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> | |
15 | #include <linux/writeback.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/blkdev.h> | |
18 | #include <linux/bio.h> | |
690e4a3e | 19 | #include <linux/prefetch.h> |
eb47b800 JK |
20 | |
21 | #include "f2fs.h" | |
22 | #include "node.h" | |
23 | #include "segment.h" | |
24 | ||
0a8165d7 | 25 | /* |
eb47b800 JK |
26 | * Lock ordering for the change of data block address: |
27 | * ->data_page | |
28 | * ->node_page | |
29 | * update block addresses in the node page | |
30 | */ | |
31 | static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr) | |
32 | { | |
33 | struct f2fs_node *rn; | |
34 | __le32 *addr_array; | |
35 | struct page *node_page = dn->node_page; | |
36 | unsigned int ofs_in_node = dn->ofs_in_node; | |
37 | ||
38 | wait_on_page_writeback(node_page); | |
39 | ||
40 | rn = (struct f2fs_node *)page_address(node_page); | |
41 | ||
42 | /* Get physical address of data block */ | |
43 | addr_array = blkaddr_in_node(rn); | |
44 | addr_array[ofs_in_node] = cpu_to_le32(new_addr); | |
45 | set_page_dirty(node_page); | |
46 | } | |
47 | ||
48 | int reserve_new_block(struct dnode_of_data *dn) | |
49 | { | |
50 | struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); | |
51 | ||
52 | if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)) | |
53 | return -EPERM; | |
54 | if (!inc_valid_block_count(sbi, dn->inode, 1)) | |
55 | return -ENOSPC; | |
56 | ||
57 | __set_data_blkaddr(dn, NEW_ADDR); | |
58 | dn->data_blkaddr = NEW_ADDR; | |
59 | sync_inode_page(dn); | |
60 | return 0; | |
61 | } | |
62 | ||
63 | static int check_extent_cache(struct inode *inode, pgoff_t pgofs, | |
64 | struct buffer_head *bh_result) | |
65 | { | |
66 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
67 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
68 | pgoff_t start_fofs, end_fofs; | |
69 | block_t start_blkaddr; | |
70 | ||
71 | read_lock(&fi->ext.ext_lock); | |
72 | if (fi->ext.len == 0) { | |
73 | read_unlock(&fi->ext.ext_lock); | |
74 | return 0; | |
75 | } | |
76 | ||
77 | sbi->total_hit_ext++; | |
78 | start_fofs = fi->ext.fofs; | |
79 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
80 | start_blkaddr = fi->ext.blk_addr; | |
81 | ||
82 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
83 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
84 | size_t count; | |
85 | ||
86 | clear_buffer_new(bh_result); | |
87 | map_bh(bh_result, inode->i_sb, | |
88 | start_blkaddr + pgofs - start_fofs); | |
89 | count = end_fofs - pgofs + 1; | |
90 | if (count < (UINT_MAX >> blkbits)) | |
91 | bh_result->b_size = (count << blkbits); | |
92 | else | |
93 | bh_result->b_size = UINT_MAX; | |
94 | ||
95 | sbi->read_hit_ext++; | |
96 | read_unlock(&fi->ext.ext_lock); | |
97 | return 1; | |
98 | } | |
99 | read_unlock(&fi->ext.ext_lock); | |
100 | return 0; | |
101 | } | |
102 | ||
103 | void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn) | |
104 | { | |
105 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
106 | pgoff_t fofs, start_fofs, end_fofs; | |
107 | block_t start_blkaddr, end_blkaddr; | |
108 | ||
109 | BUG_ON(blk_addr == NEW_ADDR); | |
110 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node; | |
111 | ||
112 | /* Update the page address in the parent node */ | |
113 | __set_data_blkaddr(dn, blk_addr); | |
114 | ||
115 | write_lock(&fi->ext.ext_lock); | |
116 | ||
117 | start_fofs = fi->ext.fofs; | |
118 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
119 | start_blkaddr = fi->ext.blk_addr; | |
120 | end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1; | |
121 | ||
122 | /* Drop and initialize the matched extent */ | |
123 | if (fi->ext.len == 1 && fofs == start_fofs) | |
124 | fi->ext.len = 0; | |
125 | ||
126 | /* Initial extent */ | |
127 | if (fi->ext.len == 0) { | |
128 | if (blk_addr != NULL_ADDR) { | |
129 | fi->ext.fofs = fofs; | |
