]>
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 | |
0a8165d7 | 27 | /* |
eb47b800 JK |
28 | * Lock ordering for the change of data block address: |
29 | * ->data_page | |
30 | * ->node_page | |
31 | * update block addresses in the node page | |
32 | */ | |
33 | static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr) | |
34 | { | |
35 | struct f2fs_node *rn; | |
36 | __le32 *addr_array; | |
37 | struct page *node_page = dn->node_page; | |
38 | unsigned int ofs_in_node = dn->ofs_in_node; | |
39 | ||
a569469e | 40 | f2fs_wait_on_page_writeback(node_page, NODE, false); |
eb47b800 | 41 | |
45590710 | 42 | rn = F2FS_NODE(node_page); |
eb47b800 JK |
43 | |
44 | /* Get physical address of data block */ | |
45 | addr_array = blkaddr_in_node(rn); | |
46 | addr_array[ofs_in_node] = cpu_to_le32(new_addr); | |
47 | set_page_dirty(node_page); | |
48 | } | |
49 | ||
50 | int reserve_new_block(struct dnode_of_data *dn) | |
51 | { | |
52 | struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); | |
53 | ||
54 | if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)) | |
55 | return -EPERM; | |
56 | if (!inc_valid_block_count(sbi, dn->inode, 1)) | |
57 | return -ENOSPC; | |
58 | ||
c01e2853 NJ |
59 | trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node); |
60 | ||
eb47b800 JK |
61 | __set_data_blkaddr(dn, NEW_ADDR); |
62 | dn->data_blkaddr = NEW_ADDR; | |
63 | sync_inode_page(dn); | |
64 | return 0; | |
65 | } | |
66 | ||
b600965c HL |
67 | int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) |
68 | { | |
69 | bool need_put = dn->inode_page ? false : true; | |
70 | int err; | |
71 | ||
72 | err = get_dnode_of_data(dn, index, ALLOC_NODE); | |
73 | if (err) | |
74 | return err; | |
75 | if (dn->data_blkaddr == NULL_ADDR) | |
76 | err = reserve_new_block(dn); | |
77 | ||
78 | if (need_put) | |
79 | f2fs_put_dnode(dn); | |
80 | return err; | |
81 | } | |
82 | ||
eb47b800 JK |
83 | static int check_extent_cache(struct inode *inode, pgoff_t pgofs, |
84 | struct buffer_head *bh_result) | |
85 | { | |
86 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
eb47b800 JK |
87 | pgoff_t start_fofs, end_fofs; |
88 | block_t start_blkaddr; | |
89 | ||
c11abd1a JK |
90 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
91 | return 0; | |
92 | ||
eb47b800 JK |
93 | read_lock(&fi->ext.ext_lock); |
94 | if (fi->ext.len == 0) { | |
95 | read_unlock(&fi->ext.ext_lock); | |
96 | return 0; | |
97 | } | |
98 | ||
dcdfff65 JK |
99 | stat_inc_total_hit(inode->i_sb); |
100 | ||
eb47b800 JK |
101 | start_fofs = fi->ext.fofs; |
102 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
103 | start_blkaddr = fi->ext.blk_addr; | |
104 | ||
105 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
106 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
107 | size_t count; | |
108 | ||
109 | clear_buffer_new(bh_result); | |
110 | map_bh(bh_result, inode->i_sb, | |
111 | start_blkaddr + pgofs - start_fofs); | |
112 | count = end_fofs - pgofs + 1; | |
113 | if (count < (UINT_MAX >> blkbits)) | |
114 | bh_result->b_size = (count << blkbits); | |
115 | else | |
116 | bh_result->b_size = UINT_MAX; | |
117 | ||
dcdfff65 | 118 | stat_inc_read_hit(inode->i_sb); |
eb47b800 JK |
119 | read_unlock(&fi->ext.ext_lock); |
120 | return 1; | |
121 | } | |
122 | read_unlock(&fi->ext.ext_lock); | |
123 | return 0; | |
124 | } | |
125 | ||
126 | void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn) | |
127 | { | |
128 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
129 | pgoff_t fofs, start_fofs, end_fofs; | |
130 | block_t start_blkaddr, end_blkaddr; | |
