Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6.git] / fs / xfs / xfs_vnodeops.c
1 /*
2  * Copyright (c) 2000-2005 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
19 #include <linux/capability.h>
20
21 #include "xfs.h"
22 #include "xfs_fs.h"
23 #include "xfs_types.h"
24 #include "xfs_bit.h"
25 #include "xfs_log.h"
26 #include "xfs_inum.h"
27 #include "xfs_trans.h"
28 #include "xfs_sb.h"
29 #include "xfs_ag.h"
30 #include "xfs_dir.h"
31 #include "xfs_dir2.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_da_btree.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_alloc_btree.h"
37 #include "xfs_ialloc_btree.h"
38 #include "xfs_dir_sf.h"
39 #include "xfs_dir2_sf.h"
40 #include "xfs_attr_sf.h"
41 #include "xfs_dinode.h"
42 #include "xfs_inode.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_dir_leaf.h"
45 #include "xfs_itable.h"
46 #include "xfs_btree.h"
47 #include "xfs_ialloc.h"
48 #include "xfs_alloc.h"
49 #include "xfs_bmap.h"
50 #include "xfs_attr.h"
51 #include "xfs_rw.h"
52 #include "xfs_error.h"
53 #include "xfs_quota.h"
54 #include "xfs_utils.h"
55 #include "xfs_rtalloc.h"
56 #include "xfs_refcache.h"
57 #include "xfs_trans_space.h"
58 #include "xfs_log_priv.h"
59 #include "xfs_mac.h"
60
61
62 /*
63  * The maximum pathlen is 1024 bytes. Since the minimum file system
64  * blocksize is 512 bytes, we can get a max of 2 extents back from
65  * bmapi.
66  */
67 #define SYMLINK_MAPS 2
68
69 /*
70  * For xfs, we check that the file isn't too big to be opened by this kernel.
71  * No other open action is required for regular files.  Devices are handled
72  * through the specfs file system, pipes through fifofs.  Device and
73  * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
74  * when a new vnode is first looked up or created.
75  */
76 STATIC int
77 xfs_open(
78         bhv_desc_t      *bdp,
79         cred_t          *credp)
80 {
81         int             mode;
82         vnode_t         *vp;
83         xfs_inode_t     *ip;
84
85         vp = BHV_TO_VNODE(bdp);
86         ip = XFS_BHVTOI(bdp);
87
88         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
89                 return XFS_ERROR(EIO);
90
91         /*
92          * If it's a directory with any blocks, read-ahead block 0
93          * as we're almost certain to have the next operation be a read there.
94          */
95         if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
96                 mode = xfs_ilock_map_shared(ip);
97                 if (ip->i_d.di_nextents > 0)
98                         (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
99                 xfs_iunlock(ip, mode);
100         }
101         return 0;
102 }
103
104
105 /*
106  * xfs_getattr
107  */
108 STATIC int
109 xfs_getattr(
110         bhv_desc_t      *bdp,
111         vattr_t         *vap,
112         int             flags,
113         cred_t          *credp)
114 {
115         xfs_inode_t     *ip;
116         xfs_mount_t     *mp;
117         vnode_t         *vp;
118
119         vp  = BHV_TO_VNODE(bdp);
120         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
121
122         ip = XFS_BHVTOI(bdp);
123         mp = ip->i_mount;
124
125         if (XFS_FORCED_SHUTDOWN(mp))
126                 return XFS_ERROR(EIO);
127
128         if (!(flags & ATTR_LAZY))
129                 xfs_ilock(ip, XFS_ILOCK_SHARED);
130
131         vap->va_size = ip->i_d.di_size;
132         if (vap->va_mask == XFS_AT_SIZE)
133                 goto all_done;
134
135         vap->va_nblocks =
136                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
137         vap->va_nodeid = ip->i_ino;
138 #if XFS_BIG_INUMS
139         vap->va_nodeid += mp->m_inoadd;
140 #endif
141         vap->va_nlink = ip->i_d.di_nlink;
142
143         /*
144          * Quick exit for non-stat callers
145          */
146         if ((vap->va_mask &
147             ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
148               XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
149                 goto all_done;
150
151         /*
152          * Copy from in-core inode.
153          */
154         vap->va_mode = ip->i_d.di_mode;
155         vap->va_uid = ip->i_d.di_uid;
156         vap->va_gid = ip->i_d.di_gid;
157         vap->va_projid = ip->i_d.di_projid;
158
159         /*
160          * Check vnode type block/char vs. everything else.
161          */
162         switch (ip->i_d.di_mode & S_IFMT) {
163         case S_IFBLK:
164         case S_IFCHR:
165                 vap->va_rdev = ip->i_df.if_u2.if_rdev;
166                 vap->va_blocksize = BLKDEV_IOSIZE;
167                 break;
168         default:
169                 vap->va_rdev = 0;
170
171                 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
172                         vap->va_blocksize = xfs_preferred_iosize(mp);
173                 } else {
174
175                         /*
176                          * If the file blocks are being allocated from a
177                          * realtime partition, then return the inode's
178                          * realtime extent size or the realtime volume's
179                          * extent size.
180                          */
181                         vap->va_blocksize = ip->i_d.di_extsize ?
182                                 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
183                                 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
184                 }
185                 break;
186         }
187
188         vn_atime_to_timespec(vp, &vap->va_atime);
189         vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
190         vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
191         vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
192         vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
193
194         /*
195          * Exit for stat callers.  See if any of the rest of the fields
196          * to be filled in are needed.
197          */
198         if ((vap->va_mask &
199              (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
200               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
201                 goto all_done;
202
203         /*
204          * Convert di_flags to xflags.
205          */
206         vap->va_xflags = xfs_ip2xflags(ip);
207
208         /*
209          * Exit for inode revalidate.  See if any of the rest of
210          * the fields to be filled in are needed.
211          */
212         if ((vap->va_mask &
213              (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
214               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
215                 goto all_done;
216
217         vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
218         vap->va_nextents =
219                 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
220                         ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
221                         ip->i_d.di_nextents;
222         if (ip->i_afp)
223                 vap->va_anextents =
224                         (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
225                                 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
226                                  ip->i_d.di_anextents;
227         else
228                 vap->va_anextents = 0;
229         vap->va_gen = ip->i_d.di_gen;
230
231  all_done:
232         if (!(flags & ATTR_LAZY))
233                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
234         return 0;
235 }
236
237
238 /*
239  * xfs_setattr
240  */
241 int
242 xfs_setattr(
243         bhv_desc_t              *bdp,
244         vattr_t                 *vap,
245         int                     flags,
246         cred_t                  *credp)
247 {
248         xfs_inode_t             *ip;
249         xfs_trans_t             *tp;
250         xfs_mount_t             *mp;
251         int                     mask;
252         int                     code;
253         uint                    lock_flags;
254         uint                    commit_flags=0;
255         uid_t                   uid=0, iuid=0;
256         gid_t                   gid=0, igid=0;
257         int                     timeflags = 0;
258         vnode_t                 *vp;
259         xfs_prid_t              projid=0, iprojid=0;
260         int                     mandlock_before, mandlock_after;
261         struct xfs_dquot        *udqp, *gdqp, *olddquot1, *olddquot2;
262         int                     file_owner;
263         int                     need_iolock = 1;
264
265         vp = BHV_TO_VNODE(bdp);
266         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
267
268         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
269                 return XFS_ERROR(EROFS);
270
271         /*
272          * Cannot set certain attributes.
273          */
274         mask = vap->va_mask;
275         if (mask & XFS_AT_NOSET) {
276                 return XFS_ERROR(EINVAL);
277         }
278
279         ip = XFS_BHVTOI(bdp);
280         mp = ip->i_mount;
281
282         if (XFS_FORCED_SHUTDOWN(mp))
283                 return XFS_ERROR(EIO);
284
285         /*
286          * Timestamps do not need to be logged and hence do not
287          * need to be done within a transaction.
288          */
289         if (mask & XFS_AT_UPDTIMES) {
290                 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
291                 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
292                             ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
293                             ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
294                 xfs_ichgtime(ip, timeflags);
295                 return 0;
296         }
297
298         olddquot1 = olddquot2 = NULL;
299         udqp = gdqp = NULL;
300
301         /*
302          * If disk quotas is on, we make sure that the dquots do exist on disk,
303          * before we start any other transactions. Trying to do this later
304          * is messy. We don't care to take a readlock to look at the ids
305          * in inode here, because we can't hold it across the trans_reserve.
306          * If the IDs do change before we take the ilock, we're covered
307          * because the i_*dquot fields will get updated anyway.
308          */
309         if (XFS_IS_QUOTA_ON(mp) &&
310             (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
311                 uint    qflags = 0;
312
313                 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
314                         uid = vap->va_uid;
315                         qflags |= XFS_QMOPT_UQUOTA;
316                 } else {
317                         uid = ip->i_d.di_uid;
318                 }
319                 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
320                         gid = vap->va_gid;
321                         qflags |= XFS_QMOPT_GQUOTA;
322                 }  else {
323                         gid = ip->i_d.di_gid;
324                 }
325                 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
326                         projid = vap->va_projid;
327                         qflags |= XFS_QMOPT_PQUOTA;
328                 }  else {
329                         projid = ip->i_d.di_projid;
330                 }
331                 /*
332                  * We take a reference when we initialize udqp and gdqp,
333                  * so it is important that we never blindly double trip on
334                  * the same variable. See xfs_create() for an example.
335                  */
336                 ASSERT(udqp == NULL);
337                 ASSERT(gdqp == NULL);
338                 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
339                                          &udqp, &gdqp);
340                 if (code)
341                         return (code);
342         }
343
344         /*
345          * For the other attributes, we acquire the inode lock and
346          * first do an error checking pass.
347          */
348         tp = NULL;
349         lock_flags = XFS_ILOCK_EXCL;
350         ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
351         if (flags & ATTR_NOLOCK)
352                 need_iolock = 0;
353         if (!(mask & XFS_AT_SIZE)) {
354                 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
355                     (mp->m_flags & XFS_MOUNT_WSYNC)) {
356                         tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
357                         commit_flags = 0;
358                         if ((code = xfs_trans_reserve(tp, 0,
359                                                      XFS_ICHANGE_LOG_RES(mp), 0,
360                                                      0, 0))) {
361                                 lock_flags = 0;
362                                 goto error_return;
363                         }
364                 }
365         } else {
366                 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
367                     !(flags & ATTR_DMI)) {
368                         int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
369                         code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
370                                 vap->va_size, 0, dmflags, NULL);
371                         if (code) {
372                                 lock_flags = 0;
373                                 goto error_return;
374                         }
375                 }
376                 if (need_iolock)
377                         lock_flags |= XFS_IOLOCK_EXCL;
378         }
379
380         xfs_ilock(ip, lock_flags);
381
382         /* boolean: are we the file owner? */
383         file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
384
385         /*
386          * Change various properties of a file.
387          * Only the owner or users with CAP_FOWNER
388          * capability may do these things.
389          */
390         if (mask &
391             (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
392              XFS_AT_GID|XFS_AT_PROJID)) {
393                 /*
394                  * CAP_FOWNER overrides the following restrictions:
395                  *
396                  * The user ID of the calling process must be equal
397                  * to the file owner ID, except in cases where the
398                  * CAP_FSETID capability is applicable.
399                  */
400                 if (!file_owner && !capable(CAP_FOWNER)) {
401                         code = XFS_ERROR(EPERM);
402                         goto error_return;
403                 }
404
405                 /*
406                  * CAP_FSETID overrides the following restrictions:
407                  *
408                  * The effective user ID of the calling process shall match
409                  * the file owner when setting the set-user-ID and
410                  * set-group-ID bits on that file.
411                  *
412                  * The effective group ID or one of the supplementary group
413                  * IDs of the calling process shall match the group owner of
414                  * the file when setting the set-group-ID bit on that file
415                  */
416                 if (mask & XFS_AT_MODE) {
417                         mode_t m = 0;
418
419                         if ((vap->va_mode & S_ISUID) && !file_owner)
420                                 m |= S_ISUID;
421                         if ((vap->va_mode & S_ISGID) &&
422                             !in_group_p((gid_t)ip->i_d.di_gid))
423                                 m |= S_ISGID;
424 #if 0
425                         /* Linux allows this, Irix doesn't. */
426                         if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
427                                 m |= S_ISVTX;
428 #endif
429                         if (m && !capable(CAP_FSETID))
430                                 vap->va_mode &= ~m;
431                 }
432         }
433
434         /*
435          * Change file ownership.  Must be the owner or privileged.
436          * If the system was configured with the "restricted_chown"
437          * option, the owner is not permitted to give away the file,
438          * and can change the group id only to a group of which he
439          * or she is a member.
440          */
441         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
442                 /*
443                  * These IDs could have changed since we last looked at them.
444                  * But, we're assured that if the ownership did change
445                  * while we didn't have the inode locked, inode's dquot(s)
446                  * would have changed also.
447                  */
448                 iuid = ip->i_d.di_uid;
449                 iprojid = ip->i_d.di_projid;
450                 igid = ip->i_d.di_gid;
451                 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
452                 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
453                 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
454                          iprojid;
455
456                 /*
457                  * CAP_CHOWN overrides the following restrictions:
458                  *
459                  * If _POSIX_CHOWN_RESTRICTED is defined, this capability
460                  * shall override the restriction that a process cannot
461                  * change the user ID of a file it owns and the restriction
462                  * that the group ID supplied to the chown() function
463                  * shall be equal to either the group ID or one of the
464                  * supplementary group IDs of the calling process.
465                  */
466                 if (restricted_chown &&
467                     (iuid != uid || (igid != gid &&
468                                      !in_group_p((gid_t)gid))) &&
469                     !capable(CAP_CHOWN)) {
470                         code = XFS_ERROR(EPERM);
471                         goto error_return;
472                 }
473                 /*
474                  * Do a quota reservation only if uid/projid/gid is actually
475                  * going to change.
476                  */
477                 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
478                     (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
479                     (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
480                         ASSERT(tp);
481                         code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
482                                                 capable(CAP_FOWNER) ?
483                                                 XFS_QMOPT_FORCE_RES : 0);
484                         if (code)       /* out of quota */
485                                 goto error_return;
486                 }
487         }
488
489         /*
490          * Truncate file.  Must have write permission and not be a directory.
491          */
492         if (mask & XFS_AT_SIZE) {
493                 /* Short circuit the truncate case for zero length files */
494                 if ((vap->va_size == 0) &&
495                    (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
496                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
497                         lock_flags &= ~XFS_ILOCK_EXCL;
498                         if (mask & XFS_AT_CTIME)
499                                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
500                         code = 0;
501                         goto error_return;
502                 }
503
504                 if (VN_ISDIR(vp)) {
505                         code = XFS_ERROR(EISDIR);
506                         goto error_return;
507                 } else if (!VN_ISREG(vp)) {
508                         code = XFS_ERROR(EINVAL);
509                         goto error_return;
510                 }
511                 /*
512                  * Make sure that the dquots are attached to the inode.
513                  */
514                 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
515                         goto error_return;
516         }
517
518         /*
519          * Change file access or modified times.
520          */
521         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
522                 if (!file_owner) {
523                         if ((flags & ATTR_UTIME) &&
524                             !capable(CAP_FOWNER)) {
525                                 code = XFS_ERROR(EPERM);
526                                 goto error_return;
527                         }
528                 }
529         }
530
531         /*
532          * Change extent size or realtime flag.
533          */
534         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
535                 /*
536                  * Can't change extent size if any extents are allocated.
537                  */
538                 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
539                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
540                      vap->va_extsize) ) {
541                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
542                         goto error_return;
543                 }
544
545                 /*
546                  * Can't change realtime flag if any extents are allocated.
547                  */
548                 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
549                     (mask & XFS_AT_XFLAGS) &&
550                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
551                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
552                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
553                         goto error_return;
554                 }
555                 /*
556                  * Extent size must be a multiple of the appropriate block
557                  * size, if set at all.
558                  */
559                 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
560                         xfs_extlen_t    size;
561
562                         if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
563                             ((mask & XFS_AT_XFLAGS) &&
564                             (vap->va_xflags & XFS_XFLAG_REALTIME))) {
565                                 size = mp->m_sb.sb_rextsize <<
566                                        mp->m_sb.sb_blocklog;
567                         } else {
568                                 size = mp->m_sb.sb_blocksize;
569                         }
570                         if (vap->va_extsize % size) {
571                                 code = XFS_ERROR(EINVAL);
572                                 goto error_return;
573                         }
574                 }
575                 /*
576                  * If realtime flag is set then must have realtime data.
577                  */
578                 if ((mask & XFS_AT_XFLAGS) &&
579                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
580                         if ((mp->m_sb.sb_rblocks == 0) ||
581                             (mp->m_sb.sb_rextsize == 0) ||
582                             (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
583                                 code = XFS_ERROR(EINVAL);
584                                 goto error_return;
585                         }
586                 }
587
588                 /*
589                  * Can't modify an immutable/append-only file unless
590                  * we have appropriate permission.
591                  */
592                 if ((mask & XFS_AT_XFLAGS) &&
593                     (ip->i_d.di_flags &
594                                 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
595                      (vap->va_xflags &
596                                 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
597                     !capable(CAP_LINUX_IMMUTABLE)) {
598                         code = XFS_ERROR(EPERM);
599                         goto error_return;
600                 }
601         }
602
603         /*
604          * Now we can make the changes.  Before we join the inode
605          * to the transaction, if XFS_AT_SIZE is set then take care of
606          * the part of the truncation that must be done without the
607          * inode lock.  This needs to be done before joining the inode
608          * to the transaction, because the inode cannot be unlocked
609          * once it is a part of the transaction.
