xref: /openbmc/linux/fs/xfs/libxfs/xfs_ialloc.c (revision fb960bd2)
1 /*
2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_ialloc.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_alloc.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_errortag.h"
35 #include "xfs_error.h"
36 #include "xfs_bmap.h"
37 #include "xfs_cksum.h"
38 #include "xfs_trans.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_icreate_item.h"
41 #include "xfs_icache.h"
42 #include "xfs_trace.h"
43 #include "xfs_log.h"
44 #include "xfs_rmap.h"
45 
46 
47 /*
48  * Allocation group level functions.
49  */
50 int
51 xfs_ialloc_cluster_alignment(
52 	struct xfs_mount	*mp)
53 {
54 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
55 	    mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
56 		return mp->m_sb.sb_inoalignmt;
57 	return 1;
58 }
59 
60 /*
61  * Lookup a record by ino in the btree given by cur.
62  */
63 int					/* error */
64 xfs_inobt_lookup(
65 	struct xfs_btree_cur	*cur,	/* btree cursor */
66 	xfs_agino_t		ino,	/* starting inode of chunk */
67 	xfs_lookup_t		dir,	/* <=, >=, == */
68 	int			*stat)	/* success/failure */
69 {
70 	cur->bc_rec.i.ir_startino = ino;
71 	cur->bc_rec.i.ir_holemask = 0;
72 	cur->bc_rec.i.ir_count = 0;
73 	cur->bc_rec.i.ir_freecount = 0;
74 	cur->bc_rec.i.ir_free = 0;
75 	return xfs_btree_lookup(cur, dir, stat);
76 }
77 
78 /*
79  * Update the record referred to by cur to the value given.
80  * This either works (return 0) or gets an EFSCORRUPTED error.
81  */
82 STATIC int				/* error */
83 xfs_inobt_update(
84 	struct xfs_btree_cur	*cur,	/* btree cursor */
85 	xfs_inobt_rec_incore_t	*irec)	/* btree record */
86 {
87 	union xfs_btree_rec	rec;
88 
89 	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
90 	if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
91 		rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
92 		rec.inobt.ir_u.sp.ir_count = irec->ir_count;
93 		rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
94 	} else {
95 		/* ir_holemask/ir_count not supported on-disk */
96 		rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
97 	}
98 	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
99 	return xfs_btree_update(cur, &rec);
100 }
101 
102 /* Convert on-disk btree record to incore inobt record. */
103 void
104 xfs_inobt_btrec_to_irec(
105 	struct xfs_mount		*mp,
106 	union xfs_btree_rec		*rec,
107 	struct xfs_inobt_rec_incore	*irec)
108 {
109 	irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
110 	if (xfs_sb_version_hassparseinodes(&mp->m_sb)) {
111 		irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
112 		irec->ir_count = rec->inobt.ir_u.sp.ir_count;
113 		irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
114 	} else {
115 		/*
116 		 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
117 		 * values for full inode chunks.
118 		 */
119 		irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
120 		irec->ir_count = XFS_INODES_PER_CHUNK;
121 		irec->ir_freecount =
122 				be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
123 	}
124 	irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
125 }
126 
127 /*
128  * Get the data from the pointed-to record.
129  */
130 int
131 xfs_inobt_get_rec(
132 	struct xfs_btree_cur		*cur,
133 	struct xfs_inobt_rec_incore	*irec,
134 	int				*stat)
135 {
136 	union xfs_btree_rec		*rec;
137 	int				error;
138 
139 	error = xfs_btree_get_rec(cur, &rec, stat);
140 	if (error || *stat == 0)
141 		return error;
142 
143 	xfs_inobt_btrec_to_irec(cur->bc_mp, rec, irec);
144 
145 	return 0;
146 }
147 
148 /*
149  * Insert a single inobt record. Cursor must already point to desired location.
150  */
151 STATIC int
152 xfs_inobt_insert_rec(
153 	struct xfs_btree_cur	*cur,
154 	uint16_t		holemask,
155 	uint8_t			count,
156 	int32_t			freecount,
157 	xfs_inofree_t		free,
158 	int			*stat)
159 {
160 	cur->bc_rec.i.ir_holemask = holemask;
161 	cur->bc_rec.i.ir_count = count;
162 	cur->bc_rec.i.ir_freecount = freecount;
163 	cur->bc_rec.i.ir_free = free;
164 	return xfs_btree_insert(cur, stat);
165 }
166 
167 /*
168  * Insert records describing a newly allocated inode chunk into the inobt.
169  */
170 STATIC int
171 xfs_inobt_insert(
172 	struct xfs_mount	*mp,
173 	struct xfs_trans	*tp,
174 	struct xfs_buf		*agbp,
175 	xfs_agino_t		newino,
176 	xfs_agino_t		newlen,
177 	xfs_btnum_t		btnum)
178 {
179 	struct xfs_btree_cur	*cur;
180 	struct xfs_agi		*agi = XFS_BUF_TO_AGI(agbp);
181 	xfs_agnumber_t		agno = be32_to_cpu(agi->agi_seqno);
182 	xfs_agino_t		thisino;
183 	int			i;
184 	int			error;
185 
186 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
187 
188 	for (thisino = newino;
189 	     thisino < newino + newlen;
190 	     thisino += XFS_INODES_PER_CHUNK) {
191 		error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
192 		if (error) {
193 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
194 			return error;
195 		}
196 		ASSERT(i == 0);
197 
198 		error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
199 					     XFS_INODES_PER_CHUNK,
200 					     XFS_INODES_PER_CHUNK,
201 					     XFS_INOBT_ALL_FREE, &i);
202 		if (error) {
203 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
204 			return error;
205 		}
206 		ASSERT(i == 1);
207 	}
208 
209 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
210 
211 	return 0;
212 }
213 
214 /*
215  * Verify that the number of free inodes in the AGI is correct.
216  */
217 #ifdef DEBUG
218 STATIC int
219 xfs_check_agi_freecount(
220 	struct xfs_btree_cur	*cur,
221 	struct xfs_agi		*agi)
222 {
223 	if (cur->bc_nlevels == 1) {
224 		xfs_inobt_rec_incore_t rec;
225 		int		freecount = 0;
226 		int		error;
227 		int		i;
228 
229 		error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
230 		if (error)
231 			return error;
232 
233 		do {
234 			error = xfs_inobt_get_rec(cur, &rec, &i);
235 			if (error)
236 				return error;
237 
238 			if (i) {
239 				freecount += rec.ir_freecount;
240 				error = xfs_btree_increment(cur, 0, &i);
241 				if (error)
242 					return error;
243 			}
244 		} while (i == 1);
245 
246 		if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
247 			ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
248 	}
249 	return 0;
250 }
251 #else
252 #define xfs_check_agi_freecount(cur, agi)	0
253 #endif
254 
255 /*
256  * Initialise a new set of inodes. When called without a transaction context
257  * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
258  * than logging them (which in a transaction context puts them into the AIL
259  * for writeback rather than the xfsbufd queue).
260  */
261 int
262 xfs_ialloc_inode_init(
263 	struct xfs_mount	*mp,
264 	struct xfs_trans	*tp,
265 	struct list_head	*buffer_list,
266 	int			icount,
267 	xfs_agnumber_t		agno,
268 	xfs_agblock_t		agbno,
269 	xfs_agblock_t		length,
270 	unsigned int		gen)
271 {
272 	struct xfs_buf		*fbuf;
273 	struct xfs_dinode	*free;
274 	int			nbufs, blks_per_cluster, inodes_per_cluster;
275 	int			version;
276 	int			i, j;
277 	xfs_daddr_t		d;
278 	xfs_ino_t		ino = 0;
279 
280 	/*
281 	 * Loop over the new block(s), filling in the inodes.  For small block
282 	 * sizes, manipulate the inodes in buffers  which are multiples of the
283 	 * blocks size.
284 	 */
285 	blks_per_cluster = xfs_icluster_size_fsb(mp);
286 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
287 	nbufs = length / blks_per_cluster;
288 
289 	/*
290 	 * Figure out what version number to use in the inodes we create.  If
291 	 * the superblock version has caught up to the one that supports the new
292 	 * inode format, then use the new inode version.  Otherwise use the old
293 	 * version so that old kernels will continue to be able to use the file
294 	 * system.
295 	 *
296 	 * For v3 inodes, we also need to write the inode number into the inode,
297 	 * so calculate the first inode number of the chunk here as
298 	 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
299 	 * across multiple filesystem blocks (such as a cluster) and so cannot
300 	 * be used in the cluster buffer loop below.
301 	 *
302 	 * Further, because we are writing the inode directly into the buffer
303 	 * and calculating a CRC on the entire inode, we have ot log the entire
304 	 * inode so that the entire range the CRC covers is present in the log.
305 	 * That means for v3 inode we log the entire buffer rather than just the
306 	 * inode cores.
307 	 */
308 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
309 		version = 3;
310 		ino = XFS_AGINO_TO_INO(mp, agno,
311 				       XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
312 
313 		/*
314 		 * log the initialisation that is about to take place as an
315 		 * logical operation. This means the transaction does not
316 		 * need to log the physical changes to the inode buffers as log
317 		 * recovery will know what initialisation is actually needed.
318 		 * Hence we only need to log the buffers as "ordered" buffers so
319 		 * they track in the AIL as if they were physically logged.
320 		 */
321 		if (tp)
322 			xfs_icreate_log(tp, agno, agbno, icount,
323 					mp->m_sb.sb_inodesize, length, gen);
324 	} else
325 		version = 2;
326 
327 	for (j = 0; j < nbufs; j++) {
328 		/*
329 		 * Get the block.
330 		 */
331 		d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
332 		fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
333 					 mp->m_bsize * blks_per_cluster,
334 					 XBF_UNMAPPED);
335 		if (!fbuf)
336 			return -ENOMEM;
337 
338 		/* Initialize the inode buffers and log them appropriately. */
339 		fbuf->b_ops = &xfs_inode_buf_ops;
340 		xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
341 		for (i = 0; i < inodes_per_cluster; i++) {
342 			int	ioffset = i << mp->m_sb.sb_inodelog;
343 			uint	isize = xfs_dinode_size(version);
344 
345 			free = xfs_make_iptr(mp, fbuf, i);
346 			free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
347 			free->di_version = version;
348 			free->di_gen = cpu_to_be32(gen);
349 			free->di_next_unlinked = cpu_to_be32(NULLAGINO);
350 
351 			if (version == 3) {
352 				free->di_ino = cpu_to_be64(ino);
353 				ino++;
354 				uuid_copy(&free->di_uuid,
355 					  &mp->m_sb.sb_meta_uuid);
356 				xfs_dinode_calc_crc(mp, free);
357 			} else if (tp) {
358 				/* just log the inode core */
359 				xfs_trans_log_buf(tp, fbuf, ioffset,
360 						  ioffset + isize - 1);
361 			}
362 		}
363 
364 		if (tp) {
365 			/*
366 			 * Mark the buffer as an inode allocation buffer so it
367 			 * sticks in AIL at the point of this allocation
368 			 * transaction. This ensures the they are on disk before
369 			 * the tail of the log can be moved past this
370 			 * transaction (i.e. by preventing relogging from moving
371 			 * it forward in the log).
372 			 */
373 			xfs_trans_inode_alloc_buf(tp, fbuf);
374 			if (version == 3) {
375 				/*
376 				 * Mark the buffer as ordered so that they are
377 				 * not physically logged in the transaction but
378 				 * still tracked in the AIL as part of the
379 				 * transaction and pin the log appropriately.
