blob: a6132a73096740fc37286ea124d94e79d4206cc8 [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>
Theodore Ts'obd2d0212010-10-27 21:30:10 -040011#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>
Jan Kara1ae48a62013-01-28 09:32:54 -050025#include <linux/mm.h>
Theodore Ts'oc9af28f2016-03-26 16:14:34 -040026#include <linux/backing-dev.h>
Jaegeuk Kima7550b32016-07-10 14:01:03 -040027#include <linux/fscrypto.h>
Theodore Ts'obd2d0212010-10-27 21:30:10 -040028
29#include "ext4_jbd2.h"
30#include "xattr.h"
31#include "acl.h"
Theodore Ts'obd2d0212010-10-27 21:30:10 -040032
Jan Kara0058f962013-04-11 23:48:32 -040033static struct kmem_cache *io_end_cachep;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040034
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040035int __init ext4_init_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040036{
Theodore Ts'obd2d0212010-10-27 21:30:10 -040037 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
Jan Kara0058f962013-04-11 23:48:32 -040038 if (io_end_cachep == NULL)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040039 return -ENOMEM;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040040 return 0;
41}
42
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040043void ext4_exit_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040044{
45 kmem_cache_destroy(io_end_cachep);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040046}
47
Theodore Ts'o1ada47d2013-03-20 09:39:42 -040048/*
Jan Karab0857d32013-06-04 14:23:41 -040049 * Print an buffer I/O error compatible with the fs/buffer.c. This
50 * provides compatibility with dmesg scrapers that look for a specific
51 * buffer I/O error message. We really need a unified error reporting
52 * structure to userspace ala Digital Unix's uerf system, but it's
53 * probably not going to happen in my lifetime, due to LKML politics...
54 */
55static void buffer_io_error(struct buffer_head *bh)
56{
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +040057 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
58 bh->b_bdev,
Jan Karab0857d32013-06-04 14:23:41 -040059 (unsigned long long)bh->b_blocknr);
60}
61
62static void ext4_finish_bio(struct bio *bio)
63{
64 int i;
Kent Overstreet2c30c712013-11-07 12:20:26 -080065 struct bio_vec *bvec;
Jan Karab0857d32013-06-04 14:23:41 -040066
Kent Overstreet2c30c712013-11-07 12:20:26 -080067 bio_for_each_segment_all(bvec, bio, i) {
Jan Karab0857d32013-06-04 14:23:41 -040068 struct page *page = bvec->bv_page;
Michael Halcrow2058f832015-04-12 00:55:10 -040069#ifdef CONFIG_EXT4_FS_ENCRYPTION
70 struct page *data_page = NULL;
Michael Halcrow2058f832015-04-12 00:55:10 -040071#endif
Jan Karab0857d32013-06-04 14:23:41 -040072 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
Michael Halcrow2058f832015-04-12 00:55:10 -040081#ifdef CONFIG_EXT4_FS_ENCRYPTION
82 if (!page->mapping) {
83 /* The bounce data pages are unmapped. */
84 data_page = page;
Jaegeuk Kima7550b32016-07-10 14:01:03 -040085 fscrypt_pullback_bio_page(&page, false);
Michael Halcrow2058f832015-04-12 00:55:10 -040086 }
87#endif
88
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020089 if (bio->bi_error) {
Jan Karab0857d32013-06-04 14:23:41 -040090 SetPageError(page);
91 set_bit(AS_EIO, &page->mapping->flags);
92 }
93 bh = head = page_buffers(page);
94 /*
95 * We check all buffers in the page under BH_Uptodate_Lock
96 * to avoid races with other end io clearing async_write flags
97 */
98 local_irq_save(flags);
99 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
100 do {
101 if (bh_offset(bh) < bio_start ||
102 bh_offset(bh) + bh->b_size > bio_end) {
103 if (buffer_async_write(bh))
104 under_io++;
105 continue;
106 }
107 clear_buffer_async_write(bh);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200108 if (bio->bi_error)
Jan Karab0857d32013-06-04 14:23:41 -0400109 buffer_io_error(bh);
110 } while ((bh = bh->b_this_page) != head);
111 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
112 local_irq_restore(flags);
Michael Halcrow2058f832015-04-12 00:55:10 -0400113 if (!under_io) {
