xref: /linux/fs/xfs/xfs_itable.c (revision d8327c784b51b57dac2c26cfad87dce0d68dfd98)
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_dir.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_error.h"
42 #include "xfs_btree.h"
43 
44 #ifndef HAVE_USERACC
45 #define useracc(ubuffer, size, flags, foo) (0)
46 #define unuseracc(ubuffer, size, flags)
47 #endif
48 
49 STATIC int
50 xfs_bulkstat_one_iget(
51 	xfs_mount_t	*mp,		/* mount point for filesystem */
52 	xfs_ino_t	ino,		/* inode number to get data for */
53 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
54 	xfs_bstat_t	*buf,		/* return buffer */
55 	int		*stat)		/* BULKSTAT_RV_... */
56 {
57 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
58 	xfs_inode_t	*ip;		/* incore inode pointer */
59 	vnode_t		*vp;
60 	int		error;
61 
62 	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
63 	if (error) {
64 		*stat = BULKSTAT_RV_NOTHING;
65 		return error;
66 	}
67 
68 	ASSERT(ip != NULL);
69 	ASSERT(ip->i_blkno != (xfs_daddr_t)0);
70 	if (ip->i_d.di_mode == 0) {
71 		*stat = BULKSTAT_RV_NOTHING;
72 		error = XFS_ERROR(ENOENT);
73 		goto out_iput;
74 	}
75 
76 	vp = XFS_ITOV(ip);
77 	dic = &ip->i_d;
78 
79 	/* xfs_iget returns the following without needing
80 	 * further change.
81 	 */
82 	buf->bs_nlink = dic->di_nlink;
83 	buf->bs_projid = dic->di_projid;
84 	buf->bs_ino = ino;
85 	buf->bs_mode = dic->di_mode;
86 	buf->bs_uid = dic->di_uid;
87 	buf->bs_gid = dic->di_gid;
88 	buf->bs_size = dic->di_size;
89 	vn_atime_to_bstime(vp, &buf->bs_atime);
90 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
91 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
92 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
93 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
94 	buf->bs_xflags = xfs_ip2xflags(ip);
95 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
96 	buf->bs_extents = dic->di_nextents;
97 	buf->bs_gen = dic->di_gen;
98 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
99 	buf->bs_dmevmask = dic->di_dmevmask;
100 	buf->bs_dmstate = dic->di_dmstate;
101 	buf->bs_aextents = dic->di_anextents;
102 
103 	switch (dic->di_format) {
104 	case XFS_DINODE_FMT_DEV:
105 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
106 		buf->bs_blksize = BLKDEV_IOSIZE;
107 		buf->bs_blocks = 0;
108 		break;
109 	case XFS_DINODE_FMT_LOCAL:
110 	case XFS_DINODE_FMT_UUID:
111 		buf->bs_rdev = 0;
112 		buf->bs_blksize = mp->m_sb.sb_blocksize;
113 		buf->bs_blocks = 0;
114 		break;
115 	case XFS_DINODE_FMT_EXTENTS:
116 	case XFS_DINODE_FMT_BTREE:
117 		buf->bs_rdev = 0;
118 		buf->bs_blksize = mp->m_sb.sb_blocksize;
119 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
120 		break;
121 	}
122 
123  out_iput:
124 	xfs_iput(ip, XFS_ILOCK_SHARED);
125 	return error;
126 }
127 
128 STATIC int
129 xfs_bulkstat_one_dinode(
130 	xfs_mount_t	*mp,		/* mount point for filesystem */
131 	xfs_ino_t	ino,		/* inode number to get data for */
132 	xfs_dinode_t	*dip,		/* dinode inode pointer */
133 	xfs_bstat_t	*buf)		/* return buffer */
134 {
135 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
136 
137 	dic = &dip->di_core;
138 
139 	/*
140 	 * The inode format changed when we moved the link count and
141 	 * made it 32 bits long.  If this is an old format inode,
142 	 * convert it in memory to look like a new one.  If it gets
143 	 * flushed to disk we will convert back before flushing or
144 	 * logging it.  We zero out the new projid field and the old link
145 	 * count field.  We'll handle clearing the pad field (the remains
146 	 * of the old uuid field) when we actually convert the inode to
147 	 * the new format. We don't change the version number so that we
148 	 * can distinguish this from a real new format inode.
