nfs_clport.c revision 317525
1/*-
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD: stable/10/sys/fs/nfsclient/nfs_clport.c 317525 2017-04-27 21:37:39Z rmacklem $");
36
37#include "opt_inet6.h"
38#include "opt_kdtrace.h"
39
40#include <sys/capsicum.h>
41
42/*
43 * generally, I don't like #includes inside .h files, but it seems to
44 * be the easiest way to handle the port.
45 */
46#include <sys/hash.h>
47#include <fs/nfs/nfsport.h>
48#include <netinet/if_ether.h>
49#include <net/if_types.h>
50
51#include <fs/nfsclient/nfs_kdtrace.h>
52
53#ifdef KDTRACE_HOOKS
54dtrace_nfsclient_attrcache_flush_probe_func_t
55		dtrace_nfscl_attrcache_flush_done_probe;
56uint32_t	nfscl_attrcache_flush_done_id;
57
58dtrace_nfsclient_attrcache_get_hit_probe_func_t
59		dtrace_nfscl_attrcache_get_hit_probe;
60uint32_t	nfscl_attrcache_get_hit_id;
61
62dtrace_nfsclient_attrcache_get_miss_probe_func_t
63		dtrace_nfscl_attrcache_get_miss_probe;
64uint32_t	nfscl_attrcache_get_miss_id;
65
66dtrace_nfsclient_attrcache_load_probe_func_t
67		dtrace_nfscl_attrcache_load_done_probe;
68uint32_t	nfscl_attrcache_load_done_id;
69#endif /* !KDTRACE_HOOKS */
70
71extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
72extern struct vop_vector newnfs_vnodeops;
73extern struct vop_vector newnfs_fifoops;
74extern uma_zone_t newnfsnode_zone;
75extern struct buf_ops buf_ops_newnfs;
76extern int ncl_pbuf_freecnt;
77extern short nfsv4_cbport;
78extern int nfscl_enablecallb;
79extern int nfs_numnfscbd;
80extern int nfscl_inited;
81struct mtx ncl_iod_mutex;
82NFSDLOCKMUTEX;
83
84extern void (*ncl_call_invalcaches)(struct vnode *);
85
86/*
87 * Comparison function for vfs_hash functions.
88 */
89int
90newnfs_vncmpf(struct vnode *vp, void *arg)
91{
92	struct nfsfh *nfhp = (struct nfsfh *)arg;
93	struct nfsnode *np = VTONFS(vp);
94
95	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
96	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
97		return (1);
98	return (0);
99}
100
101/*
102 * Look up a vnode/nfsnode by file handle.
103 * Callers must check for mount points!!
104 * In all cases, a pointer to a
105 * nfsnode structure is returned.
106 * This variant takes a "struct nfsfh *" as second argument and uses
107 * that structure up, either by hanging off the nfsnode or FREEing it.
108 */
109int
110nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
111    struct componentname *cnp, struct thread *td, struct nfsnode **npp,
112    void *stuff, int lkflags)
113{
114	struct nfsnode *np, *dnp;
115	struct vnode *vp, *nvp;
116	struct nfsv4node *newd, *oldd;
117	int error;
118	u_int hash;
119	struct nfsmount *nmp;
120
121	nmp = VFSTONFS(mntp);
122	dnp = VTONFS(dvp);
123	*npp = NULL;
124
125	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
126
127	error = vfs_hash_get(mntp, hash, lkflags,
128	    td, &nvp, newnfs_vncmpf, nfhp);
129	if (error == 0 && nvp != NULL) {
130		/*
131		 * I believe there is a slight chance that vgonel() could
132		 * get called on this vnode between when NFSVOPLOCK() drops
133		 * the VI_LOCK() and vget() acquires it again, so that it
134		 * hasn't yet had v_usecount incremented. If this were to
135		 * happen, the VI_DOOMED flag would be set, so check for
136		 * that here. Since we now have the v_usecount incremented,
137		 * we should be ok until we vrele() it, if the VI_DOOMED
138		 * flag isn't set now.
139		 */
140		VI_LOCK(nvp);
141		if ((nvp->v_iflag & VI_DOOMED)) {
142			VI_UNLOCK(nvp);
143			vrele(nvp);
144			error = ENOENT;
145		} else {
146			VI_UNLOCK(nvp);
147		}
148	}
149	if (error) {
150		FREE((caddr_t)nfhp, M_NFSFH);
151		return (error);
152	}
153	if (nvp != NULL) {
154		np = VTONFS(nvp);
155		/*
156		 * For NFSv4, check to see if it is the same name and
157		 * replace the name, if it is different.
