xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision e23feb16)
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_types.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc_btree.h"
29 #include "xfs_ialloc_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_itable.h"
34 #include "xfs_error.h"
35 #include "xfs_btree.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 
39 STATIC int
40 xfs_internal_inum(
41 	xfs_mount_t	*mp,
42 	xfs_ino_t	ino)
43 {
44 	return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
45 		(xfs_sb_version_hasquota(&mp->m_sb) &&
46 		 xfs_is_quota_inode(&mp->m_sb, ino)));
47 }
48 
49 /*
50  * Return stat information for one inode.
51  * Return 0 if ok, else errno.
52  */
53 int
54 xfs_bulkstat_one_int(
55 	struct xfs_mount	*mp,		/* mount point for filesystem */
56 	xfs_ino_t		ino,		/* inode to get data for */
57 	void __user		*buffer,	/* buffer to place output in */
58 	int			ubsize,		/* size of buffer */
59 	bulkstat_one_fmt_pf	formatter,	/* formatter, copy to user */
60 	int			*ubused,	/* bytes used by me */
61 	int			*stat)		/* BULKSTAT_RV_... */
62 {
63 	struct xfs_icdinode	*dic;		/* dinode core info pointer */
64 	struct xfs_inode	*ip;		/* incore inode pointer */
65 	struct xfs_bstat	*buf;		/* return buffer */
66 	int			error = 0;	/* error value */
67 
68 	*stat = BULKSTAT_RV_NOTHING;
69 
70 	if (!buffer || xfs_internal_inum(mp, ino))
71 		return XFS_ERROR(EINVAL);
72 
73 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP | KM_MAYFAIL);
74 	if (!buf)
75 		return XFS_ERROR(ENOMEM);
76 
77 	error = xfs_iget(mp, NULL, ino,
78 			 (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED),
79 			 XFS_ILOCK_SHARED, &ip);
80 	if (error) {
81 		*stat = BULKSTAT_RV_NOTHING;
82 		goto out_free;
83 	}
84 
85 	ASSERT(ip != NULL);
86 	ASSERT(ip->i_imap.im_blkno != 0);
87 
88 	dic = &ip->i_d;
89 
90 	/* xfs_iget returns the following without needing
91 	 * further change.
92 	 */
93 	buf->bs_nlink = dic->di_nlink;
94 	buf->bs_projid_lo = dic->di_projid_lo;
95 	buf->bs_projid_hi = dic->di_projid_hi;
96 	buf->bs_ino = ino;
97 	buf->bs_mode = dic->di_mode;
98 	buf->bs_uid = dic->di_uid;
99 	buf->bs_gid = dic->di_gid;
100 	buf->bs_size = dic->di_size;
101 	buf->bs_atime.tv_sec = dic->di_atime.t_sec;
102 	buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
103 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
104 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
105 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
106 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
107 	buf->bs_xflags = xfs_ip2xflags(ip);
108 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
109 	buf->bs_extents = dic->di_nextents;
110 	buf->bs_gen = dic->di_gen;
111 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
112 	buf->bs_dmevmask = dic->di_dmevmask;
113 	buf->bs_dmstate = dic->di_dmstate;
114 	buf->bs_aextents = dic->di_anextents;
115 	buf->bs_forkoff = XFS_IFORK_BOFF(ip);
116 
117 	switch (dic->di_format) {
118 	case XFS_DINODE_FMT_DEV:
119 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
120 		buf->bs_blksize = BLKDEV_IOSIZE;
121 		buf->bs_blocks = 0;
122 		break;
123 	case XFS_DINODE_FMT_LOCAL:
124 	case XFS_DINODE_FMT_UUID:
125 		buf->bs_rdev = 0;
126 		buf->bs_blksize = mp->m_sb.sb_blocksize;
127 		buf->bs_blocks = 0;
128 		break;
129 	case XFS_DINODE_FMT_EXTENTS:
130 	case XFS_DINODE_FMT_BTREE:
131 		buf->bs_rdev = 0;
132 		buf->bs_blksize = mp->m_sb.sb_blocksize;
133 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
134 		break;
135 	}
136 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
137 	IRELE(ip);
138 
139 	error = formatter(buffer, ubsize, ubused, buf);
140 
141 	if (!error)
142 		*stat = BULKSTAT_RV_DIDONE;
143 
144  out_free:
145 	kmem_free(buf);
146 	return error;
147 }
148 
149 /* Return 0 on success or positive error */
150 STATIC int
151 xfs_bulkstat_one_fmt(
152 	void			__user *ubuffer,
153 	int			ubsize,
154 	int			*ubused,
155 	const xfs_bstat_t	*buffer)
156 {
157 	if (ubsize < sizeof(*buffer))
158 		return XFS_ERROR(ENOMEM);
159 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
160 		return XFS_ERROR(EFAULT);
161 	if (ubused)
162 		*ubused = sizeof(*buffer);
163 	return 0;
164 }
165 
166 int
167 xfs_bulkstat_one(
168 	xfs_mount_t	*mp,		/* mount point for filesystem */
169 	xfs_ino_t	ino,		/* inode number to get data for */
170 	void		__user *buffer,	/* buffer to place output in */
171 	int		ubsize,		/* size of buffer */
172 	int		*ubused,	/* bytes used by me */
173 	int		*stat)		/* BULKSTAT_RV_... */
174 {
175 	return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
176 				    xfs_bulkstat_one_fmt, ubused, stat);
177 }
178 
179 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
180 
181 /*
182  * Return stat information in bulk (by-inode) for the filesystem.
