Merge ../to-linus-stable/
[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 (vp->v_type == VDIR && 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_type = vp->v_type;
167         vap->va_mode = ip->i_d.di_mode & MODEMASK;
168         vap->va_uid = ip->i_d.di_uid;
169         vap->va_gid = ip->i_d.di_gid;
170         vap->va_projid = ip->i_d.di_projid;
171
172         /*
173          * Check vnode type block/char vs. everything else.
174          * Do it with bitmask because that's faster than looking
175          * for multiple values individually.
176          */
177         if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
178                 vap->va_rdev = 0;
179
180                 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
181
182 #if 0
183                         /* Large block sizes confuse various
184                          * user space programs, so letting the
185                          * stripe size through is not a good
186                          * idea for now.
187                          */
188                         vap->va_blocksize = mp->m_swidth ?
189                                 /*
190                                  * If the underlying volume is a stripe, then
191                                  * return the stripe width in bytes as the
192                                  * recommended I/O size.
193                                  */
194                                 (mp->m_swidth << mp->m_sb.sb_blocklog) :
195                                 /*
196                                  * Return the largest of the preferred buffer
197                                  * sizes since doing small I/Os into larger
198                                  * buffers causes buffers to be decommissioned.
199                                  * The value returned is in bytes.
200                                  */
201                                 (1 << (int)MAX(mp->m_readio_log,
202                                                mp->m_writeio_log));
203
204 #else
205                         vap->va_blocksize =
206                                 /*
207                                  * Return the largest of the preferred buffer
208                                  * sizes since doing small I/Os into larger
209                                  * buffers causes buffers to be decommissioned.
210                                  * The value returned is in bytes.
211                                  */
212                                 1 << (int)MAX(mp->m_readio_log,
213                                                mp->m_writeio_log);
214 #endif
215                 } else {
216
217                         /*
218                          * If the file blocks are being allocated from a
219                          * realtime partition, then return the inode's
220                          * realtime extent size or the realtime volume's
221                          * extent size.
222                          */
223                         vap->va_blocksize = ip->i_d.di_extsize ?
224                                 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
225                                 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
226                 }
227         } else {
228                 vap->va_rdev = ip->i_df.if_u2.if_rdev;
229                 vap->va_blocksize = BLKDEV_IOSIZE;
230         }
231
232         vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
233         vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
234         vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
235         vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
236         vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
237         vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
238
239         /*
240          * Exit for stat callers.  See if any of the rest of the fields
241          * to be filled in are needed.
242          */
243         if ((vap->va_mask &
244              (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
245               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
246                 goto all_done;
247
248         /*
249          * Convert di_flags to xflags.
250          */
251         vap->va_xflags = xfs_ip2xflags(ip);
252
253         /*
254          * Exit for inode revalidate.  See if any of the rest of
255          * the fields to be filled in are needed.
256          */
257         if ((vap->va_mask &
258              (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
259               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
260                 goto all_done;
261
262         vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
263         vap->va_nextents =
264                 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
265                         ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
266                         ip->i_d.di_nextents;
267         if (ip->i_afp)
268                 vap->va_anextents =
269                         (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
270                                 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
271                                  ip->i_d.di_anextents;
272         else
273                 vap->va_anextents = 0;
274         vap->va_gen = ip->i_d.di_gen;
275
276  all_done:
277         if (!(flags & ATTR_LAZY))
278                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
279         return 0;
280 }
281
282
283 /*
284  * xfs_setattr
285  */
286 int
287 xfs_setattr(
288         bhv_desc_t              *bdp,
289         vattr_t                 *vap,
290         int                     flags,
291         cred_t                  *credp)
292 {
293         xfs_inode_t             *ip;
294         xfs_trans_t             *tp;
295         xfs_mount_t             *mp;
296         int                     mask;
297         int                     code;
298         uint                    lock_flags;
299         uint                    commit_flags=0;
300         uid_t                   uid=0, iuid=0;
301         gid_t                   gid=0, igid=0;
302         int                     timeflags = 0;
303         vnode_t                 *vp;
304         xfs_prid_t              projid=0, iprojid=0;
305         int                     mandlock_before, mandlock_after;
306         struct xfs_dquot        *udqp, *gdqp, *olddquot1, *olddquot2;
307         int                     file_owner;
308         int                     need_iolock = 1;
309
310         vp = BHV_TO_VNODE(bdp);
311         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
312
313         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
314                 return XFS_ERROR(EROFS);
315
316         /*
317          * Cannot set certain attributes.
318          */
319         mask = vap->va_mask;
320         if (mask & XFS_AT_NOSET) {
321                 return XFS_ERROR(EINVAL);
322         }
323
324         ip = XFS_BHVTOI(bdp);
325         mp = ip->i_mount;
326
327         if (XFS_FORCED_SHUTDOWN(mp))
328                 return XFS_ERROR(EIO);
329
330         /*
331          * Timestamps do not need to be logged and hence do not
332          * need to be done within a transaction.
333          */
334         if (mask & XFS_AT_UPDTIMES) {
335                 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
336                 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
337                             ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
338                             ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
339                 xfs_ichgtime(ip, timeflags);
340                 return 0;
341         }
342
343         olddquot1 = olddquot2 = NULL;
344         udqp = gdqp = NULL;
345
346         /*
347          * If disk quotas is on, we make sure that the dquots do exist on disk,
348          * before we start any other transactions. Trying to do this later
349          * is messy. We don't care to take a readlock to look at the ids
350          * in inode here, because we can't hold it across the trans_reserve.
351          * If the IDs do change before we take the ilock, we're covered
352          * because the i_*dquot fields will get updated anyway.
353          */
354         if (XFS_IS_QUOTA_ON(mp) &&
355             (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
356                 uint    qflags = 0;
357
358                 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
359                         uid = vap->va_uid;
360                         qflags |= XFS_QMOPT_UQUOTA;
361                 } else {
362                         uid = ip->i_d.di_uid;
363                 }
364                 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
365                         gid = vap->va_gid;
366                         qflags |= XFS_QMOPT_GQUOTA;
367                 }  else {
368                         gid = ip->i_d.di_gid;
369                 }
370                 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
371                         projid = vap->va_projid;
372                         qflags |= XFS_QMOPT_PQUOTA;
373                 }  else {
374                         projid = ip->i_d.di_projid;
375                 }
376                 /*
377                  * We take a reference when we initialize udqp and gdqp,
378                  * so it is important that we never blindly double trip on
379                  * the same variable. See xfs_create() for an example.
380                  */
381                 ASSERT(udqp == NULL);
382                 ASSERT(gdqp == NULL);
383                 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
384                                          &udqp, &gdqp);
385                 if (code)
386                         return (code);
387         }
388
389         /*
390          * For the other attributes, we acquire the inode lock and
391          * first do an error checking pass.
392          */
393         tp = NULL;
394         lock_flags = XFS_ILOCK_EXCL;
395         ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
396         if (flags & ATTR_NOLOCK)
397                 need_iolock = 0;
398         if (!(mask & XFS_AT_SIZE)) {
399                 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
400                     (mp->m_flags & XFS_MOUNT_WSYNC)) {
401                         tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
402                         commit_flags = 0;
403                         if ((code = xfs_trans_reserve(tp, 0,
404                                                      XFS_ICHANGE_LOG_RES(mp), 0,
405                                                      0, 0))) {
406                                 lock_flags = 0;
407                                 goto error_return;
408                         }
409                 }
410         } else {
411                 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
412                     !(flags & ATTR_DMI)) {
413                         int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
414                         code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
415                                 vap->va_size, 0, dmflags, NULL);
416                         if (code) {
417                                 lock_flags = 0;
418                                 goto error_return;
419                         }
420                 }
421                 if (need_iolock)
422                         lock_flags |= XFS_IOLOCK_EXCL;
423         }
424
425         xfs_ilock(ip, lock_flags);
426
427         /* boolean: are we the file owner? */
428         file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
429
430         /*
431          * Change various properties of a file.
432          * Only the owner or users with CAP_FOWNER
433          * capability may do these things.
434          */
435         if (mask &
436             (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
437              XFS_AT_GID|XFS_AT_PROJID)) {
438                 /*
439                  * CAP_FOWNER overrides the following restrictions:
440                  *
441                  * The user ID of the calling process must be equal
442                  * to the file owner ID, except in cases where the
443                  * CAP_FSETID capability is applicable.
444                  */
445                 if (!file_owner && !capable(CAP_FOWNER)) {
446                         code = XFS_ERROR(EPERM);
447                         goto error_return;
448                 }
449
450                 /*
451                  * CAP_FSETID overrides the following restrictions:
452                  *
453                  * The effective user ID of the calling process shall match
454                  * the file owner when setting the set-user-ID and
455                  * set-group-ID bits on that file.
456                  *
457                  * The effective group ID or one of the supplementary group
458                  * IDs of the calling process shall match the group owner of
459                  * the file when setting the set-group-ID bit on that file
460                  */
461                 if (mask & XFS_AT_MODE) {
462                         mode_t m = 0;
463
464                         if ((vap->va_mode & S_ISUID) && !file_owner)
465                                 m |= S_ISUID;
466                         if ((vap->va_mode & S_ISGID) &&
467                             !in_group_p((gid_t)ip->i_d.di_gid))
468                                 m |= S_ISGID;
469 #if 0
470                         /* Linux allows this, Irix doesn't. */
471                         if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
472                                 m |= S_ISVTX;
473 #endif
474                         if (m && !capable(CAP_FSETID))
475                                 vap->va_mode &= ~m;
476                 }
477         }
478
479         /*
480          * Change file ownership.  Must be the owner or privileged.
481          * If the system was configured with the "restricted_chown"
482          * option, the owner is not permitted to give away the file,
483          * and can change the group id only to a group of which he
484          * or she is a member.
485          */
486         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
487                 /*
488                  * These IDs could have changed since we last looked at them.
489                  * But, we're assured that if the ownership did change
490                  * while we didn't have the inode locked, inode's dquot(s)
491                  * would have changed also.
492                  */
493                 iuid = ip->i_d.di_uid;
494                 iprojid = ip->i_d.di_projid;
495                 igid = ip->i_d.di_gid;
496                 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
497                 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
498                 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
499                          iprojid;
500
501                 /*
502                  * CAP_CHOWN overrides the following restrictions:
503                  *
504                  * If _POSIX_CHOWN_RESTRICTED is defined, this capability
505                  * shall override the restriction that a process cannot
506                  * change the user ID of a file it owns and the restriction
507                  * that the group ID supplied to the chown() function
508                  * shall be equal to either the group ID or one of the
509                  * supplementary group IDs of the calling process.
510                  */
511                 if (restricted_chown &&
512                     (iuid != uid || (igid != gid &&
513                                      !in_group_p((gid_t)gid))) &&
514                     !capable(CAP_CHOWN)) {
515                         code = XFS_ERROR(EPERM);
516                         goto error_return;
517                 }
518                 /*
519                  * Do a quota reservation only if uid/projid/gid is actually
520                  * going to change.
521                  */
522                 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
523                     (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
524                     (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
525                         ASSERT(tp);
526                         code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
527                                                 capable(CAP_FOWNER) ?
528                                                 XFS_QMOPT_FORCE_RES : 0);
529                         if (code)       /* out of quota */
530                                 goto error_return;
531                 }
532         }
533
534         /*
535          * Truncate file.  Must have write permission and not be a directory.
536          */
537         if (mask & XFS_AT_SIZE) {
538                 /* Short circuit the truncate case for zero length files */
539                 if ((vap->va_size == 0) &&
540                    (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
541                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
542                         lock_flags &= ~XFS_ILOCK_EXCL;
543                         if (mask & XFS_AT_CTIME)
544                                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
545                         code = 0;
546                         goto error_return;
547                 }
548
549                 if (vp->v_type == VDIR) {
550                         code = XFS_ERROR(EISDIR);
551                         goto error_return;
552                 } else if (vp->v_type != VREG) {
553                         code = XFS_ERROR(EINVAL);
554                         goto error_return;
555                 }
556                 /*
557                  * Make sure that the dquots are attached to the inode.
558                  */
559                 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
560                         goto error_return;
561         }
562
563         /*
564          * Change file access or modified times.
565          */
566         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
567                 if (!file_owner) {
568                         if ((flags & ATTR_UTIME) &&
569                             !capable(CAP_FOWNER)) {
570                                 code = XFS_ERROR(EPERM);
571                                 goto error_return;
572                         }
573                 }
574         }
575
576         /*
577          * Change extent size or realtime flag.
578          */
579         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
580                 /*
581                  * Can't change extent size if any extents are allocated.
582                  */
583                 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
584                     (mask & XFS_AT_EXTSIZE) &&
585                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
586                      vap->va_extsize) ) {
587                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
588                         goto error_return;
589                 }
590
591                 /*
592                  * Can't set extent size unless the file is marked, or
593                  * about to be marked as a realtime file.
594                  *
595                  * This check will be removed when fixed size extents
596                  * with buffered data writes is implemented.
597                  *
598                  */
599                 if ((mask & XFS_AT_EXTSIZE)                     &&
600                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
601                      vap->va_extsize) &&
602                     (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
603                        ((mask & XFS_AT_XFLAGS) &&
604                         (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
605                         code = XFS_ERROR(EINVAL);
606                         goto error_return;
607                 }
608
609                 /*
610                  * Can't change realtime flag if any extents are allocated.
611                  */
612                 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
613                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
614                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
615                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
616                         goto error_return;
617                 }
618                 /*
619                  * Extent size must be a multiple of the appropriate block
620                  * size, if set at all.
621                  */
622                 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
623                         xfs_extlen_t    size;
624
625                         if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
626                             ((mask & XFS_AT_XFLAGS) &&
627                             (vap->va_xflags & XFS_XFLAG_REALTIME))) {
628                                 size = mp->m_sb.sb_rextsize <<
629                                        mp->m_sb.sb_blocklog;
630                         } else {
631                                 size = mp->m_sb.sb_blocksize;
632                         }
633                         if (vap->va_extsize % size) {
634                                 code = XFS_ERROR(EINVAL);
635                                 goto error_return;
636                         }
637                 }
638                 /*
639                  * If realtime flag is set then must have realtime data.
