coda_vfsops.c revision 1.45 1 /* $NetBSD: coda_vfsops.c,v 1.45 2005/08/30 22:24:11 xtraeme 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 1.45 2005/08/30 22:24:11 xtraeme 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(struct mount *vfsp, /* Allocated and initialized by mount(2) */
154 const char *path, /* path covered: ignored by the fs-layer */
155 void *data, /* Need to define a data type for this in netbsd? */
156 struct nameidata *ndp, /* Clobber this to lookup the device name */
157 struct proc *p) /* The ever-famous proc pointer */
158 {
159 struct vnode *dvp;
160 struct cnode *cp;
161 dev_t dev;
162 struct coda_mntinfo *mi;
163 struct vnode *rtvp;
164 const struct cdevsw *cdev;
165 CodaFid rootfid = INVAL_FID;
166 CodaFid ctlfid = CTL_FID;
167 int error;
168
169 if (vfsp->mnt_flag & MNT_GETARGS)
170 return 0;
171 ENTRY;
172
173 coda_vfsopstats_init();
174 coda_vnodeopstats_init();
175
176 MARK_ENTRY(CODA_MOUNT_STATS);
177 if (CODA_MOUNTED(vfsp)) {
178 MARK_INT_FAIL(CODA_MOUNT_STATS);
179 return(EBUSY);
180 }
181
182 /* Validate mount device. Similar to getmdev(). */
183
184 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, data, p);
185 error = namei(ndp);
186 dvp = ndp->ni_vp;
187
188 if (error) {
189 MARK_INT_FAIL(CODA_MOUNT_STATS);
190 return (error);
191 }
192 if (dvp->v_type != VCHR) {
193 MARK_INT_FAIL(CODA_MOUNT_STATS);
194 vrele(dvp);
195 return(ENXIO);
196 }
197 dev = dvp->v_specinfo->si_rdev;
198 vrele(dvp);
199 cdev = cdevsw_lookup(dev);
200 if (cdev == NULL) {
201 MARK_INT_FAIL(CODA_MOUNT_STATS);
202 return(ENXIO);
203 }
204
205 /*
206 * See if the device table matches our expectations.
207 */
208 if (cdev != &vcoda_cdevsw)
209 {
210 MARK_INT_FAIL(CODA_MOUNT_STATS);
211 return(ENXIO);
212 }
213
214 if (minor(dev) >= NVCODA || minor(dev) < 0) {
215 MARK_INT_FAIL(CODA_MOUNT_STATS);
216 return(ENXIO);
217 }
218
219 /*
220 * Initialize the mount record and link it to the vfs struct
221 */
222 mi = &coda_mnttbl[minor(dev)];
223
224 if (!VC_OPEN(&mi->mi_vcomm)) {
225 MARK_INT_FAIL(CODA_MOUNT_STATS);
226 return(ENODEV);
227 }
228
229 /* No initialization (here) of mi_vcomm! */
230 vfsp->mnt_data = mi;
231 vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
232 vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
233 vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
234 vfsp->mnt_stat.f_namemax = MAXNAMLEN;
235 mi->mi_vfsp = vfsp;
236
237 /*
238 * Make a root vnode to placate the Vnode interface, but don't
239 * actually make the CODA_ROOT call to venus until the first call
240 * to coda_root in case a server is down while venus is starting.
241 */
242 cp = make_coda_node(&rootfid, vfsp, VDIR);
243 rtvp = CTOV(cp);
244 rtvp->v_flag |= VROOT;
245
246 /* cp = make_coda_node(&ctlfid, vfsp, VCHR);
247 The above code seems to cause a loop in the cnode links.
248 I don't totally understand when it happens, it is caught
249 when closing down the system.
