Commit | Line | Data |
---|---|---|
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> | |
a27bb332 | 15 | #include <linux/aio.h> |
eb47b800 JK |
16 | #include <linux/writeback.h> |
17 | #include <linux/backing-dev.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/bio.h> | |
690e4a3e | 20 | #include <linux/prefetch.h> |
eb47b800 JK |
21 | |
22 | #include "f2fs.h" | |
23 | #include "node.h" | |
24 | #include "segment.h" | |
db9f7c1a | 25 | #include "trace.h" |
848753aa | 26 | #include <trace/events/f2fs.h> |
eb47b800 | 27 | |
93dfe2ac JK |
28 | static void f2fs_read_end_io(struct bio *bio, int err) |
29 | { | |
f568849e LT |
30 | struct bio_vec *bvec; |
31 | int i; | |
93dfe2ac | 32 | |
f568849e | 33 | bio_for_each_segment_all(bvec, bio, i) { |
93dfe2ac JK |
34 | struct page *page = bvec->bv_page; |
35 | ||
f568849e LT |
36 | if (!err) { |
37 | SetPageUptodate(page); | |
38 | } else { | |
93dfe2ac JK |
39 | ClearPageUptodate(page); |
40 | SetPageError(page); | |
41 | } | |
42 | unlock_page(page); | |
f568849e | 43 | } |
93dfe2ac JK |
44 | bio_put(bio); |
45 | } | |
46 | ||
47 | static void f2fs_write_end_io(struct bio *bio, int err) | |
48 | { | |
1b1f559f | 49 | struct f2fs_sb_info *sbi = bio->bi_private; |
f568849e LT |
50 | struct bio_vec *bvec; |
51 | int i; | |
93dfe2ac | 52 | |
f568849e | 53 | bio_for_each_segment_all(bvec, bio, i) { |
93dfe2ac JK |
54 | struct page *page = bvec->bv_page; |
55 | ||
f568849e | 56 | if (unlikely(err)) { |
cf779cab | 57 | set_page_dirty(page); |
93dfe2ac | 58 | set_bit(AS_EIO, &page->mapping->flags); |
744602cf | 59 | f2fs_stop_checkpoint(sbi); |
93dfe2ac JK |
60 | } |
61 | end_page_writeback(page); | |
62 | dec_page_count(sbi, F2FS_WRITEBACK); | |
f568849e | 63 | } |
93dfe2ac | 64 | |
93dfe2ac JK |
65 | if (!get_pages(sbi, F2FS_WRITEBACK) && |
66 | !list_empty(&sbi->cp_wait.task_list)) | |
67 | wake_up(&sbi->cp_wait); | |
68 | ||
69 | bio_put(bio); | |
70 | } | |
71 | ||
940a6d34 GZ |
72 | /* |
73 | * Low-level block read/write IO operations. | |
74 | */ | |
75 | static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr, | |
76 | int npages, bool is_read) | |
77 | { | |
78 | struct bio *bio; | |
79 | ||
80 | /* No failure on bio allocation */ | |
81 | bio = bio_alloc(GFP_NOIO, npages); | |
82 | ||
83 | bio->bi_bdev = sbi->sb->s_bdev; | |
55cf9cb6 | 84 | bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); |
940a6d34 | 85 | bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io; |
1b1f559f | 86 | bio->bi_private = sbi; |
940a6d34 GZ |
87 | |
88 | return bio; | |
89 | } | |
90 | ||
458e6197 | 91 | static void __submit_merged_bio(struct f2fs_bio_info *io) |
93dfe2ac | 92 | { |
458e6197 | 93 | struct f2fs_io_info *fio = &io->fio; |
93dfe2ac JK |
94 | |
95 | if (!io->bio) | |
96 | return; | |
97 | ||
6a8f8ca5 | 98 | if (is_read_io(fio->rw)) |
2ace38e0 | 99 | trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio); |
6a8f8ca5 | 100 | else |
2ace38e0 | 101 | trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio); |
940a6d34 | 102 | |
6a8f8ca5 | 103 | submit_bio(fio->rw, io->bio); |
93dfe2ac JK |
104 | io->bio = NULL; |
105 | } | |
106 | ||
107 | void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, | |
458e6197 | 108 | enum page_type type, int rw) |
93dfe2ac JK |
109 | { |
110 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
111 | struct f2fs_bio_info *io; | |
112 | ||
113 | io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype]; | |
114 | ||
df0f8dc0 | 115 | down_write(&io->io_rwsem); |
458e6197 JK |
116 | |
117 | /* change META to META_FLUSH in the checkpoint procedure */ | |
118 | if (type >= META_FLUSH) { | |
119 | io->fio.type = META_FLUSH; | |
0f7b2abd JK |
120 | if (test_opt(sbi, NOBARRIER)) |
121 | io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO; | |
122 | else | |
123 | io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO; | |
458e6197 JK |
124 | } |
125 | __submit_merged_bio(io); | |
df0f8dc0 | 126 | up_write(&io->io_rwsem); |
93dfe2ac JK |
127 | } |
128 | ||
129 | /* | |
130 | * Fill the locked page with data located in the block address. | |
131 | * Return unlocked page. | |
132 | */ | |
133 | int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page, | |
cf04e8eb | 134 | struct f2fs_io_info *fio) |
93dfe2ac | 135 | { |
93dfe2ac JK |
136 | struct bio *bio; |
137 | ||
2ace38e0 | 138 | trace_f2fs_submit_page_bio(page, fio); |
db9f7c1a | 139 | f2fs_trace_ios(page, fio, 0); |
93dfe2ac JK |
140 | |
141 | /* Allocate a new bio */ | |
cf04e8eb | 142 | bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw)); |
93dfe2ac JK |
143 | |
144 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { | |
145 | bio_put(bio); | |
146 | f2fs_put_page(page, 1); | |
147 | return -EFAULT; | |
148 | } | |
149 | ||
cf04e8eb | 150 | submit_bio(fio->rw, bio); |
93dfe2ac JK |
151 | return 0; |
152 | } | |
153 | ||
154 | void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page, | |
cf04e8eb | 155 | struct f2fs_io_info *fio) |
93dfe2ac | 156 | { |
458e6197 | 157 | enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); |
93dfe2ac | 158 | struct f2fs_bio_info *io; |
940a6d34 | 159 | bool is_read = is_read_io(fio->rw); |
93dfe2ac | 160 | |
940a6d34 | 161 | io = is_read ? &sbi->read_io : &sbi->write_io[btype]; |
93dfe2ac | 162 | |
cf04e8eb | 163 | verify_block_addr(sbi, fio->blk_addr); |
93dfe2ac | 164 | |
df0f8dc0 | 165 | down_write(&io->io_rwsem); |
93dfe2ac | 166 | |
940a6d34 | 167 | if (!is_read) |
93dfe2ac JK |
168 | inc_page_count(sbi, F2FS_WRITEBACK); |
169 | ||
cf04e8eb | 170 | if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 || |
458e6197 JK |
171 | io->fio.rw != fio->rw)) |