158		 */
159		oldd = newd = NULL;
160		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
161		    nvp->v_type == VREG &&
162		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
163		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
164		     cnp->cn_namelen) ||
165		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
166		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
167		     dnp->n_fhp->nfh_len))) {
168		    MALLOC(newd, struct nfsv4node *,
169			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
170			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
171		    NFSLOCKNODE(np);
172		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
173			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
174			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
175			 cnp->cn_namelen) ||
176			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
177			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
178			 dnp->n_fhp->nfh_len))) {
179			oldd = np->n_v4;
180			np->n_v4 = newd;
181			newd = NULL;
182			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
183			np->n_v4->n4_namelen = cnp->cn_namelen;
184			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
185			    dnp->n_fhp->nfh_len);
186			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
187			    cnp->cn_namelen);
188		    }
189		    NFSUNLOCKNODE(np);
190		}
191		if (newd != NULL)
192			FREE((caddr_t)newd, M_NFSV4NODE);
193		if (oldd != NULL)
194			FREE((caddr_t)oldd, M_NFSV4NODE);
195		*npp = np;
196		FREE((caddr_t)nfhp, M_NFSFH);
197		return (0);
198	}
199	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
200
201	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
202	if (error) {
203		uma_zfree(newnfsnode_zone, np);
204		FREE((caddr_t)nfhp, M_NFSFH);
205		return (error);
206	}
207	vp = nvp;
208	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
209	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
210	vp->v_data = np;
211	np->n_vnode = vp;
212	/*
213	 * Initialize the mutex even if the vnode is going to be a loser.
214	 * This simplifies the logic in reclaim, which can then unconditionally
215	 * destroy the mutex (in the case of the loser, or if hash_insert
216	 * happened to return an error no special casing is needed).
217	 */
218	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
219
220	/*
221	 * Are we getting the root? If so, make sure the vnode flags
222	 * are correct
223	 */
224	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
225	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
226		if (vp->v_type == VNON)
227			vp->v_type = VDIR;
228		vp->v_vflag |= VV_ROOT;
229	}
230
231	np->n_fhp = nfhp;
232	/*
233	 * For NFSv4, we have to attach the directory file handle and
234	 * file name, so that Open Ops can be done later.
235	 */
236	if (nmp->nm_flag & NFSMNT_NFSV4) {
237		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
238		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
239		    M_WAITOK);
240		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
241		np->n_v4->n4_namelen = cnp->cn_namelen;
242		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
243		    dnp->n_fhp->nfh_len);
244		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
245		    cnp->cn_namelen);
246	} else {
247		np->n_v4 = NULL;
248	}
249
250	/*
251	 * NFS supports recursive and shared locking.
252	 */
253	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
254	VN_LOCK_AREC(vp);
255	VN_LOCK_ASHARE(vp);
256	error = insmntque(vp, mntp);
257	if (error != 0) {
258		*npp = NULL;
259		mtx_destroy(&np->n_mtx);
260		FREE((caddr_t)nfhp, M_NFSFH);
261		if (np->n_v4 != NULL)
262			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
263		uma_zfree(newnfsnode_zone, np);
264		return (error);
265	}
266	error = vfs_hash_insert(vp, hash, lkflags,
267	    td, &nvp, newnfs_vncmpf, nfhp);
268	if (error)
269		return (error);
270	if (nvp != NULL) {
271		*npp = VTONFS(nvp);
272		/* vfs_hash_insert() vput()'s the losing vnode */
273		return (0);
274	}
275	*npp = np;
276
277	return (0);
278}
279
280/*
281 * Another variant of nfs_nget(). This one is only used by reopen. It
282 * takes almost the same args as nfs_nget(), but only succeeds if an entry
283 * exists in the cache. (Since files should already be "open" with a
284 * vnode ref cnt on the node when reopen calls this, it should always
285 * succeed.)
286 * Also, don't get a vnode lock, since it may already be locked by some
287 * other process that is handling it. This is ok, since all other threads
288 * on the client are blocked by the nfsc_lock being exclusively held by the
289 * caller of this function.
290 */
291int
292nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
293    struct thread *td, struct nfsnode **npp)
294{
295	struct vnode *nvp;
296	u_int hash;
297	struct nfsfh *nfhp;
298	int error;
299
300	*npp = NULL;
301	/* For forced dismounts, just return error. */
302	if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
303		return (EINTR);
304	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
305	    M_NFSFH, M_WAITOK);
306	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
307	nfhp->nfh_len = fhsize;
308
309	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
310
311	/*
312	 * First, try to get the vnode locked, but don't block for the lock.
313	 */
314	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
315	    newnfs_vncmpf, nfhp);
316	if (error == 0 && nvp != NULL) {
317		NFSVOPUNLOCK(nvp, 0);
318	} else if (error == EBUSY) {
319		/*
320		 * It is safe so long as a vflush() with
321		 * FORCECLOSE has not been done. Since the Renew thread is
322		 * stopped and the MNTK_UNMOUNTF flag is set before doing
323		 * a vflush() with FORCECLOSE, we should be ok here.
324		 */
325		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
326			error = EINTR;
327		else {
328			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
329			if (nvp == NULL) {
330				error = ENOENT;
331			} else if ((nvp->v_iflag & VI_DOOMED) != 0) {
332				error = ENOENT;
333				vrele(nvp);
334			} else {
335				error = 0;
336			}
337		}
338	}
339	FREE(nfhp, M_NFSFH);
340	if (error)
341		return (error);
342	if (nvp != NULL) {
343		*npp = VTONFS(nvp);
344		return (0);
345	}
346	return (EINVAL);
347}
348
349/*
350 * Load the attribute cache (that lives in the nfsnode entry) with
351 * the attributes of the second argument and
352 * Iff vaper not NULL
353 *    copy the attributes to *vaper
354 * Similar to nfs_loadattrcache(), except the attributes are passed in
355 * instead of being parsed out of the mbuf list.
