Merge master.kernel.org:/pub/scm/linux/kernel/git/dwmw2/audit-2.6
[linux-2.6] / fs / xfs / xfs_vnodeops.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32
33 #include "xfs.h"
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
36 #include "xfs_inum.h"
37 #include "xfs_log.h"
38 #include "xfs_trans.h"
39 #include "xfs_sb.h"
40 #include "xfs_ag.h"
41 #include "xfs_dir.h"
42 #include "xfs_dir2.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_itable.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_alloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
58 #include "xfs_bmap.h"
59 #include "xfs_da_btree.h"
60 #include "xfs_attr.h"
61 #include "xfs_rw.h"
62 #include "xfs_refcache.h"
63 #include "xfs_error.h"
64 #include "xfs_bit.h"
65 #include "xfs_rtalloc.h"
66 #include "xfs_quota.h"
67 #include "xfs_utils.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_dir_leaf.h"
70 #include "xfs_mac.h"
71 #include "xfs_log_priv.h"
72
73
74 /*
75  * The maximum pathlen is 1024 bytes. Since the minimum file system
76  * blocksize is 512 bytes, we can get a max of 2 extents back from
77  * bmapi.
78  */
79 #define SYMLINK_MAPS 2
80
81 /*
82  * For xfs, we check that the file isn't too big to be opened by this kernel.
83  * No other open action is required for regular files.  Devices are handled
84  * through the specfs file system, pipes through fifofs.  Device and
85  * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
86  * when a new vnode is first looked up or created.
87  */
88 STATIC int
89 xfs_open(
90         bhv_desc_t      *bdp,
91         cred_t          *credp)
92 {
93         int             mode;
94         vnode_t         *vp;
95         xfs_inode_t     *ip;
96
97         vp = BHV_TO_VNODE(bdp);
98         ip = XFS_BHVTOI(bdp);
99
100         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
101                 return XFS_ERROR(EIO);
102
103         /*
104          * If it's a directory with any blocks, read-ahead block 0
105          * as we're almost certain to have the next operation be a read there.
106          */
107         if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
108                 mode = xfs_ilock_map_shared(ip);
109                 if (ip->i_d.di_nextents > 0)
110                         (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
111                 xfs_iunlock(ip, mode);
112         }
113         return 0;
114 }
115
116
117 /*
118  * xfs_getattr
119  */
120 STATIC int
121 xfs_getattr(
122         bhv_desc_t      *bdp,
123         vattr_t         *vap,
124         int             flags,
125         cred_t          *credp)
126 {
127         xfs_inode_t     *ip;
128         xfs_mount_t     *mp;
129         vnode_t         *vp;
130
131         vp  = BHV_TO_VNODE(bdp);
132         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
133
134         ip = XFS_BHVTOI(bdp);
135         mp = ip->i_mount;
136
137         if (XFS_FORCED_SHUTDOWN(mp))
138                 return XFS_ERROR(EIO);
139
140         if (!(flags & ATTR_LAZY))
141                 xfs_ilock(ip, XFS_ILOCK_SHARED);
142
143         vap->va_size = ip->i_d.di_size;
144         if (vap->va_mask == XFS_AT_SIZE)
145                 goto all_done;
146
147         vap->va_nblocks =
148                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
149         vap->va_nodeid = ip->i_ino;
150 #if XFS_BIG_INUMS
151         vap->va_nodeid += mp->m_inoadd;
152 #endif
153         vap->va_nlink = ip->i_d.di_nlink;
154
155         /*
156          * Quick exit for non-stat callers
157          */
158         if ((vap->va_mask &
159             ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
160               XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
161                 goto all_done;
162
163         /*
164          * Copy from in-core inode.
165          */
166         vap->va_mode = ip->i_d.di_mode;
167         vap->va_uid = ip->i_d.di_uid;
168         vap->va_gid = ip->i_d.di_gid;
169         vap->va_projid = ip->i_d.di_projid;
170
171         /*
172          * Check vnode type block/char vs. everything else.
173          */
174         switch (ip->i_d.di_mode & S_IFMT) {
175         case S_IFBLK:
176         case S_IFCHR:
177                 vap->va_rdev = ip->i_df.if_u2.if_rdev;
178                 vap->va_blocksize = BLKDEV_IOSIZE;
179                 break;
180         default:
181                 vap->va_rdev = 0;
182
183                 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
184
185 #if 0
186                         /* Large block sizes confuse various
187                          * user space programs, so letting the
188                          * stripe size through is not a good
189                          * idea for now.
190                          */
191                         vap->va_blocksize = mp->m_swidth ?
192                                 /*
193                                  * If the underlying volume is a stripe, then
194                                  * return the stripe width in bytes as the
195                                  * recommended I/O size.
196                                  */
197                                 (mp->m_swidth << mp->m_sb.sb_blocklog) :
198                                 /*
199                                  * Return the largest of the preferred buffer
200                                  * sizes since doing small I/Os into larger
201                                  * buffers causes buffers to be decommissioned.
202                                  * The value returned is in bytes.
203                                  */
204                                 (1 << (int)MAX(mp->m_readio_log,
205                                                mp->m_writeio_log));
206
207 #else
208                         vap->va_blocksize =
209                                 /*
210                                  * Return the largest of the preferred buffer
211                                  * sizes since doing small I/Os into larger
212                                  * buffers causes buffers to be decommissioned.
213                                  * The value returned is in bytes.
214                                  */
215                                 1 << (int)MAX(mp->m_readio_log,
216                                                mp->m_writeio_log);
217 #endif
218                 } else {
219
220                         /*
221                          * If the file blocks are being allocated from a
222                          * realtime partition, then return the inode's
223                          * realtime extent size or the realtime volume's
224                          * extent size.
225                          */
226                         vap->va_blocksize = ip->i_d.di_extsize ?
227                                 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
228                                 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
229                 }
230                 break;
231         }
232
233         vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
234         vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
235         vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
236         vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
237         vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
238         vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
239
240         /*
241          * Exit for stat callers.  See if any of the rest of the fields
242          * to be filled in are needed.
243          */
244         if ((vap->va_mask &
245              (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
246               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
247                 goto all_done;
248
249         /*
250          * Convert di_flags to xflags.
251          */
252         vap->va_xflags = xfs_ip2xflags(ip);
253
254         /*
255          * Exit for inode revalidate.  See if any of the rest of
256          * the fields to be filled in are needed.
257          */
258         if ((vap->va_mask &
259              (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
260               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
261                 goto all_done;
262
263         vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
264         vap->va_nextents =
265                 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
266                         ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
267                         ip->i_d.di_nextents;
268         if (ip->i_afp)
269                 vap->va_anextents =
270                         (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
271                                 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
272                                  ip->i_d.di_anextents;
273         else
274                 vap->va_anextents = 0;
275         vap->va_gen = ip->i_d.di_gen;
276
277  all_done:
278         if (!(flags & ATTR_LAZY))
279                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
280         return 0;
281 }
282
283
284 /*
285  * xfs_setattr
286  */
287 int
288 xfs_setattr(
289         bhv_desc_t              *bdp,
290         vattr_t                 *vap,
291         int                     flags,
292         cred_t                  *credp)
293 {
294         xfs_inode_t             *ip;
295         xfs_trans_t             *tp;
296         xfs_mount_t             *mp;
297         int                     mask;
298         int                     code;
299         uint                    lock_flags;
300         uint                    commit_flags=0;
301         uid_t                   uid=0, iuid=0;
302         gid_t                   gid=0, igid=0;
303         int                     timeflags = 0;
304         vnode_t                 *vp;
305         xfs_prid_t              projid=0, iprojid=0;
306         int                     mandlock_before, mandlock_after;
307         struct xfs_dquot        *udqp, *gdqp, *olddquot1, *olddquot2;
308         int                     file_owner;
309         int                     need_iolock = 1;
310
311         vp = BHV_TO_VNODE(bdp);
312         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
313
314         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
315                 return XFS_ERROR(EROFS);
316
317         /*
318          * Cannot set certain attributes.
319          */
320         mask = vap->va_mask;
321         if (mask & XFS_AT_NOSET) {
322                 return XFS_ERROR(EINVAL);
323         }
324
325         ip = XFS_BHVTOI(bdp);
326         mp = ip->i_mount;
327
328         if (XFS_FORCED_SHUTDOWN(mp))
329                 return XFS_ERROR(EIO);
330
331         /*
332          * Timestamps do not need to be logged and hence do not
333          * need to be done within a transaction.
334          */
335         if (mask & XFS_AT_UPDTIMES) {
336                 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
337                 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
338                             ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
339                             ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
340                 xfs_ichgtime(ip, timeflags);
341                 return 0;
342         }
343
344         olddquot1 = olddquot2 = NULL;
345         udqp = gdqp = NULL;
346
347         /*
348          * If disk quotas is on, we make sure that the dquots do exist on disk,
349          * before we start any other transactions. Trying to do this later
350          * is messy. We don't care to take a readlock to look at the ids
351          * in inode here, because we can't hold it across the trans_reserve.
352          * If the IDs do change before we take the ilock, we're covered
353          * because the i_*dquot fields will get updated anyway.
354          */
355         if (XFS_IS_QUOTA_ON(mp) &&
356             (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
357                 uint    qflags = 0;
358
359                 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
360                         uid = vap->va_uid;
361                         qflags |= XFS_QMOPT_UQUOTA;
362                 } else {
363                         uid = ip->i_d.di_uid;
364                 }
365                 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
366                         gid = vap->va_gid;
367                         qflags |= XFS_QMOPT_GQUOTA;
368                 }  else {
369                         gid = ip->i_d.di_gid;
370                 }
371                 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
372                         projid = vap->va_projid;
373                         qflags |= XFS_QMOPT_PQUOTA;
374                 }  else {
375                         projid = ip->i_d.di_projid;
376                 }
377                 /*
378                  * We take a reference when we initialize udqp and gdqp,
379                  * so it is important that we never blindly double trip on
380                  * the same variable. See xfs_create() for an example.
381                  */
382                 ASSERT(udqp == NULL);
383                 ASSERT(gdqp == NULL);
384                 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
385                                          &udqp, &gdqp);
386                 if (code)
387                         return (code);
388         }
389
390         /*
391          * For the other attributes, we acquire the inode lock and
392          * first do an error checking pass.
393          */
394         tp = NULL;
395         lock_flags = XFS_ILOCK_EXCL;
396         ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
397         if (flags & ATTR_NOLOCK)
398                 need_iolock = 0;
399         if (!(mask & XFS_AT_SIZE)) {
400                 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
401                     (mp->m_flags & XFS_MOUNT_WSYNC)) {
402                         tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
403                         commit_flags = 0;
404                         if ((code = xfs_trans_reserve(tp, 0,
405                                                      XFS_ICHANGE_LOG_RES(mp), 0,
406                                                      0, 0))) {
407                                 lock_flags = 0;
408                                 goto error_return;
409                         }
410                 }
411         } else {
412                 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
413                     !(flags & ATTR_DMI)) {
414                         int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
415                         code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
416                                 vap->va_size, 0, dmflags, NULL);
417                         if (code) {
418                                 lock_flags = 0;
419                                 goto error_return;
420                         }
421                 }
422                 if (need_iolock)
423                         lock_flags |= XFS_IOLOCK_EXCL;
424         }
425
426         xfs_ilock(ip, lock_flags);
427
428         /* boolean: are we the file owner? */
429         file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
430
431         /*
432          * Change various properties of a file.
433          * Only the owner or users with CAP_FOWNER
434          * capability may do these things.
435          */
436         if (mask &
437             (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
438              XFS_AT_GID|XFS_AT_PROJID)) {
439                 /*
440                  * CAP_FOWNER overrides the following restrictions:
441                  *
442                  * The user ID of the calling process must be equal
443                  * to the file owner ID, except in cases where the
444                  * CAP_FSETID capability is applicable.
445                  */
446                 if (!file_owner && !capable(CAP_FOWNER)) {
447                         code = XFS_ERROR(EPERM);
448                         goto error_return;
449                 }
450
451                 /*
452                  * CAP_FSETID overrides the following restrictions:
453                  *
454                  * The effective user ID of the calling process shall match
455                  * the file owner when setting the set-user-ID and
456                  * set-group-ID bits on that file.
457                  *
458                  * The effective group ID or one of the supplementary group
459                  * IDs of the calling process shall match the group owner of
460                  * the file when setting the set-group-ID bit on that file
461                  */
462                 if (mask & XFS_AT_MODE) {
463                         mode_t m = 0;
464
465                         if ((vap->va_mode & S_ISUID) && !file_owner)
466                                 m |= S_ISUID;
467                         if ((vap->va_mode & S_ISGID) &&
468                             !in_group_p((gid_t)ip->i_d.di_gid))
469                                 m |= S_ISGID;
470 #if 0
471                         /* Linux allows this, Irix doesn't. */
472                         if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
473                                 m |= S_ISVTX;
474 #endif
475                         if (m && !capable(CAP_FSETID))
476                                 vap->va_mode &= ~m;
477                 }
478         }
479
480         /*
481          * Change file ownership.  Must be the owner or privileged.
482          * If the system was configured with the "restricted_chown"
483          * option, the owner is not permitted to give away the file,
484          * and can change the group id only to a group of which he
485          * or she is a member.
486          */
487         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
488                 /*
489                  * These IDs could have changed since we last looked at them.
490                  * But, we're assured that if the ownership did change
491                  * while we didn't have the inode locked, inode's dquot(s)
492                  * would have changed also.
493                  */
494                 iuid = ip->i_d.di_uid;
495                 iprojid = ip->i_d.di_projid;
496                 igid = ip->i_d.di_gid;
497                 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
498                 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
499                 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
500                          iprojid;
501
502                 /*
503                  * CAP_CHOWN overrides the following restrictions:
504                  *
505                  * If _POSIX_CHOWN_RESTRICTED is defined, this capability
506                  * shall override the restriction that a process cannot
507                  * change the user ID of a file it owns and the restriction
508                  * that the group ID supplied to the chown() function
509                  * shall be equal to either the group ID or one of the
510                  * supplementary group IDs of the calling process.
511                  */
512                 if (restricted_chown &&
513                     (iuid != uid || (igid != gid &&
514                                      !in_group_p((gid_t)gid))) &&
515                     !capable(CAP_CHOWN)) {
516                         code = XFS_ERROR(EPERM);
517                         goto error_return;
518                 }
519                 /*
520                  * Do a quota reservation only if uid/projid/gid is actually
521                  * going to change.
522                  */
523                 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
524                     (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
525                     (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
526                         ASSERT(tp);
527                         code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
528                                                 capable(CAP_FOWNER) ?
529                                                 XFS_QMOPT_FORCE_RES : 0);
530                         if (code)       /* out of quota */
531                                 goto error_return;
532                 }
533         }
534
535         /*
536          * Truncate file.  Must have write permission and not be a directory.
537          */
538         if (mask & XFS_AT_SIZE) {
539                 /* Short circuit the truncate case for zero length files */
540                 if ((vap->va_size == 0) &&
541                    (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
542                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
543                         lock_flags &= ~XFS_ILOCK_EXCL;
544                         if (mask & XFS_AT_CTIME)
545                                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
546                         code = 0;
547                         goto error_return;
548                 }
549
550                 if (VN_ISDIR(vp)) {
551                         code = XFS_ERROR(EISDIR);
552                         goto error_return;
553                 } else if (!VN_ISREG(vp)) {
554                         code = XFS_ERROR(EINVAL);
555                         goto error_return;
556                 }
557                 /*
558                  * Make sure that the dquots are attached to the inode.
559                  */
560                 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
561                         goto error_return;
562         }
563
564         /*
565          * Change file access or modified times.
566          */
567         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
568                 if (!file_owner) {
569                         if ((flags & ATTR_UTIME) &&
570                             !capable(CAP_FOWNER)) {
571                                 code = XFS_ERROR(EPERM);
572                                 goto error_return;
573                         }
574                 }
575         }
576
577         /*
578          * Change extent size or realtime flag.
579          */
580         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
581                 /*
582                  * Can't change extent size if any extents are allocated.
583                  */
584                 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
585                     (mask & XFS_AT_EXTSIZE) &&
586                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
587                      vap->va_extsize) ) {
588                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
589                         goto error_return;
590                 }
591
592                 /*
593                  * Can't set extent size unless the file is marked, or
594                  * about to be marked as a realtime file.
595                  *
596                  * This check will be removed when fixed size extents
597                  * with buffered data writes is implemented.
598                  *
599                  */
600                 if ((mask & XFS_AT_EXTSIZE)                     &&
601                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
602                      vap->va_extsize) &&
603                     (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
604                        ((mask & XFS_AT_XFLAGS) &&
605                         (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
606                         code = XFS_ERROR(EINVAL);
607                         goto error_return;
608                 }
609
610                 /*
611                  * Can't change realtime flag if any extents are allocated.
612                  */
613                 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
614                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
615                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
616                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
617                         goto error_return;
618                 }
619                 /*
620                  * Extent size must be a multiple of the appropriate block
621                  * size, if set at all.
622                  */
623                 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
624                         xfs_extlen_t    size;
625
626                         if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
627                             ((mask & XFS_AT_XFLAGS) &&
628                             (vap->va_xflags & XFS_XFLAG_REALTIME))) {
629                                 size = mp->m_sb.sb_rextsize <<
630                                        mp->m_sb.sb_blocklog;
631                         } else {
632                                 size = mp->m_sb.sb_blocksize;
633                         }
634                         if (vap->va_extsize % size) {
635                                 code = XFS_ERROR(EINVAL);
636                                 goto error_return;
637                         }
638                 }
639                 /*
640                  * If realtime flag is set then must have realtime data.
