2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
43 * Functions that perform the vfs operations required by the routines in
44 * nfsd_serv.c. It is hoped that this change will make the server more
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/filio.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
56 FEATURE(nfsd, "NFSv4 server");
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern struct mount nfsv4root_mnt;
62 extern struct nfsrv_stablefirst nfsrv_stablefirst;
63 extern void (*nfsd_call_servertimer)(void);
64 extern SVCPOOL *nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
82 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
83 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
84 struct mtx nfsrc_udpmtx;
85 struct mtx nfs_v4root_mutex;
86 struct mtx nfsrv_dontlistlock_mtx;
87 struct mtx nfsrv_recalllock_mtx;
88 struct nfsrvfh nfs_rootfh, nfs_pubfh;
89 int nfs_pubfhset = 0, nfs_rootfhset = 0;
90 struct proc *nfsd_master_proc = NULL;
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;
98 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
101 int nfsrv_enable_crossmntpt = 1;
102 static int nfs_commit_blks;
103 static int nfs_commit_miss;
104 extern int nfsrv_issuedelegs;
105 extern int nfsrv_dolocallocks;
106 extern int nfsd_enable_stringtouid;
107 extern struct nfsdevicehead nfsrv_devidhead;
109 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
111 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
113 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
115 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
116 int *, char *, fhandle_t *);
117 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
119 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
120 struct thread *, int, struct mbuf **, char *, struct mbuf **,
121 struct nfsvattr *, struct acl *, off_t *, int, bool *);
122 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
123 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
124 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
125 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
126 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
128 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
129 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
130 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
131 struct vnode *, struct nfsmount **, int, struct acl *, int *);
132 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
133 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
134 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
135 struct vnode *, struct nfsmount *, struct nfsvattr *);
136 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
137 NFSPROC_T *, struct nfsmount *);
138 static int nfsrv_putfhname(fhandle_t *, char *);
139 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
140 struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
141 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
142 struct vnode *, NFSPROC_T *);
143 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
144 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
146 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
148 int nfs_pnfsio(task_fn_t *, void *);
150 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
152 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
153 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
154 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
158 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
159 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
160 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
161 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
162 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
163 0, "Debug level for NFS server");
164 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
165 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
166 static int nfsrv_pnfsgetdsattr = 1;
167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
168 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
171 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
173 * The dsN subdirectories for the increased values must have been created
174 * on all DS servers before this increase is done.
176 u_int nfsrv_dsdirsize = 20;
178 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
180 int error, newdsdirsize;
182 newdsdirsize = nfsrv_dsdirsize;
183 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
184 if (error != 0 || req->newptr == NULL)
186 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
189 nfsrv_dsdirsize = newdsdirsize;
192 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
193 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
194 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
196 #define MAX_REORDERED_RPC 16
197 #define NUM_HEURISTIC 1031
198 #define NHUSE_INIT 64
200 #define NHUSE_MAX 2048
202 static struct nfsheur {
203 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
204 off_t nh_nextoff; /* next offset for sequential detection */
205 int nh_use; /* use count for selection */
206 int nh_seqcount; /* heuristic */
207 } nfsheur[NUM_HEURISTIC];
211 * Heuristic to detect sequential operation.
213 static struct nfsheur *
214 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
219 /* Locate best candidate. */
221 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
224 if (nfsheur[hi].nh_vp == vp) {
228 if (nfsheur[hi].nh_use > 0)
229 --nfsheur[hi].nh_use;
230 hi = (hi + 1) % NUM_HEURISTIC;
231 if (nfsheur[hi].nh_use < nh->nh_use)
235 /* Initialize hint if this is a new file. */
236 if (nh->nh_vp != vp) {
238 nh->nh_nextoff = uio->uio_offset;
239 nh->nh_use = NHUSE_INIT;
240 if (uio->uio_offset == 0)
246 /* Calculate heuristic. */
247 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
248 uio->uio_offset == nh->nh_nextoff) {
249 /* See comments in vfs_vnops.c:sequential_heuristic(). */
250 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
251 if (nh->nh_seqcount > IO_SEQMAX)
252 nh->nh_seqcount = IO_SEQMAX;
253 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
254 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
255 /* Probably a reordered RPC, leave seqcount alone. */
256 } else if (nh->nh_seqcount > 1) {
257 nh->nh_seqcount /= 2;
261 nh->nh_use += NHUSE_INC;
262 if (nh->nh_use > NHUSE_MAX)
263 nh->nh_use = NHUSE_MAX;
268 * Get attributes into nfsvattr structure.
