xref: /netbsd-src/sys/coda/coda_vfsops.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: coda_vfsops.c,v 1.62 2007/12/08 19:29:37 pooka 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 1.62 2007/12/08 19:29:37 pooka 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     (void *)eopnotsupp,	/* vfs_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 {
158     struct lwp *l = curlwp;
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 default size for the filename buffer.
190      */
191     /* Ensure that namei() doesn't run off the filename buffer */
192     ((char *)data)[*data_len - 1] = 0;
193     NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, data);
194     error = namei(&nd);
195     dvp = nd.ni_vp;
196 
197     if (error) {
198 	MARK_INT_FAIL(CODA_MOUNT_STATS);
199 	return (error);
200     }
201     if (dvp->v_type != VCHR) {
202 	MARK_INT_FAIL(CODA_MOUNT_STATS);
203 	vrele(dvp);
204 	return(ENXIO);
205     }
206     dev = dvp->v_specinfo->si_rdev;
207     vrele(dvp);
208     cdev = cdevsw_lookup(dev);
209     if (cdev == NULL) {
210 	MARK_INT_FAIL(CODA_MOUNT_STATS);
211 	return(ENXIO);
212     }
213 
214     /*
215      * See if the device table matches our expectations.
216      */
217     if (cdev != &vcoda_cdevsw)
218     {
219 	MARK_INT_FAIL(CODA_MOUNT_STATS);
220 	return(ENXIO);
221     }
222 
223     if (minor(dev) >= NVCODA || minor(dev) < 0) {
224 	MARK_INT_FAIL(CODA_MOUNT_STATS);
225 	return(ENXIO);
226     }
227 
228     /*
229      * Initialize the mount record and link it to the vfs struct
230      */
231     mi = &coda_mnttbl[minor(dev)];
232 
233     if (!VC_OPEN(&mi->mi_vcomm)) {
234 	MARK_INT_FAIL(CODA_MOUNT_STATS);
235 	return(ENODEV);
236     }
237 
238     /* No initialization (here) of mi_vcomm! */
239     vfsp->mnt_data = mi;
240     vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
241     vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
242     vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
243     vfsp->mnt_stat.f_namemax = MAXNAMLEN;
244     mi->mi_vfsp = vfsp;
245 
246     /*
247      * Make a root vnode to placate the Vnode interface, but don't
248      * actually make the CODA_ROOT call to venus until the first call
249      * to coda_root in case a server is down while venus is starting.
250      */
251     cp = make_coda_node(&rootfid, vfsp, VDIR);
252     rtvp = CTOV(cp);
253     rtvp->v_vflag |= VV_ROOT;
254 
255 /*  cp = make_coda_node(&ctlfid, vfsp, VCHR);
256     The above code seems to cause a loop in the cnode links.
257     I don't totally understand when it happens, it is caught
258     when closing down the system.
259  */
260     cp = make_coda_node(&ctlfid, 0, VCHR);
261 
262     coda_ctlvp = CTOV(cp);
263 
264     /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
265     mi->mi_vfsp = vfsp;
266     mi->mi_rootvp = rtvp;
267 
268     /* set filesystem block size */
269     vfsp->mnt_stat.f_bsize = 8192;	    /* XXX -JJK */
270     vfsp->mnt_stat.f_frsize = 8192;	    /* XXX -JJK */
271 
272     /* error is currently guaranteed to be zero, but in case some
273        code changes... */
274     CODADEBUG(1,
275 	     myprintf(("coda_mount returned %d\n",error)););
276     if (error)
277 	MARK_INT_FAIL(CODA_MOUNT_STATS);
278     else
279 	MARK_INT_SAT(CODA_MOUNT_STATS);
280 
281     return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE,
282 	vfsp->mnt_op->vfs_name, vfsp, l);
283 }
284 
285 int
286 coda_start(struct mount *vfsp, int flags)
287 {
288     ENTRY;
