xref: /openbmc/linux/fs/xfs/libxfs/xfs_alloc.c (revision d37cf9b63113f13d742713881ce691fc615d8b3b)
1  // SPDX-License-Identifier: GPL-2.0
2  /*
3   * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4   * All Rights Reserved.
5   */
6  #include "xfs.h"
7  #include "xfs_fs.h"
8  #include "xfs_format.h"
9  #include "xfs_log_format.h"
10  #include "xfs_shared.h"
11  #include "xfs_trans_resv.h"
12  #include "xfs_bit.h"
13  #include "xfs_mount.h"
14  #include "xfs_defer.h"
15  #include "xfs_btree.h"
16  #include "xfs_rmap.h"
17  #include "xfs_alloc_btree.h"
18  #include "xfs_alloc.h"
19  #include "xfs_extent_busy.h"
20  #include "xfs_errortag.h"
21  #include "xfs_error.h"
22  #include "xfs_trace.h"
23  #include "xfs_trans.h"
24  #include "xfs_buf_item.h"
25  #include "xfs_log.h"
26  #include "xfs_ag.h"
27  #include "xfs_ag_resv.h"
28  #include "xfs_bmap.h"
29  
30  struct kmem_cache	*xfs_extfree_item_cache;
31  
32  struct workqueue_struct *xfs_alloc_wq;
33  
34  #define XFS_ABSDIFF(a,b)	(((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
35  
36  #define	XFSA_FIXUP_BNO_OK	1
37  #define	XFSA_FIXUP_CNT_OK	2
38  
39  /*
40   * Size of the AGFL.  For CRC-enabled filesystes we steal a couple of slots in
41   * the beginning of the block for a proper header with the location information
42   * and CRC.
43   */
44  unsigned int
xfs_agfl_size(struct xfs_mount * mp)45  xfs_agfl_size(
46  	struct xfs_mount	*mp)
47  {
48  	unsigned int		size = mp->m_sb.sb_sectsize;
49  
50  	if (xfs_has_crc(mp))
51  		size -= sizeof(struct xfs_agfl);
52  
53  	return size / sizeof(xfs_agblock_t);
54  }
55  
56  unsigned int
xfs_refc_block(struct xfs_mount * mp)57  xfs_refc_block(
58  	struct xfs_mount	*mp)
59  {
60  	if (xfs_has_rmapbt(mp))
61  		return XFS_RMAP_BLOCK(mp) + 1;
62  	if (xfs_has_finobt(mp))
63  		return XFS_FIBT_BLOCK(mp) + 1;
64  	return XFS_IBT_BLOCK(mp) + 1;
65  }
66  
67  xfs_extlen_t
xfs_prealloc_blocks(struct xfs_mount * mp)68  xfs_prealloc_blocks(
69  	struct xfs_mount	*mp)
70  {
71  	if (xfs_has_reflink(mp))
72  		return xfs_refc_block(mp) + 1;
73  	if (xfs_has_rmapbt(mp))
74  		return XFS_RMAP_BLOCK(mp) + 1;
75  	if (xfs_has_finobt(mp))
76  		return XFS_FIBT_BLOCK(mp) + 1;
77  	return XFS_IBT_BLOCK(mp) + 1;
78  }
79  
80  /*
81   * The number of blocks per AG that we withhold from xfs_mod_fdblocks to
82   * guarantee that we can refill the AGFL prior to allocating space in a nearly
83   * full AG.  Although the space described by the free space btrees, the
84   * blocks used by the freesp btrees themselves, and the blocks owned by the
85   * AGFL are counted in the ondisk fdblocks, it's a mistake to let the ondisk
86   * free space in the AG drop so low that the free space btrees cannot refill an
87   * empty AGFL up to the minimum level.  Rather than grind through empty AGs
88   * until the fs goes down, we subtract this many AG blocks from the incore
89   * fdblocks to ensure user allocation does not overcommit the space the
90   * filesystem needs for the AGFLs.  The rmap btree uses a per-AG reservation to
91   * withhold space from xfs_mod_fdblocks, so we do not account for that here.
92   */
93  #define XFS_ALLOCBT_AGFL_RESERVE	4
94  
95  /*
96   * Compute the number of blocks that we set aside to guarantee the ability to
97   * refill the AGFL and handle a full bmap btree split.
98   *
99   * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of
100   * AGF buffer (PV 947395), we place constraints on the relationship among
101   * actual allocations for data blocks, freelist blocks, and potential file data
102   * bmap btree blocks. However, these restrictions may result in no actual space
103   * allocated for a delayed extent, for example, a data block in a certain AG is
104   * allocated but there is no additional block for the additional bmap btree
105   * block due to a split of the bmap btree of the file. The result of this may
106   * lead to an infinite loop when the file gets flushed to disk and all delayed
107   * extents need to be actually allocated. To get around this, we explicitly set
108   * aside a few blocks which will not be reserved in delayed allocation.
109   *
110   * For each AG, we need to reserve enough blocks to replenish a totally empty
111   * AGFL and 4 more to handle a potential split of the file's bmap btree.
112   */
113  unsigned int
xfs_alloc_set_aside(struct xfs_mount * mp)114  xfs_alloc_set_aside(
115  	struct xfs_mount	*mp)
116  {
117  	return mp->m_sb.sb_agcount * (XFS_ALLOCBT_AGFL_RESERVE + 4);
118  }
119  
120  /*
121   * When deciding how much space to allocate out of an AG, we limit the
122   * allocation maximum size to the size the AG. However, we cannot use all the
123   * blocks in the AG - some are permanently used by metadata. These
124   * blocks are generally:
125   *	- the AG superblock, AGF, AGI and AGFL
126   *	- the AGF (bno and cnt) and AGI btree root blocks, and optionally
127   *	  the AGI free inode and rmap btree root blocks.
128   *	- blocks on the AGFL according to xfs_alloc_set_aside() limits
129   *	- the rmapbt root block
130   *
131   * The AG headers are sector sized, so the amount of space they take up is
132   * dependent on filesystem geometry. The others are all single blocks.
133   */
134  unsigned int
xfs_alloc_ag_max_usable(struct xfs_mount * mp)135  xfs_alloc_ag_max_usable(
136  	struct xfs_mount	*mp)
137  {
138  	unsigned int		blocks;
139  
140  	blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */
141  	blocks += XFS_ALLOCBT_AGFL_RESERVE;
142  	blocks += 3;			/* AGF, AGI btree root blocks */
143  	if (xfs_has_finobt(mp))
144  		blocks++;		/* finobt root block */
145  	if (xfs_has_rmapbt(mp))
146  		blocks++;		/* rmap root block */
147  	if (xfs_has_reflink(mp))
148  		blocks++;		/* refcount root block */
149  
150  	return mp->m_sb.sb_agblocks - blocks;
151  }
152  
153  /*
154   * Lookup the record equal to [bno, len] in the btree given by cur.
155   */
156  STATIC int				/* error */
xfs_alloc_lookup_eq(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len,int * stat)157  xfs_alloc_lookup_eq(
158  	struct xfs_btree_cur	*cur,	/* btree cursor */
159  	xfs_agblock_t		bno,	/* starting block of extent */
160  	xfs_extlen_t		len,	/* length of extent */
161  	int			*stat)	/* success/failure */
162  {
163  	int			error;
164  
165  	cur->bc_rec.a.ar_startblock = bno;
166  	cur->bc_rec.a.ar_blockcount = len;
167  	error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
168  	cur->bc_ag.abt.active = (*stat == 1);
169  	return error;
170  }
171  
172  /*
173   * Lookup the first record greater than or equal to [bno, len]
174   * in the btree given by cur.
175   */
176  int				/* error */
xfs_alloc_lookup_ge(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len,int * stat)177  xfs_alloc_lookup_ge(
178  	struct xfs_btree_cur	*cur,	/* btree cursor */
179  	xfs_agblock_t		bno,	/* starting block of extent */
180  	xfs_extlen_t		len,	/* length of extent */
181  	int			*stat)	/* success/failure */
182  {
183  	int			error;
184  
185  	cur->bc_rec.a.ar_startblock = bno;
186  	cur->bc_rec.a.ar_blockcount = len;
187  	error = xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
188  	cur->bc_ag.abt.active = (*stat == 1);
189  	return error;
190  }
191  
192  /*
193   * Lookup the first record less than or equal to [bno, len]
194   * in the btree given by cur.
195   */
196  int					/* error */
xfs_alloc_lookup_le(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len,int * stat)197  xfs_alloc_lookup_le(
198  	struct xfs_btree_cur	*cur,	/* btree cursor */
199  	xfs_agblock_t		bno,	/* starting block of extent */
200  	xfs_extlen_t		len,	/* length of extent */
201  	int			*stat)	/* success/failure */
202  {
203  	int			error;
204  	cur->bc_rec.a.ar_startblock = bno;
205  	cur->bc_rec.a.ar_blockcount = len;
206  	error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
207  	cur->bc_ag.abt.active = (*stat == 1);
208  	return error;
209  }
210  
211  static inline bool
xfs_alloc_cur_active(struct xfs_btree_cur * cur)212  xfs_alloc_cur_active(
213  	struct xfs_btree_cur	*cur)
214  {
215  	return cur && cur->bc_ag.abt.active;
216  }
217  
218  /*
219   * Update the record referred to by cur to the value given
220   * by [bno, len].
221   * This either works (return 0) or gets an EFSCORRUPTED error.
222   */
223  STATIC int				/* error */
xfs_alloc_update(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len)224  xfs_alloc_update(
225  	struct xfs_btree_cur	*cur,	/* btree cursor */
226  	xfs_agblock_t		bno,	/* starting block of extent */
227  	xfs_extlen_t		len)	/* length of extent */
228  {
229  	union xfs_btree_rec	rec;
230  
231  	rec.alloc.ar_startblock = cpu_to_be32(bno);
232  	rec.alloc.ar_blockcount = cpu_to_be32(len);
233  	return xfs_btree_update(cur, &rec);
234  }
235  
236  /* Convert the ondisk btree record to its incore representation. */
237  void
xfs_alloc_btrec_to_irec(const union xfs_btree_rec * rec,struct xfs_alloc_rec_incore * irec)238  xfs_alloc_btrec_to_irec(
239  	const union xfs_btree_rec	*rec,
240  	struct xfs_alloc_rec_incore	*irec)
241  {
242  	irec->ar_startblock = be32_to_cpu(rec->alloc.ar_startblock);
243  	irec->ar_blockcount = be32_to_cpu(rec->alloc.ar_blockcount);
244  }
245  
246  /* Simple checks for free space records. */
247  xfs_failaddr_t
xfs_alloc_check_irec(struct xfs_btree_cur * cur,const struct xfs_alloc_rec_incore * irec)248  xfs_alloc_check_irec(
249  	struct xfs_btree_cur		*cur,
250  	const struct xfs_alloc_rec_incore *irec)
251  {
252  	struct xfs_perag		*pag = cur->bc_ag.pag;
253  
254  	if (irec->ar_blockcount == 0)
255  		return __this_address;
256  
257  	/* check for valid extent range, including overflow */
258  	if (!xfs_verify_agbext(pag, irec->ar_startblock, irec->ar_blockcount))
259  		return __this_address;
260  
261  	return NULL;
262  }
263  
264  static inline int
xfs_alloc_complain_bad_rec(struct xfs_btree_cur * cur,xfs_failaddr_t fa,const struct xfs_alloc_rec_incore * irec)265  xfs_alloc_complain_bad_rec(
266  	struct xfs_btree_cur		*cur,
267  	xfs_failaddr_t			fa,
268  	const struct xfs_alloc_rec_incore *irec)
269  {
270  	struct xfs_mount		*mp = cur->bc_mp;
271  
272  	xfs_warn(mp,
273  		"%s Freespace BTree record corruption in AG %d detected at %pS!",
274  		cur->bc_btnum == XFS_BTNUM_BNO ? "Block" : "Size",
275  		cur->bc_ag.pag->pag_agno, fa);
276  	xfs_warn(mp,
277  		"start block 0x%x block count 0x%x", irec->ar_startblock,
278  		irec->ar_blockcount);
279  	return -EFSCORRUPTED;
280  }
281  
282  /*
283   * Get the data from the pointed-to record.
284   */
285  int					/* error */
xfs_alloc_get_rec(struct xfs_btree_cur * cur,xfs_agblock_t * bno,xfs_extlen_t * len,int * stat)286  xfs_alloc_get_rec(
287  	struct xfs_btree_cur	*cur,	/* btree cursor */
288  	xfs_agblock_t		*bno,	/* output: starting block of extent */
289  	xfs_extlen_t		*len,	/* output: length of extent */
290  	int			*stat)	/* output: success/failure */
291  {
292  	struct xfs_alloc_rec_incore irec;
293  	union xfs_btree_rec	*rec;
294  	xfs_failaddr_t		fa;
295  	int			error;
296  
297  	error = xfs_btree_get_rec(cur, &rec, stat);
298  	if (error || !(*stat))
299  		return error;
300  
301  	xfs_alloc_btrec_to_irec(rec, &irec);
302  	fa = xfs_alloc_check_irec(cur, &irec);
303  	if (fa)
304  		return xfs_alloc_complain_bad_rec(cur, fa, &irec);
305  
306  	*bno = irec.ar_startblock;
307  	*len = irec.ar_blockcount;
308  	return 0;
309  }
310  
311  /*
312   * Compute aligned version of the found extent.
313   * Takes alignment and min length into account.
314   */
315  STATIC bool
xfs_alloc_compute_aligned(xfs_alloc_arg_t * args,xfs_agblock_t foundbno,xfs_extlen_t foundlen,xfs_agblock_t * resbno,xfs_extlen_t * reslen,unsigned * busy_gen)316  xfs_alloc_compute_aligned(
317  	xfs_alloc_arg_t	*args,		/* allocation argument structure */
318  	xfs_agblock_t	foundbno,	/* starting block in found extent */
319  	xfs_extlen_t	foundlen,	/* length in found extent */
320  	xfs_agblock_t	*resbno,	/* result block number */
321  	xfs_extlen_t	*reslen,	/* result length */
322  	unsigned	*busy_gen)
323  {
324  	xfs_agblock_t	bno = foundbno;
325  	xfs_extlen_t	len = foundlen;
326  	xfs_extlen_t	diff;
327  	bool		busy;
328  
329  	/* Trim busy sections out of found extent */
330  	busy = xfs_extent_busy_trim(args, &bno, &len, busy_gen);
331  
332  	/*
333  	 * If we have a largish extent that happens to start before min_agbno,
334  	 * see if we can shift it into range...
335  	 */
336  	if (bno < args->min_agbno && bno + len > args->min_agbno) {
337  		diff = args->min_agbno - bno;
338  		if (len > diff) {
339  			bno += diff;
340  			len -= diff;
341  		}
342  	}
343  
344  	if (args->alignment > 1 && len >= args->minlen) {
345  		xfs_agblock_t	aligned_bno = roundup(bno, args->alignment);
346  
347  		diff = aligned_bno - bno;
348  
349  		*resbno = aligned_bno;
350  		*reslen = diff >= len ? 0 : len - diff;
351  	} else {
352  		*resbno = bno;
353  		*reslen = len;
354  	}
355  
356  	return busy;
357  }
358  
359  /*
360   * Compute best start block and diff for "near" allocations.
361   * freelen >= wantlen already checked by caller.
362   */
363  STATIC xfs_extlen_t			/* difference value (absolute) */
xfs_alloc_compute_diff(xfs_agblock_t wantbno,xfs_extlen_t wantlen,xfs_extlen_t alignment,int datatype,xfs_agblock_t freebno,xfs_extlen_t freelen,xfs_agblock_t * newbnop)364  xfs_alloc_compute_diff(
365  	xfs_agblock_t	wantbno,	/* target starting block */
366  	xfs_extlen_t	wantlen,	/* target length */
367  	xfs_extlen_t	alignment,	/* target alignment */
368  	int		datatype,	/* are we allocating data? */
369  	xfs_agblock_t	freebno,	/* freespace's starting block */
370  	xfs_extlen_t	freelen,	/* freespace's length */
371  	xfs_agblock_t	*newbnop)	/* result: best start block from free */
372  {
373  	xfs_agblock_t	freeend;	/* end of freespace extent */
374  	xfs_agblock_t	newbno1;	/* return block number */
375  	xfs_agblock_t	newbno2;	/* other new block number */
376  	xfs_extlen_t	newlen1=0;	/* length with newbno1 */
377  	xfs_extlen_t	newlen2=0;	/* length with newbno2 */
378  	xfs_agblock_t	wantend;	/* end of target extent */
379  	bool		userdata = datatype & XFS_ALLOC_USERDATA;
380  
381  	ASSERT(freelen >= wantlen);
382  	freeend = freebno + freelen;
383  	wantend = wantbno + wantlen;
384  	/*
385  	 * We want to allocate from the start of a free extent if it is past
386  	 * the desired block or if we are allocating user data and the free
387  	 * extent is before desired block. The second case is there to allow
388  	 * for contiguous allocation from the remaining free space if the file
389  	 * grows in the short term.
390  	 */
391  	if (freebno >= wantbno || (userdata && freeend < wantend)) {
392  		if ((newbno1 = roundup(freebno, alignment)) >= freeend)
393  			newbno1 = NULLAGBLOCK;
394  	} else if (freeend >= wantend && alignment > 1) {
395  		newbno1 = roundup(wantbno, alignment);
396  		newbno2 = newbno1 - alignment;
397  		if (newbno1 >= freeend)
398  			newbno1 = NULLAGBLOCK;
399  		else
400  			newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
401  		if (newbno2 < freebno)
402  			newbno2 = NULLAGBLOCK;
403  		else
404  			newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
405  		if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
406  			if (newlen1 < newlen2 ||
407  			    (newlen1 == newlen2 &&
408  			     XFS_ABSDIFF(newbno1, wantbno) >
409  			     XFS_ABSDIFF(newbno2, wantbno)))
410  				newbno1 = newbno2;
411  		} else if (newbno2 != NULLAGBLOCK)
412  			newbno1 = newbno2;
413  	} else if (freeend >= wantend) {
414  		newbno1 = wantbno;
415  	} else if (alignment > 1) {
416  		newbno1 = roundup(freeend - wantlen, alignment);
417  		if (newbno1 > freeend - wantlen &&
418  		    newbno1 - alignment >= freebno)
419  			newbno1 -= alignment;
420  		else if (newbno1 >= freeend)
421  			newbno1 = NULLAGBLOCK;
422  	} else
423  		newbno1 = freeend - wantlen;
424  	*newbnop = newbno1;
425  	return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
426  }
427  
428  /*
429   * Fix up the length, based on mod and prod.
430   * len should be k * prod + mod for some k.
431   * If len is too small it is returned unchanged.
432   * If len hits maxlen it is left alone.
433   */
434  STATIC void
xfs_alloc_fix_len(xfs_alloc_arg_t * args)435  xfs_alloc_fix_len(
436  	xfs_alloc_arg_t	*args)		/* allocation argument structure */
437  {
438  	xfs_extlen_t	k;
439  	xfs_extlen_t	rlen;
440  
441  	ASSERT(args->mod < args->prod);
442  	rlen = args->len;
443  	ASSERT(rlen >= args->minlen);
444  	ASSERT(rlen <= args->maxlen);
445  	if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
446  	    (args->mod == 0 && rlen < args->prod))
447  		return;
448  	k = rlen % args->prod;
449  	if (k == args->mod)
450  		return;
451  	if (k > args->mod)
452  		rlen = rlen - (k - args->mod);
453  	else
454  		rlen = rlen - args->prod + (args->mod - k);
455  	/* casts to (int) catch length underflows */
456  	if ((int)rlen < (int)args->minlen)
457  		return;
458  	ASSERT(rlen >= args->minlen && rlen <= args->maxlen);
459  	ASSERT(rlen % args->prod == args->mod);
460  	ASSERT(args->pag->pagf_freeblks + args->pag->pagf_flcount >=
461  		rlen + args->minleft);
462  	args->len = rlen;
463  }
464  
465  /*
466   * Update the two btrees, logically removing from freespace the extent
467   * starting at rbno, rlen blocks.  The extent is contained within the
468   * actual (current) free extent fbno for flen blocks.