130 | fi->ext.blk_addr = blk_addr; | |
131 | fi->ext.len = 1; | |
132 | } | |
133 | goto end_update; | |
134 | } | |
135 | ||
136 | /* Frone merge */ | |
137 | if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) { | |
138 | fi->ext.fofs--; | |
139 | fi->ext.blk_addr--; | |
140 | fi->ext.len++; | |
141 | goto end_update; | |
142 | } | |
143 | ||
144 | /* Back merge */ | |
145 | if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) { | |
146 | fi->ext.len++; | |
147 | goto end_update; | |
148 | } | |
149 | ||
150 | /* Split the existing extent */ | |
151 | if (fi->ext.len > 1 && | |
152 | fofs >= start_fofs && fofs <= end_fofs) { | |
153 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
154 | fi->ext.len = fofs - start_fofs; | |
155 | } else { | |
156 | fi->ext.fofs = fofs + 1; | |
157 | fi->ext.blk_addr = start_blkaddr + | |
158 | fofs - start_fofs + 1; | |
159 | fi->ext.len -= fofs - start_fofs + 1; | |
160 | } | |
161 | goto end_update; | |
162 | } | |
163 | write_unlock(&fi->ext.ext_lock); | |
164 | return; | |
165 | ||
166 | end_update: | |
167 | write_unlock(&fi->ext.ext_lock); | |
168 | sync_inode_page(dn); | |
169 | return; | |
170 | } | |
171 | ||
172 | struct page *find_data_page(struct inode *inode, pgoff_t index) | |
173 | { | |
174 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
175 | struct address_space *mapping = inode->i_mapping; | |
176 | struct dnode_of_data dn; | |
177 | struct page *page; | |
178 | int err; | |
179 | ||
180 | page = find_get_page(mapping, index); | |
181 | if (page && PageUptodate(page)) | |
182 | return page; | |
183 | f2fs_put_page(page, 0); | |
184 | ||
185 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 186 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
187 | if (err) |
188 | return ERR_PTR(err); | |
189 | f2fs_put_dnode(&dn); | |
190 | ||
191 | if (dn.data_blkaddr == NULL_ADDR) | |
192 | return ERR_PTR(-ENOENT); | |
193 | ||
194 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
195 | if (dn.data_blkaddr == NEW_ADDR) | |
196 | return ERR_PTR(-EINVAL); | |
197 | ||
198 | page = grab_cache_page(mapping, index); | |
199 | if (!page) | |
200 | return ERR_PTR(-ENOMEM); | |
201 | ||
393ff91f JK |
202 | if (PageUptodate(page)) { |
203 | unlock_page(page); | |
204 | return page; | |
205 | } | |
206 | ||
eb47b800 | 207 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); |
393ff91f JK |
208 | wait_on_page_locked(page); |
209 | if (!PageUptodate(page)) { | |
210 | f2fs_put_page(page, 0); | |
211 | return ERR_PTR(-EIO); | |
eb47b800 | 212 | } |
eb47b800 JK |
213 | return page; |
214 | } | |
215 | ||
0a8165d7 | 216 | /* |
eb47b800 JK |
217 | * If it tries to access a hole, return an error. |
218 | * Because, the callers, functions in dir.c and GC, should be able to know | |
219 | * whether this page exists or not. | |
220 | */ | |
221 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
222 | { | |
223 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
224 | struct address_space *mapping = inode->i_mapping; | |
225 | struct dnode_of_data dn; | |
226 | struct page *page; | |
227 | int err; | |
228 | ||
229 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 230 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
231 | if (err) |
232 | return ERR_PTR(err); | |
233 | f2fs_put_dnode(&dn); | |
234 | ||
235 | if (dn.data_blkaddr == NULL_ADDR) | |
236 | return ERR_PTR(-ENOENT); | |
237 | ||
238 | page = grab_cache_page(mapping, index); | |
239 | if (!page) | |
240 | return ERR_PTR(-ENOMEM); | |
241 | ||
242 | if (PageUptodate(page)) | |
243 | return page; | |
244 | ||
245 | BUG_ON(dn.data_blkaddr == NEW_ADDR); | |
246 | BUG_ON(dn.data_blkaddr == NULL_ADDR); | |
247 | ||
248 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 249 | if (err) |
eb47b800 | 250 | return ERR_PTR(err); |
393ff91f JK |
251 | |
252 | lock_page(page); | |
253 | if (!PageUptodate(page)) { | |
254 | f2fs_put_page(page, 1); | |
