]> CyberLeo.Net >> Repos - FreeBSD/releng/9.0.git/blob - sys/ufs/ffs/ffs_vfsops.c
Copy stable/9 to releng/9.0 as part of the FreeBSD 9.0-RELEASE release
[FreeBSD/releng/9.0.git] / sys / ufs / ffs / ffs_vfsops.c
1 /*-
2  * Copyright (c) 1989, 1991, 1993, 1994
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *      @(#)ffs_vfsops.c        8.31 (Berkeley) 5/20/95
30  */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_quota.h"
36 #include "opt_ufs.h"
37 #include "opt_ffs.h"
38 #include "opt_ddb.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54
55 #include <security/mac/mac_framework.h>
56
57 #include <ufs/ufs/extattr.h>
58 #include <ufs/ufs/gjournal.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/ufsmount.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufs_extern.h>
63
64 #include <ufs/ffs/fs.h>
65 #include <ufs/ffs/ffs_extern.h>
66
67 #include <vm/vm.h>
68 #include <vm/uma.h>
69 #include <vm/vm_page.h>
70
71 #include <geom/geom.h>
72 #include <geom/geom_vfs.h>
73
74 #include <ddb/ddb.h>
75
76 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
77
78 static int      ffs_reload(struct mount *, struct thread *);
79 static int      ffs_mountfs(struct vnode *, struct mount *, struct thread *);
80 static void     ffs_oldfscompat_read(struct fs *, struct ufsmount *,
81                     ufs2_daddr_t);
82 static void     ffs_ifree(struct ufsmount *ump, struct inode *ip);
83 static vfs_init_t ffs_init;
84 static vfs_uninit_t ffs_uninit;
85 static vfs_extattrctl_t ffs_extattrctl;
86 static vfs_cmount_t ffs_cmount;
87 static vfs_unmount_t ffs_unmount;
88 static vfs_mount_t ffs_mount;
89 static vfs_statfs_t ffs_statfs;
90 static vfs_fhtovp_t ffs_fhtovp;
91 static vfs_sync_t ffs_sync;
92
93 static struct vfsops ufs_vfsops = {
94         .vfs_extattrctl =       ffs_extattrctl,
95         .vfs_fhtovp =           ffs_fhtovp,
96         .vfs_init =             ffs_init,
97         .vfs_mount =            ffs_mount,
98         .vfs_cmount =           ffs_cmount,
99         .vfs_quotactl =         ufs_quotactl,
100         .vfs_root =             ufs_root,
101         .vfs_statfs =           ffs_statfs,
102         .vfs_sync =             ffs_sync,
103         .vfs_uninit =           ffs_uninit,
104         .vfs_unmount =          ffs_unmount,
105         .vfs_vget =             ffs_vget,
106         .vfs_susp_clean =       process_deferred_inactive,
107 };
108
109 VFS_SET(ufs_vfsops, ufs, 0);
110 MODULE_VERSION(ufs, 1);
111
112 static b_strategy_t ffs_geom_strategy;
113 static b_write_t ffs_bufwrite;
114
115 static struct buf_ops ffs_ops = {
116         .bop_name =     "FFS",
117         .bop_write =    ffs_bufwrite,
118         .bop_strategy = ffs_geom_strategy,
119         .bop_sync =     bufsync,
120 #ifdef NO_FFS_SNAPSHOT
121         .bop_bdflush =  bufbdflush,
122 #else
123         .bop_bdflush =  ffs_bdflush,
124 #endif
125 };
126
127 /*
128  * Note that userquota and groupquota options are not currently used
129  * by UFS/FFS code and generally mount(8) does not pass those options
130  * from userland, but they can be passed by loader(8) via
131  * vfs.root.mountfrom.options.
132  */
133 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
134     "noclusterw", "noexec", "export", "force", "from", "groupquota",
135     "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
136     "nosymfollow", "sync", "union", "userquota", NULL };
137
138 static int
139 ffs_mount(struct mount *mp)
140 {
141         struct vnode *devvp;
142         struct thread *td;
143         struct ufsmount *ump = 0;
144         struct fs *fs;
145         pid_t fsckpid = 0;
146         int error, flags;
147         u_int mntorflags;
148         accmode_t accmode;
149         struct nameidata ndp;
150         char *fspec;
151
152         td = curthread;
153         if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
154                 return (EINVAL);
155         if (uma_inode == NULL) {
156                 uma_inode = uma_zcreate("FFS inode",
157                     sizeof(struct inode), NULL, NULL, NULL, NULL,
158                     UMA_ALIGN_PTR, 0);
159                 uma_ufs1 = uma_zcreate("FFS1 dinode",
160                     sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
161                     UMA_ALIGN_PTR, 0);
162                 uma_ufs2 = uma_zcreate("FFS2 dinode",
163                     sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
164                     UMA_ALIGN_PTR, 0);
165         }
166
167         vfs_deleteopt(mp->mnt_optnew, "groupquota");
168         vfs_deleteopt(mp->mnt_optnew, "userquota");
169
170         fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
171         if (error)
172                 return (error);
173
174         mntorflags = 0;
175         if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
176                 mntorflags |= MNT_ACLS;
177
178         if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
179                 mntorflags |= MNT_SNAPSHOT;
180                 /*
181                  * Once we have set the MNT_SNAPSHOT flag, do not
182                  * persist "snapshot" in the options list.
183                  */
184                 vfs_deleteopt(mp->mnt_optnew, "snapshot");
185                 vfs_deleteopt(mp->mnt_opt, "snapshot");
186         }
187
188         if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
189             vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
190                 /*
191                  * Once we have set the restricted PID, do not
192                  * persist "fsckpid" in the options list.
193                  */
194                 vfs_deleteopt(mp->mnt_optnew, "fsckpid");
195                 vfs_deleteopt(mp->mnt_opt, "fsckpid");
196                 if (mp->mnt_flag & MNT_UPDATE) {
197                         if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
198                              vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
199                                 printf("Checker enable: Must be read-only\n");
200                                 return (EINVAL);
201                         }
202                 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
203                         printf("Checker enable: Must be read-only\n");
204                         return (EINVAL);
205                 }
206                 /* Set to -1 if we are done */
207                 if (fsckpid == 0)
208                         fsckpid = -1;
209         }
210
211         if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
212                 if (mntorflags & MNT_ACLS) {
213                         printf("WARNING: \"acls\" and \"nfsv4acls\" "
214                             "options are mutually exclusive\n");
215                         return (EINVAL);
216                 }
217                 mntorflags |= MNT_NFS4ACLS;
218         }
219
220         MNT_ILOCK(mp);
221         mp->mnt_flag |= mntorflags;
222         MNT_IUNLOCK(mp);
223         /*
224          * If updating, check whether changing from read-only to
225          * read/write; if there is no device name, that's all we do.
226          */
227         if (mp->mnt_flag & MNT_UPDATE) {
228                 ump = VFSTOUFS(mp);
229                 fs = ump->um_fs;
230                 devvp = ump->um_devvp;
231                 if (fsckpid == -1 && ump->um_fsckpid > 0) {
232                         if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
233                             (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
234                                 return (error);
235                         DROP_GIANT();
236                         g_topology_lock();
237                         /*
238                          * Return to normal read-only mode.
239                          */
240                         error = g_access(ump->um_cp, 0, -1, 0);
241                         g_topology_unlock();
242                         PICKUP_GIANT();
243                         ump->um_fsckpid = 0;
244                 }
245                 if (fs->fs_ronly == 0 &&
246                     vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
247                         /*
248                          * Flush any dirty data and suspend filesystem.
249                          */
250                         if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
251                                 return (error);
252                         for (;;) {
253                                 vn_finished_write(mp);
254                                 if ((error = vfs_write_suspend(mp)) != 0)
255                                         return (error);
256                                 MNT_ILOCK(mp);
257                                 if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
258                                         /*
259                                          * Allow the secondary writes
260                                          * to proceed.
261                                          */
262                                         mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
263                                             MNTK_SUSPEND2);
264                                         wakeup(&mp->mnt_flag);
265                                         MNT_IUNLOCK(mp);
266                                         /*
267                                          * Allow the curthread to
268                                          * ignore the suspension to
269                                          * synchronize on-disk state.
270                                          */
271                                         td->td_pflags |= TDP_IGNSUSP;
272                                         break;
273                                 }
274                                 MNT_IUNLOCK(mp);
275                                 vn_start_write(NULL, &mp, V_WAIT);
276                         }
277                         /*
278                          * Check for and optionally get rid of files open
279                          * for writing.
280                          */
281                         flags = WRITECLOSE;
282                         if (mp->mnt_flag & MNT_FORCE)
283                                 flags |= FORCECLOSE;
284                         if (MOUNTEDSOFTDEP(mp)) {
285                                 error = softdep_flushfiles(mp, flags, td);
286                         } else {
287                                 error = ffs_flushfiles(mp, flags, td);
288                         }
289                         if (error) {
290                                 vfs_write_resume(mp);
291                                 return (error);
292                         }
293                         if (fs->fs_pendingblocks != 0 ||
294                             fs->fs_pendinginodes != 0) {
295                                 printf("%s: %s: blocks %jd files %d\n",
296                                     fs->fs_fsmnt, "update error",
297                                     (intmax_t)fs->fs_pendingblocks,
298                                     fs->fs_pendinginodes);
299                                 fs->fs_pendingblocks = 0;
300                                 fs->fs_pendinginodes = 0;
301                         }
302                         if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
303                                 fs->fs_clean = 1;
304                         if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
305                                 fs->fs_ronly = 0;
306                                 fs->fs_clean = 0;
307                                 vfs_write_resume(mp);
308                                 return (error);
309                         }
310                         if (MOUNTEDSOFTDEP(mp))
311                                 softdep_unmount(mp);
312                         DROP_GIANT();
313                         g_topology_lock();
314                         /*
315                          * Drop our write and exclusive access.
