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 | ||
67 | static int check_extent_cache(struct inode *inode, pgoff_t pgofs, | |
68 | struct buffer_head *bh_result) | |
69 | { | |
70 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
eb47b800 JK |
71 | pgoff_t start_fofs, end_fofs; |
72 | block_t start_blkaddr; | |
73 | ||
c11abd1a JK |
74 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
75 | return 0; | |
76 | ||
eb47b800 JK |
77 | read_lock(&fi->ext.ext_lock); |
78 | if (fi->ext.len == 0) { | |
79 | read_unlock(&fi->ext.ext_lock); | |
80 | return 0; | |
81 | } | |
82 | ||
dcdfff65 JK |
83 | stat_inc_total_hit(inode->i_sb); |
84 | ||
eb47b800 JK |
85 | start_fofs = fi->ext.fofs; |
86 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
87 | start_blkaddr = fi->ext.blk_addr; | |
88 | ||
89 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
90 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
91 | size_t count; | |
92 | ||
93 | clear_buffer_new(bh_result); | |
94 | map_bh(bh_result, inode->i_sb, | |
95 | start_blkaddr + pgofs - start_fofs); | |
96 | count = end_fofs - pgofs + 1; | |
97 | if (count < (UINT_MAX >> blkbits)) | |
98 | bh_result->b_size = (count << blkbits); | |
99 | else | |
100 | bh_result->b_size = UINT_MAX; | |
101 | ||
dcdfff65 | 102 | stat_inc_read_hit(inode->i_sb); |
eb47b800 JK |
103 | read_unlock(&fi->ext.ext_lock); |
104 | return 1; | |
105 | } | |
106 | read_unlock(&fi->ext.ext_lock); | |
107 | return 0; | |
108 | } | |
109 | ||
110 | void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn) | |
111 | { | |
112 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
113 | pgoff_t fofs, start_fofs, end_fofs; | |
114 | block_t start_blkaddr, end_blkaddr; | |
c11abd1a | 115 | int need_update = true; |
eb47b800 | 116 | |
5d56b671 | 117 | f2fs_bug_on(blk_addr == NEW_ADDR); |
de93653f JK |
118 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + |
119 | dn->ofs_in_node; | |
eb47b800 JK |
120 | |
121 | /* Update the page address in the parent node */ | |
122 | __set_data_blkaddr(dn, blk_addr); | |
123 | ||
c11abd1a JK |
124 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
125 | return; | |
126 | ||
eb47b800 JK |
127 | write_lock(&fi->ext.ext_lock); |
128 | ||
129 | start_fofs = fi->ext.fofs; | |
130 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
131 | start_blkaddr = fi->ext.blk_addr; | |
132 | end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1; | |
133 | ||
134 | /* Drop and initialize the matched extent */ | |
135 | if (fi->ext.len == 1 && fofs == start_fofs) | |
136 | fi->ext.len = 0; | |
137 | ||
138 | /* Initial extent */ | |
139 | if (fi->ext.len == 0) { | |
140 | if (blk_addr != NULL_ADDR) { | |
141 | fi->ext.fofs = fofs; | |
142 | fi->ext.blk_addr = blk_addr; | |
143 | fi->ext.len = 1; | |
144 | } | |
145 | goto end_update; | |
146 | } | |
147 | ||
6224da87 | 148 | /* Front merge */ |
eb47b800 JK |
149 | if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) { |
150 | fi->ext.fofs--; | |
151 | fi->ext.blk_addr--; | |
152 | fi->ext.len++; | |
153 | goto end_update; | |
154 | } | |
155 | ||
156 | /* Back merge */ | |
157 | if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) { | |
158 | fi->ext.len++; | |
159 | goto end_update; | |
160 | } | |
161 | ||
162 | /* Split the existing extent */ | |
163 | if (fi->ext.len > 1 && | |
164 | fofs >= start_fofs && fofs <= end_fofs) { | |
