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" | |
848753aa | 25 | #include <trace/events/f2fs.h> |
eb47b800 | 26 | |
93dfe2ac JK |
27 | /* |
28 | * Low-level block read/write IO operations. | |
29 | */ | |
30 | static struct bio *__bio_alloc(struct block_device *bdev, int npages) | |
31 | { | |
32 | struct bio *bio; | |
33 | ||
34 | /* No failure on bio allocation */ | |
35 | bio = bio_alloc(GFP_NOIO, npages); | |
36 | bio->bi_bdev = bdev; | |
37 | bio->bi_private = NULL; | |
38 | return bio; | |
39 | } | |
40 | ||
41 | static void f2fs_read_end_io(struct bio *bio, int err) | |
42 | { | |
43 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
44 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
45 | ||
46 | do { | |
47 | struct page *page = bvec->bv_page; | |
48 | ||
49 | if (--bvec >= bio->bi_io_vec) | |
50 | prefetchw(&bvec->bv_page->flags); | |
51 | ||
52 | if (uptodate) { | |
53 | SetPageUptodate(page); | |
54 | } else { | |
55 | ClearPageUptodate(page); | |
56 | SetPageError(page); | |
57 | } | |
58 | unlock_page(page); | |
59 | } while (bvec >= bio->bi_io_vec); | |
60 | ||
61 | bio_put(bio); | |
62 | } | |
63 | ||
64 | static void f2fs_write_end_io(struct bio *bio, int err) | |
65 | { | |
66 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
67 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
68 | struct f2fs_sb_info *sbi = F2FS_SB(bvec->bv_page->mapping->host->i_sb); | |
69 | ||
70 | do { | |
71 | struct page *page = bvec->bv_page; | |
72 | ||
73 | if (--bvec >= bio->bi_io_vec) | |
74 | prefetchw(&bvec->bv_page->flags); | |
75 | ||
76 | if (!uptodate) { | |
77 | SetPageError(page); | |
78 | set_bit(AS_EIO, &page->mapping->flags); | |
79 | set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG); | |
80 | sbi->sb->s_flags |= MS_RDONLY; | |
81 | } | |
82 | end_page_writeback(page); | |
83 | dec_page_count(sbi, F2FS_WRITEBACK); | |
84 | } while (bvec >= bio->bi_io_vec); | |
85 | ||
86 | if (bio->bi_private) | |
87 | complete(bio->bi_private); | |
88 | ||
89 | if (!get_pages(sbi, F2FS_WRITEBACK) && | |
90 | !list_empty(&sbi->cp_wait.task_list)) | |
91 | wake_up(&sbi->cp_wait); | |
92 | ||
93 | bio_put(bio); | |
94 | } | |
95 | ||
96 | static void __submit_merged_bio(struct f2fs_sb_info *sbi, | |
97 | struct f2fs_bio_info *io, | |
98 | enum page_type type, bool sync, int rw) | |
99 | { | |
100 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
101 | ||
102 | if (!io->bio) | |
103 | return; | |
104 | ||
105 | if (btype == META) | |
106 | rw |= REQ_META; | |
107 | ||
108 | if (is_read_io(rw)) { | |
109 | if (sync) | |
110 | rw |= READ_SYNC; | |
111 | submit_bio(rw, io->bio); | |
112 | trace_f2fs_submit_read_bio(sbi->sb, rw, type, io->bio); | |
113 | io->bio = NULL; | |
114 | return; | |
115 | } | |
116 | ||
117 | if (sync) | |
118 | rw |= WRITE_SYNC; | |
119 | if (type >= META_FLUSH) | |
120 | rw |= WRITE_FLUSH_FUA; | |
121 | ||
122 | /* | |
123 | * META_FLUSH is only from the checkpoint procedure, and we should wait | |
124 | * this metadata bio for FS consistency. | |
125 | */ | |
126 | if (type == META_FLUSH) { | |
127 | DECLARE_COMPLETION_ONSTACK(wait); | |
128 | io->bio->bi_private = &wait; | |
129 | submit_bio(rw, io->bio); | |
130 | wait_for_completion(&wait); | |
131 | } else { | |
132 | submit_bio(rw, io->bio); | |
133 | } | |
134 | trace_f2fs_submit_write_bio(sbi->sb, rw, btype, io->bio); | |
135 | io->bio = NULL; | |
136 | } | |
137 | ||
138 | void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, | |
139 | enum page_type type, bool sync, int rw) | |
140 | { | |
141 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
142 | struct f2fs_bio_info *io; | |
143 | ||
144 | io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype]; | |
145 | ||
146 | mutex_lock(&io->io_mutex); | |
147 | __submit_merged_bio(sbi, io, type, sync, rw); | |
148 | mutex_unlock(&io->io_mutex); | |
149 | } | |
150 | ||
151 | /* | |
152 | * Fill the locked page with data located in the block address. | |
