xref: /freebsd/sys/contrib/openzfs/module/os/linux/zfs/zpl_file.c (revision 5c65a0a9163cc00389d8527ee12c4e69df07ea42)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2011, Lawrence Livermore National Security, LLC.
23  * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
24  */
25 
26 
27 #ifdef CONFIG_COMPAT
28 #include <linux/compat.h>
29 #endif
30 #include <linux/fs.h>
31 #include <linux/migrate.h>
32 #include <sys/file.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/zfs_znode.h>
35 #include <sys/zfs_vfsops.h>
36 #include <sys/zfs_vnops.h>
37 #include <sys/zfs_project.h>
38 #if defined(HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS) || \
39     defined(HAVE_VFS_FILEMAP_DIRTY_FOLIO)
40 #include <linux/pagemap.h>
41 #endif
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 int
zpl_fsync(struct file * filp,loff_t start,loff_t end,int datasync)112 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
113 {
114 	struct inode *inode = filp->f_mapping->host;
115 	znode_t *zp = ITOZ(inode);
116 	zfsvfs_t *zfsvfs = ITOZSB(inode);
117 	cred_t *cr = CRED();
118 	int error;
119 	fstrans_cookie_t cookie;
120 
121 	/*
122 	 * The variables z_sync_writes_cnt and z_async_writes_cnt work in
123 	 * tandem so that sync writes can detect if there are any non-sync
124 	 * writes going on and vice-versa. The "vice-versa" part to this logic
125 	 * is located in zfs_putpage() where non-sync writes check if there are
126 	 * any ongoing sync writes. If any sync and non-sync writes overlap,
127 	 * we do a commit to complete the non-sync writes since the latter can
128 	 * potentially take several seconds to complete and thus block sync
129 	 * writes in the upcoming call to filemap_write_and_wait_range().
130 	 */
131 	atomic_inc_32(&zp->z_sync_writes_cnt);
132 	/*
133 	 * If the following check does not detect an overlapping non-sync write
134 	 * (say because it's just about to start), then it is guaranteed that
135 	 * the non-sync write will detect this sync write. This is because we
136 	 * always increment z_sync_writes_cnt / z_async_writes_cnt before doing
137 	 * the check on z_async_writes_cnt / z_sync_writes_cnt here and in
138 	 * zfs_putpage() respectively.
139 	 */
140 	if (atomic_load_32(&zp->z_async_writes_cnt) > 0) {
141 		if ((error = zpl_enter(zfsvfs, FTAG)) != 0) {
142 			atomic_dec_32(&zp->z_sync_writes_cnt);
143 			return (error);
144 		}
145 		zil_commit(zfsvfs->z_log, zp->z_id);
146 		zpl_exit(zfsvfs, FTAG);
147 	}
148 
149 	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
150 
151 	/*
152 	 * The sync write is not complete yet but we decrement
153 	 * z_sync_writes_cnt since zfs_fsync() increments and decrements
154 	 * it internally. If a non-sync write starts just after the decrement
155 	 * operation but before we call zfs_fsync(), it may not detect this
156 	 * overlapping sync write but it does not matter since we have already
157 	 * gone past filemap_write_and_wait_range() and we won't block due to
158 	 * the non-sync write.
159 	 */
160 	atomic_dec_32(&zp->z_sync_writes_cnt);
161 
162 	if (error)
163 		return (error);
164 
165 	crhold(cr);
166 	cookie = spl_fstrans_mark();
167 	error = -zfs_fsync(zp, datasync, cr);
168 	spl_fstrans_unmark(cookie);
169 	crfree(cr);
170 	ASSERT3S(error, <=, 0);
171 
172 	return (error);
173 }
174 
175 static inline int
zfs_io_flags(struct kiocb * kiocb)176 zfs_io_flags(struct kiocb *kiocb)
177 {
178 	int flags = 0;
179 
180 #if defined(IOCB_DSYNC)
181 	if (kiocb->ki_flags & IOCB_DSYNC)
182 		flags |= O_DSYNC;
183 #endif
184 #if defined(IOCB_SYNC)
185 	if (kiocb->ki_flags & IOCB_SYNC)
186 		flags |= O_SYNC;
187 #endif
188 #if defined(IOCB_APPEND)
189 	if (kiocb->ki_flags & IOCB_APPEND)
190 		flags |= O_APPEND;
191 #endif
192 #if defined(IOCB_DIRECT)
193 	if (kiocb->ki_flags & IOCB_DIRECT)
194 		flags |= O_DIRECT;
195 #endif
196 	return (flags);
197 }
198 
199 /*
200  * If relatime is enabled, call file_accessed() if zfs_relatime_need_update()
201  * is true.  This is needed since datasets with inherited "relatime" property
202  * aren't necessarily mounted with the MNT_RELATIME flag (e.g. after
203  * `zfs set relatime=...`), which is what relatime test in VFS by
204  * relatime_need_update() is based on.
205  */
206 static inline void
zpl_file_accessed(struct file * filp)207 zpl_file_accessed(struct file *filp)
208 {
209 	struct inode *ip = filp->f_mapping->host;
210 
211 	if (!IS_NOATIME(ip) && ITOZSB(ip)->z_relatime) {
212 		if (zfs_relatime_need_update(ip))
213 			file_accessed(filp);
214 	} else {
215 		file_accessed(filp);
216 	}
217 }
218 
219 /*
220  * When HAVE_VFS_IOV_ITER is defined the iov_iter structure supports
221  * iovecs, kvevs, bvecs and pipes, plus all the required interfaces to
222  * manipulate the iov_iter are available.  In which case the full iov_iter
223  * can be attached to the uio and correctly handled in the lower layers.
