xref: /linux/fs/xfs/libxfs/xfs_ag.c (revision 0b0128e64af056a7dd29fa3bc780af654e53f861)
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_platform.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 #include "xfs_group.h"
34 
35 /*
36  * xfs_initialize_perag_data
37  *
38  * Read in each per-ag structure so we can count up the number of
39  * allocated inodes, free inodes and used filesystem blocks as this
40  * information is no longer persistent in the superblock. Once we have
41  * this information, write it into the in-core superblock structure.
42  */
43 int
xfs_initialize_perag_data(struct xfs_mount * mp,xfs_agnumber_t agcount)44 xfs_initialize_perag_data(
45 	struct xfs_mount	*mp,
46 	xfs_agnumber_t		agcount)
47 {
48 	xfs_agnumber_t		index;
49 	struct xfs_perag	*pag;
50 	struct xfs_sb		*sbp = &mp->m_sb;
51 	uint64_t		ifree = 0;
52 	uint64_t		ialloc = 0;
53 	uint64_t		bfree = 0;
54 	uint64_t		bfreelst = 0;
55 	uint64_t		btree = 0;
56 	uint64_t		fdblocks;
57 	int			error = 0;
58 
59 	for (index = 0; index < agcount; index++) {
60 		/*
61 		 * Read the AGF and AGI buffers to populate the per-ag
62 		 * structures for us.
63 		 */
64 		pag = xfs_perag_get(mp, index);
65 		error = xfs_alloc_read_agf(pag, NULL, 0, NULL);
66 		if (!error)
67 			error = xfs_ialloc_read_agi(pag, NULL, 0, NULL);
68 		if (error) {
69 			xfs_perag_put(pag);
70 			return error;
71 		}
72 
73 		ifree += pag->pagi_freecount;
74 		ialloc += pag->pagi_count;
75 		bfree += pag->pagf_freeblks;
76 		bfreelst += pag->pagf_flcount;
77 		btree += pag->pagf_btreeblks;
78 		xfs_perag_put(pag);
79 	}
80 	fdblocks = bfree + bfreelst + btree;
81 
82 	/*
83 	 * If the new summary counts are obviously incorrect, fail the
84 	 * mount operation because that implies the AGFs are also corrupt.
85 	 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
86 	 * will prevent xfs_repair from fixing anything.
87 	 */
88 	if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
89 		xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
90 		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
91 		error = -EFSCORRUPTED;
92 		goto out;
93 	}
94 
95 	/* Overwrite incore superblock counters with just-read data */
96 	spin_lock(&mp->m_sb_lock);
97 	sbp->sb_ifree = ifree;
98 	sbp->sb_icount = ialloc;
99 	sbp->sb_fdblocks = fdblocks;
100 	spin_unlock(&mp->m_sb_lock);
101 
102 	xfs_reinit_percpu_counters(mp);
103 out:
104 	xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
105 	return error;
106 }
107 
108 static void
xfs_perag_uninit(struct xfs_group * xg)109 xfs_perag_uninit(
110 	struct xfs_group	*xg)
111 {
112 #ifdef __KERNEL__
113 	cancel_delayed_work_sync(&to_perag(xg)->pag_blockgc_work);
114 #endif
115 }
116 
117 /*
118  * Free up the per-ag resources  within the specified AG range.
119  */
120 void
xfs_free_perag_range(struct xfs_mount * mp,xfs_agnumber_t first_agno,xfs_agnumber_t end_agno)121 xfs_free_perag_range(
122 	struct xfs_mount	*mp,
123 	xfs_agnumber_t		first_agno,
124 	xfs_agnumber_t		end_agno)
125 
126 {
127 	xfs_agnumber_t		agno;
128 
129 	for (agno = first_agno; agno < end_agno; agno++)
130 		xfs_group_free(mp, agno, XG_TYPE_AG, xfs_perag_uninit);
131 }
132 
133 /* Find the size of the AG, in blocks. */
134 static xfs_agblock_t
__xfs_ag_block_count(struct xfs_mount * mp,xfs_agnumber_t agno,xfs_agnumber_t agcount,xfs_rfsblock_t dblocks)135 __xfs_ag_block_count(
136 	struct xfs_mount	*mp,
137 	xfs_agnumber_t		agno,
138 	xfs_agnumber_t		agcount,
139 	xfs_rfsblock_t		dblocks)
140 {
141 	ASSERT(agno < agcount);
142 
143 	if (agno < agcount - 1)
144 		return mp->m_sb.sb_agblocks;
145 	return dblocks - (agno * mp->m_sb.sb_agblocks);
146 }
147 
148 xfs_agblock_t
xfs_ag_block_count(struct xfs_mount * mp,xfs_agnumber_t agno)149 xfs_ag_block_count(
150 	struct xfs_mount	*mp,
151 	xfs_agnumber_t		agno)
152 {
153 	return __xfs_ag_block_count(mp, agno, mp->m_sb.sb_agcount,
154 			mp->m_sb.sb_dblocks);
155 }
156 
157 /* Calculate the first and last possible inode number in an AG. */
158 static void
__xfs_agino_range(struct xfs_mount * mp,xfs_agblock_t eoag,xfs_agino_t * first,xfs_agino_t * last)159 __xfs_agino_range(
160 	struct xfs_mount	*mp,
161 	xfs_agblock_t		eoag,
162 	xfs_agino_t		*first,
163 	xfs_agino_t		*last)
164 {
165 	xfs_agblock_t		bno;
166 
167 	/*
168 	 * Calculate the first inode, which will be in the first
169 	 * cluster-aligned block after the AGFL.