380 				 */
381 				xfs_trans_ordered_buf(tp, fbuf);
382 			}
383 		} else {
384 			fbuf->b_flags |= XBF_DONE;
385 			xfs_buf_delwri_queue(fbuf, buffer_list);
386 			xfs_buf_relse(fbuf);
387 		}
388 	}
389 	return 0;
390 }
391 
392 /*
393  * Align startino and allocmask for a recently allocated sparse chunk such that
394  * they are fit for insertion (or merge) into the on-disk inode btrees.
395  *
396  * Background:
397  *
398  * When enabled, sparse inode support increases the inode alignment from cluster
399  * size to inode chunk size. This means that the minimum range between two
400  * non-adjacent inode records in the inobt is large enough for a full inode
401  * record. This allows for cluster sized, cluster aligned block allocation
402  * without need to worry about whether the resulting inode record overlaps with
403  * another record in the tree. Without this basic rule, we would have to deal
404  * with the consequences of overlap by potentially undoing recent allocations in
405  * the inode allocation codepath.
406  *
407  * Because of this alignment rule (which is enforced on mount), there are two
408  * inobt possibilities for newly allocated sparse chunks. One is that the
409  * aligned inode record for the chunk covers a range of inodes not already
410  * covered in the inobt (i.e., it is safe to insert a new sparse record). The
411  * other is that a record already exists at the aligned startino that considers
412  * the newly allocated range as sparse. In the latter case, record content is
413  * merged in hope that sparse inode chunks fill to full chunks over time.
414  */
415 STATIC void
416 xfs_align_sparse_ino(
417 	struct xfs_mount		*mp,
418 	xfs_agino_t			*startino,
419 	uint16_t			*allocmask)
420 {
421 	xfs_agblock_t			agbno;
422 	xfs_agblock_t			mod;
423 	int				offset;
424 
425 	agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
426 	mod = agbno % mp->m_sb.sb_inoalignmt;
427 	if (!mod)
428 		return;
429 
430 	/* calculate the inode offset and align startino */
431 	offset = mod << mp->m_sb.sb_inopblog;
432 	*startino -= offset;
433 
434 	/*
435 	 * Since startino has been aligned down, left shift allocmask such that
436 	 * it continues to represent the same physical inodes relative to the
437 	 * new startino.
438 	 */
439 	*allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
440 }
441 
442 /*
443  * Determine whether the source inode record can merge into the target. Both
444  * records must be sparse, the inode ranges must match and there must be no
445  * allocation overlap between the records.
446  */
447 STATIC bool
448 __xfs_inobt_can_merge(
449 	struct xfs_inobt_rec_incore	*trec,	/* tgt record */
450 	struct xfs_inobt_rec_incore	*srec)	/* src record */
451 {
452 	uint64_t			talloc;
453 	uint64_t			salloc;
454 
455 	/* records must cover the same inode range */
456 	if (trec->ir_startino != srec->ir_startino)
457 		return false;
458 
459 	/* both records must be sparse */
460 	if (!xfs_inobt_issparse(trec->ir_holemask) ||
461 	    !xfs_inobt_issparse(srec->ir_holemask))
462 		return false;
463 
464 	/* both records must track some inodes */
465 	if (!trec->ir_count || !srec->ir_count)
466 		return false;
467 
468 	/* can't exceed capacity of a full record */
469 	if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
470 		return false;
471 
472 	/* verify there is no allocation overlap */
473 	talloc = xfs_inobt_irec_to_allocmask(trec);
474 	salloc = xfs_inobt_irec_to_allocmask(srec);
475 	if (talloc & salloc)
476 		return false;
477 
478 	return true;
479 }
480 
481 /*
482  * Merge the source inode record into the target. The caller must call
483  * __xfs_inobt_can_merge() to ensure the merge is valid.
484  */
485 STATIC void
486 __xfs_inobt_rec_merge(
487 	struct xfs_inobt_rec_incore	*trec,	/* target */
488 	struct xfs_inobt_rec_incore	*srec)	/* src */
489 {
490 	ASSERT(trec->ir_startino == srec->ir_startino);
491 
492 	/* combine the counts */
493 	trec->ir_count += srec->ir_count;
494 	trec->ir_freecount += srec->ir_freecount;
495 
496 	/*
497 	 * Merge the holemask and free mask. For both fields, 0 bits refer to
498 	 * allocated inodes. We combine the allocated ranges with bitwise AND.
499 	 */
500 	trec->ir_holemask &= srec->ir_holemask;
501 	trec->ir_free &= srec->ir_free;
502 }
503 
504 /*
505  * Insert a new sparse inode chunk into the associated inode btree. The inode
506  * record for the sparse chunk is pre-aligned to a startino that should match
507  * any pre-existing sparse inode record in the tree. This allows sparse chunks
508  * to fill over time.
509  *
510  * This function supports two modes of handling preexisting records depending on
511  * the merge flag. If merge is true, the provided record is merged with the
512  * existing record and updated in place. The merged record is returned in nrec.
513  * If merge is false, an existing record is replaced with the provided record.
514  * If no preexisting record exists, the provided record is always inserted.
515  *
516  * It is considered corruption if a merge is requested and not possible. Given
517  * the sparse inode alignment constraints, this should never happen.
518  */
519 STATIC int
520 xfs_inobt_insert_sprec(
521 	struct xfs_mount		*mp,
522 	struct xfs_trans		*tp,
523 	struct xfs_buf			*agbp,
524 	int				btnum,
525 	struct xfs_inobt_rec_incore	*nrec,	/* in/out: new/merged rec. */
526 	bool				merge)	/* merge or replace */
527 {
528 	struct xfs_btree_cur		*cur;
529 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
530 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
531 	int				error;
532 	int				i;
533 	struct xfs_inobt_rec_incore	rec;
534 
535 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
536 
537 	/* the new record is pre-aligned so we know where to look */
538 	error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
539 	if (error)
540 		goto error;
541 	/* if nothing there, insert a new record and return */
542 	if (i == 0) {
543 		error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
544 					     nrec->ir_count, nrec->ir_freecount,
545 					     nrec->ir_free, &i);
546 		if (error)
547 			goto error;
548 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
549 
550 		goto out;
551 	}
552 
553 	/*
554 	 * A record exists at this startino. Merge or replace the record
555 	 * depending on what we've been asked to do.
556 	 */
557 	if (merge) {
558 		error = xfs_inobt_get_rec(cur, &rec, &i);
559 		if (error)
560 			goto error;
561 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
562 		XFS_WANT_CORRUPTED_GOTO(mp,
563 					rec.ir_startino == nrec->ir_startino,
564 					error);
565 
566 		/*
567 		 * This should never fail. If we have coexisting records that
568 		 * cannot merge, something is seriously wrong.
569 		 */
570 		XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
571 					error);
572 
573 		trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
574 					 rec.ir_holemask, nrec->ir_startino,
575 					 nrec->ir_holemask);
576 
577 		/* merge to nrec to output the updated record */
578 		__xfs_inobt_rec_merge(nrec, &rec);
579 
580 		trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
581 					  nrec->ir_holemask);
582 
583 		error = xfs_inobt_rec_check_count(mp, nrec);
584 		if (error)
585 			goto error;
586 	}
587 
588 	error = xfs_inobt_update(cur, nrec);
589 	if (error)
590 		goto error;
591 
592 out:
593 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
594 	return 0;
595 error:
596 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
597 	return error;
598 }
599 
600 /*
601  * Allocate new inodes in the allocation group specified by agbp.
602  * Return 0 for success, else error code.
603  */
604 STATIC int				/* error code or 0 */
605 xfs_ialloc_ag_alloc(
606 	xfs_trans_t	*tp,		/* transaction pointer */
607 	xfs_buf_t	*agbp,		/* alloc group buffer */
608 	int		*alloc)
609 {
610 	xfs_agi_t	*agi;		/* allocation group header */
611 	xfs_alloc_arg_t	args;		/* allocation argument structure */
612 	xfs_agnumber_t	agno;
613 	int		error;
614 	xfs_agino_t	newino;		/* new first inode's number */
615 	xfs_agino_t	newlen;		/* new number of inodes */
616 	int		isaligned = 0;	/* inode allocation at stripe unit */
617 					/* boundary */
618 	uint16_t	allocmask = (uint16_t) -1; /* init. to full chunk */
619 	struct xfs_inobt_rec_incore rec;
620 	struct xfs_perag *pag;
621 	int		do_sparse = 0;
622 
623 	memset(&args, 0, sizeof(args));
624 	args.tp = tp;
625 	args.mp = tp->t_mountp;
626 	args.fsbno = NULLFSBLOCK;
627 	xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES);
628 
629 #ifdef DEBUG
630 	/* randomly do sparse inode allocations */
631 	if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
632 	    args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks)
633 		do_sparse = prandom_u32() & 1;
634 #endif
635 
636 	/*
637 	 * Locking will ensure that we don't have two callers in here
638 	 * at one time.
639 	 */
640 	newlen = args.mp->m_ialloc_inos;
641 	if (args.mp->m_maxicount &&
642 	    percpu_counter_read_positive(&args.mp->m_icount) + newlen >
643 							args.mp->m_maxicount)
644 		return -ENOSPC;
645 	args.minlen = args.maxlen = args.mp->m_ialloc_blks;
646 	/*
647 	 * First try to allocate inodes contiguous with the last-allocated
648 	 * chunk of inodes.  If the filesystem is striped, this will fill
649 	 * an entire stripe unit with inodes.
650 	 */
651 	agi = XFS_BUF_TO_AGI(agbp);
652 	newino = be32_to_cpu(agi->agi_newino);
653 	agno = be32_to_cpu(agi->agi_seqno);
654 	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
655 		     args.mp->m_ialloc_blks;
656 	if (do_sparse)
657 		goto sparse_alloc;
658 	if (likely(newino != NULLAGINO &&
659 		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
660 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
661 		args.type = XFS_ALLOCTYPE_THIS_BNO;
662 		args.prod = 1;
663 
664 		/*
665 		 * We need to take into account alignment here to ensure that
666 		 * we don't modify the free list if we fail to have an exact
667 		 * block. If we don't have an exact match, and every oher
668 		 * attempt allocation attempt fails, we'll end up cancelling
669 		 * a dirty transaction and shutting down.
670 		 *
671 		 * For an exact allocation, alignment must be 1,
672 		 * however we need to take cluster alignment into account when
673 		 * fixing up the freelist. Use the minalignslop field to
674 		 * indicate that extra blocks might be required for alignment,
675 		 * but not to use them in the actual exact allocation.
676 		 */
677 		args.alignment = 1;
678 		args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1;
679 
680 		/* Allow space for the inode btree to split. */
681 		args.minleft = args.mp->m_in_maxlevels - 1;
682 		if ((error = xfs_alloc_vextent(&args)))
683 			return error;
684 
685 		/*
686 		 * This request might have dirtied the transaction if the AG can
687 		 * satisfy the request, but the exact block was not available.
688 		 * If the allocation did fail, subsequent requests will relax
689 		 * the exact agbno requirement and increase the alignment
690 		 * instead. It is critical that the total size of the request
691 		 * (len + alignment + slop) does not increase from this point
692 		 * on, so reset minalignslop to ensure it is not included in
693 		 * subsequent requests.
694 		 */
695 		args.minalignslop = 0;
696 	}
697 
698 	if (unlikely(args.fsbno == NULLFSBLOCK)) {
699 		/*
700 		 * Set the alignment for the allocation.
