GNU Linux-libre 4.9.308-gnu1
[releases.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_hmsmr(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         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
244                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
245                 return -EFAULT;
246
247         trace_f2fs_submit_page_bio(page, fio);
248         f2fs_trace_ios(fio, 0);
249
250         /* Allocate a new bio */
251         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
252
253         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
254                 bio_put(bio);
255                 return -EFAULT;
256         }
257         bio_set_op_attrs(bio, fio->op, fio->op_flags);
258
259         __submit_bio(fio->sbi, bio, fio->type);
260         return 0;
261 }
262
263 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
264 {
265         struct f2fs_sb_info *sbi = fio->sbi;
266         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
267         struct f2fs_bio_info *io;
268         bool is_read = is_read_io(fio->op);
269         struct page *bio_page;
270
271         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
272
273         if (__is_valid_data_blkaddr(fio->old_blkaddr))
274                 verify_block_addr(fio, fio->old_blkaddr);
275         verify_block_addr(fio, fio->new_blkaddr);
276
277         down_write(&io->io_rwsem);
278
279         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
280             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
281                 __submit_merged_bio(io);
282 alloc_new:
283         if (io->bio == NULL) {
284                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
285
286                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
287                                                 bio_blocks, is_read);
288                 io->fio = *fio;
289         }
290
291         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
292
293         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
294                                                         PAGE_SIZE) {
295                 __submit_merged_bio(io);
296                 goto alloc_new;
297         }
298
299         io->last_block_in_bio = fio->new_blkaddr;
300         f2fs_trace_ios(fio, 0);
301
302         up_write(&io->io_rwsem);
303         trace_f2fs_submit_page_mbio(fio->page, fio);
304 }
305
306 static void __set_data_blkaddr(struct dnode_of_data *dn)
307 {
308         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
309         __le32 *addr_array;
310
311         /* Get physical address of data block */
312         addr_array = blkaddr_in_node(rn);
313         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
314 }
315
316 /*
317  * Lock ordering for the change of data block address:
318  * ->data_page
319  *  ->node_page
320  *    update block addresses in the node page
321  */
322 void set_data_blkaddr(struct dnode_of_data *dn)
323 {
324         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
325         __set_data_blkaddr(dn);
326         if (set_page_dirty(dn->node_page))
327                 dn->node_changed = true;
328 }
329
330 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
331 {
332         dn->data_blkaddr = blkaddr;
333         set_data_blkaddr(dn);
334         f2fs_update_extent_cache(dn);
335 }
336
337 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
338 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
339 {
340         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
341
342         if (!count)
343                 return 0;
344
345         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
346                 return -EPERM;
347         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
348                 return -ENOSPC;
349
350         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
351                                                 dn->ofs_in_node, count);
352
353         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
354
355         for (; count > 0; dn->ofs_in_node++) {
356                 block_t blkaddr =
357                         datablock_addr(dn->node_page, dn->ofs_in_node);
358                 if (blkaddr == NULL_ADDR) {
359                         dn->data_blkaddr = NEW_ADDR;
360                         __set_data_blkaddr(dn);
361                         count--;
362                 }
363         }
364
365         if (set_page_dirty(dn->node_page))
366                 dn->node_changed = true;
367         return 0;
368 }
369
370 /* Should keep dn->ofs_in_node unchanged */
371 int reserve_new_block(struct dnode_of_data *dn)
372 {
373         unsigned int ofs_in_node = dn->ofs_in_node;
374         int ret;
375
376         ret = reserve_new_blocks(dn, 1);
377         dn->ofs_in_node = ofs_in_node;
378         return ret;
379 }
380
381 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
382 {
383         bool need_put = dn->inode_page ? false : true;
384         int err;
385
386         err = get_dnode_of_data(dn, index, ALLOC_NODE);
387         if (err)
388                 return err;
389
390         if (dn->data_blkaddr == NULL_ADDR)
391                 err = reserve_new_block(dn);
392         if (err || need_put)
393                 f2fs_put_dnode(dn);
394         return err;
395 }
396
397 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
398 {
399         struct extent_info ei;
400         struct inode *inode = dn->inode;
401
402         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
403                 dn->data_blkaddr = ei.blk + index - ei.fofs;
404                 return 0;
405         }
406
407         return f2fs_reserve_block(dn, index);
408 }
409
410 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
411                                                 int op_flags, bool for_write)
412 {
413         struct address_space *mapping = inode->i_mapping;
414         struct dnode_of_data dn;
415         struct page *page;
416         struct extent_info ei;
417         int err;
418         struct f2fs_io_info fio = {
419                 .sbi = F2FS_I_SB(inode),
420                 .type = DATA,
421                 .op = REQ_OP_READ,
422                 .op_flags = op_flags,
423                 .encrypted_page = NULL,
424         };
425
426         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
427                 return read_mapping_page(mapping, index, NULL);
428
429         page = f2fs_grab_cache_page(mapping, index, for_write);
430         if (!page)
431                 return ERR_PTR(-ENOMEM);
432
433         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
434                 dn.data_blkaddr = ei.blk + index - ei.fofs;
435                 goto got_it;
436         }
437
438         set_new_dnode(&dn, inode, NULL, NULL, 0);
439         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
440         if (err)
441                 goto put_err;
442         f2fs_put_dnode(&dn);
443
444         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
445                 err = -ENOENT;
446                 goto put_err;
447         }
448 got_it:
449         if (PageUptodate(page)) {
450                 unlock_page(page);
451                 return page;
452         }
453
454         /*
455          * A new dentry page is allocated but not able to be written, since its
456          * new inode page couldn't be allocated due to -ENOSPC.
457          * In such the case, its blkaddr can be remained as NEW_ADDR.
458          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
459          */
460         if (dn.data_blkaddr == NEW_ADDR) {
461                 zero_user_segment(page, 0, PAGE_SIZE);
462                 if (!PageUptodate(page))
463                         SetPageUptodate(page);
464                 unlock_page(page);
465                 return page;
466         }
467
468         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
469         fio.page = page;
470         err = f2fs_submit_page_bio(&fio);
471         if (err)
472                 goto put_err;
473         return page;
474
475 put_err:
476         f2fs_put_page(page, 1);
477         return ERR_PTR(err);
478 }
479
480 struct page *find_data_page(struct inode *inode, pgoff_t index)
481 {
482         struct address_space *mapping = inode->i_mapping;
483         struct page *page;
484
485         page = find_get_page(mapping, index);
486         if (page && PageUptodate(page))
487                 return page;
488         f2fs_put_page(page, 0);
489
490         page = get_read_data_page(inode, index, READ_SYNC, false);
491         if (IS_ERR(page))
492                 return page;
493
494         if (PageUptodate(page))
495                 return page;
496
497         wait_on_page_locked(page);
498         if (unlikely(!PageUptodate(page))) {
499                 f2fs_put_page(page, 0);
500                 return ERR_PTR(-EIO);
501         }
502         return page;
503 }
504
505 /*
506  * If it tries to access a hole, return an error.
