• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/slab.h>
3 #include <linux/stat.h>
4 #include <linux/sched/xacct.h>
5 #include <linux/fcntl.h>
6 #include <linux/file.h>
7 #include <linux/uio.h>
8 #include <linux/fsnotify.h>
9 #include <linux/security.h>
10 #include <linux/export.h>
11 #include <linux/syscalls.h>
12 #include <linux/pagemap.h>
13 #include <linux/splice.h>
14 #include <linux/compat.h>
15 #include <linux/mount.h>
16 #include <linux/fs.h>
17 #include "internal.h"
18 
19 #include <linux/uaccess.h>
20 #include <asm/unistd.h>
21 
22 /*
23  * Performs necessary checks before doing a clone.
24  *
25  * Can adjust amount of bytes to clone via @req_count argument.
26  * Returns appropriate error code that caller should return or
27  * zero in case the clone should be allowed.
28  */
generic_remap_checks(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t * req_count,unsigned int remap_flags)29 static int generic_remap_checks(struct file *file_in, loff_t pos_in,
30 				struct file *file_out, loff_t pos_out,
31 				loff_t *req_count, unsigned int remap_flags)
32 {
33 	struct inode *inode_in = file_in->f_mapping->host;
34 	struct inode *inode_out = file_out->f_mapping->host;
35 	uint64_t count = *req_count;
36 	uint64_t bcount;
37 	loff_t size_in, size_out;
38 	loff_t bs = inode_out->i_sb->s_blocksize;
39 	int ret;
40 
41 	/* The start of both ranges must be aligned to an fs block. */
42 	if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
43 		return -EINVAL;
44 
45 	/* Ensure offsets don't wrap. */
46 	if (pos_in + count < pos_in || pos_out + count < pos_out)
47 		return -EINVAL;
48 
49 	size_in = i_size_read(inode_in);
50 	size_out = i_size_read(inode_out);
51 
52 	/* Dedupe requires both ranges to be within EOF. */
53 	if ((remap_flags & REMAP_FILE_DEDUP) &&
54 	    (pos_in >= size_in || pos_in + count > size_in ||
55 	     pos_out >= size_out || pos_out + count > size_out))
56 		return -EINVAL;
57 
58 	/* Ensure the infile range is within the infile. */
59 	if (pos_in >= size_in)
60 		return -EINVAL;
61 	count = min(count, size_in - (uint64_t)pos_in);
62 
63 	ret = generic_write_check_limits(file_out, pos_out, &count);
64 	if (ret)
65 		return ret;
66 
67 	/*
68 	 * If the user wanted us to link to the infile's EOF, round up to the
69 	 * next block boundary for this check.
70 	 *
71 	 * Otherwise, make sure the count is also block-aligned, having
72 	 * already confirmed the starting offsets' block alignment.
73 	 */
74 	if (pos_in + count == size_in &&
75 	    (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
76 		bcount = ALIGN(size_in, bs) - pos_in;
77 	} else {
78 		if (!IS_ALIGNED(count, bs))
79 			count = ALIGN_DOWN(count, bs);
80 		bcount = count;
81 	}
82 
83 	/* Don't allow overlapped cloning within the same file. */
84 	if (inode_in == inode_out &&
85 	    pos_out + bcount > pos_in &&
86 	    pos_out < pos_in + bcount)
87 		return -EINVAL;
88 
89 	/*
90 	 * We shortened the request but the caller can't deal with that, so
91 	 * bounce the request back to userspace.
92 	 */
93 	if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
94 		return -EINVAL;
95 
96 	*req_count = count;
97 	return 0;
98 }
99 
remap_verify_area(struct file * file,loff_t pos,loff_t len,bool write)100 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
101 			     bool write)
102 {
103 	if (unlikely(pos < 0 || len < 0))
104 		return -EINVAL;
105 
106 	if (unlikely((loff_t) (pos + len) < 0))
107 		return -EINVAL;
108 
109 	return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
110 }
111 
112 /*
113  * Ensure that we don't remap a partial EOF block in the middle of something
114  * else.  Assume that the offsets have already been checked for block
115  * alignment.
116  *
117  * For clone we only link a partial EOF block above or at the destination file's
118  * EOF.  For deduplication we accept a partial EOF block only if it ends at the
119  * destination file's EOF (can not link it into the middle of a file).
120  *
121  * Shorten the request if possible.
