coda_vfsops.c revision 1.40 1 /* $NetBSD: coda_vfsops.c,v 1.40 2004/10/15 09:09:09 skrll Exp $ */
2
3 /*
4 *
5 * Coda: an Experimental Distributed File System
6 * Release 3.1
7 *
8 * Copyright (c) 1987-1998 Carnegie Mellon University
9 * All Rights Reserved
10 *
11 * Permission to use, copy, modify and distribute this software and its
12 * documentation is hereby granted, provided that both the copyright
13 * notice and this permission notice appear in all copies of the
14 * software, derivative works or modified versions, and any portions
15 * thereof, and that both notices appear in supporting documentation, and
16 * that credit is given to Carnegie Mellon University in all documents
17 * and publicity pertaining to direct or indirect use of this code or its
18 * derivatives.
19 *
20 * CODA IS AN EXPERIMENTAL SOFTWARE SYSTEM AND IS KNOWN TO HAVE BUGS,
21 * SOME OF WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON ALLOWS
22 * FREE USE OF THIS SOFTWARE IN ITS "AS IS" CONDITION. CARNEGIE MELLON
23 * DISCLAIMS ANY LIABILITY OF ANY KIND FOR ANY DAMAGES WHATSOEVER
24 * RESULTING DIRECTLY OR INDIRECTLY FROM THE USE OF THIS SOFTWARE OR OF
25 * ANY DERIVATIVE WORK.
26 *
27 * Carnegie Mellon encourages users of this software to return any
28 * improvements or extensions that they make, and to grant Carnegie
29 * Mellon the rights to redistribute these changes without encumbrance.
30 *
31 * @(#) cfs/coda_vfsops.c,v 1.1.1.1 1998/08/29 21:26:45 rvb Exp $
32 */
33
34 /*
35 * Mach Operating System
36 * Copyright (c) 1989 Carnegie-Mellon University
37 * All rights reserved. The CMU software License Agreement specifies
38 * the terms and conditions for use and redistribution.
39 */
40
41 /*
42 * This code was written for the Coda file system at Carnegie Mellon
43 * University. Contributers include David Steere, James Kistler, and
44 * M. Satyanarayanan.
45 */
46
47 #include <sys/cdefs.h>
48 __KERNEL_RCSID(0, "$NetBSD: coda_vfsops.c,v 1.40 2004/10/15 09:09:09 skrll Exp $");
49
50 #ifdef _LKM
51 #define NVCODA 4
52 #else
53 #include <vcoda.h>
54 #endif
55
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/sysctl.h>
59 #include <sys/malloc.h>
60 #include <sys/conf.h>
61 #include <sys/namei.h>
62 #include <sys/dirent.h>
63 #include <sys/mount.h>
64 #include <sys/proc.h>
65 #include <sys/select.h>
66
67 #include <coda/coda.h>
68 #include <coda/cnode.h>
69 #include <coda/coda_vfsops.h>
70 #include <coda/coda_venus.h>
71 #include <coda/coda_subr.h>
72 #include <coda/coda_opstats.h>
73 /* for VN_RDEV */
74 #include <miscfs/specfs/specdev.h>
75
76 MALLOC_DEFINE(M_CODA, "coda", "Coda file system structures and tables");
77
78 int codadebug = 0;
79
80 int coda_vfsop_print_entry = 0;
81 #define ENTRY if(coda_vfsop_print_entry) myprintf(("Entered %s\n",__func__))
82
83 struct vnode *coda_ctlvp;
84 struct coda_mntinfo coda_mnttbl[NVCODA]; /* indexed by minor device number */
85
86 /* structure to keep statistics of internally generated/satisfied calls */
87
88 struct coda_op_stats coda_vfsopstats[CODA_VFSOPS_SIZE];
89
90 #define MARK_ENTRY(op) (coda_vfsopstats[op].entries++)
91 #define MARK_INT_SAT(op) (coda_vfsopstats[op].sat_intrn++)
92 #define MARK_INT_FAIL(op) (coda_vfsopstats[op].unsat_intrn++)
93 #define MRAK_INT_GEN(op) (coda_vfsopstats[op].gen_intrn++)
94
95 extern const struct cdevsw vcoda_cdevsw;
96 extern const struct vnodeopv_desc coda_vnodeop_opv_desc;
97
98 const struct vnodeopv_desc * const coda_vnodeopv_descs[] = {
99 &coda_vnodeop_opv_desc,
100 NULL,
101 };
102
103 struct vfsops coda_vfsops = {
104 MOUNT_CODA,
105 coda_mount,
106 coda_start,
107 coda_unmount,
108 coda_root,
109 coda_quotactl,
110 coda_nb_statvfs,
111 coda_sync,
112 coda_vget,
113 (int (*) (struct mount *, struct fid *, struct vnode ** ))
114 eopnotsupp,
115 (int (*) (struct vnode *, struct fid *)) eopnotsupp,
116 coda_init,
117 NULL,
118 coda_done,
119 NULL,
120 (int (*)(void)) eopnotsupp,
121 (int (*)(struct mount *, struct mbuf *, int *, struct ucred **))
122 eopnotsupp,
123 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
124 coda_vnodeopv_descs,
125 0
126 };
127
128
129 int
130 coda_vfsopstats_init(void)
131 {
132 int i;
133
134 for (i=0;i<CODA_VFSOPS_SIZE;i++) {
135 coda_vfsopstats[i].opcode = i;
136 coda_vfsopstats[i].entries = 0;
137 coda_vfsopstats[i].sat_intrn = 0;
138 coda_vfsopstats[i].unsat_intrn = 0;
139 coda_vfsopstats[i].gen_intrn = 0;
140 }
141
142 return 0;
143 }
144
145 /*
146 * cfs mount vfsop
147 * Set up mount info record and attach it to vfs struct.
