]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/f2fs/data.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
f2fs: do not preallocate blocks which has wrong buffer
[mirror_ubuntu-artful-kernel.git] / fs / f2fs / data.c
CommitLineData
0a8165d7 1/*
eb47b800
JK		 2 * fs/f2fs/data.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/buffer_head.h>
14#include <linux/mpage.h>
15#include <linux/writeback.h>
16#include <linux/backing-dev.h>
8f46dcae 17#include <linux/pagevec.h>
eb47b800
JK		 18#include <linux/blkdev.h>
19#include <linux/bio.h>
690e4a3e 20#include <linux/prefetch.h>
e2e40f2c 21#include <linux/uio.h>
fe76b796
JK		 22#include <linux/mm.h>
23#include <linux/memcontrol.h>
f1e88660 24#include <linux/cleancache.h>
eb47b800
JK		 25
26#include "f2fs.h"
27#include "node.h"
28#include "segment.h"
db9f7c1a 29#include "trace.h"
848753aa 30#include <trace/events/f2fs.h>
eb47b800 31
36951b38
CY		 32static bool __is_cp_guaranteed(struct page *page)
33{
34 struct address_space *mapping = page->mapping;
35 struct inode *inode;
36 struct f2fs_sb_info *sbi;
37
38 if (!mapping)
39 return false;
40
41 inode = mapping->host;
42 sbi = F2FS_I_SB(inode);
43
44 if (inode->i_ino == F2FS_META_INO(sbi) ||
45 inode->i_ino == F2FS_NODE_INO(sbi) ||
46 S_ISDIR(inode->i_mode) ||
47 is_cold_data(page))
48 return true;
49 return false;
50}
51
4246a0b6 52static void f2fs_read_end_io(struct bio *bio)
93dfe2ac 53{
f568849e
LT		 54 struct bio_vec *bvec;
55 int i;
93dfe2ac 56
8b038c70 57#ifdef CONFIG_F2FS_FAULT_INJECTION
1ecc0c5c 58 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO))
8b038c70
CY		 59 bio->bi_error = -EIO;
60#endif
61
4375a336 62 if (f2fs_bio_encrypted(bio)) {
4246a0b6 63 if (bio->bi_error) {
0b81d077 64 fscrypt_release_ctx(bio->bi_private);
4375a336 65 } else {
0b81d077 66 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
4375a336
JK		 67 return;
68 }
69 }
70
12377024
CY		 71 bio_for_each_segment_all(bvec, bio, i) {
72 struct page *page = bvec->bv_page;
f1e88660 73
4246a0b6 74 if (!bio->bi_error) {
237c0790
JK		 75 if (!PageUptodate(page))
76 SetPageUptodate(page);
f1e88660
JK		 77 } else {
78 ClearPageUptodate(page);
79 SetPageError(page);
80 }
81 unlock_page(page);
82 }
f1e88660
JK		 83 bio_put(bio);
84}
85
4246a0b6 86static void f2fs_write_end_io(struct bio *bio)
93dfe2ac 87{
1b1f559f 88 struct f2fs_sb_info *sbi = bio->bi_private;
f568849e
LT		 89 struct bio_vec *bvec;
90 int i;
93dfe2ac 91
f568849e 92 bio_for_each_segment_all(bvec, bio, i) {
93dfe2ac 93 struct page *page = bvec->bv_page;
36951b38 94 enum count_type type = WB_DATA_TYPE(page);
93dfe2ac 95
0a595eba
JK		 96 if (IS_DUMMY_WRITTEN_PAGE(page)) {
97 set_page_private(page, (unsigned long)NULL);
98 ClearPagePrivate(page);
99 unlock_page(page);
100 mempool_free(page, sbi->write_io_dummy);
101
102 if (unlikely(bio->bi_error))
103 f2fs_stop_checkpoint(sbi, true);
104 continue;
105 }
106
0b81d077 107 fscrypt_pullback_bio_page(&page, true);
4375a336 108
4246a0b6 109 if (unlikely(bio->bi_error)) {
5114a97a 110 mapping_set_error(page->mapping, -EIO);
38f91ca8 111 f2fs_stop_checkpoint(sbi, true);
93dfe2ac 112 }
36951b38
CY		 113 dec_page_count(sbi, type);
114 clear_cold_data(page);
93dfe2ac 115 end_page_writeback(page);
f568849e 116 }
36951b38 117 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
f5730184 118 wq_has_sleeper(&sbi->cp_wait))
93dfe2ac
JK		 119 wake_up(&sbi->cp_wait);
120
121 bio_put(bio);
122}
123
3c62be17
JK		 124/*
125 * Return true, if pre_bio's bdev is same as its target device.
126 */
127struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
128 block_t blk_addr, struct bio *bio)
129{
130 struct block_device *bdev = sbi->sb->s_bdev;
131 int i;
132
133 for (i = 0; i < sbi->s_ndevs; i++) {
134 if (FDEV(i).start_blk <= blk_addr &&
135 FDEV(i).end_blk >= blk_addr) {
136 blk_addr -= FDEV(i).start_blk;
137 bdev = FDEV(i).bdev;
138 break;
139 }
140 }
141 if (bio) {
142 bio->bi_bdev = bdev;
143 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
144 }
145 return bdev;
146}
147
148int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
149{
150 int i;
151
152 for (i = 0; i < sbi->s_ndevs; i++)
153 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
154 return i;
155 return 0;
156}
157
158static bool __same_bdev(struct f2fs_sb_info *sbi,
159 block_t blk_addr, struct bio *bio)
160{
161 return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
162}
163
940a6d34
GZ		 164/*
165 * Low-level block read/write IO operations.
166 */
167static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
168 int npages, bool is_read)
169{
170 struct bio *bio;
171
740432f8 172 bio = f2fs_bio_alloc(npages);
940a6d34 173
3c62be17 174 f2fs_target_device(sbi, blk_addr, bio);
940a6d34 175 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
12377024 176 bio->bi_private = is_read ? NULL : sbi;
940a6d34
GZ		 177
178 return bio;
179}
180
4fc29c1a
LT		 181static inline void __submit_bio(struct f2fs_sb_info *sbi,
182 struct bio *bio, enum page_type type)
f5730184 183{
4fc29c1a 184 if (!is_read_io(bio_op(bio))) {
0a595eba
JK		 185 unsigned int start;
186
0bfd7a09 187 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
52763a4b 188 current->plug && (type == DATA || type == NODE))
19a5f5e2 189 blk_finish_plug(current->plug);
0a595eba
JK		 190
191 if (type != DATA && type != NODE)
192 goto submit_io;
193
194 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
195 start %= F2FS_IO_SIZE(sbi);
196
197 if (start == 0)
198 goto submit_io;
199
200 /* fill dummy pages */
201 for (; start < F2FS_IO_SIZE(sbi); start++) {
202 struct page *page =
203 mempool_alloc(sbi->write_io_dummy,
204 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
205 f2fs_bug_on(sbi, !page);
206
207 SetPagePrivate(page);
208 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
209 lock_page(page);
210 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
211 f2fs_bug_on(sbi, 1);
212 }
213 /*
214 * In the NODE case, we lose next block address chain. So, we
215 * need to do checkpoint in f2fs_sync_file.
216 */
217 if (type == NODE)
218 set_sbi_flag(sbi, SBI_NEED_CP);
19a5f5e2 219 }
0a595eba 220submit_io:
554b5125
JK		 221 if (is_read_io(bio_op(bio)))
222 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
223 else
224 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
4e49ea4a 225 submit_bio(bio);
f5730184
JK		 226}
227
458e6197 228static void __submit_merged_bio(struct f2fs_bio_info *io)
93dfe2ac 229{
458e6197 230 struct f2fs_io_info *fio = &io->fio;
93dfe2ac
JK		 231
232 if (!io->bio)
233 return;
234
554b5125
JK		 235 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
236
04d328de 237 if (is_read_io(fio->op))
554b5125 238 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
6a8f8ca5 239 else
554b5125 240 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
04d328de 241
4fc29c1a 242 __submit_bio(io->sbi, io->bio, fio->type);
93dfe2ac
JK		 243 io->bio = NULL;
244}
245
0c3a5797
CY		 246static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
247 struct page *page, nid_t ino)
0fd785eb 248{
0fd785eb
CY		 249 struct bio_vec *bvec;
250 struct page *target;
251 int i;
252
0c3a5797 253 if (!io->bio)
0fd785eb 254 return false;
0c3a5797
CY		 255
256 if (!inode && !page && !ino)
257 return true;
0fd785eb
CY		 258
259 bio_for_each_segment_all(bvec, io->bio, i) {
260
0b81d077 261 if (bvec->bv_page->mapping)
0fd785eb 262 target = bvec->bv_page;
0b81d077
JK		 263 else
264 target = fscrypt_control_page(bvec->bv_page);
0fd785eb 265
0c3a5797
CY		 266 if (inode && inode == target->mapping->host)
267 return true;
268 if (page && page == target)
269 return true;
270 if (ino && ino == ino_of_node(target))
0fd785eb 271 return true;
0fd785eb
CY		 272 }
273
0fd785eb
CY		 274 return false;
275}
276
0c3a5797
CY		 277static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
278 struct page *page, nid_t ino,
279 enum page_type type)
280{
281 enum page_type btype = PAGE_TYPE_OF_BIO(type);
282 struct f2fs_bio_info *io = &sbi->write_io[btype];
283 bool ret;
284
285 down_read(&io->io_rwsem);
286 ret = __has_merged_page(io, inode, page, ino);
287 up_read(&io->io_rwsem);
288 return ret;
289}
290
291static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
292 struct inode *inode, struct page *page,
293 nid_t ino, enum page_type type, int rw)
93dfe2ac
JK		 294{
295 enum page_type btype = PAGE_TYPE_OF_BIO(type);
296 struct f2fs_bio_info *io;
297
298 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
299
df0f8dc0 300 down_write(&io->io_rwsem);
458e6197 301
0c3a5797
CY		 302 if (!__has_merged_page(io, inode, page, ino))
303 goto out;
304
458e6197
JK		 305 /* change META to META_FLUSH in the checkpoint procedure */
306 if (type >= META_FLUSH) {
307 io->fio.type = META_FLUSH;
04d328de 308 io->fio.op = REQ_OP_WRITE;
70fd7614
CH		 309 io->fio.op_flags = REQ_PREFLUSH | REQ_META | REQ_PRIO;
310 if (!test_opt(sbi, NOBARRIER))
311 io->fio.op_flags |= REQ_FUA;
458e6197
JK		 312 }
313 __submit_merged_bio(io);
0c3a5797 314out:
df0f8dc0 315 up_write(&io->io_rwsem);
93dfe2ac
JK		 316}
317
0c3a5797
CY		 318void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
319 int rw)
320{
321 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
322}
323
324void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
325 struct inode *inode, struct page *page,
326 nid_t ino, enum page_type type, int rw)
327{
328 if (has_merged_page(sbi, inode, page, ino, type))
329 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
330}
331
406657dd
CY		 332void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
333{
334 f2fs_submit_merged_bio(sbi, DATA, WRITE);
335 f2fs_submit_merged_bio(sbi, NODE, WRITE);
336 f2fs_submit_merged_bio(sbi, META, WRITE);
337}
338
93dfe2ac
JK		 339/*
340 * Fill the locked page with data located in the block address.
