xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
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_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_itable.h"
39 #include "xfs_error.h"
40 #include "xfs_btree.h"
41 
42 int
43 xfs_internal_inum(
44 	xfs_mount_t	*mp,
45 	xfs_ino_t	ino)
46 {
47 	return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
48 		(XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
49 		 (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino)));
50 }
51 
52 STATIC int
53 xfs_bulkstat_one_iget(
54 	xfs_mount_t	*mp,		/* mount point for filesystem */
55 	xfs_ino_t	ino,		/* inode number to get data for */
56 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
57 	xfs_bstat_t	*buf,		/* return buffer */
58 	int		*stat)		/* BULKSTAT_RV_... */
59 {
60 	xfs_icdinode_t	*dic;	/* dinode core info pointer */
61 	xfs_inode_t	*ip;		/* incore inode pointer */
62 	bhv_vnode_t	*vp;
63 	int		error;
64 
65 	error = xfs_iget(mp, NULL, ino,
66 			 XFS_IGET_BULKSTAT, XFS_ILOCK_SHARED, &ip, bno);
67 	if (error) {
68 		*stat = BULKSTAT_RV_NOTHING;
69 		return error;
70 	}
71 
72 	ASSERT(ip != NULL);
73 	ASSERT(ip->i_blkno != (xfs_daddr_t)0);
74 	if (ip->i_d.di_mode == 0) {
75 		*stat = BULKSTAT_RV_NOTHING;
76 		error = XFS_ERROR(ENOENT);
77 		goto out_iput;
78 	}
79 
80 	vp = XFS_ITOV(ip);
81 	dic = &ip->i_d;
82 
83 	/* xfs_iget returns the following without needing
84 	 * further change.
85 	 */
86 	buf->bs_nlink = dic->di_nlink;
87 	buf->bs_projid = dic->di_projid;
88 	buf->bs_ino = ino;
89 	buf->bs_mode = dic->di_mode;
90 	buf->bs_uid = dic->di_uid;
91 	buf->bs_gid = dic->di_gid;
92 	buf->bs_size = dic->di_size;
93 	vn_atime_to_bstime(vp, &buf->bs_atime);
94 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
95 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
96 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
97 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
98 	buf->bs_xflags = xfs_ip2xflags(ip);
99 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
100 	buf->bs_extents = dic->di_nextents;
101 	buf->bs_gen = dic->di_gen;
102 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
103 	buf->bs_dmevmask = dic->di_dmevmask;
104 	buf->bs_dmstate = dic->di_dmstate;
105 	buf->bs_aextents = dic->di_anextents;
106 
107 	switch (dic->di_format) {
108 	case XFS_DINODE_FMT_DEV:
109 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
110 		buf->bs_blksize = BLKDEV_IOSIZE;
111 		buf->bs_blocks = 0;
112 		break;
113 	case XFS_DINODE_FMT_LOCAL:
114 	case XFS_DINODE_FMT_UUID:
115 		buf->bs_rdev = 0;
116 		buf->bs_blksize = mp->m_sb.sb_blocksize;
117 		buf->bs_blocks = 0;
118 		break;
119 	case XFS_DINODE_FMT_EXTENTS:
120 	case XFS_DINODE_FMT_BTREE:
121 		buf->bs_rdev = 0;
122 		buf->bs_blksize = mp->m_sb.sb_blocksize;
123 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
124 		break;
125 	}
126 
127  out_iput:
128 	xfs_iput(ip, XFS_ILOCK_SHARED);
129 	return error;
130 }
131 
132 STATIC int
133 xfs_bulkstat_one_dinode(
134 	xfs_mount_t	*mp,		/* mount point for filesystem */
135 	xfs_ino_t	ino,		/* inode number to get data for */
136 	xfs_dinode_t	*dip,		/* dinode inode pointer */
137 	xfs_bstat_t	*buf)		/* return buffer */
138 {
139 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
140 
141 	dic = &dip->di_core;
142 
143 	/*
144 	 * The inode format changed when we moved the link count and
145 	 * made it 32 bits long.  If this is an old format inode,
146 	 * convert it in memory to look like a new one.  If it gets
147 	 * flushed to disk we will convert back before flushing or
148 	 * logging it.  We zero out the new projid field and the old link
149 	 * count field.  We'll handle clearing the pad field (the remains
150 	 * of the old uuid field) when we actually convert the inode to
151 	 * the new format. We don't change the version number so that we
152 	 * can distinguish this from a real new format inode.
