1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22 /*
23 * Copyright (c) 2011, Lawrence Livermore National Security, LLC.
24 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
25 */
26
27
28 #ifdef CONFIG_COMPAT
29 #include <linux/compat.h>
30 #endif
31 #include <linux/fs.h>
32 #include <linux/migrate.h>
33 #include <sys/file.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/zfs_znode.h>
36 #include <sys/zfs_vfsops.h>
37 #include <sys/zfs_vnops.h>
38 #include <sys/zfs_project.h>
39 #if defined(HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS) || \
40 defined(HAVE_VFS_FILEMAP_DIRTY_FOLIO)
41 #include <linux/pagemap.h>
42 #endif
43 #include <linux/fadvise.h>
44 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
45 #include <linux/writeback.h>
46 #endif
47
48 /*
49 * When using fallocate(2) to preallocate space, inflate the requested
50 * capacity check by 10% to account for the required metadata blocks.
51 */
52 static unsigned int zfs_fallocate_reserve_percent = 110;
53
54 static int
zpl_open(struct inode * ip,struct file * filp)55 zpl_open(struct inode *ip, struct file *filp)
56 {
57 cred_t *cr = CRED();
58 int error;
59 fstrans_cookie_t cookie;
60
61 error = generic_file_open(ip, filp);
62 if (error)
63 return (error);
64
65 crhold(cr);
66 cookie = spl_fstrans_mark();
67 error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
68 spl_fstrans_unmark(cookie);
69 crfree(cr);
70 ASSERT3S(error, <=, 0);
71
72 return (error);
73 }
74
75 static int
zpl_release(struct inode * ip,struct file * filp)76 zpl_release(struct inode *ip, struct file *filp)
77 {
78 cred_t *cr = CRED();
79 int error;
80 fstrans_cookie_t cookie;
81
82 cookie = spl_fstrans_mark();
83 if (ITOZ(ip)->z_atime_dirty)
84 zfs_mark_inode_dirty(ip);
85
86 crhold(cr);
87 error = -zfs_close(ip, filp->f_flags, cr);
88 spl_fstrans_unmark(cookie);
89 crfree(cr);
90 ASSERT3S(error, <=, 0);
91
92 return (error);
93 }
94
95 static int
zpl_iterate(struct file * filp,struct dir_context * ctx)96 zpl_iterate(struct file *filp, struct dir_context *ctx)
97 {
98 cred_t *cr = CRED();
99 int error;
100 fstrans_cookie_t cookie;
101
102 crhold(cr);
103 cookie = spl_fstrans_mark();
104 error = -zfs_readdir(file_inode(filp), ctx, cr);
105 spl_fstrans_unmark(cookie);
106 crfree(cr);
107 ASSERT3S(error, <=, 0);
108
109 return (error);
110 }
111
112 static int
zpl_fsync(struct file * filp,loff_t start,loff_t end,int datasync)113 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
114 {
115 struct inode *inode = filp->f_mapping->host;
116 znode_t *zp = ITOZ(inode);
117 zfsvfs_t *zfsvfs = ITOZSB(inode);
118 cred_t *cr = CRED();
119 int error;
120 fstrans_cookie_t cookie;
121
122 /*
123 * The variables z_sync_writes_cnt and z_async_writes_cnt work in
124 * tandem so that sync writes can detect if there are any non-sync
125 * writes going on and vice-versa. The "vice-versa" part to this logic
126 * is located in zfs_putpage() where non-sync writes check if there are
127 * any ongoing sync writes. If any sync and non-sync writes overlap,
128 * we do a commit to complete the non-sync writes since the latter can
129 * potentially take several seconds to complete and thus block sync
130 * writes in the upcoming call to filemap_write_and_wait_range().
131 */
132 atomic_inc_32(&zp->z_sync_writes_cnt);
133 /*
134 * If the following check does not detect an overlapping non-sync write
135 * (say because it's just about to start), then it is guaranteed that
136 * the non-sync write will detect this sync write. This is because we
137 * always increment z_sync_writes_cnt / z_async_writes_cnt before doing
138 * the check on z_async_writes_cnt / z_sync_writes_cnt here and in
139 * zfs_putpage() respectively.
140 */
141 if (atomic_load_32(&zp->z_async_writes_cnt) > 0) {
142 if ((error = zpl_enter(zfsvfs, FTAG)) != 0) {
143 atomic_dec_32(&zp->z_sync_writes_cnt);
144 return (error);
145 }
146 zil_commit(zfsvfs->z_log, zp->z_id);
147 zpl_exit(zfsvfs, FTAG);
148 }
149
150 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
151
152 /*
153 * The sync write is not complete yet but we decrement
154 * z_sync_writes_cnt since zfs_fsync() increments and decrements
155 * it internally. If a non-sync write starts just after the decrement
156 * operation but before we call zfs_fsync(), it may not detect this
157 * overlapping sync write but it does not matter since we have already
158 * gone past filemap_write_and_wait_range() and we won't block due to
159 * the non-sync write.
