1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2020-2024 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
5 */
6 #include "xfs.h"
7 #include "xfs_shared.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_defer.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_quota.h"
16 #include "xfs_bmap_util.h"
17 #include "xfs_reflink.h"
18 #include "xfs_trace.h"
19 #include "xfs_exchrange.h"
20 #include "xfs_exchmaps.h"
21 #include "xfs_sb.h"
22 #include "xfs_icache.h"
23 #include "xfs_log.h"
24 #include "xfs_rtbitmap.h"
25 #include <linux/fsnotify.h>
26
27 /* Lock (and optionally join) two inodes for a file range exchange. */
28 void
xfs_exchrange_ilock(struct xfs_trans * tp,struct xfs_inode * ip1,struct xfs_inode * ip2)29 xfs_exchrange_ilock(
30 struct xfs_trans *tp,
31 struct xfs_inode *ip1,
32 struct xfs_inode *ip2)
33 {
34 if (ip1 != ip2)
35 xfs_lock_two_inodes(ip1, XFS_ILOCK_EXCL,
36 ip2, XFS_ILOCK_EXCL);
37 else
38 xfs_ilock(ip1, XFS_ILOCK_EXCL);
39 if (tp) {
40 xfs_trans_ijoin(tp, ip1, 0);
41 if (ip2 != ip1)
42 xfs_trans_ijoin(tp, ip2, 0);
43 }
44
45 }
46
47 /* Unlock two inodes after a file range exchange operation. */
48 void
xfs_exchrange_iunlock(struct xfs_inode * ip1,struct xfs_inode * ip2)49 xfs_exchrange_iunlock(
50 struct xfs_inode *ip1,
51 struct xfs_inode *ip2)
52 {
53 if (ip2 != ip1)
54 xfs_iunlock(ip2, XFS_ILOCK_EXCL);
55 xfs_iunlock(ip1, XFS_ILOCK_EXCL);
56 }
57
58 /*
59 * Estimate the resource requirements to exchange file contents between the two
60 * files. The caller is required to hold the IOLOCK and the MMAPLOCK and to
61 * have flushed both inodes' pagecache and active direct-ios.
62 */
63 int
xfs_exchrange_estimate(struct xfs_exchmaps_req * req)64 xfs_exchrange_estimate(
65 struct xfs_exchmaps_req *req)
66 {
67 int error;
68
69 xfs_exchrange_ilock(NULL, req->ip1, req->ip2);
70 error = xfs_exchmaps_estimate(req);
71 xfs_exchrange_iunlock(req->ip1, req->ip2);
72 return error;
73 }
74
75 /*
76 * Check that file2's metadata agree with the snapshot that we took for the
77 * range commit request.
78 *
79 * This should be called after the filesystem has locked /all/ inode metadata
80 * against modification.
81 */
82 STATIC int
xfs_exchrange_check_freshness(const struct xfs_exchrange * fxr,struct xfs_inode * ip2)83 xfs_exchrange_check_freshness(
84 const struct xfs_exchrange *fxr,
85 struct xfs_inode *ip2)
86 {
87 struct inode *inode2 = VFS_I(ip2);
88 struct timespec64 ctime = inode_get_ctime(inode2);
89 struct timespec64 mtime = inode_get_mtime(inode2);
90
91 trace_xfs_exchrange_freshness(fxr, ip2);
92
93 /* Check that file2 hasn't otherwise been modified. */
94 if (fxr->file2_ino != ip2->i_ino ||
95 fxr->file2_gen != inode2->i_generation ||
96 !timespec64_equal(&fxr->file2_ctime, &ctime) ||
97 !timespec64_equal(&fxr->file2_mtime, &mtime))
98 return -EBUSY;
99
100 return 0;
101 }
102
103 #define QRETRY_IP1 (0x1)
104 #define QRETRY_IP2 (0x2)
105
106 /*
107 * Obtain a quota reservation to make sure we don't hit EDQUOT. We can skip
108 * this if quota enforcement is disabled or if both inodes' dquots are the
109 * same. The qretry structure must be initialized to zeroes before the first
110 * call to this function.
111 */
112 STATIC int
xfs_exchrange_reserve_quota(struct xfs_trans * tp,const struct xfs_exchmaps_req * req,unsigned int * qretry)113 xfs_exchrange_reserve_quota(
114 struct xfs_trans *tp,
115 const struct xfs_exchmaps_req *req,
116 unsigned int *qretry)
117 {
118 int64_t ddelta, rdelta;
119 int ip1_error = 0;
120 int error;
121
122 /*
123 * Don't bother with a quota reservation if we're not enforcing them
124 * or the two inodes have the same dquots.