153 	 */
154 	if (dic->di_version == XFS_DINODE_VERSION_1) {
155 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
156 		buf->bs_projid = 0;
157 	} else {
158 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
159 		buf->bs_projid = be16_to_cpu(dic->di_projid);
160 	}
161 
162 	buf->bs_ino = ino;
163 	buf->bs_mode = be16_to_cpu(dic->di_mode);
164 	buf->bs_uid = be32_to_cpu(dic->di_uid);
165 	buf->bs_gid = be32_to_cpu(dic->di_gid);
166 	buf->bs_size = be64_to_cpu(dic->di_size);
167 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
168 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
169 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
170 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
171 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
172 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
173 	buf->bs_xflags = xfs_dic2xflags(dic);
174 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
175 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
176 	buf->bs_gen = be32_to_cpu(dic->di_gen);
177 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
178 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
179 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
180 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
181 
182 	switch (dic->di_format) {
183 	case XFS_DINODE_FMT_DEV:
184 		buf->bs_rdev = be32_to_cpu(dip->di_u.di_dev);
185 		buf->bs_blksize = BLKDEV_IOSIZE;
186 		buf->bs_blocks = 0;
187 		break;
188 	case XFS_DINODE_FMT_LOCAL:
189 	case XFS_DINODE_FMT_UUID:
190 		buf->bs_rdev = 0;
191 		buf->bs_blksize = mp->m_sb.sb_blocksize;
192 		buf->bs_blocks = 0;
193 		break;
194 	case XFS_DINODE_FMT_EXTENTS:
195 	case XFS_DINODE_FMT_BTREE:
196 		buf->bs_rdev = 0;
197 		buf->bs_blksize = mp->m_sb.sb_blocksize;
198 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
199 		break;
200 	}
201 
202 	return 0;
203 }
204 
205 STATIC int
206 xfs_bulkstat_one_fmt(
207 	void			__user *ubuffer,
208 	const xfs_bstat_t	*buffer)
209 {
210 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
211 		return -EFAULT;
212 	return sizeof(*buffer);
213 }
214 
215 /*
216  * Return stat information for one inode.
217  * Return 0 if ok, else errno.
218  */
219 int		       		/* error status */
220 xfs_bulkstat_one(
221 	xfs_mount_t	*mp,		/* mount point for filesystem */
222 	xfs_ino_t	ino,		/* inode number to get data for */
223 	void		__user *buffer,	/* buffer to place output in */
224 	int		ubsize,		/* size of buffer */
225 	void		*private_data,	/* my private data */
226 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
227 	int		*ubused,	/* bytes used by me */
228 	void		*dibuff,	/* on-disk inode buffer */
229 	int		*stat)		/* BULKSTAT_RV_... */
230 {
231 	xfs_bstat_t	*buf;		/* return buffer */
232 	int		error = 0;	/* error value */
233 	xfs_dinode_t	*dip;		/* dinode inode pointer */
234 	bulkstat_one_fmt_pf formatter = private_data ? : xfs_bulkstat_one_fmt;
235 
236 	dip = (xfs_dinode_t *)dibuff;
237 	*stat = BULKSTAT_RV_NOTHING;
238 
239 	if (!buffer || xfs_internal_inum(mp, ino))
240 		return XFS_ERROR(EINVAL);
241 	if (ubsize < sizeof(*buf))
242 		return XFS_ERROR(ENOMEM);
243 
244 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
245 
246 	if (dip == NULL) {
247 		/* We're not being passed a pointer to a dinode.  This happens
248 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
249 		 */
250 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
251 		if (error)
252 			goto out_free;
253 	} else {
254 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
255 	}
256 
257 	error = formatter(buffer, buf);
258 	if (error < 0)  {
259 		error = EFAULT;
260 		goto out_free;
261 	}
262 
263 	*stat = BULKSTAT_RV_DIDONE;
264 	if (ubused)
265 		*ubused = error;
266 
267  out_free:
268 	kmem_free(buf, sizeof(*buf));
269 	return error;
270 }
271 
272 /*
273  * Test to see whether we can use the ondisk inode directly, based
274  * on the given bulkstat flags, filling in dipp accordingly.
275  * Returns zero if the inode is dodgey.