469   * Flags are passed in indicating whether the cursors are set to the
470   * relevant records.
471   */
472  STATIC int				/* error code */
xfs_alloc_fixup_trees(struct xfs_btree_cur * cnt_cur,struct xfs_btree_cur * bno_cur,xfs_agblock_t fbno,xfs_extlen_t flen,xfs_agblock_t rbno,xfs_extlen_t rlen,int flags)473  xfs_alloc_fixup_trees(
474  	struct xfs_btree_cur *cnt_cur,	/* cursor for by-size btree */
475  	struct xfs_btree_cur *bno_cur,	/* cursor for by-block btree */
476  	xfs_agblock_t	fbno,		/* starting block of free extent */
477  	xfs_extlen_t	flen,		/* length of free extent */
478  	xfs_agblock_t	rbno,		/* starting block of returned extent */
479  	xfs_extlen_t	rlen,		/* length of returned extent */
480  	int		flags)		/* flags, XFSA_FIXUP_... */
481  {
482  	int		error;		/* error code */
483  	int		i;		/* operation results */
484  	xfs_agblock_t	nfbno1;		/* first new free startblock */
485  	xfs_agblock_t	nfbno2;		/* second new free startblock */
486  	xfs_extlen_t	nflen1=0;	/* first new free length */
487  	xfs_extlen_t	nflen2=0;	/* second new free length */
488  	struct xfs_mount *mp;
489  
490  	mp = cnt_cur->bc_mp;
491  
492  	/*
493  	 * Look up the record in the by-size tree if necessary.
494  	 */
495  	if (flags & XFSA_FIXUP_CNT_OK) {
496  #ifdef DEBUG
497  		if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
498  			return error;
499  		if (XFS_IS_CORRUPT(mp,
500  				   i != 1 ||
501  				   nfbno1 != fbno ||
502  				   nflen1 != flen))
503  			return -EFSCORRUPTED;
504  #endif
505  	} else {
506  		if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
507  			return error;
508  		if (XFS_IS_CORRUPT(mp, i != 1))
509  			return -EFSCORRUPTED;
510  	}
511  	/*
512  	 * Look up the record in the by-block tree if necessary.
513  	 */
514  	if (flags & XFSA_FIXUP_BNO_OK) {
515  #ifdef DEBUG
516  		if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
517  			return error;
518  		if (XFS_IS_CORRUPT(mp,
519  				   i != 1 ||
520  				   nfbno1 != fbno ||
521  				   nflen1 != flen))
522  			return -EFSCORRUPTED;
523  #endif
524  	} else {
525  		if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
526  			return error;
527  		if (XFS_IS_CORRUPT(mp, i != 1))
528  			return -EFSCORRUPTED;
529  	}
530  
531  #ifdef DEBUG
532  	if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
533  		struct xfs_btree_block	*bnoblock;
534  		struct xfs_btree_block	*cntblock;
535  
536  		bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_levels[0].bp);
537  		cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_levels[0].bp);
538  
539  		if (XFS_IS_CORRUPT(mp,
540  				   bnoblock->bb_numrecs !=
541  				   cntblock->bb_numrecs))
542  			return -EFSCORRUPTED;
543  	}
544  #endif
545  
546  	/*
547  	 * Deal with all four cases: the allocated record is contained
548  	 * within the freespace record, so we can have new freespace
549  	 * at either (or both) end, or no freespace remaining.
550  	 */
551  	if (rbno == fbno && rlen == flen)
552  		nfbno1 = nfbno2 = NULLAGBLOCK;
553  	else if (rbno == fbno) {
554  		nfbno1 = rbno + rlen;
555  		nflen1 = flen - rlen;
556  		nfbno2 = NULLAGBLOCK;
557  	} else if (rbno + rlen == fbno + flen) {
558  		nfbno1 = fbno;
559  		nflen1 = flen - rlen;
560  		nfbno2 = NULLAGBLOCK;
561  	} else {
562  		nfbno1 = fbno;
563  		nflen1 = rbno - fbno;
564  		nfbno2 = rbno + rlen;
565  		nflen2 = (fbno + flen) - nfbno2;
566  	}
567  	/*
568  	 * Delete the entry from the by-size btree.
569  	 */
570  	if ((error = xfs_btree_delete(cnt_cur, &i)))
571  		return error;
572  	if (XFS_IS_CORRUPT(mp, i != 1))
573  		return -EFSCORRUPTED;
574  	/*
575  	 * Add new by-size btree entry(s).
576  	 */
577  	if (nfbno1 != NULLAGBLOCK) {
578  		if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
579  			return error;
580  		if (XFS_IS_CORRUPT(mp, i != 0))
581  			return -EFSCORRUPTED;
582  		if ((error = xfs_btree_insert(cnt_cur, &i)))
583  			return error;
584  		if (XFS_IS_CORRUPT(mp, i != 1))
585  			return -EFSCORRUPTED;
586  	}
587  	if (nfbno2 != NULLAGBLOCK) {
588  		if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
589  			return error;
590  		if (XFS_IS_CORRUPT(mp, i != 0))
591  			return -EFSCORRUPTED;
592  		if ((error = xfs_btree_insert(cnt_cur, &i)))
593  			return error;
594  		if (XFS_IS_CORRUPT(mp, i != 1))
595  			return -EFSCORRUPTED;
596  	}
597  	/*
598  	 * Fix up the by-block btree entry(s).
599  	 */
600  	if (nfbno1 == NULLAGBLOCK) {
601  		/*
602  		 * No remaining freespace, just delete the by-block tree entry.
603  		 */
604  		if ((error = xfs_btree_delete(bno_cur, &i)))
605  			return error;
606  		if (XFS_IS_CORRUPT(mp, i != 1))
607  			return -EFSCORRUPTED;
608  	} else {
609  		/*
610  		 * Update the by-block entry to start later|be shorter.
611  		 */
612  		if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
613  			return error;
614  	}
615  	if (nfbno2 != NULLAGBLOCK) {
616  		/*
617  		 * 2 resulting free entries, need to add one.
618  		 */
619  		if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
620  			return error;
621  		if (XFS_IS_CORRUPT(mp, i != 0))
622  			return -EFSCORRUPTED;
623  		if ((error = xfs_btree_insert(bno_cur, &i)))
624  			return error;
625  		if (XFS_IS_CORRUPT(mp, i != 1))
626  			return -EFSCORRUPTED;
627  	}
628  	return 0;
629  }
630  
631  /*
632   * We do not verify the AGFL contents against AGF-based index counters here,
633   * even though we may have access to the perag that contains shadow copies. We
634   * don't know if the AGF based counters have been checked, and if they have they
635   * still may be inconsistent because they haven't yet been reset on the first
636   * allocation after the AGF has been read in.
637   *
638   * This means we can only check that all agfl entries contain valid or null
639   * values because we can't reliably determine the active range to exclude
640   * NULLAGBNO as a valid value.
641   *
642   * However, we can't even do that for v4 format filesystems because there are
643   * old versions of mkfs out there that does not initialise the AGFL to known,
644   * verifiable values. HEnce we can't tell the difference between a AGFL block
645   * allocated by mkfs and a corrupted AGFL block here on v4 filesystems.
646   *
647   * As a result, we can only fully validate AGFL block numbers when we pull them
648   * from the freelist in xfs_alloc_get_freelist().
649   */
650  static xfs_failaddr_t
xfs_agfl_verify(struct xfs_buf * bp)651  xfs_agfl_verify(
652  	struct xfs_buf	*bp)
653  {
654  	struct xfs_mount *mp = bp->b_mount;
655  	struct xfs_agfl	*agfl = XFS_BUF_TO_AGFL(bp);
656  	__be32		*agfl_bno = xfs_buf_to_agfl_bno(bp);
657  	int		i;
658  
659  	if (!xfs_has_crc(mp))
660  		return NULL;
661  
662  	if (!xfs_verify_magic(bp, agfl->agfl_magicnum))
663  		return __this_address;
664  	if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid))
665  		return __this_address;
666  	/*
667  	 * during growfs operations, the perag is not fully initialised,
668  	 * so we can't use it for any useful checking. growfs ensures we can't
669  	 * use it by using uncached buffers that don't have the perag attached
670  	 * so we can detect and avoid this problem.
671  	 */
672  	if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno)
673  		return __this_address;
674  
675  	for (i = 0; i < xfs_agfl_size(mp); i++) {
676  		if (be32_to_cpu(agfl_bno[i]) != NULLAGBLOCK &&
677  		    be32_to_cpu(agfl_bno[i]) >= mp->m_sb.sb_agblocks)
678  			return __this_address;
679  	}
680  
681  	if (!xfs_log_check_lsn(mp, be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)))
682  		return __this_address;
683  	return NULL;
684  }
685  
686  static void
xfs_agfl_read_verify(struct xfs_buf * bp)687  xfs_agfl_read_verify(
688  	struct xfs_buf	*bp)
689  {
690  	struct xfs_mount *mp = bp->b_mount;
691  	xfs_failaddr_t	fa;
692  
693  	/*
694  	 * There is no verification of non-crc AGFLs because mkfs does not
695  	 * initialise the AGFL to zero or NULL. Hence the only valid part of the
696  	 * AGFL is what the AGF says is active. We can't get to the AGF, so we
697  	 * can't verify just those entries are valid.
698  	 */
699  	if (!xfs_has_crc(mp))
700  		return;
701  
702  	if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF))
703  		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
704  	else {
705  		fa = xfs_agfl_verify(bp);
706  		if (fa)
707  			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
708  	}
709  }
710  
711  static void
xfs_agfl_write_verify(struct xfs_buf * bp)712  xfs_agfl_write_verify(
713  	struct xfs_buf	*bp)
714  {
715  	struct xfs_mount	*mp = bp->b_mount;
716  	struct xfs_buf_log_item	*bip = bp->b_log_item;
717  	xfs_failaddr_t		fa;
718  
719  	/* no verification of non-crc AGFLs */
720  	if (!xfs_has_crc(mp))
721  		return;
722  
723  	fa = xfs_agfl_verify(bp);
724  	if (fa) {
725  		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
726  		return;
727  	}
728  
729  	if (bip)
730  		XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn);
731  
732  	xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF);
733  }
734  
735  const struct xfs_buf_ops xfs_agfl_buf_ops = {
736  	.name = "xfs_agfl",
737  	.magic = { cpu_to_be32(XFS_AGFL_MAGIC), cpu_to_be32(XFS_AGFL_MAGIC) },
738  	.verify_read = xfs_agfl_read_verify,
739  	.verify_write = xfs_agfl_write_verify,
740  	.verify_struct = xfs_agfl_verify,
741  };
742  
743  /*
744   * Read in the allocation group free block array.
745   */
746  int
xfs_alloc_read_agfl(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf ** bpp)747  xfs_alloc_read_agfl(
748  	struct xfs_perag	*pag,
749  	struct xfs_trans	*tp,
750  	struct xfs_buf		**bpp)
751  {
752  	struct xfs_mount	*mp = pag->pag_mount;
753  	struct xfs_buf		*bp;
754  	int			error;
755  
756  	error = xfs_trans_read_buf(
757  			mp, tp, mp->m_ddev_targp,
758  			XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGFL_DADDR(mp)),
759  			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops);
760  	if (error)
761  		return error;
762  	xfs_buf_set_ref(bp, XFS_AGFL_REF);
763  	*bpp = bp;
764  	return 0;
765  }
766  
767  STATIC int
xfs_alloc_update_counters(struct xfs_trans * tp,struct xfs_buf * agbp,long len)768  xfs_alloc_update_counters(
769  	struct xfs_trans	*tp,
770  	struct xfs_buf		*agbp,
771  	long			len)
772  {
773  	struct xfs_agf		*agf = agbp->b_addr;
774  
775  	agbp->b_pag->pagf_freeblks += len;
776  	be32_add_cpu(&agf->agf_freeblks, len);
777  
778  	if (unlikely(be32_to_cpu(agf->agf_freeblks) >
779  		     be32_to_cpu(agf->agf_length))) {
780  		xfs_buf_mark_corrupt(agbp);
781  		return -EFSCORRUPTED;
782  	}
783  
784  	xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
785  	return 0;
786  }
787  
788  /*
789   * Block allocation algorithm and data structures.
790   */
791  struct xfs_alloc_cur {
792  	struct xfs_btree_cur		*cnt;	/* btree cursors */
793  	struct xfs_btree_cur		*bnolt;
794  	struct xfs_btree_cur		*bnogt;
795  	xfs_extlen_t			cur_len;/* current search length */
796  	xfs_agblock_t			rec_bno;/* extent startblock */
797  	xfs_extlen_t			rec_len;/* extent length */
798  	xfs_agblock_t			bno;	/* alloc bno */
799  	xfs_extlen_t			len;	/* alloc len */
800  	xfs_extlen_t			diff;	/* diff from search bno */
801  	unsigned int			busy_gen;/* busy state */
802  	bool				busy;
803  };
804  
805  /*
806   * Set up cursors, etc. in the extent allocation cursor. This function can be
807   * called multiple times to reset an initialized structure without having to
808   * reallocate cursors.
809   */
810  static int
xfs_alloc_cur_setup(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur)811  xfs_alloc_cur_setup(
812  	struct xfs_alloc_arg	*args,
813  	struct xfs_alloc_cur	*acur)
814  {
815  	int			error;
816  	int			i;
817  
818  	acur->cur_len = args->maxlen;
819  	acur->rec_bno = 0;
820  	acur->rec_len = 0;
821  	acur->bno = 0;
822  	acur->len = 0;
823  	acur->diff = -1;
824  	acur->busy = false;
825  	acur->busy_gen = 0;
826  
827  	/*
828  	 * Perform an initial cntbt lookup to check for availability of maxlen
829  	 * extents. If this fails, we'll return -ENOSPC to signal the caller to
830  	 * attempt a small allocation.
831  	 */
832  	if (!acur->cnt)
833  		acur->cnt = xfs_allocbt_init_cursor(args->mp, args->tp,
834  					args->agbp, args->pag, XFS_BTNUM_CNT);
835  	error = xfs_alloc_lookup_ge(acur->cnt, 0, args->maxlen, &i);
836  	if (error)
837  		return error;
838  
839  	/*
840  	 * Allocate the bnobt left and right search cursors.
841  	 */
842  	if (!acur->bnolt)
843  		acur->bnolt = xfs_allocbt_init_cursor(args->mp, args->tp,
844  					args->agbp, args->pag, XFS_BTNUM_BNO);
845  	if (!acur->bnogt)
846  		acur->bnogt = xfs_allocbt_init_cursor(args->mp, args->tp,
847  					args->agbp, args->pag, XFS_BTNUM_BNO);
848  	return i == 1 ? 0 : -ENOSPC;
849  }
850  
851  static void
xfs_alloc_cur_close(struct xfs_alloc_cur * acur,bool error)852  xfs_alloc_cur_close(
853  	struct xfs_alloc_cur	*acur,
854  	bool			error)
855  {
856  	int			cur_error = XFS_BTREE_NOERROR;
857  
858  	if (error)
859  		cur_error = XFS_BTREE_ERROR;
860  
861  	if (acur->cnt)
862  		xfs_btree_del_cursor(acur->cnt, cur_error);
863  	if (acur->bnolt)
864  		xfs_btree_del_cursor(acur->bnolt, cur_error);
865  	if (acur->bnogt)
866  		xfs_btree_del_cursor(acur->bnogt, cur_error);
867  	acur->cnt = acur->bnolt = acur->bnogt = NULL;
868  }
869  
870  /*
871   * Check an extent for allocation and track the best available candidate in the
872   * allocation structure. The cursor is deactivated if it has entered an out of
873   * range state based on allocation arguments. Optionally return the extent
874   * extent geometry and allocation status if requested by the caller.
875   */
876  static int
xfs_alloc_cur_check(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur,struct xfs_btree_cur * cur,int * new)877  xfs_alloc_cur_check(
878  	struct xfs_alloc_arg	*args,
879  	struct xfs_alloc_cur	*acur,
880  	struct xfs_btree_cur	*cur,
881  	int			*new)
882  {
883  	int			error, i;
884  	xfs_agblock_t		bno, bnoa, bnew;
885  	xfs_extlen_t		len, lena, diff = -1;
886  	bool			busy;
887  	unsigned		busy_gen = 0;
888  	bool			deactivate = false;
889  	bool			isbnobt = cur->bc_btnum == XFS_BTNUM_BNO;
890  
891  	*new = 0;
892  
893  	error = xfs_alloc_get_rec(cur, &bno, &len, &i);
894  	if (error)
895  		return error;
896  	if (XFS_IS_CORRUPT(args->mp, i != 1))
897  		return -EFSCORRUPTED;
898  
899  	/*
900  	 * Check minlen and deactivate a cntbt cursor if out of acceptable size
901  	 * range (i.e., walking backwards looking for a minlen extent).
902  	 */
903  	if (len < args->minlen) {
904  		deactivate = !isbnobt;
905  		goto out;
906  	}
907  
908  	busy = xfs_alloc_compute_aligned(args, bno, len, &bnoa, &lena,
909  					 &busy_gen);
910  	acur->busy |= busy;
911  	if (busy)
912  		acur->busy_gen = busy_gen;
913  	/* deactivate a bnobt cursor outside of locality range */
914  	if (bnoa < args->min_agbno || bnoa > args->max_agbno) {
915  		deactivate = isbnobt;
916  		goto out;
917  	}
918  	if (lena < args->minlen)
919  		goto out;
920  
921  	args->len = XFS_EXTLEN_MIN(lena, args->maxlen);
922  	xfs_alloc_fix_len(args);
923  	ASSERT(args->len >= args->minlen);
924  	if (args->len < acur->len)
925  		goto out;
926  
927  	/*
928  	 * We have an aligned record that satisfies minlen and beats or matches
929  	 * the candidate extent size. Compare locality for near allocation mode.
930  	 */
931  	diff = xfs_alloc_compute_diff(args->agbno, args->len,
932  				      args->alignment, args->datatype,
933  				      bnoa, lena, &bnew);
934  	if (bnew == NULLAGBLOCK)
935  		goto out;
936  
937  	/*
938  	 * Deactivate a bnobt cursor with worse locality than the current best.
939  	 */
940  	if (diff > acur->diff) {
941  		deactivate = isbnobt;
942  		goto out;
943  	}
944  
945  	ASSERT(args->len > acur->len ||
946  	       (args->len == acur->len && diff <= acur->diff));
947  	acur->rec_bno = bno;
948  	acur->rec_len = len;
949  	acur->bno = bnew;
950  	acur->len = args->len;
951  	acur->diff = diff;
952  	*new = 1;
953  
954  	/*
955  	 * We're done if we found a perfect allocation. This only deactivates
956  	 * the current cursor, but this is just an optimization to terminate a
957  	 * cntbt search that otherwise runs to the edge of the tree.
958  	 */
959  	if (acur->diff == 0 && acur->len == args->maxlen)
960  		deactivate = true;
961  out:
962  	if (deactivate)
963  		cur->bc_ag.abt.active = false;
964  	trace_xfs_alloc_cur_check(args->mp, cur->bc_btnum, bno, len, diff,
965  				  *new);
966  	return 0;
967  }
968  
969  /*
970   * Complete an allocation of a candidate extent. Remove the extent from both
971   * trees and update the args structure.
