xref: /openbmc/linux/fs/xfs/libxfs/xfs_ialloc.c (revision faa16bc4)
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_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_ialloc_btree.h"
20 #include "xfs_alloc.h"
21 #include "xfs_rtalloc.h"
22 #include "xfs_errortag.h"
23 #include "xfs_error.h"
24 #include "xfs_bmap.h"
25 #include "xfs_cksum.h"
26 #include "xfs_trans.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_icreate_item.h"
29 #include "xfs_icache.h"
30 #include "xfs_trace.h"
31 #include "xfs_log.h"
32 #include "xfs_rmap.h"
33 
34 
35 /*
36  * Allocation group level functions.
37  */
38 int
39 xfs_ialloc_cluster_alignment(
40 	struct xfs_mount	*mp)
41 {
42 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
43 	    mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
44 		return mp->m_sb.sb_inoalignmt;
45 	return 1;
46 }
47 
48 /*
49  * Lookup a record by ino in the btree given by cur.
50  */
51 int					/* error */
52 xfs_inobt_lookup(
53 	struct xfs_btree_cur	*cur,	/* btree cursor */
54 	xfs_agino_t		ino,	/* starting inode of chunk */
55 	xfs_lookup_t		dir,	/* <=, >=, == */
56 	int			*stat)	/* success/failure */
57 {
58 	cur->bc_rec.i.ir_startino = ino;
59 	cur->bc_rec.i.ir_holemask = 0;
60 	cur->bc_rec.i.ir_count = 0;
61 	cur->bc_rec.i.ir_freecount = 0;
62 	cur->bc_rec.i.ir_free = 0;
63 	return xfs_btree_lookup(cur, dir, stat);
64 }
65 
66 /*
67  * Update the record referred to by cur to the value given.
68  * This either works (return 0) or gets an EFSCORRUPTED error.
69  */
70 STATIC int				/* error */
71 xfs_inobt_update(
72 	struct xfs_btree_cur	*cur,	/* btree cursor */
73 	xfs_inobt_rec_incore_t	*irec)	/* btree record */
74 {
75 	union xfs_btree_rec	rec;
76 
77 	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
78 	if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
79 		rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
80 		rec.inobt.ir_u.sp.ir_count = irec->ir_count;
81 		rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
82 	} else {
83 		/* ir_holemask/ir_count not supported on-disk */
84 		rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
85 	}
86 	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
87 	return xfs_btree_update(cur, &rec);
88 }
89 
90 /* Convert on-disk btree record to incore inobt record. */
91 void
92 xfs_inobt_btrec_to_irec(
93 	struct xfs_mount		*mp,
94 	union xfs_btree_rec		*rec,
95 	struct xfs_inobt_rec_incore	*irec)
96 {
97 	irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
98 	if (xfs_sb_version_hassparseinodes(&mp->m_sb)) {
99 		irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
100 		irec->ir_count = rec->inobt.ir_u.sp.ir_count;
101 		irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
102 	} else {
103 		/*
104 		 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
105 		 * values for full inode chunks.
106 		 */
107 		irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
108 		irec->ir_count = XFS_INODES_PER_CHUNK;
109 		irec->ir_freecount =
110 				be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
111 	}
112 	irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
113 }
114 
115 /*
116  * Get the data from the pointed-to record.
117  */
118 int
119 xfs_inobt_get_rec(
120 	struct xfs_btree_cur		*cur,
121 	struct xfs_inobt_rec_incore	*irec,
122 	int				*stat)
123 {
124 	struct xfs_mount		*mp = cur->bc_mp;
125 	xfs_agnumber_t			agno = cur->bc_private.a.agno;
126 	union xfs_btree_rec		*rec;
127 	int				error;
128 	uint64_t			realfree;
129 
130 	error = xfs_btree_get_rec(cur, &rec, stat);
131 	if (error || *stat == 0)
132 		return error;
133 
134 	xfs_inobt_btrec_to_irec(mp, rec, irec);
135 
136 	if (!xfs_verify_agino(mp, agno, irec->ir_startino))
137 		goto out_bad_rec;
138 	if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
139 	    irec->ir_count > XFS_INODES_PER_CHUNK)
140 		goto out_bad_rec;
141 	if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
142 		goto out_bad_rec;
143 
144 	/* if there are no holes, return the first available offset */
145 	if (!xfs_inobt_issparse(irec->ir_holemask))
146 		realfree = irec->ir_free;
147 	else
148 		realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
149 	if (hweight64(realfree) != irec->ir_freecount)
150 		goto out_bad_rec;
151 
152 	return 0;
153 
154 out_bad_rec:
155 	xfs_warn(mp,
156 		"%s Inode BTree record corruption in AG %d detected!",
157 		cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", agno);
158 	xfs_warn(mp,
159 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
160 		irec->ir_startino, irec->ir_count, irec->ir_freecount,
161 		irec->ir_free, irec->ir_holemask);
162 	return -EFSCORRUPTED;
163 }
164 
165 /*
166  * Insert a single inobt record. Cursor must already point to desired location.
167  */
168 int
169 xfs_inobt_insert_rec(
170 	struct xfs_btree_cur	*cur,
171 	uint16_t		holemask,
172 	uint8_t			count,
173 	int32_t			freecount,
174 	xfs_inofree_t		free,
175 	int			*stat)
176 {
177 	cur->bc_rec.i.ir_holemask = holemask;
178 	cur->bc_rec.i.ir_count = count;
179 	cur->bc_rec.i.ir_freecount = freecount;
180 	cur->bc_rec.i.ir_free = free;
181 	return xfs_btree_insert(cur, stat);
182 }
183 
184 /*
185  * Insert records describing a newly allocated inode chunk into the inobt.
186  */
187 STATIC int
188 xfs_inobt_insert(
189 	struct xfs_mount	*mp,
190 	struct xfs_trans	*tp,
191 	struct xfs_buf		*agbp,
192 	xfs_agino_t		newino,
193 	xfs_agino_t		newlen,
194 	xfs_btnum_t		btnum)
195 {
196 	struct xfs_btree_cur	*cur;
197 	struct xfs_agi		*agi = XFS_BUF_TO_AGI(agbp);
198 	xfs_agnumber_t		agno = be32_to_cpu(agi->agi_seqno);
199 	xfs_agino_t		thisino;
200 	int			i;
201 	int			error;
202 
203 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
204 
205 	for (thisino = newino;
206 	     thisino < newino + newlen;
207 	     thisino += XFS_INODES_PER_CHUNK) {
208 		error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
209 		if (error) {
210 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
211 			return error;
212 		}
213 		ASSERT(i == 0);
214 
215 		error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
216 					     XFS_INODES_PER_CHUNK,
217 					     XFS_INODES_PER_CHUNK,
218 					     XFS_INOBT_ALL_FREE, &i);
219 		if (error) {
220 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
221 			return error;
222 		}
223 		ASSERT(i == 1);
224 	}
225 
226 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
227 
228 	return 0;
229 }
230 
231 /*
232  * Verify that the number of free inodes in the AGI is correct.
233  */
234 #ifdef DEBUG
235 STATIC int
236 xfs_check_agi_freecount(
237 	struct xfs_btree_cur	*cur,
238 	struct xfs_agi		*agi)
239 {
240 	if (cur->bc_nlevels == 1) {
241 		xfs_inobt_rec_incore_t rec;
242 		int		freecount = 0;
243 		int		error;
244 		int		i;
245 
246 		error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
247 		if (error)
248 			return error;
249 
250 		do {
251 			error = xfs_inobt_get_rec(cur, &rec, &i);
252 			if (error)
253 				return error;
254 
255 			if (i) {
256 				freecount += rec.ir_freecount;
257 				error = xfs_btree_increment(cur, 0, &i);
258 				if (error)
259 					return error;
260 			}
261 		} while (i == 1);
262 
263 		if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
264 			ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
265 	}
266 	return 0;
267 }
268 #else
269 #define xfs_check_agi_freecount(cur, agi)	0
270 #endif
271 
272 /*
273  * Initialise a new set of inodes. When called without a transaction context
274  * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
275  * than logging them (which in a transaction context puts them into the AIL
276  * for writeback rather than the xfsbufd queue).
277  */
278 int
279 xfs_ialloc_inode_init(
280 	struct xfs_mount	*mp,
281 	struct xfs_trans	*tp,
282 	struct list_head	*buffer_list,
283 	int			icount,
284 	xfs_agnumber_t		agno,
285 	xfs_agblock_t		agbno,
286 	xfs_agblock_t		length,
287 	unsigned int		gen)
288 {
289 	struct xfs_buf		*fbuf;
290 	struct xfs_dinode	*free;
291 	int			nbufs, blks_per_cluster, inodes_per_cluster;
292 	int			version;
293 	int			i, j;
294 	xfs_daddr_t		d;
295 	xfs_ino_t		ino = 0;
296 
297 	/*
298 	 * Loop over the new block(s), filling in the inodes.  For small block
299 	 * sizes, manipulate the inodes in buffers  which are multiples of the
300 	 * blocks size.
301 	 */
302 	blks_per_cluster = xfs_icluster_size_fsb(mp);
303 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
304 	nbufs = length / blks_per_cluster;
305 
306 	/*
307 	 * Figure out what version number to use in the inodes we create.  If
308 	 * the superblock version has caught up to the one that supports the new
309 	 * inode format, then use the new inode version.  Otherwise use the old
310 	 * version so that old kernels will continue to be able to use the file
311 	 * system.
312 	 *
313 	 * For v3 inodes, we also need to write the inode number into the inode,
314 	 * so calculate the first inode number of the chunk here as
315 	 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
316 	 * across multiple filesystem blocks (such as a cluster) and so cannot
317 	 * be used in the cluster buffer loop below.
318 	 *
319 	 * Further, because we are writing the inode directly into the buffer
320 	 * and calculating a CRC on the entire inode, we have ot log the entire
321 	 * inode so that the entire range the CRC covers is present in the log.
322 	 * That means for v3 inode we log the entire buffer rather than just the
323 	 * inode cores.
324 	 */
325 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
326 		version = 3;
327 		ino = XFS_AGINO_TO_INO(mp, agno,
328 				       XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
329 
330 		/*
331 		 * log the initialisation that is about to take place as an
332 		 * logical operation. This means the transaction does not
333 		 * need to log the physical changes to the inode buffers as log
334 		 * recovery will know what initialisation is actually needed.
335 		 * Hence we only need to log the buffers as "ordered" buffers so
336 		 * they track in the AIL as if they were physically logged.
337 		 */
338 		if (tp)
339 			xfs_icreate_log(tp, agno, agbno, icount,
340 					mp->m_sb.sb_inodesize, length, gen);
341 	} else
342 		version = 2;
343 
344 	for (j = 0; j < nbufs; j++) {
345 		/*
346 		 * Get the block.
347 		 */
348 		d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
349 		fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
350 					 mp->m_bsize * blks_per_cluster,
351 					 XBF_UNMAPPED);
352 		if (!fbuf)
353 			return -ENOMEM;
354 
355 		/* Initialize the inode buffers and log them appropriately. */
356 		fbuf->b_ops = &xfs_inode_buf_ops;
357 		xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
358 		for (i = 0; i < inodes_per_cluster; i++) {
359 			int	ioffset = i << mp->m_sb.sb_inodelog;
360 			uint	isize = xfs_dinode_size(version);
361 
362 			free = xfs_make_iptr(mp, fbuf, i);
363 			free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
364 			free->di_version = version;
365 			free->di_gen = cpu_to_be32(gen);
366 			free->di_next_unlinked = cpu_to_be32(NULLAGINO);
367 
368 			if (version == 3) {
369 				free->di_ino = cpu_to_be64(ino);
370 				ino++;
371 				uuid_copy(&free->di_uuid,
372 					  &mp->m_sb.sb_meta_uuid);
373 				xfs_dinode_calc_crc(mp, free);
374 			} else if (tp) {
375 				/* just log the inode core */
376 				xfs_trans_log_buf(tp, fbuf, ioffset,
377 						  ioffset + isize - 1);
378 			}
379 		}
380 
381 		if (tp) {
382 			/*
383 			 * Mark the buffer as an inode allocation buffer so it
384 			 * sticks in AIL at the point of this allocation
385 			 * transaction. This ensures the they are on disk before
386 			 * the tail of the log can be moved past this
387 			 * transaction (i.e. by preventing relogging from moving
388 			 * it forward in the log).
