]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/f2fs/data.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: revert segment allocation for direct IO
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / data.c
1 /*
2 * fs/f2fs/data.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/mm.h>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
25
26 #include "f2fs.h"
27 #include "node.h"
28 #include "segment.h"
29 #include "trace.h"
30 #include <trace/events/f2fs.h>
31
32 static void f2fs_read_end_io(struct bio *bio)
33 {
34 struct bio_vec *bvec;
35 int i;
36
37 #ifdef CONFIG_F2FS_FAULT_INJECTION
38 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO))
39 bio->bi_error = -EIO;
40 #endif
41
42 if (f2fs_bio_encrypted(bio)) {
43 if (bio->bi_error) {
44 fscrypt_release_ctx(bio->bi_private);
45 } else {
46 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
47 return;
48 }
49 }
50
51 bio_for_each_segment_all(bvec, bio, i) {
52 struct page *page = bvec->bv_page;
53
54 if (!bio->bi_error) {
55 if (!PageUptodate(page))
56 SetPageUptodate(page);
57 } else {
58 ClearPageUptodate(page);
59 SetPageError(page);
60 }
61 unlock_page(page);
62 }
63 bio_put(bio);
64 }
65
66 static void f2fs_write_end_io(struct bio *bio)
67 {
68 struct f2fs_sb_info *sbi = bio->bi_private;
69 struct bio_vec *bvec;
70 int i;
71
72 bio_for_each_segment_all(bvec, bio, i) {
73 struct page *page = bvec->bv_page;
74
75 fscrypt_pullback_bio_page(&page, true);
76
77 if (unlikely(bio->bi_error)) {
78 mapping_set_error(page->mapping, -EIO);
79 f2fs_stop_checkpoint(sbi, true);
80 }
81 end_page_writeback(page);
82 }
83 if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
84 wq_has_sleeper(&sbi->cp_wait))
85 wake_up(&sbi->cp_wait);
86
87 bio_put(bio);
88 }
89
90 /*
91 * Low-level block read/write IO operations.
92 */
93 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
94 int npages, bool is_read)
95 {
96 struct bio *bio;
97
98 bio = f2fs_bio_alloc(npages);
99
100 bio->bi_bdev = sbi->sb->s_bdev;
101 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
102 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
103 bio->bi_private = is_read ? NULL : sbi;
104
105 return bio;
106 }
107
108 static inline void __submit_bio(struct f2fs_sb_info *sbi,
109 struct bio *bio, enum page_type type)
110 {
111 if (!is_read_io(bio_op(bio))) {
112 atomic_inc(&sbi->nr_wb_bios);
113 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
114 current->plug && (type == DATA || type == NODE))
115 blk_finish_plug(current->plug);
116 }
117 submit_bio(bio);
118 }
119
120 static void __submit_merged_bio(struct f2fs_bio_info *io)
121 {
122 struct f2fs_io_info *fio = &io->fio;
123
124 if (!io->bio)
125 return;
126
127 if (is_read_io(fio->op))
128 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
129 else
130 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
131
132 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
133
134 __submit_bio(io->sbi, io->bio, fio->type);
135 io->bio = NULL;
136 }
137
138 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
139 struct page *page, nid_t ino)
140 {
141 struct bio_vec *bvec;
142 struct page *target;
143 int i;
144
145 if (!io->bio)
146 return false;
147
148 if (!inode && !page && !ino)
149 return true;
150
151 bio_for_each_segment_all(bvec, io->bio, i) {
152
153 if (bvec->bv_page->mapping)
154 target = bvec->bv_page;
155 else
156 target = fscrypt_control_page(bvec->bv_page);
157
158 if (inode && inode == target->mapping->host)
159 return true;
160 if (page && page == target)
161 return true;
162 if (ino && ino == ino_of_node(target))
163 return true;
164 }
165
166 return false;
167 }
168
169 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
170 struct page *page, nid_t ino,
171 enum page_type type)
172 {
173 enum page_type btype = PAGE_TYPE_OF_BIO(type);
174 struct f2fs_bio_info *io = &sbi->write_io[btype];
175 bool ret;
176
177 down_read(&io->io_rwsem);
178 ret = __has_merged_page(io, inode, page, ino);
179 up_read(&io->io_rwsem);
180 return ret;
181 }
182
183 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
184 struct inode *inode, struct page *page,
185 nid_t ino, enum page_type type, int rw)
186 {
187 enum page_type btype = PAGE_TYPE_OF_BIO(type);
188 struct f2fs_bio_info *io;
189
190 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
191
192 down_write(&io->io_rwsem);
193
194 if (!__has_merged_page(io, inode, page, ino))
195 goto out;
196
197 /* change META to META_FLUSH in the checkpoint procedure */
198 if (type >= META_FLUSH) {
199 io->fio.type = META_FLUSH;
200 io->fio.op = REQ_OP_WRITE;
201 if (test_opt(sbi, NOBARRIER))
202 io->fio.op_flags = WRITE_FLUSH | REQ_META | REQ_PRIO;
203 else
204 io->fio.op_flags = WRITE_FLUSH_FUA | REQ_META |
205 REQ_PRIO;
206 }
207 __submit_merged_bio(io);
208 out:
209 up_write(&io->io_rwsem);
210 }
211
212 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
213 int rw)
214 {
215 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
216 }
217
218 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
219 struct inode *inode, struct page *page,
220 nid_t ino, enum page_type type, int rw)
221 {
222 if (has_merged_page(sbi, inode, page, ino, type))
223 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
224 }
225
226 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
227 {
228 f2fs_submit_merged_bio(sbi, DATA, WRITE);
229 f2fs_submit_merged_bio(sbi, NODE, WRITE);
230 f2fs_submit_merged_bio(sbi, META, WRITE);
231 }
232
233 /*
234 * Fill the locked page with data located in the block address.
235 * Return unlocked page.
236 */
237 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
238 {
239 struct bio *bio;
240 struct page *page = fio->encrypted_page ?