640                  */
641                 if ((mask & XFS_AT_XFLAGS) &&
642                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
643                         if ((mp->m_sb.sb_rblocks == 0) ||
644                             (mp->m_sb.sb_rextsize == 0) ||
645                             (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
646                                 code = XFS_ERROR(EINVAL);
647                                 goto error_return;
648                         }
649                 }
650
651                 /*
652                  * Can't modify an immutable/append-only file unless
653                  * we have appropriate permission.
654                  */
655                 if ((mask & XFS_AT_XFLAGS) &&
656                     (ip->i_d.di_flags &
657                                 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
658                      (vap->va_xflags &
659                                 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
660                     !capable(CAP_LINUX_IMMUTABLE)) {
661                         code = XFS_ERROR(EPERM);
662                         goto error_return;
663                 }
664         }
665
666         /*
667          * Now we can make the changes.  Before we join the inode
668          * to the transaction, if XFS_AT_SIZE is set then take care of
669          * the part of the truncation that must be done without the
670          * inode lock.  This needs to be done before joining the inode
671          * to the transaction, because the inode cannot be unlocked
672          * once it is a part of the transaction.
673          */
674         if (mask & XFS_AT_SIZE) {
675                 code = 0;
676                 if (vap->va_size > ip->i_d.di_size)
677                         code = xfs_igrow_start(ip, vap->va_size, credp);
678                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
679                 if (!code)
680                         code = xfs_itruncate_data(ip, vap->va_size);
681                 if (code) {
682                         ASSERT(tp == NULL);
683                         lock_flags &= ~XFS_ILOCK_EXCL;
684                         ASSERT(lock_flags == XFS_IOLOCK_EXCL);
685                         goto error_return;
686                 }
687                 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
688                 if ((code = xfs_trans_reserve(tp, 0,
689                                              XFS_ITRUNCATE_LOG_RES(mp), 0,
690                                              XFS_TRANS_PERM_LOG_RES,
691                                              XFS_ITRUNCATE_LOG_COUNT))) {
692                         xfs_trans_cancel(tp, 0);
693                         if (need_iolock)
694                                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
695                         return code;
696                 }
697                 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
698                 xfs_ilock(ip, XFS_ILOCK_EXCL);
699         }
700
701         if (tp) {
702                 xfs_trans_ijoin(tp, ip, lock_flags);
703                 xfs_trans_ihold(tp, ip);
704         }
705
706         /* determine whether mandatory locking mode changes */
707         mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
708
709         /*
710          * Truncate file.  Must have write permission and not be a directory.
711          */
712         if (mask & XFS_AT_SIZE) {
713                 if (vap->va_size > ip->i_d.di_size) {
714                         xfs_igrow_finish(tp, ip, vap->va_size,
715                             !(flags & ATTR_DMI));
716                 } else if ((vap->va_size <= ip->i_d.di_size) ||
717                            ((vap->va_size == 0) && ip->i_d.di_nextents)) {
718                         /*
719                          * signal a sync transaction unless
720                          * we're truncating an already unlinked
721                          * file on a wsync filesystem
722                          */
723                         code = xfs_itruncate_finish(&tp, ip,
724                                             (xfs_fsize_t)vap->va_size,
725                                             XFS_DATA_FORK,
726                                             ((ip->i_d.di_nlink != 0 ||
727                                               !(mp->m_flags & XFS_MOUNT_WSYNC))
728                                              ? 1 : 0));
729                         if (code) {
730                                 goto abort_return;
731                         }
732                 }
733                 /*
734                  * Have to do this even if the file's size doesn't change.
735                  */
736                 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
737         }
738
739         /*
740          * Change file access modes.
741          */
742         if (mask & XFS_AT_MODE) {
743                 ip->i_d.di_mode &= S_IFMT;
744                 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
745
746                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
747                 timeflags |= XFS_ICHGTIME_CHG;
748         }
749
750         /*
751          * Change file ownership.  Must be the owner or privileged.
752          * If the system was configured with the "restricted_chown"
753          * option, the owner is not permitted to give away the file,
754          * and can change the group id only to a group of which he
755          * or she is a member.
756          */
757         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
758                 /*
759                  * CAP_FSETID overrides the following restrictions:
760                  *
761                  * The set-user-ID and set-group-ID bits of a file will be
762                  * cleared upon successful return from chown()
763                  */
764                 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
765                     !capable(CAP_FSETID)) {
766                         ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
767                 }
768
769                 /*
770                  * Change the ownerships and register quota modifications
771                  * in the transaction.
772                  */
773                 if (iuid != uid) {
774                         if (XFS_IS_UQUOTA_ON(mp)) {
775                                 ASSERT(mask & XFS_AT_UID);
776                                 ASSERT(udqp);
777                                 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
778                                                         &ip->i_udquot, udqp);
779                         }
780                         ip->i_d.di_uid = uid;
781                 }
782                 if (igid != gid) {
783                         if (XFS_IS_GQUOTA_ON(mp)) {
784                                 ASSERT(!XFS_IS_PQUOTA_ON(mp));
785                                 ASSERT(mask & XFS_AT_GID);
786                                 ASSERT(gdqp);
787                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
788                                                         &ip->i_gdquot, gdqp);
789                         }
790                         ip->i_d.di_gid = gid;
791                 }
792                 if (iprojid != projid) {
793                         if (XFS_IS_PQUOTA_ON(mp)) {
794                                 ASSERT(!XFS_IS_GQUOTA_ON(mp));
795                                 ASSERT(mask & XFS_AT_PROJID);
796                                 ASSERT(gdqp);
797                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
798                                                         &ip->i_gdquot, gdqp);
799                         }
800                         ip->i_d.di_projid = projid;
801                         /*
802                          * We may have to rev the inode as well as
803                          * the superblock version number since projids didn't
804                          * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
805                          */
806                         if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
807                                 xfs_bump_ino_vers2(tp, ip);
808                 }
809
810                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
811                 timeflags |= XFS_ICHGTIME_CHG;
812         }
813
814
815         /*
816          * Change file access or modified times.
817          */
818         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
819                 if (mask & XFS_AT_ATIME) {
820                         ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
821                         ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
822                         ip->i_update_core = 1;
823                         timeflags &= ~XFS_ICHGTIME_ACC;
824                 }
825                 if (mask & XFS_AT_MTIME) {
826                         ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
827                         ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
828                         timeflags &= ~XFS_ICHGTIME_MOD;
829                         timeflags |= XFS_ICHGTIME_CHG;
830                 }
831                 if (tp && (flags & ATTR_UTIME))
832                         xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
833         }
834
835         /*
836          * Change XFS-added attributes.
837          */
838         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
839                 if (mask & XFS_AT_EXTSIZE) {
840                         /*
841                          * Converting bytes to fs blocks.
842                          */
843                         ip->i_d.di_extsize = vap->va_extsize >>
844                                 mp->m_sb.sb_blocklog;
845                 }
846                 if (mask & XFS_AT_XFLAGS) {
847                         uint    di_flags;
848
849                         /* can't set PREALLOC this way, just preserve it */
850                         di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
851                         if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
852                                 di_flags |= XFS_DIFLAG_IMMUTABLE;
853                         if (vap->va_xflags & XFS_XFLAG_APPEND)
854                                 di_flags |= XFS_DIFLAG_APPEND;
855                         if (vap->va_xflags & XFS_XFLAG_SYNC)
856                                 di_flags |= XFS_DIFLAG_SYNC;
857                         if (vap->va_xflags & XFS_XFLAG_NOATIME)
858                                 di_flags |= XFS_DIFLAG_NOATIME;
859                         if (vap->va_xflags & XFS_XFLAG_NODUMP)
860                                 di_flags |= XFS_DIFLAG_NODUMP;
861                         if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
862                                 di_flags |= XFS_DIFLAG_PROJINHERIT;
863                         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
864                                 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
865                                         di_flags |= XFS_DIFLAG_RTINHERIT;
866                                 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
867                                         di_flags |= XFS_DIFLAG_NOSYMLINKS;
868                         } else {
869                                 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
870                                         di_flags |= XFS_DIFLAG_REALTIME;
871                                         ip->i_iocore.io_flags |= XFS_IOCORE_RT;
872                                 } else {
873                                         ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
874                                 }
875                         }
876                         ip->i_d.di_flags = di_flags;
877                 }
878                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
879                 timeflags |= XFS_ICHGTIME_CHG;
880         }
881
882         /*
883          * Change file inode change time only if XFS_AT_CTIME set
884          * AND we have been called by a DMI function.
885          */
886
887         if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
888                 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
889                 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
890                 ip->i_update_core = 1;
891                 timeflags &= ~XFS_ICHGTIME_CHG;
892         }
893
894         /*
895          * Send out timestamp changes that need to be set to the
896          * current time.  Not done when called by a DMI function.
897          */
898         if (timeflags && !(flags & ATTR_DMI))
899                 xfs_ichgtime(ip, timeflags);
900
901         XFS_STATS_INC(xs_ig_attrchg);
902
903         /*
904          * If this is a synchronous mount, make sure that the
905          * transaction goes to disk before returning to the user.
906          * This is slightly sub-optimal in that truncates require
907          * two sync transactions instead of one for wsync filesytems.
908          * One for the truncate and one for the timestamps since we
909          * don't want to change the timestamps unless we're sure the
910          * truncate worked.  Truncates are less than 1% of the laddis
911          * mix so this probably isn't worth the trouble to optimize.
912          */
913         code = 0;
914         if (tp) {
915                 if (mp->m_flags & XFS_MOUNT_WSYNC)
916                         xfs_trans_set_sync(tp);
917
918                 code = xfs_trans_commit(tp, commit_flags, NULL);
919         }
920
921         /*
922          * If the (regular) file's mandatory locking mode changed, then
923          * notify the vnode.  We do this under the inode lock to prevent
924          * racing calls to vop_vnode_change.
925          */
926         mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
927         if (mandlock_before != mandlock_after) {
928                 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
929                                  mandlock_after);
930         }
931
932         xfs_iunlock(ip, lock_flags);
933
934         /*
935          * Release any dquot(s) the inode had kept before chown.
936          */
937         XFS_QM_DQRELE(mp, olddquot1);
938         XFS_QM_DQRELE(mp, olddquot2);
939         XFS_QM_DQRELE(mp, udqp);
940         XFS_QM_DQRELE(mp, gdqp);
941
942         if (code) {
943                 return code;
944         }
945
946         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
947             !(flags & ATTR_DMI)) {
948                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
949                                         NULL, DM_RIGHT_NULL, NULL, NULL,
950                                         0, 0, AT_DELAY_FLAG(flags));
951         }
952         return 0;
953
954  abort_return:
955         commit_flags |= XFS_TRANS_ABORT;
956         /* FALLTHROUGH */
957  error_return:
958         XFS_QM_DQRELE(mp, udqp);
959         XFS_QM_DQRELE(mp, gdqp);
960         if (tp) {
961                 xfs_trans_cancel(tp, commit_flags);
962         }
963         if (lock_flags != 0) {
964                 xfs_iunlock(ip, lock_flags);
965         }
966         return code;
967 }
968
969
970 /*
971  * xfs_access
972  * Null conversion from vnode mode bits to inode mode bits, as in efs.
973  */
974 STATIC int
975 xfs_access(
976         bhv_desc_t      *bdp,
977         int             mode,
978         cred_t          *credp)
979 {
980         xfs_inode_t     *ip;
981         int             error;
982
983         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
984                                                (inst_t *)__return_address);
985
986         ip = XFS_BHVTOI(bdp);
987         xfs_ilock(ip, XFS_ILOCK_SHARED);
988         error = xfs_iaccess(ip, mode, credp);
989         xfs_iunlock(ip, XFS_ILOCK_SHARED);
990         return error;
991 }
992
993
994 /*
995  * xfs_readlink
996  *
997  */
998 STATIC int
999 xfs_readlink(
1000         bhv_desc_t      *bdp,
1001         uio_t           *uiop,
1002         int             ioflags,
1003         cred_t          *credp)
1004 {
1005         xfs_inode_t     *ip;
1006         int             count;
1007         xfs_off_t       offset;
1008         int             pathlen;
1009         vnode_t         *vp;
1010         int             error = 0;
1011         xfs_mount_t     *mp;
1012         int             nmaps;
1013         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1014         xfs_daddr_t     d;
1015         int             byte_cnt;
1016         int             n;
1017         xfs_buf_t       *bp;
1018
1019         vp = BHV_TO_VNODE(bdp);
1020         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1021
1022         ip = XFS_BHVTOI(bdp);
1023         mp = ip->i_mount;
1024
1025         if (XFS_FORCED_SHUTDOWN(mp))
1026                 return XFS_ERROR(EIO);
1027
1028         xfs_ilock(ip, XFS_ILOCK_SHARED);
1029
1030         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1031
1032         offset = uiop->uio_offset;
1033         count = uiop->uio_resid;
1034
1035         if (offset < 0) {
1036                 error = XFS_ERROR(EINVAL);
1037                 goto error_return;
1038         }
1039         if (count <= 0) {
1040                 error = 0;
1041                 goto error_return;
1042         }
1043
1044         if (!(ioflags & IO_INVIS)) {
1045                 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1046         }
1047
1048         /*
1049          * See if the symlink is stored inline.
1050          */
1051         pathlen = (int)ip->i_d.di_size;
1052
1053         if (ip->i_df.if_flags & XFS_IFINLINE) {
1054                 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1055         }
1056         else {
1057                 /*
1058                  * Symlink not inline.  Call bmap to get it in.
