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 | */ | |
216a620a | 203 | 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; |
216a620a | 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 | ||
3402e87c | 260 | clear_buffer_new(bh_result); |
a2e7d1bf CY |
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 | ||
0c872e2d | 276 | read_lock(&fi->ext_lock); |
eb47b800 | 277 | if (fi->ext.len == 0) { |
0c872e2d | 278 | read_unlock(&fi->ext_lock); |
7e4dde79 | 279 | return false; |
eb47b800 JK |
280 | } |
281 | ||
dcdfff65 JK |
282 | stat_inc_total_hit(inode->i_sb); |
283 | ||
eb47b800 JK |
284 | start_fofs = fi->ext.fofs; |
285 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
4d0b0bd4 | 286 | start_blkaddr = fi->ext.blk; |
eb47b800 JK |
287 | |
288 | if (pgofs >= start_fofs && pgofs <= end_fofs) { | |
a2e7d1bf | 289 | *ei = fi->ext; |
dcdfff65 | 290 | stat_inc_read_hit(inode->i_sb); |
0c872e2d | 291 | read_unlock(&fi->ext_lock); |
7e4dde79 | 292 | return true; |
eb47b800 | 293 | } |
0c872e2d | 294 | read_unlock(&fi->ext_lock); |
7e4dde79 | 295 | return false; |
eb47b800 JK |
296 | } |
297 | ||
7e4dde79 CY |
298 | static bool update_extent_info(struct inode *inode, pgoff_t fofs, |
299 | block_t blkaddr) | |
eb47b800 | 300 | { |
7e4dde79 CY |
301 | struct f2fs_inode_info *fi = F2FS_I(inode); |
302 | pgoff_t start_fofs, end_fofs; | |
eb47b800 | 303 | block_t start_blkaddr, end_blkaddr; |
c11abd1a | 304 | int need_update = true; |
eb47b800 | 305 | |
0c872e2d | 306 | write_lock(&fi->ext_lock); |
eb47b800 JK |
307 | |
308 | start_fofs = fi->ext.fofs; | |
309 | end_fofs = fi->ext.fofs + fi->ext.len - 1; | |
4d0b0bd4 CY |
310 | start_blkaddr = fi->ext.blk; |
311 | end_blkaddr = fi->ext.blk + fi->ext.len - 1; | |
eb47b800 JK |
312 | |
313 | /* Drop and initialize the matched extent */ | |
314 | if (fi->ext.len == 1 && fofs == start_fofs) | |
315 | fi->ext.len = 0; | |
316 | ||
317 | /* Initial extent */ | |
318 | if (fi->ext.len == 0) { | |
7e4dde79 | 319 | if (blkaddr != NULL_ADDR) { |
eb47b800 | 320 | fi->ext.fofs = fofs; |
7e4dde79 | 321 | fi->ext.blk = blkaddr; |
eb47b800 JK |
322 | fi->ext.len = 1; |
323 | } | |
324 | goto end_update; | |
325 | } | |
326 | ||
6224da87 | 327 | /* Front merge */ |
7e4dde79 | 328 | if (fofs == start_fofs - 1 && blkaddr == start_blkaddr - 1) { |
eb47b800 | 329 | fi->ext.fofs--; |
4d0b0bd4 | 330 | fi->ext.blk--; |
eb47b800 JK |
331 | fi->ext.len++; |
332 | goto end_update; | |
333 | } | |
334 | ||
335 | /* Back merge */ | |
7e4dde79 | 336 | if (fofs == end_fofs + 1 && blkaddr == end_blkaddr + 1) { |
eb47b800 JK |
337 | fi->ext.len++; |
338 | goto end_update; | |
339 | } | |
340 | ||
341 | /* Split the existing extent */ | |
342 | if (fi->ext.len > 1 && | |
343 | fofs >= start_fofs && fofs <= end_fofs) { | |
344 | if ((end_fofs - fofs) < (fi->ext.len >> 1)) { | |
345 | fi->ext.len = fofs - start_fofs; | |
346 | } else { | |
347 | fi->ext.fofs = fofs + 1; | |
4d0b0bd4 | 348 | fi->ext.blk = start_blkaddr + fofs - start_fofs + 1; |
eb47b800 JK |
349 | fi->ext.len -= fofs - start_fofs + 1; |
350 | } | |
c11abd1a JK |
351 | } else { |
352 | need_update = false; | |
eb47b800 | 353 | } |
eb47b800 | 354 | |
c11abd1a JK |
355 | /* Finally, if the extent is very fragmented, let's drop the cache. */ |
356 | if (fi->ext.len < F2FS_MIN_EXTENT_LEN) { | |
357 | fi->ext.len = 0; | |
358 | set_inode_flag(fi, FI_NO_EXTENT); | |
359 | need_update = true; | |
360 | } | |
eb47b800 | 361 | end_update: |
0c872e2d | 362 | write_unlock(&fi->ext_lock); |
7e4dde79 CY |
363 | return need_update; |
364 | } | |
365 | ||
429511cd CY |
366 | static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi, |
367 | struct extent_tree *et, struct extent_info *ei, | |
368 | struct rb_node *parent, struct rb_node **p) | |
369 | { | |
370 | struct extent_node *en; | |
371 | ||
372 | en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC); | |
373 | if (!en) | |
374 | return NULL; | |
375 | ||
376 | en->ei = *ei; | |
377 | INIT_LIST_HEAD(&en->list); | |
378 | ||
379 | rb_link_node(&en->rb_node, parent, p); | |
380 | rb_insert_color(&en->rb_node, &et->root); | |
381 | et->count++; | |
382 | atomic_inc(&sbi->total_ext_node); | |
383 | return en; | |
384 | } | |
385 | ||
386 | static void __detach_extent_node(struct f2fs_sb_info *sbi, | |
387 | struct extent_tree *et, struct extent_node *en) | |
388 | { | |
389 | rb_erase(&en->rb_node, &et->root); | |
390 | et->count--; | |
391 | atomic_dec(&sbi->total_ext_node); | |
62c8af65 CY |
392 | |
393 | if (et->cached_en == en) | |
394 | et->cached_en = NULL; | |
429511cd CY |
395 | } |
396 | ||
93dfc526 CY |
397 | static struct extent_tree *__find_extent_tree(struct f2fs_sb_info *sbi, |
398 | nid_t ino) | |
399 | { | |
400 | struct extent_tree *et; | |
401 | ||
402 | down_read(&sbi->extent_tree_lock); | |
403 | et = radix_tree_lookup(&sbi->extent_tree_root, ino); | |
404 | if (!et) { | |
405 | up_read(&sbi->extent_tree_lock); | |
406 | return NULL; | |
407 | } | |
408 | atomic_inc(&et->refcount); | |
409 | up_read(&sbi->extent_tree_lock); | |
410 | ||
411 | return et; | |
412 | } | |
413 | ||
414 | static struct extent_tree *__grab_extent_tree(struct inode *inode) | |
415 | { | |
416 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
417 | struct extent_tree *et; | |
418 | nid_t ino = inode->i_ino; | |
419 | ||
420 | down_write(&sbi->extent_tree_lock); | |
421 | et = radix_tree_lookup(&sbi->extent_tree_root, ino); | |
422 | if (!et) { | |
423 | et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS); | |
424 | f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et); | |
425 | memset(et, 0, sizeof(struct extent_tree)); | |
426 | et->ino = ino; | |
427 | et->root = RB_ROOT; | |
428 | et->cached_en = NULL; | |
429 | rwlock_init(&et->lock); | |
430 | atomic_set(&et->refcount, 0); | |
431 | et->count = 0; | |
432 | sbi->total_ext_tree++; | |
433 | } | |
434 | atomic_inc(&et->refcount); | |
435 | up_write(&sbi->extent_tree_lock); | |
436 | ||
437 | return et; | |
438 | } | |
439 | ||
429511cd CY |
440 | static struct extent_node *__lookup_extent_tree(struct extent_tree *et, |
441 | unsigned int fofs) | |
442 | { | |
443 | struct rb_node *node = et->root.rb_node; | |
444 | struct extent_node *en; | |
445 | ||
62c8af65 CY |
446 | if (et->cached_en) { |
447 | struct extent_info *cei = &et->cached_en->ei; | |
448 | ||
449 | if (cei->fofs <= fofs && cei->fofs + cei->len > fofs) | |
450 | return et->cached_en; | |
451 | } | |
452 | ||
429511cd CY |
453 | while (node) { |
454 | en = rb_entry(node, struct extent_node, rb_node); | |
455 | ||
62c8af65 | 456 | if (fofs < en->ei.fofs) { |
429511cd | 457 | node = node->rb_left; |
62c8af65 | 458 | } else if (fofs >= en->ei.fofs + en->ei.len) { |
429511cd | 459 | node = node->rb_right; |
62c8af65 CY |
460 | } else { |
461 | et->cached_en = en; | |
429511cd | 462 | return en; |
62c8af65 | 463 | } |
429511cd CY |
464 | } |
465 | return NULL; | |
466 | } | |
467 | ||
468 | static struct extent_node *__try_back_merge(struct f2fs_sb_info *sbi, | |
469 | struct extent_tree *et, struct extent_node *en) | |
470 | { | |
471 | struct extent_node *prev; | |
472 | struct rb_node *node; | |
473 | ||
474 | node = rb_prev(&en->rb_node); | |