241 fio->encrypted_page : fio->page;
242
243 trace_f2fs_submit_page_bio(page, fio);
244 f2fs_trace_ios(fio, 0);
245
246 /* Allocate a new bio */
247 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
248
249 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
250 bio_put(bio);
251 return -EFAULT;
252 }
253 bio_set_op_attrs(bio, fio->op, fio->op_flags);
254
255 __submit_bio(fio->sbi, bio, fio->type);
256 return 0;
257 }
258
259 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
260 {
261 struct f2fs_sb_info *sbi = fio->sbi;
262 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
263 struct f2fs_bio_info *io;
264 bool is_read = is_read_io(fio->op);
265 struct page *bio_page;
266
267 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
268
269 if (fio->old_blkaddr != NEW_ADDR)
270 verify_block_addr(sbi, fio->old_blkaddr);
271 verify_block_addr(sbi, fio->new_blkaddr);
272
273 down_write(&io->io_rwsem);
274
275 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
276 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
277 __submit_merged_bio(io);
278 alloc_new:
279 if (io->bio == NULL) {
280 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
281 BIO_MAX_PAGES, is_read);
282 io->fio = *fio;
283 }
284
285 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
286
287 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
288 PAGE_SIZE) {
289 __submit_merged_bio(io);
290 goto alloc_new;
291 }
292
293 io->last_block_in_bio = fio->new_blkaddr;
294 f2fs_trace_ios(fio, 0);
295
296 up_write(&io->io_rwsem);
297 trace_f2fs_submit_page_mbio(fio->page, fio);
298 }
299
300 static void __set_data_blkaddr(struct dnode_of_data *dn)
301 {
302 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
303 __le32 *addr_array;
304
305 /* Get physical address of data block */
306 addr_array = blkaddr_in_node(rn);
307 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
308 }
309
310 /*
311 * Lock ordering for the change of data block address:
312 * ->data_page
313 * ->node_page
314 * update block addresses in the node page
315 */
316 void set_data_blkaddr(struct dnode_of_data *dn)
317 {
318 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
319 __set_data_blkaddr(dn);
320 if (set_page_dirty(dn->node_page))
321 dn->node_changed = true;
322 }
323
324 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
325 {
326 dn->data_blkaddr = blkaddr;
327 set_data_blkaddr(dn);
328 f2fs_update_extent_cache(dn);
329 }
330
331 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
332 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
333 {
334 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
335
336 if (!count)
337 return 0;
338
339 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
340 return -EPERM;
341 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
342 return -ENOSPC;
343
344 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
345 dn->ofs_in_node, count);
346
347 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
348
349 for (; count > 0; dn->ofs_in_node++) {
350 block_t blkaddr =
351 datablock_addr(dn->node_page, dn->ofs_in_node);
352 if (blkaddr == NULL_ADDR) {
353 dn->data_blkaddr = NEW_ADDR;
354 __set_data_blkaddr(dn);
355 count--;
356 }
357 }
358
359 if (set_page_dirty(dn->node_page))
360 dn->node_changed = true;
361 return 0;
362 }
363
364 /* Should keep dn->ofs_in_node unchanged */
365 int reserve_new_block(struct dnode_of_data *dn)
366 {
367 unsigned int ofs_in_node = dn->ofs_in_node;
368 int ret;
369
370 ret = reserve_new_blocks(dn, 1);
371 dn->ofs_in_node = ofs_in_node;
372 return ret;
373 }
374
375 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
376 {
377 bool need_put = dn->inode_page ? false : true;
378 int err;
379
380 err = get_dnode_of_data(dn, index, ALLOC_NODE);
381 if (err)
382 return err;
383
384 if (dn->data_blkaddr == NULL_ADDR)
385 err = reserve_new_block(dn);
386 if (err || need_put)
387 f2fs_put_dnode(dn);
388 return err;
389 }
390
391 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
392 {
393 struct extent_info ei;
394 struct inode *inode = dn->inode;
395
396 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
397 dn->data_blkaddr = ei.blk + index - ei.fofs;
398 return 0;
399 }
400
401 return f2fs_reserve_block(dn, index);
402 }
403
404 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
405 int op_flags, bool for_write)
406 {
407 struct address_space *mapping = inode->i_mapping;
408 struct dnode_of_data dn;
409 struct page *page;
410 struct extent_info ei;
411 int err;
412 struct f2fs_io_info fio = {
413 .sbi = F2FS_I_SB(inode),
414 .type = DATA,
415 .op = REQ_OP_READ,
416 .op_flags = op_flags,
417 .encrypted_page = NULL,
418 };
419
420 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
421 return read_mapping_page(mapping, index, NULL);
422
423 page = f2fs_grab_cache_page(mapping, index, for_write);
424 if (!page)
425 return ERR_PTR(-ENOMEM);
426
427 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
428 dn.data_blkaddr = ei.blk + index - ei.fofs;
429 goto got_it;
430 }
431
432 set_new_dnode(&dn, inode, NULL, NULL, 0);
433 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
434 if (err)
435 goto put_err;
436 f2fs_put_dnode(&dn);
437
438 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
439 err = -ENOENT;
440 goto put_err;
441 }
442 got_it:
443 if (PageUptodate(page)) {
444 unlock_page(page);
445 return page;
446 }
447
448 /*
449 * A new dentry page is allocated but not able to be written, since its
450 * new inode page couldn't be allocated due to -ENOSPC.
451 * In such the case, its blkaddr can be remained as NEW_ADDR.
452 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
453 */
454 if (dn.data_blkaddr == NEW_ADDR) {
455 zero_user_segment(page, 0, PAGE_SIZE);
456 if (!PageUptodate(page))
457 SetPageUptodate(page);
458 unlock_page(page);
459 return page;
460 }
461
462 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
463 fio.page = page;
464 err = f2fs_submit_page_bio(&fio);
465 if (err)
466 goto put_err;
467 return page;
468
469 put_err:
470 f2fs_put_page(page, 1);
471 return ERR_PTR(err);
472 }
473
474 struct page *find_data_page(struct inode *inode, pgoff_t index)
475 {
476 struct address_space *mapping = inode->i_mapping;
477 struct page *page;
478
479 page = find_get_page(mapping, index);
480 if (page && PageUptodate(page))
481 return page;
482 f2fs_put_page(page, 0);
483
484 page = get_read_data_page(inode, index, READ_SYNC, false);
485 if (IS_ERR(page))
486 return page;
487
488 if (PageUptodate(page))
489 return page;
490
491 wait_on_page_locked(page);
492 if (unlikely(!PageUptodate(page))) {
493 f2fs_put_page(page, 0);
494 return ERR_PTR(-EIO);
495 }
496 return page;
497 }
498
499 /*
500 * If it tries to access a hole, return an error.
501 * Because, the callers, functions in dir.c and GC, should be able to know
502 * whether this page exists or not.
503 */
504 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
505 bool for_write)
506 {
507 struct address_space *mapping = inode->i_mapping;
508 struct page *page;
509 repeat:
510 page = get_read_data_page(inode, index, READ_SYNC, for_write);
511 if (IS_ERR(page))
512 return page;
513
514 /* wait for read completion */
515 lock_page(page);
516 if (unlikely(page->mapping != mapping)) {
517 f2fs_put_page(page, 1);
518 goto repeat;
519 }
520 if (unlikely(!PageUptodate(page))) {
521 f2fs_put_page(page, 1);
522 return ERR_PTR(-EIO);
523 }
524 return page;
525 }
526
527 /*
528 * Caller ensures that this data page is never allocated.
529 * A new zero-filled data page is allocated in the page cache.
530 *
531 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
532 * f2fs_unlock_op().
533 * Note that, ipage is set only by make_empty_dir, and if any error occur,
534 * ipage should be released by this function.
535 */
536 struct page *get_new_data_page(struct inode *inode,
537 struct page *ipage, pgoff_t index, bool new_i_size)
538 {
539 struct address_space *mapping = inode->i_mapping;
540 struct page *page;
541 struct dnode_of_data dn;
542 int err;
543
544 page = f2fs_grab_cache_page(mapping, index, true);
545 if (!page) {
546 /*
547 * before exiting, we should make sure ipage will be released
548 * if any error occur.
