Loading...
1/*
2 * linux/fs/ext4/file.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/file.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * ext4 fs regular file handling primitives
16 *
17 * 64-bit file support on 64-bit platforms by Jakub Jelinek
18 * (jj@sunsite.ms.mff.cuni.cz)
19 */
20
21#include <linux/time.h>
22#include <linux/fs.h>
23#include <linux/mount.h>
24#include <linux/path.h>
25#include <linux/dax.h>
26#include <linux/quotaops.h>
27#include <linux/pagevec.h>
28#include <linux/uio.h>
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34/*
35 * Called when an inode is released. Note that this is different
36 * from ext4_file_open: open gets called at every open, but release
37 * gets called only when /all/ the files are closed.
38 */
39static int ext4_release_file(struct inode *inode, struct file *filp)
40{
41 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
42 ext4_alloc_da_blocks(inode);
43 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
44 }
45 /* if we are the last writer on the inode, drop the block reservation */
46 if ((filp->f_mode & FMODE_WRITE) &&
47 (atomic_read(&inode->i_writecount) == 1) &&
48 !EXT4_I(inode)->i_reserved_data_blocks)
49 {
50 down_write(&EXT4_I(inode)->i_data_sem);
51 ext4_discard_preallocations(inode);
52 up_write(&EXT4_I(inode)->i_data_sem);
53 }
54 if (is_dx(inode) && filp->private_data)
55 ext4_htree_free_dir_info(filp->private_data);
56
57 return 0;
58}
59
60static void ext4_unwritten_wait(struct inode *inode)
61{
62 wait_queue_head_t *wq = ext4_ioend_wq(inode);
63
64 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
65}
66
67/*
68 * This tests whether the IO in question is block-aligned or not.
69 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
70 * are converted to written only after the IO is complete. Until they are
71 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
72 * it needs to zero out portions of the start and/or end block. If 2 AIO
73 * threads are at work on the same unwritten block, they must be synchronized
74 * or one thread will zero the other's data, causing corruption.
75 */
76static int
77ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
78{
79 struct super_block *sb = inode->i_sb;
80 int blockmask = sb->s_blocksize - 1;
81
82 if (pos >= i_size_read(inode))
83 return 0;
84
85 if ((pos | iov_iter_alignment(from)) & blockmask)
86 return 1;
87
88 return 0;
89}
90
91static ssize_t
92ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
93{
94 struct file *file = iocb->ki_filp;
95 struct inode *inode = file_inode(iocb->ki_filp);
96 struct blk_plug plug;
97 int o_direct = iocb->ki_flags & IOCB_DIRECT;
98 int unaligned_aio = 0;
99 int overwrite = 0;
100 ssize_t ret;
101
102 inode_lock(inode);
103 ret = generic_write_checks(iocb, from);
104 if (ret <= 0)
105 goto out;
106
107 /*
108 * Unaligned direct AIO must be serialized among each other as zeroing
109 * of partial blocks of two competing unaligned AIOs can result in data
110 * corruption.
111 */
112 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
113 !is_sync_kiocb(iocb) &&
114 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
115 unaligned_aio = 1;
116 ext4_unwritten_wait(inode);
117 }
118
119 /*
120 * If we have encountered a bitmap-format file, the size limit
121 * is smaller than s_maxbytes, which is for extent-mapped files.
122 */
123 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
124 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
125
126 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
127 ret = -EFBIG;
128 goto out;
129 }
130 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
131 }
132
133 iocb->private = &overwrite;
134 if (o_direct) {
135 size_t length = iov_iter_count(from);
136 loff_t pos = iocb->ki_pos;
137 blk_start_plug(&plug);
138
139 /* check whether we do a DIO overwrite or not */
140 if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
141 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
142 struct ext4_map_blocks map;
143 unsigned int blkbits = inode->i_blkbits;
144 int err, len;
145
146 map.m_lblk = pos >> blkbits;
147 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
148 - map.m_lblk;
149 len = map.m_len;
150
151 err = ext4_map_blocks(NULL, inode, &map, 0);
152 /*
153 * 'err==len' means that all of blocks has
154 * been preallocated no matter they are
155 * initialized or not. For excluding
156 * unwritten extents, we need to check
157 * m_flags. There are two conditions that
158 * indicate for initialized extents. 1) If we
159 * hit extent cache, EXT4_MAP_MAPPED flag is
160 * returned; 2) If we do a real lookup,
161 * non-flags are returned. So we should check
162 * these two conditions.