1059                  */
1060                 nmaps = SYMLINK_MAPS;
1061
1062                 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1063                                   0, NULL, 0, mval, &nmaps, NULL);
1064
1065                 if (error) {
1066                         goto error_return;
1067                 }
1068
1069                 for (n = 0; n < nmaps; n++) {
1070                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1071                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1072                         bp = xfs_buf_read(mp->m_ddev_targp, d,
1073                                       BTOBB(byte_cnt), 0);
1074                         error = XFS_BUF_GETERROR(bp);
1075                         if (error) {
1076                                 xfs_ioerror_alert("xfs_readlink",
1077                                           ip->i_mount, bp, XFS_BUF_ADDR(bp));
1078                                 xfs_buf_relse(bp);
1079                                 goto error_return;
1080                         }
1081                         if (pathlen < byte_cnt)
1082                                 byte_cnt = pathlen;
1083                         pathlen -= byte_cnt;
1084
1085                         error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1086                         xfs_buf_relse (bp);
1087                 }
1088
1089         }
1090
1091
1092 error_return:
1093
1094         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1095
1096         return error;
1097 }
1098
1099
1100 /*
1101  * xfs_fsync
1102  *
1103  * This is called to sync the inode and its data out to disk.
1104  * We need to hold the I/O lock while flushing the data, and
1105  * the inode lock while flushing the inode.  The inode lock CANNOT
1106  * be held while flushing the data, so acquire after we're done
1107  * with that.
1108  */
1109 STATIC int
1110 xfs_fsync(
1111         bhv_desc_t      *bdp,
1112         int             flag,
1113         cred_t          *credp,
1114         xfs_off_t       start,
1115         xfs_off_t       stop)
1116 {
1117         xfs_inode_t     *ip;
1118         xfs_trans_t     *tp;
1119         int             error;
1120
1121         vn_trace_entry(BHV_TO_VNODE(bdp),
1122                         __FUNCTION__, (inst_t *)__return_address);
1123
1124         ip = XFS_BHVTOI(bdp);
1125
1126         ASSERT(start >= 0 && stop >= -1);
1127
1128         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1129                 return XFS_ERROR(EIO);
1130
1131         /*
1132          * We always need to make sure that the required inode state
1133          * is safe on disk.  The vnode might be clean but because
1134          * of committed transactions that haven't hit the disk yet.
1135          * Likewise, there could be unflushed non-transactional
1136          * changes to the inode core that have to go to disk.
1137          *
1138          * The following code depends on one assumption:  that
1139          * any transaction that changes an inode logs the core
1140          * because it has to change some field in the inode core
1141          * (typically nextents or nblocks).  That assumption
1142          * implies that any transactions against an inode will
1143          * catch any non-transactional updates.  If inode-altering
1144          * transactions exist that violate this assumption, the
1145          * code breaks.  Right now, it figures that if the involved
1146          * update_* field is clear and the inode is unpinned, the
1147          * inode is clean.  Either it's been flushed or it's been
1148          * committed and the commit has hit the disk unpinning the inode.
1149          * (Note that xfs_inode_item_format() called at commit clears
1150          * the update_* fields.)
1151          */
1152         xfs_ilock(ip, XFS_ILOCK_SHARED);
1153
1154         /* If we are flushing data then we care about update_size
1155          * being set, otherwise we care about update_core
1156          */
1157         if ((flag & FSYNC_DATA) ?
1158                         (ip->i_update_size == 0) :
1159                         (ip->i_update_core == 0)) {
1160                 /*
1161                  * Timestamps/size haven't changed since last inode
1162                  * flush or inode transaction commit.  That means
1163                  * either nothing got written or a transaction
1164                  * committed which caught the updates.  If the
1165                  * latter happened and the transaction hasn't
1166                  * hit the disk yet, the inode will be still
1167                  * be pinned.  If it is, force the log.
1168                  */
1169
1170                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1171
1172                 if (xfs_ipincount(ip)) {
1173                         xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1174                                       XFS_LOG_FORCE |
1175                                       ((flag & FSYNC_WAIT)
1176                                        ? XFS_LOG_SYNC : 0));
1177                 }
1178                 error = 0;
1179         } else  {
1180                 /*
1181                  * Kick off a transaction to log the inode
1182                  * core to get the updates.  Make it
1183                  * sync if FSYNC_WAIT is passed in (which
1184                  * is done by everybody but specfs).  The
1185                  * sync transaction will also force the log.
1186                  */
1187                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1188                 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1189                 if ((error = xfs_trans_reserve(tp, 0,
1190                                 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1191                                 0, 0, 0)))  {
1192                         xfs_trans_cancel(tp, 0);
1193                         return error;
1194                 }
1195                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1196
1197                 /*
1198                  * Note - it's possible that we might have pushed
1199                  * ourselves out of the way during trans_reserve
1200                  * which would flush the inode.  But there's no
1201                  * guarantee that the inode buffer has actually
1202                  * gone out yet (it's delwri).  Plus the buffer
1203                  * could be pinned anyway if it's part of an
1204                  * inode in another recent transaction.  So we
1205                  * play it safe and fire off the transaction anyway.
1206                  */
1207                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1208                 xfs_trans_ihold(tp, ip);
1209                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1210                 if (flag & FSYNC_WAIT)
1211                         xfs_trans_set_sync(tp);
1212                 error = xfs_trans_commit(tp, 0, NULL);
1213
1214                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1215         }
1216         return error;
1217 }
1218
1219 /*
1220  * This is called by xfs_inactive to free any blocks beyond eof,
1221  * when the link count isn't zero.
1222  */
1223 STATIC int
1224 xfs_inactive_free_eofblocks(
1225         xfs_mount_t     *mp,
1226         xfs_inode_t     *ip)
1227 {
1228         xfs_trans_t     *tp;
1229         int             error;
1230         xfs_fileoff_t   end_fsb;
1231         xfs_fileoff_t   last_fsb;
1232         xfs_filblks_t   map_len;
1233         int             nimaps;
1234         xfs_bmbt_irec_t imap;
1235
1236         /*
1237          * Figure out if there are any blocks beyond the end
1238          * of the file.  If not, then there is nothing to do.
1239          */
1240         end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1241         last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1242         map_len = last_fsb - end_fsb;
1243         if (map_len <= 0)
1244                 return (0);
1245
1246         nimaps = 1;
1247         xfs_ilock(ip, XFS_ILOCK_SHARED);
1248         error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1249                           NULL, 0, &imap, &nimaps, NULL);
1250         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1251
1252         if (!error && (nimaps != 0) &&
1253             (imap.br_startblock != HOLESTARTBLOCK)) {
1254                 /*
1255                  * Attach the dquots to the inode up front.
1256                  */
1257                 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1258                         return (error);
1259
1260                 /*
1261                  * There are blocks after the end of file.
1262                  * Free them up now by truncating the file to
1263                  * its current size.
1264                  */
1265                 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1266
1267                 /*
1268                  * Do the xfs_itruncate_start() call before
1269                  * reserving any log space because
1270                  * itruncate_start will call into the buffer
1271                  * cache and we can't
1272                  * do that within a transaction.
1273                  */
1274                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1275                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1276                                     ip->i_d.di_size);
1277
1278                 error = xfs_trans_reserve(tp, 0,
1279                                           XFS_ITRUNCATE_LOG_RES(mp),
1280                                           0, XFS_TRANS_PERM_LOG_RES,
1281                                           XFS_ITRUNCATE_LOG_COUNT);
1282                 if (error) {
1283                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1284                         xfs_trans_cancel(tp, 0);
1285                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1286                         return (error);
1287                 }
1288
1289                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1290                 xfs_trans_ijoin(tp, ip,
1291                                 XFS_IOLOCK_EXCL |
1292                                 XFS_ILOCK_EXCL);
1293                 xfs_trans_ihold(tp, ip);
1294
1295                 error = xfs_itruncate_finish(&tp, ip,
1296                                              ip->i_d.di_size,
1297                                              XFS_DATA_FORK,
1298                                              0);
1299                 /*
1300                  * If we get an error at this point we
1301                  * simply don't bother truncating the file.
1302                  */
1303                 if (error) {
1304                         xfs_trans_cancel(tp,
1305                                          (XFS_TRANS_RELEASE_LOG_RES |
1306                                           XFS_TRANS_ABORT));
1307                 } else {
1308                         error = xfs_trans_commit(tp,
1309                                                 XFS_TRANS_RELEASE_LOG_RES,
1310                                                 NULL);
1311                 }
1312                 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1313         }
1314         return (error);
1315 }
1316
1317 /*
1318  * Free a symlink that has blocks associated with it.
1319  */
1320 STATIC int
1321 xfs_inactive_symlink_rmt(
1322         xfs_inode_t     *ip,
1323         xfs_trans_t     **tpp)
1324 {
1325         xfs_buf_t       *bp;
1326         int             committed;
1327         int             done;
1328         int             error;
1329         xfs_fsblock_t   first_block;
1330         xfs_bmap_free_t free_list;
1331         int             i;
1332         xfs_mount_t     *mp;
1333         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1334         int             nmaps;
1335         xfs_trans_t     *ntp;
1336         int             size;
1337         xfs_trans_t     *tp;
1338
1339         tp = *tpp;
1340         mp = ip->i_mount;
1341         ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1342         /*
1343          * We're freeing a symlink that has some
1344          * blocks allocated to it.  Free the
1345          * blocks here.  We know that we've got
1346          * either 1 or 2 extents and that we can
1347          * free them all in one bunmapi call.
1348          */
1349         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1350         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1351                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1352                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1353                 xfs_trans_cancel(tp, 0);
1354                 *tpp = NULL;
1355                 return error;
1356         }
1357         /*
1358          * Lock the inode, fix the size, and join it to the transaction.
1359          * Hold it so in the normal path, we still have it locked for
1360          * the second transaction.  In the error paths we need it
1361          * held so the cancel won't rele it, see below.
1362          */
1363         xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1364         size = (int)ip->i_d.di_size;
1365         ip->i_d.di_size = 0;
1366         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1367         xfs_trans_ihold(tp, ip);
1368         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1369         /*
1370          * Find the block(s) so we can inval and unmap them.
1371          */
1372         done = 0;
1373         XFS_BMAP_INIT(&free_list, &first_block);
1374         nmaps = sizeof(mval) / sizeof(mval[0]);
1375         if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1376                         XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1377                         &free_list)))
1378                 goto error0;
1379         /*
1380          * Invalidate the block(s).
1381          */
1382         for (i = 0; i < nmaps; i++) {
1383                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1384                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1385                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1386                 xfs_trans_binval(tp, bp);
1387         }
1388         /*
1389          * Unmap the dead block(s) to the free_list.
1390          */
1391         if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1392                         &first_block, &free_list, &done)))
1393                 goto error1;
1394         ASSERT(done);
1395         /*
1396          * Commit the first transaction.  This logs the EFI and the inode.
1397          */
1398         if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1399                 goto error1;
1400         /*
1401          * The transaction must have been committed, since there were
1402          * actually extents freed by xfs_bunmapi.  See xfs_bmap_finish.
1403          * The new tp has the extent freeing and EFDs.
1404          */
1405         ASSERT(committed);
1406         /*
1407          * The first xact was committed, so add the inode to the new one.
1408          * Mark it dirty so it will be logged and moved forward in the log as
1409          * part of every commit.
1410          */
1411         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1412         xfs_trans_ihold(tp, ip);
1413         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1414         /*
1415          * Get a new, empty transaction to return to our caller.
1416          */
1417         ntp = xfs_trans_dup(tp);
1418         /*
1419          * Commit the transaction containing extent freeing and EFD's.
1420          * If we get an error on the commit here or on the reserve below,
1421          * we need to unlock the inode since the new transaction doesn't
1422          * have the inode attached.
1423          */
1424         error = xfs_trans_commit(tp, 0, NULL);
1425         tp = ntp;
1426         if (error) {
1427                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1428                 goto error0;
1429         }
1430         /*
1431          * Remove the memory for extent descriptions (just bookkeeping).
1432          */
1433         if (ip->i_df.if_bytes)
1434                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1435         ASSERT(ip->i_df.if_bytes == 0);
1436         /*
1437          * Put an itruncate log reservation in the new transaction
1438          * for our caller.
1439          */
1440         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1441                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1442                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1443                 goto error0;
1444         }
1445         /*
1446          * Return with the inode locked but not joined to the transaction.
1447          */
1448         *tpp = tp;
1449         return 0;
1450
1451  error1:
1452         xfs_bmap_cancel(&free_list);
1453  error0:
1454         /*
1455          * Have to come here with the inode locked and either
1456          * (held and in the transaction) or (not in the transaction).
1457          * If the inode isn't held then cancel would iput it, but
1458          * that's wrong since this is inactive and the vnode ref
1459          * count is 0 already.
1460          * Cancel won't do anything to the inode if held, but it still
1461          * needs to be locked until the cancel is done, if it was
1462          * joined to the transaction.
1463          */
1464         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1465         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1466         *tpp = NULL;
1467         return error;
1468
1469 }
1470
1471 STATIC int
1472 xfs_inactive_symlink_local(
1473         xfs_inode_t     *ip,
1474         xfs_trans_t     **tpp)
1475 {
1476         int             error;
1477
1478         ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1479         /*
1480          * We're freeing a symlink which fit into
1481          * the inode.  Just free the memory used
1482          * to hold the old symlink.
1483          */
1484         error = xfs_trans_reserve(*tpp, 0,
1485                                   XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1486                                   0, XFS_TRANS_PERM_LOG_RES,
1487                                   XFS_ITRUNCATE_LOG_COUNT);
1488
1489         if (error) {
1490                 xfs_trans_cancel(*tpp, 0);
1491                 *tpp = NULL;
1492                 return (error);
1493         }
1494         xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1495
1496         /*
1497          * Zero length symlinks _can_ exist.