160 */
161 atomic_dec_32(&zp->z_sync_writes_cnt);
162
163 if (error)
164 return (error);
165
166 crhold(cr);
167 cookie = spl_fstrans_mark();
168 error = -zfs_fsync(zp, datasync, cr);
169 spl_fstrans_unmark(cookie);
170 crfree(cr);
171 ASSERT3S(error, <=, 0);
172
173 return (error);
174 }
175
176 static inline int
zfs_io_flags(struct kiocb * kiocb)177 zfs_io_flags(struct kiocb *kiocb)
178 {
179 int flags = 0;
180
181 #if defined(IOCB_DSYNC)
182 if (kiocb->ki_flags & IOCB_DSYNC)
183 flags |= O_DSYNC;
184 #endif
185 #if defined(IOCB_SYNC)
186 if (kiocb->ki_flags & IOCB_SYNC)
187 flags |= O_SYNC;
188 #endif
189 #if defined(IOCB_APPEND)
190 if (kiocb->ki_flags & IOCB_APPEND)
191 flags |= O_APPEND;
192 #endif
193 #if defined(IOCB_DIRECT)
194 if (kiocb->ki_flags & IOCB_DIRECT)
195 flags |= O_DIRECT;
196 #endif
197 return (flags);
198 }
199
200 /*
201 * If relatime is enabled, call file_accessed() if zfs_relatime_need_update()
202 * is true. This is needed since datasets with inherited "relatime" property
203 * aren't necessarily mounted with the MNT_RELATIME flag (e.g. after
204 * `zfs set relatime=...`), which is what relatime test in VFS by
205 * relatime_need_update() is based on.
206 */
207 static inline void
zpl_file_accessed(struct file * filp)208 zpl_file_accessed(struct file *filp)
209 {
210 struct inode *ip = filp->f_mapping->host;
211
212 if (!IS_NOATIME(ip) && ITOZSB(ip)->z_relatime) {
213 if (zfs_relatime_need_update(ip))
214 file_accessed(filp);
215 } else {
216 file_accessed(filp);
217 }
218 }
219
220 static ssize_t
zpl_iter_read(struct kiocb * kiocb,struct iov_iter * to)221 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
222 {
223 cred_t *cr = CRED();
224 fstrans_cookie_t cookie;
225 struct file *filp = kiocb->ki_filp;
226 ssize_t count = iov_iter_count(to);
227 zfs_uio_t uio;
228
229 zfs_uio_iov_iter_init(&uio, to, kiocb->ki_pos, count, 0);
230
231 crhold(cr);
232 cookie = spl_fstrans_mark();
233
234 ssize_t ret = -zfs_read(ITOZ(filp->f_mapping->host), &uio,
235 filp->f_flags | zfs_io_flags(kiocb), cr);
236
237 spl_fstrans_unmark(cookie);
238 crfree(cr);
239
240 if (ret < 0)
241 return (ret);
242
243 ssize_t read = count - uio.uio_resid;
244 kiocb->ki_pos += read;
245
246 zpl_file_accessed(filp);
247
248 return (read);
249 }
250
251 static inline ssize_t
zpl_generic_write_checks(struct kiocb * kiocb,struct iov_iter * from,size_t * countp)252 zpl_generic_write_checks(struct kiocb *kiocb, struct iov_iter *from,
253 size_t *countp)
254 {
255 ssize_t ret = generic_write_checks(kiocb, from);
256 if (ret <= 0)
257 return (ret);
258
259 *countp = ret;
260
261 return (0);
262 }
263
264 static ssize_t
zpl_iter_write(struct kiocb * kiocb,struct iov_iter * from)265 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
266 {
267 cred_t *cr = CRED();
268 fstrans_cookie_t cookie;
269 struct file *filp = kiocb->ki_filp;
270 struct inode *ip = filp->f_mapping->host;
271 zfs_uio_t uio;
272 size_t count = 0;
273 ssize_t ret;
274
275 ret = zpl_generic_write_checks(kiocb, from, &count);
276 if (ret)
277 return (ret);
278
279 zfs_uio_iov_iter_init(&uio, from, kiocb->ki_pos, count,
280 from->iov_offset);
281
282 crhold(cr);
283 cookie = spl_fstrans_mark();
284
285 ret = -zfs_write(ITOZ(ip), &uio,
286 filp->f_flags | zfs_io_flags(kiocb), cr);
287
288 spl_fstrans_unmark(cookie);
289 crfree(cr);
290
291 if (ret < 0)
292 return (ret);
293
294 ssize_t wrote = count - uio.uio_resid;
295 kiocb->ki_pos += wrote;
296
297 return (wrote);
298 }
299
300 static ssize_t
zpl_direct_IO(struct kiocb * kiocb,struct iov_iter * iter)301 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter)
302 {
303 /*
304 * All O_DIRECT requests should be handled by
305 * zpl_iter_write/read}(). There is no way kernel generic code should
306 * call the direct_IO address_space_operations function. We set this
307 * code path to be fatal if it is executed.
