coda_vfsops.c revision 1.44 1 /* $NetBSD: coda_vfsops.c,v 1.44 2005/07/02 07:05:27 blymn 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 1.44 2005/07/02 07:05:27 blymn 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 vfs_stdextattrctl,
125 coda_vnodeopv_descs,
126 0
127 };
128
129 VFS_ATTACH(coda_vfsops);
130
131 int
132 coda_vfsopstats_init(void)
133 {
134 int i;
135
136 for (i=0;i<CODA_VFSOPS_SIZE;i++) {
137 coda_vfsopstats[i].opcode = i;
138 coda_vfsopstats[i].entries = 0;
139 coda_vfsopstats[i].sat_intrn = 0;
140 coda_vfsopstats[i].unsat_intrn = 0;
141 coda_vfsopstats[i].gen_intrn = 0;
142 }
143
144 return 0;
145 }
146
147 /*
148 * cfs mount vfsop
149 * Set up mount info record and attach it to vfs struct.
150 */
151 /*ARGSUSED*/
152 int
153 coda_mount(vfsp, path, data, ndp, p)
154 struct mount *vfsp; /* Allocated and initialized by mount(2) */
155 const char *path; /* path covered: ignored by the fs-layer */
156 void *data; /* Need to define a data type for this in netbsd? */
157 struct nameidata *ndp; /* Clobber this to lookup the device name */
158 struct proc *p; /* The ever-famous proc pointer */
159 {
160 struct vnode *dvp;
161 struct cnode *cp;
162 dev_t dev;
163 struct coda_mntinfo *mi;
164 struct vnode *rtvp;
165 const struct cdevsw *cdev;
166 CodaFid rootfid = INVAL_FID;
167 CodaFid ctlfid = CTL_FID;
168 int error;
169
170 if (vfsp->mnt_flag & MNT_GETARGS)
171 return 0;
172 ENTRY;
173
174 coda_vfsopstats_init();
175 coda_vnodeopstats_init();
176
177 MARK_ENTRY(CODA_MOUNT_STATS);
178 if (CODA_MOUNTED(vfsp)) {
179 MARK_INT_FAIL(CODA_MOUNT_STATS);
180 return(EBUSY);
181 }
182
183 /* Validate mount device. Similar to getmdev(). */
184
185 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, data, p);
186 error = namei(ndp);
187 dvp = ndp->ni_vp;
188
189 if (error) {
190 MARK_INT_FAIL(CODA_MOUNT_STATS);
191 return (error);
192 }
193 if (dvp->v_type != VCHR) {
194 MARK_INT_FAIL(CODA_MOUNT_STATS);
195 vrele(dvp);
196 return(ENXIO);
197 }
198 dev = dvp->v_specinfo->si_rdev;
199 vrele(dvp);
200 cdev = cdevsw_lookup(dev);
201 if (cdev == NULL) {
202 MARK_INT_FAIL(CODA_MOUNT_STATS);
203 return(ENXIO);
204 }
205
206 /*
207 * See if the device table matches our expectations.
208 */
209 if (cdev != &vcoda_cdevsw)
210 {
211 MARK_INT_FAIL(CODA_MOUNT_STATS);
212 return(ENXIO);
213 }
214
215 if (minor(dev) >= NVCODA || minor(dev) < 0) {
216 MARK_INT_FAIL(CODA_MOUNT_STATS);
217 return(ENXIO);
218 }
219
220 /*
221 * Initialize the mount record and link it to the vfs struct
222 */
223 mi = &coda_mnttbl[minor(dev)];
224
225 if (!VC_OPEN(&mi->mi_vcomm)) {
226 MARK_INT_FAIL(CODA_MOUNT_STATS);
227 return(ENODEV);
228 }
229
230 /* No initialization (here) of mi_vcomm! */
231 vfsp->mnt_data = mi;
232 vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
233 vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
234 vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
235 vfsp->mnt_stat.f_namemax = MAXNAMLEN;
236 mi->mi_vfsp = vfsp;
237
238 /*
239 * Make a root vnode to placate the Vnode interface, but don't
240 * actually make the CODA_ROOT call to venus until the first call
241 * to coda_root in case a server is down while venus is starting.
242 */
243 cp = make_coda_node(&rootfid, vfsp, VDIR);
244 rtvp = CTOV(cp);
245 rtvp->v_flag |= VROOT;
246
247 /* cp = make_coda_node(&ctlfid, vfsp, VCHR);
248 The above code seems to cause a loop in the cnode links.
249 I don't totally understand when it happens, it is caught
250 when closing down the system.
