xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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_dinode_core_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 (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
155 		buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
156 		buf->bs_projid = 0;
157 	} else {
158 		buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
159 		buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
160 	}
161 
162 	buf->bs_ino = ino;
163 	buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
164 	buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
165 	buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
166 	buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
167 	buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
168 	buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
169 	buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
170 	buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
171 	buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
172 	buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
173 	buf->bs_xflags = xfs_dic2xflags(dic);
174 	buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
175 	buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
176 	buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
177 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
178 	buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
179 	buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
180 	buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
181 
182 	switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
183 	case XFS_DINODE_FMT_DEV:
184 		buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
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 = INT_GET(dic->di_nblocks, ARCH_CONVERT);
199 		break;
200 	}
201 
202 	return 0;
203 }
204 
205 /*
206  * Return stat information for one inode.
207  * Return 0 if ok, else errno.
208  */
209 int		       		/* error status */
210 xfs_bulkstat_one(
211 	xfs_mount_t	*mp,		/* mount point for filesystem */
212 	xfs_ino_t	ino,		/* inode number to get data for */
213 	void		__user *buffer,	/* buffer to place output in */
214 	int		ubsize,		/* size of buffer */
215 	void		*private_data,	/* my private data */
216 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
217 	int		*ubused,	/* bytes used by me */
218 	void		*dibuff,	/* on-disk inode buffer */
219 	int		*stat)		/* BULKSTAT_RV_... */
220 {
221 	xfs_bstat_t	*buf;		/* return buffer */
222 	int		error = 0;	/* error value */
223 	xfs_dinode_t	*dip;		/* dinode inode pointer */
224 
225 	dip = (xfs_dinode_t *)dibuff;
226 	*stat = BULKSTAT_RV_NOTHING;
227 
228 	if (!buffer || xfs_internal_inum(mp, ino))
229 		return XFS_ERROR(EINVAL);
230 	if (ubsize < sizeof(*buf))
231 		return XFS_ERROR(ENOMEM);
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 		error = EFAULT;
248 		goto out_free;
249 	}
250 
251 	*stat = BULKSTAT_RV_DIDONE;
252 	if (ubused)
253 		*ubused = sizeof(*buf);
254 
255  out_free:
256 	kmem_free(buf, sizeof(*buf));
257 	return error;
258 }
259 
260 /*
261  * Test to see whether we can use the ondisk inode directly, based
262  * on the given bulkstat flags, filling in dipp accordingly.
263  * Returns zero if the inode is dodgey.
264  */
265 STATIC int
266 xfs_bulkstat_use_dinode(
267 	xfs_mount_t	*mp,
268 	int		flags,
269 	xfs_buf_t	*bp,
270 	int		clustidx,
271 	xfs_dinode_t	**dipp)
272 {
273 	xfs_dinode_t	*dip;
274 	unsigned int	aformat;
275 
276 	*dipp = NULL;
277 	if (!bp || (flags & BULKSTAT_FG_IGET))
278 		return 1;
279 	dip = (xfs_dinode_t *)
280 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
281 	if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC ||
282 	    !XFS_DINODE_GOOD_VERSION(
283 			INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
284 		return 0;
285 	if (flags & BULKSTAT_FG_QUICK) {
286 		*dipp = dip;
287 		return 1;
288 	}
289 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
290 	aformat = INT_GET(dip->di_core.di_aformat, ARCH_CONVERT);
291 	if ((XFS_CFORK_Q(&dip->di_core) == 0) ||
292 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
293 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
294 		*dipp = dip;
295 		return 1;
296 	}
297 	return 1;
298 }
299 
300 /*
301  * Return stat information in bulk (by-inode) for the filesystem.
