xref: /netbsd-src/sys/coda/coda_vfsops.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: coda_vfsops.c,v 1.69 2009/06/29 05:08:15 dholland 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 1.69 2009/06/29 05:08:15 dholland Exp $");
49 
50 #ifndef _KERNEL_OPT
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 #include <sys/module.h>
68 
69 #include <coda/coda.h>
70 #include <coda/cnode.h>
71 #include <coda/coda_vfsops.h>
72 #include <coda/coda_venus.h>
73 #include <coda/coda_subr.h>
74 #include <coda/coda_opstats.h>
75 /* for VN_RDEV */
76 #include <miscfs/specfs/specdev.h>
77 #include <miscfs/genfs/genfs.h>
78 
79 MODULE(MODULE_CLASS_VFS, coda, NULL);
80 
81 MALLOC_DEFINE(M_CODA, "coda", "Coda file system structures and tables");
82 
83 int codadebug = 0;
84 
85 int coda_vfsop_print_entry = 0;
86 #define ENTRY if(coda_vfsop_print_entry) myprintf(("Entered %s\n",__func__))
87 
88 struct vnode *coda_ctlvp;
89 struct coda_mntinfo coda_mnttbl[NVCODA]; /* indexed by minor device number */
90 
91 /* structure to keep statistics of internally generated/satisfied calls */
92 
93 struct coda_op_stats coda_vfsopstats[CODA_VFSOPS_SIZE];
94 
95 #define MARK_ENTRY(op) (coda_vfsopstats[op].entries++)
96 #define MARK_INT_SAT(op) (coda_vfsopstats[op].sat_intrn++)
97 #define MARK_INT_FAIL(op) (coda_vfsopstats[op].unsat_intrn++)
98 #define MRAK_INT_GEN(op) (coda_vfsopstats[op].gen_intrn++)
99 
100 extern const struct cdevsw vcoda_cdevsw;
101 extern const struct vnodeopv_desc coda_vnodeop_opv_desc;
102 
103 const struct vnodeopv_desc * const coda_vnodeopv_descs[] = {
104 	&coda_vnodeop_opv_desc,
105 	NULL,
106 };
107 
108 struct vfsops coda_vfsops = {
109     MOUNT_CODA,
110     256,		/* This is the pathname, unlike every other fs */
111     coda_mount,
112     coda_start,
113     coda_unmount,
114     coda_root,
115     (void *)eopnotsupp,	/* vfs_quotactl */
116     coda_nb_statvfs,
117     coda_sync,
118     coda_vget,
119     (void *)eopnotsupp,	/* vfs_fhtovp */
120     (void *)eopnotsupp,	/* vfs_vptofh */
121     coda_init,
122     NULL,		/* vfs_reinit */
123     coda_done,
124     (int (*)(void)) eopnotsupp,
125     (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
126     vfs_stdextattrctl,
127     (void *)eopnotsupp,	/* vfs_suspendctl */
128     genfs_renamelock_enter,
129     genfs_renamelock_exit,
130 	(void *)eopnotsupp,
131     coda_vnodeopv_descs,
132     0,			/* vfs_refcount */
133     { NULL, NULL },	/* vfs_list */
134 };
135 
136 static int
137 coda_modcmd(modcmd_t cmd, void *arg)
138 {
139 
140 	switch (cmd) {
141 	case MODULE_CMD_INIT:
142 		return vfs_attach(&coda_vfsops);
143 	case MODULE_CMD_FINI:
144 		return vfs_detach(&coda_vfsops);
145 	default:
146 		return ENOTTY;
147 	}
148 }
149 
150 int
151 coda_vfsopstats_init(void)
152 {
153 	int i;
154 
155 	for (i=0;i<CODA_VFSOPS_SIZE;i++) {
156 		coda_vfsopstats[i].opcode = i;
157 		coda_vfsopstats[i].entries = 0;
158 		coda_vfsopstats[i].sat_intrn = 0;
159 		coda_vfsopstats[i].unsat_intrn = 0;
160 		coda_vfsopstats[i].gen_intrn = 0;
161 	}
162 
163 	return 0;
164 }
165 
166 /*
167  * cfs mount vfsop
168  * Set up mount info record and attach it to vfs struct.