165 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
166 | fi->ext.len = fofs - start_fofs; | |
167 | } else { | |
168 | fi->ext.fofs = fofs + 1; | |
169 | fi->ext.blk_addr = start_blkaddr + | |
170 | fofs - start_fofs + 1; | |
171 | fi->ext.len -= fofs - start_fofs + 1; | |
172 | } | |
c11abd1a JK |
173 | } else { |
174 | need_update = false; | |
eb47b800 | 175 | } |
eb47b800 | 176 | |
c11abd1a JK |
177 | /* Finally, if the extent is very fragmented, let's drop the cache. */ |
178 | if (fi->ext.len < F2FS_MIN_EXTENT_LEN) { | |
179 | fi->ext.len = 0; | |
180 | set_inode_flag(fi, FI_NO_EXTENT); | |
181 | need_update = true; | |
182 | } | |
eb47b800 JK |
183 | end_update: |
184 | write_unlock(&fi->ext.ext_lock); | |
c11abd1a JK |
185 | if (need_update) |
186 | sync_inode_page(dn); | |
187 | return; | |
eb47b800 JK |
188 | } |
189 | ||
c718379b | 190 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 JK |
191 | { |
192 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
193 | struct address_space *mapping = inode->i_mapping; | |
194 | struct dnode_of_data dn; | |
195 | struct page *page; | |
196 | int err; | |
197 | ||
198 | page = find_get_page(mapping, index); | |
199 | if (page && PageUptodate(page)) | |
200 | return page; | |
201 | f2fs_put_page(page, 0); | |
202 | ||
203 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 204 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
205 | if (err) |
206 | return ERR_PTR(err); | |
207 | f2fs_put_dnode(&dn); | |
208 | ||
209 | if (dn.data_blkaddr == NULL_ADDR) | |
210 | return ERR_PTR(-ENOENT); | |
211 | ||
212 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
213 | if (dn.data_blkaddr == NEW_ADDR) | |
214 | return ERR_PTR(-EINVAL); | |
215 | ||
6f85b352 | 216 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
eb47b800 JK |
217 | if (!page) |
218 | return ERR_PTR(-ENOMEM); | |
219 | ||
393ff91f JK |
220 | if (PageUptodate(page)) { |
221 | unlock_page(page); | |
222 | return page; | |
223 | } | |
224 | ||
c718379b JK |
225 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, |
226 | sync ? READ_SYNC : READA); | |
227 | if (sync) { | |
228 | wait_on_page_locked(page); | |
229 | if (!PageUptodate(page)) { | |
230 | f2fs_put_page(page, 0); | |
231 | return ERR_PTR(-EIO); | |
232 | } | |
eb47b800 | 233 | } |
eb47b800 JK |
234 | return page; |
235 | } | |
236 | ||
0a8165d7 | 237 | /* |
eb47b800 JK |
238 | * If it tries to access a hole, return an error. |
239 | * Because, the callers, functions in dir.c and GC, should be able to know | |
240 | * whether this page exists or not. | |
241 | */ | |
242 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
243 | { | |
244 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
245 | struct address_space *mapping = inode->i_mapping; | |
246 | struct dnode_of_data dn; | |
247 | struct page *page; | |
248 | int err; | |
249 | ||
650495de | 250 | repeat: |
6f85b352 | 251 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
650495de JK |
252 | if (!page) |
253 | return ERR_PTR(-ENOMEM); | |
254 | ||
eb47b800 | 255 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 256 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
257 | if (err) { |
258 | f2fs_put_page(page, 1); | |
eb47b800 | 259 | return ERR_PTR(err); |
650495de | 260 | } |
eb47b800 JK |
261 | f2fs_put_dnode(&dn); |
262 | ||
650495de JK |
263 | if (dn.data_blkaddr == NULL_ADDR) { |
264 | f2fs_put_page(page, 1); | |
