projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: avoid punch_hole overhead when releasing volatile data
[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / fs / f2fs / file.c
1 /*
2 * fs/f2fs/file.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/stat.h>
14 #include <linux/buffer_head.h>
15 #include <linux/writeback.h>
16 #include <linux/blkdev.h>
17 #include <linux/falloc.h>
18 #include <linux/types.h>
19 #include <linux/compat.h>
20 #include <linux/uaccess.h>
21 #include <linux/mount.h>
22 #include <linux/pagevec.h>
23
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "xattr.h"
28 #include "acl.h"
29 #include "trace.h"
30 #include <trace/events/f2fs.h>
31
32 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma,
33 struct vm_fault *vmf)
34 {
35 struct page *page = vmf->page;
36 struct inode *inode = file_inode(vma->vm_file);
37 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
38 struct dnode_of_data dn;
39 int err;
40
41 f2fs_balance_fs(sbi);
42
43 sb_start_pagefault(inode->i_sb);
44
45 f2fs_bug_on(sbi, f2fs_has_inline_data(inode));
46
47 /* block allocation */
48 f2fs_lock_op(sbi);
49 set_new_dnode(&dn, inode, NULL, NULL, 0);
50 err = f2fs_reserve_block(&dn, page->index);
51 if (err) {
52 f2fs_unlock_op(sbi);
53 goto out;
54 }
55 f2fs_put_dnode(&dn);
56 f2fs_unlock_op(sbi);
57
58 file_update_time(vma->vm_file);
59 lock_page(page);
60 if (unlikely(page->mapping != inode->i_mapping ||
61 page_offset(page) > i_size_read(inode) ||
62 !PageUptodate(page))) {
63 unlock_page(page);
64 err = -EFAULT;
65 goto out;
66 }
67
68 /*
69 * check to see if the page is mapped already (no holes)
70 */
71 if (PageMappedToDisk(page))
72 goto mapped;
73
74 /* page is wholly or partially inside EOF */
75 if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) {
76 unsigned offset;
77 offset = i_size_read(inode) & ~PAGE_CACHE_MASK;
78 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
79 }
80 set_page_dirty(page);
81 SetPageUptodate(page);
82
83 trace_f2fs_vm_page_mkwrite(page, DATA);
84 mapped:
85 /* fill the page */
86 f2fs_wait_on_page_writeback(page, DATA);
87 out:
88 sb_end_pagefault(inode->i_sb);
89 return block_page_mkwrite_return(err);
90 }
91
92 static const struct vm_operations_struct f2fs_file_vm_ops = {
93 .fault = filemap_fault,
94 .map_pages = filemap_map_pages,
95 .page_mkwrite = f2fs_vm_page_mkwrite,
96 };
97
98 static int get_parent_ino(struct inode *inode, nid_t *pino)
99 {
100 struct dentry *dentry;
101
102 inode = igrab(inode);
103 dentry = d_find_any_alias(inode);
104 iput(inode);
105 if (!dentry)
106 return 0;
107
108 if (update_dent_inode(inode, &dentry->d_name)) {
109 dput(dentry);
110 return 0;
111 }
112
113 *pino = parent_ino(dentry);
114 dput(dentry);
115 return 1;
116 }
117
118 static inline bool need_do_checkpoint(struct inode *inode)
119 {
120 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
121 bool need_cp = false;
122
123 if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1)
124 need_cp = true;
125 else if (file_wrong_pino(inode))
126 need_cp = true;
127 else if (!space_for_roll_forward(sbi))
128 need_cp = true;
129 else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino))
130 need_cp = true;
131 else if (F2FS_I(inode)->xattr_ver == cur_cp_version(F2FS_CKPT(sbi)))
132 need_cp = true;
133 else if (test_opt(sbi, FASTBOOT))
134 need_cp = true;
135 else if (sbi->active_logs == 2)
136 need_cp = true;
137
138 return need_cp;
139 }
140
141 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino)
142 {
143 struct page *i = find_get_page(NODE_MAPPING(sbi), ino);
144 bool ret = false;
145 /* But we need to avoid that there are some inode updates */
146 if ((i && PageDirty(i)) || need_inode_block_update(sbi, ino))
147 ret = true;
148 f2fs_put_page(i, 0);
149 return ret;
150 }
151
152 static void try_to_fix_pino(struct inode *inode)