250 */
251 cp = make_coda_node(&ctlfid, 0, VCHR);
252
253 coda_ctlvp = CTOV(cp);
254
255 /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
256 mi->mi_vfsp = vfsp;
257 mi->mi_rootvp = rtvp;
258
259 /* set filesystem block size */
260 vfsp->mnt_stat.f_bsize = 8192; /* XXX -JJK */
261 vfsp->mnt_stat.f_frsize = 8192; /* XXX -JJK */
262
263 /* error is currently guaranteed to be zero, but in case some
264 code changes... */
265 CODADEBUG(1,
266 myprintf(("coda_mount returned %d\n",error)););
267 if (error)
268 MARK_INT_FAIL(CODA_MOUNT_STATS);
269 else
270 MARK_INT_SAT(CODA_MOUNT_STATS);
271
272 return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE, vfsp,
273 p);
274 }
275
276 int
277 coda_start(struct mount *vfsp, int flags, struct proc *p)
278 {
279 ENTRY;
280 vftomi(vfsp)->mi_started = 1;
281 return (0);
282 }
283
284 int
285 coda_unmount(struct mount *vfsp, int mntflags, struct proc *p)
286 {
287 struct coda_mntinfo *mi = vftomi(vfsp);
288 int active, error = 0;
289
290 ENTRY;
291 MARK_ENTRY(CODA_UMOUNT_STATS);
292 if (!CODA_MOUNTED(vfsp)) {
293 MARK_INT_FAIL(CODA_UMOUNT_STATS);
294 return(EINVAL);
295 }
296
297 if (mi->mi_vfsp == vfsp) { /* We found the victim */
298 if (!IS_UNMOUNTING(VTOC(mi->mi_rootvp)))
299 return (EBUSY); /* Venus is still running */
300
301 #ifdef DEBUG
302 printf("coda_unmount: ROOT: vp %p, cp %p\n", mi->mi_rootvp, VTOC(mi->mi_rootvp));
303 #endif
304 mi->mi_started = 0;
305
306 vrele(mi->mi_rootvp);
307
308 active = coda_kill(vfsp, NOT_DOWNCALL);
309 mi->mi_rootvp->v_flag &= ~VROOT;
310 error = vflush(mi->mi_vfsp, NULLVP, FORCECLOSE);
311 printf("coda_unmount: active = %d, vflush active %d\n", active, error);
312 error = 0;
313
314 /* I'm going to take this out to allow lookups to go through. I'm
315 * not sure it's important anyway. -- DCS 2/2/94
316 */
317 /* vfsp->VFS_DATA = NULL; */
318
319 /* No more vfsp's to hold onto */
320 mi->mi_vfsp = NULL;
321 mi->mi_rootvp = NULL;
322
323 if (error)
324 MARK_INT_FAIL(CODA_UMOUNT_STATS);
325 else
326 MARK_INT_SAT(CODA_UMOUNT_STATS);
327
328 return(error);
329 }
330 return (EINVAL);
331 }
332
333 /*
334 * find root of cfs
335 */
336 int
337 coda_root(struct mount *vfsp, struct vnode **vpp)
338 {
339 struct coda_mntinfo *mi = vftomi(vfsp);
340 int error;
341 struct proc *p = curproc; /* XXX - bnoble */
342 CodaFid VFid;
343 static const CodaFid invalfid = INVAL_FID;
344
345 ENTRY;
346 MARK_ENTRY(CODA_ROOT_STATS);
347
348 if (vfsp == mi->mi_vfsp) {
349 if (memcmp(&VTOC(mi->mi_rootvp)->c_fid, &invalfid, sizeof(CodaFid)))
350 { /* Found valid root. */
351 *vpp = mi->mi_rootvp;
352 /* On Mach, this is vref. On NetBSD, VOP_LOCK */
353 vref(*vpp);
354 vn_lock(*vpp, LK_EXCLUSIVE);
355 MARK_INT_SAT(CODA_ROOT_STATS);
356 return(0);
357 }
358 }
359
360 error = venus_root(vftomi(vfsp), p->p_cred->pc_ucred, p, &VFid);
361
362 if (!error) {
363 /*
364 * Save the new rootfid in the cnode, and rehash the cnode into the
365 * cnode hash with the new fid key.