341 * Return unlocked page.
342 */
05ca3632 343int f2fs_submit_page_bio(struct f2fs_io_info *fio)
93dfe2ac 344{
93dfe2ac 345 struct bio *bio;
0b81d077
JK		 346 struct page *page = fio->encrypted_page ?
347 fio->encrypted_page : fio->page;
93dfe2ac 348
2ace38e0 349 trace_f2fs_submit_page_bio(page, fio);
05ca3632 350 f2fs_trace_ios(fio, 0);
93dfe2ac
JK		 351
352 /* Allocate a new bio */
04d328de 353 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
93dfe2ac 354
09cbfeaf 355 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
93dfe2ac 356 bio_put(bio);
93dfe2ac
JK		 357 return -EFAULT;
358 }
04d328de 359 bio_set_op_attrs(bio, fio->op, fio->op_flags);
93dfe2ac 360
4fc29c1a 361 __submit_bio(fio->sbi, bio, fio->type);
93dfe2ac
JK		 362 return 0;
363}
364
0a595eba 365int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
93dfe2ac 366{
05ca3632 367 struct f2fs_sb_info *sbi = fio->sbi;
458e6197 368 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
93dfe2ac 369 struct f2fs_bio_info *io;
04d328de 370 bool is_read = is_read_io(fio->op);
4375a336 371 struct page *bio_page;
0a595eba 372 int err = 0;
93dfe2ac 373
940a6d34 374 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
93dfe2ac 375
7a9d7548
CY		 376 if (fio->old_blkaddr != NEW_ADDR)
377 verify_block_addr(sbi, fio->old_blkaddr);
378 verify_block_addr(sbi, fio->new_blkaddr);
93dfe2ac 379
36951b38
CY		 380 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
381
382 if (!is_read)
383 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
384
df0f8dc0 385 down_write(&io->io_rwsem);
93dfe2ac 386
7a9d7548 387 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
3c62be17
JK		 388 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
389 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
458e6197 390 __submit_merged_bio(io);
93dfe2ac
JK		 391alloc_new:
392 if (io->bio == NULL) {
0a595eba
JK		 393 if ((fio->type == DATA || fio->type == NODE) &&
394 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
395 err = -EAGAIN;
396 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
397 goto out_fail;
398 }
7a9d7548 399 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
664ba972 400 BIO_MAX_PAGES, is_read);
458e6197 401 io->fio = *fio;
93dfe2ac
JK		 402 }
403
09cbfeaf
KS		 404 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
405 PAGE_SIZE) {
458e6197 406 __submit_merged_bio(io);
93dfe2ac
JK		 407 goto alloc_new;
408 }
409
7a9d7548 410 io->last_block_in_bio = fio->new_blkaddr;
05ca3632 411 f2fs_trace_ios(fio, 0);
0a595eba 412out_fail:
df0f8dc0 413 up_write(&io->io_rwsem);
05ca3632 414 trace_f2fs_submit_page_mbio(fio->page, fio);
0a595eba 415 return err;
93dfe2ac
JK		 416}
417
46008c6d
CY		 418static void __set_data_blkaddr(struct dnode_of_data *dn)
419{
420 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
421 __le32 *addr_array;
422
423 /* Get physical address of data block */
424 addr_array = blkaddr_in_node(rn);
425 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
426}
427
0a8165d7 428/*
eb47b800
JK		 429 * Lock ordering for the change of data block address:
430 * ->data_page
431 * ->node_page
432 * update block addresses in the node page
433 */
216a620a 434void set_data_blkaddr(struct dnode_of_data *dn)
eb47b800 435{
46008c6d
CY		 436 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
437 __set_data_blkaddr(dn);
438 if (set_page_dirty(dn->node_page))
12719ae1 439 dn->node_changed = true;
eb47b800
JK		 440}
441
f28b3434
CY		 442void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
443{
444 dn->data_blkaddr = blkaddr;
445 set_data_blkaddr(dn);
446 f2fs_update_extent_cache(dn);
447}
448
46008c6d
CY		 449/* dn->ofs_in_node will be returned with up-to-date last block pointer */
450int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
eb47b800 451{
4081363f 452 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
eb47b800 453
46008c6d
CY		 454 if (!count)
455 return 0;
456
91942321 457 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
eb47b800 458 return -EPERM;
46008c6d 459 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
eb47b800
JK		 460 return -ENOSPC;
461
46008c6d
CY		 462 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
463 dn->ofs_in_node, count);
464
465 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
466
467 for (; count > 0; dn->ofs_in_node++) {
468 block_t blkaddr =
469 datablock_addr(dn->node_page, dn->ofs_in_node);
470 if (blkaddr == NULL_ADDR) {
471 dn->data_blkaddr = NEW_ADDR;
472 __set_data_blkaddr(dn);
473 count--;
474 }
475 }
476
477 if (set_page_dirty(dn->node_page))
478 dn->node_changed = true;
eb47b800
JK		 479 return 0;
480}
481
46008c6d
CY		 482/* Should keep dn->ofs_in_node unchanged */
483int reserve_new_block(struct dnode_of_data *dn)
484{
485 unsigned int ofs_in_node = dn->ofs_in_node;
486 int ret;
487
488 ret = reserve_new_blocks(dn, 1);
489 dn->ofs_in_node = ofs_in_node;
490 return ret;
491}
492
b600965c
HL		 493int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
494{
495 bool need_put = dn->inode_page ? false : true;
496 int err;
497
498 err = get_dnode_of_data(dn, index, ALLOC_NODE);
499 if (err)
500 return err;
a8865372 501
b600965c
HL		 502 if (dn->data_blkaddr == NULL_ADDR)
503 err = reserve_new_block(dn);
a8865372 504 if (err || need_put)
b600965c
HL		 505 f2fs_put_dnode(dn);
506 return err;
507}
508
759af1c9 509int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
eb47b800 510{
028a41e8 511 struct extent_info ei;
759af1c9 512 struct inode *inode = dn->inode;
028a41e8 513
759af1c9
FL		 514 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
515 dn->data_blkaddr = ei.blk + index - ei.fofs;
516 return 0;
429511cd 517 }
028a41e8 518
759af1c9 519 return f2fs_reserve_block(dn, index);
eb47b800
JK		 520}
521
a56c7c6f 522struct page *get_read_data_page(struct inode *inode, pgoff_t index,
04d328de 523 int op_flags, bool for_write)
eb47b800 524{
eb47b800
JK		 525 struct address_space *mapping = inode->i_mapping;
526 struct dnode_of_data dn;
527 struct page *page;
cb3bc9ee 528 struct extent_info ei;
eb47b800 529 int err;
cf04e8eb 530 struct f2fs_io_info fio = {
05ca3632 531 .sbi = F2FS_I_SB(inode),
cf04e8eb 532 .type = DATA,
04d328de
MC		 533 .op = REQ_OP_READ,
534 .op_flags = op_flags,
4375a336 535 .encrypted_page = NULL,
cf04e8eb 536 };
eb47b800 537
4375a336
JK		 538 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
539 return read_mapping_page(mapping, index, NULL);
540
a56c7c6f 541 page = f2fs_grab_cache_page(mapping, index, for_write);
650495de
JK		 542 if (!page)
543 return ERR_PTR(-ENOMEM);
544
cb3bc9ee
CY		 545 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
546 dn.data_blkaddr = ei.blk + index - ei.fofs;
547 goto got_it;
548 }
549
eb47b800 550 set_new_dnode(&dn, inode, NULL, NULL, 0);
266e97a8 551 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
86531d6b
JK		 552 if (err)
553 goto put_err;
eb47b800
JK		 554 f2fs_put_dnode(&dn);
555
6bacf52f 556 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
86531d6b
JK		 557 err = -ENOENT;
558 goto put_err;
650495de 559 }
cb3bc9ee 560got_it:
43f3eae1
JK		 561 if (PageUptodate(page)) {
562 unlock_page(page);
eb47b800 563 return page;
43f3eae1 564 }
eb47b800 565
d59ff4df
JK		 566 /*
567 * A new dentry page is allocated but not able to be written, since its
568 * new inode page couldn't be allocated due to -ENOSPC.
569 * In such the case, its blkaddr can be remained as NEW_ADDR.
