xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 545e4006)
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 
75 	vp = XFS_ITOV(ip);
76 	dic = &ip->i_d;
77 
78 	/* xfs_iget returns the following without needing
79 	 * further change.
80 	 */
81 	buf->bs_nlink = dic->di_nlink;
82 	buf->bs_projid = dic->di_projid;
83 	buf->bs_ino = ino;
84 	buf->bs_mode = dic->di_mode;
85 	buf->bs_uid = dic->di_uid;
86 	buf->bs_gid = dic->di_gid;
87 	buf->bs_size = dic->di_size;
88 	vn_atime_to_bstime(vp, &buf->bs_atime);
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 	xfs_iput(ip, XFS_ILOCK_SHARED);
123 	return error;
124 }
125 
126 STATIC void
127 xfs_bulkstat_one_dinode(
128 	xfs_mount_t	*mp,		/* mount point for filesystem */
129 	xfs_ino_t	ino,		/* inode number to get data for */
130 	xfs_dinode_t	*dip,		/* dinode inode pointer */
131 	xfs_bstat_t	*buf)		/* return buffer */
132 {
133 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
134 
135 	dic = &dip->di_core;
136 
137 	/*
138 	 * The inode format changed when we moved the link count and
139 	 * made it 32 bits long.  If this is an old format inode,
140 	 * convert it in memory to look like a new one.  If it gets
141 	 * flushed to disk we will convert back before flushing or
142 	 * logging it.  We zero out the new projid field and the old link
143 	 * count field.  We'll handle clearing the pad field (the remains
144 	 * of the old uuid field) when we actually convert the inode to
145 	 * the new format. We don't change the version number so that we
146 	 * can distinguish this from a real new format inode.
147 	 */
148 	if (dic->di_version == XFS_DINODE_VERSION_1) {
149 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
150 		buf->bs_projid = 0;
151 	} else {
152 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
153 		buf->bs_projid = be16_to_cpu(dic->di_projid);
154 	}
155 
156 	buf->bs_ino = ino;
157 	buf->bs_mode = be16_to_cpu(dic->di_mode);
158 	buf->bs_uid = be32_to_cpu(dic->di_uid);
159 	buf->bs_gid = be32_to_cpu(dic->di_gid);
160 	buf->bs_size = be64_to_cpu(dic->di_size);
161 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
162 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
163 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
164 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
165 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
166 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
167 	buf->bs_xflags = xfs_dic2xflags(dip);
168 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
169 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
170 	buf->bs_gen = be32_to_cpu(dic->di_gen);
171 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
172 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
173 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
174 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
175 
176 	switch (dic->di_format) {
177 	case XFS_DINODE_FMT_DEV:
178 		buf->bs_rdev = be32_to_cpu(dip->di_u.di_dev);
179 		buf->bs_blksize = BLKDEV_IOSIZE;
180 		buf->bs_blocks = 0;
181 		break;
182 	case XFS_DINODE_FMT_LOCAL:
183 	case XFS_DINODE_FMT_UUID:
184 		buf->bs_rdev = 0;
185 		buf->bs_blksize = mp->m_sb.sb_blocksize;
186 		buf->bs_blocks = 0;
187 		break;
188 	case XFS_DINODE_FMT_EXTENTS:
189 	case XFS_DINODE_FMT_BTREE:
190 		buf->bs_rdev = 0;
191 		buf->bs_blksize = mp->m_sb.sb_blocksize;
192 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
193 		break;
194 	}
195 }
196 
197 STATIC int
198 xfs_bulkstat_one_fmt(
199 	void			__user *ubuffer,
200 	const xfs_bstat_t	*buffer)
201 {
202 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
203 		return -EFAULT;
204 	return sizeof(*buffer);
205 }
206 
207 /*
208  * Return stat information for one inode.
209  * Return 0 if ok, else errno.