475 | if (!node) | |
476 | return NULL; | |
477 | ||
478 | prev = rb_entry(node, struct extent_node, rb_node); | |
479 | if (__is_back_mergeable(&en->ei, &prev->ei)) { | |
480 | en->ei.fofs = prev->ei.fofs; | |
481 | en->ei.blk = prev->ei.blk; | |
482 | en->ei.len += prev->ei.len; | |
483 | __detach_extent_node(sbi, et, prev); | |
484 | return prev; | |
485 | } | |
486 | return NULL; | |
487 | } | |
488 | ||
489 | static struct extent_node *__try_front_merge(struct f2fs_sb_info *sbi, | |
490 | struct extent_tree *et, struct extent_node *en) | |
491 | { | |
492 | struct extent_node *next; | |
493 | struct rb_node *node; | |
494 | ||
495 | node = rb_next(&en->rb_node); | |
496 | if (!node) | |
497 | return NULL; | |
498 | ||
499 | next = rb_entry(node, struct extent_node, rb_node); | |
500 | if (__is_front_mergeable(&en->ei, &next->ei)) { | |
501 | en->ei.len += next->ei.len; | |
502 | __detach_extent_node(sbi, et, next); | |
503 | return next; | |
504 | } | |
505 | return NULL; | |
506 | } | |
507 | ||
508 | static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi, | |
509 | struct extent_tree *et, struct extent_info *ei, | |
510 | struct extent_node **den) | |
511 | { | |
512 | struct rb_node **p = &et->root.rb_node; | |
513 | struct rb_node *parent = NULL; | |
514 | struct extent_node *en; | |
515 | ||
516 | while (*p) { | |
517 | parent = *p; | |
518 | en = rb_entry(parent, struct extent_node, rb_node); | |
519 | ||
520 | if (ei->fofs < en->ei.fofs) { | |
521 | if (__is_front_mergeable(ei, &en->ei)) { | |
522 | f2fs_bug_on(sbi, !den); | |
523 | en->ei.fofs = ei->fofs; | |
524 | en->ei.blk = ei->blk; | |
525 | en->ei.len += ei->len; | |
526 | *den = __try_back_merge(sbi, et, en); | |
527 | return en; | |
528 | } | |
529 | p = &(*p)->rb_left; | |
530 | } else if (ei->fofs >= en->ei.fofs + en->ei.len) { | |
531 | if (__is_back_mergeable(ei, &en->ei)) { | |
532 | f2fs_bug_on(sbi, !den); | |
533 | en->ei.len += ei->len; | |
534 | *den = __try_front_merge(sbi, et, en); | |
535 | return en; | |
536 | } | |
537 | p = &(*p)->rb_right; | |
538 | } else { | |
539 | f2fs_bug_on(sbi, 1); | |
540 | } | |
541 | } | |
542 | ||
543 | return __attach_extent_node(sbi, et, ei, parent, p); | |
544 | } | |
545 | ||
546 | static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi, | |
547 | struct extent_tree *et, bool free_all) | |
548 | { | |
549 | struct rb_node *node, *next; | |
550 | struct extent_node *en; | |
551 | unsigned int count = et->count; | |
552 | ||
553 | node = rb_first(&et->root); | |
554 | while (node) { | |
555 | next = rb_next(node); | |
556 | en = rb_entry(node, struct extent_node, rb_node); | |
557 | ||
558 | if (free_all) { | |
559 | spin_lock(&sbi->extent_lock); | |
560 | if (!list_empty(&en->list)) | |
561 | list_del_init(&en->list); | |
562 | spin_unlock(&sbi->extent_lock); | |
563 | } | |
564 | ||
565 | if (free_all || list_empty(&en->list)) { | |
566 | __detach_extent_node(sbi, et, en); | |
567 | kmem_cache_free(extent_node_slab, en); | |
568 | } | |
569 | node = next; | |
570 | } | |
571 | ||
572 | return count - et->count; | |
573 | } | |
574 | ||
028a41e8 CY |
575 | static void f2fs_init_extent_tree(struct inode *inode, |
576 | struct f2fs_extent *i_ext) | |
577 | { | |
578 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
579 | struct extent_tree *et; | |
580 | struct extent_node *en; | |
581 | struct extent_info ei; | |
582 | ||
583 | if (le32_to_cpu(i_ext->len) < F2FS_MIN_EXTENT_LEN) | |
584 | return; | |
585 | ||
586 | et = __grab_extent_tree(inode); | |
587 | ||
588 | write_lock(&et->lock); | |
589 | if (et->count) | |
590 | goto out; | |
591 | ||
592 | set_extent_info(&ei, le32_to_cpu(i_ext->fofs), | |
593 | le32_to_cpu(i_ext->blk), le32_to_cpu(i_ext->len)); | |
594 | ||
595 | en = __insert_extent_tree(sbi, et, &ei, NULL); | |
596 | if (en) { | |
597 | et->cached_en = en; | |
598 | ||
599 | spin_lock(&sbi->extent_lock); | |
600 | list_add_tail(&en->list, &sbi->extent_list); | |
601 | spin_unlock(&sbi->extent_lock); | |
602 | } | |
603 | out: | |
604 | write_unlock(&et->lock); | |
605 | atomic_dec(&et->refcount); | |
606 | } | |
607 | ||
429511cd CY |
608 | static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs, |
609 | struct extent_info *ei) | |
610 | { | |
611 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
612 | struct extent_tree *et; | |
613 | struct extent_node *en; | |
614 | ||
1ec4610c CY |
615 | trace_f2fs_lookup_extent_tree_start(inode, pgofs); |
616 | ||
93dfc526 CY |
617 | et = __find_extent_tree(sbi, inode->i_ino); |
618 | if (!et) | |
429511cd | 619 | return false; |
429511cd CY |
620 | |
621 | read_lock(&et->lock); | |
622 | en = __lookup_extent_tree(et, pgofs); | |
623 | if (en) { | |
624 | *ei = en->ei; | |
625 | spin_lock(&sbi->extent_lock); | |
626 | if (!list_empty(&en->list)) | |
627 | list_move_tail(&en->list, &sbi->extent_list); | |
628 | spin_unlock(&sbi->extent_lock); | |
629 | stat_inc_read_hit(sbi->sb); | |
630 | } | |
631 | stat_inc_total_hit(sbi->sb); | |
632 | read_unlock(&et->lock); | |
633 | ||
1ec4610c CY |
634 | trace_f2fs_lookup_extent_tree_end(inode, pgofs, en); |
635 | ||
429511cd CY |
636 | atomic_dec(&et->refcount); |
637 | return en ? true : false; | |
638 | } | |
639 | ||
640 | static void f2fs_update_extent_tree(struct inode *inode, pgoff_t fofs, | |
641 | block_t blkaddr) | |
642 | { | |
643 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
429511cd CY |
644 | struct extent_tree *et; |
645 | struct extent_node *en = NULL, *en1 = NULL, *en2 = NULL, *en3 = NULL; | |
646 | struct extent_node *den = NULL; | |
647 | struct extent_info ei, dei; | |
648 | unsigned int endofs; | |
649 | ||
1ec4610c CY |
650 | trace_f2fs_update_extent_tree(inode, fofs, blkaddr); |
651 | ||
93dfc526 | 652 | et = __grab_extent_tree(inode); |
429511cd CY |
653 | |
654 | write_lock(&et->lock); | |
655 | ||
656 | /* 1. lookup and remove existing extent info in cache */ | |
657 | en = __lookup_extent_tree(et, fofs); | |
658 | if (!en) | |
659 | goto update_extent; | |
660 | ||
661 | dei = en->ei; | |
662 | __detach_extent_node(sbi, et, en); | |
663 | ||
664 | /* 2. if extent can be split more, split and insert the left part */ | |
665 | if (dei.len > 1) { | |
666 | /* insert left part of split extent into cache */ | |
667 | if (fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN) { | |
668 | set_extent_info(&ei, dei.fofs, dei.blk, | |
669 | fofs - dei.fofs); | |
670 | en1 = __insert_extent_tree(sbi, et, &ei, NULL); | |
671 | } | |
672 | ||
673 | /* insert right part of split extent into cache */ | |
674 | endofs = dei.fofs + dei.len - 1; | |
675 | if (endofs - fofs >= F2FS_MIN_EXTENT_LEN) { | |
676 | set_extent_info(&ei, fofs + 1, | |
677 | fofs - dei.fofs + dei.blk, endofs - fofs); | |
678 | en2 = __insert_extent_tree(sbi, et, &ei, NULL); | |
679 | } | |
680 | } | |
681 | ||
682 | update_extent: | |
683 | /* 3. update extent in extent cache */ | |
684 | if (blkaddr) { | |
685 | set_extent_info(&ei, fofs, blkaddr, 1); | |
686 | en3 = __insert_extent_tree(sbi, et, &ei, &den); | |
687 | } | |
688 | ||
689 | /* 4. update in global extent list */ | |
690 | spin_lock(&sbi->extent_lock); | |
691 | if (en && !list_empty(&en->list)) | |
