xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 2c186228096a01f940fcc323814826572e0a156b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/capsicum.h>
37 #include <sys/extattr.h>
38 
39 /*
40  * Functions that perform the vfs operations required by the routines in
41  * nfsd_serv.c. It is hoped that this change will make the server more
42  * portable.
43  */
44 
45 #include <fs/nfs/nfsport.h>
46 #include <security/mac/mac_framework.h>
47 #include <sys/callout.h>
48 #include <sys/filio.h>
49 #include <sys/hash.h>
50 #include <sys/osd.h>
51 #include <sys/sysctl.h>
52 #include <nlm/nlm_prot.h>
53 #include <nlm/nlm.h>
54 #include <vm/vm_param.h>
55 #include <vm/vnode_pager.h>
56 
57 FEATURE(nfsd, "NFSv4 server");
58 
59 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
60 extern int nfsrv_useacl;
61 extern int newnfs_numnfsd;
62 extern int nfsrv_sessionhashsize;
63 extern struct nfslayouthash *nfslayouthash;
64 extern int nfsrv_layouthashsize;
65 extern struct mtx nfsrv_dslock_mtx;
66 extern int nfs_pnfsiothreads;
67 extern volatile int nfsrv_devidcnt;
68 extern int nfsrv_maxpnfsmirror;
69 extern uint32_t nfs_srvmaxio;
70 extern int nfs_bufpackets;
71 extern u_long sb_max_adj;
72 extern struct nfsv4lock nfsv4rootfs_lock;
73 
74 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
75 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst);
76 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool);
77 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash);
78 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash);
79 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash);
80 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock);
81 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
82 
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct mtx nfsrc_udpmtx;
86 struct mtx nfs_v4root_mutex;
87 struct mtx nfsrv_dontlistlock_mtx;
88 struct mtx nfsrv_recalllock_mtx;
89 struct nfsrvfh nfs_pubfh;
90 int nfs_pubfhset = 0;
91 int nfsd_debuglevel = 0;
92 static pid_t nfsd_master_pid = (pid_t)-1;
93 static char nfsd_master_comm[MAXCOMLEN + 1];
94 static struct timeval nfsd_master_start;
95 static uint32_t nfsv4_sysid = 0;
96 static fhandle_t zerofh;
97 
98 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL;
99 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl);
100 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table);
101 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table);
102 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh);
103 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0;
104 NFSD_VNET_DEFINE(struct callout, nfsd_callout);
105 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt);
106 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt);
107 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt);
108 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false;
109 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false;
110 
111 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
112     struct ucred *);
113 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *);
114 
115 int nfsrv_enable_crossmntpt = 1;
116 static int nfs_commit_blks;
117 static int nfs_commit_miss;
118 extern int nfsrv_issuedelegs;
119 extern int nfsrv_dolocallocks;
120 extern struct nfsdevicehead nfsrv_devidhead;
121 
122 /* Map d_type to vnode type. */
123 static uint8_t dtype_to_vnode[DT_WHT + 1] = { VNON, VFIFO, VCHR, VNON, VDIR,
124     VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON };
125 
126 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
127     struct iovec **);
128 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
129     struct mbuf **, struct iovec **);
130 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
131     int *);
132 static void nfs_dtypetovtype(struct nfsvattr *, struct vnode *, uint8_t);
133 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
134     NFSPROC_T *);
135 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
136     int *, char *, fhandle_t *);
137 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
138     NFSPROC_T *);
139 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
140     struct thread *, int, struct mbuf **, char *, struct mbuf **,
141     struct nfsvattr *, struct acl *, off_t *, int, bool *);
142 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
143 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
144     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
145 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
146     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
147     char *, int *);
148 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
149     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
150 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
151     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
152 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
153     struct vnode *, struct nfsmount **, int, struct acl *, int *);
154 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
155     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
156 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
157     struct vnode *, struct nfsmount *, struct nfsvattr *);
158 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
159     NFSPROC_T *, struct nfsmount *);
160 static int nfsrv_putfhname(fhandle_t *, char *);
161 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
162     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
163 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
164     struct vnode *, NFSPROC_T *);
165 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
166 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
167     NFSPROC_T *);
168 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
169 
170 int nfs_pnfsio(task_fn_t *, void *);
171 
172 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
173     "NFS server");
174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
175     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
176 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
177     0, "");
178 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
179     0, "");
180 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
181     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
182 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
183     0, "Debug level for NFS server");
184 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid);
185 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid,
186     CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid),
187     0, "Enable nfsd to accept numeric owner_names");
188 static int nfsrv_pnfsgetdsattr = 1;
189 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
190     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
191 static bool nfsrv_recalldeleg = false;
192 SYSCTL_BOOL(_vfs_nfsd, OID_AUTO, recalldeleg, CTLFLAG_RW,
193     &nfsrv_recalldeleg, 0,
194     "When set remove/rename recalls delegations for same client");
195 
196 /*
197  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
198  * not running.
199  * The dsN subdirectories for the increased values must have been created
200  * on all DS servers before this increase is done.
201  */
202 u_int	nfsrv_dsdirsize = 20;
203 static int
sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)204 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
205 {
206 	int error, newdsdirsize;
207 
208 	newdsdirsize = nfsrv_dsdirsize;
209 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
210 	if (error != 0 || req->newptr == NULL)
211 		return (error);
212 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
213 	    newnfs_numnfsd != 0)
214 		return (EINVAL);
215 	nfsrv_dsdirsize = newdsdirsize;
216 	return (0);
217 }
218 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
219     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
220     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
221 
222 /*
223  * nfs_srvmaxio can only be increased and only when the nfsd threads are
224  * not running.  The setting must be a power of 2, with the current limit of
225  * 1Mbyte.
226  */
227 static int
sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)228 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
229 {
230 	int error;
231 	u_int newsrvmaxio;
232 	uint64_t tval;
233 
234 	newsrvmaxio = nfs_srvmaxio;
235 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
236 	if (error != 0 || req->newptr == NULL)
237 		return (error);
238 	if (newsrvmaxio == nfs_srvmaxio)
239 		return (0);
240 	if (newsrvmaxio < nfs_srvmaxio) {
241 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
242 		return (EINVAL);
243 	}
244 	if (newsrvmaxio > 1048576) {
245 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
246 		return (EINVAL);
247 	}
248 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
249 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
250 		return (EINVAL);
251 	}
252 
253 	/*
254 	 * Check that kern.ipc.maxsockbuf is large enough for
255 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
256 	 */
257 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
258 	    sb_max_adj) {
259 		/*
260 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
261 		 * sb_max_adj.
262 		 */
263 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
264 
265 		/*
266 		 * Convert suggested sb_max_adj value to a suggested
267 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
268 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
269 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
270 		 *     (MSIZE + MCLBYTES);
271 		 * XXX If the calculation of sb_max_adj from sb_max changes,
272 		 *     this calculation must be changed as well.
273 		 */
274 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
275 		tval += MCLBYTES - 1;        /* Round up divide. */
276 		tval /= MCLBYTES;
277 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
278 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
279 		    newsrvmaxio);
280 		return (EINVAL);
281 	}
282 
283 	NFSD_LOCK();
284 	if (newnfs_numnfsd != 0) {
285 		NFSD_UNLOCK();
286 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
287 		    "threads are running\n");
288 		return (EINVAL);
289 	}
290 
291 
292 	nfs_srvmaxio = newsrvmaxio;
293 	NFSD_UNLOCK();
294 	return (0);
295 }
296 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
297     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
298     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
299 
300 static int
sysctl_dolocallocks(SYSCTL_HANDLER_ARGS)301 sysctl_dolocallocks(SYSCTL_HANDLER_ARGS)
302 {
303 	int error, igotlock, newdolocallocks;
304 
305 	newdolocallocks = nfsrv_dolocallocks;
306 	error = sysctl_handle_int(oidp, &newdolocallocks, 0, req);
307 	if (error != 0 || req->newptr == NULL)
308 		return (error);
309 	if (newdolocallocks == nfsrv_dolocallocks)
310 		return (0);
311 	if (jailed(curthread->td_ucred))
312 		return (EINVAL);
313 
314 	NFSLOCKV4ROOTMUTEX();
315 	do {
316 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
317 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
318 	} while (!igotlock);
319 	NFSUNLOCKV4ROOTMUTEX();
320 
321 	nfsrv_dolocallocks = newdolocallocks;
322 
323 	NFSLOCKV4ROOTMUTEX();
324 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
325 	NFSUNLOCKV4ROOTMUTEX();
326 	return (0);
327 }
328 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, enable_locallocks,
329     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
330     sysctl_dolocallocks, "IU", "Enable nfsd to acquire local locks on files");
331 
332 #define	MAX_REORDERED_RPC	16
333 #define	NUM_HEURISTIC		1031
334 #define	NHUSE_INIT		64
335 #define	NHUSE_INC		16
336 #define	NHUSE_MAX		2048
337 
338 static struct nfsheur {
339 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
340 	off_t nh_nextoff;	/* next offset for sequential detection */
341 	int nh_use;		/* use count for selection */
342 	int nh_seqcount;	/* heuristic */
343 } nfsheur[NUM_HEURISTIC];
344 
345 /*
346  * Heuristic to detect sequential operation.
347  */
348 static struct nfsheur *
nfsrv_sequential_heuristic(struct uio * uio,struct vnode * vp)349 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
350 {
351 	struct nfsheur *nh;
352 	int hi, try;
353 
354 	/* Locate best candidate. */
355 	try = 32;
356 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
357 	nh = &nfsheur[hi];
358 	while (try--) {
359 		if (nfsheur[hi].nh_vp == vp) {
360 			nh = &nfsheur[hi];
361 			break;
362 		}
363 		if (nfsheur[hi].nh_use > 0)
364 			--nfsheur[hi].nh_use;
365 		hi = (hi + 1) % NUM_HEURISTIC;
366 		if (nfsheur[hi].nh_use < nh->nh_use)
367 			nh = &nfsheur[hi];
368 	}
369 
370 	/* Initialize hint if this is a new file. */
371 	if (nh->nh_vp != vp) {
372 		nh->nh_vp = vp;
373 		nh->nh_nextoff = uio->uio_offset;
374 		nh->nh_use = NHUSE_INIT;
375 		if (uio->uio_offset == 0)
376 			nh->nh_seqcount = 4;
377 		else
378 			nh->nh_seqcount = 1;
379 	}
380 
381 	/* Calculate heuristic. */
382 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
383 	    uio->uio_offset == nh->nh_nextoff) {
384 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
385 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
386 		if (nh->nh_seqcount > IO_SEQMAX)
387 			nh->nh_seqcount = IO_SEQMAX;
388 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
389 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
390 		/* Probably a reordered RPC, leave seqcount alone. */
391 	} else if (nh->nh_seqcount > 1) {
392 		nh->nh_seqcount /= 2;
393 	} else {
394 		nh->nh_seqcount = 0;
395 	}
396 	nh->nh_use += NHUSE_INC;
397 	if (nh->nh_use > NHUSE_MAX)
398 		nh->nh_use = NHUSE_MAX;
399 	return (nh);
400 }
401 
402 /*
403  * Get attributes into nfsvattr structure.
404  */
405 int
nfsvno_getattr(struct vnode * vp,struct nfsvattr * nvap,struct nfsrv_descript * nd,struct thread * p,int vpislocked,nfsattrbit_t * attrbitp)406 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
407     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
408     nfsattrbit_t *attrbitp)
409 {
410 	int error, gotattr, lockedit = 0;
411 	struct nfsvattr na;
412 
413 	if (vpislocked == 0) {
414 		/*
415 		 * When vpislocked == 0, the vnode is either exclusively
416 		 * locked by this thread or not locked by this thread.
417 		 * As such, shared lock it, if not exclusively locked.
418 		 */
419 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
420 			lockedit = 1;
421 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
422 		}
423 	}
424 
425 	/*
426 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
427 	 * attributes, as required.
428 	 * This needs to be done for regular files if:
429 	 * - non-NFSv4 RPCs or
430 	 * - when attrbitp == NULL or
431 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
432 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
433 	 * these attributes.  nfsrv_proxyds() will return an error if the
434 	 * server is not a pNFS one.
435 	 */
436 	gotattr = 0;
437 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
438 	    (nd->nd_flag & ND_NFSV4) == 0 ||
439 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
440 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
441 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
442 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
443 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
444 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
445 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
446 		    NULL);
447 		if (error == 0)
448 			gotattr = 1;
449 	}
450 
451 	nvap->na_bsdflags = 0;
452 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
453 	if (lockedit != 0)
454 		NFSVOPUNLOCK(vp);
455 
456 	/*
457 	 * If we got the Change, Size and Modify Time from the DS,
458 	 * replace them.
459 	 */
460 	if (gotattr != 0) {
461 		nvap->na_atime = na.na_atime;
462 		nvap->na_mtime = na.na_mtime;
463 		nvap->na_filerev = na.na_filerev;
464 		nvap->na_size = na.na_size;
465 		nvap->na_bytes = na.na_bytes;
466 	}
467 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
468 	    error, (uintmax_t)na.na_filerev);
469 
470 	NFSEXITCODE(error);
471 	return (error);
472 }
473 
474 /*
475  * Get a file handle for a vnode.
476  */
477 int
nfsvno_getfh(struct vnode * vp,fhandle_t * fhp,struct thread * p)478 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
479 {
480 	int error;
481 
482 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
483 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
484 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
485 
486 	NFSEXITCODE(error);
487 	return (error);
488 }
489 
490 /*
491  * Perform access checking for vnodes obtained from file handles that would
492  * refer to files already opened by a Unix client. You cannot just use
493  * vn_writechk() and VOP_ACCESSX() for two reasons.
494  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
495  *     case.
496  * 2 - The owner is to be given access irrespective of mode bits for some
497  *     operations, so that processes that chmod after opening a file don't
498  *     break.
499  */
500 int
nfsvno_accchk(struct vnode * vp,accmode_t accmode,struct ucred * cred,struct nfsexstuff * exp,struct thread * p,int override,int vpislocked,u_int32_t * supportedtypep)501 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
502     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
503     u_int32_t *supportedtypep)
504 {
505 	struct vattr vattr;
506 	int error = 0, getret = 0;
507 
508 	if (vpislocked == 0) {
509 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
510 			error = EPERM;
511 			goto out;
512 		}
513 	}
514 	if (accmode & VWRITE) {
515 		/* Just vn_writechk() changed to check rdonly */
516 		/*
517 		 * Disallow write attempts on read-only file systems;
518 		 * unless the file is a socket or a block or character
519 		 * device resident on the file system.
520 		 */
521 		if (NFSVNO_EXRDONLY(exp) ||
522 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
523 			switch (vp->v_type) {
524 			case VREG:
525 			case VDIR:
526 			case VLNK:
527 				error = EROFS;
528 			default:
529 				break;
530 			}
531 		}
532 		/*
533 		 * If there's shared text associated with
534 		 * the inode, try to free it up once.  If
535 		 * we fail, we can't allow writing.
536 		 */
537 		if (VOP_IS_TEXT(vp) && error == 0)
538 			error = ETXTBSY;
539 	}
540 	if (error != 0) {
541 		if (vpislocked == 0)
542 			NFSVOPUNLOCK(vp);
543 		goto out;
544 	}
545 
546 	/*
547 	 * Should the override still be applied when ACLs are enabled?
548 	 */
549 	error = VOP_ACCESSX(vp, accmode, cred, p);
550 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
551 		/*
552 		 * Try again with VEXPLICIT_DENY, to see if the test for
553 		 * deletion is supported.
554 		 */
555 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
556 		if (error == 0) {
557 			if (vp->v_type == VDIR) {
558 				accmode &= ~(VDELETE | VDELETE_CHILD);
559 				accmode |= VWRITE;
560 				error = VOP_ACCESSX(vp, accmode, cred, p);
561 			} else if (supportedtypep != NULL) {
562 				*supportedtypep &= ~NFSACCESS_DELETE;
563 			}
564 		}
565 	}
566 
567 	/*
568 	 * Allow certain operations for the owner (reads and writes
569 	 * on files that are already open).
570 	 */
571 	if (override != NFSACCCHK_NOOVERRIDE &&
572 	    (error == EPERM || error == EACCES)) {
573 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
574 			error = 0;
575 		else if (override & NFSACCCHK_ALLOWOWNER) {
576 			getret = VOP_GETATTR(vp, &vattr, cred);
577 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
578 				error = 0;
579 		}
580 	}
581 	if (vpislocked == 0)
582 		NFSVOPUNLOCK(vp);
583 
584 out:
585 	NFSEXITCODE(error);
586 	return (error);
587 }
588 
589 /*
590  * Set attribute(s) vnop.
591  */
592 int
nfsvno_setattr(struct vnode * vp,struct nfsvattr * nvap,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)593 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
594     struct thread *p, struct nfsexstuff *exp)
595 {
596 	u_quad_t savsize = 0;
597 	int error, savedit;
598 	time_t savbtime;
599 
600 	/*
601 	 * If this is an exported file system and a pNFS service is running,
602 	 * don't VOP_SETATTR() of size for the MDS file system.
603 	 */
604 	savedit = 0;
605 	error = 0;
606 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
607 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
608 	    nvap->na_vattr.va_size > 0) {
609 		savsize = nvap->na_vattr.va_size;
610 		nvap->na_vattr.va_size = VNOVAL;
611 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
612 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
613 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
614 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
615 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
616 			savedit = 1;
617 		else
618 			savedit = 2;
619 	}
620 	if (savedit != 2)
621 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
622 	if (savedit != 0)
623 		nvap->na_vattr.va_size = savsize;
624 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
625 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
626 	    nvap->na_vattr.va_size != VNOVAL ||
627 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
628 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
629 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
630 		/* Never modify birthtime on a DS file. */
631 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
632 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
633 		/* For a pNFS server, set the attributes on the DS file. */
634 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
635 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
636 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
637 		if (error == ENOENT)
638 			error = 0;
639 	}
640 	NFSEXITCODE(error);
641 	return (error);
642 }
643 
644 /*
645  * Set up nameidata for a lookup() call and do it.
646  */
647 int
nfsvno_namei(struct nfsrv_descript * nd,struct nameidata * ndp,struct vnode * dp,int islocked,struct nfsexstuff * exp,struct vnode ** retdirp)648 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
649     struct vnode *dp, int islocked, struct nfsexstuff *exp,
650     struct vnode **retdirp)
651 {
652 	struct componentname *cnp = &ndp->ni_cnd;
653 	int i;
654 	struct iovec aiov;
655 	struct uio auio;
656 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
657 	int error = 0;
658 	char *cp;
659 
660 	*retdirp = NULL;
661 	cnp->cn_nameptr = cnp->cn_pnbuf;
662 	ndp->ni_lcf = 0;
663 	/*
664 	 * Extract and set starting directory.
665 	 */
666 	if (dp->v_type != VDIR) {
667 		if (islocked)
668 			vput(dp);
669 		else
670 			vrele(dp);
671 		nfsvno_relpathbuf(ndp);
672 		error = ENOTDIR;
673 		goto out1;
674 	}
675 	if (islocked)
676 		NFSVOPUNLOCK(dp);
677 	VREF(dp);
678 	*retdirp = dp;
679 	if (NFSVNO_EXRDONLY(exp))
680 		cnp->cn_flags |= RDONLY;
681 	ndp->ni_segflg = UIO_SYSSPACE;
682 
683 	if (nd->nd_flag & ND_PUBLOOKUP) {
684 		ndp->ni_loopcnt = 0;
685 		if (cnp->cn_pnbuf[0] == '/') {
686 			vrele(dp);
687 			/*
688 			 * Check for degenerate pathnames here, since lookup()
689 			 * panics on them.
690 			 */
691 			for (i = 1; i < ndp->ni_pathlen; i++)
692 				if (cnp->cn_pnbuf[i] != '/')
693 					break;
694 			if (i == ndp->ni_pathlen) {
695 				error = NFSERR_ACCES;
696 				goto out;
697 			}
698 			dp = rootvnode;
699 			VREF(dp);
700 		}
701 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
702 	    (nd->nd_flag & ND_NFSV4) == 0) {
703 		/*
704 		 * Only cross mount points for NFSv4 when doing a
705 		 * mount while traversing the file system above
706 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
707 		 */
708 		cnp->cn_flags |= NOCROSSMOUNT;
709 	}
710 
711 	/*
712 	 * Initialize for scan, set ni_startdir and bump ref on dp again
713 	 * because lookup() will dereference ni_startdir.
714 	 */
715 
716 	ndp->ni_startdir = dp;
717 	ndp->ni_rootdir = rootvnode;
718 	ndp->ni_topdir = NULL;
719 
720 	if (!lockleaf)
721 		cnp->cn_flags |= LOCKLEAF;
722 	for (;;) {
723 		cnp->cn_nameptr = cnp->cn_pnbuf;
724 		/*
725 		 * Call lookup() to do the real work.  If an error occurs,
726 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
727 		 * we do not have to dereference anything before returning.
728 		 * In either case ni_startdir will be dereferenced and NULLed
729 		 * out.
730 		 */
731 		error = vfs_lookup(ndp);
732 		if (error)
733 			break;
734 
735 		/*
736 		 * Check for encountering a symbolic link.  Trivial
737 		 * termination occurs if no symlink encountered.
738 		 */
739 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
740 			if (ndp->ni_vp && !lockleaf)
741 				NFSVOPUNLOCK(ndp->ni_vp);
742 			break;
743 		}
744 
745 		/*
746 		 * Validate symlink
747 		 */
748 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
749 			NFSVOPUNLOCK(ndp->ni_dvp);
750 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
751 			error = EINVAL;
752 			goto badlink2;
753 		}
754 
755 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
756 			error = ELOOP;
757 			goto badlink2;
758 		}
759 		if (ndp->ni_pathlen > 1)
760 			cp = uma_zalloc(namei_zone, M_WAITOK);
761 		else
762 			cp = cnp->cn_pnbuf;
763 		aiov.iov_base = cp;
764 		aiov.iov_len = MAXPATHLEN;
765 		auio.uio_iov = &aiov;
766 		auio.uio_iovcnt = 1;
767 		auio.uio_offset = 0;
768 		auio.uio_rw = UIO_READ;
769 		auio.uio_segflg = UIO_SYSSPACE;
770 		auio.uio_td = NULL;
771 		auio.uio_resid = MAXPATHLEN;
772 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
773 		if (error) {
774 		badlink1:
775 			if (ndp->ni_pathlen > 1)
776 				uma_zfree(namei_zone, cp);
777 		badlink2:
778 			vrele(ndp->ni_dvp);
779 			vput(ndp->ni_vp);
780 			break;
781 		}
782 		linklen = MAXPATHLEN - auio.uio_resid;
783 		if (linklen == 0) {
784 			error = ENOENT;
785 			goto badlink1;
786 		}
787 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
788 			error = ENAMETOOLONG;
789 			goto badlink1;
790 		}
791 
792 		/*
793 		 * Adjust or replace path
794 		 */
795 		if (ndp->ni_pathlen > 1) {
796 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
797 			uma_zfree(namei_zone, cnp->cn_pnbuf);
798 			cnp->cn_pnbuf = cp;
799 		} else
800 			cnp->cn_pnbuf[linklen] = '\0';
801 		ndp->ni_pathlen += linklen;
802 
803 		/*
804 		 * Cleanup refs for next loop and check if root directory
805 		 * should replace current directory.  Normally ni_dvp
806 		 * becomes the new base directory and is cleaned up when
807 		 * we loop.  Explicitly null pointers after invalidation
808 		 * to clarify operation.
809 		 */
810 		vput(ndp->ni_vp);
811 		ndp->ni_vp = NULL;
812 
813 		if (cnp->cn_pnbuf[0] == '/') {
814 			vrele(ndp->ni_dvp);
815 			ndp->ni_dvp = ndp->ni_rootdir;
816 			VREF(ndp->ni_dvp);
817 		}
818 		ndp->ni_startdir = ndp->ni_dvp;
819 		ndp->ni_dvp = NULL;
820 	}
821 	if (!lockleaf)
822 		cnp->cn_flags &= ~LOCKLEAF;
823 
824 out:
825 	if (error) {
826 		nfsvno_relpathbuf(ndp);
827 		ndp->ni_vp = NULL;
828 		ndp->ni_dvp = NULL;
829 		ndp->ni_startdir = NULL;
830 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
831 		ndp->ni_dvp = NULL;
832 	}
833 
834 out1:
835 	NFSEXITCODE2(error, nd);
836 	return (error);
837 }
838 
839 /*
840  * Set up a pathname buffer and return a pointer to it and, optionally
841  * set a hash pointer.
842  */
843 void
nfsvno_setpathbuf(struct nameidata * ndp,char ** bufpp,u_long ** hashpp)844 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
845 {
846 	struct componentname *cnp = &ndp->ni_cnd;
847 
848 	cnp->cn_flags |= (NOMACCHECK);
849 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
850 	if (hashpp != NULL)
851 		*hashpp = NULL;
852 	*bufpp = cnp->cn_pnbuf;
853 }
854 
855 /*
856  * Release the above path buffer, if not released by nfsvno_namei().
857  */
858 void
nfsvno_relpathbuf(struct nameidata * ndp)859 nfsvno_relpathbuf(struct nameidata *ndp)
860 {
861 
862 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
863 	ndp->ni_cnd.cn_pnbuf = NULL;
864 }
865 
866 /*
867  * Readlink vnode op into an mbuf list.
868  */
869 int
nfsvno_readlink(struct vnode * vp,struct ucred * cred,int maxextsiz,struct thread * p,struct mbuf ** mpp,struct mbuf ** mpendp,int * lenp)870 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
871     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
872 {
873 	struct iovec *iv;
874 	struct uio io, *uiop = &io;
875 	struct mbuf *mp, *mp3;
876 	int len, tlen, error = 0;
877 
878 	len = NFS_MAXPATHLEN;
879 	if (maxextsiz > 0)
880 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
881 		    &mp3, &mp, &iv);
882 	else
883 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
884 	uiop->uio_iov = iv;
885 	uiop->uio_offset = 0;
886 	uiop->uio_resid = len;
887 	uiop->uio_rw = UIO_READ;
888 	uiop->uio_segflg = UIO_SYSSPACE;
889 	uiop->uio_td = NULL;
890 	error = VOP_READLINK(vp, uiop, cred);
891 	free(iv, M_TEMP);
892 	if (error) {
893 		m_freem(mp3);
894 		*lenp = 0;
895 		goto out;
896 	}
897 	if (uiop->uio_resid > 0) {
898 		len -= uiop->uio_resid;
899 		tlen = NFSM_RNDUP(len);
900 		if (tlen == 0) {
901 			m_freem(mp3);
902 			mp3 = mp = NULL;
903 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
904 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
905 			    tlen - len);
906 	}
907 	*lenp = len;
908 	*mpp = mp3;
909 	*mpendp = mp;
910 
911 out:
912 	NFSEXITCODE(error);
913 	return (error);
914 }
915 
916 /*
917  * Create an mbuf chain and an associated iovec that can be used to Read
918  * or Getextattr of data.
919  * Upon success, return pointers to the first and last mbufs in the chain
920  * plus the malloc'd iovec and its iovlen.
921  */
922 static int
nfsrv_createiovec(int len,struct mbuf ** mpp,struct mbuf ** mpendp,struct iovec ** ivp)923 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
924     struct iovec **ivp)
925 {
926 	struct mbuf *m, *m2 = NULL, *m3;
927 	struct iovec *iv;
928 	int i, left, siz;
929 
930 	left = len;
931 	m3 = NULL;
932 	/*
933 	 * Generate the mbuf list with the uio_iov ref. to it.
934 	 */
935 	i = 0;
936 	while (left > 0) {
937 		NFSMGET(m);
938 		MCLGET(m, M_WAITOK);
939 		m->m_len = 0;
940 		siz = min(M_TRAILINGSPACE(m), left);
941 		left -= siz;
942 		i++;
943 		if (m3)
944 			m2->m_next = m;
945 		else
946 			m3 = m;
947 		m2 = m;
948 	}
949 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
950 	m = m3;
951 	left = len;
952 	i = 0;
953 	while (left > 0) {
954 		if (m == NULL)
955 			panic("nfsrv_createiovec iov");
956 		siz = min(M_TRAILINGSPACE(m), left);
957 		if (siz > 0) {
958 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
959 			iv->iov_len = siz;
960 			m->m_len += siz;
961 			left -= siz;
962 			iv++;
963 			i++;
964 		}
965 		m = m->m_next;
966 	}
967 	*mpp = m3;
968 	*mpendp = m2;
969 	return (i);
970 }
971 
972 /*
973  * Create an mbuf chain and an associated iovec that can be used to Read
974  * or Getextattr of data.
975  * Upon success, return pointers to the first and last mbufs in the chain
976  * plus the malloc'd iovec and its iovlen.
977  * Same as above, but creates ext_pgs mbuf(s).
978  */
979 static int
nfsrv_createiovec_extpgs(int len,int maxextsiz,struct mbuf ** mpp,struct mbuf ** mpendp,struct iovec ** ivp)980 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
981     struct mbuf **mpendp, struct iovec **ivp)
982 {
983 	struct mbuf *m, *m2 = NULL, *m3;
984 	struct iovec *iv;
985 	int i, left, pgno, siz;
986 
987 	left = len;
988 	m3 = NULL;
989 	/*
990 	 * Generate the mbuf list with the uio_iov ref. to it.
991 	 */
992 	i = 0;
993 	while (left > 0) {
994 		siz = min(left, maxextsiz);
995 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
996 		left -= siz;
997 		i += m->m_epg_npgs;
998 		if (m3 != NULL)
999 			m2->m_next = m;
1000 		else
1001 			m3 = m;
1002 		m2 = m;
1003 	}
1004 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1005 	m = m3;
1006 	left = len;
1007 	i = 0;
1008 	pgno = 0;
1009 	while (left > 0) {
1010 		if (m == NULL)
1011 			panic("nfsvno_createiovec_extpgs iov");
1012 		siz = min(PAGE_SIZE, left);
1013 		if (siz > 0) {
1014 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
1015 			iv->iov_len = siz;
1016 			m->m_len += siz;
1017 			if (pgno == m->m_epg_npgs - 1)
1018 				m->m_epg_last_len = siz;
1019 			left -= siz;
1020 			iv++;
1021 			i++;
1022 			pgno++;
1023 		}
1024 		if (pgno == m->m_epg_npgs && left > 0) {
1025 			m = m->m_next;
1026 			if (m == NULL)
1027 				panic("nfsvno_createiovec_extpgs iov");
1028 			pgno = 0;
1029 		}
1030 	}
1031 	*mpp = m3;
1032 	*mpendp = m2;
1033 	return (i);
1034 }
1035 
1036 /*
1037  * Read vnode op call into mbuf list.
