]>
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" | |
db9f7c1a | 25 | #include "trace.h" |
848753aa | 26 | #include <trace/events/f2fs.h> |
eb47b800 | 27 | |
429511cd CY |
28 | static struct kmem_cache *extent_tree_slab; |
29 | static struct kmem_cache *extent_node_slab; | |
30 | ||
93dfe2ac JK |
31 | static void f2fs_read_end_io(struct bio *bio, int err) |
32 | { | |
f568849e LT |
33 | struct bio_vec *bvec; |
34 | int i; | |
93dfe2ac | 35 | |
f568849e | 36 | bio_for_each_segment_all(bvec, bio, i) { |
93dfe2ac JK |
37 | struct page *page = bvec->bv_page; |
38 | ||
f568849e LT |
39 | if (!err) { |
40 | SetPageUptodate(page); | |
41 | } else { | |
93dfe2ac JK |
42 | ClearPageUptodate(page); |
43 | SetPageError(page); | |
44 | } | |
45 | unlock_page(page); | |
f568849e | 46 | } |
93dfe2ac JK |
47 | bio_put(bio); |
48 | } | |
49 | ||
50 | static void f2fs_write_end_io(struct bio *bio, int err) | |
51 | { | |
1b1f559f | 52 | struct f2fs_sb_info *sbi = bio->bi_private; |
f568849e LT |
53 | struct bio_vec *bvec; |
54 | int i; | |
93dfe2ac | 55 | |
f568849e | 56 | bio_for_each_segment_all(bvec, bio, i) { |
93dfe2ac JK |
57 | struct page *page = bvec->bv_page; |
58 | ||
f568849e | 59 | if (unlikely(err)) { |
cf779cab | 60 | set_page_dirty(page); |
93dfe2ac | 61 | set_bit(AS_EIO, &page->mapping->flags); |
744602cf | 62 | f2fs_stop_checkpoint(sbi); |
93dfe2ac JK |
63 | } |
64 | end_page_writeback(page); | |
65 | dec_page_count(sbi, F2FS_WRITEBACK); | |
f568849e | 66 | } |
93dfe2ac | 67 | |
93dfe2ac JK |
68 | if (!get_pages(sbi, F2FS_WRITEBACK) && |
69 | !list_empty(&sbi->cp_wait.task_list)) | |
70 | wake_up(&sbi->cp_wait); | |
71 | ||
72 | bio_put(bio); | |
73 | } | |
74 | ||
940a6d34 GZ |
75 | /* |
76 | * Low-level block read/write IO operations. | |
77 | */ | |
78 | static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr, | |
79 | int npages, bool is_read) | |
80 | { | |
81 | struct bio *bio; | |
82 | ||
83 | /* No failure on bio allocation */ | |
84 | bio = bio_alloc(GFP_NOIO, npages); | |
85 | ||
86 | bio->bi_bdev = sbi->sb->s_bdev; | |
55cf9cb6 | 87 | bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr); |
940a6d34 | 88 | bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io; |
1b1f559f | 89 | bio->bi_private = sbi; |
940a6d34 GZ |
90 | |
91 | return bio; | |
92 | } | |
93 | ||
458e6197 | 94 | static void __submit_merged_bio(struct f2fs_bio_info *io) |
93dfe2ac | 95 | { |
458e6197 | 96 | struct f2fs_io_info *fio = &io->fio; |
93dfe2ac JK |
97 | |
98 | if (!io->bio) | |
99 | return; | |
100 | ||
6a8f8ca5 | 101 | if (is_read_io(fio->rw)) |
2ace38e0 | 102 | trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio); |
6a8f8ca5 | 103 | else |
2ace38e0 | 104 | trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio); |
940a6d34 | 105 | |
6a8f8ca5 | 106 | submit_bio(fio->rw, io->bio); |
93dfe2ac JK |
107 | io->bio = NULL; |
108 | } | |
109 | ||
110 | void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, | |
458e6197 | 111 | enum page_type type, int rw) |
93dfe2ac JK |
112 | { |
113 | enum page_type btype = PAGE_TYPE_OF_BIO(type); | |
114 | struct f2fs_bio_info *io; | |
115 | ||
116 | io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype]; | |
117 | ||
df0f8dc0 | 118 | down_write(&io->io_rwsem); |
458e6197 JK |
119 | |
120 | /* change META to META_FLUSH in the checkpoint procedure */ | |
121 | if (type >= META_FLUSH) { | |
122 | io->fio.type = META_FLUSH; | |
0f7b2abd JK |
123 | if (test_opt(sbi, NOBARRIER)) |
124 | io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO; | |
125 | else | |
126 | io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO; | |
458e6197 JK |
127 | } |
128 | __submit_merged_bio(io); | |
df0f8dc0 | 129 | up_write(&io->io_rwsem); |
93dfe2ac JK |
130 | } |
131 | ||
132 | /* | |
133 | * Fill the locked page with data located in the block address. | |
134 | * Return unlocked page. | |
135 | */ | |
136 | int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page, | |
cf04e8eb | 137 | struct f2fs_io_info *fio) |
93dfe2ac | 138 | { |
93dfe2ac JK |
139 | struct bio *bio; |
140 | ||
2ace38e0 | 141 | trace_f2fs_submit_page_bio(page, fio); |
db9f7c1a | 142 | f2fs_trace_ios(page, fio, 0); |
93dfe2ac JK |
143 | |
144 | /* Allocate a new bio */ | |
cf04e8eb | 145 | bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw)); |
93dfe2ac JK |
146 | |
147 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { | |
148 | bio_put(bio); | |
149 | f2fs_put_page(page, 1); | |
150 | return -EFAULT; | |
151 | } | |
152 | ||
cf04e8eb | 153 | submit_bio(fio->rw, bio); |
93dfe2ac JK |
154 | return 0; |
155 | } | |
156 | ||
157 | void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page, | |
cf04e8eb | 158 | struct f2fs_io_info *fio) |
93dfe2ac | 159 | { |
458e6197 | 160 | enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); |
93dfe2ac | 161 | struct f2fs_bio_info *io; |
940a6d34 | 162 | bool is_read = is_read_io(fio->rw); |
93dfe2ac | 163 | |
940a6d34 | 164 | io = is_read ? &sbi->read_io : &sbi->write_io[btype]; |
93dfe2ac | 165 | |
cf04e8eb | 166 | verify_block_addr(sbi, fio->blk_addr); |
93dfe2ac | 167 | |
df0f8dc0 | 168 | down_write(&io->io_rwsem); |
93dfe2ac | 169 | |
940a6d34 | 170 | if (!is_read) |
93dfe2ac JK |
171 | inc_page_count(sbi, F2FS_WRITEBACK); |
172 | ||
cf04e8eb | 173 | if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 || |
458e6197 JK |
174 | io->fio.rw != fio->rw)) |
175 | __submit_merged_bio(io); | |
93dfe2ac JK |
176 | alloc_new: |
177 | if (io->bio == NULL) { | |
90a893c7 | 178 | int bio_blocks = MAX_BIO_BLOCKS(sbi); |
940a6d34 | 179 | |
cf04e8eb | 180 | io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read); |
458e6197 | 181 | io->fio = *fio; |
93dfe2ac JK |
182 | } |
183 | ||
184 | if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) < | |
185 | PAGE_CACHE_SIZE) { | |
458e6197 | 186 | __submit_merged_bio(io); |
93dfe2ac JK |
187 | goto alloc_new; |
188 | } | |
189 | ||
cf04e8eb | 190 | io->last_block_in_bio = fio->blk_addr; |
db9f7c1a | 191 | f2fs_trace_ios(page, fio, 0); |
93dfe2ac | 192 | |
df0f8dc0 | 193 | up_write(&io->io_rwsem); |
2ace38e0 | 194 | trace_f2fs_submit_page_mbio(page, fio); |
93dfe2ac JK |
195 | } |
196 | ||
0a8165d7 | 197 | /* |
eb47b800 JK |
198 | * Lock ordering for the change of data block address: |
199 | * ->data_page | |
200 | * ->node_page | |
201 | * update block addresses in the node page | |
202 | */ | |
e1509cf2 | 203 | static void __set_data_blkaddr(struct dnode_of_data *dn) |
eb47b800 JK |
204 | { |
205 | struct f2fs_node *rn; | |
206 | __le32 *addr_array; | |
207 | struct page *node_page = dn->node_page; | |
208 | unsigned int ofs_in_node = dn->ofs_in_node; | |
209 | ||
5514f0aa | 210 | f2fs_wait_on_page_writeback(node_page, NODE); |
eb47b800 | 211 | |
45590710 | 212 | rn = F2FS_NODE(node_page); |
eb47b800 JK |
213 | |
214 | /* Get physical address of data block */ | |
215 | addr_array = blkaddr_in_node(rn); | |
e1509cf2 | 216 | addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr); |
eb47b800 JK |
217 | set_page_dirty(node_page); |
218 | } | |
219 | ||
220 | int reserve_new_block(struct dnode_of_data *dn) | |
221 | { | |
4081363f | 222 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
eb47b800 | 223 | |
6bacf52f | 224 | if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))) |
eb47b800 | 225 | return -EPERM; |
cfb271d4 | 226 | if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1))) |
eb47b800 JK |
227 | return -ENOSPC; |
228 | ||
c01e2853 NJ |
229 | trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node); |
230 | ||
eb47b800 | 231 | dn->data_blkaddr = NEW_ADDR; |
e1509cf2 | 232 | __set_data_blkaddr(dn); |
a18ff063 | 233 | mark_inode_dirty(dn->inode); |
eb47b800 JK |
234 | sync_inode_page(dn); |
