1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2018 Red Hat, Inc.
5 * All rights reserved.
6 */
7
8 #include "xfs.h"
9 #include "xfs_fs.h"
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_sb.h"
15 #include "xfs_mount.h"
16 #include "xfs_btree.h"
17 #include "xfs_alloc_btree.h"
18 #include "xfs_rmap_btree.h"
19 #include "xfs_alloc.h"
20 #include "xfs_ialloc.h"
21 #include "xfs_rmap.h"
22 #include "xfs_ag.h"
23 #include "xfs_ag_resv.h"
24 #include "xfs_health.h"
25 #include "xfs_error.h"
26 #include "xfs_bmap.h"
27 #include "xfs_defer.h"
28 #include "xfs_log_format.h"
29 #include "xfs_trans.h"
30 #include "xfs_trace.h"
31 #include "xfs_inode.h"
32 #include "xfs_icache.h"
33
34
35 /*
36 * Passive reference counting access wrappers to the perag structures. If the
37 * per-ag structure is to be freed, the freeing code is responsible for cleaning
38 * up objects with passive references before freeing the structure. This is
39 * things like cached buffers.
40 */
41 struct xfs_perag *
xfs_perag_get(struct xfs_mount * mp,xfs_agnumber_t agno)42 xfs_perag_get(
43 struct xfs_mount *mp,
44 xfs_agnumber_t agno)
45 {
46 struct xfs_perag *pag;
47
48 rcu_read_lock();
49 pag = xa_load(&mp->m_perags, agno);
50 if (pag) {
51 trace_xfs_perag_get(pag, _RET_IP_);
52 ASSERT(atomic_read(&pag->pag_ref) >= 0);
53 atomic_inc(&pag->pag_ref);
54 }
55 rcu_read_unlock();
56 return pag;
57 }
58
59 /* Get a passive reference to the given perag. */
60 struct xfs_perag *
xfs_perag_hold(struct xfs_perag * pag)61 xfs_perag_hold(
62 struct xfs_perag *pag)
63 {
64 ASSERT(atomic_read(&pag->pag_ref) > 0 ||
65 atomic_read(&pag->pag_active_ref) > 0);
66
67 trace_xfs_perag_hold(pag, _RET_IP_);
68 atomic_inc(&pag->pag_ref);
69 return pag;
70 }
71
72 void
xfs_perag_put(struct xfs_perag * pag)73 xfs_perag_put(
74 struct xfs_perag *pag)
75 {
76 trace_xfs_perag_put(pag, _RET_IP_);
77 ASSERT(atomic_read(&pag->pag_ref) > 0);
78 atomic_dec(&pag->pag_ref);
79 }
80
81 /*
82 * Active references for perag structures. This is for short term access to the
83 * per ag structures for walking trees or accessing state. If an AG is being
84 * shrunk or is offline, then this will fail to find that AG and return NULL
85 * instead.
86 */
87 struct xfs_perag *
xfs_perag_grab(struct xfs_mount * mp,xfs_agnumber_t agno)88 xfs_perag_grab(
89 struct xfs_mount *mp,
90 xfs_agnumber_t agno)
91 {
92 struct xfs_perag *pag;
93
94 rcu_read_lock();
95 pag = xa_load(&mp->m_perags, agno);
96 if (pag) {
97 trace_xfs_perag_grab(pag, _RET_IP_);
98 if (!atomic_inc_not_zero(&pag->pag_active_ref))
99 pag = NULL;
100 }
101 rcu_read_unlock();
102 return pag;
103 }
104
105 void
xfs_perag_rele(struct xfs_perag * pag)106 xfs_perag_rele(
107 struct xfs_perag *pag)
108 {
109 trace_xfs_perag_rele(pag, _RET_IP_);
110 if (atomic_dec_and_test(&pag->pag_active_ref))
111 wake_up(&pag->pag_active_wq);
112 }
113
114 /*
115 * xfs_initialize_perag_data
116 *
117 * Read in each per-ag structure so we can count up the number of
118 * allocated inodes, free inodes and used filesystem blocks as this
119 * information is no longer persistent in the superblock. Once we have
120 * this information, write it into the in-core superblock structure.
121 */
122 int
xfs_initialize_perag_data(struct xfs_mount * mp,xfs_agnumber_t agcount)123 xfs_initialize_perag_data(
124 struct xfs_mount *mp,
125 xfs_agnumber_t agcount)
126 {
127 xfs_agnumber_t index;
128 struct xfs_perag *pag;
129 struct xfs_sb *sbp = &mp->m_sb;
130 uint64_t ifree = 0;
131 uint64_t ialloc = 0;
132 uint64_t bfree = 0;
133 uint64_t bfreelst = 0;
134 uint64_t btree = 0;
135 uint64_t fdblocks;
136 int error = 0;
137
138 for (index = 0; index < agcount; index++) {
139 /*
140 * Read the AGF and AGI buffers to populate the per-ag
141 * structures for us.