1038  */
1039 int
nfsvno_read(struct vnode * vp,off_t off,int cnt,struct ucred * cred,int maxextsiz,struct thread * p,struct mbuf ** mpp,struct mbuf ** mpendp)1040 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
1041     int maxextsiz, struct thread *p, struct mbuf **mpp,
1042     struct mbuf **mpendp)
1043 {
1044 	struct mbuf *m;
1045 	struct iovec *iv;
1046 	int error = 0, len, tlen, ioflag = 0;
1047 	struct mbuf *m3;
1048 	struct uio io, *uiop = &io;
1049 	struct nfsheur *nh;
1050 
1051 	/*
1052 	 * Attempt to read from a DS file. A return of ENOENT implies
1053 	 * there is no DS file to read.
1054 	 */
1055 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1056 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1057 	if (error != ENOENT)
1058 		return (error);
1059 
1060 	len = NFSM_RNDUP(cnt);
1061 	if (maxextsiz > 0)
1062 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1063 		    &m3, &m, &iv);
1064 	else
1065 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1066 	uiop->uio_iov = iv;
1067 	uiop->uio_offset = off;
1068 	uiop->uio_resid = len;
1069 	uiop->uio_rw = UIO_READ;
1070 	uiop->uio_segflg = UIO_SYSSPACE;
1071 	uiop->uio_td = NULL;
1072 	nh = nfsrv_sequential_heuristic(uiop, vp);
1073 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1074 	/* XXX KDM make this more systematic? */
1075 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1076 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1077 	free(iv, M_TEMP);
1078 	if (error) {
1079 		m_freem(m3);
1080 		*mpp = NULL;
1081 		goto out;
1082 	}
1083 	nh->nh_nextoff = uiop->uio_offset;
1084 	tlen = len - uiop->uio_resid;
1085 	cnt = cnt < tlen ? cnt : tlen;
1086 	tlen = NFSM_RNDUP(cnt);
1087 	if (tlen == 0) {
1088 		m_freem(m3);
1089 		m3 = m = NULL;
1090 	} else if (len != tlen || tlen != cnt)
1091 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1092 	*mpp = m3;
1093 	*mpendp = m;
1094 
1095 out:
1096 	NFSEXITCODE(error);
1097 	return (error);
1098 }
1099 
1100 /*
1101  * Create the iovec for the mbuf chain passed in as an argument.
1102  * The "cp" argument is where the data starts within the first mbuf in
1103  * the chain. It returns the iovec and the iovcnt.
1104  */
1105 static int
nfsrv_createiovecw(int retlen,struct mbuf * m,char * cp,struct iovec ** ivpp,int * iovcntp)1106 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1107     int *iovcntp)
1108 {
1109 	struct mbuf *mp;
1110 	struct iovec *ivp;
1111 	int cnt, i, len;
1112 
1113 	/*
1114 	 * Loop through the mbuf chain, counting how many mbufs are a
1115 	 * part of this write operation, so the iovec size is known.
1116 	 */
1117 	cnt = 0;
1118 	len = retlen;
1119 	mp = m;
1120 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1121 	while (len > 0) {
1122 		if (i > 0) {
1123 			len -= i;
1124 			cnt++;
1125 		}
1126 		mp = mp->m_next;
1127 		if (!mp) {
1128 			if (len > 0)
1129 				return (EBADRPC);
1130 		} else
1131 			i = mp->m_len;
1132 	}
1133 
1134 	/* Now, create the iovec. */
1135 	mp = m;
1136 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1137 	    M_WAITOK);
1138 	*iovcntp = cnt;
1139 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1140 	len = retlen;
1141 	while (len > 0) {
1142 		if (mp == NULL)
1143 			panic("nfsrv_createiovecw");
1144 		if (i > 0) {
1145 			i = min(i, len);
1146 			ivp->iov_base = cp;
1147 			ivp->iov_len = i;
1148 			ivp++;
1149 			len -= i;
1150 		}
1151 		mp = mp->m_next;
1152 		if (mp) {
1153 			i = mp->m_len;
1154 			cp = mtod(mp, caddr_t);
1155 		}
1156 	}
1157 	return (0);
1158 }
1159 
1160 /*
1161  * Write vnode op from an mbuf list.
1162  */
1163 int
nfsvno_write(struct vnode * vp,off_t off,int retlen,int * stable,struct mbuf * mp,char * cp,struct ucred * cred,struct thread * p)1164 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1165     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1166 {
1167 	struct iovec *iv;
1168 	int cnt, ioflags, error;
1169 	struct uio io, *uiop = &io;
1170 	struct nfsheur *nh;
1171 
1172 	/*
1173 	 * Attempt to write to a DS file. A return of ENOENT implies
1174 	 * there is no DS file to write.
1175 	 */
1176 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1177 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1178 	if (error != ENOENT) {
1179 		*stable = NFSWRITE_FILESYNC;
1180 		return (error);
1181 	}
1182 
1183 	if (*stable == NFSWRITE_UNSTABLE)
1184 		ioflags = IO_NODELOCKED;
1185 	else
1186 		ioflags = (IO_SYNC | IO_NODELOCKED);
1187 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1188 	if (error != 0)
1189 		return (error);
1190 	uiop->uio_iov = iv;
1191 	uiop->uio_iovcnt = cnt;
1192 	uiop->uio_resid = retlen;
1193 	uiop->uio_rw = UIO_WRITE;
1194 	uiop->uio_segflg = UIO_SYSSPACE;
1195 	NFSUIOPROC(uiop, p);
1196 	uiop->uio_offset = off;
1197 	nh = nfsrv_sequential_heuristic(uiop, vp);
1198 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1199 	/* XXX KDM make this more systematic? */
1200 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1201 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1202 	if (error == 0)
1203 		nh->nh_nextoff = uiop->uio_offset;
1204 	free(iv, M_TEMP);
1205 
1206 	NFSEXITCODE(error);
1207 	return (error);
1208 }
1209 
1210 /*
1211  * Common code for creating a regular file (plus special files for V2).
1212  */
1213 int
nfsvno_createsub(struct nfsrv_descript * nd,struct nameidata * ndp,struct vnode ** vpp,struct nfsvattr * nvap,int * exclusive_flagp,int32_t * cverf,NFSDEV_T rdev,struct nfsexstuff * exp)1214 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1215     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1216     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1217 {
1218 	u_quad_t tempsize;
1219 	int error;
1220 	struct thread *p = curthread;
1221 
1222 	error = nd->nd_repstat;
1223 	if (!error && ndp->ni_vp == NULL) {
1224 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1225 			error = VOP_CREATE(ndp->ni_dvp,
1226 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1227 			/* For a pNFS server, create the data file on a DS. */
1228 			if (error == 0 && nvap->na_type == VREG) {
1229 				/*
1230 				 * Create a data file on a DS for a pNFS server.
1231 				 * This function just returns if not
1232 				 * running a pNFS DS or the creation fails.
1233 				 */
1234 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1235 				    nd->nd_cred, p);
1236 			}
1237 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1238 			    NULL, false);
1239 			nfsvno_relpathbuf(ndp);
1240 			if (!error) {
1241 				if (*exclusive_flagp) {
1242 					*exclusive_flagp = 0;
1243 					NFSVNO_ATTRINIT(nvap);
1244 					nvap->na_atime.tv_sec = cverf[0];
1245 					nvap->na_atime.tv_nsec = cverf[1];
1246 					error = VOP_SETATTR(ndp->ni_vp,
1247 					    &nvap->na_vattr, nd->nd_cred);
1248 					if (error != 0) {
1249 						vput(ndp->ni_vp);
1250 						ndp->ni_vp = NULL;
1251 						error = NFSERR_NOTSUPP;
1252 					}
1253 				}
1254 			}
1255 		/*
1256 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1257 		 * (This implies, just get out on an error.)
1258 		 */
1259 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1260 			nvap->na_type == VFIFO) {
1261 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1262 				nvap->na_type = VFIFO;
1263                         if (nvap->na_type != VFIFO &&
1264 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1265 				nfsvno_relpathbuf(ndp);
1266 				vput(ndp->ni_dvp);
1267 				goto out;
1268 			}
1269 			nvap->na_rdev = rdev;
1270 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1271 			    &ndp->ni_cnd, &nvap->na_vattr);
1272 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1273 			    NULL, false);
1274 			nfsvno_relpathbuf(ndp);
1275 			if (error)
1276 				goto out;
1277 		} else {
1278 			nfsvno_relpathbuf(ndp);
1279 			vput(ndp->ni_dvp);
1280 			error = ENXIO;
1281 			goto out;
1282 		}
1283 		*vpp = ndp->ni_vp;
1284 	} else {
1285 		/*
1286 		 * Handle cases where error is already set and/or
1287 		 * the file exists.
1288 		 * 1 - clean up the lookup
1289 		 * 2 - iff !error and na_size set, truncate it
1290 		 */
1291 		nfsvno_relpathbuf(ndp);
1292 		*vpp = ndp->ni_vp;
1293 		if (ndp->ni_dvp == *vpp)
1294 			vrele(ndp->ni_dvp);
1295 		else
1296 			vput(ndp->ni_dvp);
1297 		if (!error && nvap->na_size != VNOVAL) {
1298 			error = nfsvno_accchk(*vpp, VWRITE,
1299 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1300 			    NFSACCCHK_VPISLOCKED, NULL);
1301 			if (!error) {
1302 				tempsize = nvap->na_size;
1303 				NFSVNO_ATTRINIT(nvap);
1304 				nvap->na_size = tempsize;
1305 				error = nfsvno_setattr(*vpp, nvap,
1306 				    nd->nd_cred, p, exp);
1307 			}
1308 		}
1309 		if (error)
1310 			vput(*vpp);
1311 	}
1312 
1313 out:
1314 	NFSEXITCODE(error);
1315 	return (error);
1316 }
1317 
1318 /*
1319  * Do a mknod vnode op.
1320  */
1321 int
nfsvno_mknod(struct nameidata * ndp,struct nfsvattr * nvap,struct ucred * cred,struct thread * p)1322 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1323     struct thread *p)
1324 {
1325 	int error = 0;
1326 	__enum_uint8(vtype) vtyp;
1327 
1328 	vtyp = nvap->na_type;
1329 	/*
1330 	 * Iff doesn't exist, create it.
1331 	 */
1332 	if (ndp->ni_vp) {
1333 		nfsvno_relpathbuf(ndp);
1334 		vput(ndp->ni_dvp);
1335 		vrele(ndp->ni_vp);
1336 		error = EEXIST;
1337 		goto out;
1338 	}
1339 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1340 		nfsvno_relpathbuf(ndp);
1341 		vput(ndp->ni_dvp);
1342 		error = NFSERR_BADTYPE;
1343 		goto out;
1344 	}
1345 	if (vtyp == VSOCK) {
1346 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1347 		    &ndp->ni_cnd, &nvap->na_vattr);
1348 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1349 		    false);
1350 		nfsvno_relpathbuf(ndp);
1351 	} else {
1352 		if (nvap->na_type != VFIFO &&
1353 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1354 			nfsvno_relpathbuf(ndp);
1355 			vput(ndp->ni_dvp);
1356 			goto out;
1357 		}
1358 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1359 		    &ndp->ni_cnd, &nvap->na_vattr);
1360 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1361 		    false);
1362 		nfsvno_relpathbuf(ndp);
1363 		/*
1364 		 * Since VOP_MKNOD returns the ni_vp, I can't
1365 		 * see any reason to do the lookup.
1366 		 */
1367 	}
1368 
1369 out:
1370 	NFSEXITCODE(error);
1371 	return (error);
1372 }
1373 
1374 /*
1375  * Mkdir vnode op.
1376  */
1377 int
nfsvno_mkdir(struct nameidata * ndp,struct nfsvattr * nvap,uid_t saved_uid,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1378 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1379     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1380 {
1381 	int error = 0;
1382 
1383 	if (ndp->ni_vp != NULL) {
1384 		if (ndp->ni_dvp == ndp->ni_vp)
1385 			vrele(ndp->ni_dvp);
1386 		else
1387 			vput(ndp->ni_dvp);
1388 		vrele(ndp->ni_vp);
1389 		nfsvno_relpathbuf(ndp);
1390 		error = EEXIST;
1391 		goto out;
1392 	}
1393 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1394 	    &nvap->na_vattr);
1395 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1396 	nfsvno_relpathbuf(ndp);
1397 
1398 out:
1399 	NFSEXITCODE(error);
1400 	return (error);
1401 }
1402 
1403 /*
1404  * symlink vnode op.
1405  */
1406 int
nfsvno_symlink(struct nameidata * ndp,struct nfsvattr * nvap,char * pathcp,int pathlen,int not_v2,uid_t saved_uid,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1407 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1408     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1409     struct nfsexstuff *exp)
1410 {
1411 	int error = 0;
1412 
1413 	if (ndp->ni_vp) {
1414 		nfsvno_relpathbuf(ndp);
1415 		if (ndp->ni_dvp == ndp->ni_vp)
1416 			vrele(ndp->ni_dvp);
1417 		else
1418 			vput(ndp->ni_dvp);
1419 		vrele(ndp->ni_vp);
1420 		error = EEXIST;
1421 		goto out;
1422 	}
1423 
1424 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1425 	    &nvap->na_vattr, pathcp);
1426 	/*
1427 	 * Although FreeBSD still had the lookup code in
1428 	 * it for 7/current, there doesn't seem to be any
1429 	 * point, since VOP_SYMLINK() returns the ni_vp.
1430 	 * Just vput it for v2.
1431 	 */
1432 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1433 	nfsvno_relpathbuf(ndp);
1434 
1435 out:
1436 	NFSEXITCODE(error);
1437 	return (error);
1438 }
1439 
1440 /*
1441  * Parse symbolic link arguments.
1442  * This function has an ugly side effect. It will malloc() an area for
1443  * the symlink and set iov_base to point to it, only if it succeeds.
1444  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1445  * be FREE'd later.
1446  */
1447 int
nfsvno_getsymlink(struct nfsrv_descript * nd,struct nfsvattr * nvap,struct thread * p,char ** pathcpp,int * lenp)1448 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1449     struct thread *p, char **pathcpp, int *lenp)
1450 {
1451 	u_int32_t *tl;
1452 	char *pathcp = NULL;
1453 	int error = 0, len;
1454 	struct nfsv2_sattr *sp;
1455 
1456 	*pathcpp = NULL;
1457 	*lenp = 0;
1458 	if ((nd->nd_flag & ND_NFSV3) &&
1459 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1460 		goto nfsmout;
1461 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1462 	len = fxdr_unsigned(int, *tl);
1463 	if (len > NFS_MAXPATHLEN || len <= 0) {
1464 		error = EBADRPC;
1465 		goto nfsmout;
1466 	}
1467 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1468 	error = nfsrv_mtostr(nd, pathcp, len);
1469 	if (error)
1470 		goto nfsmout;
1471 	if (nd->nd_flag & ND_NFSV2) {
1472 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1473 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1474 	}
1475 	*pathcpp = pathcp;
1476 	*lenp = len;
1477 	NFSEXITCODE2(0, nd);
1478 	return (0);
1479 nfsmout:
1480 	if (pathcp)
1481 		free(pathcp, M_TEMP);
1482 	NFSEXITCODE2(error, nd);
1483 	return (error);
1484 }
1485 
1486 /*
1487  * Remove a non-directory object.
1488  */
1489 int
nfsvno_removesub(struct nameidata * ndp,bool is_v4,struct nfsrv_descript * nd,struct thread * p,struct nfsexstuff * exp)1490 nfsvno_removesub(struct nameidata *ndp, bool is_v4, struct nfsrv_descript *nd,
1491     struct thread *p, struct nfsexstuff *exp)
1492 {
1493 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS], *newvp;
1494 	struct mount *mp;
1495 	int error = 0, mirrorcnt, ret;
1496 	char fname[PNFS_FILENAME_LEN + 1];
1497 	fhandle_t fh;
1498 
1499 	vp = ndp->ni_vp;
1500 	dsdvp[0] = NULL;
1501 	if (vp->v_type == VDIR) {
1502 		error = NFSERR_ISDIR;
1503 	} else if (is_v4) {
1504 		if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0)
1505 			error = nfsrv_checkremove(vp, 1, NULL,
1506 			    (nfsquad_t)((u_quad_t)0), p);
1507 		else
1508 			error = nfsrv_checkremove(vp, 1, NULL, nd->nd_clientid,
1509 			    p);
1510 	}
1511 	if (error == 0)
1512 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1513 	if (!error)
1514 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1515 	if (error == 0 && dsdvp[0] != NULL)
1516 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1517 	if (is_v4 && (nd->nd_flag & ND_NFSV41) != 0 && error == 0)
1518 		error = nfsvno_getfh(vp, &fh, p);
1519 	if (ndp->ni_dvp == vp)
1520 		vrele(ndp->ni_dvp);
1521 	else
1522 		vput(ndp->ni_dvp);
1523 	vput(vp);
1524 
1525 	/* Use ret to determine if the file still exists. */
1526 	if (is_v4 && (nd->nd_flag & ND_NFSV41) != 0 && error == 0) {
1527 		mp = vfs_busyfs(&fh.fh_fsid);
1528 		if (mp != NULL) {
1529 			/* Find out if the file still exists. */
1530 			ret = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &newvp);
1531 			if (ret == 0)
1532 				vput(newvp);
1533 			else
1534 				ret = ESTALE;
1535 			vfs_unbusy(mp);
1536 		} else {
1537 			ret = ESTALE;
1538 		}
1539 		if (ret == ESTALE) {
1540 			/* Get rid of any delegation. */
1541 			nfsrv_removedeleg(&fh, nd, p);
1542 		}
1543 	}
1544 
1545 	nfsvno_relpathbuf(ndp);
1546 	NFSEXITCODE(error);
1547 	return (error);
1548 }
1549 
1550 /*
1551  * Remove a directory.
1552  */
1553 int
nfsvno_rmdirsub(struct nameidata * ndp,int is_v4,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1554 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1555     struct thread *p, struct nfsexstuff *exp)
1556 {
1557 	struct vnode *vp;
1558 	int error = 0;
1559 
1560 	vp = ndp->ni_vp;
1561 	if (vp->v_type != VDIR) {
1562 		error = ENOTDIR;
1563 		goto out;
1564 	}
1565 	/*
1566 	 * No rmdir "." please.
1567 	 */
1568 	if (ndp->ni_dvp == vp) {
1569 		error = EINVAL;
1570 		goto out;
1571 	}
1572 	/*
1573 	 * The root of a mounted filesystem cannot be deleted.
1574 	 */
1575 	if (vp->v_vflag & VV_ROOT)
1576 		error = EBUSY;
1577 out:
1578 	if (!error)
1579 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1580 	if (ndp->ni_dvp == vp)
1581 		vrele(ndp->ni_dvp);
1582 	else
1583 		vput(ndp->ni_dvp);
1584 	vput(vp);
1585 	nfsvno_relpathbuf(ndp);
1586 	NFSEXITCODE(error);
1587 	return (error);
1588 }
1589 
1590 /*
1591  * Rename vnode op.
1592  */
1593 int
nfsvno_rename(struct nameidata * fromndp,struct nameidata * tondp,struct nfsrv_descript * nd,struct thread * p)1594 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1595     struct nfsrv_descript *nd, struct thread *p)
1596 {
1597 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS], *newvp;
1598 	struct mount *mp;
1599 	int error = 0, mirrorcnt, ret;
1600 	char fname[PNFS_FILENAME_LEN + 1];
1601 	fhandle_t fh, fh2;
1602 
1603 	dsdvp[0] = NULL;
1604 	fvp = fromndp->ni_vp;
1605 	if (nd->nd_repstat != 0) {
1606 		vrele(fromndp->ni_dvp);
1607 		vrele(fvp);
1608 		error = nd->nd_repstat;
1609 		goto out1;
1610 	}
1611 	tdvp = tondp->ni_dvp;
1612 	tvp = tondp->ni_vp;
1613 	if (tvp != NULL) {
1614 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1615 			error = (nd->nd_flag & ND_NFSV2) ? EISDIR : EEXIST;
1616 			goto out;
1617 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1618 			error = (nd->nd_flag & ND_NFSV2) ? ENOTDIR : EEXIST;
1619 			goto out;
1620 		}
1621 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1622 			error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1623 			goto out;
1624 		}
1625 
1626 		/*
1627 		 * A rename to '.' or '..' results in a prematurely
1628 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1629 		 * here.
1630 		 */
1631 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1632 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1633 		    (tondp->ni_cnd.cn_namelen == 2 &&
1634 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1635 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1636 			error = EINVAL;
1637 			goto out;
1638 		}
1639 	}
1640 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1641 		error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1642 		goto out;
1643 	}
1644 	if (fvp->v_mount != tdvp->v_mount) {
1645 		error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1646 		goto out;
1647 	}
1648 	if (fvp == tdvp) {
1649 		error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1650 		goto out;
1651 	}
1652 	if (fvp == tvp) {
1653 		/*
1654 		 * If source and destination are the same, there is nothing to
1655 		 * do. Set error to -1 to indicate this.
1656 		 */
1657 		error = -1;
1658 		goto out;
1659 	}
1660 	if (nd->nd_flag & ND_NFSV4) {
1661 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1662 			if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0)
1663 				error = nfsrv_checkremove(fvp, 0, NULL,
1664 				    (nfsquad_t)((u_quad_t)0), p);
1665 			else
1666 				error = nfsrv_checkremove(fvp, 0, NULL,
1667 				    nd->nd_clientid, p);
1668 			NFSVOPUNLOCK(fvp);
1669 		} else
1670 			error = EPERM;
1671 		if (tvp && !error) {
1672 			if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0)
1673 				error = nfsrv_checkremove(tvp, 1, NULL,
1674 				    (nfsquad_t)((u_quad_t)0), p);
1675 			else
1676 				error = nfsrv_checkremove(tvp, 1, NULL,
1677 				    nd->nd_clientid, p);
1678 		}
1679 	} else {
1680 		/*
1681 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1682 		 * that the NFSv4 client won't be confused by the rename.
1683 		 * Since nfsd_recalldelegation() can only be called on an
1684 		 * unlocked vnode at this point and fvp is the file that will
1685 		 * still exist after the rename, just do fvp.
1686 		 */
1687 		nfsd_recalldelegation(fvp, p);
1688 	}
1689 	if (error == 0 && tvp != NULL) {
1690 		if ((nd->nd_flag & ND_NFSV41) != 0)
1691 			error = nfsvno_getfh(tvp, &fh2, p);
1692 		if (error == 0)
1693 			nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname,
1694 			    &fh);
1695 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1696 		    " dsdvp=%p\n", dsdvp[0]);
1697 	}
1698 out:
1699 	if (!error) {
1700 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1701 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1702 		    &tondp->ni_cnd);
1703 	} else {
1704 		if (tdvp == tvp)
1705 			vrele(tdvp);
1706 		else
1707 			vput(tdvp);
1708 		if (tvp)
1709 			vput(tvp);
1710 		vrele(fromndp->ni_dvp);
1711 		vrele(fvp);
1712 		if (error == -1)
1713 			error = 0;
1714 	}
1715 
1716 	/*
1717 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1718 	 * if the rename succeeded, the DS file for the tvp needs to be
1719 	 * removed.
1720 	 */
1721 	if (error == 0 && dsdvp[0] != NULL) {
1722 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1723 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1724 	}
1725 
1726 	/* Use ret to determine if the file still exists. */
1727 	if ((nd->nd_flag & ND_NFSV41) != 0 && error == 0) {
1728 		mp = vfs_busyfs(&fh2.fh_fsid);
1729 		if (mp != NULL) {
1730 			/* Find out if the file still exists. */
1731 			ret = VFS_FHTOVP(mp, &fh2.fh_fid, LK_SHARED, &newvp);
1732 			if (ret == 0)
1733 				vput(newvp);
1734 			else
1735 				ret = ESTALE;
1736 			vfs_unbusy(mp);
1737 		} else {
1738 			ret = ESTALE;
1739 		}
1740 		if (ret == ESTALE) {
1741 			/* Get rid of any delegation. */
1742 			nfsrv_removedeleg(&fh2, nd, p);
1743 		}
1744 	}
1745 
1746 	nfsvno_relpathbuf(tondp);
1747 out1:
1748 	nfsvno_relpathbuf(fromndp);
1749 	NFSEXITCODE(error);
1750 	return (error);
1751 }
1752 
1753 /*
1754  * Link vnode op.
1755  */
1756 int
nfsvno_link(struct nameidata * ndp,struct vnode * vp,nfsquad_t clientid,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1757 nfsvno_link(struct nameidata *ndp, struct vnode *vp, nfsquad_t clientid,
1758     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1759 {
1760 	struct vnode *xp;
1761 	int error = 0;
1762 
1763 	xp = ndp->ni_vp;
1764 	if (xp != NULL) {
1765 		error = EEXIST;
1766 	} else {
1767 		xp = ndp->ni_dvp;
1768 		if (vp->v_mount != xp->v_mount)
1769 			error = EXDEV;
1770 	}
1771 	if (!error) {
1772 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1773 		if (!VN_IS_DOOMED(vp)) {
1774 			error = nfsrv_checkremove(vp, 0, NULL, clientid, p);
1775 			if (error == 0)
1776 				error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1777 		} else
1778 			error = EPERM;
1779 		if (ndp->ni_dvp == vp) {
1780 			vrele(ndp->ni_dvp);
1781 			NFSVOPUNLOCK(vp);
1782 		} else {
1783 			vref(vp);
1784 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1785 		}
1786 	} else {
1787 		if (ndp->ni_dvp == ndp->ni_vp)
1788 			vrele(ndp->ni_dvp);
1789 		else
1790 			vput(ndp->ni_dvp);
1791 		if (ndp->ni_vp)
1792 			vrele(ndp->ni_vp);
1793 	}
1794 	nfsvno_relpathbuf(ndp);
1795 	NFSEXITCODE(error);
1796 	return (error);
1797 }
1798 
1799 /*
1800  * Do the fsync() appropriate for the commit.
1801  */
1802 int
nfsvno_fsync(struct vnode * vp,u_int64_t off,int cnt,struct ucred * cred,struct thread * td)1803 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1804     struct thread *td)
1805 {
1806 	int error = 0;
1807 
1808 	/*
1809 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1810 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1811 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1812 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1813 	 * File systems that do not use the buffer cache (as indicated
1814 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1815 	 */
1816 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1817 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1818 		/*
1819 		 * Give up and do the whole thing
1820 		 */
1821 		vnode_pager_clean_sync(vp);
1822 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1823 	} else {
1824 		/*
1825 		 * Locate and synchronously write any buffers that fall
1826 		 * into the requested range.  Note:  we are assuming that
1827 		 * f_iosize is a power of 2.
1828 		 */
1829 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1830 		int iomask = iosize - 1;
1831 		struct bufobj *bo;
1832 		daddr_t lblkno;
1833 
1834 		/*
1835 		 * Align to iosize boundary, super-align to page boundary.
1836 		 */
1837 		if (off & iomask) {
1838 			cnt += off & iomask;
1839 			off &= ~(u_quad_t)iomask;
1840 		}
1841 		if (off & PAGE_MASK) {
1842 			cnt += off & PAGE_MASK;
1843 			off &= ~(u_quad_t)PAGE_MASK;
1844 		}
1845 		lblkno = off / iosize;
1846 
1847 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1848 			VM_OBJECT_WLOCK(vp->v_object);
1849 			vm_object_page_clean(vp->v_object, off, off + cnt,
1850 			    OBJPC_SYNC);
1851 			VM_OBJECT_WUNLOCK(vp->v_object);
1852 		}
1853 
1854 		bo = &vp->v_bufobj;
1855 		BO_LOCK(bo);
1856 		while (cnt > 0) {
1857 			struct buf *bp;
1858 
1859 			/*
1860 			 * If we have a buffer and it is marked B_DELWRI we
1861 			 * have to lock and write it.  Otherwise the prior
1862 			 * write is assumed to have already been committed.
1863 			 *
1864 			 * gbincore() can return invalid buffers now so we
1865 			 * have to check that bit as well (though B_DELWRI
1866 			 * should not be set if B_INVAL is set there could be
1867 			 * a race here since we haven't locked the buffer).
1868 			 */
1869 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1870 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1871 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1872 					BO_LOCK(bo);
1873 					continue; /* retry */
1874 				}
1875 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1876 				    B_DELWRI) {
1877 					bremfree(bp);
1878 					bp->b_flags &= ~B_ASYNC;
1879 					bwrite(bp);
1880 					++nfs_commit_miss;
1881 				} else
1882 					BUF_UNLOCK(bp);
1883 				BO_LOCK(bo);
1884 			}
1885 			++nfs_commit_blks;
1886 			if (cnt < iosize)
1887 				break;
1888 			cnt -= iosize;
1889 			++lblkno;
1890 		}
1891 		BO_UNLOCK(bo);
1892 	}
1893 	NFSEXITCODE(error);
1894 	return (error);
1895 }
1896 
1897 /*
1898  * Statfs vnode op.
1899  */
1900 int
nfsvno_statfs(struct vnode * vp,struct statfs * sf)1901 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1902 {
1903 	struct statfs *tsf;
1904 	int error;
1905 
1906 	tsf = NULL;
1907 	if (nfsrv_devidcnt > 0) {
1908 		/* For a pNFS service, get the DS numbers. */
1909 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1910 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1911 		if (error != 0) {
1912 			free(tsf, M_TEMP);
1913 			tsf = NULL;
1914 		}
1915 	}
1916 	error = VFS_STATFS(vp->v_mount, sf);
1917 	if (error == 0) {
1918 		if (tsf != NULL) {
1919 			sf->f_blocks = tsf->f_blocks;
1920 			sf->f_bavail = tsf->f_bavail;
1921 			sf->f_bfree = tsf->f_bfree;
1922 			sf->f_bsize = tsf->f_bsize;
1923 		}
1924 		/*
1925 		 * Since NFS handles these values as unsigned on the
1926 		 * wire, there is no way to represent negative values,
1927 		 * so set them to 0. Without this, they will appear
1928 		 * to be very large positive values for clients like
1929 		 * Solaris10.
1930 		 */
1931 		if (sf->f_bavail < 0)
1932 			sf->f_bavail = 0;
1933 		if (sf->f_ffree < 0)
1934 			sf->f_ffree = 0;
1935 	}
1936 	free(tsf, M_TEMP);
1937 	NFSEXITCODE(error);
1938 	return (error);
1939 }
1940 
1941 /*
1942  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1943  * must handle nfsrv_opencheck() calls after any other access checks.
