xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision f42b3800)
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 	bhv_vnode_t	*vp;
63 	int		error;
64 
65 	error = xfs_iget(mp, NULL, ino,
66 			 XFS_IGET_BULKSTAT, XFS_ILOCK_SHARED, &ip, bno);
67 	if (error) {
68 		*stat = BULKSTAT_RV_NOTHING;
69 		return error;
70 	}
71 
72 	ASSERT(ip != NULL);
73 	ASSERT(ip->i_blkno != (xfs_daddr_t)0);
74 	if (ip->i_d.di_mode == 0) {
75 		*stat = BULKSTAT_RV_NOTHING;
76 		error = XFS_ERROR(ENOENT);
77 		goto out_iput;
78 	}
79 
80 	vp = XFS_ITOV(ip);
81 	dic = &ip->i_d;
82 
83 	/* xfs_iget returns the following without needing
84 	 * further change.
85 	 */
86 	buf->bs_nlink = dic->di_nlink;
87 	buf->bs_projid = dic->di_projid;
88 	buf->bs_ino = ino;
89 	buf->bs_mode = dic->di_mode;
90 	buf->bs_uid = dic->di_uid;
91 	buf->bs_gid = dic->di_gid;
92 	buf->bs_size = dic->di_size;
93 	vn_atime_to_bstime(vp, &buf->bs_atime);
94 	buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
95 	buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
96 	buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
97 	buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
98 	buf->bs_xflags = xfs_ip2xflags(ip);
99 	buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
100 	buf->bs_extents = dic->di_nextents;
101 	buf->bs_gen = dic->di_gen;
102 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
103 	buf->bs_dmevmask = dic->di_dmevmask;
104 	buf->bs_dmstate = dic->di_dmstate;
105 	buf->bs_aextents = dic->di_anextents;
106 
107 	switch (dic->di_format) {
108 	case XFS_DINODE_FMT_DEV:
109 		buf->bs_rdev = ip->i_df.if_u2.if_rdev;
110 		buf->bs_blksize = BLKDEV_IOSIZE;
111 		buf->bs_blocks = 0;
112 		break;
113 	case XFS_DINODE_FMT_LOCAL:
114 	case XFS_DINODE_FMT_UUID:
115 		buf->bs_rdev = 0;
116 		buf->bs_blksize = mp->m_sb.sb_blocksize;
117 		buf->bs_blocks = 0;
118 		break;
119 	case XFS_DINODE_FMT_EXTENTS:
120 	case XFS_DINODE_FMT_BTREE:
121 		buf->bs_rdev = 0;
122 		buf->bs_blksize = mp->m_sb.sb_blocksize;
123 		buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
124 		break;
125 	}
126 
127  out_iput:
128 	xfs_iput(ip, XFS_ILOCK_SHARED);
129 	return error;
130 }
131 
132 STATIC void
133 xfs_bulkstat_one_dinode(
134 	xfs_mount_t	*mp,		/* mount point for filesystem */
135 	xfs_ino_t	ino,		/* inode number to get data for */
136 	xfs_dinode_t	*dip,		/* dinode inode pointer */
137 	xfs_bstat_t	*buf)		/* return buffer */
138 {
139 	xfs_dinode_core_t *dic;		/* dinode core info pointer */
140 
141 	dic = &dip->di_core;
142 
143 	/*
144 	 * The inode format changed when we moved the link count and
145 	 * made it 32 bits long.  If this is an old format inode,
146 	 * convert it in memory to look like a new one.  If it gets
147 	 * flushed to disk we will convert back before flushing or
148 	 * logging it.  We zero out the new projid field and the old link
149 	 * count field.  We'll handle clearing the pad field (the remains
150 	 * of the old uuid field) when we actually convert the inode to
151 	 * the new format. We don't change the version number so that we
152 	 * can distinguish this from a real new format inode.