255 | return ERR_PTR(-EIO); | |
eb47b800 JK |
256 | } |
257 | return page; | |
258 | } | |
259 | ||
0a8165d7 | 260 | /* |
eb47b800 JK |
261 | * Caller ensures that this data page is never allocated. |
262 | * A new zero-filled data page is allocated in the page cache. | |
263 | */ | |
264 | struct page *get_new_data_page(struct inode *inode, pgoff_t index, | |
265 | bool new_i_size) | |
266 | { | |
267 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
268 | struct address_space *mapping = inode->i_mapping; | |
269 | struct page *page; | |
270 | struct dnode_of_data dn; | |
271 | int err; | |
272 | ||
273 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 274 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
eb47b800 JK |
275 | if (err) |
276 | return ERR_PTR(err); | |
277 | ||
278 | if (dn.data_blkaddr == NULL_ADDR) { | |
279 | if (reserve_new_block(&dn)) { | |
280 | f2fs_put_dnode(&dn); | |
281 | return ERR_PTR(-ENOSPC); | |
282 | } | |
283 | } | |
284 | f2fs_put_dnode(&dn); | |
285 | ||
286 | page = grab_cache_page(mapping, index); | |
287 | if (!page) | |
288 | return ERR_PTR(-ENOMEM); | |
289 | ||
290 | if (PageUptodate(page)) | |
291 | return page; | |
292 | ||
293 | if (dn.data_blkaddr == NEW_ADDR) { | |
294 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 295 | SetPageUptodate(page); |
eb47b800 JK |
296 | } else { |
297 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 298 | if (err) |
eb47b800 | 299 | return ERR_PTR(err); |
393ff91f JK |
300 | lock_page(page); |
301 | if (!PageUptodate(page)) { | |
302 | f2fs_put_page(page, 1); | |
303 | return ERR_PTR(-EIO); | |
eb47b800 JK |
304 | } |
305 | } | |
eb47b800 JK |
306 | |
307 | if (new_i_size && | |
308 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
309 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
310 | mark_inode_dirty_sync(inode); | |
311 | } | |
312 | return page; | |
313 | } | |
314 | ||
315 | static void read_end_io(struct bio *bio, int err) | |
316 | { | |
317 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
318 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
319 | ||
320 | do { | |
321 | struct page *page = bvec->bv_page; | |
322 | ||
323 | if (--bvec >= bio->bi_io_vec) | |
324 | prefetchw(&bvec->bv_page->flags); | |
325 | ||
326 | if (uptodate) { | |
327 | SetPageUptodate(page); | |
328 | } else { | |
329 | ClearPageUptodate(page); | |
330 | SetPageError(page); | |
331 | } | |
332 | unlock_page(page); | |
333 | } while (bvec >= bio->bi_io_vec); | |
334 | kfree(bio->bi_private); | |
335 | bio_put(bio); | |
336 | } | |
337 | ||
0a8165d7 | 338 | /* |
eb47b800 | 339 | * Fill the locked page with data located in the block address. |
393ff91f | 340 | * Return unlocked page. |
eb47b800 JK |
341 | */ |
342 | int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page, | |
343 | block_t blk_addr, int type) | |
344 | { | |
345 | struct block_device *bdev = sbi->sb->s_bdev; | |
eb47b800 JK |
346 | struct bio *bio; |
347 | ||
eb47b800 JK |
348 | down_read(&sbi->bio_sem); |
349 | ||
350 | /* Allocate a new bio */ | |
3cd8a239 | 351 | bio = f2fs_bio_alloc(bdev, 1); |
eb47b800 JK |
352 | |
353 | /* Initialize the bio */ | |
3cd8a239 | 354 | bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr); |
eb47b800 | 355 | bio->bi_end_io = read_end_io; |
3cd8a239 | 356 | |
eb47b800 JK |
357 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { |
358 | kfree(bio->bi_private); | |
359 | bio_put(bio); | |
360 | up_read(&sbi->bio_sem); | |
393ff91f | 361 | f2fs_put_page(page, 1); |
eb47b800 JK |
362 | return -EFAULT; |
363 | } | |
364 | ||
365 | submit_bio(type, bio); | |
366 | up_read(&sbi->bio_sem); | |
eb47b800 JK |
367 | return 0; |
368 | } | |
369 | ||
0a8165d7 | 370 | /* |
eb47b800 JK |
371 | * This function should be used by the data read flow only where it |
372 | * does not check the "create" flag that indicates block allocation. | |
