cgroup: superblock can't be released with active dentries
[firefly-linux-kernel-4.4.55.git] / fs / xfs / xfs_itable.c
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_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_ialloc.h"
34 #include "xfs_itable.h"
35 #include "xfs_error.h"
36 #include "xfs_btree.h"
37 #include "xfs_trace.h"
38
39 STATIC int
40 xfs_internal_inum(
41         xfs_mount_t     *mp,
42         xfs_ino_t       ino)
43 {
44         return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
45                 (xfs_sb_version_hasquota(&mp->m_sb) &&
46                  (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino)));
47 }
48
49 /*
50  * Return stat information for one inode.
51  * Return 0 if ok, else errno.
52  */
53 int
54 xfs_bulkstat_one_int(
55         struct xfs_mount        *mp,            /* mount point for filesystem */
56         xfs_ino_t               ino,            /* inode to get data for */
57         void __user             *buffer,        /* buffer to place output in */
58         int                     ubsize,         /* size of buffer */
59         bulkstat_one_fmt_pf     formatter,      /* formatter, copy to user */
60         int                     *ubused,        /* bytes used by me */
61         int                     *stat)          /* BULKSTAT_RV_... */
62 {
63         struct xfs_icdinode     *dic;           /* dinode core info pointer */
64         struct xfs_inode        *ip;            /* incore inode pointer */
65         struct xfs_bstat        *buf;           /* return buffer */
66         int                     error = 0;      /* error value */
67
68         *stat = BULKSTAT_RV_NOTHING;
69
70         if (!buffer || xfs_internal_inum(mp, ino))
71                 return XFS_ERROR(EINVAL);
72
73         buf = kmem_alloc(sizeof(*buf), KM_SLEEP | KM_MAYFAIL);
74         if (!buf)
75                 return XFS_ERROR(ENOMEM);
76
77         error = xfs_iget(mp, NULL, ino,
78                          (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED),
79                          XFS_ILOCK_SHARED, &ip);
80         if (error) {
81                 *stat = BULKSTAT_RV_NOTHING;
82                 goto out_free;
83         }
84
85         ASSERT(ip != NULL);
86         ASSERT(ip->i_imap.im_blkno != 0);
87
88         dic = &ip->i_d;
89
90         /* xfs_iget returns the following without needing
91          * further change.
92          */
93         buf->bs_nlink = dic->di_nlink;
94         buf->bs_projid_lo = dic->di_projid_lo;
95         buf->bs_projid_hi = dic->di_projid_hi;
96         buf->bs_ino = ino;
97         buf->bs_mode = dic->di_mode;
98         buf->bs_uid = dic->di_uid;
99         buf->bs_gid = dic->di_gid;
100         buf->bs_size = dic->di_size;
101         buf->bs_atime.tv_sec = dic->di_atime.t_sec;
102         buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
103         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
104         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
105         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
106         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
107         buf->bs_xflags = xfs_ip2xflags(ip);
108         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
109         buf->bs_extents = dic->di_nextents;
110         buf->bs_gen = dic->di_gen;
111         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
112         buf->bs_dmevmask = dic->di_dmevmask;
113         buf->bs_dmstate = dic->di_dmstate;
114         buf->bs_aextents = dic->di_anextents;
115         buf->bs_forkoff = XFS_IFORK_BOFF(ip);
116
117         switch (dic->di_format) {
118         case XFS_DINODE_FMT_DEV:
119                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
120                 buf->bs_blksize = BLKDEV_IOSIZE;
121                 buf->bs_blocks = 0;
122                 break;
123         case XFS_DINODE_FMT_LOCAL:
124         case XFS_DINODE_FMT_UUID:
125                 buf->bs_rdev = 0;
126                 buf->bs_blksize = mp->m_sb.sb_blocksize;
127                 buf->bs_blocks = 0;
128                 break;
129         case XFS_DINODE_FMT_EXTENTS:
130         case XFS_DINODE_FMT_BTREE:
131                 buf->bs_rdev = 0;
132                 buf->bs_blksize = mp->m_sb.sb_blocksize;
133                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
134                 break;
135         }
136         xfs_iunlock(ip, XFS_ILOCK_SHARED);
137         IRELE(ip);
138
139         error = formatter(buffer, ubsize, ubused, buf);
140