1498          */
1499         if (ip->i_df.if_bytes > 0) {
1500                 xfs_idata_realloc(ip,
1501                                   -(ip->i_df.if_bytes),
1502                                   XFS_DATA_FORK);
1503                 ASSERT(ip->i_df.if_bytes == 0);
1504         }
1505         return (0);
1506 }
1507
1508 /*
1509  *
1510  */
1511 STATIC int
1512 xfs_inactive_attrs(
1513         xfs_inode_t     *ip,
1514         xfs_trans_t     **tpp)
1515 {
1516         xfs_trans_t     *tp;
1517         int             error;
1518         xfs_mount_t     *mp;
1519
1520         ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1521         tp = *tpp;
1522         mp = ip->i_mount;
1523         ASSERT(ip->i_d.di_forkoff != 0);
1524         xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1525         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1526
1527         error = xfs_attr_inactive(ip);
1528         if (error) {
1529                 *tpp = NULL;
1530                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1531                 return (error); /* goto out*/
1532         }
1533
1534         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1535         error = xfs_trans_reserve(tp, 0,
1536                                   XFS_IFREE_LOG_RES(mp),
1537                                   0, XFS_TRANS_PERM_LOG_RES,
1538                                   XFS_INACTIVE_LOG_COUNT);
1539         if (error) {
1540                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1541                 xfs_trans_cancel(tp, 0);
1542                 *tpp = NULL;
1543                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1544                 return (error);
1545         }
1546
1547         xfs_ilock(ip, XFS_ILOCK_EXCL);
1548         xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1549         xfs_trans_ihold(tp, ip);
1550         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1551
1552         ASSERT(ip->i_d.di_anextents == 0);
1553
1554         *tpp = tp;
1555         return (0);
1556 }
1557
1558 STATIC int
1559 xfs_release(
1560         bhv_desc_t      *bdp)
1561 {
1562         xfs_inode_t     *ip;
1563         vnode_t         *vp;
1564         xfs_mount_t     *mp;
1565         int             error;
1566
1567         vp = BHV_TO_VNODE(bdp);
1568         ip = XFS_BHVTOI(bdp);
1569
1570         if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1571                 return 0;
1572         }
1573
1574         /* If this is a read-only mount, don't do this (would generate I/O) */
1575         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1576                 return 0;
1577
1578 #ifdef HAVE_REFCACHE
1579         /* If we are in the NFS reference cache then don't do this now */
1580         if (ip->i_refcache)
1581                 return 0;
1582 #endif
1583
1584         mp = ip->i_mount;
1585
1586         if (ip->i_d.di_nlink != 0) {
1587                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1588                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1589                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1590                     (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1591                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1592                                 return (error);
1593                         /* Update linux inode block count after free above */
1594                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1595                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1596                 }
1597         }
1598
1599         return 0;
1600 }
1601
1602 /*
1603  * xfs_inactive
1604  *
1605  * This is called when the vnode reference count for the vnode
1606  * goes to zero.  If the file has been unlinked, then it must
1607  * now be truncated.  Also, we clear all of the read-ahead state
1608  * kept for the inode here since the file is now closed.
1609  */
1610 STATIC int
1611 xfs_inactive(
1612         bhv_desc_t      *bdp,
1613         cred_t          *credp)
1614 {
1615         xfs_inode_t     *ip;
1616         vnode_t         *vp;
1617         xfs_bmap_free_t free_list; 
1618         xfs_fsblock_t   first_block;
1619         int             committed;
1620         xfs_trans_t     *tp;
1621         xfs_mount_t     *mp;
1622         int             error;
1623         int             truncate;
1624
1625         vp = BHV_TO_VNODE(bdp);
1626         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1627
1628         ip = XFS_BHVTOI(bdp);
1629
1630         /*
1631          * If the inode is already free, then there can be nothing
1632          * to clean up here.
1633          */
1634         if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1635                 ASSERT(ip->i_df.if_real_bytes == 0);
1636                 ASSERT(ip->i_df.if_broot_bytes == 0);
1637                 return VN_INACTIVE_CACHE;
1638         }
1639
1640         /*
1641          * Only do a truncate if it's a regular file with
1642          * some actual space in it.  It's OK to look at the
1643          * inode's fields without the lock because we're the
1644          * only one with a reference to the inode.
1645          */
1646         truncate = ((ip->i_d.di_nlink == 0) &&
1647             ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1648             ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1649
1650         mp = ip->i_mount;
1651
1652         if (ip->i_d.di_nlink == 0 &&
1653             DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1654                 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1655         }
1656
1657         error = 0;
1658
1659         /* If this is a read-only mount, don't do this (would generate I/O) */
1660         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1661                 goto out;
1662
1663         if (ip->i_d.di_nlink != 0) {
1664                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1665                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1666                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1667                     (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1668                      (ip->i_delayed_blks != 0))) {
1669                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1670                                 return (VN_INACTIVE_CACHE);
1671                         /* Update linux inode block count after free above */
1672                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1673                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1674                 }
1675                 goto out;
1676         }
1677
1678         ASSERT(ip->i_d.di_nlink == 0);
1679
1680         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1681                 return (VN_INACTIVE_CACHE);
1682
1683         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1684         if (truncate) {
1685                 /*
1686                  * Do the xfs_itruncate_start() call before
1687                  * reserving any log space because itruncate_start
1688                  * will call into the buffer cache and we can't
1689                  * do that within a transaction.
1690                  */
1691                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1692
1693                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1694
1695                 error = xfs_trans_reserve(tp, 0,
1696                                           XFS_ITRUNCATE_LOG_RES(mp),
1697                                           0, XFS_TRANS_PERM_LOG_RES,
1698                                           XFS_ITRUNCATE_LOG_COUNT);
1699                 if (error) {
1700                         /* Don't call itruncate_cleanup */
1701                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1702                         xfs_trans_cancel(tp, 0);
1703                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1704                         return (VN_INACTIVE_CACHE);
1705                 }
1706
1707                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1708                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1709                 xfs_trans_ihold(tp, ip);
1710
1711                 /*
1712                  * normally, we have to run xfs_itruncate_finish sync.
1713                  * But if filesystem is wsync and we're in the inactive
1714                  * path, then we know that nlink == 0, and that the
1715                  * xaction that made nlink == 0 is permanently committed
1716                  * since xfs_remove runs as a synchronous transaction.
1717                  */
1718                 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1719                                 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1720
1721                 if (error) {
1722                         xfs_trans_cancel(tp,
1723                                 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1724                         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1725                         return (VN_INACTIVE_CACHE);
1726                 }
1727         } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1728
1729                 /*
1730                  * If we get an error while cleaning up a
1731                  * symlink we bail out.
1732                  */
1733                 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1734                         xfs_inactive_symlink_rmt(ip, &tp) :
1735                         xfs_inactive_symlink_local(ip, &tp);
1736
1737                 if (error) {
1738                         ASSERT(tp == NULL);
1739                         return (VN_INACTIVE_CACHE);
1740                 }
1741
1742                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1743                 xfs_trans_ihold(tp, ip);
1744         } else {
1745                 error = xfs_trans_reserve(tp, 0,
1746                                           XFS_IFREE_LOG_RES(mp),
1747                                           0, XFS_TRANS_PERM_LOG_RES,
1748                                           XFS_INACTIVE_LOG_COUNT);
1749                 if (error) {
1750                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1751                         xfs_trans_cancel(tp, 0);
1752                         return (VN_INACTIVE_CACHE);
1753                 }
1754
1755                 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1756                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1757                 xfs_trans_ihold(tp, ip);
1758         }
1759
1760         /*
1761          * If there are attributes associated with the file
1762          * then blow them away now.  The code calls a routine
1763          * that recursively deconstructs the attribute fork.
1764          * We need to just commit the current transaction
1765          * because we can't use it for xfs_attr_inactive().
1766          */
1767         if (ip->i_d.di_anextents > 0) {
1768                 error = xfs_inactive_attrs(ip, &tp);
1769                 /*
1770                  * If we got an error, the transaction is already
1771                  * cancelled, and the inode is unlocked. Just get out.
1772                  */
1773                  if (error)
1774                          return (VN_INACTIVE_CACHE);
1775         } else if (ip->i_afp) {
1776                 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1777         }
1778
1779         /*
1780          * Free the inode.
1781          */
1782         XFS_BMAP_INIT(&free_list, &first_block);
1783         error = xfs_ifree(tp, ip, &free_list);
1784         if (error) {
1785                 /*
1786                  * If we fail to free the inode, shut down.  The cancel
1787                  * might do that, we need to make sure.  Otherwise the
1788                  * inode might be lost for a long time or forever.
1789                  */
1790                 if (!XFS_FORCED_SHUTDOWN(mp)) {
1791                         cmn_err(CE_NOTE,
1792                 "xfs_inactive:  xfs_ifree() returned an error = %d on %s",
1793                                 error, mp->m_fsname);
1794                         xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1795                 }
1796                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1797         } else {
1798                 /*
1799                  * Credit the quota account(s). The inode is gone.
1800                  */
1801                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1802
1803                 /*
1804                  * Just ignore errors at this point.  There is
1805                  * nothing we can do except to try to keep going.
1806                  */
1807                 (void) xfs_bmap_finish(&tp,  &free_list, first_block,
1808                                        &committed);
1809                 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1810         }
1811         /*
1812          * Release the dquots held by inode, if any.
1813          */
1814         XFS_QM_DQDETACH(mp, ip);
1815
1816         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1817
1818  out:
1819         return VN_INACTIVE_CACHE;
1820 }
1821
1822
1823 /*
1824  * xfs_lookup
1825  */
1826 STATIC int
1827 xfs_lookup(
1828         bhv_desc_t              *dir_bdp,
1829         vname_t                 *dentry,
1830         vnode_t                 **vpp,
1831         int                     flags,
1832         vnode_t                 *rdir,
1833         cred_t                  *credp)
1834 {
1835         xfs_inode_t             *dp, *ip;
1836         xfs_ino_t               e_inum;
1837         int                     error;
1838         uint                    lock_mode;
1839         vnode_t                 *dir_vp;
1840
1841         dir_vp = BHV_TO_VNODE(dir_bdp);
1842         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1843
1844         dp = XFS_BHVTOI(dir_bdp);
1845
1846         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1847                 return XFS_ERROR(EIO);
1848
1849         lock_mode = xfs_ilock_map_shared(dp);
1850         error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1851         if (!error) {
1852                 *vpp = XFS_ITOV(ip);
1853                 ITRACE(ip);
1854         }
1855         xfs_iunlock_map_shared(dp, lock_mode);
1856         return error;
1857 }
1858
1859
1860 /*
1861  * xfs_create (create a new file).
1862  */
1863 STATIC int
1864 xfs_create(
1865         bhv_desc_t              *dir_bdp,
1866         vname_t                 *dentry,
1867         vattr_t                 *vap,
1868         vnode_t                 **vpp,
1869         cred_t                  *credp)
1870 {
1871         char                    *name = VNAME(dentry);
1872         vnode_t                 *dir_vp;
1873         xfs_inode_t             *dp, *ip;
1874         vnode_t                 *vp=NULL;
1875         xfs_trans_t             *tp;
1876         xfs_mount_t             *mp;
1877         xfs_dev_t               rdev;
1878         int                     error;
1879         xfs_bmap_free_t         free_list;
1880         xfs_fsblock_t           first_block;
1881         boolean_t               dp_joined_to_trans;
1882         int                     dm_event_sent = 0;
1883         uint                    cancel_flags;
1884         int                     committed;
1885         xfs_prid_t              prid;
1886         struct xfs_dquot        *udqp, *gdqp;
1887         uint                    resblks;
1888         int                     dm_di_mode;
1889         int                     namelen;
1890
1891         ASSERT(!*vpp);
1892         dir_vp = BHV_TO_VNODE(dir_bdp);
1893         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1894
1895         dp = XFS_BHVTOI(dir_bdp);
1896         mp = dp->i_mount;
1897
1898         dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1899         namelen = VNAMELEN(dentry);
1900
1901         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1902                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1903                                 dir_vp, DM_RIGHT_NULL, NULL,
1904                                 DM_RIGHT_NULL, name, NULL,
1905                                 dm_di_mode, 0, 0);
1906
1907                 if (error)
1908                         return error;
1909                 dm_event_sent = 1;
1910         }
1911
1912         if (XFS_FORCED_SHUTDOWN(mp))
1913                 return XFS_ERROR(EIO);
1914
1915         /* Return through std_return after this point. */
1916
1917         udqp = gdqp = NULL;
1918         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1919                 prid = dp->i_d.di_projid;
1920         else if (vap->va_mask & XFS_AT_PROJID)
1921                 prid = (xfs_prid_t)vap->va_projid;
1922         else
1923                 prid = (xfs_prid_t)dfltprid;
1924
1925         /*
1926          * Make sure that we have allocated dquot(s) on disk.
1927          */
1928         error = XFS_QM_DQVOPALLOC(mp, dp,
1929                         current_fsuid(credp), current_fsgid(credp), prid,
1930                         XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1931         if (error)
1932                 goto std_return;
1933
1934         ip = NULL;
1935         dp_joined_to_trans = B_FALSE;
1936
1937         tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1938         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1939         resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1940         /*
1941          * Initially assume that the file does not exist and
1942          * reserve the resources for that case.  If that is not
1943          * the case we'll drop the one we have and get a more
1944          * appropriate transaction later.
1945          */
1946         error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1947                         XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1948         if (error == ENOSPC) {
1949                 resblks = 0;
1950                 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1951                                 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1952         }
1953         if (error) {
1954                 cancel_flags = 0;
1955                 dp = NULL;
1956                 goto error_return;
1957         }
1958
1959         xfs_ilock(dp, XFS_ILOCK_EXCL);
1960
1961         XFS_BMAP_INIT(&free_list, &first_block);
1962
1963         ASSERT(ip == NULL);
1964
1965         /*
1966          * Reserve disk quota and the inode.
1967          */
1968         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1969         if (error)
1970                 goto error_return;
1971
1972         if (resblks == 0 &&
1973             (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1974                 goto error_return;
1975         rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1976         error = xfs_dir_ialloc(&tp, dp,
1977                         MAKEIMODE(vap->va_type,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 (src_vp->v_type == VDIR)
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|VTTOIF(vap->va_type);
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,
2883                         MAKEIMODE(vap->va_type,vap->va_mode), 2,
2884                         0, credp, prid, resblks > 0,
2885                 &cdp, NULL);
2886         if (error) {
2887                 if (error == ENOSPC)
2888                         goto error_return;
2889                 goto abort_return;
2890         }
2891         ITRACE(cdp);
2892
2893         /*
2894          * Now we add the directory inode to the transaction.