eb47b800 | 265 | return ERR_PTR(-ENOENT); |
650495de | 266 | } |
eb47b800 JK |
267 | |
268 | if (PageUptodate(page)) | |
269 | return page; | |
270 | ||
d59ff4df JK |
271 | /* |
272 | * A new dentry page is allocated but not able to be written, since its | |
273 | * new inode page couldn't be allocated due to -ENOSPC. | |
274 | * In such the case, its blkaddr can be remained as NEW_ADDR. | |
275 | * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. | |
276 | */ | |
277 | if (dn.data_blkaddr == NEW_ADDR) { | |
278 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
279 | SetPageUptodate(page); | |
280 | return page; | |
281 | } | |
eb47b800 JK |
282 | |
283 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 284 | if (err) |
eb47b800 | 285 | return ERR_PTR(err); |
393ff91f JK |
286 | |
287 | lock_page(page); | |
288 | if (!PageUptodate(page)) { | |
289 | f2fs_put_page(page, 1); | |
290 | return ERR_PTR(-EIO); | |
eb47b800 | 291 | } |
afcb7ca0 JK |
292 | if (page->mapping != mapping) { |
293 | f2fs_put_page(page, 1); | |
294 | goto repeat; | |
eb47b800 JK |
295 | } |
296 | return page; | |
297 | } | |
298 | ||
0a8165d7 | 299 | /* |
eb47b800 JK |
300 | * Caller ensures that this data page is never allocated. |
301 | * A new zero-filled data page is allocated in the page cache. | |
39936837 JK |
302 | * |
303 | * Also, caller should grab and release a mutex by calling mutex_lock_op() and | |
304 | * mutex_unlock_op(). | |
44a83ff6 | 305 | * Note that, npage is set only by make_empty_dir. |
eb47b800 | 306 | */ |
64aa7ed9 JK |
307 | struct page *get_new_data_page(struct inode *inode, |
308 | struct page *npage, pgoff_t index, bool new_i_size) | |
eb47b800 JK |
309 | { |
310 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
311 | struct address_space *mapping = inode->i_mapping; | |
312 | struct page *page; | |
313 | struct dnode_of_data dn; | |
314 | int err; | |
315 | ||
64aa7ed9 | 316 | set_new_dnode(&dn, inode, npage, npage, 0); |
266e97a8 | 317 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
eb47b800 JK |
318 | if (err) |
319 | return ERR_PTR(err); | |
320 | ||
321 | if (dn.data_blkaddr == NULL_ADDR) { | |
322 | if (reserve_new_block(&dn)) { | |
64aa7ed9 JK |
323 | if (!npage) |
324 | f2fs_put_dnode(&dn); | |
eb47b800 JK |
325 | return ERR_PTR(-ENOSPC); |
326 | } | |
327 | } | |
64aa7ed9 JK |
328 | if (!npage) |
329 | f2fs_put_dnode(&dn); | |
afcb7ca0 | 330 | repeat: |
eb47b800 JK |
331 | page = grab_cache_page(mapping, index); |
332 | if (!page) | |
333 | return ERR_PTR(-ENOMEM); | |
334 | ||
335 | if (PageUptodate(page)) | |
336 | return page; | |
337 | ||
338 | if (dn.data_blkaddr == NEW_ADDR) { | |
339 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 340 | SetPageUptodate(page); |
eb47b800 JK |
341 | } else { |
342 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 343 | if (err) |
eb47b800 | 344 | return ERR_PTR(err); |
393ff91f JK |
345 | lock_page(page); |
346 | if (!PageUptodate(page)) { | |
347 | f2fs_put_page(page, 1); | |
348 | return ERR_PTR(-EIO); | |
eb47b800 | 349 | } |
afcb7ca0 JK |
350 | if (page->mapping != mapping) { |
351 | f2fs_put_page(page, 1); | |
352 | goto repeat; | |
eb47b800 JK |
353 | } |
354 | } | |
eb47b800 JK |
355 | |
356 | if (new_i_size && | |
357 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
358 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
359 | /* Only the directory inode sets new_i_size */ |
