xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision 51ce16d519da0bc3c548e0facef7cb3aab1ac8cc)
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	*dic,		/* dinode inode pointer */
129 	xfs_bstat_t	*buf)		/* return buffer */
130 {
131 	/*
132 	 * The inode format changed when we moved the link count and
133 	 * made it 32 bits long.  If this is an old format inode,
134 	 * convert it in memory to look like a new one.  If it gets
135 	 * flushed to disk we will convert back before flushing or
136 	 * logging it.  We zero out the new projid field and the old link
137 	 * count field.  We'll handle clearing the pad field (the remains
138 	 * of the old uuid field) when we actually convert the inode to
139 	 * the new format. We don't change the version number so that we
140 	 * can distinguish this from a real new format inode.
141 	 */
142 	if (dic->di_version == 1) {
143 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
144 		buf->bs_projid = 0;
145 	} else {
146 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
147 		buf->bs_projid = be16_to_cpu(dic->di_projid);
148 	}
149 
150 	buf->bs_ino = ino;
151 	buf->bs_mode = be16_to_cpu(dic->di_mode);
152 	buf->bs_uid = be32_to_cpu(dic->di_uid);
153 	buf->bs_gid = be32_to_cpu(dic->di_gid);
154 	buf->bs_size = be64_to_cpu(dic->di_size);
155 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
156 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
157 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
158 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
159 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
160 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
161 	buf->bs_xflags = xfs_dic2xflags(dic);
162 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
163 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
164 	buf->bs_gen = be32_to_cpu(dic->di_gen);
165 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
166 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
167 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
168 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
169 
170 	switch (dic->di_format) {
171 	case XFS_DINODE_FMT_DEV:
172 		buf->bs_rdev = xfs_dinode_get_rdev(dic);
173 		buf->bs_blksize = BLKDEV_IOSIZE;
174 		buf->bs_blocks = 0;
175 		break;
176 	case XFS_DINODE_FMT_LOCAL:
177 	case XFS_DINODE_FMT_UUID:
178 		buf->bs_rdev = 0;
179 		buf->bs_blksize = mp->m_sb.sb_blocksize;
180 		buf->bs_blocks = 0;
181 		break;
182 	case XFS_DINODE_FMT_EXTENTS:
183 	case XFS_DINODE_FMT_BTREE:
184 		buf->bs_rdev = 0;
185 		buf->bs_blksize = mp->m_sb.sb_blocksize;
186 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
187 		break;
188 	}
189 }
190 
191 STATIC int
192 xfs_bulkstat_one_fmt(
193 	void			__user *ubuffer,
194 	const xfs_bstat_t	*buffer)
195 {
196 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
197 		return -EFAULT;
198 	return sizeof(*buffer);
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 	bulkstat_one_fmt_pf formatter = private_data ? : xfs_bulkstat_one_fmt;
221 
222 	dip = (xfs_dinode_t *)dibuff;
223 	*stat = BULKSTAT_RV_NOTHING;
224 
225 	if (!buffer || xfs_internal_inum(mp, ino))
226 		return XFS_ERROR(EINVAL);
227 	if (ubsize < sizeof(*buf))
228 		return XFS_ERROR(ENOMEM);
229 
230 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
231 
232 	if (dip == NULL) {
233 		/* We're not being passed a pointer to a dinode.  This happens
234 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
235 		 */
236 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
237 		if (error)
238 			goto out_free;
239 	} else {
240 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
241 	}
242 
243 	error = formatter(buffer, buf);
244 	if (error < 0)  {
245 		error = EFAULT;
246 		goto out_free;
247 	}
248 
249 	*stat = BULKSTAT_RV_DIDONE;
250 	if (ubused)
251 		*ubused = error;
252 
253  out_free:
254 	kmem_free(buf);
255 	return error;
256 }
257 
258 /*
259  * Test to see whether we can use the ondisk inode directly, based
260  * on the given bulkstat flags, filling in dipp accordingly.
261  * Returns zero if the inode is dodgey.