251 */
252 cp = make_coda_node(&ctlfid, 0, VCHR);
253
254 coda_ctlvp = CTOV(cp);
255
256 /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
257 mi->mi_vfsp = vfsp;
258 mi->mi_rootvp = rtvp;
259
260 /* set filesystem block size */
261 vfsp->mnt_stat.f_bsize = 8192; /* XXX -JJK */
262 vfsp->mnt_stat.f_frsize = 8192; /* XXX -JJK */
263
264 /* error is currently guaranteed to be zero, but in case some
265 code changes... */
266 CODADEBUG(1,
267 myprintf(("coda_mount returned %d\n",error)););
268 if (error)
269 MARK_INT_FAIL(CODA_MOUNT_STATS);
270 else
271 MARK_INT_SAT(CODA_MOUNT_STATS);
272
273 return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE, vfsp,
274 p);
275 }
276
277 int
278 coda_start(vfsp, flags, p)
279 struct mount *vfsp;
280 int flags;
281 struct proc *p;
282 {
283 ENTRY;
284 vftomi(vfsp)->mi_started = 1;
285 return (0);
286 }
287
288 int
289 coda_unmount(vfsp, mntflags, p)
290 struct mount *vfsp;
291 int mntflags;
292 struct proc *p;
293 {
294 struct coda_mntinfo *mi = vftomi(vfsp);
295 int active, error = 0;
296
297 ENTRY;
298 MARK_ENTRY(CODA_UMOUNT_STATS);
299 if (!CODA_MOUNTED(vfsp)) {
300 MARK_INT_FAIL(CODA_UMOUNT_STATS);
301 return(EINVAL);
302 }
303
304 if (mi->mi_vfsp == vfsp) { /* We found the victim */
305 if (!IS_UNMOUNTING(VTOC(mi->mi_rootvp)))
306 return (EBUSY); /* Venus is still running */
307
308 #ifdef DEBUG
309 printf("coda_unmount: ROOT: vp %p, cp %p\n", mi->mi_rootvp, VTOC(mi->mi_rootvp));
310 #endif
311 mi->mi_started = 0;
312
313 vrele(mi->mi_rootvp);
314
315 active = coda_kill(vfsp, NOT_DOWNCALL);
316 mi->mi_rootvp->v_flag &= ~VROOT;
317 error = vflush(mi->mi_vfsp, NULLVP, FORCECLOSE);
318 printf("coda_unmount: active = %d, vflush active %d\n", active, error);
319 error = 0;
320
321 /* I'm going to take this out to allow lookups to go through. I'm
322 * not sure it's important anyway. -- DCS 2/2/94
323 */
324 /* vfsp->VFS_DATA = NULL; */
325
326 /* No more vfsp's to hold onto */
327 mi->mi_vfsp = NULL;
328 mi->mi_rootvp = NULL;
329
330 if (error)
331 MARK_INT_FAIL(CODA_UMOUNT_STATS);
332 else
333 MARK_INT_SAT(CODA_UMOUNT_STATS);
334
335 return(error);
336 }
337 return (EINVAL);
338 }
339
340 /*
341 * find root of cfs
342 */
343 int
344 coda_root(vfsp, vpp)
345 struct mount *vfsp;
346 struct vnode **vpp;
347 {
348 struct coda_mntinfo *mi = vftomi(vfsp);
349 int error;
350 struct proc *p = curproc; /* XXX - bnoble */
351 CodaFid VFid;
352 static const CodaFid invalfid = INVAL_FID;
353
354 ENTRY;
355 MARK_ENTRY(CODA_ROOT_STATS);
356
357 if (vfsp == mi->mi_vfsp) {
358 if (memcmp(&VTOC(mi->mi_rootvp)->c_fid, &invalfid, sizeof(CodaFid)))
359 { /* Found valid root. */
360 *vpp = mi->mi_rootvp;
361 /* On Mach, this is vref. On NetBSD, VOP_LOCK */
362 vref(*vpp);
363 vn_lock(*vpp, LK_EXCLUSIVE);
364 MARK_INT_SAT(CODA_ROOT_STATS);
365 return(0);
366 }
367 }
368
369 error = venus_root(vftomi(vfsp), p->p_cred->pc_ucred, p, &VFid);
370
371 if (!error) {
372 /*
373 * Save the new rootfid in the cnode, and rehash the cnode into the
374 * cnode hash with the new fid key.
