xref: /freebsd/sys/contrib/openzfs/module/os/linux/zfs/zpl_file.c (revision d0abb9a6399accc9053e2808052be00a6754ecef)
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  */
27 
28 
29 #ifdef CONFIG_COMPAT
30 #include <linux/compat.h>
31 #endif
32 #include <linux/fs.h>
33 #include <linux/migrate.h>
34 #include <sys/file.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/zfs_znode.h>
37 #include <sys/zfs_vfsops.h>
38 #include <sys/zfs_vnops.h>
39 #include <sys/zfs_project.h>
40 #include <linux/pagemap_compat.h>
41 #include <linux/fadvise.h>
42 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
43 #include <linux/writeback.h>
44 #endif
45 
46 /*
47  * When using fallocate(2) to preallocate space, inflate the requested
48  * capacity check by 10% to account for the required metadata blocks.
49  */
50 static unsigned int zfs_fallocate_reserve_percent = 110;
51 
52 static int
zpl_open(struct inode * ip,struct file * filp)53 zpl_open(struct inode *ip, struct file *filp)
54 {
55 	cred_t *cr = CRED();
56 	int error;
57 	fstrans_cookie_t cookie;
58 
59 	error = generic_file_open(ip, filp);
60 	if (error)
61 		return (error);
62 
63 	crhold(cr);
64 	cookie = spl_fstrans_mark();
65 	error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
66 	spl_fstrans_unmark(cookie);
67 	crfree(cr);
68 	ASSERT3S(error, <=, 0);
69 
70 	return (error);
71 }
72 
73 static int
zpl_release(struct inode * ip,struct file * filp)74 zpl_release(struct inode *ip, struct file *filp)
75 {
76 	cred_t *cr = CRED();
77 	int error;
78 	fstrans_cookie_t cookie;
79 
80 	cookie = spl_fstrans_mark();
81 	if (ITOZ(ip)->z_atime_dirty)
82 		zfs_mark_inode_dirty(ip);
83 
84 	crhold(cr);
85 	error = -zfs_close(ip, filp->f_flags, cr);
86 	spl_fstrans_unmark(cookie);
87 	crfree(cr);
88 	ASSERT3S(error, <=, 0);
89 
90 	return (error);
91 }
92 
93 static int
zpl_iterate(struct file * filp,struct dir_context * ctx)94 zpl_iterate(struct file *filp, struct dir_context *ctx)
95 {
96 	cred_t *cr = CRED();
97 	int error;
98 	fstrans_cookie_t cookie;
99 
100 	crhold(cr);
101 	cookie = spl_fstrans_mark();
102 	error = -zfs_readdir(file_inode(filp), ctx, cr);
103 	spl_fstrans_unmark(cookie);
104 	crfree(cr);
105 	ASSERT3S(error, <=, 0);
106 
107 	return (error);
108 }
109 
110 static inline int
111 zpl_write_cache_pages(struct address_space *mapping,
112     struct writeback_control *wbc, void *data);
113 
114 static int
zpl_fsync(struct file * filp,loff_t start,loff_t end,int datasync)115 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
116 {
117 	struct inode *inode = filp->f_mapping->host;
118 	znode_t *zp = ITOZ(inode);
119 	cred_t *cr = CRED();
120 	int error;
121 	fstrans_cookie_t cookie;
122 
123 	/*
124 	 * Force dirty pages in the range out to the DMU and the log, ready
125 	 * for zil_commit() to write down.
126 	 *
127 	 * We call write_cache_pages() directly to ensure that zpl_putpage() is
128 	 * called with the flags we need. We need WB_SYNC_NONE to avoid a call
129 	 * to zil_commit() (since we're doing this as a kind of pre-sync); but
130 	 * we do need for_sync so that the pages remain in writeback until
131 	 * they're on disk, and so that we get an error if the DMU write fails.
132 	 */
133 	if (filemap_range_has_page(inode->i_mapping, start, end)) {
134 		int for_sync = 1;
135 		struct writeback_control wbc = {
136 			.sync_mode = WB_SYNC_NONE,
137 			.nr_to_write = LONG_MAX,
138 			.range_start = start,
139 			.range_end = end,
140 		};
141 		error =
142 		    zpl_write_cache_pages(inode->i_mapping, &wbc, &for_sync);
143 		if (error != 0) {
144 			/*
145 			 * Unclear what state things are in. zfs_putpage() will
146 			 * ensure the pages remain dirty if they haven't been
147 			 * written down to the DMU, but because there may be
148 			 * nothing logged, we can't assume that zfs_sync() ->
149 			 * zil_commit() will give us a useful error. It's
150 			 * safest if we just error out here.
151 			 */
152 			return (error);
153 		}
154 	}
155 
156 	crhold(cr);
157 	cookie = spl_fstrans_mark();
158 	error = -zfs_fsync(zp, datasync, cr);
159 	spl_fstrans_unmark(cookie);
160 	crfree(cr);
161 	ASSERT3S(error, <=, 0);
162 
163 	return (error);
164 }
165 
166 static inline int
zfs_io_flags(struct kiocb * kiocb)167 zfs_io_flags(struct kiocb *kiocb)
168 {
169 	int flags = 0;
170 
171 #if defined(IOCB_DSYNC)
172 	if (kiocb->ki_flags & IOCB_DSYNC)
173 		flags |= O_DSYNC;
174 #endif
175 #if defined(IOCB_SYNC)
176 	if (kiocb->ki_flags & IOCB_SYNC)
177 		flags |= O_SYNC;
178 #endif
179 #if defined(IOCB_APPEND)
180 	if (kiocb->ki_flags & IOCB_APPEND)
181 		flags |= O_APPEND;
182 #endif
183 #if defined(IOCB_DIRECT)
184 	if (kiocb->ki_flags & IOCB_DIRECT)
185 		flags |= O_DIRECT;
186 #endif
187 	return (flags);
188 }
189 
190 /*
191  * If relatime is enabled, call file_accessed() if zfs_relatime_need_update()
192  * is true.  This is needed since datasets with inherited "relatime" property
193  * aren't necessarily mounted with the MNT_RELATIME flag (e.g. after
194  * `zfs set relatime=...`), which is what relatime test in VFS by
195  * relatime_need_update() is based on.
