4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2011, Lawrence Livermore National Security, LLC.
23 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
28 #include <linux/compat.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/zfs_znode.h>
34 #include <sys/zfs_vfsops.h>
35 #include <sys/zfs_vnops.h>
36 #include <sys/zfs_project.h>
37 #if defined(HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS) || \
38 defined(HAVE_VFS_FILEMAP_DIRTY_FOLIO)
39 #include <linux/pagemap.h>
41 #ifdef HAVE_FILE_FADVISE
42 #include <linux/fadvise.h>
44 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
45 #include <linux/writeback.h>
49 * When using fallocate(2) to preallocate space, inflate the requested
50 * capacity check by 10% to account for the required metadata blocks.
52 static unsigned int zfs_fallocate_reserve_percent
= 110;
55 zpl_open(struct inode
*ip
, struct file
*filp
)
59 fstrans_cookie_t cookie
;
61 error
= generic_file_open(ip
, filp
);
66 cookie
= spl_fstrans_mark();
67 error
= -zfs_open(ip
, filp
->f_mode
, filp
->f_flags
, cr
);
68 spl_fstrans_unmark(cookie
);
70 ASSERT3S(error
, <=, 0);
76 zpl_release(struct inode
*ip
, struct file
*filp
)
80 fstrans_cookie_t cookie
;
82 cookie
= spl_fstrans_mark();
83 if (ITOZ(ip
)->z_atime_dirty
)
84 zfs_mark_inode_dirty(ip
);
87 error
= -zfs_close(ip
, filp
->f_flags
, cr
);
88 spl_fstrans_unmark(cookie
);
90 ASSERT3S(error
, <=, 0);
96 zpl_iterate(struct file
*filp
, zpl_dir_context_t
*ctx
)
100 fstrans_cookie_t cookie
;
103 cookie
= spl_fstrans_mark();
104 error
= -zfs_readdir(file_inode(filp
), ctx
, cr
);
105 spl_fstrans_unmark(cookie
);
107 ASSERT3S(error
, <=, 0);
112 #if !defined(HAVE_VFS_ITERATE) && !defined(HAVE_VFS_ITERATE_SHARED)
114 zpl_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
116 zpl_dir_context_t ctx
=
117 ZPL_DIR_CONTEXT_INIT(dirent
, filldir
, filp
->f_pos
);
120 error
= zpl_iterate(filp
, &ctx
);
121 filp
->f_pos
= ctx
.pos
;
125 #endif /* !HAVE_VFS_ITERATE && !HAVE_VFS_ITERATE_SHARED */
127 #if defined(HAVE_FSYNC_WITHOUT_DENTRY)
129 * Linux 2.6.35 - 3.0 API,
130 * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
131 * redundant. The dentry is still accessible via filp->f_path.dentry,
132 * and we are guaranteed that filp will never be NULL.
135 zpl_fsync(struct file
*filp
, int datasync
)
137 struct inode
*inode
= filp
->f_mapping
->host
;
140 fstrans_cookie_t cookie
;
143 cookie
= spl_fstrans_mark();
144 error
= -zfs_fsync(ITOZ(inode
), datasync
, cr
);
145 spl_fstrans_unmark(cookie
);
147 ASSERT3S(error
, <=, 0);
152 #ifdef HAVE_FILE_AIO_FSYNC
154 zpl_aio_fsync(struct kiocb
*kiocb
, int datasync
)
156 return (zpl_fsync(kiocb
->ki_filp
, datasync
));
160 #elif defined(HAVE_FSYNC_RANGE)
163 * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
164 * been pushed down in to the .fsync() vfs hook. Additionally, the i_mutex
165 * lock is no longer held by the caller, for zfs we don't require the lock
166 * to be held so we don't acquire it.
169 zpl_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
171 struct inode
*inode
= filp
->f_mapping
->host
;
172 znode_t
*zp
= ITOZ(inode
);
173 zfsvfs_t
*zfsvfs
= ITOZSB(inode
);
176 fstrans_cookie_t cookie
;
179 * The variables z_sync_writes_cnt and z_async_writes_cnt work in
180 * tandem so that sync writes can detect if there are any non-sync
181 * writes going on and vice-versa. The "vice-versa" part to this logic
182 * is located in zfs_putpage() where non-sync writes check if there are
183 * any ongoing sync writes. If any sync and non-sync writes overlap,
184 * we do a commit to complete the non-sync writes since the latter can
185 * potentially take several seconds to complete and thus block sync
186 * writes in the upcoming call to filemap_write_and_wait_range().
188 atomic_inc_32(&zp
->z_sync_writes_cnt
);
190 * If the following check does not detect an overlapping non-sync write
191 * (say because it's just about to start), then it is guaranteed that
192 * the non-sync write will detect this sync write. This is because we
193 * always increment z_sync_writes_cnt / z_async_writes_cnt before doing
194 * the check on z_async_writes_cnt / z_sync_writes_cnt here and in
195 * zfs_putpage() respectively.
197 if (atomic_load_32(&zp
->z_async_writes_cnt
) > 0) {
199 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
203 error
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
206 * The sync write is not complete yet but we decrement
207 * z_sync_writes_cnt since zfs_fsync() increments and decrements
208 * it internally. If a non-sync write starts just after the decrement
209 * operation but before we call zfs_fsync(), it may not detect this
210 * overlapping sync write but it does not matter since we have already
211 * gone past filemap_write_and_wait_range() and we won't block due to
212 * the non-sync write.