149 	 */
150 	if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
151 		buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
152 		buf->bs_projid = 0;
153 	} else {
154 		buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
155 		buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
156 	}
157 
158 	buf->bs_ino = ino;
159 	buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
160 	buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
161 	buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
162 	buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
163 	buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
164 	buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
165 	buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
166 	buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
167 	buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
168 	buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
169 	buf->bs_xflags = xfs_dic2xflags(dic);
170 	buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
171 	buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
172 	buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
173 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
174 	buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
175 	buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
176 	buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
177 
178 	switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
179 	case XFS_DINODE_FMT_DEV:
180 		buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
181 		buf->bs_blksize = BLKDEV_IOSIZE;
182 		buf->bs_blocks = 0;
183 		break;
184 	case XFS_DINODE_FMT_LOCAL:
185 	case XFS_DINODE_FMT_UUID:
186 		buf->bs_rdev = 0;
187 		buf->bs_blksize = mp->m_sb.sb_blocksize;
188 		buf->bs_blocks = 0;
189 		break;
190 	case XFS_DINODE_FMT_EXTENTS:
191 	case XFS_DINODE_FMT_BTREE:
192 		buf->bs_rdev = 0;
193 		buf->bs_blksize = mp->m_sb.sb_blocksize;
194 		buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
195 		break;
196 	}
197 
198 	return 0;
199 }
200 
201 /*
202  * Return stat information for one inode.
203  * Return 0 if ok, else errno.
204  */
205 int		       		/* error status */
206 xfs_bulkstat_one(
207 	xfs_mount_t	*mp,		/* mount point for filesystem */
208 	xfs_ino_t	ino,		/* inode number to get data for */
209 	void		__user *buffer,	/* buffer to place output in */
210 	int		ubsize,		/* size of buffer */
211 	void		*private_data,	/* my private data */
212 	xfs_daddr_t	bno,		/* starting bno of inode cluster */
213 	int		*ubused,	/* bytes used by me */
214 	void		*dibuff,	/* on-disk inode buffer */
215 	int		*stat)		/* BULKSTAT_RV_... */
216 {
217 	xfs_bstat_t	*buf;		/* return buffer */
218 	int		error = 0;	/* error value */
219 	xfs_dinode_t	*dip;		/* dinode inode pointer */
220 
221 	dip = (xfs_dinode_t *)dibuff;
222 
223 	if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
224 	    (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
225 	     (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
226 		*stat = BULKSTAT_RV_NOTHING;
227 		return XFS_ERROR(EINVAL);
228 	}
229 	if (ubsize < sizeof(*buf)) {
230 		*stat = BULKSTAT_RV_NOTHING;
231 		return XFS_ERROR(ENOMEM);
232 	}
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 	if (copy_to_user(buffer, buf, sizeof(*buf)))  {
248 		*stat = BULKSTAT_RV_NOTHING;
249 		error =  EFAULT;
250 		goto out_free;
251 	}
252 
253 	*stat = BULKSTAT_RV_DIDONE;
254 	if (ubused)
255 		*ubused = sizeof(*buf);
256 
257  out_free:
258 	kmem_free(buf, sizeof(*buf));
259 	return error;
260 }
261 
262 /*
263  * Return stat information in bulk (by-inode) for the filesystem.
