long long)bh->b_blocknr);
59 }
60 
61 static void ext4_finish_bio(struct bio *bio)
62 {
63 	int i;
64 	struct bio_vec *bvec;
65 
66 	bio_for_each_segment_all(bvec, bio, i) {
67 		struct page *page = bvec->bv_page;
68 #ifdef CONFIG_EXT4_FS_ENCRYPTION
69 		struct page *data_page = NULL;
70 		struct ext4_crypto_ctx *ctx = NULL;
71 #endif
72 		struct buffer_head *bh, *head;
73 		unsigned bio_start = bvec->bv_offset;
74 		unsigned bio_end = bio_start + bvec->bv_len;
75 		unsigned under_io = 0;
76 		unsigned long flags;
77 
78 		if (!page)
79 			continue;
80 
81 #ifdef CONFIG_EXT4_FS_ENCRYPTION
82 		if (!page->mapping) {
83 			/* The bounce data pages are unmapped. */
84 			data_page = page;
85 			ctx = (struct ext4_crypto_ctx *)page_private(data_page);
86 			page = ctx->w.control_page;
87 		}
88 #endif
89 
90 		if (bio->bi_error) {
91 			SetPageError(page);
92 			set_bit(AS_EIO, &page->mapping->flags);
93 		}
94 		bh = head = page_buffers(page);
95 		/*
96 		 * We check all buffers in the page under BH_Uptodate_Lock
97 		 * to avoid races with other end io clearing async_write flags
98 		 */
99 		local_irq_save(flags);
100 		bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
101 		do {
102 			if (bh_offset(bh) < bio_start ||
103 			    bh_offset(bh) + bh->b_size > bio_end) {
104 				if (buffer_async_write(bh))
105 					under_io++;
106 				continue;
107 			}
108 			clear_buffer_async_write(bh);
109 			if (bio->bi_error)
110 				buffer_io_error(bh);
111 		} while ((bh = bh->b_this_page) != head);
112 		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
113 		local_irq_restore(flags);
114 		if (!under_io) {
115 #ifdef CONFIG_EXT4_FS_ENCRYPTION
116 			if (ctx)
117 				ext4_restore_control_page(data_page);
118 #endif
119 			end_page_writeback(page);
120 		}
121 	}
122 }
123 
124 static void ext4_release_io_end(ext4_io_end_t *io_end)
125 {
126 	struct bio *bio, *next_bio;
127 
128 	BUG_ON(!list_empty(&io_end->list));
129 	BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
130 	WARN_ON(io_end->handle);
131 
132 	for (bio = io_end->bio; bio; bio = next_bio) {
133 		next_bio = bio->bi_private;
134 		ext4_finish_bio(bio);
135 		bio_put(bio);
136 	}
137 	kmem_cache_free(io_end_cachep, io_end);
138 }
139 
140 /*
141  * Check a range of space and convert unwritten extents to written. Note that
142  * we are protected from truncate touching same part of extent tree by the
143  * fact that truncate code waits for all DIO to finish (thus exclusion from
144  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
145  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
146  * completed (happens from ext4_free_ioend()).
147  */
148 static int ext4_end_io(ext4_io_end_t *io)
149 {
150 	struct inode *inode = io->inode;
151 	loff_t offset = io->offset;
152 	ssize_t size = io->size;
153 	handle_t *handle = io->handle;
154 	int ret = 0;
155 
156 	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
157 		   "list->prev 0x%p\n",
158 		   io, inode->i_ino, io->list.next, io->list.prev);
159 
160 	io->handle = NULL;	/* Following call will use up the handle */
161 	ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
162 	if (ret < 0) {
163 		ext4_msg(inode->i_sb, KERN_EMERG,
164 			 "failed to convert unwritten extents to written "
165 			 "extents -- potential data loss!  "
166 			 "(inode %lu, offset %llu, size %zd, error %d)",
167 			 inode->i_ino, offset, size, ret);
168 	}
169 	ext4_clear_io_unwritten_flag(io);
170 	ext4_release_io_end(io);
171 	return ret;
172 }
173 
174 static void dump_completed_IO(struct inode *inode, struct list_head *head)
175 {
176 #ifdef	EXT4FS_DEBUG
177 	struct list_head *cur, *before, *after;
178 	ext4_io_end_t *io, *io0, *io1;
179 
180 	if (list_empty(head))
181 		return;
182 
183 	ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
184 	list_for_each_entry(io, head, list) {
185 		cur = &io->list;
186 		before = cur->prev;
187 		io0 = container_of(before, ext4_io_end_t, list);
188 		after = cur->next;
189 		io1 = container_of(after, ext4_io_end_t, list);
190 
191 		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
192 			    io, inode->i_ino, io0, io1);
193 	}
194 #endif
195 }
196 
197 /* Add the io_end to per-inode completed end_io list. */
198 static void ext4_add_complete_io(ext4_io_end_t *io_end)
199 {