356 */
357int
358nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
359    void *stuff, int writeattr, int dontshrink)
360{
361	struct vnode *vp = *vpp;
362	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
363	struct nfsnode *np;
364	struct nfsmount *nmp;
365	struct timespec mtime_save;
366	u_quad_t nsize;
367	int setnsize;
368
369	/*
370	 * If v_type == VNON it is a new node, so fill in the v_type,
371	 * n_mtime fields. Check to see if it represents a special
372	 * device, and if so, check for a possible alias. Once the
373	 * correct vnode has been obtained, fill in the rest of the
374	 * information.
375	 */
376	np = VTONFS(vp);
377	NFSLOCKNODE(np);
378	if (vp->v_type != nvap->va_type) {
379		vp->v_type = nvap->va_type;
380		if (vp->v_type == VFIFO)
381			vp->v_op = &newnfs_fifoops;
382		np->n_mtime = nvap->va_mtime;
383	}
384	nmp = VFSTONFS(vp->v_mount);
385	vap = &np->n_vattr.na_vattr;
386	mtime_save = vap->va_mtime;
387	if (writeattr) {
388		np->n_vattr.na_filerev = nap->na_filerev;
389		np->n_vattr.na_size = nap->na_size;
390		np->n_vattr.na_mtime = nap->na_mtime;
391		np->n_vattr.na_ctime = nap->na_ctime;
392		np->n_vattr.na_fsid = nap->na_fsid;
393		np->n_vattr.na_mode = nap->na_mode;
394	} else {
395		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
396		    sizeof (struct nfsvattr));
397	}
398
399	/*
400	 * For NFSv4, if the node's fsid is not equal to the mount point's
401	 * fsid, return the low order 32bits of the node's fsid. This
402	 * allows getcwd(3) to work. There is a chance that the fsid might
403	 * be the same as a local fs, but since this is in an NFS mount
404	 * point, I don't think that will cause any problems?
405	 */
406	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
407	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
408	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
409		/*
410		 * va_fsid needs to be set to some value derived from
411		 * np->n_vattr.na_filesid that is not equal
412		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
413		 * from the value used for the top level server volume
414		 * in the mounted subtree.
415		 */
416		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
417		    (uint32_t)np->n_vattr.na_filesid[0])
418			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
419		else
420			vap->va_fsid = (uint32_t)hash32_buf(
421			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
422	} else
423		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
424	np->n_attrstamp = time_second;
425	setnsize = 0;
426	nsize = 0;
427	if (vap->va_size != np->n_size) {
428		if (vap->va_type == VREG) {
429			if (dontshrink && vap->va_size < np->n_size) {
430				/*
431				 * We've been told not to shrink the file;
432				 * zero np->n_attrstamp to indicate that
433				 * the attributes are stale.
434				 */
435				vap->va_size = np->n_size;
436				np->n_attrstamp = 0;
437				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
438				vnode_pager_setsize(vp, np->n_size);
439			} else if (np->n_flag & NMODIFIED) {
440				/*
441				 * We've modified the file: Use the larger
442				 * of our size, and the server's size.
443				 */
444				if (vap->va_size < np->n_size) {
445					vap->va_size = np->n_size;
446				} else {
447					np->n_size = vap->va_size;
448					np->n_flag |= NSIZECHANGED;
449				}
450				vnode_pager_setsize(vp, np->n_size);
451			} else if (vap->va_size < np->n_size) {
452				/*
453				 * When shrinking the size, the call to
454				 * vnode_pager_setsize() cannot be done
455				 * with the mutex held, so delay it until
456				 * after the mtx_unlock call.
457				 */
458				nsize = np->n_size = vap->va_size;
459				np->n_flag |= NSIZECHANGED;
460				setnsize = 1;
461			} else {
462				np->n_size = vap->va_size;
463				np->n_flag |= NSIZECHANGED;
464				vnode_pager_setsize(vp, np->n_size);
465			}
466		} else {
467			np->n_size = vap->va_size;
468		}
469	}
470	/*
471	 * The following checks are added to prevent a race between (say)
472	 * a READDIR+ and a WRITE.
473	 * READDIR+, WRITE requests sent out.
474	 * READDIR+ resp, WRITE resp received on client.
475	 * However, the WRITE resp was handled before the READDIR+ resp
476	 * causing the post op attrs from the write to be loaded first
477	 * and the attrs from the READDIR+ to be loaded later. If this
478	 * happens, we have stale attrs loaded into the attrcache.
479	 * We detect this by for the mtime moving back. We invalidate the
480	 * attrcache when this happens.
481	 */
482	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
483		/* Size changed or mtime went backwards */
484		np->n_attrstamp = 0;
485		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
486	}
487	if (vaper != NULL) {
488		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
489		if (np->n_flag & NCHG) {
490			if (np->n_flag & NACC)
491				vaper->va_atime = np->n_atim;
492			if (np->n_flag & NUPD)
493				vaper->va_mtime = np->n_mtim;
494		}
495	}
496#ifdef KDTRACE_HOOKS
497	if (np->n_attrstamp != 0)
498		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
499#endif
500	NFSUNLOCKNODE(np);
501	if (setnsize)
502		vnode_pager_setsize(vp, nsize);
503	return (0);
504}
505
506/*
507 * Fill in the client id name. For these bytes:
508 * 1 - they must be unique
509 * 2 - they should be persistent across client reboots
510 * 1 is more critical than 2
511 * Use the mount point's unique id plus either the uuid or, if that
512 * isn't set, random junk.
