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 | ||
40 | wait_on_page_writeback(node_page); | |
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); | |
179 | return; | |
180 | } | |
181 | ||
c718379b | 182 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 JK |
183 | { |
184 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
185 | struct address_space *mapping = inode->i_mapping; | |
186 | struct dnode_of_data dn; | |
187 | struct page *page; | |
188 | int err; | |
189 | ||
190 | page = find_get_page(mapping, index); | |
191 | if (page && PageUptodate(page)) | |
192 | return page; | |
193 | f2fs_put_page(page, 0); | |
194 | ||
195 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 196 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
197 | if (err) |
198 | return ERR_PTR(err); | |
199 | f2fs_put_dnode(&dn); | |
200 | ||
201 | if (dn.data_blkaddr == NULL_ADDR) | |
202 | return ERR_PTR(-ENOENT); | |
203 | ||
204 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
205 | if (dn.data_blkaddr == NEW_ADDR) | |
206 | return ERR_PTR(-EINVAL); | |
207 | ||
6f85b352 | 208 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
eb47b800 JK |
209 | if (!page) |
210 | return ERR_PTR(-ENOMEM); | |
211 | ||
393ff91f JK |
212 | if (PageUptodate(page)) { |
213 | unlock_page(page); | |
214 | return page; | |
215 | } | |
216 | ||
c718379b JK |
217 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, |
218 | sync ? READ_SYNC : READA); | |
219 | if (sync) { | |
220 | wait_on_page_locked(page); | |
221 | if (!PageUptodate(page)) { | |
222 | f2fs_put_page(page, 0); | |
223 | return ERR_PTR(-EIO); | |
224 | } | |
eb47b800 | 225 | } |
eb47b800 JK |
226 | return page; |
227 | } | |
228 | ||
0a8165d7 | 229 | /* |
eb47b800 JK |
230 | * If it tries to access a hole, return an error. |
231 | * Because, the callers, functions in dir.c and GC, should be able to know | |
232 | * whether this page exists or not. | |
233 | */ | |
234 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
235 | { | |
236 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
237 | struct address_space *mapping = inode->i_mapping; | |
238 | struct dnode_of_data dn; | |
239 | struct page *page; | |
240 | int err; | |
241 | ||
650495de | 242 | repeat: |
6f85b352 | 243 | page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS); |
650495de JK |
244 | if (!page) |
245 | return ERR_PTR(-ENOMEM); | |
246 | ||
eb47b800 | 247 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 248 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
249 | if (err) { |
250 | f2fs_put_page(page, 1); | |
eb47b800 | 251 | return ERR_PTR(err); |
650495de | 252 | } |
eb47b800 JK |
253 | f2fs_put_dnode(&dn); |
254 | ||
650495de JK |
255 | if (dn.data_blkaddr == NULL_ADDR) { |
256 | f2fs_put_page(page, 1); | |
eb47b800 | 257 | return ERR_PTR(-ENOENT); |
650495de | 258 | } |
eb47b800 JK |
259 | |
260 | if (PageUptodate(page)) | |
261 | return page; | |
262 | ||
263 | BUG_ON(dn.data_blkaddr == NEW_ADDR); | |
264 | BUG_ON(dn.data_blkaddr == NULL_ADDR); | |
265 | ||
266 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 267 | if (err) |
eb47b800 | 268 | return ERR_PTR(err); |
393ff91f JK |
269 | |
270 | lock_page(page); | |
271 | if (!PageUptodate(page)) { | |
272 | f2fs_put_page(page, 1); | |
273 | return ERR_PTR(-EIO); | |
eb47b800 | 274 | } |
afcb7ca0 JK |
275 | if (page->mapping != mapping) { |
276 | f2fs_put_page(page, 1); | |
277 | goto repeat; | |
eb47b800 JK |
278 | } |
279 | return page; | |
280 | } | |
281 | ||
0a8165d7 | 282 | /* |
eb47b800 JK |