114#ifdef CONFIG_EXT4_FS_ENCRYPTION
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400115 if (data_page)
116 fscrypt_restore_control_page(data_page);
Michael Halcrow2058f832015-04-12 00:55:10 -0400117#endif
Jan Karab0857d32013-06-04 14:23:41 -0400118 end_page_writeback(page);
Michael Halcrow2058f832015-04-12 00:55:10 -0400119 }
Jan Karab0857d32013-06-04 14:23:41 -0400120 }
121}
122
Jan Kara97a851e2013-06-04 11:58:58 -0400123static void ext4_release_io_end(ext4_io_end_t *io_end)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400124{
Jan Karab0857d32013-06-04 14:23:41 -0400125 struct bio *bio, *next_bio;
126
Jan Kara97a851e2013-06-04 11:58:58 -0400127 BUG_ON(!list_empty(&io_end->list));
128 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
Jan Kara6b523df2013-06-04 13:21:11 -0400129 WARN_ON(io_end->handle);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400130
Jan Karab0857d32013-06-04 14:23:41 -0400131 for (bio = io_end->bio; bio; bio = next_bio) {
132 next_bio = bio->bi_private;
133 ext4_finish_bio(bio);
134 bio_put(bio);
135 }
Jan Kara97a851e2013-06-04 11:58:58 -0400136 kmem_cache_free(io_end_cachep, io_end);
137}
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 io->inode = inode;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400256 INIT_LIST_HEAD(&io->list);
Jan Kara97a851e2013-06-04 11:58:58 -0400257 atomic_set(&io->count, 1);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400258 }
259 return io;
260}
261
Jan Kara97a851e2013-06-04 11:58:58 -0400262void ext4_put_io_end_defer(ext4_io_end_t *io_end)
263{
264 if (atomic_dec_and_test(&io_end->count)) {
265 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
266 ext4_release_io_end(io_end);
267 return;
268 }
269 ext4_add_complete_io(io_end);
270 }
271}
272
273int ext4_put_io_end(ext4_io_end_t *io_end)
274{
275 int err = 0;
276
277 if (atomic_dec_and_test(&io_end->count)) {
278 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
Jan Kara6b523df2013-06-04 13:21:11 -0400279 err = ext4_convert_unwritten_extents(io_end->handle,
280 io_end->inode, io_end->offset,
281 io_end->size);
282 io_end->handle = NULL;
Jan Kara97a851e2013-06-04 11:58:58 -0400283 ext4_clear_io_unwritten_flag(io_end);
284 }
285 ext4_release_io_end(io_end);
286 }
287 return err;
288}
289
290ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
291{
292 atomic_inc(&io_end->count);
293 return io_end;
294}
295
Jan Kara822dbba2013-07-10 21:31:04 -0400296/* BIO completion function for page writeback */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200297static void ext4_end_bio(struct bio *bio)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400298{
299 ext4_io_end_t *io_end = bio->bi_private;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700300 sector_t bi_sector = bio->bi_iter.bi_sector;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400301
302 BUG_ON(!io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400303 bio->bi_end_io = NULL;
Jan Kara0058f962013-04-11 23:48:32 -0400304
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200305 if (bio->bi_error) {
Jan Karab0857d32013-06-04 14:23:41 -0400306 struct inode *inode = io_end->inode;
307
Matthew Wilcox9503c672014-04-07 10:54:20 -0400308 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500309 "(offset %llu size %ld starting block %llu)",
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200310 bio->bi_error, inode->i_ino,
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500311 (unsigned long long) io_end->offset,
312 (long) io_end->size,
313 (unsigned long long)
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500314 bi_sector >> (inode->i_blkbits - 9));
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200315 mapping_set_error(inode->i_mapping, bio->bi_error);
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500316 }
Jan Kara822dbba2013-07-10 21:31:04 -0400317
318 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
319 /*
320 * Link bio into list hanging from io_end. We have to do it
321 * atomically as bio completions can be racing against each
322 * other.