172 | __submit_merged_bio(io); | |
93dfe2ac JK |
173 | alloc_new: |
174 | if (io->bio == NULL) { | |
90a893c7 | 175 | int bio_blocks = MAX_BIO_BLOCKS(sbi); |
940a6d34 | 176 | |
cf04e8eb | 177 | io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read); |
458e6197 | 178 | io->fio = *fio; |
93dfe2ac JK |
179 | } |
180 | ||
181 | if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) < | |
182 | PAGE_CACHE_SIZE) { | |
458e6197 | 183 | __submit_merged_bio(io); |
93dfe2ac JK |
184 | goto alloc_new; |
185 | } | |
186 | ||
cf04e8eb | 187 | io->last_block_in_bio = fio->blk_addr; |
db9f7c1a | 188 | f2fs_trace_ios(page, fio, 0); |
93dfe2ac | 189 | |
df0f8dc0 | 190 | up_write(&io->io_rwsem); |
2ace38e0 | 191 | trace_f2fs_submit_page_mbio(page, fio); |
93dfe2ac JK |
192 | } |
193 | ||
0a8165d7 | 194 | /* |
eb47b800 JK |
195 | * Lock ordering for the change of data block address: |
196 | * ->data_page | |
197 | * ->node_page | |
198 | * update block addresses in the node page | |
199 | */ | |
e1509cf2 | 200 | static void __set_data_blkaddr(struct dnode_of_data *dn) |
eb47b800 JK |
201 | { |
202 | struct f2fs_node *rn; | |
203 | __le32 *addr_array; | |
204 | struct page *node_page = dn->node_page; | |
205 | unsigned int ofs_in_node = dn->ofs_in_node; | |
206 | ||
5514f0aa | 207 | f2fs_wait_on_page_writeback(node_page, NODE); |
eb47b800 | 208 | |
45590710 | 209 | rn = F2FS_NODE(node_page); |
eb47b800 JK |
210 | |
211 | /* Get physical address of data block */ | |
212 | addr_array = blkaddr_in_node(rn); | |
e1509cf2 | 213 | addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr); |
eb47b800 JK |
214 | set_page_dirty(node_page); |
215 | } | |
216 | ||
217 | int reserve_new_block(struct dnode_of_data *dn) | |
218 | { | |
4081363f | 219 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
eb47b800 | 220 | |
6bacf52f | 221 | if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))) |
eb47b800 | 222 | return -EPERM; |
cfb271d4 | 223 | if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1))) |
eb47b800 JK |
224 | return -ENOSPC; |
225 | ||
c01e2853 NJ |
226 | trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node); |
227 | ||
eb47b800 | 228 | dn->data_blkaddr = NEW_ADDR; |
e1509cf2 | 229 | __set_data_blkaddr(dn); |
a18ff063 | 230 | mark_inode_dirty(dn->inode); |
eb47b800 JK |
231 | sync_inode_page(dn); |
232 | return 0; | |
233 | } | |
234 | ||
b600965c HL |
235 | int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) |
236 | { | |
237 | bool need_put = dn->inode_page ? false : true; | |
238 | int err; | |
239 | ||
240 | err = get_dnode_of_data(dn, index, ALLOC_NODE); | |
241 | if (err) | |
242 | return err; | |
a8865372 | 243 | |
b600965c HL |
244 | if (dn->data_blkaddr == NULL_ADDR) |
245 | err = reserve_new_block(dn); | |
a8865372 | 246 | if (err || need_put) |
b600965c HL |
247 | f2fs_put_dnode(dn); |
248 | return err; | |
249 | } | |
250 | ||
eb47b800 JK |
251 | static int check_extent_cache(struct inode *inode, pgoff_t pgofs, |
252 | struct buffer_head *bh_result) | |
253 | { | |
254 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
eb47b800 JK |
255 | pgoff_t start_fofs, end_fofs; |
256 | block_t start_blkaddr; | |
257 | ||
c11abd1a JK |
258 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
259 | return 0; | |
260 | ||
eb47b800 JK |
261 | read_lock(&fi->ext.ext_lock); |
262 | if (fi->ext.len == 0) { | |
263 | read_unlock(&fi->ext.ext_lock); | |
264 | return 0; | |
265 | } | |
266 | ||
dcdfff65 JK |
267 | stat_inc_total_hit(inode->i_sb); |
268 | ||
eb47b800 JK |
269 | start_fofs = fi->ext.fofs; |
270 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
271 | start_blkaddr = fi->ext.blk_addr; | |
272 | ||
273 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
274 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
275 | size_t count; | |
276 | ||
da17eece | 277 | set_buffer_new(bh_result); |
eb47b800 JK |
278 | map_bh(bh_result, inode->i_sb, |
279 | start_blkaddr + pgofs - start_fofs); | |
280 | count = end_fofs - pgofs + 1; | |
281 | if (count < (UINT_MAX >> blkbits)) | |
282 | bh_result->b_size = (count << blkbits); | |
283 | else | |
284 | bh_result->b_size = UINT_MAX; | |
285 | ||
dcdfff65 | 286 | stat_inc_read_hit(inode->i_sb); |
eb47b800 JK |
287 | read_unlock(&fi->ext.ext_lock); |
288 | return 1; | |
289 | } | |
290 | read_unlock(&fi->ext.ext_lock); | |
291 | return 0; | |
292 | } | |
293 | ||
e1509cf2 | 294 | void update_extent_cache(struct dnode_of_data *dn) |
eb47b800 JK |
295 | { |
296 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
297 | pgoff_t fofs, start_fofs, end_fofs; | |
298 | block_t start_blkaddr, end_blkaddr; | |
c11abd1a | 299 | int need_update = true; |
eb47b800 | 300 | |
e1509cf2 | 301 | f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR); |
eb47b800 JK |
302 | |
303 | /* Update the page address in the parent node */ | |
e1509cf2 | 304 | __set_data_blkaddr(dn); |
eb47b800 | 305 | |
c11abd1a JK |
306 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
307 | return; | |
308 | ||
3547ea96 JK |
309 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + |
310 | dn->ofs_in_node; | |
311 | ||
eb47b800 JK |
312 | write_lock(&fi->ext.ext_lock); |
313 | ||
314 | start_fofs = fi->ext.fofs; | |
315 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
316 | start_blkaddr = fi->ext.blk_addr; | |
317 | end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1; | |
318 | ||
319 | /* Drop and initialize the matched extent */ | |
320 | if (fi->ext.len == 1 && fofs == start_fofs) | |
321 | fi->ext.len = 0; | |
322 | ||
323 | /* Initial extent */ | |
324 | if (fi->ext.len == 0) { | |
e1509cf2 | 325 | if (dn->data_blkaddr != NULL_ADDR) { |