375 */
376 coda_unsave(VTOC(mi->mi_rootvp));
377 VTOC(mi->mi_rootvp)->c_fid = VFid;
378 coda_save(VTOC(mi->mi_rootvp));
379
380 *vpp = mi->mi_rootvp;
381 vref(*vpp);
382 vn_lock(*vpp, LK_EXCLUSIVE);
383 MARK_INT_SAT(CODA_ROOT_STATS);
384 goto exit;
385 } else if (error == ENODEV || error == EINTR) {
386 /* Gross hack here! */
387 /*
388 * If Venus fails to respond to the CODA_ROOT call, coda_call returns
389 * ENODEV. Return the uninitialized root vnode to allow vfs
390 * operations such as unmount to continue. Without this hack,
391 * there is no way to do an unmount if Venus dies before a
392 * successful CODA_ROOT call is done. All vnode operations
393 * will fail.
394 */
395 *vpp = mi->mi_rootvp;
396 vref(*vpp);
397 vn_lock(*vpp, LK_EXCLUSIVE);
398 MARK_INT_FAIL(CODA_ROOT_STATS);
399 error = 0;
400 goto exit;
401 } else {
402 CODADEBUG( CODA_ROOT, myprintf(("error %d in CODA_ROOT\n", error)); );
403 MARK_INT_FAIL(CODA_ROOT_STATS);
404
405 goto exit;
406 }
407 exit:
408 return(error);
409 }
410
411 int
412 coda_quotactl(vfsp, cmd, uid, arg, p)
413 struct mount *vfsp;
414 int cmd;
415 uid_t uid;
416 void *arg;
417 struct proc *p;
418 {
419 ENTRY;
420 return (EOPNOTSUPP);
421 }
422
423 /*
424 * Get file system statistics.
425 */
426 int
427 coda_nb_statvfs(vfsp, sbp, p)
428 struct mount *vfsp;
429 struct statvfs *sbp;
430 struct proc *p;
431 {
432 struct coda_statfs fsstat;
433 int error;
434
435 ENTRY;
436 MARK_ENTRY(CODA_STATFS_STATS);
437 if (!CODA_MOUNTED(vfsp)) {
438 /* MARK_INT_FAIL(CODA_STATFS_STATS); */
439 return(EINVAL);
440 }
441
442 /* XXX - what to do about f_flags, others? --bnoble */
443 /* Below This is what AFS does
444 #define NB_SFS_SIZ 0x895440
445 */
446 /* Note: Normal fs's have a bsize of 0x400 == 1024 */
447
448 error = venus_statfs(vftomi(vfsp), p->p_cred->pc_ucred, p, &fsstat);
449
450 if (!error) {
451 sbp->f_bsize = 8192; /* XXX */
452 sbp->f_frsize = 8192; /* XXX */
453 sbp->f_iosize = 8192; /* XXX */
454 sbp->f_blocks = fsstat.f_blocks;
455 sbp->f_bfree = fsstat.f_bfree;
456 sbp->f_bavail = fsstat.f_bavail;
457 sbp->f_bresvd = 0;
458 sbp->f_files = fsstat.f_files;
459 sbp->f_ffree = fsstat.f_ffree;
460 sbp->f_favail = fsstat.f_ffree;
461 sbp->f_fresvd = 0;
462 copy_statvfs_info(sbp, vfsp);
463 }
464
465 MARK_INT_SAT(CODA_STATFS_STATS);
466 return(error);
467 }
468
469 /*
470 * Flush any pending I/O.
471 */
472 int
473 coda_sync(vfsp, waitfor, cred, p)
474 struct mount *vfsp;
475 int waitfor;
476 struct ucred *cred;
477 struct proc *p;
478 {
479 ENTRY;
480 MARK_ENTRY(CODA_SYNC_STATS);
481 MARK_INT_SAT(CODA_SYNC_STATS);
482 return(0);
483 }
484
485 int
486 coda_vget(vfsp, ino, vpp)
487 struct mount *vfsp;
488 ino_t ino;
489 struct vnode **vpp;
490 {
491 ENTRY;
492 return (EOPNOTSUPP);
493 }
494
495 /*
496 * fhtovp is now what vget used to be in 4.3-derived systems. For
497 * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
498 * a type-specific fid.
