xref: /linux/fs/xfs/xfs_itable.c (revision 858259cf7d1c443c836a2022b78cb281f0a9b95e)
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_dir.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_error.h"
42 #include "xfs_btree.h"
43 
44 #ifndef HAVE_USERACC
45 #define useracc(ubuffer, size, flags, foo) (0)
46 #define unuseracc(ubuffer, size, flags)
47 #endif
48 
49 STATIC int
50 xfs_bulkstat_one_iget(
51 	xfs_mount_t	*mp,		/* mount point for filesystem */
52 	xfs_ino_t	ino,		/* inode number to get data for */
53 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
54 	xfs_bstat_t	*buf,		/* return buffer */
55 	int		*stat)		/* BULKSTAT_RV_... */
56 {
57 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
58 	xfs_inode_t	*ip;		/* incore inode pointer */
59 	int		error;
60 
61 	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
62 	if (error) {
63 		*stat = BULKSTAT_RV_NOTHING;
64 		return error;
65 	}
66 
67 	ASSERT(ip != NULL);
68 	ASSERT(ip->i_blkno != (xfs_daddr_t)0);
69 	if (ip->i_d.di_mode == 0) {
70 		*stat = BULKSTAT_RV_NOTHING;
71 		error = XFS_ERROR(ENOENT);
72 		goto out_iput;
73 	}
74 
75 	dic = &ip->i_d;
76 
77 	/* xfs_iget returns the following without needing
78 	 * further change.
79 	 */
80 	buf->bs_nlink = dic->di_nlink;
81 	buf->bs_projid = dic->di_projid;
82 	buf->bs_ino = ino;
83 	buf->bs_mode = dic->di_mode;
84 	buf->bs_uid = dic->di_uid;
85 	buf->bs_gid = dic->di_gid;
86 	buf->bs_size = dic->di_size;
87 	buf->bs_atime.tv_sec = dic->di_atime.t_sec;
88 	buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
89 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
90 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
91 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
92 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
93 	buf->bs_xflags = xfs_ip2xflags(ip);
94 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
95 	buf->bs_extents = dic->di_nextents;
96 	buf->bs_gen = dic->di_gen;
97 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
98 	buf->bs_dmevmask = dic->di_dmevmask;
99 	buf->bs_dmstate = dic->di_dmstate;
100 	buf->bs_aextents = dic->di_anextents;
101 
102 	switch (dic->di_format) {
103 	case XFS_DINODE_FMT_DEV:
104 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
105 		buf->bs_blksize = BLKDEV_IOSIZE;
106 		buf->bs_blocks = 0;
107 		break;
108 	case XFS_DINODE_FMT_LOCAL:
109 	case XFS_DINODE_FMT_UUID:
110 		buf->bs_rdev = 0;
111 		buf->bs_blksize = mp->m_sb.sb_blocksize;
112 		buf->bs_blocks = 0;
113 		break;
114 	case XFS_DINODE_FMT_EXTENTS:
115 	case XFS_DINODE_FMT_BTREE:
116 		buf->bs_rdev = 0;
117 		buf->bs_blksize = mp->m_sb.sb_blocksize;
118 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
119 		break;
120 	}
121 
122  out_iput:
123 	xfs_iput(ip, XFS_ILOCK_SHARED);
124 	return error;
125 }
126 
127 STATIC int
128 xfs_bulkstat_one_dinode(
129 	xfs_mount_t	*mp,		/* mount point for filesystem */
130 	xfs_ino_t	ino,		/* inode number to get data for */
131 	xfs_dinode_t	*dip,		/* dinode inode pointer */
132 	xfs_bstat_t	*buf)		/* return buffer */
133 {
134 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
135 
136 	dic = &dip->di_core;
137 
138 	/*
139 	 * The inode format changed when we moved the link count and
140 	 * made it 32 bits long.  If this is an old format inode,
141 	 * convert it in memory to look like a new one.  If it gets
142 	 * flushed to disk we will convert back before flushing or
143 	 * logging it.  We zero out the new projid field and the old link
144 	 * count field.  We'll handle clearing the pad field (the remains
145 	 * of the old uuid field) when we actually convert the inode to
146 	 * the new format. We don't change the version number so that we
147 	 * can distinguish this from a real new format inode.