360 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
361 | mark_inode_dirty_sync(inode); |
362 | } | |
363 | return page; | |
364 | } | |
365 | ||
366 | static void read_end_io(struct bio *bio, int err) | |
367 | { | |
368 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
369 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
370 | ||
371 | do { | |
372 | struct page *page = bvec->bv_page; | |
373 | ||
374 | if (--bvec >= bio->bi_io_vec) | |
375 | prefetchw(&bvec->bv_page->flags); | |
376 | ||
377 | if (uptodate) { | |
378 | SetPageUptodate(page); | |
379 | } else { | |
380 | ClearPageUptodate(page); | |
381 | SetPageError(page); | |
382 | } | |
383 | unlock_page(page); | |
384 | } while (bvec >= bio->bi_io_vec); | |
eb47b800 JK |
385 | bio_put(bio); |
386 | } | |
387 | ||
0a8165d7 | 388 | /* |
eb47b800 | 389 | * Fill the locked page with data located in the block address. |
393ff91f | 390 | * Return unlocked page. |
eb47b800 JK |
391 | */ |
392 | int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page, | |
393 | block_t blk_addr, int type) | |
394 | { | |
395 | struct block_device *bdev = sbi->sb->s_bdev; | |
eb47b800 JK |
396 | struct bio *bio; |
397 | ||
848753aa | 398 | trace_f2fs_readpage(page, blk_addr, type); |
eb47b800 | 399 | |
eb47b800 | 400 | /* Allocate a new bio */ |
3cd8a239 | 401 | bio = f2fs_bio_alloc(bdev, 1); |
eb47b800 JK |
402 | |
403 | /* Initialize the bio */ | |
3cd8a239 | 404 | bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); |
eb47b800 | 405 | bio->bi_end_io = read_end_io; |
3cd8a239 | 406 | |
eb47b800 | 407 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { |
eb47b800 | 408 | bio_put(bio); |
393ff91f | 409 | f2fs_put_page(page, 1); |
eb47b800 JK |
410 | return -EFAULT; |
411 | } | |
412 | ||
413 | submit_bio(type, bio); | |
eb47b800 JK |
414 | return 0; |
415 | } | |
416 | ||
924b720b CY |
417 | void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, int rw) |
418 | { | |
419 | struct f2fs_bio_info *io = &sbi->read_io; | |
420 | ||
421 | if (!io->bio) | |
422 | return; | |
423 | ||
d4d288bc CY |
424 | trace_f2fs_submit_read_bio(sbi->sb, rw, META, io->bio); |
425 | ||
924b720b CY |
426 | mutex_lock(&io->io_mutex); |
427 | if (io->bio) { | |
428 | submit_bio(rw, io->bio); | |
429 | io->bio = NULL; | |
430 | } | |
431 | mutex_unlock(&io->io_mutex); | |
432 | } | |
433 | ||
434 | void submit_read_page(struct f2fs_sb_info *sbi, struct page *page, | |
435 | block_t blk_addr, int rw) | |
436 | { | |
437 | struct block_device *bdev = sbi->sb->s_bdev; | |
438 | struct f2fs_bio_info *io = &sbi->read_io; | |
439 | int bio_blocks; | |
440 | ||
441 | verify_block_addr(sbi, blk_addr); | |
442 | ||
443 | mutex_lock(&io->io_mutex); | |
444 | ||
445 | if (io->bio && io->last_block_in_bio != blk_addr - 1) { | |
446 | submit_bio(rw, io->bio); | |
447 | io->bio = NULL; | |
448 | } | |
449 | alloc_new: | |
450 | if (io->bio == NULL) { | |
451 | bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); | |
452 | io->bio = f2fs_bio_alloc(bdev, bio_blocks); | |
453 | io->bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); | |
454 | io->bio->bi_end_io = read_end_io; | |
455 | } | |
456 | ||
457 | if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) < | |
458 | PAGE_CACHE_SIZE) { | |
459 | submit_bio(rw, io->bio); | |
460 | io->bio = NULL; | |
461 | goto alloc_new; | |
462 | } | |
463 | ||
464 | io->last_block_in_bio = blk_addr; | |
465 | ||