214 atomic_dec_32(&zp
->z_sync_writes_cnt
);
220 cookie
= spl_fstrans_mark();
221 error
= -zfs_fsync(zp
, datasync
, cr
);
222 spl_fstrans_unmark(cookie
);
224 ASSERT3S(error
, <=, 0);
229 #ifdef HAVE_FILE_AIO_FSYNC
231 zpl_aio_fsync(struct kiocb
*kiocb
, int datasync
)
233 return (zpl_fsync(kiocb
->ki_filp
, kiocb
->ki_pos
, -1, datasync
));
238 #error "Unsupported fops->fsync() implementation"
242 zfs_io_flags(struct kiocb
*kiocb
)
246 #if defined(IOCB_DSYNC)
247 if (kiocb
->ki_flags
& IOCB_DSYNC
)
250 #if defined(IOCB_SYNC)
251 if (kiocb
->ki_flags
& IOCB_SYNC
)
254 #if defined(IOCB_APPEND)
255 if (kiocb
->ki_flags
& IOCB_APPEND
)
258 #if defined(IOCB_DIRECT)
259 if (kiocb
->ki_flags
& IOCB_DIRECT
)
266 * If relatime is enabled, call file_accessed() if zfs_relatime_need_update()
267 * is true. This is needed since datasets with inherited "relatime" property
268 * aren't necessarily mounted with the MNT_RELATIME flag (e.g. after
269 * `zfs set relatime=...`), which is what relatime test in VFS by
270 * relatime_need_update() is based on.
273 zpl_file_accessed(struct file
*filp
)
275 struct inode
*ip
= filp
->f_mapping
->host
;
277 if (!IS_NOATIME(ip
) && ITOZSB(ip
)->z_relatime
) {
278 if (zfs_relatime_need_update(ip
))
285 #if defined(HAVE_VFS_RW_ITERATE)
288 * When HAVE_VFS_IOV_ITER is defined the iov_iter structure supports
289 * iovecs, kvevs, bvecs and pipes, plus all the required interfaces to
290 * manipulate the iov_iter are available. In which case the full iov_iter
291 * can be attached to the uio and correctly handled in the lower layers.
292 * Otherwise, for older kernels extract the iovec and pass it instead.
295 zpl_uio_init(zfs_uio_t
*uio
, struct kiocb
*kiocb
, struct iov_iter
*to
,
296 loff_t pos
, ssize_t count
, size_t skip
)
298 #if defined(HAVE_VFS_IOV_ITER)
299 zfs_uio_iov_iter_init(uio
, to
, pos
, count
, skip
);
301 #ifdef HAVE_IOV_ITER_TYPE
302 zfs_uio_iovec_init(uio
, to
->iov
, to
->nr_segs
, pos
,
303 iov_iter_type(to
) & ITER_KVEC
? UIO_SYSSPACE
: UIO_USERSPACE
,
306 zfs_uio_iovec_init(uio
, to
->iov
, to
->nr_segs
, pos
,
307 to
->type
& ITER_KVEC
? UIO_SYSSPACE
: UIO_USERSPACE
,
314 zpl_iter_read(struct kiocb
*kiocb
, struct iov_iter
*to
)
317 fstrans_cookie_t cookie
;
318 struct file
*filp
= kiocb
->ki_filp
;
319 ssize_t count
= iov_iter_count(to
);
322 zpl_uio_init(&uio
, kiocb
, to
, kiocb
->ki_pos
, count
, 0);
325 cookie
= spl_fstrans_mark();
327 int error
= -zfs_read(ITOZ(filp
->f_mapping
->host
), &uio
,
328 filp
->f_flags
| zfs_io_flags(kiocb
), cr
);
330 spl_fstrans_unmark(cookie
);
336 ssize_t read
= count
- uio
.uio_resid
;
337 kiocb
->ki_pos
+= read
;
339 zpl_file_accessed(filp
);
344 static inline ssize_t
345 zpl_generic_write_checks(struct kiocb
*kiocb
, struct iov_iter
*from
,
348 #ifdef HAVE_GENERIC_WRITE_CHECKS_KIOCB
349 ssize_t ret
= generic_write_checks(kiocb
, from
);
355 struct file
*file
= kiocb
->ki_filp
;
356 struct address_space
*mapping
= file
->f_mapping
;
357 struct inode
*ip
= mapping
->host
;
358 int isblk
= S_ISBLK(ip
->i_mode
);
360 *countp
= iov_iter_count(from
);
361 ssize_t ret
= generic_write_checks(file
, &kiocb
->ki_pos
, countp
, isblk
);
370 zpl_iter_write(struct kiocb
*kiocb
, struct iov_iter
*from
)
373 fstrans_cookie_t cookie
;
374 struct file
*filp
= kiocb
->ki_filp
;
375 struct inode
*ip
= filp
->f_mapping
->host
;
380 ret
= zpl_generic_write_checks(kiocb
, from
, &count
);
384 zpl_uio_init(&uio
, kiocb
, from
, kiocb
->ki_pos
, count
, from
->iov_offset
);
387 cookie
= spl_fstrans_mark();
389 int error
= -zfs_write(ITOZ(ip
), &uio
,
390 filp
->f_flags
| zfs_io_flags(kiocb
), cr
);
392 spl_fstrans_unmark(cookie
);
398 ssize_t wrote
= count
- uio
.uio_resid
;
399 kiocb
->ki_pos
+= wrote
;
404 #else /* !HAVE_VFS_RW_ITERATE */
