xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 600a711c)
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 
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 		 (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino)));
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 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
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 		bp = NULL;
266 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
267 		if (error) {
268 			/*
269 			 * Skip this allocation group and go to the next one.
270 			 */
271 			agno++;
272 			agino = 0;
273 			continue;
274 		}
275 		agi = XFS_BUF_TO_AGI(agbp);
276 		/*
277 		 * Allocate and initialize a btree cursor for ialloc btree.
278 		 */
279 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
280 		irbp = irbuf;
281 		irbufend = irbuf + nirbuf;
282 		end_of_ag = 0;
283 		/*
284 		 * If we're returning in the middle of an allocation group,
285 		 * we need to get the remainder of the chunk we're in.
286 		 */
287 		if (agino > 0) {
288 			xfs_inobt_rec_incore_t r;
289 
290 			/*
291 			 * Lookup the inode chunk that this inode lives in.
292 			 */
293 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
294 						 &tmp);
295 			if (!error &&	/* no I/O error */
296 			    tmp &&	/* lookup succeeded */
297 					/* got the record, should always work */
298 			    !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
299 			    i == 1 &&
300 					/* this is the right chunk */
301 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
302 					/* lastino was not last in chunk */
303 			    (chunkidx = agino - r.ir_startino + 1) <
304 				    XFS_INODES_PER_CHUNK &&
305 					/* there are some left allocated */
306 			    xfs_inobt_maskn(chunkidx,
307 				    XFS_INODES_PER_CHUNK - chunkidx) &
308 				    ~r.ir_free) {
309 				/*
310 				 * Grab the chunk record.  Mark all the
311 				 * uninteresting inodes (because they're
312 				 * before our start point) free.
313 				 */
314 				for (i = 0; i < chunkidx; i++) {
315 					if (XFS_INOBT_MASK(i) & ~r.ir_free)
316 						r.ir_freecount++;
317 				}
318 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
319 				irbp->ir_startino = r.ir_startino;
320 				irbp->ir_freecount = r.ir_freecount;
321 				irbp->ir_free = r.ir_free;
322 				irbp++;
323 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
324 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
325 			} else {
326 				/*
327 				 * If any of those tests failed, bump the
328 				 * inode number (just in case).
329 				 */
330 				agino++;
331 				icount = 0;
332 			}
333 			/*
334 			 * In any case, increment to the next record.
335 			 */
336 			if (!error)
337 				error = xfs_btree_increment(cur, 0, &tmp);
338 		} else {
339 			/*
340 			 * Start of ag.  Lookup the first inode chunk.
341 			 */
342 			error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
343 			icount = 0;
344 		}
345 		/*
346 		 * Loop through inode btree records in this ag,
347 		 * until we run out of inodes or space in the buffer.
348 		 */
349 		while (irbp < irbufend && icount < ubcount) {
350 			xfs_inobt_rec_incore_t r;
351 
352 			/*
353 			 * Loop as long as we're unable to read the
354 			 * inode btree.
355 			 */
356 			while (error) {
357 				agino += XFS_INODES_PER_CHUNK;
358 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
359 						be32_to_cpu(agi->agi_length))
360 					break;
361 				error = xfs_inobt_lookup(cur, agino,
362 							 XFS_LOOKUP_GE, &tmp);
363 				cond_resched();
364 			}
365 			/*
366 			 * If ran off the end of the ag either with an error,
367 			 * or the normal way, set end and stop collecting.
368 			 */
369 			if (error) {
370 				end_of_ag = 1;
371 				break;
372 			}
373 
374 			error = xfs_inobt_get_rec(cur, &r, &i);
375 			if (error || i == 0) {
376 				end_of_ag = 1;
377 				break;
378 			}
379 
380 			/*
381 			 * If this chunk has any allocated inodes, save it.
382 			 * Also start read-ahead now for this chunk.
383 			 */
384 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
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 				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 				}
400 				irbp->ir_startino = r.ir_startino;
401 				irbp->ir_freecount = r.ir_freecount;
402 				irbp->ir_free = r.ir_free;
403 				irbp++;
404 				icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
405 			}
406 			/*
407 			 * Set agino to after this chunk and bump the cursor.
408 			 */
409 			agino = r.ir_startino + XFS_INODES_PER_CHUNK;
410 			error = xfs_btree_increment(cur, 0, &tmp);
411 			cond_resched();
412 		}
413 		/*
414 		 * Drop the btree buffers and the agi buffer.
415 		 * We can't hold any of the locks these represent
416 		 * when calling iget.
417 		 */
418 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
419 		xfs_buf_relse(agbp);
420 		/*
421 		 * Now format all the good inodes into the user's buffer.