271 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
272 struct nfsrv_descript *nd, struct thread *p, int vpislocked,
273 nfsattrbit_t *attrbitp)
275 int error, gotattr, lockedit = 0;
278 if (vpislocked == 0) {
280 * When vpislocked == 0, the vnode is either exclusively
281 * locked by this thread or not locked by this thread.
282 * As such, shared lock it, if not exclusively locked.
284 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
286 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
291 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
292 * attributes, as required.
293 * This needs to be done for regular files if:
294 * - non-NFSv4 RPCs or
295 * - when attrbitp == NULL or
296 * - an NFSv4 RPC with any of the above attributes in attrbitp.
297 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
298 * these attributes. nfsrv_proxyds() will return an error if the
299 * server is not a pNFS one.
302 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
303 (nd->nd_flag & ND_NFSV4) == 0 ||
304 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
305 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
306 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
307 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
308 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
309 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
310 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
316 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
321 * If we got the Change, Size and Modify Time from the DS,
325 nvap->na_atime = na.na_atime;
326 nvap->na_mtime = na.na_mtime;
327 nvap->na_filerev = na.na_filerev;
328 nvap->na_size = na.na_size;
329 nvap->na_bytes = na.na_bytes;
331 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
332 error, (uintmax_t)na.na_filerev);
339 * Get a file handle for a vnode.
342 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
346 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
347 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
348 error = VOP_VPTOFH(vp, &fhp->fh_fid);
355 * Perform access checking for vnodes obtained from file handles that would
356 * refer to files already opened by a Unix client. You cannot just use
357 * vn_writechk() and VOP_ACCESSX() for two reasons.
358 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
360 * 2 - The owner is to be given access irrespective of mode bits for some
361 * operations, so that processes that chmod after opening a file don't
365 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
366 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
367 u_int32_t *supportedtypep)
370 int error = 0, getret = 0;
372 if (vpislocked == 0) {
373 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
378 if (accmode & VWRITE) {
379 /* Just vn_writechk() changed to check rdonly */
381 * Disallow write attempts on read-only file systems;
382 * unless the file is a socket or a block or character
383 * device resident on the file system.
385 if (NFSVNO_EXRDONLY(exp) ||
386 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
387 switch (vp->v_type) {
397 * If there's shared text associated with
398 * the inode, try to free it up once. If
399 * we fail, we can't allow writing.
401 if (VOP_IS_TEXT(vp) && error == 0)
411 * Should the override still be applied when ACLs are enabled?
413 error = VOP_ACCESSX(vp, accmode, cred, p);
414 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
416 * Try again with VEXPLICIT_DENY, to see if the test for
417 * deletion is supported.
419 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
421 if (vp->v_type == VDIR) {
422 accmode &= ~(VDELETE | VDELETE_CHILD);
424 error = VOP_ACCESSX(vp, accmode, cred, p);
425 } else if (supportedtypep != NULL) {
426 *supportedtypep &= ~NFSACCESS_DELETE;
432 * Allow certain operations for the owner (reads and writes
433 * on files that are already open).
435 if (override != NFSACCCHK_NOOVERRIDE &&
436 (error == EPERM || error == EACCES)) {
437 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
439 else if (override & NFSACCCHK_ALLOWOWNER) {
440 getret = VOP_GETATTR(vp, &vattr, cred);
441 if (getret == 0 && cred->cr_uid == vattr.va_uid)
454 * Set attribute(s) vnop.
457 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
458 struct thread *p, struct nfsexstuff *exp)
460 u_quad_t savsize = 0;
464 * If this is an exported file system and a pNFS service is running,
465 * don't VOP_SETATTR() of size for the MDS file system.
469 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
470 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
471 nvap->na_vattr.va_size > 0) {
472 savsize = nvap->na_vattr.va_size;
473 nvap->na_vattr.va_size = VNOVAL;
474 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
475 nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
476 nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
477 nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
478 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
484 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
486 nvap->na_vattr.va_size = savsize;
487 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
488 nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
489 nvap->na_vattr.va_size != VNOVAL ||
490 nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
491 nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
492 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
493 /* For a pNFS server, set the attributes on the DS file. */
494 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
495 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
504 * Set up nameidata for a lookup() call and do it.
507 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
508 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
509 struct vnode **retdirp)
511 struct componentname *cnp = &ndp->ni_cnd;
515 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
520 cnp->cn_nameptr = cnp->cn_pnbuf;
523 * Extract and set starting directory.