153 | * Return unlocked page. | |
154 | */ | |
155 | int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page, | |
156 | block_t blk_addr, int rw) | |
157 | { | |
158 | struct block_device *bdev = sbi->sb->s_bdev; | |
159 | struct bio *bio; | |
160 | ||
161 | trace_f2fs_submit_page_bio(page, blk_addr, rw); | |
162 | ||
163 | /* Allocate a new bio */ | |
164 | bio = __bio_alloc(bdev, 1); | |
165 | ||
166 | /* Initialize the bio */ | |
167 | bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); | |
168 | bio->bi_end_io = is_read_io(rw) ? f2fs_read_end_io : f2fs_write_end_io; | |
169 | ||
170 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { | |
171 | bio_put(bio); | |
172 | f2fs_put_page(page, 1); | |
173 | return -EFAULT; | |
174 | } | |
175 | ||
176 | submit_bio(rw, bio); | |
177 | return 0; | |
178 | } | |
179 | ||
180 | void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page, | |
181 | block_t blk_addr, enum page_type type, int rw) | |
182 | { | |
183 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
184 | struct block_device *bdev = sbi->sb->s_bdev; | |
185 | struct f2fs_bio_info *io; | |
186 | int bio_blocks; | |
187 | ||
188 | io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype]; | |
189 | ||
190 | verify_block_addr(sbi, blk_addr); | |
191 | ||
192 | mutex_lock(&io->io_mutex); | |
193 | ||
194 | if (!is_read_io(rw)) | |
195 | inc_page_count(sbi, F2FS_WRITEBACK); | |
196 | ||
197 | if (io->bio && io->last_block_in_bio != blk_addr - 1) | |
198 | __submit_merged_bio(sbi, io, type, true, rw); | |
199 | alloc_new: | |
200 | if (io->bio == NULL) { | |
201 | bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); | |
202 | io->bio = __bio_alloc(bdev, bio_blocks); | |
203 | io->bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); | |
204 | io->bio->bi_end_io = is_read_io(rw) ? f2fs_read_end_io : | |
205 | f2fs_write_end_io; | |
206 | /* | |
207 | * The end_io will be assigned at the sumbission phase. | |
208 | * Until then, let bio_add_page() merge consecutive IOs as much | |
209 | * as possible. | |
210 | */ | |
211 | } | |
212 | ||
213 | if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) < | |
214 | PAGE_CACHE_SIZE) { | |
215 | __submit_merged_bio(sbi, io, type, true, rw); | |
216 | goto alloc_new; | |
217 | } | |
218 | ||
219 | io->last_block_in_bio = blk_addr; | |
220 | ||
221 | mutex_unlock(&io->io_mutex); | |
222 | trace_f2fs_submit_page_mbio(page, rw, type, blk_addr); | |
223 | } | |
224 | ||
0a8165d7 | 225 | /* |
eb47b800 JK |
226 | * Lock ordering for the change of data block address: |
227 | * ->data_page | |
228 | * ->node_page | |
229 | * update block addresses in the node page | |
230 | */ | |
231 | static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr) | |
232 | { | |
233 | struct f2fs_node *rn; | |
234 | __le32 *addr_array; | |
235 | struct page *node_page = dn->node_page; | |
236 | unsigned int ofs_in_node = dn->ofs_in_node; | |
237 | ||
a569469e | 238 | f2fs_wait_on_page_writeback(node_page, NODE, false); |
eb47b800 | 239 | |
45590710 | 240 | rn = F2FS_NODE(node_page); |
eb47b800 JK |
241 | |
242 | /* Get physical address of data block */ | |
243 | addr_array = blkaddr_in_node(rn); | |
244 | addr_array[ofs_in_node] = cpu_to_le32(new_addr); | |
245 | set_page_dirty(node_page); | |
246 | } | |
247 | ||
248 | int reserve_new_block(struct dnode_of_data *dn) | |
249 | { | |
250 | struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); | |
251 | ||
252 | if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)) | |
253 | return -EPERM; | |
254 | if (!inc_valid_block_count(sbi, dn->inode, 1)) | |
255 | return -ENOSPC; | |
256 | ||
c01e2853 NJ |
257 | trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node); |
258 | ||
eb47b800 JK |
259 | __set_data_blkaddr(dn, NEW_ADDR); |
260 | dn->data_blkaddr = NEW_ADDR; | |
261 | sync_inode_page(dn); | |
262 | return 0; | |
263 | } | |
264 | ||
b600965c HL |
265 | int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) |
266 | { | |