153 	 */
154 	if (dic->di_version == XFS_DINODE_VERSION_1) {
155 		buf->bs_nlink = be16_to_cpu(dic->di_onlink);
156 		buf->bs_projid = 0;
157 	} else {
158 		buf->bs_nlink = be32_to_cpu(dic->di_nlink);
159 		buf->bs_projid = be16_to_cpu(dic->di_projid);
160 	}
161 
162 	buf->bs_ino = ino;
163 	buf->bs_mode = be16_to_cpu(dic->di_mode);
164 	buf->bs_uid = be32_to_cpu(dic->di_uid);
165 	buf->bs_gid = be32_to_cpu(dic->di_gid);
166 	buf->bs_size = be64_to_cpu(dic->di_size);
167 	buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
168 	buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
169 	buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
170 	buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
171 	buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
172 	buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
173 	buf->bs_xflags = xfs_dic2xflags(dip);
174 	buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
175 	buf->bs_extents = be32_to_cpu(dic->di_nextents);
176 	buf->bs_gen = be32_to_cpu(dic->di_gen);
177 	memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
178 	buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
179 	buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
180 	buf->bs_aextents = be16_to_cpu(dic->di_anextents);
181 
182 	switch (dic->di_format) {
183 	case XFS_DINODE_FMT_DEV:
184 		buf->bs_rdev = be32_to_cpu(dip->di_u.di_dev);
185 		buf->bs_blksize = BLKDEV_IOSIZE;
186 		buf->bs_blocks = 0;
187 		break;
188 	case XFS_DINODE_FMT_LOCAL:
189 	case XFS_DINODE_FMT_UUID:
190 		buf->bs_rdev = 0;
191 		buf->bs_blksize = mp->m_sb.sb_blocksize;
192 		buf->bs_blocks = 0;
193 		break;
194 	case XFS_DINODE_FMT_EXTENTS:
195 	case XFS_DINODE_FMT_BTREE:
196 		buf->bs_rdev = 0;
197 		buf->bs_blksize = mp->m_sb.sb_blocksize;
198 		buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
199 		break;
200 	}
201 }
202 
203 STATIC int
204 xfs_bulkstat_one_fmt(
205 	void			__user *ubuffer,
206 	const xfs_bstat_t	*buffer)
207 {
208 	if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
209 		return -EFAULT;
210 	return sizeof(*buffer);
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(
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 	void		*private_data,	/* my private data */
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 	bulkstat_one_fmt_pf formatter = private_data ? : xfs_bulkstat_one_fmt;
233 
234 	dip = (xfs_dinode_t *)dibuff;
235 	*stat = BULKSTAT_RV_NOTHING;
236 
237 	if (!buffer || xfs_internal_inum(mp, ino))
238 		return XFS_ERROR(EINVAL);
239 	if (ubsize < sizeof(*buf))
240 		return XFS_ERROR(ENOMEM);
241 
242 	buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
243 
244 	if (dip == NULL) {
245 		/* We're not being passed a pointer to a dinode.  This happens
246 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
247 		 */
248 		error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
249 		if (error)
250 			goto out_free;
251 	} else {
252 		xfs_bulkstat_one_dinode(mp, ino, dip, buf);
253 	}
254 
255 	error = formatter(buffer, buf);
256 	if (error < 0)  {
257 		error = EFAULT;
258 		goto out_free;
259 	}
260 
261 	*stat = BULKSTAT_RV_DIDONE;
262 	if (ubused)
263 		*ubused = error;
264 
265  out_free:
266 	kmem_free(buf, sizeof(*buf));
267 	return error;
268 }
269 
270 /*
271  * Test to see whether we can use the ondisk inode directly, based
272  * on the given bulkstat flags, filling in dipp accordingly.
273  * Returns zero if the inode is dodgey.
