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 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
37 struct inode *inode = file_inode(iocb->ki_filp);
40 inode_lock_shared(inode);
42 * Recheck under inode lock - at this point we are sure it cannot
46 inode_unlock_shared(inode);
47 /* Fallback to buffered IO in case we cannot support DAX */
48 return generic_file_read_iter(iocb, to);
50 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
51 inode_unlock_shared(inode);
53 file_accessed(iocb->ki_filp);
58 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
60 if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
63 if (!iov_iter_count(to))
64 return 0; /* skip atime */
67 if (IS_DAX(file_inode(iocb->ki_filp)))
68 return ext4_dax_read_iter(iocb, to);
70 return generic_file_read_iter(iocb, to);
74 * Called when an inode is released. Note that this is different
75 * from ext4_file_open: open gets called at every open, but release
76 * gets called only when /all/ the files are closed.
78 static int ext4_release_file(struct inode *inode, struct file *filp)
80 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
81 ext4_alloc_da_blocks(inode);
82 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
84 /* if we are the last writer on the inode, drop the block reservation */
85 if ((filp->f_mode & FMODE_WRITE) &&
86 (atomic_read(&inode->i_writecount) == 1) &&
87 !EXT4_I(inode)->i_reserved_data_blocks)
89 down_write(&EXT4_I(inode)->i_data_sem);
90 ext4_discard_preallocations(inode);
91 up_write(&EXT4_I(inode)->i_data_sem);
93 if (is_dx(inode) && filp->private_data)
94 ext4_htree_free_dir_info(filp->private_data);
99 static void ext4_unwritten_wait(struct inode *inode)
101 wait_queue_head_t *wq = ext4_ioend_wq(inode);
103 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
107 * This tests whether the IO in question is block-aligned or not.
108 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
109 * are converted to written only after the IO is complete. Until they are
110 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
111 * it needs to zero out portions of the start and/or end block. If 2 AIO
112 * threads are at work on the same unwritten block, they must be synchronized
113 * or one thread will zero the other's data, causing corruption.
116 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
118 struct super_block *sb = inode->i_sb;
119 int blockmask = sb->s_blocksize - 1;
121 if (pos >= i_size_read(inode))
124 if ((pos | iov_iter_alignment(from)) & blockmask)
130 /* Is IO overwriting allocated and initialized blocks? */
131 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
133 struct ext4_map_blocks map;
134 unsigned int blkbits = inode->i_blkbits;
137 if (pos + len > i_size_read(inode))
140 map.m_lblk = pos >> blkbits;
141 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
144 err = ext4_map_blocks(NULL, inode, &map, 0);
146 * 'err==len' means that all of the blocks have been preallocated,
147 * regardless of whether they have been initialized or not. To exclude
148 * unwritten extents, we need to check m_flags.
150 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
153 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
155 struct inode *inode = file_inode(iocb->ki_filp);
158 ret = generic_write_checks(iocb, from);
162 * If we have encountered a bitmap-format file, the size limit
163 * is smaller than s_maxbytes, which is for extent-mapped files.
165 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
166 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
168 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
170 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
172 return iov_iter_count(from);
177 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
179 struct inode *inode = file_inode(iocb->ki_filp);
181 bool overwrite = false;
184 ret = ext4_write_checks(iocb, from);
187 ret = file_remove_privs(iocb->ki_filp);
190 ret = file_update_time(iocb->ki_filp);
194 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
196 downgrade_write(&inode->i_rwsem);
198 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
203 inode_unlock_shared(inode);
205 ret = generic_write_sync(iocb, ret);
211 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
213 struct inode *inode = file_inode(iocb->ki_filp);
214 int o_direct = iocb->ki_flags & IOCB_DIRECT;
215 int unaligned_aio = 0;
219 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
224 return ext4_dax_write_iter(iocb, from);
228 ret = ext4_write_checks(iocb, from);
233 * Unaligned direct AIO must be serialized among each other as zeroing
234 * of partial blocks of two competing unaligned AIOs can result in data
237 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
238 !is_sync_kiocb(iocb) &&
239 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
241 ext4_unwritten_wait(inode);
244 iocb->private = &overwrite;
245 /* Check whether we do a DIO overwrite or not */
246 if (o_direct && ext4_should_dioread_nolock(inode) && !unaligned_aio &&
247 ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from)))
250 ret = __generic_file_write_iter(iocb, from);
254 ret = generic_write_sync(iocb, ret);
264 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