125 */
126 if (!XFS_IS_QUOTA_ON(tp->t_mountp) || req->ip1 == req->ip2 ||
127 (req->ip1->i_udquot == req->ip2->i_udquot &&
128 req->ip1->i_gdquot == req->ip2->i_gdquot &&
129 req->ip1->i_pdquot == req->ip2->i_pdquot))
130 return 0;
131
132 *qretry = 0;
133
134 /*
135 * For each file, compute the net gain in the number of regular blocks
136 * that will be mapped into that file and reserve that much quota. The
137 * quota counts must be able to absorb at least that much space.
138 */
139 ddelta = req->ip2_bcount - req->ip1_bcount;
140 rdelta = req->ip2_rtbcount - req->ip1_rtbcount;
141 if (ddelta > 0 || rdelta > 0) {
142 error = xfs_trans_reserve_quota_nblks(tp, req->ip1,
143 ddelta > 0 ? ddelta : 0,
144 rdelta > 0 ? rdelta : 0,
145 false);
146 if (error == -EDQUOT || error == -ENOSPC) {
147 /*
148 * Save this error and see what happens if we try to
149 * reserve quota for ip2. Then report both.
150 */
151 *qretry |= QRETRY_IP1;
152 ip1_error = error;
153 error = 0;
154 }
155 if (error)
156 return error;
157 }
158 if (ddelta < 0 || rdelta < 0) {
159 error = xfs_trans_reserve_quota_nblks(tp, req->ip2,
160 ddelta < 0 ? -ddelta : 0,
161 rdelta < 0 ? -rdelta : 0,
162 false);
163 if (error == -EDQUOT || error == -ENOSPC)
164 *qretry |= QRETRY_IP2;
165 if (error)
166 return error;
167 }
168 if (ip1_error)
169 return ip1_error;
170
171 /*
172 * For each file, forcibly reserve the gross gain in mapped blocks so
173 * that we don't trip over any quota block reservation assertions.
174 * We must reserve the gross gain because the quota code subtracts from
175 * bcount the number of blocks that we unmap; it does not add that
176 * quantity back to the quota block reservation.
177 */
178 error = xfs_trans_reserve_quota_nblks(tp, req->ip1, req->ip1_bcount,
179 req->ip1_rtbcount, true);
180 if (error)
181 return error;
182
183 return xfs_trans_reserve_quota_nblks(tp, req->ip2, req->ip2_bcount,
184 req->ip2_rtbcount, true);
185 }
186
187 /* Exchange the mappings (and hence the contents) of two files' forks. */
188 STATIC int
xfs_exchrange_mappings(const struct xfs_exchrange * fxr,struct xfs_inode * ip1,struct xfs_inode * ip2)189 xfs_exchrange_mappings(
190 const struct xfs_exchrange *fxr,
191 struct xfs_inode *ip1,
192 struct xfs_inode *ip2)
193 {
194 struct xfs_mount *mp = ip1->i_mount;
195 struct xfs_exchmaps_req req = {
196 .ip1 = ip1,
197 .ip2 = ip2,
198 .startoff1 = XFS_B_TO_FSBT(mp, fxr->file1_offset),
199 .startoff2 = XFS_B_TO_FSBT(mp, fxr->file2_offset),
200 .blockcount = XFS_B_TO_FSB(mp, fxr->length),
201 };
202 struct xfs_trans *tp;
203 unsigned int qretry;
204 bool retried = false;
205 int error;
206
207 trace_xfs_exchrange_mappings(fxr, ip1, ip2);
208
209 if (fxr->flags & XFS_EXCHANGE_RANGE_TO_EOF)
210 req.flags |= XFS_EXCHMAPS_SET_SIZES;
211 if (fxr->flags & XFS_EXCHANGE_RANGE_FILE1_WRITTEN)
212 req.flags |= XFS_EXCHMAPS_INO1_WRITTEN;
213
214 /*
215 * Round the request length up to the nearest file allocation unit.
216 * The prep function already checked that the request offsets and
217 * length in @fxr are safe to round up.
218 */
219 if (xfs_inode_has_bigrtalloc(ip2))
220 req.blockcount = xfs_rtb_roundup_rtx(mp, req.blockcount);
221
222 error = xfs_exchrange_estimate(&req);
223 if (error)
224 return error;
225
226 retry:
227 /* Allocate the transaction, lock the inodes, and join them. */
228 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, req.resblks, 0,
229 XFS_TRANS_RES_FDBLKS, &tp);
230 if (error)
231 return error;
232
233 xfs_exchrange_ilock(tp, ip1, ip2);
234
235 trace_xfs_exchrange_before(ip2, 2);
236 trace_xfs_exchrange_before(ip1, 1);
237
238 error = xfs_exchmaps_check_forks(mp, &req);
239 if (error)
240 goto out_trans_cancel;
241
242 /*
243 * Reserve ourselves some quota if any of them are in enforcing mode.