264  */
265 int					/* error status */
266 xfs_bulkstat(
267 	xfs_mount_t		*mp,	/* mount point for filesystem */
268 	xfs_ino_t		*lastinop, /* last inode returned */
269 	int			*ubcountp, /* size of buffer/count returned */
270 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
271 	void			*private_data,/* private data for formatter */
272 	size_t			statstruct_size, /* sizeof struct filling */
273 	char			__user *ubuffer, /* buffer with inode stats */
274 	int			flags,	/* defined in xfs_itable.h */
275 	int			*done)	/* 1 if there're more stats to get */
276 {
277 	xfs_agblock_t		agbno=0;/* allocation group block number */
278 	xfs_buf_t		*agbp;	/* agi header buffer */
279 	xfs_agi_t		*agi;	/* agi header data */
280 	xfs_agino_t		agino;	/* inode # in allocation group */
281 	xfs_agnumber_t		agno;	/* allocation group number */
282 	xfs_daddr_t		bno;	/* inode cluster start daddr */
283 	int			chunkidx; /* current index into inode chunk */
284 	int			clustidx; /* current index into inode cluster */
285 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
286 	int			end_of_ag; /* set if we've seen the ag end */
287 	int			error;	/* error code */
288 	int                     fmterror;/* bulkstat formatter result */
289 	__int32_t		gcnt;	/* current btree rec's count */
290 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
291 	xfs_agino_t		gino;	/* current btree rec's start inode */
292 	int			i;	/* loop index */
293 	int			icount;	/* count of inodes good in irbuf */
294 	xfs_ino_t		ino;	/* inode number (filesystem) */
295 	xfs_inobt_rec_t		*irbp;	/* current irec buffer pointer */
296 	xfs_inobt_rec_t		*irbuf;	/* start of irec buffer */
297 	xfs_inobt_rec_t		*irbufend; /* end of good irec buffer entries */
298 	xfs_ino_t		lastino=0; /* last inode number returned */
299 	int			nbcluster; /* # of blocks in a cluster */
300 	int			nicluster; /* # of inodes in a cluster */
301 	int			nimask;	/* mask for inode clusters */
302 	int			nirbuf;	/* size of irbuf */
303 	int			rval;	/* return value error code */
304 	int			tmp;	/* result value from btree calls */
305 	int			ubcount; /* size of user's buffer */
306 	int			ubleft;	/* bytes left in user's buffer */
307 	char			__user *ubufp;	/* pointer into user's buffer */
308 	int			ubelem;	/* spaces used in user's buffer */
309 	int			ubused;	/* bytes used by formatter */
310 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
311 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
312 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
313 
314 	/*
315 	 * Get the last inode value, see if there's nothing to do.
316 	 */
317 	ino = (xfs_ino_t)*lastinop;
318 	dip = NULL;
319 	agno = XFS_INO_TO_AGNO(mp, ino);
320 	agino = XFS_INO_TO_AGINO(mp, ino);
321 	if (agno >= mp->m_sb.sb_agcount ||
322 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
323 		*done = 1;
324 		*ubcountp = 0;
325 		return 0;
326 	}
327 	ubcount = *ubcountp; /* statstruct's */
328 	ubleft = ubcount * statstruct_size; /* bytes */
329 	*ubcountp = ubelem = 0;
330 	*done = 0;
331 	fmterror = 0;
332 	ubufp = ubuffer;
333 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
334 		mp->m_sb.sb_inopblock :
335 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
336 	nimask = ~(nicluster - 1);
337 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
338 	/*
339 	 * Lock down the user's buffer. If a buffer was not sent, as in the case
340 	 * disk quota code calls here, we skip this.
341 	 */
342 	if (ubuffer &&
343 	    (error = useracc(ubuffer, ubcount * statstruct_size,
344 			(B_READ|B_PHYS), NULL))) {
345 		return error;
346 	}
347 	/*
348 	 * Allocate a page-sized buffer for inode btree records.
349 	 * We could try allocating something smaller, but for normal
350 	 * calls we'll always (potentially) need the whole page.
351 	 */
352 	irbuf = kmem_alloc(NBPC, KM_SLEEP);
353 	nirbuf = NBPC / sizeof(*irbuf);
354 	/*
355 	 * Loop over the allocation groups, starting from the last
356 	 * inode returned; 0 means start of the allocation group.