422 		 */
423 		irbufend = irbp;
424 		for (irbp = irbuf;
425 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
426 			/*
427 			 * Now process this chunk of inodes.
428 			 */
429 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
430 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
431 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
432 			     chunkidx++, clustidx++, agino++) {
433 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
434 				/*
435 				 * Recompute agbno if this is the
436 				 * first inode of the cluster.
437 				 *
438 				 * Careful with clustidx.   There can be
439 				 * multiple clusters per chunk, a single
440 				 * cluster per chunk or a cluster that has
441 				 * inodes represented from several different
442 				 * chunks (if blocksize is large).
443 				 *
444 				 * Because of this, the starting clustidx is
445 				 * initialized to zero in this loop but must
446 				 * later be reset after reading in the cluster
447 				 * buffer.
448 				 */
449 				if ((chunkidx & (nicluster - 1)) == 0) {
450 					agbno = XFS_AGINO_TO_AGBNO(mp,
451 							irbp->ir_startino) +
452 						((chunkidx & nimask) >>
453 						 mp->m_sb.sb_inopblog);
454 				}
455 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
456 				/*
457 				 * Skip if this inode is free.
458 				 */
459 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
460 					lastino = ino;
461 					continue;
462 				}
463 				/*
464 				 * Count used inodes as free so we can tell
465 				 * when the chunk is used up.
466 				 */
467 				irbp->ir_freecount++;
468 
469 				/*
470 				 * Get the inode and fill in a single buffer.
471 				 */
472 				ubused = statstruct_size;
473 				error = formatter(mp, ino, ubufp, ubleft,
474 						  &ubused, &fmterror);
475 				if (fmterror == BULKSTAT_RV_NOTHING) {
476 					if (error && error != ENOENT &&
477 						error != EINVAL) {
478 						ubleft = 0;
479 						rval = error;
480 						break;
481 					}
482 					lastino = ino;
483 					continue;
484 				}
485 				if (fmterror == BULKSTAT_RV_GIVEUP) {
486 					ubleft = 0;
487 					ASSERT(error);
488 					rval = error;
489 					break;
490 				}
491 				if (ubufp)
492 					ubufp += ubused;
493 				ubleft -= ubused;
494 				ubelem++;
495 				lastino = ino;
496 			}
497 
498 			cond_resched();
499 		}
500 
501 		if (bp)
502 			xfs_buf_relse(bp);
503 
504 		/*
505 		 * Set up for the next loop iteration.
506 		 */
507 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
508 			if (end_of_ag) {
509 				agno++;
510 				agino = 0;
511 			} else
512 				agino = XFS_INO_TO_AGINO(mp, lastino);
513 		} else
514 			break;
515 	}
516 	/*
517 	 * Done, we're either out of filesystem or space to put the data.
518 	 */
519 	kmem_free_large(irbuf);
520 	*ubcountp = ubelem;
521 	/*
522 	 * Found some inodes, return them now and return the error next time.
523 	 */
524 	if (ubelem)
525 		rval = 0;
526 	if (agno >= mp->m_sb.sb_agcount) {
527 		/*
528 		 * If we ran out of filesystem, mark lastino as off
529 		 * the end of the filesystem, so the next call
530 		 * will return immediately.
531 		 */
532 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
533 		*done = 1;
534 	} else
535 		*lastinop = (xfs_ino_t)lastino;
536 
537 	return rval;
538 }
539 
540 /*
541  * Return stat information in bulk (by-inode) for the filesystem.
542  * Special case for non-sequential one inode bulkstat.
543  */
544 int					/* error status */
545 xfs_bulkstat_single(
546 	xfs_mount_t		*mp,	/* mount point for filesystem */
547 	xfs_ino_t		*lastinop, /* inode to return */
548 	char			__user *buffer, /* buffer with inode stats */
549 	int			*done)	/* 1 if there are more stats to get */
550 {
551 	int			count;	/* count value for bulkstat call */
552 	int			error;	/* return value */
553 	xfs_ino_t		ino;	/* filesystem inode number */
554 	int			res;	/* result from bs1 */
555 
556 	/*
557 	 * note that requesting valid inode numbers which are not allocated
558 	 * to inodes will most likely cause xfs_imap_to_bp to generate warning
559 	 * messages about bad magic numbers. This is ok. The fact that
560 	 * the inode isn't actually an inode is handled by the
561 	 * error check below. Done this way to make the usual case faster
562 	 * at the expense of the error case.
563 	 */
564 
565 	ino = (xfs_ino_t)*lastinop;
566 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t), 0, &res);
567 	if (error) {
568 		/*
569 		 * Special case way failed, do it the "long" way
570 		 * to see if that works.
