Loading...
Note: File does not exist in v3.5.6.
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2022-2023 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_trans_resv.h"
11#include "xfs_mount.h"
12#include "xfs_btree.h"
13#include "xfs_btree_staging.h"
14#include "xfs_log_format.h"
15#include "xfs_trans.h"
16#include "xfs_sb.h"
17#include "xfs_inode.h"
18#include "xfs_alloc.h"
19#include "xfs_rmap.h"
20#include "xfs_ag.h"
21#include "xfs_defer.h"
22#include "scrub/scrub.h"
23#include "scrub/common.h"
24#include "scrub/trace.h"
25#include "scrub/repair.h"
26#include "scrub/newbt.h"
27
28/*
29 * Estimate proper slack values for a btree that's being reloaded.
30 *
31 * Under most circumstances, we'll take whatever default loading value the
32 * btree bulk loading code calculates for us. However, there are some
33 * exceptions to this rule:
34 *
35 * (0) If someone turned one of the debug knobs.
36 * (1) If this is a per-AG btree and the AG has less than 10% space free.
37 * (2) If this is an inode btree and the FS has less than 10% space free.
38
39 * In either case, format the new btree blocks almost completely full to
40 * minimize space usage.
41 */
42static void
43xrep_newbt_estimate_slack(
44 struct xrep_newbt *xnr)
45{
46 struct xfs_scrub *sc = xnr->sc;
47 struct xfs_btree_bload *bload = &xnr->bload;
48 uint64_t free;
49 uint64_t sz;
50
51 /*
52 * The xfs_globals values are set to -1 (i.e. take the bload defaults)
53 * unless someone has set them otherwise, so we just pull the values
54 * here.
55 */
56 bload->leaf_slack = xfs_globals.bload_leaf_slack;
57 bload->node_slack = xfs_globals.bload_node_slack;
58
59 if (sc->ops->type == ST_PERAG) {
60 free = sc->sa.pag->pagf_freeblks;
61 sz = xfs_ag_block_count(sc->mp, sc->sa.pag->pag_agno);
62 } else {
63 free = percpu_counter_sum(&sc->mp->m_fdblocks);
64 sz = sc->mp->m_sb.sb_dblocks;
65 }
66
67 /* No further changes if there's more than 10% free space left. */
68 if (free >= div_u64(sz, 10))
69 return;
70
71 /*
72 * We're low on space; load the btrees as tightly as possible. Leave
73 * a couple of open slots in each btree block so that we don't end up
74 * splitting the btrees like crazy after a mount.
75 */
76 if (bload->leaf_slack < 0)
77 bload->leaf_slack = 2;
78 if (bload->node_slack < 0)
79 bload->node_slack = 2;
80}
81
82/* Initialize accounting resources for staging a new AG btree. */
83void
84xrep_newbt_init_ag(
85 struct xrep_newbt *xnr,
86 struct xfs_scrub *sc,
87 const struct xfs_owner_info *oinfo,
88 xfs_fsblock_t alloc_hint,
89 enum xfs_ag_resv_type resv)
90{
91 memset(xnr, 0, sizeof(struct xrep_newbt));
92 xnr->sc = sc;
93 xnr->oinfo = *oinfo; /* structure copy */
94 xnr->alloc_hint = alloc_hint;
95 xnr->resv = resv;
96 INIT_LIST_HEAD(&xnr->resv_list);
97 xnr->bload.max_dirty = XFS_B_TO_FSBT(sc->mp, 256U << 10); /* 256K */
98 xrep_newbt_estimate_slack(xnr);
99}
100
101/* Initialize accounting resources for staging a new inode fork btree. */
102int
103xrep_newbt_init_inode(
104 struct xrep_newbt *xnr,
105 struct xfs_scrub *sc,
106 int whichfork,
107 const struct xfs_owner_info *oinfo)
108{
109 struct xfs_ifork *ifp;
110
111 ifp = kmem_cache_zalloc(xfs_ifork_cache, XCHK_GFP_FLAGS);
112 if (!ifp)
113 return -ENOMEM;
114
115 xrep_newbt_init_ag(xnr, sc, oinfo,
116 XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
117 XFS_AG_RESV_NONE);
118 xnr->ifake.if_fork = ifp;
119 xnr->ifake.if_fork_size = xfs_inode_fork_size(sc->ip, whichfork);
120 return 0;
121}
122
123/*
124 * Initialize accounting resources for staging a new btree. Callers are
125 * expected to add their own reservations (and clean them up) manually.
126 */
127void
128xrep_newbt_init_bare(
129 struct xrep_newbt *xnr,
130 struct xfs_scrub *sc)
131{
132 xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
133 XFS_AG_RESV_NONE);
134}
135
136/*
137 * Designate specific blocks to be used to build our new btree. @pag must be
138 * a passive reference.