357 	 */
358 	rval = 0;
359 	while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
360 		bp = NULL;
361 		down_read(&mp->m_peraglock);
362 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
363 		up_read(&mp->m_peraglock);
364 		if (error) {
365 			/*
366 			 * Skip this allocation group and go to the next one.
367 			 */
368 			agno++;
369 			agino = 0;
370 			continue;
371 		}
372 		agi = XFS_BUF_TO_AGI(agbp);
373 		/*
374 		 * Allocate and initialize a btree cursor for ialloc btree.
375 		 */
376 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
377 			(xfs_inode_t *)0, 0);
378 		irbp = irbuf;
379 		irbufend = irbuf + nirbuf;
380 		end_of_ag = 0;
381 		/*
382 		 * If we're returning in the middle of an allocation group,
383 		 * we need to get the remainder of the chunk we're in.
384 		 */
385 		if (agino > 0) {
386 			/*
387 			 * Lookup the inode chunk that this inode lives in.
388 			 */
389 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
390 			if (!error &&	/* no I/O error */
391 			    tmp &&	/* lookup succeeded */
392 					/* got the record, should always work */
393 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
394 				    &gfree, &i)) &&
395 			    i == 1 &&
396 					/* this is the right chunk */
397 			    agino < gino + XFS_INODES_PER_CHUNK &&
398 					/* lastino was not last in chunk */
399 			    (chunkidx = agino - gino + 1) <
400 				    XFS_INODES_PER_CHUNK &&
401 					/* there are some left allocated */
402 			    XFS_INOBT_MASKN(chunkidx,
403 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
404 				/*
405 				 * Grab the chunk record.  Mark all the
406 				 * uninteresting inodes (because they're
407 				 * before our start point) free.
408 				 */
409 				for (i = 0; i < chunkidx; i++) {
410 					if (XFS_INOBT_MASK(i) & ~gfree)
411 						gcnt++;
412 				}
413 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
414 				INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
415 				INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
416 				INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
417 				irbp++;
418 				agino = gino + XFS_INODES_PER_CHUNK;
419 				icount = XFS_INODES_PER_CHUNK - gcnt;
420 			} else {
421 				/*
422 				 * If any of those tests failed, bump the
423 				 * inode number (just in case).
424 				 */
425 				agino++;
426 				icount = 0;
427 			}
428 			/*
429 			 * In any case, increment to the next record.
430 			 */
431 			if (!error)
432 				error = xfs_inobt_increment(cur, 0, &tmp);
433 		} else {
434 			/*
435 			 * Start of ag.  Lookup the first inode chunk.
436 			 */
437 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
438 			icount = 0;
439 		}
440 		/*
441 		 * Loop through inode btree records in this ag,
442 		 * until we run out of inodes or space in the buffer.
443 		 */
444 		while (irbp < irbufend && icount < ubcount) {
445 			/*
446 			 * Loop as long as we're unable to read the
447 			 * inode btree.
448 			 */
449 			while (error) {
450 				agino += XFS_INODES_PER_CHUNK;
451 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
452 						be32_to_cpu(agi->agi_length))
453 					break;
454 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
455 							    &tmp);
456 			}
457 			/*
458 			 * If ran off the end of the ag either with an error,
459 			 * or the normal way, set end and stop collecting.
460 			 */
461 			if (error ||
462 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
463 				    &gfree, &i)) ||
464 			    i == 0) {
465 				end_of_ag = 1;
466 				break;
467 			}
468 			/*
469 			 * If this chunk has any allocated inodes, save it.
470 			 */
471 			if (gcnt < XFS_INODES_PER_CHUNK) {
472 				INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
473 				INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
474 				INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
475 				irbp++;
476 				icount += XFS_INODES_PER_CHUNK - gcnt;
477 			}
478 			/*
479 			 * Set agino to after this chunk and bump the cursor.