148 	 */
149 	if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
150 		buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
151 		buf->bs_projid = 0;
152 	} else {
153 		buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
154 		buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
155 	}
156 
157 	buf->bs_ino = ino;
158 	buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
159 	buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
160 	buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
161 	buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
162 	buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
163 	buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
164 	buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
165 	buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
166 	buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
167 	buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
168 	buf->bs_xflags = xfs_dic2xflags(dic);
169 	buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
170 	buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
171 	buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
172 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
173 	buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
174 	buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
175 	buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
176 
177 	switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
178 	case XFS_DINODE_FMT_DEV:
179 		buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
180 		buf->bs_blksize = BLKDEV_IOSIZE;
181 		buf->bs_blocks = 0;
182 		break;
183 	case XFS_DINODE_FMT_LOCAL:
184 	case XFS_DINODE_FMT_UUID:
185 		buf->bs_rdev = 0;
186 		buf->bs_blksize = mp->m_sb.sb_blocksize;
187 		buf->bs_blocks = 0;
188 		break;
189 	case XFS_DINODE_FMT_EXTENTS:
190 	case XFS_DINODE_FMT_BTREE:
191 		buf->bs_rdev = 0;
192 		buf->bs_blksize = mp->m_sb.sb_blocksize;
193 		buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
194 		break;
195 	}
196 
197 	return 0;
198 }
199 
200 /*
201  * Return stat information for one inode.
202  * Return 0 if ok, else errno.
203  */
204 int		       		/* error status */
205 xfs_bulkstat_one(
206 	xfs_mount_t	*mp,		/* mount point for filesystem */
207 	xfs_ino_t	ino,		/* inode number to get data for */
208 	void		__user *buffer,	/* buffer to place output in */
209 	int		ubsize,		/* size of buffer */
210 	void		*private_data,	/* my private data */
211 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
212 	int		*ubused,	/* bytes used by me */
213 	void		*dibuff,	/* on-disk inode buffer */
214 	int		*stat)		/* BULKSTAT_RV_... */
215 {
216 	xfs_bstat_t	*buf;		/* return buffer */
217 	int		error = 0;	/* error value */
218 	xfs_dinode_t	*dip;		/* dinode inode pointer */
219 
220 	dip = (xfs_dinode_t *)dibuff;
221 
222 	if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
223 	    (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
224 	     (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
225 		*stat = BULKSTAT_RV_NOTHING;
226 		return XFS_ERROR(EINVAL);
227 	}
228 	if (ubsize < sizeof(*buf)) {
229 		*stat = BULKSTAT_RV_NOTHING;
230 		return XFS_ERROR(ENOMEM);
231 	}
232 
233 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
234 
235 	if (dip == NULL) {
236 		/* We're not being passed a pointer to a dinode.  This happens
237 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
238 		 */
239 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
240 		if (error)
241 			goto out_free;
242 	} else {
243 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
244 	}
245 
246 	if (copy_to_user(buffer, buf, sizeof(*buf)))  {
247 		*stat = BULKSTAT_RV_NOTHING;
248 		error =  EFAULT;
249 		goto out_free;
250 	}
251 
252 	*stat = BULKSTAT_RV_DIDONE;
253 	if (ubused)
254 		*ubused = sizeof(*buf);
255 
256  out_free:
257 	kmem_free(buf, sizeof(*buf));
258 	return error;
259 }
260 
261 /*
262  * Return stat information in bulk (by-inode) for the filesystem.
