xref: /linux/fs/xfs/xfs_itable.c (revision 367b8112fe2ea5c39a7bb4d263dcdd9b612fae18)
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_blkno != (xfs_daddr_t)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 	vn_atime_to_bstime(VFS_I(ip), &buf->bs_atime);
87 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
88 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
89 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
90 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
91 	buf->bs_xflags = xfs_ip2xflags(ip);
92 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
93 	buf->bs_extents = dic->di_nextents;
94 	buf->bs_gen = dic->di_gen;
95 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
96 	buf->bs_dmevmask = dic->di_dmevmask;
97 	buf->bs_dmstate = dic->di_dmstate;
98 	buf->bs_aextents = dic->di_anextents;
99 
100 	switch (dic->di_format) {
101 	case XFS_DINODE_FMT_DEV:
102 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
103 		buf->bs_blksize = BLKDEV_IOSIZE;
104 		buf->bs_blocks = 0;
105 		break;
106 	case XFS_DINODE_FMT_LOCAL:
107 	case XFS_DINODE_FMT_UUID:
108 		buf->bs_rdev = 0;
109 		buf->bs_blksize = mp->m_sb.sb_blocksize;
110 		buf->bs_blocks = 0;
111 		break;
112 	case XFS_DINODE_FMT_EXTENTS:
113 	case XFS_DINODE_FMT_BTREE:
114 		buf->bs_rdev = 0;
115 		buf->bs_blksize = mp->m_sb.sb_blocksize;
116 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
117 		break;
118 	}
119 
120 	xfs_iput(ip, XFS_ILOCK_SHARED);
121 	return error;
122 }
123 
124 STATIC void
125 xfs_bulkstat_one_dinode(
126 	xfs_mount_t	*mp,		/* mount point for filesystem */
127 	xfs_ino_t	ino,		/* inode number to get data for */
128 	xfs_dinode_t	*dip,		/* dinode inode pointer */
129 	xfs_bstat_t	*buf)		/* return buffer */
130 {
131 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
132 
133 	dic = &dip->di_core;
134 
135 	/*
136 	 * The inode format changed when we moved the link count and
137 	 * made it 32 bits long.  If this is an old format inode,
138 	 * convert it in memory to look like a new one.  If it gets
139 	 * flushed to disk we will convert back before flushing or
140 	 * logging it.  We zero out the new projid field and the old link
141 	 * count field.  We'll handle clearing the pad field (the remains
142 	 * of the old uuid field) when we actually convert the inode to
143 	 * the new format. We don't change the version number so that we
144 	 * can distinguish this from a real new format inode.
145 	 */
146 	if (dic->di_version == XFS_DINODE_VERSION_1) {
147 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
148 		buf->bs_projid = 0;
149 	} else {
150 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
151 		buf->bs_projid = be16_to_cpu(dic->di_projid);
152 	}
153 
154 	buf->bs_ino = ino;
155 	buf->bs_mode = be16_to_cpu(dic->di_mode);
156 	buf->bs_uid = be32_to_cpu(dic->di_uid);
157 	buf->bs_gid = be32_to_cpu(dic->di_gid);
158 	buf->bs_size = be64_to_cpu(dic->di_size);
159 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
160 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
161 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
162 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
163 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
164 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
165 	buf->bs_xflags = xfs_dic2xflags(dip);
166 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
167 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
168 	buf->bs_gen = be32_to_cpu(dic->di_gen);
169 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
170 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
171 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
172 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
173 
174 	switch (dic->di_format) {
175 	case XFS_DINODE_FMT_DEV:
176 		buf->bs_rdev = be32_to_cpu(dip->di_u.di_dev);
177 		buf->bs_blksize = BLKDEV_IOSIZE;
178 		buf->bs_blocks = 0;
179 		break;
180 	case XFS_DINODE_FMT_LOCAL:
181 	case XFS_DINODE_FMT_UUID:
182 		buf->bs_rdev = 0;
183 		buf->bs_blksize = mp->m_sb.sb_blocksize;
184 		buf->bs_blocks = 0;
185 		break;
186 	case XFS_DINODE_FMT_EXTENTS:
187 	case XFS_DINODE_FMT_BTREE:
188 		buf->bs_rdev = 0;
189 		buf->bs_blksize = mp->m_sb.sb_blocksize;
190 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
191 		break;
192 	}
193 }
194 
195 STATIC int
196 xfs_bulkstat_one_fmt(
197 	void			__user *ubuffer,
198 	const xfs_bstat_t	*buffer)
199 {
200 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
201 		return -EFAULT;
202 	return sizeof(*buffer);
203 }
204 
205 /*
206  * Return stat information for one inode.