549 */
550 f2fs_put_page(ipage, 1);
551 return ERR_PTR(-ENOMEM);
552 }
553
554 set_new_dnode(&dn, inode, ipage, NULL, 0);
555 err = f2fs_reserve_block(&dn, index);
556 if (err) {
557 f2fs_put_page(page, 1);
558 return ERR_PTR(err);
559 }
560 if (!ipage)
561 f2fs_put_dnode(&dn);
562
563 if (PageUptodate(page))
564 goto got_it;
565
566 if (dn.data_blkaddr == NEW_ADDR) {
567 zero_user_segment(page, 0, PAGE_SIZE);
568 if (!PageUptodate(page))
569 SetPageUptodate(page);
570 } else {
571 f2fs_put_page(page, 1);
572
573 /* if ipage exists, blkaddr should be NEW_ADDR */
574 f2fs_bug_on(F2FS_I_SB(inode), ipage);
575 page = get_lock_data_page(inode, index, true);
576 if (IS_ERR(page))
577 return page;
578 }
579 got_it:
580 if (new_i_size && i_size_read(inode) <
581 ((loff_t)(index + 1) << PAGE_SHIFT))
582 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
583 return page;
584 }
585
586 static int __allocate_data_block(struct dnode_of_data *dn)
587 {
588 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
589 struct f2fs_summary sum;
590 struct node_info ni;
591 pgoff_t fofs;
592 blkcnt_t count = 1;
593
594 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
595 return -EPERM;
596
597 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
598 if (dn->data_blkaddr == NEW_ADDR)
599 goto alloc;
600
601 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
602 return -ENOSPC;
603
604 alloc:
605 get_node_info(sbi, dn->nid, &ni);
606 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
607
608 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
609 &sum, CURSEG_WARM_DATA);
610 set_data_blkaddr(dn);
611
612 /* update i_size */
613 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
614 dn->ofs_in_node;
615 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
616 f2fs_i_size_write(dn->inode,
617 ((loff_t)(fofs + 1) << PAGE_SHIFT));
618 return 0;
619 }
620
621 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
622 {
623 struct inode *inode = file_inode(iocb->ki_filp);
624 struct f2fs_map_blocks map;
625 ssize_t ret = 0;
626
627 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
628 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
629 if (map.m_len > map.m_lblk)
630 map.m_len -= map.m_lblk;
631 else
632 map.m_len = 0;
633
634 map.m_next_pgofs = NULL;
635
636 if (iocb->ki_flags & IOCB_DIRECT) {
637 ret = f2fs_convert_inline_inode(inode);
638 if (ret)
639 return ret;
640 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
641 }
642 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
643 ret = f2fs_convert_inline_inode(inode);
644 if (ret)
645 return ret;
646 }
647 if (!f2fs_has_inline_data(inode))
648 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
649 return ret;
650 }
651
652 /*
653 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
654 * f2fs_map_blocks structure.
655 * If original data blocks are allocated, then give them to blockdev.
656 * Otherwise,
657 * a. preallocate requested block addresses
658 * b. do not use extent cache for better performance
659 * c. give the block addresses to blockdev
660 */
661 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
662 int create, int flag)
663 {
664 unsigned int maxblocks = map->m_len;
665 struct dnode_of_data dn;
666 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
667 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
668 pgoff_t pgofs, end_offset, end;
669 int err = 0, ofs = 1;
670 unsigned int ofs_in_node, last_ofs_in_node;
671 blkcnt_t prealloc;
672 struct extent_info ei;
673 block_t blkaddr;
674
675 if (!maxblocks)
676 return 0;
677
678 map->m_len = 0;
679 map->m_flags = 0;
680
681 /* it only supports block size == page size */
682 pgofs = (pgoff_t)map->m_lblk;
683 end = pgofs + maxblocks;
684
685 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
686 map->m_pblk = ei.blk + pgofs - ei.fofs;
687 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
688 map->m_flags = F2FS_MAP_MAPPED;
689 goto out;
690 }
691
692 next_dnode:
693 if (create)
694 f2fs_lock_op(sbi);
695
696 /* When reading holes, we need its node page */
697 set_new_dnode(&dn, inode, NULL, NULL, 0);
698 err = get_dnode_of_data(&dn, pgofs, mode);
699 if (err) {
700 if (flag == F2FS_GET_BLOCK_BMAP)
701 map->m_pblk = 0;
702 if (err == -ENOENT) {
703 err = 0;
704 if (map->m_next_pgofs)
705 *map->m_next_pgofs =
706 get_next_page_offset(&dn, pgofs);
707 }
708 goto unlock_out;
709 }
710
711 prealloc = 0;
712 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
713 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
714
715 next_block:
716 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
717
718 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
719 if (create) {
720 if (unlikely(f2fs_cp_error(sbi))) {
721 err = -EIO;
722 goto sync_out;
723 }
724 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
725 if (blkaddr == NULL_ADDR) {
726 prealloc++;
727 last_ofs_in_node = dn.ofs_in_node;
728 }
729 } else {
730 err = __allocate_data_block(&dn);
731 if (!err)
732 set_inode_flag(inode, FI_APPEND_WRITE);
733 }
734 if (err)
735 goto sync_out;
736 map->m_flags = F2FS_MAP_NEW;
737 blkaddr = dn.data_blkaddr;
738 } else {
739 if (flag == F2FS_GET_BLOCK_BMAP) {
740 map->m_pblk = 0;
741 goto sync_out;
742 }
743 if (flag == F2FS_GET_BLOCK_FIEMAP &&
744 blkaddr == NULL_ADDR) {
745 if (map->m_next_pgofs)
746 *map->m_next_pgofs = pgofs + 1;
747 }
748 if (flag != F2FS_GET_BLOCK_FIEMAP ||
749 blkaddr != NEW_ADDR)
750 goto sync_out;
751 }
752 }
753
754 if (flag == F2FS_GET_BLOCK_PRE_AIO)
755 goto skip;
756
757 if (map->m_len == 0) {
758 /* preallocated unwritten block should be mapped for fiemap. */
759 if (blkaddr == NEW_ADDR)
760 map->m_flags |= F2FS_MAP_UNWRITTEN;
761 map->m_flags |= F2FS_MAP_MAPPED;
762
763 map->m_pblk = blkaddr;
764 map->m_len = 1;
765 } else if ((map->m_pblk != NEW_ADDR &&
766 blkaddr == (map->m_pblk + ofs)) ||
767 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
768 flag == F2FS_GET_BLOCK_PRE_DIO) {
769 ofs++;
770 map->m_len++;
771 } else {
772 goto sync_out;
773 }
774
775 skip:
776 dn.ofs_in_node++;
777 pgofs++;
778
779 /* preallocate blocks in batch for one dnode page */
780 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
781 (pgofs == end || dn.ofs_in_node == end_offset)) {
782
783 dn.ofs_in_node = ofs_in_node;
784 err = reserve_new_blocks(&dn, prealloc);
785 if (err)
786 goto sync_out;
787
788 map->m_len += dn.ofs_in_node - ofs_in_node;
