xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision fd589a8f)
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 	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 	int			i;	/* loop index */
357 	int			icount;	/* count of inodes good in irbuf */
358 	size_t			irbsize; /* size of irec buffer in bytes */
359 	xfs_ino_t		ino;	/* inode number (filesystem) */
360 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
361 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
362 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
363 	xfs_ino_t		lastino; /* last inode number returned */
364 	int			nbcluster; /* # of blocks in a cluster */
365 	int			nicluster; /* # of inodes in a cluster */
366 	int			nimask;	/* mask for inode clusters */
367 	int			nirbuf;	/* size of irbuf */
368 	int			rval;	/* return value error code */
369 	int			tmp;	/* result value from btree calls */
370 	int			ubcount; /* size of user's buffer */
371 	int			ubleft;	/* bytes left in user's buffer */
372 	char			__user *ubufp;	/* pointer into user's buffer */
373 	int			ubelem;	/* spaces used in user's buffer */
374 	int			ubused;	/* bytes used by formatter */
375 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
376 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
377 
378 	/*
379 	 * Get the last inode value, see if there's nothing to do.
380 	 */
381 	ino = (xfs_ino_t)*lastinop;
382 	lastino = ino;
383 	dip = NULL;
384 	agno = XFS_INO_TO_AGNO(mp, ino);
385 	agino = XFS_INO_TO_AGINO(mp, ino);
386 	if (agno >= mp->m_sb.sb_agcount ||
387 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
388 		*done = 1;
389 		*ubcountp = 0;
390 		return 0;
391 	}
392 	if (!ubcountp || *ubcountp <= 0) {
393 		return EINVAL;
394 	}
395 	ubcount = *ubcountp; /* statstruct's */
396 	ubleft = ubcount * statstruct_size; /* bytes */
397 	*ubcountp = ubelem = 0;
398 	*done = 0;
399 	fmterror = 0;
400 	ubufp = ubuffer;
401 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
402 		mp->m_sb.sb_inopblock :
403 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
404 	nimask = ~(nicluster - 1);
405 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
406 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
407 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
408 	nirbuf = irbsize / sizeof(*irbuf);
409 
410 	/*
411 	 * Loop over the allocation groups, starting from the last
412 	 * inode returned; 0 means start of the allocation group.
413 	 */
414 	rval = 0;
415 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
416 		cond_resched();
417 		bp = NULL;
418 		down_read(&mp->m_peraglock);
419 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
420 		up_read(&mp->m_peraglock);
421 		if (error) {
422 			/*
423 			 * Skip this allocation group and go to the next one.
424 			 */
425 			agno++;
426 			agino = 0;
427 			continue;
428 		}
429 		agi = XFS_BUF_TO_AGI(agbp);
430 		/*
431 		 * Allocate and initialize a btree cursor for ialloc btree.
432 		 */
433 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
434 		irbp = irbuf;
435 		irbufend = irbuf + nirbuf;
436 		end_of_ag = 0;
437 		/*
438 		 * If we're returning in the middle of an allocation group,
439 		 * we need to get the remainder of the chunk we're in.
440 		 */
441 		if (agino > 0) {
442 			xfs_inobt_rec_incore_t r;
443 
444 			/*
445 			 * Lookup the inode chunk that this inode lives in.
446 			 */
447 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
448 						 &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, &r, &i)) &&
453 			    i == 1 &&
454 					/* this is the right chunk */
455 			    agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
456 					/* lastino was not last in chunk */
457 			    (chunkidx = agino - r.ir_startino + 1) <
458 				    XFS_INODES_PER_CHUNK &&
459 					/* there are some left allocated */
460 			    xfs_inobt_maskn(chunkidx,
461 				    XFS_INODES_PER_CHUNK - chunkidx) &
462 				    ~r.ir_free) {
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) & ~r.ir_free)
470 						r.ir_freecount++;
471 				}
472 				r.ir_free |= xfs_inobt_maskn(0, chunkidx);
473 				irbp->ir_startino = r.ir_startino;
474 				irbp->ir_freecount = r.ir_freecount;
475 				irbp->ir_free = r.ir_free;
476 				irbp++;
477 				agino = r.ir_startino + XFS_INODES_PER_CHUNK;
478 				icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
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(cur, 0, XFS_LOOKUP_GE, &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 			xfs_inobt_rec_incore_t r;
505 
506 			/*
507 			 * Loop as long as we're unable to read the
508 			 * inode btree.