525 if (dp->v_type != VDIR) {
530 nfsvno_relpathbuf(ndp);
538 if (NFSVNO_EXRDONLY(exp))
539 cnp->cn_flags |= RDONLY;
540 ndp->ni_segflg = UIO_SYSSPACE;
542 if (nd->nd_flag & ND_PUBLOOKUP) {
544 if (cnp->cn_pnbuf[0] == '/') {
547 * Check for degenerate pathnames here, since lookup()
550 for (i = 1; i < ndp->ni_pathlen; i++)
551 if (cnp->cn_pnbuf[i] != '/')
553 if (i == ndp->ni_pathlen) {
554 error = NFSERR_ACCES;
560 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
561 (nd->nd_flag & ND_NFSV4) == 0) {
563 * Only cross mount points for NFSv4 when doing a
564 * mount while traversing the file system above
565 * the mount point, unless nfsrv_enable_crossmntpt is set.
567 cnp->cn_flags |= NOCROSSMOUNT;
571 * Initialize for scan, set ni_startdir and bump ref on dp again
572 * because lookup() will dereference ni_startdir.
576 ndp->ni_startdir = dp;
577 ndp->ni_rootdir = rootvnode;
578 ndp->ni_topdir = NULL;
581 cnp->cn_flags |= LOCKLEAF;
583 cnp->cn_nameptr = cnp->cn_pnbuf;
585 * Call lookup() to do the real work. If an error occurs,
586 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
587 * we do not have to dereference anything before returning.
588 * In either case ni_startdir will be dereferenced and NULLed
596 * Check for encountering a symbolic link. Trivial
597 * termination occurs if no symlink encountered.
599 if ((cnp->cn_flags & ISSYMLINK) == 0) {
600 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
601 nfsvno_relpathbuf(ndp);
602 if (ndp->ni_vp && !lockleaf)
603 NFSVOPUNLOCK(ndp->ni_vp);
610 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
611 NFSVOPUNLOCK(ndp->ni_dvp);
612 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
617 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
621 if (ndp->ni_pathlen > 1)
622 cp = uma_zalloc(namei_zone, M_WAITOK);
626 aiov.iov_len = MAXPATHLEN;
627 auio.uio_iov = &aiov;
630 auio.uio_rw = UIO_READ;
631 auio.uio_segflg = UIO_SYSSPACE;
633 auio.uio_resid = MAXPATHLEN;
634 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
637 if (ndp->ni_pathlen > 1)
638 uma_zfree(namei_zone, cp);
644 linklen = MAXPATHLEN - auio.uio_resid;
649 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
650 error = ENAMETOOLONG;
655 * Adjust or replace path
657 if (ndp->ni_pathlen > 1) {
658 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
659 uma_zfree(namei_zone, cnp->cn_pnbuf);
662 cnp->cn_pnbuf[linklen] = '\0';
663 ndp->ni_pathlen += linklen;
666 * Cleanup refs for next loop and check if root directory
667 * should replace current directory. Normally ni_dvp
668 * becomes the new base directory and is cleaned up when
669 * we loop. Explicitly null pointers after invalidation
670 * to clarify operation.
675 if (cnp->cn_pnbuf[0] == '/') {
677 ndp->ni_dvp = ndp->ni_rootdir;
680 ndp->ni_startdir = ndp->ni_dvp;
684 cnp->cn_flags &= ~LOCKLEAF;
688 nfsvno_relpathbuf(ndp);
691 ndp->ni_startdir = NULL;
692 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
697 NFSEXITCODE2(error, nd);
702 * Set up a pathname buffer and return a pointer to it and, optionally
703 * set a hash pointer.
706 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
708 struct componentname *cnp = &ndp->ni_cnd;
710 cnp->cn_flags |= (NOMACCHECK | HASBUF);
711 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
714 *bufpp = cnp->cn_pnbuf;
718 * Release the above path buffer, if not released by nfsvno_namei().
721 nfsvno_relpathbuf(struct nameidata *ndp)
724 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
726 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
727 ndp->ni_cnd.cn_flags &= ~HASBUF;
731 * Readlink vnode op into an mbuf list.
734 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
735 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
738 struct uio io, *uiop = &io;
739 struct mbuf *mp, *mp3;
740 int len, tlen, error = 0;
742 len = NFS_MAXPATHLEN;
743 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
745 uiop->uio_offset = 0;
746 uiop->uio_resid = len;
747 uiop->uio_rw = UIO_READ;
748 uiop->uio_segflg = UIO_SYSSPACE;
750 error = VOP_READLINK(vp, uiop, cred);
757 if (uiop->uio_resid > 0) {
758 len -= uiop->uio_resid;
759 tlen = NFSM_RNDUP(len);
760 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
772 * Create an mbuf chain and an associated iovec that can be used to Read
773 * or Getextattr of data.
774 * Upon success, return pointers to the first and last mbufs in the chain
775 * plus the malloc'd iovec and its iovlen.
