1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2012 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_trans.h"
19 #include "xfs_alloc.h"
20 #include "xfs_bmap.h"
21 #include "xfs_bmap_util.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_rtalloc.h"
24 #include "xfs_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_trans_space.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_iomap.h"
30 #include "xfs_reflink.h"
31
32 /* Kernel only BMAP related definitions and functions */
33
34 /*
35 * Convert the given file system block to a disk block. We have to treat it
36 * differently based on whether the file is a real time file or not, because the
37 * bmap code does.
38 */
39 xfs_daddr_t
xfs_fsb_to_db(struct xfs_inode * ip,xfs_fsblock_t fsb)40 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
41 {
42 if (XFS_IS_REALTIME_INODE(ip))
43 return XFS_FSB_TO_BB(ip->i_mount, fsb);
44 return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
45 }
46
47 /*
48 * Routine to zero an extent on disk allocated to the specific inode.
49 *
50 * The VFS functions take a linearised filesystem block offset, so we have to
51 * convert the sparse xfs fsb to the right format first.
52 * VFS types are real funky, too.
53 */
54 int
xfs_zero_extent(struct xfs_inode * ip,xfs_fsblock_t start_fsb,xfs_off_t count_fsb)55 xfs_zero_extent(
56 struct xfs_inode *ip,
57 xfs_fsblock_t start_fsb,
58 xfs_off_t count_fsb)
59 {
60 struct xfs_mount *mp = ip->i_mount;
61 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
62 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
63 sector_t block = XFS_BB_TO_FSBT(mp, sector);
64
65 return blkdev_issue_zeroout(target->bt_bdev,
66 block << (mp->m_super->s_blocksize_bits - 9),
67 count_fsb << (mp->m_super->s_blocksize_bits - 9),
68 GFP_NOFS, 0);
69 }
70
71 #ifdef CONFIG_XFS_RT
72 int
xfs_bmap_rtalloc(struct xfs_bmalloca * ap)73 xfs_bmap_rtalloc(
74 struct xfs_bmalloca *ap)
75 {
76 struct xfs_mount *mp = ap->ip->i_mount;
77 xfs_fileoff_t orig_offset = ap->offset;
78 xfs_rtblock_t rtb;
79 xfs_extlen_t prod = 0; /* product factor for allocators */
80 xfs_extlen_t mod = 0; /* product factor for allocators */
81 xfs_extlen_t ralen = 0; /* realtime allocation length */
82 xfs_extlen_t align; /* minimum allocation alignment */
83 xfs_extlen_t orig_length = ap->length;
84 xfs_extlen_t minlen = mp->m_sb.sb_rextsize;
85 xfs_extlen_t raminlen;
86 bool rtlocked = false;
87 bool ignore_locality = false;
88 int error;
89
90 align = xfs_get_extsz_hint(ap->ip);
91 retry:
92 prod = align / mp->m_sb.sb_rextsize;
93 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
94 align, 1, ap->eof, 0,
95 ap->conv, &ap->offset, &ap->length);
96 if (error)
97 return error;
98 ASSERT(ap->length);
99 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
100
101 /*
102 * If we shifted the file offset downward to satisfy an extent size
103 * hint, increase minlen by that amount so that the allocator won't
104 * give us an allocation that's too short to cover at least one of the
105 * blocks that the caller asked for.
106 */
107 if (ap->offset != orig_offset)
108 minlen += orig_offset - ap->offset;
109
110 /*
111 * If the offset & length are not perfectly aligned
112 * then kill prod, it will just get us in trouble.
113 */
114 div_u64_rem(ap->offset, align, &mod);
115 if (mod || ap->length % align)
116 prod = 1;
117 /*
118 * Set ralen to be the actual requested length in rtextents.
119 */
120 ralen = ap->length / mp->m_sb.sb_rextsize;
121 /*
122 * If the old value was close enough to XFS_BMBT_MAX_EXTLEN that
123 * we rounded up to it, cut it back so it's valid again.
124 * Note that if it's a really large request (bigger than
125 * XFS_BMBT_MAX_EXTLEN), we don't hear about that number, and can't
126 * adjust the starting point to match it.
127 */
128 if (ralen * mp->m_sb.sb_rextsize >= XFS_MAX_BMBT_EXTLEN)
129 ralen = XFS_MAX_BMBT_EXTLEN / mp->m_sb.sb_rextsize;
130
131 /*
132 * Lock out modifications to both the RT bitmap and summary inodes
133 */
134 if (!rtlocked) {
135 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
136 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
137 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
138 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
139 rtlocked = true;
140 }
141
142 /*
143 * If it's an allocation to an empty file at offset 0,
144 * pick an extent that will space things out in the rt area.
145 */
146 if (ap->eof && ap->offset == 0) {
147 xfs_rtblock_t rtx; /* realtime extent no */
148
149 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
150 if (error)
151 return error;
152 ap->blkno = rtx * mp->m_sb.sb_rextsize;
153 } else {
154 ap->blkno = 0;
155 }
156
157 xfs_bmap_adjacent(ap);
158
159 /*
160 * Realtime allocation, done through xfs_rtallocate_extent.
161 */
162 if (ignore_locality)
163 ap->blkno = 0;
164 else
165 do_div(ap->blkno, mp->m_sb.sb_rextsize);
166 rtb = ap->blkno;
167 ap->length = ralen;
168 raminlen = max_t(xfs_extlen_t, 1, minlen / mp->m_sb.sb_rextsize);
169 error = xfs_rtallocate_extent(ap->tp, ap->blkno, raminlen, ap->length,
170 &ralen, ap->wasdel, prod, &rtb);
171 if (error)
172 return error;
173
174 if (rtb != NULLRTBLOCK) {
175 ap->blkno = rtb * mp->m_sb.sb_rextsize;
176 ap->length = ralen * mp->m_sb.sb_rextsize;
177 ap->ip->i_nblocks += ap->length;
178 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
179 if (ap->wasdel)
180 ap->ip->i_delayed_blks -= ap->length;
181 /*
182 * Adjust the disk quota also. This was reserved
183 * earlier.
184 */
185 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
186 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
187 XFS_TRANS_DQ_RTBCOUNT, ap->length);
188 return 0;
189 }
190
191 if (align > mp->m_sb.sb_rextsize) {
192 /*
193 * We previously enlarged the request length to try to satisfy
194 * an extent size hint. The allocator didn't return anything,
195 * so reset the parameters to the original values and try again
196 * without alignment criteria.
197 */
198 ap->offset = orig_offset;
199 ap->length = orig_length;
200 minlen = align = mp->m_sb.sb_rextsize;
201 goto retry;
202 }
203
204 if (!ignore_locality && ap->blkno != 0) {
205 /*
206 * If we can't allocate near a specific rt extent, try again
207 * without locality criteria.
208 */
209 ignore_locality = true;
210 goto retry;
211 }
212
213 ap->blkno = NULLFSBLOCK;
214 ap->length = 0;
215 return 0;
216 }
217 #endif /* CONFIG_XFS_RT */
218
219 /*
220 * Extent tree block counting routines.
221 */
222
223 /*
224 * Count leaf blocks given a range of extent records. Delayed allocation
225 * extents are not counted towards the totals.