170 	 */
171 	bno = round_up(XFS_AGFL_BLOCK(mp) + 1, M_IGEO(mp)->cluster_align);
172 	*first = XFS_AGB_TO_AGINO(mp, bno);
173 
174 	/*
175 	 * Calculate the last inode, which will be at the end of the
176 	 * last (aligned) cluster that can be allocated in the AG.
177 	 */
178 	bno = round_down(eoag, M_IGEO(mp)->cluster_align);
179 	*last = XFS_AGB_TO_AGINO(mp, bno) - 1;
180 }
181 
182 void
xfs_agino_range(struct xfs_mount * mp,xfs_agnumber_t agno,xfs_agino_t * first,xfs_agino_t * last)183 xfs_agino_range(
184 	struct xfs_mount	*mp,
185 	xfs_agnumber_t		agno,
186 	xfs_agino_t		*first,
187 	xfs_agino_t		*last)
188 {
189 	return __xfs_agino_range(mp, xfs_ag_block_count(mp, agno), first, last);
190 }
191 
192 /*
193  * Update the perag of the previous tail AG if it has been changed during
194  * recovery (i.e. recovery of a growfs).
195  */
196 int
xfs_update_last_ag_size(struct xfs_mount * mp,xfs_agnumber_t prev_agcount)197 xfs_update_last_ag_size(
198 	struct xfs_mount	*mp,
199 	xfs_agnumber_t		prev_agcount)
200 {
201 	struct xfs_perag	*pag = xfs_perag_grab(mp, prev_agcount - 1);
202 
203 	if (!pag)
204 		return -EFSCORRUPTED;
205 	pag_group(pag)->xg_block_count = __xfs_ag_block_count(mp,
206 			prev_agcount - 1, mp->m_sb.sb_agcount,
207 			mp->m_sb.sb_dblocks);
208 	__xfs_agino_range(mp, pag_group(pag)->xg_block_count, &pag->agino_min,
209 			&pag->agino_max);
210 	xfs_perag_rele(pag);
211 	return 0;
212 }
213 
214 static int
xfs_perag_alloc(struct xfs_mount * mp,xfs_agnumber_t index,xfs_agnumber_t agcount,xfs_rfsblock_t dblocks)215 xfs_perag_alloc(
216 	struct xfs_mount	*mp,
217 	xfs_agnumber_t		index,
218 	xfs_agnumber_t		agcount,
219 	xfs_rfsblock_t		dblocks)
220 {
221 	struct xfs_perag	*pag;
222 	int			error;
223 
224 	pag = kzalloc_obj(*pag);
225 	if (!pag)
226 		return -ENOMEM;
227 
228 #ifdef __KERNEL__
229 	/* Place kernel structure only init below this point. */
230 	spin_lock_init(&pag->pag_ici_lock);
231 	INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
232 	INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
233 #endif /* __KERNEL__ */
234 
235 	/*
236 	 * Pre-calculated geometry
237 	 */
238 	pag_group(pag)->xg_block_count = __xfs_ag_block_count(mp, index, agcount,
239 				dblocks);
240 	pag_group(pag)->xg_min_gbno = XFS_AGFL_BLOCK(mp) + 1;
241 	__xfs_agino_range(mp, pag_group(pag)->xg_block_count, &pag->agino_min,
242 			&pag->agino_max);
243 
244 	error = xfs_group_insert(mp, pag_group(pag), index, XG_TYPE_AG);
245 	if (error)
246 		goto out_free_perag;
247 
248 	return 0;
249 
250 out_free_perag:
251 	kfree(pag);
252 	return error;
253 }
254 
255 int
xfs_initialize_perag(struct xfs_mount * mp,xfs_agnumber_t orig_agcount,xfs_agnumber_t new_agcount,xfs_rfsblock_t dblocks,xfs_agnumber_t * maxagi)256 xfs_initialize_perag(
257 	struct xfs_mount	*mp,
258 	xfs_agnumber_t		orig_agcount,
259 	xfs_agnumber_t		new_agcount,
260 	xfs_rfsblock_t		dblocks,
261 	xfs_agnumber_t		*maxagi)
262 {
263 	xfs_agnumber_t		index;
264 	int			error;
265 
266 	if (orig_agcount >= new_agcount)
267 		return 0;
268 
269 	for (index = orig_agcount; index < new_agcount; index++) {
270 		error = xfs_perag_alloc(mp, index, new_agcount, dblocks);
271 		if (error)
272 			goto out_unwind_new_pags;
273 	}
274 
275 	*maxagi = xfs_set_inode_alloc(mp, new_agcount);
276 	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
277 	return 0;
278 
279 out_unwind_new_pags:
280 	xfs_free_perag_range(mp, orig_agcount, index);
281 	return error;
282 }
283 