778 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
781 struct mbuf *m, *m2 = NULL, *m3;
788 * Generate the mbuf list with the uio_iov ref. to it.
795 siz = min(M_TRAILINGSPACE(m), left);
804 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
810 panic("nfsvno_read iov");
811 siz = min(M_TRAILINGSPACE(m), left);
813 iv->iov_base = mtod(m, caddr_t) + m->m_len;
828 * Read vnode op call into mbuf list.
831 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
832 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
836 int error = 0, len, tlen, ioflag = 0;
838 struct uio io, *uiop = &io;
842 * Attempt to read from a DS file. A return of ENOENT implies
843 * there is no DS file to read.
845 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
846 NULL, mpendp, NULL, NULL, NULL, 0, NULL);
850 len = NFSM_RNDUP(cnt);
851 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
853 uiop->uio_offset = off;
854 uiop->uio_resid = len;
855 uiop->uio_rw = UIO_READ;
856 uiop->uio_segflg = UIO_SYSSPACE;
858 nh = nfsrv_sequential_heuristic(uiop, vp);
859 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
860 /* XXX KDM make this more systematic? */
861 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
862 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
869 nh->nh_nextoff = uiop->uio_offset;
870 tlen = len - uiop->uio_resid;
871 cnt = cnt < tlen ? cnt : tlen;
872 tlen = NFSM_RNDUP(cnt);
876 } else if (len != tlen || tlen != cnt)
877 nfsrv_adj(m3, len - tlen, tlen - cnt);
887 * Create the iovec for the mbuf chain passed in as an argument.
888 * The "cp" argument is where the data starts within the first mbuf in
889 * the chain. It returns the iovec and the iovcnt.
892 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
900 * Loop through the mbuf chain, counting how many mbufs are a
901 * part of this write operation, so the iovec size is known.
906 i = mtod(mp, caddr_t) + mp->m_len - cp;
920 /* Now, create the iovec. */
922 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
925 i = mtod(mp, caddr_t) + mp->m_len - cp;
929 panic("nfsvno_write");
940 cp = mtod(mp, caddr_t);
947 * Write vnode op from an mbuf list.
950 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
951 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
954 int cnt, ioflags, error;
955 struct uio io, *uiop = &io;
959 * Attempt to write to a DS file. A return of ENOENT implies
960 * there is no DS file to write.
962 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
963 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
964 if (error != ENOENT) {
965 *stable = NFSWRITE_FILESYNC;
970 if (*stable == NFSWRITE_UNSTABLE)
971 ioflags = IO_NODELOCKED;
973 ioflags = (IO_SYNC | IO_NODELOCKED);
974 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
978 uiop->uio_iovcnt = cnt;
979 uiop->uio_resid = retlen;
980 uiop->uio_rw = UIO_WRITE;
981 uiop->uio_segflg = UIO_SYSSPACE;
983 uiop->uio_offset = off;
984 nh = nfsrv_sequential_heuristic(uiop, vp);
985 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
986 /* XXX KDM make this more systematic? */
987 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
988 error = VOP_WRITE(vp, uiop, ioflags, cred);
990 nh->nh_nextoff = uiop->uio_offset;
998 * Common code for creating a regular file (plus special files for V2).
1001 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1002 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1003 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1007 struct thread *p = curthread;
1009 error = nd->nd_repstat;
1010 if (!error && ndp->ni_vp == NULL) {
1011 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1012 vrele(ndp->ni_startdir);
1013 error = VOP_CREATE(ndp->ni_dvp,
1014 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1015 /* For a pNFS server, create the data file on a DS. */
1016 if (error == 0 && nvap->na_type == VREG) {
1018 * Create a data file on a DS for a pNFS server.
1019 * This function just returns if not
1020 * running a pNFS DS or the creation fails.
1022 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1026 nfsvno_relpathbuf(ndp);
1028 if (*exclusive_flagp) {
1029 *exclusive_flagp = 0;
1030 NFSVNO_ATTRINIT(nvap);
1031 nvap->na_atime.tv_sec = cverf[0];
1032 nvap->na_atime.tv_nsec = cverf[1];
1033 error = VOP_SETATTR(ndp->ni_vp,
1034 &nvap->na_vattr, nd->nd_cred);
1038 error = NFSERR_NOTSUPP;
1043 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1044 * (This implies, just get out on an error.)
