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); | |
35b09d82 | 71 | #ifdef CONFIG_F2FS_STAT_FS |
eb47b800 | 72 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); |
35b09d82 | 73 | #endif |
eb47b800 JK |
74 | pgoff_t start_fofs, end_fofs; |
75 | block_t start_blkaddr; | |
76 | ||
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 | ||
35b09d82 | 83 | #ifdef CONFIG_F2FS_STAT_FS |
eb47b800 | 84 | sbi->total_hit_ext++; |
35b09d82 | 85 | #endif |
eb47b800 JK |
86 | start_fofs = fi->ext.fofs; |
87 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
88 | start_blkaddr = fi->ext.blk_addr; | |
89 | ||
90 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
91 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
92 | size_t count; | |
93 | ||
94 | clear_buffer_new(bh_result); | |
95 | map_bh(bh_result, inode->i_sb, | |
96 | start_blkaddr + pgofs - start_fofs); | |
97 | count = end_fofs - pgofs + 1; | |
98 | if (count < (UINT_MAX >> blkbits)) | |
99 | bh_result->b_size = (count << blkbits); | |
100 | else | |
101 | bh_result->b_size = UINT_MAX; | |
102 | ||
35b09d82 | 103 | #ifdef CONFIG_F2FS_STAT_FS |
eb47b800 | 104 | sbi->read_hit_ext++; |
35b09d82 | 105 | #endif |
eb47b800 JK |
106 | read_unlock(&fi->ext.ext_lock); |
107 | return 1; | |
108 | } | |
109 | read_unlock(&fi->ext.ext_lock); | |
110 | return 0; | |
111 | } | |
112 | ||
113 | void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn) | |
114 | { | |
115 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
116 | pgoff_t fofs, start_fofs, end_fofs; | |
117 | block_t start_blkaddr, end_blkaddr; | |
118 | ||
119 | BUG_ON(blk_addr == NEW_ADDR); | |
120 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node; | |
121 | ||
122 | /* Update the page address in the parent node */ | |
123 | __set_data_blkaddr(dn, blk_addr); | |
124 | ||
125 | write_lock(&fi->ext.ext_lock); | |
126 | ||
127 | start_fofs = fi->ext.fofs; | |
128 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
129 | start_blkaddr = fi->ext.blk_addr; | |
130 | end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1; | |
131 | ||
132 | /* Drop and initialize the matched extent */ | |
133 | if (fi->ext.len == 1 && fofs == start_fofs) | |
134 | fi->ext.len = 0; | |
135 | ||
136 | /* Initial extent */ | |
137 | if (fi->ext.len == 0) { | |
138 | if (blk_addr != NULL_ADDR) { | |
139 | fi->ext.fofs = fofs; | |
140 | fi->ext.blk_addr = blk_addr; | |
141 | fi->ext.len = 1; | |
142 | } | |
143 | goto end_update; | |
144 | } | |
145 | ||
6224da87 | 146 | /* Front merge */ |
eb47b800 JK |
147 | if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) { |
148 | fi->ext.fofs--; | |
149 | fi->ext.blk_addr--; | |
150 | fi->ext.len++; | |
151 | goto end_update; | |
152 | } | |
153 | ||
154 | /* Back merge */ | |
155 | if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) { | |
156 | fi->ext.len++; | |
157 | goto end_update; | |
158 | } | |
159 | ||
160 | /* Split the existing extent */ | |
161 | if (fi->ext.len > 1 && | |
162 | fofs >= start_fofs && fofs <= end_fofs) { | |
163 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
164 | fi->ext.len = fofs - start_fofs; | |
165 | } else { | |
166 | fi->ext.fofs = fofs + 1; | |
167 | fi->ext.blk_addr = start_blkaddr + | |
168 | fofs - start_fofs + 1; | |
169 | fi->ext.len -= fofs - start_fofs + 1; | |
170 | } | |
171 | goto end_update; | |
172 | } | |
173 | write_unlock(&fi->ext.ext_lock); | |
174 | return; | |
175 | ||
176 | end_update: | |
177 | write_unlock(&fi->ext.ext_lock); | |