163 */
164 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
165 overwrite = 1;
166 }
167 }
168
169 ret = __generic_file_write_iter(iocb, from);
170 inode_unlock(inode);
171
172 if (ret > 0) {
173 ssize_t err;
174
175 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
176 if (err < 0)
177 ret = err;
178 }
179 if (o_direct)
180 blk_finish_plug(&plug);
181
182 return ret;
183
184out:
185 inode_unlock(inode);
186 return ret;
187}
188
189#ifdef CONFIG_FS_DAX
190static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
191{
192 int result;
193 handle_t *handle = NULL;
194 struct inode *inode = file_inode(vma->vm_file);
195 struct super_block *sb = inode->i_sb;
196 bool write = vmf->flags & FAULT_FLAG_WRITE;
197
198 if (write) {
199 sb_start_pagefault(sb);
200 file_update_time(vma->vm_file);
201 down_read(&EXT4_I(inode)->i_mmap_sem);
202 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
203 EXT4_DATA_TRANS_BLOCKS(sb));
204 } else
205 down_read(&EXT4_I(inode)->i_mmap_sem);
206
207 if (IS_ERR(handle))
208 result = VM_FAULT_SIGBUS;
209 else
210 result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL);
211
212 if (write) {
213 if (!IS_ERR(handle))
214 ext4_journal_stop(handle);
215 up_read(&EXT4_I(inode)->i_mmap_sem);
216 sb_end_pagefault(sb);
217 } else
218 up_read(&EXT4_I(inode)->i_mmap_sem);
219
220 return result;
221}
222
223static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
224 pmd_t *pmd, unsigned int flags)
225{
226 int result;
227 handle_t *handle = NULL;
228 struct inode *inode = file_inode(vma->vm_file);
229 struct super_block *sb = inode->i_sb;
230 bool write = flags & FAULT_FLAG_WRITE;
231
232 if (write) {
233 sb_start_pagefault(sb);
234 file_update_time(vma->vm_file);
235 down_read(&EXT4_I(inode)->i_mmap_sem);
236 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
237 ext4_chunk_trans_blocks(inode,
238 PMD_SIZE / PAGE_SIZE));
239 } else
240 down_read(&EXT4_I(inode)->i_mmap_sem);
241
242 if (IS_ERR(handle))
243 result = VM_FAULT_SIGBUS;
244 else
245 result = __dax_pmd_fault(vma, addr, pmd, flags,
246 ext4_dax_mmap_get_block, NULL);
247
248 if (write) {
249 if (!IS_ERR(handle))
250 ext4_journal_stop(handle);
251 up_read(&EXT4_I(inode)->i_mmap_sem);
252 sb_end_pagefault(sb);
253 } else
254 up_read(&EXT4_I(inode)->i_mmap_sem);
255
256 return result;
257}
258
259/*
260 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
261 * handler we check for races agaist truncate. Note that since we cycle through
262 * i_mmap_sem, we are sure that also any hole punching that began before we
263 * were called is finished by now and so if it included part of the file we
264 * are working on, our pte will get unmapped and the check for pte_same() in
265 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
266 * desired.
267 */
268static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
269 struct vm_fault *vmf)
270{
271 struct inode *inode = file_inode(vma->vm_file);
272 struct super_block *sb = inode->i_sb;
273 loff_t size;
274 int ret;
275
276 sb_start_pagefault(sb);
277 file_update_time(vma->vm_file);
278 down_read(&EXT4_I(inode)->i_mmap_sem);
279 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
280 if (vmf->pgoff >= size)
281 ret = VM_FAULT_SIGBUS;
282 else
283 ret = dax_pfn_mkwrite(vma, vmf);
284 up_read(&EXT4_I(inode)->i_mmap_sem);
285 sb_end_pagefault(sb);
286
287 return ret;
288}
289
290static const struct vm_operations_struct ext4_dax_vm_ops = {
291 .fault = ext4_dax_fault,
292 .pmd_fault = ext4_dax_pmd_fault,
293 .page_mkwrite = ext4_dax_fault,
294 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
295};
296#else
297#define ext4_dax_vm_ops ext4_file_vm_ops
298#endif
299
300static const struct vm_operations_struct ext4_file_vm_ops = {
301 .fault = ext4_filemap_fault,
302 .map_pages = filemap_map_pages,
303 .page_mkwrite = ext4_page_mkwrite,
304};
305
306static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
307{
308 struct inode *inode = file->f_mapping->host;
309
310 if (ext4_encrypted_inode(inode)) {
311 int err = ext4_get_encryption_info(inode);
312 if (err)
313 return 0;
314 if (ext4_encryption_info(inode) == NULL)
315 return -ENOKEY;
316 }
317 file_accessed(file);
318 if (IS_DAX(file_inode(file))) {
319 vma->vm_ops = &ext4_dax_vm_ops;
320 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
321 } else {
322 vma->vm_ops = &ext4_file_vm_ops;
323 }
324 return 0;
325}
326
327static int ext4_file_open(struct inode * inode, struct file * filp)
328{
329 struct super_block *sb = inode->i_sb;
330 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
331 struct vfsmount *mnt = filp->f_path.mnt;
332 struct dentry *dir;
333 struct path path;
334 char buf[64], *cp;
335 int ret;
336
337 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
338 !(sb->s_flags & MS_RDONLY))) {
339 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
340 /*
341 * Sample where the filesystem has been mounted and
342 * store it in the superblock for sysadmin convenience
343 * when trying to sort through large numbers of block
344 * devices or filesystem images.