389 			 */
390 			xfs_trans_inode_alloc_buf(tp, fbuf);
391 			if (version == 3) {
392 				/*
393 				 * Mark the buffer as ordered so that they are
394 				 * not physically logged in the transaction but
395 				 * still tracked in the AIL as part of the
396 				 * transaction and pin the log appropriately.
397 				 */
398 				xfs_trans_ordered_buf(tp, fbuf);
399 			}
400 		} else {
401 			fbuf->b_flags |= XBF_DONE;
402 			xfs_buf_delwri_queue(fbuf, buffer_list);
403 			xfs_buf_relse(fbuf);
404 		}
405 	}
406 	return 0;
407 }
408 
409 /*
410  * Align startino and allocmask for a recently allocated sparse chunk such that
411  * they are fit for insertion (or merge) into the on-disk inode btrees.
412  *
413  * Background:
414  *
415  * When enabled, sparse inode support increases the inode alignment from cluster
416  * size to inode chunk size. This means that the minimum range between two
417  * non-adjacent inode records in the inobt is large enough for a full inode
418  * record. This allows for cluster sized, cluster aligned block allocation
419  * without need to worry about whether the resulting inode record overlaps with
420  * another record in the tree. Without this basic rule, we would have to deal
421  * with the consequences of overlap by potentially undoing recent allocations in
422  * the inode allocation codepath.
423  *
424  * Because of this alignment rule (which is enforced on mount), there are two
425  * inobt possibilities for newly allocated sparse chunks. One is that the
426  * aligned inode record for the chunk covers a range of inodes not already
427  * covered in the inobt (i.e., it is safe to insert a new sparse record). The
428  * other is that a record already exists at the aligned startino that considers
429  * the newly allocated range as sparse. In the latter case, record content is
430  * merged in hope that sparse inode chunks fill to full chunks over time.
431  */
432 STATIC void
433 xfs_align_sparse_ino(
434 	struct xfs_mount		*mp,
435 	xfs_agino_t			*startino,
436 	uint16_t			*allocmask)
437 {
438 	xfs_agblock_t			agbno;
439 	xfs_agblock_t			mod;
440 	int				offset;
441 
442 	agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
443 	mod = agbno % mp->m_sb.sb_inoalignmt;
444 	if (!mod)
445 		return;
446 
447 	/* calculate the inode offset and align startino */
448 	offset = mod << mp->m_sb.sb_inopblog;
449 	*startino -= offset;
450 
451 	/*
452 	 * Since startino has been aligned down, left shift allocmask such that
453 	 * it continues to represent the same physical inodes relative to the
454 	 * new startino.
455 	 */
456 	*allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
457 }
458 
459 /*
460  * Determine whether the source inode record can merge into the target. Both
461  * records must be sparse, the inode ranges must match and there must be no
462  * allocation overlap between the records.
463  */
464 STATIC bool
465 __xfs_inobt_can_merge(
466 	struct xfs_inobt_rec_incore	*trec,	/* tgt record */
467 	struct xfs_inobt_rec_incore	*srec)	/* src record */
468 {
469 	uint64_t			talloc;
470 	uint64_t			salloc;
471 
472 	/* records must cover the same inode range */
473 	if (trec->ir_startino != srec->ir_startino)
474 		return false;
475 
476 	/* both records must be sparse */
477 	if (!xfs_inobt_issparse(trec->ir_holemask) ||
478 	    !xfs_inobt_issparse(srec->ir_holemask))
479 		return false;
480 
481 	/* both records must track some inodes */
482 	if (!trec->ir_count || !srec->ir_count)
483 		return false;
484 
485 	/* can't exceed capacity of a full record */
486 	if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
487 		return false;
488 
489 	/* verify there is no allocation overlap */
490 	talloc = xfs_inobt_irec_to_allocmask(trec);
491 	salloc = xfs_inobt_irec_to_allocmask(srec);
492 	if (talloc & salloc)
493 		return false;
494 
495 	return true;
496 }
497 
498 /*
499  * Merge the source inode record into the target. The caller must call
500  * __xfs_inobt_can_merge() to ensure the merge is valid.
501  */
502 STATIC void
503 __xfs_inobt_rec_merge(
504 	struct xfs_inobt_rec_incore	*trec,	/* target */
505 	struct xfs_inobt_rec_incore	*srec)	/* src */
506 {
507 	ASSERT(trec->ir_startino == srec->ir_startino);
508 
509 	/* combine the counts */
510 	trec->ir_count += srec->ir_count;
511 	trec->ir_freecount += srec->ir_freecount;
512 
513 	/*
514 	 * Merge the holemask and free mask. For both fields, 0 bits refer to
515 	 * allocated inodes. We combine the allocated ranges with bitwise AND.
516 	 */
517 	trec->ir_holemask &= srec->ir_holemask;
518 	trec->ir_free &= srec->ir_free;
519 }
520 
521 /*
522  * Insert a new sparse inode chunk into the associated inode btree. The inode
523  * record for the sparse chunk is pre-aligned to a startino that should match
524  * any pre-existing sparse inode record in the tree. This allows sparse chunks
525  * to fill over time.
526  *
527  * This function supports two modes of handling preexisting records depending on
528  * the merge flag. If merge is true, the provided record is merged with the
529  * existing record and updated in place. The merged record is returned in nrec.
530  * If merge is false, an existing record is replaced with the provided record.
531  * If no preexisting record exists, the provided record is always inserted.
532  *
533  * It is considered corruption if a merge is requested and not possible. Given
534  * the sparse inode alignment constraints, this should never happen.
535  */
536 STATIC int
537 xfs_inobt_insert_sprec(
538 	struct xfs_mount		*mp,
539 	struct xfs_trans		*tp,
540 	struct xfs_buf			*agbp,
541 	int				btnum,
542 	struct xfs_inobt_rec_incore	*nrec,	/* in/out: new/merged rec. */
543 	bool				merge)	/* merge or replace */
544 {
545 	struct xfs_btree_cur		*cur;
546 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
547 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
548 	int				error;
549 	int				i;
550 	struct xfs_inobt_rec_incore	rec;
551 
552 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
553 
554 	/* the new record is pre-aligned so we know where to look */
555 	error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
556 	if (error)
557 		goto error;
558 	/* if nothing there, insert a new record and return */
559 	if (i == 0) {
560 		error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
561 					     nrec->ir_count, nrec->ir_freecount,
562 					     nrec->ir_free, &i);
563 		if (error)
564 			goto error;
565 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
566 
567 		goto out;
568 	}
569 
570 	/*
571 	 * A record exists at this startino. Merge or replace the record
572 	 * depending on what we've been asked to do.
573 	 */
574 	if (merge) {
575 		error = xfs_inobt_get_rec(cur, &rec, &i);
576 		if (error)
577 			goto error;
578 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
579 		XFS_WANT_CORRUPTED_GOTO(mp,
580 					rec.ir_startino == nrec->ir_startino,
581 					error);
582 
583 		/*
584 		 * This should never fail. If we have coexisting records that
585 		 * cannot merge, something is seriously wrong.
586 		 */
587 		XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
588 					error);
589 
590 		trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
591 					 rec.ir_holemask, nrec->ir_startino,
592 					 nrec->ir_holemask);
593 
594 		/* merge to nrec to output the updated record */
595 		__xfs_inobt_rec_merge(nrec, &rec);
596 
597 		trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
598 					  nrec->ir_holemask);
599 
600 		error = xfs_inobt_rec_check_count(mp, nrec);
601 		if (error)
602 			goto error;
603 	}
604 
605 	error = xfs_inobt_update(cur, nrec);
606 	if (error)
607 		goto error;
608 
609 out:
610 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
611 	return 0;
612 error:
613 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
614 	return error;
615 }
616 
617 /*
618  * Allocate new inodes in the allocation group specified by agbp.
619  * Return 0 for success, else error code.
620  */
621 STATIC int				/* error code or 0 */
622 xfs_ialloc_ag_alloc(
623 	xfs_trans_t	*tp,		/* transaction pointer */
624 	xfs_buf_t	*agbp,		/* alloc group buffer */
625 	int		*alloc)
626 {
627 	xfs_agi_t	*agi;		/* allocation group header */
628 	xfs_alloc_arg_t	args;		/* allocation argument structure */
629 	xfs_agnumber_t	agno;
630 	int		error;
631 	xfs_agino_t	newino;		/* new first inode's number */
632 	xfs_agino_t	newlen;		/* new number of inodes */
633 	int		isaligned = 0;	/* inode allocation at stripe unit */
634 					/* boundary */
635 	uint16_t	allocmask = (uint16_t) -1; /* init. to full chunk */
636 	struct xfs_inobt_rec_incore rec;
637 	struct xfs_perag *pag;
638 	int		do_sparse = 0;
639 
640 	memset(&args, 0, sizeof(args));
641 	args.tp = tp;
642 	args.mp = tp->t_mountp;
643 	args.fsbno = NULLFSBLOCK;
644 	xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES);
645 
646 #ifdef DEBUG
647 	/* randomly do sparse inode allocations */
648 	if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
649 	    args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks)
650 		do_sparse = prandom_u32() & 1;
651 #endif
652 
653 	/*
654 	 * Locking will ensure that we don't have two callers in here
655 	 * at one time.
656 	 */
657 	newlen = args.mp->m_ialloc_inos;
658 	if (args.mp->m_maxicount &&
659 	    percpu_counter_read_positive(&args.mp->m_icount) + newlen >
660 							args.mp->m_maxicount)
661 		return -ENOSPC;
662 	args.minlen = args.maxlen = args.mp->m_ialloc_blks;
663 	/*
664 	 * First try to allocate inodes contiguous with the last-allocated
665 	 * chunk of inodes.  If the filesystem is striped, this will fill
666 	 * an entire stripe unit with inodes.
667 	 */
668 	agi = XFS_BUF_TO_AGI(agbp);
669 	newino = be32_to_cpu(agi->agi_newino);
670 	agno = be32_to_cpu(agi->agi_seqno);
671 	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
672 		     args.mp->m_ialloc_blks;
673 	if (do_sparse)
674 		goto sparse_alloc;
675 	if (likely(newino != NULLAGINO &&
676 		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
677 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
678 		args.type = XFS_ALLOCTYPE_THIS_BNO;
679 		args.prod = 1;
680 
681 		/*
682 		 * We need to take into account alignment here to ensure that
683 		 * we don't modify the free list if we fail to have an exact
684 		 * block. If we don't have an exact match, and every oher
685 		 * attempt allocation attempt fails, we'll end up cancelling
686 		 * a dirty transaction and shutting down.
687 		 *
688 		 * For an exact allocation, alignment must be 1,
689 		 * however we need to take cluster alignment into account when
690 		 * fixing up the freelist. Use the minalignslop field to
691 		 * indicate that extra blocks might be required for alignment,
692 		 * but not to use them in the actual exact allocation.
