xref: /linux/fs/xfs/libxfs/xfs_alloc_btree.c (revision 3a6541e97c035dba90cdf37169d73b2d8057e55d)
1  // SPDX-License-Identifier: GPL-2.0
2  /*
3   * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
4   * All Rights Reserved.
5   */
6  #include "xfs.h"
7  #include "xfs_fs.h"
8  #include "xfs_shared.h"
9  #include "xfs_format.h"
10  #include "xfs_log_format.h"
11  #include "xfs_trans_resv.h"
12  #include "xfs_mount.h"
13  #include "xfs_btree.h"
14  #include "xfs_btree_staging.h"
15  #include "xfs_alloc_btree.h"
16  #include "xfs_alloc.h"
17  #include "xfs_extent_busy.h"
18  #include "xfs_error.h"
19  #include "xfs_trace.h"
20  #include "xfs_trans.h"
21  #include "xfs_ag.h"
22  
23  
24  STATIC struct xfs_btree_cur *
25  xfs_allocbt_dup_cursor(
26  	struct xfs_btree_cur	*cur)
27  {
28  	return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
29  			cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
30  }
31  
32  STATIC void
33  xfs_allocbt_set_root(
34  	struct xfs_btree_cur	*cur,
35  	union xfs_btree_ptr	*ptr,
36  	int			inc)
37  {
38  	struct xfs_buf		*agbp = cur->bc_ag.agbp;
39  	struct xfs_agf		*agf = agbp->b_addr;
40  	int			btnum = cur->bc_btnum;
41  
42  	ASSERT(ptr->s != 0);
43  
44  	agf->agf_roots[btnum] = ptr->s;
45  	be32_add_cpu(&agf->agf_levels[btnum], inc);
46  	cur->bc_ag.pag->pagf_levels[btnum] += inc;
47  
48  	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
49  }
50  
51  STATIC int
52  xfs_allocbt_alloc_block(
53  	struct xfs_btree_cur	*cur,
54  	union xfs_btree_ptr	*start,
55  	union xfs_btree_ptr	*new,
56  	int			*stat)
57  {
58  	int			error;
59  	xfs_agblock_t		bno;
60  
61  	/* Allocate the new block from the freelist. If we can't, give up.  */
62  	error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_ag.agbp,
63  				       &bno, 1);
64  	if (error)
65  		return error;
66  
67  	if (bno == NULLAGBLOCK) {
68  		*stat = 0;
69  		return 0;
70  	}
71  
72  	atomic64_inc(&cur->bc_mp->m_allocbt_blks);
73  	xfs_extent_busy_reuse(cur->bc_mp, cur->bc_ag.agbp->b_pag, bno, 1, false);
74  
75  	new->s = cpu_to_be32(bno);
76  
77  	*stat = 1;
78  	return 0;
79  }
80  
81  STATIC int
82  xfs_allocbt_free_block(
83  	struct xfs_btree_cur	*cur,
84  	struct xfs_buf		*bp)
85  {
86  	struct xfs_buf		*agbp = cur->bc_ag.agbp;
87  	xfs_agblock_t		bno;
88  	int			error;
89  
90  	bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
91  	error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
92  	if (error)
93  		return error;
94  
95  	atomic64_dec(&cur->bc_mp->m_allocbt_blks);
96  	xfs_extent_busy_insert(cur->bc_tp, agbp->b_pag, bno, 1,
97  			      XFS_EXTENT_BUSY_SKIP_DISCARD);
98  	return 0;
99  }
100  
101  /*
102   * Update the longest extent in the AGF
103   */
104  STATIC void
105  xfs_allocbt_update_lastrec(
106  	struct xfs_btree_cur	*cur,
107  	struct xfs_btree_block	*block,
108  	union xfs_btree_rec	*rec,
109  	int			ptr,
110  	int			reason)
111  {
112  	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
113  	struct xfs_perag	*pag;
114  	__be32			len;
115  	int			numrecs;
116  
117  	ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
118  
119  	switch (reason) {
120  	case LASTREC_UPDATE:
121  		/*
122  		 * If this is the last leaf block and it's the last record,
123  		 * then update the size of the longest extent in the AG.