196  */
197 static inline void
zpl_file_accessed(struct file * filp)198 zpl_file_accessed(struct file *filp)
199 {
200 	struct inode *ip = filp->f_mapping->host;
201 
202 	if (!IS_NOATIME(ip) && ITOZSB(ip)->z_relatime) {
203 		if (zfs_relatime_need_update(ip))
204 			file_accessed(filp);
205 	} else {
206 		file_accessed(filp);
207 	}
208 }
209 
210 static ssize_t
zpl_iter_read(struct kiocb * kiocb,struct iov_iter * to)211 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
212 {
213 	cred_t *cr = CRED();
214 	fstrans_cookie_t cookie;
215 	struct file *filp = kiocb->ki_filp;
216 	ssize_t count = iov_iter_count(to);
217 	zfs_uio_t uio;
218 
219 	zfs_uio_iov_iter_init(&uio, to, kiocb->ki_pos, count);
220 
221 	crhold(cr);
222 	cookie = spl_fstrans_mark();
223 
224 	ssize_t ret = -zfs_read(ITOZ(filp->f_mapping->host), &uio,
225 	    filp->f_flags | zfs_io_flags(kiocb), cr);
226 
227 	spl_fstrans_unmark(cookie);
228 	crfree(cr);
229 
230 	if (ret < 0)
231 		return (ret);
232 
233 	ssize_t read = count - uio.uio_resid;
234 	kiocb->ki_pos += read;
235 
236 	zpl_file_accessed(filp);
237 
238 	return (read);
239 }
240 
241 static inline ssize_t
zpl_generic_write_checks(struct kiocb * kiocb,struct iov_iter * from,size_t * countp)242 zpl_generic_write_checks(struct kiocb *kiocb, struct iov_iter *from,
243     size_t *countp)
244 {
245 	ssize_t ret = generic_write_checks(kiocb, from);
246 	if (ret <= 0)
247 		return (ret);
248 
249 	*countp = ret;
250 
251 	return (0);
252 }
253 
254 static ssize_t
zpl_iter_write(struct kiocb * kiocb,struct iov_iter * from)255 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
256 {
257 	cred_t *cr = CRED();
258 	fstrans_cookie_t cookie;
259 	struct file *filp = kiocb->ki_filp;
260 	struct inode *ip = filp->f_mapping->host;
261 	zfs_uio_t uio;
262 	size_t count = 0;
263 	ssize_t ret;
264 
265 	ret = zpl_generic_write_checks(kiocb, from, &count);
266 	if (ret)
267 		return (ret);
268 
269 	zfs_uio_iov_iter_init(&uio, from, kiocb->ki_pos, count);
270 
271 	crhold(cr);
272 	cookie = spl_fstrans_mark();
273 
274 	ret = -zfs_write(ITOZ(ip), &uio,
275 	    filp->f_flags | zfs_io_flags(kiocb), cr);
276 
277 	spl_fstrans_unmark(cookie);
278 	crfree(cr);
279 
280 	if (ret < 0)
281 		return (ret);
282 
283 	ssize_t wrote = count - uio.uio_resid;
284 	kiocb->ki_pos += wrote;
285 
286 	return (wrote);
287 }
288 
289 static ssize_t
zpl_direct_IO(struct kiocb * kiocb,struct iov_iter * iter)290 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter)
291 {
292 	/*
293 	 * All O_DIRECT requests should be handled by
294 	 * zpl_iter_write/read}(). There is no way kernel generic code should
295 	 * call the direct_IO address_space_operations function. We set this
296 	 * code path to be fatal if it is executed.
297 	 */
298 	PANIC(0);
299 	return (0);
300 }
301 
302 static loff_t
zpl_llseek(struct file * filp,loff_t offset,int whence)303 zpl_llseek(struct file *filp, loff_t offset, int whence)
304 {
305 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
306 	fstrans_cookie_t cookie;
307 
308 	if (whence == SEEK_DATA || whence == SEEK_HOLE) {
309 		struct inode *ip = filp->f_mapping->host;
310 		loff_t maxbytes = ip->i_sb->s_maxbytes;
311 		loff_t error;
312 
313 		spl_inode_lock_shared(ip);
314 		cookie = spl_fstrans_mark();
315 		error = -zfs_holey(ITOZ(ip), whence, &offset);
316 		spl_fstrans_unmark(cookie);
317 		if (error == 0)
318 			error = lseek_execute(filp, ip, offset, maxbytes);
319 		spl_inode_unlock_shared(ip);
320 
321 		return (error);
322 	}
323 #endif /* SEEK_HOLE && SEEK_DATA */
324 
325 	return (generic_file_llseek(filp, offset, whence));
326 }
327 
328 /*
329  * It's worth taking a moment to describe how mmap is implemented
330  * for zfs because it differs considerably from other Linux filesystems.