571 		 */
572 		(*lastinop)--;
573 		count = 1;
574 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
575 				sizeof(xfs_bstat_t), buffer, done))
576 			return error;
577 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
578 			return error == EFSCORRUPTED ?
579 				XFS_ERROR(EINVAL) : error;
580 		else
581 			return 0;
582 	}
583 	*done = 0;
584 	return 0;
585 }
586 
587 int
588 xfs_inumbers_fmt(
589 	void			__user *ubuffer, /* buffer to write to */
590 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
591 	long			count,		/* # of elements to read */
592 	long			*written)	/* # of bytes written */
593 {
594 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
595 		return -EFAULT;
596 	*written = count * sizeof(*buffer);
597 	return 0;
598 }
599 
600 /*
601  * Return inode number table for the filesystem.
602  */
603 int					/* error status */
604 xfs_inumbers(
605 	xfs_mount_t	*mp,		/* mount point for filesystem */
606 	xfs_ino_t	*lastino,	/* last inode returned */
607 	int		*count,		/* size of buffer/count returned */
608 	void		__user *ubuffer,/* buffer with inode descriptions */
609 	inumbers_fmt_pf	formatter)
610 {
611 	xfs_buf_t	*agbp;
612 	xfs_agino_t	agino;
613 	xfs_agnumber_t	agno;
614 	int		bcount;
615 	xfs_inogrp_t	*buffer;
616 	int		bufidx;
617 	xfs_btree_cur_t	*cur;
618 	int		error;
619 	xfs_inobt_rec_incore_t r;
620 	int		i;
621 	xfs_ino_t	ino;
622 	int		left;
623 	int		tmp;
624 
625 	ino = (xfs_ino_t)*lastino;
626 	agno = XFS_INO_TO_AGNO(mp, ino);
627 	agino = XFS_INO_TO_AGINO(mp, ino);
628 	left = *count;
629 	*count = 0;
630 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
631 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
632 	error = bufidx = 0;
633 	cur = NULL;
634 	agbp = NULL;
635 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
636 		if (agbp == NULL) {
637 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
638 			if (error) {
639 				/*
640 				 * If we can't read the AGI of this ag,
641 				 * then just skip to the next one.
642 				 */
643 				ASSERT(cur == NULL);
644 				agbp = NULL;
645 				agno++;
646 				agino = 0;
647 				continue;
648 			}
649 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
650 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
651 						 &tmp);
652 			if (error) {
653 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
654 				cur = NULL;
655 				xfs_buf_relse(agbp);
656 				agbp = NULL;
657 				/*
658 				 * Move up the last inode in the current
659 				 * chunk.  The lookup_ge will always get
660 				 * us the first inode in the next chunk.
661 				 */
662 				agino += XFS_INODES_PER_CHUNK - 1;
663 				continue;
664 			}
665 		}
666 		error = xfs_inobt_get_rec(cur, &r, &i);
667 		if (error || i == 0) {
668 			xfs_buf_relse(agbp);
669 			agbp = NULL;
670 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
671 			cur = NULL;
672 			agno++;
673 			agino = 0;
674 			continue;
675 		}
676 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
677 		buffer[bufidx].xi_startino =
678 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
679 		buffer[bufidx].xi_alloccount =
680 			XFS_INODES_PER_CHUNK - r.ir_freecount;
681 		buffer[bufidx].xi_allocmask = ~r.ir_free;
682 		bufidx++;
683 		left--;
684 		if (bufidx == bcount) {
685 			long written;
686 			if (formatter(ubuffer, buffer, bufidx, &written)) {
687 				error = XFS_ERROR(EFAULT);
688 				break;
689 			}
690 			ubuffer += written;
691 			*count += bufidx;
692 			bufidx = 0;
693 		}
694 		if (left) {
695 			error = xfs_btree_increment(cur, 0, &tmp);
696 			if (error) {
697 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
698 				cur = NULL;
699 				xfs_buf_relse(agbp);
700 				agbp = NULL;
701 				/*
702 				 * The agino value has already been bumped.
703 				 * Just try to skip up to it.
704 				 */
705 				agino += XFS_INODES_PER_CHUNK;
706 				continue;
707 			}
708 		}
709 	}
710 	if (!error) {
711 		if (bufidx) {
712 			long written;
713 			if (formatter(ubuffer, buffer, bufidx, &written))
714 				error = XFS_ERROR(EFAULT);
715 			else
716 				*count += bufidx;
717 		}
718 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
719 	}
720 	kmem_free(buffer);
721 	if (cur)
722 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
723 					   XFS_BTREE_NOERROR));
724 	if (agbp)
725 		xfs_buf_relse(agbp);
726 	return error;
727 }
728