Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0+
2 : /*
3 : * Copyright (C) 2016 Oracle. All Rights Reserved.
4 : * Author: Darrick J. Wong <darrick.wong@oracle.com>
5 : */
6 : #include "xfs.h"
7 : #include "xfs_fs.h"
8 : #include "xfs_shared.h"
9 : #include "xfs_format.h"
10 : #include "xfs_log_format.h"
11 : #include "xfs_trans_resv.h"
12 : #include "xfs_mount.h"
13 : #include "xfs_defer.h"
14 : #include "xfs_btree.h"
15 : #include "xfs_bmap.h"
16 : #include "xfs_refcount_btree.h"
17 : #include "xfs_alloc.h"
18 : #include "xfs_errortag.h"
19 : #include "xfs_error.h"
20 : #include "xfs_trace.h"
21 : #include "xfs_trans.h"
22 : #include "xfs_bit.h"
23 : #include "xfs_refcount.h"
24 : #include "xfs_rmap.h"
25 : #include "xfs_ag.h"
26 : #include "xfs_health.h"
27 : #include "xfs_rtgroup.h"
28 : #include "xfs_rtalloc.h"
29 : #include "xfs_rtrefcount_btree.h"
30 :
31 : struct kmem_cache *xfs_refcount_intent_cache;
32 :
33 : /* Allowable refcount adjustment amounts. */
34 : enum xfs_refc_adjust_op {
35 : XFS_REFCOUNT_ADJUST_INCREASE = 1,
36 : XFS_REFCOUNT_ADJUST_DECREASE = -1,
37 : XFS_REFCOUNT_ADJUST_COW_ALLOC = 0,
38 : XFS_REFCOUNT_ADJUST_COW_FREE = -1,
39 : };
40 :
41 : STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
42 : xfs_agblock_t agbno, xfs_extlen_t aglen);
43 : STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
44 : xfs_agblock_t agbno, xfs_extlen_t aglen);
45 :
46 : /* Return the maximum startblock number of the refcountbt. */
47 : static inline xfs_agblock_t
48 : xrefc_max_startblock(
49 : struct xfs_btree_cur *cur)
50 : {
51 27715245 : if (cur->bc_btnum == XFS_BTNUM_RTREFC)
52 8571772 : return cur->bc_mp->m_sb.sb_rgblocks;
53 19143473 : return cur->bc_mp->m_sb.sb_agblocks;
54 : }
55 :
56 : /*
57 : * Look up the first record less than or equal to [bno, len] in the btree
58 : * given by cur.
59 : */
60 : int
61 3845981807 : xfs_refcount_lookup_le(
62 : struct xfs_btree_cur *cur,
63 : enum xfs_refc_domain domain,
64 : xfs_agblock_t bno,
65 : int *stat)
66 : {
67 3934859906 : trace_xfs_refcount_lookup(cur,
68 : xfs_refcount_encode_startblock(bno, domain),
69 : XFS_LOOKUP_LE);
70 3844203593 : cur->bc_rec.rc.rc_startblock = bno;
71 3844203593 : cur->bc_rec.rc.rc_blockcount = 0;
72 3844203593 : cur->bc_rec.rc.rc_domain = domain;
73 3844203593 : return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
74 : }
75 :
76 : /*
77 : * Look up the first record greater than or equal to [bno, len] in the btree
78 : * given by cur.
79 : */
80 : int
81 667794530 : xfs_refcount_lookup_ge(
82 : struct xfs_btree_cur *cur,
83 : enum xfs_refc_domain domain,
84 : xfs_agblock_t bno,
85 : int *stat)
86 : {
87 739939302 : trace_xfs_refcount_lookup(cur,
88 : xfs_refcount_encode_startblock(bno, domain),
89 : XFS_LOOKUP_GE);
90 667794641 : cur->bc_rec.rc.rc_startblock = bno;
91 667794641 : cur->bc_rec.rc.rc_blockcount = 0;
92 667794641 : cur->bc_rec.rc.rc_domain = domain;
93 667794641 : return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
94 : }
95 :
96 : /*
97 : * Look up the first record equal to [bno, len] in the btree
98 : * given by cur.
99 : */
100 : int
101 0 : xfs_refcount_lookup_eq(
102 : struct xfs_btree_cur *cur,
103 : enum xfs_refc_domain domain,
104 : xfs_agblock_t bno,
105 : int *stat)
106 : {
107 0 : trace_xfs_refcount_lookup(cur,
108 : xfs_refcount_encode_startblock(bno, domain),
109 : XFS_LOOKUP_LE);
110 0 : cur->bc_rec.rc.rc_startblock = bno;
111 0 : cur->bc_rec.rc.rc_blockcount = 0;
112 0 : cur->bc_rec.rc.rc_domain = domain;
113 0 : return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
114 : }
115 :
116 : /* Convert on-disk record to in-core format. */
117 : void
118 5266479924 : xfs_refcount_btrec_to_irec(
119 : const union xfs_btree_rec *rec,
120 : struct xfs_refcount_irec *irec)
121 : {
122 5266479924 : uint32_t start;
123 :
124 5266479924 : start = be32_to_cpu(rec->refc.rc_startblock);
125 5266479924 : if (start & XFS_REFC_COWFLAG) {
126 251578958 : start &= ~XFS_REFC_COWFLAG;
127 251578958 : irec->rc_domain = XFS_REFC_DOMAIN_COW;
128 : } else {
129 5014900966 : irec->rc_domain = XFS_REFC_DOMAIN_SHARED;
130 : }
131 :
132 5266479924 : irec->rc_startblock = start;
133 5266479924 : irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
134 5266479924 : irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
135 5266479924 : }
136 :
137 : inline xfs_failaddr_t
138 3129360573 : xfs_refcount_check_perag_irec(
139 : struct xfs_perag *pag,
140 : const struct xfs_refcount_irec *irec)
141 : {
142 3129360573 : if (irec->rc_blockcount == 0 || irec->rc_blockcount > XFS_REFC_LEN_MAX)
143 0 : return __this_address;
144 :
145 3129360573 : if (!xfs_refcount_check_domain(irec))
146 0 : return __this_address;
147 :
148 : /* check for valid extent range, including overflow */
149 3129360573 : if (!xfs_verify_agbext(pag, irec->rc_startblock, irec->rc_blockcount))
150 10 : return __this_address;
151 :
152 3129360563 : if (irec->rc_refcount == 0 || irec->rc_refcount > XFS_REFC_REFCOUNT_MAX)
153 0 : return __this_address;
154 :
155 : return NULL;
156 : }
157 :
158 : inline xfs_failaddr_t
159 2150235325 : xfs_refcount_check_rtgroup_irec(
160 : struct xfs_rtgroup *rtg,
161 : const struct xfs_refcount_irec *irec)
162 : {
163 2150235325 : if (irec->rc_blockcount == 0 || irec->rc_blockcount > XFS_REFC_LEN_MAX)
164 0 : return __this_address;
165 :
166 2150235325 : if (!xfs_refcount_check_domain(irec))
167 0 : return __this_address;
168 :
169 : /* check for valid extent range, including overflow */
170 2150235325 : if (!xfs_verify_rgbext(rtg, irec->rc_startblock, irec->rc_blockcount))
171 0 : return __this_address;
172 :
173 2150235325 : if (irec->rc_refcount == 0 || irec->rc_refcount > XFS_REFC_REFCOUNT_MAX)
174 0 : return __this_address;
175 :
176 : return NULL;
177 : }
178 :
179 : /* Simple checks for refcount records. */
180 : xfs_failaddr_t
181 5266521087 : xfs_refcount_check_irec(
182 : struct xfs_btree_cur *cur,
183 : const struct xfs_refcount_irec *irec)
184 : {
185 5266521087 : if (cur->bc_btnum == XFS_BTNUM_RTREFC)
186 2144478694 : return xfs_refcount_check_rtgroup_irec(cur->bc_ino.rtg, irec);
187 3122042393 : return xfs_refcount_check_perag_irec(cur->bc_ag.pag, irec);
188 : }
189 :
190 : static inline int
191 10 : xfs_refcount_complain_bad_rec(
192 : struct xfs_btree_cur *cur,
193 : xfs_failaddr_t fa,
194 : const struct xfs_refcount_irec *irec)
195 : {
196 10 : struct xfs_mount *mp = cur->bc_mp;
197 :
198 10 : if (cur->bc_btnum == XFS_BTNUM_RTREFC) {
199 0 : xfs_warn(mp,
200 : "RT Refcount BTree record corruption in rtgroup %u detected at %pS!",
201 : cur->bc_ino.rtg->rtg_rgno, fa);
202 : } else {
203 10 : xfs_warn(mp,
204 : "Refcount BTree record corruption in AG %d detected at %pS!",
205 : cur->bc_ag.pag->pag_agno, fa);
206 : }
207 10 : xfs_warn(mp,
208 : "Start block 0x%x, block count 0x%x, references 0x%x",
209 : irec->rc_startblock, irec->rc_blockcount, irec->rc_refcount);
210 10 : xfs_btree_mark_sick(cur);
211 10 : return -EFSCORRUPTED;
212 : }
213 :
214 : /*
215 : * Get the data from the pointed-to record.
216 : */
217 : int
218 5166512020 : xfs_refcount_get_rec(
219 : struct xfs_btree_cur *cur,
220 : struct xfs_refcount_irec *irec,
221 : int *stat)
222 : {
223 5166512020 : union xfs_btree_rec *rec;
224 5166512020 : xfs_failaddr_t fa;
225 5166512020 : int error;
226 :
227 5166512020 : error = xfs_btree_get_rec(cur, &rec, stat);
228 5162685367 : if (error || !*stat)
229 : return error;
230 :
231 5145916506 : xfs_refcount_btrec_to_irec(rec, irec);
232 5146756821 : fa = xfs_refcount_check_irec(cur, irec);
233 5146434750 : if (fa)
234 10 : return xfs_refcount_complain_bad_rec(cur, fa, irec);
235 :
236 5146434740 : trace_xfs_refcount_get(cur, irec);
237 5146434740 : return 0;
238 : }
239 :
240 : /*
241 : * Update the record referred to by cur to the value given
242 : * by [bno, len, refcount].
243 : * This either works (return 0) or gets an EFSCORRUPTED error.
244 : */
245 : STATIC int
246 259987250 : xfs_refcount_update(
247 : struct xfs_btree_cur *cur,
248 : struct xfs_refcount_irec *irec)
249 : {
250 259987250 : union xfs_btree_rec rec;
251 259987250 : uint32_t start;
252 259987250 : int error;
253 :
254 259987250 : trace_xfs_refcount_update(cur, irec);
255 :
256 259986920 : start = xfs_refcount_encode_startblock(irec->rc_startblock,
257 : irec->rc_domain);
258 259986920 : rec.refc.rc_startblock = cpu_to_be32(start);
259 259986920 : rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
260 259986920 : rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
261 :
262 259986920 : error = xfs_btree_update(cur, &rec);
263 259987289 : if (error)
264 0 : trace_xfs_refcount_update_error(cur, error, _RET_IP_);
265 259987289 : return error;
266 : }
267 :
268 : /*
269 : * Insert the record referred to by cur to the value given
270 : * by [bno, len, refcount].
271 : * This either works (return 0) or gets an EFSCORRUPTED error.