142 */
143 pag = xfs_perag_get(mp, index);
144 error = xfs_alloc_read_agf(pag, NULL, 0, NULL);
145 if (!error)
146 error = xfs_ialloc_read_agi(pag, NULL, 0, NULL);
147 if (error) {
148 xfs_perag_put(pag);
149 return error;
150 }
151
152 ifree += pag->pagi_freecount;
153 ialloc += pag->pagi_count;
154 bfree += pag->pagf_freeblks;
155 bfreelst += pag->pagf_flcount;
156 btree += pag->pagf_btreeblks;
157 xfs_perag_put(pag);
158 }
159 fdblocks = bfree + bfreelst + btree;
160
161 /*
162 * If the new summary counts are obviously incorrect, fail the
163 * mount operation because that implies the AGFs are also corrupt.
164 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
165 * will prevent xfs_repair from fixing anything.
166 */
167 if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
168 xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
169 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
170 error = -EFSCORRUPTED;
171 goto out;
172 }
173
174 /* Overwrite incore superblock counters with just-read data */
175 spin_lock(&mp->m_sb_lock);
176 sbp->sb_ifree = ifree;
177 sbp->sb_icount = ialloc;
178 sbp->sb_fdblocks = fdblocks;
179 spin_unlock(&mp->m_sb_lock);
180
181 xfs_reinit_percpu_counters(mp);
182 out:
183 xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
184 return error;
185 }
186
187 /*
188 * Free up the per-ag resources associated with the mount structure.
189 */
190 void
xfs_free_perag(struct xfs_mount * mp)191 xfs_free_perag(
192 struct xfs_mount *mp)
193 {
194 struct xfs_perag *pag;
195 xfs_agnumber_t agno;
196
197 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
198 pag = xa_erase(&mp->m_perags, agno);
199 ASSERT(pag);
200 XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_ref) != 0);
201 xfs_defer_drain_free(&pag->pag_intents_drain);
202
203 cancel_delayed_work_sync(&pag->pag_blockgc_work);
204 xfs_buf_cache_destroy(&pag->pag_bcache);
205
206 /* drop the mount's active reference */
207 xfs_perag_rele(pag);
208 XFS_IS_CORRUPT(pag->pag_mount,
209 atomic_read(&pag->pag_active_ref) != 0);
210 kfree_rcu_mightsleep(pag);
211 }
212 }
213
214 /* Find the size of the AG, in blocks. */
215 static xfs_agblock_t
__xfs_ag_block_count(struct xfs_mount * mp,xfs_agnumber_t agno,xfs_agnumber_t agcount,xfs_rfsblock_t dblocks)216 __xfs_ag_block_count(
217 struct xfs_mount *mp,
218 xfs_agnumber_t agno,
219 xfs_agnumber_t agcount,
220 xfs_rfsblock_t dblocks)
221 {
222 ASSERT(agno < agcount);
223
224 if (agno < agcount - 1)
225 return mp->m_sb.sb_agblocks;
226 return dblocks - (agno * mp->m_sb.sb_agblocks);
227 }
228
229 xfs_agblock_t
xfs_ag_block_count(struct xfs_mount * mp,xfs_agnumber_t agno)230 xfs_ag_block_count(
231 struct xfs_mount *mp,
232 xfs_agnumber_t agno)
233 {
234 return __xfs_ag_block_count(mp, agno, mp->m_sb.sb_agcount,
235 mp->m_sb.sb_dblocks);
236 }
237
238 /* Calculate the first and last possible inode number in an AG. */
239 static void
__xfs_agino_range(struct xfs_mount * mp,xfs_agblock_t eoag,xfs_agino_t * first,xfs_agino_t * last)240 __xfs_agino_range(
241 struct xfs_mount *mp,
242 xfs_agblock_t eoag,
243 xfs_agino_t *first,
244 xfs_agino_t *last)
245 {
246 xfs_agblock_t bno;
247
248 /*
249 * Calculate the first inode, which will be in the first
250 * cluster-aligned block after the AGFL.
251 */
252 bno = round_up(XFS_AGFL_BLOCK(mp) + 1, M_IGEO(mp)->cluster_align);
253 *first = XFS_AGB_TO_AGINO(mp, bno);
254
255 /*
256 * Calculate the last inode, which will be at the end of the
257 * last (aligned) cluster that can be allocated in the AG.