641                  */
642                 if ((mask & XFS_AT_XFLAGS) &&
643                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
644                         if ((mp->m_sb.sb_rblocks == 0) ||
645                             (mp->m_sb.sb_rextsize == 0) ||
646                             (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
647                                 code = XFS_ERROR(EINVAL);
648                                 goto error_return;
649                         }
650                 }
651
652                 /*
653                  * Can't modify an immutable/append-only file unless
654                  * we have appropriate permission.
655                  */
656                 if ((mask & XFS_AT_XFLAGS) &&
657                     (ip->i_d.di_flags &
658                                 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
659                      (vap->va_xflags &
660                                 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
661                     !capable(CAP_LINUX_IMMUTABLE)) {
662                         code = XFS_ERROR(EPERM);
663                         goto error_return;
664                 }
665         }
666
667         /*
668          * Now we can make the changes.  Before we join the inode
669          * to the transaction, if XFS_AT_SIZE is set then take care of
670          * the part of the truncation that must be done without the
671          * inode lock.  This needs to be done before joining the inode
672          * to the transaction, because the inode cannot be unlocked
673          * once it is a part of the transaction.
674          */
675         if (mask & XFS_AT_SIZE) {
676                 code = 0;
677                 if (vap->va_size > ip->i_d.di_size)
678                         code = xfs_igrow_start(ip, vap->va_size, credp);
679                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
680                 if (!code)
681                         code = xfs_itruncate_data(ip, vap->va_size);
682                 if (code) {
683                         ASSERT(tp == NULL);
684                         lock_flags &= ~XFS_ILOCK_EXCL;
685                         ASSERT(lock_flags == XFS_IOLOCK_EXCL);
686                         goto error_return;
687                 }
688                 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
689                 if ((code = xfs_trans_reserve(tp, 0,
690                                              XFS_ITRUNCATE_LOG_RES(mp), 0,
691                                              XFS_TRANS_PERM_LOG_RES,
692                                              XFS_ITRUNCATE_LOG_COUNT))) {
693                         xfs_trans_cancel(tp, 0);
694                         if (need_iolock)
695                                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
696                         return code;
697                 }
698                 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
699                 xfs_ilock(ip, XFS_ILOCK_EXCL);
700         }
701
702         if (tp) {
703                 xfs_trans_ijoin(tp, ip, lock_flags);
704                 xfs_trans_ihold(tp, ip);
705         }
706
707         /* determine whether mandatory locking mode changes */
708         mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
709
710         /*
711          * Truncate file.  Must have write permission and not be a directory.
712          */
713         if (mask & XFS_AT_SIZE) {
714                 if (vap->va_size > ip->i_d.di_size) {
715                         xfs_igrow_finish(tp, ip, vap->va_size,
716                             !(flags & ATTR_DMI));
717                 } else if ((vap->va_size <= ip->i_d.di_size) ||
718                            ((vap->va_size == 0) && ip->i_d.di_nextents)) {
719                         /*
720                          * signal a sync transaction unless
721                          * we're truncating an already unlinked
722                          * file on a wsync filesystem
723                          */
724                         code = xfs_itruncate_finish(&tp, ip,
725                                             (xfs_fsize_t)vap->va_size,
726                                             XFS_DATA_FORK,
727                                             ((ip->i_d.di_nlink != 0 ||
728                                               !(mp->m_flags & XFS_MOUNT_WSYNC))
729                                              ? 1 : 0));
730                         if (code) {
731                                 goto abort_return;
732                         }
733                 }
734                 /*
735                  * Have to do this even if the file's size doesn't change.
736                  */
737                 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
738         }
739
740         /*
741          * Change file access modes.
742          */
743         if (mask & XFS_AT_MODE) {
744                 ip->i_d.di_mode &= S_IFMT;
745                 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
746
747                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
748                 timeflags |= XFS_ICHGTIME_CHG;
749         }
750
751         /*
752          * Change file ownership.  Must be the owner or privileged.
753          * If the system was configured with the "restricted_chown"
754          * option, the owner is not permitted to give away the file,
755          * and can change the group id only to a group of which he
756          * or she is a member.
757          */
758         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
759                 /*
760                  * CAP_FSETID overrides the following restrictions:
761                  *
762                  * The set-user-ID and set-group-ID bits of a file will be
763                  * cleared upon successful return from chown()
764                  */
765                 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
766                     !capable(CAP_FSETID)) {
767                         ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
768                 }
769
770                 /*
771                  * Change the ownerships and register quota modifications
772                  * in the transaction.
773                  */
774                 if (iuid != uid) {
775                         if (XFS_IS_UQUOTA_ON(mp)) {
776                                 ASSERT(mask & XFS_AT_UID);
777                                 ASSERT(udqp);
778                                 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
779                                                         &ip->i_udquot, udqp);
780                         }
781                         ip->i_d.di_uid = uid;
782                 }
783                 if (igid != gid) {
784                         if (XFS_IS_GQUOTA_ON(mp)) {
785                                 ASSERT(!XFS_IS_PQUOTA_ON(mp));
786                                 ASSERT(mask & XFS_AT_GID);
787                                 ASSERT(gdqp);
788                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
789                                                         &ip->i_gdquot, gdqp);
790                         }
791                         ip->i_d.di_gid = gid;
792                 }
793                 if (iprojid != projid) {
794                         if (XFS_IS_PQUOTA_ON(mp)) {
795                                 ASSERT(!XFS_IS_GQUOTA_ON(mp));
796                                 ASSERT(mask & XFS_AT_PROJID);
797                                 ASSERT(gdqp);
798                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
799                                                         &ip->i_gdquot, gdqp);
800                         }
801                         ip->i_d.di_projid = projid;
802                         /*
803                          * We may have to rev the inode as well as
804                          * the superblock version number since projids didn't
805                          * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
806                          */
807                         if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
808                                 xfs_bump_ino_vers2(tp, ip);
809                 }
810
811                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
812                 timeflags |= XFS_ICHGTIME_CHG;
813         }
814
815
816         /*
817          * Change file access or modified times.
818          */
819         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
820                 if (mask & XFS_AT_ATIME) {
821                         ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
822                         ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
823                         ip->i_update_core = 1;
824                         timeflags &= ~XFS_ICHGTIME_ACC;
825                 }
826                 if (mask & XFS_AT_MTIME) {
827                         ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
828                         ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
829                         timeflags &= ~XFS_ICHGTIME_MOD;
830                         timeflags |= XFS_ICHGTIME_CHG;
831                 }
832                 if (tp && (flags & ATTR_UTIME))
833                         xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
834         }
835
836         /*
837          * Change XFS-added attributes.
838          */
839         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
840                 if (mask & XFS_AT_EXTSIZE) {
841                         /*
842                          * Converting bytes to fs blocks.
843                          */
844                         ip->i_d.di_extsize = vap->va_extsize >>
845                                 mp->m_sb.sb_blocklog;
846                 }
847                 if (mask & XFS_AT_XFLAGS) {
848                         uint    di_flags;
849
850                         /* can't set PREALLOC this way, just preserve it */
851                         di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
852                         if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
853                                 di_flags |= XFS_DIFLAG_IMMUTABLE;
854                         if (vap->va_xflags & XFS_XFLAG_APPEND)
855                                 di_flags |= XFS_DIFLAG_APPEND;
856                         if (vap->va_xflags & XFS_XFLAG_SYNC)
857                                 di_flags |= XFS_DIFLAG_SYNC;
858                         if (vap->va_xflags & XFS_XFLAG_NOATIME)
859                                 di_flags |= XFS_DIFLAG_NOATIME;
860                         if (vap->va_xflags & XFS_XFLAG_NODUMP)
861                                 di_flags |= XFS_DIFLAG_NODUMP;
862                         if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
863                                 di_flags |= XFS_DIFLAG_PROJINHERIT;
864                         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
865                                 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
866                                         di_flags |= XFS_DIFLAG_RTINHERIT;
867                                 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
868                                         di_flags |= XFS_DIFLAG_NOSYMLINKS;
869                         } else {
870                                 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
871                                         di_flags |= XFS_DIFLAG_REALTIME;
872                                         ip->i_iocore.io_flags |= XFS_IOCORE_RT;
873                                 } else {
874                                         ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
875                                 }
876                         }
877                         ip->i_d.di_flags = di_flags;
878                 }
879                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
880                 timeflags |= XFS_ICHGTIME_CHG;
881         }
882
883         /*
884          * Change file inode change time only if XFS_AT_CTIME set
885          * AND we have been called by a DMI function.
886          */
887
888         if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
889                 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
890                 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
891                 ip->i_update_core = 1;
892                 timeflags &= ~XFS_ICHGTIME_CHG;
893         }
894
895         /*
896          * Send out timestamp changes that need to be set to the
897          * current time.  Not done when called by a DMI function.
898          */
899         if (timeflags && !(flags & ATTR_DMI))
900                 xfs_ichgtime(ip, timeflags);
901
902         XFS_STATS_INC(xs_ig_attrchg);
903
904         /*
905          * If this is a synchronous mount, make sure that the
906          * transaction goes to disk before returning to the user.
907          * This is slightly sub-optimal in that truncates require
908          * two sync transactions instead of one for wsync filesytems.
909          * One for the truncate and one for the timestamps since we
910          * don't want to change the timestamps unless we're sure the
911          * truncate worked.  Truncates are less than 1% of the laddis
912          * mix so this probably isn't worth the trouble to optimize.
913          */
914         code = 0;
915         if (tp) {
916                 if (mp->m_flags & XFS_MOUNT_WSYNC)
917                         xfs_trans_set_sync(tp);
918
919                 code = xfs_trans_commit(tp, commit_flags, NULL);
920         }
921
922         /*
923          * If the (regular) file's mandatory locking mode changed, then
924          * notify the vnode.  We do this under the inode lock to prevent
925          * racing calls to vop_vnode_change.
926          */
927         mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
928         if (mandlock_before != mandlock_after) {
929                 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
930                                  mandlock_after);
931         }
932
933         xfs_iunlock(ip, lock_flags);
934
935         /*
936          * Release any dquot(s) the inode had kept before chown.
937          */
938         XFS_QM_DQRELE(mp, olddquot1);
939         XFS_QM_DQRELE(mp, olddquot2);
940         XFS_QM_DQRELE(mp, udqp);
941         XFS_QM_DQRELE(mp, gdqp);
942
943         if (code) {
944                 return code;
945         }
946
947         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
948             !(flags & ATTR_DMI)) {
949                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
950                                         NULL, DM_RIGHT_NULL, NULL, NULL,
951                                         0, 0, AT_DELAY_FLAG(flags));
952         }
953         return 0;
954
955  abort_return:
956         commit_flags |= XFS_TRANS_ABORT;
957         /* FALLTHROUGH */
958  error_return:
959         XFS_QM_DQRELE(mp, udqp);
960         XFS_QM_DQRELE(mp, gdqp);
961         if (tp) {
962                 xfs_trans_cancel(tp, commit_flags);
963         }
964         if (lock_flags != 0) {
965                 xfs_iunlock(ip, lock_flags);
966         }
967         return code;
968 }
969
970
971 /*
972  * xfs_access
973  * Null conversion from vnode mode bits to inode mode bits, as in efs.
974  */
975 STATIC int
976 xfs_access(
977         bhv_desc_t      *bdp,
978         int             mode,
979         cred_t          *credp)
980 {
981         xfs_inode_t     *ip;
982         int             error;
983
984         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
985                                                (inst_t *)__return_address);
986
987         ip = XFS_BHVTOI(bdp);
988         xfs_ilock(ip, XFS_ILOCK_SHARED);
989         error = xfs_iaccess(ip, mode, credp);
990         xfs_iunlock(ip, XFS_ILOCK_SHARED);
991         return error;
992 }
993
994
995 /*
996  * xfs_readlink
997  *
998  */
999 STATIC int
1000 xfs_readlink(
1001         bhv_desc_t      *bdp,
1002         uio_t           *uiop,
1003         int             ioflags,
1004         cred_t          *credp)
1005 {
1006         xfs_inode_t     *ip;
1007         int             count;
1008         xfs_off_t       offset;
1009         int             pathlen;
1010         vnode_t         *vp;
1011         int             error = 0;
1012         xfs_mount_t     *mp;
1013         int             nmaps;
1014         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1015         xfs_daddr_t     d;
1016         int             byte_cnt;
1017         int             n;
1018         xfs_buf_t       *bp;
1019
1020         vp = BHV_TO_VNODE(bdp);
1021         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1022
1023         ip = XFS_BHVTOI(bdp);
1024         mp = ip->i_mount;
1025
1026         if (XFS_FORCED_SHUTDOWN(mp))
1027                 return XFS_ERROR(EIO);
1028
1029         xfs_ilock(ip, XFS_ILOCK_SHARED);
1030
1031         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1032
1033         offset = uiop->uio_offset;
1034         count = uiop->uio_resid;
1035
1036         if (offset < 0) {
1037                 error = XFS_ERROR(EINVAL);
1038                 goto error_return;
1039         }
1040         if (count <= 0) {
1041                 error = 0;
1042                 goto error_return;
1043         }
1044
1045         if (!(ioflags & IO_INVIS)) {
1046                 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1047         }
1048
1049         /*
1050          * See if the symlink is stored inline.
1051          */
1052         pathlen = (int)ip->i_d.di_size;
1053
1054         if (ip->i_df.if_flags & XFS_IFINLINE) {
1055                 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1056         }
1057         else {
1058                 /*
1059                  * Symlink not inline.  Call bmap to get it in.
1060                  */
1061                 nmaps = SYMLINK_MAPS;
1062
1063                 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1064                                   0, NULL, 0, mval, &nmaps, NULL);
1065
1066                 if (error) {
1067                         goto error_return;
1068                 }
1069
1070                 for (n = 0; n < nmaps; n++) {
1071                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1072                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1073                         bp = xfs_buf_read(mp->m_ddev_targp, d,
1074                                       BTOBB(byte_cnt), 0);
1075                         error = XFS_BUF_GETERROR(bp);
1076                         if (error) {
1077                                 xfs_ioerror_alert("xfs_readlink",
1078                                           ip->i_mount, bp, XFS_BUF_ADDR(bp));
1079                                 xfs_buf_relse(bp);
1080                                 goto error_return;
1081                         }
1082                         if (pathlen < byte_cnt)
1083                                 byte_cnt = pathlen;
1084                         pathlen -= byte_cnt;
1085
1086                         error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1087                         xfs_buf_relse (bp);
1088                 }
1089
1090         }
1091
1092
1093 error_return:
1094
1095         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1096
1097         return error;
1098 }
1099
1100
1101 /*
1102  * xfs_fsync
1103  *
1104  * This is called to sync the inode and its data out to disk.
1105  * We need to hold the I/O lock while flushing the data, and
1106  * the inode lock while flushing the inode.  The inode lock CANNOT
1107  * be held while flushing the data, so acquire after we're done
1108  * with that.
1109  */
1110 STATIC int
1111 xfs_fsync(
1112         bhv_desc_t      *bdp,
1113         int             flag,
1114         cred_t          *credp,
1115         xfs_off_t       start,
1116         xfs_off_t       stop)
1117 {
1118         xfs_inode_t     *ip;
1119         xfs_trans_t     *tp;
1120         int             error;
1121
1122         vn_trace_entry(BHV_TO_VNODE(bdp),
1123                         __FUNCTION__, (inst_t *)__return_address);
1124
1125         ip = XFS_BHVTOI(bdp);
1126
1127         ASSERT(start >= 0 && stop >= -1);
1128
1129         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1130                 return XFS_ERROR(EIO);
1131
1132         /*
1133          * We always need to make sure that the required inode state
1134          * is safe on disk.  The vnode might be clean but because
1135          * of committed transactions that haven't hit the disk yet.
1136          * Likewise, there could be unflushed non-transactional
1137          * changes to the inode core that have to go to disk.
1138          *
1139          * The following code depends on one assumption:  that
1140          * any transaction that changes an inode logs the core
1141          * because it has to change some field in the inode core
1142          * (typically nextents or nblocks).  That assumption
1143          * implies that any transactions against an inode will
1144          * catch any non-transactional updates.  If inode-altering
1145          * transactions exist that violate this assumption, the
1146          * code breaks.  Right now, it figures that if the involved
1147          * update_* field is clear and the inode is unpinned, the
1148          * inode is clean.  Either it's been flushed or it's been
1149          * committed and the commit has hit the disk unpinning the inode.
1150          * (Note that xfs_inode_item_format() called at commit clears
1151          * the update_* fields.)
1152          */
1153         xfs_ilock(ip, XFS_ILOCK_SHARED);
1154
1155         /* If we are flushing data then we care about update_size
1156          * being set, otherwise we care about update_core
1157          */
1158         if ((flag & FSYNC_DATA) ?
1159                         (ip->i_update_size == 0) :
1160                         (ip->i_update_core == 0)) {
1161                 /*
1162                  * Timestamps/size haven't changed since last inode
1163                  * flush or inode transaction commit.  That means
1164                  * either nothing got written or a transaction
1165                  * committed which caught the updates.  If the
1166                  * latter happened and the transaction hasn't
1167                  * hit the disk yet, the inode will be still
1168                  * be pinned.  If it is, force the log.
1169                  */
1170
1171                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1172
1173                 if (xfs_ipincount(ip)) {
1174                         xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1175                                       XFS_LOG_FORCE |
1176                                       ((flag & FSYNC_WAIT)
1177                                        ? XFS_LOG_SYNC : 0));
1178                 }
1179                 error = 0;
1180         } else  {
1181                 /*
1182                  * Kick off a transaction to log the inode
1183                  * core to get the updates.  Make it
1184                  * sync if FSYNC_WAIT is passed in (which
1185                  * is done by everybody but specfs).  The
1186                  * sync transaction will also force the log.