316                          */
317                         g_access(ump->um_cp, 0, -1, -1);
318                         g_topology_unlock();
319                         PICKUP_GIANT();
320                         fs->fs_ronly = 1;
321                         MNT_ILOCK(mp);
322                         mp->mnt_flag |= MNT_RDONLY;
323                         MNT_IUNLOCK(mp);
324                         /*
325                          * Allow the writers to note that filesystem
326                          * is ro now.
327                          */
328                         vfs_write_resume(mp);
329                 }
330                 if ((mp->mnt_flag & MNT_RELOAD) &&
331                     (error = ffs_reload(mp, td)) != 0)
332                         return (error);
333                 if (fs->fs_ronly &&
334                     !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
335                         /*
336                          * If we are running a checker, do not allow upgrade.
337                          */
338                         if (ump->um_fsckpid > 0) {
339                                 printf("Active checker, cannot rw upgrade\n");
340                                 return (EINVAL);
341                         }
342                         /*
343                          * If upgrade to read-write by non-root, then verify
344                          * that user has necessary permissions on the device.
345                          */
346                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
347                         error = VOP_ACCESS(devvp, VREAD | VWRITE,
348                             td->td_ucred, td);
349                         if (error)
350                                 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
351                         if (error) {
352                                 VOP_UNLOCK(devvp, 0);
353                                 return (error);
354                         }
355                         VOP_UNLOCK(devvp, 0);
356                         fs->fs_flags &= ~FS_UNCLEAN;
357                         if (fs->fs_clean == 0) {
358                                 fs->fs_flags |= FS_UNCLEAN;
359                                 if ((mp->mnt_flag & MNT_FORCE) ||
360                                     ((fs->fs_flags &
361                                      (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
362                                      (fs->fs_flags & FS_DOSOFTDEP))) {
363                                         printf("WARNING: %s was not %s\n",
364                                            fs->fs_fsmnt, "properly dismounted");
365                                 } else {
366                                         printf(
367 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
368                                             fs->fs_fsmnt);
369                                         if (fs->fs_flags & FS_SUJ)
370                                                 printf(
371 "WARNING: Forced mount will invalidate journal contents\n");
372                                         return (EPERM);
373                                 }
374                         }
375                         DROP_GIANT();
376                         g_topology_lock();
377                         /*
378                          * Request exclusive write access.
379                          */
380                         error = g_access(ump->um_cp, 0, 1, 1);
381                         g_topology_unlock();
382                         PICKUP_GIANT();
383                         if (error)
384                                 return (error);
385                         if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
386                                 return (error);
387                         fs->fs_ronly = 0;
388                         MNT_ILOCK(mp);
389                         mp->mnt_flag &= ~MNT_RDONLY;
390                         MNT_IUNLOCK(mp);
391                         fs->fs_mtime = time_second;
392                         /* check to see if we need to start softdep */
393                         if ((fs->fs_flags & FS_DOSOFTDEP) &&
394                             (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
395                                 vn_finished_write(mp);
396                                 return (error);
397                         }
398                         fs->fs_clean = 0;
399                         if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
400                                 vn_finished_write(mp);
401                                 return (error);
402                         }
403                         if (fs->fs_snapinum[0] != 0)
404                                 ffs_snapshot_mount(mp);
405                         vn_finished_write(mp);
406                 }
407                 /*
408                  * Soft updates is incompatible with "async",
409                  * so if we are doing softupdates stop the user
410                  * from setting the async flag in an update.
411                  * Softdep_mount() clears it in an initial mount
412                  * or ro->rw remount.
413                  */
414                 if (MOUNTEDSOFTDEP(mp)) {
415                         /* XXX: Reset too late ? */
416                         MNT_ILOCK(mp);
417                         mp->mnt_flag &= ~MNT_ASYNC;
418                         MNT_IUNLOCK(mp);
419                 }
420                 /*
421                  * Keep MNT_ACLS flag if it is stored in superblock.
422                  */
423                 if ((fs->fs_flags & FS_ACLS) != 0) {
424                         /* XXX: Set too late ? */
425                         MNT_ILOCK(mp);
426                         mp->mnt_flag |= MNT_ACLS;
427                         MNT_IUNLOCK(mp);
428                 }
429
430                 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
431                         /* XXX: Set too late ? */
432                         MNT_ILOCK(mp);
433                         mp->mnt_flag |= MNT_NFS4ACLS;
434                         MNT_IUNLOCK(mp);
435                 }
436                 /*
437                  * If this is a request from fsck to clean up the filesystem,
438                  * then allow the specified pid to proceed.
439                  */
440                 if (fsckpid > 0) {
441                         if (ump->um_fsckpid != 0) {
442                                 printf("Active checker already running on %s\n",
443                                     fs->fs_fsmnt);
444                                 return (EINVAL);
445                         }
446                         KASSERT(MOUNTEDSOFTDEP(mp) == 0,
447                             ("soft updates enabled on read-only file system"));
448                         DROP_GIANT();
449                         g_topology_lock();
450                         /*
451                          * Request write access.
452                          */
453                         error = g_access(ump->um_cp, 0, 1, 0);
454                         g_topology_unlock();
455                         PICKUP_GIANT();
456                         if (error) {
457                                 printf("Checker activation failed on %s\n",
458                                     fs->fs_fsmnt);
459                                 return (error);
460                         }
461                         ump->um_fsckpid = fsckpid;
462                         if (fs->fs_snapinum[0] != 0)
463                                 ffs_snapshot_mount(mp);
464                         fs->fs_mtime = time_second;
465                         fs->fs_fmod = 1;
466                         fs->fs_clean = 0;
467                         (void) ffs_sbupdate(ump, MNT_WAIT, 0);
468                 }
469
470                 /*
471                  * If this is a snapshot request, take the snapshot.
472                  */
473                 if (mp->mnt_flag & MNT_SNAPSHOT)
474                         return (ffs_snapshot(mp, fspec));
475         }
476
477         /*
478          * Not an update, or updating the name: look up the name
479          * and verify that it refers to a sensible disk device.
480          */
481         NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
482         if ((error = namei(&ndp)) != 0)
483                 return (error);
484         NDFREE(&ndp, NDF_ONLY_PNBUF);
485         devvp = ndp.ni_vp;
486         if (!vn_isdisk(devvp, &error)) {
487                 vput(devvp);
488                 return (error);
489         }
490
491         /*
492          * If mount by non-root, then verify that user has necessary
493          * permissions on the device.
494          */
495         accmode = VREAD;
496         if ((mp->mnt_flag & MNT_RDONLY) == 0)
497                 accmode |= VWRITE;
498         error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
499         if (error)
500                 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
501         if (error) {
502                 vput(devvp);
503                 return (error);
504         }
505
506         if (mp->mnt_flag & MNT_UPDATE) {
507                 /*
508                  * Update only
509                  *
510                  * If it's not the same vnode, or at least the same device
511                  * then it's not correct.
512                  */
513
514                 if (devvp->v_rdev != ump->um_devvp->v_rdev)
515                         error = EINVAL; /* needs translation */
516                 vput(devvp);
517                 if (error)
518                         return (error);
519         } else {
520                 /*
521                  * New mount
522                  *
523                  * We need the name for the mount point (also used for
524                  * "last mounted on") copied in. If an error occurs,
525                  * the mount point is discarded by the upper level code.
526                  * Note that vfs_mount() populates f_mntonname for us.
527                  */
528                 if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
529                         vrele(devvp);
530                         return (error);
531                 }
532                 if (fsckpid > 0) {
533                         KASSERT(MOUNTEDSOFTDEP(mp) == 0,
534                             ("soft updates enabled on read-only file system"));
535                         ump = VFSTOUFS(mp);
536                         fs = ump->um_fs;
537                         DROP_GIANT();
538                         g_topology_lock();
539                         /*
540                          * Request write access.
541                          */
542                         error = g_access(ump->um_cp, 0, 1, 0);
543                         g_topology_unlock();
544                         PICKUP_GIANT();
545                         if (error) {
546                                 printf("Checker activation failed on %s\n",
547                                     fs->fs_fsmnt);
548                         } else { 
549                                 ump->um_fsckpid = fsckpid;
550                                 if (fs->fs_snapinum[0] != 0)
551                                         ffs_snapshot_mount(mp);
552                                 fs->fs_mtime = time_second;
553                                 fs->fs_clean = 0;
554                                 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
555                         }
556                 }
557         }
558         vfs_mountedfrom(mp, fspec);
559         return (0);
560 }
561
562 /*
563  * Compatibility with old mount system call.