244 * In theory we only need enough to satisfy the change in the number
245 * of blocks between the two ranges being remapped.
246 */
247 error = xfs_exchrange_reserve_quota(tp, &req, &qretry);
248 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
249 xfs_trans_cancel(tp);
250 xfs_exchrange_iunlock(ip1, ip2);
251 if (qretry & QRETRY_IP1)
252 xfs_blockgc_free_quota(ip1, 0);
253 if (qretry & QRETRY_IP2)
254 xfs_blockgc_free_quota(ip2, 0);
255 retried = true;
256 goto retry;
257 }
258 if (error)
259 goto out_trans_cancel;
260
261 /* If we got this far on a dry run, all parameters are ok. */
262 if (fxr->flags & XFS_EXCHANGE_RANGE_DRY_RUN)
263 goto out_trans_cancel;
264
265 /* Update the mtime and ctime of both files. */
266 if (fxr->flags & __XFS_EXCHANGE_RANGE_UPD_CMTIME1)
267 xfs_trans_ichgtime(tp, ip1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
268 if (fxr->flags & __XFS_EXCHANGE_RANGE_UPD_CMTIME2)
269 xfs_trans_ichgtime(tp, ip2, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
270
271 xfs_exchange_mappings(tp, &req);
272
273 /*
274 * Force the log to persist metadata updates if the caller or the
275 * administrator requires this. The generic prep function already
276 * flushed the relevant parts of the page cache.
277 */
278 if (xfs_has_wsync(mp) || (fxr->flags & XFS_EXCHANGE_RANGE_DSYNC))
279 xfs_trans_set_sync(tp);
280
281 error = xfs_trans_commit(tp);
282
283 trace_xfs_exchrange_after(ip2, 2);
284 trace_xfs_exchrange_after(ip1, 1);
285
286 if (error)
287 goto out_unlock;
288
289 /*
290 * If the caller wanted us to exchange the contents of two complete
291 * files of unequal length, exchange the incore sizes now. This should
292 * be safe because we flushed both files' page caches, exchanged all
293 * the mappings, and updated the ondisk sizes.
294 */
295 if (fxr->flags & XFS_EXCHANGE_RANGE_TO_EOF) {
296 loff_t temp;
297
298 temp = i_size_read(VFS_I(ip2));
299 i_size_write(VFS_I(ip2), i_size_read(VFS_I(ip1)));
300 i_size_write(VFS_I(ip1), temp);
301 }
302
303 out_unlock:
304 xfs_exchrange_iunlock(ip1, ip2);
305 return error;
306
307 out_trans_cancel:
308 xfs_trans_cancel(tp);
309 goto out_unlock;
310 }
311
312 /*
313 * Generic code for exchanging ranges of two files via XFS_IOC_EXCHANGE_RANGE.
314 * This part deals with struct file objects and byte ranges and does not deal
315 * with XFS-specific data structures such as xfs_inodes and block ranges. This
316 * separation may some day facilitate porting to another filesystem.
317 *
318 * The goal is to exchange fxr.length bytes starting at fxr.file1_offset in
319 * file1 with the same number of bytes starting at fxr.file2_offset in file2.
320 * Implementations must call xfs_exchange_range_prep to prepare the two
321 * files prior to taking locks; and they must update the inode change and mod
322 * times of both files as part of the metadata update. The timestamp update
323 * and freshness checks must be done atomically as part of the data exchange
324 * operation to ensure correctness of the freshness check.
325 * xfs_exchange_range_finish must be called after the operation completes
326 * successfully but before locks are dropped.
327 */
328
329 /*
330 * Performs necessary checks before doing a range exchange, having stabilized
331 * mutable inode attributes via i_rwsem.
332 */
333 static inline int
xfs_exchange_range_checks(struct xfs_exchrange * fxr,unsigned int alloc_unit)334 xfs_exchange_range_checks(
335 struct xfs_exchrange *fxr,
336 unsigned int alloc_unit)
337 {
338 struct inode *inode1 = file_inode(fxr->file1);
339 loff_t size1 = i_size_read(inode1);
340 struct inode *inode2 = file_inode(fxr->file2);
341 loff_t size2 = i_size_read(inode2);
342 uint64_t allocmask = alloc_unit - 1;
343 int64_t test_len;
344 uint64_t blen;
345 loff_t tmp;
346 int error;
347
348 /* Don't touch certain kinds of inodes */
349 if (IS_IMMUTABLE(inode1) || IS_IMMUTABLE(inode2))
350 return -EPERM;
351 if (IS_SWAPFILE(inode1) || IS_SWAPFILE(inode2))
352 return -ETXTBSY;
353
354 /* Ranges cannot start after EOF. */
355 if (fxr->file1_offset > size1 || fxr->file2_offset > size2)
356 return -EINVAL;
357
358 if (fxr->flags & XFS_EXCHANGE_RANGE_TO_EOF) {
359 /*
360 * If the caller said to exchange to EOF, we set the length of
361 * the request large enough to cover everything to the end of
362 * both files.
