coda_vfsops.c revision 1.37 1 /* $NetBSD: coda_vfsops.c,v 1.37 2004/05/25 04:44:43 atatat 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 1.37 2004/05/25 04:44:43 atatat 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 coda_vnodeopv_descs,
124 0
125 };
126
127
128 int
129 coda_vfsopstats_init(void)
130 {
131 int i;
132
133 for (i=0;i<CODA_VFSOPS_SIZE;i++) {
134 coda_vfsopstats[i].opcode = i;
135 coda_vfsopstats[i].entries = 0;
136 coda_vfsopstats[i].sat_intrn = 0;
137 coda_vfsopstats[i].unsat_intrn = 0;
138 coda_vfsopstats[i].gen_intrn = 0;
139 }
140
141 return 0;
142 }
143
144 /*
145 * cfs mount vfsop
146 * Set up mount info record and attach it to vfs struct.
147 */
148 /*ARGSUSED*/
149 int
150 coda_mount(vfsp, path, data, ndp, p)
151 struct mount *vfsp; /* Allocated and initialized by mount(2) */
152 const char *path; /* path covered: ignored by the fs-layer */
153 void *data; /* Need to define a data type for this in netbsd? */
154 struct nameidata *ndp; /* Clobber this to lookup the device name */
155 struct proc *p; /* The ever-famous proc pointer */
156 {
157 struct vnode *dvp;
158 struct cnode *cp;
159 dev_t dev;
160 struct coda_mntinfo *mi;
161 struct vnode *rootvp;
162 const struct cdevsw *cdev;
163 CodaFid rootfid = INVAL_FID;
164 CodaFid ctlfid = CTL_FID;
165 int error;
166
167 if (vfsp->mnt_flag & MNT_GETARGS)
168 return 0;
169 ENTRY;
170
171 coda_vfsopstats_init();
172 coda_vnodeopstats_init();
173
174 MARK_ENTRY(CODA_MOUNT_STATS);
175 if (CODA_MOUNTED(vfsp)) {
176 MARK_INT_FAIL(CODA_MOUNT_STATS);
177 return(EBUSY);
178 }
179
180 /* Validate mount device. Similar to getmdev(). */
181
182 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, data, p);
183 error = namei(ndp);
184 dvp = ndp->ni_vp;
185
186 if (error) {
187 MARK_INT_FAIL(CODA_MOUNT_STATS);
188 return (error);
189 }
190 if (dvp->v_type != VCHR) {
191 MARK_INT_FAIL(CODA_MOUNT_STATS);
192 vrele(dvp);
193 return(ENXIO);
194 }
195 dev = dvp->v_specinfo->si_rdev;
196 vrele(dvp);
197 cdev = cdevsw_lookup(dev);
198 if (cdev == NULL) {
199 MARK_INT_FAIL(CODA_MOUNT_STATS);
200 return(ENXIO);
201 }
202
203 /*
204 * See if the device table matches our expectations.
205 */
206 if (cdev != &vcoda_cdevsw)
207 {
208 MARK_INT_FAIL(CODA_MOUNT_STATS);
209 return(ENXIO);
210 }
211
212 if (minor(dev) >= NVCODA || minor(dev) < 0) {
213 MARK_INT_FAIL(CODA_MOUNT_STATS);
214 return(ENXIO);
215 }
216
217 /*
218 * Initialize the mount record and link it to the vfs struct
219 */
220 mi = &coda_mnttbl[minor(dev)];
221
222 if (!VC_OPEN(&mi->mi_vcomm)) {
223 MARK_INT_FAIL(CODA_MOUNT_STATS);
224 return(ENODEV);
225 }
226
227 /* No initialization (here) of mi_vcomm! */
228 vfsp->mnt_data = mi;
229 vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
230 vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
231 vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
232 vfsp->mnt_stat.f_namemax = MAXNAMLEN;
233 mi->mi_vfsp = vfsp;
234
235 /*
236 * Make a root vnode to placate the Vnode interface, but don't
237 * actually make the CODA_ROOT call to venus until the first call
238 * to coda_root in case a server is down while venus is starting.
239 */
240 cp = make_coda_node(&rootfid, vfsp, VDIR);
241 rootvp = CTOV(cp);
242 rootvp->v_flag |= VROOT;
243
244 /* cp = make_coda_node(&ctlfid, vfsp, VCHR);
245 The above code seems to cause a loop in the cnode links.
246 I don't totally understand when it happens, it is caught
247 when closing down the system.