eb47b800 | 326 | fi->ext.fofs = fofs; |
e1509cf2 | 327 | fi->ext.blk_addr = dn->data_blkaddr; |
eb47b800 JK |
328 | fi->ext.len = 1; |
329 | } | |
330 | goto end_update; | |
331 | } | |
332 | ||
6224da87 | 333 | /* Front merge */ |
e1509cf2 | 334 | if (fofs == start_fofs - 1 && dn->data_blkaddr == start_blkaddr - 1) { |
eb47b800 JK |
335 | fi->ext.fofs--; |
336 | fi->ext.blk_addr--; | |
337 | fi->ext.len++; | |
338 | goto end_update; | |
339 | } | |
340 | ||
341 | /* Back merge */ | |
e1509cf2 | 342 | if (fofs == end_fofs + 1 && dn->data_blkaddr == end_blkaddr + 1) { |
eb47b800 JK |
343 | fi->ext.len++; |
344 | goto end_update; | |
345 | } | |
346 | ||
347 | /* Split the existing extent */ | |
348 | if (fi->ext.len > 1 && | |
349 | fofs >= start_fofs && fofs <= end_fofs) { | |
350 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
351 | fi->ext.len = fofs - start_fofs; | |
352 | } else { | |
353 | fi->ext.fofs = fofs + 1; | |
354 | fi->ext.blk_addr = start_blkaddr + | |
355 | fofs - start_fofs + 1; | |
356 | fi->ext.len -= fofs - start_fofs + 1; | |
357 | } | |
c11abd1a JK |
358 | } else { |
359 | need_update = false; | |
eb47b800 | 360 | } |
eb47b800 | 361 | |
c11abd1a JK |
362 | /* Finally, if the extent is very fragmented, let's drop the cache. */ |
363 | if (fi->ext.len < F2FS_MIN_EXTENT_LEN) { | |
364 | fi->ext.len = 0; | |
365 | set_inode_flag(fi, FI_NO_EXTENT); | |
366 | need_update = true; | |
367 | } | |
eb47b800 JK |
368 | end_update: |
369 | write_unlock(&fi->ext.ext_lock); | |
c11abd1a JK |
370 | if (need_update) |
371 | sync_inode_page(dn); | |
372 | return; | |
eb47b800 JK |
373 | } |
374 | ||
c718379b | 375 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 | 376 | { |
eb47b800 JK |
377 | struct address_space *mapping = inode->i_mapping; |
378 | struct dnode_of_data dn; | |
379 | struct page *page; | |
380 | int err; | |
cf04e8eb JK |
381 | struct f2fs_io_info fio = { |
382 | .type = DATA, | |
383 | .rw = sync ? READ_SYNC : READA, | |
384 | }; | |
eb47b800 JK |
385 | |
386 | page = find_get_page(mapping, index); | |
387 | if (page && PageUptodate(page)) | |
388 | return page; | |
389 | f2fs_put_page(page, 0); | |
390 | ||
391 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 392 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
393 | if (err) |
394 | return ERR_PTR(err); | |
395 | f2fs_put_dnode(&dn); | |
396 | ||
397 | if (dn.data_blkaddr == NULL_ADDR) | |
398 | return ERR_PTR(-ENOENT); | |
399 | ||
400 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
6bacf52f | 401 | if (unlikely(dn.data_blkaddr == NEW_ADDR)) |
eb47b800 JK |
402 | return ERR_PTR(-EINVAL); |
403 | ||
9ac1349a | 404 | page = grab_cache_page(mapping, index); |
eb47b800 JK |
405 | if (!page) |
406 | return ERR_PTR(-ENOMEM); | |
407 | ||
393ff91f JK |
408 | if (PageUptodate(page)) { |
409 | unlock_page(page); | |
410 | return page; | |
411 | } | |
412 | ||
cf04e8eb JK |
413 | fio.blk_addr = dn.data_blkaddr; |
414 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
1069bbf7 CY |
415 | if (err) |
416 | return ERR_PTR(err); | |
417 | ||
c718379b JK |
418 | if (sync) { |
419 | wait_on_page_locked(page); | |
6bacf52f | 420 | if (unlikely(!PageUptodate(page))) { |
c718379b JK |
421 | f2fs_put_page(page, 0); |
422 | return ERR_PTR(-EIO); | |
423 | } | |
eb47b800 | 424 | } |
eb47b800 JK |
425 | return page; |
426 | } | |
427 | ||
0a8165d7 | 428 | /* |
eb47b800 JK |
429 | * If it tries to access a hole, return an error. |
430 | * Because, the callers, functions in dir.c and GC, should be able to know | |
431 | * whether this page exists or not. | |
432 | */ | |
433 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
434 | { | |
eb47b800 JK |
435 | struct address_space *mapping = inode->i_mapping; |
436 | struct dnode_of_data dn; | |
437 | struct page *page; | |
438 | int err; | |
cf04e8eb JK |
439 | struct f2fs_io_info fio = { |
440 | .type = DATA, | |
441 | .rw = READ_SYNC, | |
442 | }; | |
650495de | 443 | repeat: |
9ac1349a | 444 | page = grab_cache_page(mapping, index); |
650495de JK |
445 | if (!page) |
446 | return ERR_PTR(-ENOMEM); | |
447 | ||
eb47b800 | 448 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 449 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
450 | if (err) { |
451 | f2fs_put_page(page, 1); | |
eb47b800 | 452 | return ERR_PTR(err); |
650495de | 453 | } |
eb47b800 JK |
454 | f2fs_put_dnode(&dn); |
455 | ||
6bacf52f | 456 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
650495de | 457 | f2fs_put_page(page, 1); |
eb47b800 | 458 | return ERR_PTR(-ENOENT); |
650495de | 459 | } |
eb47b800 JK |
460 | |
461 | if (PageUptodate(page)) | |
462 | return page; | |
463 | ||
d59ff4df JK |
464 | /* |
465 | * A new dentry page is allocated but not able to be written, since its | |
466 | * new inode page couldn't be allocated due to -ENOSPC. | |
467 | * In such the case, its blkaddr can be remained as NEW_ADDR. | |
468 | * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. | |
469 | */ | |
470 | if (dn.data_blkaddr == NEW_ADDR) { | |
471 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
472 | SetPageUptodate(page); | |
473 | return page; | |
474 | } | |
eb47b800 | 475 | |
cf04e8eb JK |
476 | fio.blk_addr = dn.data_blkaddr; |
477 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
393ff91f | 478 | if (err) |
eb47b800 | 479 | return ERR_PTR(err); |
393ff91f JK |
480 | |
481 | lock_page(page); | |
6bacf52f | 482 | if (unlikely(!PageUptodate(page))) { |
393ff91f JK |
483 | f2fs_put_page(page, 1); |
484 | return ERR_PTR(-EIO); | |
eb47b800 | 485 | } |
6bacf52f | 486 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