366 */
367 coda_unsave(VTOC(mi->mi_rootvp));
368 VTOC(mi->mi_rootvp)->c_fid = VFid;
369 coda_save(VTOC(mi->mi_rootvp));
370
371 *vpp = mi->mi_rootvp;
372 vref(*vpp);
373 vn_lock(*vpp, LK_EXCLUSIVE);
374 MARK_INT_SAT(CODA_ROOT_STATS);
375 goto exit;
376 } else if (error == ENODEV || error == EINTR) {
377 /* Gross hack here! */
378 /*
379 * If Venus fails to respond to the CODA_ROOT call, coda_call returns
380 * ENODEV. Return the uninitialized root vnode to allow vfs
381 * operations such as unmount to continue. Without this hack,
382 * there is no way to do an unmount if Venus dies before a
383 * successful CODA_ROOT call is done. All vnode operations
384 * will fail.
385 */
386 *vpp = mi->mi_rootvp;
387 vref(*vpp);
388 vn_lock(*vpp, LK_EXCLUSIVE);
389 MARK_INT_FAIL(CODA_ROOT_STATS);
390 error = 0;
391 goto exit;
392 } else {
393 CODADEBUG( CODA_ROOT, myprintf(("error %d in CODA_ROOT\n", error)); );
394 MARK_INT_FAIL(CODA_ROOT_STATS);
395
396 goto exit;
397 }
398 exit:
399 return(error);
400 }
401
402 int
403 coda_quotactl(struct mount *vfsp, int cmd, uid_t uid, void *arg,
404 struct proc *p)
405 {
406 ENTRY;
407 return (EOPNOTSUPP);
408 }
409
410 /*
411 * Get file system statistics.
412 */
413 int
414 coda_nb_statvfs(struct mount *vfsp, struct statvfs *sbp, struct proc *p)
415 {
416 struct coda_statfs fsstat;
417 int error;
418
419 ENTRY;
420 MARK_ENTRY(CODA_STATFS_STATS);
421 if (!CODA_MOUNTED(vfsp)) {
422 /* MARK_INT_FAIL(CODA_STATFS_STATS); */
423 return(EINVAL);
424 }
425
426 /* XXX - what to do about f_flags, others? --bnoble */
427 /* Below This is what AFS does
428 #define NB_SFS_SIZ 0x895440
429 */
430 /* Note: Normal fs's have a bsize of 0x400 == 1024 */
431
432 error = venus_statfs(vftomi(vfsp), p->p_cred->pc_ucred, p, &fsstat);
433
434 if (!error) {
435 sbp->f_bsize = 8192; /* XXX */
436 sbp->f_frsize = 8192; /* XXX */
437 sbp->f_iosize = 8192; /* XXX */
438 sbp->f_blocks = fsstat.f_blocks;
439 sbp->f_bfree = fsstat.f_bfree;
440 sbp->f_bavail = fsstat.f_bavail;
441 sbp->f_bresvd = 0;
442 sbp->f_files = fsstat.f_files;
443 sbp->f_ffree = fsstat.f_ffree;
444 sbp->f_favail = fsstat.f_ffree;
445 sbp->f_fresvd = 0;
446 copy_statvfs_info(sbp, vfsp);
447 }
448
449 MARK_INT_SAT(CODA_STATFS_STATS);
450 return(error);
451 }
452
453 /*
454 * Flush any pending I/O.
455 */
456 int
457 coda_sync(struct mount *vfsp, int waitfor, struct ucred *cred, struct proc *p)
458 {
459 ENTRY;
460 MARK_ENTRY(CODA_SYNC_STATS);
461 MARK_INT_SAT(CODA_SYNC_STATS);
462 return(0);
463 }
464
465 int
466 coda_vget(struct mount *vfsp, ino_t ino, struct vnode **vpp)
467 {
468 ENTRY;
469 return (EOPNOTSUPP);
470 }
471
472 /*
473 * fhtovp is now what vget used to be in 4.3-derived systems. For
474 * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
475 * a type-specific fid.