224  * Otherwise, for older kernels extract the iovec and pass it instead.
225  */
226 static void
zpl_uio_init(zfs_uio_t * uio,struct kiocb * kiocb,struct iov_iter * to,loff_t pos,ssize_t count,size_t skip)227 zpl_uio_init(zfs_uio_t *uio, struct kiocb *kiocb, struct iov_iter *to,
228     loff_t pos, ssize_t count, size_t skip)
229 {
230 #if defined(HAVE_VFS_IOV_ITER)
231 	zfs_uio_iov_iter_init(uio, to, pos, count, skip);
232 #else
233 	zfs_uio_iovec_init(uio, zfs_uio_iter_iov(to), to->nr_segs, pos,
234 	    zfs_uio_iov_iter_type(to) & ITER_KVEC ?
235 	    UIO_SYSSPACE : UIO_USERSPACE,
236 	    count, skip);
237 #endif
238 }
239 
240 static ssize_t
zpl_iter_read(struct kiocb * kiocb,struct iov_iter * to)241 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
242 {
243 	cred_t *cr = CRED();
244 	fstrans_cookie_t cookie;
245 	struct file *filp = kiocb->ki_filp;
246 	ssize_t count = iov_iter_count(to);
247 	zfs_uio_t uio;
248 
249 	zpl_uio_init(&uio, kiocb, to, kiocb->ki_pos, count, 0);
250 
251 	crhold(cr);
252 	cookie = spl_fstrans_mark();
253 
254 	ssize_t ret = -zfs_read(ITOZ(filp->f_mapping->host), &uio,
255 	    filp->f_flags | zfs_io_flags(kiocb), cr);
256 
257 	spl_fstrans_unmark(cookie);
258 	crfree(cr);
259 
260 	if (ret < 0)
261 		return (ret);
262 
263 	ssize_t read = count - uio.uio_resid;
264 	kiocb->ki_pos += read;
265 
266 	zpl_file_accessed(filp);
267 
268 	return (read);
269 }
270 
271 static inline ssize_t
zpl_generic_write_checks(struct kiocb * kiocb,struct iov_iter * from,size_t * countp)272 zpl_generic_write_checks(struct kiocb *kiocb, struct iov_iter *from,
273     size_t *countp)
274 {
275 	ssize_t ret = generic_write_checks(kiocb, from);
276 	if (ret <= 0)
277 		return (ret);
278 
279 	*countp = ret;
280 
281 	return (0);
282 }
283 
284 static ssize_t
zpl_iter_write(struct kiocb * kiocb,struct iov_iter * from)285 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
286 {
287 	cred_t *cr = CRED();
288 	fstrans_cookie_t cookie;
289 	struct file *filp = kiocb->ki_filp;
290 	struct inode *ip = filp->f_mapping->host;
291 	zfs_uio_t uio;
292 	size_t count = 0;
293 	ssize_t ret;
294 
295 	ret = zpl_generic_write_checks(kiocb, from, &count);
296 	if (ret)
297 		return (ret);
298 
299 	zpl_uio_init(&uio, kiocb, from, kiocb->ki_pos, count, from->iov_offset);
300 
301 	crhold(cr);
302 	cookie = spl_fstrans_mark();
303 
304 	ret = -zfs_write(ITOZ(ip), &uio,
305 	    filp->f_flags | zfs_io_flags(kiocb), cr);
306 
307 	spl_fstrans_unmark(cookie);
308 	crfree(cr);
309 
310 	if (ret < 0)
311 		return (ret);
312 
313 	ssize_t wrote = count - uio.uio_resid;
314 	kiocb->ki_pos += wrote;
315 
316 	return (wrote);
317 }
318 
319 static ssize_t
zpl_direct_IO_impl(void)320 zpl_direct_IO_impl(void)
321 {
322 	/*
323 	 * All O_DIRECT requests should be handled by
324 	 * zpl_{iter/aio}_{write/read}(). There is no way kernel generic code
325 	 * should call the direct_IO address_space_operations function. We set
326 	 * this code path to be fatal if it is executed.
327 	 */
328 	PANIC(0);
329 	return (0);
330 }
331 
332 #if defined(HAVE_VFS_DIRECT_IO_ITER)
333 static ssize_t
zpl_direct_IO(struct kiocb * kiocb,struct iov_iter * iter)334 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter)
335 {
336 	return (zpl_direct_IO_impl());
337 }
338 #elif defined(HAVE_VFS_DIRECT_IO_ITER_OFFSET)
339 static ssize_t
zpl_direct_IO(struct kiocb * kiocb,struct iov_iter * iter,loff_t pos)340 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter, loff_t pos)
341 {
342 	return (zpl_direct_IO_impl());
343 }
344 #else
345 #error "Unknown Direct I/O interface"
346 #endif
347 
348 static loff_t
zpl_llseek(struct file * filp,loff_t offset,int whence)349 zpl_llseek(struct file *filp, loff_t offset, int whence)
350 {
351 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
352 	fstrans_cookie_t cookie;
353 
354 	if (whence == SEEK_DATA || whence == SEEK_HOLE) {
355 		struct inode *ip = filp->f_mapping->host;
356 		loff_t maxbytes = ip->i_sb->s_maxbytes;
357 		loff_t error;
358 
359 		spl_inode_lock_shared(ip);
360 		cookie = spl_fstrans_mark();
361 		error = -zfs_holey(ITOZ(ip), whence, &offset);
362 		spl_fstrans_unmark(cookie);
363 		if (error == 0)
364 			error = lseek_execute(filp, ip, offset, maxbytes);
365 		spl_inode_unlock_shared(ip);
366 
367 		return (error);
368 	}
369 #endif /* SEEK_HOLE && SEEK_DATA */
370 
371 	return (generic_file_llseek(filp, offset, whence));
372 }
373 
374 /*
375  * It's worth taking a moment to describe how mmap is implemented
376  * for zfs because it differs considerably from other Linux filesystems.