507  * Because, the callers, functions in dir.c and GC, should be able to know
508  * whether this page exists or not.
509  */
510 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
511                                                         bool for_write)
512 {
513         struct address_space *mapping = inode->i_mapping;
514         struct page *page;
515 repeat:
516         page = get_read_data_page(inode, index, READ_SYNC, for_write);
517         if (IS_ERR(page))
518                 return page;
519
520         /* wait for read completion */
521         lock_page(page);
522         if (unlikely(page->mapping != mapping)) {
523                 f2fs_put_page(page, 1);
524                 goto repeat;
525         }
526         if (unlikely(!PageUptodate(page))) {
527                 f2fs_put_page(page, 1);
528                 return ERR_PTR(-EIO);
529         }
530         return page;
531 }
532
533 /*
534  * Caller ensures that this data page is never allocated.
535  * A new zero-filled data page is allocated in the page cache.
536  *
537  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
538  * f2fs_unlock_op().
539  * Note that, ipage is set only by make_empty_dir, and if any error occur,
540  * ipage should be released by this function.
541  */
542 struct page *get_new_data_page(struct inode *inode,
543                 struct page *ipage, pgoff_t index, bool new_i_size)
544 {
545         struct address_space *mapping = inode->i_mapping;
546         struct page *page;
547         struct dnode_of_data dn;
548         int err;
549
550         page = f2fs_grab_cache_page(mapping, index, true);
551         if (!page) {
552                 /*
553                  * before exiting, we should make sure ipage will be released
554                  * if any error occur.
555                  */
556                 f2fs_put_page(ipage, 1);
557                 return ERR_PTR(-ENOMEM);
558         }
559
560         set_new_dnode(&dn, inode, ipage, NULL, 0);
561         err = f2fs_reserve_block(&dn, index);
562         if (err) {
563                 f2fs_put_page(page, 1);
564                 return ERR_PTR(err);
565         }
566         if (!ipage)
567                 f2fs_put_dnode(&dn);
568
569         if (PageUptodate(page))
570                 goto got_it;
571
572         if (dn.data_blkaddr == NEW_ADDR) {
573                 zero_user_segment(page, 0, PAGE_SIZE);
574                 if (!PageUptodate(page))
575                         SetPageUptodate(page);
576         } else {
577                 f2fs_put_page(page, 1);
578
579                 /* if ipage exists, blkaddr should be NEW_ADDR */
580                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
581                 page = get_lock_data_page(inode, index, true);
582                 if (IS_ERR(page))
583                         return page;
584         }
585 got_it:
586         if (new_i_size && i_size_read(inode) <
587                                 ((loff_t)(index + 1) << PAGE_SHIFT))
588                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
589         return page;
590 }
591
592 static int __allocate_data_block(struct dnode_of_data *dn)
593 {
594         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
595         struct f2fs_summary sum;
596         struct node_info ni;
597         int seg = CURSEG_WARM_DATA;
598         pgoff_t fofs;
599         blkcnt_t count = 1;
600
601         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
602                 return -EPERM;
603
604         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
605         if (dn->data_blkaddr == NEW_ADDR)
606                 goto alloc;
607
608         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
609                 return -ENOSPC;
610
611 alloc:
612         get_node_info(sbi, dn->nid, &ni);
613         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
614
615         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
616                 seg = CURSEG_DIRECT_IO;
617
618         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
619                                                                 &sum, seg);
620         set_data_blkaddr(dn);
621
622         /* update i_size */
623         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
624                                                         dn->ofs_in_node;
625         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
626                 f2fs_i_size_write(dn->inode,
627                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
628         return 0;
629 }
630
631 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
632 {
633         struct inode *inode = file_inode(iocb->ki_filp);
634         struct f2fs_map_blocks map;
635         ssize_t ret = 0;
636
637         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
638         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
639         if (map.m_len > map.m_lblk)
640                 map.m_len -= map.m_lblk;
641         else
642                 map.m_len = 0;
643
644         map.m_next_pgofs = NULL;
645
646         if (iocb->ki_flags & IOCB_DIRECT) {
647                 ret = f2fs_convert_inline_inode(inode);
648                 if (ret)
649                         return ret;
650                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
651         }
652         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
653                 ret = f2fs_convert_inline_inode(inode);
654                 if (ret)
655                         return ret;
656         }
657         if (!f2fs_has_inline_data(inode))
658                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
659         return ret;
660 }
661
662 /*
663  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
664  * f2fs_map_blocks structure.
665  * If original data blocks are allocated, then give them to blockdev.
