xref: /openbmc/linux/fs/xfs/libxfs/xfs_alloc.c (revision bc33f5e5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_shared.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_btree.h"
16 #include "xfs_rmap.h"
17 #include "xfs_alloc_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_extent_busy.h"
20 #include "xfs_errortag.h"
21 #include "xfs_error.h"
22 #include "xfs_trace.h"
23 #include "xfs_trans.h"
24 #include "xfs_buf_item.h"
25 #include "xfs_log.h"
26 #include "xfs_ag.h"
27 #include "xfs_ag_resv.h"
28 #include "xfs_bmap.h"
29 
30 struct kmem_cache	*xfs_extfree_item_cache;
31 
32 struct workqueue_struct *xfs_alloc_wq;
33 
34 #define XFS_ABSDIFF(a,b)	(((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
35 
36 #define	XFSA_FIXUP_BNO_OK	1
37 #define	XFSA_FIXUP_CNT_OK	2
38 
39 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_has_crc(mp))
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_has_rmapbt(mp))
65 		return XFS_RMAP_BLOCK(mp) + 1;
66 	if (xfs_has_finobt(mp))
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_has_reflink(mp))
76 		return xfs_refc_block(mp) + 1;
77 	if (xfs_has_rmapbt(mp))
78 		return XFS_RMAP_BLOCK(mp) + 1;
79 	if (xfs_has_finobt(mp))
80 		return XFS_FIBT_BLOCK(mp) + 1;
81 	return XFS_IBT_BLOCK(mp) + 1;
82 }
83 
84 /*
85  * The number of blocks per AG that we withhold from xfs_mod_fdblocks to
86  * guarantee that we can refill the AGFL prior to allocating space in a nearly
87  * full AG.  Although the space described by the free space btrees, the
88  * blocks used by the freesp btrees themselves, and the blocks owned by the
89  * AGFL are counted in the ondisk fdblocks, it's a mistake to let the ondisk
90  * free space in the AG drop so low that the free space btrees cannot refill an
91  * empty AGFL up to the minimum level.  Rather than grind through empty AGs
92  * until the fs goes down, we subtract this many AG blocks from the incore
93  * fdblocks to ensure user allocation does not overcommit the space the
94  * filesystem needs for the AGFLs.  The rmap btree uses a per-AG reservation to
95  * withhold space from xfs_mod_fdblocks, so we do not account for that here.
96  */
97 #define XFS_ALLOCBT_AGFL_RESERVE	4
98 
99 /*
100  * Compute the number of blocks that we set aside to guarantee the ability to
101  * refill the AGFL and handle a full bmap btree split.
102  *
103  * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of
104  * AGF buffer (PV 947395), we place constraints on the relationship among
105  * actual allocations for data blocks, freelist blocks, and potential file data
106  * bmap btree blocks. However, these restrictions may result in no actual space
107  * allocated for a delayed extent, for example, a data block in a certain AG is
108  * allocated but there is no additional block for the additional bmap btree
109  * block due to a split of the bmap btree of the file. The result of this may
110  * lead to an infinite loop when the file gets flushed to disk and all delayed
111  * extents need to be actually allocated. To get around this, we explicitly set
112  * aside a few blocks which will not be reserved in delayed allocation.
113  *
114  * For each AG, we need to reserve enough blocks to replenish a totally empty
115  * AGFL and 4 more to handle a potential split of the file's bmap btree.
116  */
117 unsigned int
118 xfs_alloc_set_aside(
119 	struct xfs_mount	*mp)
120 {
121 	return mp->m_sb.sb_agcount * (XFS_ALLOCBT_AGFL_RESERVE + 4);
122 }
123 
124 /*
125  * When deciding how much space to allocate out of an AG, we limit the
126  * allocation maximum size to the size the AG. However, we cannot use all the
127  * blocks in the AG - some are permanently used by metadata. These
128  * blocks are generally:
129  *	- the AG superblock, AGF, AGI and AGFL
130  *	- the AGF (bno and cnt) and AGI btree root blocks, and optionally
131  *	  the AGI free inode and rmap btree root blocks.
132  *	- blocks on the AGFL according to xfs_alloc_set_aside() limits
133  *	- the rmapbt root block
134  *
135  * The AG headers are sector sized, so the amount of space they take up is
136  * dependent on filesystem geometry. The others are all single blocks.
137  */
138 unsigned int
139 xfs_alloc_ag_max_usable(
140 	struct xfs_mount	*mp)
141 {
142 	unsigned int		blocks;
143 
144 	blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */
145 	blocks += XFS_ALLOCBT_AGFL_RESERVE;
146 	blocks += 3;			/* AGF, AGI btree root blocks */
147 	if (xfs_has_finobt(mp))
148 		blocks++;		/* finobt root block */
149 	if (xfs_has_rmapbt(mp))
150 		blocks++;		/* rmap root block */
151 	if (xfs_has_reflink(mp))
152 		blocks++;		/* refcount root block */
153 
154 	return mp->m_sb.sb_agblocks - blocks;
155 }
156 
157 /*
158  * Lookup the record equal to [bno, len] in the btree given by cur.
159  */
160 STATIC int				/* error */
161 xfs_alloc_lookup_eq(
162 	struct xfs_btree_cur	*cur,	/* btree cursor */
163 	xfs_agblock_t		bno,	/* starting block of extent */
164 	xfs_extlen_t		len,	/* length of extent */
165 	int			*stat)	/* success/failure */
166 {
167 	int			error;
168 
169 	cur->bc_rec.a.ar_startblock = bno;
170 	cur->bc_rec.a.ar_blockcount = len;
171 	error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
172 	cur->bc_ag.abt.active = (*stat == 1);
173 	return error;
174 }
175 
176 /*
177  * Lookup the first record greater than or equal to [bno, len]
178  * in the btree given by cur.
179  */
180 int				/* error */
181 xfs_alloc_lookup_ge(
182 	struct xfs_btree_cur	*cur,	/* btree cursor */
183 	xfs_agblock_t		bno,	/* starting block of extent */
184 	xfs_extlen_t		len,	/* length of extent */
185 	int			*stat)	/* success/failure */
186 {
187 	int			error;
188 
189 	cur->bc_rec.a.ar_startblock = bno;
190 	cur->bc_rec.a.ar_blockcount = len;
191 	error = xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
192 	cur->bc_ag.abt.active = (*stat == 1);
193 	return error;
194 }
195 
196 /*
197  * Lookup the first record less than or equal to [bno, len]
198  * in the btree given by cur.
199  */
200 int					/* error */
201 xfs_alloc_lookup_le(
202 	struct xfs_btree_cur	*cur,	/* btree cursor */
203 	xfs_agblock_t		bno,	/* starting block of extent */
204 	xfs_extlen_t		len,	/* length of extent */
205 	int			*stat)	/* success/failure */
206 {
207 	int			error;
208 	cur->bc_rec.a.ar_startblock = bno;
209 	cur->bc_rec.a.ar_blockcount = len;
210 	error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
211 	cur->bc_ag.abt.active = (*stat == 1);
212 	return error;
213 }
214 
215 static inline bool
216 xfs_alloc_cur_active(
217 	struct xfs_btree_cur	*cur)
218 {
219 	return cur && cur->bc_ag.abt.active;
220 }
221 
222 /*
223  * Update the record referred to by cur to the value given
224  * by [bno, len].
225  * This either works (return 0) or gets an EFSCORRUPTED error.
226  */
227 STATIC int				/* error */
228 xfs_alloc_update(
229 	struct xfs_btree_cur	*cur,	/* btree cursor */
230 	xfs_agblock_t		bno,	/* starting block of extent */
231 	xfs_extlen_t		len)	/* length of extent */
232 {
233 	union xfs_btree_rec	rec;
234 
235 	rec.alloc.ar_startblock = cpu_to_be32(bno);
236 	rec.alloc.ar_blockcount = cpu_to_be32(len);
237 	return xfs_btree_update(cur, &rec);
238 }
239 
240 /*
241  * Get the data from the pointed-to record.
242  */
243 int					/* error */
244 xfs_alloc_get_rec(
245 	struct xfs_btree_cur	*cur,	/* btree cursor */
246 	xfs_agblock_t		*bno,	/* output: starting block of extent */
247 	xfs_extlen_t		*len,	/* output: length of extent */
248 	int			*stat)	/* output: success/failure */
249 {
250 	struct xfs_mount	*mp = cur->bc_mp;
251 	struct xfs_perag	*pag = cur->bc_ag.pag;
252 	union xfs_btree_rec	*rec;
253 	int			error;
254 
255 	error = xfs_btree_get_rec(cur, &rec, stat);
256 	if (error || !(*stat))
257 		return error;
258 
259 	*bno = be32_to_cpu(rec->alloc.ar_startblock);
260 	*len = be32_to_cpu(rec->alloc.ar_blockcount);
261 
262 	if (*len == 0)
263 		goto out_bad_rec;
264 
265 	/* check for valid extent range, including overflow */
266 	if (!xfs_verify_agbext(pag, *bno, *len))
267 		goto out_bad_rec;
268 
269 	return 0;
270 
271 out_bad_rec:
272 	xfs_warn(mp,
273 		"%s Freespace BTree record corruption in AG %d detected!",
274 		cur->bc_btnum == XFS_BTNUM_BNO ? "Block" : "Size",
275 		pag->pag_agno);
276 	xfs_warn(mp,
277 		"start block 0x%x block count 0x%x", *bno, *len);
278 	return -EFSCORRUPTED;
279 }
280 
281 /*
282  * Compute aligned version of the found extent.
283  * Takes alignment and min length into account.
284  */
285 STATIC bool
286 xfs_alloc_compute_aligned(
287 	xfs_alloc_arg_t	*args,		/* allocation argument structure */
288 	xfs_agblock_t	foundbno,	/* starting block in found extent */
289 	xfs_extlen_t	foundlen,	/* length in found extent */
290 	xfs_agblock_t	*resbno,	/* result block number */
291 	xfs_extlen_t	*reslen,	/* result length */
292 	unsigned	*busy_gen)
293 {
294 	xfs_agblock_t	bno = foundbno;
295 	xfs_extlen_t	len = foundlen;
296 	xfs_extlen_t	diff;
297 	bool		busy;
298 
299 	/* Trim busy sections out of found extent */
300 	busy = xfs_extent_busy_trim(args, &bno, &len, busy_gen);
301 
302 	/*
303 	 * If we have a largish extent that happens to start before min_agbno,
304 	 * see if we can shift it into range...
305 	 */
306 	if (bno < args->min_agbno && bno + len > args->min_agbno) {
307 		diff = args->min_agbno - bno;
308 		if (len > diff) {
309 			bno += diff;
310 			len -= diff;
311 		}
312 	}
313 
314 	if (args->alignment > 1 && len >= args->minlen) {
315 		xfs_agblock_t	aligned_bno = roundup(bno, args->alignment);
316 
317 		diff = aligned_bno - bno;
318 
319 		*resbno = aligned_bno;
320 		*reslen = diff >= len ? 0 : len - diff;
321 	} else {
322 		*resbno = bno;
323 		*reslen = len;
324 	}
325 
326 	return busy;
327 }
328 
329 /*
330  * Compute best start block and diff for "near" allocations.
331  * freelen >= wantlen already checked by caller.
332  */
333 STATIC xfs_extlen_t			/* difference value (absolute) */
334 xfs_alloc_compute_diff(
335 	xfs_agblock_t	wantbno,	/* target starting block */
336 	xfs_extlen_t	wantlen,	/* target length */
337 	xfs_extlen_t	alignment,	/* target alignment */
338 	int		datatype,	/* are we allocating data? */
339 	xfs_agblock_t	freebno,	/* freespace's starting block */
340 	xfs_extlen_t	freelen,	/* freespace's length */
341 	xfs_agblock_t	*newbnop)	/* result: best start block from free */
342 {
343 	xfs_agblock_t	freeend;	/* end of freespace extent */
344 	xfs_agblock_t	newbno1;	/* return block number */
345 	xfs_agblock_t	newbno2;	/* other new block number */
346 	xfs_extlen_t	newlen1=0;	/* length with newbno1 */
347 	xfs_extlen_t	newlen2=0;	/* length with newbno2 */
348 	xfs_agblock_t	wantend;	/* end of target extent */
349 	bool		userdata = datatype & XFS_ALLOC_USERDATA;
350 
351 	ASSERT(freelen >= wantlen);
352 	freeend = freebno + freelen;
353 	wantend = wantbno + wantlen;
354 	/*
355 	 * We want to allocate from the start of a free extent if it is past
356 	 * the desired block or if we are allocating user data and the free
357 	 * extent is before desired block. The second case is there to allow
358 	 * for contiguous allocation from the remaining free space if the file
359 	 * grows in the short term.
360 	 */
361 	if (freebno >= wantbno || (userdata && freeend < wantend)) {
362 		if ((newbno1 = roundup(freebno, alignment)) >= freeend)
363 			newbno1 = NULLAGBLOCK;
364 	} else if (freeend >= wantend && alignment > 1) {
365 		newbno1 = roundup(wantbno, alignment);
366 		newbno2 = newbno1 - alignment;
367 		if (newbno1 >= freeend)
368 			newbno1 = NULLAGBLOCK;
369 		else
370 			newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
371 		if (newbno2 < freebno)
372 			newbno2 = NULLAGBLOCK;
373 		else
374 			newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
375 		if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
376 			if (newlen1 < newlen2 ||
377 			    (newlen1 == newlen2 &&
378 			     XFS_ABSDIFF(newbno1, wantbno) >
379 			     XFS_ABSDIFF(newbno2, wantbno)))
380 				newbno1 = newbno2;
381 		} else if (newbno2 != NULLAGBLOCK)
382 			newbno1 = newbno2;
383 	} else if (freeend >= wantend) {
384 		newbno1 = wantbno;
385 	} else if (alignment > 1) {
386 		newbno1 = roundup(freeend - wantlen, alignment);
387 		if (newbno1 > freeend - wantlen &&
388 		    newbno1 - alignment >= freebno)
389 			newbno1 -= alignment;
390 		else if (newbno1 >= freeend)
391 			newbno1 = NULLAGBLOCK;
392 	} else
393 		newbno1 = freeend - wantlen;
394 	*newbnop = newbno1;
395 	return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
396 }
397 
398 /*
399  * Fix up the length, based on mod and prod.
400  * len should be k * prod + mod for some k.
401  * If len is too small it is returned unchanged.
402  * If len hits maxlen it is left alone.
403  */
404 STATIC void
405 xfs_alloc_fix_len(
406 	xfs_alloc_arg_t	*args)		/* allocation argument structure */
407 {
408 	xfs_extlen_t	k;
409 	xfs_extlen_t	rlen;
410 
411 	ASSERT(args->mod < args->prod);
412 	rlen = args->len;
413 	ASSERT(rlen >= args->minlen);
414 	ASSERT(rlen <= args->maxlen);
415 	if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
416 	    (args->mod == 0 && rlen < args->prod))
417 		return;
418 	k = rlen % args->prod;
419 	if (k == args->mod)
420 		return;
421 	if (k > args->mod)
422 		rlen = rlen - (k - args->mod);
423 	else
424 		rlen = rlen - args->prod + (args->mod - k);
425 	/* casts to (int) catch length underflows */
426 	if ((int)rlen < (int)args->minlen)
427 		return;
428 	ASSERT(rlen >= args->minlen && rlen <= args->maxlen);
429 	ASSERT(rlen % args->prod == args->mod);
430 	ASSERT(args->pag->pagf_freeblks + args->pag->pagf_flcount >=
431 		rlen + args->minleft);
432 	args->len = rlen;
433 }
434 
435 /*
436  * Update the two btrees, logically removing from freespace the extent
437  * starting at rbno, rlen blocks.  The extent is contained within the
438  * actual (current) free extent fbno for flen blocks.
439  * Flags are passed in indicating whether the cursors are set to the
440  * relevant records.
441  */
442 STATIC int				/* error code */
443 xfs_alloc_fixup_trees(
444 	struct xfs_btree_cur *cnt_cur,	/* cursor for by-size btree */
445 	struct xfs_btree_cur *bno_cur,	/* cursor for by-block btree */
446 	xfs_agblock_t	fbno,		/* starting block of free extent */
447 	xfs_extlen_t	flen,		/* length of free extent */
448 	xfs_agblock_t	rbno,		/* starting block of returned extent */
449 	xfs_extlen_t	rlen,		/* length of returned extent */
450 	int		flags)		/* flags, XFSA_FIXUP_... */
451 {
452 	int		error;		/* error code */
453 	int		i;		/* operation results */
454 	xfs_agblock_t	nfbno1;		/* first new free startblock */
455 	xfs_agblock_t	nfbno2;		/* second new free startblock */
456 	xfs_extlen_t	nflen1=0;	/* first new free length */
457 	xfs_extlen_t	nflen2=0;	/* second new free length */
458 	struct xfs_mount *mp;
459 
460 	mp = cnt_cur->bc_mp;
461 
462 	/*
463 	 * Look up the record in the by-size tree if necessary.
