xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs_common.c (revision 4445fffbbb1ea25fd0e9ea68b9380dd7a6709025)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011 Bayard G. Bell. All rights reserved.
24  */
25 
26 /*
27  *	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T
28  *		All rights reserved.
29  */
30 
31 #include <sys/errno.h>
32 #include <sys/param.h>
33 #include <sys/types.h>
34 #include <sys/user.h>
35 #include <sys/stat.h>
36 #include <sys/time.h>
37 #include <sys/utsname.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/pathname.h>
42 #include <sys/bootconf.h>
43 #include <fs/fs_subr.h>
44 #include <rpc/types.h>
45 #include <nfs/nfs.h>
46 #include <nfs/nfs4.h>
47 #include <nfs/nfs_clnt.h>
48 #include <nfs/rnode.h>
49 #include <nfs/mount.h>
50 #include <nfs/nfssys.h>
51 #include <sys/debug.h>
52 #include <sys/cmn_err.h>
53 #include <sys/file.h>
54 #include <sys/fcntl.h>
55 #include <sys/zone.h>
56 
57 /*
58  * This is the loadable module wrapper.
59  */
60 #include <sys/systm.h>
61 #include <sys/modctl.h>
62 #include <sys/syscall.h>
63 #include <sys/ddi.h>
64 
65 #include <rpc/types.h>
66 #include <rpc/auth.h>
67 #include <rpc/clnt.h>
68 #include <rpc/svc.h>
69 
70 /*
71  * The pseudo NFS filesystem to allow diskless booting to dynamically
72  * mount either a NFS V2, NFS V3, or NFS V4 filesystem.  This only implements
73  * the VFS_MOUNTROOT op and is only intended to be used by the
74  * diskless booting code until the real root filesystem is mounted.
75  * Nothing else should ever call this!
76  *
77  * The strategy is that if the initial rootfs type is set to "nfsdyn"
78  * by loadrootmodules() this filesystem is called to mount the
79  * root filesystem.  It first attempts to mount a V4 filesystem, and if that
80  * fails due to an RPC version mismatch it tries V3 and finally V2.
81  * Once the real mount succeeds the vfsops and rootfs name are changed
82  * to reflect the real filesystem type.
83  */
84 static int nfsdyninit(int, char *);
85 static int nfsdyn_mountroot(vfs_t *, whymountroot_t);
86 
87 vfsops_t *nfsdyn_vfsops;
88 
89 /*
90  * The following data structures are used to configure the NFS
91  * system call, the NFS Version 2 client VFS, and the NFS Version
92  * 3 client VFS into the system.  The NFS Version 4 structures are defined in
93  * nfs4_common.c
94  */
95 
96 /*
97  * The NFS system call.
98  */
99 static struct sysent nfssysent = {
100 	2,
101 	SE_32RVAL1 | SE_ARGC | SE_NOUNLOAD,
102 	nfssys
103 };
104 
105 static struct modlsys modlsys = {
106 	&mod_syscallops,
107 	"NFS syscall, client, and common",
108 	&nfssysent
109 };
110 
111 #ifdef _SYSCALL32_IMPL
112 static struct modlsys modlsys32 = {
113 	&mod_syscallops32,
114 	"NFS syscall, client, and common (32-bit)",
115 	&nfssysent
116 };
117 #endif /* _SYSCALL32_IMPL */
118 
119 /*
120  * The NFS Dynamic client VFS.
121  */
122 static vfsdef_t vfw = {
123 	VFSDEF_VERSION,
124 	"nfsdyn",
125 	nfsdyninit,
126 	VSW_ZMOUNT,
127 	NULL
128 };
129 
130 static struct modlfs modlfs = {
131 	&mod_fsops,
132 	"network filesystem",
133 	&vfw
134 };
135 
136 /*
137  * The NFS Version 2 client VFS.
138  */
139 static vfsdef_t vfw2 = {
140 	VFSDEF_VERSION,
141 	"nfs",
142 	nfsinit,
143 	VSW_CANREMOUNT|VSW_NOTZONESAFE|VSW_STATS|VSW_ZMOUNT,
144 	NULL
145 };
146 
147 static struct modlfs modlfs2 = {
148 	&mod_fsops,
149 	"network filesystem version 2",
150 	&vfw2
151 };
152 
153 /*
154  * The NFS Version 3 client VFS.