1046 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1047 nvap->na_type == VFIFO) {
1048 if (nvap->na_type == VCHR && rdev == 0xffffffff)
1049 nvap->na_type = VFIFO;
1050 if (nvap->na_type != VFIFO &&
1051 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1052 vrele(ndp->ni_startdir);
1053 nfsvno_relpathbuf(ndp);
1057 nvap->na_rdev = rdev;
1058 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1059 &ndp->ni_cnd, &nvap->na_vattr);
1061 nfsvno_relpathbuf(ndp);
1062 vrele(ndp->ni_startdir);
1066 vrele(ndp->ni_startdir);
1067 nfsvno_relpathbuf(ndp);
1075 * Handle cases where error is already set and/or
1077 * 1 - clean up the lookup
1078 * 2 - iff !error and na_size set, truncate it
1080 vrele(ndp->ni_startdir);
1081 nfsvno_relpathbuf(ndp);
1083 if (ndp->ni_dvp == *vpp)
1087 if (!error && nvap->na_size != VNOVAL) {
1088 error = nfsvno_accchk(*vpp, VWRITE,
1089 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1090 NFSACCCHK_VPISLOCKED, NULL);
1092 tempsize = nvap->na_size;
1093 NFSVNO_ATTRINIT(nvap);
1094 nvap->na_size = tempsize;
1095 error = VOP_SETATTR(*vpp,
1096 &nvap->na_vattr, nd->nd_cred);
1109 * Do a mknod vnode op.
1112 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1118 vtyp = nvap->na_type;
1120 * Iff doesn't exist, create it.
1123 vrele(ndp->ni_startdir);
1124 nfsvno_relpathbuf(ndp);
1130 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1131 vrele(ndp->ni_startdir);
1132 nfsvno_relpathbuf(ndp);
1134 error = NFSERR_BADTYPE;
1137 if (vtyp == VSOCK) {
1138 vrele(ndp->ni_startdir);
1139 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1140 &ndp->ni_cnd, &nvap->na_vattr);
1142 nfsvno_relpathbuf(ndp);
1144 if (nvap->na_type != VFIFO &&
1145 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1146 vrele(ndp->ni_startdir);
1147 nfsvno_relpathbuf(ndp);
1151 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1152 &ndp->ni_cnd, &nvap->na_vattr);
1154 nfsvno_relpathbuf(ndp);
1155 vrele(ndp->ni_startdir);
1157 * Since VOP_MKNOD returns the ni_vp, I can't
1158 * see any reason to do the lookup.
1171 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1172 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1176 if (ndp->ni_vp != NULL) {
1177 if (ndp->ni_dvp == ndp->ni_vp)
1182 nfsvno_relpathbuf(ndp);
1186 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1189 nfsvno_relpathbuf(ndp);
1200 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1201 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1202 struct nfsexstuff *exp)
1207 vrele(ndp->ni_startdir);
1208 nfsvno_relpathbuf(ndp);
1209 if (ndp->ni_dvp == ndp->ni_vp)
1218 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1219 &nvap->na_vattr, pathcp);
1221 vrele(ndp->ni_startdir);
1222 nfsvno_relpathbuf(ndp);
1224 * Although FreeBSD still had the lookup code in
1225 * it for 7/current, there doesn't seem to be any
1226 * point, since VOP_SYMLINK() returns the ni_vp.
1227 * Just vput it for v2.
1229 if (!not_v2 && !error)
1238 * Parse symbolic link arguments.
1239 * This function has an ugly side effect. It will malloc() an area for
1240 * the symlink and set iov_base to point to it, only if it succeeds.
1241 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1245 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1246 struct thread *p, char **pathcpp, int *lenp)
1249 char *pathcp = NULL;
1251 struct nfsv2_sattr *sp;
1255 if ((nd->nd_flag & ND_NFSV3) &&
1256 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1258 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1259 len = fxdr_unsigned(int, *tl);
1260 if (len > NFS_MAXPATHLEN || len <= 0) {
1264 pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1265 error = nfsrv_mtostr(nd, pathcp, len);
1268 if (nd->nd_flag & ND_NFSV2) {
1269 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1270 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1274 NFSEXITCODE2(0, nd);
1278 free(pathcp, M_TEMP);
1279 NFSEXITCODE2(error, nd);
1284 * Remove a non-directory object.
1287 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1288 struct thread *p, struct nfsexstuff *exp)
1290 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1291 int error = 0, mirrorcnt;
1292 char fname[PNFS_FILENAME_LEN + 1];
1297 if (vp->v_type == VDIR)
1298 error = NFSERR_ISDIR;
1300 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1303 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1305 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1306 if (error == 0 && dsdvp[0] != NULL)
1307 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1308 if (ndp->ni_dvp == vp)
1313 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1314 nfsvno_relpathbuf(ndp);
1320 * Remove a directory.
1323 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1324 struct thread *p, struct nfsexstuff *exp)
1330 if (vp->v_type != VDIR) {
1335 * No rmdir "." please.
