xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision c21b37f6)
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 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 	if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC ||
294 	    !XFS_DINODE_GOOD_VERSION(
295 			INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
296 		return 0;
297 	if (flags & BULKSTAT_FG_QUICK) {
298 		*dipp = dip;
299 		return 1;
300 	}
301 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
302 	aformat = INT_GET(dip->di_core.di_aformat, ARCH_CONVERT);
303 	if ((XFS_CFORK_Q(&dip->di_core) == 0) ||
304 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
305 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
306 		*dipp = dip;
307 		return 1;
308 	}
309 	return 1;
310 }
311 
312 /*
313  * Return stat information in bulk (by-inode) for the filesystem.
314  */
315 int					/* error status */
316 xfs_bulkstat(
317 	xfs_mount_t		*mp,	/* mount point for filesystem */
318 	xfs_ino_t		*lastinop, /* last inode returned */
319 	int			*ubcountp, /* size of buffer/count returned */
320 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
321 	void			*private_data,/* private data for formatter */
322 	size_t			statstruct_size, /* sizeof struct filling */
323 	char			__user *ubuffer, /* buffer with inode stats */
324 	int			flags,	/* defined in xfs_itable.h */
325 	int			*done)	/* 1 if there are more stats to get */
326 {
327 	xfs_agblock_t		agbno=0;/* allocation group block number */
328 	xfs_buf_t		*agbp;	/* agi header buffer */
329 	xfs_agi_t		*agi;	/* agi header data */
330 	xfs_agino_t		agino;	/* inode # in allocation group */
331 	xfs_agnumber_t		agno;	/* allocation group number */
332 	xfs_daddr_t		bno;	/* inode cluster start daddr */
333 	int			chunkidx; /* current index into inode chunk */
334 	int			clustidx; /* current index into inode cluster */
335 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
336 	int			end_of_ag; /* set if we've seen the ag end */
337 	int			error;	/* error code */
338 	int                     fmterror;/* bulkstat formatter result */
339 	__int32_t		gcnt;	/* current btree rec's count */
340 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
341 	xfs_agino_t		gino;	/* current btree rec's start inode */
342 	int			i;	/* loop index */
343 	int			icount;	/* count of inodes good in irbuf */
344 	size_t			irbsize; /* size of irec buffer in bytes */
345 	xfs_ino_t		ino;	/* inode number (filesystem) */
346 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
347 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
348 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
349 	xfs_ino_t		lastino=0; /* last inode number returned */
350 	int			nbcluster; /* # of blocks in a cluster */
351 	int			nicluster; /* # of inodes in a cluster */
352 	int			nimask;	/* mask for inode clusters */
353 	int			nirbuf;	/* size of irbuf */
354 	int			rval;	/* return value error code */
355 	int			tmp;	/* result value from btree calls */
356 	int			ubcount; /* size of user's buffer */
357 	int			ubleft;	/* bytes left in user's buffer */
358 	char			__user *ubufp;	/* pointer into user's buffer */
359 	int			ubelem;	/* spaces used in user's buffer */
360 	int			ubused;	/* bytes used by formatter */
361 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
362 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
363 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
364 
365 	/*
366 	 * Get the last inode value, see if there's nothing to do.
367 	 */
368 	ino = (xfs_ino_t)*lastinop;
369 	dip = NULL;
370 	agno = XFS_INO_TO_AGNO(mp, ino);
371 	agino = XFS_INO_TO_AGINO(mp, ino);
372 	if (agno >= mp->m_sb.sb_agcount ||
373 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
374 		*done = 1;
375 		*ubcountp = 0;
376 		return 0;
377 	}
378 	ubcount = *ubcountp; /* statstruct's */
379 	ubleft = ubcount * statstruct_size; /* bytes */
380 	*ubcountp = ubelem = 0;
381 	*done = 0;
382 	fmterror = 0;
383 	ubufp = ubuffer;
384 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
385 		mp->m_sb.sb_inopblock :
386 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
387 	nimask = ~(nicluster - 1);
388 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
389 	irbuf = kmem_zalloc_greedy(&irbsize, NBPC, NBPC * 4,
390 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
391 	nirbuf = irbsize / sizeof(*irbuf);
392 
393 	/*
394 	 * Loop over the allocation groups, starting from the last
395 	 * inode returned; 0 means start of the allocation group.