789 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
790 err = -ENOSPC;
791 goto sync_out;
792 }
793 dn.ofs_in_node = end_offset;
794 }
795
796 if (pgofs >= end)
797 goto sync_out;
798 else if (dn.ofs_in_node < end_offset)
799 goto next_block;
800
801 f2fs_put_dnode(&dn);
802
803 if (create) {
804 f2fs_unlock_op(sbi);
805 f2fs_balance_fs(sbi, dn.node_changed);
806 }
807 goto next_dnode;
808
809 sync_out:
810 f2fs_put_dnode(&dn);
811 unlock_out:
812 if (create) {
813 f2fs_unlock_op(sbi);
814 f2fs_balance_fs(sbi, dn.node_changed);
815 }
816 out:
817 trace_f2fs_map_blocks(inode, map, err);
818 return err;
819 }
820
821 static int __get_data_block(struct inode *inode, sector_t iblock,
822 struct buffer_head *bh, int create, int flag,
823 pgoff_t *next_pgofs)
824 {
825 struct f2fs_map_blocks map;
826 int ret;
827
828 map.m_lblk = iblock;
829 map.m_len = bh->b_size >> inode->i_blkbits;
830 map.m_next_pgofs = next_pgofs;
831
832 ret = f2fs_map_blocks(inode, &map, create, flag);
833 if (!ret) {
834 map_bh(bh, inode->i_sb, map.m_pblk);
835 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
836 bh->b_size = map.m_len << inode->i_blkbits;
837 }
838 return ret;
839 }
840
841 static int get_data_block(struct inode *inode, sector_t iblock,
842 struct buffer_head *bh_result, int create, int flag,
843 pgoff_t *next_pgofs)
844 {
845 return __get_data_block(inode, iblock, bh_result, create,
846 flag, next_pgofs);
847 }
848
849 static int get_data_block_dio(struct inode *inode, sector_t iblock,
850 struct buffer_head *bh_result, int create)
851 {
852 return __get_data_block(inode, iblock, bh_result, create,
853 F2FS_GET_BLOCK_DIO, NULL);
854 }
855
856 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
857 struct buffer_head *bh_result, int create)
858 {
859 /* Block number less than F2FS MAX BLOCKS */
860 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
861 return -EFBIG;
862
863 return __get_data_block(inode, iblock, bh_result, create,
864 F2FS_GET_BLOCK_BMAP, NULL);
865 }
866
867 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
868 {
869 return (offset >> inode->i_blkbits);
870 }
871
872 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
873 {
874 return (blk << inode->i_blkbits);
875 }
876
877 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
878 u64 start, u64 len)
879 {
880 struct buffer_head map_bh;
881 sector_t start_blk, last_blk;
882 pgoff_t next_pgofs;
883 u64 logical = 0, phys = 0, size = 0;
884 u32 flags = 0;
885 int ret = 0;
886
887 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
888 if (ret)
889 return ret;
890
891 if (f2fs_has_inline_data(inode)) {
892 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
893 if (ret != -EAGAIN)
894 return ret;
895 }
896
897 inode_lock(inode);
898
899 if (logical_to_blk(inode, len) == 0)
900 len = blk_to_logical(inode, 1);
901
902 start_blk = logical_to_blk(inode, start);
903 last_blk = logical_to_blk(inode, start + len - 1);
904
905 next:
906 memset(&map_bh, 0, sizeof(struct buffer_head));
907 map_bh.b_size = len;
908
909 ret = get_data_block(inode, start_blk, &map_bh, 0,
910 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
911 if (ret)
912 goto out;
913
914 /* HOLE */
915 if (!buffer_mapped(&map_bh)) {
916 start_blk = next_pgofs;
917
918 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
919 F2FS_I_SB(inode)->max_file_blocks))
920 goto prep_next;
921
922 flags |= FIEMAP_EXTENT_LAST;
923 }
924
925 if (size) {
926 if (f2fs_encrypted_inode(inode))
927 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
928
929 ret = fiemap_fill_next_extent(fieinfo, logical,
930 phys, size, flags);
931 }
932
933 if (start_blk > last_blk || ret)
934 goto out;
935
936 logical = blk_to_logical(inode, start_blk);
937 phys = blk_to_logical(inode, map_bh.b_blocknr);
938 size = map_bh.b_size;
939 flags = 0;
940 if (buffer_unwritten(&map_bh))
941 flags = FIEMAP_EXTENT_UNWRITTEN;
942
943 start_blk += logical_to_blk(inode, size);
944
945 prep_next:
946 cond_resched();
947 if (fatal_signal_pending(current))
948 ret = -EINTR;
949 else
950 goto next;
951 out:
952 if (ret == 1)
953 ret = 0;
954
955 inode_unlock(inode);
956 return ret;
957 }
958
959 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
960 unsigned nr_pages)
961 {
962 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
963 struct fscrypt_ctx *ctx = NULL;
964 struct block_device *bdev = sbi->sb->s_bdev;
965 struct bio *bio;
966
967 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
968 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
969 if (IS_ERR(ctx))
970 return ERR_CAST(ctx);
971
972 /* wait the page to be moved by cleaning */
973 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
974 }
975
976 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
977 if (!bio) {
978 if (ctx)
979 fscrypt_release_ctx(ctx);
980 return ERR_PTR(-ENOMEM);
981 }
982 bio->bi_bdev = bdev;
983 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
984 bio->bi_end_io = f2fs_read_end_io;
985 bio->bi_private = ctx;
986
987 return bio;
988 }
989
990 /*
991 * This function was originally taken from fs/mpage.c, and customized for f2fs.
992 * Major change was from block_size == page_size in f2fs by default.
993 */
994 static int f2fs_mpage_readpages(struct address_space *mapping,
995 struct list_head *pages, struct page *page,
996 unsigned nr_pages)
997 {
998 struct bio *bio = NULL;
999 unsigned page_idx;
1000 sector_t last_block_in_bio = 0;
1001 struct inode *inode = mapping->host;
1002 const unsigned blkbits = inode->i_blkbits;
1003 const unsigned blocksize = 1 << blkbits;
1004 sector_t block_in_file;
1005 sector_t last_block;
1006 sector_t last_block_in_file;
1007 sector_t block_nr;
1008 struct f2fs_map_blocks map;
1009
1010 map.m_pblk = 0;
1011 map.m_lblk = 0;
1012 map.m_len = 0;
1013 map.m_flags = 0;
1014 map.m_next_pgofs = NULL;
1015
1016 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1017
1018 prefetchw(&page->flags);
1019 if (pages) {
1020 page = list_entry(pages->prev, struct page, lru);
1021 list_del(&page->lru);
1022 if (add_to_page_cache_lru(page, mapping,
1023 page->index,
1024 readahead_gfp_mask(mapping)))
1025 goto next_page;
1026 }
1027
1028 block_in_file = (sector_t)page->index;
1029 last_block = block_in_file + nr_pages;
1030 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1031 blkbits;
1032 if (last_block > last_block_in_file)
1033 last_block = last_block_in_file;
1034
1035 /*
1036 * Map blocks using the previous result first.