c11abd1a | 131 | int need_update = true; |
eb47b800 | 132 | |
5d56b671 | 133 | f2fs_bug_on(blk_addr == NEW_ADDR); |
de93653f JK |
134 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + |
135 | dn->ofs_in_node; | |
eb47b800 JK |
136 | |
137 | /* Update the page address in the parent node */ | |
138 | __set_data_blkaddr(dn, blk_addr); | |
139 | ||
c11abd1a JK |
140 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
141 | return; | |
142 | ||
eb47b800 JK |
143 | write_lock(&fi->ext.ext_lock); |
144 | ||
145 | start_fofs = fi->ext.fofs; | |
146 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
147 | start_blkaddr = fi->ext.blk_addr; | |
148 | end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1; | |
149 | ||
150 | /* Drop and initialize the matched extent */ | |
151 | if (fi->ext.len == 1 && fofs == start_fofs) | |
152 | fi->ext.len = 0; | |
153 | ||
154 | /* Initial extent */ | |
155 | if (fi->ext.len == 0) { | |
156 | if (blk_addr != NULL_ADDR) { | |
157 | fi->ext.fofs = fofs; | |
158 | fi->ext.blk_addr = blk_addr; | |
159 | fi->ext.len = 1; | |
160 | } | |
161 | goto end_update; | |
162 | } | |
163 | ||
6224da87 | 164 | /* Front merge */ |
eb47b800 JK |
165 | if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) { |
166 | fi->ext.fofs--; | |
167 | fi->ext.blk_addr--; | |
168 | fi->ext.len++; | |
169 | goto end_update; | |
170 | } | |
171 | ||
172 | /* Back merge */ | |
173 | if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) { | |
174 | fi->ext.len++; | |
175 | goto end_update; | |
176 | } | |
177 | ||
178 | /* Split the existing extent */ | |
179 | if (fi->ext.len > 1 && | |
180 | fofs >= start_fofs && fofs <= end_fofs) { | |
181 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
182 | fi->ext.len = fofs - start_fofs; | |
183 | } else { | |
184 | fi->ext.fofs = fofs + 1; | |
185 | fi->ext.blk_addr = start_blkaddr + | |
186 | fofs - start_fofs + 1; | |
187 | fi->ext.len -= fofs - start_fofs + 1; | |
188 | } | |
c11abd1a JK |
189 | } else { |
190 | need_update = false; | |
eb47b800 | 191 | } |
eb47b800 | 192 | |
c11abd1a JK |
193 | /* Finally, if the extent is very fragmented, let's drop the cache. */ |
194 | if (fi->ext.len < F2FS_MIN_EXTENT_LEN) { | |
195 | fi->ext.len = 0; | |
196 | set_inode_flag(fi, FI_NO_EXTENT); | |
197 | need_update = true; | |
198 | } | |
eb47b800 JK |
199 | end_update: |
200 | write_unlock(&fi->ext.ext_lock); | |
c11abd1a JK |
201 | if (need_update) |
202 | sync_inode_page(dn); | |
203 | return; | |
eb47b800 JK |
204 | } |
205 | ||
c718379b | 206 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 JK |
207 | { |
208 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
209 | struct address_space *mapping = inode->i_mapping; | |
210 | struct dnode_of_data dn; | |
211 | struct page *page; | |
212 | int err; | |
213 | ||
214 | page = find_get_page(mapping, index); | |
215 | if (page && PageUptodate(page)) | |
216 | return page; | |
217 | f2fs_put_page(page, 0); | |
218 | ||
219 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 220 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
221 | if (err) |
222 | return ERR_PTR(err); | |
223 | f2fs_put_dnode(&dn); | |
224 | ||
225 | if (dn.data_blkaddr == NULL_ADDR) | |
226 | return ERR_PTR(-ENOENT); | |
227 | ||
228 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
229 | if (dn.data_blkaddr == NEW_ADDR) | |
230 | return ERR_PTR(-EINVAL); | |
231 | ||
6f85b352 | 232 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
eb47b800 JK |
233 | if (!page) |
234 | return ERR_PTR(-ENOMEM); | |
235 | ||
393ff91f JK |