407 zpl_aio_read(struct kiocb
*kiocb
, const struct iovec
*iov
,
408 unsigned long nr_segs
, loff_t pos
)
411 fstrans_cookie_t cookie
;
412 struct file
*filp
= kiocb
->ki_filp
;
416 ret
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
421 zfs_uio_iovec_init(&uio
, iov
, nr_segs
, kiocb
->ki_pos
, UIO_USERSPACE
,
425 cookie
= spl_fstrans_mark();
427 int error
= -zfs_read(ITOZ(filp
->f_mapping
->host
), &uio
,
428 filp
->f_flags
| zfs_io_flags(kiocb
), cr
);
430 spl_fstrans_unmark(cookie
);
436 ssize_t read
= count
- uio
.uio_resid
;
437 kiocb
->ki_pos
+= read
;
439 zpl_file_accessed(filp
);
445 zpl_aio_write(struct kiocb
*kiocb
, const struct iovec
*iov
,
446 unsigned long nr_segs
, loff_t pos
)
449 fstrans_cookie_t cookie
;
450 struct file
*filp
= kiocb
->ki_filp
;
451 struct inode
*ip
= filp
->f_mapping
->host
;
455 ret
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_READ
);
459 ret
= generic_write_checks(filp
, &pos
, &count
, S_ISBLK(ip
->i_mode
));
466 zfs_uio_iovec_init(&uio
, iov
, nr_segs
, kiocb
->ki_pos
, UIO_USERSPACE
,
470 cookie
= spl_fstrans_mark();
472 int error
= -zfs_write(ITOZ(ip
), &uio
,
473 filp
->f_flags
| zfs_io_flags(kiocb
), cr
);
475 spl_fstrans_unmark(cookie
);
481 ssize_t wrote
= count
- uio
.uio_resid
;
482 kiocb
->ki_pos
+= wrote
;
486 #endif /* HAVE_VFS_RW_ITERATE */
488 #if defined(HAVE_VFS_RW_ITERATE)
490 zpl_direct_IO_impl(int rw
, struct kiocb
*kiocb
, struct iov_iter
*iter
)
493 return (zpl_iter_write(kiocb
, iter
));
495 return (zpl_iter_read(kiocb
, iter
));
497 #if defined(HAVE_VFS_DIRECT_IO_ITER)
499 zpl_direct_IO(struct kiocb
*kiocb
, struct iov_iter
*iter
)
501 return (zpl_direct_IO_impl(iov_iter_rw(iter
), kiocb
, iter
));
503 #elif defined(HAVE_VFS_DIRECT_IO_ITER_OFFSET)
505 zpl_direct_IO(struct kiocb
*kiocb
, struct iov_iter
*iter
, loff_t pos
)
507 ASSERT3S(pos
, ==, kiocb
->ki_pos
);
508 return (zpl_direct_IO_impl(iov_iter_rw(iter
), kiocb
, iter
));
510 #elif defined(HAVE_VFS_DIRECT_IO_ITER_RW_OFFSET)
512 zpl_direct_IO(int rw
, struct kiocb
*kiocb
, struct iov_iter
*iter
, loff_t pos
)
514 ASSERT3S(pos
, ==, kiocb
->ki_pos
);
515 return (zpl_direct_IO_impl(rw
, kiocb
, iter
));
518 #error "Unknown direct IO interface"
521 #else /* HAVE_VFS_RW_ITERATE */
523 #if defined(HAVE_VFS_DIRECT_IO_IOVEC)
525 zpl_direct_IO(int rw
, struct kiocb
*kiocb
, const struct iovec
*iov
,
526 loff_t pos
, unsigned long nr_segs
)
529 return (zpl_aio_write(kiocb
, iov
, nr_segs
, pos
));
531 return (zpl_aio_read(kiocb
, iov
, nr_segs
, pos
));
533 #elif defined(HAVE_VFS_DIRECT_IO_ITER_RW_OFFSET)
535 zpl_direct_IO(int rw
, struct kiocb
*kiocb
, struct iov_iter
*iter
, loff_t pos
)
537 const struct iovec
*iovp
= iov_iter_iovec(iter
);
538 unsigned long nr_segs
= iter
->nr_segs
;
540 ASSERT3S(pos
, ==, kiocb
->ki_pos
);
542 return (zpl_aio_write(kiocb
, iovp
, nr_segs
, pos
));
544 return (zpl_aio_read(kiocb
, iovp
, nr_segs
, pos
));
547 #error "Unknown direct IO interface"
550 #endif /* HAVE_VFS_RW_ITERATE */
553 zpl_llseek(struct file
*filp
, loff_t offset
, int whence
)
555 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
556 fstrans_cookie_t cookie
;
558 if (whence
== SEEK_DATA
|| whence
== SEEK_HOLE
) {
559 struct inode
*ip
= filp
->f_mapping
->host
;
560 loff_t maxbytes
= ip
->i_sb
->s_maxbytes
;
563 spl_inode_lock_shared(ip
);
564 cookie
= spl_fstrans_mark();
565 error
= -zfs_holey(ITOZ(ip
), whence
, &offset
);
566 spl_fstrans_unmark(cookie
);
568 error
= lseek_execute(filp
, ip
, offset
, maxbytes
);
569 spl_inode_unlock_shared(ip
);
573 #endif /* SEEK_HOLE && SEEK_DATA */
575 return (generic_file_llseek(filp
, offset
, whence
));
579 * It's worth taking a moment to describe how mmap is implemented
580 * for zfs because it differs considerably from other Linux filesystems.