666  * Otherwise,
667  *     a. preallocate requested block addresses
668  *     b. do not use extent cache for better performance
669  *     c. give the block addresses to blockdev
670  */
671 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
672                                                 int create, int flag)
673 {
674         unsigned int maxblocks = map->m_len;
675         struct dnode_of_data dn;
676         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
677         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
678         pgoff_t pgofs, end_offset, end;
679         int err = 0, ofs = 1;
680         unsigned int ofs_in_node, last_ofs_in_node;
681         blkcnt_t prealloc;
682         struct extent_info ei;
683         bool allocated = false;
684         block_t blkaddr;
685
686         if (!maxblocks)
687                 return 0;
688
689         map->m_len = 0;
690         map->m_flags = 0;
691
692         /* it only supports block size == page size */
693         pgofs = (pgoff_t)map->m_lblk;
694         end = pgofs + maxblocks;
695
696         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
697                 map->m_pblk = ei.blk + pgofs - ei.fofs;
698                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
699                 map->m_flags = F2FS_MAP_MAPPED;
700                 goto out;
701         }
702
703 next_dnode:
704         if (create)
705                 f2fs_lock_op(sbi);
706
707         /* When reading holes, we need its node page */
708         set_new_dnode(&dn, inode, NULL, NULL, 0);
709         err = get_dnode_of_data(&dn, pgofs, mode);
710         if (err) {
711                 if (flag == F2FS_GET_BLOCK_BMAP)
712                         map->m_pblk = 0;
713                 if (err == -ENOENT) {
714                         err = 0;
715                         if (map->m_next_pgofs)
716                                 *map->m_next_pgofs =
717                                         get_next_page_offset(&dn, pgofs);
718                 }
719                 goto unlock_out;
720         }
721
722         prealloc = 0;
723         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
724         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
725
726 next_block:
727         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
728
729         if (__is_valid_data_blkaddr(blkaddr) &&
730                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
731                 err = -EFAULT;
732                 goto sync_out;
733         }
734
735         if (!is_valid_data_blkaddr(sbi, blkaddr)) {
736                 if (create) {
737                         if (unlikely(f2fs_cp_error(sbi))) {
738                                 err = -EIO;
739                                 goto sync_out;
740                         }
741                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
742                                 if (blkaddr == NULL_ADDR) {
743                                         prealloc++;
744                                         last_ofs_in_node = dn.ofs_in_node;
745                                 }
746                         } else {
747                                 err = __allocate_data_block(&dn);
748                                 if (!err) {
749                                         set_inode_flag(inode, FI_APPEND_WRITE);
750                                         allocated = true;
751                                 }
752                         }
753                         if (err)
754                                 goto sync_out;
755                         map->m_flags = F2FS_MAP_NEW;
756                         blkaddr = dn.data_blkaddr;
757                 } else {
758                         if (flag == F2FS_GET_BLOCK_BMAP) {
759                                 map->m_pblk = 0;
760                                 goto sync_out;
761                         }
762                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
763                                                 blkaddr == NULL_ADDR) {
764                                 if (map->m_next_pgofs)
765                                         *map->m_next_pgofs = pgofs + 1;
766                         }
767                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
768                                                 blkaddr != NEW_ADDR)
769                                 goto sync_out;
770                 }
771         }
772
773         if (flag == F2FS_GET_BLOCK_PRE_AIO)
774                 goto skip;
775
776         if (map->m_len == 0) {
777                 /* preallocated unwritten block should be mapped for fiemap. */
778                 if (blkaddr == NEW_ADDR)
779                         map->m_flags |= F2FS_MAP_UNWRITTEN;
780                 map->m_flags |= F2FS_MAP_MAPPED;
781
782                 map->m_pblk = blkaddr;
783                 map->m_len = 1;
784         } else if ((map->m_pblk != NEW_ADDR &&
785                         blkaddr == (map->m_pblk + ofs)) ||
786                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
787                         flag == F2FS_GET_BLOCK_PRE_DIO) {
788                 ofs++;
789                 map->m_len++;
790         } else {
791                 goto sync_out;
792         }
793
794 skip:
795         dn.ofs_in_node++;
796         pgofs++;
797
798         /* preallocate blocks in batch for one dnode page */
799         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
800                         (pgofs == end || dn.ofs_in_node == end_offset)) {
801
802                 dn.ofs_in_node = ofs_in_node;
803                 err = reserve_new_blocks(&dn, prealloc);
804                 if (err)
805                         goto sync_out;
806                 allocated = dn.node_changed;
807
808                 map->m_len += dn.ofs_in_node - ofs_in_node;
809                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
810                         err = -ENOSPC;
811                         goto sync_out;
812                 }
813                 dn.ofs_in_node = end_offset;
814         }
815
816         if (pgofs >= end)
817                 goto sync_out;
818         else if (dn.ofs_in_node < end_offset)
819                 goto next_block;
820
821         f2fs_put_dnode(&dn);
822
823         if (create) {
824                 f2fs_unlock_op(sbi);
825                 f2fs_balance_fs(sbi, allocated);
826         }
827         allocated = false;
828         goto next_dnode;
829
830 sync_out:
831         f2fs_put_dnode(&dn);
832 unlock_out:
833         if (create) {
834                 f2fs_unlock_op(sbi);
835                 f2fs_balance_fs(sbi, allocated);
836         }
837 out:
838         trace_f2fs_map_blocks(inode, map, err);
839         return err;
840 }
841
842 static int __get_data_block(struct inode *inode, sector_t iblock,
843                         struct buffer_head *bh, int create, int flag,
844                         pgoff_t *next_pgofs)
845 {
846         struct f2fs_map_blocks map;
847         int ret;
848
849         map.m_lblk = iblock;
850         map.m_len = bh->b_size >> inode->i_blkbits;
851         map.m_next_pgofs = next_pgofs;
852
853         ret = f2fs_map_blocks(inode, &map, create, flag);
854         if (!ret) {
855                 map_bh(bh, inode->i_sb, map.m_pblk);
856                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
857                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
858         }
859         return ret;
860 }
861
862 static int get_data_block(struct inode *inode, sector_t iblock,
863                         struct buffer_head *bh_result, int create, int flag,
864                         pgoff_t *next_pgofs)
865 {
866         return __get_data_block(inode, iblock, bh_result, create,
867                                                         flag, next_pgofs);
868 }
869
870 static int get_data_block_dio(struct inode *inode, sector_t iblock,
871                         struct buffer_head *bh_result, int create)
872 {
873         return __get_data_block(inode, iblock, bh_result, create,
874                                                 F2FS_GET_BLOCK_DIO, NULL);
875 }
876
877 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
878                         struct buffer_head *bh_result, int create)
879 {
880         /* Block number less than F2FS MAX BLOCKS */
881         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
882                 return -EFBIG;
883
884         return __get_data_block(inode, iblock, bh_result, create,
885                                                 F2FS_GET_BLOCK_BMAP, NULL);
886 }
887
888 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
889 {
890         return (offset >> inode->i_blkbits);
891 }
892
893 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
894 {
895         return (blk << inode->i_blkbits);
896 }
897
898 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
899                 u64 start, u64 len)
900 {
901         struct buffer_head map_bh;
902         sector_t start_blk, last_blk;
903         pgoff_t next_pgofs;
904         loff_t isize;
905         u64 logical = 0, phys = 0, size = 0;
906         u32 flags = 0;
907         int ret = 0;
908
909         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
910         if (ret)
911                 return ret;
912
913         if (f2fs_has_inline_data(inode)) {
914                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
915                 if (ret != -EAGAIN)
916                         return ret;
917         }
918
919         inode_lock(inode);
920
921         isize = i_size_read(inode);
922         if (start >= isize)
923                 goto out;
924
925         if (start + len > isize)
926                 len = isize - start;
927
928         if (logical_to_blk(inode, len) == 0)
929                 len = blk_to_logical(inode, 1);
930
931         start_blk = logical_to_blk(inode, start);
932         last_blk = logical_to_blk(inode, start + len - 1);
933
934 next:
935         memset(&map_bh, 0, sizeof(struct buffer_head));
936         map_bh.b_size = len;
937
938         ret = get_data_block(inode, start_blk, &map_bh, 0,
939                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
940         if (ret)
941                 goto out;
942
943         /* HOLE */
944         if (!buffer_mapped(&map_bh)) {
945                 start_blk = next_pgofs;
946                 /* Go through holes util pass the EOF */
947                 if (blk_to_logical(inode, start_blk) < isize)
948                         goto prep_next;
949                 /* Found a hole beyond isize means no more extents.