124  		 */
125  		if (ptr != xfs_btree_get_numrecs(block))
126  			return;
127  		len = rec->alloc.ar_blockcount;
128  		break;
129  	case LASTREC_INSREC:
130  		if (be32_to_cpu(rec->alloc.ar_blockcount) <=
131  		    be32_to_cpu(agf->agf_longest))
132  			return;
133  		len = rec->alloc.ar_blockcount;
134  		break;
135  	case LASTREC_DELREC:
136  		numrecs = xfs_btree_get_numrecs(block);
137  		if (ptr <= numrecs)
138  			return;
139  		ASSERT(ptr == numrecs + 1);
140  
141  		if (numrecs) {
142  			xfs_alloc_rec_t *rrp;
143  
144  			rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
145  			len = rrp->ar_blockcount;
146  		} else {
147  			len = 0;
148  		}
149  
150  		break;
151  	default:
152  		ASSERT(0);
153  		return;
154  	}
155  
156  	agf->agf_longest = len;
157  	pag = cur->bc_ag.agbp->b_pag;
158  	pag->pagf_longest = be32_to_cpu(len);
159  	xfs_alloc_log_agf(cur->bc_tp, cur->bc_ag.agbp, XFS_AGF_LONGEST);
160  }
161  
162  STATIC int
163  xfs_allocbt_get_minrecs(
164  	struct xfs_btree_cur	*cur,
165  	int			level)
166  {
167  	return cur->bc_mp->m_alloc_mnr[level != 0];
168  }
169  
170  STATIC int
171  xfs_allocbt_get_maxrecs(
172  	struct xfs_btree_cur	*cur,
173  	int			level)
174  {
175  	return cur->bc_mp->m_alloc_mxr[level != 0];
176  }
177  
178  STATIC void
179  xfs_allocbt_init_key_from_rec(
180  	union xfs_btree_key	*key,
181  	union xfs_btree_rec	*rec)
182  {
183  	key->alloc.ar_startblock = rec->alloc.ar_startblock;
184  	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
185  }
186  
187  STATIC void
188  xfs_bnobt_init_high_key_from_rec(
189  	union xfs_btree_key	*key,
190  	union xfs_btree_rec	*rec)
191  {
192  	__u32			x;
193  
194  	x = be32_to_cpu(rec->alloc.ar_startblock);
195  	x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
196  	key->alloc.ar_startblock = cpu_to_be32(x);
197  	key->alloc.ar_blockcount = 0;
198  }
199  
200  STATIC void
201  xfs_cntbt_init_high_key_from_rec(
202  	union xfs_btree_key	*key,
203  	union xfs_btree_rec	*rec)
204  {
205  	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
206  	key->alloc.ar_startblock = 0;
207  }
208  
209  STATIC void
210  xfs_allocbt_init_rec_from_cur(
211  	struct xfs_btree_cur	*cur,
212  	union xfs_btree_rec	*rec)
213  {
214  	rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
215  	rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
216  }
217  
218  STATIC void
219  xfs_allocbt_init_ptr_from_cur(
220  	struct xfs_btree_cur	*cur,
221  	union xfs_btree_ptr	*ptr)
222  {
223  	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
224  
225  	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
226  
227  	ptr->s = agf->agf_roots[cur->bc_btnum];
228  }
229  
230  STATIC int64_t
231  xfs_bnobt_key_diff(
232  	struct xfs_btree_cur	*cur,
233  	union xfs_btree_key	*key)
234  {
235  	xfs_alloc_rec_incore_t	*rec = &cur->bc_rec.a;
236  	xfs_alloc_key_t		*kp = &key->alloc;
237  
238  	return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
239  }
240  
241  STATIC int64_t
242  xfs_cntbt_key_diff(
243  	struct xfs_btree_cur	*cur,
244  	union xfs_btree_key	*key)
245  {
246  	xfs_alloc_rec_incore_t	*rec = &cur->bc_rec.a;
247  	xfs_alloc_key_t		*kp = &key->alloc;
248  	int64_t			diff;
249  
250  	diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
251  	if (diff)
252  		return diff;
253  
254  	return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
255  }
256  
257  STATIC int64_t
258  xfs_bnobt_diff_two_keys(
259  	struct xfs_btree_cur	*cur,
260  	union xfs_btree_key	*k1,
