xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 7dd65feb)
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 STATIC 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 	struct inode	*inode;
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_imap.im_blkno != 0);
74 
75 	dic = &ip->i_d;
76 	inode = VFS_I(ip);
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 
89 	/*
90 	 * We need to read the timestamps from the Linux inode because
91 	 * the VFS keeps writing directly into the inode structure instead
92 	 * of telling us about the updates.
93 	 */
94 	buf->bs_atime.tv_sec = inode->i_atime.tv_sec;
95 	buf->bs_atime.tv_nsec = inode->i_atime.tv_nsec;
96 	buf->bs_mtime.tv_sec = inode->i_mtime.tv_sec;
97 	buf->bs_mtime.tv_nsec = inode->i_mtime.tv_nsec;
98 	buf->bs_ctime.tv_sec = inode->i_ctime.tv_sec;
99 	buf->bs_ctime.tv_nsec = inode->i_ctime.tv_nsec;
100 
101 	buf->bs_xflags = xfs_ip2xflags(ip);
102 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
103 	buf->bs_extents = dic->di_nextents;
104 	buf->bs_gen = dic->di_gen;
105 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
106 	buf->bs_dmevmask = dic->di_dmevmask;
107 	buf->bs_dmstate = dic->di_dmstate;
108 	buf->bs_aextents = dic->di_anextents;
109 
110 	switch (dic->di_format) {
111 	case XFS_DINODE_FMT_DEV:
112 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
113 		buf->bs_blksize = BLKDEV_IOSIZE;
114 		buf->bs_blocks = 0;
115 		break;
116 	case XFS_DINODE_FMT_LOCAL:
117 	case XFS_DINODE_FMT_UUID:
118 		buf->bs_rdev = 0;
119 		buf->bs_blksize = mp->m_sb.sb_blocksize;
120 		buf->bs_blocks = 0;
121 		break;
122 	case XFS_DINODE_FMT_EXTENTS:
123 	case XFS_DINODE_FMT_BTREE:
124 		buf->bs_rdev = 0;
125 		buf->bs_blksize = mp->m_sb.sb_blocksize;
126 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
127 		break;
128 	}
129 
130 	xfs_iput(ip, XFS_ILOCK_SHARED);
131 	return error;
132 }
133 
134 STATIC void
135 xfs_bulkstat_one_dinode(
136 	xfs_mount_t	*mp,		/* mount point for filesystem */
137 	xfs_ino_t	ino,		/* inode number to get data for */
138 	xfs_dinode_t	*dic,		/* dinode inode pointer */
139 	xfs_bstat_t	*buf)		/* return buffer */
140 {
141 	/*
142 	 * The inode format changed when we moved the link count and
143 	 * made it 32 bits long.  If this is an old format inode,
144 	 * convert it in memory to look like a new one.  If it gets
145 	 * flushed to disk we will convert back before flushing or
146 	 * logging it.  We zero out the new projid field and the old link
147 	 * count field.  We'll handle clearing the pad field (the remains
148 	 * of the old uuid field) when we actually convert the inode to
149 	 * the new format. We don't change the version number so that we
150 	 * can distinguish this from a real new format inode.