377  * However, this issue is handled the same way under OpenSolaris.
378  *
379  * The issue is that by design zfs bypasses the Linux page cache and
380  * leaves all caching up to the ARC.  This has been shown to work
381  * well for the common read(2)/write(2) case.  However, mmap(2)
382  * is problem because it relies on being tightly integrated with the
383  * page cache.  To handle this we cache mmap'ed files twice, once in
384  * the ARC and a second time in the page cache.  The code is careful
385  * to keep both copies synchronized.
386  *
387  * When a file with an mmap'ed region is written to using write(2)
388  * both the data in the ARC and existing pages in the page cache
389  * are updated.  For a read(2) data will be read first from the page
390  * cache then the ARC if needed.  Neither a write(2) or read(2) will
391  * will ever result in new pages being added to the page cache.
392  *
393  * New pages are added to the page cache only via .readpage() which
394  * is called when the vfs needs to read a page off disk to back the
395  * virtual memory region.  These pages may be modified without
396  * notifying the ARC and will be written out periodically via
397  * .writepage().  This will occur due to either a sync or the usual
398  * page aging behavior.  Note because a read(2) of a mmap'ed file
399  * will always check the page cache first even when the ARC is out
400  * of date correct data will still be returned.
401  *
402  * While this implementation ensures correct behavior it does have
403  * have some drawbacks.  The most obvious of which is that it
404  * increases the required memory footprint when access mmap'ed
405  * files.  It also adds additional complexity to the code keeping
406  * both caches synchronized.
407  *
408  * Longer term it may be possible to cleanly resolve this wart by
409  * mapping page cache pages directly on to the ARC buffers.  The
410  * Linux address space operations are flexible enough to allow
411  * selection of which pages back a particular index.  The trick
412  * would be working out the details of which subsystem is in
413  * charge, the ARC, the page cache, or both.  It may also prove
414  * helpful to move the ARC buffers to a scatter-gather lists
415  * rather than a vmalloc'ed region.
416  */
417 static int
zpl_mmap(struct file * filp,struct vm_area_struct * vma)418 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
419 {
420 	struct inode *ip = filp->f_mapping->host;
421 	int error;
422 	fstrans_cookie_t cookie;
423 
424 	cookie = spl_fstrans_mark();
425 	error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
426 	    (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
427 	spl_fstrans_unmark(cookie);
428 
429 	if (error)
430 		return (error);
431 
432 	error = generic_file_mmap(filp, vma);
433 	if (error)
434 		return (error);
435 
436 	return (error);
437 }
438 
439 /*
440  * Populate a page with data for the Linux page cache.  This function is
441  * only used to support mmap(2).  There will be an identical copy of the
442  * data in the ARC which is kept up to date via .write() and .writepage().
443  */
444 static inline int
zpl_readpage_common(struct page * pp)445 zpl_readpage_common(struct page *pp)
446 {
447 	fstrans_cookie_t cookie;
448 
449 	ASSERT(PageLocked(pp));
450 
451 	cookie = spl_fstrans_mark();
452 	int error = -zfs_getpage(pp->mapping->host, pp);
453 	spl_fstrans_unmark(cookie);
454 
455 	unlock_page(pp);
456 
457 	return (error);
458 }
459 
460 #ifdef HAVE_VFS_READ_FOLIO
461 static int
zpl_read_folio(struct file * filp,struct folio * folio)462 zpl_read_folio(struct file *filp, struct folio *folio)
463 {
464 	return (zpl_readpage_common(&folio->page));
465 }
466 #else
467 static int
zpl_readpage(struct file * filp,struct page * pp)468 zpl_readpage(struct file *filp, struct page *pp)
469 {
470 	return (zpl_readpage_common(pp));
471 }
472 #endif
473 
474 static int
zpl_readpage_filler(void * data,struct page * pp)475 zpl_readpage_filler(void *data, struct page *pp)
476 {
477 	return (zpl_readpage_common(pp));
478 }
479 
480 /*
481  * Populate a set of pages with data for the Linux page cache.  This
482  * function will only be called for read ahead and never for demand
483  * paging.  For simplicity, the code relies on read_cache_pages() to
484  * correctly lock each page for IO and call zpl_readpage().
