xref: /openbmc/linux/fs/xfs/libxfs/xfs_alloc.c (revision 62975d27)
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_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_btree.h"
17 #include "xfs_rmap.h"
18 #include "xfs_alloc_btree.h"
19 #include "xfs_alloc.h"
20 #include "xfs_extent_busy.h"
21 #include "xfs_errortag.h"
22 #include "xfs_error.h"
23 #include "xfs_trace.h"
24 #include "xfs_trans.h"
25 #include "xfs_buf_item.h"
26 #include "xfs_log.h"
27 #include "xfs_ag_resv.h"
28 #include "xfs_bmap.h"
29 
30 extern kmem_zone_t	*xfs_bmap_free_item_zone;
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 STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
40 STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
41 STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
42 
43 /*
44  * Size of the AGFL.  For CRC-enabled filesystes we steal a couple of slots in
45  * the beginning of the block for a proper header with the location information
46  * and CRC.
47  */
48 unsigned int
49 xfs_agfl_size(
50 	struct xfs_mount	*mp)
51 {
52 	unsigned int		size = mp->m_sb.sb_sectsize;
53 
54 	if (xfs_sb_version_hascrc(&mp->m_sb))
55 		size -= sizeof(struct xfs_agfl);
56 
57 	return size / sizeof(xfs_agblock_t);
58 }
59 
60 unsigned int
61 xfs_refc_block(
62 	struct xfs_mount	*mp)
63 {
64 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
65 		return XFS_RMAP_BLOCK(mp) + 1;
66 	if (xfs_sb_version_hasfinobt(&mp->m_sb))
67 		return XFS_FIBT_BLOCK(mp) + 1;
68 	return XFS_IBT_BLOCK(mp) + 1;
69 }
70 
71 xfs_extlen_t
72 xfs_prealloc_blocks(
73 	struct xfs_mount	*mp)
74 {
75 	if (xfs_sb_version_hasreflink(&mp->m_sb))
76 		return xfs_refc_block(mp) + 1;
77 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
78 		return XFS_RMAP_BLOCK(mp) + 1;
79 	if (xfs_sb_version_hasfinobt(&mp->m_sb))
80 		return XFS_FIBT_BLOCK(mp) + 1;
81 	return XFS_IBT_BLOCK(mp) + 1;
82 }
83 
84 /*
85  * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of
86  * AGF buffer (PV 947395), we place constraints on the relationship among
87  * actual allocations for data blocks, freelist blocks, and potential file data
88  * bmap btree blocks. However, these restrictions may result in no actual space
89  * allocated for a delayed extent, for example, a data block in a certain AG is
90  * allocated but there is no additional block for the additional bmap btree
91  * block due to a split of the bmap btree of the file. The result of this may
92  * lead to an infinite loop when the file gets flushed to disk and all delayed
93  * extents need to be actually allocated. To get around this, we explicitly set
94  * aside a few blocks which will not be reserved in delayed allocation.
95  *
96  * We need to reserve 4 fsbs _per AG_ for the freelist and 4 more to handle a
97  * potential split of the file's bmap btree.
98  */
99 unsigned int
100 xfs_alloc_set_aside(
101 	struct xfs_mount	*mp)
102 {
103 	return mp->m_sb.sb_agcount * (XFS_ALLOC_AGFL_RESERVE + 4);
104 }
105 
106 /*
107  * When deciding how much space to allocate out of an AG, we limit the
108  * allocation maximum size to the size the AG. However, we cannot use all the
109  * blocks in the AG - some are permanently used by metadata. These
110  * blocks are generally:
111  *	- the AG superblock, AGF, AGI and AGFL
112  *	- the AGF (bno and cnt) and AGI btree root blocks, and optionally
113  *	  the AGI free inode and rmap btree root blocks.
114  *	- blocks on the AGFL according to xfs_alloc_set_aside() limits
115  *	- the rmapbt root block
116  *
117  * The AG headers are sector sized, so the amount of space they take up is
118  * dependent on filesystem geometry. The others are all single blocks.
119  */
120 unsigned int
121 xfs_alloc_ag_max_usable(
122 	struct xfs_mount	*mp)
123 {
124 	unsigned int		blocks;
125 
126 	blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */
127 	blocks += XFS_ALLOC_AGFL_RESERVE;
128 	blocks += 3;			/* AGF, AGI btree root blocks */
129 	if (xfs_sb_version_hasfinobt(&mp->m_sb))
130 		blocks++;		/* finobt root block */
131 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
132 		blocks++; 		/* rmap root block */
133 	if (xfs_sb_version_hasreflink(&mp->m_sb))
134 		blocks++;		/* refcount root block */
135 
136 	return mp->m_sb.sb_agblocks - blocks;
137 }
138 
139 /*
140  * Lookup the record equal to [bno, len] in the btree given by cur.
141  */
142 STATIC int				/* error */
143 xfs_alloc_lookup_eq(
144 	struct xfs_btree_cur	*cur,	/* btree cursor */
145 	xfs_agblock_t		bno,	/* starting block of extent */
146 	xfs_extlen_t		len,	/* length of extent */
147 	int			*stat)	/* success/failure */
148 {
149 	int			error;
150 
151 	cur->bc_rec.a.ar_startblock = bno;
152 	cur->bc_rec.a.ar_blockcount = len;
153 	error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
154 	cur->bc_ag.abt.active = (*stat == 1);
155 	return error;
156 }
157 
158 /*
159  * Lookup the first record greater than or equal to [bno, len]
160  * in the btree given by cur.
161  */
162 int				/* error */
163 xfs_alloc_lookup_ge(
164 	struct xfs_btree_cur	*cur,	/* btree cursor */
165 	xfs_agblock_t		bno,	/* starting block of extent */
166 	xfs_extlen_t		len,	/* length of extent */
167 	int			*stat)	/* success/failure */
168 {
169 	int			error;
170 
171 	cur->bc_rec.a.ar_startblock = bno;
172 	cur->bc_rec.a.ar_blockcount = len;
173 	error = xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
174 	cur->bc_ag.abt.active = (*stat == 1);
175 	return error;
176 }
177 
178 /*
179  * Lookup the first record less than or equal to [bno, len]
180  * in the btree given by cur.
181  */
182 int					/* error */
183 xfs_alloc_lookup_le(
184 	struct xfs_btree_cur	*cur,	/* btree cursor */
185 	xfs_agblock_t		bno,	/* starting block of extent */
186 	xfs_extlen_t		len,	/* length of extent */
187 	int			*stat)	/* success/failure */
188 {
189 	int			error;
190 	cur->bc_rec.a.ar_startblock = bno;
191 	cur->bc_rec.a.ar_blockcount = len;
192 	error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
193 	cur->bc_ag.abt.active = (*stat == 1);
194 	return error;
195 }
196 
197 static inline bool
198 xfs_alloc_cur_active(
199 	struct xfs_btree_cur	*cur)
200 {
201 	return cur && cur->bc_ag.abt.active;
202 }
203 
204 /*
205  * Update the record referred to by cur to the value given
206  * by [bno, len].
207  * This either works (return 0) or gets an EFSCORRUPTED error.
208  */
209 STATIC int				/* error */
210 xfs_alloc_update(
211 	struct xfs_btree_cur	*cur,	/* btree cursor */
212 	xfs_agblock_t		bno,	/* starting block of extent */
213 	xfs_extlen_t		len)	/* length of extent */
214 {
215 	union xfs_btree_rec	rec;
216 
217 	rec.alloc.ar_startblock = cpu_to_be32(bno);
218 	rec.alloc.ar_blockcount = cpu_to_be32(len);
219 	return xfs_btree_update(cur, &rec);
220 }
221 
222 /*
223  * Get the data from the pointed-to record.
224  */
225 int					/* error */
226 xfs_alloc_get_rec(
227 	struct xfs_btree_cur	*cur,	/* btree cursor */
228 	xfs_agblock_t		*bno,	/* output: starting block of extent */
229 	xfs_extlen_t		*len,	/* output: length of extent */
230 	int			*stat)	/* output: success/failure */
231 {
232 	struct xfs_mount	*mp = cur->bc_mp;
233 	xfs_agnumber_t		agno = cur->bc_ag.agno;
234 	union xfs_btree_rec	*rec;
235 	int			error;
236 
237 	error = xfs_btree_get_rec(cur, &rec, stat);
238 	if (error || !(*stat))
239 		return error;
240 
241 	*bno = be32_to_cpu(rec->alloc.ar_startblock);
242 	*len = be32_to_cpu(rec->alloc.ar_blockcount);
243 
244 	if (*len == 0)
245 		goto out_bad_rec;
246 
247 	/* check for valid extent range, including overflow */
248 	if (!xfs_verify_agbno(mp, agno, *bno))
249 		goto out_bad_rec;
250 	if (*bno > *bno + *len)
251 		goto out_bad_rec;
252 	if (!xfs_verify_agbno(mp, agno, *bno + *len - 1))
253 		goto out_bad_rec;
254 
255 	return 0;
256 
257 out_bad_rec:
258 	xfs_warn(mp,
259 		"%s Freespace BTree record corruption in AG %d detected!",
260 		cur->bc_btnum == XFS_BTNUM_BNO ? "Block" : "Size", agno);
261 	xfs_warn(mp,
262 		"start block 0x%x block count 0x%x", *bno, *len);
263 	return -EFSCORRUPTED;
264 }
265 
266 /*
267  * Compute aligned version of the found extent.
268  * Takes alignment and min length into account.
269  */
270 STATIC bool
271 xfs_alloc_compute_aligned(
272 	xfs_alloc_arg_t	*args,		/* allocation argument structure */
273 	xfs_agblock_t	foundbno,	/* starting block in found extent */
274 	xfs_extlen_t	foundlen,	/* length in found extent */
275 	xfs_agblock_t	*resbno,	/* result block number */
276 	xfs_extlen_t	*reslen,	/* result length */
277 	unsigned	*busy_gen)
278 {
279 	xfs_agblock_t	bno = foundbno;
280 	xfs_extlen_t	len = foundlen;
281 	xfs_extlen_t	diff;
282 	bool		busy;
283 
284 	/* Trim busy sections out of found extent */
285 	busy = xfs_extent_busy_trim(args, &bno, &len, busy_gen);
286 
287 	/*
288 	 * If we have a largish extent that happens to start before min_agbno,
289 	 * see if we can shift it into range...
290 	 */
291 	if (bno < args->min_agbno && bno + len > args->min_agbno) {
292 		diff = args->min_agbno - bno;
293 		if (len > diff) {
294 			bno += diff;
295 			len -= diff;
296 		}
297 	}
298 
299 	if (args->alignment > 1 && len >= args->minlen) {
300 		xfs_agblock_t	aligned_bno = roundup(bno, args->alignment);
301 
302 		diff = aligned_bno - bno;
303 
304 		*resbno = aligned_bno;
305 		*reslen = diff >= len ? 0 : len - diff;
306 	} else {
307 		*resbno = bno;
308 		*reslen = len;
309 	}
310 
311 	return busy;
312 }
313 
314 /*
315  * Compute best start block and diff for "near" allocations.
316  * freelen >= wantlen already checked by caller.
317  */
318 STATIC xfs_extlen_t			/* difference value (absolute) */
319 xfs_alloc_compute_diff(
320 	xfs_agblock_t	wantbno,	/* target starting block */
321 	xfs_extlen_t	wantlen,	/* target length */
322 	xfs_extlen_t	alignment,	/* target alignment */
323 	int		datatype,	/* are we allocating data? */
324 	xfs_agblock_t	freebno,	/* freespace's starting block */
325 	xfs_extlen_t	freelen,	/* freespace's length */
326 	xfs_agblock_t	*newbnop)	/* result: best start block from free */
327 {
328 	xfs_agblock_t	freeend;	/* end of freespace extent */
329 	xfs_agblock_t	newbno1;	/* return block number */
330 	xfs_agblock_t	newbno2;	/* other new block number */
331 	xfs_extlen_t	newlen1=0;	/* length with newbno1 */
332 	xfs_extlen_t	newlen2=0;	/* length with newbno2 */
333 	xfs_agblock_t	wantend;	/* end of target extent */
334 	bool		userdata = datatype & XFS_ALLOC_USERDATA;
335 
336 	ASSERT(freelen >= wantlen);
337 	freeend = freebno + freelen;
338 	wantend = wantbno + wantlen;
339 	/*
340 	 * We want to allocate from the start of a free extent if it is past
341 	 * the desired block or if we are allocating user data and the free
342 	 * extent is before desired block. The second case is there to allow
343 	 * for contiguous allocation from the remaining free space if the file
344 	 * grows in the short term.
345 	 */
346 	if (freebno >= wantbno || (userdata && freeend < wantend)) {
347 		if ((newbno1 = roundup(freebno, alignment)) >= freeend)
348 			newbno1 = NULLAGBLOCK;
349 	} else if (freeend >= wantend && alignment > 1) {
350 		newbno1 = roundup(wantbno, alignment);
351 		newbno2 = newbno1 - alignment;
352 		if (newbno1 >= freeend)
353 			newbno1 = NULLAGBLOCK;
354 		else
355 			newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
356 		if (newbno2 < freebno)
357 			newbno2 = NULLAGBLOCK;
358 		else
359 			newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
360 		if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
361 			if (newlen1 < newlen2 ||
362 			    (newlen1 == newlen2 &&
363 			     XFS_ABSDIFF(newbno1, wantbno) >
364 			     XFS_ABSDIFF(newbno2, wantbno)))
365 				newbno1 = newbno2;
366 		} else if (newbno2 != NULLAGBLOCK)
367 			newbno1 = newbno2;
368 	} else if (freeend >= wantend) {
369 		newbno1 = wantbno;
370 	} else if (alignment > 1) {
371 		newbno1 = roundup(freeend - wantlen, alignment);
372 		if (newbno1 > freeend - wantlen &&
373 		    newbno1 - alignment >= freebno)
374 			newbno1 -= alignment;
375 		else if (newbno1 >= freeend)
376 			newbno1 = NULLAGBLOCK;
377 	} else
378 		newbno1 = freeend - wantlen;
379 	*newbnop = newbno1;
380 	return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
381 }
382 
383 /*
384  * Fix up the length, based on mod and prod.
385  * len should be k * prod + mod for some k.
386  * If len is too small it is returned unchanged.
387  * If len hits maxlen it is left alone.
388  */
389 STATIC void
390 xfs_alloc_fix_len(
391 	xfs_alloc_arg_t	*args)		/* allocation argument structure */
392 {
393 	xfs_extlen_t	k;
394 	xfs_extlen_t	rlen;
395 
396 	ASSERT(args->mod < args->prod);
397 	rlen = args->len;
398 	ASSERT(rlen >= args->minlen);
399 	ASSERT(rlen <= args->maxlen);
400 	if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
401 	    (args->mod == 0 && rlen < args->prod))
402 		return;
403 	k = rlen % args->prod;
404 	if (k == args->mod)
405 		return;
406 	if (k > args->mod)
407 		rlen = rlen - (k - args->mod);
408 	else
409 		rlen = rlen - args->prod + (args->mod - k);
410 	/* casts to (int) catch length underflows */
411 	if ((int)rlen < (int)args->minlen)
412 		return;
413 	ASSERT(rlen >= args->minlen && rlen <= args->maxlen);
414 	ASSERT(rlen % args->prod == args->mod);
415 	ASSERT(args->pag->pagf_freeblks + args->pag->pagf_flcount >=
416 		rlen + args->minleft);
417 	args->len = rlen;
418 }
419 
420 /*
421  * Update the two btrees, logically removing from freespace the extent
422  * starting at rbno, rlen blocks.  The extent is contained within the
423  * actual (current) free extent fbno for flen blocks.
424  * Flags are passed in indicating whether the cursors are set to the
425  * relevant records.