1944  */
1945 void
nfsvno_open(struct nfsrv_descript * nd,struct nameidata * ndp,nfsquad_t clientid,nfsv4stateid_t * stateidp,struct nfsstate * stp,int * exclusive_flagp,struct nfsvattr * nvap,int32_t * cverf,int create,NFSACL_T * aclp,nfsattrbit_t * attrbitp,struct ucred * cred,bool done_namei,struct nfsexstuff * exp,struct vnode ** vpp)1946 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1947     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1948     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1949     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei,
1950     struct nfsexstuff *exp, struct vnode **vpp)
1951 {
1952 	struct vnode *vp = NULL;
1953 	u_quad_t tempsize;
1954 	struct nfsexstuff nes;
1955 	struct thread *p = curthread;
1956 	uint32_t oldrepstat;
1957 
1958 	if (ndp->ni_vp == NULL) {
1959 		/*
1960 		 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be
1961 		 * set true, since cleanup after nfsvno_namei() is needed.
1962 		 */
1963 		oldrepstat = nd->nd_repstat;
1964 		nd->nd_repstat = nfsrv_opencheck(clientid,
1965 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1966 		if (nd->nd_repstat != 0 && oldrepstat == 0)
1967 			done_namei = true;
1968 	}
1969 	if (!nd->nd_repstat) {
1970 		if (ndp->ni_vp == NULL) {
1971 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1972 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1973 			/* For a pNFS server, create the data file on a DS. */
1974 			if (nd->nd_repstat == 0) {
1975 				/*
1976 				 * Create a data file on a DS for a pNFS server.
1977 				 * This function just returns if not
1978 				 * running a pNFS DS or the creation fails.
1979 				 */
1980 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1981 				    cred, p);
1982 			}
1983 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1984 			    &ndp->ni_vp : NULL, false);
1985 			nfsvno_relpathbuf(ndp);
1986 			if (!nd->nd_repstat) {
1987 				if (*exclusive_flagp) {
1988 					*exclusive_flagp = 0;
1989 					NFSVNO_ATTRINIT(nvap);
1990 					nvap->na_atime.tv_sec = cverf[0];
1991 					nvap->na_atime.tv_nsec = cverf[1];
1992 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1993 					    &nvap->na_vattr, cred);
1994 					if (nd->nd_repstat != 0) {
1995 						vput(ndp->ni_vp);
1996 						ndp->ni_vp = NULL;
1997 						nd->nd_repstat = NFSERR_NOTSUPP;
1998 					} else
1999 						NFSSETBIT_ATTRBIT(attrbitp,
2000 						    NFSATTRBIT_TIMEACCESS);
2001 				} else {
2002 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
2003 					    aclp, p, attrbitp, exp);
2004 				}
2005 			}
2006 			vp = ndp->ni_vp;
2007 		} else {
2008 			nfsvno_relpathbuf(ndp);
2009 			vp = ndp->ni_vp;
2010 			if (create == NFSV4OPEN_CREATE) {
2011 				if (ndp->ni_dvp == vp)
2012 					vrele(ndp->ni_dvp);
2013 				else
2014 					vput(ndp->ni_dvp);
2015 			}
2016 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
2017 				if (ndp->ni_cnd.cn_flags & RDONLY)
2018 					NFSVNO_SETEXRDONLY(&nes);
2019 				else
2020 					NFSVNO_EXINIT(&nes);
2021 				nd->nd_repstat = nfsvno_accchk(vp,
2022 				    VWRITE, cred, &nes, p,
2023 				    NFSACCCHK_NOOVERRIDE,
2024 				    NFSACCCHK_VPISLOCKED, NULL);
2025 				nd->nd_repstat = nfsrv_opencheck(clientid,
2026 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
2027 				if (!nd->nd_repstat) {
2028 					tempsize = nvap->na_size;
2029 					NFSVNO_ATTRINIT(nvap);
2030 					nvap->na_size = tempsize;
2031 					nd->nd_repstat = nfsvno_setattr(vp,
2032 					    nvap, cred, p, exp);
2033 				}
2034 			} else if (vp->v_type == VREG) {
2035 				nd->nd_repstat = nfsrv_opencheck(clientid,
2036 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
2037 			}
2038 		}
2039 	} else if (done_namei) {
2040 		KASSERT(create == NFSV4OPEN_CREATE,
2041 		    ("nfsvno_open: not create"));
2042 		/*
2043 		 * done_namei is set when nfsvno_namei() has completed
2044 		 * successfully, but a subsequent error was set in
2045 		 * nd_repstat.  As such, cleanup of the nfsvno_namei()
2046 		 * results is required.
2047 		 */
2048 		nfsvno_relpathbuf(ndp);
2049 		if (ndp->ni_dvp == ndp->ni_vp)
2050 			vrele(ndp->ni_dvp);
2051 		else
2052 			vput(ndp->ni_dvp);
2053 		if (ndp->ni_vp)
2054 			vput(ndp->ni_vp);
2055 	}
2056 	*vpp = vp;
2057 
2058 	NFSEXITCODE2(0, nd);
2059 }
2060 
2061 /*
2062  * Updates the file rev and sets the mtime and ctime
2063  * to the current clock time, returning the va_filerev and va_Xtime
2064  * values.
2065  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
2066  */
2067 int
nfsvno_updfilerev(struct vnode * vp,struct nfsvattr * nvap,struct nfsrv_descript * nd,struct thread * p)2068 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
2069     struct nfsrv_descript *nd, struct thread *p)
2070 {
2071 	struct vattr va;
2072 
2073 	VATTR_NULL(&va);
2074 	vfs_timestamp(&va.va_mtime);
2075 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
2076 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
2077 		if (VN_IS_DOOMED(vp))
2078 			return (ESTALE);
2079 	}
2080 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
2081 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
2082 	return (0);
2083 }
2084 
2085 /*
2086  * Glue routine to nfsv4_fillattr().
2087  */
2088 int
nfsvno_fillattr(struct nfsrv_descript * nd,struct mount * mp,struct vnode * vp,struct nfsvattr * nvap,fhandle_t * fhp,int rderror,nfsattrbit_t * attrbitp,struct ucred * cred,struct thread * p,int isdgram,int reterr,int supports_nfsv4acls,int at_root,uint64_t mounted_on_fileno)2089 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
2090     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
2091     struct ucred *cred, struct thread *p, int isdgram, int reterr,
2092     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
2093 {
2094 	struct statfs *sf;
2095 	int error;
2096 
2097 	sf = NULL;
2098 	if (nfsrv_devidcnt > 0 &&
2099 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
2100 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
2101 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
2102 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2103 		error = nfsrv_pnfsstatfs(sf, mp);
2104 		if (error != 0) {
2105 			free(sf, M_TEMP);
2106 			sf = NULL;
2107 		}
2108 	}
2109 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2110 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2111 	    mounted_on_fileno, sf);
2112 	free(sf, M_TEMP);
2113 	NFSEXITCODE2(0, nd);
2114 	return (error);
2115 }
2116 
2117 /*
2118  * Convert a dirent d_type to a vnode type.
2119  */
nfs_dtypetovtype(struct nfsvattr * nvap,struct vnode * vp,uint8_t dtype)2120 static void nfs_dtypetovtype(struct nfsvattr *nvap, struct vnode *vp,
2121     uint8_t dtype)
2122 {
2123 
2124 	if ((vn_irflag_read(vp) & VIRF_NAMEDDIR) != 0) {
2125 		nvap->na_type = VREG;
2126 		nvap->na_bsdflags |= SFBSD_NAMEDATTR;
2127 	} else if (dtype <= DT_WHT) {
2128 		nvap->na_type = dtype_to_vnode[dtype];
2129 	} else {
2130 		nvap->na_type = VNON;
2131 	}
2132 }
2133 
2134 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2135 /*
2136  * nfs readdir service
2137  * - mallocs what it thinks is enough to read
2138  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2139  * - calls VOP_READDIR()
2140  * - loops around building the reply
2141  *	if the output generated exceeds count break out of loop
2142  *	The NFSM_CLGET macro is used here so that the reply will be packed
2143  *	tightly in mbuf clusters.
2144  * - it trims out records with d_fileno == 0
2145  *	this doesn't matter for Unix clients, but they might confuse clients
2146  *	for other os'.
2147  * - it trims out records with d_type == DT_WHT
2148  *	these cannot be seen through NFS (unless we extend the protocol)
2149  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2150  * PS: The NFS protocol spec. does not clarify what the "count" byte
2151  *	argument is a count of.. just name strings and file id's or the
2152  *	entire reply rpc or ...
2153  *	I tried just file name and id sizes and it confused the Sun client,
2154  *	so I am using the full rpc size now. The "paranoia.." comment refers
2155  *	to including the status longwords that are not a part of the dir.
2156  *	"entry" structures, but are in the rpc.
2157  */
2158 int
nfsrvd_readdir(struct nfsrv_descript * nd,int isdgram,struct vnode * vp,struct nfsexstuff * exp)2159 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2160     struct vnode *vp, struct nfsexstuff *exp)
2161 {
2162 	struct dirent *dp;
2163 	u_int32_t *tl;
2164 	int dirlen;
2165 	char *cpos, *cend, *rbuf;
2166 	struct nfsvattr at;
2167 	int nlen, error = 0, getret = 1;
2168 	int siz, cnt, fullsiz, eofflag, ncookies;
2169 	u_int64_t off, toff, verf __unused;
2170 	uint64_t *cookies = NULL, *cookiep;
2171 	struct uio io;
2172 	struct iovec iv;
2173 	int is_ufs;
2174 	struct thread *p = curthread;
2175 
2176 	if (nd->nd_repstat) {
2177 		nfsrv_postopattr(nd, getret, &at);
2178 		goto out;
2179 	}
2180 	if (nd->nd_flag & ND_NFSV2) {
2181 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2182 		off = fxdr_unsigned(u_quad_t, *tl++);
2183 	} else {
2184 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2185 		off = fxdr_hyper(tl);
2186 		tl += 2;
2187 		verf = fxdr_hyper(tl);
2188 		tl += 2;
2189 	}
2190 	toff = off;
2191 	cnt = fxdr_unsigned(int, *tl);
2192 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2193 		cnt = NFS_SRVMAXDATA(nd);
2194 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2195 	fullsiz = siz;
2196 	if (nd->nd_flag & ND_NFSV3) {
2197 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2198 		    NULL);
2199 #if 0
2200 		/*
2201 		 * va_filerev is not sufficient as a cookie verifier,
2202 		 * since it is not supposed to change when entries are
2203 		 * removed/added unless that offset cookies returned to
2204 		 * the client are no longer valid.
2205 		 */
2206 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2207 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2208 #endif
2209 	}
2210 	if (!nd->nd_repstat && vp->v_type != VDIR)
2211 		nd->nd_repstat = NFSERR_NOTDIR;
2212 	if (nd->nd_repstat == 0 && cnt == 0) {
2213 		if (nd->nd_flag & ND_NFSV2)
2214 			/* NFSv2 does not have NFSERR_TOOSMALL */
2215 			nd->nd_repstat = EPERM;
2216 		else
2217 			nd->nd_repstat = NFSERR_TOOSMALL;
2218 	}
2219 	if (!nd->nd_repstat)
2220 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2221 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2222 		    NFSACCCHK_VPISLOCKED, NULL);
2223 	if (nd->nd_repstat) {
2224 		vput(vp);
2225 		if (nd->nd_flag & ND_NFSV3)
2226 			nfsrv_postopattr(nd, getret, &at);
2227 		goto out;
2228 	}
2229 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2230 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2231 again:
2232 	eofflag = 0;
2233 	if (cookies) {
2234 		free(cookies, M_TEMP);
2235 		cookies = NULL;
2236 	}
2237 
2238 	iv.iov_base = rbuf;
2239 	iv.iov_len = siz;
2240 	io.uio_iov = &iv;
2241 	io.uio_iovcnt = 1;
2242 	io.uio_offset = (off_t)off;
2243 	io.uio_resid = siz;
2244 	io.uio_segflg = UIO_SYSSPACE;
2245 	io.uio_rw = UIO_READ;
2246 	io.uio_td = NULL;
2247 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2248 	    &cookies);
2249 	off = (u_int64_t)io.uio_offset;
2250 	if (io.uio_resid)
2251 		siz -= io.uio_resid;
2252 
2253 	if (!cookies && !nd->nd_repstat)
2254 		nd->nd_repstat = NFSERR_PERM;
2255 	if (nd->nd_flag & ND_NFSV3) {
2256 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2257 		if (!nd->nd_repstat)
2258 			nd->nd_repstat = getret;
2259 	}
2260 
2261 	/*
2262 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2263 	 */
2264 	if (nd->nd_repstat) {
2265 		vput(vp);
2266 		free(rbuf, M_TEMP);
2267 		if (cookies)
2268 			free(cookies, M_TEMP);
2269 		if (nd->nd_flag & ND_NFSV3)
2270 			nfsrv_postopattr(nd, getret, &at);
2271 		goto out;
2272 	}
2273 	/*
2274 	 * If nothing read, return eof
2275 	 * rpc reply
2276 	 */
2277 	if (siz == 0) {
2278 		vput(vp);
2279 		if (nd->nd_flag & ND_NFSV2) {
2280 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2281 		} else {
2282 			nfsrv_postopattr(nd, getret, &at);
2283 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2284 			txdr_hyper(at.na_filerev, tl);
2285 			tl += 2;
2286 		}
2287 		*tl++ = newnfs_false;
2288 		*tl = newnfs_true;
2289 		free(rbuf, M_TEMP);
2290 		free(cookies, M_TEMP);
2291 		goto out;
2292 	}
2293 
2294 	/*
2295 	 * Check for degenerate cases of nothing useful read.
2296 	 * If so go try again
2297 	 */
2298 	cpos = rbuf;
2299 	cend = rbuf + siz;
2300 	dp = (struct dirent *)cpos;
2301 	cookiep = cookies;
2302 
2303 	/*
2304 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2305 	 * directory offset up to a block boundary, so it is necessary to
2306 	 * skip over the records that precede the requested offset. This
2307 	 * requires the assumption that file offset cookies monotonically
2308 	 * increase.
2309 	 */
2310 	while (cpos < cend && ncookies > 0 &&
2311 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2312 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2313 		cpos += dp->d_reclen;
2314 		dp = (struct dirent *)cpos;
2315 		cookiep++;
2316 		ncookies--;
2317 	}
2318 	if (cpos >= cend || ncookies == 0) {
2319 		siz = fullsiz;
2320 		toff = off;
2321 		goto again;
2322 	}
2323 	vput(vp);
2324 
2325 	/*
2326 	 * If cnt > MCLBYTES and the reply will not be saved, use
2327 	 * ext_pgs mbufs for TLS.
2328 	 * For NFSv4.0, we do not know for sure if the reply will
2329 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2330 	 */
2331 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2332 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2333 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2334 		nd->nd_flag |= ND_EXTPG;
2335 
2336 	/*
2337 	 * dirlen is the size of the reply, including all XDR and must
2338 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2339 	 * if the XDR should be included in "count", but to be safe, we do.
2340 	 * (Include the two booleans at the end of the reply in dirlen now.)
2341 	 */
2342 	if (nd->nd_flag & ND_NFSV3) {
2343 		nfsrv_postopattr(nd, getret, &at);
2344 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2345 		txdr_hyper(at.na_filerev, tl);
2346 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2347 	} else {
2348 		dirlen = 2 * NFSX_UNSIGNED;
2349 	}
2350 
2351 	/* Loop through the records and build reply */
2352 	while (cpos < cend && ncookies > 0) {
2353 		nlen = dp->d_namlen;
2354 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2355 			nlen <= NFS_MAXNAMLEN) {
2356 			if (nd->nd_flag & ND_NFSV3)
2357 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2358 			else
2359 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2360 			if (dirlen > cnt) {
2361 				eofflag = 0;
2362 				break;
2363 			}
2364 
2365 			/*
2366 			 * Build the directory record xdr from
2367 			 * the dirent entry.
2368 			 */
2369 			if (nd->nd_flag & ND_NFSV3) {
2370 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2371 				*tl++ = newnfs_true;
2372 				txdr_hyper(dp->d_fileno, tl);
2373 			} else {
2374 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2375 				*tl++ = newnfs_true;
2376 				*tl = txdr_unsigned(dp->d_fileno);
2377 			}
2378 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2379 			if (nd->nd_flag & ND_NFSV3) {
2380 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2381 				txdr_hyper(*cookiep, tl);
2382 			} else {
2383 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2384 				*tl = txdr_unsigned(*cookiep);
2385 			}
2386 		}
2387 		cpos += dp->d_reclen;
2388 		dp = (struct dirent *)cpos;
2389 		cookiep++;
2390 		ncookies--;
2391 	}
2392 	if (cpos < cend)
2393 		eofflag = 0;
2394 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2395 	*tl++ = newnfs_false;
2396 	if (eofflag)
2397 		*tl = newnfs_true;
2398 	else
2399 		*tl = newnfs_false;
2400 	free(rbuf, M_TEMP);
2401 	free(cookies, M_TEMP);
2402 
2403 out:
2404 	NFSEXITCODE2(0, nd);
2405 	return (0);
2406 nfsmout:
2407 	vput(vp);
2408 	NFSEXITCODE2(error, nd);
2409 	return (error);
2410 }
2411 
2412 /*
2413  * Readdirplus for V3 and Readdir for V4.
2414  */
2415 int
nfsrvd_readdirplus(struct nfsrv_descript * nd,int isdgram,struct vnode * vp,struct nfsexstuff * exp)2416 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2417     struct vnode *vp, struct nfsexstuff *exp)
2418 {
2419 	struct dirent *dp;
2420 	u_int32_t *tl;
2421 	int dirlen;
2422 	char *cpos, *cend, *rbuf;
2423 	struct vnode *nvp;
2424 	fhandle_t nfh;
2425 	struct nfsvattr nva, at, *nvap = &nva;
2426 	struct mbuf *mb0, *mb1;
2427 	struct nfsreferral *refp;
2428 	int nlen, r, error = 0, getret = 1, usevget = 1;
2429 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2430 	caddr_t bpos0, bpos1;
2431 	u_int64_t off, toff, verf __unused;
2432 	uint64_t *cookies = NULL, *cookiep;
2433 	nfsattrbit_t attrbits, rderrbits, savbits, refbits;
2434 	struct uio io;
2435 	struct iovec iv;
2436 	struct componentname cn;
2437 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2438 	struct mount *mp, *new_mp;
2439 	uint64_t mounted_on_fileno;
2440 	struct thread *p = curthread;
2441 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2442 
2443 	if (nd->nd_repstat) {
2444 		nfsrv_postopattr(nd, getret, &at);
2445 		goto out;
2446 	}
2447 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2448 	off = fxdr_hyper(tl);
2449 	toff = off;
2450 	tl += 2;
2451 	verf = fxdr_hyper(tl);
2452 	tl += 2;
2453 	siz = fxdr_unsigned(int, *tl++);
2454 	cnt = fxdr_unsigned(int, *tl);
2455 
2456 	/*
2457 	 * Use the server's maximum data transfer size as the upper bound
2458 	 * on reply datalen.
2459 	 */
2460 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2461 		cnt = NFS_SRVMAXDATA(nd);
2462 
2463 	/*
2464 	 * siz is a "hint" of how much directory information (name, fileid,
2465 	 * cookie) should be in the reply. At least one client "hints" 0,
2466 	 * so I set it to cnt for that case. I also round it up to the
2467 	 * next multiple of DIRBLKSIZ.
2468 	 * Since the size of a Readdirplus directory entry reply will always
2469 	 * be greater than a directory entry returned by VOP_READDIR(), it
2470 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2471 	 * VOP_READDIR().
2472 	 */
2473 	if (siz <= 0)
2474 		siz = cnt;
2475 	else if (siz > NFS_SRVMAXDATA(nd))
2476 		siz = NFS_SRVMAXDATA(nd);
2477 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2478 
2479 	if (nd->nd_flag & ND_NFSV4) {
2480 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2481 		if (error)
2482 			goto nfsmout;
2483 		NFSSET_ATTRBIT(&savbits, &attrbits);
2484 		NFSSET_ATTRBIT(&refbits, &attrbits);
2485 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2486 		NFSZERO_ATTRBIT(&rderrbits);
2487 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2488 		/*
2489 		 * If these 4 bits are the only attributes requested by the
2490 		 * client, they can be satisfied without acquiring the vnode
2491 		 * for the file object unless it is a directory.
2492 		 * This will be indicated by savbits being all 0s.
2493 		 */
2494 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_TYPE);
2495 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_FILEID);
2496 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_MOUNTEDONFILEID);
2497 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_RDATTRERROR);
2498 	} else {
2499 		NFSZERO_ATTRBIT(&attrbits);
2500 	}
2501 	fullsiz = siz;
2502 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2503 #if 0
2504 	if (!nd->nd_repstat) {
2505 	    if (off && verf != at.na_filerev) {
2506 		/*
2507 		 * va_filerev is not sufficient as a cookie verifier,
2508 		 * since it is not supposed to change when entries are
2509 		 * removed/added unless that offset cookies returned to
2510 		 * the client are no longer valid.
2511 		 */
2512 		if (nd->nd_flag & ND_NFSV4) {
2513 			nd->nd_repstat = NFSERR_NOTSAME;
2514 		} else {
2515 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2516 		}
2517 	    }
2518 	}
2519 #endif
2520 	if (!nd->nd_repstat && vp->v_type != VDIR)
2521 		nd->nd_repstat = NFSERR_NOTDIR;
2522 	if (!nd->nd_repstat && cnt == 0)
2523 		nd->nd_repstat = NFSERR_TOOSMALL;
2524 	if (!nd->nd_repstat)
2525 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2526 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2527 		    NFSACCCHK_VPISLOCKED, NULL);
2528 	if (nd->nd_repstat) {
2529 		vput(vp);
2530 		if (nd->nd_flag & ND_NFSV3)
2531 			nfsrv_postopattr(nd, getret, &at);
2532 		goto out;
2533 	}
2534 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2535 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2536 
2537 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2538 again:
2539 	eofflag = 0;
2540 	if (cookies) {
2541 		free(cookies, M_TEMP);
2542 		cookies = NULL;
2543 	}
2544 
2545 	iv.iov_base = rbuf;
2546 	iv.iov_len = siz;
2547 	io.uio_iov = &iv;
2548 	io.uio_iovcnt = 1;
2549 	io.uio_offset = (off_t)off;
2550 	io.uio_resid = siz;
2551 	io.uio_segflg = UIO_SYSSPACE;
2552 	io.uio_rw = UIO_READ;
2553 	io.uio_td = NULL;
2554 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2555 	    &cookies);
2556 	off = (u_int64_t)io.uio_offset;
2557 	if (io.uio_resid)
2558 		siz -= io.uio_resid;
2559 
2560 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2561 
2562 	if (!cookies && !nd->nd_repstat)
2563 		nd->nd_repstat = NFSERR_PERM;
2564 	if (!nd->nd_repstat)
2565 		nd->nd_repstat = getret;
2566 	if (nd->nd_repstat) {
2567 		vput(vp);
2568 		if (cookies)
2569 			free(cookies, M_TEMP);
2570 		free(rbuf, M_TEMP);
2571 		if (nd->nd_flag & ND_NFSV3)
2572 			nfsrv_postopattr(nd, getret, &at);
2573 		goto out;
2574 	}
2575 	/*
2576 	 * If nothing read, return eof
2577 	 * rpc reply
2578 	 */
2579 	if (siz == 0) {
2580 		vput(vp);
2581 		if (nd->nd_flag & ND_NFSV3)
2582 			nfsrv_postopattr(nd, getret, &at);
2583 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2584 		txdr_hyper(at.na_filerev, tl);
2585 		tl += 2;
2586 		*tl++ = newnfs_false;
2587 		*tl = newnfs_true;
2588 		free(cookies, M_TEMP);
2589 		free(rbuf, M_TEMP);
2590 		goto out;
2591 	}
2592 
2593 	/*
2594 	 * Check for degenerate cases of nothing useful read.
2595 	 * If so go try again
2596 	 */
2597 	cpos = rbuf;
2598 	cend = rbuf + siz;
2599 	dp = (struct dirent *)cpos;
2600 	cookiep = cookies;
2601 
2602 	/*
2603 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2604 	 * directory offset up to a block boundary, so it is necessary to
2605 	 * skip over the records that precede the requested offset. This
2606 	 * requires the assumption that file offset cookies monotonically
2607 	 * increase.
2608 	 */
2609 	while (cpos < cend && ncookies > 0 &&
2610 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2611 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2612 	   ((nd->nd_flag & ND_NFSV4) &&
2613 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2614 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2615 		cpos += dp->d_reclen;
2616 		dp = (struct dirent *)cpos;
2617 		cookiep++;
2618 		ncookies--;
2619 	}
2620 	if (cpos >= cend || ncookies == 0) {
2621 		siz = fullsiz;
2622 		toff = off;
2623 		goto again;
2624 	}
2625 
2626 	/*
2627 	 * Busy the file system so that the mount point won't go away
2628 	 * and, as such, VFS_VGET() can be used safely.
2629 	 */
2630 	mp = vp->v_mount;
2631 	vfs_ref(mp);
2632 	NFSVOPUNLOCK(vp);
2633 	nd->nd_repstat = vfs_busy(mp, 0);
2634 	vfs_rel(mp);
2635 	if (nd->nd_repstat != 0) {
2636 		vrele(vp);
2637 		free(cookies, M_TEMP);
2638 		free(rbuf, M_TEMP);
2639 		if (nd->nd_flag & ND_NFSV3)
2640 			nfsrv_postopattr(nd, getret, &at);
2641 		goto out;
2642 	}
2643 
2644 	/*
2645 	 * Check to see if entries in this directory can be safely acquired
2646 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2647 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2648 	 * automount of the snapshot directory that is required will
2649 	 * be done.
2650 	 * This needs to be done here for NFSv4, since NFSv4 never does
2651 	 * a VFS_VGET() for "." or "..".
2652 	 */
2653 	if (is_zfs == 1) {
2654 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2655 		if (r == EOPNOTSUPP) {
2656 			usevget = 0;
2657 			cn.cn_nameiop = LOOKUP;
2658 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2659 			cn.cn_cred = nd->nd_cred;
2660 		} else if (r == 0)
2661 			vput(nvp);
2662 	}
2663 
2664 	/*
2665 	 * If the reply is likely to exceed MCLBYTES and the reply will
2666 	 * not be saved, use ext_pgs mbufs for TLS.
2667 	 * It is difficult to predict how large each entry will be and
2668 	 * how many entries have been read, so just assume the directory
2669 	 * entries grow by a factor of 4 when attributes are included.
2670 	 * For NFSv4.0, we do not know for sure if the reply will
2671 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2672 	 */
2673 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2674 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2675 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2676 		nd->nd_flag |= ND_EXTPG;
2677 
2678 	/*
2679 	 * Save this position, in case there is an error before one entry
2680 	 * is created.
2681 	 */
2682 	mb0 = nd->nd_mb;
2683 	bpos0 = nd->nd_bpos;
2684 	bextpg0 = nd->nd_bextpg;
2685 	bextpgsiz0 = nd->nd_bextpgsiz;
2686 
2687 	/*
2688 	 * Fill in the first part of the reply.
2689 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2690 	 * (Include the two booleans at the end of the reply in dirlen now,
2691 	 *  so we recognize when we have exceeded cnt.)
2692 	 */
2693 	if (nd->nd_flag & ND_NFSV3) {
2694 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2695 		nfsrv_postopattr(nd, getret, &at);
2696 	} else {
2697 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2698 	}
2699 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2700 	txdr_hyper(at.na_filerev, tl);
2701 
2702 	/*
2703 	 * Save this position, in case there is an empty reply needed.
2704 	 */
2705 	mb1 = nd->nd_mb;
2706 	bpos1 = nd->nd_bpos;
2707 	bextpg1 = nd->nd_bextpg;
2708 	bextpgsiz1 = nd->nd_bextpgsiz;
2709 
2710 	/* Loop through the records and build reply */
2711 	entrycnt = 0;
2712 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2713 		nlen = dp->d_namlen;
2714 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2715 		    nlen <= NFS_MAXNAMLEN &&
2716 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2717 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2718 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2719 			/*
2720 			 * Save the current position in the reply, in case
2721 			 * this entry exceeds cnt.
2722 			 */
2723 			mb1 = nd->nd_mb;
2724 			bpos1 = nd->nd_bpos;
2725 			bextpg1 = nd->nd_bextpg;
2726 			bextpgsiz1 = nd->nd_bextpgsiz;
2727 
2728 			/*
2729 			 * For readdir_and_lookup get the vnode using
2730 			 * the file number.
2731 			 */
2732 			nvp = NULL;
2733 			refp = NULL;
2734 			r = 0;
2735 			at_root = 0;
2736 			needs_unbusy = 0;
2737 			new_mp = mp;
2738 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2739 			if ((nd->nd_flag & ND_NFSV3) ||
2740 			    NFSNONZERO_ATTRBIT(&savbits) ||
2741 			    dp->d_type == DT_UNKNOWN ||
2742 			    (dp->d_type == DT_DIR &&
2743 			     nfsrv_enable_crossmntpt != 0)) {
2744 				if (nd->nd_flag & ND_NFSV4)
2745 					refp = nfsv4root_getreferral(NULL,
2746 					    vp, dp->d_fileno);
2747 				if (refp == NULL) {
2748 					if (usevget)
2749 						r = VFS_VGET(mp, dp->d_fileno,
2750 						    LK_SHARED, &nvp);
2751 					else
2752 						r = EOPNOTSUPP;
2753 					if (r == 0 && (vn_irflag_read(vp) &
2754 					    VIRF_NAMEDDIR) != 0)
2755 						vn_irflag_set_cond(nvp,
2756 						    VIRF_NAMEDATTR);
2757 					if (r == EOPNOTSUPP) {
2758 						if (usevget) {
2759 							usevget = 0;
2760 							cn.cn_nameiop = LOOKUP;
2761 							cn.cn_lkflags =
2762 							    LK_SHARED |
2763 							    LK_RETRY;
2764 							cn.cn_cred =
2765 							    nd->nd_cred;
2766 						}
2767 						cn.cn_nameptr = dp->d_name;
2768 						cn.cn_namelen = nlen;
2769 						cn.cn_flags = ISLASTCN |
2770 						    NOFOLLOW | LOCKLEAF;
2771 						if ((vn_irflag_read(vp) &
2772 						    VIRF_NAMEDDIR) != 0)
2773 							cn.cn_flags |=
2774 							    OPENNAMED;
2775 						if (nlen == 2 &&
2776 						    dp->d_name[0] == '.' &&
2777 						    dp->d_name[1] == '.')