692 | list_del(&en->list); | |
693 | /* | |
694 | * en1 and en2 split from en, they will become more and more smaller | |
695 | * fragments after splitting several times. So if the length is smaller | |
696 | * than F2FS_MIN_EXTENT_LEN, we will not add them into extent tree. | |
697 | */ | |
698 | if (en1) | |
699 | list_add_tail(&en1->list, &sbi->extent_list); | |
700 | if (en2) | |
701 | list_add_tail(&en2->list, &sbi->extent_list); | |
702 | if (en3) { | |
703 | if (list_empty(&en3->list)) | |
704 | list_add_tail(&en3->list, &sbi->extent_list); | |
705 | else | |
706 | list_move_tail(&en3->list, &sbi->extent_list); | |
707 | } | |
708 | if (den && !list_empty(&den->list)) | |
709 | list_del(&den->list); | |
710 | spin_unlock(&sbi->extent_lock); | |
711 | ||
712 | /* 5. release extent node */ | |
713 | if (en) | |
714 | kmem_cache_free(extent_node_slab, en); | |
715 | if (den) | |
716 | kmem_cache_free(extent_node_slab, den); | |
717 | ||
718 | write_unlock(&et->lock); | |
719 | atomic_dec(&et->refcount); | |
720 | } | |
721 | ||
0bdee482 CY |
722 | void f2fs_preserve_extent_tree(struct inode *inode) |
723 | { | |
724 | struct extent_tree *et; | |
725 | struct extent_info *ext = &F2FS_I(inode)->ext; | |
726 | bool sync = false; | |
727 | ||
728 | if (!test_opt(F2FS_I_SB(inode), EXTENT_CACHE)) | |
729 | return; | |
730 | ||
731 | et = __find_extent_tree(F2FS_I_SB(inode), inode->i_ino); | |
732 | if (!et) { | |
733 | if (ext->len) { | |
734 | ext->len = 0; | |
735 | update_inode_page(inode); | |
736 | } | |
737 | return; | |
738 | } | |
739 | ||
740 | read_lock(&et->lock); | |
741 | if (et->count) { | |
742 | struct extent_node *en; | |
743 | ||
744 | if (et->cached_en) { | |
745 | en = et->cached_en; | |
746 | } else { | |
747 | struct rb_node *node = rb_first(&et->root); | |
748 | ||
749 | if (!node) | |
750 | node = rb_last(&et->root); | |
751 | en = rb_entry(node, struct extent_node, rb_node); | |
752 | } | |
753 | ||
754 | if (__is_extent_same(ext, &en->ei)) | |
755 | goto out; | |
756 | ||
757 | *ext = en->ei; | |
758 | sync = true; | |
759 | } else if (ext->len) { | |
760 | ext->len = 0; | |
761 | sync = true; | |
762 | } | |
763 | out: | |
764 | read_unlock(&et->lock); | |
765 | atomic_dec(&et->refcount); | |
766 | ||
767 | if (sync) | |
768 | update_inode_page(inode); | |
769 | } | |
770 | ||
429511cd CY |
771 | void f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink) |
772 | { | |
773 | struct extent_tree *treevec[EXT_TREE_VEC_SIZE]; | |
774 | struct extent_node *en, *tmp; | |
775 | unsigned long ino = F2FS_ROOT_INO(sbi); | |
776 | struct radix_tree_iter iter; | |
777 | void **slot; | |
778 | unsigned int found; | |
1ec4610c | 779 | unsigned int node_cnt = 0, tree_cnt = 0; |
429511cd | 780 | |
1dcc336b CY |
781 | if (!test_opt(sbi, EXTENT_CACHE)) |
782 | return; | |
783 | ||
429511cd CY |
784 | if (available_free_memory(sbi, EXTENT_CACHE)) |
785 | return; | |
786 | ||
787 | spin_lock(&sbi->extent_lock); | |
788 | list_for_each_entry_safe(en, tmp, &sbi->extent_list, list) { | |
789 | if (!nr_shrink--) | |
790 | break; | |
791 | list_del_init(&en->list); | |
792 | } | |
793 | spin_unlock(&sbi->extent_lock); | |
794 | ||
795 | down_read(&sbi->extent_tree_lock); | |
796 | while ((found = radix_tree_gang_lookup(&sbi->extent_tree_root, | |
797 | (void **)treevec, ino, EXT_TREE_VEC_SIZE))) { | |
798 | unsigned i; | |
799 | ||
800 | ino = treevec[found - 1]->ino + 1; | |
801 | for (i = 0; i < found; i++) { | |
802 | struct extent_tree *et = treevec[i]; | |
803 | ||
804 | atomic_inc(&et->refcount); | |
805 | write_lock(&et->lock); | |
1ec4610c | 806 | node_cnt += __free_extent_tree(sbi, et, false); |
429511cd CY |
807 | write_unlock(&et->lock); |
808 | atomic_dec(&et->refcount); | |
809 | } | |
810 | } | |
811 | up_read(&sbi->extent_tree_lock); | |
812 | ||
813 | down_write(&sbi->extent_tree_lock); | |
814 | radix_tree_for_each_slot(slot, &sbi->extent_tree_root, &iter, | |
815 | F2FS_ROOT_INO(sbi)) { | |
816 | struct extent_tree *et = (struct extent_tree *)*slot; | |
817 | ||
818 | if (!atomic_read(&et->refcount) && !et->count) { | |
819 | radix_tree_delete(&sbi->extent_tree_root, et->ino); | |
820 | kmem_cache_free(extent_tree_slab, et); | |
821 | sbi->total_ext_tree--; | |
1ec4610c | 822 | tree_cnt++; |
429511cd CY |
823 | } |
824 | } | |
825 | up_write(&sbi->extent_tree_lock); | |
1ec4610c CY |
826 | |
827 | trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt); | |
429511cd CY |
828 | } |
829 | ||
830 | void f2fs_destroy_extent_tree(struct inode *inode) | |
831 | { | |
832 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
833 | struct extent_tree *et; | |
1ec4610c | 834 | unsigned int node_cnt = 0; |
429511cd | 835 | |
1dcc336b CY |
836 | if (!test_opt(sbi, EXTENT_CACHE)) |
837 | return; | |
838 | ||
93dfc526 CY |
839 | et = __find_extent_tree(sbi, inode->i_ino); |
840 | if (!et) | |
429511cd | 841 | goto out; |
429511cd CY |
842 | |
843 | /* free all extent info belong to this extent tree */ | |
844 | write_lock(&et->lock); | |
1ec4610c | 845 | node_cnt = __free_extent_tree(sbi, et, true); |
429511cd CY |
846 | write_unlock(&et->lock); |
847 | ||
848 | atomic_dec(&et->refcount); | |
849 | ||
850 | /* try to find and delete extent tree entry in radix tree */ | |
851 | down_write(&sbi->extent_tree_lock); | |
852 | et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino); | |
853 | if (!et) { | |
854 | up_write(&sbi->extent_tree_lock); | |
855 | goto out; | |
856 | } | |
857 | f2fs_bug_on(sbi, atomic_read(&et->refcount) || et->count); | |
858 | radix_tree_delete(&sbi->extent_tree_root, inode->i_ino); | |
859 | kmem_cache_free(extent_tree_slab, et); | |
860 | sbi->total_ext_tree--; | |
861 | up_write(&sbi->extent_tree_lock); | |
862 | out: | |
1ec4610c | 863 | trace_f2fs_destroy_extent_tree(inode, node_cnt); |
429511cd CY |
864 | return; |
865 | } | |
866 | ||
028a41e8 CY |
867 | void f2fs_init_extent_cache(struct inode *inode, struct f2fs_extent *i_ext) |
868 | { | |
869 | if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE)) | |
870 | f2fs_init_extent_tree(inode, i_ext); | |
871 | ||
872 | write_lock(&F2FS_I(inode)->ext_lock); | |
873 | get_extent_info(&F2FS_I(inode)->ext, *i_ext); | |
874 | write_unlock(&F2FS_I(inode)->ext_lock); | |
875 | } | |
876 | ||
7e4dde79 CY |
877 | static bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs, |
878 | struct extent_info *ei) | |
879 | { | |
91c5d9bc CY |
880 | if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT)) |
881 | return false; | |
882 | ||
1dcc336b CY |
883 | if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE)) |
884 | return f2fs_lookup_extent_tree(inode, pgofs, ei); | |
885 | ||
7e4dde79 CY |
886 | return lookup_extent_info(inode, pgofs, ei); |
887 | } | |
888 | ||
889 | void f2fs_update_extent_cache(struct dnode_of_data *dn) | |
890 | { | |
891 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); | |
892 | pgoff_t fofs; | |
893 | ||
894 | f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR); | |
895 | ||
91c5d9bc CY |
896 | if (is_inode_flag_set(fi, FI_NO_EXTENT)) |
897 | return; | |
898 | ||
7e4dde79 CY |
899 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + |
900 | dn->ofs_in_node; | |
901 | ||
1dcc336b CY |