426  */
427 STATIC int				/* error code */
428 xfs_alloc_fixup_trees(
429 	xfs_btree_cur_t	*cnt_cur,	/* cursor for by-size btree */
430 	xfs_btree_cur_t	*bno_cur,	/* cursor for by-block btree */
431 	xfs_agblock_t	fbno,		/* starting block of free extent */
432 	xfs_extlen_t	flen,		/* length of free extent */
433 	xfs_agblock_t	rbno,		/* starting block of returned extent */
434 	xfs_extlen_t	rlen,		/* length of returned extent */
435 	int		flags)		/* flags, XFSA_FIXUP_... */
436 {
437 	int		error;		/* error code */
438 	int		i;		/* operation results */
439 	xfs_agblock_t	nfbno1;		/* first new free startblock */
440 	xfs_agblock_t	nfbno2;		/* second new free startblock */
441 	xfs_extlen_t	nflen1=0;	/* first new free length */
442 	xfs_extlen_t	nflen2=0;	/* second new free length */
443 	struct xfs_mount *mp;
444 
445 	mp = cnt_cur->bc_mp;
446 
447 	/*
448 	 * Look up the record in the by-size tree if necessary.
449 	 */
450 	if (flags & XFSA_FIXUP_CNT_OK) {
451 #ifdef DEBUG
452 		if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
453 			return error;
454 		if (XFS_IS_CORRUPT(mp,
455 				   i != 1 ||
456 				   nfbno1 != fbno ||
457 				   nflen1 != flen))
458 			return -EFSCORRUPTED;
459 #endif
460 	} else {
461 		if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
462 			return error;
463 		if (XFS_IS_CORRUPT(mp, i != 1))
464 			return -EFSCORRUPTED;
465 	}
466 	/*
467 	 * Look up the record in the by-block tree if necessary.
468 	 */
469 	if (flags & XFSA_FIXUP_BNO_OK) {
470 #ifdef DEBUG
471 		if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
472 			return error;
473 		if (XFS_IS_CORRUPT(mp,
474 				   i != 1 ||
475 				   nfbno1 != fbno ||
476 				   nflen1 != flen))
477 			return -EFSCORRUPTED;
478 #endif
479 	} else {
480 		if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
481 			return error;
482 		if (XFS_IS_CORRUPT(mp, i != 1))
483 			return -EFSCORRUPTED;
484 	}
485 
486 #ifdef DEBUG
487 	if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
488 		struct xfs_btree_block	*bnoblock;
489 		struct xfs_btree_block	*cntblock;
490 
491 		bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]);
492 		cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]);
493 
494 		if (XFS_IS_CORRUPT(mp,
495 				   bnoblock->bb_numrecs !=
496 				   cntblock->bb_numrecs))
497 			return -EFSCORRUPTED;
498 	}
499 #endif
500 
501 	/*
502 	 * Deal with all four cases: the allocated record is contained
503 	 * within the freespace record, so we can have new freespace
504 	 * at either (or both) end, or no freespace remaining.
505 	 */
506 	if (rbno == fbno && rlen == flen)
507 		nfbno1 = nfbno2 = NULLAGBLOCK;
508 	else if (rbno == fbno) {
509 		nfbno1 = rbno + rlen;
510 		nflen1 = flen - rlen;
511 		nfbno2 = NULLAGBLOCK;
512 	} else if (rbno + rlen == fbno + flen) {
513 		nfbno1 = fbno;
514 		nflen1 = flen - rlen;
515 		nfbno2 = NULLAGBLOCK;
516 	} else {
517 		nfbno1 = fbno;
518 		nflen1 = rbno - fbno;
519 		nfbno2 = rbno + rlen;
520 		nflen2 = (fbno + flen) - nfbno2;
521 	}
522 	/*
523 	 * Delete the entry from the by-size btree.
524 	 */
525 	if ((error = xfs_btree_delete(cnt_cur, &i)))
526 		return error;
527 	if (XFS_IS_CORRUPT(mp, i != 1))
528 		return -EFSCORRUPTED;
529 	/*
530 	 * Add new by-size btree entry(s).
531 	 */
532 	if (nfbno1 != NULLAGBLOCK) {
533 		if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
534 			return error;
535 		if (XFS_IS_CORRUPT(mp, i != 0))
536 			return -EFSCORRUPTED;
537 		if ((error = xfs_btree_insert(cnt_cur, &i)))
538 			return error;
539 		if (XFS_IS_CORRUPT(mp, i != 1))
540 			return -EFSCORRUPTED;
541 	}
542 	if (nfbno2 != NULLAGBLOCK) {
543 		if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
544 			return error;
545 		if (XFS_IS_CORRUPT(mp, i != 0))
546 			return -EFSCORRUPTED;
547 		if ((error = xfs_btree_insert(cnt_cur, &i)))
548 			return error;
549 		if (XFS_IS_CORRUPT(mp, i != 1))
550 			return -EFSCORRUPTED;
551 	}
552 	/*
553 	 * Fix up the by-block btree entry(s).
554 	 */
555 	if (nfbno1 == NULLAGBLOCK) {
556 		/*
557 		 * No remaining freespace, just delete the by-block tree entry.
558 		 */
559 		if ((error = xfs_btree_delete(bno_cur, &i)))
560 			return error;
561 		if (XFS_IS_CORRUPT(mp, i != 1))
562 			return -EFSCORRUPTED;
563 	} else {
564 		/*
565 		 * Update the by-block entry to start later|be shorter.
566 		 */
567 		if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
568 			return error;
569 	}
570 	if (nfbno2 != NULLAGBLOCK) {
571 		/*
572 		 * 2 resulting free entries, need to add one.
573 		 */
574 		if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
575 			return error;
576 		if (XFS_IS_CORRUPT(mp, i != 0))
577 			return -EFSCORRUPTED;
578 		if ((error = xfs_btree_insert(bno_cur, &i)))
579 			return error;
580 		if (XFS_IS_CORRUPT(mp, i != 1))
581 			return -EFSCORRUPTED;
582 	}
583 	return 0;
584 }
585 
586 static xfs_failaddr_t
587 xfs_agfl_verify(
588 	struct xfs_buf	*bp)
589 {
590 	struct xfs_mount *mp = bp->b_mount;
591 	struct xfs_agfl	*agfl = XFS_BUF_TO_AGFL(bp);
592 	__be32		*agfl_bno = xfs_buf_to_agfl_bno(bp);
593 	int		i;
594 
595 	/*
596 	 * There is no verification of non-crc AGFLs because mkfs does not
597 	 * initialise the AGFL to zero or NULL. Hence the only valid part of the
598 	 * AGFL is what the AGF says is active. We can't get to the AGF, so we
599 	 * can't verify just those entries are valid.
600 	 */
601 	if (!xfs_sb_version_hascrc(&mp->m_sb))
602 		return NULL;
603 
604 	if (!xfs_verify_magic(bp, agfl->agfl_magicnum))
605 		return __this_address;
606 	if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid))
607 		return __this_address;
608 	/*
609 	 * during growfs operations, the perag is not fully initialised,
610 	 * so we can't use it for any useful checking. growfs ensures we can't
611 	 * use it by using uncached buffers that don't have the perag attached
612 	 * so we can detect and avoid this problem.
613 	 */
614 	if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno)
615 		return __this_address;
616 
617 	for (i = 0; i < xfs_agfl_size(mp); i++) {
618 		if (be32_to_cpu(agfl_bno[i]) != NULLAGBLOCK &&
619 		    be32_to_cpu(agfl_bno[i]) >= mp->m_sb.sb_agblocks)
620 			return __this_address;
621 	}
622 
623 	if (!xfs_log_check_lsn(mp, be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)))
624 		return __this_address;
625 	return NULL;
626 }
627 
628 static void
629 xfs_agfl_read_verify(
630 	struct xfs_buf	*bp)
631 {
632 	struct xfs_mount *mp = bp->b_mount;
633 	xfs_failaddr_t	fa;
634 
635 	/*
636 	 * There is no verification of non-crc AGFLs because mkfs does not
637 	 * initialise the AGFL to zero or NULL. Hence the only valid part of the
638 	 * AGFL is what the AGF says is active. We can't get to the AGF, so we
639 	 * can't verify just those entries are valid.
640 	 */
641 	if (!xfs_sb_version_hascrc(&mp->m_sb))
642 		return;
643 
644 	if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF))
645 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
646 	else {
647 		fa = xfs_agfl_verify(bp);
648 		if (fa)
649 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
650 	}
651 }
652 
653 static void
654 xfs_agfl_write_verify(
655 	struct xfs_buf	*bp)
656 {
657 	struct xfs_mount	*mp = bp->b_mount;
658 	struct xfs_buf_log_item	*bip = bp->b_log_item;
659 	xfs_failaddr_t		fa;
660 
661 	/* no verification of non-crc AGFLs */
662 	if (!xfs_sb_version_hascrc(&mp->m_sb))
663 		return;
664 
665 	fa = xfs_agfl_verify(bp);
666 	if (fa) {
667 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
668 		return;
669 	}
670 
671 	if (bip)
672 		XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn);
673 
674 	xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF);
675 }
676 
677 const struct xfs_buf_ops xfs_agfl_buf_ops = {
678 	.name = "xfs_agfl",
679 	.magic = { cpu_to_be32(XFS_AGFL_MAGIC), cpu_to_be32(XFS_AGFL_MAGIC) },
680 	.verify_read = xfs_agfl_read_verify,
681 	.verify_write = xfs_agfl_write_verify,
682 	.verify_struct = xfs_agfl_verify,
683 };
684 
685 /*
686  * Read in the allocation group free block array.
687  */
688 int					/* error */
689 xfs_alloc_read_agfl(
690 	xfs_mount_t	*mp,		/* mount point structure */
691 	xfs_trans_t	*tp,		/* transaction pointer */
692 	xfs_agnumber_t	agno,		/* allocation group number */
693 	xfs_buf_t	**bpp)		/* buffer for the ag free block array */
694 {
695 	xfs_buf_t	*bp;		/* return value */
696 	int		error;
697 
698 	ASSERT(agno != NULLAGNUMBER);
699 	error = xfs_trans_read_buf(
700 			mp, tp, mp->m_ddev_targp,
701 			XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
702 			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops);
703 	if (error)
704 		return error;
705 	xfs_buf_set_ref(bp, XFS_AGFL_REF);
706 	*bpp = bp;
707 	return 0;
708 }
709 
710 STATIC int
711 xfs_alloc_update_counters(
712 	struct xfs_trans	*tp,
713 	struct xfs_perag	*pag,
714 	struct xfs_buf		*agbp,
715 	long			len)
716 {
717 	struct xfs_agf		*agf = agbp->b_addr;
718 
719 	pag->pagf_freeblks += len;
720 	be32_add_cpu(&agf->agf_freeblks, len);
721 
722 	xfs_trans_agblocks_delta(tp, len);
723 	if (unlikely(be32_to_cpu(agf->agf_freeblks) >
724 		     be32_to_cpu(agf->agf_length))) {
725 		xfs_buf_mark_corrupt(agbp);
726 		return -EFSCORRUPTED;
727 	}
728 
729 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
730 	return 0;
731 }
732 
733 /*
734  * Block allocation algorithm and data structures.
735  */
736 struct xfs_alloc_cur {
737 	struct xfs_btree_cur		*cnt;	/* btree cursors */
738 	struct xfs_btree_cur		*bnolt;
739 	struct xfs_btree_cur		*bnogt;
740 	xfs_extlen_t			cur_len;/* current search length */
741 	xfs_agblock_t			rec_bno;/* extent startblock */
742 	xfs_extlen_t			rec_len;/* extent length */
743 	xfs_agblock_t			bno;	/* alloc bno */
744 	xfs_extlen_t			len;	/* alloc len */
745 	xfs_extlen_t			diff;	/* diff from search bno */
746 	unsigned int			busy_gen;/* busy state */
747 	bool				busy;
748 };
749 
750 /*
751  * Set up cursors, etc. in the extent allocation cursor. This function can be
752  * called multiple times to reset an initialized structure without having to
753  * reallocate cursors.
754  */
755 static int
756 xfs_alloc_cur_setup(
757 	struct xfs_alloc_arg	*args,
758 	struct xfs_alloc_cur	*acur)
759 {
760 	int			error;
761 	int			i;
762 
763 	ASSERT(args->alignment == 1 || args->type != XFS_ALLOCTYPE_THIS_BNO);
764 
765 	acur->cur_len = args->maxlen;
766 	acur->rec_bno = 0;
767 	acur->rec_len = 0;
768 	acur->bno = 0;
769 	acur->len = 0;
770 	acur->diff = -1;
771 	acur->busy = false;
772 	acur->busy_gen = 0;
773 
774 	/*
775 	 * Perform an initial cntbt lookup to check for availability of maxlen
776 	 * extents. If this fails, we'll return -ENOSPC to signal the caller to
777 	 * attempt a small allocation.
778 	 */
779 	if (!acur->cnt)
780 		acur->cnt = xfs_allocbt_init_cursor(args->mp, args->tp,
781 					args->agbp, args->agno, XFS_BTNUM_CNT);
782 	error = xfs_alloc_lookup_ge(acur->cnt, 0, args->maxlen, &i);
783 	if (error)
784 		return error;
785 
786 	/*
787 	 * Allocate the bnobt left and right search cursors.
788 	 */
789 	if (!acur->bnolt)
790 		acur->bnolt = xfs_allocbt_init_cursor(args->mp, args->tp,
791 					args->agbp, args->agno, XFS_BTNUM_BNO);
792 	if (!acur->bnogt)
793 		acur->bnogt = xfs_allocbt_init_cursor(args->mp, args->tp,
794 					args->agbp, args->agno, XFS_BTNUM_BNO);
795 	return i == 1 ? 0 : -ENOSPC;
796 }
797 
798 static void
799 xfs_alloc_cur_close(
800 	struct xfs_alloc_cur	*acur,
801 	bool			error)
802 {
803 	int			cur_error = XFS_BTREE_NOERROR;
804 
805 	if (error)
806 		cur_error = XFS_BTREE_ERROR;
807 
808 	if (acur->cnt)
809 		xfs_btree_del_cursor(acur->cnt, cur_error);
810 	if (acur->bnolt)
811 		xfs_btree_del_cursor(acur->bnolt, cur_error);
812 	if (acur->bnogt)
813 		xfs_btree_del_cursor(acur->bnogt, cur_error);
814 	acur->cnt = acur->bnolt = acur->bnogt = NULL;
815 }
816 
817 /*
818  * Check an extent for allocation and track the best available candidate in the
819  * allocation structure. The cursor is deactivated if it has entered an out of
820  * range state based on allocation arguments. Optionally return the extent
821  * extent geometry and allocation status if requested by the caller.
822  */
823 static int
824 xfs_alloc_cur_check(
825 	struct xfs_alloc_arg	*args,
826 	struct xfs_alloc_cur	*acur,
827 	struct xfs_btree_cur	*cur,
828 	int			*new)
829 {
830 	int			error, i;
831 	xfs_agblock_t		bno, bnoa, bnew;
832 	xfs_extlen_t		len, lena, diff = -1;
833 	bool			busy;
834 	unsigned		busy_gen = 0;
835 	bool			deactivate = false;
836 	bool			isbnobt = cur->bc_btnum == XFS_BTNUM_BNO;
837 
838 	*new = 0;
839 
840 	error = xfs_alloc_get_rec(cur, &bno, &len, &i);
841 	if (error)
842 		return error;
843 	if (XFS_IS_CORRUPT(args->mp, i != 1))
844 		return -EFSCORRUPTED;
845 
846 	/*
847 	 * Check minlen and deactivate a cntbt cursor if out of acceptable size
848 	 * range (i.e., walking backwards looking for a minlen extent).
