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
443 xfs_ilock(ip, XFS_IOLOCK_SHARED);
444 switch (whichfork) {
445 case XFS_ATTR_FORK:
446 lock = xfs_ilock_attr_map_shared(ip);
447 if (!xfs_inode_has_attr_fork(ip))
448 goto out_unlock_ilock;
449
450 max_len = 1LL << 32;
451 break;
452 case XFS_COW_FORK:
453 lock = XFS_ILOCK_SHARED;
454 xfs_ilock(ip, lock);
455
456 /* No CoW fork? Just return */
457 if (!xfs_ifork_ptr(ip, whichfork))
458 goto out_unlock_ilock;
459
460 if (xfs_get_cowextsz_hint(ip))
461 max_len = mp->m_super->s_maxbytes;
462 else
463 max_len = XFS_ISIZE(ip);
464 break;
465 case XFS_DATA_FORK:
466 if (!(iflags & BMV_IF_DELALLOC) &&
467 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) {
468 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
469 if (error)
470 goto out_unlock_iolock;
471
472 /*
473 * Even after flushing the inode, there can still be
474 * delalloc blocks on the inode beyond EOF due to
475 * speculative preallocation. These are not removed
476 * until the release function is called or the inode
477 * is inactivated. Hence we cannot assert here that
478 * ip->i_delayed_blks == 0.
479 */
480 }
481
482 if (xfs_get_extsz_hint(ip) ||
483 (ip->i_diflags &
484 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
485 max_len = mp->m_super->s_maxbytes;
486 else
487 max_len = XFS_ISIZE(ip);
488
489 lock = xfs_ilock_data_map_shared(ip);
490 break;
491 }
492
493 ifp = xfs_ifork_ptr(ip, whichfork);
494
495 switch (ifp->if_format) {
496 case XFS_DINODE_FMT_EXTENTS:
497 case XFS_DINODE_FMT_BTREE:
498 break;
499 case XFS_DINODE_FMT_LOCAL:
500 /* Local format inode forks report no extents. */
501 goto out_unlock_ilock;
502 default:
503 error = -EINVAL;
504 goto out_unlock_ilock;
505 }
506
507 if (bmv->bmv_length == -1) {
508 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
509 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
510 }
511
512 bmv_end = bmv->bmv_offset + bmv->bmv_length;
513
514 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
515 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
516
517 error = xfs_iread_extents(NULL, ip, whichfork);
518 if (error)
519 goto out_unlock_ilock;
520
521 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
522 /*
523 * Report a whole-file hole if the delalloc flag is set to
524 * stay compatible with the old implementation.
525 */
526 if (iflags & BMV_IF_DELALLOC)
527 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
528 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
529 goto out_unlock_ilock;
530 }
531
532 while (!xfs_getbmap_full(bmv)) {
533 xfs_trim_extent(&got, first_bno, len);
534
535 /*
536 * Report an entry for a hole if this extent doesn't directly
537 * follow the previous one.
538 */
539 if (got.br_startoff > bno) {
540 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
541 got.br_startoff);
542 if (xfs_getbmap_full(bmv))
543 break;
544 }
545
546 /*
547 * In order to report shared extents accurately, we report each
548 * distinct shared / unshared part of a single bmbt record with
549 * an individual getbmapx record.
550 */
551 bno = got.br_startoff + got.br_blockcount;
552 rec = got;
553 do {
554 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
555 &rec);
556 if (error || xfs_getbmap_full(bmv))
557 goto out_unlock_ilock;
558 } while (xfs_getbmap_next_rec(&rec, bno));
559
560 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
561 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
562
563 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
564
565 if (whichfork != XFS_ATTR_FORK && bno < end &&
566 !xfs_getbmap_full(bmv)) {
567 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
568 bno, end);
569 }
570 break;
571 }
572
573 if (bno >= first_bno + len)
574 break;
575 }
576
577 out_unlock_ilock:
578 xfs_iunlock(ip, lock);
579 out_unlock_iolock:
580 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
581 return error;
582 }
583
584 /*
585 * Dead simple method of punching delalyed allocation blocks from a range in
586 * the inode. This will always punch out both the start and end blocks, even
587 * if the ranges only partially overlap them, so it is up to the caller to
588 * ensure that partial blocks are not passed in.
589 */
590 int
xfs_bmap_punch_delalloc_range(struct xfs_inode * ip,xfs_fileoff_t start_fsb,xfs_fileoff_t length)591 xfs_bmap_punch_delalloc_range(
592 struct xfs_inode *ip,
593 xfs_fileoff_t start_fsb,
594 xfs_fileoff_t length)
595 {
596 struct xfs_ifork *ifp = &ip->i_df;
597 xfs_fileoff_t end_fsb = start_fsb + length;
598 struct xfs_bmbt_irec got, del;
599 struct xfs_iext_cursor icur;
600 int error = 0;
601
602 ASSERT(!xfs_need_iread_extents(ifp));
603
604 xfs_ilock(ip, XFS_ILOCK_EXCL);
605 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
606 goto out_unlock;
607
608 while (got.br_startoff + got.br_blockcount > start_fsb) {
609 del = got;
610 xfs_trim_extent(&del, start_fsb, length);
611
612 /*
613 * A delete can push the cursor forward. Step back to the
614 * previous extent on non-delalloc or extents outside the
615 * target range.