950                  * Note that the premise is that filesystems don't
951                  * punch holes beyond isize and keep size unchanged.
952                  */
953                 flags |= FIEMAP_EXTENT_LAST;
954         }
955
956         if (size) {
957                 if (f2fs_encrypted_inode(inode))
958                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
959
960                 ret = fiemap_fill_next_extent(fieinfo, logical,
961                                 phys, size, flags);
962         }
963
964         if (start_blk > last_blk || ret)
965                 goto out;
966
967         logical = blk_to_logical(inode, start_blk);
968         phys = blk_to_logical(inode, map_bh.b_blocknr);
969         size = map_bh.b_size;
970         flags = 0;
971         if (buffer_unwritten(&map_bh))
972                 flags = FIEMAP_EXTENT_UNWRITTEN;
973
974         start_blk += logical_to_blk(inode, size);
975
976 prep_next:
977         cond_resched();
978         if (fatal_signal_pending(current))
979                 ret = -EINTR;
980         else
981                 goto next;
982 out:
983         if (ret == 1)
984                 ret = 0;
985
986         inode_unlock(inode);
987         return ret;
988 }
989
990 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
991                                  unsigned nr_pages)
992 {
993         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
994         struct fscrypt_ctx *ctx = NULL;
995         struct block_device *bdev = sbi->sb->s_bdev;
996         struct bio *bio;
997
998         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
999                 return ERR_PTR(-EFAULT);
1000
1001         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1002                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1003                 if (IS_ERR(ctx))
1004                         return ERR_CAST(ctx);
1005
1006                 /* wait the page to be moved by cleaning */
1007                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1008         }
1009
1010         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1011         if (!bio) {
1012                 if (ctx)
1013                         fscrypt_release_ctx(ctx);
1014                 return ERR_PTR(-ENOMEM);
1015         }
1016         bio->bi_bdev = bdev;
1017         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
1018         bio->bi_end_io = f2fs_read_end_io;
1019         bio->bi_private = ctx;
1020
1021         return bio;
1022 }
1023
1024 /*
1025  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1026  * Major change was from block_size == page_size in f2fs by default.
1027  */
1028 static int f2fs_mpage_readpages(struct address_space *mapping,
1029                         struct list_head *pages, struct page *page,
1030                         unsigned nr_pages)
1031 {
1032         struct bio *bio = NULL;
1033         unsigned page_idx;
1034         sector_t last_block_in_bio = 0;
1035         struct inode *inode = mapping->host;
1036         const unsigned blkbits = inode->i_blkbits;
1037         const unsigned blocksize = 1 << blkbits;
1038         sector_t block_in_file;
1039         sector_t last_block;
1040         sector_t last_block_in_file;
1041         sector_t block_nr;
1042         struct f2fs_map_blocks map;
1043
1044         map.m_pblk = 0;
1045         map.m_lblk = 0;
1046         map.m_len = 0;
1047         map.m_flags = 0;
1048         map.m_next_pgofs = NULL;
1049
1050         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1051
1052                 prefetchw(&page->flags);
1053                 if (pages) {
1054                         page = list_entry(pages->prev, struct page, lru);
1055                         list_del(&page->lru);
1056                         if (add_to_page_cache_lru(page, mapping,
1057                                                   page->index,
1058                                                   readahead_gfp_mask(mapping)))
1059                                 goto next_page;
1060                 }
1061
1062                 block_in_file = (sector_t)page->index;
1063                 last_block = block_in_file + nr_pages;
1064                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1065                                                                 blkbits;
1066                 if (last_block > last_block_in_file)
1067                         last_block = last_block_in_file;
1068
1069                 /*
1070                  * Map blocks using the previous result first.
1071                  */
1072                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1073                                 block_in_file > map.m_lblk &&
1074                                 block_in_file < (map.m_lblk + map.m_len))
1075                         goto got_it;
1076
1077                 /*
1078                  * Then do more f2fs_map_blocks() calls until we are
1079                  * done with this page.
1080                  */
1081                 map.m_flags = 0;
1082
1083                 if (block_in_file < last_block) {
1084                         map.m_lblk = block_in_file;
1085                         map.m_len = last_block - block_in_file;
1086
1087                         if (f2fs_map_blocks(inode, &map, 0,
1088                                                 F2FS_GET_BLOCK_READ))
1089                                 goto set_error_page;
1090                 }
1091 got_it:
1092                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1093                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1094                         SetPageMappedToDisk(page);
1095
1096                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1097                                 SetPageUptodate(page);
1098                                 goto confused;
1099                         }
1100
1101                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1102                                                                 DATA_GENERIC))
1103                                 goto set_error_page;
1104                 } else {
1105                         zero_user_segment(page, 0, PAGE_SIZE);
1106                         if (!PageUptodate(page))
1107                                 SetPageUptodate(page);
1108                         unlock_page(page);
1109                         goto next_page;
1110                 }
1111
1112                 /*
1113                  * This page will go to BIO.  Do we need to send this
1114                  * BIO off first?