464 	 */
465 	if (flags & XFSA_FIXUP_CNT_OK) {
466 #ifdef DEBUG
467 		if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
468 			return error;
469 		if (XFS_IS_CORRUPT(mp,
470 				   i != 1 ||
471 				   nfbno1 != fbno ||
472 				   nflen1 != flen))
473 			return -EFSCORRUPTED;
474 #endif
475 	} else {
476 		if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
477 			return error;
478 		if (XFS_IS_CORRUPT(mp, i != 1))
479 			return -EFSCORRUPTED;
480 	}
481 	/*
482 	 * Look up the record in the by-block tree if necessary.
483 	 */
484 	if (flags & XFSA_FIXUP_BNO_OK) {
485 #ifdef DEBUG
486 		if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
487 			return error;
488 		if (XFS_IS_CORRUPT(mp,
489 				   i != 1 ||
490 				   nfbno1 != fbno ||
491 				   nflen1 != flen))
492 			return -EFSCORRUPTED;
493 #endif
494 	} else {
495 		if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
496 			return error;
497 		if (XFS_IS_CORRUPT(mp, i != 1))
498 			return -EFSCORRUPTED;
499 	}
500 
501 #ifdef DEBUG
502 	if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
503 		struct xfs_btree_block	*bnoblock;
504 		struct xfs_btree_block	*cntblock;
505 
506 		bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_levels[0].bp);
507 		cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_levels[0].bp);
508 
509 		if (XFS_IS_CORRUPT(mp,
510 				   bnoblock->bb_numrecs !=
511 				   cntblock->bb_numrecs))
512 			return -EFSCORRUPTED;
513 	}
514 #endif
515 
516 	/*
517 	 * Deal with all four cases: the allocated record is contained
518 	 * within the freespace record, so we can have new freespace
519 	 * at either (or both) end, or no freespace remaining.
520 	 */
521 	if (rbno == fbno && rlen == flen)
522 		nfbno1 = nfbno2 = NULLAGBLOCK;
523 	else if (rbno == fbno) {
524 		nfbno1 = rbno + rlen;
525 		nflen1 = flen - rlen;
526 		nfbno2 = NULLAGBLOCK;
527 	} else if (rbno + rlen == fbno + flen) {
528 		nfbno1 = fbno;
529 		nflen1 = flen - rlen;
530 		nfbno2 = NULLAGBLOCK;
531 	} else {
532 		nfbno1 = fbno;
533 		nflen1 = rbno - fbno;
534 		nfbno2 = rbno + rlen;
535 		nflen2 = (fbno + flen) - nfbno2;
536 	}
537 	/*
538 	 * Delete the entry from the by-size btree.
539 	 */
540 	if ((error = xfs_btree_delete(cnt_cur, &i)))
541 		return error;
542 	if (XFS_IS_CORRUPT(mp, i != 1))
543 		return -EFSCORRUPTED;
544 	/*
545 	 * Add new by-size btree entry(s).
546 	 */
547 	if (nfbno1 != NULLAGBLOCK) {
548 		if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
549 			return error;
550 		if (XFS_IS_CORRUPT(mp, i != 0))
551 			return -EFSCORRUPTED;
552 		if ((error = xfs_btree_insert(cnt_cur, &i)))
553 			return error;
554 		if (XFS_IS_CORRUPT(mp, i != 1))
555 			return -EFSCORRUPTED;
556 	}
557 	if (nfbno2 != NULLAGBLOCK) {
558 		if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
559 			return error;
560 		if (XFS_IS_CORRUPT(mp, i != 0))
561 			return -EFSCORRUPTED;
562 		if ((error = xfs_btree_insert(cnt_cur, &i)))
563 			return error;
564 		if (XFS_IS_CORRUPT(mp, i != 1))
565 			return -EFSCORRUPTED;
566 	}
567 	/*
568 	 * Fix up the by-block btree entry(s).
569 	 */
570 	if (nfbno1 == NULLAGBLOCK) {
571 		/*
572 		 * No remaining freespace, just delete the by-block tree entry.
573 		 */
574 		if ((error = xfs_btree_delete(bno_cur, &i)))
575 			return error;
576 		if (XFS_IS_CORRUPT(mp, i != 1))
577 			return -EFSCORRUPTED;
578 	} else {
579 		/*
580 		 * Update the by-block entry to start later|be shorter.
581 		 */
582 		if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
583 			return error;
584 	}
585 	if (nfbno2 != NULLAGBLOCK) {
586 		/*
587 		 * 2 resulting free entries, need to add one.
588 		 */
589 		if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
590 			return error;
591 		if (XFS_IS_CORRUPT(mp, i != 0))
592 			return -EFSCORRUPTED;
593 		if ((error = xfs_btree_insert(bno_cur, &i)))
594 			return error;
595 		if (XFS_IS_CORRUPT(mp, i != 1))
596 			return -EFSCORRUPTED;
597 	}
598 	return 0;
599 }
600 
601 static xfs_failaddr_t
602 xfs_agfl_verify(
603 	struct xfs_buf	*bp)
604 {
605 	struct xfs_mount *mp = bp->b_mount;
606 	struct xfs_agfl	*agfl = XFS_BUF_TO_AGFL(bp);
607 	__be32		*agfl_bno = xfs_buf_to_agfl_bno(bp);
608 	int		i;
609 
610 	/*
611 	 * There is no verification of non-crc AGFLs because mkfs does not
612 	 * initialise the AGFL to zero or NULL. Hence the only valid part of the
613 	 * AGFL is what the AGF says is active. We can't get to the AGF, so we
614 	 * can't verify just those entries are valid.
615 	 */
616 	if (!xfs_has_crc(mp))
617 		return NULL;
618 
619 	if (!xfs_verify_magic(bp, agfl->agfl_magicnum))
620 		return __this_address;
621 	if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid))
622 		return __this_address;
623 	/*
624 	 * during growfs operations, the perag is not fully initialised,
625 	 * so we can't use it for any useful checking. growfs ensures we can't
626 	 * use it by using uncached buffers that don't have the perag attached
627 	 * so we can detect and avoid this problem.
628 	 */
629 	if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno)
630 		return __this_address;
631 
632 	for (i = 0; i < xfs_agfl_size(mp); i++) {
633 		if (be32_to_cpu(agfl_bno[i]) != NULLAGBLOCK &&
634 		    be32_to_cpu(agfl_bno[i]) >= mp->m_sb.sb_agblocks)
635 			return __this_address;
636 	}
637 
638 	if (!xfs_log_check_lsn(mp, be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)))
639 		return __this_address;
640 	return NULL;
641 }
642 
643 static void
644 xfs_agfl_read_verify(
645 	struct xfs_buf	*bp)
646 {
647 	struct xfs_mount *mp = bp->b_mount;
648 	xfs_failaddr_t	fa;
649 
650 	/*
651 	 * There is no verification of non-crc AGFLs because mkfs does not
652 	 * initialise the AGFL to zero or NULL. Hence the only valid part of the
653 	 * AGFL is what the AGF says is active. We can't get to the AGF, so we
654 	 * can't verify just those entries are valid.
655 	 */
656 	if (!xfs_has_crc(mp))
657 		return;
658 
659 	if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF))
660 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
661 	else {
662 		fa = xfs_agfl_verify(bp);
663 		if (fa)
664 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
665 	}
666 }
667 
668 static void
669 xfs_agfl_write_verify(
670 	struct xfs_buf	*bp)
671 {
672 	struct xfs_mount	*mp = bp->b_mount;
673 	struct xfs_buf_log_item	*bip = bp->b_log_item;
674 	xfs_failaddr_t		fa;
675 
676 	/* no verification of non-crc AGFLs */
677 	if (!xfs_has_crc(mp))
678 		return;
679 
680 	fa = xfs_agfl_verify(bp);
681 	if (fa) {
682 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
683 		return;
684 	}
685 
686 	if (bip)
687 		XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn);
688 
689 	xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF);
690 }
691 
692 const struct xfs_buf_ops xfs_agfl_buf_ops = {
693 	.name = "xfs_agfl",
694 	.magic = { cpu_to_be32(XFS_AGFL_MAGIC), cpu_to_be32(XFS_AGFL_MAGIC) },
695 	.verify_read = xfs_agfl_read_verify,
696 	.verify_write = xfs_agfl_write_verify,
697 	.verify_struct = xfs_agfl_verify,
698 };
699 
700 /*
701  * Read in the allocation group free block array.
702  */
703 int
704 xfs_alloc_read_agfl(
705 	struct xfs_perag	*pag,
706 	struct xfs_trans	*tp,
707 	struct xfs_buf		**bpp)
708 {
709 	struct xfs_mount	*mp = pag->pag_mount;
710 	struct xfs_buf		*bp;
711 	int			error;
712 
713 	error = xfs_trans_read_buf(
714 			mp, tp, mp->m_ddev_targp,
715 			XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGFL_DADDR(mp)),
716 			XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops);
717 	if (error)
718 		return error;
719 	xfs_buf_set_ref(bp, XFS_AGFL_REF);
720 	*bpp = bp;
721 	return 0;
722 }
723 
724 STATIC int
725 xfs_alloc_update_counters(
726 	struct xfs_trans	*tp,
727 	struct xfs_buf		*agbp,
728 	long			len)
729 {
730 	struct xfs_agf		*agf = agbp->b_addr;
731 
732 	agbp->b_pag->pagf_freeblks += len;
733 	be32_add_cpu(&agf->agf_freeblks, len);
734 
735 	if (unlikely(be32_to_cpu(agf->agf_freeblks) >
736 		     be32_to_cpu(agf->agf_length))) {
737 		xfs_buf_mark_corrupt(agbp);
738 		return -EFSCORRUPTED;
739 	}
740 
741 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
742 	return 0;
743 }
744 
745 /*
746  * Block allocation algorithm and data structures.
747  */
748 struct xfs_alloc_cur {
749 	struct xfs_btree_cur		*cnt;	/* btree cursors */
750 	struct xfs_btree_cur		*bnolt;
751 	struct xfs_btree_cur		*bnogt;
752 	xfs_extlen_t			cur_len;/* current search length */
753 	xfs_agblock_t			rec_bno;/* extent startblock */
754 	xfs_extlen_t			rec_len;/* extent length */
755 	xfs_agblock_t			bno;	/* alloc bno */
756 	xfs_extlen_t			len;	/* alloc len */
757 	xfs_extlen_t			diff;	/* diff from search bno */
758 	unsigned int			busy_gen;/* busy state */
759 	bool				busy;
760 };
761 
762 /*
763  * Set up cursors, etc. in the extent allocation cursor. This function can be
764  * called multiple times to reset an initialized structure without having to
765  * reallocate cursors.
766  */
767 static int
768 xfs_alloc_cur_setup(
769 	struct xfs_alloc_arg	*args,
770 	struct xfs_alloc_cur	*acur)
771 {
772 	int			error;
773 	int			i;
774 
775 	ASSERT(args->alignment == 1 || args->type != XFS_ALLOCTYPE_THIS_BNO);
776 
777 	acur->cur_len = args->maxlen;
778 	acur->rec_bno = 0;
779 	acur->rec_len = 0;
780 	acur->bno = 0;
781 	acur->len = 0;
782 	acur->diff = -1;
783 	acur->busy = false;
784 	acur->busy_gen = 0;
785 
786 	/*
787 	 * Perform an initial cntbt lookup to check for availability of maxlen
788 	 * extents. If this fails, we'll return -ENOSPC to signal the caller to
789 	 * attempt a small allocation.
790 	 */
791 	if (!acur->cnt)
792 		acur->cnt = xfs_allocbt_init_cursor(args->mp, args->tp,
793 					args->agbp, args->pag, XFS_BTNUM_CNT);
794 	error = xfs_alloc_lookup_ge(acur->cnt, 0, args->maxlen, &i);
795 	if (error)
796 		return error;
797 
798 	/*
799 	 * Allocate the bnobt left and right search cursors.
800 	 */
801 	if (!acur->bnolt)
802 		acur->bnolt = xfs_allocbt_init_cursor(args->mp, args->tp,
803 					args->agbp, args->pag, XFS_BTNUM_BNO);
804 	if (!acur->bnogt)
805 		acur->bnogt = xfs_allocbt_init_cursor(args->mp, args->tp,
806 					args->agbp, args->pag, XFS_BTNUM_BNO);
807 	return i == 1 ? 0 : -ENOSPC;
808 }
809 
810 static void
811 xfs_alloc_cur_close(
812 	struct xfs_alloc_cur	*acur,
813 	bool			error)
814 {
815 	int			cur_error = XFS_BTREE_NOERROR;
816 
817 	if (error)
818 		cur_error = XFS_BTREE_ERROR;
819 
820 	if (acur->cnt)
821 		xfs_btree_del_cursor(acur->cnt, cur_error);
822 	if (acur->bnolt)
823 		xfs_btree_del_cursor(acur->bnolt, cur_error);
824 	if (acur->bnogt)
825 		xfs_btree_del_cursor(acur->bnogt, cur_error);
826 	acur->cnt = acur->bnolt = acur->bnogt = NULL;
827 }
828 
829 /*
830  * Check an extent for allocation and track the best available candidate in the
831  * allocation structure. The cursor is deactivated if it has entered an out of
832  * range state based on allocation arguments. Optionally return the extent
833  * extent geometry and allocation status if requested by the caller.
834  */
835 static int
836 xfs_alloc_cur_check(
837 	struct xfs_alloc_arg	*args,
838 	struct xfs_alloc_cur	*acur,
839 	struct xfs_btree_cur	*cur,
840 	int			*new)
841 {
842 	int			error, i;
843 	xfs_agblock_t		bno, bnoa, bnew;
844 	xfs_extlen_t		len, lena, diff = -1;
845 	bool			busy;
846 	unsigned		busy_gen = 0;
847 	bool			deactivate = false;
848 	bool			isbnobt = cur->bc_btnum == XFS_BTNUM_BNO;
849 
850 	*new = 0;
851 
852 	error = xfs_alloc_get_rec(cur, &bno, &len, &i);
853 	if (error)
854 		return error;
855 	if (XFS_IS_CORRUPT(args->mp, i != 1))
856 		return -EFSCORRUPTED;
857 
858 	/*
859 	 * Check minlen and deactivate a cntbt cursor if out of acceptable size
860 	 * range (i.e., walking backwards looking for a minlen extent).
861 	 */
862 	if (len < args->minlen) {
863 		deactivate = !isbnobt;
864 		goto out;
865 	}
866 
867 	busy = xfs_alloc_compute_aligned(args, bno, len, &bnoa, &lena,
868 					 &busy_gen);
869 	acur->busy |= busy;
870 	if (busy)
871 		acur->busy_gen = busy_gen;
872 	/* deactivate a bnobt cursor outside of locality range */
873 	if (bnoa < args->min_agbno || bnoa > args->max_agbno) {
874 		deactivate = isbnobt;
875 		goto out;
876 	}
877 	if (lena < args->minlen)
878 		goto out;
879 
880 	args->len = XFS_EXTLEN_MIN(lena, args->maxlen);
881 	xfs_alloc_fix_len(args);
882 	ASSERT(args->len >= args->minlen);
883 	if (args->len < acur->len)
884 		goto out;
885 
886 	/*
887 	 * We have an aligned record that satisfies minlen and beats or matches
888 	 * the candidate extent size. Compare locality for near allocation mode.
889 	 */
890 	ASSERT(args->type == XFS_ALLOCTYPE_NEAR_BNO);
891 	diff = xfs_alloc_compute_diff(args->agbno, args->len,
892 				      args->alignment, args->datatype,
893 				      bnoa, lena, &bnew);
894 	if (bnew == NULLAGBLOCK)
895 		goto out;
896 
897 	/*
898 	 * Deactivate a bnobt cursor with worse locality than the current best.
899 	 */
900 	if (diff > acur->diff) {
901 		deactivate = isbnobt;
902 		goto out;
903 	}
904 
905 	ASSERT(args->len > acur->len ||
906 	       (args->len == acur->len && diff <= acur->diff));
907 	acur->rec_bno = bno;
908 	acur->rec_len = len;
909 	acur->bno = bnew;
910 	acur->len = args->len;
911 	acur->diff = diff;
912 	*new = 1;
913 
914 	/*
915 	 * We're done if we found a perfect allocation. This only deactivates
916 	 * the current cursor, but this is just an optimization to terminate a
917 	 * cntbt search that otherwise runs to the edge of the tree.