1187                  */
1188                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1189                 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1190                 if ((error = xfs_trans_reserve(tp, 0,
1191                                 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1192                                 0, 0, 0)))  {
1193                         xfs_trans_cancel(tp, 0);
1194                         return error;
1195                 }
1196                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1197
1198                 /*
1199                  * Note - it's possible that we might have pushed
1200                  * ourselves out of the way during trans_reserve
1201                  * which would flush the inode.  But there's no
1202                  * guarantee that the inode buffer has actually
1203                  * gone out yet (it's delwri).  Plus the buffer
1204                  * could be pinned anyway if it's part of an
1205                  * inode in another recent transaction.  So we
1206                  * play it safe and fire off the transaction anyway.
1207                  */
1208                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1209                 xfs_trans_ihold(tp, ip);
1210                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1211                 if (flag & FSYNC_WAIT)
1212                         xfs_trans_set_sync(tp);
1213                 error = xfs_trans_commit(tp, 0, NULL);
1214
1215                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1216         }
1217         return error;
1218 }
1219
1220 /*
1221  * This is called by xfs_inactive to free any blocks beyond eof,
1222  * when the link count isn't zero.
1223  */
1224 STATIC int
1225 xfs_inactive_free_eofblocks(
1226         xfs_mount_t     *mp,
1227         xfs_inode_t     *ip)
1228 {
1229         xfs_trans_t     *tp;
1230         int             error;
1231         xfs_fileoff_t   end_fsb;
1232         xfs_fileoff_t   last_fsb;
1233         xfs_filblks_t   map_len;
1234         int             nimaps;
1235         xfs_bmbt_irec_t imap;
1236
1237         /*
1238          * Figure out if there are any blocks beyond the end
1239          * of the file.  If not, then there is nothing to do.
1240          */
1241         end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1242         last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1243         map_len = last_fsb - end_fsb;
1244         if (map_len <= 0)
1245                 return (0);
1246
1247         nimaps = 1;
1248         xfs_ilock(ip, XFS_ILOCK_SHARED);
1249         error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1250                           NULL, 0, &imap, &nimaps, NULL);
1251         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1252
1253         if (!error && (nimaps != 0) &&
1254             (imap.br_startblock != HOLESTARTBLOCK)) {
1255                 /*
1256                  * Attach the dquots to the inode up front.
1257                  */
1258                 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1259                         return (error);
1260
1261                 /*
1262                  * There are blocks after the end of file.
1263                  * Free them up now by truncating the file to
1264                  * its current size.
1265                  */
1266                 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1267
1268                 /*
1269                  * Do the xfs_itruncate_start() call before
1270                  * reserving any log space because
1271                  * itruncate_start will call into the buffer
1272                  * cache and we can't
1273                  * do that within a transaction.
1274                  */
1275                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1276                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1277                                     ip->i_d.di_size);
1278
1279                 error = xfs_trans_reserve(tp, 0,
1280                                           XFS_ITRUNCATE_LOG_RES(mp),
1281                                           0, XFS_TRANS_PERM_LOG_RES,
1282                                           XFS_ITRUNCATE_LOG_COUNT);
1283                 if (error) {
1284                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1285                         xfs_trans_cancel(tp, 0);
1286                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1287                         return (error);
1288                 }
1289
1290                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1291                 xfs_trans_ijoin(tp, ip,
1292                                 XFS_IOLOCK_EXCL |
1293                                 XFS_ILOCK_EXCL);
1294                 xfs_trans_ihold(tp, ip);
1295
1296                 error = xfs_itruncate_finish(&tp, ip,
1297                                              ip->i_d.di_size,
1298                                              XFS_DATA_FORK,
1299                                              0);
1300                 /*
1301                  * If we get an error at this point we
1302                  * simply don't bother truncating the file.
1303                  */
1304                 if (error) {
1305                         xfs_trans_cancel(tp,
1306                                          (XFS_TRANS_RELEASE_LOG_RES |
1307                                           XFS_TRANS_ABORT));
1308                 } else {
1309                         error = xfs_trans_commit(tp,
1310                                                 XFS_TRANS_RELEASE_LOG_RES,
1311                                                 NULL);
1312                 }
1313                 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1314         }
1315         return (error);
1316 }
1317
1318 /*
1319  * Free a symlink that has blocks associated with it.
1320  */
1321 STATIC int
1322 xfs_inactive_symlink_rmt(
1323         xfs_inode_t     *ip,
1324         xfs_trans_t     **tpp)
1325 {
1326         xfs_buf_t       *bp;
1327         int             committed;
1328         int             done;
1329         int             error;
1330         xfs_fsblock_t   first_block;
1331         xfs_bmap_free_t free_list;
1332         int             i;
1333         xfs_mount_t     *mp;
1334         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1335         int             nmaps;
1336         xfs_trans_t     *ntp;
1337         int             size;
1338         xfs_trans_t     *tp;
1339
1340         tp = *tpp;
1341         mp = ip->i_mount;
1342         ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1343         /*
1344          * We're freeing a symlink that has some
1345          * blocks allocated to it.  Free the
1346          * blocks here.  We know that we've got
1347          * either 1 or 2 extents and that we can
1348          * free them all in one bunmapi call.
1349          */
1350         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1351         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1352                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1353                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1354                 xfs_trans_cancel(tp, 0);
1355                 *tpp = NULL;
1356                 return error;
1357         }
1358         /*
1359          * Lock the inode, fix the size, and join it to the transaction.
1360          * Hold it so in the normal path, we still have it locked for
1361          * the second transaction.  In the error paths we need it
1362          * held so the cancel won't rele it, see below.
1363          */
1364         xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1365         size = (int)ip->i_d.di_size;
1366         ip->i_d.di_size = 0;
1367         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1368         xfs_trans_ihold(tp, ip);
1369         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1370         /*
1371          * Find the block(s) so we can inval and unmap them.
1372          */
1373         done = 0;
1374         XFS_BMAP_INIT(&free_list, &first_block);
1375         nmaps = sizeof(mval) / sizeof(mval[0]);
1376         if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1377                         XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1378                         &free_list)))
1379                 goto error0;
1380         /*
1381          * Invalidate the block(s).
1382          */
1383         for (i = 0; i < nmaps; i++) {
1384                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1385                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1386                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1387                 xfs_trans_binval(tp, bp);
1388         }
1389         /*
1390          * Unmap the dead block(s) to the free_list.
1391          */
1392         if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1393                         &first_block, &free_list, &done)))
1394                 goto error1;
1395         ASSERT(done);
1396         /*
1397          * Commit the first transaction.  This logs the EFI and the inode.
1398          */
1399         if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1400                 goto error1;
1401         /*
1402          * The transaction must have been committed, since there were
1403          * actually extents freed by xfs_bunmapi.  See xfs_bmap_finish.
1404          * The new tp has the extent freeing and EFDs.
1405          */
1406         ASSERT(committed);
1407         /*
1408          * The first xact was committed, so add the inode to the new one.
1409          * Mark it dirty so it will be logged and moved forward in the log as
1410          * part of every commit.
1411          */
1412         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1413         xfs_trans_ihold(tp, ip);
1414         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1415         /*
1416          * Get a new, empty transaction to return to our caller.
1417          */
1418         ntp = xfs_trans_dup(tp);
1419         /*
1420          * Commit the transaction containing extent freeing and EFD's.
1421          * If we get an error on the commit here or on the reserve below,
1422          * we need to unlock the inode since the new transaction doesn't
1423          * have the inode attached.
1424          */
1425         error = xfs_trans_commit(tp, 0, NULL);
1426         tp = ntp;
1427         if (error) {
1428                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1429                 goto error0;
1430         }
1431         /*
1432          * Remove the memory for extent descriptions (just bookkeeping).
1433          */
1434         if (ip->i_df.if_bytes)
1435                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1436         ASSERT(ip->i_df.if_bytes == 0);
1437         /*
1438          * Put an itruncate log reservation in the new transaction
1439          * for our caller.
1440          */
1441         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1442                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1443                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1444                 goto error0;
1445         }
1446         /*
1447          * Return with the inode locked but not joined to the transaction.
1448          */
1449         *tpp = tp;
1450         return 0;
1451
1452  error1:
1453         xfs_bmap_cancel(&free_list);
1454  error0:
1455         /*
1456          * Have to come here with the inode locked and either
1457          * (held and in the transaction) or (not in the transaction).
1458          * If the inode isn't held then cancel would iput it, but
1459          * that's wrong since this is inactive and the vnode ref
1460          * count is 0 already.
1461          * Cancel won't do anything to the inode if held, but it still
1462          * needs to be locked until the cancel is done, if it was
1463          * joined to the transaction.
1464          */
1465         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1466         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1467         *tpp = NULL;
1468         return error;
1469
1470 }
1471
1472 STATIC int
1473 xfs_inactive_symlink_local(
1474         xfs_inode_t     *ip,
1475         xfs_trans_t     **tpp)
1476 {
1477         int             error;
1478
1479         ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1480         /*
1481          * We're freeing a symlink which fit into
1482          * the inode.  Just free the memory used
1483          * to hold the old symlink.
1484          */
1485         error = xfs_trans_reserve(*tpp, 0,
1486                                   XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1487                                   0, XFS_TRANS_PERM_LOG_RES,
1488                                   XFS_ITRUNCATE_LOG_COUNT);
1489
1490         if (error) {
1491                 xfs_trans_cancel(*tpp, 0);
1492                 *tpp = NULL;
1493                 return (error);
1494         }
1495         xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1496
1497         /*
1498          * Zero length symlinks _can_ exist.
1499          */
1500         if (ip->i_df.if_bytes > 0) {
1501                 xfs_idata_realloc(ip,
1502                                   -(ip->i_df.if_bytes),
1503                                   XFS_DATA_FORK);
1504                 ASSERT(ip->i_df.if_bytes == 0);
1505         }
1506         return (0);
1507 }
1508
1509 /*
1510  *
1511  */
1512 STATIC int
1513 xfs_inactive_attrs(
1514         xfs_inode_t     *ip,
1515         xfs_trans_t     **tpp)
1516 {
1517         xfs_trans_t     *tp;
1518         int             error;
1519         xfs_mount_t     *mp;
1520
1521         ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1522         tp = *tpp;
1523         mp = ip->i_mount;
1524         ASSERT(ip->i_d.di_forkoff != 0);
1525         xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1526         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527
1528         error = xfs_attr_inactive(ip);
1529         if (error) {
1530                 *tpp = NULL;
1531                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1532                 return (error); /* goto out*/
1533         }
1534
1535         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1536         error = xfs_trans_reserve(tp, 0,
1537                                   XFS_IFREE_LOG_RES(mp),
1538                                   0, XFS_TRANS_PERM_LOG_RES,
1539                                   XFS_INACTIVE_LOG_COUNT);
1540         if (error) {
1541                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1542                 xfs_trans_cancel(tp, 0);
1543                 *tpp = NULL;
1544                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1545                 return (error);
1546         }
1547
1548         xfs_ilock(ip, XFS_ILOCK_EXCL);
1549         xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1550         xfs_trans_ihold(tp, ip);
1551         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1552
1553         ASSERT(ip->i_d.di_anextents == 0);
1554
1555         *tpp = tp;
1556         return (0);
1557 }
1558
1559 STATIC int
1560 xfs_release(
1561         bhv_desc_t      *bdp)
1562 {
1563         xfs_inode_t     *ip;
1564         vnode_t         *vp;
1565         xfs_mount_t     *mp;
1566         int             error;
1567
1568         vp = BHV_TO_VNODE(bdp);
1569         ip = XFS_BHVTOI(bdp);
1570
1571         if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1572                 return 0;
1573         }
1574
1575         /* If this is a read-only mount, don't do this (would generate I/O) */
1576         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1577                 return 0;
1578
1579 #ifdef HAVE_REFCACHE
1580         /* If we are in the NFS reference cache then don't do this now */
1581         if (ip->i_refcache)
1582                 return 0;
1583 #endif
1584
1585         mp = ip->i_mount;
1586
1587         if (ip->i_d.di_nlink != 0) {
1588                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1589                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1590                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1591                     (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1592                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1593                                 return (error);
1594                         /* Update linux inode block count after free above */
1595                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1596                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1597                 }
1598         }
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * xfs_inactive
1605  *
1606  * This is called when the vnode reference count for the vnode
1607  * goes to zero.  If the file has been unlinked, then it must
1608  * now be truncated.  Also, we clear all of the read-ahead state
1609  * kept for the inode here since the file is now closed.
1610  */
1611 STATIC int
1612 xfs_inactive(
1613         bhv_desc_t      *bdp,
1614         cred_t          *credp)
1615 {
1616         xfs_inode_t     *ip;
1617         vnode_t         *vp;
1618         xfs_bmap_free_t free_list; 
1619         xfs_fsblock_t   first_block;
1620         int             committed;
1621         xfs_trans_t     *tp;
1622         xfs_mount_t     *mp;
1623         int             error;
1624         int             truncate;
1625
1626         vp = BHV_TO_VNODE(bdp);
1627         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1628
1629         ip = XFS_BHVTOI(bdp);
1630
1631         /*
1632          * If the inode is already free, then there can be nothing
1633          * to clean up here.
1634          */
1635         if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1636                 ASSERT(ip->i_df.if_real_bytes == 0);
1637                 ASSERT(ip->i_df.if_broot_bytes == 0);
1638                 return VN_INACTIVE_CACHE;
1639         }
1640
1641         /*
1642          * Only do a truncate if it's a regular file with
1643          * some actual space in it.  It's OK to look at the
1644          * inode's fields without the lock because we're the
1645          * only one with a reference to the inode.
1646          */
1647         truncate = ((ip->i_d.di_nlink == 0) &&
1648             ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1649             ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1650
1651         mp = ip->i_mount;
1652
1653         if (ip->i_d.di_nlink == 0 &&
1654             DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1655                 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1656         }
1657
1658         error = 0;
1659
1660         /* If this is a read-only mount, don't do this (would generate I/O) */
1661         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1662                 goto out;
1663
1664         if (ip->i_d.di_nlink != 0) {
1665                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1666                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1667                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1668                     (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1669                      (ip->i_delayed_blks != 0))) {
1670                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1671                                 return (VN_INACTIVE_CACHE);
1672                         /* Update linux inode block count after free above */
1673                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1674                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1675                 }
1676                 goto out;
1677         }
1678
1679         ASSERT(ip->i_d.di_nlink == 0);
1680
1681         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1682                 return (VN_INACTIVE_CACHE);
1683
1684         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1685         if (truncate) {
1686                 /*
1687                  * Do the xfs_itruncate_start() call before
1688                  * reserving any log space because itruncate_start
1689                  * will call into the buffer cache and we can't
1690                  * do that within a transaction.
1691                  */
1692                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1693
1694                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1695
1696                 error = xfs_trans_reserve(tp, 0,
1697                                           XFS_ITRUNCATE_LOG_RES(mp),
1698                                           0, XFS_TRANS_PERM_LOG_RES,
1699                                           XFS_ITRUNCATE_LOG_COUNT);
1700                 if (error) {
1701                         /* Don't call itruncate_cleanup */
1702                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1703                         xfs_trans_cancel(tp, 0);
1704                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1705                         return (VN_INACTIVE_CACHE);
1706                 }
1707
1708                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1709                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1710                 xfs_trans_ihold(tp, ip);
1711
1712                 /*
1713                  * normally, we have to run xfs_itruncate_finish sync.
1714                  * But if filesystem is wsync and we're in the inactive
1715                  * path, then we know that nlink == 0, and that the
1716                  * xaction that made nlink == 0 is permanently committed
1717                  * since xfs_remove runs as a synchronous transaction.
1718                  */
1719                 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1720                                 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1721
1722                 if (error) {
1723                         xfs_trans_cancel(tp,
1724                                 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1725                         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1726                         return (VN_INACTIVE_CACHE);
1727                 }
1728         } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1729
1730                 /*
1731                  * If we get an error while cleaning up a
1732                  * symlink we bail out.
1733                  */
1734                 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1735                         xfs_inactive_symlink_rmt(ip, &tp) :
1736                         xfs_inactive_symlink_local(ip, &tp);
1737
1738                 if (error) {
1739                         ASSERT(tp == NULL);
1740                         return (VN_INACTIVE_CACHE);
1741                 }
1742
1743                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1744                 xfs_trans_ihold(tp, ip);
1745         } else {
1746                 error = xfs_trans_reserve(tp, 0,
1747                                           XFS_IFREE_LOG_RES(mp),
1748                                           0, XFS_TRANS_PERM_LOG_RES,
1749                                           XFS_INACTIVE_LOG_COUNT);
1750                 if (error) {
1751                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1752                         xfs_trans_cancel(tp, 0);
1753                         return (VN_INACTIVE_CACHE);
1754                 }
1755
1756                 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1757                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1758                 xfs_trans_ihold(tp, ip);
1759         }
1760
1761         /*
1762          * If there are attributes associated with the file
1763          * then blow them away now.  The code calls a routine
1764          * that recursively deconstructs the attribute fork.
1765          * We need to just commit the current transaction
1766          * because we can't use it for xfs_attr_inactive().
1767          */
1768         if (ip->i_d.di_anextents > 0) {
1769                 error = xfs_inactive_attrs(ip, &tp);
1770                 /*
1771                  * If we got an error, the transaction is already
1772                  * cancelled, and the inode is unlocked. Just get out.
1773                  */
1774                  if (error)
1775                          return (VN_INACTIVE_CACHE);
1776         } else if (ip->i_afp) {
1777                 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1778         }
1779
1780         /*
1781          * Free the inode.