151 	 */
152 	if (dic->di_version == 1) {
153 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
154 		buf->bs_projid = 0;
155 	} else {
156 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
157 		buf->bs_projid = be16_to_cpu(dic->di_projid);
158 	}
159 
160 	buf->bs_ino = ino;
161 	buf->bs_mode = be16_to_cpu(dic->di_mode);
162 	buf->bs_uid = be32_to_cpu(dic->di_uid);
163 	buf->bs_gid = be32_to_cpu(dic->di_gid);
164 	buf->bs_size = be64_to_cpu(dic->di_size);
165 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
166 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
167 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
168 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
169 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
170 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
171 	buf->bs_xflags = xfs_dic2xflags(dic);
172 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
173 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
174 	buf->bs_gen = be32_to_cpu(dic->di_gen);
175 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
176 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
177 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
178 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
179 
180 	switch (dic->di_format) {
181 	case XFS_DINODE_FMT_DEV:
182 		buf->bs_rdev = xfs_dinode_get_rdev(dic);
183 		buf->bs_blksize = BLKDEV_IOSIZE;
184 		buf->bs_blocks = 0;
185 		break;
186 	case XFS_DINODE_FMT_LOCAL:
187 	case XFS_DINODE_FMT_UUID:
188 		buf->bs_rdev = 0;
189 		buf->bs_blksize = mp->m_sb.sb_blocksize;
190 		buf->bs_blocks = 0;
191 		break;
192 	case XFS_DINODE_FMT_EXTENTS:
193 	case XFS_DINODE_FMT_BTREE:
194 		buf->bs_rdev = 0;
195 		buf->bs_blksize = mp->m_sb.sb_blocksize;
196 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
197 		break;
198 	}
199 }
200 
201 /* Return 0 on success or positive error */
202 STATIC int
203 xfs_bulkstat_one_fmt(
204 	void			__user *ubuffer,
205 	int			ubsize,
206 	int			*ubused,
207 	const xfs_bstat_t	*buffer)
208 {
209 	if (ubsize < sizeof(*buffer))
210 		return XFS_ERROR(ENOMEM);
211 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
212 		return XFS_ERROR(EFAULT);
213 	if (ubused)
214 		*ubused = sizeof(*buffer);
215 	return 0;
216 }
217 
218 /*
219  * Return stat information for one inode.
220  * Return 0 if ok, else errno.
221  */
222 int		   	    		/* error status */
223 xfs_bulkstat_one_int(
224 	xfs_mount_t	*mp,		/* mount point for filesystem */
225 	xfs_ino_t	ino,		/* inode number to get data for */
226 	void		__user *buffer,	/* buffer to place output in */
227 	int		ubsize,		/* size of buffer */
228 	bulkstat_one_fmt_pf formatter,	/* formatter, copy to user */
229 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
230 	int		*ubused,	/* bytes used by me */
231 	void		*dibuff,	/* on-disk inode buffer */
232 	int		*stat)		/* BULKSTAT_RV_... */
233 {
234 	xfs_bstat_t	*buf;		/* return buffer */
235 	int		error = 0;	/* error value */
236 	xfs_dinode_t	*dip;		/* dinode inode pointer */
237 
238 	dip = (xfs_dinode_t *)dibuff;
239 	*stat = BULKSTAT_RV_NOTHING;
240 
241 	if (!buffer || xfs_internal_inum(mp, ino))
242 		return XFS_ERROR(EINVAL);
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, ubsize, ubused, buf);
258 	if (error)
259 		goto out_free;
260 
261 	*stat = BULKSTAT_RV_DIDONE;
262 
263  out_free:
264 	kmem_free(buf);
265 	return error;
266 }
267 
268 int
269 xfs_bulkstat_one(
270 	xfs_mount_t	*mp,		/* mount point for filesystem */
271 	xfs_ino_t	ino,		/* inode number to get data for */
272 	void		__user *buffer,	/* buffer to place output in */
273 	int		ubsize,		/* size of buffer */
274 	void		*private_data,	/* my private data */
275 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
276 	int		*ubused,	/* bytes used by me */
277 	void		*dibuff,	/* on-disk inode buffer */
278 	int		*stat)		/* BULKSTAT_RV_... */
279 {
280 	return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
281 				    xfs_bulkstat_one_fmt, bno,
282 				    ubused, dibuff, stat);
283 }
284 
285 /*
286  * Test to see whether we can use the ondisk inode directly, based
287  * on the given bulkstat flags, filling in dipp accordingly.
288  * Returns zero if the inode is dodgey.
289  */
290 STATIC int
291 xfs_bulkstat_use_dinode(
292 	xfs_mount_t	*mp,
293 	int		flags,
294 	xfs_buf_t	*bp,
295 	int		clustidx,
296 	xfs_dinode_t	**dipp)
297 {
298 	xfs_dinode_t	*dip;
299 	unsigned int	aformat;
300 
301 	*dipp = NULL;
302 	if (!bp || (flags & BULKSTAT_FG_IGET))
303 		return 1;
304 	dip = (xfs_dinode_t *)
305 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
306 	/*
307 	 * Check the buffer containing the on-disk inode for di_mode == 0.