263  */
264 int					/* error status */
265 xfs_bulkstat(
266 	xfs_mount_t		*mp,	/* mount point for filesystem */
267 	xfs_ino_t		*lastinop, /* last inode returned */
268 	int			*ubcountp, /* size of buffer/count returned */
269 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
270 	void			*private_data,/* private data for formatter */
271 	size_t			statstruct_size, /* sizeof struct filling */
272 	char			__user *ubuffer, /* buffer with inode stats */
273 	int			flags,	/* defined in xfs_itable.h */
274 	int			*done)	/* 1 if there're more stats to get */
275 {
276 	xfs_agblock_t		agbno=0;/* allocation group block number */
277 	xfs_buf_t		*agbp;	/* agi header buffer */
278 	xfs_agi_t		*agi;	/* agi header data */
279 	xfs_agino_t		agino;	/* inode # in allocation group */
280 	xfs_agnumber_t		agno;	/* allocation group number */
281 	xfs_daddr_t		bno;	/* inode cluster start daddr */
282 	int			chunkidx; /* current index into inode chunk */
283 	int			clustidx; /* current index into inode cluster */
284 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
285 	int			end_of_ag; /* set if we've seen the ag end */
286 	int			error;	/* error code */
287 	int                     fmterror;/* bulkstat formatter result */
288 	__int32_t		gcnt;	/* current btree rec's count */
289 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
290 	xfs_agino_t		gino;	/* current btree rec's start inode */
291 	int			i;	/* loop index */
292 	int			icount;	/* count of inodes good in irbuf */
293 	xfs_ino_t		ino;	/* inode number (filesystem) */
294 	xfs_inobt_rec_t		*irbp;	/* current irec buffer pointer */
295 	xfs_inobt_rec_t		*irbuf;	/* start of irec buffer */
296 	xfs_inobt_rec_t		*irbufend; /* end of good irec buffer entries */
297 	xfs_ino_t		lastino=0; /* last inode number returned */
298 	int			nbcluster; /* # of blocks in a cluster */
299 	int			nicluster; /* # of inodes in a cluster */
300 	int			nimask;	/* mask for inode clusters */
301 	int			nirbuf;	/* size of irbuf */
302 	int			rval;	/* return value error code */
303 	int			tmp;	/* result value from btree calls */
304 	int			ubcount; /* size of user's buffer */
305 	int			ubleft;	/* bytes left in user's buffer */
306 	char			__user *ubufp;	/* pointer into user's buffer */
307 	int			ubelem;	/* spaces used in user's buffer */
308 	int			ubused;	/* bytes used by formatter */
309 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
310 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
311 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
312 
313 	/*
314 	 * Get the last inode value, see if there's nothing to do.
315 	 */
316 	ino = (xfs_ino_t)*lastinop;
317 	dip = NULL;
318 	agno = XFS_INO_TO_AGNO(mp, ino);
319 	agino = XFS_INO_TO_AGINO(mp, ino);
320 	if (agno >= mp->m_sb.sb_agcount ||
321 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
322 		*done = 1;
323 		*ubcountp = 0;
324 		return 0;
325 	}
326 	ubcount = *ubcountp; /* statstruct's */
327 	ubleft = ubcount * statstruct_size; /* bytes */
328 	*ubcountp = ubelem = 0;
329 	*done = 0;
330 	fmterror = 0;
331 	ubufp = ubuffer;
332 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
333 		mp->m_sb.sb_inopblock :
334 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
335 	nimask = ~(nicluster - 1);
336 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
337 	/*
338 	 * Lock down the user's buffer. If a buffer was not sent, as in the case
339 	 * disk quota code calls here, we skip this.
340 	 */
341 	if (ubuffer &&
342 	    (error = useracc(ubuffer, ubcount * statstruct_size,
343 			(B_READ|B_PHYS), NULL))) {
344 		return error;
345 	}
346 	/*
347 	 * Allocate a page-sized buffer for inode btree records.
348 	 * We could try allocating something smaller, but for normal
349 	 * calls we'll always (potentially) need the whole page.
350 	 */
351 	irbuf = kmem_alloc(NBPC, KM_SLEEP);
352 	nirbuf = NBPC / sizeof(*irbuf);
353 	/*
354 	 * Loop over the allocation groups, starting from the last
355 	 * inode returned; 0 means start of the allocation group.