373 | * The reason for this special functionality is to exploit VFS readahead | |
374 | * mechanism. | |
375 | */ | |
376 | static int get_data_block_ro(struct inode *inode, sector_t iblock, | |
377 | struct buffer_head *bh_result, int create) | |
378 | { | |
379 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
380 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
381 | struct dnode_of_data dn; | |
382 | pgoff_t pgofs; | |
383 | int err; | |
384 | ||
385 | /* Get the page offset from the block offset(iblock) */ | |
386 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
387 | ||
388 | if (check_extent_cache(inode, pgofs, bh_result)) | |
389 | return 0; | |
390 | ||
391 | /* When reading holes, we need its node page */ | |
392 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 393 | err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA); |
eb47b800 JK |
394 | if (err) |
395 | return (err == -ENOENT) ? 0 : err; | |
396 | ||
397 | /* It does not support data allocation */ | |
398 | BUG_ON(create); | |
399 | ||
400 | if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) { | |
401 | int i; | |
402 | unsigned int end_offset; | |
403 | ||
404 | end_offset = IS_INODE(dn.node_page) ? | |
405 | ADDRS_PER_INODE : | |
406 | ADDRS_PER_BLOCK; | |
407 | ||
408 | clear_buffer_new(bh_result); | |
409 | ||
410 | /* Give more consecutive addresses for the read ahead */ | |
411 | for (i = 0; i < end_offset - dn.ofs_in_node; i++) | |
412 | if (((datablock_addr(dn.node_page, | |
413 | dn.ofs_in_node + i)) | |
414 | != (dn.data_blkaddr + i)) || maxblocks == i) | |
415 | break; | |
416 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); | |
417 | bh_result->b_size = (i << blkbits); | |
418 | } | |
419 | f2fs_put_dnode(&dn); | |
420 | return 0; | |
421 | } | |
422 | ||
423 | static int f2fs_read_data_page(struct file *file, struct page *page) | |
424 | { | |
425 | return mpage_readpage(page, get_data_block_ro); | |
426 | } | |
427 | ||
428 | static int f2fs_read_data_pages(struct file *file, | |
429 | struct address_space *mapping, | |
430 | struct list_head *pages, unsigned nr_pages) | |
431 | { | |
432 | return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro); | |
433 | } | |
434 | ||
435 | int do_write_data_page(struct page *page) | |
436 | { | |
437 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
438 | block_t old_blk_addr, new_blk_addr; |
439 | struct dnode_of_data dn; | |
440 | int err = 0; | |
441 | ||
442 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 443 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
444 | if (err) |
445 | return err; | |
446 | ||
447 | old_blk_addr = dn.data_blkaddr; | |
448 | ||
449 | /* This page is already truncated */ | |
450 | if (old_blk_addr == NULL_ADDR) | |
451 | goto out_writepage; | |
452 | ||
453 | set_page_writeback(page); | |
454 | ||
455 | /* | |
456 | * If current allocation needs SSR, | |
457 | * it had better in-place writes for updated data. | |
458 | */ | |
459 | if (old_blk_addr != NEW_ADDR && !is_cold_data(page) && | |
460 | need_inplace_update(inode)) { | |
461 | rewrite_data_page(F2FS_SB(inode->i_sb), page, | |
462 | old_blk_addr); | |
463 | } else { | |
464 | write_data_page(inode, page, &dn, | |
465 | old_blk_addr, &new_blk_addr); | |
466 | update_extent_cache(new_blk_addr, &dn); | |
eb47b800 JK |
467 | } |
468 | out_writepage: | |
469 | f2fs_put_dnode(&dn); | |
470 | return err; | |
471 | } | |
472 | ||
473 | static int f2fs_write_data_page(struct page *page, | |
474 | struct writeback_control *wbc) | |
475 | { | |
476 | struct inode *inode = page->mapping->host; | |
477 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
478 | loff_t i_size = i_size_read(inode); | |
479 | const pgoff_t end_index = ((unsigned long long) i_size) | |
480 | >> PAGE_CACHE_SHIFT; | |
481 | unsigned offset; | |
482 | int err = 0; | |
483 | ||
484 | if (page->index < end_index) | |
485 | goto out; | |
486 | ||
