xref: /openbmc/linux/fs/xfs/libxfs/xfs_alloc_btree.c (revision e2c75e76)
1 /*
2  * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_mount.h"
26 #include "xfs_btree.h"
27 #include "xfs_alloc_btree.h"
28 #include "xfs_alloc.h"
29 #include "xfs_extent_busy.h"
30 #include "xfs_error.h"
31 #include "xfs_trace.h"
32 #include "xfs_cksum.h"
33 #include "xfs_trans.h"
34 
35 
36 STATIC struct xfs_btree_cur *
37 xfs_allocbt_dup_cursor(
38 	struct xfs_btree_cur	*cur)
39 {
40 	return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
41 			cur->bc_private.a.agbp, cur->bc_private.a.agno,
42 			cur->bc_btnum);
43 }
44 
45 STATIC void
46 xfs_allocbt_set_root(
47 	struct xfs_btree_cur	*cur,
48 	union xfs_btree_ptr	*ptr,
49 	int			inc)
50 {
51 	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
52 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
53 	xfs_agnumber_t		seqno = be32_to_cpu(agf->agf_seqno);
54 	int			btnum = cur->bc_btnum;
55 	struct xfs_perag	*pag = xfs_perag_get(cur->bc_mp, seqno);
56 
57 	ASSERT(ptr->s != 0);
58 
59 	agf->agf_roots[btnum] = ptr->s;
60 	be32_add_cpu(&agf->agf_levels[btnum], inc);
61 	pag->pagf_levels[btnum] += inc;
62 	xfs_perag_put(pag);
63 
64 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
65 }
66 
67 STATIC int
68 xfs_allocbt_alloc_block(
69 	struct xfs_btree_cur	*cur,
70 	union xfs_btree_ptr	*start,
71 	union xfs_btree_ptr	*new,
72 	int			*stat)
73 {
74 	int			error;
75 	xfs_agblock_t		bno;
76 
77 	XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
78 
79 	/* Allocate the new block from the freelist. If we can't, give up.  */
80 	error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp,
81 				       &bno, 1);
82 	if (error) {
83 		XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
84 		return error;
85 	}
86 
87 	if (bno == NULLAGBLOCK) {
88 		XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
89 		*stat = 0;
90 		return 0;
91 	}
92 
93 	xfs_extent_busy_reuse(cur->bc_mp, cur->bc_private.a.agno, bno, 1, false);
94 
95 	xfs_trans_agbtree_delta(cur->bc_tp, 1);
96 	new->s = cpu_to_be32(bno);
97 
98 	XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
99 	*stat = 1;
100 	return 0;
101 }
102 
103 STATIC int
104 xfs_allocbt_free_block(
105 	struct xfs_btree_cur	*cur,
106 	struct xfs_buf		*bp)
107 {
108 	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
109 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
110 	xfs_agblock_t		bno;
111 	int			error;
112 
113 	bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
114 	error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
115 	if (error)
116 		return error;
117 
118 	xfs_extent_busy_insert(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1,
119 			      XFS_EXTENT_BUSY_SKIP_DISCARD);
120 	xfs_trans_agbtree_delta(cur->bc_tp, -1);
121 	return 0;
122 }
123 
124 /*
125  * Update the longest extent in the AGF
126  */
127 STATIC void
128 xfs_allocbt_update_lastrec(
129 	struct xfs_btree_cur	*cur,
130 	struct xfs_btree_block	*block,
131 	union xfs_btree_rec	*rec,
132 	int			ptr,
133 	int			reason)
134 {
135 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
136 	xfs_agnumber_t		seqno = be32_to_cpu(agf->agf_seqno);
137 	struct xfs_perag	*pag;
138 	__be32			len;
139 	int			numrecs;
140 
141 	ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
142 
143 	switch (reason) {
144 	case LASTREC_UPDATE:
145 		/*
146 		 * If this is the last leaf block and it's the last record,
147 		 * then update the size of the longest extent in the AG.