564  */
565
566 static int
567 ffs_cmount(struct mntarg *ma, void *data, int flags)
568 {
569         struct ufs_args args;
570         struct export_args exp;
571         int error;
572
573         if (data == NULL)
574                 return (EINVAL);
575         error = copyin(data, &args, sizeof args);
576         if (error)
577                 return (error);
578         vfs_oexport_conv(&args.export, &exp);
579
580         ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
581         ma = mount_arg(ma, "export", &exp, sizeof(exp));
582         error = kernel_mount(ma, flags);
583
584         return (error);
585 }
586
587 /*
588  * Reload all incore data for a filesystem (used after running fsck on
589  * the root filesystem and finding things to fix). The filesystem must
590  * be mounted read-only.
591  *
592  * Things to do to update the mount:
593  *      1) invalidate all cached meta-data.
594  *      2) re-read superblock from disk.
595  *      3) re-read summary information from disk.
596  *      4) invalidate all inactive vnodes.
597  *      5) invalidate all cached file data.
598  *      6) re-read inode data for all active vnodes.
599  */
600 static int
601 ffs_reload(struct mount *mp, struct thread *td)
602 {
603         struct vnode *vp, *mvp, *devvp;
604         struct inode *ip;
605         void *space;
606         struct buf *bp;
607         struct fs *fs, *newfs;
608         struct ufsmount *ump;
609         ufs2_daddr_t sblockloc;
610         int i, blks, size, error;
611         int32_t *lp;
612
613         if ((mp->mnt_flag & MNT_RDONLY) == 0)
614                 return (EINVAL);
615         ump = VFSTOUFS(mp);
616         /*
617          * Step 1: invalidate all cached meta-data.
618          */
619         devvp = VFSTOUFS(mp)->um_devvp;
620         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
621         if (vinvalbuf(devvp, 0, 0, 0) != 0)
622                 panic("ffs_reload: dirty1");
623         VOP_UNLOCK(devvp, 0);
624
625         /*
626          * Step 2: re-read superblock from disk.
627          */
628         fs = VFSTOUFS(mp)->um_fs;
629         if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
630             NOCRED, &bp)) != 0)
631                 return (error);
632         newfs = (struct fs *)bp->b_data;
633         if ((newfs->fs_magic != FS_UFS1_MAGIC &&
634              newfs->fs_magic != FS_UFS2_MAGIC) ||
635             newfs->fs_bsize > MAXBSIZE ||
636             newfs->fs_bsize < sizeof(struct fs)) {
637                         brelse(bp);
638                         return (EIO);           /* XXX needs translation */
639         }
640         /*
641          * Copy pointer fields back into superblock before copying in   XXX
642          * new superblock. These should really be in the ufsmount.      XXX
643          * Note that important parameters (eg fs_ncg) are unchanged.
644          */
645         newfs->fs_csp = fs->fs_csp;
646         newfs->fs_maxcluster = fs->fs_maxcluster;
647         newfs->fs_contigdirs = fs->fs_contigdirs;
648         newfs->fs_active = fs->fs_active;
649         /* The file system is still read-only. */
650         newfs->fs_ronly = 1;
651         sblockloc = fs->fs_sblockloc;
652         bcopy(newfs, fs, (u_int)fs->fs_sbsize);
653         brelse(bp);
654         mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
655         ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
656         UFS_LOCK(ump);
657         if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
658                 printf("%s: reload pending error: blocks %jd files %d\n",
659                     fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
660                     fs->fs_pendinginodes);
661                 fs->fs_pendingblocks = 0;
662                 fs->fs_pendinginodes = 0;
663         }
664         UFS_UNLOCK(ump);
665
666         /*
667          * Step 3: re-read summary information from disk.
668          */
669         blks = howmany(fs->fs_cssize, fs->fs_fsize);
670         space = fs->fs_csp;
671         for (i = 0; i < blks; i += fs->fs_frag) {
672                 size = fs->fs_bsize;
673                 if (i + fs->fs_frag > blks)
674                         size = (blks - i) * fs->fs_fsize;
675                 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
676                     NOCRED, &bp);
677                 if (error)
678                         return (error);
679                 bcopy(bp->b_data, space, (u_int)size);
680                 space = (char *)space + size;
681                 brelse(bp);
682         }
683         /*
684          * We no longer know anything about clusters per cylinder group.
685          */
686         if (fs->fs_contigsumsize > 0) {
687                 lp = fs->fs_maxcluster;
688                 for (i = 0; i < fs->fs_ncg; i++)
689                         *lp++ = fs->fs_contigsumsize;
690         }
691
692 loop:
693         MNT_ILOCK(mp);
694         MNT_VNODE_FOREACH(vp, mp, mvp) {
695                 VI_LOCK(vp);
696                 if (vp->v_iflag & VI_DOOMED) {
697                         VI_UNLOCK(vp);
698                         continue;
699                 }
700                 MNT_IUNLOCK(mp);
701                 /*
702                  * Step 4: invalidate all cached file data.
703                  */
704                 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
705                         MNT_VNODE_FOREACH_ABORT(mp, mvp);
706                         goto loop;
707                 }
708                 if (vinvalbuf(vp, 0, 0, 0))
709                         panic("ffs_reload: dirty2");
710                 /*
711                  * Step 5: re-read inode data for all active vnodes.
712                  */
713                 ip = VTOI(vp);
714                 error =
715                     bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
716                     (int)fs->fs_bsize, NOCRED, &bp);
717                 if (error) {
718                         VOP_UNLOCK(vp, 0);
719                         vrele(vp);
720                         MNT_VNODE_FOREACH_ABORT(mp, mvp);
721                         return (error);
722                 }
723                 ffs_load_inode(bp, ip, fs, ip->i_number);
724                 ip->i_effnlink = ip->i_nlink;
725                 brelse(bp);
726                 VOP_UNLOCK(vp, 0);
727                 vrele(vp);
728                 MNT_ILOCK(mp);
729         }
730         MNT_IUNLOCK(mp);
731         return (0);
732 }
733
734 /*
735  * Possible superblock locations ordered from most to least likely.
736  */
737 static int sblock_try[] = SBLOCKSEARCH;
738
739 /*
740  * Common code for mount and mountroot
741  */
742 static int
743 ffs_mountfs(devvp, mp, td)
744         struct vnode *devvp;
745         struct mount *mp;
746         struct thread *td;
747 {
748         struct ufsmount *ump;
749         struct buf *bp;
750         struct fs *fs;
751         struct cdev *dev;
752         void *space;
753         ufs2_daddr_t sblockloc;
754         int error, i, blks, size, ronly;
755         int32_t *lp;
756         struct ucred *cred;
757         struct g_consumer *cp;
758         struct mount *nmp;
759
760         bp = NULL;
761         ump = NULL;
762         cred = td ? td->td_ucred : NOCRED;
763         ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
764
765         dev = devvp->v_rdev;
766         dev_ref(dev);
767         DROP_GIANT();
768         g_topology_lock();
769         error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
770         g_topology_unlock();
771         PICKUP_GIANT();
772         VOP_UNLOCK(devvp, 0);
773         if (error)
774                 goto out;
775         if (devvp->v_rdev->si_iosize_max != 0)
776                 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
777         if (mp->mnt_iosize_max > MAXPHYS)
778                 mp->mnt_iosize_max = MAXPHYS;
779
780         devvp->v_bufobj.bo_ops = &ffs_ops;
781
782         fs = NULL;
783         sblockloc = 0;
784         /*
785          * Try reading the superblock in each of its possible locations.
786          */
787         for (i = 0; sblock_try[i] != -1; i++) {
788                 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
789                         error = EINVAL;
790                         vfs_mount_error(mp,
791                             "Invalid sectorsize %d for superblock size %d",
792                             cp->provider->sectorsize, SBLOCKSIZE);
793                         goto out;
794                 }
795                 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
796                     cred, &bp)) != 0)
797                         goto out;
798                 fs = (struct fs *)bp->b_data;
799                 sblockloc = sblock_try[i];
800                 if ((fs->fs_magic == FS_UFS1_MAGIC ||
801                      (fs->fs_magic == FS_UFS2_MAGIC &&
802                       (fs->fs_sblockloc == sblockloc ||
803                        (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
804                     fs->fs_bsize <= MAXBSIZE &&
805                     fs->fs_bsize >= sizeof(struct fs))
806                         break;
807                 brelse(bp);
808                 bp = NULL;
809         }
810         if (sblock_try[i] == -1) {
811                 error = EINVAL;         /* XXX needs translation */
812                 goto out;
813         }
814         fs->fs_fmod = 0;
815         fs->fs_flags &= ~FS_INDEXDIRS;  /* no support for directory indicies */
816         fs->fs_flags &= ~FS_UNCLEAN;
817         if (fs->fs_clean == 0) {
818                 fs->fs_flags |= FS_UNCLEAN;
819                 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
820                     ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
821                      (fs->fs_flags & FS_DOSOFTDEP))) {
822                         printf("WARNING: %s was not properly dismounted\n",
823                             fs->fs_fsmnt);
824                 } else {
825                         printf(
826 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
827                             fs->fs_fsmnt);
828                         if (fs->fs_flags & FS_SUJ)
829                                 printf(
830 "WARNING: Forced mount will invalidate journal contents\n");
831                         error = EPERM;
832                         goto out;
833                 }
834                 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
835                     (mp->mnt_flag & MNT_FORCE)) {
836                         printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
837                             (intmax_t)fs->fs_pendingblocks,
838                             fs->fs_pendinginodes);
839                         fs->fs_pendingblocks = 0;
840                         fs->fs_pendinginodes = 0;
841                 }
842         }
843         if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
844                 printf("%s: mount pending error: blocks %jd files %d\n",
845                     fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
846                     fs->fs_pendinginodes);
847                 fs->fs_pendingblocks = 0;
848                 fs->fs_pendinginodes = 0;
849         }
850         if ((fs->fs_flags & FS_GJOURNAL) != 0) {
851 #ifdef UFS_GJOURNAL
852                 /*
853                  * Get journal provider name.
