Loading...
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
27#include "ext4_jbd2.h"
28#include "xattr.h"
29#include "acl.h"
30#include "ext4_extents.h"
31
32static struct kmem_cache *io_page_cachep, *io_end_cachep;
33
34int __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
47void ext4_exit_pageio(void)
48{
49 kmem_cache_destroy(io_end_cachep);
50 kmem_cache_destroy(io_page_cachep);
51}
52
53void 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
60static 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
69void ext4_free_io_end(ext4_io_end_t *io)
70{
71 int i;
72
73 BUG_ON(!io);
74 if (io->page)
75 put_page(io->page);
76 for (i = 0; i < io->num_io_pages; i++)
77 put_io_page(io->pages[i]);
78 io->num_io_pages = 0;
79 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count))
80 wake_up_all(ext4_ioend_wq(io->inode));
81 kmem_cache_free(io_end_cachep, io);
82}
83
84/*
85 * check a range of space and convert unwritten extents to written.
86 *
87 * Called with inode->i_mutex; we depend on this when we manipulate
88 * io->flag, since we could otherwise race with ext4_flush_completed_IO()
89 */
90int ext4_end_io_nolock(ext4_io_end_t *io)
91{
92 struct inode *inode = io->inode;
93 loff_t offset = io->offset;
94 ssize_t size = io->size;
95 int ret = 0;
96
97 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
98 "list->prev 0x%p\n",
99 io, inode->i_ino, io->list.next, io->list.prev);
100
101 ret = ext4_convert_unwritten_extents(inode, offset, size);
102 if (ret < 0) {
103 ext4_msg(inode->i_sb, KERN_EMERG,
104 "failed to convert unwritten extents to written "
105 "extents -- potential data loss! "
106 "(inode %lu, offset %llu, size %zd, error %d)",
107 inode->i_ino, offset, size, ret);
108 }
109
110 if (io->iocb)
111 aio_complete(io->iocb, io->result, 0);
112
113 if (io->flag & EXT4_IO_END_DIRECT)
114 inode_dio_done(inode);
115 /* Wake up anyone waiting on unwritten extent conversion */
116 if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten))
117 wake_up_all(ext4_ioend_wq(io->inode));
118 return ret;
119}
120
121/*
122 * work on completed aio dio IO, to convert unwritten extents to extents
123 */
124static void ext4_end_io_work(struct work_struct *work)
125{
126 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
127 struct inode *inode = io->inode;
128 struct ext4_inode_info *ei = EXT4_I(inode);
129 unsigned long flags;
130
131 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
132 if (io->flag & EXT4_IO_END_IN_FSYNC)
133 goto requeue;
134 if (list_empty(&io->list)) {
135 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
136 goto free;
137 }
138
139 if (!mutex_trylock(&inode->i_mutex)) {
140 bool was_queued;
141requeue:
142 was_queued = !!(io->flag & EXT4_IO_END_QUEUED);
143 io->flag |= EXT4_IO_END_QUEUED;
144 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
145 /*
146 * Requeue the work instead of waiting so that the work
147 * items queued after this can be processed.
148 */
149 queue_work(EXT4_SB(inode->i_sb)->dio_unwritten_wq, &io->work);
150 /*
151 * To prevent the ext4-dio-unwritten thread from keeping
152 * requeueing end_io requests and occupying cpu for too long,
153 * yield the cpu if it sees an end_io request that has already
154 * been requeued.
155 */
156 if (was_queued)
157 yield();
158 return;
159 }
160 list_del_init(&io->list);
161 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
162 (void) ext4_end_io_nolock(io);
163 mutex_unlock(&inode->i_mutex);
164free:
165 ext4_free_io_end(io);
166}
167
168ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
169{
170 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
171 if (io) {
172 atomic_inc(&EXT4_I(inode)->i_ioend_count);
173 io->inode = inode;
174 INIT_WORK(&io->work, ext4_end_io_work);
175 INIT_LIST_HEAD(&io->list);
176 }
177 return io;
178}
179
180/*
181 * Print an buffer I/O error compatible with the fs/buffer.c. This
182 * provides compatibility with dmesg scrapers that look for a specific
183 * buffer I/O error message. We really need a unified error reporting
184 * structure to userspace ala Digital Unix's uerf system, but it's
185 * probably not going to happen in my lifetime, due to LKML politics...