122  */
generic_remap_check_len(struct inode * inode_in,struct inode * inode_out,loff_t pos_out,loff_t * len,unsigned int remap_flags)123 static int generic_remap_check_len(struct inode *inode_in,
124 				   struct inode *inode_out,
125 				   loff_t pos_out,
126 				   loff_t *len,
127 				   unsigned int remap_flags)
128 {
129 	u64 blkmask = i_blocksize(inode_in) - 1;
130 	loff_t new_len = *len;
131 
132 	if ((*len & blkmask) == 0)
133 		return 0;
134 
135 	if (pos_out + *len < i_size_read(inode_out))
136 		new_len &= ~blkmask;
137 
138 	if (new_len == *len)
139 		return 0;
140 
141 	if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
142 		*len = new_len;
143 		return 0;
144 	}
145 
146 	return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
147 }
148 
149 /* Read a page's worth of file data into the page cache. */
vfs_dedupe_get_page(struct inode * inode,loff_t offset)150 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
151 {
152 	struct page *page;
153 
154 	page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
155 	if (IS_ERR(page))
156 		return page;
157 	if (!PageUptodate(page)) {
158 		put_page(page);
159 		return ERR_PTR(-EIO);
160 	}
161 	return page;
162 }
163 
164 /*
165  * Lock two pages, ensuring that we lock in offset order if the pages are from
166  * the same file.
167  */
vfs_lock_two_pages(struct page * page1,struct page * page2)168 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
169 {
170 	/* Always lock in order of increasing index. */
171 	if (page1->index > page2->index)
172 		swap(page1, page2);
173 
174 	lock_page(page1);
175 	if (page1 != page2)
176 		lock_page(page2);
177 }
178 
179 /* Unlock two pages, being careful not to unlock the same page twice. */
vfs_unlock_two_pages(struct page * page1,struct page * page2)180 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
181 {
182 	unlock_page(page1);
183 	if (page1 != page2)
184 		unlock_page(page2);
185 }
186 
187 /*
188  * Compare extents of two files to see if they are the same.
189  * Caller must have locked both inodes to prevent write races.
190  */
vfs_dedupe_file_range_compare(struct inode * src,loff_t srcoff,struct inode * dest,loff_t destoff,loff_t len,bool * is_same)191 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
192 					 struct inode *dest, loff_t destoff,
193 					 loff_t len, bool *is_same)
194 {
195 	loff_t src_poff;
196 	loff_t dest_poff;
197 	void *src_addr;
198 	void *dest_addr;
199 	struct page *src_page;
200 	struct page *dest_page;
201 	loff_t cmp_len;
202 	bool same;
203 	int error;
204 
205 	error = -EINVAL;
206 	same = true;
207 	while (len) {
208 		src_poff = srcoff & (PAGE_SIZE - 1);
209 		dest_poff = destoff & (PAGE_SIZE - 1);
210 		cmp_len = min(PAGE_SIZE - src_poff,
211 			      PAGE_SIZE - dest_poff);
212 		cmp_len = min(cmp_len, len);
213 		if (cmp_len <= 0)
214 			goto out_error;
215 
216 		src_page = vfs_dedupe_get_page(src, srcoff);
217 		if (IS_ERR(src_page)) {
218 			error = PTR_ERR(src_page);
219 			goto out_error;
220 		}
221 		dest_page = vfs_dedupe_get_page(dest, destoff);
222 		if (IS_ERR(dest_page)) {
223 			error = PTR_ERR(dest_page);
224 			put_page(src_page);
225 			goto out_error;
226 		}
227 
228 		vfs_lock_two_pages(src_page, dest_page);
229 
230 		/*
231 		 * Now that we've locked both pages, make sure they're still
232 		 * mapped to the file data we're interested in.  If not,
233 		 * someone is invalidating pages on us and we lose.
234 		 */
235 		if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
236 		    src_page->mapping != src->i_mapping ||
237 		    dest_page->mapping != dest->i_mapping) {
238 			same = false;
239 			goto unlock;
240 		}
241 
242 		src_addr = kmap_atomic(src_page);
243 		dest_addr = kmap_atomic(dest_page);
244 
245 		flush_dcache_page(src_page);
246 		flush_dcache_page(dest_page);
247 
248 		if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
249 			same = false;
250 
251 		kunmap_atomic(dest_addr);
252 		kunmap_atomic(src_addr);
253 unlock:
254 		vfs_unlock_two_pages(src_page, dest_page);
255 		put_page(dest_page);
256 		put_page(src_page);
257 
258 		if (!same)
259 			break;
260 
261 		srcoff += cmp_len;
262 		destoff += cmp_len;
263 		len -= cmp_len;
264 	}
265 
266 	*is_same = same;
267 	return 0;
268 
269 out_error:
270 	return error;
271 }
272 
273 /*
274  * Check that the two inodes are eligible for cloning, the ranges make
275  * sense, and then flush all dirty data.  Caller must ensure that the
276  * inodes have been locked against any other modifications.
277  *
278  * If there's an error, then the usual negative error code is returned.