480 			 */
481 			agino = gino + XFS_INODES_PER_CHUNK;
482 			error = xfs_inobt_increment(cur, 0, &tmp);
483 		}
484 		/*
485 		 * Drop the btree buffers and the agi buffer.
486 		 * We can't hold any of the locks these represent
487 		 * when calling iget.
488 		 */
489 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
490 		xfs_buf_relse(agbp);
491 		/*
492 		 * Now format all the good inodes into the user's buffer.
493 		 */
494 		irbufend = irbp;
495 		for (irbp = irbuf;
496 		     irbp < irbufend && ubleft >= statstruct_size; irbp++) {
497 			/*
498 			 * Read-ahead the next chunk's worth of inodes.
499 			 */
500 			if (&irbp[1] < irbufend) {
501 				/*
502 				 * Loop over all clusters in the next chunk.
503 				 * Do a readahead if there are any allocated
504 				 * inodes in that cluster.
505 				 */
506 				for (agbno = XFS_AGINO_TO_AGBNO(mp,
507 							INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
508 				     chunkidx = 0;
509 				     chunkidx < XFS_INODES_PER_CHUNK;
510 				     chunkidx += nicluster,
511 				     agbno += nbcluster) {
512 					if (XFS_INOBT_MASKN(chunkidx,
513 							    nicluster) &
514 					    ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
515 						xfs_btree_reada_bufs(mp, agno,
516 							agbno, nbcluster);
517 				}
518 			}
519 			/*
520 			 * Now process this chunk of inodes.
521 			 */
522 			for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
523 			     ubleft > 0 &&
524 				INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
525 			     chunkidx++, clustidx++, agino++) {
526 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
527 				/*
528 				 * Recompute agbno if this is the
529 				 * first inode of the cluster.
530 				 *
531 				 * Careful with clustidx.   There can be
532 				 * multple clusters per chunk, a single
533 				 * cluster per chunk or a cluster that has
534 				 * inodes represented from several different
535 				 * chunks (if blocksize is large).
536 				 *
537 				 * Because of this, the starting clustidx is
538 				 * initialized to zero in this loop but must
539 				 * later be reset after reading in the cluster
540 				 * buffer.
541 				 */
542 				if ((chunkidx & (nicluster - 1)) == 0) {
543 					agbno = XFS_AGINO_TO_AGBNO(mp,
544 							INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
545 						((chunkidx & nimask) >>
546 						 mp->m_sb.sb_inopblog);
547 
548 					if (flags & BULKSTAT_FG_QUICK) {
549 						ino = XFS_AGINO_TO_INO(mp, agno,
550 								       agino);
551 						bno = XFS_AGB_TO_DADDR(mp, agno,
552 								       agbno);
553 
554 						/*
555 						 * Get the inode cluster buffer
556 						 */
557 						ASSERT(xfs_inode_zone != NULL);
558 						ip = kmem_zone_zalloc(xfs_inode_zone,
559 								      KM_SLEEP);
560 						ip->i_ino = ino;
561 						ip->i_mount = mp;
562 						if (bp)
563 							xfs_buf_relse(bp);
564 						error = xfs_itobp(mp, NULL, ip,
565 								  &dip, &bp, bno);
566 						if (!error)
567 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
568 						kmem_zone_free(xfs_inode_zone, ip);
569 						if (XFS_TEST_ERROR(error != 0,
570 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
571 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
572 							bp = NULL;
573 							break;
574 						}
575 					}
576 				}
577 				/*
578 				 * Skip if this inode is free.
579 				 */
580 				if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
581 					continue;
582 				/*
583 				 * Count used inodes as free so we can tell
584 				 * when the chunk is used up.
585 				 */
586 				INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
587 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
588 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
589 				if (flags & BULKSTAT_FG_QUICK) {
590 					dip = (xfs_dinode_t *)xfs_buf_offset(bp,
591 					      (clustidx << mp->m_sb.sb_inodelog));
592 
593 					if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
594 						    != XFS_DINODE_MAGIC
595 					    || !XFS_DINODE_GOOD_VERSION(
596 						    INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
597 						continue;
598 				}
599 
600 				/*
601 				 * Get the inode and fill in a single buffer.
