blob: e4fc8ea45d7888fe3677f052e1af8dff39d443ab [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>
Theodore Ts'obd2d0212010-10-27 21:30:10 -040027
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
Theodore Ts'obd2d0212010-10-27 21:30:10 -040031
Jan Kara0058f962013-04-11 23:48:32 -040032static struct kmem_cache *io_end_cachep;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040033
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040034int __init ext4_init_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040035{
Theodore Ts'obd2d0212010-10-27 21:30:10 -040036 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
Jan Kara0058f962013-04-11 23:48:32 -040037 if (io_end_cachep == NULL)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040038 return -ENOMEM;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040039 return 0;
40}
41
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040042void ext4_exit_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040043{
44 kmem_cache_destroy(io_end_cachep);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040045}
46
Theodore Ts'o1ada47d2013-03-20 09:39:42 -040047/*
Jan Karab0857d32013-06-04 14:23:41 -040048 * 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 */
54static void buffer_io_error(struct buffer_head *bh)
55{
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +040056 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
57 bh->b_bdev,
Jan Karab0857d32013-06-04 14:23:41 -040058 (unsigned long long)bh->b_blocknr);
59}
60
61static void ext4_finish_bio(struct bio *bio)
62{
63 int i;
Kent Overstreet2c30c712013-11-07 12:20:26 -080064 struct bio_vec *bvec;
Jan Karab0857d32013-06-04 14:23:41 -040065
Kent Overstreet2c30c712013-11-07 12:20:26 -080066 bio_for_each_segment_all(bvec, bio, i) {
Jan Karab0857d32013-06-04 14:23:41 -040067 struct page *page = bvec->bv_page;
Michael Halcrow2058f832015-04-12 00:55:10 -040068#ifdef CONFIG_EXT4_FS_ENCRYPTION
69 struct page *data_page = NULL;
70 struct ext4_crypto_ctx *ctx = NULL;
71#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;
85 ctx = (struct ext4_crypto_ctx *)page_private(data_page);
Theodore Ts'o614def72015-05-31 13:31:34 -040086 page = ctx->w.control_page;
Michael Halcrow2058f832015-04-12 00:55:10 -040087 }
88#endif
89
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020090 if (bio->bi_error) {
Jan Karab0857d32013-06-04 14:23:41 -040091 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);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200109 if (bio->bi_error)
Jan Karab0857d32013-06-04 14:23:41 -0400110 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);
Michael Halcrow2058f832015-04-12 00:55:10 -0400114 if (!under_io) {
115#ifdef CONFIG_EXT4_FS_ENCRYPTION
116 if (ctx)
117 ext4_restore_control_page(data_page);
118#endif
Jan Karab0857d32013-06-04 14:23:41 -0400119 end_page_writeback(page);
Michael Halcrow2058f832015-04-12 00:55:10 -0400120 }
Jan Karab0857d32013-06-04 14:23:41 -0400121 }
122}
123
Jan Kara97a851e2013-06-04 11:58:58 -0400124static void ext4_release_io_end(ext4_io_end_t *io_end)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400125{
Jan Karab0857d32013-06-04 14:23:41 -0400126 struct bio *bio, *next_bio;
127
Jan Kara97a851e2013-06-04 11:58:58 -0400128 BUG_ON(!list_empty(&io_end->list));
129 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
Jan Kara6b523df2013-06-04 13:21:11 -0400130 WARN_ON(io_end->handle);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400131
Jan Karab0857d32013-06-04 14:23:41 -0400132 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 }
Jan Kara97a851e2013-06-04 11:58:58 -0400137 kmem_cache_free(io_end_cachep, io_end);
138}
139
Jan Karaa115f742013-06-04 14:30:00 -0400140/*
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 */
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400148static int ext4_end_io(ext4_io_end_t *io)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400149{
150 struct inode *inode = io->inode;
151 loff_t offset = io->offset;
152 ssize_t size = io->size;
Jan Kara6b523df2013-06-04 13:21:11 -0400153 handle_t *handle = io->handle;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400154 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
Jan Kara6b523df2013-06-04 13:21:11 -0400160 io->handle = NULL; /* Following call will use up the handle */
161 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400162 if (ret < 0) {
Theodore Ts'ob82e3842011-10-31 10:56:32 -0400163 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);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400168 }
Jan Kara97a851e2013-06-04 11:58:58 -0400169 ext4_clear_io_unwritten_flag(io);
170 ext4_release_io_end(io);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400171 return ret;
172}
173
Jan Kara2e8fa542013-06-04 14:21:02 -0400174static void dump_completed_IO(struct inode *inode, struct list_head *head)
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400175{
176#ifdef EXT4FS_DEBUG