235 | return 0; | |
236 | } | |
237 | ||
b600965c HL |
238 | int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) |
239 | { | |
240 | bool need_put = dn->inode_page ? false : true; | |
241 | int err; | |
242 | ||
243 | err = get_dnode_of_data(dn, index, ALLOC_NODE); | |
244 | if (err) | |
245 | return err; | |
a8865372 | 246 | |
b600965c HL |
247 | if (dn->data_blkaddr == NULL_ADDR) |
248 | err = reserve_new_block(dn); | |
a8865372 | 249 | if (err || need_put) |
b600965c HL |
250 | f2fs_put_dnode(dn); |
251 | return err; | |
252 | } | |
253 | ||
a2e7d1bf CY |
254 | static void f2fs_map_bh(struct super_block *sb, pgoff_t pgofs, |
255 | struct extent_info *ei, struct buffer_head *bh_result) | |
256 | { | |
257 | unsigned int blkbits = sb->s_blocksize_bits; | |
258 | size_t count; | |
259 | ||
260 | set_buffer_new(bh_result); | |
261 | map_bh(bh_result, sb, ei->blk + pgofs - ei->fofs); | |
262 | count = ei->fofs + ei->len - pgofs; | |
263 | if (count < (UINT_MAX >> blkbits)) | |
264 | bh_result->b_size = (count << blkbits); | |
265 | else | |
266 | bh_result->b_size = UINT_MAX; | |
267 | } | |
268 | ||
7e4dde79 CY |
269 | static bool lookup_extent_info(struct inode *inode, pgoff_t pgofs, |
270 | struct extent_info *ei) | |
eb47b800 JK |
271 | { |
272 | struct f2fs_inode_info *fi = F2FS_I(inode); | |
eb47b800 JK |
273 | pgoff_t start_fofs, end_fofs; |
274 | block_t start_blkaddr; | |
275 | ||
c11abd1a | 276 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
7e4dde79 | 277 | return false; |
c11abd1a | 278 | |
0c872e2d | 279 | read_lock(&fi->ext_lock); |
eb47b800 | 280 | if (fi->ext.len == 0) { |
0c872e2d | 281 | read_unlock(&fi->ext_lock); |
7e4dde79 | 282 | return false; |
eb47b800 JK |
283 | } |
284 | ||
dcdfff65 JK |
285 | stat_inc_total_hit(inode->i_sb); |
286 | ||
eb47b800 JK |
287 | start_fofs = fi->ext.fofs; |
288 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
4d0b0bd4 | 289 | start_blkaddr = fi->ext.blk; |
eb47b800 JK |
290 | |
291 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
a2e7d1bf | 292 | *ei = fi->ext; |
dcdfff65 | 293 | stat_inc_read_hit(inode->i_sb); |
0c872e2d | 294 | read_unlock(&fi->ext_lock); |
7e4dde79 | 295 | return true; |
eb47b800 | 296 | } |
0c872e2d | 297 | read_unlock(&fi->ext_lock); |
7e4dde79 | 298 | return false; |
eb47b800 JK |
299 | } |
300 | ||
7e4dde79 CY |
301 | static bool update_extent_info(struct inode *inode, pgoff_t fofs, |
302 | block_t blkaddr) | |
eb47b800 | 303 | { |
7e4dde79 CY |
304 | struct f2fs_inode_info *fi = F2FS_I(inode); |
305 | pgoff_t start_fofs, end_fofs; | |
eb47b800 | 306 | block_t start_blkaddr, end_blkaddr; |
c11abd1a | 307 | int need_update = true; |
eb47b800 | 308 | |
c11abd1a | 309 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
7e4dde79 | 310 | return false; |
3547ea96 | 311 | |
0c872e2d | 312 | write_lock(&fi->ext_lock); |
eb47b800 JK |
313 | |
314 | start_fofs = fi->ext.fofs; | |
315 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
4d0b0bd4 CY |
316 | start_blkaddr = fi->ext.blk; |
317 | end_blkaddr = fi->ext.blk + fi->ext.len - 1; | |
eb47b800 JK |
318 | |
319 | /* Drop and initialize the matched extent */ | |
320 | if (fi->ext.len == 1 && fofs == start_fofs) | |
321 | fi->ext.len = 0; | |
322 | ||
323 | /* Initial extent */ | |
324 | if (fi->ext.len == 0) { | |
7e4dde79 | 325 | if (blkaddr != NULL_ADDR) { |
eb47b800 | 326 | fi->ext.fofs = fofs; |
7e4dde79 | 327 | fi->ext.blk = blkaddr; |
eb47b800 JK |
328 | fi->ext.len = 1; |
329 | } | |
330 | goto end_update; | |
331 | } | |
332 | ||
6224da87 | 333 | /* Front merge */ |
7e4dde79 | 334 | if (fofs == start_fofs - 1 && blkaddr == start_blkaddr - 1) { |
eb47b800 | 335 | fi->ext.fofs--; |
4d0b0bd4 | 336 | fi->ext.blk--; |
eb47b800 JK |
337 | fi->ext.len++; |
338 | goto end_update; | |
339 | } | |
340 | ||
341 | /* Back merge */ | |
7e4dde79 | 342 | if (fofs == end_fofs + 1 && blkaddr == end_blkaddr + 1) { |
eb47b800 JK |
343 | fi->ext.len++; |
344 | goto end_update; | |
345 | } | |
346 | ||
347 | /* Split the existing extent */ | |
348 | if (fi->ext.len > 1 && | |
349 | fofs >= start_fofs && fofs <= end_fofs) { | |
350 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
351 | fi->ext.len = fofs - start_fofs; | |
352 | } else { | |
353 | fi->ext.fofs = fofs + 1; | |
4d0b0bd4 | 354 | fi->ext.blk = start_blkaddr + fofs - start_fofs + 1; |
eb47b800 JK |
355 | fi->ext.len -= fofs - start_fofs + 1; |
356 | } | |
c11abd1a JK |
357 | } else { |
358 | need_update = false; | |
eb47b800 | 359 | } |
eb47b800 | 360 | |
c11abd1a JK |
361 | /* Finally, if the extent is very fragmented, let's drop the cache. */ |
362 | if (fi->ext.len < F2FS_MIN_EXTENT_LEN) { | |
363 | fi->ext.len = 0; | |
364 | set_inode_flag(fi, FI_NO_EXTENT); | |
365 | need_update = true; | |
366 | } | |
eb47b800 | 367 | end_update: |
0c872e2d | 368 | write_unlock(&fi->ext_lock); |
7e4dde79 CY |
369 | return need_update; |
370 | } | |
371 | ||
429511cd CY |
372 | static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi, |
373 | struct extent_tree *et, struct extent_info *ei, | |
374 | struct rb_node *parent, struct rb_node **p) | |
375 | { | |
376 | struct extent_node *en; | |
377 | ||
378 | en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC); | |
379 | if (!en) | |
380 | return NULL; | |
381 | ||
382 | en->ei = *ei; | |
383 | INIT_LIST_HEAD(&en->list); | |
384 | ||
385 | rb_link_node(&en->rb_node, parent, p); | |
386 | rb_insert_color(&en->rb_node, &et->root); | |
387 | et->count++; | |
388 | atomic_inc(&sbi->total_ext_node); | |
389 | return en; | |
390 | } | |
391 | ||
392 | static void __detach_extent_node(struct f2fs_sb_info *sbi, | |
393 | struct extent_tree *et, struct extent_node *en) | |
394 | { | |
395 | rb_erase(&en->rb_node, &et->root); | |
396 | et->count--; | |
397 | atomic_dec(&sbi->total_ext_node); | |
62c8af65 CY |
398 | |
399 | if (et->cached_en == en) | |
400 | et->cached_en = NULL; | |
429511cd CY |
401 | } |
402 | ||
403 | static struct extent_node *__lookup_extent_tree(struct extent_tree *et, | |
404 | unsigned int fofs) | |
405 | { | |
406 | struct rb_node *node = et->root.rb_node; | |
407 | struct extent_node *en; | |
408 | ||
62c8af65 CY |
409 | if (et->cached_en) { |
410 | struct extent_info *cei = &et->cached_en->ei; | |
411 | ||
412 | if (cei->fofs <= fofs && cei->fofs + cei->len > fofs) | |
413 | return et->cached_en; | |
414 | } | |
415 | ||
429511cd CY |
416 | while (node) { |
417 | en = rb_entry(node, struct extent_node, rb_node); | |
418 | ||
62c8af65 | 419 | if (fofs < en->ei.fofs) { |
429511cd | 420 | node = node->rb_left; |
62c8af65 | 421 | } else if (fofs >= en->ei.fofs + en->ei.len) { |
429511cd | 422 | node = node->rb_right; |
62c8af65 CY |
423 | } else { |
424 | et->cached_en = en; | |
429511cd | 425 | return en; |
62c8af65 | 426 | } |
429511cd CY |
427 | } |
428 | return NULL; | |
429 | } | |
430 | ||
431 | static struct extent_node *__try_back_merge(struct f2fs_sb_info *sbi, | |
432 | struct extent_tree *et, struct extent_node *en) | |
433 | { | |
434 | struct extent_node *prev; | |
435 | struct rb_node *node; | |
436 | ||
437 | node = rb_prev(&en->rb_node); | |
438 | if (!node) | |
439 | return NULL; | |
440 | ||
441 | prev = rb_entry(node, struct extent_node, rb_node); | |
442 | if (__is_back_mergeable(&en->ei, &prev->ei)) { | |
443 | en->ei.fofs = prev->ei.fofs; | |
444 | en->ei.blk = prev->ei.blk; | |
445 | en->ei.len += prev->ei.len; | |
446 | __detach_extent_node(sbi, et, prev); | |
447 | return prev; | |
448 | } | |
449 | return NULL; | |
450 | } | |
451 | ||
452 | static struct extent_node *__try_front_merge(struct f2fs_sb_info *sbi, | |
453 | struct extent_tree *et, struct extent_node *en) | |
454 | { | |
455 | struct extent_node *next; | |