153 {
154 struct f2fs_inode_info *fi = F2FS_I(inode);
155 nid_t pino;
156
157 down_write(&fi->i_sem);
158 fi->xattr_ver = 0;
159 if (file_wrong_pino(inode) && inode->i_nlink == 1 &&
160 get_parent_ino(inode, &pino)) {
161 fi->i_pino = pino;
162 file_got_pino(inode);
163 up_write(&fi->i_sem);
164
165 mark_inode_dirty_sync(inode);
166 f2fs_write_inode(inode, NULL);
167 } else {
168 up_write(&fi->i_sem);
169 }
170 }
171
172 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
173 {
174 struct inode *inode = file->f_mapping->host;
175 struct f2fs_inode_info *fi = F2FS_I(inode);
176 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
177 nid_t ino = inode->i_ino;
178 int ret = 0;
179 bool need_cp = false;
180 struct writeback_control wbc = {
181 .sync_mode = WB_SYNC_ALL,
182 .nr_to_write = LONG_MAX,
183 .for_reclaim = 0,
184 };
185
186 if (unlikely(f2fs_readonly(inode->i_sb)))
187 return 0;
188
189 trace_f2fs_sync_file_enter(inode);
190
191 /* if fdatasync is triggered, let's do in-place-update */
192 if (get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
193 set_inode_flag(fi, FI_NEED_IPU);
194 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
195 clear_inode_flag(fi, FI_NEED_IPU);
196
197 if (ret) {
198 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
199 return ret;
200 }
201
202 /* if the inode is dirty, let's recover all the time */
203 if (!datasync && is_inode_flag_set(fi, FI_DIRTY_INODE)) {
204 update_inode_page(inode);
205 goto go_write;
206 }
207
208 /*
209 * if there is no written data, don't waste time to write recovery info.
210 */
211 if (!is_inode_flag_set(fi, FI_APPEND_WRITE) &&
212 !exist_written_data(sbi, ino, APPEND_INO)) {
213
214 /* it may call write_inode just prior to fsync */
215 if (need_inode_page_update(sbi, ino))
216 goto go_write;
217
218 if (is_inode_flag_set(fi, FI_UPDATE_WRITE) ||
219 exist_written_data(sbi, ino, UPDATE_INO))
220 goto flush_out;
221 goto out;
222 }
223 go_write:
224 /* guarantee free sections for fsync */
225 f2fs_balance_fs(sbi);
226
227 /*
228 * Both of fdatasync() and fsync() are able to be recovered from
229 * sudden-power-off.
230 */
231 down_read(&fi->i_sem);
232 need_cp = need_do_checkpoint(inode);
233 up_read(&fi->i_sem);
234
235 if (need_cp) {
236 /* all the dirty node pages should be flushed for POR */
237 ret = f2fs_sync_fs(inode->i_sb, 1);
238
239 /*
240 * We've secured consistency through sync_fs. Following pino
241 * will be used only for fsynced inodes after checkpoint.
242 */
243 try_to_fix_pino(inode);
244 clear_inode_flag(fi, FI_APPEND_WRITE);
245 clear_inode_flag(fi, FI_UPDATE_WRITE);
246 goto out;
247 }
248 sync_nodes:
249 sync_node_pages(sbi, ino, &wbc);
250
251 /* if cp_error was enabled, we should avoid infinite loop */
252 if (unlikely(f2fs_cp_error(sbi)))
253 goto out;
254
255 if (need_inode_block_update(sbi, ino)) {
256 mark_inode_dirty_sync(inode);
257 f2fs_write_inode(inode, NULL);
258 goto sync_nodes;
259 }
260
261 ret = wait_on_node_pages_writeback(sbi, ino);
262 if (ret)
263 goto out;
264
265 /* once recovery info is written, don't need to tack this */
266 remove_dirty_inode(sbi, ino, APPEND_INO);
267 clear_inode_flag(fi, FI_APPEND_WRITE);
268 flush_out:
269 remove_dirty_inode(sbi, ino, UPDATE_INO);
270 clear_inode_flag(fi, FI_UPDATE_WRITE);
271 ret = f2fs_issue_flush(sbi);
272 out:
273 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
274 f2fs_trace_ios(NULL, NULL, 1);
275 return ret;
276 }
277
278 static pgoff_t __get_first_dirty_index(struct address_space *mapping,
279 pgoff_t pgofs, int whence)
280 {
281 struct pagevec pvec;
282 int nr_pages;
283
284 if (whence != SEEK_DATA)
285 return 0;
286
287 /* find first dirty page index */
288 pagevec_init(&pvec, 0);
289 nr_pages = pagevec_lookup_tag(&pvec, mapping, &pgofs,
290 PAGECACHE_TAG_DIRTY, 1);