272 : */
273 : int
274 99937231 : xfs_refcount_insert(
275 : struct xfs_btree_cur *cur,
276 : struct xfs_refcount_irec *irec,
277 : int *i)
278 : {
279 99937231 : int error;
280 :
281 99937231 : trace_xfs_refcount_insert(cur, irec);
282 :
283 99936948 : cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
284 99936948 : cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
285 99936948 : cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
286 99936948 : cur->bc_rec.rc.rc_domain = irec->rc_domain;
287 :
288 99936948 : error = xfs_btree_insert(cur, i);
289 99937124 : if (error)
290 70 : goto out_error;
291 99937054 : if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
292 0 : xfs_btree_mark_sick(cur);
293 0 : error = -EFSCORRUPTED;
294 0 : goto out_error;
295 : }
296 :
297 99937124 : out_error:
298 99937124 : if (error)
299 70 : trace_xfs_refcount_insert_error(cur, error, _RET_IP_);
300 99937124 : return error;
301 : }
302 :
303 : /*
304 : * Remove the record referred to by cur, then set the pointer to the spot
305 : * where the record could be re-inserted, in case we want to increment or
306 : * decrement the cursor.
307 : * This either works (return 0) or gets an EFSCORRUPTED error.
308 : */
309 : STATIC int
310 89774871 : xfs_refcount_delete(
311 : struct xfs_btree_cur *cur,
312 : int *i)
313 : {
314 89774871 : struct xfs_refcount_irec irec;
315 89774871 : int found_rec;
316 89774871 : int error;
317 :
318 89774871 : error = xfs_refcount_get_rec(cur, &irec, &found_rec);
319 89776782 : if (error)
320 0 : goto out_error;
321 89776782 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
322 0 : xfs_btree_mark_sick(cur);
323 0 : error = -EFSCORRUPTED;
324 0 : goto out_error;
325 : }
326 89776782 : trace_xfs_refcount_delete(cur, &irec);
327 89776261 : error = xfs_btree_delete(cur, i);
328 89774191 : if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
329 0 : xfs_btree_mark_sick(cur);
330 0 : error = -EFSCORRUPTED;
331 0 : goto out_error;
332 : }
333 89774191 : if (error)
334 0 : goto out_error;
335 89774191 : error = xfs_refcount_lookup_ge(cur, irec.rc_domain, irec.rc_startblock,
336 : &found_rec);
337 89776691 : out_error:
338 89776691 : if (error)
339 0 : trace_xfs_refcount_delete_error(cur, error, _RET_IP_);
340 89776691 : return error;
341 : }
342 :
343 : /*
344 : * Adjusting the Reference Count
345 : *
346 : * As stated elsewhere, the reference count btree (refcbt) stores
347 : * >1 reference counts for extents of physical blocks. In this
348 : * operation, we're either raising or lowering the reference count of
349 : * some subrange stored in the tree:
350 : *
351 : * <------ adjustment range ------>
352 : * ----+ +---+-----+ +--+--------+---------
353 : * 2 | | 3 | 4 | |17| 55 | 10
354 : * ----+ +---+-----+ +--+--------+---------
355 : * X axis is physical blocks number;
356 : * reference counts are the numbers inside the rectangles
357 : *
358 : * The first thing we need to do is to ensure that there are no
359 : * refcount extents crossing either boundary of the range to be
360 : * adjusted. For any extent that does cross a boundary, split it into
361 : * two extents so that we can increment the refcount of one of the
362 : * pieces later:
363 : *
364 : * <------ adjustment range ------>
365 : * ----+ +---+-----+ +--+--------+----+----
366 : * 2 | | 3 | 2 | |17| 55 | 10 | 10
367 : * ----+ +---+-----+ +--+--------+----+----
368 : *
369 : * For this next step, let's assume that all the physical blocks in
370 : * the adjustment range are mapped to a file and are therefore in use
371 : * at least once. Therefore, we can infer that any gap in the
372 : * refcount tree within the adjustment range represents a physical
373 : * extent with refcount == 1:
374 : *
375 : * <------ adjustment range ------>
376 : * ----+---+---+-----+-+--+--------+----+----
377 : * 2 |"1"| 3 | 2 |1|17| 55 | 10 | 10
378 : * ----+---+---+-----+-+--+--------+----+----
379 : * ^
380 : *
381 : * For each extent that falls within the interval range, figure out
382 : * which extent is to the left or the right of that extent. Now we
383 : * have a left, current, and right extent. If the new reference count
384 : * of the center extent enables us to merge left, center, and right
385 : * into one record covering all three, do so. If the center extent is
386 : * at the left end of the range, abuts the left extent, and its new
387 : * reference count matches the left extent's record, then merge them.
388 : * If the center extent is at the right end of the range, abuts the
389 : * right extent, and the reference counts match, merge those. In the
390 : * example, we can left merge (assuming an increment operation):
391 : *
392 : * <------ adjustment range ------>
393 : * --------+---+-----+-+--+--------+----+----
394 : * 2 | 3 | 2 |1|17| 55 | 10 | 10
395 : * --------+---+-----+-+--+--------+----+----
396 : * ^
397 : *
398 : * For all other extents within the range, adjust the reference count
399 : * or delete it if the refcount falls below 2. If we were
400 : * incrementing, the end result looks like this:
401 : *
402 : * <------ adjustment range ------>
403 : * --------+---+-----+-+--+--------+----+----
404 : * 2 | 4 | 3 |2|18| 56 | 11 | 10
405 : * --------+---+-----+-+--+--------+----+----
406 : *
407 : * The result of a decrement operation looks as such:
408 : *
409 : * <------ adjustment range ------>
410 : * ----+ +---+ +--+--------+----+----
411 : * 2 | | 2 | |16| 54 | 9 | 10
412 : * ----+ +---+ +--+--------+----+----
413 : * DDDD 111111DD
414 : *
415 : * The blocks marked "D" are freed; the blocks marked "1" are only
416 : * referenced once and therefore the record is removed from the
417 : * refcount btree.
418 : */
419 :
420 : /* Next block after this extent. */
421 : static inline xfs_agblock_t
422 : xfs_refc_next(
423 : struct xfs_refcount_irec *rc)
424 : {
425 696996806 : return rc->rc_startblock + rc->rc_blockcount;
426 : }
427 :
428 : /*
429 : * Split a refcount extent that crosses agbno.
430 : */
431 : STATIC int
432 589406151 : xfs_refcount_split_extent(
433 : struct xfs_btree_cur *cur,
434 : enum xfs_refc_domain domain,
435 : xfs_agblock_t agbno,
436 : bool *shape_changed)
437 : {
438 589406151 : struct xfs_refcount_irec rcext, tmp;
439 589406151 : int found_rec;
440 589406151 : int error;
441 :
442 589406151 : *shape_changed = false;
443 589406151 : error = xfs_refcount_lookup_le(cur, domain, agbno, &found_rec);
444 589420148 : if (error)
445 174 : goto out_error;
446 589419974 : if (!found_rec)
447 : return 0;
448 :
449 568012690 : error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
450 568003824 : if (error)
451 0 : goto out_error;
452 568003824 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
453 0 : xfs_btree_mark_sick(cur);
454 0 : error = -EFSCORRUPTED;
455 0 : goto out_error;
456 : }
457 568003824 : if (rcext.rc_domain != domain)
458 : return 0;
459 566757381 : if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
460 : return 0;
461 :
462 64453308 : *shape_changed = true;
463 64453308 : trace_xfs_refcount_split_extent(cur, &rcext, agbno);
464 :
465 : /* Establish the right extent. */
466 64453209 : tmp = rcext;
467 64453209 : tmp.rc_startblock = agbno;
468 64453209 : tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
469 64453209 : error = xfs_refcount_update(cur, &tmp);
470 64453236 : if (error)
471 0 : goto out_error;
472 :
473 : /* Insert the left extent. */
474 64453236 : tmp = rcext;
475 64453236 : tmp.rc_blockcount = agbno - rcext.rc_startblock;
476 64453236 : error = xfs_refcount_insert(cur, &tmp, &found_rec);
477 64453140 : if (error)
478 31 : goto out_error;
479 64453109 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
480 0 : xfs_btree_mark_sick(cur);
481 0 : error = -EFSCORRUPTED;
482 0 : goto out_error;
483 : }
484 : return error;
485 :
486 205 : out_error:
487 205 : trace_xfs_refcount_split_extent_error(cur, error, _RET_IP_);
488 205 : return error;
489 : }
490 :
491 : /*
492 : * Merge the left, center, and right extents.
493 : */
494 : STATIC int
495 2525919 : xfs_refcount_merge_center_extents(
496 : struct xfs_btree_cur *cur,
497 : struct xfs_refcount_irec *left,
498 : struct xfs_refcount_irec *center,
499 : struct xfs_refcount_irec *right,
500 : unsigned long long extlen,
501 : xfs_extlen_t *aglen)
502 : {
503 2525919 : int error;
504 2525919 : int found_rec;
505 :
506 2525919 : trace_xfs_refcount_merge_center_extents(cur, left, center, right);
507 :
508 2525919 : ASSERT(left->rc_domain == center->rc_domain);
509 2525919 : ASSERT(right->rc_domain == center->rc_domain);
510 :
511 : /*
512 : * Make sure the center and right extents are not in the btree.
513 : * If the center extent was synthesized, the first delete call
514 : * removes the right extent and we skip the second deletion.
515 : * If center and right were in the btree, then the first delete
516 : * call removes the center and the second one removes the right
517 : * extent.
518 : */
519 2525919 : error = xfs_refcount_lookup_ge(cur, center->rc_domain,
520 : center->rc_startblock, &found_rec);
521 2525919 : if (error)
522 0 : goto out_error;
523 2525919 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
524 0 : xfs_btree_mark_sick(cur);
525 0 : error = -EFSCORRUPTED;
526 0 : goto out_error;
527 : }
528 :
529 2525919 : error = xfs_refcount_delete(cur, &found_rec);
530 2525918 : if (error)
531 0 : goto out_error;
532 2525918 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
533 0 : xfs_btree_mark_sick(cur);
534 0 : error = -EFSCORRUPTED;
535 0 : goto out_error;
536 : }
537 :
538 2525918 : if (center->rc_refcount > 1) {
539 660864 : error = xfs_refcount_delete(cur, &found_rec);
540 660864 : if (error)
541 0 : goto out_error;
542 660864 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
543 0 : xfs_btree_mark_sick(cur);
544 0 : error = -EFSCORRUPTED;
545 0 : goto out_error;
546 : }
547 : }
548 :
549 : /* Enlarge the left extent. */
550 2525918 : error = xfs_refcount_lookup_le(cur, left->rc_domain,
551 : left->rc_startblock, &found_rec);
552 2525918 : if (error)
553 0 : goto out_error;
554 2525918 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
555 0 : xfs_btree_mark_sick(cur);
556 0 : error = -EFSCORRUPTED;
557 0 : goto out_error;
558 : }
559 :
560 2525918 : left->rc_blockcount = extlen;
561 2525918 : error = xfs_refcount_update(cur, left);
562 2525919 : if (error)
563 0 : goto out_error;
564 :
565 2525919 : *aglen = 0;
566 2525919 : return error;
567 :
568 0 : out_error:
569 0 : trace_xfs_refcount_merge_center_extents_error(cur, error, _RET_IP_);
570 0 : return error;
571 : }
572 :
573 : /*
574 : * Merge with the left extent.