466 | mutex_unlock(&io->io_mutex); | |
87b8872d | 467 | trace_f2fs_submit_read_page(page, rw, META, blk_addr); |
924b720b CY |
468 | } |
469 | ||
0a8165d7 | 470 | /* |
eb47b800 JK |
471 | * This function should be used by the data read flow only where it |
472 | * does not check the "create" flag that indicates block allocation. | |
473 | * The reason for this special functionality is to exploit VFS readahead | |
474 | * mechanism. | |
475 | */ | |
476 | static int get_data_block_ro(struct inode *inode, sector_t iblock, | |
477 | struct buffer_head *bh_result, int create) | |
478 | { | |
479 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
480 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
481 | struct dnode_of_data dn; | |
482 | pgoff_t pgofs; | |
483 | int err; | |
484 | ||
485 | /* Get the page offset from the block offset(iblock) */ | |
486 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
487 | ||
848753aa NJ |
488 | if (check_extent_cache(inode, pgofs, bh_result)) { |
489 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); | |
eb47b800 | 490 | return 0; |
848753aa | 491 | } |
eb47b800 JK |
492 | |
493 | /* When reading holes, we need its node page */ | |
494 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 495 | err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA); |
848753aa NJ |
496 | if (err) { |
497 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
eb47b800 | 498 | return (err == -ENOENT) ? 0 : err; |
848753aa | 499 | } |
eb47b800 JK |
500 | |
501 | /* It does not support data allocation */ | |
5d56b671 | 502 | f2fs_bug_on(create); |
eb47b800 JK |
503 | |
504 | if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) { | |
505 | int i; | |
506 | unsigned int end_offset; | |
507 | ||
508 | end_offset = IS_INODE(dn.node_page) ? | |
de93653f | 509 | ADDRS_PER_INODE(F2FS_I(inode)) : |
eb47b800 JK |
510 | ADDRS_PER_BLOCK; |
511 | ||
512 | clear_buffer_new(bh_result); | |
513 | ||
514 | /* Give more consecutive addresses for the read ahead */ | |
515 | for (i = 0; i < end_offset - dn.ofs_in_node; i++) | |
516 | if (((datablock_addr(dn.node_page, | |
517 | dn.ofs_in_node + i)) | |
518 | != (dn.data_blkaddr + i)) || maxblocks == i) | |
519 | break; | |
520 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); | |
521 | bh_result->b_size = (i << blkbits); | |
522 | } | |
523 | f2fs_put_dnode(&dn); | |
848753aa | 524 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); |
eb47b800 JK |
525 | return 0; |
526 | } | |
527 | ||
528 | static int f2fs_read_data_page(struct file *file, struct page *page) | |
529 | { | |
530 | return mpage_readpage(page, get_data_block_ro); | |
531 | } | |
532 | ||
533 | static int f2fs_read_data_pages(struct file *file, | |
534 | struct address_space *mapping, | |
535 | struct list_head *pages, unsigned nr_pages) | |
536 | { | |
537 | return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro); | |
538 | } | |
539 | ||
540 | int do_write_data_page(struct page *page) | |
541 | { | |
542 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
543 | block_t old_blk_addr, new_blk_addr; |
544 | struct dnode_of_data dn; | |
545 | int err = 0; | |
546 | ||
547 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 548 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
549 | if (err) |
550 | return err; | |
551 | ||
552 | old_blk_addr = dn.data_blkaddr; | |
553 | ||
554 | /* This page is already truncated */ | |