210  */
211 int		       		/* error status */
212 xfs_bulkstat_one(
213 	xfs_mount_t	*mp,		/* mount point for filesystem */
214 	xfs_ino_t	ino,		/* inode number to get data for */
215 	void		__user *buffer,	/* buffer to place output in */
216 	int		ubsize,		/* size of buffer */
217 	void		*private_data,	/* my private data */
218 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
219 	int		*ubused,	/* bytes used by me */
220 	void		*dibuff,	/* on-disk inode buffer */
221 	int		*stat)		/* BULKSTAT_RV_... */
222 {
223 	xfs_bstat_t	*buf;		/* return buffer */
224 	int		error = 0;	/* error value */
225 	xfs_dinode_t	*dip;		/* dinode inode pointer */
226 	bulkstat_one_fmt_pf formatter = private_data ? : xfs_bulkstat_one_fmt;
227 
228 	dip = (xfs_dinode_t *)dibuff;
229 	*stat = BULKSTAT_RV_NOTHING;
230 
231 	if (!buffer || xfs_internal_inum(mp, ino))
232 		return XFS_ERROR(EINVAL);
233 	if (ubsize < sizeof(*buf))
234 		return XFS_ERROR(ENOMEM);
235 
236 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
237 
238 	if (dip == NULL) {
239 		/* We're not being passed a pointer to a dinode.  This happens
240 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
241 		 */
242 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
243 		if (error)
244 			goto out_free;
245 	} else {
246 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
247 	}
248 
249 	error = formatter(buffer, buf);
250 	if (error < 0)  {
251 		error = EFAULT;
252 		goto out_free;
253 	}
254 
255 	*stat = BULKSTAT_RV_DIDONE;
256 	if (ubused)
257 		*ubused = error;
258 
259  out_free:
260 	kmem_free(buf, sizeof(*buf));
261 	return error;
262 }
263 
264 /*
265  * Test to see whether we can use the ondisk inode directly, based
266  * on the given bulkstat flags, filling in dipp accordingly.
267  * Returns zero if the inode is dodgey.
268  */
269 STATIC int
270 xfs_bulkstat_use_dinode(
271 	xfs_mount_t	*mp,
272 	int		flags,
273 	xfs_buf_t	*bp,
274 	int		clustidx,
275 	xfs_dinode_t	**dipp)
276 {
277 	xfs_dinode_t	*dip;
278 	unsigned int	aformat;
279 
280 	*dipp = NULL;
281 	if (!bp || (flags & BULKSTAT_FG_IGET))
282 		return 1;
283 	dip = (xfs_dinode_t *)
284 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
285 	/*
286 	 * Check the buffer containing the on-disk inode for di_mode == 0.
287 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
288 	 * inodes that have their in-core state initialized but not flushed
289 	 * to disk yet. This is a temporary hack that would require a proper
290 	 * fix in the future.
291 	 */
292 	if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC ||
293 	    !XFS_DINODE_GOOD_VERSION(dip->di_core.di_version) ||
294 	    !dip->di_core.di_mode)
295 		return 0;
296 	if (flags & BULKSTAT_FG_QUICK) {
297 		*dipp = dip;
298 		return 1;
299 	}
300 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
301 	aformat = dip->di_core.di_aformat;
302 	if ((XFS_DFORK_Q(dip) == 0) ||
303 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
304 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
305 		*dipp = dip;
306 		return 1;
307 	}
308 	return 1;
309 }
310 
311 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
312 
313 /*
314  * Return stat information in bulk (by-inode) for the filesystem.
