| 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/blk-crypto.h> |
| 11 |
#include <linux/fs.h> |
| 12 |
#include <linux/time.h> |
| 13 |
#include <linux/highuid.h> |
| 14 |
#include <linux/pagemap.h> |
| 15 |
#include <linux/quotaops.h> |
| 16 |
#include <linux/string.h> |
| 17 |
#include <linux/buffer_head.h> |
| 18 |
#include <linux/writeback.h> |
| 19 |
#include <linux/pagevec.h> |
| 20 |
#include <linux/mpage.h> |
| 21 |
#include <linux/namei.h> |
| 22 |
#include <linux/uio.h> |
| 23 |
#include <linux/bio.h> |
| 24 |
#include <linux/workqueue.h> |
| 25 |
#include <linux/kernel.h> |
| 26 |
#include <linux/slab.h> |
| 27 |
#include <linux/mm.h> |
| 28 |
#include <linux/sched/mm.h> |
| 29 |
|
| 30 |
#include "ext4_jbd2.h" |
| 31 |
#include "xattr.h" |
| 32 |
#include "acl.h" |
| 33 |
|
| 34 |
static struct kmem_cache *io_end_cachep; |
| 35 |
static struct kmem_cache *io_end_vec_cachep; |
| 36 |
|
| 37 |
int __init ext4_init_pageio(void) |
| 38 |
{ |
| 39 |
io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT); |
| 40 |
if (io_end_cachep == NULL) |
| 41 |
return -ENOMEM; |
| 42 |
|
| 43 |
io_end_vec_cachep = KMEM_CACHE(ext4_io_end_vec, 0); |
| 44 |
if (io_end_vec_cachep == NULL) { |
| 45 |
kmem_cache_destroy(io_end_cachep); |
| 46 |
return -ENOMEM; |
| 47 |
} |
| 48 |
return 0; |
| 49 |
} |
| 50 |
|
| 51 |
void ext4_exit_pageio(void) |
| 52 |
{ |
| 53 |
kmem_cache_destroy(io_end_cachep); |
| 54 |
kmem_cache_destroy(io_end_vec_cachep); |
| 55 |
} |
| 56 |
|
| 57 |
struct ext4_io_end_vec *ext4_alloc_io_end_vec(ext4_io_end_t *io_end) |
| 58 |
{ |
| 59 |
struct ext4_io_end_vec *io_end_vec; |
| 60 |
|
| 61 |
io_end_vec = kmem_cache_zalloc(io_end_vec_cachep, GFP_NOFS); |
| 62 |
if (!io_end_vec) |
| 63 |
return ERR_PTR(-ENOMEM); |
| 64 |
INIT_LIST_HEAD(&io_end_vec->list); |
| 65 |
list_add_tail(&io_end_vec->list, &io_end->list_vec); |
| 66 |
return io_end_vec; |
| 67 |
} |
| 68 |
|
| 69 |
static void ext4_free_io_end_vec(ext4_io_end_t *io_end) |
| 70 |
{ |
| 71 |
struct ext4_io_end_vec *io_end_vec, *tmp; |
| 72 |
|
| 73 |
if (list_empty(&io_end->list_vec)) |
| 74 |
return; |
| 75 |
list_for_each_entry_safe(io_end_vec, tmp, &io_end->list_vec, list) { |
| 76 |
list_del(&io_end_vec->list); |
| 77 |
kmem_cache_free(io_end_vec_cachep, io_end_vec); |
| 78 |
} |
| 79 |
} |
| 80 |
|
| 81 |
struct ext4_io_end_vec *ext4_last_io_end_vec(ext4_io_end_t *io_end) |
| 82 |
{ |
| 83 |
BUG_ON(list_empty(&io_end->list_vec)); |
| 84 |
return list_last_entry(&io_end->list_vec, struct ext4_io_end_vec, list); |
| 85 |
} |
| 86 |
|
| 87 |
/* |
| 88 |
* Print an buffer I/O error compatible with the fs/buffer.c. This |
| 89 |
* provides compatibility with dmesg scrapers that look for a specific |
| 90 |
* buffer I/O error message. We really need a unified error reporting |
| 91 |
* structure to userspace ala Digital Unix's uerf system, but it's |
| 92 |
* probably not going to happen in my lifetime, due to LKML politics... |
| 93 |
*/ |
| 94 |
static void buffer_io_error(struct buffer_head *bh) |
