xref: /linux/fs/xfs/libxfs/xfs_ag.c (revision 7ff836f064e2c814a7504c91a4464eea45d475bd)
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_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_btree.h"
16 #include "xfs_alloc_btree.h"
17 #include "xfs_rmap_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_ialloc.h"
20 #include "xfs_rmap.h"
21 #include "xfs_ag.h"
22 #include "xfs_ag_resv.h"
23 #include "xfs_health.h"
24 
25 static struct xfs_buf *
26 xfs_get_aghdr_buf(
27 	struct xfs_mount	*mp,
28 	xfs_daddr_t		blkno,
29 	size_t			numblks,
30 	int			flags,
31 	const struct xfs_buf_ops *ops)
32 {
33 	struct xfs_buf		*bp;
34 
35 	bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
36 	if (!bp)
37 		return NULL;
38 
39 	xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
40 	bp->b_bn = blkno;
41 	bp->b_maps[0].bm_bn = blkno;
42 	bp->b_ops = ops;
43 
44 	return bp;
45 }
46 
47 /*
48  * Generic btree root block init function
49  */
50 static void
51 xfs_btroot_init(
52 	struct xfs_mount	*mp,
53 	struct xfs_buf		*bp,
54 	struct aghdr_init_data	*id)
55 {
56 	xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno, 0);
57 }
58 
59 /*
60  * Alloc btree root block init functions
61  */
62 static void
63 xfs_bnoroot_init(
64 	struct xfs_mount	*mp,
65 	struct xfs_buf		*bp,
66 	struct aghdr_init_data	*id)
67 {
68 	struct xfs_alloc_rec	*arec;
69 
70 	xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno, 0);
71 	arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
72 	arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
73 	arec->ar_blockcount = cpu_to_be32(id->agsize -
74 					  be32_to_cpu(arec->ar_startblock));
75 }
76 
77 static void
78 xfs_cntroot_init(
79 	struct xfs_mount	*mp,
80 	struct xfs_buf		*bp,
81 	struct aghdr_init_data	*id)
82 {
83 	struct xfs_alloc_rec	*arec;
84 
85 	xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno, 0);
86 	arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
87 	arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
88 	arec->ar_blockcount = cpu_to_be32(id->agsize -
89 					  be32_to_cpu(arec->ar_startblock));
90 }
91 
92 /*
93  * Reverse map root block init
94  */
95 static void
96 xfs_rmaproot_init(
97 	struct xfs_mount	*mp,
98 	struct xfs_buf		*bp,
99 	struct aghdr_init_data	*id)
100 {
101 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
102 	struct xfs_rmap_rec	*rrec;
103 
104 	xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno, 0);
105 
106 	/*
107 	 * mark the AG header regions as static metadata The BNO
108 	 * btree block is the first block after the headers, so
109 	 * it's location defines the size of region the static
110 	 * metadata consumes.
111 	 *
112 	 * Note: unlike mkfs, we never have to account for log
113 	 * space when growing the data regions
114 	 */
115 	rrec = XFS_RMAP_REC_ADDR(block, 1);
116 	rrec->rm_startblock = 0;
117 	rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
118 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
119 	rrec->rm_offset = 0;
120 
121 	/* account freespace btree root blocks */
122 	rrec = XFS_RMAP_REC_ADDR(block, 2);
123 	rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
124 	rrec->rm_blockcount = cpu_to_be32(2);
125 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
126 	rrec->rm_offset = 0;
127 
128 	/* account inode btree root blocks */
129 	rrec = XFS_RMAP_REC_ADDR(block, 3);
130 	rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
131 	rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
132 					  XFS_IBT_BLOCK(mp));
133 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
134 	rrec->rm_offset = 0;
135 
136 	/* account for rmap btree root */
137 	rrec = XFS_RMAP_REC_ADDR(block, 4);
138 	rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
139 	rrec->rm_blockcount = cpu_to_be32(1);
140 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
141 	rrec->rm_offset = 0;
142 
143 	/* account for refc btree root */
144 	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
145 		rrec = XFS_RMAP_REC_ADDR(block, 5);
146 		rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
147 		rrec->rm_blockcount = cpu_to_be32(1);
148 		rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
149 		rrec->rm_offset = 0;
150 		be16_add_cpu(&block->bb_numrecs, 1);
151 	}
152 }
153 
154 /*
155  * Initialise new secondary superblocks with the pre-grow geometry, but mark
156  * them as "in progress" so we know they haven't yet been activated. This will
157  * get cleared when the update with the new geometry information is done after
158  * changes to the primary are committed. This isn't strictly necessary, but we
159  * get it for free with the delayed buffer write lists and it means we can tell
160  * if a grow operation didn't complete properly after the fact.