315  */
316 int					/* error status */
317 xfs_bulkstat(
318 	xfs_mount_t		*mp,	/* mount point for filesystem */
319 	xfs_ino_t		*lastinop, /* last inode returned */
320 	int			*ubcountp, /* size of buffer/count returned */
321 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
322 	void			*private_data,/* private data for formatter */
323 	size_t			statstruct_size, /* sizeof struct filling */
324 	char			__user *ubuffer, /* buffer with inode stats */
325 	int			flags,	/* defined in xfs_itable.h */
326 	int			*done)	/* 1 if there are more stats to get */
327 {
328 	xfs_agblock_t		agbno=0;/* allocation group block number */
329 	xfs_buf_t		*agbp;	/* agi header buffer */
330 	xfs_agi_t		*agi;	/* agi header data */
331 	xfs_agino_t		agino;	/* inode # in allocation group */
332 	xfs_agnumber_t		agno;	/* allocation group number */
333 	xfs_daddr_t		bno;	/* inode cluster start daddr */
334 	int			chunkidx; /* current index into inode chunk */
335 	int			clustidx; /* current index into inode cluster */
336 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
337 	int			end_of_ag; /* set if we've seen the ag end */
338 	int			error;	/* error code */
339 	int                     fmterror;/* bulkstat formatter result */
340 	__int32_t		gcnt;	/* current btree rec's count */
341 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
342 	xfs_agino_t		gino;	/* current btree rec's start inode */
343 	int			i;	/* loop index */
344 	int			icount;	/* count of inodes good in irbuf */
345 	size_t			irbsize; /* size of irec buffer in bytes */
346 	xfs_ino_t		ino;	/* inode number (filesystem) */
347 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
348 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
349 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
350 	xfs_ino_t		lastino; /* last inode number returned */
351 	int			nbcluster; /* # of blocks in a cluster */
352 	int			nicluster; /* # of inodes in a cluster */
353 	int			nimask;	/* mask for inode clusters */
354 	int			nirbuf;	/* size of irbuf */
355 	int			rval;	/* return value error code */
356 	int			tmp;	/* result value from btree calls */
357 	int			ubcount; /* size of user's buffer */
358 	int			ubleft;	/* bytes left in user's buffer */
359 	char			__user *ubufp;	/* pointer into user's buffer */
360 	int			ubelem;	/* spaces used in user's buffer */
361 	int			ubused;	/* bytes used by formatter */
362 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
363 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
364 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
365 
366 	/*
367 	 * Get the last inode value, see if there's nothing to do.
368 	 */
369 	ino = (xfs_ino_t)*lastinop;
370 	lastino = ino;
371 	dip = NULL;
372 	agno = XFS_INO_TO_AGNO(mp, ino);
373 	agino = XFS_INO_TO_AGINO(mp, ino);
374 	if (agno >= mp->m_sb.sb_agcount ||
375 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
376 		*done = 1;
377 		*ubcountp = 0;
378 		return 0;
379 	}
380 	if (!ubcountp || *ubcountp <= 0) {
381 		return EINVAL;
382 	}
383 	ubcount = *ubcountp; /* statstruct's */
384 	ubleft = ubcount * statstruct_size; /* bytes */
385 	*ubcountp = ubelem = 0;
386 	*done = 0;
387 	fmterror = 0;
388 	ubufp = ubuffer;
389 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
390 		mp->m_sb.sb_inopblock :
391 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
392 	nimask = ~(nicluster - 1);
393 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
394 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
395 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
396 	nirbuf = irbsize / sizeof(*irbuf);
397 
398 	/*
399 	 * Loop over the allocation groups, starting from the last
400 	 * inode returned; 0 means start of the allocation group.
401 	 */
402 	rval = 0;
403 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
404 		cond_resched();
405 		bp = NULL;
406 		down_read(&mp->m_peraglock);
407 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
408 		up_read(&mp->m_peraglock);
409 		if (error) {
410 			/*
411 			 * Skip this allocation group and go to the next one.
412 			 */
413 			agno++;
414 			agino = 0;
415 			continue;
416 		}
417 		agi = XFS_BUF_TO_AGI(agbp);
418 		/*
419 		 * Allocate and initialize a btree cursor for ialloc btree.
420 		 */
421 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
422 						(xfs_inode_t *)0, 0);
423 		irbp = irbuf;
424 		irbufend = irbuf + nirbuf;
425 		end_of_ag = 0;
426 		/*
427 		 * If we're returning in the middle of an allocation group,
428 		 * we need to get the remainder of the chunk we're in.
429 		 */
430 		if (agino > 0) {
431 			/*
432 			 * Lookup the inode chunk that this inode lives in.
433 			 */
434 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
435 			if (!error &&	/* no I/O error */
436 			    tmp &&	/* lookup succeeded */
437 					/* got the record, should always work */
438 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
439 				    &gfree, &i)) &&
440 			    i == 1 &&
441 					/* this is the right chunk */
442 			    agino < gino + XFS_INODES_PER_CHUNK &&
443 					/* lastino was not last in chunk */
444 			    (chunkidx = agino - gino + 1) <
445 				    XFS_INODES_PER_CHUNK &&
446 					/* there are some left allocated */
447 			    XFS_INOBT_MASKN(chunkidx,
448 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
449 				/*
450 				 * Grab the chunk record.  Mark all the
451 				 * uninteresting inodes (because they're
452 				 * before our start point) free.