1037 */
1038 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1039 block_in_file > map.m_lblk &&
1040 block_in_file < (map.m_lblk + map.m_len))
1041 goto got_it;
1042
1043 /*
1044 * Then do more f2fs_map_blocks() calls until we are
1045 * done with this page.
1046 */
1047 map.m_flags = 0;
1048
1049 if (block_in_file < last_block) {
1050 map.m_lblk = block_in_file;
1051 map.m_len = last_block - block_in_file;
1052
1053 if (f2fs_map_blocks(inode, &map, 0,
1054 F2FS_GET_BLOCK_READ))
1055 goto set_error_page;
1056 }
1057 got_it:
1058 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1059 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1060 SetPageMappedToDisk(page);
1061
1062 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1063 SetPageUptodate(page);
1064 goto confused;
1065 }
1066 } else {
1067 zero_user_segment(page, 0, PAGE_SIZE);
1068 if (!PageUptodate(page))
1069 SetPageUptodate(page);
1070 unlock_page(page);
1071 goto next_page;
1072 }
1073
1074 /*
1075 * This page will go to BIO. Do we need to send this
1076 * BIO off first?
1077 */
1078 if (bio && (last_block_in_bio != block_nr - 1)) {
1079 submit_and_realloc:
1080 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1081 bio = NULL;
1082 }
1083 if (bio == NULL) {
1084 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1085 if (IS_ERR(bio)) {
1086 bio = NULL;
1087 goto set_error_page;
1088 }
1089 bio_set_op_attrs(bio, REQ_OP_READ, 0);
1090 }
1091
1092 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1093 goto submit_and_realloc;
1094
1095 last_block_in_bio = block_nr;
1096 goto next_page;
1097 set_error_page:
1098 SetPageError(page);
1099 zero_user_segment(page, 0, PAGE_SIZE);
1100 unlock_page(page);
1101 goto next_page;
1102 confused:
1103 if (bio) {
1104 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1105 bio = NULL;
1106 }
1107 unlock_page(page);
1108 next_page:
1109 if (pages)
1110 put_page(page);
1111 }
1112 BUG_ON(pages && !list_empty(pages));
1113 if (bio)
1114 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1115 return 0;
1116 }
1117
1118 static int f2fs_read_data_page(struct file *file, struct page *page)
1119 {
1120 struct inode *inode = page->mapping->host;
1121 int ret = -EAGAIN;
1122
1123 trace_f2fs_readpage(page, DATA);
1124
1125 /* If the file has inline data, try to read it directly */
1126 if (f2fs_has_inline_data(inode))
1127 ret = f2fs_read_inline_data(inode, page);
1128 if (ret == -EAGAIN)
1129 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1130 return ret;
1131 }
1132
1133 static int f2fs_read_data_pages(struct file *file,
1134 struct address_space *mapping,
1135 struct list_head *pages, unsigned nr_pages)
1136 {
1137 struct inode *inode = file->f_mapping->host;
1138 struct page *page = list_entry(pages->prev, struct page, lru);
1139
1140 trace_f2fs_readpages(inode, page, nr_pages);
1141
1142 /* If the file has inline data, skip readpages */
1143 if (f2fs_has_inline_data(inode))
1144 return 0;
1145
1146 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1147 }
1148
1149 int do_write_data_page(struct f2fs_io_info *fio)
1150 {
1151 struct page *page = fio->page;
1152 struct inode *inode = page->mapping->host;
1153 struct dnode_of_data dn;
1154 int err = 0;
1155
1156 set_new_dnode(&dn, inode, NULL, NULL, 0);
1157 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1158 if (err)
1159 return err;
1160
1161 fio->old_blkaddr = dn.data_blkaddr;
1162
1163 /* This page is already truncated */
1164 if (fio->old_blkaddr == NULL_ADDR) {
1165 ClearPageUptodate(page);
1166 goto out_writepage;
1167 }
1168
1169 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1170 gfp_t gfp_flags = GFP_NOFS;
1171
1172 /* wait for GCed encrypted page writeback */
1173 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1174 fio->old_blkaddr);
1175 retry_encrypt:
1176 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1177 gfp_flags);
1178 if (IS_ERR(fio->encrypted_page)) {
1179 err = PTR_ERR(fio->encrypted_page);
1180 if (err == -ENOMEM) {
1181 /* flush pending ios and wait for a while */
1182 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1183 congestion_wait(BLK_RW_ASYNC, HZ/50);
1184 gfp_flags |= __GFP_NOFAIL;
1185 err = 0;
1186 goto retry_encrypt;
1187 }
1188 goto out_writepage;
1189 }
1190 }
1191
1192 set_page_writeback(page);
1193
1194 /*
1195 * If current allocation needs SSR,
1196 * it had better in-place writes for updated data.
1197 */
1198 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1199 !is_cold_data(page) &&
1200 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1201 need_inplace_update(inode))) {
1202 rewrite_data_page(fio);
1203 set_inode_flag(inode, FI_UPDATE_WRITE);
1204 trace_f2fs_do_write_data_page(page, IPU);
1205 } else {
1206 write_data_page(&dn, fio);
1207 trace_f2fs_do_write_data_page(page, OPU);
1208 set_inode_flag(inode, FI_APPEND_WRITE);
1209 if (page->index == 0)
1210 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1211 }
1212 out_writepage:
1213 f2fs_put_dnode(&dn);
1214 return err;
1215 }
1216
1217 static int f2fs_write_data_page(struct page *page,
1218 struct writeback_control *wbc)
1219 {
1220 struct inode *inode = page->mapping->host;
1221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1222 loff_t i_size = i_size_read(inode);
1223 const pgoff_t end_index = ((unsigned long long) i_size)
1224 >> PAGE_SHIFT;
1225 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1226 unsigned offset = 0;
1227 bool need_balance_fs = false;
1228 int err = 0;
1229 struct f2fs_io_info fio = {
1230 .sbi = sbi,
1231 .type = DATA,
1232 .op = REQ_OP_WRITE,
1233 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1234 .page = page,
1235 .encrypted_page = NULL,
1236 };
1237
1238 trace_f2fs_writepage(page, DATA);
1239
1240 if (page->index < end_index)
1241 goto write;
1242
1243 /*
1244 * If the offset is out-of-range of file size,
1245 * this page does not have to be written to disk.