289     vftomi(vfsp)->mi_started = 1;
290     return (0);
291 }
292 
293 int
294 coda_unmount(struct mount *vfsp, int mntflags)
295 {
296     struct coda_mntinfo *mi = vftomi(vfsp);
297     int active, error = 0;
298 
299     ENTRY;
300     MARK_ENTRY(CODA_UMOUNT_STATS);
301     if (!CODA_MOUNTED(vfsp)) {
302 	MARK_INT_FAIL(CODA_UMOUNT_STATS);
303 	return(EINVAL);
304     }
305 
306     if (mi->mi_vfsp == vfsp) {	/* We found the victim */
307 	if (!IS_UNMOUNTING(VTOC(mi->mi_rootvp)))
308 	    return (EBUSY); 	/* Venus is still running */
309 
310 #ifdef	DEBUG
311 	printf("coda_unmount: ROOT: vp %p, cp %p\n", mi->mi_rootvp, VTOC(mi->mi_rootvp));
312 #endif
313 	mi->mi_started = 0;
314 
315 	vrele(mi->mi_rootvp);
316 
317 	active = coda_kill(vfsp, NOT_DOWNCALL);
318 	mi->mi_rootvp->v_vflag &= ~VV_ROOT;
319 	error = vflush(mi->mi_vfsp, NULLVP, FORCECLOSE);
320 	printf("coda_unmount: active = %d, vflush active %d\n", active, error);
321 	error = 0;
322 
323 	/* I'm going to take this out to allow lookups to go through. I'm
324 	 * not sure it's important anyway. -- DCS 2/2/94
325 	 */
326 	/* vfsp->VFS_DATA = NULL; */
327 
328 	/* No more vfsp's to hold onto */
329 	mi->mi_vfsp = NULL;
330 	mi->mi_rootvp = NULL;
331 
332 	if (error)
333 	    MARK_INT_FAIL(CODA_UMOUNT_STATS);
334 	else
335 	    MARK_INT_SAT(CODA_UMOUNT_STATS);
336 
337 	return(error);
338     }
339     return (EINVAL);
340 }
341 
342 /*
343  * find root of cfs
344  */
345 int
346 coda_root(struct mount *vfsp, struct vnode **vpp)
347 {
348     struct coda_mntinfo *mi = vftomi(vfsp);
349     int error;
350     struct lwp *l = curlwp;    /* 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), l->l_cred, l->l_proc, &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 /*
412  * Get file system statistics.
413  */
414 int
415 coda_nb_statvfs(struct mount *vfsp, struct statvfs *sbp)
416 {
417     struct lwp *l = curlwp;
418     struct coda_statfs fsstat;
419     int error;
420 
421     ENTRY;
422     MARK_ENTRY(CODA_STATFS_STATS);
423     if (!CODA_MOUNTED(vfsp)) {
424 /*	MARK_INT_FAIL(CODA_STATFS_STATS); */
425 	return(EINVAL);
426     }
427 
428     /* XXX - what to do about f_flags, others? --bnoble */
429     /* Below This is what AFS does
430     	#define NB_SFS_SIZ 0x895440
431      */
432     /* Note: Normal fs's have a bsize of 0x400 == 1024 */
433 
434     error = venus_statfs(vftomi(vfsp), l->l_cred, l, &fsstat);
435 
436     if (!error) {
437 	sbp->f_bsize = 8192; /* XXX */
438 	sbp->f_frsize = 8192; /* XXX */
439 	sbp->f_iosize = 8192; /* XXX */
440 	sbp->f_blocks = fsstat.f_blocks;
441 	sbp->f_bfree  = fsstat.f_bfree;
442 	sbp->f_bavail = fsstat.f_bavail;
443 	sbp->f_bresvd = 0;
444 	sbp->f_files  = fsstat.f_files;
445 	sbp->f_ffree  = fsstat.f_ffree;
446 	sbp->f_favail = fsstat.f_ffree;
447 	sbp->f_fresvd = 0;
448 	copy_statvfs_info(sbp, vfsp);
449     }
450 
451     MARK_INT_SAT(CODA_STATFS_STATS);
452     return(error);
453 }
454 
455 /*
456  * Flush any pending I/O.
457  */
458 int
459 coda_sync(struct mount *vfsp, int waitfor,
460     kauth_cred_t cred)
461 {
462     ENTRY;
463     MARK_ENTRY(CODA_SYNC_STATS);
464     MARK_INT_SAT(CODA_SYNC_STATS);
465     return(0);
466 }
467 
468 int
469 coda_vget(struct mount *vfsp, ino_t ino,
470     struct vnode **vpp)
471 {
472     ENTRY;
473     return (EOPNOTSUPP);
474 }
475 
476 /*
477  * fhtovp is now what vget used to be in 4.3-derived systems.  For
478  * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
479  * a type-specific fid.