267 | bool need_put = dn->inode_page ? false : true; | |
268 | int err; | |
269 | ||
270 | err = get_dnode_of_data(dn, index, ALLOC_NODE); | |
271 | if (err) | |
272 | return err; | |
273 | if (dn->data_blkaddr == NULL_ADDR) | |
274 | err = reserve_new_block(dn); | |
275 | ||
276 | if (need_put) | |
277 | f2fs_put_dnode(dn); | |
278 | return err; | |
279 | } | |
280 | ||
eb47b800 JK |
281 | static int check_extent_cache(struct inode *inode, pgoff_t pgofs, |
282 | struct buffer_head *bh_result) | |
283 | { | |
284 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
eb47b800 JK |
285 | pgoff_t start_fofs, end_fofs; |
286 | block_t start_blkaddr; | |
287 | ||
c11abd1a JK |
288 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
289 | return 0; | |
290 | ||
eb47b800 JK |
291 | read_lock(&fi->ext.ext_lock); |
292 | if (fi->ext.len == 0) { | |
293 | read_unlock(&fi->ext.ext_lock); | |
294 | return 0; | |
295 | } | |
296 | ||
dcdfff65 JK |
297 | stat_inc_total_hit(inode->i_sb); |
298 | ||
eb47b800 JK |
299 | start_fofs = fi->ext.fofs; |
300 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
301 | start_blkaddr = fi->ext.blk_addr; | |
302 | ||
303 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
304 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
305 | size_t count; | |
306 | ||
307 | clear_buffer_new(bh_result); | |
308 | map_bh(bh_result, inode->i_sb, | |
309 | start_blkaddr + pgofs - start_fofs); | |
310 | count = end_fofs - pgofs + 1; | |
311 | if (count < (UINT_MAX >> blkbits)) | |
312 | bh_result->b_size = (count << blkbits); | |
313 | else | |
314 | bh_result->b_size = UINT_MAX; | |
315 | ||
dcdfff65 | 316 | stat_inc_read_hit(inode->i_sb); |
eb47b800 JK |
317 | read_unlock(&fi->ext.ext_lock); |
318 | return 1; | |
319 | } | |
320 | read_unlock(&fi->ext.ext_lock); | |
321 | return 0; | |
322 | } | |
323 | ||
324 | void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn) | |
325 | { | |
326 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
327 | pgoff_t fofs, start_fofs, end_fofs; | |
328 | block_t start_blkaddr, end_blkaddr; | |
c11abd1a | 329 | int need_update = true; |
eb47b800 | 330 | |
5d56b671 | 331 | f2fs_bug_on(blk_addr == NEW_ADDR); |
de93653f JK |
332 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + |
333 | dn->ofs_in_node; | |
eb47b800 JK |
334 | |
335 | /* Update the page address in the parent node */ | |
336 | __set_data_blkaddr(dn, blk_addr); | |
337 | ||
c11abd1a JK |
338 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
339 | return; | |
340 | ||
eb47b800 JK |
341 | write_lock(&fi->ext.ext_lock); |
342 | ||
343 | start_fofs = fi->ext.fofs; | |
344 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
345 | start_blkaddr = fi->ext.blk_addr; | |
346 | end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1; | |
347 | ||
348 | /* Drop and initialize the matched extent */ | |
349 | if (fi->ext.len == 1 && fofs == start_fofs) | |
350 | fi->ext.len = 0; | |
351 | ||
352 | /* Initial extent */ | |
353 | if (fi->ext.len == 0) { | |
354 | if (blk_addr != NULL_ADDR) { | |
355 | fi->ext.fofs = fofs; | |
356 | fi->ext.blk_addr = blk_addr; | |
357 | fi->ext.len = 1; | |
358 | } | |
359 | goto end_update; | |
360 | } | |
361 | ||
6224da87 | 362 | /* Front merge */ |
eb47b800 JK |
363 | if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) { |
364 | fi->ext.fofs--; | |
365 | fi->ext.blk_addr--; | |
366 | fi->ext.len++; | |
367 | goto end_update; | |
368 | } | |
369 | ||
370 | /* Back merge */ | |
371 | if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) { | |
372 | fi->ext.len++; | |
373 | goto end_update; | |
374 | } | |
375 | ||
376 | /* Split the existing extent */ | |
377 | if (fi->ext.len > 1 && | |
378 | fofs >= start_fofs && fofs <= end_fofs) { | |
379 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
380 | fi->ext.len = fofs - start_fofs; | |
381 | } else { | |
382 | fi->ext.fofs = fofs + 1; | |
383 | fi->ext.blk_addr = start_blkaddr + | |
384 | fofs - start_fofs + 1; | |
