xref: /linux/fs/ext4/page-io.c (revision fc0460d0dfb19d4887f509a2573197ca0f417ce9)
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/jbd2.h>
12 #include <linux/highuid.h>
13 #include <linux/pagemap.h>
14 #include <linux/quotaops.h>
15 #include <linux/string.h>
16 #include <linux/buffer_head.h>
17 #include <linux/writeback.h>
18 #include <linux/pagevec.h>
19 #include <linux/mpage.h>
20 #include <linux/namei.h>
21 #include <linux/uio.h>
22 #include <linux/bio.h>
23 #include <linux/workqueue.h>
24 #include <linux/kernel.h>
25 #include <linux/slab.h>
26 #include <linux/mm.h>
27 
28 #include "ext4_jbd2.h"
29 #include "xattr.h"
30 #include "acl.h"
31 
32 static struct kmem_cache *io_page_cachep, *io_end_cachep;
33 
34 int __init ext4_init_pageio(void)
35 {
36 	io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
37 	if (io_page_cachep == NULL)
38 		return -ENOMEM;
39 	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
40 	if (io_end_cachep == NULL) {
41 		kmem_cache_destroy(io_page_cachep);
42 		return -ENOMEM;
43 	}
44 	return 0;
45 }
46 
47 void ext4_exit_pageio(void)
48 {
49 	kmem_cache_destroy(io_end_cachep);
50 	kmem_cache_destroy(io_page_cachep);
51 }
52 
53 void ext4_ioend_wait(struct inode *inode)
54 {
55 	wait_queue_head_t *wq = ext4_ioend_wq(inode);
56 
57 	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
58 }
59 
60 static void put_io_page(struct ext4_io_page *io_page)
61 {
62 	if (atomic_dec_and_test(&io_page->p_count)) {
63 		end_page_writeback(io_page->p_page);
64 		put_page(io_page->p_page);
65 		kmem_cache_free(io_page_cachep, io_page);
66 	}
67 }
68 
69 void ext4_free_io_end(ext4_io_end_t *io)
70 {
71 	int i;
72 
73 	BUG_ON(!io);
74 	BUG_ON(!list_empty(&io->list));
75 	BUG_ON(io->flag & EXT4_IO_END_UNWRITTEN);
76 
77 	for (i = 0; i < io->num_io_pages; i++)
78 		put_io_page(io->pages[i]);
79 	io->num_io_pages = 0;
80 	if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count))
81 		wake_up_all(ext4_ioend_wq(io->inode));
82 	kmem_cache_free(io_end_cachep, io);
83 }
84 
85 /* check a range of space and convert unwritten extents to written. */
86 static int ext4_end_io(ext4_io_end_t *io)
87 {
88 	struct inode *inode = io->inode;
89 	loff_t offset = io->offset;
90 	ssize_t size = io->size;
91 	int ret = 0;
92 
93 	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
94 		   "list->prev 0x%p\n",
95 		   io, inode->i_ino, io->list.next, io->list.prev);
96 
97 	ret = ext4_convert_unwritten_extents(inode, offset, size);
98 	if (ret < 0) {
99 		ext4_msg(inode->i_sb, KERN_EMERG,
100 			 "failed to convert unwritten extents to written "
101 			 "extents -- potential data loss!  "
102 			 "(inode %lu, offset %llu, size %zd, error %d)",
103 			 inode->i_ino, offset, size, ret);
104 	}
105 	/* Wake up anyone waiting on unwritten extent conversion */
106 	if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
107 		wake_up_all(ext4_ioend_wq(inode));
108 	if (io->flag & EXT4_IO_END_DIRECT)
109 		inode_dio_done(inode);
110 	if (io->iocb)
111 		aio_complete(io->iocb, io->result, 0);
112 	return ret;
113 }
114 
115 static void dump_completed_IO(struct inode *inode)
116 {
117 #ifdef	EXT4FS_DEBUG
118 	struct list_head *cur, *before, *after;
119 	ext4_io_end_t *io, *io0, *io1;
120 
121 	if (list_empty(&EXT4_I(inode)->i_completed_io_list)) {
122 		ext4_debug("inode %lu completed_io list is empty\n",
123 			   inode->i_ino);
124 		return;
125 	}
126 
127 	ext4_debug("Dump inode %lu completed_io list\n", inode->i_ino);
128 	list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list) {
129 		cur = &io->list;
130 		before = cur->prev;
131 		io0 = container_of(before, ext4_io_end_t, list);
132 		after = cur->next;
133 		io1 = container_of(after, ext4_io_end_t, list);
134 
135 		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
136 			    io, inode->i_ino, io0, io1);
137 	}
138 #endif
139 }
140 
141 /* Add the io_end to per-inode completed end_io list. */
142 void ext4_add_complete_io(ext4_io_end_t *io_end)
143 {
144 	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
145 	struct workqueue_struct *wq;
146 	unsigned long flags;
147 
148 	BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
149 	wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
150 
151 	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
152 	if (list_empty(&ei->i_completed_io_list))
153 		queue_work(wq, &ei->i_unwritten_work);
154 	list_add_tail(&io_end->list, &ei->i_completed_io_list);
155 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
156 }
157 
158 static int ext4_do_flush_completed_IO(struct inode *inode)
159 {