345 */
346 memset(buf, 0, sizeof(buf));
347 path.mnt = mnt;
348 path.dentry = mnt->mnt_root;
349 cp = d_path(&path, buf, sizeof(buf));
350 if (!IS_ERR(cp)) {
351 handle_t *handle;
352 int err;
353
354 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
355 if (IS_ERR(handle))
356 return PTR_ERR(handle);
357 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
358 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
359 if (err) {
360 ext4_journal_stop(handle);
361 return err;
362 }
363 strlcpy(sbi->s_es->s_last_mounted, cp,
364 sizeof(sbi->s_es->s_last_mounted));
365 ext4_handle_dirty_super(handle, sb);
366 ext4_journal_stop(handle);
367 }
368 }
369 if (ext4_encrypted_inode(inode)) {
370 ret = ext4_get_encryption_info(inode);
371 if (ret)
372 return -EACCES;
373 if (ext4_encryption_info(inode) == NULL)
374 return -ENOKEY;
375 }
376
377 dir = dget_parent(file_dentry(filp));
378 if (ext4_encrypted_inode(d_inode(dir)) &&
379 !ext4_is_child_context_consistent_with_parent(d_inode(dir), inode)) {
380 ext4_warning(inode->i_sb,
381 "Inconsistent encryption contexts: %lu/%lu\n",
382 (unsigned long) d_inode(dir)->i_ino,
383 (unsigned long) inode->i_ino);
384 dput(dir);
385 return -EPERM;
386 }
387 dput(dir);
388 /*
389 * Set up the jbd2_inode if we are opening the inode for
390 * writing and the journal is present
391 */
392 if (filp->f_mode & FMODE_WRITE) {
393 ret = ext4_inode_attach_jinode(inode);
394 if (ret < 0)
395 return ret;
396 }
397 return dquot_file_open(inode, filp);
398}
399
400/*
401 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
402 * file rather than ext4_ext_walk_space() because we can introduce
403 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
404 * function. When extent status tree has been fully implemented, it will
405 * track all extent status for a file and we can directly use it to
406 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
407 */
408
409/*
410 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
411 * lookup page cache to check whether or not there has some data between
412 * [startoff, endoff] because, if this range contains an unwritten extent,
413 * we determine this extent as a data or a hole according to whether the
414 * page cache has data or not.
415 */
416static int ext4_find_unwritten_pgoff(struct inode *inode,
417 int whence,
418 ext4_lblk_t end_blk,
419 loff_t *offset)
420{
421 struct pagevec pvec;
422 unsigned int blkbits;
423 pgoff_t index;
424 pgoff_t end;
425 loff_t endoff;
426 loff_t startoff;
427 loff_t lastoff;
428 int found = 0;
429
430 blkbits = inode->i_sb->s_blocksize_bits;
431 startoff = *offset;
432 lastoff = startoff;
433 endoff = (loff_t)end_blk << blkbits;
434
435 index = startoff >> PAGE_SHIFT;
436 end = endoff >> PAGE_SHIFT;
437
438 pagevec_init(&pvec, 0);
439 do {
440 int i, num;
441 unsigned long nr_pages;
442
443 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
444 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
445 (pgoff_t)num);
446 if (nr_pages == 0) {
447 if (whence == SEEK_DATA)
448 break;
449
450 BUG_ON(whence != SEEK_HOLE);
451 /*
452 * If this is the first time to go into the loop and
453 * offset is not beyond the end offset, it will be a
454 * hole at this offset
455 */
456 if (lastoff == startoff || lastoff < endoff)
457 found = 1;
458 break;
459 }
460
461 /*
462 * If this is the first time to go into the loop and
463 * offset is smaller than the first page offset, it will be a
464 * hole at this offset.
465 */
466 if (lastoff == startoff && whence == SEEK_HOLE &&
467 lastoff < page_offset(pvec.pages[0])) {
468 found = 1;
469 break;
470 }
471
472 for (i = 0; i < nr_pages; i++) {
473 struct page *page = pvec.pages[i];
474 struct buffer_head *bh, *head;
475
476 /*
477 * If the current offset is not beyond the end of given
478 * range, it will be a hole.