701 		 * If stripe alignment is turned on then align at stripe unit
702 		 * boundary.
703 		 * If the cluster size is smaller than a filesystem block
704 		 * then we're doing I/O for inodes in filesystem block size
705 		 * pieces, so don't need alignment anyway.
706 		 */
707 		isaligned = 0;
708 		if (args.mp->m_sinoalign) {
709 			ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
710 			args.alignment = args.mp->m_dalign;
711 			isaligned = 1;
712 		} else
713 			args.alignment = xfs_ialloc_cluster_alignment(args.mp);
714 		/*
715 		 * Need to figure out where to allocate the inode blocks.
716 		 * Ideally they should be spaced out through the a.g.
717 		 * For now, just allocate blocks up front.
718 		 */
719 		args.agbno = be32_to_cpu(agi->agi_root);
720 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
721 		/*
722 		 * Allocate a fixed-size extent of inodes.
723 		 */
724 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
725 		args.prod = 1;
726 		/*
727 		 * Allow space for the inode btree to split.
728 		 */
729 		args.minleft = args.mp->m_in_maxlevels - 1;
730 		if ((error = xfs_alloc_vextent(&args)))
731 			return error;
732 	}
733 
734 	/*
735 	 * If stripe alignment is turned on, then try again with cluster
736 	 * alignment.
737 	 */
738 	if (isaligned && args.fsbno == NULLFSBLOCK) {
739 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
740 		args.agbno = be32_to_cpu(agi->agi_root);
741 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
742 		args.alignment = xfs_ialloc_cluster_alignment(args.mp);
743 		if ((error = xfs_alloc_vextent(&args)))
744 			return error;
745 	}
746 
747 	/*
748 	 * Finally, try a sparse allocation if the filesystem supports it and
749 	 * the sparse allocation length is smaller than a full chunk.
750 	 */
751 	if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
752 	    args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks &&
753 	    args.fsbno == NULLFSBLOCK) {
754 sparse_alloc:
755 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
756 		args.agbno = be32_to_cpu(agi->agi_root);
757 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
758 		args.alignment = args.mp->m_sb.sb_spino_align;
759 		args.prod = 1;
760 
761 		args.minlen = args.mp->m_ialloc_min_blks;
762 		args.maxlen = args.minlen;
763 
764 		/*
765 		 * The inode record will be aligned to full chunk size. We must
766 		 * prevent sparse allocation from AG boundaries that result in
767 		 * invalid inode records, such as records that start at agbno 0
768 		 * or extend beyond the AG.
769 		 *
770 		 * Set min agbno to the first aligned, non-zero agbno and max to
771 		 * the last aligned agbno that is at least one full chunk from
772 		 * the end of the AG.
773 		 */
774 		args.min_agbno = args.mp->m_sb.sb_inoalignmt;
775 		args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
776 					    args.mp->m_sb.sb_inoalignmt) -
777 				 args.mp->m_ialloc_blks;
778 
779 		error = xfs_alloc_vextent(&args);
780 		if (error)
781 			return error;
782 
783 		newlen = args.len << args.mp->m_sb.sb_inopblog;
784 		ASSERT(newlen <= XFS_INODES_PER_CHUNK);
785 		allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
786 	}
787 
788 	if (args.fsbno == NULLFSBLOCK) {
789 		*alloc = 0;
790 		return 0;
791 	}
792 	ASSERT(args.len == args.minlen);
793 
794 	/*
795 	 * Stamp and write the inode buffers.
796 	 *
797 	 * Seed the new inode cluster with a random generation number. This
798 	 * prevents short-term reuse of generation numbers if a chunk is
799 	 * freed and then immediately reallocated. We use random numbers
800 	 * rather than a linear progression to prevent the next generation
801 	 * number from being easily guessable.
802 	 */
803 	error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
804 			args.agbno, args.len, prandom_u32());
805 
806 	if (error)
807 		return error;
808 	/*
809 	 * Convert the results.
810 	 */
811 	newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
812 
813 	if (xfs_inobt_issparse(~allocmask)) {
814 		/*
815 		 * We've allocated a sparse chunk. Align the startino and mask.
816 		 */
817 		xfs_align_sparse_ino(args.mp, &newino, &allocmask);
818 
819 		rec.ir_startino = newino;
820 		rec.ir_holemask = ~allocmask;
821 		rec.ir_count = newlen;
822 		rec.ir_freecount = newlen;
823 		rec.ir_free = XFS_INOBT_ALL_FREE;
824 
825 		/*
826 		 * Insert the sparse record into the inobt and allow for a merge
827 		 * if necessary. If a merge does occur, rec is updated to the
828 		 * merged record.
829 		 */
830 		error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
831 					       &rec, true);
832 		if (error == -EFSCORRUPTED) {
833 			xfs_alert(args.mp,
834 	"invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
835 				  XFS_AGINO_TO_INO(args.mp, agno,
836 						   rec.ir_startino),
837 				  rec.ir_holemask, rec.ir_count);
838 			xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
839 		}
840 		if (error)
841 			return error;
842 
843 		/*
844 		 * We can't merge the part we've just allocated as for the inobt
845 		 * due to finobt semantics. The original record may or may not
846 		 * exist independent of whether physical inodes exist in this
847 		 * sparse chunk.
848 		 *
849 		 * We must update the finobt record based on the inobt record.
850 		 * rec contains the fully merged and up to date inobt record
851 		 * from the previous call. Set merge false to replace any
852 		 * existing record with this one.
853 		 */
854 		if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
855 			error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
856 						       XFS_BTNUM_FINO, &rec,
857 						       false);
858 			if (error)
859 				return error;
860 		}
861 	} else {
862 		/* full chunk - insert new records to both btrees */
863 		error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
864 					 XFS_BTNUM_INO);
865 		if (error)
866 			return error;
867 
868 		if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
869 			error = xfs_inobt_insert(args.mp, tp, agbp, newino,
870 						 newlen, XFS_BTNUM_FINO);
871 			if (error)
872 				return error;
873 		}
874 	}
875 
876 	/*
877 	 * Update AGI counts and newino.
878 	 */
879 	be32_add_cpu(&agi->agi_count, newlen);
880 	be32_add_cpu(&agi->agi_freecount, newlen);
881 	pag = xfs_perag_get(args.mp, agno);
882 	pag->pagi_freecount += newlen;
883 	xfs_perag_put(pag);
884 	agi->agi_newino = cpu_to_be32(newino);
885 
886 	/*
887 	 * Log allocation group header fields
888 	 */
889 	xfs_ialloc_log_agi(tp, agbp,
890 		XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
891 	/*
892 	 * Modify/log superblock values for inode count and inode free count.
893 	 */
894 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
895 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
896 	*alloc = 1;
897 	return 0;
898 }
899 
900 STATIC xfs_agnumber_t
901 xfs_ialloc_next_ag(
902 	xfs_mount_t	*mp)
903 {
904 	xfs_agnumber_t	agno;
905 
906 	spin_lock(&mp->m_agirotor_lock);
907 	agno = mp->m_agirotor;
908 	if (++mp->m_agirotor >= mp->m_maxagi)
909 		mp->m_agirotor = 0;
910 	spin_unlock(&mp->m_agirotor_lock);
911 
912 	return agno;
913 }
914 
915 /*
916  * Select an allocation group to look for a free inode in, based on the parent
917  * inode and the mode.  Return the allocation group buffer.
918  */
919 STATIC xfs_agnumber_t
920 xfs_ialloc_ag_select(
921 	xfs_trans_t	*tp,		/* transaction pointer */
922 	xfs_ino_t	parent,		/* parent directory inode number */
923 	umode_t		mode,		/* bits set to indicate file type */
924 	int		okalloc)	/* ok to allocate more space */
925 {
926 	xfs_agnumber_t	agcount;	/* number of ag's in the filesystem */
927 	xfs_agnumber_t	agno;		/* current ag number */
928 	int		flags;		/* alloc buffer locking flags */
929 	xfs_extlen_t	ineed;		/* blocks needed for inode allocation */
930 	xfs_extlen_t	longest = 0;	/* longest extent available */
931 	xfs_mount_t	*mp;		/* mount point structure */
932 	int		needspace;	/* file mode implies space allocated */
933 	xfs_perag_t	*pag;		/* per allocation group data */
934 	xfs_agnumber_t	pagno;		/* parent (starting) ag number */
935 	int		error;
936 
937 	/*
938 	 * Files of these types need at least one block if length > 0
939 	 * (and they won't fit in the inode, but that's hard to figure out).
940 	 */
941 	needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
942 	mp = tp->t_mountp;
943 	agcount = mp->m_maxagi;
944 	if (S_ISDIR(mode))
945 		pagno = xfs_ialloc_next_ag(mp);
946 	else {
947 		pagno = XFS_INO_TO_AGNO(mp, parent);
948 		if (pagno >= agcount)
949 			pagno = 0;
950 	}
951 
952 	ASSERT(pagno < agcount);
953 
954 	/*
955 	 * Loop through allocation groups, looking for one with a little
956 	 * free space in it.  Note we don't look for free inodes, exactly.
957 	 * Instead, we include whether there is a need to allocate inodes
958 	 * to mean that blocks must be allocated for them,
959 	 * if none are currently free.
960 	 */
961 	agno = pagno;
962 	flags = XFS_ALLOC_FLAG_TRYLOCK;
963 	for (;;) {
964 		pag = xfs_perag_get(mp, agno);
965 		if (!pag->pagi_inodeok) {
966 			xfs_ialloc_next_ag(mp);
967 			goto nextag;
968 		}
969 
970 		if (!pag->pagi_init) {
971 			error = xfs_ialloc_pagi_init(mp, tp, agno);
972 			if (error)
973 				goto nextag;
974 		}
975 
976 		if (pag->pagi_freecount) {
977 			xfs_perag_put(pag);
978 			return agno;
979 		}
980 
981 		if (!okalloc)
982 			goto nextag;
983 
984 		if (!pag->pagf_init) {
985 			error = xfs_alloc_pagf_init(mp, tp, agno, flags);
986 			if (error)
987 				goto nextag;
988 		}
989 
990 		/*
991 		 * Check that there is enough free space for the file plus a
992 		 * chunk of inodes if we need to allocate some. If this is the
993 		 * first pass across the AGs, take into account the potential
994 		 * space needed for alignment of inode chunks when checking the
995 		 * longest contiguous free space in the AG - this prevents us
996 		 * from getting ENOSPC because we have free space larger than
997 		 * m_ialloc_blks but alignment constraints prevent us from using
998 		 * it.
999 		 *
1000 		 * If we can't find an AG with space for full alignment slack to
1001 		 * be taken into account, we must be near ENOSPC in all AGs.
1002 		 * Hence we don't include alignment for the second pass and so
1003 		 * if we fail allocation due to alignment issues then it is most
1004 		 * likely a real ENOSPC condition.
1005 		 */
1006 		ineed = mp->m_ialloc_min_blks;
1007 		if (flags && ineed > 1)
1008 			ineed += xfs_ialloc_cluster_alignment(mp);
1009 		longest = pag->pagf_longest;
1010 		if (!longest)
1011 			longest = pag->pagf_flcount > 0;
1012 
1013 		if (pag->pagf_freeblks >= needspace + ineed &&
1014 		    longest >= ineed) {
1015 			xfs_perag_put(pag);
1016 			return agno;
1017 		}
1018 nextag:
1019 		xfs_perag_put(pag);
1020 		/*
1021 		 * No point in iterating over the rest, if we're shutting
1022 		 * down.