| 95 |
{ |
| 96 |
printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n", |
| 97 |
bh->b_bdev, |
| 98 |
(unsigned long long)bh->b_blocknr); |
| 99 |
} |
| 100 |
|
| 101 |
static void ext4_finish_bio(struct bio *bio) |
| 102 |
{ |
| 103 |
struct folio_iter fi; |
| 104 |
|
| 105 |
bio_for_each_folio_all(fi, bio) { |
| 106 |
struct folio *folio = fi.folio; |
| 107 |
struct folio *io_folio = NULL; |
| 108 |
struct buffer_head *bh, *head; |
| 109 |
size_t bio_start = fi.offset; |
| 110 |
size_t bio_end = bio_start + fi.length; |
| 111 |
unsigned under_io = 0; |
| 112 |
unsigned long flags; |
| 113 |
|
| 114 |
if (fscrypt_is_bounce_folio(folio)) { |
| 115 |
io_folio = folio; |
| 116 |
folio = fscrypt_pagecache_folio(folio); |
| 117 |
} |
| 118 |
|
| 119 |
if (bio->bi_status) { |
| 120 |
int err = blk_status_to_errno(bio->bi_status); |
| 121 |
mapping_set_error(folio->mapping, err); |
| 122 |
} |
| 123 |
bh = head = folio_buffers(folio); |
| 124 |
/* |
| 125 |
* We check all buffers in the folio under b_uptodate_lock |
| 126 |
* to avoid races with other end io clearing async_write flags |
| 127 |
*/ |
| 128 |
spin_lock_irqsave(&head->b_uptodate_lock, flags); |
| 129 |
do { |
| 130 |
if (bh_offset(bh) < bio_start || |
| 131 |
bh_offset(bh) + bh->b_size > bio_end) { |
| 132 |
if (buffer_async_write(bh)) |
| 133 |
under_io++; |
| 134 |
continue; |
| 135 |
} |
| 136 |
clear_buffer_async_write(bh); |
| 137 |
if (bio->bi_status) { |
| 138 |
set_buffer_write_io_error(bh); |
| 139 |
buffer_io_error(bh); |
| 140 |
} |
| 141 |
} while ((bh = bh->b_this_page) != head); |
| 142 |
spin_unlock_irqrestore(&head->b_uptodate_lock, flags); |
| 143 |
if (!under_io) { |
| 144 |
fscrypt_free_bounce_page(&io_folio->page); |
| 145 |
folio_end_writeback(folio); |
| 146 |
} |
| 147 |
} |
| 148 |
} |
| 149 |
|
| 150 |
static void ext4_release_io_end(ext4_io_end_t *io_end) |
| 151 |
{ |
| 152 |
struct bio *bio, *next_bio; |
| 153 |
|
| 154 |
BUG_ON(!list_empty(&io_end->list)); |
| 155 |
BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); |
| 156 |
WARN_ON(io_end->handle); |
| 157 |
|
| 158 |
for (bio = io_end->bio; bio; bio = next_bio) { |
| 159 |
next_bio = bio->bi_private; |
| 160 |
ext4_finish_bio(bio); |
| 161 |
bio_put(bio); |
| 162 |
} |
| 163 |
ext4_free_io_end_vec(io_end); |
| 164 |
kmem_cache_free(io_end_cachep, io_end); |
| 165 |
} |
| 166 |
|
| 167 |
/* |
| 168 |
* On successful IO, check a range of space and convert unwritten extents to |
| 169 |
* written. On IO failure, check if journal abort is needed. 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 |
*/ |
| 176 |
static 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 |
struct super_block *sb = inode->i_sb; |
| 181 |
int ret = 0; |
| 182 |
|
| 183 |
ext4_debug("ext4_end_io_nolock: io_end 0x%p from inode %lu,list->next 0x%p," |
| 184 |
"list->prev 0x%p\n", |
| 185 |
io_end, inode->i_ino, io_end->list.next, io_end->list.prev); |
| 186 |
|
| 187 |
/* |
| 188 |
