xref: /netbsd-src/sys/coda/coda_vfsops.c (revision 7fa608457b817eca6e0977b37f758ae064f3c99c)
1 /*	$NetBSD: coda_vfsops.c,v 1.59 2007/10/10 20:42:21 ad 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 1.59 2007/10/10 20:42:21 ad 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 #include <sys/kauth.h>
67 
68 #include <coda/coda.h>
69 #include <coda/cnode.h>
70 #include <coda/coda_vfsops.h>
71 #include <coda/coda_venus.h>
72 #include <coda/coda_subr.h>
73 #include <coda/coda_opstats.h>
74 /* for VN_RDEV */
75 #include <miscfs/specfs/specdev.h>
76 
77 MALLOC_DEFINE(M_CODA, "coda", "Coda file system structures and tables");
78 
79 int codadebug = 0;
80 
81 int coda_vfsop_print_entry = 0;
82 #define ENTRY if(coda_vfsop_print_entry) myprintf(("Entered %s\n",__func__))
83 
84 struct vnode *coda_ctlvp;
85 struct coda_mntinfo coda_mnttbl[NVCODA]; /* indexed by minor device number */
86 
87 /* structure to keep statistics of internally generated/satisfied calls */
88 
89 struct coda_op_stats coda_vfsopstats[CODA_VFSOPS_SIZE];
90 
91 #define MARK_ENTRY(op) (coda_vfsopstats[op].entries++)
92 #define MARK_INT_SAT(op) (coda_vfsopstats[op].sat_intrn++)
93 #define MARK_INT_FAIL(op) (coda_vfsopstats[op].unsat_intrn++)
94 #define MRAK_INT_GEN(op) (coda_vfsopstats[op].gen_intrn++)
95 
96 extern const struct cdevsw vcoda_cdevsw;
97 extern const struct vnodeopv_desc coda_vnodeop_opv_desc;
98 
99 const struct vnodeopv_desc * const coda_vnodeopv_descs[] = {
100 	&coda_vnodeop_opv_desc,
101 	NULL,
102 };
103 
104 struct vfsops coda_vfsops = {
105     MOUNT_CODA,
106     256,		/* This is the pathname, unlike every other fs */
107     coda_mount,
108     coda_start,
109     coda_unmount,
110     coda_root,
111     coda_quotactl,
112     coda_nb_statvfs,
113     coda_sync,
114     coda_vget,
115     (void *)eopnotsupp,	/* vfs_fhtovp */
116     (void *)eopnotsupp,	/* vfs_vptofh */
117     coda_init,
118     NULL,		/* vfs_reinit */
119     coda_done,
120     (int (*)(void)) eopnotsupp,
121     (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
122     vfs_stdextattrctl,
123     (void *)eopnotsupp,	/* vfs_suspendctl */
124     coda_vnodeopv_descs,
125     0,			/* vfs_refcount */
126     { NULL, NULL },	/* vfs_list */
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     size_t *data_len,
157     struct lwp *l)		/* The ever-famous lwp pointer */
158 {
159     struct nameidata nd;
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 EINVAL;
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     /*
186      * XXX: coda passes the mount device as the entire mount args,
187      * All other fs pass a structure contining a pointer.
188      * In order to get sys_mount() to do the copyin() we've set a
189      * fixed size for the filename buffer.
190      */
191     NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, data, l);
192     error = namei(&nd);
193     dvp = nd.ni_vp;
194 
195     if (error) {
196 	MARK_INT_FAIL(CODA_MOUNT_STATS);
197 	return (error);
198     }
199     if (dvp->v_type != VCHR) {
200 	MARK_INT_FAIL(CODA_MOUNT_STATS);
201 	vrele(dvp);
202 	return(ENXIO);
203     }
204     dev = dvp->v_specinfo->si_rdev;
205     vrele(dvp);
206     cdev = cdevsw_lookup(dev);
207     if (cdev == NULL) {
208 	MARK_INT_FAIL(CODA_MOUNT_STATS);
209 	return(ENXIO);
210     }
211 
212     /*
213      * See if the device table matches our expectations.
214      */
215     if (cdev != &vcoda_cdevsw)
216     {
217 	MARK_INT_FAIL(CODA_MOUNT_STATS);
218 	return(ENXIO);
219     }
220 
221     if (minor(dev) >= NVCODA || minor(dev) < 0) {
222 	MARK_INT_FAIL(CODA_MOUNT_STATS);
223 	return(ENXIO);
224     }
225 
226     /*
227      * Initialize the mount record and link it to the vfs struct
228      */
229     mi = &coda_mnttbl[minor(dev)];
230 
231     if (!VC_OPEN(&mi->mi_vcomm)) {
232 	MARK_INT_FAIL(CODA_MOUNT_STATS);
233 	return(ENODEV);
234     }
235 
236     /* No initialization (here) of mi_vcomm! */
237     vfsp->mnt_data = mi;
238     vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
239     vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
240     vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
241     vfsp->mnt_stat.f_namemax = MAXNAMLEN;
242     mi->mi_vfsp = vfsp;
243 
244     /*
245      * Make a root vnode to placate the Vnode interface, but don't
246      * actually make the CODA_ROOT call to venus until the first call
247      * to coda_root in case a server is down while venus is starting.