302  */
303 int					/* error status */
304 xfs_bulkstat(
305 	xfs_mount_t		*mp,	/* mount point for filesystem */
306 	xfs_ino_t		*lastinop, /* last inode returned */
307 	int			*ubcountp, /* size of buffer/count returned */
308 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
309 	void			*private_data,/* private data for formatter */
310 	size_t			statstruct_size, /* sizeof struct filling */
311 	char			__user *ubuffer, /* buffer with inode stats */
312 	int			flags,	/* defined in xfs_itable.h */
313 	int			*done)	/* 1 if there are more stats to get */
314 {
315 	xfs_agblock_t		agbno=0;/* allocation group block number */
316 	xfs_buf_t		*agbp;	/* agi header buffer */
317 	xfs_agi_t		*agi;	/* agi header data */
318 	xfs_agino_t		agino;	/* inode # in allocation group */
319 	xfs_agnumber_t		agno;	/* allocation group number */
320 	xfs_daddr_t		bno;	/* inode cluster start daddr */
321 	int			chunkidx; /* current index into inode chunk */
322 	int			clustidx; /* current index into inode cluster */
323 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
324 	int			end_of_ag; /* set if we've seen the ag end */
325 	int			error;	/* error code */
326 	int                     fmterror;/* bulkstat formatter result */
327 	__int32_t		gcnt;	/* current btree rec's count */
328 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
329 	xfs_agino_t		gino;	/* current btree rec's start inode */
330 	int			i;	/* loop index */
331 	int			icount;	/* count of inodes good in irbuf */
332 	size_t			irbsize; /* size of irec buffer in bytes */
333 	xfs_ino_t		ino;	/* inode number (filesystem) */
334 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
335 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
336 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
337 	xfs_ino_t		lastino=0; /* last inode number returned */
338 	int			nbcluster; /* # of blocks in a cluster */
339 	int			nicluster; /* # of inodes in a cluster */
340 	int			nimask;	/* mask for inode clusters */
341 	int			nirbuf;	/* size of irbuf */
342 	int			rval;	/* return value error code */
343 	int			tmp;	/* result value from btree calls */
344 	int			ubcount; /* size of user's buffer */
345 	int			ubleft;	/* bytes left in user's buffer */
346 	char			__user *ubufp;	/* pointer into user's buffer */
347 	int			ubelem;	/* spaces used in user's buffer */
348 	int			ubused;	/* bytes used by formatter */
349 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
350 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
351 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
352 
353 	/*
354 	 * Get the last inode value, see if there's nothing to do.
355 	 */
356 	ino = (xfs_ino_t)*lastinop;
357 	dip = NULL;
358 	agno = XFS_INO_TO_AGNO(mp, ino);
359 	agino = XFS_INO_TO_AGINO(mp, ino);
360 	if (agno >= mp->m_sb.sb_agcount ||
361 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
362 		*done = 1;
363 		*ubcountp = 0;
364 		return 0;
365 	}
366 	ubcount = *ubcountp; /* statstruct's */
367 	ubleft = ubcount * statstruct_size; /* bytes */
368 	*ubcountp = ubelem = 0;
369 	*done = 0;
370 	fmterror = 0;
371 	ubufp = ubuffer;
372 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
373 		mp->m_sb.sb_inopblock :
374 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
375 	nimask = ~(nicluster - 1);
376 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
377 	irbuf = kmem_zalloc_greedy(&irbsize, NBPC, NBPC * 4,
378 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
379 	nirbuf = irbsize / sizeof(*irbuf);
380 
381 	/*
382 	 * Loop over the allocation groups, starting from the last
383 	 * inode returned; 0 means start of the allocation group.
384 	 */
385 	rval = 0;
386 	while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
387 		bp = NULL;
388 		down_read(&mp->m_peraglock);
389 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
390 		up_read(&mp->m_peraglock);
391 		if (error) {
392 			/*
393 			 * Skip this allocation group and go to the next one.
394 			 */
395 			agno++;
396 			agino = 0;
397 			continue;
398 		}
399 		agi = XFS_BUF_TO_AGI(agbp);
400 		/*
401 		 * Allocate and initialize a btree cursor for ialloc btree.