161  */
162 static void
163 xfs_sbblock_init(
164 	struct xfs_mount	*mp,
165 	struct xfs_buf		*bp,
166 	struct aghdr_init_data	*id)
167 {
168 	struct xfs_dsb		*dsb = XFS_BUF_TO_SBP(bp);
169 
170 	xfs_sb_to_disk(dsb, &mp->m_sb);
171 	dsb->sb_inprogress = 1;
172 }
173 
174 static void
175 xfs_agfblock_init(
176 	struct xfs_mount	*mp,
177 	struct xfs_buf		*bp,
178 	struct aghdr_init_data	*id)
179 {
180 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(bp);
181 	xfs_extlen_t		tmpsize;
182 
183 	agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
184 	agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
185 	agf->agf_seqno = cpu_to_be32(id->agno);
186 	agf->agf_length = cpu_to_be32(id->agsize);
187 	agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
188 	agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
189 	agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
190 	agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
191 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
192 		agf->agf_roots[XFS_BTNUM_RMAPi] =
193 					cpu_to_be32(XFS_RMAP_BLOCK(mp));
194 		agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
195 		agf->agf_rmap_blocks = cpu_to_be32(1);
196 	}
197 
198 	agf->agf_flfirst = cpu_to_be32(1);
199 	agf->agf_fllast = 0;
200 	agf->agf_flcount = 0;
201 	tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
202 	agf->agf_freeblks = cpu_to_be32(tmpsize);
203 	agf->agf_longest = cpu_to_be32(tmpsize);
204 	if (xfs_sb_version_hascrc(&mp->m_sb))
205 		uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
206 	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
207 		agf->agf_refcount_root = cpu_to_be32(
208 				xfs_refc_block(mp));
209 		agf->agf_refcount_level = cpu_to_be32(1);
210 		agf->agf_refcount_blocks = cpu_to_be32(1);
211 	}
212 }
213 
214 static void
215 xfs_agflblock_init(
216 	struct xfs_mount	*mp,
217 	struct xfs_buf		*bp,
218 	struct aghdr_init_data	*id)
219 {
220 	struct xfs_agfl		*agfl = XFS_BUF_TO_AGFL(bp);
221 	__be32			*agfl_bno;
222 	int			bucket;
223 
224 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
225 		agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
226 		agfl->agfl_seqno = cpu_to_be32(id->agno);
227 		uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
228 	}
229 
230 	agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
231 	for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
232 		agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
233 }
234 
235 static void
236 xfs_agiblock_init(
237 	struct xfs_mount	*mp,
238 	struct xfs_buf		*bp,
239 	struct aghdr_init_data	*id)
240 {
241 	struct xfs_agi		*agi = XFS_BUF_TO_AGI(bp);
242 	int			bucket;
243 
244 	agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
245 	agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
246 	agi->agi_seqno = cpu_to_be32(id->agno);
247 	agi->agi_length = cpu_to_be32(id->agsize);
248 	agi->agi_count = 0;
249 	agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
250 	agi->agi_level = cpu_to_be32(1);
251 	agi->agi_freecount = 0;
252 	agi->agi_newino = cpu_to_be32(NULLAGINO);
253 	agi->agi_dirino = cpu_to_be32(NULLAGINO);
254 	if (xfs_sb_version_hascrc(&mp->m_sb))
255 		uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
256 	if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
257 		agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
258 		agi->agi_free_level = cpu_to_be32(1);
259 	}
260 	for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
261 		agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
262 }
263 
264 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
265 				  struct aghdr_init_data *id);
266 static int
267 xfs_ag_init_hdr(
268 	struct xfs_mount	*mp,
269 	struct aghdr_init_data	*id,
270 	aghdr_init_work_f	work,
271 	const struct xfs_buf_ops *ops)
272 
273 {
274 	struct xfs_buf		*bp;
275 
276 	bp = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, 0, ops);
277 	if (!bp)
278 		return -ENOMEM;
279 
280 	(*work)(mp, bp, id);
281 
282 	xfs_buf_delwri_queue(bp, &id->buffer_list);
283 	xfs_buf_relse(bp);
284 	return 0;
285 }
286 
287 struct xfs_aghdr_grow_data {
288 	xfs_daddr_t		daddr;
289 	size_t			numblks;
290 	const struct xfs_buf_ops *ops;
291 	aghdr_init_work_f	work;
292 	xfs_btnum_t		type;
293 	bool			need_init;
294 };
295 
296 /*
297  * Prepare new AG headers to be written to disk. We use uncached buffers here,
298  * as it is assumed these new AG headers are currently beyond the currently
299  * valid filesystem address space. Using cached buffers would trip over EOFS
300  * corruption detection alogrithms in the buffer cache lookup routines.