291 pgofs = nr_pages ? pvec.pages[0]->index : LONG_MAX;
292 pagevec_release(&pvec);
293 return pgofs;
294 }
295
296 static bool __found_offset(block_t blkaddr, pgoff_t dirty, pgoff_t pgofs,
297 int whence)
298 {
299 switch (whence) {
300 case SEEK_DATA:
301 if ((blkaddr == NEW_ADDR && dirty == pgofs) ||
302 (blkaddr != NEW_ADDR && blkaddr != NULL_ADDR))
303 return true;
304 break;
305 case SEEK_HOLE:
306 if (blkaddr == NULL_ADDR)
307 return true;
308 break;
309 }
310 return false;
311 }
312
313 static loff_t f2fs_seek_block(struct file *file, loff_t offset, int whence)
314 {
315 struct inode *inode = file->f_mapping->host;
316 loff_t maxbytes = inode->i_sb->s_maxbytes;
317 struct dnode_of_data dn;
318 pgoff_t pgofs, end_offset, dirty;
319 loff_t data_ofs = offset;
320 loff_t isize;
321 int err = 0;
322
323 mutex_lock(&inode->i_mutex);
324
325 isize = i_size_read(inode);
326 if (offset >= isize)
327 goto fail;
328
329 /* handle inline data case */
330 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
331 if (whence == SEEK_HOLE)
332 data_ofs = isize;
333 goto found;
334 }
335
336 pgofs = (pgoff_t)(offset >> PAGE_CACHE_SHIFT);
337
338 dirty = __get_first_dirty_index(inode->i_mapping, pgofs, whence);
339
340 for (; data_ofs < isize; data_ofs = pgofs << PAGE_CACHE_SHIFT) {
341 set_new_dnode(&dn, inode, NULL, NULL, 0);
342 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
343 if (err && err != -ENOENT) {
344 goto fail;
345 } else if (err == -ENOENT) {
346 /* direct node does not exists */
347 if (whence == SEEK_DATA) {
348 pgofs = PGOFS_OF_NEXT_DNODE(pgofs,
349 F2FS_I(inode));
350 continue;
351 } else {
352 goto found;
353 }
354 }
355
356 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
357
358 /* find data/hole in dnode block */
359 for (; dn.ofs_in_node < end_offset;
360 dn.ofs_in_node++, pgofs++,
361 data_ofs = (loff_t)pgofs << PAGE_CACHE_SHIFT) {
362 block_t blkaddr;
363 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
364
365 if (__found_offset(blkaddr, dirty, pgofs, whence)) {
366 f2fs_put_dnode(&dn);
367 goto found;
368 }
369 }
370 f2fs_put_dnode(&dn);
371 }
372
373 if (whence == SEEK_DATA)
374 goto fail;
375 found:
376 if (whence == SEEK_HOLE && data_ofs > isize)
377 data_ofs = isize;
378 mutex_unlock(&inode->i_mutex);
379 return vfs_setpos(file, data_ofs, maxbytes);
380 fail:
381 mutex_unlock(&inode->i_mutex);
382 return -ENXIO;
383 }
384
385 static loff_t f2fs_llseek(struct file *file, loff_t offset, int whence)
386 {
387 struct inode *inode = file->f_mapping->host;
388 loff_t maxbytes = inode->i_sb->s_maxbytes;
389
390 switch (whence) {
391 case SEEK_SET:
392 case SEEK_CUR:
393 case SEEK_END:
394 return generic_file_llseek_size(file, offset, whence,
395 maxbytes, i_size_read(inode));
396 case SEEK_DATA:
397 case SEEK_HOLE:
398 if (offset < 0)
399 return -ENXIO;
400 return f2fs_seek_block(file, offset, whence);
401 }
402
403 return -EINVAL;
404 }
405
406 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma)
407 {
408 struct inode *inode = file_inode(file);
409
410 /* we don't need to use inline_data strictly */
411 if (f2fs_has_inline_data(inode)) {
412 int err = f2fs_convert_inline_inode(inode);
413 if (err)
414 return err;
415 }
416
417 file_accessed(file);
418 vma->vm_ops = &f2fs_file_vm_ops;
419 return 0;
420 }
421
422 int truncate_data_blocks_range(struct dnode_of_data *dn, int count)
423 {
424 int nr_free = 0, ofs = dn->ofs_in_node;
425 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
426 struct f2fs_node *raw_node;
427 __le32 *addr;
428
429 raw_node = F2FS_NODE(dn->node_page);
430 addr = blkaddr_in_node(raw_node) + ofs;
431
432 for (; count > 0; count--, addr++, dn->ofs_in_node++) {
433 block_t blkaddr = le32_to_cpu(*addr);
434 if (blkaddr == NULL_ADDR)
435 continue;
436
437 dn->data_blkaddr = NULL_ADDR;
438 f2fs_update_extent_cache(dn);