308 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
309 	 * inodes that have their in-core state initialized but not flushed
310 	 * to disk yet. This is a temporary hack that would require a proper
311 	 * fix in the future.
312 	 */
313 	if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
314 	    !XFS_DINODE_GOOD_VERSION(dip->di_version) ||
315 	    !dip->di_mode)
316 		return 0;
317 	if (flags & BULKSTAT_FG_QUICK) {
318 		*dipp = dip;
319 		return 1;
320 	}
321 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
322 	aformat = dip->di_aformat;
323 	if ((XFS_DFORK_Q(dip) == 0) ||
324 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
325 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_anextents)) {
326 		*dipp = dip;
327 		return 1;
328 	}
329 	return 1;
330 }
331 
332 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
333 
334 /*
335  * Return stat information in bulk (by-inode) for the filesystem.
336  */
337 int					/* error status */
338 xfs_bulkstat(
339 	xfs_mount_t		*mp,	/* mount point for filesystem */
340 	xfs_ino_t		*lastinop, /* last inode returned */
341 	int			*ubcountp, /* size of buffer/count returned */
342 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
343 	void			*private_data,/* private data for formatter */
344 	size_t			statstruct_size, /* sizeof struct filling */
345 	char			__user *ubuffer, /* buffer with inode stats */
346 	int			flags,	/* defined in xfs_itable.h */
347 	int			*done)	/* 1 if there are more stats to get */
348 {
349 	xfs_agblock_t		agbno=0;/* allocation group block number */
350 	xfs_buf_t		*agbp;	/* agi header buffer */
351 	xfs_agi_t		*agi;	/* agi header data */
352 	xfs_agino_t		agino;	/* inode # in allocation group */
353 	xfs_agnumber_t		agno;	/* allocation group number */
354 	xfs_daddr_t		bno;	/* inode cluster start daddr */
355 	int			chunkidx; /* current index into inode chunk */
356 	int			clustidx; /* current index into inode cluster */
357 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
358 	int			end_of_ag; /* set if we've seen the ag end */
359 	int			error;	/* error code */
360 	int                     fmterror;/* bulkstat formatter result */
361 	int			i;	/* loop index */
362 	int			icount;	/* count of inodes good in irbuf */
363 	size_t			irbsize; /* size of irec buffer in bytes */
364 	xfs_ino_t		ino;	/* inode number (filesystem) */
365 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
366 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
367 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
368 	xfs_ino_t		lastino; /* last inode number returned */
369 	int			nbcluster; /* # of blocks in a cluster */
370 	int			nicluster; /* # of inodes in a cluster */
371 	int			nimask;	/* mask for inode clusters */
372 	int			nirbuf;	/* size of irbuf */
373 	int			rval;	/* return value error code */
374 	int			tmp;	/* result value from btree calls */
375 	int			ubcount; /* size of user's buffer */
376 	int			ubleft;	/* bytes left in user's buffer */
377 	char			__user *ubufp;	/* pointer into user's buffer */
378 	int			ubelem;	/* spaces used in user's buffer */
379 	int			ubused;	/* bytes used by formatter */
380 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
381 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
382 
383 	/*
384 	 * Get the last inode value, see if there's nothing to do.
385 	 */
386 	ino = (xfs_ino_t)*lastinop;
387 	lastino = ino;
388 	dip = NULL;
389 	agno = XFS_INO_TO_AGNO(mp, ino);
390 	agino = XFS_INO_TO_AGINO(mp, ino);
391 	if (agno >= mp->m_sb.sb_agcount ||
392 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
393 		*done = 1;
394 		*ubcountp = 0;
395 		return 0;
396 	}
397 	if (!ubcountp || *ubcountp <= 0) {
398 		return EINVAL;
399 	}
400 	ubcount = *ubcountp; /* statstruct's */
401 	ubleft = ubcount * statstruct_size; /* bytes */
402 	*ubcountp = ubelem = 0;
403 	*done = 0;
404 	fmterror = 0;
405 	ubufp = ubuffer;
406 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
407 		mp->m_sb.sb_inopblock :
408 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
409 	nimask = ~(nicluster - 1);
410 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
411 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
412 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
413 	nirbuf = irbsize / sizeof(*irbuf);
414 
415 	/*
416 	 * Loop over the allocation groups, starting from the last
417 	 * inode returned; 0 means start of the allocation group.