276  */
277 STATIC int
278 xfs_bulkstat_use_dinode(
279 	xfs_mount_t	*mp,
280 	int		flags,
281 	xfs_buf_t	*bp,
282 	int		clustidx,
283 	xfs_dinode_t	**dipp)
284 {
285 	xfs_dinode_t	*dip;
286 	unsigned int	aformat;
287 
288 	*dipp = NULL;
289 	if (!bp || (flags & BULKSTAT_FG_IGET))
290 		return 1;
291 	dip = (xfs_dinode_t *)
292 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
293 	/*
294 	 * Check the buffer containing the on-disk inode for di_nlink == 0.
295 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
296 	 * inodes that have their in-core state initialized but not flushed
297 	 * to disk yet. This is a temporary hack that would require a proper
298 	 * fix in the future.
299 	 */
300 	if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC ||
301 	    !XFS_DINODE_GOOD_VERSION(dip->di_core.di_version) ||
302 	    !dip->di_core.di_nlink)
303 		return 0;
304 	if (flags & BULKSTAT_FG_QUICK) {
305 		*dipp = dip;
306 		return 1;
307 	}
308 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
309 	aformat = dip->di_core.di_aformat;
310 	if ((XFS_CFORK_Q(&dip->di_core) == 0) ||
311 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
312 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
313 		*dipp = dip;
314 		return 1;
315 	}
316 	return 1;
317 }
318 
319 /*
320  * Return stat information in bulk (by-inode) for the filesystem.
321  */
322 int					/* error status */
323 xfs_bulkstat(
324 	xfs_mount_t		*mp,	/* mount point for filesystem */
325 	xfs_ino_t		*lastinop, /* last inode returned */
326 	int			*ubcountp, /* size of buffer/count returned */
327 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
328 	void			*private_data,/* private data for formatter */
329 	size_t			statstruct_size, /* sizeof struct filling */
330 	char			__user *ubuffer, /* buffer with inode stats */
331 	int			flags,	/* defined in xfs_itable.h */
332 	int			*done)	/* 1 if there are more stats to get */
333 {
334 	xfs_agblock_t		agbno=0;/* allocation group block number */
335 	xfs_buf_t		*agbp;	/* agi header buffer */
336 	xfs_agi_t		*agi;	/* agi header data */
337 	xfs_agino_t		agino;	/* inode # in allocation group */
338 	xfs_agnumber_t		agno;	/* allocation group number */
339 	xfs_daddr_t		bno;	/* inode cluster start daddr */
340 	int			chunkidx; /* current index into inode chunk */
341 	int			clustidx; /* current index into inode cluster */
342 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
343 	int			end_of_ag; /* set if we've seen the ag end */
344 	int			error;	/* error code */
345 	int                     fmterror;/* bulkstat formatter result */
346 	__int32_t		gcnt;	/* current btree rec's count */
347 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
348 	xfs_agino_t		gino;	/* current btree rec's start inode */
349 	int			i;	/* loop index */
350 	int			icount;	/* count of inodes good in irbuf */
351 	size_t			irbsize; /* size of irec buffer in bytes */
352 	xfs_ino_t		ino;	/* inode number (filesystem) */
353 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
354 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
355 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
356 	xfs_ino_t		lastino=0; /* last inode number returned */
357 	int			nbcluster; /* # of blocks in a cluster */
358 	int			nicluster; /* # of inodes in a cluster */
359 	int			nimask;	/* mask for inode clusters */
360 	int			nirbuf;	/* size of irbuf */
361 	int			rval;	/* return value error code */
362 	int			tmp;	/* result value from btree calls */
363 	int			ubcount; /* size of user's buffer */
364 	int			ubleft;	/* bytes left in user's buffer */
365 	char			__user *ubufp;	/* pointer into user's buffer */
366 	int			ubelem;	/* spaces used in user's buffer */
367 	int			ubused;	/* bytes used by formatter */
368 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
369 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
370 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
371 
372 	/*
373 	 * Get the last inode value, see if there's nothing to do.