169  */
170 /*ARGSUSED*/
171 int
172 coda_mount(struct mount *vfsp,	/* Allocated and initialized by mount(2) */
173     const char *path,	/* path covered: ignored by the fs-layer */
174     void *data,		/* Need to define a data type for this in netbsd? */
175     size_t *data_len)
176 {
177     struct lwp *l = curlwp;
178     struct vnode *dvp;
179     struct cnode *cp;
180     dev_t dev;
181     struct coda_mntinfo *mi;
182     struct vnode *rtvp;
183     const struct cdevsw *cdev;
184     CodaFid rootfid = INVAL_FID;
185     CodaFid ctlfid = CTL_FID;
186     int error;
187 
188     if (vfsp->mnt_flag & MNT_GETARGS)
189 	return EINVAL;
190     ENTRY;
191 
192     coda_vfsopstats_init();
193     coda_vnodeopstats_init();
194 
195     MARK_ENTRY(CODA_MOUNT_STATS);
196     if (CODA_MOUNTED(vfsp)) {
197 	MARK_INT_FAIL(CODA_MOUNT_STATS);
198 	return(EBUSY);
199     }
200 
201     /* Validate mount device.  Similar to getmdev(). */
202 
203     /*
204      * XXX: coda passes the mount device as the entire mount args,
205      * All other fs pass a structure contining a pointer.
206      * In order to get sys_mount() to do the copyin() we've set a
207      * fixed default size for the filename buffer.
208      */
209     /* Ensure that namei() doesn't run off the filename buffer */
210     ((char *)data)[*data_len - 1] = 0;
211     error = namei_simple_kernel((char *)data, NSM_FOLLOW_NOEMULROOT,
212 		&dvp);
213 
214     if (error) {
215 	MARK_INT_FAIL(CODA_MOUNT_STATS);
216 	return (error);
217     }
218     if (dvp->v_type != VCHR) {
219 	MARK_INT_FAIL(CODA_MOUNT_STATS);
220 	vrele(dvp);
221 	return(ENXIO);
222     }
223     dev = dvp->v_rdev;
224     vrele(dvp);
225     cdev = cdevsw_lookup(dev);
226     if (cdev == NULL) {
227 	MARK_INT_FAIL(CODA_MOUNT_STATS);
228 	return(ENXIO);
229     }
230 
231     /*
232      * See if the device table matches our expectations.
233      */
234     if (cdev != &vcoda_cdevsw)
235     {
236 	MARK_INT_FAIL(CODA_MOUNT_STATS);
237 	return(ENXIO);
238     }
239 
240     if (minor(dev) >= NVCODA) {
241 	MARK_INT_FAIL(CODA_MOUNT_STATS);
242 	return(ENXIO);
243     }
244 
245     /*
246      * Initialize the mount record and link it to the vfs struct
247      */
248     mi = &coda_mnttbl[minor(dev)];
249 
250     if (!VC_OPEN(&mi->mi_vcomm)) {
251 	MARK_INT_FAIL(CODA_MOUNT_STATS);
252 	return(ENODEV);
253     }
254 
255     /* No initialization (here) of mi_vcomm! */
256     vfsp->mnt_data = mi;
257     vfsp->mnt_stat.f_fsidx.__fsid_val[0] = 0;
258     vfsp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_CODA);
259     vfsp->mnt_stat.f_fsid = vfsp->mnt_stat.f_fsidx.__fsid_val[0];
260     vfsp->mnt_stat.f_namemax = MAXNAMLEN;
261     mi->mi_vfsp = vfsp;
262 
263     /*
264      * Make a root vnode to placate the Vnode interface, but don't
265      * actually make the CODA_ROOT call to venus until the first call
266      * to coda_root in case a server is down while venus is starting.
267      */
268     cp = make_coda_node(&rootfid, vfsp, VDIR);
269     rtvp = CTOV(cp);
270     rtvp->v_vflag |= VV_ROOT;
271 
272 /*  cp = make_coda_node(&ctlfid, vfsp, VCHR);
273     The above code seems to cause a loop in the cnode links.
274     I don't totally understand when it happens, it is caught
275     when closing down the system.