308 */
309 PANIC(0);
310 return (0);
311 }
312
313 static loff_t
zpl_llseek(struct file * filp,loff_t offset,int whence)314 zpl_llseek(struct file *filp, loff_t offset, int whence)
315 {
316 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
317 fstrans_cookie_t cookie;
318
319 if (whence == SEEK_DATA || whence == SEEK_HOLE) {
320 struct inode *ip = filp->f_mapping->host;
321 loff_t maxbytes = ip->i_sb->s_maxbytes;
322 loff_t error;
323
324 spl_inode_lock_shared(ip);
325 cookie = spl_fstrans_mark();
326 error = -zfs_holey(ITOZ(ip), whence, &offset);
327 spl_fstrans_unmark(cookie);
328 if (error == 0)
329 error = lseek_execute(filp, ip, offset, maxbytes);
330 spl_inode_unlock_shared(ip);
331
332 return (error);
333 }
334 #endif /* SEEK_HOLE && SEEK_DATA */
335
336 return (generic_file_llseek(filp, offset, whence));
337 }
338
339 /*
340 * It's worth taking a moment to describe how mmap is implemented
341 * for zfs because it differs considerably from other Linux filesystems.
342 * However, this issue is handled the same way under OpenSolaris.
343 *
344 * The issue is that by design zfs bypasses the Linux page cache and
345 * leaves all caching up to the ARC. This has been shown to work
346 * well for the common read(2)/write(2) case. However, mmap(2)
347 * is problem because it relies on being tightly integrated with the
348 * page cache. To handle this we cache mmap'ed files twice, once in
349 * the ARC and a second time in the page cache. The code is careful
350 * to keep both copies synchronized.
351 *
352 * When a file with an mmap'ed region is written to using write(2)
353 * both the data in the ARC and existing pages in the page cache
354 * are updated. For a read(2) data will be read first from the page
355 * cache then the ARC if needed. Neither a write(2) or read(2) will
356 * will ever result in new pages being added to the page cache.
357 *
358 * New pages are added to the page cache only via .readpage() which
359 * is called when the vfs needs to read a page off disk to back the
360 * virtual memory region. These pages may be modified without
361 * notifying the ARC and will be written out periodically via
362 * .writepage(). This will occur due to either a sync or the usual
363 * page aging behavior. Note because a read(2) of a mmap'ed file
364 * will always check the page cache first even when the ARC is out
365 * of date correct data will still be returned.
366 *
367 * While this implementation ensures correct behavior it does have
368 * have some drawbacks. The most obvious of which is that it
369 * increases the required memory footprint when access mmap'ed
370 * files. It also adds additional complexity to the code keeping
371 * both caches synchronized.
372 *
373 * Longer term it may be possible to cleanly resolve this wart by
374 * mapping page cache pages directly on to the ARC buffers. The
375 * Linux address space operations are flexible enough to allow
376 * selection of which pages back a particular index. The trick
377 * would be working out the details of which subsystem is in
378 * charge, the ARC, the page cache, or both. It may also prove
379 * helpful to move the ARC buffers to a scatter-gather lists
380 * rather than a vmalloc'ed region.
381 */
382 static int
zpl_mmap(struct file * filp,struct vm_area_struct * vma)383 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
384 {
385 struct inode *ip = filp->f_mapping->host;
386 int error;
387 fstrans_cookie_t cookie;
388
389 cookie = spl_fstrans_mark();
390 error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
391 (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
392 spl_fstrans_unmark(cookie);
393
394 if (error)
395 return (error);
396
397 error = generic_file_mmap(filp, vma);
398 if (error)
399 return (error);
400
401 return (error);
402 }
403
404 /*
405 * Populate a page with data for the Linux page cache. This function is
406 * only used to support mmap(2). There will be an identical copy of the
407 * data in the ARC which is kept up to date via .write() and .writepage().
408 */
409 static inline int
zpl_readpage_common(struct page * pp)410 zpl_readpage_common(struct page *pp)
411 {
412 fstrans_cookie_t cookie;
413
414 ASSERT(PageLocked(pp));
415
416 cookie = spl_fstrans_mark();
417 int error = -zfs_getpage(pp->mapping->host, pp);
418 spl_fstrans_unmark(cookie);
419
420 unlock_page(pp);
421
422 return (error);
423 }
424
425 #ifdef HAVE_VFS_READ_FOLIO
426 static int
zpl_read_folio(struct file * filp,struct folio * folio)427 zpl_read_folio(struct file *filp, struct folio *folio)
428 {
429 return (zpl_readpage_common(&folio->page));
430 }
431 #else
432 static int
zpl_readpage(struct file * filp,struct page * pp)433 zpl_readpage(struct file *filp, struct page *pp)
434 {
435 return (zpl_readpage_common(pp));
436 }
437 #endif
438
439 static int
zpl_readpage_filler(void * data,struct page * pp)440 zpl_readpage_filler(void *data, struct page *pp)
441 {
442 return (zpl_readpage_common(pp));
443 }
444
445 /*
446 * Populate a set of pages with data for the Linux page cache. This
447 * function will only be called for read ahead and never for demand
448 * paging. For simplicity, the code relies on read_cache_pages() to
449 * correctly lock each page for IO and call zpl_readpage().