487 | f2fs_put_page(page, 1); |
488 | goto repeat; | |
eb47b800 JK |
489 | } |
490 | return page; | |
491 | } | |
492 | ||
0a8165d7 | 493 | /* |
eb47b800 JK |
494 | * Caller ensures that this data page is never allocated. |
495 | * A new zero-filled data page is allocated in the page cache. | |
39936837 | 496 | * |
4f4124d0 CY |
497 | * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and |
498 | * f2fs_unlock_op(). | |
a8865372 | 499 | * Note that, ipage is set only by make_empty_dir. |
eb47b800 | 500 | */ |
64aa7ed9 | 501 | struct page *get_new_data_page(struct inode *inode, |
a8865372 | 502 | struct page *ipage, pgoff_t index, bool new_i_size) |
eb47b800 | 503 | { |
eb47b800 JK |
504 | struct address_space *mapping = inode->i_mapping; |
505 | struct page *page; | |
506 | struct dnode_of_data dn; | |
507 | int err; | |
508 | ||
a8865372 | 509 | set_new_dnode(&dn, inode, ipage, NULL, 0); |
b600965c | 510 | err = f2fs_reserve_block(&dn, index); |
eb47b800 JK |
511 | if (err) |
512 | return ERR_PTR(err); | |
afcb7ca0 | 513 | repeat: |
eb47b800 | 514 | page = grab_cache_page(mapping, index); |
a8865372 JK |
515 | if (!page) { |
516 | err = -ENOMEM; | |
517 | goto put_err; | |
518 | } | |
eb47b800 JK |
519 | |
520 | if (PageUptodate(page)) | |
521 | return page; | |
522 | ||
523 | if (dn.data_blkaddr == NEW_ADDR) { | |
524 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 525 | SetPageUptodate(page); |
eb47b800 | 526 | } else { |
cf04e8eb JK |
527 | struct f2fs_io_info fio = { |
528 | .type = DATA, | |
529 | .rw = READ_SYNC, | |
530 | .blk_addr = dn.data_blkaddr, | |
531 | }; | |
532 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
393ff91f | 533 | if (err) |
a8865372 JK |
534 | goto put_err; |
535 | ||
393ff91f | 536 | lock_page(page); |
6bacf52f | 537 | if (unlikely(!PageUptodate(page))) { |
393ff91f | 538 | f2fs_put_page(page, 1); |
a8865372 JK |
539 | err = -EIO; |
540 | goto put_err; | |
eb47b800 | 541 | } |
6bacf52f | 542 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
543 | f2fs_put_page(page, 1); |
544 | goto repeat; | |
eb47b800 JK |
545 | } |
546 | } | |
eb47b800 JK |
547 | |
548 | if (new_i_size && | |
549 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
550 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
551 | /* Only the directory inode sets new_i_size */ |
552 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
553 | } |
554 | return page; | |
a8865372 JK |
555 | |
556 | put_err: | |
557 | f2fs_put_dnode(&dn); | |
558 | return ERR_PTR(err); | |
eb47b800 JK |
559 | } |
560 | ||
bfad7c2d JK |
561 | static int __allocate_data_block(struct dnode_of_data *dn) |
562 | { | |
4081363f | 563 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
976e4c50 | 564 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); |
bfad7c2d | 565 | struct f2fs_summary sum; |
bfad7c2d | 566 | struct node_info ni; |
38aa0889 | 567 | int seg = CURSEG_WARM_DATA; |
976e4c50 | 568 | pgoff_t fofs; |
bfad7c2d JK |
569 | |
570 | if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))) | |
571 | return -EPERM; | |
572 | if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1))) | |
573 | return -ENOSPC; | |
574 | ||
bfad7c2d JK |
575 | get_node_info(sbi, dn->nid, &ni); |
576 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
577 | ||
38aa0889 JK |
578 | if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page) |
579 | seg = CURSEG_DIRECT_IO; | |
580 | ||
581 | allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg); | |
bfad7c2d JK |
582 | |
583 | /* direct IO doesn't use extent cache to maximize the performance */ | |
41ef94b3 | 584 | __set_data_blkaddr(dn); |
bfad7c2d | 585 | |
976e4c50 JK |
586 | /* update i_size */ |
587 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + | |
588 | dn->ofs_in_node; | |
589 | if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT)) | |
590 | i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT)); | |
591 | ||
bfad7c2d JK |
592 | return 0; |
593 | } | |
594 | ||
59b802e5 JK |
595 | static void __allocate_data_blocks(struct inode *inode, loff_t offset, |
596 | size_t count) | |
597 | { | |
598 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
599 | struct dnode_of_data dn; | |
600 | u64 start = F2FS_BYTES_TO_BLK(offset); | |
601 | u64 len = F2FS_BYTES_TO_BLK(count); | |
602 | bool allocated; | |
603 | u64 end_offset; | |
604 | ||
605 | while (len) { | |
606 | f2fs_balance_fs(sbi); | |
607 | f2fs_lock_op(sbi); | |
608 | ||
609 | /* When reading holes, we need its node page */ | |
610 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
611 | if (get_dnode_of_data(&dn, start, ALLOC_NODE)) | |
612 | goto out; | |
613 | ||
614 | allocated = false; | |
615 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); | |
616 | ||
617 | while (dn.ofs_in_node < end_offset && len) { | |
618 | if (dn.data_blkaddr == NULL_ADDR) { | |
619 | if (__allocate_data_block(&dn)) | |
620 | goto sync_out; | |
621 | allocated = true; | |
622 | } | |
623 | len--; | |
624 | start++; | |
625 | dn.ofs_in_node++; | |
626 | } | |
627 | ||
628 | if (allocated) | |
629 | sync_inode_page(&dn); | |
630 | ||
631 | f2fs_put_dnode(&dn); | |
632 | f2fs_unlock_op(sbi); | |
633 | } | |
634 | return; | |
635 | ||
636 | sync_out: | |
637 | if (allocated) | |
638 | sync_inode_page(&dn); | |
639 | f2fs_put_dnode(&dn); | |
640 | out: | |
641 | f2fs_unlock_op(sbi); | |
642 | return; | |
643 | } | |
644 | ||
0a8165d7 | 645 | /* |
4f4124d0 CY |
646 | * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh. |
647 | * If original data blocks are allocated, then give them to blockdev. | |
648 | * Otherwise, | |
649 | * a. preallocate requested block addresses | |