284 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)285 xfs_get_aghdr_buf(
286 	struct xfs_mount	*mp,
287 	xfs_daddr_t		blkno,
288 	size_t			numblks,
289 	struct xfs_buf		**bpp,
290 	const struct xfs_buf_ops *ops)
291 {
292 	struct xfs_buf		*bp;
293 	int			error;
294 
295 	error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, &bp);
296 	if (error)
297 		return error;
298 
299 	bp->b_maps[0].bm_bn = blkno;
300 	bp->b_ops = ops;
301 
302 	*bpp = bp;
303 	return 0;
304 }
305 
306 /*
307  * Generic btree root block init function
308  */
309 static void
xfs_btroot_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)310 xfs_btroot_init(
311 	struct xfs_mount	*mp,
312 	struct xfs_buf		*bp,
313 	struct aghdr_init_data	*id)
314 {
315 	xfs_btree_init_buf(mp, bp, id->bc_ops, 0, 0, id->agno);
316 }
317 
318 /* Finish initializing a free space btree. */
319 static void
xfs_freesp_init_recs(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)320 xfs_freesp_init_recs(
321 	struct xfs_mount	*mp,
322 	struct xfs_buf		*bp,
323 	struct aghdr_init_data	*id)
324 {
325 	struct xfs_alloc_rec	*arec;
326 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
327 
328 	arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
329 	arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
330 
331 	if (xfs_ag_contains_log(mp, id->agno)) {
332 		struct xfs_alloc_rec	*nrec;
333 		xfs_agblock_t		start = XFS_FSB_TO_AGBNO(mp,
334 							mp->m_sb.sb_logstart);
335 
336 		ASSERT(start >= mp->m_ag_prealloc_blocks);
337 		if (start != mp->m_ag_prealloc_blocks) {
338 			/*
339 			 * Modify first record to pad stripe align of log and
340 			 * bump the record count.
341 			 */
342 			arec->ar_blockcount = cpu_to_be32(start -
343 						mp->m_ag_prealloc_blocks);
344 			be16_add_cpu(&block->bb_numrecs, 1);
345 			nrec = arec + 1;
346 
347 			/*
348 			 * Insert second record at start of internal log
349 			 * which then gets trimmed.
350 			 */
351 			nrec->ar_startblock = cpu_to_be32(
352 					be32_to_cpu(arec->ar_startblock) +
353 					be32_to_cpu(arec->ar_blockcount));
354 			arec = nrec;
355 		}
356 		/*
357 		 * Change record start to after the internal log
358 		 */
359 		be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
360 	}
361 
362 	/*
363 	 * Calculate the block count of this record; if it is nonzero,
364 	 * increment the record count.
365 	 */
366 	arec->ar_blockcount = cpu_to_be32(id->agsize -
367 					  be32_to_cpu(arec->ar_startblock));
368 	if (arec->ar_blockcount)
369 		be16_add_cpu(&block->bb_numrecs, 1);
370 }
371 
372 /*
373  * bnobt/cntbt btree root block init functions
374  */
375 static void
xfs_bnoroot_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)376 xfs_bnoroot_init(
377 	struct xfs_mount	*mp,
378 	struct xfs_buf		*bp,
379 	struct aghdr_init_data	*id)
380 {
381 	xfs_btree_init_buf(mp, bp, id->bc_ops, 0, 0, id->agno);
382 	xfs_freesp_init_recs(mp, bp, id);
383 }
384 
385 /*
386  * Reverse map root block init
387  */
388 static void
xfs_rmaproot_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)389 xfs_rmaproot_init(
390 	struct xfs_mount	*mp,
391 	struct xfs_buf		*bp,
392 	struct aghdr_init_data	*id)
393 {
394 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
395 	struct xfs_rmap_rec	*rrec;
396 
397 	xfs_btree_init_buf(mp, bp, id->bc_ops, 0, 4, id->agno);
398 
399 	/*
400 	 * mark the AG header regions as static metadata The BNO
401 	 * btree block is the first block after the headers, so
402 	 * it's location defines the size of region the static
403 	 * metadata consumes.