226 */
227 xfs_extnum_t
xfs_bmap_count_leaves(struct xfs_ifork * ifp,xfs_filblks_t * count)228 xfs_bmap_count_leaves(
229 struct xfs_ifork *ifp,
230 xfs_filblks_t *count)
231 {
232 struct xfs_iext_cursor icur;
233 struct xfs_bmbt_irec got;
234 xfs_extnum_t numrecs = 0;
235
236 for_each_xfs_iext(ifp, &icur, &got) {
237 if (!isnullstartblock(got.br_startblock)) {
238 *count += got.br_blockcount;
239 numrecs++;
240 }
241 }
242
243 return numrecs;
244 }
245
246 /*
247 * Count fsblocks of the given fork. Delayed allocation extents are
248 * not counted towards the totals.
249 */
250 int
xfs_bmap_count_blocks(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,xfs_extnum_t * nextents,xfs_filblks_t * count)251 xfs_bmap_count_blocks(
252 struct xfs_trans *tp,
253 struct xfs_inode *ip,
254 int whichfork,
255 xfs_extnum_t *nextents,
256 xfs_filblks_t *count)
257 {
258 struct xfs_mount *mp = ip->i_mount;
259 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
260 struct xfs_btree_cur *cur;
261 xfs_extlen_t btblocks = 0;
262 int error;
263
264 *nextents = 0;
265 *count = 0;
266
267 if (!ifp)
268 return 0;
269
270 switch (ifp->if_format) {
271 case XFS_DINODE_FMT_BTREE:
272 error = xfs_iread_extents(tp, ip, whichfork);
273 if (error)
274 return error;
275
276 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
277 error = xfs_btree_count_blocks(cur, &btblocks);
278 xfs_btree_del_cursor(cur, error);
279 if (error)
280 return error;
281
282 /*
283 * xfs_btree_count_blocks includes the root block contained in
284 * the inode fork in @btblocks, so subtract one because we're
285 * only interested in allocated disk blocks.
286 */
287 *count += btblocks - 1;
288
289 fallthrough;
290 case XFS_DINODE_FMT_EXTENTS:
291 *nextents = xfs_bmap_count_leaves(ifp, count);
292 break;
293 }
294
295 return 0;
296 }
297
298 static int
xfs_getbmap_report_one(struct xfs_inode * ip,struct getbmapx * bmv,struct kgetbmap * out,int64_t bmv_end,struct xfs_bmbt_irec * got)299 xfs_getbmap_report_one(
300 struct xfs_inode *ip,
301 struct getbmapx *bmv,
302 struct kgetbmap *out,
303 int64_t bmv_end,
304 struct xfs_bmbt_irec *got)
305 {
306 struct kgetbmap *p = out + bmv->bmv_entries;
307 bool shared = false;
308 int error;
309
310 error = xfs_reflink_trim_around_shared(ip, got, &shared);
311 if (error)
312 return error;
313
314 if (isnullstartblock(got->br_startblock) ||
315 got->br_startblock == DELAYSTARTBLOCK) {
316 /*
317 * Delalloc extents that start beyond EOF can occur due to
318 * speculative EOF allocation when the delalloc extent is larger
319 * than the largest freespace extent at conversion time. These
320 * extents cannot be converted by data writeback, so can exist
321 * here even if we are not supposed to be finding delalloc
322 * extents.
323 */
324 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
325 ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
326
327 p->bmv_oflags |= BMV_OF_DELALLOC;
328 p->bmv_block = -2;
329 } else {
330 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
331 }
332
333 if (got->br_state == XFS_EXT_UNWRITTEN &&
334 (bmv->bmv_iflags & BMV_IF_PREALLOC))
335 p->bmv_oflags |= BMV_OF_PREALLOC;
336
337 if (shared)
338 p->bmv_oflags |= BMV_OF_SHARED;
339
340 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
341 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
342
343 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
344 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
345 bmv->bmv_entries++;
346 return 0;
347 }
348
349 static void
xfs_getbmap_report_hole(struct xfs_inode * ip,struct getbmapx * bmv,struct kgetbmap * out,int64_t bmv_end,xfs_fileoff_t bno,xfs_fileoff_t end)350 xfs_getbmap_report_hole(
351 struct xfs_inode *ip,
352 struct getbmapx *bmv,
353 struct kgetbmap *out,
354 int64_t bmv_end,
355 xfs_fileoff_t bno,
356 xfs_fileoff_t end)
357 {
358 struct kgetbmap *p = out + bmv->bmv_entries;
359
360 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
361 return;
362
363 p->bmv_block = -1;
364 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
365 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
366
367 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
368 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
369 bmv->bmv_entries++;
370 }
371
372 static inline bool
xfs_getbmap_full(struct getbmapx * bmv)373 xfs_getbmap_full(
374 struct getbmapx *bmv)
375 {
376 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
377 }
378
379 static bool
xfs_getbmap_next_rec(struct xfs_bmbt_irec * rec,xfs_fileoff_t total_end)380 xfs_getbmap_next_rec(
381 struct xfs_bmbt_irec *rec,
382 xfs_fileoff_t total_end)
383 {
384 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
385
386 if (end == total_end)
387 return false;
388
389 rec->br_startoff += rec->br_blockcount;
390 if (!isnullstartblock(rec->br_startblock) &&
391 rec->br_startblock != DELAYSTARTBLOCK)
392 rec->br_startblock += rec->br_blockcount;
393 rec->br_blockcount = total_end - end;
394 return true;
395 }
396
397 /*
398 * Get inode's extents as described in bmv, and format for output.
399 * Calls formatter to fill the user's buffer until all extents
400 * are mapped, until the passed-in bmv->bmv_count slots have
401 * been filled, or until the formatter short-circuits the loop,
402 * if it is tracking filled-in extents on its own.
403 */
404 int /* error code */
xfs_getbmap(struct xfs_inode * ip,struct getbmapx * bmv,struct kgetbmap * out)405 xfs_getbmap(
406 struct xfs_inode *ip,
407 struct getbmapx *bmv, /* user bmap structure */
408 struct kgetbmap *out)
409 {
410 struct xfs_mount *mp = ip->i_mount;
411 int iflags = bmv->bmv_iflags;
412 int whichfork, lock, error = 0;
413 int64_t bmv_end, max_len;
414 xfs_fileoff_t bno, first_bno;
415 struct xfs_ifork *ifp;
416 struct xfs_bmbt_irec got, rec;
417 xfs_filblks_t len;
418 struct xfs_iext_cursor icur;
419
420 if (bmv->bmv_iflags & ~BMV_IF_VALID)
421 return -EINVAL;
422 #ifndef DEBUG
423 /* Only allow CoW fork queries if we're debugging. */
424 if (iflags & BMV_IF_COWFORK)
425 return -EINVAL;
426 #endif
427 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
428 return -EINVAL;
429
430 if (bmv->bmv_length < -1)
431 return -EINVAL;
432 bmv->bmv_entries = 0;
433 if (bmv->bmv_length == 0)
434 return 0;
435
436 if (iflags & BMV_IF_ATTRFORK)
437 whichfork = XFS_ATTR_FORK;
438 else if (iflags & BMV_IF_COWFORK)
439 whichfork = XFS_COW_FORK;
440 else
441 whichfork = XFS_DATA_FORK;
442 ifp = XFS_IFORK_PTR(ip, whichfork);
443
444 xfs_ilock(ip, XFS_IOLOCK_SHARED);
445 switch (whichfork) {
446 case XFS_ATTR_FORK:
447 if (!XFS_IFORK_Q(ip))
448 goto out_unlock_iolock;
449
450 max_len = 1LL << 32;
451 lock = xfs_ilock_attr_map_shared(ip);
452 break;
453 case XFS_COW_FORK:
454 /* No CoW fork? Just return */
455 if (!ifp)
456 goto out_unlock_iolock;
457
458 if (xfs_get_cowextsz_hint(ip))
459 max_len = mp->m_super->s_maxbytes;
460 else
461 max_len = XFS_ISIZE(ip);
462
463 lock = XFS_ILOCK_SHARED;
464 xfs_ilock(ip, lock);
465 break;
466 case XFS_DATA_FORK:
467 if (!(iflags & BMV_IF_DELALLOC) &&
468 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) {
469 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
470 if (error)
471 goto out_unlock_iolock;
472
473 /*
474 * Even after flushing the inode, there can still be
475 * delalloc blocks on the inode beyond EOF due to
476 * speculative preallocation. These are not removed
477 * until the release function is called or the inode
478 * is inactivated. Hence we cannot assert here that
479 * ip->i_delayed_blks == 0.