248 */
249 cp = make_coda_node(&ctlfid, 0, VCHR);
250
251 coda_ctlvp = CTOV(cp);
252
253 /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
254 mi->mi_vfsp = vfsp;
255 mi->mi_rootvp = rootvp;
256
257 /* set filesystem block size */
258 vfsp->mnt_stat.f_bsize = 8192; /* XXX -JJK */
259 vfsp->mnt_stat.f_frsize = 8192; /* XXX -JJK */
260
261 /* error is currently guaranteed to be zero, but in case some
262 code changes... */
263 CODADEBUG(1,
264 myprintf(("coda_mount returned %d\n",error)););
265 if (error)
266 MARK_INT_FAIL(CODA_MOUNT_STATS);
267 else
268 MARK_INT_SAT(CODA_MOUNT_STATS);
269
270 return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE, vfsp,
271 p);
272 }
273
274 int
275 coda_start(vfsp, flags, p)
276 struct mount *vfsp;
277 int flags;
278 struct proc *p;
279 {
280 ENTRY;
281 vftomi(vfsp)->mi_started = 1;
282 return (0);
283 }
284
285 int
286 coda_unmount(vfsp, mntflags, p)
287 struct mount *vfsp;
288 int mntflags;
289 struct proc *p;
290 {
291 struct coda_mntinfo *mi = vftomi(vfsp);
292 int active, error = 0;
293
294 ENTRY;
295 MARK_ENTRY(CODA_UMOUNT_STATS);
296 if (!CODA_MOUNTED(vfsp)) {
297 MARK_INT_FAIL(CODA_UMOUNT_STATS);
298 return(EINVAL);
299 }
300
301 if (mi->mi_vfsp == vfsp) { /* We found the victim */
302 if (!IS_UNMOUNTING(VTOC(mi->mi_rootvp)))
303 return (EBUSY); /* Venus is still running */
304
305 #ifdef DEBUG
306 printf("coda_unmount: ROOT: vp %p, cp %p\n", mi->mi_rootvp, VTOC(mi->mi_rootvp));
307 #endif
308 mi->mi_started = 0;
309
310 vrele(mi->mi_rootvp);
311
312 active = coda_kill(vfsp, NOT_DOWNCALL);
313 mi->mi_rootvp->v_flag &= ~VROOT;
314 error = vflush(mi->mi_vfsp, NULLVP, FORCECLOSE);
315 printf("coda_unmount: active = %d, vflush active %d\n", active, error);
316 error = 0;
317
318 /* I'm going to take this out to allow lookups to go through. I'm
319 * not sure it's important anyway. -- DCS 2/2/94
320 */
321 /* vfsp->VFS_DATA = NULL; */
322
323 /* No more vfsp's to hold onto */
324 mi->mi_vfsp = NULL;
325 mi->mi_rootvp = NULL;
326
327 if (error)
328 MARK_INT_FAIL(CODA_UMOUNT_STATS);
329 else
330 MARK_INT_SAT(CODA_UMOUNT_STATS);
331
332 return(error);
333 }
334 return (EINVAL);
335 }
336
337 /*
338 * find root of cfs
339 */
340 int
341 coda_root(vfsp, vpp)
342 struct mount *vfsp;
343 struct vnode **vpp;
344 {
345 struct coda_mntinfo *mi = vftomi(vfsp);
346 int error;
347 struct proc *p = curproc; /* XXX - bnoble */
348 CodaFid VFid;
349 static const CodaFid invalfid = INVAL_FID;
350
351 ENTRY;
352 MARK_ENTRY(CODA_ROOT_STATS);
353
354 if (vfsp == mi->mi_vfsp) {
355 if (memcmp(&VTOC(mi->mi_rootvp)->c_fid, &invalfid, sizeof(CodaFid)))
356 { /* Found valid root. */
357 *vpp = mi->mi_rootvp;
358 /* On Mach, this is vref. On NetBSD, VOP_LOCK */
359 vref(*vpp);
360 vn_lock(*vpp, LK_EXCLUSIVE);
361 MARK_INT_SAT(CODA_ROOT_STATS);
362 return(0);
363 }
364 }
365
366 error = venus_root(vftomi(vfsp), p->p_cred->pc_ucred, p, &VFid);
367
368 if (!error) {
369 /*
370 * Save the new rootfid in the cnode, and rehash the cnode into the
371 * cnode hash with the new fid key.