236 | if (PageUptodate(page)) { |
237 | unlock_page(page); | |
238 | return page; | |
239 | } | |
240 | ||
c718379b JK |
241 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, |
242 | sync ? READ_SYNC : READA); | |
243 | if (sync) { | |
244 | wait_on_page_locked(page); | |
245 | if (!PageUptodate(page)) { | |
246 | f2fs_put_page(page, 0); | |
247 | return ERR_PTR(-EIO); | |
248 | } | |
eb47b800 | 249 | } |
eb47b800 JK |
250 | return page; |
251 | } | |
252 | ||
0a8165d7 | 253 | /* |
eb47b800 JK |
254 | * If it tries to access a hole, return an error. |
255 | * Because, the callers, functions in dir.c and GC, should be able to know | |
256 | * whether this page exists or not. | |
257 | */ | |
258 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
259 | { | |
260 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
261 | struct address_space *mapping = inode->i_mapping; | |
262 | struct dnode_of_data dn; | |
263 | struct page *page; | |
264 | int err; | |
265 | ||
650495de | 266 | repeat: |
6f85b352 | 267 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
650495de JK |
268 | if (!page) |
269 | return ERR_PTR(-ENOMEM); | |
270 | ||
eb47b800 | 271 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 272 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
273 | if (err) { |
274 | f2fs_put_page(page, 1); | |
eb47b800 | 275 | return ERR_PTR(err); |
650495de | 276 | } |
eb47b800 JK |
277 | f2fs_put_dnode(&dn); |
278 | ||
650495de JK |
279 | if (dn.data_blkaddr == NULL_ADDR) { |
280 | f2fs_put_page(page, 1); | |
eb47b800 | 281 | return ERR_PTR(-ENOENT); |
650495de | 282 | } |
eb47b800 JK |
283 | |
284 | if (PageUptodate(page)) | |
285 | return page; | |
286 | ||
d59ff4df JK |
287 | /* |
288 | * A new dentry page is allocated but not able to be written, since its | |
289 | * new inode page couldn't be allocated due to -ENOSPC. | |
290 | * In such the case, its blkaddr can be remained as NEW_ADDR. | |
291 | * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. | |
292 | */ | |
293 | if (dn.data_blkaddr == NEW_ADDR) { | |
294 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
295 | SetPageUptodate(page); | |
296 | return page; | |
297 | } | |
eb47b800 JK |
298 | |
299 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 300 | if (err) |
eb47b800 | 301 | return ERR_PTR(err); |
393ff91f JK |
302 | |
303 | lock_page(page); | |
304 | if (!PageUptodate(page)) { | |
305 | f2fs_put_page(page, 1); | |
306 | return ERR_PTR(-EIO); | |
eb47b800 | 307 | } |
afcb7ca0 JK |
308 | if (page->mapping != mapping) { |
309 | f2fs_put_page(page, 1); | |
310 | goto repeat; | |
eb47b800 JK |
311 | } |
312 | return page; | |
313 | } | |
314 | ||
0a8165d7 | 315 | /* |
eb47b800 JK |
316 | * Caller ensures that this data page is never allocated. |
317 | * A new zero-filled data page is allocated in the page cache. | |
39936837 JK |
318 | * |
319 | * Also, caller should grab and release a mutex by calling mutex_lock_op() and | |
320 | * mutex_unlock_op(). | |
44a83ff6 | 321 | * Note that, npage is set only by make_empty_dir. |
eb47b800 | 322 | */ |
64aa7ed9 JK |
323 | struct page *get_new_data_page(struct inode *inode, |
324 | struct page *npage, pgoff_t index, bool new_i_size) | |
eb47b800 JK |
325 | { |
326 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
327 | struct address_space *mapping = inode->i_mapping; | |
328 | struct page *page; | |
329 | struct dnode_of_data dn; | |
330 | int err; | |
331 | ||
64aa7ed9 | 332 | set_new_dnode(&dn, inode, npage, npage, 0); |
b600965c | 333 | err = f2fs_reserve_block(&dn, index); |
eb47b800 JK |