439 invalidate_blocks(sbi, blkaddr);
440 if (dn->ofs_in_node == 0 && IS_INODE(dn->node_page))
441 clear_inode_flag(F2FS_I(dn->inode),
442 FI_FIRST_BLOCK_WRITTEN);
443 nr_free++;
444 }
445 if (nr_free) {
446 dec_valid_block_count(sbi, dn->inode, nr_free);
447 set_page_dirty(dn->node_page);
448 sync_inode_page(dn);
449 }
450 dn->ofs_in_node = ofs;
451
452 trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid,
453 dn->ofs_in_node, nr_free);
454 return nr_free;
455 }
456
457 void truncate_data_blocks(struct dnode_of_data *dn)
458 {
459 truncate_data_blocks_range(dn, ADDRS_PER_BLOCK);
460 }
461
462 static int truncate_partial_data_page(struct inode *inode, u64 from,
463 bool force)
464 {
465 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
466 struct page *page;
467
468 if (!offset && !force)
469 return 0;
470
471 page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, force);
472 if (IS_ERR(page))
473 return 0;
474
475 lock_page(page);
476 if (unlikely(!PageUptodate(page) ||
477 page->mapping != inode->i_mapping))
478 goto out;
479
480 f2fs_wait_on_page_writeback(page, DATA);
481 zero_user(page, offset, PAGE_CACHE_SIZE - offset);
482 if (!force)
483 set_page_dirty(page);
484 out:
485 f2fs_put_page(page, 1);
486 return 0;
487 }
488
489 int truncate_blocks(struct inode *inode, u64 from, bool lock)
490 {
491 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
492 unsigned int blocksize = inode->i_sb->s_blocksize;
493 struct dnode_of_data dn;
494 pgoff_t free_from;
495 int count = 0, err = 0;
496 struct page *ipage;
497 bool truncate_page = false;
498
499 trace_f2fs_truncate_blocks_enter(inode, from);
500
501 free_from = (pgoff_t)F2FS_BYTES_TO_BLK(from + blocksize - 1);
502
503 if (lock)
504 f2fs_lock_op(sbi);
505
506 ipage = get_node_page(sbi, inode->i_ino);
507 if (IS_ERR(ipage)) {
508 err = PTR_ERR(ipage);
509 goto out;
510 }
511
512 if (f2fs_has_inline_data(inode)) {
513 if (truncate_inline_inode(ipage, from))
514 set_page_dirty(ipage);
515 f2fs_put_page(ipage, 1);
516 truncate_page = true;
517 goto out;
518 }
519
520 set_new_dnode(&dn, inode, ipage, NULL, 0);
521 err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE);
522 if (err) {
523 if (err == -ENOENT)
524 goto free_next;
525 goto out;
526 }
527
528 count = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
529
530 count -= dn.ofs_in_node;
531 f2fs_bug_on(sbi, count < 0);
532
533 if (dn.ofs_in_node || IS_INODE(dn.node_page)) {
534 truncate_data_blocks_range(&dn, count);
535 free_from += count;
536 }
537
538 f2fs_put_dnode(&dn);
539 free_next:
540 err = truncate_inode_blocks(inode, free_from);
541 out:
542 if (lock)
543 f2fs_unlock_op(sbi);
544
545 /* lastly zero out the first data page */
546 if (!err)
547 err = truncate_partial_data_page(inode, from, truncate_page);
548
549 trace_f2fs_truncate_blocks_exit(inode, err);
550 return err;
551 }
552
553 void f2fs_truncate(struct inode *inode)
554 {
555 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
556 S_ISLNK(inode->i_mode)))
557 return;
558
559 trace_f2fs_truncate(inode);
560
561 /* we should check inline_data size */
562 if (f2fs_has_inline_data(inode) && !f2fs_may_inline(inode)) {
563 if (f2fs_convert_inline_inode(inode))
564 return;
565 }
566
567 if (!truncate_blocks(inode, i_size_read(inode), true)) {
568 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
569 mark_inode_dirty(inode);
570 }
571 }
572
573 int f2fs_getattr(struct vfsmount *mnt,
574 struct dentry *dentry, struct kstat *stat)
575 {
576 struct inode *inode = dentry->d_inode;
577 generic_fillattr(inode, stat);
578 stat->blocks <<= 3;
579 return 0;
580 }
581
582 #ifdef CONFIG_F2FS_FS_POSIX_ACL
583 static void __setattr_copy(struct inode *inode, const struct iattr *attr)
584 {
585 struct f2fs_inode_info *fi = F2FS_I(inode);
586 unsigned int ia_valid = attr->ia_valid;