372 */
373 coda_unsave(VTOC(mi->mi_rootvp));
374 VTOC(mi->mi_rootvp)->c_fid = VFid;
375 coda_save(VTOC(mi->mi_rootvp));
376
377 *vpp = mi->mi_rootvp;
378 vref(*vpp);
379 vn_lock(*vpp, LK_EXCLUSIVE);
380 MARK_INT_SAT(CODA_ROOT_STATS);
381 goto exit;
382 } else if (error == ENODEV || error == EINTR) {
383 /* Gross hack here! */
384 /*
385 * If Venus fails to respond to the CODA_ROOT call, coda_call returns
386 * ENODEV. Return the uninitialized root vnode to allow vfs
387 * operations such as unmount to continue. Without this hack,
388 * there is no way to do an unmount if Venus dies before a
389 * successful CODA_ROOT call is done. All vnode operations
390 * will fail.
391 */
392 *vpp = mi->mi_rootvp;
393 vref(*vpp);
394 vn_lock(*vpp, LK_EXCLUSIVE);
395 MARK_INT_FAIL(CODA_ROOT_STATS);
396 error = 0;
397 goto exit;
398 } else {
399 CODADEBUG( CODA_ROOT, myprintf(("error %d in CODA_ROOT\n", error)); );
400 MARK_INT_FAIL(CODA_ROOT_STATS);
401
402 goto exit;
403 }
404 exit:
405 return(error);
406 }
407
408 int
409 coda_quotactl(vfsp, cmd, uid, arg, p)
410 struct mount *vfsp;
411 int cmd;
412 uid_t uid;
413 void *arg;
414 struct proc *p;
415 {
416 ENTRY;
417 return (EOPNOTSUPP);
418 }
419
420 /*
421 * Get file system statistics.
422 */
423 int
424 coda_nb_statvfs(vfsp, sbp, p)
425 struct mount *vfsp;
426 struct statvfs *sbp;
427 struct proc *p;
428 {
429 struct coda_statfs fsstat;
430 int error;
431
432 ENTRY;
433 MARK_ENTRY(CODA_STATFS_STATS);
434 if (!CODA_MOUNTED(vfsp)) {
435 /* MARK_INT_FAIL(CODA_STATFS_STATS); */
436 return(EINVAL);
437 }
438
439 /* XXX - what to do about f_flags, others? --bnoble */
440 /* Below This is what AFS does
441 #define NB_SFS_SIZ 0x895440
442 */
443 /* Note: Normal fs's have a bsize of 0x400 == 1024 */
444
445 error = venus_statfs(vftomi(vfsp), p->p_cred->pc_ucred, p, &fsstat);
446
447 if (!error) {
448 sbp->f_bsize = 8192; /* XXX */
449 sbp->f_frsize = 8192; /* XXX */
450 sbp->f_iosize = 8192; /* XXX */
451 sbp->f_blocks = fsstat.f_blocks;
452 sbp->f_bfree = fsstat.f_bfree;
453 sbp->f_bavail = fsstat.f_bavail;
454 sbp->f_bresvd = 0;
455 sbp->f_files = fsstat.f_files;
456 sbp->f_ffree = fsstat.f_ffree;
457 sbp->f_favail = fsstat.f_ffree;
458 sbp->f_fresvd = 0;
459 copy_statvfs_info(sbp, vfsp);
460 }
461
462 MARK_INT_SAT(CODA_STATFS_STATS);
463 return(error);
464 }
465
466 /*
467 * Flush any pending I/O.
468 */
469 int
470 coda_sync(vfsp, waitfor, cred, p)
471 struct mount *vfsp;
472 int waitfor;
473 struct ucred *cred;
474 struct proc *p;
475 {
476 ENTRY;
477 MARK_ENTRY(CODA_SYNC_STATS);
478 MARK_INT_SAT(CODA_SYNC_STATS);
479 return(0);
480 }
481
482 int
483 coda_vget(vfsp, ino, vpp)
484 struct mount *vfsp;
485 ino_t ino;
486 struct vnode **vpp;
487 {
488 ENTRY;
489 return (EOPNOTSUPP);
490 }
491
492 /*
493 * fhtovp is now what vget used to be in 4.3-derived systems. For
494 * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
495 * a type-specific fid.