374 	 */
375 	ino = (xfs_ino_t)*lastinop;
376 	dip = NULL;
377 	agno = XFS_INO_TO_AGNO(mp, ino);
378 	agino = XFS_INO_TO_AGINO(mp, ino);
379 	if (agno >= mp->m_sb.sb_agcount ||
380 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
381 		*done = 1;
382 		*ubcountp = 0;
383 		return 0;
384 	}
385 	ubcount = *ubcountp; /* statstruct's */
386 	ubleft = ubcount * statstruct_size; /* bytes */
387 	*ubcountp = ubelem = 0;
388 	*done = 0;
389 	fmterror = 0;
390 	ubufp = ubuffer;
391 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
392 		mp->m_sb.sb_inopblock :
393 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
394 	nimask = ~(nicluster - 1);
395 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
396 	irbuf = kmem_zalloc_greedy(&irbsize, NBPC, NBPC * 4,
397 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
398 	nirbuf = irbsize / sizeof(*irbuf);
399 
400 	/*
401 	 * Loop over the allocation groups, starting from the last
402 	 * inode returned; 0 means start of the allocation group.
403 	 */
404 	rval = 0;
405 	while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
406 		bp = NULL;
407 		down_read(&mp->m_peraglock);
408 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
409 		up_read(&mp->m_peraglock);
410 		if (error) {
411 			/*
412 			 * Skip this allocation group and go to the next one.
413 			 */
414 			agno++;
415 			agino = 0;
416 			continue;
417 		}
418 		agi = XFS_BUF_TO_AGI(agbp);
419 		/*
420 		 * Allocate and initialize a btree cursor for ialloc btree.
421 		 */
422 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
423 						(xfs_inode_t *)0, 0);
424 		irbp = irbuf;
425 		irbufend = irbuf + nirbuf;
426 		end_of_ag = 0;
427 		/*
428 		 * If we're returning in the middle of an allocation group,
429 		 * we need to get the remainder of the chunk we're in.
430 		 */
431 		if (agino > 0) {
432 			/*
433 			 * Lookup the inode chunk that this inode lives in.
434 			 */
435 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
436 			if (!error &&	/* no I/O error */
437 			    tmp &&	/* lookup succeeded */
438 					/* got the record, should always work */
439 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
440 				    &gfree, &i)) &&
441 			    i == 1 &&
442 					/* this is the right chunk */
443 			    agino < gino + XFS_INODES_PER_CHUNK &&
444 					/* lastino was not last in chunk */
445 			    (chunkidx = agino - gino + 1) <
446 				    XFS_INODES_PER_CHUNK &&
447 					/* there are some left allocated */
448 			    XFS_INOBT_MASKN(chunkidx,
449 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
450 				/*
451 				 * Grab the chunk record.  Mark all the
452 				 * uninteresting inodes (because they're
453 				 * before our start point) free.
454 				 */
455 				for (i = 0; i < chunkidx; i++) {
456 					if (XFS_INOBT_MASK(i) & ~gfree)
457 						gcnt++;
458 				}
459 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
460 				irbp->ir_startino = gino;
461 				irbp->ir_freecount = gcnt;
462 				irbp->ir_free = gfree;
463 				irbp++;
464 				agino = gino + XFS_INODES_PER_CHUNK;
465 				icount = XFS_INODES_PER_CHUNK - gcnt;
466 			} else {
467 				/*
468 				 * If any of those tests failed, bump the
469 				 * inode number (just in case).
470 				 */
471 				agino++;
472 				icount = 0;
473 			}
474 			/*
475 			 * In any case, increment to the next record.
476 			 */
477 			if (!error)
478 				error = xfs_inobt_increment(cur, 0, &tmp);
479 		} else {
480 			/*
481 			 * Start of ag.  Lookup the first inode chunk.
482 			 */
483 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
484 			icount = 0;
485 		}
486 		/*
487 		 * Loop through inode btree records in this ag,
488 		 * until we run out of inodes or space in the buffer.
489 		 */
490 		while (irbp < irbufend && icount < ubcount) {
491 			/*
492 			 * Loop as long as we're unable to read the
493 			 * inode btree.
494 			 */
495 			while (error) {
496 				agino += XFS_INODES_PER_CHUNK;
497 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
498 						be32_to_cpu(agi->agi_length))
499 					break;
500 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
501 							    &tmp);
502 			}
503 			/*
504 			 * If ran off the end of the ag either with an error,
505 			 * or the normal way, set end and stop collecting.
506 			 */
507 			if (error ||
508 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
509 				    &gfree, &i)) ||
510 			    i == 0) {
511 				end_of_ag = 1;
512 				break;
513 			}
514 			/*
515 			 * If this chunk has any allocated inodes, save it.
