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)
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>
23 #include <linux/jbd2.h>
24 #include <linux/mount.h>
25 #include <linux/path.h>
26 #include <linux/quotaops.h>
27 #include <linux/pagevec.h>
28 #include <linux/uio.h>
30 #include "ext4_jbd2.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);
60 static 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.
77 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
79 struct super_block *sb = inode->i_sb;
80 int blockmask = sb->s_blocksize - 1;
82 if (pos >= i_size_read(inode))
85 if ((pos | iov_iter_alignment(from)) & blockmask)
92 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
94 struct file *file = iocb->ki_filp;
95 struct inode *inode = file_inode(iocb->ki_filp);
96 struct mutex *aio_mutex = NULL;
98 int o_direct = io_is_direct(file);
100 size_t length = iov_iter_count(from);
102 loff_t pos = iocb->ki_pos;
105 * Unaligned direct AIO must be serialized; see comment above
106 * In the case of O_APPEND, assume that we must always serialize
109 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
110 !is_sync_kiocb(iocb) &&
111 (file->f_flags & O_APPEND ||
112 ext4_unaligned_aio(inode, from, pos))) {
113 aio_mutex = ext4_aio_mutex(inode);
114 mutex_lock(aio_mutex);
115 ext4_unwritten_wait(inode);
118 mutex_lock(&inode->i_mutex);
119 if (file->f_flags & O_APPEND)
120 iocb->ki_pos = pos = i_size_read(inode);
123 * If we have encountered a bitmap-format file, the size limit
124 * is smaller than s_maxbytes, which is for extent-mapped files.
126 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
127 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
129 if ((pos > sbi->s_bitmap_maxbytes) ||
130 (pos == sbi->s_bitmap_maxbytes && length > 0)) {
131 mutex_unlock(&inode->i_mutex);
135 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
136 length = iov_iter_count(from);
139 iocb->private = &overwrite;
141 blk_start_plug(&plug);
144 /* check whether we do a DIO overwrite or not */
145 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
146 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
147 struct ext4_map_blocks map;
148 unsigned int blkbits = inode->i_blkbits;
151 map.m_lblk = pos >> blkbits;
152 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
156 err = ext4_map_blocks(NULL, inode, &map, 0);
158 * 'err==len' means that all of blocks has
159 * been preallocated no matter they are
160 * initialized or not. For excluding
161 * unwritten extents, we need to check
162 * m_flags. There are two conditions that
163 * indicate for initialized extents. 1) If we
164 * hit extent cache, EXT4_MAP_MAPPED flag is
165 * returned; 2) If we do a real lookup,
166 * non-flags are returned. So we should check
167 * these two conditions.
169 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
174 ret = generic_write_checks(file, &iocb->ki_pos, &length);
181 iov_iter_truncate(from, length);
182 ret = __generic_file_write_iter(iocb, from);
184 mutex_unlock(&inode->i_mutex);
189 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
194 blk_finish_plug(&plug);
198 mutex_unlock(aio_mutex);
203 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
205 return dax_fault(vma, vmf, ext4_get_block);
206 /* Is this the right get_block? */
209 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
211 return dax_mkwrite(vma, vmf, ext4_get_block);
214 static const struct vm_operations_struct ext4_dax_vm_ops = {
215 .fault = ext4_dax_fault,
216 .page_mkwrite = ext4_dax_mkwrite,
219 #define ext4_dax_vm_ops ext4_file_vm_ops
222 static const struct vm_operations_struct ext4_file_vm_ops = {
223 .fault = filemap_fault,
224 .map_pages = filemap_map_pages,
225 .page_mkwrite = ext4_page_mkwrite,
228 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
231 if (IS_DAX(file_inode(file))) {
232 vma->vm_ops = &ext4_dax_vm_ops;
233 vma->vm_flags |= VM_MIXEDMAP;
235 vma->vm_ops = &ext4_file_vm_ops;
240 static int ext4_file_open(struct inode * inode, struct file * filp)
242 struct super_block *sb = inode->i_sb;
243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
244 struct vfsmount *mnt = filp->f_path.mnt;
248 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
249 !(sb->s_flags & MS_RDONLY))) {
250 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
252 * Sample where the filesystem has been mounted and
253 * store it in the superblock for sysadmin convenience
254 * when trying to sort through large numbers of block
255 * devices or filesystem images.