918 	 */
919 	if (acur->diff == 0 && acur->len == args->maxlen)
920 		deactivate = true;
921 out:
922 	if (deactivate)
923 		cur->bc_ag.abt.active = false;
924 	trace_xfs_alloc_cur_check(args->mp, cur->bc_btnum, bno, len, diff,
925 				  *new);
926 	return 0;
927 }
928 
929 /*
930  * Complete an allocation of a candidate extent. Remove the extent from both
931  * trees and update the args structure.
932  */
933 STATIC int
934 xfs_alloc_cur_finish(
935 	struct xfs_alloc_arg	*args,
936 	struct xfs_alloc_cur	*acur)
937 {
938 	struct xfs_agf __maybe_unused *agf = args->agbp->b_addr;
939 	int			error;
940 
941 	ASSERT(acur->cnt && acur->bnolt);
942 	ASSERT(acur->bno >= acur->rec_bno);
943 	ASSERT(acur->bno + acur->len <= acur->rec_bno + acur->rec_len);
944 	ASSERT(acur->rec_bno + acur->rec_len <= be32_to_cpu(agf->agf_length));
945 
946 	error = xfs_alloc_fixup_trees(acur->cnt, acur->bnolt, acur->rec_bno,
947 				      acur->rec_len, acur->bno, acur->len, 0);
948 	if (error)
949 		return error;
950 
951 	args->agbno = acur->bno;
952 	args->len = acur->len;
953 	args->wasfromfl = 0;
954 
955 	trace_xfs_alloc_cur(args);
956 	return 0;
957 }
958 
959 /*
960  * Locality allocation lookup algorithm. This expects a cntbt cursor and uses
961  * bno optimized lookup to search for extents with ideal size and locality.
962  */
963 STATIC int
964 xfs_alloc_cntbt_iter(
965 	struct xfs_alloc_arg		*args,
966 	struct xfs_alloc_cur		*acur)
967 {
968 	struct xfs_btree_cur	*cur = acur->cnt;
969 	xfs_agblock_t		bno;
970 	xfs_extlen_t		len, cur_len;
971 	int			error;
972 	int			i;
973 
974 	if (!xfs_alloc_cur_active(cur))
975 		return 0;
976 
977 	/* locality optimized lookup */
978 	cur_len = acur->cur_len;
979 	error = xfs_alloc_lookup_ge(cur, args->agbno, cur_len, &i);
980 	if (error)
981 		return error;
982 	if (i == 0)
983 		return 0;
984 	error = xfs_alloc_get_rec(cur, &bno, &len, &i);
985 	if (error)
986 		return error;
987 
988 	/* check the current record and update search length from it */
989 	error = xfs_alloc_cur_check(args, acur, cur, &i);
990 	if (error)
991 		return error;
992 	ASSERT(len >= acur->cur_len);
993 	acur->cur_len = len;
994 
995 	/*
996 	 * We looked up the first record >= [agbno, len] above. The agbno is a
997 	 * secondary key and so the current record may lie just before or after
998 	 * agbno. If it is past agbno, check the previous record too so long as
999 	 * the length matches as it may be closer. Don't check a smaller record
1000 	 * because that could deactivate our cursor.
1001 	 */
1002 	if (bno > args->agbno) {
1003 		error = xfs_btree_decrement(cur, 0, &i);
1004 		if (!error && i) {
1005 			error = xfs_alloc_get_rec(cur, &bno, &len, &i);
1006 			if (!error && i && len == acur->cur_len)
1007 				error = xfs_alloc_cur_check(args, acur, cur,
1008 							    &i);
1009 		}
1010 		if (error)
1011 			return error;
1012 	}
1013 
1014 	/*
1015 	 * Increment the search key until we find at least one allocation
1016 	 * candidate or if the extent we found was larger. Otherwise, double the
1017 	 * search key to optimize the search. Efficiency is more important here
1018 	 * than absolute best locality.
1019 	 */
1020 	cur_len <<= 1;
1021 	if (!acur->len || acur->cur_len >= cur_len)
1022 		acur->cur_len++;
1023 	else
1024 		acur->cur_len = cur_len;
1025 
1026 	return error;
1027 }
1028 
1029 /*
1030  * Deal with the case where only small freespaces remain. Either return the
1031  * contents of the last freespace record, or allocate space from the freelist if
1032  * there is nothing in the tree.
1033  */
1034 STATIC int			/* error */
1035 xfs_alloc_ag_vextent_small(
1036 	struct xfs_alloc_arg	*args,	/* allocation argument structure */
1037 	struct xfs_btree_cur	*ccur,	/* optional by-size cursor */
1038 	xfs_agblock_t		*fbnop,	/* result block number */
1039 	xfs_extlen_t		*flenp,	/* result length */
1040 	int			*stat)	/* status: 0-freelist, 1-normal/none */
1041 {
1042 	struct xfs_agf		*agf = args->agbp->b_addr;
1043 	int			error = 0;
1044 	xfs_agblock_t		fbno = NULLAGBLOCK;
1045 	xfs_extlen_t		flen = 0;
1046 	int			i = 0;
1047 
1048 	/*
1049 	 * If a cntbt cursor is provided, try to allocate the largest record in
1050 	 * the tree. Try the AGFL if the cntbt is empty, otherwise fail the
1051 	 * allocation. Make sure to respect minleft even when pulling from the
1052 	 * freelist.
1053 	 */
1054 	if (ccur)
1055 		error = xfs_btree_decrement(ccur, 0, &i);
1056 	if (error)
1057 		goto error;
1058 	if (i) {
1059 		error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i);
1060 		if (error)
1061 			goto error;
1062 		if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1063 			error = -EFSCORRUPTED;
1064 			goto error;
1065 		}
1066 		goto out;
1067 	}
1068 
1069 	if (args->minlen != 1 || args->alignment != 1 ||
1070 	    args->resv == XFS_AG_RESV_AGFL ||
1071 	    be32_to_cpu(agf->agf_flcount) <= args->minleft)
1072 		goto out;
1073 
1074 	error = xfs_alloc_get_freelist(args->pag, args->tp, args->agbp,
1075 			&fbno, 0);
1076 	if (error)
1077 		goto error;
1078 	if (fbno == NULLAGBLOCK)
1079 		goto out;
1080 
1081 	xfs_extent_busy_reuse(args->mp, args->pag, fbno, 1,
1082 			      (args->datatype & XFS_ALLOC_NOBUSY));
1083 
1084 	if (args->datatype & XFS_ALLOC_USERDATA) {
1085 		struct xfs_buf	*bp;
1086 
1087 		error = xfs_trans_get_buf(args->tp, args->mp->m_ddev_targp,
1088 				XFS_AGB_TO_DADDR(args->mp, args->agno, fbno),
1089 				args->mp->m_bsize, 0, &bp);
1090 		if (error)
1091 			goto error;
1092 		xfs_trans_binval(args->tp, bp);
1093 	}
1094 	*fbnop = args->agbno = fbno;
1095 	*flenp = args->len = 1;
1096 	if (XFS_IS_CORRUPT(args->mp, fbno >= be32_to_cpu(agf->agf_length))) {
1097 		error = -EFSCORRUPTED;
1098 		goto error;
1099 	}
1100 	args->wasfromfl = 1;
1101 	trace_xfs_alloc_small_freelist(args);
1102 
1103 	/*
1104 	 * If we're feeding an AGFL block to something that doesn't live in the
1105 	 * free space, we need to clear out the OWN_AG rmap.
1106 	 */
1107 	error = xfs_rmap_free(args->tp, args->agbp, args->pag, fbno, 1,
1108 			      &XFS_RMAP_OINFO_AG);
1109 	if (error)
1110 		goto error;
1111 
1112 	*stat = 0;
1113 	return 0;
1114 
1115 out:
1116 	/*
1117 	 * Can't do the allocation, give up.
1118 	 */
1119 	if (flen < args->minlen) {
1120 		args->agbno = NULLAGBLOCK;
1121 		trace_xfs_alloc_small_notenough(args);
1122 		flen = 0;
1123 	}
1124 	*fbnop = fbno;
1125 	*flenp = flen;
1126 	*stat = 1;
1127 	trace_xfs_alloc_small_done(args);
1128 	return 0;
1129 
1130 error:
1131 	trace_xfs_alloc_small_error(args);
1132 	return error;
1133 }
1134 
1135 /*
1136  * Allocate a variable extent in the allocation group agno.
1137  * Type and bno are used to determine where in the allocation group the
1138  * extent will start.
1139  * Extent's length (returned in *len) will be between minlen and maxlen,
1140  * and of the form k * prod + mod unless there's nothing that large.
1141  * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1142  */
1143 STATIC int			/* error */
1144 xfs_alloc_ag_vextent(
1145 	xfs_alloc_arg_t	*args)	/* argument structure for allocation */
1146 {
1147 	int		error=0;
1148 
1149 	ASSERT(args->minlen > 0);
1150 	ASSERT(args->maxlen > 0);
1151 	ASSERT(args->minlen <= args->maxlen);
1152 	ASSERT(args->mod < args->prod);
1153 	ASSERT(args->alignment > 0);
1154 
1155 	/*
1156 	 * Branch to correct routine based on the type.
1157 	 */
1158 	args->wasfromfl = 0;
1159 	switch (args->type) {
1160 	case XFS_ALLOCTYPE_THIS_AG:
1161 		error = xfs_alloc_ag_vextent_size(args);
1162 		break;
1163 	case XFS_ALLOCTYPE_NEAR_BNO:
1164 		error = xfs_alloc_ag_vextent_near(args);
1165 		break;
1166 	case XFS_ALLOCTYPE_THIS_BNO:
1167 		error = xfs_alloc_ag_vextent_exact(args);
1168 		break;
1169 	default:
1170 		ASSERT(0);
1171 		/* NOTREACHED */
1172 	}
1173 
1174 	if (error || args->agbno == NULLAGBLOCK)
1175 		return error;
1176 
1177 	ASSERT(args->len >= args->minlen);
1178 	ASSERT(args->len <= args->maxlen);
1179 	ASSERT(!args->wasfromfl || args->resv != XFS_AG_RESV_AGFL);
1180 	ASSERT(args->agbno % args->alignment == 0);
1181 
1182 	/* if not file data, insert new block into the reverse map btree */
1183 	if (!xfs_rmap_should_skip_owner_update(&args->oinfo)) {
1184 		error = xfs_rmap_alloc(args->tp, args->agbp, args->pag,
1185 				       args->agbno, args->len, &args->oinfo);
1186 		if (error)
1187 			return error;
1188 	}
1189 
1190 	if (!args->wasfromfl) {
1191 		error = xfs_alloc_update_counters(args->tp, args->agbp,
1192 						  -((long)(args->len)));
1193 		if (error)
1194 			return error;
1195 
1196 		ASSERT(!xfs_extent_busy_search(args->mp, args->pag,
1197 					      args->agbno, args->len));
1198 	}
1199 
1200 	xfs_ag_resv_alloc_extent(args->pag, args->resv, args);
1201 
1202 	XFS_STATS_INC(args->mp, xs_allocx);
1203 	XFS_STATS_ADD(args->mp, xs_allocb, args->len);
1204 	return error;
1205 }
1206 
1207 /*
1208  * Allocate a variable extent at exactly agno/bno.
1209  * Extent's length (returned in *len) will be between minlen and maxlen,
1210  * and of the form k * prod + mod unless there's nothing that large.
1211  * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
1212  */
1213 STATIC int			/* error */
1214 xfs_alloc_ag_vextent_exact(
1215 	xfs_alloc_arg_t	*args)	/* allocation argument structure */
1216 {
1217 	struct xfs_agf __maybe_unused *agf = args->agbp->b_addr;
1218 	struct xfs_btree_cur *bno_cur;/* by block-number btree cursor */
1219 	struct xfs_btree_cur *cnt_cur;/* by count btree cursor */
1220 	int		error;
1221 	xfs_agblock_t	fbno;	/* start block of found extent */
1222 	xfs_extlen_t	flen;	/* length of found extent */
1223 	xfs_agblock_t	tbno;	/* start block of busy extent */
1224 	xfs_extlen_t	tlen;	/* length of busy extent */
1225 	xfs_agblock_t	tend;	/* end block of busy extent */
1226 	int		i;	/* success/failure of operation */
1227 	unsigned	busy_gen;
1228 
1229 	ASSERT(args->alignment == 1);
1230 
1231 	/*
1232 	 * Allocate/initialize a cursor for the by-number freespace btree.
1233 	 */
1234 	bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1235 					  args->pag, XFS_BTNUM_BNO);
1236 
1237 	/*
1238 	 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
1239 	 * Look for the closest free block <= bno, it must contain bno
1240 	 * if any free block does.
1241 	 */
1242 	error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
1243 	if (error)
1244 		goto error0;
1245 	if (!i)
1246 		goto not_found;
1247 
1248 	/*
1249 	 * Grab the freespace record.
1250 	 */
1251 	error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
1252 	if (error)
1253 		goto error0;
1254 	if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1255 		error = -EFSCORRUPTED;
1256 		goto error0;
1257 	}
1258 	ASSERT(fbno <= args->agbno);
1259 
1260 	/*
1261 	 * Check for overlapping busy extents.
1262 	 */
1263 	tbno = fbno;
1264 	tlen = flen;
1265 	xfs_extent_busy_trim(args, &tbno, &tlen, &busy_gen);
1266 
1267 	/*
1268 	 * Give up if the start of the extent is busy, or the freespace isn't
1269 	 * long enough for the minimum request.
1270 	 */
1271 	if (tbno > args->agbno)
1272 		goto not_found;
1273 	if (tlen < args->minlen)
1274 		goto not_found;
1275 	tend = tbno + tlen;
1276 	if (tend < args->agbno + args->minlen)
1277 		goto not_found;
1278 
1279 	/*
1280 	 * End of extent will be smaller of the freespace end and the
1281 	 * maximal requested end.
1282 	 *
1283 	 * Fix the length according to mod and prod if given.
1284 	 */
1285 	args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
1286 						- args->agbno;
1287 	xfs_alloc_fix_len(args);
1288 	ASSERT(args->agbno + args->len <= tend);
1289 
1290 	/*
1291 	 * We are allocating agbno for args->len
1292 	 * Allocate/initialize a cursor for the by-size btree.
1293 	 */
1294 	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1295 					args->pag, XFS_BTNUM_CNT);
1296 	ASSERT(args->agbno + args->len <= be32_to_cpu(agf->agf_length));
1297 	error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
1298 				      args->len, XFSA_FIXUP_BNO_OK);
1299 	if (error) {
1300 		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1301 		goto error0;
1302 	}
1303 
1304 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1305 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1306 
1307 	args->wasfromfl = 0;
1308 	trace_xfs_alloc_exact_done(args);
1309 	return 0;
1310 
1311 not_found:
1312 	/* Didn't find it, return null. */
1313 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1314 	args->agbno = NULLAGBLOCK;
1315 	trace_xfs_alloc_exact_notfound(args);
1316 	return 0;
1317 
1318 error0:
1319 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1320 	trace_xfs_alloc_exact_error(args);
1321 	return error;
1322 }
1323 
1324 /*
1325  * Search a given number of btree records in a given direction. Check each
1326  * record against the good extent we've already found.
1327  */
1328 STATIC int
1329 xfs_alloc_walk_iter(
1330 	struct xfs_alloc_arg	*args,
1331 	struct xfs_alloc_cur	*acur,
1332 	struct xfs_btree_cur	*cur,
1333 	bool			increment,
1334 	bool			find_one, /* quit on first candidate */
1335 	int			count,    /* rec count (-1 for infinite) */
1336 	int			*stat)
1337 {
1338 	int			error;
1339 	int			i;
1340 
1341 	*stat = 0;
1342 
1343 	/*
1344 	 * Search so long as the cursor is active or we find a better extent.
1345 	 * The cursor is deactivated if it extends beyond the range of the
1346 	 * current allocation candidate.