148 */
149 /*ARGSUSED*/
150 int
151 coda_mount(vfsp, path, data, ndp, p)
152 struct mount *vfsp; /* Allocated and initialized by mount(2) */
153 const char *path; /* path covered: ignored by the fs-layer */
154 void *data; /* Need to define a data type for this in netbsd? */
155 struct nameidata *ndp; /* Clobber this to lookup the device name */
156 struct proc *p; /* The ever-famous proc pointer */
157 {
158 struct vnode *dvp;
159 struct cnode *cp;
160 dev_t dev;
161 struct coda_mntinfo *mi;
162 struct vnode *rootvp;
163 const struct cdevsw *cdev;
164 CodaFid rootfid = INVAL_FID;
165 CodaFid ctlfid = CTL_FID;
166 int error;
167
168 if (vfsp->mnt_flag & MNT_GETARGS)
169 return 0;
170 ENTRY;
171
172 coda_vfsopstats_init();
173 coda_vnodeopstats_init();
174
175 MARK_ENTRY(CODA_MOUNT_STATS);
176 if (CODA_MOUNTED(vfsp)) {
177 MARK_INT_FAIL(CODA_MOUNT_STATS);
178 return(EBUSY);
179 }
180
181 /* Validate mount device. Similar to getmdev(). */
182
183 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, data, p);
184 error = namei(ndp);
185 dvp = ndp->ni_vp;
186
187 if (error) {
188 MARK_INT_FAIL(CODA_MOUNT_STATS);
189 return (error);
190 }
191 if (dvp->v_type != VCHR) {
192 MARK_INT_FAIL(CODA_MOUNT_STATS);
193 vrele(dvp);
194 return(ENXIO);
195 }
196 dev = dvp->v_specinfo->si_rdev;
197 vrele(dvp);
198 cdev = cdevsw_lookup(dev);
199 if (cdev == NULL) {
200 MARK_INT_FAIL(CODA_MOUNT_STATS);
201 return(ENXIO);
202 }
203
204 /*
205 * See if the device table matches our expectations.
206 */
207 if (cdev != &vcoda_cdevsw)
208 {
209 MARK_INT_FAIL(CODA_MOUNT_STATS);
210 return(ENXIO);
211 }
212
213 if (minor(dev) >= NVCODA || minor(dev) < 0) {
214 MARK_INT_FAIL(CODA_MOUNT_STATS);
215 return(ENXIO);
216 }
217
218 /*
219 * Initialize the mount record and link it to the vfs struct
220 */
221 mi = &coda_mnttbl[minor(dev)];
222
223 if (!VC_OPEN(&mi->mi_vcomm)) {
224 MARK_INT_FAIL(CODA_MOUNT_STATS);
225 return(ENODEV);
226 }
227
228 /* No initialization (here) of mi_vcomm! */
229 vfsp->mnt_data = mi;
230 vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
231 vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
232 vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
233 vfsp->mnt_stat.f_namemax = MAXNAMLEN;
234 mi->mi_vfsp = vfsp;
235
236 /*
237 * Make a root vnode to placate the Vnode interface, but don't
238 * actually make the CODA_ROOT call to venus until the first call
239 * to coda_root in case a server is down while venus is starting.