* Do not convert the unwritten extents if data writeback fails, |
| 189 |
* or stale data may be exposed. |
| 190 |
*/ |
| 191 |
io_end->handle = NULL; /* Following call will use up the handle */ |
| 192 |
if (unlikely(io_end->flag & EXT4_IO_END_FAILED)) { |
| 193 |
ret = -EIO; |
| 194 |
if (handle) |
| 195 |
jbd2_journal_free_reserved(handle); |
| 196 |
|
| 197 |
if (test_opt(sb, DATA_ERR_ABORT)) |
| 198 |
jbd2_journal_abort(EXT4_SB(sb)->s_journal, ret); |
| 199 |
} else { |
| 200 |
ret = ext4_convert_unwritten_io_end_vec(handle, io_end); |
| 201 |
} |
| 202 |
if (ret < 0 && !ext4_emergency_state(sb) && |
| 203 |
io_end->flag & EXT4_IO_END_UNWRITTEN) { |
| 204 |
ext4_msg(sb, KERN_EMERG, |
| 205 |
"failed to convert unwritten extents to written " |
| 206 |
"extents -- potential data loss! " |
| 207 |
"(inode %lu, error %d)", inode->i_ino, ret); |
| 208 |
} |
| 209 |
|
| 210 |
ext4_clear_io_unwritten_flag(io_end); |
| 211 |
ext4_release_io_end(io_end); |
| 212 |
return ret; |
| 213 |
} |
| 214 |
|
| 215 |
static void dump_completed_IO(struct inode *inode, struct list_head *head) |
| 216 |
{ |
| 217 |
#ifdef EXT4FS_DEBUG |
| 218 |
struct list_head *cur, *before, *after; |
| 219 |
ext4_io_end_t *io_end, *io_end0, *io_end1; |
| 220 |
|
| 221 |
if (list_empty(head)) |
| 222 |
return; |
| 223 |
|
| 224 |
ext4_debug("Dump inode %lu completed io list\n", inode->i_ino); |
| 225 |
list_for_each_entry(io_end, head, list) { |
| 226 |
cur = &io_end->list; |
| 227 |
before = cur->prev; |
| 228 |
io_end0 = container_of(before, ext4_io_end_t, list); |
| 229 |
after = cur->next; |
| 230 |
io_end1 = container_of(after, ext4_io_end_t, list); |
| 231 |
|
| 232 |
ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", |
| 233 |
io_end, inode->i_ino, io_end0, io_end1); |
| 234 |
} |
| 235 |
#endif |
| 236 |
} |
| 237 |
|
| 238 |
static bool ext4_io_end_defer_completion(ext4_io_end_t *io_end) |
| 239 |
{ |
| 240 |
if (io_end->flag & EXT4_IO_END_UNWRITTEN && |
| 241 |
!list_empty(&io_end->list_vec)) |
| 242 |
return true; |
| 243 |
if (test_opt(io_end->inode->i_sb, DATA_ERR_ABORT) && |
| 244 |
io_end->flag & EXT4_IO_END_FAILED && |
| 245 |
!ext4_emergency_state(io_end->inode->i_sb)) |
| 246 |
return true; |
| 247 |
return false; |
| 248 |
} |
| 249 |
|
| 250 |
/* Add the io_end to per-inode completed end_io list. */ |
| 251 |
static void ext4_add_complete_io(ext4_io_end_t *io_end) |
| 252 |
{ |
| 253 |
struct ext4_inode_info *ei = EXT4_I(io_end->inode); |
| 254 |
struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb); |
| 255 |
struct workqueue_struct *wq; |
| 256 |
unsigned long flags; |
| 257 |
|
| 258 |
/* Only reserved conversions or pending IO errors will enter here. */ |
| 259 |
WARN_ON(!(io_end->flag & EXT4_IO_END_DEFER_COMPLETION)); |
| 260 |
WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN && |
| 261 |
!io_end->handle && sbi->s_journal); |
| 262 |
WARN_ON(!io_end->bio); |
| 263 |
|
| 264 |
spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
| 265 |
wq = sbi->rsv_conversion_wq; |