274  */
275 STATIC int
276 xfs_bulkstat_use_dinode(
277 	xfs_mount_t	*mp,
278 	int		flags,
279 	xfs_buf_t	*bp,
280 	int		clustidx,
281 	xfs_dinode_t	**dipp)
282 {
283 	xfs_dinode_t	*dip;
284 	unsigned int	aformat;
285 
286 	*dipp = NULL;
287 	if (!bp || (flags & BULKSTAT_FG_IGET))
288 		return 1;
289 	dip = (xfs_dinode_t *)
290 			xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
291 	/*
292 	 * Check the buffer containing the on-disk inode for di_mode == 0.
293 	 * This is to prevent xfs_bulkstat from picking up just reclaimed
294 	 * inodes that have their in-core state initialized but not flushed
295 	 * to disk yet. This is a temporary hack that would require a proper
296 	 * fix in the future.
297 	 */
298 	if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC ||
299 	    !XFS_DINODE_GOOD_VERSION(dip->di_core.di_version) ||
300 	    !dip->di_core.di_mode)
301 		return 0;
302 	if (flags & BULKSTAT_FG_QUICK) {
303 		*dipp = dip;
304 		return 1;
305 	}
306 	/* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
307 	aformat = dip->di_core.di_aformat;
308 	if ((XFS_DFORK_Q(dip) == 0) ||
309 	    (aformat == XFS_DINODE_FMT_LOCAL) ||
310 	    (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
311 		*dipp = dip;
312 		return 1;
313 	}
314 	return 1;
315 }
316 
317 #define XFS_BULKSTAT_UBLEFT(ubleft)	((ubleft) >= statstruct_size)
318 
319 /*
320  * Return stat information in bulk (by-inode) for the filesystem.
321  */
322 int					/* error status */
323 xfs_bulkstat(
324 	xfs_mount_t		*mp,	/* mount point for filesystem */
325 	xfs_ino_t		*lastinop, /* last inode returned */
326 	int			*ubcountp, /* size of buffer/count returned */
327 	bulkstat_one_pf		formatter, /* func that'd fill a single buf */
328 	void			*private_data,/* private data for formatter */
329 	size_t			statstruct_size, /* sizeof struct filling */
330 	char			__user *ubuffer, /* buffer with inode stats */
331 	int			flags,	/* defined in xfs_itable.h */
332 	int			*done)	/* 1 if there are more stats to get */
333 {
334 	xfs_agblock_t		agbno=0;/* allocation group block number */
335 	xfs_buf_t		*agbp;	/* agi header buffer */
336 	xfs_agi_t		*agi;	/* agi header data */
337 	xfs_agino_t		agino;	/* inode # in allocation group */
338 	xfs_agnumber_t		agno;	/* allocation group number */
339 	xfs_daddr_t		bno;	/* inode cluster start daddr */
340 	int			chunkidx; /* current index into inode chunk */
341 	int			clustidx; /* current index into inode cluster */
342 	xfs_btree_cur_t		*cur;	/* btree cursor for ialloc btree */
343 	int			end_of_ag; /* set if we've seen the ag end */
344 	int			error;	/* error code */
345 	int                     fmterror;/* bulkstat formatter result */
346 	__int32_t		gcnt;	/* current btree rec's count */
347 	xfs_inofree_t		gfree;	/* current btree rec's free mask */
348 	xfs_agino_t		gino;	/* current btree rec's start inode */
349 	int			i;	/* loop index */
350 	int			icount;	/* count of inodes good in irbuf */
351 	size_t			irbsize; /* size of irec buffer in bytes */
352 	xfs_ino_t		ino;	/* inode number (filesystem) */
353 	xfs_inobt_rec_incore_t	*irbp;	/* current irec buffer pointer */
354 	xfs_inobt_rec_incore_t	*irbuf;	/* start of irec buffer */
355 	xfs_inobt_rec_incore_t	*irbufend; /* end of good irec buffer entries */
356 	xfs_ino_t		lastino; /* last inode number returned */
357 	int			nbcluster; /* # of blocks in a cluster */
358 	int			nicluster; /* # of inodes in a cluster */
359 	int			nimask;	/* mask for inode clusters */
360 	int			nirbuf;	/* size of irbuf */
361 	int			rval;	/* return value error code */
362 	int			tmp;	/* result value from btree calls */
363 	int			ubcount; /* size of user's buffer */
364 	int			ubleft;	/* bytes left in user's buffer */
365 	char			__user *ubufp;	/* pointer into user's buffer */
366 	int			ubelem;	/* spaces used in user's buffer */
367 	int			ubused;	/* bytes used by formatter */
368 	xfs_buf_t		*bp;	/* ptr to on-disk inode cluster buf */
369 	xfs_dinode_t		*dip;	/* ptr into bp for specific inode */
370 	xfs_inode_t		*ip;	/* ptr to in-core inode struct */
371 
372 	/*
373 	 * Get the last inode value, see if there's nothing to do.