1347 	 */
1348 	while (xfs_alloc_cur_active(cur) && count) {
1349 		error = xfs_alloc_cur_check(args, acur, cur, &i);
1350 		if (error)
1351 			return error;
1352 		if (i == 1) {
1353 			*stat = 1;
1354 			if (find_one)
1355 				break;
1356 		}
1357 		if (!xfs_alloc_cur_active(cur))
1358 			break;
1359 
1360 		if (increment)
1361 			error = xfs_btree_increment(cur, 0, &i);
1362 		else
1363 			error = xfs_btree_decrement(cur, 0, &i);
1364 		if (error)
1365 			return error;
1366 		if (i == 0)
1367 			cur->bc_ag.abt.active = false;
1368 
1369 		if (count > 0)
1370 			count--;
1371 	}
1372 
1373 	return 0;
1374 }
1375 
1376 /*
1377  * Search the by-bno and by-size btrees in parallel in search of an extent with
1378  * ideal locality based on the NEAR mode ->agbno locality hint.
1379  */
1380 STATIC int
1381 xfs_alloc_ag_vextent_locality(
1382 	struct xfs_alloc_arg	*args,
1383 	struct xfs_alloc_cur	*acur,
1384 	int			*stat)
1385 {
1386 	struct xfs_btree_cur	*fbcur = NULL;
1387 	int			error;
1388 	int			i;
1389 	bool			fbinc;
1390 
1391 	ASSERT(acur->len == 0);
1392 	ASSERT(args->type == XFS_ALLOCTYPE_NEAR_BNO);
1393 
1394 	*stat = 0;
1395 
1396 	error = xfs_alloc_lookup_ge(acur->cnt, args->agbno, acur->cur_len, &i);
1397 	if (error)
1398 		return error;
1399 	error = xfs_alloc_lookup_le(acur->bnolt, args->agbno, 0, &i);
1400 	if (error)
1401 		return error;
1402 	error = xfs_alloc_lookup_ge(acur->bnogt, args->agbno, 0, &i);
1403 	if (error)
1404 		return error;
1405 
1406 	/*
1407 	 * Search the bnobt and cntbt in parallel. Search the bnobt left and
1408 	 * right and lookup the closest extent to the locality hint for each
1409 	 * extent size key in the cntbt. The entire search terminates
1410 	 * immediately on a bnobt hit because that means we've found best case
1411 	 * locality. Otherwise the search continues until the cntbt cursor runs
1412 	 * off the end of the tree. If no allocation candidate is found at this
1413 	 * point, give up on locality, walk backwards from the end of the cntbt
1414 	 * and take the first available extent.
1415 	 *
1416 	 * The parallel tree searches balance each other out to provide fairly
1417 	 * consistent performance for various situations. The bnobt search can
1418 	 * have pathological behavior in the worst case scenario of larger
1419 	 * allocation requests and fragmented free space. On the other hand, the
1420 	 * bnobt is able to satisfy most smaller allocation requests much more
1421 	 * quickly than the cntbt. The cntbt search can sift through fragmented
1422 	 * free space and sets of free extents for larger allocation requests
1423 	 * more quickly than the bnobt. Since the locality hint is just a hint
1424 	 * and we don't want to scan the entire bnobt for perfect locality, the
1425 	 * cntbt search essentially bounds the bnobt search such that we can
1426 	 * find good enough locality at reasonable performance in most cases.
1427 	 */
1428 	while (xfs_alloc_cur_active(acur->bnolt) ||
1429 	       xfs_alloc_cur_active(acur->bnogt) ||
1430 	       xfs_alloc_cur_active(acur->cnt)) {
1431 
1432 		trace_xfs_alloc_cur_lookup(args);
1433 
1434 		/*
1435 		 * Search the bnobt left and right. In the case of a hit, finish
1436 		 * the search in the opposite direction and we're done.
1437 		 */
1438 		error = xfs_alloc_walk_iter(args, acur, acur->bnolt, false,
1439 					    true, 1, &i);
1440 		if (error)
1441 			return error;
1442 		if (i == 1) {
1443 			trace_xfs_alloc_cur_left(args);
1444 			fbcur = acur->bnogt;
1445 			fbinc = true;
1446 			break;
1447 		}
1448 		error = xfs_alloc_walk_iter(args, acur, acur->bnogt, true, true,
1449 					    1, &i);
1450 		if (error)
1451 			return error;
1452 		if (i == 1) {
1453 			trace_xfs_alloc_cur_right(args);
1454 			fbcur = acur->bnolt;
1455 			fbinc = false;
1456 			break;
1457 		}
1458 
1459 		/*
1460 		 * Check the extent with best locality based on the current
1461 		 * extent size search key and keep track of the best candidate.
1462 		 */
1463 		error = xfs_alloc_cntbt_iter(args, acur);
1464 		if (error)
1465 			return error;
1466 		if (!xfs_alloc_cur_active(acur->cnt)) {
1467 			trace_xfs_alloc_cur_lookup_done(args);
1468 			break;
1469 		}
1470 	}
1471 
1472 	/*
1473 	 * If we failed to find anything due to busy extents, return empty
1474 	 * handed so the caller can flush and retry. If no busy extents were
1475 	 * found, walk backwards from the end of the cntbt as a last resort.
1476 	 */
1477 	if (!xfs_alloc_cur_active(acur->cnt) && !acur->len && !acur->busy) {
1478 		error = xfs_btree_decrement(acur->cnt, 0, &i);
1479 		if (error)
1480 			return error;
1481 		if (i) {
1482 			acur->cnt->bc_ag.abt.active = true;
1483 			fbcur = acur->cnt;
1484 			fbinc = false;
1485 		}
1486 	}
1487 
1488 	/*
1489 	 * Search in the opposite direction for a better entry in the case of
1490 	 * a bnobt hit or walk backwards from the end of the cntbt.
1491 	 */
1492 	if (fbcur) {
1493 		error = xfs_alloc_walk_iter(args, acur, fbcur, fbinc, true, -1,
1494 					    &i);
1495 		if (error)
1496 			return error;
1497 	}
1498 
1499 	if (acur->len)
1500 		*stat = 1;
1501 
1502 	return 0;
1503 }
1504 
1505 /* Check the last block of the cnt btree for allocations. */
1506 static int
1507 xfs_alloc_ag_vextent_lastblock(
1508 	struct xfs_alloc_arg	*args,
1509 	struct xfs_alloc_cur	*acur,
1510 	xfs_agblock_t		*bno,
1511 	xfs_extlen_t		*len,
1512 	bool			*allocated)
1513 {
1514 	int			error;
1515 	int			i;
1516 
1517 #ifdef DEBUG
1518 	/* Randomly don't execute the first algorithm. */
1519 	if (prandom_u32_max(2))
1520 		return 0;
1521 #endif
1522 
1523 	/*
1524 	 * Start from the entry that lookup found, sequence through all larger
1525 	 * free blocks.  If we're actually pointing at a record smaller than
1526 	 * maxlen, go to the start of this block, and skip all those smaller
1527 	 * than minlen.
1528 	 */
1529 	if (*len || args->alignment > 1) {
1530 		acur->cnt->bc_levels[0].ptr = 1;
1531 		do {
1532 			error = xfs_alloc_get_rec(acur->cnt, bno, len, &i);
1533 			if (error)
1534 				return error;
1535 			if (XFS_IS_CORRUPT(args->mp, i != 1))
1536 				return -EFSCORRUPTED;
1537 			if (*len >= args->minlen)
1538 				break;
1539 			error = xfs_btree_increment(acur->cnt, 0, &i);
1540 			if (error)
1541 				return error;
1542 		} while (i);
1543 		ASSERT(*len >= args->minlen);
1544 		if (!i)
1545 			return 0;
1546 	}
1547 
1548 	error = xfs_alloc_walk_iter(args, acur, acur->cnt, true, false, -1, &i);
1549 	if (error)
1550 		return error;
1551 
1552 	/*
1553 	 * It didn't work.  We COULD be in a case where there's a good record
1554 	 * somewhere, so try again.
1555 	 */
1556 	if (acur->len == 0)
1557 		return 0;
1558 
1559 	trace_xfs_alloc_near_first(args);
1560 	*allocated = true;
1561 	return 0;
1562 }
1563 
1564 /*
1565  * Allocate a variable extent near bno in the allocation group agno.
1566  * Extent's length (returned in len) will be between minlen and maxlen,
1567  * and of the form k * prod + mod unless there's nothing that large.
1568  * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1569  */
1570 STATIC int
1571 xfs_alloc_ag_vextent_near(
1572 	struct xfs_alloc_arg	*args)
1573 {
1574 	struct xfs_alloc_cur	acur = {};
1575 	int			error;		/* error code */
1576 	int			i;		/* result code, temporary */
1577 	xfs_agblock_t		bno;
1578 	xfs_extlen_t		len;
1579 
1580 	/* handle uninitialized agbno range so caller doesn't have to */
1581 	if (!args->min_agbno && !args->max_agbno)
1582 		args->max_agbno = args->mp->m_sb.sb_agblocks - 1;
1583 	ASSERT(args->min_agbno <= args->max_agbno);
1584 
1585 	/* clamp agbno to the range if it's outside */
1586 	if (args->agbno < args->min_agbno)
1587 		args->agbno = args->min_agbno;
1588 	if (args->agbno > args->max_agbno)
1589 		args->agbno = args->max_agbno;
1590 
1591 restart:
1592 	len = 0;
1593 
1594 	/*
1595 	 * Set up cursors and see if there are any free extents as big as
1596 	 * maxlen. If not, pick the last entry in the tree unless the tree is
1597 	 * empty.
1598 	 */
1599 	error = xfs_alloc_cur_setup(args, &acur);
1600 	if (error == -ENOSPC) {
1601 		error = xfs_alloc_ag_vextent_small(args, acur.cnt, &bno,
1602 				&len, &i);
1603 		if (error)
1604 			goto out;
1605 		if (i == 0 || len == 0) {
1606 			trace_xfs_alloc_near_noentry(args);
1607 			goto out;
1608 		}
1609 		ASSERT(i == 1);
1610 	} else if (error) {
1611 		goto out;
1612 	}
1613 
1614 	/*
1615 	 * First algorithm.
1616 	 * If the requested extent is large wrt the freespaces available
1617 	 * in this a.g., then the cursor will be pointing to a btree entry
1618 	 * near the right edge of the tree.  If it's in the last btree leaf
1619 	 * block, then we just examine all the entries in that block
1620 	 * that are big enough, and pick the best one.
1621 	 */
1622 	if (xfs_btree_islastblock(acur.cnt, 0)) {
1623 		bool		allocated = false;
1624 
1625 		error = xfs_alloc_ag_vextent_lastblock(args, &acur, &bno, &len,
1626 				&allocated);
1627 		if (error)
1628 			goto out;
1629 		if (allocated)
1630 			goto alloc_finish;
1631 	}
1632 
1633 	/*
1634 	 * Second algorithm. Combined cntbt and bnobt search to find ideal
1635 	 * locality.
1636 	 */
1637 	error = xfs_alloc_ag_vextent_locality(args, &acur, &i);
1638 	if (error)
1639 		goto out;
1640 
1641 	/*
1642 	 * If we couldn't get anything, give up.
1643 	 */
1644 	if (!acur.len) {
1645 		if (acur.busy) {
1646 			trace_xfs_alloc_near_busy(args);
1647 			xfs_extent_busy_flush(args->mp, args->pag,
1648 					      acur.busy_gen);
1649 			goto restart;
1650 		}
1651 		trace_xfs_alloc_size_neither(args);
1652 		args->agbno = NULLAGBLOCK;
1653 		goto out;
1654 	}
1655 
1656 alloc_finish:
1657 	/* fix up btrees on a successful allocation */
1658 	error = xfs_alloc_cur_finish(args, &acur);
1659 
1660 out:
1661 	xfs_alloc_cur_close(&acur, error);
1662 	return error;
1663 }
1664 
1665 /*
1666  * Allocate a variable extent anywhere in the allocation group agno.
1667  * Extent's length (returned in len) will be between minlen and maxlen,
1668  * and of the form k * prod + mod unless there's nothing that large.
1669  * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1670  */
1671 STATIC int				/* error */
1672 xfs_alloc_ag_vextent_size(
1673 	xfs_alloc_arg_t	*args)		/* allocation argument structure */
1674 {
1675 	struct xfs_agf	*agf = args->agbp->b_addr;
1676 	struct xfs_btree_cur *bno_cur;	/* cursor for bno btree */
1677 	struct xfs_btree_cur *cnt_cur;	/* cursor for cnt btree */
1678 	int		error;		/* error result */
1679 	xfs_agblock_t	fbno;		/* start of found freespace */
1680 	xfs_extlen_t	flen;		/* length of found freespace */
1681 	int		i;		/* temp status variable */
1682 	xfs_agblock_t	rbno;		/* returned block number */
1683 	xfs_extlen_t	rlen;		/* length of returned extent */
1684 	bool		busy;
1685 	unsigned	busy_gen;
1686 
1687 restart:
1688 	/*
1689 	 * Allocate and initialize a cursor for the by-size btree.
1690 	 */
1691 	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1692 					args->pag, XFS_BTNUM_CNT);
1693 	bno_cur = NULL;
1694 
1695 	/*
1696 	 * Look for an entry >= maxlen+alignment-1 blocks.
1697 	 */
1698 	if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1699 			args->maxlen + args->alignment - 1, &i)))
1700 		goto error0;
1701 
1702 	/*
1703 	 * If none then we have to settle for a smaller extent. In the case that
1704 	 * there are no large extents, this will return the last entry in the
1705 	 * tree unless the tree is empty. In the case that there are only busy
1706 	 * large extents, this will return the largest small extent unless there
1707 	 * are no smaller extents available.
1708 	 */
1709 	if (!i) {
1710 		error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1711 						   &fbno, &flen, &i);
1712 		if (error)
1713 			goto error0;
1714 		if (i == 0 || flen == 0) {
1715 			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1716 			trace_xfs_alloc_size_noentry(args);
1717 			return 0;
1718 		}
1719 		ASSERT(i == 1);
1720 		busy = xfs_alloc_compute_aligned(args, fbno, flen, &rbno,
1721 				&rlen, &busy_gen);
1722 	} else {
1723 		/*
1724 		 * Search for a non-busy extent that is large enough.
1725 		 */
1726 		for (;;) {
1727 			error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1728 			if (error)
1729 				goto error0;
1730 			if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1731 				error = -EFSCORRUPTED;
1732 				goto error0;
1733 			}
1734 
1735 			busy = xfs_alloc_compute_aligned(args, fbno, flen,
1736 					&rbno, &rlen, &busy_gen);
1737 
1738 			if (rlen >= args->maxlen)
1739 				break;
1740 
1741 			error = xfs_btree_increment(cnt_cur, 0, &i);
1742 			if (error)
1743 				goto error0;
1744 			if (i == 0) {
1745 				/*
1746 				 * Our only valid extents must have been busy.
1747 				 * Make it unbusy by forcing the log out and
1748 				 * retrying.
1749 				 */
1750 				xfs_btree_del_cursor(cnt_cur,
1751 						     XFS_BTREE_NOERROR);
1752 				trace_xfs_alloc_size_busy(args);
1753 				xfs_extent_busy_flush(args->mp,
1754 							args->pag, busy_gen);
1755 				goto restart;
1756 			}
1757 		}
1758 	}
1759 
1760 	/*
1761 	 * In the first case above, we got the last entry in the
1762 	 * by-size btree.  Now we check to see if the space hits maxlen
1763 	 * once aligned; if not, we search left for something better.
1764 	 * This can't happen in the second case above.
1765 	 */
1766 	rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1767 	if (XFS_IS_CORRUPT(args->mp,
1768 			   rlen != 0 &&
1769 			   (rlen > flen ||
1770 			    rbno + rlen > fbno + flen))) {
1771 		error = -EFSCORRUPTED;
1772 		goto error0;
1773 	}
1774 	if (rlen < args->maxlen) {
1775 		xfs_agblock_t	bestfbno;
1776 		xfs_extlen_t	bestflen;
1777 		xfs_agblock_t	bestrbno;
1778 		xfs_extlen_t	bestrlen;
1779 
1780 		bestrlen = rlen;
1781 		bestrbno = rbno;
1782 		bestflen = flen;
1783 		bestfbno = fbno;
1784 		for (;;) {
1785 			if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1786 				goto error0;
1787 			if (i == 0)
1788 				break;
1789 			if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1790 					&i)))
1791 				goto error0;
1792 			if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1793 				error = -EFSCORRUPTED;
1794 				goto error0;
1795 			}
1796 			if (flen < bestrlen)
1797 				break;
1798 			busy = xfs_alloc_compute_aligned(args, fbno, flen,
1799 					&rbno, &rlen, &busy_gen);
1800 			rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1801 			if (XFS_IS_CORRUPT(args->mp,
1802 					   rlen != 0 &&
1803 					   (rlen > flen ||
1804 					    rbno + rlen > fbno + flen))) {
1805 				error = -EFSCORRUPTED;
1806 				goto error0;
1807 			}
1808 			if (rlen > bestrlen) {
1809 				bestrlen = rlen;
1810 				bestrbno = rbno;
1811 				bestflen = flen;
1812 				bestfbno = fbno;
1813 				if (rlen == args->maxlen)
1814 					break;
1815 			}
1816 		}
1817 		if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1818 				&i)))
1819 			goto error0;
1820 		if (XFS_IS_CORRUPT(args->mp, i != 1)) {
1821 			error = -EFSCORRUPTED;
1822 			goto error0;
1823 		}
1824 		rlen = bestrlen;
1825 		rbno = bestrbno;
1826 		flen = bestflen;
1827 		fbno = bestfbno;
1828 	}
1829 	args->wasfromfl = 0;
1830 	/*
1831 	 * Fix up the length.
