xref: /linux/fs/xfs/libxfs/xfs_alloc_btree.c (revision fbc872c38c8fed31948c85683b5326ee5ab9fccc)
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 	ASSERT(rec->alloc.ar_startblock != 0);
209 
210 	key->alloc.ar_startblock = rec->alloc.ar_startblock;
211 	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
212 }
213 
214 STATIC void
215 xfs_allocbt_init_rec_from_key(
216 	union xfs_btree_key	*key,
217 	union xfs_btree_rec	*rec)
218 {
219 	ASSERT(key->alloc.ar_startblock != 0);
220 
221 	rec->alloc.ar_startblock = key->alloc.ar_startblock;
222 	rec->alloc.ar_blockcount = key->alloc.ar_blockcount;
223 }
224 
225 STATIC void
226 xfs_allocbt_init_rec_from_cur(
227 	struct xfs_btree_cur	*cur,
228 	union xfs_btree_rec	*rec)
229 {
230 	ASSERT(cur->bc_rec.a.ar_startblock != 0);
231 
232 	rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
233 	rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
234 }
235 
236 STATIC void
237 xfs_allocbt_init_ptr_from_cur(
238 	struct xfs_btree_cur	*cur,
239 	union xfs_btree_ptr	*ptr)
240 {
241 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
242 
243 	ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
244 	ASSERT(agf->agf_roots[cur->bc_btnum] != 0);
245 
246 	ptr->s = agf->agf_roots[cur->bc_btnum];
247 }
248 
249 STATIC __int64_t
250 xfs_allocbt_key_diff(
251 	struct xfs_btree_cur	*cur,
252 	union xfs_btree_key	*key)
253 {
254 	xfs_alloc_rec_incore_t	*rec = &cur->bc_rec.a;
255 	xfs_alloc_key_t		*kp = &key->alloc;
256 	__int64_t		diff;
257 
258 	if (cur->bc_btnum == XFS_BTNUM_BNO) {
259 		return (__int64_t)be32_to_cpu(kp->ar_startblock) -
260 				rec->ar_startblock;
261 	}
262 
263 	diff = (__int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
264 	if (diff)
265 		return diff;
266 
267 	return (__int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
268 }
269 
270 static bool
271 xfs_allocbt_verify(
272 	struct xfs_buf		*bp)
273 {
274 	struct xfs_mount	*mp = bp->b_target->bt_mount;
275 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
276 	struct xfs_perag	*pag = bp->b_pag;
277 	unsigned int		level;
278 
279 	/*
280 	 * magic number and level verification
281 	 *
282 	 * During growfs operations, we can't verify the exact level or owner as
283 	 * the perag is not fully initialised and hence not attached to the
284 	 * buffer.  In this case, check against the maximum tree depth.
285 	 *
286 	 * Similarly, during log recovery we will have a perag structure
287 	 * attached, but the agf information will not yet have been initialised
288 	 * from the on disk AGF. Again, we can only check against maximum limits
289 	 * in this case.
290 	 */
291 	level = be16_to_cpu(block->bb_level);
292 	switch (block->bb_magic) {
293 	case cpu_to_be32(XFS_ABTB_CRC_MAGIC):
294 		if (!xfs_btree_sblock_v5hdr_verify(bp))
295 			return false;
296 		/* fall through */
297 	case cpu_to_be32(XFS_ABTB_MAGIC):
298 		if (pag && pag->pagf_init) {
299 			if (level >= pag->pagf_levels[XFS_BTNUM_BNOi])
300 				return false;
301 		} else if (level >= mp->m_ag_maxlevels)
302 			return false;
303 		break;
304 	case cpu_to_be32(XFS_ABTC_CRC_MAGIC):
305 		if (!xfs_btree_sblock_v5hdr_verify(bp))
306 			return false;
307 		/* fall through */
308 	case cpu_to_be32(XFS_ABTC_MAGIC):
309 		if (pag && pag->pagf_init) {
310 			if (level >= pag->pagf_levels[XFS_BTNUM_CNTi])
311 				return false;
312 		} else if (level >= mp->m_ag_maxlevels)
313 			return false;
314 		break;
315 	default:
316 		return false;
317 	}
318 
319 	return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
320 }
321 
322 static void
323 xfs_allocbt_read_verify(
324 	struct xfs_buf	*bp)
325 {
326 	if (!xfs_btree_sblock_verify_crc(bp))
327 		xfs_buf_ioerror(bp, -EFSBADCRC);
328 	else if (!xfs_allocbt_verify(bp))
329 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
330 
331 	if (bp->b_error) {
332 		trace_xfs_btree_corrupt(bp, _RET_IP_);
333 		xfs_verifier_error(bp);
334 	}
335 }
336 
337 static void
338 xfs_allocbt_write_verify(
339 	struct xfs_buf	*bp)
340 {
341 	if (!xfs_allocbt_verify(bp)) {
342 		trace_xfs_btree_corrupt(bp, _RET_IP_);
343 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