402 		 */
403 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
404 						(xfs_inode_t *)0, 0);
405 		irbp = irbuf;
406 		irbufend = irbuf + nirbuf;
407 		end_of_ag = 0;
408 		/*
409 		 * If we're returning in the middle of an allocation group,
410 		 * we need to get the remainder of the chunk we're in.
411 		 */
412 		if (agino > 0) {
413 			/*
414 			 * Lookup the inode chunk that this inode lives in.
415 			 */
416 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
417 			if (!error &&	/* no I/O error */
418 			    tmp &&	/* lookup succeeded */
419 					/* got the record, should always work */
420 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
421 				    &gfree, &i)) &&
422 			    i == 1 &&
423 					/* this is the right chunk */
424 			    agino < gino + XFS_INODES_PER_CHUNK &&
425 					/* lastino was not last in chunk */
426 			    (chunkidx = agino - gino + 1) <
427 				    XFS_INODES_PER_CHUNK &&
428 					/* there are some left allocated */
429 			    XFS_INOBT_MASKN(chunkidx,
430 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
431 				/*
432 				 * Grab the chunk record.  Mark all the
433 				 * uninteresting inodes (because they're
434 				 * before our start point) free.
435 				 */
436 				for (i = 0; i < chunkidx; i++) {
437 					if (XFS_INOBT_MASK(i) & ~gfree)
438 						gcnt++;
439 				}
440 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
441 				irbp->ir_startino = gino;
442 				irbp->ir_freecount = gcnt;
443 				irbp->ir_free = gfree;
444 				irbp++;
445 				agino = gino + XFS_INODES_PER_CHUNK;
446 				icount = XFS_INODES_PER_CHUNK - gcnt;
447 			} else {
448 				/*
449 				 * If any of those tests failed, bump the
450 				 * inode number (just in case).
451 				 */
452 				agino++;
453 				icount = 0;
454 			}
455 			/*
456 			 * In any case, increment to the next record.
457 			 */
458 			if (!error)
459 				error = xfs_inobt_increment(cur, 0, &tmp);
460 		} else {
461 			/*
462 			 * Start of ag.  Lookup the first inode chunk.
463 			 */
464 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
465 			icount = 0;
466 		}
467 		/*
468 		 * Loop through inode btree records in this ag,
469 		 * until we run out of inodes or space in the buffer.
470 		 */
471 		while (irbp < irbufend && icount < ubcount) {
472 			/*
473 			 * Loop as long as we're unable to read the
474 			 * inode btree.
475 			 */
476 			while (error) {
477 				agino += XFS_INODES_PER_CHUNK;
478 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
479 						be32_to_cpu(agi->agi_length))
480 					break;
481 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
482 							    &tmp);
483 			}
484 			/*
485 			 * If ran off the end of the ag either with an error,
486 			 * or the normal way, set end and stop collecting.
487 			 */
488 			if (error ||
489 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
490 				    &gfree, &i)) ||
491 			    i == 0) {
492 				end_of_ag = 1;
493 				break;
494 			}
495 			/*
496 			 * If this chunk has any allocated inodes, save it.
497 			 * Also start read-ahead now for this chunk.
498 			 */
499 			if (gcnt < XFS_INODES_PER_CHUNK) {
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, gino),
506 				     chunkidx = 0;
507 				     chunkidx < XFS_INODES_PER_CHUNK;
508 				     chunkidx += nicluster,
509 				     agbno += nbcluster) {
510 					if (XFS_INOBT_MASKN(chunkidx,
511 							    nicluster) & ~gfree)
512 						xfs_btree_reada_bufs(mp, agno,
513 							agbno, nbcluster);
514 				}
515 				irbp->ir_startino = gino;
516 				irbp->ir_freecount = gcnt;
517 				irbp->ir_free = gfree;
518 				irbp++;
519 				icount += XFS_INODES_PER_CHUNK - gcnt;
520 			}
521 			/*
522 			 * Set agino to after this chunk and bump the cursor.