509 			 */
510 			while (error) {
511 				agino += XFS_INODES_PER_CHUNK;
512 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
513 						be32_to_cpu(agi->agi_length))
514 					break;
515 				error = xfs_inobt_lookup(cur, agino,
516 							 XFS_LOOKUP_GE, &tmp);
517 				cond_resched();
518 			}
519 			/*
520 			 * If ran off the end of the ag either with an error,
521 			 * or the normal way, set end and stop collecting.
522 			 */
523 			if (error) {
524 				end_of_ag = 1;
525 				break;
526 			}
527 
528 			error = xfs_inobt_get_rec(cur, &r, &i);
529 			if (error || i == 0) {
530 				end_of_ag = 1;
531 				break;
532 			}
533 
534 			/*
535 			 * If this chunk has any allocated inodes, save it.
536 			 * Also start read-ahead now for this chunk.
537 			 */
538 			if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
539 				/*
540 				 * Loop over all clusters in the next chunk.
541 				 * Do a readahead if there are any allocated
542 				 * inodes in that cluster.
543 				 */
544 				agbno = XFS_AGINO_TO_AGBNO(mp, r.ir_startino);
545 				for (chunkidx = 0;
546 				     chunkidx < XFS_INODES_PER_CHUNK;
547 				     chunkidx += nicluster,
548 				     agbno += nbcluster) {
549 					if (xfs_inobt_maskn(chunkidx, nicluster)
550 							& ~r.ir_free)
551 						xfs_btree_reada_bufs(mp, agno,
552 							agbno, nbcluster);
553 				}
554 				irbp->ir_startino = r.ir_startino;
555 				irbp->ir_freecount = r.ir_freecount;
556 				irbp->ir_free = r.ir_free;
557 				irbp++;
558 				icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
559 			}
560 			/*
561 			 * Set agino to after this chunk and bump the cursor.
562 			 */
563 			agino = r.ir_startino + XFS_INODES_PER_CHUNK;
564 			error = xfs_btree_increment(cur, 0, &tmp);
565 			cond_resched();
566 		}
567 		/*
568 		 * Drop the btree buffers and the agi buffer.
569 		 * We can't hold any of the locks these represent
570 		 * when calling iget.
571 		 */
572 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
573 		xfs_buf_relse(agbp);
574 		/*
575 		 * Now format all the good inodes into the user's buffer.
576 		 */
577 		irbufend = irbp;
578 		for (irbp = irbuf;
579 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
580 			/*
581 			 * Now process this chunk of inodes.
582 			 */
583 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
584 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
585 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
586 			     chunkidx++, clustidx++, agino++) {
587 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
588 				/*
589 				 * Recompute agbno if this is the
590 				 * first inode of the cluster.
591 				 *
592 				 * Careful with clustidx.   There can be
593 				 * multiple clusters per chunk, a single
594 				 * cluster per chunk or a cluster that has
595 				 * inodes represented from several different
596 				 * chunks (if blocksize is large).
597 				 *
598 				 * Because of this, the starting clustidx is
599 				 * initialized to zero in this loop but must
600 				 * later be reset after reading in the cluster
601 				 * buffer.
602 				 */
603 				if ((chunkidx & (nicluster - 1)) == 0) {
604 					agbno = XFS_AGINO_TO_AGBNO(mp,
605 							irbp->ir_startino) +
606 						((chunkidx & nimask) >>
607 						 mp->m_sb.sb_inopblog);
608 
609 					if (flags & (BULKSTAT_FG_QUICK |
610 						     BULKSTAT_FG_INLINE)) {
611 						int offset;
612 
613 						ino = XFS_AGINO_TO_INO(mp, agno,
614 								       agino);
615 						bno = XFS_AGB_TO_DADDR(mp, agno,
616 								       agbno);
617 
618 						/*
619 						 * Get the inode cluster buffer
620 						 */
621 						if (bp)
622 							xfs_buf_relse(bp);
623 
624 						error = xfs_inotobp(mp, NULL, ino, &dip,
625 								    &bp, &offset,
626 								    XFS_IGET_BULKSTAT);
627 
628 						if (!error)
629 							clustidx = offset / mp->m_sb.sb_inodesize;
630 						if (XFS_TEST_ERROR(error != 0,
631 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
632 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
633 							bp = NULL;
634 							ubleft = 0;
635 							rval = error;
636 							break;
637 						}
638 					}
639 				}
640 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
641 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
642 				/*
643 				 * Skip if this inode is free.
