xref: /linux/fs/xfs/libxfs/xfs_ag.c (revision 18f90d372cf35b387663f1567de701e5393f6eb5)
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 
23 static struct xfs_buf *
24 xfs_get_aghdr_buf(
25 	struct xfs_mount	*mp,
26 	xfs_daddr_t		blkno,
27 	size_t			numblks,
28 	int			flags,
29 	const struct xfs_buf_ops *ops)
30 {
31 	struct xfs_buf		*bp;
32 
33 	bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
34 	if (!bp)
35 		return NULL;
36 
37 	xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
38 	bp->b_bn = blkno;
39 	bp->b_maps[0].bm_bn = blkno;
40 	bp->b_ops = ops;
41 
42 	return bp;
43 }
44 
45 /*
46  * Generic btree root block init function
47  */
48 static void
49 xfs_btroot_init(
50 	struct xfs_mount	*mp,
51 	struct xfs_buf		*bp,
52 	struct aghdr_init_data	*id)
53 {
54 	xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno, 0);
55 }
56 
57 /*
58  * Alloc btree root block init functions
59  */
60 static void
61 xfs_bnoroot_init(
62 	struct xfs_mount	*mp,
63 	struct xfs_buf		*bp,
64 	struct aghdr_init_data	*id)
65 {
66 	struct xfs_alloc_rec	*arec;
67 
68 	xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno, 0);
69 	arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
70 	arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
71 	arec->ar_blockcount = cpu_to_be32(id->agsize -
72 					  be32_to_cpu(arec->ar_startblock));
73 }
74 
75 static void
76 xfs_cntroot_init(
77 	struct xfs_mount	*mp,
78 	struct xfs_buf		*bp,
79 	struct aghdr_init_data	*id)
80 {
81 	struct xfs_alloc_rec	*arec;
82 
83 	xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno, 0);
84 	arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
85 	arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
86 	arec->ar_blockcount = cpu_to_be32(id->agsize -
87 					  be32_to_cpu(arec->ar_startblock));
88 }
89 
90 /*
91  * Reverse map root block init
92  */
93 static void
94 xfs_rmaproot_init(
95 	struct xfs_mount	*mp,
96 	struct xfs_buf		*bp,
97 	struct aghdr_init_data	*id)
98 {
99 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
100 	struct xfs_rmap_rec	*rrec;
101 
102 	xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno, 0);
103 
104 	/*
105 	 * mark the AG header regions as static metadata The BNO
106 	 * btree block is the first block after the headers, so
107 	 * it's location defines the size of region the static
108 	 * metadata consumes.
109 	 *
110 	 * Note: unlike mkfs, we never have to account for log
111 	 * space when growing the data regions
112 	 */
113 	rrec = XFS_RMAP_REC_ADDR(block, 1);
114 	rrec->rm_startblock = 0;
115 	rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
116 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
117 	rrec->rm_offset = 0;
118 
119 	/* account freespace btree root blocks */
120 	rrec = XFS_RMAP_REC_ADDR(block, 2);
121 	rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
122 	rrec->rm_blockcount = cpu_to_be32(2);
123 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
124 	rrec->rm_offset = 0;
125 
126 	/* account inode btree root blocks */
127 	rrec = XFS_RMAP_REC_ADDR(block, 3);
128 	rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
129 	rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
130 					  XFS_IBT_BLOCK(mp));
131 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
132 	rrec->rm_offset = 0;
133 
134 	/* account for rmap btree root */
135 	rrec = XFS_RMAP_REC_ADDR(block, 4);
136 	rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
137 	rrec->rm_blockcount = cpu_to_be32(1);
138 	rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
139 	rrec->rm_offset = 0;
140 
141 	/* account for refc btree root */
142 	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
143 		rrec = XFS_RMAP_REC_ADDR(block, 5);
144 		rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
145 		rrec->rm_blockcount = cpu_to_be32(1);
146 		rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
147 		rrec->rm_offset = 0;
148 		be16_add_cpu(&block->bb_numrecs, 1);
149 	}
150 }
151 
152 /*
153  * Initialise new secondary superblocks with the pre-grow geometry, but mark
154  * them as "in progress" so we know they haven't yet been activated. This will
155  * get cleared when the update with the new geometry information is done after
156  * changes to the primary are committed. This isn't strictly necessary, but we
157  * get it for free with the delayed buffer write lists and it means we can tell
158  * if a grow operation didn't complete properly after the fact.