356 	 */
357 	rval = 0;
358 	while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
359 		bp = NULL;
360 		down_read(&mp->m_peraglock);
361 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
362 		up_read(&mp->m_peraglock);
363 		if (error) {
364 			/*
365 			 * Skip this allocation group and go to the next one.
366 			 */
367 			agno++;
368 			agino = 0;
369 			continue;
370 		}
371 		agi = XFS_BUF_TO_AGI(agbp);
372 		/*
373 		 * Allocate and initialize a btree cursor for ialloc btree.
374 		 */
375 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
376 			(xfs_inode_t *)0, 0);
377 		irbp = irbuf;
378 		irbufend = irbuf + nirbuf;
379 		end_of_ag = 0;
380 		/*
381 		 * If we're returning in the middle of an allocation group,
382 		 * we need to get the remainder of the chunk we're in.
383 		 */
384 		if (agino > 0) {
385 			/*
386 			 * Lookup the inode chunk that this inode lives in.
387 			 */
388 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
389 			if (!error &&	/* no I/O error */
390 			    tmp &&	/* lookup succeeded */
391 					/* got the record, should always work */
392 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
393 				    &gfree, &i)) &&
394 			    i == 1 &&
395 					/* this is the right chunk */
396 			    agino < gino + XFS_INODES_PER_CHUNK &&
397 					/* lastino was not last in chunk */
398 			    (chunkidx = agino - gino + 1) <
399 				    XFS_INODES_PER_CHUNK &&
400 					/* there are some left allocated */
401 			    XFS_INOBT_MASKN(chunkidx,
402 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
403 				/*
404 				 * Grab the chunk record.  Mark all the
405 				 * uninteresting inodes (because they're
406 				 * before our start point) free.
407 				 */
408 				for (i = 0; i < chunkidx; i++) {
409 					if (XFS_INOBT_MASK(i) & ~gfree)
410 						gcnt++;
411 				}
412 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
413 				INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
414 				INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
415 				INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
416 				irbp++;
417 				agino = gino + XFS_INODES_PER_CHUNK;
418 				icount = XFS_INODES_PER_CHUNK - gcnt;
419 			} else {
420 				/*
421 				 * If any of those tests failed, bump the
422 				 * inode number (just in case).
423 				 */
424 				agino++;
425 				icount = 0;
426 			}
427 			/*
428 			 * In any case, increment to the next record.
429 			 */
430 			if (!error)
431 				error = xfs_inobt_increment(cur, 0, &tmp);
432 		} else {
433 			/*
434 			 * Start of ag.  Lookup the first inode chunk.
435 			 */
436 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
437 			icount = 0;
438 		}
439 		/*
440 		 * Loop through inode btree records in this ag,
441 		 * until we run out of inodes or space in the buffer.
442 		 */
443 		while (irbp < irbufend && icount < ubcount) {
444 			/*
445 			 * Loop as long as we're unable to read the
446 			 * inode btree.
447 			 */
448 			while (error) {
449 				agino += XFS_INODES_PER_CHUNK;
450 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
451 						be32_to_cpu(agi->agi_length))
452 					break;
453 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
454 							    &tmp);
455 			}
456 			/*
457 			 * If ran off the end of the ag either with an error,
458 			 * or the normal way, set end and stop collecting.
459 			 */
460 			if (error ||
461 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
462 				    &gfree, &i)) ||
463 			    i == 0) {
464 				end_of_ag = 1;
465 				break;
466 			}
467 			/*
468 			 * If this chunk has any allocated inodes, save it.
469 			 */
470 			if (gcnt < XFS_INODES_PER_CHUNK) {
471 				INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
472 				INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
473 				INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
474 				irbp++;
475 				icount += XFS_INODES_PER_CHUNK - gcnt;
476 			}
477 			/*
478 			 * Set agino to after this chunk and bump the cursor.