1832 	 */
1833 	args->len = rlen;
1834 	if (rlen < args->minlen) {
1835 		if (busy) {
1836 			xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1837 			trace_xfs_alloc_size_busy(args);
1838 			xfs_extent_busy_flush(args->mp, args->pag, busy_gen);
1839 			goto restart;
1840 		}
1841 		goto out_nominleft;
1842 	}
1843 	xfs_alloc_fix_len(args);
1844 
1845 	rlen = args->len;
1846 	if (XFS_IS_CORRUPT(args->mp, rlen > flen)) {
1847 		error = -EFSCORRUPTED;
1848 		goto error0;
1849 	}
1850 	/*
1851 	 * Allocate and initialize a cursor for the by-block tree.
1852 	 */
1853 	bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1854 					args->pag, XFS_BTNUM_BNO);
1855 	if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1856 			rbno, rlen, XFSA_FIXUP_CNT_OK)))
1857 		goto error0;
1858 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1859 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1860 	cnt_cur = bno_cur = NULL;
1861 	args->len = rlen;
1862 	args->agbno = rbno;
1863 	if (XFS_IS_CORRUPT(args->mp,
1864 			   args->agbno + args->len >
1865 			   be32_to_cpu(agf->agf_length))) {
1866 		error = -EFSCORRUPTED;
1867 		goto error0;
1868 	}
1869 	trace_xfs_alloc_size_done(args);
1870 	return 0;
1871 
1872 error0:
1873 	trace_xfs_alloc_size_error(args);
1874 	if (cnt_cur)
1875 		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1876 	if (bno_cur)
1877 		xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1878 	return error;
1879 
1880 out_nominleft:
1881 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1882 	trace_xfs_alloc_size_nominleft(args);
1883 	args->agbno = NULLAGBLOCK;
1884 	return 0;
1885 }
1886 
1887 /*
1888  * Free the extent starting at agno/bno for length.
1889  */
1890 STATIC int
1891 xfs_free_ag_extent(
1892 	struct xfs_trans		*tp,
1893 	struct xfs_buf			*agbp,
1894 	xfs_agnumber_t			agno,
1895 	xfs_agblock_t			bno,
1896 	xfs_extlen_t			len,
1897 	const struct xfs_owner_info	*oinfo,
1898 	enum xfs_ag_resv_type		type)
1899 {
1900 	struct xfs_mount		*mp;
1901 	struct xfs_btree_cur		*bno_cur;
1902 	struct xfs_btree_cur		*cnt_cur;
1903 	xfs_agblock_t			gtbno; /* start of right neighbor */
1904 	xfs_extlen_t			gtlen; /* length of right neighbor */
1905 	xfs_agblock_t			ltbno; /* start of left neighbor */
1906 	xfs_extlen_t			ltlen; /* length of left neighbor */
1907 	xfs_agblock_t			nbno; /* new starting block of freesp */
1908 	xfs_extlen_t			nlen; /* new length of freespace */
1909 	int				haveleft; /* have a left neighbor */
1910 	int				haveright; /* have a right neighbor */
1911 	int				i;
1912 	int				error;
1913 	struct xfs_perag		*pag = agbp->b_pag;
1914 
1915 	bno_cur = cnt_cur = NULL;
1916 	mp = tp->t_mountp;
1917 
1918 	if (!xfs_rmap_should_skip_owner_update(oinfo)) {
1919 		error = xfs_rmap_free(tp, agbp, pag, bno, len, oinfo);
1920 		if (error)
1921 			goto error0;
1922 	}
1923 
1924 	/*
1925 	 * Allocate and initialize a cursor for the by-block btree.
1926 	 */
1927 	bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_BNO);
1928 	/*
1929 	 * Look for a neighboring block on the left (lower block numbers)
1930 	 * that is contiguous with this space.
1931 	 */
1932 	if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1933 		goto error0;
1934 	if (haveleft) {
1935 		/*
1936 		 * There is a block to our left.
1937 		 */
1938 		if ((error = xfs_alloc_get_rec(bno_cur, &ltbno, &ltlen, &i)))
1939 			goto error0;
1940 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1941 			error = -EFSCORRUPTED;
1942 			goto error0;
1943 		}
1944 		/*
1945 		 * It's not contiguous, though.
1946 		 */
1947 		if (ltbno + ltlen < bno)
1948 			haveleft = 0;
1949 		else {
1950 			/*
1951 			 * If this failure happens the request to free this
1952 			 * space was invalid, it's (partly) already free.
1953 			 * Very bad.
1954 			 */
1955 			if (XFS_IS_CORRUPT(mp, ltbno + ltlen > bno)) {
1956 				error = -EFSCORRUPTED;
1957 				goto error0;
1958 			}
1959 		}
1960 	}
1961 	/*
1962 	 * Look for a neighboring block on the right (higher block numbers)
1963 	 * that is contiguous with this space.
1964 	 */
1965 	if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1966 		goto error0;
1967 	if (haveright) {
1968 		/*
1969 		 * There is a block to our right.
1970 		 */
1971 		if ((error = xfs_alloc_get_rec(bno_cur, &gtbno, &gtlen, &i)))
1972 			goto error0;
1973 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1974 			error = -EFSCORRUPTED;
1975 			goto error0;
1976 		}
1977 		/*
1978 		 * It's not contiguous, though.
1979 		 */
1980 		if (bno + len < gtbno)
1981 			haveright = 0;
1982 		else {
1983 			/*
1984 			 * If this failure happens the request to free this
1985 			 * space was invalid, it's (partly) already free.
1986 			 * Very bad.
1987 			 */
1988 			if (XFS_IS_CORRUPT(mp, bno + len > gtbno)) {
1989 				error = -EFSCORRUPTED;
1990 				goto error0;
1991 			}
1992 		}
1993 	}
1994 	/*
1995 	 * Now allocate and initialize a cursor for the by-size tree.
1996 	 */
1997 	cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_CNT);
1998 	/*
1999 	 * Have both left and right contiguous neighbors.
2000 	 * Merge all three into a single free block.
2001 	 */
2002 	if (haveleft && haveright) {
2003 		/*
2004 		 * Delete the old by-size entry on the left.
2005 		 */
2006 		if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
2007 			goto error0;
2008 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2009 			error = -EFSCORRUPTED;
2010 			goto error0;
2011 		}
2012 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2013 			goto error0;
2014 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2015 			error = -EFSCORRUPTED;
2016 			goto error0;
2017 		}
2018 		/*
2019 		 * Delete the old by-size entry on the right.
2020 		 */
2021 		if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
2022 			goto error0;
2023 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2024 			error = -EFSCORRUPTED;
2025 			goto error0;
2026 		}
2027 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2028 			goto error0;
2029 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2030 			error = -EFSCORRUPTED;
2031 			goto error0;
2032 		}
2033 		/*
2034 		 * Delete the old by-block entry for the right block.
2035 		 */
2036 		if ((error = xfs_btree_delete(bno_cur, &i)))
2037 			goto error0;
2038 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2039 			error = -EFSCORRUPTED;
2040 			goto error0;
2041 		}
2042 		/*
2043 		 * Move the by-block cursor back to the left neighbor.
2044 		 */
2045 		if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
2046 			goto error0;
2047 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2048 			error = -EFSCORRUPTED;
2049 			goto error0;
2050 		}
2051 #ifdef DEBUG
2052 		/*
2053 		 * Check that this is the right record: delete didn't
2054 		 * mangle the cursor.
2055 		 */
2056 		{
2057 			xfs_agblock_t	xxbno;
2058 			xfs_extlen_t	xxlen;
2059 
2060 			if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
2061 					&i)))
2062 				goto error0;
2063 			if (XFS_IS_CORRUPT(mp,
2064 					   i != 1 ||
2065 					   xxbno != ltbno ||
2066 					   xxlen != ltlen)) {
2067 				error = -EFSCORRUPTED;
2068 				goto error0;
2069 			}
2070 		}
2071 #endif
2072 		/*
2073 		 * Update remaining by-block entry to the new, joined block.
2074 		 */
2075 		nbno = ltbno;
2076 		nlen = len + ltlen + gtlen;
2077 		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2078 			goto error0;
2079 	}
2080 	/*
2081 	 * Have only a left contiguous neighbor.
2082 	 * Merge it together with the new freespace.
2083 	 */
2084 	else if (haveleft) {
2085 		/*
2086 		 * Delete the old by-size entry on the left.
2087 		 */
2088 		if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
2089 			goto error0;
2090 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2091 			error = -EFSCORRUPTED;
2092 			goto error0;
2093 		}
2094 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2095 			goto error0;
2096 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2097 			error = -EFSCORRUPTED;
2098 			goto error0;
2099 		}
2100 		/*
2101 		 * Back up the by-block cursor to the left neighbor, and
2102 		 * update its length.
2103 		 */
2104 		if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
2105 			goto error0;
2106 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2107 			error = -EFSCORRUPTED;
2108 			goto error0;
2109 		}
2110 		nbno = ltbno;
2111 		nlen = len + ltlen;
2112 		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2113 			goto error0;
2114 	}
2115 	/*
2116 	 * Have only a right contiguous neighbor.
2117 	 * Merge it together with the new freespace.
2118 	 */
2119 	else if (haveright) {
2120 		/*
2121 		 * Delete the old by-size entry on the right.
2122 		 */
2123 		if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
2124 			goto error0;
2125 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2126 			error = -EFSCORRUPTED;
2127 			goto error0;
2128 		}
2129 		if ((error = xfs_btree_delete(cnt_cur, &i)))
2130 			goto error0;
2131 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2132 			error = -EFSCORRUPTED;
2133 			goto error0;
2134 		}
2135 		/*
2136 		 * Update the starting block and length of the right
2137 		 * neighbor in the by-block tree.
2138 		 */
2139 		nbno = bno;
2140 		nlen = len + gtlen;
2141 		if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
2142 			goto error0;
2143 	}
2144 	/*
2145 	 * No contiguous neighbors.
2146 	 * Insert the new freespace into the by-block tree.
2147 	 */
2148 	else {
2149 		nbno = bno;
2150 		nlen = len;
2151 		if ((error = xfs_btree_insert(bno_cur, &i)))
2152 			goto error0;
2153 		if (XFS_IS_CORRUPT(mp, i != 1)) {
2154 			error = -EFSCORRUPTED;
2155 			goto error0;
2156 		}
2157 	}
2158 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
2159 	bno_cur = NULL;
2160 	/*
2161 	 * In all cases we need to insert the new freespace in the by-size tree.
2162 	 */
2163 	if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
2164 		goto error0;
2165 	if (XFS_IS_CORRUPT(mp, i != 0)) {
2166 		error = -EFSCORRUPTED;
2167 		goto error0;
2168 	}
2169 	if ((error = xfs_btree_insert(cnt_cur, &i)))
2170 		goto error0;
2171 	if (XFS_IS_CORRUPT(mp, i != 1)) {
2172 		error = -EFSCORRUPTED;
2173 		goto error0;
2174 	}
2175 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
2176 	cnt_cur = NULL;
2177 
2178 	/*
2179 	 * Update the freespace totals in the ag and superblock.
2180 	 */
2181 	error = xfs_alloc_update_counters(tp, agbp, len);
2182 	xfs_ag_resv_free_extent(agbp->b_pag, type, tp, len);
2183 	if (error)
2184 		goto error0;
2185 
2186 	XFS_STATS_INC(mp, xs_freex);
2187 	XFS_STATS_ADD(mp, xs_freeb, len);
2188 
2189 	trace_xfs_free_extent(mp, agno, bno, len, type, haveleft, haveright);
2190 
2191 	return 0;
2192 
2193  error0:
2194 	trace_xfs_free_extent(mp, agno, bno, len, type, -1, -1);
2195 	if (bno_cur)
2196 		xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
2197 	if (cnt_cur)
2198 		xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
2199 	return error;
2200 }
2201 
2202 /*
2203  * Visible (exported) allocation/free functions.
2204  * Some of these are used just by xfs_alloc_btree.c and this file.
2205  */
2206 
2207 /*
2208  * Compute and fill in value of m_alloc_maxlevels.
2209  */
2210 void
2211 xfs_alloc_compute_maxlevels(
2212 	xfs_mount_t	*mp)	/* file system mount structure */
2213 {
2214 	mp->m_alloc_maxlevels = xfs_btree_compute_maxlevels(mp->m_alloc_mnr,
2215 			(mp->m_sb.sb_agblocks + 1) / 2);
2216 	ASSERT(mp->m_alloc_maxlevels <= xfs_allocbt_maxlevels_ondisk());
2217 }
2218 
2219 /*
2220  * Find the length of the longest extent in an AG.  The 'need' parameter
2221  * specifies how much space we're going to need for the AGFL and the
2222  * 'reserved' parameter tells us how many blocks in this AG are reserved for
2223  * other callers.
2224  */
2225 xfs_extlen_t
2226 xfs_alloc_longest_free_extent(
2227 	struct xfs_perag	*pag,
2228 	xfs_extlen_t		need,
2229 	xfs_extlen_t		reserved)
2230 {
2231 	xfs_extlen_t		delta = 0;
2232 
2233 	/*
2234 	 * If the AGFL needs a recharge, we'll have to subtract that from the
2235 	 * longest extent.
2236 	 */
2237 	if (need > pag->pagf_flcount)
2238 		delta = need - pag->pagf_flcount;
2239 
2240 	/*
2241 	 * If we cannot maintain others' reservations with space from the
2242 	 * not-longest freesp extents, we'll have to subtract /that/ from
2243 	 * the longest extent too.
2244 	 */
2245 	if (pag->pagf_freeblks - pag->pagf_longest < reserved)
2246 		delta += reserved - (pag->pagf_freeblks - pag->pagf_longest);
2247 
2248 	/*
2249 	 * If the longest extent is long enough to satisfy all the
2250 	 * reservations and AGFL rules in place, we can return this extent.
2251 	 */
2252 	if (pag->pagf_longest > delta)
2253 		return min_t(xfs_extlen_t, pag->pag_mount->m_ag_max_usable,
2254 				pag->pagf_longest - delta);
2255 
2256 	/* Otherwise, let the caller try for 1 block if there's space. */
2257 	return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
2258 }
2259 
2260 /*
2261  * Compute the minimum length of the AGFL in the given AG.  If @pag is NULL,
2262  * return the largest possible minimum length.
2263  */
2264 unsigned int
2265 xfs_alloc_min_freelist(
2266 	struct xfs_mount	*mp,
2267 	struct xfs_perag	*pag)
2268 {
2269 	/* AG btrees have at least 1 level. */
2270 	static const uint8_t	fake_levels[XFS_BTNUM_AGF] = {1, 1, 1};
2271 	const uint8_t		*levels = pag ? pag->pagf_levels : fake_levels;
2272 	unsigned int		min_free;
2273 
2274 	ASSERT(mp->m_alloc_maxlevels > 0);
2275 
2276 	/* space needed by-bno freespace btree */
2277 	min_free = min_t(unsigned int, levels[XFS_BTNUM_BNOi] + 1,
2278 				       mp->m_alloc_maxlevels);
2279 	/* space needed by-size freespace btree */
2280 	min_free += min_t(unsigned int, levels[XFS_BTNUM_CNTi] + 1,
2281 				       mp->m_alloc_maxlevels);
2282 	/* space needed reverse mapping used space btree */
2283 	if (xfs_has_rmapbt(mp))
2284 		min_free += min_t(unsigned int, levels[XFS_BTNUM_RMAPi] + 1,
2285 						mp->m_rmap_maxlevels);
2286 
2287 	return min_free;
2288 }
2289 
2290 /*
2291  * Check if the operation we are fixing up the freelist for should go ahead or
2292  * not. If we are freeing blocks, we always allow it, otherwise the allocation
2293  * is dependent on whether the size and shape of free space available will
2294  * permit the requested allocation to take place.