262  */
263 STATIC int
264 xfs_bulkstat_use_dinode(
265 	xfs_mount_t	*mp,
266 	int		flags,
267 	xfs_buf_t	*bp,
268 	int		clustidx,
269 	xfs_dinode_t	**dipp)
270 {
271 	xfs_dinode_t	*dip;
272 	unsigned int	aformat;
273 
274 	*dipp = NULL;
275 	if (!bp || (flags & BULKSTAT_FG_IGET))
276 		return 1;
277 	dip = (xfs_dinode_t *)
278 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
279 	/*
280 	 * Check the buffer containing the on-disk inode for di_mode == 0.
281 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
282 	 * inodes that have their in-core state initialized but not flushed
283 	 * to disk yet. This is a temporary hack that would require a proper
284 	 * fix in the future.
285 	 */
286 	if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
287 	    !XFS_DINODE_GOOD_VERSION(dip->di_version) ||
288 	    !dip->di_mode)
289 		return 0;
290 	if (flags & BULKSTAT_FG_QUICK) {
291 		*dipp = dip;
292 		return 1;
293 	}
294 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
295 	aformat = dip->di_aformat;
296 	if ((XFS_DFORK_Q(dip) == 0) ||
297 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
298 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_anextents)) {
299 		*dipp = dip;
300 		return 1;
301 	}
302 	return 1;
303 }
304 
305 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
306 
307 /*
308  * Return stat information in bulk (by-inode) for the filesystem.
309  */
310 int					/* error status */
311 xfs_bulkstat(
312 	xfs_mount_t		*mp,	/* mount point for filesystem */
313 	xfs_ino_t		*lastinop, /* last inode returned */
314 	int			*ubcountp, /* size of buffer/count returned */
315 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
316 	void			*private_data,/* private data for formatter */
317 	size_t			statstruct_size, /* sizeof struct filling */
318 	char			__user *ubuffer, /* buffer with inode stats */
319 	int			flags,	/* defined in xfs_itable.h */
320 	int			*done)	/* 1 if there are more stats to get */
321 {
322 	xfs_agblock_t		agbno=0;/* allocation group block number */
323 	xfs_buf_t		*agbp;	/* agi header buffer */
324 	xfs_agi_t		*agi;	/* agi header data */
325 	xfs_agino_t		agino;	/* inode # in allocation group */
326 	xfs_agnumber_t		agno;	/* allocation group number */
327 	xfs_daddr_t		bno;	/* inode cluster start daddr */
328 	int			chunkidx; /* current index into inode chunk */
329 	int			clustidx; /* current index into inode cluster */
330 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
331 	int			end_of_ag; /* set if we've seen the ag end */
332 	int			error;	/* error code */
333 	int                     fmterror;/* bulkstat formatter result */
334 	__int32_t		gcnt;	/* current btree rec's count */
335 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
336 	xfs_agino_t		gino;	/* current btree rec's start inode */
337 	int			i;	/* loop index */
338 	int			icount;	/* count of inodes good in irbuf */
339 	size_t			irbsize; /* size of irec buffer in bytes */
340 	xfs_ino_t		ino;	/* inode number (filesystem) */
341 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
342 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
343 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
344 	xfs_ino_t		lastino; /* last inode number returned */
345 	int			nbcluster; /* # of blocks in a cluster */
346 	int			nicluster; /* # of inodes in a cluster */
347 	int			nimask;	/* mask for inode clusters */
348 	int			nirbuf;	/* size of irbuf */
349 	int			rval;	/* return value error code */
350 	int			tmp;	/* result value from btree calls */
351 	int			ubcount; /* size of user's buffer */
352 	int			ubleft;	/* bytes left in user's buffer */
353 	char			__user *ubufp;	/* pointer into user's buffer */
354 	int			ubelem;	/* spaces used in user's buffer */
355 	int			ubused;	/* bytes used by formatter */
356 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
357 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
358 
359 	/*
360 	 * Get the last inode value, see if there's nothing to do.