1023 		 */
1024 		if (XFS_FORCED_SHUTDOWN(mp))
1025 			return NULLAGNUMBER;
1026 		agno++;
1027 		if (agno >= agcount)
1028 			agno = 0;
1029 		if (agno == pagno) {
1030 			if (flags == 0)
1031 				return NULLAGNUMBER;
1032 			flags = 0;
1033 		}
1034 	}
1035 }
1036 
1037 /*
1038  * Try to retrieve the next record to the left/right from the current one.
1039  */
1040 STATIC int
1041 xfs_ialloc_next_rec(
1042 	struct xfs_btree_cur	*cur,
1043 	xfs_inobt_rec_incore_t	*rec,
1044 	int			*done,
1045 	int			left)
1046 {
1047 	int                     error;
1048 	int			i;
1049 
1050 	if (left)
1051 		error = xfs_btree_decrement(cur, 0, &i);
1052 	else
1053 		error = xfs_btree_increment(cur, 0, &i);
1054 
1055 	if (error)
1056 		return error;
1057 	*done = !i;
1058 	if (i) {
1059 		error = xfs_inobt_get_rec(cur, rec, &i);
1060 		if (error)
1061 			return error;
1062 		XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 STATIC int
1069 xfs_ialloc_get_rec(
1070 	struct xfs_btree_cur	*cur,
1071 	xfs_agino_t		agino,
1072 	xfs_inobt_rec_incore_t	*rec,
1073 	int			*done)
1074 {
1075 	int                     error;
1076 	int			i;
1077 
1078 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1079 	if (error)
1080 		return error;
1081 	*done = !i;
1082 	if (i) {
1083 		error = xfs_inobt_get_rec(cur, rec, &i);
1084 		if (error)
1085 			return error;
1086 		XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1087 	}
1088 
1089 	return 0;
1090 }
1091 
1092 /*
1093  * Return the offset of the first free inode in the record. If the inode chunk
1094  * is sparsely allocated, we convert the record holemask to inode granularity
1095  * and mask off the unallocated regions from the inode free mask.
1096  */
1097 STATIC int
1098 xfs_inobt_first_free_inode(
1099 	struct xfs_inobt_rec_incore	*rec)
1100 {
1101 	xfs_inofree_t			realfree;
1102 
1103 	/* if there are no holes, return the first available offset */
1104 	if (!xfs_inobt_issparse(rec->ir_holemask))
1105 		return xfs_lowbit64(rec->ir_free);
1106 
1107 	realfree = xfs_inobt_irec_to_allocmask(rec);
1108 	realfree &= rec->ir_free;
1109 
1110 	return xfs_lowbit64(realfree);
1111 }
1112 
1113 /*
1114  * Allocate an inode using the inobt-only algorithm.
1115  */
1116 STATIC int
1117 xfs_dialloc_ag_inobt(
1118 	struct xfs_trans	*tp,
1119 	struct xfs_buf		*agbp,
1120 	xfs_ino_t		parent,
1121 	xfs_ino_t		*inop)
1122 {
1123 	struct xfs_mount	*mp = tp->t_mountp;
1124 	struct xfs_agi		*agi = XFS_BUF_TO_AGI(agbp);
1125 	xfs_agnumber_t		agno = be32_to_cpu(agi->agi_seqno);
1126 	xfs_agnumber_t		pagno = XFS_INO_TO_AGNO(mp, parent);
1127 	xfs_agino_t		pagino = XFS_INO_TO_AGINO(mp, parent);
1128 	struct xfs_perag	*pag;
1129 	struct xfs_btree_cur	*cur, *tcur;
1130 	struct xfs_inobt_rec_incore rec, trec;
1131 	xfs_ino_t		ino;
1132 	int			error;
1133 	int			offset;
1134 	int			i, j;
1135 	int			searchdistance = 10;
1136 
1137 	pag = xfs_perag_get(mp, agno);
1138 
1139 	ASSERT(pag->pagi_init);
1140 	ASSERT(pag->pagi_inodeok);
1141 	ASSERT(pag->pagi_freecount > 0);
1142 
1143  restart_pagno:
1144 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1145 	/*
1146 	 * If pagino is 0 (this is the root inode allocation) use newino.
1147 	 * This must work because we've just allocated some.
1148 	 */
1149 	if (!pagino)
1150 		pagino = be32_to_cpu(agi->agi_newino);
1151 
1152 	error = xfs_check_agi_freecount(cur, agi);
1153 	if (error)
1154 		goto error0;
1155 
1156 	/*
1157 	 * If in the same AG as the parent, try to get near the parent.
1158 	 */
1159 	if (pagno == agno) {
1160 		int		doneleft;	/* done, to the left */
1161 		int		doneright;	/* done, to the right */
1162 
1163 		error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1164 		if (error)
1165 			goto error0;
1166 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1167 
1168 		error = xfs_inobt_get_rec(cur, &rec, &j);
1169 		if (error)
1170 			goto error0;
1171 		XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
1172 
1173 		if (rec.ir_freecount > 0) {
1174 			/*
1175 			 * Found a free inode in the same chunk
1176 			 * as the parent, done.
1177 			 */
1178 			goto alloc_inode;
1179 		}
1180 
1181 
1182 		/*
1183 		 * In the same AG as parent, but parent's chunk is full.
1184 		 */
1185 
1186 		/* duplicate the cursor, search left & right simultaneously */
1187 		error = xfs_btree_dup_cursor(cur, &tcur);
1188 		if (error)
1189 			goto error0;
1190 
1191 		/*
1192 		 * Skip to last blocks looked up if same parent inode.
1193 		 */
1194 		if (pagino != NULLAGINO &&
1195 		    pag->pagl_pagino == pagino &&
1196 		    pag->pagl_leftrec != NULLAGINO &&
1197 		    pag->pagl_rightrec != NULLAGINO) {
1198 			error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1199 						   &trec, &doneleft);
1200 			if (error)
1201 				goto error1;
1202 
1203 			error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1204 						   &rec, &doneright);
1205 			if (error)
1206 				goto error1;
1207 		} else {
1208 			/* search left with tcur, back up 1 record */
1209 			error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1210 			if (error)
1211 				goto error1;
1212 
1213 			/* search right with cur, go forward 1 record. */
1214 			error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1215 			if (error)
1216 				goto error1;
1217 		}
1218 
1219 		/*
1220 		 * Loop until we find an inode chunk with a free inode.
1221 		 */
1222 		while (--searchdistance > 0 && (!doneleft || !doneright)) {
1223 			int	useleft;  /* using left inode chunk this time */
1224 
1225 			/* figure out the closer block if both are valid. */
1226 			if (!doneleft && !doneright) {
1227 				useleft = pagino -
1228 				 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1229 				  rec.ir_startino - pagino;
1230 			} else {
1231 				useleft = !doneleft;
1232 			}
1233 
1234 			/* free inodes to the left? */
1235 			if (useleft && trec.ir_freecount) {
1236 				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1237 				cur = tcur;
1238 
1239 				pag->pagl_leftrec = trec.ir_startino;
1240 				pag->pagl_rightrec = rec.ir_startino;
1241 				pag->pagl_pagino = pagino;
1242 				rec = trec;
1243 				goto alloc_inode;
1244 			}
1245 
1246 			/* free inodes to the right? */
1247 			if (!useleft && rec.ir_freecount) {
1248 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1249 
1250 				pag->pagl_leftrec = trec.ir_startino;
1251 				pag->pagl_rightrec = rec.ir_startino;
1252 				pag->pagl_pagino = pagino;
1253 				goto alloc_inode;
1254 			}
1255 
1256 			/* get next record to check */
1257 			if (useleft) {
1258 				error = xfs_ialloc_next_rec(tcur, &trec,
1259 								 &doneleft, 1);
1260 			} else {
1261 				error = xfs_ialloc_next_rec(cur, &rec,
1262 								 &doneright, 0);
1263 			}
1264 			if (error)
1265 				goto error1;
1266 		}
1267 
1268 		if (searchdistance <= 0) {
1269 			/*
1270 			 * Not in range - save last search
1271 			 * location and allocate a new inode
1272 			 */
1273 			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1274 			pag->pagl_leftrec = trec.ir_startino;
1275 			pag->pagl_rightrec = rec.ir_startino;
1276 			pag->pagl_pagino = pagino;
1277 
1278 		} else {
1279 			/*
1280 			 * We've reached the end of the btree. because
1281 			 * we are only searching a small chunk of the
1282 			 * btree each search, there is obviously free
1283 			 * inodes closer to the parent inode than we
1284 			 * are now. restart the search again.
1285 			 */
1286 			pag->pagl_pagino = NULLAGINO;
1287 			pag->pagl_leftrec = NULLAGINO;
1288 			pag->pagl_rightrec = NULLAGINO;
1289 			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1290 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1291 			goto restart_pagno;
1292 		}
1293 	}
1294 
1295 	/*
1296 	 * In a different AG from the parent.
1297 	 * See if the most recently allocated block has any free.
1298 	 */
1299 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1300 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1301 					 XFS_LOOKUP_EQ, &i);
1302 		if (error)
1303 			goto error0;
1304 
1305 		if (i == 1) {
1306 			error = xfs_inobt_get_rec(cur, &rec, &j);
1307 			if (error)
1308 				goto error0;
1309 
1310 			if (j == 1 && rec.ir_freecount > 0) {
1311 				/*
1312 				 * The last chunk allocated in the group
1313 				 * still has a free inode.
1314 				 */
1315 				goto alloc_inode;
1316 			}
1317 		}
1318 	}
1319 
1320 	/*
1321 	 * None left in the last group, search the whole AG
1322 	 */
1323 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1324 	if (error)
1325 		goto error0;
1326 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1327 
1328 	for (;;) {
1329 		error = xfs_inobt_get_rec(cur, &rec, &i);
1330 		if (error)
1331 			goto error0;
1332 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1333 		if (rec.ir_freecount > 0)
1334 			break;
1335 		error = xfs_btree_increment(cur, 0, &i);
1336 		if (error)
1337 			goto error0;
1338 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1339 	}
1340 
1341 alloc_inode:
1342 	offset = xfs_inobt_first_free_inode(&rec);
1343 	ASSERT(offset >= 0);
1344 	ASSERT(offset < XFS_INODES_PER_CHUNK);
1345 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1346 				   XFS_INODES_PER_CHUNK) == 0);
1347 	ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1348 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1349 	rec.ir_freecount--;
1350 	error = xfs_inobt_update(cur, &rec);
1351 	if (error)
1352 		goto error0;
1353 	be32_add_cpu(&agi->agi_freecount, -1);
1354 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1355 	pag->pagi_freecount--;
1356 
1357 	error = xfs_check_agi_freecount(cur, agi);
1358 	if (error)
1359 		goto error0;
1360 
1361 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1362 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1363 	xfs_perag_put(pag);
1364 	*inop = ino;
1365 	return 0;
1366 error1:
1367 	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1368 error0:
1369 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1370 	xfs_perag_put(pag);
1371 	return error;
1372 }
1373 
1374 /*
1375  * Use the free inode btree to allocate an inode based on distance from the
1376  * parent. Note that the provided cursor may be deleted and replaced.