453 				 */
454 				for (i = 0; i < chunkidx; i++) {
455 					if (XFS_INOBT_MASK(i) & ~gfree)
456 						gcnt++;
457 				}
458 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
459 				irbp->ir_startino = gino;
460 				irbp->ir_freecount = gcnt;
461 				irbp->ir_free = gfree;
462 				irbp++;
463 				agino = gino + XFS_INODES_PER_CHUNK;
464 				icount = XFS_INODES_PER_CHUNK - gcnt;
465 			} else {
466 				/*
467 				 * If any of those tests failed, bump the
468 				 * inode number (just in case).
469 				 */
470 				agino++;
471 				icount = 0;
472 			}
473 			/*
474 			 * In any case, increment to the next record.
475 			 */
476 			if (!error)
477 				error = xfs_inobt_increment(cur, 0, &tmp);
478 		} else {
479 			/*
480 			 * Start of ag.  Lookup the first inode chunk.
481 			 */
482 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
483 			icount = 0;
484 		}
485 		/*
486 		 * Loop through inode btree records in this ag,
487 		 * until we run out of inodes or space in the buffer.
488 		 */
489 		while (irbp < irbufend && icount < ubcount) {
490 			/*
491 			 * Loop as long as we're unable to read the
492 			 * inode btree.
493 			 */
494 			while (error) {
495 				agino += XFS_INODES_PER_CHUNK;
496 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
497 						be32_to_cpu(agi->agi_length))
498 					break;
499 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
500 							    &tmp);
501 				cond_resched();
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 			cond_resched();
546 		}
547 		/*
548 		 * Drop the btree buffers and the agi buffer.
549 		 * We can't hold any of the locks these represent
550 		 * when calling iget.
551 		 */
552 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
553 		xfs_buf_relse(agbp);
554 		/*
555 		 * Now format all the good inodes into the user's buffer.
556 		 */
557 		irbufend = irbp;
558 		for (irbp = irbuf;
559 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
560 			/*
561 			 * Now process this chunk of inodes.
562 			 */
563 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
564 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
565 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
566 			     chunkidx++, clustidx++, agino++) {
567 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
568 				/*
569 				 * Recompute agbno if this is the
570 				 * first inode of the cluster.
571 				 *
572 				 * Careful with clustidx.   There can be
573 				 * multple clusters per chunk, a single
574 				 * cluster per chunk or a cluster that has
575 				 * inodes represented from several different
576 				 * chunks (if blocksize is large).
577 				 *
578 				 * Because of this, the starting clustidx is
579 				 * initialized to zero in this loop but must
580 				 * later be reset after reading in the cluster
581 				 * buffer.
582 				 */
583 				if ((chunkidx & (nicluster - 1)) == 0) {
584 					agbno = XFS_AGINO_TO_AGBNO(mp,
585 							irbp->ir_startino) +
586 						((chunkidx & nimask) >>
587 						 mp->m_sb.sb_inopblog);
588 
589 					if (flags & (BULKSTAT_FG_QUICK |
590 						     BULKSTAT_FG_INLINE)) {
591 						ino = XFS_AGINO_TO_INO(mp, agno,
592 								       agino);
593 						bno = XFS_AGB_TO_DADDR(mp, agno,
594 								       agbno);
595 
596 						/*
597 						 * Get the inode cluster buffer
598 						 */
599 						ASSERT(xfs_inode_zone != NULL);
600 						ip = kmem_zone_zalloc(xfs_inode_zone,
601 								      KM_SLEEP);
602 						ip->i_ino = ino;
603 						ip->i_mount = mp;
604 						spin_lock_init(&ip->i_flags_lock);
605 						if (bp)
606 							xfs_buf_relse(bp);
607 						error = xfs_itobp(mp, NULL, ip,
608 								&dip, &bp, bno,
609 								XFS_IMAP_BULKSTAT,
610 								XFS_BUF_LOCK);
611 						if (!error)
612 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
613 						kmem_zone_free(xfs_inode_zone, ip);
614 						if (XFS_TEST_ERROR(error != 0,
615 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
616 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
617 							bp = NULL;
618 							ubleft = 0;
619 							rval = error;
620 							break;
621 						}
622 					}
623 				}
624 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
625 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
626 				/*
627 				 * Skip if this inode is free.