487 | /* | |
488 | * If the offset is out-of-range of file size, | |
489 | * this page does not have to be written to disk. | |
490 | */ | |
491 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
492 | if ((page->index >= end_index + 1) || !offset) { | |
493 | if (S_ISDIR(inode->i_mode)) { | |
494 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
495 | inode_dec_dirty_dents(inode); | |
496 | } | |
497 | goto unlock_out; | |
498 | } | |
499 | ||
500 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
501 | out: | |
502 | if (sbi->por_doing) | |
503 | goto redirty_out; | |
504 | ||
505 | if (wbc->for_reclaim && !S_ISDIR(inode->i_mode) && !is_cold_data(page)) | |
506 | goto redirty_out; | |
507 | ||
508 | mutex_lock_op(sbi, DATA_WRITE); | |
509 | if (S_ISDIR(inode->i_mode)) { | |
510 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
511 | inode_dec_dirty_dents(inode); | |
512 | } | |
513 | err = do_write_data_page(page); | |
514 | if (err && err != -ENOENT) { | |
515 | wbc->pages_skipped++; | |
516 | set_page_dirty(page); | |
517 | } | |
518 | mutex_unlock_op(sbi, DATA_WRITE); | |
519 | ||
520 | if (wbc->for_reclaim) | |
521 | f2fs_submit_bio(sbi, DATA, true); | |
522 | ||
523 | if (err == -ENOENT) | |
524 | goto unlock_out; | |
525 | ||
526 | clear_cold_data(page); | |
527 | unlock_page(page); | |
528 | ||
529 | if (!wbc->for_reclaim && !S_ISDIR(inode->i_mode)) | |
530 | f2fs_balance_fs(sbi); | |
531 | return 0; | |
532 | ||
533 | unlock_out: | |
534 | unlock_page(page); | |
535 | return (err == -ENOENT) ? 0 : err; | |
536 | ||
537 | redirty_out: | |
538 | wbc->pages_skipped++; | |
539 | set_page_dirty(page); | |
540 | return AOP_WRITEPAGE_ACTIVATE; | |
541 | } | |
542 | ||
543 | #define MAX_DESIRED_PAGES_WP 4096 | |
544 | ||
fa9150a8 NJ |
545 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
546 | void *data) | |
547 | { | |
548 | struct address_space *mapping = data; | |
549 | int ret = mapping->a_ops->writepage(page, wbc); | |
550 | mapping_set_error(mapping, ret); | |
551 | return ret; | |
552 | } | |
553 | ||
25ca923b | 554 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
555 | struct writeback_control *wbc) |
556 | { | |
557 | struct inode *inode = mapping->host; | |
558 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
559 | int ret; | |
560 | long excess_nrtw = 0, desired_nrtw; | |
561 | ||
cfb185a1 | 562 | /* deal with chardevs and other special file */ |
563 | if (!mapping->a_ops->writepage) | |
564 | return 0; | |
565 | ||
eb47b800 JK |
566 | if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) { |
567 | desired_nrtw = MAX_DESIRED_PAGES_WP; | |
568 | excess_nrtw = desired_nrtw - wbc->nr_to_write; | |
569 | wbc->nr_to_write = desired_nrtw; | |
570 | } | |
571 | ||
572 | if (!S_ISDIR(inode->i_mode)) | |
573 | mutex_lock(&sbi->writepages); | |
fa9150a8 | 574 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
eb47b800 JK |
575 | if (!S_ISDIR(inode->i_mode)) |
576 | mutex_unlock(&sbi->writepages); | |
577 | f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL)); | |
578 | ||
579 | remove_dirty_dir_inode(inode); | |
580 | ||
581 | wbc->nr_to_write -= excess_nrtw; | |
582 | return ret; | |
583 | } | |
584 | ||
585 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, | |
586 | loff_t pos, unsigned len, unsigned flags, | |
587 | struct page **pagep, void **fsdata) | |
588 | { | |
589 | struct inode *inode = mapping->host; | |
590 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
591 | struct page *page; | |
592 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; | |
593 | struct dnode_of_data dn; | |
594 | int err = 0; | |
595 | ||
596 | /* for nobh_write_end */ | |
597 | *fsdata = NULL; | |
598 | ||
599 | f2fs_balance_fs(sbi); | |
600 | ||
601 | page = grab_cache_page_write_begin(mapping, index, flags); | |
602 | if (!page) | |
603 | return -ENOMEM; | |
604 | *pagep = page; | |
605 | ||
606 | mutex_lock_op(sbi, DATA_NEW); | |