902 | if (test_opt(F2FS_I_SB(dn->inode), EXTENT_CACHE)) |
903 | return f2fs_update_extent_tree(dn->inode, fofs, | |
904 | dn->data_blkaddr); | |
905 | ||
7e4dde79 | 906 | if (update_extent_info(dn->inode, fofs, dn->data_blkaddr)) |
c11abd1a | 907 | sync_inode_page(dn); |
eb47b800 JK |
908 | } |
909 | ||
c718379b | 910 | struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync) |
eb47b800 | 911 | { |
eb47b800 JK |
912 | struct address_space *mapping = inode->i_mapping; |
913 | struct dnode_of_data dn; | |
914 | struct page *page; | |
cb3bc9ee | 915 | struct extent_info ei; |
eb47b800 | 916 | int err; |
cf04e8eb JK |
917 | struct f2fs_io_info fio = { |
918 | .type = DATA, | |
919 | .rw = sync ? READ_SYNC : READA, | |
920 | }; | |
eb47b800 | 921 | |
b7f204cc JK |
922 | /* |
923 | * If sync is false, it needs to check its block allocation. | |
924 | * This is need and triggered by two flows: | |
925 | * gc and truncate_partial_data_page. | |
926 | */ | |
927 | if (!sync) | |
928 | goto search; | |
929 | ||
eb47b800 JK |
930 | page = find_get_page(mapping, index); |
931 | if (page && PageUptodate(page)) | |
932 | return page; | |
933 | f2fs_put_page(page, 0); | |
b7f204cc | 934 | search: |
cb3bc9ee CY |
935 | if (f2fs_lookup_extent_cache(inode, index, &ei)) { |
936 | dn.data_blkaddr = ei.blk + index - ei.fofs; | |
937 | goto got_it; | |
938 | } | |
939 | ||
eb47b800 | 940 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 941 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
eb47b800 JK |
942 | if (err) |
943 | return ERR_PTR(err); | |
944 | f2fs_put_dnode(&dn); | |
945 | ||
946 | if (dn.data_blkaddr == NULL_ADDR) | |
947 | return ERR_PTR(-ENOENT); | |
948 | ||
949 | /* By fallocate(), there is no cached page, but with NEW_ADDR */ | |
6bacf52f | 950 | if (unlikely(dn.data_blkaddr == NEW_ADDR)) |
eb47b800 JK |
951 | return ERR_PTR(-EINVAL); |
952 | ||
cb3bc9ee | 953 | got_it: |
9ac1349a | 954 | page = grab_cache_page(mapping, index); |
eb47b800 JK |
955 | if (!page) |
956 | return ERR_PTR(-ENOMEM); | |
957 | ||
393ff91f JK |
958 | if (PageUptodate(page)) { |
959 | unlock_page(page); | |
960 | return page; | |
961 | } | |
962 | ||
cf04e8eb JK |
963 | fio.blk_addr = dn.data_blkaddr; |
964 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
1069bbf7 CY |
965 | if (err) |
966 | return ERR_PTR(err); | |
967 | ||
c718379b JK |
968 | if (sync) { |
969 | wait_on_page_locked(page); | |
6bacf52f | 970 | if (unlikely(!PageUptodate(page))) { |
c718379b JK |
971 | f2fs_put_page(page, 0); |
972 | return ERR_PTR(-EIO); | |
973 | } | |
eb47b800 | 974 | } |
eb47b800 JK |
975 | return page; |
976 | } | |
977 | ||
0a8165d7 | 978 | /* |
eb47b800 JK |
979 | * If it tries to access a hole, return an error. |
980 | * Because, the callers, functions in dir.c and GC, should be able to know | |
981 | * whether this page exists or not. | |
982 | */ | |
983 | struct page *get_lock_data_page(struct inode *inode, pgoff_t index) | |
984 | { | |
eb47b800 JK |
985 | struct address_space *mapping = inode->i_mapping; |
986 | struct dnode_of_data dn; | |
987 | struct page *page; | |
cb3bc9ee | 988 | struct extent_info ei; |
eb47b800 | 989 | int err; |
cf04e8eb JK |
990 | struct f2fs_io_info fio = { |
991 | .type = DATA, | |
992 | .rw = READ_SYNC, | |
993 | }; | |
650495de | 994 | repeat: |
9ac1349a | 995 | page = grab_cache_page(mapping, index); |
650495de JK |
996 | if (!page) |
997 | return ERR_PTR(-ENOMEM); | |
998 | ||
cb3bc9ee CY |
999 | if (f2fs_lookup_extent_cache(inode, index, &ei)) { |
1000 | dn.data_blkaddr = ei.blk + index - ei.fofs; | |
1001 | goto got_it; | |
1002 | } | |
1003 | ||
eb47b800 | 1004 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
266e97a8 | 1005 | err = get_dnode_of_data(&dn, index, LOOKUP_NODE); |
650495de JK |
1006 | if (err) { |
1007 | f2fs_put_page(page, 1); | |
eb47b800 | 1008 | return ERR_PTR(err); |
650495de | 1009 | } |
eb47b800 JK |
1010 | f2fs_put_dnode(&dn); |
1011 | ||
6bacf52f | 1012 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
650495de | 1013 | f2fs_put_page(page, 1); |
eb47b800 | 1014 | return ERR_PTR(-ENOENT); |
650495de | 1015 | } |
eb47b800 | 1016 | |
cb3bc9ee | 1017 | got_it: |
eb47b800 JK |
1018 | if (PageUptodate(page)) |
1019 | return page; | |
1020 | ||
d59ff4df JK |
1021 | /* |
1022 | * A new dentry page is allocated but not able to be written, since its | |
1023 | * new inode page couldn't be allocated due to -ENOSPC. | |
1024 | * In such the case, its blkaddr can be remained as NEW_ADDR. | |
1025 | * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata. | |
1026 | */ | |
1027 | if (dn.data_blkaddr == NEW_ADDR) { | |
1028 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
1029 | SetPageUptodate(page); | |
1030 | return page; | |
1031 | } | |
eb47b800 | 1032 | |
cf04e8eb JK |
1033 | fio.blk_addr = dn.data_blkaddr; |
1034 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
393ff91f | 1035 | if (err) |
eb47b800 | 1036 | return ERR_PTR(err); |
393ff91f JK |
1037 | |
1038 | lock_page(page); | |
6bacf52f | 1039 | if (unlikely(!PageUptodate(page))) { |
393ff91f JK |
1040 | f2fs_put_page(page, 1); |
1041 | return ERR_PTR(-EIO); | |
eb47b800 | 1042 | } |
6bacf52f | 1043 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
1044 | f2fs_put_page(page, 1); |
1045 | goto repeat; | |
eb47b800 JK |
1046 | } |
1047 | return page; | |
1048 | } | |
1049 | ||
0a8165d7 | 1050 | /* |
eb47b800 JK |
1051 | * Caller ensures that this data page is never allocated. |
1052 | * A new zero-filled data page is allocated in the page cache. | |
39936837 | 1053 | * |
4f4124d0 CY |
1054 | * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and |
1055 | * f2fs_unlock_op(). | |
a8865372 | 1056 | * Note that, ipage is set only by make_empty_dir. |
eb47b800 | 1057 | */ |
64aa7ed9 | 1058 | struct page *get_new_data_page(struct inode *inode, |
a8865372 | 1059 | struct page *ipage, pgoff_t index, bool new_i_size) |
eb47b800 | 1060 | { |
eb47b800 JK |
1061 | struct address_space *mapping = inode->i_mapping; |
1062 | struct page *page; | |
1063 | struct dnode_of_data dn; | |
1064 | int err; | |
1065 | ||
a8865372 | 1066 | set_new_dnode(&dn, inode, ipage, NULL, 0); |
b600965c | 1067 | err = f2fs_reserve_block(&dn, index); |
eb47b800 JK |
1068 | if (err) |
1069 | return ERR_PTR(err); | |
afcb7ca0 | 1070 | repeat: |
eb47b800 | 1071 | page = grab_cache_page(mapping, index); |
a8865372 JK |
1072 | if (!page) { |
1073 | err = -ENOMEM; | |
1074 | goto put_err; | |
1075 | } | |
eb47b800 JK |
1076 | |
1077 | if (PageUptodate(page)) | |
1078 | return page; | |
1079 | ||
1080 | if (dn.data_blkaddr == NEW_ADDR) { | |
1081 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
393ff91f | 1082 | SetPageUptodate(page); |
eb47b800 | 1083 | } else { |
cf04e8eb JK |
1084 | struct f2fs_io_info fio = { |
1085 | .type = DATA, | |
1086 | .rw = READ_SYNC, | |
1087 | .blk_addr = dn.data_blkaddr, | |
1088 | }; | |
1089 | err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio); | |
393ff91f | 1090 | if (err) |
a8865372 JK |
1091 | goto put_err; |
1092 | ||
393ff91f | 1093 | lock_page(page); |
6bacf52f | 1094 | if (unlikely(!PageUptodate(page))) { |
393ff91f | 1095 | f2fs_put_page(page, 1); |
a8865372 JK |