693 		 */
694 		args.alignment = 1;
695 		args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1;
696 
697 		/* Allow space for the inode btree to split. */
698 		args.minleft = args.mp->m_in_maxlevels - 1;
699 		if ((error = xfs_alloc_vextent(&args)))
700 			return error;
701 
702 		/*
703 		 * This request might have dirtied the transaction if the AG can
704 		 * satisfy the request, but the exact block was not available.
705 		 * If the allocation did fail, subsequent requests will relax
706 		 * the exact agbno requirement and increase the alignment
707 		 * instead. It is critical that the total size of the request
708 		 * (len + alignment + slop) does not increase from this point
709 		 * on, so reset minalignslop to ensure it is not included in
710 		 * subsequent requests.
711 		 */
712 		args.minalignslop = 0;
713 	}
714 
715 	if (unlikely(args.fsbno == NULLFSBLOCK)) {
716 		/*
717 		 * Set the alignment for the allocation.
718 		 * If stripe alignment is turned on then align at stripe unit
719 		 * boundary.
720 		 * If the cluster size is smaller than a filesystem block
721 		 * then we're doing I/O for inodes in filesystem block size
722 		 * pieces, so don't need alignment anyway.
723 		 */
724 		isaligned = 0;
725 		if (args.mp->m_sinoalign) {
726 			ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
727 			args.alignment = args.mp->m_dalign;
728 			isaligned = 1;
729 		} else
730 			args.alignment = xfs_ialloc_cluster_alignment(args.mp);
731 		/*
732 		 * Need to figure out where to allocate the inode blocks.
733 		 * Ideally they should be spaced out through the a.g.
734 		 * For now, just allocate blocks up front.
735 		 */
736 		args.agbno = be32_to_cpu(agi->agi_root);
737 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
738 		/*
739 		 * Allocate a fixed-size extent of inodes.
740 		 */
741 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
742 		args.prod = 1;
743 		/*
744 		 * Allow space for the inode btree to split.
745 		 */
746 		args.minleft = args.mp->m_in_maxlevels - 1;
747 		if ((error = xfs_alloc_vextent(&args)))
748 			return error;
749 	}
750 
751 	/*
752 	 * If stripe alignment is turned on, then try again with cluster
753 	 * alignment.
754 	 */
755 	if (isaligned && args.fsbno == NULLFSBLOCK) {
756 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
757 		args.agbno = be32_to_cpu(agi->agi_root);
758 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
759 		args.alignment = xfs_ialloc_cluster_alignment(args.mp);
760 		if ((error = xfs_alloc_vextent(&args)))
761 			return error;
762 	}
763 
764 	/*
765 	 * Finally, try a sparse allocation if the filesystem supports it and
766 	 * the sparse allocation length is smaller than a full chunk.
767 	 */
768 	if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
769 	    args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks &&
770 	    args.fsbno == NULLFSBLOCK) {
771 sparse_alloc:
772 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
773 		args.agbno = be32_to_cpu(agi->agi_root);
774 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
775 		args.alignment = args.mp->m_sb.sb_spino_align;
776 		args.prod = 1;
777 
778 		args.minlen = args.mp->m_ialloc_min_blks;
779 		args.maxlen = args.minlen;
780 
781 		/*
782 		 * The inode record will be aligned to full chunk size. We must
783 		 * prevent sparse allocation from AG boundaries that result in
784 		 * invalid inode records, such as records that start at agbno 0
785 		 * or extend beyond the AG.
786 		 *
787 		 * Set min agbno to the first aligned, non-zero agbno and max to
788 		 * the last aligned agbno that is at least one full chunk from
789 		 * the end of the AG.
790 		 */
791 		args.min_agbno = args.mp->m_sb.sb_inoalignmt;
792 		args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
793 					    args.mp->m_sb.sb_inoalignmt) -
794 				 args.mp->m_ialloc_blks;
795 
796 		error = xfs_alloc_vextent(&args);
797 		if (error)
798 			return error;
799 
800 		newlen = args.len << args.mp->m_sb.sb_inopblog;
801 		ASSERT(newlen <= XFS_INODES_PER_CHUNK);
802 		allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
803 	}
804 
805 	if (args.fsbno == NULLFSBLOCK) {
806 		*alloc = 0;
807 		return 0;
808 	}
809 	ASSERT(args.len == args.minlen);
810 
811 	/*
812 	 * Stamp and write the inode buffers.
813 	 *
814 	 * Seed the new inode cluster with a random generation number. This
815 	 * prevents short-term reuse of generation numbers if a chunk is
816 	 * freed and then immediately reallocated. We use random numbers
817 	 * rather than a linear progression to prevent the next generation
818 	 * number from being easily guessable.
819 	 */
820 	error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
821 			args.agbno, args.len, prandom_u32());
822 
823 	if (error)
824 		return error;
825 	/*
826 	 * Convert the results.
827 	 */
828 	newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
829 
830 	if (xfs_inobt_issparse(~allocmask)) {
831 		/*
832 		 * We've allocated a sparse chunk. Align the startino and mask.
833 		 */
834 		xfs_align_sparse_ino(args.mp, &newino, &allocmask);
835 
836 		rec.ir_startino = newino;
837 		rec.ir_holemask = ~allocmask;
838 		rec.ir_count = newlen;
839 		rec.ir_freecount = newlen;
840 		rec.ir_free = XFS_INOBT_ALL_FREE;
841 
842 		/*
843 		 * Insert the sparse record into the inobt and allow for a merge
844 		 * if necessary. If a merge does occur, rec is updated to the
845 		 * merged record.
846 		 */
847 		error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
848 					       &rec, true);
849 		if (error == -EFSCORRUPTED) {
850 			xfs_alert(args.mp,
851 	"invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
852 				  XFS_AGINO_TO_INO(args.mp, agno,
853 						   rec.ir_startino),
854 				  rec.ir_holemask, rec.ir_count);
855 			xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
856 		}
857 		if (error)
858 			return error;
859 
860 		/*
861 		 * We can't merge the part we've just allocated as for the inobt
862 		 * due to finobt semantics. The original record may or may not
863 		 * exist independent of whether physical inodes exist in this
864 		 * sparse chunk.
865 		 *
866 		 * We must update the finobt record based on the inobt record.
867 		 * rec contains the fully merged and up to date inobt record
868 		 * from the previous call. Set merge false to replace any
869 		 * existing record with this one.
870 		 */
871 		if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
872 			error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
873 						       XFS_BTNUM_FINO, &rec,
874 						       false);
875 			if (error)
876 				return error;
877 		}
878 	} else {
879 		/* full chunk - insert new records to both btrees */
880 		error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
881 					 XFS_BTNUM_INO);
882 		if (error)
883 			return error;
884 
885 		if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
886 			error = xfs_inobt_insert(args.mp, tp, agbp, newino,
887 						 newlen, XFS_BTNUM_FINO);
888 			if (error)
889 				return error;
890 		}
891 	}
892 
893 	/*
894 	 * Update AGI counts and newino.
895 	 */
896 	be32_add_cpu(&agi->agi_count, newlen);
897 	be32_add_cpu(&agi->agi_freecount, newlen);
898 	pag = xfs_perag_get(args.mp, agno);
899 	pag->pagi_freecount += newlen;
900 	pag->pagi_count += newlen;
901 	xfs_perag_put(pag);
902 	agi->agi_newino = cpu_to_be32(newino);
903 
904 	/*
905 	 * Log allocation group header fields
906 	 */
907 	xfs_ialloc_log_agi(tp, agbp,
908 		XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
909 	/*
910 	 * Modify/log superblock values for inode count and inode free count.
911 	 */
912 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
913 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
914 	*alloc = 1;
915 	return 0;
916 }
917 
918 STATIC xfs_agnumber_t
919 xfs_ialloc_next_ag(
920 	xfs_mount_t	*mp)
921 {
922 	xfs_agnumber_t	agno;
923 
924 	spin_lock(&mp->m_agirotor_lock);
925 	agno = mp->m_agirotor;
926 	if (++mp->m_agirotor >= mp->m_maxagi)
927 		mp->m_agirotor = 0;
928 	spin_unlock(&mp->m_agirotor_lock);
929 
930 	return agno;
931 }
932 
933 /*
934  * Select an allocation group to look for a free inode in, based on the parent
935  * inode and the mode.  Return the allocation group buffer.
936  */
937 STATIC xfs_agnumber_t
938 xfs_ialloc_ag_select(
939 	xfs_trans_t	*tp,		/* transaction pointer */
940 	xfs_ino_t	parent,		/* parent directory inode number */
941 	umode_t		mode)		/* bits set to indicate file type */
942 {
943 	xfs_agnumber_t	agcount;	/* number of ag's in the filesystem */
944 	xfs_agnumber_t	agno;		/* current ag number */
945 	int		flags;		/* alloc buffer locking flags */
946 	xfs_extlen_t	ineed;		/* blocks needed for inode allocation */
947 	xfs_extlen_t	longest = 0;	/* longest extent available */
948 	xfs_mount_t	*mp;		/* mount point structure */
949 	int		needspace;	/* file mode implies space allocated */
950 	xfs_perag_t	*pag;		/* per allocation group data */
951 	xfs_agnumber_t	pagno;		/* parent (starting) ag number */
952 	int		error;
953 
954 	/*
955 	 * Files of these types need at least one block if length > 0
956 	 * (and they won't fit in the inode, but that's hard to figure out).
957 	 */
958 	needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
959 	mp = tp->t_mountp;
960 	agcount = mp->m_maxagi;
961 	if (S_ISDIR(mode))
962 		pagno = xfs_ialloc_next_ag(mp);
963 	else {
964 		pagno = XFS_INO_TO_AGNO(mp, parent);
965 		if (pagno >= agcount)
966 			pagno = 0;
967 	}
968 
969 	ASSERT(pagno < agcount);
970 
971 	/*
972 	 * Loop through allocation groups, looking for one with a little
973 	 * free space in it.  Note we don't look for free inodes, exactly.
974 	 * Instead, we include whether there is a need to allocate inodes
975 	 * to mean that blocks must be allocated for them,
976 	 * if none are currently free.
977 	 */
978 	agno = pagno;
979 	flags = XFS_ALLOC_FLAG_TRYLOCK;
980 	for (;;) {
981 		pag = xfs_perag_get(mp, agno);
982 		if (!pag->pagi_inodeok) {
983 			xfs_ialloc_next_ag(mp);
984 			goto nextag;
985 		}
986 
987 		if (!pag->pagi_init) {
988 			error = xfs_ialloc_pagi_init(mp, tp, agno);
989 			if (error)
990 				goto nextag;
991 		}
992 
993 		if (pag->pagi_freecount) {
994 			xfs_perag_put(pag);
995 			return agno;
996 		}
997 
998 		if (!pag->pagf_init) {
999 			error = xfs_alloc_pagf_init(mp, tp, agno, flags);
1000 			if (error)
1001 				goto nextag;
1002 		}
1003 
1004 		/*
1005 		 * Check that there is enough free space for the file plus a
1006 		 * chunk of inodes if we need to allocate some. If this is the
1007 		 * first pass across the AGs, take into account the potential
1008 		 * space needed for alignment of inode chunks when checking the
1009 		 * longest contiguous free space in the AG - this prevents us
1010 		 * from getting ENOSPC because we have free space larger than
1011 		 * m_ialloc_blks but alignment constraints prevent us from using
1012 		 * it.
1013 		 *
1014 		 * If we can't find an AG with space for full alignment slack to
1015 		 * be taken into account, we must be near ENOSPC in all AGs.
