xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 94c7b6fc)
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_inum.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.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_itable.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_dinode.h"
37 
38 STATIC int
39 xfs_internal_inum(
40 	xfs_mount_t	*mp,
41 	xfs_ino_t	ino)
42 {
43 	return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
44 		(xfs_sb_version_hasquota(&mp->m_sb) &&
45 		 xfs_is_quota_inode(&mp->m_sb, ino)));
46 }
47 
48 /*
49  * Return stat information for one inode.
50  * Return 0 if ok, else errno.
51  */
52 int
53 xfs_bulkstat_one_int(
54 	struct xfs_mount	*mp,		/* mount point for filesystem */
55 	xfs_ino_t		ino,		/* inode to get data for */
56 	void __user		*buffer,	/* buffer to place output in */
57 	int			ubsize,		/* size of buffer */
58 	bulkstat_one_fmt_pf	formatter,	/* formatter, copy to user */
59 	int			*ubused,	/* bytes used by me */
60 	int			*stat)		/* BULKSTAT_RV_... */
61 {
62 	struct xfs_icdinode	*dic;		/* dinode core info pointer */
63 	struct xfs_inode	*ip;		/* incore inode pointer */
64 	struct xfs_bstat	*buf;		/* return buffer */
65 	int			error = 0;	/* error value */
66 
67 	*stat = BULKSTAT_RV_NOTHING;
68 
69 	if (!buffer || xfs_internal_inum(mp, ino))
70 		return XFS_ERROR(EINVAL);
71 
72 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP | KM_MAYFAIL);
73 	if (!buf)
74 		return XFS_ERROR(ENOMEM);
75 
76 	error = xfs_iget(mp, NULL, ino,
77 			 (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED),
78 			 XFS_ILOCK_SHARED, &ip);
79 	if (error) {
80 		*stat = BULKSTAT_RV_NOTHING;
81 		goto out_free;
82 	}
83 
84 	ASSERT(ip != NULL);
85 	ASSERT(ip->i_imap.im_blkno != 0);
86 
87 	dic = &ip->i_d;
88 
89 	/* xfs_iget returns the following without needing
90 	 * further change.
91 	 */
92 	buf->bs_nlink = dic->di_nlink;
93 	buf->bs_projid_lo = dic->di_projid_lo;
94 	buf->bs_projid_hi = dic->di_projid_hi;
95 	buf->bs_ino = ino;
96 	buf->bs_mode = dic->di_mode;
97 	buf->bs_uid = dic->di_uid;
98 	buf->bs_gid = dic->di_gid;
99 	buf->bs_size = dic->di_size;
100 	buf->bs_atime.tv_sec = dic->di_atime.t_sec;
101 	buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
102 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
103 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
104 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
105 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
106 	buf->bs_xflags = xfs_ip2xflags(ip);
107 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
108 	buf->bs_extents = dic->di_nextents;
109 	buf->bs_gen = dic->di_gen;
110 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
111 	buf->bs_dmevmask = dic->di_dmevmask;
112 	buf->bs_dmstate = dic->di_dmstate;
113 	buf->bs_aextents = dic->di_anextents;
114 	buf->bs_forkoff = XFS_IFORK_BOFF(ip);
115 
116 	switch (dic->di_format) {
117 	case XFS_DINODE_FMT_DEV:
118 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
119 		buf->bs_blksize = BLKDEV_IOSIZE;
120 		buf->bs_blocks = 0;
121 		break;
122 	case XFS_DINODE_FMT_LOCAL:
123 	case XFS_DINODE_FMT_UUID:
124 		buf->bs_rdev = 0;
125 		buf->bs_blksize = mp->m_sb.sb_blocksize;
126 		buf->bs_blocks = 0;
127 		break;
128 	case XFS_DINODE_FMT_EXTENTS:
129 	case XFS_DINODE_FMT_BTREE:
130 		buf->bs_rdev = 0;
131 		buf->bs_blksize = mp->m_sb.sb_blocksize;
132 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
133 		break;
134 	}
135 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
136 	IRELE(ip);
137 
138 	error = formatter(buffer, ubsize, ubused, buf);
139 
140 	if (!error)
141 		*stat = BULKSTAT_RV_DIDONE;
142 
143  out_free:
144 	kmem_free(buf);
145 	return error;
146 }
147 
148 /* Return 0 on success or positive error */
149 STATIC int
150 xfs_bulkstat_one_fmt(
151 	void			__user *ubuffer,
152 	int			ubsize,
153 	int			*ubused,
154 	const xfs_bstat_t	*buffer)
155 {
156 	if (ubsize < sizeof(*buffer))
157 		return XFS_ERROR(ENOMEM);
158 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
159 		return XFS_ERROR(EFAULT);
160 	if (ubused)
161 		*ubused = sizeof(*buffer);
162 	return 0;
163 }
164 
165 int
166 xfs_bulkstat_one(
167 	xfs_mount_t	*mp,		/* mount point for filesystem */
168 	xfs_ino_t	ino,		/* inode number to get data for */
169 	void		__user *buffer,	/* buffer to place output in */
170 	int		ubsize,		/* size of buffer */
171 	int		*ubused,	/* bytes used by me */
172 	int		*stat)		/* BULKSTAT_RV_... */
173 {
174 	return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
175 				    xfs_bulkstat_one_fmt, ubused, stat);
176 }
177 
178 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
179 
180 /*
181  * Return stat information in bulk (by-inode) for the filesystem.