581 * However, this issue is handled the same way under OpenSolaris.
583 * The issue is that by design zfs bypasses the Linux page cache and
584 * leaves all caching up to the ARC. This has been shown to work
585 * well for the common read(2)/write(2) case. However, mmap(2)
586 * is problem because it relies on being tightly integrated with the
587 * page cache. To handle this we cache mmap'ed files twice, once in
588 * the ARC and a second time in the page cache. The code is careful
589 * to keep both copies synchronized.
591 * When a file with an mmap'ed region is written to using write(2)
592 * both the data in the ARC and existing pages in the page cache
593 * are updated. For a read(2) data will be read first from the page
594 * cache then the ARC if needed. Neither a write(2) or read(2) will
595 * will ever result in new pages being added to the page cache.
597 * New pages are added to the page cache only via .readpage() which
598 * is called when the vfs needs to read a page off disk to back the
599 * virtual memory region. These pages may be modified without
600 * notifying the ARC and will be written out periodically via
601 * .writepage(). This will occur due to either a sync or the usual
602 * page aging behavior. Note because a read(2) of a mmap'ed file
603 * will always check the page cache first even when the ARC is out
604 * of date correct data will still be returned.
606 * While this implementation ensures correct behavior it does have
607 * have some drawbacks. The most obvious of which is that it
608 * increases the required memory footprint when access mmap'ed
609 * files. It also adds additional complexity to the code keeping
610 * both caches synchronized.
612 * Longer term it may be possible to cleanly resolve this wart by
613 * mapping page cache pages directly on to the ARC buffers. The
614 * Linux address space operations are flexible enough to allow
615 * selection of which pages back a particular index. The trick
616 * would be working out the details of which subsystem is in
617 * charge, the ARC, the page cache, or both. It may also prove
618 * helpful to move the ARC buffers to a scatter-gather lists
619 * rather than a vmalloc'ed region.
622 zpl_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
624 struct inode
*ip
= filp
->f_mapping
->host
;
625 znode_t
*zp
= ITOZ(ip
);
627 fstrans_cookie_t cookie
;
629 cookie
= spl_fstrans_mark();
630 error
= -zfs_map(ip
, vma
->vm_pgoff
, (caddr_t
*)vma
->vm_start
,
631 (size_t)(vma
->vm_end
- vma
->vm_start
), vma
->vm_flags
);
632 spl_fstrans_unmark(cookie
);
636 error
= generic_file_mmap(filp
, vma
);
640 mutex_enter(&zp
->z_lock
);
641 zp
->z_is_mapped
= B_TRUE
;
642 mutex_exit(&zp
->z_lock
);
648 * Populate a page with data for the Linux page cache. This function is
649 * only used to support mmap(2). There will be an identical copy of the
650 * data in the ARC which is kept up to date via .write() and .writepage().
653 zpl_readpage_common(struct page
*pp
)
658 fstrans_cookie_t cookie
;
660 ASSERT(PageLocked(pp
));
661 ip
= pp
->mapping
->host
;
664 cookie
= spl_fstrans_mark();
665 error
= -zfs_getpage(ip
, pl
, 1);
666 spl_fstrans_unmark(cookie
);
670 ClearPageUptodate(pp
);
674 flush_dcache_page(pp
);
681 #ifdef HAVE_VFS_READ_FOLIO
683 zpl_read_folio(struct file
*filp
, struct folio
*folio
)
685 return (zpl_readpage_common(&folio
->page
));
689 zpl_readpage(struct file
*filp
, struct page
*pp
)
691 return (zpl_readpage_common(pp
));
696 zpl_readpage_filler(void *data
, struct page
*pp
)
698 return (zpl_readpage_common(pp
));
702 * Populate a set of pages with data for the Linux page cache. This
703 * function will only be called for read ahead and never for demand
704 * paging. For simplicity, the code relies on read_cache_pages() to
705 * correctly lock each page for IO and call zpl_readpage().