1337 if (ndp->ni_dvp == vp) {
1342 * The root of a mounted filesystem cannot be deleted.
1344 if (vp->v_vflag & VV_ROOT)
1348 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1349 if (ndp->ni_dvp == vp)
1354 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1355 nfsvno_relpathbuf(ndp);
1364 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1365 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1367 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1368 int error = 0, mirrorcnt;
1369 char fname[PNFS_FILENAME_LEN + 1];
1373 fvp = fromndp->ni_vp;
1375 vrele(fromndp->ni_dvp);
1380 tdvp = tondp->ni_dvp;
1383 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1384 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1386 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1387 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1390 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1391 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1396 * A rename to '.' or '..' results in a prematurely
1397 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1400 if ((tondp->ni_cnd.cn_namelen == 1 &&
1401 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1402 (tondp->ni_cnd.cn_namelen == 2 &&
1403 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1404 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1409 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1410 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1413 if (fvp->v_mount != tdvp->v_mount) {
1414 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1418 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1423 * If source and destination are the same, there is nothing to
1424 * do. Set error to -1 to indicate this.
1429 if (ndflag & ND_NFSV4) {
1430 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1431 error = nfsrv_checkremove(fvp, 0, NULL,
1432 (nfsquad_t)((u_quad_t)0), p);
1437 error = nfsrv_checkremove(tvp, 1, NULL,
1438 (nfsquad_t)((u_quad_t)0), p);
1441 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1442 * that the NFSv4 client won't be confused by the rename.
1443 * Since nfsd_recalldelegation() can only be called on an
1444 * unlocked vnode at this point and fvp is the file that will
1445 * still exist after the rename, just do fvp.
1447 nfsd_recalldelegation(fvp, p);
1449 if (error == 0 && tvp != NULL) {
1450 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1451 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1452 " dsdvp=%p\n", dsdvp[0]);
1456 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1457 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1466 vrele(fromndp->ni_dvp);
1473 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1474 * if the rename succeeded, the DS file for the tvp needs to be
1477 if (error == 0 && dsdvp[0] != NULL) {
1478 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1479 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1482 vrele(tondp->ni_startdir);
1483 nfsvno_relpathbuf(tondp);
1485 vrele(fromndp->ni_startdir);
1486 nfsvno_relpathbuf(fromndp);
1495 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1496 struct thread *p, struct nfsexstuff *exp)
1506 if (vp->v_mount != xp->v_mount)
1510 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1511 if (!VN_IS_DOOMED(vp))
1512 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1515 if (ndp->ni_dvp == vp)
1521 if (ndp->ni_dvp == ndp->ni_vp)
1528 nfsvno_relpathbuf(ndp);
1534 * Do the fsync() appropriate for the commit.
1537 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1543 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1544 * file is done. At this time VOP_FSYNC does not accept offset and
1545 * byte count parameters so call VOP_FSYNC the whole file for now.
1546 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1547 * File systems that do not use the buffer cache (as indicated
1548 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1550 if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1551 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1553 * Give up and do the whole thing
1555 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1556 VM_OBJECT_WLOCK(vp->v_object);
1557 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1558 VM_OBJECT_WUNLOCK(vp->v_object);
1560 error = VOP_FSYNC(vp, MNT_WAIT, td);
1563 * Locate and synchronously write any buffers that fall
1564 * into the requested range. Note: we are assuming that
1565 * f_iosize is a power of 2.
1567 int iosize = vp->v_mount->mnt_stat.f_iosize;
1568 int iomask = iosize - 1;
1573 * Align to iosize boundary, super-align to page boundary.
1576 cnt += off & iomask;
1577 off &= ~(u_quad_t)iomask;
1579 if (off & PAGE_MASK) {
1580 cnt += off & PAGE_MASK;
1581 off &= ~(u_quad_t)PAGE_MASK;
1583 lblkno = off / iosize;
1585 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1586 VM_OBJECT_WLOCK(vp->v_object);
1587 vm_object_page_clean(vp->v_object, off, off + cnt,
1589 VM_OBJECT_WUNLOCK(vp->v_object);
1598 * If we have a buffer and it is marked B_DELWRI we
1599 * have to lock and write it. Otherwise the prior
1600 * write is assumed to have already been committed.
1602 * gbincore() can return invalid buffers now so we
1603 * have to check that bit as well (though B_DELWRI
1604 * should not be set if B_INVAL is set there could be
1605 * a race here since we haven't locked the buffer).