610          */
611         if (mask & XFS_AT_SIZE) {
612                 code = 0;
613                 if ((vap->va_size > ip->i_d.di_size) && 
614                     (flags & ATTR_NOSIZETOK) == 0) {
615                         code = xfs_igrow_start(ip, vap->va_size, credp);
616                 }
617                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
618                 if (!code)
619                         code = xfs_itruncate_data(ip, vap->va_size);
620                 if (code) {
621                         ASSERT(tp == NULL);
622                         lock_flags &= ~XFS_ILOCK_EXCL;
623                         ASSERT(lock_flags == XFS_IOLOCK_EXCL);
624                         goto error_return;
625                 }
626                 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
627                 if ((code = xfs_trans_reserve(tp, 0,
628                                              XFS_ITRUNCATE_LOG_RES(mp), 0,
629                                              XFS_TRANS_PERM_LOG_RES,
630                                              XFS_ITRUNCATE_LOG_COUNT))) {
631                         xfs_trans_cancel(tp, 0);
632                         if (need_iolock)
633                                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
634                         return code;
635                 }
636                 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
637                 xfs_ilock(ip, XFS_ILOCK_EXCL);
638         }
639
640         if (tp) {
641                 xfs_trans_ijoin(tp, ip, lock_flags);
642                 xfs_trans_ihold(tp, ip);
643         }
644
645         /* determine whether mandatory locking mode changes */
646         mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
647
648         /*
649          * Truncate file.  Must have write permission and not be a directory.
650          */
651         if (mask & XFS_AT_SIZE) {
652                 if (vap->va_size > ip->i_d.di_size) {
653                         xfs_igrow_finish(tp, ip, vap->va_size,
654                             !(flags & ATTR_DMI));
655                 } else if ((vap->va_size <= ip->i_d.di_size) ||
656                            ((vap->va_size == 0) && ip->i_d.di_nextents)) {
657                         /*
658                          * signal a sync transaction unless
659                          * we're truncating an already unlinked
660                          * file on a wsync filesystem
661                          */
662                         code = xfs_itruncate_finish(&tp, ip,
663                                             (xfs_fsize_t)vap->va_size,
664                                             XFS_DATA_FORK,
665                                             ((ip->i_d.di_nlink != 0 ||
666                                               !(mp->m_flags & XFS_MOUNT_WSYNC))
667                                              ? 1 : 0));
668                         if (code) {
669                                 goto abort_return;
670                         }
671                 }
672                 /*
673                  * Have to do this even if the file's size doesn't change.
674                  */
675                 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
676         }
677
678         /*
679          * Change file access modes.
680          */
681         if (mask & XFS_AT_MODE) {
682                 ip->i_d.di_mode &= S_IFMT;
683                 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
684
685                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
686                 timeflags |= XFS_ICHGTIME_CHG;
687         }
688
689         /*
690          * Change file ownership.  Must be the owner or privileged.
691          * If the system was configured with the "restricted_chown"
692          * option, the owner is not permitted to give away the file,
693          * and can change the group id only to a group of which he
694          * or she is a member.
695          */
696         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
697                 /*
698                  * CAP_FSETID overrides the following restrictions:
699                  *
700                  * The set-user-ID and set-group-ID bits of a file will be
701                  * cleared upon successful return from chown()
702                  */
703                 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
704                     !capable(CAP_FSETID)) {
705                         ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
706                 }
707
708                 /*
709                  * Change the ownerships and register quota modifications
710                  * in the transaction.
711                  */
712                 if (iuid != uid) {
713                         if (XFS_IS_UQUOTA_ON(mp)) {
714                                 ASSERT(mask & XFS_AT_UID);
715                                 ASSERT(udqp);
716                                 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
717                                                         &ip->i_udquot, udqp);
718                         }
719                         ip->i_d.di_uid = uid;
720                 }
721                 if (igid != gid) {
722                         if (XFS_IS_GQUOTA_ON(mp)) {
723                                 ASSERT(!XFS_IS_PQUOTA_ON(mp));
724                                 ASSERT(mask & XFS_AT_GID);
725                                 ASSERT(gdqp);
726                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
727                                                         &ip->i_gdquot, gdqp);
728                         }
729                         ip->i_d.di_gid = gid;
730                 }
731                 if (iprojid != projid) {
732                         if (XFS_IS_PQUOTA_ON(mp)) {
733                                 ASSERT(!XFS_IS_GQUOTA_ON(mp));
734                                 ASSERT(mask & XFS_AT_PROJID);
735                                 ASSERT(gdqp);
736                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
737                                                         &ip->i_gdquot, gdqp);
738                         }
739                         ip->i_d.di_projid = projid;
740                         /*
741                          * We may have to rev the inode as well as
742                          * the superblock version number since projids didn't
743                          * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
744                          */
745                         if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
746                                 xfs_bump_ino_vers2(tp, ip);
747                 }
748
749                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
750                 timeflags |= XFS_ICHGTIME_CHG;
751         }
752
753
754         /*
755          * Change file access or modified times.
756          */
757         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
758                 if (mask & XFS_AT_ATIME) {
759                         ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
760                         ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
761                         ip->i_update_core = 1;
762                         timeflags &= ~XFS_ICHGTIME_ACC;
763                 }
764                 if (mask & XFS_AT_MTIME) {
765                         ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
766                         ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
767                         timeflags &= ~XFS_ICHGTIME_MOD;
768                         timeflags |= XFS_ICHGTIME_CHG;
769                 }
770                 if (tp && (flags & ATTR_UTIME))
771                         xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
772         }
773
774         /*
775          * Change XFS-added attributes.
776          */
777         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
778                 if (mask & XFS_AT_EXTSIZE) {
779                         /*
780                          * Converting bytes to fs blocks.
781                          */
782                         ip->i_d.di_extsize = vap->va_extsize >>
783                                 mp->m_sb.sb_blocklog;
784                 }
785                 if (mask & XFS_AT_XFLAGS) {
786                         uint    di_flags;
787
788                         /* can't set PREALLOC this way, just preserve it */
789                         di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
790                         if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
791                                 di_flags |= XFS_DIFLAG_IMMUTABLE;
792                         if (vap->va_xflags & XFS_XFLAG_APPEND)
793                                 di_flags |= XFS_DIFLAG_APPEND;
794                         if (vap->va_xflags & XFS_XFLAG_SYNC)
795                                 di_flags |= XFS_DIFLAG_SYNC;
796                         if (vap->va_xflags & XFS_XFLAG_NOATIME)
797                                 di_flags |= XFS_DIFLAG_NOATIME;
798                         if (vap->va_xflags & XFS_XFLAG_NODUMP)
799                                 di_flags |= XFS_DIFLAG_NODUMP;
800                         if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
801                                 di_flags |= XFS_DIFLAG_PROJINHERIT;
802                         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
803                                 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
804                                         di_flags |= XFS_DIFLAG_RTINHERIT;
805                                 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
806                                         di_flags |= XFS_DIFLAG_NOSYMLINKS;
807                                 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
808                                         di_flags |= XFS_DIFLAG_EXTSZINHERIT;
809                         } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
810                                 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
811                                         di_flags |= XFS_DIFLAG_REALTIME;
812                                         ip->i_iocore.io_flags |= XFS_IOCORE_RT;
813                                 } else {
814                                         ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
815                                 }
816                                 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
817                                         di_flags |= XFS_DIFLAG_EXTSIZE;
818                         }
819                         ip->i_d.di_flags = di_flags;
820                 }
821                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
822                 timeflags |= XFS_ICHGTIME_CHG;
823         }
824
825         /*
826          * Change file inode change time only if XFS_AT_CTIME set
827          * AND we have been called by a DMI function.
828          */
829
830         if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
831                 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
832                 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
833                 ip->i_update_core = 1;
834                 timeflags &= ~XFS_ICHGTIME_CHG;
835         }
836
837         /*
838          * Send out timestamp changes that need to be set to the
839          * current time.  Not done when called by a DMI function.
840          */
841         if (timeflags && !(flags & ATTR_DMI))
842                 xfs_ichgtime(ip, timeflags);
843
844         XFS_STATS_INC(xs_ig_attrchg);
845
846         /*
847          * If this is a synchronous mount, make sure that the
848          * transaction goes to disk before returning to the user.
849          * This is slightly sub-optimal in that truncates require
850          * two sync transactions instead of one for wsync filesytems.
851          * One for the truncate and one for the timestamps since we
852          * don't want to change the timestamps unless we're sure the
853          * truncate worked.  Truncates are less than 1% of the laddis
854          * mix so this probably isn't worth the trouble to optimize.
855          */
856         code = 0;
857         if (tp) {
858                 if (mp->m_flags & XFS_MOUNT_WSYNC)
859                         xfs_trans_set_sync(tp);
860
861                 code = xfs_trans_commit(tp, commit_flags, NULL);
862         }
863
864         /*
865          * If the (regular) file's mandatory locking mode changed, then
866          * notify the vnode.  We do this under the inode lock to prevent
867          * racing calls to vop_vnode_change.
868          */
869         mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
870         if (mandlock_before != mandlock_after) {
871                 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
872                                  mandlock_after);
873         }
874
875         xfs_iunlock(ip, lock_flags);
876
877         /*
878          * Release any dquot(s) the inode had kept before chown.
879          */
880         XFS_QM_DQRELE(mp, olddquot1);
881         XFS_QM_DQRELE(mp, olddquot2);
882         XFS_QM_DQRELE(mp, udqp);
883         XFS_QM_DQRELE(mp, gdqp);
884
885         if (code) {
886                 return code;
887         }
888
889         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
890             !(flags & ATTR_DMI)) {
891                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
892                                         NULL, DM_RIGHT_NULL, NULL, NULL,
893                                         0, 0, AT_DELAY_FLAG(flags));
894         }
895         return 0;
896
897  abort_return:
898         commit_flags |= XFS_TRANS_ABORT;
899         /* FALLTHROUGH */
900  error_return:
901         XFS_QM_DQRELE(mp, udqp);
902         XFS_QM_DQRELE(mp, gdqp);
903         if (tp) {
904                 xfs_trans_cancel(tp, commit_flags);
905         }
906         if (lock_flags != 0) {
907                 xfs_iunlock(ip, lock_flags);
908         }
909         return code;
910 }
911
912
913 /*
914  * xfs_access
915  * Null conversion from vnode mode bits to inode mode bits, as in efs.
916  */
917 STATIC int
918 xfs_access(
919         bhv_desc_t      *bdp,
920         int             mode,
921         cred_t          *credp)
922 {
923         xfs_inode_t     *ip;
924         int             error;
925
926         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
927                                                (inst_t *)__return_address);
928
929         ip = XFS_BHVTOI(bdp);
930         xfs_ilock(ip, XFS_ILOCK_SHARED);
931         error = xfs_iaccess(ip, mode, credp);
932         xfs_iunlock(ip, XFS_ILOCK_SHARED);
933         return error;
934 }
935
936
937 /*
938  * xfs_readlink
939  *
940  */
941 STATIC int
942 xfs_readlink(
943         bhv_desc_t      *bdp,
944         uio_t           *uiop,
945         int             ioflags,
946         cred_t          *credp)
947 {
948         xfs_inode_t     *ip;
949         int             count;
950         xfs_off_t       offset;
951         int             pathlen;
952         vnode_t         *vp;
953         int             error = 0;
954         xfs_mount_t     *mp;
955         int             nmaps;
956         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
957         xfs_daddr_t     d;
958         int             byte_cnt;
959         int             n;
960         xfs_buf_t       *bp;
961
962         vp = BHV_TO_VNODE(bdp);
963         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
964
965         ip = XFS_BHVTOI(bdp);
966         mp = ip->i_mount;
967
968         if (XFS_FORCED_SHUTDOWN(mp))
969                 return XFS_ERROR(EIO);
970
971         xfs_ilock(ip, XFS_ILOCK_SHARED);
972
973         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
974
975         offset = uiop->uio_offset;
976         count = uiop->uio_resid;
977
978         if (offset < 0) {
979                 error = XFS_ERROR(EINVAL);
980                 goto error_return;
981         }
982         if (count <= 0) {
983                 error = 0;
984                 goto error_return;
985         }
986
987         /*
988          * See if the symlink is stored inline.
989          */
990         pathlen = (int)ip->i_d.di_size;
991
992         if (ip->i_df.if_flags & XFS_IFINLINE) {
993                 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
994         }
995         else {
996                 /*
997                  * Symlink not inline.  Call bmap to get it in.
998                  */
999                 nmaps = SYMLINK_MAPS;
1000
1001                 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1002                                   0, NULL, 0, mval, &nmaps, NULL);
1003
1004                 if (error) {
1005                         goto error_return;
1006                 }
1007
1008                 for (n = 0; n < nmaps; n++) {
1009                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1010                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1011                         bp = xfs_buf_read(mp->m_ddev_targp, d,
1012                                       BTOBB(byte_cnt), 0);
1013                         error = XFS_BUF_GETERROR(bp);
1014                         if (error) {
1015                                 xfs_ioerror_alert("xfs_readlink",
1016                                           ip->i_mount, bp, XFS_BUF_ADDR(bp));
1017                                 xfs_buf_relse(bp);
1018                                 goto error_return;
1019                         }
1020                         if (pathlen < byte_cnt)
1021                                 byte_cnt = pathlen;
1022                         pathlen -= byte_cnt;
1023
1024                         error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1025                         xfs_buf_relse (bp);
1026                 }
1027
1028         }
1029
1030
1031 error_return:
1032
1033         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1034
1035         return error;
1036 }
1037
1038
1039 /*
1040  * xfs_fsync
1041  *
1042  * This is called to sync the inode and its data out to disk.
1043  * We need to hold the I/O lock while flushing the data, and
1044  * the inode lock while flushing the inode.  The inode lock CANNOT
1045  * be held while flushing the data, so acquire after we're done
1046  * with that.
1047  */
1048 STATIC int
1049 xfs_fsync(
1050         bhv_desc_t      *bdp,
1051         int             flag,
1052         cred_t          *credp,
1053         xfs_off_t       start,
1054         xfs_off_t       stop)
1055 {
1056         xfs_inode_t     *ip;
1057         xfs_trans_t     *tp;
1058         int             error;
1059         int             log_flushed = 0, changed = 1;
1060
1061         vn_trace_entry(BHV_TO_VNODE(bdp),
1062                         __FUNCTION__, (inst_t *)__return_address);
1063
1064         ip = XFS_BHVTOI(bdp);
1065
1066         ASSERT(start >= 0 && stop >= -1);
1067
1068         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1069                 return XFS_ERROR(EIO);
1070
1071         /*
1072          * We always need to make sure that the required inode state
1073          * is safe on disk.  The vnode might be clean but because
1074          * of committed transactions that haven't hit the disk yet.
1075          * Likewise, there could be unflushed non-transactional
1076          * changes to the inode core that have to go to disk.
1077          *
1078          * The following code depends on one assumption:  that
1079          * any transaction that changes an inode logs the core
1080          * because it has to change some field in the inode core
1081          * (typically nextents or nblocks).  That assumption
1082          * implies that any transactions against an inode will
1083          * catch any non-transactional updates.  If inode-altering
1084          * transactions exist that violate this assumption, the
1085          * code breaks.  Right now, it figures that if the involved
1086          * update_* field is clear and the inode is unpinned, the
1087          * inode is clean.  Either it's been flushed or it's been
1088          * committed and the commit has hit the disk unpinning the inode.
1089          * (Note that xfs_inode_item_format() called at commit clears
1090          * the update_* fields.)
1091          */
1092         xfs_ilock(ip, XFS_ILOCK_SHARED);
1093
1094         /* If we are flushing data then we care about update_size
1095          * being set, otherwise we care about update_core
1096          */
1097         if ((flag & FSYNC_DATA) ?
1098                         (ip->i_update_size == 0) :
1099                         (ip->i_update_core == 0)) {
1100                 /*
1101                  * Timestamps/size haven't changed since last inode
1102                  * flush or inode transaction commit.  That means
1103                  * either nothing got written or a transaction
1104                  * committed which caught the updates.  If the
1105                  * latter happened and the transaction hasn't
1106                  * hit the disk yet, the inode will be still
1107                  * be pinned.  If it is, force the log.
1108                  */
1109
1110                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1111
1112                 if (xfs_ipincount(ip)) {
1113                         _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1114                                       XFS_LOG_FORCE |
1115                                       ((flag & FSYNC_WAIT)
1116                                        ? XFS_LOG_SYNC : 0),
1117                                       &log_flushed);
1118                 } else {
1119                         /*
1120                          * If the inode is not pinned and nothing
1121                          * has changed we don't need to flush the
1122                          * cache.
1123                          */
1124                         changed = 0;
1125                 }
1126                 error = 0;
1127         } else  {
1128                 /*
1129                  * Kick off a transaction to log the inode
1130                  * core to get the updates.  Make it
1131                  * sync if FSYNC_WAIT is passed in (which
1132                  * is done by everybody but specfs).  The
1133                  * sync transaction will also force the log.
1134                  */
1135                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1136                 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1137                 if ((error = xfs_trans_reserve(tp, 0,
1138                                 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1139                                 0, 0, 0)))  {
1140                         xfs_trans_cancel(tp, 0);
1141                         return error;
1142                 }
1143                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1144
1145                 /*
1146                  * Note - it's possible that we might have pushed
1147                  * ourselves out of the way during trans_reserve
1148                  * which would flush the inode.  But there's no
1149                  * guarantee that the inode buffer has actually
1150                  * gone out yet (it's delwri).  Plus the buffer
1151                  * could be pinned anyway if it's part of an
1152                  * inode in another recent transaction.  So we
1153                  * play it safe and fire off the transaction anyway.