516 			 * Also start read-ahead now for this chunk.
517 			 */
518 			if (gcnt < XFS_INODES_PER_CHUNK) {
519 				/*
520 				 * Loop over all clusters in the next chunk.
521 				 * Do a readahead if there are any allocated
522 				 * inodes in that cluster.
523 				 */
524 				for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
525 				     chunkidx = 0;
526 				     chunkidx < XFS_INODES_PER_CHUNK;
527 				     chunkidx += nicluster,
528 				     agbno += nbcluster) {
529 					if (XFS_INOBT_MASKN(chunkidx,
530 							    nicluster) & ~gfree)
531 						xfs_btree_reada_bufs(mp, agno,
532 							agbno, nbcluster);
533 				}
534 				irbp->ir_startino = gino;
535 				irbp->ir_freecount = gcnt;
536 				irbp->ir_free = gfree;
537 				irbp++;
538 				icount += XFS_INODES_PER_CHUNK - gcnt;
539 			}
540 			/*
541 			 * Set agino to after this chunk and bump the cursor.
542 			 */
543 			agino = gino + XFS_INODES_PER_CHUNK;
544 			error = xfs_inobt_increment(cur, 0, &tmp);
545 		}
546 		/*
547 		 * Drop the btree buffers and the agi buffer.
548 		 * We can't hold any of the locks these represent
549 		 * when calling iget.
550 		 */
551 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
552 		xfs_buf_relse(agbp);
553 		/*
554 		 * Now format all the good inodes into the user's buffer.
555 		 */
556 		irbufend = irbp;
557 		for (irbp = irbuf;
558 		     irbp < irbufend && ubleft >= statstruct_size; irbp++) {
559 			/*
560 			 * Now process this chunk of inodes.
561 			 */
562 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
563 			     ubleft > 0 &&
564 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
565 			     chunkidx++, clustidx++, agino++) {
566 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
567 				/*
568 				 * Recompute agbno if this is the
569 				 * first inode of the cluster.
570 				 *
571 				 * Careful with clustidx.   There can be
572 				 * multple clusters per chunk, a single
573 				 * cluster per chunk or a cluster that has
574 				 * inodes represented from several different
575 				 * chunks (if blocksize is large).
576 				 *
577 				 * Because of this, the starting clustidx is
578 				 * initialized to zero in this loop but must
579 				 * later be reset after reading in the cluster
580 				 * buffer.
581 				 */
582 				if ((chunkidx & (nicluster - 1)) == 0) {
583 					agbno = XFS_AGINO_TO_AGBNO(mp,
584 							irbp->ir_startino) +
585 						((chunkidx & nimask) >>
586 						 mp->m_sb.sb_inopblog);
587 
588 					if (flags & (BULKSTAT_FG_QUICK |
589 						     BULKSTAT_FG_INLINE)) {
590 						ino = XFS_AGINO_TO_INO(mp, agno,
591 								       agino);
592 						bno = XFS_AGB_TO_DADDR(mp, agno,
593 								       agbno);
594 
595 						/*
596 						 * Get the inode cluster buffer
597 						 */
598 						ASSERT(xfs_inode_zone != NULL);
599 						ip = kmem_zone_zalloc(xfs_inode_zone,
600 								      KM_SLEEP);
601 						ip->i_ino = ino;
602 						ip->i_mount = mp;
603 						spin_lock_init(&ip->i_flags_lock);
604 						if (bp)
605 							xfs_buf_relse(bp);
606 						error = xfs_itobp(mp, NULL, ip,
607 								&dip, &bp, bno,
608 								XFS_IMAP_BULKSTAT);
609 						if (!error)
610 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
611 						kmem_zone_free(xfs_inode_zone, ip);
612 						if (XFS_TEST_ERROR(error != 0,
613 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
614 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
615 							bp = NULL;
616 							ubleft = 0;
617 							rval = error;
618 							break;
619 						}
620 					}
621 				}
622 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
623 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
624 				/*
625 				 * Skip if this inode is free.
626 				 */
627 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
628 					lastino = ino;
629 					continue;
630 				}
631 				/*
632 				 * Count used inodes as free so we can tell
633 				 * when the chunk is used up.