1115                  */
1116                 if (bio && (last_block_in_bio != block_nr - 1)) {
1117 submit_and_realloc:
1118                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1119                         bio = NULL;
1120                 }
1121                 if (bio == NULL) {
1122                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1123                         if (IS_ERR(bio)) {
1124                                 bio = NULL;
1125                                 goto set_error_page;
1126                         }
1127                         bio_set_op_attrs(bio, REQ_OP_READ, 0);
1128                 }
1129
1130                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1131                         goto submit_and_realloc;
1132
1133                 last_block_in_bio = block_nr;
1134                 goto next_page;
1135 set_error_page:
1136                 SetPageError(page);
1137                 zero_user_segment(page, 0, PAGE_SIZE);
1138                 unlock_page(page);
1139                 goto next_page;
1140 confused:
1141                 if (bio) {
1142                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1143                         bio = NULL;
1144                 }
1145                 unlock_page(page);
1146 next_page:
1147                 if (pages)
1148                         put_page(page);
1149         }
1150         BUG_ON(pages && !list_empty(pages));
1151         if (bio)
1152                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1153         return 0;
1154 }
1155
1156 static int f2fs_read_data_page(struct file *file, struct page *page)
1157 {
1158         struct inode *inode = page->mapping->host;
1159         int ret = -EAGAIN;
1160
1161         trace_f2fs_readpage(page, DATA);
1162
1163         /* If the file has inline data, try to read it directly */
1164         if (f2fs_has_inline_data(inode))
1165                 ret = f2fs_read_inline_data(inode, page);
1166         if (ret == -EAGAIN)
1167                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1168         return ret;
1169 }
1170
1171 static int f2fs_read_data_pages(struct file *file,
1172                         struct address_space *mapping,
1173                         struct list_head *pages, unsigned nr_pages)
1174 {
1175         struct inode *inode = file->f_mapping->host;
1176         struct page *page = list_entry(pages->prev, struct page, lru);
1177
1178         trace_f2fs_readpages(inode, page, nr_pages);
1179
1180         /* If the file has inline data, skip readpages */
1181         if (f2fs_has_inline_data(inode))
1182                 return 0;
1183
1184         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1185 }
1186
1187 int do_write_data_page(struct f2fs_io_info *fio)
1188 {
1189         struct page *page = fio->page;
1190         struct inode *inode = page->mapping->host;
1191         struct dnode_of_data dn;
1192         int err = 0;
1193
1194         set_new_dnode(&dn, inode, NULL, NULL, 0);
1195         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1196         if (err)
1197                 return err;
1198
1199         fio->old_blkaddr = dn.data_blkaddr;
1200
1201         /* This page is already truncated */
1202         if (fio->old_blkaddr == NULL_ADDR) {
1203                 ClearPageUptodate(page);
1204                 clear_cold_data(page);
1205                 goto out_writepage;
1206         }
1207
1208         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1209                 gfp_t gfp_flags = GFP_NOFS;
1210
1211                 /* wait for GCed encrypted page writeback */
1212                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1213                                                         fio->old_blkaddr);
1214 retry_encrypt:
1215                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1216                                                                 gfp_flags);
1217                 if (IS_ERR(fio->encrypted_page)) {
1218                         err = PTR_ERR(fio->encrypted_page);
1219                         if (err == -ENOMEM) {
1220                                 /* flush pending ios and wait for a while */
1221                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1222                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1223                                 gfp_flags |= __GFP_NOFAIL;
1224                                 err = 0;
1225                                 goto retry_encrypt;
1226                         }
1227                         goto out_writepage;
1228                 }
1229         }
1230
1231         set_page_writeback(page);
1232
1233         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1234                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1235                                                         DATA_GENERIC)) {
1236                 err = -EFAULT;
1237                 goto out_writepage;
1238         }
1239         /*
1240          * If current allocation needs SSR,
1241          * it had better in-place writes for updated data.
1242          */
1243         if (unlikely(is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1244                         !is_cold_data(page) &&
1245                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1246                         need_inplace_update(inode))) {
1247                 rewrite_data_page(fio);
1248                 set_inode_flag(inode, FI_UPDATE_WRITE);
1249                 trace_f2fs_do_write_data_page(page, IPU);
1250         } else {
1251                 write_data_page(&dn, fio);
1252                 trace_f2fs_do_write_data_page(page, OPU);
1253                 set_inode_flag(inode, FI_APPEND_WRITE);
1254                 if (page->index == 0)
1255                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1256         }
1257 out_writepage:
1258         f2fs_put_dnode(&dn);
1259         return err;
1260 }
1261
1262 static int f2fs_write_data_page(struct page *page,
1263                                         struct writeback_control *wbc)
1264 {
1265         struct inode *inode = page->mapping->host;
1266         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1267         loff_t i_size = i_size_read(inode);
1268         const pgoff_t end_index = ((unsigned long long) i_size)
1269                                                         >> PAGE_SHIFT;
1270         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
1271         unsigned offset = 0;
1272         bool need_balance_fs = false;
1273         int err = 0;
1274         struct f2fs_io_info fio = {
1275                 .sbi = sbi,
1276                 .type = DATA,
1277                 .op = REQ_OP_WRITE,
1278                 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1279                 .page = page,
1280                 .encrypted_page = NULL,
1281         };
1282
1283         trace_f2fs_writepage(page, DATA);
1284
1285         if (page->index < end_index)
1286                 goto write;
1287
1288         /*
1289          * If the offset is out-of-range of file size,
1290          * this page does not have to be written to disk.
1291          */
1292         offset = i_size & (PAGE_SIZE - 1);
1293         if ((page->index >= end_index + 1) || !offset)
1294                 goto out;
1295
1296         zero_user_segment(page, offset, PAGE_SIZE);
1297 write:
1298         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1299                 goto redirty_out;
1300         if (f2fs_is_drop_cache(inode))
1301                 goto out;
1302         /* we should not write 0'th page having journal header */
1303         if (f2fs_is_volatile_file(inode) && (!page->index ||
1304                         (!wbc->for_reclaim &&
1305                         available_free_memory(sbi, BASE_CHECK))))
1306                 goto redirty_out;
1307
1308         /* we should bypass data pages to proceed the kworkder jobs */
1309         if (unlikely(f2fs_cp_error(sbi))) {
1310                 mapping_set_error(page->mapping, -EIO);
1311                 goto out;
1312         }
1313
1314         /* Dentry blocks are controlled by checkpoint */
1315         if (S_ISDIR(inode->i_mode)) {
1316                 err = do_write_data_page(&fio);
1317                 goto done;
1318         }
1319
1320         if (!wbc->for_reclaim)
1321                 need_balance_fs = true;
1322         else if (has_not_enough_free_secs(sbi, 0, 0))
1323                 goto redirty_out;
1324
1325         err = -EAGAIN;
1326         f2fs_lock_op(sbi);
1327         if (f2fs_has_inline_data(inode))
1328                 err = f2fs_write_inline_data(inode, page);
1329         if (err == -EAGAIN)
1330                 err = do_write_data_page(&fio);
1331         if (F2FS_I(inode)->last_disk_size < psize)
1332                 F2FS_I(inode)->last_disk_size = psize;
1333         f2fs_unlock_op(sbi);
1334 done:
1335         if (err && err != -ENOENT)
1336                 goto redirty_out;
1337
1338         clear_cold_data(page);
1339 out:
1340         inode_dec_dirty_pages(inode);
1341         if (err) {
1342                 ClearPageUptodate(page);
1343                 clear_cold_data(page);
1344         }
1345
1346         if (wbc->for_reclaim) {
1347                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1348                 remove_dirty_inode(inode);
1349         }
1350
1351         unlock_page(page);
1352         f2fs_balance_fs(sbi, need_balance_fs);
1353
1354         if (unlikely(f2fs_cp_error(sbi)))
1355                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1356
1357         return 0;
1358
1359 redirty_out:
1360         redirty_page_for_writepage(wbc, page);
1361         unlock_page(page);
1362         return err;
1363 }
1364
1365 /*
1366  * This function was copied from write_cche_pages from mm/page-writeback.c.