485  */
486 #ifdef HAVE_VFS_READPAGES
487 static int
zpl_readpages(struct file * filp,struct address_space * mapping,struct list_head * pages,unsigned nr_pages)488 zpl_readpages(struct file *filp, struct address_space *mapping,
489     struct list_head *pages, unsigned nr_pages)
490 {
491 	return (read_cache_pages(mapping, pages, zpl_readpage_filler, NULL));
492 }
493 #else
494 static void
zpl_readahead(struct readahead_control * ractl)495 zpl_readahead(struct readahead_control *ractl)
496 {
497 	struct page *page;
498 
499 	while ((page = readahead_page(ractl)) != NULL) {
500 		int ret;
501 
502 		ret = zpl_readpage_filler(NULL, page);
503 		put_page(page);
504 		if (ret)
505 			break;
506 	}
507 }
508 #endif
509 
510 static int
zpl_putpage(struct page * pp,struct writeback_control * wbc,void * data)511 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
512 {
513 	boolean_t *for_sync = data;
514 	fstrans_cookie_t cookie;
515 	int ret;
516 
517 	ASSERT(PageLocked(pp));
518 	ASSERT(!PageWriteback(pp));
519 
520 	cookie = spl_fstrans_mark();
521 	ret = zfs_putpage(pp->mapping->host, pp, wbc, *for_sync);
522 	spl_fstrans_unmark(cookie);
523 
524 	return (ret);
525 }
526 
527 #ifdef HAVE_WRITEPAGE_T_FOLIO
528 static int
zpl_putfolio(struct folio * pp,struct writeback_control * wbc,void * data)529 zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data)
530 {
531 	return (zpl_putpage(&pp->page, wbc, data));
532 }
533 #endif
534 
535 static inline int
zpl_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,void * data)536 zpl_write_cache_pages(struct address_space *mapping,
537     struct writeback_control *wbc, void *data)
538 {
539 	int result;
540 
541 #ifdef HAVE_WRITEPAGE_T_FOLIO
542 	result = write_cache_pages(mapping, wbc, zpl_putfolio, data);
543 #else
544 	result = write_cache_pages(mapping, wbc, zpl_putpage, data);
545 #endif
546 	return (result);
547 }
548 
549 static int
zpl_writepages(struct address_space * mapping,struct writeback_control * wbc)550 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
551 {
552 	znode_t		*zp = ITOZ(mapping->host);
553 	zfsvfs_t	*zfsvfs = ITOZSB(mapping->host);
554 	enum writeback_sync_modes sync_mode;
555 	int result;
556 
557 	if ((result = zpl_enter(zfsvfs, FTAG)) != 0)
558 		return (result);
559 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
560 		wbc->sync_mode = WB_SYNC_ALL;
561 	zpl_exit(zfsvfs, FTAG);
562 	sync_mode = wbc->sync_mode;
563 
564 	/*
565 	 * We don't want to run write_cache_pages() in SYNC mode here, because
566 	 * that would make putpage() wait for a single page to be committed to
567 	 * disk every single time, resulting in atrocious performance. Instead
568 	 * we run it once in non-SYNC mode so that the ZIL gets all the data,
569 	 * and then we commit it all in one go.
570 	 */
571 	boolean_t for_sync = (sync_mode == WB_SYNC_ALL);
572 	wbc->sync_mode = WB_SYNC_NONE;
573 	result = zpl_write_cache_pages(mapping, wbc, &for_sync);
574 	if (sync_mode != wbc->sync_mode) {
575 		if ((result = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
576 			return (result);
577 		if (zfsvfs->z_log != NULL)
578 			zil_commit(zfsvfs->z_log, zp->z_id);
579 		zpl_exit(zfsvfs, FTAG);
580 
581 		/*
582 		 * We need to call write_cache_pages() again (we can't just
583 		 * return after the commit) because the previous call in
584 		 * non-SYNC mode does not guarantee that we got all the dirty
585 		 * pages (see the implementation of write_cache_pages() for
586 		 * details). That being said, this is a no-op in most cases.
587 		 */
588 		wbc->sync_mode = sync_mode;
589 		result = zpl_write_cache_pages(mapping, wbc, &for_sync);
590 	}
591 	return (result);
592 }
593 
594 /*
595  * Write out dirty pages to the ARC, this function is only required to
596  * support mmap(2).  Mapped pages may be dirtied by memory operations
597  * which never call .write().  These dirty pages are kept in sync with
598  * the ARC buffers via this hook.
599  */
600 static int
zpl_writepage(struct page * pp,struct writeback_control * wbc)601 zpl_writepage(struct page *pp, struct writeback_control *wbc)
602 {
603 	if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
604 		wbc->sync_mode = WB_SYNC_ALL;
605 
606 	boolean_t for_sync = (wbc->sync_mode == WB_SYNC_ALL);
607 
608 	return (zpl_putpage(pp, wbc, &for_sync));
609 }
610 
611 /*
612  * The flag combination which matches the behavior of zfs_space() is
613  * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
614  * flag was introduced in the 2.6.38 kernel.
615  *
616  * The original mode=0 (allocate space) behavior can be reasonably emulated
617  * by checking if enough space exists and creating a sparse file, as real
618  * persistent space reservation is not possible due to COW, snapshots, etc.