2778 							cn.cn_flags |=
2779 							    ISDOTDOT;
2780 						if (NFSVOPLOCK(vp, LK_SHARED)
2781 						    != 0) {
2782 							nd->nd_repstat = EPERM;
2783 							break;
2784 						}
2785 						if ((vp->v_vflag & VV_ROOT) != 0
2786 						    && (cn.cn_flags & ISDOTDOT)
2787 						    != 0) {
2788 							vref(vp);
2789 							nvp = vp;
2790 							r = 0;
2791 						} else {
2792 							r = VOP_LOOKUP(vp, &nvp,
2793 							    &cn);
2794 							if (vp != nvp)
2795 								NFSVOPUNLOCK(vp);
2796 						}
2797 					}
2798 
2799 					/*
2800 					 * For NFSv4, check to see if nvp is
2801 					 * a mount point and get the mount
2802 					 * point vnode, as required.
2803 					 */
2804 					if (r == 0 &&
2805 					    nfsrv_enable_crossmntpt != 0 &&
2806 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2807 					    nvp->v_type == VDIR &&
2808 					    nvp->v_mountedhere != NULL) {
2809 						new_mp = nvp->v_mountedhere;
2810 						r = vfs_busy(new_mp, 0);
2811 						vput(nvp);
2812 						nvp = NULL;
2813 						if (r == 0) {
2814 							r = VFS_ROOT(new_mp,
2815 							    LK_SHARED, &nvp);
2816 							needs_unbusy = 1;
2817 							if (r == 0)
2818 								at_root = 1;
2819 						}
2820 					}
2821 				}
2822 
2823 				/*
2824 				 * If we failed to look up the entry, then it
2825 				 * has become invalid, most likely removed.
2826 				 */
2827 				if (r != 0) {
2828 					if (needs_unbusy)
2829 						vfs_unbusy(new_mp);
2830 					goto invalid;
2831 				}
2832 				KASSERT(refp != NULL || nvp != NULL,
2833 				    ("%s: undetected lookup error", __func__));
2834 
2835 				if (refp == NULL &&
2836 				    ((nd->nd_flag & ND_NFSV3) ||
2837 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2838 					r = nfsvno_getfh(nvp, &nfh, p);
2839 					if (!r)
2840 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2841 						1, &attrbits);
2842 					if (r == 0 && is_zfs == 1 &&
2843 					    nfsrv_enable_crossmntpt != 0 &&
2844 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2845 					    nvp->v_type == VDIR &&
2846 					    vp->v_mount != nvp->v_mount) {
2847 					    /*
2848 					     * For a ZFS snapshot, there is a
2849 					     * pseudo mount that does not set
2850 					     * v_mountedhere, so it needs to
2851 					     * be detected via a different
2852 					     * mount structure.
2853 					     */
2854 					    at_root = 1;
2855 					    if (new_mp == mp)
2856 						new_mp = nvp->v_mount;
2857 					}
2858 				}
2859 
2860 				/*
2861 				 * If we failed to get attributes of the entry,
2862 				 * then just skip it for NFSv3 (the traditional
2863 				 * behavior in the old NFS server).
2864 				 * For NFSv4 the behavior is controlled by
2865 				 * RDATTRERROR: we either ignore the error or
2866 				 * fail the request.
2867 				 * The exception is EOPNOTSUPP, which can be
2868 				 * returned by nfsvno_getfh() for certain
2869 				 * file systems, such as devfs.  This indicates
2870 				 * that the file system cannot be exported,
2871 				 * so just skip over the entry.
2872 				 * Note that RDATTRERROR is never set for NFSv3.
2873 				 */
2874 				if (r != 0) {
2875 					if (!NFSISSET_ATTRBIT(&attrbits,
2876 					    NFSATTRBIT_RDATTRERROR) ||
2877 					    r == EOPNOTSUPP) {
2878 						vput(nvp);
2879 						if (needs_unbusy != 0)
2880 							vfs_unbusy(new_mp);
2881 						if ((nd->nd_flag & ND_NFSV3) ||
2882 						    r == EOPNOTSUPP)
2883 							goto invalid;
2884 						nd->nd_repstat = r;
2885 						break;
2886 					}
2887 				}
2888 			} else if (NFSNONZERO_ATTRBIT(&attrbits)) {
2889 				/* Only need Type and/or Fileid. */
2890 				VATTR_NULL(&nvap->na_vattr);
2891 				nvap->na_fileid = dp->d_fileno;
2892 				nfs_dtypetovtype(nvap, vp, dp->d_type);
2893 			}
2894 
2895 			/*
2896 			 * Build the directory record xdr
2897 			 */
2898 			if (nd->nd_flag & ND_NFSV3) {
2899 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2900 				*tl++ = newnfs_true;
2901 				txdr_hyper(dp->d_fileno, tl);
2902 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2903 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2904 				txdr_hyper(*cookiep, tl);
2905 				nfsrv_postopattr(nd, 0, nvap);
2906 				dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh,
2907 				    0, 1);
2908 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2909 				if (nvp != NULL)
2910 					vput(nvp);
2911 			} else {
2912 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2913 				*tl++ = newnfs_true;
2914 				txdr_hyper(*cookiep, tl);
2915 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2916 				if (nvp != NULL) {
2917 					supports_nfsv4acls =
2918 					    nfs_supportsnfsv4acls(nvp);
2919 					NFSVOPUNLOCK(nvp);
2920 				} else
2921 					supports_nfsv4acls = 0;
2922 				if (refp != NULL) {
2923 					dirlen += nfsrv_putreferralattr(nd,
2924 					    &refbits, refp, 0,
2925 					    &nd->nd_repstat);
2926 					if (nd->nd_repstat) {
2927 						if (nvp != NULL)
2928 							vrele(nvp);
2929 						if (needs_unbusy != 0)
2930 							vfs_unbusy(new_mp);
2931 						break;
2932 					}
2933 				} else if (r) {
2934 					dirlen += nfsvno_fillattr(nd, new_mp,
2935 					    nvp, nvap, &nfh, r, &rderrbits,
2936 					    nd->nd_cred, p, isdgram, 0,
2937 					    supports_nfsv4acls, at_root,
2938 					    mounted_on_fileno);
2939 				} else {
2940 					dirlen += nfsvno_fillattr(nd, new_mp,
2941 					    nvp, nvap, &nfh, r, &attrbits,
2942 					    nd->nd_cred, p, isdgram, 0,
2943 					    supports_nfsv4acls, at_root,
2944 					    mounted_on_fileno);
2945 				}
2946 				if (nvp != NULL)
2947 					vrele(nvp);
2948 				dirlen += (3 * NFSX_UNSIGNED);
2949 			}
2950 			if (needs_unbusy != 0)
2951 				vfs_unbusy(new_mp);
2952 			if (dirlen <= cnt)
2953 				entrycnt++;
2954 		}
2955 invalid:
2956 		cpos += dp->d_reclen;
2957 		dp = (struct dirent *)cpos;
2958 		cookiep++;
2959 		ncookies--;
2960 	}
2961 	vrele(vp);
2962 	vfs_unbusy(mp);
2963 
2964 	/*
2965 	 * If dirlen > cnt, we must strip off the last entry. If that
2966 	 * results in an empty reply, report NFSERR_TOOSMALL.
2967 	 */
2968 	if (dirlen > cnt || nd->nd_repstat) {
2969 		if (!nd->nd_repstat && entrycnt == 0)
2970 			nd->nd_repstat = NFSERR_TOOSMALL;
2971 		if (nd->nd_repstat) {
2972 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2973 			if (nd->nd_flag & ND_NFSV3)
2974 				nfsrv_postopattr(nd, getret, &at);
2975 		} else
2976 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2977 		eofflag = 0;
2978 	} else if (cpos < cend)
2979 		eofflag = 0;
2980 	if (!nd->nd_repstat) {
2981 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2982 		*tl++ = newnfs_false;
2983 		if (eofflag)
2984 			*tl = newnfs_true;
2985 		else
2986 			*tl = newnfs_false;
2987 	}
2988 	free(cookies, M_TEMP);
2989 	free(rbuf, M_TEMP);
2990 
2991 out:
2992 	NFSEXITCODE2(0, nd);
2993 	return (0);
2994 nfsmout:
2995 	vput(vp);
2996 	NFSEXITCODE2(error, nd);
2997 	return (error);
2998 }
2999 
3000 /*
3001  * Get the settable attributes out of the mbuf list.
3002  * (Return 0 or EBADRPC)
3003  */
3004 int
nfsrv_sattr(struct nfsrv_descript * nd,vnode_t vp,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,NFSACL_T * aclp,struct thread * p)3005 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
3006     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
3007 {
3008 	u_int32_t *tl;
3009 	struct nfsv2_sattr *sp;
3010 	int error = 0, toclient = 0;
3011 
3012 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
3013 	case ND_NFSV2:
3014 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
3015 		/*
3016 		 * Some old clients didn't fill in the high order 16bits.
3017 		 * --> check the low order 2 bytes for 0xffff
3018 		 */
3019 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
3020 			nvap->na_mode = nfstov_mode(sp->sa_mode);
3021 		if (sp->sa_uid != newnfs_xdrneg1)
3022 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
3023 		if (sp->sa_gid != newnfs_xdrneg1)
3024 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
3025 		if (sp->sa_size != newnfs_xdrneg1)
3026 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
3027 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
3028 #ifdef notyet
3029 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
3030 #else
3031 			nvap->na_atime.tv_sec =
3032 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
3033 			nvap->na_atime.tv_nsec = 0;
3034 #endif
3035 		}
3036 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
3037 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
3038 		break;
3039 	case ND_NFSV3:
3040 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3041 		if (*tl == newnfs_true) {
3042 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3043 			nvap->na_mode = nfstov_mode(*tl);
3044 		}
3045 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3046 		if (*tl == newnfs_true) {
3047 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3048 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
3049 		}
3050 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3051 		if (*tl == newnfs_true) {
3052 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3053 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
3054 		}
3055 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3056 		if (*tl == newnfs_true) {
3057 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3058 			nvap->na_size = fxdr_hyper(tl);
3059 		}
3060 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3061 		switch (fxdr_unsigned(int, *tl)) {
3062 		case NFSV3SATTRTIME_TOCLIENT:
3063 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3064 			fxdr_nfsv3time(tl, &nvap->na_atime);
3065 			toclient = 1;
3066 			break;
3067 		case NFSV3SATTRTIME_TOSERVER:
3068 			vfs_timestamp(&nvap->na_atime);
3069 			nvap->na_vaflags |= VA_UTIMES_NULL;
3070 			break;
3071 		}
3072 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3073 		switch (fxdr_unsigned(int, *tl)) {
3074 		case NFSV3SATTRTIME_TOCLIENT:
3075 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3076 			fxdr_nfsv3time(tl, &nvap->na_mtime);
3077 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
3078 			break;
3079 		case NFSV3SATTRTIME_TOSERVER:
3080 			vfs_timestamp(&nvap->na_mtime);
3081 			if (!toclient)
3082 				nvap->na_vaflags |= VA_UTIMES_NULL;
3083 			break;
3084 		}
3085 		break;
3086 	case ND_NFSV4:
3087 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
3088 	}
3089 nfsmout:
3090 	NFSEXITCODE2(error, nd);
3091 	return (error);
3092 }
3093 
3094 /*
3095  * Handle the setable attributes for V4.
3096  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
3097  */
3098 int
nfsv4_sattr(struct nfsrv_descript * nd,vnode_t vp,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,NFSACL_T * aclp,struct thread * p)3099 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
3100     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
3101 {
3102 	u_int32_t *tl;
3103 	int attrsum = 0;
3104 	int i, j;
3105 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
3106 	int moderet, toclient = 0;
3107 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
3108 	uid_t uid;
3109 	gid_t gid;
3110 	u_short mode, mask;		/* Same type as va_mode. */
3111 	struct vattr va;
3112 
3113 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
3114 	if (error)
3115 		goto nfsmout;
3116 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3117 	attrsize = fxdr_unsigned(int, *tl);
3118 
3119 	/*
3120 	 * Loop around getting the setable attributes. If an unsupported
3121 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
3122 	 * Once nd_repstat != 0, do not set the attribute value, but keep
3123 	 * parsing the attribute(s).
3124 	 */
3125 	if (retnotsup) {
3126 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3127 		bitpos = NFSATTRBIT_MAX;
3128 	} else {
3129 		bitpos = 0;
3130 	}
3131 	moderet = 0;
3132 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
3133 	    if (attrsum > attrsize) {
3134 		error = NFSERR_BADXDR;
3135 		goto nfsmout;
3136 	    }
3137 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
3138 		switch (bitpos) {
3139 		case NFSATTRBIT_SIZE:
3140 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
3141 			if (!nd->nd_repstat) {
3142 				if (vp != NULL && vp->v_type != VREG)
3143 					nd->nd_repstat = (vp->v_type == VDIR) ?
3144 					    NFSERR_ISDIR : NFSERR_INVAL;
3145 				else
3146 					nvap->na_size = fxdr_hyper(tl);
3147 			}
3148 			attrsum += NFSX_HYPER;
3149 			break;
3150 		case NFSATTRBIT_ACL:
3151 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
3152 			    &aclsize, p);
3153 			if (error)
3154 				goto nfsmout;
3155 			if (aceerr && !nd->nd_repstat)
3156 				nd->nd_repstat = aceerr;
3157 			attrsum += aclsize;
3158 			break;
3159 		case NFSATTRBIT_ARCHIVE:
3160 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3161 			if (!nd->nd_repstat)
3162 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3163 			attrsum += NFSX_UNSIGNED;
3164 			break;
3165 		case NFSATTRBIT_HIDDEN:
3166 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3167 			if (!nd->nd_repstat)
3168 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3169 			attrsum += NFSX_UNSIGNED;
3170 			break;
3171 		case NFSATTRBIT_MIMETYPE:
3172 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3173 			i = fxdr_unsigned(int, *tl);
3174 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3175 			if (error)
3176 				goto nfsmout;
3177 			if (!nd->nd_repstat)
3178 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3179 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3180 			break;
3181 		case NFSATTRBIT_MODE:
3182 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3183 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3184 			if (!nd->nd_repstat)
3185 				nvap->na_mode = nfstov_mode(*tl);
3186 			attrsum += NFSX_UNSIGNED;
3187 			break;
3188 		case NFSATTRBIT_OWNER:
3189 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3190 			j = fxdr_unsigned(int, *tl);
3191 			if (j < 0) {
3192 				error = NFSERR_BADXDR;
3193 				goto nfsmout;
3194 			}
3195 			if (j > NFSV4_SMALLSTR)
3196 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3197 			else
3198 				cp = namestr;
3199 			error = nfsrv_mtostr(nd, cp, j);
3200 			if (error) {
3201 				if (j > NFSV4_SMALLSTR)
3202 					free(cp, M_NFSSTRING);
3203 				goto nfsmout;
3204 			}
3205 			if (!nd->nd_repstat) {
3206 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3207 				    &uid);
3208 				if (!nd->nd_repstat)
3209 					nvap->na_uid = uid;
3210 			}
3211 			if (j > NFSV4_SMALLSTR)
3212 				free(cp, M_NFSSTRING);
3213 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3214 			break;
3215 		case NFSATTRBIT_OWNERGROUP:
3216 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3217 			j = fxdr_unsigned(int, *tl);
3218 			if (j < 0) {
3219 				error = NFSERR_BADXDR;
3220 				goto nfsmout;
3221 			}
3222 			if (j > NFSV4_SMALLSTR)
3223 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3224 			else
3225 				cp = namestr;
3226 			error = nfsrv_mtostr(nd, cp, j);
3227 			if (error) {
3228 				if (j > NFSV4_SMALLSTR)
3229 					free(cp, M_NFSSTRING);
3230 				goto nfsmout;
3231 			}
3232 			if (!nd->nd_repstat) {
3233 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3234 				    &gid);
3235 				if (!nd->nd_repstat)
3236 					nvap->na_gid = gid;
3237 			}
3238 			if (j > NFSV4_SMALLSTR)
3239 				free(cp, M_NFSSTRING);
3240 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3241 			break;
3242 		case NFSATTRBIT_SYSTEM:
3243 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3244 			if (!nd->nd_repstat)
3245 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3246 			attrsum += NFSX_UNSIGNED;
3247 			break;
3248 		case NFSATTRBIT_TIMEACCESSSET:
3249 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3250 			attrsum += NFSX_UNSIGNED;
3251 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3252 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3253 			    if (!nd->nd_repstat)
3254 				fxdr_nfsv4time(tl, &nvap->na_atime);
3255 			    toclient = 1;
3256 			    attrsum += NFSX_V4TIME;
3257 			} else if (!nd->nd_repstat) {
3258 			    vfs_timestamp(&nvap->na_atime);
3259 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3260 			}
3261 			break;
3262 		case NFSATTRBIT_TIMEBACKUP:
3263 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3264 			if (!nd->nd_repstat)
3265 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3266 			attrsum += NFSX_V4TIME;
3267 			break;
3268 		case NFSATTRBIT_TIMECREATE:
3269 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3270 			if (!nd->nd_repstat)
3271 				fxdr_nfsv4time(tl, &nvap->na_btime);
3272 			attrsum += NFSX_V4TIME;
3273 			break;
3274 		case NFSATTRBIT_TIMEMODIFYSET:
3275 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3276 			attrsum += NFSX_UNSIGNED;
3277 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3278 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3279 			    if (!nd->nd_repstat)
3280 				fxdr_nfsv4time(tl, &nvap->na_mtime);
3281 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3282 			    attrsum += NFSX_V4TIME;
3283 			} else if (!nd->nd_repstat) {
3284 			    vfs_timestamp(&nvap->na_mtime);
3285 			    if (!toclient)
3286 				nvap->na_vaflags |= VA_UTIMES_NULL;
3287 			}
3288 			break;
3289 		case NFSATTRBIT_MODESETMASKED:
3290 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3291 			mode = fxdr_unsigned(u_short, *tl++);
3292 			mask = fxdr_unsigned(u_short, *tl);
3293 			/*
3294 			 * vp == NULL implies an Open/Create operation.
3295 			 * This attribute can only be used for Setattr and
3296 			 * only for NFSv4.1 or higher.
3297 			 * If moderet != 0, a mode attribute has also been
3298 			 * specified and this attribute cannot be done in the
3299 			 * same Setattr operation.
3300 			 */
3301 			if (!nd->nd_repstat) {
3302 				if ((nd->nd_flag & ND_NFSV41) == 0)
3303 					nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3304 				else if ((mode & ~07777) != 0 ||
3305 				    (mask & ~07777) != 0 || vp == NULL)
3306 					nd->nd_repstat = NFSERR_INVAL;
3307 				else if (moderet == 0)
3308 					moderet = VOP_GETATTR(vp, &va,
3309 					    nd->nd_cred);
3310 				if (moderet == 0)
3311 					nvap->na_mode = (mode & mask) |
3312 					    (va.va_mode & ~mask);
3313 				else
3314 					nd->nd_repstat = moderet;
3315 			}
3316 			attrsum += 2 * NFSX_UNSIGNED;
3317 			break;
3318 		case NFSATTRBIT_MODEUMASK:
3319 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3320 			mode = fxdr_unsigned(u_short, *tl++);
3321 			mask = fxdr_unsigned(u_short, *tl);
3322 			/*
3323 			 * If moderet != 0, mode has already been done.
3324 			 * If vp != NULL, this is not a file object creation.
3325 			 */
3326 			if (!nd->nd_repstat) {
3327 				if ((nd->nd_flag & ND_NFSV42) == 0)
3328 					nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3329 				else if ((mask & ~0777) != 0 || vp != NULL ||
3330 				    moderet != 0)
3331 					nd->nd_repstat = NFSERR_INVAL;
3332 				else
3333 					nvap->na_mode = (mode & ~mask);
3334 			}
3335 			attrsum += 2 * NFSX_UNSIGNED;
3336 			break;
3337 		default:
3338 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3339 			/*
3340 			 * set bitpos so we drop out of the loop.
3341 			 */
3342 			bitpos = NFSATTRBIT_MAX;
3343 			break;
3344 		}
3345 	}
3346 
3347 	/*
3348 	 * some clients pad the attrlist, so we need to skip over the
3349 	 * padding.  This also skips over unparsed non-supported attributes.
3350 	 */
3351 	if (attrsum > attrsize) {
3352 		error = NFSERR_BADXDR;
3353 	} else {
3354 		attrsize = NFSM_RNDUP(attrsize);
3355 		if (attrsum < attrsize)
3356 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3357 	}
3358 nfsmout:
3359 	NFSEXITCODE2(error, nd);
3360 	return (error);
3361 }
3362 
3363 /*
3364  * Check/setup export credentials.
3365  */
3366 int
nfsd_excred(struct nfsrv_descript * nd,struct nfsexstuff * exp,struct ucred * credanon,bool testsec)3367 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3368     struct ucred *credanon, bool testsec)
3369 {
3370 	int error;
3371 
3372 	/*
3373 	 * Check/setup credentials.
3374 	 */
3375 	if (nd->nd_flag & ND_GSS)
3376 		exp->nes_exflag &= ~MNT_EXPORTANON;
3377 
3378 	/*
3379 	 * Check to see if the operation is allowed for this security flavor.
3380 	 */
3381 	error = 0;
3382 	if (testsec) {
3383 		error = nfsvno_testexp(nd, exp);
3384 		if (error != 0)
3385 			goto out;
3386 	}
3387 
3388 	/*
3389 	 * Check to see if the file system is exported V4 only.
3390 	 */
3391 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3392 		error = NFSERR_PROGNOTV4;
3393 		goto out;
3394 	}
3395 
3396 	/*
3397 	 * Now, map the user credentials.
3398 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3399 	 *  Fsinfo RPC. If set for anything else, this code might need
3400 	 *  to change.)
3401 	 */
3402 	if (NFSVNO_EXPORTED(exp)) {
3403 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3404 		     NFSVNO_EXPORTANON(exp) ||
3405 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3406 			nd->nd_cred->cr_uid = credanon->cr_uid;
3407 			/*
3408 			 * 'credanon' is already a 'struct ucred' that was built
3409 			 * internally with calls to crsetgroups_fallback(), so
3410 			 * we don't need a fallback here.
3411 			 */
3412 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3413 			    credanon->cr_groups);
3414 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3415 			/*
3416 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3417 			 * if there is a replacement credential with a group
3418 			 * list set up by "nfsuserd -manage-gids".
3419 			 * If there is no replacement, nfsrv_getgrpscred()
3420 			 * simply returns its argument.
3421 			 */
3422 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3423 		}
3424 	}
3425 
3426 out:
3427 	NFSEXITCODE2(error, nd);
3428 	return (error);
3429 }
3430 
3431 /*
3432  * Check exports.
3433  */
3434 int
nfsvno_checkexp(struct mount * mp,struct sockaddr * nam,struct nfsexstuff * exp,struct ucred ** credp)3435 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3436     struct ucred **credp)
3437 {
3438 	int error;
3439 
3440 	error = 0;
3441 	*credp = NULL;
3442 	MNT_ILOCK(mp);
3443 	if (mp->mnt_exjail == NULL ||
3444 	    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3445 		error = EACCES;
3446 	MNT_IUNLOCK(mp);
3447 	if (error == 0)
3448 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3449 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3450 	if (error) {
3451 		if (NFSD_VNET(nfs_rootfhset)) {
3452 			exp->nes_exflag = 0;
3453 			exp->nes_numsecflavor = 0;
3454 			error = 0;
3455 		}
3456 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3457 	    MAXSECFLAVORS) {
3458 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3459 		exp->nes_numsecflavor = 0;
3460 		error = EACCES;
3461 	}
3462 	NFSEXITCODE(error);
3463 	return (error);
3464 }
3465 
3466 /*
3467  * Get a vnode for a file handle and export stuff.
3468  */
3469 int
nfsvno_fhtovp(struct mount * mp,fhandle_t * fhp,struct sockaddr * nam,int lktype,struct vnode ** vpp,struct nfsexstuff * exp,struct ucred ** credp)3470 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3471     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3472     struct ucred **credp)
3473 {
3474 	int error;
3475 
3476 	*credp = NULL;
3477 	exp->nes_numsecflavor = 0;
3478 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3479 	if (error != 0)
3480 		/* Make sure the server replies ESTALE to the client. */
3481 		error = ESTALE;
3482 	if (nam && !error) {
3483 		MNT_ILOCK(mp);
3484 		if (mp->mnt_exjail == NULL ||
3485 		    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3486 			error = EACCES;
3487 		MNT_IUNLOCK(mp);
3488 		if (error == 0)
3489 			error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3490 			    &exp->nes_numsecflavor, exp->nes_secflavors);
3491 		if (error) {
3492 			if (NFSD_VNET(nfs_rootfhset)) {
3493 				exp->nes_exflag = 0;
3494 				exp->nes_numsecflavor = 0;
3495 				error = 0;
3496 			} else {
3497 				vput(*vpp);
3498 			}
3499 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3500 		    MAXSECFLAVORS) {
3501 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3502 			exp->nes_numsecflavor = 0;
3503 			error = EACCES;
3504 			vput(*vpp);
3505 		}
3506 	}
3507 	NFSEXITCODE(error);
3508 	return (error);
3509 }
3510 
3511 /*
3512  * nfsd_fhtovp() - convert a fh to a vnode ptr
3513  * 	- look up fsid in mount list (if not found ret error)
3514  *	- get vp and export rights by calling nfsvno_fhtovp()
3515  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3516  *	  for AUTH_SYS
3517  *	- if mpp != NULL, return the mount point so that it can
3518  *	  be used for vn_finished_write() by the caller
3519  */
3520 void
nfsd_fhtovp(struct nfsrv_descript * nd,struct nfsrvfh * nfp,int lktype,struct vnode ** vpp,struct nfsexstuff * exp,struct mount ** mpp,int startwrite,int nextop)3521 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3522     struct vnode **vpp, struct nfsexstuff *exp,
3523     struct mount **mpp, int startwrite, int nextop)
3524 {
3525 	struct mount *mp, *mpw;
3526 	struct ucred *credanon;
3527 	fhandle_t *fhp;
3528 	int error;
3529 
3530 	if (mpp != NULL)
3531 		*mpp = NULL;
3532 	*vpp = NULL;
3533 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3534 	mp = vfs_busyfs(&fhp->fh_fsid);
3535 	if (mp == NULL) {
3536 		nd->nd_repstat = ESTALE;
3537 		goto out;
3538 	}
3539 
3540 	if (startwrite) {
3541 		mpw = mp;
3542 		error = vn_start_write(NULL, &mpw, V_WAIT);
3543 		if (error != 0) {
3544 			mpw = NULL;
3545 			vfs_unbusy(mp);
3546 			nd->nd_repstat = ESTALE;
3547 			goto out;
3548 		}
3549 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3550 			lktype = LK_EXCLUSIVE;
3551 	} else
3552 		mpw = NULL;
3553 
3554 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3555 	    &credanon);
3556 	vfs_unbusy(mp);
3557 
3558 	if (nd->nd_repstat == 0 &&
3559 	    nfp->nfsrvfh_len >= NFSX_MYFH + NFSX_V4NAMEDDIRFH &&
3560 	    nfp->nfsrvfh_len <= NFSX_MYFH + NFSX_V4NAMEDATTRFH) {
3561 		if (nfp->nfsrvfh_len == NFSX_MYFH + NFSX_V4NAMEDDIRFH)
3562 			vn_irflag_set_cond(*vpp, VIRF_NAMEDDIR);
3563 		else
3564 			vn_irflag_set_cond(*vpp, VIRF_NAMEDATTR);
3565 	}
3566 
3567 	/*
3568 	 * For NFSv4 without a pseudo root fs, unexported file handles
3569 	 * can be returned, so that Lookup works everywhere.
3570 	 */
3571 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3572 	    !(nd->nd_flag & ND_NFSV4)) {
3573 		vput(*vpp);
3574 		*vpp = NULL;
3575 		nd->nd_repstat = EACCES;
3576 	}
3577 
3578 	/*
3579 	 * Personally, I've never seen any point in requiring a
3580 	 * reserved port#, since only in the rare case where the
3581 	 * clients are all boxes with secure system privileges,
3582 	 * does it provide any enhanced security, but... some people
3583 	 * believe it to be useful and keep putting this code back in.
3584 	 * (There is also some "security checker" out there that
3585 	 *  complains if the nfs server doesn't enforce this.)
3586 	 * However, note the following:
3587 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3588 	 *	required.
3589 	 * RFC2623 recommends that, if a reserved port# is checked for,
3590 	 *	that there be a way to turn that off--> ifdef'd.
3591 	 */
3592 #ifdef NFS_REQRSVPORT
3593 	if (!nd->nd_repstat) {
3594 		struct sockaddr_in *saddr;
3595 		struct sockaddr_in6 *saddr6;
3596 
3597 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3598 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3599 		if (!(nd->nd_flag & ND_NFSV4) &&
3600 		    ((saddr->sin_family == AF_INET &&
3601 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3602 		     (saddr6->sin6_family == AF_INET6 &&
3603 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3604 			vput(*vpp);
3605 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3606 		}
3607 	}
3608 #endif	/* NFS_REQRSVPORT */
3609 
3610 	/*
3611 	 * Check/setup credentials.
3612 	 */
3613 	if (!nd->nd_repstat) {
3614 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3615 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3616 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3617 		if (nd->nd_repstat)
3618 			vput(*vpp);
3619 	}
3620 	if (credanon != NULL)
3621 		crfree(credanon);
3622 	if (nd->nd_repstat) {
3623 		vn_finished_write(mpw);
3624 		*vpp = NULL;
3625 	} else if (mpp != NULL) {
3626 		*mpp = mpw;
3627 	}
3628 
3629 out:
3630 	NFSEXITCODE2(0, nd);
3631 }
3632 
3633 /*
3634  * glue for fp.
3635  */
3636 static int
fp_getfvp(struct thread * p,int fd,struct file ** fpp,struct vnode ** vpp)3637 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3638 {
3639 	struct filedesc *fdp;
3640 	struct file *fp;
3641 	int error = 0;
3642 
3643 	fdp = p->td_proc->p_fd;
3644 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3645 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3646 		error = EBADF;
3647 		goto out;
3648 	}
3649 	*fpp = fp;
3650 
3651 out:
3652 	NFSEXITCODE(error);
3653 	return (error);
3654 }
3655 
3656 /*
3657  * Called from nfssvc() to update the exports list. Just call
3658  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3659  * in the mount list.
3660  */
3661 int
nfsrv_v4rootexport(void * argp,struct ucred * cred,struct thread * p)3662 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3663 {
3664 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3665 	int error = 0;
3666 	struct nameidata nd;
3667 	fhandle_t fh;
3668 
3669 	error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false);
3670 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3671 		NFSD_VNET(nfs_rootfhset) = 0;
3672 	else if (error == 0) {
3673 		if (nfsexargp->fspec == NULL) {
3674 			error = EPERM;
3675 			goto out;
3676 		}
3677 		/*
3678 		 * If fspec != NULL, this is the v4root path.