570 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
571 */
572 if (dn.data_blkaddr == NEW_ADDR) {
09cbfeaf 573 zero_user_segment(page, 0, PAGE_SIZE);
237c0790
JK		 574 if (!PageUptodate(page))
575 SetPageUptodate(page);
43f3eae1 576 unlock_page(page);
d59ff4df
JK		 577 return page;
578 }
eb47b800 579
7a9d7548 580 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
05ca3632
JK		 581 fio.page = page;
582 err = f2fs_submit_page_bio(&fio);
393ff91f 583 if (err)
86531d6b 584 goto put_err;
43f3eae1 585 return page;
86531d6b
JK		 586
587put_err:
588 f2fs_put_page(page, 1);
589 return ERR_PTR(err);
43f3eae1
JK		 590}
591
592struct page *find_data_page(struct inode *inode, pgoff_t index)
593{
594 struct address_space *mapping = inode->i_mapping;
595 struct page *page;
596
597 page = find_get_page(mapping, index);
598 if (page && PageUptodate(page))
599 return page;
600 f2fs_put_page(page, 0);
601
70fd7614 602 page = get_read_data_page(inode, index, 0, false);
43f3eae1
JK		 603 if (IS_ERR(page))
604 return page;
605
606 if (PageUptodate(page))
607 return page;
608
609 wait_on_page_locked(page);
610 if (unlikely(!PageUptodate(page))) {
611 f2fs_put_page(page, 0);
612 return ERR_PTR(-EIO);
613 }
614 return page;
615}
616
617/*
618 * If it tries to access a hole, return an error.
619 * Because, the callers, functions in dir.c and GC, should be able to know
620 * whether this page exists or not.
621 */
a56c7c6f
JK		 622struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
623 bool for_write)
43f3eae1
JK		 624{
625 struct address_space *mapping = inode->i_mapping;
626 struct page *page;
627repeat:
70fd7614 628 page = get_read_data_page(inode, index, 0, for_write);
43f3eae1
JK		 629 if (IS_ERR(page))
630 return page;
393ff91f 631
43f3eae1 632 /* wait for read completion */
393ff91f 633 lock_page(page);
6bacf52f 634 if (unlikely(page->mapping != mapping)) {
afcb7ca0
JK		 635 f2fs_put_page(page, 1);
636 goto repeat;
eb47b800 637 }
1563ac75
CY		 638 if (unlikely(!PageUptodate(page))) {
639 f2fs_put_page(page, 1);
640 return ERR_PTR(-EIO);
641 }
eb47b800
JK		 642 return page;
643}
644
0a8165d7 645/*
eb47b800
JK		 646 * Caller ensures that this data page is never allocated.
647 * A new zero-filled data page is allocated in the page cache.
39936837 648 *
4f4124d0
CY		 649 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
650 * f2fs_unlock_op().
470f00e9
CY		 651 * Note that, ipage is set only by make_empty_dir, and if any error occur,
652 * ipage should be released by this function.
eb47b800 653 */
64aa7ed9 654struct page *get_new_data_page(struct inode *inode,
a8865372 655 struct page *ipage, pgoff_t index, bool new_i_size)
eb47b800 656{
eb47b800
JK		 657 struct address_space *mapping = inode->i_mapping;
658 struct page *page;
659 struct dnode_of_data dn;
660 int err;
7612118a 661
a56c7c6f 662 page = f2fs_grab_cache_page(mapping, index, true);
470f00e9
CY		 663 if (!page) {
664 /*
665 * before exiting, we should make sure ipage will be released
666 * if any error occur.
667 */
668 f2fs_put_page(ipage, 1);
01f28610 669 return ERR_PTR(-ENOMEM);
470f00e9 670 }
eb47b800 671
a8865372 672 set_new_dnode(&dn, inode, ipage, NULL, 0);
b600965c 673 err = f2fs_reserve_block(&dn, index);
01f28610
JK		 674 if (err) {
675 f2fs_put_page(page, 1);
eb47b800 676 return ERR_PTR(err);
a8865372 677 }
01f28610
JK		 678 if (!ipage)
679 f2fs_put_dnode(&dn);
eb47b800
JK		 680
681 if (PageUptodate(page))
01f28610 682 goto got_it;
eb47b800
JK		 683
684 if (dn.data_blkaddr == NEW_ADDR) {
09cbfeaf 685 zero_user_segment(page, 0, PAGE_SIZE);
237c0790
JK		 686 if (!PageUptodate(page))
687 SetPageUptodate(page);
eb47b800 688 } else {
4375a336 689 f2fs_put_page(page, 1);
a8865372 690
7612118a
JK		 691 /* if ipage exists, blkaddr should be NEW_ADDR */
692 f2fs_bug_on(F2FS_I_SB(inode), ipage);
693 page = get_lock_data_page(inode, index, true);
4375a336 694 if (IS_ERR(page))
7612118a 695 return page;
eb47b800 696 }
01f28610 697got_it:
9edcdabf 698 if (new_i_size && i_size_read(inode) <
ee6d182f 699 ((loff_t)(index + 1) << PAGE_SHIFT))
fc9581c8 700 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
eb47b800
JK		 701 return page;
702}
703
bfad7c2d
JK		 704static int __allocate_data_block(struct dnode_of_data *dn)
705{
4081363f 706 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
bfad7c2d 707 struct f2fs_summary sum;
bfad7c2d 708 struct node_info ni;
976e4c50 709 pgoff_t fofs;
46008c6d 710 blkcnt_t count = 1;
bfad7c2d 711
91942321 712 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
bfad7c2d 713 return -EPERM;
df6136ef
CY		 714
715 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
716 if (dn->data_blkaddr == NEW_ADDR)
717 goto alloc;
718
46008c6d 719 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
bfad7c2d
JK		 720 return -ENOSPC;
721
df6136ef 722alloc:
bfad7c2d
JK		 723 get_node_info(sbi, dn->nid, &ni);
724 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
725
df6136ef 726 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
6ae1be13 727 &sum, CURSEG_WARM_DATA);
216a620a 728 set_data_blkaddr(dn);
bfad7c2d 729
976e4c50 730 /* update i_size */
81ca7350 731 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
976e4c50 732 dn->ofs_in_node;
09cbfeaf 733 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
fc9581c8 734 f2fs_i_size_write(dn->inode,
09cbfeaf 735 ((loff_t)(fofs + 1) << PAGE_SHIFT));
bfad7c2d
JK		 736 return 0;
737}
738
c040ff9d
JK		 739static inline bool __force_buffered_io(struct inode *inode, int rw)
740{
741 return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) ||
742 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
743 F2FS_I_SB(inode)->s_ndevs);
744}
745
a7de6086 746int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
59b802e5 747{
b439b103 748 struct inode *inode = file_inode(iocb->ki_filp);
5b8db7fa 749 struct f2fs_map_blocks map;
a7de6086 750 int err = 0;
59b802e5 751
dc91de78
JK		 752 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
753 return 0;
754
0080c507 755 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
dfd02e4d
CY		 756 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
757 if (map.m_len > map.m_lblk)
758 map.m_len -= map.m_lblk;
759 else
760 map.m_len = 0;
761
da85985c 762 map.m_next_pgofs = NULL;
2a340760 763
24b84912 764 if (iocb->ki_flags & IOCB_DIRECT) {
a7de6086
JK		 765 err = f2fs_convert_inline_inode(inode);
766 if (err)
767 return err;
c040ff9d
JK		 768 return f2fs_map_blocks(inode, &map, 1,
769 __force_buffered_io(inode, WRITE) ?
770 F2FS_GET_BLOCK_PRE_AIO :
771 F2FS_GET_BLOCK_PRE_DIO);
24b84912
JK		 772 }
773 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
a7de6086
JK		 774 err = f2fs_convert_inline_inode(inode);
775 if (err)
776 return err;
b439b103 777 }
24b84912
JK		 778 if (!f2fs_has_inline_data(inode))
779 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
a7de6086 780 return err;
59b802e5
JK		 781}
782
0a8165d7 783/*
003a3e1d
JK		 784 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
785 * f2fs_map_blocks structure.
4f4124d0
CY		 786 * If original data blocks are allocated, then give them to blockdev.