160 	ext4_io_end_t *io;
161 	struct list_head unwritten;
162 	unsigned long flags;
163 	struct ext4_inode_info *ei = EXT4_I(inode);
164 	int err, ret = 0;
165 
166 	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
167 	dump_completed_IO(inode);
168 	list_replace_init(&ei->i_completed_io_list, &unwritten);
169 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
170 
171 	while (!list_empty(&unwritten)) {
172 		io = list_entry(unwritten.next, ext4_io_end_t, list);
173 		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
174 		list_del_init(&io->list);
175 
176 		err = ext4_end_io(io);
177 		if (unlikely(!ret && err))
178 			ret = err;
179 		io->flag &= ~EXT4_IO_END_UNWRITTEN;
180 		ext4_free_io_end(io);
181 	}
182 	return ret;
183 }
184 
185 /*
186  * work on completed aio dio IO, to convert unwritten extents to extents
187  */
188 void ext4_end_io_work(struct work_struct *work)
189 {
190 	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
191 						  i_unwritten_work);
192 	ext4_do_flush_completed_IO(&ei->vfs_inode);
193 }
194 
195 int ext4_flush_unwritten_io(struct inode *inode)
196 {
197 	int ret;
198 	WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) &&
199 		     !(inode->i_state & I_FREEING));
200 	ret = ext4_do_flush_completed_IO(inode);
201 	ext4_unwritten_wait(inode);
202 	return ret;
203 }
204 
205 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
206 {
207 	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
208 	if (io) {
209 		atomic_inc(&EXT4_I(inode)->i_ioend_count);
210 		io->inode = inode;
211 		INIT_LIST_HEAD(&io->list);
212 	}
213 	return io;
214 }
215 
216 /*
217  * Print an buffer I/O error compatible with the fs/buffer.c.  This
218  * provides compatibility with dmesg scrapers that look for a specific
219  * buffer I/O error message.  We really need a unified error reporting
220  * structure to userspace ala Digital Unix's uerf system, but it's
221  * probably not going to happen in my lifetime, due to LKML politics...
222  */
223 static void buffer_io_error(struct buffer_head *bh)
224 {
225 	char b[BDEVNAME_SIZE];
226 	printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
227 			bdevname(bh->b_bdev, b),
228 			(unsigned long long)bh->b_blocknr);
229 }
230 
231 static void ext4_end_bio(struct bio *bio, int error)
232 {
233 	ext4_io_end_t *io_end = bio->bi_private;
234 	struct inode *inode;
235 	int i;
236 	sector_t bi_sector = bio->bi_sector;
237 
238 	BUG_ON(!io_end);
239 	bio->bi_private = NULL;
240 	bio->bi_end_io = NULL;
241 	if (test_bit(BIO_UPTODATE, &bio->bi_flags))
242 		error = 0;
243 	bio_put(bio);
244 
245 	for (i = 0; i < io_end->num_io_pages; i++) {
246 		struct page *page = io_end->pages[i]->p_page;
247 		struct buffer_head *bh, *head;
248 		loff_t offset;
249 		loff_t io_end_offset;
250 
251 		if (error) {
252 			SetPageError(page);
253 			set_bit(AS_EIO, &page->mapping->flags);
254 			head = page_buffers(page);
255 			BUG_ON(!head);
256 
257 			io_end_offset = io_end->offset + io_end->size;
258 
259 			offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
260 			bh = head;
261 			do {
262 				if ((offset >= io_end->offset) &&
263 				    (offset+bh->b_size <= io_end_offset))
264 					buffer_io_error(bh);
265 
266 				offset += bh->b_size;
267 				bh = bh->b_this_page;
268 			} while (bh != head);
269 		}
270 
271 		put_io_page(io_end->pages[i]);
272 	}
273 	io_end->num_io_pages = 0;
274 	inode = io_end->inode;
275 
276 	if (error) {
277 		io_end->flag |= EXT4_IO_END_ERROR;
278 		ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
279 			     "(offset %llu size %ld starting block %llu)",
280 			     inode->i_ino,
281 			     (unsigned long long) io_end->offset,
282 			     (long) io_end->size,
283 			     (unsigned long long)
284 			     bi_sector >> (inode->i_blkbits - 9));
285 	}
286 
287 	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
288 		ext4_free_io_end(io_end);
289 		return;
290 	}
291 
292 	ext4_add_complete_io(io_end);
293 }
294 
295 void ext4_io_submit(struct ext4_io_submit *io)
296 {
297 	struct bio *bio = io->io_bio;
298 
299 	if (bio) {
300 		bio_get(io->io_bio);
301 		submit_bio(io->io_op, io->io_bio);
302 		BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
303 		bio_put(io->io_bio);
304 	}
305 	io->io_bio = NULL;
306 	io->io_op = 0;
307 	io->io_end = NULL;
308 }
309 
310 static int io_submit_init(struct ext4_io_submit *io,
311 			  struct inode *inode,
312 			  struct writeback_control *wbc,
313 			  struct buffer_head *bh)
314 {
315 	ext4_io_end_t *io_end;
316 	struct page *page = bh->b_page;
317 	int nvecs = bio_get_nr_vecs(bh->b_bdev);