418 	 */
419 	rval = 0;
420 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
421 		cond_resched();
422 		bp = NULL;
423 		down_read(&mp->m_peraglock);
424 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
425 		up_read(&mp->m_peraglock);
426 		if (error) {
427 			/*
428 			 * Skip this allocation group and go to the next one.
429 			 */
430 			agno++;
431 			agino = 0;
432 			continue;
433 		}
434 		agi = XFS_BUF_TO_AGI(agbp);
435 		/*
436 		 * Allocate and initialize a btree cursor for ialloc btree.
437 		 */
438 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
439 		irbp = irbuf;
440 		irbufend = irbuf + nirbuf;
441 		end_of_ag = 0;
442 		/*
443 		 * If we're returning in the middle of an allocation group,
444 		 * we need to get the remainder of the chunk we're in.
445 		 */
446 		if (agino > 0) {
447 			xfs_inobt_rec_incore_t r;
448 
449 			/*
450 			 * Lookup the inode chunk that this inode lives in.
451 			 */
452 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
453 						 &tmp);
454 			if (!error &&	/* no I/O error */
455 			    tmp &&	/* lookup succeeded */
456 					/* got the record, should always work */
457 			    !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
458 			    i == 1 &&
459 					/* this is the right chunk */
460 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
461 					/* lastino was not last in chunk */
462 			    (chunkidx = agino - r.ir_startino + 1) <
463 				    XFS_INODES_PER_CHUNK &&
464 					/* there are some left allocated */
465 			    xfs_inobt_maskn(chunkidx,
466 				    XFS_INODES_PER_CHUNK - chunkidx) &
467 				    ~r.ir_free) {
468 				/*
469 				 * Grab the chunk record.  Mark all the
470 				 * uninteresting inodes (because they're
471 				 * before our start point) free.
472 				 */
473 				for (i = 0; i < chunkidx; i++) {
474 					if (XFS_INOBT_MASK(i) & ~r.ir_free)
475 						r.ir_freecount++;
476 				}
477 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
478 				irbp->ir_startino = r.ir_startino;
479 				irbp->ir_freecount = r.ir_freecount;
480 				irbp->ir_free = r.ir_free;
481 				irbp++;
482 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
483 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
484 			} else {
485 				/*
486 				 * If any of those tests failed, bump the
487 				 * inode number (just in case).
488 				 */
489 				agino++;
490 				icount = 0;
491 			}
492 			/*
493 			 * In any case, increment to the next record.
494 			 */
495 			if (!error)
496 				error = xfs_btree_increment(cur, 0, &tmp);
497 		} else {
498 			/*
499 			 * Start of ag.  Lookup the first inode chunk.
500 			 */
501 			error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
502 			icount = 0;
503 		}
504 		/*
505 		 * Loop through inode btree records in this ag,
506 		 * until we run out of inodes or space in the buffer.
507 		 */
508 		while (irbp < irbufend && icount < ubcount) {
509 			xfs_inobt_rec_incore_t r;
510 
511 			/*
512 			 * Loop as long as we're unable to read the
513 			 * inode btree.
514 			 */
515 			while (error) {
516 				agino += XFS_INODES_PER_CHUNK;
517 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
518 						be32_to_cpu(agi->agi_length))
519 					break;
520 				error = xfs_inobt_lookup(cur, agino,
521 							 XFS_LOOKUP_GE, &tmp);
522 				cond_resched();
523 			}
524 			/*
525 			 * If ran off the end of the ag either with an error,
526 			 * or the normal way, set end and stop collecting.
527 			 */
528 			if (error) {
529 				end_of_ag = 1;
530 				break;
531 			}
532 
533 			error = xfs_inobt_get_rec(cur, &r, &i);
534 			if (error || i == 0) {
535 				end_of_ag = 1;
536 				break;
537 			}
538 
539 			/*
540 			 * If this chunk has any allocated inodes, save it.
541 			 * Also start read-ahead now for this chunk.
542 			 */
543 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
544 				/*
545 				 * Loop over all clusters in the next chunk.