404 	 *
405 	 * Note: unlike mkfs, we never have to account for log
406 	 * space when growing the data regions
407 	 */
408 	rrec = XFS_RMAP_REC_ADDR(block, 1);
409 	rrec->rm_startblock = 0;
410 	rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
411 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
412 	rrec->rm_offset = 0;
413 
414 	/* account freespace btree root blocks */
415 	rrec = XFS_RMAP_REC_ADDR(block, 2);
416 	rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
417 	rrec->rm_blockcount = cpu_to_be32(2);
418 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
419 	rrec->rm_offset = 0;
420 
421 	/* account inode btree root blocks */
422 	rrec = XFS_RMAP_REC_ADDR(block, 3);
423 	rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
424 	rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
425 					  XFS_IBT_BLOCK(mp));
426 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
427 	rrec->rm_offset = 0;
428 
429 	/* account for rmap btree root */
430 	rrec = XFS_RMAP_REC_ADDR(block, 4);
431 	rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
432 	rrec->rm_blockcount = cpu_to_be32(1);
433 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
434 	rrec->rm_offset = 0;
435 
436 	/* account for refc btree root */
437 	if (xfs_has_reflink(mp)) {
438 		rrec = XFS_RMAP_REC_ADDR(block, 5);
439 		rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
440 		rrec->rm_blockcount = cpu_to_be32(1);
441 		rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
442 		rrec->rm_offset = 0;
443 		be16_add_cpu(&block->bb_numrecs, 1);
444 	}
445 
446 	/* account for the log space */
447 	if (xfs_ag_contains_log(mp, id->agno)) {
448 		rrec = XFS_RMAP_REC_ADDR(block,
449 				be16_to_cpu(block->bb_numrecs) + 1);
450 		rrec->rm_startblock = cpu_to_be32(
451 				XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
452 		rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
453 		rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
454 		rrec->rm_offset = 0;
455 		be16_add_cpu(&block->bb_numrecs, 1);
456 	}
457 }
458 
459 /*
460  * Initialise new secondary superblocks with the pre-grow geometry, but mark
461  * them as "in progress" so we know they haven't yet been activated. This will
462  * get cleared when the update with the new geometry information is done after
463  * changes to the primary are committed. This isn't strictly necessary, but we
464  * get it for free with the delayed buffer write lists and it means we can tell
465  * if a grow operation didn't complete properly after the fact.
466  */
467 static void
xfs_sbblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)468 xfs_sbblock_init(
469 	struct xfs_mount	*mp,
470 	struct xfs_buf		*bp,
471 	struct aghdr_init_data	*id)
472 {
473 	struct xfs_dsb		*dsb = bp->b_addr;
474 
475 	xfs_sb_to_disk(dsb, &mp->m_sb);
476 	dsb->sb_inprogress = 1;
477 }
478 
479 static void
xfs_agfblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)480 xfs_agfblock_init(
481 	struct xfs_mount	*mp,
482 	struct xfs_buf		*bp,
483 	struct aghdr_init_data	*id)
484 {
485 	struct xfs_agf		*agf = bp->b_addr;
486 	xfs_extlen_t		tmpsize;
487 
488 	agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
489 	agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
490 	agf->agf_seqno = cpu_to_be32(id->agno);
491 	agf->agf_length = cpu_to_be32(id->agsize);
492 	agf->agf_bno_root = cpu_to_be32(XFS_BNO_BLOCK(mp));
493 	agf->agf_cnt_root = cpu_to_be32(XFS_CNT_BLOCK(mp));
494 	agf->agf_bno_level = cpu_to_be32(1);
495 	agf->agf_cnt_level = cpu_to_be32(1);
496 	if (xfs_has_rmapbt(mp)) {
497 		agf->agf_rmap_root = cpu_to_be32(XFS_RMAP_BLOCK(mp));
498 		agf->agf_rmap_level = cpu_to_be32(1);
499 		agf->agf_rmap_blocks = cpu_to_be32(1);
500 	}
501 
502 	agf->agf_flfirst = cpu_to_be32(1);
503 	agf->agf_fllast = 0;
504 	agf->agf_flcount = 0;
505 	tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
506 	agf->agf_freeblks = cpu_to_be32(tmpsize);