155  */
156 static vfsdef_t vfw3 = {
157 	VFSDEF_VERSION,
158 	"nfs3",
159 	nfs3init,
160 	VSW_CANREMOUNT|VSW_NOTZONESAFE|VSW_STATS|VSW_ZMOUNT,
161 	NULL
162 };
163 
164 static struct modlfs modlfs3 = {
165 	&mod_fsops,
166 	"network filesystem version 3",
167 	&vfw3
168 };
169 
170 extern struct modlfs modlfs4;
171 
172 /*
173  * We have too many linkage structures so we define our own XXX
174  */
175 struct modlinkage_big {
176 	int		ml_rev;		/* rev of loadable modules system */
177 	void		*ml_linkage[7];	/* NULL terminated list of */
178 					/* linkage structures */
179 };
180 
181 /*
182  * All of the module configuration linkages required to configure
183  * the system call and client VFS's into the system.
184  */
185 static struct modlinkage_big modlinkage = {
186 	MODREV_1,
187 	&modlsys,
188 #ifdef _SYSCALL32_IMPL
189 	&modlsys32,
190 #endif
191 	&modlfs,
192 	&modlfs2,
193 	&modlfs3,
194 	&modlfs4,
195 	NULL
196 };
197 
198 /*
199  * This routine is invoked automatically when the kernel module
200  * containing this routine is loaded.  This allows module specific
201  * initialization to be done when the module is loaded.
202  */
203 int
204 _init(void)
205 {
206 	int status;
207 
208 	if ((status = nfs_clntinit()) != 0) {
209 		cmn_err(CE_WARN, "_init: nfs_clntinit failed");
210 		return (status);
211 	}
212 
213 	/*
214 	 * Create the version specific kstats.
215 	 *
216 	 * PSARC 2001/697 Contract Private Interface
217 	 * All nfs kstats are under SunMC contract
218 	 * Please refer to the PSARC listed above and contact
219 	 * SunMC before making any changes!
220 	 *
221 	 * Changes must be reviewed by Solaris File Sharing
222 	 * Changes must be communicated to contract-2001-697@sun.com
223 	 *
224 	 */
225 
226 	zone_key_create(&nfsstat_zone_key, nfsstat_zone_init, NULL,
227 	    nfsstat_zone_fini);
228 	status = mod_install((struct modlinkage *)&modlinkage);
229 
230 	if (status)  {
231 		(void) zone_key_delete(nfsstat_zone_key);
232 
233 		/*
234 		 * Failed to install module, cleanup previous
235 		 * initialization work.
236 		 */
237 		nfs_clntfini();
238 
239 		/*
240 		 * Clean up work performed indirectly by mod_installfs()
241 		 * as a result of our call to mod_install().
242 		 */
243 		nfs4fini();
244 		nfs3fini();
245 		nfsfini();
246 	}
247 	return (status);
248 }
249 
250 int
251 _fini(void)
252 {
253 	/* Don't allow module to be unloaded */
254 	return (EBUSY);
255 }
256 
257 int
258 _info(struct modinfo *modinfop)
259 {
260 	return (mod_info((struct modlinkage *)&modlinkage, modinfop));
261 }
262 
263 /*
264  * General utilities
265  */
266 
267 /*
268  * Returns the preferred transfer size in bytes based on
269  * what network interfaces are available.
270  */
271 int
272 nfstsize(void)
273 {
274 	/*
275 	 * For the moment, just return NFS_MAXDATA until we can query the
276 	 * appropriate transport.
277 	 */
278 	return (NFS_MAXDATA);
279 }
280 
281 /*
282  * Returns the preferred transfer size in bytes based on
283  * what network interfaces are available.
284  */
285 
286 /* this should reflect the largest transfer size possible */
287 static int nfs3_max_transfer_size = 1024 * 1024;
288 
289 int
290 nfs3tsize(void)
291 {
292 	/*
293 	 * For the moment, just return nfs3_max_transfer_size until we
294 	 * can query the appropriate transport.