334 | if (err) |
335 | return ERR_PTR(err); | |
336 | ||
afcb7ca0 | 337 | repeat: |
eb47b800 JK |
338 | page = grab_cache_page(mapping, index); |
339 | if (!page) | |
340 | return ERR_PTR(-ENOMEM); | |
341 | ||
342 | if (PageUptodate(page)) | |
343 | return page; | |
344 | ||
345 | if (dn.data_blkaddr == NEW_ADDR) { | |
346 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 347 | SetPageUptodate(page); |
eb47b800 JK |
348 | } else { |
349 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 350 | if (err) |
eb47b800 | 351 | return ERR_PTR(err); |
393ff91f JK |
352 | lock_page(page); |
353 | if (!PageUptodate(page)) { | |
354 | f2fs_put_page(page, 1); | |
355 | return ERR_PTR(-EIO); | |
eb47b800 | 356 | } |
afcb7ca0 JK |
357 | if (page->mapping != mapping) { |
358 | f2fs_put_page(page, 1); | |
359 | goto repeat; | |
eb47b800 JK |
360 | } |
361 | } | |
eb47b800 JK |
362 | |
363 | if (new_i_size && | |
364 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
365 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
366 | /* Only the directory inode sets new_i_size */ |
367 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
368 | mark_inode_dirty_sync(inode); |
369 | } | |
370 | return page; | |
371 | } | |
372 | ||
373 | static void read_end_io(struct bio *bio, int err) | |
374 | { | |
375 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
376 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
377 | ||
378 | do { | |
379 | struct page *page = bvec->bv_page; | |
380 | ||
381 | if (--bvec >= bio->bi_io_vec) | |
382 | prefetchw(&bvec->bv_page->flags); | |
383 | ||
384 | if (uptodate) { | |
385 | SetPageUptodate(page); | |
386 | } else { | |
387 | ClearPageUptodate(page); | |
388 | SetPageError(page); | |
389 | } | |
390 | unlock_page(page); | |
391 | } while (bvec >= bio->bi_io_vec); | |
eb47b800 JK |
392 | bio_put(bio); |
393 | } | |
394 | ||
0a8165d7 | 395 | /* |
eb47b800 | 396 | * Fill the locked page with data located in the block address. |
393ff91f | 397 | * Return unlocked page. |
eb47b800 JK |
398 | */ |
399 | int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page, | |
400 | block_t blk_addr, int type) | |
401 | { | |
402 | struct block_device *bdev = sbi->sb->s_bdev; | |
eb47b800 JK |
403 | struct bio *bio; |
404 | ||
848753aa | 405 | trace_f2fs_readpage(page, blk_addr, type); |
eb47b800 | 406 | |
eb47b800 | 407 | /* Allocate a new bio */ |
3cd8a239 | 408 | bio = f2fs_bio_alloc(bdev, 1); |
eb47b800 JK |
409 | |
410 | /* Initialize the bio */ | |
3cd8a239 | 411 | bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); |
eb47b800 | 412 | bio->bi_end_io = read_end_io; |
3cd8a239 | 413 | |
eb47b800 | 414 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { |
eb47b800 | 415 | bio_put(bio); |
393ff91f | 416 | f2fs_put_page(page, 1); |
eb47b800 JK |
417 | return -EFAULT; |
418 | } | |
419 | ||
420 | submit_bio(type, bio); | |
eb47b800 JK |
421 | return 0; |
422 | } | |
423 | ||
924b720b CY |
424 | void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, int rw) |
425 | { | |
426 | struct f2fs_bio_info *io = &sbi->read_io; | |
427 | ||
428 | if (!io->bio) | |
429 | return; | |
430 | ||
d4d288bc CY |
431 | trace_f2fs_submit_read_bio(sbi->sb, rw, META, io->bio); |
432 | ||
924b720b CY |
433 | mutex_lock(&io->io_mutex); |
434 | if (io->bio) { | |
435 | submit_bio(rw, io->bio); | |
436 | io->bio = NULL; | |
437 | } | |
438 | mutex_unlock(&io->io_mutex); | |
439 | } | |
440 | ||
441 | void submit_read_page(struct f2fs_sb_info *sbi, struct page *page, | |
442 | block_t blk_addr, int rw) | |
443 | { | |