479 			 */
480 			agino = gino + XFS_INODES_PER_CHUNK;
481 			error = xfs_inobt_increment(cur, 0, &tmp);
482 		}
483 		/*
484 		 * Drop the btree buffers and the agi buffer.
485 		 * We can't hold any of the locks these represent
486 		 * when calling iget.
487 		 */
488 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
489 		xfs_buf_relse(agbp);
490 		/*
491 		 * Now format all the good inodes into the user's buffer.
492 		 */
493 		irbufend = irbp;
494 		for (irbp = irbuf;
495 		     irbp < irbufend && ubleft >= statstruct_size; irbp++) {
496 			/*
497 			 * Read-ahead the next chunk's worth of inodes.
498 			 */
499 			if (&irbp[1] < irbufend) {
500 				/*
501 				 * Loop over all clusters in the next chunk.
502 				 * Do a readahead if there are any allocated
503 				 * inodes in that cluster.
504 				 */
505 				for (agbno = XFS_AGINO_TO_AGBNO(mp,
506 							INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
507 				     chunkidx = 0;
508 				     chunkidx < XFS_INODES_PER_CHUNK;
509 				     chunkidx += nicluster,
510 				     agbno += nbcluster) {
511 					if (XFS_INOBT_MASKN(chunkidx,
512 							    nicluster) &
513 					    ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
514 						xfs_btree_reada_bufs(mp, agno,
515 							agbno, nbcluster);
516 				}
517 			}
518 			/*
519 			 * Now process this chunk of inodes.
520 			 */
521 			for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
522 			     ubleft > 0 &&
523 				INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
524 			     chunkidx++, clustidx++, agino++) {
525 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
526 				/*
527 				 * Recompute agbno if this is the
528 				 * first inode of the cluster.
529 				 *
530 				 * Careful with clustidx.   There can be
531 				 * multple clusters per chunk, a single
532 				 * cluster per chunk or a cluster that has
533 				 * inodes represented from several different
534 				 * chunks (if blocksize is large).
535 				 *
536 				 * Because of this, the starting clustidx is
537 				 * initialized to zero in this loop but must
538 				 * later be reset after reading in the cluster
539 				 * buffer.
540 				 */
541 				if ((chunkidx & (nicluster - 1)) == 0) {
542 					agbno = XFS_AGINO_TO_AGBNO(mp,
543 							INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
544 						((chunkidx & nimask) >>
545 						 mp->m_sb.sb_inopblog);
546 
547 					if (flags & BULKSTAT_FG_QUICK) {
548 						ino = XFS_AGINO_TO_INO(mp, agno,
549 								       agino);
550 						bno = XFS_AGB_TO_DADDR(mp, agno,
551 								       agbno);
552 
553 						/*
554 						 * Get the inode cluster buffer
555 						 */
556 						ASSERT(xfs_inode_zone != NULL);
557 						ip = kmem_zone_zalloc(xfs_inode_zone,
558 								      KM_SLEEP);
559 						ip->i_ino = ino;
560 						ip->i_mount = mp;
561 						if (bp)
562 							xfs_buf_relse(bp);
563 						error = xfs_itobp(mp, NULL, ip,
564 								  &dip, &bp, bno);
565 						if (!error)
566 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
567 						kmem_zone_free(xfs_inode_zone, ip);
568 						if (XFS_TEST_ERROR(error != 0,
569 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
570 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
571 							bp = NULL;
572 							break;
573 						}
574 					}
575 				}
576 				/*
577 				 * Skip if this inode is free.
578 				 */
579 				if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
580 					continue;
581 				/*
582 				 * Count used inodes as free so we can tell
583 				 * when the chunk is used up.
584 				 */
585 				INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
586 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
587 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
588 				if (flags & BULKSTAT_FG_QUICK) {
589 					dip = (xfs_dinode_t *)xfs_buf_offset(bp,
590 					      (clustidx << mp->m_sb.sb_inodelog));
591 
592 					if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
593 						    != XFS_DINODE_MAGIC
594 					    || !XFS_DINODE_GOOD_VERSION(
595 						    INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
596 						continue;
597 				}
598 
599 				/*
600 				 * Get the inode and fill in a single buffer.