301  *
302  * This is a non-transactional function, but the prepared buffers are added to a
303  * delayed write buffer list supplied by the caller so they can submit them to
304  * disk and wait on them as required.
305  */
306 int
307 xfs_ag_init_headers(
308 	struct xfs_mount	*mp,
309 	struct aghdr_init_data	*id)
310 
311 {
312 	struct xfs_aghdr_grow_data aghdr_data[] = {
313 	{ /* SB */
314 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
315 		.numblks = XFS_FSS_TO_BB(mp, 1),
316 		.ops = &xfs_sb_buf_ops,
317 		.work = &xfs_sbblock_init,
318 		.need_init = true
319 	},
320 	{ /* AGF */
321 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
322 		.numblks = XFS_FSS_TO_BB(mp, 1),
323 		.ops = &xfs_agf_buf_ops,
324 		.work = &xfs_agfblock_init,
325 		.need_init = true
326 	},
327 	{ /* AGFL */
328 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
329 		.numblks = XFS_FSS_TO_BB(mp, 1),
330 		.ops = &xfs_agfl_buf_ops,
331 		.work = &xfs_agflblock_init,
332 		.need_init = true
333 	},
334 	{ /* AGI */
335 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
336 		.numblks = XFS_FSS_TO_BB(mp, 1),
337 		.ops = &xfs_agi_buf_ops,
338 		.work = &xfs_agiblock_init,
339 		.need_init = true
340 	},
341 	{ /* BNO root block */
342 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
343 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
344 		.ops = &xfs_bnobt_buf_ops,
345 		.work = &xfs_bnoroot_init,
346 		.need_init = true
347 	},
348 	{ /* CNT root block */
349 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
350 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
351 		.ops = &xfs_cntbt_buf_ops,
352 		.work = &xfs_cntroot_init,
353 		.need_init = true
354 	},
355 	{ /* INO root block */
356 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
357 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
358 		.ops = &xfs_inobt_buf_ops,
359 		.work = &xfs_btroot_init,
360 		.type = XFS_BTNUM_INO,
361 		.need_init = true
362 	},
363 	{ /* FINO root block */
364 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
365 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
366 		.ops = &xfs_finobt_buf_ops,
367 		.work = &xfs_btroot_init,
368 		.type = XFS_BTNUM_FINO,
369 		.need_init =  xfs_sb_version_hasfinobt(&mp->m_sb)
370 	},
371 	{ /* RMAP root block */
372 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
373 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
374 		.ops = &xfs_rmapbt_buf_ops,
375 		.work = &xfs_rmaproot_init,
376 		.need_init = xfs_sb_version_hasrmapbt(&mp->m_sb)
377 	},
378 	{ /* REFC root block */
379 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
380 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
381 		.ops = &xfs_refcountbt_buf_ops,
382 		.work = &xfs_btroot_init,
383 		.type = XFS_BTNUM_REFC,
384 		.need_init = xfs_sb_version_hasreflink(&mp->m_sb)
385 	},
386 	{ /* NULL terminating block */
387 		.daddr = XFS_BUF_DADDR_NULL,
388 	}
389 	};
390 	struct  xfs_aghdr_grow_data *dp;
391 	int			error = 0;
392 
393 	/* Account for AG free space in new AG */
394 	id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
395 	for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
396 		if (!dp->need_init)
397 			continue;
398 
399 		id->daddr = dp->daddr;
400 		id->numblks = dp->numblks;
401 		id->type = dp->type;
402 		error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
403 		if (error)
404 			break;
405 	}
406 	return error;
407 }
408 
409 /*
410  * Extent the AG indicated by the @id by the length passed in
411  */
412 int
413 xfs_ag_extend_space(
414 	struct xfs_mount	*mp,
415 	struct xfs_trans	*tp,
416 	struct aghdr_init_data	*id,
417 	xfs_extlen_t		len)
418 {
419 	struct xfs_buf		*bp;
420 	struct xfs_agi		*agi;
421 	struct xfs_agf		*agf;
422 	int			error;
423 
424 	/*
425 	 * Change the agi length.