178 | sync_inode_page(dn); | |
eb47b800 JK |
179 | } |
180 | ||
c718379b | 181 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 JK |
182 | { |
183 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
184 | struct address_space *mapping = inode->i_mapping; | |
185 | struct dnode_of_data dn; | |
186 | struct page *page; | |
187 | int err; | |
188 | ||
189 | page = find_get_page(mapping, index); | |
190 | if (page && PageUptodate(page)) | |
191 | return page; | |
192 | f2fs_put_page(page, 0); | |
193 | ||
194 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 195 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
196 | if (err) |
197 | return ERR_PTR(err); | |
198 | f2fs_put_dnode(&dn); | |
199 | ||
200 | if (dn.data_blkaddr == NULL_ADDR) | |
201 | return ERR_PTR(-ENOENT); | |
202 | ||
203 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
204 | if (dn.data_blkaddr == NEW_ADDR) | |
205 | return ERR_PTR(-EINVAL); | |
206 | ||
6f85b352 | 207 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
eb47b800 JK |
208 | if (!page) |
209 | return ERR_PTR(-ENOMEM); | |
210 | ||
393ff91f JK |
211 | if (PageUptodate(page)) { |
212 | unlock_page(page); | |
213 | return page; | |
214 | } | |
215 | ||
c718379b JK |
216 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, |
217 | sync ? READ_SYNC : READA); | |
218 | if (sync) { | |
219 | wait_on_page_locked(page); | |
220 | if (!PageUptodate(page)) { | |
221 | f2fs_put_page(page, 0); | |
222 | return ERR_PTR(-EIO); | |
223 | } | |
eb47b800 | 224 | } |
eb47b800 JK |
225 | return page; |
226 | } | |
227 | ||
0a8165d7 | 228 | /* |
eb47b800 JK |
229 | * If it tries to access a hole, return an error. |
230 | * Because, the callers, functions in dir.c and GC, should be able to know | |
231 | * whether this page exists or not. | |
232 | */ | |
233 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
234 | { | |
235 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
236 | struct address_space *mapping = inode->i_mapping; | |
237 | struct dnode_of_data dn; | |
238 | struct page *page; | |
239 | int err; | |
240 | ||
650495de | 241 | repeat: |
6f85b352 | 242 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
650495de JK |
243 | if (!page) |
244 | return ERR_PTR(-ENOMEM); | |
245 | ||
eb47b800 | 246 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 247 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
248 | if (err) { |
249 | f2fs_put_page(page, 1); | |
eb47b800 | 250 | return ERR_PTR(err); |
650495de | 251 | } |
eb47b800 JK |
252 | f2fs_put_dnode(&dn); |
253 | ||
650495de JK |
254 | if (dn.data_blkaddr == NULL_ADDR) { |
255 | f2fs_put_page(page, 1); | |
eb47b800 | 256 | return ERR_PTR(-ENOENT); |
650495de | 257 | } |
eb47b800 JK |
258 | |
259 | if (PageUptodate(page)) | |
260 | return page; | |
261 | ||
262 | BUG_ON(dn.data_blkaddr == NEW_ADDR); | |
263 | BUG_ON(dn.data_blkaddr == NULL_ADDR); | |
264 | ||
265 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 266 | if (err) |
eb47b800 | 267 | return ERR_PTR(err); |
393ff91f JK |
268 | |
269 | lock_page(page); | |
270 | if (!PageUptodate(page)) { | |
271 | f2fs_put_page(page, 1); | |
272 | return ERR_PTR(-EIO); | |
eb47b800 | 273 | } |
afcb7ca0 JK |
274 | if (page->mapping != mapping) { |
275 | f2fs_put_page(page, 1); | |
276 | goto repeat; | |
eb47b800 JK |
277 | } |
278 | return page; | |
279 | } | |
280 | ||
0a8165d7 | 281 | /* |
eb47b800 JK |
282 | * Caller ensures that this data page is never allocated. |
283 | * A new zero-filled data page is allocated in the page cache. | |