361 	 */
362 	ino = (xfs_ino_t)*lastinop;
363 	lastino = ino;
364 	dip = NULL;
365 	agno = XFS_INO_TO_AGNO(mp, ino);
366 	agino = XFS_INO_TO_AGINO(mp, ino);
367 	if (agno >= mp->m_sb.sb_agcount ||
368 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
369 		*done = 1;
370 		*ubcountp = 0;
371 		return 0;
372 	}
373 	if (!ubcountp || *ubcountp <= 0) {
374 		return EINVAL;
375 	}
376 	ubcount = *ubcountp; /* statstruct's */
377 	ubleft = ubcount * statstruct_size; /* bytes */
378 	*ubcountp = ubelem = 0;
379 	*done = 0;
380 	fmterror = 0;
381 	ubufp = ubuffer;
382 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
383 		mp->m_sb.sb_inopblock :
384 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
385 	nimask = ~(nicluster - 1);
386 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
387 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
388 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
389 	nirbuf = irbsize / sizeof(*irbuf);
390 
391 	/*
392 	 * Loop over the allocation groups, starting from the last
393 	 * inode returned; 0 means start of the allocation group.
394 	 */
395 	rval = 0;
396 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
397 		cond_resched();
398 		bp = NULL;
399 		down_read(&mp->m_peraglock);
400 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
401 		up_read(&mp->m_peraglock);
402 		if (error) {
403 			/*
404 			 * Skip this allocation group and go to the next one.
405 			 */
406 			agno++;
407 			agino = 0;
408 			continue;
409 		}
410 		agi = XFS_BUF_TO_AGI(agbp);
411 		/*
412 		 * Allocate and initialize a btree cursor for ialloc btree.
413 		 */
414 		cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
415 		irbp = irbuf;
416 		irbufend = irbuf + nirbuf;
417 		end_of_ag = 0;
418 		/*
419 		 * If we're returning in the middle of an allocation group,
420 		 * we need to get the remainder of the chunk we're in.
421 		 */
422 		if (agino > 0) {
423 			/*
424 			 * Lookup the inode chunk that this inode lives in.
425 			 */
426 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
427 			if (!error &&	/* no I/O error */
428 			    tmp &&	/* lookup succeeded */
429 					/* got the record, should always work */
430 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
431 				    &gfree, &i)) &&
432 			    i == 1 &&
433 					/* this is the right chunk */
434 			    agino < gino + XFS_INODES_PER_CHUNK &&
435 					/* lastino was not last in chunk */
436 			    (chunkidx = agino - gino + 1) <
437 				    XFS_INODES_PER_CHUNK &&
438 					/* there are some left allocated */
439 			    XFS_INOBT_MASKN(chunkidx,
440 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
441 				/*
442 				 * Grab the chunk record.  Mark all the
443 				 * uninteresting inodes (because they're
444 				 * before our start point) free.
445 				 */
446 				for (i = 0; i < chunkidx; i++) {
447 					if (XFS_INOBT_MASK(i) & ~gfree)
448 						gcnt++;
449 				}
450 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
451 				irbp->ir_startino = gino;
452 				irbp->ir_freecount = gcnt;
453 				irbp->ir_free = gfree;
454 				irbp++;
455 				agino = gino + XFS_INODES_PER_CHUNK;
456 				icount = XFS_INODES_PER_CHUNK - gcnt;
457 			} else {
458 				/*
459 				 * If any of those tests failed, bump the
460 				 * inode number (just in case).
461 				 */
462 				agino++;
463 				icount = 0;
464 			}
465 			/*
466 			 * In any case, increment to the next record.
467 			 */
468 			if (!error)
469 				error = xfs_btree_increment(cur, 0, &tmp);
470 		} else {
471 			/*
472 			 * Start of ag.  Lookup the first inode chunk.
473 			 */
474 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
475 			icount = 0;
476 		}
477 		/*
478 		 * Loop through inode btree records in this ag,
479 		 * until we run out of inodes or space in the buffer.
480 		 */
481 		while (irbp < irbufend && icount < ubcount) {
482 			/*
483 			 * Loop as long as we're unable to read the
484 			 * inode btree.
485 			 */
486 			while (error) {
487 				agino += XFS_INODES_PER_CHUNK;
488 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
489 						be32_to_cpu(agi->agi_length))
490 					break;
491 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
492 							    &tmp);
493 				cond_resched();
494 			}
495 			/*
496 			 * If ran off the end of the ag either with an error,
497 			 * or the normal way, set end and stop collecting.
498 			 */
499 			if (error ||
500 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
501 				    &gfree, &i)) ||
502 			    i == 0) {
503 				end_of_ag = 1;
504 				break;
505 			}
506 			/*
507 			 * If this chunk has any allocated inodes, save it.