141         if (!error)
142                 *stat = BULKSTAT_RV_DIDONE;
143
144  out_free:
145         kmem_free(buf);
146         return error;
147 }
148
149 /* Return 0 on success or positive error */
150 STATIC int
151 xfs_bulkstat_one_fmt(
152         void                    __user *ubuffer,
153         int                     ubsize,
154         int                     *ubused,
155         const xfs_bstat_t       *buffer)
156 {
157         if (ubsize < sizeof(*buffer))
158                 return XFS_ERROR(ENOMEM);
159         if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
160                 return XFS_ERROR(EFAULT);
161         if (ubused)
162                 *ubused = sizeof(*buffer);
163         return 0;
164 }
165
166 int
167 xfs_bulkstat_one(
168         xfs_mount_t     *mp,            /* mount point for filesystem */
169         xfs_ino_t       ino,            /* inode number to get data for */
170         void            __user *buffer, /* buffer to place output in */
171         int             ubsize,         /* size of buffer */
172         int             *ubused,        /* bytes used by me */
173         int             *stat)          /* BULKSTAT_RV_... */
174 {
175         return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
176                                     xfs_bulkstat_one_fmt, ubused, stat);
177 }
178
179 #define XFS_BULKSTAT_UBLEFT(ubleft)     ((ubleft) >= statstruct_size)
180
181 /*
182  * Return stat information in bulk (by-inode) for the filesystem.
183  */
184 int                                     /* error status */
185 xfs_bulkstat(
186         xfs_mount_t             *mp,    /* mount point for filesystem */
187         xfs_ino_t               *lastinop, /* last inode returned */
188         int                     *ubcountp, /* size of buffer/count returned */
189         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
190         size_t                  statstruct_size, /* sizeof struct filling */
191         char                    __user *ubuffer, /* buffer with inode stats */
192         int                     *done)  /* 1 if there are more stats to get */
193 {
194         xfs_agblock_t           agbno=0;/* allocation group block number */
195         xfs_buf_t               *agbp;  /* agi header buffer */
196         xfs_agi_t               *agi;   /* agi header data */
197         xfs_agino_t             agino;  /* inode # in allocation group */
198         xfs_agnumber_t          agno;   /* allocation group number */
199         int                     chunkidx; /* current index into inode chunk */
200         int                     clustidx; /* current index into inode cluster */
201         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
202         int                     end_of_ag; /* set if we've seen the ag end */
203         int                     error;  /* error code */
204         int                     fmterror;/* bulkstat formatter result */
205         int                     i;      /* loop index */
206         int                     icount; /* count of inodes good in irbuf */
207         size_t                  irbsize; /* size of irec buffer in bytes */
208         xfs_ino_t               ino;    /* inode number (filesystem) */
209         xfs_inobt_rec_incore_t  *irbp;  /* current irec buffer pointer */
210         xfs_inobt_rec_incore_t  *irbuf; /* start of irec buffer */
211         xfs_inobt_rec_incore_t  *irbufend; /* end of good irec buffer entries */
212         xfs_ino_t               lastino; /* last inode number returned */
213         int                     nbcluster; /* # of blocks in a cluster */
214         int                     nicluster; /* # of inodes in a cluster */
215         int                     nimask; /* mask for inode clusters */
216         int                     nirbuf; /* size of irbuf */
217         int                     rval;   /* return value error code */
218         int                     tmp;    /* result value from btree calls */
219         int                     ubcount; /* size of user's buffer */
220         int                     ubleft; /* bytes left in user's buffer */
221         char                    __user *ubufp;  /* pointer into user's buffer */
222         int                     ubelem; /* spaces used in user's buffer */
223         int                     ubused; /* bytes used by formatter */
224         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
225
226         /*
227          * Get the last inode value, see if there's nothing to do.