426 	 */
427 	error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
428 	if (error)
429 		return error;
430 
431 	agi = XFS_BUF_TO_AGI(bp);
432 	be32_add_cpu(&agi->agi_length, len);
433 	ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
434 	       be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
435 	xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
436 
437 	/*
438 	 * Change agf length.
439 	 */
440 	error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
441 	if (error)
442 		return error;
443 
444 	agf = XFS_BUF_TO_AGF(bp);
445 	be32_add_cpu(&agf->agf_length, len);
446 	ASSERT(agf->agf_length == agi->agi_length);
447 	xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
448 
449 	/*
450 	 * Free the new space.
451 	 *
452 	 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
453 	 * this doesn't actually exist in the rmap btree.
454 	 */
455 	error = xfs_rmap_free(tp, bp, id->agno,
456 				be32_to_cpu(agf->agf_length) - len,
457 				len, &XFS_RMAP_OINFO_SKIP_UPDATE);
458 	if (error)
459 		return error;
460 
461 	return  xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
462 					be32_to_cpu(agf->agf_length) - len),
463 				len, &XFS_RMAP_OINFO_SKIP_UPDATE,
464 				XFS_AG_RESV_NONE);
465 }
466 
467 /* Retrieve AG geometry. */
468 int
469 xfs_ag_get_geometry(
470 	struct xfs_mount	*mp,
471 	xfs_agnumber_t		agno,
472 	struct xfs_ag_geometry	*ageo)
473 {
474 	struct xfs_buf		*agi_bp;
475 	struct xfs_buf		*agf_bp;
476 	struct xfs_agi		*agi;
477 	struct xfs_agf		*agf;
478 	struct xfs_perag	*pag;
479 	unsigned int		freeblks;
480 	int			error;
481 
482 	if (agno >= mp->m_sb.sb_agcount)
483 		return -EINVAL;
484 
485 	/* Lock the AG headers. */
486 	error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
487 	if (error)
488 		return error;
489 	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
490 	if (error)
491 		goto out_agi;
492 	pag = xfs_perag_get(mp, agno);
493 
494 	/* Fill out form. */
495 	memset(ageo, 0, sizeof(*ageo));
496 	ageo->ag_number = agno;
497 
498 	agi = XFS_BUF_TO_AGI(agi_bp);
499 	ageo->ag_icount = be32_to_cpu(agi->agi_count);
500 	ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
501 
502 	agf = XFS_BUF_TO_AGF(agf_bp);
503 	ageo->ag_length = be32_to_cpu(agf->agf_length);
504 	freeblks = pag->pagf_freeblks +
505 		   pag->pagf_flcount +
506 		   pag->pagf_btreeblks -
507 		   xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
508 	ageo->ag_freeblks = freeblks;
509 	xfs_ag_geom_health(pag, ageo);
510 
511 	/* Release resources. */
512 	xfs_perag_put(pag);
513 	xfs_buf_relse(agf_bp);
514 out_agi:
515 	xfs_buf_relse(agi_bp);
516 	return error;
517 }
518