616 */
617 if (!del.br_blockcount ||
618 !isnullstartblock(del.br_startblock)) {
619 if (!xfs_iext_prev_extent(ifp, &icur, &got))
620 break;
621 continue;
622 }
623
624 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
625 &got, &del);
626 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
627 break;
628 }
629
630 out_unlock:
631 xfs_iunlock(ip, XFS_ILOCK_EXCL);
632 return error;
633 }
634
635 /*
636 * Test whether it is appropriate to check an inode for and free post EOF
637 * blocks. The 'force' parameter determines whether we should also consider
638 * regular files that are marked preallocated or append-only.
639 */
640 bool
xfs_can_free_eofblocks(struct xfs_inode * ip,bool force)641 xfs_can_free_eofblocks(
642 struct xfs_inode *ip,
643 bool force)
644 {
645 struct xfs_bmbt_irec imap;
646 struct xfs_mount *mp = ip->i_mount;
647 xfs_fileoff_t end_fsb;
648 xfs_fileoff_t last_fsb;
649 int nimaps = 1;
650 int error;
651
652 /*
653 * Caller must either hold the exclusive io lock; or be inactivating
654 * the inode, which guarantees there are no other users of the inode.
655 */
656 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL) ||
657 (VFS_I(ip)->i_state & I_FREEING));
658
659 /* prealloc/delalloc exists only on regular files */
660 if (!S_ISREG(VFS_I(ip)->i_mode))
661 return false;
662
663 /*
664 * Zero sized files with no cached pages and delalloc blocks will not
665 * have speculative prealloc/delalloc blocks to remove.
666 */
667 if (VFS_I(ip)->i_size == 0 &&
668 VFS_I(ip)->i_mapping->nrpages == 0 &&
669 ip->i_delayed_blks == 0)
670 return false;
671
672 /* If we haven't read in the extent list, then don't do it now. */
673 if (xfs_need_iread_extents(&ip->i_df))
674 return false;
675
676 /*
677 * Do not free real preallocated or append-only files unless the file
678 * has delalloc blocks and we are forced to remove them.
679 */
680 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
681 if (!force || ip->i_delayed_blks == 0)
682 return false;
683
684 /*
685 * Do not try to free post-EOF blocks if EOF is beyond the end of the
686 * range supported by the page cache, because the truncation will loop
687 * forever.
688 */
689 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
690 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1)
691 end_fsb = roundup_64(end_fsb, mp->m_sb.sb_rextsize);
692 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
693 if (last_fsb <= end_fsb)
694 return false;
695
696 /*
697 * Look up the mapping for the first block past EOF. If we can't find
698 * it, there's nothing to free.
699 */
700 xfs_ilock(ip, XFS_ILOCK_SHARED);
701 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps,
702 0);
703 xfs_iunlock(ip, XFS_ILOCK_SHARED);
704 if (error || nimaps == 0)
705 return false;
706
707 /*
708 * If there's a real mapping there or there are delayed allocation
709 * reservations, then we have post-EOF blocks to try to free.
710 */
711 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks;
712 }
713
714 /*
715 * This is called to free any blocks beyond eof. The caller must hold
716 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
717 * reference to the inode.
718 */
719 int
xfs_free_eofblocks(struct xfs_inode * ip)720 xfs_free_eofblocks(
721 struct xfs_inode *ip)
722 {
723 struct xfs_trans *tp;
724 struct xfs_mount *mp = ip->i_mount;
725 int error;
726
727 /* Attach the dquots to the inode up front. */
728 error = xfs_qm_dqattach(ip);
729 if (error)
730 return error;
731
732 /* Wait on dio to ensure i_size has settled. */
733 inode_dio_wait(VFS_I(ip));
734
735 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
736 if (error) {
737 ASSERT(xfs_is_shutdown(mp));
738 return error;
739 }
740
741 xfs_ilock(ip, XFS_ILOCK_EXCL);
742 xfs_trans_ijoin(tp, ip, 0);
743
744 /*
745 * Do not update the on-disk file size. If we update the on-disk file
746 * size and then the system crashes before the contents of the file are
747 * flushed to disk then the files may be full of holes (ie NULL files
748 * bug).
749 */
750 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
751 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
752 if (error)
753 goto err_cancel;
754
755 error = xfs_trans_commit(tp);
756 if (error)
757 goto out_unlock;
758
759 xfs_inode_clear_eofblocks_tag(ip);
760 goto out_unlock;
761
762 err_cancel:
763 /*
764 * If we get an error at this point we simply don't
765 * bother truncating the file.
766 */
767 xfs_trans_cancel(tp);
768 out_unlock:
769 xfs_iunlock(ip, XFS_ILOCK_EXCL);
770 return error;
771 }
772
773 int
xfs_alloc_file_space(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)774 xfs_alloc_file_space(
775 struct xfs_inode *ip,
776 xfs_off_t offset,
777 xfs_off_t len)
778 {
779 xfs_mount_t *mp = ip->i_mount;
780 xfs_off_t count;
781 xfs_filblks_t allocated_fsb;
782 xfs_filblks_t allocatesize_fsb;
783 xfs_extlen_t extsz, temp;
784 xfs_fileoff_t startoffset_fsb;
785 xfs_fileoff_t endoffset_fsb;
786 int nimaps;
787 int rt;
788 xfs_trans_t *tp;
789 xfs_bmbt_irec_t imaps[1], *imapp;
790 int error;
791
792 trace_xfs_alloc_file_space(ip);
793
794 if (xfs_is_shutdown(mp))
795 return -EIO;
796
797 error = xfs_qm_dqattach(ip);
798 if (error)
799 return error;
800
801 if (len <= 0)
802 return -EINVAL;
803
804 rt = XFS_IS_REALTIME_INODE(ip);
805 extsz = xfs_get_extsz_hint(ip);
806
807 count = len;
808 imapp = &imaps[0];
809 nimaps = 1;
810 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
811 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
812 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
813
814 /*
815 * Allocate file space until done or until there is an error
816 */
817 while (allocatesize_fsb && !error) {
818 xfs_fileoff_t s, e;
819 unsigned int dblocks, rblocks, resblks;
820
821 /*
822 * Determine space reservations for data/realtime.