546 				 * Do a readahead if there are any allocated
547 				 * inodes in that cluster.
548 				 */
549 				agbno = XFS_AGINO_TO_AGBNO(mp, r.ir_startino);
550 				for (chunkidx = 0;
551 				     chunkidx < XFS_INODES_PER_CHUNK;
552 				     chunkidx += nicluster,
553 				     agbno += nbcluster) {
554 					if (xfs_inobt_maskn(chunkidx, nicluster)
555 							& ~r.ir_free)
556 						xfs_btree_reada_bufs(mp, agno,
557 							agbno, nbcluster);
558 				}
559 				irbp->ir_startino = r.ir_startino;
560 				irbp->ir_freecount = r.ir_freecount;
561 				irbp->ir_free = r.ir_free;
562 				irbp++;
563 				icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
564 			}
565 			/*
566 			 * Set agino to after this chunk and bump the cursor.
567 			 */
568 			agino = r.ir_startino + XFS_INODES_PER_CHUNK;
569 			error = xfs_btree_increment(cur, 0, &tmp);
570 			cond_resched();
571 		}
572 		/*
573 		 * Drop the btree buffers and the agi buffer.
574 		 * We can't hold any of the locks these represent
575 		 * when calling iget.
576 		 */
577 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
578 		xfs_buf_relse(agbp);
579 		/*
580 		 * Now format all the good inodes into the user's buffer.
581 		 */
582 		irbufend = irbp;
583 		for (irbp = irbuf;
584 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
585 			/*
586 			 * Now process this chunk of inodes.
587 			 */
588 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
589 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
590 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
591 			     chunkidx++, clustidx++, agino++) {
592 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
593 				/*
594 				 * Recompute agbno if this is the
595 				 * first inode of the cluster.
596 				 *
597 				 * Careful with clustidx.   There can be
598 				 * multiple clusters per chunk, a single
599 				 * cluster per chunk or a cluster that has
600 				 * inodes represented from several different
601 				 * chunks (if blocksize is large).
602 				 *
603 				 * Because of this, the starting clustidx is
604 				 * initialized to zero in this loop but must
605 				 * later be reset after reading in the cluster
606 				 * buffer.
607 				 */
608 				if ((chunkidx & (nicluster - 1)) == 0) {
609 					agbno = XFS_AGINO_TO_AGBNO(mp,
610 							irbp->ir_startino) +
611 						((chunkidx & nimask) >>
612 						 mp->m_sb.sb_inopblog);
613 
614 					if (flags & (BULKSTAT_FG_QUICK |
615 						     BULKSTAT_FG_INLINE)) {
616 						int offset;
617 
618 						ino = XFS_AGINO_TO_INO(mp, agno,
619 								       agino);
620 						bno = XFS_AGB_TO_DADDR(mp, agno,
621 								       agbno);
622 
623 						/*
624 						 * Get the inode cluster buffer
625 						 */
626 						if (bp)
627 							xfs_buf_relse(bp);
628 
629 						error = xfs_inotobp(mp, NULL, ino, &dip,
630 								    &bp, &offset,
631 								    XFS_IGET_BULKSTAT);
632 
633 						if (!error)
634 							clustidx = offset / mp->m_sb.sb_inodesize;
635 						if (XFS_TEST_ERROR(error != 0,
636 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
637 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
638 							bp = NULL;
639 							ubleft = 0;
640 							rval = error;
641 							break;
642 						}
643 					}
644 				}
645 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
646 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
647 				/*
648 				 * Skip if this inode is free.
649 				 */
650 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
651 					lastino = ino;
652 					continue;
653 				}
654 				/*
655 				 * Count used inodes as free so we can tell
656 				 * when the chunk is used up.
657 				 */
658 				irbp->ir_freecount++;
659 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
660 							     clustidx, &dip)) {
661 					lastino = ino;
662 					continue;
663 				}
664 				/*
665 				 * If we need to do an iget, cannot hold bp.
666 				 * Drop it, until starting the next cluster.
667 				 */
668 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
669 					if (bp)
670 						xfs_buf_relse(bp);
671 					bp = NULL;
672 				}
673 
674 				/*
675 				 * Get the inode and fill in a single buffer.
676 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
677 				 * BULKSTAT_FG_IGET uses igets.