207  * Return 0 if ok, else errno.
208  */
209 int		       		/* error status */
210 xfs_bulkstat_one(
211 	xfs_mount_t	*mp,		/* mount point for filesystem */
212 	xfs_ino_t	ino,		/* inode number to get data for */
213 	void		__user *buffer,	/* buffer to place output in */
214 	int		ubsize,		/* size of buffer */
215 	void		*private_data,	/* my private data */
216 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
217 	int		*ubused,	/* bytes used by me */
218 	void		*dibuff,	/* on-disk inode buffer */
219 	int		*stat)		/* BULKSTAT_RV_... */
220 {
221 	xfs_bstat_t	*buf;		/* return buffer */
222 	int		error = 0;	/* error value */
223 	xfs_dinode_t	*dip;		/* dinode inode pointer */
224 	bulkstat_one_fmt_pf formatter = private_data ? : xfs_bulkstat_one_fmt;
225 
226 	dip = (xfs_dinode_t *)dibuff;
227 	*stat = BULKSTAT_RV_NOTHING;
228 
229 	if (!buffer || xfs_internal_inum(mp, ino))
230 		return XFS_ERROR(EINVAL);
231 	if (ubsize < sizeof(*buf))
232 		return XFS_ERROR(ENOMEM);
233 
234 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
235 
236 	if (dip == NULL) {
237 		/* We're not being passed a pointer to a dinode.  This happens
238 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
239 		 */
240 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
241 		if (error)
242 			goto out_free;
243 	} else {
244 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
245 	}
246 
247 	error = formatter(buffer, buf);
248 	if (error < 0)  {
249 		error = EFAULT;
250 		goto out_free;
251 	}
252 
253 	*stat = BULKSTAT_RV_DIDONE;
254 	if (ubused)
255 		*ubused = error;
256 
257  out_free:
258 	kmem_free(buf);
259 	return error;
260 }
261 
262 /*
263  * Test to see whether we can use the ondisk inode directly, based
264  * on the given bulkstat flags, filling in dipp accordingly.
265  * Returns zero if the inode is dodgey.
266  */
267 STATIC int
268 xfs_bulkstat_use_dinode(
269 	xfs_mount_t	*mp,
270 	int		flags,
271 	xfs_buf_t	*bp,
272 	int		clustidx,
273 	xfs_dinode_t	**dipp)
274 {
275 	xfs_dinode_t	*dip;
276 	unsigned int	aformat;
277 
278 	*dipp = NULL;
279 	if (!bp || (flags & BULKSTAT_FG_IGET))
280 		return 1;
281 	dip = (xfs_dinode_t *)
282 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
283 	/*
284 	 * Check the buffer containing the on-disk inode for di_mode == 0.
285 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
286 	 * inodes that have their in-core state initialized but not flushed
287 	 * to disk yet. This is a temporary hack that would require a proper
288 	 * fix in the future.
289 	 */
290 	if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC ||
291 	    !XFS_DINODE_GOOD_VERSION(dip->di_core.di_version) ||
292 	    !dip->di_core.di_mode)
293 		return 0;
294 	if (flags & BULKSTAT_FG_QUICK) {
295 		*dipp = dip;
296 		return 1;
297 	}
298 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
299 	aformat = dip->di_core.di_aformat;
300 	if ((XFS_DFORK_Q(dip) == 0) ||
301 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
302 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
303 		*dipp = dip;
304 		return 1;
305 	}
306 	return 1;
307 }
308 
309 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
310 
311 /*
312  * Return stat information in bulk (by-inode) for the filesystem.