363 */
364 fxr->length = max_t(int64_t, size1 - fxr->file1_offset,
365 size2 - fxr->file2_offset);
366 } else {
367 /*
368 * Otherwise we require both ranges to end within EOF.
369 */
370 if (fxr->file1_offset + fxr->length > size1 ||
371 fxr->file2_offset + fxr->length > size2)
372 return -EINVAL;
373 }
374
375 /*
376 * The start of both ranges must be aligned to the file allocation
377 * unit.
378 */
379 if (!IS_ALIGNED(fxr->file1_offset, alloc_unit) ||
380 !IS_ALIGNED(fxr->file2_offset, alloc_unit))
381 return -EINVAL;
382
383 /* Ensure offsets don't wrap. */
384 if (check_add_overflow(fxr->file1_offset, fxr->length, &tmp) ||
385 check_add_overflow(fxr->file2_offset, fxr->length, &tmp))
386 return -EINVAL;
387
388 /*
389 * Make sure we don't hit any file size limits. If we hit any size
390 * limits such that test_length was adjusted, we abort the whole
391 * operation.
392 */
393 test_len = fxr->length;
394 error = generic_write_check_limits(fxr->file2, fxr->file2_offset,
395 &test_len);
396 if (error)
397 return error;
398 error = generic_write_check_limits(fxr->file1, fxr->file1_offset,
399 &test_len);
400 if (error)
401 return error;
402 if (test_len != fxr->length)
403 return -EINVAL;
404
405 /*
406 * If the user wanted us to exchange up to the infile's EOF, round up
407 * to the next allocation unit boundary for this check. Do the same
408 * for the outfile.
409 *
410 * Otherwise, reject the range length if it's not aligned to an
411 * allocation unit.
412 */
413 if (fxr->file1_offset + fxr->length == size1)
414 blen = ALIGN(size1, alloc_unit) - fxr->file1_offset;
415 else if (fxr->file2_offset + fxr->length == size2)
416 blen = ALIGN(size2, alloc_unit) - fxr->file2_offset;
417 else if (!IS_ALIGNED(fxr->length, alloc_unit))
418 return -EINVAL;
419 else
420 blen = fxr->length;
421
422 /* Don't allow overlapped exchanges within the same file. */
423 if (inode1 == inode2 &&
424 fxr->file2_offset + blen > fxr->file1_offset &&
425 fxr->file1_offset + blen > fxr->file2_offset)
426 return -EINVAL;
427
428 /*
429 * Ensure that we don't exchange a partial EOF block into the middle of
430 * another file.
431 */
432 if ((fxr->length & allocmask) == 0)
433 return 0;
434
435 blen = fxr->length;
436 if (fxr->file2_offset + blen < size2)
437 blen &= ~allocmask;
438
439 if (fxr->file1_offset + blen < size1)
440 blen &= ~allocmask;
441
442 return blen == fxr->length ? 0 : -EINVAL;
443 }
444
445 /*
446 * Check that the two inodes are eligible for range exchanges, the ranges make
447 * sense, and then flush all dirty data. Caller must ensure that the inodes
448 * have been locked against any other modifications.
449 */
450 static inline int
xfs_exchange_range_prep(struct xfs_exchrange * fxr,unsigned int alloc_unit)451 xfs_exchange_range_prep(
452 struct xfs_exchrange *fxr,
453 unsigned int alloc_unit)
454 {
455 struct inode *inode1 = file_inode(fxr->file1);
456 struct inode *inode2 = file_inode(fxr->file2);
457 bool same_inode = (inode1 == inode2);
458 int error;
459
460 /* Check that we don't violate system file offset limits. */
461 error = xfs_exchange_range_checks(fxr, alloc_unit);
462 if (error || fxr->length == 0)
463 return error;
464
465 /* Wait for the completion of any pending IOs on both files */
466 inode_dio_wait(inode1);
467 if (!same_inode)
468 inode_dio_wait(inode2);
469
470 error = filemap_write_and_wait_range(inode1->i_mapping,
471 fxr->file1_offset,
472 fxr->file1_offset + fxr->length - 1);
473 if (error)
474 return error;
475
476 error = filemap_write_and_wait_range(inode2->i_mapping,
477 fxr->file2_offset,
478 fxr->file2_offset + fxr->length - 1);
479 if (error)
480 return error;
481
482 /*
483 * If the files or inodes involved require synchronous writes, amend
484 * the request to force the filesystem to flush all data and metadata
485 * to disk after the operation completes.