849 	 */
850 	if (len < args->minlen) {
851 		deactivate = !isbnobt;
852 		goto out;
853 	}
854 
855 	busy = xfs_alloc_compute_aligned(args, bno, len, &bnoa, &lena,
856 					 &busy_gen);
857 	acur->busy |= busy;
858 	if (busy)
859 		acur->busy_gen = busy_gen;
860 	/* deactivate a bnobt cursor outside of locality range */
861 	if (bnoa < args->min_agbno || bnoa > args->max_agbno) {
862 		deactivate = isbnobt;
863 		goto out;
864 	}
865 	if (lena < args->minlen)
866 		goto out;
867 
868 	args->len = XFS_EXTLEN_MIN(lena, args->maxlen);
869 	xfs_alloc_fix_len(args);
870 	ASSERT(args->len >= args->minlen);
871 	if (args->len < acur->len)
872 		goto out;
873 
874 	/*
875 	 * We have an aligned record that satisfies minlen and beats or matches
876 	 * the candidate extent size. Compare locality for near allocation mode.
877 	 */
878 	ASSERT(args->type == XFS_ALLOCTYPE_NEAR_BNO);
879 	diff = xfs_alloc_compute_diff(args->agbno, args->len,
880 				      args->alignment, args->datatype,
881 				      bnoa, lena, &bnew);
882 	if (bnew == NULLAGBLOCK)
883 		goto out;
884 
885 	/*
886 	 * Deactivate a bnobt cursor with worse locality than the current best.
887 	 */
888 	if (diff > acur->diff) {
889 		deactivate = isbnobt;
890 		goto out;
891 	}
892 
893 	ASSERT(args->len > acur->len ||
894 	       (args->len == acur->len && diff <= acur->diff));
895 	acur->rec_bno = bno;
896 	acur->rec_len = len;
897 	acur->bno = bnew;
898 	acur->len = args->len;
899 	acur->diff = diff;
900 	*new = 1;
901 
902 	/*
903 	 * We're done if we found a perfect allocation. This only deactivates
904 	 * the current cursor, but this is just an optimization to terminate a
905 	 * cntbt search that otherwise runs to the edge of the tree.
906 	 */
907 	if (acur->diff == 0 && acur->len == args->maxlen)
908 		deactivate = true;
909 out:
910 	if (deactivate)
911 		cur->bc_ag.abt.active = false;
912 	trace_xfs_alloc_cur_check(args->mp, cur->bc_btnum, bno, len, diff,
913 				  *new);
914 	return 0;
915 }
916 
917 /*
918  * Complete an allocation of a candidate extent. Remove the extent from both
919  * trees and update the args structure.
920  */
921 STATIC int
922 xfs_alloc_cur_finish(
923 	struct xfs_alloc_arg	*args,
924 	struct xfs_alloc_cur	*acur)
925 {
926 	struct xfs_agf __maybe_unused *agf = args->agbp->b_addr;
927 	int			error;
928 
929 	ASSERT(acur->cnt && acur->bnolt);
930 	ASSERT(acur->bno >= acur->rec_bno);
931 	ASSERT(acur->bno + acur->len <= acur->rec_bno + acur->rec_len);
932 	ASSERT(acur->rec_bno + acur->rec_len <= be32_to_cpu(agf->agf_length));
933 
934 	error = xfs_alloc_fixup_trees(acur->cnt, acur->bnolt, acur->rec_bno,
935 				      acur->rec_len, acur->bno, acur->len, 0);
936 	if (error)
937 		return error;
938 
939 	args->agbno = acur->bno;
940 	args->len = acur->len;
941 	args->wasfromfl = 0;
942 
943 	trace_xfs_alloc_cur(args);
944 	return 0;
945 }
946 
947 /*
948  * Locality allocation lookup algorithm. This expects a cntbt cursor and uses
949  * bno optimized lookup to search for extents with ideal size and locality.
950  */
951 STATIC int
952 xfs_alloc_cntbt_iter(
953 	struct xfs_alloc_arg		*args,
954 	struct xfs_alloc_cur		*acur)
955 {
956 	struct xfs_btree_cur	*cur = acur->cnt;
957 	xfs_agblock_t		bno;
958 	xfs_extlen_t		len, cur_len;
959 	int			error;
960 	int			i;
961 
962 	if (!xfs_alloc_cur_active(cur))
963 		return 0;
964 
965 	/* locality optimized lookup */
966 	cur_len = acur->cur_len;
967 	error = xfs_alloc_lookup_ge(cur, args->agbno, cur_len, &i);
968 	if (error)
969 		return error;
970 	if (i == 0)
971 		return 0;
972 	error = xfs_alloc_get_rec(cur, &bno, &len, &i);
973 	if (error)
974 		return error;
975 
976 	/* check the current record and update search length from it */
977 	error = xfs_alloc_cur_check(args, acur, cur, &i);
978 	if (error)
979 		return error;
980 	ASSERT(len >= acur->cur_len);
981 	acur->cur_len = len;
982 
983 	/*
984 	 * We looked up the first record >= [agbno, len] above. The agbno is a
985 	 * secondary key and so the current record may lie just before or after
986 	 * agbno. If it is past agbno, check the previous record too so long as
987 	 * the length matches as it may be closer. Don't check a smaller record
988 	 * because that could deactivate our cursor.
989 	 */
990 	if (bno > args->agbno) {
991 		error = xfs_btree_decrement(cur, 0, &i);
992 		if (!error && i) {
993 			error = xfs_alloc_get_rec(cur, &bno, &len, &i);
994 			if (!error && i && len == acur->cur_len)
995 				error = xfs_alloc_cur_check(args, acur, cur,
996 							    &i);
997 		}
998 		if (error)
999 			return error;
1000 	}
1001 
1002 	/*
1003 	 * Increment the search key until we find at least one allocation
1004 	 * candidate or if the extent we found was larger. Otherwise, double the
1005 	 * search key to optimize the search. Efficiency is more important here
1006 	 * than absolute best locality.
1007 	 */
1008 	cur_len <<= 1;
1009 	if (!acur->len || acur->cur_len >= cur_len)
1010 		acur->cur_len++;
1011 	else
1012 		acur->cur_len = cur_len;
1013 
1014 	return error;
1015 }
1016 
1017 /*
1018  * Deal with the case where only small freespaces remain. Either return the
1019  * contents of the last freespace record, or allocate space from the freelist if
1020  * there is nothing in the tree.
1021  */
1022 STATIC int			/* error */
1023 xfs_alloc_ag_vextent_small(
1024 	struct xfs_alloc_arg	*args,	/* allocation argument structure */
1025 	struct xfs_btree_cur	*ccur,	/* optional by-size cursor */
1026 	xfs_agblock_t		*fbnop,	/* result block number */
1027 	xfs_extlen_t		*flenp,	/* result length */
1028 	int			*stat)	/* status: 0-freelist, 1-normal/none */
1029 {
1030 	struct xfs_agf		*agf = args->agbp->b_addr;
1031 	int			error = 0;
1032 	xfs_agblock_t		fbno = NULLAGBLOCK;
1033 	xfs_extlen_t		flen = 0;
1034 	int			i = 0;
1035 
1036 	/*
1037 	 * If a cntbt cursor is provided, try to allocate the largest record in
1038 	 * the tree. Try the AGFL if the cntbt is empty, otherwise fail the
1039 	 * allocation. Make sure to respect minleft even when pulling from the
1040 	 * freelist.
1041 	 */
1042 	if (ccur)
1043 		error = xfs_btree_decrement(ccur, 0, &i);
1044 	if (error)
1045 		goto error;
1046 	if (i) {
1047 		error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i);
1048 		if (error)
1049 			goto error;
1050 		if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1051 			error = -EFSCORRUPTED;
1052 			goto error;
1053 		}
1054 		goto out;
1055 	}
1056 
1057 	if (args->minlen != 1 || args->alignment != 1 ||
1058 	    args->resv == XFS_AG_RESV_AGFL ||
1059 	    be32_to_cpu(agf->agf_flcount) <= args->minleft)
1060 		goto out;
1061 
1062 	error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1063 	if (error)
1064 		goto error;
1065 	if (fbno == NULLAGBLOCK)
1066 		goto out;
1067 
1068 	xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1,
1069 			      (args->datatype & XFS_ALLOC_NOBUSY));
1070 
1071 	if (args->datatype & XFS_ALLOC_USERDATA) {
1072 		struct xfs_buf	*bp;
1073 
1074 		error = xfs_trans_get_buf(args->tp, args->mp->m_ddev_targp,
1075 				XFS_AGB_TO_DADDR(args->mp, args->agno, fbno),
1076 				args->mp->m_bsize, 0, &bp);
1077 		if (error)
1078 			goto error;
1079 		xfs_trans_binval(args->tp, bp);
1080 	}
1081 	*fbnop = args->agbno = fbno;
1082 	*flenp = args->len = 1;
1083 	if (XFS_IS_CORRUPT(args->mp, fbno >= be32_to_cpu(agf->agf_length))) {
1084 		error = -EFSCORRUPTED;
1085 		goto error;
1086 	}
1087 	args->wasfromfl = 1;
1088 	trace_xfs_alloc_small_freelist(args);
1089 
1090 	/*
1091 	 * If we're feeding an AGFL block to something that doesn't live in the
1092 	 * free space, we need to clear out the OWN_AG rmap.
1093 	 */
1094 	error = xfs_rmap_free(args->tp, args->agbp, args->agno, fbno, 1,
1095 			      &XFS_RMAP_OINFO_AG);
1096 	if (error)
1097 		goto error;
1098 
1099 	*stat = 0;
1100 	return 0;
1101 
1102 out:
1103 	/*
1104 	 * Can't do the allocation, give up.
1105 	 */
1106 	if (flen < args->minlen) {
1107 		args->agbno = NULLAGBLOCK;
1108 		trace_xfs_alloc_small_notenough(args);
1109 		flen = 0;
1110 	}
1111 	*fbnop = fbno;
1112 	*flenp = flen;
1113 	*stat = 1;
1114 	trace_xfs_alloc_small_done(args);
1115 	return 0;
1116 
1117 error:
1118 	trace_xfs_alloc_small_error(args);
1119 	return error;
1120 }
1121 
1122 /*
1123  * Allocate a variable extent in the allocation group agno.
1124  * Type and bno are used to determine where in the allocation group the
1125  * extent will start.
1126  * Extent's length (returned in *len) will be between minlen and maxlen,
1127  * and of the form k * prod + mod unless there's nothing that large.
1128  * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1129  */
1130 STATIC int			/* error */
1131 xfs_alloc_ag_vextent(
1132 	xfs_alloc_arg_t	*args)	/* argument structure for allocation */
1133 {
1134 	int		error=0;
1135 
1136 	ASSERT(args->minlen > 0);
1137 	ASSERT(args->maxlen > 0);
1138 	ASSERT(args->minlen <= args->maxlen);
1139 	ASSERT(args->mod < args->prod);
1140 	ASSERT(args->alignment > 0);
1141 
1142 	/*
1143 	 * Branch to correct routine based on the type.
1144 	 */
1145 	args->wasfromfl = 0;
1146 	switch (args->type) {
1147 	case XFS_ALLOCTYPE_THIS_AG:
1148 		error = xfs_alloc_ag_vextent_size(args);
1149 		break;
1150 	case XFS_ALLOCTYPE_NEAR_BNO:
1151 		error = xfs_alloc_ag_vextent_near(args);
1152 		break;
1153 	case XFS_ALLOCTYPE_THIS_BNO:
1154 		error = xfs_alloc_ag_vextent_exact(args);
1155 		break;
1156 	default:
1157 		ASSERT(0);
1158 		/* NOTREACHED */
1159 	}
1160 
1161 	if (error || args->agbno == NULLAGBLOCK)
1162 		return error;
1163 
1164 	ASSERT(args->len >= args->minlen);
1165 	ASSERT(args->len <= args->maxlen);
1166 	ASSERT(!args->wasfromfl || args->resv != XFS_AG_RESV_AGFL);
1167 	ASSERT(args->agbno % args->alignment == 0);
1168 
1169 	/* if not file data, insert new block into the reverse map btree */
1170 	if (!xfs_rmap_should_skip_owner_update(&args->oinfo)) {
1171 		error = xfs_rmap_alloc(args->tp, args->agbp, args->agno,
1172 				       args->agbno, args->len, &args->oinfo);
1173 		if (error)
1174 			return error;
1175 	}
1176 
1177 	if (!args->wasfromfl) {
1178 		error = xfs_alloc_update_counters(args->tp, args->pag,
1179 						  args->agbp,
1180 						  -((long)(args->len)));
1181 		if (error)
1182 			return error;
1183 
1184 		ASSERT(!xfs_extent_busy_search(args->mp, args->agno,
1185 					      args->agbno, args->len));
1186 	}
1187 
1188 	xfs_ag_resv_alloc_extent(args->pag, args->resv, args);
1189 
1190 	XFS_STATS_INC(args->mp, xs_allocx);
1191 	XFS_STATS_ADD(args->mp, xs_allocb, args->len);
1192 	return error;
1193 }
1194 
1195 /*
1196  * Allocate a variable extent at exactly agno/bno.
1197  * Extent's length (returned in *len) will be between minlen and maxlen,
1198  * and of the form k * prod + mod unless there's nothing that large.
1199  * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
1200  */
1201 STATIC int			/* error */
1202 xfs_alloc_ag_vextent_exact(
1203 	xfs_alloc_arg_t	*args)	/* allocation argument structure */
1204 {
1205 	struct xfs_agf __maybe_unused *agf = args->agbp->b_addr;
1206 	xfs_btree_cur_t	*bno_cur;/* by block-number btree cursor */
1207 	xfs_btree_cur_t	*cnt_cur;/* by count btree cursor */
1208 	int		error;
1209 	xfs_agblock_t	fbno;	/* start block of found extent */
1210 	xfs_extlen_t	flen;	/* length of found extent */
1211 	xfs_agblock_t	tbno;	/* start block of busy extent */
1212 	xfs_extlen_t	tlen;	/* length of busy extent */
1213 	xfs_agblock_t	tend;	/* end block of busy extent */
1214 	int		i;	/* success/failure of operation */
1215 	unsigned	busy_gen;
1216 
1217 	ASSERT(args->alignment == 1);
1218 
1219 	/*
1220 	 * Allocate/initialize a cursor for the by-number freespace btree.
1221 	 */
1222 	bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1223 					  args->agno, XFS_BTNUM_BNO);
1224 
1225 	/*
1226 	 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
1227 	 * Look for the closest free block <= bno, it must contain bno
1228 	 * if any free block does.
1229 	 */
1230 	error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
1231 	if (error)
1232 		goto error0;
1233 	if (!i)
1234 		goto not_found;
1235 
1236 	/*
1237 	 * Grab the freespace record.
1238 	 */
1239 	error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
1240 	if (error)
1241 		goto error0;
1242 	if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1243 		error = -EFSCORRUPTED;
1244 		goto error0;
1245 	}
1246 	ASSERT(fbno <= args->agbno);
1247 
1248 	/*
1249 	 * Check for overlapping busy extents.
1250 	 */
1251 	tbno = fbno;
1252 	tlen = flen;
1253 	xfs_extent_busy_trim(args, &tbno, &tlen, &busy_gen);
1254 
1255 	/*
1256 	 * Give up if the start of the extent is busy, or the freespace isn't
1257 	 * long enough for the minimum request.
1258 	 */
1259 	if (tbno > args->agbno)
1260 		goto not_found;
1261 	if (tlen < args->minlen)
1262 		goto not_found;
1263 	tend = tbno + tlen;
1264 	if (tend < args->agbno + args->minlen)
1265 		goto not_found;
1266 
1267 	/*
1268 	 * End of extent will be smaller of the freespace end and the
1269 	 * maximal requested end.
1270 	 *
1271 	 * Fix the length according to mod and prod if given.
1272 	 */
1273 	args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
1274 						- args->agbno;
1275 	xfs_alloc_fix_len(args);
1276 	ASSERT(args->agbno + args->len <= tend);
1277 
1278 	/*
1279 	 * We are allocating agbno for args->len
1280 	 * Allocate/initialize a cursor for the by-size btree.