707 #ifdef HAVE_VFS_READPAGES
709 zpl_readpages(struct file
*filp
, struct address_space
*mapping
,
710 struct list_head
*pages
, unsigned nr_pages
)
712 return (read_cache_pages(mapping
, pages
, zpl_readpage_filler
, NULL
));
716 zpl_readahead(struct readahead_control
*ractl
)
720 while ((page
= readahead_page(ractl
)) != NULL
) {
723 ret
= zpl_readpage_filler(NULL
, page
);
732 zpl_putpage(struct page
*pp
, struct writeback_control
*wbc
, void *data
)
734 boolean_t
*for_sync
= data
;
735 fstrans_cookie_t cookie
;
737 ASSERT(PageLocked(pp
));
738 ASSERT(!PageWriteback(pp
));
740 cookie
= spl_fstrans_mark();
741 (void) zfs_putpage(pp
->mapping
->host
, pp
, wbc
, *for_sync
);
742 spl_fstrans_unmark(cookie
);
748 zpl_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
750 znode_t
*zp
= ITOZ(mapping
->host
);
751 zfsvfs_t
*zfsvfs
= ITOZSB(mapping
->host
);
752 enum writeback_sync_modes sync_mode
;
756 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
757 wbc
->sync_mode
= WB_SYNC_ALL
;
759 sync_mode
= wbc
->sync_mode
;
762 * We don't want to run write_cache_pages() in SYNC mode here, because
763 * that would make putpage() wait for a single page to be committed to
764 * disk every single time, resulting in atrocious performance. Instead
765 * we run it once in non-SYNC mode so that the ZIL gets all the data,
766 * and then we commit it all in one go.
768 boolean_t for_sync
= (sync_mode
== WB_SYNC_ALL
);
769 wbc
->sync_mode
= WB_SYNC_NONE
;
770 result
= write_cache_pages(mapping
, wbc
, zpl_putpage
, &for_sync
);
771 if (sync_mode
!= wbc
->sync_mode
) {
774 if (zfsvfs
->z_log
!= NULL
)
775 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
779 * We need to call write_cache_pages() again (we can't just
780 * return after the commit) because the previous call in
781 * non-SYNC mode does not guarantee that we got all the dirty
782 * pages (see the implementation of write_cache_pages() for
783 * details). That being said, this is a no-op in most cases.
785 wbc
->sync_mode
= sync_mode
;
786 result
= write_cache_pages(mapping
, wbc
, zpl_putpage
,
793 * Write out dirty pages to the ARC, this function is only required to
794 * support mmap(2). Mapped pages may be dirtied by memory operations
795 * which never call .write(). These dirty pages are kept in sync with
796 * the ARC buffers via this hook.
799 zpl_writepage(struct page
*pp
, struct writeback_control
*wbc
)
801 if (ITOZSB(pp
->mapping
->host
)->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
802 wbc
->sync_mode
= WB_SYNC_ALL
;
804 boolean_t for_sync
= (wbc
->sync_mode
== WB_SYNC_ALL
);
806 return (zpl_putpage(pp
, wbc
, &for_sync
));
810 * The flag combination which matches the behavior of zfs_space() is
811 * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE. The FALLOC_FL_PUNCH_HOLE
812 * flag was introduced in the 2.6.38 kernel.
814 * The original mode=0 (allocate space) behavior can be reasonably emulated
815 * by checking if enough space exists and creating a sparse file, as real
816 * persistent space reservation is not possible due to COW, snapshots, etc.
819 zpl_fallocate_common(struct inode
*ip
, int mode
, loff_t offset
, loff_t len
)
823 fstrans_cookie_t cookie
;
826 int test_mode
= FALLOC_FL_PUNCH_HOLE
;
827 #ifdef HAVE_FALLOC_FL_ZERO_RANGE
828 test_mode
|= FALLOC_FL_ZERO_RANGE
;
831 if ((mode
& ~(FALLOC_FL_KEEP_SIZE
| test_mode
)) != 0)
832 return (-EOPNOTSUPP
);
834 if (offset
< 0 || len
<= 0)
838 olen
= i_size_read(ip
);
841 cookie
= spl_fstrans_mark();
842 if (mode
& (test_mode
)) {
845 if (mode
& FALLOC_FL_KEEP_SIZE
) {
849 if (offset
+ len
> olen
)
853 bf
.l_whence
= SEEK_SET
;
858 error
= -zfs_space(ITOZ(ip
), F_FREESP
, &bf
, O_RDWR
, offset
, cr
);
859 } else if ((mode
& ~FALLOC_FL_KEEP_SIZE
) == 0) {
860 unsigned int percent
= zfs_fallocate_reserve_percent
;
861 struct kstatfs statfs
;
863 /* Legacy mode, disable fallocate compatibility. */
870 * Use zfs_statvfs() instead of dmu_objset_space() since it
871 * also checks project quota limits, which are relevant here.
873 error
= zfs_statvfs(ip
, &statfs
);
878 * Shrink available space a bit to account for overhead/races.
879 * We know the product previously fit into availbytes from
880 * dmu_objset_space(), so the smaller product will also fit.