1154                  */
1155                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1156                 xfs_trans_ihold(tp, ip);
1157                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1158                 if (flag & FSYNC_WAIT)
1159                         xfs_trans_set_sync(tp);
1160                 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1161
1162                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1163         }
1164
1165         if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1166                 /*
1167                  * If the log write didn't issue an ordered tag we need
1168                  * to flush the disk cache for the data device now.
1169                  */
1170                 if (!log_flushed)
1171                         xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1172
1173                 /*
1174                  * If this inode is on the RT dev we need to flush that
1175                  * cache aswell.
1176                  */
1177                 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1178                         xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1179         }
1180
1181         return error;
1182 }
1183
1184 /*
1185  * This is called by xfs_inactive to free any blocks beyond eof,
1186  * when the link count isn't zero.
1187  */
1188 STATIC int
1189 xfs_inactive_free_eofblocks(
1190         xfs_mount_t     *mp,
1191         xfs_inode_t     *ip)
1192 {
1193         xfs_trans_t     *tp;
1194         int             error;
1195         xfs_fileoff_t   end_fsb;
1196         xfs_fileoff_t   last_fsb;
1197         xfs_filblks_t   map_len;
1198         int             nimaps;
1199         xfs_bmbt_irec_t imap;
1200
1201         /*
1202          * Figure out if there are any blocks beyond the end
1203          * of the file.  If not, then there is nothing to do.
1204          */
1205         end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1206         last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1207         map_len = last_fsb - end_fsb;
1208         if (map_len <= 0)
1209                 return (0);
1210
1211         nimaps = 1;
1212         xfs_ilock(ip, XFS_ILOCK_SHARED);
1213         error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1214                           NULL, 0, &imap, &nimaps, NULL);
1215         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1216
1217         if (!error && (nimaps != 0) &&
1218             (imap.br_startblock != HOLESTARTBLOCK ||
1219              ip->i_delayed_blks)) {
1220                 /*
1221                  * Attach the dquots to the inode up front.
1222                  */
1223                 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1224                         return (error);
1225
1226                 /*
1227                  * There are blocks after the end of file.
1228                  * Free them up now by truncating the file to
1229                  * its current size.
1230                  */
1231                 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1232
1233                 /*
1234                  * Do the xfs_itruncate_start() call before
1235                  * reserving any log space because
1236                  * itruncate_start will call into the buffer
1237                  * cache and we can't
1238                  * do that within a transaction.
1239                  */
1240                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1241                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1242                                     ip->i_d.di_size);
1243
1244                 error = xfs_trans_reserve(tp, 0,
1245                                           XFS_ITRUNCATE_LOG_RES(mp),
1246                                           0, XFS_TRANS_PERM_LOG_RES,
1247                                           XFS_ITRUNCATE_LOG_COUNT);
1248                 if (error) {
1249                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1250                         xfs_trans_cancel(tp, 0);
1251                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1252                         return (error);
1253                 }
1254
1255                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1256                 xfs_trans_ijoin(tp, ip,
1257                                 XFS_IOLOCK_EXCL |
1258                                 XFS_ILOCK_EXCL);
1259                 xfs_trans_ihold(tp, ip);
1260
1261                 error = xfs_itruncate_finish(&tp, ip,
1262                                              ip->i_d.di_size,
1263                                              XFS_DATA_FORK,
1264                                              0);
1265                 /*
1266                  * If we get an error at this point we
1267                  * simply don't bother truncating the file.
1268                  */
1269                 if (error) {
1270                         xfs_trans_cancel(tp,
1271                                          (XFS_TRANS_RELEASE_LOG_RES |
1272                                           XFS_TRANS_ABORT));
1273                 } else {
1274                         error = xfs_trans_commit(tp,
1275                                                 XFS_TRANS_RELEASE_LOG_RES,
1276                                                 NULL);
1277                 }
1278                 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1279         }
1280         return (error);
1281 }
1282
1283 /*
1284  * Free a symlink that has blocks associated with it.
1285  */
1286 STATIC int
1287 xfs_inactive_symlink_rmt(
1288         xfs_inode_t     *ip,
1289         xfs_trans_t     **tpp)
1290 {
1291         xfs_buf_t       *bp;
1292         int             committed;
1293         int             done;
1294         int             error;
1295         xfs_fsblock_t   first_block;
1296         xfs_bmap_free_t free_list;
1297         int             i;
1298         xfs_mount_t     *mp;
1299         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1300         int             nmaps;
1301         xfs_trans_t     *ntp;
1302         int             size;
1303         xfs_trans_t     *tp;
1304
1305         tp = *tpp;
1306         mp = ip->i_mount;
1307         ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1308         /*
1309          * We're freeing a symlink that has some
1310          * blocks allocated to it.  Free the
1311          * blocks here.  We know that we've got
1312          * either 1 or 2 extents and that we can
1313          * free them all in one bunmapi call.
1314          */
1315         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1316         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1317                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1318                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1319                 xfs_trans_cancel(tp, 0);
1320                 *tpp = NULL;
1321                 return error;
1322         }
1323         /*
1324          * Lock the inode, fix the size, and join it to the transaction.
1325          * Hold it so in the normal path, we still have it locked for
1326          * the second transaction.  In the error paths we need it
1327          * held so the cancel won't rele it, see below.
1328          */
1329         xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1330         size = (int)ip->i_d.di_size;
1331         ip->i_d.di_size = 0;
1332         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1333         xfs_trans_ihold(tp, ip);
1334         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1335         /*
1336          * Find the block(s) so we can inval and unmap them.
1337          */
1338         done = 0;
1339         XFS_BMAP_INIT(&free_list, &first_block);
1340         nmaps = sizeof(mval) / sizeof(mval[0]);
1341         if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1342                         XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1343                         &free_list)))
1344                 goto error0;
1345         /*
1346          * Invalidate the block(s).
1347          */
1348         for (i = 0; i < nmaps; i++) {
1349                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1350                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1351                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1352                 xfs_trans_binval(tp, bp);
1353         }
1354         /*
1355          * Unmap the dead block(s) to the free_list.
1356          */
1357         if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1358                         &first_block, &free_list, &done)))
1359                 goto error1;
1360         ASSERT(done);
1361         /*
1362          * Commit the first transaction.  This logs the EFI and the inode.
1363          */
1364         if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1365                 goto error1;
1366         /*
1367          * The transaction must have been committed, since there were
1368          * actually extents freed by xfs_bunmapi.  See xfs_bmap_finish.
1369          * The new tp has the extent freeing and EFDs.
1370          */
1371         ASSERT(committed);
1372         /*
1373          * The first xact was committed, so add the inode to the new one.
1374          * Mark it dirty so it will be logged and moved forward in the log as
1375          * part of every commit.
1376          */
1377         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1378         xfs_trans_ihold(tp, ip);
1379         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1380         /*
1381          * Get a new, empty transaction to return to our caller.
1382          */
1383         ntp = xfs_trans_dup(tp);
1384         /*
1385          * Commit the transaction containing extent freeing and EFD's.
1386          * If we get an error on the commit here or on the reserve below,
1387          * we need to unlock the inode since the new transaction doesn't
1388          * have the inode attached.
1389          */
1390         error = xfs_trans_commit(tp, 0, NULL);
1391         tp = ntp;
1392         if (error) {
1393                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1394                 goto error0;
1395         }
1396         /*
1397          * Remove the memory for extent descriptions (just bookkeeping).
1398          */
1399         if (ip->i_df.if_bytes)
1400                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1401         ASSERT(ip->i_df.if_bytes == 0);
1402         /*
1403          * Put an itruncate log reservation in the new transaction
1404          * for our caller.
1405          */
1406         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1407                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1408                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1409                 goto error0;
1410         }
1411         /*
1412          * Return with the inode locked but not joined to the transaction.
1413          */
1414         *tpp = tp;
1415         return 0;
1416
1417  error1:
1418         xfs_bmap_cancel(&free_list);
1419  error0:
1420         /*
1421          * Have to come here with the inode locked and either
1422          * (held and in the transaction) or (not in the transaction).
1423          * If the inode isn't held then cancel would iput it, but
1424          * that's wrong since this is inactive and the vnode ref
1425          * count is 0 already.
1426          * Cancel won't do anything to the inode if held, but it still
1427          * needs to be locked until the cancel is done, if it was
1428          * joined to the transaction.
1429          */
1430         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1431         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1432         *tpp = NULL;
1433         return error;
1434
1435 }
1436
1437 STATIC int
1438 xfs_inactive_symlink_local(
1439         xfs_inode_t     *ip,
1440         xfs_trans_t     **tpp)
1441 {
1442         int             error;
1443
1444         ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1445         /*
1446          * We're freeing a symlink which fit into
1447          * the inode.  Just free the memory used
1448          * to hold the old symlink.
1449          */
1450         error = xfs_trans_reserve(*tpp, 0,
1451                                   XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1452                                   0, XFS_TRANS_PERM_LOG_RES,
1453                                   XFS_ITRUNCATE_LOG_COUNT);
1454
1455         if (error) {
1456                 xfs_trans_cancel(*tpp, 0);
1457                 *tpp = NULL;
1458                 return (error);
1459         }
1460         xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1461
1462         /*
1463          * Zero length symlinks _can_ exist.
1464          */
1465         if (ip->i_df.if_bytes > 0) {
1466                 xfs_idata_realloc(ip,
1467                                   -(ip->i_df.if_bytes),
1468                                   XFS_DATA_FORK);
1469                 ASSERT(ip->i_df.if_bytes == 0);
1470         }
1471         return (0);
1472 }
1473
1474 /*
1475  *
1476  */
1477 STATIC int
1478 xfs_inactive_attrs(
1479         xfs_inode_t     *ip,
1480         xfs_trans_t     **tpp)
1481 {
1482         xfs_trans_t     *tp;
1483         int             error;
1484         xfs_mount_t     *mp;
1485
1486         ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1487         tp = *tpp;
1488         mp = ip->i_mount;
1489         ASSERT(ip->i_d.di_forkoff != 0);
1490         xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1491         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1492
1493         error = xfs_attr_inactive(ip);
1494         if (error) {
1495                 *tpp = NULL;
1496                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1497                 return (error); /* goto out*/
1498         }
1499
1500         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1501         error = xfs_trans_reserve(tp, 0,
1502                                   XFS_IFREE_LOG_RES(mp),
1503                                   0, XFS_TRANS_PERM_LOG_RES,
1504                                   XFS_INACTIVE_LOG_COUNT);
1505         if (error) {
1506                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1507                 xfs_trans_cancel(tp, 0);
1508                 *tpp = NULL;
1509                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1510                 return (error);
1511         }
1512
1513         xfs_ilock(ip, XFS_ILOCK_EXCL);
1514         xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1515         xfs_trans_ihold(tp, ip);
1516         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1517
1518         ASSERT(ip->i_d.di_anextents == 0);
1519
1520         *tpp = tp;
1521         return (0);
1522 }
1523
1524 STATIC int
1525 xfs_release(
1526         bhv_desc_t      *bdp)
1527 {
1528         xfs_inode_t     *ip;
1529         vnode_t         *vp;
1530         xfs_mount_t     *mp;
1531         int             error;
1532
1533         vp = BHV_TO_VNODE(bdp);
1534         ip = XFS_BHVTOI(bdp);
1535
1536         if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1537                 return 0;
1538         }
1539
1540         /* If this is a read-only mount, don't do this (would generate I/O) */
1541         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1542                 return 0;
1543
1544 #ifdef HAVE_REFCACHE
1545         /* If we are in the NFS reference cache then don't do this now */
1546         if (ip->i_refcache)
1547                 return 0;
1548 #endif
1549
1550         mp = ip->i_mount;
1551
1552         if (ip->i_d.di_nlink != 0) {
1553                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1554                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1555                        ip->i_delayed_blks > 0)) &&
1556                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1557                     (!(ip->i_d.di_flags &
1558                                 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1559                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1560                                 return (error);
1561                         /* Update linux inode block count after free above */
1562                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1563                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1564                 }
1565         }
1566
1567         return 0;
1568 }
1569
1570 /*
1571  * xfs_inactive
1572  *
1573  * This is called when the vnode reference count for the vnode
1574  * goes to zero.  If the file has been unlinked, then it must
1575  * now be truncated.  Also, we clear all of the read-ahead state
1576  * kept for the inode here since the file is now closed.
1577  */
1578 STATIC int
1579 xfs_inactive(
1580         bhv_desc_t      *bdp,
1581         cred_t          *credp)
1582 {
1583         xfs_inode_t     *ip;
1584         vnode_t         *vp;
1585         xfs_bmap_free_t free_list; 
1586         xfs_fsblock_t   first_block;
1587         int             committed;
1588         xfs_trans_t     *tp;
1589         xfs_mount_t     *mp;
1590         int             error;
1591         int             truncate;
1592
1593         vp = BHV_TO_VNODE(bdp);
1594         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1595
1596         ip = XFS_BHVTOI(bdp);
1597
1598         /*
1599          * If the inode is already free, then there can be nothing
1600          * to clean up here.
1601          */
1602         if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1603                 ASSERT(ip->i_df.if_real_bytes == 0);
1604                 ASSERT(ip->i_df.if_broot_bytes == 0);
1605                 return VN_INACTIVE_CACHE;
1606         }
1607
1608         /*
1609          * Only do a truncate if it's a regular file with
1610          * some actual space in it.  It's OK to look at the
1611          * inode's fields without the lock because we're the
1612          * only one with a reference to the inode.
1613          */
1614         truncate = ((ip->i_d.di_nlink == 0) &&
1615             ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1616              (ip->i_delayed_blks > 0)) &&
1617             ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1618
1619         mp = ip->i_mount;
1620
1621         if (ip->i_d.di_nlink == 0 &&
1622             DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1623                 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1624         }
1625
1626         error = 0;
1627
1628         /* If this is a read-only mount, don't do this (would generate I/O) */
1629         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1630                 goto out;
1631
1632         if (ip->i_d.di_nlink != 0) {
1633                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1634                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1635                        ip->i_delayed_blks > 0)) &&
1636                       (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1637                      (!(ip->i_d.di_flags &
1638                                 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1639                       (ip->i_delayed_blks != 0)))) {
1640                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1641                                 return (VN_INACTIVE_CACHE);
1642                         /* Update linux inode block count after free above */
1643                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1644                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1645                 }
1646                 goto out;
1647         }
1648
1649         ASSERT(ip->i_d.di_nlink == 0);
1650
1651         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1652                 return (VN_INACTIVE_CACHE);
1653
1654         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1655         if (truncate) {
1656                 /*
1657                  * Do the xfs_itruncate_start() call before
1658                  * reserving any log space because itruncate_start
1659                  * will call into the buffer cache and we can't
1660                  * do that within a transaction.
1661                  */
1662                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1663
1664                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1665
1666                 error = xfs_trans_reserve(tp, 0,
1667                                           XFS_ITRUNCATE_LOG_RES(mp),
1668                                           0, XFS_TRANS_PERM_LOG_RES,
1669                                           XFS_ITRUNCATE_LOG_COUNT);
1670                 if (error) {
1671                         /* Don't call itruncate_cleanup */
1672                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1673                         xfs_trans_cancel(tp, 0);
1674                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1675                         return (VN_INACTIVE_CACHE);
1676                 }
1677
1678                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1679                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1680                 xfs_trans_ihold(tp, ip);
1681
1682                 /*
1683                  * normally, we have to run xfs_itruncate_finish sync.
1684                  * But if filesystem is wsync and we're in the inactive
1685                  * path, then we know that nlink == 0, and that the
1686                  * xaction that made nlink == 0 is permanently committed
1687                  * since xfs_remove runs as a synchronous transaction.
1688                  */
1689                 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1690                                 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1691
1692                 if (error) {
1693                         xfs_trans_cancel(tp,
1694                                 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1695                         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1696                         return (VN_INACTIVE_CACHE);
1697                 }
1698         } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1699
1700                 /*
1701                  * If we get an error while cleaning up a
1702                  * symlink we bail out.
1703                  */
1704                 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1705                         xfs_inactive_symlink_rmt(ip, &tp) :
1706                         xfs_inactive_symlink_local(ip, &tp);
1707
1708                 if (error) {
1709                         ASSERT(tp == NULL);
1710                         return (VN_INACTIVE_CACHE);
1711                 }
1712
1713                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1714                 xfs_trans_ihold(tp, ip);
1715         } else {
1716                 error = xfs_trans_reserve(tp, 0,
1717                                           XFS_IFREE_LOG_RES(mp),
1718                                           0, XFS_TRANS_PERM_LOG_RES,
1719                                           XFS_INACTIVE_LOG_COUNT);
1720                 if (error) {
1721                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1722                         xfs_trans_cancel(tp, 0);
1723                         return (VN_INACTIVE_CACHE);
1724                 }
1725
1726                 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1727                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1728                 xfs_trans_ihold(tp, ip);
1729         }
1730
1731         /*
1732          * If there are attributes associated with the file
1733          * then blow them away now.  The code calls a routine
1734          * that recursively deconstructs the attribute fork.
1735          * We need to just commit the current transaction
1736          * because we can't use it for xfs_attr_inactive().
1737          */
1738         if (ip->i_d.di_anextents > 0) {
1739                 error = xfs_inactive_attrs(ip, &tp);
1740                 /*
1741                  * If we got an error, the transaction is already
1742                  * cancelled, and the inode is unlocked. Just get out.
1743                  */
1744                  if (error)
1745                          return (VN_INACTIVE_CACHE);
1746         } else if (ip->i_afp) {
1747                 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1748         }
1749
1750         /*
1751          * Free the inode.
1752          */
1753         XFS_BMAP_INIT(&free_list, &first_block);
1754         error = xfs_ifree(tp, ip, &free_list);
1755         if (error) {
1756                 /*
1757                  * If we fail to free the inode, shut down.  The cancel
1758                  * might do that, we need to make sure.  Otherwise the
1759                  * inode might be lost for a long time or forever.
