blob: b24a2541a9baaa0d4c22e80a75050af2517a417d [file] [log] [blame]
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001/*
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
Theodore Ts'obd2d0212010-10-27 21:30:10 -04009#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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070021#include <linux/aio.h>
Theodore Ts'obd2d0212010-10-27 21:30:10 -040022#include <linux/uio.h>
23#include <linux/bio.h>
24#include <linux/workqueue.h>
25#include <linux/kernel.h>
26#include <linux/slab.h>
Jan Kara1ae48a62013-01-28 09:32:54 -050027#include <linux/mm.h>
Anatol Pomozove8974c32013-07-11 22:42:42 -040028#include <linux/ratelimit.h>
Theodore Ts'obd2d0212010-10-27 21:30:10 -040029
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
Theodore Ts'obd2d0212010-10-27 21:30:10 -040033
Jan Kara0058f962013-04-11 23:48:32 -040034static struct kmem_cache *io_end_cachep;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040035
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040036int __init ext4_init_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040037{
Theodore Ts'obd2d0212010-10-27 21:30:10 -040038 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
Jan Kara0058f962013-04-11 23:48:32 -040039 if (io_end_cachep == NULL)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040040 return -ENOMEM;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040041 return 0;
42}
43
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040044void ext4_exit_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040045{
46 kmem_cache_destroy(io_end_cachep);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040047}
48
Theodore Ts'o1ada47d2013-03-20 09:39:42 -040049/*
Jan Karab0857d32013-06-04 14:23:41 -040050 * Print an buffer I/O error compatible with the fs/buffer.c. This
51 * provides compatibility with dmesg scrapers that look for a specific
52 * buffer I/O error message. We really need a unified error reporting
53 * structure to userspace ala Digital Unix's uerf system, but it's
54 * probably not going to happen in my lifetime, due to LKML politics...
55 */
56static void buffer_io_error(struct buffer_head *bh)
57{
58 char b[BDEVNAME_SIZE];
Anatol Pomozove8974c32013-07-11 22:42:42 -040059 printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
Jan Karab0857d32013-06-04 14:23:41 -040060 bdevname(bh->b_bdev, b),
61 (unsigned long long)bh->b_blocknr);
62}
63
64static void ext4_finish_bio(struct bio *bio)
65{
66 int i;
67 int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
Kent Overstreet2c30c712013-11-07 12:20:26 -080068 struct bio_vec *bvec;
Jan Karab0857d32013-06-04 14:23:41 -040069
Kent Overstreet2c30c712013-11-07 12:20:26 -080070 bio_for_each_segment_all(bvec, bio, i) {
Jan Karab0857d32013-06-04 14:23:41 -040071 struct page *page = bvec->bv_page;
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 if (error) {
82 SetPageError(page);
83 set_bit(AS_EIO, &page->mapping->flags);
84 }
85 bh = head = page_buffers(page);
86 /*
87 * We check all buffers in the page under BH_Uptodate_Lock
88 * to avoid races with other end io clearing async_write flags
89 */
90 local_irq_save(flags);
91 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
92 do {
93 if (bh_offset(bh) < bio_start ||
94 bh_offset(bh) + bh->b_size > bio_end) {
95 if (buffer_async_write(bh))
96 under_io++;
97 continue;
98 }
99 clear_buffer_async_write(bh);
100 if (error)
101 buffer_io_error(bh);
102 } while ((bh = bh->b_this_page) != head);
103 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
104 local_irq_restore(flags);
105 if (!under_io)
106 end_page_writeback(page);
107 }
108}
109
Jan Kara97a851e2013-06-04 11:58:58 -0400110static void ext4_release_io_end(ext4_io_end_t *io_end)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400111{
Jan Karab0857d32013-06-04 14:23:41 -0400112 struct bio *bio, *next_bio;
113
Jan Kara97a851e2013-06-04 11:58:58 -0400114 BUG_ON(!list_empty(&io_end->list));
115 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
Jan Kara6b523df2013-06-04 13:21:11 -0400116 WARN_ON(io_end->handle);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400117
Jan Kara97a851e2013-06-04 11:58:58 -0400118 if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
119 wake_up_all(ext4_ioend_wq(io_end->inode));
Jan Karab0857d32013-06-04 14:23:41 -0400120
121 for (bio = io_end->bio; bio; bio = next_bio) {
122 next_bio = bio->bi_private;
123 ext4_finish_bio(bio);
124 bio_put(bio);
125 }
Jan Kara97a851e2013-06-04 11:58:58 -0400126 kmem_cache_free(io_end_cachep, io_end);
127}
128
129static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
130{
131 struct inode *inode = io_end->inode;
132
133 io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
134 /* Wake up anyone waiting on unwritten extent conversion */
135 if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
136 wake_up_all(ext4_ioend_wq(inode));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400137}
138
Jan Karaa115f742013-06-04 14:30:00 -0400139/*
140 * Check a range of space and convert unwritten extents to written. Note that
141 * we are protected from truncate touching same part of extent tree by the
142 * fact that truncate code waits for all DIO to finish (thus exclusion from
143 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
144 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
145 * completed (happens from ext4_free_ioend()).
