xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 77d84ff8)
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			nbcluster; /* # of blocks in a cluster */
213 	int			nicluster; /* # of inodes in a cluster */
214 	int			nimask;	/* mask for inode clusters */
215 	int			nirbuf;	/* size of irbuf */
216 	int			rval;	/* return value error code */
217 	int			tmp;	/* result value from btree calls */
218 	int			ubcount; /* size of user's buffer */
219 	int			ubleft;	/* bytes left in user's buffer */
220 	char			__user *ubufp;	/* pointer into user's buffer */
221 	int			ubelem;	/* spaces used in user's buffer */
222 	int			ubused;	/* bytes used by formatter */
223 
224 	/*
225 	 * Get the last inode value, see if there's nothing to do.
226 	 */
227 	ino = (xfs_ino_t)*lastinop;
228 	lastino = ino;
229 	agno = XFS_INO_TO_AGNO(mp, ino);
230 	agino = XFS_INO_TO_AGINO(mp, ino);
231 	if (agno >= mp->m_sb.sb_agcount ||
232 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
233 		*done = 1;
234 		*ubcountp = 0;
235 		return 0;
236 	}
237 	if (!ubcountp || *ubcountp <= 0) {
238 		return EINVAL;
239 	}
240 	ubcount = *ubcountp; /* statstruct's */
241 	ubleft = ubcount * statstruct_size; /* bytes */
242 	*ubcountp = ubelem = 0;
243 	*done = 0;
244 	fmterror = 0;
245 	ubufp = ubuffer;
246 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
247 		mp->m_sb.sb_inopblock :
248 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
249 	nimask = ~(nicluster - 1);
250 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
251 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4);
252 	if (!irbuf)
253 		return ENOMEM;
254 
255 	nirbuf = irbsize / sizeof(*irbuf);
256 
257 	/*
258 	 * Loop over the allocation groups, starting from the last
259 	 * inode returned; 0 means start of the allocation group.
260 	 */
261 	rval = 0;
262 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
263 		cond_resched();
264 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
265 		if (error) {
266 			/*
267 			 * Skip this allocation group and go to the next one.
268 			 */
269 			agno++;
270 			agino = 0;
271 			continue;
272 		}
273 		agi = XFS_BUF_TO_AGI(agbp);
274 		/*
275 		 * Allocate and initialize a btree cursor for ialloc btree.
276 		 */
277 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
278 		irbp = irbuf;
279 		irbufend = irbuf + nirbuf;
280 		end_of_ag = 0;
281 		/*
282 		 * If we're returning in the middle of an allocation group,
283 		 * we need to get the remainder of the chunk we're in.
284 		 */
285 		if (agino > 0) {
286 			xfs_inobt_rec_incore_t r;
287 
288 			/*
289 			 * Lookup the inode chunk that this inode lives in.
290 			 */
291 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
292 						 &tmp);
293 			if (!error &&	/* no I/O error */
294 			    tmp &&	/* lookup succeeded */
295 					/* got the record, should always work */
296 			    !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
297 			    i == 1 &&
298 					/* this is the right chunk */
299 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
300 					/* lastino was not last in chunk */
301 			    (chunkidx = agino - r.ir_startino + 1) <
302 				    XFS_INODES_PER_CHUNK &&
303 					/* there are some left allocated */
304 			    xfs_inobt_maskn(chunkidx,
305 				    XFS_INODES_PER_CHUNK - chunkidx) &
306 				    ~r.ir_free) {
307 				/*
308 				 * Grab the chunk record.  Mark all the
309 				 * uninteresting inodes (because they're
310 				 * before our start point) free.
311 				 */
312 				for (i = 0; i < chunkidx; i++) {
313 					if (XFS_INOBT_MASK(i) & ~r.ir_free)
314 						r.ir_freecount++;
315 				}
316 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
317 				irbp->ir_startino = r.ir_startino;
318 				irbp->ir_freecount = r.ir_freecount;
319 				irbp->ir_free = r.ir_free;
320 				irbp++;
321 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
322 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
323 			} else {
324 				/*
325 				 * If any of those tests failed, bump the
326 				 * inode number (just in case).