601 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
602 				 * BULKSTAT_FG_IGET uses igets.
603 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
604 				 * This is also used to count inodes/blks, etc
605 				 * in xfs_qm_quotacheck.
606 				 */
607 				ubused = statstruct_size;
608 				error = formatter(mp, ino, ubufp,
609 						ubleft, private_data,
610 						bno, &ubused, dip, &fmterror);
611 				if (fmterror == BULKSTAT_RV_NOTHING) {
612 					if (error == ENOMEM)
613 						ubleft = 0;
614 					continue;
615 				}
616 				if (fmterror == BULKSTAT_RV_GIVEUP) {
617 					ubleft = 0;
618 					ASSERT(error);
619 					rval = error;
620 					break;
621 				}
622 				if (ubufp)
623 					ubufp += ubused;
624 				ubleft -= ubused;
625 				ubelem++;
626 				lastino = ino;
627 			}
628 		}
629 
630 		if (bp)
631 			xfs_buf_relse(bp);
632 
633 		/*
634 		 * Set up for the next loop iteration.
635 		 */
636 		if (ubleft > 0) {
637 			if (end_of_ag) {
638 				agno++;
639 				agino = 0;
640 			} else
641 				agino = XFS_INO_TO_AGINO(mp, lastino);
642 		} else
643 			break;
644 	}
645 	/*
646 	 * Done, we're either out of filesystem or space to put the data.
647 	 */
648 	kmem_free(irbuf, NBPC);
649 	if (ubuffer)
650 		unuseracc(ubuffer, ubcount * statstruct_size, (B_READ|B_PHYS));
651 	*ubcountp = ubelem;
652 	if (agno >= mp->m_sb.sb_agcount) {
653 		/*
654 		 * If we ran out of filesystem, mark lastino as off
655 		 * the end of the filesystem, so the next call
656 		 * will return immediately.
657 		 */
658 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
659 		*done = 1;
660 	} else
661 		*lastinop = (xfs_ino_t)lastino;
662 
663 	return rval;
664 }
665 
666 /*
667  * Return stat information in bulk (by-inode) for the filesystem.
668  * Special case for non-sequential one inode bulkstat.
669  */
670 int					/* error status */
671 xfs_bulkstat_single(
672 	xfs_mount_t		*mp,	/* mount point for filesystem */
673 	xfs_ino_t		*lastinop, /* inode to return */
674 	char			__user *buffer, /* buffer with inode stats */
675 	int			*done)	/* 1 if there're more stats to get */
676 {
677 	int			count;	/* count value for bulkstat call */
678 	int			error;	/* return value */
679 	xfs_ino_t		ino;	/* filesystem inode number */
680 	int			res;	/* result from bs1 */
681 
682 	/*
683 	 * note that requesting valid inode numbers which are not allocated
684 	 * to inodes will most likely cause xfs_itobp to generate warning
685 	 * messages about bad magic numbers. This is ok. The fact that
686 	 * the inode isn't actually an inode is handled by the
687 	 * error check below. Done this way to make the usual case faster
688 	 * at the expense of the error case.
689 	 */
690 
691 	ino = (xfs_ino_t)*lastinop;
692 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
693 				 NULL, 0, NULL, NULL, &res);
694 	if (error) {
695 		/*
696 		 * Special case way failed, do it the "long" way
697 		 * to see if that works.
698 		 */
699 		(*lastinop)--;
700 		count = 1;
701 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
702 				NULL, sizeof(xfs_bstat_t), buffer,
703 				BULKSTAT_FG_IGET, done))
704 			return error;
705 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
706 			return error == EFSCORRUPTED ?
707 				XFS_ERROR(EINVAL) : error;
708 		else
709 			return 0;
710 	}
711 	*done = 0;
712 	return 0;
713 }
714 
715 /*
716  * Return inode number table for the filesystem.