283 | * Caller ensures that this data page is never allocated. |
284 | * A new zero-filled data page is allocated in the page cache. | |
39936837 JK |
285 | * |
286 | * Also, caller should grab and release a mutex by calling mutex_lock_op() and | |
287 | * mutex_unlock_op(). | |
44a83ff6 | 288 | * Note that, npage is set only by make_empty_dir. |
eb47b800 | 289 | */ |
64aa7ed9 JK |
290 | struct page *get_new_data_page(struct inode *inode, |
291 | struct page *npage, pgoff_t index, bool new_i_size) | |
eb47b800 JK |
292 | { |
293 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
294 | struct address_space *mapping = inode->i_mapping; | |
295 | struct page *page; | |
296 | struct dnode_of_data dn; | |
297 | int err; | |
298 | ||
64aa7ed9 | 299 | set_new_dnode(&dn, inode, npage, npage, 0); |
266e97a8 | 300 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
eb47b800 JK |
301 | if (err) |
302 | return ERR_PTR(err); | |
303 | ||
304 | if (dn.data_blkaddr == NULL_ADDR) { | |
305 | if (reserve_new_block(&dn)) { | |
64aa7ed9 JK |
306 | if (!npage) |
307 | f2fs_put_dnode(&dn); | |
eb47b800 JK |
308 | return ERR_PTR(-ENOSPC); |
309 | } | |
310 | } | |
64aa7ed9 JK |
311 | if (!npage) |
312 | f2fs_put_dnode(&dn); | |
afcb7ca0 | 313 | repeat: |
eb47b800 JK |
314 | page = grab_cache_page(mapping, index); |
315 | if (!page) | |
316 | return ERR_PTR(-ENOMEM); | |
317 | ||
318 | if (PageUptodate(page)) | |
319 | return page; | |
320 | ||
321 | if (dn.data_blkaddr == NEW_ADDR) { | |
322 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 323 | SetPageUptodate(page); |
eb47b800 JK |
324 | } else { |
325 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 326 | if (err) |
eb47b800 | 327 | return ERR_PTR(err); |
393ff91f JK |
328 | lock_page(page); |
329 | if (!PageUptodate(page)) { | |
330 | f2fs_put_page(page, 1); | |
331 | return ERR_PTR(-EIO); | |
eb47b800 | 332 | } |
afcb7ca0 JK |
333 | if (page->mapping != mapping) { |
334 | f2fs_put_page(page, 1); | |
335 | goto repeat; | |
eb47b800 JK |
336 | } |
337 | } | |
eb47b800 JK |
338 | |
339 | if (new_i_size && | |
340 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
341 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
342 | /* Only the directory inode sets new_i_size */ |
343 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
344 | mark_inode_dirty_sync(inode); |
345 | } | |
346 | return page; | |
347 | } | |
348 | ||
349 | static void read_end_io(struct bio *bio, int err) | |
350 | { | |
351 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
352 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
353 | ||
354 | do { | |
355 | struct page *page = bvec->bv_page; | |
356 | ||
357 | if (--bvec >= bio->bi_io_vec) | |
358 | prefetchw(&bvec->bv_page->flags); | |
359 | ||
360 | if (uptodate) { | |
361 | SetPageUptodate(page); | |
362 | } else { | |
363 | ClearPageUptodate(page); | |
364 | SetPageError(page); | |
365 | } | |
366 | unlock_page(page); | |
367 | } while (bvec >= bio->bi_io_vec); | |
eb47b800 JK |
368 | bio_put(bio); |
369 | } | |
370 | ||
0a8165d7 | 371 | /* |
eb47b800 | 372 | * Fill the locked page with data located in the block address. |
393ff91f | 373 | * Return unlocked page. |
eb47b800 JK |
374 | */ |
375 | int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page, | |
376 | block_t blk_addr, int type) | |
377 | { | |
378 | struct block_device *bdev = sbi->sb->s_bdev; | |
eb47b800 JK |
379 | struct bio *bio; |
380 | ||
848753aa | 381 | trace_f2fs_readpage(page, blk_addr, type); |
eb47b800 JK |
382 | |