331  * However, this issue is handled the same way under OpenSolaris.
332  *
333  * The issue is that by design zfs bypasses the Linux page cache and
334  * leaves all caching up to the ARC.  This has been shown to work
335  * well for the common read(2)/write(2) case.  However, mmap(2)
336  * is problem because it relies on being tightly integrated with the
337  * page cache.  To handle this we cache mmap'ed files twice, once in
338  * the ARC and a second time in the page cache.  The code is careful
339  * to keep both copies synchronized.
340  *
341  * When a file with an mmap'ed region is written to using write(2)
342  * both the data in the ARC and existing pages in the page cache
343  * are updated.  For a read(2) data will be read first from the page
344  * cache then the ARC if needed.  Neither a write(2) or read(2) will
345  * will ever result in new pages being added to the page cache.
346  *
347  * New pages are added to the page cache only via .readpage() which
348  * is called when the vfs needs to read a page off disk to back the
349  * virtual memory region.  These pages may be modified without
350  * notifying the ARC and will be written out periodically via
351  * .writepage().  This will occur due to either a sync or the usual
352  * page aging behavior.  Note because a read(2) of a mmap'ed file
353  * will always check the page cache first even when the ARC is out
354  * of date correct data will still be returned.
355  *
356  * While this implementation ensures correct behavior it does have
357  * have some drawbacks.  The most obvious of which is that it
358  * increases the required memory footprint when access mmap'ed
359  * files.  It also adds additional complexity to the code keeping
360  * both caches synchronized.
361  *
362  * Longer term it may be possible to cleanly resolve this wart by
363  * mapping page cache pages directly on to the ARC buffers.  The
364  * Linux address space operations are flexible enough to allow
365  * selection of which pages back a particular index.  The trick
366  * would be working out the details of which subsystem is in
367  * charge, the ARC, the page cache, or both.  It may also prove
368  * helpful to move the ARC buffers to a scatter-gather lists
369  * rather than a vmalloc'ed region.
370  */
371 static int
zpl_mmap(struct file * filp,struct vm_area_struct * vma)372 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
373 {
374 	struct inode *ip = filp->f_mapping->host;
375 	int error;
376 	fstrans_cookie_t cookie;
377 
378 	cookie = spl_fstrans_mark();
379 	error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
380 	    (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
381 	spl_fstrans_unmark(cookie);
382 
383 	if (error)
384 		return (error);
385 
386 	error = generic_file_mmap(filp, vma);
387 	if (error)
388 		return (error);
389 
390 	return (error);
391 }
392 
393 /*
394  * Populate a page with data for the Linux page cache.  This function is
395  * only used to support mmap(2).  There will be an identical copy of the
396  * data in the ARC which is kept up to date via .write() and .writepage().
397  */
398 static inline int
zpl_readpage_common(struct page * pp)399 zpl_readpage_common(struct page *pp)
400 {
401 	fstrans_cookie_t cookie;
402 
403 	ASSERT(PageLocked(pp));
404 
405 	cookie = spl_fstrans_mark();
406 	int error = -zfs_getpage(pp->mapping->host, pp);
407 	spl_fstrans_unmark(cookie);
408 
409 	unlock_page(pp);
410 
411 	return (error);
412 }
413 
414 #ifdef HAVE_VFS_READ_FOLIO
415 static int
zpl_read_folio(struct file * filp,struct folio * folio)416 zpl_read_folio(struct file *filp, struct folio *folio)
417 {
418 	return (zpl_readpage_common(&folio->page));
419 }
420 #else
421 static int
zpl_readpage(struct file * filp,struct page * pp)422 zpl_readpage(struct file *filp, struct page *pp)
423 {
424 	return (zpl_readpage_common(pp));
425 }
426 #endif
427 
428 static int
zpl_readpage_filler(void * data,struct page * pp)429 zpl_readpage_filler(void *data, struct page *pp)
430 {
431 	return (zpl_readpage_common(pp));
432 }
433 
434 /*
435  * Populate a set of pages with data for the Linux page cache.  This
436  * function will only be called for read ahead and never for demand
437  * paging.  For simplicity, the code relies on read_cache_pages() to
438  * correctly lock each page for IO and call zpl_readpage().