276  */
277     cp = make_coda_node(&ctlfid, 0, VCHR);
278 
279     coda_ctlvp = CTOV(cp);
280 
281     /* Add vfs and rootvp to chain of vfs hanging off mntinfo */
282     mi->mi_vfsp = vfsp;
283     mi->mi_rootvp = rtvp;
284 
285     /* set filesystem block size */
286     vfsp->mnt_stat.f_bsize = 8192;	    /* XXX -JJK */
287     vfsp->mnt_stat.f_frsize = 8192;	    /* XXX -JJK */
288 
289     /* error is currently guaranteed to be zero, but in case some
290        code changes... */
291     CODADEBUG(1,
292 	     myprintf(("coda_mount returned %d\n",error)););
293     if (error)
294 	MARK_INT_FAIL(CODA_MOUNT_STATS);
295     else
296 	MARK_INT_SAT(CODA_MOUNT_STATS);
297 
298     return set_statvfs_info("/coda", UIO_SYSSPACE, "CODA", UIO_SYSSPACE,
299 	vfsp->mnt_op->vfs_name, vfsp, l);
300 }
301 
302 int
303 coda_start(struct mount *vfsp, int flags)
304 {
305     ENTRY;
306     vftomi(vfsp)->mi_started = 1;
307     return (0);
308 }
309 
310 int
311 coda_unmount(struct mount *vfsp, int mntflags)
312 {
313     struct coda_mntinfo *mi = vftomi(vfsp);
314     int active, error = 0;
315 
316     ENTRY;
317     MARK_ENTRY(CODA_UMOUNT_STATS);
318     if (!CODA_MOUNTED(vfsp)) {
319 	MARK_INT_FAIL(CODA_UMOUNT_STATS);
320 	return(EINVAL);
321     }
322 
323     if (mi->mi_vfsp == vfsp) {	/* We found the victim */
324 	if (!IS_UNMOUNTING(VTOC(mi->mi_rootvp)))
325 	    return (EBUSY); 	/* Venus is still running */
326 
327 #ifdef	DEBUG
328 	printf("coda_unmount: ROOT: vp %p, cp %p\n", mi->mi_rootvp, VTOC(mi->mi_rootvp));
329 #endif
330 	mi->mi_started = 0;
331 
332 	vrele(mi->mi_rootvp);
333 
334 	active = coda_kill(vfsp, NOT_DOWNCALL);
335 	mi->mi_rootvp->v_vflag &= ~VV_ROOT;
336 	error = vflush(mi->mi_vfsp, NULLVP, FORCECLOSE);
337 	printf("coda_unmount: active = %d, vflush active %d\n", active, error);
338 	error = 0;
339 
340 	/* I'm going to take this out to allow lookups to go through. I'm
341 	 * not sure it's important anyway. -- DCS 2/2/94
342 	 */
343 	/* vfsp->VFS_DATA = NULL; */
344 
345 	/* No more vfsp's to hold onto */
346 	mi->mi_vfsp = NULL;
347 	mi->mi_rootvp = NULL;
348 
349 	if (error)
350 	    MARK_INT_FAIL(CODA_UMOUNT_STATS);
351 	else
352 	    MARK_INT_SAT(CODA_UMOUNT_STATS);
353 
354 	return(error);
355     }
356     return (EINVAL);
357 }
358 
359 /*
360  * find root of cfs
361  */
362 int
363 coda_root(struct mount *vfsp, struct vnode **vpp)
364 {
365     struct coda_mntinfo *mi = vftomi(vfsp);
366     int error;
367     struct lwp *l = curlwp;    /* XXX - bnoble */
368     CodaFid VFid;
369     static const CodaFid invalfid = INVAL_FID;
370 
371     ENTRY;
372     MARK_ENTRY(CODA_ROOT_STATS);
373 
374     if (vfsp == mi->mi_vfsp) {
375     	if (memcmp(&VTOC(mi->mi_rootvp)->c_fid, &invalfid, sizeof(CodaFid)))
376 	    { /* Found valid root. */
377 		*vpp = mi->mi_rootvp;
378 		/* On Mach, this is vref.  On NetBSD, VOP_LOCK */
379 		vref(*vpp);
380 		vn_lock(*vpp, LK_EXCLUSIVE);
381 		MARK_INT_SAT(CODA_ROOT_STATS);
382 		return(0);
383 	    }
384     }
385 
386     error = venus_root(vftomi(vfsp), l->l_cred, l->l_proc, &VFid);
387 
388     if (!error) {
389 	/*
390 	 * Save the new rootfid in the cnode, and rehash the cnode into the
391 	 * cnode hash with the new fid key.