619  */
620 static long
zpl_fallocate_common(struct inode * ip,int mode,loff_t offset,loff_t len)621 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
622 {
623 	cred_t *cr = CRED();
624 	loff_t olen;
625 	fstrans_cookie_t cookie;
626 	int error = 0;
627 
628 	int test_mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE;
629 
630 	if ((mode & ~(FALLOC_FL_KEEP_SIZE | test_mode)) != 0)
631 		return (-EOPNOTSUPP);
632 
633 	if (offset < 0 || len <= 0)
634 		return (-EINVAL);
635 
636 	spl_inode_lock(ip);
637 	olen = i_size_read(ip);
638 
639 	crhold(cr);
640 	cookie = spl_fstrans_mark();
641 	if (mode & (test_mode)) {
642 		flock64_t bf;
643 
644 		if (mode & FALLOC_FL_KEEP_SIZE) {
645 			if (offset > olen)
646 				goto out_unmark;
647 
648 			if (offset + len > olen)
649 				len = olen - offset;
650 		}
651 		bf.l_type = F_WRLCK;
652 		bf.l_whence = SEEK_SET;
653 		bf.l_start = offset;
654 		bf.l_len = len;
655 		bf.l_pid = 0;
656 
657 		error = -zfs_space(ITOZ(ip), F_FREESP, &bf, O_RDWR, offset, cr);
658 	} else if ((mode & ~FALLOC_FL_KEEP_SIZE) == 0) {
659 		unsigned int percent = zfs_fallocate_reserve_percent;
660 		struct kstatfs statfs;
661 
662 		/* Legacy mode, disable fallocate compatibility. */
663 		if (percent == 0) {
664 			error = -EOPNOTSUPP;
665 			goto out_unmark;
666 		}
667 
668 		/*
669 		 * Use zfs_statvfs() instead of dmu_objset_space() since it
670 		 * also checks project quota limits, which are relevant here.
671 		 */
672 		error = zfs_statvfs(ip, &statfs);
673 		if (error)
674 			goto out_unmark;
675 
676 		/*
677 		 * Shrink available space a bit to account for overhead/races.
678 		 * We know the product previously fit into availbytes from
679 		 * dmu_objset_space(), so the smaller product will also fit.
680 		 */
681 		if (len > statfs.f_bavail * (statfs.f_bsize * 100 / percent)) {
682 			error = -ENOSPC;
683 			goto out_unmark;
684 		}
685 		if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > olen)
686 			error = zfs_freesp(ITOZ(ip), offset + len, 0, 0, FALSE);
687 	}
688 out_unmark:
689 	spl_fstrans_unmark(cookie);
690 	spl_inode_unlock(ip);
691 
692 	crfree(cr);
693 
694 	return (error);
695 }
696 
697 static long
zpl_fallocate(struct file * filp,int mode,loff_t offset,loff_t len)698 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
699 {
700 	return zpl_fallocate_common(file_inode(filp),
701 	    mode, offset, len);
702 }
703 
704 static int
zpl_ioctl_getversion(struct file * filp,void __user * arg)705 zpl_ioctl_getversion(struct file *filp, void __user *arg)
706 {
707 	uint32_t generation = file_inode(filp)->i_generation;
708 
709 	return (copy_to_user(arg, &generation, sizeof (generation)));
710 }
711 
712 static int
zpl_fadvise(struct file * filp,loff_t offset,loff_t len,int advice)713 zpl_fadvise(struct file *filp, loff_t offset, loff_t len, int advice)
714 {
715 	struct inode *ip = file_inode(filp);
716 	znode_t *zp = ITOZ(ip);
717 	zfsvfs_t *zfsvfs = ITOZSB(ip);
718 	objset_t *os = zfsvfs->z_os;
719 	int error = 0;
720 
721 	if (S_ISFIFO(ip->i_mode))
722 		return (-ESPIPE);
723 
724 	if (offset < 0 || len < 0)
725 		return (-EINVAL);
726 
727 	if ((error = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
728 		return (error);
729 
730 	switch (advice) {
731 	case POSIX_FADV_SEQUENTIAL:
732 	case POSIX_FADV_WILLNEED:
733 #ifdef HAVE_GENERIC_FADVISE
734 		if (zn_has_cached_data(zp, offset, offset + len - 1))
735 			error = generic_fadvise(filp, offset, len, advice);
736 #endif
737 		/*
738 		 * Pass on the caller's size directly, but note that
739 		 * dmu_prefetch_max will effectively cap it.  If there
740 		 * really is a larger sequential access pattern, perhaps
741 		 * dmu_zfetch will detect it.
742 		 */
743 		if (len == 0)
744 			len = i_size_read(ip) - offset;
745 
746 		dmu_prefetch(os, zp->z_id, 0, offset, len,
747 		    ZIO_PRIORITY_ASYNC_READ);
748 		break;
749 	case POSIX_FADV_NORMAL:
750 	case POSIX_FADV_RANDOM:
751 	case POSIX_FADV_DONTNEED:
752 	case POSIX_FADV_NOREUSE:
753 		/* ignored for now */
754 		break;
755 	default:
756 		error = -EINVAL;
757 		break;
758 	}
759 
760 	zfs_exit(zfsvfs, FTAG);
761 
762 	return (error);
763 }
764 
765 #define	ZFS_FL_USER_VISIBLE	(FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
766 #define	ZFS_FL_USER_MODIFIABLE	(FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
767 
768 static uint32_t
__zpl_ioctl_getflags(struct inode * ip)769 __zpl_ioctl_getflags(struct inode *ip)
770 {
771 	uint64_t zfs_flags = ITOZ(ip)->z_pflags;
772 	uint32_t ioctl_flags = 0;
773 
774 	if (zfs_flags & ZFS_IMMUTABLE)
775 		ioctl_flags |= FS_IMMUTABLE_FL;
776 
777 	if (zfs_flags & ZFS_APPENDONLY)
778 		ioctl_flags |= FS_APPEND_FL;
779 
780 	if (zfs_flags & ZFS_NODUMP)
781 		ioctl_flags |= FS_NODUMP_FL;
782 
783 	if (zfs_flags & ZFS_PROJINHERIT)
784 		ioctl_flags |= ZFS_PROJINHERIT_FL;
785 
786 	return (ioctl_flags & ZFS_FL_USER_VISIBLE);
787 }
788 
789 /*
790  * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
791  * attributes common to both Linux and Solaris are mapped.