248      */
249     cp = make_coda_node(&rootfid, vfsp, VDIR);
250     rtvp = CTOV(cp);
251     rtvp->v_vflag |= VV_ROOT;
252 
253 /*  cp = make_coda_node(&ctlfid, vfsp, VCHR);
254     The above code seems to cause a loop in the cnode links.
255     I don't totally understand when it happens, it is caught
256     when closing down the system.
257  */
258     cp = make_coda_node(&ctlfid, 0, VCHR);
259 
260     coda_ctlvp = CTOV(cp);
261 
262     /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
263     mi->mi_vfsp = vfsp;
264     mi->mi_rootvp = rtvp;
265 
266     /* set filesystem block size */
267     vfsp->mnt_stat.f_bsize = 8192;	    /* XXX -JJK */
268     vfsp->mnt_stat.f_frsize = 8192;	    /* XXX -JJK */
269 
270     /* error is currently guaranteed to be zero, but in case some
271        code changes... */
272     CODADEBUG(1,
273 	     myprintf(("coda_mount returned %d\n",error)););
274     if (error)
275 	MARK_INT_FAIL(CODA_MOUNT_STATS);
276     else
277 	MARK_INT_SAT(CODA_MOUNT_STATS);
278 
279     return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE,
280 	vfsp->mnt_op->vfs_name, vfsp, l);
281 }
282 
283 int
284 coda_start(struct mount *vfsp, int flags, struct lwp *l)
285 {
286     ENTRY;
287     vftomi(vfsp)->mi_started = 1;
288     return (0);
289 }
290 
291 int
292 coda_unmount(struct mount *vfsp, int mntflags, struct lwp *l)
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_vflag &= ~VV_ROOT;
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(struct mount *vfsp, struct vnode **vpp)
345 {
346     struct coda_mntinfo *mi = vftomi(vfsp);
347     int error;
348     struct lwp *l = curlwp;    /* 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), l->l_cred, l->l_proc, &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(struct mount *vfsp, int cmd, uid_t uid,
411     void *arg, struct lwp *l)
412 {
413     ENTRY;
414     return (EOPNOTSUPP);
415 }
416 
417 /*
418  * Get file system statistics.
419  */
420 int
421 coda_nb_statvfs(struct mount *vfsp, struct statvfs *sbp, struct lwp *l)
422 {
423     struct coda_statfs fsstat;
424     int error;
425 
426     ENTRY;
427     MARK_ENTRY(CODA_STATFS_STATS);
428     if (!CODA_MOUNTED(vfsp)) {
429 /*	MARK_INT_FAIL(CODA_STATFS_STATS); */
430 	return(EINVAL);
431     }
432 
433     /* XXX - what to do about f_flags, others? --bnoble */
434     /* Below This is what AFS does
435     	#define NB_SFS_SIZ 0x895440
436      */
437     /* Note: Normal fs's have a bsize of 0x400 == 1024 */
438 
439     error = venus_statfs(vftomi(vfsp), l->l_cred, l, &fsstat);
440 
441     if (!error) {
442 	sbp->f_bsize = 8192; /* XXX */
443 	sbp->f_frsize = 8192; /* XXX */
444 	sbp->f_iosize = 8192; /* XXX */
445 	sbp->f_blocks = fsstat.f_blocks;
446 	sbp->f_bfree  = fsstat.f_bfree;
447 	sbp->f_bavail = fsstat.f_bavail;
448 	sbp->f_bresvd = 0;
449 	sbp->f_files  = fsstat.f_files;
450 	sbp->f_ffree  = fsstat.f_ffree;
451 	sbp->f_favail = fsstat.f_ffree;
452 	sbp->f_fresvd = 0;
453 	copy_statvfs_info(sbp, vfsp);
454     }
455 
456     MARK_INT_SAT(CODA_STATFS_STATS);
457     return(error);
458 }
459 
460 /*
461  * Flush any pending I/O.
462  */
463 int
464 coda_sync(struct mount *vfsp, int waitfor,
465     kauth_cred_t cred, struct lwp *l)
466 {
467     ENTRY;
468     MARK_ENTRY(CODA_SYNC_STATS);
469     MARK_INT_SAT(CODA_SYNC_STATS);
470     return(0);
471 }
472 
473 int
474 coda_vget(struct mount *vfsp, ino_t ino,
475     struct vnode **vpp)
476 {
477     ENTRY;
478     return (EOPNOTSUPP);
479 }
480 
481 /*
482  * fhtovp is now what vget used to be in 4.3-derived systems.  For
483  * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
484  * a type-specific fid.