392 	 */
393 	coda_unsave(VTOC(mi->mi_rootvp));
394 	VTOC(mi->mi_rootvp)->c_fid = VFid;
395 	coda_save(VTOC(mi->mi_rootvp));
396 
397 	*vpp = mi->mi_rootvp;
398 	vref(*vpp);
399 	vn_lock(*vpp, LK_EXCLUSIVE);
400 	MARK_INT_SAT(CODA_ROOT_STATS);
401 	goto exit;
402     } else if (error == ENODEV || error == EINTR) {
403 	/* Gross hack here! */
404 	/*
405 	 * If Venus fails to respond to the CODA_ROOT call, coda_call returns
406 	 * ENODEV. Return the uninitialized root vnode to allow vfs
407 	 * operations such as unmount to continue. Without this hack,
408 	 * there is no way to do an unmount if Venus dies before a
409 	 * successful CODA_ROOT call is done. All vnode operations
410 	 * will fail.
411 	 */
412 	*vpp = mi->mi_rootvp;
413 	vref(*vpp);
414 	vn_lock(*vpp, LK_EXCLUSIVE);
415 	MARK_INT_FAIL(CODA_ROOT_STATS);
416 	error = 0;
417 	goto exit;
418     } else {
419 	CODADEBUG( CODA_ROOT, myprintf(("error %d in CODA_ROOT\n", error)); );
420 	MARK_INT_FAIL(CODA_ROOT_STATS);
421 
422 	goto exit;
423     }
424  exit:
425     return(error);
426 }
427 
428 /*
429  * Get file system statistics.
430  */
431 int
432 coda_nb_statvfs(struct mount *vfsp, struct statvfs *sbp)
433 {
434     struct lwp *l = curlwp;
435     struct coda_statfs fsstat;
436     int error;
437 
438     ENTRY;
439     MARK_ENTRY(CODA_STATFS_STATS);
440     if (!CODA_MOUNTED(vfsp)) {
441 /*	MARK_INT_FAIL(CODA_STATFS_STATS); */
442 	return(EINVAL);
443     }
444 
445     /* XXX - what to do about f_flags, others? --bnoble */
446     /* Below This is what AFS does
447     	#define NB_SFS_SIZ 0x895440
448      */
449     /* Note: Normal fs's have a bsize of 0x400 == 1024 */
450 
451     error = venus_statfs(vftomi(vfsp), l->l_cred, l, &fsstat);
452 
453     if (!error) {
454 	sbp->f_bsize = 8192; /* XXX */
455 	sbp->f_frsize = 8192; /* XXX */
456 	sbp->f_iosize = 8192; /* XXX */
457 	sbp->f_blocks = fsstat.f_blocks;
458 	sbp->f_bfree  = fsstat.f_bfree;
459 	sbp->f_bavail = fsstat.f_bavail;
460 	sbp->f_bresvd = 0;
461 	sbp->f_files  = fsstat.f_files;
462 	sbp->f_ffree  = fsstat.f_ffree;
463 	sbp->f_favail = fsstat.f_ffree;
464 	sbp->f_fresvd = 0;
465 	copy_statvfs_info(sbp, vfsp);
466     }
467 
468     MARK_INT_SAT(CODA_STATFS_STATS);
469     return(error);
470 }
471 
472 /*
473  * Flush any pending I/O.
474  */
475 int
476 coda_sync(struct mount *vfsp, int waitfor,
477     kauth_cred_t cred)
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(struct mount *vfsp, ino_t ino,
487     struct vnode **vpp)
488 {
489     ENTRY;
490     return (EOPNOTSUPP);
491 }
492 
493 /*
494  * fhtovp is now what vget used to be in 4.3-derived systems.  For
495  * some silly reason, vget is now keyed by a 32 bit ino_t, rather than
496  * a type-specific fid.