344 		xfs_verifier_error(bp);
345 		return;
346 	}
347 	xfs_btree_sblock_calc_crc(bp);
348 
349 }
350 
351 const struct xfs_buf_ops xfs_allocbt_buf_ops = {
352 	.name = "xfs_allocbt",
353 	.verify_read = xfs_allocbt_read_verify,
354 	.verify_write = xfs_allocbt_write_verify,
355 };
356 
357 
358 #if defined(DEBUG) || defined(XFS_WARN)
359 STATIC int
360 xfs_allocbt_keys_inorder(
361 	struct xfs_btree_cur	*cur,
362 	union xfs_btree_key	*k1,
363 	union xfs_btree_key	*k2)
364 {
365 	if (cur->bc_btnum == XFS_BTNUM_BNO) {
366 		return be32_to_cpu(k1->alloc.ar_startblock) <
367 		       be32_to_cpu(k2->alloc.ar_startblock);
368 	} else {
369 		return be32_to_cpu(k1->alloc.ar_blockcount) <
370 			be32_to_cpu(k2->alloc.ar_blockcount) ||
371 			(k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
372 			 be32_to_cpu(k1->alloc.ar_startblock) <
373 			 be32_to_cpu(k2->alloc.ar_startblock));
374 	}
375 }
376 
377 STATIC int
378 xfs_allocbt_recs_inorder(
379 	struct xfs_btree_cur	*cur,
380 	union xfs_btree_rec	*r1,
381 	union xfs_btree_rec	*r2)
382 {
383 	if (cur->bc_btnum == XFS_BTNUM_BNO) {
384 		return be32_to_cpu(r1->alloc.ar_startblock) +
385 			be32_to_cpu(r1->alloc.ar_blockcount) <=
386 			be32_to_cpu(r2->alloc.ar_startblock);
387 	} else {
388 		return be32_to_cpu(r1->alloc.ar_blockcount) <
389 			be32_to_cpu(r2->alloc.ar_blockcount) ||
390 			(r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
391 			 be32_to_cpu(r1->alloc.ar_startblock) <
392 			 be32_to_cpu(r2->alloc.ar_startblock));
393 	}
394 }
395 #endif	/* DEBUG */
396 
397 static const struct xfs_btree_ops xfs_allocbt_ops = {
398 	.rec_len		= sizeof(xfs_alloc_rec_t),
399 	.key_len		= sizeof(xfs_alloc_key_t),
400 
401 	.dup_cursor		= xfs_allocbt_dup_cursor,
402 	.set_root		= xfs_allocbt_set_root,
403 	.alloc_block		= xfs_allocbt_alloc_block,
404 	.free_block		= xfs_allocbt_free_block,
405 	.update_lastrec		= xfs_allocbt_update_lastrec,
406 	.get_minrecs		= xfs_allocbt_get_minrecs,
407 	.get_maxrecs		= xfs_allocbt_get_maxrecs,
408 	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
409 	.init_rec_from_key	= xfs_allocbt_init_rec_from_key,
410 	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
411 	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
412 	.key_diff		= xfs_allocbt_key_diff,
413 	.buf_ops		= &xfs_allocbt_buf_ops,
414 #if defined(DEBUG) || defined(XFS_WARN)
415 	.keys_inorder		= xfs_allocbt_keys_inorder,
416 	.recs_inorder		= xfs_allocbt_recs_inorder,
417 #endif
418 };
419 
420 /*
421  * Allocate a new allocation btree cursor.
422  */
423 struct xfs_btree_cur *			/* new alloc btree cursor */
424 xfs_allocbt_init_cursor(
425 	struct xfs_mount	*mp,		/* file system mount point */
426 	struct xfs_trans	*tp,		/* transaction pointer */
427 	struct xfs_buf		*agbp,		/* buffer for agf structure */
428 	xfs_agnumber_t		agno,		/* allocation group number */
429 	xfs_btnum_t		btnum)		/* btree identifier */
430 {
431 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
432 	struct xfs_btree_cur	*cur;
433 
434 	ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
435 
436 	cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_SLEEP);
437 
438 	cur->bc_tp = tp;
439 	cur->bc_mp = mp;
440 	cur->bc_btnum = btnum;
441 	cur->bc_blocklog = mp->m_sb.sb_blocklog;
442 	cur->bc_ops = &xfs_allocbt_ops;
443 
444 	if (btnum == XFS_BTNUM_CNT) {
445 		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
446 		cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
447 	} else {
448 		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
449 	}
450 
451 	cur->bc_private.a.agbp = agbp;
452 	cur->bc_private.a.agno = agno;
453 
454 	if (xfs_sb_version_hascrc(&mp->m_sb))
455 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
456 
457 	return cur;
458 }
459 
460 /*
461  * Calculate number of records in an alloc btree block.
462  */
463 int
464 xfs_allocbt_maxrecs(
465 	struct xfs_mount	*mp,
466 	int			blocklen,
467 	int			leaf)
468 {
469 	blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
470 
471 	if (leaf)
472 		return blocklen / sizeof(xfs_alloc_rec_t);
473 	return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
474 }
475