183  */
184 int					/* error status */
185 xfs_bulkstat(
186 	xfs_mount_t		*mp,	/* mount point for filesystem */
187 	xfs_ino_t		*lastinop, /* last inode returned */
188 	int			*ubcountp, /* size of buffer/count returned */
189 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
190 	size_t			statstruct_size, /* sizeof struct filling */
191 	char			__user *ubuffer, /* buffer with inode stats */
192 	int			*done)	/* 1 if there are more stats to get */
193 {
194 	xfs_agblock_t		agbno=0;/* allocation group block number */
195 	xfs_buf_t		*agbp;	/* agi header buffer */
196 	xfs_agi_t		*agi;	/* agi header data */
197 	xfs_agino_t		agino;	/* inode # in allocation group */
198 	xfs_agnumber_t		agno;	/* allocation group number */
199 	int			chunkidx; /* current index into inode chunk */
200 	int			clustidx; /* current index into inode cluster */
201 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
202 	int			end_of_ag; /* set if we've seen the ag end */
203 	int			error;	/* error code */
204 	int                     fmterror;/* bulkstat formatter result */
205 	int			i;	/* loop index */
206 	int			icount;	/* count of inodes good in irbuf */
207 	size_t			irbsize; /* size of irec buffer in bytes */
208 	xfs_ino_t		ino;	/* inode number (filesystem) */
209 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
210 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
211 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
212 	xfs_ino_t		lastino; /* last inode number returned */
213 	int			nbcluster; /* # of blocks in a cluster */
214 	int			nicluster; /* # of inodes in a cluster */
215 	int			nimask;	/* mask for inode clusters */
216 	int			nirbuf;	/* size of irbuf */
217 	int			rval;	/* return value error code */
218 	int			tmp;	/* result value from btree calls */
219 	int			ubcount; /* size of user's buffer */
220 	int			ubleft;	/* bytes left in user's buffer */
221 	char			__user *ubufp;	/* pointer into user's buffer */
222 	int			ubelem;	/* spaces used in user's buffer */
223 	int			ubused;	/* bytes used by formatter */
224 
225 	/*
226 	 * Get the last inode value, see if there's nothing to do.
227 	 */
228 	ino = (xfs_ino_t)*lastinop;
229 	lastino = ino;
230 	agno = XFS_INO_TO_AGNO(mp, ino);
231 	agino = XFS_INO_TO_AGINO(mp, ino);
232 	if (agno >= mp->m_sb.sb_agcount ||
233 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
234 		*done = 1;
235 		*ubcountp = 0;
236 		return 0;
237 	}
238 	if (!ubcountp || *ubcountp <= 0) {
239 		return EINVAL;
240 	}
241 	ubcount = *ubcountp; /* statstruct's */
242 	ubleft = ubcount * statstruct_size; /* bytes */
243 	*ubcountp = ubelem = 0;
244 	*done = 0;
245 	fmterror = 0;
246 	ubufp = ubuffer;
247 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
248 		mp->m_sb.sb_inopblock :
249 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
250 	nimask = ~(nicluster - 1);
251 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
252 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4);
253 	if (!irbuf)
254 		return ENOMEM;
255 
256 	nirbuf = irbsize / sizeof(*irbuf);
257 
258 	/*
259 	 * Loop over the allocation groups, starting from the last
260 	 * inode returned; 0 means start of the allocation group.
261 	 */
262 	rval = 0;
263 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
264 		cond_resched();
265 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
266 		if (error) {
267 			/*
268 			 * Skip this allocation group and go to the next one.