313  */
314 int					/* error status */
315 xfs_bulkstat(
316 	xfs_mount_t		*mp,	/* mount point for filesystem */
317 	xfs_ino_t		*lastinop, /* last inode returned */
318 	int			*ubcountp, /* size of buffer/count returned */
319 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
320 	void			*private_data,/* private data for formatter */
321 	size_t			statstruct_size, /* sizeof struct filling */
322 	char			__user *ubuffer, /* buffer with inode stats */
323 	int			flags,	/* defined in xfs_itable.h */
324 	int			*done)	/* 1 if there are more stats to get */
325 {
326 	xfs_agblock_t		agbno=0;/* allocation group block number */
327 	xfs_buf_t		*agbp;	/* agi header buffer */
328 	xfs_agi_t		*agi;	/* agi header data */
329 	xfs_agino_t		agino;	/* inode # in allocation group */
330 	xfs_agnumber_t		agno;	/* allocation group number */
331 	xfs_daddr_t		bno;	/* inode cluster start daddr */
332 	int			chunkidx; /* current index into inode chunk */
333 	int			clustidx; /* current index into inode cluster */
334 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
335 	int			end_of_ag; /* set if we've seen the ag end */
336 	int			error;	/* error code */
337 	int                     fmterror;/* bulkstat formatter result */
338 	__int32_t		gcnt;	/* current btree rec's count */
339 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
340 	xfs_agino_t		gino;	/* current btree rec's start inode */
341 	int			i;	/* loop index */
342 	int			icount;	/* count of inodes good in irbuf */
343 	size_t			irbsize; /* size of irec buffer in bytes */
344 	xfs_ino_t		ino;	/* inode number (filesystem) */
345 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
346 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
347 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
348 	xfs_ino_t		lastino; /* last inode number returned */
349 	int			nbcluster; /* # of blocks in a cluster */
350 	int			nicluster; /* # of inodes in a cluster */
351 	int			nimask;	/* mask for inode clusters */
352 	int			nirbuf;	/* size of irbuf */
353 	int			rval;	/* return value error code */
354 	int			tmp;	/* result value from btree calls */
355 	int			ubcount; /* size of user's buffer */
356 	int			ubleft;	/* bytes left in user's buffer */
357 	char			__user *ubufp;	/* pointer into user's buffer */
358 	int			ubelem;	/* spaces used in user's buffer */
359 	int			ubused;	/* bytes used by formatter */
360 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
361 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
362 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
363 
364 	/*
365 	 * Get the last inode value, see if there's nothing to do.
366 	 */
367 	ino = (xfs_ino_t)*lastinop;
368 	lastino = ino;
369 	dip = NULL;
370 	agno = XFS_INO_TO_AGNO(mp, ino);
371 	agino = XFS_INO_TO_AGINO(mp, ino);
372 	if (agno >= mp->m_sb.sb_agcount ||
373 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
374 		*done = 1;
375 		*ubcountp = 0;
376 		return 0;
377 	}
378 	if (!ubcountp || *ubcountp <= 0) {
379 		return EINVAL;
380 	}
381 	ubcount = *ubcountp; /* statstruct's */
382 	ubleft = ubcount * statstruct_size; /* bytes */
383 	*ubcountp = ubelem = 0;
384 	*done = 0;
385 	fmterror = 0;
386 	ubufp = ubuffer;
387 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
388 		mp->m_sb.sb_inopblock :
389 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
390 	nimask = ~(nicluster - 1);
391 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
392 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
393 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
394 	nirbuf = irbsize / sizeof(*irbuf);
395 
396 	/*
397 	 * Loop over the allocation groups, starting from the last
398 	 * inode returned; 0 means start of the allocation group.
399 	 */
400 	rval = 0;
401 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
402 		cond_resched();
403 		bp = NULL;
404 		down_read(&mp->m_peraglock);
405 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
406 		up_read(&mp->m_peraglock);
407 		if (error) {
408 			/*
409 			 * Skip this allocation group and go to the next one.