396 	 */
397 	rval = 0;
398 	while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
399 		bp = NULL;
400 		down_read(&mp->m_peraglock);
401 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
402 		up_read(&mp->m_peraglock);
403 		if (error) {
404 			/*
405 			 * Skip this allocation group and go to the next one.
406 			 */
407 			agno++;
408 			agino = 0;
409 			continue;
410 		}
411 		agi = XFS_BUF_TO_AGI(agbp);
412 		/*
413 		 * Allocate and initialize a btree cursor for ialloc btree.
414 		 */
415 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
416 						(xfs_inode_t *)0, 0);
417 		irbp = irbuf;
418 		irbufend = irbuf + nirbuf;
419 		end_of_ag = 0;
420 		/*
421 		 * If we're returning in the middle of an allocation group,
422 		 * we need to get the remainder of the chunk we're in.
423 		 */
424 		if (agino > 0) {
425 			/*
426 			 * Lookup the inode chunk that this inode lives in.
427 			 */
428 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
429 			if (!error &&	/* no I/O error */
430 			    tmp &&	/* lookup succeeded */
431 					/* got the record, should always work */
432 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
433 				    &gfree, &i)) &&
434 			    i == 1 &&
435 					/* this is the right chunk */
436 			    agino < gino + XFS_INODES_PER_CHUNK &&
437 					/* lastino was not last in chunk */
438 			    (chunkidx = agino - gino + 1) <
439 				    XFS_INODES_PER_CHUNK &&
440 					/* there are some left allocated */
441 			    XFS_INOBT_MASKN(chunkidx,
442 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
443 				/*
444 				 * Grab the chunk record.  Mark all the
445 				 * uninteresting inodes (because they're
446 				 * before our start point) free.
447 				 */
448 				for (i = 0; i < chunkidx; i++) {
449 					if (XFS_INOBT_MASK(i) & ~gfree)
450 						gcnt++;
451 				}
452 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
453 				irbp->ir_startino = gino;
454 				irbp->ir_freecount = gcnt;
455 				irbp->ir_free = gfree;
456 				irbp++;
457 				agino = gino + XFS_INODES_PER_CHUNK;
458 				icount = XFS_INODES_PER_CHUNK - gcnt;
459 			} else {
460 				/*
461 				 * If any of those tests failed, bump the
462 				 * inode number (just in case).
463 				 */
464 				agino++;
465 				icount = 0;
466 			}
467 			/*
468 			 * In any case, increment to the next record.
469 			 */
470 			if (!error)
471 				error = xfs_inobt_increment(cur, 0, &tmp);
472 		} else {
473 			/*
474 			 * Start of ag.  Lookup the first inode chunk.
475 			 */
476 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
477 			icount = 0;
478 		}
479 		/*
480 		 * Loop through inode btree records in this ag,
481 		 * until we run out of inodes or space in the buffer.
482 		 */
483 		while (irbp < irbufend && icount < ubcount) {
484 			/*
485 			 * Loop as long as we're unable to read the
486 			 * inode btree.
487 			 */
488 			while (error) {
489 				agino += XFS_INODES_PER_CHUNK;
490 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
491 						be32_to_cpu(agi->agi_length))
492 					break;
493 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
494 							    &tmp);
495 			}
496 			/*
497 			 * If ran off the end of the ag either with an error,
498 			 * or the normal way, set end and stop collecting.
499 			 */
500 			if (error ||
501 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
502 				    &gfree, &i)) ||
503 			    i == 0) {
504 				end_of_ag = 1;
505 				break;
506 			}
507 			/*
508 			 * If this chunk has any allocated inodes, save it.
509 			 * Also start read-ahead now for this chunk.
510 			 */
511 			if (gcnt < XFS_INODES_PER_CHUNK) {
512 				/*
513 				 * Loop over all clusters in the next chunk.
514 				 * Do a readahead if there are any allocated
515 				 * inodes in that cluster.
