xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision b04b4f78)
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 	int		error;
63 
64 	error = xfs_iget(mp, NULL, ino,
65 			 XFS_IGET_BULKSTAT, XFS_ILOCK_SHARED, &ip, bno);
66 	if (error) {
67 		*stat = BULKSTAT_RV_NOTHING;
68 		return error;
69 	}
70 
71 	ASSERT(ip != NULL);
72 	ASSERT(ip->i_imap.im_blkno != 0);
73 
74 	dic = &ip->i_d;
75 
76 	/* xfs_iget returns the following without needing
77 	 * further change.
78 	 */
79 	buf->bs_nlink = dic->di_nlink;
80 	buf->bs_projid = dic->di_projid;
81 	buf->bs_ino = ino;
82 	buf->bs_mode = dic->di_mode;
83 	buf->bs_uid = dic->di_uid;
84 	buf->bs_gid = dic->di_gid;
85 	buf->bs_size = dic->di_size;
86 	/*
87 	 * We are reading the atime from the Linux inode because the
88 	 * dinode might not be uptodate.
89 	 */
90 	buf->bs_atime.tv_sec = VFS_I(ip)->i_atime.tv_sec;
91 	buf->bs_atime.tv_nsec = VFS_I(ip)->i_atime.tv_nsec;
92 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
93 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
94 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
95 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
96 	buf->bs_xflags = xfs_ip2xflags(ip);
97 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
98 	buf->bs_extents = dic->di_nextents;
99 	buf->bs_gen = dic->di_gen;
100 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
101 	buf->bs_dmevmask = dic->di_dmevmask;
102 	buf->bs_dmstate = dic->di_dmstate;
103 	buf->bs_aextents = dic->di_anextents;
104 
105 	switch (dic->di_format) {
106 	case XFS_DINODE_FMT_DEV:
107 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
108 		buf->bs_blksize = BLKDEV_IOSIZE;
109 		buf->bs_blocks = 0;
110 		break;
111 	case XFS_DINODE_FMT_LOCAL:
112 	case XFS_DINODE_FMT_UUID:
113 		buf->bs_rdev = 0;
114 		buf->bs_blksize = mp->m_sb.sb_blocksize;
115 		buf->bs_blocks = 0;
116 		break;
117 	case XFS_DINODE_FMT_EXTENTS:
118 	case XFS_DINODE_FMT_BTREE:
119 		buf->bs_rdev = 0;
120 		buf->bs_blksize = mp->m_sb.sb_blocksize;
121 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
122 		break;
123 	}
124 
125 	xfs_iput(ip, XFS_ILOCK_SHARED);
126 	return error;
127 }
128 
129 STATIC void
130 xfs_bulkstat_one_dinode(
131 	xfs_mount_t	*mp,		/* mount point for filesystem */
132 	xfs_ino_t	ino,		/* inode number to get data for */
133 	xfs_dinode_t	*dic,		/* dinode inode pointer */
134 	xfs_bstat_t	*buf)		/* return buffer */
135 {
136 	/*
137 	 * The inode format changed when we moved the link count and
138 	 * made it 32 bits long.  If this is an old format inode,
139 	 * convert it in memory to look like a new one.  If it gets
140 	 * flushed to disk we will convert back before flushing or
141 	 * logging it.  We zero out the new projid field and the old link
142 	 * count field.  We'll handle clearing the pad field (the remains
143 	 * of the old uuid field) when we actually convert the inode to
144 	 * the new format. We don't change the version number so that we
145 	 * can distinguish this from a real new format inode.