385 | fi->ext.len -= fofs - start_fofs + 1; | |
386 | } | |
c11abd1a JK |
387 | } else { |
388 | need_update = false; | |
eb47b800 | 389 | } |
eb47b800 | 390 | |
c11abd1a JK |
391 | /* Finally, if the extent is very fragmented, let's drop the cache. */ |
392 | if (fi->ext.len < F2FS_MIN_EXTENT_LEN) { | |
393 | fi->ext.len = 0; | |
394 | set_inode_flag(fi, FI_NO_EXTENT); | |
395 | need_update = true; | |
396 | } | |
eb47b800 JK |
397 | end_update: |
398 | write_unlock(&fi->ext.ext_lock); | |
c11abd1a JK |
399 | if (need_update) |
400 | sync_inode_page(dn); | |
401 | return; | |
eb47b800 JK |
402 | } |
403 | ||
c718379b | 404 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 JK |
405 | { |
406 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
407 | struct address_space *mapping = inode->i_mapping; | |
408 | struct dnode_of_data dn; | |
409 | struct page *page; | |
410 | int err; | |
411 | ||
412 | page = find_get_page(mapping, index); | |
413 | if (page && PageUptodate(page)) | |
414 | return page; | |
415 | f2fs_put_page(page, 0); | |
416 | ||
417 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 418 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
419 | if (err) |
420 | return ERR_PTR(err); | |
421 | f2fs_put_dnode(&dn); | |
422 | ||
423 | if (dn.data_blkaddr == NULL_ADDR) | |
424 | return ERR_PTR(-ENOENT); | |
425 | ||
426 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
427 | if (dn.data_blkaddr == NEW_ADDR) | |
428 | return ERR_PTR(-EINVAL); | |
429 | ||
6f85b352 | 430 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
eb47b800 JK |
431 | if (!page) |
432 | return ERR_PTR(-ENOMEM); | |
433 | ||
393ff91f JK |
434 | if (PageUptodate(page)) { |
435 | unlock_page(page); | |
436 | return page; | |
437 | } | |
438 | ||
93dfe2ac | 439 | err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, |
c718379b | 440 | sync ? READ_SYNC : READA); |
1069bbf7 CY |
441 | if (err) |
442 | return ERR_PTR(err); | |
443 | ||
c718379b JK |
444 | if (sync) { |
445 | wait_on_page_locked(page); | |
446 | if (!PageUptodate(page)) { | |
447 | f2fs_put_page(page, 0); | |
448 | return ERR_PTR(-EIO); | |
449 | } | |
eb47b800 | 450 | } |
eb47b800 JK |
451 | return page; |
452 | } | |
453 | ||
0a8165d7 | 454 | /* |
eb47b800 JK |
455 | * If it tries to access a hole, return an error. |
456 | * Because, the callers, functions in dir.c and GC, should be able to know | |
457 | * whether this page exists or not. | |
458 | */ | |
459 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
460 | { | |
461 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
462 | struct address_space *mapping = inode->i_mapping; | |
463 | struct dnode_of_data dn; | |
464 | struct page *page; | |
465 | int err; | |
466 | ||
650495de | 467 | repeat: |
6f85b352 | 468 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
650495de JK |
469 | if (!page) |
470 | return ERR_PTR(-ENOMEM); | |
471 | ||
eb47b800 | 472 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 473 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
474 | if (err) { |
475 | f2fs_put_page(page, 1); | |
eb47b800 | 476 | return ERR_PTR(err); |
650495de | 477 | } |
eb47b800 JK |
478 | f2fs_put_dnode(&dn); |
479 | ||
650495de JK |
480 | if (dn.data_blkaddr == NULL_ADDR) { |
481 | f2fs_put_page(page, 1); | |
eb47b800 | 482 | return ERR_PTR(-ENOENT); |
650495de | 483 | } |
eb47b800 JK |
484 | |
485 | if (PageUptodate(page)) | |
486 | return page; | |
487 | ||
d59ff4df JK |
488 | /* |
489 | * A new dentry page is allocated but not able to be written, since its | |
490 | * new inode page couldn't be allocated due to -ENOSPC. | |
491 | * In such the case, its blkaddr can be remained as NEW_ADDR. | |
492 | * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. | |
493 | */ | |
494 | if (dn.data_blkaddr == NEW_ADDR) { | |
495 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
496 | SetPageUptodate(page); | |
497 | return page; | |
498 | } | |
eb47b800 | 499 | |
93dfe2ac | 500 | err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC); |