792  */
793 static int
zpl_ioctl_getflags(struct file * filp,void __user * arg)794 zpl_ioctl_getflags(struct file *filp, void __user *arg)
795 {
796 	uint32_t flags;
797 	int err;
798 
799 	flags = __zpl_ioctl_getflags(file_inode(filp));
800 	err = copy_to_user(arg, &flags, sizeof (flags));
801 
802 	return (err);
803 }
804 
805 /*
806  * fchange() is a helper macro to detect if we have been asked to change a
807  * flag. This is ugly, but the requirement that we do this is a consequence of
808  * how the Linux file attribute interface was designed. Another consequence is
809  * that concurrent modification of files suffers from a TOCTOU race. Neither
810  * are things we can fix without modifying the kernel-userland interface, which
811  * is outside of our jurisdiction.
812  */
813 
814 #define	fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
815 
816 static int
__zpl_ioctl_setflags(struct inode * ip,uint32_t ioctl_flags,xvattr_t * xva)817 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva)
818 {
819 	uint64_t zfs_flags = ITOZ(ip)->z_pflags;
820 	xoptattr_t *xoap;
821 
822 	if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL |
823 	    ZFS_PROJINHERIT_FL))
824 		return (-EOPNOTSUPP);
825 
826 	if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE)
827 		return (-EACCES);
828 
829 	if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
830 	    fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
831 	    !capable(CAP_LINUX_IMMUTABLE))
832 		return (-EPERM);
833 
834 	if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
835 		return (-EACCES);
836 
837 	xva_init(xva);
838 	xoap = xva_getxoptattr(xva);
839 
840 #define	FLAG_CHANGE(iflag, zflag, xflag, xfield)	do {	\
841 	if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) ||	\
842 	    ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) {	\
843 		XVA_SET_REQ(xva, (xflag));	\
844 		(xfield) = ((ioctl_flags & (iflag)) != 0);	\
845 	}	\
846 } while (0)
847 
848 	FLAG_CHANGE(FS_IMMUTABLE_FL, ZFS_IMMUTABLE, XAT_IMMUTABLE,
849 	    xoap->xoa_immutable);
850 	FLAG_CHANGE(FS_APPEND_FL, ZFS_APPENDONLY, XAT_APPENDONLY,
851 	    xoap->xoa_appendonly);
852 	FLAG_CHANGE(FS_NODUMP_FL, ZFS_NODUMP, XAT_NODUMP,
853 	    xoap->xoa_nodump);
854 	FLAG_CHANGE(ZFS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT,
855 	    xoap->xoa_projinherit);
856 
857 #undef	FLAG_CHANGE
858 
859 	return (0);
860 }
861 
862 static int
zpl_ioctl_setflags(struct file * filp,void __user * arg)863 zpl_ioctl_setflags(struct file *filp, void __user *arg)
864 {
865 	struct inode *ip = file_inode(filp);
866 	uint32_t flags;
867 	cred_t *cr = CRED();
868 	xvattr_t xva;
869 	int err;
870 	fstrans_cookie_t cookie;
871 
872 	if (copy_from_user(&flags, arg, sizeof (flags)))
873 		return (-EFAULT);
874 
875 	err = __zpl_ioctl_setflags(ip, flags, &xva);
876 	if (err)
877 		return (err);
878 
879 	crhold(cr);
880 	cookie = spl_fstrans_mark();
881 	err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
882 	spl_fstrans_unmark(cookie);
883 	crfree(cr);
884 
885 	return (err);
886 }
887 
888 static int
zpl_ioctl_getxattr(struct file * filp,void __user * arg)889 zpl_ioctl_getxattr(struct file *filp, void __user *arg)
890 {
891 	zfsxattr_t fsx = { 0 };
892 	struct inode *ip = file_inode(filp);
893 	int err;
894 
895 	fsx.fsx_xflags = __zpl_ioctl_getflags(ip);
896 	fsx.fsx_projid = ITOZ(ip)->z_projid;
897 	err = copy_to_user(arg, &fsx, sizeof (fsx));
898 
899 	return (err);
900 }
901 
902 static int
zpl_ioctl_setxattr(struct file * filp,void __user * arg)903 zpl_ioctl_setxattr(struct file *filp, void __user *arg)
904 {
905 	struct inode *ip = file_inode(filp);
906 	zfsxattr_t fsx;
907 	cred_t *cr = CRED();
908 	xvattr_t xva;
909 	xoptattr_t *xoap;
910 	int err;
911 	fstrans_cookie_t cookie;
912 
913 	if (copy_from_user(&fsx, arg, sizeof (fsx)))
914 		return (-EFAULT);
915 
916 	if (!zpl_is_valid_projid(fsx.fsx_projid))
917 		return (-EINVAL);
918 
919 	err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva);
920 	if (err)
921 		return (err);
922 
923 	xoap = xva_getxoptattr(&xva);
924 	XVA_SET_REQ(&xva, XAT_PROJID);
925 	xoap->xoa_projid = fsx.fsx_projid;
926 
927 	crhold(cr);
928 	cookie = spl_fstrans_mark();
929 	err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
930 	spl_fstrans_unmark(cookie);
931 	crfree(cr);
932 
933 	return (err);
934 }
935 
936 /*
937  * Expose Additional File Level Attributes of ZFS.