| 266 |
if (list_empty(&ei->i_rsv_conversion_list)) |
| 267 |
queue_work(wq, &ei->i_rsv_conversion_work); |
| 268 |
list_add_tail(&io_end->list, &ei->i_rsv_conversion_list); |
| 269 |
spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
| 270 |
} |
| 271 |
|
| 272 |
static int ext4_do_flush_completed_IO(struct inode *inode, |
| 273 |
struct list_head *head) |
| 274 |
{ |
| 275 |
ext4_io_end_t *io_end; |
| 276 |
struct list_head unwritten; |
| 277 |
unsigned long flags; |
| 278 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 279 |
int err, ret = 0; |
| 280 |
|
| 281 |
spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
| 282 |
dump_completed_IO(inode, head); |
| 283 |
list_replace_init(head, &unwritten); |
| 284 |
spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
| 285 |
|
| 286 |
while (!list_empty(&unwritten)) { |
| 287 |
io_end = list_entry(unwritten.next, ext4_io_end_t, list); |
| 288 |
BUG_ON(!(io_end->flag & EXT4_IO_END_DEFER_COMPLETION)); |
| 289 |
list_del_init(&io_end->list); |
| 290 |
|
| 291 |
err = ext4_end_io_end(io_end); |
| 292 |
if (unlikely(!ret && err)) |
| 293 |
ret = err; |
| 294 |
} |
| 295 |
return ret; |
| 296 |
} |
| 297 |
|
| 298 |
/* |
| 299 |
* Used to convert unwritten extents to written extents upon IO completion, |
| 300 |
* or used to abort the journal upon IO errors. |
| 301 |
*/ |
| 302 |
void ext4_end_io_rsv_work(struct work_struct *work) |
| 303 |
{ |
| 304 |
struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info, |
| 305 |
i_rsv_conversion_work); |
| 306 |
ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list); |
| 307 |
} |
| 308 |
|
| 309 |
ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) |
| 310 |
{ |
| 311 |
ext4_io_end_t *io_end = kmem_cache_zalloc(io_end_cachep, flags); |
| 312 |
|
| 313 |
if (io_end) { |
| 314 |
io_end->inode = inode; |
| 315 |
INIT_LIST_HEAD(&io_end->list); |
| 316 |
INIT_LIST_HEAD(&io_end->list_vec); |
| 317 |
refcount_set(&io_end->count, 1); |
| 318 |
} |
| 319 |
return io_end; |
| 320 |
} |
| 321 |
|
| 322 |
void ext4_put_io_end_defer(ext4_io_end_t *io_end) |
| 323 |
{ |
| 324 |
if (refcount_dec_and_test(&io_end->count)) { |
| 325 |
if (ext4_io_end_defer_completion(io_end)) |
| 326 |
return ext4_add_complete_io(io_end); |
| 327 |
|
| 328 |
ext4_release_io_end(io_end); |
| 329 |
} |
| 330 |
} |
| 331 |
|
| 332 |
int ext4_put_io_end(ext4_io_end_t *io_end) |
| 333 |
{ |
| 334 |
if (refcount_dec_and_test(&io_end->count)) { |
| 335 |
if (ext4_io_end_defer_completion(io_end)) |
| 336 |
return ext4_end_io_end(io_end); |
| 337 |
|
| 338 |
ext4_release_io_end(io_end); |
| 339 |
} |
| 340 |
return 0; |
| 341 |
} |
| 342 |
|
| 343 |
ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end) |
| 344 |
{ |
| 345 |
refcount_inc(&io_end->count); |
| 346 |
return io_end; |
| 347 |
} |
| 348 |
|
| 349 |
/* BIO completion function for page writeback */ |
| 350 |
static void ext4_end_bio(struct bio *bio) |
| 351 |
{ |
| 352 |
ext4_io_end_t *io_end = bio->bi_private; |
| 353 |