182  */
183 int					/* error status */
184 xfs_bulkstat(
185 	xfs_mount_t		*mp,	/* mount point for filesystem */
186 	xfs_ino_t		*lastinop, /* last inode returned */
187 	int			*ubcountp, /* size of buffer/count returned */
188 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
189 	size_t			statstruct_size, /* sizeof struct filling */
190 	char			__user *ubuffer, /* buffer with inode stats */
191 	int			*done)	/* 1 if there are more stats to get */
192 {
193 	xfs_agblock_t		agbno=0;/* allocation group block number */
194 	xfs_buf_t		*agbp;	/* agi header buffer */
195 	xfs_agi_t		*agi;	/* agi header data */
196 	xfs_agino_t		agino;	/* inode # in allocation group */
197 	xfs_agnumber_t		agno;	/* allocation group number */
198 	int			chunkidx; /* current index into inode chunk */
199 	int			clustidx; /* current index into inode cluster */
200 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
201 	int			end_of_ag; /* set if we've seen the ag end */
202 	int			error;	/* error code */
203 	int                     fmterror;/* bulkstat formatter result */
204 	int			i;	/* loop index */
205 	int			icount;	/* count of inodes good in irbuf */
206 	size_t			irbsize; /* size of irec buffer in bytes */
207 	xfs_ino_t		ino;	/* inode number (filesystem) */
208 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
209 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
210 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
211 	xfs_ino_t		lastino; /* last inode number returned */
212 	int			blks_per_cluster; /* # of blocks per cluster */
213 	int			inodes_per_cluster;/* # of inodes per cluster */
214 	int			nirbuf;	/* size of irbuf */
215 	int			rval;	/* return value error code */
216 	int			tmp;	/* result value from btree calls */
217 	int			ubcount; /* size of user's buffer */
218 	int			ubleft;	/* bytes left in user's buffer */
219 	char			__user *ubufp;	/* pointer into user's buffer */
220 	int			ubelem;	/* spaces used in user's buffer */
221 	int			ubused;	/* bytes used by formatter */
222 
223 	/*
224 	 * Get the last inode value, see if there's nothing to do.
225 	 */
226 	ino = (xfs_ino_t)*lastinop;
227 	lastino = ino;
228 	agno = XFS_INO_TO_AGNO(mp, ino);
229 	agino = XFS_INO_TO_AGINO(mp, ino);
230 	if (agno >= mp->m_sb.sb_agcount ||
231 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
232 		*done = 1;
233 		*ubcountp = 0;
234 		return 0;
235 	}
236 	if (!ubcountp || *ubcountp <= 0) {
237 		return EINVAL;
238 	}
239 	ubcount = *ubcountp; /* statstruct's */
240 	ubleft = ubcount * statstruct_size; /* bytes */
241 	*ubcountp = ubelem = 0;
242 	*done = 0;
243 	fmterror = 0;
244 	ubufp = ubuffer;
245 	blks_per_cluster = xfs_icluster_size_fsb(mp);
246 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
247 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4);
248 	if (!irbuf)
249 		return ENOMEM;
250 
251 	nirbuf = irbsize / sizeof(*irbuf);
252 
253 	/*
254 	 * Loop over the allocation groups, starting from the last
255 	 * inode returned; 0 means start of the allocation group.
256 	 */
257 	rval = 0;
258 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
259 		cond_resched();
260 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
261 		if (error) {
262 			/*
263 			 * Skip this allocation group and go to the next one.
264 			 */
265 			agno++;
266 			agino = 0;
267 			continue;
268 		}
269 		agi = XFS_BUF_TO_AGI(agbp);
270 		/*
271 		 * Allocate and initialize a btree cursor for ialloc btree.