3679 		 */
3680 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3681 		if ((error = namei(&nd)) != 0)
3682 			goto out;
3683 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3684 		vrele(nd.ni_vp);
3685 		if (!error) {
3686 			NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH;
3687 			NFSBCOPY((caddr_t)&fh,
3688 			    NFSD_VNET(nfs_rootfh).nfsrvfh_data,
3689 			    sizeof (fhandle_t));
3690 			NFSD_VNET(nfs_rootfhset) = 1;
3691 		}
3692 	}
3693 
3694 out:
3695 	NFSEXITCODE(error);
3696 	return (error);
3697 }
3698 
3699 /*
3700  * This function needs to test to see if the system is near its limit
3701  * for memory allocation via malloc() or mget() and return True iff
3702  * either of these resources are near their limit.
3703  * XXX (For now, this is just a stub.)
3704  */
3705 int nfsrv_testmalloclimit = 0;
3706 int
nfsrv_mallocmget_limit(void)3707 nfsrv_mallocmget_limit(void)
3708 {
3709 	static int printmesg = 0;
3710 	static int testval = 1;
3711 
3712 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3713 		if ((printmesg++ % 100) == 0)
3714 			printf("nfsd: malloc/mget near limit\n");
3715 		return (1);
3716 	}
3717 	return (0);
3718 }
3719 
3720 /*
3721  * BSD specific initialization of a mount point.
3722  */
3723 void
nfsd_mntinit(void)3724 nfsd_mntinit(void)
3725 {
3726 
3727 	NFSD_LOCK();
3728 	if (NFSD_VNET(nfsrv_mntinited)) {
3729 		NFSD_UNLOCK();
3730 		return;
3731 	}
3732 	NFSD_VNET(nfsrv_mntinited) = true;
3733 	nfsrvd_init(0);
3734 	NFSD_UNLOCK();
3735 
3736 	NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP,
3737 	    M_WAITOK | M_ZERO);
3738 	NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3739 	mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3740 	lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0);
3741 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist);
3742 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist);
3743 	NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
3744 	TAILQ_INIT(&NFSD_VNET(nfsv4root_opt));
3745 	TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt));
3746 	NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt);
3747 	NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt);
3748 	NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0;
3749 	NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0;
3750 	callout_init(&NFSD_VNET(nfsd_callout), 1);
3751 
3752 	nfsrvd_initcache();
3753 	nfsd_init();
3754 }
3755 
3756 static void
nfsd_timer(void * arg)3757 nfsd_timer(void *arg)
3758 {
3759 	struct vnet *vnetp;
3760 
3761 	vnetp = (struct vnet *)arg;
3762 	NFSD_CURVNET_SET_QUIET(vnetp);
3763 	nfsrv_servertimer(vnetp);
3764 	callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer,
3765 	    arg, 0);
3766 	NFSD_CURVNET_RESTORE();
3767 }
3768 
3769 /*
3770  * Get a vnode for a file handle, without checking exports, etc.
3771  */
3772 struct vnode *
nfsvno_getvp(fhandle_t * fhp)3773 nfsvno_getvp(fhandle_t *fhp)
3774 {
3775 	struct mount *mp;
3776 	struct vnode *vp;
3777 	int error;
3778 
3779 	mp = vfs_busyfs(&fhp->fh_fsid);
3780 	if (mp == NULL)
3781 		return (NULL);
3782 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3783 	vfs_unbusy(mp);
3784 	if (error)
3785 		return (NULL);
3786 	return (vp);
3787 }
3788 
3789 /*
3790  * Do a local VOP_ADVLOCK().
3791  */
3792 int
nfsvno_advlock(struct vnode * vp,int ftype,u_int64_t first,u_int64_t end,struct thread * td)3793 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3794     u_int64_t end, struct thread *td)
3795 {
3796 	int error = 0;
3797 	struct flock fl;
3798 	u_int64_t tlen;
3799 
3800 	if (nfsrv_dolocallocks == 0)
3801 		goto out;
3802 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3803 
3804 	fl.l_whence = SEEK_SET;
3805 	fl.l_type = ftype;
3806 	fl.l_start = (off_t)first;
3807 	if (end == NFS64BITSSET) {
3808 		fl.l_len = 0;
3809 	} else {
3810 		tlen = end - first;
3811 		fl.l_len = (off_t)tlen;
3812 	}
3813 	/*
3814 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3815 	 * values for all calls, so that all locks will be held by the
3816 	 * nfsd server. (The nfsd server handles conflicts between the
3817 	 * various clients.)
3818 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3819 	 * bytes, so it can't be put in l_sysid.
3820 	 */
3821 	if (nfsv4_sysid == 0)
3822 		nfsv4_sysid = nlm_acquire_next_sysid();
3823 	fl.l_pid = (pid_t)0;
3824 	fl.l_sysid = (int)nfsv4_sysid;
3825 
3826 	if (ftype == F_UNLCK)
3827 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3828 		    (F_POSIX | F_REMOTE));
3829 	else
3830 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3831 		    (F_POSIX | F_REMOTE));
3832 
3833 out:
3834 	NFSEXITCODE(error);
3835 	return (error);
3836 }
3837 
3838 /*
3839  * Check the nfsv4 root exports.
3840  */
3841 int
nfsvno_v4rootexport(struct nfsrv_descript * nd)3842 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3843 {
3844 	struct ucred *credanon;
3845 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3846 	uint64_t exflags;
3847 
3848 	error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags,
3849 	    &credanon, &numsecflavor, secflavors);
3850 	if (error) {
3851 		error = NFSERR_PROGUNAVAIL;
3852 		goto out;
3853 	}
3854 	if (credanon != NULL)
3855 		crfree(credanon);
3856 	for (i = 0; i < numsecflavor; i++) {
3857 		if (secflavors[i] == AUTH_SYS)
3858 			nd->nd_flag |= ND_EXAUTHSYS;
3859 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3860 			nd->nd_flag |= ND_EXGSS;
3861 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3862 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3863 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3864 			nd->nd_flag |= ND_EXGSSPRIVACY;
3865 	}
3866 
3867 	/* And set ND_EXxx flags for TLS. */
3868 	if ((exflags & MNT_EXTLS) != 0) {
3869 		nd->nd_flag |= ND_EXTLS;
3870 		if ((exflags & MNT_EXTLSCERT) != 0)
3871 			nd->nd_flag |= ND_EXTLSCERT;
3872 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3873 			nd->nd_flag |= ND_EXTLSCERTUSER;
3874 	}
3875 
3876 out:
3877 	NFSEXITCODE(error);
3878 	return (error);
3879 }
3880 
3881 /*
3882  * Nfs server pseudo system call for the nfsd's
3883  */
3884 /*
3885  * MPSAFE
3886  */
3887 static int
nfssvc_nfsd(struct thread * td,struct nfssvc_args * uap)3888 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3889 {
3890 	struct file *fp;
3891 	struct nfsd_addsock_args sockarg;
3892 	struct nfsd_nfsd_args nfsdarg;
3893 	struct nfsd_nfsd_oargs onfsdarg;
3894 	struct nfsd_pnfsd_args pnfsdarg;
3895 	struct vnode *vp, *nvp, *curdvp;
3896 	struct pnfsdsfile *pf;
3897 	struct nfsdevice *ds, *fds;
3898 	cap_rights_t rights;
3899 	int buflen, error, ret;
3900 	char *buf, *cp, *cp2, *cp3;
3901 	char fname[PNFS_FILENAME_LEN + 1];
3902 
3903 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
3904 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3905 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3906 		if (error)
3907 			goto out;
3908 		/*
3909 		 * Since we don't know what rights might be required,
3910 		 * pretend that we need them all. It is better to be too
3911 		 * careful than too reckless.
3912 		 */
3913 		error = fget(td, sockarg.sock,
3914 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3915 		if (error != 0)
3916 			goto out;
3917 		if (fp->f_type != DTYPE_SOCKET) {
3918 			fdrop(fp, td);
3919 			error = EPERM;
3920 			goto out;
3921 		}
3922 		error = nfsrvd_addsock(fp);
3923 		fdrop(fp, td);
3924 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3925 		if (uap->argp == NULL) {
3926 			error = EINVAL;
3927 			goto out;
3928 		}
3929 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3930 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3931 			if (error == 0) {
3932 				nfsdarg.principal = onfsdarg.principal;
3933 				nfsdarg.minthreads = onfsdarg.minthreads;
3934 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3935 				nfsdarg.version = 1;
3936 				nfsdarg.addr = NULL;
3937 				nfsdarg.addrlen = 0;
3938 				nfsdarg.dnshost = NULL;
3939 				nfsdarg.dnshostlen = 0;
3940 				nfsdarg.dspath = NULL;
3941 				nfsdarg.dspathlen = 0;
3942 				nfsdarg.mdspath = NULL;
3943 				nfsdarg.mdspathlen = 0;
3944 				nfsdarg.mirrorcnt = 1;
3945 			}
3946 		} else
3947 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3948 		if (error)
3949 			goto out;
3950 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3951 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3952 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3953 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3954 		    nfsdarg.mirrorcnt >= 1 &&
3955 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3956 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3957 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3958 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3959 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3960 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3961 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3962 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3963 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3964 			if (error != 0) {
3965 				free(cp, M_TEMP);
3966 				goto out;
3967 			}
3968 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3969 			nfsdarg.addr = cp;
3970 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3971 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3972 			if (error != 0) {
3973 				free(nfsdarg.addr, M_TEMP);
3974 				free(cp, M_TEMP);
3975 				goto out;
3976 			}
3977 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3978 			nfsdarg.dnshost = cp;
3979 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3980 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3981 			if (error != 0) {
3982 				free(nfsdarg.addr, M_TEMP);
3983 				free(nfsdarg.dnshost, M_TEMP);
3984 				free(cp, M_TEMP);
3985 				goto out;
3986 			}
3987 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3988 			nfsdarg.dspath = cp;
3989 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3990 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3991 			if (error != 0) {
3992 				free(nfsdarg.addr, M_TEMP);
3993 				free(nfsdarg.dnshost, M_TEMP);
3994 				free(nfsdarg.dspath, M_TEMP);
3995 				free(cp, M_TEMP);
3996 				goto out;
3997 			}
3998 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3999 			nfsdarg.mdspath = cp;
4000 		} else {
4001 			nfsdarg.addr = NULL;
4002 			nfsdarg.addrlen = 0;
4003 			nfsdarg.dnshost = NULL;
4004 			nfsdarg.dnshostlen = 0;
4005 			nfsdarg.dspath = NULL;
4006 			nfsdarg.dspathlen = 0;
4007 			nfsdarg.mdspath = NULL;
4008 			nfsdarg.mdspathlen = 0;
4009 			nfsdarg.mirrorcnt = 1;
4010 		}
4011 		nfsd_timer(NFSD_TD_TO_VNET(td));
4012 		error = nfsrvd_nfsd(td, &nfsdarg);
4013 		callout_drain(&NFSD_VNET(nfsd_callout));
4014 		free(nfsdarg.addr, M_TEMP);
4015 		free(nfsdarg.dnshost, M_TEMP);
4016 		free(nfsdarg.dspath, M_TEMP);
4017 		free(nfsdarg.mdspath, M_TEMP);
4018 	} else if (uap->flag & NFSSVC_PNFSDS) {
4019 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
4020 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
4021 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
4022 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
4023 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
4024 			    NULL);
4025 			if (error == 0)
4026 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
4027 			free(cp, M_TEMP);
4028 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
4029 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
4030 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
4031 			buf = malloc(buflen, M_TEMP, M_WAITOK);
4032 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
4033 			    NULL);
4034 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
4035 			if (error == 0 && pnfsdarg.dspath != NULL) {
4036 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
4037 				error = copyinstr(pnfsdarg.dspath, cp2,
4038 				    PATH_MAX + 1, NULL);
4039 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
4040 				    error);
4041 			} else
4042 				cp2 = NULL;
4043 			if (error == 0 && pnfsdarg.curdspath != NULL) {
4044 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
4045 				error = copyinstr(pnfsdarg.curdspath, cp3,
4046 				    PATH_MAX + 1, NULL);
4047 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
4048 				    error);
4049 			} else
4050 				cp3 = NULL;
4051 			curdvp = NULL;
4052 			fds = NULL;
4053 			if (error == 0)
4054 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
4055 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
4056 				    &fds);
4057 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
4058 			if (error == 0) {
4059 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
4060 					printf("copymr: dsdir out of range\n");
4061 					pf->dsf_dir = 0;
4062 				}
4063 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
4064 				error = nfsrv_copymr(vp, nvp,
4065 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
4066 				    (struct pnfsdsfile *)buf,
4067 				    buflen / sizeof(*pf), td->td_ucred, td);
4068 				vput(vp);
4069 				vput(nvp);
4070 				if (fds != NULL && error == 0) {
4071 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
4072 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
4073 					if (ret == 0) {
4074 						nfsrv_dsremove(curdvp, fname,
4075 						    td->td_ucred, td);
4076 						NFSVOPUNLOCK(curdvp);
4077 					}
4078 				}
4079 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
4080 			}
4081 			free(cp, M_TEMP);
4082 			free(cp2, M_TEMP);
4083 			free(cp3, M_TEMP);
4084 			free(buf, M_TEMP);
4085 		}
4086 	} else {
4087 		error = nfssvc_srvcall(td, uap, td->td_ucred);
4088 	}
4089 
4090 out:
4091 	NFSD_CURVNET_RESTORE();
4092 	NFSEXITCODE(error);
4093 	return (error);
4094 }
4095 
4096 static int
nfssvc_srvcall(struct thread * p,struct nfssvc_args * uap,struct ucred * cred)4097 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
4098 {
4099 	struct nfsex_args export;
4100 	struct nfsex_oldargs oexp;
4101 	struct file *fp = NULL;
4102 	int stablefd, i, len;
4103 	struct nfsd_clid adminrevoke;
4104 	struct nfsd_dumplist dumplist;
4105 	struct nfsd_dumpclients *dumpclients;
4106 	struct nfsd_dumplocklist dumplocklist;
4107 	struct nfsd_dumplocks *dumplocks;
4108 	struct nameidata nd;
4109 	vnode_t vp;
4110 	int error = EINVAL, igotlock;
4111 	struct proc *procp;
4112 	gid_t *grps;
4113 
4114 	if (uap->flag & NFSSVC_PUBLICFH) {
4115 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
4116 		    sizeof (fhandle_t));
4117 		error = copyin(uap->argp,
4118 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
4119 		if (!error)
4120 			nfs_pubfhset = 1;
4121 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
4122 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
4123 		error = copyin(uap->argp,(caddr_t)&export,
4124 		    sizeof (struct nfsex_args));
4125 		if (!error) {
4126 			grps = NULL;
4127 			if (export.export.ex_ngroups > NGROUPS_MAX ||
4128 			    export.export.ex_ngroups < 0)
4129 				error = EINVAL;
4130 			else if (export.export.ex_ngroups > 0) {
4131 				grps = malloc(export.export.ex_ngroups *
4132 				    sizeof(gid_t), M_TEMP, M_WAITOK);
4133 				error = copyin(export.export.ex_groups, grps,
4134 				    export.export.ex_ngroups * sizeof(gid_t));
4135 				export.export.ex_groups = grps;
4136 			} else
4137 				export.export.ex_groups = NULL;
4138 			if (!error)
4139 				error = nfsrv_v4rootexport(&export, cred, p);
4140 			free(grps, M_TEMP);
4141 		}
4142 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
4143 	    NFSSVC_V4ROOTEXPORT) {
4144 		error = copyin(uap->argp,(caddr_t)&oexp,
4145 		    sizeof (struct nfsex_oldargs));
4146 		if (!error) {
4147 			memset(&export.export, 0, sizeof(export.export));
4148 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
4149 			export.export.ex_root = oexp.export.ex_root;
4150 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
4151 			export.export.ex_ngroups =
4152 			    oexp.export.ex_anon.cr_ngroups;
4153 			export.export.ex_groups = NULL;
4154 			if (export.export.ex_ngroups > XU_NGROUPS ||
4155 			    export.export.ex_ngroups < 0)
4156 				error = EINVAL;
4157 			else if (export.export.ex_ngroups > 0) {
4158 				export.export.ex_groups = malloc(
4159 				    export.export.ex_ngroups * sizeof(gid_t),
4160 				    M_TEMP, M_WAITOK);
4161 				for (i = 0; i < export.export.ex_ngroups; i++)
4162 					export.export.ex_groups[i] =
4163 					    oexp.export.ex_anon.cr_groups[i];
4164 			}
4165 			export.export.ex_addr = oexp.export.ex_addr;
4166 			export.export.ex_addrlen = oexp.export.ex_addrlen;
4167 			export.export.ex_mask = oexp.export.ex_mask;
4168 			export.export.ex_masklen = oexp.export.ex_masklen;
4169 			export.export.ex_indexfile = oexp.export.ex_indexfile;
4170 			export.export.ex_numsecflavors =
4171 			    oexp.export.ex_numsecflavors;
4172 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
4173 			    export.export.ex_numsecflavors < 0)
4174 				error = EINVAL;
4175 			else {
4176 				for (i = 0; i < export.export.ex_numsecflavors;
4177 				    i++)
4178 					export.export.ex_secflavors[i] =
4179 					    oexp.export.ex_secflavors[i];
4180 			}
4181 			export.fspec = oexp.fspec;
4182 			if (error == 0)
4183 				error = nfsrv_v4rootexport(&export, cred, p);
4184 			free(export.export.ex_groups, M_TEMP);
4185 		}
4186 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
4187 		nfs_pubfhset = 0;
4188 		error = 0;
4189 	} else if (uap->flag & NFSSVC_STABLERESTART) {
4190 		error = copyin(uap->argp, (caddr_t)&stablefd,
4191 		    sizeof (int));
4192 		if (!error)
4193 			error = fp_getfvp(p, stablefd, &fp, &vp);
4194 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
4195 			error = EBADF;
4196 		if (!error && NFSD_VNET(nfsrv_numnfsd) != 0)
4197 			error = ENXIO;
4198 		if (!error) {
4199 			NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp;
4200 			nfsrv_setupstable(p);
4201 		}
4202 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
4203 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
4204 		    sizeof (struct nfsd_clid));
4205 		if (!error)
4206 			error = nfsrv_adminrevoke(&adminrevoke, p);
4207 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
4208 		error = copyin(uap->argp, (caddr_t)&dumplist,
4209 		    sizeof (struct nfsd_dumplist));
4210 		if (!error && (dumplist.ndl_size < 1 ||
4211 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
4212 			error = EPERM;
4213 		if (!error) {
4214 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
4215 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4216 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
4217 		    error = copyout(dumpclients, dumplist.ndl_list, len);
4218 		    free(dumpclients, M_TEMP);
4219 		}
4220 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
4221 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
4222 		    sizeof (struct nfsd_dumplocklist));
4223 		if (!error && (dumplocklist.ndllck_size < 1 ||
4224 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
4225 			error = EPERM;
4226 		if (!error)
4227 			error = nfsrv_lookupfilename(&nd,
4228 				dumplocklist.ndllck_fname, p);
4229 		if (!error) {
4230 			len = sizeof (struct nfsd_dumplocks) *
4231 				dumplocklist.ndllck_size;
4232 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4233 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4234 			    dumplocklist.ndllck_size, p);
4235 			vput(nd.ni_vp);
4236 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4237 			    len);
4238 			free(dumplocks, M_TEMP);
4239 		}
4240 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4241 		procp = p->td_proc;
4242 		PROC_LOCK(procp);
4243 		nfsd_master_pid = procp->p_pid;
4244 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4245 		nfsd_master_start = procp->p_stats->p_start;
4246 		NFSD_VNET(nfsd_master_proc) = procp;
4247 		PROC_UNLOCK(procp);
4248 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4249 		NFSLOCKV4ROOTMUTEX();
4250 		if (!NFSD_VNET(nfsrv_suspend_nfsd)) {
4251 			/* Lock out all nfsd threads */
4252 			do {
4253 				igotlock = nfsv4_lock(
4254 				    &NFSD_VNET(nfsd_suspend_lock), 1, NULL,
4255 				    NFSV4ROOTLOCKMUTEXPTR, NULL);
4256 			} while (igotlock == 0 &&
4257 			    !NFSD_VNET(nfsrv_suspend_nfsd));
4258 			NFSD_VNET(nfsrv_suspend_nfsd) = true;
4259 		}
4260 		NFSUNLOCKV4ROOTMUTEX();
4261 		error = 0;
4262 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4263 		NFSLOCKV4ROOTMUTEX();
4264 		if (NFSD_VNET(nfsrv_suspend_nfsd)) {
4265 			nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0);
4266 			NFSD_VNET(nfsrv_suspend_nfsd) = false;
4267 		}
4268 		NFSUNLOCKV4ROOTMUTEX();
4269 		error = 0;
4270 	}
4271 
4272 	NFSEXITCODE(error);
4273 	return (error);
4274 }
4275 
4276 /*
4277  * Check exports.
4278  * Returns 0 if ok, 1 otherwise.
4279  */
4280 int
nfsvno_testexp(struct nfsrv_descript * nd,struct nfsexstuff * exp)4281 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4282 {
4283 	int i;
4284 
4285 	if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4286 	    (NFSVNO_EXTLSCERT(exp) &&
4287 	     (nd->nd_flag & ND_TLSCERT) == 0) ||
4288 	    (NFSVNO_EXTLSCERTUSER(exp) &&
4289 	     (nd->nd_flag & ND_TLSCERTUSER) == 0)) {
4290 		if ((nd->nd_flag & ND_NFSV4) != 0)
4291 			return (NFSERR_WRONGSEC);
4292 #ifdef notnow
4293 		/* There is currently no auth_stat for this. */
4294 		else if ((nd->nd_flag & ND_TLS) == 0)
4295 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4296 		else
4297 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4298 #endif
4299 		else
4300 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4301 	}
4302 
4303 	/*
4304 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4305 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4306 	 */
4307 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4308 		return (0);
4309 
4310 	/*
4311 	 * This seems odd, but allow the case where the security flavor
4312 	 * list is empty. This happens when NFSv4 is traversing non-exported
4313 	 * file systems. Exported file systems should always have a non-empty
4314 	 * security flavor list.
4315 	 */
4316 	if (exp->nes_numsecflavor == 0)
4317 		return (0);
4318 
4319 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4320 		/*
4321 		 * The tests for privacy and integrity must be first,
4322 		 * since ND_GSS is set for everything but AUTH_SYS.
4323 		 */
4324 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4325 		    (nd->nd_flag & ND_GSSPRIVACY))
4326 			return (0);
4327 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4328 		    (nd->nd_flag & ND_GSSINTEGRITY))
4329 			return (0);
4330 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4331 		    (nd->nd_flag & ND_GSS))
4332 			return (0);
4333 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4334 		    (nd->nd_flag & ND_GSS) == 0)
4335 			return (0);
4336 	}
4337 	if ((nd->nd_flag & ND_NFSV4) != 0)
4338 		return (NFSERR_WRONGSEC);
4339 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4340 }
4341 
4342 /*
4343  * Calculate a hash value for the fid in a file handle.
4344  */
4345 uint32_t
nfsrv_hashfh(fhandle_t * fhp)4346 nfsrv_hashfh(fhandle_t *fhp)
4347 {
4348 	uint32_t hashval;
4349 
4350 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4351 	return (hashval);
4352 }
4353 
4354 /*
4355  * Calculate a hash value for the sessionid.
4356  */
4357 uint32_t
nfsrv_hashsessionid(uint8_t * sessionid)4358 nfsrv_hashsessionid(uint8_t *sessionid)
4359 {
4360 	uint32_t hashval;
4361 
4362 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4363 	return (hashval);
4364 }
4365 
4366 /*
4367  * Signal the userland master nfsd to backup the stable restart file.
4368  */
4369 void
nfsrv_backupstable(void)4370 nfsrv_backupstable(void)
4371 {
4372 	struct proc *procp;
4373 
4374 	if (NFSD_VNET(nfsd_master_proc) != NULL) {
4375 		procp = pfind(nfsd_master_pid);
4376 		/* Try to make sure it is the correct process. */
4377 		if (procp == NFSD_VNET(nfsd_master_proc) &&
4378 		    procp->p_stats->p_start.tv_sec ==
4379 		    nfsd_master_start.tv_sec &&
4380 		    procp->p_stats->p_start.tv_usec ==
4381 		    nfsd_master_start.tv_usec &&
4382 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4383 			kern_psignal(procp, SIGUSR2);
4384 		else
4385 			NFSD_VNET(nfsd_master_proc) = NULL;
4386 
4387 		if (procp != NULL)
4388 			PROC_UNLOCK(procp);
4389 	}
4390 }
4391 
4392 /*
4393  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4394  * The arguments are in a structure, so that they can be passed through
4395  * taskqueue for a kernel process to execute this function.
4396  */
4397 struct nfsrvdscreate {
4398 	int			done;
4399 	int			inprog;
4400 	struct task		tsk;
4401 	struct ucred		*tcred;
4402 	struct vnode		*dvp;
4403 	NFSPROC_T		*p;
4404 	struct pnfsdsfile	*pf;
4405 	int			err;
4406 	fhandle_t		fh;
4407 	struct vattr		va;
4408 	struct vattr		createva;
4409 };
4410 
4411 int
nfsrv_dscreate(struct vnode * dvp,struct vattr * vap,struct vattr * nvap,fhandle_t * fhp,struct pnfsdsfile * pf,struct pnfsdsattr * dsa,char * fnamep,struct ucred * tcred,NFSPROC_T * p,struct vnode ** nvpp)4412 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4413     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4414     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4415 {
4416 	struct vnode *nvp;
4417 	struct nameidata named;
4418 	struct vattr va;
4419 	char *bufp;
4420 	u_long *hashp;
4421 	struct nfsnode *np;
4422 	struct nfsmount *nmp;
4423 	int error;
4424 
4425 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4426 	    LOCKPARENT | LOCKLEAF | NOCACHE);
4427 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4428 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4429 	named.ni_cnd.cn_nameptr = bufp;
4430 	if (fnamep != NULL) {
4431 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4432 		named.ni_cnd.cn_namelen = strlen(bufp);
4433 	} else
4434 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4435 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4436 
4437 	/* Create the date file in the DS mount. */
4438 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4439 	if (error == 0) {
4440 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4441 		vref(dvp);
4442 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4443 		if (error == 0) {
4444 			/* Set the ownership of the file. */
4445 			error = VOP_SETATTR(nvp, nvap, tcred);
4446 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4447 			    " setattr-uid=%d\n", error);
4448 			if (error != 0)
4449 				vput(nvp);
4450 		}
4451 		if (error != 0)
4452 			printf("pNFS: pnfscreate failed=%d\n", error);
4453 	} else
4454 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4455 	if (error == 0) {
4456 		np = VTONFS(nvp);
4457 		nmp = VFSTONFS(nvp->v_mount);
4458 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4459 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4460 		    struct sockaddr_in6) ||
4461 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4462 			printf("Bad DS file: fstype=%s salen=%d"
4463 			    " fhlen=%d\n",
4464 			    nvp->v_mount->mnt_vfc->vfc_name,
4465 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4466 			error = ENOENT;
4467 		}
4468 
4469 		/* Set extattrs for the DS on the MDS file. */
4470 		if (error == 0) {
4471 			if (dsa != NULL) {
4472 				error = VOP_GETATTR(nvp, &va, tcred);
4473 				if (error == 0) {
4474 					dsa->dsa_filerev = va.va_filerev;
4475 					dsa->dsa_size = va.va_size;
4476 					dsa->dsa_atime = va.va_atime;
4477 					dsa->dsa_mtime = va.va_mtime;
4478 					dsa->dsa_bytes = va.va_bytes;
4479 				}
4480 			}
4481 			if (error == 0) {
4482 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4483 				    NFSX_MYFH);
4484 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4485 				    nmp->nm_nam->sa_len);
4486 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4487 				    pf->dsf_filename,
4488 				    sizeof(pf->dsf_filename));
4489 			}
4490 		} else
4491 			printf("pNFS: pnfscreate can't get DS"
4492 			    " attr=%d\n", error);
4493 		if (nvpp != NULL && error == 0)
4494 			*nvpp = nvp;
4495 		else
4496 			vput(nvp);
4497 	}
4498 	nfsvno_relpathbuf(&named);
4499 	return (error);
4500 }
4501 
4502 /*
4503  * Start up the thread that will execute nfsrv_dscreate().
4504  */
4505 static void
start_dscreate(void * arg,int pending)4506 start_dscreate(void *arg, int pending)
4507 {
4508 	struct nfsrvdscreate *dsc;
4509 
4510 	dsc = (struct nfsrvdscreate *)arg;
4511 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4512 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4513 	dsc->done = 1;
4514 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4515 }
4516 
4517 /*
4518  * Create a pNFS data file on the Data Server(s).
4519  */
4520 static void
nfsrv_pnfscreate(struct vnode * vp,struct vattr * vap,struct ucred * cred,NFSPROC_T * p)4521 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4522     NFSPROC_T *p)
4523 {
4524 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4525 	struct nfsdevice *ds, *tds, *fds;
4526 	struct mount *mp;
4527 	struct pnfsdsfile *pf, *tpf;
4528 	struct pnfsdsattr dsattr;
4529 	struct vattr va;
4530 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4531 	struct nfsmount *nmp;
4532 	fhandle_t fh;
4533 	uid_t vauid;
4534 	gid_t vagid;
4535 	u_short vamode;
4536 	struct ucred *tcred;
4537 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4538 	int failpos, timo;
4539 
4540 	/* Get a DS server directory in a round-robin order. */
4541 	mirrorcnt = 1;
4542 	mp = vp->v_mount;
4543 	ds = fds = NULL;
4544 	NFSDDSLOCK();
4545 	/*
4546 	 * Search for the first entry that handles this MDS fs, but use the
4547 	 * first entry for all MDS fs's otherwise.