sector_t bi_sector = bio->bi_iter.bi_sector; |
| 354 |
|
| 355 |
if (WARN_ONCE(!io_end, "io_end is NULL: %pg: sector %Lu len %u err %d\n", |
| 356 |
bio->bi_bdev, |
| 357 |
(long long) bio->bi_iter.bi_sector, |
| 358 |
(unsigned) bio_sectors(bio), |
| 359 |
bio->bi_status)) { |
| 360 |
ext4_finish_bio(bio); |
| 361 |
bio_put(bio); |
| 362 |
return; |
| 363 |
} |
| 364 |
bio->bi_end_io = NULL; |
| 365 |
|
| 366 |
if (bio->bi_status) { |
| 367 |
struct inode *inode = io_end->inode; |
| 368 |
|
| 369 |
ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu " |
| 370 |
"starting block %llu)", |
| 371 |
bio->bi_status, inode->i_ino, |
| 372 |
(unsigned long long) |
| 373 |
bi_sector >> (inode->i_blkbits - 9)); |
| 374 |
io_end->flag |= EXT4_IO_END_FAILED; |
| 375 |
mapping_set_error(inode->i_mapping, |
| 376 |
blk_status_to_errno(bio->bi_status)); |
| 377 |
} |
| 378 |
|
| 379 |
if (ext4_io_end_defer_completion(io_end)) { |
| 380 |
/* |
| 381 |
* Link bio into list hanging from io_end. We have to do it |
| 382 |
* atomically as bio completions can be racing against each |
| 383 |
* other. |
| 384 |
*/ |
| 385 |
bio->bi_private = xchg(&io_end->bio, bio); |
| 386 |
ext4_put_io_end_defer(io_end); |
| 387 |
} else { |
| 388 |
/* |
| 389 |
* Drop io_end reference early. Inode can get freed once |
| 390 |
* we finish the bio. |
| 391 |
*/ |
| 392 |
ext4_put_io_end_defer(io_end); |
| 393 |
ext4_finish_bio(bio); |
| 394 |
bio_put(bio); |
| 395 |
} |
| 396 |
} |
| 397 |
|
| 398 |
void ext4_io_submit(struct ext4_io_submit *io) |
| 399 |
{ |
| 400 |
struct bio *bio = io->io_bio; |
| 401 |
|
| 402 |
if (bio) { |
| 403 |
if (io->io_wbc->sync_mode == WB_SYNC_ALL) |
| 404 |
io->io_bio->bi_opf |= REQ_SYNC; |
| 405 |
blk_crypto_submit_bio(io->io_bio); |
| 406 |
} |
| 407 |
io->io_bio = NULL; |
| 408 |
} |
| 409 |
|
| 410 |
void ext4_io_submit_init(struct ext4_io_submit *io, |
| 411 |
struct writeback_control *wbc) |
| 412 |
{ |
| 413 |
io->io_wbc = wbc; |
| 414 |
io->io_bio = NULL; |
| 415 |
io->io_end = NULL; |
| 416 |
} |
| 417 |
|
| 418 |
static void io_submit_init_bio(struct ext4_io_submit *io, |
| 419 |
struct buffer_head *bh) |
| 420 |
{ |
| 421 |
struct bio *bio; |
| 422 |
|
| 423 |
/* |
| 424 |
* bio_alloc will _always_ be able to allocate a bio if |
| 425 |
* __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset(). |
| 426 |
*/ |
| 427 |
bio = bio_alloc(bh->b_bdev, BIO_MAX_VECS, REQ_OP_WRITE, GFP_NOIO); |
| 428 |
fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); |
| 429 |
bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
| 430 |
bio->bi_end_io = ext4_end_bio; |
| 431 |
bio->bi_private = ext4_get_io_end(io->io_end); |
| 432 |
io->io_bio = bio; |
| 433 |
io->io_next_block = bh->b_blocknr; |
| 434 |
wbc_init_bio(io->io_wbc, bio); |
| 435 |
} |
| 436 |
|
| 437 |
static void io_submit_add_bh(struct ext4_io_submit *io, |
| 438 |
struct inode *inode, |
| 439 |
struct folio *folio, |
| 440 |
struct folio *io_folio, |
| 441 |
struct buffer_head *bh) |
| 442 |
{ |