186 */
187static void buffer_io_error(struct buffer_head *bh)
188{
189 char b[BDEVNAME_SIZE];
190 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
191 bdevname(bh->b_bdev, b),
192 (unsigned long long)bh->b_blocknr);
193}
194
195static void ext4_end_bio(struct bio *bio, int error)
196{
197 ext4_io_end_t *io_end = bio->bi_private;
198 struct workqueue_struct *wq;
199 struct inode *inode;
200 unsigned long flags;
201 int i;
202 sector_t bi_sector = bio->bi_sector;
203
204 BUG_ON(!io_end);
205 bio->bi_private = NULL;
206 bio->bi_end_io = NULL;
207 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
208 error = 0;
209 bio_put(bio);
210
211 for (i = 0; i < io_end->num_io_pages; i++) {
212 struct page *page = io_end->pages[i]->p_page;
213 struct buffer_head *bh, *head;
214 loff_t offset;
215 loff_t io_end_offset;
216
217 if (error) {
218 SetPageError(page);
219 set_bit(AS_EIO, &page->mapping->flags);
220 head = page_buffers(page);
221 BUG_ON(!head);
222
223 io_end_offset = io_end->offset + io_end->size;
224
225 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
226 bh = head;
227 do {
228 if ((offset >= io_end->offset) &&
229 (offset+bh->b_size <= io_end_offset))
230 buffer_io_error(bh);
231
232 offset += bh->b_size;
233 bh = bh->b_this_page;
234 } while (bh != head);
235 }
236
237 put_io_page(io_end->pages[i]);
238 }
239 io_end->num_io_pages = 0;
240 inode = io_end->inode;
241
242 if (error) {
243 io_end->flag |= EXT4_IO_END_ERROR;
244 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
245 "(offset %llu size %ld starting block %llu)",
246 inode->i_ino,
247 (unsigned long long) io_end->offset,
248 (long) io_end->size,
249 (unsigned long long)
250 bi_sector >> (inode->i_blkbits - 9));
251 }
252
253 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
254 ext4_free_io_end(io_end);
255 return;
256 }
257
258 /* Add the io_end to per-inode completed io list*/
259 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
260 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
261 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
262
263 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
264 /* queue the work to convert unwritten extents to written */
265 queue_work(wq, &io_end->work);
266}
267
268void ext4_io_submit(struct ext4_io_submit *io)
269{
270 struct bio *bio = io->io_bio;
271
272 if (bio) {
273 bio_get(io->io_bio);
274 submit_bio(io->io_op, io->io_bio);
275 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
276 bio_put(io->io_bio);
277 }
278 io->io_bio = NULL;
279 io->io_op = 0;
280 io->io_end = NULL;
281}
282
283static int io_submit_init(struct ext4_io_submit *io,
284 struct inode *inode,
285 struct writeback_control *wbc,
286 struct buffer_head *bh)
287{
288 ext4_io_end_t *io_end;
289 struct page *page = bh->b_page;
290 int nvecs = bio_get_nr_vecs(bh->b_bdev);
291 struct bio *bio;
292
293 io_end = ext4_init_io_end(inode, GFP_NOFS);
294 if (!io_end)
295 return -ENOMEM;
296 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
297 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
298 bio->bi_bdev = bh->b_bdev;
299 bio->bi_private = io->io_end = io_end;
300 bio->bi_end_io = ext4_end_bio;
301
302 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
303
304 io->io_bio = bio;
305 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
306 io->io_next_block = bh->b_blocknr;
307 return 0;
308}
309
310static int io_submit_add_bh(struct ext4_io_submit *io,
311 struct ext4_io_page *io_page,
312 struct inode *inode,
313 struct writeback_control *wbc,
314 struct buffer_head *bh)
315{