587
588 if (ia_valid & ATTR_UID)
589 inode->i_uid = attr->ia_uid;
590 if (ia_valid & ATTR_GID)
591 inode->i_gid = attr->ia_gid;
592 if (ia_valid & ATTR_ATIME)
593 inode->i_atime = timespec_trunc(attr->ia_atime,
594 inode->i_sb->s_time_gran);
595 if (ia_valid & ATTR_MTIME)
596 inode->i_mtime = timespec_trunc(attr->ia_mtime,
597 inode->i_sb->s_time_gran);
598 if (ia_valid & ATTR_CTIME)
599 inode->i_ctime = timespec_trunc(attr->ia_ctime,
600 inode->i_sb->s_time_gran);
601 if (ia_valid & ATTR_MODE) {
602 umode_t mode = attr->ia_mode;
603
604 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
605 mode &= ~S_ISGID;
606 set_acl_inode(fi, mode);
607 }
608 }
609 #else
610 #define __setattr_copy setattr_copy
611 #endif
612
613 int f2fs_setattr(struct dentry *dentry, struct iattr *attr)
614 {
615 struct inode *inode = dentry->d_inode;
616 struct f2fs_inode_info *fi = F2FS_I(inode);
617 int err;
618
619 err = inode_change_ok(inode, attr);
620 if (err)
621 return err;
622
623 if (attr->ia_valid & ATTR_SIZE) {
624 if (attr->ia_size != i_size_read(inode)) {
625 truncate_setsize(inode, attr->ia_size);
626 f2fs_truncate(inode);
627 f2fs_balance_fs(F2FS_I_SB(inode));
628 } else {
629 /*
630 * giving a chance to truncate blocks past EOF which
631 * are fallocated with FALLOC_FL_KEEP_SIZE.
632 */
633 f2fs_truncate(inode);
634 }
635 }
636
637 __setattr_copy(inode, attr);
638
639 if (attr->ia_valid & ATTR_MODE) {
640 err = posix_acl_chmod(inode, get_inode_mode(inode));
641 if (err || is_inode_flag_set(fi, FI_ACL_MODE)) {
642 inode->i_mode = fi->i_acl_mode;
643 clear_inode_flag(fi, FI_ACL_MODE);
644 }
645 }
646
647 mark_inode_dirty(inode);
648 return err;
649 }
650
651 const struct inode_operations f2fs_file_inode_operations = {
652 .getattr = f2fs_getattr,
653 .setattr = f2fs_setattr,
654 .get_acl = f2fs_get_acl,
655 .set_acl = f2fs_set_acl,
656 #ifdef CONFIG_F2FS_FS_XATTR
657 .setxattr = generic_setxattr,
658 .getxattr = generic_getxattr,
659 .listxattr = f2fs_listxattr,
660 .removexattr = generic_removexattr,
661 #endif
662 .fiemap = f2fs_fiemap,
663 };
664
665 static void fill_zero(struct inode *inode, pgoff_t index,
666 loff_t start, loff_t len)
667 {
668 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
669 struct page *page;
670
671 if (!len)
672 return;
673
674 f2fs_balance_fs(sbi);
675
676 f2fs_lock_op(sbi);
677 page = get_new_data_page(inode, NULL, index, false);
678 f2fs_unlock_op(sbi);
679
680 if (!IS_ERR(page)) {
681 f2fs_wait_on_page_writeback(page, DATA);
682 zero_user(page, start, len);
683 set_page_dirty(page);
684 f2fs_put_page(page, 1);
685 }
686 }
687
688 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end)
689 {
690 pgoff_t index;
691 int err;
692
693 for (index = pg_start; index < pg_end; index++) {
694 struct dnode_of_data dn;
695
696 set_new_dnode(&dn, inode, NULL, NULL, 0);
697 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
698 if (err) {
699 if (err == -ENOENT)
700 continue;
701 return err;
702 }
703
704 if (dn.data_blkaddr != NULL_ADDR)
705 truncate_data_blocks_range(&dn, 1);
706 f2fs_put_dnode(&dn);
707 }
708 return 0;
709 }
710
711 static int punch_hole(struct inode *inode, loff_t offset, loff_t len)
712 {
713 pgoff_t pg_start, pg_end;
714 loff_t off_start, off_end;
715 int ret = 0;
716
717 if (!S_ISREG(inode->i_mode))
718 return -EOPNOTSUPP;
719
720 /* skip punching hole beyond i_size */
721 if (offset >= inode->i_size)
722 return ret;
723
724 if (f2fs_has_inline_data(inode)) {
725 ret = f2fs_convert_inline_inode(inode);
726 if (ret)
727 return ret;
728 }
729
730 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
731 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
732
733 off_start = offset & (PAGE_CACHE_SIZE - 1);
734 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
735
736 if (pg_start == pg_end) {