410 			 */
411 			agno++;
412 			agino = 0;
413 			continue;
414 		}
415 		agi = XFS_BUF_TO_AGI(agbp);
416 		/*
417 		 * Allocate and initialize a btree cursor for ialloc btree.
418 		 */
419 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
420 						(xfs_inode_t *)0, 0);
421 		irbp = irbuf;
422 		irbufend = irbuf + nirbuf;
423 		end_of_ag = 0;
424 		/*
425 		 * If we're returning in the middle of an allocation group,
426 		 * we need to get the remainder of the chunk we're in.
427 		 */
428 		if (agino > 0) {
429 			/*
430 			 * Lookup the inode chunk that this inode lives in.
431 			 */
432 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
433 			if (!error &&	/* no I/O error */
434 			    tmp &&	/* lookup succeeded */
435 					/* got the record, should always work */
436 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
437 				    &gfree, &i)) &&
438 			    i == 1 &&
439 					/* this is the right chunk */
440 			    agino < gino + XFS_INODES_PER_CHUNK &&
441 					/* lastino was not last in chunk */
442 			    (chunkidx = agino - gino + 1) <
443 				    XFS_INODES_PER_CHUNK &&
444 					/* there are some left allocated */
445 			    XFS_INOBT_MASKN(chunkidx,
446 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
447 				/*
448 				 * Grab the chunk record.  Mark all the
449 				 * uninteresting inodes (because they're
450 				 * before our start point) free.
451 				 */
452 				for (i = 0; i < chunkidx; i++) {
453 					if (XFS_INOBT_MASK(i) & ~gfree)
454 						gcnt++;
455 				}
456 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
457 				irbp->ir_startino = gino;
458 				irbp->ir_freecount = gcnt;
459 				irbp->ir_free = gfree;
460 				irbp++;
461 				agino = gino + XFS_INODES_PER_CHUNK;
462 				icount = XFS_INODES_PER_CHUNK - gcnt;
463 			} else {
464 				/*
465 				 * If any of those tests failed, bump the
466 				 * inode number (just in case).
467 				 */
468 				agino++;
469 				icount = 0;
470 			}
471 			/*
472 			 * In any case, increment to the next record.
473 			 */
474 			if (!error)
475 				error = xfs_inobt_increment(cur, 0, &tmp);
476 		} else {
477 			/*
478 			 * Start of ag.  Lookup the first inode chunk.
479 			 */
480 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
481 			icount = 0;
482 		}
483 		/*
484 		 * Loop through inode btree records in this ag,
485 		 * until we run out of inodes or space in the buffer.
486 		 */
487 		while (irbp < irbufend && icount < ubcount) {
488 			/*
489 			 * Loop as long as we're unable to read the
490 			 * inode btree.
491 			 */
492 			while (error) {
493 				agino += XFS_INODES_PER_CHUNK;
494 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
495 						be32_to_cpu(agi->agi_length))
496 					break;
497 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
498 							    &tmp);
499 				cond_resched();
500 			}
501 			/*
502 			 * If ran off the end of the ag either with an error,
503 			 * or the normal way, set end and stop collecting.
504 			 */
505 			if (error ||
506 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
507 				    &gfree, &i)) ||
508 			    i == 0) {
509 				end_of_ag = 1;
510 				break;
511 			}
512 			/*
513 			 * If this chunk has any allocated inodes, save it.
514 			 * Also start read-ahead now for this chunk.
515 			 */
516 			if (gcnt < XFS_INODES_PER_CHUNK) {
517 				/*
518 				 * Loop over all clusters in the next chunk.
519 				 * Do a readahead if there are any allocated
520 				 * inodes in that cluster.
521 				 */
522 				for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
523 				     chunkidx = 0;
524 				     chunkidx < XFS_INODES_PER_CHUNK;
525 				     chunkidx += nicluster,
526 				     agbno += nbcluster) {
527 					if (XFS_INOBT_MASKN(chunkidx,
528 							    nicluster) & ~gfree)
529 						xfs_btree_reada_bufs(mp, agno,
530 							agbno, nbcluster);
531 				}
532 				irbp->ir_startino = gino;
533 				irbp->ir_freecount = gcnt;
534 				irbp->ir_free = gfree;
535 				irbp++;
536 				icount += XFS_INODES_PER_CHUNK - gcnt;
537 			}
538 			/*
539 			 * Set agino to after this chunk and bump the cursor.