644 				 */
645 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
646 					lastino = ino;
647 					continue;
648 				}
649 				/*
650 				 * Count used inodes as free so we can tell
651 				 * when the chunk is used up.
652 				 */
653 				irbp->ir_freecount++;
654 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
655 							     clustidx, &dip)) {
656 					lastino = ino;
657 					continue;
658 				}
659 				/*
660 				 * If we need to do an iget, cannot hold bp.
661 				 * Drop it, until starting the next cluster.
662 				 */
663 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
664 					if (bp)
665 						xfs_buf_relse(bp);
666 					bp = NULL;
667 				}
668 
669 				/*
670 				 * Get the inode and fill in a single buffer.
671 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
672 				 * BULKSTAT_FG_IGET uses igets.
673 				 * BULKSTAT_FG_INLINE uses dip if we have an
674 				 * inline attr fork, else igets.
675 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
676 				 * This is also used to count inodes/blks, etc
677 				 * in xfs_qm_quotacheck.
678 				 */
679 				ubused = statstruct_size;
680 				error = formatter(mp, ino, ubufp,
681 						ubleft, private_data,
682 						bno, &ubused, dip, &fmterror);
683 				if (fmterror == BULKSTAT_RV_NOTHING) {
684 					if (error && error != ENOENT &&
685 						error != EINVAL) {
686 						ubleft = 0;
687 						rval = error;
688 						break;
689 					}
690 					lastino = ino;
691 					continue;
692 				}
693 				if (fmterror == BULKSTAT_RV_GIVEUP) {
694 					ubleft = 0;
695 					ASSERT(error);
696 					rval = error;
697 					break;
698 				}
699 				if (ubufp)
700 					ubufp += ubused;
701 				ubleft -= ubused;
702 				ubelem++;
703 				lastino = ino;
704 			}
705 
706 			cond_resched();
707 		}
708 
709 		if (bp)
710 			xfs_buf_relse(bp);
711 
712 		/*
713 		 * Set up for the next loop iteration.
714 		 */
715 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
716 			if (end_of_ag) {
717 				agno++;
718 				agino = 0;
719 			} else
720 				agino = XFS_INO_TO_AGINO(mp, lastino);
721 		} else
722 			break;
723 	}
724 	/*
725 	 * Done, we're either out of filesystem or space to put the data.
726 	 */
727 	kmem_free(irbuf);
728 	*ubcountp = ubelem;
729 	/*
730 	 * Found some inodes, return them now and return the error next time.
731 	 */
732 	if (ubelem)
733 		rval = 0;
734 	if (agno >= mp->m_sb.sb_agcount) {
735 		/*
736 		 * If we ran out of filesystem, mark lastino as off
737 		 * the end of the filesystem, so the next call
738 		 * will return immediately.
739 		 */
740 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
741 		*done = 1;
742 	} else
743 		*lastinop = (xfs_ino_t)lastino;
744 
745 	return rval;
746 }
747 
748 /*
749  * Return stat information in bulk (by-inode) for the filesystem.
750  * Special case for non-sequential one inode bulkstat.
751  */
752 int					/* error status */
753 xfs_bulkstat_single(
754 	xfs_mount_t		*mp,	/* mount point for filesystem */
755 	xfs_ino_t		*lastinop, /* inode to return */
756 	char			__user *buffer, /* buffer with inode stats */
757 	int			*done)	/* 1 if there are more stats to get */
758 {
759 	int			count;	/* count value for bulkstat call */
760 	int			error;	/* return value */
761 	xfs_ino_t		ino;	/* filesystem inode number */
762 	int			res;	/* result from bs1 */
763 
764 	/*
765 	 * note that requesting valid inode numbers which are not allocated
766 	 * to inodes will most likely cause xfs_itobp to generate warning
767 	 * messages about bad magic numbers. This is ok. The fact that
768 	 * the inode isn't actually an inode is handled by the
769 	 * error check below. Done this way to make the usual case faster
770 	 * at the expense of the error case.
771 	 */
772 
773 	ino = (xfs_ino_t)*lastinop;
774 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
775 				 NULL, 0, NULL, NULL, &res);
776 	if (error) {
777 		/*
778 		 * Special case way failed, do it the "long" way
779 		 * to see if that works.
780 		 */
781 		(*lastinop)--;
782 		count = 1;
783 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
784 				NULL, sizeof(xfs_bstat_t), buffer,
785 				BULKSTAT_FG_IGET, done))
786 			return error;
787 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
788 			return error == EFSCORRUPTED ?