4548 	 */
4549 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4550 		if (tds->nfsdev_nmp != NULL) {
4551 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4552 				ds = tds;
4553 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4554 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4555 				ds = fds = tds;
4556 				break;
4557 			}
4558 		}
4559 	}
4560 	if (ds == NULL) {
4561 		NFSDDSUNLOCK();
4562 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4563 		return;
4564 	}
4565 	i = dsdir[0] = ds->nfsdev_nextdir;
4566 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4567 	dvp[0] = ds->nfsdev_dsdir[i];
4568 	tds = TAILQ_NEXT(ds, nfsdev_list);
4569 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4570 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4571 			if (tds->nfsdev_nmp != NULL &&
4572 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4573 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4574 			      fsidcmp(&mp->mnt_stat.f_fsid,
4575 			      &tds->nfsdev_mdsfsid) == 0))) {
4576 				dsdir[mirrorcnt] = i;
4577 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4578 				mirrorcnt++;
4579 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4580 					break;
4581 			}
4582 		}
4583 	}
4584 	/* Put at end of list to implement round-robin usage. */
4585 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4586 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4587 	NFSDDSUNLOCK();
4588 	dsc = NULL;
4589 	if (mirrorcnt > 1)
4590 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4591 		    M_WAITOK | M_ZERO);
4592 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4593 	    M_ZERO);
4594 
4595 	error = nfsvno_getfh(vp, &fh, p);
4596 	if (error == 0)
4597 		error = VOP_GETATTR(vp, &va, cred);
4598 	if (error == 0) {
4599 		/* Set the attributes for "vp" to Setattr the DS vp. */
4600 		vauid = va.va_uid;
4601 		vagid = va.va_gid;
4602 		vamode = va.va_mode;
4603 		VATTR_NULL(&va);
4604 		va.va_uid = vauid;
4605 		va.va_gid = vagid;
4606 		va.va_mode = vamode;
4607 		va.va_size = 0;
4608 	} else
4609 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4610 
4611 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4612 	    cred->cr_gid);
4613 	/* Make data file name based on FH. */
4614 	tcred = newnfs_getcred();
4615 
4616 	/*
4617 	 * Create the file on each DS mirror, using kernel process(es) for the
4618 	 * additional mirrors.
4619 	 */
4620 	failpos = -1;
4621 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4622 		tpf->dsf_dir = dsdir[i];
4623 		tdsc->tcred = tcred;
4624 		tdsc->p = p;
4625 		tdsc->pf = tpf;
4626 		tdsc->createva = *vap;
4627 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4628 		tdsc->va = va;
4629 		tdsc->dvp = dvp[i];
4630 		tdsc->done = 0;
4631 		tdsc->inprog = 0;
4632 		tdsc->err = 0;
4633 		ret = EIO;
4634 		if (nfs_pnfsiothreads != 0) {
4635 			ret = nfs_pnfsio(start_dscreate, tdsc);
4636 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4637 		}
4638 		if (ret != 0) {
4639 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4640 			    NULL, tcred, p, NULL);
4641 			if (ret != 0) {
4642 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4643 				    error));
4644 				if (failpos == -1 && nfsds_failerr(ret))
4645 					failpos = i;
4646 				else
4647 					error = ret;
4648 			}
4649 		}
4650 	}
4651 	if (error == 0) {
4652 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4653 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4654 		    &dsattr, NULL, tcred, p, NULL);
4655 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4656 			failpos = mirrorcnt - 1;
4657 			error = 0;
4658 		}
4659 	}
4660 	timo = hz / 50;		/* Wait for 20msec. */
4661 	if (timo < 1)
4662 		timo = 1;
4663 	/* Wait for kernel task(s) to complete. */
4664 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4665 		while (tdsc->inprog != 0 && tdsc->done == 0)
4666 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4667 		if (tdsc->err != 0) {
4668 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4669 				failpos = i;
4670 			else if (error == 0)
4671 				error = tdsc->err;
4672 		}
4673 	}
4674 
4675 	/*
4676 	 * If failpos has been set, that mirror has failed, so it needs
4677 	 * to be disabled.
4678 	 */
4679 	if (failpos >= 0) {
4680 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4681 		NFSLOCKMNT(nmp);
4682 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4683 		     NFSMNTP_CANCELRPCS)) == 0) {
4684 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4685 			NFSUNLOCKMNT(nmp);
4686 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4687 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4688 			    ds);
4689 			if (ds != NULL)
4690 				nfsrv_killrpcs(nmp);
4691 			NFSLOCKMNT(nmp);
4692 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4693 			wakeup(nmp);
4694 		}
4695 		NFSUNLOCKMNT(nmp);
4696 	}
4697 
4698 	NFSFREECRED(tcred);
4699 	if (error == 0) {
4700 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4701 
4702 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4703 		    mirrorcnt, nfsrv_maxpnfsmirror);
4704 		/*
4705 		 * For all mirrors that couldn't be created, fill in the
4706 		 * *pf structure, but with an IP address == 0.0.0.0.
4707 		 */
4708 		tpf = pf + mirrorcnt;
4709 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4710 			*tpf = *pf;
4711 			tpf->dsf_sin.sin_family = AF_INET;
4712 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4713 			tpf->dsf_sin.sin_addr.s_addr = 0;
4714 			tpf->dsf_sin.sin_port = 0;
4715 		}
4716 
4717 		error = vn_extattr_set(vp, IO_NODELOCKED,
4718 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4719 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4720 		if (error == 0)
4721 			error = vn_extattr_set(vp, IO_NODELOCKED,
4722 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4723 			    sizeof(dsattr), (char *)&dsattr, p);
4724 		if (error != 0)
4725 			printf("pNFS: pnfscreate setextattr=%d\n",
4726 			    error);
4727 	} else
4728 		printf("pNFS: pnfscreate=%d\n", error);
4729 	free(pf, M_TEMP);
4730 	free(dsc, M_TEMP);
4731 }
4732 
4733 /*
4734  * Get the information needed to remove the pNFS Data Server file from the
4735  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4736  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4737  */
4738 static void
nfsrv_pnfsremovesetup(struct vnode * vp,NFSPROC_T * p,struct vnode ** dvpp,int * mirrorcntp,char * fname,fhandle_t * fhp)4739 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4740     int *mirrorcntp, char *fname, fhandle_t *fhp)
4741 {
4742 	struct vattr va;
4743 	struct ucred *tcred;
4744 	char *buf;
4745 	int buflen, error;
4746 
4747 	dvpp[0] = NULL;
4748 	/* If not an exported regular file or not a pNFS server, just return. */
4749 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4750 	    nfsrv_devidcnt == 0)
4751 		return;
4752 
4753 	/* Check to see if this is the last hard link. */
4754 	tcred = newnfs_getcred();
4755 	error = VOP_GETATTR(vp, &va, tcred);
4756 	NFSFREECRED(tcred);
4757 	if (error != 0) {
4758 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4759 		return;
4760 	}
4761 	if (va.va_nlink > 1)
4762 		return;
4763 
4764 	error = nfsvno_getfh(vp, fhp, p);
4765 	if (error != 0) {
4766 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4767 		return;
4768 	}
4769 
4770 	buflen = 1024;
4771 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4772 	/* Get the directory vnode for the DS mount and the file handle. */
4773 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4774 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4775 	free(buf, M_TEMP);
4776 	if (error != 0)
4777 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4778 }
4779 
4780 /*
4781  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4782  * The arguments are in a structure, so that they can be passed through
4783  * taskqueue for a kernel process to execute this function.
4784  */
4785 struct nfsrvdsremove {
4786 	int			done;
4787 	int			inprog;
4788 	struct task		tsk;
4789 	struct ucred		*tcred;
4790 	struct vnode		*dvp;
4791 	NFSPROC_T		*p;
4792 	int			err;
4793 	char			fname[PNFS_FILENAME_LEN + 1];
4794 };
4795 
4796 static int
nfsrv_dsremove(struct vnode * dvp,char * fname,struct ucred * tcred,NFSPROC_T * p)4797 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4798     NFSPROC_T *p)
4799 {
4800 	struct nameidata named;
4801 	struct vnode *nvp;
4802 	char *bufp;
4803 	u_long *hashp;
4804 	int error;
4805 
4806 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4807 	if (error != 0)
4808 		return (error);
4809 	named.ni_cnd.cn_nameiop = DELETE;
4810 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4811 	named.ni_cnd.cn_cred = tcred;
4812 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
4813 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4814 	named.ni_cnd.cn_nameptr = bufp;
4815 	named.ni_cnd.cn_namelen = strlen(fname);
4816 	strlcpy(bufp, fname, NAME_MAX);
4817 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4818 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4819 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4820 	if (error == 0) {
4821 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4822 		vput(nvp);
4823 	}
4824 	NFSVOPUNLOCK(dvp);
4825 	nfsvno_relpathbuf(&named);
4826 	if (error != 0)
4827 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4828 	return (error);
4829 }
4830 
4831 /*
4832  * Start up the thread that will execute nfsrv_dsremove().
4833  */
4834 static void
start_dsremove(void * arg,int pending)4835 start_dsremove(void *arg, int pending)
4836 {
4837 	struct nfsrvdsremove *dsrm;
4838 
4839 	dsrm = (struct nfsrvdsremove *)arg;
4840 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4841 	    dsrm->p);
4842 	dsrm->done = 1;
4843 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4844 }
4845 
4846 /*
4847  * Remove a pNFS data file from a Data Server.
4848  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4849  * removed to set up the dvp and fill in the FH.
4850  */
4851 static void
nfsrv_pnfsremove(struct vnode ** dvp,int mirrorcnt,char * fname,fhandle_t * fhp,NFSPROC_T * p)4852 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4853     NFSPROC_T *p)
4854 {
4855 	struct ucred *tcred;
4856 	struct nfsrvdsremove *dsrm, *tdsrm;
4857 	struct nfsdevice *ds;
4858 	struct nfsmount *nmp;
4859 	int failpos, i, ret, timo;
4860 
4861 	tcred = newnfs_getcred();
4862 	dsrm = NULL;
4863 	if (mirrorcnt > 1)
4864 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4865 	/*
4866 	 * Remove the file on each DS mirror, using kernel process(es) for the
4867 	 * additional mirrors.
4868 	 */
4869 	failpos = -1;
4870 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4871 		tdsrm->tcred = tcred;
4872 		tdsrm->p = p;
4873 		tdsrm->dvp = dvp[i];
4874 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4875 		tdsrm->inprog = 0;
4876 		tdsrm->done = 0;
4877 		tdsrm->err = 0;
4878 		ret = EIO;
4879 		if (nfs_pnfsiothreads != 0) {
4880 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4881 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4882 		}
4883 		if (ret != 0) {
4884 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4885 			if (failpos == -1 && nfsds_failerr(ret))
4886 				failpos = i;
4887 		}
4888 	}
4889 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4890 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4891 		failpos = mirrorcnt - 1;
4892 	timo = hz / 50;		/* Wait for 20msec. */
4893 	if (timo < 1)
4894 		timo = 1;
4895 	/* Wait for kernel task(s) to complete. */
4896 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4897 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4898 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4899 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4900 			failpos = i;
4901 	}
4902 
4903 	/*
4904 	 * If failpos has been set, that mirror has failed, so it needs
4905 	 * to be disabled.
4906 	 */
4907 	if (failpos >= 0) {
4908 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4909 		NFSLOCKMNT(nmp);
4910 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4911 		     NFSMNTP_CANCELRPCS)) == 0) {
4912 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4913 			NFSUNLOCKMNT(nmp);
4914 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4915 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4916 			    ds);
4917 			if (ds != NULL)
4918 				nfsrv_killrpcs(nmp);
4919 			NFSLOCKMNT(nmp);
4920 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4921 			wakeup(nmp);
4922 		}
4923 		NFSUNLOCKMNT(nmp);
4924 	}
4925 
4926 	/* Get rid all layouts for the file. */
4927 	nfsrv_freefilelayouts(fhp);
4928 
4929 	NFSFREECRED(tcred);
4930 	free(dsrm, M_TEMP);
4931 }
4932 
4933 /*
4934  * Generate a file name based on the file handle and put it in *bufp.
4935  * Return the number of bytes generated.
4936  */
4937 static int
nfsrv_putfhname(fhandle_t * fhp,char * bufp)4938 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4939 {
4940 	int i;
4941 	uint8_t *cp;
4942 	const uint8_t *hexdigits = "0123456789abcdef";
4943 
4944 	cp = (uint8_t *)fhp;
4945 	for (i = 0; i < sizeof(*fhp); i++) {
4946 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4947 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4948 	}
4949 	bufp[2 * i] = '\0';
4950 	return (2 * i);
4951 }
4952 
4953 /*
4954  * Update the Metadata file's attributes from the DS file when a Read/Write
4955  * layout is returned.
4956  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4957  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4958  */
4959 int
nfsrv_updatemdsattr(struct vnode * vp,struct nfsvattr * nap,NFSPROC_T * p)4960 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4961 {
4962 	struct ucred *tcred;
4963 	int error;
4964 
4965 	/* Do this as root so that it won't fail with EACCES. */
4966 	tcred = newnfs_getcred();
4967 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4968 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4969 	NFSFREECRED(tcred);
4970 	return (error);
4971 }
4972 
4973 /*
4974  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4975  */
4976 static int
nfsrv_dssetacl(struct vnode * vp,struct acl * aclp,struct ucred * cred,NFSPROC_T * p)4977 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4978     NFSPROC_T *p)
4979 {
4980 	int error;
4981 
4982 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4983 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4984 	return (error);
4985 }
4986 
4987 static int
nfsrv_proxyds(struct vnode * vp,off_t off,int cnt,struct ucred * cred,struct thread * p,int ioproc,struct mbuf ** mpp,char * cp,struct mbuf ** mpp2,struct nfsvattr * nap,struct acl * aclp,off_t * offp,int content,bool * eofp)4988 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4989     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4990     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4991     off_t *offp, int content, bool *eofp)
4992 {
4993 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4994 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4995 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4996 	struct nfsdevice *ds;
4997 	struct pnfsdsattr dsattr;
4998 	struct opnfsdsattr odsattr;
4999 	char *buf;
5000 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
5001 
5002 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
5003 	/*
5004 	 * If not a regular file, not exported or not a pNFS server,
5005 	 * just return ENOENT.
5006 	 */
5007 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
5008 	    nfsrv_devidcnt == 0)
5009 		return (ENOENT);
5010 
5011 	buflen = 1024;
5012 	buf = malloc(buflen, M_TEMP, M_WAITOK);
5013 	error = 0;
5014 
5015 	/*
5016 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
5017 	 * from the MetaData file's extended attribute.
5018 	 */
5019 	if (ioproc == NFSPROC_GETATTR) {
5020 		error = vn_extattr_get(vp, IO_NODELOCKED,
5021 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
5022 		    p);
5023 		if (error == 0) {
5024 			if (buflen == sizeof(odsattr)) {
5025 				NFSBCOPY(buf, &odsattr, buflen);
5026 				nap->na_filerev = odsattr.dsa_filerev;
5027 				nap->na_size = odsattr.dsa_size;
5028 				nap->na_atime = odsattr.dsa_atime;
5029 				nap->na_mtime = odsattr.dsa_mtime;
5030 				/*
5031 				 * Fake na_bytes by rounding up na_size.
5032 				 * Since we don't know the block size, just
5033 				 * use BLKDEV_IOSIZE.
5034 				 */
5035 				nap->na_bytes = (odsattr.dsa_size +
5036 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
5037 			} else if (buflen == sizeof(dsattr)) {
5038 				NFSBCOPY(buf, &dsattr, buflen);
5039 				nap->na_filerev = dsattr.dsa_filerev;
5040 				nap->na_size = dsattr.dsa_size;
5041 				nap->na_atime = dsattr.dsa_atime;
5042 				nap->na_mtime = dsattr.dsa_mtime;
5043 				nap->na_bytes = dsattr.dsa_bytes;
5044 			} else
5045 				error = ENXIO;
5046 		}
5047 		if (error == 0) {
5048 			/*
5049 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
5050 			 * returns 0, just return now.  nfsrv_checkdsattr()
5051 			 * returns 0 if there is no Read/Write layout
5052 			 * plus either an Open/Write_access or Write
5053 			 * delegation issued to a client for the file.
5054 			 */
5055 			if (nfsrv_pnfsgetdsattr == 0 ||
5056 			    nfsrv_checkdsattr(vp, p) == 0) {
5057 				free(buf, M_TEMP);
5058 				return (error);
5059 			}
5060 		}
5061 
5062 		/*
5063 		 * Clear ENOATTR so the code below will attempt to do a
5064 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
5065 		 * the extended attribute.
5066 		 */
5067 		if (error == ENOATTR)
5068 			error = 0;
5069 	}
5070 
5071 	origmircnt = -1;
5072 	trycnt = 0;
5073 tryagain:
5074 	if (error == 0) {
5075 		buflen = 1024;
5076 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
5077 		    LK_EXCLUSIVE)
5078 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
5079 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
5080 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
5081 		    NULL, NULL);
5082 		if (error == 0) {
5083 			for (i = 0; i < mirrorcnt; i++)
5084 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
5085 		} else
5086 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
5087 	} else
5088 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
5089 	if (error == 0) {
5090 		failpos = -1;
5091 		if (origmircnt == -1)
5092 			origmircnt = mirrorcnt;
5093 		/*
5094 		 * If failpos is set to a mirror#, then that mirror has
5095 		 * failed and will be disabled. For Read, Getattr and Seek, the
5096 		 * function only tries one mirror, so if that mirror has
5097 		 * failed, it will need to be retried. As such, increment
5098 		 * tryitagain for these cases.
5099 		 * For Write, Setattr and Setacl, the function tries all
5100 		 * mirrors and will not return an error for the case where
5101 		 * one mirror has failed. For these cases, the functioning
5102 		 * mirror(s) will have been modified, so a retry isn't
5103 		 * necessary. These functions will set failpos for the
5104 		 * failed mirror#.
5105 		 */
5106 		if (ioproc == NFSPROC_READDS) {
5107 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
5108 			    mpp, mpp2);
5109 			if (nfsds_failerr(error) && mirrorcnt > 1) {
5110 				/*
5111 				 * Setting failpos will cause the mirror
5112 				 * to be disabled and then a retry of this
5113 				 * read is required.
5114 				 */
5115 				failpos = 0;
5116 				error = 0;
5117 				trycnt++;
5118 			}
5119 		} else if (ioproc == NFSPROC_WRITEDS)
5120 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
5121 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
5122 		else if (ioproc == NFSPROC_SETATTR)
5123 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
5124 			    mirrorcnt, nap, &failpos);
5125 		else if (ioproc == NFSPROC_SETACL)
5126 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
5127 			    mirrorcnt, aclp, &failpos);
5128 		else if (ioproc == NFSPROC_SEEKDS) {
5129 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
5130 			    p, nmp[0]);
5131 			if (nfsds_failerr(error) && mirrorcnt > 1) {
5132 				/*
5133 				 * Setting failpos will cause the mirror
5134 				 * to be disabled and then a retry of this
5135 				 * read is required.
5136 				 */
5137 				failpos = 0;
5138 				error = 0;
5139 				trycnt++;
5140 			}
5141 		} else if (ioproc == NFSPROC_ALLOCATE)
5142 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
5143 			    &nmp[0], mirrorcnt, &failpos);
5144 		else if (ioproc == NFSPROC_DEALLOCATE)
5145 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
5146 			    vp, &nmp[0], mirrorcnt, &failpos);
5147 		else {
5148 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
5149 			    vp, nmp[mirrorcnt - 1], nap);
5150 			if (nfsds_failerr(error) && mirrorcnt > 1) {
5151 				/*
5152 				 * Setting failpos will cause the mirror
5153 				 * to be disabled and then a retry of this
5154 				 * getattr is required.
5155 				 */
5156 				failpos = mirrorcnt - 1;
5157 				error = 0;
5158 				trycnt++;
5159 			}
5160 		}
5161 		ds = NULL;
5162 		if (failpos >= 0) {
5163 			failnmp = nmp[failpos];
5164 			NFSLOCKMNT(failnmp);
5165 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
5166 			     NFSMNTP_CANCELRPCS)) == 0) {
5167 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
5168 				NFSUNLOCKMNT(failnmp);
5169 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
5170 				    failnmp, p);
5171 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
5172 				    failpos, ds);
5173 				if (ds != NULL)
5174 					nfsrv_killrpcs(failnmp);
5175 				NFSLOCKMNT(failnmp);
5176 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
5177 				wakeup(failnmp);
5178 			}
5179 			NFSUNLOCKMNT(failnmp);
5180 		}
5181 		for (i = 0; i < mirrorcnt; i++)
5182 			NFSVOPUNLOCK(dvp[i]);
5183 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
5184 		    trycnt);
5185 		/* Try the Read/Getattr again if a mirror was deleted. */
5186 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
5187 			goto tryagain;
5188 	} else {
5189 		/* Return ENOENT for any Extended Attribute error. */
5190 		error = ENOENT;
5191 	}
5192 	free(buf, M_TEMP);
5193 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
5194 	return (error);
5195 }
5196 
5197 /*
5198  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
5199  * attribute.
5200  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
5201  *           to be checked.  If it points to a NULL nmp, then it returns
5202  *           a suitable destination.
5203  * curnmp - If non-NULL, it is the source mount for the copy.
5204  */
5205 int
nfsrv_dsgetsockmnt(struct vnode * vp,int lktype,char * buf,int * buflenp,int * mirrorcntp,NFSPROC_T * p,struct vnode ** dvpp,fhandle_t * fhp,char * devid,char * fnamep,struct vnode ** nvpp,struct nfsmount ** newnmpp,struct nfsmount * curnmp,int * ippos,int * dsdirp)5206 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
5207     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
5208     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
5209     struct nfsmount *curnmp, int *ippos, int *dsdirp)
5210 {
5211 	struct vnode *dvp, *nvp = NULL, **tdvpp;
5212 	struct mount *mp;
5213 	struct nfsmount *nmp, *newnmp;
5214 	struct sockaddr *sad;
5215 	struct sockaddr_in *sin;
5216 	struct nfsdevice *ds, *tds, *fndds;
5217 	struct pnfsdsfile *pf;
5218 	uint32_t dsdir;
5219 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
5220 
5221 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
5222 	*mirrorcntp = 1;
5223 	tdvpp = dvpp;
5224 	if (nvpp != NULL)
5225 		*nvpp = NULL;
5226 	if (dvpp != NULL)
5227 		*dvpp = NULL;
5228 	if (ippos != NULL)
5229 		*ippos = -1;
5230 	if (newnmpp != NULL)
5231 		newnmp = *newnmpp;
5232 	else
5233 		newnmp = NULL;
5234 	mp = vp->v_mount;
5235 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5236 	    "pnfsd.dsfile", buflenp, buf, p);
5237 	mirrorcnt = *buflenp / sizeof(*pf);
5238 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
5239 	    *buflenp != sizeof(*pf) * mirrorcnt))
5240 		error = ENOATTR;
5241 
5242 	pf = (struct pnfsdsfile *)buf;
5243 	/* If curnmp != NULL, check for a match in the mirror list. */
5244 	if (curnmp != NULL && error == 0) {
5245 		fnd = 0;
5246 		for (i = 0; i < mirrorcnt; i++, pf++) {
5247 			sad = (struct sockaddr *)&pf->dsf_sin;
5248 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5249 				if (ippos != NULL)
5250 					*ippos = i;
5251 				fnd = 1;
5252 				break;
5253 			}
5254 		}
5255 		if (fnd == 0)
5256 			error = ENXIO;
5257 	}
5258 
5259 	gotone = 0;
5260 	pf = (struct pnfsdsfile *)buf;
5261 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5262 	    error);
5263 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5264 		fhiszero = 0;
5265 		sad = (struct sockaddr *)&pf->dsf_sin;
5266 		sin = &pf->dsf_sin;
5267 		dsdir = pf->dsf_dir;
5268 		if (dsdir >= nfsrv_dsdirsize) {
5269 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5270 			error = ENOATTR;
5271 		} else if (nvpp != NULL && newnmp != NULL &&
5272 		    nfsaddr2_match(sad, newnmp->nm_nam))
5273 			error = EEXIST;
5274 		if (error == 0) {
5275 			if (ippos != NULL && curnmp == NULL &&
5276 			    sad->sa_family == AF_INET &&
5277 			    sin->sin_addr.s_addr == 0)
5278 				*ippos = i;
5279 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5280 				fhiszero = 1;
5281 			/* Use the socket address to find the mount point. */
5282 			fndds = NULL;
5283 			NFSDDSLOCK();
5284 			/* Find a match for the IP address. */
5285 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5286 				if (ds->nfsdev_nmp != NULL) {
5287 					dvp = ds->nfsdev_dvp;
5288 					nmp = VFSTONFS(dvp->v_mount);
5289 					if (nmp != ds->nfsdev_nmp)
5290 						printf("different2 nmp %p %p\n",
5291 						    nmp, ds->nfsdev_nmp);
5292 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5293 						fndds = ds;
5294 						break;
5295 					}
5296 				}
5297 			}
5298 			if (fndds != NULL && newnmpp != NULL &&
5299 			    newnmp == NULL) {
5300 				/* Search for a place to make a mirror copy. */
5301 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5302 				    nfsdev_list) {
5303 					if (tds->nfsdev_nmp != NULL &&
5304 					    fndds != tds &&
5305 					    ((tds->nfsdev_mdsisset == 0 &&
5306 					      fndds->nfsdev_mdsisset == 0) ||
5307 					     (tds->nfsdev_mdsisset != 0 &&
5308 					      fndds->nfsdev_mdsisset != 0 &&
5309 					      fsidcmp(&tds->nfsdev_mdsfsid,
5310 					      &mp->mnt_stat.f_fsid) == 0))) {
5311 						*newnmpp = tds->nfsdev_nmp;
5312 						break;
5313 					}
5314 				}
5315 				if (tds != NULL) {
5316 					/*
5317 					 * Move this entry to the end of the
5318 					 * list, so it won't be selected as
5319 					 * easily the next time.
5320 					 */
5321 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5322 					    nfsdev_list);
5323 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5324 					    nfsdev_list);
5325 				}
5326 			}
5327 			NFSDDSUNLOCK();
5328 			if (fndds != NULL) {
5329 				dvp = fndds->nfsdev_dsdir[dsdir];
5330 				if (lktype != 0 || fhiszero != 0 ||
5331 				    (nvpp != NULL && *nvpp == NULL)) {
5332 					if (fhiszero != 0)
5333 						error = vn_lock(dvp,
5334 						    LK_EXCLUSIVE);
5335 					else if (lktype != 0)
5336 						error = vn_lock(dvp, lktype);
5337 					else
5338 						error = vn_lock(dvp, LK_SHARED);
5339 					/*
5340 					 * If the file handle is all 0's, try to
5341 					 * do a Lookup against the DS to acquire
5342 					 * it.
5343 					 * If dvpp == NULL or the Lookup fails,
5344 					 * unlock dvp after the call.
5345 					 */
5346 					if (error == 0 && (fhiszero != 0 ||
5347 					    (nvpp != NULL && *nvpp == NULL))) {
5348 						error = nfsrv_pnfslookupds(vp,
5349 						    dvp, pf, &nvp, p);
5350 						if (error == 0) {
5351 							if (fhiszero != 0)
5352 								nfsrv_pnfssetfh(
5353 								    vp, pf,
5354 								    devid,
5355 								    fnamep,
5356 								    nvp, p);
5357 							if (nvpp != NULL &&
5358 							    *nvpp == NULL) {
5359 								*nvpp = nvp;
5360 								*dsdirp = dsdir;
5361 							} else
5362 								vput(nvp);
5363 						}
5364 						if (error != 0 || lktype == 0)
5365 							NFSVOPUNLOCK(dvp);
5366 					}
5367 				}
5368 				if (error == 0) {
5369 					gotone++;
5370 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5371 					if (devid != NULL) {
5372 						NFSBCOPY(fndds->nfsdev_deviceid,
5373 						    devid, NFSX_V4DEVICEID);
5374 						devid += NFSX_V4DEVICEID;
5375 					}
5376 					if (dvpp != NULL)
5377 						*tdvpp++ = dvp;
5378 					if (fhp != NULL)
5379 						NFSBCOPY(&pf->dsf_fh, fhp++,
5380 						    NFSX_MYFH);
5381 					if (fnamep != NULL && gotone == 1)
5382 						strlcpy(fnamep,
5383 						    pf->dsf_filename,
5384 						    sizeof(pf->dsf_filename));
5385 				} else
5386 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5387 					    "err=%d\n", error);
5388 			}
5389 		}
5390 	}
5391 	if (error == 0 && gotone == 0)
5392 		error = ENOENT;
5393 
5394 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5395 	    error);
5396 	if (error == 0)
5397 		*mirrorcntp = gotone;
5398 	else {
5399 		if (gotone > 0 && dvpp != NULL) {
5400 			/*
5401 			 * If the error didn't occur on the first one and
5402 			 * dvpp != NULL, the one(s) prior to the failure will
5403 			 * have locked dvp's that need to be unlocked.
5404 			 */
5405 			for (i = 0; i < gotone; i++) {
5406 				NFSVOPUNLOCK(*dvpp);
5407 				*dvpp++ = NULL;
5408 			}
5409 		}
5410 		/*
5411 		 * If it found the vnode to be copied from before a failure,
5412 		 * it needs to be vput()'d.
5413 		 */
5414 		if (nvpp != NULL && *nvpp != NULL) {
5415 			vput(*nvpp);
5416 			*nvpp = NULL;
5417 		}
5418 	}
5419 	return (error);
5420 }
5421 
5422 /*
5423  * Set the extended attribute for the Change attribute.
5424  */
5425 static int
nfsrv_setextattr(struct vnode * vp,struct nfsvattr * nap,NFSPROC_T * p)5426 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5427 {
5428 	struct pnfsdsattr dsattr;
5429 	int error;
5430 
5431 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5432 	dsattr.dsa_filerev = nap->na_filerev;
5433 	dsattr.dsa_size = nap->na_size;
5434 	dsattr.dsa_atime = nap->na_atime;
5435 	dsattr.dsa_mtime = nap->na_mtime;
5436 	dsattr.dsa_bytes = nap->na_bytes;
5437 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5438 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5439 	if (error != 0)
5440 		printf("pNFS: setextattr=%d\n", error);
5441 	return (error);
5442 }
5443 
5444 static int
nfsrv_readdsrpc(fhandle_t * fhp,off_t off,int len,struct ucred * cred,NFSPROC_T * p,struct nfsmount * nmp,struct mbuf ** mpp,struct mbuf ** mpendp)5445 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5446     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5447 {
5448 	uint32_t *tl;
5449 	struct nfsrv_descript *nd;
5450 	nfsv4stateid_t st;
5451 	struct mbuf *m, *m2;
5452 	int error = 0, retlen, tlen, trimlen;
5453 
5454 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5455 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5456 	*mpp = NULL;
5457 	/*
5458 	 * Use a stateid where other is an alternating 01010 pattern and
5459 	 * seqid is 0xffffffff.  This value is not defined as special by
5460 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5461 	 * MDS->DS proxy operation.