146 	 */
147 	if (dic->di_version == 1) {
148 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
149 		buf->bs_projid = 0;
150 	} else {
151 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
152 		buf->bs_projid = be16_to_cpu(dic->di_projid);
153 	}
154 
155 	buf->bs_ino = ino;
156 	buf->bs_mode = be16_to_cpu(dic->di_mode);
157 	buf->bs_uid = be32_to_cpu(dic->di_uid);
158 	buf->bs_gid = be32_to_cpu(dic->di_gid);
159 	buf->bs_size = be64_to_cpu(dic->di_size);
160 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
161 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
162 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
163 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
164 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
165 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
166 	buf->bs_xflags = xfs_dic2xflags(dic);
167 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
168 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
169 	buf->bs_gen = be32_to_cpu(dic->di_gen);
170 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
171 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
172 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
173 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
174 
175 	switch (dic->di_format) {
176 	case XFS_DINODE_FMT_DEV:
177 		buf->bs_rdev = xfs_dinode_get_rdev(dic);
178 		buf->bs_blksize = BLKDEV_IOSIZE;
179 		buf->bs_blocks = 0;
180 		break;
181 	case XFS_DINODE_FMT_LOCAL:
182 	case XFS_DINODE_FMT_UUID:
183 		buf->bs_rdev = 0;
184 		buf->bs_blksize = mp->m_sb.sb_blocksize;
185 		buf->bs_blocks = 0;
186 		break;
187 	case XFS_DINODE_FMT_EXTENTS:
188 	case XFS_DINODE_FMT_BTREE:
189 		buf->bs_rdev = 0;
190 		buf->bs_blksize = mp->m_sb.sb_blocksize;
191 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
192 		break;
193 	}
194 }
195 
196 /* Return 0 on success or positive error */
197 STATIC int
198 xfs_bulkstat_one_fmt(
199 	void			__user *ubuffer,
200 	int			ubsize,
201 	int			*ubused,
202 	const xfs_bstat_t	*buffer)
203 {
204 	if (ubsize < sizeof(*buffer))
205 		return XFS_ERROR(ENOMEM);
206 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
207 		return XFS_ERROR(EFAULT);
208 	if (ubused)
209 		*ubused = sizeof(*buffer);
210 	return 0;
211 }
212 
213 /*
214  * Return stat information for one inode.
215  * Return 0 if ok, else errno.
216  */
217 int		   	    		/* error status */
218 xfs_bulkstat_one_int(
219 	xfs_mount_t	*mp,		/* mount point for filesystem */
220 	xfs_ino_t	ino,		/* inode number to get data for */
221 	void		__user *buffer,	/* buffer to place output in */
222 	int		ubsize,		/* size of buffer */
223 	bulkstat_one_fmt_pf formatter,	/* formatter, copy to user */
224 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
225 	int		*ubused,	/* bytes used by me */
226 	void		*dibuff,	/* on-disk inode buffer */
227 	int		*stat)		/* BULKSTAT_RV_... */
228 {
229 	xfs_bstat_t	*buf;		/* return buffer */
230 	int		error = 0;	/* error value */
231 	xfs_dinode_t	*dip;		/* dinode inode pointer */
232 
233 	dip = (xfs_dinode_t *)dibuff;
234 	*stat = BULKSTAT_RV_NOTHING;
235 
236 	if (!buffer || xfs_internal_inum(mp, ino))
237 		return XFS_ERROR(EINVAL);
238 
239 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
240 
241 	if (dip == NULL) {
242 		/* We're not being passed a pointer to a dinode.  This happens
243 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
244 		 */
245 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
246 		if (error)
247 			goto out_free;
248 	} else {
249 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
250 	}
251 
252 	error = formatter(buffer, ubsize, ubused, buf);
253 	if (error)
254 		goto out_free;
255 
256 	*stat = BULKSTAT_RV_DIDONE;
257 
258  out_free:
259 	kmem_free(buf);
260 	return error;
261 }
262 
263 int
264 xfs_bulkstat_one(
265 	xfs_mount_t	*mp,		/* mount point for filesystem */
266 	xfs_ino_t	ino,		/* inode number to get data for */
267 	void		__user *buffer,	/* buffer to place output in */
268 	int		ubsize,		/* size of buffer */
269 	void		*private_data,	/* my private data */
270 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
271 	int		*ubused,	/* bytes used by me */
272 	void		*dibuff,	/* on-disk inode buffer */
273 	int		*stat)		/* BULKSTAT_RV_... */
274 {
275 	return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
276 				    xfs_bulkstat_one_fmt, bno,
277 				    ubused, dibuff, stat);
278 }
279 
280 /*
281  * Test to see whether we can use the ondisk inode directly, based
282  * on the given bulkstat flags, filling in dipp accordingly.
283  * Returns zero if the inode is dodgey.