374 	 */
375 	ino = (xfs_ino_t)*lastinop;
376 	lastino = ino;
377 	dip = NULL;
378 	agno = XFS_INO_TO_AGNO(mp, ino);
379 	agino = XFS_INO_TO_AGINO(mp, ino);
380 	if (agno >= mp->m_sb.sb_agcount ||
381 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
382 		*done = 1;
383 		*ubcountp = 0;
384 		return 0;
385 	}
386 	if (!ubcountp || *ubcountp <= 0) {
387 		return EINVAL;
388 	}
389 	ubcount = *ubcountp; /* statstruct's */
390 	ubleft = ubcount * statstruct_size; /* bytes */
391 	*ubcountp = ubelem = 0;
392 	*done = 0;
393 	fmterror = 0;
394 	ubufp = ubuffer;
395 	nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
396 		mp->m_sb.sb_inopblock :
397 		(XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
398 	nimask = ~(nicluster - 1);
399 	nbcluster = nicluster >> mp->m_sb.sb_inopblog;
400 	irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
401 				   KM_SLEEP | KM_MAYFAIL | KM_LARGE);
402 	nirbuf = irbsize / sizeof(*irbuf);
403 
404 	/*
405 	 * Loop over the allocation groups, starting from the last
406 	 * inode returned; 0 means start of the allocation group.
407 	 */
408 	rval = 0;
409 	while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
410 		cond_resched();
411 		bp = NULL;
412 		down_read(&mp->m_peraglock);
413 		error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
414 		up_read(&mp->m_peraglock);
415 		if (error) {
416 			/*
417 			 * Skip this allocation group and go to the next one.
418 			 */
419 			agno++;
420 			agino = 0;
421 			continue;
422 		}
423 		agi = XFS_BUF_TO_AGI(agbp);
424 		/*
425 		 * Allocate and initialize a btree cursor for ialloc btree.
426 		 */
427 		cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
428 						(xfs_inode_t *)0, 0);
429 		irbp = irbuf;
430 		irbufend = irbuf + nirbuf;
431 		end_of_ag = 0;
432 		/*
433 		 * If we're returning in the middle of an allocation group,
434 		 * we need to get the remainder of the chunk we're in.
435 		 */
436 		if (agino > 0) {
437 			/*
438 			 * Lookup the inode chunk that this inode lives in.
439 			 */
440 			error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
441 			if (!error &&	/* no I/O error */
442 			    tmp &&	/* lookup succeeded */
443 					/* got the record, should always work */
444 			    !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
445 				    &gfree, &i)) &&
446 			    i == 1 &&
447 					/* this is the right chunk */
448 			    agino < gino + XFS_INODES_PER_CHUNK &&
449 					/* lastino was not last in chunk */
450 			    (chunkidx = agino - gino + 1) <
451 				    XFS_INODES_PER_CHUNK &&
452 					/* there are some left allocated */
453 			    XFS_INOBT_MASKN(chunkidx,
454 				    XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
455 				/*
456 				 * Grab the chunk record.  Mark all the
457 				 * uninteresting inodes (because they're
458 				 * before our start point) free.