2895          * We waited until now since xfs_dir_ialloc might start
2896          * a new transaction.  Had we joined the transaction
2897          * earlier, the locks might have gotten released.
2898          */
2899         VN_HOLD(dir_vp);
2900         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2901         dp_joined_to_trans = B_TRUE;
2902
2903         XFS_BMAP_INIT(&free_list, &first_block);
2904
2905         error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2906                         cdp->i_ino, &first_block, &free_list,
2907                         resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2908         if (error) {
2909                 ASSERT(error != ENOSPC);
2910                 goto error1;
2911         }
2912         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2913
2914         /*
2915          * Bump the in memory version number of the parent directory
2916          * so that other processes accessing it will recognize that
2917          * the directory has changed.
2918          */
2919         dp->i_gen++;
2920
2921         error = XFS_DIR_INIT(mp, tp, cdp, dp);
2922         if (error) {
2923                 goto error2;
2924         }
2925
2926         cdp->i_gen = 1;
2927         error = xfs_bumplink(tp, dp);
2928         if (error) {
2929                 goto error2;
2930         }
2931
2932         cvp = XFS_ITOV(cdp);
2933
2934         created = B_TRUE;
2935
2936         *vpp = cvp;
2937         IHOLD(cdp);
2938
2939         /*
2940          * Attach the dquots to the new inode and modify the icount incore.
2941          */
2942         XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2943
2944         /*
2945          * If this is a synchronous mount, make sure that the
2946          * mkdir transaction goes to disk before returning to
2947          * the user.
2948          */
2949         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2950                 xfs_trans_set_sync(tp);
2951         }
2952
2953         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2954         if (error) {
2955                 IRELE(cdp);
2956                 goto error2;
2957         }
2958
2959         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2960         XFS_QM_DQRELE(mp, udqp);
2961         XFS_QM_DQRELE(mp, gdqp);
2962         if (error) {
2963                 IRELE(cdp);
2964         }
2965
2966         /* Fall through to std_return with error = 0 or errno from
2967          * xfs_trans_commit. */
2968
2969 std_return:
2970         if ( (created || (error != 0 && dm_event_sent != 0)) &&
2971                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2972                                                 DM_EVENT_POSTCREATE)) {
2973                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2974                                         dir_vp, DM_RIGHT_NULL,
2975                                         created ? XFS_ITOV(cdp):NULL,
2976                                         DM_RIGHT_NULL,
2977                                         dir_name, NULL,
2978                                         dm_di_mode, error, 0);
2979         }
2980         return error;
2981
2982  error2:
2983  error1:
2984         xfs_bmap_cancel(&free_list);
2985  abort_return:
2986         cancel_flags |= XFS_TRANS_ABORT;
2987  error_return:
2988         xfs_trans_cancel(tp, cancel_flags);
2989         XFS_QM_DQRELE(mp, udqp);
2990         XFS_QM_DQRELE(mp, gdqp);
2991
2992         if (!dp_joined_to_trans && (dp != NULL)) {
2993                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2994         }
2995
2996         goto std_return;
2997 }
2998
2999
3000 /*
3001  * xfs_rmdir
3002  *
3003  */
3004 STATIC int
3005 xfs_rmdir(
3006         bhv_desc_t              *dir_bdp,
3007         vname_t                 *dentry,
3008         cred_t                  *credp)
3009 {
3010         char                    *name = VNAME(dentry);
3011         xfs_inode_t             *dp;
3012         xfs_inode_t             *cdp;   /* child directory */
3013         xfs_trans_t             *tp;
3014         xfs_mount_t             *mp;
3015         int                     error;
3016         xfs_bmap_free_t         free_list;
3017         xfs_fsblock_t           first_block;
3018         int                     cancel_flags;
3019         int                     committed;
3020         vnode_t                 *dir_vp;
3021         int                     dm_di_mode = 0;
3022         int                     last_cdp_link;
3023         int                     namelen;
3024         uint                    resblks;
3025
3026         dir_vp = BHV_TO_VNODE(dir_bdp);
3027         dp = XFS_BHVTOI(dir_bdp);
3028         mp = dp->i_mount;
3029
3030         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3031
3032         if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3033                 return XFS_ERROR(EIO);
3034         namelen = VNAMELEN(dentry);
3035
3036         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3037                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3038                                         dir_vp, DM_RIGHT_NULL,
3039                                         NULL, DM_RIGHT_NULL,
3040                                         name, NULL, 0, 0, 0);
3041                 if (error)
3042                         return XFS_ERROR(error);
3043         }
3044
3045         /* Return through std_return after this point. */
3046
3047         cdp = NULL;
3048
3049         /*
3050          * We need to get a reference to cdp before we get our log
3051          * reservation.  The reason for this is that we cannot call
3052          * xfs_iget for an inode for which we do not have a reference
3053          * once we've acquired a log reservation.  This is because the
3054          * inode we are trying to get might be in xfs_inactive going
3055          * for a log reservation.  Since we'll have to wait for the
3056          * inactive code to complete before returning from xfs_iget,
3057          * we need to make sure that we don't have log space reserved
3058          * when we call xfs_iget.  Instead we get an unlocked referece
3059          * to the inode before getting our log reservation.
3060          */
3061         error = xfs_get_dir_entry(dentry, &cdp);
3062         if (error) {
3063                 REMOVE_DEBUG_TRACE(__LINE__);
3064                 goto std_return;
3065         }
3066         mp = dp->i_mount;
3067         dm_di_mode = cdp->i_d.di_mode;
3068
3069         /*
3070          * Get the dquots for the inodes.
3071          */
3072         error = XFS_QM_DQATTACH(mp, dp, 0);
3073         if (!error && dp != cdp)
3074                 error = XFS_QM_DQATTACH(mp, cdp, 0);
3075         if (error) {
3076                 IRELE(cdp);
3077                 REMOVE_DEBUG_TRACE(__LINE__);
3078                 goto std_return;
3079         }
3080
3081         tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3082         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3083         /*
3084          * We try to get the real space reservation first,
3085          * allowing for directory btree deletion(s) implying
3086          * possible bmap insert(s).  If we can't get the space
3087          * reservation then we use 0 instead, and avoid the bmap
3088          * btree insert(s) in the directory code by, if the bmap
3089          * insert tries to happen, instead trimming the LAST
3090          * block from the directory.
3091          */
3092         resblks = XFS_REMOVE_SPACE_RES(mp);
3093         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3094                         XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3095         if (error == ENOSPC) {
3096                 resblks = 0;
3097                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3098                                 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3099         }
3100         if (error) {
3101                 ASSERT(error != ENOSPC);
3102                 cancel_flags = 0;
3103                 IRELE(cdp);
3104                 goto error_return;
3105         }
3106         XFS_BMAP_INIT(&free_list, &first_block);
3107
3108         /*
3109          * Now lock the child directory inode and the parent directory
3110          * inode in the proper order.  This will take care of validating
3111          * that the directory entry for the child directory inode has
3112          * not changed while we were obtaining a log reservation.
3113          */
3114         error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3115         if (error) {
3116                 xfs_trans_cancel(tp, cancel_flags);
3117                 IRELE(cdp);
3118                 goto std_return;
3119         }
3120
3121         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3122         if (dp != cdp) {
3123                 /*
3124                  * Only increment the parent directory vnode count if
3125                  * we didn't bump it in looking up cdp.  The only time
3126                  * we don't bump it is when we're looking up ".".
3127                  */
3128                 VN_HOLD(dir_vp);
3129         }
3130
3131         ITRACE(cdp);
3132         xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3133
3134         ASSERT(cdp->i_d.di_nlink >= 2);
3135         if (cdp->i_d.di_nlink != 2) {
3136                 error = XFS_ERROR(ENOTEMPTY);
3137                 goto error_return;
3138         }
3139         if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3140                 error = XFS_ERROR(ENOTEMPTY);
3141                 goto error_return;
3142         }
3143
3144         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3145                 &first_block, &free_list, resblks);
3146         if (error) {
3147                 goto error1;
3148         }
3149
3150         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3151
3152         /*
3153          * Bump the in memory generation count on the parent
3154          * directory so that other can know that it has changed.
3155          */
3156         dp->i_gen++;
3157
3158         /*
3159          * Drop the link from cdp's "..".
3160          */
3161         error = xfs_droplink(tp, dp);
3162         if (error) {
3163                 goto error1;
3164         }
3165
3166         /*
3167          * Drop the link from dp to cdp.
3168          */
3169         error = xfs_droplink(tp, cdp);
3170         if (error) {
3171                 goto error1;
3172         }
3173
3174         /*
3175          * Drop the "." link from cdp to self.
3176          */
3177         error = xfs_droplink(tp, cdp);
3178         if (error) {
3179                 goto error1;
3180         }
3181
3182         /* Determine these before committing transaction */
3183         last_cdp_link = (cdp)->i_d.di_nlink==0;
3184
3185         /*
3186          * Take an extra ref on the child vnode so that it
3187          * does not go to xfs_inactive() from within the commit.
3188          */
3189         IHOLD(cdp);
3190
3191         /*
3192          * If this is a synchronous mount, make sure that the
3193          * rmdir transaction goes to disk before returning to
3194          * the user.
3195          */
3196         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3197                 xfs_trans_set_sync(tp);
3198         }
3199
3200         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3201         if (error) {
3202                 xfs_bmap_cancel(&free_list);
3203                 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3204                                  XFS_TRANS_ABORT));
3205                 IRELE(cdp);
3206                 goto std_return;
3207         }
3208
3209         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3210         if (error) {
3211                 IRELE(cdp);
3212                 goto std_return;
3213         }
3214
3215
3216         /*
3217          * Let interposed file systems know about removed links.
3218          */
3219         VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3220
3221         IRELE(cdp);
3222
3223         /* Fall through to std_return with error = 0 or the errno
3224          * from xfs_trans_commit. */
3225 std_return:
3226         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3227                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3228                                         dir_vp, DM_RIGHT_NULL,
3229                                         NULL, DM_RIGHT_NULL,
3230                                         name, NULL, dm_di_mode,
3231                                         error, 0);
3232         }
3233         return error;
3234
3235  error1:
3236         xfs_bmap_cancel(&free_list);
3237         cancel_flags |= XFS_TRANS_ABORT;
3238  error_return:
3239         xfs_trans_cancel(tp, cancel_flags);
3240         goto std_return;
3241 }
3242
3243
3244 /*
3245  * xfs_readdir
3246  *
3247  * Read dp's entries starting at uiop->uio_offset and translate them into
3248  * bufsize bytes worth of struct dirents starting at bufbase.
3249  */
3250 STATIC int
3251 xfs_readdir(
3252         bhv_desc_t      *dir_bdp,
3253         uio_t           *uiop,
3254         cred_t          *credp,
3255         int             *eofp)
3256 {
3257         xfs_inode_t     *dp;
3258         xfs_trans_t     *tp = NULL;
3259         int             error = 0;
3260         uint            lock_mode;
3261         xfs_off_t       start_offset;
3262
3263         vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3264                                                (inst_t *)__return_address);
3265         dp = XFS_BHVTOI(dir_bdp);
3266
3267         if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3268                 return XFS_ERROR(EIO);
3269         }
3270
3271         lock_mode = xfs_ilock_map_shared(dp);
3272         start_offset = uiop->uio_offset;
3273         error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3274         if (start_offset != uiop->uio_offset) {
3275                 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3276         }
3277         xfs_iunlock_map_shared(dp, lock_mode);
3278         return error;
3279 }
3280
3281
3282 /*
3283  * xfs_symlink
3284  *
3285  */
3286 STATIC int
3287 xfs_symlink(
3288         bhv_desc_t              *dir_bdp,
3289         vname_t                 *dentry,
3290         vattr_t                 *vap,
3291         char                    *target_path,
3292         vnode_t                 **vpp,
3293         cred_t                  *credp)
3294 {
3295         xfs_trans_t             *tp;
3296         xfs_mount_t             *mp;
3297         xfs_inode_t             *dp;
3298         xfs_inode_t             *ip;
3299         int                     error;
3300         int                     pathlen;
3301         xfs_bmap_free_t         free_list;
3302         xfs_fsblock_t           first_block;
3303         boolean_t               dp_joined_to_trans;
3304         vnode_t                 *dir_vp;
3305         uint                    cancel_flags;
3306         int                     committed;
3307         xfs_fileoff_t           first_fsb;
3308         xfs_filblks_t           fs_blocks;
3309         int                     nmaps;
3310         xfs_bmbt_irec_t         mval[SYMLINK_MAPS];
3311         xfs_daddr_t             d;
3312         char                    *cur_chunk;
3313         int                     byte_cnt;
3314         int                     n;
3315         xfs_buf_t               *bp;
3316         xfs_prid_t              prid;
3317         struct xfs_dquot        *udqp, *gdqp;
3318         uint                    resblks;
3319         char                    *link_name = VNAME(dentry);
3320         int                     link_namelen;
3321
3322         *vpp = NULL;
3323         dir_vp = BHV_TO_VNODE(dir_bdp);
3324         dp = XFS_BHVTOI(dir_bdp);
3325         dp_joined_to_trans = B_FALSE;
3326         error = 0;
3327         ip = NULL;
3328         tp = NULL;
3329
3330         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3331
3332         mp = dp->i_mount;
3333
3334         if (XFS_FORCED_SHUTDOWN(mp))
3335                 return XFS_ERROR(EIO);
3336
3337         link_namelen = VNAMELEN(dentry);
3338
3339         /*
3340          * Check component lengths of the target path name.
3341          */
3342         pathlen = strlen(target_path);
3343         if (pathlen >= MAXPATHLEN)      /* total string too long */
3344                 return XFS_ERROR(ENAMETOOLONG);
3345         if (pathlen >= MAXNAMELEN) {    /* is any component too long? */
3346                 int len, total;
3347                 char *path;
3348
3349                 for(total = 0, path = target_path; total < pathlen;) {
3350                         /*
3351                          * Skip any slashes.
3352                          */
3353                         while(*path == '/') {
3354                                 total++;
3355                                 path++;
3356                         }
3357
3358                         /*
3359                          * Count up to the next slash or end of path.