146 */
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400147static int ext4_end_io(ext4_io_end_t *io)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400148{
149 struct inode *inode = io->inode;
150 loff_t offset = io->offset;
151 ssize_t size = io->size;
Jan Kara6b523df2013-06-04 13:21:11 -0400152 handle_t *handle = io->handle;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400153 int ret = 0;
154
155 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
156 "list->prev 0x%p\n",
157 io, inode->i_ino, io->list.next, io->list.prev);
158
Jan Kara6b523df2013-06-04 13:21:11 -0400159 io->handle = NULL; /* Following call will use up the handle */
160 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400161 if (ret < 0) {
Theodore Ts'ob82e3842011-10-31 10:56:32 -0400162 ext4_msg(inode->i_sb, KERN_EMERG,
163 "failed to convert unwritten extents to written "
164 "extents -- potential data loss! "
165 "(inode %lu, offset %llu, size %zd, error %d)",
166 inode->i_ino, offset, size, ret);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400167 }
Jan Kara97a851e2013-06-04 11:58:58 -0400168 ext4_clear_io_unwritten_flag(io);
169 ext4_release_io_end(io);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400170 return ret;
171}
172
Jan Kara2e8fa542013-06-04 14:21:02 -0400173static void dump_completed_IO(struct inode *inode, struct list_head *head)
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400174{
175#ifdef EXT4FS_DEBUG
176 struct list_head *cur, *before, *after;
177 ext4_io_end_t *io, *io0, *io1;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400178
Jan Kara2e8fa542013-06-04 14:21:02 -0400179 if (list_empty(head))
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400180 return;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400181
Jan Kara2e8fa542013-06-04 14:21:02 -0400182 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
183 list_for_each_entry(io, head, list) {
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400184 cur = &io->list;
185 before = cur->prev;
186 io0 = container_of(before, ext4_io_end_t, list);
187 after = cur->next;
188 io1 = container_of(after, ext4_io_end_t, list);
189
190 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
191 io, inode->i_ino, io0, io1);
192 }
193#endif
194}
195
196/* Add the io_end to per-inode completed end_io list. */
Jan Kara97a851e2013-06-04 11:58:58 -0400197static void ext4_add_complete_io(ext4_io_end_t *io_end)
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400198{
199 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
Jan Kara78371a42013-10-16 08:25:11 -0400200 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400201 struct workqueue_struct *wq;
202 unsigned long flags;
203
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200204 /* Only reserved conversions from writeback should enter here */
205 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
Jan Kara78371a42013-10-16 08:25:11 -0400206 WARN_ON(!io_end->handle && sbi->s_journal);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400207 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
Jan Kara78371a42013-10-16 08:25:11 -0400208 wq = sbi->rsv_conversion_wq;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200209 if (list_empty(&ei->i_rsv_conversion_list))
210 queue_work(wq, &ei->i_rsv_conversion_work);
211 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400212 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
213}
214
Jan Kara2e8fa542013-06-04 14:21:02 -0400215static int ext4_do_flush_completed_IO(struct inode *inode,
216 struct list_head *head)
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400217{
218 ext4_io_end_t *io;
Jan Kara002bd7f2013-01-28 09:49:15 -0500219 struct list_head unwritten;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400220 unsigned long flags;
221 struct ext4_inode_info *ei = EXT4_I(inode);
222 int err, ret = 0;
223
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400224 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
Jan Kara2e8fa542013-06-04 14:21:02 -0400225 dump_completed_IO(inode, head);
226 list_replace_init(head, &unwritten);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400227 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
228
229 while (!list_empty(&unwritten)) {
230 io = list_entry(unwritten.next, ext4_io_end_t, list);
231 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
232 list_del_init(&io->list);
233
234 err = ext4_end_io(io);
235 if (unlikely(!ret && err))
236 ret = err;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400237 }
238 return ret;
239}
240
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400241/*
Jan Kara2e8fa542013-06-04 14:21:02 -0400242 * work on completed IO, to convert unwritten extents to extents
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400243 */
Jan Kara2e8fa542013-06-04 14:21:02 -0400244void ext4_end_io_rsv_work(struct work_struct *work)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400245{
Jan Kara84c17542013-01-28 09:43:46 -0500246 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