393ff91f | 501 | if (err) |
eb47b800 | 502 | return ERR_PTR(err); |
393ff91f JK |
503 | |
504 | lock_page(page); | |
505 | if (!PageUptodate(page)) { | |
506 | f2fs_put_page(page, 1); | |
507 | return ERR_PTR(-EIO); | |
eb47b800 | 508 | } |
afcb7ca0 JK |
509 | if (page->mapping != mapping) { |
510 | f2fs_put_page(page, 1); | |
511 | goto repeat; | |
eb47b800 JK |
512 | } |
513 | return page; | |
514 | } | |
515 | ||
0a8165d7 | 516 | /* |
eb47b800 JK |
517 | * Caller ensures that this data page is never allocated. |
518 | * A new zero-filled data page is allocated in the page cache. | |
39936837 JK |
519 | * |
520 | * Also, caller should grab and release a mutex by calling mutex_lock_op() and | |
521 | * mutex_unlock_op(). | |
44a83ff6 | 522 | * Note that, npage is set only by make_empty_dir. |
eb47b800 | 523 | */ |
64aa7ed9 JK |
524 | struct page *get_new_data_page(struct inode *inode, |
525 | struct page *npage, pgoff_t index, bool new_i_size) | |
eb47b800 JK |
526 | { |
527 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
528 | struct address_space *mapping = inode->i_mapping; | |
529 | struct page *page; | |
530 | struct dnode_of_data dn; | |
531 | int err; | |
532 | ||
64aa7ed9 | 533 | set_new_dnode(&dn, inode, npage, npage, 0); |
b600965c | 534 | err = f2fs_reserve_block(&dn, index); |
eb47b800 JK |
535 | if (err) |
536 | return ERR_PTR(err); | |
537 | ||
afcb7ca0 | 538 | repeat: |
eb47b800 JK |
539 | page = grab_cache_page(mapping, index); |
540 | if (!page) | |
541 | return ERR_PTR(-ENOMEM); | |
542 | ||
543 | if (PageUptodate(page)) | |
544 | return page; | |
545 | ||
546 | if (dn.data_blkaddr == NEW_ADDR) { | |
547 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 548 | SetPageUptodate(page); |
eb47b800 | 549 | } else { |
93dfe2ac JK |
550 | err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, |
551 | READ_SYNC); | |
393ff91f | 552 | if (err) |
eb47b800 | 553 | return ERR_PTR(err); |
393ff91f JK |
554 | lock_page(page); |
555 | if (!PageUptodate(page)) { | |
556 | f2fs_put_page(page, 1); | |
557 | return ERR_PTR(-EIO); | |
eb47b800 | 558 | } |
afcb7ca0 JK |
559 | if (page->mapping != mapping) { |
560 | f2fs_put_page(page, 1); | |
561 | goto repeat; | |
eb47b800 JK |
562 | } |
563 | } | |
eb47b800 JK |
564 | |
565 | if (new_i_size && | |
566 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
567 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
568 | /* Only the directory inode sets new_i_size */ |
569 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
570 | mark_inode_dirty_sync(inode); |
571 | } | |
572 | return page; | |
573 | } | |
574 | ||
0a8165d7 | 575 | /* |
eb47b800 JK |
576 | * This function should be used by the data read flow only where it |
577 | * does not check the "create" flag that indicates block allocation. | |
578 | * The reason for this special functionality is to exploit VFS readahead | |
579 | * mechanism. | |
580 | */ | |
581 | static int get_data_block_ro(struct inode *inode, sector_t iblock, | |
582 | struct buffer_head *bh_result, int create) | |
583 | { | |
584 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
585 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
586 | struct dnode_of_data dn; | |
587 | pgoff_t pgofs; | |
588 | int err; | |
589 | ||
590 | /* Get the page offset from the block offset(iblock) */ | |
591 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
592 | ||
848753aa NJ |
593 | if (check_extent_cache(inode, pgofs, bh_result)) { |
594 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); | |
eb47b800 | 595 | return 0; |
848753aa | 596 | } |
eb47b800 JK |
597 | |
598 | /* When reading holes, we need its node page */ | |
599 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 600 | err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA); |
848753aa NJ |
601 | if (err) { |
602 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
eb47b800 | 603 | return (err == -ENOENT) ? 0 : err; |
848753aa | 604 | } |
eb47b800 JK |