327 				 */
328 				agino++;
329 				icount = 0;
330 			}
331 			/*
332 			 * In any case, increment to the next record.
333 			 */
334 			if (!error)
335 				error = xfs_btree_increment(cur, 0, &tmp);
336 		} else {
337 			/*
338 			 * Start of ag.  Lookup the first inode chunk.
339 			 */
340 			error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
341 			icount = 0;
342 		}
343 		/*
344 		 * Loop through inode btree records in this ag,
345 		 * until we run out of inodes or space in the buffer.
346 		 */
347 		while (irbp < irbufend && icount < ubcount) {
348 			xfs_inobt_rec_incore_t r;
349 
350 			/*
351 			 * Loop as long as we're unable to read the
352 			 * inode btree.
353 			 */
354 			while (error) {
355 				agino += XFS_INODES_PER_CHUNK;
356 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
357 						be32_to_cpu(agi->agi_length))
358 					break;
359 				error = xfs_inobt_lookup(cur, agino,
360 							 XFS_LOOKUP_GE, &tmp);
361 				cond_resched();
362 			}
363 			/*
364 			 * If ran off the end of the ag either with an error,
365 			 * or the normal way, set end and stop collecting.
366 			 */
367 			if (error) {
368 				end_of_ag = 1;
369 				break;
370 			}
371 
372 			error = xfs_inobt_get_rec(cur, &r, &i);
373 			if (error || i == 0) {
374 				end_of_ag = 1;
375 				break;
376 			}
377 
378 			/*
379 			 * If this chunk has any allocated inodes, save it.
380 			 * Also start read-ahead now for this chunk.
381 			 */
382 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
383 				struct blk_plug	plug;
384 				/*
385 				 * Loop over all clusters in the next chunk.
386 				 * Do a readahead if there are any allocated
387 				 * inodes in that cluster.
388 				 */
389 				blk_start_plug(&plug);
390 				agbno = XFS_AGINO_TO_AGBNO(mp, r.ir_startino);
391 				for (chunkidx = 0;
392 				     chunkidx < XFS_INODES_PER_CHUNK;
393 				     chunkidx += nicluster,
394 				     agbno += nbcluster) {
395 					if (xfs_inobt_maskn(chunkidx, nicluster)
396 							& ~r.ir_free)
397 						xfs_btree_reada_bufs(mp, agno,
398 							agbno, nbcluster,
399 							&xfs_inode_buf_ops);
400 				}
401 				blk_finish_plug(&plug);
402 				irbp->ir_startino = r.ir_startino;
403 				irbp->ir_freecount = r.ir_freecount;
404 				irbp->ir_free = r.ir_free;
405 				irbp++;
406 				icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
407 			}
408 			/*
409 			 * Set agino to after this chunk and bump the cursor.
410 			 */
411 			agino = r.ir_startino + XFS_INODES_PER_CHUNK;
412 			error = xfs_btree_increment(cur, 0, &tmp);
413 			cond_resched();
414 		}
415 		/*
416 		 * Drop the btree buffers and the agi buffer.
417 		 * We can't hold any of the locks these represent
418 		 * when calling iget.
419 		 */
420 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
421 		xfs_buf_relse(agbp);
422 		/*
423 		 * Now format all the good inodes into the user's buffer.
424 		 */
425 		irbufend = irbp;
426 		for (irbp = irbuf;
427 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
428 			/*
429 			 * Now process this chunk of inodes.
430 			 */
431 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
432 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
433 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
434 			     chunkidx++, clustidx++, agino++) {
435 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
436 
437 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
438 				/*
439 				 * Skip if this inode is free.
440 				 */
441 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
442 					lastino = ino;
443 					continue;
444 				}
445 				/*
446 				 * Count used inodes as free so we can tell
447 				 * when the chunk is used up.