479 */
480 if (lastoff < endoff && whence == SEEK_HOLE &&
481 page->index > end) {
482 found = 1;
483 *offset = lastoff;
484 goto out;
485 }
486
487 lock_page(page);
488
489 if (unlikely(page->mapping != inode->i_mapping)) {
490 unlock_page(page);
491 continue;
492 }
493
494 if (!page_has_buffers(page)) {
495 unlock_page(page);
496 continue;
497 }
498
499 if (page_has_buffers(page)) {
500 lastoff = page_offset(page);
501 bh = head = page_buffers(page);
502 do {
503 if (buffer_uptodate(bh) ||
504 buffer_unwritten(bh)) {
505 if (whence == SEEK_DATA)
506 found = 1;
507 } else {
508 if (whence == SEEK_HOLE)
509 found = 1;
510 }
511 if (found) {
512 *offset = max_t(loff_t,
513 startoff, lastoff);
514 unlock_page(page);
515 goto out;
516 }
517 lastoff += bh->b_size;
518 bh = bh->b_this_page;
519 } while (bh != head);
520 }
521
522 lastoff = page_offset(page) + PAGE_SIZE;
523 unlock_page(page);
524 }
525
526 /*
527 * The no. of pages is less than our desired, that would be a
528 * hole in there.
529 */
530 if (nr_pages < num && whence == SEEK_HOLE) {
531 found = 1;
532 *offset = lastoff;
533 break;
534 }
535
536 index = pvec.pages[i - 1]->index + 1;
537 pagevec_release(&pvec);
538 } while (index <= end);
539
540out:
541 pagevec_release(&pvec);
542 return found;
543}
544
545/*
546 * ext4_seek_data() retrieves the offset for SEEK_DATA.
547 */
548static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
549{
550 struct inode *inode = file->f_mapping->host;
551 struct extent_status es;
552 ext4_lblk_t start, last, end;
553 loff_t dataoff, isize;
554 int blkbits;
555 int ret;
556
557 inode_lock(inode);
558
559 isize = i_size_read(inode);
560 if (offset >= isize) {
561 inode_unlock(inode);
562 return -ENXIO;
563 }
564
565 blkbits = inode->i_sb->s_blocksize_bits;
566 start = offset >> blkbits;
567 last = start;
568 end = isize >> blkbits;
569 dataoff = offset;
570
571 do {
572 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
573 if (ret <= 0) {
574 /* No extent found -> no data */
575 if (ret == 0)
576 ret = -ENXIO;
577 inode_unlock(inode);
578 return ret;
579 }
580
581 last = es.es_lblk;
582 if (last != start)
583 dataoff = (loff_t)last << blkbits;
584 if (!ext4_es_is_unwritten(&es))
585 break;
586
587 /*
588 * If there is a unwritten extent at this offset,
589 * it will be as a data or a hole according to page
590 * cache that has data or not.
591 */
592 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
593 es.es_lblk + es.es_len, &dataoff))
594 break;
595 last += es.es_len;
596 dataoff = (loff_t)last << blkbits;
597 cond_resched();
598 } while (last <= end);
599
600 inode_unlock(inode);
601
602 if (dataoff > isize)
603 return -ENXIO;
604
605 return vfs_setpos(file, dataoff, maxsize);
606}
607
608/*
609 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
610 */
611static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
612{
613 struct inode *inode = file->f_mapping->host;
614 struct extent_status es;
615 ext4_lblk_t start, last, end;
616 loff_t holeoff, isize;
617 int blkbits;
618 int ret;
619
620 inode_lock(inode);
621
622 isize = i_size_read(inode);
623 if (offset >= isize) {
624 inode_unlock(inode);
625 return -ENXIO;
626 }
627
628 blkbits = inode->i_sb->s_blocksize_bits;
629 start = offset >> blkbits;
630 last = start;
631 end = isize >> blkbits;
632 holeoff = offset;
633
634 do {
635 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
636 if (ret < 0) {
637 inode_unlock(inode);
638 return ret;
639 }
640 /* Found a hole? */
641 if (ret == 0 || es.es_lblk > last) {
642 if (last != start)
643 holeoff = (loff_t)last << blkbits;
644 break;
645 }
646 /*
647 * If there is a unwritten extent at this offset,
648 * it will be as a data or a hole according to page
649 * cache that has data or not.
650 */
651 if (ext4_es_is_unwritten(&es) &&
652 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
653 last + es.es_len, &holeoff))
654 break;
655
656 last += es.es_len;
657 holeoff = (loff_t)last << blkbits;
658 cond_resched();
659 } while (last <= end);
660
661 inode_unlock(inode);
662
663 if (holeoff > isize)
664 holeoff = isize;
665
666 return vfs_setpos(file, holeoff, maxsize);
667}
668
669/*
670 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
671 * by calling generic_file_llseek_size() with the appropriate maxbytes
672 * value for each.