486 */
487 if (((fxr->file1->f_flags | fxr->file2->f_flags) & O_SYNC) ||
488 IS_SYNC(inode1) || IS_SYNC(inode2))
489 fxr->flags |= XFS_EXCHANGE_RANGE_DSYNC;
490
491 return 0;
492 }
493
494 /*
495 * Finish a range exchange operation, if it was successful. Caller must ensure
496 * that the inodes are still locked against any other modifications.
497 */
498 static inline int
xfs_exchange_range_finish(struct xfs_exchrange * fxr)499 xfs_exchange_range_finish(
500 struct xfs_exchrange *fxr)
501 {
502 int error;
503
504 error = file_remove_privs(fxr->file1);
505 if (error)
506 return error;
507 if (file_inode(fxr->file1) == file_inode(fxr->file2))
508 return 0;
509
510 return file_remove_privs(fxr->file2);
511 }
512
513 /*
514 * Check the alignment of an exchange request when the allocation unit size
515 * isn't a power of two. The generic file-level helpers use (fast)
516 * bitmask-based alignment checks, but here we have to use slow long division.
517 */
518 static int
xfs_exchrange_check_rtalign(const struct xfs_exchrange * fxr,struct xfs_inode * ip1,struct xfs_inode * ip2,unsigned int alloc_unit)519 xfs_exchrange_check_rtalign(
520 const struct xfs_exchrange *fxr,
521 struct xfs_inode *ip1,
522 struct xfs_inode *ip2,
523 unsigned int alloc_unit)
524 {
525 uint64_t length = fxr->length;
526 uint64_t blen;
527 loff_t size1, size2;
528
529 size1 = i_size_read(VFS_I(ip1));
530 size2 = i_size_read(VFS_I(ip2));
531
532 /* The start of both ranges must be aligned to a rt extent. */
533 if (!isaligned_64(fxr->file1_offset, alloc_unit) ||
534 !isaligned_64(fxr->file2_offset, alloc_unit))
535 return -EINVAL;
536
537 if (fxr->flags & XFS_EXCHANGE_RANGE_TO_EOF)
538 length = max_t(int64_t, size1 - fxr->file1_offset,
539 size2 - fxr->file2_offset);
540
541 /*
542 * If the user wanted us to exchange up to the infile's EOF, round up
543 * to the next rt extent boundary for this check. Do the same for the
544 * outfile.
545 *
546 * Otherwise, reject the range length if it's not rt extent aligned.
547 * We already confirmed the starting offsets' rt extent block
548 * alignment.
549 */
550 if (fxr->file1_offset + length == size1)
551 blen = roundup_64(size1, alloc_unit) - fxr->file1_offset;
552 else if (fxr->file2_offset + length == size2)
553 blen = roundup_64(size2, alloc_unit) - fxr->file2_offset;
554 else if (!isaligned_64(length, alloc_unit))
555 return -EINVAL;
556 else
557 blen = length;
558
559 /* Don't allow overlapped exchanges within the same file. */
560 if (ip1 == ip2 &&
561 fxr->file2_offset + blen > fxr->file1_offset &&
562 fxr->file1_offset + blen > fxr->file2_offset)
563 return -EINVAL;
564
565 /*
566 * Ensure that we don't exchange a partial EOF rt extent into the
567 * middle of another file.
568 */
569 if (isaligned_64(length, alloc_unit))
570 return 0;
571
572 blen = length;
573 if (fxr->file2_offset + length < size2)
574 blen = rounddown_64(blen, alloc_unit);
575
576 if (fxr->file1_offset + blen < size1)
577 blen = rounddown_64(blen, alloc_unit);
578
579 return blen == length ? 0 : -EINVAL;
580 }
581
582 /* Prepare two files to have their data exchanged. */
583 STATIC int
xfs_exchrange_prep(struct xfs_exchrange * fxr,struct xfs_inode * ip1,struct xfs_inode * ip2)584 xfs_exchrange_prep(
585 struct xfs_exchrange *fxr,
586 struct xfs_inode *ip1,
587 struct xfs_inode *ip2)
588 {
589 struct xfs_mount *mp = ip2->i_mount;
590 unsigned int alloc_unit = xfs_inode_alloc_unitsize(ip2);
591 int error;
592
593 trace_xfs_exchrange_prep(fxr, ip1, ip2);
594
595 /* Verify both files are either real-time or non-realtime */
596 if (XFS_IS_REALTIME_INODE(ip1) != XFS_IS_REALTIME_INODE(ip2))
597 return -EINVAL;
598
599 /* Check non-power of two alignment issues, if necessary. */
600 if (!is_power_of_2(alloc_unit)) {
601 error = xfs_exchrange_check_rtalign(fxr, ip1, ip2, alloc_unit);
602 if (error)
603 return error;
604
605 /*
606 * Do the generic file-level checks with the regular block
607 * alignment.