1281 	 */
1282 	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1283 		args->agno, XFS_BTNUM_CNT);
1284 	ASSERT(args->agbno + args->len <= be32_to_cpu(agf->agf_length));
1285 	error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
1286 				      args->len, XFSA_FIXUP_BNO_OK);
1287 	if (error) {
1288 		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1289 		goto error0;
1290 	}
1291 
1292 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1293 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1294 
1295 	args->wasfromfl = 0;
1296 	trace_xfs_alloc_exact_done(args);
1297 	return 0;
1298 
1299 not_found:
1300 	/* Didn't find it, return null. */
1301 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1302 	args->agbno = NULLAGBLOCK;
1303 	trace_xfs_alloc_exact_notfound(args);
1304 	return 0;
1305 
1306 error0:
1307 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1308 	trace_xfs_alloc_exact_error(args);
1309 	return error;
1310 }
1311 
1312 /*
1313  * Search a given number of btree records in a given direction. Check each
1314  * record against the good extent we've already found.
1315  */
1316 STATIC int
1317 xfs_alloc_walk_iter(
1318 	struct xfs_alloc_arg	*args,
1319 	struct xfs_alloc_cur	*acur,
1320 	struct xfs_btree_cur	*cur,
1321 	bool			increment,
1322 	bool			find_one, /* quit on first candidate */
1323 	int			count,    /* rec count (-1 for infinite) */
1324 	int			*stat)
1325 {
1326 	int			error;
1327 	int			i;
1328 
1329 	*stat = 0;
1330 
1331 	/*
1332 	 * Search so long as the cursor is active or we find a better extent.
1333 	 * The cursor is deactivated if it extends beyond the range of the
1334 	 * current allocation candidate.
1335 	 */
1336 	while (xfs_alloc_cur_active(cur) && count) {
1337 		error = xfs_alloc_cur_check(args, acur, cur, &i);
1338 		if (error)
1339 			return error;
1340 		if (i == 1) {
1341 			*stat = 1;
1342 			if (find_one)
1343 				break;
1344 		}
1345 		if (!xfs_alloc_cur_active(cur))
1346 			break;
1347 
1348 		if (increment)
1349 			error = xfs_btree_increment(cur, 0, &i);
1350 		else
1351 			error = xfs_btree_decrement(cur, 0, &i);
1352 		if (error)
1353 			return error;
1354 		if (i == 0)
1355 			cur->bc_ag.abt.active = false;
1356 
1357 		if (count > 0)
1358 			count--;
1359 	}
1360 
1361 	return 0;
1362 }
1363 
1364 /*
1365  * Search the by-bno and by-size btrees in parallel in search of an extent with
1366  * ideal locality based on the NEAR mode ->agbno locality hint.
1367  */
1368 STATIC int
1369 xfs_alloc_ag_vextent_locality(
1370 	struct xfs_alloc_arg	*args,
1371 	struct xfs_alloc_cur	*acur,
1372 	int			*stat)
1373 {
1374 	struct xfs_btree_cur	*fbcur = NULL;
1375 	int			error;
1376 	int			i;
1377 	bool			fbinc;
1378 
1379 	ASSERT(acur->len == 0);
1380 	ASSERT(args->type == XFS_ALLOCTYPE_NEAR_BNO);
1381 
1382 	*stat = 0;
1383 
1384 	error = xfs_alloc_lookup_ge(acur->cnt, args->agbno, acur->cur_len, &i);
1385 	if (error)
1386 		return error;
1387 	error = xfs_alloc_lookup_le(acur->bnolt, args->agbno, 0, &i);
1388 	if (error)
1389 		return error;
1390 	error = xfs_alloc_lookup_ge(acur->bnogt, args->agbno, 0, &i);
1391 	if (error)
1392 		return error;
1393 
1394 	/*
1395 	 * Search the bnobt and cntbt in parallel. Search the bnobt left and
1396 	 * right and lookup the closest extent to the locality hint for each
1397 	 * extent size key in the cntbt. The entire search terminates
1398 	 * immediately on a bnobt hit because that means we've found best case
1399 	 * locality. Otherwise the search continues until the cntbt cursor runs
1400 	 * off the end of the tree. If no allocation candidate is found at this
1401 	 * point, give up on locality, walk backwards from the end of the cntbt
1402 	 * and take the first available extent.
1403 	 *
1404 	 * The parallel tree searches balance each other out to provide fairly
1405 	 * consistent performance for various situations. The bnobt search can
1406 	 * have pathological behavior in the worst case scenario of larger
1407 	 * allocation requests and fragmented free space. On the other hand, the
1408 	 * bnobt is able to satisfy most smaller allocation requests much more
1409 	 * quickly than the cntbt. The cntbt search can sift through fragmented
1410 	 * free space and sets of free extents for larger allocation requests
1411 	 * more quickly than the bnobt. Since the locality hint is just a hint
1412 	 * and we don't want to scan the entire bnobt for perfect locality, the
1413 	 * cntbt search essentially bounds the bnobt search such that we can
1414 	 * find good enough locality at reasonable performance in most cases.
1415 	 */
1416 	while (xfs_alloc_cur_active(acur->bnolt) ||
1417 	       xfs_alloc_cur_active(acur->bnogt) ||
1418 	       xfs_alloc_cur_active(acur->cnt)) {
1419 
1420 		trace_xfs_alloc_cur_lookup(args);
1421 
1422 		/*
1423 		 * Search the bnobt left and right. In the case of a hit, finish
1424 		 * the search in the opposite direction and we're done.
1425 		 */
1426 		error = xfs_alloc_walk_iter(args, acur, acur->bnolt, false,
1427 					    true, 1, &i);
1428 		if (error)
1429 			return error;
1430 		if (i == 1) {
1431 			trace_xfs_alloc_cur_left(args);
1432 			fbcur = acur->bnogt;
1433 			fbinc = true;
1434 			break;
1435 		}
1436 		error = xfs_alloc_walk_iter(args, acur, acur->bnogt, true, true,
1437 					    1, &i);
1438 		if (error)
1439 			return error;
1440 		if (i == 1) {
1441 			trace_xfs_alloc_cur_right(args);
1442 			fbcur = acur->bnolt;
1443 			fbinc = false;
1444 			break;
1445 		}
1446 
1447 		/*
1448 		 * Check the extent with best locality based on the current
1449 		 * extent size search key and keep track of the best candidate.
1450 		 */
1451 		error = xfs_alloc_cntbt_iter(args, acur);
1452 		if (error)
1453 			return error;
1454 		if (!xfs_alloc_cur_active(acur->cnt)) {
1455 			trace_xfs_alloc_cur_lookup_done(args);
1456 			break;
1457 		}
1458 	}
1459 
1460 	/*
1461 	 * If we failed to find anything due to busy extents, return empty
1462 	 * handed so the caller can flush and retry. If no busy extents were
1463 	 * found, walk backwards from the end of the cntbt as a last resort.
1464 	 */
1465 	if (!xfs_alloc_cur_active(acur->cnt) && !acur->len && !acur->busy) {
1466 		error = xfs_btree_decrement(acur->cnt, 0, &i);
1467 		if (error)
1468 			return error;
1469 		if (i) {
1470 			acur->cnt->bc_ag.abt.active = true;
1471 			fbcur = acur->cnt;
1472 			fbinc = false;
1473 		}
1474 	}
1475 
1476 	/*
1477 	 * Search in the opposite direction for a better entry in the case of
1478 	 * a bnobt hit or walk backwards from the end of the cntbt.
1479 	 */
1480 	if (fbcur) {
1481 		error = xfs_alloc_walk_iter(args, acur, fbcur, fbinc, true, -1,
1482 					    &i);
1483 		if (error)
1484 			return error;
1485 	}
1486 
1487 	if (acur->len)
1488 		*stat = 1;
1489 
1490 	return 0;
1491 }
1492 
1493 /* Check the last block of the cnt btree for allocations. */
1494 static int
1495 xfs_alloc_ag_vextent_lastblock(
1496 	struct xfs_alloc_arg	*args,
1497 	struct xfs_alloc_cur	*acur,
1498 	xfs_agblock_t		*bno,
1499 	xfs_extlen_t		*len,
1500 	bool			*allocated)
1501 {
1502 	int			error;
1503 	int			i;
1504 
1505 #ifdef DEBUG
1506 	/* Randomly don't execute the first algorithm. */
1507 	if (prandom_u32() & 1)
1508 		return 0;
1509 #endif
1510 
1511 	/*
1512 	 * Start from the entry that lookup found, sequence through all larger
1513 	 * free blocks.  If we're actually pointing at a record smaller than
1514 	 * maxlen, go to the start of this block, and skip all those smaller
1515 	 * than minlen.
1516 	 */
1517 	if (*len || args->alignment > 1) {
1518 		acur->cnt->bc_ptrs[0] = 1;
1519 		do {
1520 			error = xfs_alloc_get_rec(acur->cnt, bno, len, &i);
1521 			if (error)
1522 				return error;
1523 			if (XFS_IS_CORRUPT(args->mp, i != 1))
1524 				return -EFSCORRUPTED;
1525 			if (*len >= args->minlen)
1526 				break;
1527 			error = xfs_btree_increment(acur->cnt, 0, &i);
1528 			if (error)
1529 				return error;
1530 		} while (i);
1531 		ASSERT(*len >= args->minlen);
1532 		if (!i)
1533 			return 0;
1534 	}
1535 
1536 	error = xfs_alloc_walk_iter(args, acur, acur->cnt, true, false, -1, &i);
1537 	if (error)
1538 		return error;
1539 
1540 	/*
1541 	 * It didn't work.  We COULD be in a case where there's a good record
1542 	 * somewhere, so try again.
1543 	 */
1544 	if (acur->len == 0)
1545 		return 0;
1546 
1547 	trace_xfs_alloc_near_first(args);
1548 	*allocated = true;
1549 	return 0;
1550 }
1551 
1552 /*
1553  * Allocate a variable extent near bno in the allocation group agno.
1554  * Extent's length (returned in len) will be between minlen and maxlen,
1555  * and of the form k * prod + mod unless there's nothing that large.
1556  * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1557  */
1558 STATIC int
1559 xfs_alloc_ag_vextent_near(
1560 	struct xfs_alloc_arg	*args)
1561 {
1562 	struct xfs_alloc_cur	acur = {};
1563 	int			error;		/* error code */
1564 	int			i;		/* result code, temporary */
1565 	xfs_agblock_t		bno;
1566 	xfs_extlen_t		len;
1567 
1568 	/* handle uninitialized agbno range so caller doesn't have to */
1569 	if (!args->min_agbno && !args->max_agbno)
1570 		args->max_agbno = args->mp->m_sb.sb_agblocks - 1;
1571 	ASSERT(args->min_agbno <= args->max_agbno);
1572 
1573 	/* clamp agbno to the range if it's outside */
1574 	if (args->agbno < args->min_agbno)
1575 		args->agbno = args->min_agbno;
1576 	if (args->agbno > args->max_agbno)
1577 		args->agbno = args->max_agbno;
1578 
1579 restart:
1580 	len = 0;
1581 
1582 	/*
1583 	 * Set up cursors and see if there are any free extents as big as
1584 	 * maxlen. If not, pick the last entry in the tree unless the tree is
1585 	 * empty.
1586 	 */
1587 	error = xfs_alloc_cur_setup(args, &acur);
1588 	if (error == -ENOSPC) {
1589 		error = xfs_alloc_ag_vextent_small(args, acur.cnt, &bno,
1590 				&len, &i);
1591 		if (error)
1592 			goto out;
1593 		if (i == 0 || len == 0) {
1594 			trace_xfs_alloc_near_noentry(args);
1595 			goto out;
1596 		}
1597 		ASSERT(i == 1);
1598 	} else if (error) {
1599 		goto out;
1600 	}
1601 
1602 	/*
1603 	 * First algorithm.
1604 	 * If the requested extent is large wrt the freespaces available
1605 	 * in this a.g., then the cursor will be pointing to a btree entry
1606 	 * near the right edge of the tree.  If it's in the last btree leaf
1607 	 * block, then we just examine all the entries in that block
1608 	 * that are big enough, and pick the best one.
1609 	 */
1610 	if (xfs_btree_islastblock(acur.cnt, 0)) {
1611 		bool		allocated = false;
1612 
1613 		error = xfs_alloc_ag_vextent_lastblock(args, &acur, &bno, &len,
1614 				&allocated);
1615 		if (error)
1616 			goto out;
1617 		if (allocated)
1618 			goto alloc_finish;
1619 	}
1620 
1621 	/*
1622 	 * Second algorithm. Combined cntbt and bnobt search to find ideal
1623 	 * locality.
1624 	 */
1625 	error = xfs_alloc_ag_vextent_locality(args, &acur, &i);
1626 	if (error)
1627 		goto out;
1628 
1629 	/*
1630 	 * If we couldn't get anything, give up.
1631 	 */
1632 	if (!acur.len) {
1633 		if (acur.busy) {
1634 			trace_xfs_alloc_near_busy(args);
1635 			xfs_extent_busy_flush(args->mp, args->pag,
1636 					      acur.busy_gen);
1637 			goto restart;
1638 		}
1639 		trace_xfs_alloc_size_neither(args);
1640 		args->agbno = NULLAGBLOCK;
1641 		goto out;
1642 	}
1643 
1644 alloc_finish:
1645 	/* fix up btrees on a successful allocation */
1646 	error = xfs_alloc_cur_finish(args, &acur);
1647 
1648 out:
1649 	xfs_alloc_cur_close(&acur, error);
1650 	return error;
1651 }
1652 
1653 /*
1654  * Allocate a variable extent anywhere in the allocation group agno.
1655  * Extent's length (returned in len) will be between minlen and maxlen,
1656  * and of the form k * prod + mod unless there's nothing that large.
1657  * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1658  */
1659 STATIC int				/* error */
1660 xfs_alloc_ag_vextent_size(
1661 	xfs_alloc_arg_t	*args)		/* allocation argument structure */
1662 {
1663 	struct xfs_agf	*agf = args->agbp->b_addr;
1664 	xfs_btree_cur_t	*bno_cur;	/* cursor for bno btree */
1665 	xfs_btree_cur_t	*cnt_cur;	/* cursor for cnt btree */
1666 	int		error;		/* error result */
1667 	xfs_agblock_t	fbno;		/* start of found freespace */
1668 	xfs_extlen_t	flen;		/* length of found freespace */
1669 	int		i;		/* temp status variable */
1670 	xfs_agblock_t	rbno;		/* returned block number */
1671 	xfs_extlen_t	rlen;		/* length of returned extent */
1672 	bool		busy;
1673 	unsigned	busy_gen;
1674 
1675 restart:
1676 	/*
1677 	 * Allocate and initialize a cursor for the by-size btree.
1678 	 */
1679 	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1680 		args->agno, XFS_BTNUM_CNT);
1681 	bno_cur = NULL;
1682 	busy = false;
1683 
1684 	/*
1685 	 * Look for an entry >= maxlen+alignment-1 blocks.
1686 	 */
1687 	if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1688 			args->maxlen + args->alignment - 1, &i)))
1689 		goto error0;
1690 
1691 	/*
1692 	 * If none then we have to settle for a smaller extent. In the case that
1693 	 * there are no large extents, this will return the last entry in the
1694 	 * tree unless the tree is empty. In the case that there are only busy
1695 	 * large extents, this will return the largest small extent unless there
1696 	 * are no smaller extents available.
1697 	 */
1698 	if (!i) {
1699 		error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1700 						   &fbno, &flen, &i);
1701 		if (error)
1702 			goto error0;
1703 		if (i == 0 || flen == 0) {
1704 			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1705 			trace_xfs_alloc_size_noentry(args);
1706 			return 0;
1707 		}
1708 		ASSERT(i == 1);
1709 		busy = xfs_alloc_compute_aligned(args, fbno, flen, &rbno,
1710 				&rlen, &busy_gen);
1711 	} else {
1712 		/*
1713 		 * Search for a non-busy extent that is large enough.
1714 		 */
1715 		for (;;) {
1716 			error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1717 			if (error)
1718 				goto error0;
1719 			if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1720 				error = -EFSCORRUPTED;
1721 				goto error0;
1722 			}
1723 
1724 			busy = xfs_alloc_compute_aligned(args, fbno, flen,
1725 					&rbno, &rlen, &busy_gen);
1726 
1727 			if (rlen >= args->maxlen)
1728 				break;
1729 
1730 			error = xfs_btree_increment(cnt_cur, 0, &i);
1731 			if (error)
1732 				goto error0;
1733 			if (i == 0) {
1734 				/*
1735 				 * Our only valid extents must have been busy.