650 | * b. do not use extent cache for better performance | |
651 | * c. give the block addresses to blockdev | |
eb47b800 | 652 | */ |
ccfb3000 JK |
653 | static int __get_data_block(struct inode *inode, sector_t iblock, |
654 | struct buffer_head *bh_result, int create, bool fiemap) | |
eb47b800 JK |
655 | { |
656 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
657 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
658 | struct dnode_of_data dn; | |
bfad7c2d JK |
659 | int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA; |
660 | pgoff_t pgofs, end_offset; | |
661 | int err = 0, ofs = 1; | |
662 | bool allocated = false; | |
eb47b800 JK |
663 | |
664 | /* Get the page offset from the block offset(iblock) */ | |
665 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
666 | ||
bfad7c2d JK |
667 | if (check_extent_cache(inode, pgofs, bh_result)) |
668 | goto out; | |
669 | ||
59b802e5 | 670 | if (create) |
4081363f | 671 | f2fs_lock_op(F2FS_I_SB(inode)); |
eb47b800 JK |
672 | |
673 | /* When reading holes, we need its node page */ | |
674 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
bfad7c2d | 675 | err = get_dnode_of_data(&dn, pgofs, mode); |
1ec79083 | 676 | if (err) { |
bfad7c2d JK |
677 | if (err == -ENOENT) |
678 | err = 0; | |
679 | goto unlock_out; | |
848753aa | 680 | } |
ccfb3000 | 681 | if (dn.data_blkaddr == NEW_ADDR && !fiemap) |
1ec79083 | 682 | goto put_out; |
eb47b800 | 683 | |
bfad7c2d | 684 | if (dn.data_blkaddr != NULL_ADDR) { |
da17eece | 685 | set_buffer_new(bh_result); |
bfad7c2d JK |
686 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); |
687 | } else if (create) { | |
688 | err = __allocate_data_block(&dn); | |
689 | if (err) | |
690 | goto put_out; | |
691 | allocated = true; | |
da17eece | 692 | set_buffer_new(bh_result); |
bfad7c2d JK |
693 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); |
694 | } else { | |
695 | goto put_out; | |
696 | } | |
697 | ||
6403eb1f | 698 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); |
bfad7c2d JK |
699 | bh_result->b_size = (((size_t)1) << blkbits); |
700 | dn.ofs_in_node++; | |
701 | pgofs++; | |
702 | ||
703 | get_next: | |
704 | if (dn.ofs_in_node >= end_offset) { | |
705 | if (allocated) | |
706 | sync_inode_page(&dn); | |
707 | allocated = false; | |
708 | f2fs_put_dnode(&dn); | |
709 | ||
710 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
711 | err = get_dnode_of_data(&dn, pgofs, mode); | |
1ec79083 | 712 | if (err) { |
bfad7c2d JK |
713 | if (err == -ENOENT) |
714 | err = 0; | |
715 | goto unlock_out; | |
716 | } | |
ccfb3000 | 717 | if (dn.data_blkaddr == NEW_ADDR && !fiemap) |
1ec79083 JK |
718 | goto put_out; |
719 | ||
6403eb1f | 720 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); |
bfad7c2d | 721 | } |
eb47b800 | 722 | |
bfad7c2d JK |
723 | if (maxblocks > (bh_result->b_size >> blkbits)) { |
724 | block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node); | |
725 | if (blkaddr == NULL_ADDR && create) { | |
726 | err = __allocate_data_block(&dn); | |
727 | if (err) | |
728 | goto sync_out; | |
729 | allocated = true; | |
730 | blkaddr = dn.data_blkaddr; | |
731 | } | |
e1c42045 | 732 | /* Give more consecutive addresses for the readahead */ |
bfad7c2d JK |
733 | if (blkaddr == (bh_result->b_blocknr + ofs)) { |
734 | ofs++; | |
735 | dn.ofs_in_node++; | |
736 | pgofs++; | |
737 | bh_result->b_size += (((size_t)1) << blkbits); | |
738 | goto get_next; | |
739 | } | |
eb47b800 | 740 | } |
bfad7c2d JK |
741 | sync_out: |
742 | if (allocated) | |
743 | sync_inode_page(&dn); | |
744 | put_out: | |
eb47b800 | 745 | f2fs_put_dnode(&dn); |
bfad7c2d JK |
746 | unlock_out: |
747 | if (create) | |
4081363f | 748 | f2fs_unlock_op(F2FS_I_SB(inode)); |
bfad7c2d JK |
749 | out: |
750 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
751 | return err; | |
eb47b800 JK |
752 | } |
753 | ||
ccfb3000 JK |
754 | static int get_data_block(struct inode *inode, sector_t iblock, |
755 | struct buffer_head *bh_result, int create) | |
756 | { | |
757 | return __get_data_block(inode, iblock, bh_result, create, false); | |
758 | } | |
759 | ||
760 | static int get_data_block_fiemap(struct inode *inode, sector_t iblock, | |
761 | struct buffer_head *bh_result, int create) | |
762 | { | |
763 | return __get_data_block(inode, iblock, bh_result, create, true); | |
764 | } | |
765 | ||
9ab70134 JK |
766 | int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
767 | u64 start, u64 len) | |
768 | { | |
ccfb3000 JK |
769 | return generic_block_fiemap(inode, fieinfo, |
770 | start, len, get_data_block_fiemap); | |
9ab70134 JK |
771 | } |
772 | ||
eb47b800 JK |
773 | static int f2fs_read_data_page(struct file *file, struct page *page) |
774 | { | |
9ffe0fb5 | 775 | struct inode *inode = page->mapping->host; |
b3d208f9 | 776 | int ret = -EAGAIN; |
9ffe0fb5 | 777 | |
c20e89cd CY |
778 | trace_f2fs_readpage(page, DATA); |
779 | ||
e1c42045 | 780 | /* If the file has inline data, try to read it directly */ |
9ffe0fb5 HL |
781 | if (f2fs_has_inline_data(inode)) |
782 | ret = f2fs_read_inline_data(inode, page); | |
b3d208f9 | 783 | if (ret == -EAGAIN) |
9ffe0fb5 HL |
784 | ret = mpage_readpage(page, get_data_block); |
785 | ||
786 | return ret; | |
eb47b800 JK |
787 | } |
788 | ||
789 | static int f2fs_read_data_pages(struct file *file, | |
790 | struct address_space *mapping, | |
791 | struct list_head *pages, unsigned nr_pages) | |
792 | { | |
9ffe0fb5 HL |
793 | struct inode *inode = file->f_mapping->host; |
794 | ||
795 | /* If the file has inline data, skip readpages */ | |
796 | if (f2fs_has_inline_data(inode)) | |
797 | return 0; | |