516 				 */
517 				for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
518 				     chunkidx = 0;
519 				     chunkidx < XFS_INODES_PER_CHUNK;
520 				     chunkidx += nicluster,
521 				     agbno += nbcluster) {
522 					if (XFS_INOBT_MASKN(chunkidx,
523 							    nicluster) & ~gfree)
524 						xfs_btree_reada_bufs(mp, agno,
525 							agbno, nbcluster);
526 				}
527 				irbp->ir_startino = gino;
528 				irbp->ir_freecount = gcnt;
529 				irbp->ir_free = gfree;
530 				irbp++;
531 				icount += XFS_INODES_PER_CHUNK - gcnt;
532 			}
533 			/*
534 			 * Set agino to after this chunk and bump the cursor.
535 			 */
536 			agino = gino + XFS_INODES_PER_CHUNK;
537 			error = xfs_inobt_increment(cur, 0, &tmp);
538 		}
539 		/*
540 		 * Drop the btree buffers and the agi buffer.
541 		 * We can't hold any of the locks these represent
542 		 * when calling iget.
543 		 */
544 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
545 		xfs_buf_relse(agbp);
546 		/*
547 		 * Now format all the good inodes into the user's buffer.
548 		 */
549 		irbufend = irbp;
550 		for (irbp = irbuf;
551 		     irbp < irbufend && ubleft >= statstruct_size; irbp++) {
552 			/*
553 			 * Now process this chunk of inodes.
554 			 */
555 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
556 			     ubleft > 0 &&
557 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
558 			     chunkidx++, clustidx++, agino++) {
559 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
560 				/*
561 				 * Recompute agbno if this is the
562 				 * first inode of the cluster.
563 				 *
564 				 * Careful with clustidx.   There can be
565 				 * multple clusters per chunk, a single
566 				 * cluster per chunk or a cluster that has
567 				 * inodes represented from several different
568 				 * chunks (if blocksize is large).
569 				 *
570 				 * Because of this, the starting clustidx is
571 				 * initialized to zero in this loop but must
572 				 * later be reset after reading in the cluster
573 				 * buffer.
574 				 */
575 				if ((chunkidx & (nicluster - 1)) == 0) {
576 					agbno = XFS_AGINO_TO_AGBNO(mp,
577 							irbp->ir_startino) +
578 						((chunkidx & nimask) >>
579 						 mp->m_sb.sb_inopblog);
580 
581 					if (flags & (BULKSTAT_FG_QUICK |
582 						     BULKSTAT_FG_INLINE)) {
583 						ino = XFS_AGINO_TO_INO(mp, agno,
584 								       agino);
585 						bno = XFS_AGB_TO_DADDR(mp, agno,
586 								       agbno);
587 
588 						/*
589 						 * Get the inode cluster buffer
590 						 */
591 						ASSERT(xfs_inode_zone != NULL);
592 						ip = kmem_zone_zalloc(xfs_inode_zone,
593 								      KM_SLEEP);
594 						ip->i_ino = ino;
595 						ip->i_mount = mp;
596 						spin_lock_init(&ip->i_flags_lock);
597 						if (bp)
598 							xfs_buf_relse(bp);
599 						error = xfs_itobp(mp, NULL, ip,
600 								&dip, &bp, bno,
601 								XFS_IMAP_BULKSTAT);
602 						if (!error)
603 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
604 						kmem_zone_free(xfs_inode_zone, ip);
605 						if (XFS_TEST_ERROR(error != 0,
606 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
607 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
608 							bp = NULL;
609 							ubleft = 0;
610 							rval = error;
611 							break;
612 						}
613 					}
614 				}
615 				/*
616 				 * Skip if this inode is free.
617 				 */
618 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free)
619 					continue;
620 				/*
621 				 * Count used inodes as free so we can tell
622 				 * when the chunk is used up.
623 				 */
624 				irbp->ir_freecount++;
625 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
626 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
627 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
628 							     clustidx, &dip))
629 					continue;
630 				/*
631 				 * If we need to do an iget, cannot hold bp.
632 				 * Drop it, until starting the next cluster.
633 				 */
634 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
635 					if (bp)
636 						xfs_buf_relse(bp);
637 					bp = NULL;
638 				}
639 
640 				/*
641 				 * Get the inode and fill in a single buffer.