508 			 * Also start read-ahead now for this chunk.
509 			 */
510 			if (gcnt < XFS_INODES_PER_CHUNK) {
511 				/*
512 				 * Loop over all clusters in the next chunk.
513 				 * Do a readahead if there are any allocated
514 				 * inodes in that cluster.
515 				 */
516 				for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
517 				     chunkidx = 0;
518 				     chunkidx < XFS_INODES_PER_CHUNK;
519 				     chunkidx += nicluster,
520 				     agbno += nbcluster) {
521 					if (XFS_INOBT_MASKN(chunkidx,
522 							    nicluster) & ~gfree)
523 						xfs_btree_reada_bufs(mp, agno,
524 							agbno, nbcluster);
525 				}
526 				irbp->ir_startino = gino;
527 				irbp->ir_freecount = gcnt;
528 				irbp->ir_free = gfree;
529 				irbp++;
530 				icount += XFS_INODES_PER_CHUNK - gcnt;
531 			}
532 			/*
533 			 * Set agino to after this chunk and bump the cursor.
534 			 */
535 			agino = gino + XFS_INODES_PER_CHUNK;
536 			error = xfs_btree_increment(cur, 0, &tmp);
537 			cond_resched();
538 		}
539 		/*
540 		 * Drop the btree buffers and the agi buffer.
541 		 * We can't hold any of the locks these represent
542 		 * when calling iget.
543 		 */
544 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
545 		xfs_buf_relse(agbp);
546 		/*
547 		 * Now format all the good inodes into the user's buffer.
548 		 */
549 		irbufend = irbp;
550 		for (irbp = irbuf;
551 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
552 			/*
553 			 * Now process this chunk of inodes.
554 			 */
555 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
556 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
557 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
558 			     chunkidx++, clustidx++, agino++) {
559 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
560 				/*
561 				 * Recompute agbno if this is the
562 				 * first inode of the cluster.
563 				 *
564 				 * Careful with clustidx.   There can be
565 				 * multple clusters per chunk, a single
566 				 * cluster per chunk or a cluster that has
567 				 * inodes represented from several different
568 				 * chunks (if blocksize is large).
569 				 *
570 				 * Because of this, the starting clustidx is
571 				 * initialized to zero in this loop but must
572 				 * later be reset after reading in the cluster
573 				 * buffer.
574 				 */
575 				if ((chunkidx & (nicluster - 1)) == 0) {
576 					agbno = XFS_AGINO_TO_AGBNO(mp,
577 							irbp->ir_startino) +
578 						((chunkidx & nimask) >>
579 						 mp->m_sb.sb_inopblog);
580 
581 					if (flags & (BULKSTAT_FG_QUICK |
582 						     BULKSTAT_FG_INLINE)) {
583 						int offset;
584 
585 						ino = XFS_AGINO_TO_INO(mp, agno,
586 								       agino);
587 						bno = XFS_AGB_TO_DADDR(mp, agno,
588 								       agbno);
589 
590 						/*
591 						 * Get the inode cluster buffer
592 						 */
593 						if (bp)
594 							xfs_buf_relse(bp);
595 
596 						error = xfs_inotobp(mp, NULL, ino, &dip,
597 								    &bp, &offset,
598 								    XFS_IMAP_BULKSTAT);
599 
600 						if (!error)
601 							clustidx = offset / mp->m_sb.sb_inodesize;
602 						if (XFS_TEST_ERROR(error != 0,
603 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
604 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
605 							bp = NULL;
606 							ubleft = 0;
607 							rval = error;
608 							break;
609 						}
610 					}
611 				}
612 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
613 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
614 				/*
615 				 * Skip if this inode is free.
616 				 */
617 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
618 					lastino = ino;
619 					continue;
620 				}
621 				/*
622 				 * Count used inodes as free so we can tell
623 				 * when the chunk is used up.
624 				 */
625 				irbp->ir_freecount++;
626 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
627 							     clustidx, &dip)) {
628 					lastino = ino;
629 					continue;
630 				}
631 				/*
632 				 * If we need to do an iget, cannot hold bp.
633 				 * Drop it, until starting the next cluster.