267 struct inode *inode = file_inode(vma->vm_file);
268 struct super_block *sb = inode->i_sb;
269 bool write = vmf->flags & FAULT_FLAG_WRITE;
272 sb_start_pagefault(sb);
273 file_update_time(vma->vm_file);
275 down_read(&EXT4_I(inode)->i_mmap_sem);
276 result = dax_iomap_fault(vma, vmf, &ext4_iomap_ops);
277 up_read(&EXT4_I(inode)->i_mmap_sem);
279 sb_end_pagefault(sb);
284 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
285 pmd_t *pmd, unsigned int flags)
288 struct inode *inode = file_inode(vma->vm_file);
289 struct super_block *sb = inode->i_sb;
290 bool write = flags & FAULT_FLAG_WRITE;
293 sb_start_pagefault(sb);
294 file_update_time(vma->vm_file);
296 down_read(&EXT4_I(inode)->i_mmap_sem);
297 result = dax_iomap_pmd_fault(vma, addr, pmd, flags,
299 up_read(&EXT4_I(inode)->i_mmap_sem);
301 sb_end_pagefault(sb);
307 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
308 * handler we check for races agaist truncate. Note that since we cycle through
309 * i_mmap_sem, we are sure that also any hole punching that began before we
310 * were called is finished by now and so if it included part of the file we
311 * are working on, our pte will get unmapped and the check for pte_same() in
312 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
315 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
316 struct vm_fault *vmf)
318 struct inode *inode = file_inode(vma->vm_file);
319 struct super_block *sb = inode->i_sb;
323 sb_start_pagefault(sb);
324 file_update_time(vma->vm_file);
325 down_read(&EXT4_I(inode)->i_mmap_sem);
326 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
327 if (vmf->pgoff >= size)
328 ret = VM_FAULT_SIGBUS;
330 ret = dax_pfn_mkwrite(vma, vmf);
331 up_read(&EXT4_I(inode)->i_mmap_sem);
332 sb_end_pagefault(sb);
337 static const struct vm_operations_struct ext4_dax_vm_ops = {
338 .fault = ext4_dax_fault,
339 .pmd_fault = ext4_dax_pmd_fault,
340 .page_mkwrite = ext4_dax_fault,
341 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
344 #define ext4_dax_vm_ops ext4_file_vm_ops
347 static const struct vm_operations_struct ext4_file_vm_ops = {
348 .fault = ext4_filemap_fault,
349 .map_pages = filemap_map_pages,
350 .page_mkwrite = ext4_page_mkwrite,
353 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
355 struct inode *inode = file->f_mapping->host;
357 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
360 if (ext4_encrypted_inode(inode)) {
361 int err = fscrypt_get_encryption_info(inode);
364 if (!fscrypt_has_encryption_key(inode))
368 if (IS_DAX(file_inode(file))) {
369 vma->vm_ops = &ext4_dax_vm_ops;
370 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
372 vma->vm_ops = &ext4_file_vm_ops;
377 static int ext4_file_open(struct inode * inode, struct file * filp)
379 struct super_block *sb = inode->i_sb;
380 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
381 struct vfsmount *mnt = filp->f_path.mnt;
387 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
390 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
391 !(sb->s_flags & MS_RDONLY))) {
392 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
394 * Sample where the filesystem has been mounted and
395 * store it in the superblock for sysadmin convenience
396 * when trying to sort through large numbers of block
397 * devices or filesystem images.
399 memset(buf, 0, sizeof(buf));
401 path.dentry = mnt->mnt_root;
402 cp = d_path(&path, buf, sizeof(buf));
407 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
409 return PTR_ERR(handle);
410 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
411 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
413 ext4_journal_stop(handle);
416 strlcpy(sbi->s_es->s_last_mounted, cp,
417 sizeof(sbi->s_es->s_last_mounted));
418 ext4_handle_dirty_super(handle, sb);
419 ext4_journal_stop(handle);
422 if (ext4_encrypted_inode(inode)) {
423 ret = fscrypt_get_encryption_info(inode);
426 if (!fscrypt_has_encryption_key(inode))
430 dir = dget_parent(file_dentry(filp));
431 if (ext4_encrypted_inode(d_inode(dir)) &&
432 !fscrypt_has_permitted_context(d_inode(dir), inode)) {
433 ext4_warning(inode->i_sb,
434 "Inconsistent encryption contexts: %lu/%lu",
435 (unsigned long) d_inode(dir)->i_ino,
436 (unsigned long) inode->i_ino);
442 * Set up the jbd2_inode if we are opening the inode for
443 * writing and the journal is present
445 if (filp->f_mode & FMODE_WRITE) {
446 ret = ext4_inode_attach_jinode(inode);
450 return dquot_file_open(inode, filp);
454 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
455 * file rather than ext4_ext_walk_space() because we can introduce
456 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
457 * function. When extent status tree has been fully implemented, it will
458 * track all extent status for a file and we can directly use it to
459 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
463 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
464 * lookup page cache to check whether or not there has some data between
465 * [startoff, endoff] because, if this range contains an unwritten extent,
466 * we determine this extent as a data or a hole according to whether the
467 * page cache has data or not.
