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 * Copyright (c) 2025, Klara, Inc.
26 * Copyright (c) 2025, Rob Norris <robn@despairlabs.com>
27 */
28
29
30 #ifdef CONFIG_COMPAT
31 #include <linux/compat.h>
32 #endif
33 #include <linux/fs.h>
34 #include <linux/migrate.h>
35 #include <sys/file.h>
36 #include <sys/dmu_objset.h>
37 #include <sys/zfs_znode.h>
38 #include <sys/zfs_vfsops.h>
39 #include <sys/zfs_vnops.h>
40 #include <sys/zfs_project.h>
41 #include <linux/pagemap_compat.h>
42 #include <linux/fadvise.h>
43 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
44 #include <linux/writeback.h>
45 #endif
46
47 /*
48 * When using fallocate(2) to preallocate space, inflate the requested
49 * capacity check by 10% to account for the required metadata blocks.
50 */
51 static unsigned int zfs_fallocate_reserve_percent = 110;
52
53 static int
zpl_open(struct inode * ip,struct file * filp)54 zpl_open(struct inode *ip, struct file *filp)
55 {
56 cred_t *cr = CRED();
57 int error;
58 fstrans_cookie_t cookie;
59
60 error = generic_file_open(ip, filp);
61 if (error)
62 return (error);
63
64 crhold(cr);
65 cookie = spl_fstrans_mark();
66 error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
67 spl_fstrans_unmark(cookie);
68 crfree(cr);
69 ASSERT3S(error, <=, 0);
70
71 return (error);
72 }
73
74 static int
zpl_release(struct inode * ip,struct file * filp)75 zpl_release(struct inode *ip, struct file *filp)
76 {
77 cred_t *cr = CRED();
78 int error;
79 fstrans_cookie_t cookie;
80
81 cookie = spl_fstrans_mark();
82 if (ITOZ(ip)->z_atime_dirty)
83 zfs_mark_inode_dirty(ip);
84
85 crhold(cr);
86 error = -zfs_close(ip, filp->f_flags, cr);
87 spl_fstrans_unmark(cookie);
88 crfree(cr);
89 ASSERT3S(error, <=, 0);
90
91 return (error);
92 }
93
94 static int
zpl_iterate(struct file * filp,struct dir_context * ctx)95 zpl_iterate(struct file *filp, struct dir_context *ctx)
96 {
97 cred_t *cr = CRED();
98 int error;
99 fstrans_cookie_t cookie;
100
101 crhold(cr);
102 cookie = spl_fstrans_mark();
103 error = -zfs_readdir(file_inode(filp), ctx, cr);
104 spl_fstrans_unmark(cookie);
105 crfree(cr);
106 ASSERT3S(error, <=, 0);
107
108 return (error);
109 }
110
111 static inline int
112 zpl_write_cache_pages(struct address_space *mapping,
113 struct writeback_control *wbc, void *data);
114
115 static int
zpl_fsync(struct file * filp,loff_t start,loff_t end,int datasync)116 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
117 {
118 struct inode *inode = filp->f_mapping->host;
119 znode_t *zp = ITOZ(inode);
120 cred_t *cr = CRED();
121 int error;
122 fstrans_cookie_t cookie;
123
124 /*
125 * Force dirty pages in the range out to the DMU and the log, ready
126 * for zil_commit() to write down.
127 *
128 * We call write_cache_pages() directly to ensure that zpl_putpage() is
129 * called with the flags we need. We need WB_SYNC_NONE to avoid a call
130 * to zil_commit() (since we're doing this as a kind of pre-sync); but
131 * we do need for_sync so that the pages remain in writeback until
132 * they're on disk, and so that we get an error if the DMU write fails.
133 */
134 if (filemap_range_has_page(inode->i_mapping, start, end)) {
135 int for_sync = 1;
136 struct writeback_control wbc = {
137 .sync_mode = WB_SYNC_NONE,
138 .nr_to_write = LONG_MAX,
139 .range_start = start,
140 .range_end = end,
141 };
142 error =
143 zpl_write_cache_pages(inode->i_mapping, &wbc, &for_sync);
144 if (error != 0) {
145 /*
146 * Unclear what state things are in. zfs_putpage() will
147 * ensure the pages remain dirty if they haven't been
148 * written down to the DMU, but because there may be
149 * nothing logged, we can't assume that zfs_sync() ->
150 * zil_commit() will give us a useful error. It's
151 * safest if we just error out here.
152 */
153 return (error);
154 }
155 }
156
157 crhold(cr);
158 cookie = spl_fstrans_mark();
159 error = -zfs_fsync(zp, datasync, cr);
160 spl_fstrans_unmark(cookie);
161 crfree(cr);
162 ASSERT3S(error, <=, 0);
163
164 return (error);
165 }
166
167 static inline int
zfs_io_flags(struct kiocb * kiocb)168 zfs_io_flags(struct kiocb *kiocb)
169 {
170 int flags = 0;
171
172 #if defined(IOCB_DSYNC)
173 if (kiocb->ki_flags & IOCB_DSYNC)
174 flags |= O_DSYNC;
175 #endif
176 #if defined(IOCB_SYNC)
177 if (kiocb->ki_flags & IOCB_SYNC)
178 flags |= O_SYNC;
179 #endif
180 #if defined(IOCB_APPEND)
181 if (kiocb->ki_flags & IOCB_APPEND)
182 flags |= O_APPEND;
183 #endif
184 #if defined(IOCB_DIRECT)
185 if (kiocb->ki_flags & IOCB_DIRECT)
186 flags |= O_DIRECT;
187 #endif
188 return (flags);
189 }
190
191 /*
192 * If relatime is enabled, call file_accessed() if zfs_relatime_need_update()
193 * is true. This is needed since datasets with inherited "relatime" property
194 * aren't necessarily mounted with the MNT_RELATIME flag (e.g. after
195 * `zfs set relatime=...`), which is what relatime test in VFS by
196 * relatime_need_update() is based on.