279  * Otherwise returns 0 with *len set to the request length.
280  */
generic_remap_file_range_prep(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t * len,unsigned int remap_flags)281 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
282 				  struct file *file_out, loff_t pos_out,
283 				  loff_t *len, unsigned int remap_flags)
284 {
285 	struct inode *inode_in = file_inode(file_in);
286 	struct inode *inode_out = file_inode(file_out);
287 	bool same_inode = (inode_in == inode_out);
288 	int ret;
289 
290 	/* Don't touch certain kinds of inodes */
291 	if (IS_IMMUTABLE(inode_out))
292 		return -EPERM;
293 
294 	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
295 		return -ETXTBSY;
296 
297 	/* Don't reflink dirs, pipes, sockets... */
298 	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
299 		return -EISDIR;
300 	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
301 		return -EINVAL;
302 
303 	/* Zero length dedupe exits immediately; reflink goes to EOF. */
304 	if (*len == 0) {
305 		loff_t isize = i_size_read(inode_in);
306 
307 		if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
308 			return 0;
309 		if (pos_in > isize)
310 			return -EINVAL;
311 		*len = isize - pos_in;
312 		if (*len == 0)
313 			return 0;
314 	}
315 
316 	/* Check that we don't violate system file offset limits. */
317 	ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
318 			remap_flags);
319 	if (ret)
320 		return ret;
321 
322 	/* Wait for the completion of any pending IOs on both files */
323 	inode_dio_wait(inode_in);
324 	if (!same_inode)
325 		inode_dio_wait(inode_out);
326 
327 	ret = filemap_write_and_wait_range(inode_in->i_mapping,
328 			pos_in, pos_in + *len - 1);
329 	if (ret)
330 		return ret;
331 
332 	ret = filemap_write_and_wait_range(inode_out->i_mapping,
333 			pos_out, pos_out + *len - 1);
334 	if (ret)
335 		return ret;
336 
337 	/*
338 	 * Check that the extents are the same.
339 	 */
340 	if (remap_flags & REMAP_FILE_DEDUP) {
341 		bool		is_same = false;
342 
343 		ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
344 				inode_out, pos_out, *len, &is_same);
345 		if (ret)
346 			return ret;
347 		if (!is_same)
348 			return -EBADE;
349 	}
350 
351 	ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
352 			remap_flags);
353 	if (ret)
354 		return ret;
355 
356 	/* If can't alter the file contents, we're done. */
357 	if (!(remap_flags & REMAP_FILE_DEDUP))
358 		ret = file_modified(file_out);
359 
360 	return ret;
361 }
362 EXPORT_SYMBOL(generic_remap_file_range_prep);
363 
do_clone_file_range(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t len,unsigned int remap_flags)364 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
365 			   struct file *file_out, loff_t pos_out,
366 			   loff_t len, unsigned int remap_flags)
367 {
368 	loff_t ret;
369 
370 	WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
371 
372 	/*
373 	 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
374 	 * the same mount. Practically, they only need to be on the same file
375 	 * system.
376 	 */
377 	if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
378 		return -EXDEV;
379 
380 	ret = generic_file_rw_checks(file_in, file_out);
381 	if (ret < 0)
382 		return ret;
383 
384 	if (!file_in->f_op->remap_file_range)
385 		return -EOPNOTSUPP;
386 
387 	ret = remap_verify_area(file_in, pos_in, len, false);
388 	if (ret)
389 		return ret;
390 
391 	ret = remap_verify_area(file_out, pos_out, len, true);
392 	if (ret)
393 		return ret;
394 
395 	ret = file_in->f_op->remap_file_range(file_in, pos_in,
396 			file_out, pos_out, len, remap_flags);
397 	if (ret < 0)
398 		return ret;
399 
400 	fsnotify_access(file_in);
401 	fsnotify_modify(file_out);
402 	return ret;
403 }
404 EXPORT_SYMBOL(do_clone_file_range);
405 
vfs_clone_file_range(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t len,unsigned int remap_flags)406 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
407 			    struct file *file_out, loff_t pos_out,
408 			    loff_t len, unsigned int remap_flags)
409 {
410 	loff_t ret;
411 
412 	file_start_write(file_out);
413 	ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
414 				  remap_flags);
415 	file_end_write(file_out);
416 
417 	return ret;
418 }
419 EXPORT_SYMBOL(vfs_clone_file_range);
420 
421 /* Check whether we are allowed to dedupe the destination file */
allow_file_dedupe(struct file * file)422 static bool allow_file_dedupe(struct file *file)
423 {
424 	struct user_namespace *mnt_userns = file_mnt_user_ns(file);
425 	struct inode *inode = file_inode(file);
426 
427 	if (capable(CAP_SYS_ADMIN))
428 		return true;
429 	if (file->f_mode & FMODE_WRITE)
430 		return true;
431 	if (uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
432 		return true;
433 	if (!inode_permission(mnt_userns, inode, MAY_WRITE))
434 		return true;
435 	return false;
436 }
437 
vfs_dedupe_file_range_one(struct file * src_file,loff_t src_pos,struct file * dst_file,loff_t dst_pos,loff_t len,unsigned int remap_flags)438 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
439 				 struct file *dst_file, loff_t dst_pos,
440 				 loff_t len, unsigned int remap_flags)
441 {
442 	loff_t ret;
443 
444 	WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
445 				     REMAP_FILE_CAN_SHORTEN));
446 
447 	ret = mnt_want_write_file(dst_file);
448 	if (ret)
449 		return ret;
450 
451 	/*
452 	 * This is redundant if called from vfs_dedupe_file_range(), but other
453 	 * callers need it and it's not performance sesitive...