159  */
160 static void
161 xfs_sbblock_init(
162 	struct xfs_mount	*mp,
163 	struct xfs_buf		*bp,
164 	struct aghdr_init_data	*id)
165 {
166 	struct xfs_dsb		*dsb = XFS_BUF_TO_SBP(bp);
167 
168 	xfs_sb_to_disk(dsb, &mp->m_sb);
169 	dsb->sb_inprogress = 1;
170 }
171 
172 static void
173 xfs_agfblock_init(
174 	struct xfs_mount	*mp,
175 	struct xfs_buf		*bp,
176 	struct aghdr_init_data	*id)
177 {
178 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(bp);
179 	xfs_extlen_t		tmpsize;
180 
181 	agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
182 	agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
183 	agf->agf_seqno = cpu_to_be32(id->agno);
184 	agf->agf_length = cpu_to_be32(id->agsize);
185 	agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
186 	agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
187 	agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
188 	agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
189 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
190 		agf->agf_roots[XFS_BTNUM_RMAPi] =
191 					cpu_to_be32(XFS_RMAP_BLOCK(mp));
192 		agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
193 		agf->agf_rmap_blocks = cpu_to_be32(1);
194 	}
195 
196 	agf->agf_flfirst = cpu_to_be32(1);
197 	agf->agf_fllast = 0;
198 	agf->agf_flcount = 0;
199 	tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
200 	agf->agf_freeblks = cpu_to_be32(tmpsize);
201 	agf->agf_longest = cpu_to_be32(tmpsize);
202 	if (xfs_sb_version_hascrc(&mp->m_sb))
203 		uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
204 	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
205 		agf->agf_refcount_root = cpu_to_be32(
206 				xfs_refc_block(mp));
207 		agf->agf_refcount_level = cpu_to_be32(1);
208 		agf->agf_refcount_blocks = cpu_to_be32(1);
209 	}
210 }
211 
212 static void
213 xfs_agflblock_init(
214 	struct xfs_mount	*mp,
215 	struct xfs_buf		*bp,
216 	struct aghdr_init_data	*id)
217 {
218 	struct xfs_agfl		*agfl = XFS_BUF_TO_AGFL(bp);
219 	__be32			*agfl_bno;
220 	int			bucket;
221 
222 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
223 		agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
224 		agfl->agfl_seqno = cpu_to_be32(id->agno);
225 		uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
226 	}
227 
228 	agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
229 	for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
230 		agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
231 }
232 
233 static void
234 xfs_agiblock_init(
235 	struct xfs_mount	*mp,
236 	struct xfs_buf		*bp,
237 	struct aghdr_init_data	*id)
238 {
239 	struct xfs_agi		*agi = XFS_BUF_TO_AGI(bp);
240 	int			bucket;
241 
242 	agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
243 	agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
244 	agi->agi_seqno = cpu_to_be32(id->agno);
245 	agi->agi_length = cpu_to_be32(id->agsize);
246 	agi->agi_count = 0;
247 	agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
248 	agi->agi_level = cpu_to_be32(1);
249 	agi->agi_freecount = 0;
250 	agi->agi_newino = cpu_to_be32(NULLAGINO);
251 	agi->agi_dirino = cpu_to_be32(NULLAGINO);
252 	if (xfs_sb_version_hascrc(&mp->m_sb))
253 		uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
254 	if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
255 		agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
256 		agi->agi_free_level = cpu_to_be32(1);
257 	}
258 	for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
259 		agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
260 }
261 
262 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
263 				  struct aghdr_init_data *id);
264 static int
265 xfs_ag_init_hdr(
266 	struct xfs_mount	*mp,
267 	struct aghdr_init_data	*id,
268 	aghdr_init_work_f	work,
269 	const struct xfs_buf_ops *ops)
270 
271 {
272 	struct xfs_buf		*bp;
273 
274 	bp = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, 0, ops);
275 	if (!bp)
276 		return -ENOMEM;
277 
278 	(*work)(mp, bp, id);
279 
280 	xfs_buf_delwri_queue(bp, &id->buffer_list);
281 	xfs_buf_relse(bp);
282 	return 0;
283 }
284 
285 struct xfs_aghdr_grow_data {
286 	xfs_daddr_t		daddr;
287 	size_t			numblks;
288 	const struct xfs_buf_ops *ops;
289 	aghdr_init_work_f	work;
290 	xfs_btnum_t		type;
291 	bool			need_init;
292 };
293 
294 /*
295  * Prepare new AG headers to be written to disk. We use uncached buffers here,
296  * as it is assumed these new AG headers are currently beyond the currently
297  * valid filesystem address space. Using cached buffers would trip over EOFS
298  * corruption detection alogrithms in the buffer cache lookup routines.
299  *
300  * This is a non-transactional function, but the prepared buffers are added to a
301  * delayed write buffer list supplied by the caller so they can submit them to
302  * disk and wait on them as required.
