Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-2.6.git] / fs / ext4 / file.c
blob6f4cc567c382b7289e2b35a7038756a581b4cd95
1 /*
2 * linux/fs/ext4/file.c
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)
9 * from
11 * linux/fs/minix/file.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * ext4 fs regular file handling primitives
17 * 64-bit file support on 64-bit platforms by Jakub Jelinek
18 * (jj@sunsite.ms.mff.cuni.cz)
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"
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.
39 static int ext4_release_file(struct inode *inode, struct file *filp)
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);
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)
50 down_write(&EXT4_I(inode)->i_data_sem);
51 ext4_discard_preallocations(inode);
52 up_write(&EXT4_I(inode)->i_data_sem);
54 if (is_dx(inode) && filp->private_data)
55 ext4_htree_free_dir_info(filp->private_data);
57 return 0;
60 void ext4_unwritten_wait(struct inode *inode)
62 wait_queue_head_t *wq = ext4_ioend_wq(inode);
64 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
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.
76 static int
77 ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
78 unsigned long nr_segs, loff_t pos)
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;
85 if (pos >= inode->i_size)
86 return 0;
88 if ((pos & blockmask) || (final_size & blockmask))
89 return 1;
91 return 0;
94 static ssize_t
95 ext4_file_dio_write(struct kiocb *iocb, const struct iovec *iov,
96 unsigned long nr_segs, loff_t pos)
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);
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);
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);
116 BUG_ON(iocb->ki_pos != pos);
118 mutex_lock(&inode->i_mutex);
119 blk_start_plug(&plug);
121 iocb->private = &overwrite;
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;
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;
135 err = ext4_map_blocks(NULL, inode, &map, 0);
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.
145 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
146 overwrite = 1;
149 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
150 mutex_unlock(&inode->i_mutex);
152 if (ret > 0 || ret == -EIOCBQUEUED) {
153 ssize_t err;
155 err = generic_write_sync(file, pos, ret);
156 if (err < 0 && ret > 0)
157 ret = err;
159 blk_finish_plug(&plug);
161 if (unaligned_aio)
162 mutex_unlock(ext4_aio_mutex(inode));
164 return ret;
167 static ssize_t
168 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
171 struct inode *inode = file_inode(iocb->ki_filp);
172 ssize_t ret;
175 * If we have encountered a bitmap-format file, the size limit
176 * is smaller than s_maxbytes, which is for extent-mapped files.
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);
183 if ((pos > sbi->s_bitmap_maxbytes ||
184 (pos == sbi->s_bitmap_maxbytes && length > 0)))
185 return -EFBIG;
187 if (pos + length > sbi->s_bitmap_maxbytes) {
188 nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
189 sbi->s_bitmap_maxbytes - pos);
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);
198 return ret;
201 static const struct vm_operations_struct ext4_file_vm_ops = {
202 .fault = filemap_fault,
203 .page_mkwrite = ext4_page_mkwrite,
204 .remap_pages = generic_file_remap_pages,
207 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
209 struct address_space *mapping = file->f_mapping;
211 if (!mapping->a_ops->readpage)
212 return -ENOEXEC;
213 file_accessed(file);
214 vma->vm_ops = &ext4_file_vm_ops;
215 return 0;
218 static int ext4_file_open(struct inode * inode, struct file * filp)
220 struct super_block *sb = inode->i_sb;
221 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
222 struct ext4_inode_info *ei = EXT4_I(inode);
223 struct vfsmount *mnt = filp->f_path.mnt;
224 struct path path;
225 char buf[64], *cp;
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;
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.
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;
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;
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);
259 * Set up the jbd2_inode if we are opening the inode for
260 * writing and the journal is present
262 if (sbi->s_journal && !ei->jinode && (filp->f_mode & FMODE_WRITE)) {
263 struct jbd2_inode *jinode = jbd2_alloc_inode(GFP_KERNEL);
265 spin_lock(&inode->i_lock);
266 if (!ei->jinode) {
267 if (!jinode) {
268 spin_unlock(&inode->i_lock);
269 return -ENOMEM;
271 ei->jinode = jinode;
272 jbd2_journal_init_jbd_inode(ei->jinode, inode);
273 jinode = NULL;
275 spin_unlock(&inode->i_lock);
276 if (unlikely(jinode != NULL))
277 jbd2_free_inode(jinode);
279 return dquot_file_open(inode, filp);
283 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
284 * file rather than ext4_ext_walk_space() because we can introduce
285 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
286 * function. When extent status tree has been fully implemented, it will
287 * track all extent status for a file and we can directly use it to
288 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
292 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
293 * lookup page cache to check whether or not there has some data between
294 * [startoff, endoff] because, if this range contains an unwritten extent,
295 * we determine this extent as a data or a hole according to whether the
296 * page cache has data or not.