513 */
514void
515nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
516{
517	int uuidlen;
518
519	/*
520	 * First, put in the 64bit mount point identifier.
521	 */
522	if (idlen >= sizeof (u_int64_t)) {
523		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
524		cp += sizeof (u_int64_t);
525		idlen -= sizeof (u_int64_t);
526	}
527
528	/*
529	 * If uuid is non-zero length, use it.
530	 */
531	uuidlen = strlen(uuid);
532	if (uuidlen > 0 && idlen >= uuidlen) {
533		NFSBCOPY(uuid, cp, uuidlen);
534		cp += uuidlen;
535		idlen -= uuidlen;
536	}
537
538	/*
539	 * This only normally happens if the uuid isn't set.
540	 */
541	while (idlen > 0) {
542		*cp++ = (u_int8_t)(arc4random() % 256);
543		idlen--;
544	}
545}
546
547/*
548 * Fill in a lock owner name. For now, pid + the process's creation time.
549 */
550void
551nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
552{
553	union {
554		u_int32_t	lval;
555		u_int8_t	cval[4];
556	} tl;
557	struct proc *p;
558
559	if (id == NULL) {
560		/* Return the single open_owner of all 0 bytes. */
561		bzero(cp, NFSV4CL_LOCKNAMELEN);
562		return;
563	}
564	if ((flags & F_POSIX) != 0) {
565		p = (struct proc *)id;
566		tl.lval = p->p_pid;
567		*cp++ = tl.cval[0];
568		*cp++ = tl.cval[1];
569		*cp++ = tl.cval[2];
570		*cp++ = tl.cval[3];
571		tl.lval = p->p_stats->p_start.tv_sec;
572		*cp++ = tl.cval[0];
573		*cp++ = tl.cval[1];
574		*cp++ = tl.cval[2];
575		*cp++ = tl.cval[3];
576		tl.lval = p->p_stats->p_start.tv_usec;
577		*cp++ = tl.cval[0];
578		*cp++ = tl.cval[1];
579		*cp++ = tl.cval[2];
580		*cp = tl.cval[3];
581	} else if ((flags & F_FLOCK) != 0) {
582		bcopy(&id, cp, sizeof(id));
583		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
584	} else {
585		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
586		bzero(cp, NFSV4CL_LOCKNAMELEN);
587	}
588}
589
590/*
591 * Find the parent process for the thread passed in as an argument.
592 * If none exists, return NULL, otherwise return a thread for the parent.
593 * (Can be any of the threads, since it is only used for td->td_proc.)
594 */
595NFSPROC_T *
596nfscl_getparent(struct thread *td)
597{
598	struct proc *p;
599	struct thread *ptd;
600
601	if (td == NULL)
602		return (NULL);
603	p = td->td_proc;
604	if (p->p_pid == 0)
605		return (NULL);
606	p = p->p_pptr;
607	if (p == NULL)
608		return (NULL);
609	ptd = TAILQ_FIRST(&p->p_threads);
610	return (ptd);
611}
612
613/*
614 * Start up the renew kernel thread.
615 */
616static void
617start_nfscl(void *arg)
618{
619	struct nfsclclient *clp;
620	struct thread *td;
621
622	clp = (struct nfsclclient *)arg;
623	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
624	nfscl_renewthread(clp, td);
625	kproc_exit(0);
626}
627
628void
629nfscl_start_renewthread(struct nfsclclient *clp)
630{
631
632	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
633	    "nfscl");
634}
635
636/*
637 * Handle wcc_data.
638 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
639 * as the first Op after PutFH.
640 * (For NFSv4, the postop attributes are after the Op, so they can't be
641 *  parsed here. A separate call to nfscl_postop_attr() is required.)
642 */
643int
644nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
645    struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
646{
647	u_int32_t *tl;
648	struct nfsnode *np = VTONFS(vp);
649	struct nfsvattr nfsva;
650	int error = 0;
651
652	if (wccflagp != NULL)
653		*wccflagp = 0;
654	if (nd->nd_flag & ND_NFSV3) {
655		*flagp = 0;
656		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
657		if (*tl == newnfs_true) {
658			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
659			if (wccflagp != NULL) {
660				mtx_lock(&np->n_mtx);
661				*wccflagp = (np->n_mtime.tv_sec ==
662				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
663				    np->n_mtime.tv_nsec ==
664				    fxdr_unsigned(u_int32_t, *(tl + 3)));
665				mtx_unlock(&np->n_mtx);
666			}
667		}
668		error = nfscl_postop_attr(nd, nap, flagp, stuff);
669	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
670	    == (ND_NFSV4 | ND_V4WCCATTR)) {
671		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
672		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
673		    NULL, NULL, NULL, NULL, NULL);
674		if (error)
675			return (error);
676		/*
677		 * Get rid of Op# and status for next op.
678		 */
679		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
680		if (*++tl)
681			nd->nd_flag |= ND_NOMOREDATA;
682		if (wccflagp != NULL &&
683		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
684			mtx_lock(&np->n_mtx);
685			*wccflagp = (np->n_mtime.tv_sec ==
686			    nfsva.na_vattr.va_mtime.tv_sec &&
687			    np->n_mtime.tv_nsec ==
688			    nfsva.na_vattr.va_mtime.tv_sec);
689			mtx_unlock(&np->n_mtx);
690		}
691	}
692nfsmout:
693	return (error);
694}
695
696/*
697 * Get postop attributes.