1367  * The major change is making write step of cold data page separately from
1368  * warm/hot data page.
1369  */
1370 static int f2fs_write_cache_pages(struct address_space *mapping,
1371                                         struct writeback_control *wbc)
1372 {
1373         int ret = 0;
1374         int done = 0;
1375         struct pagevec pvec;
1376         int nr_pages;
1377         pgoff_t uninitialized_var(writeback_index);
1378         pgoff_t index;
1379         pgoff_t end;            /* Inclusive */
1380         pgoff_t done_index;
1381         int cycled;
1382         int range_whole = 0;
1383         int tag;
1384         int nwritten = 0;
1385
1386         pagevec_init(&pvec, 0);
1387
1388         if (wbc->range_cyclic) {
1389                 writeback_index = mapping->writeback_index; /* prev offset */
1390                 index = writeback_index;
1391                 if (index == 0)
1392                         cycled = 1;
1393                 else
1394                         cycled = 0;
1395                 end = -1;
1396         } else {
1397                 index = wbc->range_start >> PAGE_SHIFT;
1398                 end = wbc->range_end >> PAGE_SHIFT;
1399                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1400                         range_whole = 1;
1401                 cycled = 1; /* ignore range_cyclic tests */
1402         }
1403         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1404                 tag = PAGECACHE_TAG_TOWRITE;
1405         else
1406                 tag = PAGECACHE_TAG_DIRTY;
1407 retry:
1408         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1409                 tag_pages_for_writeback(mapping, index, end);
1410         done_index = index;
1411         while (!done && (index <= end)) {
1412                 int i;
1413
1414                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1415                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1416                 if (nr_pages == 0)
1417                         break;
1418
1419                 for (i = 0; i < nr_pages; i++) {
1420                         struct page *page = pvec.pages[i];
1421
1422                         if (page->index > end) {
1423                                 done = 1;
1424                                 break;
1425                         }
1426
1427                         done_index = page->index;
1428
1429                         lock_page(page);
1430
1431                         if (unlikely(page->mapping != mapping)) {
1432 continue_unlock:
1433                                 unlock_page(page);
1434                                 continue;
1435                         }
1436
1437                         if (!PageDirty(page)) {
1438                                 /* someone wrote it for us */
1439                                 goto continue_unlock;
1440                         }
1441
1442                         if (PageWriteback(page)) {
1443                                 if (wbc->sync_mode != WB_SYNC_NONE)
1444                                         f2fs_wait_on_page_writeback(page,
1445                                                                 DATA, true);
1446                                 else
1447                                         goto continue_unlock;
1448                         }
1449
1450                         BUG_ON(PageWriteback(page));
1451                         if (!clear_page_dirty_for_io(page))
1452                                 goto continue_unlock;
1453
1454                         ret = mapping->a_ops->writepage(page, wbc);
1455                         if (unlikely(ret)) {
1456                                 done_index = page->index + 1;
1457                                 done = 1;
1458                                 break;
1459                         } else {
1460                                 nwritten++;
1461                         }
1462
1463                         if (--wbc->nr_to_write <= 0 &&
1464                             wbc->sync_mode == WB_SYNC_NONE) {
1465                                 done = 1;
1466                                 break;
1467                         }
1468                 }
1469                 pagevec_release(&pvec);
1470                 cond_resched();
1471         }
1472
1473         if (!cycled && !done) {
1474                 cycled = 1;
1475                 index = 0;
1476                 end = writeback_index - 1;
1477                 goto retry;
1478         }
1479         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1480                 mapping->writeback_index = done_index;
1481
1482         if (nwritten)
1483                 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1484                                                         NULL, 0, DATA, WRITE);
1485
1486         return ret;
1487 }
1488
1489 static int f2fs_write_data_pages(struct address_space *mapping,
1490                             struct writeback_control *wbc)
1491 {
1492         struct inode *inode = mapping->host;
1493         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1494         struct blk_plug plug;
1495         int ret;
1496
1497         /* deal with chardevs and other special file */
1498         if (!mapping->a_ops->writepage)
1499                 return 0;
1500
1501         /* skip writing if there is no dirty page in this inode */
1502         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1503                 return 0;
1504
1505         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1506                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1507                         available_free_memory(sbi, DIRTY_DENTS))
1508                 goto skip_write;
1509
1510         /* skip writing during file defragment */
1511         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1512                 goto skip_write;
1513
1514         /* during POR, we don't need to trigger writepage at all. */
1515         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1516                 goto skip_write;
1517
1518         trace_f2fs_writepages(mapping->host, wbc, DATA);
1519
1520         blk_start_plug(&plug);
1521         ret = f2fs_write_cache_pages(mapping, wbc);
1522         blk_finish_plug(&plug);
1523         /*
1524          * if some pages were truncated, we cannot guarantee its mapping->host
1525          * to detect pending bios.
1526          */
1527
1528         remove_dirty_inode(inode);
1529         return ret;
1530
1531 skip_write:
1532         wbc->pages_skipped += get_dirty_pages(inode);
1533         trace_f2fs_writepages(mapping->host, wbc, DATA);
1534         return 0;
1535 }
1536
1537 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1538 {
1539         struct inode *inode = mapping->host;
1540         loff_t i_size = i_size_read(inode);
1541
1542         if (to > i_size) {
1543                 truncate_pagecache(inode, i_size);
1544                 truncate_blocks(inode, i_size, true);
1545         }
1546 }
1547
1548 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1549                         struct page *page, loff_t pos, unsigned len,
1550                         block_t *blk_addr, bool *node_changed)
1551 {
1552         struct inode *inode = page->mapping->host;
1553         pgoff_t index = page->index;
1554         struct dnode_of_data dn;
1555         struct page *ipage;
1556         bool locked = false;
1557         struct extent_info ei;
1558         int err = 0;
1559
1560         /*
1561          * we already allocated all the blocks, so we don't need to get
1562          * the block addresses when there is no need to fill the page.