628 				 */
629 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
630 					lastino = ino;
631 					continue;
632 				}
633 				/*
634 				 * Count used inodes as free so we can tell
635 				 * when the chunk is used up.
636 				 */
637 				irbp->ir_freecount++;
638 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
639 							     clustidx, &dip)) {
640 					lastino = ino;
641 					continue;
642 				}
643 				/*
644 				 * If we need to do an iget, cannot hold bp.
645 				 * Drop it, until starting the next cluster.
646 				 */
647 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
648 					if (bp)
649 						xfs_buf_relse(bp);
650 					bp = NULL;
651 				}
652 
653 				/*
654 				 * Get the inode and fill in a single buffer.
655 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
656 				 * BULKSTAT_FG_IGET uses igets.
657 				 * BULKSTAT_FG_INLINE uses dip if we have an
658 				 * inline attr fork, else igets.
659 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
660 				 * This is also used to count inodes/blks, etc
661 				 * in xfs_qm_quotacheck.
662 				 */
663 				ubused = statstruct_size;
664 				error = formatter(mp, ino, ubufp,
665 						ubleft, private_data,
666 						bno, &ubused, dip, &fmterror);
667 				if (fmterror == BULKSTAT_RV_NOTHING) {
668 					if (error && error != ENOENT &&
669 						error != EINVAL) {
670 						ubleft = 0;
671 						rval = error;
672 						break;
673 					}
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 			cond_resched();
691 		}
692 
693 		if (bp)
694 			xfs_buf_relse(bp);
695 
696 		/*
697 		 * Set up for the next loop iteration.
698 		 */
699 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
700 			if (end_of_ag) {
701 				agno++;
702 				agino = 0;
703 			} else
704 				agino = XFS_INO_TO_AGINO(mp, lastino);
705 		} else
706 			break;
707 	}
708 	/*
709 	 * Done, we're either out of filesystem or space to put the data.
710 	 */
711 	kmem_free(irbuf, irbsize);
712 	*ubcountp = ubelem;
713 	/*
714 	 * Found some inodes, return them now and return the error next time.
715 	 */
716 	if (ubelem)
717 		rval = 0;
718 	if (agno >= mp->m_sb.sb_agcount) {
719 		/*
720 		 * If we ran out of filesystem, mark lastino as off
721 		 * the end of the filesystem, so the next call
722 		 * will return immediately.
723 		 */
724 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
725 		*done = 1;
726 	} else
727 		*lastinop = (xfs_ino_t)lastino;
728 
729 	return rval;
730 }
731 
732 /*
733  * Return stat information in bulk (by-inode) for the filesystem.
734  * Special case for non-sequential one inode bulkstat.
735  */
736 int					/* error status */
737 xfs_bulkstat_single(
738 	xfs_mount_t		*mp,	/* mount point for filesystem */
739 	xfs_ino_t		*lastinop, /* inode to return */
740 	char			__user *buffer, /* buffer with inode stats */
741 	int			*done)	/* 1 if there are more stats to get */
742 {
743 	int			count;	/* count value for bulkstat call */
744 	int			error;	/* return value */
745 	xfs_ino_t		ino;	/* filesystem inode number */
746 	int			res;	/* result from bs1 */
747 
748 	/*
749 	 * note that requesting valid inode numbers which are not allocated
750 	 * to inodes will most likely cause xfs_itobp to generate warning
751 	 * messages about bad magic numbers. This is ok. The fact that
752 	 * the inode isn't actually an inode is handled by the
753 	 * error check below. Done this way to make the usual case faster
754 	 * at the expense of the error case.
755 	 */
756 
757 	ino = (xfs_ino_t)*lastinop;
758 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
759 				 NULL, 0, NULL, NULL, &res);
760 	if (error) {
761 		/*
762 		 * Special case way failed, do it the "long" way
763 		 * to see if that works.
764 		 */
765 		(*lastinop)--;
766 		count = 1;
767 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
768 				NULL, sizeof(xfs_bstat_t), buffer,
769 				BULKSTAT_FG_IGET, done))
770 			return error;
771 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
772 			return error == EFSCORRUPTED ?