1246 */
1247 offset = i_size & (PAGE_SIZE - 1);
1248 if ((page->index >= end_index + 1) || !offset)
1249 goto out;
1250
1251 zero_user_segment(page, offset, PAGE_SIZE);
1252 write:
1253 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1254 goto redirty_out;
1255 if (f2fs_is_drop_cache(inode))
1256 goto out;
1257 /* we should not write 0'th page having journal header */
1258 if (f2fs_is_volatile_file(inode) && (!page->index ||
1259 (!wbc->for_reclaim &&
1260 available_free_memory(sbi, BASE_CHECK))))
1261 goto redirty_out;
1262
1263 /* we should bypass data pages to proceed the kworkder jobs */
1264 if (unlikely(f2fs_cp_error(sbi))) {
1265 mapping_set_error(page->mapping, -EIO);
1266 goto out;
1267 }
1268
1269 /* Dentry blocks are controlled by checkpoint */
1270 if (S_ISDIR(inode->i_mode)) {
1271 err = do_write_data_page(&fio);
1272 goto done;
1273 }
1274
1275 if (!wbc->for_reclaim)
1276 need_balance_fs = true;
1277 else if (has_not_enough_free_secs(sbi, 0, 0))
1278 goto redirty_out;
1279
1280 err = -EAGAIN;
1281 f2fs_lock_op(sbi);
1282 if (f2fs_has_inline_data(inode))
1283 err = f2fs_write_inline_data(inode, page);
1284 if (err == -EAGAIN)
1285 err = do_write_data_page(&fio);
1286 if (F2FS_I(inode)->last_disk_size < psize)
1287 F2FS_I(inode)->last_disk_size = psize;
1288 f2fs_unlock_op(sbi);
1289 done:
1290 if (err && err != -ENOENT)
1291 goto redirty_out;
1292
1293 clear_cold_data(page);
1294 out:
1295 inode_dec_dirty_pages(inode);
1296 if (err)
1297 ClearPageUptodate(page);
1298
1299 if (wbc->for_reclaim) {
1300 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1301 remove_dirty_inode(inode);
1302 }
1303
1304 unlock_page(page);
1305 f2fs_balance_fs(sbi, need_balance_fs);
1306
1307 if (unlikely(f2fs_cp_error(sbi)))
1308 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1309
1310 return 0;
1311
1312 redirty_out:
1313 redirty_page_for_writepage(wbc, page);
1314 unlock_page(page);
1315 return err;
1316 }
1317
1318 /*
1319 * This function was copied from write_cche_pages from mm/page-writeback.c.
1320 * The major change is making write step of cold data page separately from
1321 * warm/hot data page.
1322 */
1323 static int f2fs_write_cache_pages(struct address_space *mapping,
1324 struct writeback_control *wbc)
1325 {
1326 int ret = 0;
1327 int done = 0;
1328 struct pagevec pvec;
1329 int nr_pages;
1330 pgoff_t uninitialized_var(writeback_index);
1331 pgoff_t index;
1332 pgoff_t end; /* Inclusive */
1333 pgoff_t done_index;
1334 int cycled;
1335 int range_whole = 0;
1336 int tag;
1337 int nwritten = 0;
1338
1339 pagevec_init(&pvec, 0);
1340
1341 if (wbc->range_cyclic) {
1342 writeback_index = mapping->writeback_index; /* prev offset */
1343 index = writeback_index;
1344 if (index == 0)
1345 cycled = 1;
1346 else
1347 cycled = 0;
1348 end = -1;
1349 } else {
1350 index = wbc->range_start >> PAGE_SHIFT;
1351 end = wbc->range_end >> PAGE_SHIFT;
1352 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1353 range_whole = 1;
1354 cycled = 1; /* ignore range_cyclic tests */
1355 }
1356 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1357 tag = PAGECACHE_TAG_TOWRITE;
1358 else
1359 tag = PAGECACHE_TAG_DIRTY;
1360 retry:
1361 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1362 tag_pages_for_writeback(mapping, index, end);
1363 done_index = index;
1364 while (!done && (index <= end)) {
1365 int i;
1366
1367 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1368 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1369 if (nr_pages == 0)
1370 break;
1371
1372 for (i = 0; i < nr_pages; i++) {
1373 struct page *page = pvec.pages[i];
1374
1375 if (page->index > end) {
1376 done = 1;
1377 break;
1378 }
1379
1380 done_index = page->index;
1381
1382 lock_page(page);
1383
1384 if (unlikely(page->mapping != mapping)) {
1385 continue_unlock:
1386 unlock_page(page);
1387 continue;
1388 }
1389
1390 if (!PageDirty(page)) {
1391 /* someone wrote it for us */
1392 goto continue_unlock;
1393 }
1394
1395 if (PageWriteback(page)) {
1396 if (wbc->sync_mode != WB_SYNC_NONE)
1397 f2fs_wait_on_page_writeback(page,
1398 DATA, true);
1399 else
1400 goto continue_unlock;
1401 }
1402
1403 BUG_ON(PageWriteback(page));
1404 if (!clear_page_dirty_for_io(page))
1405 goto continue_unlock;
1406
1407 ret = mapping->a_ops->writepage(page, wbc);
1408 if (unlikely(ret)) {
1409 done_index = page->index + 1;
1410 done = 1;
1411 break;
1412 } else {
1413 nwritten++;
1414 }
1415
1416 if (--wbc->nr_to_write <= 0 &&
1417 wbc->sync_mode == WB_SYNC_NONE) {
1418 done = 1;
1419 break;
1420 }
1421 }
1422 pagevec_release(&pvec);
1423 cond_resched();
1424 }
1425
1426 if (!cycled && !done) {
1427 cycled = 1;
1428 index = 0;
1429 end = writeback_index - 1;
1430 goto retry;
1431 }
1432 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1433 mapping->writeback_index = done_index;
1434
1435 if (nwritten)
1436 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1437 NULL, 0, DATA, WRITE);
1438
1439 return ret;
1440 }
1441
1442 static int f2fs_write_data_pages(struct address_space *mapping,
1443 struct writeback_control *wbc)
1444 {
1445 struct inode *inode = mapping->host;
1446 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1447 struct blk_plug plug;
1448 int ret;
1449
1450 /* deal with chardevs and other special file */
1451 if (!mapping->a_ops->writepage)
1452 return 0;
1453
1454 /* skip writing if there is no dirty page in this inode */
1455 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1456 return 0;
1457
1458 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1459 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1460 available_free_memory(sbi, DIRTY_DENTS))
1461 goto skip_write;
1462
1463 /* skip writing during file defragment */
1464 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1465 goto skip_write;
1466
1467 /* during POR, we don't need to trigger writepage at all. */
1468 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1469 goto skip_write;
1470
1471 trace_f2fs_writepages(mapping->host, wbc, DATA);
1472
1473 blk_start_plug(&plug);
1474 ret = f2fs_write_cache_pages(mapping, wbc);
1475 blk_finish_plug(&plug);
1476 /*
1477 * if some pages were truncated, we cannot guarantee its mapping->host
1478 * to detect pending bios.