605 | |
606 | /* It does not support data allocation */ | |
5d56b671 | 607 | f2fs_bug_on(create); |
eb47b800 JK |
608 | |
609 | if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) { | |
610 | int i; | |
611 | unsigned int end_offset; | |
612 | ||
613 | end_offset = IS_INODE(dn.node_page) ? | |
de93653f | 614 | ADDRS_PER_INODE(F2FS_I(inode)) : |
eb47b800 JK |
615 | ADDRS_PER_BLOCK; |
616 | ||
617 | clear_buffer_new(bh_result); | |
618 | ||
619 | /* Give more consecutive addresses for the read ahead */ | |
620 | for (i = 0; i < end_offset - dn.ofs_in_node; i++) | |
621 | if (((datablock_addr(dn.node_page, | |
622 | dn.ofs_in_node + i)) | |
623 | != (dn.data_blkaddr + i)) || maxblocks == i) | |
624 | break; | |
625 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); | |
f9a4e6df | 626 | bh_result->b_size = (((size_t)i) << blkbits); |
eb47b800 JK |
627 | } |
628 | f2fs_put_dnode(&dn); | |
848753aa | 629 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); |
eb47b800 JK |
630 | return 0; |
631 | } | |
632 | ||
633 | static int f2fs_read_data_page(struct file *file, struct page *page) | |
634 | { | |
635 | return mpage_readpage(page, get_data_block_ro); | |
636 | } | |
637 | ||
638 | static int f2fs_read_data_pages(struct file *file, | |
639 | struct address_space *mapping, | |
640 | struct list_head *pages, unsigned nr_pages) | |
641 | { | |
642 | return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro); | |
643 | } | |
644 | ||
645 | int do_write_data_page(struct page *page) | |
646 | { | |
647 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
648 | block_t old_blk_addr, new_blk_addr; |
649 | struct dnode_of_data dn; | |
650 | int err = 0; | |
651 | ||
652 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 653 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
654 | if (err) |
655 | return err; | |
656 | ||
657 | old_blk_addr = dn.data_blkaddr; | |
658 | ||
659 | /* This page is already truncated */ | |
660 | if (old_blk_addr == NULL_ADDR) | |
661 | goto out_writepage; | |
662 | ||
663 | set_page_writeback(page); | |
664 | ||
665 | /* | |
666 | * If current allocation needs SSR, | |
667 | * it had better in-place writes for updated data. | |
668 | */ | |
b25958b6 HL |
669 | if (unlikely(old_blk_addr != NEW_ADDR && |
670 | !is_cold_data(page) && | |
671 | need_inplace_update(inode))) { | |
eb47b800 JK |
672 | rewrite_data_page(F2FS_SB(inode->i_sb), page, |
673 | old_blk_addr); | |
674 | } else { | |
675 | write_data_page(inode, page, &dn, | |
676 | old_blk_addr, &new_blk_addr); | |
677 | update_extent_cache(new_blk_addr, &dn); | |
eb47b800 JK |
678 | } |
679 | out_writepage: | |
680 | f2fs_put_dnode(&dn); | |
681 | return err; | |
682 | } | |
683 | ||
684 | static int f2fs_write_data_page(struct page *page, | |
685 | struct writeback_control *wbc) | |
686 | { | |
687 | struct inode *inode = page->mapping->host; | |
688 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
689 | loff_t i_size = i_size_read(inode); | |
690 | const pgoff_t end_index = ((unsigned long long) i_size) | |
691 | >> PAGE_CACHE_SHIFT; | |
692 | unsigned offset; | |
39936837 | 693 | bool need_balance_fs = false; |
eb47b800 JK |
694 | int err = 0; |
695 | ||
696 | if (page->index < end_index) | |
39936837 | 697 | goto write; |
eb47b800 JK |
698 | |
699 | /* | |
700 | * If the offset is out-of-range of file size, | |
701 | * this page does not have to be written to disk. | |
702 | */ | |
703 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
704 | if ((page->index >= end_index + 1) || !offset) { | |
705 | if (S_ISDIR(inode->i_mode)) { | |
706 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
707 | inode_dec_dirty_dents(inode); | |
708 | } | |
39936837 | 709 | goto out; |
eb47b800 JK |
710 | } |
711 | ||
712 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 JK |
713 | write: |
714 | if (sbi->por_doing) { | |
715 | err = AOP_WRITEPAGE_ACTIVATE; | |
eb47b800 | 716 | goto redirty_out; |
39936837 | 717 | } |
eb47b800 | 718 | |