938  */
939 static int
zpl_ioctl_getdosflags(struct file * filp,void __user * arg)940 zpl_ioctl_getdosflags(struct file *filp, void __user *arg)
941 {
942 	struct inode *ip = file_inode(filp);
943 	uint64_t dosflags = ITOZ(ip)->z_pflags;
944 	dosflags &= ZFS_DOS_FL_USER_VISIBLE;
945 	int err = copy_to_user(arg, &dosflags, sizeof (dosflags));
946 
947 	return (err);
948 }
949 
950 static int
__zpl_ioctl_setdosflags(struct inode * ip,uint64_t ioctl_flags,xvattr_t * xva)951 __zpl_ioctl_setdosflags(struct inode *ip, uint64_t ioctl_flags, xvattr_t *xva)
952 {
953 	uint64_t zfs_flags = ITOZ(ip)->z_pflags;
954 	xoptattr_t *xoap;
955 
956 	if (ioctl_flags & (~ZFS_DOS_FL_USER_VISIBLE))
957 		return (-EOPNOTSUPP);
958 
959 	if ((fchange(ioctl_flags, zfs_flags, ZFS_IMMUTABLE, ZFS_IMMUTABLE) ||
960 	    fchange(ioctl_flags, zfs_flags, ZFS_APPENDONLY, ZFS_APPENDONLY)) &&
961 	    !capable(CAP_LINUX_IMMUTABLE))
962 		return (-EPERM);
963 
964 	if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
965 		return (-EACCES);
966 
967 	xva_init(xva);
968 	xoap = xva_getxoptattr(xva);
969 
970 #define	FLAG_CHANGE(iflag, xflag, xfield)	do {	\
971 	if (((ioctl_flags & (iflag)) && !(zfs_flags & (iflag))) ||	\
972 	    ((zfs_flags & (iflag)) && !(ioctl_flags & (iflag)))) {	\
973 		XVA_SET_REQ(xva, (xflag));	\
974 		(xfield) = ((ioctl_flags & (iflag)) != 0);	\
975 	}	\
976 } while (0)
977 
978 	FLAG_CHANGE(ZFS_IMMUTABLE, XAT_IMMUTABLE, xoap->xoa_immutable);
979 	FLAG_CHANGE(ZFS_APPENDONLY, XAT_APPENDONLY, xoap->xoa_appendonly);
980 	FLAG_CHANGE(ZFS_NODUMP, XAT_NODUMP, xoap->xoa_nodump);
981 	FLAG_CHANGE(ZFS_READONLY, XAT_READONLY, xoap->xoa_readonly);
982 	FLAG_CHANGE(ZFS_HIDDEN, XAT_HIDDEN, xoap->xoa_hidden);
983 	FLAG_CHANGE(ZFS_SYSTEM, XAT_SYSTEM, xoap->xoa_system);
984 	FLAG_CHANGE(ZFS_ARCHIVE, XAT_ARCHIVE, xoap->xoa_archive);
985 	FLAG_CHANGE(ZFS_NOUNLINK, XAT_NOUNLINK, xoap->xoa_nounlink);
986 	FLAG_CHANGE(ZFS_REPARSE, XAT_REPARSE, xoap->xoa_reparse);
987 	FLAG_CHANGE(ZFS_OFFLINE, XAT_OFFLINE, xoap->xoa_offline);
988 	FLAG_CHANGE(ZFS_SPARSE, XAT_SPARSE, xoap->xoa_sparse);
989 
990 #undef	FLAG_CHANGE
991 
992 	return (0);
993 }
994 
995 /*
996  * Set Additional File Level Attributes of ZFS.