773 				XFS_ERROR(EINVAL) : error;
774 		else
775 			return 0;
776 	}
777 	*done = 0;
778 	return 0;
779 }
780 
781 int
782 xfs_inumbers_fmt(
783 	void			__user *ubuffer, /* buffer to write to */
784 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
785 	long			count,		/* # of elements to read */
786 	long			*written)	/* # of bytes written */
787 {
788 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
789 		return -EFAULT;
790 	*written = count * sizeof(*buffer);
791 	return 0;
792 }
793 
794 /*
795  * Return inode number table for the filesystem.
796  */
797 int					/* error status */
798 xfs_inumbers(
799 	xfs_mount_t	*mp,		/* mount point for filesystem */
800 	xfs_ino_t	*lastino,	/* last inode returned */
801 	int		*count,		/* size of buffer/count returned */
802 	void		__user *ubuffer,/* buffer with inode descriptions */
803 	inumbers_fmt_pf	formatter)
804 {
805 	xfs_buf_t	*agbp;
806 	xfs_agino_t	agino;
807 	xfs_agnumber_t	agno;
808 	int		bcount;
809 	xfs_inogrp_t	*buffer;
810 	int		bufidx;
811 	xfs_btree_cur_t	*cur;
812 	int		error;
813 	__int32_t	gcnt;
814 	xfs_inofree_t	gfree;
815 	xfs_agino_t	gino;
816 	int		i;
817 	xfs_ino_t	ino;
818 	int		left;
819 	int		tmp;
820 
821 	ino = (xfs_ino_t)*lastino;
822 	agno = XFS_INO_TO_AGNO(mp, ino);
823 	agino = XFS_INO_TO_AGINO(mp, ino);
824 	left = *count;
825 	*count = 0;
826 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
827 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
828 	error = bufidx = 0;
829 	cur = NULL;
830 	agbp = NULL;
831 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
832 		if (agbp == NULL) {
833 			down_read(&mp->m_peraglock);
834 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
835 			up_read(&mp->m_peraglock);
836 			if (error) {
837 				/*
838 				 * If we can't read the AGI of this ag,
839 				 * then just skip to the next one.
840 				 */
841 				ASSERT(cur == NULL);
842 				agbp = NULL;
843 				agno++;
844 				agino = 0;
845 				continue;
846 			}
847 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
848 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
849 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
850 			if (error) {
851 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
852 				cur = NULL;
853 				xfs_buf_relse(agbp);
854 				agbp = NULL;
855 				/*
856 				 * Move up the last inode in the current
857 				 * chunk.  The lookup_ge will always get
858 				 * us the first inode in the next chunk.
859 				 */
860 				agino += XFS_INODES_PER_CHUNK - 1;
861 				continue;
862 			}
863 		}
864 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
865 			&i)) ||
866 		    i == 0) {
867 			xfs_buf_relse(agbp);
868 			agbp = NULL;
869 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
870 			cur = NULL;
871 			agno++;
872 			agino = 0;
873 			continue;
874 		}
875 		agino = gino + XFS_INODES_PER_CHUNK - 1;
876 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
877 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
878 		buffer[bufidx].xi_allocmask = ~gfree;
879 		bufidx++;
880 		left--;
881 		if (bufidx == bcount) {
882 			long written;
883 			if (formatter(ubuffer, buffer, bufidx, &written)) {
884 				error = XFS_ERROR(EFAULT);
885 				break;
886 			}
887 			ubuffer += written;
888 			*count += bufidx;
889 			bufidx = 0;
890 		}
891 		if (left) {
892 			error = xfs_inobt_increment(cur, 0, &tmp);
893 			if (error) {
894 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
895 				cur = NULL;
896 				xfs_buf_relse(agbp);
897 				agbp = NULL;
898 				/*
899 				 * The agino value has already been bumped.
900 				 * Just try to skip up to it.
901 				 */
902 				agino += XFS_INODES_PER_CHUNK;
903 				continue;
904 			}
905 		}
906 	}
907 	if (!error) {
908 		if (bufidx) {
909 			long written;
910 			if (formatter(ubuffer, buffer, bufidx, &written))
911 				error = XFS_ERROR(EFAULT);
912 			else
913 				*count += bufidx;
914 		}
915 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
916 	}
917 	kmem_free(buffer, bcount * sizeof(*buffer));
918 	if (cur)
919 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
920 					   XFS_BTREE_NOERROR));
921 	if (agbp)
922 		xfs_buf_relse(agbp);
923 	return error;
924 }
925