523 			 */
524 			agino = gino + XFS_INODES_PER_CHUNK;
525 			error = xfs_inobt_increment(cur, 0, &tmp);
526 		}
527 		/*
528 		 * Drop the btree buffers and the agi buffer.
529 		 * We can't hold any of the locks these represent
530 		 * when calling iget.
531 		 */
532 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
533 		xfs_buf_relse(agbp);
534 		/*
535 		 * Now format all the good inodes into the user's buffer.
536 		 */
537 		irbufend = irbp;
538 		for (irbp = irbuf;
539 		     irbp < irbufend && ubleft >= statstruct_size; irbp++) {
540 			/*
541 			 * Now process this chunk of inodes.
542 			 */
543 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
544 			     ubleft > 0 &&
545 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
546 			     chunkidx++, clustidx++, agino++) {
547 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
548 				/*
549 				 * Recompute agbno if this is the
550 				 * first inode of the cluster.
551 				 *
552 				 * Careful with clustidx.   There can be
553 				 * multple clusters per chunk, a single
554 				 * cluster per chunk or a cluster that has
555 				 * inodes represented from several different
556 				 * chunks (if blocksize is large).
557 				 *
558 				 * Because of this, the starting clustidx is
559 				 * initialized to zero in this loop but must
560 				 * later be reset after reading in the cluster
561 				 * buffer.
562 				 */
563 				if ((chunkidx & (nicluster - 1)) == 0) {
564 					agbno = XFS_AGINO_TO_AGBNO(mp,
565 							irbp->ir_startino) +
566 						((chunkidx & nimask) >>
567 						 mp->m_sb.sb_inopblog);
568 
569 					if (flags & (BULKSTAT_FG_QUICK |
570 						     BULKSTAT_FG_INLINE)) {
571 						ino = XFS_AGINO_TO_INO(mp, agno,
572 								       agino);
573 						bno = XFS_AGB_TO_DADDR(mp, agno,
574 								       agbno);
575 
576 						/*
577 						 * Get the inode cluster buffer
578 						 */
579 						ASSERT(xfs_inode_zone != NULL);
580 						ip = kmem_zone_zalloc(xfs_inode_zone,
581 								      KM_SLEEP);
582 						ip->i_ino = ino;
583 						ip->i_mount = mp;
584 						spin_lock_init(&ip->i_flags_lock);
585 						if (bp)
586 							xfs_buf_relse(bp);
587 						error = xfs_itobp(mp, NULL, ip,
588 								&dip, &bp, bno,
589 								XFS_IMAP_BULKSTAT);
590 						if (!error)
591 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
592 						kmem_zone_free(xfs_inode_zone, ip);
593 						if (XFS_TEST_ERROR(error != 0,
594 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
595 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
596 							bp = NULL;
597 							ubleft = 0;
598 							rval = error;
599 							break;
600 						}
601 					}
602 				}
603 				/*
604 				 * Skip if this inode is free.
605 				 */
606 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free)
607 					continue;
608 				/*
609 				 * Count used inodes as free so we can tell
610 				 * when the chunk is used up.
611 				 */
612 				irbp->ir_freecount++;
613 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
614 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
615 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
616 							     clustidx, &dip))
617 					continue;
618 				/*
619 				 * If we need to do an iget, cannot hold bp.
620 				 * Drop it, until starting the next cluster.
621 				 */
622 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
623 					if (bp)
624 						xfs_buf_relse(bp);
625 					bp = NULL;
626 				}
627 
628 				/*
629 				 * Get the inode and fill in a single buffer.
630 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
631 				 * BULKSTAT_FG_IGET uses igets.
632 				 * BULKSTAT_FG_INLINE uses dip if we have an
633 				 * inline attr fork, else igets.
