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);
136 if (pos + length > sbi->s_bitmap_maxbytes)
137 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
140 iocb->private = &overwrite;
142 blk_start_plug(&plug);
145 /* check whether we do a DIO overwrite or not */
146 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
147 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
148 struct ext4_map_blocks map;
149 unsigned int blkbits = inode->i_blkbits;
152 map.m_lblk = pos >> blkbits;
153 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
157 err = ext4_map_blocks(NULL, inode, &map, 0);
159 * 'err==len' means that all of blocks has
160 * been preallocated no matter they are
161 * initialized or not. For excluding
162 * unwritten extents, we need to check
163 * m_flags. There are two conditions that
164 * indicate for initialized extents. 1) If we
165 * hit extent cache, EXT4_MAP_MAPPED flag is
166 * returned; 2) If we do a real lookup,
167 * non-flags are returned. So we should check
168 * these two conditions.
170 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
175 ret = __generic_file_write_iter(iocb, from);
176 mutex_unlock(&inode->i_mutex);
181 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
186 blk_finish_plug(&plug);
190 mutex_unlock(aio_mutex);
195 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
197 return dax_fault(vma, vmf, ext4_get_block);
198 /* Is this the right get_block? */
201 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
203 return dax_mkwrite(vma, vmf, ext4_get_block);
206 static const struct vm_operations_struct ext4_dax_vm_ops = {
207 .fault = ext4_dax_fault,
208 .page_mkwrite = ext4_dax_mkwrite,
209 .pfn_mkwrite = dax_pfn_mkwrite,
212 #define ext4_dax_vm_ops ext4_file_vm_ops
215 static const struct vm_operations_struct ext4_file_vm_ops = {
216 .fault = filemap_fault,
217 .map_pages = filemap_map_pages,
218 .page_mkwrite = ext4_page_mkwrite,
221 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
224 if (IS_DAX(file_inode(file))) {
225 vma->vm_ops = &ext4_dax_vm_ops;
226 vma->vm_flags |= VM_MIXEDMAP;
228 vma->vm_ops = &ext4_file_vm_ops;
233 static int ext4_file_open(struct inode * inode, struct file * filp)
235 struct super_block *sb = inode->i_sb;
236 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
237 struct vfsmount *mnt = filp->f_path.mnt;
241 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
242 !(sb->s_flags & MS_RDONLY))) {
243 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
245 * Sample where the filesystem has been mounted and
246 * store it in the superblock for sysadmin convenience
247 * when trying to sort through large numbers of block
248 * devices or filesystem images.
250 memset(buf, 0, sizeof(buf));
252 path.dentry = mnt->mnt_root;
253 cp = d_path(&path, buf, sizeof(buf));
258 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
260 return PTR_ERR(handle);
261 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
262 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
264 ext4_journal_stop(handle);
267 strlcpy(sbi->s_es->s_last_mounted, cp,
268 sizeof(sbi->s_es->s_last_mounted));
269 ext4_handle_dirty_super(handle, sb);
270 ext4_journal_stop(handle);
274 * Set up the jbd2_inode if we are opening the inode for
275 * writing and the journal is present
277 if (filp->f_mode & FMODE_WRITE) {
278 int ret = ext4_inode_attach_jinode(inode);
282 return dquot_file_open(inode, filp);
286 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
287 * file rather than ext4_ext_walk_space() because we can introduce
288 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
289 * function. When extent status tree has been fully implemented, it will
290 * track all extent status for a file and we can directly use it to
291 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
295 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
296 * lookup page cache to check whether or not there has some data between
297 * [startoff, endoff] because, if this range contains an unwritten extent,
298 * we determine this extent as a data or a hole according to whether the
299 * page cache has data or not.
301 static int ext4_find_unwritten_pgoff(struct inode *inode,
303 struct ext4_map_blocks *map,
307 unsigned int blkbits;
315 blkbits = inode->i_sb->s_blocksize_bits;
318 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
320 index = startoff >> PAGE_CACHE_SHIFT;
321 end = endoff >> PAGE_CACHE_SHIFT;
323 pagevec_init(&pvec, 0);
326 unsigned long nr_pages;
328 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
329 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
332 if (whence == SEEK_DATA)
335 BUG_ON(whence != SEEK_HOLE);
337 * If this is the first time to go into the loop and
338 * offset is not beyond the end offset, it will be a
339 * hole at this offset
341 if (lastoff == startoff || lastoff < endoff)
347 * If this is the first time to go into the loop and
348 * offset is smaller than the first page offset, it will be a
349 * hole at this offset.
351 if (lastoff == startoff && whence == SEEK_HOLE &&
352 lastoff < page_offset(pvec.pages[0])) {
357 for (i = 0; i < nr_pages; i++) {
358 struct page *page = pvec.pages[i];
359 struct buffer_head *bh, *head;
362 * If the current offset is not beyond the end of given
363 * range, it will be a hole.