1760                  */
1761                 if (!XFS_FORCED_SHUTDOWN(mp)) {
1762                         cmn_err(CE_NOTE,
1763                 "xfs_inactive:  xfs_ifree() returned an error = %d on %s",
1764                                 error, mp->m_fsname);
1765                         xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1766                 }
1767                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1768         } else {
1769                 /*
1770                  * Credit the quota account(s). The inode is gone.
1771                  */
1772                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1773
1774                 /*
1775                  * Just ignore errors at this point.  There is
1776                  * nothing we can do except to try to keep going.
1777                  */
1778                 (void) xfs_bmap_finish(&tp,  &free_list, first_block,
1779                                        &committed);
1780                 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1781         }
1782         /*
1783          * Release the dquots held by inode, if any.
1784          */
1785         XFS_QM_DQDETACH(mp, ip);
1786
1787         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1788
1789  out:
1790         return VN_INACTIVE_CACHE;
1791 }
1792
1793
1794 /*
1795  * xfs_lookup
1796  */
1797 STATIC int
1798 xfs_lookup(
1799         bhv_desc_t              *dir_bdp,
1800         vname_t                 *dentry,
1801         vnode_t                 **vpp,
1802         int                     flags,
1803         vnode_t                 *rdir,
1804         cred_t                  *credp)
1805 {
1806         xfs_inode_t             *dp, *ip;
1807         xfs_ino_t               e_inum;
1808         int                     error;
1809         uint                    lock_mode;
1810         vnode_t                 *dir_vp;
1811
1812         dir_vp = BHV_TO_VNODE(dir_bdp);
1813         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1814
1815         dp = XFS_BHVTOI(dir_bdp);
1816
1817         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1818                 return XFS_ERROR(EIO);
1819
1820         lock_mode = xfs_ilock_map_shared(dp);
1821         error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1822         if (!error) {
1823                 *vpp = XFS_ITOV(ip);
1824                 ITRACE(ip);
1825         }
1826         xfs_iunlock_map_shared(dp, lock_mode);
1827         return error;
1828 }
1829
1830
1831 /*
1832  * xfs_create (create a new file).
1833  */
1834 STATIC int
1835 xfs_create(
1836         bhv_desc_t              *dir_bdp,
1837         vname_t                 *dentry,
1838         vattr_t                 *vap,
1839         vnode_t                 **vpp,
1840         cred_t                  *credp)
1841 {
1842         char                    *name = VNAME(dentry);
1843         vnode_t                 *dir_vp;
1844         xfs_inode_t             *dp, *ip;
1845         vnode_t                 *vp=NULL;
1846         xfs_trans_t             *tp;
1847         xfs_mount_t             *mp;
1848         xfs_dev_t               rdev;
1849         int                     error;
1850         xfs_bmap_free_t         free_list;
1851         xfs_fsblock_t           first_block;
1852         boolean_t               dp_joined_to_trans;
1853         int                     dm_event_sent = 0;
1854         uint                    cancel_flags;
1855         int                     committed;
1856         xfs_prid_t              prid;
1857         struct xfs_dquot        *udqp, *gdqp;
1858         uint                    resblks;
1859         int                     dm_di_mode;
1860         int                     namelen;
1861
1862         ASSERT(!*vpp);
1863         dir_vp = BHV_TO_VNODE(dir_bdp);
1864         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1865
1866         dp = XFS_BHVTOI(dir_bdp);
1867         mp = dp->i_mount;
1868
1869         dm_di_mode = vap->va_mode;
1870         namelen = VNAMELEN(dentry);
1871
1872         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1873                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1874                                 dir_vp, DM_RIGHT_NULL, NULL,
1875                                 DM_RIGHT_NULL, name, NULL,
1876                                 dm_di_mode, 0, 0);
1877
1878                 if (error)
1879                         return error;
1880                 dm_event_sent = 1;
1881         }
1882
1883         if (XFS_FORCED_SHUTDOWN(mp))
1884                 return XFS_ERROR(EIO);
1885
1886         /* Return through std_return after this point. */
1887
1888         udqp = gdqp = NULL;
1889         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1890                 prid = dp->i_d.di_projid;
1891         else if (vap->va_mask & XFS_AT_PROJID)
1892                 prid = (xfs_prid_t)vap->va_projid;
1893         else
1894                 prid = (xfs_prid_t)dfltprid;
1895
1896         /*
1897          * Make sure that we have allocated dquot(s) on disk.
1898          */
1899         error = XFS_QM_DQVOPALLOC(mp, dp,
1900                         current_fsuid(credp), current_fsgid(credp), prid,
1901                         XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1902         if (error)
1903                 goto std_return;
1904
1905         ip = NULL;
1906         dp_joined_to_trans = B_FALSE;
1907
1908         tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1909         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1910         resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1911         /*
1912          * Initially assume that the file does not exist and
1913          * reserve the resources for that case.  If that is not
1914          * the case we'll drop the one we have and get a more
1915          * appropriate transaction later.
1916          */
1917         error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1918                         XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1919         if (error == ENOSPC) {
1920                 resblks = 0;
1921                 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1922                                 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1923         }
1924         if (error) {
1925                 cancel_flags = 0;
1926                 dp = NULL;
1927                 goto error_return;
1928         }
1929
1930         xfs_ilock(dp, XFS_ILOCK_EXCL);
1931
1932         XFS_BMAP_INIT(&free_list, &first_block);
1933
1934         ASSERT(ip == NULL);
1935
1936         /*
1937          * Reserve disk quota and the inode.
1938          */
1939         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1940         if (error)
1941                 goto error_return;
1942
1943         if (resblks == 0 &&
1944             (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1945                 goto error_return;
1946         rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1947         error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1948                         rdev, credp, prid, resblks > 0,
1949                         &ip, &committed);
1950         if (error) {
1951                 if (error == ENOSPC)
1952                         goto error_return;
1953                 goto abort_return;
1954         }
1955         ITRACE(ip);
1956
1957         /*
1958          * At this point, we've gotten a newly allocated inode.
1959          * It is locked (and joined to the transaction).
1960          */
1961
1962         ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1963
1964         /*
1965          * Now we join the directory inode to the transaction.
1966          * We do not do it earlier because xfs_dir_ialloc
1967          * might commit the previous transaction (and release
1968          * all the locks).
1969          */
1970
1971         VN_HOLD(dir_vp);
1972         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1973         dp_joined_to_trans = B_TRUE;
1974
1975         error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1976                 &first_block, &free_list,
1977                 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1978         if (error) {
1979                 ASSERT(error != ENOSPC);
1980                 goto abort_return;
1981         }
1982         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1983         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1984
1985         /*
1986          * If this is a synchronous mount, make sure that the
1987          * create transaction goes to disk before returning to
1988          * the user.
1989          */
1990         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1991                 xfs_trans_set_sync(tp);
1992         }
1993
1994         dp->i_gen++;
1995
1996         /*
1997          * Attach the dquot(s) to the inodes and modify them incore.
1998          * These ids of the inode couldn't have changed since the new
1999          * inode has been locked ever since it was created.
2000          */
2001         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2002
2003         /*
2004          * xfs_trans_commit normally decrements the vnode ref count
2005          * when it unlocks the inode. Since we want to return the
2006          * vnode to the caller, we bump the vnode ref count now.
2007          */
2008         IHOLD(ip);
2009         vp = XFS_ITOV(ip);
2010
2011         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2012         if (error) {
2013                 xfs_bmap_cancel(&free_list);
2014                 goto abort_rele;
2015         }
2016
2017         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2018         if (error) {
2019                 IRELE(ip);
2020                 tp = NULL;
2021                 goto error_return;
2022         }
2023
2024         XFS_QM_DQRELE(mp, udqp);
2025         XFS_QM_DQRELE(mp, gdqp);
2026
2027         /*
2028          * Propogate the fact that the vnode changed after the
2029          * xfs_inode locks have been released.
2030          */
2031         VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2032
2033         *vpp = vp;
2034
2035         /* Fallthrough to std_return with error = 0  */
2036
2037 std_return:
2038         if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2039                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2040                                                         DM_EVENT_POSTCREATE)) {
2041                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2042                         dir_vp, DM_RIGHT_NULL,
2043                         *vpp ? vp:NULL,
2044                         DM_RIGHT_NULL, name, NULL,
2045                         dm_di_mode, error, 0);
2046         }
2047         return error;
2048
2049  abort_return:
2050         cancel_flags |= XFS_TRANS_ABORT;
2051         /* FALLTHROUGH */
2052  error_return:
2053
2054         if (tp != NULL)
2055                 xfs_trans_cancel(tp, cancel_flags);
2056
2057         if (!dp_joined_to_trans && (dp != NULL))
2058                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2059         XFS_QM_DQRELE(mp, udqp);
2060         XFS_QM_DQRELE(mp, gdqp);
2061
2062         goto std_return;
2063
2064  abort_rele:
2065         /*
2066          * Wait until after the current transaction is aborted to
2067          * release the inode.  This prevents recursive transactions
2068          * and deadlocks from xfs_inactive.
2069          */
2070         cancel_flags |= XFS_TRANS_ABORT;
2071         xfs_trans_cancel(tp, cancel_flags);
2072         IRELE(ip);
2073
2074         XFS_QM_DQRELE(mp, udqp);
2075         XFS_QM_DQRELE(mp, gdqp);
2076
2077         goto std_return;
2078 }
2079
2080 #ifdef DEBUG
2081 /*
2082  * Some counters to see if (and how often) we are hitting some deadlock
2083  * prevention code paths.
2084  */
2085
2086 int xfs_rm_locks;
2087 int xfs_rm_lock_delays;
2088 int xfs_rm_attempts;
2089 #endif
2090
2091 /*
2092  * The following routine will lock the inodes associated with the
2093  * directory and the named entry in the directory. The locks are
2094  * acquired in increasing inode number.
2095  *
2096  * If the entry is "..", then only the directory is locked. The
2097  * vnode ref count will still include that from the .. entry in
2098  * this case.
2099  *
2100  * There is a deadlock we need to worry about. If the locked directory is
2101  * in the AIL, it might be blocking up the log. The next inode we lock
2102  * could be already locked by another thread waiting for log space (e.g
2103  * a permanent log reservation with a long running transaction (see
2104  * xfs_itruncate_finish)). To solve this, we must check if the directory
2105  * is in the ail and use lock_nowait. If we can't lock, we need to
2106  * drop the inode lock on the directory and try again. xfs_iunlock will
2107  * potentially push the tail if we were holding up the log.
2108  */
2109 STATIC int
2110 xfs_lock_dir_and_entry(
2111         xfs_inode_t     *dp,
2112         vname_t         *dentry,
2113         xfs_inode_t     *ip)    /* inode of entry 'name' */
2114 {
2115         int             attempts;
2116         xfs_ino_t       e_inum;
2117         xfs_inode_t     *ips[2];
2118         xfs_log_item_t  *lp;
2119
2120 #ifdef DEBUG
2121         xfs_rm_locks++;
2122 #endif
2123         attempts = 0;
2124
2125 again:
2126         xfs_ilock(dp, XFS_ILOCK_EXCL);
2127
2128         e_inum = ip->i_ino;
2129
2130         ITRACE(ip);
2131
2132         /*
2133          * We want to lock in increasing inum. Since we've already
2134          * acquired the lock on the directory, we may need to release
2135          * if if the inum of the entry turns out to be less.
2136          */
2137         if (e_inum > dp->i_ino) {
2138                 /*
2139                  * We are already in the right order, so just
2140                  * lock on the inode of the entry.
2141                  * We need to use nowait if dp is in the AIL.
2142                  */
2143
2144                 lp = (xfs_log_item_t *)dp->i_itemp;
2145                 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2146                         if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2147                                 attempts++;
2148 #ifdef DEBUG
2149                                 xfs_rm_attempts++;
2150 #endif
2151
2152                                 /*
2153                                  * Unlock dp and try again.
2154                                  * xfs_iunlock will try to push the tail
2155                                  * if the inode is in the AIL.
2156                                  */
2157
2158                                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2159
2160                                 if ((attempts % 5) == 0) {
2161                                         delay(1); /* Don't just spin the CPU */
2162 #ifdef DEBUG
2163                                         xfs_rm_lock_delays++;
2164 #endif
2165                                 }
2166                                 goto again;
2167                         }
2168                 } else {
2169                         xfs_ilock(ip, XFS_ILOCK_EXCL);
2170                 }
2171         } else if (e_inum < dp->i_ino) {
2172                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2173
2174                 ips[0] = ip;
2175                 ips[1] = dp;
2176                 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2177         }
2178         /* else  e_inum == dp->i_ino */
2179         /*     This can happen if we're asked to lock /x/..
2180          *     the entry is "..", which is also the parent directory.
2181          */
2182
2183         return 0;
2184 }
2185
2186 #ifdef DEBUG
2187 int xfs_locked_n;
2188 int xfs_small_retries;
2189 int xfs_middle_retries;
2190 int xfs_lots_retries;
2191 int xfs_lock_delays;
2192 #endif
2193
2194 /*
2195  * The following routine will lock n inodes in exclusive mode.
2196  * We assume the caller calls us with the inodes in i_ino order.
2197  *
2198  * We need to detect deadlock where an inode that we lock
2199  * is in the AIL and we start waiting for another inode that is locked
2200  * by a thread in a long running transaction (such as truncate). This can
2201  * result in deadlock since the long running trans might need to wait
2202  * for the inode we just locked in order to push the tail and free space
2203  * in the log.
2204  */
2205 void
2206 xfs_lock_inodes(
2207         xfs_inode_t     **ips,
2208         int             inodes,
2209         int             first_locked,
2210         uint            lock_mode)
2211 {
2212         int             attempts = 0, i, j, try_lock;
2213         xfs_log_item_t  *lp;
2214
2215         ASSERT(ips && (inodes >= 2)); /* we need at least two */
2216
2217         if (first_locked) {
2218                 try_lock = 1;
2219                 i = 1;
2220         } else {
2221                 try_lock = 0;
2222                 i = 0;
2223         }
2224
2225 again:
2226         for (; i < inodes; i++) {
2227                 ASSERT(ips[i]);
2228
2229                 if (i && (ips[i] == ips[i-1]))  /* Already locked */
2230                         continue;
2231
2232                 /*
2233                  * If try_lock is not set yet, make sure all locked inodes
2234                  * are not in the AIL.
2235                  * If any are, set try_lock to be used later.
2236                  */
2237
2238                 if (!try_lock) {
2239                         for (j = (i - 1); j >= 0 && !try_lock; j--) {
2240                                 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2241                                 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2242                                         try_lock++;
2243                                 }
2244                         }
2245                 }
2246
2247                 /*
2248                  * If any of the previous locks we have locked is in the AIL,
2249                  * we must TRY to get the second and subsequent locks. If
2250                  * we can't get any, we must release all we have
2251                  * and try again.
2252                  */
2253
2254                 if (try_lock) {
2255                         /* try_lock must be 0 if i is 0. */
2256                         /*
2257                          * try_lock means we have an inode locked
2258                          * that is in the AIL.
2259                          */
2260                         ASSERT(i != 0);
2261                         if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2262                                 attempts++;
2263
2264                                 /*
2265                                  * Unlock all previous guys and try again.
2266                                  * xfs_iunlock will try to push the tail
2267                                  * if the inode is in the AIL.
2268                                  */
2269
2270                                 for(j = i - 1; j >= 0; j--) {
2271
2272                                         /*
2273                                          * Check to see if we've already
2274                                          * unlocked this one.
2275                                          * Not the first one going back,
2276                                          * and the inode ptr is the same.
2277                                          */
2278                                         if ((j != (i - 1)) && ips[j] ==
2279                                                                 ips[j+1])
2280                                                 continue;
2281
2282                                         xfs_iunlock(ips[j], lock_mode);
2283                                 }
2284
2285                                 if ((attempts % 5) == 0) {
2286                                         delay(1); /* Don't just spin the CPU */
2287 #ifdef DEBUG
2288                                         xfs_lock_delays++;
2289 #endif
2290                                 }
2291                                 i = 0;
2292                                 try_lock = 0;
2293                                 goto again;
2294                         }
2295                 } else {
2296                         xfs_ilock(ips[i], lock_mode);
2297                 }
2298         }
2299
2300 #ifdef DEBUG
2301         if (attempts) {
2302                 if (attempts < 5) xfs_small_retries++;
2303                 else if (attempts < 100) xfs_middle_retries++;
2304                 else xfs_lots_retries++;
2305         } else {
2306                 xfs_locked_n++;
2307         }
2308 #endif
2309 }
2310
2311 #ifdef  DEBUG
2312 #define REMOVE_DEBUG_TRACE(x)   {remove_which_error_return = (x);}
2313 int remove_which_error_return = 0;
2314 #else /* ! DEBUG */
2315 #define REMOVE_DEBUG_TRACE(x)
2316 #endif  /* ! DEBUG */
2317
2318
2319 /*
2320  * xfs_remove
2321  *
2322  */
2323 STATIC int
2324 xfs_remove(
2325         bhv_desc_t              *dir_bdp,
2326         vname_t                 *dentry,
2327         cred_t                  *credp)
2328 {
2329         vnode_t                 *dir_vp;
2330         char                    *name = VNAME(dentry);
2331         xfs_inode_t             *dp, *ip;
2332         xfs_trans_t             *tp = NULL;
2333         xfs_mount_t             *mp;
2334         int                     error = 0;
2335         xfs_bmap_free_t         free_list;
2336         xfs_fsblock_t           first_block;
2337         int                     cancel_flags;
2338         int                     committed;
2339         int                     dm_di_mode = 0;
2340         int                     link_zero;
2341         uint                    resblks;
2342         int                     namelen;
2343
2344         dir_vp = BHV_TO_VNODE(dir_bdp);
2345         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2346
2347         dp = XFS_BHVTOI(dir_bdp);
2348         mp = dp->i_mount;
2349
2350         if (XFS_FORCED_SHUTDOWN(mp))
2351                 return XFS_ERROR(EIO);
2352
2353         namelen = VNAMELEN(dentry);
2354
2355         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2356                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2357                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2358                                         name, NULL, 0, 0, 0);
2359                 if (error)
2360                         return error;
2361         }
2362
2363         /* From this point on, return through std_return */
2364         ip = NULL;
2365
2366         /*
2367          * We need to get a reference to ip before we get our log
2368          * reservation. The reason for this is that we cannot call
2369          * xfs_iget for an inode for which we do not have a reference
2370          * once we've acquired a log reservation. This is because the
2371          * inode we are trying to get might be in xfs_inactive going
2372          * for a log reservation. Since we'll have to wait for the
2373          * inactive code to complete before returning from xfs_iget,
2374          * we need to make sure that we don't have log space reserved
2375          * when we call xfs_iget.  Instead we get an unlocked referece
2376          * to the inode before getting our log reservation.