450 */
451 #ifdef HAVE_VFS_READPAGES
452 static int
zpl_readpages(struct file * filp,struct address_space * mapping,struct list_head * pages,unsigned nr_pages)453 zpl_readpages(struct file *filp, struct address_space *mapping,
454 struct list_head *pages, unsigned nr_pages)
455 {
456 return (read_cache_pages(mapping, pages, zpl_readpage_filler, NULL));
457 }
458 #else
459 static void
zpl_readahead(struct readahead_control * ractl)460 zpl_readahead(struct readahead_control *ractl)
461 {
462 struct page *page;
463
464 while ((page = readahead_page(ractl)) != NULL) {
465 int ret;
466
467 ret = zpl_readpage_filler(NULL, page);
468 put_page(page);
469 if (ret)
470 break;
471 }
472 }
473 #endif
474
475 static int
zpl_putpage(struct page * pp,struct writeback_control * wbc,void * data)476 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
477 {
478 boolean_t *for_sync = data;
479 fstrans_cookie_t cookie;
480 int ret;
481
482 ASSERT(PageLocked(pp));
483 ASSERT(!PageWriteback(pp));
484
485 cookie = spl_fstrans_mark();
486 ret = zfs_putpage(pp->mapping->host, pp, wbc, *for_sync);
487 spl_fstrans_unmark(cookie);
488
489 return (ret);
490 }
491
492 #ifdef HAVE_WRITEPAGE_T_FOLIO
493 static int
zpl_putfolio(struct folio * pp,struct writeback_control * wbc,void * data)494 zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data)
495 {
496 return (zpl_putpage(&pp->page, wbc, data));
497 }
498 #endif
499
500 static inline int
zpl_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,void * data)501 zpl_write_cache_pages(struct address_space *mapping,
502 struct writeback_control *wbc, void *data)
503 {
504 int result;
505
506 #ifdef HAVE_WRITEPAGE_T_FOLIO
507 result = write_cache_pages(mapping, wbc, zpl_putfolio, data);
508 #else
509 result = write_cache_pages(mapping, wbc, zpl_putpage, data);
510 #endif
511 return (result);
512 }
513
514 static int
zpl_writepages(struct address_space * mapping,struct writeback_control * wbc)515 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
516 {
517 znode_t *zp = ITOZ(mapping->host);
518 zfsvfs_t *zfsvfs = ITOZSB(mapping->host);
519 enum writeback_sync_modes sync_mode;
520 int result;
521
522 if ((result = zpl_enter(zfsvfs, FTAG)) != 0)
523 return (result);
524 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
525 wbc->sync_mode = WB_SYNC_ALL;
526 zpl_exit(zfsvfs, FTAG);
527 sync_mode = wbc->sync_mode;
528
529 /*
530 * We don't want to run write_cache_pages() in SYNC mode here, because
531 * that would make putpage() wait for a single page to be committed to
532 * disk every single time, resulting in atrocious performance. Instead
533 * we run it once in non-SYNC mode so that the ZIL gets all the data,
534 * and then we commit it all in one go.
535 */
536 boolean_t for_sync = (sync_mode == WB_SYNC_ALL);
537 wbc->sync_mode = WB_SYNC_NONE;
538 result = zpl_write_cache_pages(mapping, wbc, &for_sync);
539 if (sync_mode != wbc->sync_mode) {
540 if ((result = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
541 return (result);
542 if (zfsvfs->z_log != NULL)
543 zil_commit(zfsvfs->z_log, zp->z_id);
544 zpl_exit(zfsvfs, FTAG);
545
546 /*
547 * We need to call write_cache_pages() again (we can't just
548 * return after the commit) because the previous call in
549 * non-SYNC mode does not guarantee that we got all the dirty
550 * pages (see the implementation of write_cache_pages() for
551 * details). That being said, this is a no-op in most cases.
552 */
553 wbc->sync_mode = sync_mode;
554 result = zpl_write_cache_pages(mapping, wbc, &for_sync);
555 }
556 return (result);
557 }
558
559 /*
560 * Write out dirty pages to the ARC, this function is only required to
561 * support mmap(2). Mapped pages may be dirtied by memory operations
562 * which never call .write(). These dirty pages are kept in sync with
563 * the ARC buffers via this hook.
564 */
565 static int
zpl_writepage(struct page * pp,struct writeback_control * wbc)566 zpl_writepage(struct page *pp, struct writeback_control *wbc)
567 {
568 if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
569 wbc->sync_mode = WB_SYNC_ALL;
570
571 boolean_t for_sync = (wbc->sync_mode == WB_SYNC_ALL);
572
573 return (zpl_putpage(pp, wbc, &for_sync));
574 }
575
576 /*
577 * The flag combination which matches the behavior of zfs_space() is
578 * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE. The FALLOC_FL_PUNCH_HOLE
579 * flag was introduced in the 2.6.38 kernel.
580 *
581 * The original mode=0 (allocate space) behavior can be reasonably emulated
582 * by checking if enough space exists and creating a sparse file, as real
583 * persistent space reservation is not possible due to COW, snapshots, etc.