882 if (len
> statfs
.f_bavail
* (statfs
.f_bsize
* 100 / percent
)) {
886 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> olen
)
887 error
= zfs_freesp(ITOZ(ip
), offset
+ len
, 0, 0, FALSE
);
890 spl_fstrans_unmark(cookie
);
891 spl_inode_unlock(ip
);
899 zpl_fallocate(struct file
*filp
, int mode
, loff_t offset
, loff_t len
)
901 return zpl_fallocate_common(file_inode(filp
),
906 zpl_ioctl_getversion(struct file
*filp
, void __user
*arg
)
908 uint32_t generation
= file_inode(filp
)->i_generation
;
910 return (copy_to_user(arg
, &generation
, sizeof (generation
)));
913 #ifdef HAVE_FILE_FADVISE
915 zpl_fadvise(struct file
*filp
, loff_t offset
, loff_t len
, int advice
)
917 struct inode
*ip
= file_inode(filp
);
918 znode_t
*zp
= ITOZ(ip
);
919 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
920 objset_t
*os
= zfsvfs
->z_os
;
923 if (S_ISFIFO(ip
->i_mode
))
926 if (offset
< 0 || len
< 0)
933 case POSIX_FADV_SEQUENTIAL
:
934 case POSIX_FADV_WILLNEED
:
935 #ifdef HAVE_GENERIC_FADVISE
936 if (zn_has_cached_data(zp
))
937 error
= generic_fadvise(filp
, offset
, len
, advice
);
940 * Pass on the caller's size directly, but note that
941 * dmu_prefetch_max will effectively cap it. If there
942 * really is a larger sequential access pattern, perhaps
943 * dmu_zfetch will detect it.
946 len
= i_size_read(ip
) - offset
;
948 dmu_prefetch(os
, zp
->z_id
, 0, offset
, len
,
949 ZIO_PRIORITY_ASYNC_READ
);
951 case POSIX_FADV_NORMAL
:
952 case POSIX_FADV_RANDOM
:
953 case POSIX_FADV_DONTNEED
:
954 case POSIX_FADV_NOREUSE
:
955 /* ignored for now */
966 #endif /* HAVE_FILE_FADVISE */
968 #define ZFS_FL_USER_VISIBLE (FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
969 #define ZFS_FL_USER_MODIFIABLE (FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
972 __zpl_ioctl_getflags(struct inode
*ip
)
974 uint64_t zfs_flags
= ITOZ(ip
)->z_pflags
;
975 uint32_t ioctl_flags
= 0;
977 if (zfs_flags
& ZFS_IMMUTABLE
)
978 ioctl_flags
|= FS_IMMUTABLE_FL
;
980 if (zfs_flags
& ZFS_APPENDONLY
)
981 ioctl_flags
|= FS_APPEND_FL
;
983 if (zfs_flags
& ZFS_NODUMP
)
984 ioctl_flags
|= FS_NODUMP_FL
;
986 if (zfs_flags
& ZFS_PROJINHERIT
)
987 ioctl_flags
|= ZFS_PROJINHERIT_FL
;
989 return (ioctl_flags
& ZFS_FL_USER_VISIBLE
);
993 * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
994 * attributes common to both Linux and Solaris are mapped.
997 zpl_ioctl_getflags(struct file
*filp
, void __user
*arg
)
1002 flags
= __zpl_ioctl_getflags(file_inode(filp
));
1003 err
= copy_to_user(arg
, &flags
, sizeof (flags
));
1009 * fchange() is a helper macro to detect if we have been asked to change a
1010 * flag. This is ugly, but the requirement that we do this is a consequence of
1011 * how the Linux file attribute interface was designed. Another consequence is
1012 * that concurrent modification of files suffers from a TOCTOU race. Neither
1013 * are things we can fix without modifying the kernel-userland interface, which
1014 * is outside of our jurisdiction.
1017 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
1020 __zpl_ioctl_setflags(struct inode
*ip
, uint32_t ioctl_flags
, xvattr_t
*xva
)
1022 uint64_t zfs_flags
= ITOZ(ip
)->z_pflags
;
1025 if (ioctl_flags
& ~(FS_IMMUTABLE_FL
| FS_APPEND_FL
| FS_NODUMP_FL
|
1026 ZFS_PROJINHERIT_FL
))
1027 return (-EOPNOTSUPP
);
1029 if (ioctl_flags
& ~ZFS_FL_USER_MODIFIABLE
)
1032 if ((fchange(ioctl_flags
, zfs_flags
, FS_IMMUTABLE_FL
, ZFS_IMMUTABLE
) ||
1033 fchange(ioctl_flags
, zfs_flags
, FS_APPEND_FL
, ZFS_APPENDONLY
)) &&
1034 !capable(CAP_LINUX_IMMUTABLE
))
1037 if (!zpl_inode_owner_or_capable(kcred
->user_ns
, ip
))
1041 xoap
= xva_getxoptattr(xva
);
1043 #define FLAG_CHANGE(iflag, zflag, xflag, xfield) do { \