1607 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1608 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1609 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1611 continue; /* retry */
1613 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1616 bp->b_flags &= ~B_ASYNC;
1639 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1645 if (nfsrv_devidcnt > 0) {
1646 /* For a pNFS service, get the DS numbers. */
1647 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1648 error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1654 error = VFS_STATFS(vp->v_mount, sf);
1657 sf->f_blocks = tsf->f_blocks;
1658 sf->f_bavail = tsf->f_bavail;
1659 sf->f_bfree = tsf->f_bfree;
1660 sf->f_bsize = tsf->f_bsize;
1663 * Since NFS handles these values as unsigned on the
1664 * wire, there is no way to represent negative values,
1665 * so set them to 0. Without this, they will appear
1666 * to be very large positive values for clients like
1669 if (sf->f_bavail < 0)
1671 if (sf->f_ffree < 0)
1680 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1681 * must handle nfsrv_opencheck() calls after any other access checks.
1684 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1685 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1686 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1687 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1688 struct nfsexstuff *exp, struct vnode **vpp)
1690 struct vnode *vp = NULL;
1692 struct nfsexstuff nes;
1693 struct thread *p = curthread;
1695 if (ndp->ni_vp == NULL)
1696 nd->nd_repstat = nfsrv_opencheck(clientid,
1697 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1698 if (!nd->nd_repstat) {
1699 if (ndp->ni_vp == NULL) {
1700 vrele(ndp->ni_startdir);
1701 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1702 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1703 /* For a pNFS server, create the data file on a DS. */
1704 if (nd->nd_repstat == 0) {
1706 * Create a data file on a DS for a pNFS server.
1707 * This function just returns if not
1708 * running a pNFS DS or the creation fails.
1710 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1714 nfsvno_relpathbuf(ndp);
1715 if (!nd->nd_repstat) {
1716 if (*exclusive_flagp) {
1717 *exclusive_flagp = 0;
1718 NFSVNO_ATTRINIT(nvap);
1719 nvap->na_atime.tv_sec = cverf[0];
1720 nvap->na_atime.tv_nsec = cverf[1];
1721 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1722 &nvap->na_vattr, cred);
1723 if (nd->nd_repstat != 0) {
1726 nd->nd_repstat = NFSERR_NOTSUPP;
1728 NFSSETBIT_ATTRBIT(attrbitp,
1729 NFSATTRBIT_TIMEACCESS);
1731 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1732 aclp, p, attrbitp, exp);
1737 if (ndp->ni_startdir)
1738 vrele(ndp->ni_startdir);
1739 nfsvno_relpathbuf(ndp);
1741 if (create == NFSV4OPEN_CREATE) {
1742 if (ndp->ni_dvp == vp)
1747 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1748 if (ndp->ni_cnd.cn_flags & RDONLY)
1749 NFSVNO_SETEXRDONLY(&nes);
1751 NFSVNO_EXINIT(&nes);
1752 nd->nd_repstat = nfsvno_accchk(vp,
1753 VWRITE, cred, &nes, p,
1754 NFSACCCHK_NOOVERRIDE,
1755 NFSACCCHK_VPISLOCKED, NULL);
1756 nd->nd_repstat = nfsrv_opencheck(clientid,
1757 stateidp, stp, vp, nd, p, nd->nd_repstat);
1758 if (!nd->nd_repstat) {
1759 tempsize = nvap->na_size;
1760 NFSVNO_ATTRINIT(nvap);
1761 nvap->na_size = tempsize;
1762 nd->nd_repstat = VOP_SETATTR(vp,
1763 &nvap->na_vattr, cred);
1765 } else if (vp->v_type == VREG) {
1766 nd->nd_repstat = nfsrv_opencheck(clientid,
1767 stateidp, stp, vp, nd, p, nd->nd_repstat);
1771 if (ndp->ni_cnd.cn_flags & HASBUF)
1772 nfsvno_relpathbuf(ndp);
1773 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1774 vrele(ndp->ni_startdir);
1775 if (ndp->ni_dvp == ndp->ni_vp)
1785 NFSEXITCODE2(0, nd);
1789 * Updates the file rev and sets the mtime and ctime
1790 * to the current clock time, returning the va_filerev and va_Xtime
1792 * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1795 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1796 struct nfsrv_descript *nd, struct thread *p)
1801 vfs_timestamp(&va.va_mtime);
1802 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1803 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1804 if (VN_IS_DOOMED(vp))
1807 (void) VOP_SETATTR(vp, &va, nd->nd_cred);
1808 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1813 * Glue routine to nfsv4_fillattr().
1816 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1817 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1818 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1819 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1825 if (nfsrv_devidcnt > 0 &&
1826 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1827 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1828 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1829 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1830 error = nfsrv_pnfsstatfs(sf, mp);
1836 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1837 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1838 mounted_on_fileno, sf);
1840 NFSEXITCODE2(0, nd);
1844 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1846 * nfs readdir service
1847 * - mallocs what it thinks is enough to read
1848 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1849 * - calls VOP_READDIR()
1850 * - loops around building the reply
1851 * if the output generated exceeds count break out of loop
1852 * The NFSM_CLGET macro is used here so that the reply will be packed
1853 * tightly in mbuf clusters.