584 */
585 static long
zpl_fallocate_common(struct inode * ip,int mode,loff_t offset,loff_t len)586 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
587 {
588 cred_t *cr = CRED();
589 loff_t olen;
590 fstrans_cookie_t cookie;
591 int error = 0;
592
593 int test_mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE;
594
595 if ((mode & ~(FALLOC_FL_KEEP_SIZE | test_mode)) != 0)
596 return (-EOPNOTSUPP);
597
598 if (offset < 0 || len <= 0)
599 return (-EINVAL);
600
601 spl_inode_lock(ip);
602 olen = i_size_read(ip);
603
604 crhold(cr);
605 cookie = spl_fstrans_mark();
606 if (mode & (test_mode)) {
607 flock64_t bf;
608
609 if (mode & FALLOC_FL_KEEP_SIZE) {
610 if (offset > olen)
611 goto out_unmark;
612
613 if (offset + len > olen)
614 len = olen - offset;
615 }
616 bf.l_type = F_WRLCK;
617 bf.l_whence = SEEK_SET;
618 bf.l_start = offset;
619 bf.l_len = len;
620 bf.l_pid = 0;
621
622 error = -zfs_space(ITOZ(ip), F_FREESP, &bf, O_RDWR, offset, cr);
623 } else if ((mode & ~FALLOC_FL_KEEP_SIZE) == 0) {
624 unsigned int percent = zfs_fallocate_reserve_percent;
625 struct kstatfs statfs;
626
627 /* Legacy mode, disable fallocate compatibility. */
628 if (percent == 0) {
629 error = -EOPNOTSUPP;
630 goto out_unmark;
631 }
632
633 /*
634 * Use zfs_statvfs() instead of dmu_objset_space() since it
635 * also checks project quota limits, which are relevant here.
636 */
637 error = zfs_statvfs(ip, &statfs);
638 if (error)
639 goto out_unmark;
640
641 /*
642 * Shrink available space a bit to account for overhead/races.
643 * We know the product previously fit into availbytes from
644 * dmu_objset_space(), so the smaller product will also fit.
645 */
646 if (len > statfs.f_bavail * (statfs.f_bsize * 100 / percent)) {
647 error = -ENOSPC;
648 goto out_unmark;
649 }
650 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > olen)
651 error = zfs_freesp(ITOZ(ip), offset + len, 0, 0, FALSE);
652 }
653 out_unmark:
654 spl_fstrans_unmark(cookie);
655 spl_inode_unlock(ip);
656
657 crfree(cr);
658
659 return (error);
660 }
661
662 static long
zpl_fallocate(struct file * filp,int mode,loff_t offset,loff_t len)663 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
664 {
665 return zpl_fallocate_common(file_inode(filp),
666 mode, offset, len);
667 }
668
669 static int
zpl_ioctl_getversion(struct file * filp,void __user * arg)670 zpl_ioctl_getversion(struct file *filp, void __user *arg)
671 {
672 uint32_t generation = file_inode(filp)->i_generation;
673
674 return (copy_to_user(arg, &generation, sizeof (generation)));
675 }
676
677 static int
zpl_fadvise(struct file * filp,loff_t offset,loff_t len,int advice)678 zpl_fadvise(struct file *filp, loff_t offset, loff_t len, int advice)
679 {
680 struct inode *ip = file_inode(filp);
681 znode_t *zp = ITOZ(ip);
682 zfsvfs_t *zfsvfs = ITOZSB(ip);
683 objset_t *os = zfsvfs->z_os;
684 int error = 0;
685
686 if (S_ISFIFO(ip->i_mode))
687 return (-ESPIPE);
688
689 if (offset < 0 || len < 0)
690 return (-EINVAL);
691
692 if ((error = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
693 return (error);
694
695 switch (advice) {
696 case POSIX_FADV_SEQUENTIAL:
697 case POSIX_FADV_WILLNEED:
698 #ifdef HAVE_GENERIC_FADVISE
699 if (zn_has_cached_data(zp, offset, offset + len - 1))
700 error = generic_fadvise(filp, offset, len, advice);
701 #endif
702 /*
703 * Pass on the caller's size directly, but note that
704 * dmu_prefetch_max will effectively cap it. If there
705 * really is a larger sequential access pattern, perhaps
706 * dmu_zfetch will detect it.
707 */
708 if (len == 0)
709 len = i_size_read(ip) - offset;
710
711 dmu_prefetch(os, zp->z_id, 0, offset, len,
712 ZIO_PRIORITY_ASYNC_READ);
713 break;
714 case POSIX_FADV_NORMAL:
715 case POSIX_FADV_RANDOM:
716 case POSIX_FADV_DONTNEED:
717 case POSIX_FADV_NOREUSE:
718 /* ignored for now */
719 break;
720 default:
721 error = -EINVAL;
722 break;
723 }
724
725 zfs_exit(zfsvfs, FTAG);
726
727 return (error);
728 }
729
730 #define ZFS_FL_USER_VISIBLE (FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
731 #define ZFS_FL_USER_MODIFIABLE (FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
732
733 static uint32_t
__zpl_ioctl_getflags(struct inode * ip)734 __zpl_ioctl_getflags(struct inode *ip)
735 {
736 uint64_t zfs_flags = ITOZ(ip)->z_pflags;
737 uint32_t ioctl_flags = 0;
738
739 if (zfs_flags & ZFS_IMMUTABLE)
740 ioctl_flags |= FS_IMMUTABLE_FL;
741
742 if (zfs_flags & ZFS_APPENDONLY)
743 ioctl_flags |= FS_APPEND_FL;
744
745 if (zfs_flags & ZFS_NODUMP)
746 ioctl_flags |= FS_NODUMP_FL;
747
748 if (zfs_flags & ZFS_PROJINHERIT)
749 ioctl_flags |= ZFS_PROJINHERIT_FL;
750
751 return (ioctl_flags & ZFS_FL_USER_VISIBLE);
752 }
753
754 /*
755 * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
756 * attributes common to both Linux and Solaris are mapped.