284  */
285 STATIC int
286 xfs_bulkstat_use_dinode(
287 	xfs_mount_t	*mp,
288 	int		flags,
289 	xfs_buf_t	*bp,
290 	int		clustidx,
291 	xfs_dinode_t	**dipp)
292 {
293 	xfs_dinode_t	*dip;
294 	unsigned int	aformat;
295 
296 	*dipp = NULL;
297 	if (!bp || (flags & BULKSTAT_FG_IGET))
298 		return 1;
299 	dip = (xfs_dinode_t *)
300 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
301 	/*
302 	 * Check the buffer containing the on-disk inode for di_mode == 0.
303 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
304 	 * inodes that have their in-core state initialized but not flushed
305 	 * to disk yet. This is a temporary hack that would require a proper
306 	 * fix in the future.
307 	 */
308 	if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
309 	    !XFS_DINODE_GOOD_VERSION(dip->di_version) ||
310 	    !dip->di_mode)
311 		return 0;
312 	if (flags & BULKSTAT_FG_QUICK) {
313 		*dipp = dip;
314 		return 1;
315 	}
316 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
317 	aformat = dip->di_aformat;
318 	if ((XFS_DFORK_Q(dip) == 0) ||
319 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
320 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_anextents)) {
321 		*dipp = dip;
322 		return 1;
323 	}
324 	return 1;
325 }
326 
327 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
328 
329 /*
330  * Return stat information in bulk (by-inode) for the filesystem.
331  */
332 int					/* error status */
333 xfs_bulkstat(
334 	xfs_mount_t		*mp,	/* mount point for filesystem */
335 	xfs_ino_t		*lastinop, /* last inode returned */
336 	int			*ubcountp, /* size of buffer/count returned */
337 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
338 	void			*private_data,/* private data for formatter */
339 	size_t			statstruct_size, /* sizeof struct filling */
340 	char			__user *ubuffer, /* buffer with inode stats */
341 	int			flags,	/* defined in xfs_itable.h */
342 	int			*done)	/* 1 if there are more stats to get */
343 {
344 	xfs_agblock_t		agbno=0;/* allocation group block number */
345 	xfs_buf_t		*agbp;	/* agi header buffer */
346 	xfs_agi_t		*agi;	/* agi header data */
347 	xfs_agino_t		agino;	/* inode # in allocation group */
348 	xfs_agnumber_t		agno;	/* allocation group number */
349 	xfs_daddr_t		bno;	/* inode cluster start daddr */
350 	int			chunkidx; /* current index into inode chunk */
351 	int			clustidx; /* current index into inode cluster */
352 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
353 	int			end_of_ag; /* set if we've seen the ag end */
354 	int			error;	/* error code */
355 	int                     fmterror;/* bulkstat formatter result */
356 	__int32_t		gcnt;	/* current btree rec's count */
357 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
358 	xfs_agino_t		gino;	/* current btree rec's start inode */
359 	int			i;	/* loop index */
360 	int			icount;	/* count of inodes good in irbuf */
361 	size_t			irbsize; /* size of irec buffer in bytes */
362 	xfs_ino_t		ino;	/* inode number (filesystem) */
363 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
364 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
365 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
366 	xfs_ino_t		lastino; /* last inode number returned */
367 	int			nbcluster; /* # of blocks in a cluster */
368 	int			nicluster; /* # of inodes in a cluster */
369 	int			nimask;	/* mask for inode clusters */
370 	int			nirbuf;	/* size of irbuf */
371 	int			rval;	/* return value error code */
372 	int			tmp;	/* result value from btree calls */
373 	int			ubcount; /* size of user's buffer */
374 	int			ubleft;	/* bytes left in user's buffer */
375 	char			__user *ubufp;	/* pointer into user's buffer */
376 	int			ubelem;	/* spaces used in user's buffer */
377 	int			ubused;	/* bytes used by formatter */
378 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
379 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
380 
381 	/*
382 	 * Get the last inode value, see if there's nothing to do.