634 				 */
635 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
636 					if (bp)
637 						xfs_buf_relse(bp);
638 					bp = NULL;
639 				}
640 
641 				/*
642 				 * Get the inode and fill in a single buffer.
643 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
644 				 * BULKSTAT_FG_IGET uses igets.
645 				 * BULKSTAT_FG_INLINE uses dip if we have an
646 				 * inline attr fork, else igets.
647 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
648 				 * This is also used to count inodes/blks, etc
649 				 * in xfs_qm_quotacheck.
650 				 */
651 				ubused = statstruct_size;
652 				error = formatter(mp, ino, ubufp,
653 						ubleft, private_data,
654 						bno, &ubused, dip, &fmterror);
655 				if (fmterror == BULKSTAT_RV_NOTHING) {
656 					if (error && error != ENOENT &&
657 						error != EINVAL) {
658 						ubleft = 0;
659 						rval = error;
660 						break;
661 					}
662 					lastino = ino;
663 					continue;
664 				}
665 				if (fmterror == BULKSTAT_RV_GIVEUP) {
666 					ubleft = 0;
667 					ASSERT(error);
668 					rval = error;
669 					break;
670 				}
671 				if (ubufp)
672 					ubufp += ubused;
673 				ubleft -= ubused;
674 				ubelem++;
675 				lastino = ino;
676 			}
677 
678 			cond_resched();
679 		}
680 
681 		if (bp)
682 			xfs_buf_relse(bp);
683 
684 		/*
685 		 * Set up for the next loop iteration.
686 		 */
687 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
688 			if (end_of_ag) {
689 				agno++;
690 				agino = 0;
691 			} else
692 				agino = XFS_INO_TO_AGINO(mp, lastino);
693 		} else
694 			break;
695 	}
696 	/*
697 	 * Done, we're either out of filesystem or space to put the data.
698 	 */
699 	kmem_free(irbuf);
700 	*ubcountp = ubelem;
701 	/*
702 	 * Found some inodes, return them now and return the error next time.
703 	 */
704 	if (ubelem)
705 		rval = 0;
706 	if (agno >= mp->m_sb.sb_agcount) {
707 		/*
708 		 * If we ran out of filesystem, mark lastino as off
709 		 * the end of the filesystem, so the next call
710 		 * will return immediately.
711 		 */
712 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
713 		*done = 1;
714 	} else
715 		*lastinop = (xfs_ino_t)lastino;
716 
717 	return rval;
718 }
719 
720 /*
721  * Return stat information in bulk (by-inode) for the filesystem.
722  * Special case for non-sequential one inode bulkstat.
723  */
724 int					/* error status */
725 xfs_bulkstat_single(
726 	xfs_mount_t		*mp,	/* mount point for filesystem */
727 	xfs_ino_t		*lastinop, /* inode to return */
728 	char			__user *buffer, /* buffer with inode stats */
729 	int			*done)	/* 1 if there are more stats to get */
730 {
731 	int			count;	/* count value for bulkstat call */
732 	int			error;	/* return value */
733 	xfs_ino_t		ino;	/* filesystem inode number */
734 	int			res;	/* result from bs1 */
735 
736 	/*
737 	 * note that requesting valid inode numbers which are not allocated
738 	 * to inodes will most likely cause xfs_itobp to generate warning
739 	 * messages about bad magic numbers. This is ok. The fact that
740 	 * the inode isn't actually an inode is handled by the
741 	 * error check below. Done this way to make the usual case faster
742 	 * at the expense of the error case.
743 	 */
744 
745 	ino = (xfs_ino_t)*lastinop;
746 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
747 				 NULL, 0, NULL, NULL, &res);
748 	if (error) {
749 		/*
750 		 * Special case way failed, do it the "long" way
751 		 * to see if that works.
752 		 */
753 		(*lastinop)--;
754 		count = 1;
755 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
756 				NULL, sizeof(xfs_bstat_t), buffer,
757 				BULKSTAT_FG_IGET, done))
758 			return error;
759 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
760 			return error == EFSCORRUPTED ?