1736 				 * Make it unbusy by forcing the log out and
1737 				 * retrying.
1738 				 */
1739 				xfs_btree_del_cursor(cnt_cur,
1740 						     XFS_BTREE_NOERROR);
1741 				trace_xfs_alloc_size_busy(args);
1742 				xfs_extent_busy_flush(args->mp,
1743 							args->pag, busy_gen);
1744 				goto restart;
1745 			}
1746 		}
1747 	}
1748 
1749 	/*
1750 	 * In the first case above, we got the last entry in the
1751 	 * by-size btree.  Now we check to see if the space hits maxlen
1752 	 * once aligned; if not, we search left for something better.
1753 	 * This can't happen in the second case above.
1754 	 */
1755 	rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1756 	if (XFS_IS_CORRUPT(args->mp,
1757 			   rlen != 0 &&
1758 			   (rlen > flen ||
1759 			    rbno + rlen > fbno + flen))) {
1760 		error = -EFSCORRUPTED;
1761 		goto error0;
1762 	}
1763 	if (rlen < args->maxlen) {
1764 		xfs_agblock_t	bestfbno;
1765 		xfs_extlen_t	bestflen;
1766 		xfs_agblock_t	bestrbno;
1767 		xfs_extlen_t	bestrlen;
1768 
1769 		bestrlen = rlen;
1770 		bestrbno = rbno;
1771 		bestflen = flen;
1772 		bestfbno = fbno;
1773 		for (;;) {
1774 			if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1775 				goto error0;
1776 			if (i == 0)
1777 				break;
1778 			if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1779 					&i)))
1780 				goto error0;
1781 			if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1782 				error = -EFSCORRUPTED;
1783 				goto error0;
1784 			}
1785 			if (flen < bestrlen)
1786 				break;
1787 			busy = xfs_alloc_compute_aligned(args, fbno, flen,
1788 					&rbno, &rlen, &busy_gen);
1789 			rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1790 			if (XFS_IS_CORRUPT(args->mp,
1791 					   rlen != 0 &&
1792 					   (rlen > flen ||
1793 					    rbno + rlen > fbno + flen))) {
1794 				error = -EFSCORRUPTED;
1795 				goto error0;
1796 			}
1797 			if (rlen > bestrlen) {
1798 				bestrlen = rlen;
1799 				bestrbno = rbno;
1800 				bestflen = flen;
1801 				bestfbno = fbno;
1802 				if (rlen == args->maxlen)
1803 					break;
1804 			}
1805 		}
1806 		if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1807 				&i)))
1808 			goto error0;
1809 		if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1810 			error = -EFSCORRUPTED;
1811 			goto error0;
1812 		}
1813 		rlen = bestrlen;
1814 		rbno = bestrbno;
1815 		flen = bestflen;
1816 		fbno = bestfbno;
1817 	}
1818 	args->wasfromfl = 0;
1819 	/*
1820 	 * Fix up the length.
1821 	 */
1822 	args->len = rlen;
1823 	if (rlen < args->minlen) {
1824 		if (busy) {
1825 			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1826 			trace_xfs_alloc_size_busy(args);
1827 			xfs_extent_busy_flush(args->mp, args->pag, busy_gen);
1828 			goto restart;
1829 		}
1830 		goto out_nominleft;
1831 	}
1832 	xfs_alloc_fix_len(args);
1833 
1834 	rlen = args->len;
1835 	if (XFS_IS_CORRUPT(args->mp, rlen > flen)) {
1836 		error = -EFSCORRUPTED;
1837 		goto error0;
1838 	}
1839 	/*
1840 	 * Allocate and initialize a cursor for the by-block tree.
1841 	 */
1842 	bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1843 		args->agno, XFS_BTNUM_BNO);
1844 	if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1845 			rbno, rlen, XFSA_FIXUP_CNT_OK)))
1846 		goto error0;
1847 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1848 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1849 	cnt_cur = bno_cur = NULL;
1850 	args->len = rlen;
1851 	args->agbno = rbno;
1852 	if (XFS_IS_CORRUPT(args->mp,
1853 			   args->agbno + args->len >
1854 			   be32_to_cpu(agf->agf_length))) {
1855 		error = -EFSCORRUPTED;
1856 		goto error0;
1857 	}
1858 	trace_xfs_alloc_size_done(args);
1859 	return 0;
1860 
1861 error0:
1862 	trace_xfs_alloc_size_error(args);
1863 	if (cnt_cur)
1864 		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1865 	if (bno_cur)
1866 		xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1867 	return error;
1868 
1869 out_nominleft:
1870 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1871 	trace_xfs_alloc_size_nominleft(args);
1872 	args->agbno = NULLAGBLOCK;
1873 	return 0;
1874 }
1875 
1876 /*
1877  * Free the extent starting at agno/bno for length.
1878  */
1879 STATIC int
1880 xfs_free_ag_extent(
1881 	struct xfs_trans		*tp,
1882 	struct xfs_buf			*agbp,
1883 	xfs_agnumber_t			agno,
1884 	xfs_agblock_t			bno,
1885 	xfs_extlen_t			len,
1886 	const struct xfs_owner_info	*oinfo,
1887 	enum xfs_ag_resv_type		type)
1888 {
1889 	struct xfs_mount		*mp;
1890 	struct xfs_perag		*pag;
1891 	struct xfs_btree_cur		*bno_cur;
1892 	struct xfs_btree_cur		*cnt_cur;
1893 	xfs_agblock_t			gtbno; /* start of right neighbor */
1894 	xfs_extlen_t			gtlen; /* length of right neighbor */
1895 	xfs_agblock_t			ltbno; /* start of left neighbor */
1896 	xfs_extlen_t			ltlen; /* length of left neighbor */
1897 	xfs_agblock_t			nbno; /* new starting block of freesp */
1898 	xfs_extlen_t			nlen; /* new length of freespace */
1899 	int				haveleft; /* have a left neighbor */
1900 	int				haveright; /* have a right neighbor */
1901 	int				i;
1902 	int				error;
1903 
1904 	bno_cur = cnt_cur = NULL;
1905 	mp = tp->t_mountp;
1906 
1907 	if (!xfs_rmap_should_skip_owner_update(oinfo)) {
1908 		error = xfs_rmap_free(tp, agbp, agno, bno, len, oinfo);
1909 		if (error)
1910 			goto error0;
1911 	}
1912 
1913 	/*
1914 	 * Allocate and initialize a cursor for the by-block btree.
1915 	 */
1916 	bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1917 	/*
1918 	 * Look for a neighboring block on the left (lower block numbers)
1919 	 * that is contiguous with this space.
1920 	 */
1921 	if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1922 		goto error0;
1923 	if (haveleft) {
1924 		/*
1925 		 * There is a block to our left.
1926 		 */
1927 		if ((error = xfs_alloc_get_rec(bno_cur, &ltbno, &ltlen, &i)))
1928 			goto error0;
1929 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1930 			error = -EFSCORRUPTED;
1931 			goto error0;
1932 		}
1933 		/*
1934 		 * It's not contiguous, though.
1935 		 */
1936 		if (ltbno + ltlen < bno)
1937 			haveleft = 0;
1938 		else {
1939 			/*
1940 			 * If this failure happens the request to free this
1941 			 * space was invalid, it's (partly) already free.
1942 			 * Very bad.
1943 			 */
1944 			if (XFS_IS_CORRUPT(mp, ltbno + ltlen > bno)) {
1945 				error = -EFSCORRUPTED;
1946 				goto error0;
1947 			}
1948 		}
1949 	}
1950 	/*
1951 	 * Look for a neighboring block on the right (higher block numbers)
1952 	 * that is contiguous with this space.
1953 	 */
1954 	if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1955 		goto error0;
1956 	if (haveright) {
1957 		/*
1958 		 * There is a block to our right.
1959 		 */
1960 		if ((error = xfs_alloc_get_rec(bno_cur, &gtbno, &gtlen, &i)))
1961 			goto error0;
1962 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1963 			error = -EFSCORRUPTED;
1964 			goto error0;
1965 		}
1966 		/*
1967 		 * It's not contiguous, though.
1968 		 */
1969 		if (bno + len < gtbno)
1970 			haveright = 0;
1971 		else {
1972 			/*
1973 			 * If this failure happens the request to free this
1974 			 * space was invalid, it's (partly) already free.
1975 			 * Very bad.
1976 			 */
1977 			if (XFS_IS_CORRUPT(mp, bno + len > gtbno)) {
1978 				error = -EFSCORRUPTED;
1979 				goto error0;
1980 			}
1981 		}
1982 	}
1983 	/*
1984 	 * Now allocate and initialize a cursor for the by-size tree.
1985 	 */
1986 	cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1987 	/*
1988 	 * Have both left and right contiguous neighbors.
1989 	 * Merge all three into a single free block.
1990 	 */
1991 	if (haveleft && haveright) {
1992 		/*
1993 		 * Delete the old by-size entry on the left.
1994 		 */
1995 		if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1996 			goto error0;
1997 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1998 			error = -EFSCORRUPTED;
1999 			goto error0;
2000 		}
2001 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2002 			goto error0;
2003 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2004 			error = -EFSCORRUPTED;
2005 			goto error0;
2006 		}
2007 		/*
2008 		 * Delete the old by-size entry on the right.
2009 		 */
2010 		if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
2011 			goto error0;
2012 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2013 			error = -EFSCORRUPTED;
2014 			goto error0;
2015 		}
2016 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2017 			goto error0;
2018 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2019 			error = -EFSCORRUPTED;
2020 			goto error0;
2021 		}
2022 		/*
2023 		 * Delete the old by-block entry for the right block.
2024 		 */
2025 		if ((error = xfs_btree_delete(bno_cur, &i)))
2026 			goto error0;
2027 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2028 			error = -EFSCORRUPTED;
2029 			goto error0;
2030 		}
2031 		/*
2032 		 * Move the by-block cursor back to the left neighbor.
2033 		 */
2034 		if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
2035 			goto error0;
2036 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2037 			error = -EFSCORRUPTED;
2038 			goto error0;
2039 		}
2040 #ifdef DEBUG
2041 		/*
2042 		 * Check that this is the right record: delete didn't
2043 		 * mangle the cursor.
2044 		 */
2045 		{
2046 			xfs_agblock_t	xxbno;
2047 			xfs_extlen_t	xxlen;
2048 
2049 			if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
2050 					&i)))
2051 				goto error0;
2052 			if (XFS_IS_CORRUPT(mp,
2053 					   i != 1 ||
2054 					   xxbno != ltbno ||
2055 					   xxlen != ltlen)) {
2056 				error = -EFSCORRUPTED;
2057 				goto error0;
2058 			}
2059 		}
2060 #endif
2061 		/*
2062 		 * Update remaining by-block entry to the new, joined block.
2063 		 */
2064 		nbno = ltbno;
2065 		nlen = len + ltlen + gtlen;
2066 		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2067 			goto error0;
2068 	}
2069 	/*
2070 	 * Have only a left contiguous neighbor.
2071 	 * Merge it together with the new freespace.
2072 	 */
2073 	else if (haveleft) {
2074 		/*
2075 		 * Delete the old by-size entry on the left.
2076 		 */
2077 		if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
2078 			goto error0;
2079 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2080 			error = -EFSCORRUPTED;
2081 			goto error0;
2082 		}
2083 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2084 			goto error0;
2085 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2086 			error = -EFSCORRUPTED;
2087 			goto error0;
2088 		}
2089 		/*
2090 		 * Back up the by-block cursor to the left neighbor, and
2091 		 * update its length.
2092 		 */
2093 		if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
2094 			goto error0;
2095 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2096 			error = -EFSCORRUPTED;
2097 			goto error0;
2098 		}
2099 		nbno = ltbno;
2100 		nlen = len + ltlen;
2101 		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2102 			goto error0;
2103 	}
2104 	/*
2105 	 * Have only a right contiguous neighbor.
2106 	 * Merge it together with the new freespace.
2107 	 */
2108 	else if (haveright) {
2109 		/*
2110 		 * Delete the old by-size entry on the right.
2111 		 */
2112 		if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
2113 			goto error0;
2114 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2115 			error = -EFSCORRUPTED;
2116 			goto error0;
2117 		}
2118 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2119 			goto error0;
2120 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2121 			error = -EFSCORRUPTED;
2122 			goto error0;
2123 		}
2124 		/*
2125 		 * Update the starting block and length of the right
2126 		 * neighbor in the by-block tree.
2127 		 */
2128 		nbno = bno;
2129 		nlen = len + gtlen;
2130 		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2131 			goto error0;
2132 	}
2133 	/*
2134 	 * No contiguous neighbors.
2135 	 * Insert the new freespace into the by-block tree.
2136 	 */
2137 	else {
2138 		nbno = bno;
2139 		nlen = len;
2140 		if ((error = xfs_btree_insert(bno_cur, &i)))
2141 			goto error0;
2142 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2143 			error = -EFSCORRUPTED;
2144 			goto error0;
2145 		}
2146 	}
2147 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
2148 	bno_cur = NULL;
2149 	/*
2150 	 * In all cases we need to insert the new freespace in the by-size tree.
2151 	 */
2152 	if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
2153 		goto error0;
2154 	if (XFS_IS_CORRUPT(mp, i != 0)) {
2155 		error = -EFSCORRUPTED;
2156 		goto error0;
2157 	}
2158 	if ((error = xfs_btree_insert(cnt_cur, &i)))
2159 		goto error0;
2160 	if (XFS_IS_CORRUPT(mp, i != 1)) {
2161 		error = -EFSCORRUPTED;
2162 		goto error0;
2163 	}
2164 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
2165 	cnt_cur = NULL;
2166 
2167 	/*
2168 	 * Update the freespace totals in the ag and superblock.
2169 	 */
2170 	pag = xfs_perag_get(mp, agno);
2171 	error = xfs_alloc_update_counters(tp, pag, agbp, len);
2172 	xfs_ag_resv_free_extent(pag, type, tp, len);
2173 	xfs_perag_put(pag);
2174 	if (error)
2175 		goto error0;
2176 
2177 	XFS_STATS_INC(mp, xs_freex);
2178 	XFS_STATS_ADD(mp, xs_freeb, len);
2179 
2180 	trace_xfs_free_extent(mp, agno, bno, len, type, haveleft, haveright);
2181 
2182 	return 0;
2183 
2184  error0:
2185 	trace_xfs_free_extent(mp, agno, bno, len, type, -1, -1);
2186 	if (bno_cur)
2187 		xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
2188 	if (cnt_cur)
2189 		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
2190 	return error;
2191 }
2192 
2193 /*
2194  * Visible (exported) allocation/free functions.
2195  * Some of these are used just by xfs_alloc_btree.c and this file.
2196  */
2197 
2198 /*
2199  * Compute and fill in value of m_ag_maxlevels.
2200  */
2201 void
2202 xfs_alloc_compute_maxlevels(
2203 	xfs_mount_t	*mp)	/* file system mount structure */
2204 {
2205 	mp->m_ag_maxlevels = xfs_btree_compute_maxlevels(mp->m_alloc_mnr,
2206 			(mp->m_sb.sb_agblocks + 1) / 2);
2207 }
2208 
2209 /*
2210  * Find the length of the longest extent in an AG.  The 'need' parameter
2211  * specifies how much space we're going to need for the AGFL and the
2212  * 'reserved' parameter tells us how many blocks in this AG are reserved for
2213  * other callers.
2214  */
2215 xfs_extlen_t
2216 xfs_alloc_longest_free_extent(
2217 	struct xfs_perag	*pag,
2218 	xfs_extlen_t		need,
2219 	xfs_extlen_t		reserved)
2220 {
2221 	xfs_extlen_t		delta = 0;
2222 
2223 	/*
2224 	 * If the AGFL needs a recharge, we'll have to subtract that from the
2225 	 * longest extent.