485  */
486 int
487 coda_fhtovp(struct mount *vfsp, struct fid *fhp, struct mbuf *nam,
488     struct vnode **vpp, int *exflagsp,
489     kauth_cred_t *creadanonp)
490 {
491     struct cfid *cfid = (struct cfid *)fhp;
492     struct cnode *cp = 0;
493     int error;
494     struct lwp *l = curlwp; /* XXX -mach */
495     CodaFid VFid;
496     int vtype;
497 
498     ENTRY;
499 
500     MARK_ENTRY(CODA_VGET_STATS);
501     /* Check for vget of control object. */
502     if (IS_CTL_FID(&cfid->cfid_fid)) {
503 	*vpp = coda_ctlvp;
504 	vref(coda_ctlvp);
505 	MARK_INT_SAT(CODA_VGET_STATS);
506 	return(0);
507     }
508 
509     error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, l->l_cred, l->l_proc, &VFid, &vtype);
510 
511     if (error) {
512 	CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
513 	    *vpp = (struct vnode *)0;
514     } else {
515 	CODADEBUG(CODA_VGET,
516 		 myprintf(("vget: %s type %d result %d\n",
517 			coda_f2s(&VFid), vtype, error)); )
518 
519 	cp = make_coda_node(&VFid, vfsp, vtype);
520 	*vpp = CTOV(cp);
521     }
522     return(error);
523 }
524 
525 int
526 coda_vptofh(struct vnode *vnp, struct fid *fidp)
527 {
528     ENTRY;
529     return (EOPNOTSUPP);
530 }
531 
532 void
533 coda_init(void)
534 {
535     ENTRY;
536 }
537 
538 void
539 coda_done(void)
540 {
541     ENTRY;
542 }
543 
544 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
545 {
546 	sysctl_createv(clog, 0, NULL, NULL,
547 		       CTLFLAG_PERMANENT,
548 		       CTLTYPE_NODE, "vfs", NULL,
549 		       NULL, 0, NULL, 0,
550 		       CTL_VFS, CTL_EOL);
551 	sysctl_createv(clog, 0, NULL, NULL,
552 		       CTLFLAG_PERMANENT,
553 		       CTLTYPE_NODE, "coda",
554 		       SYSCTL_DESCR("code vfs options"),
555 		       NULL, 0, NULL, 0,
556 		       CTL_VFS, 18, CTL_EOL);
557 	/*
558 	 * XXX the "18" above could be dynamic, thereby eliminating
559 	 * one more instance of the "number to vfs" mapping problem,
560 	 * but "18" is the order as taken from sys/mount.h
561 	 */
562 
563 /*
564 	sysctl_createv(clog, 0, NULL, NULL,
565 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
566 		       CTLTYPE_INT, "clusterread",
567 		       SYSCTL_DESCR( anyone? ),
568 		       NULL, 0, &doclusterread, 0,
569 		       CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
570 */
571 }
572 
573 /*
574  * To allow for greater ease of use, some vnodes may be orphaned when
575  * Venus dies.  Certain operations should still be allowed to go
576  * through, but without propagating orphan-ness.  So this function will
577  * get a new vnode for the file from the current run of Venus.
578  */
579 
580 int
581 getNewVnode(struct vnode **vpp)
582 {
583     struct cfid cfid;
584     struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
585 
586     ENTRY;
587 
588     cfid.cfid_len = (short)sizeof(CodaFid);
589     cfid.cfid_fid = VTOC(*vpp)->c_fid;	/* Structure assignment. */
590     /* XXX ? */
591 
592     /* We're guessing that if set, the 1st element on the list is a
593      * valid vnode to use. If not, return ENODEV as venus is dead.
594      */
595     if (mi->mi_vfsp == NULL)
596 	return ENODEV;
597 
598     return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
599 		      NULL, NULL);
600 }
601 
602 #include <ufs/ufs/quota.h>
603 #include <ufs/ufs/ufsmount.h>
604 /* get the mount structure corresponding to a given device.  Assume
605  * device corresponds to a UFS. Return NULL if no device is found.
606  */
607 struct mount *devtomp(dev_t dev)
608 {
609     struct mount *mp, *nmp;
610 
611     for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
612 	nmp = mp->mnt_list.cqe_next;
613 	if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
614 	    ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
615 	    /* mount corresponds to UFS and the device matches one we want */
616 	    return(mp);
617 	}
618     }
619     /* mount structure wasn't found */
620     return(NULL);
621 }
622