197 */
198 static inline void
zpl_file_accessed(struct file * filp)199 zpl_file_accessed(struct file *filp)
200 {
201 struct inode *ip = filp->f_mapping->host;
202
203 if (!IS_NOATIME(ip) && ITOZSB(ip)->z_relatime) {
204 if (zfs_relatime_need_update(ip))
205 file_accessed(filp);
206 } else {
207 file_accessed(filp);
208 }
209 }
210
211 static ssize_t
zpl_iter_read(struct kiocb * kiocb,struct iov_iter * to)212 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
213 {
214 cred_t *cr = CRED();
215 fstrans_cookie_t cookie;
216 struct file *filp = kiocb->ki_filp;
217 ssize_t count = iov_iter_count(to);
218 zfs_uio_t uio;
219
220 zfs_uio_iov_iter_init(&uio, to, kiocb->ki_pos, count);
221
222 crhold(cr);
223 cookie = spl_fstrans_mark();
224
225 ssize_t ret = -zfs_read(ITOZ(filp->f_mapping->host), &uio,
226 filp->f_flags | zfs_io_flags(kiocb), cr);
227
228 spl_fstrans_unmark(cookie);
229 crfree(cr);
230
231 if (ret < 0)
232 return (ret);
233
234 ssize_t read = count - uio.uio_resid;
235 kiocb->ki_pos += read;
236
237 zpl_file_accessed(filp);
238
239 return (read);
240 }
241
242 static inline ssize_t
zpl_generic_write_checks(struct kiocb * kiocb,struct iov_iter * from,size_t * countp)243 zpl_generic_write_checks(struct kiocb *kiocb, struct iov_iter *from,
244 size_t *countp)
245 {
246 ssize_t ret = generic_write_checks(kiocb, from);
247 if (ret <= 0)
248 return (ret);
249
250 *countp = ret;
251
252 return (0);
253 }
254
255 static ssize_t
zpl_iter_write(struct kiocb * kiocb,struct iov_iter * from)256 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
257 {
258 cred_t *cr = CRED();
259 fstrans_cookie_t cookie;
260 struct file *filp = kiocb->ki_filp;
261 struct inode *ip = filp->f_mapping->host;
262 zfs_uio_t uio;
263 size_t count = 0;
264 ssize_t ret;
265
266 ret = zpl_generic_write_checks(kiocb, from, &count);
267 if (ret)
268 return (ret);
269
270 zfs_uio_iov_iter_init(&uio, from, kiocb->ki_pos, count);
271
272 crhold(cr);
273 cookie = spl_fstrans_mark();
274
275 ret = -zfs_write(ITOZ(ip), &uio,
276 filp->f_flags | zfs_io_flags(kiocb), cr);
277
278 spl_fstrans_unmark(cookie);
279 crfree(cr);
280
281 if (ret < 0)
282 return (ret);
283
284 ssize_t wrote = count - uio.uio_resid;
285 kiocb->ki_pos += wrote;
286
287 return (wrote);
288 }
289
290 static ssize_t
zpl_direct_IO(struct kiocb * kiocb,struct iov_iter * iter)291 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter)
292 {
293 /*
294 * All O_DIRECT requests should be handled by
295 * zpl_iter_write/read}(). There is no way kernel generic code should
296 * call the direct_IO address_space_operations function. We set this
297 * code path to be fatal if it is executed.
298 */
299 PANIC(0);
300 return (0);
301 }
302
303 static loff_t
zpl_llseek(struct file * filp,loff_t offset,int whence)304 zpl_llseek(struct file *filp, loff_t offset, int whence)
305 {
306 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
307 fstrans_cookie_t cookie;
308
309 if (whence == SEEK_DATA || whence == SEEK_HOLE) {
310 struct inode *ip = filp->f_mapping->host;
311 loff_t maxbytes = ip->i_sb->s_maxbytes;
312 loff_t error;
313
314 spl_inode_lock_shared(ip);
315 cookie = spl_fstrans_mark();
316 error = -zfs_holey(ITOZ(ip), whence, &offset);
317 spl_fstrans_unmark(cookie);
318 if (error == 0)
319 error = lseek_execute(filp, ip, offset, maxbytes);
320 spl_inode_unlock_shared(ip);
321
322 return (error);
323 }
324 #endif /* SEEK_HOLE && SEEK_DATA */
325
326 return (generic_file_llseek(filp, offset, whence));
327 }
328
329 /*
330 * It's worth taking a moment to describe how mmap is implemented
331 * for zfs because it differs considerably from other Linux filesystems.
332 * However, this issue is handled the same way under OpenSolaris.
333 *
334 * The issue is that by design zfs bypasses the Linux page cache and
335 * leaves all caching up to the ARC. This has been shown to work
336 * well for the common read(2)/write(2) case. However, mmap(2)
337 * is problem because it relies on being tightly integrated with the
338 * page cache. To handle this we cache mmap'ed files twice, once in
339 * the ARC and a second time in the page cache. The code is careful
340 * to keep both copies synchronized.