139 */
140STATIC int
141xrep_newbt_add_blocks(
142 struct xrep_newbt *xnr,
143 struct xfs_perag *pag,
144 const struct xfs_alloc_arg *args)
145{
146 struct xfs_mount *mp = xnr->sc->mp;
147 struct xrep_newbt_resv *resv;
148 int error;
149
150 resv = kmalloc(sizeof(struct xrep_newbt_resv), XCHK_GFP_FLAGS);
151 if (!resv)
152 return -ENOMEM;
153
154 INIT_LIST_HEAD(&resv->list);
155 resv->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
156 resv->len = args->len;
157 resv->used = 0;
158 resv->pag = xfs_perag_hold(pag);
159
160 if (args->tp) {
161 ASSERT(xnr->oinfo.oi_offset == 0);
162
163 error = xfs_alloc_schedule_autoreap(args, true, &resv->autoreap);
164 if (error)
165 goto out_pag;
166 }
167
168 list_add_tail(&resv->list, &xnr->resv_list);
169 return 0;
170out_pag:
171 xfs_perag_put(resv->pag);
172 kfree(resv);
173 return error;
174}
175
176/*
177 * Add an extent to the new btree reservation pool. Callers are required to
178 * reap this reservation manually if the repair is cancelled. @pag must be a
179 * passive reference.
180 */
181int
182xrep_newbt_add_extent(
183 struct xrep_newbt *xnr,
184 struct xfs_perag *pag,
185 xfs_agblock_t agbno,
186 xfs_extlen_t len)
187{
188 struct xfs_mount *mp = xnr->sc->mp;
189 struct xfs_alloc_arg args = {
190 .tp = NULL, /* no autoreap */
191 .oinfo = xnr->oinfo,
192 .fsbno = XFS_AGB_TO_FSB(mp, pag->pag_agno, agbno),
193 .len = len,
194 .resv = xnr->resv,
195 };
196
197 return xrep_newbt_add_blocks(xnr, pag, &args);
198}
199
200/* Don't let our allocation hint take us beyond this AG */
201static inline void
202xrep_newbt_validate_ag_alloc_hint(
203 struct xrep_newbt *xnr)
204{
205 struct xfs_scrub *sc = xnr->sc;
206 xfs_agnumber_t agno = XFS_FSB_TO_AGNO(sc->mp, xnr->alloc_hint);
207
208 if (agno == sc->sa.pag->pag_agno &&
209 xfs_verify_fsbno(sc->mp, xnr->alloc_hint))
210 return;
211
212 xnr->alloc_hint = XFS_AGB_TO_FSB(sc->mp, sc->sa.pag->pag_agno,
213 XFS_AGFL_BLOCK(sc->mp) + 1);
214}
215
216/* Allocate disk space for a new per-AG btree. */
217STATIC int
218xrep_newbt_alloc_ag_blocks(
219 struct xrep_newbt *xnr,
220 uint64_t nr_blocks)
221{
222 struct xfs_scrub *sc = xnr->sc;
223 struct xfs_mount *mp = sc->mp;
224 int error = 0;
225
226 ASSERT(sc->sa.pag != NULL);
227
228 while (nr_blocks > 0) {
229 struct xfs_alloc_arg args = {
230 .tp = sc->tp,
231 .mp = mp,
232 .oinfo = xnr->oinfo,
233 .minlen = 1,
234 .maxlen = nr_blocks,
235 .prod = 1,
236 .resv = xnr->resv,
237 };
238 xfs_agnumber_t agno;
239
240 xrep_newbt_validate_ag_alloc_hint(xnr);
241
242 error = xfs_alloc_vextent_near_bno(&args, xnr->alloc_hint);
243 if (error)
244 return error;
245 if (args.fsbno == NULLFSBLOCK)
246 return -ENOSPC;
247
248 agno = XFS_FSB_TO_AGNO(mp, args.fsbno);
249
250 trace_xrep_newbt_alloc_ag_blocks(mp, agno,
251 XFS_FSB_TO_AGBNO(mp, args.fsbno), args.len,
252 xnr->oinfo.oi_owner);
253
254 if (agno != sc->sa.pag->pag_agno) {
255 ASSERT(agno == sc->sa.pag->pag_agno);
256 return -EFSCORRUPTED;
257 }
258
259 error = xrep_newbt_add_blocks(xnr, sc->sa.pag, &args);
260 if (error)
261 return error;
262
263 nr_blocks -= args.len;
264 xnr->alloc_hint = args.fsbno + args.len;
265
266 error = xrep_defer_finish(sc);
267 if (error)
268 return error;
269 }
270
271 return 0;
272}
273
274/* Don't let our allocation hint take us beyond EOFS */
275static inline void
276xrep_newbt_validate_file_alloc_hint(