761 				XFS_ERROR(EINVAL) : error;
762 		else
763 			return 0;
764 	}
765 	*done = 0;
766 	return 0;
767 }
768 
769 int
770 xfs_inumbers_fmt(
771 	void			__user *ubuffer, /* buffer to write to */
772 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
773 	long			count,		/* # of elements to read */
774 	long			*written)	/* # of bytes written */
775 {
776 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
777 		return -EFAULT;
778 	*written = count * sizeof(*buffer);
779 	return 0;
780 }
781 
782 /*
783  * Return inode number table for the filesystem.
784  */
785 int					/* error status */
786 xfs_inumbers(
787 	xfs_mount_t	*mp,		/* mount point for filesystem */
788 	xfs_ino_t	*lastino,	/* last inode returned */
789 	int		*count,		/* size of buffer/count returned */
790 	void		__user *ubuffer,/* buffer with inode descriptions */
791 	inumbers_fmt_pf	formatter)
792 {
793 	xfs_buf_t	*agbp;
794 	xfs_agino_t	agino;
795 	xfs_agnumber_t	agno;
796 	int		bcount;
797 	xfs_inogrp_t	*buffer;
798 	int		bufidx;
799 	xfs_btree_cur_t	*cur;
800 	int		error;
801 	__int32_t	gcnt;
802 	xfs_inofree_t	gfree;
803 	xfs_agino_t	gino;
804 	int		i;
805 	xfs_ino_t	ino;
806 	int		left;
807 	int		tmp;
808 
809 	ino = (xfs_ino_t)*lastino;
810 	agno = XFS_INO_TO_AGNO(mp, ino);
811 	agino = XFS_INO_TO_AGINO(mp, ino);
812 	left = *count;
813 	*count = 0;
814 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
815 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
816 	error = bufidx = 0;
817 	cur = NULL;
818 	agbp = NULL;
819 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
820 		if (agbp == NULL) {
821 			down_read(&mp->m_peraglock);
822 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
823 			up_read(&mp->m_peraglock);
824 			if (error) {
825 				/*
826 				 * If we can't read the AGI of this ag,
827 				 * then just skip to the next one.
828 				 */
829 				ASSERT(cur == NULL);
830 				agbp = NULL;
831 				agno++;
832 				agino = 0;
833 				continue;
834 			}
835 			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
836 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
837 			if (error) {
838 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
839 				cur = NULL;
840 				xfs_buf_relse(agbp);
841 				agbp = NULL;
842 				/*
843 				 * Move up the last inode in the current
844 				 * chunk.  The lookup_ge will always get
845 				 * us the first inode in the next chunk.
846 				 */
847 				agino += XFS_INODES_PER_CHUNK - 1;
848 				continue;
849 			}
850 		}
851 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
852 			&i)) ||
853 		    i == 0) {
854 			xfs_buf_relse(agbp);
855 			agbp = NULL;
856 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
857 			cur = NULL;
858 			agno++;
859 			agino = 0;
860 			continue;
861 		}
862 		agino = gino + XFS_INODES_PER_CHUNK - 1;
863 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
864 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
865 		buffer[bufidx].xi_allocmask = ~gfree;
866 		bufidx++;
867 		left--;
868 		if (bufidx == bcount) {
869 			long written;
870 			if (formatter(ubuffer, buffer, bufidx, &written)) {
871 				error = XFS_ERROR(EFAULT);
872 				break;
873 			}
874 			ubuffer += written;
875 			*count += bufidx;
876 			bufidx = 0;
877 		}
878 		if (left) {
879 			error = xfs_btree_increment(cur, 0, &tmp);
880 			if (error) {
881 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
882 				cur = NULL;
883 				xfs_buf_relse(agbp);
884 				agbp = NULL;
885 				/*
886 				 * The agino value has already been bumped.
887 				 * Just try to skip up to it.
888 				 */
889 				agino += XFS_INODES_PER_CHUNK;
890 				continue;
891 			}
892 		}
893 	}
894 	if (!error) {
895 		if (bufidx) {
896 			long written;
897 			if (formatter(ubuffer, buffer, bufidx, &written))
898 				error = XFS_ERROR(EFAULT);
899 			else
900 				*count += bufidx;
901 		}
902 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
903 	}
904 	kmem_free(buffer);
905 	if (cur)
906 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
907 					   XFS_BTREE_NOERROR));
908 	if (agbp)
909 		xfs_buf_relse(agbp);
910 	return error;
911 }
912