439  */
440 #ifdef HAVE_VFS_READPAGES
441 static int
zpl_readpages(struct file * filp,struct address_space * mapping,struct list_head * pages,unsigned nr_pages)442 zpl_readpages(struct file *filp, struct address_space *mapping,
443     struct list_head *pages, unsigned nr_pages)
444 {
445 	return (read_cache_pages(mapping, pages, zpl_readpage_filler, NULL));
446 }
447 #else
448 static void
zpl_readahead(struct readahead_control * ractl)449 zpl_readahead(struct readahead_control *ractl)
450 {
451 	struct page *page;
452 
453 	while ((page = readahead_page(ractl)) != NULL) {
454 		int ret;
455 
456 		ret = zpl_readpage_filler(NULL, page);
457 		put_page(page);
458 		if (ret)
459 			break;
460 	}
461 }
462 #endif
463 
464 static int
zpl_putpage(struct page * pp,struct writeback_control * wbc,void * data)465 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
466 {
467 	boolean_t *for_sync = data;
468 	fstrans_cookie_t cookie;
469 	int ret;
470 
471 	ASSERT(PageLocked(pp));
472 	ASSERT(!PageWriteback(pp));
473 
474 	cookie = spl_fstrans_mark();
475 	ret = zfs_putpage(pp->mapping->host, pp, wbc, *for_sync);
476 	spl_fstrans_unmark(cookie);
477 
478 	return (ret);
479 }
480 
481 #ifdef HAVE_WRITEPAGE_T_FOLIO
482 static int
zpl_putfolio(struct folio * pp,struct writeback_control * wbc,void * data)483 zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data)
484 {
485 	return (zpl_putpage(&pp->page, wbc, data));
486 }
487 #endif
488 
489 static inline int
zpl_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,void * data)490 zpl_write_cache_pages(struct address_space *mapping,
491     struct writeback_control *wbc, void *data)
492 {
493 	int result;
494 
495 #ifdef HAVE_WRITEPAGE_T_FOLIO
496 	result = write_cache_pages(mapping, wbc, zpl_putfolio, data);
497 #else
498 	result = write_cache_pages(mapping, wbc, zpl_putpage, data);
499 #endif
500 	return (result);
501 }
502 
503 static int
zpl_writepages(struct address_space * mapping,struct writeback_control * wbc)504 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
505 {
506 	znode_t		*zp = ITOZ(mapping->host);
507 	zfsvfs_t	*zfsvfs = ITOZSB(mapping->host);
508 	enum writeback_sync_modes sync_mode;
509 	int result;
510 
511 	if ((result = zpl_enter(zfsvfs, FTAG)) != 0)
512 		return (result);
513 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
514 		wbc->sync_mode = WB_SYNC_ALL;
515 	zpl_exit(zfsvfs, FTAG);
516 	sync_mode = wbc->sync_mode;
517 
518 	/*
519 	 * We don't want to run write_cache_pages() in SYNC mode here, because
520 	 * that would make putpage() wait for a single page to be committed to
521 	 * disk every single time, resulting in atrocious performance. Instead
522 	 * we run it once in non-SYNC mode so that the ZIL gets all the data,
523 	 * and then we commit it all in one go.
524 	 */
525 	boolean_t for_sync = (sync_mode == WB_SYNC_ALL);
526 	wbc->sync_mode = WB_SYNC_NONE;
527 	result = zpl_write_cache_pages(mapping, wbc, &for_sync);
528 	if (sync_mode != wbc->sync_mode) {
529 		if ((result = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
530 			return (result);
531 
532 		if (zfsvfs->z_log != NULL) {
533 			/*
534 			 * We don't want to block here if the pool suspends,
535 			 * because this is not a syncing op by itself, but
536 			 * might be part of one that the caller will
537 			 * coordinate.
538 			 */
539 			result = -zil_commit_flags(zfsvfs->z_log, zp->z_id,
540 			    ZIL_COMMIT_NOW);
541 		}
542 
543 		zpl_exit(zfsvfs, FTAG);
544 
545 		/*
546 		 * If zil_commit_flags() failed, it's unclear what state things
547 		 * are currently in. putpage() has written back out what it can
548 		 * to the DMU, but it may not be on disk. We have little choice
549 		 * but to escape.
550 		 */
551 		if (result != 0)
552 			return (result);
553 
554 		/*
555 		 * We need to call write_cache_pages() again (we can't just
556 		 * return after the commit) because the previous call in
557 		 * non-SYNC mode does not guarantee that we got all the dirty
558 		 * pages (see the implementation of write_cache_pages() for
559 		 * details). That being said, this is a no-op in most cases.
560 		 */
561 		wbc->sync_mode = sync_mode;
562 		result = zpl_write_cache_pages(mapping, wbc, &for_sync);
563 	}
564 	return (result);
565 }
566 
567 #ifdef HAVE_VFS_WRITEPAGE
568 /*
569  * Write out dirty pages to the ARC, this function is only required to
570  * support mmap(2).  Mapped pages may be dirtied by memory operations
571  * which never call .write().  These dirty pages are kept in sync with
572  * the ARC buffers via this hook.
573  */
574 static int
zpl_writepage(struct page * pp,struct writeback_control * wbc)575 zpl_writepage(struct page *pp, struct writeback_control *wbc)
576 {
577 	if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
578 		wbc->sync_mode = WB_SYNC_ALL;
579 
580 	boolean_t for_sync = (wbc->sync_mode == WB_SYNC_ALL);
581 
582 	return (zpl_putpage(pp, wbc, &for_sync));
583 }
584 #endif
585 
586 /*
587  * The flag combination which matches the behavior of zfs_space() is
588  * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
589  * flag was introduced in the 2.6.38 kernel.
590  *
591  * The original mode=0 (allocate space) behavior can be reasonably emulated
592  * by checking if enough space exists and creating a sparse file, as real
593  * persistent space reservation is not possible due to COW, snapshots, etc.