972   */
973  STATIC int
xfs_alloc_cur_finish(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur)974  xfs_alloc_cur_finish(
975  	struct xfs_alloc_arg	*args,
976  	struct xfs_alloc_cur	*acur)
977  {
978  	struct xfs_agf __maybe_unused *agf = args->agbp->b_addr;
979  	int			error;
980  
981  	ASSERT(acur->cnt && acur->bnolt);
982  	ASSERT(acur->bno >= acur->rec_bno);
983  	ASSERT(acur->bno + acur->len <= acur->rec_bno + acur->rec_len);
984  	ASSERT(acur->rec_bno + acur->rec_len <= be32_to_cpu(agf->agf_length));
985  
986  	error = xfs_alloc_fixup_trees(acur->cnt, acur->bnolt, acur->rec_bno,
987  				      acur->rec_len, acur->bno, acur->len, 0);
988  	if (error)
989  		return error;
990  
991  	args->agbno = acur->bno;
992  	args->len = acur->len;
993  	args->wasfromfl = 0;
994  
995  	trace_xfs_alloc_cur(args);
996  	return 0;
997  }
998  
999  /*
1000   * Locality allocation lookup algorithm. This expects a cntbt cursor and uses
1001   * bno optimized lookup to search for extents with ideal size and locality.
1002   */
1003  STATIC int
xfs_alloc_cntbt_iter(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur)1004  xfs_alloc_cntbt_iter(
1005  	struct xfs_alloc_arg		*args,
1006  	struct xfs_alloc_cur		*acur)
1007  {
1008  	struct xfs_btree_cur	*cur = acur->cnt;
1009  	xfs_agblock_t		bno;
1010  	xfs_extlen_t		len, cur_len;
1011  	int			error;
1012  	int			i;
1013  
1014  	if (!xfs_alloc_cur_active(cur))
1015  		return 0;
1016  
1017  	/* locality optimized lookup */
1018  	cur_len = acur->cur_len;
1019  	error = xfs_alloc_lookup_ge(cur, args->agbno, cur_len, &i);
1020  	if (error)
1021  		return error;
1022  	if (i == 0)
1023  		return 0;
1024  	error = xfs_alloc_get_rec(cur, &bno, &len, &i);
1025  	if (error)
1026  		return error;
1027  
1028  	/* check the current record and update search length from it */
1029  	error = xfs_alloc_cur_check(args, acur, cur, &i);
1030  	if (error)
1031  		return error;
1032  	ASSERT(len >= acur->cur_len);
1033  	acur->cur_len = len;
1034  
1035  	/*
1036  	 * We looked up the first record >= [agbno, len] above. The agbno is a
1037  	 * secondary key and so the current record may lie just before or after
1038  	 * agbno. If it is past agbno, check the previous record too so long as
1039  	 * the length matches as it may be closer. Don't check a smaller record
1040  	 * because that could deactivate our cursor.
1041  	 */
1042  	if (bno > args->agbno) {
1043  		error = xfs_btree_decrement(cur, 0, &i);
1044  		if (!error && i) {
1045  			error = xfs_alloc_get_rec(cur, &bno, &len, &i);
1046  			if (!error && i && len == acur->cur_len)
1047  				error = xfs_alloc_cur_check(args, acur, cur,
1048  							    &i);
1049  		}
1050  		if (error)
1051  			return error;
1052  	}
1053  
1054  	/*
1055  	 * Increment the search key until we find at least one allocation
1056  	 * candidate or if the extent we found was larger. Otherwise, double the
1057  	 * search key to optimize the search. Efficiency is more important here
1058  	 * than absolute best locality.
1059  	 */
1060  	cur_len <<= 1;
1061  	if (!acur->len || acur->cur_len >= cur_len)
1062  		acur->cur_len++;
1063  	else
1064  		acur->cur_len = cur_len;
1065  
1066  	return error;
1067  }
1068  
1069  /*
1070   * Deal with the case where only small freespaces remain. Either return the
1071   * contents of the last freespace record, or allocate space from the freelist if
1072   * there is nothing in the tree.
1073   */
1074  STATIC int			/* error */
xfs_alloc_ag_vextent_small(struct xfs_alloc_arg * args,struct xfs_btree_cur * ccur,xfs_agblock_t * fbnop,xfs_extlen_t * flenp,int * stat)1075  xfs_alloc_ag_vextent_small(
1076  	struct xfs_alloc_arg	*args,	/* allocation argument structure */
1077  	struct xfs_btree_cur	*ccur,	/* optional by-size cursor */
1078  	xfs_agblock_t		*fbnop,	/* result block number */
1079  	xfs_extlen_t		*flenp,	/* result length */
1080  	int			*stat)	/* status: 0-freelist, 1-normal/none */
1081  {
1082  	struct xfs_agf		*agf = args->agbp->b_addr;
1083  	int			error = 0;
1084  	xfs_agblock_t		fbno = NULLAGBLOCK;
1085  	xfs_extlen_t		flen = 0;
1086  	int			i = 0;
1087  
1088  	/*
1089  	 * If a cntbt cursor is provided, try to allocate the largest record in
1090  	 * the tree. Try the AGFL if the cntbt is empty, otherwise fail the
1091  	 * allocation. Make sure to respect minleft even when pulling from the
1092  	 * freelist.
1093  	 */
1094  	if (ccur)
1095  		error = xfs_btree_decrement(ccur, 0, &i);
1096  	if (error)
1097  		goto error;
1098  	if (i) {
1099  		error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i);
1100  		if (error)
1101  			goto error;
1102  		if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1103  			error = -EFSCORRUPTED;
1104  			goto error;
1105  		}
1106  		goto out;
1107  	}
1108  
1109  	if (args->minlen != 1 || args->alignment != 1 ||
1110  	    args->resv == XFS_AG_RESV_AGFL ||
1111  	    be32_to_cpu(agf->agf_flcount) <= args->minleft)
1112  		goto out;
1113  
1114  	error = xfs_alloc_get_freelist(args->pag, args->tp, args->agbp,
1115  			&fbno, 0);
1116  	if (error)
1117  		goto error;
1118  	if (fbno == NULLAGBLOCK)
1119  		goto out;
1120  
1121  	xfs_extent_busy_reuse(args->mp, args->pag, fbno, 1,
1122  			      (args->datatype & XFS_ALLOC_NOBUSY));
1123  
1124  	if (args->datatype & XFS_ALLOC_USERDATA) {
1125  		struct xfs_buf	*bp;
1126  
1127  		error = xfs_trans_get_buf(args->tp, args->mp->m_ddev_targp,
1128  				XFS_AGB_TO_DADDR(args->mp, args->agno, fbno),
1129  				args->mp->m_bsize, 0, &bp);
1130  		if (error)
1131  			goto error;
1132  		xfs_trans_binval(args->tp, bp);
1133  	}
1134  	*fbnop = args->agbno = fbno;
1135  	*flenp = args->len = 1;
1136  	if (XFS_IS_CORRUPT(args->mp, fbno >= be32_to_cpu(agf->agf_length))) {
1137  		error = -EFSCORRUPTED;
1138  		goto error;
1139  	}
1140  	args->wasfromfl = 1;
1141  	trace_xfs_alloc_small_freelist(args);
1142  
1143  	/*
1144  	 * If we're feeding an AGFL block to something that doesn't live in the
1145  	 * free space, we need to clear out the OWN_AG rmap.
1146  	 */
1147  	error = xfs_rmap_free(args->tp, args->agbp, args->pag, fbno, 1,
1148  			      &XFS_RMAP_OINFO_AG);
1149  	if (error)
1150  		goto error;
1151  
1152  	*stat = 0;
1153  	return 0;
1154  
1155  out:
1156  	/*
1157  	 * Can't do the allocation, give up.
1158  	 */
1159  	if (flen < args->minlen) {
1160  		args->agbno = NULLAGBLOCK;
1161  		trace_xfs_alloc_small_notenough(args);
1162  		flen = 0;
1163  	}
1164  	*fbnop = fbno;
1165  	*flenp = flen;
1166  	*stat = 1;
1167  	trace_xfs_alloc_small_done(args);
1168  	return 0;
1169  
1170  error:
1171  	trace_xfs_alloc_small_error(args);
1172  	return error;
1173  }
1174  
1175  /*
1176   * Allocate a variable extent at exactly agno/bno.
1177   * Extent's length (returned in *len) will be between minlen and maxlen,
1178   * and of the form k * prod + mod unless there's nothing that large.
1179   * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
1180   */
1181  STATIC int			/* error */
xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t * args)1182  xfs_alloc_ag_vextent_exact(
1183  	xfs_alloc_arg_t	*args)	/* allocation argument structure */
1184  {
1185  	struct xfs_agf __maybe_unused *agf = args->agbp->b_addr;
1186  	struct xfs_btree_cur *bno_cur;/* by block-number btree cursor */
1187  	struct xfs_btree_cur *cnt_cur;/* by count btree cursor */
1188  	int		error;
1189  	xfs_agblock_t	fbno;	/* start block of found extent */
1190  	xfs_extlen_t	flen;	/* length of found extent */
1191  	xfs_agblock_t	tbno;	/* start block of busy extent */
1192  	xfs_extlen_t	tlen;	/* length of busy extent */
1193  	xfs_agblock_t	tend;	/* end block of busy extent */
1194  	int		i;	/* success/failure of operation */
1195  	unsigned	busy_gen;
1196  
1197  	ASSERT(args->alignment == 1);
1198  
1199  	/*
1200  	 * Allocate/initialize a cursor for the by-number freespace btree.
1201  	 */
1202  	bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1203  					  args->pag, XFS_BTNUM_BNO);
1204  
1205  	/*
1206  	 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
1207  	 * Look for the closest free block <= bno, it must contain bno
1208  	 * if any free block does.
1209  	 */
1210  	error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
1211  	if (error)
1212  		goto error0;
1213  	if (!i)
1214  		goto not_found;
1215  
1216  	/*
1217  	 * Grab the freespace record.
1218  	 */
1219  	error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
1220  	if (error)
1221  		goto error0;
1222  	if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1223  		error = -EFSCORRUPTED;
1224  		goto error0;
1225  	}
1226  	ASSERT(fbno <= args->agbno);
1227  
1228  	/*
1229  	 * Check for overlapping busy extents.
1230  	 */
1231  	tbno = fbno;
1232  	tlen = flen;
1233  	xfs_extent_busy_trim(args, &tbno, &tlen, &busy_gen);
1234  
1235  	/*
1236  	 * Give up if the start of the extent is busy, or the freespace isn't
1237  	 * long enough for the minimum request.
1238  	 */
1239  	if (tbno > args->agbno)
1240  		goto not_found;
1241  	if (tlen < args->minlen)
1242  		goto not_found;
1243  	tend = tbno + tlen;
1244  	if (tend < args->agbno + args->minlen)
1245  		goto not_found;
1246  
1247  	/*
1248  	 * End of extent will be smaller of the freespace end and the
1249  	 * maximal requested end.
1250  	 *
1251  	 * Fix the length according to mod and prod if given.
1252  	 */
1253  	args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
1254  						- args->agbno;
1255  	xfs_alloc_fix_len(args);
1256  	ASSERT(args->agbno + args->len <= tend);
1257  
1258  	/*
1259  	 * We are allocating agbno for args->len
1260  	 * Allocate/initialize a cursor for the by-size btree.
1261  	 */
1262  	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1263  					args->pag, XFS_BTNUM_CNT);
1264  	ASSERT(args->agbno + args->len <= be32_to_cpu(agf->agf_length));
1265  	error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
1266  				      args->len, XFSA_FIXUP_BNO_OK);
1267  	if (error) {
1268  		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1269  		goto error0;
1270  	}
1271  
1272  	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1273  	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1274  
1275  	args->wasfromfl = 0;
1276  	trace_xfs_alloc_exact_done(args);
1277  	return 0;
1278  
1279  not_found:
1280  	/* Didn't find it, return null. */
1281  	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1282  	args->agbno = NULLAGBLOCK;
1283  	trace_xfs_alloc_exact_notfound(args);
1284  	return 0;
1285  
1286  error0:
1287  	xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1288  	trace_xfs_alloc_exact_error(args);
1289  	return error;
1290  }
1291  
1292  /*
1293   * Search a given number of btree records in a given direction. Check each
1294   * record against the good extent we've already found.
1295   */
1296  STATIC int
xfs_alloc_walk_iter(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur,struct xfs_btree_cur * cur,bool increment,bool find_one,int count,int * stat)1297  xfs_alloc_walk_iter(
1298  	struct xfs_alloc_arg	*args,
1299  	struct xfs_alloc_cur	*acur,
1300  	struct xfs_btree_cur	*cur,
1301  	bool			increment,
1302  	bool			find_one, /* quit on first candidate */
1303  	int			count,    /* rec count (-1 for infinite) */
1304  	int			*stat)
1305  {
1306  	int			error;
1307  	int			i;
1308  
1309  	*stat = 0;
1310  
1311  	/*
1312  	 * Search so long as the cursor is active or we find a better extent.
1313  	 * The cursor is deactivated if it extends beyond the range of the
1314  	 * current allocation candidate.
1315  	 */
1316  	while (xfs_alloc_cur_active(cur) && count) {
1317  		error = xfs_alloc_cur_check(args, acur, cur, &i);
1318  		if (error)
1319  			return error;
1320  		if (i == 1) {
1321  			*stat = 1;
1322  			if (find_one)
1323  				break;
1324  		}
1325  		if (!xfs_alloc_cur_active(cur))
1326  			break;
1327  
1328  		if (increment)
1329  			error = xfs_btree_increment(cur, 0, &i);
1330  		else
1331  			error = xfs_btree_decrement(cur, 0, &i);
1332  		if (error)
1333  			return error;
1334  		if (i == 0)
1335  			cur->bc_ag.abt.active = false;
1336  
1337  		if (count > 0)
1338  			count--;
1339  	}
1340  
1341  	return 0;
1342  }
1343  
1344  /*
1345   * Search the by-bno and by-size btrees in parallel in search of an extent with
1346   * ideal locality based on the NEAR mode ->agbno locality hint.
1347   */
1348  STATIC int
xfs_alloc_ag_vextent_locality(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur,int * stat)1349  xfs_alloc_ag_vextent_locality(
1350  	struct xfs_alloc_arg	*args,
1351  	struct xfs_alloc_cur	*acur,
1352  	int			*stat)
1353  {
1354  	struct xfs_btree_cur	*fbcur = NULL;
1355  	int			error;
1356  	int			i;
1357  	bool			fbinc;
1358  
1359  	ASSERT(acur->len == 0);
1360  
1361  	*stat = 0;
1362  
1363  	error = xfs_alloc_lookup_ge(acur->cnt, args->agbno, acur->cur_len, &i);
1364  	if (error)
1365  		return error;
1366  	error = xfs_alloc_lookup_le(acur->bnolt, args->agbno, 0, &i);
1367  	if (error)
1368  		return error;
1369  	error = xfs_alloc_lookup_ge(acur->bnogt, args->agbno, 0, &i);
1370  	if (error)
1371  		return error;
1372  
1373  	/*
1374  	 * Search the bnobt and cntbt in parallel. Search the bnobt left and
1375  	 * right and lookup the closest extent to the locality hint for each
1376  	 * extent size key in the cntbt. The entire search terminates
1377  	 * immediately on a bnobt hit because that means we've found best case
1378  	 * locality. Otherwise the search continues until the cntbt cursor runs
1379  	 * off the end of the tree. If no allocation candidate is found at this
1380  	 * point, give up on locality, walk backwards from the end of the cntbt
1381  	 * and take the first available extent.
1382  	 *
1383  	 * The parallel tree searches balance each other out to provide fairly
1384  	 * consistent performance for various situations. The bnobt search can
1385  	 * have pathological behavior in the worst case scenario of larger
1386  	 * allocation requests and fragmented free space. On the other hand, the
1387  	 * bnobt is able to satisfy most smaller allocation requests much more
1388  	 * quickly than the cntbt. The cntbt search can sift through fragmented
1389  	 * free space and sets of free extents for larger allocation requests
1390  	 * more quickly than the bnobt. Since the locality hint is just a hint
1391  	 * and we don't want to scan the entire bnobt for perfect locality, the
1392  	 * cntbt search essentially bounds the bnobt search such that we can
1393  	 * find good enough locality at reasonable performance in most cases.
1394  	 */
1395  	while (xfs_alloc_cur_active(acur->bnolt) ||
1396  	       xfs_alloc_cur_active(acur->bnogt) ||
1397  	       xfs_alloc_cur_active(acur->cnt)) {
1398  
1399  		trace_xfs_alloc_cur_lookup(args);
1400  
1401  		/*
1402  		 * Search the bnobt left and right. In the case of a hit, finish
1403  		 * the search in the opposite direction and we're done.
1404  		 */
1405  		error = xfs_alloc_walk_iter(args, acur, acur->bnolt, false,
1406  					    true, 1, &i);
1407  		if (error)
1408  			return error;
1409  		if (i == 1) {
1410  			trace_xfs_alloc_cur_left(args);
1411  			fbcur = acur->bnogt;
1412  			fbinc = true;
1413  			break;
1414  		}
1415  		error = xfs_alloc_walk_iter(args, acur, acur->bnogt, true, true,
1416  					    1, &i);
1417  		if (error)
1418  			return error;
1419  		if (i == 1) {
1420  			trace_xfs_alloc_cur_right(args);
1421  			fbcur = acur->bnolt;
1422  			fbinc = false;
1423  			break;
1424  		}
1425  
1426  		/*
1427  		 * Check the extent with best locality based on the current
1428  		 * extent size search key and keep track of the best candidate.
1429  		 */
1430  		error = xfs_alloc_cntbt_iter(args, acur);
1431  		if (error)
1432  			return error;
1433  		if (!xfs_alloc_cur_active(acur->cnt)) {
1434  			trace_xfs_alloc_cur_lookup_done(args);
1435  			break;
1436  		}
1437  	}
1438  
1439  	/*
1440  	 * If we failed to find anything due to busy extents, return empty
1441  	 * handed so the caller can flush and retry. If no busy extents were
1442  	 * found, walk backwards from the end of the cntbt as a last resort.
1443  	 */
1444  	if (!xfs_alloc_cur_active(acur->cnt) && !acur->len && !acur->busy) {
1445  		error = xfs_btree_decrement(acur->cnt, 0, &i);
1446  		if (error)
1447  			return error;
1448  		if (i) {
1449  			acur->cnt->bc_ag.abt.active = true;
1450  			fbcur = acur->cnt;
1451  			fbinc = false;
1452  		}
1453  	}
1454  
1455  	/*
1456  	 * Search in the opposite direction for a better entry in the case of
1457  	 * a bnobt hit or walk backwards from the end of the cntbt.
1458  	 */
1459  	if (fbcur) {
1460  		error = xfs_alloc_walk_iter(args, acur, fbcur, fbinc, true, -1,
1461  					    &i);
1462  		if (error)
1463  			return error;
1464  	}
1465  
1466  	if (acur->len)
1467  		*stat = 1;
1468  
1469  	return 0;
1470  }
1471  
1472  /* Check the last block of the cnt btree for allocations. */
1473  static int
xfs_alloc_ag_vextent_lastblock(struct xfs_alloc_arg * args,struct xfs_alloc_cur * acur,xfs_agblock_t * bno,xfs_extlen_t * len,bool * allocated)1474  xfs_alloc_ag_vextent_lastblock(
1475  	struct xfs_alloc_arg	*args,
1476  	struct xfs_alloc_cur	*acur,
1477  	xfs_agblock_t		*bno,
1478  	xfs_extlen_t		*len,
1479  	bool			*allocated)
1480  {
1481  	int			error;
1482  	int			i;
1483  
1484  #ifdef DEBUG
1485  	/* Randomly don't execute the first algorithm. */
1486  	if (get_random_u32_below(2))
1487  		return 0;
1488  #endif
1489  
1490  	/*
1491  	 * Start from the entry that lookup found, sequence through all larger
1492  	 * free blocks.  If we're actually pointing at a record smaller than
1493  	 * maxlen, go to the start of this block, and skip all those smaller
1494  	 * than minlen.