823 */
824 if (unlikely(extsz)) {
825 s = startoffset_fsb;
826 do_div(s, extsz);
827 s *= extsz;
828 e = startoffset_fsb + allocatesize_fsb;
829 div_u64_rem(startoffset_fsb, extsz, &temp);
830 if (temp)
831 e += temp;
832 div_u64_rem(e, extsz, &temp);
833 if (temp)
834 e += extsz - temp;
835 } else {
836 s = 0;
837 e = allocatesize_fsb;
838 }
839
840 /*
841 * The transaction reservation is limited to a 32-bit block
842 * count, hence we need to limit the number of blocks we are
843 * trying to reserve to avoid an overflow. We can't allocate
844 * more than @nimaps extents, and an extent is limited on disk
845 * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the
846 * limit.
847 */
848 resblks = min_t(xfs_fileoff_t, (e - s),
849 (XFS_MAX_BMBT_EXTLEN * nimaps));
850 if (unlikely(rt)) {
851 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
852 rblocks = resblks;
853 } else {
854 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
855 rblocks = 0;
856 }
857
858 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
859 dblocks, rblocks, false, &tp);
860 if (error)
861 break;
862
863 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
864 XFS_IEXT_ADD_NOSPLIT_CNT);
865 if (error == -EFBIG)
866 error = xfs_iext_count_upgrade(tp, ip,
867 XFS_IEXT_ADD_NOSPLIT_CNT);
868 if (error)
869 goto error;
870
871 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
872 allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp,
873 &nimaps);
874 if (error)
875 goto error;
876
877 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
878 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
879
880 error = xfs_trans_commit(tp);
881 xfs_iunlock(ip, XFS_ILOCK_EXCL);
882 if (error)
883 break;
884
885 allocated_fsb = imapp->br_blockcount;
886
887 if (nimaps == 0) {
888 error = -ENOSPC;
889 break;
890 }
891
892 startoffset_fsb += allocated_fsb;
893 allocatesize_fsb -= allocated_fsb;
894 }
895
896 return error;
897
898 error:
899 xfs_trans_cancel(tp);
900 xfs_iunlock(ip, XFS_ILOCK_EXCL);
901 return error;
902 }
903
904 static int
xfs_unmap_extent(struct xfs_inode * ip,xfs_fileoff_t startoffset_fsb,xfs_filblks_t len_fsb,int * done)905 xfs_unmap_extent(
906 struct xfs_inode *ip,
907 xfs_fileoff_t startoffset_fsb,
908 xfs_filblks_t len_fsb,
909 int *done)
910 {
911 struct xfs_mount *mp = ip->i_mount;
912 struct xfs_trans *tp;
913 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
914 int error;
915
916 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
917 false, &tp);
918 if (error)
919 return error;
920
921 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
922 XFS_IEXT_PUNCH_HOLE_CNT);
923 if (error == -EFBIG)
924 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
925 if (error)
926 goto out_trans_cancel;
927
928 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
929 if (error)
930 goto out_trans_cancel;
931
932 error = xfs_trans_commit(tp);
933 out_unlock:
934 xfs_iunlock(ip, XFS_ILOCK_EXCL);
935 return error;
936
937 out_trans_cancel:
938 xfs_trans_cancel(tp);
939 goto out_unlock;
940 }
941
942 /* Caller must first wait for the completion of any pending DIOs if required. */
943 int
xfs_flush_unmap_range(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)944 xfs_flush_unmap_range(
945 struct xfs_inode *ip,
946 xfs_off_t offset,
947 xfs_off_t len)
948 {
949 struct xfs_mount *mp = ip->i_mount;
950 struct inode *inode = VFS_I(ip);
951 xfs_off_t rounding, start, end;
952 int error;
953
954 rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE);
955 start = round_down(offset, rounding);
956 end = round_up(offset + len, rounding) - 1;
957
958 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
959 if (error)
960 return error;
961 truncate_pagecache_range(inode, start, end);
962 return 0;
963 }
964
965 int
xfs_free_file_space(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)966 xfs_free_file_space(
967 struct xfs_inode *ip,
968 xfs_off_t offset,
969 xfs_off_t len)
970 {
971 struct xfs_mount *mp = ip->i_mount;
972 xfs_fileoff_t startoffset_fsb;
973 xfs_fileoff_t endoffset_fsb;
974 int done = 0, error;
975
976 trace_xfs_free_file_space(ip);
977
978 error = xfs_qm_dqattach(ip);
979 if (error)
980 return error;
981
982 if (len <= 0) /* if nothing being freed */
983 return 0;
984
985 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
986 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
987
988 /* We can only free complete realtime extents. */
989 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
990 startoffset_fsb = roundup_64(startoffset_fsb,
991 mp->m_sb.sb_rextsize);
992 endoffset_fsb = rounddown_64(endoffset_fsb,
993 mp->m_sb.sb_rextsize);
994 }
995
996 /*
997 * Need to zero the stuff we're not freeing, on disk.