148 		 */
149 		if (ptr != xfs_btree_get_numrecs(block))
150 			return;
151 		len = rec->alloc.ar_blockcount;
152 		break;
153 	case LASTREC_INSREC:
154 		if (be32_to_cpu(rec->alloc.ar_blockcount) <=
155 		    be32_to_cpu(agf->agf_longest))
156 			return;
157 		len = rec->alloc.ar_blockcount;
158 		break;
159 	case LASTREC_DELREC:
160 		numrecs = xfs_btree_get_numrecs(block);
161 		if (ptr <= numrecs)
162 			return;
163 		ASSERT(ptr == numrecs + 1);
164 
165 		if (numrecs) {
166 			xfs_alloc_rec_t *rrp;
167 
168 			rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
169 			len = rrp->ar_blockcount;
170 		} else {
171 			len = 0;
172 		}
173 
174 		break;
175 	default:
176 		ASSERT(0);
177 		return;
178 	}
179 
180 	agf->agf_longest = len;
181 	pag = xfs_perag_get(cur->bc_mp, seqno);
182 	pag->pagf_longest = be32_to_cpu(len);
183 	xfs_perag_put(pag);
184 	xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, XFS_AGF_LONGEST);
185 }
186 
187 STATIC int
188 xfs_allocbt_get_minrecs(
189 	struct xfs_btree_cur	*cur,
190 	int			level)
191 {
192 	return cur->bc_mp->m_alloc_mnr[level != 0];
193 }
194 
195 STATIC int
196 xfs_allocbt_get_maxrecs(
197 	struct xfs_btree_cur	*cur,
198 	int			level)
199 {
200 	return cur->bc_mp->m_alloc_mxr[level != 0];
201 }
202 
203 STATIC void
204 xfs_allocbt_init_key_from_rec(
205 	union xfs_btree_key	*key,
206 	union xfs_btree_rec	*rec)
207 {
208 	key->alloc.ar_startblock = rec->alloc.ar_startblock;
209 	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
210 }
211 
212 STATIC void
213 xfs_bnobt_init_high_key_from_rec(
214 	union xfs_btree_key	*key,
215 	union xfs_btree_rec	*rec)
216 {
217 	__u32			x;
218 
219 	x = be32_to_cpu(rec->alloc.ar_startblock);
220 	x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
221 	key->alloc.ar_startblock = cpu_to_be32(x);
222 	key->alloc.ar_blockcount = 0;
223 }
224 
225 STATIC void
226 xfs_cntbt_init_high_key_from_rec(
227 	union xfs_btree_key	*key,
228 	union xfs_btree_rec	*rec)
229 {
230 	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
231 	key->alloc.ar_startblock = 0;
232 }
233 
234 STATIC void
235 xfs_allocbt_init_rec_from_cur(
236 	struct xfs_btree_cur	*cur,
237 	union xfs_btree_rec	*rec)
238 {
239 	rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
240 	rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
241 }
242 
243 STATIC void
244 xfs_allocbt_init_ptr_from_cur(
245 	struct xfs_btree_cur	*cur,
246 	union xfs_btree_ptr	*ptr)
247 {
248 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
249 
250 	ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
251 	ASSERT(agf->agf_roots[cur->bc_btnum] != 0);
252 
253 	ptr->s = agf->agf_roots[cur->bc_btnum];
254 }
255 
256 STATIC int64_t
257 xfs_bnobt_key_diff(
258 	struct xfs_btree_cur	*cur,
259 	union xfs_btree_key	*key)
260 {
261 	xfs_alloc_rec_incore_t	*rec = &cur->bc_rec.a;
262 	xfs_alloc_key_t		*kp = &key->alloc;
263 
264 	return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
265 }
266 
267 STATIC int64_t
268 xfs_cntbt_key_diff(
269 	struct xfs_btree_cur	*cur,
270 	union xfs_btree_key	*key)
271 {
272 	xfs_alloc_rec_incore_t	*rec = &cur->bc_rec.a;
273 	xfs_alloc_key_t		*kp = &key->alloc;
274 	int64_t			diff;
275 
276 	diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
277 	if (diff)
278 		return diff;
279 
280 	return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
281 }
282 
283 STATIC int64_t
284 xfs_bnobt_diff_two_keys(
285 	struct xfs_btree_cur	*cur,
286 	union xfs_btree_key	*k1,
287 	union xfs_btree_key	*k2)
288 {
289 	return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
290 			  be32_to_cpu(k2->alloc.ar_startblock);
291 }
292 
293 STATIC int64_t
294 xfs_cntbt_diff_two_keys(
295 	struct xfs_btree_cur	*cur,
296 	union xfs_btree_key	*k1,
297 	union xfs_btree_key	*k2)
298 {
299 	int64_t			diff;
300 
301 	diff =  be32_to_cpu(k1->alloc.ar_blockcount) -
302 		be32_to_cpu(k2->alloc.ar_blockcount);
303 	if (diff)
304 		return diff;
305 
306 	return  be32_to_cpu(k1->alloc.ar_startblock) -
307 		be32_to_cpu(k2->alloc.ar_startblock);
308 }
309 
310 static xfs_failaddr_t
311 xfs_allocbt_verify(
312 	struct xfs_buf		*bp)
313 {
314 	struct xfs_mount	*mp = bp->b_target->bt_mount;
315 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
316 	struct xfs_perag	*pag = bp->b_pag;
317 	xfs_failaddr_t		fa;
318 	unsigned int		level;
319 
320 	/*
321 	 * magic number and level verification
322 	 *
323 	 * During growfs operations, we can't verify the exact level or owner as
324 	 * the perag is not fully initialised and hence not attached to the
325 	 * buffer.  In this case, check against the maximum tree depth.