1782          */
1783         XFS_BMAP_INIT(&free_list, &first_block);
1784         error = xfs_ifree(tp, ip, &free_list);
1785         if (error) {
1786                 /*
1787                  * If we fail to free the inode, shut down.  The cancel
1788                  * might do that, we need to make sure.  Otherwise the
1789                  * inode might be lost for a long time or forever.
1790                  */
1791                 if (!XFS_FORCED_SHUTDOWN(mp)) {
1792                         cmn_err(CE_NOTE,
1793                 "xfs_inactive:  xfs_ifree() returned an error = %d on %s",
1794                                 error, mp->m_fsname);
1795                         xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1796                 }
1797                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1798         } else {
1799                 /*
1800                  * Credit the quota account(s). The inode is gone.
1801                  */
1802                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1803
1804                 /*
1805                  * Just ignore errors at this point.  There is
1806                  * nothing we can do except to try to keep going.
1807                  */
1808                 (void) xfs_bmap_finish(&tp,  &free_list, first_block,
1809                                        &committed);
1810                 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1811         }
1812         /*
1813          * Release the dquots held by inode, if any.
1814          */
1815         XFS_QM_DQDETACH(mp, ip);
1816
1817         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1818
1819  out:
1820         return VN_INACTIVE_CACHE;
1821 }
1822
1823
1824 /*
1825  * xfs_lookup
1826  */
1827 STATIC int
1828 xfs_lookup(
1829         bhv_desc_t              *dir_bdp,
1830         vname_t                 *dentry,
1831         vnode_t                 **vpp,
1832         int                     flags,
1833         vnode_t                 *rdir,
1834         cred_t                  *credp)
1835 {
1836         xfs_inode_t             *dp, *ip;
1837         xfs_ino_t               e_inum;
1838         int                     error;
1839         uint                    lock_mode;
1840         vnode_t                 *dir_vp;
1841
1842         dir_vp = BHV_TO_VNODE(dir_bdp);
1843         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1844
1845         dp = XFS_BHVTOI(dir_bdp);
1846
1847         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1848                 return XFS_ERROR(EIO);
1849
1850         lock_mode = xfs_ilock_map_shared(dp);
1851         error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1852         if (!error) {
1853                 *vpp = XFS_ITOV(ip);
1854                 ITRACE(ip);
1855         }
1856         xfs_iunlock_map_shared(dp, lock_mode);
1857         return error;
1858 }
1859
1860
1861 /*
1862  * xfs_create (create a new file).
1863  */
1864 STATIC int
1865 xfs_create(
1866         bhv_desc_t              *dir_bdp,
1867         vname_t                 *dentry,
1868         vattr_t                 *vap,
1869         vnode_t                 **vpp,
1870         cred_t                  *credp)
1871 {
1872         char                    *name = VNAME(dentry);
1873         vnode_t                 *dir_vp;
1874         xfs_inode_t             *dp, *ip;
1875         vnode_t                 *vp=NULL;
1876         xfs_trans_t             *tp;
1877         xfs_mount_t             *mp;
1878         xfs_dev_t               rdev;
1879         int                     error;
1880         xfs_bmap_free_t         free_list;
1881         xfs_fsblock_t           first_block;
1882         boolean_t               dp_joined_to_trans;
1883         int                     dm_event_sent = 0;
1884         uint                    cancel_flags;
1885         int                     committed;
1886         xfs_prid_t              prid;
1887         struct xfs_dquot        *udqp, *gdqp;
1888         uint                    resblks;
1889         int                     dm_di_mode;
1890         int                     namelen;
1891
1892         ASSERT(!*vpp);
1893         dir_vp = BHV_TO_VNODE(dir_bdp);
1894         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1895
1896         dp = XFS_BHVTOI(dir_bdp);
1897         mp = dp->i_mount;
1898
1899         dm_di_mode = vap->va_mode;
1900         namelen = VNAMELEN(dentry);
1901
1902         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1903                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1904                                 dir_vp, DM_RIGHT_NULL, NULL,
1905                                 DM_RIGHT_NULL, name, NULL,
1906                                 dm_di_mode, 0, 0);
1907
1908                 if (error)
1909                         return error;
1910                 dm_event_sent = 1;
1911         }
1912
1913         if (XFS_FORCED_SHUTDOWN(mp))
1914                 return XFS_ERROR(EIO);
1915
1916         /* Return through std_return after this point. */
1917
1918         udqp = gdqp = NULL;
1919         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1920                 prid = dp->i_d.di_projid;
1921         else if (vap->va_mask & XFS_AT_PROJID)
1922                 prid = (xfs_prid_t)vap->va_projid;
1923         else
1924                 prid = (xfs_prid_t)dfltprid;
1925
1926         /*
1927          * Make sure that we have allocated dquot(s) on disk.
1928          */
1929         error = XFS_QM_DQVOPALLOC(mp, dp,
1930                         current_fsuid(credp), current_fsgid(credp), prid,
1931                         XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1932         if (error)
1933                 goto std_return;
1934
1935         ip = NULL;
1936         dp_joined_to_trans = B_FALSE;
1937
1938         tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1939         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1940         resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1941         /*
1942          * Initially assume that the file does not exist and
1943          * reserve the resources for that case.  If that is not
1944          * the case we'll drop the one we have and get a more
1945          * appropriate transaction later.
1946          */
1947         error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1948                         XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1949         if (error == ENOSPC) {
1950                 resblks = 0;
1951                 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1952                                 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1953         }
1954         if (error) {
1955                 cancel_flags = 0;
1956                 dp = NULL;
1957                 goto error_return;
1958         }
1959
1960         xfs_ilock(dp, XFS_ILOCK_EXCL);
1961
1962         XFS_BMAP_INIT(&free_list, &first_block);
1963
1964         ASSERT(ip == NULL);
1965
1966         /*
1967          * Reserve disk quota and the inode.
1968          */
1969         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1970         if (error)
1971                 goto error_return;
1972
1973         if (resblks == 0 &&
1974             (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1975                 goto error_return;
1976         rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1977         error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1978                         rdev, credp, prid, resblks > 0,
1979                         &ip, &committed);
1980         if (error) {
1981                 if (error == ENOSPC)
1982                         goto error_return;
1983                 goto abort_return;
1984         }
1985         ITRACE(ip);
1986
1987         /*
1988          * At this point, we've gotten a newly allocated inode.
1989          * It is locked (and joined to the transaction).
1990          */
1991
1992         ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1993
1994         /*
1995          * Now we join the directory inode to the transaction.
1996          * We do not do it earlier because xfs_dir_ialloc
1997          * might commit the previous transaction (and release
1998          * all the locks).
1999          */
2000
2001         VN_HOLD(dir_vp);
2002         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2003         dp_joined_to_trans = B_TRUE;
2004
2005         error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
2006                 &first_block, &free_list,
2007                 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2008         if (error) {
2009                 ASSERT(error != ENOSPC);
2010                 goto abort_return;
2011         }
2012         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2013         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2014
2015         /*
2016          * If this is a synchronous mount, make sure that the
2017          * create transaction goes to disk before returning to
2018          * the user.
2019          */
2020         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2021                 xfs_trans_set_sync(tp);
2022         }
2023
2024         dp->i_gen++;
2025
2026         /*
2027          * Attach the dquot(s) to the inodes and modify them incore.
2028          * These ids of the inode couldn't have changed since the new
2029          * inode has been locked ever since it was created.
2030          */
2031         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2032
2033         /*
2034          * xfs_trans_commit normally decrements the vnode ref count
2035          * when it unlocks the inode. Since we want to return the
2036          * vnode to the caller, we bump the vnode ref count now.
2037          */
2038         IHOLD(ip);
2039         vp = XFS_ITOV(ip);
2040
2041         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2042         if (error) {
2043                 xfs_bmap_cancel(&free_list);
2044                 goto abort_rele;
2045         }
2046
2047         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2048         if (error) {
2049                 IRELE(ip);
2050                 tp = NULL;
2051                 goto error_return;
2052         }
2053
2054         XFS_QM_DQRELE(mp, udqp);
2055         XFS_QM_DQRELE(mp, gdqp);
2056
2057         /*
2058          * Propogate the fact that the vnode changed after the
2059          * xfs_inode locks have been released.
2060          */
2061         VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2062
2063         *vpp = vp;
2064
2065         /* Fallthrough to std_return with error = 0  */
2066
2067 std_return:
2068         if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2069                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2070                                                         DM_EVENT_POSTCREATE)) {
2071                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2072                         dir_vp, DM_RIGHT_NULL,
2073                         *vpp ? vp:NULL,
2074                         DM_RIGHT_NULL, name, NULL,
2075                         dm_di_mode, error, 0);
2076         }
2077         return error;
2078
2079  abort_return:
2080         cancel_flags |= XFS_TRANS_ABORT;
2081         /* FALLTHROUGH */
2082  error_return:
2083
2084         if (tp != NULL)
2085                 xfs_trans_cancel(tp, cancel_flags);
2086
2087         if (!dp_joined_to_trans && (dp != NULL))
2088                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2089         XFS_QM_DQRELE(mp, udqp);
2090         XFS_QM_DQRELE(mp, gdqp);
2091
2092         goto std_return;
2093
2094  abort_rele:
2095         /*
2096          * Wait until after the current transaction is aborted to
2097          * release the inode.  This prevents recursive transactions
2098          * and deadlocks from xfs_inactive.
2099          */
2100         cancel_flags |= XFS_TRANS_ABORT;
2101         xfs_trans_cancel(tp, cancel_flags);
2102         IRELE(ip);
2103
2104         XFS_QM_DQRELE(mp, udqp);
2105         XFS_QM_DQRELE(mp, gdqp);
2106
2107         goto std_return;
2108 }
2109
2110 #ifdef DEBUG
2111 /*
2112  * Some counters to see if (and how often) we are hitting some deadlock
2113  * prevention code paths.
2114  */
2115
2116 int xfs_rm_locks;
2117 int xfs_rm_lock_delays;
2118 int xfs_rm_attempts;
2119 #endif
2120
2121 /*
2122  * The following routine will lock the inodes associated with the
2123  * directory and the named entry in the directory. The locks are
2124  * acquired in increasing inode number.
2125  *
2126  * If the entry is "..", then only the directory is locked. The
2127  * vnode ref count will still include that from the .. entry in
2128  * this case.
2129  *
2130  * There is a deadlock we need to worry about. If the locked directory is
2131  * in the AIL, it might be blocking up the log. The next inode we lock
2132  * could be already locked by another thread waiting for log space (e.g
2133  * a permanent log reservation with a long running transaction (see
2134  * xfs_itruncate_finish)). To solve this, we must check if the directory
2135  * is in the ail and use lock_nowait. If we can't lock, we need to
2136  * drop the inode lock on the directory and try again. xfs_iunlock will
2137  * potentially push the tail if we were holding up the log.
2138  */
2139 STATIC int
2140 xfs_lock_dir_and_entry(
2141         xfs_inode_t     *dp,
2142         vname_t         *dentry,
2143         xfs_inode_t     *ip)    /* inode of entry 'name' */
2144 {
2145         int             attempts;
2146         xfs_ino_t       e_inum;
2147         xfs_inode_t     *ips[2];
2148         xfs_log_item_t  *lp;
2149
2150 #ifdef DEBUG
2151         xfs_rm_locks++;
2152 #endif
2153         attempts = 0;
2154
2155 again:
2156         xfs_ilock(dp, XFS_ILOCK_EXCL);
2157
2158         e_inum = ip->i_ino;
2159
2160         ITRACE(ip);
2161
2162         /*
2163          * We want to lock in increasing inum. Since we've already
2164          * acquired the lock on the directory, we may need to release
2165          * if if the inum of the entry turns out to be less.
2166          */
2167         if (e_inum > dp->i_ino) {
2168                 /*
2169                  * We are already in the right order, so just
2170                  * lock on the inode of the entry.
2171                  * We need to use nowait if dp is in the AIL.
2172                  */
2173
2174                 lp = (xfs_log_item_t *)dp->i_itemp;
2175                 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2176                         if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2177                                 attempts++;
2178 #ifdef DEBUG
2179                                 xfs_rm_attempts++;
2180 #endif
2181
2182                                 /*
2183                                  * Unlock dp and try again.
2184                                  * xfs_iunlock will try to push the tail
2185                                  * if the inode is in the AIL.
2186                                  */
2187
2188                                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2189
2190                                 if ((attempts % 5) == 0) {
2191                                         delay(1); /* Don't just spin the CPU */
2192 #ifdef DEBUG
2193                                         xfs_rm_lock_delays++;
2194 #endif
2195                                 }
2196                                 goto again;
2197                         }
2198                 } else {
2199                         xfs_ilock(ip, XFS_ILOCK_EXCL);
2200                 }
2201         } else if (e_inum < dp->i_ino) {
2202                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2203
2204                 ips[0] = ip;
2205                 ips[1] = dp;
2206                 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2207         }
2208         /* else  e_inum == dp->i_ino */
2209         /*     This can happen if we're asked to lock /x/..
2210          *     the entry is "..", which is also the parent directory.
2211          */
2212
2213         return 0;
2214 }
2215
2216 #ifdef DEBUG
2217 int xfs_locked_n;
2218 int xfs_small_retries;
2219 int xfs_middle_retries;
2220 int xfs_lots_retries;
2221 int xfs_lock_delays;
2222 #endif
2223
2224 /*
2225  * The following routine will lock n inodes in exclusive mode.
2226  * We assume the caller calls us with the inodes in i_ino order.
2227  *
2228  * We need to detect deadlock where an inode that we lock
2229  * is in the AIL and we start waiting for another inode that is locked
2230  * by a thread in a long running transaction (such as truncate). This can
2231  * result in deadlock since the long running trans might need to wait
2232  * for the inode we just locked in order to push the tail and free space
2233  * in the log.
2234  */
2235 void
2236 xfs_lock_inodes(
2237         xfs_inode_t     **ips,
2238         int             inodes,
2239         int             first_locked,
2240         uint            lock_mode)
2241 {
2242         int             attempts = 0, i, j, try_lock;
2243         xfs_log_item_t  *lp;
2244
2245         ASSERT(ips && (inodes >= 2)); /* we need at least two */
2246
2247         if (first_locked) {
2248                 try_lock = 1;
2249                 i = 1;
2250         } else {
2251                 try_lock = 0;
2252                 i = 0;
2253         }
2254
2255 again:
2256         for (; i < inodes; i++) {
2257                 ASSERT(ips[i]);
2258
2259                 if (i && (ips[i] == ips[i-1]))  /* Already locked */
2260                         continue;
2261
2262                 /*
2263                  * If try_lock is not set yet, make sure all locked inodes
2264                  * are not in the AIL.
2265                  * If any are, set try_lock to be used later.
2266                  */
2267
2268                 if (!try_lock) {
2269                         for (j = (i - 1); j >= 0 && !try_lock; j--) {
2270                                 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2271                                 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2272                                         try_lock++;
2273                                 }
2274                         }
2275                 }
2276
2277                 /*
2278                  * If any of the previous locks we have locked is in the AIL,
2279                  * we must TRY to get the second and subsequent locks. If
2280                  * we can't get any, we must release all we have
2281                  * and try again.
2282                  */
2283
2284                 if (try_lock) {
2285                         /* try_lock must be 0 if i is 0. */
2286                         /*
2287                          * try_lock means we have an inode locked
2288                          * that is in the AIL.
2289                          */
2290                         ASSERT(i != 0);
2291                         if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2292                                 attempts++;
2293
2294                                 /*
2295                                  * Unlock all previous guys and try again.
2296                                  * xfs_iunlock will try to push the tail
2297                                  * if the inode is in the AIL.
2298                                  */
2299
2300                                 for(j = i - 1; j >= 0; j--) {
2301
2302                                         /*
2303                                          * Check to see if we've already
2304                                          * unlocked this one.
2305                                          * Not the first one going back,
2306                                          * and the inode ptr is the same.
2307                                          */
2308                                         if ((j != (i - 1)) && ips[j] ==
2309                                                                 ips[j+1])
2310                                                 continue;
2311
2312                                         xfs_iunlock(ips[j], lock_mode);
2313                                 }
2314
2315                                 if ((attempts % 5) == 0) {
2316                                         delay(1); /* Don't just spin the CPU */
2317 #ifdef DEBUG
2318                                         xfs_lock_delays++;
2319 #endif
2320                                 }
2321                                 i = 0;
2322                                 try_lock = 0;
2323                                 goto again;
2324                         }
2325                 } else {
2326                         xfs_ilock(ips[i], lock_mode);
2327                 }
2328         }
2329
2330 #ifdef DEBUG
2331         if (attempts) {
2332                 if (attempts < 5) xfs_small_retries++;
2333                 else if (attempts < 100) xfs_middle_retries++;
2334                 else xfs_lots_retries++;
2335         } else {
2336                 xfs_locked_n++;
2337         }
2338 #endif
2339 }
2340
2341 #ifdef  DEBUG
2342 #define REMOVE_DEBUG_TRACE(x)   {remove_which_error_return = (x);}
2343 int remove_which_error_return = 0;
2344 #else /* ! DEBUG */
2345 #define REMOVE_DEBUG_TRACE(x)
2346 #endif  /* ! DEBUG */
2347
2348
2349 /*
2350  * xfs_remove
2351  *
2352  */
2353 STATIC int
2354 xfs_remove(
2355         bhv_desc_t              *dir_bdp,
2356         vname_t                 *dentry,
2357         cred_t                  *credp)
2358 {
2359         vnode_t                 *dir_vp;
2360         char                    *name = VNAME(dentry);
2361         xfs_inode_t             *dp, *ip;
2362         xfs_trans_t             *tp = NULL;
2363         xfs_mount_t             *mp;
2364         int                     error = 0;
2365         xfs_bmap_free_t         free_list;
2366         xfs_fsblock_t           first_block;
2367         int                     cancel_flags;
2368         int                     committed;
2369         int                     dm_di_mode = 0;
2370         int                     link_zero;
2371         uint                    resblks;
2372         int                     namelen;
2373
2374         dir_vp = BHV_TO_VNODE(dir_bdp);
2375         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2376
2377         dp = XFS_BHVTOI(dir_bdp);
2378         mp = dp->i_mount;
2379
2380         if (XFS_FORCED_SHUTDOWN(mp))
2381                 return XFS_ERROR(EIO);
2382
2383         namelen = VNAMELEN(dentry);
2384
2385         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2386                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2387                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2388                                         name, NULL, 0, 0, 0);
2389                 if (error)
2390                         return error;
2391         }
2392
2393         /* From this point on, return through std_return */
2394         ip = NULL;
2395
2396         /*
2397          * We need to get a reference to ip before we get our log
2398          * reservation. The reason for this is that we cannot call
2399          * xfs_iget for an inode for which we do not have a reference
2400          * once we've acquired a log reservation. This is because the
2401          * inode we are trying to get might be in xfs_inactive going
2402          * for a log reservation. Since we'll have to wait for the
2403          * inactive code to complete before returning from xfs_iget,
2404          * we need to make sure that we don't have log space reserved
2405          * when we call xfs_iget.  Instead we get an unlocked referece
2406          * to the inode before getting our log reservation.