480 */
481 }
482
483 if (xfs_get_extsz_hint(ip) ||
484 (ip->i_diflags &
485 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
486 max_len = mp->m_super->s_maxbytes;
487 else
488 max_len = XFS_ISIZE(ip);
489
490 lock = xfs_ilock_data_map_shared(ip);
491 break;
492 }
493
494 switch (ifp->if_format) {
495 case XFS_DINODE_FMT_EXTENTS:
496 case XFS_DINODE_FMT_BTREE:
497 break;
498 case XFS_DINODE_FMT_LOCAL:
499 /* Local format inode forks report no extents. */
500 goto out_unlock_ilock;
501 default:
502 error = -EINVAL;
503 goto out_unlock_ilock;
504 }
505
506 if (bmv->bmv_length == -1) {
507 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
508 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
509 }
510
511 bmv_end = bmv->bmv_offset + bmv->bmv_length;
512
513 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
514 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
515
516 error = xfs_iread_extents(NULL, ip, whichfork);
517 if (error)
518 goto out_unlock_ilock;
519
520 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
521 /*
522 * Report a whole-file hole if the delalloc flag is set to
523 * stay compatible with the old implementation.
524 */
525 if (iflags & BMV_IF_DELALLOC)
526 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
527 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
528 goto out_unlock_ilock;
529 }
530
531 while (!xfs_getbmap_full(bmv)) {
532 xfs_trim_extent(&got, first_bno, len);
533
534 /*
535 * Report an entry for a hole if this extent doesn't directly
536 * follow the previous one.
537 */
538 if (got.br_startoff > bno) {
539 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
540 got.br_startoff);
541 if (xfs_getbmap_full(bmv))
542 break;
543 }
544
545 /*
546 * In order to report shared extents accurately, we report each
547 * distinct shared / unshared part of a single bmbt record with
548 * an individual getbmapx record.
549 */
550 bno = got.br_startoff + got.br_blockcount;
551 rec = got;
552 do {
553 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
554 &rec);
555 if (error || xfs_getbmap_full(bmv))
556 goto out_unlock_ilock;
557 } while (xfs_getbmap_next_rec(&rec, bno));
558
559 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
560 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
561
562 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
563
564 if (whichfork != XFS_ATTR_FORK && bno < end &&
565 !xfs_getbmap_full(bmv)) {
566 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
567 bno, end);
568 }
569 break;
570 }
571
572 if (bno >= first_bno + len)
573 break;
574 }
575
576 out_unlock_ilock:
577 xfs_iunlock(ip, lock);
578 out_unlock_iolock:
579 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
580 return error;
581 }
582
583 /*
584 * Dead simple method of punching delalyed allocation blocks from a range in
585 * the inode. This will always punch out both the start and end blocks, even
586 * if the ranges only partially overlap them, so it is up to the caller to
587 * ensure that partial blocks are not passed in.
588 */
589 int
xfs_bmap_punch_delalloc_range(struct xfs_inode * ip,xfs_fileoff_t start_fsb,xfs_fileoff_t length)590 xfs_bmap_punch_delalloc_range(
591 struct xfs_inode *ip,
592 xfs_fileoff_t start_fsb,
593 xfs_fileoff_t length)
594 {
595 struct xfs_ifork *ifp = &ip->i_df;
596 xfs_fileoff_t end_fsb = start_fsb + length;
597 struct xfs_bmbt_irec got, del;
598 struct xfs_iext_cursor icur;
599 int error = 0;
600
601 ASSERT(!xfs_need_iread_extents(ifp));
602
603 xfs_ilock(ip, XFS_ILOCK_EXCL);
604 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
605 goto out_unlock;
606
607 while (got.br_startoff + got.br_blockcount > start_fsb) {
608 del = got;
609 xfs_trim_extent(&del, start_fsb, length);
610
611 /*
612 * A delete can push the cursor forward. Step back to the
613 * previous extent on non-delalloc or extents outside the
614 * target range.
615 */
616 if (!del.br_blockcount ||
617 !isnullstartblock(del.br_startblock)) {
618 if (!xfs_iext_prev_extent(ifp, &icur, &got))
619 break;
620 continue;
621 }
622
623 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
624 &got, &del);
625 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
626 break;
627 }
628
629 out_unlock:
630 xfs_iunlock(ip, XFS_ILOCK_EXCL);
631 return error;
632 }
633
634 /*
635 * Test whether it is appropriate to check an inode for and free post EOF
636 * blocks. The 'force' parameter determines whether we should also consider
637 * regular files that are marked preallocated or append-only.
638 */
639 bool
xfs_can_free_eofblocks(struct xfs_inode * ip,bool force)640 xfs_can_free_eofblocks(
641 struct xfs_inode *ip,
642 bool force)
643 {
644 struct xfs_bmbt_irec imap;
645 struct xfs_mount *mp = ip->i_mount;
646 xfs_fileoff_t end_fsb;
647 xfs_fileoff_t last_fsb;
648 int nimaps = 1;
649 int error;
650
651 /*
652 * Caller must either hold the exclusive io lock; or be inactivating
653 * the inode, which guarantees there are no other users of the inode.
654 */
655 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL) ||
656 (VFS_I(ip)->i_state & I_FREEING));
657
658 /* prealloc/delalloc exists only on regular files */
659 if (!S_ISREG(VFS_I(ip)->i_mode))
660 return false;
661
662 /*
663 * Zero sized files with no cached pages and delalloc blocks will not
664 * have speculative prealloc/delalloc blocks to remove.
665 */
666 if (VFS_I(ip)->i_size == 0 &&
667 VFS_I(ip)->i_mapping->nrpages == 0 &&
668 ip->i_delayed_blks == 0)
669 return false;
670
671 /* If we haven't read in the extent list, then don't do it now. */
672 if (xfs_need_iread_extents(&ip->i_df))
673 return false;
674
675 /*
676 * Do not free real preallocated or append-only files unless the file
677 * has delalloc blocks and we are forced to remove them.
678 */
679 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
680 if (!force || ip->i_delayed_blks == 0)
681 return false;
682
683 /*
684 * Do not try to free post-EOF blocks if EOF is beyond the end of the
685 * range supported by the page cache, because the truncation will loop
686 * forever.