496 */
497 int
498 coda_fhtovp(vfsp, fhp, nam, vpp, exflagsp, creadanonp)
499 struct mount *vfsp;
500 struct fid *fhp;
501 struct mbuf *nam;
502 struct vnode **vpp;
503 int *exflagsp;
504 struct ucred **creadanonp;
505 {
506 struct cfid *cfid = (struct cfid *)fhp;
507 struct cnode *cp = 0;
508 int error;
509 struct proc *p = curproc; /* XXX -mach */
510 CodaFid VFid;
511 int vtype;
512
513 ENTRY;
514
515 MARK_ENTRY(CODA_VGET_STATS);
516 /* Check for vget of control object. */
517 if (IS_CTL_FID(&cfid->cfid_fid)) {
518 *vpp = coda_ctlvp;
519 vref(coda_ctlvp);
520 MARK_INT_SAT(CODA_VGET_STATS);
521 return(0);
522 }
523
524 error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, p->p_cred->pc_ucred, p, &VFid, &vtype);
525
526 if (error) {
527 CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
528 *vpp = (struct vnode *)0;
529 } else {
530 CODADEBUG(CODA_VGET,
531 myprintf(("vget: %s type %d result %d\n",
532 coda_f2s(&VFid), vtype, error)); )
533
534 cp = make_coda_node(&VFid, vfsp, vtype);
535 *vpp = CTOV(cp);
536 }
537 return(error);
538 }
539
540 int
541 coda_vptofh(vnp, fidp)
542 struct vnode *vnp;
543 struct fid *fidp;
544 {
545 ENTRY;
546 return (EOPNOTSUPP);
547 }
548
549 void
550 coda_init(void)
551 {
552 ENTRY;
553 }
554
555 void
556 coda_done(void)
557 {
558 ENTRY;
559 }
560
561 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
562 {
563 sysctl_createv(clog, 0, NULL, NULL,
564 CTLFLAG_PERMANENT,
565 CTLTYPE_NODE, "vfs", NULL,
566 NULL, 0, NULL, 0,
567 CTL_VFS, CTL_EOL);
568 sysctl_createv(clog, 0, NULL, NULL,
569 CTLFLAG_PERMANENT,
570 CTLTYPE_NODE, "coda",
571 SYSCTL_DESCR("code vfs options"),
572 NULL, 0, NULL, 0,
573 CTL_VFS, 18, CTL_EOL);
574 /*
575 * XXX the "18" above could be dynamic, thereby eliminating
576 * one more instance of the "number to vfs" mapping problem,
577 * but "18" is the order as taken from sys/mount.h
578 */
579
580 /*
581 sysctl_createv(clog, 0, NULL, NULL,
582 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
583 CTLTYPE_INT, "clusterread",
584 SYSCTL_DESCR( anyone? ),
585 NULL, 0, &doclusterread, 0,
586 CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
587 */
588 }
589
590 /*
591 * To allow for greater ease of use, some vnodes may be orphaned when
592 * Venus dies. Certain operations should still be allowed to go
593 * through, but without propagating orphan-ness. So this function will
594 * get a new vnode for the file from the current run of Venus.
595 */
596
597 int
598 getNewVnode(vpp)
599 struct vnode **vpp;
600 {
601 struct cfid cfid;
602 struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
603
604 ENTRY;
605
606 cfid.cfid_len = (short)sizeof(CodaFid);
607 cfid.cfid_fid = VTOC(*vpp)->c_fid; /* Structure assignment. */
608 /* XXX ? */
609
610 /* We're guessing that if set, the 1st element on the list is a
611 * valid vnode to use. If not, return ENODEV as venus is dead.
612 */
613 if (mi->mi_vfsp == NULL)
614 return ENODEV;
615
616 return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
617 NULL, NULL);
618 }
619
620 #include <ufs/ufs/quota.h>
621 #include <ufs/ufs/ufsmount.h>
622 /* get the mount structure corresponding to a given device. Assume
623 * device corresponds to a UFS. Return NULL if no device is found.
624 */
625 struct mount *devtomp(dev)
626 dev_t dev;
627 {
628 struct mount *mp, *nmp;
629
630 for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
631 nmp = mp->mnt_list.cqe_next;
632 if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
633 ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
634 /* mount corresponds to UFS and the device matches one we want */
635 return(mp);
636 }
637 }
638 /* mount structure wasn't found */
639 return(NULL);
640 }
641