634 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
635 				 * This is also used to count inodes/blks, etc
636 				 * in xfs_qm_quotacheck.
637 				 */
638 				ubused = statstruct_size;
639 				error = formatter(mp, ino, ubufp,
640 						ubleft, private_data,
641 						bno, &ubused, dip, &fmterror);
642 				if (fmterror == BULKSTAT_RV_NOTHING) {
643                                         if (error == EFAULT) {
644                                                 ubleft = 0;
645                                                 rval = error;
646                                                 break;
647                                         }
648 					else if (error == ENOMEM)
649 						ubleft = 0;
650 					else
651 						lastino = ino;
652 					continue;
653 				}
654 				if (fmterror == BULKSTAT_RV_GIVEUP) {
655 					ubleft = 0;
656 					ASSERT(error);
657 					rval = error;
658 					break;
659 				}
660 				if (ubufp)
661 					ubufp += ubused;
662 				ubleft -= ubused;
663 				ubelem++;
664 				lastino = ino;
665 			}
666 		}
667 
668 		if (bp)
669 			xfs_buf_relse(bp);
670 
671 		/*
672 		 * Set up for the next loop iteration.
673 		 */
674 		if (ubleft > 0) {
675 			if (end_of_ag) {
676 				agno++;
677 				agino = 0;
678 			} else
679 				agino = XFS_INO_TO_AGINO(mp, lastino);
680 		} else
681 			break;
682 	}
683 	/*
684 	 * Done, we're either out of filesystem or space to put the data.
685 	 */
686 	kmem_free(irbuf, irbsize);
687 	*ubcountp = ubelem;
688 	if (agno >= mp->m_sb.sb_agcount) {
689 		/*
690 		 * If we ran out of filesystem, mark lastino as off
691 		 * the end of the filesystem, so the next call
692 		 * will return immediately.
693 		 */
694 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
695 		*done = 1;
696 	} else
697 		*lastinop = (xfs_ino_t)lastino;
698 
699 	return rval;
700 }
701 
702 /*
703  * Return stat information in bulk (by-inode) for the filesystem.
704  * Special case for non-sequential one inode bulkstat.
705  */
706 int					/* error status */
707 xfs_bulkstat_single(
708 	xfs_mount_t		*mp,	/* mount point for filesystem */
709 	xfs_ino_t		*lastinop, /* inode to return */
710 	char			__user *buffer, /* buffer with inode stats */
711 	int			*done)	/* 1 if there are more stats to get */
712 {
713 	int			count;	/* count value for bulkstat call */
714 	int			error;	/* return value */
715 	xfs_ino_t		ino;	/* filesystem inode number */
716 	int			res;	/* result from bs1 */
717 
718 	/*
719 	 * note that requesting valid inode numbers which are not allocated
720 	 * to inodes will most likely cause xfs_itobp to generate warning
721 	 * messages about bad magic numbers. This is ok. The fact that
722 	 * the inode isn't actually an inode is handled by the
723 	 * error check below. Done this way to make the usual case faster
724 	 * at the expense of the error case.
725 	 */
726 
727 	ino = (xfs_ino_t)*lastinop;
728 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
729 				 NULL, 0, NULL, NULL, &res);
730 	if (error) {
731 		/*
732 		 * Special case way failed, do it the "long" way
733 		 * to see if that works.
734 		 */
735 		(*lastinop)--;
736 		count = 1;
737 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
738 				NULL, sizeof(xfs_bstat_t), buffer,
739 				BULKSTAT_FG_IGET, done))
740 			return error;
741 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
742 			return error == EFSCORRUPTED ?
743 				XFS_ERROR(EINVAL) : error;
744 		else
745 			return 0;
746 	}
747 	*done = 0;
748 	return 0;
749 }
750 
751 /*
752  * Return inode number table for the filesystem.