240 */
241 cp = make_coda_node(&rootfid, vfsp, VDIR);
242 rootvp = CTOV(cp);
243 rootvp->v_flag |= VROOT;
244
245 /* cp = make_coda_node(&ctlfid, vfsp, VCHR);
246 The above code seems to cause a loop in the cnode links.
247 I don't totally understand when it happens, it is caught
248 when closing down the system.
249 */
250 cp = make_coda_node(&ctlfid, 0, VCHR);
251
252 coda_ctlvp = CTOV(cp);
253
254 /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
255 mi->mi_vfsp = vfsp;
256 mi->mi_rootvp = rootvp;
257
258 /* set filesystem block size */
259 vfsp->mnt_stat.f_bsize = 8192; /* XXX -JJK */
260 vfsp->mnt_stat.f_frsize = 8192; /* XXX -JJK */
261
262 /* error is currently guaranteed to be zero, but in case some
263 code changes... */
264 CODADEBUG(1,
265 myprintf(("coda_mount returned %d\n",error)););
266 if (error)
267 MARK_INT_FAIL(CODA_MOUNT_STATS);
268 else
269 MARK_INT_SAT(CODA_MOUNT_STATS);
270
271 return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE, vfsp,
272 p);
273 }
274
275 int
276 coda_start(vfsp, flags, p)
277 struct mount *vfsp;
278 int flags;
279 struct proc *p;
280 {
281 ENTRY;
282 vftomi(vfsp)->mi_started = 1;
283 return (0);
284 }
285
286 int
287 coda_unmount(vfsp, mntflags, p)
288 struct mount *vfsp;
289 int mntflags;
290 struct proc *p;
291 {
292 struct coda_mntinfo *mi = vftomi(vfsp);
293 int active, error = 0;
294
295 ENTRY;
296 MARK_ENTRY(CODA_UMOUNT_STATS);
297 if (!CODA_MOUNTED(vfsp)) {
298 MARK_INT_FAIL(CODA_UMOUNT_STATS);
299 return(EINVAL);
300 }
301
302 if (mi->mi_vfsp == vfsp) { /* We found the victim */
303 if (!IS_UNMOUNTING(VTOC(mi->mi_rootvp)))
304 return (EBUSY); /* Venus is still running */
305
306 #ifdef DEBUG
307 printf("coda_unmount: ROOT: vp %p, cp %p\n", mi->mi_rootvp, VTOC(mi->mi_rootvp));
308 #endif
309 mi->mi_started = 0;
310
311 vrele(mi->mi_rootvp);
312
313 active = coda_kill(vfsp, NOT_DOWNCALL);
314 mi->mi_rootvp->v_flag &= ~VROOT;
315 error = vflush(mi->mi_vfsp, NULLVP, FORCECLOSE);
316 printf("coda_unmount: active = %d, vflush active %d\n", active, error);
317 error = 0;
318
319 /* I'm going to take this out to allow lookups to go through. I'm
320 * not sure it's important anyway. -- DCS 2/2/94
321 */
322 /* vfsp->VFS_DATA = NULL; */
323
324 /* No more vfsp's to hold onto */
325 mi->mi_vfsp = NULL;
326 mi->mi_rootvp = NULL;
327
328 if (error)
329 MARK_INT_FAIL(CODA_UMOUNT_STATS);
330 else
331 MARK_INT_SAT(CODA_UMOUNT_STATS);
332
333 return(error);
334 }
335 return (EINVAL);
336 }
337
338 /*
339 * find root of cfs
340 */
341 int
342 coda_root(vfsp, vpp)
343 struct mount *vfsp;
344 struct vnode **vpp;
345 {
346 struct coda_mntinfo *mi = vftomi(vfsp);
347 int error;
348 struct proc *p = curproc; /* XXX - bnoble */
349 CodaFid VFid;
350 static const CodaFid invalfid = INVAL_FID;
351
352 ENTRY;
353 MARK_ENTRY(CODA_ROOT_STATS);
354
355 if (vfsp == mi->mi_vfsp) {
356 if (memcmp(&VTOC(mi->mi_rootvp)->c_fid, &invalfid, sizeof(CodaFid)))
357 { /* Found valid root. */
358 *vpp = mi->mi_rootvp;
359 /* On Mach, this is vref. On NetBSD, VOP_LOCK */
360 vref(*vpp);
361 vn_lock(*vpp, LK_EXCLUSIVE);
362 MARK_INT_SAT(CODA_ROOT_STATS);
363 return(0);
364 }
365 }
366
367 error = venus_root(vftomi(vfsp), p->p_cred->pc_ucred, p, &VFid);
368
369 if (!error) {
370 /*
371 * Save the new rootfid in the cnode, and rehash the cnode into the
372 * cnode hash with the new fid key.