2407          */
2408         error = xfs_get_dir_entry(dentry, &ip);
2409         if (error) {
2410                 REMOVE_DEBUG_TRACE(__LINE__);
2411                 goto std_return;
2412         }
2413
2414         dm_di_mode = ip->i_d.di_mode;
2415
2416         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2417
2418         ITRACE(ip);
2419
2420         error = XFS_QM_DQATTACH(mp, dp, 0);
2421         if (!error && dp != ip)
2422                 error = XFS_QM_DQATTACH(mp, ip, 0);
2423         if (error) {
2424                 REMOVE_DEBUG_TRACE(__LINE__);
2425                 IRELE(ip);
2426                 goto std_return;
2427         }
2428
2429         tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2430         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2431         /*
2432          * We try to get the real space reservation first,
2433          * allowing for directory btree deletion(s) implying
2434          * possible bmap insert(s).  If we can't get the space
2435          * reservation then we use 0 instead, and avoid the bmap
2436          * btree insert(s) in the directory code by, if the bmap
2437          * insert tries to happen, instead trimming the LAST
2438          * block from the directory.
2439          */
2440         resblks = XFS_REMOVE_SPACE_RES(mp);
2441         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2442                         XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2443         if (error == ENOSPC) {
2444                 resblks = 0;
2445                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2446                                 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2447         }
2448         if (error) {
2449                 ASSERT(error != ENOSPC);
2450                 REMOVE_DEBUG_TRACE(__LINE__);
2451                 xfs_trans_cancel(tp, 0);
2452                 IRELE(ip);
2453                 return error;
2454         }
2455
2456         error = xfs_lock_dir_and_entry(dp, dentry, ip);
2457         if (error) {
2458                 REMOVE_DEBUG_TRACE(__LINE__);
2459                 xfs_trans_cancel(tp, cancel_flags);
2460                 IRELE(ip);
2461                 goto std_return;
2462         }
2463
2464         /*
2465          * At this point, we've gotten both the directory and the entry
2466          * inodes locked.
2467          */
2468         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2469         if (dp != ip) {
2470                 /*
2471                  * Increment vnode ref count only in this case since
2472                  * there's an extra vnode reference in the case where
2473                  * dp == ip.
2474                  */
2475                 IHOLD(dp);
2476                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2477         }
2478
2479         /*
2480          * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2481          */
2482         XFS_BMAP_INIT(&free_list, &first_block);
2483         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2484                 &first_block, &free_list, 0);
2485         if (error) {
2486                 ASSERT(error != ENOENT);
2487                 REMOVE_DEBUG_TRACE(__LINE__);
2488                 goto error1;
2489         }
2490         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2491
2492         dp->i_gen++;
2493         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2494
2495         error = xfs_droplink(tp, ip);
2496         if (error) {
2497                 REMOVE_DEBUG_TRACE(__LINE__);
2498                 goto error1;
2499         }
2500
2501         /* Determine if this is the last link while
2502          * we are in the transaction.
2503          */
2504         link_zero = (ip)->i_d.di_nlink==0;
2505
2506         /*
2507          * Take an extra ref on the inode so that it doesn't
2508          * go to xfs_inactive() from within the commit.
2509          */
2510         IHOLD(ip);
2511
2512         /*
2513          * If this is a synchronous mount, make sure that the
2514          * remove transaction goes to disk before returning to
2515          * the user.
2516          */
2517         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2518                 xfs_trans_set_sync(tp);
2519         }
2520
2521         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2522         if (error) {
2523                 REMOVE_DEBUG_TRACE(__LINE__);
2524                 goto error_rele;
2525         }
2526
2527         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2528         if (error) {
2529                 IRELE(ip);
2530                 goto std_return;
2531         }
2532
2533         /*
2534          * Before we drop our extra reference to the inode, purge it
2535          * from the refcache if it is there.  By waiting until afterwards
2536          * to do the IRELE, we ensure that we won't go inactive in the
2537          * xfs_refcache_purge_ip routine (although that would be OK).
2538          */
2539         xfs_refcache_purge_ip(ip);
2540
2541         vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2542
2543         /*
2544          * Let interposed file systems know about removed links.
2545          */
2546         VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2547
2548         IRELE(ip);
2549
2550 /*      Fall through to std_return with error = 0 */
2551  std_return:
2552         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2553                                                 DM_EVENT_POSTREMOVE)) {
2554                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2555                                 dir_vp, DM_RIGHT_NULL,
2556                                 NULL, DM_RIGHT_NULL,
2557                                 name, NULL, dm_di_mode, error, 0);
2558         }
2559         return error;
2560
2561  error1:
2562         xfs_bmap_cancel(&free_list);
2563         cancel_flags |= XFS_TRANS_ABORT;
2564         xfs_trans_cancel(tp, cancel_flags);
2565         goto std_return;
2566
2567  error_rele:
2568         /*
2569          * In this case make sure to not release the inode until after
2570          * the current transaction is aborted.  Releasing it beforehand
2571          * can cause us to go to xfs_inactive and start a recursive
2572          * transaction which can easily deadlock with the current one.
2573          */
2574         xfs_bmap_cancel(&free_list);
2575         cancel_flags |= XFS_TRANS_ABORT;
2576         xfs_trans_cancel(tp, cancel_flags);
2577
2578         /*
2579          * Before we drop our extra reference to the inode, purge it
2580          * from the refcache if it is there.  By waiting until afterwards
2581          * to do the IRELE, we ensure that we won't go inactive in the
2582          * xfs_refcache_purge_ip routine (although that would be OK).
2583          */
2584         xfs_refcache_purge_ip(ip);
2585
2586         IRELE(ip);
2587
2588         goto std_return;
2589 }
2590
2591
2592 /*
2593  * xfs_link
2594  *
2595  */
2596 STATIC int
2597 xfs_link(
2598         bhv_desc_t              *target_dir_bdp,
2599         vnode_t                 *src_vp,
2600         vname_t                 *dentry,
2601         cred_t                  *credp)
2602 {
2603         xfs_inode_t             *tdp, *sip;
2604         xfs_trans_t             *tp;
2605         xfs_mount_t             *mp;
2606         xfs_inode_t             *ips[2];
2607         int                     error;
2608         xfs_bmap_free_t         free_list;
2609         xfs_fsblock_t           first_block;
2610         int                     cancel_flags;
2611         int                     committed;
2612         vnode_t                 *target_dir_vp;
2613         bhv_desc_t              *src_bdp;
2614         int                     resblks;
2615         char                    *target_name = VNAME(dentry);
2616         int                     target_namelen;
2617
2618         target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2619         vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2620         vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2621
2622         target_namelen = VNAMELEN(dentry);
2623         if (VN_ISDIR(src_vp))
2624                 return XFS_ERROR(EPERM);
2625
2626         src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2627         sip = XFS_BHVTOI(src_bdp);
2628         tdp = XFS_BHVTOI(target_dir_bdp);
2629         mp = tdp->i_mount;
2630         if (XFS_FORCED_SHUTDOWN(mp))
2631                 return XFS_ERROR(EIO);
2632
2633         if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2634                 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2635                                         target_dir_vp, DM_RIGHT_NULL,
2636                                         src_vp, DM_RIGHT_NULL,
2637                                         target_name, NULL, 0, 0, 0);
2638                 if (error)
2639                         return error;
2640         }
2641
2642         /* Return through std_return after this point. */
2643
2644         error = XFS_QM_DQATTACH(mp, sip, 0);
2645         if (!error && sip != tdp)
2646                 error = XFS_QM_DQATTACH(mp, tdp, 0);
2647         if (error)
2648                 goto std_return;
2649
2650         tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2651         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2652         resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2653         error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2654                         XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2655         if (error == ENOSPC) {
2656                 resblks = 0;
2657                 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2658                                 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2659         }
2660         if (error) {
2661                 cancel_flags = 0;
2662                 goto error_return;
2663         }
2664
2665         if (sip->i_ino < tdp->i_ino) {
2666                 ips[0] = sip;
2667                 ips[1] = tdp;
2668         } else {
2669                 ips[0] = tdp;
2670                 ips[1] = sip;
2671         }
2672
2673         xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2674
2675         /*
2676          * Increment vnode ref counts since xfs_trans_commit &
2677          * xfs_trans_cancel will both unlock the inodes and
2678          * decrement the associated ref counts.
2679          */
2680         VN_HOLD(src_vp);
2681         VN_HOLD(target_dir_vp);
2682         xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2683         xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2684
2685         /*
2686          * If the source has too many links, we can't make any more to it.
2687          */
2688         if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2689                 error = XFS_ERROR(EMLINK);
2690                 goto error_return;
2691         }
2692
2693         /*
2694          * If we are using project inheritance, we only allow hard link
2695          * creation in our tree when the project IDs are the same; else
2696          * the tree quota mechanism could be circumvented.
2697          */
2698         if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2699                      (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2700                 error = XFS_ERROR(EPERM);
2701                 goto error_return;
2702         }
2703
2704         if (resblks == 0 &&
2705             (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2706                         target_namelen)))
2707                 goto error_return;
2708
2709         XFS_BMAP_INIT(&free_list, &first_block);
2710
2711         error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2712                                    sip->i_ino, &first_block, &free_list,
2713                                    resblks);
2714         if (error)
2715                 goto abort_return;
2716         xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2717         tdp->i_gen++;
2718         xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2719
2720         error = xfs_bumplink(tp, sip);
2721         if (error) {
2722                 goto abort_return;
2723         }
2724
2725         /*
2726          * If this is a synchronous mount, make sure that the
2727          * link transaction goes to disk before returning to
2728          * the user.
2729          */
2730         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2731                 xfs_trans_set_sync(tp);
2732         }
2733
2734         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2735         if (error) {
2736                 xfs_bmap_cancel(&free_list);
2737                 goto abort_return;
2738         }
2739
2740         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2741         if (error) {
2742                 goto std_return;
2743         }
2744
2745         /* Fall through to std_return with error = 0. */
2746 std_return:
2747         if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2748                                                 DM_EVENT_POSTLINK)) {
2749                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2750                                 target_dir_vp, DM_RIGHT_NULL,
2751                                 src_vp, DM_RIGHT_NULL,
2752                                 target_name, NULL, 0, error, 0);
2753         }
2754         return error;
2755
2756  abort_return:
2757         cancel_flags |= XFS_TRANS_ABORT;
2758         /* FALLTHROUGH */
2759  error_return:
2760         xfs_trans_cancel(tp, cancel_flags);
2761
2762         goto std_return;
2763 }
2764 /*
2765  * xfs_mkdir
2766  *
2767  */
2768 STATIC int
2769 xfs_mkdir(
2770         bhv_desc_t              *dir_bdp,
2771         vname_t                 *dentry,
2772         vattr_t                 *vap,
2773         vnode_t                 **vpp,
2774         cred_t                  *credp)
2775 {
2776         char                    *dir_name = VNAME(dentry);
2777         xfs_inode_t             *dp;
2778         xfs_inode_t             *cdp;   /* inode of created dir */
2779         vnode_t                 *cvp;   /* vnode of created dir */
2780         xfs_trans_t             *tp;
2781         xfs_mount_t             *mp;
2782         int                     cancel_flags;
2783         int                     error;
2784         int                     committed;
2785         xfs_bmap_free_t         free_list;
2786         xfs_fsblock_t           first_block;
2787         vnode_t                 *dir_vp;
2788         boolean_t               dp_joined_to_trans;
2789         boolean_t               created = B_FALSE;
2790         int                     dm_event_sent = 0;
2791         xfs_prid_t              prid;
2792         struct xfs_dquot        *udqp, *gdqp;
2793         uint                    resblks;
2794         int                     dm_di_mode;
2795         int                     dir_namelen;
2796
2797         dir_vp = BHV_TO_VNODE(dir_bdp);
2798         dp = XFS_BHVTOI(dir_bdp);
2799         mp = dp->i_mount;
2800
2801         if (XFS_FORCED_SHUTDOWN(mp))
2802                 return XFS_ERROR(EIO);
2803
2804         dir_namelen = VNAMELEN(dentry);
2805
2806         tp = NULL;
2807         dp_joined_to_trans = B_FALSE;
2808         dm_di_mode = vap->va_mode;
2809
2810         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2811                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2812                                         dir_vp, DM_RIGHT_NULL, NULL,
2813                                         DM_RIGHT_NULL, dir_name, NULL,
2814                                         dm_di_mode, 0, 0);
2815                 if (error)
2816                         return error;
2817                 dm_event_sent = 1;
2818         }
2819
2820         /* Return through std_return after this point. */
2821
2822         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2823
2824         mp = dp->i_mount;
2825         udqp = gdqp = NULL;
2826         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2827                 prid = dp->i_d.di_projid;
2828         else if (vap->va_mask & XFS_AT_PROJID)
2829                 prid = (xfs_prid_t)vap->va_projid;
2830         else
2831                 prid = (xfs_prid_t)dfltprid;
2832
2833         /*
2834          * Make sure that we have allocated dquot(s) on disk.
2835          */
2836         error = XFS_QM_DQVOPALLOC(mp, dp,
2837                         current_fsuid(credp), current_fsgid(credp), prid,
2838                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2839         if (error)
2840                 goto std_return;
2841
2842         tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2843         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2844         resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2845         error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2846                                   XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2847         if (error == ENOSPC) {
2848                 resblks = 0;
2849                 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2850                                           XFS_TRANS_PERM_LOG_RES,
2851                                           XFS_MKDIR_LOG_COUNT);
2852         }
2853         if (error) {
2854                 cancel_flags = 0;
2855                 dp = NULL;
2856                 goto error_return;
2857         }
2858
2859         xfs_ilock(dp, XFS_ILOCK_EXCL);
2860
2861         /*
2862          * Check for directory link count overflow.
2863          */
2864         if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2865                 error = XFS_ERROR(EMLINK);
2866                 goto error_return;
2867         }
2868
2869         /*
2870          * Reserve disk quota and the inode.
2871          */
2872         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2873         if (error)
2874                 goto error_return;
2875
2876         if (resblks == 0 &&
2877             (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2878                 goto error_return;
2879         /*
2880          * create the directory inode.
2881          */
2882         error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2883                         0, credp, prid, resblks > 0,
2884                 &cdp, NULL);
2885         if (error) {
2886                 if (error == ENOSPC)
2887                         goto error_return;
2888                 goto abort_return;
2889         }
2890         ITRACE(cdp);
2891
2892         /*
2893          * Now we add the directory inode to the transaction.
2894          * We waited until now since xfs_dir_ialloc might start
2895          * a new transaction.  Had we joined the transaction
2896          * earlier, the locks might have gotten released.
2897          */
2898         VN_HOLD(dir_vp);
2899         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2900         dp_joined_to_trans = B_TRUE;
2901
2902         XFS_BMAP_INIT(&free_list, &first_block);
2903
2904         error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2905                         cdp->i_ino, &first_block, &free_list,
2906                         resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2907         if (error) {
2908                 ASSERT(error != ENOSPC);
2909                 goto error1;
2910         }
2911         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2912
2913         /*
2914          * Bump the in memory version number of the parent directory
2915          * so that other processes accessing it will recognize that
2916          * the directory has changed.
2917          */
2918         dp->i_gen++;
2919
2920         error = XFS_DIR_INIT(mp, tp, cdp, dp);
2921         if (error) {
2922                 goto error2;
2923         }
2924
2925         cdp->i_gen = 1;
2926         error = xfs_bumplink(tp, dp);
2927         if (error) {
2928                 goto error2;
2929         }
2930
2931         cvp = XFS_ITOV(cdp);
2932
2933         created = B_TRUE;
2934
2935         *vpp = cvp;
2936         IHOLD(cdp);
2937
2938         /*
2939          * Attach the dquots to the new inode and modify the icount incore.
2940          */
2941         XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2942
2943         /*
2944          * If this is a synchronous mount, make sure that the
2945          * mkdir transaction goes to disk before returning to
2946          * the user.