540 			 */
541 			agino = gino + XFS_INODES_PER_CHUNK;
542 			error = xfs_inobt_increment(cur, 0, &tmp);
543 			cond_resched();
544 		}
545 		/*
546 		 * Drop the btree buffers and the agi buffer.
547 		 * We can't hold any of the locks these represent
548 		 * when calling iget.
549 		 */
550 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
551 		xfs_buf_relse(agbp);
552 		/*
553 		 * Now format all the good inodes into the user's buffer.
554 		 */
555 		irbufend = irbp;
556 		for (irbp = irbuf;
557 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
558 			/*
559 			 * Now process this chunk of inodes.
560 			 */
561 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
562 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
563 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
564 			     chunkidx++, clustidx++, agino++) {
565 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
566 				/*
567 				 * Recompute agbno if this is the
568 				 * first inode of the cluster.
569 				 *
570 				 * Careful with clustidx.   There can be
571 				 * multple clusters per chunk, a single
572 				 * cluster per chunk or a cluster that has
573 				 * inodes represented from several different
574 				 * chunks (if blocksize is large).
575 				 *
576 				 * Because of this, the starting clustidx is
577 				 * initialized to zero in this loop but must
578 				 * later be reset after reading in the cluster
579 				 * buffer.
580 				 */
581 				if ((chunkidx & (nicluster - 1)) == 0) {
582 					agbno = XFS_AGINO_TO_AGBNO(mp,
583 							irbp->ir_startino) +
584 						((chunkidx & nimask) >>
585 						 mp->m_sb.sb_inopblog);
586 
587 					if (flags & (BULKSTAT_FG_QUICK |
588 						     BULKSTAT_FG_INLINE)) {
589 						ino = XFS_AGINO_TO_INO(mp, agno,
590 								       agino);
591 						bno = XFS_AGB_TO_DADDR(mp, agno,
592 								       agbno);
593 
594 						/*
595 						 * Get the inode cluster buffer
596 						 */
597 						ASSERT(xfs_inode_zone != NULL);
598 						ip = kmem_zone_zalloc(xfs_inode_zone,
599 								      KM_SLEEP);
600 						ip->i_ino = ino;
601 						ip->i_mount = mp;
602 						spin_lock_init(&ip->i_flags_lock);
603 						if (bp)
604 							xfs_buf_relse(bp);
605 						error = xfs_itobp(mp, NULL, ip,
606 								&dip, &bp, bno,
607 								XFS_IMAP_BULKSTAT,
608 								XFS_BUF_LOCK);
609 						if (!error)
610 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
611 						kmem_zone_free(xfs_inode_zone, ip);
612 						if (XFS_TEST_ERROR(error != 0,
613 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
614 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
615 							bp = NULL;
616 							ubleft = 0;
617 							rval = error;
618 							break;
619 						}
620 					}
621 				}
622 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
623 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
624 				/*
625 				 * Skip if this inode is free.
626 				 */
627 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
628 					lastino = ino;
629 					continue;
630 				}
631 				/*
632 				 * Count used inodes as free so we can tell
633 				 * when the chunk is used up.
634 				 */
635 				irbp->ir_freecount++;
636 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
637 							     clustidx, &dip)) {
638 					lastino = ino;
639 					continue;
640 				}
641 				/*
642 				 * If we need to do an iget, cannot hold bp.
643 				 * Drop it, until starting the next cluster.
644 				 */
645 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
646 					if (bp)
647 						xfs_buf_relse(bp);
648 					bp = NULL;
649 				}
650 
651 				/*
652 				 * Get the inode and fill in a single buffer.
653 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
654 				 * BULKSTAT_FG_IGET uses igets.
655 				 * BULKSTAT_FG_INLINE uses dip if we have an
656 				 * inline attr fork, else igets.