258 */
259 bno = round_down(eoag, M_IGEO(mp)->cluster_align);
260 *last = XFS_AGB_TO_AGINO(mp, bno) - 1;
261 }
262
263 void
xfs_agino_range(struct xfs_mount * mp,xfs_agnumber_t agno,xfs_agino_t * first,xfs_agino_t * last)264 xfs_agino_range(
265 struct xfs_mount *mp,
266 xfs_agnumber_t agno,
267 xfs_agino_t *first,
268 xfs_agino_t *last)
269 {
270 return __xfs_agino_range(mp, xfs_ag_block_count(mp, agno), first, last);
271 }
272
273 /*
274 * Free perag within the specified AG range, it is only used to free unused
275 * perags under the error handling path.
276 */
277 void
xfs_free_unused_perag_range(struct xfs_mount * mp,xfs_agnumber_t agstart,xfs_agnumber_t agend)278 xfs_free_unused_perag_range(
279 struct xfs_mount *mp,
280 xfs_agnumber_t agstart,
281 xfs_agnumber_t agend)
282 {
283 struct xfs_perag *pag;
284 xfs_agnumber_t index;
285
286 for (index = agstart; index < agend; index++) {
287 pag = xa_erase(&mp->m_perags, index);
288 if (!pag)
289 break;
290 xfs_buf_cache_destroy(&pag->pag_bcache);
291 xfs_defer_drain_free(&pag->pag_intents_drain);
292 kfree(pag);
293 }
294 }
295
296 int
xfs_initialize_perag(struct xfs_mount * mp,xfs_agnumber_t agcount,xfs_rfsblock_t dblocks,xfs_agnumber_t * maxagi)297 xfs_initialize_perag(
298 struct xfs_mount *mp,
299 xfs_agnumber_t agcount,
300 xfs_rfsblock_t dblocks,
301 xfs_agnumber_t *maxagi)
302 {
303 struct xfs_perag *pag;
304 xfs_agnumber_t index;
305 xfs_agnumber_t first_initialised = NULLAGNUMBER;
306 int error;
307
308 /*
309 * Walk the current per-ag tree so we don't try to initialise AGs
310 * that already exist (growfs case). Allocate and insert all the
311 * AGs we don't find ready for initialisation.
312 */
313 for (index = 0; index < agcount; index++) {
314 pag = xfs_perag_get(mp, index);
315 if (pag) {
316 xfs_perag_put(pag);
317 continue;
318 }
319
320 pag = kzalloc(sizeof(*pag), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
321 if (!pag) {
322 error = -ENOMEM;
323 goto out_unwind_new_pags;
324 }
325 pag->pag_agno = index;
326 pag->pag_mount = mp;
327
328 error = xa_insert(&mp->m_perags, index, pag, GFP_KERNEL);
329 if (error) {
330 WARN_ON_ONCE(error == -EBUSY);
331 goto out_free_pag;
332 }
333
334 #ifdef __KERNEL__
335 /* Place kernel structure only init below this point. */
336 spin_lock_init(&pag->pag_ici_lock);
337 spin_lock_init(&pag->pagb_lock);
338 spin_lock_init(&pag->pag_state_lock);
339 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
340 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
341 xfs_defer_drain_init(&pag->pag_intents_drain);
342 init_waitqueue_head(&pag->pagb_wait);
343 init_waitqueue_head(&pag->pag_active_wq);
344 pag->pagb_count = 0;
345 pag->pagb_tree = RB_ROOT;
346 xfs_hooks_init(&pag->pag_rmap_update_hooks);
347 #endif /* __KERNEL__ */
348
349 error = xfs_buf_cache_init(&pag->pag_bcache);
350 if (error)
351 goto out_remove_pag;
352
353 /* Active ref owned by mount indicates AG is online. */
354 atomic_set(&pag->pag_active_ref, 1);
355
356 /* first new pag is fully initialized */
357 if (first_initialised == NULLAGNUMBER)
358 first_initialised = index;
359
360 /*
361 * Pre-calculated geometry
362 */
363 pag->block_count = __xfs_ag_block_count(mp, index, agcount,
364 dblocks);
365 pag->min_block = XFS_AGFL_BLOCK(mp);
366 __xfs_agino_range(mp, pag->block_count, &pag->agino_min,
367 &pag->agino_max);
368 }
369
370 index = xfs_set_inode_alloc(mp, agcount);
371
372 if (maxagi)
373 *maxagi = index;
374
375 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
376 return 0;
377
378 out_remove_pag:
379 xfs_defer_drain_free(&pag->pag_intents_drain);
380 pag = xa_erase(&mp->m_perags, index);
381 out_free_pag:
382 kfree(pag);
383 out_unwind_new_pags:
384 /* unwind any prior newly initialized pags */
385 xfs_free_unused_perag_range(mp, first_initialised, agcount);
386 return error;
387 }
388
389 static int
xfs_get_aghdr_buf(struct xfs_mount * mp,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp,const struct xfs_buf_ops * ops)390 xfs_get_aghdr_buf(
391 struct xfs_mount *mp,
392 xfs_daddr_t blkno,
393 size_t numblks,
394 struct xfs_buf **bpp,
395 const struct xfs_buf_ops *ops)
396 {
397 struct xfs_buf *bp;
398 int error;
399
400 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
401 if (error)
402 return error;
403
404 bp->b_maps[0].bm_bn = blkno;
405 bp->b_ops = ops;
406
407 *bpp = bp;
408 return 0;
409 }
410
411 /*
412 * Generic btree root block init function
413 */
414 static void
xfs_btroot_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)415 xfs_btroot_init(
416 struct xfs_mount *mp,
417 struct xfs_buf *bp,
418 struct aghdr_init_data *id)
419 {
420 xfs_btree_init_buf(mp, bp, id->bc_ops, 0, 0, id->agno);
421 }
422
423 /* Finish initializing a free space btree. */
424 static void
xfs_freesp_init_recs(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)425 xfs_freesp_init_recs(