295 	 */
296 	return (nfs3_max_transfer_size);
297 }
298 
299 static uint_t nfs3_max_transfer_size_clts = 32 * 1024;
300 static uint_t nfs3_max_transfer_size_cots = 1024 * 1024;
301 static uint_t nfs3_max_transfer_size_rdma = 1024 * 1024;
302 
303 uint_t
304 nfs3_tsize(struct knetconfig *knp)
305 {
306 
307 	if (knp->knc_semantics == NC_TPI_COTS_ORD ||
308 	    knp->knc_semantics == NC_TPI_COTS)
309 		return (nfs3_max_transfer_size_cots);
310 	if (knp->knc_semantics == NC_TPI_RDMA)
311 		return (nfs3_max_transfer_size_rdma);
312 	return (nfs3_max_transfer_size_clts);
313 }
314 
315 uint_t
316 rfs3_tsize(struct svc_req *req)
317 {
318 
319 	if (req->rq_xprt->xp_type == T_COTS_ORD ||
320 	    req->rq_xprt->xp_type == T_COTS)
321 		return (nfs3_max_transfer_size_cots);
322 	if (req->rq_xprt->xp_type == T_RDMA)
323 		return (nfs3_max_transfer_size_rdma);
324 	return (nfs3_max_transfer_size_clts);
325 }
326 
327 /* ARGSUSED */
328 static int
329 nfsdyninit(int fstyp, char *name)
330 {
331 	static const fs_operation_def_t nfsdyn_vfsops_template[] = {
332 		VFSNAME_MOUNTROOT, { .vfs_mountroot = nfsdyn_mountroot },
333 		NULL, NULL
334 	};
335 	int error;
336 
337 	error = vfs_setfsops(fstyp, nfsdyn_vfsops_template, &nfsdyn_vfsops);
338 	if (error != 0)
339 		return (error);
340 
341 	return (0);
342 }
343 
344 /* ARGSUSED */
345 static int
346 nfsdyn_mountroot(vfs_t *vfsp, whymountroot_t why)
347 {
348 	char root_hostname[SYS_NMLN+1];
349 	struct servinfo *svp;
350 	int error;
351 	int vfsflags;
352 	char *root_path;
353 	struct pathname pn;
354 	char *name;
355 	static char token[10];
356 	struct nfs_args args;		/* nfs mount arguments */
357 
358 	bzero(&args, sizeof (args));
359 
360 	/* do this BEFORE getfile which causes xid stamps to be initialized */
361 	clkset(-1L);		/* hack for now - until we get time svc? */
362 
363 	if (why == ROOT_REMOUNT) {
364 		/*
365 		 * Shouldn't happen.
366 		 */
367 		panic("nfs3_mountroot: why == ROOT_REMOUNT\n");
368 	}
369 
370 	if (why == ROOT_UNMOUNT) {
371 		/*
372 		 * Nothing to do for NFS.
373 		 */
374 		return (0);
375 	}
376 
377 	/*
378 	 * why == ROOT_INIT
379 	 */
380 
381 	name = token;
382 	*name = 0;
383 	getfsname("root", name, sizeof (token));
384 
385 	pn_alloc(&pn);
386 	root_path = pn.pn_path;
387 
388 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
389 	mutex_init(&svp->sv_lock, NULL, MUTEX_DEFAULT, NULL);
390 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
391 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
392 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
393 
394 	/*
395 	 * First try version 4
396 	 */
397 	vfs_setops(vfsp, nfs4_vfsops);
398 	args.addr = &svp->sv_addr;
399 	args.fh = (char *)&svp->sv_fhandle;
400 	args.knconf = svp->sv_knconf;
401 	args.hostname = root_hostname;
402 	vfsflags = 0;
403 
404 	if (error = mount_root(*name ? name : "root", root_path, NFS_V4,