272 		 */
273 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno,
274 					    XFS_BTNUM_INO);
275 		irbp = irbuf;
276 		irbufend = irbuf + nirbuf;
277 		end_of_ag = 0;
278 		/*
279 		 * If we're returning in the middle of an allocation group,
280 		 * we need to get the remainder of the chunk we're in.
281 		 */
282 		if (agino > 0) {
283 			xfs_inobt_rec_incore_t r;
284 
285 			/*
286 			 * Lookup the inode chunk that this inode lives in.
287 			 */
288 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
289 						 &tmp);
290 			if (!error &&	/* no I/O error */
291 			    tmp &&	/* lookup succeeded */
292 					/* got the record, should always work */
293 			    !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
294 			    i == 1 &&
295 					/* this is the right chunk */
296 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
297 					/* lastino was not last in chunk */
298 			    (chunkidx = agino - r.ir_startino + 1) <
299 				    XFS_INODES_PER_CHUNK &&
300 					/* there are some left allocated */
301 			    xfs_inobt_maskn(chunkidx,
302 				    XFS_INODES_PER_CHUNK - chunkidx) &
303 				    ~r.ir_free) {
304 				/*
305 				 * Grab the chunk record.  Mark all the
306 				 * uninteresting inodes (because they're
307 				 * before our start point) free.
308 				 */
309 				for (i = 0; i < chunkidx; i++) {
310 					if (XFS_INOBT_MASK(i) & ~r.ir_free)
311 						r.ir_freecount++;
312 				}
313 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
314 				irbp->ir_startino = r.ir_startino;
315 				irbp->ir_freecount = r.ir_freecount;
316 				irbp->ir_free = r.ir_free;
317 				irbp++;
318 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
319 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
320 			} else {
321 				/*
322 				 * If any of those tests failed, bump the
323 				 * inode number (just in case).
324 				 */
325 				agino++;
326 				icount = 0;
327 			}
328 			/*
329 			 * In any case, increment to the next record.
330 			 */
331 			if (!error)
332 				error = xfs_btree_increment(cur, 0, &tmp);
333 		} else {
334 			/*
335 			 * Start of ag.  Lookup the first inode chunk.
336 			 */
337 			error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
338 			icount = 0;
339 		}
340 		/*
341 		 * Loop through inode btree records in this ag,
342 		 * until we run out of inodes or space in the buffer.
343 		 */
344 		while (irbp < irbufend && icount < ubcount) {
345 			xfs_inobt_rec_incore_t r;
346 
347 			/*
348 			 * Loop as long as we're unable to read the
349 			 * inode btree.
350 			 */
351 			while (error) {
352 				agino += XFS_INODES_PER_CHUNK;
353 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
354 						be32_to_cpu(agi->agi_length))
355 					break;
356 				error = xfs_inobt_lookup(cur, agino,
357 							 XFS_LOOKUP_GE, &tmp);
358 				cond_resched();
359 			}
360 			/*
361 			 * If ran off the end of the ag either with an error,
362 			 * or the normal way, set end and stop collecting.
363 			 */
364 			if (error) {
365 				end_of_ag = 1;
366 				break;
367 			}
368 
369 			error = xfs_inobt_get_rec(cur, &r, &i);
370 			if (error || i == 0) {
371 				end_of_ag = 1;
372 				break;
373 			}
374 
375 			/*
376 			 * If this chunk has any allocated inodes, save it.
377 			 * Also start read-ahead now for this chunk.
378 			 */
379 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
380 				struct blk_plug	plug;
381 				/*
382 				 * Loop over all clusters in the next chunk.
383 				 * Do a readahead if there are any allocated
384 				 * inodes in that cluster.
385 				 */
386 				blk_start_plug(&plug);
387 				agbno = XFS_AGINO_TO_AGBNO(mp, r.ir_startino);
388 				for (chunkidx = 0;
389 				     chunkidx < XFS_INODES_PER_CHUNK;
390 				     chunkidx += inodes_per_cluster,
391 				     agbno += blks_per_cluster) {
392 					if (xfs_inobt_maskn(chunkidx,
393 					    inodes_per_cluster) & ~r.ir_free)
394 						xfs_btree_reada_bufs(mp, agno,
395 							agbno, blks_per_cluster,
396 							&xfs_inode_buf_ops);
397 				}
398 				blk_finish_plug(&plug);
399 				irbp->ir_startino = r.ir_startino;
400 				irbp->ir_freecount = r.ir_freecount;
401 				irbp->ir_free = r.ir_free;
402 				irbp++;
403 				icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
404 			}
405 			/*
406 			 * Set agino to after this chunk and bump the cursor.