1495  	 */
1496  	if (*len || args->alignment > 1) {
1497  		acur->cnt->bc_levels[0].ptr = 1;
1498  		do {
1499  			error = xfs_alloc_get_rec(acur->cnt, bno, len, &i);
1500  			if (error)
1501  				return error;
1502  			if (XFS_IS_CORRUPT(args->mp, i != 1))
1503  				return -EFSCORRUPTED;
1504  			if (*len >= args->minlen)
1505  				break;
1506  			error = xfs_btree_increment(acur->cnt, 0, &i);
1507  			if (error)
1508  				return error;
1509  		} while (i);
1510  		ASSERT(*len >= args->minlen);
1511  		if (!i)
1512  			return 0;
1513  	}
1514  
1515  	error = xfs_alloc_walk_iter(args, acur, acur->cnt, true, false, -1, &i);
1516  	if (error)
1517  		return error;
1518  
1519  	/*
1520  	 * It didn't work.  We COULD be in a case where there's a good record
1521  	 * somewhere, so try again.
1522  	 */
1523  	if (acur->len == 0)
1524  		return 0;
1525  
1526  	trace_xfs_alloc_near_first(args);
1527  	*allocated = true;
1528  	return 0;
1529  }
1530  
1531  /*
1532   * Allocate a variable extent near bno in the allocation group agno.
1533   * Extent's length (returned in len) will be between minlen and maxlen,
1534   * and of the form k * prod + mod unless there's nothing that large.
1535   * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1536   */
1537  STATIC int
xfs_alloc_ag_vextent_near(struct xfs_alloc_arg * args,uint32_t alloc_flags)1538  xfs_alloc_ag_vextent_near(
1539  	struct xfs_alloc_arg	*args,
1540  	uint32_t		alloc_flags)
1541  {
1542  	struct xfs_alloc_cur	acur = {};
1543  	int			error;		/* error code */
1544  	int			i;		/* result code, temporary */
1545  	xfs_agblock_t		bno;
1546  	xfs_extlen_t		len;
1547  
1548  	/* handle uninitialized agbno range so caller doesn't have to */
1549  	if (!args->min_agbno && !args->max_agbno)
1550  		args->max_agbno = args->mp->m_sb.sb_agblocks - 1;
1551  	ASSERT(args->min_agbno <= args->max_agbno);
1552  
1553  	/* clamp agbno to the range if it's outside */
1554  	if (args->agbno < args->min_agbno)
1555  		args->agbno = args->min_agbno;
1556  	if (args->agbno > args->max_agbno)
1557  		args->agbno = args->max_agbno;
1558  
1559  	/* Retry once quickly if we find busy extents before blocking. */
1560  	alloc_flags |= XFS_ALLOC_FLAG_TRYFLUSH;
1561  restart:
1562  	len = 0;
1563  
1564  	/*
1565  	 * Set up cursors and see if there are any free extents as big as
1566  	 * maxlen. If not, pick the last entry in the tree unless the tree is
1567  	 * empty.
1568  	 */
1569  	error = xfs_alloc_cur_setup(args, &acur);
1570  	if (error == -ENOSPC) {
1571  		error = xfs_alloc_ag_vextent_small(args, acur.cnt, &bno,
1572  				&len, &i);
1573  		if (error)
1574  			goto out;
1575  		if (i == 0 || len == 0) {
1576  			trace_xfs_alloc_near_noentry(args);
1577  			goto out;
1578  		}
1579  		ASSERT(i == 1);
1580  	} else if (error) {
1581  		goto out;
1582  	}
1583  
1584  	/*
1585  	 * First algorithm.
1586  	 * If the requested extent is large wrt the freespaces available
1587  	 * in this a.g., then the cursor will be pointing to a btree entry
1588  	 * near the right edge of the tree.  If it's in the last btree leaf
1589  	 * block, then we just examine all the entries in that block
1590  	 * that are big enough, and pick the best one.
1591  	 */
1592  	if (xfs_btree_islastblock(acur.cnt, 0)) {
1593  		bool		allocated = false;
1594  
1595  		error = xfs_alloc_ag_vextent_lastblock(args, &acur, &bno, &len,
1596  				&allocated);
1597  		if (error)
1598  			goto out;
1599  		if (allocated)
1600  			goto alloc_finish;
1601  	}
1602  
1603  	/*
1604  	 * Second algorithm. Combined cntbt and bnobt search to find ideal
1605  	 * locality.
1606  	 */
1607  	error = xfs_alloc_ag_vextent_locality(args, &acur, &i);
1608  	if (error)
1609  		goto out;
1610  
1611  	/*
1612  	 * If we couldn't get anything, give up.
1613  	 */
1614  	if (!acur.len) {
1615  		if (acur.busy) {
1616  			/*
1617  			 * Our only valid extents must have been busy. Flush and
1618  			 * retry the allocation again. If we get an -EAGAIN
1619  			 * error, we're being told that a deadlock was avoided
1620  			 * and the current transaction needs committing before
1621  			 * the allocation can be retried.
1622  			 */
1623  			trace_xfs_alloc_near_busy(args);
1624  			error = xfs_extent_busy_flush(args->tp, args->pag,
1625  					acur.busy_gen, alloc_flags);
1626  			if (error)
1627  				goto out;
1628  
1629  			alloc_flags &= ~XFS_ALLOC_FLAG_TRYFLUSH;
1630  			goto restart;
1631  		}
1632  		trace_xfs_alloc_size_neither(args);
1633  		args->agbno = NULLAGBLOCK;
1634  		goto out;
1635  	}
1636  
1637  alloc_finish:
1638  	/* fix up btrees on a successful allocation */
1639  	error = xfs_alloc_cur_finish(args, &acur);
1640  
1641  out:
1642  	xfs_alloc_cur_close(&acur, error);
1643  	return error;
1644  }
1645  
1646  /*
1647   * Allocate a variable extent anywhere in the allocation group agno.
1648   * Extent's length (returned in len) will be between minlen and maxlen,
1649   * and of the form k * prod + mod unless there's nothing that large.
1650   * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1651   */
1652  static int
xfs_alloc_ag_vextent_size(struct xfs_alloc_arg * args,uint32_t alloc_flags)1653  xfs_alloc_ag_vextent_size(
1654  	struct xfs_alloc_arg	*args,
1655  	uint32_t		alloc_flags)
1656  {
1657  	struct xfs_agf		*agf = args->agbp->b_addr;
1658  	struct xfs_btree_cur	*bno_cur;
1659  	struct xfs_btree_cur	*cnt_cur;
1660  	xfs_agblock_t		fbno;		/* start of found freespace */
1661  	xfs_extlen_t		flen;		/* length of found freespace */
1662  	xfs_agblock_t		rbno;		/* returned block number */
1663  	xfs_extlen_t		rlen;		/* length of returned extent */
1664  	bool			busy;
1665  	unsigned		busy_gen;
1666  	int			error;
1667  	int			i;
1668  
1669  	/* Retry once quickly if we find busy extents before blocking. */
1670  	alloc_flags |= XFS_ALLOC_FLAG_TRYFLUSH;
1671  restart:
1672  	/*
1673  	 * Allocate and initialize a cursor for the by-size btree.
1674  	 */
1675  	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1676  					args->pag, XFS_BTNUM_CNT);
1677  	bno_cur = NULL;
1678  
1679  	/*
1680  	 * Look for an entry >= maxlen+alignment-1 blocks.
1681  	 */
1682  	if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1683  			args->maxlen + args->alignment - 1, &i)))
1684  		goto error0;
1685  
1686  	/*
1687  	 * If none then we have to settle for a smaller extent. In the case that
1688  	 * there are no large extents, this will return the last entry in the
1689  	 * tree unless the tree is empty. In the case that there are only busy
1690  	 * large extents, this will return the largest small extent unless there
1691  	 * are no smaller extents available.
1692  	 */
1693  	if (!i) {
1694  		error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1695  						   &fbno, &flen, &i);
1696  		if (error)
1697  			goto error0;
1698  		if (i == 0 || flen == 0) {
1699  			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1700  			trace_xfs_alloc_size_noentry(args);
1701  			return 0;
1702  		}
1703  		ASSERT(i == 1);
1704  		busy = xfs_alloc_compute_aligned(args, fbno, flen, &rbno,
1705  				&rlen, &busy_gen);
1706  	} else {
1707  		/*
1708  		 * Search for a non-busy extent that is large enough.
1709  		 */
1710  		for (;;) {
1711  			error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1712  			if (error)
1713  				goto error0;
1714  			if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1715  				error = -EFSCORRUPTED;
1716  				goto error0;
1717  			}
1718  
1719  			busy = xfs_alloc_compute_aligned(args, fbno, flen,
1720  					&rbno, &rlen, &busy_gen);
1721  
1722  			if (rlen >= args->maxlen)
1723  				break;
1724  
1725  			error = xfs_btree_increment(cnt_cur, 0, &i);
1726  			if (error)
1727  				goto error0;
1728  			if (i)
1729  				continue;
1730  
1731  			/*
1732  			 * Our only valid extents must have been busy. Flush and
1733  			 * retry the allocation again. If we get an -EAGAIN
1734  			 * error, we're being told that a deadlock was avoided
1735  			 * and the current transaction needs committing before
1736  			 * the allocation can be retried.
1737  			 */
1738  			trace_xfs_alloc_size_busy(args);
1739  			error = xfs_extent_busy_flush(args->tp, args->pag,
1740  					busy_gen, alloc_flags);
1741  			if (error)
1742  				goto error0;
1743  
1744  			alloc_flags &= ~XFS_ALLOC_FLAG_TRYFLUSH;
1745  			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1746  			goto restart;
1747  		}
1748  	}
1749  
1750  	/*
1751  	 * In the first case above, we got the last entry in the
1752  	 * by-size btree.  Now we check to see if the space hits maxlen
1753  	 * once aligned; if not, we search left for something better.
1754  	 * This can't happen in the second case above.
1755  	 */
1756  	rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1757  	if (XFS_IS_CORRUPT(args->mp,
1758  			   rlen != 0 &&
1759  			   (rlen > flen ||
1760  			    rbno + rlen > fbno + flen))) {
1761  		error = -EFSCORRUPTED;
1762  		goto error0;
1763  	}
1764  	if (rlen < args->maxlen) {
1765  		xfs_agblock_t	bestfbno;
1766  		xfs_extlen_t	bestflen;
1767  		xfs_agblock_t	bestrbno;
1768  		xfs_extlen_t	bestrlen;
1769  
1770  		bestrlen = rlen;
1771  		bestrbno = rbno;
1772  		bestflen = flen;
1773  		bestfbno = fbno;
1774  		for (;;) {
1775  			if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1776  				goto error0;
1777  			if (i == 0)
1778  				break;
1779  			if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1780  					&i)))
1781  				goto error0;
1782  			if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1783  				error = -EFSCORRUPTED;
1784  				goto error0;
1785  			}
1786  			if (flen <= bestrlen)
1787  				break;
1788  			busy = xfs_alloc_compute_aligned(args, fbno, flen,
1789  					&rbno, &rlen, &busy_gen);
1790  			rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1791  			if (XFS_IS_CORRUPT(args->mp,
1792  					   rlen != 0 &&
1793  					   (rlen > flen ||
1794  					    rbno + rlen > fbno + flen))) {
1795  				error = -EFSCORRUPTED;
1796  				goto error0;
1797  			}
1798  			if (rlen > bestrlen) {
1799  				bestrlen = rlen;
1800  				bestrbno = rbno;
1801  				bestflen = flen;
1802  				bestfbno = fbno;
1803  				if (rlen == args->maxlen)
1804  					break;
1805  			}
1806  		}
1807  		if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1808  				&i)))
1809  			goto error0;
1810  		if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1811  			error = -EFSCORRUPTED;
1812  			goto error0;
1813  		}
1814  		rlen = bestrlen;
1815  		rbno = bestrbno;
1816  		flen = bestflen;
1817  		fbno = bestfbno;
1818  	}
1819  	args->wasfromfl = 0;
1820  	/*
1821  	 * Fix up the length.
1822  	 */
1823  	args->len = rlen;
1824  	if (rlen < args->minlen) {
1825  		if (busy) {
1826  			/*
1827  			 * Our only valid extents must have been busy. Flush and
1828  			 * retry the allocation again. If we get an -EAGAIN
1829  			 * error, we're being told that a deadlock was avoided
1830  			 * and the current transaction needs committing before
1831  			 * the allocation can be retried.
1832  			 */
1833  			trace_xfs_alloc_size_busy(args);
1834  			error = xfs_extent_busy_flush(args->tp, args->pag,
1835  					busy_gen, alloc_flags);
1836  			if (error)
1837  				goto error0;
1838  
1839  			alloc_flags &= ~XFS_ALLOC_FLAG_TRYFLUSH;
1840  			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1841  			goto restart;
1842  		}
1843  		goto out_nominleft;
1844  	}
1845  	xfs_alloc_fix_len(args);
1846  
1847  	rlen = args->len;
1848  	if (XFS_IS_CORRUPT(args->mp, rlen > flen)) {
1849  		error = -EFSCORRUPTED;
1850  		goto error0;
1851  	}
1852  	/*
1853  	 * Allocate and initialize a cursor for the by-block tree.
1854  	 */
1855  	bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1856  					args->pag, XFS_BTNUM_BNO);
1857  	if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1858  			rbno, rlen, XFSA_FIXUP_CNT_OK)))
1859  		goto error0;
1860  	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1861  	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1862  	cnt_cur = bno_cur = NULL;
1863  	args->len = rlen;
1864  	args->agbno = rbno;
1865  	if (XFS_IS_CORRUPT(args->mp,
1866  			   args->agbno + args->len >
1867  			   be32_to_cpu(agf->agf_length))) {
1868  		error = -EFSCORRUPTED;
1869  		goto error0;
1870  	}
1871  	trace_xfs_alloc_size_done(args);
1872  	return 0;
1873  
1874  error0:
1875  	trace_xfs_alloc_size_error(args);
1876  	if (cnt_cur)
1877  		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1878  	if (bno_cur)
1879  		xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1880  	return error;
1881  
1882  out_nominleft:
1883  	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1884  	trace_xfs_alloc_size_nominleft(args);
1885  	args->agbno = NULLAGBLOCK;
1886  	return 0;
1887  }
1888  
1889  /*
1890   * Free the extent starting at agno/bno for length.
1891   */
1892  STATIC int
xfs_free_ag_extent(struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agnumber_t agno,xfs_agblock_t bno,xfs_extlen_t len,const struct xfs_owner_info * oinfo,enum xfs_ag_resv_type type)1893  xfs_free_ag_extent(
1894  	struct xfs_trans		*tp,
1895  	struct xfs_buf			*agbp,
1896  	xfs_agnumber_t			agno,
1897  	xfs_agblock_t			bno,
1898  	xfs_extlen_t			len,
1899  	const struct xfs_owner_info	*oinfo,
1900  	enum xfs_ag_resv_type		type)
1901  {
1902  	struct xfs_mount		*mp;
1903  	struct xfs_btree_cur		*bno_cur;
1904  	struct xfs_btree_cur		*cnt_cur;
1905  	xfs_agblock_t			gtbno; /* start of right neighbor */
1906  	xfs_extlen_t			gtlen; /* length of right neighbor */
1907  	xfs_agblock_t			ltbno; /* start of left neighbor */
1908  	xfs_extlen_t			ltlen; /* length of left neighbor */
1909  	xfs_agblock_t			nbno; /* new starting block of freesp */
1910  	xfs_extlen_t			nlen; /* new length of freespace */
1911  	int				haveleft; /* have a left neighbor */
1912  	int				haveright; /* have a right neighbor */
1913  	int				i;
1914  	int				error;
1915  	struct xfs_perag		*pag = agbp->b_pag;
1916  
1917  	bno_cur = cnt_cur = NULL;
1918  	mp = tp->t_mountp;
1919  
1920  	if (!xfs_rmap_should_skip_owner_update(oinfo)) {
1921  		error = xfs_rmap_free(tp, agbp, pag, bno, len, oinfo);
1922  		if (error)
1923  			goto error0;
1924  	}
1925  
1926  	/*
1927  	 * Allocate and initialize a cursor for the by-block btree.
1928  	 */
1929  	bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_BNO);
1930  	/*
1931  	 * Look for a neighboring block on the left (lower block numbers)
1932  	 * that is contiguous with this space.
1933  	 */
1934  	if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1935  		goto error0;
1936  	if (haveleft) {
1937  		/*
1938  		 * There is a block to our left.
1939  		 */
1940  		if ((error = xfs_alloc_get_rec(bno_cur, &ltbno, &ltlen, &i)))
1941  			goto error0;
1942  		if (XFS_IS_CORRUPT(mp, i != 1)) {
1943  			error = -EFSCORRUPTED;
1944  			goto error0;
1945  		}
1946  		/*
1947  		 * It's not contiguous, though.
1948  		 */
1949  		if (ltbno + ltlen < bno)
1950  			haveleft = 0;
1951  		else {
1952  			/*
1953  			 * If this failure happens the request to free this
1954  			 * space was invalid, it's (partly) already free.
1955  			 * Very bad.
1956  			 */
1957  			if (XFS_IS_CORRUPT(mp, ltbno + ltlen > bno)) {
1958  				error = -EFSCORRUPTED;
1959  				goto error0;
1960  			}
1961  		}
1962  	}
1963  	/*
1964  	 * Look for a neighboring block on the right (higher block numbers)
1965  	 * that is contiguous with this space.
1966  	 */
1967  	if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1968  		goto error0;
1969  	if (haveright) {
1970  		/*
1971  		 * There is a block to our right.
1972  		 */
1973  		if ((error = xfs_alloc_get_rec(bno_cur, &gtbno, &gtlen, &i)))
1974  			goto error0;
1975  		if (XFS_IS_CORRUPT(mp, i != 1)) {
1976  			error = -EFSCORRUPTED;
1977  			goto error0;
1978  		}
1979  		/*
1980  		 * It's not contiguous, though.
1981  		 */
1982  		if (bno + len < gtbno)
1983  			haveright = 0;
1984  		else {
1985  			/*
1986  			 * If this failure happens the request to free this
1987  			 * space was invalid, it's (partly) already free.
1988  			 * Very bad.
1989  			 */
1990  			if (XFS_IS_CORRUPT(mp, bno + len > gtbno)) {
1991  				error = -EFSCORRUPTED;
1992  				goto error0;
1993  			}
1994  		}
1995  	}
1996  	/*
1997  	 * Now allocate and initialize a cursor for the by-size tree.
1998  	 */
1999  	cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_CNT);
2000  	/*
2001  	 * Have both left and right contiguous neighbors.
2002  	 * Merge all three into a single free block.
2003  	 */
2004  	if (haveleft && haveright) {
2005  		/*
2006  		 * Delete the old by-size entry on the left.
2007  		 */
2008  		if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
2009  			goto error0;
2010  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2011  			error = -EFSCORRUPTED;
2012  			goto error0;
2013  		}
2014  		if ((error = xfs_btree_delete(cnt_cur, &i)))
2015  			goto error0;
2016  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2017  			error = -EFSCORRUPTED;
2018  			goto error0;
2019  		}
2020  		/*
2021  		 * Delete the old by-size entry on the right.
2022  		 */
2023  		if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
2024  			goto error0;
2025  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2026  			error = -EFSCORRUPTED;
2027  			goto error0;
2028  		}
2029  		if ((error = xfs_btree_delete(cnt_cur, &i)))
2030  			goto error0;
2031  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2032  			error = -EFSCORRUPTED;
2033  			goto error0;
2034  		}
2035  		/*
2036  		 * Delete the old by-block entry for the right block.
2037  		 */
2038  		if ((error = xfs_btree_delete(bno_cur, &i)))
2039  			goto error0;
2040  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2041  			error = -EFSCORRUPTED;
2042  			goto error0;
2043  		}
2044  		/*
2045  		 * Move the by-block cursor back to the left neighbor.
2046  		 */
2047  		if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
2048  			goto error0;
2049  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2050  			error = -EFSCORRUPTED;
2051  			goto error0;
2052  		}
2053  #ifdef DEBUG
2054  		/*
2055  		 * Check that this is the right record: delete didn't
2056  		 * mangle the cursor.