737 fill_zero(inode, pg_start, off_start,
738 off_end - off_start);
739 } else {
740 if (off_start)
741 fill_zero(inode, pg_start++, off_start,
742 PAGE_CACHE_SIZE - off_start);
743 if (off_end)
744 fill_zero(inode, pg_end, 0, off_end);
745
746 if (pg_start < pg_end) {
747 struct address_space *mapping = inode->i_mapping;
748 loff_t blk_start, blk_end;
749 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
750
751 f2fs_balance_fs(sbi);
752
753 blk_start = pg_start << PAGE_CACHE_SHIFT;
754 blk_end = pg_end << PAGE_CACHE_SHIFT;
755 truncate_inode_pages_range(mapping, blk_start,
756 blk_end - 1);
757
758 f2fs_lock_op(sbi);
759 ret = truncate_hole(inode, pg_start, pg_end);
760 f2fs_unlock_op(sbi);
761 }
762 }
763
764 return ret;
765 }
766
767 static int expand_inode_data(struct inode *inode, loff_t offset,
768 loff_t len, int mode)
769 {
770 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
771 pgoff_t index, pg_start, pg_end;
772 loff_t new_size = i_size_read(inode);
773 loff_t off_start, off_end;
774 int ret = 0;
775
776 f2fs_balance_fs(sbi);
777
778 ret = inode_newsize_ok(inode, (len + offset));
779 if (ret)
780 return ret;
781
782 if (f2fs_has_inline_data(inode)) {
783 ret = f2fs_convert_inline_inode(inode);
784 if (ret)
785 return ret;
786 }
787
788 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
789 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
790
791 off_start = offset & (PAGE_CACHE_SIZE - 1);
792 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
793
794 f2fs_lock_op(sbi);
795
796 for (index = pg_start; index <= pg_end; index++) {
797 struct dnode_of_data dn;
798
799 if (index == pg_end && !off_end)
800 goto noalloc;
801
802 set_new_dnode(&dn, inode, NULL, NULL, 0);
803 ret = f2fs_reserve_block(&dn, index);
804 if (ret)
805 break;
806 noalloc:
807 if (pg_start == pg_end)
808 new_size = offset + len;
809 else if (index == pg_start && off_start)
810 new_size = (index + 1) << PAGE_CACHE_SHIFT;
811 else if (index == pg_end)
812 new_size = (index << PAGE_CACHE_SHIFT) + off_end;
813 else
814 new_size += PAGE_CACHE_SIZE;
815 }
816
817 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
818 i_size_read(inode) < new_size) {
819 i_size_write(inode, new_size);
820 mark_inode_dirty(inode);
821 update_inode_page(inode);
822 }
823 f2fs_unlock_op(sbi);
824
825 return ret;
826 }
827
828 static long f2fs_fallocate(struct file *file, int mode,
829 loff_t offset, loff_t len)
830 {
831 struct inode *inode = file_inode(file);
832 long ret;
833
834 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
835 return -EOPNOTSUPP;
836
837 mutex_lock(&inode->i_mutex);
838
839 if (mode & FALLOC_FL_PUNCH_HOLE)
840 ret = punch_hole(inode, offset, len);
841 else
842 ret = expand_inode_data(inode, offset, len, mode);
843
844 if (!ret) {
845 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
846 mark_inode_dirty(inode);
847 }
848
849 mutex_unlock(&inode->i_mutex);
850
851 trace_f2fs_fallocate(inode, mode, offset, len, ret);
852 return ret;
853 }
854
855 static int f2fs_release_file(struct inode *inode, struct file *filp)
856 {
857 /* some remained atomic pages should discarded */
858 if (f2fs_is_atomic_file(inode))
859 commit_inmem_pages(inode, true);
860 if (f2fs_is_volatile_file(inode)) {
861 set_inode_flag(F2FS_I(inode), FI_DROP_CACHE);
862 filemap_fdatawrite(inode->i_mapping);
863 clear_inode_flag(F2FS_I(inode), FI_DROP_CACHE);
864 }
865 return 0;
866 }
867
868 #define F2FS_REG_FLMASK (~(FS_DIRSYNC_FL | FS_TOPDIR_FL))
869 #define F2FS_OTHER_FLMASK (FS_NODUMP_FL | FS_NOATIME_FL)
870
871 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags)
872 {
873 if (S_ISDIR(mode))
874 return flags;
875 else if (S_ISREG(mode))
876 return flags & F2FS_REG_FLMASK;
877 else
878 return flags & F2FS_OTHER_FLMASK;
879 }
880
881 static int f2fs_ioc_getflags(struct file *filp, unsigned long arg)