854                  */
855                 size = 1024;
856                 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
857                 if (g_io_getattr("GJOURNAL::provider", cp, &size,
858                     mp->mnt_gjprovider) == 0) {
859                         mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
860                             M_UFSMNT, M_WAITOK);
861                         MNT_ILOCK(mp);
862                         mp->mnt_flag |= MNT_GJOURNAL;
863                         MNT_IUNLOCK(mp);
864                 } else {
865                         printf(
866 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n",
867                             mp->mnt_stat.f_mntonname);
868                         free(mp->mnt_gjprovider, M_UFSMNT);
869                         mp->mnt_gjprovider = NULL;
870                 }
871 #else
872                 printf(
873 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n",
874                     mp->mnt_stat.f_mntonname);
875 #endif
876         } else {
877                 mp->mnt_gjprovider = NULL;
878         }
879         ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
880         ump->um_cp = cp;
881         ump->um_bo = &devvp->v_bufobj;
882         ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
883         if (fs->fs_magic == FS_UFS1_MAGIC) {
884                 ump->um_fstype = UFS1;
885                 ump->um_balloc = ffs_balloc_ufs1;
886         } else {
887                 ump->um_fstype = UFS2;
888                 ump->um_balloc = ffs_balloc_ufs2;
889         }
890         ump->um_blkatoff = ffs_blkatoff;
891         ump->um_truncate = ffs_truncate;
892         ump->um_update = ffs_update;
893         ump->um_valloc = ffs_valloc;
894         ump->um_vfree = ffs_vfree;
895         ump->um_ifree = ffs_ifree;
896         ump->um_rdonly = ffs_rdonly;
897         ump->um_snapgone = ffs_snapgone;
898         mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
899         bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
900         if (fs->fs_sbsize < SBLOCKSIZE)
901                 bp->b_flags |= B_INVAL | B_NOCACHE;
902         brelse(bp);
903         bp = NULL;
904         fs = ump->um_fs;
905         ffs_oldfscompat_read(fs, ump, sblockloc);
906         fs->fs_ronly = ronly;
907         size = fs->fs_cssize;
908         blks = howmany(size, fs->fs_fsize);
909         if (fs->fs_contigsumsize > 0)
910                 size += fs->fs_ncg * sizeof(int32_t);
911         size += fs->fs_ncg * sizeof(u_int8_t);
912         space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
913         fs->fs_csp = space;
914         for (i = 0; i < blks; i += fs->fs_frag) {
915                 size = fs->fs_bsize;
916                 if (i + fs->fs_frag > blks)
917                         size = (blks - i) * fs->fs_fsize;
918                 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
919                     cred, &bp)) != 0) {
920                         free(fs->fs_csp, M_UFSMNT);
921                         goto out;
922                 }
923                 bcopy(bp->b_data, space, (u_int)size);
924                 space = (char *)space + size;
925                 brelse(bp);
926                 bp = NULL;
927         }
928         if (fs->fs_contigsumsize > 0) {
929                 fs->fs_maxcluster = lp = space;
930                 for (i = 0; i < fs->fs_ncg; i++)
931                         *lp++ = fs->fs_contigsumsize;
932                 space = lp;
933         }
934         size = fs->fs_ncg * sizeof(u_int8_t);
935         fs->fs_contigdirs = (u_int8_t *)space;
936         bzero(fs->fs_contigdirs, size);
937         fs->fs_active = NULL;
938         mp->mnt_data = ump;
939         mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
940         mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
941         nmp = NULL;
942         if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
943             (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
944                 if (nmp)
945                         vfs_rel(nmp);
946                 vfs_getnewfsid(mp);
947         }
948         mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
949         MNT_ILOCK(mp);
950         mp->mnt_flag |= MNT_LOCAL;
951         MNT_IUNLOCK(mp);
952         if ((fs->fs_flags & FS_MULTILABEL) != 0) {
953 #ifdef MAC
954                 MNT_ILOCK(mp);
955                 mp->mnt_flag |= MNT_MULTILABEL;
956                 MNT_IUNLOCK(mp);
957 #else
958                 printf(
959 "WARNING: %s: multilabel flag on fs but no MAC support\n",
960                     mp->mnt_stat.f_mntonname);
961 #endif
962         }
963         if ((fs->fs_flags & FS_ACLS) != 0) {
964 #ifdef UFS_ACL
965                 MNT_ILOCK(mp);
966
967                 if (mp->mnt_flag & MNT_NFS4ACLS)
968                         printf("WARNING: ACLs flag on fs conflicts with "
969                             "\"nfsv4acls\" mount option; option ignored\n");
970                 mp->mnt_flag &= ~MNT_NFS4ACLS;
971                 mp->mnt_flag |= MNT_ACLS;
972
973                 MNT_IUNLOCK(mp);
974 #else
975                 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
976                     mp->mnt_stat.f_mntonname);
977 #endif
978         }
979         if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
980 #ifdef UFS_ACL
981                 MNT_ILOCK(mp);
982
983                 if (mp->mnt_flag & MNT_ACLS)
984                         printf("WARNING: NFSv4 ACLs flag on fs conflicts with "
985                             "\"acls\" mount option; option ignored\n");
986                 mp->mnt_flag &= ~MNT_ACLS;
987                 mp->mnt_flag |= MNT_NFS4ACLS;
988
989                 MNT_IUNLOCK(mp);
990 #else
991                 printf(
992 "WARNING: %s: NFSv4 ACLs flag on fs but no ACLs support\n",
993                     mp->mnt_stat.f_mntonname);
994 #endif
995         }
996         if ((fs->fs_flags & FS_TRIM) != 0) {
997                 size = sizeof(int);
998                 if (g_io_getattr("GEOM::candelete", cp, &size,
999                     &ump->um_candelete) == 0) {
1000                         if (!ump->um_candelete)
1001                                 printf(
1002 "WARNING: %s: TRIM flag on fs but disk does not support TRIM\n",
1003                                     mp->mnt_stat.f_mntonname);
1004                 } else {
1005                         printf(
1006 "WARNING: %s: TRIM flag on fs but cannot get whether disk supports TRIM\n",
1007                             mp->mnt_stat.f_mntonname);
1008                         ump->um_candelete = 0;
1009                 }
1010         }
1011
1012         ump->um_mountp = mp;
1013         ump->um_dev = dev;
1014         ump->um_devvp = devvp;
1015         ump->um_nindir = fs->fs_nindir;
1016         ump->um_bptrtodb = fs->fs_fsbtodb;
1017         ump->um_seqinc = fs->fs_frag;
1018         for (i = 0; i < MAXQUOTAS; i++)
1019                 ump->um_quotas[i] = NULLVP;
1020 #ifdef UFS_EXTATTR
1021         ufs_extattr_uepm_init(&ump->um_extattr);
1022 #endif
1023         /*
1024          * Set FS local "last mounted on" information (NULL pad)
1025          */
1026         bzero(fs->fs_fsmnt, MAXMNTLEN);
1027         strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1028         mp->mnt_stat.f_iosize = fs->fs_bsize;
1029
1030         if (mp->mnt_flag & MNT_ROOTFS) {
1031                 /*
1032                  * Root mount; update timestamp in mount structure.
1033                  * this will be used by the common root mount code
1034                  * to update the system clock.
1035                  */
1036                 mp->mnt_time = fs->fs_time;
1037         }
1038
1039         if (ronly == 0) {
1040                 fs->fs_mtime = time_second;
1041                 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1042                     (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1043                         free(fs->fs_csp, M_UFSMNT);
1044                         ffs_flushfiles(mp, FORCECLOSE, td);
1045                         goto out;
1046                 }
1047                 if (fs->fs_snapinum[0] != 0)
1048                         ffs_snapshot_mount(mp);
1049                 fs->fs_fmod = 1;
1050                 fs->fs_clean = 0;
1051                 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1052         }
1053         /*
1054          * Initialize filesystem stat information in mount struct.
1055          */
1056         MNT_ILOCK(mp);
1057         mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED |
1058             MNTK_EXTENDED_SHARED;
1059         MNT_IUNLOCK(mp);
1060 #ifdef UFS_EXTATTR
1061 #ifdef UFS_EXTATTR_AUTOSTART
1062         /*
1063          *
1064          * Auto-starting does the following:
1065          *      - check for /.attribute in the fs, and extattr_start if so
1066          *      - for each file in .attribute, enable that file with
1067          *        an attribute of the same name.
1068          * Not clear how to report errors -- probably eat them.