407 			 */
408 			agino = r.ir_startino + XFS_INODES_PER_CHUNK;
409 			error = xfs_btree_increment(cur, 0, &tmp);
410 			cond_resched();
411 		}
412 		/*
413 		 * Drop the btree buffers and the agi buffer.
414 		 * We can't hold any of the locks these represent
415 		 * when calling iget.
416 		 */
417 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
418 		xfs_buf_relse(agbp);
419 		/*
420 		 * Now format all the good inodes into the user's buffer.
421 		 */
422 		irbufend = irbp;
423 		for (irbp = irbuf;
424 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
425 			/*
426 			 * Now process this chunk of inodes.
427 			 */
428 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
429 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
430 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
431 			     chunkidx++, clustidx++, agino++) {
432 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
433 
434 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
435 				/*
436 				 * Skip if this inode is free.
437 				 */
438 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
439 					lastino = ino;
440 					continue;
441 				}
442 				/*
443 				 * Count used inodes as free so we can tell
444 				 * when the chunk is used up.
445 				 */
446 				irbp->ir_freecount++;
447 
448 				/*
449 				 * Get the inode and fill in a single buffer.
450 				 */
451 				ubused = statstruct_size;
452 				error = formatter(mp, ino, ubufp, ubleft,
453 						  &ubused, &fmterror);
454 				if (fmterror == BULKSTAT_RV_NOTHING) {
455 					if (error && error != ENOENT &&
456 						error != EINVAL) {
457 						ubleft = 0;
458 						rval = error;
459 						break;
460 					}
461 					lastino = ino;
462 					continue;
463 				}
464 				if (fmterror == BULKSTAT_RV_GIVEUP) {
465 					ubleft = 0;
466 					ASSERT(error);
467 					rval = error;
468 					break;
469 				}
470 				if (ubufp)
471 					ubufp += ubused;
472 				ubleft -= ubused;
473 				ubelem++;
474 				lastino = ino;
475 			}
476 
477 			cond_resched();
478 		}
479 		/*
480 		 * Set up for the next loop iteration.
481 		 */
482 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
483 			if (end_of_ag) {
484 				agno++;
485 				agino = 0;
486 			} else
487 				agino = XFS_INO_TO_AGINO(mp, lastino);
488 		} else
489 			break;
490 	}
491 	/*
492 	 * Done, we're either out of filesystem or space to put the data.
493 	 */
494 	kmem_free(irbuf);
495 	*ubcountp = ubelem;
496 	/*
497 	 * Found some inodes, return them now and return the error next time.
498 	 */
499 	if (ubelem)
500 		rval = 0;
501 	if (agno >= mp->m_sb.sb_agcount) {
502 		/*
503 		 * If we ran out of filesystem, mark lastino as off
504 		 * the end of the filesystem, so the next call
505 		 * will return immediately.
506 		 */
507 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
508 		*done = 1;
509 	} else
510 		*lastinop = (xfs_ino_t)lastino;
511 
512 	return rval;
513 }
514 
515 /*
516  * Return stat information in bulk (by-inode) for the filesystem.
517  * Special case for non-sequential one inode bulkstat.
518  */
519 int					/* error status */
520 xfs_bulkstat_single(
521 	xfs_mount_t		*mp,	/* mount point for filesystem */
522 	xfs_ino_t		*lastinop, /* inode to return */
523 	char			__user *buffer, /* buffer with inode stats */
524 	int			*done)	/* 1 if there are more stats to get */
525 {
526 	int			count;	/* count value for bulkstat call */
527 	int			error;	/* return value */
528 	xfs_ino_t		ino;	/* filesystem inode number */
529 	int			res;	/* result from bs1 */
530 
531 	/*
532 	 * note that requesting valid inode numbers which are not allocated
533 	 * to inodes will most likely cause xfs_imap_to_bp to generate warning
534 	 * messages about bad magic numbers. This is ok. The fact that
535 	 * the inode isn't actually an inode is handled by the
536 	 * error check below. Done this way to make the usual case faster
537 	 * at the expense of the error case.
538 	 */
539 
540 	ino = *lastinop;
541 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
542 				 NULL, &res);
543 	if (error) {
544 		/*
545 		 * Special case way failed, do it the "long" way
546 		 * to see if that works.
547 		 */
548 		(*lastinop)--;
549 		count = 1;
550 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
551 				sizeof(xfs_bstat_t), buffer, done))
552 			return error;
553 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
554 			return error == EFSCORRUPTED ?