456 | struct rb_node *node; | |
457 | ||
458 | node = rb_next(&en->rb_node); | |
459 | if (!node) | |
460 | return NULL; | |
461 | ||
462 | next = rb_entry(node, struct extent_node, rb_node); | |
463 | if (__is_front_mergeable(&en->ei, &next->ei)) { | |
464 | en->ei.len += next->ei.len; | |
465 | __detach_extent_node(sbi, et, next); | |
466 | return next; | |
467 | } | |
468 | return NULL; | |
469 | } | |
470 | ||
471 | static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi, | |
472 | struct extent_tree *et, struct extent_info *ei, | |
473 | struct extent_node **den) | |
474 | { | |
475 | struct rb_node **p = &et->root.rb_node; | |
476 | struct rb_node *parent = NULL; | |
477 | struct extent_node *en; | |
478 | ||
479 | while (*p) { | |
480 | parent = *p; | |
481 | en = rb_entry(parent, struct extent_node, rb_node); | |
482 | ||
483 | if (ei->fofs < en->ei.fofs) { | |
484 | if (__is_front_mergeable(ei, &en->ei)) { | |
485 | f2fs_bug_on(sbi, !den); | |
486 | en->ei.fofs = ei->fofs; | |
487 | en->ei.blk = ei->blk; | |
488 | en->ei.len += ei->len; | |
489 | *den = __try_back_merge(sbi, et, en); | |
490 | return en; | |
491 | } | |
492 | p = &(*p)->rb_left; | |
493 | } else if (ei->fofs >= en->ei.fofs + en->ei.len) { | |
494 | if (__is_back_mergeable(ei, &en->ei)) { | |
495 | f2fs_bug_on(sbi, !den); | |
496 | en->ei.len += ei->len; | |
497 | *den = __try_front_merge(sbi, et, en); | |
498 | return en; | |
499 | } | |
500 | p = &(*p)->rb_right; | |
501 | } else { | |
502 | f2fs_bug_on(sbi, 1); | |
503 | } | |
504 | } | |
505 | ||
506 | return __attach_extent_node(sbi, et, ei, parent, p); | |
507 | } | |
508 | ||
509 | static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi, | |
510 | struct extent_tree *et, bool free_all) | |
511 | { | |
512 | struct rb_node *node, *next; | |
513 | struct extent_node *en; | |
514 | unsigned int count = et->count; | |
515 | ||
516 | node = rb_first(&et->root); | |
517 | while (node) { | |
518 | next = rb_next(node); | |
519 | en = rb_entry(node, struct extent_node, rb_node); | |
520 | ||
521 | if (free_all) { | |
522 | spin_lock(&sbi->extent_lock); | |
523 | if (!list_empty(&en->list)) | |
524 | list_del_init(&en->list); | |
525 | spin_unlock(&sbi->extent_lock); | |
526 | } | |
527 | ||
528 | if (free_all || list_empty(&en->list)) { | |
529 | __detach_extent_node(sbi, et, en); | |
530 | kmem_cache_free(extent_node_slab, en); | |
531 | } | |
532 | node = next; | |
533 | } | |
534 | ||
535 | return count - et->count; | |
536 | } | |
537 | ||
538 | static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs, | |
539 | struct extent_info *ei) | |
540 | { | |
541 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
542 | struct extent_tree *et; | |
543 | struct extent_node *en; | |
544 | ||
545 | if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT)) | |
546 | return false; | |
547 | ||
1ec4610c CY |
548 | trace_f2fs_lookup_extent_tree_start(inode, pgofs); |
549 | ||
429511cd CY |
550 | down_read(&sbi->extent_tree_lock); |
551 | et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino); | |
552 | if (!et) { | |
553 | up_read(&sbi->extent_tree_lock); | |
554 | return false; | |
555 | } | |
556 | atomic_inc(&et->refcount); | |
557 | up_read(&sbi->extent_tree_lock); | |
558 | ||
559 | read_lock(&et->lock); | |
560 | en = __lookup_extent_tree(et, pgofs); | |
561 | if (en) { | |
562 | *ei = en->ei; | |
563 | spin_lock(&sbi->extent_lock); | |
564 | if (!list_empty(&en->list)) | |
565 | list_move_tail(&en->list, &sbi->extent_list); | |
566 | spin_unlock(&sbi->extent_lock); | |
567 | stat_inc_read_hit(sbi->sb); | |
568 | } | |
569 | stat_inc_total_hit(sbi->sb); | |
570 | read_unlock(&et->lock); | |
571 | ||
1ec4610c CY |
572 | trace_f2fs_lookup_extent_tree_end(inode, pgofs, en); |
573 | ||
429511cd CY |
574 | atomic_dec(&et->refcount); |
575 | return en ? true : false; | |
576 | } | |
577 | ||
578 | static void f2fs_update_extent_tree(struct inode *inode, pgoff_t fofs, | |
579 | block_t blkaddr) | |
580 | { | |
581 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
582 | nid_t ino = inode->i_ino; | |
583 | struct extent_tree *et; | |
584 | struct extent_node *en = NULL, *en1 = NULL, *en2 = NULL, *en3 = NULL; | |
585 | struct extent_node *den = NULL; | |
586 | struct extent_info ei, dei; | |
587 | unsigned int endofs; | |
588 | ||
589 | if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT)) | |
590 | return; | |
591 | ||
1ec4610c CY |
592 | trace_f2fs_update_extent_tree(inode, fofs, blkaddr); |
593 | ||
429511cd CY |
594 | down_write(&sbi->extent_tree_lock); |
595 | et = radix_tree_lookup(&sbi->extent_tree_root, ino); | |
596 | if (!et) { | |
597 | et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS); | |
598 | f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et); | |
599 | memset(et, 0, sizeof(struct extent_tree)); | |
600 | et->ino = ino; | |
601 | et->root = RB_ROOT; | |
62c8af65 | 602 | et->cached_en = NULL; |
429511cd CY |
603 | rwlock_init(&et->lock); |
604 | atomic_set(&et->refcount, 0); | |
605 | et->count = 0; | |
606 | sbi->total_ext_tree++; | |
607 | } | |
608 | atomic_inc(&et->refcount); | |
609 | up_write(&sbi->extent_tree_lock); | |
610 | ||
611 | write_lock(&et->lock); | |
612 | ||
613 | /* 1. lookup and remove existing extent info in cache */ | |
614 | en = __lookup_extent_tree(et, fofs); | |
615 | if (!en) | |
616 | goto update_extent; | |
617 | ||
618 | dei = en->ei; | |
619 | __detach_extent_node(sbi, et, en); | |
620 | ||
621 | /* 2. if extent can be split more, split and insert the left part */ | |
622 | if (dei.len > 1) { | |
623 | /* insert left part of split extent into cache */ | |
624 | if (fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN) { | |
625 | set_extent_info(&ei, dei.fofs, dei.blk, | |
626 | fofs - dei.fofs); | |
627 | en1 = __insert_extent_tree(sbi, et, &ei, NULL); | |
628 | } | |
629 | ||
630 | /* insert right part of split extent into cache */ | |
631 | endofs = dei.fofs + dei.len - 1; | |
632 | if (endofs - fofs >= F2FS_MIN_EXTENT_LEN) { | |
633 | set_extent_info(&ei, fofs + 1, | |
634 | fofs - dei.fofs + dei.blk, endofs - fofs); | |
635 | en2 = __insert_extent_tree(sbi, et, &ei, NULL); | |
636 | } | |
637 | } | |
638 | ||
639 | update_extent: | |
640 | /* 3. update extent in extent cache */ | |
641 | if (blkaddr) { | |
642 | set_extent_info(&ei, fofs, blkaddr, 1); | |
643 | en3 = __insert_extent_tree(sbi, et, &ei, &den); | |
644 | } | |
645 | ||
646 | /* 4. update in global extent list */ | |
647 | spin_lock(&sbi->extent_lock); | |
648 | if (en && !list_empty(&en->list)) | |
649 | list_del(&en->list); | |
650 | /* | |
651 | * en1 and en2 split from en, they will become more and more smaller | |
652 | * fragments after splitting several times. So if the length is smaller | |
653 | * than F2FS_MIN_EXTENT_LEN, we will not add them into extent tree. | |
654 | */ | |
655 | if (en1) | |
656 | list_add_tail(&en1->list, &sbi->extent_list); | |
657 | if (en2) | |
658 | list_add_tail(&en2->list, &sbi->extent_list); | |
659 | if (en3) { | |
660 | if (list_empty(&en3->list)) | |
661 | list_add_tail(&en3->list, &sbi->extent_list); | |
662 | else | |
663 | list_move_tail(&en3->list, &sbi->extent_list); | |
664 | } | |
665 | if (den && !list_empty(&den->list)) | |
666 | list_del(&den->list); | |
667 | spin_unlock(&sbi->extent_lock); | |
668 | ||
669 | /* 5. release extent node */ | |
670 | if (en) | |
671 | kmem_cache_free(extent_node_slab, en); | |
672 | if (den) | |
673 | kmem_cache_free(extent_node_slab, den); | |
674 | ||
675 | write_unlock(&et->lock); | |
676 | atomic_dec(&et->refcount); | |
677 | } | |
678 | ||
679 | void f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink) | |
680 | { | |