426 struct xfs_mount *mp,
427 struct xfs_buf *bp,
428 struct aghdr_init_data *id)
429 {
430 struct xfs_alloc_rec *arec;
431 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
432
433 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
434 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
435
436 if (xfs_ag_contains_log(mp, id->agno)) {
437 struct xfs_alloc_rec *nrec;
438 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
439 mp->m_sb.sb_logstart);
440
441 ASSERT(start >= mp->m_ag_prealloc_blocks);
442 if (start != mp->m_ag_prealloc_blocks) {
443 /*
444 * Modify first record to pad stripe align of log and
445 * bump the record count.
446 */
447 arec->ar_blockcount = cpu_to_be32(start -
448 mp->m_ag_prealloc_blocks);
449 be16_add_cpu(&block->bb_numrecs, 1);
450 nrec = arec + 1;
451
452 /*
453 * Insert second record at start of internal log
454 * which then gets trimmed.
455 */
456 nrec->ar_startblock = cpu_to_be32(
457 be32_to_cpu(arec->ar_startblock) +
458 be32_to_cpu(arec->ar_blockcount));
459 arec = nrec;
460 }
461 /*
462 * Change record start to after the internal log
463 */
464 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
465 }
466
467 /*
468 * Calculate the block count of this record; if it is nonzero,
469 * increment the record count.
470 */
471 arec->ar_blockcount = cpu_to_be32(id->agsize -
472 be32_to_cpu(arec->ar_startblock));
473 if (arec->ar_blockcount)
474 be16_add_cpu(&block->bb_numrecs, 1);
475 }
476
477 /*
478 * bnobt/cntbt btree root block init functions
479 */
480 static void
xfs_bnoroot_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)481 xfs_bnoroot_init(
482 struct xfs_mount *mp,
483 struct xfs_buf *bp,
484 struct aghdr_init_data *id)
485 {
486 xfs_btree_init_buf(mp, bp, id->bc_ops, 0, 0, id->agno);
487 xfs_freesp_init_recs(mp, bp, id);
488 }
489
490 /*
491 * Reverse map root block init
492 */
493 static void
xfs_rmaproot_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)494 xfs_rmaproot_init(
495 struct xfs_mount *mp,
496 struct xfs_buf *bp,
497 struct aghdr_init_data *id)
498 {
499 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
500 struct xfs_rmap_rec *rrec;
501
502 xfs_btree_init_buf(mp, bp, id->bc_ops, 0, 4, id->agno);
503
504 /*
505 * mark the AG header regions as static metadata The BNO
506 * btree block is the first block after the headers, so
507 * it's location defines the size of region the static
508 * metadata consumes.
509 *
510 * Note: unlike mkfs, we never have to account for log
511 * space when growing the data regions
512 */
513 rrec = XFS_RMAP_REC_ADDR(block, 1);
514 rrec->rm_startblock = 0;
515 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
516 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
517 rrec->rm_offset = 0;
518
519 /* account freespace btree root blocks */
520 rrec = XFS_RMAP_REC_ADDR(block, 2);
521 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
522 rrec->rm_blockcount = cpu_to_be32(2);
523 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
524 rrec->rm_offset = 0;
525
526 /* account inode btree root blocks */
527 rrec = XFS_RMAP_REC_ADDR(block, 3);
528 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
529 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
530 XFS_IBT_BLOCK(mp));
531 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
532 rrec->rm_offset = 0;
533
534 /* account for rmap btree root */
535 rrec = XFS_RMAP_REC_ADDR(block, 4);
536 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
537 rrec->rm_blockcount = cpu_to_be32(1);
538 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
539 rrec->rm_offset = 0;
540
541 /* account for refc btree root */
542 if (xfs_has_reflink(mp)) {
543 rrec = XFS_RMAP_REC_ADDR(block, 5);
544 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
545 rrec->rm_blockcount = cpu_to_be32(1);
546 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
547 rrec->rm_offset = 0;
548 be16_add_cpu(&block->bb_numrecs, 1);
549 }
550
551 /* account for the log space */
552 if (xfs_ag_contains_log(mp, id->agno)) {
553 rrec = XFS_RMAP_REC_ADDR(block,
554 be16_to_cpu(block->bb_numrecs) + 1);
555 rrec->rm_startblock = cpu_to_be32(
556 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
557 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
558 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
559 rrec->rm_offset = 0;
560 be16_add_cpu(&block->bb_numrecs, 1);
561 }
562 }
563
564 /*
565 * Initialise new secondary superblocks with the pre-grow geometry, but mark
566 * them as "in progress" so we know they haven't yet been activated. This will
567 * get cleared when the update with the new geometry information is done after
568 * changes to the primary are committed. This isn't strictly necessary, but we
569 * get it for free with the delayed buffer write lists and it means we can tell
570 * if a grow operation didn't complete properly after the fact.