1016 		 * Hence we don't include alignment for the second pass and so
1017 		 * if we fail allocation due to alignment issues then it is most
1018 		 * likely a real ENOSPC condition.
1019 		 */
1020 		ineed = mp->m_ialloc_min_blks;
1021 		if (flags && ineed > 1)
1022 			ineed += xfs_ialloc_cluster_alignment(mp);
1023 		longest = pag->pagf_longest;
1024 		if (!longest)
1025 			longest = pag->pagf_flcount > 0;
1026 
1027 		if (pag->pagf_freeblks >= needspace + ineed &&
1028 		    longest >= ineed) {
1029 			xfs_perag_put(pag);
1030 			return agno;
1031 		}
1032 nextag:
1033 		xfs_perag_put(pag);
1034 		/*
1035 		 * No point in iterating over the rest, if we're shutting
1036 		 * down.
1037 		 */
1038 		if (XFS_FORCED_SHUTDOWN(mp))
1039 			return NULLAGNUMBER;
1040 		agno++;
1041 		if (agno >= agcount)
1042 			agno = 0;
1043 		if (agno == pagno) {
1044 			if (flags == 0)
1045 				return NULLAGNUMBER;
1046 			flags = 0;
1047 		}
1048 	}
1049 }
1050 
1051 /*
1052  * Try to retrieve the next record to the left/right from the current one.
1053  */
1054 STATIC int
1055 xfs_ialloc_next_rec(
1056 	struct xfs_btree_cur	*cur,
1057 	xfs_inobt_rec_incore_t	*rec,
1058 	int			*done,
1059 	int			left)
1060 {
1061 	int                     error;
1062 	int			i;
1063 
1064 	if (left)
1065 		error = xfs_btree_decrement(cur, 0, &i);
1066 	else
1067 		error = xfs_btree_increment(cur, 0, &i);
1068 
1069 	if (error)
1070 		return error;
1071 	*done = !i;
1072 	if (i) {
1073 		error = xfs_inobt_get_rec(cur, rec, &i);
1074 		if (error)
1075 			return error;
1076 		XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1077 	}
1078 
1079 	return 0;
1080 }
1081 
1082 STATIC int
1083 xfs_ialloc_get_rec(
1084 	struct xfs_btree_cur	*cur,
1085 	xfs_agino_t		agino,
1086 	xfs_inobt_rec_incore_t	*rec,
1087 	int			*done)
1088 {
1089 	int                     error;
1090 	int			i;
1091 
1092 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1093 	if (error)
1094 		return error;
1095 	*done = !i;
1096 	if (i) {
1097 		error = xfs_inobt_get_rec(cur, rec, &i);
1098 		if (error)
1099 			return error;
1100 		XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1101 	}
1102 
1103 	return 0;
1104 }
1105 
1106 /*
1107  * Return the offset of the first free inode in the record. If the inode chunk
1108  * is sparsely allocated, we convert the record holemask to inode granularity
1109  * and mask off the unallocated regions from the inode free mask.
1110  */
1111 STATIC int
1112 xfs_inobt_first_free_inode(
1113 	struct xfs_inobt_rec_incore	*rec)
1114 {
1115 	xfs_inofree_t			realfree;
1116 
1117 	/* if there are no holes, return the first available offset */
1118 	if (!xfs_inobt_issparse(rec->ir_holemask))
1119 		return xfs_lowbit64(rec->ir_free);
1120 
1121 	realfree = xfs_inobt_irec_to_allocmask(rec);
1122 	realfree &= rec->ir_free;
1123 
1124 	return xfs_lowbit64(realfree);
1125 }
1126 
1127 /*
1128  * Allocate an inode using the inobt-only algorithm.
1129  */
1130 STATIC int
1131 xfs_dialloc_ag_inobt(
1132 	struct xfs_trans	*tp,
1133 	struct xfs_buf		*agbp,
1134 	xfs_ino_t		parent,
1135 	xfs_ino_t		*inop)
1136 {
1137 	struct xfs_mount	*mp = tp->t_mountp;
1138 	struct xfs_agi		*agi = XFS_BUF_TO_AGI(agbp);
1139 	xfs_agnumber_t		agno = be32_to_cpu(agi->agi_seqno);
1140 	xfs_agnumber_t		pagno = XFS_INO_TO_AGNO(mp, parent);
1141 	xfs_agino_t		pagino = XFS_INO_TO_AGINO(mp, parent);
1142 	struct xfs_perag	*pag;
1143 	struct xfs_btree_cur	*cur, *tcur;
1144 	struct xfs_inobt_rec_incore rec, trec;
1145 	xfs_ino_t		ino;
1146 	int			error;
1147 	int			offset;
1148 	int			i, j;
1149 	int			searchdistance = 10;
1150 
1151 	pag = xfs_perag_get(mp, agno);
1152 
1153 	ASSERT(pag->pagi_init);
1154 	ASSERT(pag->pagi_inodeok);
1155 	ASSERT(pag->pagi_freecount > 0);
1156 
1157  restart_pagno:
1158 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1159 	/*
1160 	 * If pagino is 0 (this is the root inode allocation) use newino.
1161 	 * This must work because we've just allocated some.
1162 	 */
1163 	if (!pagino)
1164 		pagino = be32_to_cpu(agi->agi_newino);
1165 
1166 	error = xfs_check_agi_freecount(cur, agi);
1167 	if (error)
1168 		goto error0;
1169 
1170 	/*
1171 	 * If in the same AG as the parent, try to get near the parent.
1172 	 */
1173 	if (pagno == agno) {
1174 		int		doneleft;	/* done, to the left */
1175 		int		doneright;	/* done, to the right */
1176 
1177 		error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1178 		if (error)
1179 			goto error0;
1180 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1181 
1182 		error = xfs_inobt_get_rec(cur, &rec, &j);
1183 		if (error)
1184 			goto error0;
1185 		XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
1186 
1187 		if (rec.ir_freecount > 0) {
1188 			/*
1189 			 * Found a free inode in the same chunk
1190 			 * as the parent, done.
1191 			 */
1192 			goto alloc_inode;
1193 		}
1194 
1195 
1196 		/*
1197 		 * In the same AG as parent, but parent's chunk is full.
1198 		 */
1199 
1200 		/* duplicate the cursor, search left & right simultaneously */
1201 		error = xfs_btree_dup_cursor(cur, &tcur);
1202 		if (error)
1203 			goto error0;
1204 
1205 		/*
1206 		 * Skip to last blocks looked up if same parent inode.
1207 		 */
1208 		if (pagino != NULLAGINO &&
1209 		    pag->pagl_pagino == pagino &&
1210 		    pag->pagl_leftrec != NULLAGINO &&
1211 		    pag->pagl_rightrec != NULLAGINO) {
1212 			error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1213 						   &trec, &doneleft);
1214 			if (error)
1215 				goto error1;
1216 
1217 			error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1218 						   &rec, &doneright);
1219 			if (error)
1220 				goto error1;
1221 		} else {
1222 			/* search left with tcur, back up 1 record */
1223 			error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1224 			if (error)
1225 				goto error1;
1226 
1227 			/* search right with cur, go forward 1 record. */
1228 			error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1229 			if (error)
1230 				goto error1;
1231 		}
1232 
1233 		/*
1234 		 * Loop until we find an inode chunk with a free inode.
1235 		 */
1236 		while (--searchdistance > 0 && (!doneleft || !doneright)) {
1237 			int	useleft;  /* using left inode chunk this time */
1238 
1239 			/* figure out the closer block if both are valid. */
1240 			if (!doneleft && !doneright) {
1241 				useleft = pagino -
1242 				 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1243 				  rec.ir_startino - pagino;
1244 			} else {
1245 				useleft = !doneleft;
1246 			}
1247 
1248 			/* free inodes to the left? */
1249 			if (useleft && trec.ir_freecount) {
1250 				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1251 				cur = tcur;
1252 
1253 				pag->pagl_leftrec = trec.ir_startino;
1254 				pag->pagl_rightrec = rec.ir_startino;
1255 				pag->pagl_pagino = pagino;
1256 				rec = trec;
1257 				goto alloc_inode;
1258 			}
1259 
1260 			/* free inodes to the right? */
1261 			if (!useleft && rec.ir_freecount) {
1262 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1263 
1264 				pag->pagl_leftrec = trec.ir_startino;
1265 				pag->pagl_rightrec = rec.ir_startino;
1266 				pag->pagl_pagino = pagino;
1267 				goto alloc_inode;
1268 			}
1269 
1270 			/* get next record to check */
1271 			if (useleft) {
1272 				error = xfs_ialloc_next_rec(tcur, &trec,
1273 								 &doneleft, 1);
1274 			} else {
1275 				error = xfs_ialloc_next_rec(cur, &rec,
1276 								 &doneright, 0);
1277 			}
1278 			if (error)
1279 				goto error1;
1280 		}
1281 
1282 		if (searchdistance <= 0) {
1283 			/*
1284 			 * Not in range - save last search
1285 			 * location and allocate a new inode
1286 			 */
1287 			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1288 			pag->pagl_leftrec = trec.ir_startino;
1289 			pag->pagl_rightrec = rec.ir_startino;
1290 			pag->pagl_pagino = pagino;
1291 
1292 		} else {
1293 			/*
1294 			 * We've reached the end of the btree. because
1295 			 * we are only searching a small chunk of the
1296 			 * btree each search, there is obviously free
1297 			 * inodes closer to the parent inode than we
1298 			 * are now. restart the search again.
1299 			 */
1300 			pag->pagl_pagino = NULLAGINO;
1301 			pag->pagl_leftrec = NULLAGINO;
1302 			pag->pagl_rightrec = NULLAGINO;
1303 			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1304 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1305 			goto restart_pagno;
1306 		}
1307 	}
1308 
1309 	/*
1310 	 * In a different AG from the parent.
1311 	 * See if the most recently allocated block has any free.
1312 	 */
1313 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1314 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1315 					 XFS_LOOKUP_EQ, &i);
1316 		if (error)
1317 			goto error0;
1318 
1319 		if (i == 1) {
1320 			error = xfs_inobt_get_rec(cur, &rec, &j);
1321 			if (error)
1322 				goto error0;
1323 
1324 			if (j == 1 && rec.ir_freecount > 0) {
1325 				/*
1326 				 * The last chunk allocated in the group
1327 				 * still has a free inode.
1328 				 */
1329 				goto alloc_inode;
1330 			}
1331 		}
1332 	}
1333 
1334 	/*
1335 	 * None left in the last group, search the whole AG
1336 	 */
1337 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1338 	if (error)
1339 		goto error0;
1340 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1341 
1342 	for (;;) {
1343 		error = xfs_inobt_get_rec(cur, &rec, &i);
1344 		if (error)
1345 			goto error0;
1346 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1347 		if (rec.ir_freecount > 0)
1348 			break;
1349 		error = xfs_btree_increment(cur, 0, &i);
1350 		if (error)
1351 			goto error0;
1352 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1353 	}
1354 
1355 alloc_inode:
1356 	offset = xfs_inobt_first_free_inode(&rec);
1357 	ASSERT(offset >= 0);
1358 	ASSERT(offset < XFS_INODES_PER_CHUNK);
1359 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1360 				   XFS_INODES_PER_CHUNK) == 0);
1361 	ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1362 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1363 	rec.ir_freecount--;
1364 	error = xfs_inobt_update(cur, &rec);
1365 	if (error)
1366 		goto error0;
1367 	be32_add_cpu(&agi->agi_freecount, -1);
1368 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1369 	pag->pagi_freecount--;
1370 
1371 	error = xfs_check_agi_freecount(cur, agi);
1372 	if (error)
1373 		goto error0;
1374 
1375 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1376 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1377 	xfs_perag_put(pag);
1378 	*inop = ino;
1379 	return 0;
1380 error1:
1381 	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1382 error0:
1383 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1384 	xfs_perag_put(pag);
1385 	return error;
1386 }
1387 
1388 /*
1389  * Use the free inode btree to allocate an inode based on distance from the
1390  * parent. Note that the provided cursor may be deleted and replaced.