Jan Kara2e8fa542013-06-04 14:21:02 -0400247 i_rsv_conversion_work);
248 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
249}
250
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400251ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
252{
Jesper Juhlb17b35e2010-12-19 21:41:55 -0500253 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400254 if (io) {
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500255 atomic_inc(&EXT4_I(inode)->i_ioend_count);
256 io->inode = inode;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400257 INIT_LIST_HEAD(&io->list);
Jan Kara97a851e2013-06-04 11:58:58 -0400258 atomic_set(&io->count, 1);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400259 }
260 return io;
261}
262
Jan Kara97a851e2013-06-04 11:58:58 -0400263void 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
274int 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) {
Jan Kara6b523df2013-06-04 13:21:11 -0400280 err = ext4_convert_unwritten_extents(io_end->handle,
281 io_end->inode, io_end->offset,
282 io_end->size);
283 io_end->handle = NULL;
Jan Kara97a851e2013-06-04 11:58:58 -0400284 ext4_clear_io_unwritten_flag(io_end);
285 }
286 ext4_release_io_end(io_end);
287 }
288 return err;
289}
290
291ext4_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
Jan Kara822dbba2013-07-10 21:31:04 -0400297/* BIO completion function for page writeback */
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400298static void ext4_end_bio(struct bio *bio, int error)
299{
300 ext4_io_end_t *io_end = bio->bi_private;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700301 sector_t bi_sector = bio->bi_iter.bi_sector;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400302
303 BUG_ON(!io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400304 bio->bi_end_io = NULL;
305 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
306 error = 0;
Jan Kara0058f962013-04-11 23:48:32 -0400307
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500308 if (error) {
Jan Karab0857d32013-06-04 14:23:41 -0400309 struct inode *inode = io_end->inode;
310
Matthew Wilcox9503c672014-04-07 10:54:20 -0400311 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500312 "(offset %llu size %ld starting block %llu)",
Matthew Wilcox9503c672014-04-07 10:54:20 -0400313 error, inode->i_ino,
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500314 (unsigned long long) io_end->offset,
315 (long) io_end->size,
316 (unsigned long long)
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500317 bi_sector >> (inode->i_blkbits - 9));
Matthew Wilcox9503c672014-04-07 10:54:20 -0400318 mapping_set_error(inode->i_mapping, error);
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500319 }
Jan Kara822dbba2013-07-10 21:31:04 -0400320
321 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
322 /*
323 * Link bio into list hanging from io_end. We have to do it
324 * atomically as bio completions can be racing against each
325 * other.
326 */
327 bio->bi_private = xchg(&io_end->bio, bio);
328 ext4_put_io_end_defer(io_end);
329 } else {
330 /*
331 * Drop io_end reference early. Inode can get freed once
332 * we finish the bio.
333 */
334 ext4_put_io_end_defer(io_end);
335 ext4_finish_bio(bio);
336 bio_put(bio);
337 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400338}
339
340void ext4_io_submit(struct ext4_io_submit *io)
341{
342 struct bio *bio = io->io_bio;
343
344 if (bio) {
345 bio_get(io->io_bio);
346 submit_bio(io->io_op, io->io_bio);
347 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
348 bio_put(io->io_bio);
349 }
Peter Huewe7dc57612011-02-21 21:01:42 -0500350 io->io_bio = NULL;
Jan Kara97a851e2013-06-04 11:58:58 -0400351}
352
353void ext4_io_submit_init(struct ext4_io_submit *io,
354 struct writeback_control *wbc)
355{
356 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
357 io->io_bio = NULL;
Peter Huewe7dc57612011-02-21 21:01:42 -0500358 io->io_end = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400359}
360
Jan Kara97a851e2013-06-04 11:58:58 -0400361static int io_submit_init_bio(struct ext4_io_submit *io,
362 struct buffer_head *bh)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400363{
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400364 int nvecs = bio_get_nr_vecs(bh->b_bdev);
365 struct bio *bio;
366
Theodore Ts'o275d3ba2011-06-29 21:44:45 -0400367 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
Theodore Ts'oa1d8d9a72013-06-06 10:18:22 -0400368 if (!bio)
369 return -ENOMEM;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700370 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400371 bio->bi_bdev = bh->b_bdev;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400372 bio->bi_end_io = ext4_end_bio;
Jan Kara97a851e2013-06-04 11:58:58 -0400373 bio->bi_private = ext4_get_io_end(io->io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400374 io->io_bio = bio;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400375 io->io_next_block = bh->b_blocknr;
376 return 0;
377}
378
379static int io_submit_add_bh(struct ext4_io_submit *io,