571 */
572 static void
xfs_sbblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)573 xfs_sbblock_init(
574 struct xfs_mount *mp,
575 struct xfs_buf *bp,
576 struct aghdr_init_data *id)
577 {
578 struct xfs_dsb *dsb = bp->b_addr;
579
580 xfs_sb_to_disk(dsb, &mp->m_sb);
581 dsb->sb_inprogress = 1;
582 }
583
584 static void
xfs_agfblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)585 xfs_agfblock_init(
586 struct xfs_mount *mp,
587 struct xfs_buf *bp,
588 struct aghdr_init_data *id)
589 {
590 struct xfs_agf *agf = bp->b_addr;
591 xfs_extlen_t tmpsize;
592
593 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
594 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
595 agf->agf_seqno = cpu_to_be32(id->agno);
596 agf->agf_length = cpu_to_be32(id->agsize);
597 agf->agf_bno_root = cpu_to_be32(XFS_BNO_BLOCK(mp));
598 agf->agf_cnt_root = cpu_to_be32(XFS_CNT_BLOCK(mp));
599 agf->agf_bno_level = cpu_to_be32(1);
600 agf->agf_cnt_level = cpu_to_be32(1);
601 if (xfs_has_rmapbt(mp)) {
602 agf->agf_rmap_root = cpu_to_be32(XFS_RMAP_BLOCK(mp));
603 agf->agf_rmap_level = cpu_to_be32(1);
604 agf->agf_rmap_blocks = cpu_to_be32(1);
605 }
606
607 agf->agf_flfirst = cpu_to_be32(1);
608 agf->agf_fllast = 0;
609 agf->agf_flcount = 0;
610 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
611 agf->agf_freeblks = cpu_to_be32(tmpsize);
612 agf->agf_longest = cpu_to_be32(tmpsize);
613 if (xfs_has_crc(mp))
614 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
615 if (xfs_has_reflink(mp)) {
616 agf->agf_refcount_root = cpu_to_be32(
617 xfs_refc_block(mp));
618 agf->agf_refcount_level = cpu_to_be32(1);
619 agf->agf_refcount_blocks = cpu_to_be32(1);
620 }
621
622 if (xfs_ag_contains_log(mp, id->agno)) {
623 int64_t logblocks = mp->m_sb.sb_logblocks;
624
625 be32_add_cpu(&agf->agf_freeblks, -logblocks);
626 agf->agf_longest = cpu_to_be32(id->agsize -
627 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
628 }
629 }
630
631 static void
xfs_agflblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)632 xfs_agflblock_init(
633 struct xfs_mount *mp,
634 struct xfs_buf *bp,
635 struct aghdr_init_data *id)
636 {
637 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
638 __be32 *agfl_bno;
639 int bucket;
640
641 if (xfs_has_crc(mp)) {
642 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
643 agfl->agfl_seqno = cpu_to_be32(id->agno);
644 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
645 }
646
647 agfl_bno = xfs_buf_to_agfl_bno(bp);
648 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
649 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
650 }
651
652 static void
xfs_agiblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)653 xfs_agiblock_init(
654 struct xfs_mount *mp,
655 struct xfs_buf *bp,
656 struct aghdr_init_data *id)
657 {
658 struct xfs_agi *agi = bp->b_addr;
659 int bucket;
660
661 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
662 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
663 agi->agi_seqno = cpu_to_be32(id->agno);
664 agi->agi_length = cpu_to_be32(id->agsize);
665 agi->agi_count = 0;
666 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
667 agi->agi_level = cpu_to_be32(1);
668 agi->agi_freecount = 0;
669 agi->agi_newino = cpu_to_be32(NULLAGINO);
670 agi->agi_dirino = cpu_to_be32(NULLAGINO);
671 if (xfs_has_crc(mp))
672 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
673 if (xfs_has_finobt(mp)) {
674 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
675 agi->agi_free_level = cpu_to_be32(1);
676 }
677 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
678 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
679 if (xfs_has_inobtcounts(mp)) {
680 agi->agi_iblocks = cpu_to_be32(1);
681 if (xfs_has_finobt(mp))
682 agi->agi_fblocks = cpu_to_be32(1);
683 }
684 }
685
686 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
687 struct aghdr_init_data *id);
688 static int