757 */
758 static int
zpl_ioctl_getflags(struct file * filp,void __user * arg)759 zpl_ioctl_getflags(struct file *filp, void __user *arg)
760 {
761 uint32_t flags;
762 int err;
763
764 flags = __zpl_ioctl_getflags(file_inode(filp));
765 err = copy_to_user(arg, &flags, sizeof (flags));
766
767 return (err);
768 }
769
770 /*
771 * fchange() is a helper macro to detect if we have been asked to change a
772 * flag. This is ugly, but the requirement that we do this is a consequence of
773 * how the Linux file attribute interface was designed. Another consequence is
774 * that concurrent modification of files suffers from a TOCTOU race. Neither
775 * are things we can fix without modifying the kernel-userland interface, which
776 * is outside of our jurisdiction.
777 */
778
779 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
780
781 static int
__zpl_ioctl_setflags(struct inode * ip,uint32_t ioctl_flags,xvattr_t * xva)782 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva)
783 {
784 uint64_t zfs_flags = ITOZ(ip)->z_pflags;
785 xoptattr_t *xoap;
786
787 if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL |
788 ZFS_PROJINHERIT_FL))
789 return (-EOPNOTSUPP);
790
791 if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE)
792 return (-EACCES);
793
794 if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
795 fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
796 !capable(CAP_LINUX_IMMUTABLE))
797 return (-EPERM);
798
799 if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
800 return (-EACCES);
801
802 xva_init(xva);
803 xoap = xva_getxoptattr(xva);
804
805 #define FLAG_CHANGE(iflag, zflag, xflag, xfield) do { \
806 if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) || \
807 ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) { \
808 XVA_SET_REQ(xva, (xflag)); \
809 (xfield) = ((ioctl_flags & (iflag)) != 0); \
810 } \
811 } while (0)
812
813 FLAG_CHANGE(FS_IMMUTABLE_FL, ZFS_IMMUTABLE, XAT_IMMUTABLE,
814 xoap->xoa_immutable);
815 FLAG_CHANGE(FS_APPEND_FL, ZFS_APPENDONLY, XAT_APPENDONLY,
816 xoap->xoa_appendonly);
817 FLAG_CHANGE(FS_NODUMP_FL, ZFS_NODUMP, XAT_NODUMP,
818 xoap->xoa_nodump);
819 FLAG_CHANGE(ZFS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT,
820 xoap->xoa_projinherit);
821
822 #undef FLAG_CHANGE
823
824 return (0);
825 }
826
827 static int
zpl_ioctl_setflags(struct file * filp,void __user * arg)828 zpl_ioctl_setflags(struct file *filp, void __user *arg)
829 {
830 struct inode *ip = file_inode(filp);
831 uint32_t flags;
832 cred_t *cr = CRED();
833 xvattr_t xva;
834 int err;
835 fstrans_cookie_t cookie;
836
837 if (copy_from_user(&flags, arg, sizeof (flags)))
838 return (-EFAULT);
839
840 err = __zpl_ioctl_setflags(ip, flags, &xva);
841 if (err)
842 return (err);
843
844 crhold(cr);
845 cookie = spl_fstrans_mark();
846 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
847 spl_fstrans_unmark(cookie);
848 crfree(cr);
849
850 return (err);
851 }
852
853 static int
zpl_ioctl_getxattr(struct file * filp,void __user * arg)854 zpl_ioctl_getxattr(struct file *filp, void __user *arg)
855 {
856 zfsxattr_t fsx = { 0 };
857 struct inode *ip = file_inode(filp);
858 int err;
859
860 fsx.fsx_xflags = __zpl_ioctl_getflags(ip);
861 fsx.fsx_projid = ITOZ(ip)->z_projid;
862 err = copy_to_user(arg, &fsx, sizeof (fsx));
863
864 return (err);
865 }
866
867 static int
zpl_ioctl_setxattr(struct file * filp,void __user * arg)868 zpl_ioctl_setxattr(struct file *filp, void __user *arg)
869 {
870 struct inode *ip = file_inode(filp);
871 zfsxattr_t fsx;
872 cred_t *cr = CRED();
873 xvattr_t xva;
874 xoptattr_t *xoap;
875 int err;
876 fstrans_cookie_t cookie;
877
878 if (copy_from_user(&fsx, arg, sizeof (fsx)))
879 return (-EFAULT);
880
881 if (!zpl_is_valid_projid(fsx.fsx_projid))
882 return (-EINVAL);
883
884 err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva);
885 if (err)
886 return (err);
887
888 xoap = xva_getxoptattr(&xva);
889 XVA_SET_REQ(&xva, XAT_PROJID);
890 xoap->xoa_projid = fsx.fsx_projid;
891
892 crhold(cr);
893 cookie = spl_fstrans_mark();
894 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
895 spl_fstrans_unmark(cookie);
896 crfree(cr);
897
898 return (err);
899 }
900
901 /*
902 * Expose Additional File Level Attributes of ZFS.