277 struct xrep_newbt *xnr)
278{
279 struct xfs_scrub *sc = xnr->sc;
280
281 if (xfs_verify_fsbno(sc->mp, xnr->alloc_hint))
282 return;
283
284 xnr->alloc_hint = XFS_AGB_TO_FSB(sc->mp, 0, XFS_AGFL_BLOCK(sc->mp) + 1);
285}
286
287/* Allocate disk space for our new file-based btree. */
288STATIC int
289xrep_newbt_alloc_file_blocks(
290 struct xrep_newbt *xnr,
291 uint64_t nr_blocks)
292{
293 struct xfs_scrub *sc = xnr->sc;
294 struct xfs_mount *mp = sc->mp;
295 int error = 0;
296
297 while (nr_blocks > 0) {
298 struct xfs_alloc_arg args = {
299 .tp = sc->tp,
300 .mp = mp,
301 .oinfo = xnr->oinfo,
302 .minlen = 1,
303 .maxlen = nr_blocks,
304 .prod = 1,
305 .resv = xnr->resv,
306 };
307 struct xfs_perag *pag;
308 xfs_agnumber_t agno;
309
310 xrep_newbt_validate_file_alloc_hint(xnr);
311
312 error = xfs_alloc_vextent_start_ag(&args, xnr->alloc_hint);
313 if (error)
314 return error;
315 if (args.fsbno == NULLFSBLOCK)
316 return -ENOSPC;
317
318 agno = XFS_FSB_TO_AGNO(mp, args.fsbno);
319
320 trace_xrep_newbt_alloc_file_blocks(mp, agno,
321 XFS_FSB_TO_AGBNO(mp, args.fsbno), args.len,
322 xnr->oinfo.oi_owner);
323
324 pag = xfs_perag_get(mp, agno);
325 if (!pag) {
326 ASSERT(0);
327 return -EFSCORRUPTED;
328 }
329
330 error = xrep_newbt_add_blocks(xnr, pag, &args);
331 xfs_perag_put(pag);
332 if (error)
333 return error;
334
335 nr_blocks -= args.len;
336 xnr->alloc_hint = args.fsbno + args.len;
337
338 error = xrep_defer_finish(sc);
339 if (error)
340 return error;
341 }
342
343 return 0;
344}
345
346/* Allocate disk space for our new btree. */
347int
348xrep_newbt_alloc_blocks(
349 struct xrep_newbt *xnr,
350 uint64_t nr_blocks)
351{
352 if (xnr->sc->ip)
353 return xrep_newbt_alloc_file_blocks(xnr, nr_blocks);
354 return xrep_newbt_alloc_ag_blocks(xnr, nr_blocks);
355}
356
357/*
358 * Free the unused part of a space extent that was reserved for a new ondisk
359 * structure. Returns the number of EFIs logged or a negative errno.
360 */
361STATIC int
362xrep_newbt_free_extent(
363 struct xrep_newbt *xnr,
364 struct xrep_newbt_resv *resv,
365 bool btree_committed)
366{
367 struct xfs_scrub *sc = xnr->sc;
368 xfs_agblock_t free_agbno = resv->agbno;
369 xfs_extlen_t free_aglen = resv->len;
370 xfs_fsblock_t fsbno;
371 int error;
372
373 if (!btree_committed || resv->used == 0) {
374 /*
375 * If we're not committing a new btree or we didn't use the
376 * space reservation, let the existing EFI free the entire
377 * space extent.
378 */
379 trace_xrep_newbt_free_blocks(sc->mp, resv->pag->pag_agno,
380 free_agbno, free_aglen, xnr->oinfo.oi_owner);
381 xfs_alloc_commit_autoreap(sc->tp, &resv->autoreap);
382 return 1;
383 }
384
385 /*
386 * We used space and committed the btree. Cancel the autoreap, remove
387 * the written blocks from the reservation, and possibly log a new EFI
388 * to free any unused reservation space.
389 */
390 xfs_alloc_cancel_autoreap(sc->tp, &resv->autoreap);
391 free_agbno += resv->used;
392 free_aglen -= resv->used;
393
394 if (free_aglen == 0)
395 return 0;
396
397 trace_xrep_newbt_free_blocks(sc->mp, resv->pag->pag_agno, free_agbno,
398 free_aglen, xnr->oinfo.oi_owner);
399
400 ASSERT(xnr->resv != XFS_AG_RESV_AGFL);
401 ASSERT(xnr->resv != XFS_AG_RESV_IGNORE);
402
403 /*
404 * Use EFIs to free the reservations. This reduces the chance
405 * that we leak blocks if the system goes down.