642 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
643 				 * BULKSTAT_FG_IGET uses igets.
644 				 * BULKSTAT_FG_INLINE uses dip if we have an
645 				 * inline attr fork, else igets.
646 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
647 				 * This is also used to count inodes/blks, etc
648 				 * in xfs_qm_quotacheck.
649 				 */
650 				ubused = statstruct_size;
651 				error = formatter(mp, ino, ubufp,
652 						ubleft, private_data,
653 						bno, &ubused, dip, &fmterror);
654 				if (fmterror == BULKSTAT_RV_NOTHING) {
655                                         if (error == EFAULT) {
656                                                 ubleft = 0;
657                                                 rval = error;
658                                                 break;
659                                         }
660 					else if (error == ENOMEM)
661 						ubleft = 0;
662 					else
663 						lastino = ino;
664 					continue;
665 				}
666 				if (fmterror == BULKSTAT_RV_GIVEUP) {
667 					ubleft = 0;
668 					ASSERT(error);
669 					rval = error;
670 					break;
671 				}
672 				if (ubufp)
673 					ubufp += ubused;
674 				ubleft -= ubused;
675 				ubelem++;
676 				lastino = ino;
677 			}
678 		}
679 
680 		if (bp)
681 			xfs_buf_relse(bp);
682 
683 		/*
684 		 * Set up for the next loop iteration.
685 		 */
686 		if (ubleft > 0) {
687 			if (end_of_ag) {
688 				agno++;
689 				agino = 0;
690 			} else
691 				agino = XFS_INO_TO_AGINO(mp, lastino);
692 		} else
693 			break;
694 	}
695 	/*
696 	 * Done, we're either out of filesystem or space to put the data.
697 	 */
698 	kmem_free(irbuf, irbsize);
699 	*ubcountp = ubelem;
700 	if (agno >= mp->m_sb.sb_agcount) {
701 		/*
702 		 * If we ran out of filesystem, mark lastino as off
703 		 * the end of the filesystem, so the next call
704 		 * will return immediately.
705 		 */
706 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
707 		*done = 1;
708 	} else
709 		*lastinop = (xfs_ino_t)lastino;
710 
711 	return rval;
712 }
713 
714 /*
715  * Return stat information in bulk (by-inode) for the filesystem.
716  * Special case for non-sequential one inode bulkstat.
717  */
718 int					/* error status */
719 xfs_bulkstat_single(
720 	xfs_mount_t		*mp,	/* mount point for filesystem */
721 	xfs_ino_t		*lastinop, /* inode to return */
722 	char			__user *buffer, /* buffer with inode stats */
723 	int			*done)	/* 1 if there are more stats to get */
724 {
725 	int			count;	/* count value for bulkstat call */
726 	int			error;	/* return value */
727 	xfs_ino_t		ino;	/* filesystem inode number */
728 	int			res;	/* result from bs1 */
729 
730 	/*
731 	 * note that requesting valid inode numbers which are not allocated
732 	 * to inodes will most likely cause xfs_itobp to generate warning
733 	 * messages about bad magic numbers. This is ok. The fact that
734 	 * the inode isn't actually an inode is handled by the
735 	 * error check below. Done this way to make the usual case faster
736 	 * at the expense of the error case.
737 	 */
738 
739 	ino = (xfs_ino_t)*lastinop;
740 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
741 				 NULL, 0, NULL, NULL, &res);
742 	if (error) {
743 		/*
744 		 * Special case way failed, do it the "long" way
745 		 * to see if that works.
746 		 */
747 		(*lastinop)--;
748 		count = 1;
749 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
750 				NULL, sizeof(xfs_bstat_t), buffer,
751 				BULKSTAT_FG_IGET, done))
752 			return error;
753 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
754 			return error == EFSCORRUPTED ?
755 				XFS_ERROR(EINVAL) : error;
756 		else
757 			return 0;
758 	}
759 	*done = 0;
760 	return 0;
761 }
762 
763 int
764 xfs_inumbers_fmt(
765 	void			__user *ubuffer, /* buffer to write to */
766 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
767 	long			count,		/* # of elements to read */
768 	long			*written)	/* # of bytes written */
769 {
770 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
771 		return -EFAULT;
772 	*written = count * sizeof(*buffer);
773 	return 0;
774 }
775 
776 /*
777  * Return inode number table for the filesystem.