507 	agf->agf_longest = cpu_to_be32(tmpsize);
508 	if (xfs_has_crc(mp))
509 		uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
510 	if (xfs_has_reflink(mp)) {
511 		agf->agf_refcount_root = cpu_to_be32(
512 				xfs_refc_block(mp));
513 		agf->agf_refcount_level = cpu_to_be32(1);
514 		agf->agf_refcount_blocks = cpu_to_be32(1);
515 	}
516 
517 	if (xfs_ag_contains_log(mp, id->agno)) {
518 		int64_t	logblocks = mp->m_sb.sb_logblocks;
519 
520 		be32_add_cpu(&agf->agf_freeblks, -logblocks);
521 		agf->agf_longest = cpu_to_be32(id->agsize -
522 			XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
523 	}
524 }
525 
526 static void
xfs_agflblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)527 xfs_agflblock_init(
528 	struct xfs_mount	*mp,
529 	struct xfs_buf		*bp,
530 	struct aghdr_init_data	*id)
531 {
532 	struct xfs_agfl		*agfl = XFS_BUF_TO_AGFL(bp);
533 	__be32			*agfl_bno;
534 	int			bucket;
535 
536 	if (xfs_has_crc(mp)) {
537 		agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
538 		agfl->agfl_seqno = cpu_to_be32(id->agno);
539 		uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
540 	}
541 
542 	agfl_bno = xfs_buf_to_agfl_bno(bp);
543 	for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
544 		agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
545 }
546 
547 static void
xfs_agiblock_init(struct xfs_mount * mp,struct xfs_buf * bp,struct aghdr_init_data * id)548 xfs_agiblock_init(
549 	struct xfs_mount	*mp,
550 	struct xfs_buf		*bp,
551 	struct aghdr_init_data	*id)
552 {
553 	struct xfs_agi		*agi = bp->b_addr;
554 	int			bucket;
555 
556 	agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
557 	agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
558 	agi->agi_seqno = cpu_to_be32(id->agno);
559 	agi->agi_length = cpu_to_be32(id->agsize);
560 	agi->agi_count = 0;
561 	agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
562 	agi->agi_level = cpu_to_be32(1);
563 	agi->agi_freecount = 0;
564 	agi->agi_newino = cpu_to_be32(NULLAGINO);
565 	agi->agi_dirino = cpu_to_be32(NULLAGINO);
566 	if (xfs_has_crc(mp))
567 		uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
568 	if (xfs_has_finobt(mp)) {
569 		agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
570 		agi->agi_free_level = cpu_to_be32(1);
571 	}
572 	for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
573 		agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
574 	if (xfs_has_inobtcounts(mp)) {
575 		agi->agi_iblocks = cpu_to_be32(1);
576 		if (xfs_has_finobt(mp))
577 			agi->agi_fblocks = cpu_to_be32(1);
578 	}
579 }
580 
581 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
582 				  struct aghdr_init_data *id);
583 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)584 xfs_ag_init_hdr(
585 	struct xfs_mount	*mp,
586 	struct aghdr_init_data	*id,
587 	aghdr_init_work_f	work,
588 	const struct xfs_buf_ops *ops)
589 {
590 	struct xfs_buf		*bp;
591 	int			error;
592 
593 	error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
594 	if (error)
595 		return error;
596 
597 	(*work)(mp, bp, id);
598 
599 	xfs_buf_delwri_queue(bp, &id->buffer_list);
600 	xfs_buf_relse(bp);
601 	return 0;
602 }
603 
604 struct xfs_aghdr_grow_data {
605 	xfs_daddr_t		daddr;
606 	size_t			numblks;
607 	const struct xfs_buf_ops *ops;
608 	aghdr_init_work_f	work;
609 	const struct xfs_btree_ops *bc_ops;
610 	bool			need_init;
611 };
612 
613 /*
614  * Prepare new AG headers to be written to disk. We use uncached buffers here,
615  * as it is assumed these new AG headers are currently beyond the currently
616  * valid filesystem address space. Using cached buffers would trip over EOFS
617  * corruption detection alogrithms in the buffer cache lookup routines.
618  *
619  * This is a non-transactional function, but the prepared buffers are added to a
620  * delayed write buffer list supplied by the caller so they can submit them to
621  * disk and wait on them as required.