228          */
229         ino = (xfs_ino_t)*lastinop;
230         lastino = ino;
231         agno = XFS_INO_TO_AGNO(mp, ino);
232         agino = XFS_INO_TO_AGINO(mp, ino);
233         if (agno >= mp->m_sb.sb_agcount ||
234             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
235                 *done = 1;
236                 *ubcountp = 0;
237                 return 0;
238         }
239         if (!ubcountp || *ubcountp <= 0) {
240                 return EINVAL;
241         }
242         ubcount = *ubcountp; /* statstruct's */
243         ubleft = ubcount * statstruct_size; /* bytes */
244         *ubcountp = ubelem = 0;
245         *done = 0;
246         fmterror = 0;
247         ubufp = ubuffer;
248         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
249                 mp->m_sb.sb_inopblock :
250                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
251         nimask = ~(nicluster - 1);
252         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
253         irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4);
254         if (!irbuf)
255                 return ENOMEM;
256
257         nirbuf = irbsize / sizeof(*irbuf);
258
259         /*
260          * Loop over the allocation groups, starting from the last
261          * inode returned; 0 means start of the allocation group.
262          */
263         rval = 0;
264         while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
265                 cond_resched();
266                 bp = NULL;
267                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
268                 if (error) {
269                         /*
270                          * Skip this allocation group and go to the next one.
271                          */
272                         agno++;
273                         agino = 0;
274                         continue;
275                 }
276                 agi = XFS_BUF_TO_AGI(agbp);
277                 /*
278                  * Allocate and initialize a btree cursor for ialloc btree.
279                  */
280                 cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
281                 irbp = irbuf;
282                 irbufend = irbuf + nirbuf;
283                 end_of_ag = 0;
284                 /*
285                  * If we're returning in the middle of an allocation group,
286                  * we need to get the remainder of the chunk we're in.
287                  */
288                 if (agino > 0) {
289                         xfs_inobt_rec_incore_t r;
290
291                         /*
292                          * Lookup the inode chunk that this inode lives in.
293                          */
294                         error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE,
295                                                  &tmp);
296                         if (!error &&   /* no I/O error */
297                             tmp &&      /* lookup succeeded */
298                                         /* got the record, should always work */
299                             !(error = xfs_inobt_get_rec(cur, &r, &i)) &&
300                             i == 1 &&
301                                         /* this is the right chunk */
302                             agino < r.ir_startino + XFS_INODES_PER_CHUNK &&
303                                         /* lastino was not last in chunk */
304                             (chunkidx = agino - r.ir_startino + 1) <
305                                     XFS_INODES_PER_CHUNK &&
306                                         /* there are some left allocated */
307                             xfs_inobt_maskn(chunkidx,
308                                     XFS_INODES_PER_CHUNK - chunkidx) &
309                                     ~r.ir_free) {
310                                 /*
311                                  * Grab the chunk record.  Mark all the
312                                  * uninteresting inodes (because they're
313                                  * before our start point) free.
314                                  */
315                                 for (i = 0; i < chunkidx; i++) {
316                                         if (XFS_INOBT_MASK(i) & ~r.ir_free)
317                                                 r.ir_freecount++;
318                                 }
319                                 r.ir_free |= xfs_inobt_maskn(0, chunkidx);
320                                 irbp->ir_startino = r.ir_startino;
321                                 irbp->ir_freecount = r.ir_freecount;
322                                 irbp->ir_free = r.ir_free;
323                                 irbp++;
324                                 agino = r.ir_startino + XFS_INODES_PER_CHUNK;
325                                 icount = XFS_INODES_PER_CHUNK - r.ir_freecount;
326                         } else {
327                                 /*
328                                  * If any of those tests failed, bump the
329                                  * inode number (just in case).
330                                  */
331                                 agino++;
332                                 icount = 0;
333                         }
334                         /*
335                          * In any case, increment to the next record.
336                          */
337                         if (!error)
338                                 error = xfs_btree_increment(cur, 0, &tmp);
339                 } else {
340                         /*
341                          * Start of ag.  Lookup the first inode chunk.
342                          */
343                         error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &tmp);
344                         icount = 0;
345                 }
346                 /*
347                  * Loop through inode btree records in this ag,
348                  * until we run out of inodes or space in the buffer.