1391  */
1392 STATIC int
1393 xfs_dialloc_ag_finobt_near(
1394 	xfs_agino_t			pagino,
1395 	struct xfs_btree_cur		**ocur,
1396 	struct xfs_inobt_rec_incore	*rec)
1397 {
1398 	struct xfs_btree_cur		*lcur = *ocur;	/* left search cursor */
1399 	struct xfs_btree_cur		*rcur;	/* right search cursor */
1400 	struct xfs_inobt_rec_incore	rrec;
1401 	int				error;
1402 	int				i, j;
1403 
1404 	error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1405 	if (error)
1406 		return error;
1407 
1408 	if (i == 1) {
1409 		error = xfs_inobt_get_rec(lcur, rec, &i);
1410 		if (error)
1411 			return error;
1412 		XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
1413 
1414 		/*
1415 		 * See if we've landed in the parent inode record. The finobt
1416 		 * only tracks chunks with at least one free inode, so record
1417 		 * existence is enough.
1418 		 */
1419 		if (pagino >= rec->ir_startino &&
1420 		    pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1421 			return 0;
1422 	}
1423 
1424 	error = xfs_btree_dup_cursor(lcur, &rcur);
1425 	if (error)
1426 		return error;
1427 
1428 	error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1429 	if (error)
1430 		goto error_rcur;
1431 	if (j == 1) {
1432 		error = xfs_inobt_get_rec(rcur, &rrec, &j);
1433 		if (error)
1434 			goto error_rcur;
1435 		XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
1436 	}
1437 
1438 	XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
1439 	if (i == 1 && j == 1) {
1440 		/*
1441 		 * Both the left and right records are valid. Choose the closer
1442 		 * inode chunk to the target.
1443 		 */
1444 		if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1445 		    (rrec.ir_startino - pagino)) {
1446 			*rec = rrec;
1447 			xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1448 			*ocur = rcur;
1449 		} else {
1450 			xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1451 		}
1452 	} else if (j == 1) {
1453 		/* only the right record is valid */
1454 		*rec = rrec;
1455 		xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1456 		*ocur = rcur;
1457 	} else if (i == 1) {
1458 		/* only the left record is valid */
1459 		xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1460 	}
1461 
1462 	return 0;
1463 
1464 error_rcur:
1465 	xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1466 	return error;
1467 }
1468 
1469 /*
1470  * Use the free inode btree to find a free inode based on a newino hint. If
1471  * the hint is NULL, find the first free inode in the AG.
1472  */
1473 STATIC int
1474 xfs_dialloc_ag_finobt_newino(
1475 	struct xfs_agi			*agi,
1476 	struct xfs_btree_cur		*cur,
1477 	struct xfs_inobt_rec_incore	*rec)
1478 {
1479 	int error;
1480 	int i;
1481 
1482 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1483 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1484 					 XFS_LOOKUP_EQ, &i);
1485 		if (error)
1486 			return error;
1487 		if (i == 1) {
1488 			error = xfs_inobt_get_rec(cur, rec, &i);
1489 			if (error)
1490 				return error;
1491 			XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1492 			return 0;
1493 		}
1494 	}
1495 
1496 	/*
1497 	 * Find the first inode available in the AG.
1498 	 */
1499 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1500 	if (error)
1501 		return error;
1502 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1503 
1504 	error = xfs_inobt_get_rec(cur, rec, &i);
1505 	if (error)
1506 		return error;
1507 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1508 
1509 	return 0;
1510 }
1511 
1512 /*
1513  * Update the inobt based on a modification made to the finobt. Also ensure that
1514  * the records from both trees are equivalent post-modification.
1515  */
1516 STATIC int
1517 xfs_dialloc_ag_update_inobt(
1518 	struct xfs_btree_cur		*cur,	/* inobt cursor */
1519 	struct xfs_inobt_rec_incore	*frec,	/* finobt record */
1520 	int				offset) /* inode offset */
1521 {
1522 	struct xfs_inobt_rec_incore	rec;
1523 	int				error;
1524 	int				i;
1525 
1526 	error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1527 	if (error)
1528 		return error;
1529 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1530 
1531 	error = xfs_inobt_get_rec(cur, &rec, &i);
1532 	if (error)
1533 		return error;
1534 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1535 	ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1536 				   XFS_INODES_PER_CHUNK) == 0);
1537 
1538 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1539 	rec.ir_freecount--;
1540 
1541 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
1542 				  (rec.ir_freecount == frec->ir_freecount));
1543 
1544 	return xfs_inobt_update(cur, &rec);
1545 }
1546 
1547 /*
1548  * Allocate an inode using the free inode btree, if available. Otherwise, fall
1549  * back to the inobt search algorithm.
1550  *
1551  * The caller selected an AG for us, and made sure that free inodes are
1552  * available.
1553  */
1554 STATIC int
1555 xfs_dialloc_ag(
1556 	struct xfs_trans	*tp,
1557 	struct xfs_buf		*agbp,
1558 	xfs_ino_t		parent,
1559 	xfs_ino_t		*inop)
1560 {
1561 	struct xfs_mount		*mp = tp->t_mountp;
1562 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
1563 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
1564 	xfs_agnumber_t			pagno = XFS_INO_TO_AGNO(mp, parent);
1565 	xfs_agino_t			pagino = XFS_INO_TO_AGINO(mp, parent);
1566 	struct xfs_perag		*pag;
1567 	struct xfs_btree_cur		*cur;	/* finobt cursor */
1568 	struct xfs_btree_cur		*icur;	/* inobt cursor */
1569 	struct xfs_inobt_rec_incore	rec;
1570 	xfs_ino_t			ino;
1571 	int				error;
1572 	int				offset;
1573 	int				i;
1574 
1575 	if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1576 		return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1577 
1578 	pag = xfs_perag_get(mp, agno);
1579 
1580 	/*
1581 	 * If pagino is 0 (this is the root inode allocation) use newino.
1582 	 * This must work because we've just allocated some.
1583 	 */
1584 	if (!pagino)
1585 		pagino = be32_to_cpu(agi->agi_newino);
1586 
1587 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1588 
1589 	error = xfs_check_agi_freecount(cur, agi);
1590 	if (error)
1591 		goto error_cur;
1592 
1593 	/*
1594 	 * The search algorithm depends on whether we're in the same AG as the
1595 	 * parent. If so, find the closest available inode to the parent. If
1596 	 * not, consider the agi hint or find the first free inode in the AG.
1597 	 */
1598 	if (agno == pagno)
1599 		error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1600 	else
1601 		error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1602 	if (error)
1603 		goto error_cur;
1604 
1605 	offset = xfs_inobt_first_free_inode(&rec);
1606 	ASSERT(offset >= 0);
1607 	ASSERT(offset < XFS_INODES_PER_CHUNK);
1608 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1609 				   XFS_INODES_PER_CHUNK) == 0);
1610 	ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1611 
1612 	/*
1613 	 * Modify or remove the finobt record.
1614 	 */
1615 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1616 	rec.ir_freecount--;
1617 	if (rec.ir_freecount)
1618 		error = xfs_inobt_update(cur, &rec);
1619 	else
1620 		error = xfs_btree_delete(cur, &i);
1621 	if (error)
1622 		goto error_cur;
1623 
1624 	/*
1625 	 * The finobt has now been updated appropriately. We haven't updated the
1626 	 * agi and superblock yet, so we can create an inobt cursor and validate
1627 	 * the original freecount. If all is well, make the equivalent update to
1628 	 * the inobt using the finobt record and offset information.
1629 	 */
1630 	icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1631 
1632 	error = xfs_check_agi_freecount(icur, agi);
1633 	if (error)
1634 		goto error_icur;
1635 
1636 	error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1637 	if (error)
1638 		goto error_icur;
1639 
1640 	/*
1641 	 * Both trees have now been updated. We must update the perag and
1642 	 * superblock before we can check the freecount for each btree.
1643 	 */
1644 	be32_add_cpu(&agi->agi_freecount, -1);
1645 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1646 	pag->pagi_freecount--;
1647 
1648 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1649 
1650 	error = xfs_check_agi_freecount(icur, agi);
1651 	if (error)
1652 		goto error_icur;
1653 	error = xfs_check_agi_freecount(cur, agi);
1654 	if (error)
1655 		goto error_icur;
1656 
1657 	xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1658 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1659 	xfs_perag_put(pag);
1660 	*inop = ino;
1661 	return 0;
1662 
1663 error_icur:
1664 	xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1665 error_cur:
1666 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1667 	xfs_perag_put(pag);
1668 	return error;
1669 }
1670 
1671 /*
1672  * Allocate an inode on disk.
1673  *
1674  * Mode is used to tell whether the new inode will need space, and whether it
1675  * is a directory.
1676  *
1677  * This function is designed to be called twice if it has to do an allocation
1678  * to make more free inodes.  On the first call, *IO_agbp should be set to NULL.
1679  * If an inode is available without having to performn an allocation, an inode
1680  * number is returned.  In this case, *IO_agbp is set to NULL.  If an allocation
1681  * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1682  * The caller should then commit the current transaction, allocate a
1683  * new transaction, and call xfs_dialloc() again, passing in the previous value
1684  * of *IO_agbp.  IO_agbp should be held across the transactions. Since the AGI
1685  * buffer is locked across the two calls, the second call is guaranteed to have
1686  * a free inode available.
1687  *
1688  * Once we successfully pick an inode its number is returned and the on-disk
1689  * data structures are updated.  The inode itself is not read in, since doing so
1690  * would break ordering constraints with xfs_reclaim.
1691  */
1692 int
1693 xfs_dialloc(
1694 	struct xfs_trans	*tp,
1695 	xfs_ino_t		parent,
1696 	umode_t			mode,
1697 	struct xfs_buf		**IO_agbp,
1698 	xfs_ino_t		*inop)
1699 {
1700 	struct xfs_mount	*mp = tp->t_mountp;
1701 	struct xfs_buf		*agbp;
1702 	xfs_agnumber_t		agno;
1703 	int			error;
1704 	int			ialloced;
1705 	int			noroom = 0;
1706 	xfs_agnumber_t		start_agno;
1707 	struct xfs_perag	*pag;
1708 	int			okalloc = 1;
1709 
1710 	if (*IO_agbp) {
1711 		/*
1712 		 * If the caller passes in a pointer to the AGI buffer,
1713 		 * continue where we left off before.  In this case, we
1714 		 * know that the allocation group has free inodes.
1715 		 */
1716 		agbp = *IO_agbp;
1717 		goto out_alloc;
1718 	}
1719 
1720 	/*
1721 	 * We do not have an agbp, so select an initial allocation
1722 	 * group for inode allocation.
1723 	 */
1724 	start_agno = xfs_ialloc_ag_select(tp, parent, mode);
1725 	if (start_agno == NULLAGNUMBER) {
1726 		*inop = NULLFSINO;
1727 		return 0;
1728 	}
1729 
1730 	/*
1731 	 * If we have already hit the ceiling of inode blocks then clear
1732 	 * okalloc so we scan all available agi structures for a free
1733 	 * inode.
1734 	 *
1735 	 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1736 	 * which will sacrifice the preciseness but improve the performance.