177 struct list_head *cur, *before, *after;
178 ext4_io_end_t *io, *io0, *io1;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400179
Jan Kara2e8fa542013-06-04 14:21:02 -0400180 if (list_empty(head))
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400181 return;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400182
Jan Kara2e8fa542013-06-04 14:21:02 -0400183 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
184 list_for_each_entry(io, head, list) {
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400185 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. */
Jan Kara97a851e2013-06-04 11:58:58 -0400198static void ext4_add_complete_io(ext4_io_end_t *io_end)
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400199{
200 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
Jan Kara78371a42013-10-16 08:25:11 -0400201 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400202 struct workqueue_struct *wq;
203 unsigned long flags;
204
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200205 /* Only reserved conversions from writeback should enter here */
206 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
Jan Kara78371a42013-10-16 08:25:11 -0400207 WARN_ON(!io_end->handle && sbi->s_journal);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400208 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
Jan Kara78371a42013-10-16 08:25:11 -0400209 wq = sbi->rsv_conversion_wq;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200210 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);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400213 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
214}
215
Jan Kara2e8fa542013-06-04 14:21:02 -0400216static int ext4_do_flush_completed_IO(struct inode *inode,
217 struct list_head *head)
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400218{
219 ext4_io_end_t *io;
Jan Kara002bd7f2013-01-28 09:49:15 -0500220 struct list_head unwritten;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400221 unsigned long flags;
222 struct ext4_inode_info *ei = EXT4_I(inode);
223 int err, ret = 0;
224
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400225 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
Jan Kara2e8fa542013-06-04 14:21:02 -0400226 dump_completed_IO(inode, head);
227 list_replace_init(head, &unwritten);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400228 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;
Dmitry Monakhov28a535f2012-09-29 00:14:55 -0400238 }
239 return ret;
240}
241
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400242/*
Jan Kara2e8fa542013-06-04 14:21:02 -0400243 * work on completed IO, to convert unwritten extents to extents
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400244 */
Jan Kara2e8fa542013-06-04 14:21:02 -0400245void ext4_end_io_rsv_work(struct work_struct *work)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400246{
Jan Kara84c17542013-01-28 09:43:46 -0500247 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
Jan Kara2e8fa542013-06-04 14:21:02 -0400248 i_rsv_conversion_work);
249 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
250}
251
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400252ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
253{
Jesper Juhlb17b35e2010-12-19 21:41:55 -0500254 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400255 if (io) {
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500256 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 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200298static void ext4_end_bio(struct bio *bio)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400299{
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;
Jan Kara0058f962013-04-11 23:48:32 -0400305
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200306 if (bio->bi_error) {
Jan Karab0857d32013-06-04 14:23:41 -0400307 struct inode *inode = io_end->inode;
308
Matthew Wilcox9503c672014-04-07 10:54:20 -0400309 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500310 "(offset %llu size %ld starting block %llu)",
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200311 bio->bi_error, inode->i_ino,
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500312 (unsigned long long) io_end->offset,
313 (long) io_end->size,
314 (unsigned long long)
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500315 bi_sector >> (inode->i_blkbits - 9));
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200316 mapping_set_error(inode->i_mapping, bio->bi_error);
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500317 }
Jan Kara822dbba2013-07-10 21:31:04 -0400318
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 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400336}
337
338void ext4_io_submit(struct ext4_io_submit *io)
339{
340 struct bio *bio = io->io_bio;
341
342 if (bio) {
Tejun Heo5a339112015-07-21 23:50:24 -0400343 int io_op = io->io_wbc->sync_mode == WB_SYNC_ALL ?
