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/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>
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 = iocb->ki_flags & IOCB_DIRECT;
103 * Unaligned direct AIO must be serialized; see comment above
104 * In the case of O_APPEND, assume that we must always serialize
107 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
108 !is_sync_kiocb(iocb) &&
109 (iocb->ki_flags & IOCB_APPEND ||
110 ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
111 aio_mutex = ext4_aio_mutex(inode);
112 mutex_lock(aio_mutex);
113 ext4_unwritten_wait(inode);
117 ret = generic_write_checks(iocb, from);
122 * If we have encountered a bitmap-format file, the size limit
123 * is smaller than s_maxbytes, which is for extent-mapped files.
125 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
126 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
128 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
132 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
135 iocb->private = &overwrite;
137 size_t length = iov_iter_count(from);
138 loff_t pos = iocb->ki_pos;
139 blk_start_plug(&plug);
141 /* check whether we do a DIO overwrite or not */
142 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
143 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
144 struct ext4_map_blocks map;
145 unsigned int blkbits = inode->i_blkbits;
148 map.m_lblk = pos >> blkbits;
149 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
153 err = ext4_map_blocks(NULL, inode, &map, 0);
155 * 'err==len' means that all of blocks has
156 * been preallocated no matter they are
157 * initialized or not. For excluding
158 * unwritten extents, we need to check
159 * m_flags. There are two conditions that
160 * indicate for initialized extents. 1) If we
161 * hit extent cache, EXT4_MAP_MAPPED flag is
162 * returned; 2) If we do a real lookup,
163 * non-flags are returned. So we should check
164 * these two conditions.
166 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
171 ret = __generic_file_write_iter(iocb, from);
177 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
182 blk_finish_plug(&plug);
185 mutex_unlock(aio_mutex);
191 mutex_unlock(aio_mutex);
196 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
199 handle_t *handle = NULL;
200 struct inode *inode = file_inode(vma->vm_file);
201 struct super_block *sb = inode->i_sb;
202 bool write = vmf->flags & FAULT_FLAG_WRITE;
205 sb_start_pagefault(sb);
206 file_update_time(vma->vm_file);
207 down_read(&EXT4_I(inode)->i_mmap_sem);
208 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
209 EXT4_DATA_TRANS_BLOCKS(sb));
211 down_read(&EXT4_I(inode)->i_mmap_sem);
214 result = VM_FAULT_SIGBUS;
216 result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL);
220 ext4_journal_stop(handle);
221 up_read(&EXT4_I(inode)->i_mmap_sem);
222 sb_end_pagefault(sb);
224 up_read(&EXT4_I(inode)->i_mmap_sem);
229 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
230 pmd_t *pmd, unsigned int flags)
233 handle_t *handle = NULL;
234 struct inode *inode = file_inode(vma->vm_file);
235 struct super_block *sb = inode->i_sb;
236 bool write = flags & FAULT_FLAG_WRITE;
239 sb_start_pagefault(sb);
240 file_update_time(vma->vm_file);
241 down_read(&EXT4_I(inode)->i_mmap_sem);
242 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
243 ext4_chunk_trans_blocks(inode,
244 PMD_SIZE / PAGE_SIZE));
246 down_read(&EXT4_I(inode)->i_mmap_sem);
249 result = VM_FAULT_SIGBUS;
251 result = __dax_pmd_fault(vma, addr, pmd, flags,
252 ext4_dax_mmap_get_block, NULL);
256 ext4_journal_stop(handle);
257 up_read(&EXT4_I(inode)->i_mmap_sem);
258 sb_end_pagefault(sb);
260 up_read(&EXT4_I(inode)->i_mmap_sem);
265 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
268 struct inode *inode = file_inode(vma->vm_file);
270 sb_start_pagefault(inode->i_sb);
271 file_update_time(vma->vm_file);
272 down_read(&EXT4_I(inode)->i_mmap_sem);
273 err = __dax_mkwrite(vma, vmf, ext4_dax_mmap_get_block, NULL);
274 up_read(&EXT4_I(inode)->i_mmap_sem);
275 sb_end_pagefault(inode->i_sb);
281 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite()
282 * handler we check for races agaist truncate. Note that since we cycle through
283 * i_mmap_sem, we are sure that also any hole punching that began before we
284 * were called is finished by now and so if it included part of the file we
285 * are working on, our pte will get unmapped and the check for pte_same() in
286 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
289 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
290 struct vm_fault *vmf)
292 struct inode *inode = file_inode(vma->vm_file);
293 struct super_block *sb = inode->i_sb;
297 sb_start_pagefault(sb);
298 file_update_time(vma->vm_file);