687 */
688 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
689 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1)
690 end_fsb = roundup_64(end_fsb, mp->m_sb.sb_rextsize);
691 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
692 if (last_fsb <= end_fsb)
693 return false;
694
695 /*
696 * Look up the mapping for the first block past EOF. If we can't find
697 * it, there's nothing to free.
698 */
699 xfs_ilock(ip, XFS_ILOCK_SHARED);
700 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps,
701 0);
702 xfs_iunlock(ip, XFS_ILOCK_SHARED);
703 if (error || nimaps == 0)
704 return false;
705
706 /*
707 * If there's a real mapping there or there are delayed allocation
708 * reservations, then we have post-EOF blocks to try to free.
709 */
710 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks;
711 }
712
713 /*
714 * This is called to free any blocks beyond eof. The caller must hold
715 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
716 * reference to the inode.
717 */
718 int
xfs_free_eofblocks(struct xfs_inode * ip)719 xfs_free_eofblocks(
720 struct xfs_inode *ip)
721 {
722 struct xfs_trans *tp;
723 struct xfs_mount *mp = ip->i_mount;
724 int error;
725
726 /* Attach the dquots to the inode up front. */
727 error = xfs_qm_dqattach(ip);
728 if (error)
729 return error;
730
731 /* Wait on dio to ensure i_size has settled. */
732 inode_dio_wait(VFS_I(ip));
733
734 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
735 if (error) {
736 ASSERT(xfs_is_shutdown(mp));
737 return error;
738 }
739
740 xfs_ilock(ip, XFS_ILOCK_EXCL);
741 xfs_trans_ijoin(tp, ip, 0);
742
743 /*
744 * Do not update the on-disk file size. If we update the on-disk file
745 * size and then the system crashes before the contents of the file are
746 * flushed to disk then the files may be full of holes (ie NULL files
747 * bug).
748 */
749 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
750 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
751 if (error)
752 goto err_cancel;
753
754 error = xfs_trans_commit(tp);
755 if (error)
756 goto out_unlock;
757
758 xfs_inode_clear_eofblocks_tag(ip);
759 goto out_unlock;
760
761 err_cancel:
762 /*
763 * If we get an error at this point we simply don't
764 * bother truncating the file.
765 */
766 xfs_trans_cancel(tp);
767 out_unlock:
768 xfs_iunlock(ip, XFS_ILOCK_EXCL);
769 return error;
770 }
771
772 int
xfs_alloc_file_space(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)773 xfs_alloc_file_space(
774 struct xfs_inode *ip,
775 xfs_off_t offset,
776 xfs_off_t len)
777 {
778 xfs_mount_t *mp = ip->i_mount;
779 xfs_off_t count;
780 xfs_filblks_t allocated_fsb;
781 xfs_filblks_t allocatesize_fsb;
782 xfs_extlen_t extsz, temp;
783 xfs_fileoff_t startoffset_fsb;
784 xfs_fileoff_t endoffset_fsb;
785 int nimaps;
786 int rt;
787 xfs_trans_t *tp;
788 xfs_bmbt_irec_t imaps[1], *imapp;
789 int error;
790
791 trace_xfs_alloc_file_space(ip);
792
793 if (xfs_is_shutdown(mp))
794 return -EIO;
795
796 error = xfs_qm_dqattach(ip);
797 if (error)
798 return error;
799
800 if (len <= 0)
801 return -EINVAL;
802
803 rt = XFS_IS_REALTIME_INODE(ip);
804 extsz = xfs_get_extsz_hint(ip);
805
806 count = len;
807 imapp = &imaps[0];
808 nimaps = 1;
809 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
810 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
811 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
812
813 /*
814 * Allocate file space until done or until there is an error
815 */
816 while (allocatesize_fsb && !error) {
817 xfs_fileoff_t s, e;
818 unsigned int dblocks, rblocks, resblks;
819
820 /*
821 * Determine space reservations for data/realtime.
822 */
823 if (unlikely(extsz)) {
824 s = startoffset_fsb;
825 do_div(s, extsz);
826 s *= extsz;
827 e = startoffset_fsb + allocatesize_fsb;
828 div_u64_rem(startoffset_fsb, extsz, &temp);
829 if (temp)
830 e += temp;
831 div_u64_rem(e, extsz, &temp);
832 if (temp)
833 e += extsz - temp;
834 } else {
835 s = 0;
836 e = allocatesize_fsb;
837 }
838
839 /*
840 * The transaction reservation is limited to a 32-bit block
841 * count, hence we need to limit the number of blocks we are
842 * trying to reserve to avoid an overflow. We can't allocate
843 * more than @nimaps extents, and an extent is limited on disk
844 * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the
845 * limit.
846 */
847 resblks = min_t(xfs_fileoff_t, (e - s),
848 (XFS_MAX_BMBT_EXTLEN * nimaps));
849 if (unlikely(rt)) {
850 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
851 rblocks = resblks;
852 } else {
853 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
854 rblocks = 0;
855 }
856
857 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
858 dblocks, rblocks, false, &tp);
859 if (error)
860 break;
861
862 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
863 XFS_IEXT_ADD_NOSPLIT_CNT);
864 if (error == -EFBIG)
865 error = xfs_iext_count_upgrade(tp, ip,
866 XFS_IEXT_ADD_NOSPLIT_CNT);
867 if (error)
868 goto error;
869
870 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
871 allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp,
872 &nimaps);
873 if (error)
874 goto error;
875
876 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
877 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
878
879 error = xfs_trans_commit(tp);
880 xfs_iunlock(ip, XFS_ILOCK_EXCL);
881 if (error)
882 break;
883
884 allocated_fsb = imapp->br_blockcount;
885
886 if (nimaps == 0) {
887 error = -ENOSPC;
888 break;
889 }
890
891 startoffset_fsb += allocated_fsb;
892 allocatesize_fsb -= allocated_fsb;
893 }
894
895 return error;
896
897 error:
898 xfs_trans_cancel(tp);
899 xfs_iunlock(ip, XFS_ILOCK_EXCL);
900 return error;
901 }
902
903 static int
xfs_unmap_extent(struct xfs_inode * ip,xfs_fileoff_t startoffset_fsb,xfs_filblks_t len_fsb,int * done)904 xfs_unmap_extent(
905 struct xfs_inode *ip,
906 xfs_fileoff_t startoffset_fsb,
907 xfs_filblks_t len_fsb,
908 int *done)
909 {
910 struct xfs_mount *mp = ip->i_mount;
911 struct xfs_trans *tp;
912 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
913 int error;
914
915 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
916 false, &tp);
917 if (error)
918 return error;
919
920 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
921 XFS_IEXT_PUNCH_HOLE_CNT);
922 if (error == -EFBIG)
923 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
924 if (error)
925 goto out_trans_cancel;
926
927 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
928 if (error)
929 goto out_trans_cancel;
930
931 error = xfs_trans_commit(tp);
932 out_unlock:
933 xfs_iunlock(ip, XFS_ILOCK_EXCL);
934 return error;
935
936 out_trans_cancel:
937 xfs_trans_cancel(tp);
938 goto out_unlock;
939 }
940
941 /* Caller must first wait for the completion of any pending DIOs if required. */
942 int
xfs_flush_unmap_range(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)943 xfs_flush_unmap_range(
944 struct xfs_inode *ip,
945 xfs_off_t offset,
946 xfs_off_t len)
947 {
948 struct xfs_mount *mp = ip->i_mount;
949 struct inode *inode = VFS_I(ip);
950 xfs_off_t rounding, start, end;
951 int error;
952
953 rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE);
954 start = round_down(offset, rounding);
955 end = round_up(offset + len, rounding) - 1;
956
957 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
958 if (error)
959 return error;
960 truncate_pagecache_range(inode, start, end);
961 return 0;
962 }
963
964 int
xfs_free_file_space(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)965 xfs_free_file_space(
966 struct xfs_inode *ip,
967 xfs_off_t offset,
968 xfs_off_t len)
969 {
970 struct xfs_mount *mp = ip->i_mount;
971 xfs_fileoff_t startoffset_fsb;
972 xfs_fileoff_t endoffset_fsb;
973 int done = 0, error;
974
975 trace_xfs_free_file_space(ip);
976
977 error = xfs_qm_dqattach(ip);
978 if (error)
979 return error;
980
981 if (len <= 0) /* if nothing being freed */
982 return 0;
983
984 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
985 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
986
987 /* We can only free complete realtime extents. */
988 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
989 startoffset_fsb = roundup_64(startoffset_fsb,
990 mp->m_sb.sb_rextsize);
991 endoffset_fsb = rounddown_64(endoffset_fsb,
992 mp->m_sb.sb_rextsize);
993 }
994
995 /*
996 * Need to zero the stuff we're not freeing, on disk.