1737 	 */
1738 	if (mp->m_maxicount &&
1739 	    percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos
1740 							> mp->m_maxicount) {
1741 		noroom = 1;
1742 		okalloc = 0;
1743 	}
1744 
1745 	/*
1746 	 * Loop until we find an allocation group that either has free inodes
1747 	 * or in which we can allocate some inodes.  Iterate through the
1748 	 * allocation groups upward, wrapping at the end.
1749 	 */
1750 	agno = start_agno;
1751 	for (;;) {
1752 		pag = xfs_perag_get(mp, agno);
1753 		if (!pag->pagi_inodeok) {
1754 			xfs_ialloc_next_ag(mp);
1755 			goto nextag;
1756 		}
1757 
1758 		if (!pag->pagi_init) {
1759 			error = xfs_ialloc_pagi_init(mp, tp, agno);
1760 			if (error)
1761 				goto out_error;
1762 		}
1763 
1764 		/*
1765 		 * Do a first racy fast path check if this AG is usable.
1766 		 */
1767 		if (!pag->pagi_freecount && !okalloc)
1768 			goto nextag;
1769 
1770 		/*
1771 		 * Then read in the AGI buffer and recheck with the AGI buffer
1772 		 * lock held.
1773 		 */
1774 		error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1775 		if (error)
1776 			goto out_error;
1777 
1778 		if (pag->pagi_freecount) {
1779 			xfs_perag_put(pag);
1780 			goto out_alloc;
1781 		}
1782 
1783 		if (!okalloc)
1784 			goto nextag_relse_buffer;
1785 
1786 
1787 		error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1788 		if (error) {
1789 			xfs_trans_brelse(tp, agbp);
1790 
1791 			if (error != -ENOSPC)
1792 				goto out_error;
1793 
1794 			xfs_perag_put(pag);
1795 			*inop = NULLFSINO;
1796 			return 0;
1797 		}
1798 
1799 		if (ialloced) {
1800 			/*
1801 			 * We successfully allocated some inodes, return
1802 			 * the current context to the caller so that it
1803 			 * can commit the current transaction and call
1804 			 * us again where we left off.
1805 			 */
1806 			ASSERT(pag->pagi_freecount > 0);
1807 			xfs_perag_put(pag);
1808 
1809 			*IO_agbp = agbp;
1810 			*inop = NULLFSINO;
1811 			return 0;
1812 		}
1813 
1814 nextag_relse_buffer:
1815 		xfs_trans_brelse(tp, agbp);
1816 nextag:
1817 		xfs_perag_put(pag);
1818 		if (++agno == mp->m_sb.sb_agcount)
1819 			agno = 0;
1820 		if (agno == start_agno) {
1821 			*inop = NULLFSINO;
1822 			return noroom ? -ENOSPC : 0;
1823 		}
1824 	}
1825 
1826 out_alloc:
1827 	*IO_agbp = NULL;
1828 	return xfs_dialloc_ag(tp, agbp, parent, inop);
1829 out_error:
1830 	xfs_perag_put(pag);
1831 	return error;
1832 }
1833 
1834 /*
1835  * Free the blocks of an inode chunk. We must consider that the inode chunk
1836  * might be sparse and only free the regions that are allocated as part of the
1837  * chunk.
1838  */
1839 STATIC void
1840 xfs_difree_inode_chunk(
1841 	struct xfs_mount		*mp,
1842 	xfs_agnumber_t			agno,
1843 	struct xfs_inobt_rec_incore	*rec,
1844 	struct xfs_defer_ops		*dfops)
1845 {
1846 	xfs_agblock_t	sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino);
1847 	int		startidx, endidx;
1848 	int		nextbit;
1849 	xfs_agblock_t	agbno;
1850 	int		contigblk;
1851 	struct xfs_owner_info	oinfo;
1852 	DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1853 	xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES);
1854 
1855 	if (!xfs_inobt_issparse(rec->ir_holemask)) {
1856 		/* not sparse, calculate extent info directly */
1857 		xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno),
1858 				  mp->m_ialloc_blks, &oinfo);
1859 		return;
1860 	}
1861 
1862 	/* holemask is only 16-bits (fits in an unsigned long) */
1863 	ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1864 	holemask[0] = rec->ir_holemask;
1865 
1866 	/*
1867 	 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1868 	 * holemask and convert the start/end index of each range to an extent.
1869 	 * We start with the start and end index both pointing at the first 0 in
1870 	 * the mask.
1871 	 */
1872 	startidx = endidx = find_first_zero_bit(holemask,
1873 						XFS_INOBT_HOLEMASK_BITS);
1874 	nextbit = startidx + 1;
1875 	while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1876 		nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1877 					     nextbit);
1878 		/*
1879 		 * If the next zero bit is contiguous, update the end index of
1880 		 * the current range and continue.
1881 		 */
1882 		if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1883 		    nextbit == endidx + 1) {
1884 			endidx = nextbit;
1885 			goto next;
1886 		}
1887 
1888 		/*
1889 		 * nextbit is not contiguous with the current end index. Convert
1890 		 * the current start/end to an extent and add it to the free
1891 		 * list.
1892 		 */
1893 		agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1894 				  mp->m_sb.sb_inopblock;
1895 		contigblk = ((endidx - startidx + 1) *
1896 			     XFS_INODES_PER_HOLEMASK_BIT) /
1897 			    mp->m_sb.sb_inopblock;
1898 
1899 		ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1900 		ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1901 		xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno),
1902 				  contigblk, &oinfo);
1903 
1904 		/* reset range to current bit and carry on... */
1905 		startidx = endidx = nextbit;
1906 
1907 next:
1908 		nextbit++;
1909 	}
1910 }
1911 
1912 STATIC int
1913 xfs_difree_inobt(
1914 	struct xfs_mount		*mp,
1915 	struct xfs_trans		*tp,
1916 	struct xfs_buf			*agbp,
1917 	xfs_agino_t			agino,
1918 	struct xfs_defer_ops		*dfops,
1919 	struct xfs_icluster		*xic,
1920 	struct xfs_inobt_rec_incore	*orec)
1921 {
1922 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
1923 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
1924 	struct xfs_perag		*pag;
1925 	struct xfs_btree_cur		*cur;
1926 	struct xfs_inobt_rec_incore	rec;
1927 	int				ilen;
1928 	int				error;
1929 	int				i;
1930 	int				off;
1931 
1932 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1933 	ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1934 
1935 	/*
1936 	 * Initialize the cursor.
1937 	 */
1938 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1939 
1940 	error = xfs_check_agi_freecount(cur, agi);
1941 	if (error)
1942 		goto error0;
1943 
1944 	/*
1945 	 * Look for the entry describing this inode.
1946 	 */
1947 	if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1948 		xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1949 			__func__, error);
1950 		goto error0;
1951 	}
1952 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1953 	error = xfs_inobt_get_rec(cur, &rec, &i);
1954 	if (error) {
1955 		xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1956 			__func__, error);
1957 		goto error0;
1958 	}
1959 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1960 	/*
1961 	 * Get the offset in the inode chunk.
1962 	 */
1963 	off = agino - rec.ir_startino;
1964 	ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1965 	ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1966 	/*
1967 	 * Mark the inode free & increment the count.
1968 	 */
1969 	rec.ir_free |= XFS_INOBT_MASK(off);
1970 	rec.ir_freecount++;
1971 
1972 	/*
1973 	 * When an inode chunk is free, it becomes eligible for removal. Don't
1974 	 * remove the chunk if the block size is large enough for multiple inode
1975 	 * chunks (that might not be free).
1976 	 */
1977 	if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1978 	    rec.ir_free == XFS_INOBT_ALL_FREE &&
1979 	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1980 		xic->deleted = true;
1981 		xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1982 		xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1983 
1984 		/*
1985 		 * Remove the inode cluster from the AGI B+Tree, adjust the
1986 		 * AGI and Superblock inode counts, and mark the disk space
1987 		 * to be freed when the transaction is committed.
1988 		 */
1989 		ilen = rec.ir_freecount;
1990 		be32_add_cpu(&agi->agi_count, -ilen);
1991 		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1992 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1993 		pag = xfs_perag_get(mp, agno);
1994 		pag->pagi_freecount -= ilen - 1;
1995 		pag->pagi_count -= ilen;
1996 		xfs_perag_put(pag);
1997 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1998 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1999 
2000 		if ((error = xfs_btree_delete(cur, &i))) {
2001 			xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2002 				__func__, error);
2003 			goto error0;
2004 		}
2005 
2006 		xfs_difree_inode_chunk(mp, agno, &rec, dfops);
2007 	} else {
2008 		xic->deleted = false;
2009 
2010 		error = xfs_inobt_update(cur, &rec);
2011 		if (error) {
2012 			xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2013 				__func__, error);
2014 			goto error0;
2015 		}
2016 
2017 		/*
2018 		 * Change the inode free counts and log the ag/sb changes.
2019 		 */
2020 		be32_add_cpu(&agi->agi_freecount, 1);
2021 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2022 		pag = xfs_perag_get(mp, agno);
2023 		pag->pagi_freecount++;
2024 		xfs_perag_put(pag);
2025 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2026 	}
2027 
2028 	error = xfs_check_agi_freecount(cur, agi);
2029 	if (error)
2030 		goto error0;
2031 
2032 	*orec = rec;
2033 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2034 	return 0;
2035 
2036 error0:
2037 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2038 	return error;
2039 }
2040 
2041 /*
2042  * Free an inode in the free inode btree.
2043  */
2044 STATIC int
2045 xfs_difree_finobt(
2046 	struct xfs_mount		*mp,
2047 	struct xfs_trans		*tp,
2048 	struct xfs_buf			*agbp,
2049 	xfs_agino_t			agino,
2050 	struct xfs_inobt_rec_incore	*ibtrec) /* inobt record */
2051 {
2052 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
2053 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
2054 	struct xfs_btree_cur		*cur;
2055 	struct xfs_inobt_rec_incore	rec;
2056 	int				offset = agino - ibtrec->ir_startino;
2057 	int				error;
2058 	int				i;
2059 
2060 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2061 
2062 	error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2063 	if (error)
2064 		goto error;
2065 	if (i == 0) {
2066 		/*
2067 		 * If the record does not exist in the finobt, we must have just
2068 		 * freed an inode in a previously fully allocated chunk. If not,
2069 		 * something is out of sync.
2070 		 */
2071 		XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
2072 
2073 		error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2074 					     ibtrec->ir_count,
2075 					     ibtrec->ir_freecount,
2076 					     ibtrec->ir_free, &i);
2077 		if (error)
2078 			goto error;
2079 		ASSERT(i == 1);
2080 
2081 		goto out;
2082 	}
2083 
2084 	/*
2085 	 * Read and update the existing record. We could just copy the ibtrec
2086 	 * across here, but that would defeat the purpose of having redundant
2087 	 * metadata. By making the modifications independently, we can catch
2088 	 * corruptions that we wouldn't see if we just copied from one record
2089 	 * to another.
2090 	 */
2091 	error = xfs_inobt_get_rec(cur, &rec, &i);
2092 	if (error)
2093 		goto error;
2094 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
2095 
2096 	rec.ir_free |= XFS_INOBT_MASK(offset);
2097 	rec.ir_freecount++;
2098 
2099 	XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
2100 				(rec.ir_freecount == ibtrec->ir_freecount),
2101 				error);
2102 
2103 	/*
2104 	 * The content of inobt records should always match between the inobt
2105 	 * and finobt. The lifecycle of records in the finobt is different from
2106 	 * the inobt in that the finobt only tracks records with at least one
2107 	 * free inode. Hence, if all of the inodes are free and we aren't
2108 	 * keeping inode chunks permanently on disk, remove the record.