3360                          * Error out if the component is bigger than MAXNAMELEN.
3361                          */
3362                         for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3363                                 if (++len >= MAXNAMELEN) {
3364                                         error = ENAMETOOLONG;
3365                                         return error;
3366                                 }
3367                         }
3368                 }
3369         }
3370
3371         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3372                 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3373                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3374                                         link_name, target_path, 0, 0, 0);
3375                 if (error)
3376                         return error;
3377         }
3378
3379         /* Return through std_return after this point. */
3380
3381         udqp = gdqp = NULL;
3382         if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3383                 prid = dp->i_d.di_projid;
3384         else if (vap->va_mask & XFS_AT_PROJID)
3385                 prid = (xfs_prid_t)vap->va_projid;
3386         else
3387                 prid = (xfs_prid_t)dfltprid;
3388
3389         /*
3390          * Make sure that we have allocated dquot(s) on disk.
3391          */
3392         error = XFS_QM_DQVOPALLOC(mp, dp,
3393                         current_fsuid(credp), current_fsgid(credp), prid,
3394                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3395         if (error)
3396                 goto std_return;
3397
3398         tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3399         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3400         /*
3401          * The symlink will fit into the inode data fork?
3402          * There can't be any attributes so we get the whole variable part.
3403          */
3404         if (pathlen <= XFS_LITINO(mp))
3405                 fs_blocks = 0;
3406         else
3407                 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3408         resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3409         error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3410                         XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3411         if (error == ENOSPC && fs_blocks == 0) {
3412                 resblks = 0;
3413                 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3414                                 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3415         }
3416         if (error) {
3417                 cancel_flags = 0;
3418                 dp = NULL;
3419                 goto error_return;
3420         }
3421
3422         xfs_ilock(dp, XFS_ILOCK_EXCL);
3423
3424         /*
3425          * Check whether the directory allows new symlinks or not.
3426          */
3427         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3428                 error = XFS_ERROR(EPERM);
3429                 goto error_return;
3430         }
3431
3432         /*
3433          * Reserve disk quota : blocks and inode.
3434          */
3435         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3436         if (error)
3437                 goto error_return;
3438
3439         /*
3440          * Check for ability to enter directory entry, if no space reserved.
3441          */
3442         if (resblks == 0 &&
3443             (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3444                 goto error_return;
3445         /*
3446          * Initialize the bmap freelist prior to calling either
3447          * bmapi or the directory create code.
3448          */
3449         XFS_BMAP_INIT(&free_list, &first_block);
3450
3451         /*
3452          * Allocate an inode for the symlink.
3453          */
3454         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3455                                1, 0, credp, prid, resblks > 0, &ip, NULL);
3456         if (error) {
3457                 if (error == ENOSPC)
3458                         goto error_return;
3459                 goto error1;
3460         }
3461         ITRACE(ip);
3462
3463         VN_HOLD(dir_vp);
3464         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3465         dp_joined_to_trans = B_TRUE;
3466
3467         /*
3468          * Also attach the dquot(s) to it, if applicable.
3469          */
3470         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3471
3472         if (resblks)
3473                 resblks -= XFS_IALLOC_SPACE_RES(mp);
3474         /*
3475          * If the symlink will fit into the inode, write it inline.
3476          */
3477         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3478                 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3479                 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3480                 ip->i_d.di_size = pathlen;
3481
3482                 /*
3483                  * The inode was initially created in extent format.
3484                  */
3485                 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3486                 ip->i_df.if_flags |= XFS_IFINLINE;
3487
3488                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3489                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3490
3491         } else {
3492                 first_fsb = 0;
3493                 nmaps = SYMLINK_MAPS;
3494
3495                 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3496                                   XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3497                                   &first_block, resblks, mval, &nmaps,
3498                                   &free_list);
3499                 if (error) {
3500                         goto error1;
3501                 }
3502
3503                 if (resblks)
3504                         resblks -= fs_blocks;
3505                 ip->i_d.di_size = pathlen;
3506                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3507
3508                 cur_chunk = target_path;
3509                 for (n = 0; n < nmaps; n++) {
3510                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3511                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3512                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3513                                                BTOBB(byte_cnt), 0);
3514                         ASSERT(bp && !XFS_BUF_GETERROR(bp));
3515                         if (pathlen < byte_cnt) {
3516                                 byte_cnt = pathlen;
3517                         }
3518                         pathlen -= byte_cnt;
3519
3520                         memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3521                         cur_chunk += byte_cnt;
3522
3523                         xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3524                 }
3525         }
3526
3527         /*
3528          * Create the directory entry for the symlink.
3529          */
3530         error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3531                         ip->i_ino, &first_block, &free_list, resblks);
3532         if (error) {
3533                 goto error1;
3534         }
3535         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3536         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3537
3538         /*
3539          * Bump the in memory version number of the parent directory
3540          * so that other processes accessing it will recognize that
3541          * the directory has changed.
3542          */
3543         dp->i_gen++;
3544
3545         /*
3546          * If this is a synchronous mount, make sure that the
3547          * symlink transaction goes to disk before returning to
3548          * the user.
3549          */
3550         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3551                 xfs_trans_set_sync(tp);
3552         }
3553
3554         /*
3555          * xfs_trans_commit normally decrements the vnode ref count
3556          * when it unlocks the inode. Since we want to return the
3557          * vnode to the caller, we bump the vnode ref count now.
3558          */
3559         IHOLD(ip);
3560
3561         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3562         if (error) {
3563                 goto error2;
3564         }
3565         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3566         XFS_QM_DQRELE(mp, udqp);
3567         XFS_QM_DQRELE(mp, gdqp);
3568
3569         /* Fall through to std_return with error = 0 or errno from
3570          * xfs_trans_commit     */
3571 std_return:
3572         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3573                              DM_EVENT_POSTSYMLINK)) {
3574                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3575                                         dir_vp, DM_RIGHT_NULL,
3576                                         error ? NULL : XFS_ITOV(ip),
3577                                         DM_RIGHT_NULL, link_name, target_path,
3578                                         0, error, 0);
3579         }
3580
3581         if (!error) {
3582                 vnode_t *vp;
3583
3584                 ASSERT(ip);
3585                 vp = XFS_ITOV(ip);
3586                 *vpp = vp;
3587         }
3588         return error;
3589
3590  error2:
3591         IRELE(ip);
3592  error1:
3593         xfs_bmap_cancel(&free_list);
3594         cancel_flags |= XFS_TRANS_ABORT;
3595  error_return:
3596         xfs_trans_cancel(tp, cancel_flags);
3597         XFS_QM_DQRELE(mp, udqp);
3598         XFS_QM_DQRELE(mp, gdqp);
3599
3600         if (!dp_joined_to_trans && (dp != NULL)) {
3601                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3602         }
3603
3604         goto std_return;
3605 }
3606
3607
3608 /*
3609  * xfs_fid2
3610  *
3611  * A fid routine that takes a pointer to a previously allocated
3612  * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3613  */
3614 STATIC int
3615 xfs_fid2(
3616         bhv_desc_t      *bdp,
3617         fid_t           *fidp)
3618 {
3619         xfs_inode_t     *ip;
3620         xfs_fid2_t      *xfid;
3621
3622         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3623                                        (inst_t *)__return_address);
3624         ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3625
3626         xfid = (xfs_fid2_t *)fidp;
3627         ip = XFS_BHVTOI(bdp);
3628         xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3629         xfid->fid_pad = 0;
3630         /*
3631          * use memcpy because the inode is a long long and there's no
3632          * assurance that xfid->fid_ino is properly aligned.
3633          */
3634         memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3635         xfid->fid_gen = ip->i_d.di_gen;
3636
3637         return 0;
3638 }
3639
3640
3641 /*
3642  * xfs_rwlock
3643  */
3644 int
3645 xfs_rwlock(
3646         bhv_desc_t      *bdp,
3647         vrwlock_t       locktype)
3648 {
3649         xfs_inode_t     *ip;
3650         vnode_t         *vp;
3651
3652         vp = BHV_TO_VNODE(bdp);
3653         if (vp->v_type == VDIR)
3654                 return 1;
3655         ip = XFS_BHVTOI(bdp);
3656         if (locktype == VRWLOCK_WRITE) {
3657                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3658         } else if (locktype == VRWLOCK_TRY_READ) {
3659                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3660         } else if (locktype == VRWLOCK_TRY_WRITE) {
3661                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3662         } else {
3663                 ASSERT((locktype == VRWLOCK_READ) ||
3664                        (locktype == VRWLOCK_WRITE_DIRECT));
3665                 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3666         }
3667
3668         return 1;
3669 }
3670
3671
3672 /*
3673  * xfs_rwunlock
3674  */
3675 void
3676 xfs_rwunlock(
3677         bhv_desc_t      *bdp,
3678         vrwlock_t       locktype)
3679 {
3680         xfs_inode_t     *ip;
3681         vnode_t         *vp;
3682
3683         vp = BHV_TO_VNODE(bdp);
3684         if (vp->v_type == VDIR)
3685                 return;
3686         ip = XFS_BHVTOI(bdp);
3687         if (locktype == VRWLOCK_WRITE) {
3688                 /*
3689                  * In the write case, we may have added a new entry to
3690                  * the reference cache.  This might store a pointer to
3691                  * an inode to be released in this inode.  If it is there,
3692                  * clear the pointer and release the inode after unlocking
3693                  * this one.
3694                  */
3695                 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3696         } else {
3697                 ASSERT((locktype == VRWLOCK_READ) ||
3698                        (locktype == VRWLOCK_WRITE_DIRECT));
3699                 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3700         }
3701         return;
3702 }
3703
3704 STATIC int
3705 xfs_inode_flush(
3706         bhv_desc_t      *bdp,
3707         int             flags)
3708 {
3709         xfs_inode_t     *ip;
3710         xfs_mount_t     *mp;
3711         xfs_inode_log_item_t *iip;
3712         int             error = 0;
3713
3714         ip = XFS_BHVTOI(bdp);
3715         mp = ip->i_mount;
3716         iip = ip->i_itemp;
3717
3718         if (XFS_FORCED_SHUTDOWN(mp))
3719                 return XFS_ERROR(EIO);
3720
3721         /*
3722          * Bypass inodes which have already been cleaned by
3723          * the inode flush clustering code inside xfs_iflush
3724          */
3725         if ((ip->i_update_core == 0) &&
3726             ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3727                 return 0;
3728
3729         if (flags & FLUSH_LOG) {
3730                 if (iip && iip->ili_last_lsn) {
3731                         xlog_t          *log = mp->m_log;
3732                         xfs_lsn_t       sync_lsn;
3733                         int             s, log_flags = XFS_LOG_FORCE;
3734
3735                         s = GRANT_LOCK(log);
3736                         sync_lsn = log->l_last_sync_lsn;
3737                         GRANT_UNLOCK(log, s);
3738
3739                         if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3740                                 return 0;
3741
3742                         if (flags & FLUSH_SYNC)
3743                                 log_flags |= XFS_LOG_SYNC;
3744                         return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3745                 }
3746         }
3747
3748         /*
3749          * We make this non-blocking if the inode is contended,
3750          * return EAGAIN to indicate to the caller that they
3751          * did not succeed. This prevents the flush path from
3752          * blocking on inodes inside another operation right
3753          * now, they get caught later by xfs_sync.
3754          */
3755         if (flags & FLUSH_INODE) {
3756                 int     flush_flags;
3757
3758                 if (xfs_ipincount(ip))
3759                         return EAGAIN;
3760
3761                 if (flags & FLUSH_SYNC) {
3762                         xfs_ilock(ip, XFS_ILOCK_SHARED);
3763                         xfs_iflock(ip);
3764                 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3765                         if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3766                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3767                                 return EAGAIN;
3768                         }
3769                 } else {
3770                         return EAGAIN;
3771                 }
3772
3773                 if (flags & FLUSH_SYNC)
3774                         flush_flags = XFS_IFLUSH_SYNC;
3775                 else
3776                         flush_flags = XFS_IFLUSH_ASYNC;
3777
3778                 error = xfs_iflush(ip, flush_flags);
3779                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3780         }
3781
3782         return error;
3783 }
3784
3785
3786 int
3787 xfs_set_dmattrs (
3788         bhv_desc_t      *bdp,
3789         u_int           evmask,
3790         u_int16_t       state,
3791         cred_t          *credp)
3792 {
3793         xfs_inode_t     *ip;
3794         xfs_trans_t     *tp;
3795         xfs_mount_t     *mp;
3796         int             error;
3797
3798         if (!capable(CAP_SYS_ADMIN))
3799                 return XFS_ERROR(EPERM);
3800
3801         ip = XFS_BHVTOI(bdp);
3802         mp = ip->i_mount;
3803
3804         if (XFS_FORCED_SHUTDOWN(mp))
3805                 return XFS_ERROR(EIO);
3806
3807         tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3808         error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3809         if (error) {
3810                 xfs_trans_cancel(tp, 0);
3811                 return error;
3812         }
3813         xfs_ilock(ip, XFS_ILOCK_EXCL);
3814         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3815
3816         ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3817         ip->i_iocore.io_dmstate  = ip->i_d.di_dmstate  = state;
3818
3819         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3820         IHOLD(ip);
3821         error = xfs_trans_commit(tp, 0, NULL);
3822
3823         return error;
3824 }
3825
3826
3827 /*
3828  * xfs_reclaim
3829  */
3830 STATIC int
3831 xfs_reclaim(
3832         bhv_desc_t      *bdp)
3833 {
3834         xfs_inode_t     *ip;
3835         vnode_t         *vp;
3836
3837         vp = BHV_TO_VNODE(bdp);
3838         ip = XFS_BHVTOI(bdp);
3839
3840         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3841
3842         ASSERT(!VN_MAPPED(vp));
3843
3844         /* bad inode, get out here ASAP */
3845         if (VN_BAD(vp)) {
3846                 xfs_ireclaim(ip);
3847                 return 0;
3848         }
3849
3850         if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3851                 if (ip->i_d.di_size > 0) {
3852                         /*
3853                          * Flush and invalidate any data left around that is
3854                          * a part of this file.