1096 | err = -EIO; |
1097 | goto put_err; | |
eb47b800 | 1098 | } |
6bacf52f | 1099 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
1100 | f2fs_put_page(page, 1); |
1101 | goto repeat; | |
eb47b800 JK |
1102 | } |
1103 | } | |
eb47b800 JK |
1104 | |
1105 | if (new_i_size && | |
1106 | i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) { | |
1107 | i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT)); | |
699489bb JK |
1108 | /* Only the directory inode sets new_i_size */ |
1109 | set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR); | |
eb47b800 JK |
1110 | } |
1111 | return page; | |
a8865372 JK |
1112 | |
1113 | put_err: | |
1114 | f2fs_put_dnode(&dn); | |
1115 | return ERR_PTR(err); | |
eb47b800 JK |
1116 | } |
1117 | ||
bfad7c2d JK |
1118 | static int __allocate_data_block(struct dnode_of_data *dn) |
1119 | { | |
4081363f | 1120 | struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); |
976e4c50 | 1121 | struct f2fs_inode_info *fi = F2FS_I(dn->inode); |
bfad7c2d | 1122 | struct f2fs_summary sum; |
bfad7c2d | 1123 | struct node_info ni; |
38aa0889 | 1124 | int seg = CURSEG_WARM_DATA; |
976e4c50 | 1125 | pgoff_t fofs; |
bfad7c2d JK |
1126 | |
1127 | if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))) | |
1128 | return -EPERM; | |
1129 | if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1))) | |
1130 | return -ENOSPC; | |
1131 | ||
bfad7c2d JK |
1132 | get_node_info(sbi, dn->nid, &ni); |
1133 | set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); | |
1134 | ||
38aa0889 JK |
1135 | if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page) |
1136 | seg = CURSEG_DIRECT_IO; | |
1137 | ||
1138 | allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg); | |
bfad7c2d JK |
1139 | |
1140 | /* direct IO doesn't use extent cache to maximize the performance */ | |
216a620a | 1141 | set_data_blkaddr(dn); |
bfad7c2d | 1142 | |
976e4c50 JK |
1143 | /* update i_size */ |
1144 | fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) + | |
1145 | dn->ofs_in_node; | |
1146 | if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT)) | |
1147 | i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT)); | |
1148 | ||
bfad7c2d JK |
1149 | return 0; |
1150 | } | |
1151 | ||
59b802e5 JK |
1152 | static void __allocate_data_blocks(struct inode *inode, loff_t offset, |
1153 | size_t count) | |
1154 | { | |
1155 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
1156 | struct dnode_of_data dn; | |
1157 | u64 start = F2FS_BYTES_TO_BLK(offset); | |
1158 | u64 len = F2FS_BYTES_TO_BLK(count); | |
1159 | bool allocated; | |
1160 | u64 end_offset; | |
1161 | ||
1162 | while (len) { | |
1163 | f2fs_balance_fs(sbi); | |
1164 | f2fs_lock_op(sbi); | |
1165 | ||
1166 | /* When reading holes, we need its node page */ | |
1167 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
1168 | if (get_dnode_of_data(&dn, start, ALLOC_NODE)) | |
1169 | goto out; | |
1170 | ||
1171 | allocated = false; | |
1172 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); | |
1173 | ||
1174 | while (dn.ofs_in_node < end_offset && len) { | |
d6d4f1cb CY |
1175 | block_t blkaddr; |
1176 | ||
1177 | blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node); | |
1178 | if (blkaddr == NULL_ADDR) { | |
59b802e5 JK |
1179 | if (__allocate_data_block(&dn)) |
1180 | goto sync_out; | |
1181 | allocated = true; | |
1182 | } | |
1183 | len--; | |
1184 | start++; | |
1185 | dn.ofs_in_node++; | |
1186 | } | |
1187 | ||
1188 | if (allocated) | |
1189 | sync_inode_page(&dn); | |
1190 | ||
1191 | f2fs_put_dnode(&dn); | |
1192 | f2fs_unlock_op(sbi); | |
1193 | } | |
1194 | return; | |
1195 | ||
1196 | sync_out: | |
1197 | if (allocated) | |
1198 | sync_inode_page(&dn); | |
1199 | f2fs_put_dnode(&dn); | |
1200 | out: | |
1201 | f2fs_unlock_op(sbi); | |
1202 | return; | |
1203 | } | |
1204 | ||
0a8165d7 | 1205 | /* |
4f4124d0 CY |
1206 | * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh. |
1207 | * If original data blocks are allocated, then give them to blockdev. | |
1208 | * Otherwise, | |
1209 | * a. preallocate requested block addresses | |
1210 | * b. do not use extent cache for better performance | |
1211 | * c. give the block addresses to blockdev | |
eb47b800 | 1212 | */ |
ccfb3000 JK |
1213 | static int __get_data_block(struct inode *inode, sector_t iblock, |
1214 | struct buffer_head *bh_result, int create, bool fiemap) | |
eb47b800 JK |
1215 | { |
1216 | unsigned int blkbits = inode->i_sb->s_blocksize_bits; | |
1217 | unsigned maxblocks = bh_result->b_size >> blkbits; | |
1218 | struct dnode_of_data dn; | |
bfad7c2d JK |
1219 | int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA; |
1220 | pgoff_t pgofs, end_offset; | |
1221 | int err = 0, ofs = 1; | |
a2e7d1bf | 1222 | struct extent_info ei; |
bfad7c2d | 1223 | bool allocated = false; |
eb47b800 JK |
1224 | |
1225 | /* Get the page offset from the block offset(iblock) */ | |
1226 | pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits)); | |
1227 | ||
7e4dde79 | 1228 | if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) { |
a2e7d1bf | 1229 | f2fs_map_bh(inode->i_sb, pgofs, &ei, bh_result); |
bfad7c2d | 1230 | goto out; |
a2e7d1bf | 1231 | } |
bfad7c2d | 1232 | |
59b802e5 | 1233 | if (create) |
4081363f | 1234 | f2fs_lock_op(F2FS_I_SB(inode)); |
eb47b800 JK |
1235 | |
1236 | /* When reading holes, we need its node page */ | |
1237 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
bfad7c2d | 1238 | err = get_dnode_of_data(&dn, pgofs, mode); |
1ec79083 | 1239 | if (err) { |
bfad7c2d JK |
1240 | if (err == -ENOENT) |
1241 | err = 0; | |
1242 | goto unlock_out; | |
848753aa | 1243 | } |
ccfb3000 | 1244 | if (dn.data_blkaddr == NEW_ADDR && !fiemap) |
1ec79083 | 1245 | goto put_out; |
eb47b800 | 1246 | |
bfad7c2d | 1247 | if (dn.data_blkaddr != NULL_ADDR) { |
3402e87c | 1248 | clear_buffer_new(bh_result); |
bfad7c2d JK |
1249 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); |
1250 | } else if (create) { | |
1251 | err = __allocate_data_block(&dn); | |
1252 | if (err) | |
1253 | goto put_out; | |
1254 | allocated = true; | |
da17eece | 1255 | set_buffer_new(bh_result); |
bfad7c2d JK |
1256 | map_bh(bh_result, inode->i_sb, dn.data_blkaddr); |
1257 | } else { | |
1258 | goto put_out; | |
1259 | } | |
1260 | ||
6403eb1f | 1261 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); |
bfad7c2d JK |
1262 | bh_result->b_size = (((size_t)1) << blkbits); |
1263 | dn.ofs_in_node++; | |
1264 | pgofs++; | |
1265 | ||
1266 | get_next: | |
1267 | if (dn.ofs_in_node >= end_offset) { | |
1268 | if (allocated) | |
1269 | sync_inode_page(&dn); | |
1270 | allocated = false; | |
1271 | f2fs_put_dnode(&dn); | |
1272 | ||
1273 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
1274 | err = get_dnode_of_data(&dn, pgofs, mode); | |
1ec79083 | 1275 | if (err) { |
bfad7c2d JK |
1276 | if (err == -ENOENT) |
1277 | err = 0; | |
1278 | goto unlock_out; | |
1279 | } | |
ccfb3000 | 1280 | if (dn.data_blkaddr == NEW_ADDR && !fiemap) |
1ec79083 JK |
1281 | goto put_out; |
1282 | ||
6403eb1f | 1283 | end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode)); |
bfad7c2d | 1284 | } |
eb47b800 | 1285 | |
bfad7c2d JK |
1286 | if (maxblocks > (bh_result->b_size >> blkbits)) { |