594  */
595 static long
zpl_fallocate_common(struct inode * ip,int mode,loff_t offset,loff_t len)596 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
597 {
598 	cred_t *cr = CRED();
599 	loff_t olen;
600 	fstrans_cookie_t cookie;
601 	int error = 0;
602 
603 	int test_mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE;
604 
605 	if ((mode & ~(FALLOC_FL_KEEP_SIZE | test_mode)) != 0)
606 		return (-EOPNOTSUPP);
607 
608 	if (offset < 0 || len <= 0)
609 		return (-EINVAL);
610 
611 	spl_inode_lock(ip);
612 	olen = i_size_read(ip);
613 
614 	crhold(cr);
615 	cookie = spl_fstrans_mark();
616 	if (mode & (test_mode)) {
617 		flock64_t bf;
618 
619 		if (mode & FALLOC_FL_KEEP_SIZE) {
620 			if (offset > olen)
621 				goto out_unmark;
622 
623 			if (offset + len > olen)
624 				len = olen - offset;
625 		}
626 		bf.l_type = F_WRLCK;
627 		bf.l_whence = SEEK_SET;
628 		bf.l_start = offset;
629 		bf.l_len = len;
630 		bf.l_pid = 0;
631 
632 		error = -zfs_space(ITOZ(ip), F_FREESP, &bf, O_RDWR, offset, cr);
633 	} else if ((mode & ~FALLOC_FL_KEEP_SIZE) == 0) {
634 		unsigned int percent = zfs_fallocate_reserve_percent;
635 		struct kstatfs statfs;
636 
637 		/* Legacy mode, disable fallocate compatibility. */
638 		if (percent == 0) {
639 			error = -EOPNOTSUPP;
640 			goto out_unmark;
641 		}
642 
643 		/*
644 		 * Use zfs_statvfs() instead of dmu_objset_space() since it
645 		 * also checks project quota limits, which are relevant here.
646 		 */
647 		error = zfs_statvfs(ip, &statfs);
648 		if (error)
649 			goto out_unmark;
650 
651 		/*
652 		 * Shrink available space a bit to account for overhead/races.
653 		 * We know the product previously fit into availbytes from
654 		 * dmu_objset_space(), so the smaller product will also fit.
655 		 */
656 		if (len > statfs.f_bavail * (statfs.f_bsize * 100 / percent)) {
657 			error = -ENOSPC;
658 			goto out_unmark;
659 		}
660 		if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > olen)
661 			error = zfs_freesp(ITOZ(ip), offset + len, 0, 0, FALSE);
662 	}
663 out_unmark:
664 	spl_fstrans_unmark(cookie);
665 	spl_inode_unlock(ip);
666 
667 	crfree(cr);
668 
669 	return (error);
670 }
671 
672 static long
zpl_fallocate(struct file * filp,int mode,loff_t offset,loff_t len)673 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
674 {
675 	return zpl_fallocate_common(file_inode(filp),
676 	    mode, offset, len);
677 }
678 
679 static int
zpl_ioctl_getversion(struct file * filp,void __user * arg)680 zpl_ioctl_getversion(struct file *filp, void __user *arg)
681 {
682 	uint32_t generation = file_inode(filp)->i_generation;
683 
684 	return (copy_to_user(arg, &generation, sizeof (generation)));
685 }
686 
687 static int
zpl_fadvise(struct file * filp,loff_t offset,loff_t len,int advice)688 zpl_fadvise(struct file *filp, loff_t offset, loff_t len, int advice)
689 {
690 	struct inode *ip = file_inode(filp);
691 	znode_t *zp = ITOZ(ip);
692 	zfsvfs_t *zfsvfs = ITOZSB(ip);
693 	objset_t *os = zfsvfs->z_os;
694 	int error = 0;
695 
696 	if (S_ISFIFO(ip->i_mode))
697 		return (-ESPIPE);
698 
699 	if (offset < 0 || len < 0)
700 		return (-EINVAL);
701 
702 	if ((error = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
703 		return (error);
704 
705 	switch (advice) {
706 	case POSIX_FADV_SEQUENTIAL:
707 	case POSIX_FADV_WILLNEED:
708 #ifdef HAVE_GENERIC_FADVISE
709 		if (zn_has_cached_data(zp, offset, offset + len - 1))
710 			error = generic_fadvise(filp, offset, len, advice);
711 #endif
712 		/*
713 		 * Pass on the caller's size directly, but note that
714 		 * dmu_prefetch_max will effectively cap it.  If there
715 		 * really is a larger sequential access pattern, perhaps
716 		 * dmu_zfetch will detect it.
717 		 */
718 		if (len == 0)
719 			len = i_size_read(ip) - offset;
720 
721 		dmu_prefetch(os, zp->z_id, 0, offset, len,
722 		    ZIO_PRIORITY_ASYNC_READ);
723 		break;
724 	case POSIX_FADV_NORMAL:
725 	case POSIX_FADV_RANDOM:
726 	case POSIX_FADV_DONTNEED:
727 	case POSIX_FADV_NOREUSE:
728 		/* ignored for now */
729 		break;
730 	default:
731 		error = -EINVAL;
732 		break;
733 	}
734 
735 	zfs_exit(zfsvfs, FTAG);
736 
737 	return (error);
738 }
739 
740 #define	ZFS_FL_USER_VISIBLE	(FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
741 #define	ZFS_FL_USER_MODIFIABLE	(FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
742 
743 static uint32_t
__zpl_ioctl_getflags(struct inode * ip)744 __zpl_ioctl_getflags(struct inode *ip)
745 {
746 	uint64_t zfs_flags = ITOZ(ip)->z_pflags;
747 	uint32_t ioctl_flags = 0;
748 
749 	if (zfs_flags & ZFS_IMMUTABLE)
750 		ioctl_flags |= FS_IMMUTABLE_FL;
751 
752 	if (zfs_flags & ZFS_APPENDONLY)
753 		ioctl_flags |= FS_APPEND_FL;
754 
755 	if (zfs_flags & ZFS_NODUMP)
756 		ioctl_flags |= FS_NODUMP_FL;
757 
758 	if (zfs_flags & ZFS_PROJINHERIT)
759 		ioctl_flags |= ZFS_PROJINHERIT_FL;
760 
761 	return (ioctl_flags & ZFS_FL_USER_VISIBLE);
762 }
763 
764 /*
765  * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
766  * attributes common to both Linux and Solaris are mapped.
