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