200 	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
201 	struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
202 	struct workqueue_struct *wq;
203 	unsigned long flags;
204 
205 	/* Only reserved conversions from writeback should enter here */
206 	WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
207 	WARN_ON(!io_end->handle && sbi->s_journal);
208 	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
209 	wq = sbi->rsv_conversion_wq;
210 	if (list_empty(&ei->i_rsv_conversion_list))
211 		queue_work(wq, &ei->i_rsv_conversion_work);
212 	list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
213 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
214 }
215 
216 static int ext4_do_flush_completed_IO(struct inode *inode,
217 				      struct list_head *head)
218 {
219 	ext4_io_end_t *io;
220 	struct list_head unwritten;
221 	unsigned long flags;
222 	struct ext4_inode_info *ei = EXT4_I(inode);
223 	int err, ret = 0;
224 
225 	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
226 	dump_completed_IO(inode, head);
227 	list_replace_init(head, &unwritten);
228 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
229 
230 	while (!list_empty(&unwritten)) {
231 		io = list_entry(unwritten.next, ext4_io_end_t, list);
232 		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
233 		list_del_init(&io->list);
234 
235 		err = ext4_end_io(io);
236 		if (unlikely(!ret && err))
237 			ret = err;
238 	}
239 	return ret;
240 }
241 
242 /*
243  * work on completed IO, to convert unwritten extents to extents
244  */
245 void ext4_end_io_rsv_work(struct work_struct *work)
246 {
247 	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
248 						  i_rsv_conversion_work);
249 	ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
250 }
251 
252 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
253 {
254 	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
255 	if (io) {
256 		io->inode = inode;
257 		INIT_LIST_HEAD(&io->list);
258 		atomic_set(&io->count, 1);
259 	}
260 	return io;
261 }
262 
263 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
264 {
265 	if (atomic_dec_and_test(&io_end->count)) {
266 		if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
267 			ext4_release_io_end(io_end);
268 			return;
269 		}
270 		ext4_add_complete_io(io_end);
271 	}
272 }
273 
274 int ext4_put_io_end(ext4_io_end_t *io_end)
275 {
276 	int err = 0;
277 
278 	if (atomic_dec_and_test(&io_end->count)) {
279 		if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
280 			err = ext4_convert_unwritten_extents(io_end->handle,
281 						io_end->inode, io_end->offset,
282 						io_end->size);
283 			io_end->handle = NULL;
284 			ext4_clear_io_unwritten_flag(io_end);
285 		}
286 		ext4_release_io_end(io_end);
287 	}
288 	return err;
289 }
290 
291 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
292 {
293 	atomic_inc(&io_end->count);
294 	return io_end;
295 }
296 
297 /* BIO completion function for page writeback */
298 static void ext4_end_bio(struct bio *bio)
299 {
300 	ext4_io_end_t *io_end = bio->bi_private;
301 	sector_t bi_sector = bio->bi_iter.bi_sector;
302 
303 	BUG_ON(!io_end);
304 	bio->bi_end_io = NULL;
305 
306 	if (bio->bi_error) {
307 		struct inode *inode = io_end->inode;
308 
309 		ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
310 			     "(offset %llu size %ld starting block %llu)",
311 			     bio->bi_error, inode->i_ino,
312 			     (unsigned long long) io_end->offset,
313 			     (long) io_end->size,
314 			     (unsigned long long)
315 			     bi_sector >> (inode->i_blkbits - 9));
316 		mapping_set_error(inode->i_mapping, bio->bi_error);
317 	}
318 
319 	if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
320 		/*
321 		 * Link bio into list hanging from io_end. We have to do it
322 		 * atomically as bio completions can be racing against each
323 		 * other.
324 		 */
325 		bio->bi_private = xchg(&io_end->bio, bio);
326 		ext4_put_io_end_defer(io_end);
327 	} else {
328 		/*
329 		 * Drop io_end reference early. Inode can get freed once
330 		 * we finish the bio.
331 		 */
332 		ext4_put_io_end_defer(io_end);
333 		ext4_finish_bio(bio);
334 		bio_put(bio);
335 	}
336 }
337 
338 void ext4_io_submit(struct ext4_io_submit *io)
339 {
340 	struct bio *bio = io->io_bio;
341 
342 	if (bio) {
343 		int io_op = io->io_wbc->sync_mode == WB_SYNC_ALL ?