698 */
699int
700nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
701    void *stuff)
702{
703	u_int32_t *tl;
704	int error = 0;
705
706	*retp = 0;
707	if (nd->nd_flag & ND_NOMOREDATA)
708		return (error);
709	if (nd->nd_flag & ND_NFSV3) {
710		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
711		*retp = fxdr_unsigned(int, *tl);
712	} else if (nd->nd_flag & ND_NFSV4) {
713		/*
714		 * For NFSv4, the postop attr are at the end, so no point
715		 * in looking if nd_repstat != 0.
716		 */
717		if (!nd->nd_repstat) {
718			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
719			if (*(tl + 1))
720				/* should never happen since nd_repstat != 0 */
721				nd->nd_flag |= ND_NOMOREDATA;
722			else
723				*retp = 1;
724		}
725	} else if (!nd->nd_repstat) {
726		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
727		*retp = 1;
728	}
729	if (*retp) {
730		error = nfsm_loadattr(nd, nap);
731		if (error)
732			*retp = 0;
733	}
734nfsmout:
735	return (error);
736}
737
738/*
739 * Fill in the setable attributes. The full argument indicates whether
740 * to fill in them all or just mode and time.
741 */
742void
743nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
744    struct vnode *vp, int flags, u_int32_t rdev)
745{
746	u_int32_t *tl;
747	struct nfsv2_sattr *sp;
748	nfsattrbit_t attrbits;
749
750	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
751	case ND_NFSV2:
752		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
753		if (vap->va_mode == (mode_t)VNOVAL)
754			sp->sa_mode = newnfs_xdrneg1;
755		else
756			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
757		if (vap->va_uid == (uid_t)VNOVAL)
758			sp->sa_uid = newnfs_xdrneg1;
759		else
760			sp->sa_uid = txdr_unsigned(vap->va_uid);
761		if (vap->va_gid == (gid_t)VNOVAL)
762			sp->sa_gid = newnfs_xdrneg1;
763		else
764			sp->sa_gid = txdr_unsigned(vap->va_gid);
765		if (flags & NFSSATTR_SIZE0)
766			sp->sa_size = 0;
767		else if (flags & NFSSATTR_SIZENEG1)
768			sp->sa_size = newnfs_xdrneg1;
769		else if (flags & NFSSATTR_SIZERDEV)
770			sp->sa_size = txdr_unsigned(rdev);
771		else
772			sp->sa_size = txdr_unsigned(vap->va_size);
773		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
774		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
775		break;
776	case ND_NFSV3:
777		if (vap->va_mode != (mode_t)VNOVAL) {
778			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
779			*tl++ = newnfs_true;
780			*tl = txdr_unsigned(vap->va_mode);
781		} else {
782			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
783			*tl = newnfs_false;
784		}
785		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
786			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
787			*tl++ = newnfs_true;
788			*tl = txdr_unsigned(vap->va_uid);
789		} else {
790			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
791			*tl = newnfs_false;
792		}
793		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
794			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
795			*tl++ = newnfs_true;
796			*tl = txdr_unsigned(vap->va_gid);
797		} else {
798			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
799			*tl = newnfs_false;
800		}
801		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
802			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
803			*tl++ = newnfs_true;
804			txdr_hyper(vap->va_size, tl);
805		} else {
806			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
807			*tl = newnfs_false;
808		}
809		if (vap->va_atime.tv_sec != VNOVAL) {
810			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
811				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
812				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
813				txdr_nfsv3time(&vap->va_atime, tl);
814			} else {
815				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
816				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
817			}
818		} else {
819			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
820			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
821		}
822		if (vap->va_mtime.tv_sec != VNOVAL) {
823			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
824				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
825				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
826				txdr_nfsv3time(&vap->va_mtime, tl);
827			} else {
828				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
829				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
830			}
831		} else {
832			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
833			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
834		}
835		break;
836	case ND_NFSV4:
837		NFSZERO_ATTRBIT(&attrbits);
838		if (vap->va_mode != (mode_t)VNOVAL)
839			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
840		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
841			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
842		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
843			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
844		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
845			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
846		if (vap->va_atime.tv_sec != VNOVAL)
847			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
848		if (vap->va_mtime.tv_sec != VNOVAL)
849			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
850		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
851		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
852		break;
853	};
854}
855
856/*
857 * nfscl_request() - mostly a wrapper for newnfs_request().
858 */
859int
860nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
861    struct ucred *cred, void *stuff)
862{
863	int ret, vers;
864	struct nfsmount *nmp;
865
866	nmp = VFSTONFS(vp->v_mount);
867	if (nd->nd_flag & ND_NFSV4)
868		vers = NFS_VER4;
869	else if (nd->nd_flag & ND_NFSV3)
870		vers = NFS_VER3;
871	else
872		vers = NFS_VER2;
873	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
874		NFS_PROG, vers, NULL, 1, NULL, NULL);
875	return (ret);
876}
877
878/*
879 * fill in this bsden's variant of statfs using nfsstatfs.