2226 	 */
2227 	if (need > pag->pagf_flcount)
2228 		delta = need - pag->pagf_flcount;
2229 
2230 	/*
2231 	 * If we cannot maintain others' reservations with space from the
2232 	 * not-longest freesp extents, we'll have to subtract /that/ from
2233 	 * the longest extent too.
2234 	 */
2235 	if (pag->pagf_freeblks - pag->pagf_longest < reserved)
2236 		delta += reserved - (pag->pagf_freeblks - pag->pagf_longest);
2237 
2238 	/*
2239 	 * If the longest extent is long enough to satisfy all the
2240 	 * reservations and AGFL rules in place, we can return this extent.
2241 	 */
2242 	if (pag->pagf_longest > delta)
2243 		return min_t(xfs_extlen_t, pag->pag_mount->m_ag_max_usable,
2244 				pag->pagf_longest - delta);
2245 
2246 	/* Otherwise, let the caller try for 1 block if there's space. */
2247 	return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
2248 }
2249 
2250 /*
2251  * Compute the minimum length of the AGFL in the given AG.  If @pag is NULL,
2252  * return the largest possible minimum length.
2253  */
2254 unsigned int
2255 xfs_alloc_min_freelist(
2256 	struct xfs_mount	*mp,
2257 	struct xfs_perag	*pag)
2258 {
2259 	/* AG btrees have at least 1 level. */
2260 	static const uint8_t	fake_levels[XFS_BTNUM_AGF] = {1, 1, 1};
2261 	const uint8_t		*levels = pag ? pag->pagf_levels : fake_levels;
2262 	unsigned int		min_free;
2263 
2264 	ASSERT(mp->m_ag_maxlevels > 0);
2265 
2266 	/* space needed by-bno freespace btree */
2267 	min_free = min_t(unsigned int, levels[XFS_BTNUM_BNOi] + 1,
2268 				       mp->m_ag_maxlevels);
2269 	/* space needed by-size freespace btree */
2270 	min_free += min_t(unsigned int, levels[XFS_BTNUM_CNTi] + 1,
2271 				       mp->m_ag_maxlevels);
2272 	/* space needed reverse mapping used space btree */
2273 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2274 		min_free += min_t(unsigned int, levels[XFS_BTNUM_RMAPi] + 1,
2275 						mp->m_rmap_maxlevels);
2276 
2277 	return min_free;
2278 }
2279 
2280 /*
2281  * Check if the operation we are fixing up the freelist for should go ahead or
2282  * not. If we are freeing blocks, we always allow it, otherwise the allocation
2283  * is dependent on whether the size and shape of free space available will
2284  * permit the requested allocation to take place.
2285  */
2286 static bool
2287 xfs_alloc_space_available(
2288 	struct xfs_alloc_arg	*args,
2289 	xfs_extlen_t		min_free,
2290 	int			flags)
2291 {
2292 	struct xfs_perag	*pag = args->pag;
2293 	xfs_extlen_t		alloc_len, longest;
2294 	xfs_extlen_t		reservation; /* blocks that are still reserved */
2295 	int			available;
2296 	xfs_extlen_t		agflcount;
2297 
2298 	if (flags & XFS_ALLOC_FLAG_FREEING)
2299 		return true;
2300 
2301 	reservation = xfs_ag_resv_needed(pag, args->resv);
2302 
2303 	/* do we have enough contiguous free space for the allocation? */
2304 	alloc_len = args->minlen + (args->alignment - 1) + args->minalignslop;
2305 	longest = xfs_alloc_longest_free_extent(pag, min_free, reservation);
2306 	if (longest < alloc_len)
2307 		return false;
2308 
2309 	/*
2310 	 * Do we have enough free space remaining for the allocation? Don't
2311 	 * account extra agfl blocks because we are about to defer free them,
2312 	 * making them unavailable until the current transaction commits.
2313 	 */
2314 	agflcount = min_t(xfs_extlen_t, pag->pagf_flcount, min_free);
2315 	available = (int)(pag->pagf_freeblks + agflcount -
2316 			  reservation - min_free - args->minleft);
2317 	if (available < (int)max(args->total, alloc_len))
2318 		return false;
2319 
2320 	/*
2321 	 * Clamp maxlen to the amount of free space available for the actual
2322 	 * extent allocation.
2323 	 */
2324 	if (available < (int)args->maxlen && !(flags & XFS_ALLOC_FLAG_CHECK)) {
2325 		args->maxlen = available;
2326 		ASSERT(args->maxlen > 0);
2327 		ASSERT(args->maxlen >= args->minlen);
2328 	}
2329 
2330 	return true;
2331 }
2332 
2333 int
2334 xfs_free_agfl_block(
2335 	struct xfs_trans	*tp,
2336 	xfs_agnumber_t		agno,
2337 	xfs_agblock_t		agbno,
2338 	struct xfs_buf		*agbp,
2339 	struct xfs_owner_info	*oinfo)
2340 {
2341 	int			error;
2342 	struct xfs_buf		*bp;
2343 
2344 	error = xfs_free_ag_extent(tp, agbp, agno, agbno, 1, oinfo,
2345 				   XFS_AG_RESV_AGFL);
2346 	if (error)
2347 		return error;
2348 
2349 	error = xfs_trans_get_buf(tp, tp->t_mountp->m_ddev_targp,
2350 			XFS_AGB_TO_DADDR(tp->t_mountp, agno, agbno),
2351 			tp->t_mountp->m_bsize, 0, &bp);
2352 	if (error)
2353 		return error;
2354 	xfs_trans_binval(tp, bp);
2355 
2356 	return 0;
2357 }
2358 
2359 /*
2360  * Check the agfl fields of the agf for inconsistency or corruption. The purpose
2361  * is to detect an agfl header padding mismatch between current and early v5
2362  * kernels. This problem manifests as a 1-slot size difference between the
2363  * on-disk flcount and the active [first, last] range of a wrapped agfl. This
2364  * may also catch variants of agfl count corruption unrelated to padding. Either
2365  * way, we'll reset the agfl and warn the user.
2366  *
2367  * Return true if a reset is required before the agfl can be used, false
2368  * otherwise.
2369  */
2370 static bool
2371 xfs_agfl_needs_reset(
2372 	struct xfs_mount	*mp,
2373 	struct xfs_agf		*agf)
2374 {
2375 	uint32_t		f = be32_to_cpu(agf->agf_flfirst);
2376 	uint32_t		l = be32_to_cpu(agf->agf_fllast);
2377 	uint32_t		c = be32_to_cpu(agf->agf_flcount);
2378 	int			agfl_size = xfs_agfl_size(mp);
2379 	int			active;
2380 
2381 	/* no agfl header on v4 supers */
2382 	if (!xfs_sb_version_hascrc(&mp->m_sb))
2383 		return false;
2384 
2385 	/*
2386 	 * The agf read verifier catches severe corruption of these fields.
2387 	 * Repeat some sanity checks to cover a packed -> unpacked mismatch if
2388 	 * the verifier allows it.
2389 	 */
2390 	if (f >= agfl_size || l >= agfl_size)
2391 		return true;
2392 	if (c > agfl_size)
2393 		return true;
2394 
2395 	/*
2396 	 * Check consistency between the on-disk count and the active range. An
2397 	 * agfl padding mismatch manifests as an inconsistent flcount.
2398 	 */
2399 	if (c && l >= f)
2400 		active = l - f + 1;
2401 	else if (c)
2402 		active = agfl_size - f + l + 1;
2403 	else
2404 		active = 0;
2405 
2406 	return active != c;
2407 }
2408 
2409 /*
2410  * Reset the agfl to an empty state. Ignore/drop any existing blocks since the
2411  * agfl content cannot be trusted. Warn the user that a repair is required to
2412  * recover leaked blocks.
2413  *
2414  * The purpose of this mechanism is to handle filesystems affected by the agfl
2415  * header padding mismatch problem. A reset keeps the filesystem online with a
2416  * relatively minor free space accounting inconsistency rather than suffer the
2417  * inevitable crash from use of an invalid agfl block.
2418  */
2419 static void
2420 xfs_agfl_reset(
2421 	struct xfs_trans	*tp,
2422 	struct xfs_buf		*agbp,
2423 	struct xfs_perag	*pag)
2424 {
2425 	struct xfs_mount	*mp = tp->t_mountp;
2426 	struct xfs_agf		*agf = agbp->b_addr;
2427 
2428 	ASSERT(pag->pagf_agflreset);
2429 	trace_xfs_agfl_reset(mp, agf, 0, _RET_IP_);
2430 
2431 	xfs_warn(mp,
2432 	       "WARNING: Reset corrupted AGFL on AG %u. %d blocks leaked. "
2433 	       "Please unmount and run xfs_repair.",
2434 	         pag->pag_agno, pag->pagf_flcount);
2435 
2436 	agf->agf_flfirst = 0;
2437 	agf->agf_fllast = cpu_to_be32(xfs_agfl_size(mp) - 1);
2438 	agf->agf_flcount = 0;
2439 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLLAST |
2440 				    XFS_AGF_FLCOUNT);
2441 
2442 	pag->pagf_flcount = 0;
2443 	pag->pagf_agflreset = false;
2444 }
2445 
2446 /*
2447  * Defer an AGFL block free. This is effectively equivalent to
2448  * xfs_bmap_add_free() with some special handling particular to AGFL blocks.
2449  *
2450  * Deferring AGFL frees helps prevent log reservation overruns due to too many
2451  * allocation operations in a transaction. AGFL frees are prone to this problem
2452  * because for one they are always freed one at a time. Further, an immediate
2453  * AGFL block free can cause a btree join and require another block free before
2454  * the real allocation can proceed. Deferring the free disconnects freeing up
2455  * the AGFL slot from freeing the block.
2456  */
2457 STATIC void
2458 xfs_defer_agfl_block(
2459 	struct xfs_trans		*tp,
2460 	xfs_agnumber_t			agno,
2461 	xfs_fsblock_t			agbno,
2462 	struct xfs_owner_info		*oinfo)
2463 {
2464 	struct xfs_mount		*mp = tp->t_mountp;
2465 	struct xfs_extent_free_item	*new;		/* new element */
2466 
2467 	ASSERT(xfs_bmap_free_item_zone != NULL);
2468 	ASSERT(oinfo != NULL);
2469 
2470 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, 0);
2471 	new->xefi_startblock = XFS_AGB_TO_FSB(mp, agno, agbno);
2472 	new->xefi_blockcount = 1;
2473 	new->xefi_oinfo = *oinfo;
2474 
2475 	trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1);
2476 
2477 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_AGFL_FREE, &new->xefi_list);
2478 }
2479 
2480 /*
2481  * Decide whether to use this allocation group for this allocation.
2482  * If so, fix up the btree freelist's size.
2483  */
2484 int			/* error */
2485 xfs_alloc_fix_freelist(
2486 	struct xfs_alloc_arg	*args,	/* allocation argument structure */
2487 	int			flags)	/* XFS_ALLOC_FLAG_... */
2488 {
2489 	struct xfs_mount	*mp = args->mp;
2490 	struct xfs_perag	*pag = args->pag;
2491 	struct xfs_trans	*tp = args->tp;
2492 	struct xfs_buf		*agbp = NULL;
2493 	struct xfs_buf		*agflbp = NULL;
2494 	struct xfs_alloc_arg	targs;	/* local allocation arguments */
2495 	xfs_agblock_t		bno;	/* freelist block */
2496 	xfs_extlen_t		need;	/* total blocks needed in freelist */
2497 	int			error = 0;
2498 
2499 	/* deferred ops (AGFL block frees) require permanent transactions */
2500 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
2501 
2502 	if (!pag->pagf_init) {
2503 		error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp);
2504 		if (error) {
2505 			/* Couldn't lock the AGF so skip this AG. */
2506 			if (error == -EAGAIN)
2507 				error = 0;
2508 			goto out_no_agbp;
2509 		}
2510 	}
2511 
2512 	/*
2513 	 * If this is a metadata preferred pag and we are user data then try
2514 	 * somewhere else if we are not being asked to try harder at this
2515 	 * point
2516 	 */
2517 	if (pag->pagf_metadata && (args->datatype & XFS_ALLOC_USERDATA) &&
2518 	    (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
2519 		ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
2520 		goto out_agbp_relse;
2521 	}
2522 
2523 	need = xfs_alloc_min_freelist(mp, pag);
2524 	if (!xfs_alloc_space_available(args, need, flags |
2525 			XFS_ALLOC_FLAG_CHECK))
2526 		goto out_agbp_relse;
2527 
2528 	/*
2529 	 * Get the a.g. freespace buffer.
2530 	 * Can fail if we're not blocking on locks, and it's held.
2531 	 */
2532 	if (!agbp) {
2533 		error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp);
2534 		if (error) {
2535 			/* Couldn't lock the AGF so skip this AG. */
2536 			if (error == -EAGAIN)
2537 				error = 0;
2538 			goto out_no_agbp;
2539 		}
2540 	}
2541 
2542 	/* reset a padding mismatched agfl before final free space check */
2543 	if (pag->pagf_agflreset)
2544 		xfs_agfl_reset(tp, agbp, pag);
2545 
2546 	/* If there isn't enough total space or single-extent, reject it. */
2547 	need = xfs_alloc_min_freelist(mp, pag);
2548 	if (!xfs_alloc_space_available(args, need, flags))
2549 		goto out_agbp_relse;
2550 
2551 	/*
2552 	 * Make the freelist shorter if it's too long.
2553 	 *
2554 	 * Note that from this point onwards, we will always release the agf and
2555 	 * agfl buffers on error. This handles the case where we error out and
2556 	 * the buffers are clean or may not have been joined to the transaction
2557 	 * and hence need to be released manually. If they have been joined to
2558 	 * the transaction, then xfs_trans_brelse() will handle them
2559 	 * appropriately based on the recursion count and dirty state of the
2560 	 * buffer.
2561 	 *
2562 	 * XXX (dgc): When we have lots of free space, does this buy us
2563 	 * anything other than extra overhead when we need to put more blocks
2564 	 * back on the free list? Maybe we should only do this when space is
2565 	 * getting low or the AGFL is more than half full?
2566 	 *
2567 	 * The NOSHRINK flag prevents the AGFL from being shrunk if it's too
2568 	 * big; the NORMAP flag prevents AGFL expand/shrink operations from
2569 	 * updating the rmapbt.  Both flags are used in xfs_repair while we're
2570 	 * rebuilding the rmapbt, and neither are used by the kernel.  They're
2571 	 * both required to ensure that rmaps are correctly recorded for the
2572 	 * regenerated AGFL, bnobt, and cntbt.  See repair/phase5.c and
2573 	 * repair/rmap.c in xfsprogs for details.
2574 	 */
2575 	memset(&targs, 0, sizeof(targs));
2576 	/* struct copy below */
2577 	if (flags & XFS_ALLOC_FLAG_NORMAP)
2578 		targs.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
2579 	else
2580 		targs.oinfo = XFS_RMAP_OINFO_AG;
2581 	while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) {
2582 		error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
2583 		if (error)
2584 			goto out_agbp_relse;
2585 
2586 		/* defer agfl frees */
2587 		xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
2588 	}
2589 
2590 	targs.tp = tp;
2591 	targs.mp = mp;
2592 	targs.agbp = agbp;
2593 	targs.agno = args->agno;
2594 	targs.alignment = targs.minlen = targs.prod = 1;
2595 	targs.type = XFS_ALLOCTYPE_THIS_AG;
2596 	targs.pag = pag;
2597 	error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp);
2598 	if (error)
2599 		goto out_agbp_relse;
2600 
2601 	/* Make the freelist longer if it's too short. */
2602 	while (pag->pagf_flcount < need) {
2603 		targs.agbno = 0;
2604 		targs.maxlen = need - pag->pagf_flcount;
2605 		targs.resv = XFS_AG_RESV_AGFL;
2606 
2607 		/* Allocate as many blocks as possible at once. */
2608 		error = xfs_alloc_ag_vextent(&targs);
2609 		if (error)
2610 			goto out_agflbp_relse;
2611 
2612 		/*
2613 		 * Stop if we run out.  Won't happen if callers are obeying
2614 		 * the restrictions correctly.  Can happen for free calls
2615 		 * on a completely full ag.
