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