717  */
718 int					/* error status */
719 xfs_inumbers(
720 	xfs_mount_t	*mp,		/* mount point for filesystem */
721 	xfs_ino_t	*lastino,	/* last inode returned */
722 	int		*count,		/* size of buffer/count returned */
723 	xfs_inogrp_t	__user *ubuffer)/* buffer with inode descriptions */
724 {
725 	xfs_buf_t	*agbp;
726 	xfs_agino_t	agino;
727 	xfs_agnumber_t	agno;
728 	int		bcount;
729 	xfs_inogrp_t	*buffer;
730 	int		bufidx;
731 	xfs_btree_cur_t	*cur;
732 	int		error;
733 	__int32_t	gcnt;
734 	xfs_inofree_t	gfree;
735 	xfs_agino_t	gino;
736 	int		i;
737 	xfs_ino_t	ino;
738 	int		left;
739 	int		tmp;
740 
741 	ino = (xfs_ino_t)*lastino;
742 	agno = XFS_INO_TO_AGNO(mp, ino);
743 	agino = XFS_INO_TO_AGINO(mp, ino);
744 	left = *count;
745 	*count = 0;
746 	bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
747 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
748 	error = bufidx = 0;
749 	cur = NULL;
750 	agbp = NULL;
751 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
752 		if (agbp == NULL) {
753 			down_read(&mp->m_peraglock);
754 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
755 			up_read(&mp->m_peraglock);
756 			if (error) {
757 				/*
758 				 * If we can't read the AGI of this ag,
759 				 * then just skip to the next one.
760 				 */
761 				ASSERT(cur == NULL);
762 				agbp = NULL;
763 				agno++;
764 				agino = 0;
765 				continue;
766 			}
767 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
768 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
769 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
770 			if (error) {
771 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
772 				cur = NULL;
773 				xfs_buf_relse(agbp);
774 				agbp = NULL;
775 				/*
776 				 * Move up the the last inode in the current
777 				 * chunk.  The lookup_ge will always get
778 				 * us the first inode in the next chunk.
779 				 */
780 				agino += XFS_INODES_PER_CHUNK - 1;
781 				continue;
782 			}
783 		}
784 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
785 			&i)) ||
786 		    i == 0) {
787 			xfs_buf_relse(agbp);
788 			agbp = NULL;
789 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
790 			cur = NULL;
791 			agno++;
792 			agino = 0;
793 			continue;
794 		}
795 		agino = gino + XFS_INODES_PER_CHUNK - 1;
796 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
797 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
798 		buffer[bufidx].xi_allocmask = ~gfree;
799 		bufidx++;
800 		left--;
801 		if (bufidx == bcount) {
802 			if (copy_to_user(ubuffer, buffer,
803 					bufidx * sizeof(*buffer))) {
804 				error = XFS_ERROR(EFAULT);
805 				break;
806 			}
807 			ubuffer += bufidx;
808 			*count += bufidx;
809 			bufidx = 0;
810 		}
811 		if (left) {
812 			error = xfs_inobt_increment(cur, 0, &tmp);
813 			if (error) {
814 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
815 				cur = NULL;
816 				xfs_buf_relse(agbp);
817 				agbp = NULL;
818 				/*
819 				 * The agino value has already been bumped.
820 				 * Just try to skip up to it.
821 				 */
822 				agino += XFS_INODES_PER_CHUNK;
823 				continue;
824 			}
825 		}
826 	}
827 	if (!error) {
828 		if (bufidx) {
829 			if (copy_to_user(ubuffer, buffer,
830 					bufidx * sizeof(*buffer)))
831 				error = XFS_ERROR(EFAULT);
832 			else
833 				*count += bufidx;
834 		}
835 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
836 	}
837 	kmem_free(buffer, bcount * sizeof(*buffer));
838 	if (cur)
839 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
840 					   XFS_BTREE_NOERROR));
841 	if (agbp)
842 		xfs_buf_relse(agbp);
843 	return error;
844 }
845