787 * Otherwise,
788 * a. preallocate requested block addresses
789 * b. do not use extent cache for better performance
790 * c. give the block addresses to blockdev
eb47b800 791 */
d323d005 792int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
e2b4e2bc 793 int create, int flag)
eb47b800 794{
003a3e1d 795 unsigned int maxblocks = map->m_len;
eb47b800 796 struct dnode_of_data dn;
f9811703 797 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
ac6f1999 798 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
46008c6d 799 pgoff_t pgofs, end_offset, end;
bfad7c2d 800 int err = 0, ofs = 1;
46008c6d
CY		 801 unsigned int ofs_in_node, last_ofs_in_node;
802 blkcnt_t prealloc;
a2e7d1bf 803 struct extent_info ei;
7df3a431 804 block_t blkaddr;
eb47b800 805
dfd02e4d
CY		 806 if (!maxblocks)
807 return 0;
808
003a3e1d
JK		 809 map->m_len = 0;
810 map->m_flags = 0;
811
812 /* it only supports block size == page size */
813 pgofs = (pgoff_t)map->m_lblk;
46008c6d 814 end = pgofs + maxblocks;
eb47b800 815
24b84912 816 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
003a3e1d
JK		 817 map->m_pblk = ei.blk + pgofs - ei.fofs;
818 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
819 map->m_flags = F2FS_MAP_MAPPED;
bfad7c2d 820 goto out;
a2e7d1bf 821 }
bfad7c2d 822
4fe71e88 823next_dnode:
59b802e5 824 if (create)
3104af35 825 f2fs_lock_op(sbi);
eb47b800
JK		 826
827 /* When reading holes, we need its node page */
828 set_new_dnode(&dn, inode, NULL, NULL, 0);
bfad7c2d 829 err = get_dnode_of_data(&dn, pgofs, mode);
1ec79083 830 if (err) {
43473f96
CY		 831 if (flag == F2FS_GET_BLOCK_BMAP)
832 map->m_pblk = 0;
da85985c 833 if (err == -ENOENT) {
bfad7c2d 834 err = 0;
da85985c
CY		 835 if (map->m_next_pgofs)
836 *map->m_next_pgofs =
837 get_next_page_offset(&dn, pgofs);
838 }
bfad7c2d 839 goto unlock_out;
848753aa 840 }
973163fc 841
46008c6d 842 prealloc = 0;
230436b3 843 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
81ca7350 844 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
4fe71e88
CY		 845
846next_block:
847 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
848
849 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
973163fc 850 if (create) {
f9811703
CY		 851 if (unlikely(f2fs_cp_error(sbi))) {
852 err = -EIO;
4fe71e88 853 goto sync_out;
f9811703 854 }
24b84912 855 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
46008c6d
CY		 856 if (blkaddr == NULL_ADDR) {
857 prealloc++;
858 last_ofs_in_node = dn.ofs_in_node;
859 }
24b84912
JK		 860 } else {
861 err = __allocate_data_block(&dn);
6f2d8ed6 862 if (!err)
91942321 863 set_inode_flag(inode, FI_APPEND_WRITE);
24b84912 864 }
973163fc 865 if (err)
4fe71e88 866 goto sync_out;
973163fc 867 map->m_flags = F2FS_MAP_NEW;
4fe71e88 868 blkaddr = dn.data_blkaddr;
973163fc 869 } else {
43473f96
CY		 870 if (flag == F2FS_GET_BLOCK_BMAP) {
871 map->m_pblk = 0;
872 goto sync_out;
873 }
da85985c
CY		 874 if (flag == F2FS_GET_BLOCK_FIEMAP &&
875 blkaddr == NULL_ADDR) {
876 if (map->m_next_pgofs)
877 *map->m_next_pgofs = pgofs + 1;
878 }
973163fc 879 if (flag != F2FS_GET_BLOCK_FIEMAP ||
43473f96 880 blkaddr != NEW_ADDR)
4fe71e88 881 goto sync_out;
e2b4e2bc 882 }
e2b4e2bc 883 }
eb47b800 884
46008c6d
CY		 885 if (flag == F2FS_GET_BLOCK_PRE_AIO)
886 goto skip;
887
4fe71e88
CY		 888 if (map->m_len == 0) {
889 /* preallocated unwritten block should be mapped for fiemap. */
890 if (blkaddr == NEW_ADDR)
891 map->m_flags |= F2FS_MAP_UNWRITTEN;
892 map->m_flags |= F2FS_MAP_MAPPED;
893
894 map->m_pblk = blkaddr;
895 map->m_len = 1;
896 } else if ((map->m_pblk != NEW_ADDR &&
897 blkaddr == (map->m_pblk + ofs)) ||
b439b103 898 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
46008c6d 899 flag == F2FS_GET_BLOCK_PRE_DIO) {
4fe71e88
CY		 900 ofs++;
901 map->m_len++;
902 } else {
903 goto sync_out;
904 }
bfad7c2d 905
46008c6d 906skip:
bfad7c2d
JK		 907 dn.ofs_in_node++;
908 pgofs++;
909
46008c6d
CY		 910 /* preallocate blocks in batch for one dnode page */
911 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
912 (pgofs == end || dn.ofs_in_node == end_offset)) {
7df3a431 913
46008c6d
CY		 914 dn.ofs_in_node = ofs_in_node;
915 err = reserve_new_blocks(&dn, prealloc);
916 if (err)
917 goto sync_out;
bfad7c2d 918
46008c6d
CY		 919 map->m_len += dn.ofs_in_node - ofs_in_node;
920 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
921 err = -ENOSPC;
922 goto sync_out;
3104af35 923 }
46008c6d
CY		 924 dn.ofs_in_node = end_offset;
925 }
926
927 if (pgofs >= end)
928 goto sync_out;
929 else if (dn.ofs_in_node < end_offset)
930 goto next_block;
931
46008c6d
CY		 932 f2fs_put_dnode(&dn);
933
934 if (create) {
935 f2fs_unlock_op(sbi);
6f2d8ed6 936 f2fs_balance_fs(sbi, dn.node_changed);
eb47b800 937 }
46008c6d 938 goto next_dnode;
7df3a431 939
bfad7c2d 940sync_out:
eb47b800 941 f2fs_put_dnode(&dn);
bfad7c2d 942unlock_out:
2a340760 943 if (create) {
3104af35 944 f2fs_unlock_op(sbi);
6f2d8ed6 945 f2fs_balance_fs(sbi, dn.node_changed);
2a340760 946 }
bfad7c2d 947out:
003a3e1d 948 trace_f2fs_map_blocks(inode, map, err);
bfad7c2d 949 return err;
eb47b800
JK		 950}
951
003a3e1d 952static int __get_data_block(struct inode *inode, sector_t iblock,
da85985c
CY		 953 struct buffer_head *bh, int create, int flag,
954 pgoff_t *next_pgofs)
003a3e1d
JK		 955{
956 struct f2fs_map_blocks map;
a7de6086 957 int err;
003a3e1d
JK		 958
959 map.m_lblk = iblock;
960 map.m_len = bh->b_size >> inode->i_blkbits;
da85985c 961 map.m_next_pgofs = next_pgofs;
003a3e1d 962
a7de6086
JK		 963 err = f2fs_map_blocks(inode, &map, create, flag);
964 if (!err) {
003a3e1d
JK		 965 map_bh(bh, inode->i_sb, map.m_pblk);
966 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
967 bh->b_size = map.m_len << inode->i_blkbits;
968 }
a7de6086 969 return err;
003a3e1d
JK		 970}
971
ccfb3000 972static int get_data_block(struct inode *inode, sector_t iblock,
da85985c
CY		 973 struct buffer_head *bh_result, int create, int flag,
974 pgoff_t *next_pgofs)
e2b4e2bc 975{
da85985c
CY		 976 return __get_data_block(inode, iblock, bh_result, create,
977 flag, next_pgofs);
e2b4e2bc
CY		 978}
979
980static int get_data_block_dio(struct inode *inode, sector_t iblock,
ccfb3000
JK		 981 struct buffer_head *bh_result, int create)
982{
e2b4e2bc 983 return __get_data_block(inode, iblock, bh_result, create,
da85985c 984 F2FS_GET_BLOCK_DIO, NULL);
ccfb3000
JK		 985}
986
e2b4e2bc 987static int get_data_block_bmap(struct inode *inode, sector_t iblock,
ccfb3000
JK		 988 struct buffer_head *bh_result, int create)
989{
179448bf 990 /* Block number less than F2FS MAX BLOCKS */
e0afc4d6 991 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
179448bf
YH		 992 return -EFBIG;
993
e2b4e2bc 994 return __get_data_block(inode, iblock, bh_result, create,
da85985c 995 F2FS_GET_BLOCK_BMAP, NULL);
ccfb3000
JK		 996}
997
7f63eb77
JK		 998static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
999{
1000 return (offset >> inode->i_blkbits);
1001}
1002
1003static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1004{
1005 return (blk << inode->i_blkbits);
1006}
1007
9ab70134
JK		 1008int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1009 u64 start, u64 len)
1010{
7f63eb77
JK		 1011 struct buffer_head map_bh;
1012 sector_t start_blk, last_blk;
da85985c 1013 pgoff_t next_pgofs;
7f63eb77
JK		 1014 u64 logical = 0, phys = 0, size = 0;
1015 u32 flags = 0;
7f63eb77
JK		 1016 int ret = 0;
1017
1018 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1019 if (ret)
1020 return ret;
1021
67f8cf3c
JK		 1022 if (f2fs_has_inline_data(inode)) {
1023 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1024 if (ret != -EAGAIN)
1025 return ret;
1026 }
1027
5955102c 1028 inode_lock(inode);
de1475cc 1029
7f63eb77
JK		 1030 if (logical_to_blk(inode, len) == 0)
1031 len = blk_to_logical(inode, 1);
1032
1033 start_blk = logical_to_blk(inode, start);
1034 last_blk = logical_to_blk(inode, start + len - 1);
9a950d52 1035
7f63eb77
JK		 1036next:
1037 memset(&map_bh, 0, sizeof(struct buffer_head));
1038 map_bh.b_size = len;
1039
e2b4e2bc 1040 ret = get_data_block(inode, start_blk, &map_bh, 0,
da85985c 1041 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
7f63eb77
JK		 1042 if (ret)
1043 goto out;
1044
1045 /* HOLE */
1046 if (!buffer_mapped(&map_bh)) {
da85985c 1047 start_blk = next_pgofs;
58736fa6
CY		 1048
1049 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1050 F2FS_I_SB(inode)->max_file_blocks))
9a950d52 1051 goto prep_next;
58736fa6 1052
9a950d52
FL		 1053 flags |= FIEMAP_EXTENT_LAST;
1054 }
7f63eb77 1055
da5af127
CY		 1056 if (size) {
1057 if (f2fs_encrypted_inode(inode))
1058 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1059
9a950d52
FL		 1060 ret = fiemap_fill_next_extent(fieinfo, logical,
1061 phys, size, flags);
da5af127 1062 }
7f63eb77 1063
9a950d52
FL		 1064 if (start_blk > last_blk || ret)
1065 goto out;
7f63eb77 1066
9a950d52
FL		 1067 logical = blk_to_logical(inode, start_blk);
1068 phys = blk_to_logical(inode, map_bh.b_blocknr);
1069 size = map_bh.b_size;
1070 flags = 0;
1071 if (buffer_unwritten(&map_bh))
1072 flags = FIEMAP_EXTENT_UNWRITTEN;
7f63eb77 1073
9a950d52 1074 start_blk += logical_to_blk(inode, size);
7f63eb77 1075
9a950d52 1076prep_next:
7f63eb77
JK		 1077 cond_resched();
1078 if (fatal_signal_pending(current))
1079 ret = -EINTR;
1080 else
1081 goto next;
1082out:
1083 if (ret == 1)
1084 ret = 0;
1085
5955102c 1086 inode_unlock(inode);
7f63eb77 1087 return ret;
9ab70134
JK		 1088}
1089
6a7a3aed
WY		 1090static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1091 unsigned nr_pages)
78682f79
CY		 1092{
1093 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1094 struct fscrypt_ctx *ctx = NULL;
78682f79
CY		 1095 struct bio *bio;
1096
1097 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1098 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1099 if (IS_ERR(ctx))
1100 return ERR_CAST(ctx);
1101
1102 /* wait the page to be moved by cleaning */
1103 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1104 }
1105
1106 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1107 if (!bio) {
1108 if (ctx)
1109 fscrypt_release_ctx(ctx);
1110 return ERR_PTR(-ENOMEM);
1111 }
3c62be17 1112 f2fs_target_device(sbi, blkaddr, bio);
78682f79
CY		 1113 bio->bi_end_io = f2fs_read_end_io;
1114 bio->bi_private = ctx;
1115
1116 return bio;
1117}
1118
f1e88660
JK		 1119/*
1120 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1121 * Major change was from block_size == page_size in f2fs by default.