753  */
754 int					/* error status */
755 xfs_inumbers(
756 	xfs_mount_t	*mp,		/* mount point for filesystem */
757 	xfs_ino_t	*lastino,	/* last inode returned */
758 	int		*count,		/* size of buffer/count returned */
759 	xfs_inogrp_t	__user *ubuffer)/* buffer with inode descriptions */
760 {
761 	xfs_buf_t	*agbp;
762 	xfs_agino_t	agino;
763 	xfs_agnumber_t	agno;
764 	int		bcount;
765 	xfs_inogrp_t	*buffer;
766 	int		bufidx;
767 	xfs_btree_cur_t	*cur;
768 	int		error;
769 	__int32_t	gcnt;
770 	xfs_inofree_t	gfree;
771 	xfs_agino_t	gino;
772 	int		i;
773 	xfs_ino_t	ino;
774 	int		left;
775 	int		tmp;
776 
777 	ino = (xfs_ino_t)*lastino;
778 	agno = XFS_INO_TO_AGNO(mp, ino);
779 	agino = XFS_INO_TO_AGINO(mp, ino);
780 	left = *count;
781 	*count = 0;
782 	bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
783 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
784 	error = bufidx = 0;
785 	cur = NULL;
786 	agbp = NULL;
787 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
788 		if (agbp == NULL) {
789 			down_read(&mp->m_peraglock);
790 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
791 			up_read(&mp->m_peraglock);
792 			if (error) {
793 				/*
794 				 * If we can't read the AGI of this ag,
795 				 * then just skip to the next one.
796 				 */
797 				ASSERT(cur == NULL);
798 				agbp = NULL;
799 				agno++;
800 				agino = 0;
801 				continue;
802 			}
803 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
804 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
805 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
806 			if (error) {
807 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
808 				cur = NULL;
809 				xfs_buf_relse(agbp);
810 				agbp = NULL;
811 				/*
812 				 * Move up the last inode in the current
813 				 * chunk.  The lookup_ge will always get
814 				 * us the first inode in the next chunk.
815 				 */
816 				agino += XFS_INODES_PER_CHUNK - 1;
817 				continue;
818 			}
819 		}
820 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
821 			&i)) ||
822 		    i == 0) {
823 			xfs_buf_relse(agbp);
824 			agbp = NULL;
825 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
826 			cur = NULL;
827 			agno++;
828 			agino = 0;
829 			continue;
830 		}
831 		agino = gino + XFS_INODES_PER_CHUNK - 1;
832 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
833 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
834 		buffer[bufidx].xi_allocmask = ~gfree;
835 		bufidx++;
836 		left--;
837 		if (bufidx == bcount) {
838 			if (copy_to_user(ubuffer, buffer,
839 					bufidx * sizeof(*buffer))) {
840 				error = XFS_ERROR(EFAULT);
841 				break;
842 			}
843 			ubuffer += bufidx;
844 			*count += bufidx;
845 			bufidx = 0;
846 		}
847 		if (left) {
848 			error = xfs_inobt_increment(cur, 0, &tmp);
849 			if (error) {
850 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
851 				cur = NULL;
852 				xfs_buf_relse(agbp);
853 				agbp = NULL;
854 				/*
855 				 * The agino value has already been bumped.
856 				 * Just try to skip up to it.
857 				 */
858 				agino += XFS_INODES_PER_CHUNK;
859 				continue;
860 			}
861 		}
862 	}
863 	if (!error) {
864 		if (bufidx) {
865 			if (copy_to_user(ubuffer, buffer,
866 					bufidx * sizeof(*buffer)))
867 				error = XFS_ERROR(EFAULT);
868 			else
869 				*count += bufidx;
870 		}
871 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
872 	}
873 	kmem_free(buffer, bcount * sizeof(*buffer));
874 	if (cur)
875 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
876 					   XFS_BTREE_NOERROR));
877 	if (agbp)
878 		xfs_buf_relse(agbp);
879 	return error;
880 }
881