673 */
674loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
675{
676 struct inode *inode = file->f_mapping->host;
677 loff_t maxbytes;
678
679 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
680 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
681 else
682 maxbytes = inode->i_sb->s_maxbytes;
683
684 switch (whence) {
685 case SEEK_SET:
686 case SEEK_CUR:
687 case SEEK_END:
688 return generic_file_llseek_size(file, offset, whence,
689 maxbytes, i_size_read(inode));
690 case SEEK_DATA:
691 return ext4_seek_data(file, offset, maxbytes);
692 case SEEK_HOLE:
693 return ext4_seek_hole(file, offset, maxbytes);
694 }
695
696 return -EINVAL;
697}
698
699const struct file_operations ext4_file_operations = {
700 .llseek = ext4_llseek,
701 .read_iter = generic_file_read_iter,
702 .write_iter = ext4_file_write_iter,
703 .unlocked_ioctl = ext4_ioctl,
704#ifdef CONFIG_COMPAT
705 .compat_ioctl = ext4_compat_ioctl,
706#endif
707 .mmap = ext4_file_mmap,
708 .open = ext4_file_open,
709 .release = ext4_release_file,
710 .fsync = ext4_sync_file,
711 .splice_read = generic_file_splice_read,
712 .splice_write = iter_file_splice_write,
713 .fallocate = ext4_fallocate,
714};
715
716const struct inode_operations ext4_file_inode_operations = {
717 .setattr = ext4_setattr,
718 .getattr = ext4_getattr,
719 .setxattr = generic_setxattr,
720 .getxattr = generic_getxattr,
721 .listxattr = ext4_listxattr,
722 .removexattr = generic_removexattr,
723 .get_acl = ext4_get_acl,
724 .set_acl = ext4_set_acl,
725 .fiemap = ext4_fiemap,
726};
727
1/*
2 * linux/fs/ext4/file.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/file.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * ext4 fs regular file handling primitives
16 *
17 * 64-bit file support on 64-bit platforms by Jakub Jelinek
18 * (jj@sunsite.ms.mff.cuni.cz)
19 */
20
21#include <linux/time.h>
22#include <linux/fs.h>
23#include <linux/jbd2.h>
24#include <linux/mount.h>
25#include <linux/path.h>
26#include <linux/aio.h>
27#include <linux/quotaops.h>
28#include <linux/pagevec.h>
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34/*
35 * Called when an inode is released. Note that this is different
36 * from ext4_file_open: open gets called at every open, but release
37 * gets called only when /all/ the files are closed.
38 */
39static int ext4_release_file(struct inode *inode, struct file *filp)
40{
41 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
42 ext4_alloc_da_blocks(inode);
43 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
44 }
45 /* if we are the last writer on the inode, drop the block reservation */
46 if ((filp->f_mode & FMODE_WRITE) &&
47 (atomic_read(&inode->i_writecount) == 1) &&
48 !EXT4_I(inode)->i_reserved_data_blocks)
49 {
50 down_write(&EXT4_I(inode)->i_data_sem);
51 ext4_discard_preallocations(inode);
52 up_write(&EXT4_I(inode)->i_data_sem);
53 }
54 if (is_dx(inode) && filp->private_data)
55 ext4_htree_free_dir_info(filp->private_data);
56
57 return 0;
58}
59
60void ext4_unwritten_wait(struct inode *inode)
61{
62 wait_queue_head_t *wq = ext4_ioend_wq(inode);
63
64 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
65}
66
67/*
68 * This tests whether the IO in question is block-aligned or not.
69 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
70 * are converted to written only after the IO is complete. Until they are
71 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
72 * it needs to zero out portions of the start and/or end block. If 2 AIO
73 * threads are at work on the same unwritten block, they must be synchronized
74 * or one thread will zero the other's data, causing corruption.
75 */
76static int
77ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
78 unsigned long nr_segs, loff_t pos)
79{
80 struct super_block *sb = inode->i_sb;
81 int blockmask = sb->s_blocksize - 1;
82 size_t count = iov_length(iov, nr_segs);
83 loff_t final_size = pos + count;
84
85 if (pos >= i_size_read(inode))
86 return 0;
87
88 if ((pos & blockmask) || (final_size & blockmask))
89 return 1;
90
91 return 0;
92}
93
94static ssize_t
95ext4_file_dio_write(struct kiocb *iocb, const struct iovec *iov,
96 unsigned long nr_segs, loff_t pos)
97{
98 struct file *file = iocb->ki_filp;
99 struct inode *inode = file->f_mapping->host;
100 struct blk_plug plug;
101 int unaligned_aio = 0;
102 ssize_t ret;
103 int overwrite = 0;
104 size_t length = iov_length(iov, nr_segs);
105
106 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
107 !is_sync_kiocb(iocb))
108 unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
109
110 /* Unaligned direct AIO must be serialized; see comment above */
111 if (unaligned_aio) {
112 mutex_lock(ext4_aio_mutex(inode));
113 ext4_unwritten_wait(inode);
114 }
115
116 BUG_ON(iocb->ki_pos != pos);
117
118 mutex_lock(&inode->i_mutex);
119 blk_start_plug(&plug);
120
121 iocb->private = &overwrite;
122
123 /* check whether we do a DIO overwrite or not */
124 if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
125 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
126 struct ext4_map_blocks map;
127 unsigned int blkbits = inode->i_blkbits;
128 int err, len;
129
130 map.m_lblk = pos >> blkbits;
131 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
132 - map.m_lblk;
133 len = map.m_len;
134
135 err = ext4_map_blocks(NULL, inode, &map, 0);
136 /*
137 * 'err==len' means that all of blocks has been preallocated no
138 * matter they are initialized or not. For excluding
139 * uninitialized extents, we need to check m_flags. There are
140 * two conditions that indicate for initialized extents.