555 | if (old_blk_addr == NULL_ADDR) | |
556 | goto out_writepage; | |
557 | ||
558 | set_page_writeback(page); | |
559 | ||
560 | /* | |
561 | * If current allocation needs SSR, | |
562 | * it had better in-place writes for updated data. | |
563 | */ | |
b25958b6 HL |
564 | if (unlikely(old_blk_addr != NEW_ADDR && |
565 | !is_cold_data(page) && | |
566 | need_inplace_update(inode))) { | |
eb47b800 JK |
567 | rewrite_data_page(F2FS_SB(inode->i_sb), page, |
568 | old_blk_addr); | |
569 | } else { | |
570 | write_data_page(inode, page, &dn, | |
571 | old_blk_addr, &new_blk_addr); | |
572 | update_extent_cache(new_blk_addr, &dn); | |
eb47b800 JK |
573 | } |
574 | out_writepage: | |
575 | f2fs_put_dnode(&dn); | |
576 | return err; | |
577 | } | |
578 | ||
579 | static int f2fs_write_data_page(struct page *page, | |
580 | struct writeback_control *wbc) | |
581 | { | |
582 | struct inode *inode = page->mapping->host; | |
583 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
584 | loff_t i_size = i_size_read(inode); | |
585 | const pgoff_t end_index = ((unsigned long long) i_size) | |
586 | >> PAGE_CACHE_SHIFT; | |
587 | unsigned offset; | |
39936837 | 588 | bool need_balance_fs = false; |
eb47b800 JK |
589 | int err = 0; |
590 | ||
591 | if (page->index < end_index) | |
39936837 | 592 | goto write; |
eb47b800 JK |
593 | |
594 | /* | |
595 | * If the offset is out-of-range of file size, | |
596 | * this page does not have to be written to disk. | |
597 | */ | |
598 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
599 | if ((page->index >= end_index + 1) || !offset) { | |
600 | if (S_ISDIR(inode->i_mode)) { | |
601 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
602 | inode_dec_dirty_dents(inode); | |
603 | } | |
39936837 | 604 | goto out; |
eb47b800 JK |
605 | } |
606 | ||
607 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 JK |
608 | write: |
609 | if (sbi->por_doing) { | |
610 | err = AOP_WRITEPAGE_ACTIVATE; | |
eb47b800 | 611 | goto redirty_out; |
39936837 | 612 | } |
eb47b800 | 613 | |
39936837 | 614 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 JK |
615 | if (S_ISDIR(inode->i_mode)) { |
616 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
617 | inode_dec_dirty_dents(inode); | |
39936837 JK |
618 | err = do_write_data_page(page); |
619 | } else { | |
e479556b | 620 | f2fs_lock_op(sbi); |
39936837 | 621 | err = do_write_data_page(page); |
e479556b | 622 | f2fs_unlock_op(sbi); |
39936837 | 623 | need_balance_fs = true; |
eb47b800 | 624 | } |
39936837 JK |
625 | if (err == -ENOENT) |
626 | goto out; | |
627 | else if (err) | |
628 | goto redirty_out; | |
eb47b800 JK |
629 | |
630 | if (wbc->for_reclaim) | |
631 | f2fs_submit_bio(sbi, DATA, true); | |
632 | ||
eb47b800 | 633 | clear_cold_data(page); |
39936837 | 634 | out: |
eb47b800 | 635 | unlock_page(page); |
39936837 | 636 | if (need_balance_fs) |
eb47b800 JK |
637 | f2fs_balance_fs(sbi); |
638 | return 0; | |
639 | ||
eb47b800 JK |
640 | redirty_out: |
641 | wbc->pages_skipped++; | |
642 | set_page_dirty(page); | |
39936837 | 643 | return err; |
eb47b800 JK |
644 | } |
645 | ||
646 | #define MAX_DESIRED_PAGES_WP 4096 | |
647 | ||
fa9150a8 NJ |
648 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
649 | void *data) | |
650 | { | |
651 | struct address_space *mapping = data; | |
652 | int ret = mapping->a_ops->writepage(page, wbc); | |
653 | mapping_set_error(mapping, ret); | |
