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 if (!inode_trylock_shared(inode)) {
41 if (iocb->ki_flags & IOCB_NOWAIT)
43 inode_lock_shared(inode);
46 * Recheck under inode lock - at this point we are sure it cannot
50 inode_unlock_shared(inode);
51 /* Fallback to buffered IO in case we cannot support DAX */
52 return generic_file_read_iter(iocb, to);
54 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
55 inode_unlock_shared(inode);
57 file_accessed(iocb->ki_filp);
62 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
64 if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
67 if (!iov_iter_count(to))
68 return 0; /* skip atime */
71 if (IS_DAX(file_inode(iocb->ki_filp)))
72 return ext4_dax_read_iter(iocb, to);
74 return generic_file_read_iter(iocb, to);
78 * Called when an inode is released. Note that this is different
79 * from ext4_file_open: open gets called at every open, but release
80 * gets called only when /all/ the files are closed.
82 static int ext4_release_file(struct inode *inode, struct file *filp)
84 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
85 ext4_alloc_da_blocks(inode);
86 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
88 /* if we are the last writer on the inode, drop the block reservation */
89 if ((filp->f_mode & FMODE_WRITE) &&
90 (atomic_read(&inode->i_writecount) == 1) &&
91 !EXT4_I(inode)->i_reserved_data_blocks)
93 down_write(&EXT4_I(inode)->i_data_sem);
94 ext4_discard_preallocations(inode);
95 up_write(&EXT4_I(inode)->i_data_sem);
97 if (is_dx(inode) && filp->private_data)
98 ext4_htree_free_dir_info(filp->private_data);
103 static void ext4_unwritten_wait(struct inode *inode)
105 wait_queue_head_t *wq = ext4_ioend_wq(inode);
107 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
111 * This tests whether the IO in question is block-aligned or not.
112 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
113 * are converted to written only after the IO is complete. Until they are
114 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
115 * it needs to zero out portions of the start and/or end block. If 2 AIO
116 * threads are at work on the same unwritten block, they must be synchronized
117 * or one thread will zero the other's data, causing corruption.
120 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
122 struct super_block *sb = inode->i_sb;
123 int blockmask = sb->s_blocksize - 1;
125 if (pos >= i_size_read(inode))
128 if ((pos | iov_iter_alignment(from)) & blockmask)
134 /* Is IO overwriting allocated and initialized blocks? */
135 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
137 struct ext4_map_blocks map;
138 unsigned int blkbits = inode->i_blkbits;
141 if (pos + len > i_size_read(inode))
144 map.m_lblk = pos >> blkbits;
145 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
148 err = ext4_map_blocks(NULL, inode, &map, 0);
150 * 'err==len' means that all of the blocks have been preallocated,
151 * regardless of whether they have been initialized or not. To exclude
152 * unwritten extents, we need to check m_flags.
154 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
157 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
159 struct inode *inode = file_inode(iocb->ki_filp);
162 ret = generic_write_checks(iocb, from);
166 * If we have encountered a bitmap-format file, the size limit
167 * is smaller than s_maxbytes, which is for extent-mapped files.