2057  		 */
2058  		{
2059  			xfs_agblock_t	xxbno;
2060  			xfs_extlen_t	xxlen;
2061  
2062  			if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
2063  					&i)))
2064  				goto error0;
2065  			if (XFS_IS_CORRUPT(mp,
2066  					   i != 1 ||
2067  					   xxbno != ltbno ||
2068  					   xxlen != ltlen)) {
2069  				error = -EFSCORRUPTED;
2070  				goto error0;
2071  			}
2072  		}
2073  #endif
2074  		/*
2075  		 * Update remaining by-block entry to the new, joined block.
2076  		 */
2077  		nbno = ltbno;
2078  		nlen = len + ltlen + gtlen;
2079  		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2080  			goto error0;
2081  	}
2082  	/*
2083  	 * Have only a left contiguous neighbor.
2084  	 * Merge it together with the new freespace.
2085  	 */
2086  	else if (haveleft) {
2087  		/*
2088  		 * Delete the old by-size entry on the left.
2089  		 */
2090  		if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
2091  			goto error0;
2092  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2093  			error = -EFSCORRUPTED;
2094  			goto error0;
2095  		}
2096  		if ((error = xfs_btree_delete(cnt_cur, &i)))
2097  			goto error0;
2098  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2099  			error = -EFSCORRUPTED;
2100  			goto error0;
2101  		}
2102  		/*
2103  		 * Back up the by-block cursor to the left neighbor, and
2104  		 * update its length.
2105  		 */
2106  		if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
2107  			goto error0;
2108  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2109  			error = -EFSCORRUPTED;
2110  			goto error0;
2111  		}
2112  		nbno = ltbno;
2113  		nlen = len + ltlen;
2114  		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2115  			goto error0;
2116  	}
2117  	/*
2118  	 * Have only a right contiguous neighbor.
2119  	 * Merge it together with the new freespace.
2120  	 */
2121  	else if (haveright) {
2122  		/*
2123  		 * Delete the old by-size entry on the right.
2124  		 */
2125  		if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
2126  			goto error0;
2127  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2128  			error = -EFSCORRUPTED;
2129  			goto error0;
2130  		}
2131  		if ((error = xfs_btree_delete(cnt_cur, &i)))
2132  			goto error0;
2133  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2134  			error = -EFSCORRUPTED;
2135  			goto error0;
2136  		}
2137  		/*
2138  		 * Update the starting block and length of the right
2139  		 * neighbor in the by-block tree.
2140  		 */
2141  		nbno = bno;
2142  		nlen = len + gtlen;
2143  		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2144  			goto error0;
2145  	}
2146  	/*
2147  	 * No contiguous neighbors.
2148  	 * Insert the new freespace into the by-block tree.
2149  	 */
2150  	else {
2151  		nbno = bno;
2152  		nlen = len;
2153  		if ((error = xfs_btree_insert(bno_cur, &i)))
2154  			goto error0;
2155  		if (XFS_IS_CORRUPT(mp, i != 1)) {
2156  			error = -EFSCORRUPTED;
2157  			goto error0;
2158  		}
2159  	}
2160  	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
2161  	bno_cur = NULL;
2162  	/*
2163  	 * In all cases we need to insert the new freespace in the by-size tree.
2164  	 */
2165  	if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
2166  		goto error0;
2167  	if (XFS_IS_CORRUPT(mp, i != 0)) {
2168  		error = -EFSCORRUPTED;
2169  		goto error0;
2170  	}
2171  	if ((error = xfs_btree_insert(cnt_cur, &i)))
2172  		goto error0;
2173  	if (XFS_IS_CORRUPT(mp, i != 1)) {
2174  		error = -EFSCORRUPTED;
2175  		goto error0;
2176  	}
2177  	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
2178  	cnt_cur = NULL;
2179  
2180  	/*
2181  	 * Update the freespace totals in the ag and superblock.
2182  	 */
2183  	error = xfs_alloc_update_counters(tp, agbp, len);
2184  	xfs_ag_resv_free_extent(agbp->b_pag, type, tp, len);
2185  	if (error)
2186  		goto error0;
2187  
2188  	XFS_STATS_INC(mp, xs_freex);
2189  	XFS_STATS_ADD(mp, xs_freeb, len);
2190  
2191  	trace_xfs_free_extent(mp, agno, bno, len, type, haveleft, haveright);
2192  
2193  	return 0;
2194  
2195   error0:
2196  	trace_xfs_free_extent(mp, agno, bno, len, type, -1, -1);
2197  	if (bno_cur)
2198  		xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
2199  	if (cnt_cur)
2200  		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
2201  	return error;
2202  }
2203  
2204  /*
2205   * Visible (exported) allocation/free functions.
2206   * Some of these are used just by xfs_alloc_btree.c and this file.
2207   */
2208  
2209  /*
2210   * Compute and fill in value of m_alloc_maxlevels.
2211   */
2212  void
xfs_alloc_compute_maxlevels(xfs_mount_t * mp)2213  xfs_alloc_compute_maxlevels(
2214  	xfs_mount_t	*mp)	/* file system mount structure */
2215  {
2216  	mp->m_alloc_maxlevels = xfs_btree_compute_maxlevels(mp->m_alloc_mnr,
2217  			(mp->m_sb.sb_agblocks + 1) / 2);
2218  	ASSERT(mp->m_alloc_maxlevels <= xfs_allocbt_maxlevels_ondisk());
2219  }
2220  
2221  /*
2222   * Find the length of the longest extent in an AG.  The 'need' parameter
2223   * specifies how much space we're going to need for the AGFL and the
2224   * 'reserved' parameter tells us how many blocks in this AG are reserved for
2225   * other callers.
2226   */
2227  xfs_extlen_t
xfs_alloc_longest_free_extent(struct xfs_perag * pag,xfs_extlen_t need,xfs_extlen_t reserved)2228  xfs_alloc_longest_free_extent(
2229  	struct xfs_perag	*pag,
2230  	xfs_extlen_t		need,
2231  	xfs_extlen_t		reserved)
2232  {
2233  	xfs_extlen_t		delta = 0;
2234  
2235  	/*
2236  	 * If the AGFL needs a recharge, we'll have to subtract that from the
2237  	 * longest extent.
2238  	 */
2239  	if (need > pag->pagf_flcount)
2240  		delta = need - pag->pagf_flcount;
2241  
2242  	/*
2243  	 * If we cannot maintain others' reservations with space from the
2244  	 * not-longest freesp extents, we'll have to subtract /that/ from
2245  	 * the longest extent too.
2246  	 */
2247  	if (pag->pagf_freeblks - pag->pagf_longest < reserved)
2248  		delta += reserved - (pag->pagf_freeblks - pag->pagf_longest);
2249  
2250  	/*
2251  	 * If the longest extent is long enough to satisfy all the
2252  	 * reservations and AGFL rules in place, we can return this extent.
2253  	 */
2254  	if (pag->pagf_longest > delta)
2255  		return min_t(xfs_extlen_t, pag->pag_mount->m_ag_max_usable,
2256  				pag->pagf_longest - delta);
2257  
2258  	/* Otherwise, let the caller try for 1 block if there's space. */
2259  	return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
2260  }
2261  
2262  /*
2263   * Compute the minimum length of the AGFL in the given AG.  If @pag is NULL,
2264   * return the largest possible minimum length.
2265   */
2266  unsigned int
xfs_alloc_min_freelist(struct xfs_mount * mp,struct xfs_perag * pag)2267  xfs_alloc_min_freelist(
2268  	struct xfs_mount	*mp,
2269  	struct xfs_perag	*pag)
2270  {
2271  	/* AG btrees have at least 1 level. */
2272  	static const uint8_t	fake_levels[XFS_BTNUM_AGF] = {1, 1, 1};
2273  	const uint8_t		*levels = pag ? pag->pagf_levels : fake_levels;
2274  	unsigned int		min_free;
2275  
2276  	ASSERT(mp->m_alloc_maxlevels > 0);
2277  
2278  	/*
2279  	 * For a btree shorter than the maximum height, the worst case is that
2280  	 * every level gets split and a new level is added, then while inserting
2281  	 * another entry to refill the AGFL, every level under the old root gets
2282  	 * split again. This is:
2283  	 *
2284  	 *   (full height split reservation) + (AGFL refill split height)
2285  	 * = (current height + 1) + (current height - 1)
2286  	 * = (new height) + (new height - 2)
2287  	 * = 2 * new height - 2
2288  	 *
2289  	 * For a btree of maximum height, the worst case is that every level
2290  	 * under the root gets split, then while inserting another entry to
2291  	 * refill the AGFL, every level under the root gets split again. This is
2292  	 * also:
2293  	 *
2294  	 *   2 * (current height - 1)
2295  	 * = 2 * (new height - 1)
2296  	 * = 2 * new height - 2
2297  	 */
2298  
2299  	/* space needed by-bno freespace btree */
2300  	min_free = min_t(unsigned int, levels[XFS_BTNUM_BNOi] + 1,
2301  				       mp->m_alloc_maxlevels) * 2 - 2;
2302  	/* space needed by-size freespace btree */
2303  	min_free += min_t(unsigned int, levels[XFS_BTNUM_CNTi] + 1,
2304  				       mp->m_alloc_maxlevels) * 2 - 2;
2305  	/* space needed reverse mapping used space btree */
2306  	if (xfs_has_rmapbt(mp))
2307  		min_free += min_t(unsigned int, levels[XFS_BTNUM_RMAPi] + 1,
2308  						mp->m_rmap_maxlevels) * 2 - 2;
2309  
2310  	return min_free;
2311  }
2312  
2313  /*
2314   * Check if the operation we are fixing up the freelist for should go ahead or
2315   * not. If we are freeing blocks, we always allow it, otherwise the allocation
2316   * is dependent on whether the size and shape of free space available will
2317   * permit the requested allocation to take place.
2318   */
2319  static bool
xfs_alloc_space_available(struct xfs_alloc_arg * args,xfs_extlen_t min_free,int flags)2320  xfs_alloc_space_available(
2321  	struct xfs_alloc_arg	*args,
2322  	xfs_extlen_t		min_free,
2323  	int			flags)
2324  {
2325  	struct xfs_perag	*pag = args->pag;
2326  	xfs_extlen_t		alloc_len, longest;
2327  	xfs_extlen_t		reservation; /* blocks that are still reserved */
2328  	int			available;
2329  	xfs_extlen_t		agflcount;
2330  
2331  	if (flags & XFS_ALLOC_FLAG_FREEING)
2332  		return true;
2333  
2334  	reservation = xfs_ag_resv_needed(pag, args->resv);
2335  
2336  	/* do we have enough contiguous free space for the allocation? */
2337  	alloc_len = args->minlen + (args->alignment - 1) + args->minalignslop;
2338  	longest = xfs_alloc_longest_free_extent(pag, min_free, reservation);
2339  	if (longest < alloc_len)
2340  		return false;
2341  
2342  	/*
2343  	 * Do we have enough free space remaining for the allocation? Don't
2344  	 * account extra agfl blocks because we are about to defer free them,
2345  	 * making them unavailable until the current transaction commits.
2346  	 */
2347  	agflcount = min_t(xfs_extlen_t, pag->pagf_flcount, min_free);
2348  	available = (int)(pag->pagf_freeblks + agflcount -
2349  			  reservation - min_free - args->minleft);
2350  	if (available < (int)max(args->total, alloc_len))
2351  		return false;
2352  
2353  	/*
2354  	 * Clamp maxlen to the amount of free space available for the actual
2355  	 * extent allocation.
2356  	 */
2357  	if (available < (int)args->maxlen && !(flags & XFS_ALLOC_FLAG_CHECK)) {
2358  		args->maxlen = available;
2359  		ASSERT(args->maxlen > 0);
2360  		ASSERT(args->maxlen >= args->minlen);
2361  	}
2362  
2363  	return true;
2364  }
2365  
2366  int
xfs_free_agfl_block(struct xfs_trans * tp,xfs_agnumber_t agno,xfs_agblock_t agbno,struct xfs_buf * agbp,struct xfs_owner_info * oinfo)2367  xfs_free_agfl_block(
2368  	struct xfs_trans	*tp,
2369  	xfs_agnumber_t		agno,
2370  	xfs_agblock_t		agbno,
2371  	struct xfs_buf		*agbp,
2372  	struct xfs_owner_info	*oinfo)
2373  {
2374  	int			error;
2375  	struct xfs_buf		*bp;
2376  
2377  	error = xfs_free_ag_extent(tp, agbp, agno, agbno, 1, oinfo,
2378  				   XFS_AG_RESV_AGFL);
2379  	if (error)
2380  		return error;
2381  
2382  	error = xfs_trans_get_buf(tp, tp->t_mountp->m_ddev_targp,
2383  			XFS_AGB_TO_DADDR(tp->t_mountp, agno, agbno),
2384  			tp->t_mountp->m_bsize, 0, &bp);
2385  	if (error)
2386  		return error;
2387  	xfs_trans_binval(tp, bp);
2388  
2389  	return 0;
2390  }
2391  
2392  /*
2393   * Check the agfl fields of the agf for inconsistency or corruption.
2394   *
2395   * The original purpose was to detect an agfl header padding mismatch between
2396   * current and early v5 kernels. This problem manifests as a 1-slot size
2397   * difference between the on-disk flcount and the active [first, last] range of
2398   * a wrapped agfl.
2399   *
2400   * However, we need to use these same checks to catch agfl count corruptions
2401   * unrelated to padding. This could occur on any v4 or v5 filesystem, so either
2402   * way, we need to reset the agfl and warn the user.
2403   *
2404   * Return true if a reset is required before the agfl can be used, false
2405   * otherwise.
2406   */
2407  static bool
xfs_agfl_needs_reset(struct xfs_mount * mp,struct xfs_agf * agf)2408  xfs_agfl_needs_reset(
2409  	struct xfs_mount	*mp,
2410  	struct xfs_agf		*agf)
2411  {
2412  	uint32_t		f = be32_to_cpu(agf->agf_flfirst);
2413  	uint32_t		l = be32_to_cpu(agf->agf_fllast);
2414  	uint32_t		c = be32_to_cpu(agf->agf_flcount);
2415  	int			agfl_size = xfs_agfl_size(mp);
2416  	int			active;
2417  
2418  	/*
2419  	 * The agf read verifier catches severe corruption of these fields.
2420  	 * Repeat some sanity checks to cover a packed -> unpacked mismatch if
2421  	 * the verifier allows it.
2422  	 */
2423  	if (f >= agfl_size || l >= agfl_size)
2424  		return true;
2425  	if (c > agfl_size)
2426  		return true;
2427  
2428  	/*
2429  	 * Check consistency between the on-disk count and the active range. An
2430  	 * agfl padding mismatch manifests as an inconsistent flcount.
2431  	 */
2432  	if (c && l >= f)
2433  		active = l - f + 1;
2434  	else if (c)
2435  		active = agfl_size - f + l + 1;
2436  	else
2437  		active = 0;
2438  
2439  	return active != c;
2440  }
2441  
2442  /*
2443   * Reset the agfl to an empty state. Ignore/drop any existing blocks since the
2444   * agfl content cannot be trusted. Warn the user that a repair is required to
2445   * recover leaked blocks.
2446   *
2447   * The purpose of this mechanism is to handle filesystems affected by the agfl
2448   * header padding mismatch problem. A reset keeps the filesystem online with a
2449   * relatively minor free space accounting inconsistency rather than suffer the
2450   * inevitable crash from use of an invalid agfl block.
2451   */
2452  static void
xfs_agfl_reset(struct xfs_trans * tp,struct xfs_buf * agbp,struct xfs_perag * pag)2453  xfs_agfl_reset(
2454  	struct xfs_trans	*tp,
2455  	struct xfs_buf		*agbp,
2456  	struct xfs_perag	*pag)
2457  {
2458  	struct xfs_mount	*mp = tp->t_mountp;
2459  	struct xfs_agf		*agf = agbp->b_addr;
2460  
2461  	ASSERT(xfs_perag_agfl_needs_reset(pag));
2462  	trace_xfs_agfl_reset(mp, agf, 0, _RET_IP_);
2463  
2464  	xfs_warn(mp,
2465  	       "WARNING: Reset corrupted AGFL on AG %u. %d blocks leaked. "
2466  	       "Please unmount and run xfs_repair.",
2467  	         pag->pag_agno, pag->pagf_flcount);
2468  
2469  	agf->agf_flfirst = 0;
2470  	agf->agf_fllast = cpu_to_be32(xfs_agfl_size(mp) - 1);
2471  	agf->agf_flcount = 0;
2472  	xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLLAST |
2473  				    XFS_AGF_FLCOUNT);
2474  
2475  	pag->pagf_flcount = 0;
2476  	clear_bit(XFS_AGSTATE_AGFL_NEEDS_RESET, &pag->pag_opstate);
2477  }
2478  
2479  /*
2480   * Defer an AGFL block free. This is effectively equivalent to
2481   * xfs_free_extent_later() with some special handling particular to AGFL blocks.
2482   *
2483   * Deferring AGFL frees helps prevent log reservation overruns due to too many
2484   * allocation operations in a transaction. AGFL frees are prone to this problem
2485   * because for one they are always freed one at a time. Further, an immediate
2486   * AGFL block free can cause a btree join and require another block free before
2487   * the real allocation can proceed. Deferring the free disconnects freeing up
2488   * the AGFL slot from freeing the block.
2489   */
2490  static int
xfs_defer_agfl_block(struct xfs_trans * tp,xfs_agnumber_t agno,xfs_agblock_t agbno,struct xfs_owner_info * oinfo)2491  xfs_defer_agfl_block(
2492  	struct xfs_trans		*tp,
2493  	xfs_agnumber_t			agno,
2494  	xfs_agblock_t			agbno,
2495  	struct xfs_owner_info		*oinfo)
2496  {
2497  	struct xfs_mount		*mp = tp->t_mountp;
2498  	struct xfs_extent_free_item	*xefi;
2499  	xfs_fsblock_t			fsbno = XFS_AGB_TO_FSB(mp, agno, agbno);
2500  
2501  	ASSERT(xfs_extfree_item_cache != NULL);
2502  	ASSERT(oinfo != NULL);
2503  
2504  	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, fsbno)))
2505  		return -EFSCORRUPTED;
2506  
2507  	xefi = kmem_cache_zalloc(xfs_extfree_item_cache,
2508  			       GFP_KERNEL | __GFP_NOFAIL);
2509  	xefi->xefi_startblock = fsbno;
2510  	xefi->xefi_blockcount = 1;
2511  	xefi->xefi_owner = oinfo->oi_owner;
2512  	xefi->xefi_agresv = XFS_AG_RESV_AGFL;
2513  
2514  	trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1);
2515  
2516  	xfs_extent_free_get_group(mp, xefi);
2517  	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_AGFL_FREE, &xefi->xefi_list);
2518  	return 0;
2519  }
2520  
2521  /*
2522   * Add the extent to the list of extents to be free at transaction end.
2523   * The list is maintained sorted (by block number).