789 				XFS_ERROR(EINVAL) : error;
790 		else
791 			return 0;
792 	}
793 	*done = 0;
794 	return 0;
795 }
796 
797 int
798 xfs_inumbers_fmt(
799 	void			__user *ubuffer, /* buffer to write to */
800 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
801 	long			count,		/* # of elements to read */
802 	long			*written)	/* # of bytes written */
803 {
804 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
805 		return -EFAULT;
806 	*written = count * sizeof(*buffer);
807 	return 0;
808 }
809 
810 /*
811  * Return inode number table for the filesystem.
812  */
813 int					/* error status */
814 xfs_inumbers(
815 	xfs_mount_t	*mp,		/* mount point for filesystem */
816 	xfs_ino_t	*lastino,	/* last inode returned */
817 	int		*count,		/* size of buffer/count returned */
818 	void		__user *ubuffer,/* buffer with inode descriptions */
819 	inumbers_fmt_pf	formatter)
820 {
821 	xfs_buf_t	*agbp;
822 	xfs_agino_t	agino;
823 	xfs_agnumber_t	agno;
824 	int		bcount;
825 	xfs_inogrp_t	*buffer;
826 	int		bufidx;
827 	xfs_btree_cur_t	*cur;
828 	int		error;
829 	xfs_inobt_rec_incore_t r;
830 	int		i;
831 	xfs_ino_t	ino;
832 	int		left;
833 	int		tmp;
834 
835 	ino = (xfs_ino_t)*lastino;
836 	agno = XFS_INO_TO_AGNO(mp, ino);
837 	agino = XFS_INO_TO_AGINO(mp, ino);
838 	left = *count;
839 	*count = 0;
840 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
841 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
842 	error = bufidx = 0;
843 	cur = NULL;
844 	agbp = NULL;
845 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
846 		if (agbp == NULL) {
847 			down_read(&mp->m_peraglock);
848 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
849 			up_read(&mp->m_peraglock);
850 			if (error) {
851 				/*
852 				 * If we can't read the AGI of this ag,
853 				 * then just skip to the next one.
854 				 */
855 				ASSERT(cur == NULL);
856 				agbp = NULL;
857 				agno++;
858 				agino = 0;
859 				continue;
860 			}
861 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
862 			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
863 						 &tmp);
864 			if (error) {
865 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
866 				cur = NULL;
867 				xfs_buf_relse(agbp);
868 				agbp = NULL;
869 				/*
870 				 * Move up the last inode in the current
871 				 * chunk.  The lookup_ge will always get
872 				 * us the first inode in the next chunk.
873 				 */
874 				agino += XFS_INODES_PER_CHUNK - 1;
875 				continue;
876 			}
877 		}
878 		error = xfs_inobt_get_rec(cur, &r, &i);
879 		if (error || i == 0) {
880 			xfs_buf_relse(agbp);
881 			agbp = NULL;
882 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
883 			cur = NULL;
884 			agno++;
885 			agino = 0;
886 			continue;
887 		}
888 		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
889 		buffer[bufidx].xi_startino =
890 			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
891 		buffer[bufidx].xi_alloccount =
892 			XFS_INODES_PER_CHUNK - r.ir_freecount;
893 		buffer[bufidx].xi_allocmask = ~r.ir_free;
894 		bufidx++;
895 		left--;
896 		if (bufidx == bcount) {
897 			long written;
898 			if (formatter(ubuffer, buffer, bufidx, &written)) {
899 				error = XFS_ERROR(EFAULT);
900 				break;
901 			}
902 			ubuffer += written;
903 			*count += bufidx;
904 			bufidx = 0;
905 		}
906 		if (left) {
907 			error = xfs_btree_increment(cur, 0, &tmp);
908 			if (error) {
909 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
910 				cur = NULL;
911 				xfs_buf_relse(agbp);
912 				agbp = NULL;
913 				/*
914 				 * The agino value has already been bumped.
915 				 * Just try to skip up to it.
916 				 */
917 				agino += XFS_INODES_PER_CHUNK;
918 				continue;
919 			}
920 		}
921 	}
922 	if (!error) {
923 		if (bufidx) {
924 			long written;
925 			if (formatter(ubuffer, buffer, bufidx, &written))
926 				error = XFS_ERROR(EFAULT);
927 			else
928 				*count += bufidx;
929 		}
930 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
931 	}
932 	kmem_free(buffer);
933 	if (cur)
934 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
935 					   XFS_BTREE_NOERROR));
936 	if (agbp)
937 		xfs_buf_relse(agbp);
938 	return error;
939 }
940