767  */
768 static int
zpl_ioctl_getflags(struct file * filp,void __user * arg)769 zpl_ioctl_getflags(struct file *filp, void __user *arg)
770 {
771 	uint32_t flags;
772 	int err;
773 
774 	flags = __zpl_ioctl_getflags(file_inode(filp));
775 	err = copy_to_user(arg, &flags, sizeof (flags));
776 
777 	return (err);
778 }
779 
780 /*
781  * fchange() is a helper macro to detect if we have been asked to change a
782  * flag. This is ugly, but the requirement that we do this is a consequence of
783  * how the Linux file attribute interface was designed. Another consequence is
784  * that concurrent modification of files suffers from a TOCTOU race. Neither
785  * are things we can fix without modifying the kernel-userland interface, which
786  * is outside of our jurisdiction.
787  */
788 
789 #define	fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
790 
791 static int
__zpl_ioctl_setflags(struct inode * ip,uint32_t ioctl_flags,xvattr_t * xva)792 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva)
793 {
794 	uint64_t zfs_flags = ITOZ(ip)->z_pflags;
795 	xoptattr_t *xoap;
796 
797 	if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL |
798 	    ZFS_PROJINHERIT_FL))
799 		return (-EOPNOTSUPP);
800 
801 	if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE)
802 		return (-EACCES);
803 
804 	if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
805 	    fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
806 	    !capable(CAP_LINUX_IMMUTABLE))
807 		return (-EPERM);
808 
809 	if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
810 		return (-EACCES);
811 
812 	xva_init(xva);
813 	xoap = xva_getxoptattr(xva);
814 
815 #define	FLAG_CHANGE(iflag, zflag, xflag, xfield)	do {	\
816 	if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) ||	\
817 	    ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) {	\
818 		XVA_SET_REQ(xva, (xflag));	\
819 		(xfield) = ((ioctl_flags & (iflag)) != 0);	\
820 	}	\
821 } while (0)
822 
823 	FLAG_CHANGE(FS_IMMUTABLE_FL, ZFS_IMMUTABLE, XAT_IMMUTABLE,
824 	    xoap->xoa_immutable);
825 	FLAG_CHANGE(FS_APPEND_FL, ZFS_APPENDONLY, XAT_APPENDONLY,
826 	    xoap->xoa_appendonly);
827 	FLAG_CHANGE(FS_NODUMP_FL, ZFS_NODUMP, XAT_NODUMP,
828 	    xoap->xoa_nodump);
829 	FLAG_CHANGE(ZFS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT,
830 	    xoap->xoa_projinherit);
831 
832 #undef	FLAG_CHANGE
833 
834 	return (0);
835 }
836 
837 static int
zpl_ioctl_setflags(struct file * filp,void __user * arg)838 zpl_ioctl_setflags(struct file *filp, void __user *arg)
839 {
840 	struct inode *ip = file_inode(filp);
841 	uint32_t flags;
842 	cred_t *cr = CRED();
843 	xvattr_t xva;
844 	int err;
845 	fstrans_cookie_t cookie;
846 
847 	if (copy_from_user(&flags, arg, sizeof (flags)))
848 		return (-EFAULT);
849 
850 	err = __zpl_ioctl_setflags(ip, flags, &xva);
851 	if (err)
852 		return (err);
853 
854 	crhold(cr);
855 	cookie = spl_fstrans_mark();
856 	err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
857 	spl_fstrans_unmark(cookie);
858 	crfree(cr);
859 
860 	return (err);
861 }
862 
863 static int
zpl_ioctl_getxattr(struct file * filp,void __user * arg)864 zpl_ioctl_getxattr(struct file *filp, void __user *arg)
865 {
866 	zfsxattr_t fsx = { 0 };
867 	struct inode *ip = file_inode(filp);
868 	int err;
869 
870 	fsx.fsx_xflags = __zpl_ioctl_getflags(ip);
871 	fsx.fsx_projid = ITOZ(ip)->z_projid;
872 	err = copy_to_user(arg, &fsx, sizeof (fsx));
873 
874 	return (err);
875 }
876 
877 static int
zpl_ioctl_setxattr(struct file * filp,void __user * arg)878 zpl_ioctl_setxattr(struct file *filp, void __user *arg)
879 {
880 	struct inode *ip = file_inode(filp);
881 	zfsxattr_t fsx;
882 	cred_t *cr = CRED();
883 	xvattr_t xva;
884 	xoptattr_t *xoap;
885 	int err;
886 	fstrans_cookie_t cookie;
887 
888 	if (copy_from_user(&fsx, arg, sizeof (fsx)))
889 		return (-EFAULT);
890 
891 	if (!zpl_is_valid_projid(fsx.fsx_projid))
892 		return (-EINVAL);
893 
894 	err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva);
895 	if (err)
896 		return (err);
897 
898 	xoap = xva_getxoptattr(&xva);
899 	XVA_SET_REQ(&xva, XAT_PROJID);
900 	xoap->xoa_projid = fsx.fsx_projid;
901 
902 	crhold(cr);
903 	cookie = spl_fstrans_mark();
904 	err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
905 	spl_fstrans_unmark(cookie);
906 	crfree(cr);
907 
908 	return (err);
909 }
910 
911 /*
912  * Expose Additional File Level Attributes of ZFS.