169 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
170 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
172 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
174 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
176 return iov_iter_count(from);
181 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
183 struct inode *inode = file_inode(iocb->ki_filp);
186 if (!inode_trylock(inode)) {
187 if (iocb->ki_flags & IOCB_NOWAIT)
191 ret = ext4_write_checks(iocb, from);
194 ret = file_remove_privs(iocb->ki_filp);
197 ret = file_update_time(iocb->ki_filp);
201 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
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);
227 if (!inode_trylock(inode)) {
228 if (iocb->ki_flags & IOCB_NOWAIT)
233 ret = ext4_write_checks(iocb, from);
238 * Unaligned direct AIO must be serialized among each other as zeroing
239 * of partial blocks of two competing unaligned AIOs can result in data
242 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
243 !is_sync_kiocb(iocb) &&
244 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
246 ext4_unwritten_wait(inode);
249 iocb->private = &overwrite;
250 /* Check whether we do a DIO overwrite or not */
251 if (o_direct && !unaligned_aio) {
252 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
253 if (ext4_should_dioread_nolock(inode))
255 } else if (iocb->ki_flags & IOCB_NOWAIT) {
261 ret = __generic_file_write_iter(iocb, from);
265 ret = generic_write_sync(iocb, ret);
275 static int ext4_dax_huge_fault(struct vm_fault *vmf,
276 enum page_entry_size pe_size)
279 handle_t *handle = NULL;
280 struct inode *inode = file_inode(vmf->vma->vm_file);
281 struct super_block *sb = inode->i_sb;
282 bool write = vmf->flags & FAULT_FLAG_WRITE;
285 sb_start_pagefault(sb);
286 file_update_time(vmf->vma->vm_file);
287 down_read(&EXT4_I(inode)->i_mmap_sem);
288 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
289 EXT4_DATA_TRANS_BLOCKS(sb));
291 down_read(&EXT4_I(inode)->i_mmap_sem);
294 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
296 result = VM_FAULT_SIGBUS;
299 ext4_journal_stop(handle);
300 up_read(&EXT4_I(inode)->i_mmap_sem);
301 sb_end_pagefault(sb);
303 up_read(&EXT4_I(inode)->i_mmap_sem);
309 static int ext4_dax_fault(struct vm_fault *vmf)
311 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
315 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
316 * handler we check for races agaist truncate. Note that since we cycle through
317 * i_mmap_sem, we are sure that also any hole punching that began before we
318 * were called is finished by now and so if it included part of the file we
319 * are working on, our pte will get unmapped and the check for pte_same() in
320 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
323 static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf)
325 struct inode *inode = file_inode(vmf->vma->vm_file);
326 struct super_block *sb = inode->i_sb;
330 sb_start_pagefault(sb);
331 file_update_time(vmf->vma->vm_file);
332 down_read(&EXT4_I(inode)->i_mmap_sem);
333 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 if (vmf->pgoff >= size)
335 ret = VM_FAULT_SIGBUS;
337 ret = dax_pfn_mkwrite(vmf);
338 up_read(&EXT4_I(inode)->i_mmap_sem);
339 sb_end_pagefault(sb);
344 static const struct vm_operations_struct ext4_dax_vm_ops = {
345 .fault = ext4_dax_fault,
346 .huge_fault = ext4_dax_huge_fault,
347 .page_mkwrite = ext4_dax_fault,
348 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
351 #define ext4_dax_vm_ops ext4_file_vm_ops
354 static const struct vm_operations_struct ext4_file_vm_ops = {
355 .fault = ext4_filemap_fault,
356 .map_pages = filemap_map_pages,
357 .page_mkwrite = ext4_page_mkwrite,
360 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
362 struct inode *inode = file->f_mapping->host;
364 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
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);
451 /* Set the flags to support nowait AIO */
452 filp->f_mode |= FMODE_AIO_NOWAIT;
454 return dquot_file_open(inode, filp);
458 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
459 * file rather than ext4_ext_walk_space() because we can introduce
460 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
461 * function. When extent status tree has been fully implemented, it will
462 * track all extent status for a file and we can directly use it to
463 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
467 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
468 * lookup page cache to check whether or not there has some data between
469 * [startoff, endoff] because, if this range contains an unwritten extent,
470 * we determine this extent as a data or a hole according to whether the
471 * page cache has data or not.
473 static int ext4_find_unwritten_pgoff(struct inode *inode,
479 unsigned int blkbits;
487 blkbits = inode->i_sb->s_blocksize_bits;
490 endoff = (loff_t)end_blk << blkbits;
492 index = startoff >> PAGE_SHIFT;
493 end = (endoff - 1) >> PAGE_SHIFT;
495 pagevec_init(&pvec, 0);
498 unsigned long nr_pages;
500 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1;
501 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
506 for (i = 0; i < nr_pages; i++) {
507 struct page *page = pvec.pages[i];
508 struct buffer_head *bh, *head;
511 * If current offset is smaller than the page offset,
512 * there is a hole at this offset.