575 : */
576 : STATIC int
577 16865594 : xfs_refcount_merge_left_extent(
578 : struct xfs_btree_cur *cur,
579 : struct xfs_refcount_irec *left,
580 : struct xfs_refcount_irec *cleft,
581 : xfs_agblock_t *agbno,
582 : xfs_extlen_t *aglen)
583 : {
584 16865594 : int error;
585 16865594 : int found_rec;
586 :
587 16865594 : trace_xfs_refcount_merge_left_extent(cur, left, cleft);
588 :
589 16865592 : ASSERT(left->rc_domain == cleft->rc_domain);
590 :
591 : /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
592 16865592 : if (cleft->rc_refcount > 1) {
593 10491408 : error = xfs_refcount_lookup_le(cur, cleft->rc_domain,
594 : cleft->rc_startblock, &found_rec);
595 10491408 : if (error)
596 0 : goto out_error;
597 10491408 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
598 0 : xfs_btree_mark_sick(cur);
599 0 : error = -EFSCORRUPTED;
600 0 : goto out_error;
601 : }
602 :
603 10491408 : error = xfs_refcount_delete(cur, &found_rec);
604 10491408 : if (error)
605 0 : goto out_error;
606 10491408 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
607 0 : xfs_btree_mark_sick(cur);
608 0 : error = -EFSCORRUPTED;
609 0 : goto out_error;
610 : }
611 : }
612 :
613 : /* Enlarge the left extent. */
614 16865592 : error = xfs_refcount_lookup_le(cur, left->rc_domain,
615 : left->rc_startblock, &found_rec);
616 16865629 : if (error)
617 0 : goto out_error;
618 16865629 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
619 0 : xfs_btree_mark_sick(cur);
620 0 : error = -EFSCORRUPTED;
621 0 : goto out_error;
622 : }
623 :
624 16865629 : left->rc_blockcount += cleft->rc_blockcount;
625 16865629 : error = xfs_refcount_update(cur, left);
626 16865620 : if (error)
627 0 : goto out_error;
628 :
629 16865620 : *agbno += cleft->rc_blockcount;
630 16865620 : *aglen -= cleft->rc_blockcount;
631 16865620 : return error;
632 :
633 0 : out_error:
634 0 : trace_xfs_refcount_merge_left_extent_error(cur, error, _RET_IP_);
635 0 : return error;
636 : }
637 :
638 : /*
639 : * Merge with the right extent.
640 : */
641 : STATIC int
642 4584383 : xfs_refcount_merge_right_extent(
643 : struct xfs_btree_cur *cur,
644 : struct xfs_refcount_irec *right,
645 : struct xfs_refcount_irec *cright,
646 : xfs_extlen_t *aglen)
647 : {
648 4584383 : int error;
649 4584383 : int found_rec;
650 :
651 4584383 : trace_xfs_refcount_merge_right_extent(cur, cright, right);
652 :
653 4584383 : ASSERT(right->rc_domain == cright->rc_domain);
654 :
655 : /*
656 : * If the extent ending at agbno+aglen (cright) wasn't synthesized,
657 : * remove it.
658 : */
659 4584383 : if (cright->rc_refcount > 1) {
660 2364599 : error = xfs_refcount_lookup_le(cur, cright->rc_domain,
661 : cright->rc_startblock, &found_rec);
662 2364599 : if (error)
663 0 : goto out_error;
664 2364599 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
665 0 : xfs_btree_mark_sick(cur);
666 0 : error = -EFSCORRUPTED;
667 0 : goto out_error;
668 : }
669 :
670 2364599 : error = xfs_refcount_delete(cur, &found_rec);
671 2364599 : if (error)
672 0 : goto out_error;
673 2364599 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
674 0 : xfs_btree_mark_sick(cur);
675 0 : error = -EFSCORRUPTED;
676 0 : goto out_error;
677 : }
678 : }
679 :
680 : /* Enlarge the right extent. */
681 4584383 : error = xfs_refcount_lookup_le(cur, right->rc_domain,
682 : right->rc_startblock, &found_rec);
683 4584384 : if (error)
684 0 : goto out_error;
685 4584384 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
686 0 : xfs_btree_mark_sick(cur);
687 0 : error = -EFSCORRUPTED;
688 0 : goto out_error;
689 : }
690 :
691 4584384 : right->rc_startblock -= cright->rc_blockcount;
692 4584384 : right->rc_blockcount += cright->rc_blockcount;
693 4584384 : error = xfs_refcount_update(cur, right);
694 4584384 : if (error)
695 0 : goto out_error;
696 :
697 4584384 : *aglen -= cright->rc_blockcount;
698 4584384 : return error;
699 :
700 0 : out_error:
701 0 : trace_xfs_refcount_merge_right_extent_error(cur, error, _RET_IP_);
702 0 : return error;
703 : }
704 :
705 : /*
706 : * Find the left extent and the one after it (cleft). This function assumes
707 : * that we've already split any extent crossing agbno.
708 : */
709 : STATIC int
710 294709111 : xfs_refcount_find_left_extents(
711 : struct xfs_btree_cur *cur,
712 : struct xfs_refcount_irec *left,
713 : struct xfs_refcount_irec *cleft,
714 : enum xfs_refc_domain domain,
715 : xfs_agblock_t agbno,
716 : xfs_extlen_t aglen)
717 : {
718 294709111 : struct xfs_refcount_irec tmp;
719 294709111 : int error;
720 294709111 : int found_rec;
721 :
722 294709111 : left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
723 294709111 : error = xfs_refcount_lookup_le(cur, domain, agbno - 1, &found_rec);
724 294711420 : if (error)
725 0 : goto out_error;
726 294711420 : if (!found_rec)
727 : return 0;
728 :
729 250611120 : error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
730 250611131 : if (error)
731 0 : goto out_error;
732 250611131 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
733 0 : xfs_btree_mark_sick(cur);
734 0 : error = -EFSCORRUPTED;
735 0 : goto out_error;
736 : }
737 :
738 250611131 : if (tmp.rc_domain != domain)
739 : return 0;
740 247989654 : if (xfs_refc_next(&tmp) != agbno)
741 : return 0;
742 : /* We have a left extent; retrieve (or invent) the next right one */
743 68139542 : *left = tmp;
744 :
745 68139542 : error = xfs_btree_increment(cur, 0, &found_rec);
746 68139491 : if (error)
747 0 : goto out_error;
748 68139491 : if (found_rec) {
749 66352235 : error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
750 66352292 : if (error)
751 0 : goto out_error;
752 66352292 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
753 0 : xfs_btree_mark_sick(cur);
754 0 : error = -EFSCORRUPTED;
755 0 : goto out_error;
756 : }
757 :
758 66352292 : if (tmp.rc_domain != domain)
759 271240 : goto not_found;
760 :
761 : /* if tmp starts at the end of our range, just use that */
762 66081052 : if (tmp.rc_startblock == agbno)
763 59299193 : *cleft = tmp;
764 : else {
765 : /*
766 : * There's a gap in the refcntbt at the start of the
767 : * range we're interested in (refcount == 1) so
768 : * synthesize the implied extent and pass it back.
769 : * We assume here that the agbno/aglen range was
770 : * passed in from a data fork extent mapping and
771 : * therefore is allocated to exactly one owner.
772 : */
773 6781859 : cleft->rc_startblock = agbno;
774 6781859 : cleft->rc_blockcount = min(aglen,
775 : tmp.rc_startblock - agbno);
776 6781859 : cleft->rc_refcount = 1;
777 6781859 : cleft->rc_domain = domain;
778 : }
779 : } else {
780 1787256 : not_found:
781 : /*
782 : * No extents, so pretend that there's one covering the whole
783 : * range.
784 : */
785 2058496 : cleft->rc_startblock = agbno;
786 2058496 : cleft->rc_blockcount = aglen;
787 2058496 : cleft->rc_refcount = 1;
788 2058496 : cleft->rc_domain = domain;
789 : }
790 68139548 : trace_xfs_refcount_find_left_extent(cur, left, cleft, agbno);
791 68139548 : return error;
792 :
793 0 : out_error:
794 0 : trace_xfs_refcount_find_left_extent_error(cur, error, _RET_IP_);
795 0 : return error;
796 : }
797 :
798 : /*
799 : * Find the right extent and the one before it (cright). This function
800 : * assumes that we've already split any extents crossing agbno + aglen.
801 : */
802 : STATIC int
803 294709911 : xfs_refcount_find_right_extents(
804 : struct xfs_btree_cur *cur,
805 : struct xfs_refcount_irec *right,
806 : struct xfs_refcount_irec *cright,
807 : enum xfs_refc_domain domain,
808 : xfs_agblock_t agbno,
809 : xfs_extlen_t aglen)
810 : {
811 294709911 : struct xfs_refcount_irec tmp;
812 294709911 : int error;
813 294709911 : int found_rec;
814 :
815 294709911 : right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
816 294709911 : error = xfs_refcount_lookup_ge(cur, domain, agbno + aglen, &found_rec);
817 294711098 : if (error)
818 0 : goto out_error;
819 294711098 : if (!found_rec)
820 : return 0;
821 :
822 241838087 : error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
823 241838468 : if (error)
824 10 : goto out_error;
825 241838458 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
826 0 : xfs_btree_mark_sick(cur);
827 0 : error = -EFSCORRUPTED;
828 0 : goto out_error;
829 : }
830 :
831 241838458 : if (tmp.rc_domain != domain)
832 : return 0;
833 229693882 : if (tmp.rc_startblock != agbno + aglen)
834 : return 0;
835 : /* We have a right extent; retrieve (or invent) the next left one */
836 71578481 : *right = tmp;
837 :
838 71578481 : error = xfs_btree_decrement(cur, 0, &found_rec);
839 71578374 : if (error)
840 0 : goto out_error;
841 71578374 : if (found_rec) {
842 71460986 : error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
843 71460929 : if (error)
844 0 : goto out_error;
845 71460929 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
846 0 : xfs_btree_mark_sick(cur);
847 0 : error = -EFSCORRUPTED;
848 0 : goto out_error;
849 : }
850 :
851 71460929 : if (tmp.rc_domain != domain)
852 8319 : goto not_found;
853 :
854 : /* if tmp ends at the end of our range, just use that */
855 71452610 : if (xfs_refc_next(&tmp) == agbno + aglen)
856 66989720 : *cright = tmp;
857 : else {
858 : /*
859 : * There's a gap in the refcntbt at the end of the
860 : * range we're interested in (refcount == 1) so
861 : * create the implied extent and pass it back.
862 : * We assume here that the agbno/aglen range was
863 : * passed in from a data fork extent mapping and
864 : * therefore is allocated to exactly one owner.
865 : */
866 4462890 : cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
867 4462890 : cright->rc_blockcount = right->rc_startblock -
868 : cright->rc_startblock;
869 4462890 : cright->rc_refcount = 1;
870 4462890 : cright->rc_domain = domain;
871 : }
872 : } else {
873 117388 : not_found:
874 : /*
875 : * No extents, so pretend that there's one covering the whole
876 : * range.
877 : */
878 125707 : cright->rc_startblock = agbno;
879 125707 : cright->rc_blockcount = aglen;
880 125707 : cright->rc_refcount = 1;
881 125707 : cright->rc_domain = domain;
882 : }
883 71578317 : trace_xfs_refcount_find_right_extent(cur, cright, right,
884 : agbno + aglen);
885 71578317 : return error;
886 :
887 10 : out_error:
888 10 : trace_xfs_refcount_find_right_extent_error(cur, error, _RET_IP_);
889 10 : return error;
890 : }
891 :
892 : /* Is this extent valid? */
893 : static inline bool
894 : xfs_refc_valid(
895 : const struct xfs_refcount_irec *rc)
896 : {
897 1080185179 : return rc->rc_startblock != NULLAGBLOCK;
898 : }
899 :
900 : static inline xfs_nlink_t
901 : xfs_refc_merge_refcount(
902 : const struct xfs_refcount_irec *irec,
903 : enum xfs_refc_adjust_op adjust)
904 : {
905 : /* Once a record hits XFS_REFC_REFCOUNT_MAX, it is pinned forever */
906 157432367 : if (irec->rc_refcount == XFS_REFC_REFCOUNT_MAX)
907 : return XFS_REFC_REFCOUNT_MAX;
908 157432317 : return irec->rc_refcount + adjust;
909 : }
910 :
911 : static inline bool
912 95462690 : xfs_refc_want_merge_center(
913 : const struct xfs_refcount_irec *left,
914 : const struct xfs_refcount_irec *cleft,
915 : const struct xfs_refcount_irec *cright,
916 : const struct xfs_refcount_irec *right,
917 : bool cleft_is_cright,
918 : enum xfs_refc_adjust_op adjust,
919 : unsigned long long *ulenp)
920 : {
921 95462690 : unsigned long long ulen = left->rc_blockcount;
922 95462690 : xfs_nlink_t new_refcount;
923 :
924 : /*
925 : * To merge with a center record, both shoulder records must be
926 : * adjacent to the record we want to adjust. This is only true if
927 : * find_left and find_right made all four records valid.