405 	    &args, &vfsflags)) {
406 		if (error != EPROTONOSUPPORT) {
407 			nfs_cmn_err(error, CE_WARN,
408 			    "Unable to mount NFS root filesystem: %m");
409 			sv_free(svp);
410 			pn_free(&pn);
411 			vfs_setops(vfsp, nfsdyn_vfsops);
412 			return (error);
413 		}
414 
415 		/*
416 		 * Then try version 3
417 		 */
418 		bzero(&args, sizeof (args));
419 		vfs_setops(vfsp, nfs3_vfsops);
420 		args.addr = &svp->sv_addr;
421 		args.fh = (char *)&svp->sv_fhandle;
422 		args.knconf = svp->sv_knconf;
423 		args.hostname = root_hostname;
424 		vfsflags = 0;
425 
426 		if (error = mount_root(*name ? name : "root", root_path,
427 		    NFS_V3, &args, &vfsflags)) {
428 			if (error != EPROTONOSUPPORT) {
429 				nfs_cmn_err(error, CE_WARN,
430 				    "Unable to mount NFS root filesystem: %m");
431 				sv_free(svp);
432 				pn_free(&pn);
433 				vfs_setops(vfsp, nfsdyn_vfsops);
434 				return (error);
435 			}
436 
437 			/*
438 			 * Finally, try version 2
439 			 */
440 			bzero(&args, sizeof (args));
441 			args.addr = &svp->sv_addr;
442 			args.fh = (char *)&svp->sv_fhandle.fh_buf;
443 			args.knconf = svp->sv_knconf;
444 			args.hostname = root_hostname;
445 			vfsflags = 0;
446 
447 			vfs_setops(vfsp, nfs_vfsops);
448 
449 			if (error = mount_root(*name ? name : "root",
450 			    root_path, NFS_VERSION, &args, &vfsflags)) {
451 				nfs_cmn_err(error, CE_WARN,
452 				    "Unable to mount NFS root filesystem: %m");
453 				sv_free(svp);
454 				pn_free(&pn);
455 				vfs_setops(vfsp, nfsdyn_vfsops);
456 				return (error);
457 			}
458 		}
459 	}
460 
461 	sv_free(svp);
462 	pn_free(&pn);
463 	return (VFS_MOUNTROOT(vfsp, why));
464 }
465 
466 int
467 nfs_setopts(vnode_t *vp, model_t model, struct nfs_args *buf)
468 {
469 	mntinfo_t *mi;			/* mount info, pointed at by vfs */
470 	STRUCT_HANDLE(nfs_args, args);
471 	int flags;
472 
473 #ifdef lint
474 	model = model;
475 #endif
476 
477 	STRUCT_SET_HANDLE(args, model, buf);
478 
479 	flags = STRUCT_FGET(args, flags);
480 
481 	/*
482 	 * Set option fields in mount info record
483 	 */
484 	mi = VTOMI(vp);
485 
486 	if (flags & NFSMNT_NOAC) {
487 		mi->mi_flags |= MI_NOAC;
488 		PURGE_ATTRCACHE(vp);
489 	}
490 	if (flags & NFSMNT_NOCTO)
491 		mi->mi_flags |= MI_NOCTO;
492 	if (flags & NFSMNT_LLOCK)
493 		mi->mi_flags |= MI_LLOCK;
494 	if (flags & NFSMNT_GRPID)
495 		mi->mi_flags |= MI_GRPID;
496 	if (flags & NFSMNT_RETRANS) {
497 		if (STRUCT_FGET(args, retrans) < 0)
498 			return (EINVAL);
499 		mi->mi_retrans = STRUCT_FGET(args, retrans);
500 	}
501 	if (flags & NFSMNT_TIMEO) {
502 		if (STRUCT_FGET(args, timeo) <= 0)
503 			return (EINVAL);
504 		mi->mi_timeo = STRUCT_FGET(args, timeo);
505 		/*
506 		 * The following scales the standard deviation and
507 		 * and current retransmission timer to match the
508 		 * initial value for the timeout specified.