880 */
881void
882nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
883{
884	struct statfs *sbp = (struct statfs *)statfs;
885
886	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
887		sbp->f_bsize = NFS_FABLKSIZE;
888		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
889		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
890		/*
891		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
892		 * the value after dividing by NFS_FABLKSIZE is small
893		 * enough that it will fit in 63bits, so it is ok to
894		 * assign it to f_bavail without fear that it will become
895		 * negative.
896		 */
897		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
898		sbp->f_files = sfp->sf_tfiles;
899		/* Since f_ffree is int64_t, clip it to 63bits. */
900		if (sfp->sf_ffiles > INT64_MAX)
901			sbp->f_ffree = INT64_MAX;
902		else
903			sbp->f_ffree = sfp->sf_ffiles;
904	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
905		/*
906		 * The type casts to (int32_t) ensure that this code is
907		 * compatible with the old NFS client, in that it will
908		 * propagate bit31 to the high order bits. This may or may
909		 * not be correct for NFSv2, but since it is a legacy
910		 * environment, I'd rather retain backwards compatibility.
911		 */
912		sbp->f_bsize = (int32_t)sfp->sf_bsize;
913		sbp->f_blocks = (int32_t)sfp->sf_blocks;
914		sbp->f_bfree = (int32_t)sfp->sf_bfree;
915		sbp->f_bavail = (int32_t)sfp->sf_bavail;
916		sbp->f_files = 0;
917		sbp->f_ffree = 0;
918	}
919}
920
921/*
922 * Use the fsinfo stuff to update the mount point.
923 */
924void
925nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
926{
927
928	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
929	    fsp->fs_wtpref >= NFS_FABLKSIZE)
930		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
931		    ~(NFS_FABLKSIZE - 1);
932	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
933		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
934		if (nmp->nm_wsize == 0)
935			nmp->nm_wsize = fsp->fs_wtmax;
936	}
937	if (nmp->nm_wsize < NFS_FABLKSIZE)
938		nmp->nm_wsize = NFS_FABLKSIZE;
939	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
940	    fsp->fs_rtpref >= NFS_FABLKSIZE)
941		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
942		    ~(NFS_FABLKSIZE - 1);
943	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
944		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
945		if (nmp->nm_rsize == 0)
946			nmp->nm_rsize = fsp->fs_rtmax;
947	}
948	if (nmp->nm_rsize < NFS_FABLKSIZE)
949		nmp->nm_rsize = NFS_FABLKSIZE;
950	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
951	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
952		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
953		    ~(NFS_DIRBLKSIZ - 1);
954	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
955		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
956		if (nmp->nm_readdirsize == 0)
957			nmp->nm_readdirsize = fsp->fs_rtmax;
958	}
959	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
960		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
961	if (fsp->fs_maxfilesize > 0 &&
962	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
963		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
964	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
965	nmp->nm_state |= NFSSTA_GOTFSINFO;
966}
967
968/*
969 * Get a pointer to my IP addrress and return it.
970 * Return NULL if you can't find one.
971 */
972u_int8_t *
973nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
974{
975	struct sockaddr_in sad, *sin;
976	struct rtentry *rt;
977	u_int8_t *retp = NULL;
978	static struct in_addr laddr;
979
980	*isinet6p = 0;
981	/*
982	 * Loop up a route for the destination address.
983	 */
984	if (nmp->nm_nam->sa_family == AF_INET) {
985		bzero(&sad, sizeof (sad));
986		sin = (struct sockaddr_in *)nmp->nm_nam;
987		sad.sin_family = AF_INET;
988		sad.sin_len = sizeof (struct sockaddr_in);
989		sad.sin_addr.s_addr = sin->sin_addr.s_addr;
990		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
991		rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
992		     curthread->td_proc->p_fibnum);
993		if (rt != NULL) {
994			if (rt->rt_ifp != NULL &&
995			    rt->rt_ifa != NULL &&
996			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
997			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
998				sin = (struct sockaddr_in *)
999				    rt->rt_ifa->ifa_addr;
1000				laddr.s_addr = sin->sin_addr.s_addr;
1001				retp = (u_int8_t *)&laddr;
1002			}
1003			RTFREE_LOCKED(rt);
1004		}
1005		CURVNET_RESTORE();
1006#ifdef INET6
1007	} else if (nmp->nm_nam->sa_family == AF_INET6) {
1008		struct sockaddr_in6 sad6, *sin6;
1009		static struct in6_addr laddr6;
1010
1011		bzero(&sad6, sizeof (sad6));
1012		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1013		sad6.sin6_family = AF_INET6;
1014		sad6.sin6_len = sizeof (struct sockaddr_in6);
1015		sad6.sin6_addr = sin6->sin6_addr;
1016		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1017		rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1018		     curthread->td_proc->p_fibnum);
1019		if (rt != NULL) {
1020			if (rt->rt_ifp != NULL &&
1021			    rt->rt_ifa != NULL &&
1022			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1023			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1024				sin6 = (struct sockaddr_in6 *)
1025				    rt->rt_ifa->ifa_addr;
1026				laddr6 = sin6->sin6_addr;
1027				retp = (u_int8_t *)&laddr6;
1028				*isinet6p = 1;
1029			}
1030			RTFREE_LOCKED(rt);
1031		}
1032		CURVNET_RESTORE();
1033#endif
1034	}
1035	return (retp);
1036}
1037
1038/*
1039 * Copy NFS uid, gids from the cred structure.