349                  */
350                 while (irbp < irbufend && icount < ubcount) {
351                         xfs_inobt_rec_incore_t r;
352
353                         /*
354                          * Loop as long as we're unable to read the
355                          * inode btree.
356                          */
357                         while (error) {
358                                 agino += XFS_INODES_PER_CHUNK;
359                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
360                                                 be32_to_cpu(agi->agi_length))
361                                         break;
362                                 error = xfs_inobt_lookup(cur, agino,
363                                                          XFS_LOOKUP_GE, &tmp);
364                                 cond_resched();
365                         }
366                         /*
367                          * If ran off the end of the ag either with an error,
368                          * or the normal way, set end and stop collecting.
369                          */
370                         if (error) {
371                                 end_of_ag = 1;
372                                 break;
373                         }
374
375                         error = xfs_inobt_get_rec(cur, &r, &i);
376                         if (error || i == 0) {
377                                 end_of_ag = 1;
378                                 break;
379                         }
380
381                         /*
382                          * If this chunk has any allocated inodes, save it.
383                          * Also start read-ahead now for this chunk.
384                          */
385                         if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
386                                 /*
387                                  * Loop over all clusters in the next chunk.
388                                  * Do a readahead if there are any allocated
389                                  * inodes in that cluster.
390                                  */
391                                 agbno = XFS_AGINO_TO_AGBNO(mp, r.ir_startino);
392                                 for (chunkidx = 0;
393                                      chunkidx < XFS_INODES_PER_CHUNK;
394                                      chunkidx += nicluster,
395                                      agbno += nbcluster) {
396                                         if (xfs_inobt_maskn(chunkidx, nicluster)
397                                                         & ~r.ir_free)
398                                                 xfs_btree_reada_bufs(mp, agno,
399                                                         agbno, nbcluster);
400                                 }
401                                 irbp->ir_startino = r.ir_startino;
402                                 irbp->ir_freecount = r.ir_freecount;
403                                 irbp->ir_free = r.ir_free;
404                                 irbp++;
405                                 icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
406                         }
407                         /*
408                          * Set agino to after this chunk and bump the cursor.
409                          */
410                         agino = r.ir_startino + XFS_INODES_PER_CHUNK;
411                         error = xfs_btree_increment(cur, 0, &tmp);
412                         cond_resched();
413                 }
414                 /*
415                  * Drop the btree buffers and the agi buffer.
416                  * We can't hold any of the locks these represent
417                  * when calling iget.
418                  */
419                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
420                 xfs_buf_relse(agbp);
421                 /*
422                  * Now format all the good inodes into the user's buffer.
423                  */
424                 irbufend = irbp;
425                 for (irbp = irbuf;
426                      irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
427                         /*
428                          * Now process this chunk of inodes.
429                          */
430                         for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
431                              XFS_BULKSTAT_UBLEFT(ubleft) &&
432                                 irbp->ir_freecount < XFS_INODES_PER_CHUNK;
433                              chunkidx++, clustidx++, agino++) {
434                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
435                                 /*
436                                  * Recompute agbno if this is the
437                                  * first inode of the cluster.
438                                  *
439                                  * Careful with clustidx.   There can be
440                                  * multiple clusters per chunk, a single
441                                  * cluster per chunk or a cluster that has
442                                  * inodes represented from several different
443                                  * chunks (if blocksize is large).
444                                  *
445                                  * Because of this, the starting clustidx is
446                                  * initialized to zero in this loop but must
447                                  * later be reset after reading in the cluster
448                                  * buffer.
449                                  */
450                                 if ((chunkidx & (nicluster - 1)) == 0) {
451                                         agbno = XFS_AGINO_TO_AGBNO(mp,
452                                                         irbp->ir_startino) +
453                                                 ((chunkidx & nimask) >>
454                                                  mp->m_sb.sb_inopblog);
455                                 }
456                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
457                                 /*
458                                  * Skip if this inode is free.
459                                  */
460                                 if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
461                                         lastino = ino;
462                                         continue;
463                                 }
464                                 /*
465                                  * Count used inodes as free so we can tell
466                                  * when the chunk is used up.