1069          * This would all happen while the filesystem was busy/not
1070          * available, so would effectively be "atomic".
1071          */
1072         (void) ufs_extattr_autostart(mp, td);
1073 #endif /* !UFS_EXTATTR_AUTOSTART */
1074 #endif /* !UFS_EXTATTR */
1075         return (0);
1076 out:
1077         if (bp)
1078                 brelse(bp);
1079         if (cp != NULL) {
1080                 DROP_GIANT();
1081                 g_topology_lock();
1082                 g_vfs_close(cp);
1083                 g_topology_unlock();
1084                 PICKUP_GIANT();
1085         }
1086         if (ump) {
1087                 mtx_destroy(UFS_MTX(ump));
1088                 if (mp->mnt_gjprovider != NULL) {
1089                         free(mp->mnt_gjprovider, M_UFSMNT);
1090                         mp->mnt_gjprovider = NULL;
1091                 }
1092                 free(ump->um_fs, M_UFSMNT);
1093                 free(ump, M_UFSMNT);
1094                 mp->mnt_data = NULL;
1095         }
1096         dev_rel(dev);
1097         return (error);
1098 }
1099
1100 #include <sys/sysctl.h>
1101 static int bigcgs = 0;
1102 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1103
1104 /*
1105  * Sanity checks for loading old filesystem superblocks.
1106  * See ffs_oldfscompat_write below for unwound actions.
1107  *
1108  * XXX - Parts get retired eventually.
1109  * Unfortunately new bits get added.
1110  */
1111 static void
1112 ffs_oldfscompat_read(fs, ump, sblockloc)
1113         struct fs *fs;
1114         struct ufsmount *ump;
1115         ufs2_daddr_t sblockloc;
1116 {
1117         off_t maxfilesize;
1118
1119         /*
1120          * If not yet done, update fs_flags location and value of fs_sblockloc.
1121          */
1122         if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1123                 fs->fs_flags = fs->fs_old_flags;
1124                 fs->fs_old_flags |= FS_FLAGS_UPDATED;
1125                 fs->fs_sblockloc = sblockloc;
1126         }
1127         /*
1128          * If not yet done, update UFS1 superblock with new wider fields.
1129          */
1130         if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1131                 fs->fs_maxbsize = fs->fs_bsize;
1132                 fs->fs_time = fs->fs_old_time;
1133                 fs->fs_size = fs->fs_old_size;
1134                 fs->fs_dsize = fs->fs_old_dsize;
1135                 fs->fs_csaddr = fs->fs_old_csaddr;
1136                 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1137                 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1138                 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1139                 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1140         }
1141         if (fs->fs_magic == FS_UFS1_MAGIC &&
1142             fs->fs_old_inodefmt < FS_44INODEFMT) {
1143                 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1144                 fs->fs_qbmask = ~fs->fs_bmask;
1145                 fs->fs_qfmask = ~fs->fs_fmask;
1146         }
1147         if (fs->fs_magic == FS_UFS1_MAGIC) {
1148                 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1149                 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1150                 if (fs->fs_maxfilesize > maxfilesize)
1151                         fs->fs_maxfilesize = maxfilesize;
1152         }
1153         /* Compatibility for old filesystems */
1154         if (fs->fs_avgfilesize <= 0)
1155                 fs->fs_avgfilesize = AVFILESIZ;
1156         if (fs->fs_avgfpdir <= 0)
1157                 fs->fs_avgfpdir = AFPDIR;
1158         if (bigcgs) {
1159                 fs->fs_save_cgsize = fs->fs_cgsize;
1160                 fs->fs_cgsize = fs->fs_bsize;
1161         }
1162 }
1163
1164 /*
1165  * Unwinding superblock updates for old filesystems.
1166  * See ffs_oldfscompat_read above for details.
1167  *
1168  * XXX - Parts get retired eventually.
1169  * Unfortunately new bits get added.
1170  */
1171 void
1172 ffs_oldfscompat_write(fs, ump)
1173         struct fs *fs;
1174         struct ufsmount *ump;
1175 {
1176
1177         /*
1178          * Copy back UFS2 updated fields that UFS1 inspects.
1179          */
1180         if (fs->fs_magic == FS_UFS1_MAGIC) {
1181                 fs->fs_old_time = fs->fs_time;
1182                 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1183                 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1184                 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1185                 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1186                 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1187         }
1188         if (bigcgs) {
1189                 fs->fs_cgsize = fs->fs_save_cgsize;
1190                 fs->fs_save_cgsize = 0;
1191         }
1192 }
1193
1194 /*
1195  * unmount system call
1196  */
1197 static int
1198 ffs_unmount(mp, mntflags)
1199         struct mount *mp;
1200         int mntflags;
1201 {
1202         struct thread *td;
1203         struct ufsmount *ump = VFSTOUFS(mp);
1204         struct fs *fs;
1205         int error, flags, susp;
1206 #ifdef UFS_EXTATTR
1207         int e_restart;
1208 #endif
1209
1210         flags = 0;
1211         td = curthread;
1212         fs = ump->um_fs;
1213         if (mntflags & MNT_FORCE) {
1214                 flags |= FORCECLOSE;
1215                 susp = fs->fs_ronly != 0;
1216         } else
1217                 susp = 0;
1218 #ifdef UFS_EXTATTR
1219         if ((error = ufs_extattr_stop(mp, td))) {
1220                 if (error != EOPNOTSUPP)
1221                         printf("ffs_unmount: ufs_extattr_stop returned %d\n",
1222                             error);
1223                 e_restart = 0;
1224         } else {
1225                 ufs_extattr_uepm_destroy(&ump->um_extattr);
1226                 e_restart = 1;
1227         }
1228 #endif
1229         if (susp) {
1230                 /*
1231                  * dounmount already called vn_start_write().
1232                  */
1233                 for (;;) {
1234                         vn_finished_write(mp);
1235                         if ((error = vfs_write_suspend(mp)) != 0)
1236                                 return (error);
1237                         MNT_ILOCK(mp);
1238                         if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
1239                                 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
1240                                     MNTK_SUSPEND2);
1241                                 wakeup(&mp->mnt_flag);
1242                                 MNT_IUNLOCK(mp);
1243                                 td->td_pflags |= TDP_IGNSUSP;
1244                                 break;
1245                         }
1246                         MNT_IUNLOCK(mp);
1247                         vn_start_write(NULL, &mp, V_WAIT);
1248                 }
1249         }
1250         if (MOUNTEDSOFTDEP(mp))
1251                 error = softdep_flushfiles(mp, flags, td);
1252         else
1253                 error = ffs_flushfiles(mp, flags, td);
1254         if (error != 0 && error != ENXIO)
1255                 goto fail;
1256
1257         UFS_LOCK(ump);
1258         if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1259                 printf("%s: unmount pending error: blocks %jd files %d\n",
1260                     fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1261                     fs->fs_pendinginodes);
1262                 fs->fs_pendingblocks = 0;
1263                 fs->fs_pendinginodes = 0;
1264         }
1265         UFS_UNLOCK(ump);
1266         softdep_unmount(mp);
1267         if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1268                 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1269                 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1270                 if (error && error != ENXIO) {
1271                         fs->fs_clean = 0;
1272                         goto fail;
1273                 }
1274         }
1275         if (susp) {
1276                 vfs_write_resume(mp);
1277                 vn_start_write(NULL, &mp, V_WAIT);
1278         }
1279         DROP_GIANT();
1280         g_topology_lock();
1281         if (ump->um_fsckpid > 0) {
1282                 /*
1283                  * Return to normal read-only mode.
1284                  */
1285                 error = g_access(ump->um_cp, 0, -1, 0);
1286                 ump->um_fsckpid = 0;
1287         }
1288         g_vfs_close(ump->um_cp);
1289         g_topology_unlock();
1290         PICKUP_GIANT();
1291         vrele(ump->um_devvp);
1292         dev_rel(ump->um_dev);
1293         mtx_destroy(UFS_MTX(ump));
1294         if (mp->mnt_gjprovider != NULL) {
1295                 free(mp->mnt_gjprovider, M_UFSMNT);
1296                 mp->mnt_gjprovider = NULL;
1297         }
1298         free(fs->fs_csp, M_UFSMNT);
1299         free(fs, M_UFSMNT);
1300         free(ump, M_UFSMNT);
1301         mp->mnt_data = NULL;
1302         MNT_ILOCK(mp);
1303         mp->mnt_flag &= ~MNT_LOCAL;
1304         MNT_IUNLOCK(mp);
1305         return (error);
1306
1307 fail:
1308         if (susp) {
1309                 vfs_write_resume(mp);
1310                 vn_start_write(NULL, &mp, V_WAIT);
1311         }
1312 #ifdef UFS_EXTATTR
1313         if (e_restart) {
1314                 ufs_extattr_uepm_init(&ump->um_extattr);
1315 #ifdef UFS_EXTATTR_AUTOSTART
1316                 (void) ufs_extattr_autostart(mp, td);
1317 #endif
1318         }
1319 #endif
1320
1321         return (error);
1322 }
1323
1324 /*
1325  * Flush out all the files in a filesystem.