1122 */
1123static int f2fs_mpage_readpages(struct address_space *mapping,
1124 struct list_head *pages, struct page *page,
1125 unsigned nr_pages)
1126{
1127 struct bio *bio = NULL;
1128 unsigned page_idx;
1129 sector_t last_block_in_bio = 0;
1130 struct inode *inode = mapping->host;
1131 const unsigned blkbits = inode->i_blkbits;
1132 const unsigned blocksize = 1 << blkbits;
1133 sector_t block_in_file;
1134 sector_t last_block;
1135 sector_t last_block_in_file;
1136 sector_t block_nr;
f1e88660
JK		 1137 struct f2fs_map_blocks map;
1138
1139 map.m_pblk = 0;
1140 map.m_lblk = 0;
1141 map.m_len = 0;
1142 map.m_flags = 0;
da85985c 1143 map.m_next_pgofs = NULL;
f1e88660
JK		 1144
1145 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1146
1147 prefetchw(&page->flags);
1148 if (pages) {
939afa94 1149 page = list_last_entry(pages, struct page, lru);
f1e88660
JK		 1150 list_del(&page->lru);
1151 if (add_to_page_cache_lru(page, mapping,
8a5c743e
MH		 1152 page->index,
1153 readahead_gfp_mask(mapping)))
f1e88660
JK		 1154 goto next_page;
1155 }
1156
1157 block_in_file = (sector_t)page->index;
1158 last_block = block_in_file + nr_pages;
1159 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1160 blkbits;
1161 if (last_block > last_block_in_file)
1162 last_block = last_block_in_file;
1163
1164 /*
1165 * Map blocks using the previous result first.
1166 */
1167 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1168 block_in_file > map.m_lblk &&
1169 block_in_file < (map.m_lblk + map.m_len))
1170 goto got_it;
1171
1172 /*
1173 * Then do more f2fs_map_blocks() calls until we are
1174 * done with this page.
1175 */
1176 map.m_flags = 0;
1177
1178 if (block_in_file < last_block) {
1179 map.m_lblk = block_in_file;
1180 map.m_len = last_block - block_in_file;
1181
46c9e141 1182 if (f2fs_map_blocks(inode, &map, 0,
da85985c 1183 F2FS_GET_BLOCK_READ))
f1e88660
JK		 1184 goto set_error_page;
1185 }
1186got_it:
1187 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1188 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1189 SetPageMappedToDisk(page);
1190
1191 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1192 SetPageUptodate(page);
1193 goto confused;
1194 }
1195 } else {
09cbfeaf 1196 zero_user_segment(page, 0, PAGE_SIZE);
237c0790
JK		 1197 if (!PageUptodate(page))
1198 SetPageUptodate(page);
f1e88660
JK		 1199 unlock_page(page);
1200 goto next_page;
1201 }
1202
1203 /*
1204 * This page will go to BIO. Do we need to send this
1205 * BIO off first?
1206 */
3c62be17
JK		 1207 if (bio && (last_block_in_bio != block_nr - 1 ||
1208 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
f1e88660 1209submit_and_realloc:
4fc29c1a 1210 __submit_bio(F2FS_I_SB(inode), bio, DATA);
f1e88660
JK		 1211 bio = NULL;
1212 }
1213 if (bio == NULL) {
78682f79 1214 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1d353eb7
JK		 1215 if (IS_ERR(bio)) {
1216 bio = NULL;
f1e88660 1217 goto set_error_page;
4375a336 1218 }
04d328de 1219 bio_set_op_attrs(bio, REQ_OP_READ, 0);
f1e88660
JK		 1220 }
1221
1222 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1223 goto submit_and_realloc;
1224
1225 last_block_in_bio = block_nr;
1226 goto next_page;
1227set_error_page:
1228 SetPageError(page);
09cbfeaf 1229 zero_user_segment(page, 0, PAGE_SIZE);
f1e88660
JK		 1230 unlock_page(page);
1231 goto next_page;
1232confused:
1233 if (bio) {
4fc29c1a 1234 __submit_bio(F2FS_I_SB(inode), bio, DATA);
f1e88660
JK		 1235 bio = NULL;
1236 }
1237 unlock_page(page);
1238next_page:
1239 if (pages)
09cbfeaf 1240 put_page(page);
f1e88660
JK		 1241 }
1242 BUG_ON(pages && !list_empty(pages));
1243 if (bio)
4fc29c1a 1244 __submit_bio(F2FS_I_SB(inode), bio, DATA);
f1e88660
JK		 1245 return 0;
1246}
1247
eb47b800
JK		 1248static int f2fs_read_data_page(struct file *file, struct page *page)
1249{
9ffe0fb5 1250 struct inode *inode = page->mapping->host;
b3d208f9 1251 int ret = -EAGAIN;
9ffe0fb5 1252
c20e89cd
CY		 1253 trace_f2fs_readpage(page, DATA);
1254
e1c42045 1255 /* If the file has inline data, try to read it directly */
9ffe0fb5
HL		 1256 if (f2fs_has_inline_data(inode))
1257 ret = f2fs_read_inline_data(inode, page);
b3d208f9 1258 if (ret == -EAGAIN)
f1e88660 1259 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
9ffe0fb5 1260 return ret;
eb47b800
JK		 1261}
1262
1263static int f2fs_read_data_pages(struct file *file,
1264 struct address_space *mapping,
1265 struct list_head *pages, unsigned nr_pages)
1266{
9ffe0fb5 1267 struct inode *inode = file->f_mapping->host;
939afa94 1268 struct page *page = list_last_entry(pages, struct page, lru);
b8c29400
CY		 1269
1270 trace_f2fs_readpages(inode, page, nr_pages);
9ffe0fb5
HL		 1271
1272 /* If the file has inline data, skip readpages */
1273 if (f2fs_has_inline_data(inode))
1274 return 0;
1275
f1e88660 1276 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
eb47b800
JK		 1277}
1278
05ca3632 1279int do_write_data_page(struct f2fs_io_info *fio)
eb47b800 1280{
05ca3632 1281 struct page *page = fio->page;
eb47b800 1282 struct inode *inode = page->mapping->host;
eb47b800
JK		 1283 struct dnode_of_data dn;
1284 int err = 0;
1285
1286 set_new_dnode(&dn, inode, NULL, NULL, 0);
266e97a8 1287 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
eb47b800
JK		 1288 if (err)
1289 return err;
1290
28bc106b 1291 fio->old_blkaddr = dn.data_blkaddr;
eb47b800
JK		 1292
1293 /* This page is already truncated */
7a9d7548 1294 if (fio->old_blkaddr == NULL_ADDR) {
2bca1e23 1295 ClearPageUptodate(page);
eb47b800 1296 goto out_writepage;
2bca1e23 1297 }
eb47b800 1298
4375a336 1299 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
b32e4482 1300 gfp_t gfp_flags = GFP_NOFS;
08b39fbd
CY		 1301
1302 /* wait for GCed encrypted page writeback */
1303 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
7a9d7548 1304 fio->old_blkaddr);
b32e4482
JK		 1305retry_encrypt:
1306 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
9c4bb8a3
DG		 1307 PAGE_SIZE, 0,
1308 fio->page->index,
1309 gfp_flags);
4375a336
JK		 1310 if (IS_ERR(fio->encrypted_page)) {
1311 err = PTR_ERR(fio->encrypted_page);
b32e4482
JK		 1312 if (err == -ENOMEM) {
1313 /* flush pending ios and wait for a while */
1314 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1315 congestion_wait(BLK_RW_ASYNC, HZ/50);
1316 gfp_flags |= __GFP_NOFAIL;
1317 err = 0;
1318 goto retry_encrypt;
1319 }
4375a336
JK		 1320 goto out_writepage;
1321 }
1322 }
1323
eb47b800
JK		 1324 set_page_writeback(page);
1325
1326 /*
1327 * If current allocation needs SSR,
1328 * it had better in-place writes for updated data.