798 | ||
bfad7c2d | 799 | return mpage_readpages(mapping, pages, nr_pages, get_data_block); |
eb47b800 JK |
800 | } |
801 | ||
458e6197 | 802 | int do_write_data_page(struct page *page, struct f2fs_io_info *fio) |
eb47b800 JK |
803 | { |
804 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
805 | struct dnode_of_data dn; |
806 | int err = 0; | |
807 | ||
808 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 809 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
810 | if (err) |
811 | return err; | |
812 | ||
cf04e8eb | 813 | fio->blk_addr = dn.data_blkaddr; |
eb47b800 JK |
814 | |
815 | /* This page is already truncated */ | |
cf04e8eb | 816 | if (fio->blk_addr == NULL_ADDR) |
eb47b800 JK |
817 | goto out_writepage; |
818 | ||
819 | set_page_writeback(page); | |
820 | ||
821 | /* | |
822 | * If current allocation needs SSR, | |
823 | * it had better in-place writes for updated data. | |
824 | */ | |
cf04e8eb | 825 | if (unlikely(fio->blk_addr != NEW_ADDR && |
b25958b6 HL |
826 | !is_cold_data(page) && |
827 | need_inplace_update(inode))) { | |
cf04e8eb | 828 | rewrite_data_page(page, fio); |
fff04f90 | 829 | set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE); |
eb47b800 | 830 | } else { |
cf04e8eb | 831 | write_data_page(page, &dn, fio); |
e1509cf2 | 832 | update_extent_cache(&dn); |
fff04f90 | 833 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); |
eb47b800 JK |
834 | } |
835 | out_writepage: | |
836 | f2fs_put_dnode(&dn); | |
837 | return err; | |
838 | } | |
839 | ||
840 | static int f2fs_write_data_page(struct page *page, | |
841 | struct writeback_control *wbc) | |
842 | { | |
843 | struct inode *inode = page->mapping->host; | |
4081363f | 844 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
eb47b800 JK |
845 | loff_t i_size = i_size_read(inode); |
846 | const pgoff_t end_index = ((unsigned long long) i_size) | |
847 | >> PAGE_CACHE_SHIFT; | |
9ffe0fb5 | 848 | unsigned offset = 0; |
39936837 | 849 | bool need_balance_fs = false; |
eb47b800 | 850 | int err = 0; |
458e6197 JK |
851 | struct f2fs_io_info fio = { |
852 | .type = DATA, | |
6c311ec6 | 853 | .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE, |
458e6197 | 854 | }; |
eb47b800 | 855 | |
ecda0de3 CY |
856 | trace_f2fs_writepage(page, DATA); |
857 | ||
eb47b800 | 858 | if (page->index < end_index) |
39936837 | 859 | goto write; |
eb47b800 JK |
860 | |
861 | /* | |
862 | * If the offset is out-of-range of file size, | |
863 | * this page does not have to be written to disk. | |
864 | */ | |
865 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
76f60268 | 866 | if ((page->index >= end_index + 1) || !offset) |
39936837 | 867 | goto out; |
eb47b800 JK |
868 | |
869 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 | 870 | write: |
caf0047e | 871 | if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) |
eb47b800 | 872 | goto redirty_out; |
1e84371f JK |
873 | if (f2fs_is_drop_cache(inode)) |
874 | goto out; | |
875 | if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim && | |
876 | available_free_memory(sbi, BASE_CHECK)) | |
877 | goto redirty_out; | |
eb47b800 | 878 | |
39936837 | 879 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 | 880 | if (S_ISDIR(inode->i_mode)) { |
cf779cab JK |
881 | if (unlikely(f2fs_cp_error(sbi))) |
882 | goto redirty_out; | |
458e6197 | 883 | err = do_write_data_page(page, &fio); |
8618b881 JK |
884 | goto done; |
885 | } | |
9ffe0fb5 | 886 | |
cf779cab JK |
887 | /* we should bypass data pages to proceed the kworkder jobs */ |
888 | if (unlikely(f2fs_cp_error(sbi))) { | |
889 | SetPageError(page); | |
a7ffdbe2 | 890 | goto out; |
cf779cab JK |
891 | } |
892 | ||
8618b881 | 893 | if (!wbc->for_reclaim) |
39936837 | 894 | need_balance_fs = true; |
8618b881 | 895 | else if (has_not_enough_free_secs(sbi, 0)) |
39936837 | 896 | goto redirty_out; |
eb47b800 | 897 | |
b3d208f9 | 898 | err = -EAGAIN; |
8618b881 | 899 | f2fs_lock_op(sbi); |
b3d208f9 JK |
900 | if (f2fs_has_inline_data(inode)) |
901 | err = f2fs_write_inline_data(inode, page); | |
902 | if (err == -EAGAIN) | |
8618b881 JK |
903 | err = do_write_data_page(page, &fio); |
904 | f2fs_unlock_op(sbi); | |
905 | done: | |
906 | if (err && err != -ENOENT) | |
907 | goto redirty_out; | |
eb47b800 | 908 | |
eb47b800 | 909 | clear_cold_data(page); |
39936837 | 910 | out: |
a7ffdbe2 | 911 | inode_dec_dirty_pages(inode); |
eb47b800 | 912 | unlock_page(page); |
39936837 | 913 | if (need_balance_fs) |
eb47b800 | 914 | f2fs_balance_fs(sbi); |
2aea39ec JK |
915 | if (wbc->for_reclaim) |
916 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
eb47b800 JK |
917 | return 0; |
918 | ||
eb47b800 | 919 | redirty_out: |
76f60268 | 920 | redirty_page_for_writepage(wbc, page); |
8618b881 | 921 | return AOP_WRITEPAGE_ACTIVATE; |
eb47b800 JK |
922 | } |
923 | ||
fa9150a8 NJ |
924 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
925 | void *data) | |
926 | { | |
927 | struct address_space *mapping = data; | |
928 | int ret = mapping->a_ops->writepage(page, wbc); | |
929 | mapping_set_error(mapping, ret); | |
930 | return ret; | |
931 | } | |
932 | ||
25ca923b | 933 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
934 | struct writeback_control *wbc) |
935 | { | |
936 | struct inode *inode = mapping->host; | |
4081363f | 937 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
531ad7d5 | 938 | bool locked = false; |
eb47b800 | 939 | int ret; |
50c8cdb3 | 940 | long diff; |
eb47b800 | 941 | |
e5748434 CY |
942 | trace_f2fs_writepages(mapping->host, wbc, DATA); |
943 | ||
cfb185a1 | 944 | /* deal with chardevs and other special file */ |
945 | if (!mapping->a_ops->writepage) | |