299 down_read(&EXT4_I(inode)->i_mmap_sem);
300 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
301 if (vmf->pgoff >= size)
302 ret = VM_FAULT_SIGBUS;
304 ret = dax_pfn_mkwrite(vma, vmf);
305 up_read(&EXT4_I(inode)->i_mmap_sem);
306 sb_end_pagefault(sb);
311 static const struct vm_operations_struct ext4_dax_vm_ops = {
312 .fault = ext4_dax_fault,
313 .pmd_fault = ext4_dax_pmd_fault,
314 .page_mkwrite = ext4_dax_mkwrite,
315 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
318 #define ext4_dax_vm_ops ext4_file_vm_ops
321 static const struct vm_operations_struct ext4_file_vm_ops = {
322 .fault = ext4_filemap_fault,
323 .map_pages = filemap_map_pages,
324 .page_mkwrite = ext4_page_mkwrite,
327 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
329 struct inode *inode = file->f_mapping->host;
331 if (ext4_encrypted_inode(inode)) {
332 int err = ext4_get_encryption_info(inode);
335 if (ext4_encryption_info(inode) == NULL)
339 if (IS_DAX(file_inode(file))) {
340 vma->vm_ops = &ext4_dax_vm_ops;
341 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
343 vma->vm_ops = &ext4_file_vm_ops;
348 static int ext4_file_open(struct inode * inode, struct file * filp)
350 struct super_block *sb = inode->i_sb;
351 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
352 struct vfsmount *mnt = filp->f_path.mnt;
353 struct inode *dir = filp->f_path.dentry->d_parent->d_inode;
358 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
359 !(sb->s_flags & MS_RDONLY))) {
360 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
362 * Sample where the filesystem has been mounted and
363 * store it in the superblock for sysadmin convenience
364 * when trying to sort through large numbers of block
365 * devices or filesystem images.
367 memset(buf, 0, sizeof(buf));
369 path.dentry = mnt->mnt_root;
370 cp = d_path(&path, buf, sizeof(buf));
375 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
377 return PTR_ERR(handle);
378 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
379 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
381 ext4_journal_stop(handle);
384 strlcpy(sbi->s_es->s_last_mounted, cp,
385 sizeof(sbi->s_es->s_last_mounted));
386 ext4_handle_dirty_super(handle, sb);
387 ext4_journal_stop(handle);
390 if (ext4_encrypted_inode(inode)) {
391 ret = ext4_get_encryption_info(inode);
394 if (ext4_encryption_info(inode) == NULL)
397 if (ext4_encrypted_inode(dir) &&
398 !ext4_is_child_context_consistent_with_parent(dir, inode)) {
399 ext4_warning(inode->i_sb,
400 "Inconsistent encryption contexts: %lu/%lu\n",
401 (unsigned long) dir->i_ino,
402 (unsigned long) inode->i_ino);
406 * Set up the jbd2_inode if we are opening the inode for
407 * writing and the journal is present
409 if (filp->f_mode & FMODE_WRITE) {
410 ret = ext4_inode_attach_jinode(inode);
414 return dquot_file_open(inode, filp);
418 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
419 * file rather than ext4_ext_walk_space() because we can introduce
420 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
421 * function. When extent status tree has been fully implemented, it will
422 * track all extent status for a file and we can directly use it to
423 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
427 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
428 * lookup page cache to check whether or not there has some data between
429 * [startoff, endoff] because, if this range contains an unwritten extent,
430 * we determine this extent as a data or a hole according to whether the
431 * page cache has data or not.
433 static int ext4_find_unwritten_pgoff(struct inode *inode,
435 struct ext4_map_blocks *map,
439 unsigned int blkbits;
447 blkbits = inode->i_sb->s_blocksize_bits;
450 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
452 index = startoff >> PAGE_CACHE_SHIFT;
453 end = endoff >> PAGE_CACHE_SHIFT;
455 pagevec_init(&pvec, 0);
458 unsigned long nr_pages;
460 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
461 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
464 if (whence == SEEK_DATA)
467 BUG_ON(whence != SEEK_HOLE);
469 * If this is the first time to go into the loop and
470 * offset is not beyond the end offset, it will be a
471 * hole at this offset
473 if (lastoff == startoff || lastoff < endoff)
479 * If this is the first time to go into the loop and
480 * offset is smaller than the first page offset, it will be a
481 * hole at this offset.
483 if (lastoff == startoff && whence == SEEK_HOLE &&
484 lastoff < page_offset(pvec.pages[0])) {
489 for (i = 0; i < nr_pages; i++) {
490 struct page *page = pvec.pages[i];
491 struct buffer_head *bh, *head;
494 * If the current offset is not beyond the end of given
495 * range, it will be a hole.