341 *
342 * When a file with an mmap'ed region is written to using write(2)
343 * both the data in the ARC and existing pages in the page cache
344 * are updated. For a read(2) data will be read first from the page
345 * cache then the ARC if needed. Neither a write(2) or read(2) will
346 * will ever result in new pages being added to the page cache.
347 *
348 * New pages are added to the page cache only via .readpage() which
349 * is called when the vfs needs to read a page off disk to back the
350 * virtual memory region. These pages may be modified without
351 * notifying the ARC and will be written out periodically via
352 * .writepage(). This will occur due to either a sync or the usual
353 * page aging behavior. Note because a read(2) of a mmap'ed file
354 * will always check the page cache first even when the ARC is out
355 * of date correct data will still be returned.
356 *
357 * While this implementation ensures correct behavior it does have
358 * have some drawbacks. The most obvious of which is that it
359 * increases the required memory footprint when access mmap'ed
360 * files. It also adds additional complexity to the code keeping
361 * both caches synchronized.
362 *
363 * Longer term it may be possible to cleanly resolve this wart by
364 * mapping page cache pages directly on to the ARC buffers. The
365 * Linux address space operations are flexible enough to allow
366 * selection of which pages back a particular index. The trick
367 * would be working out the details of which subsystem is in
368 * charge, the ARC, the page cache, or both. It may also prove
369 * helpful to move the ARC buffers to a scatter-gather lists
370 * rather than a vmalloc'ed region.
371 */
372 static int
zpl_mmap(struct file * filp,struct vm_area_struct * vma)373 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
374 {
375 struct inode *ip = filp->f_mapping->host;
376 int error;
377 fstrans_cookie_t cookie;
378
379 cookie = spl_fstrans_mark();
380 error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
381 (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
382 spl_fstrans_unmark(cookie);
383
384 if (error)
385 return (error);
386
387 error = generic_file_mmap(filp, vma);
388 if (error)
389 return (error);
390
391 return (error);
392 }
393
394 /*
395 * Populate a page with data for the Linux page cache. This function is
396 * only used to support mmap(2). There will be an identical copy of the
397 * data in the ARC which is kept up to date via .write() and .writepage().
398 */
399 static inline int
zpl_readpage_common(struct page * pp)400 zpl_readpage_common(struct page *pp)
401 {
402 fstrans_cookie_t cookie;
403
404 ASSERT(PageLocked(pp));
405
406 cookie = spl_fstrans_mark();
407 int error = -zfs_getpage(pp->mapping->host, pp);
408 spl_fstrans_unmark(cookie);
409
410 unlock_page(pp);
411
412 return (error);
413 }
414
415 #ifdef HAVE_VFS_READ_FOLIO
416 static int
zpl_read_folio(struct file * filp,struct folio * folio)417 zpl_read_folio(struct file *filp, struct folio *folio)
418 {
419 return (zpl_readpage_common(&folio->page));
420 }
421 #else
422 static int
zpl_readpage(struct file * filp,struct page * pp)423 zpl_readpage(struct file *filp, struct page *pp)
424 {
425 return (zpl_readpage_common(pp));
426 }
427 #endif
428
429 static int
zpl_readpage_filler(void * data,struct page * pp)430 zpl_readpage_filler(void *data, struct page *pp)
431 {
432 return (zpl_readpage_common(pp));
433 }
434
435 /*
436 * Populate a set of pages with data for the Linux page cache. This
437 * function will only be called for read ahead and never for demand
438 * paging. For simplicity, the code relies on read_cache_pages() to
439 * correctly lock each page for IO and call zpl_readpage().
440 */
441 #ifdef HAVE_VFS_READPAGES
442 static int
zpl_readpages(struct file * filp,struct address_space * mapping,struct list_head * pages,unsigned nr_pages)443 zpl_readpages(struct file *filp, struct address_space *mapping,
444 struct list_head *pages, unsigned nr_pages)
445 {
446 return (read_cache_pages(mapping, pages, zpl_readpage_filler, NULL));
447 }
448 #else
449 static void
zpl_readahead(struct readahead_control * ractl)450 zpl_readahead(struct readahead_control *ractl)
451 {
452 struct page *page;
453
454 while ((page = readahead_page(ractl)) != NULL) {
455 int ret;
456
457 ret = zpl_readpage_filler(NULL, page);
458 put_page(page);
459 if (ret)
460 break;
461 }
462 }
463 #endif
464
465 static int
zpl_putpage(struct page * pp,struct writeback_control * wbc,void * data)466 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
467 {
468 boolean_t *for_sync = data;
469 fstrans_cookie_t cookie;
470 int ret;
471
472 ASSERT(PageLocked(pp));
473 ASSERT(!PageWriteback(pp));
474
475 cookie = spl_fstrans_mark();
476 ret = zfs_putpage(pp->mapping->host, pp, wbc, *for_sync);
477 spl_fstrans_unmark(cookie);
478
479 return (ret);
480 }
481
482 #ifdef HAVE_WRITE_CACHE_PAGES
483 #ifdef HAVE_WRITEPAGE_T_FOLIO
484 static int
zpl_putfolio(struct folio * pp,struct writeback_control * wbc,void * data)485 zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data)
486 {
487 return (zpl_putpage(&pp->page, wbc, data));
488 }
489 #endif
490
491 static inline int
zpl_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,void * data)492 zpl_write_cache_pages(struct address_space *mapping,
493 struct writeback_control *wbc, void *data)
494 {
495 int result;
496
497 #ifdef HAVE_WRITEPAGE_T_FOLIO
498 result = write_cache_pages(mapping, wbc, zpl_putfolio, data);
499 #else
500 result = write_cache_pages(mapping, wbc, zpl_putpage, data);
501 #endif
502 return (result);
503 }
504 #else
505 static inline int
zpl_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,void * data)506 zpl_write_cache_pages(struct address_space *mapping,
507 struct writeback_control *wbc, void *data)
508 {
509 pgoff_t start = wbc->range_start >> PAGE_SHIFT;
510 pgoff_t end = wbc->range_end >> PAGE_SHIFT;
511
512 struct folio_batch fbatch;
513 folio_batch_init(&fbatch);
514
515 /*
516 * This atomically (-ish) tags all DIRTY pages in the range with
517 * TOWRITE, allowing users to continue dirtying or undirtying pages
518 * while we get on with writeback, without us treading on each other.