383 	 */
384 	ino = (xfs_ino_t)*lastinop;
385 	lastino = ino;
386 	dip = NULL;
387 	agno = XFS_INO_TO_AGNO(mp, ino);
388 	agino = XFS_INO_TO_AGINO(mp, ino);
389 	if (agno >= mp->m_sb.sb_agcount ||
390 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
391 		*done = 1;
392 		*ubcountp = 0;
393 		return 0;
394 	}
395 	if (!ubcountp || *ubcountp <= 0) {
396 		return EINVAL;
397 	}
398 	ubcount = *ubcountp; /* statstruct's */
399 	ubleft = ubcount * statstruct_size; /* bytes */
400 	*ubcountp = ubelem = 0;
401 	*done = 0;
402 	fmterror = 0;
403 	ubufp = ubuffer;
404 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
405 		mp->m_sb.sb_inopblock :
406 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
407 	nimask = ~(nicluster - 1);
408 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
409 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
410 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
411 	nirbuf = irbsize / sizeof(*irbuf);
412 
413 	/*
414 	 * Loop over the allocation groups, starting from the last
415 	 * inode returned; 0 means start of the allocation group.
416 	 */
417 	rval = 0;
418 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
419 		cond_resched();
420 		bp = NULL;
421 		down_read(&mp->m_peraglock);
422 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
423 		up_read(&mp->m_peraglock);
424 		if (error) {
425 			/*
426 			 * Skip this allocation group and go to the next one.
427 			 */
428 			agno++;
429 			agino = 0;
430 			continue;
431 		}
432 		agi = XFS_BUF_TO_AGI(agbp);
433 		/*
434 		 * Allocate and initialize a btree cursor for ialloc btree.
435 		 */
436 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
437 		irbp = irbuf;
438 		irbufend = irbuf + nirbuf;
439 		end_of_ag = 0;
440 		/*
441 		 * If we're returning in the middle of an allocation group,
442 		 * we need to get the remainder of the chunk we're in.
443 		 */
444 		if (agino > 0) {
445 			/*
446 			 * Lookup the inode chunk that this inode lives in.
447 			 */
448 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
449 			if (!error &&	/* no I/O error */
450 			    tmp &&	/* lookup succeeded */
451 					/* got the record, should always work */
452 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
453 				    &gfree, &i)) &&
454 			    i == 1 &&
455 					/* this is the right chunk */
456 			    agino < gino + XFS_INODES_PER_CHUNK &&
457 					/* lastino was not last in chunk */
458 			    (chunkidx = agino - gino + 1) <
459 				    XFS_INODES_PER_CHUNK &&
460 					/* there are some left allocated */
461 			    xfs_inobt_maskn(chunkidx,
462 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
463 				/*
464 				 * Grab the chunk record.  Mark all the
465 				 * uninteresting inodes (because they're
466 				 * before our start point) free.
467 				 */
468 				for (i = 0; i < chunkidx; i++) {
469 					if (XFS_INOBT_MASK(i) & ~gfree)
470 						gcnt++;
471 				}
472 				gfree |= xfs_inobt_maskn(0, chunkidx);
473 				irbp->ir_startino = gino;
474 				irbp->ir_freecount = gcnt;
475 				irbp->ir_free = gfree;
476 				irbp++;
477 				agino = gino + XFS_INODES_PER_CHUNK;
478 				icount = XFS_INODES_PER_CHUNK - gcnt;
479 			} else {
480 				/*
481 				 * If any of those tests failed, bump the
482 				 * inode number (just in case).
483 				 */
484 				agino++;
485 				icount = 0;
486 			}
487 			/*
488 			 * In any case, increment to the next record.
489 			 */
490 			if (!error)
491 				error = xfs_btree_increment(cur, 0, &tmp);
492 		} else {
493 			/*
494 			 * Start of ag.  Lookup the first inode chunk.
495 			 */
496 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
497 			icount = 0;
498 		}
499 		/*
500 		 * Loop through inode btree records in this ag,
501 		 * until we run out of inodes or space in the buffer.
502 		 */
503 		while (irbp < irbufend && icount < ubcount) {
504 			/*
505 			 * Loop as long as we're unable to read the
506 			 * inode btree.
507 			 */
508 			while (error) {
509 				agino += XFS_INODES_PER_CHUNK;
510 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
511 						be32_to_cpu(agi->agi_length))
512 					break;
513 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
514 							    &tmp);
515 				cond_resched();
516 			}
517 			/*
518 			 * If ran off the end of the ag either with an error,
519 			 * or the normal way, set end and stop collecting.