39936837 | 719 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 JK |
720 | if (S_ISDIR(inode->i_mode)) { |
721 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
722 | inode_dec_dirty_dents(inode); | |
39936837 JK |
723 | err = do_write_data_page(page); |
724 | } else { | |
e479556b | 725 | f2fs_lock_op(sbi); |
39936837 | 726 | err = do_write_data_page(page); |
e479556b | 727 | f2fs_unlock_op(sbi); |
39936837 | 728 | need_balance_fs = true; |
eb47b800 | 729 | } |
39936837 JK |
730 | if (err == -ENOENT) |
731 | goto out; | |
732 | else if (err) | |
733 | goto redirty_out; | |
eb47b800 JK |
734 | |
735 | if (wbc->for_reclaim) | |
93dfe2ac | 736 | f2fs_submit_merged_bio(sbi, DATA, true, WRITE); |
eb47b800 | 737 | |
eb47b800 | 738 | clear_cold_data(page); |
39936837 | 739 | out: |
eb47b800 | 740 | unlock_page(page); |
39936837 | 741 | if (need_balance_fs) |
eb47b800 JK |
742 | f2fs_balance_fs(sbi); |
743 | return 0; | |
744 | ||
eb47b800 JK |
745 | redirty_out: |
746 | wbc->pages_skipped++; | |
747 | set_page_dirty(page); | |
39936837 | 748 | return err; |
eb47b800 JK |
749 | } |
750 | ||
751 | #define MAX_DESIRED_PAGES_WP 4096 | |
752 | ||
fa9150a8 NJ |
753 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
754 | void *data) | |
755 | { | |
756 | struct address_space *mapping = data; | |
757 | int ret = mapping->a_ops->writepage(page, wbc); | |
758 | mapping_set_error(mapping, ret); | |
759 | return ret; | |
760 | } | |
761 | ||
25ca923b | 762 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
763 | struct writeback_control *wbc) |
764 | { | |
765 | struct inode *inode = mapping->host; | |
766 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
531ad7d5 | 767 | bool locked = false; |
eb47b800 JK |
768 | int ret; |
769 | long excess_nrtw = 0, desired_nrtw; | |
770 | ||
cfb185a1 | 771 | /* deal with chardevs and other special file */ |
772 | if (!mapping->a_ops->writepage) | |
773 | return 0; | |
774 | ||
eb47b800 JK |
775 | if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) { |
776 | desired_nrtw = MAX_DESIRED_PAGES_WP; | |
777 | excess_nrtw = desired_nrtw - wbc->nr_to_write; | |
778 | wbc->nr_to_write = desired_nrtw; | |
779 | } | |
780 | ||
531ad7d5 | 781 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 782 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
783 | locked = true; |
784 | } | |
fa9150a8 | 785 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 786 | if (locked) |
eb47b800 | 787 | mutex_unlock(&sbi->writepages); |
93dfe2ac | 788 | f2fs_submit_merged_bio(sbi, DATA, wbc->sync_mode == WB_SYNC_ALL, WRITE); |
eb47b800 JK |
789 | |
790 | remove_dirty_dir_inode(inode); | |
791 | ||
792 | wbc->nr_to_write -= excess_nrtw; | |
793 | return ret; | |
794 | } | |
795 | ||
796 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, | |
797 | loff_t pos, unsigned len, unsigned flags, | |
798 | struct page **pagep, void **fsdata) | |
799 | { | |
800 | struct inode *inode = mapping->host; | |
801 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
802 | struct page *page; | |
803 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; | |
804 | struct dnode_of_data dn; | |
805 | int err = 0; | |
806 | ||
eb47b800 | 807 | f2fs_balance_fs(sbi); |
afcb7ca0 | 808 | repeat: |
eb47b800 JK |
809 | page = grab_cache_page_write_begin(mapping, index, flags); |
810 | if (!page) | |
811 | return -ENOMEM; | |
812 | *pagep = page; | |
813 | ||
e479556b | 814 | f2fs_lock_op(sbi); |
eb47b800 | 815 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
b600965c | 816 | err = f2fs_reserve_block(&dn, index); |
e479556b | 817 | f2fs_unlock_op(sbi); |
eb47b800 | 818 | |
b600965c HL |
819 | if (err) { |
820 | f2fs_put_page(page, 1); | |
821 | return err; | |
822 | } | |
823 | ||
eb47b800 JK |
824 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) |
825 | return 0; | |
826 | ||
827 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
828 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
829 | unsigned end = start + len; | |