519 */
520 tag_pages_for_writeback(mapping, start, end);
521
522 int err = 0;
523 unsigned int npages;
524
525 /*
526 * Grab references to the TOWRITE pages just flagged. This may not get
527 * all of them, so we do it in a loop until there are none left.
528 */
529 while ((npages = filemap_get_folios_tag(mapping, &start, end,
530 PAGECACHE_TAG_TOWRITE, &fbatch)) != 0) {
531
532 /* Loop over each page and write it out. */
533 struct folio *folio;
534 while ((folio = folio_batch_next(&fbatch)) != NULL) {
535 folio_lock(folio);
536
537 /*
538 * If the folio has been remapped, or is no longer
539 * dirty, then there's nothing to do.
540 */
541 if (folio->mapping != mapping ||
542 !folio_test_dirty(folio)) {
543 folio_unlock(folio);
544 continue;
545 }
546
547 /*
548 * If writeback is already in progress, wait for it to
549 * finish. We continue after this even if the page
550 * ends up clean; zfs_putpage() will skip it if no
551 * further work is required.
552 */
553 while (folio_test_writeback(folio))
554 folio_wait_bit(folio, PG_writeback);
555
556 /*
557 * Write it out and collect any error. zfs_putpage()
558 * will clear the TOWRITE and DIRTY flags, and return
559 * with the page unlocked.
560 */
561 int ferr = zpl_putpage(&folio->page, wbc, data);
562 if (err == 0 && ferr != 0)
563 err = ferr;
564
565 /* Housekeeping for the caller. */
566 wbc->nr_to_write -= folio_nr_pages(folio);
567 }
568
569 /* Release any remaining references on the batch. */
570 folio_batch_release(&fbatch);
571 }
572
573 return (err);
574 }
575 #endif
576
577 static int
zpl_writepages(struct address_space * mapping,struct writeback_control * wbc)578 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
579 {
580 znode_t *zp = ITOZ(mapping->host);
581 zfsvfs_t *zfsvfs = ITOZSB(mapping->host);
582 enum writeback_sync_modes sync_mode;
583 int result;
584
585 if ((result = zpl_enter(zfsvfs, FTAG)) != 0)
586 return (result);
587 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
588 wbc->sync_mode = WB_SYNC_ALL;
589 zpl_exit(zfsvfs, FTAG);
590 sync_mode = wbc->sync_mode;
591
592 /*
593 * We don't want to run write_cache_pages() in SYNC mode here, because
594 * that would make putpage() wait for a single page to be committed to
595 * disk every single time, resulting in atrocious performance. Instead
596 * we run it once in non-SYNC mode so that the ZIL gets all the data,
597 * and then we commit it all in one go.
598 */
599 boolean_t for_sync = (sync_mode == WB_SYNC_ALL);
600 wbc->sync_mode = WB_SYNC_NONE;
601 result = zpl_write_cache_pages(mapping, wbc, &for_sync);
602 if (sync_mode != wbc->sync_mode) {
603 if ((result = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
604 return (result);
605
606 if (zfsvfs->z_log != NULL) {
607 /*
608 * We don't want to block here if the pool suspends,
609 * because this is not a syncing op by itself, but
610 * might be part of one that the caller will
611 * coordinate.
612 */
613 result = -zil_commit_flags(zfsvfs->z_log, zp->z_id,
614 ZIL_COMMIT_NOW);
615 }
616
617 zpl_exit(zfsvfs, FTAG);
618
619 /*
620 * If zil_commit_flags() failed, it's unclear what state things
621 * are currently in. putpage() has written back out what it can
622 * to the DMU, but it may not be on disk. We have little choice
623 * but to escape.
624 */
625 if (result != 0)
626 return (result);
627
628 /*
629 * We need to call write_cache_pages() again (we can't just
630 * return after the commit) because the previous call in
631 * non-SYNC mode does not guarantee that we got all the dirty
632 * pages (see the implementation of write_cache_pages() for
633 * details). That being said, this is a no-op in most cases.
634 */
635 wbc->sync_mode = sync_mode;
636 result = zpl_write_cache_pages(mapping, wbc, &for_sync);
637 }
638 return (result);
639 }
640
641 #ifdef HAVE_VFS_WRITEPAGE
642 /*
643 * Write out dirty pages to the ARC, this function is only required to
644 * support mmap(2). Mapped pages may be dirtied by memory operations
645 * which never call .write(). These dirty pages are kept in sync with
646 * the ARC buffers via this hook.