344 			    WRITE_SYNC : WRITE;
345 		submit_bio(io_op, io->io_bio);
346 	}
347 	io->io_bio = NULL;
348 }
349 
350 void ext4_io_submit_init(struct ext4_io_submit *io,
351 			 struct writeback_control *wbc)
352 {
353 	io->io_wbc = wbc;
354 	io->io_bio = NULL;
355 	io->io_end = NULL;
356 }
357 
358 static int io_submit_init_bio(struct ext4_io_submit *io,
359 			      struct buffer_head *bh)
360 {
361 	struct bio *bio;
362 
363 	bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
364 	if (!bio)
365 		return -ENOMEM;
366 	wbc_init_bio(io->io_wbc, bio);
367 	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
368 	bio->bi_bdev = bh->b_bdev;
369 	bio->bi_end_io = ext4_end_bio;
370 	bio->bi_private = ext4_get_io_end(io->io_end);
371 	io->io_bio = bio;
372 	io->io_next_block = bh->b_blocknr;
373 	return 0;
374 }
375 
376 static int io_submit_add_bh(struct ext4_io_submit *io,
377 			    struct inode *inode,
378 			    struct page *page,
379 			    struct buffer_head *bh)
380 {
381 	int ret;
382 
383 	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
384 submit_and_retry:
385 		ext4_io_submit(io);
386 	}
387 	if (io->io_bio == NULL) {
388 		ret = io_submit_init_bio(io, bh);
389 		if (ret)
390 			return ret;
391 	}
392 	ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
393 	if (ret != bh->b_size)
394 		goto submit_and_retry;
395 	wbc_account_io(io->io_wbc, page, bh->b_size);
396 	io->io_next_block++;
397 	return 0;
398 }
399 
400 int ext4_bio_write_page(struct ext4_io_submit *io,
401 			struct page *page,
402 			int len,
403 			struct writeback_control *wbc,
404 			bool keep_towrite)
405 {
406 	struct page *data_page = NULL;
407 	struct inode *inode = page->mapping->host;
408 	unsigned block_start, blocksize;
409 	struct buffer_head *bh, *head;
410 	int ret = 0;
411 	int nr_submitted = 0;
412 	int nr_to_submit = 0;
413 
414 	blocksize = 1 << inode->i_blkbits;
415 
416 	BUG_ON(!PageLocked(page));
417 	BUG_ON(PageWriteback(page));
418 
419 	if (keep_towrite)
420 		set_page_writeback_keepwrite(page);
421 	else
422 		set_page_writeback(page);
423 	ClearPageError(page);
424 
425 	/*
426 	 * Comments copied from block_write_full_page:
427 	 *
428 	 * The page straddles i_size.  It must be zeroed out on each and every
429 	 * writepage invocation because it may be mmapped.  "A file is mapped
430 	 * in multiples of the page size.  For a file that is not a multiple of
431 	 * the page size, the remaining memory is zeroed when mapped, and
432 	 * writes to that region are not written out to the file."
433 	 */
434 	if (len < PAGE_SIZE)
435 		zero_user_segment(page, len, PAGE_SIZE);
436 	/*
437 	 * In the first loop we prepare and mark buffers to submit. We have to
438 	 * mark all buffers in the page before submitting so that
439 	 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
440 	 * on the first buffer finishes and we are still working on submitting
441 	 * the second buffer.
442 	 */
443 	bh = head = page_buffers(page);
444 	do {
445 		block_start = bh_offset(bh);
446 		if (block_start >= len) {
447 			clear_buffer_dirty(bh);
448 			set_buffer_uptodate(bh);
449 			continue;
450 		}
451 		if (!buffer_dirty(bh) || buffer_delay(bh) ||
452 		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
453 			/* A hole? We can safely clear the dirty bit */
454 			if (!buffer_mapped(bh))
455 				clear_buffer_dirty(bh);
456 			if (io->io_bio)
457 				ext4_io_submit(io);
458 			continue;
459 		}
460 		if (buffer_new(bh)) {
461 			clear_buffer_new(bh);
462 			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
463 		}
464 		set_buffer_async_write(bh);
465 		nr_to_submit++;
466 	} while ((bh = bh->b_this_page) != head);
467 
468 	bh = head = page_buffers(page);
469 
470 	if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
471 	    nr_to_submit) {
472 		gfp_t gfp_flags = GFP_NOFS;
473 
474 	retry_encrypt:
475 		data_page = ext4_encrypt(inode, page, gfp_flags);
476 		if (IS_ERR(data_page)) {
477 			ret = PTR_ERR(data_page);
478 			if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
479 				if (io->io_bio) {
480 					ext4_io_submit(io);
481 					congestion_wait(BLK_RW_ASYNC, HZ/50);
482 				}
483 				gfp_flags |= __GFP_NOFAIL;
484 				goto retry_encrypt;
485 			}
486 			data_page = NULL;
487 			goto out;
488 		}
489 	}
490 
491 	/* Now submit buffers to write */
492 	do {
493 		if (!buffer_async_write(bh))
494 			continue;
495 		ret = io_submit_add_bh(io, inode,
496 				       data_page ? data_page : page, bh);
497 		if (ret) {
498 			/*
499 			 * We only get here on ENOMEM.  Not much else
500 			 * we can do but mark the page as dirty, and
501 			 * better luck next time.
502 			 */
503 			break;
504 		}
505 		nr_submitted++;
506 		clear_buffer_dirty(bh);
507 	} while ((bh = bh->b_this_page) != head);
508 
509 	/* Error stopped previous loop? Clean up buffers... */
510 	if (ret) {
511 	out:
512 		if (data_page)
513 			ext4_restore_control_page(data_page);
514 		printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
515 		redirty_page_for_writepage(wbc, page);
516 		do {
517 			clear_buffer_async_write(bh);
518 			bh = bh->b_this_page;
519 		} while (bh != head);
520 	}
521 	unlock_page(page);
522 	/* Nothing submitted - we have to end page writeback */
523 	if (!nr_submitted)
524 		end_page_writeback(page);
525 	return ret;
526 }
527