634 				 */
635 				irbp->ir_freecount++;
636 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
637 							     clustidx, &dip)) {
638 					lastino = ino;
639 					continue;
640 				}
641 				/*
642 				 * If we need to do an iget, cannot hold bp.
643 				 * Drop it, until starting the next cluster.
644 				 */
645 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
646 					if (bp)
647 						xfs_buf_relse(bp);
648 					bp = NULL;
649 				}
650 
651 				/*
652 				 * Get the inode and fill in a single buffer.
653 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
654 				 * BULKSTAT_FG_IGET uses igets.
655 				 * BULKSTAT_FG_INLINE uses dip if we have an
656 				 * inline attr fork, else igets.
657 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
658 				 * This is also used to count inodes/blks, etc
659 				 * in xfs_qm_quotacheck.
660 				 */
661 				ubused = statstruct_size;
662 				error = formatter(mp, ino, ubufp,
663 						ubleft, private_data,
664 						bno, &ubused, dip, &fmterror);
665 				if (fmterror == BULKSTAT_RV_NOTHING) {
666                                         if (error == EFAULT) {
667                                                 ubleft = 0;
668                                                 rval = error;
669                                                 break;
670                                         }
671 					else if (error == ENOMEM)
672 						ubleft = 0;
673 					else
674 						lastino = ino;
675 					continue;
676 				}
677 				if (fmterror == BULKSTAT_RV_GIVEUP) {
678 					ubleft = 0;
679 					ASSERT(error);
680 					rval = error;
681 					break;
682 				}
683 				if (ubufp)
684 					ubufp += ubused;
685 				ubleft -= ubused;
686 				ubelem++;
687 				lastino = ino;
688 			}
689 		}
690 
691 		if (bp)
692 			xfs_buf_relse(bp);
693 
694 		/*
695 		 * Set up for the next loop iteration.
696 		 */
697 		if (ubleft > 0) {
698 			if (end_of_ag) {
699 				agno++;
700 				agino = 0;
701 			}
702 		} else
703 			break;
704 	}
705 	/*
706 	 * Done, we're either out of filesystem or space to put the data.
707 	 */
708 	kmem_free(irbuf, irbsize);
709 	*ubcountp = ubelem;
710 	if (agno >= mp->m_sb.sb_agcount) {
711 		/*
712 		 * If we ran out of filesystem, mark lastino as off
713 		 * the end of the filesystem, so the next call
714 		 * will return immediately.
715 		 */
716 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
717 		*done = 1;
718 	} else
719 		*lastinop = (xfs_ino_t)lastino;
720 
721 	return rval;
722 }
723 
724 /*
725  * Return stat information in bulk (by-inode) for the filesystem.
726  * Special case for non-sequential one inode bulkstat.
727  */
728 int					/* error status */
729 xfs_bulkstat_single(
730 	xfs_mount_t		*mp,	/* mount point for filesystem */
731 	xfs_ino_t		*lastinop, /* inode to return */
732 	char			__user *buffer, /* buffer with inode stats */
733 	int			*done)	/* 1 if there are more stats to get */
734 {
735 	int			count;	/* count value for bulkstat call */
736 	int			error;	/* return value */
737 	xfs_ino_t		ino;	/* filesystem inode number */
738 	int			res;	/* result from bs1 */
739 
740 	/*
741 	 * note that requesting valid inode numbers which are not allocated
742 	 * to inodes will most likely cause xfs_itobp to generate warning
743 	 * messages about bad magic numbers. This is ok. The fact that
744 	 * the inode isn't actually an inode is handled by the
745 	 * error check below. Done this way to make the usual case faster
746 	 * at the expense of the error case.
747 	 */
748 
749 	ino = (xfs_ino_t)*lastinop;
750 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
751 				 NULL, 0, NULL, NULL, &res);
752 	if (error) {
753 		/*
754 		 * Special case way failed, do it the "long" way
755 		 * to see if that works.
756 		 */
757 		(*lastinop)--;
758 		count = 1;
759 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
760 				NULL, sizeof(xfs_bstat_t), buffer,
761 				BULKSTAT_FG_IGET, done))
762 			return error;
763 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
764 			return error == EFSCORRUPTED ?