1377  */
1378 STATIC int
1379 xfs_dialloc_ag_finobt_near(
1380 	xfs_agino_t			pagino,
1381 	struct xfs_btree_cur		**ocur,
1382 	struct xfs_inobt_rec_incore	*rec)
1383 {
1384 	struct xfs_btree_cur		*lcur = *ocur;	/* left search cursor */
1385 	struct xfs_btree_cur		*rcur;	/* right search cursor */
1386 	struct xfs_inobt_rec_incore	rrec;
1387 	int				error;
1388 	int				i, j;
1389 
1390 	error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1391 	if (error)
1392 		return error;
1393 
1394 	if (i == 1) {
1395 		error = xfs_inobt_get_rec(lcur, rec, &i);
1396 		if (error)
1397 			return error;
1398 		XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
1399 
1400 		/*
1401 		 * See if we've landed in the parent inode record. The finobt
1402 		 * only tracks chunks with at least one free inode, so record
1403 		 * existence is enough.
1404 		 */
1405 		if (pagino >= rec->ir_startino &&
1406 		    pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1407 			return 0;
1408 	}
1409 
1410 	error = xfs_btree_dup_cursor(lcur, &rcur);
1411 	if (error)
1412 		return error;
1413 
1414 	error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1415 	if (error)
1416 		goto error_rcur;
1417 	if (j == 1) {
1418 		error = xfs_inobt_get_rec(rcur, &rrec, &j);
1419 		if (error)
1420 			goto error_rcur;
1421 		XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
1422 	}
1423 
1424 	XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
1425 	if (i == 1 && j == 1) {
1426 		/*
1427 		 * Both the left and right records are valid. Choose the closer
1428 		 * inode chunk to the target.
1429 		 */
1430 		if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1431 		    (rrec.ir_startino - pagino)) {
1432 			*rec = rrec;
1433 			xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1434 			*ocur = rcur;
1435 		} else {
1436 			xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1437 		}
1438 	} else if (j == 1) {
1439 		/* only the right record is valid */
1440 		*rec = rrec;
1441 		xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1442 		*ocur = rcur;
1443 	} else if (i == 1) {
1444 		/* only the left record is valid */
1445 		xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1446 	}
1447 
1448 	return 0;
1449 
1450 error_rcur:
1451 	xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1452 	return error;
1453 }
1454 
1455 /*
1456  * Use the free inode btree to find a free inode based on a newino hint. If
1457  * the hint is NULL, find the first free inode in the AG.
1458  */
1459 STATIC int
1460 xfs_dialloc_ag_finobt_newino(
1461 	struct xfs_agi			*agi,
1462 	struct xfs_btree_cur		*cur,
1463 	struct xfs_inobt_rec_incore	*rec)
1464 {
1465 	int error;
1466 	int i;
1467 
1468 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1469 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1470 					 XFS_LOOKUP_EQ, &i);
1471 		if (error)
1472 			return error;
1473 		if (i == 1) {
1474 			error = xfs_inobt_get_rec(cur, rec, &i);
1475 			if (error)
1476 				return error;
1477 			XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1478 			return 0;
1479 		}
1480 	}
1481 
1482 	/*
1483 	 * Find the first inode available in the AG.
1484 	 */
1485 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1486 	if (error)
1487 		return error;
1488 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1489 
1490 	error = xfs_inobt_get_rec(cur, rec, &i);
1491 	if (error)
1492 		return error;
1493 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1494 
1495 	return 0;
1496 }
1497 
1498 /*
1499  * Update the inobt based on a modification made to the finobt. Also ensure that
1500  * the records from both trees are equivalent post-modification.
1501  */
1502 STATIC int
1503 xfs_dialloc_ag_update_inobt(
1504 	struct xfs_btree_cur		*cur,	/* inobt cursor */
1505 	struct xfs_inobt_rec_incore	*frec,	/* finobt record */
1506 	int				offset) /* inode offset */
1507 {
1508 	struct xfs_inobt_rec_incore	rec;
1509 	int				error;
1510 	int				i;
1511 
1512 	error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1513 	if (error)
1514 		return error;
1515 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1516 
1517 	error = xfs_inobt_get_rec(cur, &rec, &i);
1518 	if (error)
1519 		return error;
1520 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1521 	ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1522 				   XFS_INODES_PER_CHUNK) == 0);
1523 
1524 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1525 	rec.ir_freecount--;
1526 
1527 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
1528 				  (rec.ir_freecount == frec->ir_freecount));
1529 
1530 	return xfs_inobt_update(cur, &rec);
1531 }
1532 
1533 /*
1534  * Allocate an inode using the free inode btree, if available. Otherwise, fall
1535  * back to the inobt search algorithm.
1536  *
1537  * The caller selected an AG for us, and made sure that free inodes are
1538  * available.
1539  */
1540 STATIC int
1541 xfs_dialloc_ag(
1542 	struct xfs_trans	*tp,
1543 	struct xfs_buf		*agbp,
1544 	xfs_ino_t		parent,
1545 	xfs_ino_t		*inop)
1546 {
1547 	struct xfs_mount		*mp = tp->t_mountp;
1548 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
1549 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
1550 	xfs_agnumber_t			pagno = XFS_INO_TO_AGNO(mp, parent);
1551 	xfs_agino_t			pagino = XFS_INO_TO_AGINO(mp, parent);
1552 	struct xfs_perag		*pag;
1553 	struct xfs_btree_cur		*cur;	/* finobt cursor */
1554 	struct xfs_btree_cur		*icur;	/* inobt cursor */
1555 	struct xfs_inobt_rec_incore	rec;
1556 	xfs_ino_t			ino;
1557 	int				error;
1558 	int				offset;
1559 	int				i;
1560 
1561 	if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1562 		return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1563 
1564 	pag = xfs_perag_get(mp, agno);
1565 
1566 	/*
1567 	 * If pagino is 0 (this is the root inode allocation) use newino.
1568 	 * This must work because we've just allocated some.
1569 	 */
1570 	if (!pagino)
1571 		pagino = be32_to_cpu(agi->agi_newino);
1572 
1573 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1574 
1575 	error = xfs_check_agi_freecount(cur, agi);
1576 	if (error)
1577 		goto error_cur;
1578 
1579 	/*
1580 	 * The search algorithm depends on whether we're in the same AG as the
1581 	 * parent. If so, find the closest available inode to the parent. If
1582 	 * not, consider the agi hint or find the first free inode in the AG.
1583 	 */
1584 	if (agno == pagno)
1585 		error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1586 	else
1587 		error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1588 	if (error)
1589 		goto error_cur;
1590 
1591 	offset = xfs_inobt_first_free_inode(&rec);
1592 	ASSERT(offset >= 0);
1593 	ASSERT(offset < XFS_INODES_PER_CHUNK);
1594 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1595 				   XFS_INODES_PER_CHUNK) == 0);
1596 	ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1597 
1598 	/*
1599 	 * Modify or remove the finobt record.
1600 	 */
1601 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1602 	rec.ir_freecount--;
1603 	if (rec.ir_freecount)
1604 		error = xfs_inobt_update(cur, &rec);
1605 	else
1606 		error = xfs_btree_delete(cur, &i);
1607 	if (error)
1608 		goto error_cur;
1609 
1610 	/*
1611 	 * The finobt has now been updated appropriately. We haven't updated the
1612 	 * agi and superblock yet, so we can create an inobt cursor and validate
1613 	 * the original freecount. If all is well, make the equivalent update to
1614 	 * the inobt using the finobt record and offset information.
1615 	 */
1616 	icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1617 
1618 	error = xfs_check_agi_freecount(icur, agi);
1619 	if (error)
1620 		goto error_icur;
1621 
1622 	error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1623 	if (error)
1624 		goto error_icur;
1625 
1626 	/*
1627 	 * Both trees have now been updated. We must update the perag and
1628 	 * superblock before we can check the freecount for each btree.
1629 	 */
1630 	be32_add_cpu(&agi->agi_freecount, -1);
1631 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1632 	pag->pagi_freecount--;
1633 
1634 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1635 
1636 	error = xfs_check_agi_freecount(icur, agi);
1637 	if (error)
1638 		goto error_icur;
1639 	error = xfs_check_agi_freecount(cur, agi);
1640 	if (error)
1641 		goto error_icur;
1642 
1643 	xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1644 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1645 	xfs_perag_put(pag);
1646 	*inop = ino;
1647 	return 0;
1648 
1649 error_icur:
1650 	xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1651 error_cur:
1652 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1653 	xfs_perag_put(pag);
1654 	return error;
1655 }
1656 
1657 /*
1658  * Allocate an inode on disk.
1659  *
1660  * Mode is used to tell whether the new inode will need space, and whether it
1661  * is a directory.
1662  *
1663  * This function is designed to be called twice if it has to do an allocation
1664  * to make more free inodes.  On the first call, *IO_agbp should be set to NULL.
1665  * If an inode is available without having to performn an allocation, an inode
1666  * number is returned.  In this case, *IO_agbp is set to NULL.  If an allocation
1667  * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1668  * The caller should then commit the current transaction, allocate a
1669  * new transaction, and call xfs_dialloc() again, passing in the previous value
1670  * of *IO_agbp.  IO_agbp should be held across the transactions. Since the AGI
1671  * buffer is locked across the two calls, the second call is guaranteed to have
1672  * a free inode available.
1673  *
1674  * Once we successfully pick an inode its number is returned and the on-disk
1675  * data structures are updated.  The inode itself is not read in, since doing so
1676  * would break ordering constraints with xfs_reclaim.
1677  */
1678 int
1679 xfs_dialloc(
1680 	struct xfs_trans	*tp,
1681 	xfs_ino_t		parent,
1682 	umode_t			mode,
1683 	int			okalloc,
1684 	struct xfs_buf		**IO_agbp,
1685 	xfs_ino_t		*inop)
1686 {
1687 	struct xfs_mount	*mp = tp->t_mountp;
1688 	struct xfs_buf		*agbp;
1689 	xfs_agnumber_t		agno;
1690 	int			error;
1691 	int			ialloced;
1692 	int			noroom = 0;
1693 	xfs_agnumber_t		start_agno;
1694 	struct xfs_perag	*pag;
1695 
1696 	if (*IO_agbp) {
1697 		/*
1698 		 * If the caller passes in a pointer to the AGI buffer,
1699 		 * continue where we left off before.  In this case, we
1700 		 * know that the allocation group has free inodes.
1701 		 */
1702 		agbp = *IO_agbp;
1703 		goto out_alloc;
1704 	}
1705 
1706 	/*
1707 	 * We do not have an agbp, so select an initial allocation
1708 	 * group for inode allocation.
1709 	 */
1710 	start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1711 	if (start_agno == NULLAGNUMBER) {
1712 		*inop = NULLFSINO;
1713 		return 0;
1714 	}
1715 
1716 	/*
1717 	 * If we have already hit the ceiling of inode blocks then clear
1718 	 * okalloc so we scan all available agi structures for a free
1719 	 * inode.
1720 	 *
1721 	 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1722 	 * which will sacrifice the preciseness but improve the performance.
1723 	 */
1724 	if (mp->m_maxicount &&
1725 	    percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos
1726 							> mp->m_maxicount) {
1727 		noroom = 1;
1728 		okalloc = 0;
1729 	}
1730 
1731 	/*
1732 	 * Loop until we find an allocation group that either has free inodes
1733 	 * or in which we can allocate some inodes.  Iterate through the
1734 	 * allocation groups upward, wrapping at the end.