2377          */
2378         error = xfs_get_dir_entry(dentry, &ip);
2379         if (error) {
2380                 REMOVE_DEBUG_TRACE(__LINE__);
2381                 goto std_return;
2382         }
2383
2384         dm_di_mode = ip->i_d.di_mode;
2385
2386         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2387
2388         ITRACE(ip);
2389
2390         error = XFS_QM_DQATTACH(mp, dp, 0);
2391         if (!error && dp != ip)
2392                 error = XFS_QM_DQATTACH(mp, ip, 0);
2393         if (error) {
2394                 REMOVE_DEBUG_TRACE(__LINE__);
2395                 IRELE(ip);
2396                 goto std_return;
2397         }
2398
2399         tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2400         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2401         /*
2402          * We try to get the real space reservation first,
2403          * allowing for directory btree deletion(s) implying
2404          * possible bmap insert(s).  If we can't get the space
2405          * reservation then we use 0 instead, and avoid the bmap
2406          * btree insert(s) in the directory code by, if the bmap
2407          * insert tries to happen, instead trimming the LAST
2408          * block from the directory.
2409          */
2410         resblks = XFS_REMOVE_SPACE_RES(mp);
2411         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2412                         XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2413         if (error == ENOSPC) {
2414                 resblks = 0;
2415                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2416                                 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2417         }
2418         if (error) {
2419                 ASSERT(error != ENOSPC);
2420                 REMOVE_DEBUG_TRACE(__LINE__);
2421                 xfs_trans_cancel(tp, 0);
2422                 IRELE(ip);
2423                 return error;
2424         }
2425
2426         error = xfs_lock_dir_and_entry(dp, dentry, ip);
2427         if (error) {
2428                 REMOVE_DEBUG_TRACE(__LINE__);
2429                 xfs_trans_cancel(tp, cancel_flags);
2430                 IRELE(ip);
2431                 goto std_return;
2432         }
2433
2434         /*
2435          * At this point, we've gotten both the directory and the entry
2436          * inodes locked.
2437          */
2438         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2439         if (dp != ip) {
2440                 /*
2441                  * Increment vnode ref count only in this case since
2442                  * there's an extra vnode reference in the case where
2443                  * dp == ip.
2444                  */
2445                 IHOLD(dp);
2446                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2447         }
2448
2449         /*
2450          * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2451          */
2452         XFS_BMAP_INIT(&free_list, &first_block);
2453         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2454                 &first_block, &free_list, 0);
2455         if (error) {
2456                 ASSERT(error != ENOENT);
2457                 REMOVE_DEBUG_TRACE(__LINE__);
2458                 goto error1;
2459         }
2460         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2461
2462         dp->i_gen++;
2463         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2464
2465         error = xfs_droplink(tp, ip);
2466         if (error) {
2467                 REMOVE_DEBUG_TRACE(__LINE__);
2468                 goto error1;
2469         }
2470
2471         /* Determine if this is the last link while
2472          * we are in the transaction.
2473          */
2474         link_zero = (ip)->i_d.di_nlink==0;
2475
2476         /*
2477          * Take an extra ref on the inode so that it doesn't
2478          * go to xfs_inactive() from within the commit.
2479          */
2480         IHOLD(ip);
2481
2482         /*
2483          * If this is a synchronous mount, make sure that the
2484          * remove transaction goes to disk before returning to
2485          * the user.
2486          */
2487         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2488                 xfs_trans_set_sync(tp);
2489         }
2490
2491         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2492         if (error) {
2493                 REMOVE_DEBUG_TRACE(__LINE__);
2494                 goto error_rele;
2495         }
2496
2497         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2498         if (error) {
2499                 IRELE(ip);
2500                 goto std_return;
2501         }
2502
2503         /*
2504          * Before we drop our extra reference to the inode, purge it
2505          * from the refcache if it is there.  By waiting until afterwards
2506          * to do the IRELE, we ensure that we won't go inactive in the
2507          * xfs_refcache_purge_ip routine (although that would be OK).
2508          */
2509         xfs_refcache_purge_ip(ip);
2510
2511         vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2512
2513         /*
2514          * Let interposed file systems know about removed links.
2515          */
2516         VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2517
2518         IRELE(ip);
2519
2520 /*      Fall through to std_return with error = 0 */
2521  std_return:
2522         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2523                                                 DM_EVENT_POSTREMOVE)) {
2524                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2525                                 dir_vp, DM_RIGHT_NULL,
2526                                 NULL, DM_RIGHT_NULL,
2527                                 name, NULL, dm_di_mode, error, 0);
2528         }
2529         return error;
2530
2531  error1:
2532         xfs_bmap_cancel(&free_list);
2533         cancel_flags |= XFS_TRANS_ABORT;
2534         xfs_trans_cancel(tp, cancel_flags);
2535         goto std_return;
2536
2537  error_rele:
2538         /*
2539          * In this case make sure to not release the inode until after
2540          * the current transaction is aborted.  Releasing it beforehand
2541          * can cause us to go to xfs_inactive and start a recursive
2542          * transaction which can easily deadlock with the current one.
2543          */
2544         xfs_bmap_cancel(&free_list);
2545         cancel_flags |= XFS_TRANS_ABORT;
2546         xfs_trans_cancel(tp, cancel_flags);
2547
2548         /*
2549          * Before we drop our extra reference to the inode, purge it
2550          * from the refcache if it is there.  By waiting until afterwards
2551          * to do the IRELE, we ensure that we won't go inactive in the
2552          * xfs_refcache_purge_ip routine (although that would be OK).
2553          */
2554         xfs_refcache_purge_ip(ip);
2555
2556         IRELE(ip);
2557
2558         goto std_return;
2559 }
2560
2561
2562 /*
2563  * xfs_link
2564  *
2565  */
2566 STATIC int
2567 xfs_link(
2568         bhv_desc_t              *target_dir_bdp,
2569         vnode_t                 *src_vp,
2570         vname_t                 *dentry,
2571         cred_t                  *credp)
2572 {
2573         xfs_inode_t             *tdp, *sip;
2574         xfs_trans_t             *tp;
2575         xfs_mount_t             *mp;
2576         xfs_inode_t             *ips[2];
2577         int                     error;
2578         xfs_bmap_free_t         free_list;
2579         xfs_fsblock_t           first_block;
2580         int                     cancel_flags;
2581         int                     committed;
2582         vnode_t                 *target_dir_vp;
2583         int                     resblks;
2584         char                    *target_name = VNAME(dentry);
2585         int                     target_namelen;
2586
2587         target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2588         vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2589         vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2590
2591         target_namelen = VNAMELEN(dentry);
2592         if (VN_ISDIR(src_vp))
2593                 return XFS_ERROR(EPERM);
2594
2595         sip = xfs_vtoi(src_vp);
2596         tdp = XFS_BHVTOI(target_dir_bdp);
2597         mp = tdp->i_mount;
2598         if (XFS_FORCED_SHUTDOWN(mp))
2599                 return XFS_ERROR(EIO);
2600
2601         if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2602                 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2603                                         target_dir_vp, DM_RIGHT_NULL,
2604                                         src_vp, DM_RIGHT_NULL,
2605                                         target_name, NULL, 0, 0, 0);
2606                 if (error)
2607                         return error;
2608         }
2609
2610         /* Return through std_return after this point. */
2611
2612         error = XFS_QM_DQATTACH(mp, sip, 0);
2613         if (!error && sip != tdp)
2614                 error = XFS_QM_DQATTACH(mp, tdp, 0);
2615         if (error)
2616                 goto std_return;
2617
2618         tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2619         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2620         resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2621         error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2622                         XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2623         if (error == ENOSPC) {
2624                 resblks = 0;
2625                 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2626                                 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2627         }
2628         if (error) {
2629                 cancel_flags = 0;
2630                 goto error_return;
2631         }
2632
2633         if (sip->i_ino < tdp->i_ino) {
2634                 ips[0] = sip;
2635                 ips[1] = tdp;
2636         } else {
2637                 ips[0] = tdp;
2638                 ips[1] = sip;
2639         }
2640
2641         xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2642
2643         /*
2644          * Increment vnode ref counts since xfs_trans_commit &
2645          * xfs_trans_cancel will both unlock the inodes and
2646          * decrement the associated ref counts.
2647          */
2648         VN_HOLD(src_vp);
2649         VN_HOLD(target_dir_vp);
2650         xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2651         xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2652
2653         /*
2654          * If the source has too many links, we can't make any more to it.
2655          */
2656         if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2657                 error = XFS_ERROR(EMLINK);
2658                 goto error_return;
2659         }
2660
2661         /*
2662          * If we are using project inheritance, we only allow hard link
2663          * creation in our tree when the project IDs are the same; else
2664          * the tree quota mechanism could be circumvented.
2665          */
2666         if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2667                      (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2668                 error = XFS_ERROR(EPERM);
2669                 goto error_return;
2670         }
2671
2672         if (resblks == 0 &&
2673             (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2674                         target_namelen)))
2675                 goto error_return;
2676
2677         XFS_BMAP_INIT(&free_list, &first_block);
2678
2679         error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2680                                    sip->i_ino, &first_block, &free_list,
2681                                    resblks);
2682         if (error)
2683                 goto abort_return;
2684         xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2685         tdp->i_gen++;
2686         xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2687
2688         error = xfs_bumplink(tp, sip);
2689         if (error) {
2690                 goto abort_return;
2691         }
2692
2693         /*
2694          * If this is a synchronous mount, make sure that the
2695          * link transaction goes to disk before returning to
2696          * the user.
2697          */
2698         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2699                 xfs_trans_set_sync(tp);
2700         }
2701
2702         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2703         if (error) {
2704                 xfs_bmap_cancel(&free_list);
2705                 goto abort_return;
2706         }
2707
2708         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2709         if (error) {
2710                 goto std_return;
2711         }
2712
2713         /* Fall through to std_return with error = 0. */
2714 std_return:
2715         if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2716                                                 DM_EVENT_POSTLINK)) {
2717                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2718                                 target_dir_vp, DM_RIGHT_NULL,
2719                                 src_vp, DM_RIGHT_NULL,
2720                                 target_name, NULL, 0, error, 0);
2721         }
2722         return error;
2723
2724  abort_return:
2725         cancel_flags |= XFS_TRANS_ABORT;
2726         /* FALLTHROUGH */
2727  error_return:
2728         xfs_trans_cancel(tp, cancel_flags);
2729
2730         goto std_return;
2731 }
2732 /*
2733  * xfs_mkdir
2734  *
2735  */
2736 STATIC int
2737 xfs_mkdir(
2738         bhv_desc_t              *dir_bdp,
2739         vname_t                 *dentry,
2740         vattr_t                 *vap,
2741         vnode_t                 **vpp,
2742         cred_t                  *credp)
2743 {
2744         char                    *dir_name = VNAME(dentry);
2745         xfs_inode_t             *dp;
2746         xfs_inode_t             *cdp;   /* inode of created dir */
2747         vnode_t                 *cvp;   /* vnode of created dir */
2748         xfs_trans_t             *tp;
2749         xfs_mount_t             *mp;
2750         int                     cancel_flags;
2751         int                     error;
2752         int                     committed;
2753         xfs_bmap_free_t         free_list;
2754         xfs_fsblock_t           first_block;
2755         vnode_t                 *dir_vp;
2756         boolean_t               dp_joined_to_trans;
2757         boolean_t               created = B_FALSE;
2758         int                     dm_event_sent = 0;
2759         xfs_prid_t              prid;
2760         struct xfs_dquot        *udqp, *gdqp;
2761         uint                    resblks;
2762         int                     dm_di_mode;
2763         int                     dir_namelen;
2764
2765         dir_vp = BHV_TO_VNODE(dir_bdp);
2766         dp = XFS_BHVTOI(dir_bdp);
2767         mp = dp->i_mount;
2768
2769         if (XFS_FORCED_SHUTDOWN(mp))
2770                 return XFS_ERROR(EIO);
2771
2772         dir_namelen = VNAMELEN(dentry);
2773
2774         tp = NULL;
2775         dp_joined_to_trans = B_FALSE;
2776         dm_di_mode = vap->va_mode;
2777
2778         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2779                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2780                                         dir_vp, DM_RIGHT_NULL, NULL,
2781                                         DM_RIGHT_NULL, dir_name, NULL,
2782                                         dm_di_mode, 0, 0);
2783                 if (error)
2784                         return error;
2785                 dm_event_sent = 1;
2786         }
2787
2788         /* Return through std_return after this point. */
2789
2790         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2791
2792         mp = dp->i_mount;
2793         udqp = gdqp = NULL;
2794         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2795                 prid = dp->i_d.di_projid;
2796         else if (vap->va_mask & XFS_AT_PROJID)
2797                 prid = (xfs_prid_t)vap->va_projid;
2798         else
2799                 prid = (xfs_prid_t)dfltprid;
2800
2801         /*
2802          * Make sure that we have allocated dquot(s) on disk.
2803          */
2804         error = XFS_QM_DQVOPALLOC(mp, dp,
2805                         current_fsuid(credp), current_fsgid(credp), prid,
2806                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2807         if (error)
2808                 goto std_return;
2809
2810         tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2811         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2812         resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2813         error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2814                                   XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2815         if (error == ENOSPC) {
2816                 resblks = 0;
2817                 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2818                                           XFS_TRANS_PERM_LOG_RES,
2819                                           XFS_MKDIR_LOG_COUNT);
2820         }
2821         if (error) {
2822                 cancel_flags = 0;
2823                 dp = NULL;
2824                 goto error_return;
2825         }
2826
2827         xfs_ilock(dp, XFS_ILOCK_EXCL);
2828
2829         /*
2830          * Check for directory link count overflow.
2831          */
2832         if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2833                 error = XFS_ERROR(EMLINK);
2834                 goto error_return;
2835         }
2836
2837         /*
2838          * Reserve disk quota and the inode.
2839          */
2840         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2841         if (error)
2842                 goto error_return;
2843
2844         if (resblks == 0 &&
2845             (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2846                 goto error_return;
2847         /*
2848          * create the directory inode.
2849          */
2850         error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2851                         0, credp, prid, resblks > 0,
2852                 &cdp, NULL);
2853         if (error) {
2854                 if (error == ENOSPC)
2855                         goto error_return;
2856                 goto abort_return;
2857         }
2858         ITRACE(cdp);
2859
2860         /*
2861          * Now we add the directory inode to the transaction.
2862          * We waited until now since xfs_dir_ialloc might start
2863          * a new transaction.  Had we joined the transaction
2864          * earlier, the locks might have gotten released.
2865          */
2866         VN_HOLD(dir_vp);
2867         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2868         dp_joined_to_trans = B_TRUE;
2869
2870         XFS_BMAP_INIT(&free_list, &first_block);
2871
2872         error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2873                         cdp->i_ino, &first_block, &free_list,
2874                         resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2875         if (error) {
2876                 ASSERT(error != ENOSPC);
2877                 goto error1;
2878         }
2879         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2880
2881         /*
2882          * Bump the in memory version number of the parent directory
2883          * so that other processes accessing it will recognize that
2884          * the directory has changed.
2885          */
2886         dp->i_gen++;
2887
2888         error = XFS_DIR_INIT(mp, tp, cdp, dp);
2889         if (error) {
2890                 goto error2;
2891         }
2892
2893         cdp->i_gen = 1;
2894         error = xfs_bumplink(tp, dp);
2895         if (error) {
2896                 goto error2;
2897         }
2898
2899         cvp = XFS_ITOV(cdp);
2900
2901         created = B_TRUE;
2902
2903         *vpp = cvp;
2904         IHOLD(cdp);
2905
2906         /*
2907          * Attach the dquots to the new inode and modify the icount incore.
2908          */
2909         XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2910
2911         /*
2912          * If this is a synchronous mount, make sure that the
2913          * mkdir transaction goes to disk before returning to
2914          * the user.