654 | return ret; | |
655 | } | |
656 | ||
25ca923b | 657 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
658 | struct writeback_control *wbc) |
659 | { | |
660 | struct inode *inode = mapping->host; | |
661 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
531ad7d5 | 662 | bool locked = false; |
eb47b800 JK |
663 | int ret; |
664 | long excess_nrtw = 0, desired_nrtw; | |
665 | ||
cfb185a1 | 666 | /* deal with chardevs and other special file */ |
667 | if (!mapping->a_ops->writepage) | |
668 | return 0; | |
669 | ||
eb47b800 JK |
670 | if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) { |
671 | desired_nrtw = MAX_DESIRED_PAGES_WP; | |
672 | excess_nrtw = desired_nrtw - wbc->nr_to_write; | |
673 | wbc->nr_to_write = desired_nrtw; | |
674 | } | |
675 | ||
531ad7d5 | 676 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 677 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
678 | locked = true; |
679 | } | |
fa9150a8 | 680 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 681 | if (locked) |
eb47b800 JK |
682 | mutex_unlock(&sbi->writepages); |
683 | f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL)); | |
684 | ||
685 | remove_dirty_dir_inode(inode); | |
686 | ||
687 | wbc->nr_to_write -= excess_nrtw; | |
688 | return ret; | |
689 | } | |
690 | ||
691 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, | |
692 | loff_t pos, unsigned len, unsigned flags, | |
693 | struct page **pagep, void **fsdata) | |
694 | { | |
695 | struct inode *inode = mapping->host; | |
696 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
697 | struct page *page; | |
698 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; | |
699 | struct dnode_of_data dn; | |
700 | int err = 0; | |
701 | ||
eb47b800 | 702 | f2fs_balance_fs(sbi); |
afcb7ca0 | 703 | repeat: |
eb47b800 JK |
704 | page = grab_cache_page_write_begin(mapping, index, flags); |
705 | if (!page) | |
706 | return -ENOMEM; | |
707 | *pagep = page; | |
708 | ||
e479556b | 709 | f2fs_lock_op(sbi); |
eb47b800 JK |
710 | |
711 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 712 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
39936837 JK |
713 | if (err) |
714 | goto err; | |
eb47b800 | 715 | |
39936837 | 716 | if (dn.data_blkaddr == NULL_ADDR) |
eb47b800 | 717 | err = reserve_new_block(&dn); |
39936837 | 718 | |
eb47b800 | 719 | f2fs_put_dnode(&dn); |
39936837 JK |
720 | if (err) |
721 | goto err; | |
eb47b800 | 722 | |
e479556b | 723 | f2fs_unlock_op(sbi); |
eb47b800 JK |
724 | |
725 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) | |
726 | return 0; | |
727 | ||
728 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
729 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
730 | unsigned end = start + len; | |
731 | ||
732 | /* Reading beyond i_size is simple: memset to zero */ | |
733 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 734 | goto out; |
eb47b800 JK |
735 | } |
736 | ||
737 | if (dn.data_blkaddr == NEW_ADDR) { | |
738 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
739 | } else { | |
740 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 741 | if (err) |
eb47b800 | 742 | return err; |
393ff91f JK |
743 | lock_page(page); |
744 | if (!PageUptodate(page)) { | |
745 | f2fs_put_page(page, 1); | |
746 | return -EIO; | |
eb47b800 | 747 | } |
afcb7ca0 JK |
748 | if (page->mapping != mapping) { |
749 | f2fs_put_page(page, 1); | |
750 | goto repeat; | |