383 | down_read(&sbi->bio_sem); | |
384 | ||
385 | /* Allocate a new bio */ | |
3cd8a239 | 386 | bio = f2fs_bio_alloc(bdev, 1); |
eb47b800 JK |
387 | |
388 | /* Initialize the bio */ | |
3cd8a239 | 389 | bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); |
eb47b800 | 390 | bio->bi_end_io = read_end_io; |
3cd8a239 | 391 | |
eb47b800 JK |
392 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { |
393 | kfree(bio->bi_private); | |
394 | bio_put(bio); | |
395 | up_read(&sbi->bio_sem); | |
393ff91f | 396 | f2fs_put_page(page, 1); |
eb47b800 JK |
397 | return -EFAULT; |
398 | } | |
399 | ||
400 | submit_bio(type, bio); | |
401 | up_read(&sbi->bio_sem); | |
eb47b800 JK |
402 | return 0; |
403 | } | |
404 | ||
0a8165d7 | 405 | /* |
eb47b800 JK |
406 | * This function should be used by the data read flow only where it |
407 | * does not check the "create" flag that indicates block allocation. | |
408 | * The reason for this special functionality is to exploit VFS readahead | |
409 | * mechanism. | |
410 | */ | |
411 | static int get_data_block_ro(struct inode *inode, sector_t iblock, | |
412 | struct buffer_head *bh_result, int create) | |
413 | { | |
414 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
415 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
416 | struct dnode_of_data dn; | |
417 | pgoff_t pgofs; | |
418 | int err; | |
419 | ||
420 | /* Get the page offset from the block offset(iblock) */ | |
421 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
422 | ||
848753aa NJ |
423 | if (check_extent_cache(inode, pgofs, bh_result)) { |
424 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); | |
eb47b800 | 425 | return 0; |
848753aa | 426 | } |
eb47b800 JK |
427 | |
428 | /* When reading holes, we need its node page */ | |
429 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 430 | err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA); |
848753aa NJ |
431 | if (err) { |
432 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
eb47b800 | 433 | return (err == -ENOENT) ? 0 : err; |
848753aa | 434 | } |
eb47b800 JK |
435 | |
436 | /* It does not support data allocation */ | |
437 | BUG_ON(create); | |
438 | ||
439 | if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) { | |
440 | int i; | |
441 | unsigned int end_offset; | |
442 | ||
443 | end_offset = IS_INODE(dn.node_page) ? | |
444 | ADDRS_PER_INODE : | |
445 | ADDRS_PER_BLOCK; | |
446 | ||
447 | clear_buffer_new(bh_result); | |
448 | ||
449 | /* Give more consecutive addresses for the read ahead */ | |
450 | for (i = 0; i < end_offset - dn.ofs_in_node; i++) | |
451 | if (((datablock_addr(dn.node_page, | |
452 | dn.ofs_in_node + i)) | |
453 | != (dn.data_blkaddr + i)) || maxblocks == i) | |
454 | break; | |
455 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); | |
456 | bh_result->b_size = (i << blkbits); | |
457 | } | |
458 | f2fs_put_dnode(&dn); | |
848753aa | 459 | trace_f2fs_get_data_block(inode, iblock, bh_result, 0); |
eb47b800 JK |
460 | return 0; |
461 | } | |
462 | ||
463 | static int f2fs_read_data_page(struct file *file, struct page *page) | |
464 | { | |
465 | return mpage_readpage(page, get_data_block_ro); | |
466 | } | |
467 | ||
468 | static int f2fs_read_data_pages(struct file *file, | |
469 | struct address_space *mapping, | |
470 | struct list_head *pages, unsigned nr_pages) | |
471 | { | |
472 | return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro); | |
473 | } | |
474 | ||
475 | int do_write_data_page(struct page *page) | |
476 | { | |
477 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