2295  */
2296 static bool
2297 xfs_alloc_space_available(
2298 	struct xfs_alloc_arg	*args,
2299 	xfs_extlen_t		min_free,
2300 	int			flags)
2301 {
2302 	struct xfs_perag	*pag = args->pag;
2303 	xfs_extlen_t		alloc_len, longest;
2304 	xfs_extlen_t		reservation; /* blocks that are still reserved */
2305 	int			available;
2306 	xfs_extlen_t		agflcount;
2307 
2308 	if (flags & XFS_ALLOC_FLAG_FREEING)
2309 		return true;
2310 
2311 	reservation = xfs_ag_resv_needed(pag, args->resv);
2312 
2313 	/* do we have enough contiguous free space for the allocation? */
2314 	alloc_len = args->minlen + (args->alignment - 1) + args->minalignslop;
2315 	longest = xfs_alloc_longest_free_extent(pag, min_free, reservation);
2316 	if (longest < alloc_len)
2317 		return false;
2318 
2319 	/*
2320 	 * Do we have enough free space remaining for the allocation? Don't
2321 	 * account extra agfl blocks because we are about to defer free them,
2322 	 * making them unavailable until the current transaction commits.
2323 	 */
2324 	agflcount = min_t(xfs_extlen_t, pag->pagf_flcount, min_free);
2325 	available = (int)(pag->pagf_freeblks + agflcount -
2326 			  reservation - min_free - args->minleft);
2327 	if (available < (int)max(args->total, alloc_len))
2328 		return false;
2329 
2330 	/*
2331 	 * Clamp maxlen to the amount of free space available for the actual
2332 	 * extent allocation.
2333 	 */
2334 	if (available < (int)args->maxlen && !(flags & XFS_ALLOC_FLAG_CHECK)) {
2335 		args->maxlen = available;
2336 		ASSERT(args->maxlen > 0);
2337 		ASSERT(args->maxlen >= args->minlen);
2338 	}
2339 
2340 	return true;
2341 }
2342 
2343 int
2344 xfs_free_agfl_block(
2345 	struct xfs_trans	*tp,
2346 	xfs_agnumber_t		agno,
2347 	xfs_agblock_t		agbno,
2348 	struct xfs_buf		*agbp,
2349 	struct xfs_owner_info	*oinfo)
2350 {
2351 	int			error;
2352 	struct xfs_buf		*bp;
2353 
2354 	error = xfs_free_ag_extent(tp, agbp, agno, agbno, 1, oinfo,
2355 				   XFS_AG_RESV_AGFL);
2356 	if (error)
2357 		return error;
2358 
2359 	error = xfs_trans_get_buf(tp, tp->t_mountp->m_ddev_targp,
2360 			XFS_AGB_TO_DADDR(tp->t_mountp, agno, agbno),
2361 			tp->t_mountp->m_bsize, 0, &bp);
2362 	if (error)
2363 		return error;
2364 	xfs_trans_binval(tp, bp);
2365 
2366 	return 0;
2367 }
2368 
2369 /*
2370  * Check the agfl fields of the agf for inconsistency or corruption. The purpose
2371  * is to detect an agfl header padding mismatch between current and early v5
2372  * kernels. This problem manifests as a 1-slot size difference between the
2373  * on-disk flcount and the active [first, last] range of a wrapped agfl. This
2374  * may also catch variants of agfl count corruption unrelated to padding. Either
2375  * way, we'll reset the agfl and warn the user.
2376  *
2377  * Return true if a reset is required before the agfl can be used, false
2378  * otherwise.
2379  */
2380 static bool
2381 xfs_agfl_needs_reset(
2382 	struct xfs_mount	*mp,
2383 	struct xfs_agf		*agf)
2384 {
2385 	uint32_t		f = be32_to_cpu(agf->agf_flfirst);
2386 	uint32_t		l = be32_to_cpu(agf->agf_fllast);
2387 	uint32_t		c = be32_to_cpu(agf->agf_flcount);
2388 	int			agfl_size = xfs_agfl_size(mp);
2389 	int			active;
2390 
2391 	/* no agfl header on v4 supers */
2392 	if (!xfs_has_crc(mp))
2393 		return false;
2394 
2395 	/*
2396 	 * The agf read verifier catches severe corruption of these fields.
2397 	 * Repeat some sanity checks to cover a packed -> unpacked mismatch if
2398 	 * the verifier allows it.
2399 	 */
2400 	if (f >= agfl_size || l >= agfl_size)
2401 		return true;
2402 	if (c > agfl_size)
2403 		return true;
2404 
2405 	/*
2406 	 * Check consistency between the on-disk count and the active range. An
2407 	 * agfl padding mismatch manifests as an inconsistent flcount.
2408 	 */
2409 	if (c && l >= f)
2410 		active = l - f + 1;
2411 	else if (c)
2412 		active = agfl_size - f + l + 1;
2413 	else
2414 		active = 0;
2415 
2416 	return active != c;
2417 }
2418 
2419 /*
2420  * Reset the agfl to an empty state. Ignore/drop any existing blocks since the
2421  * agfl content cannot be trusted. Warn the user that a repair is required to
2422  * recover leaked blocks.
2423  *
2424  * The purpose of this mechanism is to handle filesystems affected by the agfl
2425  * header padding mismatch problem. A reset keeps the filesystem online with a
2426  * relatively minor free space accounting inconsistency rather than suffer the
2427  * inevitable crash from use of an invalid agfl block.
2428  */
2429 static void
2430 xfs_agfl_reset(
2431 	struct xfs_trans	*tp,
2432 	struct xfs_buf		*agbp,
2433 	struct xfs_perag	*pag)
2434 {
2435 	struct xfs_mount	*mp = tp->t_mountp;
2436 	struct xfs_agf		*agf = agbp->b_addr;
2437 
2438 	ASSERT(pag->pagf_agflreset);
2439 	trace_xfs_agfl_reset(mp, agf, 0, _RET_IP_);
2440 
2441 	xfs_warn(mp,
2442 	       "WARNING: Reset corrupted AGFL on AG %u. %d blocks leaked. "
2443 	       "Please unmount and run xfs_repair.",
2444 	         pag->pag_agno, pag->pagf_flcount);
2445 
2446 	agf->agf_flfirst = 0;
2447 	agf->agf_fllast = cpu_to_be32(xfs_agfl_size(mp) - 1);
2448 	agf->agf_flcount = 0;
2449 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLLAST |
2450 				    XFS_AGF_FLCOUNT);
2451 
2452 	pag->pagf_flcount = 0;
2453 	pag->pagf_agflreset = false;
2454 }
2455 
2456 /*
2457  * Defer an AGFL block free. This is effectively equivalent to
2458  * xfs_free_extent_later() with some special handling particular to AGFL blocks.
2459  *
2460  * Deferring AGFL frees helps prevent log reservation overruns due to too many
2461  * allocation operations in a transaction. AGFL frees are prone to this problem
2462  * because for one they are always freed one at a time. Further, an immediate
2463  * AGFL block free can cause a btree join and require another block free before
2464  * the real allocation can proceed. Deferring the free disconnects freeing up
2465  * the AGFL slot from freeing the block.
2466  */
2467 STATIC void
2468 xfs_defer_agfl_block(
2469 	struct xfs_trans		*tp,
2470 	xfs_agnumber_t			agno,
2471 	xfs_fsblock_t			agbno,
2472 	struct xfs_owner_info		*oinfo)
2473 {
2474 	struct xfs_mount		*mp = tp->t_mountp;
2475 	struct xfs_extent_free_item	*new;		/* new element */
2476 
2477 	ASSERT(xfs_extfree_item_cache != NULL);
2478 	ASSERT(oinfo != NULL);
2479 
2480 	new = kmem_cache_zalloc(xfs_extfree_item_cache,
2481 			       GFP_KERNEL | __GFP_NOFAIL);
2482 	new->xefi_startblock = XFS_AGB_TO_FSB(mp, agno, agbno);
2483 	new->xefi_blockcount = 1;
2484 	new->xefi_owner = oinfo->oi_owner;
2485 
2486 	trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1);
2487 
2488 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_AGFL_FREE, &new->xefi_list);
2489 }
2490 
2491 /*
2492  * Add the extent to the list of extents to be free at transaction end.
2493  * The list is maintained sorted (by block number).
2494  */
2495 void
2496 __xfs_free_extent_later(
2497 	struct xfs_trans		*tp,
2498 	xfs_fsblock_t			bno,
2499 	xfs_filblks_t			len,
2500 	const struct xfs_owner_info	*oinfo,
2501 	bool				skip_discard)
2502 {
2503 	struct xfs_extent_free_item	*new;		/* new element */
2504 #ifdef DEBUG
2505 	struct xfs_mount		*mp = tp->t_mountp;
2506 	xfs_agnumber_t			agno;
2507 	xfs_agblock_t			agbno;
2508 
2509 	ASSERT(bno != NULLFSBLOCK);
2510 	ASSERT(len > 0);
2511 	ASSERT(len <= XFS_MAX_BMBT_EXTLEN);
2512 	ASSERT(!isnullstartblock(bno));
2513 	agno = XFS_FSB_TO_AGNO(mp, bno);
2514 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
2515 	ASSERT(agno < mp->m_sb.sb_agcount);
2516 	ASSERT(agbno < mp->m_sb.sb_agblocks);
2517 	ASSERT(len < mp->m_sb.sb_agblocks);
2518 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
2519 #endif
2520 	ASSERT(xfs_extfree_item_cache != NULL);
2521 
2522 	new = kmem_cache_zalloc(xfs_extfree_item_cache,
2523 			       GFP_KERNEL | __GFP_NOFAIL);
2524 	new->xefi_startblock = bno;
2525 	new->xefi_blockcount = (xfs_extlen_t)len;
2526 	if (skip_discard)
2527 		new->xefi_flags |= XFS_EFI_SKIP_DISCARD;
2528 	if (oinfo) {
2529 		ASSERT(oinfo->oi_offset == 0);
2530 
2531 		if (oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK)
2532 			new->xefi_flags |= XFS_EFI_ATTR_FORK;
2533 		if (oinfo->oi_flags & XFS_OWNER_INFO_BMBT_BLOCK)
2534 			new->xefi_flags |= XFS_EFI_BMBT_BLOCK;
2535 		new->xefi_owner = oinfo->oi_owner;
2536 	} else {
2537 		new->xefi_owner = XFS_RMAP_OWN_NULL;
2538 	}
2539 	trace_xfs_bmap_free_defer(tp->t_mountp,
2540 			XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
2541 			XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
2542 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
2543 }
2544 
2545 #ifdef DEBUG
2546 /*
2547  * Check if an AGF has a free extent record whose length is equal to
2548  * args->minlen.
2549  */
2550 STATIC int
2551 xfs_exact_minlen_extent_available(
2552 	struct xfs_alloc_arg	*args,
2553 	struct xfs_buf		*agbp,
2554 	int			*stat)
2555 {
2556 	struct xfs_btree_cur	*cnt_cur;
2557 	xfs_agblock_t		fbno;
2558 	xfs_extlen_t		flen;
2559 	int			error = 0;
2560 
2561 	cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, agbp,
2562 					args->pag, XFS_BTNUM_CNT);
2563 	error = xfs_alloc_lookup_ge(cnt_cur, 0, args->minlen, stat);
2564 	if (error)
2565 		goto out;
2566 
2567 	if (*stat == 0) {
2568 		error = -EFSCORRUPTED;
2569 		goto out;
2570 	}
2571 
2572 	error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, stat);
2573 	if (error)
2574 		goto out;
2575 
2576 	if (*stat == 1 && flen != args->minlen)
2577 		*stat = 0;
2578 
2579 out:
2580 	xfs_btree_del_cursor(cnt_cur, error);
2581 
2582 	return error;
2583 }
2584 #endif
2585 
2586 /*
2587  * Decide whether to use this allocation group for this allocation.
2588  * If so, fix up the btree freelist's size.
2589  */
2590 int			/* error */
2591 xfs_alloc_fix_freelist(
2592 	struct xfs_alloc_arg	*args,	/* allocation argument structure */
2593 	int			flags)	/* XFS_ALLOC_FLAG_... */
2594 {
2595 	struct xfs_mount	*mp = args->mp;
2596 	struct xfs_perag	*pag = args->pag;
2597 	struct xfs_trans	*tp = args->tp;
2598 	struct xfs_buf		*agbp = NULL;
2599 	struct xfs_buf		*agflbp = NULL;
2600 	struct xfs_alloc_arg	targs;	/* local allocation arguments */
2601 	xfs_agblock_t		bno;	/* freelist block */
2602 	xfs_extlen_t		need;	/* total blocks needed in freelist */
2603 	int			error = 0;
2604 
2605 	/* deferred ops (AGFL block frees) require permanent transactions */
2606 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
2607 
2608 	if (!pag->pagf_init) {
2609 		error = xfs_alloc_read_agf(pag, tp, flags, &agbp);
2610 		if (error) {
2611 			/* Couldn't lock the AGF so skip this AG. */
2612 			if (error == -EAGAIN)
2613 				error = 0;
2614 			goto out_no_agbp;
2615 		}
2616 	}
2617 
2618 	/*
2619 	 * If this is a metadata preferred pag and we are user data then try
2620 	 * somewhere else if we are not being asked to try harder at this
2621 	 * point
2622 	 */
2623 	if (pag->pagf_metadata && (args->datatype & XFS_ALLOC_USERDATA) &&
2624 	    (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
2625 		ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
2626 		goto out_agbp_relse;
2627 	}
2628 
2629 	need = xfs_alloc_min_freelist(mp, pag);
2630 	if (!xfs_alloc_space_available(args, need, flags |
2631 			XFS_ALLOC_FLAG_CHECK))
2632 		goto out_agbp_relse;
2633 
2634 	/*
2635 	 * Get the a.g. freespace buffer.
2636 	 * Can fail if we're not blocking on locks, and it's held.
2637 	 */
2638 	if (!agbp) {
2639 		error = xfs_alloc_read_agf(pag, tp, flags, &agbp);
2640 		if (error) {
2641 			/* Couldn't lock the AGF so skip this AG. */
2642 			if (error == -EAGAIN)
2643 				error = 0;
2644 			goto out_no_agbp;
2645 		}
2646 	}
2647 
2648 	/* reset a padding mismatched agfl before final free space check */
2649 	if (pag->pagf_agflreset)
2650 		xfs_agfl_reset(tp, agbp, pag);
2651 
2652 	/* If there isn't enough total space or single-extent, reject it. */
2653 	need = xfs_alloc_min_freelist(mp, pag);
2654 	if (!xfs_alloc_space_available(args, need, flags))
2655 		goto out_agbp_relse;
2656 
2657 #ifdef DEBUG
2658 	if (args->alloc_minlen_only) {
2659 		int stat;
2660 
2661 		error = xfs_exact_minlen_extent_available(args, agbp, &stat);
2662 		if (error || !stat)
2663 			goto out_agbp_relse;
2664 	}
2665 #endif
2666 	/*
2667 	 * Make the freelist shorter if it's too long.
2668 	 *
2669 	 * Note that from this point onwards, we will always release the agf and
2670 	 * agfl buffers on error. This handles the case where we error out and
2671 	 * the buffers are clean or may not have been joined to the transaction
2672 	 * and hence need to be released manually. If they have been joined to
2673 	 * the transaction, then xfs_trans_brelse() will handle them
2674 	 * appropriately based on the recursion count and dirty state of the
2675 	 * buffer.
2676 	 *
2677 	 * XXX (dgc): When we have lots of free space, does this buy us
2678 	 * anything other than extra overhead when we need to put more blocks
2679 	 * back on the free list? Maybe we should only do this when space is
2680 	 * getting low or the AGFL is more than half full?
2681 	 *
2682 	 * The NOSHRINK flag prevents the AGFL from being shrunk if it's too
2683 	 * big; the NORMAP flag prevents AGFL expand/shrink operations from
2684 	 * updating the rmapbt.  Both flags are used in xfs_repair while we're
2685 	 * rebuilding the rmapbt, and neither are used by the kernel.  They're
2686 	 * both required to ensure that rmaps are correctly recorded for the
2687 	 * regenerated AGFL, bnobt, and cntbt.  See repair/phase5.c and
2688 	 * repair/rmap.c in xfsprogs for details.