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400380 struct inode *inode,
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400381 struct buffer_head *bh)
382{
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400383 int ret;
384
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400385 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
386submit_and_retry:
387 ext4_io_submit(io);
388 }
389 if (io->io_bio == NULL) {
Jan Kara97a851e2013-06-04 11:58:58 -0400390 ret = io_submit_init_bio(io, bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400391 if (ret)
392 return ret;
393 }
Theodore Ts'oa5499842013-05-11 19:07:42 -0400394 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
395 if (ret != bh->b_size)
396 goto submit_and_retry;
Jan Kara97a851e2013-06-04 11:58:58 -0400397 io->io_next_block++;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400398 return 0;
399}
400
401int ext4_bio_write_page(struct ext4_io_submit *io,
402 struct page *page,
403 int len,
Namjae Jeon1c8349a2014-05-12 08:12:25 -0400404 struct writeback_control *wbc,
405 bool keep_towrite)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400406{
407 struct inode *inode = page->mapping->host;
Jan Kara0058f962013-04-11 23:48:32 -0400408 unsigned block_start, blocksize;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400409 struct buffer_head *bh, *head;
410 int ret = 0;
Jan Kara0058f962013-04-11 23:48:32 -0400411 int nr_submitted = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400412
413 blocksize = 1 << inode->i_blkbits;
414
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500415 BUG_ON(!PageLocked(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400416 BUG_ON(PageWriteback(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400417
Namjae Jeon1c8349a2014-05-12 08:12:25 -0400418 if (keep_towrite)
419 set_page_writeback_keepwrite(page);
420 else
421 set_page_writeback(page);
Theodore Ts'oa54aa762011-02-27 16:43:24 -0500422 ClearPageError(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400423
Jan Kara0058f962013-04-11 23:48:32 -0400424 /*
Linus Torvaldsf8409ab2014-06-08 13:03:35 -0700425 * Comments copied from block_write_full_page:
Jan Karaeeece462014-05-27 12:48:55 -0400426 *
427 * The page straddles i_size. It must be zeroed out on each and every
428 * writepage invocation because it may be mmapped. "A file is mapped
429 * in multiples of the page size. For a file that is not a multiple of
430 * the page size, the remaining memory is zeroed when mapped, and
431 * writes to that region are not written out to the file."
432 */
433 if (len < PAGE_CACHE_SIZE)
434 zero_user_segment(page, len, PAGE_CACHE_SIZE);
435 /*
Jan Kara0058f962013-04-11 23:48:32 -0400436 * In the first loop we prepare and mark buffers to submit. We have to
437 * mark all buffers in the page before submitting so that
438 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
439 * on the first buffer finishes and we are still working on submitting
440 * the second buffer.
441 */
442 bh = head = page_buffers(page);
443 do {
444 block_start = bh_offset(bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400445 if (block_start >= len) {
446 clear_buffer_dirty(bh);
447 set_buffer_uptodate(bh);
448 continue;
449 }
Jan Kara8a850c32013-01-28 20:53:28 -0500450 if (!buffer_dirty(bh) || buffer_delay(bh) ||
451 !buffer_mapped(bh) || buffer_unwritten(bh)) {
452 /* A hole? We can safely clear the dirty bit */
453 if (!buffer_mapped(bh))
454 clear_buffer_dirty(bh);
455 if (io->io_bio)
456 ext4_io_submit(io);
457 continue;
458 }
Jan Kara0058f962013-04-11 23:48:32 -0400459 if (buffer_new(bh)) {
460 clear_buffer_new(bh);
461 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
462 }
463 set_buffer_async_write(bh);
464 } while ((bh = bh->b_this_page) != head);
465
466 /* Now submit buffers to write */
467 bh = head = page_buffers(page);
468 do {
469 if (!buffer_async_write(bh))
470 continue;
Jan Kara97a851e2013-06-04 11:58:58 -0400471 ret = io_submit_add_bh(io, inode, bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400472 if (ret) {
473 /*
474 * We only get here on ENOMEM. Not much else
475 * we can do but mark the page as dirty, and
476 * better luck next time.
477 */
Jan Kara1ae48a62013-01-28 09:32:54 -0500478 redirty_page_for_writepage(wbc, page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400479 break;
480 }
Jan Kara0058f962013-04-11 23:48:32 -0400481 nr_submitted++;
Jan Kara1ae48a62013-01-28 09:32:54 -0500482 clear_buffer_dirty(bh);
Jan Kara0058f962013-04-11 23:48:32 -0400483 } while ((bh = bh->b_this_page) != head);
484
485 /* Error stopped previous loop? Clean up buffers... */
486 if (ret) {
487 do {
488 clear_buffer_async_write(bh);
489 bh = bh->b_this_page;
490 } while (bh != head);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400491 }
492 unlock_page(page);
Jan Kara0058f962013-04-11 23:48:32 -0400493 /* Nothing submitted - we have to end page writeback */
494 if (!nr_submitted)
495 end_page_writeback(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400496 return ret;
497}