1040 */
1041void
1042newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1043{
1044	int i;
1045
1046	KASSERT(cr->cr_ngroups >= 0,
1047	    ("newnfs_copyincred: negative cr_ngroups"));
1048	nfscr->nfsc_uid = cr->cr_uid;
1049	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1050	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1051		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1052}
1053
1054
1055/*
1056 * Do any client specific initialization.
1057 */
1058void
1059nfscl_init(void)
1060{
1061	static int inited = 0;
1062
1063	if (inited)
1064		return;
1065	inited = 1;
1066	nfscl_inited = 1;
1067	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1068}
1069
1070/*
1071 * Check each of the attributes to be set, to ensure they aren't already
1072 * the correct value. Disable setting ones already correct.
1073 */
1074int
1075nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1076{
1077
1078	if (vap->va_mode != (mode_t)VNOVAL) {
1079		if (vap->va_mode == nvap->na_mode)
1080			vap->va_mode = (mode_t)VNOVAL;
1081	}
1082	if (vap->va_uid != (uid_t)VNOVAL) {
1083		if (vap->va_uid == nvap->na_uid)
1084			vap->va_uid = (uid_t)VNOVAL;
1085	}
1086	if (vap->va_gid != (gid_t)VNOVAL) {
1087		if (vap->va_gid == nvap->na_gid)
1088			vap->va_gid = (gid_t)VNOVAL;
1089	}
1090	if (vap->va_size != VNOVAL) {
1091		if (vap->va_size == nvap->na_size)
1092			vap->va_size = VNOVAL;
1093	}
1094
1095	/*
1096	 * We are normally called with only a partially initialized
1097	 * VAP.  Since the NFSv3 spec says that server may use the
1098	 * file attributes to store the verifier, the spec requires
1099	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1100	 * in atime, but we can't really assume that all servers will
1101	 * so we ensure that our SETATTR sets both atime and mtime.
1102	 * Set the VA_UTIMES_NULL flag for this case, so that
1103	 * the server's time will be used.  This is needed to
1104	 * work around a bug in some Solaris servers, where
1105	 * setting the time TOCLIENT causes the Setattr RPC
1106	 * to return NFS_OK, but not set va_mode.
1107	 */
1108	if (vap->va_mtime.tv_sec == VNOVAL) {
1109		vfs_timestamp(&vap->va_mtime);
1110		vap->va_vaflags |= VA_UTIMES_NULL;
1111	}
1112	if (vap->va_atime.tv_sec == VNOVAL)
1113		vap->va_atime = vap->va_mtime;
1114	return (1);
1115}
1116
1117/*
1118 * Map nfsv4 errors to errno.h errors.
1119 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1120 * error should only be returned for the Open, Create and Setattr Ops.
1121 * As such, most calls can just pass in 0 for those arguments.
1122 */
1123APPLESTATIC int
1124nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1125{
1126	struct proc *p;
1127
1128	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1129		return (error);
1130	if (td != NULL)
1131		p = td->td_proc;
1132	else
1133		p = NULL;
1134	switch (error) {
1135	case NFSERR_BADOWNER:
1136		tprintf(p, LOG_INFO,
1137		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1138		    uid, gid);
1139		return (EPERM);
1140	case NFSERR_BADNAME:
1141	case NFSERR_BADCHAR:
1142		printf("nfsv4 char/name not handled by server\n");
1143		return (ENOENT);
1144	case NFSERR_STALECLIENTID:
1145	case NFSERR_STALESTATEID:
1146	case NFSERR_EXPIRED:
1147	case NFSERR_BADSTATEID:
1148	case NFSERR_BADSESSION:
1149		printf("nfsv4 recover err returned %d\n", error);
1150		return (EIO);
1151	case NFSERR_BADHANDLE:
1152	case NFSERR_SERVERFAULT:
1153	case NFSERR_BADTYPE:
1154	case NFSERR_FHEXPIRED:
1155	case NFSERR_RESOURCE:
1156	case NFSERR_MOVED:
1157	case NFSERR_NOFILEHANDLE:
1158	case NFSERR_MINORVERMISMATCH:
1159	case NFSERR_OLDSTATEID:
1160	case NFSERR_BADSEQID:
1161	case NFSERR_LEASEMOVED:
1162	case NFSERR_RECLAIMBAD:
1163	case NFSERR_BADXDR:
1164	case NFSERR_OPILLEGAL:
1165		printf("nfsv4 client/server protocol prob err=%d\n",
1166		    error);
1167		return (EIO);
1168	default:
1169		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1170		return (EIO);
1171	};
1172}
1173
1174/*
1175 * Check to see if the process for this owner exists. Return 1 if it doesn't
1176 * and 0 otherwise.
