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