467                                  */
468                                 irbp->ir_freecount++;
469
470                                 /*
471                                  * Get the inode and fill in a single buffer.
472                                  */
473                                 ubused = statstruct_size;
474                                 error = formatter(mp, ino, ubufp, ubleft,
475                                                   &ubused, &fmterror);
476                                 if (fmterror == BULKSTAT_RV_NOTHING) {
477                                         if (error && error != ENOENT &&
478                                                 error != EINVAL) {
479                                                 ubleft = 0;
480                                                 rval = error;
481                                                 break;
482                                         }
483                                         lastino = ino;
484                                         continue;
485                                 }
486                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
487                                         ubleft = 0;
488                                         ASSERT(error);
489                                         rval = error;
490                                         break;
491                                 }
492                                 if (ubufp)
493                                         ubufp += ubused;
494                                 ubleft -= ubused;
495                                 ubelem++;
496                                 lastino = ino;
497                         }
498
499                         cond_resched();
500                 }
501
502                 if (bp)
503                         xfs_buf_relse(bp);
504
505                 /*
506                  * Set up for the next loop iteration.
507                  */
508                 if (XFS_BULKSTAT_UBLEFT(ubleft)) {
509                         if (end_of_ag) {
510                                 agno++;
511                                 agino = 0;
512                         } else
513                                 agino = XFS_INO_TO_AGINO(mp, lastino);
514                 } else
515                         break;
516         }
517         /*
518          * Done, we're either out of filesystem or space to put the data.
519          */
520         kmem_free_large(irbuf);
521         *ubcountp = ubelem;
522         /*
523          * Found some inodes, return them now and return the error next time.
524          */
525         if (ubelem)
526                 rval = 0;
527         if (agno >= mp->m_sb.sb_agcount) {
528                 /*
529                  * If we ran out of filesystem, mark lastino as off
530                  * the end of the filesystem, so the next call
531                  * will return immediately.
532                  */
533                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
534                 *done = 1;
535         } else
536                 *lastinop = (xfs_ino_t)lastino;
537
538         return rval;
539 }
540
541 /*
542  * Return stat information in bulk (by-inode) for the filesystem.
543  * Special case for non-sequential one inode bulkstat.
544  */
545 int                                     /* error status */
546 xfs_bulkstat_single(
547         xfs_mount_t             *mp,    /* mount point for filesystem */
548         xfs_ino_t               *lastinop, /* inode to return */
549         char                    __user *buffer, /* buffer with inode stats */
550         int                     *done)  /* 1 if there are more stats to get */
551 {
552         int                     count;  /* count value for bulkstat call */
553         int                     error;  /* return value */
554         xfs_ino_t               ino;    /* filesystem inode number */
555         int                     res;    /* result from bs1 */
556
557         /*
558          * note that requesting valid inode numbers which are not allocated
559          * to inodes will most likely cause xfs_itobp to generate warning
560          * messages about bad magic numbers. This is ok. The fact that
561          * the inode isn't actually an inode is handled by the
562          * error check below. Done this way to make the usual case faster
563          * at the expense of the error case.
564          */
565
566         ino = (xfs_ino_t)*lastinop;
567         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t), 0, &res);
568         if (error) {
569                 /*
570                  * Special case way failed, do it the "long" way
571                  * to see if that works.
572                  */
573                 (*lastinop)--;
574                 count = 1;
575                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
576                                 sizeof(xfs_bstat_t), buffer, done))
577                         return error;
578                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
579                         return error == EFSCORRUPTED ?
580                                 XFS_ERROR(EINVAL) : error;
581                 else
582                         return 0;
583         }
584         *done = 0;
585         return 0;
586 }
587
588 int
589 xfs_inumbers_fmt(
590         void                    __user *ubuffer, /* buffer to write to */
591         const xfs_inogrp_t      *buffer,        /* buffer to read from */
592         long                    count,          /* # of elements to read */
593         long                    *written)       /* # of bytes written */
594 {
595         if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
596                 return -EFAULT;
597         *written = count * sizeof(*buffer);
598         return 0;
599 }
600
601 /*
602  * Return inode number table for the filesystem.