1563          */
1564         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE)
1565                 return 0;
1566
1567         if (f2fs_has_inline_data(inode) ||
1568                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1569                 f2fs_lock_op(sbi);
1570                 locked = true;
1571         }
1572 restart:
1573         /* check inline_data */
1574         ipage = get_node_page(sbi, inode->i_ino);
1575         if (IS_ERR(ipage)) {
1576                 err = PTR_ERR(ipage);
1577                 goto unlock_out;
1578         }
1579
1580         set_new_dnode(&dn, inode, ipage, ipage, 0);
1581
1582         if (f2fs_has_inline_data(inode)) {
1583                 if (pos + len <= MAX_INLINE_DATA) {
1584                         read_inline_data(page, ipage);
1585                         set_inode_flag(inode, FI_DATA_EXIST);
1586                         if (inode->i_nlink)
1587                                 set_inline_node(ipage);
1588                 } else {
1589                         err = f2fs_convert_inline_page(&dn, page);
1590                         if (err)
1591                                 goto out;
1592                         if (dn.data_blkaddr == NULL_ADDR)
1593                                 err = f2fs_get_block(&dn, index);
1594                 }
1595         } else if (locked) {
1596                 err = f2fs_get_block(&dn, index);
1597         } else {
1598                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1599                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1600                 } else {
1601                         /* hole case */
1602                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1603                         if (err || dn.data_blkaddr == NULL_ADDR) {
1604                                 f2fs_put_dnode(&dn);
1605                                 f2fs_lock_op(sbi);
1606                                 locked = true;
1607                                 goto restart;
1608                         }
1609                 }
1610         }
1611
1612         /* convert_inline_page can make node_changed */
1613         *blk_addr = dn.data_blkaddr;
1614         *node_changed = dn.node_changed;
1615 out:
1616         f2fs_put_dnode(&dn);
1617 unlock_out:
1618         if (locked)
1619                 f2fs_unlock_op(sbi);
1620         return err;
1621 }
1622
1623 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1624                 loff_t pos, unsigned len, unsigned flags,
1625                 struct page **pagep, void **fsdata)
1626 {
1627         struct inode *inode = mapping->host;
1628         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1629         struct page *page = NULL;
1630         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1631         bool need_balance = false;
1632         block_t blkaddr = NULL_ADDR;
1633         int err = 0;
1634
1635         trace_f2fs_write_begin(inode, pos, len, flags);
1636
1637         /*
1638          * We should check this at this moment to avoid deadlock on inode page
1639          * and #0 page. The locking rule for inline_data conversion should be:
1640          * lock_page(page #0) -> lock_page(inode_page)
1641          */
1642         if (index != 0) {
1643                 err = f2fs_convert_inline_inode(inode);
1644                 if (err)
1645                         goto fail;
1646         }
1647 repeat:
1648         /*
1649          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1650          * wait_for_stable_page. Will wait that below with our IO control.
1651          */
1652         page = pagecache_get_page(mapping, index,
1653                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1654         if (!page) {
1655                 err = -ENOMEM;
1656                 goto fail;
1657         }
1658
1659         *pagep = page;
1660
1661         err = prepare_write_begin(sbi, page, pos, len,
1662                                         &blkaddr, &need_balance);
1663         if (err)
1664                 goto fail;
1665
1666         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1667                 unlock_page(page);
1668                 f2fs_balance_fs(sbi, true);
1669                 lock_page(page);
1670                 if (page->mapping != mapping) {
1671                         /* The page got truncated from under us */
1672                         f2fs_put_page(page, 1);
1673                         goto repeat;
1674                 }
1675         }
1676
1677         f2fs_wait_on_page_writeback(page, DATA, false);
1678
1679         /* wait for GCed encrypted page writeback */
1680         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1681                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1682
1683         if (len == PAGE_SIZE || PageUptodate(page))
1684                 return 0;
1685
1686         if (blkaddr == NEW_ADDR) {
1687                 zero_user_segment(page, 0, PAGE_SIZE);
1688                 SetPageUptodate(page);
1689         } else {
1690                 struct bio *bio;
1691
1692                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1693                 if (IS_ERR(bio)) {
1694                         err = PTR_ERR(bio);
1695                         goto fail;
1696                 }
1697                 bio_set_op_attrs(bio, REQ_OP_READ, READ_SYNC);
1698                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1699                         bio_put(bio);
1700                         err = -EFAULT;
1701                         goto fail;
1702                 }
1703
1704                 __submit_bio(sbi, bio, DATA);
1705
1706                 lock_page(page);
1707                 if (unlikely(page->mapping != mapping)) {
1708                         f2fs_put_page(page, 1);
1709                         goto repeat;
1710                 }
1711                 if (unlikely(!PageUptodate(page))) {
1712                         err = -EIO;
1713                         goto fail;
1714                 }
1715         }
1716         return 0;
1717
1718 fail:
1719         f2fs_put_page(page, 1);
1720         f2fs_write_failed(mapping, pos + len);
1721         return err;
1722 }
1723
1724 static int f2fs_write_end(struct file *file,
1725                         struct address_space *mapping,
1726                         loff_t pos, unsigned len, unsigned copied,
1727                         struct page *page, void *fsdata)
1728 {
1729         struct inode *inode = page->mapping->host;
1730
1731         trace_f2fs_write_end(inode, pos, len, copied);
1732
1733         /*
1734          * This should be come from len == PAGE_SIZE, and we expect copied
1735          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1736          * let generic_perform_write() try to copy data again through copied=0.