269 			 */
270 			agno++;
271 			agino = 0;
272 			continue;
273 		}
274 		agi = XFS_BUF_TO_AGI(agbp);
275 		/*
276 		 * Allocate and initialize a btree cursor for ialloc btree.
277 		 */
278 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
279 		irbp = irbuf;
280 		irbufend = irbuf + nirbuf;
281 		end_of_ag = 0;
282 		/*
283 		 * If we're returning in the middle of an allocation group,
284 		 * we need to get the remainder of the chunk we're in.
285 		 */
286 		if (agino > 0) {
287 			xfs_inobt_rec_incore_t r;
288 
289 			/*
290 			 * Lookup the inode chunk that this inode lives in.
291 			 */
292 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
293 						 &tmp);
294 			if (!error &&	/* no I/O error */
295 			    tmp &&	/* lookup succeeded */
296 					/* got the record, should always work */
297 			    !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
298 			    i == 1 &&
299 					/* this is the right chunk */
300 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
301 					/* lastino was not last in chunk */
302 			    (chunkidx = agino - r.ir_startino + 1) <
303 				    XFS_INODES_PER_CHUNK &&
304 					/* there are some left allocated */
305 			    xfs_inobt_maskn(chunkidx,
306 				    XFS_INODES_PER_CHUNK - chunkidx) &
307 				    ~r.ir_free) {
308 				/*
309 				 * Grab the chunk record.  Mark all the
310 				 * uninteresting inodes (because they're
311 				 * before our start point) free.
312 				 */
313 				for (i = 0; i < chunkidx; i++) {
314 					if (XFS_INOBT_MASK(i) & ~r.ir_free)
315 						r.ir_freecount++;
316 				}
317 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
318 				irbp->ir_startino = r.ir_startino;
319 				irbp->ir_freecount = r.ir_freecount;
320 				irbp->ir_free = r.ir_free;
321 				irbp++;
322 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
323 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
324 			} else {
325 				/*
326 				 * If any of those tests failed, bump the
327 				 * inode number (just in case).
328 				 */
329 				agino++;
330 				icount = 0;
331 			}
332 			/*
333 			 * In any case, increment to the next record.
334 			 */
335 			if (!error)
336 				error = xfs_btree_increment(cur, 0, &tmp);
337 		} else {
338 			/*
339 			 * Start of ag.  Lookup the first inode chunk.
340 			 */
341 			error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
342 			icount = 0;
343 		}
344 		/*
345 		 * Loop through inode btree records in this ag,
346 		 * until we run out of inodes or space in the buffer.
347 		 */
348 		while (irbp < irbufend && icount < ubcount) {
349 			xfs_inobt_rec_incore_t r;
350 
351 			/*
352 			 * Loop as long as we're unable to read the
353 			 * inode btree.
354 			 */
355 			while (error) {
356 				agino += XFS_INODES_PER_CHUNK;
357 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
358 						be32_to_cpu(agi->agi_length))
359 					break;
360 				error = xfs_inobt_lookup(cur, agino,
361 							 XFS_LOOKUP_GE, &tmp);
362 				cond_resched();
363 			}
364 			/*
365 			 * If ran off the end of the ag either with an error,
366 			 * or the normal way, set end and stop collecting.
367 			 */
368 			if (error) {
369 				end_of_ag = 1;
370 				break;
371 			}
372 
373 			error = xfs_inobt_get_rec(cur, &r, &i);
374 			if (error || i == 0) {
375 				end_of_ag = 1;
376 				break;
377 			}
378 
379 			/*
380 			 * If this chunk has any allocated inodes, save it.
381 			 * Also start read-ahead now for this chunk.
382 			 */
383 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
384 				struct blk_plug	plug;
385 				/*
386 				 * Loop over all clusters in the next chunk.
387 				 * Do a readahead if there are any allocated
388 				 * inodes in that cluster.
389 				 */
390 				blk_start_plug(&plug);
391 				agbno = XFS_AGINO_TO_AGBNO(mp, r.ir_startino);
392 				for (chunkidx = 0;
393 				     chunkidx < XFS_INODES_PER_CHUNK;
394 				     chunkidx += nicluster,
395 				     agbno += nbcluster) {
396 					if (xfs_inobt_maskn(chunkidx, nicluster)
397 							& ~r.ir_free)
398 						xfs_btree_reada_bufs(mp, agno,
399 							agbno, nbcluster,
400 							&xfs_inode_buf_ops);
401 				}
402 				blk_finish_plug(&plug);
403 				irbp->ir_startino = r.ir_startino;
404 				irbp->ir_freecount = r.ir_freecount;
405 				irbp->ir_free = r.ir_free;
406 				irbp++;
407 				icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
408 			}
409 			/*
410 			 * Set agino to after this chunk and bump the cursor.