5462 	 */
5463 	st.other[0] = 0x55555555;
5464 	st.other[1] = 0x55555555;
5465 	st.other[2] = 0x55555555;
5466 	st.seqid = 0xffffffff;
5467 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5468 	    NULL, NULL, 0, 0, cred);
5469 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5470 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5471 	txdr_hyper(off, tl);
5472 	*(tl + 2) = txdr_unsigned(len);
5473 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5474 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5475 	if (error != 0) {
5476 		free(nd, M_TEMP);
5477 		return (error);
5478 	}
5479 	if (nd->nd_repstat == 0) {
5480 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5481 		NFSM_STRSIZ(retlen, len);
5482 		if (retlen > 0) {
5483 			/* Trim off the pre-data XDR from the mbuf chain. */
5484 			m = nd->nd_mrep;
5485 			while (m != NULL && m != nd->nd_md) {
5486 				if (m->m_next == nd->nd_md) {
5487 					m->m_next = NULL;
5488 					m_freem(nd->nd_mrep);
5489 					nd->nd_mrep = m = nd->nd_md;
5490 				} else
5491 					m = m->m_next;
5492 			}
5493 			if (m == NULL) {
5494 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5495 				error = ENOENT;
5496 				goto nfsmout;
5497 			}
5498 
5499 			/*
5500 			 * Now, adjust first mbuf so that any XDR before the
5501 			 * read data is skipped over.
5502 			 */
5503 			trimlen = nd->nd_dpos - mtod(m, char *);
5504 			if (trimlen > 0) {
5505 				m->m_len -= trimlen;
5506 				NFSM_DATAP(m, trimlen);
5507 			}
5508 
5509 			/*
5510 			 * Truncate the mbuf chain at retlen bytes of data,
5511 			 * plus XDR padding that brings the length up to a
5512 			 * multiple of 4.
5513 			 */
5514 			tlen = NFSM_RNDUP(retlen);
5515 			do {
5516 				if (m->m_len >= tlen) {
5517 					m->m_len = tlen;
5518 					tlen = 0;
5519 					m2 = m->m_next;
5520 					m->m_next = NULL;
5521 					m_freem(m2);
5522 					break;
5523 				}
5524 				tlen -= m->m_len;
5525 				m = m->m_next;
5526 			} while (m != NULL);
5527 			if (tlen > 0) {
5528 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5529 				error = ENOENT;
5530 				goto nfsmout;
5531 			}
5532 			*mpp = nd->nd_mrep;
5533 			*mpendp = m;
5534 			nd->nd_mrep = NULL;
5535 		}
5536 	} else
5537 		error = nd->nd_repstat;
5538 nfsmout:
5539 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5540 	m_freem(nd->nd_mrep);
5541 	free(nd, M_TEMP);
5542 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5543 	return (error);
5544 }
5545 
5546 /*
5547  * Do a write RPC on a DS data file, using this structure for the arguments,
5548  * so that this function can be executed by a separate kernel process.
5549  */
5550 struct nfsrvwritedsdorpc {
5551 	int			done;
5552 	int			inprog;
5553 	struct task		tsk;
5554 	fhandle_t		fh;
5555 	off_t			off;
5556 	int			len;
5557 	struct nfsmount		*nmp;
5558 	struct ucred		*cred;
5559 	NFSPROC_T		*p;
5560 	struct mbuf		*m;
5561 	int			err;
5562 };
5563 
5564 static int
nfsrv_writedsdorpc(struct nfsmount * nmp,fhandle_t * fhp,off_t off,int len,struct nfsvattr * nap,struct mbuf * m,struct ucred * cred,NFSPROC_T * p)5565 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5566     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5567 {
5568 	uint32_t *tl;
5569 	struct nfsrv_descript *nd;
5570 	nfsattrbit_t attrbits;
5571 	nfsv4stateid_t st;
5572 	int commit, error, retlen;
5573 
5574 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5575 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5576 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5577 
5578 	/*
5579 	 * Use a stateid where other is an alternating 01010 pattern and
5580 	 * seqid is 0xffffffff.  This value is not defined as special by
5581 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5582 	 * MDS->DS proxy operation.
5583 	 */
5584 	st.other[0] = 0x55555555;
5585 	st.other[1] = 0x55555555;
5586 	st.other[2] = 0x55555555;
5587 	st.seqid = 0xffffffff;
5588 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5589 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5590 	txdr_hyper(off, tl);
5591 	tl += 2;
5592 	/*
5593 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5594 	 * buffers.  Since clients should be accessing the DS servers directly
5595 	 * using the pNFS layouts, this just needs to work correctly as a
5596 	 * fallback.
5597 	 */
5598 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5599 	*tl = txdr_unsigned(len);
5600 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5601 
5602 	/* Put data in mbuf chain. */
5603 	nd->nd_mb->m_next = m;
5604 
5605 	/* Set nd_mb and nd_bpos to end of data. */
5606 	while (m->m_next != NULL)
5607 		m = m->m_next;
5608 	nd->nd_mb = m;
5609 	nfsm_set(nd, m->m_len);
5610 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5611 
5612 	/* Do a Getattr for the attributes that change upon writing. */
5613 	NFSZERO_ATTRBIT(&attrbits);
5614 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5615 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5616 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5617 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5618 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5619 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5620 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5621 	(void) nfsrv_putattrbit(nd, &attrbits);
5622 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5623 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5624 	if (error != 0) {
5625 		free(nd, M_TEMP);
5626 		return (error);
5627 	}
5628 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5629 	/* Get rid of weak cache consistency data for now. */
5630 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5631 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5632 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5633 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5634 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5635 		if (error != 0)
5636 			goto nfsmout;
5637 		/*
5638 		 * Get rid of Op# and status for next op.
5639 		 */
5640 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5641 		if (*++tl != 0)
5642 			nd->nd_flag |= ND_NOMOREDATA;
5643 	}
5644 	if (nd->nd_repstat == 0) {
5645 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5646 		retlen = fxdr_unsigned(int, *tl++);
5647 		commit = fxdr_unsigned(int, *tl);
5648 		if (commit != NFSWRITE_FILESYNC)
5649 			error = NFSERR_IO;
5650 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5651 		    retlen, commit, error);
5652 	} else
5653 		error = nd->nd_repstat;
5654 	/* We have no use for the Write Verifier since we use FileSync. */
5655 
5656 	/*
5657 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5658 	 * on the Metadata file, so its attributes will be what the file's
5659 	 * would be if it had been written.
5660 	 */
5661 	if (error == 0) {
5662 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5663 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5664 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5665 	}
5666 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5667 nfsmout:
5668 	m_freem(nd->nd_mrep);
5669 	free(nd, M_TEMP);
5670 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5671 	return (error);
5672 }
5673 
5674 /*
5675  * Start up the thread that will execute nfsrv_writedsdorpc().
5676  */
5677 static void
start_writedsdorpc(void * arg,int pending)5678 start_writedsdorpc(void *arg, int pending)
5679 {
5680 	struct nfsrvwritedsdorpc *drpc;
5681 
5682 	drpc = (struct nfsrvwritedsdorpc *)arg;
5683 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5684 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5685 	drpc->done = 1;
5686 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5687 }
5688 
5689 static int
nfsrv_writedsrpc(fhandle_t * fhp,off_t off,int len,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount ** nmpp,int mirrorcnt,struct mbuf ** mpp,char * cp,int * failposp)5690 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5691     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5692     struct mbuf **mpp, char *cp, int *failposp)
5693 {
5694 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5695 	struct nfsvattr na;
5696 	struct mbuf *m;
5697 	int error, i, offs, ret, timo;
5698 
5699 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5700 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5701 	drpc = NULL;
5702 	if (mirrorcnt > 1)
5703 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5704 		    M_WAITOK);
5705 
5706 	/* Calculate offset in mbuf chain that data starts. */
5707 	offs = cp - mtod(*mpp, char *);
5708 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5709 
5710 	/*
5711 	 * Do the write RPC for every DS, using a separate kernel process
5712 	 * for every DS except the last one.
5713 	 */
5714 	error = 0;
5715 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5716 		tdrpc->done = 0;
5717 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5718 		tdrpc->off = off;
5719 		tdrpc->len = len;
5720 		tdrpc->nmp = *nmpp;
5721 		tdrpc->cred = cred;
5722 		tdrpc->p = p;
5723 		tdrpc->inprog = 0;
5724 		tdrpc->err = 0;
5725 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5726 		ret = EIO;
5727 		if (nfs_pnfsiothreads != 0) {
5728 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5729 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5730 			    ret);
5731 		}
5732 		if (ret != 0) {
5733 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5734 			    tdrpc->m, cred, p);
5735 			if (nfsds_failerr(ret) && *failposp == -1)
5736 				*failposp = i;
5737 			else if (error == 0 && ret != 0)
5738 				error = ret;
5739 		}
5740 		nmpp++;
5741 		fhp++;
5742 	}
5743 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5744 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5745 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5746 		*failposp = mirrorcnt - 1;
5747 	else if (error == 0 && ret != 0)
5748 		error = ret;
5749 	if (error == 0)
5750 		error = nfsrv_setextattr(vp, &na, p);
5751 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5752 	tdrpc = drpc;
5753 	timo = hz / 50;		/* Wait for 20msec. */
5754 	if (timo < 1)
5755 		timo = 1;
5756 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5757 		/* Wait for RPCs on separate threads to complete. */
5758 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5759 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5760 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5761 			*failposp = i;
5762 		else if (error == 0 && tdrpc->err != 0)
5763 			error = tdrpc->err;
5764 	}
5765 	free(drpc, M_TEMP);
5766 	return (error);
5767 }
5768 
5769 /*
5770  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5771  * so that this function can be executed by a separate kernel process.
5772  */
5773 struct nfsrvallocatedsdorpc {
5774 	int			done;
5775 	int			inprog;
5776 	struct task		tsk;
5777 	fhandle_t		fh;
5778 	off_t			off;
5779 	off_t			len;
5780 	struct nfsmount		*nmp;
5781 	struct ucred		*cred;
5782 	NFSPROC_T		*p;
5783 	int			err;
5784 };
5785 
5786 static int
nfsrv_allocatedsdorpc(struct nfsmount * nmp,fhandle_t * fhp,off_t off,off_t len,struct nfsvattr * nap,struct ucred * cred,NFSPROC_T * p)5787 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5788     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5789 {
5790 	uint32_t *tl;
5791 	struct nfsrv_descript *nd;
5792 	nfsattrbit_t attrbits;
5793 	nfsv4stateid_t st;
5794 	int error;
5795 
5796 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5797 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5798 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5799 
5800 	/*
5801 	 * Use a stateid where other is an alternating 01010 pattern and
5802 	 * seqid is 0xffffffff.  This value is not defined as special by
5803 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5804 	 * MDS->DS proxy operation.
5805 	 */
5806 	st.other[0] = 0x55555555;
5807 	st.other[1] = 0x55555555;
5808 	st.other[2] = 0x55555555;
5809 	st.seqid = 0xffffffff;
5810 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5811 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5812 	txdr_hyper(off, tl); tl += 2;
5813 	txdr_hyper(len, tl); tl += 2;
5814 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5815 
5816 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5817 	NFSGETATTR_ATTRBIT(&attrbits);
5818 	nfsrv_putattrbit(nd, &attrbits);
5819 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5820 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5821 	if (error != 0) {
5822 		free(nd, M_TEMP);
5823 		return (error);
5824 	}
5825 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5826 	    nd->nd_repstat);
5827 	if (nd->nd_repstat == 0) {
5828 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5829 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5830 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5831 	} else
5832 		error = nd->nd_repstat;
5833 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5834 nfsmout:
5835 	m_freem(nd->nd_mrep);
5836 	free(nd, M_TEMP);
5837 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5838 	return (error);
5839 }
5840 
5841 /*
5842  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5843  */
5844 static void
start_allocatedsdorpc(void * arg,int pending)5845 start_allocatedsdorpc(void *arg, int pending)
5846 {
5847 	struct nfsrvallocatedsdorpc *drpc;
5848 
5849 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5850 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5851 	    drpc->len, NULL, drpc->cred, drpc->p);
5852 	drpc->done = 1;
5853 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5854 }
5855 
5856 static int
nfsrv_allocatedsrpc(fhandle_t * fhp,off_t off,off_t len,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount ** nmpp,int mirrorcnt,int * failposp)5857 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5858     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5859     int *failposp)
5860 {
5861 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5862 	struct nfsvattr na;
5863 	int error, i, ret, timo;
5864 
5865 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5866 	drpc = NULL;
5867 	if (mirrorcnt > 1)
5868 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5869 		    M_WAITOK);
5870 
5871 	/*
5872 	 * Do the allocate RPC for every DS, using a separate kernel process
5873 	 * for every DS except the last one.
5874 	 */
5875 	error = 0;
5876 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5877 		tdrpc->done = 0;
5878 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5879 		tdrpc->off = off;
5880 		tdrpc->len = len;
5881 		tdrpc->nmp = *nmpp;
5882 		tdrpc->cred = cred;
5883 		tdrpc->p = p;
5884 		tdrpc->inprog = 0;
5885 		tdrpc->err = 0;
5886 		ret = EIO;
5887 		if (nfs_pnfsiothreads != 0) {
5888 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5889 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5890 			    ret);
5891 		}
5892 		if (ret != 0) {
5893 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5894 			    cred, p);
5895 			if (nfsds_failerr(ret) && *failposp == -1)
5896 				*failposp = i;
5897 			else if (error == 0 && ret != 0)
5898 				error = ret;
5899 		}
5900 		nmpp++;
5901 		fhp++;
5902 	}
5903 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5904 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5905 		*failposp = mirrorcnt - 1;
5906 	else if (error == 0 && ret != 0)
5907 		error = ret;
5908 	if (error == 0)
5909 		error = nfsrv_setextattr(vp, &na, p);
5910 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5911 	tdrpc = drpc;
5912 	timo = hz / 50;		/* Wait for 20msec. */
5913 	if (timo < 1)
5914 		timo = 1;
5915 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5916 		/* Wait for RPCs on separate threads to complete. */
5917 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5918 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5919 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5920 			*failposp = i;
5921 		else if (error == 0 && tdrpc->err != 0)
5922 			error = tdrpc->err;
5923 	}
5924 	free(drpc, M_TEMP);
5925 	return (error);
5926 }
5927 
5928 /*
5929  * Do a deallocate RPC on a DS data file, using this structure for the
5930  * arguments, so that this function can be executed by a separate kernel
5931  * process.
5932  */
5933 struct nfsrvdeallocatedsdorpc {
5934 	int			done;
5935 	int			inprog;
5936 	struct task		tsk;
5937 	fhandle_t		fh;
5938 	off_t			off;
5939 	off_t			len;
5940 	struct nfsmount		*nmp;
5941 	struct ucred		*cred;
5942 	NFSPROC_T		*p;
5943 	int			err;
5944 };
5945 
5946 static int
nfsrv_deallocatedsdorpc(struct nfsmount * nmp,fhandle_t * fhp,off_t off,off_t len,struct nfsvattr * nap,struct ucred * cred,NFSPROC_T * p)5947 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5948     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5949 {
5950 	uint32_t *tl;
5951 	struct nfsrv_descript *nd;
5952 	nfsattrbit_t attrbits;
5953 	nfsv4stateid_t st;
5954 	int error;
5955 
5956 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5957 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5958 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5959 
5960 	/*
5961 	 * Use a stateid where other is an alternating 01010 pattern and
5962 	 * seqid is 0xffffffff.  This value is not defined as special by
5963 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5964 	 * MDS->DS proxy operation.
5965 	 */
5966 	st.other[0] = 0x55555555;
5967 	st.other[1] = 0x55555555;
5968 	st.other[2] = 0x55555555;
5969 	st.seqid = 0xffffffff;
5970 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5971 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5972 	txdr_hyper(off, tl); tl += 2;
5973 	txdr_hyper(len, tl); tl += 2;
5974 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5975 
5976 	/* Do a Getattr for the attributes that change upon writing. */
5977 	NFSZERO_ATTRBIT(&attrbits);
5978 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5979 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5980 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5981 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5982 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5983 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5984 	nfsrv_putattrbit(nd, &attrbits);
5985 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5986 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5987 	if (error != 0) {
5988 		free(nd, M_TEMP);
5989 		return (error);
5990 	}
5991 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5992 	    nd->nd_repstat);
5993 	/* Get rid of weak cache consistency data for now. */
5994 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5995 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5996 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5997 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5998 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5999 		if (error != 0)
6000 			goto nfsmout;
6001 		/*
6002 		 * Get rid of Op# and status for next op.
6003 		 */
6004 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6005 		if (*++tl != 0)
6006 			nd->nd_flag |= ND_NOMOREDATA;
6007 	}
6008 	if (nd->nd_repstat == 0) {
6009 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6010 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
6011 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
6012 	} else
6013 		error = nd->nd_repstat;
6014 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
6015 nfsmout:
6016 	m_freem(nd->nd_mrep);
6017 	free(nd, M_TEMP);
6018 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
6019 	return (error);
6020 }
6021 
6022 /*
6023  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
6024  */
6025 static void
start_deallocatedsdorpc(void * arg,int pending)6026 start_deallocatedsdorpc(void *arg, int pending)
6027 {
6028 	struct nfsrvdeallocatedsdorpc *drpc;
6029 
6030 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
6031 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
6032 	    drpc->len, NULL, drpc->cred, drpc->p);
6033 	drpc->done = 1;
6034 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
6035 }
6036 
6037 static int
nfsrv_deallocatedsrpc(fhandle_t * fhp,off_t off,off_t len,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount ** nmpp,int mirrorcnt,int * failposp)6038 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
6039     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
6040     int *failposp)
6041 {
6042 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
6043 	struct nfsvattr na;
6044 	int error, i, ret, timo;
6045 
6046 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
6047 	drpc = NULL;
6048 	if (mirrorcnt > 1)
6049 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6050 		    M_WAITOK);
6051 
6052 	/*
6053 	 * Do the deallocate RPC for every DS, using a separate kernel process
6054 	 * for every DS except the last one.
6055 	 */
6056 	error = 0;
6057 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6058 		tdrpc->done = 0;
6059 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6060 		tdrpc->off = off;
6061 		tdrpc->len = len;
6062 		tdrpc->nmp = *nmpp;
6063 		tdrpc->cred = cred;
6064 		tdrpc->p = p;
6065 		tdrpc->inprog = 0;
6066 		tdrpc->err = 0;
6067 		ret = EIO;
6068 		if (nfs_pnfsiothreads != 0) {
6069 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
6070 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
6071 			    ret);
6072 		}
6073 		if (ret != 0) {
6074 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
6075 			    NULL, cred, p);
6076 			if (nfsds_failerr(ret) && *failposp == -1)
6077 				*failposp = i;
6078 			else if (error == 0 && ret != 0)
6079 				error = ret;
6080 		}
6081 		nmpp++;
6082 		fhp++;
6083 	}
6084 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
6085 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6086 		*failposp = mirrorcnt - 1;
6087 	else if (error == 0 && ret != 0)
6088 		error = ret;
6089 	if (error == 0)
6090 		error = nfsrv_setextattr(vp, &na, p);
6091 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
6092 	tdrpc = drpc;
6093 	timo = hz / 50;		/* Wait for 20msec. */
6094 	if (timo < 1)
6095 		timo = 1;
6096 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6097 		/* Wait for RPCs on separate threads to complete. */
6098 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6099 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
6100 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6101 			*failposp = i;
6102 		else if (error == 0 && tdrpc->err != 0)
6103 			error = tdrpc->err;
6104 	}
6105 	free(drpc, M_TEMP);
6106 	return (error);
6107 }
6108 
6109 static int
nfsrv_setattrdsdorpc(fhandle_t * fhp,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount * nmp,struct nfsvattr * nap,struct nfsvattr * dsnap)6110 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6111     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
6112     struct nfsvattr *dsnap)
6113 {
6114 	uint32_t *tl;
6115 	struct nfsrv_descript *nd;
6116 	nfsv4stateid_t st;
6117 	nfsattrbit_t attrbits;
6118 	int error;
6119 
6120 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
6121 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6122 	/*
6123 	 * Use a stateid where other is an alternating 01010 pattern and
6124 	 * seqid is 0xffffffff.  This value is not defined as special by
6125 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6126 	 * MDS->DS proxy operation.
6127 	 */
6128 	st.other[0] = 0x55555555;
6129 	st.other[1] = 0x55555555;
6130 	st.other[2] = 0x55555555;
6131 	st.seqid = 0xffffffff;
6132 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6133 	    NULL, NULL, 0, 0, cred);
6134 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6135 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
6136 
6137 	/* Do a Getattr for the attributes that change due to writing. */
6138 	NFSZERO_ATTRBIT(&attrbits);
6139 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6140 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6141 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6142 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6143 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6144 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
6145 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
6146 	(void) nfsrv_putattrbit(nd, &attrbits);
6147 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6148 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6149 	if (error != 0) {
6150 		free(nd, M_TEMP);
6151 		return (error);
6152 	}
6153 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
6154 	    nd->nd_repstat);
6155 	/* Get rid of weak cache consistency data for now. */
6156 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
6157 	    (ND_NFSV4 | ND_V4WCCATTR)) {
6158 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
6159 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
6160 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
6161 		if (error != 0)
6162 			goto nfsmout;
6163 		/*
6164 		 * Get rid of Op# and status for next op.
6165 		 */
6166 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6167 		if (*++tl != 0)
6168 			nd->nd_flag |= ND_NOMOREDATA;
6169 	}
6170 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
6171 	if (error != 0)
6172 		goto nfsmout;
6173 	if (nd->nd_repstat != 0)
6174 		error = nd->nd_repstat;
6175 	/*
6176 	 * Get the Change, Size, Access Time and Modify Time attributes and set
6177 	 * on the Metadata file, so its attributes will be what the file's
6178 	 * would be if it had been written.
6179 	 */
6180 	if (error == 0) {
6181 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6182 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
6183 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL,
6184 		    NULL);
6185 	}
6186 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
6187 nfsmout:
6188 	m_freem(nd->nd_mrep);
6189 	free(nd, M_TEMP);
6190 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
6191 	return (error);
6192 }
6193 
6194 struct nfsrvsetattrdsdorpc {
6195 	int			done;
6196 	int			inprog;
6197 	struct task		tsk;
6198 	fhandle_t		fh;
6199 	struct nfsmount		*nmp;
6200 	struct vnode		*vp;
6201 	struct ucred		*cred;
6202 	NFSPROC_T		*p;
6203 	struct nfsvattr		na;
6204 	struct nfsvattr		dsna;
6205 	int			err;
6206 };
6207 
6208 /*
6209  * Start up the thread that will execute nfsrv_setattrdsdorpc().
6210  */
6211 static void
start_setattrdsdorpc(void * arg,int pending)6212 start_setattrdsdorpc(void *arg, int pending)
6213 {
6214 	struct nfsrvsetattrdsdorpc *drpc;
6215 
6216 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
6217 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
6218 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
6219 	drpc->done = 1;
6220 }
6221 
6222 static int
nfsrv_setattrdsrpc(fhandle_t * fhp,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount ** nmpp,int mirrorcnt,struct nfsvattr * nap,int * failposp)6223 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6224     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
6225     struct nfsvattr *nap, int *failposp)
6226 {
6227 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
6228 	struct nfsvattr na;
6229 	int error, i, ret, timo;
6230 
6231 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
6232 	drpc = NULL;
6233 	if (mirrorcnt > 1)
6234 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6235 		    M_WAITOK);
6236 
6237 	/*
6238 	 * Do the setattr RPC for every DS, using a separate kernel process
6239 	 * for every DS except the last one.
6240 	 */
6241 	error = 0;
6242 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6243 		tdrpc->done = 0;
6244 		tdrpc->inprog = 0;
6245 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6246 		tdrpc->nmp = *nmpp;
6247 		tdrpc->vp = vp;
6248 		tdrpc->cred = cred;
6249 		tdrpc->p = p;
6250 		tdrpc->na = *nap;
6251 		tdrpc->err = 0;
6252 		ret = EIO;
6253 		if (nfs_pnfsiothreads != 0) {
6254 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6255 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6256 			    ret);
6257 		}
6258 		if (ret != 0) {
6259 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6260 			    &na);
6261 			if (nfsds_failerr(ret) && *failposp == -1)
6262 				*failposp = i;
6263 			else if (error == 0 && ret != 0)
6264 				error = ret;
6265 		}
6266 		nmpp++;
6267 		fhp++;
6268 	}
6269 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6270 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6271 		*failposp = mirrorcnt - 1;
6272 	else if (error == 0 && ret != 0)
6273 		error = ret;
6274 	if (error == 0)
6275 		error = nfsrv_setextattr(vp, &na, p);
6276 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6277 	tdrpc = drpc;
6278 	timo = hz / 50;		/* Wait for 20msec. */
6279 	if (timo < 1)
6280 		timo = 1;
6281 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6282 		/* Wait for RPCs on separate threads to complete. */
6283 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6284 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6285 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6286 			*failposp = i;
6287 		else if (error == 0 && tdrpc->err != 0)
6288 			error = tdrpc->err;
6289 	}
6290 	free(drpc, M_TEMP);
6291 	return (error);
6292 }
6293 
6294 /*
6295  * Do a Setattr of an NFSv4 ACL on the DS file.
6296  */
6297 static int
nfsrv_setacldsdorpc(fhandle_t * fhp,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount * nmp,struct acl * aclp)6298 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6299     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6300 {
6301 	struct nfsrv_descript *nd;
6302 	nfsv4stateid_t st;
6303 	nfsattrbit_t attrbits;
6304 	int error;
6305 
6306 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6307 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6308 	/*
6309 	 * Use a stateid where other is an alternating 01010 pattern and
6310 	 * seqid is 0xffffffff.  This value is not defined as special by
6311 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6312 	 * MDS->DS proxy operation.
6313 	 */
6314 	st.other[0] = 0x55555555;
6315 	st.other[1] = 0x55555555;
6316 	st.other[2] = 0x55555555;
6317 	st.seqid = 0xffffffff;
6318 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6319 	    NULL, NULL, 0, 0, cred);
6320 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6321 	NFSZERO_ATTRBIT(&attrbits);
6322 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6323 	/*
6324 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6325 	 * so passing in the metadata "vp" will be ok, since it is of
6326 	 * the same type (VREG).
6327 	 */
6328 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6329 	    NULL, 0, 0, 0, 0, 0, NULL);
6330 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6331 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6332 	if (error != 0) {
6333 		free(nd, M_TEMP);
6334 		return (error);
6335 	}
6336 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6337 	    nd->nd_repstat);
6338 	error = nd->nd_repstat;
6339 	m_freem(nd->nd_mrep);
6340 	free(nd, M_TEMP);
6341 	return (error);
6342 }
6343 
6344 struct nfsrvsetacldsdorpc {
6345 	int			done;
6346 	int			inprog;
6347 	struct task		tsk;
6348 	fhandle_t		fh;
6349 	struct nfsmount		*nmp;
6350 	struct vnode		*vp;
6351 	struct ucred		*cred;
6352 	NFSPROC_T		*p;
6353 	struct acl		*aclp;
6354 	int			err;
6355 };
6356 
6357 /*
6358  * Start up the thread that will execute nfsrv_setacldsdorpc().
6359  */
6360 static void
start_setacldsdorpc(void * arg,int pending)6361 start_setacldsdorpc(void *arg, int pending)
6362 {
6363 	struct nfsrvsetacldsdorpc *drpc;
6364 
6365 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6366 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6367 	    drpc->vp, drpc->nmp, drpc->aclp);
6368 	drpc->done = 1;
6369 }
6370 
6371 static int
nfsrv_setacldsrpc(fhandle_t * fhp,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount ** nmpp,int mirrorcnt,struct acl * aclp,int * failposp)6372 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6373     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6374     int *failposp)
6375 {
6376 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6377 	int error, i, ret, timo;
6378 
6379 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6380 	drpc = NULL;
6381 	if (mirrorcnt > 1)
6382 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6383 		    M_WAITOK);
6384 
6385 	/*
6386 	 * Do the setattr RPC for every DS, using a separate kernel process
6387 	 * for every DS except the last one.
6388 	 */
6389 	error = 0;
6390 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6391 		tdrpc->done = 0;
6392 		tdrpc->inprog = 0;
6393 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6394 		tdrpc->nmp = *nmpp;
6395 		tdrpc->vp = vp;
6396 		tdrpc->cred = cred;
6397 		tdrpc->p = p;
6398 		tdrpc->aclp = aclp;
6399 		tdrpc->err = 0;
6400 		ret = EIO;
6401 		if (nfs_pnfsiothreads != 0) {
6402 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6403 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6404 			    ret);
6405 		}
6406 		if (ret != 0) {
6407 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6408 			    aclp);
6409 			if (nfsds_failerr(ret) && *failposp == -1)
6410 				*failposp = i;
6411 			else if (error == 0 && ret != 0)
6412 				error = ret;
6413 		}
6414 		nmpp++;
6415 		fhp++;
6416 	}
6417 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6418 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6419 		*failposp = mirrorcnt - 1;
6420 	else if (error == 0 && ret != 0)
6421 		error = ret;
6422 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6423 	tdrpc = drpc;
6424 	timo = hz / 50;		/* Wait for 20msec. */
6425 	if (timo < 1)
6426 		timo = 1;
6427 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6428 		/* Wait for RPCs on separate threads to complete. */
6429 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6430 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6431 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6432 			*failposp = i;
6433 		else if (error == 0 && tdrpc->err != 0)
6434 			error = tdrpc->err;
6435 	}
6436 	free(drpc, M_TEMP);
6437 	return (error);
6438 }
6439 
6440 /*
6441  * Getattr call to the DS for the attributes that change due to writing.
6442  */
6443 static int
nfsrv_getattrdsrpc(fhandle_t * fhp,struct ucred * cred,NFSPROC_T * p,struct vnode * vp,struct nfsmount * nmp,struct nfsvattr * nap)6444 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6445     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6446 {
6447 	struct nfsrv_descript *nd;
6448 	int error;
6449 	nfsattrbit_t attrbits;
6450 
6451 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6452 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6453 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6454 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6455 	NFSZERO_ATTRBIT(&attrbits);
6456 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6457 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6458 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6459 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6460 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6461 	(void) nfsrv_putattrbit(nd, &attrbits);
6462 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6463 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6464 	if (error != 0) {
6465 		free(nd, M_TEMP);
6466 		return (error);
6467 	}
6468 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6469 	    nd->nd_repstat);
6470 	if (nd->nd_repstat == 0) {
6471 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6472 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6473 		    NULL, NULL, NULL);
6474 		/*
6475 		 * We can only save the updated values in the extended
6476 		 * attribute if the vp is exclusively locked.