323 */
324 bio->bi_private = xchg(&io_end->bio, bio);
325 ext4_put_io_end_defer(io_end);
326 } else {
327 /*
328 * Drop io_end reference early. Inode can get freed once
329 * we finish the bio.
330 */
331 ext4_put_io_end_defer(io_end);
332 ext4_finish_bio(bio);
333 bio_put(bio);
334 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400335}
336
337void ext4_io_submit(struct ext4_io_submit *io)
338{
339 struct bio *bio = io->io_bio;
340
341 if (bio) {
Mike Christie95fe6c12016-06-05 14:31:48 -0500342 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
343 WRITE_SYNC : 0;
344 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
Mike Christie4e49ea42016-06-05 14:31:41 -0500345 submit_bio(io->io_bio);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400346 }
Peter Huewe7dc57612011-02-21 21:01:42 -0500347 io->io_bio = NULL;
Jan Kara97a851e2013-06-04 11:58:58 -0400348}
349
350void ext4_io_submit_init(struct ext4_io_submit *io,
351 struct writeback_control *wbc)
352{
Tejun Heo5a339112015-07-21 23:50:24 -0400353 io->io_wbc = wbc;
Jan Kara97a851e2013-06-04 11:58:58 -0400354 io->io_bio = NULL;
Peter Huewe7dc57612011-02-21 21:01:42 -0500355 io->io_end = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400356}
357
Jan Kara97a851e2013-06-04 11:58:58 -0400358static int io_submit_init_bio(struct ext4_io_submit *io,
359 struct buffer_head *bh)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400360{
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400361 struct bio *bio;
362
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200363 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
Theodore Ts'oa1d8d9a72013-06-06 10:18:22 -0400364 if (!bio)
365 return -ENOMEM;
Tejun Heo001e4a82015-07-21 23:51:26 -0400366 wbc_init_bio(io->io_wbc, bio);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700367 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400368 bio->bi_bdev = bh->b_bdev;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400369 bio->bi_end_io = ext4_end_bio;
Jan Kara97a851e2013-06-04 11:58:58 -0400370 bio->bi_private = ext4_get_io_end(io->io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400371 io->io_bio = bio;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400372 io->io_next_block = bh->b_blocknr;
373 return 0;
374}
375
376static int io_submit_add_bh(struct ext4_io_submit *io,
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400377 struct inode *inode,
Michael Halcrow2058f832015-04-12 00:55:10 -0400378 struct page *page,
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400379 struct buffer_head *bh)
380{
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400381 int ret;
382
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400383 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
384submit_and_retry:
385 ext4_io_submit(io);
386 }
387 if (io->io_bio == NULL) {
Jan Kara97a851e2013-06-04 11:58:58 -0400388 ret = io_submit_init_bio(io, bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400389 if (ret)
390 return ret;
391 }
Michael Halcrow2058f832015-04-12 00:55:10 -0400392 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
Theodore Ts'oa5499842013-05-11 19:07:42 -0400393 if (ret != bh->b_size)
394 goto submit_and_retry;
Tejun Heo001e4a82015-07-21 23:51:26 -0400395 wbc_account_io(io->io_wbc, page, bh->b_size);
Jan Kara97a851e2013-06-04 11:58:58 -0400396 io->io_next_block++;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400397 return 0;
398}
399
400int ext4_bio_write_page(struct ext4_io_submit *io,
401 struct page *page,
402 int len,
Namjae Jeon1c8349a2014-05-12 08:12:25 -0400403 struct writeback_control *wbc,
404 bool keep_towrite)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400405{
Michael Halcrow2058f832015-04-12 00:55:10 -0400406 struct page *data_page = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400407 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'o937d7b82015-10-02 23:54:58 -0400412 int nr_to_submit = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400413