xfs_ag_init_hdr(struct xfs_mount * mp,struct aghdr_init_data * id,aghdr_init_work_f work,const struct xfs_buf_ops * ops)689 xfs_ag_init_hdr(
690 struct xfs_mount *mp,
691 struct aghdr_init_data *id,
692 aghdr_init_work_f work,
693 const struct xfs_buf_ops *ops)
694 {
695 struct xfs_buf *bp;
696 int error;
697
698 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
699 if (error)
700 return error;
701
702 (*work)(mp, bp, id);
703
704 xfs_buf_delwri_queue(bp, &id->buffer_list);
705 xfs_buf_relse(bp);
706 return 0;
707 }
708
709 struct xfs_aghdr_grow_data {
710 xfs_daddr_t daddr;
711 size_t numblks;
712 const struct xfs_buf_ops *ops;
713 aghdr_init_work_f work;
714 const struct xfs_btree_ops *bc_ops;
715 bool need_init;
716 };
717
718 /*
719 * Prepare new AG headers to be written to disk. We use uncached buffers here,
720 * as it is assumed these new AG headers are currently beyond the currently
721 * valid filesystem address space. Using cached buffers would trip over EOFS
722 * corruption detection alogrithms in the buffer cache lookup routines.
723 *
724 * This is a non-transactional function, but the prepared buffers are added to a
725 * delayed write buffer list supplied by the caller so they can submit them to
726 * disk and wait on them as required.
727 */
728 int
xfs_ag_init_headers(struct xfs_mount * mp,struct aghdr_init_data * id)729 xfs_ag_init_headers(
730 struct xfs_mount *mp,
731 struct aghdr_init_data *id)
732
733 {
734 struct xfs_aghdr_grow_data aghdr_data[] = {
735 { /* SB */
736 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
737 .numblks = XFS_FSS_TO_BB(mp, 1),
738 .ops = &xfs_sb_buf_ops,
739 .work = &xfs_sbblock_init,
740 .need_init = true
741 },
742 { /* AGF */
743 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
744 .numblks = XFS_FSS_TO_BB(mp, 1),
745 .ops = &xfs_agf_buf_ops,
746 .work = &xfs_agfblock_init,
747 .need_init = true
748 },
749 { /* AGFL */
750 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
751 .numblks = XFS_FSS_TO_BB(mp, 1),
752 .ops = &xfs_agfl_buf_ops,
753 .work = &xfs_agflblock_init,
754 .need_init = true
755 },
756 { /* AGI */
757 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
758 .numblks = XFS_FSS_TO_BB(mp, 1),
759 .ops = &xfs_agi_buf_ops,
760 .work = &xfs_agiblock_init,
761 .need_init = true
762 },
763 { /* BNO root block */
764 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
765 .numblks = BTOBB(mp->m_sb.sb_blocksize),
766 .ops = &xfs_bnobt_buf_ops,
767 .work = &xfs_bnoroot_init,
768 .bc_ops = &xfs_bnobt_ops,
769 .need_init = true
770 },
771 { /* CNT root block */
772 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
773 .numblks = BTOBB(mp->m_sb.sb_blocksize),
774 .ops = &xfs_cntbt_buf_ops,
775 .work = &xfs_bnoroot_init,
776 .bc_ops = &xfs_cntbt_ops,
777 .need_init = true
778 },
779 { /* INO root block */
780 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
781 .numblks = BTOBB(mp->m_sb.sb_blocksize),
782 .ops = &xfs_inobt_buf_ops,
783 .work = &xfs_btroot_init,
784 .bc_ops = &xfs_inobt_ops,
785 .need_init = true
786 },
787 { /* FINO root block */
788 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
789 .numblks = BTOBB(mp->m_sb.sb_blocksize),
790 .ops = &xfs_finobt_buf_ops,
791 .work = &xfs_btroot_init,
792 .bc_ops = &xfs_finobt_ops,
793 .need_init = xfs_has_finobt(mp)
794 },
795 { /* RMAP root block */
796 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
797 .numblks = BTOBB(mp->m_sb.sb_blocksize),
798 .ops = &xfs_rmapbt_buf_ops,
799 .work = &xfs_rmaproot_init,
800 .bc_ops = &xfs_rmapbt_ops,
801 .need_init = xfs_has_rmapbt(mp)
802 },
803 { /* REFC root block */
804 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
805 .numblks = BTOBB(mp->m_sb.sb_blocksize),
806 .ops = &xfs_refcountbt_buf_ops,
807 .work = &xfs_btroot_init,
808 .bc_ops = &xfs_refcountbt_ops,
809 .need_init = xfs_has_reflink(mp)
810 },