1326  */
1327 int
1328 ffs_flushfiles(mp, flags, td)
1329         struct mount *mp;
1330         int flags;
1331         struct thread *td;
1332 {
1333         struct ufsmount *ump;
1334         int error;
1335
1336         ump = VFSTOUFS(mp);
1337 #ifdef QUOTA
1338         if (mp->mnt_flag & MNT_QUOTA) {
1339                 int i;
1340                 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1341                 if (error)
1342                         return (error);
1343                 for (i = 0; i < MAXQUOTAS; i++) {
1344                         quotaoff(td, mp, i);
1345                 }
1346                 /*
1347                  * Here we fall through to vflush again to ensure
1348                  * that we have gotten rid of all the system vnodes.
1349                  */
1350         }
1351 #endif
1352         ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1353         if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1354                 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1355                         return (error);
1356                 ffs_snapshot_unmount(mp);
1357                 flags |= FORCECLOSE;
1358                 /*
1359                  * Here we fall through to vflush again to ensure
1360                  * that we have gotten rid of all the system vnodes.
1361                  */
1362         }
1363         /*
1364          * Flush all the files.
1365          */
1366         if ((error = vflush(mp, 0, flags, td)) != 0)
1367                 return (error);
1368         /*
1369          * Flush filesystem metadata.
1370          */
1371         vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1372         error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1373         VOP_UNLOCK(ump->um_devvp, 0);
1374         return (error);
1375 }
1376
1377 /*
1378  * Get filesystem statistics.
1379  */
1380 static int
1381 ffs_statfs(mp, sbp)
1382         struct mount *mp;
1383         struct statfs *sbp;
1384 {
1385         struct ufsmount *ump;
1386         struct fs *fs;
1387
1388         ump = VFSTOUFS(mp);
1389         fs = ump->um_fs;
1390         if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1391                 panic("ffs_statfs");
1392         sbp->f_version = STATFS_VERSION;
1393         sbp->f_bsize = fs->fs_fsize;
1394         sbp->f_iosize = fs->fs_bsize;
1395         sbp->f_blocks = fs->fs_dsize;
1396         UFS_LOCK(ump);
1397         sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1398             fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1399         sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1400             dbtofsb(fs, fs->fs_pendingblocks);
1401         sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1402         sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1403         UFS_UNLOCK(ump);
1404         sbp->f_namemax = NAME_MAX;
1405         return (0);
1406 }
1407
1408 /*
1409  * Go through the disk queues to initiate sandbagged IO;
1410  * go through the inodes to write those that have been modified;
1411  * initiate the writing of the super block if it has been modified.
1412  *
1413  * Note: we are always called with the filesystem marked `MPBUSY'.
1414  */
1415 static int
1416 ffs_sync(mp, waitfor)
1417         struct mount *mp;
1418         int waitfor;
1419 {
1420         struct vnode *mvp, *vp, *devvp;
1421         struct thread *td;
1422         struct inode *ip;
1423         struct ufsmount *ump = VFSTOUFS(mp);
1424         struct fs *fs;
1425         int error, count, wait, lockreq, allerror = 0;
1426         int suspend;
1427         int suspended;
1428         int secondary_writes;
1429         int secondary_accwrites;
1430         int softdep_deps;
1431         int softdep_accdeps;
1432         struct bufobj *bo;
1433
1434         td = curthread;
1435         fs = ump->um_fs;
1436         if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) {
1437                 printf("fs = %s\n", fs->fs_fsmnt);
1438                 panic("ffs_sync: rofs mod");
1439         }
1440         /*
1441          * Write back each (modified) inode.
1442          */
1443         wait = 0;
1444         suspend = 0;
1445         suspended = 0;
1446         lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1447         if (waitfor == MNT_SUSPEND) {
1448                 suspend = 1;
1449                 waitfor = MNT_WAIT;
1450         }
1451         if (waitfor == MNT_WAIT) {
1452                 wait = 1;
1453                 lockreq = LK_EXCLUSIVE;
1454         }
1455         lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1456         MNT_ILOCK(mp);
1457 loop:
1458         /* Grab snapshot of secondary write counts */
1459         secondary_writes = mp->mnt_secondary_writes;
1460         secondary_accwrites = mp->mnt_secondary_accwrites;
1461
1462         /* Grab snapshot of softdep dependency counts */
1463         MNT_IUNLOCK(mp);
1464         softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1465         MNT_ILOCK(mp);
1466
1467         MNT_VNODE_FOREACH(vp, mp, mvp) {
1468                 /*
1469                  * Depend on the mntvnode_slock to keep things stable enough
1470                  * for a quick test.  Since there might be hundreds of
1471                  * thousands of vnodes, we cannot afford even a subroutine
1472                  * call unless there's a good chance that we have work to do.
1473                  */
1474                 VI_LOCK(vp);
1475                 if (vp->v_iflag & VI_DOOMED) {
1476                         VI_UNLOCK(vp);
1477                         continue;
1478                 }
1479                 ip = VTOI(vp);
1480                 if (vp->v_type == VNON || ((ip->i_flag &
1481                     (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1482                     vp->v_bufobj.bo_dirty.bv_cnt == 0)) {
1483                         VI_UNLOCK(vp);
1484                         continue;
1485                 }
1486                 MNT_IUNLOCK(mp);
1487                 if ((error = vget(vp, lockreq, td)) != 0) {
1488                         MNT_ILOCK(mp);
1489                         if (error == ENOENT || error == ENOLCK) {
1490                                 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
1491                                 goto loop;
1492                         }
1493                         continue;
1494                 }
1495                 if ((error = ffs_syncvnode(vp, waitfor)) != 0)
1496                         allerror = error;
1497                 vput(vp);
1498                 MNT_ILOCK(mp);
1499         }
1500         MNT_IUNLOCK(mp);
1501         /*
1502          * Force stale filesystem control information to be flushed.
1503          */
1504         if (waitfor == MNT_WAIT) {
1505                 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1506                         allerror = error;
1507                 /* Flushed work items may create new vnodes to clean */
1508                 if (allerror == 0 && count) {
1509                         MNT_ILOCK(mp);
1510                         goto loop;
1511                 }
1512         }
1513 #ifdef QUOTA
1514         qsync(mp);
1515 #endif
1516         devvp = ump->um_devvp;
1517         bo = &devvp->v_bufobj;
1518         BO_LOCK(bo);
1519         if (waitfor != MNT_LAZY &&
1520             (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
1521                 BO_UNLOCK(bo);
1522                 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1523                 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1524                         allerror = error;
1525                 VOP_UNLOCK(devvp, 0);
1526                 if (allerror == 0 && waitfor == MNT_WAIT) {
1527                         MNT_ILOCK(mp);
1528                         goto loop;
1529                 }
1530         } else if (suspend != 0) {
1531                 if (softdep_check_suspend(mp,
1532                                           devvp,
1533                                           softdep_deps,
1534                                           softdep_accdeps,
1535                                           secondary_writes,
1536                                           secondary_accwrites) != 0)
1537                         goto loop;      /* More work needed */
1538                 mtx_assert(MNT_MTX(mp), MA_OWNED);
1539                 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1540                 MNT_IUNLOCK(mp);
1541                 suspended = 1;
1542         } else
1543                 BO_UNLOCK(bo);
1544         /*
1545          * Write back modified superblock.
1546          */
1547         if (fs->fs_fmod != 0 &&
1548             (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1549                 allerror = error;
1550         return (allerror);
1551 }
1552
1553 int
1554 ffs_vget(mp, ino, flags, vpp)
1555         struct mount *mp;
1556         ino_t ino;
1557         int flags;
1558         struct vnode **vpp;
1559 {
1560         return (ffs_vgetf(mp, ino, flags, vpp, 0));
1561 }
1562
1563 int
1564 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1565         struct mount *mp;
1566         ino_t ino;
1567         int flags;
1568         struct vnode **vpp;
1569         int ffs_flags;
1570 {
1571         struct fs *fs;
1572         struct inode *ip;
1573         struct ufsmount *ump;
1574         struct buf *bp;
1575         struct vnode *vp;
1576         struct cdev *dev;
1577         int error;
1578
1579         error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1580         if (error || *vpp != NULL)
1581                 return (error);
1582
1583         /*
1584          * We must promote to an exclusive lock for vnode creation.  This
1585          * can happen if lookup is passed LOCKSHARED.
1586          */
1587         if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1588                 flags &= ~LK_TYPE_MASK;
1589                 flags |= LK_EXCLUSIVE;
1590         }
1591
1592         /*
1593          * We do not lock vnode creation as it is believed to be too
1594          * expensive for such rare case as simultaneous creation of vnode
1595          * for same ino by different processes. We just allow them to race
1596          * and check later to decide who wins. Let the race begin!
1597          */
1598
1599         ump = VFSTOUFS(mp);
1600         dev = ump->um_dev;
1601         fs = ump->um_fs;
1602
1603         /*
1604          * If this malloc() is performed after the getnewvnode()
1605          * it might block, leaving a vnode with a NULL v_data to be
1606          * found by ffs_sync() if a sync happens to fire right then,
1607          * which will cause a panic because ffs_sync() blindly
1608          * dereferences vp->v_data (as well it should).