303  */
304 int
305 xfs_ag_init_headers(
306 	struct xfs_mount	*mp,
307 	struct aghdr_init_data	*id)
308 
309 {
310 	struct xfs_aghdr_grow_data aghdr_data[] = {
311 	{ /* SB */
312 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
313 		.numblks = XFS_FSS_TO_BB(mp, 1),
314 		.ops = &xfs_sb_buf_ops,
315 		.work = &xfs_sbblock_init,
316 		.need_init = true
317 	},
318 	{ /* AGF */
319 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
320 		.numblks = XFS_FSS_TO_BB(mp, 1),
321 		.ops = &xfs_agf_buf_ops,
322 		.work = &xfs_agfblock_init,
323 		.need_init = true
324 	},
325 	{ /* AGFL */
326 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
327 		.numblks = XFS_FSS_TO_BB(mp, 1),
328 		.ops = &xfs_agfl_buf_ops,
329 		.work = &xfs_agflblock_init,
330 		.need_init = true
331 	},
332 	{ /* AGI */
333 		.daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
334 		.numblks = XFS_FSS_TO_BB(mp, 1),
335 		.ops = &xfs_agi_buf_ops,
336 		.work = &xfs_agiblock_init,
337 		.need_init = true
338 	},
339 	{ /* BNO root block */
340 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
341 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
342 		.ops = &xfs_bnobt_buf_ops,
343 		.work = &xfs_bnoroot_init,
344 		.need_init = true
345 	},
346 	{ /* CNT root block */
347 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
348 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
349 		.ops = &xfs_cntbt_buf_ops,
350 		.work = &xfs_cntroot_init,
351 		.need_init = true
352 	},
353 	{ /* INO root block */
354 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
355 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
356 		.ops = &xfs_inobt_buf_ops,
357 		.work = &xfs_btroot_init,
358 		.type = XFS_BTNUM_INO,
359 		.need_init = true
360 	},
361 	{ /* FINO root block */
362 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
363 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
364 		.ops = &xfs_finobt_buf_ops,
365 		.work = &xfs_btroot_init,
366 		.type = XFS_BTNUM_FINO,
367 		.need_init =  xfs_sb_version_hasfinobt(&mp->m_sb)
368 	},
369 	{ /* RMAP root block */
370 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
371 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
372 		.ops = &xfs_rmapbt_buf_ops,
373 		.work = &xfs_rmaproot_init,
374 		.need_init = xfs_sb_version_hasrmapbt(&mp->m_sb)
375 	},
376 	{ /* REFC root block */
377 		.daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
378 		.numblks = BTOBB(mp->m_sb.sb_blocksize),
379 		.ops = &xfs_refcountbt_buf_ops,
380 		.work = &xfs_btroot_init,
381 		.type = XFS_BTNUM_REFC,
382 		.need_init = xfs_sb_version_hasreflink(&mp->m_sb)
383 	},
384 	{ /* NULL terminating block */
385 		.daddr = XFS_BUF_DADDR_NULL,
386 	}
387 	};
388 	struct  xfs_aghdr_grow_data *dp;
389 	int			error = 0;
390 
391 	/* Account for AG free space in new AG */
392 	id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
393 	for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
394 		if (!dp->need_init)
395 			continue;
396 
397 		id->daddr = dp->daddr;
398 		id->numblks = dp->numblks;
399 		id->type = dp->type;
400 		error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
401 		if (error)
402 			break;
403 	}
404 	return error;
405 }
406 
407 /*
408  * Extent the AG indicated by the @id by the length passed in
409  */
410 int
411 xfs_ag_extend_space(
412 	struct xfs_mount	*mp,
413 	struct xfs_trans	*tp,
414 	struct aghdr_init_data	*id,
415 	xfs_extlen_t		len)
416 {
417 	struct xfs_buf		*bp;
418 	struct xfs_agi		*agi;
419 	struct xfs_agf		*agf;
420 	int			error;
421 
422 	/*
423 	 * Change the agi length.
424 	 */
425 	error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
426 	if (error)
427 		return error;
428 
429 	agi = XFS_BUF_TO_AGI(bp);
430 	be32_add_cpu(&agi->agi_length, len);
431 	ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
432 	       be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
433 	xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
434 
435 	/*
436 	 * Change agf length.
437 	 */
438 	error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
439 	if (error)
440 		return error;
441 
442 	agf = XFS_BUF_TO_AGF(bp);
443 	be32_add_cpu(&agf->agf_length, len);
444 	ASSERT(agf->agf_length == agi->agi_length);
445 	xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
446 
447 	/*
448 	 * Free the new space.
449 	 *
450 	 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
451 	 * this doesn't actually exist in the rmap btree.
452 	 */
453 	error = xfs_rmap_free(tp, bp, id->agno,
454 				be32_to_cpu(agf->agf_length) - len,
455 				len, &XFS_RMAP_OINFO_SKIP_UPDATE);
456 	if (error)
457 		return error;
458 
459 	return  xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
460 					be32_to_cpu(agf->agf_length) - len),
461 				len, &XFS_RMAP_OINFO_SKIP_UPDATE,
462 				XFS_AG_RESV_NONE);
463 }
464