476 */
477 int
478 coda_fhtovp(struct mount *vfsp, struct fid *fhp, struct mbuf *nam,
479 struct vnode **vpp, int *exflagsp, struct ucred **creadanonp)
480 {
481 struct cfid *cfid = (struct cfid *)fhp;
482 struct cnode *cp = 0;
483 int error;
484 struct proc *p = curproc; /* XXX -mach */
485 CodaFid VFid;
486 int vtype;
487
488 ENTRY;
489
490 MARK_ENTRY(CODA_VGET_STATS);
491 /* Check for vget of control object. */
492 if (IS_CTL_FID(&cfid->cfid_fid)) {
493 *vpp = coda_ctlvp;
494 vref(coda_ctlvp);
495 MARK_INT_SAT(CODA_VGET_STATS);
496 return(0);
497 }
498
499 error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, p->p_cred->pc_ucred, p, &VFid, &vtype);
500
501 if (error) {
502 CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
503 *vpp = (struct vnode *)0;
504 } else {
505 CODADEBUG(CODA_VGET,
506 myprintf(("vget: %s type %d result %d\n",
507 coda_f2s(&VFid), vtype, error)); )
508
509 cp = make_coda_node(&VFid, vfsp, vtype);
510 *vpp = CTOV(cp);
511 }
512 return(error);
513 }
514
515 int
516 coda_vptofh(struct vnode *vnp, struct fid *fidp)
517 {
518 ENTRY;
519 return (EOPNOTSUPP);
520 }
521
522 void
523 coda_init(void)
524 {
525 ENTRY;
526 }
527
528 void
529 coda_done(void)
530 {
531 ENTRY;
532 }
533
534 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
535 {
536 sysctl_createv(clog, 0, NULL, NULL,
537 CTLFLAG_PERMANENT,
538 CTLTYPE_NODE, "vfs", NULL,
539 NULL, 0, NULL, 0,
540 CTL_VFS, CTL_EOL);
541 sysctl_createv(clog, 0, NULL, NULL,
542 CTLFLAG_PERMANENT,
543 CTLTYPE_NODE, "coda",
544 SYSCTL_DESCR("code vfs options"),
545 NULL, 0, NULL, 0,
546 CTL_VFS, 18, CTL_EOL);
547 /*
548 * XXX the "18" above could be dynamic, thereby eliminating
549 * one more instance of the "number to vfs" mapping problem,
550 * but "18" is the order as taken from sys/mount.h
551 */
552
553 /*
554 sysctl_createv(clog, 0, NULL, NULL,
555 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
556 CTLTYPE_INT, "clusterread",
557 SYSCTL_DESCR( anyone? ),
558 NULL, 0, &doclusterread, 0,
559 CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
560 */
561 }
562
563 /*
564 * To allow for greater ease of use, some vnodes may be orphaned when
565 * Venus dies. Certain operations should still be allowed to go
566 * through, but without propagating orphan-ness. So this function will
567 * get a new vnode for the file from the current run of Venus.
568 */
569
570 int
571 getNewVnode(struct vnode **vpp)
572 {
573 struct cfid cfid;
574 struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
575
576 ENTRY;
577
578 cfid.cfid_len = (short)sizeof(CodaFid);
579 cfid.cfid_fid = VTOC(*vpp)->c_fid; /* Structure assignment. */
580 /* XXX ? */
581
582 /* We're guessing that if set, the 1st element on the list is a
583 * valid vnode to use. If not, return ENODEV as venus is dead.
584 */
585 if (mi->mi_vfsp == NULL)
586 return ENODEV;
587
588 return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
589 NULL, NULL);
590 }
591
592 #include <ufs/ufs/quota.h>
593 #include <ufs/ufs/ufsmount.h>
594 /* get the mount structure corresponding to a given device. Assume
595 * device corresponds to a UFS. Return NULL if no device is found.
596 */
597 struct mount *devtomp(dev_t dev)
598 {
599 struct mount *mp, *nmp;
600
601 for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
602 nmp = mp->mnt_list.cqe_next;
603 if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
604 ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
605 /* mount corresponds to UFS and the device matches one we want */
606 return(mp);
607 }
608 }
609 /* mount structure wasn't found */
610 return(NULL);
611 }
612