459 				 */
460 				for (i = 0; i < chunkidx; i++) {
461 					if (XFS_INOBT_MASK(i) & ~gfree)
462 						gcnt++;
463 				}
464 				gfree |= XFS_INOBT_MASKN(0, chunkidx);
465 				irbp->ir_startino = gino;
466 				irbp->ir_freecount = gcnt;
467 				irbp->ir_free = gfree;
468 				irbp++;
469 				agino = gino + XFS_INODES_PER_CHUNK;
470 				icount = XFS_INODES_PER_CHUNK - gcnt;
471 			} else {
472 				/*
473 				 * If any of those tests failed, bump the
474 				 * inode number (just in case).
475 				 */
476 				agino++;
477 				icount = 0;
478 			}
479 			/*
480 			 * In any case, increment to the next record.
481 			 */
482 			if (!error)
483 				error = xfs_inobt_increment(cur, 0, &tmp);
484 		} else {
485 			/*
486 			 * Start of ag.  Lookup the first inode chunk.
487 			 */
488 			error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
489 			icount = 0;
490 		}
491 		/*
492 		 * Loop through inode btree records in this ag,
493 		 * until we run out of inodes or space in the buffer.
494 		 */
495 		while (irbp < irbufend && icount < ubcount) {
496 			/*
497 			 * Loop as long as we're unable to read the
498 			 * inode btree.
499 			 */
500 			while (error) {
501 				agino += XFS_INODES_PER_CHUNK;
502 				if (XFS_AGINO_TO_AGBNO(mp, agino) >=
503 						be32_to_cpu(agi->agi_length))
504 					break;
505 				error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
506 							    &tmp);
507 				cond_resched();
508 			}
509 			/*
510 			 * If ran off the end of the ag either with an error,
511 			 * or the normal way, set end and stop collecting.
512 			 */
513 			if (error ||
514 			    (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
515 				    &gfree, &i)) ||
516 			    i == 0) {
517 				end_of_ag = 1;
518 				break;
519 			}
520 			/*
521 			 * If this chunk has any allocated inodes, save it.
522 			 * Also start read-ahead now for this chunk.
523 			 */
524 			if (gcnt < XFS_INODES_PER_CHUNK) {
525 				/*
526 				 * Loop over all clusters in the next chunk.
527 				 * Do a readahead if there are any allocated
528 				 * inodes in that cluster.
529 				 */
530 				for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
531 				     chunkidx = 0;
532 				     chunkidx < XFS_INODES_PER_CHUNK;
533 				     chunkidx += nicluster,
534 				     agbno += nbcluster) {
535 					if (XFS_INOBT_MASKN(chunkidx,
536 							    nicluster) & ~gfree)
537 						xfs_btree_reada_bufs(mp, agno,
538 							agbno, nbcluster);
539 				}
540 				irbp->ir_startino = gino;
541 				irbp->ir_freecount = gcnt;
542 				irbp->ir_free = gfree;
543 				irbp++;
544 				icount += XFS_INODES_PER_CHUNK - gcnt;
545 			}
546 			/*
547 			 * Set agino to after this chunk and bump the cursor.
548 			 */
549 			agino = gino + XFS_INODES_PER_CHUNK;
550 			error = xfs_inobt_increment(cur, 0, &tmp);
551 			cond_resched();
552 		}
553 		/*
554 		 * Drop the btree buffers and the agi buffer.
555 		 * We can't hold any of the locks these represent
556 		 * when calling iget.
557 		 */
558 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
559 		xfs_buf_relse(agbp);
560 		/*
561 		 * Now format all the good inodes into the user's buffer.
562 		 */
563 		irbufend = irbp;
564 		for (irbp = irbuf;
565 		     irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
566 			/*
567 			 * Now process this chunk of inodes.
568 			 */
569 			for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
570 			     XFS_BULKSTAT_UBLEFT(ubleft) &&
571 				irbp->ir_freecount < XFS_INODES_PER_CHUNK;
572 			     chunkidx++, clustidx++, agino++) {
573 				ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
574 				/*
575 				 * Recompute agbno if this is the
576 				 * first inode of the cluster.