257 memset(buf, 0, sizeof(buf));
259 path.dentry = mnt->mnt_root;
260 cp = d_path(&path, buf, sizeof(buf));
265 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
267 return PTR_ERR(handle);
268 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
269 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
271 ext4_journal_stop(handle);
274 strlcpy(sbi->s_es->s_last_mounted, cp,
275 sizeof(sbi->s_es->s_last_mounted));
276 ext4_handle_dirty_super(handle, sb);
277 ext4_journal_stop(handle);
281 * Set up the jbd2_inode if we are opening the inode for
282 * writing and the journal is present
284 if (filp->f_mode & FMODE_WRITE) {
285 int ret = ext4_inode_attach_jinode(inode);
289 return dquot_file_open(inode, filp);
293 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
294 * file rather than ext4_ext_walk_space() because we can introduce
295 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
296 * function. When extent status tree has been fully implemented, it will
297 * track all extent status for a file and we can directly use it to
298 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
302 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
303 * lookup page cache to check whether or not there has some data between
304 * [startoff, endoff] because, if this range contains an unwritten extent,
305 * we determine this extent as a data or a hole according to whether the
306 * page cache has data or not.
308 static int ext4_find_unwritten_pgoff(struct inode *inode,
310 struct ext4_map_blocks *map,
314 unsigned int blkbits;
322 blkbits = inode->i_sb->s_blocksize_bits;
325 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
327 index = startoff >> PAGE_CACHE_SHIFT;
328 end = endoff >> PAGE_CACHE_SHIFT;
330 pagevec_init(&pvec, 0);
333 unsigned long nr_pages;
335 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
336 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
339 if (whence == SEEK_DATA)
342 BUG_ON(whence != SEEK_HOLE);
344 * If this is the first time to go into the loop and
345 * offset is not beyond the end offset, it will be a
346 * hole at this offset
348 if (lastoff == startoff || lastoff < endoff)
354 * If this is the first time to go into the loop and
355 * offset is smaller than the first page offset, it will be a
356 * hole at this offset.
358 if (lastoff == startoff && whence == SEEK_HOLE &&
359 lastoff < page_offset(pvec.pages[0])) {
364 for (i = 0; i < nr_pages; i++) {
365 struct page *page = pvec.pages[i];
366 struct buffer_head *bh, *head;
369 * If the current offset is not beyond the end of given
370 * range, it will be a hole.
372 if (lastoff < endoff && whence == SEEK_HOLE &&
381 if (unlikely(page->mapping != inode->i_mapping)) {
386 if (!page_has_buffers(page)) {
391 if (page_has_buffers(page)) {
392 lastoff = page_offset(page);
393 bh = head = page_buffers(page);
395 if (buffer_uptodate(bh) ||
396 buffer_unwritten(bh)) {
397 if (whence == SEEK_DATA)
400 if (whence == SEEK_HOLE)
404 *offset = max_t(loff_t,
409 lastoff += bh->b_size;
410 bh = bh->b_this_page;
411 } while (bh != head);
414 lastoff = page_offset(page) + PAGE_SIZE;
419 * The no. of pages is less than our desired, that would be a
422 if (nr_pages < num && whence == SEEK_HOLE) {
428 index = pvec.pages[i - 1]->index + 1;
429 pagevec_release(&pvec);
430 } while (index <= end);
433 pagevec_release(&pvec);
438 * ext4_seek_data() retrieves the offset for SEEK_DATA.
440 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
442 struct inode *inode = file->f_mapping->host;
443 struct ext4_map_blocks map;
444 struct extent_status es;
445 ext4_lblk_t start, last, end;
446 loff_t dataoff, isize;
450 mutex_lock(&inode->i_mutex);
452 isize = i_size_read(inode);
453 if (offset >= isize) {
454 mutex_unlock(&inode->i_mutex);
458 blkbits = inode->i_sb->s_blocksize_bits;
459 start = offset >> blkbits;
461 end = isize >> blkbits;
466 map.m_len = end - last + 1;
467 ret = ext4_map_blocks(NULL, inode, &map, 0);
468 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
470 dataoff = (loff_t)last << blkbits;
475 * If there is a delay extent at this offset,
476 * it will be as a data.