1044 if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) || \
1045 ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) { \
1046 XVA_SET_REQ(xva, (xflag)); \
1047 (xfield) = ((ioctl_flags & (iflag)) != 0); \
1051 FLAG_CHANGE(FS_IMMUTABLE_FL
, ZFS_IMMUTABLE
, XAT_IMMUTABLE
,
1052 xoap
->xoa_immutable
);
1053 FLAG_CHANGE(FS_APPEND_FL
, ZFS_APPENDONLY
, XAT_APPENDONLY
,
1054 xoap
->xoa_appendonly
);
1055 FLAG_CHANGE(FS_NODUMP_FL
, ZFS_NODUMP
, XAT_NODUMP
,
1057 FLAG_CHANGE(ZFS_PROJINHERIT_FL
, ZFS_PROJINHERIT
, XAT_PROJINHERIT
,
1058 xoap
->xoa_projinherit
);
1066 zpl_ioctl_setflags(struct file
*filp
, void __user
*arg
)
1068 struct inode
*ip
= file_inode(filp
);
1070 cred_t
*cr
= CRED();
1073 fstrans_cookie_t cookie
;
1075 if (copy_from_user(&flags
, arg
, sizeof (flags
)))
1078 err
= __zpl_ioctl_setflags(ip
, flags
, &xva
);
1083 cookie
= spl_fstrans_mark();
1084 err
= -zfs_setattr(ITOZ(ip
), (vattr_t
*)&xva
, 0, cr
);
1085 spl_fstrans_unmark(cookie
);
1092 zpl_ioctl_getxattr(struct file
*filp
, void __user
*arg
)
1094 zfsxattr_t fsx
= { 0 };
1095 struct inode
*ip
= file_inode(filp
);
1098 fsx
.fsx_xflags
= __zpl_ioctl_getflags(ip
);
1099 fsx
.fsx_projid
= ITOZ(ip
)->z_projid
;
1100 err
= copy_to_user(arg
, &fsx
, sizeof (fsx
));
1106 zpl_ioctl_setxattr(struct file
*filp
, void __user
*arg
)
1108 struct inode
*ip
= file_inode(filp
);
1110 cred_t
*cr
= CRED();
1114 fstrans_cookie_t cookie
;
1116 if (copy_from_user(&fsx
, arg
, sizeof (fsx
)))
1119 if (!zpl_is_valid_projid(fsx
.fsx_projid
))
1122 err
= __zpl_ioctl_setflags(ip
, fsx
.fsx_xflags
, &xva
);
1126 xoap
= xva_getxoptattr(&xva
);
1127 XVA_SET_REQ(&xva
, XAT_PROJID
);
1128 xoap
->xoa_projid
= fsx
.fsx_projid
;
1131 cookie
= spl_fstrans_mark();
1132 err
= -zfs_setattr(ITOZ(ip
), (vattr_t
*)&xva
, 0, cr
);
1133 spl_fstrans_unmark(cookie
);
1140 * Expose Additional File Level Attributes of ZFS.
1143 zpl_ioctl_getdosflags(struct file
*filp
, void __user
*arg
)
1145 struct inode
*ip
= file_inode(filp
);
1146 uint64_t dosflags
= ITOZ(ip
)->z_pflags
;
1147 dosflags
&= ZFS_DOS_FL_USER_VISIBLE
;
1148 int err
= copy_to_user(arg
, &dosflags
, sizeof (dosflags
));
1154 __zpl_ioctl_setdosflags(struct inode
*ip
, uint64_t ioctl_flags
, xvattr_t
*xva
)
1156 uint64_t zfs_flags
= ITOZ(ip
)->z_pflags
;
1159 if (ioctl_flags
& (~ZFS_DOS_FL_USER_VISIBLE
))
1160 return (-EOPNOTSUPP
);
1162 if ((fchange(ioctl_flags
, zfs_flags
, ZFS_IMMUTABLE
, ZFS_IMMUTABLE
) ||
1163 fchange(ioctl_flags
, zfs_flags
, ZFS_APPENDONLY
, ZFS_APPENDONLY
)) &&
1164 !capable(CAP_LINUX_IMMUTABLE
))
1167 if (!zpl_inode_owner_or_capable(kcred
->user_ns
, ip
))
1171 xoap
= xva_getxoptattr(xva
);
1173 #define FLAG_CHANGE(iflag, xflag, xfield) do { \
1174 if (((ioctl_flags & (iflag)) && !(zfs_flags & (iflag))) || \
1175 ((zfs_flags & (iflag)) && !(ioctl_flags & (iflag)))) { \
1176 XVA_SET_REQ(xva, (xflag)); \
1177 (xfield) = ((ioctl_flags & (iflag)) != 0); \
1181 FLAG_CHANGE(ZFS_IMMUTABLE
, XAT_IMMUTABLE
, xoap
->xoa_immutable
);
1182 FLAG_CHANGE(ZFS_APPENDONLY
, XAT_APPENDONLY
, xoap
->xoa_appendonly
);
1183 FLAG_CHANGE(ZFS_NODUMP
, XAT_NODUMP
, xoap
->xoa_nodump
);
1184 FLAG_CHANGE(ZFS_READONLY
, XAT_READONLY
, xoap
->xoa_readonly
);
1185 FLAG_CHANGE(ZFS_HIDDEN
, XAT_HIDDEN
, xoap
->xoa_hidden
);
1186 FLAG_CHANGE(ZFS_SYSTEM
, XAT_SYSTEM
, xoap
->xoa_system
);
1187 FLAG_CHANGE(ZFS_ARCHIVE
, XAT_ARCHIVE
, xoap
->xoa_archive
);
1188 FLAG_CHANGE(ZFS_NOUNLINK
, XAT_NOUNLINK
, xoap
->xoa_nounlink
);
1189 FLAG_CHANGE(ZFS_REPARSE
, XAT_REPARSE
, xoap
->xoa_reparse
);
1190 FLAG_CHANGE(ZFS_OFFLINE
, XAT_OFFLINE
, xoap
->xoa_offline
);
1191 FLAG_CHANGE(ZFS_SPARSE
, XAT_SPARSE
, xoap
->xoa_sparse
);
1199 * Set Additional File Level Attributes of ZFS.