2947          */
2948         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2949                 xfs_trans_set_sync(tp);
2950         }
2951
2952         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2953         if (error) {
2954                 IRELE(cdp);
2955                 goto error2;
2956         }
2957
2958         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2959         XFS_QM_DQRELE(mp, udqp);
2960         XFS_QM_DQRELE(mp, gdqp);
2961         if (error) {
2962                 IRELE(cdp);
2963         }
2964
2965         /* Fall through to std_return with error = 0 or errno from
2966          * xfs_trans_commit. */
2967
2968 std_return:
2969         if ( (created || (error != 0 && dm_event_sent != 0)) &&
2970                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2971                                                 DM_EVENT_POSTCREATE)) {
2972                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2973                                         dir_vp, DM_RIGHT_NULL,
2974                                         created ? XFS_ITOV(cdp):NULL,
2975                                         DM_RIGHT_NULL,
2976                                         dir_name, NULL,
2977                                         dm_di_mode, error, 0);
2978         }
2979         return error;
2980
2981  error2:
2982  error1:
2983         xfs_bmap_cancel(&free_list);
2984  abort_return:
2985         cancel_flags |= XFS_TRANS_ABORT;
2986  error_return:
2987         xfs_trans_cancel(tp, cancel_flags);
2988         XFS_QM_DQRELE(mp, udqp);
2989         XFS_QM_DQRELE(mp, gdqp);
2990
2991         if (!dp_joined_to_trans && (dp != NULL)) {
2992                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2993         }
2994
2995         goto std_return;
2996 }
2997
2998
2999 /*
3000  * xfs_rmdir
3001  *
3002  */
3003 STATIC int
3004 xfs_rmdir(
3005         bhv_desc_t              *dir_bdp,
3006         vname_t                 *dentry,
3007         cred_t                  *credp)
3008 {
3009         char                    *name = VNAME(dentry);
3010         xfs_inode_t             *dp;
3011         xfs_inode_t             *cdp;   /* child directory */
3012         xfs_trans_t             *tp;
3013         xfs_mount_t             *mp;
3014         int                     error;
3015         xfs_bmap_free_t         free_list;
3016         xfs_fsblock_t           first_block;
3017         int                     cancel_flags;
3018         int                     committed;
3019         vnode_t                 *dir_vp;
3020         int                     dm_di_mode = 0;
3021         int                     last_cdp_link;
3022         int                     namelen;
3023         uint                    resblks;
3024
3025         dir_vp = BHV_TO_VNODE(dir_bdp);
3026         dp = XFS_BHVTOI(dir_bdp);
3027         mp = dp->i_mount;
3028
3029         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3030
3031         if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3032                 return XFS_ERROR(EIO);
3033         namelen = VNAMELEN(dentry);
3034
3035         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3036                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3037                                         dir_vp, DM_RIGHT_NULL,
3038                                         NULL, DM_RIGHT_NULL,
3039                                         name, NULL, 0, 0, 0);
3040                 if (error)
3041                         return XFS_ERROR(error);
3042         }
3043
3044         /* Return through std_return after this point. */
3045
3046         cdp = NULL;
3047
3048         /*
3049          * We need to get a reference to cdp before we get our log
3050          * reservation.  The reason for this is that we cannot call
3051          * xfs_iget for an inode for which we do not have a reference
3052          * once we've acquired a log reservation.  This is because the
3053          * inode we are trying to get might be in xfs_inactive going
3054          * for a log reservation.  Since we'll have to wait for the
3055          * inactive code to complete before returning from xfs_iget,
3056          * we need to make sure that we don't have log space reserved
3057          * when we call xfs_iget.  Instead we get an unlocked referece
3058          * to the inode before getting our log reservation.
3059          */
3060         error = xfs_get_dir_entry(dentry, &cdp);
3061         if (error) {
3062                 REMOVE_DEBUG_TRACE(__LINE__);
3063                 goto std_return;
3064         }
3065         mp = dp->i_mount;
3066         dm_di_mode = cdp->i_d.di_mode;
3067
3068         /*
3069          * Get the dquots for the inodes.
3070          */
3071         error = XFS_QM_DQATTACH(mp, dp, 0);
3072         if (!error && dp != cdp)
3073                 error = XFS_QM_DQATTACH(mp, cdp, 0);
3074         if (error) {
3075                 IRELE(cdp);
3076                 REMOVE_DEBUG_TRACE(__LINE__);
3077                 goto std_return;
3078         }
3079
3080         tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3081         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3082         /*
3083          * We try to get the real space reservation first,
3084          * allowing for directory btree deletion(s) implying
3085          * possible bmap insert(s).  If we can't get the space
3086          * reservation then we use 0 instead, and avoid the bmap
3087          * btree insert(s) in the directory code by, if the bmap
3088          * insert tries to happen, instead trimming the LAST
3089          * block from the directory.
3090          */
3091         resblks = XFS_REMOVE_SPACE_RES(mp);
3092         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3093                         XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3094         if (error == ENOSPC) {
3095                 resblks = 0;
3096                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3097                                 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3098         }
3099         if (error) {
3100                 ASSERT(error != ENOSPC);
3101                 cancel_flags = 0;
3102                 IRELE(cdp);
3103                 goto error_return;
3104         }
3105         XFS_BMAP_INIT(&free_list, &first_block);
3106
3107         /*
3108          * Now lock the child directory inode and the parent directory
3109          * inode in the proper order.  This will take care of validating
3110          * that the directory entry for the child directory inode has
3111          * not changed while we were obtaining a log reservation.
3112          */
3113         error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3114         if (error) {
3115                 xfs_trans_cancel(tp, cancel_flags);
3116                 IRELE(cdp);
3117                 goto std_return;
3118         }
3119
3120         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3121         if (dp != cdp) {
3122                 /*
3123                  * Only increment the parent directory vnode count if
3124                  * we didn't bump it in looking up cdp.  The only time
3125                  * we don't bump it is when we're looking up ".".
3126                  */
3127                 VN_HOLD(dir_vp);
3128         }
3129
3130         ITRACE(cdp);
3131         xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3132
3133         ASSERT(cdp->i_d.di_nlink >= 2);
3134         if (cdp->i_d.di_nlink != 2) {
3135                 error = XFS_ERROR(ENOTEMPTY);
3136                 goto error_return;
3137         }
3138         if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3139                 error = XFS_ERROR(ENOTEMPTY);
3140                 goto error_return;
3141         }
3142
3143         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3144                 &first_block, &free_list, resblks);
3145         if (error) {
3146                 goto error1;
3147         }
3148
3149         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3150
3151         /*
3152          * Bump the in memory generation count on the parent
3153          * directory so that other can know that it has changed.
3154          */
3155         dp->i_gen++;
3156
3157         /*
3158          * Drop the link from cdp's "..".
3159          */
3160         error = xfs_droplink(tp, dp);
3161         if (error) {
3162                 goto error1;
3163         }
3164
3165         /*
3166          * Drop the link from dp to cdp.
3167          */
3168         error = xfs_droplink(tp, cdp);
3169         if (error) {
3170                 goto error1;
3171         }
3172
3173         /*
3174          * Drop the "." link from cdp to self.
3175          */
3176         error = xfs_droplink(tp, cdp);
3177         if (error) {
3178                 goto error1;
3179         }
3180
3181         /* Determine these before committing transaction */
3182         last_cdp_link = (cdp)->i_d.di_nlink==0;
3183
3184         /*
3185          * Take an extra ref on the child vnode so that it
3186          * does not go to xfs_inactive() from within the commit.
3187          */
3188         IHOLD(cdp);
3189
3190         /*
3191          * If this is a synchronous mount, make sure that the
3192          * rmdir transaction goes to disk before returning to
3193          * the user.
3194          */
3195         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3196                 xfs_trans_set_sync(tp);
3197         }
3198
3199         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3200         if (error) {
3201                 xfs_bmap_cancel(&free_list);
3202                 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3203                                  XFS_TRANS_ABORT));
3204                 IRELE(cdp);
3205                 goto std_return;
3206         }
3207
3208         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3209         if (error) {
3210                 IRELE(cdp);
3211                 goto std_return;
3212         }
3213
3214
3215         /*
3216          * Let interposed file systems know about removed links.
3217          */
3218         VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3219
3220         IRELE(cdp);
3221
3222         /* Fall through to std_return with error = 0 or the errno
3223          * from xfs_trans_commit. */
3224 std_return:
3225         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3226                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3227                                         dir_vp, DM_RIGHT_NULL,
3228                                         NULL, DM_RIGHT_NULL,
3229                                         name, NULL, dm_di_mode,
3230                                         error, 0);
3231         }
3232         return error;
3233
3234  error1:
3235         xfs_bmap_cancel(&free_list);
3236         cancel_flags |= XFS_TRANS_ABORT;
3237  error_return:
3238         xfs_trans_cancel(tp, cancel_flags);
3239         goto std_return;
3240 }
3241
3242
3243 /*
3244  * xfs_readdir
3245  *
3246  * Read dp's entries starting at uiop->uio_offset and translate them into
3247  * bufsize bytes worth of struct dirents starting at bufbase.
3248  */
3249 STATIC int
3250 xfs_readdir(
3251         bhv_desc_t      *dir_bdp,
3252         uio_t           *uiop,
3253         cred_t          *credp,
3254         int             *eofp)
3255 {
3256         xfs_inode_t     *dp;
3257         xfs_trans_t     *tp = NULL;
3258         int             error = 0;
3259         uint            lock_mode;
3260         xfs_off_t       start_offset;
3261
3262         vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3263                                                (inst_t *)__return_address);
3264         dp = XFS_BHVTOI(dir_bdp);
3265
3266         if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3267                 return XFS_ERROR(EIO);
3268         }
3269
3270         lock_mode = xfs_ilock_map_shared(dp);
3271         start_offset = uiop->uio_offset;
3272         error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3273         if (start_offset != uiop->uio_offset) {
3274                 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3275         }
3276         xfs_iunlock_map_shared(dp, lock_mode);
3277         return error;
3278 }
3279
3280
3281 /*
3282  * xfs_symlink
3283  *
3284  */
3285 STATIC int
3286 xfs_symlink(
3287         bhv_desc_t              *dir_bdp,
3288         vname_t                 *dentry,
3289         vattr_t                 *vap,
3290         char                    *target_path,
3291         vnode_t                 **vpp,
3292         cred_t                  *credp)
3293 {
3294         xfs_trans_t             *tp;
3295         xfs_mount_t             *mp;
3296         xfs_inode_t             *dp;
3297         xfs_inode_t             *ip;
3298         int                     error;
3299         int                     pathlen;
3300         xfs_bmap_free_t         free_list;
3301         xfs_fsblock_t           first_block;
3302         boolean_t               dp_joined_to_trans;
3303         vnode_t                 *dir_vp;
3304         uint                    cancel_flags;
3305         int                     committed;
3306         xfs_fileoff_t           first_fsb;
3307         xfs_filblks_t           fs_blocks;
3308         int                     nmaps;
3309         xfs_bmbt_irec_t         mval[SYMLINK_MAPS];
3310         xfs_daddr_t             d;
3311         char                    *cur_chunk;
3312         int                     byte_cnt;
3313         int                     n;
3314         xfs_buf_t               *bp;
3315         xfs_prid_t              prid;
3316         struct xfs_dquot        *udqp, *gdqp;
3317         uint                    resblks;
3318         char                    *link_name = VNAME(dentry);
3319         int                     link_namelen;
3320
3321         *vpp = NULL;
3322         dir_vp = BHV_TO_VNODE(dir_bdp);
3323         dp = XFS_BHVTOI(dir_bdp);
3324         dp_joined_to_trans = B_FALSE;
3325         error = 0;
3326         ip = NULL;
3327         tp = NULL;
3328
3329         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3330
3331         mp = dp->i_mount;
3332
3333         if (XFS_FORCED_SHUTDOWN(mp))
3334                 return XFS_ERROR(EIO);
3335
3336         link_namelen = VNAMELEN(dentry);
3337
3338         /*
3339          * Check component lengths of the target path name.
3340          */
3341         pathlen = strlen(target_path);
3342         if (pathlen >= MAXPATHLEN)      /* total string too long */
3343                 return XFS_ERROR(ENAMETOOLONG);
3344         if (pathlen >= MAXNAMELEN) {    /* is any component too long? */
3345                 int len, total;
3346                 char *path;
3347
3348                 for(total = 0, path = target_path; total < pathlen;) {
3349                         /*
3350                          * Skip any slashes.
3351                          */
3352                         while(*path == '/') {
3353                                 total++;
3354                                 path++;
3355                         }
3356
3357                         /*
3358                          * Count up to the next slash or end of path.
3359                          * Error out if the component is bigger than MAXNAMELEN.
3360                          */
3361                         for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3362                                 if (++len >= MAXNAMELEN) {
3363                                         error = ENAMETOOLONG;
3364                                         return error;
3365                                 }
3366                         }
3367                 }
3368         }
3369
3370         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3371                 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3372                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3373                                         link_name, target_path, 0, 0, 0);
3374                 if (error)
3375                         return error;
3376         }
3377
3378         /* Return through std_return after this point. */
3379
3380         udqp = gdqp = NULL;
3381         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3382                 prid = dp->i_d.di_projid;
3383         else if (vap->va_mask & XFS_AT_PROJID)
3384                 prid = (xfs_prid_t)vap->va_projid;
3385         else
3386                 prid = (xfs_prid_t)dfltprid;
3387
3388         /*
3389          * Make sure that we have allocated dquot(s) on disk.
3390          */
3391         error = XFS_QM_DQVOPALLOC(mp, dp,
3392                         current_fsuid(credp), current_fsgid(credp), prid,
3393                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3394         if (error)
3395                 goto std_return;
3396
3397         tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3398         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3399         /*
3400          * The symlink will fit into the inode data fork?
3401          * There can't be any attributes so we get the whole variable part.
3402          */
3403         if (pathlen <= XFS_LITINO(mp))
3404                 fs_blocks = 0;
3405         else
3406                 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3407         resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3408         error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3409                         XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3410         if (error == ENOSPC && fs_blocks == 0) {
3411                 resblks = 0;
3412                 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3413                                 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3414         }
3415         if (error) {
3416                 cancel_flags = 0;
3417                 dp = NULL;
3418                 goto error_return;
3419         }
3420
3421         xfs_ilock(dp, XFS_ILOCK_EXCL);
3422
3423         /*
3424          * Check whether the directory allows new symlinks or not.
3425          */
3426         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3427                 error = XFS_ERROR(EPERM);
3428                 goto error_return;
3429         }
3430
3431         /*
3432          * Reserve disk quota : blocks and inode.
3433          */
3434         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3435         if (error)
3436                 goto error_return;
3437
3438         /*
3439          * Check for ability to enter directory entry, if no space reserved.
3440          */
3441         if (resblks == 0 &&
3442             (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3443                 goto error_return;
3444         /*
3445          * Initialize the bmap freelist prior to calling either
3446          * bmapi or the directory create code.
3447          */
3448         XFS_BMAP_INIT(&free_list, &first_block);
3449
3450         /*
3451          * Allocate an inode for the symlink.
3452          */
3453         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3454                                1, 0, credp, prid, resblks > 0, &ip, NULL);
3455         if (error) {
3456                 if (error == ENOSPC)
3457                         goto error_return;
3458                 goto error1;
3459         }
3460         ITRACE(ip);
3461
3462         VN_HOLD(dir_vp);
3463         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3464         dp_joined_to_trans = B_TRUE;
3465
3466         /*
3467          * Also attach the dquot(s) to it, if applicable.
3468          */
3469         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3470
3471         if (resblks)
3472                 resblks -= XFS_IALLOC_SPACE_RES(mp);
3473         /*
3474          * If the symlink will fit into the inode, write it inline.
3475          */
3476         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3477                 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3478                 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3479                 ip->i_d.di_size = pathlen;
3480
3481                 /*
3482                  * The inode was initially created in extent format.
3483                  */
3484                 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3485                 ip->i_df.if_flags |= XFS_IFINLINE;
3486
3487                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3488                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3489
3490         } else {
3491                 first_fsb = 0;
3492                 nmaps = SYMLINK_MAPS;
3493
3494                 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3495                                   XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3496                                   &first_block, resblks, mval, &nmaps,
3497                                   &free_list);
3498                 if (error) {
3499                         goto error1;
3500                 }
3501
3502                 if (resblks)
3503                         resblks -= fs_blocks;
3504                 ip->i_d.di_size = pathlen;
3505                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3506
3507                 cur_chunk = target_path;
3508                 for (n = 0; n < nmaps; n++) {
3509                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3510                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3511                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3512                                                BTOBB(byte_cnt), 0);
3513                         ASSERT(bp && !XFS_BUF_GETERROR(bp));
3514                         if (pathlen < byte_cnt) {
3515                                 byte_cnt = pathlen;
3516                         }
3517                         pathlen -= byte_cnt;
3518
3519                         memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3520                         cur_chunk += byte_cnt;
3521
3522                         xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3523                 }
3524         }
3525
3526         /*
3527          * Create the directory entry for the symlink.
3528          */
3529         error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3530                         ip->i_ino, &first_block, &free_list, resblks);
3531         if (error) {
3532                 goto error1;
3533         }
3534         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3535         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3536
3537         /*
3538          * Bump the in memory version number of the parent directory
3539          * so that other processes accessing it will recognize that
3540          * the directory has changed.
3541          */
3542         dp->i_gen++;
3543
3544         /*
3545          * If this is a synchronous mount, make sure that the
3546          * symlink transaction goes to disk before returning to
3547          * the user.
3548          */
3549         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3550                 xfs_trans_set_sync(tp);
3551         }
3552
3553         /*
3554          * xfs_trans_commit normally decrements the vnode ref count
3555          * when it unlocks the inode. Since we want to return the
3556          * vnode to the caller, we bump the vnode ref count now.