316 ext4_io_end_t *io_end;
317 int ret;
318
319 if (buffer_new(bh)) {
320 clear_buffer_new(bh);
321 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
322 }
323
324 if (!buffer_mapped(bh) || buffer_delay(bh)) {
325 if (!buffer_mapped(bh))
326 clear_buffer_dirty(bh);
327 if (io->io_bio)
328 ext4_io_submit(io);
329 return 0;
330 }
331
332 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
333submit_and_retry:
334 ext4_io_submit(io);
335 }
336 if (io->io_bio == NULL) {
337 ret = io_submit_init(io, inode, wbc, bh);
338 if (ret)
339 return ret;
340 }
341 io_end = io->io_end;
342 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
343 (io_end->pages[io_end->num_io_pages-1] != io_page))
344 goto submit_and_retry;
345 if (buffer_uninit(bh))
346 ext4_set_io_unwritten_flag(inode, io_end);
347 io->io_end->size += bh->b_size;
348 io->io_next_block++;
349 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
350 if (ret != bh->b_size)
351 goto submit_and_retry;
352 if ((io_end->num_io_pages == 0) ||
353 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
354 io_end->pages[io_end->num_io_pages++] = io_page;
355 atomic_inc(&io_page->p_count);
356 }
357 return 0;
358}
359
360int ext4_bio_write_page(struct ext4_io_submit *io,
361 struct page *page,
362 int len,
363 struct writeback_control *wbc)
364{
365 struct inode *inode = page->mapping->host;
366 unsigned block_start, block_end, blocksize;
367 struct ext4_io_page *io_page;
368 struct buffer_head *bh, *head;
369 int ret = 0;
370
371 blocksize = 1 << inode->i_blkbits;
372
373 BUG_ON(!PageLocked(page));
374 BUG_ON(PageWriteback(page));
375
376 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
377 if (!io_page) {
378 set_page_dirty(page);
379 unlock_page(page);
380 return -ENOMEM;
381 }
382 io_page->p_page = page;
383 atomic_set(&io_page->p_count, 1);
384 get_page(page);
385 set_page_writeback(page);
386 ClearPageError(page);
387
388 for (bh = head = page_buffers(page), block_start = 0;
389 bh != head || !block_start;
390 block_start = block_end, bh = bh->b_this_page) {
391
392 block_end = block_start + blocksize;
393 if (block_start >= len) {
394 /*
395 * Comments copied from block_write_full_page_endio:
396 *
397 * The page straddles i_size. It must be zeroed out on
398 * each and every writepage invocation because it may
399 * be mmapped. "A file is mapped in multiples of the
400 * page size. For a file that is not a multiple of
401 * the page size, the remaining memory is zeroed when
402 * mapped, and writes to that region are not written
403 * out to the file."
404 */
405 zero_user_segment(page, block_start, block_end);
406 clear_buffer_dirty(bh);
407 set_buffer_uptodate(bh);
408 continue;
409 }
410 clear_buffer_dirty(bh);
411 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
412 if (ret) {
413 /*
414 * We only get here on ENOMEM. Not much else
415 * we can do but mark the page as dirty, and
416 * better luck next time.
417 */
418 set_page_dirty(page);
419 break;
420 }
421 }
422 unlock_page(page);
423 /*
424 * If the page was truncated before we could do the writeback,
425 * or we had a memory allocation error while trying to write
426 * the first buffer head, we won't have submitted any pages for
427 * I/O. In that case we need to make sure we've cleared the
428 * PageWriteback bit from the page to prevent the system from
429 * wedging later on.
430 */
431 put_io_page(io_page);
432 return ret;
433}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/page-io.c
4 *
5 * This contains the new page_io functions for ext4
6 *
7 * Written by Theodore Ts'o, 2010.
