xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 93f14468)
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 		 (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino)));
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 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
225 
226 	/*
227 	 * Get the last inode value, see if there's nothing to do.
228 	 */
229 	ino = (xfs_ino_t)*lastinop;
230 	lastino = ino;
231 	agno = XFS_INO_TO_AGNO(mp, ino);
232 	agino = XFS_INO_TO_AGINO(mp, ino);
233 	if (agno >= mp->m_sb.sb_agcount ||
234 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
235 		*done = 1;
236 		*ubcountp = 0;
237 		return 0;
238 	}
239 	if (!ubcountp || *ubcountp <= 0) {
240 		return EINVAL;
241 	}
242 	ubcount = *ubcountp; /* statstruct's */
243 	ubleft = ubcount * statstruct_size; /* bytes */
244 	*ubcountp = ubelem = 0;
245 	*done = 0;
246 	fmterror = 0;
247 	ubufp = ubuffer;
248 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
249 		mp->m_sb.sb_inopblock :
250 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
251 	nimask = ~(nicluster - 1);
252 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
253 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4);
254 	if (!irbuf)
255 		return ENOMEM;
256 
257 	nirbuf = irbsize / sizeof(*irbuf);
258 
259 	/*
260 	 * Loop over the allocation groups, starting from the last
261 	 * inode returned; 0 means start of the allocation group.
262 	 */
263 	rval = 0;
264 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
265 		cond_resched();
266 		bp = NULL;
267 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
268 		if (error) {
269 			/*
270 			 * Skip this allocation group and go to the next one.
271 			 */
272 			agno++;
273 			agino = 0;
274 			continue;
275 		}
276 		agi = XFS_BUF_TO_AGI(agbp);
277 		/*
278 		 * Allocate and initialize a btree cursor for ialloc btree.
279 		 */
280 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
281 		irbp = irbuf;
282 		irbufend = irbuf + nirbuf;
283 		end_of_ag = 0;
284 		/*
285 		 * If we're returning in the middle of an allocation group,
286 		 * we need to get the remainder of the chunk we're in.
287 		 */
288 		if (agino > 0) {
289 			xfs_inobt_rec_incore_t r;
290 
291 			/*
292 			 * Lookup the inode chunk that this inode lives in.
293 			 */
294 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
295 						 &tmp);
296 			if (!error &&	/* no I/O error */
297 			    tmp &&	/* lookup succeeded */
298 					/* got the record, should always work */
299 			    !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
300 			    i == 1 &&
301 					/* this is the right chunk */
302 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
303 					/* lastino was not last in chunk */
304 			    (chunkidx = agino - r.ir_startino + 1) <
305 				    XFS_INODES_PER_CHUNK &&
306 					/* there are some left allocated */
307 			    xfs_inobt_maskn(chunkidx,
308 				    XFS_INODES_PER_CHUNK - chunkidx) &
309 				    ~r.ir_free) {
310 				/*
311 				 * Grab the chunk record.  Mark all the
312 				 * uninteresting inodes (because they're
313 				 * before our start point) free.
314 				 */
315 				for (i = 0; i < chunkidx; i++) {
316 					if (XFS_INOBT_MASK(i) & ~r.ir_free)
317 						r.ir_freecount++;
318 				}
319 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
320 				irbp->ir_startino = r.ir_startino;
321 				irbp->ir_freecount = r.ir_freecount;
322 				irbp->ir_free = r.ir_free;
323 				irbp++;
324 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
325 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
326 			} else {
327 				/*
328 				 * If any of those tests failed, bump the
329 				 * inode number (just in case).
330 				 */
331 				agino++;
332 				icount = 0;
333 			}
334 			/*
335 			 * In any case, increment to the next record.
336 			 */
337 			if (!error)
338 				error = xfs_btree_increment(cur, 0, &tmp);
339 		} else {
340 			/*
341 			 * Start of ag.  Lookup the first inode chunk.
342 			 */
343 			error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
344 			icount = 0;
345 		}
346 		/*
347 		 * Loop through inode btree records in this ag,
348 		 * until we run out of inodes or space in the buffer.
349 		 */
350 		while (irbp < irbufend && icount < ubcount) {
351 			xfs_inobt_rec_incore_t r;
352 
353 			/*
354 			 * Loop as long as we're unable to read the
355 			 * inode btree.
356 			 */
357 			while (error) {
358 				agino += XFS_INODES_PER_CHUNK;
359 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
360 						be32_to_cpu(agi->agi_length))
361 					break;
362 				error = xfs_inobt_lookup(cur, agino,
363 							 XFS_LOOKUP_GE, &tmp);
364 				cond_resched();
365 			}
366 			/*
367 			 * If ran off the end of the ag either with an error,
368 			 * or the normal way, set end and stop collecting.