6477 		 * This should happen when any of the following operations
6478 		 * occur on the vnode:
6479 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6480 		 * As such, the updated extended attribute should get saved
6481 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6482 		 * attributes to be returned without calling this function.
6483 		 */
6484 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6485 			error = nfsrv_setextattr(vp, nap, p);
6486 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6487 			    error);
6488 		}
6489 	} else
6490 		error = nd->nd_repstat;
6491 	m_freem(nd->nd_mrep);
6492 	free(nd, M_TEMP);
6493 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6494 	return (error);
6495 }
6496 
6497 /*
6498  * Seek call to a DS.
6499  */
6500 static int
nfsrv_seekdsrpc(fhandle_t * fhp,off_t * offp,int content,bool * eofp,struct ucred * cred,NFSPROC_T * p,struct nfsmount * nmp)6501 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6502     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6503 {
6504 	uint32_t *tl;
6505 	struct nfsrv_descript *nd;
6506 	nfsv4stateid_t st;
6507 	int error;
6508 
6509 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6510 	/*
6511 	 * Use a stateid where other is an alternating 01010 pattern and
6512 	 * seqid is 0xffffffff.  This value is not defined as special by
6513 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6514 	 * MDS->DS proxy operation.
6515 	 */
6516 	st.other[0] = 0x55555555;
6517 	st.other[1] = 0x55555555;
6518 	st.other[2] = 0x55555555;
6519 	st.seqid = 0xffffffff;
6520 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6521 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6522 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6523 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6524 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6525 	txdr_hyper(*offp, tl); tl += 2;
6526 	*tl = txdr_unsigned(content);
6527 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6528 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6529 	if (error != 0) {
6530 		free(nd, M_TEMP);
6531 		return (error);
6532 	}
6533 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6534 	if (nd->nd_repstat == 0) {
6535 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6536 		if (*tl++ == newnfs_true)
6537 			*eofp = true;
6538 		else
6539 			*eofp = false;
6540 		*offp = fxdr_hyper(tl);
6541 	} else
6542 		error = nd->nd_repstat;
6543 nfsmout:
6544 	m_freem(nd->nd_mrep);
6545 	free(nd, M_TEMP);
6546 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6547 	return (error);
6548 }
6549 
6550 /*
6551  * Get the device id and file handle for a DS file.
6552  */
6553 int
nfsrv_dsgetdevandfh(struct vnode * vp,NFSPROC_T * p,int * mirrorcntp,fhandle_t * fhp,char * devid)6554 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6555     fhandle_t *fhp, char *devid)
6556 {
6557 	int buflen, error;
6558 	char *buf;
6559 
6560 	buflen = 1024;
6561 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6562 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6563 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6564 	free(buf, M_TEMP);
6565 	return (error);
6566 }
6567 
6568 /*
6569  * Do a Lookup against the DS for the filename.
6570  */
6571 static int
nfsrv_pnfslookupds(struct vnode * vp,struct vnode * dvp,struct pnfsdsfile * pf,struct vnode ** nvpp,NFSPROC_T * p)6572 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6573     struct vnode **nvpp, NFSPROC_T *p)
6574 {
6575 	struct nameidata named;
6576 	struct ucred *tcred;
6577 	char *bufp;
6578 	u_long *hashp;
6579 	struct vnode *nvp;
6580 	int error;
6581 
6582 	tcred = newnfs_getcred();
6583 	named.ni_cnd.cn_nameiop = LOOKUP;
6584 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6585 	named.ni_cnd.cn_cred = tcred;
6586 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
6587 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6588 	named.ni_cnd.cn_nameptr = bufp;
6589 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6590 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6591 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6592 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6593 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6594 	NFSFREECRED(tcred);
6595 	nfsvno_relpathbuf(&named);
6596 	if (error == 0)
6597 		*nvpp = nvp;
6598 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6599 	return (error);
6600 }
6601 
6602 /*
6603  * Set the file handle to the correct one.
6604  */
6605 static void
nfsrv_pnfssetfh(struct vnode * vp,struct pnfsdsfile * pf,char * devid,char * fnamep,struct vnode * nvp,NFSPROC_T * p)6606 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6607     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6608 {
6609 	struct nfsnode *np;
6610 	int ret = 0;
6611 
6612 	np = VTONFS(nvp);
6613 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6614 	/*
6615 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6616 	 * locked and vn_start_write() has been done.  If devid != NULL or
6617 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6618 	 * may not have been done.
6619 	 * If not done now, it will be done on a future call.
6620 	 */
6621 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6622 	    LK_EXCLUSIVE)
6623 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6624 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6625 		    (char *)pf, p);
6626 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6627 }
6628 
6629 /*
6630  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6631  * when the DS has failed.
6632  */
6633 void
nfsrv_killrpcs(struct nfsmount * nmp)6634 nfsrv_killrpcs(struct nfsmount *nmp)
6635 {
6636 
6637 	/*
6638 	 * Call newnfs_nmcancelreqs() to cause
6639 	 * any RPCs in progress on the mount point to
6640 	 * fail.
6641 	 * This will cause any process waiting for an
6642 	 * RPC to complete while holding a vnode lock
6643 	 * on the mounted-on vnode (such as "df" or
6644 	 * a non-forced "umount") to fail.
6645 	 * This will unlock the mounted-on vnode so
6646 	 * a forced dismount can succeed.
6647 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6648 	 * called.
6649 	 */
6650 	newnfs_nmcancelreqs(nmp);
6651 }
6652 
6653 /*
6654  * Sum up the statfs info for each of the DSs, so that the client will
6655  * receive the total for all DSs.
6656  */
6657 static int
nfsrv_pnfsstatfs(struct statfs * sf,struct mount * mp)6658 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6659 {
6660 	struct statfs *tsf;
6661 	struct nfsdevice *ds;
6662 	struct vnode **dvpp, **tdvpp, *dvp;
6663 	uint64_t tot;
6664 	int cnt, error = 0, i;
6665 
6666 	if (nfsrv_devidcnt <= 0)
6667 		return (ENXIO);
6668 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6669 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6670 
6671 	/* Get an array of the dvps for the DSs. */
6672 	tdvpp = dvpp;
6673 	i = 0;
6674 	NFSDDSLOCK();
6675 	/* First, search for matches for same file system. */
6676 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6677 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6678 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6679 			if (++i > nfsrv_devidcnt)
6680 				break;
6681 			*tdvpp++ = ds->nfsdev_dvp;
6682 		}
6683 	}
6684 	/*
6685 	 * If no matches for same file system, total all servers not assigned
6686 	 * to a file system.
6687 	 */
6688 	if (i == 0) {
6689 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6690 			if (ds->nfsdev_nmp != NULL &&
6691 			    ds->nfsdev_mdsisset == 0) {
6692 				if (++i > nfsrv_devidcnt)
6693 					break;
6694 				*tdvpp++ = ds->nfsdev_dvp;
6695 			}
6696 		}
6697 	}
6698 	NFSDDSUNLOCK();
6699 	cnt = i;
6700 
6701 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6702 	tdvpp = dvpp;
6703 	for (i = 0; i < cnt && error == 0; i++) {
6704 		dvp = *tdvpp++;
6705 		error = VFS_STATFS(dvp->v_mount, tsf);
6706 		if (error == 0) {
6707 			if (sf->f_bsize == 0) {
6708 				if (tsf->f_bsize > 0)
6709 					sf->f_bsize = tsf->f_bsize;
6710 				else
6711 					sf->f_bsize = 8192;
6712 			}
6713 			if (tsf->f_blocks > 0) {
6714 				if (sf->f_bsize != tsf->f_bsize) {
6715 					tot = tsf->f_blocks * tsf->f_bsize;
6716 					sf->f_blocks += (tot / sf->f_bsize);
6717 				} else
6718 					sf->f_blocks += tsf->f_blocks;
6719 			}
6720 			if (tsf->f_bfree > 0) {
6721 				if (sf->f_bsize != tsf->f_bsize) {
6722 					tot = tsf->f_bfree * tsf->f_bsize;
6723 					sf->f_bfree += (tot / sf->f_bsize);
6724 				} else
6725 					sf->f_bfree += tsf->f_bfree;
6726 			}
6727 			if (tsf->f_bavail > 0) {
6728 				if (sf->f_bsize != tsf->f_bsize) {
6729 					tot = tsf->f_bavail * tsf->f_bsize;
6730 					sf->f_bavail += (tot / sf->f_bsize);
6731 				} else
6732 					sf->f_bavail += tsf->f_bavail;
6733 			}
6734 		}
6735 	}
6736 	free(tsf, M_TEMP);
6737 	free(dvpp, M_TEMP);
6738 	return (error);
6739 }
6740 
6741 /*
6742  * Set an NFSv4 acl.
6743  */
6744 int
nfsrv_setacl(struct vnode * vp,NFSACL_T * aclp,struct ucred * cred,NFSPROC_T * p)6745 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6746 {
6747 	int error;
6748 
6749 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6750 		error = NFSERR_ATTRNOTSUPP;
6751 		goto out;
6752 	}
6753 	/*
6754 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6755 	 * Make sure it has enough room for that - splitting every entry
6756 	 * into two and appending "canonical six" entries at the end.
6757 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6758 	 */
6759 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6760 		error = NFSERR_ATTRNOTSUPP;
6761 		goto out;
6762 	}
6763 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6764 	if (error == 0) {
6765 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6766 		if (error == ENOENT)
6767 			error = 0;
6768 	}
6769 
6770 out:
6771 	NFSEXITCODE(error);
6772 	return (error);
6773 }
6774 
6775 /*
6776  * Seek vnode op call (actually it is a VOP_IOCTL()).
6777  * This function is called with the vnode locked, but unlocks and vrele()s
6778  * the vp before returning.
6779  */
6780 int
nfsvno_seek(struct nfsrv_descript * nd,struct vnode * vp,u_long cmd,off_t * offp,int content,bool * eofp,struct ucred * cred,NFSPROC_T * p)6781 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6782     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6783 {
6784 	struct nfsvattr at;
6785 	int error, ret;
6786 
6787 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6788 	/*
6789 	 * Attempt to seek on a DS file. A return of ENOENT implies
6790 	 * there is no DS file to seek on.
6791 	 */
6792 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6793 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6794 	if (error != ENOENT) {
6795 		vput(vp);
6796 		return (error);
6797 	}
6798 
6799 	/*
6800 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6801 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6802 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6803 	 */
6804 	NFSVOPUNLOCK(vp);
6805 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6806 	*eofp = false;
6807 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6808 		/* Handle the cases where we might be at EOF. */
6809 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6810 		if (ret == 0 && *offp == at.na_size) {
6811 			*eofp = true;
6812 			error = 0;
6813 		}
6814 		if (ret != 0 && error == 0)
6815 			error = ret;
6816 	}
6817 	vrele(vp);
6818 	NFSEXITCODE(error);
6819 	return (error);
6820 }
6821 
6822 /*
6823  * Allocate vnode op call.
6824  */
6825 int
nfsvno_allocate(struct vnode * vp,off_t off,off_t len,struct ucred * cred,NFSPROC_T * p)6826 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6827     NFSPROC_T *p)
6828 {
6829 	int error;
6830 	off_t olen;
6831 
6832 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6833 	/*
6834 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6835 	 * there is no DS file to allocate on.
6836 	 */
6837 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6838 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6839 	if (error != ENOENT)
6840 		return (error);
6841 
6842 	/*
6843 	 * Do the actual VOP_ALLOCATE(), looping so long as
6844 	 * progress is being made, to achieve completion.
6845 	 */
6846 	do {
6847 		olen = len;
6848 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6849 		if (error == 0 && len > 0 && olen > len)
6850 			maybe_yield();
6851 	} while (error == 0 && len > 0 && olen > len);
6852 	if (error == 0 && len > 0)
6853 		error = NFSERR_IO;
6854 	NFSEXITCODE(error);
6855 	return (error);
6856 }
6857 
6858 /*
6859  * Deallocate vnode op call.
6860  */
6861 int
nfsvno_deallocate(struct vnode * vp,off_t off,off_t len,struct ucred * cred,NFSPROC_T * p)6862 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6863     NFSPROC_T *p)
6864 {
6865 	int error;
6866 	off_t olen;
6867 
6868 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6869 	/*
6870 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6871 	 * there is no DS file to deallocate on.
6872 	 */
6873 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6874 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6875 	if (error != ENOENT)
6876 		return (error);
6877 
6878 	/*
6879 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6880 	 * progress is being made, to achieve completion.
6881 	 */
6882 	do {
6883 		olen = len;
6884 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6885 		if (error == 0 && len > 0 && olen > len)
6886 			maybe_yield();
6887 	} while (error == 0 && len > 0 && olen > len);
6888 	if (error == 0 && len > 0)
6889 		error = NFSERR_IO;
6890 	NFSEXITCODE(error);
6891 	return (error);
6892 }
6893 
6894 /*
6895  * Get Extended Atribute vnode op into an mbuf list.
6896  */
6897 int
nfsvno_getxattr(struct vnode * vp,char * name,uint32_t maxresp,struct ucred * cred,uint64_t flag,int maxextsiz,struct thread * p,struct mbuf ** mpp,struct mbuf ** mpendp,int * lenp)6898 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6899     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6900     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6901 {
6902 	struct iovec *iv;
6903 	struct uio io, *uiop = &io;
6904 	struct mbuf *m, *m2;
6905 	int alen, error, len, tlen;
6906 	size_t siz;
6907 
6908 	/* First, find out the size of the extended attribute. */
6909 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6910 	    &siz, cred, p);
6911 	if (error != 0)
6912 		return (NFSERR_NOXATTR);
6913 	if (siz > maxresp - NFS_MAXXDR)
6914 		return (NFSERR_XATTR2BIG);
6915 	len = siz;
6916 	tlen = NFSM_RNDUP(len);
6917 	if (tlen > 0) {
6918 		/*
6919 		 * If cnt > MCLBYTES and the reply will not be saved, use
6920 		 * ext_pgs mbufs for TLS.
6921 		 * For NFSv4.0, we do not know for sure if the reply will
6922 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6923 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6924 		 */
6925 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6926 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6927 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6928 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6929 			    maxextsiz, &m, &m2, &iv);
6930 		else
6931 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6932 			    &iv);
6933 		uiop->uio_iov = iv;
6934 	} else {
6935 		uiop->uio_iovcnt = 0;
6936 		uiop->uio_iov = iv = NULL;
6937 		m = m2 = NULL;
6938 	}
6939 	uiop->uio_offset = 0;
6940 	uiop->uio_resid = tlen;
6941 	uiop->uio_rw = UIO_READ;
6942 	uiop->uio_segflg = UIO_SYSSPACE;
6943 	uiop->uio_td = p;
6944 #ifdef MAC
6945 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6946 	    name);
6947 	if (error != 0)
6948 		goto out;
6949 #endif
6950 
6951 	if (tlen > 0)
6952 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6953 		    NULL, cred, p);
6954 	if (error != 0)
6955 		goto out;
6956 	if (uiop->uio_resid > 0) {
6957 		alen = tlen;
6958 		len = tlen - uiop->uio_resid;
6959 		tlen = NFSM_RNDUP(len);
6960 		if (alen != tlen)
6961 			printf("nfsvno_getxattr: weird size read\n");
6962 		if (tlen == 0) {
6963 			m_freem(m);
6964 			m = m2 = NULL;
6965 		} else if (alen != tlen || tlen != len)
6966 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6967 	}
6968 	*lenp = len;
6969 	*mpp = m;
6970 	*mpendp = m2;
6971 
6972 out:
6973 	if (error != 0) {
6974 		if (m != NULL)
6975 			m_freem(m);
6976 		*lenp = 0;
6977 	}
6978 	free(iv, M_TEMP);
6979 	NFSEXITCODE(error);
6980 	return (error);
6981 }
6982 
6983 /*
6984  * Set Extended attribute vnode op from an mbuf list.
6985  */
6986 int
nfsvno_setxattr(struct vnode * vp,char * name,int len,struct mbuf * m,char * cp,struct ucred * cred,struct thread * p)6987 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6988     char *cp, struct ucred *cred, struct thread *p)
6989 {
6990 	struct iovec *iv;
6991 	struct uio uio, *uiop = &uio;
6992 	int cnt, error;
6993 
6994 	error = 0;
6995 #ifdef MAC
6996 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6997 	    name);
6998 #endif
6999 	if (error != 0)
7000 		goto out;
7001 
7002 	uiop->uio_rw = UIO_WRITE;
7003 	uiop->uio_segflg = UIO_SYSSPACE;
7004 	uiop->uio_td = p;
7005 	uiop->uio_offset = 0;
7006 	uiop->uio_resid = len;
7007 	if (len > 0) {
7008 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
7009 		uiop->uio_iov = iv;
7010 		uiop->uio_iovcnt = cnt;
7011 	} else {
7012 		uiop->uio_iov = iv = NULL;
7013 		uiop->uio_iovcnt = 0;
7014 	}
7015 	if (error == 0) {
7016 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
7017 		    cred, p);
7018 		if (error == 0) {
7019 			if (vp->v_type == VREG && nfsrv_devidcnt != 0)
7020 				nfsvno_updateds(vp, cred, p);
7021 			error = VOP_FSYNC(vp, MNT_WAIT, p);
7022 		}
7023 		free(iv, M_TEMP);
7024 	}
7025 
7026 out:
7027 	NFSEXITCODE(error);
7028 	return (error);
7029 }
7030 
7031 /*
7032  * For a pNFS server, the DS file's ctime and
7033  * va_filerev (TimeMetadata and Change) needs to
7034  * be updated.  This is a hack, but works by
7035  * flipping the S_ISGID bit in va_mode and then
7036  * flipping it back.
7037  * It does result in two MDS->DS RPCs, but creating
7038  * a custom RPC just to do this seems overkill, since
7039  * Setxattr/Rmxattr will not be done that frequently.
7040  * If it fails part way through, that is not too
7041  * serious, since the DS file is never executed.
7042  */
7043 static void
nfsvno_updateds(struct vnode * vp,struct ucred * cred,NFSPROC_T * p)7044 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p)
7045 {
7046 	struct nfsvattr nva;
7047 	int ret;
7048 	u_short tmode;
7049 
7050 	ret = VOP_GETATTR(vp, &nva.na_vattr, cred);
7051 	if (ret == 0) {
7052 		tmode = nva.na_mode;
7053 		NFSVNO_ATTRINIT(&nva);
7054 		tmode ^= S_ISGID;
7055 		NFSVNO_SETATTRVAL(&nva, mode, tmode);
7056 		ret = nfsrv_proxyds(vp, 0, 0, cred, p,
7057 		    NFSPROC_SETATTR, NULL, NULL, NULL, &nva,
7058 		    NULL, NULL, 0, NULL);
7059 		if (ret == 0) {
7060 			tmode ^= S_ISGID;
7061 			NFSVNO_SETATTRVAL(&nva, mode, tmode);
7062 			ret = nfsrv_proxyds(vp, 0, 0, cred, p,
7063 			    NFSPROC_SETATTR, NULL, NULL, NULL,
7064 			    &nva, NULL, NULL, 0, NULL);
7065 		}
7066 	}
7067 }
7068 
7069 /*
7070  * Remove Extended attribute vnode op.
7071  */
7072 int
nfsvno_rmxattr(struct nfsrv_descript * nd,struct vnode * vp,char * name,struct ucred * cred,struct thread * p)7073 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
7074     struct ucred *cred, struct thread *p)
7075 {
7076 	int error;
7077 
7078 	/*
7079 	 * Get rid of any delegations.  I am not sure why this is required,
7080 	 * but RFC-8276 says so.
7081 	 */
7082 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
7083 	if (error != 0)
7084 		goto out;
7085 #ifdef MAC
7086 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
7087 	    name);
7088 	if (error != 0)
7089 		goto out;
7090 #endif
7091 
7092 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
7093 	if (error == EOPNOTSUPP)
7094 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
7095 		    cred, p);
7096 	if (error == 0) {
7097 		if (vp->v_type == VREG && nfsrv_devidcnt != 0)
7098 			nfsvno_updateds(vp, cred, p);
7099 		error = VOP_FSYNC(vp, MNT_WAIT, p);
7100 	}
7101 out:
7102 	NFSEXITCODE(error);
7103 	return (error);
7104 }
7105 
7106 /*
7107  * List Extended Atribute vnode op into an mbuf list.
7108  */
7109 int
nfsvno_listxattr(struct vnode * vp,uint64_t cookie,struct ucred * cred,struct thread * p,u_char ** bufp,uint32_t * lenp,bool * eofp)7110 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
7111     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
7112 {
7113 	struct iovec iv;
7114 	struct uio io;
7115 	int error;
7116 	size_t siz;
7117 
7118 	*bufp = NULL;
7119 	/* First, find out the size of the extended attribute. */
7120 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
7121 	    p);
7122 	if (error != 0)
7123 		return (NFSERR_NOXATTR);
7124 	if (siz <= cookie) {
7125 		*lenp = 0;
7126 		*eofp = true;
7127 		goto out;
7128 	}
7129 	if (siz > cookie + *lenp) {
7130 		siz = cookie + *lenp;
7131 		*eofp = false;
7132 	} else
7133 		*eofp = true;
7134 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
7135 	if (siz > 10 * 1024 * 1024) {
7136 		error = NFSERR_XATTR2BIG;
7137 		goto out;
7138 	}
7139 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
7140 	iv.iov_base = *bufp;
7141 	iv.iov_len = siz;
7142 	io.uio_iovcnt = 1;
7143 	io.uio_iov = &iv;
7144 	io.uio_offset = 0;
7145 	io.uio_resid = siz;
7146 	io.uio_rw = UIO_READ;
7147 	io.uio_segflg = UIO_SYSSPACE;
7148 	io.uio_td = p;
7149 #ifdef MAC
7150 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
7151 	if (error != 0)
7152 		goto out;
7153 #endif
7154 
7155 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
7156 	    p);
7157 	if (error != 0)
7158 		goto out;
7159 	if (io.uio_resid > 0)
7160 		siz -= io.uio_resid;
7161 	*lenp = siz;
7162 
7163 out:
7164 	if (error != 0) {
7165 		free(*bufp, M_TEMP);
7166 		*bufp = NULL;
7167 	}
7168 	NFSEXITCODE(error);
7169 	return (error);
7170 }
7171 
7172 /*
7173  * Trim trailing data off the mbuf list being built.
7174  */
7175 void
nfsm_trimtrailing(struct nfsrv_descript * nd,struct mbuf * mb,char * bpos,int bextpg,int bextpgsiz)7176 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
7177     int bextpg, int bextpgsiz)
7178 {
7179 	vm_page_t pg;
7180 	int fullpgsiz, i;
7181 
7182 	if (mb->m_next != NULL) {
7183 		m_freem(mb->m_next);
7184 		mb->m_next = NULL;
7185 	}
7186 	if ((mb->m_flags & M_EXTPG) != 0) {
7187 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
7188 		    ("nfsm_trimtrailing: bextpg out of range"));
7189 		KASSERT(bpos == (char *)(void *)
7190 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
7191 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
7192 
7193 		/* First, get rid of any pages after this position. */
7194 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
7195 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
7196 			vm_page_unwire_noq(pg);
7197 			vm_page_free(pg);
7198 		}
7199 		mb->m_epg_npgs = bextpg + 1;
7200 		if (bextpg == 0)
7201 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
7202 		else
7203 			fullpgsiz = PAGE_SIZE;
7204 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
7205 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
7206 		for (i = 1; i < mb->m_epg_npgs; i++)
7207 			mb->m_len += m_epg_pagelen(mb, i, 0);
7208 		nd->nd_bextpgsiz = bextpgsiz;
7209 		nd->nd_bextpg = bextpg;
7210 	} else
7211 		mb->m_len = bpos - mtod(mb, char *);
7212 	nd->nd_mb = mb;
7213 	nd->nd_bpos = bpos;
7214 }
7215 
7216 
7217 /*
7218  * Check to see if a put file handle operation should test for
7219  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
7220  * When Open is the next operation, NFSERR_WRONGSEC cannot be
7221  * replied for the Open cases that use a component.  This can
7222  * be identified by the fact that the file handle's type is VDIR.
7223  */
7224 bool
nfsrv_checkwrongsec(struct nfsrv_descript * nd,int nextop,__enum_uint8 (vtype)vtyp)7225 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp)
7226 {
7227 
7228 	if ((nd->nd_flag & ND_NFSV4) == 0)
7229 		return (true);
7230 
7231 	if ((nd->nd_flag & ND_LASTOP) != 0)
7232 		return (false);
7233 
7234 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
7235 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
7236 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
7237 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
7238 		return (false);
7239 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
7240 		return (false);
7241 	return (true);
7242 }
7243 
7244 /*
7245  * Check DSs marked no space.
7246  */
7247 void
nfsrv_checknospc(void)7248 nfsrv_checknospc(void)
7249 {
7250 	struct statfs *tsf;
7251 	struct nfsdevice *ds;
7252 	struct vnode **dvpp, **tdvpp, *dvp;
7253 	char *devid, *tdevid;
7254 	int cnt, error = 0, i;
7255 
7256 	if (nfsrv_devidcnt <= 0)
7257 		return;
7258 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
7259 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
7260 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
7261 
7262 	/* Get an array of the dvps for the DSs. */
7263 	tdvpp = dvpp;
7264 	tdevid = devid;
7265 	i = 0;
7266 	NFSDDSLOCK();
7267 	/* First, search for matches for same file system. */
7268 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7269 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
7270 			if (++i > nfsrv_devidcnt)
7271 				break;
7272 			*tdvpp++ = ds->nfsdev_dvp;
7273 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
7274 			tdevid += NFSX_V4DEVICEID;
7275 		}
7276 	}
7277 	NFSDDSUNLOCK();
7278 
7279 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7280 	cnt = i;
7281 	tdvpp = dvpp;
7282 	tdevid = devid;
7283 	for (i = 0; i < cnt && error == 0; i++) {
7284 		dvp = *tdvpp++;
7285 		error = VFS_STATFS(dvp->v_mount, tsf);
7286 		if (error == 0 && tsf->f_bavail > 0) {
7287 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7288 			nfsrv_marknospc(tdevid, false);
7289 		}
7290 		tdevid += NFSX_V4DEVICEID;
7291 	}
7292 	free(tsf, M_TEMP);
7293 	free(dvpp, M_TEMP);
7294 	free(devid, M_TEMP);
7295 }
7296 
7297 /*
7298  * Initialize everything that needs to be initialized for a vnet.
7299  */
7300 static void
nfsrv_vnetinit(const void * unused __unused)7301 nfsrv_vnetinit(const void *unused __unused)
7302 {
7303 
7304 	nfsd_mntinit();
7305 }
7306 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7307     nfsrv_vnetinit, NULL);
7308 
7309 /*
7310  * Clean up everything that is in a vnet and needs to be
7311  * done when the jail is destroyed or the module unloaded.
7312  */
7313 static void
nfsrv_cleanup(const void * unused __unused)7314 nfsrv_cleanup(const void *unused __unused)
7315 {
7316 	int i;
7317 
7318 	NFSD_LOCK();
7319 	if (!NFSD_VNET(nfsrv_mntinited)) {
7320 		NFSD_UNLOCK();
7321 		return;
7322 	}
7323 	NFSD_VNET(nfsrv_mntinited) = false;
7324 	NFSD_UNLOCK();
7325 
7326 	/* Clean out all NFSv4 state. */
7327 	nfsrv_throwawayallstate(curthread);
7328 
7329 	/* Clean the NFS server reply cache */
7330 	nfsrvd_cleancache();
7331 
7332 	/* Clean out v4root exports. */
7333 	if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) {
7334 		vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export);
7335 		free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT);
7336 		NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
7337 	}
7338 
7339 	/* Free up the krpc server pool. */
7340 	if (NFSD_VNET(nfsrvd_pool) != NULL)
7341 		svcpool_destroy(NFSD_VNET(nfsrvd_pool));
7342 
7343 	/* and get rid of the locks */
7344 	for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7345 		mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx);
7346 		mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx);
7347 	}
7348 	mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx);
7349 	for (i = 0; i < nfsrv_sessionhashsize; i++)
7350 		mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx);
7351 	lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock);
7352 	free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE);
7353 	free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE);
7354 	free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE);
7355 	free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT);
7356 	free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE);
7357 	free(NFSD_VNET(nfssessionhash), M_NFSDSESSION);
7358 	free(NFSD_VNET(nfsv4root_mnt), M_TEMP);
7359 	NFSD_VNET(nfsv4root_mnt) = NULL;
7360 }
7361 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7362     nfsrv_cleanup, NULL);
7363 
7364 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7365 
7366 /*
7367  * Called once to initialize data structures...
7368  */
7369 static int
nfsd_modevent(module_t mod,int type,void * data)7370 nfsd_modevent(module_t mod, int type, void *data)
7371 {
7372 	int error = 0, i;
7373 	static int loaded = 0;
7374 
7375 	switch (type) {
7376 	case MOD_LOAD:
7377 		if (loaded)
7378 			goto out;
7379 		newnfs_portinit();
7380 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7381 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7382 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7383 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7384 #ifdef VV_DISABLEDELEG
7385 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7386 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7387 #endif
7388 		nfsd_call_nfsd = nfssvc_nfsd;
7389 		loaded = 1;
7390 		break;
7391 
7392 	case MOD_UNLOAD:
7393 		if (newnfs_numnfsd != 0) {
7394 			error = EBUSY;
7395 			break;
7396 		}
7397 
7398 #ifdef VV_DISABLEDELEG
7399 		vn_deleg_ops.vndeleg_recall = NULL;
7400 		vn_deleg_ops.vndeleg_disable = NULL;
7401 #endif
7402 		nfsd_call_nfsd = NULL;
7403 		mtx_destroy(&nfsrc_udpmtx);
7404 		mtx_destroy(&nfs_v4root_mutex);
7405 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7406 		mtx_destroy(&nfsrv_recalllock_mtx);
7407 		if (nfslayouthash != NULL) {
7408 			for (i = 0; i < nfsrv_layouthashsize; i++)
7409 				mtx_destroy(&nfslayouthash[i].mtx);
7410 			free(nfslayouthash, M_NFSDSESSION);
7411 		}
7412 		loaded = 0;
7413 		break;
7414 	default:
7415 		error = EOPNOTSUPP;
7416 		break;
7417 	}
7418 
7419 out:
7420 	NFSEXITCODE(error);
7421 	return (error);
7422 }
7423 static moduledata_t nfsd_mod = {
7424 	"nfsd",
7425 	nfsd_modevent,
7426 	NULL,
7427 };
7428 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7429 
7430 /* So that loader and kldload(2) can find us, wherever we are.. */
7431 MODULE_VERSION(nfsd, 1);
7432 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7433 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7434 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7435 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7436