603  */
604 int                                     /* error status */
605 xfs_inumbers(
606         xfs_mount_t     *mp,            /* mount point for filesystem */
607         xfs_ino_t       *lastino,       /* last inode returned */
608         int             *count,         /* size of buffer/count returned */
609         void            __user *ubuffer,/* buffer with inode descriptions */
610         inumbers_fmt_pf formatter)
611 {
612         xfs_buf_t       *agbp;
613         xfs_agino_t     agino;
614         xfs_agnumber_t  agno;
615         int             bcount;
616         xfs_inogrp_t    *buffer;
617         int             bufidx;
618         xfs_btree_cur_t *cur;
619         int             error;
620         xfs_inobt_rec_incore_t r;
621         int             i;
622         xfs_ino_t       ino;
623         int             left;
624         int             tmp;
625
626         ino = (xfs_ino_t)*lastino;
627         agno = XFS_INO_TO_AGNO(mp, ino);
628         agino = XFS_INO_TO_AGINO(mp, ino);
629         left = *count;
630         *count = 0;
631         bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
632         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
633         error = bufidx = 0;
634         cur = NULL;
635         agbp = NULL;
636         while (left > 0 && agno < mp->m_sb.sb_agcount) {
637                 if (agbp == NULL) {
638                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
639                         if (error) {
640                                 /*
641                                  * If we can't read the AGI of this ag,
642                                  * then just skip to the next one.
643                                  */
644                                 ASSERT(cur == NULL);
645                                 agbp = NULL;
646                                 agno++;
647                                 agino = 0;
648                                 continue;
649                         }
650                         cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
651                         error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
652                                                  &tmp);
653                         if (error) {
654                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
655                                 cur = NULL;
656                                 xfs_buf_relse(agbp);
657                                 agbp = NULL;
658                                 /*
659                                  * Move up the last inode in the current
660                                  * chunk.  The lookup_ge will always get
661                                  * us the first inode in the next chunk.
662                                  */
663                                 agino += XFS_INODES_PER_CHUNK - 1;
664                                 continue;
665                         }
666                 }
667                 error = xfs_inobt_get_rec(cur, &r, &i);
668                 if (error || i == 0) {
669                         xfs_buf_relse(agbp);
670                         agbp = NULL;
671                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
672                         cur = NULL;
673                         agno++;
674                         agino = 0;
675                         continue;
676                 }
677                 agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
678                 buffer[bufidx].xi_startino =
679                         XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
680                 buffer[bufidx].xi_alloccount =
681                         XFS_INODES_PER_CHUNK - r.ir_freecount;
682                 buffer[bufidx].xi_allocmask = ~r.ir_free;
683                 bufidx++;
684                 left--;
685                 if (bufidx == bcount) {
686                         long written;
687                         if (formatter(ubuffer, buffer, bufidx, &written)) {
688                                 error = XFS_ERROR(EFAULT);
689                                 break;
690                         }
691                         ubuffer += written;
692                         *count += bufidx;
693                         bufidx = 0;
694                 }
695                 if (left) {
696                         error = xfs_btree_increment(cur, 0, &tmp);
697                         if (error) {
698                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
699                                 cur = NULL;
700                                 xfs_buf_relse(agbp);
701                                 agbp = NULL;
702                                 /*
703                                  * The agino value has already been bumped.
704                                  * Just try to skip up to it.
705                                  */
706                                 agino += XFS_INODES_PER_CHUNK;
707                                 continue;
708                         }
709                 }
710         }
711         if (!error) {
712                 if (bufidx) {
713                         long written;
714                         if (formatter(ubuffer, buffer, bufidx, &written))
715                                 error = XFS_ERROR(EFAULT);
716                         else
717                                 *count += bufidx;
718                 }
719                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
720         }
721         kmem_free(buffer);
722         if (cur)
723                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
724                                            XFS_BTREE_NOERROR));
725         if (agbp)
726                 xfs_buf_relse(agbp);
727         return error;
728 }