913  */
914 static int
zpl_ioctl_getdosflags(struct file * filp,void __user * arg)915 zpl_ioctl_getdosflags(struct file *filp, void __user *arg)
916 {
917 	struct inode *ip = file_inode(filp);
918 	uint64_t dosflags = ITOZ(ip)->z_pflags;
919 	dosflags &= ZFS_DOS_FL_USER_VISIBLE;
920 	int err = copy_to_user(arg, &dosflags, sizeof (dosflags));
921 
922 	return (err);
923 }
924 
925 static int
__zpl_ioctl_setdosflags(struct inode * ip,uint64_t ioctl_flags,xvattr_t * xva)926 __zpl_ioctl_setdosflags(struct inode *ip, uint64_t ioctl_flags, xvattr_t *xva)
927 {
928 	uint64_t zfs_flags = ITOZ(ip)->z_pflags;
929 	xoptattr_t *xoap;
930 
931 	if (ioctl_flags & (~ZFS_DOS_FL_USER_VISIBLE))
932 		return (-EOPNOTSUPP);
933 
934 	if ((fchange(ioctl_flags, zfs_flags, ZFS_IMMUTABLE, ZFS_IMMUTABLE) ||
935 	    fchange(ioctl_flags, zfs_flags, ZFS_APPENDONLY, ZFS_APPENDONLY)) &&
936 	    !capable(CAP_LINUX_IMMUTABLE))
937 		return (-EPERM);
938 
939 	if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
940 		return (-EACCES);
941 
942 	xva_init(xva);
943 	xoap = xva_getxoptattr(xva);
944 
945 #define	FLAG_CHANGE(iflag, xflag, xfield)	do {	\
946 	if (((ioctl_flags & (iflag)) && !(zfs_flags & (iflag))) ||	\
947 	    ((zfs_flags & (iflag)) && !(ioctl_flags & (iflag)))) {	\
948 		XVA_SET_REQ(xva, (xflag));	\
949 		(xfield) = ((ioctl_flags & (iflag)) != 0);	\
950 	}	\
951 } while (0)
952 
953 	FLAG_CHANGE(ZFS_IMMUTABLE, XAT_IMMUTABLE, xoap->xoa_immutable);
954 	FLAG_CHANGE(ZFS_APPENDONLY, XAT_APPENDONLY, xoap->xoa_appendonly);
955 	FLAG_CHANGE(ZFS_NODUMP, XAT_NODUMP, xoap->xoa_nodump);
956 	FLAG_CHANGE(ZFS_READONLY, XAT_READONLY, xoap->xoa_readonly);
957 	FLAG_CHANGE(ZFS_HIDDEN, XAT_HIDDEN, xoap->xoa_hidden);
958 	FLAG_CHANGE(ZFS_SYSTEM, XAT_SYSTEM, xoap->xoa_system);
959 	FLAG_CHANGE(ZFS_ARCHIVE, XAT_ARCHIVE, xoap->xoa_archive);
960 	FLAG_CHANGE(ZFS_NOUNLINK, XAT_NOUNLINK, xoap->xoa_nounlink);
961 	FLAG_CHANGE(ZFS_REPARSE, XAT_REPARSE, xoap->xoa_reparse);
962 	FLAG_CHANGE(ZFS_OFFLINE, XAT_OFFLINE, xoap->xoa_offline);
963 	FLAG_CHANGE(ZFS_SPARSE, XAT_SPARSE, xoap->xoa_sparse);
964 
965 #undef	FLAG_CHANGE
966 
967 	return (0);
968 }
969 
970 /*
971  * Set Additional File Level Attributes of ZFS.