141 * 1) If we hit extent cache, EXT4_MAP_MAPPED flag is returned;
142 * 2) If we do a real lookup, non-flags are returned.
143 * So we should check these two conditions.
144 */
145 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
146 overwrite = 1;
147 }
148
149 ret = __generic_file_aio_write(iocb, iov, nr_segs);
150 mutex_unlock(&inode->i_mutex);
151
152 if (ret > 0) {
153 ssize_t err;
154
155 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
156 if (err < 0)
157 ret = err;
158 }
159 blk_finish_plug(&plug);
160
161 if (unaligned_aio)
162 mutex_unlock(ext4_aio_mutex(inode));
163
164 return ret;
165}
166
167static ssize_t
168ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
170{
171 struct inode *inode = file_inode(iocb->ki_filp);
172 ssize_t ret;
173
174 /*
175 * If we have encountered a bitmap-format file, the size limit
176 * is smaller than s_maxbytes, which is for extent-mapped files.
177 */
178
179 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
180 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
181 size_t length = iov_length(iov, nr_segs);
182
183 if ((pos > sbi->s_bitmap_maxbytes ||
184 (pos == sbi->s_bitmap_maxbytes && length > 0)))
185 return -EFBIG;
186
187 if (pos + length > sbi->s_bitmap_maxbytes) {
188 nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
189 sbi->s_bitmap_maxbytes - pos);
190 }
191 }
192
193 if (unlikely(iocb->ki_filp->f_flags & O_DIRECT))
194 ret = ext4_file_dio_write(iocb, iov, nr_segs, pos);
195 else
196 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
197
198 return ret;
199}
200
201static const struct vm_operations_struct ext4_file_vm_ops = {
202 .fault = filemap_fault,
203 .map_pages = filemap_map_pages,
204 .page_mkwrite = ext4_page_mkwrite,
205 .remap_pages = generic_file_remap_pages,
206};
207
208static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
209{
210 struct address_space *mapping = file->f_mapping;
211
212 if (!mapping->a_ops->readpage)
213 return -ENOEXEC;
214 file_accessed(file);
215 vma->vm_ops = &ext4_file_vm_ops;
216 return 0;
217}
218
219static int ext4_file_open(struct inode * inode, struct file * filp)
220{
221 struct super_block *sb = inode->i_sb;
222 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
223 struct vfsmount *mnt = filp->f_path.mnt;
224 struct path path;
225 char buf[64], *cp;
226
227 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
228 !(sb->s_flags & MS_RDONLY))) {
229 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
230 /*
231 * Sample where the filesystem has been mounted and
232 * store it in the superblock for sysadmin convenience
233 * when trying to sort through large numbers of block
234 * devices or filesystem images.
235 */
236 memset(buf, 0, sizeof(buf));
237 path.mnt = mnt;
238 path.dentry = mnt->mnt_root;
239 cp = d_path(&path, buf, sizeof(buf));
240 if (!IS_ERR(cp)) {
241 handle_t *handle;
242 int err;
243
244 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
245 if (IS_ERR(handle))
246 return PTR_ERR(handle);
247 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
248 if (err) {
249 ext4_journal_stop(handle);
250 return err;
251 }
252 strlcpy(sbi->s_es->s_last_mounted, cp,
253 sizeof(sbi->s_es->s_last_mounted));
254 ext4_handle_dirty_super(handle, sb);
255 ext4_journal_stop(handle);
256 }
257 }
258 /*
259 * Set up the jbd2_inode if we are opening the inode for
260 * writing and the journal is present
261 */
262 if (filp->f_mode & FMODE_WRITE) {
263 int ret = ext4_inode_attach_jinode(inode);
264 if (ret < 0)
265 return ret;
266 }
267 return dquot_file_open(inode, filp);
268}
269
270/*
271 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
272 * file rather than ext4_ext_walk_space() because we can introduce
273 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
274 * function. When extent status tree has been fully implemented, it will
275 * track all extent status for a file and we can directly use it to
276 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
277 */
278
279/*
280 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
281 * lookup page cache to check whether or not there has some data between
282 * [startoff, endoff] because, if this range contains an unwritten extent,
283 * we determine this extent as a data or a hole according to whether the
284 * page cache has data or not.