448 				 */
449 				irbp->ir_freecount++;
450 
451 				/*
452 				 * Get the inode and fill in a single buffer.
453 				 */
454 				ubused = statstruct_size;
455 				error = formatter(mp, ino, ubufp, ubleft,
456 						  &ubused, &fmterror);
457 				if (fmterror == BULKSTAT_RV_NOTHING) {
458 					if (error && error != ENOENT &&
459 						error != EINVAL) {
460 						ubleft = 0;
461 						rval = error;
462 						break;
463 					}
464 					lastino = ino;
465 					continue;
466 				}
467 				if (fmterror == BULKSTAT_RV_GIVEUP) {
468 					ubleft = 0;
469 					ASSERT(error);
470 					rval = error;
471 					break;
472 				}
473 				if (ubufp)
474 					ubufp += ubused;
475 				ubleft -= ubused;
476 				ubelem++;
477 				lastino = ino;
478 			}
479 
480 			cond_resched();
481 		}
482 		/*
483 		 * Set up for the next loop iteration.
484 		 */
485 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
486 			if (end_of_ag) {
487 				agno++;
488 				agino = 0;
489 			} else
490 				agino = XFS_INO_TO_AGINO(mp, lastino);
491 		} else
492 			break;
493 	}
494 	/*
495 	 * Done, we're either out of filesystem or space to put the data.
496 	 */
497 	kmem_free(irbuf);
498 	*ubcountp = ubelem;
499 	/*
500 	 * Found some inodes, return them now and return the error next time.
501 	 */
502 	if (ubelem)
503 		rval = 0;
504 	if (agno >= mp->m_sb.sb_agcount) {
505 		/*
506 		 * If we ran out of filesystem, mark lastino as off
507 		 * the end of the filesystem, so the next call
508 		 * will return immediately.
509 		 */
510 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
511 		*done = 1;
512 	} else
513 		*lastinop = (xfs_ino_t)lastino;
514 
515 	return rval;
516 }
517 
518 /*
519  * Return stat information in bulk (by-inode) for the filesystem.
520  * Special case for non-sequential one inode bulkstat.
521  */
522 int					/* error status */
523 xfs_bulkstat_single(
524 	xfs_mount_t		*mp,	/* mount point for filesystem */
525 	xfs_ino_t		*lastinop, /* inode to return */
526 	char			__user *buffer, /* buffer with inode stats */
527 	int			*done)	/* 1 if there are more stats to get */
528 {
529 	int			count;	/* count value for bulkstat call */
530 	int			error;	/* return value */
531 	xfs_ino_t		ino;	/* filesystem inode number */
532 	int			res;	/* result from bs1 */
533 
534 	/*
535 	 * note that requesting valid inode numbers which are not allocated
536 	 * to inodes will most likely cause xfs_imap_to_bp to generate warning
537 	 * messages about bad magic numbers. This is ok. The fact that
538 	 * the inode isn't actually an inode is handled by the
539 	 * error check below. Done this way to make the usual case faster
540 	 * at the expense of the error case.
541 	 */
542 
543 	ino = *lastinop;
544 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
545 				 NULL, &res);
546 	if (error) {
547 		/*
548 		 * Special case way failed, do it the "long" way
549 		 * to see if that works.
550 		 */
551 		(*lastinop)--;
552 		count = 1;
553 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
554 				sizeof(xfs_bstat_t), buffer, done))
555 			return error;
556 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
557 			return error == EFSCORRUPTED ?
558 				XFS_ERROR(EINVAL) : error;
559 		else
560 			return 0;
561 	}
562 	*done = 0;
563 	return 0;
564 }
565 
566 int
567 xfs_inumbers_fmt(
568 	void			__user *ubuffer, /* buffer to write to */
569 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
570 	long			count,		/* # of elements to read */
571 	long			*written)	/* # of bytes written */
572 {
573 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
574 		return -EFAULT;
575 	*written = count * sizeof(*buffer);
576 	return 0;
577 }
578 
579 /*
580  * Return inode number table for the filesystem.