997  */
998 static int
zpl_ioctl_setdosflags(struct file * filp,void __user * arg)999 zpl_ioctl_setdosflags(struct file *filp, void __user *arg)
1000 {
1001 	struct inode *ip = file_inode(filp);
1002 	uint64_t dosflags;
1003 	cred_t *cr = CRED();
1004 	xvattr_t xva;
1005 	int err;
1006 	fstrans_cookie_t cookie;
1007 
1008 	if (copy_from_user(&dosflags, arg, sizeof (dosflags)))
1009 		return (-EFAULT);
1010 
1011 	err = __zpl_ioctl_setdosflags(ip, dosflags, &xva);
1012 	if (err)
1013 		return (err);
1014 
1015 	crhold(cr);
1016 	cookie = spl_fstrans_mark();
1017 	err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap);
1018 	spl_fstrans_unmark(cookie);
1019 	crfree(cr);
1020 
1021 	return (err);
1022 }
1023 
1024 static long
zpl_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1025 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1026 {
1027 	switch (cmd) {
1028 	case FS_IOC_GETVERSION:
1029 		return (zpl_ioctl_getversion(filp, (void *)arg));
1030 	case FS_IOC_GETFLAGS:
1031 		return (zpl_ioctl_getflags(filp, (void *)arg));
1032 	case FS_IOC_SETFLAGS:
1033 		return (zpl_ioctl_setflags(filp, (void *)arg));
1034 	case ZFS_IOC_FSGETXATTR:
1035 		return (zpl_ioctl_getxattr(filp, (void *)arg));
1036 	case ZFS_IOC_FSSETXATTR:
1037 		return (zpl_ioctl_setxattr(filp, (void *)arg));
1038 	case ZFS_IOC_GETDOSFLAGS:
1039 		return (zpl_ioctl_getdosflags(filp, (void *)arg));
1040 	case ZFS_IOC_SETDOSFLAGS:
1041 		return (zpl_ioctl_setdosflags(filp, (void *)arg));
1042 	case ZFS_IOC_COMPAT_FICLONE:
1043 		return (zpl_ioctl_ficlone(filp, (void *)arg));
1044 	case ZFS_IOC_COMPAT_FICLONERANGE:
1045 		return (zpl_ioctl_ficlonerange(filp, (void *)arg));
1046 	case ZFS_IOC_COMPAT_FIDEDUPERANGE:
1047 		return (zpl_ioctl_fideduperange(filp, (void *)arg));
1048 	default:
1049 		return (-ENOTTY);
1050 	}
1051 }
1052 
1053 #ifdef CONFIG_COMPAT
1054 static long
zpl_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)1055 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1056 {
1057 	switch (cmd) {
1058 	case FS_IOC32_GETVERSION:
1059 		cmd = FS_IOC_GETVERSION;
1060 		break;
1061 	case FS_IOC32_GETFLAGS:
1062 		cmd = FS_IOC_GETFLAGS;
1063 		break;
1064 	case FS_IOC32_SETFLAGS:
1065 		cmd = FS_IOC_SETFLAGS;
1066 		break;
1067 	default:
1068 		return (-ENOTTY);
1069 	}
1070 	return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
1071 }
1072 #endif /* CONFIG_COMPAT */
1073 
1074 const struct address_space_operations zpl_address_space_operations = {
1075 #ifdef HAVE_VFS_READPAGES
1076 	.readpages	= zpl_readpages,
1077 #else
1078 	.readahead	= zpl_readahead,
1079 #endif
1080 #ifdef HAVE_VFS_READ_FOLIO
1081 	.read_folio	= zpl_read_folio,
1082 #else
1083 	.readpage	= zpl_readpage,
1084 #endif
1085 	.writepage	= zpl_writepage,
1086 	.writepages	= zpl_writepages,
1087 	.direct_IO	= zpl_direct_IO,
1088 #ifdef HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS
1089 	.set_page_dirty = __set_page_dirty_nobuffers,
1090 #endif
1091 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
1092 	.dirty_folio	= filemap_dirty_folio,
1093 #endif
1094 #ifdef HAVE_VFS_MIGRATE_FOLIO
1095 	.migrate_folio	= migrate_folio,
1096 #else
1097 	.migratepage	= migrate_page,
1098 #endif
1099 };
1100 
1101 const struct file_operations zpl_file_operations = {
1102 	.open		= zpl_open,
1103 	.release	= zpl_release,
1104 	.llseek		= zpl_llseek,
1105 	.read_iter	= zpl_iter_read,
1106 	.write_iter	= zpl_iter_write,
1107 #ifdef HAVE_VFS_IOV_ITER
1108 #ifdef HAVE_COPY_SPLICE_READ
1109 	.splice_read	= copy_splice_read,
1110 #else
1111 	.splice_read	= generic_file_splice_read,
1112 #endif
1113 	.splice_write	= iter_file_splice_write,
1114 #endif
1115 	.mmap		= zpl_mmap,
1116 	.fsync		= zpl_fsync,
1117 	.fallocate	= zpl_fallocate,
1118 	.copy_file_range	= zpl_copy_file_range,
1119 #ifdef HAVE_VFS_CLONE_FILE_RANGE
1120 	.clone_file_range	= zpl_clone_file_range,
1121 #endif
1122 #ifdef HAVE_VFS_REMAP_FILE_RANGE
1123 	.remap_file_range	= zpl_remap_file_range,
1124 #endif
1125 #ifdef HAVE_VFS_DEDUPE_FILE_RANGE
1126 	.dedupe_file_range	= zpl_dedupe_file_range,
1127 #endif
1128 	.fadvise	= zpl_fadvise,
1129 	.unlocked_ioctl	= zpl_ioctl,
1130 #ifdef CONFIG_COMPAT
1131 	.compat_ioctl	= zpl_compat_ioctl,
1132 #endif
1133 };
1134 
1135 const struct file_operations zpl_dir_file_operations = {
1136 	.llseek		= generic_file_llseek,
1137 	.read		= generic_read_dir,
1138 	.iterate_shared	= zpl_iterate,
1139 	.fsync		= zpl_fsync,
1140 	.unlocked_ioctl = zpl_ioctl,
1141 #ifdef CONFIG_COMPAT
1142 	.compat_ioctl   = zpl_compat_ioctl,
1143 #endif
1144 };
1145 
1146 /* CSTYLED */
1147 module_param(zfs_fallocate_reserve_percent, uint, 0644);
1148 MODULE_PARM_DESC(zfs_fallocate_reserve_percent,
1149 	"Percentage of length to use for the available capacity check");
1150