1202 zpl_ioctl_setdosflags(struct file
*filp
, void __user
*arg
)
1204 struct inode
*ip
= file_inode(filp
);
1206 cred_t
*cr
= CRED();
1209 fstrans_cookie_t cookie
;
1211 if (copy_from_user(&dosflags
, arg
, sizeof (dosflags
)))
1214 err
= __zpl_ioctl_setdosflags(ip
, dosflags
, &xva
);
1219 cookie
= spl_fstrans_mark();
1220 err
= -zfs_setattr(ITOZ(ip
), (vattr_t
*)&xva
, 0, cr
);
1221 spl_fstrans_unmark(cookie
);
1228 zpl_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
1231 case FS_IOC_GETVERSION
:
1232 return (zpl_ioctl_getversion(filp
, (void *)arg
));
1233 case FS_IOC_GETFLAGS
:
1234 return (zpl_ioctl_getflags(filp
, (void *)arg
));
1235 case FS_IOC_SETFLAGS
:
1236 return (zpl_ioctl_setflags(filp
, (void *)arg
));
1237 case ZFS_IOC_FSGETXATTR
:
1238 return (zpl_ioctl_getxattr(filp
, (void *)arg
));
1239 case ZFS_IOC_FSSETXATTR
:
1240 return (zpl_ioctl_setxattr(filp
, (void *)arg
));
1241 case ZFS_IOC_GETDOSFLAGS
:
1242 return (zpl_ioctl_getdosflags(filp
, (void *)arg
));
1243 case ZFS_IOC_SETDOSFLAGS
:
1244 return (zpl_ioctl_setdosflags(filp
, (void *)arg
));
1250 #ifdef CONFIG_COMPAT
1252 zpl_compat_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
1255 case FS_IOC32_GETVERSION
:
1256 cmd
= FS_IOC_GETVERSION
;
1258 case FS_IOC32_GETFLAGS
:
1259 cmd
= FS_IOC_GETFLAGS
;
1261 case FS_IOC32_SETFLAGS
:
1262 cmd
= FS_IOC_SETFLAGS
;
1267 return (zpl_ioctl(filp
, cmd
, (unsigned long)compat_ptr(arg
)));
1269 #endif /* CONFIG_COMPAT */
1272 const struct address_space_operations zpl_address_space_operations
= {
1273 #ifdef HAVE_VFS_READPAGES
1274 .readpages
= zpl_readpages
,
1276 .readahead
= zpl_readahead
,
1278 #ifdef HAVE_VFS_READ_FOLIO
1279 .read_folio
= zpl_read_folio
,
1281 .readpage
= zpl_readpage
,
1283 .writepage
= zpl_writepage
,
1284 .writepages
= zpl_writepages
,
1285 .direct_IO
= zpl_direct_IO
,
1286 #ifdef HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS
1287 .set_page_dirty
= __set_page_dirty_nobuffers
,
1289 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
1290 .dirty_folio
= filemap_dirty_folio
,
1294 const struct file_operations zpl_file_operations
= {
1296 .release
= zpl_release
,
1297 .llseek
= zpl_llseek
,
1298 #ifdef HAVE_VFS_RW_ITERATE
1299 #ifdef HAVE_NEW_SYNC_READ
1300 .read
= new_sync_read
,
1301 .write
= new_sync_write
,
1303 .read_iter
= zpl_iter_read
,
1304 .write_iter
= zpl_iter_write
,
1305 #ifdef HAVE_VFS_IOV_ITER
1306 .splice_read
= generic_file_splice_read
,
1307 .splice_write
= iter_file_splice_write
,
1310 .read
= do_sync_read
,
1311 .write
= do_sync_write
,
1312 .aio_read
= zpl_aio_read
,
1313 .aio_write
= zpl_aio_write
,
1317 #ifdef HAVE_FILE_AIO_FSYNC
1318 .aio_fsync
= zpl_aio_fsync
,
1320 .fallocate
= zpl_fallocate
,
1321 #ifdef HAVE_FILE_FADVISE
1322 .fadvise
= zpl_fadvise
,
1324 .unlocked_ioctl
= zpl_ioctl
,
1325 #ifdef CONFIG_COMPAT
1326 .compat_ioctl
= zpl_compat_ioctl
,
1330 const struct file_operations zpl_dir_file_operations
= {
1331 .llseek
= generic_file_llseek
,
1332 .read
= generic_read_dir
,
1333 #if defined(HAVE_VFS_ITERATE_SHARED)
1334 .iterate_shared
= zpl_iterate
,
1335 #elif defined(HAVE_VFS_ITERATE)
1336 .iterate
= zpl_iterate
,
1338 .readdir
= zpl_readdir
,
1341 .unlocked_ioctl
= zpl_ioctl
,
1342 #ifdef CONFIG_COMPAT
1343 .compat_ioctl
= zpl_compat_ioctl
,
1348 module_param(zfs_fallocate_reserve_percent
, uint
, 0644);
1349 MODULE_PARM_DESC(zfs_fallocate_reserve_percent
,
1350 "Percentage of length to use for the available capacity check");