499 */
500 int
501 coda_fhtovp(vfsp, fhp, nam, vpp, exflagsp, creadanonp)
502 struct mount *vfsp;
503 struct fid *fhp;
504 struct mbuf *nam;
505 struct vnode **vpp;
506 int *exflagsp;
507 struct ucred **creadanonp;
508 {
509 struct cfid *cfid = (struct cfid *)fhp;
510 struct cnode *cp = 0;
511 int error;
512 struct proc *p = curproc; /* XXX -mach */
513 CodaFid VFid;
514 int vtype;
515
516 ENTRY;
517
518 MARK_ENTRY(CODA_VGET_STATS);
519 /* Check for vget of control object. */
520 if (IS_CTL_FID(&cfid->cfid_fid)) {
521 *vpp = coda_ctlvp;
522 vref(coda_ctlvp);
523 MARK_INT_SAT(CODA_VGET_STATS);
524 return(0);
525 }
526
527 error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, p->p_cred->pc_ucred, p, &VFid, &vtype);
528
529 if (error) {
530 CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
531 *vpp = (struct vnode *)0;
532 } else {
533 CODADEBUG(CODA_VGET,
534 myprintf(("vget: %s type %d result %d\n",
535 coda_f2s(&VFid), vtype, error)); )
536
537 cp = make_coda_node(&VFid, vfsp, vtype);
538 *vpp = CTOV(cp);
539 }
540 return(error);
541 }
542
543 int
544 coda_vptofh(vnp, fidp)
545 struct vnode *vnp;
546 struct fid *fidp;
547 {
548 ENTRY;
549 return (EOPNOTSUPP);
550 }
551
552 void
553 coda_init(void)
554 {
555 ENTRY;
556 }
557
558 void
559 coda_done(void)
560 {
561 ENTRY;
562 }
563
564 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
565 {
566 sysctl_createv(clog, 0, NULL, NULL,
567 CTLFLAG_PERMANENT,
568 CTLTYPE_NODE, "vfs", NULL,
569 NULL, 0, NULL, 0,
570 CTL_VFS, CTL_EOL);
571 sysctl_createv(clog, 0, NULL, NULL,
572 CTLFLAG_PERMANENT,
573 CTLTYPE_NODE, "coda",
574 SYSCTL_DESCR("code vfs options"),
575 NULL, 0, NULL, 0,
576 CTL_VFS, 18, CTL_EOL);
577 /*
578 * XXX the "18" above could be dynamic, thereby eliminating
579 * one more instance of the "number to vfs" mapping problem,
580 * but "18" is the order as taken from sys/mount.h
581 */
582
583 /*
584 sysctl_createv(clog, 0, NULL, NULL,
585 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
586 CTLTYPE_INT, "clusterread",
587 SYSCTL_DESCR( anyone? ),
588 NULL, 0, &doclusterread, 0,
589 CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
590 */
591 }
592
593 /*
594 * To allow for greater ease of use, some vnodes may be orphaned when
595 * Venus dies. Certain operations should still be allowed to go
596 * through, but without propagating orphan-ness. So this function will
597 * get a new vnode for the file from the current run of Venus.
598 */
599
600 int
601 getNewVnode(vpp)
602 struct vnode **vpp;
603 {
604 struct cfid cfid;
605 struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
606
607 ENTRY;
608
609 cfid.cfid_len = (short)sizeof(CodaFid);
610 cfid.cfid_fid = VTOC(*vpp)->c_fid; /* Structure assignment. */
611 /* XXX ? */
612
613 /* We're guessing that if set, the 1st element on the list is a
614 * valid vnode to use. If not, return ENODEV as venus is dead.
615 */
616 if (mi->mi_vfsp == NULL)
617 return ENODEV;
618
619 return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
620 NULL, NULL);
621 }
622
623 #include <ufs/ufs/quota.h>
624 #include <ufs/ufs/ufsmount.h>
625 /* get the mount structure corresponding to a given device. Assume
626 * device corresponds to a UFS. Return NULL if no device is found.
627 */
628 struct mount *devtomp(dev)
629 dev_t dev;
630 {
631 struct mount *mp, *nmp;
632
633 for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
634 nmp = mp->mnt_list.cqe_next;
635 if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
636 ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
637 /* mount corresponds to UFS and the device matches one we want */
638 return(mp);
639 }
640 }
641 /* mount structure wasn't found */
642 return(NULL);
643 }
644