1329 */
7a9d7548 1330 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
b25958b6 1331 !is_cold_data(page) &&
2da3e027 1332 !IS_ATOMIC_WRITTEN_PAGE(page) &&
b25958b6 1333 need_inplace_update(inode))) {
05ca3632 1334 rewrite_data_page(fio);
91942321 1335 set_inode_flag(inode, FI_UPDATE_WRITE);
8ce67cb0 1336 trace_f2fs_do_write_data_page(page, IPU);
eb47b800 1337 } else {
05ca3632 1338 write_data_page(&dn, fio);
8ce67cb0 1339 trace_f2fs_do_write_data_page(page, OPU);
91942321 1340 set_inode_flag(inode, FI_APPEND_WRITE);
3c6c2beb 1341 if (page->index == 0)
91942321 1342 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
eb47b800
JK		 1343 }
1344out_writepage:
1345 f2fs_put_dnode(&dn);
1346 return err;
1347}
1348
1349static int f2fs_write_data_page(struct page *page,
1350 struct writeback_control *wbc)
1351{
1352 struct inode *inode = page->mapping->host;
4081363f 1353 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
eb47b800
JK		 1354 loff_t i_size = i_size_read(inode);
1355 const pgoff_t end_index = ((unsigned long long) i_size)
09cbfeaf 1356 >> PAGE_SHIFT;
26de9b11 1357 loff_t psize = (page->index + 1) << PAGE_SHIFT;
9ffe0fb5 1358 unsigned offset = 0;
39936837 1359 bool need_balance_fs = false;
eb47b800 1360 int err = 0;
458e6197 1361 struct f2fs_io_info fio = {
05ca3632 1362 .sbi = sbi,
458e6197 1363 .type = DATA,
04d328de 1364 .op = REQ_OP_WRITE,
7637241e 1365 .op_flags = wbc_to_write_flags(wbc),
05ca3632 1366 .page = page,
4375a336 1367 .encrypted_page = NULL,
458e6197 1368 };
eb47b800 1369
ecda0de3
CY		 1370 trace_f2fs_writepage(page, DATA);
1371
eb47b800 1372 if (page->index < end_index)
39936837 1373 goto write;
eb47b800
JK		 1374
1375 /*
1376 * If the offset is out-of-range of file size,
1377 * this page does not have to be written to disk.
1378 */
09cbfeaf 1379 offset = i_size & (PAGE_SIZE - 1);
76f60268 1380 if ((page->index >= end_index + 1) || !offset)
39936837 1381 goto out;
eb47b800 1382
09cbfeaf 1383 zero_user_segment(page, offset, PAGE_SIZE);
39936837 1384write:
caf0047e 1385 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
eb47b800 1386 goto redirty_out;
1e84371f
JK		 1387 if (f2fs_is_drop_cache(inode))
1388 goto out;
e6e5f561
JK		 1389 /* we should not write 0'th page having journal header */
1390 if (f2fs_is_volatile_file(inode) && (!page->index ||
1391 (!wbc->for_reclaim &&
1392 available_free_memory(sbi, BASE_CHECK))))
1e84371f 1393 goto redirty_out;
eb47b800 1394
cf779cab
JK		 1395 /* we should bypass data pages to proceed the kworkder jobs */
1396 if (unlikely(f2fs_cp_error(sbi))) {
7f319975 1397 mapping_set_error(page->mapping, -EIO);
a7ffdbe2 1398 goto out;
cf779cab
JK		 1399 }
1400
39936837 1401 /* Dentry blocks are controlled by checkpoint */
eb47b800 1402 if (S_ISDIR(inode->i_mode)) {
05ca3632 1403 err = do_write_data_page(&fio);
8618b881
JK		 1404 goto done;
1405 }
9ffe0fb5 1406
8618b881 1407 if (!wbc->for_reclaim)
39936837 1408 need_balance_fs = true;
7f3037a5 1409 else if (has_not_enough_free_secs(sbi, 0, 0))
39936837 1410 goto redirty_out;
eb47b800 1411
b3d208f9 1412 err = -EAGAIN;
8618b881 1413 f2fs_lock_op(sbi);
b3d208f9
JK		 1414 if (f2fs_has_inline_data(inode))
1415 err = f2fs_write_inline_data(inode, page);
1416 if (err == -EAGAIN)
05ca3632 1417 err = do_write_data_page(&fio);
26de9b11
JK		 1418 if (F2FS_I(inode)->last_disk_size < psize)
1419 F2FS_I(inode)->last_disk_size = psize;
8618b881
JK		 1420 f2fs_unlock_op(sbi);
1421done:
1422 if (err && err != -ENOENT)
1423 goto redirty_out;
eb47b800 1424
39936837 1425out:
a7ffdbe2 1426 inode_dec_dirty_pages(inode);
2bca1e23
JK		 1427 if (err)
1428 ClearPageUptodate(page);
0c3a5797
CY		 1429
1430 if (wbc->for_reclaim) {
1431 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1432 remove_dirty_inode(inode);
1433 }
1434
eb47b800 1435 unlock_page(page);
2c4db1a6 1436 f2fs_balance_fs(sbi, need_balance_fs);
0c3a5797
CY		 1437
1438 if (unlikely(f2fs_cp_error(sbi)))
2aea39ec 1439 f2fs_submit_merged_bio(sbi, DATA, WRITE);
0c3a5797 1440
eb47b800
JK		 1441 return 0;
1442
eb47b800 1443redirty_out:
76f60268 1444 redirty_page_for_writepage(wbc, page);
0002b61b
CY		 1445 if (!err)
1446 return AOP_WRITEPAGE_ACTIVATE;
b230e6ca
JK		 1447 unlock_page(page);
1448 return err;
fa9150a8
NJ		 1449}
1450
8f46dcae
CY		 1451/*
1452 * This function was copied from write_cche_pages from mm/page-writeback.c.
1453 * The major change is making write step of cold data page separately from
1454 * warm/hot data page.
1455 */
1456static int f2fs_write_cache_pages(struct address_space *mapping,
b230e6ca 1457 struct writeback_control *wbc)
8f46dcae
CY		 1458{
1459 int ret = 0;
1460 int done = 0;
1461 struct pagevec pvec;
1462 int nr_pages;
1463 pgoff_t uninitialized_var(writeback_index);
1464 pgoff_t index;
1465 pgoff_t end; /* Inclusive */
1466 pgoff_t done_index;
1467 int cycled;
1468 int range_whole = 0;
1469 int tag;
6ca56ca4 1470 int nwritten = 0;
8f46dcae
CY		 1471
1472 pagevec_init(&pvec, 0);
46ae957f 1473
8f46dcae
CY		 1474 if (wbc->range_cyclic) {
1475 writeback_index = mapping->writeback_index; /* prev offset */
1476 index = writeback_index;
1477 if (index == 0)
1478 cycled = 1;
1479 else
1480 cycled = 0;
1481 end = -1;
1482 } else {
09cbfeaf
KS		 1483 index = wbc->range_start >> PAGE_SHIFT;
1484 end = wbc->range_end >> PAGE_SHIFT;
8f46dcae
CY		 1485 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1486 range_whole = 1;
1487 cycled = 1; /* ignore range_cyclic tests */
1488 }
1489 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1490 tag = PAGECACHE_TAG_TOWRITE;
1491 else
1492 tag = PAGECACHE_TAG_DIRTY;
1493retry:
1494 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1495 tag_pages_for_writeback(mapping, index, end);
1496 done_index = index;
1497 while (!done && (index <= end)) {
1498 int i;
1499
1500 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1501 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1502 if (nr_pages == 0)
1503 break;
1504
1505 for (i = 0; i < nr_pages; i++) {
1506 struct page *page = pvec.pages[i];
1507
1508 if (page->index > end) {
1509 done = 1;
1510 break;
1511 }
1512
1513 done_index = page->index;
1514
1515 lock_page(page);
1516
1517 if (unlikely(page->mapping != mapping)) {
1518continue_unlock:
1519 unlock_page(page);
1520 continue;
1521 }
1522
1523 if (!PageDirty(page)) {
1524 /* someone wrote it for us */
1525 goto continue_unlock;
1526 }
1527
8f46dcae
CY		 1528 if (PageWriteback(page)) {
1529 if (wbc->sync_mode != WB_SYNC_NONE)
fec1d657
JK		 1530 f2fs_wait_on_page_writeback(page,
1531 DATA, true);
8f46dcae
CY		 1532 else
1533 goto continue_unlock;
1534 }
1535
1536 BUG_ON(PageWriteback(page));
1537 if (!clear_page_dirty_for_io(page))
1538 goto continue_unlock;
1539
b230e6ca 1540 ret = mapping->a_ops->writepage(page, wbc);
8f46dcae 1541 if (unlikely(ret)) {
0002b61b
CY		 1542 /*
1543 * keep nr_to_write, since vfs uses this to
1544 * get # of written pages.