997 */
998 if (endoffset_fsb > startoffset_fsb) {
999 while (!done) {
1000 error = xfs_unmap_extent(ip, startoffset_fsb,
1001 endoffset_fsb - startoffset_fsb, &done);
1002 if (error)
1003 return error;
1004 }
1005 }
1006
1007 /*
1008 * Now that we've unmap all full blocks we'll have to zero out any
1009 * partial block at the beginning and/or end. xfs_zero_range is smart
1010 * enough to skip any holes, including those we just created, but we
1011 * must take care not to zero beyond EOF and enlarge i_size.
1012 */
1013 if (offset >= XFS_ISIZE(ip))
1014 return 0;
1015 if (offset + len > XFS_ISIZE(ip))
1016 len = XFS_ISIZE(ip) - offset;
1017 error = xfs_zero_range(ip, offset, len, NULL);
1018 if (error)
1019 return error;
1020
1021 /*
1022 * If we zeroed right up to EOF and EOF straddles a page boundary we
1023 * must make sure that the post-EOF area is also zeroed because the
1024 * page could be mmap'd and xfs_zero_range doesn't do that for us.
1025 * Writeback of the eof page will do this, albeit clumsily.
1026 */
1027 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
1028 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1029 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
1030 }
1031
1032 return error;
1033 }
1034
1035 static int
xfs_prepare_shift(struct xfs_inode * ip,loff_t offset)1036 xfs_prepare_shift(
1037 struct xfs_inode *ip,
1038 loff_t offset)
1039 {
1040 struct xfs_mount *mp = ip->i_mount;
1041 int error;
1042
1043 /*
1044 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1045 * into the accessible region of the file.
1046 */
1047 if (xfs_can_free_eofblocks(ip, true)) {
1048 error = xfs_free_eofblocks(ip);
1049 if (error)
1050 return error;
1051 }
1052
1053 /*
1054 * Shift operations must stabilize the start block offset boundary along
1055 * with the full range of the operation. If we don't, a COW writeback
1056 * completion could race with an insert, front merge with the start
1057 * extent (after split) during the shift and corrupt the file. Start
1058 * with the block just prior to the start to stabilize the boundary.
1059 */
1060 offset = round_down(offset, mp->m_sb.sb_blocksize);
1061 if (offset)
1062 offset -= mp->m_sb.sb_blocksize;
1063
1064 /*
1065 * Writeback and invalidate cache for the remainder of the file as we're
1066 * about to shift down every extent from offset to EOF.
1067 */
1068 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1069 if (error)
1070 return error;
1071
1072 /*
1073 * Clean out anything hanging around in the cow fork now that
1074 * we've flushed all the dirty data out to disk to avoid having
1075 * CoW extents at the wrong offsets.
1076 */
1077 if (xfs_inode_has_cow_data(ip)) {
1078 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1079 true);
1080 if (error)
1081 return error;
1082 }
1083
1084 return 0;
1085 }
1086
1087 /*
1088 * xfs_collapse_file_space()
1089 * This routine frees disk space and shift extent for the given file.
1090 * The first thing we do is to free data blocks in the specified range
1091 * by calling xfs_free_file_space(). It would also sync dirty data
1092 * and invalidate page cache over the region on which collapse range
1093 * is working. And Shift extent records to the left to cover a hole.
1094 * RETURNS:
1095 * 0 on success
1096 * errno on error
1097 *
1098 */
1099 int
xfs_collapse_file_space(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)1100 xfs_collapse_file_space(
1101 struct xfs_inode *ip,
1102 xfs_off_t offset,
1103 xfs_off_t len)
1104 {
1105 struct xfs_mount *mp = ip->i_mount;
1106 struct xfs_trans *tp;
1107 int error;
1108 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
1109 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1110 bool done = false;
1111
1112 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1113 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1114
1115 trace_xfs_collapse_file_space(ip);
1116
1117 error = xfs_free_file_space(ip, offset, len);
1118 if (error)
1119 return error;
1120
1121 error = xfs_prepare_shift(ip, offset);
1122 if (error)
1123 return error;
1124
1125 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1126 if (error)
1127 return error;
1128
1129 xfs_ilock(ip, XFS_ILOCK_EXCL);
1130 xfs_trans_ijoin(tp, ip, 0);
1131
1132 while (!done) {
1133 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1134 &done);
1135 if (error)
1136 goto out_trans_cancel;
1137 if (done)
1138 break;
1139
1140 /* finish any deferred frees and roll the transaction */
1141 error = xfs_defer_finish(&tp);
1142 if (error)
1143 goto out_trans_cancel;
1144 }
1145
1146 error = xfs_trans_commit(tp);
1147 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1148 return error;
1149
1150 out_trans_cancel:
1151 xfs_trans_cancel(tp);
1152 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1153 return error;
1154 }
1155
1156 /*
1157 * xfs_insert_file_space()
1158 * This routine create hole space by shifting extents for the given file.
1159 * The first thing we do is to sync dirty data and invalidate page cache
1160 * over the region on which insert range is working. And split an extent
1161 * to two extents at given offset by calling xfs_bmap_split_extent.
1162 * And shift all extent records which are laying between [offset,
1163 * last allocated extent] to the right to reserve hole range.
1164 * RETURNS:
1165 * 0 on success
1166 * errno on error
1167 */
1168 int
xfs_insert_file_space(struct xfs_inode * ip,loff_t offset,loff_t len)1169 xfs_insert_file_space(
1170 struct xfs_inode *ip,
1171 loff_t offset,
1172 loff_t len)
1173 {
1174 struct xfs_mount *mp = ip->i_mount;
1175 struct xfs_trans *tp;
1176 int error;
1177 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1178 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1179 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1180 bool done = false;
1181
1182 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1183 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1184
1185 trace_xfs_insert_file_space(ip);
1186
1187 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1188 if (error)
1189 return error;
1190
1191 error = xfs_prepare_shift(ip, offset);
1192 if (error)
1193 return error;
1194
1195 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1196 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1197 if (error)
1198 return error;
1199
1200 xfs_ilock(ip, XFS_ILOCK_EXCL);
1201 xfs_trans_ijoin(tp, ip, 0);
1202
1203 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
1204 XFS_IEXT_PUNCH_HOLE_CNT);
1205 if (error == -EFBIG)
1206 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
1207 if (error)
1208 goto out_trans_cancel;
1209
1210 /*
1211 * The extent shifting code works on extent granularity. So, if stop_fsb
1212 * is not the starting block of extent, we need to split the extent at
1213 * stop_fsb.