298 static int ext4_find_unwritten_pgoff(struct inode *inode,
299 int whence,
300 struct ext4_map_blocks *map,
301 loff_t *offset)
303 struct pagevec pvec;
304 unsigned int blkbits;
305 pgoff_t index;
306 pgoff_t end;
307 loff_t endoff;
308 loff_t startoff;
309 loff_t lastoff;
310 int found = 0;
312 blkbits = inode->i_sb->s_blocksize_bits;
313 startoff = *offset;
314 lastoff = startoff;
315 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
317 index = startoff >> PAGE_CACHE_SHIFT;
318 end = endoff >> PAGE_CACHE_SHIFT;
320 pagevec_init(&pvec, 0);
321 do {
322 int i, num;
323 unsigned long nr_pages;
325 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
326 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
327 (pgoff_t)num);
328 if (nr_pages == 0) {
329 if (whence == SEEK_DATA)
330 break;
332 BUG_ON(whence != SEEK_HOLE);
334 * If this is the first time to go into the loop and
335 * offset is not beyond the end offset, it will be a
336 * hole at this offset
338 if (lastoff == startoff || lastoff < endoff)
339 found = 1;
340 break;
344 * If this is the first time to go into the loop and
345 * offset is smaller than the first page offset, it will be a
346 * hole at this offset.
348 if (lastoff == startoff && whence == SEEK_HOLE &&
349 lastoff < page_offset(pvec.pages[0])) {
350 found = 1;
351 break;
354 for (i = 0; i < nr_pages; i++) {
355 struct page *page = pvec.pages[i];
356 struct buffer_head *bh, *head;
359 * If the current offset is not beyond the end of given
360 * range, it will be a hole.
362 if (lastoff < endoff && whence == SEEK_HOLE &&
363 page->index > end) {
364 found = 1;
365 *offset = lastoff;
366 goto out;
369 lock_page(page);
371 if (unlikely(page->mapping != inode->i_mapping)) {
372 unlock_page(page);
373 continue;
376 if (!page_has_buffers(page)) {
377 unlock_page(page);
378 continue;
381 if (page_has_buffers(page)) {
382 lastoff = page_offset(page);
383 bh = head = page_buffers(page);
384 do {
385 if (buffer_uptodate(bh) ||
386 buffer_unwritten(bh)) {
387 if (whence == SEEK_DATA)
388 found = 1;
389 } else {
390 if (whence == SEEK_HOLE)
391 found = 1;
393 if (found) {
394 *offset = max_t(loff_t,
395 startoff, lastoff);
396 unlock_page(page);
397 goto out;
399 lastoff += bh->b_size;
400 bh = bh->b_this_page;
401 } while (bh != head);
404 lastoff = page_offset(page) + PAGE_SIZE;
405 unlock_page(page);
409 * The no. of pages is less than our desired, that would be a
410 * hole in there.
412 if (nr_pages < num && whence == SEEK_HOLE) {
413 found = 1;
414 *offset = lastoff;
415 break;
418 index = pvec.pages[i - 1]->index + 1;
419 pagevec_release(&pvec);
420 } while (index <= end);
422 out:
423 pagevec_release(&pvec);
424 return found;
428 * ext4_seek_data() retrieves the offset for SEEK_DATA.
430 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
432 struct inode *inode = file->f_mapping->host;
433 struct ext4_map_blocks map;
434 struct extent_status es;
435 ext4_lblk_t start, last, end;
436 loff_t dataoff, isize;
437 int blkbits;
438 int ret = 0;
440 mutex_lock(&inode->i_mutex);
442 isize = i_size_read(inode);
443 if (offset >= isize) {
444 mutex_unlock(&inode->i_mutex);
445 return -ENXIO;
448 blkbits = inode->i_sb->s_blocksize_bits;
449 start = offset >> blkbits;
450 last = start;
451 end = isize >> blkbits;
452 dataoff = offset;
454 do {
455 map.m_lblk = last;
456 map.m_len = end - last + 1;
457 ret = ext4_map_blocks(NULL, inode, &map, 0);
458 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
459 if (last != start)
460 dataoff = (loff_t)last << blkbits;
461 break;
465 * If there is a delay extent at this offset,
466 * it will be as a data.