3557          */
3558         IHOLD(ip);
3559
3560         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3561         if (error) {
3562                 goto error2;
3563         }
3564         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3565         XFS_QM_DQRELE(mp, udqp);
3566         XFS_QM_DQRELE(mp, gdqp);
3567
3568         /* Fall through to std_return with error = 0 or errno from
3569          * xfs_trans_commit     */
3570 std_return:
3571         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3572                              DM_EVENT_POSTSYMLINK)) {
3573                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3574                                         dir_vp, DM_RIGHT_NULL,
3575                                         error ? NULL : XFS_ITOV(ip),
3576                                         DM_RIGHT_NULL, link_name, target_path,
3577                                         0, error, 0);
3578         }
3579
3580         if (!error) {
3581                 vnode_t *vp;
3582
3583                 ASSERT(ip);
3584                 vp = XFS_ITOV(ip);
3585                 *vpp = vp;
3586         }
3587         return error;
3588
3589  error2:
3590         IRELE(ip);
3591  error1:
3592         xfs_bmap_cancel(&free_list);
3593         cancel_flags |= XFS_TRANS_ABORT;
3594  error_return:
3595         xfs_trans_cancel(tp, cancel_flags);
3596         XFS_QM_DQRELE(mp, udqp);
3597         XFS_QM_DQRELE(mp, gdqp);
3598
3599         if (!dp_joined_to_trans && (dp != NULL)) {
3600                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3601         }
3602
3603         goto std_return;
3604 }
3605
3606
3607 /*
3608  * xfs_fid2
3609  *
3610  * A fid routine that takes a pointer to a previously allocated
3611  * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3612  */
3613 STATIC int
3614 xfs_fid2(
3615         bhv_desc_t      *bdp,
3616         fid_t           *fidp)
3617 {
3618         xfs_inode_t     *ip;
3619         xfs_fid2_t      *xfid;
3620
3621         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3622                                        (inst_t *)__return_address);
3623         ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3624
3625         xfid = (xfs_fid2_t *)fidp;
3626         ip = XFS_BHVTOI(bdp);
3627         xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3628         xfid->fid_pad = 0;
3629         /*
3630          * use memcpy because the inode is a long long and there's no
3631          * assurance that xfid->fid_ino is properly aligned.
3632          */
3633         memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3634         xfid->fid_gen = ip->i_d.di_gen;
3635
3636         return 0;
3637 }
3638
3639
3640 /*
3641  * xfs_rwlock
3642  */
3643 int
3644 xfs_rwlock(
3645         bhv_desc_t      *bdp,
3646         vrwlock_t       locktype)
3647 {
3648         xfs_inode_t     *ip;
3649         vnode_t         *vp;
3650
3651         vp = BHV_TO_VNODE(bdp);
3652         if (VN_ISDIR(vp))
3653                 return 1;
3654         ip = XFS_BHVTOI(bdp);
3655         if (locktype == VRWLOCK_WRITE) {
3656                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3657         } else if (locktype == VRWLOCK_TRY_READ) {
3658                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3659         } else if (locktype == VRWLOCK_TRY_WRITE) {
3660                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3661         } else {
3662                 ASSERT((locktype == VRWLOCK_READ) ||
3663                        (locktype == VRWLOCK_WRITE_DIRECT));
3664                 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3665         }
3666
3667         return 1;
3668 }
3669
3670
3671 /*
3672  * xfs_rwunlock
3673  */
3674 void
3675 xfs_rwunlock(
3676         bhv_desc_t      *bdp,
3677         vrwlock_t       locktype)
3678 {
3679         xfs_inode_t     *ip;
3680         vnode_t         *vp;
3681
3682         vp = BHV_TO_VNODE(bdp);
3683         if (VN_ISDIR(vp))
3684                 return;
3685         ip = XFS_BHVTOI(bdp);
3686         if (locktype == VRWLOCK_WRITE) {
3687                 /*
3688                  * In the write case, we may have added a new entry to
3689                  * the reference cache.  This might store a pointer to
3690                  * an inode to be released in this inode.  If it is there,
3691                  * clear the pointer and release the inode after unlocking
3692                  * this one.
3693                  */
3694                 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3695         } else {
3696                 ASSERT((locktype == VRWLOCK_READ) ||
3697                        (locktype == VRWLOCK_WRITE_DIRECT));
3698                 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3699         }
3700         return;
3701 }
3702
3703 STATIC int
3704 xfs_inode_flush(
3705         bhv_desc_t      *bdp,
3706         int             flags)
3707 {
3708         xfs_inode_t     *ip;
3709         xfs_mount_t     *mp;
3710         xfs_inode_log_item_t *iip;
3711         int             error = 0;
3712
3713         ip = XFS_BHVTOI(bdp);
3714         mp = ip->i_mount;
3715         iip = ip->i_itemp;
3716
3717         if (XFS_FORCED_SHUTDOWN(mp))
3718                 return XFS_ERROR(EIO);
3719
3720         /*
3721          * Bypass inodes which have already been cleaned by
3722          * the inode flush clustering code inside xfs_iflush
3723          */
3724         if ((ip->i_update_core == 0) &&
3725             ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3726                 return 0;
3727
3728         if (flags & FLUSH_LOG) {
3729                 if (iip && iip->ili_last_lsn) {
3730                         xlog_t          *log = mp->m_log;
3731                         xfs_lsn_t       sync_lsn;
3732                         int             s, log_flags = XFS_LOG_FORCE;
3733
3734                         s = GRANT_LOCK(log);
3735                         sync_lsn = log->l_last_sync_lsn;
3736                         GRANT_UNLOCK(log, s);
3737
3738                         if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3739                                 return 0;
3740
3741                         if (flags & FLUSH_SYNC)
3742                                 log_flags |= XFS_LOG_SYNC;
3743                         return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3744                 }
3745         }
3746
3747         /*
3748          * We make this non-blocking if the inode is contended,
3749          * return EAGAIN to indicate to the caller that they
3750          * did not succeed. This prevents the flush path from
3751          * blocking on inodes inside another operation right
3752          * now, they get caught later by xfs_sync.
3753          */
3754         if (flags & FLUSH_INODE) {
3755                 int     flush_flags;
3756
3757                 if (xfs_ipincount(ip))
3758                         return EAGAIN;
3759
3760                 if (flags & FLUSH_SYNC) {
3761                         xfs_ilock(ip, XFS_ILOCK_SHARED);
3762                         xfs_iflock(ip);
3763                 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3764                         if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3765                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3766                                 return EAGAIN;
3767                         }
3768                 } else {
3769                         return EAGAIN;
3770                 }
3771
3772                 if (flags & FLUSH_SYNC)
3773                         flush_flags = XFS_IFLUSH_SYNC;
3774                 else
3775                         flush_flags = XFS_IFLUSH_ASYNC;
3776
3777                 error = xfs_iflush(ip, flush_flags);
3778                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3779         }
3780
3781         return error;
3782 }
3783
3784
3785 int
3786 xfs_set_dmattrs (
3787         bhv_desc_t      *bdp,
3788         u_int           evmask,
3789         u_int16_t       state,
3790         cred_t          *credp)
3791 {
3792         xfs_inode_t     *ip;
3793         xfs_trans_t     *tp;
3794         xfs_mount_t     *mp;
3795         int             error;
3796
3797         if (!capable(CAP_SYS_ADMIN))
3798                 return XFS_ERROR(EPERM);
3799
3800         ip = XFS_BHVTOI(bdp);
3801         mp = ip->i_mount;
3802
3803         if (XFS_FORCED_SHUTDOWN(mp))
3804                 return XFS_ERROR(EIO);
3805
3806         tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3807         error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3808         if (error) {
3809                 xfs_trans_cancel(tp, 0);
3810                 return error;
3811         }
3812         xfs_ilock(ip, XFS_ILOCK_EXCL);
3813         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3814
3815         ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3816         ip->i_iocore.io_dmstate  = ip->i_d.di_dmstate  = state;
3817
3818         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3819         IHOLD(ip);
3820         error = xfs_trans_commit(tp, 0, NULL);
3821
3822         return error;
3823 }
3824
3825
3826 /*
3827  * xfs_reclaim
3828  */
3829 STATIC int
3830 xfs_reclaim(
3831         bhv_desc_t      *bdp)
3832 {
3833         xfs_inode_t     *ip;
3834         vnode_t         *vp;
3835
3836         vp = BHV_TO_VNODE(bdp);
3837         ip = XFS_BHVTOI(bdp);
3838
3839         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3840
3841         ASSERT(!VN_MAPPED(vp));
3842
3843         /* bad inode, get out here ASAP */
3844         if (VN_BAD(vp)) {
3845                 xfs_ireclaim(ip);
3846                 return 0;
3847         }
3848
3849         vn_iowait(vp);
3850
3851         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3852         ASSERT(VN_CACHED(vp) == 0);
3853
3854         /* If we have nothing to flush with this inode then complete the
3855          * teardown now, otherwise break the link between the xfs inode
3856          * and the linux inode and clean up the xfs inode later. This
3857          * avoids flushing the inode to disk during the delete operation
3858          * itself.
3859          */
3860         if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3861                 xfs_ilock(ip, XFS_ILOCK_EXCL);
3862                 xfs_iflock(ip);
3863                 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3864         } else {
3865                 xfs_mount_t     *mp = ip->i_mount;
3866
3867                 /* Protect sync from us */
3868                 XFS_MOUNT_ILOCK(mp);
3869                 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3870                 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3871                 ip->i_flags |= XFS_IRECLAIMABLE;
3872                 XFS_MOUNT_IUNLOCK(mp);
3873         }
3874         return 0;
3875 }
3876
3877 int
3878 xfs_finish_reclaim(
3879         xfs_inode_t     *ip,
3880         int             locked,
3881         int             sync_mode)
3882 {
3883         xfs_ihash_t     *ih = ip->i_hash;
3884         vnode_t         *vp = XFS_ITOV_NULL(ip);
3885         int             error;
3886
3887         if (vp && VN_BAD(vp))
3888                 goto reclaim;
3889
3890         /* The hash lock here protects a thread in xfs_iget_core from
3891          * racing with us on linking the inode back with a vnode.
3892          * Once we have the XFS_IRECLAIM flag set it will not touch
3893          * us.
3894          */
3895         write_lock(&ih->ih_lock);
3896         if ((ip->i_flags & XFS_IRECLAIM) ||
3897             (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3898                 write_unlock(&ih->ih_lock);
3899                 if (locked) {
3900                         xfs_ifunlock(ip);
3901                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3902                 }
3903                 return(1);
3904         }
3905         ip->i_flags |= XFS_IRECLAIM;
3906         write_unlock(&ih->ih_lock);
3907
3908         /*
3909          * If the inode is still dirty, then flush it out.  If the inode
3910          * is not in the AIL, then it will be OK to flush it delwri as
3911          * long as xfs_iflush() does not keep any references to the inode.
3912          * We leave that decision up to xfs_iflush() since it has the
3913          * knowledge of whether it's OK to simply do a delwri flush of
3914          * the inode or whether we need to wait until the inode is
3915          * pulled from the AIL.
3916          * We get the flush lock regardless, though, just to make sure
3917          * we don't free it while it is being flushed.
3918          */
3919         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3920                 if (!locked) {
3921                         xfs_ilock(ip, XFS_ILOCK_EXCL);
3922                         xfs_iflock(ip);
3923                 }
3924
3925                 if (ip->i_update_core ||
3926                     ((ip->i_itemp != NULL) &&
3927                      (ip->i_itemp->ili_format.ilf_fields != 0))) {
3928                         error = xfs_iflush(ip, sync_mode);
3929                         /*
3930                          * If we hit an error, typically because of filesystem
3931                          * shutdown, we don't need to let vn_reclaim to know
3932                          * because we're gonna reclaim the inode anyway.
3933                          */
3934                         if (error) {
3935                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3936                                 goto reclaim;
3937                         }
3938                         xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3939                 }
3940
3941                 ASSERT(ip->i_update_core == 0);
3942                 ASSERT(ip->i_itemp == NULL ||
3943                        ip->i_itemp->ili_format.ilf_fields == 0);
3944                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3945         } else if (locked) {
3946                 /*
3947                  * We are not interested in doing an iflush if we're
3948                  * in the process of shutting down the filesystem forcibly.
3949                  * So, just reclaim the inode.
3950                  */
3951                 xfs_ifunlock(ip);
3952                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3953         }
3954
3955  reclaim:
3956         xfs_ireclaim(ip);
3957         return 0;
3958 }
3959
3960 int
3961 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3962 {
3963         int             purged;
3964         xfs_inode_t     *ip, *n;
3965         int             done = 0;
3966
3967         while (!done) {
3968                 purged = 0;
3969                 XFS_MOUNT_ILOCK(mp);
3970                 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3971                         if (noblock) {
3972                                 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3973                                         continue;
3974                                 if (xfs_ipincount(ip) ||
3975                                     !xfs_iflock_nowait(ip)) {
3976                                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3977                                         continue;
3978                                 }
3979                         }
3980                         XFS_MOUNT_IUNLOCK(mp);
3981                         xfs_finish_reclaim(ip, noblock,
3982                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3983                         purged = 1;
3984                         break;
3985                 }
3986
3987                 done = !purged;
3988         }
3989
3990         XFS_MOUNT_IUNLOCK(mp);
3991         return 0;
3992 }
3993
3994 /*
3995  * xfs_alloc_file_space()
3996  *      This routine allocates disk space for the given file.
3997  *
3998  *      If alloc_type == 0, this request is for an ALLOCSP type
3999  *      request which will change the file size.  In this case, no
4000  *      DMAPI event will be generated by the call.  A TRUNCATE event
4001  *      will be generated later by xfs_setattr.
4002  *
4003  *      If alloc_type != 0, this request is for a RESVSP type
4004  *      request, and a DMAPI DM_EVENT_WRITE will be generated if the
4005  *      lower block boundary byte address is less than the file's
4006  *      length.
4007  *
4008  * RETURNS:
4009  *       0 on success
4010  *      errno on error
4011  *
4012  */
4013 STATIC int
4014 xfs_alloc_file_space(
4015         xfs_inode_t             *ip,
4016         xfs_off_t               offset,
4017         xfs_off_t               len,
4018         int                     alloc_type,
4019         int                     attr_flags)
4020 {
4021         xfs_filblks_t           allocated_fsb;
4022         xfs_filblks_t           allocatesize_fsb;
4023         int                     committed;
4024         xfs_off_t               count;
4025         xfs_filblks_t           datablocks;
4026         int                     error;
4027         xfs_fsblock_t           firstfsb;
4028         xfs_bmap_free_t         free_list;
4029         xfs_bmbt_irec_t         *imapp;
4030         xfs_bmbt_irec_t         imaps[1];
4031         xfs_mount_t             *mp;
4032         int                     numrtextents;
4033         int                     reccount;
4034         uint                    resblks;
4035         int                     rt;
4036         int                     rtextsize;
4037         xfs_fileoff_t           startoffset_fsb;
4038         xfs_trans_t             *tp;
4039         int                     xfs_bmapi_flags;
4040
4041         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4042         mp = ip->i_mount;
4043
4044         if (XFS_FORCED_SHUTDOWN(mp))
4045                 return XFS_ERROR(EIO);
4046
4047         /*
4048          * determine if this is a realtime file
4049          */
4050         if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4051                 if (ip->i_d.di_extsize)
4052                         rtextsize = ip->i_d.di_extsize;
4053                 else
4054                         rtextsize = mp->m_sb.sb_rextsize;
4055         } else
4056                 rtextsize = 0;
4057
4058         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4059                 return error;
4060
4061         if (len <= 0)
4062                 return XFS_ERROR(EINVAL);
4063
4064         count = len;
4065         error = 0;
4066         imapp = &imaps[0];
4067         reccount = 1;
4068         xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4069         startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4070         allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4071
4072         /*      Generate a DMAPI event if needed.       */
4073         if (alloc_type != 0 && offset < ip->i_d.di_size &&
4074                         (attr_flags&ATTR_DMI) == 0  &&
4075                         DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4076                 xfs_off_t           end_dmi_offset;
4077
4078                 end_dmi_offset = offset+len;
4079                 if (end_dmi_offset > ip->i_d.di_size)
4080                         end_dmi_offset = ip->i_d.di_size;
4081                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4082                         offset, end_dmi_offset - offset,
4083                         0, NULL);
4084                 if (error)
4085                         return(error);
4086         }
4087
4088         /*
4089          * allocate file space until done or until there is an error
4090          */
4091 retry:
4092         while (allocatesize_fsb && !error) {
4093                 /*
4094                  * determine if reserving space on
4095                  * the data or realtime partition.
4096                  */
4097                 if (rt) {
4098                         xfs_fileoff_t s, e;
4099
4100                         s = startoffset_fsb;
4101                         do_div(s, rtextsize);
4102                         s *= rtextsize;
4103                         e = roundup_64(startoffset_fsb + allocatesize_fsb,
4104                                 rtextsize);
4105                         numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4106                         datablocks = 0;
4107                 } else {
4108                         datablocks = allocatesize_fsb;
4109                         numrtextents = 0;
4110                 }
4111
4112                 /*
4113                  * allocate and setup the transaction
4114                  */
4115                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4116                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4117                 error = xfs_trans_reserve(tp,
4118                                           resblks,
4119                                           XFS_WRITE_LOG_RES(mp),
4120                                           numrtextents,
4121                                           XFS_TRANS_PERM_LOG_RES,
4122                                           XFS_WRITE_LOG_COUNT);
4123
4124                 /*
4125                  * check for running out of space
4126                  */
4127                 if (error) {
4128                         /*
4129                          * Free the transaction structure.