928 : */
929 95462690 : if (!xfs_refc_valid(left) || !xfs_refc_valid(right) ||
930 44254933 : !xfs_refc_valid(cleft) || !xfs_refc_valid(cright))
931 : return false;
932 :
933 : /* There must only be one record for the entire range. */
934 44254933 : if (!cleft_is_cright)
935 : return false;
936 :
937 : /* The shoulder record refcounts must match the new refcount. */
938 29640084 : new_refcount = xfs_refc_merge_refcount(cleft, adjust);
939 29640084 : if (left->rc_refcount != new_refcount)
940 : return false;
941 9399289 : if (right->rc_refcount != new_refcount)
942 : return false;
943 :
944 : /*
945 : * The new record cannot exceed the max length. ulen is a ULL as the
946 : * individual record block counts can be up to (u32 - 1) in length
947 : * hence we need to catch u32 addition overflows here.
948 : */
949 2525919 : ulen += cleft->rc_blockcount + right->rc_blockcount;
950 2525919 : if (ulen >= XFS_REFC_LEN_MAX)
951 : return false;
952 :
953 2525919 : *ulenp = ulen;
954 2525919 : return true;
955 : }
956 :
957 : static inline bool
958 92936657 : xfs_refc_want_merge_left(
959 : const struct xfs_refcount_irec *left,
960 : const struct xfs_refcount_irec *cleft,
961 : enum xfs_refc_adjust_op adjust)
962 : {
963 92936657 : unsigned long long ulen = left->rc_blockcount;
964 92936657 : xfs_nlink_t new_refcount;
965 :
966 : /*
967 : * For a left merge, the left shoulder record must be adjacent to the
968 : * start of the range. If this is true, find_left made left and cleft
969 : * contain valid contents.
970 : */
971 92936657 : if (!xfs_refc_valid(left) || !xfs_refc_valid(cleft))
972 : return false;
973 :
974 : /* Left shoulder record refcount must match the new refcount. */
975 65613442 : new_refcount = xfs_refc_merge_refcount(cleft, adjust);
976 65613442 : if (left->rc_refcount != new_refcount)
977 : return false;
978 :
979 : /*
980 : * The new record cannot exceed the max length. ulen is a ULL as the
981 : * individual record block counts can be up to (u32 - 1) in length
982 : * hence we need to catch u32 addition overflows here.
983 : */
984 16865606 : ulen += cleft->rc_blockcount;
985 16865606 : if (ulen >= XFS_REFC_LEN_MAX)
986 0 : return false;
987 :
988 : return true;
989 : }
990 :
991 : static inline bool
992 86063311 : xfs_refc_want_merge_right(
993 : const struct xfs_refcount_irec *cright,
994 : const struct xfs_refcount_irec *right,
995 : enum xfs_refc_adjust_op adjust)
996 : {
997 86063311 : unsigned long long ulen = right->rc_blockcount;
998 86063311 : xfs_nlink_t new_refcount;
999 :
1000 : /*
1001 : * For a right merge, the right shoulder record must be adjacent to the
1002 : * end of the range. If this is true, find_right made cright and right
1003 : * contain valid contents.
1004 : */
1005 86063311 : if (!xfs_refc_valid(right) || !xfs_refc_valid(cright))
1006 : return false;
1007 :
1008 : /* Right shoulder record refcount must match the new refcount. */
1009 62178841 : new_refcount = xfs_refc_merge_refcount(cright, adjust);
1010 62178841 : if (right->rc_refcount != new_refcount)
1011 : return false;
1012 :
1013 : /*
1014 : * The new record cannot exceed the max length. ulen is a ULL as the
1015 : * individual record block counts can be up to (u32 - 1) in length
1016 : * hence we need to catch u32 addition overflows here.
1017 : */
1018 4584384 : ulen += cright->rc_blockcount;
1019 4584384 : if (ulen >= XFS_REFC_LEN_MAX)
1020 0 : return false;
1021 :
1022 : return true;
1023 : }
1024 :
1025 : /*
1026 : * Try to merge with any extents on the boundaries of the adjustment range.
1027 : */
1028 : STATIC int
1029 294709076 : xfs_refcount_merge_extents(
1030 : struct xfs_btree_cur *cur,
1031 : enum xfs_refc_domain domain,
1032 : xfs_agblock_t *agbno,
1033 : xfs_extlen_t *aglen,
1034 : enum xfs_refc_adjust_op adjust,
1035 : bool *shape_changed)
1036 : {
1037 294709076 : struct xfs_refcount_irec left = {0}, cleft = {0};
1038 294709076 : struct xfs_refcount_irec cright = {0}, right = {0};
1039 294709076 : int error;
1040 294709076 : unsigned long long ulen;
1041 294709076 : bool cequal;
1042 :
1043 294709076 : *shape_changed = false;
1044 : /*
1045 : * Find the extent just below agbno [left], just above agbno [cleft],
1046 : * just below (agbno + aglen) [cright], and just above (agbno + aglen)
1047 : * [right].
1048 : */
1049 294709076 : error = xfs_refcount_find_left_extents(cur, &left, &cleft, domain,
1050 : *agbno, *aglen);
1051 294710181 : if (error)
1052 : return error;
1053 294710323 : error = xfs_refcount_find_right_extents(cur, &right, &cright, domain,
1054 : *agbno, *aglen);
1055 294710178 : if (error)
1056 : return error;
1057 :
1058 : /* No left or right extent to merge; exit. */
1059 294710168 : if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
1060 : return 0;
1061 :
1062 95462629 : cequal = (cleft.rc_startblock == cright.rc_startblock) &&
1063 29640088 : (cleft.rc_blockcount == cright.rc_blockcount);
1064 :
1065 : /* Try to merge left, cleft, and right. cleft must == cright. */
1066 95462629 : if (xfs_refc_want_merge_center(&left, &cleft, &cright, &right, cequal,
1067 : adjust, &ulen)) {
1068 2525919 : *shape_changed = true;
1069 2525919 : return xfs_refcount_merge_center_extents(cur, &left, &cleft,
1070 : &right, ulen, aglen);
1071 : }
1072 :
1073 : /* Try to merge left and cleft. */
1074 92936520 : if (xfs_refc_want_merge_left(&left, &cleft, adjust)) {
1075 16865605 : *shape_changed = true;
1076 16865605 : error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
1077 : agbno, aglen);
1078 16865611 : if (error)
1079 : return error;
1080 :
1081 : /*
1082 : * If we just merged left + cleft and cleft == cright,
1083 : * we no longer have a cright to merge with right. We're done.
1084 : */
1085 16865611 : if (cequal)
1086 : return 0;
1087 : }
1088 :
1089 : /* Try to merge cright and right. */
1090 86063156 : if (xfs_refc_want_merge_right(&cright, &right, adjust)) {
1091 4584383 : *shape_changed = true;
1092 4584383 : return xfs_refcount_merge_right_extent(cur, &right, &cright,
1093 : aglen);
1094 : }
1095 :
1096 : return 0;
1097 : }
1098 :
1099 : static inline struct xbtree_refc *
1100 : xrefc_btree_state(
1101 : struct xfs_btree_cur *cur)
1102 : {
1103 1963818610 : if (cur->bc_btnum == XFS_BTNUM_RTREFC)
1104 756785217 : return &cur->bc_ino.refc;
1105 1207033393 : return &cur->bc_ag.refc;
1106 : }
1107 :
1108 : /*
1109 : * XXX: This is a pretty hand-wavy estimate. The penalty for guessing
1110 : * true incorrectly is a shutdown FS; the penalty for guessing false
1111 : * incorrectly is more transaction rolls than might be necessary.
1112 : * Be conservative here.
1113 : */
1114 : static bool
1115 332501374 : xfs_refcount_still_have_space(
1116 : struct xfs_btree_cur *cur)
1117 : {
1118 332501374 : unsigned long overhead;
1119 :
1120 : /*
1121 : * Worst case estimate: full splits of the free space and rmap btrees
1122 : * to handle each of the shape changes to the refcount btree.
1123 : */
1124 665002748 : overhead = xfs_allocfree_block_count(cur->bc_mp,
1125 : xrefc_btree_state(cur)->shape_changes);
1126 332498985 : overhead += cur->bc_maxlevels;
1127 332498985 : overhead *= cur->bc_mp->m_sb.sb_blocksize;
1128 :
1129 : /*
1130 : * Only allow 2 refcount extent updates per transaction if the
1131 : * refcount continue update "error" has been injected.
1132 : */
1133 711506650 : if (xrefc_btree_state(cur)->nr_ops > 2 &&
1134 46508690 : XFS_TEST_ERROR(false, cur->bc_mp,
1135 : XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
1136 : return false;
1137 :
1138 664997110 : if (xrefc_btree_state(cur)->nr_ops == 0)
1139 : return true;
1140 87287731 : else if (overhead > cur->bc_tp->t_log_res)
1141 : return false;
1142 87287731 : return cur->bc_tp->t_log_res - overhead >
1143 87287731 : xrefc_btree_state(cur)->nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
1144 : }
1145 :
1146 : /* Schedule an extent free. */
1147 : static int
1148 75969070 : xrefc_free_extent(
1149 : struct xfs_btree_cur *cur,
1150 : struct xfs_refcount_irec *rec)
1151 : {
1152 75969070 : xfs_fsblock_t fsbno;
1153 75969070 : unsigned int flags = 0;
1154 :
1155 75969070 : if (cur->bc_btnum == XFS_BTNUM_RTREFC) {
1156 28847902 : flags |= XFS_FREE_EXTENT_REALTIME;
1157 28847902 : fsbno = xfs_rgbno_to_rtb(cur->bc_mp, cur->bc_ino.rtg->rtg_rgno,
1158 : rec->rc_startblock);
1159 : } else {
1160 47121168 : fsbno = XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1161 : rec->rc_startblock);
1162 : }
1163 :
1164 75969068 : return xfs_free_extent_later(cur->bc_tp, fsbno, rec->rc_blockcount,
1165 : NULL, XFS_AG_RESV_NONE, flags);
1166 : }
1167 :
1168 : /*
1169 : * Adjust the refcounts of middle extents. At this point we should have
1170 : * split extents that crossed the adjustment range; merged with adjacent
1171 : * extents; and updated agbno/aglen to reflect the merges. Therefore,
1172 : * all we have to do is update the extents inside [agbno, agbno + aglen].