468 ext4_es_find_delayed_extent_range(inode, last, last, &es);
469 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
470 if (last != start)
471 dataoff = (loff_t)last << blkbits;
472 break;
476 * If there is a unwritten extent at this offset,
477 * it will be as a data or a hole according to page
478 * cache that has data or not.
480 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
481 int unwritten;
482 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
483 &map, &dataoff);
484 if (unwritten)
485 break;
488 last++;
489 dataoff = (loff_t)last << blkbits;
490 } while (last <= end);
492 mutex_unlock(&inode->i_mutex);
494 if (dataoff > isize)
495 return -ENXIO;
497 return vfs_setpos(file, dataoff, maxsize);
501 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
503 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
505 struct inode *inode = file->f_mapping->host;
506 struct ext4_map_blocks map;
507 struct extent_status es;
508 ext4_lblk_t start, last, end;
509 loff_t holeoff, isize;
510 int blkbits;
511 int ret = 0;
513 mutex_lock(&inode->i_mutex);
515 isize = i_size_read(inode);
516 if (offset >= isize) {
517 mutex_unlock(&inode->i_mutex);
518 return -ENXIO;
521 blkbits = inode->i_sb->s_blocksize_bits;
522 start = offset >> blkbits;
523 last = start;
524 end = isize >> blkbits;
525 holeoff = offset;
527 do {
528 map.m_lblk = last;
529 map.m_len = end - last + 1;
530 ret = ext4_map_blocks(NULL, inode, &map, 0);
531 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
532 last += ret;
533 holeoff = (loff_t)last << blkbits;
534 continue;
538 * If there is a delay extent at this offset,
539 * we will skip this extent.
541 ext4_es_find_delayed_extent_range(inode, last, last, &es);
542 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
543 last = es.es_lblk + es.es_len;
544 holeoff = (loff_t)last << blkbits;
545 continue;
549 * If there is a unwritten extent at this offset,
550 * it will be as a data or a hole according to page
551 * cache that has data or not.
553 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
554 int unwritten;
555 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
556 &map, &holeoff);
557 if (!unwritten) {
558 last += ret;
559 holeoff = (loff_t)last << blkbits;
560 continue;
564 /* find a hole */
565 break;
566 } while (last <= end);
568 mutex_unlock(&inode->i_mutex);
570 if (holeoff > isize)
571 holeoff = isize;
573 return vfs_setpos(file, holeoff, maxsize);
577 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
578 * by calling generic_file_llseek_size() with the appropriate maxbytes
579 * value for each.
581 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
583 struct inode *inode = file->f_mapping->host;
584 loff_t maxbytes;
586 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
587 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
588 else
589 maxbytes = inode->i_sb->s_maxbytes;
591 switch (whence) {
592 case SEEK_SET:
593 case SEEK_CUR:
594 case SEEK_END:
595 return generic_file_llseek_size(file, offset, whence,
596 maxbytes, i_size_read(inode));
597 case SEEK_DATA:
598 return ext4_seek_data(file, offset, maxbytes);
599 case SEEK_HOLE:
600 return ext4_seek_hole(file, offset, maxbytes);
603 return -EINVAL;
606 const struct file_operations ext4_file_operations = {
607 .llseek = ext4_llseek,
608 .read = do_sync_read,
609 .write = do_sync_write,
610 .aio_read = generic_file_aio_read,
611 .aio_write = ext4_file_write,
612 .unlocked_ioctl = ext4_ioctl,
613 #ifdef CONFIG_COMPAT
614 .compat_ioctl = ext4_compat_ioctl,
615 #endif
616 .mmap = ext4_file_mmap,
617 .open = ext4_file_open,
618 .release = ext4_release_file,
619 .fsync = ext4_sync_file,
620 .splice_read = generic_file_splice_read,
621 .splice_write = generic_file_splice_write,
622 .fallocate = ext4_fallocate,
625 const struct inode_operations ext4_file_inode_operations = {
626 .setattr = ext4_setattr,
627 .getattr = ext4_getattr,
628 .setxattr = generic_setxattr,
629 .getxattr = generic_getxattr,
630 .listxattr = ext4_listxattr,
631 .removexattr = generic_removexattr,
632 .get_acl = ext4_get_acl,
633 .fiemap = ext4_fiemap,