497  */
498 int
499 coda_fhtovp(struct mount *vfsp, struct fid *fhp, struct mbuf *nam,
500     struct vnode **vpp, int *exflagsp,
501     kauth_cred_t *creadanonp)
502 {
503     struct cfid *cfid = (struct cfid *)fhp;
504     struct cnode *cp = 0;
505     int error;
506     struct lwp *l = curlwp; /* XXX -mach */
507     CodaFid VFid;
508     int vtype;
509 
510     ENTRY;
511 
512     MARK_ENTRY(CODA_VGET_STATS);
513     /* Check for vget of control object. */
514     if (IS_CTL_FID(&cfid->cfid_fid)) {
515 	*vpp = coda_ctlvp;
516 	vref(coda_ctlvp);
517 	MARK_INT_SAT(CODA_VGET_STATS);
518 	return(0);
519     }
520 
521     error = venus_fhtovp(vftomi(vfsp), &cfid->cfid_fid, l->l_cred, l->l_proc, &VFid, &vtype);
522 
523     if (error) {
524 	CODADEBUG(CODA_VGET, myprintf(("vget error %d\n",error));)
525 	    *vpp = (struct vnode *)0;
526     } else {
527 	CODADEBUG(CODA_VGET,
528 		 myprintf(("vget: %s type %d result %d\n",
529 			coda_f2s(&VFid), vtype, error)); )
530 
531 	cp = make_coda_node(&VFid, vfsp, vtype);
532 	*vpp = CTOV(cp);
533     }
534     return(error);
535 }
536 
537 int
538 coda_vptofh(struct vnode *vnp, struct fid *fidp)
539 {
540     ENTRY;
541     return (EOPNOTSUPP);
542 }
543 
544 void
545 coda_init(void)
546 {
547     ENTRY;
548 }
549 
550 void
551 coda_done(void)
552 {
553     ENTRY;
554 }
555 
556 SYSCTL_SETUP(sysctl_vfs_coda_setup, "sysctl vfs.coda subtree setup")
557 {
558 	sysctl_createv(clog, 0, NULL, NULL,
559 		       CTLFLAG_PERMANENT,
560 		       CTLTYPE_NODE, "vfs", NULL,
561 		       NULL, 0, NULL, 0,
562 		       CTL_VFS, CTL_EOL);
563 	sysctl_createv(clog, 0, NULL, NULL,
564 		       CTLFLAG_PERMANENT,
565 		       CTLTYPE_NODE, "coda",
566 		       SYSCTL_DESCR("code vfs options"),
567 		       NULL, 0, NULL, 0,
568 		       CTL_VFS, 18, CTL_EOL);
569 	/*
570 	 * XXX the "18" above could be dynamic, thereby eliminating
571 	 * one more instance of the "number to vfs" mapping problem,
572 	 * but "18" is the order as taken from sys/mount.h
573 	 */
574 
575 /*
576 	sysctl_createv(clog, 0, NULL, NULL,
577 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
578 		       CTLTYPE_INT, "clusterread",
579 		       SYSCTL_DESCR( anyone? ),
580 		       NULL, 0, &doclusterread, 0,
581 		       CTL_VFS, 18, FFS_CLUSTERREAD, CTL_EOL);
582 */
583 }
584 
585 /*
586  * To allow for greater ease of use, some vnodes may be orphaned when
587  * Venus dies.  Certain operations should still be allowed to go
588  * through, but without propagating orphan-ness.  So this function will
589  * get a new vnode for the file from the current run of Venus.
590  */
591 
592 int
593 getNewVnode(struct vnode **vpp)
594 {
595     struct cfid cfid;
596     struct coda_mntinfo *mi = vftomi((*vpp)->v_mount);
597 
598     ENTRY;
599 
600     cfid.cfid_len = (short)sizeof(CodaFid);
601     cfid.cfid_fid = VTOC(*vpp)->c_fid;	/* Structure assignment. */
602     /* XXX ? */
603 
604     /* We're guessing that if set, the 1st element on the list is a
605      * valid vnode to use. If not, return ENODEV as venus is dead.
606      */
607     if (mi->mi_vfsp == NULL)
608 	return ENODEV;
609 
610     return coda_fhtovp(mi->mi_vfsp, (struct fid*)&cfid, NULL, vpp,
611 		      NULL, NULL);
612 }
613 
614 #include <ufs/ufs/quota.h>
615 #include <ufs/ufs/ufsmount.h>
616 /* get the mount structure corresponding to a given device.  Assume
617  * device corresponds to a UFS. Return NULL if no device is found.
618  */
619 struct mount *devtomp(dev_t dev)
620 {
621     struct mount *mp, *nmp;
622 
623     for (mp = mountlist.cqh_first; mp != (void*)&mountlist; mp = nmp) {
624 	nmp = mp->mnt_list.cqe_next;
625 	if ((!strcmp(mp->mnt_op->vfs_name, MOUNT_UFS)) &&
626 	    ((VFSTOUFS(mp))->um_dev == (dev_t) dev)) {
627 	    /* mount corresponds to UFS and the device matches one we want */
628 	    return(mp);
629 	}
630     }
631     /* mount structure wasn't found */
632     return(NULL);
633 }
634