8 */
9
10#include <linux/fs.h>
11#include <linux/time.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#include <linux/backing-dev.h>
28
29#include "ext4_jbd2.h"
30#include "xattr.h"
31#include "acl.h"
32
33static struct kmem_cache *io_end_cachep;
34static struct kmem_cache *io_end_vec_cachep;
35
36int __init ext4_init_pageio(void)
37{
38 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
39 if (io_end_cachep == NULL)
40 return -ENOMEM;
41
42 io_end_vec_cachep = KMEM_CACHE(ext4_io_end_vec, 0);
43 if (io_end_vec_cachep == NULL) {
44 kmem_cache_destroy(io_end_cachep);
45 return -ENOMEM;
46 }
47 return 0;
48}
49
50void ext4_exit_pageio(void)
51{
52 kmem_cache_destroy(io_end_cachep);
53 kmem_cache_destroy(io_end_vec_cachep);
54}
55
56struct ext4_io_end_vec *ext4_alloc_io_end_vec(ext4_io_end_t *io_end)
57{
58 struct ext4_io_end_vec *io_end_vec;
59
60 io_end_vec = kmem_cache_zalloc(io_end_vec_cachep, GFP_NOFS);
61 if (!io_end_vec)
62 return ERR_PTR(-ENOMEM);
63 INIT_LIST_HEAD(&io_end_vec->list);
64 list_add_tail(&io_end_vec->list, &io_end->list_vec);
65 return io_end_vec;
66}
67
68static void ext4_free_io_end_vec(ext4_io_end_t *io_end)
69{
70 struct ext4_io_end_vec *io_end_vec, *tmp;
71
72 if (list_empty(&io_end->list_vec))
73 return;
74 list_for_each_entry_safe(io_end_vec, tmp, &io_end->list_vec, list) {
75 list_del(&io_end_vec->list);
76 kmem_cache_free(io_end_vec_cachep, io_end_vec);
77 }
78}
79
80struct ext4_io_end_vec *ext4_last_io_end_vec(ext4_io_end_t *io_end)
81{
82 BUG_ON(list_empty(&io_end->list_vec));
83 return list_last_entry(&io_end->list_vec, struct ext4_io_end_vec, list);
84}
85
86/*
87 * Print an buffer I/O error compatible with the fs/buffer.c. This
88 * provides compatibility with dmesg scrapers that look for a specific
89 * buffer I/O error message. We really need a unified error reporting
90 * structure to userspace ala Digital Unix's uerf system, but it's
91 * probably not going to happen in my lifetime, due to LKML politics...
92 */
93static void buffer_io_error(struct buffer_head *bh)
94{
95 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
96 bh->b_bdev,
97 (unsigned long long)bh->b_blocknr);
98}
99
100static void ext4_finish_bio(struct bio *bio)
101{
102 struct bio_vec *bvec;
103 struct bvec_iter_all iter_all;
104
105 bio_for_each_segment_all(bvec, bio, iter_all) {
106 struct page *page = bvec->bv_page;
107 struct page *bounce_page = NULL;
108 struct buffer_head *bh, *head;
109 unsigned bio_start = bvec->bv_offset;
110 unsigned bio_end = bio_start + bvec->bv_len;
111 unsigned under_io = 0;
112 unsigned long flags;
113
114 if (!page)
115 continue;
116
117 if (fscrypt_is_bounce_page(page)) {
118 bounce_page = page;
119 page = fscrypt_pagecache_page(bounce_page);
120 }
121
122 if (bio->bi_status) {
123 SetPageError(page);
124 mapping_set_error(page->mapping, -EIO);
125 }
126 bh = head = page_buffers(page);
127 /*
128 * We check all buffers in the page under b_uptodate_lock
129 * to avoid races with other end io clearing async_write flags
130 */
131 spin_lock_irqsave(&head->b_uptodate_lock, flags);
132 do {
133 if (bh_offset(bh) < bio_start ||
134 bh_offset(bh) + bh->b_size > bio_end) {
135 if (buffer_async_write(bh))
136 under_io++;
137 continue;
138 }
139 clear_buffer_async_write(bh);
140 if (bio->bi_status)
141 buffer_io_error(bh);
142 } while ((bh = bh->b_this_page) != head);
143 spin_unlock_irqrestore(&head->b_uptodate_lock, flags);
144 if (!under_io) {
145 fscrypt_free_bounce_page(bounce_page);
146 end_page_writeback(page);
147 }
148 }
149}
150
151static void ext4_release_io_end(ext4_io_end_t *io_end)
152{
153 struct bio *bio, *next_bio;
154
155 BUG_ON(!list_empty(&io_end->list));
156 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
157 WARN_ON(io_end->handle);
158
159 for (bio = io_end->bio; bio; bio = next_bio) {
160 next_bio = bio->bi_private;
161 ext4_finish_bio(bio);
162 bio_put(bio);
163 }
164 ext4_free_io_end_vec(io_end);
165 kmem_cache_free(io_end_cachep, io_end);
166}
167
168/*
169 * Check a range of space and convert unwritten extents to written. Note that
170 * we are protected from truncate touching same part of extent tree by the
171 * fact that truncate code waits for all DIO to finish (thus exclusion from
172 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
173 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
174 * completed (happens from ext4_free_ioend()).