1173 : */
1174 : STATIC int
1175 268444441 : xfs_refcount_adjust_extents(
1176 : struct xfs_btree_cur *cur,
1177 : xfs_agblock_t *agbno,
1178 : xfs_extlen_t *aglen,
1179 : enum xfs_refc_adjust_op adj)
1180 : {
1181 268444441 : struct xfs_refcount_irec ext, tmp;
1182 268444441 : int error;
1183 268444441 : int found_rec, found_tmp;
1184 :
1185 : /* Merging did all the work already. */
1186 268444441 : if (*aglen == 0)
1187 : return 0;
1188 :
1189 247315273 : error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_SHARED, *agbno,
1190 : &found_rec);
1191 247317231 : if (error)
1192 0 : goto out_error;
1193 :
1194 471267639 : while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
1195 323262034 : error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1196 323264604 : if (error)
1197 0 : goto out_error;
1198 323264604 : if (!found_rec || ext.rc_domain != XFS_REFC_DOMAIN_SHARED) {
1199 26144361 : ext.rc_startblock = xrefc_max_startblock(cur);
1200 26144361 : ext.rc_blockcount = 0;
1201 26144361 : ext.rc_refcount = 0;
1202 26144361 : ext.rc_domain = XFS_REFC_DOMAIN_SHARED;
1203 : }
1204 :
1205 : /*
1206 : * Deal with a hole in the refcount tree; if a file maps to
1207 : * these blocks and there's no refcountbt record, pretend that
1208 : * there is one with refcount == 1.
1209 : */
1210 323264604 : if (ext.rc_startblock != *agbno) {
1211 108549542 : tmp.rc_startblock = *agbno;
1212 108549542 : tmp.rc_blockcount = min(*aglen,
1213 : ext.rc_startblock - *agbno);
1214 108549542 : tmp.rc_refcount = 1 + adj;
1215 108549542 : tmp.rc_domain = XFS_REFC_DOMAIN_SHARED;
1216 :
1217 108549542 : trace_xfs_refcount_modify_extent(cur, &tmp);
1218 :
1219 : /*
1220 : * Either cover the hole (increment) or
1221 : * delete the range (decrement).
1222 : */
1223 108548142 : xrefc_btree_state(cur)->nr_ops++;
1224 108548142 : if (tmp.rc_refcount) {
1225 32579285 : error = xfs_refcount_insert(cur, &tmp,
1226 : &found_tmp);
1227 32579276 : if (error)
1228 39 : goto out_error;
1229 32579237 : if (XFS_IS_CORRUPT(cur->bc_mp,
1230 : found_tmp != 1)) {
1231 0 : xfs_btree_mark_sick(cur);
1232 0 : error = -EFSCORRUPTED;
1233 0 : goto out_error;
1234 : }
1235 : } else {
1236 75968857 : error = xrefc_free_extent(cur, &tmp);
1237 75968766 : if (error)
1238 0 : goto out_error;
1239 : }
1240 :
1241 108548003 : (*agbno) += tmp.rc_blockcount;
1242 108548003 : (*aglen) -= tmp.rc_blockcount;
1243 :
1244 : /* Stop if there's nothing left to modify */
1245 108548003 : if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
1246 : break;
1247 :
1248 : /* Move the cursor to the start of ext. */
1249 9235409 : error = xfs_refcount_lookup_ge(cur,
1250 : XFS_REFC_DOMAIN_SHARED, *agbno,
1251 : &found_rec);
1252 9235409 : if (error)
1253 0 : goto out_error;
1254 : }
1255 :
1256 : /*
1257 : * A previous step trimmed agbno/aglen such that the end of the
1258 : * range would not be in the middle of the record. If this is
1259 : * no longer the case, something is seriously wrong with the
1260 : * btree. Make sure we never feed the synthesized record into
1261 : * the processing loop below.
1262 : */
1263 223950471 : if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount == 0) ||
1264 223950471 : XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount > *aglen)) {
1265 0 : xfs_btree_mark_sick(cur);
1266 0 : error = -EFSCORRUPTED;
1267 0 : goto out_error;
1268 : }
1269 :
1270 : /*
1271 : * Adjust the reference count and either update the tree
1272 : * (incr) or free the blocks (decr).
1273 : */
1274 223950471 : if (ext.rc_refcount == XFS_REFC_REFCOUNT_MAX)
1275 80 : goto skip;
1276 223950391 : ext.rc_refcount += adj;
1277 223950391 : trace_xfs_refcount_modify_extent(cur, &ext);
1278 223950086 : xrefc_btree_state(cur)->nr_ops++;
1279 223950086 : if (ext.rc_refcount > 1) {
1280 171558338 : error = xfs_refcount_update(cur, &ext);
1281 171558279 : if (error)
1282 0 : goto out_error;
1283 52391748 : } else if (ext.rc_refcount == 1) {
1284 52391748 : error = xfs_refcount_delete(cur, &found_rec);
1285 52391749 : if (error)
1286 0 : goto out_error;
1287 52391749 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1288 0 : xfs_btree_mark_sick(cur);
1289 0 : error = -EFSCORRUPTED;
1290 0 : goto out_error;
1291 : }
1292 52391749 : goto advloop;
1293 : } else {
1294 0 : error = xrefc_free_extent(cur, &ext);
1295 0 : if (error)
1296 0 : goto out_error;
1297 : }
1298 :
1299 0 : skip:
1300 171558359 : error = xfs_btree_increment(cur, 0, &found_rec);
1301 171558659 : if (error)
1302 0 : goto out_error;
1303 :
1304 171558659 : advloop:
1305 223950408 : (*agbno) += ext.rc_blockcount;
1306 223950408 : (*aglen) -= ext.rc_blockcount;
1307 : }
1308 :
1309 : return error;
1310 39 : out_error:
1311 39 : trace_xfs_refcount_modify_extent_error(cur, error, _RET_IP_);
1312 39 : return error;
1313 : }
1314 :
1315 : /* Adjust the reference count of a range of AG blocks. */
1316 : STATIC int
1317 268445952 : xfs_refcount_adjust(
1318 : struct xfs_btree_cur *cur,
1319 : xfs_agblock_t *agbno,
1320 : xfs_extlen_t *aglen,
1321 : enum xfs_refc_adjust_op adj)
1322 : {
1323 268445952 : bool shape_changed;
1324 268445952 : int shape_changes = 0;
1325 268445952 : int error;
1326 :
1327 268445952 : if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
1328 136258151 : trace_xfs_refcount_increase(cur, *agbno, *aglen);
1329 : else
1330 132187801 : trace_xfs_refcount_decrease(cur, *agbno, *aglen);
1331 :
1332 : /*
1333 : * Ensure that no rcextents cross the boundary of the adjustment range.
1334 : */
1335 268445273 : error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1336 : *agbno, &shape_changed);
1337 268444547 : if (error)
1338 190 : goto out_error;
1339 268444357 : if (shape_changed)
1340 19072336 : shape_changes++;
1341 :
1342 268444357 : error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1343 268444357 : *agbno + *aglen, &shape_changed);
1344 268445845 : if (error)
1345 14 : goto out_error;
1346 268445831 : if (shape_changed)
1347 26775947 : shape_changes++;
1348 :
1349 : /*
1350 : * Try to merge with the left or right extents of the range.
1351 : */
1352 268445831 : error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_SHARED,
1353 : agbno, aglen, adj, &shape_changed);
1354 268444573 : if (error)
1355 10 : goto out_error;
1356 268444563 : if (shape_changed)
1357 21938159 : shape_changes++;
1358 268444563 : if (shape_changes)
1359 96682486 : xrefc_btree_state(cur)->shape_changes++;
1360 :
1361 : /* Now that we've taken care of the ends, adjust the middle extents */
1362 268444563 : error = xfs_refcount_adjust_extents(cur, agbno, aglen, adj);
1363 268442322 : if (error)
1364 39 : goto out_error;
1365 :
1366 : return 0;
1367 :
1368 253 : out_error:
1369 253 : trace_xfs_refcount_adjust_error(cur, error, _RET_IP_);
1370 253 : return error;
1371 : }
1372 :
1373 : /* Clean up after calling xfs_refcount_finish_one. */
1374 : void
1375 292245129 : xfs_refcount_finish_one_cleanup(
1376 : struct xfs_trans *tp,
1377 : struct xfs_btree_cur *rcur,
1378 : int error)
1379 : {
1380 292245129 : struct xfs_buf *agbp = NULL;
1381 :
1382 292245129 : if (rcur == NULL)
1383 : return;
1384 292244841 : if (rcur->bc_btnum == XFS_BTNUM_REFC)
1385 188563748 : agbp = rcur->bc_ag.agbp;
1386 292244841 : xfs_btree_del_cursor(rcur, error);
1387 292247686 : if (agbp)
1388 188567834 : xfs_trans_brelse(tp, agbp);
1389 : }
1390 :
1391 : /* Does this btree cursor match the given AG? */
1392 : static inline bool
1393 : xfs_refcount_is_wrong_cursor(
1394 : struct xfs_btree_cur *cur,
1395 : struct xfs_refcount_intent *ri)
1396 : {
1397 3448348 : if (cur->bc_btnum == XFS_BTNUM_RTREFC)
1398 1395173 : return cur->bc_ino.rtg != ri->ri_rtg;
1399 2053175 : return cur->bc_ag.pag != ri->ri_pag;
1400 : }
1401 :
1402 : /*
1403 : * Set up a continuation a deferred refcount operation by updating the intent.
1404 : * Checks to make sure we're not going to run off the end of the AG or rtgroup.
1405 : */
1406 : static inline int
1407 1795 : xfs_refcount_continue_op(
1408 : struct xfs_btree_cur *cur,
1409 : struct xfs_refcount_intent *ri,
1410 : xfs_agblock_t new_agbno)
1411 : {
1412 1795 : struct xfs_mount *mp = cur->bc_mp;
1413 :
1414 1795 : if (ri->ri_realtime) {
1415 388 : struct xfs_rtgroup *rtg = ri->ri_rtg;
1416 :
1417 388 : if (XFS_IS_CORRUPT(mp, !xfs_verify_rgbext(rtg, new_agbno,
1418 : ri->ri_blockcount))) {
1419 0 : xfs_btree_mark_sick(cur);
1420 0 : return -EFSCORRUPTED;
1421 : }
1422 :
1423 388 : ri->ri_startblock = xfs_rgbno_to_rtb(mp, rtg->rtg_rgno, new_agbno);
1424 :
1425 388 : ASSERT(xfs_verify_rtbext(mp, ri->ri_startblock, ri->ri_blockcount));
1426 388 : ASSERT(rtg->rtg_rgno == xfs_rtb_to_rgno(mp, ri->ri_startblock));
1427 : } else {
1428 1407 : struct xfs_perag *pag = cur->bc_ag.pag;
1429 :
1430 1407 : if (XFS_IS_CORRUPT(mp, !xfs_verify_agbext(pag, new_agbno,
1431 : ri->ri_blockcount))) {
1432 0 : xfs_btree_mark_sick(cur);
1433 0 : return -EFSCORRUPTED;
1434 : }
1435 :
1436 1407 : ri->ri_startblock = XFS_AGB_TO_FSB(mp, pag->pag_agno, new_agbno);
1437 :
1438 1407 : ASSERT(xfs_verify_fsbext(mp, ri->ri_startblock, ri->ri_blockcount));
1439 1407 : ASSERT(pag->pag_agno == XFS_FSB_TO_AGNO(mp, ri->ri_startblock));
1440 : }
1441 :
1442 : return 0;
1443 : }
1444 :
1445 : /*
1446 : * Process one of the deferred refcount operations. We pass back the
1447 : * btree cursor to maintain our lock on the btree between calls.
1448 : * This saves time and eliminates a buffer deadlock between the
1449 : * superblock and the AGF because we'll always grab them in the same
1450 : * order.