414 blocksize = 1 << inode->i_blkbits;
415
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500416 BUG_ON(!PageLocked(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400417 BUG_ON(PageWriteback(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400418
Namjae Jeon1c8349a2014-05-12 08:12:25 -0400419 if (keep_towrite)
420 set_page_writeback_keepwrite(page);
421 else
422 set_page_writeback(page);
Theodore Ts'oa54aa762011-02-27 16:43:24 -0500423 ClearPageError(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400424
Jan Kara0058f962013-04-11 23:48:32 -0400425 /*
Linus Torvaldsf8409ab2014-06-08 13:03:35 -0700426 * Comments copied from block_write_full_page:
Jan Karaeeece462014-05-27 12:48:55 -0400427 *
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 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300434 if (len < PAGE_SIZE)
435 zero_user_segment(page, len, PAGE_SIZE);
Jan Karaeeece462014-05-27 12:48:55 -0400436 /*
Jan Kara0058f962013-04-11 23:48:32 -0400437 * 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);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400446 if (block_start >= len) {
447 clear_buffer_dirty(bh);
448 set_buffer_uptodate(bh);
449 continue;
450 }
Jan Kara8a850c32013-01-28 20:53:28 -0500451 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 }
Jan Kara0058f962013-04-11 23:48:32 -0400460 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);
Theodore Ts'o937d7b82015-10-02 23:54:58 -0400465 nr_to_submit++;
Jan Kara0058f962013-04-11 23:48:32 -0400466 } while ((bh = bh->b_this_page) != head);
467
Jan Kara0058f962013-04-11 23:48:32 -0400468 bh = head = page_buffers(page);
Michael Halcrow2058f832015-04-12 00:55:10 -0400469
Theodore Ts'o937d7b82015-10-02 23:54:58 -0400470 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
471 nr_to_submit) {
Theodore Ts'oc9af28f2016-03-26 16:14:34 -0400472 gfp_t gfp_flags = GFP_NOFS;
473
474 retry_encrypt:
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400475 data_page = fscrypt_encrypt_page(inode, page, gfp_flags);
Michael Halcrow2058f832015-04-12 00:55:10 -0400476 if (IS_ERR(data_page)) {
477 ret = PTR_ERR(data_page);
Dan Carpenter4762cc32016-04-02 18:13:38 -0400478 if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
Theodore Ts'oc9af28f2016-03-26 16:14:34 -0400479 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 }
Michael Halcrow2058f832015-04-12 00:55:10 -0400486 data_page = NULL;
487 goto out;
488 }
489 }
490
491 /* Now submit buffers to write */
Jan Kara0058f962013-04-11 23:48:32 -0400492 do {
493 if (!buffer_async_write(bh))
494 continue;
Michael Halcrow2058f832015-04-12 00:55:10 -0400495 ret = io_submit_add_bh(io, inode,
496 data_page ? data_page : page, bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400497 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 */
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400503 break;
504 }
Jan Kara0058f962013-04-11 23:48:32 -0400505 nr_submitted++;
Jan Kara1ae48a62013-01-28 09:32:54 -0500506 clear_buffer_dirty(bh);
Jan Kara0058f962013-04-11 23:48:32 -0400507 } while ((bh = bh->b_this_page) != head);
508
509 /* Error stopped previous loop? Clean up buffers... */
510 if (ret) {
Michael Halcrow2058f832015-04-12 00:55:10 -0400511 out:
512 if (data_page)
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400513 fscrypt_restore_control_page(data_page);
Michael Halcrow2058f832015-04-12 00:55:10 -0400514 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
515 redirty_page_for_writepage(wbc, page);
Jan Kara0058f962013-04-11 23:48:32 -0400516 do {
517 clear_buffer_async_write(bh);
518 bh = bh->b_this_page;
519 } while (bh != head);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400520 }
521 unlock_page(page);
Jan Kara0058f962013-04-11 23:48:32 -0400522 /* Nothing submitted - we have to end page writeback */
523 if (!nr_submitted)
524 end_page_writeback(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400525 return ret;
526}