454 	 */
455 	ret = remap_verify_area(src_file, src_pos, len, false);
456 	if (ret)
457 		goto out_drop_write;
458 
459 	ret = remap_verify_area(dst_file, dst_pos, len, true);
460 	if (ret)
461 		goto out_drop_write;
462 
463 	ret = -EPERM;
464 	if (!allow_file_dedupe(dst_file))
465 		goto out_drop_write;
466 
467 	ret = -EXDEV;
468 	if (src_file->f_path.mnt != dst_file->f_path.mnt)
469 		goto out_drop_write;
470 
471 	ret = -EISDIR;
472 	if (S_ISDIR(file_inode(dst_file)->i_mode))
473 		goto out_drop_write;
474 
475 	ret = -EINVAL;
476 	if (!dst_file->f_op->remap_file_range)
477 		goto out_drop_write;
478 
479 	if (len == 0) {
480 		ret = 0;
481 		goto out_drop_write;
482 	}
483 
484 	ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
485 			dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
486 out_drop_write:
487 	mnt_drop_write_file(dst_file);
488 
489 	return ret;
490 }
491 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
492 
vfs_dedupe_file_range(struct file * file,struct file_dedupe_range * same)493 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
494 {
495 	struct file_dedupe_range_info *info;
496 	struct inode *src = file_inode(file);
497 	u64 off;
498 	u64 len;
499 	int i;
500 	int ret;
501 	u16 count = same->dest_count;
502 	loff_t deduped;
503 
504 	if (!(file->f_mode & FMODE_READ))
505 		return -EINVAL;
506 
507 	if (same->reserved1 || same->reserved2)
508 		return -EINVAL;
509 
510 	off = same->src_offset;
511 	len = same->src_length;
512 
513 	if (S_ISDIR(src->i_mode))
514 		return -EISDIR;
515 
516 	if (!S_ISREG(src->i_mode))
517 		return -EINVAL;
518 
519 	if (!file->f_op->remap_file_range)
520 		return -EOPNOTSUPP;
521 
522 	ret = remap_verify_area(file, off, len, false);
523 	if (ret < 0)
524 		return ret;
525 	ret = 0;
526 
527 	if (off + len > i_size_read(src))
528 		return -EINVAL;
529 
530 	/* Arbitrary 1G limit on a single dedupe request, can be raised. */
531 	len = min_t(u64, len, 1 << 30);
532 
533 	/* pre-format output fields to sane values */
534 	for (i = 0; i < count; i++) {
535 		same->info[i].bytes_deduped = 0ULL;
536 		same->info[i].status = FILE_DEDUPE_RANGE_SAME;
537 	}
538 
539 	for (i = 0, info = same->info; i < count; i++, info++) {
540 		struct fd dst_fd = fdget(info->dest_fd);
541 		struct file *dst_file = dst_fd.file;
542 
543 		if (!dst_file) {
544 			info->status = -EBADF;
545 			goto next_loop;
546 		}
547 
548 		if (info->reserved) {
549 			info->status = -EINVAL;
550 			goto next_fdput;
551 		}
552 
553 		deduped = vfs_dedupe_file_range_one(file, off, dst_file,
554 						    info->dest_offset, len,
555 						    REMAP_FILE_CAN_SHORTEN);
556 		if (deduped == -EBADE)
557 			info->status = FILE_DEDUPE_RANGE_DIFFERS;
558 		else if (deduped < 0)
559 			info->status = deduped;
560 		else
561 			info->bytes_deduped = len;
562 
563 next_fdput:
564 		fdput(dst_fd);
565 next_loop:
566 		if (fatal_signal_pending(current))
567 			break;
568 	}
569 	return ret;
570 }
571 EXPORT_SYMBOL(vfs_dedupe_file_range);
572