678 				 * BULKSTAT_FG_INLINE uses dip if we have an
679 				 * inline attr fork, else igets.
680 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
681 				 * This is also used to count inodes/blks, etc
682 				 * in xfs_qm_quotacheck.
683 				 */
684 				ubused = statstruct_size;
685 				error = formatter(mp, ino, ubufp,
686 						ubleft, private_data,
687 						bno, &ubused, dip, &fmterror);
688 				if (fmterror == BULKSTAT_RV_NOTHING) {
689 					if (error && error != ENOENT &&
690 						error != EINVAL) {
691 						ubleft = 0;
692 						rval = error;
693 						break;
694 					}
695 					lastino = ino;
696 					continue;
697 				}
698 				if (fmterror == BULKSTAT_RV_GIVEUP) {
699 					ubleft = 0;
700 					ASSERT(error);
701 					rval = error;
702 					break;
703 				}
704 				if (ubufp)
705 					ubufp += ubused;
706 				ubleft -= ubused;
707 				ubelem++;
708 				lastino = ino;
709 			}
710 
711 			cond_resched();
712 		}
713 
714 		if (bp)
715 			xfs_buf_relse(bp);
716 
717 		/*
718 		 * Set up for the next loop iteration.
719 		 */
720 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
721 			if (end_of_ag) {
722 				agno++;
723 				agino = 0;
724 			} else
725 				agino = XFS_INO_TO_AGINO(mp, lastino);
726 		} else
727 			break;
728 	}
729 	/*
730 	 * Done, we're either out of filesystem or space to put the data.
731 	 */
732 	kmem_free(irbuf);
733 	*ubcountp = ubelem;
734 	/*
735 	 * Found some inodes, return them now and return the error next time.
736 	 */
737 	if (ubelem)
738 		rval = 0;
739 	if (agno >= mp->m_sb.sb_agcount) {
740 		/*
741 		 * If we ran out of filesystem, mark lastino as off
742 		 * the end of the filesystem, so the next call
743 		 * will return immediately.
744 		 */
745 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
746 		*done = 1;
747 	} else
748 		*lastinop = (xfs_ino_t)lastino;
749 
750 	return rval;
751 }
752 
753 /*
754  * Return stat information in bulk (by-inode) for the filesystem.
755  * Special case for non-sequential one inode bulkstat.
756  */
757 int					/* error status */
758 xfs_bulkstat_single(
759 	xfs_mount_t		*mp,	/* mount point for filesystem */
760 	xfs_ino_t		*lastinop, /* inode to return */
761 	char			__user *buffer, /* buffer with inode stats */
762 	int			*done)	/* 1 if there are more stats to get */
763 {
764 	int			count;	/* count value for bulkstat call */
765 	int			error;	/* return value */
766 	xfs_ino_t		ino;	/* filesystem inode number */
767 	int			res;	/* result from bs1 */
768 
769 	/*
770 	 * note that requesting valid inode numbers which are not allocated
771 	 * to inodes will most likely cause xfs_itobp to generate warning
772 	 * messages about bad magic numbers. This is ok. The fact that
773 	 * the inode isn't actually an inode is handled by the
774 	 * error check below. Done this way to make the usual case faster
775 	 * at the expense of the error case.
776 	 */
777 
778 	ino = (xfs_ino_t)*lastinop;
779 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
780 				 NULL, 0, NULL, NULL, &res);
781 	if (error) {
782 		/*
783 		 * Special case way failed, do it the "long" way
784 		 * to see if that works.
785 		 */
786 		(*lastinop)--;
787 		count = 1;
788 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
789 				NULL, sizeof(xfs_bstat_t), buffer,
790 				BULKSTAT_FG_IGET, done))
791 			return error;
792 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
793 			return error == EFSCORRUPTED ?
794 				XFS_ERROR(EINVAL) : error;
795 		else
796 			return 0;
797 	}
798 	*done = 0;
799 	return 0;
800 }
801 
802 int
803 xfs_inumbers_fmt(
804 	void			__user *ubuffer, /* buffer to write to */
805 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
806 	long			count,		/* # of elements to read */
807 	long			*written)	/* # of bytes written */
808 {
809 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
810 		return -EFAULT;
811 	*written = count * sizeof(*buffer);
812 	return 0;
813 }
814 
815 /*
816  * Return inode number table for the filesystem.