2109 	 * Otherwise, update the record with the new information.
2110 	 *
2111 	 * Note that we currently can't free chunks when the block size is large
2112 	 * enough for multiple chunks. Leave the finobt record to remain in sync
2113 	 * with the inobt.
2114 	 */
2115 	if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2116 	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
2117 	    !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2118 		error = xfs_btree_delete(cur, &i);
2119 		if (error)
2120 			goto error;
2121 		ASSERT(i == 1);
2122 	} else {
2123 		error = xfs_inobt_update(cur, &rec);
2124 		if (error)
2125 			goto error;
2126 	}
2127 
2128 out:
2129 	error = xfs_check_agi_freecount(cur, agi);
2130 	if (error)
2131 		goto error;
2132 
2133 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2134 	return 0;
2135 
2136 error:
2137 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2138 	return error;
2139 }
2140 
2141 /*
2142  * Free disk inode.  Carefully avoids touching the incore inode, all
2143  * manipulations incore are the caller's responsibility.
2144  * The on-disk inode is not changed by this operation, only the
2145  * btree (free inode mask) is changed.
2146  */
2147 int
2148 xfs_difree(
2149 	struct xfs_trans	*tp,		/* transaction pointer */
2150 	xfs_ino_t		inode,		/* inode to be freed */
2151 	struct xfs_defer_ops	*dfops,		/* extents to free */
2152 	struct xfs_icluster	*xic)	/* cluster info if deleted */
2153 {
2154 	/* REFERENCED */
2155 	xfs_agblock_t		agbno;	/* block number containing inode */
2156 	struct xfs_buf		*agbp;	/* buffer for allocation group header */
2157 	xfs_agino_t		agino;	/* allocation group inode number */
2158 	xfs_agnumber_t		agno;	/* allocation group number */
2159 	int			error;	/* error return value */
2160 	struct xfs_mount	*mp;	/* mount structure for filesystem */
2161 	struct xfs_inobt_rec_incore rec;/* btree record */
2162 
2163 	mp = tp->t_mountp;
2164 
2165 	/*
2166 	 * Break up inode number into its components.
2167 	 */
2168 	agno = XFS_INO_TO_AGNO(mp, inode);
2169 	if (agno >= mp->m_sb.sb_agcount)  {
2170 		xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2171 			__func__, agno, mp->m_sb.sb_agcount);
2172 		ASSERT(0);
2173 		return -EINVAL;
2174 	}
2175 	agino = XFS_INO_TO_AGINO(mp, inode);
2176 	if (inode != XFS_AGINO_TO_INO(mp, agno, agino))  {
2177 		xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2178 			__func__, (unsigned long long)inode,
2179 			(unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2180 		ASSERT(0);
2181 		return -EINVAL;
2182 	}
2183 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2184 	if (agbno >= mp->m_sb.sb_agblocks)  {
2185 		xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2186 			__func__, agbno, mp->m_sb.sb_agblocks);
2187 		ASSERT(0);
2188 		return -EINVAL;
2189 	}
2190 	/*
2191 	 * Get the allocation group header.
2192 	 */
2193 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2194 	if (error) {
2195 		xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2196 			__func__, error);
2197 		return error;
2198 	}
2199 
2200 	/*
2201 	 * Fix up the inode allocation btree.
2202 	 */
2203 	error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec);
2204 	if (error)
2205 		goto error0;
2206 
2207 	/*
2208 	 * Fix up the free inode btree.
2209 	 */
2210 	if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2211 		error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2212 		if (error)
2213 			goto error0;
2214 	}
2215 
2216 	return 0;
2217 
2218 error0:
2219 	return error;
2220 }
2221 
2222 STATIC int
2223 xfs_imap_lookup(
2224 	struct xfs_mount	*mp,
2225 	struct xfs_trans	*tp,
2226 	xfs_agnumber_t		agno,
2227 	xfs_agino_t		agino,
2228 	xfs_agblock_t		agbno,
2229 	xfs_agblock_t		*chunk_agbno,
2230 	xfs_agblock_t		*offset_agbno,
2231 	int			flags)
2232 {
2233 	struct xfs_inobt_rec_incore rec;
2234 	struct xfs_btree_cur	*cur;
2235 	struct xfs_buf		*agbp;
2236 	int			error;
2237 	int			i;
2238 
2239 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2240 	if (error) {
2241 		xfs_alert(mp,
2242 			"%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2243 			__func__, error, agno);
2244 		return error;
2245 	}
2246 
2247 	/*
2248 	 * Lookup the inode record for the given agino. If the record cannot be
2249 	 * found, then it's an invalid inode number and we should abort. Once
2250 	 * we have a record, we need to ensure it contains the inode number
2251 	 * we are looking up.
2252 	 */
2253 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
2254 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2255 	if (!error) {
2256 		if (i)
2257 			error = xfs_inobt_get_rec(cur, &rec, &i);
2258 		if (!error && i == 0)
2259 			error = -EINVAL;
2260 	}
2261 
2262 	xfs_trans_brelse(tp, agbp);
2263 	xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
2264 	if (error)
2265 		return error;
2266 
2267 	/* check that the returned record contains the required inode */
2268 	if (rec.ir_startino > agino ||
2269 	    rec.ir_startino + mp->m_ialloc_inos <= agino)
2270 		return -EINVAL;
2271 
2272 	/* for untrusted inodes check it is allocated first */
2273 	if ((flags & XFS_IGET_UNTRUSTED) &&
2274 	    (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2275 		return -EINVAL;
2276 
2277 	*chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2278 	*offset_agbno = agbno - *chunk_agbno;
2279 	return 0;
2280 }
2281 
2282 /*
2283  * Return the location of the inode in imap, for mapping it into a buffer.
2284  */
2285 int
2286 xfs_imap(
2287 	xfs_mount_t	 *mp,	/* file system mount structure */
2288 	xfs_trans_t	 *tp,	/* transaction pointer */
2289 	xfs_ino_t	ino,	/* inode to locate */
2290 	struct xfs_imap	*imap,	/* location map structure */
2291 	uint		flags)	/* flags for inode btree lookup */
2292 {
2293 	xfs_agblock_t	agbno;	/* block number of inode in the alloc group */
2294 	xfs_agino_t	agino;	/* inode number within alloc group */
2295 	xfs_agnumber_t	agno;	/* allocation group number */
2296 	int		blks_per_cluster; /* num blocks per inode cluster */
2297 	xfs_agblock_t	chunk_agbno;	/* first block in inode chunk */
2298 	xfs_agblock_t	cluster_agbno;	/* first block in inode cluster */
2299 	int		error;	/* error code */
2300 	int		offset;	/* index of inode in its buffer */
2301 	xfs_agblock_t	offset_agbno;	/* blks from chunk start to inode */
2302 
2303 	ASSERT(ino != NULLFSINO);
2304 
2305 	/*
2306 	 * Split up the inode number into its parts.
2307 	 */
2308 	agno = XFS_INO_TO_AGNO(mp, ino);
2309 	agino = XFS_INO_TO_AGINO(mp, ino);
2310 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2311 	if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2312 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2313 #ifdef DEBUG
2314 		/*
2315 		 * Don't output diagnostic information for untrusted inodes
2316 		 * as they can be invalid without implying corruption.
2317 		 */
2318 		if (flags & XFS_IGET_UNTRUSTED)
2319 			return -EINVAL;
2320 		if (agno >= mp->m_sb.sb_agcount) {
2321 			xfs_alert(mp,
2322 				"%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2323 				__func__, agno, mp->m_sb.sb_agcount);
2324 		}
2325 		if (agbno >= mp->m_sb.sb_agblocks) {
2326 			xfs_alert(mp,
2327 		"%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2328 				__func__, (unsigned long long)agbno,
2329 				(unsigned long)mp->m_sb.sb_agblocks);
2330 		}
2331 		if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2332 			xfs_alert(mp,
2333 		"%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2334 				__func__, ino,
2335 				XFS_AGINO_TO_INO(mp, agno, agino));
2336 		}
2337 		xfs_stack_trace();
2338 #endif /* DEBUG */
2339 		return -EINVAL;
2340 	}
2341 
2342 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2343 
2344 	/*
2345 	 * For bulkstat and handle lookups, we have an untrusted inode number
2346 	 * that we have to verify is valid. We cannot do this just by reading
2347 	 * the inode buffer as it may have been unlinked and removed leaving
2348 	 * inodes in stale state on disk. Hence we have to do a btree lookup
2349 	 * in all cases where an untrusted inode number is passed.
2350 	 */
2351 	if (flags & XFS_IGET_UNTRUSTED) {
2352 		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2353 					&chunk_agbno, &offset_agbno, flags);
2354 		if (error)
2355 			return error;
2356 		goto out_map;
2357 	}
2358 
2359 	/*
2360 	 * If the inode cluster size is the same as the blocksize or
2361 	 * smaller we get to the buffer by simple arithmetics.
2362 	 */
2363 	if (blks_per_cluster == 1) {
2364 		offset = XFS_INO_TO_OFFSET(mp, ino);
2365 		ASSERT(offset < mp->m_sb.sb_inopblock);
2366 
2367 		imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2368 		imap->im_len = XFS_FSB_TO_BB(mp, 1);
2369 		imap->im_boffset = (unsigned short)(offset <<
2370 							mp->m_sb.sb_inodelog);
2371 		return 0;
2372 	}
2373 
2374 	/*
2375 	 * If the inode chunks are aligned then use simple maths to
2376 	 * find the location. Otherwise we have to do a btree
2377 	 * lookup to find the location.
2378 	 */
2379 	if (mp->m_inoalign_mask) {
2380 		offset_agbno = agbno & mp->m_inoalign_mask;
2381 		chunk_agbno = agbno - offset_agbno;
2382 	} else {
2383 		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2384 					&chunk_agbno, &offset_agbno, flags);
2385 		if (error)
2386 			return error;
2387 	}
2388 
2389 out_map:
2390 	ASSERT(agbno >= chunk_agbno);
2391 	cluster_agbno = chunk_agbno +
2392 		((offset_agbno / blks_per_cluster) * blks_per_cluster);
2393 	offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2394 		XFS_INO_TO_OFFSET(mp, ino);
2395 
2396 	imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2397 	imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
2398 	imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2399 
2400 	/*
2401 	 * If the inode number maps to a block outside the bounds
2402 	 * of the file system then return NULL rather than calling
2403 	 * read_buf and panicing when we get an error from the
2404 	 * driver.
2405 	 */
2406 	if ((imap->im_blkno + imap->im_len) >
2407 	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2408 		xfs_alert(mp,
2409 	"%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2410 			__func__, (unsigned long long) imap->im_blkno,
2411 			(unsigned long long) imap->im_len,
2412 			XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2413 		return -EINVAL;
2414 	}
2415 	return 0;
2416 }
2417 
2418 /*
2419  * Compute and fill in value of m_in_maxlevels.
2420  */
2421 void
2422 xfs_ialloc_compute_maxlevels(
2423 	xfs_mount_t	*mp)		/* file system mount structure */
2424 {
2425 	uint		inodes;
2426 
2427 	inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2428 	mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp->m_inobt_mnr,
2429 							 inodes);
2430 }
2431 
2432 /*
2433  * Log specified fields for the ag hdr (inode section). The growth of the agi
2434  * structure over time requires that we interpret the buffer as two logical
2435  * regions delineated by the end of the unlinked list. This is due to the size
2436  * of the hash table and its location in the middle of the agi.