1451 : */
1452 : int
1453 294706913 : xfs_refcount_finish_one(
1454 : struct xfs_trans *tp,
1455 : struct xfs_refcount_intent *ri,
1456 : struct xfs_btree_cur **pcur)
1457 : {
1458 294706913 : struct xfs_mount *mp = tp->t_mountp;
1459 294706913 : struct xfs_btree_cur *rcur;
1460 294706913 : struct xfs_buf *agbp = NULL;
1461 294706913 : int error = 0;
1462 294706913 : xfs_agblock_t bno;
1463 294706913 : unsigned long nr_ops = 0;
1464 294706913 : int shape_changes = 0;
1465 :
1466 294706913 : trace_xfs_refcount_deferred(mp, ri);
1467 :
1468 294702975 : if (ri->ri_realtime) {
1469 104598888 : xfs_rgnumber_t rgno;
1470 :
1471 104598888 : bno = xfs_rtb_to_rgbno(mp, ri->ri_startblock, &rgno);
1472 : } else {
1473 190104087 : bno = XFS_FSB_TO_AGBNO(mp, ri->ri_startblock);
1474 : }
1475 :
1476 294702869 : if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1477 : return -EIO;
1478 :
1479 : /*
1480 : * If we haven't gotten a cursor or the cursor AG doesn't match
1481 : * the startblock, get one now.
1482 : */
1483 294703821 : rcur = *pcur;
1484 298152169 : if (rcur != NULL && xfs_refcount_is_wrong_cursor(rcur, ri)) {
1485 984807 : nr_ops = xrefc_btree_state(rcur)->nr_ops;
1486 984807 : shape_changes = xrefc_btree_state(rcur)->shape_changes;
1487 984807 : xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1488 984807 : rcur = NULL;
1489 984807 : *pcur = NULL;
1490 : }
1491 294703821 : if (rcur == NULL) {
1492 292240032 : if (ri->ri_realtime) {
1493 103679323 : xfs_rtgroup_lock(tp, ri->ri_rtg, XFS_RTGLOCK_REFCOUNT);
1494 103680173 : rcur = xfs_rtrefcountbt_init_cursor(mp, tp, ri->ri_rtg,
1495 103680173 : ri->ri_rtg->rtg_refcountip);
1496 : } else {
1497 188560709 : error = xfs_alloc_read_agf(ri->ri_pag, tp,
1498 : XFS_ALLOC_FLAG_FREEING, &agbp);
1499 188567209 : if (error)
1500 : return error;
1501 :
1502 188567066 : rcur = xfs_refcountbt_init_cursor(mp, tp, agbp,
1503 : ri->ri_pag);
1504 : }
1505 292247709 : xrefc_btree_state(rcur)->nr_ops = nr_ops;
1506 584495418 : xrefc_btree_state(rcur)->shape_changes = shape_changes;
1507 : }
1508 294711498 : *pcur = rcur;
1509 :
1510 294711498 : switch (ri->ri_type) {
1511 136258170 : case XFS_REFCOUNT_INCREASE:
1512 136258170 : error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1513 : XFS_REFCOUNT_ADJUST_INCREASE);
1514 136258136 : if (error)
1515 : return error;
1516 136258027 : if (ri->ri_blockcount > 0)
1517 210 : error = xfs_refcount_continue_op(rcur, ri, bno);
1518 : break;
1519 132188224 : case XFS_REFCOUNT_DECREASE:
1520 132188224 : error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1521 : XFS_REFCOUNT_ADJUST_DECREASE);
1522 132184334 : if (error)
1523 : return error;
1524 132184190 : if (ri->ri_blockcount > 0)
1525 1585 : error = xfs_refcount_continue_op(rcur, ri, bno);
1526 : break;
1527 4923613 : case XFS_REFCOUNT_ALLOC_COW:
1528 4923613 : error = __xfs_refcount_cow_alloc(rcur, bno, ri->ri_blockcount);
1529 4923401 : if (error)
1530 : return error;
1531 4923400 : ri->ri_blockcount = 0;
1532 4923400 : break;
1533 21341491 : case XFS_REFCOUNT_FREE_COW:
1534 21341491 : error = __xfs_refcount_cow_free(rcur, bno, ri->ri_blockcount);
1535 21341656 : if (error)
1536 : return error;
1537 21341656 : ri->ri_blockcount = 0;
1538 21341656 : break;
1539 0 : default:
1540 0 : ASSERT(0);
1541 0 : return -EFSCORRUPTED;
1542 : }
1543 294707273 : if (!error && ri->ri_blockcount > 0)
1544 1795 : trace_xfs_refcount_finish_one_leftover(mp, ri);
1545 : return error;
1546 : }
1547 :
1548 : /*
1549 : * Record a refcount intent for later processing.
1550 : */
1551 : static void
1552 294698667 : __xfs_refcount_add(
1553 : struct xfs_trans *tp,
1554 : enum xfs_refcount_intent_type type,
1555 : bool isrt,
1556 : xfs_fsblock_t startblock,
1557 : xfs_extlen_t blockcount)
1558 : {
1559 294698667 : struct xfs_refcount_intent *ri;
1560 294698667 : enum xfs_defer_ops_type optype;
1561 :
1562 294698667 : ri = kmem_cache_alloc(xfs_refcount_intent_cache,
1563 : GFP_NOFS | __GFP_NOFAIL);
1564 294700041 : INIT_LIST_HEAD(&ri->ri_list);
1565 294700041 : ri->ri_type = type;
1566 294700041 : ri->ri_startblock = startblock;
1567 294700041 : ri->ri_blockcount = blockcount;
1568 294700041 : ri->ri_realtime = isrt;
1569 :
1570 294700041 : trace_xfs_refcount_defer(tp->t_mountp, ri);
1571 :
1572 : /*
1573 : * Deferred refcount updates for the realtime and data sections must
1574 : * use separate transactions to finish deferred work because updates to
1575 : * realtime metadata files can lock AGFs to allocate btree blocks and
1576 : * we don't want that mixing with the AGF locks taken to finish data
1577 : * section updates.
1578 : */
1579 294695459 : if (isrt)
1580 : optype = XFS_DEFER_OPS_TYPE_REFCOUNT_RT;
1581 : else
1582 190097881 : optype = XFS_DEFER_OPS_TYPE_REFCOUNT;
1583 :
1584 294695459 : xfs_refcount_update_get_group(tp->t_mountp, ri);
1585 294708559 : xfs_defer_add(tp, optype, &ri->ri_list);
1586 294708599 : }
1587 :
1588 : /*
1589 : * Increase the reference count of the blocks backing a file's extent.
1590 : */
1591 : void
1592 136257399 : xfs_refcount_increase_extent(
1593 : struct xfs_trans *tp,
1594 : bool isrt,
1595 : struct xfs_bmbt_irec *PREV)
1596 : {
1597 136257399 : if (!xfs_has_reflink(tp->t_mountp))
1598 : return;
1599 :
1600 136257362 : __xfs_refcount_add(tp, XFS_REFCOUNT_INCREASE, isrt, PREV->br_startblock,
1601 136257362 : PREV->br_blockcount);
1602 : }
1603 :
1604 : /*
1605 : * Decrease the reference count of the blocks backing a file's extent.
1606 : */
1607 : void
1608 132182789 : xfs_refcount_decrease_extent(
1609 : struct xfs_trans *tp,
1610 : bool isrt,
1611 : struct xfs_bmbt_irec *PREV)
1612 : {
1613 132182789 : if (!xfs_has_reflink(tp->t_mountp))
1614 : return;
1615 :
1616 132182176 : __xfs_refcount_add(tp, XFS_REFCOUNT_DECREASE, isrt, PREV->br_startblock,
1617 132182176 : PREV->br_blockcount);
1618 : }
1619 :
1620 : /*
1621 : * Given an AG extent, find the lowest-numbered run of shared blocks
1622 : * within that range and return the range in fbno/flen. If
1623 : * find_end_of_shared is set, return the longest contiguous extent of
1624 : * shared blocks; if not, just return the first extent we find. If no
1625 : * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1626 : * and 0, respectively.
1627 : */
1628 : int
1629 2923361721 : xfs_refcount_find_shared(
1630 : struct xfs_btree_cur *cur,
1631 : xfs_agblock_t agbno,
1632 : xfs_extlen_t aglen,
1633 : xfs_agblock_t *fbno,
1634 : xfs_extlen_t *flen,
1635 : bool find_end_of_shared)
1636 : {
1637 2923361721 : struct xfs_refcount_irec tmp;
1638 2923361721 : int i;
1639 2923361721 : int have;
1640 2923361721 : int error;
1641 :
1642 2923361721 : trace_xfs_refcount_find_shared(cur, agbno, aglen);
1643 :
1644 : /* By default, skip the whole range */
1645 2913348921 : *fbno = NULLAGBLOCK;
1646 2913348921 : *flen = 0;
1647 :
1648 : /* Try to find a refcount extent that crosses the start */
1649 2913348921 : error = xfs_refcount_lookup_le(cur, XFS_REFC_DOMAIN_SHARED, agbno,
1650 : &have);
1651 2924394415 : if (error)
1652 67 : goto out_error;
1653 2924394348 : if (!have) {
1654 : /* No left extent, look at the next one */
1655 937091279 : error = xfs_btree_increment(cur, 0, &have);
1656 935535154 : if (error)
1657 0 : goto out_error;
1658 935535154 : if (!have)
1659 888142377 : goto done;
1660 : }
1661 2034695846 : error = xfs_refcount_get_rec(cur, &tmp, &i);
1662 2031815617 : if (error)
1663 0 : goto out_error;
1664 2031815617 : if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1665 0 : xfs_btree_mark_sick(cur);
1666 0 : error = -EFSCORRUPTED;
1667 0 : goto out_error;
1668 : }
1669 2031815617 : if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1670 3174644 : goto done;
1671 :
1672 : /* If the extent ends before the start, look at the next one */
1673 2028640973 : if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1674 1517029908 : error = xfs_btree_increment(cur, 0, &have);
1675 1518455720 : if (error)
1676 0 : goto out_error;
1677 1518455720 : if (!have)
1678 29743785 : goto done;
1679 1488711935 : error = xfs_refcount_get_rec(cur, &tmp, &i);
1680 1488258009 : if (error)
1681 0 : goto out_error;
1682 1488258009 : if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1683 0 : xfs_btree_mark_sick(cur);
1684 0 : error = -EFSCORRUPTED;
1685 0 : goto out_error;
1686 : }
1687 1488258009 : if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1688 33986020 : goto done;
1689 : }
1690 :
1691 : /* If the extent starts after the range we want, bail out */
1692 1965883054 : if (tmp.rc_startblock >= agbno + aglen)
1693 1492932692 : goto done;
1694 :
1695 : /* We found the start of a shared extent! */
1696 472950362 : if (tmp.rc_startblock < agbno) {
1697 48803553 : tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1698 48803553 : tmp.rc_startblock = agbno;
1699 : }
1700 :
1701 472950362 : *fbno = tmp.rc_startblock;
1702 472950362 : *flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1703 472950362 : if (!find_end_of_shared)
1704 468326754 : goto done;
1705 :
1706 : /* Otherwise, find the end of this shared extent */
1707 7271378 : while (*fbno + *flen < agbno + aglen) {
1708 2635566 : error = xfs_btree_increment(cur, 0, &have);
1709 2782685 : if (error)
1710 0 : goto out_error;
1711 2782685 : if (!have)
1712 : break;
1713 2773615 : error = xfs_refcount_get_rec(cur, &tmp, &i);
1714 2773617 : if (error)
1715 0 : goto out_error;
1716 2773617 : if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1717 0 : xfs_btree_mark_sick(cur);
1718 0 : error = -EFSCORRUPTED;
1719 0 : goto out_error;
1720 : }
1721 2773617 : if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED ||
1722 2759207 : tmp.rc_startblock >= agbno + aglen ||
1723 2661927 : tmp.rc_startblock != *fbno + *flen)
1724 : break;
1725 2647770 : *flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1726 : }
1727 :
1728 4786761 : done:
1729 2921093033 : trace_xfs_refcount_find_shared_result(cur, *fbno, *flen);
1730 :
1731 2918769918 : out_error:
1732 2918769918 : if (error)
1733 67 : trace_xfs_refcount_find_shared_error(cur, error, _RET_IP_);
1734 2918769918 : return error;
1735 : }
1736 :
1737 : /*
1738 : * Recovering CoW Blocks After a Crash
1739 : *
1740 : * Due to the way that the copy on write mechanism works, there's a window of
1741 : * opportunity in which we can lose track of allocated blocks during a crash.