1287 | block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node); | |
1288 | if (blkaddr == NULL_ADDR && create) { | |
1289 | err = __allocate_data_block(&dn); | |
1290 | if (err) | |
1291 | goto sync_out; | |
1292 | allocated = true; | |
3402e87c | 1293 | set_buffer_new(bh_result); |
bfad7c2d JK |
1294 | blkaddr = dn.data_blkaddr; |
1295 | } | |
e1c42045 | 1296 | /* Give more consecutive addresses for the readahead */ |
bfad7c2d JK |
1297 | if (blkaddr == (bh_result->b_blocknr + ofs)) { |
1298 | ofs++; | |
1299 | dn.ofs_in_node++; | |
1300 | pgofs++; | |
1301 | bh_result->b_size += (((size_t)1) << blkbits); | |
1302 | goto get_next; | |
1303 | } | |
eb47b800 | 1304 | } |
bfad7c2d JK |
1305 | sync_out: |
1306 | if (allocated) | |
1307 | sync_inode_page(&dn); | |
1308 | put_out: | |
eb47b800 | 1309 | f2fs_put_dnode(&dn); |
bfad7c2d JK |
1310 | unlock_out: |
1311 | if (create) | |
4081363f | 1312 | f2fs_unlock_op(F2FS_I_SB(inode)); |
bfad7c2d JK |
1313 | out: |
1314 | trace_f2fs_get_data_block(inode, iblock, bh_result, err); | |
1315 | return err; | |
eb47b800 JK |
1316 | } |
1317 | ||
ccfb3000 JK |
1318 | static int get_data_block(struct inode *inode, sector_t iblock, |
1319 | struct buffer_head *bh_result, int create) | |
1320 | { | |
1321 | return __get_data_block(inode, iblock, bh_result, create, false); | |
1322 | } | |
1323 | ||
1324 | static int get_data_block_fiemap(struct inode *inode, sector_t iblock, | |
1325 | struct buffer_head *bh_result, int create) | |
1326 | { | |
1327 | return __get_data_block(inode, iblock, bh_result, create, true); | |
1328 | } | |
1329 | ||
9ab70134 JK |
1330 | int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
1331 | u64 start, u64 len) | |
1332 | { | |
ccfb3000 JK |
1333 | return generic_block_fiemap(inode, fieinfo, |
1334 | start, len, get_data_block_fiemap); | |
9ab70134 JK |
1335 | } |
1336 | ||
eb47b800 JK |
1337 | static int f2fs_read_data_page(struct file *file, struct page *page) |
1338 | { | |
9ffe0fb5 | 1339 | struct inode *inode = page->mapping->host; |
b3d208f9 | 1340 | int ret = -EAGAIN; |
9ffe0fb5 | 1341 | |
c20e89cd CY |
1342 | trace_f2fs_readpage(page, DATA); |
1343 | ||
e1c42045 | 1344 | /* If the file has inline data, try to read it directly */ |
9ffe0fb5 HL |
1345 | if (f2fs_has_inline_data(inode)) |
1346 | ret = f2fs_read_inline_data(inode, page); | |
b3d208f9 | 1347 | if (ret == -EAGAIN) |
9ffe0fb5 HL |
1348 | ret = mpage_readpage(page, get_data_block); |
1349 | ||
1350 | return ret; | |
eb47b800 JK |
1351 | } |
1352 | ||
1353 | static int f2fs_read_data_pages(struct file *file, | |
1354 | struct address_space *mapping, | |
1355 | struct list_head *pages, unsigned nr_pages) | |
1356 | { | |
9ffe0fb5 HL |
1357 | struct inode *inode = file->f_mapping->host; |
1358 | ||
1359 | /* If the file has inline data, skip readpages */ | |
1360 | if (f2fs_has_inline_data(inode)) | |
1361 | return 0; | |
1362 | ||
bfad7c2d | 1363 | return mpage_readpages(mapping, pages, nr_pages, get_data_block); |
eb47b800 JK |
1364 | } |
1365 | ||
458e6197 | 1366 | int do_write_data_page(struct page *page, struct f2fs_io_info *fio) |
eb47b800 JK |
1367 | { |
1368 | struct inode *inode = page->mapping->host; | |
eb47b800 JK |
1369 | struct dnode_of_data dn; |
1370 | int err = 0; | |
1371 | ||
1372 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
266e97a8 | 1373 | err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE); |
eb47b800 JK |
1374 | if (err) |
1375 | return err; | |
1376 | ||
cf04e8eb | 1377 | fio->blk_addr = dn.data_blkaddr; |
eb47b800 JK |
1378 | |
1379 | /* This page is already truncated */ | |
2bca1e23 JK |
1380 | if (fio->blk_addr == NULL_ADDR) { |
1381 | ClearPageUptodate(page); | |
eb47b800 | 1382 | goto out_writepage; |
2bca1e23 | 1383 | } |
eb47b800 JK |
1384 | |
1385 | set_page_writeback(page); | |
1386 | ||
1387 | /* | |
1388 | * If current allocation needs SSR, | |
1389 | * it had better in-place writes for updated data. | |
1390 | */ | |
cf04e8eb | 1391 | if (unlikely(fio->blk_addr != NEW_ADDR && |
b25958b6 HL |
1392 | !is_cold_data(page) && |
1393 | need_inplace_update(inode))) { | |
cf04e8eb | 1394 | rewrite_data_page(page, fio); |
fff04f90 | 1395 | set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE); |
8ce67cb0 | 1396 | trace_f2fs_do_write_data_page(page, IPU); |
eb47b800 | 1397 | } else { |
cf04e8eb | 1398 | write_data_page(page, &dn, fio); |
216a620a | 1399 | set_data_blkaddr(&dn); |
7e4dde79 | 1400 | f2fs_update_extent_cache(&dn); |
8ce67cb0 | 1401 | trace_f2fs_do_write_data_page(page, OPU); |
fff04f90 | 1402 | set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE); |
3c6c2beb JK |
1403 | if (page->index == 0) |
1404 | set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN); | |
eb47b800 JK |
1405 | } |
1406 | out_writepage: | |
1407 | f2fs_put_dnode(&dn); | |
1408 | return err; | |
1409 | } | |
1410 | ||
1411 | static int f2fs_write_data_page(struct page *page, | |
1412 | struct writeback_control *wbc) | |
1413 | { | |
1414 | struct inode *inode = page->mapping->host; | |
4081363f | 1415 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
eb47b800 JK |
1416 | loff_t i_size = i_size_read(inode); |
1417 | const pgoff_t end_index = ((unsigned long long) i_size) | |
1418 | >> PAGE_CACHE_SHIFT; | |
9ffe0fb5 | 1419 | unsigned offset = 0; |
39936837 | 1420 | bool need_balance_fs = false; |
eb47b800 | 1421 | int err = 0; |
458e6197 JK |
1422 | struct f2fs_io_info fio = { |
1423 | .type = DATA, | |
6c311ec6 | 1424 | .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE, |
458e6197 | 1425 | }; |
eb47b800 | 1426 | |
ecda0de3 CY |
1427 | trace_f2fs_writepage(page, DATA); |
1428 | ||
eb47b800 | 1429 | if (page->index < end_index) |
39936837 | 1430 | goto write; |
eb47b800 JK |
1431 | |
1432 | /* | |
1433 | * If the offset is out-of-range of file size, | |
1434 | * this page does not have to be written to disk. | |
1435 | */ | |
1436 | offset = i_size & (PAGE_CACHE_SIZE - 1); | |
76f60268 | 1437 | if ((page->index >= end_index + 1) || !offset) |
39936837 | 1438 | goto out; |
eb47b800 JK |
1439 | |
1440 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
39936837 | 1441 | write: |
caf0047e | 1442 | if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) |
eb47b800 | 1443 | goto redirty_out; |
1e84371f JK |
1444 | if (f2fs_is_drop_cache(inode)) |
1445 | goto out; | |
1446 | if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim && | |
1447 | available_free_memory(sbi, BASE_CHECK)) | |
1448 | goto redirty_out; | |
eb47b800 | 1449 | |
39936837 | 1450 | /* Dentry blocks are controlled by checkpoint */ |
eb47b800 | 1451 | if (S_ISDIR(inode->i_mode)) { |
cf779cab JK |
1452 | if (unlikely(f2fs_cp_error(sbi))) |
1453 | goto redirty_out; | |
458e6197 | 1454 | err = do_write_data_page(page, &fio); |
8618b881 JK |
1455 | goto done; |
1456 | } | |
9ffe0fb5 | 1457 | |
cf779cab JK |
1458 | /* we should bypass data pages to proceed the kworkder jobs */ |
1459 | if (unlikely(f2fs_cp_error(sbi))) { | |
1460 | SetPageError(page); | |
a7ffdbe2 | 1461 | goto out; |
cf779cab JK |
1462 | } |
1463 | ||
8618b881 | 1464 | if (!wbc->for_reclaim) |
39936837 | 1465 | need_balance_fs = true; |
8618b881 | 1466 | else if (has_not_enough_free_secs(sbi, 0)) |