478 | block_t old_blk_addr, new_blk_addr; |
479 | struct dnode_of_data dn; | |
480 | int err = 0; | |
481 | ||
482 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 483 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
484 | if (err) |
485 | return err; | |
486 | ||
487 | old_blk_addr = dn.data_blkaddr; | |
488 | ||
489 | /* This page is already truncated */ | |
490 | if (old_blk_addr == NULL_ADDR) | |
491 | goto out_writepage; | |
492 | ||
493 | set_page_writeback(page); | |
494 | ||
495 | /* | |
496 | * If current allocation needs SSR, | |
497 | * it had better in-place writes for updated data. | |
498 | */ | |
b25958b6 HL |
499 | if (unlikely(old_blk_addr != NEW_ADDR && |
500 | !is_cold_data(page) && | |
501 | need_inplace_update(inode))) { | |
eb47b800 JK |
502 | rewrite_data_page(F2FS_SB(inode->i_sb), page, |
503 | old_blk_addr); | |
504 | } else { | |
505 | write_data_page(inode, page, &dn, | |
506 | old_blk_addr, &new_blk_addr); | |
507 | update_extent_cache(new_blk_addr, &dn); | |
eb47b800 JK |
508 | } |
509 | out_writepage: | |
510 | f2fs_put_dnode(&dn); | |
511 | return err; | |
512 | } | |
513 | ||
514 | static int f2fs_write_data_page(struct page *page, | |
515 | struct writeback_control *wbc) | |
516 | { | |
517 | struct inode *inode = page->mapping->host; | |
518 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
519 | loff_t i_size = i_size_read(inode); | |
520 | const pgoff_t end_index = ((unsigned long long) i_size) | |
521 | >> PAGE_CACHE_SHIFT; | |
522 | unsigned offset; | |
39936837 | 523 | bool need_balance_fs = false; |
eb47b800 JK |
524 | int err = 0; |
525 | ||
526 | if (page->index < end_index) | |
39936837 | 527 | goto write; |
eb47b800 JK |
528 | |
529 | /* | |
530 | * If the offset is out-of-range of file size, | |
531 | * this page does not have to be written to disk. | |
532 | */ | |
533 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
534 | if ((page->index >= end_index + 1) || !offset) { | |
535 | if (S_ISDIR(inode->i_mode)) { | |
536 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
537 | inode_dec_dirty_dents(inode); | |
538 | } | |
39936837 | 539 | goto out; |
eb47b800 JK |
540 | } |
541 | ||
542 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 JK |
543 | write: |
544 | if (sbi->por_doing) { | |
545 | err = AOP_WRITEPAGE_ACTIVATE; | |
eb47b800 | 546 | goto redirty_out; |
39936837 | 547 | } |
eb47b800 | 548 | |
39936837 | 549 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 JK |
550 | if (S_ISDIR(inode->i_mode)) { |
551 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
552 | inode_dec_dirty_dents(inode); | |
39936837 JK |
553 | err = do_write_data_page(page); |
554 | } else { | |
555 | int ilock = mutex_lock_op(sbi); | |
556 | err = do_write_data_page(page); | |
557 | mutex_unlock_op(sbi, ilock); | |
558 | need_balance_fs = true; | |
eb47b800 | 559 | } |
39936837 JK |
560 | if (err == -ENOENT) |
561 | goto out; | |
562 | else if (err) | |
563 | goto redirty_out; | |
eb47b800 JK |
564 | |
565 | if (wbc->for_reclaim) | |
566 | f2fs_submit_bio(sbi, DATA, true); | |
567 | ||
eb47b800 | 568 | clear_cold_data(page); |
39936837 | 569 | out: |
eb47b800 | 570 | unlock_page(page); |
39936837 | 571 | if (need_balance_fs) |
eb47b800 JK |
572 | f2fs_balance_fs(sbi); |
573 | return 0; | |
574 | ||
eb47b800 JK |
575 | redirty_out: |
576 | wbc->pages_skipped++; | |
577 | set_page_dirty(page); | |
39936837 | 578 | return err; |
eb47b800 JK |
579 | } |
580 | ||
581 | #define MAX_DESIRED_PAGES_WP 4096 | |
582 | ||
fa9150a8 NJ |