765 				XFS_ERROR(EINVAL) : error;
766 		else
767 			return 0;
768 	}
769 	*done = 0;
770 	return 0;
771 }
772 
773 int
774 xfs_inumbers_fmt(
775 	void			__user *ubuffer, /* buffer to write to */
776 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
777 	long			count,		/* # of elements to read */
778 	long			*written)	/* # of bytes written */
779 {
780 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
781 		return -EFAULT;
782 	*written = count * sizeof(*buffer);
783 	return 0;
784 }
785 
786 /*
787  * Return inode number table for the filesystem.
788  */
789 int					/* error status */
790 xfs_inumbers(
791 	xfs_mount_t	*mp,		/* mount point for filesystem */
792 	xfs_ino_t	*lastino,	/* last inode returned */
793 	int		*count,		/* size of buffer/count returned */
794 	void		__user *ubuffer,/* buffer with inode descriptions */
795 	inumbers_fmt_pf	formatter)
796 {
797 	xfs_buf_t	*agbp;
798 	xfs_agino_t	agino;
799 	xfs_agnumber_t	agno;
800 	int		bcount;
801 	xfs_inogrp_t	*buffer;
802 	int		bufidx;
803 	xfs_btree_cur_t	*cur;
804 	int		error;
805 	__int32_t	gcnt;
806 	xfs_inofree_t	gfree;
807 	xfs_agino_t	gino;
808 	int		i;
809 	xfs_ino_t	ino;
810 	int		left;
811 	int		tmp;
812 
813 	ino = (xfs_ino_t)*lastino;
814 	agno = XFS_INO_TO_AGNO(mp, ino);
815 	agino = XFS_INO_TO_AGINO(mp, ino);
816 	left = *count;
817 	*count = 0;
818 	bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
819 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
820 	error = bufidx = 0;
821 	cur = NULL;
822 	agbp = NULL;
823 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
824 		if (agbp == NULL) {
825 			down_read(&mp->m_peraglock);
826 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
827 			up_read(&mp->m_peraglock);
828 			if (error) {
829 				/*
830 				 * If we can't read the AGI of this ag,
831 				 * then just skip to the next one.
832 				 */
833 				ASSERT(cur == NULL);
834 				agbp = NULL;
835 				agno++;
836 				agino = 0;
837 				continue;
838 			}
839 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
840 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
841 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
842 			if (error) {
843 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
844 				cur = NULL;
845 				xfs_buf_relse(agbp);
846 				agbp = NULL;
847 				/*
848 				 * Move up the last inode in the current
849 				 * chunk.  The lookup_ge will always get
850 				 * us the first inode in the next chunk.
851 				 */
852 				agino += XFS_INODES_PER_CHUNK - 1;
853 				continue;
854 			}
855 		}
856 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
857 			&i)) ||
858 		    i == 0) {
859 			xfs_buf_relse(agbp);
860 			agbp = NULL;
861 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
862 			cur = NULL;
863 			agno++;
864 			agino = 0;
865 			continue;
866 		}
867 		agino = gino + XFS_INODES_PER_CHUNK - 1;
868 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
869 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
870 		buffer[bufidx].xi_allocmask = ~gfree;
871 		bufidx++;
872 		left--;
873 		if (bufidx == bcount) {
874 			long written;
875 			if (formatter(ubuffer, buffer, bufidx, &written)) {
876 				error = XFS_ERROR(EFAULT);
877 				break;
878 			}
879 			ubuffer += written;
880 			*count += bufidx;
881 			bufidx = 0;
882 		}
883 		if (left) {
884 			error = xfs_inobt_increment(cur, 0, &tmp);
885 			if (error) {
886 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
887 				cur = NULL;
888 				xfs_buf_relse(agbp);
889 				agbp = NULL;
890 				/*
891 				 * The agino value has already been bumped.
892 				 * Just try to skip up to it.
893 				 */
894 				agino += XFS_INODES_PER_CHUNK;
895 				continue;
896 			}
897 		}
898 	}
899 	if (!error) {
900 		if (bufidx) {
901 			long written;
902 			if (formatter(ubuffer, buffer, bufidx, &written))
903 				error = XFS_ERROR(EFAULT);
904 			else
905 				*count += bufidx;
906 		}
907 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
908 	}
909 	kmem_free(buffer, bcount * sizeof(*buffer));
910 	if (cur)
911 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
912 					   XFS_BTREE_NOERROR));
913 	if (agbp)
914 		xfs_buf_relse(agbp);
915 	return error;
916 }
917