577 				 *
578 				 * Careful with clustidx.   There can be
579 				 * multple clusters per chunk, a single
580 				 * cluster per chunk or a cluster that has
581 				 * inodes represented from several different
582 				 * chunks (if blocksize is large).
583 				 *
584 				 * Because of this, the starting clustidx is
585 				 * initialized to zero in this loop but must
586 				 * later be reset after reading in the cluster
587 				 * buffer.
588 				 */
589 				if ((chunkidx & (nicluster - 1)) == 0) {
590 					agbno = XFS_AGINO_TO_AGBNO(mp,
591 							irbp->ir_startino) +
592 						((chunkidx & nimask) >>
593 						 mp->m_sb.sb_inopblog);
594 
595 					if (flags & (BULKSTAT_FG_QUICK |
596 						     BULKSTAT_FG_INLINE)) {
597 						ino = XFS_AGINO_TO_INO(mp, agno,
598 								       agino);
599 						bno = XFS_AGB_TO_DADDR(mp, agno,
600 								       agbno);
601 
602 						/*
603 						 * Get the inode cluster buffer
604 						 */
605 						ASSERT(xfs_inode_zone != NULL);
606 						ip = kmem_zone_zalloc(xfs_inode_zone,
607 								      KM_SLEEP);
608 						ip->i_ino = ino;
609 						ip->i_mount = mp;
610 						spin_lock_init(&ip->i_flags_lock);
611 						if (bp)
612 							xfs_buf_relse(bp);
613 						error = xfs_itobp(mp, NULL, ip,
614 								&dip, &bp, bno,
615 								XFS_IMAP_BULKSTAT,
616 								XFS_BUF_LOCK);
617 						if (!error)
618 							clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
619 						kmem_zone_free(xfs_inode_zone, ip);
620 						if (XFS_TEST_ERROR(error != 0,
621 								   mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
622 								   XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
623 							bp = NULL;
624 							ubleft = 0;
625 							rval = error;
626 							break;
627 						}
628 					}
629 				}
630 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
631 				bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
632 				/*
633 				 * Skip if this inode is free.
634 				 */
635 				if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
636 					lastino = ino;
637 					continue;
638 				}
639 				/*
640 				 * Count used inodes as free so we can tell
641 				 * when the chunk is used up.
642 				 */
643 				irbp->ir_freecount++;
644 				if (!xfs_bulkstat_use_dinode(mp, flags, bp,
645 							     clustidx, &dip)) {
646 					lastino = ino;
647 					continue;
648 				}
649 				/*
650 				 * If we need to do an iget, cannot hold bp.
651 				 * Drop it, until starting the next cluster.
652 				 */
653 				if ((flags & BULKSTAT_FG_INLINE) && !dip) {
654 					if (bp)
655 						xfs_buf_relse(bp);
656 					bp = NULL;
657 				}
658 
659 				/*
660 				 * Get the inode and fill in a single buffer.
661 				 * BULKSTAT_FG_QUICK uses dip to fill it in.
662 				 * BULKSTAT_FG_IGET uses igets.
663 				 * BULKSTAT_FG_INLINE uses dip if we have an
664 				 * inline attr fork, else igets.
665 				 * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
666 				 * This is also used to count inodes/blks, etc
667 				 * in xfs_qm_quotacheck.
668 				 */
669 				ubused = statstruct_size;
670 				error = formatter(mp, ino, ubufp,
671 						ubleft, private_data,
672 						bno, &ubused, dip, &fmterror);
673 				if (fmterror == BULKSTAT_RV_NOTHING) {
674 					if (error && error != ENOENT &&
675 						error != EINVAL) {
676 						ubleft = 0;
677 						rval = error;
678 						break;
679 					}
680 					lastino = ino;
681 					continue;
682 				}
683 				if (fmterror == BULKSTAT_RV_GIVEUP) {
684 					ubleft = 0;
685 					ASSERT(error);
686 					rval = error;
687 					break;
688 				}
689 				if (ubufp)
690 					ubufp += ubused;
691 				ubleft -= ubused;
692 				ubelem++;
693 				lastino = ino;
694 			}
695 
696 			cond_resched();
697 		}
698 
699 		if (bp)
700 			xfs_buf_relse(bp);
701 
702 		/*
703 		 * Set up for the next loop iteration.