261  	union xfs_btree_key	*k2)
262  {
263  	return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
264  			  be32_to_cpu(k2->alloc.ar_startblock);
265  }
266  
267  STATIC int64_t
268  xfs_cntbt_diff_two_keys(
269  	struct xfs_btree_cur	*cur,
270  	union xfs_btree_key	*k1,
271  	union xfs_btree_key	*k2)
272  {
273  	int64_t			diff;
274  
275  	diff =  be32_to_cpu(k1->alloc.ar_blockcount) -
276  		be32_to_cpu(k2->alloc.ar_blockcount);
277  	if (diff)
278  		return diff;
279  
280  	return  be32_to_cpu(k1->alloc.ar_startblock) -
281  		be32_to_cpu(k2->alloc.ar_startblock);
282  }
283  
284  static xfs_failaddr_t
285  xfs_allocbt_verify(
286  	struct xfs_buf		*bp)
287  {
288  	struct xfs_mount	*mp = bp->b_mount;
289  	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
290  	struct xfs_perag	*pag = bp->b_pag;
291  	xfs_failaddr_t		fa;
292  	unsigned int		level;
293  	xfs_btnum_t		btnum = XFS_BTNUM_BNOi;
294  
295  	if (!xfs_verify_magic(bp, block->bb_magic))
296  		return __this_address;
297  
298  	if (xfs_sb_version_hascrc(&mp->m_sb)) {
299  		fa = xfs_btree_sblock_v5hdr_verify(bp);
300  		if (fa)
301  			return fa;
302  	}
303  
304  	/*
305  	 * The perag may not be attached during grow operations or fully
306  	 * initialized from the AGF during log recovery. Therefore we can only
307  	 * check against maximum tree depth from those contexts.
308  	 *
309  	 * Otherwise check against the per-tree limit. Peek at one of the
310  	 * verifier magic values to determine the type of tree we're verifying
311  	 * against.
312  	 */
313  	level = be16_to_cpu(block->bb_level);
314  	if (bp->b_ops->magic[0] == cpu_to_be32(XFS_ABTC_MAGIC))
315  		btnum = XFS_BTNUM_CNTi;
316  	if (pag && pag->pagf_init) {
317  		if (level >= pag->pagf_levels[btnum])
318  			return __this_address;
319  	} else if (level >= mp->m_ag_maxlevels)
320  		return __this_address;
321  
322  	return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
323  }
324  
325  static void
326  xfs_allocbt_read_verify(
327  	struct xfs_buf	*bp)
328  {
329  	xfs_failaddr_t	fa;
330  
331  	if (!xfs_btree_sblock_verify_crc(bp))
332  		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
333  	else {
334  		fa = xfs_allocbt_verify(bp);
335  		if (fa)
336  			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
337  	}
338  
339  	if (bp->b_error)
340  		trace_xfs_btree_corrupt(bp, _RET_IP_);
341  }
342  
343  static void
344  xfs_allocbt_write_verify(
345  	struct xfs_buf	*bp)
346  {
347  	xfs_failaddr_t	fa;
348  
349  	fa = xfs_allocbt_verify(bp);
350  	if (fa) {
351  		trace_xfs_btree_corrupt(bp, _RET_IP_);
352  		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
353  		return;
354  	}
355  	xfs_btree_sblock_calc_crc(bp);
356  
357  }
358  
359  const struct xfs_buf_ops xfs_bnobt_buf_ops = {
360  	.name = "xfs_bnobt",
361  	.magic = { cpu_to_be32(XFS_ABTB_MAGIC),
362  		   cpu_to_be32(XFS_ABTB_CRC_MAGIC) },
363  	.verify_read = xfs_allocbt_read_verify,
364  	.verify_write = xfs_allocbt_write_verify,
365  	.verify_struct = xfs_allocbt_verify,
366  };
367  
368  const struct xfs_buf_ops xfs_cntbt_buf_ops = {
369  	.name = "xfs_cntbt",
370  	.magic = { cpu_to_be32(XFS_ABTC_MAGIC),
371  		   cpu_to_be32(XFS_ABTC_CRC_MAGIC) },
372  	.verify_read = xfs_allocbt_read_verify,
373  	.verify_write = xfs_allocbt_write_verify,
374  	.verify_struct = xfs_allocbt_verify,
375  };
376  
377  STATIC int
378  xfs_bnobt_keys_inorder(
379  	struct xfs_btree_cur	*cur,
380  	union xfs_btree_key	*k1,
381  	union xfs_btree_key	*k2)
382  {
383  	return be32_to_cpu(k1->alloc.ar_startblock) <