882 {
883 struct inode *inode = file_inode(filp);
884 struct f2fs_inode_info *fi = F2FS_I(inode);
885 unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
886 return put_user(flags, (int __user *)arg);
887 }
888
889 static int f2fs_ioc_setflags(struct file *filp, unsigned long arg)
890 {
891 struct inode *inode = file_inode(filp);
892 struct f2fs_inode_info *fi = F2FS_I(inode);
893 unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
894 unsigned int oldflags;
895 int ret;
896
897 ret = mnt_want_write_file(filp);
898 if (ret)
899 return ret;
900
901 if (!inode_owner_or_capable(inode)) {
902 ret = -EACCES;
903 goto out;
904 }
905
906 if (get_user(flags, (int __user *)arg)) {
907 ret = -EFAULT;
908 goto out;
909 }
910
911 flags = f2fs_mask_flags(inode->i_mode, flags);
912
913 mutex_lock(&inode->i_mutex);
914
915 oldflags = fi->i_flags;
916
917 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
918 if (!capable(CAP_LINUX_IMMUTABLE)) {
919 mutex_unlock(&inode->i_mutex);
920 ret = -EPERM;
921 goto out;
922 }
923 }
924
925 flags = flags & FS_FL_USER_MODIFIABLE;
926 flags |= oldflags & ~FS_FL_USER_MODIFIABLE;
927 fi->i_flags = flags;
928 mutex_unlock(&inode->i_mutex);
929
930 f2fs_set_inode_flags(inode);
931 inode->i_ctime = CURRENT_TIME;
932 mark_inode_dirty(inode);
933 out:
934 mnt_drop_write_file(filp);
935 return ret;
936 }
937
938 static int f2fs_ioc_getversion(struct file *filp, unsigned long arg)
939 {
940 struct inode *inode = file_inode(filp);
941
942 return put_user(inode->i_generation, (int __user *)arg);
943 }
944
945 static int f2fs_ioc_start_atomic_write(struct file *filp)
946 {
947 struct inode *inode = file_inode(filp);
948
949 if (!inode_owner_or_capable(inode))
950 return -EACCES;
951
952 f2fs_balance_fs(F2FS_I_SB(inode));
953
954 if (f2fs_is_atomic_file(inode))
955 return 0;
956
957 set_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
958
959 return f2fs_convert_inline_inode(inode);
960 }
961
962 static int f2fs_ioc_commit_atomic_write(struct file *filp)
963 {
964 struct inode *inode = file_inode(filp);
965 int ret;
966
967 if (!inode_owner_or_capable(inode))
968 return -EACCES;
969
970 if (f2fs_is_volatile_file(inode))
971 return 0;
972
973 ret = mnt_want_write_file(filp);
974 if (ret)
975 return ret;
976
977 if (f2fs_is_atomic_file(inode))
978 commit_inmem_pages(inode, false);
979
980 ret = f2fs_sync_file(filp, 0, LONG_MAX, 0);
981 mnt_drop_write_file(filp);
982 clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
983 return ret;
984 }
985
986 static int f2fs_ioc_start_volatile_write(struct file *filp)
987 {
988 struct inode *inode = file_inode(filp);
989
990 if (!inode_owner_or_capable(inode))
991 return -EACCES;
992
993 if (f2fs_is_volatile_file(inode))
994 return 0;
995
996 set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
997
998 return f2fs_convert_inline_inode(inode);
999 }
1000
1001 static int f2fs_ioc_release_volatile_write(struct file *filp)
1002 {
1003 struct inode *inode = file_inode(filp);
1004
1005 if (!inode_owner_or_capable(inode))
1006 return -EACCES;
1007
1008 if (!f2fs_is_volatile_file(inode))
1009 return 0;
1010
1011 if (!f2fs_is_first_block_written(inode))
1012 return truncate_partial_data_page(inode, 0, true);
1013
1014 punch_hole(inode, 0, F2FS_BLKSIZE);
1015 return 0;
1016 }
1017
1018 static int f2fs_ioc_abort_volatile_write(struct file *filp)
1019 {
1020 struct inode *inode = file_inode(filp);
1021 int ret;
1022
1023 if (!inode_owner_or_capable(inode))
1024 return -EACCES;
1025
1026 ret = mnt_want_write_file(filp);
1027 if (ret)
1028 return ret;
1029
1030 f2fs_balance_fs(F2FS_I_SB(inode));
1031
1032 if (f2fs_is_atomic_file(inode)) {
1033 commit_inmem_pages(inode, false);
1034 clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
1035 }
1036
1037 if (f2fs_is_volatile_file(inode)) {