2915          */
2916         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2917                 xfs_trans_set_sync(tp);
2918         }
2919
2920         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2921         if (error) {
2922                 IRELE(cdp);
2923                 goto error2;
2924         }
2925
2926         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2927         XFS_QM_DQRELE(mp, udqp);
2928         XFS_QM_DQRELE(mp, gdqp);
2929         if (error) {
2930                 IRELE(cdp);
2931         }
2932
2933         /* Fall through to std_return with error = 0 or errno from
2934          * xfs_trans_commit. */
2935
2936 std_return:
2937         if ( (created || (error != 0 && dm_event_sent != 0)) &&
2938                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2939                                                 DM_EVENT_POSTCREATE)) {
2940                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2941                                         dir_vp, DM_RIGHT_NULL,
2942                                         created ? XFS_ITOV(cdp):NULL,
2943                                         DM_RIGHT_NULL,
2944                                         dir_name, NULL,
2945                                         dm_di_mode, error, 0);
2946         }
2947         return error;
2948
2949  error2:
2950  error1:
2951         xfs_bmap_cancel(&free_list);
2952  abort_return:
2953         cancel_flags |= XFS_TRANS_ABORT;
2954  error_return:
2955         xfs_trans_cancel(tp, cancel_flags);
2956         XFS_QM_DQRELE(mp, udqp);
2957         XFS_QM_DQRELE(mp, gdqp);
2958
2959         if (!dp_joined_to_trans && (dp != NULL)) {
2960                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2961         }
2962
2963         goto std_return;
2964 }
2965
2966
2967 /*
2968  * xfs_rmdir
2969  *
2970  */
2971 STATIC int
2972 xfs_rmdir(
2973         bhv_desc_t              *dir_bdp,
2974         vname_t                 *dentry,
2975         cred_t                  *credp)
2976 {
2977         char                    *name = VNAME(dentry);
2978         xfs_inode_t             *dp;
2979         xfs_inode_t             *cdp;   /* child directory */
2980         xfs_trans_t             *tp;
2981         xfs_mount_t             *mp;
2982         int                     error;
2983         xfs_bmap_free_t         free_list;
2984         xfs_fsblock_t           first_block;
2985         int                     cancel_flags;
2986         int                     committed;
2987         vnode_t                 *dir_vp;
2988         int                     dm_di_mode = 0;
2989         int                     last_cdp_link;
2990         int                     namelen;
2991         uint                    resblks;
2992
2993         dir_vp = BHV_TO_VNODE(dir_bdp);
2994         dp = XFS_BHVTOI(dir_bdp);
2995         mp = dp->i_mount;
2996
2997         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2998
2999         if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3000                 return XFS_ERROR(EIO);
3001         namelen = VNAMELEN(dentry);
3002
3003         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3004                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3005                                         dir_vp, DM_RIGHT_NULL,
3006                                         NULL, DM_RIGHT_NULL,
3007                                         name, NULL, 0, 0, 0);
3008                 if (error)
3009                         return XFS_ERROR(error);
3010         }
3011
3012         /* Return through std_return after this point. */
3013
3014         cdp = NULL;
3015
3016         /*
3017          * We need to get a reference to cdp before we get our log
3018          * reservation.  The reason for this is that we cannot call
3019          * xfs_iget for an inode for which we do not have a reference
3020          * once we've acquired a log reservation.  This is because the
3021          * inode we are trying to get might be in xfs_inactive going
3022          * for a log reservation.  Since we'll have to wait for the
3023          * inactive code to complete before returning from xfs_iget,
3024          * we need to make sure that we don't have log space reserved
3025          * when we call xfs_iget.  Instead we get an unlocked referece
3026          * to the inode before getting our log reservation.
3027          */
3028         error = xfs_get_dir_entry(dentry, &cdp);
3029         if (error) {
3030                 REMOVE_DEBUG_TRACE(__LINE__);
3031                 goto std_return;
3032         }
3033         mp = dp->i_mount;
3034         dm_di_mode = cdp->i_d.di_mode;
3035
3036         /*
3037          * Get the dquots for the inodes.
3038          */
3039         error = XFS_QM_DQATTACH(mp, dp, 0);
3040         if (!error && dp != cdp)
3041                 error = XFS_QM_DQATTACH(mp, cdp, 0);
3042         if (error) {
3043                 IRELE(cdp);
3044                 REMOVE_DEBUG_TRACE(__LINE__);
3045                 goto std_return;
3046         }
3047
3048         tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3049         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3050         /*
3051          * We try to get the real space reservation first,
3052          * allowing for directory btree deletion(s) implying
3053          * possible bmap insert(s).  If we can't get the space
3054          * reservation then we use 0 instead, and avoid the bmap
3055          * btree insert(s) in the directory code by, if the bmap
3056          * insert tries to happen, instead trimming the LAST
3057          * block from the directory.
3058          */
3059         resblks = XFS_REMOVE_SPACE_RES(mp);
3060         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3061                         XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3062         if (error == ENOSPC) {
3063                 resblks = 0;
3064                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3065                                 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3066         }
3067         if (error) {
3068                 ASSERT(error != ENOSPC);
3069                 cancel_flags = 0;
3070                 IRELE(cdp);
3071                 goto error_return;
3072         }
3073         XFS_BMAP_INIT(&free_list, &first_block);
3074
3075         /*
3076          * Now lock the child directory inode and the parent directory
3077          * inode in the proper order.  This will take care of validating
3078          * that the directory entry for the child directory inode has
3079          * not changed while we were obtaining a log reservation.
3080          */
3081         error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3082         if (error) {
3083                 xfs_trans_cancel(tp, cancel_flags);
3084                 IRELE(cdp);
3085                 goto std_return;
3086         }
3087
3088         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3089         if (dp != cdp) {
3090                 /*
3091                  * Only increment the parent directory vnode count if
3092                  * we didn't bump it in looking up cdp.  The only time
3093                  * we don't bump it is when we're looking up ".".
3094                  */
3095                 VN_HOLD(dir_vp);
3096         }
3097
3098         ITRACE(cdp);
3099         xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3100
3101         ASSERT(cdp->i_d.di_nlink >= 2);
3102         if (cdp->i_d.di_nlink != 2) {
3103                 error = XFS_ERROR(ENOTEMPTY);
3104                 goto error_return;
3105         }
3106         if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3107                 error = XFS_ERROR(ENOTEMPTY);
3108                 goto error_return;
3109         }
3110
3111         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3112                 &first_block, &free_list, resblks);
3113         if (error) {
3114                 goto error1;
3115         }
3116
3117         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3118
3119         /*
3120          * Bump the in memory generation count on the parent
3121          * directory so that other can know that it has changed.
3122          */
3123         dp->i_gen++;
3124
3125         /*
3126          * Drop the link from cdp's "..".
3127          */
3128         error = xfs_droplink(tp, dp);
3129         if (error) {
3130                 goto error1;
3131         }
3132
3133         /*
3134          * Drop the link from dp to cdp.
3135          */
3136         error = xfs_droplink(tp, cdp);
3137         if (error) {
3138                 goto error1;
3139         }
3140
3141         /*
3142          * Drop the "." link from cdp to self.
3143          */
3144         error = xfs_droplink(tp, cdp);
3145         if (error) {
3146                 goto error1;
3147         }
3148
3149         /* Determine these before committing transaction */
3150         last_cdp_link = (cdp)->i_d.di_nlink==0;
3151
3152         /*
3153          * Take an extra ref on the child vnode so that it
3154          * does not go to xfs_inactive() from within the commit.
3155          */
3156         IHOLD(cdp);
3157
3158         /*
3159          * If this is a synchronous mount, make sure that the
3160          * rmdir transaction goes to disk before returning to
3161          * the user.
3162          */
3163         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3164                 xfs_trans_set_sync(tp);
3165         }
3166
3167         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3168         if (error) {
3169                 xfs_bmap_cancel(&free_list);
3170                 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3171                                  XFS_TRANS_ABORT));
3172                 IRELE(cdp);
3173                 goto std_return;
3174         }
3175
3176         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3177         if (error) {
3178                 IRELE(cdp);
3179                 goto std_return;
3180         }
3181
3182
3183         /*
3184          * Let interposed file systems know about removed links.
3185          */
3186         VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3187
3188         IRELE(cdp);
3189
3190         /* Fall through to std_return with error = 0 or the errno
3191          * from xfs_trans_commit. */
3192 std_return:
3193         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3194                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3195                                         dir_vp, DM_RIGHT_NULL,
3196                                         NULL, DM_RIGHT_NULL,
3197                                         name, NULL, dm_di_mode,
3198                                         error, 0);
3199         }
3200         return error;
3201
3202  error1:
3203         xfs_bmap_cancel(&free_list);
3204         cancel_flags |= XFS_TRANS_ABORT;
3205  error_return:
3206         xfs_trans_cancel(tp, cancel_flags);
3207         goto std_return;
3208 }
3209
3210
3211 /*
3212  * xfs_readdir
3213  *
3214  * Read dp's entries starting at uiop->uio_offset and translate them into
3215  * bufsize bytes worth of struct dirents starting at bufbase.
3216  */
3217 STATIC int
3218 xfs_readdir(
3219         bhv_desc_t      *dir_bdp,
3220         uio_t           *uiop,
3221         cred_t          *credp,
3222         int             *eofp)
3223 {
3224         xfs_inode_t     *dp;
3225         xfs_trans_t     *tp = NULL;
3226         int             error = 0;
3227         uint            lock_mode;
3228
3229         vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3230                                                (inst_t *)__return_address);
3231         dp = XFS_BHVTOI(dir_bdp);
3232
3233         if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3234                 return XFS_ERROR(EIO);
3235         }
3236
3237         lock_mode = xfs_ilock_map_shared(dp);
3238         error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3239         xfs_iunlock_map_shared(dp, lock_mode);
3240         return error;
3241 }
3242
3243
3244 /*
3245  * xfs_symlink
3246  *
3247  */
3248 STATIC int
3249 xfs_symlink(
3250         bhv_desc_t              *dir_bdp,
3251         vname_t                 *dentry,
3252         vattr_t                 *vap,
3253         char                    *target_path,
3254         vnode_t                 **vpp,
3255         cred_t                  *credp)
3256 {
3257         xfs_trans_t             *tp;
3258         xfs_mount_t             *mp;
3259         xfs_inode_t             *dp;
3260         xfs_inode_t             *ip;
3261         int                     error;
3262         int                     pathlen;
3263         xfs_bmap_free_t         free_list;
3264         xfs_fsblock_t           first_block;
3265         boolean_t               dp_joined_to_trans;
3266         vnode_t                 *dir_vp;
3267         uint                    cancel_flags;
3268         int                     committed;
3269         xfs_fileoff_t           first_fsb;
3270         xfs_filblks_t           fs_blocks;
3271         int                     nmaps;
3272         xfs_bmbt_irec_t         mval[SYMLINK_MAPS];
3273         xfs_daddr_t             d;
3274         char                    *cur_chunk;
3275         int                     byte_cnt;
3276         int                     n;
3277         xfs_buf_t               *bp;
3278         xfs_prid_t              prid;
3279         struct xfs_dquot        *udqp, *gdqp;
3280         uint                    resblks;
3281         char                    *link_name = VNAME(dentry);
3282         int                     link_namelen;
3283
3284         *vpp = NULL;
3285         dir_vp = BHV_TO_VNODE(dir_bdp);
3286         dp = XFS_BHVTOI(dir_bdp);
3287         dp_joined_to_trans = B_FALSE;
3288         error = 0;
3289         ip = NULL;
3290         tp = NULL;
3291
3292         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3293
3294         mp = dp->i_mount;
3295
3296         if (XFS_FORCED_SHUTDOWN(mp))
3297                 return XFS_ERROR(EIO);
3298
3299         link_namelen = VNAMELEN(dentry);
3300
3301         /*
3302          * Check component lengths of the target path name.
3303          */
3304         pathlen = strlen(target_path);
3305         if (pathlen >= MAXPATHLEN)      /* total string too long */
3306                 return XFS_ERROR(ENAMETOOLONG);
3307         if (pathlen >= MAXNAMELEN) {    /* is any component too long? */
3308                 int len, total;
3309                 char *path;
3310
3311                 for(total = 0, path = target_path; total < pathlen;) {
3312                         /*
3313                          * Skip any slashes.
3314                          */
3315                         while(*path == '/') {
3316                                 total++;
3317                                 path++;
3318                         }
3319
3320                         /*
3321                          * Count up to the next slash or end of path.
3322                          * Error out if the component is bigger than MAXNAMELEN.
3323                          */
3324                         for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3325                                 if (++len >= MAXNAMELEN) {
3326                                         error = ENAMETOOLONG;
3327                                         return error;
3328                                 }
3329                         }
3330                 }
3331         }
3332
3333         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3334                 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3335                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3336                                         link_name, target_path, 0, 0, 0);
3337                 if (error)
3338                         return error;
3339         }
3340
3341         /* Return through std_return after this point. */
3342
3343         udqp = gdqp = NULL;
3344         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3345                 prid = dp->i_d.di_projid;
3346         else if (vap->va_mask & XFS_AT_PROJID)
3347                 prid = (xfs_prid_t)vap->va_projid;
3348         else
3349                 prid = (xfs_prid_t)dfltprid;
3350
3351         /*
3352          * Make sure that we have allocated dquot(s) on disk.
3353          */
3354         error = XFS_QM_DQVOPALLOC(mp, dp,
3355                         current_fsuid(credp), current_fsgid(credp), prid,
3356                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3357         if (error)
3358                 goto std_return;
3359
3360         tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3361         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3362         /*
3363          * The symlink will fit into the inode data fork?
3364          * There can't be any attributes so we get the whole variable part.
3365          */
3366         if (pathlen <= XFS_LITINO(mp))
3367                 fs_blocks = 0;
3368         else
3369                 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3370         resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3371         error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3372                         XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3373         if (error == ENOSPC && fs_blocks == 0) {
3374                 resblks = 0;
3375                 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3376                                 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3377         }
3378         if (error) {
3379                 cancel_flags = 0;
3380                 dp = NULL;
3381                 goto error_return;
3382         }
3383
3384         xfs_ilock(dp, XFS_ILOCK_EXCL);
3385
3386         /*
3387          * Check whether the directory allows new symlinks or not.
3388          */
3389         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3390                 error = XFS_ERROR(EPERM);
3391                 goto error_return;
3392         }
3393
3394         /*
3395          * Reserve disk quota : blocks and inode.
3396          */
3397         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3398         if (error)
3399                 goto error_return;
3400
3401         /*
3402          * Check for ability to enter directory entry, if no space reserved.
3403          */
3404         if (resblks == 0 &&
3405             (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3406                 goto error_return;
3407         /*
3408          * Initialize the bmap freelist prior to calling either
3409          * bmapi or the directory create code.
3410          */
3411         XFS_BMAP_INIT(&free_list, &first_block);
3412
3413         /*
3414          * Allocate an inode for the symlink.
3415          */
3416         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3417                                1, 0, credp, prid, resblks > 0, &ip, NULL);
3418         if (error) {
3419                 if (error == ENOSPC)
3420                         goto error_return;
3421                 goto error1;
3422         }
3423         ITRACE(ip);
3424
3425         VN_HOLD(dir_vp);
3426         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3427         dp_joined_to_trans = B_TRUE;
3428
3429         /*
3430          * Also attach the dquot(s) to it, if applicable.
3431          */
3432         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3433
3434         if (resblks)
3435                 resblks -= XFS_IALLOC_SPACE_RES(mp);
3436         /*
3437          * If the symlink will fit into the inode, write it inline.
3438          */
3439         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3440                 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3441                 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3442                 ip->i_d.di_size = pathlen;
3443
3444                 /*
3445                  * The inode was initially created in extent format.
3446                  */
3447                 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3448                 ip->i_df.if_flags |= XFS_IFINLINE;
3449
3450                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3451                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3452
3453         } else {
3454                 first_fsb = 0;
3455                 nmaps = SYMLINK_MAPS;
3456
3457                 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3458                                   XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3459                                   &first_block, resblks, mval, &nmaps,
3460                                   &free_list);
3461                 if (error) {
3462                         goto error1;
3463                 }
3464
3465                 if (resblks)
3466                         resblks -= fs_blocks;
3467                 ip->i_d.di_size = pathlen;
3468                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3469
3470                 cur_chunk = target_path;
3471                 for (n = 0; n < nmaps; n++) {
3472                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3473                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3474                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3475                                                BTOBB(byte_cnt), 0);
3476                         ASSERT(bp && !XFS_BUF_GETERROR(bp));
3477                         if (pathlen < byte_cnt) {
3478                                 byte_cnt = pathlen;
3479                         }
3480                         pathlen -= byte_cnt;
3481
3482                         memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3483                         cur_chunk += byte_cnt;
3484
3485                         xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3486                 }
3487         }
3488
3489         /*
3490          * Create the directory entry for the symlink.
3491          */
3492         error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3493                         ip->i_ino, &first_block, &free_list, resblks);
3494         if (error) {
3495                 goto error1;
3496         }
3497         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3498         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3499
3500         /*
3501          * Bump the in memory version number of the parent directory
3502          * so that other processes accessing it will recognize that
3503          * the directory has changed.
3504          */
3505         dp->i_gen++;
3506
3507         /*
3508          * If this is a synchronous mount, make sure that the
3509          * symlink transaction goes to disk before returning to
3510          * the user.
3511          */
3512         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3513                 xfs_trans_set_sync(tp);
3514         }
3515
3516         /*
3517          * xfs_trans_commit normally decrements the vnode ref count
3518          * when it unlocks the inode. Since we want to return the
3519          * vnode to the caller, we bump the vnode ref count now.