318 	struct bio *bio;
319 
320 	io_end = ext4_init_io_end(inode, GFP_NOFS);
321 	if (!io_end)
322 		return -ENOMEM;
323 	bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
324 	bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
325 	bio->bi_bdev = bh->b_bdev;
326 	bio->bi_private = io->io_end = io_end;
327 	bio->bi_end_io = ext4_end_bio;
328 
329 	io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
330 
331 	io->io_bio = bio;
332 	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
333 	io->io_next_block = bh->b_blocknr;
334 	return 0;
335 }
336 
337 static int io_submit_add_bh(struct ext4_io_submit *io,
338 			    struct ext4_io_page *io_page,
339 			    struct inode *inode,
340 			    struct writeback_control *wbc,
341 			    struct buffer_head *bh)
342 {
343 	ext4_io_end_t *io_end;
344 	int ret;
345 
346 	if (buffer_new(bh)) {
347 		clear_buffer_new(bh);
348 		unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
349 	}
350 
351 	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
352 submit_and_retry:
353 		ext4_io_submit(io);
354 	}
355 	if (io->io_bio == NULL) {
356 		ret = io_submit_init(io, inode, wbc, bh);
357 		if (ret)
358 			return ret;
359 	}
360 	io_end = io->io_end;
361 	if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
362 	    (io_end->pages[io_end->num_io_pages-1] != io_page))
363 		goto submit_and_retry;
364 	if (buffer_uninit(bh))
365 		ext4_set_io_unwritten_flag(inode, io_end);
366 	io->io_end->size += bh->b_size;
367 	io->io_next_block++;
368 	ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
369 	if (ret != bh->b_size)
370 		goto submit_and_retry;
371 	if ((io_end->num_io_pages == 0) ||
372 	    (io_end->pages[io_end->num_io_pages-1] != io_page)) {
373 		io_end->pages[io_end->num_io_pages++] = io_page;
374 		atomic_inc(&io_page->p_count);
375 	}
376 	return 0;
377 }
378 
379 int ext4_bio_write_page(struct ext4_io_submit *io,
380 			struct page *page,
381 			int len,
382 			struct writeback_control *wbc)
383 {
384 	struct inode *inode = page->mapping->host;
385 	unsigned block_start, block_end, blocksize;
386 	struct ext4_io_page *io_page;
387 	struct buffer_head *bh, *head;
388 	int ret = 0;
389 
390 	blocksize = 1 << inode->i_blkbits;
391 
392 	BUG_ON(!PageLocked(page));
393 	BUG_ON(PageWriteback(page));
394 
395 	io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
396 	if (!io_page) {
397 		redirty_page_for_writepage(wbc, page);
398 		unlock_page(page);
399 		return -ENOMEM;
400 	}
401 	io_page->p_page = page;
402 	atomic_set(&io_page->p_count, 1);
403 	get_page(page);
404 	set_page_writeback(page);
405 	ClearPageError(page);
406 
407 	for (bh = head = page_buffers(page), block_start = 0;
408 	     bh != head || !block_start;
409 	     block_start = block_end, bh = bh->b_this_page) {
410 
411 		block_end = block_start + blocksize;
412 		if (block_start >= len) {
413 			/*
414 			 * Comments copied from block_write_full_page_endio:
415 			 *
416 			 * The page straddles i_size.  It must be zeroed out on
417 			 * each and every writepage invocation because it may
418 			 * be mmapped.  "A file is mapped in multiples of the
419 			 * page size.  For a file that is not a multiple of
420 			 * the  page size, the remaining memory is zeroed when
421 			 * mapped, and writes to that region are not written
422 			 * out to the file."
423 			 */
424 			zero_user_segment(page, block_start, block_end);
425 			clear_buffer_dirty(bh);
426 			set_buffer_uptodate(bh);
427 			continue;
428 		}
429 		if (!buffer_dirty(bh) || buffer_delay(bh) ||
430 		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
431 			/* A hole? We can safely clear the dirty bit */
432 			if (!buffer_mapped(bh))
433 				clear_buffer_dirty(bh);
434 			if (io->io_bio)
435 				ext4_io_submit(io);
436 			continue;
437 		}
438 		ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
439 		if (ret) {
440 			/*
441 			 * We only get here on ENOMEM.  Not much else
442 			 * we can do but mark the page as dirty, and
443 			 * better luck next time.
444 			 */
445 			redirty_page_for_writepage(wbc, page);
446 			break;
447 		}
448 		clear_buffer_dirty(bh);
449 	}
450 	unlock_page(page);
451 	/*
452 	 * If the page was truncated before we could do the writeback,
453 	 * or we had a memory allocation error while trying to write
454 	 * the first buffer head, we won't have submitted any pages for
455 	 * I/O.  In that case we need to make sure we've cleared the
456 	 * PageWriteback bit from the page to prevent the system from
457 	 * wedging later on.
458 	 */
459 	put_io_page(io_page);
460 	return ret;
461 }
462