1038 clear_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
1039 filemap_fdatawrite(inode->i_mapping);
1040 set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
1041 }
1042 mnt_drop_write_file(filp);
1043 return ret;
1044 }
1045
1046 static int f2fs_ioc_shutdown(struct file *filp, unsigned long arg)
1047 {
1048 struct inode *inode = file_inode(filp);
1049 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1050 struct super_block *sb = sbi->sb;
1051 __u32 in;
1052
1053 if (!capable(CAP_SYS_ADMIN))
1054 return -EPERM;
1055
1056 if (get_user(in, (__u32 __user *)arg))
1057 return -EFAULT;
1058
1059 switch (in) {
1060 case F2FS_GOING_DOWN_FULLSYNC:
1061 sb = freeze_bdev(sb->s_bdev);
1062 if (sb && !IS_ERR(sb)) {
1063 f2fs_stop_checkpoint(sbi);
1064 thaw_bdev(sb->s_bdev, sb);
1065 }
1066 break;
1067 case F2FS_GOING_DOWN_METASYNC:
1068 /* do checkpoint only */
1069 f2fs_sync_fs(sb, 1);
1070 f2fs_stop_checkpoint(sbi);
1071 break;
1072 case F2FS_GOING_DOWN_NOSYNC:
1073 f2fs_stop_checkpoint(sbi);
1074 break;
1075 default:
1076 return -EINVAL;
1077 }
1078 return 0;
1079 }
1080
1081 static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg)
1082 {
1083 struct inode *inode = file_inode(filp);
1084 struct super_block *sb = inode->i_sb;
1085 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1086 struct fstrim_range range;
1087 int ret;
1088
1089 if (!capable(CAP_SYS_ADMIN))
1090 return -EPERM;
1091
1092 if (!blk_queue_discard(q))
1093 return -EOPNOTSUPP;
1094
1095 if (copy_from_user(&range, (struct fstrim_range __user *)arg,
1096 sizeof(range)))
1097 return -EFAULT;
1098
1099 range.minlen = max((unsigned int)range.minlen,
1100 q->limits.discard_granularity);
1101 ret = f2fs_trim_fs(F2FS_SB(sb), &range);
1102 if (ret < 0)
1103 return ret;
1104
1105 if (copy_to_user((struct fstrim_range __user *)arg, &range,
1106 sizeof(range)))
1107 return -EFAULT;
1108 return 0;
1109 }
1110
1111 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1112 {
1113 switch (cmd) {
1114 case F2FS_IOC_GETFLAGS:
1115 return f2fs_ioc_getflags(filp, arg);
1116 case F2FS_IOC_SETFLAGS:
1117 return f2fs_ioc_setflags(filp, arg);
1118 case F2FS_IOC_GETVERSION:
1119 return f2fs_ioc_getversion(filp, arg);
1120 case F2FS_IOC_START_ATOMIC_WRITE:
1121 return f2fs_ioc_start_atomic_write(filp);
1122 case F2FS_IOC_COMMIT_ATOMIC_WRITE:
1123 return f2fs_ioc_commit_atomic_write(filp);
1124 case F2FS_IOC_START_VOLATILE_WRITE:
1125 return f2fs_ioc_start_volatile_write(filp);
1126 case F2FS_IOC_RELEASE_VOLATILE_WRITE:
1127 return f2fs_ioc_release_volatile_write(filp);
1128 case F2FS_IOC_ABORT_VOLATILE_WRITE:
1129 return f2fs_ioc_abort_volatile_write(filp);
1130 case F2FS_IOC_SHUTDOWN:
1131 return f2fs_ioc_shutdown(filp, arg);
1132 case FITRIM:
1133 return f2fs_ioc_fitrim(filp, arg);
1134 default:
1135 return -ENOTTY;
1136 }
1137 }
1138
1139 #ifdef CONFIG_COMPAT
1140 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1141 {
1142 switch (cmd) {
1143 case F2FS_IOC32_GETFLAGS:
1144 cmd = F2FS_IOC_GETFLAGS;
1145 break;
1146 case F2FS_IOC32_SETFLAGS:
1147 cmd = F2FS_IOC_SETFLAGS;
1148 break;
1149 default:
1150 return -ENOIOCTLCMD;
1151 }
1152 return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
1153 }
1154 #endif
1155
1156 const struct file_operations f2fs_file_operations = {
1157 .llseek = f2fs_llseek,
1158 .read = new_sync_read,
1159 .write = new_sync_write,
1160 .read_iter = generic_file_read_iter,
1161 .write_iter = generic_file_write_iter,
1162 .open = generic_file_open,
1163 .release = f2fs_release_file,
1164 .mmap = f2fs_file_mmap,
1165 .fsync = f2fs_sync_file,
1166 .fallocate = f2fs_fallocate,
1167 .unlocked_ioctl = f2fs_ioctl,
1168 #ifdef CONFIG_COMPAT
1169 .compat_ioctl = f2fs_compat_ioctl,
1170 #endif
1171 .splice_read = generic_file_splice_read,
1172 .splice_write = iter_file_splice_write,
1173 };