998 */
999 if (endoffset_fsb > startoffset_fsb) {
1000 while (!done) {
1001 error = xfs_unmap_extent(ip, startoffset_fsb,
1002 endoffset_fsb - startoffset_fsb, &done);
1003 if (error)
1004 return error;
1005 }
1006 }
1007
1008 /*
1009 * Now that we've unmap all full blocks we'll have to zero out any
1010 * partial block at the beginning and/or end. xfs_zero_range is smart
1011 * enough to skip any holes, including those we just created, but we
1012 * must take care not to zero beyond EOF and enlarge i_size.
1013 */
1014 if (offset >= XFS_ISIZE(ip))
1015 return 0;
1016 if (offset + len > XFS_ISIZE(ip))
1017 len = XFS_ISIZE(ip) - offset;
1018 error = xfs_zero_range(ip, offset, len, NULL);
1019 if (error)
1020 return error;
1021
1022 /*
1023 * If we zeroed right up to EOF and EOF straddles a page boundary we
1024 * must make sure that the post-EOF area is also zeroed because the
1025 * page could be mmap'd and xfs_zero_range doesn't do that for us.
1026 * Writeback of the eof page will do this, albeit clumsily.
1027 */
1028 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
1029 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1030 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
1031 }
1032
1033 return error;
1034 }
1035
1036 static int
xfs_prepare_shift(struct xfs_inode * ip,loff_t offset)1037 xfs_prepare_shift(
1038 struct xfs_inode *ip,
1039 loff_t offset)
1040 {
1041 struct xfs_mount *mp = ip->i_mount;
1042 int error;
1043
1044 /*
1045 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1046 * into the accessible region of the file.
1047 */
1048 if (xfs_can_free_eofblocks(ip, true)) {
1049 error = xfs_free_eofblocks(ip);
1050 if (error)
1051 return error;
1052 }
1053
1054 /*
1055 * Shift operations must stabilize the start block offset boundary along
1056 * with the full range of the operation. If we don't, a COW writeback
1057 * completion could race with an insert, front merge with the start
1058 * extent (after split) during the shift and corrupt the file. Start
1059 * with the block just prior to the start to stabilize the boundary.
1060 */
1061 offset = round_down(offset, mp->m_sb.sb_blocksize);
1062 if (offset)
1063 offset -= mp->m_sb.sb_blocksize;
1064
1065 /*
1066 * Writeback and invalidate cache for the remainder of the file as we're
1067 * about to shift down every extent from offset to EOF.
1068 */
1069 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1070 if (error)
1071 return error;
1072
1073 /*
1074 * Clean out anything hanging around in the cow fork now that
1075 * we've flushed all the dirty data out to disk to avoid having
1076 * CoW extents at the wrong offsets.
1077 */
1078 if (xfs_inode_has_cow_data(ip)) {
1079 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1080 true);
1081 if (error)
1082 return error;
1083 }
1084
1085 return 0;
1086 }
1087
1088 /*
1089 * xfs_collapse_file_space()
1090 * This routine frees disk space and shift extent for the given file.
1091 * The first thing we do is to free data blocks in the specified range
1092 * by calling xfs_free_file_space(). It would also sync dirty data
1093 * and invalidate page cache over the region on which collapse range
1094 * is working. And Shift extent records to the left to cover a hole.
1095 * RETURNS:
1096 * 0 on success
1097 * errno on error
1098 *
1099 */
1100 int
xfs_collapse_file_space(struct xfs_inode * ip,xfs_off_t offset,xfs_off_t len)1101 xfs_collapse_file_space(
1102 struct xfs_inode *ip,
1103 xfs_off_t offset,
1104 xfs_off_t len)
1105 {
1106 struct xfs_mount *mp = ip->i_mount;
1107 struct xfs_trans *tp;
1108 int error;
1109 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
1110 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1111 bool done = false;
1112
1113 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1114 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1115
1116 trace_xfs_collapse_file_space(ip);
1117
1118 error = xfs_free_file_space(ip, offset, len);
1119 if (error)
1120 return error;
1121
1122 error = xfs_prepare_shift(ip, offset);
1123 if (error)
1124 return error;
1125
1126 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1127 if (error)
1128 return error;
1129
1130 xfs_ilock(ip, XFS_ILOCK_EXCL);
1131 xfs_trans_ijoin(tp, ip, 0);
1132
1133 while (!done) {
1134 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1135 &done);
1136 if (error)
1137 goto out_trans_cancel;
1138 if (done)
1139 break;
1140
1141 /* finish any deferred frees and roll the transaction */
1142 error = xfs_defer_finish(&tp);
1143 if (error)
1144 goto out_trans_cancel;
1145 }
1146
1147 error = xfs_trans_commit(tp);
1148 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1149 return error;
1150
1151 out_trans_cancel:
1152 xfs_trans_cancel(tp);
1153 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1154 return error;
1155 }
1156
1157 /*
1158 * xfs_insert_file_space()
1159 * This routine create hole space by shifting extents for the given file.
1160 * The first thing we do is to sync dirty data and invalidate page cache
1161 * over the region on which insert range is working. And split an extent
1162 * to two extents at given offset by calling xfs_bmap_split_extent.
1163 * And shift all extent records which are laying between [offset,
1164 * last allocated extent] to the right to reserve hole range.