2524   */
2525  int
__xfs_free_extent_later(struct xfs_trans * tp,xfs_fsblock_t bno,xfs_filblks_t len,const struct xfs_owner_info * oinfo,enum xfs_ag_resv_type type,bool skip_discard)2526  __xfs_free_extent_later(
2527  	struct xfs_trans		*tp,
2528  	xfs_fsblock_t			bno,
2529  	xfs_filblks_t			len,
2530  	const struct xfs_owner_info	*oinfo,
2531  	enum xfs_ag_resv_type		type,
2532  	bool				skip_discard)
2533  {
2534  	struct xfs_extent_free_item	*xefi;
2535  	struct xfs_mount		*mp = tp->t_mountp;
2536  #ifdef DEBUG
2537  	xfs_agnumber_t			agno;
2538  	xfs_agblock_t			agbno;
2539  
2540  	ASSERT(bno != NULLFSBLOCK);
2541  	ASSERT(len > 0);
2542  	ASSERT(len <= XFS_MAX_BMBT_EXTLEN);
2543  	ASSERT(!isnullstartblock(bno));
2544  	agno = XFS_FSB_TO_AGNO(mp, bno);
2545  	agbno = XFS_FSB_TO_AGBNO(mp, bno);
2546  	ASSERT(agno < mp->m_sb.sb_agcount);
2547  	ASSERT(agbno < mp->m_sb.sb_agblocks);
2548  	ASSERT(len < mp->m_sb.sb_agblocks);
2549  	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
2550  #endif
2551  	ASSERT(xfs_extfree_item_cache != NULL);
2552  	ASSERT(type != XFS_AG_RESV_AGFL);
2553  
2554  	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbext(mp, bno, len)))
2555  		return -EFSCORRUPTED;
2556  
2557  	xefi = kmem_cache_zalloc(xfs_extfree_item_cache,
2558  			       GFP_KERNEL | __GFP_NOFAIL);
2559  	xefi->xefi_startblock = bno;
2560  	xefi->xefi_blockcount = (xfs_extlen_t)len;
2561  	xefi->xefi_agresv = type;
2562  	if (skip_discard)
2563  		xefi->xefi_flags |= XFS_EFI_SKIP_DISCARD;
2564  	if (oinfo) {
2565  		ASSERT(oinfo->oi_offset == 0);
2566  
2567  		if (oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK)
2568  			xefi->xefi_flags |= XFS_EFI_ATTR_FORK;
2569  		if (oinfo->oi_flags & XFS_OWNER_INFO_BMBT_BLOCK)
2570  			xefi->xefi_flags |= XFS_EFI_BMBT_BLOCK;
2571  		xefi->xefi_owner = oinfo->oi_owner;
2572  	} else {
2573  		xefi->xefi_owner = XFS_RMAP_OWN_NULL;
2574  	}
2575  	trace_xfs_bmap_free_defer(mp,
2576  			XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
2577  			XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
2578  
2579  	xfs_extent_free_get_group(mp, xefi);
2580  	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &xefi->xefi_list);
2581  	return 0;
2582  }
2583  
2584  /*
2585   * Check if an AGF has a free extent record whose length is equal to
2586   * args->minlen.
2587   */
2588  STATIC int
xfs_exact_minlen_extent_available(struct xfs_alloc_arg * args,struct xfs_buf * agbp,int * stat)2589  xfs_exact_minlen_extent_available(
2590  	struct xfs_alloc_arg	*args,
2591  	struct xfs_buf		*agbp,
2592  	int			*stat)
2593  {
2594  	struct xfs_btree_cur	*cnt_cur;
2595  	xfs_agblock_t		fbno;
2596  	xfs_extlen_t		flen;
2597  	int			error = 0;
2598  
2599  	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, agbp,
2600  					args->pag, XFS_BTNUM_CNT);
2601  	error = xfs_alloc_lookup_ge(cnt_cur, 0, args->minlen, stat);
2602  	if (error)
2603  		goto out;
2604  
2605  	if (*stat == 0) {
2606  		error = -EFSCORRUPTED;
2607  		goto out;
2608  	}
2609  
2610  	error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, stat);
2611  	if (error)
2612  		goto out;
2613  
2614  	if (*stat == 1 && flen != args->minlen)
2615  		*stat = 0;
2616  
2617  out:
2618  	xfs_btree_del_cursor(cnt_cur, error);
2619  
2620  	return error;
2621  }
2622  
2623  /*
2624   * Decide whether to use this allocation group for this allocation.
2625   * If so, fix up the btree freelist's size.
2626   */
2627  int			/* error */
xfs_alloc_fix_freelist(struct xfs_alloc_arg * args,uint32_t alloc_flags)2628  xfs_alloc_fix_freelist(
2629  	struct xfs_alloc_arg	*args,	/* allocation argument structure */
2630  	uint32_t		alloc_flags)
2631  {
2632  	struct xfs_mount	*mp = args->mp;
2633  	struct xfs_perag	*pag = args->pag;
2634  	struct xfs_trans	*tp = args->tp;
2635  	struct xfs_buf		*agbp = NULL;
2636  	struct xfs_buf		*agflbp = NULL;
2637  	struct xfs_alloc_arg	targs;	/* local allocation arguments */
2638  	xfs_agblock_t		bno;	/* freelist block */
2639  	xfs_extlen_t		need;	/* total blocks needed in freelist */
2640  	int			error = 0;
2641  
2642  	/* deferred ops (AGFL block frees) require permanent transactions */
2643  	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
2644  
2645  	if (!xfs_perag_initialised_agf(pag)) {
2646  		error = xfs_alloc_read_agf(pag, tp, alloc_flags, &agbp);
2647  		if (error) {
2648  			/* Couldn't lock the AGF so skip this AG. */
2649  			if (error == -EAGAIN)
2650  				error = 0;
2651  			goto out_no_agbp;
2652  		}
2653  	}
2654  
2655  	/*
2656  	 * If this is a metadata preferred pag and we are user data then try
2657  	 * somewhere else if we are not being asked to try harder at this
2658  	 * point
2659  	 */
2660  	if (xfs_perag_prefers_metadata(pag) &&
2661  	    (args->datatype & XFS_ALLOC_USERDATA) &&
2662  	    (alloc_flags & XFS_ALLOC_FLAG_TRYLOCK)) {
2663  		ASSERT(!(alloc_flags & XFS_ALLOC_FLAG_FREEING));
2664  		goto out_agbp_relse;
2665  	}
2666  
2667  	need = xfs_alloc_min_freelist(mp, pag);
2668  	if (!xfs_alloc_space_available(args, need, alloc_flags |
2669  			XFS_ALLOC_FLAG_CHECK))
2670  		goto out_agbp_relse;
2671  
2672  	/*
2673  	 * Get the a.g. freespace buffer.
2674  	 * Can fail if we're not blocking on locks, and it's held.
2675  	 */
2676  	if (!agbp) {
2677  		error = xfs_alloc_read_agf(pag, tp, alloc_flags, &agbp);
2678  		if (error) {
2679  			/* Couldn't lock the AGF so skip this AG. */
2680  			if (error == -EAGAIN)
2681  				error = 0;
2682  			goto out_no_agbp;
2683  		}
2684  	}
2685  
2686  	/* reset a padding mismatched agfl before final free space check */
2687  	if (xfs_perag_agfl_needs_reset(pag))
2688  		xfs_agfl_reset(tp, agbp, pag);
2689  
2690  	/* If there isn't enough total space or single-extent, reject it. */
2691  	need = xfs_alloc_min_freelist(mp, pag);
2692  	if (!xfs_alloc_space_available(args, need, alloc_flags))
2693  		goto out_agbp_relse;
2694  
2695  	if (IS_ENABLED(CONFIG_XFS_DEBUG) && args->alloc_minlen_only) {
2696  		int stat;
2697  
2698  		error = xfs_exact_minlen_extent_available(args, agbp, &stat);
2699  		if (error || !stat)
2700  			goto out_agbp_relse;
2701  	}
2702  
2703  	/*
2704  	 * Make the freelist shorter if it's too long.
2705  	 *
2706  	 * Note that from this point onwards, we will always release the agf and
2707  	 * agfl buffers on error. This handles the case where we error out and
2708  	 * the buffers are clean or may not have been joined to the transaction
2709  	 * and hence need to be released manually. If they have been joined to
2710  	 * the transaction, then xfs_trans_brelse() will handle them
2711  	 * appropriately based on the recursion count and dirty state of the
2712  	 * buffer.
2713  	 *
2714  	 * XXX (dgc): When we have lots of free space, does this buy us
2715  	 * anything other than extra overhead when we need to put more blocks
2716  	 * back on the free list? Maybe we should only do this when space is
2717  	 * getting low or the AGFL is more than half full?
2718  	 *
2719  	 * The NOSHRINK flag prevents the AGFL from being shrunk if it's too
2720  	 * big; the NORMAP flag prevents AGFL expand/shrink operations from
2721  	 * updating the rmapbt.  Both flags are used in xfs_repair while we're
2722  	 * rebuilding the rmapbt, and neither are used by the kernel.  They're
2723  	 * both required to ensure that rmaps are correctly recorded for the
2724  	 * regenerated AGFL, bnobt, and cntbt.  See repair/phase5.c and
2725  	 * repair/rmap.c in xfsprogs for details.
2726  	 */
2727  	memset(&targs, 0, sizeof(targs));
2728  	/* struct copy below */
2729  	if (alloc_flags & XFS_ALLOC_FLAG_NORMAP)
2730  		targs.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
2731  	else
2732  		targs.oinfo = XFS_RMAP_OINFO_AG;
2733  	while (!(alloc_flags & XFS_ALLOC_FLAG_NOSHRINK) &&
2734  			pag->pagf_flcount > need) {
2735  		error = xfs_alloc_get_freelist(pag, tp, agbp, &bno, 0);
2736  		if (error)
2737  			goto out_agbp_relse;
2738  
2739  		/* defer agfl frees */
2740  		error = xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
2741  		if (error)
2742  			goto out_agbp_relse;
2743  	}
2744  
2745  	targs.tp = tp;
2746  	targs.mp = mp;
2747  	targs.agbp = agbp;
2748  	targs.agno = args->agno;
2749  	targs.alignment = targs.minlen = targs.prod = 1;
2750  	targs.pag = pag;
2751  	error = xfs_alloc_read_agfl(pag, tp, &agflbp);
2752  	if (error)
2753  		goto out_agbp_relse;
2754  
2755  	/* Make the freelist longer if it's too short. */
2756  	while (pag->pagf_flcount < need) {
2757  		targs.agbno = 0;
2758  		targs.maxlen = need - pag->pagf_flcount;
2759  		targs.resv = XFS_AG_RESV_AGFL;
2760  
2761  		/* Allocate as many blocks as possible at once. */
2762  		error = xfs_alloc_ag_vextent_size(&targs, alloc_flags);
2763  		if (error)
2764  			goto out_agflbp_relse;
2765  
2766  		/*
2767  		 * Stop if we run out.  Won't happen if callers are obeying
2768  		 * the restrictions correctly.  Can happen for free calls
2769  		 * on a completely full ag.
2770  		 */
2771  		if (targs.agbno == NULLAGBLOCK) {
2772  			if (alloc_flags & XFS_ALLOC_FLAG_FREEING)
2773  				break;
2774  			goto out_agflbp_relse;
2775  		}
2776  
2777  		if (!xfs_rmap_should_skip_owner_update(&targs.oinfo)) {
2778  			error = xfs_rmap_alloc(tp, agbp, pag,
2779  				       targs.agbno, targs.len, &targs.oinfo);
2780  			if (error)
2781  				goto out_agflbp_relse;
2782  		}
2783  		error = xfs_alloc_update_counters(tp, agbp,
2784  						  -((long)(targs.len)));
2785  		if (error)
2786  			goto out_agflbp_relse;
2787  
2788  		/*
2789  		 * Put each allocated block on the list.
2790  		 */
2791  		for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
2792  			error = xfs_alloc_put_freelist(pag, tp, agbp,
2793  							agflbp, bno, 0);
2794  			if (error)
2795  				goto out_agflbp_relse;
2796  		}
2797  	}
2798  	xfs_trans_brelse(tp, agflbp);
2799  	args->agbp = agbp;
2800  	return 0;
2801  
2802  out_agflbp_relse:
2803  	xfs_trans_brelse(tp, agflbp);
2804  out_agbp_relse:
2805  	if (agbp)
2806  		xfs_trans_brelse(tp, agbp);
2807  out_no_agbp:
2808  	args->agbp = NULL;
2809  	return error;
2810  }
2811  
2812  /*
2813   * Get a block from the freelist.
2814   * Returns with the buffer for the block gotten.
2815   */
2816  int
xfs_alloc_get_freelist(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agblock_t * bnop,int btreeblk)2817  xfs_alloc_get_freelist(
2818  	struct xfs_perag	*pag,
2819  	struct xfs_trans	*tp,
2820  	struct xfs_buf		*agbp,
2821  	xfs_agblock_t		*bnop,
2822  	int			btreeblk)
2823  {
2824  	struct xfs_agf		*agf = agbp->b_addr;
2825  	struct xfs_buf		*agflbp;
2826  	xfs_agblock_t		bno;
2827  	__be32			*agfl_bno;
2828  	int			error;
2829  	uint32_t		logflags;
2830  	struct xfs_mount	*mp = tp->t_mountp;
2831  
2832  	/*
2833  	 * Freelist is empty, give up.
2834  	 */
2835  	if (!agf->agf_flcount) {
2836  		*bnop = NULLAGBLOCK;
2837  		return 0;
2838  	}
2839  	/*
2840  	 * Read the array of free blocks.
2841  	 */
2842  	error = xfs_alloc_read_agfl(pag, tp, &agflbp);
2843  	if (error)
2844  		return error;
2845  
2846  
2847  	/*
2848  	 * Get the block number and update the data structures.
2849  	 */
2850  	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
2851  	bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2852  	if (XFS_IS_CORRUPT(tp->t_mountp, !xfs_verify_agbno(pag, bno)))
2853  		return -EFSCORRUPTED;
2854  
2855  	be32_add_cpu(&agf->agf_flfirst, 1);
2856  	xfs_trans_brelse(tp, agflbp);
2857  	if (be32_to_cpu(agf->agf_flfirst) == xfs_agfl_size(mp))
2858  		agf->agf_flfirst = 0;
2859  
2860  	ASSERT(!xfs_perag_agfl_needs_reset(pag));
2861  	be32_add_cpu(&agf->agf_flcount, -1);
2862  	pag->pagf_flcount--;
2863  
2864  	logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2865  	if (btreeblk) {
2866  		be32_add_cpu(&agf->agf_btreeblks, 1);
2867  		pag->pagf_btreeblks++;
2868  		logflags |= XFS_AGF_BTREEBLKS;
2869  	}
2870  
2871  	xfs_alloc_log_agf(tp, agbp, logflags);
2872  	*bnop = bno;
2873  
2874  	return 0;
2875  }
2876  
2877  /*
2878   * Log the given fields from the agf structure.
2879   */
2880  void
xfs_alloc_log_agf(struct xfs_trans * tp,struct xfs_buf * bp,uint32_t fields)2881  xfs_alloc_log_agf(
2882  	struct xfs_trans	*tp,
2883  	struct xfs_buf		*bp,
2884  	uint32_t		fields)
2885  {
2886  	int	first;		/* first byte offset */
2887  	int	last;		/* last byte offset */
2888  	static const short	offsets[] = {
2889  		offsetof(xfs_agf_t, agf_magicnum),
2890  		offsetof(xfs_agf_t, agf_versionnum),
2891  		offsetof(xfs_agf_t, agf_seqno),
2892  		offsetof(xfs_agf_t, agf_length),
2893  		offsetof(xfs_agf_t, agf_roots[0]),
2894  		offsetof(xfs_agf_t, agf_levels[0]),
2895  		offsetof(xfs_agf_t, agf_flfirst),
2896  		offsetof(xfs_agf_t, agf_fllast),
2897  		offsetof(xfs_agf_t, agf_flcount),
2898  		offsetof(xfs_agf_t, agf_freeblks),
2899  		offsetof(xfs_agf_t, agf_longest),
2900  		offsetof(xfs_agf_t, agf_btreeblks),
2901  		offsetof(xfs_agf_t, agf_uuid),
2902  		offsetof(xfs_agf_t, agf_rmap_blocks),
2903  		offsetof(xfs_agf_t, agf_refcount_blocks),
2904  		offsetof(xfs_agf_t, agf_refcount_root),
2905  		offsetof(xfs_agf_t, agf_refcount_level),
2906  		/* needed so that we don't log the whole rest of the structure: */
2907  		offsetof(xfs_agf_t, agf_spare64),
2908  		sizeof(xfs_agf_t)
2909  	};
2910  
2911  	trace_xfs_agf(tp->t_mountp, bp->b_addr, fields, _RET_IP_);
2912  
2913  	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF);
2914  
2915  	xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2916  	xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2917  }
2918  
2919  /*
2920   * Put the block on the freelist for the allocation group.
2921   */
2922  int
xfs_alloc_put_freelist(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,struct xfs_buf * agflbp,xfs_agblock_t bno,int btreeblk)2923  xfs_alloc_put_freelist(
2924  	struct xfs_perag	*pag,
2925  	struct xfs_trans	*tp,
2926  	struct xfs_buf		*agbp,
2927  	struct xfs_buf		*agflbp,
2928  	xfs_agblock_t		bno,
2929  	int			btreeblk)
2930  {
2931  	struct xfs_mount	*mp = tp->t_mountp;
2932  	struct xfs_agf		*agf = agbp->b_addr;
2933  	__be32			*blockp;
2934  	int			error;
2935  	uint32_t		logflags;
2936  	__be32			*agfl_bno;
2937  	int			startoff;
2938  
2939  	if (!agflbp) {
2940  		error = xfs_alloc_read_agfl(pag, tp, &agflbp);
2941  		if (error)
2942  			return error;
2943  	}
2944  
2945  	be32_add_cpu(&agf->agf_fllast, 1);
2946  	if (be32_to_cpu(agf->agf_fllast) == xfs_agfl_size(mp))
2947  		agf->agf_fllast = 0;
2948  
2949  	ASSERT(!xfs_perag_agfl_needs_reset(pag));
2950  	be32_add_cpu(&agf->agf_flcount, 1);
2951  	pag->pagf_flcount++;
2952  
2953  	logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2954  	if (btreeblk) {
2955  		be32_add_cpu(&agf->agf_btreeblks, -1);
2956  		pag->pagf_btreeblks--;
2957  		logflags |= XFS_AGF_BTREEBLKS;
2958  	}
2959  
2960  	xfs_alloc_log_agf(tp, agbp, logflags);
2961  
2962  	ASSERT(be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp));
2963  
2964  	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
2965  	blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)];
2966  	*blockp = cpu_to_be32(bno);
2967  	startoff = (char *)blockp - (char *)agflbp->b_addr;
2968  
2969  	xfs_alloc_log_agf(tp, agbp, logflags);
2970  
2971  	xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF);
2972  	xfs_trans_log_buf(tp, agflbp, startoff,
2973  			  startoff + sizeof(xfs_agblock_t) - 1);
2974  	return 0;
2975  }
2976  
2977  /*
2978   * Check that this AGF/AGI header's sequence number and length matches the AG
2979   * number and size in fsblocks.
2980   */
2981  xfs_failaddr_t
xfs_validate_ag_length(struct xfs_buf * bp,uint32_t seqno,uint32_t length)2982  xfs_validate_ag_length(
2983  	struct xfs_buf		*bp,
2984  	uint32_t		seqno,
2985  	uint32_t		length)
2986  {
2987  	struct xfs_mount	*mp = bp->b_mount;
2988  	/*
2989  	 * During growfs operations, the perag is not fully initialised,
2990  	 * so we can't use it for any useful checking. growfs ensures we can't
2991  	 * use it by using uncached buffers that don't have the perag attached
2992  	 * so we can detect and avoid this problem.
2993  	 */
2994  	if (bp->b_pag && seqno != bp->b_pag->pag_agno)
2995  		return __this_address;
2996  
2997  	/*
2998  	 * Only the last AG in the filesystem is allowed to be shorter
2999  	 * than the AG size recorded in the superblock.
3000  	 */
3001  	if (length != mp->m_sb.sb_agblocks) {
3002  		/*
3003  		 * During growfs, the new last AG can get here before we
3004  		 * have updated the superblock. Give it a pass on the seqno
3005  		 * check.
3006  		 */
3007  		if (bp->b_pag && seqno != mp->m_sb.sb_agcount - 1)
3008  			return __this_address;
3009  		if (length < XFS_MIN_AG_BLOCKS)
3010  			return __this_address;
3011  		if (length > mp->m_sb.sb_agblocks)
3012  			return __this_address;
3013  	}
3014  
3015  	return NULL;
3016  }
3017  
3018  /*
3019   * Verify the AGF is consistent.