2616 		 */
2617 		if (targs.agbno == NULLAGBLOCK) {
2618 			if (flags & XFS_ALLOC_FLAG_FREEING)
2619 				break;
2620 			goto out_agflbp_relse;
2621 		}
2622 		/*
2623 		 * Put each allocated block on the list.
2624 		 */
2625 		for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
2626 			error = xfs_alloc_put_freelist(tp, agbp,
2627 							agflbp, bno, 0);
2628 			if (error)
2629 				goto out_agflbp_relse;
2630 		}
2631 	}
2632 	xfs_trans_brelse(tp, agflbp);
2633 	args->agbp = agbp;
2634 	return 0;
2635 
2636 out_agflbp_relse:
2637 	xfs_trans_brelse(tp, agflbp);
2638 out_agbp_relse:
2639 	if (agbp)
2640 		xfs_trans_brelse(tp, agbp);
2641 out_no_agbp:
2642 	args->agbp = NULL;
2643 	return error;
2644 }
2645 
2646 /*
2647  * Get a block from the freelist.
2648  * Returns with the buffer for the block gotten.
2649  */
2650 int				/* error */
2651 xfs_alloc_get_freelist(
2652 	xfs_trans_t	*tp,	/* transaction pointer */
2653 	xfs_buf_t	*agbp,	/* buffer containing the agf structure */
2654 	xfs_agblock_t	*bnop,	/* block address retrieved from freelist */
2655 	int		btreeblk) /* destination is a AGF btree */
2656 {
2657 	struct xfs_agf	*agf = agbp->b_addr;
2658 	xfs_buf_t	*agflbp;/* buffer for a.g. freelist structure */
2659 	xfs_agblock_t	bno;	/* block number returned */
2660 	__be32		*agfl_bno;
2661 	int		error;
2662 	int		logflags;
2663 	xfs_mount_t	*mp = tp->t_mountp;
2664 	xfs_perag_t	*pag;	/* per allocation group data */
2665 
2666 	/*
2667 	 * Freelist is empty, give up.
2668 	 */
2669 	if (!agf->agf_flcount) {
2670 		*bnop = NULLAGBLOCK;
2671 		return 0;
2672 	}
2673 	/*
2674 	 * Read the array of free blocks.
2675 	 */
2676 	error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno),
2677 				    &agflbp);
2678 	if (error)
2679 		return error;
2680 
2681 
2682 	/*
2683 	 * Get the block number and update the data structures.
2684 	 */
2685 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
2686 	bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2687 	be32_add_cpu(&agf->agf_flfirst, 1);
2688 	xfs_trans_brelse(tp, agflbp);
2689 	if (be32_to_cpu(agf->agf_flfirst) == xfs_agfl_size(mp))
2690 		agf->agf_flfirst = 0;
2691 
2692 	pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2693 	ASSERT(!pag->pagf_agflreset);
2694 	be32_add_cpu(&agf->agf_flcount, -1);
2695 	xfs_trans_agflist_delta(tp, -1);
2696 	pag->pagf_flcount--;
2697 
2698 	logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2699 	if (btreeblk) {
2700 		be32_add_cpu(&agf->agf_btreeblks, 1);
2701 		pag->pagf_btreeblks++;
2702 		logflags |= XFS_AGF_BTREEBLKS;
2703 	}
2704 	xfs_perag_put(pag);
2705 
2706 	xfs_alloc_log_agf(tp, agbp, logflags);
2707 	*bnop = bno;
2708 
2709 	return 0;
2710 }
2711 
2712 /*
2713  * Log the given fields from the agf structure.
2714  */
2715 void
2716 xfs_alloc_log_agf(
2717 	xfs_trans_t	*tp,	/* transaction pointer */
2718 	xfs_buf_t	*bp,	/* buffer for a.g. freelist header */
2719 	int		fields)	/* mask of fields to be logged (XFS_AGF_...) */
2720 {
2721 	int	first;		/* first byte offset */
2722 	int	last;		/* last byte offset */
2723 	static const short	offsets[] = {
2724 		offsetof(xfs_agf_t, agf_magicnum),
2725 		offsetof(xfs_agf_t, agf_versionnum),
2726 		offsetof(xfs_agf_t, agf_seqno),
2727 		offsetof(xfs_agf_t, agf_length),
2728 		offsetof(xfs_agf_t, agf_roots[0]),
2729 		offsetof(xfs_agf_t, agf_levels[0]),
2730 		offsetof(xfs_agf_t, agf_flfirst),
2731 		offsetof(xfs_agf_t, agf_fllast),
2732 		offsetof(xfs_agf_t, agf_flcount),
2733 		offsetof(xfs_agf_t, agf_freeblks),
2734 		offsetof(xfs_agf_t, agf_longest),
2735 		offsetof(xfs_agf_t, agf_btreeblks),
2736 		offsetof(xfs_agf_t, agf_uuid),
2737 		offsetof(xfs_agf_t, agf_rmap_blocks),
2738 		offsetof(xfs_agf_t, agf_refcount_blocks),
2739 		offsetof(xfs_agf_t, agf_refcount_root),
2740 		offsetof(xfs_agf_t, agf_refcount_level),
2741 		/* needed so that we don't log the whole rest of the structure: */
2742 		offsetof(xfs_agf_t, agf_spare64),
2743 		sizeof(xfs_agf_t)
2744 	};
2745 
2746 	trace_xfs_agf(tp->t_mountp, bp->b_addr, fields, _RET_IP_);
2747 
2748 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF);
2749 
2750 	xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2751 	xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2752 }
2753 
2754 /*
2755  * Interface for inode allocation to force the pag data to be initialized.
2756  */
2757 int					/* error */
2758 xfs_alloc_pagf_init(
2759 	xfs_mount_t		*mp,	/* file system mount structure */
2760 	xfs_trans_t		*tp,	/* transaction pointer */
2761 	xfs_agnumber_t		agno,	/* allocation group number */
2762 	int			flags)	/* XFS_ALLOC_FLAGS_... */
2763 {
2764 	xfs_buf_t		*bp;
2765 	int			error;
2766 
2767 	error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp);
2768 	if (!error)
2769 		xfs_trans_brelse(tp, bp);
2770 	return error;
2771 }
2772 
2773 /*
2774  * Put the block on the freelist for the allocation group.
2775  */
2776 int					/* error */
2777 xfs_alloc_put_freelist(
2778 	xfs_trans_t		*tp,	/* transaction pointer */
2779 	xfs_buf_t		*agbp,	/* buffer for a.g. freelist header */
2780 	xfs_buf_t		*agflbp,/* buffer for a.g. free block array */
2781 	xfs_agblock_t		bno,	/* block being freed */
2782 	int			btreeblk) /* block came from a AGF btree */
2783 {
2784 	struct xfs_mount	*mp = tp->t_mountp;
2785 	struct xfs_agf		*agf = agbp->b_addr;
2786 	__be32			*blockp;/* pointer to array entry */
2787 	int			error;
2788 	int			logflags;
2789 	xfs_perag_t		*pag;	/* per allocation group data */
2790 	__be32			*agfl_bno;
2791 	int			startoff;
2792 
2793 	if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2794 			be32_to_cpu(agf->agf_seqno), &agflbp)))
2795 		return error;
2796 	be32_add_cpu(&agf->agf_fllast, 1);
2797 	if (be32_to_cpu(agf->agf_fllast) == xfs_agfl_size(mp))
2798 		agf->agf_fllast = 0;
2799 
2800 	pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2801 	ASSERT(!pag->pagf_agflreset);
2802 	be32_add_cpu(&agf->agf_flcount, 1);
2803 	xfs_trans_agflist_delta(tp, 1);
2804 	pag->pagf_flcount++;
2805 
2806 	logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2807 	if (btreeblk) {
2808 		be32_add_cpu(&agf->agf_btreeblks, -1);
2809 		pag->pagf_btreeblks--;
2810 		logflags |= XFS_AGF_BTREEBLKS;
2811 	}
2812 	xfs_perag_put(pag);
2813 
2814 	xfs_alloc_log_agf(tp, agbp, logflags);
2815 
2816 	ASSERT(be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp));
2817 
2818 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
2819 	blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)];
2820 	*blockp = cpu_to_be32(bno);
2821 	startoff = (char *)blockp - (char *)agflbp->b_addr;
2822 
2823 	xfs_alloc_log_agf(tp, agbp, logflags);
2824 
2825 	xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF);
2826 	xfs_trans_log_buf(tp, agflbp, startoff,
2827 			  startoff + sizeof(xfs_agblock_t) - 1);
2828 	return 0;
2829 }
2830 
2831 static xfs_failaddr_t
2832 xfs_agf_verify(
2833 	struct xfs_buf		*bp)
2834 {
2835 	struct xfs_mount	*mp = bp->b_mount;
2836 	struct xfs_agf		*agf = bp->b_addr;
2837 
2838 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
2839 		if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid))
2840 			return __this_address;
2841 		if (!xfs_log_check_lsn(mp, be64_to_cpu(agf->agf_lsn)))
2842 			return __this_address;
2843 	}
2844 
2845 	if (!xfs_verify_magic(bp, agf->agf_magicnum))
2846 		return __this_address;
2847 
2848 	if (!(XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2849 	      be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2850 	      be32_to_cpu(agf->agf_flfirst) < xfs_agfl_size(mp) &&
2851 	      be32_to_cpu(agf->agf_fllast) < xfs_agfl_size(mp) &&
2852 	      be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp)))
2853 		return __this_address;
2854 
2855 	if (be32_to_cpu(agf->agf_length) > mp->m_sb.sb_dblocks)
2856 		return __this_address;
2857 
2858 	if (be32_to_cpu(agf->agf_freeblks) < be32_to_cpu(agf->agf_longest) ||
2859 	    be32_to_cpu(agf->agf_freeblks) > be32_to_cpu(agf->agf_length))
2860 		return __this_address;
2861 
2862 	if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) < 1 ||
2863 	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) < 1 ||
2864 	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) > XFS_BTREE_MAXLEVELS ||
2865 	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) > XFS_BTREE_MAXLEVELS)
2866 		return __this_address;
2867 
2868 	if (xfs_sb_version_hasrmapbt(&mp->m_sb) &&
2869 	    (be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) < 1 ||
2870 	     be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) > XFS_BTREE_MAXLEVELS))
2871 		return __this_address;
2872 
2873 	if (xfs_sb_version_hasrmapbt(&mp->m_sb) &&
2874 	    be32_to_cpu(agf->agf_rmap_blocks) > be32_to_cpu(agf->agf_length))
2875 		return __this_address;
2876 
2877 	/*
2878 	 * during growfs operations, the perag is not fully initialised,
2879 	 * so we can't use it for any useful checking. growfs ensures we can't
2880 	 * use it by using uncached buffers that don't have the perag attached
2881 	 * so we can detect and avoid this problem.
2882 	 */
2883 	if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno)
2884 		return __this_address;
2885 
2886 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
2887 	    be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length))
2888 		return __this_address;
2889 
2890 	if (xfs_sb_version_hasreflink(&mp->m_sb) &&
2891 	    be32_to_cpu(agf->agf_refcount_blocks) >
2892 	    be32_to_cpu(agf->agf_length))
2893 		return __this_address;
2894 
2895 	if (xfs_sb_version_hasreflink(&mp->m_sb) &&
2896 	    (be32_to_cpu(agf->agf_refcount_level) < 1 ||
2897 	     be32_to_cpu(agf->agf_refcount_level) > XFS_BTREE_MAXLEVELS))
2898 		return __this_address;
2899 
2900 	return NULL;
2901 
2902 }
2903 
2904 static void
2905 xfs_agf_read_verify(
2906 	struct xfs_buf	*bp)
2907 {
2908 	struct xfs_mount *mp = bp->b_mount;
2909 	xfs_failaddr_t	fa;
2910 
2911 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
2912 	    !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF))
2913 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2914 	else {
2915 		fa = xfs_agf_verify(bp);
2916 		if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_ALLOC_READ_AGF))
2917 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2918 	}
2919 }
2920 
2921 static void
2922 xfs_agf_write_verify(
2923 	struct xfs_buf	*bp)
2924 {
2925 	struct xfs_mount	*mp = bp->b_mount;
2926 	struct xfs_buf_log_item	*bip = bp->b_log_item;
2927 	struct xfs_agf		*agf = bp->b_addr;
2928 	xfs_failaddr_t		fa;
2929 
2930 	fa = xfs_agf_verify(bp);
2931 	if (fa) {
2932 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2933 		return;
2934 	}
2935 
2936 	if (!xfs_sb_version_hascrc(&mp->m_sb))
2937 		return;
2938 
2939 	if (bip)
2940 		agf->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2941 
2942 	xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF);
2943 }
2944 
2945 const struct xfs_buf_ops xfs_agf_buf_ops = {
2946 	.name = "xfs_agf",
2947 	.magic = { cpu_to_be32(XFS_AGF_MAGIC), cpu_to_be32(XFS_AGF_MAGIC) },
2948 	.verify_read = xfs_agf_read_verify,
2949 	.verify_write = xfs_agf_write_verify,
2950 	.verify_struct = xfs_agf_verify,
2951 };
2952 
2953 /*
2954  * Read in the allocation group header (free/alloc section).
2955  */
2956 int					/* error */
2957 xfs_read_agf(
2958 	struct xfs_mount	*mp,	/* mount point structure */
2959 	struct xfs_trans	*tp,	/* transaction pointer */
2960 	xfs_agnumber_t		agno,	/* allocation group number */
2961 	int			flags,	/* XFS_BUF_ */
2962 	struct xfs_buf		**bpp)	/* buffer for the ag freelist header */
2963 {
2964 	int		error;
2965 
2966 	trace_xfs_read_agf(mp, agno);
2967 
2968 	ASSERT(agno != NULLAGNUMBER);
2969 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2970 			XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2971 			XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops);
2972 	if (error)
2973 		return error;
2974 
2975 	ASSERT(!(*bpp)->b_error);
2976 	xfs_buf_set_ref(*bpp, XFS_AGF_REF);
2977 	return 0;
2978 }
2979 
2980 /*
2981  * Read in the allocation group header (free/alloc section).
2982  */
2983 int					/* error */
2984 xfs_alloc_read_agf(
2985 	struct xfs_mount	*mp,	/* mount point structure */
2986 	struct xfs_trans	*tp,	/* transaction pointer */
2987 	xfs_agnumber_t		agno,	/* allocation group number */
2988 	int			flags,	/* XFS_ALLOC_FLAG_... */
2989 	struct xfs_buf		**bpp)	/* buffer for the ag freelist header */
2990 {
2991 	struct xfs_agf		*agf;		/* ag freelist header */
2992 	struct xfs_perag	*pag;		/* per allocation group data */
2993 	int			error;
2994 
2995 	trace_xfs_alloc_read_agf(mp, agno);
2996 
2997 	/* We don't support trylock when freeing. */
2998 	ASSERT((flags & (XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK)) !=
2999 			(XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK));
3000 	ASSERT(agno != NULLAGNUMBER);
3001 	error = xfs_read_agf(mp, tp, agno,
3002 			(flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
3003 			bpp);
3004 	if (error)
3005 		return error;
3006 	ASSERT(!(*bpp)->b_error);
3007 
3008 	agf = (*bpp)->b_addr;
3009 	pag = xfs_perag_get(mp, agno);
3010 	if (!pag->pagf_init) {
3011 		pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
3012 		pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
3013 		pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
3014 		pag->pagf_longest = be32_to_cpu(agf->agf_longest);
3015 		pag->pagf_levels[XFS_BTNUM_BNOi] =
3016 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
3017 		pag->pagf_levels[XFS_BTNUM_CNTi] =
3018 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
3019 		pag->pagf_levels[XFS_BTNUM_RMAPi] =
3020 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]);
3021 		pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level);
3022 		pag->pagf_init = 1;
3023 		pag->pagf_agflreset = xfs_agfl_needs_reset(mp, agf);
3024 	}
3025 #ifdef DEBUG
3026 	else if (!XFS_FORCED_SHUTDOWN(mp)) {
3027 		ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
3028 		ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
3029 		ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
3030 		ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
3031 		ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
3032 		       be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
3033 		ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
3034 		       be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
3035 	}
3036 #endif
3037 	xfs_perag_put(pag);
3038 	return 0;
3039 }
3040 
3041 /*
3042  * Allocate an extent (variable-size).