1165 * RETURNS:
1166 * 0 on success
1167 * errno on error
1168 */
1169 int
xfs_insert_file_space(struct xfs_inode * ip,loff_t offset,loff_t len)1170 xfs_insert_file_space(
1171 struct xfs_inode *ip,
1172 loff_t offset,
1173 loff_t len)
1174 {
1175 struct xfs_mount *mp = ip->i_mount;
1176 struct xfs_trans *tp;
1177 int error;
1178 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1179 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1180 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1181 bool done = false;
1182
1183 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1184 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1185
1186 trace_xfs_insert_file_space(ip);
1187
1188 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1189 if (error)
1190 return error;
1191
1192 error = xfs_prepare_shift(ip, offset);
1193 if (error)
1194 return error;
1195
1196 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1197 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1198 if (error)
1199 return error;
1200
1201 xfs_ilock(ip, XFS_ILOCK_EXCL);
1202 xfs_trans_ijoin(tp, ip, 0);
1203
1204 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
1205 XFS_IEXT_PUNCH_HOLE_CNT);
1206 if (error == -EFBIG)
1207 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
1208 if (error)
1209 goto out_trans_cancel;
1210
1211 /*
1212 * The extent shifting code works on extent granularity. So, if stop_fsb
1213 * is not the starting block of extent, we need to split the extent at
1214 * stop_fsb.
1215 */
1216 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1217 if (error)
1218 goto out_trans_cancel;
1219
1220 do {
1221 error = xfs_defer_finish(&tp);
1222 if (error)
1223 goto out_trans_cancel;
1224
1225 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1226 &done, stop_fsb);
1227 if (error)
1228 goto out_trans_cancel;
1229 } while (!done);
1230
1231 error = xfs_trans_commit(tp);
1232 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1233 return error;
1234
1235 out_trans_cancel:
1236 xfs_trans_cancel(tp);
1237 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1238 return error;
1239 }
1240
1241 /*
1242 * We need to check that the format of the data fork in the temporary inode is
1243 * valid for the target inode before doing the swap. This is not a problem with
1244 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1245 * data fork depending on the space the attribute fork is taking so we can get
1246 * invalid formats on the target inode.
1247 *
1248 * E.g. target has space for 7 extents in extent format, temp inode only has
1249 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1250 * btree, but when swapped it needs to be in extent format. Hence we can't just
1251 * blindly swap data forks on attr2 filesystems.
1252 *
1253 * Note that we check the swap in both directions so that we don't end up with
1254 * a corrupt temporary inode, either.
1255 *
1256 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1257 * inode will prevent this situation from occurring, so all we do here is
1258 * reject and log the attempt. basically we are putting the responsibility on
1259 * userspace to get this right.
1260 */
1261 static int
xfs_swap_extents_check_format(struct xfs_inode * ip,struct xfs_inode * tip)1262 xfs_swap_extents_check_format(
1263 struct xfs_inode *ip, /* target inode */
1264 struct xfs_inode *tip) /* tmp inode */
1265 {
1266 struct xfs_ifork *ifp = &ip->i_df;
1267 struct xfs_ifork *tifp = &tip->i_df;
1268
1269 /* User/group/project quota ids must match if quotas are enforced. */
1270 if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1271 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1272 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1273 ip->i_projid != tip->i_projid))
1274 return -EINVAL;
1275
1276 /* Should never get a local format */
1277 if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1278 tifp->if_format == XFS_DINODE_FMT_LOCAL)
1279 return -EINVAL;
1280
1281 /*
1282 * if the target inode has less extents that then temporary inode then
1283 * why did userspace call us?
1284 */
1285 if (ifp->if_nextents < tifp->if_nextents)
1286 return -EINVAL;
1287
1288 /*
1289 * If we have to use the (expensive) rmap swap method, we can
1290 * handle any number of extents and any format.
1291 */
1292 if (xfs_has_rmapbt(ip->i_mount))
1293 return 0;
1294
1295 /*
1296 * if the target inode is in extent form and the temp inode is in btree
1297 * form then we will end up with the target inode in the wrong format
1298 * as we already know there are less extents in the temp inode.
1299 */
1300 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1301 tifp->if_format == XFS_DINODE_FMT_BTREE)
1302 return -EINVAL;
1303
1304 /* Check temp in extent form to max in target */
1305 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1306 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1307 return -EINVAL;
1308
1309 /* Check target in extent form to max in temp */
1310 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1311 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1312 return -EINVAL;
1313
1314 /*
1315 * If we are in a btree format, check that the temp root block will fit
1316 * in the target and that it has enough extents to be in btree format
1317 * in the target.
1318 *
1319 * Note that we have to be careful to allow btree->extent conversions
1320 * (a common defrag case) which will occur when the temp inode is in
1321 * extent format...
1322 */
1323 if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1324 if (xfs_inode_has_attr_fork(ip) &&
1325 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > xfs_inode_fork_boff(ip))
1326 return -EINVAL;
1327 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1328 return -EINVAL;
1329 }
1330
1331 /* Reciprocal target->temp btree format checks */
1332 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1333 if (xfs_inode_has_attr_fork(tip) &&
1334 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > xfs_inode_fork_boff(tip))
1335 return -EINVAL;
1336 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1337 return -EINVAL;
1338 }
1339
1340 return 0;
1341 }
1342
1343 static int
xfs_swap_extent_flush(struct xfs_inode * ip)1344 xfs_swap_extent_flush(
1345 struct xfs_inode *ip)
1346 {
1347 int error;
1348
1349 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1350 if (error)
1351 return error;
1352 truncate_pagecache_range(VFS_I(ip), 0, -1);
1353
1354 /* Verify O_DIRECT for ftmp */
1355 if (VFS_I(ip)->i_mapping->nrpages)
1356 return -EINVAL;
1357 return 0;
1358 }
1359
1360 /*
1361 * Move extents from one file to another, when rmap is enabled.