657 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
658 				 * This is also used to count inodes/blks, etc
659 				 * in xfs_qm_quotacheck.
660 				 */
661 				ubused = statstruct_size;
662 				error = formatter(mp, ino, ubufp,
663 						ubleft, private_data,
664 						bno, &ubused, dip, &fmterror);
665 				if (fmterror == BULKSTAT_RV_NOTHING) {
666 					if (error && error != ENOENT &&
667 						error != EINVAL) {
668 						ubleft = 0;
669 						rval = error;
670 						break;
671 					}
672 					lastino = ino;
673 					continue;
674 				}
675 				if (fmterror == BULKSTAT_RV_GIVEUP) {
676 					ubleft = 0;
677 					ASSERT(error);
678 					rval = error;
679 					break;
680 				}
681 				if (ubufp)
682 					ubufp += ubused;
683 				ubleft -= ubused;
684 				ubelem++;
685 				lastino = ino;
686 			}
687 
688 			cond_resched();
689 		}
690 
691 		if (bp)
692 			xfs_buf_relse(bp);
693 
694 		/*
695 		 * Set up for the next loop iteration.
696 		 */
697 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
698 			if (end_of_ag) {
699 				agno++;
700 				agino = 0;
701 			} else
702 				agino = XFS_INO_TO_AGINO(mp, lastino);
703 		} else
704 			break;
705 	}
706 	/*
707 	 * Done, we're either out of filesystem or space to put the data.
708 	 */
709 	kmem_free(irbuf);
710 	*ubcountp = ubelem;
711 	/*
712 	 * Found some inodes, return them now and return the error next time.
713 	 */
714 	if (ubelem)
715 		rval = 0;
716 	if (agno >= mp->m_sb.sb_agcount) {
717 		/*
718 		 * If we ran out of filesystem, mark lastino as off
719 		 * the end of the filesystem, so the next call
720 		 * will return immediately.
721 		 */
722 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
723 		*done = 1;
724 	} else
725 		*lastinop = (xfs_ino_t)lastino;
726 
727 	return rval;
728 }
729 
730 /*
731  * Return stat information in bulk (by-inode) for the filesystem.
732  * Special case for non-sequential one inode bulkstat.
733  */
734 int					/* error status */
735 xfs_bulkstat_single(
736 	xfs_mount_t		*mp,	/* mount point for filesystem */
737 	xfs_ino_t		*lastinop, /* inode to return */
738 	char			__user *buffer, /* buffer with inode stats */
739 	int			*done)	/* 1 if there are more stats to get */
740 {
741 	int			count;	/* count value for bulkstat call */
742 	int			error;	/* return value */
743 	xfs_ino_t		ino;	/* filesystem inode number */
744 	int			res;	/* result from bs1 */
745 
746 	/*
747 	 * note that requesting valid inode numbers which are not allocated
748 	 * to inodes will most likely cause xfs_itobp to generate warning
749 	 * messages about bad magic numbers. This is ok. The fact that
750 	 * the inode isn't actually an inode is handled by the
751 	 * error check below. Done this way to make the usual case faster
752 	 * at the expense of the error case.
753 	 */
754 
755 	ino = (xfs_ino_t)*lastinop;
756 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
757 				 NULL, 0, NULL, NULL, &res);
758 	if (error) {
759 		/*
760 		 * Special case way failed, do it the "long" way
761 		 * to see if that works.
762 		 */
763 		(*lastinop)--;
764 		count = 1;
765 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
766 				NULL, sizeof(xfs_bstat_t), buffer,
767 				BULKSTAT_FG_IGET, done))
768 			return error;
769 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
770 			return error == EFSCORRUPTED ?
771 				XFS_ERROR(EINVAL) : error;
772 		else
773 			return 0;
774 	}
775 	*done = 0;
776 	return 0;
777 }
778 
779 int
780 xfs_inumbers_fmt(
781 	void			__user *ubuffer, /* buffer to write to */
782 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
783 	long			count,		/* # of elements to read */
784 	long			*written)	/* # of bytes written */
785 {
786 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
787 		return -EFAULT;
788 	*written = count * sizeof(*buffer);
789 	return 0;
790 }
791 
792 /*
793  * Return inode number table for the filesystem.