647 */
648 static int
zpl_writepage(struct page * pp,struct writeback_control * wbc)649 zpl_writepage(struct page *pp, struct writeback_control *wbc)
650 {
651 if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
652 wbc->sync_mode = WB_SYNC_ALL;
653
654 boolean_t for_sync = (wbc->sync_mode == WB_SYNC_ALL);
655
656 return (zpl_putpage(pp, wbc, &for_sync));
657 }
658 #endif
659
660 /*
661 * The flag combination which matches the behavior of zfs_space() is
662 * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE. The FALLOC_FL_PUNCH_HOLE
663 * flag was introduced in the 2.6.38 kernel.
664 *
665 * The original mode=0 (allocate space) behavior can be reasonably emulated
666 * by checking if enough space exists and creating a sparse file, as real
667 * persistent space reservation is not possible due to COW, snapshots, etc.
668 */
669 static long
zpl_fallocate_common(struct inode * ip,int mode,loff_t offset,loff_t len)670 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
671 {
672 cred_t *cr = CRED();
673 loff_t olen;
674 fstrans_cookie_t cookie;
675 int error = 0;
676
677 int test_mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE;
678
679 if ((mode & ~(FALLOC_FL_KEEP_SIZE | test_mode)) != 0)
680 return (-EOPNOTSUPP);
681
682 if (offset < 0 || len <= 0)
683 return (-EINVAL);
684
685 spl_inode_lock(ip);
686 olen = i_size_read(ip);
687
688 crhold(cr);
689 cookie = spl_fstrans_mark();
690 if (mode & (test_mode)) {
691 flock64_t bf;
692
693 if (mode & FALLOC_FL_KEEP_SIZE) {
694 if (offset > olen)
695 goto out_unmark;
696
697 if (offset + len > olen)
698 len = olen - offset;
699 }
700 bf.l_type = F_WRLCK;
701 bf.l_whence = SEEK_SET;
702 bf.l_start = offset;
703 bf.l_len = len;
704 bf.l_pid = 0;
705
706 error = -zfs_space(ITOZ(ip), F_FREESP, &bf, O_RDWR, offset, cr);
707 } else if ((mode & ~FALLOC_FL_KEEP_SIZE) == 0) {
708 unsigned int percent = zfs_fallocate_reserve_percent;
709 struct kstatfs statfs;
710
711 /* Legacy mode, disable fallocate compatibility. */
712 if (percent == 0) {
713 error = -EOPNOTSUPP;
714 goto out_unmark;
715 }
716
717 /*
718 * Use zfs_statvfs() instead of dmu_objset_space() since it
719 * also checks project quota limits, which are relevant here.
720 */
721 error = zfs_statvfs(ip, &statfs);
722 if (error)
723 goto out_unmark;
724
725 /*
726 * Shrink available space a bit to account for overhead/races.
727 * We know the product previously fit into availbytes from
728 * dmu_objset_space(), so the smaller product will also fit.
729 */
730 if (len > statfs.f_bavail * (statfs.f_bsize * 100 / percent)) {
731 error = -ENOSPC;
732 goto out_unmark;
733 }
734 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > olen)
735 error = zfs_freesp(ITOZ(ip), offset + len, 0, 0, FALSE);
736 }
737 out_unmark:
738 spl_fstrans_unmark(cookie);
739 spl_inode_unlock(ip);
740
741 crfree(cr);
742
743 return (error);
744 }
745
746 static long
zpl_fallocate(struct file * filp,int mode,loff_t offset,loff_t len)747 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
748 {
749 return zpl_fallocate_common(file_inode(filp),
750 mode, offset, len);
751 }
752
753 static int
zpl_ioctl_getversion(struct file * filp,void __user * arg)754 zpl_ioctl_getversion(struct file *filp, void __user *arg)
755 {
756 uint32_t generation = file_inode(filp)->i_generation;
757
758 return (copy_to_user(arg, &generation, sizeof (generation)));
759 }
760
761 static int
zpl_fadvise(struct file * filp,loff_t offset,loff_t len,int advice)762 zpl_fadvise(struct file *filp, loff_t offset, loff_t len, int advice)
763 {
764 struct inode *ip = file_inode(filp);
765 znode_t *zp = ITOZ(ip);
766 zfsvfs_t *zfsvfs = ITOZSB(ip);
767 objset_t *os = zfsvfs->z_os;
768 int error = 0;
769
770 if (S_ISFIFO(ip->i_mode))
771 return (-ESPIPE);
772
773 if (offset < 0 || len < 0)
774 return (-EINVAL);
775
776 if ((error = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
777 return (error);
778
779 switch (advice) {
780 case POSIX_FADV_SEQUENTIAL:
781 case POSIX_FADV_WILLNEED:
782 #ifdef HAVE_GENERIC_FADVISE
783 if (zn_has_cached_data(zp, offset, offset + len - 1))
784 error = generic_fadvise(filp, offset, len, advice);
785 #endif
786 /*
787 * Pass on the caller's size directly, but note that
788 * dmu_prefetch_max will effectively cap it. If there
789 * really is a larger sequential access pattern, perhaps
790 * dmu_zfetch will detect it.