1214 */
1215 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1216 if (error)
1217 goto out_trans_cancel;
1218
1219 do {
1220 error = xfs_defer_finish(&tp);
1221 if (error)
1222 goto out_trans_cancel;
1223
1224 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1225 &done, stop_fsb);
1226 if (error)
1227 goto out_trans_cancel;
1228 } while (!done);
1229
1230 error = xfs_trans_commit(tp);
1231 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1232 return error;
1233
1234 out_trans_cancel:
1235 xfs_trans_cancel(tp);
1236 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1237 return error;
1238 }
1239
1240 /*
1241 * We need to check that the format of the data fork in the temporary inode is
1242 * valid for the target inode before doing the swap. This is not a problem with
1243 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1244 * data fork depending on the space the attribute fork is taking so we can get
1245 * invalid formats on the target inode.
1246 *
1247 * E.g. target has space for 7 extents in extent format, temp inode only has
1248 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1249 * btree, but when swapped it needs to be in extent format. Hence we can't just
1250 * blindly swap data forks on attr2 filesystems.
1251 *
1252 * Note that we check the swap in both directions so that we don't end up with
1253 * a corrupt temporary inode, either.
1254 *
1255 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1256 * inode will prevent this situation from occurring, so all we do here is
1257 * reject and log the attempt. basically we are putting the responsibility on
1258 * userspace to get this right.
1259 */
1260 static int
xfs_swap_extents_check_format(struct xfs_inode * ip,struct xfs_inode * tip)1261 xfs_swap_extents_check_format(
1262 struct xfs_inode *ip, /* target inode */
1263 struct xfs_inode *tip) /* tmp inode */
1264 {
1265 struct xfs_ifork *ifp = &ip->i_df;
1266 struct xfs_ifork *tifp = &tip->i_df;
1267
1268 /* User/group/project quota ids must match if quotas are enforced. */
1269 if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1270 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1271 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1272 ip->i_projid != tip->i_projid))
1273 return -EINVAL;
1274
1275 /* Should never get a local format */
1276 if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1277 tifp->if_format == XFS_DINODE_FMT_LOCAL)
1278 return -EINVAL;
1279
1280 /*
1281 * if the target inode has less extents that then temporary inode then
1282 * why did userspace call us?
1283 */
1284 if (ifp->if_nextents < tifp->if_nextents)
1285 return -EINVAL;
1286
1287 /*
1288 * If we have to use the (expensive) rmap swap method, we can
1289 * handle any number of extents and any format.
1290 */
1291 if (xfs_has_rmapbt(ip->i_mount))
1292 return 0;
1293
1294 /*
1295 * if the target inode is in extent form and the temp inode is in btree
1296 * form then we will end up with the target inode in the wrong format
1297 * as we already know there are less extents in the temp inode.
1298 */
1299 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1300 tifp->if_format == XFS_DINODE_FMT_BTREE)
1301 return -EINVAL;
1302
1303 /* Check temp in extent form to max in target */
1304 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1305 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1306 return -EINVAL;
1307
1308 /* Check target in extent form to max in temp */
1309 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1310 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1311 return -EINVAL;
1312
1313 /*
1314 * If we are in a btree format, check that the temp root block will fit
1315 * in the target and that it has enough extents to be in btree format
1316 * in the target.
1317 *
1318 * Note that we have to be careful to allow btree->extent conversions
1319 * (a common defrag case) which will occur when the temp inode is in
1320 * extent format...
1321 */
1322 if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1323 if (XFS_IFORK_Q(ip) &&
1324 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > XFS_IFORK_BOFF(ip))
1325 return -EINVAL;
1326 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1327 return -EINVAL;
1328 }
1329
1330 /* Reciprocal target->temp btree format checks */
1331 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1332 if (XFS_IFORK_Q(tip) &&
1333 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1334 return -EINVAL;
1335 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1336 return -EINVAL;
1337 }
1338
1339 return 0;
1340 }
1341
1342 static int
xfs_swap_extent_flush(struct xfs_inode * ip)1343 xfs_swap_extent_flush(
1344 struct xfs_inode *ip)
1345 {
1346 int error;
1347
1348 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1349 if (error)
1350 return error;
1351 truncate_pagecache_range(VFS_I(ip), 0, -1);
1352
1353 /* Verify O_DIRECT for ftmp */
1354 if (VFS_I(ip)->i_mapping->nrpages)
1355 return -EINVAL;
1356 return 0;
1357 }
1358
1359 /*
1360 * Move extents from one file to another, when rmap is enabled.
1361 */
1362 STATIC int
xfs_swap_extent_rmap(struct xfs_trans ** tpp,struct xfs_inode * ip,struct xfs_inode * tip)1363 xfs_swap_extent_rmap(
1364 struct xfs_trans **tpp,
1365 struct xfs_inode *ip,
1366 struct xfs_inode *tip)
1367 {
1368 struct xfs_trans *tp = *tpp;
1369 struct xfs_bmbt_irec irec;
1370 struct xfs_bmbt_irec uirec;
1371 struct xfs_bmbt_irec tirec;
1372 xfs_fileoff_t offset_fsb;
1373 xfs_fileoff_t end_fsb;
1374 xfs_filblks_t count_fsb;
1375 int error;
1376 xfs_filblks_t ilen;
1377 xfs_filblks_t rlen;
1378 int nimaps;
1379 uint64_t tip_flags2;
1380
1381 /*
1382 * If the source file has shared blocks, we must flag the donor
1383 * file as having shared blocks so that we get the shared-block
1384 * rmap functions when we go to fix up the rmaps. The flags
1385 * will be switch for reals later.