2689 	 */
2690 	memset(&targs, 0, sizeof(targs));
2691 	/* struct copy below */
2692 	if (flags & XFS_ALLOC_FLAG_NORMAP)
2693 		targs.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
2694 	else
2695 		targs.oinfo = XFS_RMAP_OINFO_AG;
2696 	while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) {
2697 		error = xfs_alloc_get_freelist(pag, tp, agbp, &bno, 0);
2698 		if (error)
2699 			goto out_agbp_relse;
2700 
2701 		/* defer agfl frees */
2702 		xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
2703 	}
2704 
2705 	targs.tp = tp;
2706 	targs.mp = mp;
2707 	targs.agbp = agbp;
2708 	targs.agno = args->agno;
2709 	targs.alignment = targs.minlen = targs.prod = 1;
2710 	targs.type = XFS_ALLOCTYPE_THIS_AG;
2711 	targs.pag = pag;
2712 	error = xfs_alloc_read_agfl(pag, tp, &agflbp);
2713 	if (error)
2714 		goto out_agbp_relse;
2715 
2716 	/* Make the freelist longer if it's too short. */
2717 	while (pag->pagf_flcount < need) {
2718 		targs.agbno = 0;
2719 		targs.maxlen = need - pag->pagf_flcount;
2720 		targs.resv = XFS_AG_RESV_AGFL;
2721 
2722 		/* Allocate as many blocks as possible at once. */
2723 		error = xfs_alloc_ag_vextent(&targs);
2724 		if (error)
2725 			goto out_agflbp_relse;
2726 
2727 		/*
2728 		 * Stop if we run out.  Won't happen if callers are obeying
2729 		 * the restrictions correctly.  Can happen for free calls
2730 		 * on a completely full ag.
2731 		 */
2732 		if (targs.agbno == NULLAGBLOCK) {
2733 			if (flags & XFS_ALLOC_FLAG_FREEING)
2734 				break;
2735 			goto out_agflbp_relse;
2736 		}
2737 		/*
2738 		 * Put each allocated block on the list.
2739 		 */
2740 		for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
2741 			error = xfs_alloc_put_freelist(pag, tp, agbp,
2742 							agflbp, bno, 0);
2743 			if (error)
2744 				goto out_agflbp_relse;
2745 		}
2746 	}
2747 	xfs_trans_brelse(tp, agflbp);
2748 	args->agbp = agbp;
2749 	return 0;
2750 
2751 out_agflbp_relse:
2752 	xfs_trans_brelse(tp, agflbp);
2753 out_agbp_relse:
2754 	if (agbp)
2755 		xfs_trans_brelse(tp, agbp);
2756 out_no_agbp:
2757 	args->agbp = NULL;
2758 	return error;
2759 }
2760 
2761 /*
2762  * Get a block from the freelist.
2763  * Returns with the buffer for the block gotten.
2764  */
2765 int
2766 xfs_alloc_get_freelist(
2767 	struct xfs_perag	*pag,
2768 	struct xfs_trans	*tp,
2769 	struct xfs_buf		*agbp,
2770 	xfs_agblock_t		*bnop,
2771 	int			btreeblk)
2772 {
2773 	struct xfs_agf		*agf = agbp->b_addr;
2774 	struct xfs_buf		*agflbp;
2775 	xfs_agblock_t		bno;
2776 	__be32			*agfl_bno;
2777 	int			error;
2778 	uint32_t		logflags;
2779 	struct xfs_mount	*mp = tp->t_mountp;
2780 
2781 	/*
2782 	 * Freelist is empty, give up.
2783 	 */
2784 	if (!agf->agf_flcount) {
2785 		*bnop = NULLAGBLOCK;
2786 		return 0;
2787 	}
2788 	/*
2789 	 * Read the array of free blocks.
2790 	 */
2791 	error = xfs_alloc_read_agfl(pag, tp, &agflbp);
2792 	if (error)
2793 		return error;
2794 
2795 
2796 	/*
2797 	 * Get the block number and update the data structures.
2798 	 */
2799 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
2800 	bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2801 	be32_add_cpu(&agf->agf_flfirst, 1);
2802 	xfs_trans_brelse(tp, agflbp);
2803 	if (be32_to_cpu(agf->agf_flfirst) == xfs_agfl_size(mp))
2804 		agf->agf_flfirst = 0;
2805 
2806 	ASSERT(!pag->pagf_agflreset);
2807 	be32_add_cpu(&agf->agf_flcount, -1);
2808 	pag->pagf_flcount--;
2809 
2810 	logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2811 	if (btreeblk) {
2812 		be32_add_cpu(&agf->agf_btreeblks, 1);
2813 		pag->pagf_btreeblks++;
2814 		logflags |= XFS_AGF_BTREEBLKS;
2815 	}
2816 
2817 	xfs_alloc_log_agf(tp, agbp, logflags);
2818 	*bnop = bno;
2819 
2820 	return 0;
2821 }
2822 
2823 /*
2824  * Log the given fields from the agf structure.
2825  */
2826 void
2827 xfs_alloc_log_agf(
2828 	struct xfs_trans	*tp,
2829 	struct xfs_buf		*bp,
2830 	uint32_t		fields)
2831 {
2832 	int	first;		/* first byte offset */
2833 	int	last;		/* last byte offset */
2834 	static const short	offsets[] = {
2835 		offsetof(xfs_agf_t, agf_magicnum),
2836 		offsetof(xfs_agf_t, agf_versionnum),
2837 		offsetof(xfs_agf_t, agf_seqno),
2838 		offsetof(xfs_agf_t, agf_length),
2839 		offsetof(xfs_agf_t, agf_roots[0]),
2840 		offsetof(xfs_agf_t, agf_levels[0]),
2841 		offsetof(xfs_agf_t, agf_flfirst),
2842 		offsetof(xfs_agf_t, agf_fllast),
2843 		offsetof(xfs_agf_t, agf_flcount),
2844 		offsetof(xfs_agf_t, agf_freeblks),
2845 		offsetof(xfs_agf_t, agf_longest),
2846 		offsetof(xfs_agf_t, agf_btreeblks),
2847 		offsetof(xfs_agf_t, agf_uuid),
2848 		offsetof(xfs_agf_t, agf_rmap_blocks),
2849 		offsetof(xfs_agf_t, agf_refcount_blocks),
2850 		offsetof(xfs_agf_t, agf_refcount_root),
2851 		offsetof(xfs_agf_t, agf_refcount_level),
2852 		/* needed so that we don't log the whole rest of the structure: */
2853 		offsetof(xfs_agf_t, agf_spare64),
2854 		sizeof(xfs_agf_t)
2855 	};
2856 
2857 	trace_xfs_agf(tp->t_mountp, bp->b_addr, fields, _RET_IP_);
2858 
2859 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF);
2860 
2861 	xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2862 	xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2863 }
2864 
2865 /*
2866  * Put the block on the freelist for the allocation group.
2867  */
2868 int
2869 xfs_alloc_put_freelist(
2870 	struct xfs_perag	*pag,
2871 	struct xfs_trans	*tp,
2872 	struct xfs_buf		*agbp,
2873 	struct xfs_buf		*agflbp,
2874 	xfs_agblock_t		bno,
2875 	int			btreeblk)
2876 {
2877 	struct xfs_mount	*mp = tp->t_mountp;
2878 	struct xfs_agf		*agf = agbp->b_addr;
2879 	__be32			*blockp;
2880 	int			error;
2881 	uint32_t		logflags;
2882 	__be32			*agfl_bno;
2883 	int			startoff;
2884 
2885 	if (!agflbp) {
2886 		error = xfs_alloc_read_agfl(pag, tp, &agflbp);
2887 		if (error)
2888 			return error;
2889 	}
2890 
2891 	be32_add_cpu(&agf->agf_fllast, 1);
2892 	if (be32_to_cpu(agf->agf_fllast) == xfs_agfl_size(mp))
2893 		agf->agf_fllast = 0;
2894 
2895 	ASSERT(!pag->pagf_agflreset);
2896 	be32_add_cpu(&agf->agf_flcount, 1);
2897 	pag->pagf_flcount++;
2898 
2899 	logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2900 	if (btreeblk) {
2901 		be32_add_cpu(&agf->agf_btreeblks, -1);
2902 		pag->pagf_btreeblks--;
2903 		logflags |= XFS_AGF_BTREEBLKS;
2904 	}
2905 
2906 	xfs_alloc_log_agf(tp, agbp, logflags);
2907 
2908 	ASSERT(be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp));
2909 
2910 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
2911 	blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)];
2912 	*blockp = cpu_to_be32(bno);
2913 	startoff = (char *)blockp - (char *)agflbp->b_addr;
2914 
2915 	xfs_alloc_log_agf(tp, agbp, logflags);
2916 
2917 	xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF);
2918 	xfs_trans_log_buf(tp, agflbp, startoff,
2919 			  startoff + sizeof(xfs_agblock_t) - 1);
2920 	return 0;
2921 }
2922 
2923 static xfs_failaddr_t
2924 xfs_agf_verify(
2925 	struct xfs_buf		*bp)
2926 {
2927 	struct xfs_mount	*mp = bp->b_mount;
2928 	struct xfs_agf		*agf = bp->b_addr;
2929 
2930 	if (xfs_has_crc(mp)) {
2931 		if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid))
2932 			return __this_address;
2933 		if (!xfs_log_check_lsn(mp, be64_to_cpu(agf->agf_lsn)))
2934 			return __this_address;
2935 	}
2936 
2937 	if (!xfs_verify_magic(bp, agf->agf_magicnum))
2938 		return __this_address;
2939 
2940 	if (!(XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2941 	      be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2942 	      be32_to_cpu(agf->agf_flfirst) < xfs_agfl_size(mp) &&
2943 	      be32_to_cpu(agf->agf_fllast) < xfs_agfl_size(mp) &&
2944 	      be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp)))
2945 		return __this_address;
2946 
2947 	if (be32_to_cpu(agf->agf_length) > mp->m_sb.sb_dblocks)
2948 		return __this_address;
2949 
2950 	if (be32_to_cpu(agf->agf_freeblks) < be32_to_cpu(agf->agf_longest) ||
2951 	    be32_to_cpu(agf->agf_freeblks) > be32_to_cpu(agf->agf_length))
2952 		return __this_address;
2953 
2954 	if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) < 1 ||
2955 	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) < 1 ||
2956 	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) >
2957 						mp->m_alloc_maxlevels ||
2958 	    be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) >
2959 						mp->m_alloc_maxlevels)
2960 		return __this_address;
2961 
2962 	if (xfs_has_rmapbt(mp) &&
2963 	    (be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) < 1 ||
2964 	     be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) >
2965 						mp->m_rmap_maxlevels))
2966 		return __this_address;
2967 
2968 	if (xfs_has_rmapbt(mp) &&
2969 	    be32_to_cpu(agf->agf_rmap_blocks) > be32_to_cpu(agf->agf_length))
2970 		return __this_address;
2971 
2972 	/*
2973 	 * during growfs operations, the perag is not fully initialised,
2974 	 * so we can't use it for any useful checking. growfs ensures we can't
2975 	 * use it by using uncached buffers that don't have the perag attached
2976 	 * so we can detect and avoid this problem.
2977 	 */
2978 	if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno)
2979 		return __this_address;
2980 
2981 	if (xfs_has_lazysbcount(mp) &&
2982 	    be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length))
2983 		return __this_address;
2984 
2985 	if (xfs_has_reflink(mp) &&
2986 	    be32_to_cpu(agf->agf_refcount_blocks) >
2987 	    be32_to_cpu(agf->agf_length))
2988 		return __this_address;
2989 
2990 	if (xfs_has_reflink(mp) &&
2991 	    (be32_to_cpu(agf->agf_refcount_level) < 1 ||
2992 	     be32_to_cpu(agf->agf_refcount_level) > mp->m_refc_maxlevels))
2993 		return __this_address;
2994 
2995 	return NULL;
2996 
2997 }
2998 
2999 static void
3000 xfs_agf_read_verify(
3001 	struct xfs_buf	*bp)
3002 {
3003 	struct xfs_mount *mp = bp->b_mount;
3004 	xfs_failaddr_t	fa;
3005 
3006 	if (xfs_has_crc(mp) &&
3007 	    !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF))
3008 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
3009 	else {
3010 		fa = xfs_agf_verify(bp);
3011 		if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_ALLOC_READ_AGF))
3012 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
3013 	}
3014 }
3015 
3016 static void
3017 xfs_agf_write_verify(
3018 	struct xfs_buf	*bp)
3019 {
3020 	struct xfs_mount	*mp = bp->b_mount;
3021 	struct xfs_buf_log_item	*bip = bp->b_log_item;
3022 	struct xfs_agf		*agf = bp->b_addr;
3023 	xfs_failaddr_t		fa;
3024 
3025 	fa = xfs_agf_verify(bp);
3026 	if (fa) {
3027 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
3028 		return;
3029 	}
3030 
3031 	if (!xfs_has_crc(mp))
3032 		return;
3033 
3034 	if (bip)
3035 		agf->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn);
3036 
3037 	xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF);
3038 }
3039 
3040 const struct xfs_buf_ops xfs_agf_buf_ops = {
3041 	.name = "xfs_agf",
3042 	.magic = { cpu_to_be32(XFS_AGF_MAGIC), cpu_to_be32(XFS_AGF_MAGIC) },
3043 	.verify_read = xfs_agf_read_verify,
3044 	.verify_write = xfs_agf_write_verify,
3045 	.verify_struct = xfs_agf_verify,
3046 };
3047 
3048 /*
3049  * Read in the allocation group header (free/alloc section).
3050  */
3051 int
3052 xfs_read_agf(
3053 	struct xfs_perag	*pag,
3054 	struct xfs_trans	*tp,
3055 	int			flags,
3056 	struct xfs_buf		**agfbpp)
3057 {
3058 	struct xfs_mount	*mp = pag->pag_mount;
3059 	int			error;
3060 
3061 	trace_xfs_read_agf(pag->pag_mount, pag->pag_agno);
3062 
3063 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
3064 			XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGF_DADDR(mp)),
3065 			XFS_FSS_TO_BB(mp, 1), flags, agfbpp, &xfs_agf_buf_ops);
3066 	if (error)
3067 		return error;
3068 
3069 	xfs_buf_set_ref(*agfbpp, XFS_AGF_REF);
3070 	return 0;
3071 }
3072 
3073 /*
3074  * Read in the allocation group header (free/alloc section) and initialise the
3075  * perag structure if necessary. If the caller provides @agfbpp, then return the
3076  * locked buffer to the caller, otherwise free it.
3077  */
3078 int
3079 xfs_alloc_read_agf(
3080 	struct xfs_perag	*pag,
3081 	struct xfs_trans	*tp,
3082 	int			flags,
3083 	struct xfs_buf		**agfbpp)
3084 {
3085 	struct xfs_buf		*agfbp;
3086 	struct xfs_agf		*agf;
3087 	int			error;
3088 	int			allocbt_blks;
3089 
3090 	trace_xfs_alloc_read_agf(pag->pag_mount, pag->pag_agno);
3091 
3092 	/* We don't support trylock when freeing. */
3093 	ASSERT((flags & (XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK)) !=
3094 			(XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK));
3095 	error = xfs_read_agf(pag, tp,
3096 			(flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
3097 			&agfbp);
3098 	if (error)
3099 		return error;
3100 
3101 	agf = agfbp->b_addr;
3102 	if (!pag->pagf_init) {
3103 		pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
3104 		pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
3105 		pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
3106 		pag->pagf_longest = be32_to_cpu(agf->agf_longest);
3107 		pag->pagf_levels[XFS_BTNUM_BNOi] =
3108 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
3109 		pag->pagf_levels[XFS_BTNUM_CNTi] =
3110 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
3111 		pag->pagf_levels[XFS_BTNUM_RMAPi] =
3112 			be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]);
3113 		pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level);
3114 		pag->pagf_init = 1;
3115 		pag->pagf_agflreset = xfs_agfl_needs_reset(pag->pag_mount, agf);
3116 
3117 		/*
3118 		 * Update the in-core allocbt counter. Filter out the rmapbt
3119 		 * subset of the btreeblks counter because the rmapbt is managed
3120 		 * by perag reservation. Subtract one for the rmapbt root block
3121 		 * because the rmap counter includes it while the btreeblks
3122 		 * counter only tracks non-root blocks.