3020   *
3021   * We do not verify the AGFL indexes in the AGF are fully consistent here
3022   * because of issues with variable on-disk structure sizes. Instead, we check
3023   * the agfl indexes for consistency when we initialise the perag from the AGF
3024   * information after a read completes.
3025   *
3026   * If the index is inconsistent, then we mark the perag as needing an AGFL
3027   * reset. The first AGFL update performed then resets the AGFL indexes and
3028   * refills the AGFL with known good free blocks, allowing the filesystem to
3029   * continue operating normally at the cost of a few leaked free space blocks.
3030   */
3031  static xfs_failaddr_t
xfs_agf_verify(struct xfs_buf * bp)3032  xfs_agf_verify(
3033  	struct xfs_buf		*bp)
3034  {
3035  	struct xfs_mount	*mp = bp->b_mount;
3036  	struct xfs_agf		*agf = bp->b_addr;
3037  	xfs_failaddr_t		fa;
3038  	uint32_t		agf_seqno = be32_to_cpu(agf->agf_seqno);
3039  	uint32_t		agf_length = be32_to_cpu(agf->agf_length);
3040  
3041  	if (xfs_has_crc(mp)) {
3042  		if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid))
3043  			return __this_address;
3044  		if (!xfs_log_check_lsn(mp, be64_to_cpu(agf->agf_lsn)))
3045  			return __this_address;
3046  	}
3047  
3048  	if (!xfs_verify_magic(bp, agf->agf_magicnum))
3049  		return __this_address;
3050  
3051  	if (!XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)))
3052  		return __this_address;
3053  
3054  	/*
3055  	 * Both agf_seqno and agf_length need to validated before anything else
3056  	 * block number related in the AGF or AGFL can be checked.
3057  	 */
3058  	fa = xfs_validate_ag_length(bp, agf_seqno, agf_length);
3059  	if (fa)
3060  		return fa;
3061  
3062  	if (be32_to_cpu(agf->agf_flfirst) >= xfs_agfl_size(mp))
3063  		return __this_address;
3064  	if (be32_to_cpu(agf->agf_fllast) >= xfs_agfl_size(mp))
3065  		return __this_address;
3066  	if (be32_to_cpu(agf->agf_flcount) > xfs_agfl_size(mp))
3067  		return __this_address;
3068  
3069  	if (be32_to_cpu(agf->agf_freeblks) < be32_to_cpu(agf->agf_longest) ||
3070  	    be32_to_cpu(agf->agf_freeblks) > agf_length)
3071  		return __this_address;
3072  
3073  	if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) < 1 ||
3074  	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) < 1 ||
3075  	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) >
3076  						mp->m_alloc_maxlevels ||
3077  	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) >
3078  						mp->m_alloc_maxlevels)
3079  		return __this_address;
3080  
3081  	if (xfs_has_lazysbcount(mp) &&
3082  	    be32_to_cpu(agf->agf_btreeblks) > agf_length)
3083  		return __this_address;
3084  
3085  	if (xfs_has_rmapbt(mp)) {
3086  		if (be32_to_cpu(agf->agf_rmap_blocks) > agf_length)
3087  			return __this_address;
3088  
3089  		if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) < 1 ||
3090  		    be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) >
3091  							mp->m_rmap_maxlevels)
3092  			return __this_address;
3093  	}
3094  
3095  	if (xfs_has_reflink(mp)) {
3096  		if (be32_to_cpu(agf->agf_refcount_blocks) > agf_length)
3097  			return __this_address;
3098  
3099  		if (be32_to_cpu(agf->agf_refcount_level) < 1 ||
3100  		    be32_to_cpu(agf->agf_refcount_level) > mp->m_refc_maxlevels)
3101  			return __this_address;
3102  	}
3103  
3104  	return NULL;
3105  }
3106  
3107  static void
xfs_agf_read_verify(struct xfs_buf * bp)3108  xfs_agf_read_verify(
3109  	struct xfs_buf	*bp)
3110  {
3111  	struct xfs_mount *mp = bp->b_mount;
3112  	xfs_failaddr_t	fa;
3113  
3114  	if (xfs_has_crc(mp) &&
3115  	    !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF))
3116  		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
3117  	else {
3118  		fa = xfs_agf_verify(bp);
3119  		if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_ALLOC_READ_AGF))
3120  			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
3121  	}
3122  }
3123  
3124  static void
xfs_agf_write_verify(struct xfs_buf * bp)3125  xfs_agf_write_verify(
3126  	struct xfs_buf	*bp)
3127  {
3128  	struct xfs_mount	*mp = bp->b_mount;
3129  	struct xfs_buf_log_item	*bip = bp->b_log_item;
3130  	struct xfs_agf		*agf = bp->b_addr;
3131  	xfs_failaddr_t		fa;
3132  
3133  	fa = xfs_agf_verify(bp);
3134  	if (fa) {
3135  		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
3136  		return;
3137  	}
3138  
3139  	if (!xfs_has_crc(mp))
3140  		return;
3141  
3142  	if (bip)
3143  		agf->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn);
3144  
3145  	xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF);
3146  }
3147  
3148  const struct xfs_buf_ops xfs_agf_buf_ops = {
3149  	.name = "xfs_agf",
3150  	.magic = { cpu_to_be32(XFS_AGF_MAGIC), cpu_to_be32(XFS_AGF_MAGIC) },
3151  	.verify_read = xfs_agf_read_verify,
3152  	.verify_write = xfs_agf_write_verify,
3153  	.verify_struct = xfs_agf_verify,
3154  };
3155  
3156  /*
3157   * Read in the allocation group header (free/alloc section).
3158   */
3159  int
xfs_read_agf(struct xfs_perag * pag,struct xfs_trans * tp,int flags,struct xfs_buf ** agfbpp)3160  xfs_read_agf(
3161  	struct xfs_perag	*pag,
3162  	struct xfs_trans	*tp,
3163  	int			flags,
3164  	struct xfs_buf		**agfbpp)
3165  {
3166  	struct xfs_mount	*mp = pag->pag_mount;
3167  	int			error;
3168  
3169  	trace_xfs_read_agf(pag->pag_mount, pag->pag_agno);
3170  
3171  	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3172  			XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGF_DADDR(mp)),
3173  			XFS_FSS_TO_BB(mp, 1), flags, agfbpp, &xfs_agf_buf_ops);
3174  	if (error)
3175  		return error;
3176  
3177  	xfs_buf_set_ref(*agfbpp, XFS_AGF_REF);
3178  	return 0;
3179  }
3180  
3181  /*
3182   * Read in the allocation group header (free/alloc section) and initialise the
3183   * perag structure if necessary. If the caller provides @agfbpp, then return the
3184   * locked buffer to the caller, otherwise free it.
3185   */
3186  int
xfs_alloc_read_agf(struct xfs_perag * pag,struct xfs_trans * tp,int flags,struct xfs_buf ** agfbpp)3187  xfs_alloc_read_agf(
3188  	struct xfs_perag	*pag,
3189  	struct xfs_trans	*tp,
3190  	int			flags,
3191  	struct xfs_buf		**agfbpp)
3192  {
3193  	struct xfs_buf		*agfbp;
3194  	struct xfs_agf		*agf;
3195  	int			error;
3196  	int			allocbt_blks;
3197  
3198  	trace_xfs_alloc_read_agf(pag->pag_mount, pag->pag_agno);
3199  
3200  	/* We don't support trylock when freeing. */
3201  	ASSERT((flags & (XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK)) !=
3202  			(XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK));
3203  	error = xfs_read_agf(pag, tp,
3204  			(flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
3205  			&agfbp);
3206  	if (error)
3207  		return error;
3208  
3209  	agf = agfbp->b_addr;
3210  	if (!xfs_perag_initialised_agf(pag)) {
3211  		pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
3212  		pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
3213  		pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
3214  		pag->pagf_longest = be32_to_cpu(agf->agf_longest);
3215  		pag->pagf_levels[XFS_BTNUM_BNOi] =
3216  			be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
3217  		pag->pagf_levels[XFS_BTNUM_CNTi] =
3218  			be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
3219  		pag->pagf_levels[XFS_BTNUM_RMAPi] =
3220  			be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]);
3221  		pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level);
3222  		if (xfs_agfl_needs_reset(pag->pag_mount, agf))
3223  			set_bit(XFS_AGSTATE_AGFL_NEEDS_RESET, &pag->pag_opstate);
3224  		else
3225  			clear_bit(XFS_AGSTATE_AGFL_NEEDS_RESET, &pag->pag_opstate);
3226  
3227  		/*
3228  		 * Update the in-core allocbt counter. Filter out the rmapbt
3229  		 * subset of the btreeblks counter because the rmapbt is managed
3230  		 * by perag reservation. Subtract one for the rmapbt root block
3231  		 * because the rmap counter includes it while the btreeblks
3232  		 * counter only tracks non-root blocks.
3233  		 */
3234  		allocbt_blks = pag->pagf_btreeblks;
3235  		if (xfs_has_rmapbt(pag->pag_mount))
3236  			allocbt_blks -= be32_to_cpu(agf->agf_rmap_blocks) - 1;
3237  		if (allocbt_blks > 0)
3238  			atomic64_add(allocbt_blks,
3239  					&pag->pag_mount->m_allocbt_blks);
3240  
3241  		set_bit(XFS_AGSTATE_AGF_INIT, &pag->pag_opstate);
3242  	}
3243  #ifdef DEBUG
3244  	else if (!xfs_is_shutdown(pag->pag_mount)) {
3245  		ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
3246  		ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
3247  		ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
3248  		ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
3249  		ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
3250  		       be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
3251  		ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
3252  		       be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
3253  	}
3254  #endif
3255  	if (agfbpp)
3256  		*agfbpp = agfbp;
3257  	else
3258  		xfs_trans_brelse(tp, agfbp);
3259  	return 0;
3260  }
3261  
3262  /*
3263   * Pre-proces allocation arguments to set initial state that we don't require
3264   * callers to set up correctly, as well as bounds check the allocation args
3265   * that are set up.
3266   */
3267  static int
xfs_alloc_vextent_check_args(struct xfs_alloc_arg * args,xfs_fsblock_t target,xfs_agnumber_t * minimum_agno)3268  xfs_alloc_vextent_check_args(
3269  	struct xfs_alloc_arg	*args,
3270  	xfs_fsblock_t		target,
3271  	xfs_agnumber_t		*minimum_agno)
3272  {
3273  	struct xfs_mount	*mp = args->mp;
3274  	xfs_agblock_t		agsize;
3275  
3276  	args->fsbno = NULLFSBLOCK;
3277  
3278  	*minimum_agno = 0;
3279  	if (args->tp->t_highest_agno != NULLAGNUMBER)
3280  		*minimum_agno = args->tp->t_highest_agno;
3281  
3282  	/*
3283  	 * Just fix this up, for the case where the last a.g. is shorter
3284  	 * (or there's only one a.g.) and the caller couldn't easily figure
3285  	 * that out (xfs_bmap_alloc).
3286  	 */
3287  	agsize = mp->m_sb.sb_agblocks;
3288  	if (args->maxlen > agsize)
3289  		args->maxlen = agsize;
3290  	if (args->alignment == 0)
3291  		args->alignment = 1;
3292  
3293  	ASSERT(args->minlen > 0);
3294  	ASSERT(args->maxlen > 0);
3295  	ASSERT(args->alignment > 0);
3296  	ASSERT(args->resv != XFS_AG_RESV_AGFL);
3297  
3298  	ASSERT(XFS_FSB_TO_AGNO(mp, target) < mp->m_sb.sb_agcount);
3299  	ASSERT(XFS_FSB_TO_AGBNO(mp, target) < agsize);
3300  	ASSERT(args->minlen <= args->maxlen);
3301  	ASSERT(args->minlen <= agsize);
3302  	ASSERT(args->mod < args->prod);
3303  
3304  	if (XFS_FSB_TO_AGNO(mp, target) >= mp->m_sb.sb_agcount ||
3305  	    XFS_FSB_TO_AGBNO(mp, target) >= agsize ||
3306  	    args->minlen > args->maxlen || args->minlen > agsize ||
3307  	    args->mod >= args->prod) {
3308  		trace_xfs_alloc_vextent_badargs(args);
3309  		return -ENOSPC;
3310  	}
3311  
3312  	if (args->agno != NULLAGNUMBER && *minimum_agno > args->agno) {
3313  		trace_xfs_alloc_vextent_skip_deadlock(args);
3314  		return -ENOSPC;
3315  	}
3316  	return 0;
3317  
3318  }
3319  
3320  /*
3321   * Prepare an AG for allocation. If the AG is not prepared to accept the
3322   * allocation, return failure.
3323   *
3324   * XXX(dgc): The complexity of "need_pag" will go away as all caller paths are
3325   * modified to hold their own perag references.
3326   */
3327  static int
xfs_alloc_vextent_prepare_ag(struct xfs_alloc_arg * args,uint32_t alloc_flags)3328  xfs_alloc_vextent_prepare_ag(
3329  	struct xfs_alloc_arg	*args,
3330  	uint32_t		alloc_flags)
3331  {
3332  	bool			need_pag = !args->pag;
3333  	int			error;
3334  
3335  	if (need_pag)
3336  		args->pag = xfs_perag_get(args->mp, args->agno);
3337  
3338  	args->agbp = NULL;
3339  	error = xfs_alloc_fix_freelist(args, alloc_flags);
3340  	if (error) {
3341  		trace_xfs_alloc_vextent_nofix(args);
3342  		if (need_pag)
3343  			xfs_perag_put(args->pag);
3344  		args->agbno = NULLAGBLOCK;
3345  		return error;
3346  	}
3347  	if (!args->agbp) {
3348  		/* cannot allocate in this AG at all */
3349  		trace_xfs_alloc_vextent_noagbp(args);
3350  		args->agbno = NULLAGBLOCK;
3351  		return 0;
3352  	}
3353  	args->wasfromfl = 0;
3354  	return 0;
3355  }
3356  
3357  /*
3358   * Post-process allocation results to account for the allocation if it succeed
3359   * and set the allocated block number correctly for the caller.
3360   *
3361   * XXX: we should really be returning ENOSPC for ENOSPC, not
3362   * hiding it behind a "successful" NULLFSBLOCK allocation.
3363   */
3364  static int
xfs_alloc_vextent_finish(struct xfs_alloc_arg * args,xfs_agnumber_t minimum_agno,int alloc_error,bool drop_perag)3365  xfs_alloc_vextent_finish(
3366  	struct xfs_alloc_arg	*args,
3367  	xfs_agnumber_t		minimum_agno,
3368  	int			alloc_error,
3369  	bool			drop_perag)
3370  {
3371  	struct xfs_mount	*mp = args->mp;
3372  	int			error = 0;
3373  
3374  	/*
3375  	 * We can end up here with a locked AGF. If we failed, the caller is
3376  	 * likely going to try to allocate again with different parameters, and
3377  	 * that can widen the AGs that are searched for free space. If we have
3378  	 * to do BMBT block allocation, we have to do a new allocation.
3379  	 *
3380  	 * Hence leaving this function with the AGF locked opens up potential
3381  	 * ABBA AGF deadlocks because a future allocation attempt in this
3382  	 * transaction may attempt to lock a lower number AGF.
3383  	 *
3384  	 * We can't release the AGF until the transaction is commited, so at
3385  	 * this point we must update the "first allocation" tracker to point at
3386  	 * this AG if the tracker is empty or points to a lower AG. This allows
3387  	 * the next allocation attempt to be modified appropriately to avoid
3388  	 * deadlocks.
3389  	 */
3390  	if (args->agbp &&
3391  	    (args->tp->t_highest_agno == NULLAGNUMBER ||
3392  	     args->agno > minimum_agno))
3393  		args->tp->t_highest_agno = args->agno;
3394  
3395  	/*
3396  	 * If the allocation failed with an error or we had an ENOSPC result,
3397  	 * preserve the returned error whilst also marking the allocation result
3398  	 * as "no extent allocated". This ensures that callers that fail to
3399  	 * capture the error will still treat it as a failed allocation.
3400  	 */
3401  	if (alloc_error || args->agbno == NULLAGBLOCK) {
3402  		args->fsbno = NULLFSBLOCK;
3403  		error = alloc_error;
3404  		goto out_drop_perag;
3405  	}
3406  
3407  	args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
3408  
3409  	ASSERT(args->len >= args->minlen);
3410  	ASSERT(args->len <= args->maxlen);
3411  	ASSERT(args->agbno % args->alignment == 0);
3412  	XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), args->len);
3413  
3414  	/* if not file data, insert new block into the reverse map btree */
3415  	if (!xfs_rmap_should_skip_owner_update(&args->oinfo)) {
3416  		error = xfs_rmap_alloc(args->tp, args->agbp, args->pag,
3417  				       args->agbno, args->len, &args->oinfo);
3418  		if (error)
3419  			goto out_drop_perag;
3420  	}
3421  
3422  	if (!args->wasfromfl) {
3423  		error = xfs_alloc_update_counters(args->tp, args->agbp,
3424  						  -((long)(args->len)));
3425  		if (error)
3426  			goto out_drop_perag;
3427  
3428  		ASSERT(!xfs_extent_busy_search(mp, args->pag, args->agbno,
3429  				args->len));
3430  	}
3431  
3432  	xfs_ag_resv_alloc_extent(args->pag, args->resv, args);
3433  
3434  	XFS_STATS_INC(mp, xs_allocx);
3435  	XFS_STATS_ADD(mp, xs_allocb, args->len);
3436  
3437  	trace_xfs_alloc_vextent_finish(args);
3438  
3439  out_drop_perag:
3440  	if (drop_perag && args->pag) {
3441  		xfs_perag_rele(args->pag);
3442  		args->pag = NULL;
3443  	}
3444  	return error;
3445  }
3446  
3447  /*
3448   * Allocate within a single AG only. This uses a best-fit length algorithm so if
3449   * you need an exact sized allocation without locality constraints, this is the
3450   * fastest way to do it.
3451   *
3452   * Caller is expected to hold a perag reference in args->pag.
3453   */
3454  int
xfs_alloc_vextent_this_ag(struct xfs_alloc_arg * args,xfs_agnumber_t agno)3455  xfs_alloc_vextent_this_ag(
3456  	struct xfs_alloc_arg	*args,
3457  	xfs_agnumber_t		agno)
3458  {
3459  	struct xfs_mount	*mp = args->mp;
3460  	xfs_agnumber_t		minimum_agno;
3461  	uint32_t		alloc_flags = 0;
3462  	int			error;
3463  
3464  	ASSERT(args->pag != NULL);
3465  	ASSERT(args->pag->pag_agno == agno);
3466  
3467  	args->agno = agno;
3468  	args->agbno = 0;
3469  
3470  	trace_xfs_alloc_vextent_this_ag(args);
3471  
3472  	error = xfs_alloc_vextent_check_args(args, XFS_AGB_TO_FSB(mp, agno, 0),
3473  			&minimum_agno);
3474  	if (error) {
3475  		if (error == -ENOSPC)
3476  			return 0;
3477  		return error;
3478  	}
3479  
3480  	error = xfs_alloc_vextent_prepare_ag(args, alloc_flags);
3481  	if (!error && args->agbp)
3482  		error = xfs_alloc_ag_vextent_size(args, alloc_flags);
3483  
3484  	return xfs_alloc_vextent_finish(args, minimum_agno, error, false);
3485  }
3486  
3487  /*
3488   * Iterate all AGs trying to allocate an extent starting from @start_ag.
3489   *
3490   * If the incoming allocation type is XFS_ALLOCTYPE_NEAR_BNO, it means the
3491   * allocation attempts in @start_agno have locality information. If we fail to
3492   * allocate in that AG, then we revert to anywhere-in-AG for all the other AGs
3493   * we attempt to allocation in as there is no locality optimisation possible for
3494   * those allocations.
3495   *
3496   * On return, args->pag may be left referenced if we finish before the "all
3497   * failed" return point. The allocation finish still needs the perag, and
3498   * so the caller will release it once they've finished the allocation.