681 | struct extent_tree *treevec[EXT_TREE_VEC_SIZE]; | |
682 | struct extent_node *en, *tmp; | |
683 | unsigned long ino = F2FS_ROOT_INO(sbi); | |
684 | struct radix_tree_iter iter; | |
685 | void **slot; | |
686 | unsigned int found; | |
1ec4610c | 687 | unsigned int node_cnt = 0, tree_cnt = 0; |
429511cd | 688 | |
1dcc336b CY |
689 | if (!test_opt(sbi, EXTENT_CACHE)) |
690 | return; | |
691 | ||
429511cd CY |
692 | if (available_free_memory(sbi, EXTENT_CACHE)) |
693 | return; | |
694 | ||
695 | spin_lock(&sbi->extent_lock); | |
696 | list_for_each_entry_safe(en, tmp, &sbi->extent_list, list) { | |
697 | if (!nr_shrink--) | |
698 | break; | |
699 | list_del_init(&en->list); | |
700 | } | |
701 | spin_unlock(&sbi->extent_lock); | |
702 | ||
703 | down_read(&sbi->extent_tree_lock); | |
704 | while ((found = radix_tree_gang_lookup(&sbi->extent_tree_root, | |
705 | (void **)treevec, ino, EXT_TREE_VEC_SIZE))) { | |
706 | unsigned i; | |
707 | ||
708 | ino = treevec[found - 1]->ino + 1; | |
709 | for (i = 0; i < found; i++) { | |
710 | struct extent_tree *et = treevec[i]; | |
711 | ||
712 | atomic_inc(&et->refcount); | |
713 | write_lock(&et->lock); | |
1ec4610c | 714 | node_cnt += __free_extent_tree(sbi, et, false); |
429511cd CY |
715 | write_unlock(&et->lock); |
716 | atomic_dec(&et->refcount); | |
717 | } | |
718 | } | |
719 | up_read(&sbi->extent_tree_lock); | |
720 | ||
721 | down_write(&sbi->extent_tree_lock); | |
722 | radix_tree_for_each_slot(slot, &sbi->extent_tree_root, &iter, | |
723 | F2FS_ROOT_INO(sbi)) { | |
724 | struct extent_tree *et = (struct extent_tree *)*slot; | |
725 | ||
726 | if (!atomic_read(&et->refcount) && !et->count) { | |
727 | radix_tree_delete(&sbi->extent_tree_root, et->ino); | |
728 | kmem_cache_free(extent_tree_slab, et); | |
729 | sbi->total_ext_tree--; | |
1ec4610c | 730 | tree_cnt++; |
429511cd CY |
731 | } |
732 | } | |
733 | up_write(&sbi->extent_tree_lock); | |
1ec4610c CY |
734 | |
735 | trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt); | |
429511cd CY |
736 | } |
737 | ||
738 | void f2fs_destroy_extent_tree(struct inode *inode) | |
739 | { | |
740 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
741 | struct extent_tree *et; | |
1ec4610c | 742 | unsigned int node_cnt = 0; |
429511cd | 743 | |
1dcc336b CY |
744 | if (!test_opt(sbi, EXTENT_CACHE)) |
745 | return; | |
746 | ||
429511cd CY |
747 | down_read(&sbi->extent_tree_lock); |
748 | et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino); | |
749 | if (!et) { | |
750 | up_read(&sbi->extent_tree_lock); | |
751 | goto out; | |
752 | } | |
753 | atomic_inc(&et->refcount); | |
754 | up_read(&sbi->extent_tree_lock); | |
755 | ||
756 | /* free all extent info belong to this extent tree */ | |
757 | write_lock(&et->lock); | |
1ec4610c | 758 | node_cnt = __free_extent_tree(sbi, et, true); |
429511cd CY |
759 | write_unlock(&et->lock); |
760 | ||
761 | atomic_dec(&et->refcount); | |
762 | ||
763 | /* try to find and delete extent tree entry in radix tree */ | |
764 | down_write(&sbi->extent_tree_lock); | |
765 | et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino); | |
766 | if (!et) { | |
767 | up_write(&sbi->extent_tree_lock); | |
768 | goto out; | |
769 | } | |
770 | f2fs_bug_on(sbi, atomic_read(&et->refcount) || et->count); | |
771 | radix_tree_delete(&sbi->extent_tree_root, inode->i_ino); | |
772 | kmem_cache_free(extent_tree_slab, et); | |
773 | sbi->total_ext_tree--; | |
774 | up_write(&sbi->extent_tree_lock); | |
775 | out: | |
1ec4610c | 776 | trace_f2fs_destroy_extent_tree(inode, node_cnt); |
429511cd CY |
777 | return; |
778 | } | |
779 | ||
7e4dde79 CY |
780 | static bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs, |
781 | struct extent_info *ei) | |
782 | { | |
1dcc336b CY |
783 | if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE)) |
784 | return f2fs_lookup_extent_tree(inode, pgofs, ei); | |
785 | ||
7e4dde79 CY |
786 | return lookup_extent_info(inode, pgofs, ei); |
787 | } | |
788 | ||
789 | void f2fs_update_extent_cache(struct dnode_of_data *dn) | |
790 | { | |
791 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
792 | pgoff_t fofs; | |
793 | ||
794 | f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR); | |
795 | ||
796 | /* Update the page address in the parent node */ | |
797 | __set_data_blkaddr(dn); | |
798 | ||
799 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + | |
800 | dn->ofs_in_node; | |
801 | ||
1dcc336b CY |
802 | if (test_opt(F2FS_I_SB(dn->inode), EXTENT_CACHE)) |
803 | return f2fs_update_extent_tree(dn->inode, fofs, | |
804 | dn->data_blkaddr); | |
805 | ||
7e4dde79 | 806 | if (update_extent_info(dn->inode, fofs, dn->data_blkaddr)) |
c11abd1a | 807 | sync_inode_page(dn); |
eb47b800 JK |
808 | } |
809 | ||
c718379b | 810 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 | 811 | { |
eb47b800 JK |
812 | struct address_space *mapping = inode->i_mapping; |
813 | struct dnode_of_data dn; | |
814 | struct page *page; | |
815 | int err; | |
cf04e8eb JK |
816 | struct f2fs_io_info fio = { |
817 | .type = DATA, | |
818 | .rw = sync ? READ_SYNC : READA, | |
819 | }; | |
eb47b800 JK |
820 | |
821 | page = find_get_page(mapping, index); | |
822 | if (page && PageUptodate(page)) | |
823 | return page; | |
824 | f2fs_put_page(page, 0); | |
825 | ||
826 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 827 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
828 | if (err) |
829 | return ERR_PTR(err); | |
830 | f2fs_put_dnode(&dn); | |
831 | ||
832 | if (dn.data_blkaddr == NULL_ADDR) | |
833 | return ERR_PTR(-ENOENT); | |
834 | ||
835 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
6bacf52f | 836 | if (unlikely(dn.data_blkaddr == NEW_ADDR)) |
eb47b800 JK |
837 | return ERR_PTR(-EINVAL); |
838 | ||
9ac1349a | 839 | page = grab_cache_page(mapping, index); |
eb47b800 JK |
840 | if (!page) |
841 | return ERR_PTR(-ENOMEM); | |
842 | ||
393ff91f JK |
843 | if (PageUptodate(page)) { |
844 | unlock_page(page); | |
845 | return page; | |
846 | } | |
847 | ||
cf04e8eb JK |
848 | fio.blk_addr = dn.data_blkaddr; |
849 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
1069bbf7 CY |
850 | if (err) |
851 | return ERR_PTR(err); | |
852 | ||
c718379b JK |
853 | if (sync) { |
854 | wait_on_page_locked(page); | |
6bacf52f | 855 | if (unlikely(!PageUptodate(page))) { |
c718379b JK |
856 | f2fs_put_page(page, 0); |
857 | return ERR_PTR(-EIO); | |
858 | } | |
eb47b800 | 859 | } |
eb47b800 JK |
860 | return page; |
861 | } | |
862 | ||
0a8165d7 | 863 | /* |
eb47b800 JK |
864 | * If it tries to access a hole, return an error. |
865 | * Because, the callers, functions in dir.c and GC, should be able to know | |
866 | * whether this page exists or not. | |
867 | */ | |
868 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
869 | { | |
eb47b800 JK |
870 | struct address_space *mapping = inode->i_mapping; |
871 | struct dnode_of_data dn; | |
872 | struct page *page; | |
873 | int err; | |
cf04e8eb JK |
874 | struct f2fs_io_info fio = { |
875 | .type = DATA, | |
876 | .rw = READ_SYNC, | |
877 | }; | |
650495de | 878 | repeat: |
9ac1349a | 879 | page = grab_cache_page(mapping, index); |
650495de JK |
880 | if (!page) |
881 | return ERR_PTR(-ENOMEM); | |
882 | ||
eb47b800 | 883 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 884 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
885 | if (err) { |
886 | f2fs_put_page(page, 1); | |
eb47b800 | 887 | return ERR_PTR(err); |
650495de | 888 | } |
eb47b800 JK |
889 | f2fs_put_dnode(&dn); |
890 | ||
6bacf52f | 891 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
650495de | 892 | f2fs_put_page(page, 1); |
eb47b800 | 893 | return ERR_PTR(-ENOENT); |
650495de | 894 | } |
eb47b800 JK |
895 | |
896 | if (PageUptodate(page)) | |