406 */
407 fsbno = XFS_AGB_TO_FSB(sc->mp, resv->pag->pag_agno, free_agbno);
408 error = xfs_free_extent_later(sc->tp, fsbno, free_aglen, &xnr->oinfo,
409 xnr->resv, true);
410 if (error)
411 return error;
412
413 return 1;
414}
415
416/* Free all the accounting info and disk space we reserved for a new btree. */
417STATIC int
418xrep_newbt_free(
419 struct xrep_newbt *xnr,
420 bool btree_committed)
421{
422 struct xfs_scrub *sc = xnr->sc;
423 struct xrep_newbt_resv *resv, *n;
424 unsigned int freed = 0;
425 int error = 0;
426
427 /*
428 * If the filesystem already went down, we can't free the blocks. Skip
429 * ahead to freeing the incore metadata because we can't fix anything.
430 */
431 if (xfs_is_shutdown(sc->mp))
432 goto junkit;
433
434 list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
435 int ret;
436
437 ret = xrep_newbt_free_extent(xnr, resv, btree_committed);
438 list_del(&resv->list);
439 xfs_perag_put(resv->pag);
440 kfree(resv);
441 if (ret < 0) {
442 error = ret;
443 goto junkit;
444 }
445
446 freed += ret;
447 if (freed >= XREP_MAX_ITRUNCATE_EFIS) {
448 error = xrep_defer_finish(sc);
449 if (error)
450 goto junkit;
451 freed = 0;
452 }
453 }
454
455 if (freed)
456 error = xrep_defer_finish(sc);
457
458junkit:
459 /*
460 * If we still have reservations attached to @newbt, cleanup must have
461 * failed and the filesystem is about to go down. Clean up the incore
462 * reservations and try to commit to freeing the space we used.
463 */
464 list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
465 xfs_alloc_commit_autoreap(sc->tp, &resv->autoreap);
466 list_del(&resv->list);
467 xfs_perag_put(resv->pag);
468 kfree(resv);
469 }
470
471 if (sc->ip) {
472 kmem_cache_free(xfs_ifork_cache, xnr->ifake.if_fork);
473 xnr->ifake.if_fork = NULL;
474 }
475
476 return error;
477}
478
479/*
480 * Free all the accounting info and unused disk space allocations after
481 * committing a new btree.
482 */
483int
484xrep_newbt_commit(
485 struct xrep_newbt *xnr)
486{
487 return xrep_newbt_free(xnr, true);
488}
489
490/*
491 * Free all the accounting info and all of the disk space we reserved for a new
492 * btree that we're not going to commit. We want to try to roll things back
493 * cleanly for things like ENOSPC midway through allocation.
494 */
495void
496xrep_newbt_cancel(
497 struct xrep_newbt *xnr)
498{
499 xrep_newbt_free(xnr, false);
500}
501
502/* Feed one of the reserved btree blocks to the bulk loader. */
503int
504xrep_newbt_claim_block(
505 struct xfs_btree_cur *cur,
506 struct xrep_newbt *xnr,
507 union xfs_btree_ptr *ptr)
508{
509 struct xrep_newbt_resv *resv;
510 struct xfs_mount *mp = cur->bc_mp;
511 xfs_agblock_t agbno;
512
513 /*
514 * The first item in the list should always have a free block unless
515 * we're completely out.
516 */
517 resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
518 if (resv->used == resv->len)
519 return -ENOSPC;
520
521 /*
522 * Peel off a block from the start of the reservation. We allocate
523 * blocks in order to place blocks on disk in increasing record or key
524 * order. The block reservations tend to end up on the list in
525 * decreasing order, which hopefully results in leaf blocks ending up
526 * together.
527 */
528 agbno = resv->agbno + resv->used;
529 resv->used++;
530
531 /* If we used all the blocks in this reservation, move it to the end. */
532 if (resv->used == resv->len)
533 list_move_tail(&resv->list, &xnr->resv_list);
534
535 trace_xrep_newbt_claim_block(mp, resv->pag->pag_agno, agbno, 1,
536 xnr->oinfo.oi_owner);
537
538 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
539 ptr->l = cpu_to_be64(XFS_AGB_TO_FSB(mp, resv->pag->pag_agno,
540 agbno));
541 else
542 ptr->s = cpu_to_be32(agbno);
543
544 /* Relog all the EFIs. */
545 return xrep_defer_finish(xnr->sc);
546}
547
548/* How many reserved blocks are unused? */
549unsigned int
550xrep_newbt_unused_blocks(
551 struct xrep_newbt *xnr)
552{
553 struct xrep_newbt_resv *resv;
554 unsigned int unused = 0;
555
556 list_for_each_entry(resv, &xnr->resv_list, list)
557 unused += resv->len - resv->used;
558 return unused;
559}