444 | struct block_device *bdev = sbi->sb->s_bdev; | |
445 | struct f2fs_bio_info *io = &sbi->read_io; | |
446 | int bio_blocks; | |
447 | ||
448 | verify_block_addr(sbi, blk_addr); | |
449 | ||
450 | mutex_lock(&io->io_mutex); | |
451 | ||
452 | if (io->bio && io->last_block_in_bio != blk_addr - 1) { | |
453 | submit_bio(rw, io->bio); | |
454 | io->bio = NULL; | |
455 | } | |
456 | alloc_new: | |
457 | if (io->bio == NULL) { | |
458 | bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); | |
459 | io->bio = f2fs_bio_alloc(bdev, bio_blocks); | |
460 | io->bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); | |
461 | io->bio->bi_end_io = read_end_io; | |
462 | } | |
463 | ||
464 | if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) < | |
465 | PAGE_CACHE_SIZE) { | |
466 | submit_bio(rw, io->bio); | |
467 | io->bio = NULL; | |
468 | goto alloc_new; | |
469 | } | |
470 | ||
471 | io->last_block_in_bio = blk_addr; | |
472 | ||
473 | mutex_unlock(&io->io_mutex); | |
87b8872d | 474 | trace_f2fs_submit_read_page(page, rw, META, blk_addr); |
924b720b CY |
475 | } |
476 | ||
0a8165d7 | 477 | /* |
eb47b800 JK |
478 | * This function should be used by the data read flow only where it |
479 | * does not check the "create" flag that indicates block allocation. | |
480 | * The reason for this special functionality is to exploit VFS readahead | |
481 | * mechanism. | |
482 | */ | |
483 | static int get_data_block_ro(struct inode *inode, sector_t iblock, | |
484 | struct buffer_head *bh_result, int create) | |
485 | { | |
486 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
487 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
488 | struct dnode_of_data dn; | |
489 | pgoff_t pgofs; | |
490 | int err; | |
491 | ||
492 | /* Get the page offset from the block offset(iblock) */ | |
493 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
494 | ||
848753aa NJ |
495 | if (check_extent_cache(inode, pgofs, bh_result)) { |
496 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); | |
eb47b800 | 497 | return 0; |
848753aa | 498 | } |
eb47b800 JK |
499 | |
500 | /* When reading holes, we need its node page */ | |
501 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 502 | err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA); |
848753aa NJ |
503 | if (err) { |
504 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
eb47b800 | 505 | return (err == -ENOENT) ? 0 : err; |
848753aa | 506 | } |
eb47b800 JK |
507 | |
508 | /* It does not support data allocation */ | |
5d56b671 | 509 | f2fs_bug_on(create); |
eb47b800 JK |
510 | |
511 | if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) { | |
512 | int i; | |
513 | unsigned int end_offset; | |
514 | ||
515 | end_offset = IS_INODE(dn.node_page) ? | |
de93653f | 516 | ADDRS_PER_INODE(F2FS_I(inode)) : |
eb47b800 JK |
517 | ADDRS_PER_BLOCK; |
518 | ||
519 | clear_buffer_new(bh_result); | |
520 | ||
521 | /* Give more consecutive addresses for the read ahead */ | |
522 | for (i = 0; i < end_offset - dn.ofs_in_node; i++) | |
523 | if (((datablock_addr(dn.node_page, | |
524 | dn.ofs_in_node + i)) | |
525 | != (dn.data_blkaddr + i)) || maxblocks == i) | |
526 | break; | |
527 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); | |
f9a4e6df | 528 | bh_result->b_size = (((size_t)i) << blkbits); |
eb47b800 JK |
529 | } |
530 | f2fs_put_dnode(&dn); | |
848753aa | 531 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); |
eb47b800 JK |
532 | return 0; |
533 | } | |
534 | ||
535 | static int f2fs_read_data_page(struct file *file, struct page *page) | |
536 | { | |