897 | return page; | |
898 | ||
d59ff4df JK |
899 | /* |
900 | * A new dentry page is allocated but not able to be written, since its | |
901 | * new inode page couldn't be allocated due to -ENOSPC. | |
902 | * In such the case, its blkaddr can be remained as NEW_ADDR. | |
903 | * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. | |
904 | */ | |
905 | if (dn.data_blkaddr == NEW_ADDR) { | |
906 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
907 | SetPageUptodate(page); | |
908 | return page; | |
909 | } | |
eb47b800 | 910 | |
cf04e8eb JK |
911 | fio.blk_addr = dn.data_blkaddr; |
912 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
393ff91f | 913 | if (err) |
eb47b800 | 914 | return ERR_PTR(err); |
393ff91f JK |
915 | |
916 | lock_page(page); | |
6bacf52f | 917 | if (unlikely(!PageUptodate(page))) { |
393ff91f JK |
918 | f2fs_put_page(page, 1); |
919 | return ERR_PTR(-EIO); | |
eb47b800 | 920 | } |
6bacf52f | 921 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
922 | f2fs_put_page(page, 1); |
923 | goto repeat; | |
eb47b800 JK |
924 | } |
925 | return page; | |
926 | } | |
927 | ||
0a8165d7 | 928 | /* |
eb47b800 JK |
929 | * Caller ensures that this data page is never allocated. |
930 | * A new zero-filled data page is allocated in the page cache. | |
39936837 | 931 | * |
4f4124d0 CY |
932 | * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and |
933 | * f2fs_unlock_op(). | |
a8865372 | 934 | * Note that, ipage is set only by make_empty_dir. |
eb47b800 | 935 | */ |
64aa7ed9 | 936 | struct page *get_new_data_page(struct inode *inode, |
a8865372 | 937 | struct page *ipage, pgoff_t index, bool new_i_size) |
eb47b800 | 938 | { |
eb47b800 JK |
939 | struct address_space *mapping = inode->i_mapping; |
940 | struct page *page; | |
941 | struct dnode_of_data dn; | |
942 | int err; | |
943 | ||
a8865372 | 944 | set_new_dnode(&dn, inode, ipage, NULL, 0); |
b600965c | 945 | err = f2fs_reserve_block(&dn, index); |
eb47b800 JK |
946 | if (err) |
947 | return ERR_PTR(err); | |
afcb7ca0 | 948 | repeat: |
eb47b800 | 949 | page = grab_cache_page(mapping, index); |
a8865372 JK |
950 | if (!page) { |
951 | err = -ENOMEM; | |
952 | goto put_err; | |
953 | } | |
eb47b800 JK |
954 | |
955 | if (PageUptodate(page)) | |
956 | return page; | |
957 | ||
958 | if (dn.data_blkaddr == NEW_ADDR) { | |
959 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 960 | SetPageUptodate(page); |
eb47b800 | 961 | } else { |
cf04e8eb JK |
962 | struct f2fs_io_info fio = { |
963 | .type = DATA, | |
964 | .rw = READ_SYNC, | |
965 | .blk_addr = dn.data_blkaddr, | |
966 | }; | |
967 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
393ff91f | 968 | if (err) |
a8865372 JK |
969 | goto put_err; |
970 | ||
393ff91f | 971 | lock_page(page); |
6bacf52f | 972 | if (unlikely(!PageUptodate(page))) { |
393ff91f | 973 | f2fs_put_page(page, 1); |
a8865372 JK |
974 | err = -EIO; |
975 | goto put_err; | |
eb47b800 | 976 | } |
6bacf52f | 977 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
978 | f2fs_put_page(page, 1); |
979 | goto repeat; | |
eb47b800 JK |
980 | } |
981 | } | |
eb47b800 JK |
982 | |
983 | if (new_i_size && | |
984 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
985 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
986 | /* Only the directory inode sets new_i_size */ |
987 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
988 | } |
989 | return page; | |
a8865372 JK |
990 | |
991 | put_err: | |
992 | f2fs_put_dnode(&dn); | |
993 | return ERR_PTR(err); | |
eb47b800 JK |
994 | } |
995 | ||
bfad7c2d JK |
996 | static int __allocate_data_block(struct dnode_of_data *dn) |
997 | { | |
4081363f | 998 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
976e4c50 | 999 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); |
bfad7c2d | 1000 | struct f2fs_summary sum; |
bfad7c2d | 1001 | struct node_info ni; |
38aa0889 | 1002 | int seg = CURSEG_WARM_DATA; |
976e4c50 | 1003 | pgoff_t fofs; |
bfad7c2d JK |
1004 | |
1005 | if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))) | |
1006 | return -EPERM; | |
1007 | if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1))) | |
1008 | return -ENOSPC; | |
1009 | ||
bfad7c2d JK |
1010 | get_node_info(sbi, dn->nid, &ni); |
1011 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
1012 | ||
38aa0889 JK |
1013 | if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page) |
1014 | seg = CURSEG_DIRECT_IO; | |
1015 | ||
1016 | allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg); | |
bfad7c2d JK |
1017 | |
1018 | /* direct IO doesn't use extent cache to maximize the performance */ | |
41ef94b3 | 1019 | __set_data_blkaddr(dn); |
bfad7c2d | 1020 | |
976e4c50 JK |
1021 | /* update i_size */ |
1022 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + | |
1023 | dn->ofs_in_node; | |
1024 | if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT)) | |
1025 | i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT)); | |
1026 | ||
bfad7c2d JK |
1027 | return 0; |
1028 | } | |
1029 | ||
59b802e5 JK |
1030 | static void __allocate_data_blocks(struct inode *inode, loff_t offset, |
1031 | size_t count) | |
1032 | { | |
1033 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
1034 | struct dnode_of_data dn; | |
1035 | u64 start = F2FS_BYTES_TO_BLK(offset); | |
1036 | u64 len = F2FS_BYTES_TO_BLK(count); | |
1037 | bool allocated; | |
1038 | u64 end_offset; | |
1039 | ||
1040 | while (len) { | |
1041 | f2fs_balance_fs(sbi); | |
1042 | f2fs_lock_op(sbi); | |
1043 | ||
1044 | /* When reading holes, we need its node page */ | |
1045 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
1046 | if (get_dnode_of_data(&dn, start, ALLOC_NODE)) | |
1047 | goto out; | |
1048 | ||
1049 | allocated = false; | |
1050 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); | |
1051 | ||
1052 | while (dn.ofs_in_node < end_offset && len) { | |
1053 | if (dn.data_blkaddr == NULL_ADDR) { | |
1054 | if (__allocate_data_block(&dn)) | |
1055 | goto sync_out; | |
1056 | allocated = true; | |
1057 | } | |
1058 | len--; | |
1059 | start++; | |
1060 | dn.ofs_in_node++; | |
1061 | } | |
1062 | ||
1063 | if (allocated) | |
1064 | sync_inode_page(&dn); | |
1065 | ||
1066 | f2fs_put_dnode(&dn); | |
1067 | f2fs_unlock_op(sbi); | |
1068 | } | |
1069 | return; | |
1070 | ||
1071 | sync_out: | |
1072 | if (allocated) | |
1073 | sync_inode_page(&dn); | |
1074 | f2fs_put_dnode(&dn); | |
1075 | out: | |
1076 | f2fs_unlock_op(sbi); | |
1077 | return; | |
1078 | } | |
1079 | ||
0a8165d7 | 1080 | /* |
4f4124d0 CY |
1081 | * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh. |
1082 | * If original data blocks are allocated, then give them to blockdev. | |
1083 | * Otherwise, | |
1084 | * a. preallocate requested block addresses | |
1085 | * b. do not use extent cache for better performance | |
1086 | * c. give the block addresses to blockdev | |
eb47b800 | 1087 | */ |
ccfb3000 JK |
1088 | static int __get_data_block(struct inode *inode, sector_t iblock, |
1089 | struct buffer_head *bh_result, int create, bool fiemap) | |
eb47b800 JK |
1090 | { |
1091 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
1092 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
1093 | struct dnode_of_data dn; | |
bfad7c2d JK |
1094 | int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA; |
1095 | pgoff_t pgofs, end_offset; | |
1096 | int err = 0, ofs = 1; | |
a2e7d1bf | 1097 | struct extent_info ei; |
bfad7c2d | 1098 | bool allocated = false; |
eb47b800 JK |
1099 | |
1100 | /* Get the page offset from the block offset(iblock) */ | |
1101 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
1102 | ||
7e4dde79 | 1103 | if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) { |
a2e7d1bf | 1104 | f2fs_map_bh(inode->i_sb, pgofs, &ei, bh_result); |