373 */
374 coda_unsave(VTOC(mi->mi_rootvp));
375 VTOC(mi->mi_rootvp)->c_fid = VFid;
376 coda_save(VTOC(mi->mi_rootvp));
377
378 *vpp = mi->mi_rootvp;
379 vref(*vpp);
380 vn_lock(*vpp, LK_EXCLUSIVE);
381 MARK_INT_SAT(CODA_ROOT_STATS);
382 goto exit;
383 } else if (error == ENODEV || error == EINTR) {
384 /* Gross hack here! */
385 /*
386 * If Venus fails to respond to the CODA_ROOT call, coda_call returns
387 * ENODEV. Return the uninitialized root vnode to allow vfs
388 * operations such as unmount to continue. Without this hack,
389 * there is no way to do an unmount if Venus dies before a
390 * successful CODA_ROOT call is done. All vnode operations
391 * will fail.
392 */
393 *vpp = mi->mi_rootvp;
394 vref(*vpp);
395 vn_lock(*vpp, LK_EXCLUSIVE);
396 MARK_INT_FAIL(CODA_ROOT_STATS);
397 error = 0;
398 goto exit;
399 } else {
400 CODADEBUG( CODA_ROOT, myprintf(("error %d in CODA_ROOT\n", error)); );
401 MARK_INT_FAIL(CODA_ROOT_STATS);
402
403 goto exit;
404 }
405 exit:
406 return(error);
407 }
408
409 int
410 coda_quotactl(vfsp, cmd, uid, arg, p)
411 struct mount *vfsp;
412 int cmd;
413 uid_t uid;
414 void *arg;
415 struct proc *p;
416 {
417 ENTRY;
418 return (EOPNOTSUPP);
419 }
420
421 /*
422 * Get file system statistics.
423 */
424 int
425 coda_nb_statvfs(vfsp, sbp, p)
426 struct mount *vfsp;
427 struct statvfs *sbp;
428 struct proc *p;
429 {
430 struct coda_statfs fsstat;
431 int error;
432
433 ENTRY;
434 MARK_ENTRY(CODA_STATFS_STATS);
435 if (!CODA_MOUNTED(vfsp)) {
436 /* MARK_INT_FAIL(CODA_STATFS_STATS); */
437 return(EINVAL);
438 }
439
440 /* XXX - what to do about f_flags, others? --bnoble */
441 /* Below This is what AFS does
442 #define NB_SFS_SIZ 0x895440
443 */
444 /* Note: Normal fs's have a bsize of 0x400 == 1024 */
445
446 error = venus_statfs(vftomi(vfsp), p->p_cred->pc_ucred, p, &fsstat);
447
448 if (!error) {
449 sbp->f_bsize = 8192; /* XXX */
450 sbp->f_frsize = 8192; /* XXX */
451 sbp->f_iosize = 8192; /* XXX */
452 sbp->f_blocks = fsstat.f_blocks;
453 sbp->f_bfree = fsstat.f_bfree;
454 sbp->f_bavail = fsstat.f_bavail;
455 sbp->f_bresvd = 0;
456 sbp->f_files = fsstat.f_files;
457 sbp->f_ffree = fsstat.f_ffree;
458 sbp->f_favail = fsstat.f_ffree;
459 sbp->f_fresvd = 0;
460 copy_statvfs_info(sbp, vfsp);
461 }
462
463 MARK_INT_SAT(CODA_STATFS_STATS);
464 return(error);
465 }
466
467 /*
468 * Flush any pending I/O.
469 */
470 int
471 coda_sync(vfsp, waitfor, cred, p)
472 struct mount *vfsp;
473 int waitfor;
474 struct ucred *cred;
475 struct proc *p;
476 {
477 ENTRY;
478 MARK_ENTRY(CODA_SYNC_STATS);
479 MARK_INT_SAT(CODA_SYNC_STATS);
480 return(0);
481 }
482
483 int
484 coda_vget(vfsp, ino, vpp)
485 struct mount *vfsp;
486 ino_t ino;
487 struct vnode **vpp;
488 {
489 ENTRY;
490 return (EOPNOTSUPP);
491 }
492
493 /*
494 * fhtovp is now what vget used to be in 4.3-derived systems. For
495 * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
496 * a type-specific fid.