285 */
286static int ext4_find_unwritten_pgoff(struct inode *inode,
287 int whence,
288 struct ext4_map_blocks *map,
289 loff_t *offset)
290{
291 struct pagevec pvec;
292 unsigned int blkbits;
293 pgoff_t index;
294 pgoff_t end;
295 loff_t endoff;
296 loff_t startoff;
297 loff_t lastoff;
298 int found = 0;
299
300 blkbits = inode->i_sb->s_blocksize_bits;
301 startoff = *offset;
302 lastoff = startoff;
303 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
304
305 index = startoff >> PAGE_CACHE_SHIFT;
306 end = endoff >> PAGE_CACHE_SHIFT;
307
308 pagevec_init(&pvec, 0);
309 do {
310 int i, num;
311 unsigned long nr_pages;
312
313 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
314 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
315 (pgoff_t)num);
316 if (nr_pages == 0) {
317 if (whence == SEEK_DATA)
318 break;
319
320 BUG_ON(whence != SEEK_HOLE);
321 /*
322 * If this is the first time to go into the loop and
323 * offset is not beyond the end offset, it will be a
324 * hole at this offset
325 */
326 if (lastoff == startoff || lastoff < endoff)
327 found = 1;
328 break;
329 }
330
331 /*
332 * If this is the first time to go into the loop and
333 * offset is smaller than the first page offset, it will be a
334 * hole at this offset.
335 */
336 if (lastoff == startoff && whence == SEEK_HOLE &&
337 lastoff < page_offset(pvec.pages[0])) {
338 found = 1;
339 break;
340 }
341
342 for (i = 0; i < nr_pages; i++) {
343 struct page *page = pvec.pages[i];
344 struct buffer_head *bh, *head;
345
346 /*
347 * If the current offset is not beyond the end of given
348 * range, it will be a hole.
349 */
350 if (lastoff < endoff && whence == SEEK_HOLE &&
351 page->index > end) {
352 found = 1;
353 *offset = lastoff;
354 goto out;
355 }
356
357 lock_page(page);
358
359 if (unlikely(page->mapping != inode->i_mapping)) {
360 unlock_page(page);
361 continue;
362 }
363
364 if (!page_has_buffers(page)) {
365 unlock_page(page);
366 continue;
367 }
368
369 if (page_has_buffers(page)) {
370 lastoff = page_offset(page);
371 bh = head = page_buffers(page);
372 do {
373 if (buffer_uptodate(bh) ||
374 buffer_unwritten(bh)) {
375 if (whence == SEEK_DATA)
376 found = 1;
377 } else {
378 if (whence == SEEK_HOLE)
379 found = 1;
380 }
381 if (found) {
382 *offset = max_t(loff_t,
383 startoff, lastoff);
384 unlock_page(page);
385 goto out;
386 }
387 lastoff += bh->b_size;
388 bh = bh->b_this_page;
389 } while (bh != head);
390 }
391
392 lastoff = page_offset(page) + PAGE_SIZE;
393 unlock_page(page);
394 }
395
396 /*
397 * The no. of pages is less than our desired, that would be a
398 * hole in there.
399 */
400 if (nr_pages < num && whence == SEEK_HOLE) {
401 found = 1;
402 *offset = lastoff;
403 break;
404 }
405
406 index = pvec.pages[i - 1]->index + 1;
407 pagevec_release(&pvec);
408 } while (index <= end);
409
410out:
411 pagevec_release(&pvec);
412 return found;
413}
414
415/*
416 * ext4_seek_data() retrieves the offset for SEEK_DATA.
417 */
418static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
419{
420 struct inode *inode = file->f_mapping->host;
421 struct ext4_map_blocks map;
422 struct extent_status es;
423 ext4_lblk_t start, last, end;
424 loff_t dataoff, isize;
425 int blkbits;
426 int ret = 0;
427
428 mutex_lock(&inode->i_mutex);
429
430 isize = i_size_read(inode);
431 if (offset >= isize) {
432 mutex_unlock(&inode->i_mutex);
433 return -ENXIO;
434 }
435
436 blkbits = inode->i_sb->s_blocksize_bits;
437 start = offset >> blkbits;
438 last = start;
439 end = isize >> blkbits;
440 dataoff = offset;
441
442 do {
443 map.m_lblk = last;
444 map.m_len = end - last + 1;
445 ret = ext4_map_blocks(NULL, inode, &map, 0);
446 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
447 if (last != start)
448 dataoff = (loff_t)last << blkbits;
449 break;
450 }
451
452 /*
453 * If there is a delay extent at this offset,
454 * it will be as a data.