602 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
603 				 * BULKSTAT_FG_IGET uses igets.
604 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
605 				 * This is also used to count inodes/blks, etc
606 				 * in xfs_qm_quotacheck.
607 				 */
608 				ubused = statstruct_size;
609 				error = formatter(mp, ino, ubufp,
610 						ubleft, private_data,
611 						bno, &ubused, dip, &fmterror);
612 				if (fmterror == BULKSTAT_RV_NOTHING) {
613 					if (error == ENOMEM)
614 						ubleft = 0;
615 					continue;
616 				}
617 				if (fmterror == BULKSTAT_RV_GIVEUP) {
618 					ubleft = 0;
619 					ASSERT(error);
620 					rval = error;
621 					break;
622 				}
623 				if (ubufp)
624 					ubufp += ubused;
625 				ubleft -= ubused;
626 				ubelem++;
627 				lastino = ino;
628 			}
629 		}
630 
631 		if (bp)
632 			xfs_buf_relse(bp);
633 
634 		/*
635 		 * Set up for the next loop iteration.
636 		 */
637 		if (ubleft > 0) {
638 			if (end_of_ag) {
639 				agno++;
640 				agino = 0;
641 			} else
642 				agino = XFS_INO_TO_AGINO(mp, lastino);
643 		} else
644 			break;
645 	}
646 	/*
647 	 * Done, we're either out of filesystem or space to put the data.
648 	 */
649 	kmem_free(irbuf, NBPC);
650 	if (ubuffer)
651 		unuseracc(ubuffer, ubcount * statstruct_size, (B_READ|B_PHYS));
652 	*ubcountp = ubelem;
653 	if (agno >= mp->m_sb.sb_agcount) {
654 		/*
655 		 * If we ran out of filesystem, mark lastino as off
656 		 * the end of the filesystem, so the next call
657 		 * will return immediately.
658 		 */
659 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
660 		*done = 1;
661 	} else
662 		*lastinop = (xfs_ino_t)lastino;
663 
664 	return rval;
665 }
666 
667 /*
668  * Return stat information in bulk (by-inode) for the filesystem.
669  * Special case for non-sequential one inode bulkstat.
670  */
671 int					/* error status */
672 xfs_bulkstat_single(
673 	xfs_mount_t		*mp,	/* mount point for filesystem */
674 	xfs_ino_t		*lastinop, /* inode to return */
675 	char			__user *buffer, /* buffer with inode stats */
676 	int			*done)	/* 1 if there're more stats to get */
677 {
678 	int			count;	/* count value for bulkstat call */
679 	int			error;	/* return value */
680 	xfs_ino_t		ino;	/* filesystem inode number */
681 	int			res;	/* result from bs1 */
682 
683 	/*
684 	 * note that requesting valid inode numbers which are not allocated
685 	 * to inodes will most likely cause xfs_itobp to generate warning
686 	 * messages about bad magic numbers. This is ok. The fact that
687 	 * the inode isn't actually an inode is handled by the
688 	 * error check below. Done this way to make the usual case faster
689 	 * at the expense of the error case.
690 	 */
691 
692 	ino = (xfs_ino_t)*lastinop;
693 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
694 				 NULL, 0, NULL, NULL, &res);
695 	if (error) {
696 		/*
697 		 * Special case way failed, do it the "long" way
698 		 * to see if that works.
699 		 */
700 		(*lastinop)--;
701 		count = 1;
702 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
703 				NULL, sizeof(xfs_bstat_t), buffer,
704 				BULKSTAT_FG_IGET, done))
705 			return error;
706 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
707 			return error == EFSCORRUPTED ?
708 				XFS_ERROR(EINVAL) : error;
709 		else
710 			return 0;
711 	}
712 	*done = 0;
713 	return 0;
714 }
715 
716 /*
717  * Return inode number table for the filesystem.