509 		 */
510 		mi->mi_timers[NFS_CALLTYPES].rt_deviate =
511 		    (mi->mi_timeo * hz * 2) / 5;
512 		mi->mi_timers[NFS_CALLTYPES].rt_rtxcur =
513 		    mi->mi_timeo * hz / 10;
514 	}
515 	if (flags & NFSMNT_RSIZE) {
516 		if (STRUCT_FGET(args, rsize) <= 0)
517 			return (EINVAL);
518 		mi->mi_tsize = MIN(mi->mi_tsize, STRUCT_FGET(args, rsize));
519 		mi->mi_curread = MIN(mi->mi_curread, mi->mi_tsize);
520 	}
521 	if (flags & NFSMNT_WSIZE) {
522 		if (STRUCT_FGET(args, wsize) <= 0)
523 			return (EINVAL);
524 		mi->mi_stsize = MIN(mi->mi_stsize, STRUCT_FGET(args, wsize));
525 		mi->mi_curwrite = MIN(mi->mi_curwrite, mi->mi_stsize);
526 	}
527 	if (flags & NFSMNT_ACREGMIN) {
528 		if (STRUCT_FGET(args, acregmin) < 0)
529 			mi->mi_acregmin = ACMINMAX;
530 		else
531 			mi->mi_acregmin = MIN(STRUCT_FGET(args, acregmin),
532 			    ACMINMAX);
533 		mi->mi_acregmin = SEC2HR(mi->mi_acregmin);
534 	}
535 	if (flags & NFSMNT_ACREGMAX) {
536 		if (STRUCT_FGET(args, acregmax) < 0)
537 			mi->mi_acregmax = ACMAXMAX;
538 		else
539 			mi->mi_acregmax = MIN(STRUCT_FGET(args, acregmax),
540 			    ACMAXMAX);
541 		mi->mi_acregmax = SEC2HR(mi->mi_acregmax);
542 	}
543 	if (flags & NFSMNT_ACDIRMIN) {
544 		if (STRUCT_FGET(args, acdirmin) < 0)
545 			mi->mi_acdirmin = ACMINMAX;
546 		else
547 			mi->mi_acdirmin = MIN(STRUCT_FGET(args, acdirmin),
548 			    ACMINMAX);
549 		mi->mi_acdirmin = SEC2HR(mi->mi_acdirmin);
550 	}
551 	if (flags & NFSMNT_ACDIRMAX) {
552 		if (STRUCT_FGET(args, acdirmax) < 0)
553 			mi->mi_acdirmax = ACMAXMAX;
554 		else
555 			mi->mi_acdirmax = MIN(STRUCT_FGET(args, acdirmax),
556 			    ACMAXMAX);
557 		mi->mi_acdirmax = SEC2HR(mi->mi_acdirmax);
558 	}
559 
560 	if (flags & NFSMNT_LOOPBACK)
561 		mi->mi_flags |= MI_LOOPBACK;
562 
563 	return (0);
564 }
565 
566 /*
567  * Set or Clear direct I/O flag
568  * VOP_RWLOCK() is held for write access to prevent a race condition
569  * which would occur if a process is in the middle of a write when
570  * directio flag gets set. It is possible that all pages may not get flushed.
571  */
572 
573 /* ARGSUSED */
574 int
575 nfs_directio(vnode_t *vp, int cmd, cred_t *cr)
576 {
577 	int	error = 0;
578 	rnode_t	*rp;
579 
580 	rp = VTOR(vp);
581 
582 	if (cmd == DIRECTIO_ON) {
583 
584 		if (rp->r_flags & RDIRECTIO)
585 			return (0);
586 
587 		/*
588 		 * Flush the page cache.
589 		 */
590 
591 		(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
592 
593 		if (rp->r_flags & RDIRECTIO) {
594 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
595 			return (0);
596 		}
597 
598 		if (vn_has_cached_data(vp) &&
599 		    ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) {
600 			error = VOP_PUTPAGE(vp, (offset_t)0, (uint_t)0,
601 			    B_INVAL, cr, NULL);
602 			if (error) {
603 				if (error == ENOSPC || error == EDQUOT) {
604 					mutex_enter(&rp->r_statelock);
605 					if (!rp->r_error)
606 						rp->r_error = error;
607 					mutex_exit(&rp->r_statelock);
608 				}
609 				VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
610 				return (error);
611 			}
612 		}
613 
614 		mutex_enter(&rp->r_statelock);
615 		rp->r_flags |= RDIRECTIO;
616 		mutex_exit(&rp->r_statelock);
617 		VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
618 		return (0);
619 	}
620 
621 	if (cmd == DIRECTIO_OFF) {
622 		mutex_enter(&rp->r_statelock);
623 		rp->r_flags &= ~RDIRECTIO;	/* disable direct mode */
624 		mutex_exit(&rp->r_statelock);
625 		return (0);
626 	}
627 
628 	return (EINVAL);
629 }
630