455 */
456 ext4_es_find_delayed_extent_range(inode, last, last, &es);
457 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
458 if (last != start)
459 dataoff = (loff_t)last << blkbits;
460 break;
461 }
462
463 /*
464 * If there is a unwritten extent at this offset,
465 * it will be as a data or a hole according to page
466 * cache that has data or not.
467 */
468 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
469 int unwritten;
470 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
471 &map, &dataoff);
472 if (unwritten)
473 break;
474 }
475
476 last++;
477 dataoff = (loff_t)last << blkbits;
478 } while (last <= end);
479
480 mutex_unlock(&inode->i_mutex);
481
482 if (dataoff > isize)
483 return -ENXIO;
484
485 return vfs_setpos(file, dataoff, maxsize);
486}
487
488/*
489 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
490 */
491static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
492{
493 struct inode *inode = file->f_mapping->host;
494 struct ext4_map_blocks map;
495 struct extent_status es;
496 ext4_lblk_t start, last, end;
497 loff_t holeoff, isize;
498 int blkbits;
499 int ret = 0;
500
501 mutex_lock(&inode->i_mutex);
502
503 isize = i_size_read(inode);
504 if (offset >= isize) {
505 mutex_unlock(&inode->i_mutex);
506 return -ENXIO;
507 }
508
509 blkbits = inode->i_sb->s_blocksize_bits;
510 start = offset >> blkbits;
511 last = start;
512 end = isize >> blkbits;
513 holeoff = offset;
514
515 do {
516 map.m_lblk = last;
517 map.m_len = end - last + 1;
518 ret = ext4_map_blocks(NULL, inode, &map, 0);
519 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
520 last += ret;
521 holeoff = (loff_t)last << blkbits;
522 continue;
523 }
524
525 /*
526 * If there is a delay extent at this offset,
527 * we will skip this extent.
528 */
529 ext4_es_find_delayed_extent_range(inode, last, last, &es);
530 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
531 last = es.es_lblk + es.es_len;
532 holeoff = (loff_t)last << blkbits;
533 continue;
534 }
535
536 /*
537 * If there is a unwritten extent at this offset,
538 * it will be as a data or a hole according to page
539 * cache that has data or not.
540 */
541 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
542 int unwritten;
543 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
544 &map, &holeoff);
545 if (!unwritten) {
546 last += ret;
547 holeoff = (loff_t)last << blkbits;
548 continue;
549 }
550 }
551
552 /* find a hole */
553 break;
554 } while (last <= end);
555
556 mutex_unlock(&inode->i_mutex);
557
558 if (holeoff > isize)
559 holeoff = isize;
560
561 return vfs_setpos(file, holeoff, maxsize);
562}
563
564/*
565 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
566 * by calling generic_file_llseek_size() with the appropriate maxbytes
567 * value for each.
568 */
569loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
570{
571 struct inode *inode = file->f_mapping->host;
572 loff_t maxbytes;
573
574 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
575 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
576 else
577 maxbytes = inode->i_sb->s_maxbytes;
578
579 switch (whence) {
580 case SEEK_SET:
581 case SEEK_CUR:
582 case SEEK_END:
583 return generic_file_llseek_size(file, offset, whence,
584 maxbytes, i_size_read(inode));
585 case SEEK_DATA:
586 return ext4_seek_data(file, offset, maxbytes);
587 case SEEK_HOLE:
588 return ext4_seek_hole(file, offset, maxbytes);
589 }
590
591 return -EINVAL;
592}
593
594const struct file_operations ext4_file_operations = {
595 .llseek = ext4_llseek,
596 .read = do_sync_read,
597 .write = do_sync_write,
598 .aio_read = generic_file_aio_read,
599 .aio_write = ext4_file_write,
600 .unlocked_ioctl = ext4_ioctl,
601#ifdef CONFIG_COMPAT
602 .compat_ioctl = ext4_compat_ioctl,
603#endif
604 .mmap = ext4_file_mmap,
605 .open = ext4_file_open,
606 .release = ext4_release_file,
607 .fsync = ext4_sync_file,
608 .splice_read = generic_file_splice_read,
609 .splice_write = generic_file_splice_write,
610 .fallocate = ext4_fallocate,
611};
612
613const struct inode_operations ext4_file_inode_operations = {
614 .setattr = ext4_setattr,
615 .getattr = ext4_getattr,
616 .setxattr = generic_setxattr,
617 .getxattr = generic_getxattr,
618 .listxattr = ext4_listxattr,
619 .removexattr = generic_removexattr,
620 .get_acl = ext4_get_acl,
621 .set_acl = ext4_set_acl,
622 .fiemap = ext4_fiemap,
623};
624