1545 */
1546 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1547 unlock_page(page);
1548 ret = 0;
1549 continue;
1550 }
b230e6ca
JK		 1551 done_index = page->index + 1;
1552 done = 1;
1553 break;
6ca56ca4
CY		 1554 } else {
1555 nwritten++;
8f46dcae
CY		 1556 }
1557
1558 if (--wbc->nr_to_write <= 0 &&
1559 wbc->sync_mode == WB_SYNC_NONE) {
1560 done = 1;
1561 break;
1562 }
1563 }
1564 pagevec_release(&pvec);
1565 cond_resched();
1566 }
1567
8f46dcae
CY		 1568 if (!cycled && !done) {
1569 cycled = 1;
1570 index = 0;
1571 end = writeback_index - 1;
1572 goto retry;
1573 }
1574 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1575 mapping->writeback_index = done_index;
1576
6ca56ca4
CY		 1577 if (nwritten)
1578 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1579 NULL, 0, DATA, WRITE);
1580
8f46dcae
CY		 1581 return ret;
1582}
1583
25ca923b 1584static int f2fs_write_data_pages(struct address_space *mapping,
eb47b800
JK		 1585 struct writeback_control *wbc)
1586{
1587 struct inode *inode = mapping->host;
4081363f 1588 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
9dfa1baf 1589 struct blk_plug plug;
eb47b800 1590 int ret;
eb47b800 1591
cfb185a1 1592 /* deal with chardevs and other special file */
1593 if (!mapping->a_ops->writepage)
1594 return 0;
1595
6a290544
CY		 1596 /* skip writing if there is no dirty page in this inode */
1597 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1598 return 0;
1599
a1257023
JK		 1600 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1601 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1602 available_free_memory(sbi, DIRTY_DENTS))
1603 goto skip_write;
1604
d323d005 1605 /* skip writing during file defragment */
91942321 1606 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
d323d005
CY		 1607 goto skip_write;
1608
d5669f7b
JK		 1609 /* during POR, we don't need to trigger writepage at all. */
1610 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1611 goto skip_write;
1612
d31c7c3f
YH		 1613 trace_f2fs_writepages(mapping->host, wbc, DATA);
1614
9dfa1baf 1615 blk_start_plug(&plug);
b230e6ca 1616 ret = f2fs_write_cache_pages(mapping, wbc);
9dfa1baf 1617 blk_finish_plug(&plug);
28ea6162
JK		 1618 /*
1619 * if some pages were truncated, we cannot guarantee its mapping->host
1620 * to detect pending bios.
1621 */
458e6197 1622
c227f912 1623 remove_dirty_inode(inode);
eb47b800 1624 return ret;
d3baf95d
JK		 1625
1626skip_write:
a7ffdbe2 1627 wbc->pages_skipped += get_dirty_pages(inode);
d31c7c3f 1628 trace_f2fs_writepages(mapping->host, wbc, DATA);
d3baf95d 1629 return 0;
eb47b800
JK		 1630}
1631
3aab8f82
CY		 1632static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1633{
1634 struct inode *inode = mapping->host;
819d9153 1635 loff_t i_size = i_size_read(inode);
3aab8f82 1636
819d9153
JK		 1637 if (to > i_size) {
1638 truncate_pagecache(inode, i_size);
1639 truncate_blocks(inode, i_size, true);
3aab8f82
CY		 1640 }
1641}
1642
2aadac08
JK		 1643static int prepare_write_begin(struct f2fs_sb_info *sbi,
1644 struct page *page, loff_t pos, unsigned len,
1645 block_t *blk_addr, bool *node_changed)
1646{
1647 struct inode *inode = page->mapping->host;
1648 pgoff_t index = page->index;
1649 struct dnode_of_data dn;
1650 struct page *ipage;
b4d07a3e
JK		 1651 bool locked = false;
1652 struct extent_info ei;
2aadac08
JK		 1653 int err = 0;
1654
24b84912
JK		 1655 /*
1656 * we already allocated all the blocks, so we don't need to get
1657 * the block addresses when there is no need to fill the page.
1658 */
dc91de78
JK		 1659 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1660 !is_inode_flag_set(inode, FI_NO_PREALLOC))
24b84912
JK		 1661 return 0;
1662
b4d07a3e 1663 if (f2fs_has_inline_data(inode) ||
09cbfeaf 1664 (pos & PAGE_MASK) >= i_size_read(inode)) {
b4d07a3e
JK		 1665 f2fs_lock_op(sbi);
1666 locked = true;
1667 }
1668restart:
2aadac08
JK		 1669 /* check inline_data */
1670 ipage = get_node_page(sbi, inode->i_ino);
1671 if (IS_ERR(ipage)) {
1672 err = PTR_ERR(ipage);
1673 goto unlock_out;
1674 }
1675
1676 set_new_dnode(&dn, inode, ipage, ipage, 0);
1677
1678 if (f2fs_has_inline_data(inode)) {
1679 if (pos + len <= MAX_INLINE_DATA) {
1680 read_inline_data(page, ipage);
91942321 1681 set_inode_flag(inode, FI_DATA_EXIST);
ab47036d
CY		 1682 if (inode->i_nlink)
1683 set_inline_node(ipage);
2aadac08
JK		 1684 } else {
1685 err = f2fs_convert_inline_page(&dn, page);
1686 if (err)
b4d07a3e
JK		 1687 goto out;
1688 if (dn.data_blkaddr == NULL_ADDR)
1689 err = f2fs_get_block(&dn, index);
1690 }
1691 } else if (locked) {
1692 err = f2fs_get_block(&dn, index);
1693 } else {
1694 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1695 dn.data_blkaddr = ei.blk + index - ei.fofs;
1696 } else {
b4d07a3e
JK		 1697 /* hole case */
1698 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
4da7bf5a 1699 if (err || dn.data_blkaddr == NULL_ADDR) {
b4d07a3e
JK		 1700 f2fs_put_dnode(&dn);
1701 f2fs_lock_op(sbi);
1702 locked = true;
1703 goto restart;
1704 }
2aadac08
JK		 1705 }
1706 }
b4d07a3e 1707
2aadac08
JK		 1708 /* convert_inline_page can make node_changed */
1709 *blk_addr = dn.data_blkaddr;
1710 *node_changed = dn.node_changed;
b4d07a3e 1711out:
2aadac08
JK		 1712 f2fs_put_dnode(&dn);
1713unlock_out:
b4d07a3e
JK		 1714 if (locked)
1715 f2fs_unlock_op(sbi);
2aadac08
JK		 1716 return err;
1717}
1718
eb47b800
JK		 1719static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1720 loff_t pos, unsigned len, unsigned flags,
1721 struct page **pagep, void **fsdata)
1722{
1723 struct inode *inode = mapping->host;
4081363f 1724 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
86531d6b 1725 struct page *page = NULL;
09cbfeaf 1726 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2aadac08
JK		 1727 bool need_balance = false;
1728 block_t blkaddr = NULL_ADDR;
eb47b800
JK		 1729 int err = 0;
1730
62aed044
CY		 1731 trace_f2fs_write_begin(inode, pos, len, flags);
1732
5f727395
JK		 1733 /*
1734 * We should check this at this moment to avoid deadlock on inode page
1735 * and #0 page. The locking rule for inline_data conversion should be:
1736 * lock_page(page #0) -> lock_page(inode_page)
1737 */
1738 if (index != 0) {
1739 err = f2fs_convert_inline_inode(inode);
1740 if (err)
1741 goto fail;
1742 }
afcb7ca0 1743repeat:
eb47b800 1744 page = grab_cache_page_write_begin(mapping, index, flags);
3aab8f82
CY		 1745 if (!page) {
1746 err = -ENOMEM;
1747 goto fail;
1748 }
d5f66990 1749
eb47b800
JK		 1750 *pagep = page;
1751
2aadac08
JK		 1752 err = prepare_write_begin(sbi, page, pos, len,
1753 &blkaddr, &need_balance);
9ba69cf9 1754 if (err)
2aadac08 1755 goto fail;
9ba69cf9 1756
7f3037a5 1757 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2a340760 1758 unlock_page(page);
2c4db1a6 1759 f2fs_balance_fs(sbi, true);
2a340760
JK		 1760 lock_page(page);
1761 if (page->mapping != mapping) {
1762 /* The page got truncated from under us */
1763 f2fs_put_page(page, 1);
1764 goto repeat;
1765 }
1766 }
1767
fec1d657 1768 f2fs_wait_on_page_writeback(page, DATA, false);
b3d208f9 1769
08b39fbd
CY		 1770 /* wait for GCed encrypted page writeback */
1771 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
2aadac08 1772 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
08b39fbd 1773
649d7df2
JK		 1774 if (len == PAGE_SIZE || PageUptodate(page))
1775 return 0;
eb47b800 1776
746e2403
YH		 1777 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1778 zero_user_segment(page, len, PAGE_SIZE);
1779 return 0;
1780 }
1781
2aadac08 1782 if (blkaddr == NEW_ADDR) {
09cbfeaf 1783 zero_user_segment(page, 0, PAGE_SIZE);
649d7df2 1784 SetPageUptodate(page);
eb47b800 1785 } else {
78682f79 1786 struct bio *bio;
d54c795b 1787
78682f79
CY		 1788 bio = f2fs_grab_bio(inode, blkaddr, 1);
1789 if (IS_ERR(bio)) {
1790 err = PTR_ERR(bio);
3aab8f82 1791 goto fail;
eb47b800 1792 }
70fd7614 1793 bio->bi_opf = REQ_OP_READ;
78682f79
CY		 1794 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1795 bio_put(bio);
1796 err = -EFAULT;
1797 goto fail;
1798 }
1799
4fc29c1a 1800 __submit_bio(sbi, bio, DATA);
d54c795b 1801
393ff91f 1802 lock_page(page);
6bacf52f 1803 if (unlikely(page->mapping != mapping)) {
afcb7ca0
JK		 1804 f2fs_put_page(page, 1);
1805 goto repeat;
eb47b800 1806 }
1563ac75
CY		 1807 if (unlikely(!PageUptodate(page))) {
1808 err = -EIO;
1809 goto fail;
4375a336 1810 }
eb47b800 1811 }
eb47b800 1812 return 0;
9ba69cf9 1813
3aab8f82 1814fail:
86531d6b 1815 f2fs_put_page(page, 1);
3aab8f82
CY		 1816 f2fs_write_failed(mapping, pos + len);
1817 return err;
eb47b800
JK		 1818}
1819
a1dd3c13
JK		 1820static int f2fs_write_end(struct file *file,
1821 struct address_space *mapping,
1822 loff_t pos, unsigned len, unsigned copied,
1823 struct page *page, void *fsdata)
1824{
1825 struct inode *inode = page->mapping->host;
1826
dfb2bf38
CY		 1827 trace_f2fs_write_end(inode, pos, len, copied);
1828
649d7df2
JK		 1829 /*
1830 * This should be come from len == PAGE_SIZE, and we expect copied
1831 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1832 * let generic_perform_write() try to copy data again through copied=0.