520 			 */
521 			if (error ||
522 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
523 				    &gfree, &i)) ||
524 			    i == 0) {
525 				end_of_ag = 1;
526 				break;
527 			}
528 			/*
529 			 * If this chunk has any allocated inodes, save it.
530 			 * Also start read-ahead now for this chunk.
531 			 */
532 			if (gcnt < XFS_INODES_PER_CHUNK) {
533 				/*
534 				 * Loop over all clusters in the next chunk.
535 				 * Do a readahead if there are any allocated
536 				 * inodes in that cluster.
537 				 */
538 				for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
539 				     chunkidx = 0;
540 				     chunkidx < XFS_INODES_PER_CHUNK;
541 				     chunkidx += nicluster,
542 				     agbno += nbcluster) {
543 					if (xfs_inobt_maskn(chunkidx,
544 							    nicluster) & ~gfree)
545 						xfs_btree_reada_bufs(mp, agno,
546 							agbno, nbcluster);
547 				}
548 				irbp->ir_startino = gino;
549 				irbp->ir_freecount = gcnt;
550 				irbp->ir_free = gfree;
551 				irbp++;
552 				icount += XFS_INODES_PER_CHUNK - gcnt;
553 			}
554 			/*
555 			 * Set agino to after this chunk and bump the cursor.
556 			 */
557 			agino = gino + XFS_INODES_PER_CHUNK;
558 			error = xfs_btree_increment(cur, 0, &tmp);
559 			cond_resched();
560 		}
561 		/*
562 		 * Drop the btree buffers and the agi buffer.
563 		 * We can't hold any of the locks these represent
564 		 * when calling iget.
565 		 */
566 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
567 		xfs_buf_relse(agbp);
568 		/*
569 		 * Now format all the good inodes into the user's buffer.
570 		 */
571 		irbufend = irbp;
572 		for (irbp = irbuf;
573 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
574 			/*
575 			 * Now process this chunk of inodes.
576 			 */
577 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
578 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
579 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
580 			     chunkidx++, clustidx++, agino++) {
581 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
582 				/*
583 				 * Recompute agbno if this is the
584 				 * first inode of the cluster.
585 				 *
586 				 * Careful with clustidx.   There can be
587 				 * multiple clusters per chunk, a single
588 				 * cluster per chunk or a cluster that has
589 				 * inodes represented from several different
590 				 * chunks (if blocksize is large).
591 				 *
592 				 * Because of this, the starting clustidx is
593 				 * initialized to zero in this loop but must
594 				 * later be reset after reading in the cluster
595 				 * buffer.
596 				 */
597 				if ((chunkidx & (nicluster - 1)) == 0) {
598 					agbno = XFS_AGINO_TO_AGBNO(mp,
599 							irbp->ir_startino) +
600 						((chunkidx & nimask) >>
601 						 mp->m_sb.sb_inopblog);
602 
603 					if (flags & (BULKSTAT_FG_QUICK |
604 						     BULKSTAT_FG_INLINE)) {
605 						int offset;
606 
607 						ino = XFS_AGINO_TO_INO(mp, agno,
608 								       agino);
609 						bno = XFS_AGB_TO_DADDR(mp, agno,
610 								       agbno);
611 
612 						/*
613 						 * Get the inode cluster buffer
614 						 */
615 						if (bp)
616 							xfs_buf_relse(bp);
617 
618 						error = xfs_inotobp(mp, NULL, ino, &dip,
619 								    &bp, &offset,
620 								    XFS_IGET_BULKSTAT);
621 
622 						if (!error)
623 							clustidx = offset / mp->m_sb.sb_inodesize;
624 						if (XFS_TEST_ERROR(error != 0,
625 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
626 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
627 							bp = NULL;
628 							ubleft = 0;
629 							rval = error;
630 							break;
631 						}
632 					}
633 				}
634 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
635 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
636 				/*
637 				 * Skip if this inode is free.
638 				 */
639 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
640 					lastino = ino;
641 					continue;
642 				}
643 				/*
644 				 * Count used inodes as free so we can tell
645 				 * when the chunk is used up.
646 				 */
647 				irbp->ir_freecount++;
648 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
649 							     clustidx, &dip)) {
650 					lastino = ino;
651 					continue;
652 				}
653 				/*
654 				 * If we need to do an iget, cannot hold bp.
655 				 * Drop it, until starting the next cluster.