39936837 | 1467 | goto redirty_out; |
eb47b800 | 1468 | |
b3d208f9 | 1469 | err = -EAGAIN; |
8618b881 | 1470 | f2fs_lock_op(sbi); |
b3d208f9 JK |
1471 | if (f2fs_has_inline_data(inode)) |
1472 | err = f2fs_write_inline_data(inode, page); | |
1473 | if (err == -EAGAIN) | |
8618b881 JK |
1474 | err = do_write_data_page(page, &fio); |
1475 | f2fs_unlock_op(sbi); | |
1476 | done: | |
1477 | if (err && err != -ENOENT) | |
1478 | goto redirty_out; | |
eb47b800 | 1479 | |
eb47b800 | 1480 | clear_cold_data(page); |
39936837 | 1481 | out: |
a7ffdbe2 | 1482 | inode_dec_dirty_pages(inode); |
2bca1e23 JK |
1483 | if (err) |
1484 | ClearPageUptodate(page); | |
eb47b800 | 1485 | unlock_page(page); |
39936837 | 1486 | if (need_balance_fs) |
eb47b800 | 1487 | f2fs_balance_fs(sbi); |
2aea39ec JK |
1488 | if (wbc->for_reclaim) |
1489 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
eb47b800 JK |
1490 | return 0; |
1491 | ||
eb47b800 | 1492 | redirty_out: |
76f60268 | 1493 | redirty_page_for_writepage(wbc, page); |
8618b881 | 1494 | return AOP_WRITEPAGE_ACTIVATE; |
eb47b800 JK |
1495 | } |
1496 | ||
fa9150a8 NJ |
1497 | static int __f2fs_writepage(struct page *page, struct writeback_control *wbc, |
1498 | void *data) | |
1499 | { | |
1500 | struct address_space *mapping = data; | |
1501 | int ret = mapping->a_ops->writepage(page, wbc); | |
1502 | mapping_set_error(mapping, ret); | |
1503 | return ret; | |
1504 | } | |
1505 | ||
25ca923b | 1506 | static int f2fs_write_data_pages(struct address_space *mapping, |
eb47b800 JK |
1507 | struct writeback_control *wbc) |
1508 | { | |
1509 | struct inode *inode = mapping->host; | |
4081363f | 1510 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
eb47b800 | 1511 | int ret; |
50c8cdb3 | 1512 | long diff; |
eb47b800 | 1513 | |
e5748434 CY |
1514 | trace_f2fs_writepages(mapping->host, wbc, DATA); |
1515 | ||
cfb185a1 | 1516 | /* deal with chardevs and other special file */ |
1517 | if (!mapping->a_ops->writepage) | |
1518 | return 0; | |
1519 | ||
87d6f890 | 1520 | if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE && |
a7ffdbe2 | 1521 | get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) && |
6fb03f3a | 1522 | available_free_memory(sbi, DIRTY_DENTS)) |
d3baf95d | 1523 | goto skip_write; |
87d6f890 | 1524 | |
d5669f7b JK |
1525 | /* during POR, we don't need to trigger writepage at all. */ |
1526 | if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) | |
1527 | goto skip_write; | |
1528 | ||
50c8cdb3 | 1529 | diff = nr_pages_to_write(sbi, DATA, wbc); |
eb47b800 | 1530 | |
fa9150a8 | 1531 | ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping); |
458e6197 JK |
1532 | |
1533 | f2fs_submit_merged_bio(sbi, DATA, WRITE); | |
eb47b800 JK |
1534 | |
1535 | remove_dirty_dir_inode(inode); | |
1536 | ||
50c8cdb3 | 1537 | wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff); |
eb47b800 | 1538 | return ret; |
d3baf95d JK |
1539 | |
1540 | skip_write: | |
a7ffdbe2 | 1541 | wbc->pages_skipped += get_dirty_pages(inode); |
d3baf95d | 1542 | return 0; |
eb47b800 JK |
1543 | } |
1544 | ||
3aab8f82 CY |
1545 | static void f2fs_write_failed(struct address_space *mapping, loff_t to) |
1546 | { | |
1547 | struct inode *inode = mapping->host; | |
1548 | ||
1549 | if (to > inode->i_size) { | |
1550 | truncate_pagecache(inode, inode->i_size); | |
764aa3e9 | 1551 | truncate_blocks(inode, inode->i_size, true); |
3aab8f82 CY |
1552 | } |
1553 | } | |
1554 | ||
eb47b800 JK |
1555 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, |
1556 | loff_t pos, unsigned len, unsigned flags, | |
1557 | struct page **pagep, void **fsdata) | |
1558 | { | |
1559 | struct inode *inode = mapping->host; | |
4081363f | 1560 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
9ba69cf9 | 1561 | struct page *page, *ipage; |
eb47b800 JK |
1562 | pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT; |
1563 | struct dnode_of_data dn; | |
1564 | int err = 0; | |
1565 | ||
62aed044 CY |
1566 | trace_f2fs_write_begin(inode, pos, len, flags); |
1567 | ||
eb47b800 | 1568 | f2fs_balance_fs(sbi); |
5f727395 JK |
1569 | |
1570 | /* | |
1571 | * We should check this at this moment to avoid deadlock on inode page | |
1572 | * and #0 page. The locking rule for inline_data conversion should be: | |
1573 | * lock_page(page #0) -> lock_page(inode_page) | |
1574 | */ | |
1575 | if (index != 0) { | |
1576 | err = f2fs_convert_inline_inode(inode); | |
1577 | if (err) | |
1578 | goto fail; | |
1579 | } | |
afcb7ca0 | 1580 | repeat: |
eb47b800 | 1581 | page = grab_cache_page_write_begin(mapping, index, flags); |
3aab8f82 CY |
1582 | if (!page) { |
1583 | err = -ENOMEM; | |
1584 | goto fail; | |
1585 | } | |
d5f66990 | 1586 | |
eb47b800 JK |
1587 | *pagep = page; |
1588 | ||
e479556b | 1589 | f2fs_lock_op(sbi); |
9ba69cf9 JK |
1590 | |
1591 | /* check inline_data */ | |
1592 | ipage = get_node_page(sbi, inode->i_ino); | |
cd34e296 CY |
1593 | if (IS_ERR(ipage)) { |
1594 | err = PTR_ERR(ipage); | |
9ba69cf9 | 1595 | goto unlock_fail; |
cd34e296 | 1596 | } |
9ba69cf9 | 1597 | |
b3d208f9 JK |
1598 | set_new_dnode(&dn, inode, ipage, ipage, 0); |
1599 | ||
9ba69cf9 | 1600 | if (f2fs_has_inline_data(inode)) { |
b3d208f9 JK |
1601 | if (pos + len <= MAX_INLINE_DATA) { |
1602 | read_inline_data(page, ipage); | |
1603 | set_inode_flag(F2FS_I(inode), FI_DATA_EXIST); | |
1604 | sync_inode_page(&dn); | |
1605 | goto put_next; | |
b3d208f9 | 1606 | } |
5f727395 JK |
1607 | err = f2fs_convert_inline_page(&dn, page); |
1608 | if (err) | |
1609 | goto put_fail; | |
b600965c | 1610 | } |
9ba69cf9 JK |
1611 | err = f2fs_reserve_block(&dn, index); |
1612 | if (err) | |
8cdcb713 | 1613 | goto put_fail; |
b3d208f9 | 1614 | put_next: |
9ba69cf9 JK |
1615 | f2fs_put_dnode(&dn); |
1616 | f2fs_unlock_op(sbi); | |
1617 | ||
eb47b800 JK |
1618 | if ((len == PAGE_CACHE_SIZE) || PageUptodate(page)) |
1619 | return 0; | |
1620 | ||
b3d208f9 JK |
1621 | f2fs_wait_on_page_writeback(page, DATA); |
1622 | ||
eb47b800 JK |
1623 | if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) { |
1624 | unsigned start = pos & (PAGE_CACHE_SIZE - 1); | |
1625 | unsigned end = start + len; | |
1626 | ||
1627 | /* Reading beyond i_size is simple: memset to zero */ | |
1628 | zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE); | |
393ff91f | 1629 | goto out; |
eb47b800 JK |
1630 | } |
1631 | ||
b3d208f9 | 1632 | if (dn.data_blkaddr == NEW_ADDR) { |
eb47b800 JK |
1633 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); |
1634 | } else { | |
cf04e8eb JK |
1635 | struct f2fs_io_info fio = { |
1636 | .type = DATA, | |
1637 | .rw = READ_SYNC, | |
1638 | .blk_addr = dn.data_blkaddr, | |
1639 | }; | |
1640 | err = f2fs_submit_page_bio(sbi, page, &fio); | |
9234f319 JK |
1641 | if (err) |
1642 | goto fail; | |
d54c795b | 1643 | |
393ff91f | 1644 | lock_page(page); |
6bacf52f | 1645 | if (unlikely(!PageUptodate(page))) { |
393ff91f | 1646 | f2fs_put_page(page, 1); |
3aab8f82 CY |
1647 | err = -EIO; |
1648 | goto fail; | |
eb47b800 | 1649 | } |
6bacf52f | 1650 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
1651 | f2fs_put_page(page, 1); |
1652 | goto repeat; | |
eb47b800 JK |
1653 | } |
1654 | } | |
393ff91f | 1655 | out: |