1737          */
1738         if (!PageUptodate(page)) {
1739                 if (unlikely(copied != PAGE_SIZE))
1740                         copied = 0;
1741                 else
1742                         SetPageUptodate(page);
1743         }
1744         if (!copied)
1745                 goto unlock_out;
1746
1747         set_page_dirty(page);
1748         clear_cold_data(page);
1749
1750         if (pos + copied > i_size_read(inode))
1751                 f2fs_i_size_write(inode, pos + copied);
1752 unlock_out:
1753         f2fs_put_page(page, 1);
1754         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1755         return copied;
1756 }
1757
1758 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1759                            loff_t offset)
1760 {
1761         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1762
1763         if (offset & blocksize_mask)
1764                 return -EINVAL;
1765
1766         if (iov_iter_alignment(iter) & blocksize_mask)
1767                 return -EINVAL;
1768
1769         return 0;
1770 }
1771
1772 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1773 {
1774         struct address_space *mapping = iocb->ki_filp->f_mapping;
1775         struct inode *inode = mapping->host;
1776         size_t count = iov_iter_count(iter);
1777         loff_t offset = iocb->ki_pos;
1778         int rw = iov_iter_rw(iter);
1779         int err;
1780
1781         err = check_direct_IO(inode, iter, offset);
1782         if (err)
1783                 return err;
1784
1785         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1786                 return 0;
1787         if (test_opt(F2FS_I_SB(inode), LFS))
1788                 return 0;
1789
1790         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1791
1792         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1793         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1794         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1795
1796         if (rw == WRITE) {
1797                 if (err > 0)
1798                         set_inode_flag(inode, FI_UPDATE_WRITE);
1799                 else if (err < 0)
1800                         f2fs_write_failed(mapping, offset + count);
1801         }
1802
1803         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1804
1805         return err;
1806 }
1807
1808 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1809                                                         unsigned int length)
1810 {
1811         struct inode *inode = page->mapping->host;
1812         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1813
1814         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1815                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1816                 return;
1817
1818         if (PageDirty(page)) {
1819                 if (inode->i_ino == F2FS_META_INO(sbi))
1820                         dec_page_count(sbi, F2FS_DIRTY_META);
1821                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1822                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1823                 else
1824                         inode_dec_dirty_pages(inode);
1825         }
1826
1827         clear_cold_data(page);
1828
1829         /* This is atomic written page, keep Private */
1830         if (IS_ATOMIC_WRITTEN_PAGE(page))
1831                 return;
1832
1833         set_page_private(page, 0);
1834         ClearPagePrivate(page);
1835 }
1836
1837 int f2fs_release_page(struct page *page, gfp_t wait)
1838 {
1839         /* If this is dirty page, keep PagePrivate */
1840         if (PageDirty(page))
1841                 return 0;
1842
1843         /* This is atomic written page, keep Private */
1844         if (IS_ATOMIC_WRITTEN_PAGE(page))
1845                 return 0;
1846
1847         clear_cold_data(page);
1848         set_page_private(page, 0);
1849         ClearPagePrivate(page);
1850         return 1;
1851 }
1852
1853 /*
1854  * This was copied from __set_page_dirty_buffers which gives higher performance
1855  * in very high speed storages. (e.g., pmem)
1856  */
1857 void f2fs_set_page_dirty_nobuffers(struct page *page)
1858 {
1859         struct address_space *mapping = page->mapping;
1860         unsigned long flags;
1861
1862         if (unlikely(!mapping))
1863                 return;
1864
1865         spin_lock(&mapping->private_lock);
1866         lock_page_memcg(page);
1867         SetPageDirty(page);
1868         spin_unlock(&mapping->private_lock);
1869
1870         spin_lock_irqsave(&mapping->tree_lock, flags);
1871         WARN_ON_ONCE(!PageUptodate(page));
1872         account_page_dirtied(page, mapping);
1873         radix_tree_tag_set(&mapping->page_tree,
1874                         page_index(page), PAGECACHE_TAG_DIRTY);
1875         spin_unlock_irqrestore(&mapping->tree_lock, flags);
1876         unlock_page_memcg(page);
1877
1878         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1879         return;
1880 }
1881
1882 static int f2fs_set_data_page_dirty(struct page *page)
1883 {
1884         struct address_space *mapping = page->mapping;
1885         struct inode *inode = mapping->host;
1886
1887         trace_f2fs_set_page_dirty(page, DATA);
1888
1889         if (!PageUptodate(page))
1890                 SetPageUptodate(page);
1891
1892         if (f2fs_is_atomic_file(inode)) {
1893                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1894                         register_inmem_page(inode, page);
1895                         return 1;
1896                 }
1897                 /*
1898                  * Previously, this page has been registered, we just
1899                  * return here.
1900                  */
1901                 return 0;
1902         }
1903
1904         if (!PageDirty(page)) {
1905                 f2fs_set_page_dirty_nobuffers(page);
1906                 update_dirty_page(inode, page);
1907                 return 1;
1908         }
1909         return 0;
1910 }
1911
1912 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1913 {
1914         struct inode *inode = mapping->host;
1915
1916         if (f2fs_has_inline_data(inode))
1917                 return 0;
1918
1919         /* make sure allocating whole blocks */
1920         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1921                 filemap_write_and_wait(mapping);
1922
1923         return generic_block_bmap(mapping, block, get_data_block_bmap);
1924 }
1925
1926 #ifdef CONFIG_MIGRATION
1927 #include <linux/migrate.h>
1928
1929 int f2fs_migrate_page(struct address_space *mapping,
1930                 struct page *newpage, struct page *page, enum migrate_mode mode)
1931 {
1932         int rc, extra_count;
1933         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
1934         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
1935
1936         BUG_ON(PageWriteback(page));
1937
1938         /* migrating an atomic written page is safe with the inmem_lock hold */
1939         if (atomic_written && !mutex_trylock(&fi->inmem_lock))
1940                 return -EAGAIN;
1941
1942         /*
1943          * A reference is expected if PagePrivate set when move mapping,
1944          * however F2FS breaks this for maintaining dirty page counts when
1945          * truncating pages. So here adjusting the 'extra_count' make it work.
1946          */
1947         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
1948         rc = migrate_page_move_mapping(mapping, newpage,
1949                                 page, NULL, mode, extra_count);
1950         if (rc != MIGRATEPAGE_SUCCESS) {
1951                 if (atomic_written)
1952                         mutex_unlock(&fi->inmem_lock);
1953                 return rc;
1954         }
1955
1956         if (atomic_written) {
1957                 struct inmem_pages *cur;
1958                 list_for_each_entry(cur, &fi->inmem_pages, list)
1959                         if (cur->page == page) {
1960                                 cur->page = newpage;
1961                                 break;
1962                         }
1963                 mutex_unlock(&fi->inmem_lock);
1964                 put_page(page);
1965                 get_page(newpage);
1966         }
1967
1968         if (PagePrivate(page))
1969                 SetPagePrivate(newpage);
1970         set_page_private(newpage, page_private(page));
1971
1972         migrate_page_copy(newpage, page);
1973
1974         return MIGRATEPAGE_SUCCESS;
1975 }
1976 #endif
1977
1978 const struct address_space_operations f2fs_dblock_aops = {
1979         .readpage       = f2fs_read_data_page,
1980         .readpages      = f2fs_read_data_pages,
1981         .writepage      = f2fs_write_data_page,
1982         .writepages     = f2fs_write_data_pages,
1983         .write_begin    = f2fs_write_begin,
1984         .write_end      = f2fs_write_end,
1985         .set_page_dirty = f2fs_set_data_page_dirty,
1986         .invalidatepage = f2fs_invalidate_page,
1987         .releasepage    = f2fs_release_page,
1988         .direct_IO      = f2fs_direct_IO,
1989         .bmap           = f2fs_bmap,
1990 #ifdef CONFIG_MIGRATION
1991         .migratepage    = f2fs_migrate_page,
1992 #endif
1993 };