175 */
176static int ext4_end_io_end(ext4_io_end_t *io_end)
177{
178 struct inode *inode = io_end->inode;
179 handle_t *handle = io_end->handle;
180 int ret = 0;
181
182 ext4_debug("ext4_end_io_nolock: io_end 0x%p from inode %lu,list->next 0x%p,"
183 "list->prev 0x%p\n",
184 io_end, inode->i_ino, io_end->list.next, io_end->list.prev);
185
186 io_end->handle = NULL; /* Following call will use up the handle */
187 ret = ext4_convert_unwritten_io_end_vec(handle, io_end);
188 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
189 ext4_msg(inode->i_sb, KERN_EMERG,
190 "failed to convert unwritten extents to written "
191 "extents -- potential data loss! "
192 "(inode %lu, error %d)", inode->i_ino, ret);
193 }
194 ext4_clear_io_unwritten_flag(io_end);
195 ext4_release_io_end(io_end);
196 return ret;
197}
198
199static void dump_completed_IO(struct inode *inode, struct list_head *head)
200{
201#ifdef EXT4FS_DEBUG
202 struct list_head *cur, *before, *after;
203 ext4_io_end_t *io_end, *io_end0, *io_end1;
204
205 if (list_empty(head))
206 return;
207
208 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
209 list_for_each_entry(io_end, head, list) {
210 cur = &io_end->list;
211 before = cur->prev;
212 io_end0 = container_of(before, ext4_io_end_t, list);
213 after = cur->next;
214 io_end1 = container_of(after, ext4_io_end_t, list);
215
216 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
217 io_end, inode->i_ino, io_end0, io_end1);
218 }
219#endif
220}
221
222/* Add the io_end to per-inode completed end_io list. */
223static void ext4_add_complete_io(ext4_io_end_t *io_end)
224{
225 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
226 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
227 struct workqueue_struct *wq;
228 unsigned long flags;
229
230 /* Only reserved conversions from writeback should enter here */
231 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
232 WARN_ON(!io_end->handle && sbi->s_journal);
233 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
234 wq = sbi->rsv_conversion_wq;
235 if (list_empty(&ei->i_rsv_conversion_list))
236 queue_work(wq, &ei->i_rsv_conversion_work);
237 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
238 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
239}
240
241static int ext4_do_flush_completed_IO(struct inode *inode,
242 struct list_head *head)
243{
244 ext4_io_end_t *io_end;
245 struct list_head unwritten;
246 unsigned long flags;
247 struct ext4_inode_info *ei = EXT4_I(inode);
248 int err, ret = 0;
249
250 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
251 dump_completed_IO(inode, head);
252 list_replace_init(head, &unwritten);
253 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
254
255 while (!list_empty(&unwritten)) {
256 io_end = list_entry(unwritten.next, ext4_io_end_t, list);
257 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
258 list_del_init(&io_end->list);
259
260 err = ext4_end_io_end(io_end);
261 if (unlikely(!ret && err))
262 ret = err;
263 }
264 return ret;
265}
266
267/*
268 * work on completed IO, to convert unwritten extents to extents
269 */
270void ext4_end_io_rsv_work(struct work_struct *work)
271{
272 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
273 i_rsv_conversion_work);
274 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
275}
276
277ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
278{
279 ext4_io_end_t *io_end = kmem_cache_zalloc(io_end_cachep, flags);
280
281 if (io_end) {
282 io_end->inode = inode;
283 INIT_LIST_HEAD(&io_end->list);
284 INIT_LIST_HEAD(&io_end->list_vec);
285 atomic_set(&io_end->count, 1);
286 }
287 return io_end;
288}
289
290void ext4_put_io_end_defer(ext4_io_end_t *io_end)
291{
292 if (atomic_dec_and_test(&io_end->count)) {
293 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) ||
294 list_empty(&io_end->list_vec)) {
295 ext4_release_io_end(io_end);
296 return;
297 }
298 ext4_add_complete_io(io_end);
299 }
300}
301
302int ext4_put_io_end(ext4_io_end_t *io_end)
303{
304 int err = 0;
305
306 if (atomic_dec_and_test(&io_end->count)) {
307 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
308 err = ext4_convert_unwritten_io_end_vec(io_end->handle,
309 io_end);
310 io_end->handle = NULL;
311 ext4_clear_io_unwritten_flag(io_end);
312 }
313 ext4_release_io_end(io_end);
314 }