3520          */
3521         IHOLD(ip);
3522
3523         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3524         if (error) {
3525                 goto error2;
3526         }
3527         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3528         XFS_QM_DQRELE(mp, udqp);
3529         XFS_QM_DQRELE(mp, gdqp);
3530
3531         /* Fall through to std_return with error = 0 or errno from
3532          * xfs_trans_commit     */
3533 std_return:
3534         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3535                              DM_EVENT_POSTSYMLINK)) {
3536                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3537                                         dir_vp, DM_RIGHT_NULL,
3538                                         error ? NULL : XFS_ITOV(ip),
3539                                         DM_RIGHT_NULL, link_name, target_path,
3540                                         0, error, 0);
3541         }
3542
3543         if (!error) {
3544                 vnode_t *vp;
3545
3546                 ASSERT(ip);
3547                 vp = XFS_ITOV(ip);
3548                 *vpp = vp;
3549         }
3550         return error;
3551
3552  error2:
3553         IRELE(ip);
3554  error1:
3555         xfs_bmap_cancel(&free_list);
3556         cancel_flags |= XFS_TRANS_ABORT;
3557  error_return:
3558         xfs_trans_cancel(tp, cancel_flags);
3559         XFS_QM_DQRELE(mp, udqp);
3560         XFS_QM_DQRELE(mp, gdqp);
3561
3562         if (!dp_joined_to_trans && (dp != NULL)) {
3563                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3564         }
3565
3566         goto std_return;
3567 }
3568
3569
3570 /*
3571  * xfs_fid2
3572  *
3573  * A fid routine that takes a pointer to a previously allocated
3574  * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3575  */
3576 STATIC int
3577 xfs_fid2(
3578         bhv_desc_t      *bdp,
3579         fid_t           *fidp)
3580 {
3581         xfs_inode_t     *ip;
3582         xfs_fid2_t      *xfid;
3583
3584         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3585                                        (inst_t *)__return_address);
3586         ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3587
3588         xfid = (xfs_fid2_t *)fidp;
3589         ip = XFS_BHVTOI(bdp);
3590         xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3591         xfid->fid_pad = 0;
3592         /*
3593          * use memcpy because the inode is a long long and there's no
3594          * assurance that xfid->fid_ino is properly aligned.
3595          */
3596         memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3597         xfid->fid_gen = ip->i_d.di_gen;
3598
3599         return 0;
3600 }
3601
3602
3603 /*
3604  * xfs_rwlock
3605  */
3606 int
3607 xfs_rwlock(
3608         bhv_desc_t      *bdp,
3609         vrwlock_t       locktype)
3610 {
3611         xfs_inode_t     *ip;
3612         vnode_t         *vp;
3613
3614         vp = BHV_TO_VNODE(bdp);
3615         if (VN_ISDIR(vp))
3616                 return 1;
3617         ip = XFS_BHVTOI(bdp);
3618         if (locktype == VRWLOCK_WRITE) {
3619                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3620         } else if (locktype == VRWLOCK_TRY_READ) {
3621                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3622         } else if (locktype == VRWLOCK_TRY_WRITE) {
3623                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3624         } else {
3625                 ASSERT((locktype == VRWLOCK_READ) ||
3626                        (locktype == VRWLOCK_WRITE_DIRECT));
3627                 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3628         }
3629
3630         return 1;
3631 }
3632
3633
3634 /*
3635  * xfs_rwunlock
3636  */
3637 void
3638 xfs_rwunlock(
3639         bhv_desc_t      *bdp,
3640         vrwlock_t       locktype)
3641 {
3642         xfs_inode_t     *ip;
3643         vnode_t         *vp;
3644
3645         vp = BHV_TO_VNODE(bdp);
3646         if (VN_ISDIR(vp))
3647                 return;
3648         ip = XFS_BHVTOI(bdp);
3649         if (locktype == VRWLOCK_WRITE) {
3650                 /*
3651                  * In the write case, we may have added a new entry to
3652                  * the reference cache.  This might store a pointer to
3653                  * an inode to be released in this inode.  If it is there,
3654                  * clear the pointer and release the inode after unlocking
3655                  * this one.
3656                  */
3657                 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3658         } else {
3659                 ASSERT((locktype == VRWLOCK_READ) ||
3660                        (locktype == VRWLOCK_WRITE_DIRECT));
3661                 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3662         }
3663         return;
3664 }
3665
3666 STATIC int
3667 xfs_inode_flush(
3668         bhv_desc_t      *bdp,
3669         int             flags)
3670 {
3671         xfs_inode_t     *ip;
3672         xfs_mount_t     *mp;
3673         xfs_inode_log_item_t *iip;
3674         int             error = 0;
3675
3676         ip = XFS_BHVTOI(bdp);
3677         mp = ip->i_mount;
3678         iip = ip->i_itemp;
3679
3680         if (XFS_FORCED_SHUTDOWN(mp))
3681                 return XFS_ERROR(EIO);
3682
3683         /*
3684          * Bypass inodes which have already been cleaned by
3685          * the inode flush clustering code inside xfs_iflush
3686          */
3687         if ((ip->i_update_core == 0) &&
3688             ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3689                 return 0;
3690
3691         if (flags & FLUSH_LOG) {
3692                 if (iip && iip->ili_last_lsn) {
3693                         xlog_t          *log = mp->m_log;
3694                         xfs_lsn_t       sync_lsn;
3695                         int             s, log_flags = XFS_LOG_FORCE;
3696
3697                         s = GRANT_LOCK(log);
3698                         sync_lsn = log->l_last_sync_lsn;
3699                         GRANT_UNLOCK(log, s);
3700
3701                         if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3702                                 return 0;
3703
3704                         if (flags & FLUSH_SYNC)
3705                                 log_flags |= XFS_LOG_SYNC;
3706                         return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3707                 }
3708         }
3709
3710         /*
3711          * We make this non-blocking if the inode is contended,
3712          * return EAGAIN to indicate to the caller that they
3713          * did not succeed. This prevents the flush path from
3714          * blocking on inodes inside another operation right
3715          * now, they get caught later by xfs_sync.
3716          */
3717         if (flags & FLUSH_INODE) {
3718                 int     flush_flags;
3719
3720                 if (xfs_ipincount(ip))
3721                         return EAGAIN;
3722
3723                 if (flags & FLUSH_SYNC) {
3724                         xfs_ilock(ip, XFS_ILOCK_SHARED);
3725                         xfs_iflock(ip);
3726                 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3727                         if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3728                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3729                                 return EAGAIN;
3730                         }
3731                 } else {
3732                         return EAGAIN;
3733                 }
3734
3735                 if (flags & FLUSH_SYNC)
3736                         flush_flags = XFS_IFLUSH_SYNC;
3737                 else
3738                         flush_flags = XFS_IFLUSH_ASYNC;
3739
3740                 error = xfs_iflush(ip, flush_flags);
3741                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3742         }
3743
3744         return error;
3745 }
3746
3747
3748 int
3749 xfs_set_dmattrs (
3750         bhv_desc_t      *bdp,
3751         u_int           evmask,
3752         u_int16_t       state,
3753         cred_t          *credp)
3754 {
3755         xfs_inode_t     *ip;
3756         xfs_trans_t     *tp;
3757         xfs_mount_t     *mp;
3758         int             error;
3759
3760         if (!capable(CAP_SYS_ADMIN))
3761                 return XFS_ERROR(EPERM);
3762
3763         ip = XFS_BHVTOI(bdp);
3764         mp = ip->i_mount;
3765
3766         if (XFS_FORCED_SHUTDOWN(mp))
3767                 return XFS_ERROR(EIO);
3768
3769         tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3770         error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3771         if (error) {
3772                 xfs_trans_cancel(tp, 0);
3773                 return error;
3774         }
3775         xfs_ilock(ip, XFS_ILOCK_EXCL);
3776         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3777
3778         ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3779         ip->i_iocore.io_dmstate  = ip->i_d.di_dmstate  = state;
3780
3781         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3782         IHOLD(ip);
3783         error = xfs_trans_commit(tp, 0, NULL);
3784
3785         return error;
3786 }
3787
3788
3789 /*
3790  * xfs_reclaim
3791  */
3792 STATIC int
3793 xfs_reclaim(
3794         bhv_desc_t      *bdp)
3795 {
3796         xfs_inode_t     *ip;
3797         vnode_t         *vp;
3798
3799         vp = BHV_TO_VNODE(bdp);
3800         ip = XFS_BHVTOI(bdp);
3801
3802         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3803
3804         ASSERT(!VN_MAPPED(vp));
3805
3806         /* bad inode, get out here ASAP */
3807         if (VN_BAD(vp)) {
3808                 xfs_ireclaim(ip);
3809                 return 0;
3810         }
3811
3812         vn_iowait(vp);
3813
3814         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3815
3816         /*
3817          * Make sure the atime in the XFS inode is correct before freeing the
3818          * Linux inode.
3819          */
3820         xfs_synchronize_atime(ip);
3821
3822         /* If we have nothing to flush with this inode then complete the
3823          * teardown now, otherwise break the link between the xfs inode
3824          * and the linux inode and clean up the xfs inode later. This
3825          * avoids flushing the inode to disk during the delete operation
3826          * itself.
3827          */
3828         if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3829                 xfs_ilock(ip, XFS_ILOCK_EXCL);
3830                 xfs_iflock(ip);
3831                 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3832         } else {
3833                 xfs_mount_t     *mp = ip->i_mount;
3834
3835                 /* Protect sync from us */
3836                 XFS_MOUNT_ILOCK(mp);
3837                 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3838                 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3839                 ip->i_flags |= XFS_IRECLAIMABLE;
3840                 XFS_MOUNT_IUNLOCK(mp);
3841         }
3842         return 0;
3843 }
3844
3845 int
3846 xfs_finish_reclaim(
3847         xfs_inode_t     *ip,
3848         int             locked,
3849         int             sync_mode)
3850 {
3851         xfs_ihash_t     *ih = ip->i_hash;
3852         vnode_t         *vp = XFS_ITOV_NULL(ip);
3853         int             error;
3854
3855         if (vp && VN_BAD(vp))
3856                 goto reclaim;
3857
3858         /* The hash lock here protects a thread in xfs_iget_core from
3859          * racing with us on linking the inode back with a vnode.
3860          * Once we have the XFS_IRECLAIM flag set it will not touch
3861          * us.
3862          */
3863         write_lock(&ih->ih_lock);
3864         if ((ip->i_flags & XFS_IRECLAIM) ||
3865             (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3866                 write_unlock(&ih->ih_lock);
3867                 if (locked) {
3868                         xfs_ifunlock(ip);
3869                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3870                 }
3871                 return(1);
3872         }
3873         ip->i_flags |= XFS_IRECLAIM;
3874         write_unlock(&ih->ih_lock);
3875
3876         /*
3877          * If the inode is still dirty, then flush it out.  If the inode
3878          * is not in the AIL, then it will be OK to flush it delwri as
3879          * long as xfs_iflush() does not keep any references to the inode.
3880          * We leave that decision up to xfs_iflush() since it has the
3881          * knowledge of whether it's OK to simply do a delwri flush of
3882          * the inode or whether we need to wait until the inode is
3883          * pulled from the AIL.
3884          * We get the flush lock regardless, though, just to make sure
3885          * we don't free it while it is being flushed.
3886          */
3887         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3888                 if (!locked) {
3889                         xfs_ilock(ip, XFS_ILOCK_EXCL);
3890                         xfs_iflock(ip);
3891                 }
3892
3893                 if (ip->i_update_core ||
3894                     ((ip->i_itemp != NULL) &&
3895                      (ip->i_itemp->ili_format.ilf_fields != 0))) {
3896                         error = xfs_iflush(ip, sync_mode);
3897                         /*
3898                          * If we hit an error, typically because of filesystem
3899                          * shutdown, we don't need to let vn_reclaim to know
3900                          * because we're gonna reclaim the inode anyway.
3901                          */
3902                         if (error) {
3903                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3904                                 goto reclaim;
3905                         }
3906                         xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3907                 }
3908
3909                 ASSERT(ip->i_update_core == 0);
3910                 ASSERT(ip->i_itemp == NULL ||
3911                        ip->i_itemp->ili_format.ilf_fields == 0);
3912                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3913         } else if (locked) {
3914                 /*
3915                  * We are not interested in doing an iflush if we're
3916                  * in the process of shutting down the filesystem forcibly.
3917                  * So, just reclaim the inode.
3918                  */
3919                 xfs_ifunlock(ip);
3920                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3921         }
3922
3923  reclaim:
3924         xfs_ireclaim(ip);
3925         return 0;
3926 }
3927
3928 int
3929 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3930 {
3931         int             purged;
3932         xfs_inode_t     *ip, *n;
3933         int             done = 0;
3934
3935         while (!done) {
3936                 purged = 0;
3937                 XFS_MOUNT_ILOCK(mp);
3938                 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3939                         if (noblock) {
3940                                 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3941                                         continue;
3942                                 if (xfs_ipincount(ip) ||
3943                                     !xfs_iflock_nowait(ip)) {
3944                                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3945                                         continue;
3946                                 }
3947                         }
3948                         XFS_MOUNT_IUNLOCK(mp);
3949                         if (xfs_finish_reclaim(ip, noblock,
3950                                         XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3951                                 delay(1);
3952                         purged = 1;
3953                         break;
3954                 }
3955
3956                 done = !purged;
3957         }
3958
3959         XFS_MOUNT_IUNLOCK(mp);
3960         return 0;
3961 }
3962
3963 /*
3964  * xfs_alloc_file_space()
3965  *      This routine allocates disk space for the given file.
3966  *
3967  *      If alloc_type == 0, this request is for an ALLOCSP type
3968  *      request which will change the file size.  In this case, no
3969  *      DMAPI event will be generated by the call.  A TRUNCATE event
3970  *      will be generated later by xfs_setattr.
3971  *
3972  *      If alloc_type != 0, this request is for a RESVSP type
3973  *      request, and a DMAPI DM_EVENT_WRITE will be generated if the
3974  *      lower block boundary byte address is less than the file's
3975  *      length.
3976  *
3977  * RETURNS:
3978  *       0 on success
3979  *      errno on error
3980  *
3981  */
3982 STATIC int
3983 xfs_alloc_file_space(
3984         xfs_inode_t             *ip,
3985         xfs_off_t               offset,
3986         xfs_off_t               len,
3987         int                     alloc_type,
3988         int                     attr_flags)
3989 {
3990         xfs_mount_t             *mp = ip->i_mount;
3991         xfs_off_t               count;
3992         xfs_filblks_t           allocated_fsb;
3993         xfs_filblks_t           allocatesize_fsb;
3994         xfs_extlen_t            extsz, temp;
3995         xfs_fileoff_t           startoffset_fsb;
3996         xfs_fsblock_t           firstfsb;
3997         int                     nimaps;
3998         int                     bmapi_flag;
3999         int                     quota_flag;
4000         int                     rt;
4001         xfs_trans_t             *tp;
4002         xfs_bmbt_irec_t         imaps[1], *imapp;
4003         xfs_bmap_free_t         free_list;
4004         uint                    qblocks, resblks, resrtextents;
4005         int                     committed;
4006         int                     error;
4007
4008         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4009
4010         if (XFS_FORCED_SHUTDOWN(mp))
4011                 return XFS_ERROR(EIO);
4012
4013         rt = XFS_IS_REALTIME_INODE(ip);
4014         if (unlikely(rt)) {
4015                 if (!(extsz = ip->i_d.di_extsize))
4016                         extsz = mp->m_sb.sb_rextsize;
4017         } else {
4018                 extsz = ip->i_d.di_extsize;
4019         }
4020
4021         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4022                 return error;
4023
4024         if (len <= 0)
4025                 return XFS_ERROR(EINVAL);
4026
4027         count = len;
4028         error = 0;
4029         imapp = &imaps[0];
4030         nimaps = 1;
4031         bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4032         startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4033         allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4034
4035         /*      Generate a DMAPI event if needed.       */
4036         if (alloc_type != 0 && offset < ip->i_d.di_size &&
4037                         (attr_flags&ATTR_DMI) == 0  &&
4038                         DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4039                 xfs_off_t           end_dmi_offset;
4040
4041                 end_dmi_offset = offset+len;
4042                 if (end_dmi_offset > ip->i_d.di_size)
4043                         end_dmi_offset = ip->i_d.di_size;
4044                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4045                         offset, end_dmi_offset - offset,
4046                         0, NULL);
4047                 if (error)
4048                         return(error);
4049         }
4050
4051         /*
4052          * Allocate file space until done or until there is an error
4053          */
4054 retry:
4055         while (allocatesize_fsb && !error) {
4056                 xfs_fileoff_t   s, e;
4057
4058                 /*
4059                  * Determine space reservations for data/realtime.
4060                  */
4061                 if (unlikely(extsz)) {
4062                         s = startoffset_fsb;
4063                         do_div(s, extsz);
4064                         s *= extsz;
4065                         e = startoffset_fsb + allocatesize_fsb;
4066                         if ((temp = do_mod(startoffset_fsb, extsz)))
4067                                 e += temp;
4068                         if ((temp = do_mod(e, extsz)))
4069                                 e += extsz - temp;
4070                 } else {
4071                         s = 0;
4072                         e = allocatesize_fsb;
4073                 }
4074
4075                 if (unlikely(rt)) {
4076                         resrtextents = qblocks = (uint)(e - s);
4077                         resrtextents /= mp->m_sb.sb_rextsize;
4078                         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4079                         quota_flag = XFS_QMOPT_RES_RTBLKS;
4080                 } else {
4081                         resrtextents = 0;
4082                         resblks = qblocks = \
4083                                 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4084                         quota_flag = XFS_QMOPT_RES_REGBLKS;
4085                 }
4086
4087                 /*
4088                  * Allocate and setup the transaction.
4089                  */
4090                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4091                 error = xfs_trans_reserve(tp, resblks,
4092                                           XFS_WRITE_LOG_RES(mp), resrtextents,
4093                                           XFS_TRANS_PERM_LOG_RES,
4094                                           XFS_WRITE_LOG_COUNT);
4095                 /*
4096                  * Check for running out of space
4097                  */
4098                 if (error) {
4099                         /*
4100                          * Free the transaction structure.