903 */
904 static int
zpl_ioctl_getdosflags(struct file * filp,void __user * arg)905 zpl_ioctl_getdosflags(struct file *filp, void __user *arg)
906 {
907 struct inode *ip = file_inode(filp);
908 uint64_t dosflags = ITOZ(ip)->z_pflags;
909 dosflags &= ZFS_DOS_FL_USER_VISIBLE;
910 int err = copy_to_user(arg, &dosflags, sizeof (dosflags));
911
912 return (err);
913 }
914
915 static int
__zpl_ioctl_setdosflags(struct inode * ip,uint64_t ioctl_flags,xvattr_t * xva)916 __zpl_ioctl_setdosflags(struct inode *ip, uint64_t ioctl_flags, xvattr_t *xva)
917 {
918 uint64_t zfs_flags = ITOZ(ip)->z_pflags;
919 xoptattr_t *xoap;
920
921 if (ioctl_flags & (~ZFS_DOS_FL_USER_VISIBLE))
922 return (-EOPNOTSUPP);
923
924 if ((fchange(ioctl_flags, zfs_flags, ZFS_IMMUTABLE, ZFS_IMMUTABLE) ||
925 fchange(ioctl_flags, zfs_flags, ZFS_APPENDONLY, ZFS_APPENDONLY)) &&
926 !capable(CAP_LINUX_IMMUTABLE))
927 return (-EPERM);
928
929 if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
930 return (-EACCES);
931
932 xva_init(xva);
933 xoap = xva_getxoptattr(xva);
934
935 #define FLAG_CHANGE(iflag, xflag, xfield) do { \
936 if (((ioctl_flags & (iflag)) && !(zfs_flags & (iflag))) || \
937 ((zfs_flags & (iflag)) && !(ioctl_flags & (iflag)))) { \
938 XVA_SET_REQ(xva, (xflag)); \
939 (xfield) = ((ioctl_flags & (iflag)) != 0); \
940 } \
941 } while (0)
942
943 FLAG_CHANGE(ZFS_IMMUTABLE, XAT_IMMUTABLE, xoap->xoa_immutable);
944 FLAG_CHANGE(ZFS_APPENDONLY, XAT_APPENDONLY, xoap->xoa_appendonly);
945 FLAG_CHANGE(ZFS_NODUMP, XAT_NODUMP, xoap->xoa_nodump);
946 FLAG_CHANGE(ZFS_READONLY, XAT_READONLY, xoap->xoa_readonly);
947 FLAG_CHANGE(ZFS_HIDDEN, XAT_HIDDEN, xoap->xoa_hidden);
948 FLAG_CHANGE(ZFS_SYSTEM, XAT_SYSTEM, xoap->xoa_system);
949 FLAG_CHANGE(ZFS_ARCHIVE, XAT_ARCHIVE, xoap->xoa_archive);
950 FLAG_CHANGE(ZFS_NOUNLINK, XAT_NOUNLINK, xoap->xoa_nounlink);
951 FLAG_CHANGE(ZFS_REPARSE, XAT_REPARSE, xoap->xoa_reparse);
952 FLAG_CHANGE(ZFS_OFFLINE, XAT_OFFLINE, xoap->xoa_offline);
953 FLAG_CHANGE(ZFS_SPARSE, XAT_SPARSE, xoap->xoa_sparse);
954
955 #undef FLAG_CHANGE
956
957 return (0);
958 }
959
960 /*
961 * Set Additional File Level Attributes of ZFS.