537 | return mpage_readpage(page, get_data_block_ro); | |
538 | } | |
539 | ||
540 | static int f2fs_read_data_pages(struct file *file, | |
541 | struct address_space *mapping, | |
542 | struct list_head *pages, unsigned nr_pages) | |
543 | { | |
544 | return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro); | |
545 | } | |
546 | ||
547 | int do_write_data_page(struct page *page) | |
548 | { | |
549 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
550 | block_t old_blk_addr, new_blk_addr; |
551 | struct dnode_of_data dn; | |
552 | int err = 0; | |
553 | ||
554 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 555 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
556 | if (err) |
557 | return err; | |
558 | ||
559 | old_blk_addr = dn.data_blkaddr; | |
560 | ||
561 | /* This page is already truncated */ | |
562 | if (old_blk_addr == NULL_ADDR) | |
563 | goto out_writepage; | |
564 | ||
565 | set_page_writeback(page); | |
566 | ||
567 | /* | |
568 | * If current allocation needs SSR, | |
569 | * it had better in-place writes for updated data. | |
570 | */ | |
b25958b6 HL |
571 | if (unlikely(old_blk_addr != NEW_ADDR && |
572 | !is_cold_data(page) && | |
573 | need_inplace_update(inode))) { | |
eb47b800 JK |
574 | rewrite_data_page(F2FS_SB(inode->i_sb), page, |
575 | old_blk_addr); | |
576 | } else { | |
577 | write_data_page(inode, page, &dn, | |
578 | old_blk_addr, &new_blk_addr); | |
579 | update_extent_cache(new_blk_addr, &dn); | |
eb47b800 JK |
580 | } |
581 | out_writepage: | |
582 | f2fs_put_dnode(&dn); | |
583 | return err; | |
584 | } | |
585 | ||
586 | static int f2fs_write_data_page(struct page *page, | |
587 | struct writeback_control *wbc) | |
588 | { | |
589 | struct inode *inode = page->mapping->host; | |
590 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
591 | loff_t i_size = i_size_read(inode); | |
592 | const pgoff_t end_index = ((unsigned long long) i_size) | |
593 | >> PAGE_CACHE_SHIFT; | |
594 | unsigned offset; | |
39936837 | 595 | bool need_balance_fs = false; |
eb47b800 JK |
596 | int err = 0; |
597 | ||
598 | if (page->index < end_index) | |
39936837 | 599 | goto write; |
eb47b800 JK |
600 | |
601 | /* | |
602 | * If the offset is out-of-range of file size, | |
603 | * this page does not have to be written to disk. | |
604 | */ | |
605 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
606 | if ((page->index >= end_index + 1) || !offset) { | |
607 | if (S_ISDIR(inode->i_mode)) { | |
608 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
609 | inode_dec_dirty_dents(inode); | |
610 | } | |
39936837 | 611 | goto out; |
eb47b800 JK |
612 | } |
613 | ||
614 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 JK |
615 | write: |
616 | if (sbi->por_doing) { | |
617 | err = AOP_WRITEPAGE_ACTIVATE; | |
eb47b800 | 618 | goto redirty_out; |
39936837 | 619 | } |
eb47b800 | 620 | |
39936837 | 621 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 JK |
622 | if (S_ISDIR(inode->i_mode)) { |
623 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
624 | inode_dec_dirty_dents(inode); | |
39936837 JK |
625 | err = do_write_data_page(page); |
626 | } else { | |
e479556b | 627 | f2fs_lock_op(sbi); |
39936837 | 628 | err = do_write_data_page(page); |
e479556b | 629 | f2fs_unlock_op(sbi); |
39936837 | 630 | need_balance_fs = true; |
eb47b800 | 631 | } |
39936837 JK |
632 | if (err == -ENOENT) |
633 | goto out; | |
634 | else if (err) | |
635 | goto redirty_out; | |
eb47b800 JK |
636 | |
637 | if (wbc->for_reclaim) | |
638 | f2fs_submit_bio(sbi, DATA, true); | |
639 | ||
eb47b800 | 640 | clear_cold_data(page); |