eb47b800 JK |
751 | } |
752 | } | |
393ff91f | 753 | out: |
eb47b800 JK |
754 | SetPageUptodate(page); |
755 | clear_cold_data(page); | |
756 | return 0; | |
39936837 JK |
757 | |
758 | err: | |
e479556b | 759 | f2fs_unlock_op(sbi); |
39936837 JK |
760 | f2fs_put_page(page, 1); |
761 | return err; | |
eb47b800 JK |
762 | } |
763 | ||
a1dd3c13 JK |
764 | static int f2fs_write_end(struct file *file, |
765 | struct address_space *mapping, | |
766 | loff_t pos, unsigned len, unsigned copied, | |
767 | struct page *page, void *fsdata) | |
768 | { | |
769 | struct inode *inode = page->mapping->host; | |
770 | ||
771 | SetPageUptodate(page); | |
772 | set_page_dirty(page); | |
773 | ||
774 | if (pos + copied > i_size_read(inode)) { | |
775 | i_size_write(inode, pos + copied); | |
776 | mark_inode_dirty(inode); | |
777 | update_inode_page(inode); | |
778 | } | |
779 | ||
75c3c8bc | 780 | f2fs_put_page(page, 1); |
a1dd3c13 JK |
781 | return copied; |
782 | } | |
783 | ||
eb47b800 JK |
784 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
785 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
786 | { | |
787 | struct file *file = iocb->ki_filp; | |
788 | struct inode *inode = file->f_mapping->host; | |
789 | ||
790 | if (rw == WRITE) | |
791 | return 0; | |
792 | ||
793 | /* Needs synchronization with the cleaner */ | |
794 | return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, | |
795 | get_data_block_ro); | |
796 | } | |
797 | ||
d47992f8 LC |
798 | static void f2fs_invalidate_data_page(struct page *page, unsigned int offset, |
799 | unsigned int length) | |
eb47b800 JK |
800 | { |
801 | struct inode *inode = page->mapping->host; | |
802 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
803 | if (S_ISDIR(inode->i_mode) && PageDirty(page)) { | |
804 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
805 | inode_dec_dirty_dents(inode); | |
806 | } | |
807 | ClearPagePrivate(page); | |
808 | } | |
809 | ||
810 | static int f2fs_release_data_page(struct page *page, gfp_t wait) | |
811 | { | |
812 | ClearPagePrivate(page); | |
c3850aa1 | 813 | return 1; |
eb47b800 JK |
814 | } |
815 | ||
816 | static int f2fs_set_data_page_dirty(struct page *page) | |
817 | { | |
818 | struct address_space *mapping = page->mapping; | |
819 | struct inode *inode = mapping->host; | |
820 | ||
26c6b887 JK |
821 | trace_f2fs_set_page_dirty(page, DATA); |
822 | ||
eb47b800 JK |
823 | SetPageUptodate(page); |
824 | if (!PageDirty(page)) { | |
825 | __set_page_dirty_nobuffers(page); | |
826 | set_dirty_dir_page(inode, page); | |
827 | return 1; | |
828 | } | |
829 | return 0; | |
830 | } | |
831 | ||
c01e54b7 JK |
832 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
833 | { | |
834 | return generic_block_bmap(mapping, block, get_data_block_ro); | |
835 | } | |
836 | ||
eb47b800 JK |
837 | const struct address_space_operations f2fs_dblock_aops = { |
838 | .readpage = f2fs_read_data_page, | |
839 | .readpages = f2fs_read_data_pages, | |
840 | .writepage = f2fs_write_data_page, | |
841 | .writepages = f2fs_write_data_pages, | |
842 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 843 | .write_end = f2fs_write_end, |
eb47b800 JK |
844 | .set_page_dirty = f2fs_set_data_page_dirty, |
845 | .invalidatepage = f2fs_invalidate_data_page, | |
846 | .releasepage = f2fs_release_data_page, | |
847 | .direct_IO = f2fs_direct_IO, | |
c01e54b7 | 848 | .bmap = f2fs_bmap, |
eb47b800 | 849 | }; |