962 */
963 static int
zpl_ioctl_setdosflags(struct file * filp,void __user * arg)964 zpl_ioctl_setdosflags(struct file *filp, void __user *arg)
965 {
966 struct inode *ip = file_inode(filp);
967 uint64_t dosflags;
968 cred_t *cr = CRED();
969 xvattr_t xva;
970 int err;
971 fstrans_cookie_t cookie;
972
973 if (copy_from_user(&dosflags, arg, sizeof (dosflags)))
974 return (-EFAULT);
975
976 err = __zpl_ioctl_setdosflags(ip, dosflags, &xva);
977 if (err)
978 return (err);
979
980 crhold(cr);
981 cookie = spl_fstrans_mark();
982 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
983 spl_fstrans_unmark(cookie);
984 crfree(cr);
985
986 return (err);
987 }
988
989 static long
zpl_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)990 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
991 {
992 switch (cmd) {
993 case FS_IOC_GETVERSION:
994 return (zpl_ioctl_getversion(filp, (void *)arg));
995 case FS_IOC_GETFLAGS:
996 return (zpl_ioctl_getflags(filp, (void *)arg));
997 case FS_IOC_SETFLAGS:
998 return (zpl_ioctl_setflags(filp, (void *)arg));
999 case ZFS_IOC_FSGETXATTR:
1000 return (zpl_ioctl_getxattr(filp, (void *)arg));
1001 case ZFS_IOC_FSSETXATTR:
1002 return (zpl_ioctl_setxattr(filp, (void *)arg));
1003 case ZFS_IOC_GETDOSFLAGS:
1004 return (zpl_ioctl_getdosflags(filp, (void *)arg));
1005 case ZFS_IOC_SETDOSFLAGS:
1006 return (zpl_ioctl_setdosflags(filp, (void *)arg));
1007 case ZFS_IOC_COMPAT_FICLONE:
1008 return (zpl_ioctl_ficlone(filp, (void *)arg));
1009 case ZFS_IOC_COMPAT_FICLONERANGE:
1010 return (zpl_ioctl_ficlonerange(filp, (void *)arg));
1011 case ZFS_IOC_COMPAT_FIDEDUPERANGE:
1012 return (zpl_ioctl_fideduperange(filp, (void *)arg));
1013 default:
1014 return (-ENOTTY);
1015 }
1016 }
1017
1018 #ifdef CONFIG_COMPAT
1019 static long
zpl_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1020 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1021 {
1022 switch (cmd) {
1023 case FS_IOC32_GETVERSION:
1024 cmd = FS_IOC_GETVERSION;
1025 break;
1026 case FS_IOC32_GETFLAGS:
1027 cmd = FS_IOC_GETFLAGS;
1028 break;
1029 case FS_IOC32_SETFLAGS:
1030 cmd = FS_IOC_SETFLAGS;
1031 break;
1032 default:
1033 return (-ENOTTY);
1034 }
1035 return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
1036 }
1037 #endif /* CONFIG_COMPAT */
1038
1039 const struct address_space_operations zpl_address_space_operations = {
1040 #ifdef HAVE_VFS_READPAGES
1041 .readpages = zpl_readpages,
1042 #else
1043 .readahead = zpl_readahead,
1044 #endif
1045 #ifdef HAVE_VFS_READ_FOLIO
1046 .read_folio = zpl_read_folio,
1047 #else
1048 .readpage = zpl_readpage,
1049 #endif
1050 .writepage = zpl_writepage,
1051 .writepages = zpl_writepages,
1052 .direct_IO = zpl_direct_IO,
1053 #ifdef HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS
1054 .set_page_dirty = __set_page_dirty_nobuffers,
1055 #endif
1056 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
1057 .dirty_folio = filemap_dirty_folio,
1058 #endif
1059 #ifdef HAVE_VFS_MIGRATE_FOLIO
1060 .migrate_folio = migrate_folio,
1061 #else
1062 .migratepage = migrate_page,
1063 #endif
1064 };
1065
1066 const struct file_operations zpl_file_operations = {
1067 .open = zpl_open,
1068 .release = zpl_release,
1069 .llseek = zpl_llseek,
1070 .read_iter = zpl_iter_read,
1071 .write_iter = zpl_iter_write,
1072 #ifdef HAVE_COPY_SPLICE_READ
1073 .splice_read = copy_splice_read,
1074 #else
1075 .splice_read = generic_file_splice_read,
1076 #endif
1077 .splice_write = iter_file_splice_write,
1078 .mmap = zpl_mmap,
1079 .fsync = zpl_fsync,
1080 .fallocate = zpl_fallocate,
1081 .copy_file_range = zpl_copy_file_range,
1082 #ifdef HAVE_VFS_CLONE_FILE_RANGE
1083 .clone_file_range = zpl_clone_file_range,
1084 #endif
1085 #ifdef HAVE_VFS_REMAP_FILE_RANGE
1086 .remap_file_range = zpl_remap_file_range,
1087 #endif
1088 #ifdef HAVE_VFS_DEDUPE_FILE_RANGE
1089 .dedupe_file_range = zpl_dedupe_file_range,
1090 #endif
1091 .fadvise = zpl_fadvise,
1092 .unlocked_ioctl = zpl_ioctl,
1093 #ifdef CONFIG_COMPAT
1094 .compat_ioctl = zpl_compat_ioctl,
1095 #endif
1096 };
1097
1098 const struct file_operations zpl_dir_file_operations = {
1099 .llseek = generic_file_llseek,
1100 .read = generic_read_dir,
1101 .iterate_shared = zpl_iterate,
1102 .fsync = zpl_fsync,
1103 .unlocked_ioctl = zpl_ioctl,
1104 #ifdef CONFIG_COMPAT
1105 .compat_ioctl = zpl_compat_ioctl,
1106 #endif
1107 };
1108
1109 module_param(zfs_fallocate_reserve_percent, uint, 0644);
1110 MODULE_PARM_DESC(zfs_fallocate_reserve_percent,
1111 "Percentage of length to use for the available capacity check");
1112