704 		 */
705 		if (XFS_BULKSTAT_UBLEFT(ubleft)) {
706 			if (end_of_ag) {
707 				agno++;
708 				agino = 0;
709 			} else
710 				agino = XFS_INO_TO_AGINO(mp, lastino);
711 		} else
712 			break;
713 	}
714 	/*
715 	 * Done, we're either out of filesystem or space to put the data.
716 	 */
717 	kmem_free(irbuf, irbsize);
718 	*ubcountp = ubelem;
719 	/*
720 	 * Found some inodes, return them now and return the error next time.
721 	 */
722 	if (ubelem)
723 		rval = 0;
724 	if (agno >= mp->m_sb.sb_agcount) {
725 		/*
726 		 * If we ran out of filesystem, mark lastino as off
727 		 * the end of the filesystem, so the next call
728 		 * will return immediately.
729 		 */
730 		*lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
731 		*done = 1;
732 	} else
733 		*lastinop = (xfs_ino_t)lastino;
734 
735 	return rval;
736 }
737 
738 /*
739  * Return stat information in bulk (by-inode) for the filesystem.
740  * Special case for non-sequential one inode bulkstat.
741  */
742 int					/* error status */
743 xfs_bulkstat_single(
744 	xfs_mount_t		*mp,	/* mount point for filesystem */
745 	xfs_ino_t		*lastinop, /* inode to return */
746 	char			__user *buffer, /* buffer with inode stats */
747 	int			*done)	/* 1 if there are more stats to get */
748 {
749 	int			count;	/* count value for bulkstat call */
750 	int			error;	/* return value */
751 	xfs_ino_t		ino;	/* filesystem inode number */
752 	int			res;	/* result from bs1 */
753 
754 	/*
755 	 * note that requesting valid inode numbers which are not allocated
756 	 * to inodes will most likely cause xfs_itobp to generate warning
757 	 * messages about bad magic numbers. This is ok. The fact that
758 	 * the inode isn't actually an inode is handled by the
759 	 * error check below. Done this way to make the usual case faster
760 	 * at the expense of the error case.
761 	 */
762 
763 	ino = (xfs_ino_t)*lastinop;
764 	error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
765 				 NULL, 0, NULL, NULL, &res);
766 	if (error) {
767 		/*
768 		 * Special case way failed, do it the "long" way
769 		 * to see if that works.
770 		 */
771 		(*lastinop)--;
772 		count = 1;
773 		if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
774 				NULL, sizeof(xfs_bstat_t), buffer,
775 				BULKSTAT_FG_IGET, done))
776 			return error;
777 		if (count == 0 || (xfs_ino_t)*lastinop != ino)
778 			return error == EFSCORRUPTED ?
779 				XFS_ERROR(EINVAL) : error;
780 		else
781 			return 0;
782 	}
783 	*done = 0;
784 	return 0;
785 }
786 
787 int
788 xfs_inumbers_fmt(
789 	void			__user *ubuffer, /* buffer to write to */
790 	const xfs_inogrp_t	*buffer,	/* buffer to read from */
791 	long			count,		/* # of elements to read */
792 	long			*written)	/* # of bytes written */
793 {
794 	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
795 		return -EFAULT;
796 	*written = count * sizeof(*buffer);
797 	return 0;
798 }
799 
800 /*
801  * Return inode number table for the filesystem.