| 443 |
if (io->io_bio && (bh->b_blocknr != io->io_next_block || |
| 444 |
!fscrypt_mergeable_bio_bh(io->io_bio, bh))) { |
| 445 |
submit_and_retry: |
| 446 |
ext4_io_submit(io); |
| 447 |
} |
| 448 |
if (io->io_bio == NULL) { |
| 449 |
io_submit_init_bio(io, bh); |
| 450 |
io->io_bio->bi_write_hint = inode->i_write_hint; |
| 451 |
} |
| 452 |
if (!bio_add_folio(io->io_bio, io_folio, bh->b_size, bh_offset(bh))) |
| 453 |
goto submit_and_retry; |
| 454 |
wbc_account_cgroup_owner(io->io_wbc, folio, bh->b_size); |
| 455 |
io->io_next_block++; |
| 456 |
} |
| 457 |
|
| 458 |
int ext4_bio_write_folio(struct ext4_io_submit *io, struct folio *folio, |
| 459 |
size_t len) |
| 460 |
{ |
| 461 |
struct folio *io_folio = folio; |
| 462 |
struct inode *inode = folio->mapping->host; |
| 463 |
unsigned block_start; |
| 464 |
struct buffer_head *bh, *head; |
| 465 |
int ret = 0; |
| 466 |
int nr_to_submit = 0; |
| 467 |
struct writeback_control *wbc = io->io_wbc; |
| 468 |
bool keep_towrite = false; |
| 469 |
|
| 470 |
BUG_ON(!folio_test_locked(folio)); |
| 471 |
BUG_ON(folio_test_writeback(folio)); |
| 472 |
|
| 473 |
/* |
| 474 |
* Comments copied from block_write_full_folio: |
| 475 |
* |
| 476 |
* The folio straddles i_size. It must be zeroed out on each and every |
| 477 |
* writepage invocation because it may be mmapped. "A file is mapped |
| 478 |
* in multiples of the page size. For a file that is not a multiple of |
| 479 |
* the page size, the remaining memory is zeroed when mapped, and |
| 480 |
* writes to that region are not written out to the file." |
| 481 |
*/ |
| 482 |
if (len < folio_size(folio)) |
| 483 |
folio_zero_segment(folio, len, folio_size(folio)); |
| 484 |
/* |
| 485 |
* In the first loop we prepare and mark buffers to submit. We have to |
| 486 |
* mark all buffers in the folio before submitting so that |
| 487 |
* folio_end_writeback() cannot be called from ext4_end_bio() when IO |
| 488 |
* on the first buffer finishes and we are still working on submitting |
| 489 |
* the second buffer. |
| 490 |
*/ |
| 491 |
bh = head = folio_buffers(folio); |
| 492 |
do { |
| 493 |
block_start = bh_offset(bh); |
| 494 |
if (block_start >= len) { |
| 495 |
clear_buffer_dirty(bh); |
| 496 |
set_buffer_uptodate(bh); |
| 497 |
continue; |
| 498 |
} |
| 499 |
if (!buffer_dirty(bh) || buffer_delay(bh) || |
| 500 |
!buffer_mapped(bh) || buffer_unwritten(bh)) { |
| 501 |
/* A hole? We can safely clear the dirty bit */ |
| 502 |
if (!buffer_mapped(bh)) |
| 503 |
clear_buffer_dirty(bh); |
| 504 |
/* |
| 505 |
* Keeping dirty some buffer we cannot write? Make sure |
| 506 |
* to redirty the folio and keep TOWRITE tag so that |
| 507 |
* racing WB_SYNC_ALL writeback does not skip the folio. |
| 508 |
* This happens e.g. when doing writeout for |
| 509 |
* transaction commit or when journalled data is not |
| 510 |
* yet committed. |
| 511 |
*/ |
| 512 |
if (buffer_dirty(bh) || |
| 513 |
(buffer_jbd(bh) && buffer_jbddirty(bh))) { |
| 514 |
if (!folio_test_dirty(folio)) |