39936837 JK |
284 | * |
285 | * Also, caller should grab and release a mutex by calling mutex_lock_op() and | |
286 | * mutex_unlock_op(). | |
44a83ff6 | 287 | * Note that, npage is set only by make_empty_dir. |
eb47b800 | 288 | */ |
64aa7ed9 JK |
289 | struct page *get_new_data_page(struct inode *inode, |
290 | struct page *npage, pgoff_t index, bool new_i_size) | |
eb47b800 JK |
291 | { |
292 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
293 | struct address_space *mapping = inode->i_mapping; | |
294 | struct page *page; | |
295 | struct dnode_of_data dn; | |
296 | int err; | |
297 | ||
64aa7ed9 | 298 | set_new_dnode(&dn, inode, npage, npage, 0); |
266e97a8 | 299 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
eb47b800 JK |
300 | if (err) |
301 | return ERR_PTR(err); | |
302 | ||
303 | if (dn.data_blkaddr == NULL_ADDR) { | |
304 | if (reserve_new_block(&dn)) { | |
64aa7ed9 JK |
305 | if (!npage) |
306 | f2fs_put_dnode(&dn); | |
eb47b800 JK |
307 | return ERR_PTR(-ENOSPC); |
308 | } | |
309 | } | |
64aa7ed9 JK |
310 | if (!npage) |
311 | f2fs_put_dnode(&dn); | |
afcb7ca0 | 312 | repeat: |
eb47b800 JK |
313 | page = grab_cache_page(mapping, index); |
314 | if (!page) | |
315 | return ERR_PTR(-ENOMEM); | |
316 | ||
317 | if (PageUptodate(page)) | |
318 | return page; | |
319 | ||
320 | if (dn.data_blkaddr == NEW_ADDR) { | |
321 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 322 | SetPageUptodate(page); |
eb47b800 JK |
323 | } else { |
324 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 325 | if (err) |
eb47b800 | 326 | return ERR_PTR(err); |
393ff91f JK |
327 | lock_page(page); |
328 | if (!PageUptodate(page)) { | |
329 | f2fs_put_page(page, 1); | |
330 | return ERR_PTR(-EIO); | |
eb47b800 | 331 | } |
afcb7ca0 JK |
332 | if (page->mapping != mapping) { |
333 | f2fs_put_page(page, 1); | |
334 | goto repeat; | |
eb47b800 JK |
335 | } |
336 | } | |
eb47b800 JK |
337 | |
338 | if (new_i_size && | |
339 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
340 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
341 | /* Only the directory inode sets new_i_size */ |
342 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
343 | mark_inode_dirty_sync(inode); |
344 | } | |
345 | return page; | |
346 | } | |
347 | ||
348 | static void read_end_io(struct bio *bio, int err) | |
349 | { | |
350 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
351 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
352 | ||
353 | do { | |
354 | struct page *page = bvec->bv_page; | |
355 | ||
356 | if (--bvec >= bio->bi_io_vec) | |
357 | prefetchw(&bvec->bv_page->flags); | |
358 | ||
359 | if (uptodate) { | |
360 | SetPageUptodate(page); | |
361 | } else { | |
362 | ClearPageUptodate(page); | |
363 | SetPageError(page); | |
364 | } | |
365 | unlock_page(page); | |
366 | } while (bvec >= bio->bi_io_vec); | |
eb47b800 JK |
367 | bio_put(bio); |
368 | } | |
369 | ||
0a8165d7 | 370 | /* |
eb47b800 | 371 | * Fill the locked page with data located in the block address. |
393ff91f | 372 | * Return unlocked page. |
eb47b800 JK |
373 | */ |
374 | int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page, | |
375 | block_t blk_addr, int type) | |
376 | { | |
377 | struct block_device *bdev = sbi->sb->s_bdev; | |
eb47b800 JK |
378 | struct bio *bio; |
379 | ||
848753aa | 380 | trace_f2fs_readpage(page, blk_addr, type); |
eb47b800 JK |
381 | |
382 | down_read(&sbi->bio_sem); | |
383 | ||
384 | /* Allocate a new bio */ | |