497 if (lastoff < endoff && whence == SEEK_HOLE &&
506 if (unlikely(page->mapping != inode->i_mapping)) {
511 if (!page_has_buffers(page)) {
516 if (page_has_buffers(page)) {
517 lastoff = page_offset(page);
518 bh = head = page_buffers(page);
520 if (buffer_uptodate(bh) ||
521 buffer_unwritten(bh)) {
522 if (whence == SEEK_DATA)
525 if (whence == SEEK_HOLE)
529 *offset = max_t(loff_t,
534 lastoff += bh->b_size;
535 bh = bh->b_this_page;
536 } while (bh != head);
539 lastoff = page_offset(page) + PAGE_SIZE;
544 * The no. of pages is less than our desired, that would be a
547 if (nr_pages < num && whence == SEEK_HOLE) {
553 index = pvec.pages[i - 1]->index + 1;
554 pagevec_release(&pvec);
555 } while (index <= end);
558 pagevec_release(&pvec);
563 * ext4_seek_data() retrieves the offset for SEEK_DATA.
565 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
567 struct inode *inode = file->f_mapping->host;
568 struct ext4_map_blocks map;
569 struct extent_status es;
570 ext4_lblk_t start, last, end;
571 loff_t dataoff, isize;
577 isize = i_size_read(inode);
578 if (offset >= isize) {
583 blkbits = inode->i_sb->s_blocksize_bits;
584 start = offset >> blkbits;
586 end = isize >> blkbits;
591 map.m_len = end - last + 1;
592 ret = ext4_map_blocks(NULL, inode, &map, 0);
593 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
595 dataoff = (loff_t)last << blkbits;
600 * If there is a delay extent at this offset,
601 * it will be as a data.
603 ext4_es_find_delayed_extent_range(inode, last, last, &es);
604 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
606 dataoff = (loff_t)last << blkbits;
611 * If there is a unwritten extent at this offset,
612 * it will be as a data or a hole according to page
613 * cache that has data or not.
615 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
617 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
624 dataoff = (loff_t)last << blkbits;
625 } while (last <= end);
632 return vfs_setpos(file, dataoff, maxsize);
636 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
638 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
640 struct inode *inode = file->f_mapping->host;
641 struct ext4_map_blocks map;
642 struct extent_status es;
643 ext4_lblk_t start, last, end;
644 loff_t holeoff, isize;
650 isize = i_size_read(inode);
651 if (offset >= isize) {
656 blkbits = inode->i_sb->s_blocksize_bits;
657 start = offset >> blkbits;
659 end = isize >> blkbits;
664 map.m_len = end - last + 1;
665 ret = ext4_map_blocks(NULL, inode, &map, 0);
666 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
668 holeoff = (loff_t)last << blkbits;
673 * If there is a delay extent at this offset,
674 * we will skip this extent.
676 ext4_es_find_delayed_extent_range(inode, last, last, &es);
677 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
678 last = es.es_lblk + es.es_len;
679 holeoff = (loff_t)last << blkbits;
684 * If there is a unwritten extent at this offset,
685 * it will be as a data or a hole according to page
686 * cache that has data or not.
688 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
690 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
694 holeoff = (loff_t)last << blkbits;
701 } while (last <= end);
708 return vfs_setpos(file, holeoff, maxsize);
712 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
713 * by calling generic_file_llseek_size() with the appropriate maxbytes
716 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
718 struct inode *inode = file->f_mapping->host;
721 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
722 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
724 maxbytes = inode->i_sb->s_maxbytes;
730 return generic_file_llseek_size(file, offset, whence,
731 maxbytes, i_size_read(inode));
733 return ext4_seek_data(file, offset, maxbytes);
735 return ext4_seek_hole(file, offset, maxbytes);
741 const struct file_operations ext4_file_operations = {
742 .llseek = ext4_llseek,
743 .read_iter = generic_file_read_iter,
744 .write_iter = ext4_file_write_iter,
745 .unlocked_ioctl = ext4_ioctl,
747 .compat_ioctl = ext4_compat_ioctl,
749 .mmap = ext4_file_mmap,
750 .open = ext4_file_open,
751 .release = ext4_release_file,
752 .fsync = ext4_sync_file,
753 .splice_read = generic_file_splice_read,
754 .splice_write = iter_file_splice_write,
755 .fallocate = ext4_fallocate,
758 const struct inode_operations ext4_file_inode_operations = {
759 .setattr = ext4_setattr,
760 .getattr = ext4_getattr,
761 .setxattr = generic_setxattr,
762 .getxattr = generic_getxattr,
763 .listxattr = ext4_listxattr,
764 .removexattr = generic_removexattr,
765 .get_acl = ext4_get_acl,
766 .set_acl = ext4_set_acl,
767 .fiemap = ext4_fiemap,