4101                          */
4102                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4103                         xfs_trans_cancel(tp, 0);
4104                         break;
4105                 }
4106                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4107                 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4108                                                       qblocks, 0, quota_flag);
4109                 if (error)
4110                         goto error1;
4111
4112                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4113                 xfs_trans_ihold(tp, ip);
4114
4115                 /*
4116                  * Issue the xfs_bmapi() call to allocate the blocks
4117                  */
4118                 XFS_BMAP_INIT(&free_list, &firstfsb);
4119                 error = xfs_bmapi(tp, ip, startoffset_fsb,
4120                                   allocatesize_fsb, bmapi_flag,
4121                                   &firstfsb, 0, imapp, &nimaps,
4122                                   &free_list);
4123                 if (error) {
4124                         goto error0;
4125                 }
4126
4127                 /*
4128                  * Complete the transaction
4129                  */
4130                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4131                 if (error) {
4132                         goto error0;
4133                 }
4134
4135                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4136                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4137                 if (error) {
4138                         break;
4139                 }
4140
4141                 allocated_fsb = imapp->br_blockcount;
4142
4143                 if (nimaps == 0) {
4144                         error = XFS_ERROR(ENOSPC);
4145                         break;
4146                 }
4147
4148                 startoffset_fsb += allocated_fsb;
4149                 allocatesize_fsb -= allocated_fsb;
4150         }
4151 dmapi_enospc_check:
4152         if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4153             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4154
4155                 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4156                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4157                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4158                                 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4159                 if (error == 0)
4160                         goto retry;     /* Maybe DMAPI app. has made space */
4161                 /* else fall through with error from XFS_SEND_DATA */
4162         }
4163
4164         return error;
4165
4166 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4167         xfs_bmap_cancel(&free_list);
4168         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4169
4170 error1: /* Just cancel transaction */
4171         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4172         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4173         goto dmapi_enospc_check;
4174 }
4175
4176 /*
4177  * Zero file bytes between startoff and endoff inclusive.
4178  * The iolock is held exclusive and no blocks are buffered.
4179  */
4180 STATIC int
4181 xfs_zero_remaining_bytes(
4182         xfs_inode_t             *ip,
4183         xfs_off_t               startoff,
4184         xfs_off_t               endoff)
4185 {
4186         xfs_bmbt_irec_t         imap;
4187         xfs_fileoff_t           offset_fsb;
4188         xfs_off_t               lastoffset;
4189         xfs_off_t               offset;
4190         xfs_buf_t               *bp;
4191         xfs_mount_t             *mp = ip->i_mount;
4192         int                     nimap;
4193         int                     error = 0;
4194
4195         bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4196                                 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4197                                 mp->m_rtdev_targp : mp->m_ddev_targp);
4198
4199         for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4200                 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4201                 nimap = 1;
4202                 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4203                         &nimap, NULL);
4204                 if (error || nimap < 1)
4205                         break;
4206                 ASSERT(imap.br_blockcount >= 1);
4207                 ASSERT(imap.br_startoff == offset_fsb);
4208                 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4209                 if (lastoffset > endoff)
4210                         lastoffset = endoff;
4211                 if (imap.br_startblock == HOLESTARTBLOCK)
4212                         continue;
4213                 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4214                 if (imap.br_state == XFS_EXT_UNWRITTEN)
4215                         continue;
4216                 XFS_BUF_UNDONE(bp);
4217                 XFS_BUF_UNWRITE(bp);
4218                 XFS_BUF_READ(bp);
4219                 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4220                 xfsbdstrat(mp, bp);
4221                 if ((error = xfs_iowait(bp))) {
4222                         xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4223                                           mp, bp, XFS_BUF_ADDR(bp));
4224                         break;
4225                 }
4226                 memset(XFS_BUF_PTR(bp) +
4227                         (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4228                       0, lastoffset - offset + 1);
4229                 XFS_BUF_UNDONE(bp);
4230                 XFS_BUF_UNREAD(bp);
4231                 XFS_BUF_WRITE(bp);
4232                 xfsbdstrat(mp, bp);
4233                 if ((error = xfs_iowait(bp))) {
4234                         xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4235                                           mp, bp, XFS_BUF_ADDR(bp));
4236                         break;
4237                 }
4238         }
4239         xfs_buf_free(bp);
4240         return error;
4241 }
4242
4243 /*
4244  * xfs_free_file_space()
4245  *      This routine frees disk space for the given file.
4246  *
4247  *      This routine is only called by xfs_change_file_space
4248  *      for an UNRESVSP type call.
4249  *
4250  * RETURNS:
4251  *       0 on success
4252  *      errno on error
4253  *
4254  */
4255 STATIC int
4256 xfs_free_file_space(
4257         xfs_inode_t             *ip,
4258         xfs_off_t               offset,
4259         xfs_off_t               len,
4260         int                     attr_flags)
4261 {
4262         vnode_t                 *vp;
4263         int                     committed;
4264         int                     done;
4265         xfs_off_t               end_dmi_offset;
4266         xfs_fileoff_t           endoffset_fsb;
4267         int                     error;
4268         xfs_fsblock_t           firstfsb;
4269         xfs_bmap_free_t         free_list;
4270         xfs_off_t               ilen;
4271         xfs_bmbt_irec_t         imap;
4272         xfs_off_t               ioffset;
4273         xfs_extlen_t            mod=0;
4274         xfs_mount_t             *mp;
4275         int                     nimap;
4276         uint                    resblks;
4277         int                     rounding;
4278         int                     rt;
4279         xfs_fileoff_t           startoffset_fsb;
4280         xfs_trans_t             *tp;
4281         int                     need_iolock = 1;
4282
4283         vp = XFS_ITOV(ip);
4284         mp = ip->i_mount;
4285
4286         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4287
4288         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4289                 return error;
4290
4291         error = 0;
4292         if (len <= 0)   /* if nothing being freed */
4293                 return error;
4294         rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4295         startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4296         end_dmi_offset = offset + len;
4297         endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4298
4299         if (offset < ip->i_d.di_size &&
4300             (attr_flags & ATTR_DMI) == 0 &&
4301             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4302                 if (end_dmi_offset > ip->i_d.di_size)
4303                         end_dmi_offset = ip->i_d.di_size;
4304                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4305                                 offset, end_dmi_offset - offset,
4306                                 AT_DELAY_FLAG(attr_flags), NULL);
4307                 if (error)
4308                         return(error);
4309         }
4310
4311         ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4312         if (attr_flags & ATTR_NOLOCK)
4313                 need_iolock = 0;
4314         if (need_iolock)
4315                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4316
4317         rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4318                         (__uint8_t)NBPP);
4319         ilen = len + (offset & (rounding - 1));
4320         ioffset = offset & ~(rounding - 1);
4321         if (ilen & (rounding - 1))
4322                 ilen = (ilen + rounding) & ~(rounding - 1);
4323
4324         if (VN_CACHED(vp) != 0) {
4325                 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4326                                 ctooff(offtoct(ioffset)), -1);
4327                 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4328                                 -1, FI_REMAPF_LOCKED);
4329         }
4330
4331         /*
4332          * Need to zero the stuff we're not freeing, on disk.
4333          * If its a realtime file & can't use unwritten extents then we
4334          * actually need to zero the extent edges.  Otherwise xfs_bunmapi
4335          * will take care of it for us.
4336          */
4337         if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4338                 nimap = 1;
4339                 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4340                         &imap, &nimap, NULL);
4341                 if (error)
4342                         goto out_unlock_iolock;
4343                 ASSERT(nimap == 0 || nimap == 1);
4344                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4345                         xfs_daddr_t     block;
4346
4347                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4348                         block = imap.br_startblock;
4349                         mod = do_div(block, mp->m_sb.sb_rextsize);
4350                         if (mod)
4351                                 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4352                 }
4353                 nimap = 1;
4354                 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4355                         &imap, &nimap, NULL);
4356                 if (error)
4357                         goto out_unlock_iolock;
4358                 ASSERT(nimap == 0 || nimap == 1);
4359                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4360                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4361                         mod++;
4362                         if (mod && (mod != mp->m_sb.sb_rextsize))
4363                                 endoffset_fsb -= mod;
4364                 }
4365         }
4366         if ((done = (endoffset_fsb <= startoffset_fsb)))
4367                 /*
4368                  * One contiguous piece to clear
4369                  */
4370                 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4371         else {
4372                 /*
4373                  * Some full blocks, possibly two pieces to clear
4374                  */
4375                 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4376                         error = xfs_zero_remaining_bytes(ip, offset,
4377                                 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4378                 if (!error &&
4379                     XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4380                         error = xfs_zero_remaining_bytes(ip,
4381                                 XFS_FSB_TO_B(mp, endoffset_fsb),
4382                                 offset + len - 1);
4383         }
4384
4385         /*
4386          * free file space until done or until there is an error
4387          */
4388         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4389         while (!error && !done) {
4390
4391                 /*
4392                  * allocate and setup the transaction
4393                  */
4394                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4395                 error = xfs_trans_reserve(tp,
4396                                           resblks,
4397                                           XFS_WRITE_LOG_RES(mp),
4398                                           0,
4399                                           XFS_TRANS_PERM_LOG_RES,
4400                                           XFS_WRITE_LOG_COUNT);
4401
4402                 /*
4403                  * check for running out of space
4404                  */
4405                 if (error) {
4406                         /*
4407                          * Free the transaction structure.
4408                          */
4409                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4410                         xfs_trans_cancel(tp, 0);
4411                         break;
4412                 }
4413                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4414                 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4415                                 ip->i_udquot, ip->i_gdquot, resblks, 0,
4416                                 XFS_QMOPT_RES_REGBLKS);
4417                 if (error)
4418                         goto error1;
4419
4420                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4421                 xfs_trans_ihold(tp, ip);
4422
4423                 /*
4424                  * issue the bunmapi() call to free the blocks
4425                  */
4426                 XFS_BMAP_INIT(&free_list, &firstfsb);
4427                 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4428                                   endoffset_fsb - startoffset_fsb,
4429                                   0, 2, &firstfsb, &free_list, &done);
4430                 if (error) {
4431                         goto error0;
4432                 }
4433
4434                 /*
4435                  * complete the transaction
4436                  */
4437                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4438                 if (error) {
4439                         goto error0;
4440                 }
4441
4442                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4443                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4444         }
4445
4446  out_unlock_iolock:
4447         if (need_iolock)
4448                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4449         return error;
4450
4451  error0:
4452         xfs_bmap_cancel(&free_list);
4453  error1:
4454         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4455         xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4456                     XFS_ILOCK_EXCL);
4457         return error;
4458 }
4459
4460 /*
4461  * xfs_change_file_space()
4462  *      This routine allocates or frees disk space for the given file.
4463  *      The user specified parameters are checked for alignment and size
4464  *      limitations.
4465  *
4466  * RETURNS:
4467  *       0 on success
4468  *      errno on error
4469  *
4470  */
4471 int
4472 xfs_change_file_space(
4473         bhv_desc_t      *bdp,
4474         int             cmd,
4475         xfs_flock64_t   *bf,
4476         xfs_off_t       offset,
4477         cred_t          *credp,
4478         int             attr_flags)
4479 {
4480         int             clrprealloc;
4481         int             error;
4482         xfs_fsize_t     fsize;
4483         xfs_inode_t     *ip;
4484         xfs_mount_t     *mp;
4485         int             setprealloc;
4486         xfs_off_t       startoffset;
4487         xfs_off_t       llen;
4488         xfs_trans_t     *tp;
4489         vattr_t         va;
4490         vnode_t         *vp;
4491
4492         vp = BHV_TO_VNODE(bdp);
4493         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4494
4495         ip = XFS_BHVTOI(bdp);
4496         mp = ip->i_mount;
4497
4498         /*
4499          * must be a regular file and have write permission
4500          */
4501         if (!VN_ISREG(vp))
4502                 return XFS_ERROR(EINVAL);
4503
4504         xfs_ilock(ip, XFS_ILOCK_SHARED);
4505
4506         if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4507                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4508                 return error;
4509         }
4510
4511         xfs_iunlock(ip, XFS_ILOCK_SHARED);
4512
4513         switch (bf->l_whence) {
4514         case 0: /*SEEK_SET*/
4515                 break;
4516         case 1: /*SEEK_CUR*/
4517                 bf->l_start += offset;
4518                 break;
4519         case 2: /*SEEK_END*/
4520                 bf->l_start += ip->i_d.di_size;
4521                 break;
4522         default:
4523                 return XFS_ERROR(EINVAL);
4524         }
4525
4526         llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4527
4528         if (   (bf->l_start < 0)
4529             || (bf->l_start > XFS_MAXIOFFSET(mp))
4530             || (bf->l_start + llen < 0)
4531             || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4532                 return XFS_ERROR(EINVAL);
4533
4534         bf->l_whence = 0;
4535
4536         startoffset = bf->l_start;
4537         fsize = ip->i_d.di_size;
4538
4539         /*
4540          * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4541          * file space.
4542          * These calls do NOT zero the data space allocated to the file,
4543          * nor do they change the file size.
4544          *
4545          * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4546          * space.
4547          * These calls cause the new file data to be zeroed and the file
4548          * size to be changed.
4549          */
4550         setprealloc = clrprealloc = 0;
4551
4552         switch (cmd) {
4553         case XFS_IOC_RESVSP:
4554         case XFS_IOC_RESVSP64:
4555                 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4556                                                                 1, attr_flags);
4557                 if (error)
4558                         return error;
4559                 setprealloc = 1;
4560                 break;
4561
4562         case XFS_IOC_UNRESVSP:
4563         case XFS_IOC_UNRESVSP64:
4564                 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4565                                                                 attr_flags)))
4566                         return error;
4567                 break;
4568
4569         case XFS_IOC_ALLOCSP:
4570         case XFS_IOC_ALLOCSP64:
4571         case XFS_IOC_FREESP:
4572         case XFS_IOC_FREESP64:
4573                 if (startoffset > fsize) {
4574                         error = xfs_alloc_file_space(ip, fsize,
4575                                         startoffset - fsize, 0, attr_flags);
4576                         if (error)
4577                                 break;
4578                 }
4579
4580                 va.va_mask = XFS_AT_SIZE;
4581                 va.va_size = startoffset;
4582
4583                 error = xfs_setattr(bdp, &va, attr_flags, credp);
4584
4585                 if (error)
4586                         return error;
4587
4588                 clrprealloc = 1;
4589                 break;
4590
4591         default:
4592                 ASSERT(0);
4593                 return XFS_ERROR(EINVAL);
4594         }
4595
4596         /*
4597          * update the inode timestamp, mode, and prealloc flag bits
4598          */
4599         tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4600
4601         if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4602                                       0, 0, 0))) {
4603                 /* ASSERT(0); */
4604                 xfs_trans_cancel(tp, 0);
4605                 return error;
4606         }
4607
4608         xfs_ilock(ip, XFS_ILOCK_EXCL);
4609
4610         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4611         xfs_trans_ihold(tp, ip);
4612
4613         if ((attr_flags & ATTR_DMI) == 0) {
4614                 ip->i_d.di_mode &= ~S_ISUID;
4615
4616                 /*
4617                  * Note that we don't have to worry about mandatory
4618                  * file locking being disabled here because we only
4619                  * clear the S_ISGID bit if the Group execute bit is
4620                  * on, but if it was on then mandatory locking wouldn't
4621                  * have been enabled.
4622                  */
4623                 if (ip->i_d.di_mode & S_IXGRP)
4624                         ip->i_d.di_mode &= ~S_ISGID;
4625
4626                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4627         }
4628         if (setprealloc)
4629                 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4630         else if (clrprealloc)
4631                 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4632
4633         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4634         xfs_trans_set_sync(tp);
4635
4636         error = xfs_trans_commit(tp, 0, NULL);
4637
4638         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4639
4640         return error;
4641 }
4642
4643 vnodeops_t xfs_vnodeops = {
4644         BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4645         .vop_open               = xfs_open,
4646         .vop_read               = xfs_read,
4647 #ifdef HAVE_SENDFILE
4648         .vop_sendfile           = xfs_sendfile,
4649 #endif
4650         .vop_write              = xfs_write,
4651         .vop_ioctl              = xfs_ioctl,
4652         .vop_getattr            = xfs_getattr,
4653         .vop_setattr            = xfs_setattr,
4654         .vop_access             = xfs_access,
4655         .vop_lookup             = xfs_lookup,
4656         .vop_create             = xfs_create,
4657         .vop_remove             = xfs_remove,
4658         .vop_link               = xfs_link,
4659         .vop_rename             = xfs_rename,
4660         .vop_mkdir              = xfs_mkdir,
4661         .vop_rmdir              = xfs_rmdir,
4662         .vop_readdir            = xfs_readdir,
4663         .vop_symlink            = xfs_symlink,
4664         .vop_readlink           = xfs_readlink,
4665         .vop_fsync              = xfs_fsync,
4666         .vop_inactive           = xfs_inactive,
4667         .vop_fid2               = xfs_fid2,
4668         .vop_rwlock             = xfs_rwlock,
4669         .vop_rwunlock           = xfs_rwunlock,
4670         .vop_bmap               = xfs_bmap,
4671         .vop_reclaim            = xfs_reclaim,
4672         .vop_attr_get           = xfs_attr_get,
4673         .vop_attr_set           = xfs_attr_set,
4674         .vop_attr_remove        = xfs_attr_remove,
4675         .vop_attr_list          = xfs_attr_list,
4676         .vop_link_removed       = (vop_link_removed_t)fs_noval,
4677         .vop_vnode_change       = (vop_vnode_change_t)fs_noval,
4678         .vop_tosspages          = fs_tosspages,
4679         .vop_flushinval_pages   = fs_flushinval_pages,
4680         .vop_flush_pages        = fs_flush_pages,
4681         .vop_release            = xfs_release,
4682         .vop_iflush             = xfs_inode_flush,
4683 };