1177 */
1178int
1179nfscl_procdoesntexist(u_int8_t *own)
1180{
1181	union {
1182		u_int32_t	lval;
1183		u_int8_t	cval[4];
1184	} tl;
1185	struct proc *p;
1186	pid_t pid;
1187	int i, ret = 0;
1188
1189	/* For the single open_owner of all 0 bytes, just return 0. */
1190	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1191		if (own[i] != 0)
1192			break;
1193	if (i == NFSV4CL_LOCKNAMELEN)
1194		return (0);
1195
1196	tl.cval[0] = *own++;
1197	tl.cval[1] = *own++;
1198	tl.cval[2] = *own++;
1199	tl.cval[3] = *own++;
1200	pid = tl.lval;
1201	p = pfind_locked(pid);
1202	if (p == NULL)
1203		return (1);
1204	if (p->p_stats == NULL) {
1205		PROC_UNLOCK(p);
1206		return (0);
1207	}
1208	tl.cval[0] = *own++;
1209	tl.cval[1] = *own++;
1210	tl.cval[2] = *own++;
1211	tl.cval[3] = *own++;
1212	if (tl.lval != p->p_stats->p_start.tv_sec) {
1213		ret = 1;
1214	} else {
1215		tl.cval[0] = *own++;
1216		tl.cval[1] = *own++;
1217		tl.cval[2] = *own++;
1218		tl.cval[3] = *own;
1219		if (tl.lval != p->p_stats->p_start.tv_usec)
1220			ret = 1;
1221	}
1222	PROC_UNLOCK(p);
1223	return (ret);
1224}
1225
1226/*
1227 * - nfs pseudo system call for the client
1228 */
1229/*
1230 * MPSAFE
1231 */
1232static int
1233nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1234{
1235	struct file *fp;
1236	struct nfscbd_args nfscbdarg;
1237	struct nfsd_nfscbd_args nfscbdarg2;
1238	struct nameidata nd;
1239	struct nfscl_dumpmntopts dumpmntopts;
1240	cap_rights_t rights;
1241	char *buf;
1242	int error;
1243
1244	if (uap->flag & NFSSVC_CBADDSOCK) {
1245		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1246		if (error)
1247			return (error);
1248		/*
1249		 * Since we don't know what rights might be required,
1250		 * pretend that we need them all. It is better to be too
1251		 * careful than too reckless.
1252		 */
1253		error = fget(td, nfscbdarg.sock,
1254		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1255		if (error)
1256			return (error);
1257		if (fp->f_type != DTYPE_SOCKET) {
1258			fdrop(fp, td);
1259			return (EPERM);
1260		}
1261		error = nfscbd_addsock(fp);
1262		fdrop(fp, td);
1263		if (!error && nfscl_enablecallb == 0) {
1264			nfsv4_cbport = nfscbdarg.port;
1265			nfscl_enablecallb = 1;
1266		}
1267	} else if (uap->flag & NFSSVC_NFSCBD) {
1268		if (uap->argp == NULL)
1269			return (EINVAL);
1270		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1271		    sizeof(nfscbdarg2));
1272		if (error)
1273			return (error);
1274		error = nfscbd_nfsd(td, &nfscbdarg2);
1275	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1276		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1277		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1278		    dumpmntopts.ndmnt_blen > 1024))
1279			error = EINVAL;
1280		if (error == 0)
1281			error = nfsrv_lookupfilename(&nd,
1282			    dumpmntopts.ndmnt_fname, td);
1283		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1284		    "nfs") != 0) {
1285			vput(nd.ni_vp);
1286			error = EINVAL;
1287		}
1288		if (error == 0) {
1289			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1290			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1291			    dumpmntopts.ndmnt_blen);
1292			vput(nd.ni_vp);
1293			error = copyout(buf, dumpmntopts.ndmnt_buf,
1294			    dumpmntopts.ndmnt_blen);
1295			free(buf, M_TEMP);
1296		}
1297	} else {
1298		error = EINVAL;
1299	}
1300	return (error);
1301}
1302
1303extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1304
1305/*
1306 * Called once to initialize data structures...
1307 */
1308static int
1309nfscl_modevent(module_t mod, int type, void *data)
1310{
1311	int error = 0;
1312	static int loaded = 0;
1313
1314	switch (type) {
1315	case MOD_LOAD:
1316		if (loaded)
1317			return (0);
1318		newnfs_portinit();
1319		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1320		nfscl_init();
1321		NFSD_LOCK();
1322		nfsrvd_cbinit(0);
1323		NFSD_UNLOCK();
1324		ncl_call_invalcaches = ncl_invalcaches;
1325		nfsd_call_nfscl = nfssvc_nfscl;
1326		loaded = 1;
1327		break;
1328
1329	case MOD_UNLOAD:
1330		if (nfs_numnfscbd != 0) {
1331			error = EBUSY;
1332			break;
1333		}
1334
1335		/*
1336		 * XXX: Unloading of nfscl module is unsupported.
1337		 */
1338#if 0
1339		ncl_call_invalcaches = NULL;
1340		nfsd_call_nfscl = NULL;
1341		/* and get rid of the mutexes */
1342		mtx_destroy(&ncl_iod_mutex);
1343		loaded = 0;
1344		break;
1345#else
1346		/* FALLTHROUGH */
1347#endif
1348	default:
1349		error = EOPNOTSUPP;
1350		break;
1351	}
1352	return error;
1353}
1354static moduledata_t nfscl_mod = {
1355	"nfscl",
1356	nfscl_modevent,
1357	NULL,
1358};
1359DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1360
1361/* So that loader and kldload(2) can find us, wherever we are.. */
1362MODULE_VERSION(nfscl, 1);
1363MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1364MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1365MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1366MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1367
1368