4130                          */
4131                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4132                         xfs_trans_cancel(tp, 0);
4133                         break;
4134                 }
4135                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4136                 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4137                                 ip->i_udquot, ip->i_gdquot, resblks, 0, 0);
4138                 if (error)
4139                         goto error1;
4140
4141                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4142                 xfs_trans_ihold(tp, ip);
4143
4144                 /*
4145                  * issue the bmapi() call to allocate the blocks
4146                  */
4147                 XFS_BMAP_INIT(&free_list, &firstfsb);
4148                 error = xfs_bmapi(tp, ip, startoffset_fsb,
4149                                   allocatesize_fsb, xfs_bmapi_flags,
4150                                   &firstfsb, 0, imapp, &reccount,
4151                                   &free_list);
4152                 if (error) {
4153                         goto error0;
4154                 }
4155
4156                 /*
4157                  * complete the transaction
4158                  */
4159                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4160                 if (error) {
4161                         goto error0;
4162                 }
4163
4164                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4165                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4166                 if (error) {
4167                         break;
4168                 }
4169
4170                 allocated_fsb = imapp->br_blockcount;
4171
4172                 if (reccount == 0) {
4173                         error = XFS_ERROR(ENOSPC);
4174                         break;
4175                 }
4176
4177                 startoffset_fsb += allocated_fsb;
4178                 allocatesize_fsb -= allocated_fsb;
4179         }
4180 dmapi_enospc_check:
4181         if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4182             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4183
4184                 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4185                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4186                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4187                                 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4188                 if (error == 0)
4189                         goto retry;     /* Maybe DMAPI app. has made space */
4190                 /* else fall through with error from XFS_SEND_DATA */
4191         }
4192
4193         return error;
4194
4195  error0:
4196         xfs_bmap_cancel(&free_list);
4197  error1:
4198         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4199         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4200         goto dmapi_enospc_check;
4201 }
4202
4203 /*
4204  * Zero file bytes between startoff and endoff inclusive.
4205  * The iolock is held exclusive and no blocks are buffered.
4206  */
4207 STATIC int
4208 xfs_zero_remaining_bytes(
4209         xfs_inode_t             *ip,
4210         xfs_off_t               startoff,
4211         xfs_off_t               endoff)
4212 {
4213         xfs_bmbt_irec_t         imap;
4214         xfs_fileoff_t           offset_fsb;
4215         xfs_off_t               lastoffset;
4216         xfs_off_t               offset;
4217         xfs_buf_t               *bp;
4218         xfs_mount_t             *mp = ip->i_mount;
4219         int                     nimap;
4220         int                     error = 0;
4221
4222         bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4223                                 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4224                                 mp->m_rtdev_targp : mp->m_ddev_targp);
4225
4226         for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4227                 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4228                 nimap = 1;
4229                 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4230                         &nimap, NULL);
4231                 if (error || nimap < 1)
4232                         break;
4233                 ASSERT(imap.br_blockcount >= 1);
4234                 ASSERT(imap.br_startoff == offset_fsb);
4235                 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4236                 if (lastoffset > endoff)
4237                         lastoffset = endoff;
4238                 if (imap.br_startblock == HOLESTARTBLOCK)
4239                         continue;
4240                 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4241                 if (imap.br_state == XFS_EXT_UNWRITTEN)
4242                         continue;
4243                 XFS_BUF_UNDONE(bp);
4244                 XFS_BUF_UNWRITE(bp);
4245                 XFS_BUF_READ(bp);
4246                 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4247                 xfsbdstrat(mp, bp);
4248                 if ((error = xfs_iowait(bp))) {
4249                         xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4250                                           mp, bp, XFS_BUF_ADDR(bp));
4251                         break;
4252                 }
4253                 memset(XFS_BUF_PTR(bp) +
4254                         (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4255                       0, lastoffset - offset + 1);
4256                 XFS_BUF_UNDONE(bp);
4257                 XFS_BUF_UNREAD(bp);
4258                 XFS_BUF_WRITE(bp);
4259                 xfsbdstrat(mp, bp);
4260                 if ((error = xfs_iowait(bp))) {
4261                         xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4262                                           mp, bp, XFS_BUF_ADDR(bp));
4263                         break;
4264                 }
4265         }
4266         xfs_buf_free(bp);
4267         return error;
4268 }
4269
4270 /*
4271  * xfs_free_file_space()
4272  *      This routine frees disk space for the given file.
4273  *
4274  *      This routine is only called by xfs_change_file_space
4275  *      for an UNRESVSP type call.
4276  *
4277  * RETURNS:
4278  *       0 on success
4279  *      errno on error
4280  *
4281  */
4282 STATIC int
4283 xfs_free_file_space(
4284         xfs_inode_t             *ip,
4285         xfs_off_t               offset,
4286         xfs_off_t               len,
4287         int                     attr_flags)
4288 {
4289         vnode_t                 *vp;
4290         int                     committed;
4291         int                     done;
4292         xfs_off_t               end_dmi_offset;
4293         xfs_fileoff_t           endoffset_fsb;
4294         int                     error;
4295         xfs_fsblock_t           firstfsb;
4296         xfs_bmap_free_t         free_list;
4297         xfs_off_t               ilen;
4298         xfs_bmbt_irec_t         imap;
4299         xfs_off_t               ioffset;
4300         xfs_extlen_t            mod=0;
4301         xfs_mount_t             *mp;
4302         int                     nimap;
4303         uint                    resblks;
4304         int                     rounding;
4305         int                     rt;
4306         xfs_fileoff_t           startoffset_fsb;
4307         xfs_trans_t             *tp;
4308         int                     need_iolock = 1;
4309
4310         vp = XFS_ITOV(ip);
4311         mp = ip->i_mount;
4312
4313         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4314
4315         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4316                 return error;
4317
4318         error = 0;
4319         if (len <= 0)   /* if nothing being freed */
4320                 return error;
4321         rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4322         startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4323         end_dmi_offset = offset + len;
4324         endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4325
4326         if (offset < ip->i_d.di_size &&
4327             (attr_flags & ATTR_DMI) == 0 &&
4328             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4329                 if (end_dmi_offset > ip->i_d.di_size)
4330                         end_dmi_offset = ip->i_d.di_size;
4331                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4332                                 offset, end_dmi_offset - offset,
4333                                 AT_DELAY_FLAG(attr_flags), NULL);
4334                 if (error)
4335                         return(error);
4336         }
4337
4338         ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4339         if (attr_flags & ATTR_NOLOCK)
4340                 need_iolock = 0;
4341         if (need_iolock)
4342                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4343
4344         rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4345                         (__uint8_t)NBPP);
4346         ilen = len + (offset & (rounding - 1));
4347         ioffset = offset & ~(rounding - 1);
4348         if (ilen & (rounding - 1))
4349                 ilen = (ilen + rounding) & ~(rounding - 1);
4350
4351         if (VN_CACHED(vp) != 0) {
4352                 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4353                                 ctooff(offtoct(ioffset)), -1);
4354                 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4355                                 -1, FI_REMAPF_LOCKED);
4356         }
4357
4358         /*
4359          * Need to zero the stuff we're not freeing, on disk.
4360          * If its a realtime file & can't use unwritten extents then we
4361          * actually need to zero the extent edges.  Otherwise xfs_bunmapi
4362          * will take care of it for us.
4363          */
4364         if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4365                 nimap = 1;
4366                 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4367                         &imap, &nimap, NULL);
4368                 if (error)
4369                         goto out_unlock_iolock;
4370                 ASSERT(nimap == 0 || nimap == 1);
4371                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4372                         xfs_daddr_t     block;
4373
4374                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4375                         block = imap.br_startblock;
4376                         mod = do_div(block, mp->m_sb.sb_rextsize);
4377                         if (mod)
4378                                 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4379                 }
4380                 nimap = 1;
4381                 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4382                         &imap, &nimap, NULL);
4383                 if (error)
4384                         goto out_unlock_iolock;
4385                 ASSERT(nimap == 0 || nimap == 1);
4386                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4387                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4388                         mod++;
4389                         if (mod && (mod != mp->m_sb.sb_rextsize))
4390                                 endoffset_fsb -= mod;
4391                 }
4392         }
4393         if ((done = (endoffset_fsb <= startoffset_fsb)))
4394                 /*
4395                  * One contiguous piece to clear
4396                  */
4397                 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4398         else {
4399                 /*
4400                  * Some full blocks, possibly two pieces to clear
4401                  */
4402                 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4403                         error = xfs_zero_remaining_bytes(ip, offset,
4404                                 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4405                 if (!error &&
4406                     XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4407                         error = xfs_zero_remaining_bytes(ip,
4408                                 XFS_FSB_TO_B(mp, endoffset_fsb),
4409                                 offset + len - 1);
4410         }
4411
4412         /*
4413          * free file space until done or until there is an error
4414          */
4415         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4416         while (!error && !done) {
4417
4418                 /*
4419                  * allocate and setup the transaction
4420                  */
4421                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4422                 error = xfs_trans_reserve(tp,
4423                                           resblks,
4424                                           XFS_WRITE_LOG_RES(mp),
4425                                           0,
4426                                           XFS_TRANS_PERM_LOG_RES,
4427                                           XFS_WRITE_LOG_COUNT);
4428
4429                 /*
4430                  * check for running out of space
4431                  */
4432                 if (error) {
4433                         /*
4434                          * Free the transaction structure.
4435                          */
4436                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4437                         xfs_trans_cancel(tp, 0);
4438                         break;
4439                 }
4440                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4441                 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4442                                 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4443                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4444                 if (error)
4445                         goto error1;
4446
4447                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4448                 xfs_trans_ihold(tp, ip);
4449
4450                 /*
4451                  * issue the bunmapi() call to free the blocks
4452                  */
4453                 XFS_BMAP_INIT(&free_list, &firstfsb);
4454                 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4455                                   endoffset_fsb - startoffset_fsb,
4456                                   0, 2, &firstfsb, &free_list, &done);
4457                 if (error) {
4458                         goto error0;
4459                 }
4460
4461                 /*
4462                  * complete the transaction
4463                  */
4464                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4465                 if (error) {
4466                         goto error0;
4467                 }
4468
4469                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4470                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4471         }
4472
4473  out_unlock_iolock:
4474         if (need_iolock)
4475                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4476         return error;
4477
4478  error0:
4479         xfs_bmap_cancel(&free_list);
4480  error1:
4481         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4482         xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4483                     XFS_ILOCK_EXCL);
4484         return error;
4485 }
4486
4487 /*
4488  * xfs_change_file_space()
4489  *      This routine allocates or frees disk space for the given file.
4490  *      The user specified parameters are checked for alignment and size
4491  *      limitations.
4492  *
4493  * RETURNS:
4494  *       0 on success
4495  *      errno on error
4496  *
4497  */
4498 int
4499 xfs_change_file_space(
4500         bhv_desc_t      *bdp,
4501         int             cmd,
4502         xfs_flock64_t   *bf,
4503         xfs_off_t       offset,
4504         cred_t          *credp,
4505         int             attr_flags)
4506 {
4507         int             clrprealloc;
4508         int             error;
4509         xfs_fsize_t     fsize;
4510         xfs_inode_t     *ip;
4511         xfs_mount_t     *mp;
4512         int             setprealloc;
4513         xfs_off_t       startoffset;
4514         xfs_off_t       llen;
4515         xfs_trans_t     *tp;
4516         vattr_t         va;
4517         vnode_t         *vp;
4518
4519         vp = BHV_TO_VNODE(bdp);
4520         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4521
4522         ip = XFS_BHVTOI(bdp);
4523         mp = ip->i_mount;
4524
4525         /*
4526          * must be a regular file and have write permission
4527          */
4528         if (!VN_ISREG(vp))
4529                 return XFS_ERROR(EINVAL);
4530
4531         xfs_ilock(ip, XFS_ILOCK_SHARED);
4532
4533         if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4534                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4535                 return error;
4536         }
4537
4538         xfs_iunlock(ip, XFS_ILOCK_SHARED);
4539
4540         switch (bf->l_whence) {
4541         case 0: /*SEEK_SET*/
4542                 break;
4543         case 1: /*SEEK_CUR*/
4544                 bf->l_start += offset;
4545                 break;
4546         case 2: /*SEEK_END*/
4547                 bf->l_start += ip->i_d.di_size;
4548                 break;
4549         default:
4550                 return XFS_ERROR(EINVAL);
4551         }
4552
4553         llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4554
4555         if (   (bf->l_start < 0)
4556             || (bf->l_start > XFS_MAXIOFFSET(mp))
4557             || (bf->l_start + llen < 0)
4558             || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4559                 return XFS_ERROR(EINVAL);
4560
4561         bf->l_whence = 0;
4562
4563         startoffset = bf->l_start;
4564         fsize = ip->i_d.di_size;
4565
4566         /*
4567          * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4568          * file space.
4569          * These calls do NOT zero the data space allocated to the file,
4570          * nor do they change the file size.
4571          *
4572          * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4573          * space.
4574          * These calls cause the new file data to be zeroed and the file
4575          * size to be changed.
4576          */
4577         setprealloc = clrprealloc = 0;
4578
4579         switch (cmd) {
4580         case XFS_IOC_RESVSP:
4581         case XFS_IOC_RESVSP64:
4582                 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4583                                                                 1, attr_flags);
4584                 if (error)
4585                         return error;
4586                 setprealloc = 1;
4587                 break;
4588
4589         case XFS_IOC_UNRESVSP:
4590         case XFS_IOC_UNRESVSP64:
4591                 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4592                                                                 attr_flags)))
4593                         return error;
4594                 break;
4595
4596         case XFS_IOC_ALLOCSP:
4597         case XFS_IOC_ALLOCSP64:
4598         case XFS_IOC_FREESP:
4599         case XFS_IOC_FREESP64:
4600                 if (startoffset > fsize) {
4601                         error = xfs_alloc_file_space(ip, fsize,
4602                                         startoffset - fsize, 0, attr_flags);
4603                         if (error)
4604                                 break;
4605                 }
4606
4607                 va.va_mask = XFS_AT_SIZE;
4608                 va.va_size = startoffset;
4609
4610                 error = xfs_setattr(bdp, &va, attr_flags, credp);
4611
4612                 if (error)
4613                         return error;
4614
4615                 clrprealloc = 1;
4616                 break;
4617
4618         default:
4619                 ASSERT(0);
4620                 return XFS_ERROR(EINVAL);
4621         }
4622
4623         /*
4624          * update the inode timestamp, mode, and prealloc flag bits
4625          */
4626         tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4627
4628         if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4629                                       0, 0, 0))) {
4630                 /* ASSERT(0); */
4631                 xfs_trans_cancel(tp, 0);
4632                 return error;
4633         }
4634
4635         xfs_ilock(ip, XFS_ILOCK_EXCL);
4636
4637         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4638         xfs_trans_ihold(tp, ip);
4639
4640         if ((attr_flags & ATTR_DMI) == 0) {
4641                 ip->i_d.di_mode &= ~S_ISUID;
4642
4643                 /*
4644                  * Note that we don't have to worry about mandatory
4645                  * file locking being disabled here because we only
4646                  * clear the S_ISGID bit if the Group execute bit is
4647                  * on, but if it was on then mandatory locking wouldn't
4648                  * have been enabled.
4649                  */
4650                 if (ip->i_d.di_mode & S_IXGRP)
4651                         ip->i_d.di_mode &= ~S_ISGID;
4652
4653                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4654         }
4655         if (setprealloc)
4656                 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4657         else if (clrprealloc)
4658                 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4659
4660         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4661         xfs_trans_set_sync(tp);
4662
4663         error = xfs_trans_commit(tp, 0, NULL);
4664
4665         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4666
4667         return error;
4668 }
4669
4670 vnodeops_t xfs_vnodeops = {
4671         BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4672         .vop_open               = xfs_open,
4673         .vop_read               = xfs_read,
4674 #ifdef HAVE_SENDFILE
4675         .vop_sendfile           = xfs_sendfile,
4676 #endif
4677         .vop_write              = xfs_write,
4678         .vop_ioctl              = xfs_ioctl,
4679         .vop_getattr            = xfs_getattr,
4680         .vop_setattr            = xfs_setattr,
4681         .vop_access             = xfs_access,
4682         .vop_lookup             = xfs_lookup,
4683         .vop_create             = xfs_create,
4684         .vop_remove             = xfs_remove,
4685         .vop_link               = xfs_link,
4686         .vop_rename             = xfs_rename,
4687         .vop_mkdir              = xfs_mkdir,
4688         .vop_rmdir              = xfs_rmdir,
4689         .vop_readdir            = xfs_readdir,
4690         .vop_symlink            = xfs_symlink,
4691         .vop_readlink           = xfs_readlink,
4692         .vop_fsync              = xfs_fsync,
4693         .vop_inactive           = xfs_inactive,
4694         .vop_fid2               = xfs_fid2,
4695         .vop_rwlock             = xfs_rwlock,
4696         .vop_rwunlock           = xfs_rwunlock,
4697         .vop_bmap               = xfs_bmap,
4698         .vop_reclaim            = xfs_reclaim,
4699         .vop_attr_get           = xfs_attr_get,
4700         .vop_attr_set           = xfs_attr_set,
4701         .vop_attr_remove        = xfs_attr_remove,
4702         .vop_attr_list          = xfs_attr_list,
4703         .vop_link_removed       = (vop_link_removed_t)fs_noval,
4704         .vop_vnode_change       = (vop_vnode_change_t)fs_noval,
4705         .vop_tosspages          = fs_tosspages,
4706         .vop_flushinval_pages   = fs_flushinval_pages,
4707         .vop_flush_pages        = fs_flush_pages,
4708         .vop_release            = xfs_release,
4709         .vop_iflush             = xfs_inode_flush,
4710 };