3cd8a239 | 385 | bio = f2fs_bio_alloc(bdev, 1); |
eb47b800 JK |
386 | |
387 | /* Initialize the bio */ | |
3cd8a239 | 388 | bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); |
eb47b800 | 389 | bio->bi_end_io = read_end_io; |
3cd8a239 | 390 | |
eb47b800 | 391 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { |
eb47b800 JK |
392 | bio_put(bio); |
393 | up_read(&sbi->bio_sem); | |
393ff91f | 394 | f2fs_put_page(page, 1); |
eb47b800 JK |
395 | return -EFAULT; |
396 | } | |
397 | ||
398 | submit_bio(type, bio); | |
399 | up_read(&sbi->bio_sem); | |
eb47b800 JK |
400 | return 0; |
401 | } | |
402 | ||
0a8165d7 | 403 | /* |
eb47b800 JK |
404 | * This function should be used by the data read flow only where it |
405 | * does not check the "create" flag that indicates block allocation. | |
406 | * The reason for this special functionality is to exploit VFS readahead | |
407 | * mechanism. | |
408 | */ | |
409 | static int get_data_block_ro(struct inode *inode, sector_t iblock, | |
410 | struct buffer_head *bh_result, int create) | |
411 | { | |
412 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
413 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
414 | struct dnode_of_data dn; | |
415 | pgoff_t pgofs; | |
416 | int err; | |
417 | ||
418 | /* Get the page offset from the block offset(iblock) */ | |
419 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
420 | ||
848753aa NJ |
421 | if (check_extent_cache(inode, pgofs, bh_result)) { |
422 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); | |
eb47b800 | 423 | return 0; |
848753aa | 424 | } |
eb47b800 JK |
425 | |
426 | /* When reading holes, we need its node page */ | |
427 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 428 | err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA); |
848753aa NJ |
429 | if (err) { |
430 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
eb47b800 | 431 | return (err == -ENOENT) ? 0 : err; |
848753aa | 432 | } |
eb47b800 JK |
433 | |
434 | /* It does not support data allocation */ | |
435 | BUG_ON(create); | |
436 | ||
437 | if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) { | |
438 | int i; | |
439 | unsigned int end_offset; | |
440 | ||
441 | end_offset = IS_INODE(dn.node_page) ? | |
442 | ADDRS_PER_INODE : | |
443 | ADDRS_PER_BLOCK; | |
444 | ||
445 | clear_buffer_new(bh_result); | |
446 | ||
447 | /* Give more consecutive addresses for the read ahead */ | |
448 | for (i = 0; i < end_offset - dn.ofs_in_node; i++) | |
449 | if (((datablock_addr(dn.node_page, | |
450 | dn.ofs_in_node + i)) | |
451 | != (dn.data_blkaddr + i)) || maxblocks == i) | |
452 | break; | |
453 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); | |
454 | bh_result->b_size = (i << blkbits); | |
455 | } | |
456 | f2fs_put_dnode(&dn); | |
848753aa | 457 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); |
eb47b800 JK |
458 | return 0; |
459 | } | |
460 | ||
461 | static int f2fs_read_data_page(struct file *file, struct page *page) | |
462 | { | |
463 | return mpage_readpage(page, get_data_block_ro); | |
464 | } | |
465 | ||
466 | static int f2fs_read_data_pages(struct file *file, | |
467 | struct address_space *mapping, | |
468 | struct list_head *pages, unsigned nr_pages) | |
469 | { | |
470 | return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro); | |
471 | } | |
472 | ||
473 | int do_write_data_page(struct page *page) | |
474 | { | |
475 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
476 | block_t old_blk_addr, new_blk_addr; |