344 WRITE_SYNC : WRITE;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400345 bio_get(io->io_bio);
Tejun Heo5a339112015-07-21 23:50:24 -0400346 submit_bio(io_op, io->io_bio);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400347 bio_put(io->io_bio);
348 }
Peter Huewe7dc57612011-02-21 21:01:42 -0500349 io->io_bio = NULL;
Jan Kara97a851e2013-06-04 11:58:58 -0400350}
351
352void ext4_io_submit_init(struct ext4_io_submit *io,
353 struct writeback_control *wbc)
354{
Tejun Heo5a339112015-07-21 23:50:24 -0400355 io->io_wbc = wbc;
Jan Kara97a851e2013-06-04 11:58:58 -0400356 io->io_bio = NULL;
Peter Huewe7dc57612011-02-21 21:01:42 -0500357 io->io_end = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400358}
359
Jan Kara97a851e2013-06-04 11:58:58 -0400360static int io_submit_init_bio(struct ext4_io_submit *io,
361 struct buffer_head *bh)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400362{
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400363 struct bio *bio;
364
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200365 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
Theodore Ts'oa1d8d9a72013-06-06 10:18:22 -0400366 if (!bio)
367 return -ENOMEM;
Tejun Heo001e4a82015-07-21 23:51:26 -0400368 wbc_init_bio(io->io_wbc, bio);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700369 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400370 bio->bi_bdev = bh->b_bdev;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400371 bio->bi_end_io = ext4_end_bio;
Jan Kara97a851e2013-06-04 11:58:58 -0400372 bio->bi_private = ext4_get_io_end(io->io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400373 io->io_bio = bio;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400374 io->io_next_block = bh->b_blocknr;
375 return 0;
376}
377
378static int io_submit_add_bh(struct ext4_io_submit *io,
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400379 struct inode *inode,
Michael Halcrow2058f832015-04-12 00:55:10 -0400380 struct page *page,
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 }
Michael Halcrow2058f832015-04-12 00:55:10 -0400394 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
Theodore Ts'oa5499842013-05-11 19:07:42 -0400395 if (ret != bh->b_size)
396 goto submit_and_retry;
Tejun Heo001e4a82015-07-21 23:51:26 -0400397 wbc_account_io(io->io_wbc, page, bh->b_size);
Jan Kara97a851e2013-06-04 11:58:58 -0400398 io->io_next_block++;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400399 return 0;
400}
401
402int ext4_bio_write_page(struct ext4_io_submit *io,
403 struct page *page,
404 int len,
Namjae Jeon1c8349a2014-05-12 08:12:25 -0400405 struct writeback_control *wbc,
406 bool keep_towrite)
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400407{
Michael Halcrow2058f832015-04-12 00:55:10 -0400408 struct page *data_page = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400409 struct inode *inode = page->mapping->host;
Jan Kara0058f962013-04-11 23:48:32 -0400410 unsigned block_start, blocksize;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400411 struct buffer_head *bh, *head;
412 int ret = 0;
Jan Kara0058f962013-04-11 23:48:32 -0400413 int nr_submitted = 0;
Theodore Ts'o937d7b82015-10-02 23:54:58 -0400414 int nr_to_submit = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400415
416 blocksize = 1 << inode->i_blkbits;
417
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500418 BUG_ON(!PageLocked(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400419 BUG_ON(PageWriteback(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400420
Namjae Jeon1c8349a2014-05-12 08:12:25 -0400421 if (keep_towrite)
422 set_page_writeback_keepwrite(page);
423 else
424 set_page_writeback(page);
Theodore Ts'oa54aa762011-02-27 16:43:24 -0500425 ClearPageError(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400426
Jan Kara0058f962013-04-11 23:48:32 -0400427 /*
Linus Torvaldsf8409ab2014-06-08 13:03:35 -0700428 * Comments copied from block_write_full_page:
Jan Karaeeece462014-05-27 12:48:55 -0400429 *
430 * The page straddles i_size. It must be zeroed out on each and every
431 * writepage invocation because it may be mmapped. "A file is mapped
432 * in multiples of the page size. For a file that is not a multiple of
433 * the page size, the remaining memory is zeroed when mapped, and
434 * writes to that region are not written out to the file."