369 			 */
370 			if (error) {
371 				end_of_ag = 1;
372 				break;
373 			}
374 
375 			error = xfs_inobt_get_rec(cur, &r, &i);
376 			if (error || i == 0) {
377 				end_of_ag = 1;
378 				break;
379 			}
380 
381 			/*
382 			 * If this chunk has any allocated inodes, save it.
383 			 * Also start read-ahead now for this chunk.
384 			 */
385 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
386 				/*
387 				 * Loop over all clusters in the next chunk.
388 				 * Do a readahead if there are any allocated
389 				 * inodes in that cluster.
390 				 */
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 				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 				 * Recompute agbno if this is the
438 				 * first inode of the cluster.
439 				 *
440 				 * Careful with clustidx.   There can be
441 				 * multiple clusters per chunk, a single
442 				 * cluster per chunk or a cluster that has
443 				 * inodes represented from several different
444 				 * chunks (if blocksize is large).
445 				 *
446 				 * Because of this, the starting clustidx is
447 				 * initialized to zero in this loop but must
448 				 * later be reset after reading in the cluster
449 				 * buffer.
450 				 */
451 				if ((chunkidx & (nicluster - 1)) == 0) {
452 					agbno = XFS_AGINO_TO_AGBNO(mp,
453 							irbp->ir_startino) +
454 						((chunkidx & nimask) >>
455 						 mp->m_sb.sb_inopblog);
456 				}
457 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
458 				/*
459 				 * Skip if this inode is free.
460 				 */
461 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
462 					lastino = ino;
463 					continue;
464 				}
465 				/*
466 				 * Count used inodes as free so we can tell
467 				 * when the chunk is used up.
468 				 */
469 				irbp->ir_freecount++;
470 
471 				/*
472 				 * Get the inode and fill in a single buffer.
473 				 */
474 				ubused = statstruct_size;
475 				error = formatter(mp, ino, ubufp, ubleft,
476 						  &ubused, &fmterror);
477 				if (fmterror == BULKSTAT_RV_NOTHING) {
478 					if (error && error != ENOENT &&
479 						error != EINVAL) {
480 						ubleft = 0;
481 						rval = error;
482 						break;
483 					}
484 					lastino = ino;
485 					continue;
486 				}
487 				if (fmterror == BULKSTAT_RV_GIVEUP) {
488 					ubleft = 0;
489 					ASSERT(error);
490 					rval = error;
491 					break;
492 				}
493 				if (ubufp)
494 					ubufp += ubused;
495 				ubleft -= ubused;
496 				ubelem++;
497 				lastino = ino;
498 			}
499 
500 			cond_resched();
501 		}
502 
503 		if (bp)
504 			xfs_buf_relse(bp);
505 
506 		/*
507 		 * Set up for the next loop iteration.
508 		 */
509 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
510 			if (end_of_ag) {
511 				agno++;
512 				agino = 0;
513 			} else
514 				agino = XFS_INO_TO_AGINO(mp, lastino);
515 		} else
516 			break;
517 	}
518 	/*
519 	 * Done, we're either out of filesystem or space to put the data.
520 	 */
521 	kmem_free_large(irbuf);
522 	*ubcountp = ubelem;
523 	/*
524 	 * Found some inodes, return them now and return the error next time.
525 	 */
526 	if (ubelem)
527 		rval = 0;
528 	if (agno >= mp->m_sb.sb_agcount) {
529 		/*
530 		 * If we ran out of filesystem, mark lastino as off
531 		 * the end of the filesystem, so the next call
532 		 * will return immediately.
533 		 */
534 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
535 		*done = 1;
536 	} else
537 		*lastinop = (xfs_ino_t)lastino;
538 
539 	return rval;
540 }
541 
542 /*
543  * Return stat information in bulk (by-inode) for the filesystem.
544  * Special case for non-sequential one inode bulkstat.
545  */
546 int					/* error status */
547 xfs_bulkstat_single(
548 	xfs_mount_t		*mp,	/* mount point for filesystem */
549 	xfs_ino_t		*lastinop, /* inode to return */
550 	char			__user *buffer, /* buffer with inode stats */
551 	int			*done)	/* 1 if there are more stats to get */
552 {
553 	int			count;	/* count value for bulkstat call */
554 	int			error;	/* return value */
555 	xfs_ino_t		ino;	/* filesystem inode number */
556 	int			res;	/* result from bs1 */
557 
558 	/*
559 	 * note that requesting valid inode numbers which are not allocated
560 	 * to inodes will most likely cause xfs_imap_to_bp to generate warning
561 	 * messages about bad magic numbers. This is ok. The fact that
562 	 * the inode isn't actually an inode is handled by the
563 	 * error check below. Done this way to make the usual case faster
564 	 * at the expense of the error case.