514 if (whence == SEEK_HOLE && lastoff < endoff &&
515 lastoff < page_offset(pvec.pages[i])) {
521 if (page->index > end)
526 if (unlikely(page->mapping != inode->i_mapping)) {
531 if (!page_has_buffers(page)) {
536 if (page_has_buffers(page)) {
537 lastoff = page_offset(page);
538 bh = head = page_buffers(page);
540 if (buffer_uptodate(bh) ||
541 buffer_unwritten(bh)) {
542 if (whence == SEEK_DATA)
545 if (whence == SEEK_HOLE)
549 *offset = max_t(loff_t,
554 lastoff += bh->b_size;
555 bh = bh->b_this_page;
556 } while (bh != head);
559 lastoff = page_offset(page) + PAGE_SIZE;
563 /* The no. of pages is less than our desired, we are done. */
567 index = pvec.pages[i - 1]->index + 1;
568 pagevec_release(&pvec);
569 } while (index <= end);
571 if (whence == SEEK_HOLE && lastoff < endoff) {
576 pagevec_release(&pvec);
581 * ext4_seek_data() retrieves the offset for SEEK_DATA.
583 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
585 struct inode *inode = file->f_mapping->host;
586 struct extent_status es;
587 ext4_lblk_t start, last, end;
588 loff_t dataoff, isize;
594 isize = i_size_read(inode);
595 if (offset >= isize) {
600 blkbits = inode->i_sb->s_blocksize_bits;
601 start = offset >> blkbits;
603 end = isize >> blkbits;
607 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
609 /* No extent found -> no data */
618 dataoff = (loff_t)last << blkbits;
619 if (!ext4_es_is_unwritten(&es))
623 * If there is a unwritten extent at this offset,
624 * it will be as a data or a hole according to page
625 * cache that has data or not.
627 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
628 es.es_lblk + es.es_len, &dataoff))
631 dataoff = (loff_t)last << blkbits;
633 } while (last <= end);
640 return vfs_setpos(file, dataoff, maxsize);
644 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
646 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
648 struct inode *inode = file->f_mapping->host;
649 struct extent_status es;
650 ext4_lblk_t start, last, end;
651 loff_t holeoff, isize;
657 isize = i_size_read(inode);
658 if (offset >= isize) {
663 blkbits = inode->i_sb->s_blocksize_bits;
664 start = offset >> blkbits;
666 end = isize >> blkbits;
670 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
676 if (ret == 0 || es.es_lblk > last) {
678 holeoff = (loff_t)last << blkbits;
682 * If there is a unwritten extent at this offset,
683 * it will be as a data or a hole according to page
684 * cache that has data or not.
686 if (ext4_es_is_unwritten(&es) &&
687 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
688 last + es.es_len, &holeoff))
692 holeoff = (loff_t)last << blkbits;
694 } while (last <= end);
701 return vfs_setpos(file, holeoff, maxsize);
705 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
706 * by calling generic_file_llseek_size() with the appropriate maxbytes
709 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
711 struct inode *inode = file->f_mapping->host;
714 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
715 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
717 maxbytes = inode->i_sb->s_maxbytes;
723 return generic_file_llseek_size(file, offset, whence,
724 maxbytes, i_size_read(inode));
726 return ext4_seek_data(file, offset, maxbytes);
728 return ext4_seek_hole(file, offset, maxbytes);
734 const struct file_operations ext4_file_operations = {
735 .llseek = ext4_llseek,
736 .read_iter = ext4_file_read_iter,
737 .write_iter = ext4_file_write_iter,
738 .unlocked_ioctl = ext4_ioctl,
740 .compat_ioctl = ext4_compat_ioctl,
742 .mmap = ext4_file_mmap,
743 .open = ext4_file_open,
744 .release = ext4_release_file,
745 .fsync = ext4_sync_file,
746 .get_unmapped_area = thp_get_unmapped_area,
747 .splice_read = generic_file_splice_read,
748 .splice_write = iter_file_splice_write,
749 .fallocate = ext4_fallocate,
752 const struct inode_operations ext4_file_inode_operations = {
753 .setattr = ext4_setattr,
754 .getattr = ext4_file_getattr,
755 .listxattr = ext4_listxattr,
756 .get_acl = ext4_get_acl,
757 .set_acl = ext4_set_acl,
758 .fiemap = ext4_fiemap,