315 return err;
316}
317
318ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
319{
320 atomic_inc(&io_end->count);
321 return io_end;
322}
323
324/* BIO completion function for page writeback */
325static void ext4_end_bio(struct bio *bio)
326{
327 ext4_io_end_t *io_end = bio->bi_private;
328 sector_t bi_sector = bio->bi_iter.bi_sector;
329 char b[BDEVNAME_SIZE];
330
331 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
332 bio_devname(bio, b),
333 (long long) bio->bi_iter.bi_sector,
334 (unsigned) bio_sectors(bio),
335 bio->bi_status)) {
336 ext4_finish_bio(bio);
337 bio_put(bio);
338 return;
339 }
340 bio->bi_end_io = NULL;
341
342 if (bio->bi_status) {
343 struct inode *inode = io_end->inode;
344
345 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
346 "starting block %llu)",
347 bio->bi_status, inode->i_ino,
348 (unsigned long long)
349 bi_sector >> (inode->i_blkbits - 9));
350 mapping_set_error(inode->i_mapping,
351 blk_status_to_errno(bio->bi_status));
352 }
353
354 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
355 /*
356 * Link bio into list hanging from io_end. We have to do it
357 * atomically as bio completions can be racing against each
358 * other.
359 */
360 bio->bi_private = xchg(&io_end->bio, bio);
361 ext4_put_io_end_defer(io_end);
362 } else {
363 /*
364 * Drop io_end reference early. Inode can get freed once
365 * we finish the bio.
366 */
367 ext4_put_io_end_defer(io_end);
368 ext4_finish_bio(bio);
369 bio_put(bio);
370 }
371}
372
373void ext4_io_submit(struct ext4_io_submit *io)
374{
375 struct bio *bio = io->io_bio;
376
377 if (bio) {
378 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
379 REQ_SYNC : 0;
380 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
381 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
382 submit_bio(io->io_bio);
383 }
384 io->io_bio = NULL;
385}
386
387void ext4_io_submit_init(struct ext4_io_submit *io,
388 struct writeback_control *wbc)
389{
390 io->io_wbc = wbc;
391 io->io_bio = NULL;
392 io->io_end = NULL;
393}
394
395static void io_submit_init_bio(struct ext4_io_submit *io,
396 struct buffer_head *bh)
397{
398 struct bio *bio;
399
400 /*
401 * bio_alloc will _always_ be able to allocate a bio if
402 * __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset().
403 */
404 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
405 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
406 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
407 bio_set_dev(bio, bh->b_bdev);
408 bio->bi_end_io = ext4_end_bio;
409 bio->bi_private = ext4_get_io_end(io->io_end);
410 io->io_bio = bio;
411 io->io_next_block = bh->b_blocknr;
412 wbc_init_bio(io->io_wbc, bio);
413}
414
415static void io_submit_add_bh(struct ext4_io_submit *io,
416 struct inode *inode,
417 struct page *page,
418 struct buffer_head *bh)
419{
420 int ret;
421
422 if (io->io_bio && (bh->b_blocknr != io->io_next_block ||
423 !fscrypt_mergeable_bio_bh(io->io_bio, bh))) {
424submit_and_retry:
425 ext4_io_submit(io);
426 }
427 if (io->io_bio == NULL) {
428 io_submit_init_bio(io, bh);
429 io->io_bio->bi_write_hint = inode->i_write_hint;
430 }
431 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
432 if (ret != bh->b_size)
433 goto submit_and_retry;
434 wbc_account_cgroup_owner(io->io_wbc, page, bh->b_size);
435 io->io_next_block++;
436}
437
438int ext4_bio_write_page(struct ext4_io_submit *io,
439 struct page *page,
440 int len,
441 struct writeback_control *wbc,
442 bool keep_towrite)
443{
444 struct page *bounce_page = NULL;
445 struct inode *inode = page->mapping->host;
446 unsigned block_start;
447 struct buffer_head *bh, *head;
448 int ret = 0;
449 int nr_submitted = 0;
450 int nr_to_submit = 0;
451
452 BUG_ON(!PageLocked(page));
453 BUG_ON(PageWriteback(page));
454
455 if (keep_towrite)
456 set_page_writeback_keepwrite(page);
457 else
458 set_page_writeback(page);
459 ClearPageError(page);
460
461 /*
462 * Comments copied from block_write_full_page:
463 *
464 * The page straddles i_size. It must be zeroed out on each and every
465 * writepage invocation because it may be mmapped. "A file is mapped
466 * in multiples of the page size. For a file that is not a multiple of
467 * the page size, the remaining memory is zeroed when mapped, and
468 * writes to that region are not written out to the file."