1735 	 */
1736 	agno = start_agno;
1737 	for (;;) {
1738 		pag = xfs_perag_get(mp, agno);
1739 		if (!pag->pagi_inodeok) {
1740 			xfs_ialloc_next_ag(mp);
1741 			goto nextag;
1742 		}
1743 
1744 		if (!pag->pagi_init) {
1745 			error = xfs_ialloc_pagi_init(mp, tp, agno);
1746 			if (error)
1747 				goto out_error;
1748 		}
1749 
1750 		/*
1751 		 * Do a first racy fast path check if this AG is usable.
1752 		 */
1753 		if (!pag->pagi_freecount && !okalloc)
1754 			goto nextag;
1755 
1756 		/*
1757 		 * Then read in the AGI buffer and recheck with the AGI buffer
1758 		 * lock held.
1759 		 */
1760 		error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1761 		if (error)
1762 			goto out_error;
1763 
1764 		if (pag->pagi_freecount) {
1765 			xfs_perag_put(pag);
1766 			goto out_alloc;
1767 		}
1768 
1769 		if (!okalloc)
1770 			goto nextag_relse_buffer;
1771 
1772 
1773 		error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1774 		if (error) {
1775 			xfs_trans_brelse(tp, agbp);
1776 
1777 			if (error != -ENOSPC)
1778 				goto out_error;
1779 
1780 			xfs_perag_put(pag);
1781 			*inop = NULLFSINO;
1782 			return 0;
1783 		}
1784 
1785 		if (ialloced) {
1786 			/*
1787 			 * We successfully allocated some inodes, return
1788 			 * the current context to the caller so that it
1789 			 * can commit the current transaction and call
1790 			 * us again where we left off.
1791 			 */
1792 			ASSERT(pag->pagi_freecount > 0);
1793 			xfs_perag_put(pag);
1794 
1795 			*IO_agbp = agbp;
1796 			*inop = NULLFSINO;
1797 			return 0;
1798 		}
1799 
1800 nextag_relse_buffer:
1801 		xfs_trans_brelse(tp, agbp);
1802 nextag:
1803 		xfs_perag_put(pag);
1804 		if (++agno == mp->m_sb.sb_agcount)
1805 			agno = 0;
1806 		if (agno == start_agno) {
1807 			*inop = NULLFSINO;
1808 			return noroom ? -ENOSPC : 0;
1809 		}
1810 	}
1811 
1812 out_alloc:
1813 	*IO_agbp = NULL;
1814 	return xfs_dialloc_ag(tp, agbp, parent, inop);
1815 out_error:
1816 	xfs_perag_put(pag);
1817 	return error;
1818 }
1819 
1820 /*
1821  * Free the blocks of an inode chunk. We must consider that the inode chunk
1822  * might be sparse and only free the regions that are allocated as part of the
1823  * chunk.
1824  */
1825 STATIC void
1826 xfs_difree_inode_chunk(
1827 	struct xfs_mount		*mp,
1828 	xfs_agnumber_t			agno,
1829 	struct xfs_inobt_rec_incore	*rec,
1830 	struct xfs_defer_ops		*dfops)
1831 {
1832 	xfs_agblock_t	sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino);
1833 	int		startidx, endidx;
1834 	int		nextbit;
1835 	xfs_agblock_t	agbno;
1836 	int		contigblk;
1837 	struct xfs_owner_info	oinfo;
1838 	DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1839 	xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES);
1840 
1841 	if (!xfs_inobt_issparse(rec->ir_holemask)) {
1842 		/* not sparse, calculate extent info directly */
1843 		xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno),
1844 				  mp->m_ialloc_blks, &oinfo);
1845 		return;
1846 	}
1847 
1848 	/* holemask is only 16-bits (fits in an unsigned long) */
1849 	ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1850 	holemask[0] = rec->ir_holemask;
1851 
1852 	/*
1853 	 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1854 	 * holemask and convert the start/end index of each range to an extent.
1855 	 * We start with the start and end index both pointing at the first 0 in
1856 	 * the mask.
1857 	 */
1858 	startidx = endidx = find_first_zero_bit(holemask,
1859 						XFS_INOBT_HOLEMASK_BITS);
1860 	nextbit = startidx + 1;
1861 	while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1862 		nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1863 					     nextbit);
1864 		/*
1865 		 * If the next zero bit is contiguous, update the end index of
1866 		 * the current range and continue.
1867 		 */
1868 		if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1869 		    nextbit == endidx + 1) {
1870 			endidx = nextbit;
1871 			goto next;
1872 		}
1873 
1874 		/*
1875 		 * nextbit is not contiguous with the current end index. Convert
1876 		 * the current start/end to an extent and add it to the free
1877 		 * list.
1878 		 */
1879 		agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1880 				  mp->m_sb.sb_inopblock;
1881 		contigblk = ((endidx - startidx + 1) *
1882 			     XFS_INODES_PER_HOLEMASK_BIT) /
1883 			    mp->m_sb.sb_inopblock;
1884 
1885 		ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1886 		ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1887 		xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno),
1888 				  contigblk, &oinfo);
1889 
1890 		/* reset range to current bit and carry on... */
1891 		startidx = endidx = nextbit;
1892 
1893 next:
1894 		nextbit++;
1895 	}
1896 }
1897 
1898 STATIC int
1899 xfs_difree_inobt(
1900 	struct xfs_mount		*mp,
1901 	struct xfs_trans		*tp,
1902 	struct xfs_buf			*agbp,
1903 	xfs_agino_t			agino,
1904 	struct xfs_defer_ops		*dfops,
1905 	struct xfs_icluster		*xic,
1906 	struct xfs_inobt_rec_incore	*orec)
1907 {
1908 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
1909 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
1910 	struct xfs_perag		*pag;
1911 	struct xfs_btree_cur		*cur;
1912 	struct xfs_inobt_rec_incore	rec;
1913 	int				ilen;
1914 	int				error;
1915 	int				i;
1916 	int				off;
1917 
1918 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1919 	ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1920 
1921 	/*
1922 	 * Initialize the cursor.
1923 	 */
1924 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1925 
1926 	error = xfs_check_agi_freecount(cur, agi);
1927 	if (error)
1928 		goto error0;
1929 
1930 	/*
1931 	 * Look for the entry describing this inode.
1932 	 */
1933 	if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1934 		xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1935 			__func__, error);
1936 		goto error0;
1937 	}
1938 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1939 	error = xfs_inobt_get_rec(cur, &rec, &i);
1940 	if (error) {
1941 		xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1942 			__func__, error);
1943 		goto error0;
1944 	}
1945 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1946 	/*
1947 	 * Get the offset in the inode chunk.
1948 	 */
1949 	off = agino - rec.ir_startino;
1950 	ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1951 	ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1952 	/*
1953 	 * Mark the inode free & increment the count.
1954 	 */
1955 	rec.ir_free |= XFS_INOBT_MASK(off);
1956 	rec.ir_freecount++;
1957 
1958 	/*
1959 	 * When an inode chunk is free, it becomes eligible for removal. Don't
1960 	 * remove the chunk if the block size is large enough for multiple inode
1961 	 * chunks (that might not be free).
1962 	 */
1963 	if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1964 	    rec.ir_free == XFS_INOBT_ALL_FREE &&
1965 	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1966 		xic->deleted = true;
1967 		xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1968 		xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1969 
1970 		/*
1971 		 * Remove the inode cluster from the AGI B+Tree, adjust the
1972 		 * AGI and Superblock inode counts, and mark the disk space
1973 		 * to be freed when the transaction is committed.
1974 		 */
1975 		ilen = rec.ir_freecount;
1976 		be32_add_cpu(&agi->agi_count, -ilen);
1977 		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1978 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1979 		pag = xfs_perag_get(mp, agno);
1980 		pag->pagi_freecount -= ilen - 1;
1981 		xfs_perag_put(pag);
1982 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1983 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1984 
1985 		if ((error = xfs_btree_delete(cur, &i))) {
1986 			xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1987 				__func__, error);
1988 			goto error0;
1989 		}
1990 
1991 		xfs_difree_inode_chunk(mp, agno, &rec, dfops);
1992 	} else {
1993 		xic->deleted = false;
1994 
1995 		error = xfs_inobt_update(cur, &rec);
1996 		if (error) {
1997 			xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1998 				__func__, error);
1999 			goto error0;
2000 		}
2001 
2002 		/*
2003 		 * Change the inode free counts and log the ag/sb changes.
2004 		 */
2005 		be32_add_cpu(&agi->agi_freecount, 1);
2006 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2007 		pag = xfs_perag_get(mp, agno);
2008 		pag->pagi_freecount++;
2009 		xfs_perag_put(pag);
2010 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2011 	}
2012 
2013 	error = xfs_check_agi_freecount(cur, agi);
2014 	if (error)
2015 		goto error0;
2016 
2017 	*orec = rec;
2018 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2019 	return 0;
2020 
2021 error0:
2022 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2023 	return error;
2024 }
2025 
2026 /*
2027  * Free an inode in the free inode btree.
2028  */
2029 STATIC int
2030 xfs_difree_finobt(
2031 	struct xfs_mount		*mp,
2032 	struct xfs_trans		*tp,
2033 	struct xfs_buf			*agbp,
2034 	xfs_agino_t			agino,
2035 	struct xfs_inobt_rec_incore	*ibtrec) /* inobt record */
2036 {
2037 	struct xfs_agi			*agi = XFS_BUF_TO_AGI(agbp);
2038 	xfs_agnumber_t			agno = be32_to_cpu(agi->agi_seqno);
2039 	struct xfs_btree_cur		*cur;
2040 	struct xfs_inobt_rec_incore	rec;
2041 	int				offset = agino - ibtrec->ir_startino;
2042 	int				error;
2043 	int				i;
2044 
2045 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2046 
2047 	error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2048 	if (error)
2049 		goto error;
2050 	if (i == 0) {
2051 		/*
2052 		 * If the record does not exist in the finobt, we must have just
2053 		 * freed an inode in a previously fully allocated chunk. If not,
2054 		 * something is out of sync.
2055 		 */
2056 		XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
2057 
2058 		error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2059 					     ibtrec->ir_count,
2060 					     ibtrec->ir_freecount,
2061 					     ibtrec->ir_free, &i);
2062 		if (error)
2063 			goto error;
2064 		ASSERT(i == 1);
2065 
2066 		goto out;
2067 	}
2068 
2069 	/*
2070 	 * Read and update the existing record. We could just copy the ibtrec
2071 	 * across here, but that would defeat the purpose of having redundant
2072 	 * metadata. By making the modifications independently, we can catch
2073 	 * corruptions that we wouldn't see if we just copied from one record
2074 	 * to another.
2075 	 */
2076 	error = xfs_inobt_get_rec(cur, &rec, &i);
2077 	if (error)
2078 		goto error;
2079 	XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
2080 
2081 	rec.ir_free |= XFS_INOBT_MASK(offset);
2082 	rec.ir_freecount++;
2083 
2084 	XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
2085 				(rec.ir_freecount == ibtrec->ir_freecount),
2086 				error);
2087 
2088 	/*
2089 	 * The content of inobt records should always match between the inobt
2090 	 * and finobt. The lifecycle of records in the finobt is different from
2091 	 * the inobt in that the finobt only tracks records with at least one
2092 	 * free inode. Hence, if all of the inodes are free and we aren't
2093 	 * keeping inode chunks permanently on disk, remove the record.
2094 	 * Otherwise, update the record with the new information.
2095 	 *
2096 	 * Note that we currently can't free chunks when the block size is large
2097 	 * enough for multiple chunks. Leave the finobt record to remain in sync
2098 	 * with the inobt.