384  	       be32_to_cpu(k2->alloc.ar_startblock);
385  }
386  
387  STATIC int
388  xfs_bnobt_recs_inorder(
389  	struct xfs_btree_cur	*cur,
390  	union xfs_btree_rec	*r1,
391  	union xfs_btree_rec	*r2)
392  {
393  	return be32_to_cpu(r1->alloc.ar_startblock) +
394  		be32_to_cpu(r1->alloc.ar_blockcount) <=
395  		be32_to_cpu(r2->alloc.ar_startblock);
396  }
397  
398  STATIC int
399  xfs_cntbt_keys_inorder(
400  	struct xfs_btree_cur	*cur,
401  	union xfs_btree_key	*k1,
402  	union xfs_btree_key	*k2)
403  {
404  	return be32_to_cpu(k1->alloc.ar_blockcount) <
405  		be32_to_cpu(k2->alloc.ar_blockcount) ||
406  		(k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
407  		 be32_to_cpu(k1->alloc.ar_startblock) <
408  		 be32_to_cpu(k2->alloc.ar_startblock));
409  }
410  
411  STATIC int
412  xfs_cntbt_recs_inorder(
413  	struct xfs_btree_cur	*cur,
414  	union xfs_btree_rec	*r1,
415  	union xfs_btree_rec	*r2)
416  {
417  	return be32_to_cpu(r1->alloc.ar_blockcount) <
418  		be32_to_cpu(r2->alloc.ar_blockcount) ||
419  		(r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
420  		 be32_to_cpu(r1->alloc.ar_startblock) <
421  		 be32_to_cpu(r2->alloc.ar_startblock));
422  }
423  
424  static const struct xfs_btree_ops xfs_bnobt_ops = {
425  	.rec_len		= sizeof(xfs_alloc_rec_t),
426  	.key_len		= sizeof(xfs_alloc_key_t),
427  
428  	.dup_cursor		= xfs_allocbt_dup_cursor,
429  	.set_root		= xfs_allocbt_set_root,
430  	.alloc_block		= xfs_allocbt_alloc_block,
431  	.free_block		= xfs_allocbt_free_block,
432  	.update_lastrec		= xfs_allocbt_update_lastrec,
433  	.get_minrecs		= xfs_allocbt_get_minrecs,
434  	.get_maxrecs		= xfs_allocbt_get_maxrecs,
435  	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
436  	.init_high_key_from_rec	= xfs_bnobt_init_high_key_from_rec,
437  	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
438  	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
439  	.key_diff		= xfs_bnobt_key_diff,
440  	.buf_ops		= &xfs_bnobt_buf_ops,
441  	.diff_two_keys		= xfs_bnobt_diff_two_keys,
442  	.keys_inorder		= xfs_bnobt_keys_inorder,
443  	.recs_inorder		= xfs_bnobt_recs_inorder,
444  };
445  
446  static const struct xfs_btree_ops xfs_cntbt_ops = {
447  	.rec_len		= sizeof(xfs_alloc_rec_t),
448  	.key_len		= sizeof(xfs_alloc_key_t),
449  
450  	.dup_cursor		= xfs_allocbt_dup_cursor,
451  	.set_root		= xfs_allocbt_set_root,
452  	.alloc_block		= xfs_allocbt_alloc_block,
453  	.free_block		= xfs_allocbt_free_block,
454  	.update_lastrec		= xfs_allocbt_update_lastrec,
455  	.get_minrecs		= xfs_allocbt_get_minrecs,
456  	.get_maxrecs		= xfs_allocbt_get_maxrecs,
457  	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
458  	.init_high_key_from_rec	= xfs_cntbt_init_high_key_from_rec,
459  	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
460  	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
461  	.key_diff		= xfs_cntbt_key_diff,
462  	.buf_ops		= &xfs_cntbt_buf_ops,
463  	.diff_two_keys		= xfs_cntbt_diff_two_keys,
464  	.keys_inorder		= xfs_cntbt_keys_inorder,
465  	.recs_inorder		= xfs_cntbt_recs_inorder,
466  };
467  
468  /* Allocate most of a new allocation btree cursor. */
469  STATIC struct xfs_btree_cur *
470  xfs_allocbt_init_common(
471  	struct xfs_mount	*mp,
472  	struct xfs_trans	*tp,
473  	struct xfs_perag	*pag,
474  	xfs_btnum_t		btnum)
475  {
476  	struct xfs_btree_cur	*cur;
477  
478  	ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
479  
480  	cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
481  