411 			 */
412 			agino = r.ir_startino + XFS_INODES_PER_CHUNK;
413 			error = xfs_btree_increment(cur, 0, &tmp);
414 			cond_resched();
415 		}
416 		/*
417 		 * Drop the btree buffers and the agi buffer.
418 		 * We can't hold any of the locks these represent
419 		 * when calling iget.
420 		 */
421 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
422 		xfs_buf_relse(agbp);
423 		/*
424 		 * Now format all the good inodes into the user's buffer.
425 		 */
426 		irbufend = irbp;
427 		for (irbp = irbuf;
428 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
429 			/*
430 			 * Now process this chunk of inodes.
431 			 */
432 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
433 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
434 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
435 			     chunkidx++, clustidx++, agino++) {
436 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
437 
438 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
439 				/*
440 				 * Skip if this inode is free.
441 				 */
442 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
443 					lastino = ino;
444 					continue;
445 				}
446 				/*
447 				 * Count used inodes as free so we can tell
448 				 * when the chunk is used up.
449 				 */
450 				irbp->ir_freecount++;
451 
452 				/*
453 				 * Get the inode and fill in a single buffer.
454 				 */
455 				ubused = statstruct_size;
456 				error = formatter(mp, ino, ubufp, ubleft,
457 						  &ubused, &fmterror);
458 				if (fmterror == BULKSTAT_RV_NOTHING) {
459 					if (error && error != ENOENT &&
460 						error != EINVAL) {
461 						ubleft = 0;
462 						rval = error;
463 						break;
464 					}
465 					lastino = ino;
466 					continue;
467 				}
468 				if (fmterror == BULKSTAT_RV_GIVEUP) {
469 					ubleft = 0;
470 					ASSERT(error);
471 					rval = error;
472 					break;
473 				}
474 				if (ubufp)
475 					ubufp += ubused;
476 				ubleft -= ubused;
477 				ubelem++;
478 				lastino = ino;
479 			}
480 
481 			cond_resched();
482 		}
483 		/*
484 		 * Set up for the next loop iteration.
485 		 */
486 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
487 			if (end_of_ag) {
488 				agno++;
489 				agino = 0;
490 			} else
491 				agino = XFS_INO_TO_AGINO(mp, lastino);
492 		} else
493 			break;
494 	}
495 	/*
496 	 * Done, we're either out of filesystem or space to put the data.
497 	 */
498 	kmem_free(irbuf);
499 	*ubcountp = ubelem;
500 	/*
501 	 * Found some inodes, return them now and return the error next time.
502 	 */
503 	if (ubelem)
504 		rval = 0;
505 	if (agno >= mp->m_sb.sb_agcount) {
506 		/*
507 		 * If we ran out of filesystem, mark lastino as off
508 		 * the end of the filesystem, so the next call
509 		 * will return immediately.
510 		 */
511 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
512 		*done = 1;
513 	} else
514 		*lastinop = (xfs_ino_t)lastino;
515 
516 	return rval;
517 }
518 
519 /*
520  * Return stat information in bulk (by-inode) for the filesystem.
521  * Special case for non-sequential one inode bulkstat.
522  */
523 int					/* error status */
524 xfs_bulkstat_single(
525 	xfs_mount_t		*mp,	/* mount point for filesystem */
526 	xfs_ino_t		*lastinop, /* inode to return */
527 	char			__user *buffer, /* buffer with inode stats */
528 	int			*done)	/* 1 if there are more stats to get */
529 {
530 	int			count;	/* count value for bulkstat call */
531 	int			error;	/* return value */
532 	xfs_ino_t		ino;	/* filesystem inode number */
533 	int			res;	/* result from bs1 */
534 
535 	/*
536 	 * note that requesting valid inode numbers which are not allocated
537 	 * to inodes will most likely cause xfs_imap_to_bp to generate warning
538 	 * messages about bad magic numbers. This is ok. The fact that
539 	 * the inode isn't actually an inode is handled by the
540 	 * error check below. Done this way to make the usual case faster
541 	 * at the expense of the error case.
542 	 */
543 
544 	ino = *lastinop;
545 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
546 				 NULL, &res);
547 	if (error) {
548 		/*
549 		 * Special case way failed, do it the "long" way
550 		 * to see if that works.
551 		 */
552 		(*lastinop)--;
553 		count = 1;
554 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
555 				sizeof(xfs_bstat_t), buffer, done))
556 			return error;
557 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
558 			return error == EFSCORRUPTED ?