1386 */
1387 tip_flags2 = tip->i_diflags2;
1388 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK)
1389 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1390
1391 offset_fsb = 0;
1392 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1393 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1394
1395 while (count_fsb) {
1396 /* Read extent from the donor file */
1397 nimaps = 1;
1398 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1399 &nimaps, 0);
1400 if (error)
1401 goto out;
1402 ASSERT(nimaps == 1);
1403 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1404
1405 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1406 ilen = tirec.br_blockcount;
1407
1408 /* Unmap the old blocks in the source file. */
1409 while (tirec.br_blockcount) {
1410 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1411 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1412
1413 /* Read extent from the source file */
1414 nimaps = 1;
1415 error = xfs_bmapi_read(ip, tirec.br_startoff,
1416 tirec.br_blockcount, &irec,
1417 &nimaps, 0);
1418 if (error)
1419 goto out;
1420 ASSERT(nimaps == 1);
1421 ASSERT(tirec.br_startoff == irec.br_startoff);
1422 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1423
1424 /* Trim the extent. */
1425 uirec = tirec;
1426 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1427 tirec.br_blockcount,
1428 irec.br_blockcount);
1429 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1430
1431 if (xfs_bmap_is_real_extent(&uirec)) {
1432 error = xfs_iext_count_may_overflow(ip,
1433 XFS_DATA_FORK,
1434 XFS_IEXT_SWAP_RMAP_CNT);
1435 if (error == -EFBIG)
1436 error = xfs_iext_count_upgrade(tp, ip,
1437 XFS_IEXT_SWAP_RMAP_CNT);
1438 if (error)
1439 goto out;
1440 }
1441
1442 if (xfs_bmap_is_real_extent(&irec)) {
1443 error = xfs_iext_count_may_overflow(tip,
1444 XFS_DATA_FORK,
1445 XFS_IEXT_SWAP_RMAP_CNT);
1446 if (error == -EFBIG)
1447 error = xfs_iext_count_upgrade(tp, ip,
1448 XFS_IEXT_SWAP_RMAP_CNT);
1449 if (error)
1450 goto out;
1451 }
1452
1453 /* Remove the mapping from the donor file. */
1454 xfs_bmap_unmap_extent(tp, tip, &uirec);
1455
1456 /* Remove the mapping from the source file. */
1457 xfs_bmap_unmap_extent(tp, ip, &irec);
1458
1459 /* Map the donor file's blocks into the source file. */
1460 xfs_bmap_map_extent(tp, ip, &uirec);
1461
1462 /* Map the source file's blocks into the donor file. */
1463 xfs_bmap_map_extent(tp, tip, &irec);
1464
1465 error = xfs_defer_finish(tpp);
1466 tp = *tpp;
1467 if (error)
1468 goto out;
1469
1470 tirec.br_startoff += rlen;
1471 if (tirec.br_startblock != HOLESTARTBLOCK &&
1472 tirec.br_startblock != DELAYSTARTBLOCK)
1473 tirec.br_startblock += rlen;
1474 tirec.br_blockcount -= rlen;
1475 }
1476
1477 /* Roll on... */
1478 count_fsb -= ilen;
1479 offset_fsb += ilen;
1480 }
1481
1482 tip->i_diflags2 = tip_flags2;
1483 return 0;
1484
1485 out:
1486 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1487 tip->i_diflags2 = tip_flags2;
1488 return error;
1489 }
1490
1491 /* Swap the extents of two files by swapping data forks. */
1492 STATIC int
xfs_swap_extent_forks(struct xfs_trans * tp,struct xfs_inode * ip,struct xfs_inode * tip,int * src_log_flags,int * target_log_flags)1493 xfs_swap_extent_forks(
1494 struct xfs_trans *tp,
1495 struct xfs_inode *ip,
1496 struct xfs_inode *tip,
1497 int *src_log_flags,
1498 int *target_log_flags)
1499 {
1500 xfs_filblks_t aforkblks = 0;
1501 xfs_filblks_t taforkblks = 0;
1502 xfs_extnum_t junk;
1503 uint64_t tmp;
1504 int error;
1505
1506 /*
1507 * Count the number of extended attribute blocks
1508 */
1509 if (XFS_IFORK_Q(ip) && ip->i_afp->if_nextents > 0 &&
1510 ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1511 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1512 &aforkblks);
1513 if (error)
1514 return error;
1515 }
1516 if (XFS_IFORK_Q(tip) && tip->i_afp->if_nextents > 0 &&
1517 tip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1518 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1519 &taforkblks);
1520 if (error)
1521 return error;
1522 }
1523
1524 /*
1525 * Btree format (v3) inodes have the inode number stamped in the bmbt
1526 * block headers. We can't start changing the bmbt blocks until the
1527 * inode owner change is logged so recovery does the right thing in the
1528 * event of a crash. Set the owner change log flags now and leave the
1529 * bmbt scan as the last step.
1530 */
1531 if (xfs_has_v3inodes(ip->i_mount)) {
1532 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1533 (*target_log_flags) |= XFS_ILOG_DOWNER;
1534 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1535 (*src_log_flags) |= XFS_ILOG_DOWNER;
1536 }
1537
1538 /*
1539 * Swap the data forks of the inodes
1540 */
1541 swap(ip->i_df, tip->i_df);
1542
1543 /*
1544 * Fix the on-disk inode values
1545 */
1546 tmp = (uint64_t)ip->i_nblocks;
1547 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks;
1548 tip->i_nblocks = tmp + taforkblks - aforkblks;
1549
1550 /*
1551 * The extents in the source inode could still contain speculative
1552 * preallocation beyond EOF (e.g. the file is open but not modified
1553 * while defrag is in progress). In that case, we need to copy over the
1554 * number of delalloc blocks the data fork in the source inode is
1555 * tracking beyond EOF so that when the fork is truncated away when the
1556 * temporary inode is unlinked we don't underrun the i_delayed_blks
1557 * counter on that inode.
1558 */
1559 ASSERT(tip->i_delayed_blks == 0);
1560 tip->i_delayed_blks = ip->i_delayed_blks;
1561 ip->i_delayed_blks = 0;
1562
1563 switch (ip->i_df.if_format) {
1564 case XFS_DINODE_FMT_EXTENTS:
1565 (*src_log_flags) |= XFS_ILOG_DEXT;
1566 break;
1567 case XFS_DINODE_FMT_BTREE:
1568 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1569 (*src_log_flags & XFS_ILOG_DOWNER));
1570 (*src_log_flags) |= XFS_ILOG_DBROOT;
1571 break;
1572 }
1573
1574 switch (tip->i_df.if_format) {
1575 case XFS_DINODE_FMT_EXTENTS:
1576 (*target_log_flags) |= XFS_ILOG_DEXT;
1577 break;
1578 case XFS_DINODE_FMT_BTREE:
1579 (*target_log_flags) |= XFS_ILOG_DBROOT;
1580 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1581 (*target_log_flags & XFS_ILOG_DOWNER));
1582 break;
1583 }
1584
1585 return 0;
1586 }
1587
1588 /*
1589 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1590 * change owner scan attempts to order all modified buffers in the current
1591 * transaction. In the event of ordered buffer failure, the offending buffer is
1592 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1593 * the transaction in this case to replenish the fallback log reservation and
1594 * restart the scan. This process repeats until the scan completes.
1595 */
1596 static int
xfs_swap_change_owner(struct xfs_trans ** tpp,struct xfs_inode * ip,struct xfs_inode * tmpip)1597 xfs_swap_change_owner(
1598 struct xfs_trans **tpp,
1599 struct xfs_inode *ip,
1600 struct xfs_inode *tmpip)
1601 {
1602 int error;
1603 struct xfs_trans *tp = *tpp;
1604
1605 do {
1606 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1607 NULL);
1608 /* success or fatal error */
1609 if (error != -EAGAIN)
1610 break;
1611
1612 error = xfs_trans_roll(tpp);
1613 if (error)
1614 break;
1615 tp = *tpp;
1616
1617 /*
1618 * Redirty both inodes so they can relog and keep the log tail
1619 * moving forward.