1742 : * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1743 : * causes blocks to be allocated and stored in the CoW fork. The blocks are
1744 : * no longer in the free space btree but are not otherwise recorded anywhere
1745 : * until the write completes and the blocks are mapped into the file. A crash
1746 : * in between allocation and remapping results in the replacement blocks being
1747 : * lost. This situation is exacerbated by the CoW extent size hint because
1748 : * allocations can hang around for long time.
1749 : *
1750 : * However, there is a place where we can record these allocations before they
1751 : * become mappings -- the reference count btree. The btree does not record
1752 : * extents with refcount == 1, so we can record allocations with a refcount of
1753 : * 1. Blocks being used for CoW writeout cannot be shared, so there should be
1754 : * no conflict with shared block records. These mappings should be created
1755 : * when we allocate blocks to the CoW fork and deleted when they're removed
1756 : * from the CoW fork.
1757 : *
1758 : * Minor nit: records for in-progress CoW allocations and records for shared
1759 : * extents must never be merged, to preserve the property that (except for CoW
1760 : * allocations) there are no refcount btree entries with refcount == 1. The
1761 : * only time this could potentially happen is when unsharing a block that's
1762 : * adjacent to CoW allocations, so we must be careful to avoid this.
1763 : *
1764 : * At mount time we recover lost CoW allocations by searching the refcount
1765 : * btree for these refcount == 1 mappings. These represent CoW allocations
1766 : * that were in progress at the time the filesystem went down, so we can free
1767 : * them to get the space back.
1768 : *
1769 : * This mechanism is superior to creating EFIs for unmapped CoW extents for
1770 : * several reasons -- first, EFIs pin the tail of the log and would have to be
1771 : * periodically relogged to avoid filling up the log. Second, CoW completions
1772 : * will have to file an EFD and create new EFIs for whatever remains in the
1773 : * CoW fork; this partially takes care of (1) but extent-size reservations
1774 : * will have to periodically relog even if there's no writeout in progress.
1775 : * This can happen if the CoW extent size hint is set, which you really want.
1776 : * Third, EFIs cannot currently be automatically relogged into newer
1777 : * transactions to advance the log tail. Fourth, stuffing the log full of
1778 : * EFIs places an upper bound on the number of CoW allocations that can be
1779 : * held filesystem-wide at any given time. Recording them in the refcount
1780 : * btree doesn't require us to maintain any state in memory and doesn't pin
1781 : * the log.
1782 : */
1783 : /*
1784 : * Adjust the refcounts of CoW allocations. These allocations are "magic"
1785 : * in that they're not referenced anywhere else in the filesystem, so we
1786 : * stash them in the refcount btree with a refcount of 1 until either file
1787 : * remapping (or CoW cancellation) happens.
1788 : */
1789 : STATIC int
1790 26264211 : xfs_refcount_adjust_cow_extents(
1791 : struct xfs_btree_cur *cur,
1792 : xfs_agblock_t agbno,
1793 : xfs_extlen_t aglen,
1794 : enum xfs_refc_adjust_op adj)
1795 : {
1796 26264211 : struct xfs_refcount_irec ext, tmp;
1797 26264211 : int error;
1798 26264211 : int found_rec, found_tmp;
1799 :
1800 26264211 : if (aglen == 0)
1801 : return 0;
1802 :
1803 : /* Find any overlapping refcount records */
1804 24245346 : error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_COW, agbno,
1805 : &found_rec);
1806 24247038 : if (error)
1807 0 : goto out_error;
1808 24247038 : error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1809 24246687 : if (error)
1810 0 : goto out_error;
1811 24246687 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec &&
1812 : ext.rc_domain != XFS_REFC_DOMAIN_COW)) {
1813 0 : xfs_btree_mark_sick(cur);
1814 0 : error = -EFSCORRUPTED;
1815 0 : goto out_error;
1816 : }
1817 24246687 : if (!found_rec) {
1818 1570884 : ext.rc_startblock = xrefc_max_startblock(cur);
1819 1570884 : ext.rc_blockcount = 0;
1820 1570884 : ext.rc_refcount = 0;
1821 1570884 : ext.rc_domain = XFS_REFC_DOMAIN_COW;
1822 : }
1823 :
1824 24246687 : switch (adj) {
1825 2904844 : case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1826 : /* Adding a CoW reservation, there should be nothing here. */
1827 2904844 : if (XFS_IS_CORRUPT(cur->bc_mp,
1828 : agbno + aglen > ext.rc_startblock)) {
1829 0 : xfs_btree_mark_sick(cur);
1830 0 : error = -EFSCORRUPTED;
1831 0 : goto out_error;
1832 : }
1833 :
1834 2904844 : tmp.rc_startblock = agbno;
1835 2904844 : tmp.rc_blockcount = aglen;
1836 2904844 : tmp.rc_refcount = 1;
1837 2904844 : tmp.rc_domain = XFS_REFC_DOMAIN_COW;
1838 :
1839 2904844 : trace_xfs_refcount_modify_extent(cur, &tmp);
1840 :
1841 2904786 : error = xfs_refcount_insert(cur, &tmp,
1842 : &found_tmp);
1843 2904769 : if (error)
1844 0 : goto out_error;
1845 2904769 : if (XFS_IS_CORRUPT(cur->bc_mp, found_tmp != 1)) {
1846 0 : xfs_btree_mark_sick(cur);
1847 0 : error = -EFSCORRUPTED;
1848 0 : goto out_error;
1849 : }
1850 : break;
1851 21341843 : case XFS_REFCOUNT_ADJUST_COW_FREE:
1852 : /* Removing a CoW reservation, there should be one extent. */
1853 21341843 : if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_startblock != agbno)) {
1854 0 : xfs_btree_mark_sick(cur);
1855 0 : error = -EFSCORRUPTED;
1856 0 : goto out_error;
1857 : }
1858 21341843 : if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount != aglen)) {
1859 0 : xfs_btree_mark_sick(cur);
1860 0 : error = -EFSCORRUPTED;
1861 0 : goto out_error;
1862 : }
1863 21341843 : if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_refcount != 1)) {
1864 0 : xfs_btree_mark_sick(cur);
1865 0 : error = -EFSCORRUPTED;
1866 0 : goto out_error;
1867 : }
1868 :
1869 21341843 : ext.rc_refcount = 0;
1870 21341843 : trace_xfs_refcount_modify_extent(cur, &ext);
1871 21340316 : error = xfs_refcount_delete(cur, &found_rec);
1872 21341981 : if (error)
1873 0 : goto out_error;
1874 21341981 : if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1875 0 : xfs_btree_mark_sick(cur);
1876 0 : error = -EFSCORRUPTED;
1877 0 : goto out_error;
1878 : }
1879 : break;
1880 0 : default:
1881 0 : ASSERT(0);
1882 : }
1883 :
1884 : return error;
1885 0 : out_error:
1886 0 : trace_xfs_refcount_modify_extent_error(cur, error, _RET_IP_);
1887 0 : return error;
1888 : }
1889 :
1890 : /*
1891 : * Add or remove refcount btree entries for CoW reservations.
1892 : */
1893 : STATIC int
1894 26261862 : xfs_refcount_adjust_cow(
1895 : struct xfs_btree_cur *cur,
1896 : xfs_agblock_t agbno,
1897 : xfs_extlen_t aglen,
1898 : enum xfs_refc_adjust_op adj)
1899 : {
1900 26261862 : bool shape_changed;
1901 26261862 : int error;
1902 :
1903 : /*
1904 : * Ensure that no rcextents cross the boundary of the adjustment range.
1905 : */
1906 26261862 : error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1907 : agbno, &shape_changed);
1908 26262737 : if (error)
1909 1 : goto out_error;
1910 :
1911 26262736 : error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1912 : agbno + aglen, &shape_changed);
1913 26264994 : if (error)
1914 0 : goto out_error;
1915 :
1916 : /*
1917 : * Try to merge with the left or right extents of the range.
1918 : */
1919 26264994 : error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_COW, &agbno,
1920 : &aglen, adj, &shape_changed);
1921 26264309 : if (error)
1922 0 : goto out_error;
1923 :
1924 : /* Now that we've taken care of the ends, adjust the middle extents */
1925 26264309 : error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1926 26265183 : if (error)
1927 0 : goto out_error;
1928 :
1929 : return 0;
1930 :
1931 1 : out_error:
1932 1 : trace_xfs_refcount_adjust_cow_error(cur, error, _RET_IP_);
1933 1 : return error;
1934 : }
1935 :
1936 : /*
1937 : * Record a CoW allocation in the refcount btree.
1938 : */
1939 : STATIC int
1940 4923512 : __xfs_refcount_cow_alloc(
1941 : struct xfs_btree_cur *rcur,
1942 : xfs_agblock_t agbno,
1943 : xfs_extlen_t aglen)
1944 : {
1945 4923512 : trace_xfs_refcount_cow_increase(rcur, agbno, aglen);
1946 :
1947 : /* Add refcount btree reservation */
1948 4923400 : return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1949 : XFS_REFCOUNT_ADJUST_COW_ALLOC);
1950 : }
1951 :
1952 : /*
1953 : * Remove a CoW allocation from the refcount btree.