477 | struct dnode_of_data dn; | |
478 | int err = 0; | |
479 | ||
480 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 481 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
482 | if (err) |
483 | return err; | |
484 | ||
485 | old_blk_addr = dn.data_blkaddr; | |
486 | ||
487 | /* This page is already truncated */ | |
488 | if (old_blk_addr == NULL_ADDR) | |
489 | goto out_writepage; | |
490 | ||
491 | set_page_writeback(page); | |
492 | ||
493 | /* | |
494 | * If current allocation needs SSR, | |
495 | * it had better in-place writes for updated data. | |
496 | */ | |
b25958b6 HL |
497 | if (unlikely(old_blk_addr != NEW_ADDR && |
498 | !is_cold_data(page) && | |
499 | need_inplace_update(inode))) { | |
eb47b800 JK |
500 | rewrite_data_page(F2FS_SB(inode->i_sb), page, |
501 | old_blk_addr); | |
502 | } else { | |
503 | write_data_page(inode, page, &dn, | |
504 | old_blk_addr, &new_blk_addr); | |
505 | update_extent_cache(new_blk_addr, &dn); | |
eb47b800 JK |
506 | } |
507 | out_writepage: | |
508 | f2fs_put_dnode(&dn); | |
509 | return err; | |
510 | } | |
511 | ||
512 | static int f2fs_write_data_page(struct page *page, | |
513 | struct writeback_control *wbc) | |
514 | { | |
515 | struct inode *inode = page->mapping->host; | |
516 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
517 | loff_t i_size = i_size_read(inode); | |
518 | const pgoff_t end_index = ((unsigned long long) i_size) | |
519 | >> PAGE_CACHE_SHIFT; | |
520 | unsigned offset; | |
39936837 | 521 | bool need_balance_fs = false; |
eb47b800 JK |
522 | int err = 0; |
523 | ||
524 | if (page->index < end_index) | |
39936837 | 525 | goto write; |
eb47b800 JK |
526 | |
527 | /* | |
528 | * If the offset is out-of-range of file size, | |
529 | * this page does not have to be written to disk. | |
530 | */ | |
531 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
532 | if ((page->index >= end_index + 1) || !offset) { | |
533 | if (S_ISDIR(inode->i_mode)) { | |
534 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
535 | inode_dec_dirty_dents(inode); | |
536 | } | |
39936837 | 537 | goto out; |
eb47b800 JK |
538 | } |
539 | ||
540 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 JK |
541 | write: |
542 | if (sbi->por_doing) { | |
543 | err = AOP_WRITEPAGE_ACTIVATE; | |
eb47b800 | 544 | goto redirty_out; |
39936837 | 545 | } |
eb47b800 | 546 | |
39936837 | 547 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 JK |
548 | if (S_ISDIR(inode->i_mode)) { |
549 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
550 | inode_dec_dirty_dents(inode); | |
39936837 JK |
551 | err = do_write_data_page(page); |
552 | } else { | |
553 | int ilock = mutex_lock_op(sbi); | |
554 | err = do_write_data_page(page); | |
555 | mutex_unlock_op(sbi, ilock); | |
556 | need_balance_fs = true; | |
eb47b800 | 557 | } |
39936837 JK |
558 | if (err == -ENOENT) |
559 | goto out; | |
560 | else if (err) | |
561 | goto redirty_out; | |
eb47b800 JK |
562 | |
563 | if (wbc->for_reclaim) | |
564 | f2fs_submit_bio(sbi, DATA, true); | |
565 | ||
eb47b800 | 566 | clear_cold_data(page); |
39936837 | 567 | out: |
eb47b800 | 568 | unlock_page(page); |
39936837 | 569 | if (need_balance_fs) |
eb47b800 JK |
570 | f2fs_balance_fs(sbi); |
571 | return 0; | |
572 | ||
eb47b800 JK |
573 | redirty_out: |
574 | wbc->pages_skipped++; | |
575 | set_page_dirty(page); | |
39936837 | 576 | return err; |
eb47b800 JK |
577 | } |
578 | ||
579 | #define MAX_DESIRED_PAGES_WP 4096 | |
580 | ||
fa9150a8 NJ |
581 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