622  */
623 int
xfs_ag_init_headers(struct xfs_mount * mp,struct aghdr_init_data * id)624 xfs_ag_init_headers(
625 	struct xfs_mount	*mp,
626 	struct aghdr_init_data	*id)
627 
628 {
629 	struct xfs_aghdr_grow_data aghdr_data[] = {
630 	{ /* SB */
631 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
632 		.numblks = XFS_FSS_TO_BB(mp, 1),
633 		.ops = &xfs_sb_buf_ops,
634 		.work = &xfs_sbblock_init,
635 		.need_init = true
636 	},
637 	{ /* AGF */
638 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
639 		.numblks = XFS_FSS_TO_BB(mp, 1),
640 		.ops = &xfs_agf_buf_ops,
641 		.work = &xfs_agfblock_init,
642 		.need_init = true
643 	},
644 	{ /* AGFL */
645 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
646 		.numblks = XFS_FSS_TO_BB(mp, 1),
647 		.ops = &xfs_agfl_buf_ops,
648 		.work = &xfs_agflblock_init,
649 		.need_init = true
650 	},
651 	{ /* AGI */
652 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
653 		.numblks = XFS_FSS_TO_BB(mp, 1),
654 		.ops = &xfs_agi_buf_ops,
655 		.work = &xfs_agiblock_init,
656 		.need_init = true
657 	},
658 	{ /* BNO root block */
659 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
660 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
661 		.ops = &xfs_bnobt_buf_ops,
662 		.work = &xfs_bnoroot_init,
663 		.bc_ops = &xfs_bnobt_ops,
664 		.need_init = true
665 	},
666 	{ /* CNT root block */
667 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
668 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
669 		.ops = &xfs_cntbt_buf_ops,
670 		.work = &xfs_bnoroot_init,
671 		.bc_ops = &xfs_cntbt_ops,
672 		.need_init = true
673 	},
674 	{ /* INO root block */
675 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
676 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
677 		.ops = &xfs_inobt_buf_ops,
678 		.work = &xfs_btroot_init,
679 		.bc_ops = &xfs_inobt_ops,
680 		.need_init = true
681 	},
682 	{ /* FINO root block */
683 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
684 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
685 		.ops = &xfs_finobt_buf_ops,
686 		.work = &xfs_btroot_init,
687 		.bc_ops = &xfs_finobt_ops,
688 		.need_init =  xfs_has_finobt(mp)
689 	},
690 	{ /* RMAP root block */
691 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
692 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
693 		.ops = &xfs_rmapbt_buf_ops,
694 		.work = &xfs_rmaproot_init,
695 		.bc_ops = &xfs_rmapbt_ops,
696 		.need_init = xfs_has_rmapbt(mp)
697 	},
698 	{ /* REFC root block */
699 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
700 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
701 		.ops = &xfs_refcountbt_buf_ops,
702 		.work = &xfs_btroot_init,
703 		.bc_ops = &xfs_refcountbt_ops,
704 		.need_init = xfs_has_reflink(mp)
705 	},
706 	{ /* NULL terminating block */
707 		.daddr = XFS_BUF_DADDR_NULL,
708 	}
709 	};
710 	struct  xfs_aghdr_grow_data *dp;
711 	int			error = 0;
712 
713 	/* Account for AG free space in new AG */
714 	id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
715 	for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
716 		if (!dp->need_init)
717 			continue;
718 
719 		id->daddr = dp->daddr;
720 		id->numblks = dp->numblks;
721 		id->bc_ops = dp->bc_ops;
722 		error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
723 		if (error)
724 			break;
725 	}
726 	return error;
727 }
728 
729 int
xfs_ag_shrink_space(struct xfs_perag * pag,struct xfs_trans ** tpp,xfs_extlen_t delta)730 xfs_ag_shrink_space(
731 	struct xfs_perag	*pag,
732 	struct xfs_trans	**tpp,
733 	xfs_extlen_t		delta)
734 {
735 	struct xfs_mount	*mp = pag_mount(pag);
736 	struct xfs_alloc_arg	args = {
737 		.tp	= *tpp,
738 		.mp	= mp,
739 		.pag	= pag,
740 		.minlen = delta,
741 		.maxlen = delta,
742 		.oinfo	= XFS_RMAP_OINFO_SKIP_UPDATE,
743 		.resv	= XFS_AG_RESV_NONE,
744 		.prod	= 1
745 	};
746 	struct xfs_buf		*agibp, *agfbp;
747 	struct xfs_agi		*agi;
748 	struct xfs_agf		*agf;
749 	xfs_agblock_t		aglen;
750 	int			error, err2;
751 
752 	ASSERT(pag_agno(pag) == mp->m_sb.sb_agcount - 1);
753 	error = xfs_ialloc_read_agi(pag, *tpp, 0, &agibp);
754 	if (error)
755 		return error;
756 
757 	agi = agibp->b_addr;
758 
759 	error = xfs_alloc_read_agf(pag, *tpp, 0, &agfbp);
760 	if (error)
761 		return error;
762 
763 	agf = agfbp->b_addr;
764 	aglen = be32_to_cpu(agi->agi_length);
765 	/* some extra paranoid checks before we shrink the ag */
766 	if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length)) {
767 		xfs_ag_mark_sick(pag, XFS_SICK_AG_AGF);
768 		return -EFSCORRUPTED;
769 	}
770 	if (delta >= aglen)
771 		return -EINVAL;
772 
773 	/*
774 	 * Make sure that the last inode cluster cannot overlap with the new
775 	 * end of the AG, even if it's sparse.