39936837 | 641 | out: |
eb47b800 | 642 | unlock_page(page); |
39936837 | 643 | if (need_balance_fs) |
eb47b800 JK |
644 | f2fs_balance_fs(sbi); |
645 | return 0; | |
646 | ||
eb47b800 JK |
647 | redirty_out: |
648 | wbc->pages_skipped++; | |
649 | set_page_dirty(page); | |
39936837 | 650 | return err; |
eb47b800 JK |
651 | } |
652 | ||
653 | #define MAX_DESIRED_PAGES_WP 4096 | |
654 | ||
fa9150a8 NJ |
655 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
656 | void *data) | |
657 | { | |
658 | struct address_space *mapping = data; | |
659 | int ret = mapping->a_ops->writepage(page, wbc); | |
660 | mapping_set_error(mapping, ret); | |
661 | return ret; | |
662 | } | |
663 | ||
25ca923b | 664 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
665 | struct writeback_control *wbc) |
666 | { | |
667 | struct inode *inode = mapping->host; | |
668 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
531ad7d5 | 669 | bool locked = false; |
eb47b800 JK |
670 | int ret; |
671 | long excess_nrtw = 0, desired_nrtw; | |
672 | ||
cfb185a1 | 673 | /* deal with chardevs and other special file */ |
674 | if (!mapping->a_ops->writepage) | |
675 | return 0; | |
676 | ||
eb47b800 JK |
677 | if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) { |
678 | desired_nrtw = MAX_DESIRED_PAGES_WP; | |
679 | excess_nrtw = desired_nrtw - wbc->nr_to_write; | |
680 | wbc->nr_to_write = desired_nrtw; | |
681 | } | |
682 | ||
531ad7d5 | 683 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 684 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
685 | locked = true; |
686 | } | |
fa9150a8 | 687 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 688 | if (locked) |
eb47b800 JK |
689 | mutex_unlock(&sbi->writepages); |
690 | f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL)); | |
691 | ||
692 | remove_dirty_dir_inode(inode); | |
693 | ||
694 | wbc->nr_to_write -= excess_nrtw; | |
695 | return ret; | |
696 | } | |
697 | ||
698 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, | |
699 | loff_t pos, unsigned len, unsigned flags, | |
700 | struct page **pagep, void **fsdata) | |
701 | { | |
702 | struct inode *inode = mapping->host; | |
703 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
704 | struct page *page; | |
705 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; | |
706 | struct dnode_of_data dn; | |
707 | int err = 0; | |
708 | ||
eb47b800 | 709 | f2fs_balance_fs(sbi); |
afcb7ca0 | 710 | repeat: |
eb47b800 JK |
711 | page = grab_cache_page_write_begin(mapping, index, flags); |
712 | if (!page) | |
713 | return -ENOMEM; | |
714 | *pagep = page; | |
715 | ||
e479556b | 716 | f2fs_lock_op(sbi); |
eb47b800 | 717 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
b600965c | 718 | err = f2fs_reserve_block(&dn, index); |
e479556b | 719 | f2fs_unlock_op(sbi); |
eb47b800 | 720 | |
b600965c HL |
721 | if (err) { |
722 | f2fs_put_page(page, 1); | |
723 | return err; | |
724 | } | |
725 | ||
eb47b800 JK |
726 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) |
727 | return 0; | |
728 | ||
729 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
730 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
731 | unsigned end = start + len; | |
732 | ||
733 | /* Reading beyond i_size is simple: memset to zero */ | |
734 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 735 | goto out; |
eb47b800 JK |
736 | } |
737 | ||
738 | if (dn.data_blkaddr == NEW_ADDR) { | |
739 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
740 | } else { | |
741 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 742 | if (err) |