bfad7c2d | 1105 | goto out; |
a2e7d1bf | 1106 | } |
bfad7c2d | 1107 | |
59b802e5 | 1108 | if (create) |
4081363f | 1109 | f2fs_lock_op(F2FS_I_SB(inode)); |
eb47b800 JK |
1110 | |
1111 | /* When reading holes, we need its node page */ | |
1112 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
bfad7c2d | 1113 | err = get_dnode_of_data(&dn, pgofs, mode); |
1ec79083 | 1114 | if (err) { |
bfad7c2d JK |
1115 | if (err == -ENOENT) |
1116 | err = 0; | |
1117 | goto unlock_out; | |
848753aa | 1118 | } |
ccfb3000 | 1119 | if (dn.data_blkaddr == NEW_ADDR && !fiemap) |
1ec79083 | 1120 | goto put_out; |
eb47b800 | 1121 | |
bfad7c2d | 1122 | if (dn.data_blkaddr != NULL_ADDR) { |
da17eece | 1123 | set_buffer_new(bh_result); |
bfad7c2d JK |
1124 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); |
1125 | } else if (create) { | |
1126 | err = __allocate_data_block(&dn); | |
1127 | if (err) | |
1128 | goto put_out; | |
1129 | allocated = true; | |
da17eece | 1130 | set_buffer_new(bh_result); |
bfad7c2d JK |
1131 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); |
1132 | } else { | |
1133 | goto put_out; | |
1134 | } | |
1135 | ||
6403eb1f | 1136 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); |
bfad7c2d JK |
1137 | bh_result->b_size = (((size_t)1) << blkbits); |
1138 | dn.ofs_in_node++; | |
1139 | pgofs++; | |
1140 | ||
1141 | get_next: | |
1142 | if (dn.ofs_in_node >= end_offset) { | |
1143 | if (allocated) | |
1144 | sync_inode_page(&dn); | |
1145 | allocated = false; | |
1146 | f2fs_put_dnode(&dn); | |
1147 | ||
1148 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
1149 | err = get_dnode_of_data(&dn, pgofs, mode); | |
1ec79083 | 1150 | if (err) { |
bfad7c2d JK |
1151 | if (err == -ENOENT) |
1152 | err = 0; | |
1153 | goto unlock_out; | |
1154 | } | |
ccfb3000 | 1155 | if (dn.data_blkaddr == NEW_ADDR && !fiemap) |
1ec79083 JK |
1156 | goto put_out; |
1157 | ||
6403eb1f | 1158 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); |
bfad7c2d | 1159 | } |
eb47b800 | 1160 | |
bfad7c2d JK |
1161 | if (maxblocks > (bh_result->b_size >> blkbits)) { |
1162 | block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node); | |
1163 | if (blkaddr == NULL_ADDR && create) { | |
1164 | err = __allocate_data_block(&dn); | |
1165 | if (err) | |
1166 | goto sync_out; | |
1167 | allocated = true; | |
1168 | blkaddr = dn.data_blkaddr; | |
1169 | } | |
e1c42045 | 1170 | /* Give more consecutive addresses for the readahead */ |
bfad7c2d JK |
1171 | if (blkaddr == (bh_result->b_blocknr + ofs)) { |
1172 | ofs++; | |
1173 | dn.ofs_in_node++; | |
1174 | pgofs++; | |
1175 | bh_result->b_size += (((size_t)1) << blkbits); | |
1176 | goto get_next; | |
1177 | } | |
eb47b800 | 1178 | } |
bfad7c2d JK |
1179 | sync_out: |
1180 | if (allocated) | |
1181 | sync_inode_page(&dn); | |
1182 | put_out: | |
eb47b800 | 1183 | f2fs_put_dnode(&dn); |
bfad7c2d JK |
1184 | unlock_out: |
1185 | if (create) | |
4081363f | 1186 | f2fs_unlock_op(F2FS_I_SB(inode)); |
bfad7c2d JK |
1187 | out: |
1188 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
1189 | return err; | |
eb47b800 JK |
1190 | } |
1191 | ||
ccfb3000 JK |
1192 | static int get_data_block(struct inode *inode, sector_t iblock, |
1193 | struct buffer_head *bh_result, int create) | |
1194 | { | |
1195 | return __get_data_block(inode, iblock, bh_result, create, false); | |
1196 | } | |
1197 | ||
1198 | static int get_data_block_fiemap(struct inode *inode, sector_t iblock, | |
1199 | struct buffer_head *bh_result, int create) | |
1200 | { | |
1201 | return __get_data_block(inode, iblock, bh_result, create, true); | |
1202 | } | |
1203 | ||
9ab70134 JK |
1204 | int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
1205 | u64 start, u64 len) | |
1206 | { | |
ccfb3000 JK |
1207 | return generic_block_fiemap(inode, fieinfo, |
1208 | start, len, get_data_block_fiemap); | |
9ab70134 JK |
1209 | } |
1210 | ||
eb47b800 JK |
1211 | static int f2fs_read_data_page(struct file *file, struct page *page) |
1212 | { | |
9ffe0fb5 | 1213 | struct inode *inode = page->mapping->host; |
b3d208f9 | 1214 | int ret = -EAGAIN; |
9ffe0fb5 | 1215 | |
c20e89cd CY |
1216 | trace_f2fs_readpage(page, DATA); |
1217 | ||
e1c42045 | 1218 | /* If the file has inline data, try to read it directly */ |
9ffe0fb5 HL |
1219 | if (f2fs_has_inline_data(inode)) |
1220 | ret = f2fs_read_inline_data(inode, page); | |
b3d208f9 | 1221 | if (ret == -EAGAIN) |
9ffe0fb5 HL |
1222 | ret = mpage_readpage(page, get_data_block); |
1223 | ||
1224 | return ret; | |
eb47b800 JK |
1225 | } |
1226 | ||
1227 | static int f2fs_read_data_pages(struct file *file, | |
1228 | struct address_space *mapping, | |
1229 | struct list_head *pages, unsigned nr_pages) | |
1230 | { | |
9ffe0fb5 HL |
1231 | struct inode *inode = file->f_mapping->host; |
1232 | ||
1233 | /* If the file has inline data, skip readpages */ | |
1234 | if (f2fs_has_inline_data(inode)) | |
1235 | return 0; | |
1236 | ||
bfad7c2d | 1237 | return mpage_readpages(mapping, pages, nr_pages, get_data_block); |
eb47b800 JK |
1238 | } |
1239 | ||
458e6197 | 1240 | int do_write_data_page(struct page *page, struct f2fs_io_info *fio) |
eb47b800 JK |
1241 | { |
1242 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
1243 | struct dnode_of_data dn; |
1244 | int err = 0; | |
1245 | ||
1246 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 1247 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
1248 | if (err) |
1249 | return err; | |
1250 | ||
cf04e8eb | 1251 | fio->blk_addr = dn.data_blkaddr; |
eb47b800 JK |
1252 | |
1253 | /* This page is already truncated */ | |
cf04e8eb | 1254 | if (fio->blk_addr == NULL_ADDR) |
eb47b800 JK |
1255 | goto out_writepage; |
1256 | ||
1257 | set_page_writeback(page); | |
1258 | ||
1259 | /* | |
1260 | * If current allocation needs SSR, | |
1261 | * it had better in-place writes for updated data. | |
1262 | */ | |
cf04e8eb | 1263 | if (unlikely(fio->blk_addr != NEW_ADDR && |
b25958b6 HL |
1264 | !is_cold_data(page) && |
1265 | need_inplace_update(inode))) { | |
cf04e8eb | 1266 | rewrite_data_page(page, fio); |
fff04f90 | 1267 | set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE); |
eb47b800 | 1268 | } else { |
cf04e8eb | 1269 | write_data_page(page, &dn, fio); |
7e4dde79 | 1270 | f2fs_update_extent_cache(&dn); |
fff04f90 | 1271 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); |
eb47b800 JK |
1272 | } |
1273 | out_writepage: | |
1274 | f2fs_put_dnode(&dn); | |
1275 | return err; | |
1276 | } | |
1277 | ||
1278 | static int f2fs_write_data_page(struct page *page, | |
1279 | struct writeback_control *wbc) | |
1280 | { | |
1281 | struct inode *inode = page->mapping->host; | |
4081363f | 1282 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
eb47b800 JK |
1283 | loff_t i_size = i_size_read(inode); |
1284 | const pgoff_t end_index = ((unsigned long long) i_size) | |
1285 | >> PAGE_CACHE_SHIFT; | |
9ffe0fb5 | 1286 | unsigned offset = 0; |
39936837 | 1287 | bool need_balance_fs = false; |
eb47b800 | 1288 | int err = 0; |
458e6197 JK |
1289 | struct f2fs_io_info fio = { |
1290 | .type = DATA, | |
6c311ec6 | 1291 | .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE, |
458e6197 | 1292 | }; |
eb47b800 | 1293 | |
ecda0de3 CY |
1294 | trace_f2fs_writepage(page, DATA); |
1295 | ||
eb47b800 | 1296 | if (page->index < end_index) |
39936837 | 1297 | goto write; |
eb47b800 JK |
1298 | |
1299 | /* | |
1300 | * If the offset is out-of-range of file size, | |
1301 | * this page does not have to be written to disk. | |
1302 | */ | |
1303 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
76f60268 | 1304 | if ((page->index >= end_index + 1) || !offset) |
39936837 | 1305 | goto out; |
eb47b800 JK |
1306 | |