582 | void *data) | |
583 | { | |
584 | struct address_space *mapping = data; | |
585 | int ret = mapping->a_ops->writepage(page, wbc); | |
586 | mapping_set_error(mapping, ret); | |
587 | return ret; | |
588 | } | |
589 | ||
25ca923b | 590 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
591 | struct writeback_control *wbc) |
592 | { | |
593 | struct inode *inode = mapping->host; | |
594 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
531ad7d5 | 595 | bool locked = false; |
eb47b800 JK |
596 | int ret; |
597 | long excess_nrtw = 0, desired_nrtw; | |
598 | ||
cfb185a1 | 599 | /* deal with chardevs and other special file */ |
600 | if (!mapping->a_ops->writepage) | |
601 | return 0; | |
602 | ||
eb47b800 JK |
603 | if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) { |
604 | desired_nrtw = MAX_DESIRED_PAGES_WP; | |
605 | excess_nrtw = desired_nrtw - wbc->nr_to_write; | |
606 | wbc->nr_to_write = desired_nrtw; | |
607 | } | |
608 | ||
531ad7d5 | 609 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 610 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
611 | locked = true; |
612 | } | |
fa9150a8 | 613 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 614 | if (locked) |
eb47b800 JK |
615 | mutex_unlock(&sbi->writepages); |
616 | f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL)); | |
617 | ||
618 | remove_dirty_dir_inode(inode); | |
619 | ||
620 | wbc->nr_to_write -= excess_nrtw; | |
621 | return ret; | |
622 | } | |
623 | ||
624 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, | |
625 | loff_t pos, unsigned len, unsigned flags, | |
626 | struct page **pagep, void **fsdata) | |
627 | { | |
628 | struct inode *inode = mapping->host; | |
629 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
630 | struct page *page; | |
631 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; | |
632 | struct dnode_of_data dn; | |
633 | int err = 0; | |
39936837 | 634 | int ilock; |
eb47b800 | 635 | |
eb47b800 | 636 | f2fs_balance_fs(sbi); |
afcb7ca0 | 637 | repeat: |
eb47b800 JK |
638 | page = grab_cache_page_write_begin(mapping, index, flags); |
639 | if (!page) | |
640 | return -ENOMEM; | |
641 | *pagep = page; | |
642 | ||
39936837 | 643 | ilock = mutex_lock_op(sbi); |
eb47b800 JK |
644 | |
645 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 646 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
39936837 JK |
647 | if (err) |
648 | goto err; | |
eb47b800 | 649 | |
39936837 | 650 | if (dn.data_blkaddr == NULL_ADDR) |
eb47b800 | 651 | err = reserve_new_block(&dn); |
39936837 | 652 | |
eb47b800 | 653 | f2fs_put_dnode(&dn); |
39936837 JK |
654 | if (err) |
655 | goto err; | |
eb47b800 | 656 | |
39936837 | 657 | mutex_unlock_op(sbi, ilock); |
eb47b800 JK |
658 | |
659 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) | |
660 | return 0; | |
661 | ||
662 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
663 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
664 | unsigned end = start + len; | |
665 | ||
666 | /* Reading beyond i_size is simple: memset to zero */ | |
667 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 668 | goto out; |
eb47b800 JK |
669 | } |
670 | ||
671 | if (dn.data_blkaddr == NEW_ADDR) { | |
672 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
673 | } else { | |
674 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 675 | if (err) |
eb47b800 | 676 | return err; |
393ff91f JK |
677 | lock_page(page); |
678 | if (!PageUptodate(page)) { | |
679 | f2fs_put_page(page, 1); | |