811 { /* NULL terminating block */
812 .daddr = XFS_BUF_DADDR_NULL,
813 }
814 };
815 struct xfs_aghdr_grow_data *dp;
816 int error = 0;
817
818 /* Account for AG free space in new AG */
819 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
820 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
821 if (!dp->need_init)
822 continue;
823
824 id->daddr = dp->daddr;
825 id->numblks = dp->numblks;
826 id->bc_ops = dp->bc_ops;
827 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
828 if (error)
829 break;
830 }
831 return error;
832 }
833
834 int
xfs_ag_shrink_space(struct xfs_perag * pag,struct xfs_trans ** tpp,xfs_extlen_t delta)835 xfs_ag_shrink_space(
836 struct xfs_perag *pag,
837 struct xfs_trans **tpp,
838 xfs_extlen_t delta)
839 {
840 struct xfs_mount *mp = pag->pag_mount;
841 struct xfs_alloc_arg args = {
842 .tp = *tpp,
843 .mp = mp,
844 .pag = pag,
845 .minlen = delta,
846 .maxlen = delta,
847 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
848 .resv = XFS_AG_RESV_NONE,
849 .prod = 1
850 };
851 struct xfs_buf *agibp, *agfbp;
852 struct xfs_agi *agi;
853 struct xfs_agf *agf;
854 xfs_agblock_t aglen;
855 int error, err2;
856
857 ASSERT(pag->pag_agno == mp->m_sb.sb_agcount - 1);
858 error = xfs_ialloc_read_agi(pag, *tpp, 0, &agibp);
859 if (error)
860 return error;
861
862 agi = agibp->b_addr;
863
864 error = xfs_alloc_read_agf(pag, *tpp, 0, &agfbp);
865 if (error)
866 return error;
867
868 agf = agfbp->b_addr;
869 aglen = be32_to_cpu(agi->agi_length);
870 /* some extra paranoid checks before we shrink the ag */
871 if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length)) {
872 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGF);
873 return -EFSCORRUPTED;
874 }
875 if (delta >= aglen)
876 return -EINVAL;
877
878 /*
879 * Make sure that the last inode cluster cannot overlap with the new
880 * end of the AG, even if it's sparse.
881 */
882 error = xfs_ialloc_check_shrink(pag, *tpp, agibp, aglen - delta);
883 if (error)
884 return error;
885
886 /*
887 * Disable perag reservations so it doesn't cause the allocation request
888 * to fail. We'll reestablish reservation before we return.
889 */
890 xfs_ag_resv_free(pag);
891
892 /* internal log shouldn't also show up in the free space btrees */
893 error = xfs_alloc_vextent_exact_bno(&args,
894 XFS_AGB_TO_FSB(mp, pag->pag_agno, aglen - delta));
895 if (!error && args.agbno == NULLAGBLOCK)
896 error = -ENOSPC;
897
898 if (error) {
899 /*
900 * If extent allocation fails, need to roll the transaction to
901 * ensure that the AGFL fixup has been committed anyway.
902 *
903 * We need to hold the AGF across the roll to ensure nothing can
904 * access the AG for allocation until the shrink is fully
905 * cleaned up. And due to the resetting of the AG block
906 * reservation space needing to lock the AGI, we also have to
907 * hold that so we don't get AGI/AGF lock order inversions in
908 * the error handling path.
909 */
910 xfs_trans_bhold(*tpp, agfbp);
911 xfs_trans_bhold(*tpp, agibp);
912 err2 = xfs_trans_roll(tpp);
913 if (err2)
914 return err2;
915 xfs_trans_bjoin(*tpp, agfbp);
916 xfs_trans_bjoin(*tpp, agibp);
917 goto resv_init_out;
918 }
919
920 /*
921 * if successfully deleted from freespace btrees, need to confirm
922 * per-AG reservation works as expected.
923 */
924 be32_add_cpu(&agi->agi_length, -delta);
925 be32_add_cpu(&agf->agf_length, -delta);
926
927 err2 = xfs_ag_resv_init(pag, *tpp);
928 if (err2) {
929 be32_add_cpu(&agi->agi_length, delta);
930 be32_add_cpu(&agf->agf_length, delta);
931 if (err2 != -ENOSPC)
932 goto resv_err;
933
934 err2 = xfs_free_extent_later(*tpp, args.fsbno, delta, NULL,
935 XFS_AG_RESV_NONE, XFS_FREE_EXTENT_SKIP_DISCARD);
936 if (err2)
937 goto resv_err;
938
939 /*
940 * Roll the transaction before trying to re-init the per-ag
941 * reservation. The new transaction is clean so it will cancel
942 * without any side effects.