1307 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 | 1308 | write: |
caf0047e | 1309 | if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) |
eb47b800 | 1310 | goto redirty_out; |
1e84371f JK |
1311 | if (f2fs_is_drop_cache(inode)) |
1312 | goto out; | |
1313 | if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim && | |
1314 | available_free_memory(sbi, BASE_CHECK)) | |
1315 | goto redirty_out; | |
eb47b800 | 1316 | |
39936837 | 1317 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 | 1318 | if (S_ISDIR(inode->i_mode)) { |
cf779cab JK |
1319 | if (unlikely(f2fs_cp_error(sbi))) |
1320 | goto redirty_out; | |
458e6197 | 1321 | err = do_write_data_page(page, &fio); |
8618b881 JK |
1322 | goto done; |
1323 | } | |
9ffe0fb5 | 1324 | |
cf779cab JK |
1325 | /* we should bypass data pages to proceed the kworkder jobs */ |
1326 | if (unlikely(f2fs_cp_error(sbi))) { | |
1327 | SetPageError(page); | |
a7ffdbe2 | 1328 | goto out; |
cf779cab JK |
1329 | } |
1330 | ||
8618b881 | 1331 | if (!wbc->for_reclaim) |
39936837 | 1332 | need_balance_fs = true; |
8618b881 | 1333 | else if (has_not_enough_free_secs(sbi, 0)) |
39936837 | 1334 | goto redirty_out; |
eb47b800 | 1335 | |
b3d208f9 | 1336 | err = -EAGAIN; |
8618b881 | 1337 | f2fs_lock_op(sbi); |
b3d208f9 JK |
1338 | if (f2fs_has_inline_data(inode)) |
1339 | err = f2fs_write_inline_data(inode, page); | |
1340 | if (err == -EAGAIN) | |
8618b881 JK |
1341 | err = do_write_data_page(page, &fio); |
1342 | f2fs_unlock_op(sbi); | |
1343 | done: | |
1344 | if (err && err != -ENOENT) | |
1345 | goto redirty_out; | |
eb47b800 | 1346 | |
eb47b800 | 1347 | clear_cold_data(page); |
39936837 | 1348 | out: |
a7ffdbe2 | 1349 | inode_dec_dirty_pages(inode); |
eb47b800 | 1350 | unlock_page(page); |
39936837 | 1351 | if (need_balance_fs) |
eb47b800 | 1352 | f2fs_balance_fs(sbi); |
2aea39ec JK |
1353 | if (wbc->for_reclaim) |
1354 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
eb47b800 JK |
1355 | return 0; |
1356 | ||
eb47b800 | 1357 | redirty_out: |
76f60268 | 1358 | redirty_page_for_writepage(wbc, page); |
8618b881 | 1359 | return AOP_WRITEPAGE_ACTIVATE; |
eb47b800 JK |
1360 | } |
1361 | ||
fa9150a8 NJ |
1362 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
1363 | void *data) | |
1364 | { | |
1365 | struct address_space *mapping = data; | |
1366 | int ret = mapping->a_ops->writepage(page, wbc); | |
1367 | mapping_set_error(mapping, ret); | |
1368 | return ret; | |
1369 | } | |
1370 | ||
25ca923b | 1371 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
1372 | struct writeback_control *wbc) |
1373 | { | |
1374 | struct inode *inode = mapping->host; | |
4081363f | 1375 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
531ad7d5 | 1376 | bool locked = false; |
eb47b800 | 1377 | int ret; |
50c8cdb3 | 1378 | long diff; |
eb47b800 | 1379 | |
e5748434 CY |
1380 | trace_f2fs_writepages(mapping->host, wbc, DATA); |
1381 | ||
cfb185a1 | 1382 | /* deal with chardevs and other special file */ |
1383 | if (!mapping->a_ops->writepage) | |
1384 | return 0; | |
1385 | ||
87d6f890 | 1386 | if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && |
a7ffdbe2 | 1387 | get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && |
6fb03f3a | 1388 | available_free_memory(sbi, DIRTY_DENTS)) |
d3baf95d | 1389 | goto skip_write; |
87d6f890 | 1390 | |
50c8cdb3 | 1391 | diff = nr_pages_to_write(sbi, DATA, wbc); |
eb47b800 | 1392 | |
531ad7d5 | 1393 | if (!S_ISDIR(inode->i_mode)) { |
eb47b800 | 1394 | mutex_lock(&sbi->writepages); |
531ad7d5 JK |
1395 | locked = true; |
1396 | } | |
fa9150a8 | 1397 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
531ad7d5 | 1398 | if (locked) |
eb47b800 | 1399 | mutex_unlock(&sbi->writepages); |
458e6197 JK |
1400 | |
1401 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
eb47b800 JK |
1402 | |
1403 | remove_dirty_dir_inode(inode); | |
1404 | ||
50c8cdb3 | 1405 | wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); |
eb47b800 | 1406 | return ret; |
d3baf95d JK |
1407 | |
1408 | skip_write: | |
a7ffdbe2 | 1409 | wbc->pages_skipped += get_dirty_pages(inode); |
d3baf95d | 1410 | return 0; |
eb47b800 JK |
1411 | } |
1412 | ||
3aab8f82 CY |
1413 | static void f2fs_write_failed(struct address_space *mapping, loff_t to) |
1414 | { | |
1415 | struct inode *inode = mapping->host; | |
1416 | ||
1417 | if (to > inode->i_size) { | |
1418 | truncate_pagecache(inode, inode->i_size); | |
764aa3e9 | 1419 | truncate_blocks(inode, inode->i_size, true); |
3aab8f82 CY |
1420 | } |
1421 | } | |
1422 | ||
eb47b800 JK |
1423 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, |
1424 | loff_t pos, unsigned len, unsigned flags, | |
1425 | struct page **pagep, void **fsdata) | |
1426 | { | |
1427 | struct inode *inode = mapping->host; | |
4081363f | 1428 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
9ba69cf9 | 1429 | struct page *page, *ipage; |
eb47b800 JK |
1430 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; |
1431 | struct dnode_of_data dn; | |
1432 | int err = 0; | |
1433 | ||
62aed044 CY |
1434 | trace_f2fs_write_begin(inode, pos, len, flags); |
1435 | ||
eb47b800 | 1436 | f2fs_balance_fs(sbi); |
5f727395 JK |
1437 | |
1438 | /* | |
1439 | * We should check this at this moment to avoid deadlock on inode page | |
1440 | * and #0 page. The locking rule for inline_data conversion should be: | |
1441 | * lock_page(page #0) -> lock_page(inode_page) | |
1442 | */ | |
1443 | if (index != 0) { | |
1444 | err = f2fs_convert_inline_inode(inode); | |
1445 | if (err) | |
1446 | goto fail; | |
1447 | } | |
afcb7ca0 | 1448 | repeat: |
eb47b800 | 1449 | page = grab_cache_page_write_begin(mapping, index, flags); |
3aab8f82 CY |
1450 | if (!page) { |
1451 | err = -ENOMEM; | |
1452 | goto fail; | |
1453 | } | |
d5f66990 | 1454 | |
eb47b800 JK |
1455 | *pagep = page; |
1456 | ||
e479556b | 1457 | f2fs_lock_op(sbi); |
9ba69cf9 JK |
1458 | |
1459 | /* check inline_data */ | |
1460 | ipage = get_node_page(sbi, inode->i_ino); | |
cd34e296 CY |
1461 | if (IS_ERR(ipage)) { |
1462 | err = PTR_ERR(ipage); | |
9ba69cf9 | 1463 | goto unlock_fail; |
cd34e296 | 1464 | } |
9ba69cf9 | 1465 | |
b3d208f9 JK |
1466 | set_new_dnode(&dn, inode, ipage, ipage, 0); |
1467 | ||
9ba69cf9 | 1468 | if (f2fs_has_inline_data(inode)) { |
b3d208f9 JK |
1469 | if (pos + len <= MAX_INLINE_DATA) { |
1470 | read_inline_data(page, ipage); | |
1471 | set_inode_flag(F2FS_I(inode), FI_DATA_EXIST); | |
1472 | sync_inode_page(&dn); | |
1473 | goto put_next; | |
b3d208f9 | 1474 | } |
5f727395 JK |
1475 | err = f2fs_convert_inline_page(&dn, page); |
1476 | if (err) | |
1477 | goto put_fail; | |
b600965c | 1478 | } |
9ba69cf9 JK |
1479 | err = f2fs_reserve_block(&dn, index); |
1480 | if (err) | |
8cdcb713 | 1481 | goto put_fail; |
b3d208f9 | 1482 | put_next: |
9ba69cf9 JK |
1483 | f2fs_put_dnode(&dn); |
1484 | f2fs_unlock_op(sbi); | |
1485 | ||
eb47b800 JK |
1486 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) |
1487 | return 0; | |
1488 | ||
b3d208f9 JK |
1489 | f2fs_wait_on_page_writeback(page, DATA); |
1490 | ||
eb47b800 JK |
1491 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { |
1492 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
1493 | unsigned end = start + len; | |
1494 | ||
1495 | /* Reading beyond i_size is simple: memset to zero */ | |
1496 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 1497 | goto out; |
eb47b800 JK |
1498 | } |
1499 | ||
b3d208f9 | 1500 | if (dn.data_blkaddr == NEW_ADDR) { |
eb47b800 JK |
1501 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); |
1502 | } else { | |
cf04e8eb JK |
1503 | struct f2fs_io_info fio = { |
1504 | .type = DATA, | |
1505 | .rw = READ_SYNC, | |
1506 | .blk_addr = dn.data_blkaddr, | |
1507 | }; | |