326 	 *
327 	 * Similarly, during log recovery we will have a perag structure
328 	 * attached, but the agf information will not yet have been initialised
329 	 * from the on disk AGF. Again, we can only check against maximum limits
330 	 * in this case.
331 	 */
332 	level = be16_to_cpu(block->bb_level);
333 	switch (block->bb_magic) {
334 	case cpu_to_be32(XFS_ABTB_CRC_MAGIC):
335 		fa = xfs_btree_sblock_v5hdr_verify(bp);
336 		if (fa)
337 			return fa;
338 		/* fall through */
339 	case cpu_to_be32(XFS_ABTB_MAGIC):
340 		if (pag && pag->pagf_init) {
341 			if (level >= pag->pagf_levels[XFS_BTNUM_BNOi])
342 				return __this_address;
343 		} else if (level >= mp->m_ag_maxlevels)
344 			return __this_address;
345 		break;
346 	case cpu_to_be32(XFS_ABTC_CRC_MAGIC):
347 		fa = xfs_btree_sblock_v5hdr_verify(bp);
348 		if (fa)
349 			return fa;
350 		/* fall through */
351 	case cpu_to_be32(XFS_ABTC_MAGIC):
352 		if (pag && pag->pagf_init) {
353 			if (level >= pag->pagf_levels[XFS_BTNUM_CNTi])
354 				return __this_address;
355 		} else if (level >= mp->m_ag_maxlevels)
356 			return __this_address;
357 		break;
358 	default:
359 		return __this_address;
360 	}
361 
362 	return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
363 }
364 
365 static void
366 xfs_allocbt_read_verify(
367 	struct xfs_buf	*bp)
368 {
369 	xfs_failaddr_t	fa;
370 
371 	if (!xfs_btree_sblock_verify_crc(bp))
372 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
373 	else {
374 		fa = xfs_allocbt_verify(bp);
375 		if (fa)
376 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
377 	}
378 
379 	if (bp->b_error)
380 		trace_xfs_btree_corrupt(bp, _RET_IP_);
381 }
382 
383 static void
384 xfs_allocbt_write_verify(
385 	struct xfs_buf	*bp)
386 {
387 	xfs_failaddr_t	fa;
388 
389 	fa = xfs_allocbt_verify(bp);
390 	if (fa) {
391 		trace_xfs_btree_corrupt(bp, _RET_IP_);
392 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
393 		return;
394 	}
395 	xfs_btree_sblock_calc_crc(bp);
396 
397 }
398 
399 const struct xfs_buf_ops xfs_allocbt_buf_ops = {
400 	.name = "xfs_allocbt",
401 	.verify_read = xfs_allocbt_read_verify,
402 	.verify_write = xfs_allocbt_write_verify,
403 	.verify_struct = xfs_allocbt_verify,
404 };
405 
406 
407 STATIC int
408 xfs_bnobt_keys_inorder(
409 	struct xfs_btree_cur	*cur,
410 	union xfs_btree_key	*k1,
411 	union xfs_btree_key	*k2)
412 {
413 	return be32_to_cpu(k1->alloc.ar_startblock) <
414 	       be32_to_cpu(k2->alloc.ar_startblock);
415 }
416 
417 STATIC int
418 xfs_bnobt_recs_inorder(
419 	struct xfs_btree_cur	*cur,
420 	union xfs_btree_rec	*r1,
421 	union xfs_btree_rec	*r2)
422 {
423 	return be32_to_cpu(r1->alloc.ar_startblock) +
424 		be32_to_cpu(r1->alloc.ar_blockcount) <=
425 		be32_to_cpu(r2->alloc.ar_startblock);
426 }
427 
428 STATIC int
429 xfs_cntbt_keys_inorder(
430 	struct xfs_btree_cur	*cur,
431 	union xfs_btree_key	*k1,
432 	union xfs_btree_key	*k2)
433 {
434 	return be32_to_cpu(k1->alloc.ar_blockcount) <
435 		be32_to_cpu(k2->alloc.ar_blockcount) ||
436 		(k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
437 		 be32_to_cpu(k1->alloc.ar_startblock) <
438 		 be32_to_cpu(k2->alloc.ar_startblock));
439 }
440 
441 STATIC int
442 xfs_cntbt_recs_inorder(
443 	struct xfs_btree_cur	*cur,
444 	union xfs_btree_rec	*r1,
445 	union xfs_btree_rec	*r2)
446 {
447 	return be32_to_cpu(r1->alloc.ar_blockcount) <
448 		be32_to_cpu(r2->alloc.ar_blockcount) ||
449 		(r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
450 		 be32_to_cpu(r1->alloc.ar_startblock) <
451 		 be32_to_cpu(r2->alloc.ar_startblock));
452 }
453 
454 static const struct xfs_btree_ops xfs_bnobt_ops = {