435 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300436 if (len < PAGE_SIZE)
437 zero_user_segment(page, len, PAGE_SIZE);
Jan Karaeeece462014-05-27 12:48:55 -0400438 /*
Jan Kara0058f962013-04-11 23:48:32 -0400439 * In the first loop we prepare and mark buffers to submit. We have to
440 * mark all buffers in the page before submitting so that
441 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
442 * on the first buffer finishes and we are still working on submitting
443 * the second buffer.
444 */
445 bh = head = page_buffers(page);
446 do {
447 block_start = bh_offset(bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400448 if (block_start >= len) {
449 clear_buffer_dirty(bh);
450 set_buffer_uptodate(bh);
451 continue;
452 }
Jan Kara8a850c32013-01-28 20:53:28 -0500453 if (!buffer_dirty(bh) || buffer_delay(bh) ||
454 !buffer_mapped(bh) || buffer_unwritten(bh)) {
455 /* A hole? We can safely clear the dirty bit */
456 if (!buffer_mapped(bh))
457 clear_buffer_dirty(bh);
458 if (io->io_bio)
459 ext4_io_submit(io);
460 continue;
461 }
Jan Kara0058f962013-04-11 23:48:32 -0400462 if (buffer_new(bh)) {
463 clear_buffer_new(bh);
464 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
465 }
466 set_buffer_async_write(bh);
Theodore Ts'o937d7b82015-10-02 23:54:58 -0400467 nr_to_submit++;
Jan Kara0058f962013-04-11 23:48:32 -0400468 } while ((bh = bh->b_this_page) != head);
469
Jan Kara0058f962013-04-11 23:48:32 -0400470 bh = head = page_buffers(page);
Michael Halcrow2058f832015-04-12 00:55:10 -0400471
Theodore Ts'o937d7b82015-10-02 23:54:58 -0400472 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
473 nr_to_submit) {
Theodore Ts'oc9af28f2016-03-26 16:14:34 -0400474 gfp_t gfp_flags = GFP_NOFS;
475
476 retry_encrypt:
477 data_page = ext4_encrypt(inode, page, gfp_flags);
Michael Halcrow2058f832015-04-12 00:55:10 -0400478 if (IS_ERR(data_page)) {
479 ret = PTR_ERR(data_page);
Dan Carpenter4762cc32016-04-02 18:13:38 -0400480 if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
Theodore Ts'oc9af28f2016-03-26 16:14:34 -0400481 if (io->io_bio) {
482 ext4_io_submit(io);
483 congestion_wait(BLK_RW_ASYNC, HZ/50);
484 }
485 gfp_flags |= __GFP_NOFAIL;
486 goto retry_encrypt;
487 }
Michael Halcrow2058f832015-04-12 00:55:10 -0400488 data_page = NULL;
489 goto out;
490 }
491 }
492
493 /* Now submit buffers to write */
Jan Kara0058f962013-04-11 23:48:32 -0400494 do {
495 if (!buffer_async_write(bh))
496 continue;
Michael Halcrow2058f832015-04-12 00:55:10 -0400497 ret = io_submit_add_bh(io, inode,
498 data_page ? data_page : page, bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400499 if (ret) {
500 /*
501 * We only get here on ENOMEM. Not much else
502 * we can do but mark the page as dirty, and
503 * better luck next time.
504 */
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400505 break;
506 }
Jan Kara0058f962013-04-11 23:48:32 -0400507 nr_submitted++;
Jan Kara1ae48a62013-01-28 09:32:54 -0500508 clear_buffer_dirty(bh);
Jan Kara0058f962013-04-11 23:48:32 -0400509 } while ((bh = bh->b_this_page) != head);
510
511 /* Error stopped previous loop? Clean up buffers... */
512 if (ret) {
Michael Halcrow2058f832015-04-12 00:55:10 -0400513 out:
514 if (data_page)
515 ext4_restore_control_page(data_page);
516 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
517 redirty_page_for_writepage(wbc, page);
Jan Kara0058f962013-04-11 23:48:32 -0400518 do {
519 clear_buffer_async_write(bh);
520 bh = bh->b_this_page;
521 } while (bh != head);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400522 }
523 unlock_page(page);
Jan Kara0058f962013-04-11 23:48:32 -0400524 /* Nothing submitted - we have to end page writeback */
525 if (!nr_submitted)
526 end_page_writeback(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400527 return ret;
528}