365 if (lastoff < endoff && whence == SEEK_HOLE &&
374 if (unlikely(page->mapping != inode->i_mapping)) {
379 if (!page_has_buffers(page)) {
384 if (page_has_buffers(page)) {
385 lastoff = page_offset(page);
386 bh = head = page_buffers(page);
388 if (buffer_uptodate(bh) ||
389 buffer_unwritten(bh)) {
390 if (whence == SEEK_DATA)
393 if (whence == SEEK_HOLE)
397 *offset = max_t(loff_t,
402 lastoff += bh->b_size;
403 bh = bh->b_this_page;
404 } while (bh != head);
407 lastoff = page_offset(page) + PAGE_SIZE;
412 * The no. of pages is less than our desired, that would be a
415 if (nr_pages < num && whence == SEEK_HOLE) {
421 index = pvec.pages[i - 1]->index + 1;
422 pagevec_release(&pvec);
423 } while (index <= end);
426 pagevec_release(&pvec);
431 * ext4_seek_data() retrieves the offset for SEEK_DATA.
433 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
435 struct inode *inode = file->f_mapping->host;
436 struct ext4_map_blocks map;
437 struct extent_status es;
438 ext4_lblk_t start, last, end;
439 loff_t dataoff, isize;
443 mutex_lock(&inode->i_mutex);
445 isize = i_size_read(inode);
446 if (offset >= isize) {
447 mutex_unlock(&inode->i_mutex);
451 blkbits = inode->i_sb->s_blocksize_bits;
452 start = offset >> blkbits;
454 end = isize >> blkbits;
459 map.m_len = end - last + 1;
460 ret = ext4_map_blocks(NULL, inode, &map, 0);
461 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
463 dataoff = (loff_t)last << blkbits;
468 * If there is a delay extent at this offset,
469 * it will be as a data.
471 ext4_es_find_delayed_extent_range(inode, last, last, &es);
472 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
474 dataoff = (loff_t)last << blkbits;
479 * If there is a unwritten extent at this offset,
480 * it will be as a data or a hole according to page
481 * cache that has data or not.
483 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
485 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
492 dataoff = (loff_t)last << blkbits;
493 } while (last <= end);
495 mutex_unlock(&inode->i_mutex);
500 return vfs_setpos(file, dataoff, maxsize);
504 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
506 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
508 struct inode *inode = file->f_mapping->host;
509 struct ext4_map_blocks map;
510 struct extent_status es;
511 ext4_lblk_t start, last, end;
512 loff_t holeoff, isize;
516 mutex_lock(&inode->i_mutex);
518 isize = i_size_read(inode);
519 if (offset >= isize) {
520 mutex_unlock(&inode->i_mutex);
524 blkbits = inode->i_sb->s_blocksize_bits;
525 start = offset >> blkbits;
527 end = isize >> blkbits;
532 map.m_len = end - last + 1;
533 ret = ext4_map_blocks(NULL, inode, &map, 0);
534 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
536 holeoff = (loff_t)last << blkbits;
541 * If there is a delay extent at this offset,
542 * we will skip this extent.
544 ext4_es_find_delayed_extent_range(inode, last, last, &es);
545 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
546 last = es.es_lblk + es.es_len;
547 holeoff = (loff_t)last << blkbits;
552 * If there is a unwritten extent at this offset,
553 * it will be as a data or a hole according to page
554 * cache that has data or not.
556 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
558 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
562 holeoff = (loff_t)last << blkbits;
569 } while (last <= end);
571 mutex_unlock(&inode->i_mutex);
576 return vfs_setpos(file, holeoff, maxsize);
580 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
581 * by calling generic_file_llseek_size() with the appropriate maxbytes
584 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
586 struct inode *inode = file->f_mapping->host;
589 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
590 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
592 maxbytes = inode->i_sb->s_maxbytes;
598 return generic_file_llseek_size(file, offset, whence,
599 maxbytes, i_size_read(inode));
601 return ext4_seek_data(file, offset, maxbytes);
603 return ext4_seek_hole(file, offset, maxbytes);
609 const struct file_operations ext4_file_operations = {
610 .llseek = ext4_llseek,
611 .read_iter = generic_file_read_iter,
612 .write_iter = ext4_file_write_iter,
613 .unlocked_ioctl = ext4_ioctl,
615 .compat_ioctl = ext4_compat_ioctl,
617 .mmap = ext4_file_mmap,
618 .open = ext4_file_open,
619 .release = ext4_release_file,
620 .fsync = ext4_sync_file,
621 .splice_read = generic_file_splice_read,
622 .splice_write = iter_file_splice_write,
623 .fallocate = ext4_fallocate,
626 const struct inode_operations ext4_file_inode_operations = {
627 .setattr = ext4_setattr,
628 .getattr = ext4_getattr,
629 .setxattr = generic_setxattr,
630 .getxattr = generic_getxattr,
631 .listxattr = ext4_listxattr,
632 .removexattr = generic_removexattr,
633 .get_acl = ext4_get_acl,
634 .set_acl = ext4_set_acl,
635 .fiemap = ext4_fiemap,