478 ext4_es_find_delayed_extent_range(inode, last, last, &es);
479 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
481 dataoff = (loff_t)last << blkbits;
486 * If there is a unwritten extent at this offset,
487 * it will be as a data or a hole according to page
488 * cache that has data or not.
490 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
492 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
499 dataoff = (loff_t)last << blkbits;
500 } while (last <= end);
502 mutex_unlock(&inode->i_mutex);
507 return vfs_setpos(file, dataoff, maxsize);
511 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
513 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
515 struct inode *inode = file->f_mapping->host;
516 struct ext4_map_blocks map;
517 struct extent_status es;
518 ext4_lblk_t start, last, end;
519 loff_t holeoff, isize;
523 mutex_lock(&inode->i_mutex);
525 isize = i_size_read(inode);
526 if (offset >= isize) {
527 mutex_unlock(&inode->i_mutex);
531 blkbits = inode->i_sb->s_blocksize_bits;
532 start = offset >> blkbits;
534 end = isize >> blkbits;
539 map.m_len = end - last + 1;
540 ret = ext4_map_blocks(NULL, inode, &map, 0);
541 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
543 holeoff = (loff_t)last << blkbits;
548 * If there is a delay extent at this offset,
549 * we will skip this extent.
551 ext4_es_find_delayed_extent_range(inode, last, last, &es);
552 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
553 last = es.es_lblk + es.es_len;
554 holeoff = (loff_t)last << blkbits;
559 * If there is a unwritten extent at this offset,
560 * it will be as a data or a hole according to page
561 * cache that has data or not.
563 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
565 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
569 holeoff = (loff_t)last << blkbits;
576 } while (last <= end);
578 mutex_unlock(&inode->i_mutex);
583 return vfs_setpos(file, holeoff, maxsize);
587 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
588 * by calling generic_file_llseek_size() with the appropriate maxbytes
591 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
593 struct inode *inode = file->f_mapping->host;
596 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
597 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
599 maxbytes = inode->i_sb->s_maxbytes;
605 return generic_file_llseek_size(file, offset, whence,
606 maxbytes, i_size_read(inode));
608 return ext4_seek_data(file, offset, maxbytes);
610 return ext4_seek_hole(file, offset, maxbytes);
616 const struct file_operations ext4_file_operations = {
617 .llseek = ext4_llseek,
618 .read_iter = generic_file_read_iter,
619 .write_iter = ext4_file_write_iter,
620 .unlocked_ioctl = ext4_ioctl,
622 .compat_ioctl = ext4_compat_ioctl,
624 .mmap = ext4_file_mmap,
625 .open = ext4_file_open,
626 .release = ext4_release_file,
627 .fsync = ext4_sync_file,
628 .splice_read = generic_file_splice_read,
629 .splice_write = iter_file_splice_write,
630 .fallocate = ext4_fallocate,
634 const struct file_operations ext4_dax_file_operations = {
635 .llseek = ext4_llseek,
636 .read_iter = generic_file_read_iter,
637 .write_iter = ext4_file_write_iter,
638 .unlocked_ioctl = ext4_ioctl,
640 .compat_ioctl = ext4_compat_ioctl,
642 .mmap = ext4_file_mmap,
643 .open = ext4_file_open,
644 .release = ext4_release_file,
645 .fsync = ext4_sync_file,
646 /* Splice not yet supported with DAX */
647 .fallocate = ext4_fallocate,
651 const struct inode_operations ext4_file_inode_operations = {
652 .setattr = ext4_setattr,
653 .getattr = ext4_getattr,
654 .setxattr = generic_setxattr,
655 .getxattr = generic_getxattr,
656 .listxattr = ext4_listxattr,
657 .removexattr = generic_removexattr,
658 .get_acl = ext4_get_acl,
659 .set_acl = ext4_set_acl,
660 .fiemap = ext4_fiemap,