608 */
609 alloc_unit = mp->m_sb.sb_blocksize;
610 }
611
612 error = xfs_exchange_range_prep(fxr, alloc_unit);
613 if (error || fxr->length == 0)
614 return error;
615
616 if (fxr->flags & __XFS_EXCHANGE_RANGE_CHECK_FRESH2) {
617 error = xfs_exchrange_check_freshness(fxr, ip2);
618 if (error)
619 return error;
620 }
621
622 /* Attach dquots to both inodes before changing block maps. */
623 error = xfs_qm_dqattach(ip2);
624 if (error)
625 return error;
626 error = xfs_qm_dqattach(ip1);
627 if (error)
628 return error;
629
630 trace_xfs_exchrange_flush(fxr, ip1, ip2);
631
632 /* Flush the relevant ranges of both files. */
633 error = xfs_flush_unmap_range(ip2, fxr->file2_offset, fxr->length);
634 if (error)
635 return error;
636 error = xfs_flush_unmap_range(ip1, fxr->file1_offset, fxr->length);
637 if (error)
638 return error;
639
640 /*
641 * Cancel CoW fork preallocations for the ranges of both files. The
642 * prep function should have flushed all the dirty data, so the only
643 * CoW mappings remaining should be speculative.
644 */
645 if (xfs_inode_has_cow_data(ip1)) {
646 error = xfs_reflink_cancel_cow_range(ip1, fxr->file1_offset,
647 fxr->length, true);
648 if (error)
649 return error;
650 }
651
652 if (xfs_inode_has_cow_data(ip2)) {
653 error = xfs_reflink_cancel_cow_range(ip2, fxr->file2_offset,
654 fxr->length, true);
655 if (error)
656 return error;
657 }
658
659 return 0;
660 }
661
662 /*
663 * Exchange contents of files. This is the binding between the generic
664 * file-level concepts and the XFS inode-specific implementation.
665 */
666 STATIC int
xfs_exchrange_contents(struct xfs_exchrange * fxr)667 xfs_exchrange_contents(
668 struct xfs_exchrange *fxr)
669 {
670 struct inode *inode1 = file_inode(fxr->file1);
671 struct inode *inode2 = file_inode(fxr->file2);
672 struct xfs_inode *ip1 = XFS_I(inode1);
673 struct xfs_inode *ip2 = XFS_I(inode2);
674 struct xfs_mount *mp = ip1->i_mount;
675 int error;
676
677 if (!xfs_has_exchange_range(mp))
678 return -EOPNOTSUPP;
679
680 if (fxr->flags & ~(XFS_EXCHANGE_RANGE_ALL_FLAGS |
681 XFS_EXCHANGE_RANGE_PRIV_FLAGS))
682 return -EINVAL;
683
684 if (xfs_is_shutdown(mp))
685 return -EIO;
686
687 /* Lock both files against IO */
688 error = xfs_ilock2_io_mmap(ip1, ip2);
689 if (error)
690 goto out_err;
691
692 /* Prepare and then exchange file contents. */
693 error = xfs_exchrange_prep(fxr, ip1, ip2);
694 if (error)
695 goto out_unlock;
696
697 error = xfs_exchrange_mappings(fxr, ip1, ip2);
698 if (error)
699 goto out_unlock;
700
701 /*
702 * Finish the exchange by removing special file privileges like any
703 * other file write would do. This may involve turning on support for
704 * logged xattrs if either file has security capabilities.