583 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
584 | void *data) | |
585 | { | |
586 | struct address_space *mapping = data; | |
587 | int ret = mapping->a_ops->writepage(page, wbc); | |
588 | mapping_set_error(mapping, ret); | |
589 | return ret; | |
590 | } | |
591 | ||
25ca923b | 592 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
593 | struct writeback_control *wbc) |
594 | { | |
595 | struct inode *inode = mapping->host; | |
596 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
531ad7d5 | 597 | bool locked = false; |
eb47b800 JK |
598 | int ret; |
599 | long excess_nrtw = 0, desired_nrtw; | |
600 | ||
cfb185a1 | 601 | /* deal with chardevs and other special file */ |
602 | if (!mapping->a_ops->writepage) | |
603 | return 0; | |
604 | ||
eb47b800 JK |
605 | if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) { |
606 | desired_nrtw = MAX_DESIRED_PAGES_WP; | |
607 | excess_nrtw = desired_nrtw - wbc->nr_to_write; | |
608 | wbc->nr_to_write = desired_nrtw; | |
609 | } | |
610 | ||
531ad7d5 | 611 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 612 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
613 | locked = true; |
614 | } | |
fa9150a8 | 615 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 616 | if (locked) |
eb47b800 JK |
617 | mutex_unlock(&sbi->writepages); |
618 | f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL)); | |
619 | ||
620 | remove_dirty_dir_inode(inode); | |
621 | ||
622 | wbc->nr_to_write -= excess_nrtw; | |
623 | return ret; | |
624 | } | |
625 | ||
626 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, | |
627 | loff_t pos, unsigned len, unsigned flags, | |
628 | struct page **pagep, void **fsdata) | |
629 | { | |
630 | struct inode *inode = mapping->host; | |
631 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
632 | struct page *page; | |
633 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; | |
634 | struct dnode_of_data dn; | |
635 | int err = 0; | |
39936837 | 636 | int ilock; |
eb47b800 JK |
637 | |
638 | /* for nobh_write_end */ | |
639 | *fsdata = NULL; | |
640 | ||
641 | f2fs_balance_fs(sbi); | |
afcb7ca0 | 642 | repeat: |
eb47b800 JK |
643 | page = grab_cache_page_write_begin(mapping, index, flags); |
644 | if (!page) | |
645 | return -ENOMEM; | |
646 | *pagep = page; | |
647 | ||
39936837 | 648 | ilock = mutex_lock_op(sbi); |
eb47b800 JK |
649 | |
650 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 651 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
39936837 JK |
652 | if (err) |
653 | goto err; | |
eb47b800 | 654 | |
39936837 | 655 | if (dn.data_blkaddr == NULL_ADDR) |
eb47b800 | 656 | err = reserve_new_block(&dn); |
39936837 | 657 | |
eb47b800 | 658 | f2fs_put_dnode(&dn); |
39936837 JK |
659 | if (err) |
660 | goto err; | |
eb47b800 | 661 | |
39936837 | 662 | mutex_unlock_op(sbi, ilock); |
eb47b800 JK |
663 | |
664 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) | |
665 | return 0; | |
666 | ||
667 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
668 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
669 | unsigned end = start + len; | |
670 | ||
671 | /* Reading beyond i_size is simple: memset to zero */ | |
672 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 673 | goto out; |
eb47b800 JK |
674 | } |
675 | ||
676 | if (dn.data_blkaddr == NEW_ADDR) { | |
677 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
678 | } else { | |
679 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 680 | if (err) |
eb47b800 | 681 | return err; |
393ff91f JK |
682 | lock_page(page); |