794  */
795 int					/* error status */
796 xfs_inumbers(
797 	xfs_mount_t	*mp,		/* mount point for filesystem */
798 	xfs_ino_t	*lastino,	/* last inode returned */
799 	int		*count,		/* size of buffer/count returned */
800 	void		__user *ubuffer,/* buffer with inode descriptions */
801 	inumbers_fmt_pf	formatter)
802 {
803 	xfs_buf_t	*agbp;
804 	xfs_agino_t	agino;
805 	xfs_agnumber_t	agno;
806 	int		bcount;
807 	xfs_inogrp_t	*buffer;
808 	int		bufidx;
809 	xfs_btree_cur_t	*cur;
810 	int		error;
811 	__int32_t	gcnt;
812 	xfs_inofree_t	gfree;
813 	xfs_agino_t	gino;
814 	int		i;
815 	xfs_ino_t	ino;
816 	int		left;
817 	int		tmp;
818 
819 	ino = (xfs_ino_t)*lastino;
820 	agno = XFS_INO_TO_AGNO(mp, ino);
821 	agino = XFS_INO_TO_AGINO(mp, ino);
822 	left = *count;
823 	*count = 0;
824 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
825 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
826 	error = bufidx = 0;
827 	cur = NULL;
828 	agbp = NULL;
829 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
830 		if (agbp == NULL) {
831 			down_read(&mp->m_peraglock);
832 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
833 			up_read(&mp->m_peraglock);
834 			if (error) {
835 				/*
836 				 * If we can't read the AGI of this ag,
837 				 * then just skip to the next one.
838 				 */
839 				ASSERT(cur == NULL);
840 				agbp = NULL;
841 				agno++;
842 				agino = 0;
843 				continue;
844 			}
845 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
846 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
847 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
848 			if (error) {
849 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
850 				cur = NULL;
851 				xfs_buf_relse(agbp);
852 				agbp = NULL;
853 				/*
854 				 * Move up the last inode in the current
855 				 * chunk.  The lookup_ge will always get
856 				 * us the first inode in the next chunk.
857 				 */
858 				agino += XFS_INODES_PER_CHUNK - 1;
859 				continue;
860 			}
861 		}
862 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
863 			&i)) ||
864 		    i == 0) {
865 			xfs_buf_relse(agbp);
866 			agbp = NULL;
867 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
868 			cur = NULL;
869 			agno++;
870 			agino = 0;
871 			continue;
872 		}
873 		agino = gino + XFS_INODES_PER_CHUNK - 1;
874 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
875 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
876 		buffer[bufidx].xi_allocmask = ~gfree;
877 		bufidx++;
878 		left--;
879 		if (bufidx == bcount) {
880 			long written;
881 			if (formatter(ubuffer, buffer, bufidx, &written)) {
882 				error = XFS_ERROR(EFAULT);
883 				break;
884 			}
885 			ubuffer += written;
886 			*count += bufidx;
887 			bufidx = 0;
888 		}
889 		if (left) {
890 			error = xfs_inobt_increment(cur, 0, &tmp);
891 			if (error) {
892 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
893 				cur = NULL;
894 				xfs_buf_relse(agbp);
895 				agbp = NULL;
896 				/*
897 				 * The agino value has already been bumped.
898 				 * Just try to skip up to it.
899 				 */
900 				agino += XFS_INODES_PER_CHUNK;
901 				continue;
902 			}
903 		}
904 	}
905 	if (!error) {
906 		if (bufidx) {
907 			long written;
908 			if (formatter(ubuffer, buffer, bufidx, &written))
909 				error = XFS_ERROR(EFAULT);
910 			else
911 				*count += bufidx;
912 		}
913 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
914 	}
915 	kmem_free(buffer);
916 	if (cur)
917 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
918 					   XFS_BTREE_NOERROR));
919 	if (agbp)
920 		xfs_buf_relse(agbp);
921 	return error;
922 }
923