946 | return 0; | |
947 | ||
87d6f890 | 948 | if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && |
a7ffdbe2 | 949 | get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && |
6fb03f3a | 950 | available_free_memory(sbi, DIRTY_DENTS)) |
d3baf95d | 951 | goto skip_write; |
87d6f890 | 952 | |
50c8cdb3 | 953 | diff = nr_pages_to_write(sbi, DATA, wbc); |
eb47b800 | 954 | |
531ad7d5 | 955 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 956 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
957 | locked = true; |
958 | } | |
fa9150a8 | 959 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 960 | if (locked) |
eb47b800 | 961 | mutex_unlock(&sbi->writepages); |
458e6197 JK |
962 | |
963 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
eb47b800 JK |
964 | |
965 | remove_dirty_dir_inode(inode); | |
966 | ||
50c8cdb3 | 967 | wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); |
eb47b800 | 968 | return ret; |
d3baf95d JK |
969 | |
970 | skip_write: | |
a7ffdbe2 | 971 | wbc->pages_skipped += get_dirty_pages(inode); |
d3baf95d | 972 | return 0; |
eb47b800 JK |
973 | } |
974 | ||
3aab8f82 CY |
975 | static void f2fs_write_failed(struct address_space *mapping, loff_t to) |
976 | { | |
977 | struct inode *inode = mapping->host; | |
978 | ||
979 | if (to > inode->i_size) { | |
980 | truncate_pagecache(inode, inode->i_size); | |
764aa3e9 | 981 | truncate_blocks(inode, inode->i_size, true); |
3aab8f82 CY |
982 | } |
983 | } | |
984 | ||
eb47b800 JK |
985 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, |
986 | loff_t pos, unsigned len, unsigned flags, | |
987 | struct page **pagep, void **fsdata) | |
988 | { | |
989 | struct inode *inode = mapping->host; | |
4081363f | 990 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
9ba69cf9 | 991 | struct page *page, *ipage; |
eb47b800 JK |
992 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; |
993 | struct dnode_of_data dn; | |
994 | int err = 0; | |
995 | ||
62aed044 CY |
996 | trace_f2fs_write_begin(inode, pos, len, flags); |
997 | ||
eb47b800 | 998 | f2fs_balance_fs(sbi); |
5f727395 JK |
999 | |
1000 | /* | |
1001 | * We should check this at this moment to avoid deadlock on inode page | |
1002 | * and #0 page. The locking rule for inline_data conversion should be: | |
1003 | * lock_page(page #0) -> lock_page(inode_page) | |
1004 | */ | |
1005 | if (index != 0) { | |
1006 | err = f2fs_convert_inline_inode(inode); | |
1007 | if (err) | |
1008 | goto fail; | |
1009 | } | |
afcb7ca0 | 1010 | repeat: |
eb47b800 | 1011 | page = grab_cache_page_write_begin(mapping, index, flags); |
3aab8f82 CY |
1012 | if (!page) { |
1013 | err = -ENOMEM; | |
1014 | goto fail; | |
1015 | } | |
d5f66990 | 1016 | |
eb47b800 JK |
1017 | *pagep = page; |
1018 | ||
e479556b | 1019 | f2fs_lock_op(sbi); |
9ba69cf9 JK |
1020 | |
1021 | /* check inline_data */ | |
1022 | ipage = get_node_page(sbi, inode->i_ino); | |
cd34e296 CY |
1023 | if (IS_ERR(ipage)) { |
1024 | err = PTR_ERR(ipage); | |
9ba69cf9 | 1025 | goto unlock_fail; |
cd34e296 | 1026 | } |
9ba69cf9 | 1027 | |
b3d208f9 JK |
1028 | set_new_dnode(&dn, inode, ipage, ipage, 0); |
1029 | ||
9ba69cf9 | 1030 | if (f2fs_has_inline_data(inode)) { |
b3d208f9 JK |
1031 | if (pos + len <= MAX_INLINE_DATA) { |
1032 | read_inline_data(page, ipage); | |
1033 | set_inode_flag(F2FS_I(inode), FI_DATA_EXIST); | |
1034 | sync_inode_page(&dn); | |
1035 | goto put_next; | |
b3d208f9 | 1036 | } |
5f727395 JK |
1037 | err = f2fs_convert_inline_page(&dn, page); |
1038 | if (err) | |
1039 | goto put_fail; | |
b600965c | 1040 | } |
9ba69cf9 JK |
1041 | err = f2fs_reserve_block(&dn, index); |
1042 | if (err) | |
8cdcb713 | 1043 | goto put_fail; |
b3d208f9 | 1044 | put_next: |
9ba69cf9 JK |
1045 | f2fs_put_dnode(&dn); |
1046 | f2fs_unlock_op(sbi); | |
1047 | ||
eb47b800 JK |
1048 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) |
1049 | return 0; | |
1050 | ||
b3d208f9 JK |
1051 | f2fs_wait_on_page_writeback(page, DATA); |
1052 | ||
eb47b800 JK |
1053 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { |
1054 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
1055 | unsigned end = start + len; | |
1056 | ||
1057 | /* Reading beyond i_size is simple: memset to zero */ | |
1058 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 1059 | goto out; |
eb47b800 JK |
1060 | } |
1061 | ||
b3d208f9 | 1062 | if (dn.data_blkaddr == NEW_ADDR) { |
eb47b800 JK |
1063 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); |
1064 | } else { | |
cf04e8eb JK |
1065 | struct f2fs_io_info fio = { |
1066 | .type = DATA, | |
1067 | .rw = READ_SYNC, | |
1068 | .blk_addr = dn.data_blkaddr, | |
1069 | }; | |
1070 | err = f2fs_submit_page_bio(sbi, page, &fio); | |
9234f319 JK |
1071 | if (err) |
1072 | goto fail; | |
d54c795b | 1073 | |
393ff91f | 1074 | lock_page(page); |
6bacf52f | 1075 | if (unlikely(!PageUptodate(page))) { |
393ff91f | 1076 | f2fs_put_page(page, 1); |
3aab8f82 CY |
1077 | err = -EIO; |
1078 | goto fail; | |
eb47b800 | 1079 | } |
6bacf52f | 1080 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
1081 | f2fs_put_page(page, 1); |
1082 | goto repeat; | |
eb47b800 JK |
1083 | } |
1084 | } | |
393ff91f | 1085 | out: |
eb47b800 JK |
1086 | SetPageUptodate(page); |
1087 | clear_cold_data(page); | |
1088 | return 0; | |
9ba69cf9 | 1089 | |
8cdcb713 JK |
1090 | put_fail: |
1091 | f2fs_put_dnode(&dn); | |
9ba69cf9 JK |
1092 | unlock_fail: |
1093 | f2fs_unlock_op(sbi); | |
b3d208f9 | 1094 | f2fs_put_page(page, 1); |
3aab8f82 CY |
1095 | fail: |
1096 | f2fs_write_failed(mapping, pos + len); | |
1097 | return err; | |
eb47b800 JK |
1098 | } |
1099 | ||
a1dd3c13 JK |