eb47b800 JK |
1656 | SetPageUptodate(page); |
1657 | clear_cold_data(page); | |
1658 | return 0; | |
9ba69cf9 | 1659 | |
8cdcb713 JK |
1660 | put_fail: |
1661 | f2fs_put_dnode(&dn); | |
9ba69cf9 JK |
1662 | unlock_fail: |
1663 | f2fs_unlock_op(sbi); | |
b3d208f9 | 1664 | f2fs_put_page(page, 1); |
3aab8f82 CY |
1665 | fail: |
1666 | f2fs_write_failed(mapping, pos + len); | |
1667 | return err; | |
eb47b800 JK |
1668 | } |
1669 | ||
a1dd3c13 JK |
1670 | static int f2fs_write_end(struct file *file, |
1671 | struct address_space *mapping, | |
1672 | loff_t pos, unsigned len, unsigned copied, | |
1673 | struct page *page, void *fsdata) | |
1674 | { | |
1675 | struct inode *inode = page->mapping->host; | |
1676 | ||
dfb2bf38 CY |
1677 | trace_f2fs_write_end(inode, pos, len, copied); |
1678 | ||
34ba94ba | 1679 | set_page_dirty(page); |
a1dd3c13 JK |
1680 | |
1681 | if (pos + copied > i_size_read(inode)) { | |
1682 | i_size_write(inode, pos + copied); | |
1683 | mark_inode_dirty(inode); | |
1684 | update_inode_page(inode); | |
1685 | } | |
1686 | ||
75c3c8bc | 1687 | f2fs_put_page(page, 1); |
a1dd3c13 JK |
1688 | return copied; |
1689 | } | |
1690 | ||
944fcfc1 | 1691 | static int check_direct_IO(struct inode *inode, int rw, |
5b46f25d | 1692 | struct iov_iter *iter, loff_t offset) |
944fcfc1 JK |
1693 | { |
1694 | unsigned blocksize_mask = inode->i_sb->s_blocksize - 1; | |
944fcfc1 JK |
1695 | |
1696 | if (rw == READ) | |
1697 | return 0; | |
1698 | ||
1699 | if (offset & blocksize_mask) | |
1700 | return -EINVAL; | |
1701 | ||
5b46f25d AV |
1702 | if (iov_iter_alignment(iter) & blocksize_mask) |
1703 | return -EINVAL; | |
1704 | ||
944fcfc1 JK |
1705 | return 0; |
1706 | } | |
1707 | ||
eb47b800 | 1708 | static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb, |
d8d3d94b | 1709 | struct iov_iter *iter, loff_t offset) |
eb47b800 JK |
1710 | { |
1711 | struct file *file = iocb->ki_filp; | |
3aab8f82 CY |
1712 | struct address_space *mapping = file->f_mapping; |
1713 | struct inode *inode = mapping->host; | |
1714 | size_t count = iov_iter_count(iter); | |
1715 | int err; | |
944fcfc1 | 1716 | |
b3d208f9 JK |
1717 | /* we don't need to use inline_data strictly */ |
1718 | if (f2fs_has_inline_data(inode)) { | |
1719 | err = f2fs_convert_inline_inode(inode); | |
1720 | if (err) | |
1721 | return err; | |
1722 | } | |
9ffe0fb5 | 1723 | |
5b46f25d | 1724 | if (check_direct_IO(inode, rw, iter, offset)) |
944fcfc1 JK |
1725 | return 0; |
1726 | ||
70407fad CY |
1727 | trace_f2fs_direct_IO_enter(inode, offset, count, rw); |
1728 | ||
59b802e5 JK |
1729 | if (rw & WRITE) |
1730 | __allocate_data_blocks(inode, offset, count); | |
1731 | ||
3aab8f82 CY |
1732 | err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block); |
1733 | if (err < 0 && (rw & WRITE)) | |
1734 | f2fs_write_failed(mapping, offset + count); | |
70407fad CY |
1735 | |
1736 | trace_f2fs_direct_IO_exit(inode, offset, count, rw, err); | |
1737 | ||
3aab8f82 | 1738 | return err; |
eb47b800 JK |
1739 | } |
1740 | ||
487261f3 CY |
1741 | void f2fs_invalidate_page(struct page *page, unsigned int offset, |
1742 | unsigned int length) | |
eb47b800 JK |
1743 | { |
1744 | struct inode *inode = page->mapping->host; | |
487261f3 | 1745 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
a7ffdbe2 | 1746 | |
487261f3 CY |
1747 | if (inode->i_ino >= F2FS_ROOT_INO(sbi) && |
1748 | (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE)) | |
a7ffdbe2 JK |
1749 | return; |
1750 | ||
487261f3 CY |
1751 | if (PageDirty(page)) { |
1752 | if (inode->i_ino == F2FS_META_INO(sbi)) | |
1753 | dec_page_count(sbi, F2FS_DIRTY_META); | |
1754 | else if (inode->i_ino == F2FS_NODE_INO(sbi)) | |
1755 | dec_page_count(sbi, F2FS_DIRTY_NODES); | |
1756 | else | |
1757 | inode_dec_dirty_pages(inode); | |
1758 | } | |
eb47b800 JK |
1759 | ClearPagePrivate(page); |
1760 | } | |
1761 | ||
487261f3 | 1762 | int f2fs_release_page(struct page *page, gfp_t wait) |
eb47b800 | 1763 | { |
f68daeeb JK |
1764 | /* If this is dirty page, keep PagePrivate */ |
1765 | if (PageDirty(page)) | |
1766 | return 0; | |
1767 | ||
eb47b800 | 1768 | ClearPagePrivate(page); |
c3850aa1 | 1769 | return 1; |
eb47b800 JK |
1770 | } |
1771 | ||
1772 | static int f2fs_set_data_page_dirty(struct page *page) | |
1773 | { | |
1774 | struct address_space *mapping = page->mapping; | |
1775 | struct inode *inode = mapping->host; | |
1776 | ||
26c6b887 JK |
1777 | trace_f2fs_set_page_dirty(page, DATA); |
1778 | ||
eb47b800 | 1779 | SetPageUptodate(page); |
34ba94ba | 1780 | |
1e84371f | 1781 | if (f2fs_is_atomic_file(inode)) { |
34ba94ba JK |
1782 | register_inmem_page(inode, page); |
1783 | return 1; | |
1784 | } | |
1785 | ||
a18ff063 JK |
1786 | mark_inode_dirty(inode); |
1787 | ||
eb47b800 JK |
1788 | if (!PageDirty(page)) { |
1789 | __set_page_dirty_nobuffers(page); | |
a7ffdbe2 | 1790 | update_dirty_page(inode, page); |
eb47b800 JK |
1791 | return 1; |
1792 | } | |
1793 | return 0; | |
1794 | } | |
1795 | ||
c01e54b7 JK |
1796 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
1797 | { | |
454ae7e5 CY |
1798 | struct inode *inode = mapping->host; |
1799 | ||
b3d208f9 JK |
1800 | /* we don't need to use inline_data strictly */ |
1801 | if (f2fs_has_inline_data(inode)) { | |
1802 | int err = f2fs_convert_inline_inode(inode); | |
1803 | if (err) | |
1804 | return err; | |
1805 | } | |
bfad7c2d | 1806 | return generic_block_bmap(mapping, block, get_data_block); |
c01e54b7 JK |
1807 | } |
1808 | ||
429511cd CY |
1809 | void init_extent_cache_info(struct f2fs_sb_info *sbi) |
1810 | { | |
1811 | INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO); | |
1812 | init_rwsem(&sbi->extent_tree_lock); | |
1813 | INIT_LIST_HEAD(&sbi->extent_list); | |
1814 | spin_lock_init(&sbi->extent_lock); | |
1815 | sbi->total_ext_tree = 0; | |
1816 | atomic_set(&sbi->total_ext_node, 0); | |
1817 | } | |
1818 | ||
1819 | int __init create_extent_cache(void) | |
1820 | { | |
1821 | extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree", | |
1822 | sizeof(struct extent_tree)); | |
1823 | if (!extent_tree_slab) | |
1824 | return -ENOMEM; | |
1825 | extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node", | |
1826 | sizeof(struct extent_node)); | |
1827 | if (!extent_node_slab) { | |
1828 | kmem_cache_destroy(extent_tree_slab); | |
1829 | return -ENOMEM; | |
1830 | } | |
1831 | return 0; | |
1832 | } | |
1833 | ||
1834 | void destroy_extent_cache(void) | |
1835 | { | |
1836 | kmem_cache_destroy(extent_node_slab); | |
1837 | kmem_cache_destroy(extent_tree_slab); | |
1838 | } | |
1839 | ||
eb47b800 JK |
1840 | const struct address_space_operations f2fs_dblock_aops = { |
1841 | .readpage = f2fs_read_data_page, | |
1842 | .readpages = f2fs_read_data_pages, | |
1843 | .writepage = f2fs_write_data_page, | |
1844 | .writepages = f2fs_write_data_pages, | |
1845 | .write_begin = f2fs_write_begin, | |
a1dd3c13 | 1846 | .write_end = f2fs_write_end, |
eb47b800 | 1847 | .set_page_dirty = f2fs_set_data_page_dirty, |
487261f3 CY |
1848 | .invalidatepage = f2fs_invalidate_page, |
1849 | .releasepage = f2fs_release_page, | |
eb47b800 | 1850 | .direct_IO = f2fs_direct_IO, |
c01e54b7 | 1851 | .bmap = f2fs_bmap, |
eb47b800 | 1852 | }; |