455 	.rec_len		= sizeof(xfs_alloc_rec_t),
456 	.key_len		= sizeof(xfs_alloc_key_t),
457 
458 	.dup_cursor		= xfs_allocbt_dup_cursor,
459 	.set_root		= xfs_allocbt_set_root,
460 	.alloc_block		= xfs_allocbt_alloc_block,
461 	.free_block		= xfs_allocbt_free_block,
462 	.update_lastrec		= xfs_allocbt_update_lastrec,
463 	.get_minrecs		= xfs_allocbt_get_minrecs,
464 	.get_maxrecs		= xfs_allocbt_get_maxrecs,
465 	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
466 	.init_high_key_from_rec	= xfs_bnobt_init_high_key_from_rec,
467 	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
468 	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
469 	.key_diff		= xfs_bnobt_key_diff,
470 	.buf_ops		= &xfs_allocbt_buf_ops,
471 	.diff_two_keys		= xfs_bnobt_diff_two_keys,
472 	.keys_inorder		= xfs_bnobt_keys_inorder,
473 	.recs_inorder		= xfs_bnobt_recs_inorder,
474 };
475 
476 static const struct xfs_btree_ops xfs_cntbt_ops = {
477 	.rec_len		= sizeof(xfs_alloc_rec_t),
478 	.key_len		= sizeof(xfs_alloc_key_t),
479 
480 	.dup_cursor		= xfs_allocbt_dup_cursor,
481 	.set_root		= xfs_allocbt_set_root,
482 	.alloc_block		= xfs_allocbt_alloc_block,
483 	.free_block		= xfs_allocbt_free_block,
484 	.update_lastrec		= xfs_allocbt_update_lastrec,
485 	.get_minrecs		= xfs_allocbt_get_minrecs,
486 	.get_maxrecs		= xfs_allocbt_get_maxrecs,
487 	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
488 	.init_high_key_from_rec	= xfs_cntbt_init_high_key_from_rec,
489 	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
490 	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
491 	.key_diff		= xfs_cntbt_key_diff,
492 	.buf_ops		= &xfs_allocbt_buf_ops,
493 	.diff_two_keys		= xfs_cntbt_diff_two_keys,
494 	.keys_inorder		= xfs_cntbt_keys_inorder,
495 	.recs_inorder		= xfs_cntbt_recs_inorder,
496 };
497 
498 /*
499  * Allocate a new allocation btree cursor.
500  */
501 struct xfs_btree_cur *			/* new alloc btree cursor */
502 xfs_allocbt_init_cursor(
503 	struct xfs_mount	*mp,		/* file system mount point */
504 	struct xfs_trans	*tp,		/* transaction pointer */
505 	struct xfs_buf		*agbp,		/* buffer for agf structure */
506 	xfs_agnumber_t		agno,		/* allocation group number */
507 	xfs_btnum_t		btnum)		/* btree identifier */
508 {
509 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
510 	struct xfs_btree_cur	*cur;
511 
512 	ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
513 
514 	cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
515 
516 	cur->bc_tp = tp;
517 	cur->bc_mp = mp;
518 	cur->bc_btnum = btnum;
519 	cur->bc_blocklog = mp->m_sb.sb_blocklog;
520 
521 	if (btnum == XFS_BTNUM_CNT) {
522 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
523 		cur->bc_ops = &xfs_cntbt_ops;
524 		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
525 		cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
526 	} else {
527 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
528 		cur->bc_ops = &xfs_bnobt_ops;
529 		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
530 	}
531 
532 	cur->bc_private.a.agbp = agbp;
533 	cur->bc_private.a.agno = agno;
534 
535 	if (xfs_sb_version_hascrc(&mp->m_sb))
536 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
537 
538 	return cur;
539 }
540 
541 /*
542  * Calculate number of records in an alloc btree block.
543  */
544 int
545 xfs_allocbt_maxrecs(
546 	struct xfs_mount	*mp,
547 	int			blocklen,
548 	int			leaf)
549 {
550 	blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
551 
552 	if (leaf)
553 		return blocklen / sizeof(xfs_alloc_rec_t);
554 	return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
555 }
556