565 	 */
566 
567 	ino = (xfs_ino_t)*lastinop;
568 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t), 0, &res);
569 	if (error) {
570 		/*
571 		 * Special case way failed, do it the "long" way
572 		 * to see if that works.
573 		 */
574 		(*lastinop)--;
575 		count = 1;
576 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
577 				sizeof(xfs_bstat_t), buffer, done))
578 			return error;
579 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
580 			return error == EFSCORRUPTED ?
581 				XFS_ERROR(EINVAL) : error;
582 		else
583 			return 0;
584 	}
585 	*done = 0;
586 	return 0;
587 }
588 
589 int
590 xfs_inumbers_fmt(
591 	void			__user *ubuffer, /* buffer to write to */
592 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
593 	long			count,		/* # of elements to read */
594 	long			*written)	/* # of bytes written */
595 {
596 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
597 		return -EFAULT;
598 	*written = count * sizeof(*buffer);
599 	return 0;
600 }
601 
602 /*
603  * Return inode number table for the filesystem.
604  */
605 int					/* error status */
606 xfs_inumbers(
607 	xfs_mount_t	*mp,		/* mount point for filesystem */
608 	xfs_ino_t	*lastino,	/* last inode returned */
609 	int		*count,		/* size of buffer/count returned */
610 	void		__user *ubuffer,/* buffer with inode descriptions */
611 	inumbers_fmt_pf	formatter)
612 {
613 	xfs_buf_t	*agbp;
614 	xfs_agino_t	agino;
615 	xfs_agnumber_t	agno;
616 	int		bcount;
617 	xfs_inogrp_t	*buffer;
618 	int		bufidx;
619 	xfs_btree_cur_t	*cur;
620 	int		error;
621 	xfs_inobt_rec_incore_t r;
622 	int		i;
623 	xfs_ino_t	ino;
624 	int		left;
625 	int		tmp;
626 
627 	ino = (xfs_ino_t)*lastino;
628 	agno = XFS_INO_TO_AGNO(mp, ino);
629 	agino = XFS_INO_TO_AGINO(mp, ino);
630 	left = *count;
631 	*count = 0;
632 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
633 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
634 	error = bufidx = 0;
635 	cur = NULL;
636 	agbp = NULL;
637 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
638 		if (agbp == NULL) {
639 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
640 			if (error) {
641 				/*
642 				 * If we can't read the AGI of this ag,
643 				 * then just skip to the next one.
644 				 */
645 				ASSERT(cur == NULL);
646 				agbp = NULL;
647 				agno++;
648 				agino = 0;
649 				continue;
650 			}
651 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
652 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
653 						 &tmp);
654 			if (error) {
655 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
656 				cur = NULL;
657 				xfs_buf_relse(agbp);
658 				agbp = NULL;
659 				/*
660 				 * Move up the last inode in the current
661 				 * chunk.  The lookup_ge will always get
662 				 * us the first inode in the next chunk.
663 				 */
664 				agino += XFS_INODES_PER_CHUNK - 1;
665 				continue;
666 			}
667 		}
668 		error = xfs_inobt_get_rec(cur, &r, &i);
669 		if (error || i == 0) {
670 			xfs_buf_relse(agbp);
671 			agbp = NULL;
672 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
673 			cur = NULL;
674 			agno++;
675 			agino = 0;
676 			continue;
677 		}
678 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
679 		buffer[bufidx].xi_startino =
680 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
681 		buffer[bufidx].xi_alloccount =
682 			XFS_INODES_PER_CHUNK - r.ir_freecount;
683 		buffer[bufidx].xi_allocmask = ~r.ir_free;
684 		bufidx++;
685 		left--;
686 		if (bufidx == bcount) {
687 			long written;
688 			if (formatter(ubuffer, buffer, bufidx, &written)) {
689 				error = XFS_ERROR(EFAULT);
690 				break;
691 			}
692 			ubuffer += written;
693 			*count += bufidx;
694 			bufidx = 0;
695 		}
696 		if (left) {
697 			error = xfs_btree_increment(cur, 0, &tmp);
698 			if (error) {
699 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
700 				cur = NULL;
701 				xfs_buf_relse(agbp);
702 				agbp = NULL;
703 				/*
704 				 * The agino value has already been bumped.
705 				 * Just try to skip up to it.
706 				 */
707 				agino += XFS_INODES_PER_CHUNK;
708 				continue;
709 			}
710 		}
711 	}
712 	if (!error) {
713 		if (bufidx) {
714 			long written;
715 			if (formatter(ubuffer, buffer, bufidx, &written))
716 				error = XFS_ERROR(EFAULT);
717 			else
718 				*count += bufidx;
719 		}
720 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
721 	}
722 	kmem_free(buffer);
723 	if (cur)
724 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
725 					   XFS_BTREE_NOERROR));
726 	if (agbp)
727 		xfs_buf_relse(agbp);
728 	return error;
729 }
730