1479 */
1480
1481 remove_dirty_inode(inode);
1482 return ret;
1483
1484 skip_write:
1485 wbc->pages_skipped += get_dirty_pages(inode);
1486 trace_f2fs_writepages(mapping->host, wbc, DATA);
1487 return 0;
1488 }
1489
1490 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1491 {
1492 struct inode *inode = mapping->host;
1493 loff_t i_size = i_size_read(inode);
1494
1495 if (to > i_size) {
1496 truncate_pagecache(inode, i_size);
1497 truncate_blocks(inode, i_size, true);
1498 }
1499 }
1500
1501 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1502 struct page *page, loff_t pos, unsigned len,
1503 block_t *blk_addr, bool *node_changed)
1504 {
1505 struct inode *inode = page->mapping->host;
1506 pgoff_t index = page->index;
1507 struct dnode_of_data dn;
1508 struct page *ipage;
1509 bool locked = false;
1510 struct extent_info ei;
1511 int err = 0;
1512
1513 /*
1514 * we already allocated all the blocks, so we don't need to get
1515 * the block addresses when there is no need to fill the page.
1516 */
1517 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE)
1518 return 0;
1519
1520 if (f2fs_has_inline_data(inode) ||
1521 (pos & PAGE_MASK) >= i_size_read(inode)) {
1522 f2fs_lock_op(sbi);
1523 locked = true;
1524 }
1525 restart:
1526 /* check inline_data */
1527 ipage = get_node_page(sbi, inode->i_ino);
1528 if (IS_ERR(ipage)) {
1529 err = PTR_ERR(ipage);
1530 goto unlock_out;
1531 }
1532
1533 set_new_dnode(&dn, inode, ipage, ipage, 0);
1534
1535 if (f2fs_has_inline_data(inode)) {
1536 if (pos + len <= MAX_INLINE_DATA) {
1537 read_inline_data(page, ipage);
1538 set_inode_flag(inode, FI_DATA_EXIST);
1539 if (inode->i_nlink)
1540 set_inline_node(ipage);
1541 } else {
1542 err = f2fs_convert_inline_page(&dn, page);
1543 if (err)
1544 goto out;
1545 if (dn.data_blkaddr == NULL_ADDR)
1546 err = f2fs_get_block(&dn, index);
1547 }
1548 } else if (locked) {
1549 err = f2fs_get_block(&dn, index);
1550 } else {
1551 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1552 dn.data_blkaddr = ei.blk + index - ei.fofs;
1553 } else {
1554 /* hole case */
1555 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1556 if (err || dn.data_blkaddr == NULL_ADDR) {
1557 f2fs_put_dnode(&dn);
1558 f2fs_lock_op(sbi);
1559 locked = true;
1560 goto restart;
1561 }
1562 }
1563 }
1564
1565 /* convert_inline_page can make node_changed */
1566 *blk_addr = dn.data_blkaddr;
1567 *node_changed = dn.node_changed;
1568 out:
1569 f2fs_put_dnode(&dn);
1570 unlock_out:
1571 if (locked)
1572 f2fs_unlock_op(sbi);
1573 return err;
1574 }
1575
1576 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1577 loff_t pos, unsigned len, unsigned flags,
1578 struct page **pagep, void **fsdata)
1579 {
1580 struct inode *inode = mapping->host;
1581 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1582 struct page *page = NULL;
1583 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1584 bool need_balance = false;
1585 block_t blkaddr = NULL_ADDR;
1586 int err = 0;
1587
1588 trace_f2fs_write_begin(inode, pos, len, flags);
1589
1590 /*
1591 * We should check this at this moment to avoid deadlock on inode page
1592 * and #0 page. The locking rule for inline_data conversion should be:
1593 * lock_page(page #0) -> lock_page(inode_page)
1594 */
1595 if (index != 0) {
1596 err = f2fs_convert_inline_inode(inode);
1597 if (err)
1598 goto fail;
1599 }
1600 repeat:
1601 page = grab_cache_page_write_begin(mapping, index, flags);
1602 if (!page) {
1603 err = -ENOMEM;
1604 goto fail;
1605 }
1606
1607 *pagep = page;
1608
1609 err = prepare_write_begin(sbi, page, pos, len,
1610 &blkaddr, &need_balance);
1611 if (err)
1612 goto fail;
1613
1614 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1615 unlock_page(page);
1616 f2fs_balance_fs(sbi, true);
1617 lock_page(page);
1618 if (page->mapping != mapping) {
1619 /* The page got truncated from under us */
1620 f2fs_put_page(page, 1);
1621 goto repeat;
1622 }
1623 }
1624
1625 f2fs_wait_on_page_writeback(page, DATA, false);
1626
1627 /* wait for GCed encrypted page writeback */
1628 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1629 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1630
1631 if (len == PAGE_SIZE || PageUptodate(page))
1632 return 0;
1633
1634 if (blkaddr == NEW_ADDR) {
1635 zero_user_segment(page, 0, PAGE_SIZE);
1636 SetPageUptodate(page);
1637 } else {
1638 struct bio *bio;
1639
1640 bio = f2fs_grab_bio(inode, blkaddr, 1);
1641 if (IS_ERR(bio)) {
1642 err = PTR_ERR(bio);
1643 goto fail;
1644 }
1645 bio_set_op_attrs(bio, REQ_OP_READ, READ_SYNC);
1646 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1647 bio_put(bio);
1648 err = -EFAULT;
1649 goto fail;
1650 }
1651
1652 __submit_bio(sbi, bio, DATA);
1653
1654 lock_page(page);
1655 if (unlikely(page->mapping != mapping)) {
1656 f2fs_put_page(page, 1);
1657 goto repeat;
1658 }
1659 if (unlikely(!PageUptodate(page))) {
1660 err = -EIO;
1661 goto fail;
1662 }
1663 }
1664 return 0;
1665
1666 fail:
1667 f2fs_put_page(page, 1);
1668 f2fs_write_failed(mapping, pos + len);
1669 return err;
1670 }
1671
1672 static int f2fs_write_end(struct file *file,
1673 struct address_space *mapping,
1674 loff_t pos, unsigned len, unsigned copied,
1675 struct page *page, void *fsdata)
1676 {
1677 struct inode *inode = page->mapping->host;
1678
1679 trace_f2fs_write_end(inode, pos, len, copied);
1680
1681 /*
1682 * This should be come from len == PAGE_SIZE, and we expect copied
1683 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1684 * let generic_perform_write() try to copy data again through copied=0.