1620 */
1621 xfs_trans_ijoin(tp, ip, 0);
1622 xfs_trans_ijoin(tp, tmpip, 0);
1623 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1624 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1625 } while (true);
1626
1627 return error;
1628 }
1629
1630 int
xfs_swap_extents(struct xfs_inode * ip,struct xfs_inode * tip,struct xfs_swapext * sxp)1631 xfs_swap_extents(
1632 struct xfs_inode *ip, /* target inode */
1633 struct xfs_inode *tip, /* tmp inode */
1634 struct xfs_swapext *sxp)
1635 {
1636 struct xfs_mount *mp = ip->i_mount;
1637 struct xfs_trans *tp;
1638 struct xfs_bstat *sbp = &sxp->sx_stat;
1639 int src_log_flags, target_log_flags;
1640 int error = 0;
1641 uint64_t f;
1642 int resblks = 0;
1643 unsigned int flags = 0;
1644
1645 /*
1646 * Lock the inodes against other IO, page faults and truncate to
1647 * begin with. Then we can ensure the inodes are flushed and have no
1648 * page cache safely. Once we have done this we can take the ilocks and
1649 * do the rest of the checks.
1650 */
1651 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1652 filemap_invalidate_lock_two(VFS_I(ip)->i_mapping,
1653 VFS_I(tip)->i_mapping);
1654
1655 /* Verify that both files have the same format */
1656 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1657 error = -EINVAL;
1658 goto out_unlock;
1659 }
1660
1661 /* Verify both files are either real-time or non-realtime */
1662 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1663 error = -EINVAL;
1664 goto out_unlock;
1665 }
1666
1667 error = xfs_qm_dqattach(ip);
1668 if (error)
1669 goto out_unlock;
1670
1671 error = xfs_qm_dqattach(tip);
1672 if (error)
1673 goto out_unlock;
1674
1675 error = xfs_swap_extent_flush(ip);
1676 if (error)
1677 goto out_unlock;
1678 error = xfs_swap_extent_flush(tip);
1679 if (error)
1680 goto out_unlock;
1681
1682 if (xfs_inode_has_cow_data(tip)) {
1683 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1684 if (error)
1685 goto out_unlock;
1686 }
1687
1688 /*
1689 * Extent "swapping" with rmap requires a permanent reservation and
1690 * a block reservation because it's really just a remap operation
1691 * performed with log redo items!
1692 */
1693 if (xfs_has_rmapbt(mp)) {
1694 int w = XFS_DATA_FORK;
1695 uint32_t ipnext = ip->i_df.if_nextents;
1696 uint32_t tipnext = tip->i_df.if_nextents;
1697
1698 /*
1699 * Conceptually this shouldn't affect the shape of either bmbt,
1700 * but since we atomically move extents one by one, we reserve
1701 * enough space to rebuild both trees.
1702 */
1703 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1704 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1705
1706 /*
1707 * If either inode straddles a bmapbt block allocation boundary,
1708 * the rmapbt algorithm triggers repeated allocs and frees as
1709 * extents are remapped. This can exhaust the block reservation
1710 * prematurely and cause shutdown. Return freed blocks to the
1711 * transaction reservation to counter this behavior.
1712 */
1713 flags |= XFS_TRANS_RES_FDBLKS;
1714 }
1715 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1716 &tp);
1717 if (error)
1718 goto out_unlock;
1719
1720 /*
1721 * Lock and join the inodes to the tansaction so that transaction commit
1722 * or cancel will unlock the inodes from this point onwards.
1723 */
1724 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1725 xfs_trans_ijoin(tp, ip, 0);
1726 xfs_trans_ijoin(tp, tip, 0);
1727
1728
1729 /* Verify all data are being swapped */
1730 if (sxp->sx_offset != 0 ||
1731 sxp->sx_length != ip->i_disk_size ||
1732 sxp->sx_length != tip->i_disk_size) {
1733 error = -EFAULT;
1734 goto out_trans_cancel;
1735 }
1736
1737 trace_xfs_swap_extent_before(ip, 0);
1738 trace_xfs_swap_extent_before(tip, 1);
1739
1740 /* check inode formats now that data is flushed */
1741 error = xfs_swap_extents_check_format(ip, tip);
1742 if (error) {
1743 xfs_notice(mp,
1744 "%s: inode 0x%llx format is incompatible for exchanging.",
1745 __func__, ip->i_ino);
1746 goto out_trans_cancel;
1747 }
1748
1749 /*
1750 * Compare the current change & modify times with that
1751 * passed in. If they differ, we abort this swap.
1752 * This is the mechanism used to ensure the calling
1753 * process that the file was not changed out from
1754 * under it.
1755 */
1756 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1757 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1758 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1759 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1760 error = -EBUSY;
1761 goto out_trans_cancel;
1762 }
1763
1764 /*
1765 * Note the trickiness in setting the log flags - we set the owner log
1766 * flag on the opposite inode (i.e. the inode we are setting the new
1767 * owner to be) because once we swap the forks and log that, log
1768 * recovery is going to see the fork as owned by the swapped inode,
1769 * not the pre-swapped inodes.
1770 */
1771 src_log_flags = XFS_ILOG_CORE;
1772 target_log_flags = XFS_ILOG_CORE;
1773
1774 if (xfs_has_rmapbt(mp))
1775 error = xfs_swap_extent_rmap(&tp, ip, tip);
1776 else
1777 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1778 &target_log_flags);
1779 if (error)
1780 goto out_trans_cancel;
1781
1782 /* Do we have to swap reflink flags? */
1783 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
1784 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
1785 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1786 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1787 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1788 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1789 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
1790 }
1791
1792 /* Swap the cow forks. */
1793 if (xfs_has_reflink(mp)) {
1794 ASSERT(!ip->i_cowfp ||
1795 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1796 ASSERT(!tip->i_cowfp ||
1797 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1798
1799 swap(ip->i_cowfp, tip->i_cowfp);
1800
1801 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1802 xfs_inode_set_cowblocks_tag(ip);
1803 else
1804 xfs_inode_clear_cowblocks_tag(ip);
1805 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1806 xfs_inode_set_cowblocks_tag(tip);
1807 else
1808 xfs_inode_clear_cowblocks_tag(tip);
1809 }
1810
1811 xfs_trans_log_inode(tp, ip, src_log_flags);
1812 xfs_trans_log_inode(tp, tip, target_log_flags);
1813
1814 /*
1815 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1816 * have inode number owner values in the bmbt blocks that still refer to
1817 * the old inode. Scan each bmbt to fix up the owner values with the
1818 * inode number of the current inode.
1819 */
1820 if (src_log_flags & XFS_ILOG_DOWNER) {
1821 error = xfs_swap_change_owner(&tp, ip, tip);
1822 if (error)
1823 goto out_trans_cancel;
1824 }
1825 if (target_log_flags & XFS_ILOG_DOWNER) {
1826 error = xfs_swap_change_owner(&tp, tip, ip);
1827 if (error)
1828 goto out_trans_cancel;
1829 }
1830
1831 /*
1832 * If this is a synchronous mount, make sure that the
1833 * transaction goes to disk before returning to the user.
1834 */
1835 if (xfs_has_wsync(mp))
1836 xfs_trans_set_sync(tp);
1837
1838 error = xfs_trans_commit(tp);
1839
1840 trace_xfs_swap_extent_after(ip, 0);
1841 trace_xfs_swap_extent_after(tip, 1);
1842
1843 out_unlock_ilock:
1844 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1845 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1846 out_unlock:
1847 filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping,
1848 VFS_I(tip)->i_mapping);
1849 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1850 return error;
1851
1852 out_trans_cancel:
1853 xfs_trans_cancel(tp);
1854 goto out_unlock_ilock;
1855 }
1856