972  */
973 static int
zpl_ioctl_setdosflags(struct file * filp,void __user * arg)974 zpl_ioctl_setdosflags(struct file *filp, void __user *arg)
975 {
976 	struct inode *ip = file_inode(filp);
977 	uint64_t dosflags;
978 	cred_t *cr = CRED();
979 	xvattr_t xva;
980 	int err;
981 	fstrans_cookie_t cookie;
982 
983 	if (copy_from_user(&dosflags, arg, sizeof (dosflags)))
984 		return (-EFAULT);
985 
986 	err = __zpl_ioctl_setdosflags(ip, dosflags, &xva);
987 	if (err)
988 		return (err);
989 
990 	crhold(cr);
991 	cookie = spl_fstrans_mark();
992 	err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
993 	spl_fstrans_unmark(cookie);
994 	crfree(cr);
995 
996 	return (err);
997 }
998 
999 static int
zpl_ioctl_rewrite(struct file * filp,void __user * arg)1000 zpl_ioctl_rewrite(struct file *filp, void __user *arg)
1001 {
1002 	struct inode *ip = file_inode(filp);
1003 	zfs_rewrite_args_t args;
1004 	fstrans_cookie_t cookie;
1005 	int err;
1006 
1007 	if (copy_from_user(&args, arg, sizeof (args)))
1008 		return (-EFAULT);
1009 
1010 	if (unlikely(!(filp->f_mode & FMODE_WRITE)))
1011 		return (-EBADF);
1012 
1013 	cookie = spl_fstrans_mark();
1014 	err = -zfs_rewrite(ITOZ(ip), args.off, args.len, args.flags, args.arg);
1015 	spl_fstrans_unmark(cookie);
1016 
1017 	return (err);
1018 }
1019 
1020 static long
zpl_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1021 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1022 {
1023 	switch (cmd) {
1024 	case FS_IOC_GETVERSION:
1025 		return (zpl_ioctl_getversion(filp, (void *)arg));
1026 	case FS_IOC_GETFLAGS:
1027 		return (zpl_ioctl_getflags(filp, (void *)arg));
1028 	case FS_IOC_SETFLAGS:
1029 		return (zpl_ioctl_setflags(filp, (void *)arg));
1030 	case ZFS_IOC_FSGETXATTR:
1031 		return (zpl_ioctl_getxattr(filp, (void *)arg));
1032 	case ZFS_IOC_FSSETXATTR:
1033 		return (zpl_ioctl_setxattr(filp, (void *)arg));
1034 	case ZFS_IOC_GETDOSFLAGS:
1035 		return (zpl_ioctl_getdosflags(filp, (void *)arg));
1036 	case ZFS_IOC_SETDOSFLAGS:
1037 		return (zpl_ioctl_setdosflags(filp, (void *)arg));
1038 	case ZFS_IOC_REWRITE:
1039 		return (zpl_ioctl_rewrite(filp, (void *)arg));
1040 	default:
1041 		return (-ENOTTY);
1042 	}
1043 }
1044 
1045 #ifdef CONFIG_COMPAT
1046 static long
zpl_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1047 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1048 {
1049 	switch (cmd) {
1050 	case FS_IOC32_GETVERSION:
1051 		cmd = FS_IOC_GETVERSION;
1052 		break;
1053 	case FS_IOC32_GETFLAGS:
1054 		cmd = FS_IOC_GETFLAGS;
1055 		break;
1056 	case FS_IOC32_SETFLAGS:
1057 		cmd = FS_IOC_SETFLAGS;
1058 		break;
1059 	default:
1060 		return (-ENOTTY);
1061 	}
1062 	return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
1063 }
1064 #endif /* CONFIG_COMPAT */
1065 
1066 const struct address_space_operations zpl_address_space_operations = {
1067 #ifdef HAVE_VFS_READPAGES
1068 	.readpages	= zpl_readpages,
1069 #else
1070 	.readahead	= zpl_readahead,
1071 #endif
1072 #ifdef HAVE_VFS_READ_FOLIO
1073 	.read_folio	= zpl_read_folio,
1074 #else
1075 	.readpage	= zpl_readpage,
1076 #endif
1077 #ifdef HAVE_VFS_WRITEPAGE
1078 	.writepage	= zpl_writepage,
1079 #endif
1080 	.writepages	= zpl_writepages,
1081 	.direct_IO	= zpl_direct_IO,
1082 #ifdef HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS
1083 	.set_page_dirty = __set_page_dirty_nobuffers,
1084 #endif
1085 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
1086 	.dirty_folio	= filemap_dirty_folio,
1087 #endif
1088 #ifdef HAVE_VFS_MIGRATE_FOLIO
1089 	.migrate_folio	= migrate_folio,
1090 #elif defined(HAVE_VFS_MIGRATEPAGE)
1091 	.migratepage	= migrate_page,
1092 #endif
1093 };
1094 
1095 const struct file_operations zpl_file_operations = {
1096 	.open		= zpl_open,
1097 	.release	= zpl_release,
1098 	.llseek		= zpl_llseek,
1099 	.read_iter	= zpl_iter_read,
1100 	.write_iter	= zpl_iter_write,
1101 #ifdef HAVE_COPY_SPLICE_READ
1102 	.splice_read	= copy_splice_read,
1103 #else
1104 	.splice_read	= generic_file_splice_read,
1105 #endif
1106 	.splice_write	= iter_file_splice_write,
1107 	.mmap		= zpl_mmap,
1108 	.fsync		= zpl_fsync,
1109 	.fallocate	= zpl_fallocate,
1110 	.copy_file_range	= zpl_copy_file_range,
1111 #ifdef HAVE_VFS_CLONE_FILE_RANGE
1112 	.clone_file_range	= zpl_clone_file_range,
1113 #endif
1114 #ifdef HAVE_VFS_REMAP_FILE_RANGE
1115 	.remap_file_range	= zpl_remap_file_range,
1116 #endif
1117 #ifdef HAVE_VFS_DEDUPE_FILE_RANGE
1118 	.dedupe_file_range	= zpl_dedupe_file_range,
1119 #endif
1120 	.fadvise	= zpl_fadvise,
1121 	.unlocked_ioctl	= zpl_ioctl,
1122 #ifdef CONFIG_COMPAT
1123 	.compat_ioctl	= zpl_compat_ioctl,
1124 #endif
1125 };
1126 
1127 const struct file_operations zpl_dir_file_operations = {
1128 	.llseek		= generic_file_llseek,
1129 	.read		= generic_read_dir,
1130 	.iterate_shared	= zpl_iterate,
1131 	.fsync		= zpl_fsync,
1132 	.unlocked_ioctl = zpl_ioctl,
1133 #ifdef CONFIG_COMPAT
1134 	.compat_ioctl   = zpl_compat_ioctl,
1135 #endif
1136 };
1137 
1138 module_param(zfs_fallocate_reserve_percent, uint, 0644);
1139 MODULE_PARM_DESC(zfs_fallocate_reserve_percent,
1140 	"Percentage of length to use for the available capacity check");
1141