705 */
706 error = xfs_exchange_range_finish(fxr);
707 if (error)
708 goto out_unlock;
709
710 out_unlock:
711 xfs_iunlock2_io_mmap(ip1, ip2);
712 out_err:
713 if (error)
714 trace_xfs_exchrange_error(ip2, error, _RET_IP_);
715 return error;
716 }
717
718 /* Exchange parts of two files. */
719 static int
xfs_exchange_range(struct xfs_exchrange * fxr)720 xfs_exchange_range(
721 struct xfs_exchrange *fxr)
722 {
723 struct inode *inode1 = file_inode(fxr->file1);
724 struct inode *inode2 = file_inode(fxr->file2);
725 loff_t check_len = fxr->length;
726 int ret;
727
728 BUILD_BUG_ON(XFS_EXCHANGE_RANGE_ALL_FLAGS &
729 XFS_EXCHANGE_RANGE_PRIV_FLAGS);
730
731 /* Both files must be on the same mount/filesystem. */
732 if (fxr->file1->f_path.mnt != fxr->file2->f_path.mnt)
733 return -EXDEV;
734
735 if (fxr->flags & ~(XFS_EXCHANGE_RANGE_ALL_FLAGS |
736 __XFS_EXCHANGE_RANGE_CHECK_FRESH2))
737 return -EINVAL;
738
739 /* Userspace requests only honored for regular files. */
740 if (S_ISDIR(inode1->i_mode) || S_ISDIR(inode2->i_mode))
741 return -EISDIR;
742 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
743 return -EINVAL;
744
745 /* Both files must be opened for read and write. */
746 if (!(fxr->file1->f_mode & FMODE_READ) ||
747 !(fxr->file1->f_mode & FMODE_WRITE) ||
748 !(fxr->file2->f_mode & FMODE_READ) ||
749 !(fxr->file2->f_mode & FMODE_WRITE))
750 return -EBADF;
751
752 /* Neither file can be opened append-only. */
753 if ((fxr->file1->f_flags & O_APPEND) ||
754 (fxr->file2->f_flags & O_APPEND))
755 return -EBADF;
756
757 /*
758 * If we're exchanging to EOF we can't calculate the length until taking
759 * the iolock. Pass a 0 length to remap_verify_area similar to the
760 * FICLONE and FICLONERANGE ioctls that support cloning to EOF as well.
761 */
762 if (fxr->flags & XFS_EXCHANGE_RANGE_TO_EOF)
763 check_len = 0;
764 ret = remap_verify_area(fxr->file1, fxr->file1_offset, check_len, true);
765 if (ret)
766 return ret;
767 ret = remap_verify_area(fxr->file2, fxr->file2_offset, check_len, true);
768 if (ret)
769 return ret;
770
771 /* Update cmtime if the fd/inode don't forbid it. */
772 if (!(fxr->file1->f_mode & FMODE_NOCMTIME) && !IS_NOCMTIME(inode1))
773 fxr->flags |= __XFS_EXCHANGE_RANGE_UPD_CMTIME1;
774 if (!(fxr->file2->f_mode & FMODE_NOCMTIME) && !IS_NOCMTIME(inode2))
775 fxr->flags |= __XFS_EXCHANGE_RANGE_UPD_CMTIME2;
776
777 file_start_write(fxr->file2);
778 ret = xfs_exchrange_contents(fxr);
779 file_end_write(fxr->file2);
780 if (ret)
781 return ret;
782
783 fsnotify_modify(fxr->file1);
784 if (fxr->file2 != fxr->file1)
785 fsnotify_modify(fxr->file2);
786 return 0;
787 }
788
789 /* Collect exchange-range arguments from userspace. */
790 long
xfs_ioc_exchange_range(struct file * file,struct xfs_exchange_range __user * argp)791 xfs_ioc_exchange_range(
792 struct file *file,
793 struct xfs_exchange_range __user *argp)
794 {
795 struct xfs_exchrange fxr = {
796 .file2 = file,
797 };
798 struct xfs_exchange_range args;
799 struct fd file1;
800 int error;
801
802 if (copy_from_user(&args, argp, sizeof(args)))
803 return -EFAULT;
804 if (memchr_inv(&args.pad, 0, sizeof(args.pad)))
805 return -EINVAL;
806 if (args.flags & ~XFS_EXCHANGE_RANGE_ALL_FLAGS)
807 return -EINVAL;
808
809 fxr.file1_offset = args.file1_offset;
810 fxr.file2_offset = args.file2_offset;
811 fxr.length = args.length;
812 fxr.flags = args.flags;
813
814 file1 = fdget(args.file1_fd);
815 if (!fd_file(file1))
816 return -EBADF;
817 fxr.file1 = fd_file(file1);
818
819 error = xfs_exchange_range(&fxr);
820 fdput(file1);
821 return error;
822 }