943 */
944 error = xfs_defer_finish(tpp);
945 if (error)
946 return error;
947
948 error = -ENOSPC;
949 goto resv_init_out;
950 }
951
952 /* Update perag geometry */
953 pag->block_count -= delta;
954 __xfs_agino_range(pag->pag_mount, pag->block_count, &pag->agino_min,
955 &pag->agino_max);
956
957 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
958 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
959 return 0;
960
961 resv_init_out:
962 err2 = xfs_ag_resv_init(pag, *tpp);
963 if (!err2)
964 return error;
965 resv_err:
966 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
967 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
968 return err2;
969 }
970
971 /*
972 * Extent the AG indicated by the @id by the length passed in
973 */
974 int
xfs_ag_extend_space(struct xfs_perag * pag,struct xfs_trans * tp,xfs_extlen_t len)975 xfs_ag_extend_space(
976 struct xfs_perag *pag,
977 struct xfs_trans *tp,
978 xfs_extlen_t len)
979 {
980 struct xfs_buf *bp;
981 struct xfs_agi *agi;
982 struct xfs_agf *agf;
983 int error;
984
985 ASSERT(pag->pag_agno == pag->pag_mount->m_sb.sb_agcount - 1);
986
987 error = xfs_ialloc_read_agi(pag, tp, 0, &bp);
988 if (error)
989 return error;
990
991 agi = bp->b_addr;
992 be32_add_cpu(&agi->agi_length, len);
993 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
994
995 /*
996 * Change agf length.
997 */
998 error = xfs_alloc_read_agf(pag, tp, 0, &bp);
999 if (error)
1000 return error;
1001
1002 agf = bp->b_addr;
1003 be32_add_cpu(&agf->agf_length, len);
1004 ASSERT(agf->agf_length == agi->agi_length);
1005 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
1006
1007 /*
1008 * Free the new space.
1009 *
1010 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
1011 * this doesn't actually exist in the rmap btree.
1012 */
1013 error = xfs_rmap_free(tp, bp, pag, be32_to_cpu(agf->agf_length) - len,
1014 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
1015 if (error)
1016 return error;
1017
1018 error = xfs_free_extent(tp, pag, be32_to_cpu(agf->agf_length) - len,
1019 len, &XFS_RMAP_OINFO_SKIP_UPDATE, XFS_AG_RESV_NONE);
1020 if (error)
1021 return error;
1022
1023 /* Update perag geometry */
1024 pag->block_count = be32_to_cpu(agf->agf_length);
1025 __xfs_agino_range(pag->pag_mount, pag->block_count, &pag->agino_min,
1026 &pag->agino_max);
1027 return 0;
1028 }
1029
1030 /* Retrieve AG geometry. */
1031 int
xfs_ag_get_geometry(struct xfs_perag * pag,struct xfs_ag_geometry * ageo)1032 xfs_ag_get_geometry(
1033 struct xfs_perag *pag,
1034 struct xfs_ag_geometry *ageo)
1035 {
1036 struct xfs_buf *agi_bp;
1037 struct xfs_buf *agf_bp;
1038 struct xfs_agi *agi;
1039 struct xfs_agf *agf;
1040 unsigned int freeblks;
1041 int error;
1042
1043 /* Lock the AG headers. */
1044 error = xfs_ialloc_read_agi(pag, NULL, 0, &agi_bp);
1045 if (error)
1046 return error;
1047 error = xfs_alloc_read_agf(pag, NULL, 0, &agf_bp);
1048 if (error)
1049 goto out_agi;
1050
1051 /* Fill out form. */
1052 memset(ageo, 0, sizeof(*ageo));
1053 ageo->ag_number = pag->pag_agno;
1054
1055 agi = agi_bp->b_addr;
1056 ageo->ag_icount = be32_to_cpu(agi->agi_count);
1057 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
1058
1059 agf = agf_bp->b_addr;
1060 ageo->ag_length = be32_to_cpu(agf->agf_length);
1061 freeblks = pag->pagf_freeblks +
1062 pag->pagf_flcount +
1063 pag->pagf_btreeblks -
1064 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
1065 ageo->ag_freeblks = freeblks;
1066 xfs_ag_geom_health(pag, ageo);
1067
1068 /* Release resources. */
1069 xfs_buf_relse(agf_bp);
1070 out_agi:
1071 xfs_buf_relse(agi_bp);
1072 return error;
1073 }
1074