3855                          *
3856                          * Get the inode's i/o lock so that buffers are pushed
3857                          * out while holding the proper lock.  We can't hold
3858                          * the inode lock here since flushing out buffers may
3859                          * cause us to try to get the lock in xfs_strategy().
3860                          *
3861                          * We don't have to call remapf() here, because there
3862                          * cannot be any mapped file references to this vnode
3863                          * since it is being reclaimed.
3864                          */
3865                         xfs_ilock(ip, XFS_IOLOCK_EXCL);
3866
3867                         /*
3868                          * If we hit an IO error, we need to make sure that the
3869                          * buffer and page caches of file data for
3870                          * the file are tossed away. We don't want to use
3871                          * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3872                          * pages to stay attached to the vnode, but be
3873                          * marked P_BAD. pdflush/vnode_pagebad
3874                          * hates that.
3875                          */
3876                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3877                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3878                         } else {
3879                                 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3880                         }
3881
3882                         ASSERT(VN_CACHED(vp) == 0);
3883                         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3884                                ip->i_delayed_blks == 0);
3885                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3886                 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3887                         /*
3888                          * di_size field may not be quite accurate if we're
3889                          * shutting down.
3890                          */
3891                         VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3892                         ASSERT(VN_CACHED(vp) == 0);
3893                 }
3894         }
3895
3896         /* If we have nothing to flush with this inode then complete the
3897          * teardown now, otherwise break the link between the xfs inode
3898          * and the linux inode and clean up the xfs inode later. This
3899          * avoids flushing the inode to disk during the delete operation
3900          * itself.
3901          */
3902         if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3903                 xfs_ilock(ip, XFS_ILOCK_EXCL);
3904                 xfs_iflock(ip);
3905                 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3906         } else {
3907                 xfs_mount_t     *mp = ip->i_mount;
3908
3909                 /* Protect sync from us */
3910                 XFS_MOUNT_ILOCK(mp);
3911                 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3912                 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3913                 ip->i_flags |= XFS_IRECLAIMABLE;
3914                 XFS_MOUNT_IUNLOCK(mp);
3915         }
3916         return 0;
3917 }
3918
3919 int
3920 xfs_finish_reclaim(
3921         xfs_inode_t     *ip,
3922         int             locked,
3923         int             sync_mode)
3924 {
3925         xfs_ihash_t     *ih = ip->i_hash;
3926         vnode_t         *vp = XFS_ITOV_NULL(ip);
3927         int             error;
3928
3929         if (vp && VN_BAD(vp))
3930                 goto reclaim;
3931
3932         /* The hash lock here protects a thread in xfs_iget_core from
3933          * racing with us on linking the inode back with a vnode.
3934          * Once we have the XFS_IRECLAIM flag set it will not touch
3935          * us.
3936          */
3937         write_lock(&ih->ih_lock);
3938         if ((ip->i_flags & XFS_IRECLAIM) ||
3939             (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3940                 write_unlock(&ih->ih_lock);
3941                 if (locked) {
3942                         xfs_ifunlock(ip);
3943                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3944                 }
3945                 return(1);
3946         }
3947         ip->i_flags |= XFS_IRECLAIM;
3948         write_unlock(&ih->ih_lock);
3949
3950         /*
3951          * If the inode is still dirty, then flush it out.  If the inode
3952          * is not in the AIL, then it will be OK to flush it delwri as
3953          * long as xfs_iflush() does not keep any references to the inode.
3954          * We leave that decision up to xfs_iflush() since it has the
3955          * knowledge of whether it's OK to simply do a delwri flush of
3956          * the inode or whether we need to wait until the inode is
3957          * pulled from the AIL.
3958          * We get the flush lock regardless, though, just to make sure
3959          * we don't free it while it is being flushed.
3960          */
3961         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3962                 if (!locked) {
3963                         xfs_ilock(ip, XFS_ILOCK_EXCL);
3964                         xfs_iflock(ip);
3965                 }
3966
3967                 if (ip->i_update_core ||
3968                     ((ip->i_itemp != NULL) &&
3969                      (ip->i_itemp->ili_format.ilf_fields != 0))) {
3970                         error = xfs_iflush(ip, sync_mode);
3971                         /*
3972                          * If we hit an error, typically because of filesystem
3973                          * shutdown, we don't need to let vn_reclaim to know
3974                          * because we're gonna reclaim the inode anyway.
3975                          */
3976                         if (error) {
3977                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3978                                 goto reclaim;
3979                         }
3980                         xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3981                 }
3982
3983                 ASSERT(ip->i_update_core == 0);
3984                 ASSERT(ip->i_itemp == NULL ||
3985                        ip->i_itemp->ili_format.ilf_fields == 0);
3986                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3987         } else if (locked) {
3988                 /*
3989                  * We are not interested in doing an iflush if we're
3990                  * in the process of shutting down the filesystem forcibly.
3991                  * So, just reclaim the inode.
3992                  */
3993                 xfs_ifunlock(ip);
3994                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3995         }
3996
3997  reclaim:
3998         xfs_ireclaim(ip);
3999         return 0;
4000 }
4001
4002 int
4003 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
4004 {
4005         int             purged;
4006         xfs_inode_t     *ip, *n;
4007         int             done = 0;
4008
4009         while (!done) {
4010                 purged = 0;
4011                 XFS_MOUNT_ILOCK(mp);
4012                 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
4013                         if (noblock) {
4014                                 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4015                                         continue;
4016                                 if (xfs_ipincount(ip) ||
4017                                     !xfs_iflock_nowait(ip)) {
4018                                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4019                                         continue;
4020                                 }
4021                         }
4022                         XFS_MOUNT_IUNLOCK(mp);
4023                         xfs_finish_reclaim(ip, noblock,
4024                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4025                         purged = 1;
4026                         break;
4027                 }
4028
4029                 done = !purged;
4030         }
4031
4032         XFS_MOUNT_IUNLOCK(mp);
4033         return 0;
4034 }
4035
4036 /*
4037  * xfs_alloc_file_space()
4038  *      This routine allocates disk space for the given file.
4039  *
4040  *      If alloc_type == 0, this request is for an ALLOCSP type
4041  *      request which will change the file size.  In this case, no
4042  *      DMAPI event will be generated by the call.  A TRUNCATE event
4043  *      will be generated later by xfs_setattr.
4044  *
4045  *      If alloc_type != 0, this request is for a RESVSP type
4046  *      request, and a DMAPI DM_EVENT_WRITE will be generated if the
4047  *      lower block boundary byte address is less than the file's
4048  *      length.
4049  *
4050  * RETURNS:
4051  *       0 on success
4052  *      errno on error
4053  *
4054  */
4055 STATIC int
4056 xfs_alloc_file_space(
4057         xfs_inode_t             *ip,
4058         xfs_off_t               offset,
4059         xfs_off_t               len,
4060         int                     alloc_type,
4061         int                     attr_flags)
4062 {
4063         xfs_filblks_t           allocated_fsb;
4064         xfs_filblks_t           allocatesize_fsb;
4065         int                     committed;
4066         xfs_off_t               count;
4067         xfs_filblks_t           datablocks;
4068         int                     error;
4069         xfs_fsblock_t           firstfsb;
4070         xfs_bmap_free_t         free_list;
4071         xfs_bmbt_irec_t         *imapp;
4072         xfs_bmbt_irec_t         imaps[1];
4073         xfs_mount_t             *mp;
4074         int                     numrtextents;
4075         int                     reccount;
4076         uint                    resblks;
4077         int                     rt;
4078         int                     rtextsize;
4079         xfs_fileoff_t           startoffset_fsb;
4080         xfs_trans_t             *tp;
4081         int                     xfs_bmapi_flags;
4082
4083         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4084         mp = ip->i_mount;
4085
4086         if (XFS_FORCED_SHUTDOWN(mp))
4087                 return XFS_ERROR(EIO);
4088
4089         /*
4090          * determine if this is a realtime file
4091          */
4092         if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4093                 if (ip->i_d.di_extsize)
4094                         rtextsize = ip->i_d.di_extsize;
4095                 else
4096                         rtextsize = mp->m_sb.sb_rextsize;
4097         } else
4098                 rtextsize = 0;
4099
4100         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4101                 return error;
4102
4103         if (len <= 0)
4104                 return XFS_ERROR(EINVAL);
4105
4106         count = len;
4107         error = 0;
4108         imapp = &imaps[0];
4109         reccount = 1;
4110         xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4111         startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4112         allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4113
4114         /*      Generate a DMAPI event if needed.       */
4115         if (alloc_type != 0 && offset < ip->i_d.di_size &&
4116                         (attr_flags&ATTR_DMI) == 0  &&
4117                         DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4118                 xfs_off_t           end_dmi_offset;
4119
4120                 end_dmi_offset = offset+len;
4121                 if (end_dmi_offset > ip->i_d.di_size)
4122                         end_dmi_offset = ip->i_d.di_size;
4123                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4124                         offset, end_dmi_offset - offset,
4125                         0, NULL);
4126                 if (error)
4127                         return(error);
4128         }
4129
4130         /*
4131          * allocate file space until done or until there is an error
4132          */
4133 retry:
4134         while (allocatesize_fsb && !error) {
4135                 /*
4136                  * determine if reserving space on
4137                  * the data or realtime partition.
4138                  */
4139                 if (rt) {
4140                         xfs_fileoff_t s, e;
4141
4142                         s = startoffset_fsb;
4143                         do_div(s, rtextsize);
4144                         s *= rtextsize;
4145                         e = roundup_64(startoffset_fsb + allocatesize_fsb,
4146                                 rtextsize);
4147                         numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4148                         datablocks = 0;
4149                 } else {
4150                         datablocks = allocatesize_fsb;
4151                         numrtextents = 0;
4152                 }
4153
4154                 /*
4155                  * allocate and setup the transaction
4156                  */
4157                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4158                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4159                 error = xfs_trans_reserve(tp,
4160                                           resblks,
4161                                           XFS_WRITE_LOG_RES(mp),
4162                                           numrtextents,
4163                                           XFS_TRANS_PERM_LOG_RES,
4164                                           XFS_WRITE_LOG_COUNT);
4165
4166                 /*
4167                  * check for running out of space
4168                  */
4169                 if (error) {
4170                         /*
4171                          * Free the transaction structure.
4172                          */
4173                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4174                         xfs_trans_cancel(tp, 0);
4175                         break;
4176                 }
4177                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4178                 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4179                                 ip->i_udquot, ip->i_gdquot, resblks, 0, 0);
4180                 if (error)
4181                         goto error1;
4182
4183                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4184                 xfs_trans_ihold(tp, ip);
4185
4186                 /*
4187                  * issue the bmapi() call to allocate the blocks
4188                  */
4189                 XFS_BMAP_INIT(&free_list, &firstfsb);
4190                 error = xfs_bmapi(tp, ip, startoffset_fsb,
4191                                   allocatesize_fsb, xfs_bmapi_flags,
4192                                   &firstfsb, 0, imapp, &reccount,
4193                                   &free_list);
4194                 if (error) {
4195                         goto error0;
4196                 }
4197
4198                 /*
4199                  * complete the transaction
4200                  */
4201                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4202                 if (error) {
4203                         goto error0;
4204                 }
4205
4206                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4207                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4208                 if (error) {
4209                         break;
4210                 }
4211
4212                 allocated_fsb = imapp->br_blockcount;
4213
4214                 if (reccount == 0) {
4215                         error = XFS_ERROR(ENOSPC);
4216                         break;
4217                 }
4218
4219                 startoffset_fsb += allocated_fsb;
4220                 allocatesize_fsb -= allocated_fsb;
4221         }
4222 dmapi_enospc_check:
4223         if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4224             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4225
4226                 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4227                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4228                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4229                                 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4230                 if (error == 0)
4231                         goto retry;     /* Maybe DMAPI app. has made space */
4232                 /* else fall through with error from XFS_SEND_DATA */
4233         }
4234
4235         return error;
4236
4237  error0:
4238         xfs_bmap_cancel(&free_list);
4239  error1:
4240         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4241         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4242         goto dmapi_enospc_check;
4243 }
4244
4245 /*
4246  * Zero file bytes between startoff and endoff inclusive.
4247  * The iolock is held exclusive and no blocks are buffered.
4248  */
4249 STATIC int
4250 xfs_zero_remaining_bytes(
4251         xfs_inode_t             *ip,
4252         xfs_off_t               startoff,
4253         xfs_off_t               endoff)
4254 {
4255         xfs_bmbt_irec_t         imap;
4256         xfs_fileoff_t           offset_fsb;
4257         xfs_off_t               lastoffset;
4258         xfs_off_t               offset;
4259         xfs_buf_t               *bp;
4260         xfs_mount_t             *mp = ip->i_mount;
4261         int                     nimap;
4262         int                     error = 0;
4263
4264         bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4265                                 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4266                                 mp->m_rtdev_targp : mp->m_ddev_targp);
4267
4268         for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4269                 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4270                 nimap = 1;
4271                 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4272                         &nimap, NULL);
4273                 if (error || nimap < 1)
4274                         break;
4275                 ASSERT(imap.br_blockcount >= 1);
4276                 ASSERT(imap.br_startoff == offset_fsb);
4277                 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4278                 if (lastoffset > endoff)
4279                         lastoffset = endoff;
4280                 if (imap.br_startblock == HOLESTARTBLOCK)
4281                         continue;
4282                 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4283                 if (imap.br_state == XFS_EXT_UNWRITTEN)
4284                         continue;
4285                 XFS_BUF_UNDONE(bp);
4286                 XFS_BUF_UNWRITE(bp);
4287                 XFS_BUF_READ(bp);
4288                 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4289                 xfsbdstrat(mp, bp);
4290                 if ((error = xfs_iowait(bp))) {
4291                         xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4292                                           mp, bp, XFS_BUF_ADDR(bp));
4293                         break;
4294                 }
4295                 memset(XFS_BUF_PTR(bp) +
4296                         (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4297                       0, lastoffset - offset + 1);
4298                 XFS_BUF_UNDONE(bp);
4299                 XFS_BUF_UNREAD(bp);
4300                 XFS_BUF_WRITE(bp);
4301                 xfsbdstrat(mp, bp);
4302                 if ((error = xfs_iowait(bp))) {
4303                         xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4304                                           mp, bp, XFS_BUF_ADDR(bp));
4305                         break;
4306                 }
4307         }
4308         xfs_buf_free(bp);
4309         return error;
4310 }
4311
4312 /*
4313  * xfs_free_file_space()
4314  *      This routine frees disk space for the given file.