1362 */
1363 STATIC int
xfs_swap_extent_rmap(struct xfs_trans ** tpp,struct xfs_inode * ip,struct xfs_inode * tip)1364 xfs_swap_extent_rmap(
1365 struct xfs_trans **tpp,
1366 struct xfs_inode *ip,
1367 struct xfs_inode *tip)
1368 {
1369 struct xfs_trans *tp = *tpp;
1370 struct xfs_bmbt_irec irec;
1371 struct xfs_bmbt_irec uirec;
1372 struct xfs_bmbt_irec tirec;
1373 xfs_fileoff_t offset_fsb;
1374 xfs_fileoff_t end_fsb;
1375 xfs_filblks_t count_fsb;
1376 int error;
1377 xfs_filblks_t ilen;
1378 xfs_filblks_t rlen;
1379 int nimaps;
1380 uint64_t tip_flags2;
1381
1382 /*
1383 * If the source file has shared blocks, we must flag the donor
1384 * file as having shared blocks so that we get the shared-block
1385 * rmap functions when we go to fix up the rmaps. The flags
1386 * will be switch for reals later.
1387 */
1388 tip_flags2 = tip->i_diflags2;
1389 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK)
1390 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1391
1392 offset_fsb = 0;
1393 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1394 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1395
1396 while (count_fsb) {
1397 /* Read extent from the donor file */
1398 nimaps = 1;
1399 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1400 &nimaps, 0);
1401 if (error)
1402 goto out;
1403 ASSERT(nimaps == 1);
1404 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1405
1406 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1407 ilen = tirec.br_blockcount;
1408
1409 /* Unmap the old blocks in the source file. */
1410 while (tirec.br_blockcount) {
1411 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1412 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1413
1414 /* Read extent from the source file */
1415 nimaps = 1;
1416 error = xfs_bmapi_read(ip, tirec.br_startoff,
1417 tirec.br_blockcount, &irec,
1418 &nimaps, 0);
1419 if (error)
1420 goto out;
1421 ASSERT(nimaps == 1);
1422 ASSERT(tirec.br_startoff == irec.br_startoff);
1423 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1424
1425 /* Trim the extent. */
1426 uirec = tirec;
1427 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1428 tirec.br_blockcount,
1429 irec.br_blockcount);
1430 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1431
1432 if (xfs_bmap_is_real_extent(&uirec)) {
1433 error = xfs_iext_count_may_overflow(ip,
1434 XFS_DATA_FORK,
1435 XFS_IEXT_SWAP_RMAP_CNT);
1436 if (error == -EFBIG)
1437 error = xfs_iext_count_upgrade(tp, ip,
1438 XFS_IEXT_SWAP_RMAP_CNT);
1439 if (error)
1440 goto out;
1441 }
1442
1443 if (xfs_bmap_is_real_extent(&irec)) {
1444 error = xfs_iext_count_may_overflow(tip,
1445 XFS_DATA_FORK,
1446 XFS_IEXT_SWAP_RMAP_CNT);
1447 if (error == -EFBIG)
1448 error = xfs_iext_count_upgrade(tp, ip,
1449 XFS_IEXT_SWAP_RMAP_CNT);
1450 if (error)
1451 goto out;
1452 }
1453
1454 /* Remove the mapping from the donor file. */
1455 xfs_bmap_unmap_extent(tp, tip, &uirec);
1456
1457 /* Remove the mapping from the source file. */
1458 xfs_bmap_unmap_extent(tp, ip, &irec);
1459
1460 /* Map the donor file's blocks into the source file. */
1461 xfs_bmap_map_extent(tp, ip, &uirec);
1462
1463 /* Map the source file's blocks into the donor file. */
1464 xfs_bmap_map_extent(tp, tip, &irec);
1465
1466 error = xfs_defer_finish(tpp);
1467 tp = *tpp;
1468 if (error)
1469 goto out;
1470
1471 tirec.br_startoff += rlen;
1472 if (tirec.br_startblock != HOLESTARTBLOCK &&
1473 tirec.br_startblock != DELAYSTARTBLOCK)
1474 tirec.br_startblock += rlen;
1475 tirec.br_blockcount -= rlen;
1476 }
1477
1478 /* Roll on... */
1479 count_fsb -= ilen;
1480 offset_fsb += ilen;
1481 }
1482
1483 tip->i_diflags2 = tip_flags2;
1484 return 0;
1485
1486 out:
1487 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1488 tip->i_diflags2 = tip_flags2;
1489 return error;
1490 }
1491
1492 /* Swap the extents of two files by swapping data forks. */
1493 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)1494 xfs_swap_extent_forks(
1495 struct xfs_trans *tp,
1496 struct xfs_inode *ip,
1497 struct xfs_inode *tip,
1498 int *src_log_flags,
1499 int *target_log_flags)
1500 {
1501 xfs_filblks_t aforkblks = 0;
1502 xfs_filblks_t taforkblks = 0;
1503 xfs_extnum_t junk;
1504 uint64_t tmp;
1505 int error;
1506
1507 /*
1508 * Count the number of extended attribute blocks
1509 */
1510 if (xfs_inode_has_attr_fork(ip) && ip->i_af.if_nextents > 0 &&
1511 ip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1512 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1513 &aforkblks);
1514 if (error)
1515 return error;
1516 }
1517 if (xfs_inode_has_attr_fork(tip) && tip->i_af.if_nextents > 0 &&
1518 tip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1519 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1520 &taforkblks);
1521 if (error)
1522 return error;
1523 }
1524
1525 /*
1526 * Btree format (v3) inodes have the inode number stamped in the bmbt
1527 * block headers. We can't start changing the bmbt blocks until the
1528 * inode owner change is logged so recovery does the right thing in the
1529 * event of a crash. Set the owner change log flags now and leave the
1530 * bmbt scan as the last step.