1833 */
1834 if (!PageUptodate(page)) {
746e2403 1835 if (unlikely(copied != len))
649d7df2
JK		 1836 copied = 0;
1837 else
1838 SetPageUptodate(page);
1839 }
1840 if (!copied)
1841 goto unlock_out;
1842
34ba94ba 1843 set_page_dirty(page);
a1dd3c13 1844
fc9581c8
JK		 1845 if (pos + copied > i_size_read(inode))
1846 f2fs_i_size_write(inode, pos + copied);
649d7df2 1847unlock_out:
3024c9a1 1848 f2fs_put_page(page, 1);
d0239e1b 1849 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
a1dd3c13
JK		 1850 return copied;
1851}
1852
6f673763
OS		 1853static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1854 loff_t offset)
944fcfc1
JK		 1855{
1856 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
944fcfc1 1857
944fcfc1
JK		 1858 if (offset & blocksize_mask)
1859 return -EINVAL;
1860
5b46f25d
AV		 1861 if (iov_iter_alignment(iter) & blocksize_mask)
1862 return -EINVAL;
1863
944fcfc1
JK		 1864 return 0;
1865}
1866
c8b8e32d 1867static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
eb47b800 1868{
b439b103 1869 struct address_space *mapping = iocb->ki_filp->f_mapping;
3aab8f82
CY		 1870 struct inode *inode = mapping->host;
1871 size_t count = iov_iter_count(iter);
c8b8e32d 1872 loff_t offset = iocb->ki_pos;
82e0a5aa 1873 int rw = iov_iter_rw(iter);
3aab8f82 1874 int err;
944fcfc1 1875
b439b103 1876 err = check_direct_IO(inode, iter, offset);
b9d777b8
JK		 1877 if (err)
1878 return err;
9ffe0fb5 1879
c040ff9d 1880 if (__force_buffered_io(inode, rw))
36abef4e 1881 return 0;
fcc85a4d 1882
5302fb00 1883 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
70407fad 1884
82e0a5aa 1885 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
c8b8e32d 1886 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
82e0a5aa
CY		 1887 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1888
1889 if (rw == WRITE) {
6bfc4919 1890 if (err > 0)
91942321 1891 set_inode_flag(inode, FI_UPDATE_WRITE);
6bfc4919
JK		 1892 else if (err < 0)
1893 f2fs_write_failed(mapping, offset + count);
1894 }
70407fad 1895
5302fb00 1896 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
70407fad 1897
3aab8f82 1898 return err;
eb47b800
JK		 1899}
1900
487261f3
CY		 1901void f2fs_invalidate_page(struct page *page, unsigned int offset,
1902 unsigned int length)
eb47b800
JK		 1903{
1904 struct inode *inode = page->mapping->host;
487261f3 1905 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
a7ffdbe2 1906
487261f3 1907 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
09cbfeaf 1908 (offset % PAGE_SIZE || length != PAGE_SIZE))
a7ffdbe2
JK		 1909 return;
1910
487261f3 1911 if (PageDirty(page)) {
933439c8 1912 if (inode->i_ino == F2FS_META_INO(sbi)) {
487261f3 1913 dec_page_count(sbi, F2FS_DIRTY_META);
933439c8 1914 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
487261f3 1915 dec_page_count(sbi, F2FS_DIRTY_NODES);
933439c8 1916 } else {
487261f3 1917 inode_dec_dirty_pages(inode);
933439c8
CY		 1918 remove_dirty_inode(inode);
1919 }
487261f3 1920 }
decd36b6
CY		 1921
1922 /* This is atomic written page, keep Private */
1923 if (IS_ATOMIC_WRITTEN_PAGE(page))
1924 return;
1925
23dc974e 1926 set_page_private(page, 0);
eb47b800
JK		 1927 ClearPagePrivate(page);
1928}
1929
487261f3 1930int f2fs_release_page(struct page *page, gfp_t wait)
eb47b800 1931{
f68daeeb
JK		 1932 /* If this is dirty page, keep PagePrivate */
1933 if (PageDirty(page))
1934 return 0;
1935
decd36b6
CY		 1936 /* This is atomic written page, keep Private */
1937 if (IS_ATOMIC_WRITTEN_PAGE(page))
1938 return 0;
1939
23dc974e 1940 set_page_private(page, 0);
eb47b800 1941 ClearPagePrivate(page);
c3850aa1 1942 return 1;
eb47b800
JK		 1943}
1944
fe76b796
JK		 1945/*
1946 * This was copied from __set_page_dirty_buffers which gives higher performance
1947 * in very high speed storages. (e.g., pmem)
1948 */
1949void f2fs_set_page_dirty_nobuffers(struct page *page)
1950{
1951 struct address_space *mapping = page->mapping;
1952 unsigned long flags;
1953
1954 if (unlikely(!mapping))
1955 return;
1956
1957 spin_lock(&mapping->private_lock);
1958 lock_page_memcg(page);
1959 SetPageDirty(page);
1960 spin_unlock(&mapping->private_lock);
1961
1962 spin_lock_irqsave(&mapping->tree_lock, flags);
1963 WARN_ON_ONCE(!PageUptodate(page));
1964 account_page_dirtied(page, mapping);
1965 radix_tree_tag_set(&mapping->page_tree,
1966 page_index(page), PAGECACHE_TAG_DIRTY);
1967 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1968 unlock_page_memcg(page);
1969
1970 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1971 return;
1972}
1973
eb47b800
JK		 1974static int f2fs_set_data_page_dirty(struct page *page)
1975{
1976 struct address_space *mapping = page->mapping;
1977 struct inode *inode = mapping->host;
1978
26c6b887
JK		 1979 trace_f2fs_set_page_dirty(page, DATA);
1980
237c0790
JK		 1981 if (!PageUptodate(page))
1982 SetPageUptodate(page);
34ba94ba 1983
5fe45743 1984 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
decd36b6
CY		 1985 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1986 register_inmem_page(inode, page);
1987 return 1;
1988 }
1989 /*
1990 * Previously, this page has been registered, we just
1991 * return here.
1992 */
1993 return 0;
34ba94ba
JK		 1994 }
1995
eb47b800 1996 if (!PageDirty(page)) {
fe76b796 1997 f2fs_set_page_dirty_nobuffers(page);
a7ffdbe2 1998 update_dirty_page(inode, page);
eb47b800
JK		 1999 return 1;
2000 }
2001 return 0;
2002}
2003
c01e54b7
JK		 2004static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2005{
454ae7e5
CY		 2006 struct inode *inode = mapping->host;
2007
1d373a0e
JK		 2008 if (f2fs_has_inline_data(inode))
2009 return 0;
2010
2011 /* make sure allocating whole blocks */
2012 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2013 filemap_write_and_wait(mapping);
2014
e2b4e2bc 2015 return generic_block_bmap(mapping, block, get_data_block_bmap);
429511cd
CY		 2016}
2017
5b7a487c
WG		 2018#ifdef CONFIG_MIGRATION
2019#include <linux/migrate.h>
2020
2021int f2fs_migrate_page(struct address_space *mapping,
2022 struct page *newpage, struct page *page, enum migrate_mode mode)
2023{
2024 int rc, extra_count;
2025 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2026 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2027
2028 BUG_ON(PageWriteback(page));
2029
2030 /* migrating an atomic written page is safe with the inmem_lock hold */
2031 if (atomic_written && !mutex_trylock(&fi->inmem_lock))
2032 return -EAGAIN;
2033
2034 /*
2035 * A reference is expected if PagePrivate set when move mapping,
2036 * however F2FS breaks this for maintaining dirty page counts when
2037 * truncating pages. So here adjusting the 'extra_count' make it work.
2038 */
2039 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2040 rc = migrate_page_move_mapping(mapping, newpage,
2041 page, NULL, mode, extra_count);
2042 if (rc != MIGRATEPAGE_SUCCESS) {
2043 if (atomic_written)
2044 mutex_unlock(&fi->inmem_lock);
2045 return rc;
2046 }
2047
2048 if (atomic_written) {
2049 struct inmem_pages *cur;
2050 list_for_each_entry(cur, &fi->inmem_pages, list)
2051 if (cur->page == page) {
2052 cur->page = newpage;
2053 break;
2054 }
2055 mutex_unlock(&fi->inmem_lock);
2056 put_page(page);
2057 get_page(newpage);
2058 }
2059
2060 if (PagePrivate(page))
2061 SetPagePrivate(newpage);
2062 set_page_private(newpage, page_private(page));
2063
2064 migrate_page_copy(newpage, page);
2065
2066 return MIGRATEPAGE_SUCCESS;
2067}
2068#endif
2069
eb47b800
JK		 2070const struct address_space_operations f2fs_dblock_aops = {
2071 .readpage = f2fs_read_data_page,
2072 .readpages = f2fs_read_data_pages,
2073 .writepage = f2fs_write_data_page,
2074 .writepages = f2fs_write_data_pages,
2075 .write_begin = f2fs_write_begin,
a1dd3c13 2076 .write_end = f2fs_write_end,
eb47b800 2077 .set_page_dirty = f2fs_set_data_page_dirty,
487261f3
CY		 2078 .invalidatepage = f2fs_invalidate_page,
2079 .releasepage = f2fs_release_page,
eb47b800 2080 .direct_IO = f2fs_direct_IO,
c01e54b7 2081 .bmap = f2fs_bmap,
5b7a487c
WG		 2082#ifdef CONFIG_MIGRATION
2083 .migratepage = f2fs_migrate_page,
2084#endif
eb47b800 2085};
Ubuntu Artful kernel mirror