581  */
582 int					/* error status */
583 xfs_inumbers(
584 	xfs_mount_t	*mp,		/* mount point for filesystem */
585 	xfs_ino_t	*lastino,	/* last inode returned */
586 	int		*count,		/* size of buffer/count returned */
587 	void		__user *ubuffer,/* buffer with inode descriptions */
588 	inumbers_fmt_pf	formatter)
589 {
590 	xfs_buf_t	*agbp;
591 	xfs_agino_t	agino;
592 	xfs_agnumber_t	agno;
593 	int		bcount;
594 	xfs_inogrp_t	*buffer;
595 	int		bufidx;
596 	xfs_btree_cur_t	*cur;
597 	int		error;
598 	xfs_inobt_rec_incore_t r;
599 	int		i;
600 	xfs_ino_t	ino;
601 	int		left;
602 	int		tmp;
603 
604 	ino = (xfs_ino_t)*lastino;
605 	agno = XFS_INO_TO_AGNO(mp, ino);
606 	agino = XFS_INO_TO_AGINO(mp, ino);
607 	left = *count;
608 	*count = 0;
609 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
610 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
611 	error = bufidx = 0;
612 	cur = NULL;
613 	agbp = NULL;
614 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
615 		if (agbp == NULL) {
616 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
617 			if (error) {
618 				/*
619 				 * If we can't read the AGI of this ag,
620 				 * then just skip to the next one.
621 				 */
622 				ASSERT(cur == NULL);
623 				agbp = NULL;
624 				agno++;
625 				agino = 0;
626 				continue;
627 			}
628 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
629 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
630 						 &tmp);
631 			if (error) {
632 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
633 				cur = NULL;
634 				xfs_buf_relse(agbp);
635 				agbp = NULL;
636 				/*
637 				 * Move up the last inode in the current
638 				 * chunk.  The lookup_ge will always get
639 				 * us the first inode in the next chunk.
640 				 */
641 				agino += XFS_INODES_PER_CHUNK - 1;
642 				continue;
643 			}
644 		}
645 		error = xfs_inobt_get_rec(cur, &r, &i);
646 		if (error || i == 0) {
647 			xfs_buf_relse(agbp);
648 			agbp = NULL;
649 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
650 			cur = NULL;
651 			agno++;
652 			agino = 0;
653 			continue;
654 		}
655 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
656 		buffer[bufidx].xi_startino =
657 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
658 		buffer[bufidx].xi_alloccount =
659 			XFS_INODES_PER_CHUNK - r.ir_freecount;
660 		buffer[bufidx].xi_allocmask = ~r.ir_free;
661 		bufidx++;
662 		left--;
663 		if (bufidx == bcount) {
664 			long written;
665 			if (formatter(ubuffer, buffer, bufidx, &written)) {
666 				error = XFS_ERROR(EFAULT);
667 				break;
668 			}
669 			ubuffer += written;
670 			*count += bufidx;
671 			bufidx = 0;
672 		}
673 		if (left) {
674 			error = xfs_btree_increment(cur, 0, &tmp);
675 			if (error) {
676 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
677 				cur = NULL;
678 				xfs_buf_relse(agbp);
679 				agbp = NULL;
680 				/*
681 				 * The agino value has already been bumped.
682 				 * Just try to skip up to it.
683 				 */
684 				agino += XFS_INODES_PER_CHUNK;
685 				continue;
686 			}
687 		}
688 	}
689 	if (!error) {
690 		if (bufidx) {
691 			long written;
692 			if (formatter(ubuffer, buffer, bufidx, &written))
693 				error = XFS_ERROR(EFAULT);
694 			else
695 				*count += bufidx;
696 		}
697 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
698 	}
699 	kmem_free(buffer);
700 	if (cur)
701 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
702 					   XFS_BTREE_NOERROR));
703 	if (agbp)
704 		xfs_buf_relse(agbp);
705 	return error;
706 }
707