823
824 /* Opaque freshness blob for XFS_IOC_COMMIT_RANGE */
825 struct xfs_commit_range_fresh {
826 xfs_fsid_t fsid; /* m_fixedfsid */
827 __u64 file2_ino; /* inode number */
828 __s64 file2_mtime; /* modification time */
829 __s64 file2_ctime; /* change time */
830 __s32 file2_mtime_nsec; /* mod time, nsec */
831 __s32 file2_ctime_nsec; /* change time, nsec */
832 __u32 file2_gen; /* inode generation */
833 __u32 magic; /* zero */
834 };
835 #define XCR_FRESH_MAGIC 0x444F524B /* DORK */
836
837 /* Set up a commitrange operation by sampling file2's write-related attrs */
838 long
xfs_ioc_start_commit(struct file * file,struct xfs_commit_range __user * argp)839 xfs_ioc_start_commit(
840 struct file *file,
841 struct xfs_commit_range __user *argp)
842 {
843 struct xfs_commit_range args = { };
844 struct timespec64 ts;
845 struct xfs_commit_range_fresh *kern_f;
846 struct xfs_commit_range_fresh __user *user_f;
847 struct inode *inode2 = file_inode(file);
848 struct xfs_inode *ip2 = XFS_I(inode2);
849 const unsigned int lockflags = XFS_IOLOCK_SHARED |
850 XFS_MMAPLOCK_SHARED |
851 XFS_ILOCK_SHARED;
852
853 BUILD_BUG_ON(sizeof(struct xfs_commit_range_fresh) !=
854 sizeof(args.file2_freshness));
855
856 kern_f = (struct xfs_commit_range_fresh *)&args.file2_freshness;
857
858 memcpy(&kern_f->fsid, ip2->i_mount->m_fixedfsid, sizeof(xfs_fsid_t));
859
860 xfs_ilock(ip2, lockflags);
861 ts = inode_get_ctime(inode2);
862 kern_f->file2_ctime = ts.tv_sec;
863 kern_f->file2_ctime_nsec = ts.tv_nsec;
864 ts = inode_get_mtime(inode2);
865 kern_f->file2_mtime = ts.tv_sec;
866 kern_f->file2_mtime_nsec = ts.tv_nsec;
867 kern_f->file2_ino = ip2->i_ino;
868 kern_f->file2_gen = inode2->i_generation;
869 kern_f->magic = XCR_FRESH_MAGIC;
870 xfs_iunlock(ip2, lockflags);
871
872 user_f = (struct xfs_commit_range_fresh __user *)&argp->file2_freshness;
873 if (copy_to_user(user_f, kern_f, sizeof(*kern_f)))
874 return -EFAULT;
875
876 return 0;
877 }
878
879 /*
880 * Exchange file1 and file2 contents if file2 has not been written since the
881 * start commit operation.
882 */
883 long
xfs_ioc_commit_range(struct file * file,struct xfs_commit_range __user * argp)884 xfs_ioc_commit_range(
885 struct file *file,
886 struct xfs_commit_range __user *argp)
887 {
888 struct xfs_exchrange fxr = {
889 .file2 = file,
890 };
891 struct xfs_commit_range args;
892 struct xfs_commit_range_fresh *kern_f;
893 struct xfs_inode *ip2 = XFS_I(file_inode(file));
894 struct xfs_mount *mp = ip2->i_mount;
895 struct fd file1;
896 int error;
897
898 kern_f = (struct xfs_commit_range_fresh *)&args.file2_freshness;
899
900 if (copy_from_user(&args, argp, sizeof(args)))
901 return -EFAULT;
902 if (args.flags & ~XFS_EXCHANGE_RANGE_ALL_FLAGS)
903 return -EINVAL;
904 if (kern_f->magic != XCR_FRESH_MAGIC)
905 return -EBUSY;
906 if (memcmp(&kern_f->fsid, mp->m_fixedfsid, sizeof(xfs_fsid_t)))
907 return -EBUSY;
908
909 fxr.file1_offset = args.file1_offset;
910 fxr.file2_offset = args.file2_offset;
911 fxr.length = args.length;
912 fxr.flags = args.flags | __XFS_EXCHANGE_RANGE_CHECK_FRESH2;
913 fxr.file2_ino = kern_f->file2_ino;
914 fxr.file2_gen = kern_f->file2_gen;
915 fxr.file2_mtime.tv_sec = kern_f->file2_mtime;
916 fxr.file2_mtime.tv_nsec = kern_f->file2_mtime_nsec;
917 fxr.file2_ctime.tv_sec = kern_f->file2_ctime;
918 fxr.file2_ctime.tv_nsec = kern_f->file2_ctime_nsec;
919
920 file1 = fdget(args.file1_fd);
921 if (fd_empty(file1))
922 return -EBADF;
923 fxr.file1 = fd_file(file1);
924
925 error = xfs_exchange_range(&fxr);
926 fdput(file1);
927 return error;
928 }
929