| 515 |
folio_redirty_for_writepage(wbc, folio); |
| 516 |
keep_towrite = true; |
| 517 |
} |
| 518 |
continue; |
| 519 |
} |
| 520 |
if (buffer_new(bh)) |
| 521 |
clear_buffer_new(bh); |
| 522 |
set_buffer_async_write(bh); |
| 523 |
clear_buffer_dirty(bh); |
| 524 |
nr_to_submit++; |
| 525 |
} while ((bh = bh->b_this_page) != head); |
| 526 |
|
| 527 |
if (!nr_to_submit) { |
| 528 |
/* |
| 529 |
* We have nothing to submit. Just cycle the folio through |
| 530 |
* writeback state to properly update xarray tags. |
| 531 |
*/ |
| 532 |
__folio_start_writeback(folio, keep_towrite); |
| 533 |
folio_end_writeback(folio); |
| 534 |
return 0; |
| 535 |
} |
| 536 |
|
| 537 |
bh = head = folio_buffers(folio); |
| 538 |
|
| 539 |
/* |
| 540 |
* If any blocks are being written to an encrypted file, encrypt them |
| 541 |
* into a bounce page. For simplicity, just encrypt until the last |
| 542 |
* block which might be needed. This may cause some unneeded blocks |
| 543 |
* (e.g. holes) to be unnecessarily encrypted, but this is rare and |
| 544 |
* can't happen in the common case of blocksize == PAGE_SIZE. |
| 545 |
*/ |
| 546 |
if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
| 547 |
gfp_t gfp_flags = GFP_NOFS; |
| 548 |
unsigned int enc_bytes = round_up(len, i_blocksize(inode)); |
| 549 |
struct page *bounce_page; |
| 550 |
|
| 551 |
/* |
| 552 |
* Since bounce page allocation uses a mempool, we can only use |
| 553 |
* a waiting mask (i.e. request guaranteed allocation) on the |
| 554 |
* first page of the bio. Otherwise it can deadlock. |
| 555 |
*/ |
| 556 |
if (io->io_bio) |
| 557 |
gfp_flags = GFP_NOWAIT; |
| 558 |
retry_encrypt: |
| 559 |
bounce_page = fscrypt_encrypt_pagecache_blocks(folio, |
| 560 |
enc_bytes, 0, gfp_flags); |
| 561 |
if (IS_ERR(bounce_page)) { |
| 562 |
ret = PTR_ERR(bounce_page); |
| 563 |
if (ret == -ENOMEM && |
| 564 |
(io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) { |
| 565 |
gfp_t new_gfp_flags = GFP_NOFS; |
| 566 |
if (io->io_bio) |
| 567 |
ext4_io_submit(io); |
| 568 |
else |
| 569 |
new_gfp_flags |= __GFP_NOFAIL; |
| 570 |
memalloc_retry_wait(gfp_flags); |
| 571 |
gfp_flags = new_gfp_flags; |
| 572 |
goto retry_encrypt; |
| 573 |
} |
| 574 |
|
| 575 |
printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret); |
| 576 |
folio_redirty_for_writepage(wbc, folio); |
| 577 |
do { |
| 578 |
if (buffer_async_write(bh)) { |
| 579 |
clear_buffer_async_write(bh); |
| 580 |
set_buffer_dirty(bh); |
| 581 |
} |
| 582 |
bh = bh->b_this_page; |
| 583 |
} while (bh != head); |
| 584 |
|
| 585 |
return ret; |
| 586 |
} |
| 587 |
io_folio = page_folio(bounce_page); |
| 588 |
} |
| 589 |
|
| 590 |
__folio_start_writeback(folio, keep_towrite); |
| 591 |
|
| 592 |
/* Now submit buffers to write */ |
| 593 |
do { |
| 594 |
if (!buffer_async_write(bh)) |
| 595 |
continue; |
| 596 |
io_submit_add_bh(io, inode, folio, io_folio, bh); |
| 597 |
} while ((bh = bh->b_this_page) != head); |
| 598 |
|
| 599 |
return 0; |
| 600 |
} |