1531 */
1532 if (xfs_has_v3inodes(ip->i_mount)) {
1533 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1534 (*target_log_flags) |= XFS_ILOG_DOWNER;
1535 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1536 (*src_log_flags) |= XFS_ILOG_DOWNER;
1537 }
1538
1539 /*
1540 * Swap the data forks of the inodes
1541 */
1542 swap(ip->i_df, tip->i_df);
1543
1544 /*
1545 * Fix the on-disk inode values
1546 */
1547 tmp = (uint64_t)ip->i_nblocks;
1548 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks;
1549 tip->i_nblocks = tmp + taforkblks - aforkblks;
1550
1551 /*
1552 * The extents in the source inode could still contain speculative
1553 * preallocation beyond EOF (e.g. the file is open but not modified
1554 * while defrag is in progress). In that case, we need to copy over the
1555 * number of delalloc blocks the data fork in the source inode is
1556 * tracking beyond EOF so that when the fork is truncated away when the
1557 * temporary inode is unlinked we don't underrun the i_delayed_blks
1558 * counter on that inode.
1559 */
1560 ASSERT(tip->i_delayed_blks == 0);
1561 tip->i_delayed_blks = ip->i_delayed_blks;
1562 ip->i_delayed_blks = 0;
1563
1564 switch (ip->i_df.if_format) {
1565 case XFS_DINODE_FMT_EXTENTS:
1566 (*src_log_flags) |= XFS_ILOG_DEXT;
1567 break;
1568 case XFS_DINODE_FMT_BTREE:
1569 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1570 (*src_log_flags & XFS_ILOG_DOWNER));
1571 (*src_log_flags) |= XFS_ILOG_DBROOT;
1572 break;
1573 }
1574
1575 switch (tip->i_df.if_format) {
1576 case XFS_DINODE_FMT_EXTENTS:
1577 (*target_log_flags) |= XFS_ILOG_DEXT;
1578 break;
1579 case XFS_DINODE_FMT_BTREE:
1580 (*target_log_flags) |= XFS_ILOG_DBROOT;
1581 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1582 (*target_log_flags & XFS_ILOG_DOWNER));
1583 break;
1584 }
1585
1586 return 0;
1587 }
1588
1589 /*
1590 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1591 * change owner scan attempts to order all modified buffers in the current
1592 * transaction. In the event of ordered buffer failure, the offending buffer is
1593 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1594 * the transaction in this case to replenish the fallback log reservation and
1595 * restart the scan. This process repeats until the scan completes.
1596 */
1597 static int
xfs_swap_change_owner(struct xfs_trans ** tpp,struct xfs_inode * ip,struct xfs_inode * tmpip)1598 xfs_swap_change_owner(
1599 struct xfs_trans **tpp,
1600 struct xfs_inode *ip,
1601 struct xfs_inode *tmpip)
1602 {
1603 int error;
1604 struct xfs_trans *tp = *tpp;
1605
1606 do {
1607 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1608 NULL);
1609 /* success or fatal error */
1610 if (error != -EAGAIN)
1611 break;
1612
1613 error = xfs_trans_roll(tpp);
1614 if (error)
1615 break;
1616 tp = *tpp;
1617
1618 /*
1619 * Redirty both inodes so they can relog and keep the log tail
1620 * moving forward.
1621 */
1622 xfs_trans_ijoin(tp, ip, 0);
1623 xfs_trans_ijoin(tp, tmpip, 0);
1624 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1625 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1626 } while (true);
1627
1628 return error;
1629 }
1630
1631 int
xfs_swap_extents(struct xfs_inode * ip,struct xfs_inode * tip,struct xfs_swapext * sxp)1632 xfs_swap_extents(
1633 struct xfs_inode *ip, /* target inode */
1634 struct xfs_inode *tip, /* tmp inode */
1635 struct xfs_swapext *sxp)
1636 {
1637 struct xfs_mount *mp = ip->i_mount;
1638 struct xfs_trans *tp;
1639 struct xfs_bstat *sbp = &sxp->sx_stat;
1640 int src_log_flags, target_log_flags;
1641 int error = 0;
1642 uint64_t f;
1643 int resblks = 0;
1644 unsigned int flags = 0;
1645
1646 /*
1647 * Lock the inodes against other IO, page faults and truncate to
1648 * begin with. Then we can ensure the inodes are flushed and have no
1649 * page cache safely. Once we have done this we can take the ilocks and
1650 * do the rest of the checks.
1651 */
1652 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1653 filemap_invalidate_lock_two(VFS_I(ip)->i_mapping,
1654 VFS_I(tip)->i_mapping);
1655
1656 /* Verify that both files have the same format */
1657 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1658 error = -EINVAL;
1659 goto out_unlock;
1660 }
1661
1662 /* Verify both files are either real-time or non-realtime */
1663 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1664 error = -EINVAL;
1665 goto out_unlock;
1666 }
1667
1668 error = xfs_qm_dqattach(ip);
1669 if (error)
1670 goto out_unlock;
1671
1672 error = xfs_qm_dqattach(tip);
1673 if (error)
1674 goto out_unlock;
1675
1676 error = xfs_swap_extent_flush(ip);
1677 if (error)
1678 goto out_unlock;
1679 error = xfs_swap_extent_flush(tip);
1680 if (error)
1681 goto out_unlock;
1682
1683 if (xfs_inode_has_cow_data(tip)) {
1684 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1685 if (error)
1686 goto out_unlock;
1687 }
1688
1689 /*
1690 * Extent "swapping" with rmap requires a permanent reservation and
1691 * a block reservation because it's really just a remap operation
1692 * performed with log redo items!