1100 | static int f2fs_write_end(struct file *file, |
1101 | struct address_space *mapping, | |
1102 | loff_t pos, unsigned len, unsigned copied, | |
1103 | struct page *page, void *fsdata) | |
1104 | { | |
1105 | struct inode *inode = page->mapping->host; | |
1106 | ||
dfb2bf38 CY |
1107 | trace_f2fs_write_end(inode, pos, len, copied); |
1108 | ||
34ba94ba | 1109 | set_page_dirty(page); |
a1dd3c13 JK |
1110 | |
1111 | if (pos + copied > i_size_read(inode)) { | |
1112 | i_size_write(inode, pos + copied); | |
1113 | mark_inode_dirty(inode); | |
1114 | update_inode_page(inode); | |
1115 | } | |
1116 | ||
75c3c8bc | 1117 | f2fs_put_page(page, 1); |
a1dd3c13 JK |
1118 | return copied; |
1119 | } | |
1120 | ||
944fcfc1 | 1121 | static int check_direct_IO(struct inode *inode, int rw, |
5b46f25d | 1122 | struct iov_iter *iter, loff_t offset) |
944fcfc1 JK |
1123 | { |
1124 | unsigned blocksize_mask = inode->i_sb->s_blocksize - 1; | |
944fcfc1 JK |
1125 | |
1126 | if (rw == READ) | |
1127 | return 0; | |
1128 | ||
1129 | if (offset & blocksize_mask) | |
1130 | return -EINVAL; | |
1131 | ||
5b46f25d AV |
1132 | if (iov_iter_alignment(iter) & blocksize_mask) |
1133 | return -EINVAL; | |
1134 | ||
944fcfc1 JK |
1135 | return 0; |
1136 | } | |
1137 | ||
eb47b800 | 1138 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
d8d3d94b | 1139 | struct iov_iter *iter, loff_t offset) |
eb47b800 JK |
1140 | { |
1141 | struct file *file = iocb->ki_filp; | |
3aab8f82 CY |
1142 | struct address_space *mapping = file->f_mapping; |
1143 | struct inode *inode = mapping->host; | |
1144 | size_t count = iov_iter_count(iter); | |
1145 | int err; | |
944fcfc1 | 1146 | |
b3d208f9 JK |
1147 | /* we don't need to use inline_data strictly */ |
1148 | if (f2fs_has_inline_data(inode)) { | |
1149 | err = f2fs_convert_inline_inode(inode); | |
1150 | if (err) | |
1151 | return err; | |
1152 | } | |
9ffe0fb5 | 1153 | |
5b46f25d | 1154 | if (check_direct_IO(inode, rw, iter, offset)) |
944fcfc1 JK |
1155 | return 0; |
1156 | ||
70407fad CY |
1157 | trace_f2fs_direct_IO_enter(inode, offset, count, rw); |
1158 | ||
59b802e5 JK |
1159 | if (rw & WRITE) |
1160 | __allocate_data_blocks(inode, offset, count); | |
1161 | ||
3aab8f82 CY |
1162 | err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block); |
1163 | if (err < 0 && (rw & WRITE)) | |
1164 | f2fs_write_failed(mapping, offset + count); | |
70407fad CY |
1165 | |
1166 | trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); | |
1167 | ||
3aab8f82 | 1168 | return err; |
eb47b800 JK |
1169 | } |
1170 | ||
487261f3 CY |
1171 | void f2fs_invalidate_page(struct page *page, unsigned int offset, |
1172 | unsigned int length) | |
eb47b800 JK |
1173 | { |
1174 | struct inode *inode = page->mapping->host; | |
487261f3 | 1175 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
a7ffdbe2 | 1176 | |
487261f3 CY |
1177 | if (inode->i_ino >= F2FS_ROOT_INO(sbi) && |
1178 | (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE)) | |
a7ffdbe2 JK |
1179 | return; |
1180 | ||
487261f3 CY |
1181 | if (PageDirty(page)) { |
1182 | if (inode->i_ino == F2FS_META_INO(sbi)) | |
1183 | dec_page_count(sbi, F2FS_DIRTY_META); | |
1184 | else if (inode->i_ino == F2FS_NODE_INO(sbi)) | |
1185 | dec_page_count(sbi, F2FS_DIRTY_NODES); | |
1186 | else | |
1187 | inode_dec_dirty_pages(inode); | |
1188 | } | |
eb47b800 JK |
1189 | ClearPagePrivate(page); |
1190 | } | |
1191 | ||
487261f3 | 1192 | int f2fs_release_page(struct page *page, gfp_t wait) |
eb47b800 | 1193 | { |
f68daeeb JK |
1194 | /* If this is dirty page, keep PagePrivate */ |
1195 | if (PageDirty(page)) | |
1196 | return 0; | |
1197 | ||
eb47b800 | 1198 | ClearPagePrivate(page); |
c3850aa1 | 1199 | return 1; |
eb47b800 JK |
1200 | } |
1201 | ||
1202 | static int f2fs_set_data_page_dirty(struct page *page) | |
1203 | { | |
1204 | struct address_space *mapping = page->mapping; | |
1205 | struct inode *inode = mapping->host; | |
1206 | ||
26c6b887 JK |
1207 | trace_f2fs_set_page_dirty(page, DATA); |
1208 | ||
eb47b800 | 1209 | SetPageUptodate(page); |
34ba94ba | 1210 | |
1e84371f | 1211 | if (f2fs_is_atomic_file(inode)) { |
34ba94ba JK |
1212 | register_inmem_page(inode, page); |
1213 | return 1; | |
1214 | } | |
1215 | ||
a18ff063 JK |
1216 | mark_inode_dirty(inode); |
1217 | ||
eb47b800 JK |
1218 | if (!PageDirty(page)) { |
1219 | __set_page_dirty_nobuffers(page); | |
a7ffdbe2 | 1220 | update_dirty_page(inode, page); |
eb47b800 JK |
1221 | return 1; |
1222 | } | |
1223 | return 0; | |
1224 | } | |
1225 | ||
c01e54b7 JK |
1226 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
1227 | { | |
454ae7e5 CY |
1228 | struct inode *inode = mapping->host; |
1229 | ||
b3d208f9 JK |
1230 | /* we don't need to use inline_data strictly */ |
1231 | if (f2fs_has_inline_data(inode)) { | |
1232 | int err = f2fs_convert_inline_inode(inode); | |
1233 | if (err) | |
1234 | return err; | |
1235 | } | |
bfad7c2d | 1236 | return generic_block_bmap(mapping, block, get_data_block); |
c01e54b7 JK |
1237 | } |
1238 | ||
eb47b800 JK |
1239 | const struct address_space_operations f2fs_dblock_aops = { |
1240 | .readpage = f2fs_read_data_page, | |
1241 | .readpages = f2fs_read_data_pages, | |
1242 | .writepage = f2fs_write_data_page, | |
1243 | .writepages = f2fs_write_data_pages, | |
1244 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 1245 | .write_end = f2fs_write_end, |
eb47b800 | 1246 | .set_page_dirty = f2fs_set_data_page_dirty, |
487261f3 CY |
1247 | .invalidatepage = f2fs_invalidate_page, |
1248 | .releasepage = f2fs_release_page, | |
eb47b800 | 1249 | .direct_IO = f2fs_direct_IO, |
c01e54b7 | 1250 | .bmap = f2fs_bmap, |
eb47b800 | 1251 | }; |