817  */
818 int					/* error status */
819 xfs_inumbers(
820 	xfs_mount_t	*mp,		/* mount point for filesystem */
821 	xfs_ino_t	*lastino,	/* last inode returned */
822 	int		*count,		/* size of buffer/count returned */
823 	void		__user *ubuffer,/* buffer with inode descriptions */
824 	inumbers_fmt_pf	formatter)
825 {
826 	xfs_buf_t	*agbp;
827 	xfs_agino_t	agino;
828 	xfs_agnumber_t	agno;
829 	int		bcount;
830 	xfs_inogrp_t	*buffer;
831 	int		bufidx;
832 	xfs_btree_cur_t	*cur;
833 	int		error;
834 	xfs_inobt_rec_incore_t r;
835 	int		i;
836 	xfs_ino_t	ino;
837 	int		left;
838 	int		tmp;
839 
840 	ino = (xfs_ino_t)*lastino;
841 	agno = XFS_INO_TO_AGNO(mp, ino);
842 	agino = XFS_INO_TO_AGINO(mp, ino);
843 	left = *count;
844 	*count = 0;
845 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
846 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
847 	error = bufidx = 0;
848 	cur = NULL;
849 	agbp = NULL;
850 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
851 		if (agbp == NULL) {
852 			down_read(&mp->m_peraglock);
853 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
854 			up_read(&mp->m_peraglock);
855 			if (error) {
856 				/*
857 				 * If we can't read the AGI of this ag,
858 				 * then just skip to the next one.
859 				 */
860 				ASSERT(cur == NULL);
861 				agbp = NULL;
862 				agno++;
863 				agino = 0;
864 				continue;
865 			}
866 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
867 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
868 						 &tmp);
869 			if (error) {
870 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
871 				cur = NULL;
872 				xfs_buf_relse(agbp);
873 				agbp = NULL;
874 				/*
875 				 * Move up the last inode in the current
876 				 * chunk.  The lookup_ge will always get
877 				 * us the first inode in the next chunk.
878 				 */
879 				agino += XFS_INODES_PER_CHUNK - 1;
880 				continue;
881 			}
882 		}
883 		error = xfs_inobt_get_rec(cur, &r, &i);
884 		if (error || i == 0) {
885 			xfs_buf_relse(agbp);
886 			agbp = NULL;
887 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
888 			cur = NULL;
889 			agno++;
890 			agino = 0;
891 			continue;
892 		}
893 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
894 		buffer[bufidx].xi_startino =
895 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
896 		buffer[bufidx].xi_alloccount =
897 			XFS_INODES_PER_CHUNK - r.ir_freecount;
898 		buffer[bufidx].xi_allocmask = ~r.ir_free;
899 		bufidx++;
900 		left--;
901 		if (bufidx == bcount) {
902 			long written;
903 			if (formatter(ubuffer, buffer, bufidx, &written)) {
904 				error = XFS_ERROR(EFAULT);
905 				break;
906 			}
907 			ubuffer += written;
908 			*count += bufidx;
909 			bufidx = 0;
910 		}
911 		if (left) {
912 			error = xfs_btree_increment(cur, 0, &tmp);
913 			if (error) {
914 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
915 				cur = NULL;
916 				xfs_buf_relse(agbp);
917 				agbp = NULL;
918 				/*
919 				 * The agino value has already been bumped.
920 				 * Just try to skip up to it.
921 				 */
922 				agino += XFS_INODES_PER_CHUNK;
923 				continue;
924 			}
925 		}
926 	}
927 	if (!error) {
928 		if (bufidx) {
929 			long written;
930 			if (formatter(ubuffer, buffer, bufidx, &written))
931 				error = XFS_ERROR(EFAULT);
932 			else
933 				*count += bufidx;
934 		}
935 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
936 	}
937 	kmem_free(buffer);
938 	if (cur)
939 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
940 					   XFS_BTREE_NOERROR));
941 	if (agbp)
942 		xfs_buf_relse(agbp);
943 	return error;
944 }
945