607 | ||
608 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 609 | err = get_dnode_of_data(&dn, index, ALLOC_NODE); |
eb47b800 JK |
610 | if (err) { |
611 | mutex_unlock_op(sbi, DATA_NEW); | |
612 | f2fs_put_page(page, 1); | |
613 | return err; | |
614 | } | |
615 | ||
616 | if (dn.data_blkaddr == NULL_ADDR) { | |
617 | err = reserve_new_block(&dn); | |
618 | if (err) { | |
619 | f2fs_put_dnode(&dn); | |
620 | mutex_unlock_op(sbi, DATA_NEW); | |
621 | f2fs_put_page(page, 1); | |
622 | return err; | |
623 | } | |
624 | } | |
625 | f2fs_put_dnode(&dn); | |
626 | ||
627 | mutex_unlock_op(sbi, DATA_NEW); | |
628 | ||
629 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) | |
630 | return 0; | |
631 | ||
632 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { | |
633 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
634 | unsigned end = start + len; | |
635 | ||
636 | /* Reading beyond i_size is simple: memset to zero */ | |
637 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 638 | goto out; |
eb47b800 JK |
639 | } |
640 | ||
641 | if (dn.data_blkaddr == NEW_ADDR) { | |
642 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
643 | } else { | |
644 | err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC); | |
393ff91f | 645 | if (err) |
eb47b800 | 646 | return err; |
393ff91f JK |
647 | lock_page(page); |
648 | if (!PageUptodate(page)) { | |
649 | f2fs_put_page(page, 1); | |
650 | return -EIO; | |
eb47b800 JK |
651 | } |
652 | } | |
393ff91f | 653 | out: |
eb47b800 JK |
654 | SetPageUptodate(page); |
655 | clear_cold_data(page); | |
656 | return 0; | |
657 | } | |
658 | ||
659 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, | |
660 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
661 | { | |
662 | struct file *file = iocb->ki_filp; | |
663 | struct inode *inode = file->f_mapping->host; | |
664 | ||
665 | if (rw == WRITE) | |
666 | return 0; | |
667 | ||
668 | /* Needs synchronization with the cleaner */ | |
669 | return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, | |
670 | get_data_block_ro); | |
671 | } | |
672 | ||
673 | static void f2fs_invalidate_data_page(struct page *page, unsigned long offset) | |
674 | { | |
675 | struct inode *inode = page->mapping->host; | |
676 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
677 | if (S_ISDIR(inode->i_mode) && PageDirty(page)) { | |
678 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
679 | inode_dec_dirty_dents(inode); | |
680 | } | |
681 | ClearPagePrivate(page); | |
682 | } | |
683 | ||
684 | static int f2fs_release_data_page(struct page *page, gfp_t wait) | |
685 | { | |
686 | ClearPagePrivate(page); | |
c3850aa1 | 687 | return 1; |
eb47b800 JK |
688 | } |
689 | ||
690 | static int f2fs_set_data_page_dirty(struct page *page) | |
691 | { | |
692 | struct address_space *mapping = page->mapping; | |
693 | struct inode *inode = mapping->host; | |
694 | ||
695 | SetPageUptodate(page); | |
696 | if (!PageDirty(page)) { | |
697 | __set_page_dirty_nobuffers(page); | |
698 | set_dirty_dir_page(inode, page); | |
699 | return 1; | |
700 | } | |
701 | return 0; | |
702 | } | |
703 | ||
c01e54b7 JK |
704 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
705 | { | |
706 | return generic_block_bmap(mapping, block, get_data_block_ro); | |
707 | } | |
708 | ||
eb47b800 JK |
709 | const struct address_space_operations f2fs_dblock_aops = { |
710 | .readpage = f2fs_read_data_page, | |
711 | .readpages = f2fs_read_data_pages, | |
712 | .writepage = f2fs_write_data_page, | |
713 | .writepages = f2fs_write_data_pages, | |
714 | .write_begin = f2fs_write_begin, | |
715 | .write_end = nobh_write_end, | |
716 | .set_page_dirty = f2fs_set_data_page_dirty, | |
717 | .invalidatepage = f2fs_invalidate_data_page, | |
718 | .releasepage = f2fs_release_data_page, | |
719 | .direct_IO = f2fs_direct_IO, | |
c01e54b7 | 720 | .bmap = f2fs_bmap, |
eb47b800 | 721 | }; |