497 */
498 int
499 coda_fhtovp(vfsp, fhp, nam, vpp, exflagsp, creadanonp)
500 struct mount *vfsp;
501 struct fid *fhp;
502 struct mbuf *nam;
503 struct vnode **vpp;
504 int *exflagsp;
505 struct ucred **creadanonp;
506 {
507 struct cfid *cfid = (struct cfid *)fhp;
508 struct cnode *cp = 0;
509 int error;
510 struct proc *p = curproc; /* XXX -mach */
511 CodaFid VFid;
512 int vtype;
513
514 ENTRY;
515
516 MARK_ENTRY(CODA_VGET_STATS);
517 /* Check for vget of control object. */
518 if (IS_CTL_FID(&cfid->cfid_fid)) {
519 *vpp = coda_ctlvp;
520 vref(coda_ctlvp);
521 MARK_INT_SAT(CODA_VGET_STATS);
522 return(0);
523 }
524
525 error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, p->p_cred->pc_ucred, p, &VFid, &vtype);
526
527 if (error) {
528 CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
529 *vpp = (struct vnode *)0;
530 } else {
531 CODADEBUG(CODA_VGET,
532 myprintf(("vget: %s type %d result %d\n",
533 coda_f2s(&VFid), vtype, error)); )
534
535 cp = make_coda_node(&VFid, vfsp, vtype);
536 *vpp = CTOV(cp);
537 }
538 return(error);
539 }
540
541 int
542 coda_vptofh(vnp, fidp)
543 struct vnode *vnp;
544 struct fid *fidp;
545 {
546 ENTRY;
547 return (EOPNOTSUPP);
548 }
549
550 void
551 coda_init(void)
552 {
553 ENTRY;
554 }
555
556 void
557 coda_done(void)
558 {
559 ENTRY;
560 }
561
562 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
563 {
564 sysctl_createv(clog, 0, NULL, NULL,
565 CTLFLAG_PERMANENT,
566 CTLTYPE_NODE, "vfs", NULL,
567 NULL, 0, NULL, 0,
568 CTL_VFS, CTL_EOL);
569 sysctl_createv(clog, 0, NULL, NULL,
570 CTLFLAG_PERMANENT,
571 CTLTYPE_NODE, "coda",
572 SYSCTL_DESCR("code vfs options"),
573 NULL, 0, NULL, 0,
574 CTL_VFS, 18, CTL_EOL);
575 /*
576 * XXX the "18" above could be dynamic, thereby eliminating
577 * one more instance of the "number to vfs" mapping problem,
578 * but "18" is the order as taken from sys/mount.h
579 */
580
581 /*
582 sysctl_createv(clog, 0, NULL, NULL,
583 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
584 CTLTYPE_INT, "clusterread",
585 SYSCTL_DESCR( anyone? ),
586 NULL, 0, &doclusterread, 0,
587 CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
588 */
589 }
590
591 /*
592 * To allow for greater ease of use, some vnodes may be orphaned when
593 * Venus dies. Certain operations should still be allowed to go
594 * through, but without propagating orphan-ness. So this function will
595 * get a new vnode for the file from the current run of Venus.
596 */
597
598 int
599 getNewVnode(vpp)
600 struct vnode **vpp;
601 {
602 struct cfid cfid;
603 struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
604
605 ENTRY;
606
607 cfid.cfid_len = (short)sizeof(CodaFid);
608 cfid.cfid_fid = VTOC(*vpp)->c_fid; /* Structure assignment. */
609 /* XXX ? */
610
611 /* We're guessing that if set, the 1st element on the list is a
612 * valid vnode to use. If not, return ENODEV as venus is dead.
613 */
614 if (mi->mi_vfsp == NULL)
615 return ENODEV;
616
617 return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
618 NULL, NULL);
619 }
620
621 #include <ufs/ufs/quota.h>
622 #include <ufs/ufs/ufsmount.h>
623 /* get the mount structure corresponding to a given device. Assume
624 * device corresponds to a UFS. Return NULL if no device is found.
625 */
626 struct mount *devtomp(dev)
627 dev_t dev;
628 {
629 struct mount *mp, *nmp;
630
631 for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
632 nmp = mp->mnt_list.cqe_next;
633 if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
634 ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
635 /* mount corresponds to UFS and the device matches one we want */
636 return(mp);
637 }
638 }
639 /* mount structure wasn't found */
640 return(NULL);
641 }
642