683 | if (!PageUptodate(page)) { | |
684 | f2fs_put_page(page, 1); | |
685 | return -EIO; | |
eb47b800 | 686 | } |
afcb7ca0 JK |
687 | if (page->mapping != mapping) { |
688 | f2fs_put_page(page, 1); | |
689 | goto repeat; | |
eb47b800 JK |
690 | } |
691 | } | |
393ff91f | 692 | out: |
eb47b800 JK |
693 | SetPageUptodate(page); |
694 | clear_cold_data(page); | |
695 | return 0; | |
39936837 JK |
696 | |
697 | err: | |
698 | mutex_unlock_op(sbi, ilock); | |
699 | f2fs_put_page(page, 1); | |
700 | return err; | |
eb47b800 JK |
701 | } |
702 | ||
a1dd3c13 JK |
703 | static int f2fs_write_end(struct file *file, |
704 | struct address_space *mapping, | |
705 | loff_t pos, unsigned len, unsigned copied, | |
706 | struct page *page, void *fsdata) | |
707 | { | |
708 | struct inode *inode = page->mapping->host; | |
709 | ||
710 | SetPageUptodate(page); | |
711 | set_page_dirty(page); | |
712 | ||
713 | if (pos + copied > i_size_read(inode)) { | |
714 | i_size_write(inode, pos + copied); | |
715 | mark_inode_dirty(inode); | |
716 | update_inode_page(inode); | |
717 | } | |
718 | ||
719 | unlock_page(page); | |
720 | page_cache_release(page); | |
721 | return copied; | |
722 | } | |
723 | ||
eb47b800 JK |
724 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
725 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
726 | { | |
727 | struct file *file = iocb->ki_filp; | |
728 | struct inode *inode = file->f_mapping->host; | |
729 | ||
730 | if (rw == WRITE) | |
731 | return 0; | |
732 | ||
733 | /* Needs synchronization with the cleaner */ | |
734 | return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, | |
735 | get_data_block_ro); | |
736 | } | |
737 | ||
d47992f8 LC |
738 | static void f2fs_invalidate_data_page(struct page *page, unsigned int offset, |
739 | unsigned int length) | |
eb47b800 JK |
740 | { |
741 | struct inode *inode = page->mapping->host; | |
742 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
743 | if (S_ISDIR(inode->i_mode) && PageDirty(page)) { | |
744 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
745 | inode_dec_dirty_dents(inode); | |
746 | } | |
747 | ClearPagePrivate(page); | |
748 | } | |
749 | ||
750 | static int f2fs_release_data_page(struct page *page, gfp_t wait) | |
751 | { | |
752 | ClearPagePrivate(page); | |
c3850aa1 | 753 | return 1; |
eb47b800 JK |
754 | } |
755 | ||
756 | static int f2fs_set_data_page_dirty(struct page *page) | |
757 | { | |
758 | struct address_space *mapping = page->mapping; | |
759 | struct inode *inode = mapping->host; | |
760 | ||
761 | SetPageUptodate(page); | |
762 | if (!PageDirty(page)) { | |
763 | __set_page_dirty_nobuffers(page); | |
764 | set_dirty_dir_page(inode, page); | |
765 | return 1; | |
766 | } | |
767 | return 0; | |
768 | } | |
769 | ||
c01e54b7 JK |
770 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
771 | { | |
772 | return generic_block_bmap(mapping, block, get_data_block_ro); | |
773 | } | |
774 | ||
eb47b800 JK |
775 | const struct address_space_operations f2fs_dblock_aops = { |
776 | .readpage = f2fs_read_data_page, | |
777 | .readpages = f2fs_read_data_pages, | |
778 | .writepage = f2fs_write_data_page, | |
779 | .writepages = f2fs_write_data_pages, | |
780 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 781 | .write_end = f2fs_write_end, |
eb47b800 JK |
782 | .set_page_dirty = f2fs_set_data_page_dirty, |
783 | .invalidatepage = f2fs_invalidate_data_page, | |
784 | .releasepage = f2fs_release_data_page, | |
785 | .direct_IO = f2fs_direct_IO, | |
c01e54b7 | 786 | .bmap = f2fs_bmap, |
eb47b800 | 787 | }; |