2099 	 */
2100 	if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2101 	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
2102 	    !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2103 		error = xfs_btree_delete(cur, &i);
2104 		if (error)
2105 			goto error;
2106 		ASSERT(i == 1);
2107 	} else {
2108 		error = xfs_inobt_update(cur, &rec);
2109 		if (error)
2110 			goto error;
2111 	}
2112 
2113 out:
2114 	error = xfs_check_agi_freecount(cur, agi);
2115 	if (error)
2116 		goto error;
2117 
2118 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2119 	return 0;
2120 
2121 error:
2122 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2123 	return error;
2124 }
2125 
2126 /*
2127  * Free disk inode.  Carefully avoids touching the incore inode, all
2128  * manipulations incore are the caller's responsibility.
2129  * The on-disk inode is not changed by this operation, only the
2130  * btree (free inode mask) is changed.
2131  */
2132 int
2133 xfs_difree(
2134 	struct xfs_trans	*tp,		/* transaction pointer */
2135 	xfs_ino_t		inode,		/* inode to be freed */
2136 	struct xfs_defer_ops	*dfops,		/* extents to free */
2137 	struct xfs_icluster	*xic)	/* cluster info if deleted */
2138 {
2139 	/* REFERENCED */
2140 	xfs_agblock_t		agbno;	/* block number containing inode */
2141 	struct xfs_buf		*agbp;	/* buffer for allocation group header */
2142 	xfs_agino_t		agino;	/* allocation group inode number */
2143 	xfs_agnumber_t		agno;	/* allocation group number */
2144 	int			error;	/* error return value */
2145 	struct xfs_mount	*mp;	/* mount structure for filesystem */
2146 	struct xfs_inobt_rec_incore rec;/* btree record */
2147 
2148 	mp = tp->t_mountp;
2149 
2150 	/*
2151 	 * Break up inode number into its components.
2152 	 */
2153 	agno = XFS_INO_TO_AGNO(mp, inode);
2154 	if (agno >= mp->m_sb.sb_agcount)  {
2155 		xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2156 			__func__, agno, mp->m_sb.sb_agcount);
2157 		ASSERT(0);
2158 		return -EINVAL;
2159 	}
2160 	agino = XFS_INO_TO_AGINO(mp, inode);
2161 	if (inode != XFS_AGINO_TO_INO(mp, agno, agino))  {
2162 		xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2163 			__func__, (unsigned long long)inode,
2164 			(unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2165 		ASSERT(0);
2166 		return -EINVAL;
2167 	}
2168 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2169 	if (agbno >= mp->m_sb.sb_agblocks)  {
2170 		xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2171 			__func__, agbno, mp->m_sb.sb_agblocks);
2172 		ASSERT(0);
2173 		return -EINVAL;
2174 	}
2175 	/*
2176 	 * Get the allocation group header.
2177 	 */
2178 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2179 	if (error) {
2180 		xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2181 			__func__, error);
2182 		return error;
2183 	}
2184 
2185 	/*
2186 	 * Fix up the inode allocation btree.
2187 	 */
2188 	error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec);
2189 	if (error)
2190 		goto error0;
2191 
2192 	/*
2193 	 * Fix up the free inode btree.
2194 	 */
2195 	if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2196 		error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2197 		if (error)
2198 			goto error0;
2199 	}
2200 
2201 	return 0;
2202 
2203 error0:
2204 	return error;
2205 }
2206 
2207 STATIC int
2208 xfs_imap_lookup(
2209 	struct xfs_mount	*mp,
2210 	struct xfs_trans	*tp,
2211 	xfs_agnumber_t		agno,
2212 	xfs_agino_t		agino,
2213 	xfs_agblock_t		agbno,
2214 	xfs_agblock_t		*chunk_agbno,
2215 	xfs_agblock_t		*offset_agbno,
2216 	int			flags)
2217 {
2218 	struct xfs_inobt_rec_incore rec;
2219 	struct xfs_btree_cur	*cur;
2220 	struct xfs_buf		*agbp;
2221 	int			error;
2222 	int			i;
2223 
2224 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2225 	if (error) {
2226 		xfs_alert(mp,
2227 			"%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2228 			__func__, error, agno);
2229 		return error;
2230 	}
2231 
2232 	/*
2233 	 * Lookup the inode record for the given agino. If the record cannot be
2234 	 * found, then it's an invalid inode number and we should abort. Once
2235 	 * we have a record, we need to ensure it contains the inode number
2236 	 * we are looking up.
2237 	 */
2238 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
2239 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2240 	if (!error) {
2241 		if (i)
2242 			error = xfs_inobt_get_rec(cur, &rec, &i);
2243 		if (!error && i == 0)
2244 			error = -EINVAL;
2245 	}
2246 
2247 	xfs_trans_brelse(tp, agbp);
2248 	xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
2249 	if (error)
2250 		return error;
2251 
2252 	/* check that the returned record contains the required inode */
2253 	if (rec.ir_startino > agino ||
2254 	    rec.ir_startino + mp->m_ialloc_inos <= agino)
2255 		return -EINVAL;
2256 
2257 	/* for untrusted inodes check it is allocated first */
2258 	if ((flags & XFS_IGET_UNTRUSTED) &&
2259 	    (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2260 		return -EINVAL;
2261 
2262 	*chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2263 	*offset_agbno = agbno - *chunk_agbno;
2264 	return 0;
2265 }
2266 
2267 /*
2268  * Return the location of the inode in imap, for mapping it into a buffer.
2269  */
2270 int
2271 xfs_imap(
2272 	xfs_mount_t	 *mp,	/* file system mount structure */
2273 	xfs_trans_t	 *tp,	/* transaction pointer */
2274 	xfs_ino_t	ino,	/* inode to locate */
2275 	struct xfs_imap	*imap,	/* location map structure */
2276 	uint		flags)	/* flags for inode btree lookup */
2277 {
2278 	xfs_agblock_t	agbno;	/* block number of inode in the alloc group */
2279 	xfs_agino_t	agino;	/* inode number within alloc group */
2280 	xfs_agnumber_t	agno;	/* allocation group number */
2281 	int		blks_per_cluster; /* num blocks per inode cluster */
2282 	xfs_agblock_t	chunk_agbno;	/* first block in inode chunk */
2283 	xfs_agblock_t	cluster_agbno;	/* first block in inode cluster */
2284 	int		error;	/* error code */
2285 	int		offset;	/* index of inode in its buffer */
2286 	xfs_agblock_t	offset_agbno;	/* blks from chunk start to inode */
2287 
2288 	ASSERT(ino != NULLFSINO);
2289 
2290 	/*
2291 	 * Split up the inode number into its parts.
2292 	 */
2293 	agno = XFS_INO_TO_AGNO(mp, ino);
2294 	agino = XFS_INO_TO_AGINO(mp, ino);
2295 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2296 	if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2297 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2298 #ifdef DEBUG
2299 		/*
2300 		 * Don't output diagnostic information for untrusted inodes
2301 		 * as they can be invalid without implying corruption.
2302 		 */
2303 		if (flags & XFS_IGET_UNTRUSTED)
2304 			return -EINVAL;
2305 		if (agno >= mp->m_sb.sb_agcount) {
2306 			xfs_alert(mp,
2307 				"%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2308 				__func__, agno, mp->m_sb.sb_agcount);
2309 		}
2310 		if (agbno >= mp->m_sb.sb_agblocks) {
2311 			xfs_alert(mp,
2312 		"%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2313 				__func__, (unsigned long long)agbno,
2314 				(unsigned long)mp->m_sb.sb_agblocks);
2315 		}
2316 		if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2317 			xfs_alert(mp,
2318 		"%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2319 				__func__, ino,
2320 				XFS_AGINO_TO_INO(mp, agno, agino));
2321 		}
2322 		xfs_stack_trace();
2323 #endif /* DEBUG */
2324 		return -EINVAL;
2325 	}
2326 
2327 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2328 
2329 	/*
2330 	 * For bulkstat and handle lookups, we have an untrusted inode number
2331 	 * that we have to verify is valid. We cannot do this just by reading
2332 	 * the inode buffer as it may have been unlinked and removed leaving
2333 	 * inodes in stale state on disk. Hence we have to do a btree lookup
2334 	 * in all cases where an untrusted inode number is passed.
2335 	 */
2336 	if (flags & XFS_IGET_UNTRUSTED) {
2337 		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2338 					&chunk_agbno, &offset_agbno, flags);
2339 		if (error)
2340 			return error;
2341 		goto out_map;
2342 	}
2343 
2344 	/*
2345 	 * If the inode cluster size is the same as the blocksize or
2346 	 * smaller we get to the buffer by simple arithmetics.
2347 	 */
2348 	if (blks_per_cluster == 1) {
2349 		offset = XFS_INO_TO_OFFSET(mp, ino);
2350 		ASSERT(offset < mp->m_sb.sb_inopblock);
2351 
2352 		imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2353 		imap->im_len = XFS_FSB_TO_BB(mp, 1);
2354 		imap->im_boffset = (unsigned short)(offset <<
2355 							mp->m_sb.sb_inodelog);
2356 		return 0;
2357 	}
2358 
2359 	/*
2360 	 * If the inode chunks are aligned then use simple maths to
2361 	 * find the location. Otherwise we have to do a btree
2362 	 * lookup to find the location.
2363 	 */
2364 	if (mp->m_inoalign_mask) {
2365 		offset_agbno = agbno & mp->m_inoalign_mask;
2366 		chunk_agbno = agbno - offset_agbno;
2367 	} else {
2368 		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2369 					&chunk_agbno, &offset_agbno, flags);
2370 		if (error)
2371 			return error;
2372 	}
2373 
2374 out_map:
2375 	ASSERT(agbno >= chunk_agbno);
2376 	cluster_agbno = chunk_agbno +
2377 		((offset_agbno / blks_per_cluster) * blks_per_cluster);
2378 	offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2379 		XFS_INO_TO_OFFSET(mp, ino);
2380 
2381 	imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2382 	imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
2383 	imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2384 
2385 	/*
2386 	 * If the inode number maps to a block outside the bounds
2387 	 * of the file system then return NULL rather than calling
2388 	 * read_buf and panicing when we get an error from the
2389 	 * driver.
2390 	 */
2391 	if ((imap->im_blkno + imap->im_len) >
2392 	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2393 		xfs_alert(mp,
2394 	"%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2395 			__func__, (unsigned long long) imap->im_blkno,
2396 			(unsigned long long) imap->im_len,
2397 			XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2398 		return -EINVAL;
2399 	}
2400 	return 0;
2401 }
2402 
2403 /*
2404  * Compute and fill in value of m_in_maxlevels.
2405  */
2406 void
2407 xfs_ialloc_compute_maxlevels(
2408 	xfs_mount_t	*mp)		/* file system mount structure */
2409 {
2410 	uint		inodes;
2411 
2412 	inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2413 	mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr,
2414 							 inodes);
2415 }
2416 
2417 /*
2418  * Log specified fields for the ag hdr (inode section). The growth of the agi
2419  * structure over time requires that we interpret the buffer as two logical
2420  * regions delineated by the end of the unlinked list. This is due to the size
2421  * of the hash table and its location in the middle of the agi.
2422  *
2423  * For example, a request to log a field before agi_unlinked and a field after
2424  * agi_unlinked could cause us to log the entire hash table and use an excessive
2425  * amount of log space. To avoid this behavior, log the region up through
2426  * agi_unlinked in one call and the region after agi_unlinked through the end of
2427  * the structure in another.