1693 */
1694 if (xfs_has_rmapbt(mp)) {
1695 int w = XFS_DATA_FORK;
1696 uint32_t ipnext = ip->i_df.if_nextents;
1697 uint32_t tipnext = tip->i_df.if_nextents;
1698
1699 /*
1700 * Conceptually this shouldn't affect the shape of either bmbt,
1701 * but since we atomically move extents one by one, we reserve
1702 * enough space to rebuild both trees.
1703 */
1704 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1705 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1706
1707 /*
1708 * If either inode straddles a bmapbt block allocation boundary,
1709 * the rmapbt algorithm triggers repeated allocs and frees as
1710 * extents are remapped. This can exhaust the block reservation
1711 * prematurely and cause shutdown. Return freed blocks to the
1712 * transaction reservation to counter this behavior.
1713 */
1714 flags |= XFS_TRANS_RES_FDBLKS;
1715 }
1716 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1717 &tp);
1718 if (error)
1719 goto out_unlock;
1720
1721 /*
1722 * Lock and join the inodes to the tansaction so that transaction commit
1723 * or cancel will unlock the inodes from this point onwards.
1724 */
1725 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1726 xfs_trans_ijoin(tp, ip, 0);
1727 xfs_trans_ijoin(tp, tip, 0);
1728
1729
1730 /* Verify all data are being swapped */
1731 if (sxp->sx_offset != 0 ||
1732 sxp->sx_length != ip->i_disk_size ||
1733 sxp->sx_length != tip->i_disk_size) {
1734 error = -EFAULT;
1735 goto out_trans_cancel;
1736 }
1737
1738 trace_xfs_swap_extent_before(ip, 0);
1739 trace_xfs_swap_extent_before(tip, 1);
1740
1741 /* check inode formats now that data is flushed */
1742 error = xfs_swap_extents_check_format(ip, tip);
1743 if (error) {
1744 xfs_notice(mp,
1745 "%s: inode 0x%llx format is incompatible for exchanging.",
1746 __func__, ip->i_ino);
1747 goto out_trans_cancel;
1748 }
1749
1750 /*
1751 * Compare the current change & modify times with that
1752 * passed in. If they differ, we abort this swap.
1753 * This is the mechanism used to ensure the calling
1754 * process that the file was not changed out from
1755 * under it.
1756 */
1757 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1758 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1759 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1760 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1761 error = -EBUSY;
1762 goto out_trans_cancel;
1763 }
1764
1765 /*
1766 * Note the trickiness in setting the log flags - we set the owner log
1767 * flag on the opposite inode (i.e. the inode we are setting the new
1768 * owner to be) because once we swap the forks and log that, log
1769 * recovery is going to see the fork as owned by the swapped inode,
1770 * not the pre-swapped inodes.
1771 */
1772 src_log_flags = XFS_ILOG_CORE;
1773 target_log_flags = XFS_ILOG_CORE;
1774
1775 if (xfs_has_rmapbt(mp))
1776 error = xfs_swap_extent_rmap(&tp, ip, tip);
1777 else
1778 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1779 &target_log_flags);
1780 if (error)
1781 goto out_trans_cancel;
1782
1783 /* Do we have to swap reflink flags? */
1784 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
1785 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
1786 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1787 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1788 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1789 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1790 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
1791 }
1792
1793 /* Swap the cow forks. */
1794 if (xfs_has_reflink(mp)) {
1795 ASSERT(!ip->i_cowfp ||
1796 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1797 ASSERT(!tip->i_cowfp ||
1798 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1799
1800 swap(ip->i_cowfp, tip->i_cowfp);
1801
1802 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1803 xfs_inode_set_cowblocks_tag(ip);
1804 else
1805 xfs_inode_clear_cowblocks_tag(ip);
1806 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1807 xfs_inode_set_cowblocks_tag(tip);
1808 else
1809 xfs_inode_clear_cowblocks_tag(tip);
1810 }
1811
1812 xfs_trans_log_inode(tp, ip, src_log_flags);
1813 xfs_trans_log_inode(tp, tip, target_log_flags);
1814
1815 /*
1816 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1817 * have inode number owner values in the bmbt blocks that still refer to
1818 * the old inode. Scan each bmbt to fix up the owner values with the
1819 * inode number of the current inode.
1820 */
1821 if (src_log_flags & XFS_ILOG_DOWNER) {
1822 error = xfs_swap_change_owner(&tp, ip, tip);
1823 if (error)
1824 goto out_trans_cancel;
1825 }
1826 if (target_log_flags & XFS_ILOG_DOWNER) {
1827 error = xfs_swap_change_owner(&tp, tip, ip);
1828 if (error)
1829 goto out_trans_cancel;
1830 }
1831
1832 /*
1833 * If this is a synchronous mount, make sure that the
1834 * transaction goes to disk before returning to the user.
1835 */
1836 if (xfs_has_wsync(mp))
1837 xfs_trans_set_sync(tp);
1838
1839 error = xfs_trans_commit(tp);
1840
1841 trace_xfs_swap_extent_after(ip, 0);
1842 trace_xfs_swap_extent_after(tip, 1);
1843
1844 out_unlock_ilock:
1845 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1846 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1847 out_unlock:
1848 filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping,
1849 VFS_I(tip)->i_mapping);
1850 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1851 return error;
1852
1853 out_trans_cancel:
1854 xfs_trans_cancel(tp);
1855 goto out_unlock_ilock;
1856 }
1857