480  */
481 int
482 coda_fhtovp(struct mount *vfsp, struct fid *fhp, struct mbuf *nam,
483     struct vnode **vpp, int *exflagsp,
484     kauth_cred_t *creadanonp)
485 {
486     struct cfid *cfid = (struct cfid *)fhp;
487     struct cnode *cp = 0;
488     int error;
489     struct lwp *l = curlwp; /* XXX -mach */
490     CodaFid VFid;
491     int vtype;
492 
493     ENTRY;
494 
495     MARK_ENTRY(CODA_VGET_STATS);
496     /* Check for vget of control object. */
497     if (IS_CTL_FID(&cfid->cfid_fid)) {
498 	*vpp = coda_ctlvp;
499 	vref(coda_ctlvp);
500 	MARK_INT_SAT(CODA_VGET_STATS);
501 	return(0);
502     }
503 
504     error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, l->l_cred, l->l_proc, &VFid, &vtype);
505 
506     if (error) {
507 	CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
508 	    *vpp = (struct vnode *)0;
509     } else {
510 	CODADEBUG(CODA_VGET,
511 		 myprintf(("vget: %s type %d result %d\n",
512 			coda_f2s(&VFid), vtype, error)); )
513 
514 	cp = make_coda_node(&VFid, vfsp, vtype);
515 	*vpp = CTOV(cp);
516     }
517     return(error);
518 }
519 
520 int
521 coda_vptofh(struct vnode *vnp, struct fid *fidp)
522 {
523     ENTRY;
524     return (EOPNOTSUPP);
525 }
526 
527 void
528 coda_init(void)
529 {
530     ENTRY;
531 }
532 
533 void
534 coda_done(void)
535 {
536     ENTRY;
537 }
538 
539 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
540 {
541 	sysctl_createv(clog, 0, NULL, NULL,
542 		       CTLFLAG_PERMANENT,
543 		       CTLTYPE_NODE, "vfs", NULL,
544 		       NULL, 0, NULL, 0,
545 		       CTL_VFS, CTL_EOL);
546 	sysctl_createv(clog, 0, NULL, NULL,
547 		       CTLFLAG_PERMANENT,
548 		       CTLTYPE_NODE, "coda",
549 		       SYSCTL_DESCR("code vfs options"),
550 		       NULL, 0, NULL, 0,
551 		       CTL_VFS, 18, CTL_EOL);
552 	/*
553 	 * XXX the "18" above could be dynamic, thereby eliminating
554 	 * one more instance of the "number to vfs" mapping problem,
555 	 * but "18" is the order as taken from sys/mount.h
556 	 */
557 
558 /*
559 	sysctl_createv(clog, 0, NULL, NULL,
560 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
561 		       CTLTYPE_INT, "clusterread",
562 		       SYSCTL_DESCR( anyone? ),
563 		       NULL, 0, &doclusterread, 0,
564 		       CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
565 */
566 }
567 
568 /*
569  * To allow for greater ease of use, some vnodes may be orphaned when
570  * Venus dies.  Certain operations should still be allowed to go
571  * through, but without propagating orphan-ness.  So this function will
572  * get a new vnode for the file from the current run of Venus.
573  */
574 
575 int
576 getNewVnode(struct vnode **vpp)
577 {
578     struct cfid cfid;
579     struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
580 
581     ENTRY;
582 
583     cfid.cfid_len = (short)sizeof(CodaFid);
584     cfid.cfid_fid = VTOC(*vpp)->c_fid;	/* Structure assignment. */
585     /* XXX ? */
586 
587     /* We're guessing that if set, the 1st element on the list is a
588      * valid vnode to use. If not, return ENODEV as venus is dead.
589      */
590     if (mi->mi_vfsp == NULL)
591 	return ENODEV;
592 
593     return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
594 		      NULL, NULL);
595 }
596 
597 #include <ufs/ufs/quota.h>
598 #include <ufs/ufs/ufsmount.h>
599 /* get the mount structure corresponding to a given device.  Assume
600  * device corresponds to a UFS. Return NULL if no device is found.
601  */
602 struct mount *devtomp(dev_t dev)
603 {
604     struct mount *mp, *nmp;
605 
606     for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
607 	nmp = mp->mnt_list.cqe_next;
608 	if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
609 	    ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
610 	    /* mount corresponds to UFS and the device matches one we want */
611 	    return(mp);
612 	}
613     }
614     /* mount structure wasn't found */
615     return(NULL);
616 }
617