791 */
792 if (len == 0)
793 len = i_size_read(ip) - offset;
794
795 dmu_prefetch(os, zp->z_id, 0, offset, len,
796 ZIO_PRIORITY_ASYNC_READ);
797 break;
798 case POSIX_FADV_NORMAL:
799 case POSIX_FADV_RANDOM:
800 case POSIX_FADV_DONTNEED:
801 case POSIX_FADV_NOREUSE:
802 /* ignored for now */
803 break;
804 default:
805 error = -EINVAL;
806 break;
807 }
808
809 zfs_exit(zfsvfs, FTAG);
810
811 return (error);
812 }
813
814 #define ZFS_FL_USER_VISIBLE (FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
815 #define ZFS_FL_USER_MODIFIABLE (FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
816
817 static uint32_t
__zpl_ioctl_getflags(struct inode * ip)818 __zpl_ioctl_getflags(struct inode *ip)
819 {
820 uint64_t zfs_flags = ITOZ(ip)->z_pflags;
821 uint32_t ioctl_flags = 0;
822
823 if (zfs_flags & ZFS_IMMUTABLE)
824 ioctl_flags |= FS_IMMUTABLE_FL;
825
826 if (zfs_flags & ZFS_APPENDONLY)
827 ioctl_flags |= FS_APPEND_FL;
828
829 if (zfs_flags & ZFS_NODUMP)
830 ioctl_flags |= FS_NODUMP_FL;
831
832 if (zfs_flags & ZFS_PROJINHERIT)
833 ioctl_flags |= ZFS_PROJINHERIT_FL;
834
835 return (ioctl_flags & ZFS_FL_USER_VISIBLE);
836 }
837
838 /*
839 * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
840 * attributes common to both Linux and Solaris are mapped.
841 */
842 static int
zpl_ioctl_getflags(struct file * filp,void __user * arg)843 zpl_ioctl_getflags(struct file *filp, void __user *arg)
844 {
845 uint32_t flags;
846 int err;
847
848 flags = __zpl_ioctl_getflags(file_inode(filp));
849 err = copy_to_user(arg, &flags, sizeof (flags));
850
851 return (err);
852 }
853
854 /*
855 * fchange() is a helper macro to detect if we have been asked to change a
856 * flag. This is ugly, but the requirement that we do this is a consequence of
857 * how the Linux file attribute interface was designed. Another consequence is
858 * that concurrent modification of files suffers from a TOCTOU race. Neither
859 * are things we can fix without modifying the kernel-userland interface, which
860 * is outside of our jurisdiction.
861 */
862
863 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
864
865 static int
__zpl_ioctl_setflags(struct inode * ip,uint32_t ioctl_flags,xvattr_t * xva)866 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva)
867 {
868 uint64_t zfs_flags = ITOZ(ip)->z_pflags;
869 xoptattr_t *xoap;
870
871 if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL |
872 ZFS_PROJINHERIT_FL))
873 return (-EOPNOTSUPP);
874
875 if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE)
876 return (-EACCES);
877
878 if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
879 fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
880 !capable(CAP_LINUX_IMMUTABLE))
881 return (-EPERM);
882
883 if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
884 return (-EACCES);
885
886 xva_init(xva);
887 xoap = xva_getxoptattr(xva);
888
889 #define FLAG_CHANGE(iflag, zflag, xflag, xfield) do { \
890 if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) || \
891 ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) { \
892 XVA_SET_REQ(xva, (xflag)); \
893 (xfield) = ((ioctl_flags & (iflag)) != 0); \
894 } \
895 } while (0)
896
897 FLAG_CHANGE(FS_IMMUTABLE_FL, ZFS_IMMUTABLE, XAT_IMMUTABLE,
898 xoap->xoa_immutable);
899 FLAG_CHANGE(FS_APPEND_FL, ZFS_APPENDONLY, XAT_APPENDONLY,
900 xoap->xoa_appendonly);
901 FLAG_CHANGE(FS_NODUMP_FL, ZFS_NODUMP, XAT_NODUMP,
902 xoap->xoa_nodump);
903 FLAG_CHANGE(ZFS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT,
904 xoap->xoa_projinherit);
905
906 #undef FLAG_CHANGE
907
908 return (0);
909 }
910
911 static int
zpl_ioctl_setflags(struct file * filp,void __user * arg)912 zpl_ioctl_setflags(struct file *filp, void __user *arg)
913 {
914 struct inode *ip = file_inode(filp);
915 uint32_t flags;
916 cred_t *cr = CRED();
917 xvattr_t xva;
918 int err;
919 fstrans_cookie_t cookie;
920
921 if (copy_from_user(&flags, arg, sizeof (flags)))
922 return (-EFAULT);
923
924 err = __zpl_ioctl_setflags(ip, flags, &xva);
925 if (err)
926 return (err);
927
928 crhold(cr);
929 cookie = spl_fstrans_mark();
930 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
931 spl_fstrans_unmark(cookie);
932 crfree(cr);
933
934 return (err);
935 }
936
937 static int
zpl_ioctl_getxattr(struct file * filp,void __user * arg)938 zpl_ioctl_getxattr(struct file *filp, void __user *arg)
939 {
940 zfsxattr_t fsx = { 0 };
941 struct inode *ip = file_inode(filp);
942 int err;
943
944 fsx.fsx_xflags = __zpl_ioctl_getflags(ip);
945 fsx.fsx_projid = ITOZ(ip)->z_projid;
946 err = copy_to_user(arg, &fsx, sizeof (fsx));
947
948 return (err);
949 }
950
951 static int
zpl_ioctl_setxattr(struct file * filp,void __user * arg)952 zpl_ioctl_setxattr(struct file *filp, void __user *arg)
953 {
954 struct inode *ip = file_inode(filp);
955 zfsxattr_t fsx;
956 cred_t *cr = CRED();
957 xvattr_t xva;
958 xoptattr_t *xoap;
959 int err;
960 fstrans_cookie_t cookie;
961
962 if (copy_from_user(&fsx, arg, sizeof (fsx)))
963 return (-EFAULT);
964
965 if (!zpl_is_valid_projid(fsx.fsx_projid))
966 return (-EINVAL);
967
968 err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva);
969 if (err)
970 return (err);
971
972 xoap = xva_getxoptattr(&xva);
973 XVA_SET_REQ(&xva, XAT_PROJID);
974 xoap->xoa_projid = fsx.fsx_projid;
975
976 crhold(cr);
977 cookie = spl_fstrans_mark();
978 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
979 spl_fstrans_unmark(cookie);
980 crfree(cr);
981
982 return (err);
983 }
984
985 /*
986 * Expose Additional File Level Attributes of ZFS.