1954 : */
1955 : STATIC int
1956 21339858 : __xfs_refcount_cow_free(
1957 : struct xfs_btree_cur *rcur,
1958 : xfs_agblock_t agbno,
1959 : xfs_extlen_t aglen)
1960 : {
1961 21339858 : trace_xfs_refcount_cow_decrease(rcur, agbno, aglen);
1962 :
1963 : /* Remove refcount btree reservation */
1964 21337398 : return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1965 : XFS_REFCOUNT_ADJUST_COW_FREE);
1966 : }
1967 :
1968 : /* Record a CoW staging extent in the refcount btree. */
1969 : void
1970 4923414 : xfs_refcount_alloc_cow_extent(
1971 : struct xfs_trans *tp,
1972 : bool isrt,
1973 : xfs_fsblock_t fsb,
1974 : xfs_extlen_t len)
1975 : {
1976 4923414 : struct xfs_mount *mp = tp->t_mountp;
1977 :
1978 4923414 : if (!xfs_has_reflink(mp))
1979 : return;
1980 :
1981 4923382 : __xfs_refcount_add(tp, XFS_REFCOUNT_ALLOC_COW, isrt, fsb, len);
1982 :
1983 : /* Add rmap entry */
1984 4923509 : xfs_rmap_alloc_extent(tp, isrt, fsb, len, XFS_RMAP_OWN_COW);
1985 : }
1986 :
1987 : /* Forget a CoW staging event in the refcount btree. */
1988 : void
1989 21337515 : xfs_refcount_free_cow_extent(
1990 : struct xfs_trans *tp,
1991 : bool isrt,
1992 : xfs_fsblock_t fsb,
1993 : xfs_extlen_t len)
1994 : {
1995 21337515 : struct xfs_mount *mp = tp->t_mountp;
1996 :
1997 21337515 : if (!xfs_has_reflink(mp))
1998 : return;
1999 :
2000 : /* Remove rmap entry */
2001 21337305 : xfs_rmap_free_extent(tp, isrt, fsb, len, XFS_RMAP_OWN_COW);
2002 21339454 : __xfs_refcount_add(tp, XFS_REFCOUNT_FREE_COW, isrt, fsb, len);
2003 : }
2004 :
2005 : struct xfs_refcount_recovery {
2006 : struct list_head rr_list;
2007 : struct xfs_refcount_irec rr_rrec;
2008 : };
2009 :
2010 : /* Stuff an extent on the recovery list. */
2011 : STATIC int
2012 310417 : xfs_refcount_recover_extent(
2013 : struct xfs_btree_cur *cur,
2014 : const union xfs_btree_rec *rec,
2015 : void *priv)
2016 : {
2017 310417 : struct list_head *debris = priv;
2018 310417 : struct xfs_refcount_recovery *rr;
2019 :
2020 310417 : if (XFS_IS_CORRUPT(cur->bc_mp,
2021 : be32_to_cpu(rec->refc.rc_refcount) != 1)) {
2022 0 : xfs_btree_mark_sick(cur);
2023 0 : return -EFSCORRUPTED;
2024 : }
2025 :
2026 310417 : rr = kmalloc(sizeof(struct xfs_refcount_recovery),
2027 : GFP_KERNEL | __GFP_NOFAIL);
2028 310417 : INIT_LIST_HEAD(&rr->rr_list);
2029 310417 : xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
2030 :
2031 310417 : if (xfs_refcount_check_irec(cur, &rr->rr_rrec) != NULL ||
2032 310417 : XFS_IS_CORRUPT(cur->bc_mp,
2033 : rr->rr_rrec.rc_domain != XFS_REFC_DOMAIN_COW)) {
2034 0 : xfs_btree_mark_sick(cur);
2035 0 : kfree(rr);
2036 0 : return -EFSCORRUPTED;
2037 : }
2038 :
2039 310417 : list_add_tail(&rr->rr_list, debris);
2040 310417 : return 0;
2041 : }
2042 :
2043 : /* Find and remove leftover CoW reservations. */
2044 : static int
2045 62327 : xfs_refcount_recover_group_cow_leftovers(
2046 : struct xfs_mount *mp,
2047 : struct xfs_perag *pag,
2048 : struct xfs_rtgroup *rtg)
2049 : {
2050 62327 : struct xfs_trans *tp;
2051 62327 : struct xfs_btree_cur *cur;
2052 62327 : struct xfs_buf *agbp = NULL;
2053 62327 : struct xfs_refcount_recovery *rr, *n;
2054 62327 : struct list_head debris;
2055 62327 : union xfs_btree_irec low = {
2056 : .rc.rc_domain = XFS_REFC_DOMAIN_COW,
2057 : };
2058 62327 : union xfs_btree_irec high = {
2059 : .rc.rc_domain = XFS_REFC_DOMAIN_COW,
2060 : .rc.rc_startblock = -1U,
2061 : };
2062 62327 : xfs_fsblock_t fsb;
2063 62327 : int error;
2064 :
2065 : /* reflink filesystems mustn't have AGs larger than 2^31-1 blocks */
2066 62327 : BUILD_BUG_ON(XFS_MAX_CRC_AG_BLOCKS >= XFS_REFC_COWFLAG);
2067 62327 : if (pag && mp->m_sb.sb_agblocks > XFS_MAX_CRC_AG_BLOCKS)
2068 : return -EOPNOTSUPP;
2069 :
2070 : /* rtreflink filesystems can't have rtgroups larger than 2^31-1 blocks */
2071 62327 : BUILD_BUG_ON(XFS_MAX_RGBLOCKS >= XFS_REFC_COWFLAG);
2072 62327 : if (rtg && mp->m_sb.sb_rgblocks >= XFS_MAX_RGBLOCKS)
2073 : return -EOPNOTSUPP;
2074 :
2075 62327 : INIT_LIST_HEAD(&debris);
2076 :
2077 : /*
2078 : * In this first part, we use an empty transaction to gather up
2079 : * all the leftover CoW extents so that we can subsequently
2080 : * delete them. The empty transaction is used to avoid
2081 : * a buffer lock deadlock if there happens to be a loop in the
2082 : * refcountbt because we're allowed to re-grab a buffer that is
2083 : * already attached to our transaction. When we're done
2084 : * recording the CoW debris we cancel the (empty) transaction
2085 : * and everything goes away cleanly.
2086 : */
2087 62327 : error = xfs_trans_alloc_empty(mp, &tp);
2088 62327 : if (error)
2089 : return error;
2090 :
2091 62327 : if (rtg) {
2092 6325 : xfs_rtgroup_lock(NULL, rtg, XFS_RTGLOCK_REFCOUNT);
2093 6325 : cur = xfs_rtrefcountbt_init_cursor(mp, tp, rtg,
2094 : rtg->rtg_refcountip);
2095 : } else {
2096 56002 : error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
2097 56002 : if (error)
2098 10 : goto out_trans;
2099 55992 : cur = xfs_refcountbt_init_cursor(mp, tp, agbp, pag);
2100 : }
2101 :
2102 : /* Find all the leftover CoW staging extents. */
2103 62317 : error = xfs_btree_query_range(cur, &low, &high,
2104 : xfs_refcount_recover_extent, &debris);
2105 62317 : xfs_btree_del_cursor(cur, error);
2106 62317 : if (agbp)
2107 55992 : xfs_trans_brelse(tp, agbp);
2108 : else
2109 6325 : xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_REFCOUNT);
2110 62317 : xfs_trans_cancel(tp);
2111 62317 : if (error)
2112 20 : goto out_free;
2113 :
2114 : /* Now iterate the list to free the leftovers */
2115 372714 : list_for_each_entry_safe(rr, n, &debris, rr_list) {
2116 : /* Set up transaction. */
2117 310417 : error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
2118 310417 : if (error)
2119 0 : goto out_free;
2120 :
2121 : /* Free the orphan record */
2122 310417 : if (rtg)
2123 857 : fsb = xfs_rgbno_to_rtb(mp, rtg->rtg_rgno,
2124 : rr->rr_rrec.rc_startblock);
2125 : else
2126 309560 : fsb = XFS_AGB_TO_FSB(mp, pag->pag_agno,
2127 : rr->rr_rrec.rc_startblock);
2128 310417 : xfs_refcount_free_cow_extent(tp, rtg != NULL, fsb,
2129 : rr->rr_rrec.rc_blockcount);
2130 :
2131 : /* Free the block. */
2132 310417 : error = xfs_free_extent_later(tp, fsb,
2133 310417 : rr->rr_rrec.rc_blockcount, NULL,
2134 : XFS_AG_RESV_NONE,
2135 : rtg != NULL ? XFS_FREE_EXTENT_REALTIME : 0);
2136 310417 : if (error)
2137 0 : goto out_trans;
2138 :
2139 310417 : error = xfs_trans_commit(tp);
2140 310417 : if (error)
2141 0 : goto out_free;
2142 :
2143 310417 : list_del(&rr->rr_list);
2144 310417 : kfree(rr);
2145 : }
2146 :
2147 : return error;
2148 10 : out_trans:
2149 10 : xfs_trans_cancel(tp);
2150 30 : out_free:
2151 : /* Free the leftover list */
2152 30 : list_for_each_entry_safe(rr, n, &debris, rr_list) {
2153 0 : list_del(&rr->rr_list);
2154 0 : kfree(rr);
2155 : }
2156 : return error;
2157 : }
2158 :
2159 : int
2160 56002 : xfs_refcount_recover_cow_leftovers(
2161 : struct xfs_mount *mp,
2162 : struct xfs_perag *pag)
2163 : {
2164 56002 : return xfs_refcount_recover_group_cow_leftovers(mp, pag, NULL);
2165 : }
2166 :
2167 : int
2168 6325 : xfs_refcount_recover_rtcow_leftovers(
2169 : struct xfs_mount *mp,
2170 : struct xfs_rtgroup *rtg)
2171 : {
2172 6325 : return xfs_refcount_recover_group_cow_leftovers(mp, NULL, rtg);
2173 : }
2174 :
2175 : /*
2176 : * Scan part of the keyspace of the refcount records and tell us if the area
2177 : * has no records, is fully mapped by records, or is partially filled.
2178 : */
2179 : int
2180 3022608338 : xfs_refcount_has_records(
2181 : struct xfs_btree_cur *cur,
2182 : enum xfs_refc_domain domain,
2183 : xfs_agblock_t bno,
2184 : xfs_extlen_t len,
2185 : enum xbtree_recpacking *outcome)
2186 : {
2187 3022608338 : union xfs_btree_irec low;
2188 3022608338 : union xfs_btree_irec high;
2189 :
2190 3022608338 : memset(&low, 0, sizeof(low));
2191 3022608338 : low.rc.rc_startblock = bno;
2192 3022608338 : memset(&high, 0xFF, sizeof(high));
2193 3022608338 : high.rc.rc_startblock = bno + len - 1;
2194 3022608338 : low.rc.rc_domain = high.rc.rc_domain = domain;
2195 :
2196 3022608338 : return xfs_btree_has_records(cur, &low, &high, NULL, outcome);
2197 : }
2198 :
2199 : struct xfs_refcount_query_range_info {
2200 : xfs_refcount_query_range_fn fn;
2201 : void *priv;
2202 : };
2203 :
2204 : /* Format btree record and pass to our callback. */
2205 : STATIC int
2206 770736 : xfs_refcount_query_range_helper(
2207 : struct xfs_btree_cur *cur,
2208 : const union xfs_btree_rec *rec,
2209 : void *priv)
2210 : {
2211 770736 : struct xfs_refcount_query_range_info *query = priv;
2212 770736 : struct xfs_refcount_irec irec;
2213 770736 : xfs_failaddr_t fa;
2214 :
2215 770736 : xfs_refcount_btrec_to_irec(rec, &irec);
2216 770739 : fa = xfs_refcount_check_irec(cur, &irec);
2217 770722 : if (fa)
2218 0 : return xfs_refcount_complain_bad_rec(cur, fa, &irec);
2219 :
2220 770722 : return query->fn(cur, &irec, query->priv);
2221 : }
2222 :
2223 : /* Find all refcount records between two keys. */
2224 : int
2225 96704 : xfs_refcount_query_range(
2226 : struct xfs_btree_cur *cur,
2227 : const struct xfs_refcount_irec *low_rec,
2228 : const struct xfs_refcount_irec *high_rec,
2229 : xfs_refcount_query_range_fn fn,
2230 : void *priv)
2231 : {
2232 96704 : union xfs_btree_irec low_brec = { .rc = *low_rec };
2233 96704 : union xfs_btree_irec high_brec = { .rc = *high_rec };
2234 96704 : struct xfs_refcount_query_range_info query = { .priv = priv, .fn = fn };
2235 :
2236 96704 : return xfs_btree_query_range(cur, &low_brec, &high_brec,
2237 : xfs_refcount_query_range_helper, &query);
2238 : }
2239 :
2240 : int __init
2241 59 : xfs_refcount_intent_init_cache(void)
2242 : {
2243 59 : xfs_refcount_intent_cache = kmem_cache_create("xfs_refc_intent",
2244 : sizeof(struct xfs_refcount_intent),
2245 : 0, 0, NULL);
2246 :
2247 59 : return xfs_refcount_intent_cache != NULL ? 0 : -ENOMEM;
2248 : }
2249 :
2250 : void
2251 58 : xfs_refcount_intent_destroy_cache(void)
2252 : {
2253 58 : kmem_cache_destroy(xfs_refcount_intent_cache);
2254 58 : xfs_refcount_intent_cache = NULL;
2255 58 : }
|