1609          */
1610         ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1611
1612         /* Allocate a new vnode/inode. */
1613         if (fs->fs_magic == FS_UFS1_MAGIC)
1614                 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1615         else
1616                 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1617         if (error) {
1618                 *vpp = NULL;
1619                 uma_zfree(uma_inode, ip);
1620                 return (error);
1621         }
1622         /*
1623          * FFS supports recursive locking.
1624          */
1625         lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1626         VN_LOCK_AREC(vp);
1627         vp->v_data = ip;
1628         vp->v_bufobj.bo_bsize = fs->fs_bsize;
1629         ip->i_vnode = vp;
1630         ip->i_ump = ump;
1631         ip->i_fs = fs;
1632         ip->i_dev = dev;
1633         ip->i_number = ino;
1634         ip->i_ea_refs = 0;
1635 #ifdef QUOTA
1636         {
1637                 int i;
1638                 for (i = 0; i < MAXQUOTAS; i++)
1639                         ip->i_dquot[i] = NODQUOT;
1640         }
1641 #endif
1642
1643         if (ffs_flags & FFSV_FORCEINSMQ)
1644                 vp->v_vflag |= VV_FORCEINSMQ;
1645         error = insmntque(vp, mp);
1646         if (error != 0) {
1647                 uma_zfree(uma_inode, ip);
1648                 *vpp = NULL;
1649                 return (error);
1650         }
1651         vp->v_vflag &= ~VV_FORCEINSMQ;
1652         error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1653         if (error || *vpp != NULL)
1654                 return (error);
1655
1656         /* Read in the disk contents for the inode, copy into the inode. */
1657         error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1658             (int)fs->fs_bsize, NOCRED, &bp);
1659         if (error) {
1660                 /*
1661                  * The inode does not contain anything useful, so it would
1662                  * be misleading to leave it on its hash chain. With mode
1663                  * still zero, it will be unlinked and returned to the free
1664                  * list by vput().
1665                  */
1666                 brelse(bp);
1667                 vput(vp);
1668                 *vpp = NULL;
1669                 return (error);
1670         }
1671         if (ip->i_ump->um_fstype == UFS1)
1672                 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1673         else
1674                 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1675         ffs_load_inode(bp, ip, fs, ino);
1676         if (DOINGSOFTDEP(vp))
1677                 softdep_load_inodeblock(ip);
1678         else
1679                 ip->i_effnlink = ip->i_nlink;
1680         bqrelse(bp);
1681
1682         /*
1683          * Initialize the vnode from the inode, check for aliases.
1684          * Note that the underlying vnode may have changed.
1685          */
1686         if (ip->i_ump->um_fstype == UFS1)
1687                 error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1688         else
1689                 error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1690         if (error) {
1691                 vput(vp);
1692                 *vpp = NULL;
1693                 return (error);
1694         }
1695
1696         /*
1697          * Finish inode initialization.
1698          */
1699         if (vp->v_type != VFIFO) {
1700                 /* FFS supports shared locking for all files except fifos. */
1701                 VN_LOCK_ASHARE(vp);
1702         }
1703
1704         /*
1705          * Set up a generation number for this inode if it does not
1706          * already have one. This should only happen on old filesystems.
1707          */
1708         if (ip->i_gen == 0) {
1709                 ip->i_gen = arc4random() / 2 + 1;
1710                 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1711                         ip->i_flag |= IN_MODIFIED;
1712                         DIP_SET(ip, i_gen, ip->i_gen);
1713                 }
1714         }
1715 #ifdef MAC
1716         if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1717                 /*
1718                  * If this vnode is already allocated, and we're running
1719                  * multi-label, attempt to perform a label association
1720                  * from the extended attributes on the inode.
1721                  */
1722                 error = mac_vnode_associate_extattr(mp, vp);
1723                 if (error) {
1724                         /* ufs_inactive will release ip->i_devvp ref. */
1725                         vput(vp);
1726                         *vpp = NULL;
1727                         return (error);
1728                 }
1729         }
1730 #endif
1731
1732         *vpp = vp;
1733         return (0);
1734 }
1735
1736 /*
1737  * File handle to vnode
1738  *
1739  * Have to be really careful about stale file handles:
1740  * - check that the inode number is valid
1741  * - call ffs_vget() to get the locked inode
1742  * - check for an unallocated inode (i_mode == 0)
1743  * - check that the given client host has export rights and return
1744  *   those rights via. exflagsp and credanonp
1745  */
1746 static int
1747 ffs_fhtovp(mp, fhp, flags, vpp)
1748         struct mount *mp;
1749         struct fid *fhp;
1750         int flags;
1751         struct vnode **vpp;
1752 {
1753         struct ufid *ufhp;
1754         struct fs *fs;
1755
1756         ufhp = (struct ufid *)fhp;
1757         fs = VFSTOUFS(mp)->um_fs;
1758         if (ufhp->ufid_ino < ROOTINO ||
1759             ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1760                 return (ESTALE);
1761         return (ufs_fhtovp(mp, ufhp, flags, vpp));
1762 }
1763
1764 /*
1765  * Initialize the filesystem.
1766  */
1767 static int
1768 ffs_init(vfsp)
1769         struct vfsconf *vfsp;
1770 {
1771
1772         softdep_initialize();
1773         return (ufs_init(vfsp));
1774 }
1775
1776 /*
1777  * Undo the work of ffs_init().
1778  */
1779 static int
1780 ffs_uninit(vfsp)
1781         struct vfsconf *vfsp;
1782 {
1783         int ret;
1784
1785         ret = ufs_uninit(vfsp);
1786         softdep_uninitialize();
1787         return (ret);
1788 }
1789
1790 /*
1791  * Write a superblock and associated information back to disk.
1792  */
1793 int
1794 ffs_sbupdate(ump, waitfor, suspended)
1795         struct ufsmount *ump;
1796         int waitfor;
1797         int suspended;
1798 {
1799         struct fs *fs = ump->um_fs;
1800         struct buf *sbbp;
1801         struct buf *bp;
1802         int blks;
1803         void *space;
1804         int i, size, error, allerror = 0;
1805
1806         if (fs->fs_ronly == 1 &&
1807             (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1808             (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1809                 panic("ffs_sbupdate: write read-only filesystem");
1810         /*
1811          * We use the superblock's buf to serialize calls to ffs_sbupdate().
1812          */
1813         sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1814             (int)fs->fs_sbsize, 0, 0, 0);
1815         /*
1816          * First write back the summary information.
1817          */
1818         blks = howmany(fs->fs_cssize, fs->fs_fsize);
1819         space = fs->fs_csp;
1820         for (i = 0; i < blks; i += fs->fs_frag) {
1821                 size = fs->fs_bsize;
1822                 if (i + fs->fs_frag > blks)
1823                         size = (blks - i) * fs->fs_fsize;
1824                 bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1825                     size, 0, 0, 0);
1826                 bcopy(space, bp->b_data, (u_int)size);
1827                 space = (char *)space + size;
1828                 if (suspended)
1829                         bp->b_flags |= B_VALIDSUSPWRT;
1830                 if (waitfor != MNT_WAIT)
1831                         bawrite(bp);
1832                 else if ((error = bwrite(bp)) != 0)
1833                         allerror = error;
1834         }
1835         /*
1836          * Now write back the superblock itself. If any errors occurred
1837          * up to this point, then fail so that the superblock avoids
1838          * being written out as clean.
1839          */
1840         if (allerror) {
1841                 brelse(sbbp);
1842                 return (allerror);
1843         }
1844         bp = sbbp;
1845         if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1846             (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1847                 printf("%s: correcting fs_sblockloc from %jd to %d\n",
1848                     fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1849                 fs->fs_sblockloc = SBLOCK_UFS1;
1850         }
1851         if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1852             (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1853                 printf("%s: correcting fs_sblockloc from %jd to %d\n",
1854                     fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1855                 fs->fs_sblockloc = SBLOCK_UFS2;
1856         }
1857         fs->fs_fmod = 0;
1858         fs->fs_time = time_second;
1859         if (fs->fs_flags & FS_DOSOFTDEP)
1860                 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1861         bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1862         ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1863         if (suspended)
1864                 bp->b_flags |= B_VALIDSUSPWRT;
1865         if (waitfor != MNT_WAIT)
1866                 bawrite(bp);
1867         else if ((error = bwrite(bp)) != 0)
1868                 allerror = error;
1869         return (allerror);
1870 }
1871
1872 static int
1873 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1874         int attrnamespace, const char *attrname)
1875 {
1876
1877 #ifdef UFS_EXTATTR
1878         return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1879             attrname));
1880 #else
1881         return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1882             attrname));
1883 #endif
1884 }
1885
1886 static void
1887 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1888 {
1889
1890         if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1891                 uma_zfree(uma_ufs1, ip->i_din1);
1892         else if (ip->i_din2 != NULL)
1893                 uma_zfree(uma_ufs2, ip->i_din2);
1894         uma_zfree(uma_inode, ip);
1895 }
1896
1897 static int dobkgrdwrite = 1;
1898 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1899     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1900
1901 /*
1902  * Complete a background write started from bwrite.
1903  */
1904 static void
1905 ffs_backgroundwritedone(struct buf *bp)
1906 {
1907         struct bufobj *bufobj;
1908         struct buf *origbp;
1909
1910         /*
1911          * Find the original buffer that we are writing.