987 */
988 static int
zpl_ioctl_getdosflags(struct file * filp,void __user * arg)989 zpl_ioctl_getdosflags(struct file *filp, void __user *arg)
990 {
991 struct inode *ip = file_inode(filp);
992 uint64_t dosflags = ITOZ(ip)->z_pflags;
993 dosflags &= ZFS_DOS_FL_USER_VISIBLE;
994 int err = copy_to_user(arg, &dosflags, sizeof (dosflags));
995
996 return (err);
997 }
998
999 static int
__zpl_ioctl_setdosflags(struct inode * ip,uint64_t ioctl_flags,xvattr_t * xva)1000 __zpl_ioctl_setdosflags(struct inode *ip, uint64_t ioctl_flags, xvattr_t *xva)
1001 {
1002 uint64_t zfs_flags = ITOZ(ip)->z_pflags;
1003 xoptattr_t *xoap;
1004
1005 if (ioctl_flags & (~ZFS_DOS_FL_USER_VISIBLE))
1006 return (-EOPNOTSUPP);
1007
1008 if ((fchange(ioctl_flags, zfs_flags, ZFS_IMMUTABLE, ZFS_IMMUTABLE) ||
1009 fchange(ioctl_flags, zfs_flags, ZFS_APPENDONLY, ZFS_APPENDONLY)) &&
1010 !capable(CAP_LINUX_IMMUTABLE))
1011 return (-EPERM);
1012
1013 if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
1014 return (-EACCES);
1015
1016 xva_init(xva);
1017 xoap = xva_getxoptattr(xva);
1018
1019 #define FLAG_CHANGE(iflag, xflag, xfield) do { \
1020 if (((ioctl_flags & (iflag)) && !(zfs_flags & (iflag))) || \
1021 ((zfs_flags & (iflag)) && !(ioctl_flags & (iflag)))) { \
1022 XVA_SET_REQ(xva, (xflag)); \
1023 (xfield) = ((ioctl_flags & (iflag)) != 0); \
1024 } \
1025 } while (0)
1026
1027 FLAG_CHANGE(ZFS_IMMUTABLE, XAT_IMMUTABLE, xoap->xoa_immutable);
1028 FLAG_CHANGE(ZFS_APPENDONLY, XAT_APPENDONLY, xoap->xoa_appendonly);
1029 FLAG_CHANGE(ZFS_NODUMP, XAT_NODUMP, xoap->xoa_nodump);
1030 FLAG_CHANGE(ZFS_READONLY, XAT_READONLY, xoap->xoa_readonly);
1031 FLAG_CHANGE(ZFS_HIDDEN, XAT_HIDDEN, xoap->xoa_hidden);
1032 FLAG_CHANGE(ZFS_SYSTEM, XAT_SYSTEM, xoap->xoa_system);
1033 FLAG_CHANGE(ZFS_ARCHIVE, XAT_ARCHIVE, xoap->xoa_archive);
1034 FLAG_CHANGE(ZFS_NOUNLINK, XAT_NOUNLINK, xoap->xoa_nounlink);
1035 FLAG_CHANGE(ZFS_REPARSE, XAT_REPARSE, xoap->xoa_reparse);
1036 FLAG_CHANGE(ZFS_OFFLINE, XAT_OFFLINE, xoap->xoa_offline);
1037 FLAG_CHANGE(ZFS_SPARSE, XAT_SPARSE, xoap->xoa_sparse);
1038
1039 #undef FLAG_CHANGE
1040
1041 return (0);
1042 }
1043
1044 /*
1045 * Set Additional File Level Attributes of ZFS.