776 	 */
777 	error = xfs_ialloc_check_shrink(pag, *tpp, agibp, aglen - delta);
778 	if (error)
779 		return error;
780 
781 	/*
782 	 * Disable perag reservations so it doesn't cause the allocation request
783 	 * to fail. We'll reestablish reservation before we return.
784 	 */
785 	xfs_ag_resv_free(pag);
786 
787 	/* internal log shouldn't also show up in the free space btrees */
788 	error = xfs_alloc_vextent_exact_bno(&args,
789 			xfs_agbno_to_fsb(pag, aglen - delta));
790 	if (!error && args.agbno == NULLAGBLOCK)
791 		error = -ENOSPC;
792 
793 	if (error) {
794 		/*
795 		 * If extent allocation fails, need to roll the transaction to
796 		 * ensure that the AGFL fixup has been committed anyway.
797 		 *
798 		 * We need to hold the AGF across the roll to ensure nothing can
799 		 * access the AG for allocation until the shrink is fully
800 		 * cleaned up. And due to the resetting of the AG block
801 		 * reservation space needing to lock the AGI, we also have to
802 		 * hold that so we don't get AGI/AGF lock order inversions in
803 		 * the error handling path.
804 		 */
805 		xfs_trans_bhold(*tpp, agfbp);
806 		xfs_trans_bhold(*tpp, agibp);
807 		err2 = xfs_trans_roll(tpp);
808 		if (err2)
809 			return err2;
810 		xfs_trans_bjoin(*tpp, agfbp);
811 		xfs_trans_bjoin(*tpp, agibp);
812 		goto resv_init_out;
813 	}
814 
815 	/*
816 	 * if successfully deleted from freespace btrees, need to confirm
817 	 * per-AG reservation works as expected.
818 	 */
819 	be32_add_cpu(&agi->agi_length, -delta);
820 	be32_add_cpu(&agf->agf_length, -delta);
821 
822 	err2 = xfs_ag_resv_init(pag, *tpp);
823 	if (err2) {
824 		be32_add_cpu(&agi->agi_length, delta);
825 		be32_add_cpu(&agf->agf_length, delta);
826 		if (err2 != -ENOSPC)
827 			goto resv_err;
828 
829 		err2 = xfs_free_extent_later(*tpp, args.fsbno, delta, NULL,
830 				XFS_AG_RESV_NONE, XFS_FREE_EXTENT_SKIP_DISCARD);
831 		if (err2)
832 			goto resv_err;
833 
834 		/*
835 		 * Roll the transaction before trying to re-init the per-ag
836 		 * reservation. The new transaction is clean so it will cancel
837 		 * without any side effects.