802  */
803 int					/* error status */
804 xfs_inumbers(
805 	xfs_mount_t	*mp,		/* mount point for filesystem */
806 	xfs_ino_t	*lastino,	/* last inode returned */
807 	int		*count,		/* size of buffer/count returned */
808 	void		__user *ubuffer,/* buffer with inode descriptions */
809 	inumbers_fmt_pf	formatter)
810 {
811 	xfs_buf_t	*agbp;
812 	xfs_agino_t	agino;
813 	xfs_agnumber_t	agno;
814 	int		bcount;
815 	xfs_inogrp_t	*buffer;
816 	int		bufidx;
817 	xfs_btree_cur_t	*cur;
818 	int		error;
819 	__int32_t	gcnt;
820 	xfs_inofree_t	gfree;
821 	xfs_agino_t	gino;
822 	int		i;
823 	xfs_ino_t	ino;
824 	int		left;
825 	int		tmp;
826 
827 	ino = (xfs_ino_t)*lastino;
828 	agno = XFS_INO_TO_AGNO(mp, ino);
829 	agino = XFS_INO_TO_AGINO(mp, ino);
830 	left = *count;
831 	*count = 0;
832 	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
833 	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
834 	error = bufidx = 0;
835 	cur = NULL;
836 	agbp = NULL;
837 	while (left > 0 && agno < mp->m_sb.sb_agcount) {
838 		if (agbp == NULL) {
839 			down_read(&mp->m_peraglock);
840 			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
841 			up_read(&mp->m_peraglock);
842 			if (error) {
843 				/*
844 				 * If we can't read the AGI of this ag,
845 				 * then just skip to the next one.
846 				 */
847 				ASSERT(cur == NULL);
848 				agbp = NULL;
849 				agno++;
850 				agino = 0;
851 				continue;
852 			}
853 			cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
854 				XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
855 			error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
856 			if (error) {
857 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
858 				cur = NULL;
859 				xfs_buf_relse(agbp);
860 				agbp = NULL;
861 				/*
862 				 * Move up the last inode in the current
863 				 * chunk.  The lookup_ge will always get
864 				 * us the first inode in the next chunk.
865 				 */
866 				agino += XFS_INODES_PER_CHUNK - 1;
867 				continue;
868 			}
869 		}
870 		if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
871 			&i)) ||
872 		    i == 0) {
873 			xfs_buf_relse(agbp);
874 			agbp = NULL;
875 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
876 			cur = NULL;
877 			agno++;
878 			agino = 0;
879 			continue;
880 		}
881 		agino = gino + XFS_INODES_PER_CHUNK - 1;
882 		buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
883 		buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
884 		buffer[bufidx].xi_allocmask = ~gfree;
885 		bufidx++;
886 		left--;
887 		if (bufidx == bcount) {
888 			long written;
889 			if (formatter(ubuffer, buffer, bufidx, &written)) {
890 				error = XFS_ERROR(EFAULT);
891 				break;
892 			}
893 			ubuffer += written;
894 			*count += bufidx;
895 			bufidx = 0;
896 		}
897 		if (left) {
898 			error = xfs_inobt_increment(cur, 0, &tmp);
899 			if (error) {
900 				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
901 				cur = NULL;
902 				xfs_buf_relse(agbp);
903 				agbp = NULL;
904 				/*
905 				 * The agino value has already been bumped.
906 				 * Just try to skip up to it.
907 				 */
908 				agino += XFS_INODES_PER_CHUNK;
909 				continue;
910 			}
911 		}
912 	}
913 	if (!error) {
914 		if (bufidx) {
915 			long written;
916 			if (formatter(ubuffer, buffer, bufidx, &written))
917 				error = XFS_ERROR(EFAULT);
918 			else
919 				*count += bufidx;
920 		}
921 		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
922 	}
923 	kmem_free(buffer, bcount * sizeof(*buffer));
924 	if (cur)
925 		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
926 					   XFS_BTREE_NOERROR));
927 	if (agbp)
928 		xfs_buf_relse(agbp);
929 	return error;
930 }
931