718  */
719 int					/* error status */
720 xfs_inumbers(
721 	xfs_mount_t	*mp,		/* mount point for filesystem */
722 	xfs_ino_t	*lastino,	/* last inode returned */
723 	int		*count,		/* size of buffer/count returned */
724 	xfs_inogrp_t	__user *ubuffer)/* buffer with inode descriptions */
725 {
726 	xfs_buf_t	*agbp;
727 	xfs_agino_t	agino;
728 	xfs_agnumber_t	agno;
729 	int		bcount;
730 	xfs_inogrp_t	*buffer;
731 	int		bufidx;
732 	xfs_btree_cur_t	*cur;
733 	int		error;
734 	__int32_t	gcnt;
735 	xfs_inofree_t	gfree;
736 	xfs_agino_t	gino;
737 	int		i;
738 	xfs_ino_t	ino;
739 	int		left;
740 	int		tmp;
741 
742 	ino = (xfs_ino_t)*lastino;
743 	agno = XFS_INO_TO_AGNO(mp, ino);
744 	agino = XFS_INO_TO_AGINO(mp, ino);
745 	left = *count;
746 	*count = 0;
747 	bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
748 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
749 	error = bufidx = 0;
750 	cur = NULL;
751 	agbp = NULL;
752 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
753 		if (agbp == NULL) {
754 			down_read(&mp->m_peraglock);
755 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
756 			up_read(&mp->m_peraglock);
757 			if (error) {
758 				/*
759 				 * If we can't read the AGI of this ag,
760 				 * then just skip to the next one.
761 				 */
762 				ASSERT(cur == NULL);
763 				agbp = NULL;
764 				agno++;
765 				agino = 0;
766 				continue;
767 			}
768 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
769 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
770 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
771 			if (error) {
772 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
773 				cur = NULL;
774 				xfs_buf_relse(agbp);
775 				agbp = NULL;
776 				/*
777 				 * Move up the the last inode in the current
778 				 * chunk.  The lookup_ge will always get
779 				 * us the first inode in the next chunk.
780 				 */
781 				agino += XFS_INODES_PER_CHUNK - 1;
782 				continue;
783 			}
784 		}
785 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
786 			&i)) ||
787 		    i == 0) {
788 			xfs_buf_relse(agbp);
789 			agbp = NULL;
790 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
791 			cur = NULL;
792 			agno++;
793 			agino = 0;
794 			continue;
795 		}
796 		agino = gino + XFS_INODES_PER_CHUNK - 1;
797 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
798 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
799 		buffer[bufidx].xi_allocmask = ~gfree;
800 		bufidx++;
801 		left--;
802 		if (bufidx == bcount) {
803 			if (copy_to_user(ubuffer, buffer,
804 					bufidx * sizeof(*buffer))) {
805 				error = XFS_ERROR(EFAULT);
806 				break;
807 			}
808 			ubuffer += bufidx;
809 			*count += bufidx;
810 			bufidx = 0;
811 		}
812 		if (left) {
813 			error = xfs_inobt_increment(cur, 0, &tmp);
814 			if (error) {
815 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
816 				cur = NULL;
817 				xfs_buf_relse(agbp);
818 				agbp = NULL;
819 				/*
820 				 * The agino value has already been bumped.
821 				 * Just try to skip up to it.
822 				 */
823 				agino += XFS_INODES_PER_CHUNK;
824 				continue;
825 			}
826 		}
827 	}
828 	if (!error) {
829 		if (bufidx) {
830 			if (copy_to_user(ubuffer, buffer,
831 					bufidx * sizeof(*buffer)))
832 				error = XFS_ERROR(EFAULT);
833 			else
834 				*count += bufidx;
835 		}
836 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
837 	}
838 	kmem_free(buffer, bcount * sizeof(*buffer));
839 	if (cur)
840 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
841 					   XFS_BTREE_NOERROR));
842 	if (agbp)
843 		xfs_buf_relse(agbp);
844 	return error;
845 }
846