680 | return -EIO; | |
eb47b800 | 681 | } |
afcb7ca0 JK |
682 | if (page->mapping != mapping) { |
683 | f2fs_put_page(page, 1); | |
684 | goto repeat; | |
eb47b800 JK |
685 | } |
686 | } | |
393ff91f | 687 | out: |
eb47b800 JK |
688 | SetPageUptodate(page); |
689 | clear_cold_data(page); | |
690 | return 0; | |
39936837 JK |
691 | |
692 | err: | |
693 | mutex_unlock_op(sbi, ilock); | |
694 | f2fs_put_page(page, 1); | |
695 | return err; | |
eb47b800 JK |
696 | } |
697 | ||
a1dd3c13 JK |
698 | static int f2fs_write_end(struct file *file, |
699 | struct address_space *mapping, | |
700 | loff_t pos, unsigned len, unsigned copied, | |
701 | struct page *page, void *fsdata) | |
702 | { | |
703 | struct inode *inode = page->mapping->host; | |
704 | ||
705 | SetPageUptodate(page); | |
706 | set_page_dirty(page); | |
707 | ||
708 | if (pos + copied > i_size_read(inode)) { | |
709 | i_size_write(inode, pos + copied); | |
710 | mark_inode_dirty(inode); | |
711 | update_inode_page(inode); | |
712 | } | |
713 | ||
714 | unlock_page(page); | |
715 | page_cache_release(page); | |
716 | return copied; | |
717 | } | |
718 | ||
eb47b800 JK |
719 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
720 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
721 | { | |
722 | struct file *file = iocb->ki_filp; | |
723 | struct inode *inode = file->f_mapping->host; | |
724 | ||
725 | if (rw == WRITE) | |
726 | return 0; | |
727 | ||
728 | /* Needs synchronization with the cleaner */ | |
729 | return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, | |
730 | get_data_block_ro); | |
731 | } | |
732 | ||
d47992f8 LC |
733 | static void f2fs_invalidate_data_page(struct page *page, unsigned int offset, |
734 | unsigned int length) | |
eb47b800 JK |
735 | { |
736 | struct inode *inode = page->mapping->host; | |
737 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
738 | if (S_ISDIR(inode->i_mode) && PageDirty(page)) { | |
739 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
740 | inode_dec_dirty_dents(inode); | |
741 | } | |
742 | ClearPagePrivate(page); | |
743 | } | |
744 | ||
745 | static int f2fs_release_data_page(struct page *page, gfp_t wait) | |
746 | { | |
747 | ClearPagePrivate(page); | |
c3850aa1 | 748 | return 1; |
eb47b800 JK |
749 | } |
750 | ||
751 | static int f2fs_set_data_page_dirty(struct page *page) | |
752 | { | |
753 | struct address_space *mapping = page->mapping; | |
754 | struct inode *inode = mapping->host; | |
755 | ||
756 | SetPageUptodate(page); | |
757 | if (!PageDirty(page)) { | |
758 | __set_page_dirty_nobuffers(page); | |
759 | set_dirty_dir_page(inode, page); | |
760 | return 1; | |
761 | } | |
762 | return 0; | |
763 | } | |
764 | ||
c01e54b7 JK |
765 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
766 | { | |
767 | return generic_block_bmap(mapping, block, get_data_block_ro); | |
768 | } | |
769 | ||
eb47b800 JK |
770 | const struct address_space_operations f2fs_dblock_aops = { |
771 | .readpage = f2fs_read_data_page, | |
772 | .readpages = f2fs_read_data_pages, | |
773 | .writepage = f2fs_write_data_page, | |
774 | .writepages = f2fs_write_data_pages, | |
775 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 776 | .write_end = f2fs_write_end, |
eb47b800 JK |
777 | .set_page_dirty = f2fs_set_data_page_dirty, |
778 | .invalidatepage = f2fs_invalidate_data_page, | |
779 | .releasepage = f2fs_release_data_page, | |
780 | .direct_IO = f2fs_direct_IO, | |
c01e54b7 | 781 | .bmap = f2fs_bmap, |
eb47b800 | 782 | }; |