eb47b800 | 743 | return err; |
393ff91f JK |
744 | lock_page(page); |
745 | if (!PageUptodate(page)) { | |
746 | f2fs_put_page(page, 1); | |
747 | return -EIO; | |
eb47b800 | 748 | } |
afcb7ca0 JK |
749 | if (page->mapping != mapping) { |
750 | f2fs_put_page(page, 1); | |
751 | goto repeat; | |
eb47b800 JK |
752 | } |
753 | } | |
393ff91f | 754 | out: |
eb47b800 JK |
755 | SetPageUptodate(page); |
756 | clear_cold_data(page); | |
757 | return 0; | |
758 | } | |
759 | ||
a1dd3c13 JK |
760 | static int f2fs_write_end(struct file *file, |
761 | struct address_space *mapping, | |
762 | loff_t pos, unsigned len, unsigned copied, | |
763 | struct page *page, void *fsdata) | |
764 | { | |
765 | struct inode *inode = page->mapping->host; | |
766 | ||
767 | SetPageUptodate(page); | |
768 | set_page_dirty(page); | |
769 | ||
770 | if (pos + copied > i_size_read(inode)) { | |
771 | i_size_write(inode, pos + copied); | |
772 | mark_inode_dirty(inode); | |
773 | update_inode_page(inode); | |
774 | } | |
775 | ||
75c3c8bc | 776 | f2fs_put_page(page, 1); |
a1dd3c13 JK |
777 | return copied; |
778 | } | |
779 | ||
eb47b800 JK |
780 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
781 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
782 | { | |
783 | struct file *file = iocb->ki_filp; | |
784 | struct inode *inode = file->f_mapping->host; | |
785 | ||
786 | if (rw == WRITE) | |
787 | return 0; | |
788 | ||
789 | /* Needs synchronization with the cleaner */ | |
790 | return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, | |
791 | get_data_block_ro); | |
792 | } | |
793 | ||
d47992f8 LC |
794 | static void f2fs_invalidate_data_page(struct page *page, unsigned int offset, |
795 | unsigned int length) | |
eb47b800 JK |
796 | { |
797 | struct inode *inode = page->mapping->host; | |
798 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
799 | if (S_ISDIR(inode->i_mode) && PageDirty(page)) { | |
800 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
801 | inode_dec_dirty_dents(inode); | |
802 | } | |
803 | ClearPagePrivate(page); | |
804 | } | |
805 | ||
806 | static int f2fs_release_data_page(struct page *page, gfp_t wait) | |
807 | { | |
808 | ClearPagePrivate(page); | |
c3850aa1 | 809 | return 1; |
eb47b800 JK |
810 | } |
811 | ||
812 | static int f2fs_set_data_page_dirty(struct page *page) | |
813 | { | |
814 | struct address_space *mapping = page->mapping; | |
815 | struct inode *inode = mapping->host; | |
816 | ||
26c6b887 JK |
817 | trace_f2fs_set_page_dirty(page, DATA); |
818 | ||
eb47b800 JK |
819 | SetPageUptodate(page); |
820 | if (!PageDirty(page)) { | |
821 | __set_page_dirty_nobuffers(page); | |
822 | set_dirty_dir_page(inode, page); | |
823 | return 1; | |
824 | } | |
825 | return 0; | |
826 | } | |
827 | ||
c01e54b7 JK |
828 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
829 | { | |
830 | return generic_block_bmap(mapping, block, get_data_block_ro); | |
831 | } | |
832 | ||
eb47b800 JK |
833 | const struct address_space_operations f2fs_dblock_aops = { |
834 | .readpage = f2fs_read_data_page, | |
835 | .readpages = f2fs_read_data_pages, | |
836 | .writepage = f2fs_write_data_page, | |
837 | .writepages = f2fs_write_data_pages, | |
838 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 839 | .write_end = f2fs_write_end, |
eb47b800 JK |
840 | .set_page_dirty = f2fs_set_data_page_dirty, |
841 | .invalidatepage = f2fs_invalidate_data_page, | |
842 | .releasepage = f2fs_release_data_page, | |
843 | .direct_IO = f2fs_direct_IO, | |
c01e54b7 | 844 | .bmap = f2fs_bmap, |
eb47b800 | 845 | }; |