3043  * Depending on the allocation type, we either look in a single allocation
3044  * group or loop over the allocation groups to find the result.
3045  */
3046 int				/* error */
3047 xfs_alloc_vextent(
3048 	struct xfs_alloc_arg	*args)	/* allocation argument structure */
3049 {
3050 	xfs_agblock_t		agsize;	/* allocation group size */
3051 	int			error;
3052 	int			flags;	/* XFS_ALLOC_FLAG_... locking flags */
3053 	struct xfs_mount	*mp;	/* mount structure pointer */
3054 	xfs_agnumber_t		sagno;	/* starting allocation group number */
3055 	xfs_alloctype_t		type;	/* input allocation type */
3056 	int			bump_rotor = 0;
3057 	xfs_agnumber_t		rotorstep = xfs_rotorstep; /* inode32 agf stepper */
3058 
3059 	mp = args->mp;
3060 	type = args->otype = args->type;
3061 	args->agbno = NULLAGBLOCK;
3062 	/*
3063 	 * Just fix this up, for the case where the last a.g. is shorter
3064 	 * (or there's only one a.g.) and the caller couldn't easily figure
3065 	 * that out (xfs_bmap_alloc).
3066 	 */
3067 	agsize = mp->m_sb.sb_agblocks;
3068 	if (args->maxlen > agsize)
3069 		args->maxlen = agsize;
3070 	if (args->alignment == 0)
3071 		args->alignment = 1;
3072 	ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
3073 	ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
3074 	ASSERT(args->minlen <= args->maxlen);
3075 	ASSERT(args->minlen <= agsize);
3076 	ASSERT(args->mod < args->prod);
3077 	if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
3078 	    XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
3079 	    args->minlen > args->maxlen || args->minlen > agsize ||
3080 	    args->mod >= args->prod) {
3081 		args->fsbno = NULLFSBLOCK;
3082 		trace_xfs_alloc_vextent_badargs(args);
3083 		return 0;
3084 	}
3085 
3086 	switch (type) {
3087 	case XFS_ALLOCTYPE_THIS_AG:
3088 	case XFS_ALLOCTYPE_NEAR_BNO:
3089 	case XFS_ALLOCTYPE_THIS_BNO:
3090 		/*
3091 		 * These three force us into a single a.g.
3092 		 */
3093 		args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
3094 		args->pag = xfs_perag_get(mp, args->agno);
3095 		error = xfs_alloc_fix_freelist(args, 0);
3096 		if (error) {
3097 			trace_xfs_alloc_vextent_nofix(args);
3098 			goto error0;
3099 		}
3100 		if (!args->agbp) {
3101 			trace_xfs_alloc_vextent_noagbp(args);
3102 			break;
3103 		}
3104 		args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
3105 		if ((error = xfs_alloc_ag_vextent(args)))
3106 			goto error0;
3107 		break;
3108 	case XFS_ALLOCTYPE_START_BNO:
3109 		/*
3110 		 * Try near allocation first, then anywhere-in-ag after
3111 		 * the first a.g. fails.
3112 		 */
3113 		if ((args->datatype & XFS_ALLOC_INITIAL_USER_DATA) &&
3114 		    (mp->m_flags & XFS_MOUNT_32BITINODES)) {
3115 			args->fsbno = XFS_AGB_TO_FSB(mp,
3116 					((mp->m_agfrotor / rotorstep) %
3117 					mp->m_sb.sb_agcount), 0);
3118 			bump_rotor = 1;
3119 		}
3120 		args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
3121 		args->type = XFS_ALLOCTYPE_NEAR_BNO;
3122 		/* FALLTHROUGH */
3123 	case XFS_ALLOCTYPE_FIRST_AG:
3124 		/*
3125 		 * Rotate through the allocation groups looking for a winner.
3126 		 */
3127 		if (type == XFS_ALLOCTYPE_FIRST_AG) {
3128 			/*
3129 			 * Start with allocation group given by bno.
3130 			 */
3131 			args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
3132 			args->type = XFS_ALLOCTYPE_THIS_AG;
3133 			sagno = 0;
3134 			flags = 0;
3135 		} else {
3136 			/*
3137 			 * Start with the given allocation group.
3138 			 */
3139 			args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
3140 			flags = XFS_ALLOC_FLAG_TRYLOCK;
3141 		}
3142 		/*
3143 		 * Loop over allocation groups twice; first time with
3144 		 * trylock set, second time without.
3145 		 */
3146 		for (;;) {
3147 			args->pag = xfs_perag_get(mp, args->agno);
3148 			error = xfs_alloc_fix_freelist(args, flags);
3149 			if (error) {
3150 				trace_xfs_alloc_vextent_nofix(args);
3151 				goto error0;
3152 			}
3153 			/*
3154 			 * If we get a buffer back then the allocation will fly.
3155 			 */
3156 			if (args->agbp) {
3157 				if ((error = xfs_alloc_ag_vextent(args)))
3158 					goto error0;
3159 				break;
3160 			}
3161 
3162 			trace_xfs_alloc_vextent_loopfailed(args);
3163 
3164 			/*
3165 			 * Didn't work, figure out the next iteration.
3166 			 */
3167 			if (args->agno == sagno &&
3168 			    type == XFS_ALLOCTYPE_START_BNO)
3169 				args->type = XFS_ALLOCTYPE_THIS_AG;
3170 			/*
3171 			* For the first allocation, we can try any AG to get
3172 			* space.  However, if we already have allocated a
3173 			* block, we don't want to try AGs whose number is below
3174 			* sagno. Otherwise, we may end up with out-of-order
3175 			* locking of AGF, which might cause deadlock.
3176 			*/
3177 			if (++(args->agno) == mp->m_sb.sb_agcount) {
3178 				if (args->tp->t_firstblock != NULLFSBLOCK)
3179 					args->agno = sagno;
3180 				else
3181 					args->agno = 0;
3182 			}
3183 			/*
3184 			 * Reached the starting a.g., must either be done
3185 			 * or switch to non-trylock mode.
3186 			 */
3187 			if (args->agno == sagno) {
3188 				if (flags == 0) {
3189 					args->agbno = NULLAGBLOCK;
3190 					trace_xfs_alloc_vextent_allfailed(args);
3191 					break;
3192 				}
3193 
3194 				flags = 0;
3195 				if (type == XFS_ALLOCTYPE_START_BNO) {
3196 					args->agbno = XFS_FSB_TO_AGBNO(mp,
3197 						args->fsbno);
3198 					args->type = XFS_ALLOCTYPE_NEAR_BNO;
3199 				}
3200 			}
3201 			xfs_perag_put(args->pag);
3202 		}
3203 		if (bump_rotor) {
3204 			if (args->agno == sagno)
3205 				mp->m_agfrotor = (mp->m_agfrotor + 1) %
3206 					(mp->m_sb.sb_agcount * rotorstep);
3207 			else
3208 				mp->m_agfrotor = (args->agno * rotorstep + 1) %
3209 					(mp->m_sb.sb_agcount * rotorstep);
3210 		}
3211 		break;
3212 	default:
3213 		ASSERT(0);
3214 		/* NOTREACHED */
3215 	}
3216 	if (args->agbno == NULLAGBLOCK)
3217 		args->fsbno = NULLFSBLOCK;
3218 	else {
3219 		args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
3220 #ifdef DEBUG
3221 		ASSERT(args->len >= args->minlen);
3222 		ASSERT(args->len <= args->maxlen);
3223 		ASSERT(args->agbno % args->alignment == 0);
3224 		XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
3225 			args->len);
3226 #endif
3227 
3228 	}
3229 	xfs_perag_put(args->pag);
3230 	return 0;
3231 error0:
3232 	xfs_perag_put(args->pag);
3233 	return error;
3234 }
3235 
3236 /* Ensure that the freelist is at full capacity. */
3237 int
3238 xfs_free_extent_fix_freelist(
3239 	struct xfs_trans	*tp,
3240 	xfs_agnumber_t		agno,
3241 	struct xfs_buf		**agbp)
3242 {
3243 	struct xfs_alloc_arg	args;
3244 	int			error;
3245 
3246 	memset(&args, 0, sizeof(struct xfs_alloc_arg));
3247 	args.tp = tp;
3248 	args.mp = tp->t_mountp;
3249 	args.agno = agno;
3250 
3251 	/*
3252 	 * validate that the block number is legal - the enables us to detect
3253 	 * and handle a silent filesystem corruption rather than crashing.
3254 	 */
3255 	if (args.agno >= args.mp->m_sb.sb_agcount)
3256 		return -EFSCORRUPTED;
3257 
3258 	args.pag = xfs_perag_get(args.mp, args.agno);
3259 	ASSERT(args.pag);
3260 
3261 	error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
3262 	if (error)
3263 		goto out;
3264 
3265 	*agbp = args.agbp;
3266 out:
3267 	xfs_perag_put(args.pag);
3268 	return error;
3269 }
3270 
3271 /*
3272  * Free an extent.
3273  * Just break up the extent address and hand off to xfs_free_ag_extent
3274  * after fixing up the freelist.
3275  */
3276 int
3277 __xfs_free_extent(
3278 	struct xfs_trans		*tp,
3279 	xfs_fsblock_t			bno,
3280 	xfs_extlen_t			len,
3281 	const struct xfs_owner_info	*oinfo,
3282 	enum xfs_ag_resv_type		type,
3283 	bool				skip_discard)
3284 {
3285 	struct xfs_mount		*mp = tp->t_mountp;
3286 	struct xfs_buf			*agbp;
3287 	xfs_agnumber_t			agno = XFS_FSB_TO_AGNO(mp, bno);
3288 	xfs_agblock_t			agbno = XFS_FSB_TO_AGBNO(mp, bno);
3289 	struct xfs_agf			*agf;
3290 	int				error;
3291 	unsigned int			busy_flags = 0;
3292 
3293 	ASSERT(len != 0);
3294 	ASSERT(type != XFS_AG_RESV_AGFL);
3295 
3296 	if (XFS_TEST_ERROR(false, mp,
3297 			XFS_ERRTAG_FREE_EXTENT))
3298 		return -EIO;
3299 
3300 	error = xfs_free_extent_fix_freelist(tp, agno, &agbp);
3301 	if (error)
3302 		return error;
3303 	agf = agbp->b_addr;
3304 
3305 	if (XFS_IS_CORRUPT(mp, agbno >= mp->m_sb.sb_agblocks)) {
3306 		error = -EFSCORRUPTED;
3307 		goto err;
3308 	}
3309 
3310 	/* validate the extent size is legal now we have the agf locked */
3311 	if (XFS_IS_CORRUPT(mp, agbno + len > be32_to_cpu(agf->agf_length))) {
3312 		error = -EFSCORRUPTED;
3313 		goto err;
3314 	}
3315 
3316 	error = xfs_free_ag_extent(tp, agbp, agno, agbno, len, oinfo, type);
3317 	if (error)
3318 		goto err;
3319 
3320 	if (skip_discard)
3321 		busy_flags |= XFS_EXTENT_BUSY_SKIP_DISCARD;
3322 	xfs_extent_busy_insert(tp, agno, agbno, len, busy_flags);
3323 	return 0;
3324 
3325 err:
3326 	xfs_trans_brelse(tp, agbp);
3327 	return error;
3328 }
3329 
3330 struct xfs_alloc_query_range_info {
3331 	xfs_alloc_query_range_fn	fn;
3332 	void				*priv;
3333 };
3334 
3335 /* Format btree record and pass to our callback. */
3336 STATIC int
3337 xfs_alloc_query_range_helper(
3338 	struct xfs_btree_cur		*cur,
3339 	union xfs_btree_rec		*rec,
3340 	void				*priv)
3341 {
3342 	struct xfs_alloc_query_range_info	*query = priv;
3343 	struct xfs_alloc_rec_incore		irec;
3344 
3345 	irec.ar_startblock = be32_to_cpu(rec->alloc.ar_startblock);
3346 	irec.ar_blockcount = be32_to_cpu(rec->alloc.ar_blockcount);
3347 	return query->fn(cur, &irec, query->priv);
3348 }
3349 
3350 /* Find all free space within a given range of blocks. */
3351 int
3352 xfs_alloc_query_range(
3353 	struct xfs_btree_cur			*cur,
3354 	struct xfs_alloc_rec_incore		*low_rec,
3355 	struct xfs_alloc_rec_incore		*high_rec,
3356 	xfs_alloc_query_range_fn		fn,
3357 	void					*priv)
3358 {
3359 	union xfs_btree_irec			low_brec;
3360 	union xfs_btree_irec			high_brec;
3361 	struct xfs_alloc_query_range_info	query;
3362 
3363 	ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3364 	low_brec.a = *low_rec;
3365 	high_brec.a = *high_rec;
3366 	query.priv = priv;
3367 	query.fn = fn;
3368 	return xfs_btree_query_range(cur, &low_brec, &high_brec,
3369 			xfs_alloc_query_range_helper, &query);
3370 }
3371 
3372 /* Find all free space records. */
3373 int
3374 xfs_alloc_query_all(
3375 	struct xfs_btree_cur			*cur,
3376 	xfs_alloc_query_range_fn		fn,
3377 	void					*priv)
3378 {
3379 	struct xfs_alloc_query_range_info	query;
3380 
3381 	ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3382 	query.priv = priv;
3383 	query.fn = fn;
3384 	return xfs_btree_query_all(cur, xfs_alloc_query_range_helper, &query);
3385 }
3386 
3387 /* Is there a record covering a given extent? */
3388 int
3389 xfs_alloc_has_record(
3390 	struct xfs_btree_cur	*cur,
3391 	xfs_agblock_t		bno,
3392 	xfs_extlen_t		len,
3393 	bool			*exists)
3394 {
3395 	union xfs_btree_irec	low;
3396 	union xfs_btree_irec	high;
3397 
3398 	memset(&low, 0, sizeof(low));
3399 	low.a.ar_startblock = bno;
3400 	memset(&high, 0xFF, sizeof(high));
3401 	high.a.ar_startblock = bno + len - 1;
3402 
3403 	return xfs_btree_has_record(cur, &low, &high, exists);
3404 }
3405 
3406 /*
3407  * Walk all the blocks in the AGFL.  The @walk_fn can return any negative
3408  * error code or XFS_ITER_*.
3409  */
3410 int
3411 xfs_agfl_walk(
3412 	struct xfs_mount	*mp,
3413 	struct xfs_agf		*agf,
3414 	struct xfs_buf		*agflbp,
3415 	xfs_agfl_walk_fn	walk_fn,
3416 	void			*priv)
3417 {
3418 	__be32			*agfl_bno;
3419 	unsigned int		i;
3420 	int			error;
3421 
3422 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
3423 	i = be32_to_cpu(agf->agf_flfirst);
3424 
3425 	/* Nothing to walk in an empty AGFL. */
3426 	if (agf->agf_flcount == cpu_to_be32(0))
3427 		return 0;
3428 
3429 	/* Otherwise, walk from first to last, wrapping as needed. */
3430 	for (;;) {
3431 		error = walk_fn(mp, be32_to_cpu(agfl_bno[i]), priv);
3432 		if (error)
3433 			return error;
3434 		if (i == be32_to_cpu(agf->agf_fllast))
3435 			break;
3436 		if (++i == xfs_agfl_size(mp))
3437 			i = 0;
3438 	}
3439 
3440 	return 0;
3441 }
3442