4315  *
4316  *      This routine is only called by xfs_change_file_space
4317  *      for an UNRESVSP type call.
4318  *
4319  * RETURNS:
4320  *       0 on success
4321  *      errno on error
4322  *
4323  */
4324 STATIC int
4325 xfs_free_file_space(
4326         xfs_inode_t             *ip,
4327         xfs_off_t               offset,
4328         xfs_off_t               len,
4329         int                     attr_flags)
4330 {
4331         vnode_t                 *vp;
4332         int                     committed;
4333         int                     done;
4334         xfs_off_t               end_dmi_offset;
4335         xfs_fileoff_t           endoffset_fsb;
4336         int                     error;
4337         xfs_fsblock_t           firstfsb;
4338         xfs_bmap_free_t         free_list;
4339         xfs_off_t               ilen;
4340         xfs_bmbt_irec_t         imap;
4341         xfs_off_t               ioffset;
4342         xfs_extlen_t            mod=0;
4343         xfs_mount_t             *mp;
4344         int                     nimap;
4345         uint                    resblks;
4346         int                     rounding;
4347         int                     rt;
4348         xfs_fileoff_t           startoffset_fsb;
4349         xfs_trans_t             *tp;
4350         int                     need_iolock = 1;
4351
4352         vp = XFS_ITOV(ip);
4353         mp = ip->i_mount;
4354
4355         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4356
4357         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4358                 return error;
4359
4360         error = 0;
4361         if (len <= 0)   /* if nothing being freed */
4362                 return error;
4363         rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4364         startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4365         end_dmi_offset = offset + len;
4366         endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4367
4368         if (offset < ip->i_d.di_size &&
4369             (attr_flags & ATTR_DMI) == 0 &&
4370             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4371                 if (end_dmi_offset > ip->i_d.di_size)
4372                         end_dmi_offset = ip->i_d.di_size;
4373                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4374                                 offset, end_dmi_offset - offset,
4375                                 AT_DELAY_FLAG(attr_flags), NULL);
4376                 if (error)
4377                         return(error);
4378         }
4379
4380         ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4381         if (attr_flags & ATTR_NOLOCK)
4382                 need_iolock = 0;
4383         if (need_iolock)
4384                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4385
4386         rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4387                         (__uint8_t)NBPP);
4388         ilen = len + (offset & (rounding - 1));
4389         ioffset = offset & ~(rounding - 1);
4390         if (ilen & (rounding - 1))
4391                 ilen = (ilen + rounding) & ~(rounding - 1);
4392
4393         if (VN_CACHED(vp) != 0) {
4394                 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4395                                 ctooff(offtoct(ioffset)), -1);
4396                 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4397                                 -1, FI_REMAPF_LOCKED);
4398         }
4399
4400         /*
4401          * Need to zero the stuff we're not freeing, on disk.
4402          * If its a realtime file & can't use unwritten extents then we
4403          * actually need to zero the extent edges.  Otherwise xfs_bunmapi
4404          * will take care of it for us.
4405          */
4406         if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4407                 nimap = 1;
4408                 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4409                         &imap, &nimap, NULL);
4410                 if (error)
4411                         goto out_unlock_iolock;
4412                 ASSERT(nimap == 0 || nimap == 1);
4413                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4414                         xfs_daddr_t     block;
4415
4416                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4417                         block = imap.br_startblock;
4418                         mod = do_div(block, mp->m_sb.sb_rextsize);
4419                         if (mod)
4420                                 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4421                 }
4422                 nimap = 1;
4423                 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4424                         &imap, &nimap, NULL);
4425                 if (error)
4426                         goto out_unlock_iolock;
4427                 ASSERT(nimap == 0 || nimap == 1);
4428                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4429                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4430                         mod++;
4431                         if (mod && (mod != mp->m_sb.sb_rextsize))
4432                                 endoffset_fsb -= mod;
4433                 }
4434         }
4435         if ((done = (endoffset_fsb <= startoffset_fsb)))
4436                 /*
4437                  * One contiguous piece to clear
4438                  */
4439                 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4440         else {
4441                 /*
4442                  * Some full blocks, possibly two pieces to clear
4443                  */
4444                 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4445                         error = xfs_zero_remaining_bytes(ip, offset,
4446                                 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4447                 if (!error &&
4448                     XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4449                         error = xfs_zero_remaining_bytes(ip,
4450                                 XFS_FSB_TO_B(mp, endoffset_fsb),
4451                                 offset + len - 1);
4452         }
4453
4454         /*
4455          * free file space until done or until there is an error
4456          */
4457         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4458         while (!error && !done) {
4459
4460                 /*
4461                  * allocate and setup the transaction
4462                  */
4463                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4464                 error = xfs_trans_reserve(tp,
4465                                           resblks,
4466                                           XFS_WRITE_LOG_RES(mp),
4467                                           0,
4468                                           XFS_TRANS_PERM_LOG_RES,
4469                                           XFS_WRITE_LOG_COUNT);
4470
4471                 /*
4472                  * check for running out of space
4473                  */
4474                 if (error) {
4475                         /*
4476                          * Free the transaction structure.
4477                          */
4478                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4479                         xfs_trans_cancel(tp, 0);
4480                         break;
4481                 }
4482                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4483                 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4484                                 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4485                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4486                 if (error)
4487                         goto error1;
4488
4489                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4490                 xfs_trans_ihold(tp, ip);
4491
4492                 /*
4493                  * issue the bunmapi() call to free the blocks
4494                  */
4495                 XFS_BMAP_INIT(&free_list, &firstfsb);
4496                 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4497                                   endoffset_fsb - startoffset_fsb,
4498                                   0, 2, &firstfsb, &free_list, &done);
4499                 if (error) {
4500                         goto error0;
4501                 }
4502
4503                 /*
4504                  * complete the transaction
4505                  */
4506                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4507                 if (error) {
4508                         goto error0;
4509                 }
4510
4511                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4512                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4513         }
4514
4515  out_unlock_iolock:
4516         if (need_iolock)
4517                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4518         return error;
4519
4520  error0:
4521         xfs_bmap_cancel(&free_list);
4522  error1:
4523         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4524         xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4525                     XFS_ILOCK_EXCL);
4526         return error;
4527 }
4528
4529 /*
4530  * xfs_change_file_space()
4531  *      This routine allocates or frees disk space for the given file.
4532  *      The user specified parameters are checked for alignment and size
4533  *      limitations.
4534  *
4535  * RETURNS:
4536  *       0 on success
4537  *      errno on error
4538  *
4539  */
4540 int
4541 xfs_change_file_space(
4542         bhv_desc_t      *bdp,
4543         int             cmd,
4544         xfs_flock64_t   *bf,
4545         xfs_off_t       offset,
4546         cred_t          *credp,
4547         int             attr_flags)
4548 {
4549         int             clrprealloc;
4550         int             error;
4551         xfs_fsize_t     fsize;
4552         xfs_inode_t     *ip;
4553         xfs_mount_t     *mp;
4554         int             setprealloc;
4555         xfs_off_t       startoffset;
4556         xfs_off_t       llen;
4557         xfs_trans_t     *tp;
4558         vattr_t         va;
4559         vnode_t         *vp;
4560
4561         vp = BHV_TO_VNODE(bdp);
4562         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4563
4564         ip = XFS_BHVTOI(bdp);
4565         mp = ip->i_mount;
4566
4567         /*
4568          * must be a regular file and have write permission
4569          */
4570         if (vp->v_type != VREG)
4571                 return XFS_ERROR(EINVAL);
4572
4573         xfs_ilock(ip, XFS_ILOCK_SHARED);
4574
4575         if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4576                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4577                 return error;
4578         }
4579
4580         xfs_iunlock(ip, XFS_ILOCK_SHARED);
4581
4582         switch (bf->l_whence) {
4583         case 0: /*SEEK_SET*/
4584                 break;
4585         case 1: /*SEEK_CUR*/
4586                 bf->l_start += offset;
4587                 break;
4588         case 2: /*SEEK_END*/
4589                 bf->l_start += ip->i_d.di_size;
4590                 break;
4591         default:
4592                 return XFS_ERROR(EINVAL);
4593         }
4594
4595         llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4596
4597         if (   (bf->l_start < 0)
4598             || (bf->l_start > XFS_MAXIOFFSET(mp))
4599             || (bf->l_start + llen < 0)
4600             || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4601                 return XFS_ERROR(EINVAL);
4602
4603         bf->l_whence = 0;
4604
4605         startoffset = bf->l_start;
4606         fsize = ip->i_d.di_size;
4607
4608         /*
4609          * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4610          * file space.
4611          * These calls do NOT zero the data space allocated to the file,
4612          * nor do they change the file size.
4613          *
4614          * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4615          * space.
4616          * These calls cause the new file data to be zeroed and the file
4617          * size to be changed.
4618          */
4619         setprealloc = clrprealloc = 0;
4620
4621         switch (cmd) {
4622         case XFS_IOC_RESVSP:
4623         case XFS_IOC_RESVSP64:
4624                 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4625                                                                 1, attr_flags);
4626                 if (error)
4627                         return error;
4628                 setprealloc = 1;
4629                 break;
4630
4631         case XFS_IOC_UNRESVSP:
4632         case XFS_IOC_UNRESVSP64:
4633                 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4634                                                                 attr_flags)))
4635                         return error;
4636                 break;
4637
4638         case XFS_IOC_ALLOCSP:
4639         case XFS_IOC_ALLOCSP64:
4640         case XFS_IOC_FREESP:
4641         case XFS_IOC_FREESP64:
4642                 if (startoffset > fsize) {
4643                         error = xfs_alloc_file_space(ip, fsize,
4644                                         startoffset - fsize, 0, attr_flags);
4645                         if (error)
4646                                 break;
4647                 }
4648
4649                 va.va_mask = XFS_AT_SIZE;
4650                 va.va_size = startoffset;
4651
4652                 error = xfs_setattr(bdp, &va, attr_flags, credp);
4653
4654                 if (error)
4655                         return error;
4656
4657                 clrprealloc = 1;
4658                 break;
4659
4660         default:
4661                 ASSERT(0);
4662                 return XFS_ERROR(EINVAL);
4663         }
4664
4665         /*
4666          * update the inode timestamp, mode, and prealloc flag bits
4667          */
4668         tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4669
4670         if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4671                                       0, 0, 0))) {
4672                 /* ASSERT(0); */
4673                 xfs_trans_cancel(tp, 0);
4674                 return error;
4675         }
4676
4677         xfs_ilock(ip, XFS_ILOCK_EXCL);
4678
4679         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4680         xfs_trans_ihold(tp, ip);
4681
4682         if ((attr_flags & ATTR_DMI) == 0) {
4683                 ip->i_d.di_mode &= ~S_ISUID;
4684
4685                 /*
4686                  * Note that we don't have to worry about mandatory
4687                  * file locking being disabled here because we only
4688                  * clear the S_ISGID bit if the Group execute bit is
4689                  * on, but if it was on then mandatory locking wouldn't
4690                  * have been enabled.
4691                  */
4692                 if (ip->i_d.di_mode & S_IXGRP)
4693                         ip->i_d.di_mode &= ~S_ISGID;
4694
4695                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4696         }
4697         if (setprealloc)
4698                 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4699         else if (clrprealloc)
4700                 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4701
4702         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4703         xfs_trans_set_sync(tp);
4704
4705         error = xfs_trans_commit(tp, 0, NULL);
4706
4707         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4708
4709         return error;
4710 }
4711
4712 vnodeops_t xfs_vnodeops = {
4713         BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4714         .vop_open               = xfs_open,
4715         .vop_read               = xfs_read,
4716 #ifdef HAVE_SENDFILE
4717         .vop_sendfile           = xfs_sendfile,
4718 #endif
4719         .vop_write              = xfs_write,
4720         .vop_ioctl              = xfs_ioctl,
4721         .vop_getattr            = xfs_getattr,
4722         .vop_setattr            = xfs_setattr,
4723         .vop_access             = xfs_access,
4724         .vop_lookup             = xfs_lookup,
4725         .vop_create             = xfs_create,
4726         .vop_remove             = xfs_remove,
4727         .vop_link               = xfs_link,
4728         .vop_rename             = xfs_rename,
4729         .vop_mkdir              = xfs_mkdir,
4730         .vop_rmdir              = xfs_rmdir,
4731         .vop_readdir            = xfs_readdir,
4732         .vop_symlink            = xfs_symlink,
4733         .vop_readlink           = xfs_readlink,
4734         .vop_fsync              = xfs_fsync,
4735         .vop_inactive           = xfs_inactive,
4736         .vop_fid2               = xfs_fid2,
4737         .vop_rwlock             = xfs_rwlock,
4738         .vop_rwunlock           = xfs_rwunlock,
4739         .vop_bmap               = xfs_bmap,
4740         .vop_reclaim            = xfs_reclaim,
4741         .vop_attr_get           = xfs_attr_get,
4742         .vop_attr_set           = xfs_attr_set,
4743         .vop_attr_remove        = xfs_attr_remove,
4744         .vop_attr_list          = xfs_attr_list,
4745         .vop_link_removed       = (vop_link_removed_t)fs_noval,
4746         .vop_vnode_change       = (vop_vnode_change_t)fs_noval,
4747         .vop_tosspages          = fs_tosspages,
4748         .vop_flushinval_pages   = fs_flushinval_pages,
4749         .vop_flush_pages        = fs_flush_pages,
4750         .vop_release            = xfs_release,
4751         .vop_iflush             = xfs_inode_flush,
4752 };