2437  *
2438  * For example, a request to log a field before agi_unlinked and a field after
2439  * agi_unlinked could cause us to log the entire hash table and use an excessive
2440  * amount of log space. To avoid this behavior, log the region up through
2441  * agi_unlinked in one call and the region after agi_unlinked through the end of
2442  * the structure in another.
2443  */
2444 void
2445 xfs_ialloc_log_agi(
2446 	xfs_trans_t	*tp,		/* transaction pointer */
2447 	xfs_buf_t	*bp,		/* allocation group header buffer */
2448 	int		fields)		/* bitmask of fields to log */
2449 {
2450 	int			first;		/* first byte number */
2451 	int			last;		/* last byte number */
2452 	static const short	offsets[] = {	/* field starting offsets */
2453 					/* keep in sync with bit definitions */
2454 		offsetof(xfs_agi_t, agi_magicnum),
2455 		offsetof(xfs_agi_t, agi_versionnum),
2456 		offsetof(xfs_agi_t, agi_seqno),
2457 		offsetof(xfs_agi_t, agi_length),
2458 		offsetof(xfs_agi_t, agi_count),
2459 		offsetof(xfs_agi_t, agi_root),
2460 		offsetof(xfs_agi_t, agi_level),
2461 		offsetof(xfs_agi_t, agi_freecount),
2462 		offsetof(xfs_agi_t, agi_newino),
2463 		offsetof(xfs_agi_t, agi_dirino),
2464 		offsetof(xfs_agi_t, agi_unlinked),
2465 		offsetof(xfs_agi_t, agi_free_root),
2466 		offsetof(xfs_agi_t, agi_free_level),
2467 		sizeof(xfs_agi_t)
2468 	};
2469 #ifdef DEBUG
2470 	xfs_agi_t		*agi;	/* allocation group header */
2471 
2472 	agi = XFS_BUF_TO_AGI(bp);
2473 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2474 #endif
2475 
2476 	/*
2477 	 * Compute byte offsets for the first and last fields in the first
2478 	 * region and log the agi buffer. This only logs up through
2479 	 * agi_unlinked.
2480 	 */
2481 	if (fields & XFS_AGI_ALL_BITS_R1) {
2482 		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2483 				  &first, &last);
2484 		xfs_trans_log_buf(tp, bp, first, last);
2485 	}
2486 
2487 	/*
2488 	 * Mask off the bits in the first region and calculate the first and
2489 	 * last field offsets for any bits in the second region.
2490 	 */
2491 	fields &= ~XFS_AGI_ALL_BITS_R1;
2492 	if (fields) {
2493 		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2494 				  &first, &last);
2495 		xfs_trans_log_buf(tp, bp, first, last);
2496 	}
2497 }
2498 
2499 static xfs_failaddr_t
2500 xfs_agi_verify(
2501 	struct xfs_buf	*bp)
2502 {
2503 	struct xfs_mount *mp = bp->b_target->bt_mount;
2504 	struct xfs_agi	*agi = XFS_BUF_TO_AGI(bp);
2505 	int		i;
2506 
2507 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
2508 		if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2509 			return __this_address;
2510 		if (!xfs_log_check_lsn(mp,
2511 				be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn)))
2512 			return __this_address;
2513 	}
2514 
2515 	/*
2516 	 * Validate the magic number of the agi block.
2517 	 */
2518 	if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2519 		return __this_address;
2520 	if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2521 		return __this_address;
2522 
2523 	if (be32_to_cpu(agi->agi_level) < 1 ||
2524 	    be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2525 		return __this_address;
2526 
2527 	if (xfs_sb_version_hasfinobt(&mp->m_sb) &&
2528 	    (be32_to_cpu(agi->agi_free_level) < 1 ||
2529 	     be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS))
2530 		return __this_address;
2531 
2532 	/*
2533 	 * during growfs operations, the perag is not fully initialised,
2534 	 * so we can't use it for any useful checking. growfs ensures we can't
2535 	 * use it by using uncached buffers that don't have the perag attached
2536 	 * so we can detect and avoid this problem.
2537 	 */
2538 	if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2539 		return __this_address;
2540 
2541 	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2542 		if (agi->agi_unlinked[i] == NULLAGINO)
2543 			continue;
2544 		if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2545 			return __this_address;
2546 	}
2547 
2548 	return NULL;
2549 }
2550 
2551 static void
2552 xfs_agi_read_verify(
2553 	struct xfs_buf	*bp)
2554 {
2555 	struct xfs_mount *mp = bp->b_target->bt_mount;
2556 	xfs_failaddr_t	fa;
2557 
2558 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
2559 	    !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2560 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2561 	else {
2562 		fa = xfs_agi_verify(bp);
2563 		if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2564 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2565 	}
2566 }
2567 
2568 static void
2569 xfs_agi_write_verify(
2570 	struct xfs_buf	*bp)
2571 {
2572 	struct xfs_mount	*mp = bp->b_target->bt_mount;
2573 	struct xfs_buf_log_item	*bip = bp->b_log_item;
2574 	xfs_failaddr_t		fa;
2575 
2576 	fa = xfs_agi_verify(bp);
2577 	if (fa) {
2578 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2579 		return;
2580 	}
2581 
2582 	if (!xfs_sb_version_hascrc(&mp->m_sb))
2583 		return;
2584 
2585 	if (bip)
2586 		XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2587 	xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2588 }
2589 
2590 const struct xfs_buf_ops xfs_agi_buf_ops = {
2591 	.name = "xfs_agi",
2592 	.verify_read = xfs_agi_read_verify,
2593 	.verify_write = xfs_agi_write_verify,
2594 	.verify_struct = xfs_agi_verify,
2595 };
2596 
2597 /*
2598  * Read in the allocation group header (inode allocation section)
2599  */
2600 int
2601 xfs_read_agi(
2602 	struct xfs_mount	*mp,	/* file system mount structure */
2603 	struct xfs_trans	*tp,	/* transaction pointer */
2604 	xfs_agnumber_t		agno,	/* allocation group number */
2605 	struct xfs_buf		**bpp)	/* allocation group hdr buf */
2606 {
2607 	int			error;
2608 
2609 	trace_xfs_read_agi(mp, agno);
2610 
2611 	ASSERT(agno != NULLAGNUMBER);
2612 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2613 			XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2614 			XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2615 	if (error)
2616 		return error;
2617 	if (tp)
2618 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF);
2619 
2620 	xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2621 	return 0;
2622 }
2623 
2624 int
2625 xfs_ialloc_read_agi(
2626 	struct xfs_mount	*mp,	/* file system mount structure */
2627 	struct xfs_trans	*tp,	/* transaction pointer */
2628 	xfs_agnumber_t		agno,	/* allocation group number */
2629 	struct xfs_buf		**bpp)	/* allocation group hdr buf */
2630 {
2631 	struct xfs_agi		*agi;	/* allocation group header */
2632 	struct xfs_perag	*pag;	/* per allocation group data */
2633 	int			error;
2634 
2635 	trace_xfs_ialloc_read_agi(mp, agno);
2636 
2637 	error = xfs_read_agi(mp, tp, agno, bpp);
2638 	if (error)
2639 		return error;
2640 
2641 	agi = XFS_BUF_TO_AGI(*bpp);
2642 	pag = xfs_perag_get(mp, agno);
2643 	if (!pag->pagi_init) {
2644 		pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2645 		pag->pagi_count = be32_to_cpu(agi->agi_count);
2646 		pag->pagi_init = 1;
2647 	}
2648 
2649 	/*
2650 	 * It's possible for these to be out of sync if
2651 	 * we are in the middle of a forced shutdown.
2652 	 */
2653 	ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2654 		XFS_FORCED_SHUTDOWN(mp));
2655 	xfs_perag_put(pag);
2656 	return 0;
2657 }
2658 
2659 /*
2660  * Read in the agi to initialise the per-ag data in the mount structure
2661  */
2662 int
2663 xfs_ialloc_pagi_init(
2664 	xfs_mount_t	*mp,		/* file system mount structure */
2665 	xfs_trans_t	*tp,		/* transaction pointer */
2666 	xfs_agnumber_t	agno)		/* allocation group number */
2667 {
2668 	xfs_buf_t	*bp = NULL;
2669 	int		error;
2670 
2671 	error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2672 	if (error)
2673 		return error;
2674 	if (bp)
2675 		xfs_trans_brelse(tp, bp);
2676 	return 0;
2677 }
2678 
2679 /* Is there an inode record covering a given range of inode numbers? */
2680 int
2681 xfs_ialloc_has_inode_record(
2682 	struct xfs_btree_cur	*cur,
2683 	xfs_agino_t		low,
2684 	xfs_agino_t		high,
2685 	bool			*exists)
2686 {
2687 	struct xfs_inobt_rec_incore	irec;
2688 	xfs_agino_t		agino;
2689 	uint16_t		holemask;
2690 	int			has_record;
2691 	int			i;
2692 	int			error;
2693 
2694 	*exists = false;
2695 	error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2696 	while (error == 0 && has_record) {
2697 		error = xfs_inobt_get_rec(cur, &irec, &has_record);
2698 		if (error || irec.ir_startino > high)
2699 			break;
2700 
2701 		agino = irec.ir_startino;
2702 		holemask = irec.ir_holemask;
2703 		for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1,
2704 				i++, agino += XFS_INODES_PER_HOLEMASK_BIT) {
2705 			if (holemask & 1)
2706 				continue;
2707 			if (agino + XFS_INODES_PER_HOLEMASK_BIT > low &&
2708 					agino <= high) {
2709 				*exists = true;
2710 				return 0;
2711 			}
2712 		}
2713 
2714 		error = xfs_btree_increment(cur, 0, &has_record);
2715 	}
2716 	return error;
2717 }
2718 
2719 /* Is there an inode record covering a given extent? */
2720 int
2721 xfs_ialloc_has_inodes_at_extent(
2722 	struct xfs_btree_cur	*cur,
2723 	xfs_agblock_t		bno,
2724 	xfs_extlen_t		len,
2725 	bool			*exists)
2726 {
2727 	xfs_agino_t		low;
2728 	xfs_agino_t		high;
2729 
2730 	low = XFS_OFFBNO_TO_AGINO(cur->bc_mp, bno, 0);
2731 	high = XFS_OFFBNO_TO_AGINO(cur->bc_mp, bno + len, 0) - 1;
2732 
2733 	return xfs_ialloc_has_inode_record(cur, low, high, exists);
2734 }
2735 
2736 struct xfs_ialloc_count_inodes {
2737 	xfs_agino_t			count;
2738 	xfs_agino_t			freecount;
2739 };
2740 
2741 /* Record inode counts across all inobt records. */
2742 STATIC int
2743 xfs_ialloc_count_inodes_rec(
2744 	struct xfs_btree_cur		*cur,
2745 	union xfs_btree_rec		*rec,
2746 	void				*priv)
2747 {
2748 	struct xfs_inobt_rec_incore	irec;
2749 	struct xfs_ialloc_count_inodes	*ci = priv;
2750 
2751 	xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2752 	ci->count += irec.ir_count;
2753 	ci->freecount += irec.ir_freecount;
2754 
2755 	return 0;
2756 }
2757 
2758 /* Count allocated and free inodes under an inobt. */
2759 int
2760 xfs_ialloc_count_inodes(
2761 	struct xfs_btree_cur		*cur,
2762 	xfs_agino_t			*count,
2763 	xfs_agino_t			*freecount)
2764 {
2765 	struct xfs_ialloc_count_inodes	ci = {0};
2766 	int				error;
2767 
2768 	ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2769 	error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2770 	if (error)
2771 		return error;
2772 
2773 	*count = ci.count;
2774 	*freecount = ci.freecount;
2775 	return 0;
2776 }
2777