656 				 */
657 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
658 					if (bp)
659 						xfs_buf_relse(bp);
660 					bp = NULL;
661 				}
662 
663 				/*
664 				 * Get the inode and fill in a single buffer.
665 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
666 				 * BULKSTAT_FG_IGET uses igets.
667 				 * BULKSTAT_FG_INLINE uses dip if we have an
668 				 * inline attr fork, else igets.
669 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
670 				 * This is also used to count inodes/blks, etc
671 				 * in xfs_qm_quotacheck.
672 				 */
673 				ubused = statstruct_size;
674 				error = formatter(mp, ino, ubufp,
675 						ubleft, private_data,
676 						bno, &ubused, dip, &fmterror);
677 				if (fmterror == BULKSTAT_RV_NOTHING) {
678 					if (error && error != ENOENT &&
679 						error != EINVAL) {
680 						ubleft = 0;
681 						rval = error;
682 						break;
683 					}
684 					lastino = ino;
685 					continue;
686 				}
687 				if (fmterror == BULKSTAT_RV_GIVEUP) {
688 					ubleft = 0;
689 					ASSERT(error);
690 					rval = error;
691 					break;
692 				}
693 				if (ubufp)
694 					ubufp += ubused;
695 				ubleft -= ubused;
696 				ubelem++;
697 				lastino = ino;
698 			}
699 
700 			cond_resched();
701 		}
702 
703 		if (bp)
704 			xfs_buf_relse(bp);
705 
706 		/*
707 		 * Set up for the next loop iteration.
708 		 */
709 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
710 			if (end_of_ag) {
711 				agno++;
712 				agino = 0;
713 			} else
714 				agino = XFS_INO_TO_AGINO(mp, lastino);
715 		} else
716 			break;
717 	}
718 	/*
719 	 * Done, we're either out of filesystem or space to put the data.
720 	 */
721 	kmem_free(irbuf);
722 	*ubcountp = ubelem;
723 	/*
724 	 * Found some inodes, return them now and return the error next time.
725 	 */
726 	if (ubelem)
727 		rval = 0;
728 	if (agno >= mp->m_sb.sb_agcount) {
729 		/*
730 		 * If we ran out of filesystem, mark lastino as off
731 		 * the end of the filesystem, so the next call
732 		 * will return immediately.
733 		 */
734 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
735 		*done = 1;
736 	} else
737 		*lastinop = (xfs_ino_t)lastino;
738 
739 	return rval;
740 }
741 
742 /*
743  * Return stat information in bulk (by-inode) for the filesystem.
744  * Special case for non-sequential one inode bulkstat.
745  */
746 int					/* error status */
747 xfs_bulkstat_single(
748 	xfs_mount_t		*mp,	/* mount point for filesystem */
749 	xfs_ino_t		*lastinop, /* inode to return */
750 	char			__user *buffer, /* buffer with inode stats */
751 	int			*done)	/* 1 if there are more stats to get */
752 {
753 	int			count;	/* count value for bulkstat call */
754 	int			error;	/* return value */
755 	xfs_ino_t		ino;	/* filesystem inode number */
756 	int			res;	/* result from bs1 */
757 
758 	/*
759 	 * note that requesting valid inode numbers which are not allocated
760 	 * to inodes will most likely cause xfs_itobp to generate warning
761 	 * messages about bad magic numbers. This is ok. The fact that
762 	 * the inode isn't actually an inode is handled by the
763 	 * error check below. Done this way to make the usual case faster
764 	 * at the expense of the error case.
765 	 */
766 
767 	ino = (xfs_ino_t)*lastinop;
768 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
769 				 NULL, 0, NULL, NULL, &res);
770 	if (error) {
771 		/*
772 		 * Special case way failed, do it the "long" way
773 		 * to see if that works.
774 		 */
775 		(*lastinop)--;
776 		count = 1;
777 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
778 				NULL, sizeof(xfs_bstat_t), buffer,
779 				BULKSTAT_FG_IGET, done))
780 			return error;
781 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
782 			return error == EFSCORRUPTED ?