1508 | err = f2fs_submit_page_bio(sbi, page, &fio); | |
9234f319 JK |
1509 | if (err) |
1510 | goto fail; | |
d54c795b | 1511 | |
393ff91f | 1512 | lock_page(page); |
6bacf52f | 1513 | if (unlikely(!PageUptodate(page))) { |
393ff91f | 1514 | f2fs_put_page(page, 1); |
3aab8f82 CY |
1515 | err = -EIO; |
1516 | goto fail; | |
eb47b800 | 1517 | } |
6bacf52f | 1518 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
1519 | f2fs_put_page(page, 1); |
1520 | goto repeat; | |
eb47b800 JK |
1521 | } |
1522 | } | |
393ff91f | 1523 | out: |
eb47b800 JK |
1524 | SetPageUptodate(page); |
1525 | clear_cold_data(page); | |
1526 | return 0; | |
9ba69cf9 | 1527 | |
8cdcb713 JK |
1528 | put_fail: |
1529 | f2fs_put_dnode(&dn); | |
9ba69cf9 JK |
1530 | unlock_fail: |
1531 | f2fs_unlock_op(sbi); | |
b3d208f9 | 1532 | f2fs_put_page(page, 1); |
3aab8f82 CY |
1533 | fail: |
1534 | f2fs_write_failed(mapping, pos + len); | |
1535 | return err; | |
eb47b800 JK |
1536 | } |
1537 | ||
a1dd3c13 JK |
1538 | static int f2fs_write_end(struct file *file, |
1539 | struct address_space *mapping, | |
1540 | loff_t pos, unsigned len, unsigned copied, | |
1541 | struct page *page, void *fsdata) | |
1542 | { | |
1543 | struct inode *inode = page->mapping->host; | |
1544 | ||
dfb2bf38 CY |
1545 | trace_f2fs_write_end(inode, pos, len, copied); |
1546 | ||
34ba94ba | 1547 | set_page_dirty(page); |
a1dd3c13 JK |
1548 | |
1549 | if (pos + copied > i_size_read(inode)) { | |
1550 | i_size_write(inode, pos + copied); | |
1551 | mark_inode_dirty(inode); | |
1552 | update_inode_page(inode); | |
1553 | } | |
1554 | ||
75c3c8bc | 1555 | f2fs_put_page(page, 1); |
a1dd3c13 JK |
1556 | return copied; |
1557 | } | |
1558 | ||
944fcfc1 | 1559 | static int check_direct_IO(struct inode *inode, int rw, |
5b46f25d | 1560 | struct iov_iter *iter, loff_t offset) |
944fcfc1 JK |
1561 | { |
1562 | unsigned blocksize_mask = inode->i_sb->s_blocksize - 1; | |
944fcfc1 JK |
1563 | |
1564 | if (rw == READ) | |
1565 | return 0; | |
1566 | ||
1567 | if (offset & blocksize_mask) | |
1568 | return -EINVAL; | |
1569 | ||
5b46f25d AV |
1570 | if (iov_iter_alignment(iter) & blocksize_mask) |
1571 | return -EINVAL; | |
1572 | ||
944fcfc1 JK |
1573 | return 0; |
1574 | } | |
1575 | ||
eb47b800 | 1576 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
d8d3d94b | 1577 | struct iov_iter *iter, loff_t offset) |
eb47b800 JK |
1578 | { |
1579 | struct file *file = iocb->ki_filp; | |
3aab8f82 CY |
1580 | struct address_space *mapping = file->f_mapping; |
1581 | struct inode *inode = mapping->host; | |
1582 | size_t count = iov_iter_count(iter); | |
1583 | int err; | |
944fcfc1 | 1584 | |
b3d208f9 JK |
1585 | /* we don't need to use inline_data strictly */ |
1586 | if (f2fs_has_inline_data(inode)) { | |
1587 | err = f2fs_convert_inline_inode(inode); | |
1588 | if (err) | |
1589 | return err; | |
1590 | } | |
9ffe0fb5 | 1591 | |
5b46f25d | 1592 | if (check_direct_IO(inode, rw, iter, offset)) |
944fcfc1 JK |
1593 | return 0; |
1594 | ||
70407fad CY |
1595 | trace_f2fs_direct_IO_enter(inode, offset, count, rw); |
1596 | ||
59b802e5 JK |
1597 | if (rw & WRITE) |
1598 | __allocate_data_blocks(inode, offset, count); | |
1599 | ||
3aab8f82 CY |
1600 | err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block); |
1601 | if (err < 0 && (rw & WRITE)) | |
1602 | f2fs_write_failed(mapping, offset + count); | |
70407fad CY |
1603 | |
1604 | trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); | |
1605 | ||
3aab8f82 | 1606 | return err; |
eb47b800 JK |
1607 | } |
1608 | ||
487261f3 CY |
1609 | void f2fs_invalidate_page(struct page *page, unsigned int offset, |
1610 | unsigned int length) | |
eb47b800 JK |
1611 | { |
1612 | struct inode *inode = page->mapping->host; | |
487261f3 | 1613 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
a7ffdbe2 | 1614 | |
487261f3 CY |
1615 | if (inode->i_ino >= F2FS_ROOT_INO(sbi) && |
1616 | (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE)) | |
a7ffdbe2 JK |
1617 | return; |
1618 | ||
487261f3 CY |
1619 | if (PageDirty(page)) { |
1620 | if (inode->i_ino == F2FS_META_INO(sbi)) | |
1621 | dec_page_count(sbi, F2FS_DIRTY_META); | |
1622 | else if (inode->i_ino == F2FS_NODE_INO(sbi)) | |
1623 | dec_page_count(sbi, F2FS_DIRTY_NODES); | |
1624 | else | |
1625 | inode_dec_dirty_pages(inode); | |
1626 | } | |
eb47b800 JK |
1627 | ClearPagePrivate(page); |
1628 | } | |
1629 | ||
487261f3 | 1630 | int f2fs_release_page(struct page *page, gfp_t wait) |
eb47b800 | 1631 | { |
f68daeeb JK |
1632 | /* If this is dirty page, keep PagePrivate */ |
1633 | if (PageDirty(page)) | |
1634 | return 0; | |
1635 | ||
eb47b800 | 1636 | ClearPagePrivate(page); |
c3850aa1 | 1637 | return 1; |
eb47b800 JK |
1638 | } |
1639 | ||
1640 | static int f2fs_set_data_page_dirty(struct page *page) | |
1641 | { | |
1642 | struct address_space *mapping = page->mapping; | |
1643 | struct inode *inode = mapping->host; | |
1644 | ||
26c6b887 JK |
1645 | trace_f2fs_set_page_dirty(page, DATA); |
1646 | ||
eb47b800 | 1647 | SetPageUptodate(page); |
34ba94ba | 1648 | |
1e84371f | 1649 | if (f2fs_is_atomic_file(inode)) { |
34ba94ba JK |
1650 | register_inmem_page(inode, page); |
1651 | return 1; | |
1652 | } | |
1653 | ||
a18ff063 JK |
1654 | mark_inode_dirty(inode); |
1655 | ||
eb47b800 JK |
1656 | if (!PageDirty(page)) { |
1657 | __set_page_dirty_nobuffers(page); | |
a7ffdbe2 | 1658 | update_dirty_page(inode, page); |
eb47b800 JK |
1659 | return 1; |
1660 | } | |
1661 | return 0; | |
1662 | } | |
1663 | ||
c01e54b7 JK |
1664 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
1665 | { | |
454ae7e5 CY |
1666 | struct inode *inode = mapping->host; |
1667 | ||
b3d208f9 JK |
1668 | /* we don't need to use inline_data strictly */ |
1669 | if (f2fs_has_inline_data(inode)) { | |
1670 | int err = f2fs_convert_inline_inode(inode); | |
1671 | if (err) | |
1672 | return err; | |
1673 | } | |
bfad7c2d | 1674 | return generic_block_bmap(mapping, block, get_data_block); |
c01e54b7 JK |
1675 | } |
1676 | ||
429511cd CY |
1677 | void init_extent_cache_info(struct f2fs_sb_info *sbi) |
1678 | { | |
1679 | INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO); | |
1680 | init_rwsem(&sbi->extent_tree_lock); | |
1681 | INIT_LIST_HEAD(&sbi->extent_list); | |
1682 | spin_lock_init(&sbi->extent_lock); | |
1683 | sbi->total_ext_tree = 0; | |
1684 | atomic_set(&sbi->total_ext_node, 0); | |
1685 | } | |
1686 | ||
1687 | int __init create_extent_cache(void) | |
1688 | { | |
1689 | extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree", | |
1690 | sizeof(struct extent_tree)); | |
1691 | if (!extent_tree_slab) | |
1692 | return -ENOMEM; | |
1693 | extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node", | |
1694 | sizeof(struct extent_node)); | |
1695 | if (!extent_node_slab) { | |
1696 | kmem_cache_destroy(extent_tree_slab); | |
1697 | return -ENOMEM; | |
1698 | } | |
1699 | return 0; | |
1700 | } | |
1701 | ||
1702 | void destroy_extent_cache(void) | |
1703 | { | |
1704 | kmem_cache_destroy(extent_node_slab); | |
1705 | kmem_cache_destroy(extent_tree_slab); | |
1706 | } | |
1707 | ||
eb47b800 JK |
1708 | const struct address_space_operations f2fs_dblock_aops = { |
1709 | .readpage = f2fs_read_data_page, | |
1710 | .readpages = f2fs_read_data_pages, | |
1711 | .writepage = f2fs_write_data_page, | |
1712 | .writepages = f2fs_write_data_pages, | |
1713 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 1714 | .write_end = f2fs_write_end, |
eb47b800 | 1715 | .set_page_dirty = f2fs_set_data_page_dirty, |
487261f3 CY |
1716 | .invalidatepage = f2fs_invalidate_page, |
1717 | .releasepage = f2fs_release_page, | |
eb47b800 | 1718 | .direct_IO = f2fs_direct_IO, |
c01e54b7 | 1719 | .bmap = f2fs_bmap, |
eb47b800 | 1720 | }; |