469 static int ext4_find_unwritten_pgoff(struct inode *inode,
475 unsigned int blkbits;
483 blkbits = inode->i_sb->s_blocksize_bits;
486 endoff = (loff_t)end_blk << blkbits;
488 index = startoff >> PAGE_SHIFT;
489 end = endoff >> PAGE_SHIFT;
491 pagevec_init(&pvec, 0);
494 unsigned long nr_pages;
496 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
497 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
500 if (whence == SEEK_DATA)
503 BUG_ON(whence != SEEK_HOLE);
505 * If this is the first time to go into the loop and
506 * offset is not beyond the end offset, it will be a
507 * hole at this offset
509 if (lastoff == startoff || lastoff < endoff)
515 * If this is the first time to go into the loop and
516 * offset is smaller than the first page offset, it will be a
517 * hole at this offset.
519 if (lastoff == startoff && whence == SEEK_HOLE &&
520 lastoff < page_offset(pvec.pages[0])) {
525 for (i = 0; i < nr_pages; i++) {
526 struct page *page = pvec.pages[i];
527 struct buffer_head *bh, *head;
530 * If the current offset is not beyond the end of given
531 * range, it will be a hole.
533 if (lastoff < endoff && whence == SEEK_HOLE &&
542 if (unlikely(page->mapping != inode->i_mapping)) {
547 if (!page_has_buffers(page)) {
552 if (page_has_buffers(page)) {
553 lastoff = page_offset(page);
554 bh = head = page_buffers(page);
556 if (buffer_uptodate(bh) ||
557 buffer_unwritten(bh)) {
558 if (whence == SEEK_DATA)
561 if (whence == SEEK_HOLE)
565 *offset = max_t(loff_t,
570 lastoff += bh->b_size;
571 bh = bh->b_this_page;
572 } while (bh != head);
575 lastoff = page_offset(page) + PAGE_SIZE;
580 * The no. of pages is less than our desired, that would be a
583 if (nr_pages < num && whence == SEEK_HOLE) {
589 index = pvec.pages[i - 1]->index + 1;
590 pagevec_release(&pvec);
591 } while (index <= end);
594 pagevec_release(&pvec);
599 * ext4_seek_data() retrieves the offset for SEEK_DATA.
601 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
603 struct inode *inode = file->f_mapping->host;
604 struct extent_status es;
605 ext4_lblk_t start, last, end;
606 loff_t dataoff, isize;
612 isize = i_size_read(inode);
613 if (offset >= isize) {
618 blkbits = inode->i_sb->s_blocksize_bits;
619 start = offset >> blkbits;
621 end = isize >> blkbits;
625 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
627 /* No extent found -> no data */
636 dataoff = (loff_t)last << blkbits;
637 if (!ext4_es_is_unwritten(&es))
641 * If there is a unwritten extent at this offset,
642 * it will be as a data or a hole according to page
643 * cache that has data or not.
645 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
646 es.es_lblk + es.es_len, &dataoff))
649 dataoff = (loff_t)last << blkbits;
651 } while (last <= end);
658 return vfs_setpos(file, dataoff, maxsize);
662 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
664 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
666 struct inode *inode = file->f_mapping->host;
667 struct extent_status es;
668 ext4_lblk_t start, last, end;
669 loff_t holeoff, isize;
675 isize = i_size_read(inode);
676 if (offset >= isize) {
681 blkbits = inode->i_sb->s_blocksize_bits;
682 start = offset >> blkbits;
684 end = isize >> blkbits;
688 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
694 if (ret == 0 || es.es_lblk > last) {
696 holeoff = (loff_t)last << blkbits;
700 * If there is a unwritten extent at this offset,
701 * it will be as a data or a hole according to page
702 * cache that has data or not.
704 if (ext4_es_is_unwritten(&es) &&
705 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
706 last + es.es_len, &holeoff))
710 holeoff = (loff_t)last << blkbits;
712 } while (last <= end);
719 return vfs_setpos(file, holeoff, maxsize);
723 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
724 * by calling generic_file_llseek_size() with the appropriate maxbytes
727 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
729 struct inode *inode = file->f_mapping->host;
732 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
733 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
735 maxbytes = inode->i_sb->s_maxbytes;
741 return generic_file_llseek_size(file, offset, whence,
742 maxbytes, i_size_read(inode));
744 return ext4_seek_data(file, offset, maxbytes);
746 return ext4_seek_hole(file, offset, maxbytes);
752 const struct file_operations ext4_file_operations = {
753 .llseek = ext4_llseek,
754 .read_iter = ext4_file_read_iter,
755 .write_iter = ext4_file_write_iter,
756 .unlocked_ioctl = ext4_ioctl,
758 .compat_ioctl = ext4_compat_ioctl,
760 .mmap = ext4_file_mmap,
761 .open = ext4_file_open,
762 .release = ext4_release_file,
763 .fsync = ext4_sync_file,
764 .get_unmapped_area = thp_get_unmapped_area,
765 .splice_read = generic_file_splice_read,
766 .splice_write = iter_file_splice_write,
767 .fallocate = ext4_fallocate,
770 const struct inode_operations ext4_file_inode_operations = {
771 .setattr = ext4_setattr,
772 .getattr = ext4_getattr,
773 .listxattr = ext4_listxattr,
774 .get_acl = ext4_get_acl,
775 .set_acl = ext4_set_acl,
776 .fiemap = ext4_fiemap,