1685 */
1686 if (!PageUptodate(page)) {
1687 if (unlikely(copied != PAGE_SIZE))
1688 copied = 0;
1689 else
1690 SetPageUptodate(page);
1691 }
1692 if (!copied)
1693 goto unlock_out;
1694
1695 set_page_dirty(page);
1696 clear_cold_data(page);
1697
1698 if (pos + copied > i_size_read(inode))
1699 f2fs_i_size_write(inode, pos + copied);
1700 unlock_out:
1701 f2fs_put_page(page, 1);
1702 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1703 return copied;
1704 }
1705
1706 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1707 loff_t offset)
1708 {
1709 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1710
1711 if (offset & blocksize_mask)
1712 return -EINVAL;
1713
1714 if (iov_iter_alignment(iter) & blocksize_mask)
1715 return -EINVAL;
1716
1717 return 0;
1718 }
1719
1720 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1721 {
1722 struct address_space *mapping = iocb->ki_filp->f_mapping;
1723 struct inode *inode = mapping->host;
1724 size_t count = iov_iter_count(iter);
1725 loff_t offset = iocb->ki_pos;
1726 int rw = iov_iter_rw(iter);
1727 int err;
1728
1729 err = check_direct_IO(inode, iter, offset);
1730 if (err)
1731 return err;
1732
1733 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1734 return 0;
1735 if (test_opt(F2FS_I_SB(inode), LFS))
1736 return 0;
1737
1738 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1739
1740 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1741 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1742 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1743
1744 if (rw == WRITE) {
1745 if (err > 0)
1746 set_inode_flag(inode, FI_UPDATE_WRITE);
1747 else if (err < 0)
1748 f2fs_write_failed(mapping, offset + count);
1749 }
1750
1751 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1752
1753 return err;
1754 }
1755
1756 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1757 unsigned int length)
1758 {
1759 struct inode *inode = page->mapping->host;
1760 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1761
1762 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1763 (offset % PAGE_SIZE || length != PAGE_SIZE))
1764 return;
1765
1766 if (PageDirty(page)) {
1767 if (inode->i_ino == F2FS_META_INO(sbi)) {
1768 dec_page_count(sbi, F2FS_DIRTY_META);
1769 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
1770 dec_page_count(sbi, F2FS_DIRTY_NODES);
1771 } else {
1772 inode_dec_dirty_pages(inode);
1773 remove_dirty_inode(inode);
1774 }
1775 }
1776
1777 /* This is atomic written page, keep Private */
1778 if (IS_ATOMIC_WRITTEN_PAGE(page))
1779 return;
1780
1781 set_page_private(page, 0);
1782 ClearPagePrivate(page);
1783 }
1784
1785 int f2fs_release_page(struct page *page, gfp_t wait)
1786 {
1787 /* If this is dirty page, keep PagePrivate */
1788 if (PageDirty(page))
1789 return 0;
1790
1791 /* This is atomic written page, keep Private */
1792 if (IS_ATOMIC_WRITTEN_PAGE(page))
1793 return 0;
1794
1795 set_page_private(page, 0);
1796 ClearPagePrivate(page);
1797 return 1;
1798 }
1799
1800 /*
1801 * This was copied from __set_page_dirty_buffers which gives higher performance
1802 * in very high speed storages. (e.g., pmem)
1803 */
1804 void f2fs_set_page_dirty_nobuffers(struct page *page)
1805 {
1806 struct address_space *mapping = page->mapping;
1807 unsigned long flags;
1808
1809 if (unlikely(!mapping))
1810 return;
1811
1812 spin_lock(&mapping->private_lock);
1813 lock_page_memcg(page);
1814 SetPageDirty(page);
1815 spin_unlock(&mapping->private_lock);
1816
1817 spin_lock_irqsave(&mapping->tree_lock, flags);
1818 WARN_ON_ONCE(!PageUptodate(page));
1819 account_page_dirtied(page, mapping);
1820 radix_tree_tag_set(&mapping->page_tree,
1821 page_index(page), PAGECACHE_TAG_DIRTY);
1822 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1823 unlock_page_memcg(page);
1824
1825 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1826 return;
1827 }
1828
1829 static int f2fs_set_data_page_dirty(struct page *page)
1830 {
1831 struct address_space *mapping = page->mapping;
1832 struct inode *inode = mapping->host;
1833
1834 trace_f2fs_set_page_dirty(page, DATA);
1835
1836 if (!PageUptodate(page))
1837 SetPageUptodate(page);
1838
1839 if (f2fs_is_atomic_file(inode)) {
1840 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1841 register_inmem_page(inode, page);
1842 return 1;
1843 }
1844 /*
1845 * Previously, this page has been registered, we just
1846 * return here.
1847 */
1848 return 0;
1849 }
1850
1851 if (!PageDirty(page)) {
1852 f2fs_set_page_dirty_nobuffers(page);
1853 update_dirty_page(inode, page);
1854 return 1;
1855 }
1856 return 0;
1857 }
1858
1859 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1860 {
1861 struct inode *inode = mapping->host;
1862
1863 if (f2fs_has_inline_data(inode))
1864 return 0;
1865
1866 /* make sure allocating whole blocks */
1867 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1868 filemap_write_and_wait(mapping);
1869
1870 return generic_block_bmap(mapping, block, get_data_block_bmap);
1871 }
1872
1873 #ifdef CONFIG_MIGRATION
1874 #include <linux/migrate.h>
1875
1876 int f2fs_migrate_page(struct address_space *mapping,
1877 struct page *newpage, struct page *page, enum migrate_mode mode)
1878 {
1879 int rc, extra_count;
1880 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
1881 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
1882
1883 BUG_ON(PageWriteback(page));
1884
1885 /* migrating an atomic written page is safe with the inmem_lock hold */
1886 if (atomic_written && !mutex_trylock(&fi->inmem_lock))
1887 return -EAGAIN;
1888
1889 /*
1890 * A reference is expected if PagePrivate set when move mapping,
1891 * however F2FS breaks this for maintaining dirty page counts when
1892 * truncating pages. So here adjusting the 'extra_count' make it work.
1893 */
1894 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
1895 rc = migrate_page_move_mapping(mapping, newpage,
1896 page, NULL, mode, extra_count);
1897 if (rc != MIGRATEPAGE_SUCCESS) {
1898 if (atomic_written)
1899 mutex_unlock(&fi->inmem_lock);
1900 return rc;
1901 }
1902
1903 if (atomic_written) {
1904 struct inmem_pages *cur;
1905 list_for_each_entry(cur, &fi->inmem_pages, list)
1906 if (cur->page == page) {
1907 cur->page = newpage;
1908 break;
1909 }
1910 mutex_unlock(&fi->inmem_lock);
1911 put_page(page);
1912 get_page(newpage);
1913 }
1914
1915 if (PagePrivate(page))
1916 SetPagePrivate(newpage);
1917 set_page_private(newpage, page_private(page));
1918
1919 migrate_page_copy(newpage, page);
1920
1921 return MIGRATEPAGE_SUCCESS;
1922 }
1923 #endif
1924
1925 const struct address_space_operations f2fs_dblock_aops = {
1926 .readpage = f2fs_read_data_page,
1927 .readpages = f2fs_read_data_pages,
1928 .writepage = f2fs_write_data_page,
1929 .writepages = f2fs_write_data_pages,
1930 .write_begin = f2fs_write_begin,
1931 .write_end = f2fs_write_end,
1932 .set_page_dirty = f2fs_set_data_page_dirty,
1933 .invalidatepage = f2fs_invalidate_page,
1934 .releasepage = f2fs_release_page,
1935 .direct_IO = f2fs_direct_IO,
1936 .bmap = f2fs_bmap,
1937 #ifdef CONFIG_MIGRATION
1938 .migratepage = f2fs_migrate_page,
1939 #endif
1940 };
Ubuntu Jammy Linux kernel mirror