3123 		 */
3124 		allocbt_blks = pag->pagf_btreeblks;
3125 		if (xfs_has_rmapbt(pag->pag_mount))
3126 			allocbt_blks -= be32_to_cpu(agf->agf_rmap_blocks) - 1;
3127 		if (allocbt_blks > 0)
3128 			atomic64_add(allocbt_blks,
3129 					&pag->pag_mount->m_allocbt_blks);
3130 	}
3131 #ifdef DEBUG
3132 	else if (!xfs_is_shutdown(pag->pag_mount)) {
3133 		ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
3134 		ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
3135 		ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
3136 		ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
3137 		ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
3138 		       be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
3139 		ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
3140 		       be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
3141 	}
3142 #endif
3143 	if (agfbpp)
3144 		*agfbpp = agfbp;
3145 	else
3146 		xfs_trans_brelse(tp, agfbp);
3147 	return 0;
3148 }
3149 
3150 /*
3151  * Allocate an extent (variable-size).
3152  * Depending on the allocation type, we either look in a single allocation
3153  * group or loop over the allocation groups to find the result.
3154  */
3155 int				/* error */
3156 xfs_alloc_vextent(
3157 	struct xfs_alloc_arg	*args)	/* allocation argument structure */
3158 {
3159 	xfs_agblock_t		agsize;	/* allocation group size */
3160 	int			error;
3161 	int			flags;	/* XFS_ALLOC_FLAG_... locking flags */
3162 	struct xfs_mount	*mp;	/* mount structure pointer */
3163 	xfs_agnumber_t		sagno;	/* starting allocation group number */
3164 	xfs_alloctype_t		type;	/* input allocation type */
3165 	int			bump_rotor = 0;
3166 	xfs_agnumber_t		rotorstep = xfs_rotorstep; /* inode32 agf stepper */
3167 
3168 	mp = args->mp;
3169 	type = args->otype = args->type;
3170 	args->agbno = NULLAGBLOCK;
3171 	/*
3172 	 * Just fix this up, for the case where the last a.g. is shorter
3173 	 * (or there's only one a.g.) and the caller couldn't easily figure
3174 	 * that out (xfs_bmap_alloc).
3175 	 */
3176 	agsize = mp->m_sb.sb_agblocks;
3177 	if (args->maxlen > agsize)
3178 		args->maxlen = agsize;
3179 	if (args->alignment == 0)
3180 		args->alignment = 1;
3181 	ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
3182 	ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
3183 	ASSERT(args->minlen <= args->maxlen);
3184 	ASSERT(args->minlen <= agsize);
3185 	ASSERT(args->mod < args->prod);
3186 	if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
3187 	    XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
3188 	    args->minlen > args->maxlen || args->minlen > agsize ||
3189 	    args->mod >= args->prod) {
3190 		args->fsbno = NULLFSBLOCK;
3191 		trace_xfs_alloc_vextent_badargs(args);
3192 		return 0;
3193 	}
3194 
3195 	switch (type) {
3196 	case XFS_ALLOCTYPE_THIS_AG:
3197 	case XFS_ALLOCTYPE_NEAR_BNO:
3198 	case XFS_ALLOCTYPE_THIS_BNO:
3199 		/*
3200 		 * These three force us into a single a.g.
3201 		 */
3202 		args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
3203 		args->pag = xfs_perag_get(mp, args->agno);
3204 		error = xfs_alloc_fix_freelist(args, 0);
3205 		if (error) {
3206 			trace_xfs_alloc_vextent_nofix(args);
3207 			goto error0;
3208 		}
3209 		if (!args->agbp) {
3210 			trace_xfs_alloc_vextent_noagbp(args);
3211 			break;
3212 		}
3213 		args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
3214 		if ((error = xfs_alloc_ag_vextent(args)))
3215 			goto error0;
3216 		break;
3217 	case XFS_ALLOCTYPE_START_BNO:
3218 		/*
3219 		 * Try near allocation first, then anywhere-in-ag after
3220 		 * the first a.g. fails.
3221 		 */
3222 		if ((args->datatype & XFS_ALLOC_INITIAL_USER_DATA) &&
3223 		    xfs_is_inode32(mp)) {
3224 			args->fsbno = XFS_AGB_TO_FSB(mp,
3225 					((mp->m_agfrotor / rotorstep) %
3226 					mp->m_sb.sb_agcount), 0);
3227 			bump_rotor = 1;
3228 		}
3229 		args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
3230 		args->type = XFS_ALLOCTYPE_NEAR_BNO;
3231 		fallthrough;
3232 	case XFS_ALLOCTYPE_FIRST_AG:
3233 		/*
3234 		 * Rotate through the allocation groups looking for a winner.
3235 		 */
3236 		if (type == XFS_ALLOCTYPE_FIRST_AG) {
3237 			/*
3238 			 * Start with allocation group given by bno.
3239 			 */
3240 			args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
3241 			args->type = XFS_ALLOCTYPE_THIS_AG;
3242 			sagno = 0;
3243 			flags = 0;
3244 		} else {
3245 			/*
3246 			 * Start with the given allocation group.
3247 			 */
3248 			args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
3249 			flags = XFS_ALLOC_FLAG_TRYLOCK;
3250 		}
3251 		/*
3252 		 * Loop over allocation groups twice; first time with
3253 		 * trylock set, second time without.
3254 		 */
3255 		for (;;) {
3256 			args->pag = xfs_perag_get(mp, args->agno);
3257 			error = xfs_alloc_fix_freelist(args, flags);
3258 			if (error) {
3259 				trace_xfs_alloc_vextent_nofix(args);
3260 				goto error0;
3261 			}
3262 			/*
3263 			 * If we get a buffer back then the allocation will fly.
3264 			 */
3265 			if (args->agbp) {
3266 				if ((error = xfs_alloc_ag_vextent(args)))
3267 					goto error0;
3268 				break;
3269 			}
3270 
3271 			trace_xfs_alloc_vextent_loopfailed(args);
3272 
3273 			/*
3274 			 * Didn't work, figure out the next iteration.
3275 			 */
3276 			if (args->agno == sagno &&
3277 			    type == XFS_ALLOCTYPE_START_BNO)
3278 				args->type = XFS_ALLOCTYPE_THIS_AG;
3279 			/*
3280 			* For the first allocation, we can try any AG to get
3281 			* space.  However, if we already have allocated a
3282 			* block, we don't want to try AGs whose number is below
3283 			* sagno. Otherwise, we may end up with out-of-order
3284 			* locking of AGF, which might cause deadlock.
3285 			*/
3286 			if (++(args->agno) == mp->m_sb.sb_agcount) {
3287 				if (args->tp->t_firstblock != NULLFSBLOCK)
3288 					args->agno = sagno;
3289 				else
3290 					args->agno = 0;
3291 			}
3292 			/*
3293 			 * Reached the starting a.g., must either be done
3294 			 * or switch to non-trylock mode.
3295 			 */
3296 			if (args->agno == sagno) {
3297 				if (flags == 0) {
3298 					args->agbno = NULLAGBLOCK;
3299 					trace_xfs_alloc_vextent_allfailed(args);
3300 					break;
3301 				}
3302 
3303 				flags = 0;
3304 				if (type == XFS_ALLOCTYPE_START_BNO) {
3305 					args->agbno = XFS_FSB_TO_AGBNO(mp,
3306 						args->fsbno);
3307 					args->type = XFS_ALLOCTYPE_NEAR_BNO;
3308 				}
3309 			}
3310 			xfs_perag_put(args->pag);
3311 		}
3312 		if (bump_rotor) {
3313 			if (args->agno == sagno)
3314 				mp->m_agfrotor = (mp->m_agfrotor + 1) %
3315 					(mp->m_sb.sb_agcount * rotorstep);
3316 			else
3317 				mp->m_agfrotor = (args->agno * rotorstep + 1) %
3318 					(mp->m_sb.sb_agcount * rotorstep);
3319 		}
3320 		break;
3321 	default:
3322 		ASSERT(0);
3323 		/* NOTREACHED */
3324 	}
3325 	if (args->agbno == NULLAGBLOCK)
3326 		args->fsbno = NULLFSBLOCK;
3327 	else {
3328 		args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
3329 #ifdef DEBUG
3330 		ASSERT(args->len >= args->minlen);
3331 		ASSERT(args->len <= args->maxlen);
3332 		ASSERT(args->agbno % args->alignment == 0);
3333 		XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
3334 			args->len);
3335 #endif
3336 
3337 	}
3338 	xfs_perag_put(args->pag);
3339 	return 0;
3340 error0:
3341 	xfs_perag_put(args->pag);
3342 	return error;
3343 }
3344 
3345 /* Ensure that the freelist is at full capacity. */
3346 int
3347 xfs_free_extent_fix_freelist(
3348 	struct xfs_trans	*tp,
3349 	struct xfs_perag	*pag,
3350 	struct xfs_buf		**agbp)
3351 {
3352 	struct xfs_alloc_arg	args;
3353 	int			error;
3354 
3355 	memset(&args, 0, sizeof(struct xfs_alloc_arg));
3356 	args.tp = tp;
3357 	args.mp = tp->t_mountp;
3358 	args.agno = pag->pag_agno;
3359 	args.pag = pag;
3360 
3361 	/*
3362 	 * validate that the block number is legal - the enables us to detect
3363 	 * and handle a silent filesystem corruption rather than crashing.
3364 	 */
3365 	if (args.agno >= args.mp->m_sb.sb_agcount)
3366 		return -EFSCORRUPTED;
3367 
3368 	error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
3369 	if (error)
3370 		return error;
3371 
3372 	*agbp = args.agbp;
3373 	return 0;
3374 }
3375 
3376 /*
3377  * Free an extent.
3378  * Just break up the extent address and hand off to xfs_free_ag_extent
3379  * after fixing up the freelist.
3380  */
3381 int
3382 __xfs_free_extent(
3383 	struct xfs_trans		*tp,
3384 	xfs_fsblock_t			bno,
3385 	xfs_extlen_t			len,
3386 	const struct xfs_owner_info	*oinfo,
3387 	enum xfs_ag_resv_type		type,
3388 	bool				skip_discard)
3389 {
3390 	struct xfs_mount		*mp = tp->t_mountp;
3391 	struct xfs_buf			*agbp;
3392 	xfs_agnumber_t			agno = XFS_FSB_TO_AGNO(mp, bno);
3393 	xfs_agblock_t			agbno = XFS_FSB_TO_AGBNO(mp, bno);
3394 	struct xfs_agf			*agf;
3395 	int				error;
3396 	unsigned int			busy_flags = 0;
3397 	struct xfs_perag		*pag;
3398 
3399 	ASSERT(len != 0);
3400 	ASSERT(type != XFS_AG_RESV_AGFL);
3401 
3402 	if (XFS_TEST_ERROR(false, mp,
3403 			XFS_ERRTAG_FREE_EXTENT))
3404 		return -EIO;
3405 
3406 	pag = xfs_perag_get(mp, agno);
3407 	error = xfs_free_extent_fix_freelist(tp, pag, &agbp);
3408 	if (error)
3409 		goto err;
3410 	agf = agbp->b_addr;
3411 
3412 	if (XFS_IS_CORRUPT(mp, agbno >= mp->m_sb.sb_agblocks)) {
3413 		error = -EFSCORRUPTED;
3414 		goto err_release;
3415 	}
3416 
3417 	/* validate the extent size is legal now we have the agf locked */
3418 	if (XFS_IS_CORRUPT(mp, agbno + len > be32_to_cpu(agf->agf_length))) {
3419 		error = -EFSCORRUPTED;
3420 		goto err_release;
3421 	}
3422 
3423 	error = xfs_free_ag_extent(tp, agbp, agno, agbno, len, oinfo, type);
3424 	if (error)
3425 		goto err_release;
3426 
3427 	if (skip_discard)
3428 		busy_flags |= XFS_EXTENT_BUSY_SKIP_DISCARD;
3429 	xfs_extent_busy_insert(tp, pag, agbno, len, busy_flags);
3430 	xfs_perag_put(pag);
3431 	return 0;
3432 
3433 err_release:
3434 	xfs_trans_brelse(tp, agbp);
3435 err:
3436 	xfs_perag_put(pag);
3437 	return error;
3438 }
3439 
3440 struct xfs_alloc_query_range_info {
3441 	xfs_alloc_query_range_fn	fn;
3442 	void				*priv;
3443 };
3444 
3445 /* Format btree record and pass to our callback. */
3446 STATIC int
3447 xfs_alloc_query_range_helper(
3448 	struct xfs_btree_cur		*cur,
3449 	const union xfs_btree_rec	*rec,
3450 	void				*priv)
3451 {
3452 	struct xfs_alloc_query_range_info	*query = priv;
3453 	struct xfs_alloc_rec_incore		irec;
3454 
3455 	irec.ar_startblock = be32_to_cpu(rec->alloc.ar_startblock);
3456 	irec.ar_blockcount = be32_to_cpu(rec->alloc.ar_blockcount);
3457 	return query->fn(cur, &irec, query->priv);
3458 }
3459 
3460 /* Find all free space within a given range of blocks. */
3461 int
3462 xfs_alloc_query_range(
3463 	struct xfs_btree_cur			*cur,
3464 	const struct xfs_alloc_rec_incore	*low_rec,
3465 	const struct xfs_alloc_rec_incore	*high_rec,
3466 	xfs_alloc_query_range_fn		fn,
3467 	void					*priv)
3468 {
3469 	union xfs_btree_irec			low_brec;
3470 	union xfs_btree_irec			high_brec;
3471 	struct xfs_alloc_query_range_info	query;
3472 
3473 	ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3474 	low_brec.a = *low_rec;
3475 	high_brec.a = *high_rec;
3476 	query.priv = priv;
3477 	query.fn = fn;
3478 	return xfs_btree_query_range(cur, &low_brec, &high_brec,
3479 			xfs_alloc_query_range_helper, &query);
3480 }
3481 
3482 /* Find all free space records. */
3483 int
3484 xfs_alloc_query_all(
3485 	struct xfs_btree_cur			*cur,
3486 	xfs_alloc_query_range_fn		fn,
3487 	void					*priv)
3488 {
3489 	struct xfs_alloc_query_range_info	query;
3490 
3491 	ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3492 	query.priv = priv;
3493 	query.fn = fn;
3494 	return xfs_btree_query_all(cur, xfs_alloc_query_range_helper, &query);
3495 }
3496 
3497 /* Is there a record covering a given extent? */
3498 int
3499 xfs_alloc_has_record(
3500 	struct xfs_btree_cur	*cur,
3501 	xfs_agblock_t		bno,
3502 	xfs_extlen_t		len,
3503 	bool			*exists)
3504 {
3505 	union xfs_btree_irec	low;
3506 	union xfs_btree_irec	high;
3507 
3508 	memset(&low, 0, sizeof(low));
3509 	low.a.ar_startblock = bno;
3510 	memset(&high, 0xFF, sizeof(high));
3511 	high.a.ar_startblock = bno + len - 1;
3512 
3513 	return xfs_btree_has_record(cur, &low, &high, exists);
3514 }
3515 
3516 /*
3517  * Walk all the blocks in the AGFL.  The @walk_fn can return any negative
3518  * error code or XFS_ITER_*.
3519  */
3520 int
3521 xfs_agfl_walk(
3522 	struct xfs_mount	*mp,
3523 	struct xfs_agf		*agf,
3524 	struct xfs_buf		*agflbp,
3525 	xfs_agfl_walk_fn	walk_fn,
3526 	void			*priv)
3527 {
3528 	__be32			*agfl_bno;
3529 	unsigned int		i;
3530 	int			error;
3531 
3532 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
3533 	i = be32_to_cpu(agf->agf_flfirst);
3534 
3535 	/* Nothing to walk in an empty AGFL. */
3536 	if (agf->agf_flcount == cpu_to_be32(0))
3537 		return 0;
3538 
3539 	/* Otherwise, walk from first to last, wrapping as needed. */
3540 	for (;;) {
3541 		error = walk_fn(mp, be32_to_cpu(agfl_bno[i]), priv);
3542 		if (error)
3543 			return error;
3544 		if (i == be32_to_cpu(agf->agf_fllast))
3545 			break;
3546 		if (++i == xfs_agfl_size(mp))
3547 			i = 0;
3548 	}
3549 
3550 	return 0;
3551 }
3552 
3553 int __init
3554 xfs_extfree_intent_init_cache(void)
3555 {
3556 	xfs_extfree_item_cache = kmem_cache_create("xfs_extfree_intent",
3557 			sizeof(struct xfs_extent_free_item),
3558 			0, 0, NULL);
3559 
3560 	return xfs_extfree_item_cache != NULL ? 0 : -ENOMEM;
3561 }
3562 
3563 void
3564 xfs_extfree_intent_destroy_cache(void)
3565 {
3566 	kmem_cache_destroy(xfs_extfree_item_cache);
3567 	xfs_extfree_item_cache = NULL;
3568 }
3569