783 				XFS_ERROR(EINVAL) : error;
784 		else
785 			return 0;
786 	}
787 	*done = 0;
788 	return 0;
789 }
790 
791 int
792 xfs_inumbers_fmt(
793 	void			__user *ubuffer, /* buffer to write to */
794 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
795 	long			count,		/* # of elements to read */
796 	long			*written)	/* # of bytes written */
797 {
798 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
799 		return -EFAULT;
800 	*written = count * sizeof(*buffer);
801 	return 0;
802 }
803 
804 /*
805  * Return inode number table for the filesystem.
806  */
807 int					/* error status */
808 xfs_inumbers(
809 	xfs_mount_t	*mp,		/* mount point for filesystem */
810 	xfs_ino_t	*lastino,	/* last inode returned */
811 	int		*count,		/* size of buffer/count returned */
812 	void		__user *ubuffer,/* buffer with inode descriptions */
813 	inumbers_fmt_pf	formatter)
814 {
815 	xfs_buf_t	*agbp;
816 	xfs_agino_t	agino;
817 	xfs_agnumber_t	agno;
818 	int		bcount;
819 	xfs_inogrp_t	*buffer;
820 	int		bufidx;
821 	xfs_btree_cur_t	*cur;
822 	int		error;
823 	__int32_t	gcnt;
824 	xfs_inofree_t	gfree;
825 	xfs_agino_t	gino;
826 	int		i;
827 	xfs_ino_t	ino;
828 	int		left;
829 	int		tmp;
830 
831 	ino = (xfs_ino_t)*lastino;
832 	agno = XFS_INO_TO_AGNO(mp, ino);
833 	agino = XFS_INO_TO_AGINO(mp, ino);
834 	left = *count;
835 	*count = 0;
836 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
837 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
838 	error = bufidx = 0;
839 	cur = NULL;
840 	agbp = NULL;
841 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
842 		if (agbp == NULL) {
843 			down_read(&mp->m_peraglock);
844 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
845 			up_read(&mp->m_peraglock);
846 			if (error) {
847 				/*
848 				 * If we can't read the AGI of this ag,
849 				 * then just skip to the next one.
850 				 */
851 				ASSERT(cur == NULL);
852 				agbp = NULL;
853 				agno++;
854 				agino = 0;
855 				continue;
856 			}
857 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
858 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
859 			if (error) {
860 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
861 				cur = NULL;
862 				xfs_buf_relse(agbp);
863 				agbp = NULL;
864 				/*
865 				 * Move up the last inode in the current
866 				 * chunk.  The lookup_ge will always get
867 				 * us the first inode in the next chunk.
868 				 */
869 				agino += XFS_INODES_PER_CHUNK - 1;
870 				continue;
871 			}
872 		}
873 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
874 			&i)) ||
875 		    i == 0) {
876 			xfs_buf_relse(agbp);
877 			agbp = NULL;
878 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
879 			cur = NULL;
880 			agno++;
881 			agino = 0;
882 			continue;
883 		}
884 		agino = gino + XFS_INODES_PER_CHUNK - 1;
885 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
886 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
887 		buffer[bufidx].xi_allocmask = ~gfree;
888 		bufidx++;
889 		left--;
890 		if (bufidx == bcount) {
891 			long written;
892 			if (formatter(ubuffer, buffer, bufidx, &written)) {
893 				error = XFS_ERROR(EFAULT);
894 				break;
895 			}
896 			ubuffer += written;
897 			*count += bufidx;
898 			bufidx = 0;
899 		}
900 		if (left) {
901 			error = xfs_btree_increment(cur, 0, &tmp);
902 			if (error) {
903 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
904 				cur = NULL;
905 				xfs_buf_relse(agbp);
906 				agbp = NULL;
907 				/*
908 				 * The agino value has already been bumped.
909 				 * Just try to skip up to it.
910 				 */
911 				agino += XFS_INODES_PER_CHUNK;
912 				continue;
913 			}
914 		}
915 	}
916 	if (!error) {
917 		if (bufidx) {
918 			long written;
919 			if (formatter(ubuffer, buffer, bufidx, &written))
920 				error = XFS_ERROR(EFAULT);
921 			else
922 				*count += bufidx;
923 		}
924 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
925 	}
926 	kmem_free(buffer);
927 	if (cur)
928 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
929 					   XFS_BTREE_NOERROR));
930 	if (agbp)
931 		xfs_buf_relse(agbp);
932 	return error;
933 }
934