830 | ||
831 | /* Reading beyond i_size is simple: memset to zero */ | |
832 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 833 | goto out; |
eb47b800 JK |
834 | } |
835 | ||
836 | if (dn.data_blkaddr == NEW_ADDR) { | |
837 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
838 | } else { | |
93dfe2ac JK |
839 | err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, |
840 | READ_SYNC); | |
393ff91f | 841 | if (err) |
eb47b800 | 842 | return err; |
393ff91f JK |
843 | lock_page(page); |
844 | if (!PageUptodate(page)) { | |
845 | f2fs_put_page(page, 1); | |
846 | return -EIO; | |
eb47b800 | 847 | } |
afcb7ca0 JK |
848 | if (page->mapping != mapping) { |
849 | f2fs_put_page(page, 1); | |
850 | goto repeat; | |
eb47b800 JK |
851 | } |
852 | } | |
393ff91f | 853 | out: |
eb47b800 JK |
854 | SetPageUptodate(page); |
855 | clear_cold_data(page); | |
856 | return 0; | |
857 | } | |
858 | ||
a1dd3c13 JK |
859 | static int f2fs_write_end(struct file *file, |
860 | struct address_space *mapping, | |
861 | loff_t pos, unsigned len, unsigned copied, | |
862 | struct page *page, void *fsdata) | |
863 | { | |
864 | struct inode *inode = page->mapping->host; | |
865 | ||
866 | SetPageUptodate(page); | |
867 | set_page_dirty(page); | |
868 | ||
869 | if (pos + copied > i_size_read(inode)) { | |
870 | i_size_write(inode, pos + copied); | |
871 | mark_inode_dirty(inode); | |
872 | update_inode_page(inode); | |
873 | } | |
874 | ||
75c3c8bc | 875 | f2fs_put_page(page, 1); |
a1dd3c13 JK |
876 | return copied; |
877 | } | |
878 | ||
eb47b800 JK |
879 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
880 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
881 | { | |
882 | struct file *file = iocb->ki_filp; | |
883 | struct inode *inode = file->f_mapping->host; | |
884 | ||
885 | if (rw == WRITE) | |
886 | return 0; | |
887 | ||
888 | /* Needs synchronization with the cleaner */ | |
889 | return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, | |
890 | get_data_block_ro); | |
891 | } | |
892 | ||
d47992f8 LC |
893 | static void f2fs_invalidate_data_page(struct page *page, unsigned int offset, |
894 | unsigned int length) | |
eb47b800 JK |
895 | { |
896 | struct inode *inode = page->mapping->host; | |
897 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
898 | if (S_ISDIR(inode->i_mode) && PageDirty(page)) { | |
899 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
900 | inode_dec_dirty_dents(inode); | |
901 | } | |
902 | ClearPagePrivate(page); | |
903 | } | |
904 | ||
905 | static int f2fs_release_data_page(struct page *page, gfp_t wait) | |
906 | { | |
907 | ClearPagePrivate(page); | |
c3850aa1 | 908 | return 1; |
eb47b800 JK |
909 | } |
910 | ||
911 | static int f2fs_set_data_page_dirty(struct page *page) | |
912 | { | |
913 | struct address_space *mapping = page->mapping; | |
914 | struct inode *inode = mapping->host; | |
915 | ||
26c6b887 JK |
916 | trace_f2fs_set_page_dirty(page, DATA); |
917 | ||
eb47b800 JK |
918 | SetPageUptodate(page); |
919 | if (!PageDirty(page)) { | |
920 | __set_page_dirty_nobuffers(page); | |
921 | set_dirty_dir_page(inode, page); | |
922 | return 1; | |
923 | } | |
924 | return 0; | |
925 | } | |
926 | ||
c01e54b7 JK |
927 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
928 | { | |
929 | return generic_block_bmap(mapping, block, get_data_block_ro); | |
930 | } | |
931 | ||
eb47b800 JK |
932 | const struct address_space_operations f2fs_dblock_aops = { |
933 | .readpage = f2fs_read_data_page, | |
934 | .readpages = f2fs_read_data_pages, | |
935 | .writepage = f2fs_write_data_page, | |
936 | .writepages = f2fs_write_data_pages, | |
937 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 938 | .write_end = f2fs_write_end, |
eb47b800 JK |
939 | .set_page_dirty = f2fs_set_data_page_dirty, |
940 | .invalidatepage = f2fs_invalidate_data_page, | |
941 | .releasepage = f2fs_release_data_page, | |
942 | .direct_IO = f2fs_direct_IO, | |
c01e54b7 | 943 | .bmap = f2fs_bmap, |
eb47b800 | 944 | }; |