2428  */
2429 void
2430 xfs_ialloc_log_agi(
2431 	xfs_trans_t	*tp,		/* transaction pointer */
2432 	xfs_buf_t	*bp,		/* allocation group header buffer */
2433 	int		fields)		/* bitmask of fields to log */
2434 {
2435 	int			first;		/* first byte number */
2436 	int			last;		/* last byte number */
2437 	static const short	offsets[] = {	/* field starting offsets */
2438 					/* keep in sync with bit definitions */
2439 		offsetof(xfs_agi_t, agi_magicnum),
2440 		offsetof(xfs_agi_t, agi_versionnum),
2441 		offsetof(xfs_agi_t, agi_seqno),
2442 		offsetof(xfs_agi_t, agi_length),
2443 		offsetof(xfs_agi_t, agi_count),
2444 		offsetof(xfs_agi_t, agi_root),
2445 		offsetof(xfs_agi_t, agi_level),
2446 		offsetof(xfs_agi_t, agi_freecount),
2447 		offsetof(xfs_agi_t, agi_newino),
2448 		offsetof(xfs_agi_t, agi_dirino),
2449 		offsetof(xfs_agi_t, agi_unlinked),
2450 		offsetof(xfs_agi_t, agi_free_root),
2451 		offsetof(xfs_agi_t, agi_free_level),
2452 		sizeof(xfs_agi_t)
2453 	};
2454 #ifdef DEBUG
2455 	xfs_agi_t		*agi;	/* allocation group header */
2456 
2457 	agi = XFS_BUF_TO_AGI(bp);
2458 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2459 #endif
2460 
2461 	/*
2462 	 * Compute byte offsets for the first and last fields in the first
2463 	 * region and log the agi buffer. This only logs up through
2464 	 * agi_unlinked.
2465 	 */
2466 	if (fields & XFS_AGI_ALL_BITS_R1) {
2467 		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2468 				  &first, &last);
2469 		xfs_trans_log_buf(tp, bp, first, last);
2470 	}
2471 
2472 	/*
2473 	 * Mask off the bits in the first region and calculate the first and
2474 	 * last field offsets for any bits in the second region.
2475 	 */
2476 	fields &= ~XFS_AGI_ALL_BITS_R1;
2477 	if (fields) {
2478 		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2479 				  &first, &last);
2480 		xfs_trans_log_buf(tp, bp, first, last);
2481 	}
2482 }
2483 
2484 #ifdef DEBUG
2485 STATIC void
2486 xfs_check_agi_unlinked(
2487 	struct xfs_agi		*agi)
2488 {
2489 	int			i;
2490 
2491 	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2492 		ASSERT(agi->agi_unlinked[i]);
2493 }
2494 #else
2495 #define xfs_check_agi_unlinked(agi)
2496 #endif
2497 
2498 static bool
2499 xfs_agi_verify(
2500 	struct xfs_buf	*bp)
2501 {
2502 	struct xfs_mount *mp = bp->b_target->bt_mount;
2503 	struct xfs_agi	*agi = XFS_BUF_TO_AGI(bp);
2504 
2505 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
2506 		if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2507 			return false;
2508 		if (!xfs_log_check_lsn(mp,
2509 				be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn)))
2510 			return false;
2511 	}
2512 
2513 	/*
2514 	 * Validate the magic number of the agi block.
2515 	 */
2516 	if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2517 		return false;
2518 	if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2519 		return false;
2520 
2521 	if (be32_to_cpu(agi->agi_level) < 1 ||
2522 	    be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2523 		return false;
2524 
2525 	if (xfs_sb_version_hasfinobt(&mp->m_sb) &&
2526 	    (be32_to_cpu(agi->agi_free_level) < 1 ||
2527 	     be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS))
2528 		return false;
2529 
2530 	/*
2531 	 * during growfs operations, the perag is not fully initialised,
2532 	 * so we can't use it for any useful checking. growfs ensures we can't
2533 	 * use it by using uncached buffers that don't have the perag attached
2534 	 * so we can detect and avoid this problem.
2535 	 */
2536 	if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2537 		return false;
2538 
2539 	xfs_check_agi_unlinked(agi);
2540 	return true;
2541 }
2542 
2543 static void
2544 xfs_agi_read_verify(
2545 	struct xfs_buf	*bp)
2546 {
2547 	struct xfs_mount *mp = bp->b_target->bt_mount;
2548 
2549 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
2550 	    !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2551 		xfs_buf_ioerror(bp, -EFSBADCRC);
2552 	else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2553 				XFS_ERRTAG_IALLOC_READ_AGI))
2554 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
2555 
2556 	if (bp->b_error)
2557 		xfs_verifier_error(bp);
2558 }
2559 
2560 static void
2561 xfs_agi_write_verify(
2562 	struct xfs_buf	*bp)
2563 {
2564 	struct xfs_mount *mp = bp->b_target->bt_mount;
2565 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
2566 
2567 	if (!xfs_agi_verify(bp)) {
2568 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
2569 		xfs_verifier_error(bp);
2570 		return;
2571 	}
2572 
2573 	if (!xfs_sb_version_hascrc(&mp->m_sb))
2574 		return;
2575 
2576 	if (bip)
2577 		XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2578 	xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2579 }
2580 
2581 const struct xfs_buf_ops xfs_agi_buf_ops = {
2582 	.name = "xfs_agi",
2583 	.verify_read = xfs_agi_read_verify,
2584 	.verify_write = xfs_agi_write_verify,
2585 };
2586 
2587 /*
2588  * Read in the allocation group header (inode allocation section)
2589  */
2590 int
2591 xfs_read_agi(
2592 	struct xfs_mount	*mp,	/* file system mount structure */
2593 	struct xfs_trans	*tp,	/* transaction pointer */
2594 	xfs_agnumber_t		agno,	/* allocation group number */
2595 	struct xfs_buf		**bpp)	/* allocation group hdr buf */
2596 {
2597 	int			error;
2598 
2599 	trace_xfs_read_agi(mp, agno);
2600 
2601 	ASSERT(agno != NULLAGNUMBER);
2602 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2603 			XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2604 			XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2605 	if (error)
2606 		return error;
2607 	if (tp)
2608 		xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF);
2609 
2610 	xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2611 	return 0;
2612 }
2613 
2614 int
2615 xfs_ialloc_read_agi(
2616 	struct xfs_mount	*mp,	/* file system mount structure */
2617 	struct xfs_trans	*tp,	/* transaction pointer */
2618 	xfs_agnumber_t		agno,	/* allocation group number */
2619 	struct xfs_buf		**bpp)	/* allocation group hdr buf */
2620 {
2621 	struct xfs_agi		*agi;	/* allocation group header */
2622 	struct xfs_perag	*pag;	/* per allocation group data */
2623 	int			error;
2624 
2625 	trace_xfs_ialloc_read_agi(mp, agno);
2626 
2627 	error = xfs_read_agi(mp, tp, agno, bpp);
2628 	if (error)
2629 		return error;
2630 
2631 	agi = XFS_BUF_TO_AGI(*bpp);
2632 	pag = xfs_perag_get(mp, agno);
2633 	if (!pag->pagi_init) {
2634 		pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2635 		pag->pagi_count = be32_to_cpu(agi->agi_count);
2636 		pag->pagi_init = 1;
2637 	}
2638 
2639 	/*
2640 	 * It's possible for these to be out of sync if
2641 	 * we are in the middle of a forced shutdown.
2642 	 */
2643 	ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2644 		XFS_FORCED_SHUTDOWN(mp));
2645 	xfs_perag_put(pag);
2646 	return 0;
2647 }
2648 
2649 /*
2650  * Read in the agi to initialise the per-ag data in the mount structure
2651  */
2652 int
2653 xfs_ialloc_pagi_init(
2654 	xfs_mount_t	*mp,		/* file system mount structure */
2655 	xfs_trans_t	*tp,		/* transaction pointer */
2656 	xfs_agnumber_t	agno)		/* allocation group number */
2657 {
2658 	xfs_buf_t	*bp = NULL;
2659 	int		error;
2660 
2661 	error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2662 	if (error)
2663 		return error;
2664 	if (bp)
2665 		xfs_trans_brelse(tp, bp);
2666 	return 0;
2667 }
2668 
2669 /* Calculate the first and last possible inode number in an AG. */
2670 void
2671 xfs_ialloc_agino_range(
2672 	struct xfs_mount	*mp,
2673 	xfs_agnumber_t		agno,
2674 	xfs_agino_t		*first,
2675 	xfs_agino_t		*last)
2676 {
2677 	xfs_agblock_t		bno;
2678 	xfs_agblock_t		eoag;
2679 
2680 	eoag = xfs_ag_block_count(mp, agno);
2681 
2682 	/*
2683 	 * Calculate the first inode, which will be in the first
2684 	 * cluster-aligned block after the AGFL.
2685 	 */
2686 	bno = round_up(XFS_AGFL_BLOCK(mp) + 1,
2687 			xfs_ialloc_cluster_alignment(mp));
2688 	*first = XFS_OFFBNO_TO_AGINO(mp, bno, 0);
2689 
2690 	/*
2691 	 * Calculate the last inode, which will be at the end of the
2692 	 * last (aligned) cluster that can be allocated in the AG.
2693 	 */
2694 	bno = round_down(eoag, xfs_ialloc_cluster_alignment(mp));
2695 	*last = XFS_OFFBNO_TO_AGINO(mp, bno, 0) - 1;
2696 }
2697 
2698 /*
2699  * Verify that an AG inode number pointer neither points outside the AG
2700  * nor points at static metadata.
2701  */
2702 bool
2703 xfs_verify_agino(
2704 	struct xfs_mount	*mp,
2705 	xfs_agnumber_t		agno,
2706 	xfs_agino_t		agino)
2707 {
2708 	xfs_agino_t		first;
2709 	xfs_agino_t		last;
2710 
2711 	xfs_ialloc_agino_range(mp, agno, &first, &last);
2712 	return agino >= first && agino <= last;
2713 }
2714 
2715 /*
2716  * Verify that an FS inode number pointer neither points outside the
2717  * filesystem nor points at static AG metadata.
2718  */
2719 bool
2720 xfs_verify_ino(
2721 	struct xfs_mount	*mp,
2722 	xfs_ino_t		ino)
2723 {
2724 	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, ino);
2725 	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ino);
2726 
2727 	if (agno >= mp->m_sb.sb_agcount)
2728 		return false;
2729 	if (XFS_AGINO_TO_INO(mp, agno, agino) != ino)
2730 		return false;
2731 	return xfs_verify_agino(mp, agno, agino);
2732 }
2733 
2734 /* Is this an internal inode number? */
2735 bool
2736 xfs_internal_inum(
2737 	struct xfs_mount	*mp,
2738 	xfs_ino_t		ino)
2739 {
2740 	return ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
2741 		(xfs_sb_version_hasquota(&mp->m_sb) &&
2742 		 xfs_is_quota_inode(&mp->m_sb, ino));
2743 }
2744 
2745 /*
2746  * Verify that a directory entry's inode number doesn't point at an internal
2747  * inode, empty space, or static AG metadata.
2748  */
2749 bool
2750 xfs_verify_dir_ino(
2751 	struct xfs_mount	*mp,
2752 	xfs_ino_t		ino)
2753 {
2754 	if (xfs_internal_inum(mp, ino))
2755 		return false;
2756 	return xfs_verify_ino(mp, ino);
2757 }
2758