1912          */
1913         bufobj = bp->b_bufobj;
1914         BO_LOCK(bufobj);
1915         if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1916                 panic("backgroundwritedone: lost buffer");
1917         /* Grab an extra reference to be dropped by the bufdone() below. */
1918         bufobj_wrefl(bufobj);
1919         BO_UNLOCK(bufobj);
1920         /*
1921          * Process dependencies then return any unfinished ones.
1922          */
1923         if (!LIST_EMPTY(&bp->b_dep))
1924                 buf_complete(bp);
1925 #ifdef SOFTUPDATES
1926         if (!LIST_EMPTY(&bp->b_dep))
1927                 softdep_move_dependencies(bp, origbp);
1928 #endif
1929         /*
1930          * This buffer is marked B_NOCACHE so when it is released
1931          * by biodone it will be tossed.
1932          */
1933         bp->b_flags |= B_NOCACHE;
1934         bp->b_flags &= ~B_CACHE;
1935         bufdone(bp);
1936         BO_LOCK(bufobj);
1937         /*
1938          * Clear the BV_BKGRDINPROG flag in the original buffer
1939          * and awaken it if it is waiting for the write to complete.
1940          * If BV_BKGRDINPROG is not set in the original buffer it must
1941          * have been released and re-instantiated - which is not legal.
1942          */
1943         KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
1944             ("backgroundwritedone: lost buffer2"));
1945         origbp->b_vflags &= ~BV_BKGRDINPROG;
1946         if (origbp->b_vflags & BV_BKGRDWAIT) {
1947                 origbp->b_vflags &= ~BV_BKGRDWAIT;
1948                 wakeup(&origbp->b_xflags);
1949         }
1950         BO_UNLOCK(bufobj);
1951 }
1952
1953
1954 /*
1955  * Write, release buffer on completion.  (Done by iodone
1956  * if async).  Do not bother writing anything if the buffer
1957  * is invalid.
1958  *
1959  * Note that we set B_CACHE here, indicating that buffer is
1960  * fully valid and thus cacheable.  This is true even of NFS
1961  * now so we set it generally.  This could be set either here
1962  * or in biodone() since the I/O is synchronous.  We put it
1963  * here.
1964  */
1965 static int
1966 ffs_bufwrite(struct buf *bp)
1967 {
1968         int oldflags, s;
1969         struct buf *newbp;
1970
1971         CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1972         if (bp->b_flags & B_INVAL) {
1973                 brelse(bp);
1974                 return (0);
1975         }
1976
1977         oldflags = bp->b_flags;
1978
1979         if (!BUF_ISLOCKED(bp))
1980                 panic("bufwrite: buffer is not busy???");
1981         s = splbio();
1982         /*
1983          * If a background write is already in progress, delay
1984          * writing this block if it is asynchronous. Otherwise
1985          * wait for the background write to complete.
1986          */
1987         BO_LOCK(bp->b_bufobj);
1988         if (bp->b_vflags & BV_BKGRDINPROG) {
1989                 if (bp->b_flags & B_ASYNC) {
1990                         BO_UNLOCK(bp->b_bufobj);
1991                         splx(s);
1992                         bdwrite(bp);
1993                         return (0);
1994                 }
1995                 bp->b_vflags |= BV_BKGRDWAIT;
1996                 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0);
1997                 if (bp->b_vflags & BV_BKGRDINPROG)
1998                         panic("bufwrite: still writing");
1999         }
2000         BO_UNLOCK(bp->b_bufobj);
2001
2002         /*
2003          * If this buffer is marked for background writing and we
2004          * do not have to wait for it, make a copy and write the
2005          * copy so as to leave this buffer ready for further use.
2006          *
2007          * This optimization eats a lot of memory.  If we have a page
2008          * or buffer shortfall we can't do it.
2009          */
2010         if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2011             (bp->b_flags & B_ASYNC) &&
2012             !vm_page_count_severe() &&
2013             !buf_dirty_count_severe()) {
2014                 KASSERT(bp->b_iodone == NULL,
2015                     ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2016
2017                 /* get a new block */
2018                 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2019                 if (newbp == NULL)
2020                         goto normal_write;
2021
2022                 /*
2023                  * set it to be identical to the old block.  We have to
2024                  * set b_lblkno and BKGRDMARKER before calling bgetvp()
2025                  * to avoid confusing the splay tree and gbincore().
2026                  */
2027                 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2028                 newbp->b_lblkno = bp->b_lblkno;
2029                 newbp->b_xflags |= BX_BKGRDMARKER;
2030                 BO_LOCK(bp->b_bufobj);
2031                 bp->b_vflags |= BV_BKGRDINPROG;
2032                 bgetvp(bp->b_vp, newbp);
2033                 BO_UNLOCK(bp->b_bufobj);
2034                 newbp->b_bufobj = &bp->b_vp->v_bufobj;
2035                 newbp->b_blkno = bp->b_blkno;
2036                 newbp->b_offset = bp->b_offset;
2037                 newbp->b_iodone = ffs_backgroundwritedone;
2038                 newbp->b_flags |= B_ASYNC;
2039                 newbp->b_flags &= ~B_INVAL;
2040
2041 #ifdef SOFTUPDATES
2042                 /*
2043                  * Move over the dependencies.  If there are rollbacks,
2044                  * leave the parent buffer dirtied as it will need to
2045                  * be written again.
2046                  */
2047                 if (LIST_EMPTY(&bp->b_dep) ||
2048                     softdep_move_dependencies(bp, newbp) == 0)
2049                         bundirty(bp);
2050 #else
2051                 bundirty(bp);
2052 #endif
2053
2054                 /*
2055                  * Initiate write on the copy, release the original to
2056                  * the B_LOCKED queue so that it cannot go away until
2057                  * the background write completes. If not locked it could go
2058                  * away and then be reconstituted while it was being written.
2059                  * If the reconstituted buffer were written, we could end up
2060                  * with two background copies being written at the same time.
2061                  */
2062                 bqrelse(bp);
2063                 bp = newbp;
2064         } else
2065                 /* Mark the buffer clean */
2066                 bundirty(bp);
2067
2068
2069         /* Let the normal bufwrite do the rest for us */
2070 normal_write:
2071         return (bufwrite(bp));
2072 }
2073
2074
2075 static void
2076 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2077 {
2078         struct vnode *vp;
2079         int error;
2080         struct buf *tbp;
2081         int nocopy;
2082
2083         vp = bo->__bo_vnode;
2084         if (bp->b_iocmd == BIO_WRITE) {
2085                 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2086                     bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2087                     (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2088                         panic("ffs_geom_strategy: bad I/O");
2089                 nocopy = bp->b_flags & B_NOCOPY;
2090                 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2091                 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2092                     vp->v_rdev->si_snapdata != NULL) {
2093                         if ((bp->b_flags & B_CLUSTER) != 0) {
2094                                 runningbufwakeup(bp);
2095                                 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2096                                               b_cluster.cluster_entry) {
2097                                         error = ffs_copyonwrite(vp, tbp);
2098                                         if (error != 0 &&
2099                                             error != EOPNOTSUPP) {
2100                                                 bp->b_error = error;
2101                                                 bp->b_ioflags |= BIO_ERROR;
2102                                                 bufdone(bp);
2103                                                 return;
2104                                         }
2105                                 }
2106                                 bp->b_runningbufspace = bp->b_bufsize;
2107                                 atomic_add_long(&runningbufspace,
2108                                                bp->b_runningbufspace);
2109                         } else {
2110                                 error = ffs_copyonwrite(vp, bp);
2111                                 if (error != 0 && error != EOPNOTSUPP) {
2112                                         bp->b_error = error;
2113                                         bp->b_ioflags |= BIO_ERROR;
2114                                         bufdone(bp);
2115                                         return;
2116                                 }
2117                         }
2118                 }
2119 #ifdef SOFTUPDATES
2120                 if ((bp->b_flags & B_CLUSTER) != 0) {
2121                         TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2122                                       b_cluster.cluster_entry) {
2123                                 if (!LIST_EMPTY(&tbp->b_dep))
2124                                         buf_start(tbp);
2125                         }
2126                 } else {
2127                         if (!LIST_EMPTY(&bp->b_dep))
2128                                 buf_start(bp);
2129                 }
2130
2131 #endif
2132         }
2133         g_vfs_strategy(bo, bp);
2134 }
2135
2136 #ifdef  DDB
2137
2138 static void
2139 db_print_ffs(struct ufsmount *ump)
2140 {
2141         db_printf("mp %p %s devvp %p fs %p su_wl %d su_deps %d su_req %d\n",
2142             ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname,
2143             ump->um_devvp, ump->um_fs, ump->softdep_on_worklist,
2144             ump->softdep_deps, ump->softdep_req);
2145 }
2146
2147 DB_SHOW_COMMAND(ffs, db_show_ffs)
2148 {
2149         struct mount *mp;
2150         struct ufsmount *ump;
2151
2152         if (have_addr) {
2153                 ump = VFSTOUFS((struct mount *)addr);
2154                 db_print_ffs(ump);
2155                 return;
2156         }
2157
2158         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2159                 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2160                         db_print_ffs(VFSTOUFS(mp));
2161         }
2162 }
2163
2164 #endif  /* DDB */