1046 */
1047 static int
zpl_ioctl_setdosflags(struct file * filp,void __user * arg)1048 zpl_ioctl_setdosflags(struct file *filp, void __user *arg)
1049 {
1050 struct inode *ip = file_inode(filp);
1051 uint64_t dosflags;
1052 cred_t *cr = CRED();
1053 xvattr_t xva;
1054 int err;
1055 fstrans_cookie_t cookie;
1056
1057 if (copy_from_user(&dosflags, arg, sizeof (dosflags)))
1058 return (-EFAULT);
1059
1060 err = __zpl_ioctl_setdosflags(ip, dosflags, &xva);
1061 if (err)
1062 return (err);
1063
1064 crhold(cr);
1065 cookie = spl_fstrans_mark();
1066 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
1067 spl_fstrans_unmark(cookie);
1068 crfree(cr);
1069
1070 return (err);
1071 }
1072
1073 static int
zpl_ioctl_rewrite(struct file * filp,void __user * arg)1074 zpl_ioctl_rewrite(struct file *filp, void __user *arg)
1075 {
1076 struct inode *ip = file_inode(filp);
1077 zfs_rewrite_args_t args;
1078 fstrans_cookie_t cookie;
1079 int err;
1080
1081 if (copy_from_user(&args, arg, sizeof (args)))
1082 return (-EFAULT);
1083
1084 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
1085 return (-EBADF);
1086
1087 cookie = spl_fstrans_mark();
1088 err = -zfs_rewrite(ITOZ(ip), args.off, args.len, args.flags, args.arg);
1089 spl_fstrans_unmark(cookie);
1090
1091 return (err);
1092 }
1093
1094 static long
zpl_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1095 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1096 {
1097 switch (cmd) {
1098 case FS_IOC_GETVERSION:
1099 return (zpl_ioctl_getversion(filp, (void *)arg));
1100 case FS_IOC_GETFLAGS:
1101 return (zpl_ioctl_getflags(filp, (void *)arg));
1102 case FS_IOC_SETFLAGS:
1103 return (zpl_ioctl_setflags(filp, (void *)arg));
1104 case ZFS_IOC_FSGETXATTR:
1105 return (zpl_ioctl_getxattr(filp, (void *)arg));
1106 case ZFS_IOC_FSSETXATTR:
1107 return (zpl_ioctl_setxattr(filp, (void *)arg));
1108 case ZFS_IOC_GETDOSFLAGS:
1109 return (zpl_ioctl_getdosflags(filp, (void *)arg));
1110 case ZFS_IOC_SETDOSFLAGS:
1111 return (zpl_ioctl_setdosflags(filp, (void *)arg));
1112 case ZFS_IOC_REWRITE:
1113 return (zpl_ioctl_rewrite(filp, (void *)arg));
1114 default:
1115 return (-ENOTTY);
1116 }
1117 }
1118
1119 #ifdef CONFIG_COMPAT
1120 static long
zpl_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1121 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1122 {
1123 switch (cmd) {
1124 case FS_IOC32_GETVERSION:
1125 cmd = FS_IOC_GETVERSION;
1126 break;
1127 case FS_IOC32_GETFLAGS:
1128 cmd = FS_IOC_GETFLAGS;
1129 break;
1130 case FS_IOC32_SETFLAGS:
1131 cmd = FS_IOC_SETFLAGS;
1132 break;
1133 default:
1134 return (-ENOTTY);
1135 }
1136 return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
1137 }
1138 #endif /* CONFIG_COMPAT */
1139
1140 const struct address_space_operations zpl_address_space_operations = {
1141 #ifdef HAVE_VFS_READPAGES
1142 .readpages = zpl_readpages,
1143 #else
1144 .readahead = zpl_readahead,
1145 #endif
1146 #ifdef HAVE_VFS_READ_FOLIO
1147 .read_folio = zpl_read_folio,
1148 #else
1149 .readpage = zpl_readpage,
1150 #endif
1151 #ifdef HAVE_VFS_WRITEPAGE
1152 .writepage = zpl_writepage,
1153 #endif
1154 .writepages = zpl_writepages,
1155 .direct_IO = zpl_direct_IO,
1156 #ifdef HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS
1157 .set_page_dirty = __set_page_dirty_nobuffers,
1158 #endif
1159 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
1160 .dirty_folio = filemap_dirty_folio,
1161 #endif
1162 #ifdef HAVE_VFS_MIGRATE_FOLIO
1163 .migrate_folio = migrate_folio,
1164 #elif defined(HAVE_VFS_MIGRATEPAGE)
1165 .migratepage = migrate_page,
1166 #endif
1167 };
1168
1169 const struct file_operations zpl_file_operations = {
1170 .open = zpl_open,
1171 .release = zpl_release,
1172 .llseek = zpl_llseek,
1173 .read_iter = zpl_iter_read,
1174 .write_iter = zpl_iter_write,
1175 #ifdef HAVE_COPY_SPLICE_READ
1176 .splice_read = copy_splice_read,
1177 #else
1178 .splice_read = generic_file_splice_read,
1179 #endif
1180 .splice_write = iter_file_splice_write,
1181 .mmap = zpl_mmap,
1182 .fsync = zpl_fsync,
1183 .fallocate = zpl_fallocate,
1184 .copy_file_range = zpl_copy_file_range,
1185 #ifdef HAVE_VFS_CLONE_FILE_RANGE
1186 .clone_file_range = zpl_clone_file_range,
1187 #endif
1188 #ifdef HAVE_VFS_REMAP_FILE_RANGE
1189 .remap_file_range = zpl_remap_file_range,
1190 #endif
1191 #ifdef HAVE_VFS_DEDUPE_FILE_RANGE
1192 .dedupe_file_range = zpl_dedupe_file_range,
1193 #endif
1194 .fadvise = zpl_fadvise,
1195 .unlocked_ioctl = zpl_ioctl,
1196 #ifdef CONFIG_COMPAT
1197 .compat_ioctl = zpl_compat_ioctl,
1198 #endif
1199 };
1200
1201 const struct file_operations zpl_dir_file_operations = {
1202 .llseek = generic_file_llseek,
1203 .read = generic_read_dir,
1204 .iterate_shared = zpl_iterate,
1205 .fsync = zpl_fsync,
1206 .unlocked_ioctl = zpl_ioctl,
1207 #ifdef CONFIG_COMPAT
1208 .compat_ioctl = zpl_compat_ioctl,
1209 #endif
1210 };
1211
1212 module_param(zfs_fallocate_reserve_percent, uint, 0644);
1213 MODULE_PARM_DESC(zfs_fallocate_reserve_percent,
1214 "Percentage of length to use for the available capacity check");
1215