3499   *
3500   * When we wrap the AG iteration at the end of the filesystem, we have to be
3501   * careful not to wrap into AGs below ones we already have locked in the
3502   * transaction if we are doing a blocking iteration. This will result in an
3503   * out-of-order locking of AGFs and hence can cause deadlocks.
3504   */
3505  static int
xfs_alloc_vextent_iterate_ags(struct xfs_alloc_arg * args,xfs_agnumber_t minimum_agno,xfs_agnumber_t start_agno,xfs_agblock_t target_agbno,uint32_t alloc_flags)3506  xfs_alloc_vextent_iterate_ags(
3507  	struct xfs_alloc_arg	*args,
3508  	xfs_agnumber_t		minimum_agno,
3509  	xfs_agnumber_t		start_agno,
3510  	xfs_agblock_t		target_agbno,
3511  	uint32_t		alloc_flags)
3512  {
3513  	struct xfs_mount	*mp = args->mp;
3514  	xfs_agnumber_t		restart_agno = minimum_agno;
3515  	xfs_agnumber_t		agno;
3516  	int			error = 0;
3517  
3518  	if (alloc_flags & XFS_ALLOC_FLAG_TRYLOCK)
3519  		restart_agno = 0;
3520  restart:
3521  	for_each_perag_wrap_range(mp, start_agno, restart_agno,
3522  			mp->m_sb.sb_agcount, agno, args->pag) {
3523  		args->agno = agno;
3524  		error = xfs_alloc_vextent_prepare_ag(args, alloc_flags);
3525  		if (error)
3526  			break;
3527  		if (!args->agbp) {
3528  			trace_xfs_alloc_vextent_loopfailed(args);
3529  			continue;
3530  		}
3531  
3532  		/*
3533  		 * Allocation is supposed to succeed now, so break out of the
3534  		 * loop regardless of whether we succeed or not.
3535  		 */
3536  		if (args->agno == start_agno && target_agbno) {
3537  			args->agbno = target_agbno;
3538  			error = xfs_alloc_ag_vextent_near(args, alloc_flags);
3539  		} else {
3540  			args->agbno = 0;
3541  			error = xfs_alloc_ag_vextent_size(args, alloc_flags);
3542  		}
3543  		break;
3544  	}
3545  	if (error) {
3546  		xfs_perag_rele(args->pag);
3547  		args->pag = NULL;
3548  		return error;
3549  	}
3550  	if (args->agbp)
3551  		return 0;
3552  
3553  	/*
3554  	 * We didn't find an AG we can alloation from. If we were given
3555  	 * constraining flags by the caller, drop them and retry the allocation
3556  	 * without any constraints being set.
3557  	 */
3558  	if (alloc_flags & XFS_ALLOC_FLAG_TRYLOCK) {
3559  		alloc_flags &= ~XFS_ALLOC_FLAG_TRYLOCK;
3560  		restart_agno = minimum_agno;
3561  		goto restart;
3562  	}
3563  
3564  	ASSERT(args->pag == NULL);
3565  	trace_xfs_alloc_vextent_allfailed(args);
3566  	return 0;
3567  }
3568  
3569  /*
3570   * Iterate from the AGs from the start AG to the end of the filesystem, trying
3571   * to allocate blocks. It starts with a near allocation attempt in the initial
3572   * AG, then falls back to anywhere-in-ag after the first AG fails. It will wrap
3573   * back to zero if allowed by previous allocations in this transaction,
3574   * otherwise will wrap back to the start AG and run a second blocking pass to
3575   * the end of the filesystem.
3576   */
3577  int
xfs_alloc_vextent_start_ag(struct xfs_alloc_arg * args,xfs_fsblock_t target)3578  xfs_alloc_vextent_start_ag(
3579  	struct xfs_alloc_arg	*args,
3580  	xfs_fsblock_t		target)
3581  {
3582  	struct xfs_mount	*mp = args->mp;
3583  	xfs_agnumber_t		minimum_agno;
3584  	xfs_agnumber_t		start_agno;
3585  	xfs_agnumber_t		rotorstep = xfs_rotorstep;
3586  	bool			bump_rotor = false;
3587  	uint32_t		alloc_flags = XFS_ALLOC_FLAG_TRYLOCK;
3588  	int			error;
3589  
3590  	ASSERT(args->pag == NULL);
3591  
3592  	args->agno = NULLAGNUMBER;
3593  	args->agbno = NULLAGBLOCK;
3594  
3595  	trace_xfs_alloc_vextent_start_ag(args);
3596  
3597  	error = xfs_alloc_vextent_check_args(args, target, &minimum_agno);
3598  	if (error) {
3599  		if (error == -ENOSPC)
3600  			return 0;
3601  		return error;
3602  	}
3603  
3604  	if ((args->datatype & XFS_ALLOC_INITIAL_USER_DATA) &&
3605  	    xfs_is_inode32(mp)) {
3606  		target = XFS_AGB_TO_FSB(mp,
3607  				((mp->m_agfrotor / rotorstep) %
3608  				mp->m_sb.sb_agcount), 0);
3609  		bump_rotor = 1;
3610  	}
3611  
3612  	start_agno = max(minimum_agno, XFS_FSB_TO_AGNO(mp, target));
3613  	error = xfs_alloc_vextent_iterate_ags(args, minimum_agno, start_agno,
3614  			XFS_FSB_TO_AGBNO(mp, target), alloc_flags);
3615  
3616  	if (bump_rotor) {
3617  		if (args->agno == start_agno)
3618  			mp->m_agfrotor = (mp->m_agfrotor + 1) %
3619  				(mp->m_sb.sb_agcount * rotorstep);
3620  		else
3621  			mp->m_agfrotor = (args->agno * rotorstep + 1) %
3622  				(mp->m_sb.sb_agcount * rotorstep);
3623  	}
3624  
3625  	return xfs_alloc_vextent_finish(args, minimum_agno, error, true);
3626  }
3627  
3628  /*
3629   * Iterate from the agno indicated via @target through to the end of the
3630   * filesystem attempting blocking allocation. This does not wrap or try a second
3631   * pass, so will not recurse into AGs lower than indicated by the target.
3632   */
3633  int
xfs_alloc_vextent_first_ag(struct xfs_alloc_arg * args,xfs_fsblock_t target)3634  xfs_alloc_vextent_first_ag(
3635  	struct xfs_alloc_arg	*args,
3636  	xfs_fsblock_t		target)
3637   {
3638  	struct xfs_mount	*mp = args->mp;
3639  	xfs_agnumber_t		minimum_agno;
3640  	xfs_agnumber_t		start_agno;
3641  	uint32_t		alloc_flags = XFS_ALLOC_FLAG_TRYLOCK;
3642  	int			error;
3643  
3644  	ASSERT(args->pag == NULL);
3645  
3646  	args->agno = NULLAGNUMBER;
3647  	args->agbno = NULLAGBLOCK;
3648  
3649  	trace_xfs_alloc_vextent_first_ag(args);
3650  
3651  	error = xfs_alloc_vextent_check_args(args, target, &minimum_agno);
3652  	if (error) {
3653  		if (error == -ENOSPC)
3654  			return 0;
3655  		return error;
3656  	}
3657  
3658  	start_agno = max(minimum_agno, XFS_FSB_TO_AGNO(mp, target));
3659  	error = xfs_alloc_vextent_iterate_ags(args, minimum_agno, start_agno,
3660  			XFS_FSB_TO_AGBNO(mp, target), alloc_flags);
3661  	return xfs_alloc_vextent_finish(args, minimum_agno, error, true);
3662  }
3663  
3664  /*
3665   * Allocate at the exact block target or fail. Caller is expected to hold a
3666   * perag reference in args->pag.
3667   */
3668  int
xfs_alloc_vextent_exact_bno(struct xfs_alloc_arg * args,xfs_fsblock_t target)3669  xfs_alloc_vextent_exact_bno(
3670  	struct xfs_alloc_arg	*args,
3671  	xfs_fsblock_t		target)
3672  {
3673  	struct xfs_mount	*mp = args->mp;
3674  	xfs_agnumber_t		minimum_agno;
3675  	int			error;
3676  
3677  	ASSERT(args->pag != NULL);
3678  	ASSERT(args->pag->pag_agno == XFS_FSB_TO_AGNO(mp, target));
3679  
3680  	args->agno = XFS_FSB_TO_AGNO(mp, target);
3681  	args->agbno = XFS_FSB_TO_AGBNO(mp, target);
3682  
3683  	trace_xfs_alloc_vextent_exact_bno(args);
3684  
3685  	error = xfs_alloc_vextent_check_args(args, target, &minimum_agno);
3686  	if (error) {
3687  		if (error == -ENOSPC)
3688  			return 0;
3689  		return error;
3690  	}
3691  
3692  	error = xfs_alloc_vextent_prepare_ag(args, 0);
3693  	if (!error && args->agbp)
3694  		error = xfs_alloc_ag_vextent_exact(args);
3695  
3696  	return xfs_alloc_vextent_finish(args, minimum_agno, error, false);
3697  }
3698  
3699  /*
3700   * Allocate an extent as close to the target as possible. If there are not
3701   * viable candidates in the AG, then fail the allocation.
3702   *
3703   * Caller may or may not have a per-ag reference in args->pag.
3704   */
3705  int
xfs_alloc_vextent_near_bno(struct xfs_alloc_arg * args,xfs_fsblock_t target)3706  xfs_alloc_vextent_near_bno(
3707  	struct xfs_alloc_arg	*args,
3708  	xfs_fsblock_t		target)
3709  {
3710  	struct xfs_mount	*mp = args->mp;
3711  	xfs_agnumber_t		minimum_agno;
3712  	bool			needs_perag = args->pag == NULL;
3713  	uint32_t		alloc_flags = 0;
3714  	int			error;
3715  
3716  	if (!needs_perag)
3717  		ASSERT(args->pag->pag_agno == XFS_FSB_TO_AGNO(mp, target));
3718  
3719  	args->agno = XFS_FSB_TO_AGNO(mp, target);
3720  	args->agbno = XFS_FSB_TO_AGBNO(mp, target);
3721  
3722  	trace_xfs_alloc_vextent_near_bno(args);
3723  
3724  	error = xfs_alloc_vextent_check_args(args, target, &minimum_agno);
3725  	if (error) {
3726  		if (error == -ENOSPC)
3727  			return 0;
3728  		return error;
3729  	}
3730  
3731  	if (needs_perag)
3732  		args->pag = xfs_perag_grab(mp, args->agno);
3733  
3734  	error = xfs_alloc_vextent_prepare_ag(args, alloc_flags);
3735  	if (!error && args->agbp)
3736  		error = xfs_alloc_ag_vextent_near(args, alloc_flags);
3737  
3738  	return xfs_alloc_vextent_finish(args, minimum_agno, error, needs_perag);
3739  }
3740  
3741  /* Ensure that the freelist is at full capacity. */
3742  int
xfs_free_extent_fix_freelist(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_buf ** agbp)3743  xfs_free_extent_fix_freelist(
3744  	struct xfs_trans	*tp,
3745  	struct xfs_perag	*pag,
3746  	struct xfs_buf		**agbp)
3747  {
3748  	struct xfs_alloc_arg	args;
3749  	int			error;
3750  
3751  	memset(&args, 0, sizeof(struct xfs_alloc_arg));
3752  	args.tp = tp;
3753  	args.mp = tp->t_mountp;
3754  	args.agno = pag->pag_agno;
3755  	args.pag = pag;
3756  
3757  	/*
3758  	 * validate that the block number is legal - the enables us to detect
3759  	 * and handle a silent filesystem corruption rather than crashing.
3760  	 */
3761  	if (args.agno >= args.mp->m_sb.sb_agcount)
3762  		return -EFSCORRUPTED;
3763  
3764  	error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
3765  	if (error)
3766  		return error;
3767  
3768  	*agbp = args.agbp;
3769  	return 0;
3770  }
3771  
3772  /*
3773   * Free an extent.
3774   * Just break up the extent address and hand off to xfs_free_ag_extent
3775   * after fixing up the freelist.
3776   */
3777  int
__xfs_free_extent(struct xfs_trans * tp,struct xfs_perag * pag,xfs_agblock_t agbno,xfs_extlen_t len,const struct xfs_owner_info * oinfo,enum xfs_ag_resv_type type,bool skip_discard)3778  __xfs_free_extent(
3779  	struct xfs_trans		*tp,
3780  	struct xfs_perag		*pag,
3781  	xfs_agblock_t			agbno,
3782  	xfs_extlen_t			len,
3783  	const struct xfs_owner_info	*oinfo,
3784  	enum xfs_ag_resv_type		type,
3785  	bool				skip_discard)
3786  {
3787  	struct xfs_mount		*mp = tp->t_mountp;
3788  	struct xfs_buf			*agbp;
3789  	struct xfs_agf			*agf;
3790  	int				error;
3791  	unsigned int			busy_flags = 0;
3792  
3793  	ASSERT(len != 0);
3794  	ASSERT(type != XFS_AG_RESV_AGFL);
3795  
3796  	if (XFS_TEST_ERROR(false, mp,
3797  			XFS_ERRTAG_FREE_EXTENT))
3798  		return -EIO;
3799  
3800  	error = xfs_free_extent_fix_freelist(tp, pag, &agbp);
3801  	if (error)
3802  		return error;
3803  	agf = agbp->b_addr;
3804  
3805  	if (XFS_IS_CORRUPT(mp, agbno >= mp->m_sb.sb_agblocks)) {
3806  		error = -EFSCORRUPTED;
3807  		goto err_release;
3808  	}
3809  
3810  	/* validate the extent size is legal now we have the agf locked */
3811  	if (XFS_IS_CORRUPT(mp, agbno + len > be32_to_cpu(agf->agf_length))) {
3812  		error = -EFSCORRUPTED;
3813  		goto err_release;
3814  	}
3815  
3816  	error = xfs_free_ag_extent(tp, agbp, pag->pag_agno, agbno, len, oinfo,
3817  			type);
3818  	if (error)
3819  		goto err_release;
3820  
3821  	if (skip_discard)
3822  		busy_flags |= XFS_EXTENT_BUSY_SKIP_DISCARD;
3823  	xfs_extent_busy_insert(tp, pag, agbno, len, busy_flags);
3824  	return 0;
3825  
3826  err_release:
3827  	xfs_trans_brelse(tp, agbp);
3828  	return error;
3829  }
3830  
3831  struct xfs_alloc_query_range_info {
3832  	xfs_alloc_query_range_fn	fn;
3833  	void				*priv;
3834  };
3835  
3836  /* Format btree record and pass to our callback. */
3837  STATIC int
xfs_alloc_query_range_helper(struct xfs_btree_cur * cur,const union xfs_btree_rec * rec,void * priv)3838  xfs_alloc_query_range_helper(
3839  	struct xfs_btree_cur		*cur,
3840  	const union xfs_btree_rec	*rec,
3841  	void				*priv)
3842  {
3843  	struct xfs_alloc_query_range_info	*query = priv;
3844  	struct xfs_alloc_rec_incore		irec;
3845  	xfs_failaddr_t				fa;
3846  
3847  	xfs_alloc_btrec_to_irec(rec, &irec);
3848  	fa = xfs_alloc_check_irec(cur, &irec);
3849  	if (fa)
3850  		return xfs_alloc_complain_bad_rec(cur, fa, &irec);
3851  
3852  	return query->fn(cur, &irec, query->priv);
3853  }
3854  
3855  /* Find all free space within a given range of blocks. */
3856  int
xfs_alloc_query_range(struct xfs_btree_cur * cur,const struct xfs_alloc_rec_incore * low_rec,const struct xfs_alloc_rec_incore * high_rec,xfs_alloc_query_range_fn fn,void * priv)3857  xfs_alloc_query_range(
3858  	struct xfs_btree_cur			*cur,
3859  	const struct xfs_alloc_rec_incore	*low_rec,
3860  	const struct xfs_alloc_rec_incore	*high_rec,
3861  	xfs_alloc_query_range_fn		fn,
3862  	void					*priv)
3863  {
3864  	union xfs_btree_irec			low_brec = { .a = *low_rec };
3865  	union xfs_btree_irec			high_brec = { .a = *high_rec };
3866  	struct xfs_alloc_query_range_info	query = { .priv = priv, .fn = fn };
3867  
3868  	ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3869  	return xfs_btree_query_range(cur, &low_brec, &high_brec,
3870  			xfs_alloc_query_range_helper, &query);
3871  }
3872  
3873  /* Find all free space records. */
3874  int
xfs_alloc_query_all(struct xfs_btree_cur * cur,xfs_alloc_query_range_fn fn,void * priv)3875  xfs_alloc_query_all(
3876  	struct xfs_btree_cur			*cur,
3877  	xfs_alloc_query_range_fn		fn,
3878  	void					*priv)
3879  {
3880  	struct xfs_alloc_query_range_info	query;
3881  
3882  	ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3883  	query.priv = priv;
3884  	query.fn = fn;
3885  	return xfs_btree_query_all(cur, xfs_alloc_query_range_helper, &query);
3886  }
3887  
3888  /*
3889   * Scan part of the keyspace of the free space and tell us if the area has no
3890   * records, is fully mapped by records, or is partially filled.
3891   */
3892  int
xfs_alloc_has_records(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len,enum xbtree_recpacking * outcome)3893  xfs_alloc_has_records(
3894  	struct xfs_btree_cur	*cur,
3895  	xfs_agblock_t		bno,
3896  	xfs_extlen_t		len,
3897  	enum xbtree_recpacking	*outcome)
3898  {
3899  	union xfs_btree_irec	low;
3900  	union xfs_btree_irec	high;
3901  
3902  	memset(&low, 0, sizeof(low));
3903  	low.a.ar_startblock = bno;
3904  	memset(&high, 0xFF, sizeof(high));
3905  	high.a.ar_startblock = bno + len - 1;
3906  
3907  	return xfs_btree_has_records(cur, &low, &high, NULL, outcome);
3908  }
3909  
3910  /*
3911   * Walk all the blocks in the AGFL.  The @walk_fn can return any negative
3912   * error code or XFS_ITER_*.
3913   */
3914  int
xfs_agfl_walk(struct xfs_mount * mp,struct xfs_agf * agf,struct xfs_buf * agflbp,xfs_agfl_walk_fn walk_fn,void * priv)3915  xfs_agfl_walk(
3916  	struct xfs_mount	*mp,
3917  	struct xfs_agf		*agf,
3918  	struct xfs_buf		*agflbp,
3919  	xfs_agfl_walk_fn	walk_fn,
3920  	void			*priv)
3921  {
3922  	__be32			*agfl_bno;
3923  	unsigned int		i;
3924  	int			error;
3925  
3926  	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
3927  	i = be32_to_cpu(agf->agf_flfirst);
3928  
3929  	/* Nothing to walk in an empty AGFL. */
3930  	if (agf->agf_flcount == cpu_to_be32(0))
3931  		return 0;
3932  
3933  	/* Otherwise, walk from first to last, wrapping as needed. */
3934  	for (;;) {
3935  		error = walk_fn(mp, be32_to_cpu(agfl_bno[i]), priv);
3936  		if (error)
3937  			return error;
3938  		if (i == be32_to_cpu(agf->agf_fllast))
3939  			break;
3940  		if (++i == xfs_agfl_size(mp))
3941  			i = 0;
3942  	}
3943  
3944  	return 0;
3945  }
3946  
3947  int __init
xfs_extfree_intent_init_cache(void)3948  xfs_extfree_intent_init_cache(void)
3949  {
3950  	xfs_extfree_item_cache = kmem_cache_create("xfs_extfree_intent",
3951  			sizeof(struct xfs_extent_free_item),
3952  			0, 0, NULL);
3953  
3954  	return xfs_extfree_item_cache != NULL ? 0 : -ENOMEM;
3955  }
3956  
3957  void
xfs_extfree_intent_destroy_cache(void)3958  xfs_extfree_intent_destroy_cache(void)
3959  {
3960  	kmem_cache_destroy(xfs_extfree_item_cache);
3961  	xfs_extfree_item_cache = NULL;
3962  }
3963