559 				XFS_ERROR(EINVAL) : error;
560 		else
561 			return 0;
562 	}
563 	*done = 0;
564 	return 0;
565 }
566 
567 int
568 xfs_inumbers_fmt(
569 	void			__user *ubuffer, /* buffer to write to */
570 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
571 	long			count,		/* # of elements to read */
572 	long			*written)	/* # of bytes written */
573 {
574 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
575 		return -EFAULT;
576 	*written = count * sizeof(*buffer);
577 	return 0;
578 }
579 
580 /*
581  * Return inode number table for the filesystem.
582  */
583 int					/* error status */
584 xfs_inumbers(
585 	xfs_mount_t	*mp,		/* mount point for filesystem */
586 	xfs_ino_t	*lastino,	/* last inode returned */
587 	int		*count,		/* size of buffer/count returned */
588 	void		__user *ubuffer,/* buffer with inode descriptions */
589 	inumbers_fmt_pf	formatter)
590 {
591 	xfs_buf_t	*agbp;
592 	xfs_agino_t	agino;
593 	xfs_agnumber_t	agno;
594 	int		bcount;
595 	xfs_inogrp_t	*buffer;
596 	int		bufidx;
597 	xfs_btree_cur_t	*cur;
598 	int		error;
599 	xfs_inobt_rec_incore_t r;
600 	int		i;
601 	xfs_ino_t	ino;
602 	int		left;
603 	int		tmp;
604 
605 	ino = (xfs_ino_t)*lastino;
606 	agno = XFS_INO_TO_AGNO(mp, ino);
607 	agino = XFS_INO_TO_AGINO(mp, ino);
608 	left = *count;
609 	*count = 0;
610 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
611 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
612 	error = bufidx = 0;
613 	cur = NULL;
614 	agbp = NULL;
615 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
616 		if (agbp == NULL) {
617 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
618 			if (error) {
619 				/*
620 				 * If we can't read the AGI of this ag,
621 				 * then just skip to the next one.
622 				 */
623 				ASSERT(cur == NULL);
624 				agbp = NULL;
625 				agno++;
626 				agino = 0;
627 				continue;
628 			}
629 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
630 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
631 						 &tmp);
632 			if (error) {
633 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
634 				cur = NULL;
635 				xfs_buf_relse(agbp);
636 				agbp = NULL;
637 				/*
638 				 * Move up the last inode in the current
639 				 * chunk.  The lookup_ge will always get
640 				 * us the first inode in the next chunk.
641 				 */
642 				agino += XFS_INODES_PER_CHUNK - 1;
643 				continue;
644 			}
645 		}
646 		error = xfs_inobt_get_rec(cur, &r, &i);
647 		if (error || i == 0) {
648 			xfs_buf_relse(agbp);
649 			agbp = NULL;
650 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
651 			cur = NULL;
652 			agno++;
653 			agino = 0;
654 			continue;
655 		}
656 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
657 		buffer[bufidx].xi_startino =
658 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
659 		buffer[bufidx].xi_alloccount =
660 			XFS_INODES_PER_CHUNK - r.ir_freecount;
661 		buffer[bufidx].xi_allocmask = ~r.ir_free;
662 		bufidx++;
663 		left--;
664 		if (bufidx == bcount) {
665 			long written;
666 			if (formatter(ubuffer, buffer, bufidx, &written)) {
667 				error = XFS_ERROR(EFAULT);
668 				break;
669 			}
670 			ubuffer += written;
671 			*count += bufidx;
672 			bufidx = 0;
673 		}
674 		if (left) {
675 			error = xfs_btree_increment(cur, 0, &tmp);
676 			if (error) {
677 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
678 				cur = NULL;
679 				xfs_buf_relse(agbp);
680 				agbp = NULL;
681 				/*
682 				 * The agino value has already been bumped.
683 				 * Just try to skip up to it.
684 				 */
685 				agino += XFS_INODES_PER_CHUNK;
686 				continue;
687 			}
688 		}
689 	}
690 	if (!error) {
691 		if (bufidx) {
692 			long written;
693 			if (formatter(ubuffer, buffer, bufidx, &written))
694 				error = XFS_ERROR(EFAULT);
695 			else
696 				*count += bufidx;
697 		}
698 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
699 	}
700 	kmem_free(buffer);
701 	if (cur)
702 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
703 					   XFS_BTREE_NOERROR));
704 	if (agbp)
705 		xfs_buf_relse(agbp);
706 	return error;
707 }
708