555 				XFS_ERROR(EINVAL) : error;
556 		else
557 			return 0;
558 	}
559 	*done = 0;
560 	return 0;
561 }
562 
563 int
564 xfs_inumbers_fmt(
565 	void			__user *ubuffer, /* buffer to write to */
566 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
567 	long			count,		/* # of elements to read */
568 	long			*written)	/* # of bytes written */
569 {
570 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
571 		return -EFAULT;
572 	*written = count * sizeof(*buffer);
573 	return 0;
574 }
575 
576 /*
577  * Return inode number table for the filesystem.
578  */
579 int					/* error status */
580 xfs_inumbers(
581 	xfs_mount_t	*mp,		/* mount point for filesystem */
582 	xfs_ino_t	*lastino,	/* last inode returned */
583 	int		*count,		/* size of buffer/count returned */
584 	void		__user *ubuffer,/* buffer with inode descriptions */
585 	inumbers_fmt_pf	formatter)
586 {
587 	xfs_buf_t	*agbp;
588 	xfs_agino_t	agino;
589 	xfs_agnumber_t	agno;
590 	int		bcount;
591 	xfs_inogrp_t	*buffer;
592 	int		bufidx;
593 	xfs_btree_cur_t	*cur;
594 	int		error;
595 	xfs_inobt_rec_incore_t r;
596 	int		i;
597 	xfs_ino_t	ino;
598 	int		left;
599 	int		tmp;
600 
601 	ino = (xfs_ino_t)*lastino;
602 	agno = XFS_INO_TO_AGNO(mp, ino);
603 	agino = XFS_INO_TO_AGINO(mp, ino);
604 	left = *count;
605 	*count = 0;
606 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
607 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
608 	error = bufidx = 0;
609 	cur = NULL;
610 	agbp = NULL;
611 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
612 		if (agbp == NULL) {
613 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
614 			if (error) {
615 				/*
616 				 * If we can't read the AGI of this ag,
617 				 * then just skip to the next one.
618 				 */
619 				ASSERT(cur == NULL);
620 				agbp = NULL;
621 				agno++;
622 				agino = 0;
623 				continue;
624 			}
625 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno,
626 						    XFS_BTNUM_INO);
627 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
628 						 &tmp);
629 			if (error) {
630 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
631 				cur = NULL;
632 				xfs_buf_relse(agbp);
633 				agbp = NULL;
634 				/*
635 				 * Move up the last inode in the current
636 				 * chunk.  The lookup_ge will always get
637 				 * us the first inode in the next chunk.
638 				 */
639 				agino += XFS_INODES_PER_CHUNK - 1;
640 				continue;
641 			}
642 		}
643 		error = xfs_inobt_get_rec(cur, &r, &i);
644 		if (error || i == 0) {
645 			xfs_buf_relse(agbp);
646 			agbp = NULL;
647 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
648 			cur = NULL;
649 			agno++;
650 			agino = 0;
651 			continue;
652 		}
653 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
654 		buffer[bufidx].xi_startino =
655 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
656 		buffer[bufidx].xi_alloccount =
657 			XFS_INODES_PER_CHUNK - r.ir_freecount;
658 		buffer[bufidx].xi_allocmask = ~r.ir_free;
659 		bufidx++;
660 		left--;
661 		if (bufidx == bcount) {
662 			long written;
663 			if (formatter(ubuffer, buffer, bufidx, &written)) {
664 				error = XFS_ERROR(EFAULT);
665 				break;
666 			}
667 			ubuffer += written;
668 			*count += bufidx;
669 			bufidx = 0;
670 		}
671 		if (left) {
672 			error = xfs_btree_increment(cur, 0, &tmp);
673 			if (error) {
674 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
675 				cur = NULL;
676 				xfs_buf_relse(agbp);
677 				agbp = NULL;
678 				/*
679 				 * The agino value has already been bumped.
680 				 * Just try to skip up to it.
681 				 */
682 				agino += XFS_INODES_PER_CHUNK;
683 				continue;
684 			}
685 		}
686 	}
687 	if (!error) {
688 		if (bufidx) {
689 			long written;
690 			if (formatter(ubuffer, buffer, bufidx, &written))
691 				error = XFS_ERROR(EFAULT);
692 			else
693 				*count += bufidx;
694 		}
695 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
696 	}
697 	kmem_free(buffer);
698 	if (cur)
699 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
700 					   XFS_BTREE_NOERROR));
701 	if (agbp)
702 		xfs_buf_relse(agbp);
703 	return error;
704 }
705