838 		 */
839 		error = xfs_defer_finish(tpp);
840 		if (error)
841 			return error;
842 
843 		error = -ENOSPC;
844 		goto resv_init_out;
845 	}
846 
847 	/* Update perag geometry */
848 	pag_group(pag)->xg_block_count -= delta;
849 	__xfs_agino_range(mp, pag_group(pag)->xg_block_count, &pag->agino_min,
850 			&pag->agino_max);
851 
852 	xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
853 	xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
854 	return 0;
855 
856 resv_init_out:
857 	err2 = xfs_ag_resv_init(pag, *tpp);
858 	if (!err2)
859 		return error;
860 resv_err:
861 	xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
862 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
863 	return err2;
864 }
865 
866 void
xfs_growfs_compute_deltas(struct xfs_mount * mp,xfs_rfsblock_t nb,int64_t * deltap,xfs_agnumber_t * nagcountp)867 xfs_growfs_compute_deltas(
868 	struct xfs_mount	*mp,
869 	xfs_rfsblock_t		nb,
870 	int64_t			*deltap,
871 	xfs_agnumber_t		*nagcountp)
872 {
873 	xfs_rfsblock_t	nb_div, nb_mod;
874 	int64_t		delta;
875 	xfs_agnumber_t	nagcount;
876 
877 	nb_div = nb;
878 	nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks);
879 	if (nb_mod && nb_mod >= XFS_MIN_AG_BLOCKS)
880 		nb_div++;
881 	else if (nb_mod)
882 		nb = nb_div * mp->m_sb.sb_agblocks;
883 
884 	if (nb_div > XFS_MAX_AGNUMBER + 1) {
885 		nb_div = XFS_MAX_AGNUMBER + 1;
886 		nb = nb_div * mp->m_sb.sb_agblocks;
887 	}
888 	nagcount = nb_div;
889 	delta = nb - mp->m_sb.sb_dblocks;
890 	*deltap = delta;
891 	*nagcountp = nagcount;
892 }
893 
894 /*
895  * Extent the AG indicated by the @id by the length passed in
896  */
897 int
xfs_ag_extend_space(struct xfs_perag * pag,struct xfs_trans * tp,xfs_extlen_t len)898 xfs_ag_extend_space(
899 	struct xfs_perag	*pag,
900 	struct xfs_trans	*tp,
901 	xfs_extlen_t		len)
902 {
903 	struct xfs_mount	*mp = pag_mount(pag);
904 	struct xfs_buf		*bp;
905 	struct xfs_agi		*agi;
906 	struct xfs_agf		*agf;
907 	int			error;
908 
909 	ASSERT(pag_agno(pag) == mp->m_sb.sb_agcount - 1);
910 
911 	error = xfs_ialloc_read_agi(pag, tp, 0, &bp);
912 	if (error)
913 		return error;
914 
915 	agi = bp->b_addr;
916 	be32_add_cpu(&agi->agi_length, len);
917 	xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
918 
919 	/*
920 	 * Change agf length.
921 	 */
922 	error = xfs_alloc_read_agf(pag, tp, 0, &bp);
923 	if (error)
924 		return error;
925 
926 	agf = bp->b_addr;
927 	be32_add_cpu(&agf->agf_length, len);
928 	ASSERT(agf->agf_length == agi->agi_length);
929 	xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
930 
931 	/*
932 	 * Free the new space.
933 	 *
934 	 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
935 	 * this doesn't actually exist in the rmap btree.
936 	 */
937 	error = xfs_rmap_free(tp, bp, pag, be32_to_cpu(agf->agf_length) - len,
938 				len, &XFS_RMAP_OINFO_SKIP_UPDATE);
939 	if (error)
940 		return error;
941 
942 	error = xfs_free_extent(tp, pag, be32_to_cpu(agf->agf_length) - len,
943 			len, &XFS_RMAP_OINFO_SKIP_UPDATE, XFS_AG_RESV_NONE);
944 	if (error)
945 		return error;
946 
947 	/* Update perag geometry */
948 	pag_group(pag)->xg_block_count = be32_to_cpu(agf->agf_length);
949 	__xfs_agino_range(mp, pag_group(pag)->xg_block_count, &pag->agino_min,
950 			&pag->agino_max);
951 	return 0;
952 }
953 
954 /* Retrieve AG geometry. */
955 int
xfs_ag_get_geometry(struct xfs_perag * pag,struct xfs_ag_geometry * ageo)956 xfs_ag_get_geometry(
957 	struct xfs_perag	*pag,
958 	struct xfs_ag_geometry	*ageo)
959 {
960 	struct xfs_buf		*agi_bp;
961 	struct xfs_buf		*agf_bp;
962 	struct xfs_agi		*agi;
963 	struct xfs_agf		*agf;
964 	unsigned int		freeblks;
965 	int			error;
966 
967 	/* Lock the AG headers. */
968 	error = xfs_ialloc_read_agi(pag, NULL, 0, &agi_bp);
969 	if (error)
970 		return error;
971 	error = xfs_alloc_read_agf(pag, NULL, 0, &agf_bp);
972 	if (error)
973 		goto out_agi;
974 
975 	/* Fill out form. */
976 	memset(ageo, 0, sizeof(*ageo));
977 	ageo->ag_number = pag_agno(pag);
978 
979 	agi = agi_bp->b_addr;
980 	ageo->ag_icount = be32_to_cpu(agi->agi_count);
981 	ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
982 
983 	agf = agf_bp->b_addr;
984 	ageo->ag_length = be32_to_cpu(agf->agf_length);
985 	freeblks = pag->pagf_freeblks +
986 		   pag->pagf_flcount +
987 		   pag->pagf_btreeblks -
988 		   xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
989 	ageo->ag_freeblks = freeblks;
990 	xfs_ag_geom_health(pag, ageo);
991 
992 	/* Release resources. */
993 	xfs_buf_relse(agf_bp);
994 out_agi:
995 	xfs_buf_relse(agi_bp);
996 	return error;
997 }
998