778  */
779 int					/* error status */
780 xfs_inumbers(
781 	xfs_mount_t	*mp,		/* mount point for filesystem */
782 	xfs_ino_t	*lastino,	/* last inode returned */
783 	int		*count,		/* size of buffer/count returned */
784 	void		__user *ubuffer,/* buffer with inode descriptions */
785 	inumbers_fmt_pf	formatter)
786 {
787 	xfs_buf_t	*agbp;
788 	xfs_agino_t	agino;
789 	xfs_agnumber_t	agno;
790 	int		bcount;
791 	xfs_inogrp_t	*buffer;
792 	int		bufidx;
793 	xfs_btree_cur_t	*cur;
794 	int		error;
795 	__int32_t	gcnt;
796 	xfs_inofree_t	gfree;
797 	xfs_agino_t	gino;
798 	int		i;
799 	xfs_ino_t	ino;
800 	int		left;
801 	int		tmp;
802 
803 	ino = (xfs_ino_t)*lastino;
804 	agno = XFS_INO_TO_AGNO(mp, ino);
805 	agino = XFS_INO_TO_AGINO(mp, ino);
806 	left = *count;
807 	*count = 0;
808 	bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
809 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
810 	error = bufidx = 0;
811 	cur = NULL;
812 	agbp = NULL;
813 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
814 		if (agbp == NULL) {
815 			down_read(&mp->m_peraglock);
816 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
817 			up_read(&mp->m_peraglock);
818 			if (error) {
819 				/*
820 				 * If we can't read the AGI of this ag,
821 				 * then just skip to the next one.
822 				 */
823 				ASSERT(cur == NULL);
824 				agbp = NULL;
825 				agno++;
826 				agino = 0;
827 				continue;
828 			}
829 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
830 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
831 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
832 			if (error) {
833 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
834 				cur = NULL;
835 				xfs_buf_relse(agbp);
836 				agbp = NULL;
837 				/*
838 				 * Move up the last inode in the current
839 				 * chunk.  The lookup_ge will always get
840 				 * us the first inode in the next chunk.
841 				 */
842 				agino += XFS_INODES_PER_CHUNK - 1;
843 				continue;
844 			}
845 		}
846 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
847 			&i)) ||
848 		    i == 0) {
849 			xfs_buf_relse(agbp);
850 			agbp = NULL;
851 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
852 			cur = NULL;
853 			agno++;
854 			agino = 0;
855 			continue;
856 		}
857 		agino = gino + XFS_INODES_PER_CHUNK - 1;
858 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
859 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
860 		buffer[bufidx].xi_allocmask = ~gfree;
861 		bufidx++;
862 		left--;
863 		if (bufidx == bcount) {
864 			long written;
865 			if (formatter(ubuffer, buffer, bufidx, &written)) {
866 				error = XFS_ERROR(EFAULT);
867 				break;
868 			}
869 			ubuffer += written;
870 			*count += bufidx;
871 			bufidx = 0;
872 		}
873 		if (left) {
874 			error = xfs_inobt_increment(cur, 0, &tmp);
875 			if (error) {
876 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
877 				cur = NULL;
878 				xfs_buf_relse(agbp);
879 				agbp = NULL;
880 				/*
881 				 * The agino value has already been bumped.
882 				 * Just try to skip up to it.
883 				 */
884 				agino += XFS_INODES_PER_CHUNK;
885 				continue;
886 			}
887 		}
888 	}
889 	if (!error) {
890 		if (bufidx) {
891 			long written;
892 			if (formatter(ubuffer, buffer, bufidx, &written))
893 				error = XFS_ERROR(EFAULT);
894 			else
895 				*count += bufidx;
896 		}
897 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
898 	}
899 	kmem_free(buffer, bcount * sizeof(*buffer));
900 	if (cur)
901 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
902 					   XFS_BTREE_NOERROR));
903 	if (agbp)
904 		xfs_buf_relse(agbp);
905 	return error;
906 }
907