482  	cur->bc_tp = tp;
483  	cur->bc_mp = mp;
484  	cur->bc_btnum = btnum;
485  	cur->bc_blocklog = mp->m_sb.sb_blocklog;
486  	cur->bc_ag.abt.active = false;
487  
488  	if (btnum == XFS_BTNUM_CNT) {
489  		cur->bc_ops = &xfs_cntbt_ops;
490  		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
491  		cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
492  	} else {
493  		cur->bc_ops = &xfs_bnobt_ops;
494  		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
495  	}
496  
497  	/* take a reference for the cursor */
498  	atomic_inc(&pag->pag_ref);
499  	cur->bc_ag.pag = pag;
500  
501  	if (xfs_sb_version_hascrc(&mp->m_sb))
502  		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
503  
504  	return cur;
505  }
506  
507  /*
508   * Allocate a new allocation btree cursor.
509   */
510  struct xfs_btree_cur *			/* new alloc btree cursor */
511  xfs_allocbt_init_cursor(
512  	struct xfs_mount	*mp,		/* file system mount point */
513  	struct xfs_trans	*tp,		/* transaction pointer */
514  	struct xfs_buf		*agbp,		/* buffer for agf structure */
515  	struct xfs_perag	*pag,
516  	xfs_btnum_t		btnum)		/* btree identifier */
517  {
518  	struct xfs_agf		*agf = agbp->b_addr;
519  	struct xfs_btree_cur	*cur;
520  
521  	cur = xfs_allocbt_init_common(mp, tp, pag, btnum);
522  	if (btnum == XFS_BTNUM_CNT)
523  		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
524  	else
525  		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
526  
527  	cur->bc_ag.agbp = agbp;
528  
529  	return cur;
530  }
531  
532  /* Create a free space btree cursor with a fake root for staging. */
533  struct xfs_btree_cur *
534  xfs_allocbt_stage_cursor(
535  	struct xfs_mount	*mp,
536  	struct xbtree_afakeroot	*afake,
537  	struct xfs_perag	*pag,
538  	xfs_btnum_t		btnum)
539  {
540  	struct xfs_btree_cur	*cur;
541  
542  	cur = xfs_allocbt_init_common(mp, NULL, pag, btnum);
543  	xfs_btree_stage_afakeroot(cur, afake);
544  	return cur;
545  }
546  
547  /*
548   * Install a new free space btree root.  Caller is responsible for invalidating
549   * and freeing the old btree blocks.
550   */
551  void
552  xfs_allocbt_commit_staged_btree(
553  	struct xfs_btree_cur	*cur,
554  	struct xfs_trans	*tp,
555  	struct xfs_buf		*agbp)
556  {
557  	struct xfs_agf		*agf = agbp->b_addr;
558  	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
559  
560  	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
561  
562  	agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root);
563  	agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels);
564  	xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
565  
566  	if (cur->bc_btnum == XFS_BTNUM_BNO) {
567  		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_bnobt_ops);
568  	} else {
569  		cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE;
570  		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_cntbt_ops);
571  	}
572  }
573  
574  /*
575   * Calculate number of records in an alloc btree block.
576   */
577  int
578  xfs_allocbt_maxrecs(
579  	struct xfs_mount	*mp,
580  	int			blocklen,
581  	int			leaf)
582  {
583  	blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
584  
585  	if (leaf)
586  		return blocklen / sizeof(xfs_alloc_rec_t);
587  	return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
588  }
589  
590  /* Calculate the freespace btree size for some records. */
591  xfs_extlen_t
592  xfs_allocbt_calc_size(
593  	struct xfs_mount	*mp,
594  	unsigned long long	len)
595  {
596  	return xfs_btree_calc_size(mp->m_alloc_mnr, len);
597  }
598