469 */
470 if (len < PAGE_SIZE)
471 zero_user_segment(page, len, PAGE_SIZE);
472 /*
473 * In the first loop we prepare and mark buffers to submit. We have to
474 * mark all buffers in the page before submitting so that
475 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
476 * on the first buffer finishes and we are still working on submitting
477 * the second buffer.
478 */
479 bh = head = page_buffers(page);
480 do {
481 block_start = bh_offset(bh);
482 if (block_start >= len) {
483 clear_buffer_dirty(bh);
484 set_buffer_uptodate(bh);
485 continue;
486 }
487 if (!buffer_dirty(bh) || buffer_delay(bh) ||
488 !buffer_mapped(bh) || buffer_unwritten(bh)) {
489 /* A hole? We can safely clear the dirty bit */
490 if (!buffer_mapped(bh))
491 clear_buffer_dirty(bh);
492 if (io->io_bio)
493 ext4_io_submit(io);
494 continue;
495 }
496 if (buffer_new(bh))
497 clear_buffer_new(bh);
498 set_buffer_async_write(bh);
499 nr_to_submit++;
500 } while ((bh = bh->b_this_page) != head);
501
502 bh = head = page_buffers(page);
503
504 /*
505 * If any blocks are being written to an encrypted file, encrypt them
506 * into a bounce page. For simplicity, just encrypt until the last
507 * block which might be needed. This may cause some unneeded blocks
508 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
509 * can't happen in the common case of blocksize == PAGE_SIZE.
510 */
511 if (fscrypt_inode_uses_fs_layer_crypto(inode) && nr_to_submit) {
512 gfp_t gfp_flags = GFP_NOFS;
513 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
514
515 /*
516 * Since bounce page allocation uses a mempool, we can only use
517 * a waiting mask (i.e. request guaranteed allocation) on the
518 * first page of the bio. Otherwise it can deadlock.
519 */
520 if (io->io_bio)
521 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
522 retry_encrypt:
523 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
524 0, gfp_flags);
525 if (IS_ERR(bounce_page)) {
526 ret = PTR_ERR(bounce_page);
527 if (ret == -ENOMEM &&
528 (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
529 gfp_flags = GFP_NOFS;
530 if (io->io_bio)
531 ext4_io_submit(io);
532 else
533 gfp_flags |= __GFP_NOFAIL;
534 congestion_wait(BLK_RW_ASYNC, HZ/50);
535 goto retry_encrypt;
536 }
537
538 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
539 redirty_page_for_writepage(wbc, page);
540 do {
541 clear_buffer_async_write(bh);
542 bh = bh->b_this_page;
543 } while (bh != head);
544 goto unlock;
545 }
546 }
547
548 /* Now submit buffers to write */
549 do {
550 if (!buffer_async_write(bh))
551 continue;
552 io_submit_add_bh(io, inode,
553 bounce_page ? bounce_page : page, bh);
554 nr_submitted++;
555 clear_buffer_dirty(bh);
556 } while ((bh = bh->b_this_page) != head);
557
558unlock:
559 unlock_page(page);
560 /* Nothing submitted - we have to end page writeback */
561 if (!nr_submitted)
562 end_page_writeback(page);
563 return ret;
564}