1854 * - it trims out records with d_fileno == 0
1855 * this doesn't matter for Unix clients, but they might confuse clients
1857 * - it trims out records with d_type == DT_WHT
1858 * these cannot be seen through NFS (unless we extend the protocol)
1859 * The alternate call nfsrvd_readdirplus() does lookups as well.
1860 * PS: The NFS protocol spec. does not clarify what the "count" byte
1861 * argument is a count of.. just name strings and file id's or the
1862 * entire reply rpc or ...
1863 * I tried just file name and id sizes and it confused the Sun client,
1864 * so I am using the full rpc size now. The "paranoia.." comment refers
1865 * to including the status longwords that are not a part of the dir.
1866 * "entry" structures, but are in the rpc.
1869 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1870 struct vnode *vp, struct nfsexstuff *exp)
1875 char *cpos, *cend, *rbuf;
1877 int nlen, error = 0, getret = 1;
1878 int siz, cnt, fullsiz, eofflag, ncookies;
1879 u_int64_t off, toff, verf __unused;
1880 u_long *cookies = NULL, *cookiep;
1884 struct thread *p = curthread;
1886 if (nd->nd_repstat) {
1887 nfsrv_postopattr(nd, getret, &at);
1890 if (nd->nd_flag & ND_NFSV2) {
1891 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1892 off = fxdr_unsigned(u_quad_t, *tl++);
1894 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1895 off = fxdr_hyper(tl);
1897 verf = fxdr_hyper(tl);
1901 cnt = fxdr_unsigned(int, *tl);
1902 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1903 cnt = NFS_SRVMAXDATA(nd);
1904 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1906 if (nd->nd_flag & ND_NFSV3) {
1907 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
1911 * va_filerev is not sufficient as a cookie verifier,
1912 * since it is not supposed to change when entries are
1913 * removed/added unless that offset cookies returned to
1914 * the client are no longer valid.
1916 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1917 nd->nd_repstat = NFSERR_BAD_COOKIE;
1920 if (!nd->nd_repstat && vp->v_type != VDIR)
1921 nd->nd_repstat = NFSERR_NOTDIR;
1922 if (nd->nd_repstat == 0 && cnt == 0) {
1923 if (nd->nd_flag & ND_NFSV2)
1924 /* NFSv2 does not have NFSERR_TOOSMALL */
1925 nd->nd_repstat = EPERM;
1927 nd->nd_repstat = NFSERR_TOOSMALL;
1929 if (!nd->nd_repstat)
1930 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1931 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1932 NFSACCCHK_VPISLOCKED, NULL);
1933 if (nd->nd_repstat) {
1935 if (nd->nd_flag & ND_NFSV3)
1936 nfsrv_postopattr(nd, getret, &at);
1939 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1940 rbuf = malloc(siz, M_TEMP, M_WAITOK);
1944 free(cookies, M_TEMP);
1952 io.uio_offset = (off_t)off;
1954 io.uio_segflg = UIO_SYSSPACE;
1955 io.uio_rw = UIO_READ;
1957 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1959 off = (u_int64_t)io.uio_offset;
1961 siz -= io.uio_resid;
1963 if (!cookies && !nd->nd_repstat)
1964 nd->nd_repstat = NFSERR_PERM;
1965 if (nd->nd_flag & ND_NFSV3) {
1966 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
1967 if (!nd->nd_repstat)
1968 nd->nd_repstat = getret;
1972 * Handles the failed cases. nd->nd_repstat == 0 past here.
1974 if (nd->nd_repstat) {
1978 free(cookies, M_TEMP);
1979 if (nd->nd_flag & ND_NFSV3)
1980 nfsrv_postopattr(nd, getret, &at);
1984 * If nothing read, return eof
1989 if (nd->nd_flag & ND_NFSV2) {
1990 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1992 nfsrv_postopattr(nd, getret, &at);
1993 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1994 txdr_hyper(at.na_filerev, tl);
1997 *tl++ = newnfs_false;
2000 free(cookies, M_TEMP);
2005 * Check for degenerate cases of nothing useful read.
2006 * If so go try again
2010 dp = (struct dirent *)cpos;
2014 * For some reason FreeBSD's ufs_readdir() chooses to back the
2015 * directory offset up to a block boundary, so it is necessary to
2016 * skip over the records that precede the requested offset. This
2017 * requires the assumption that file offset cookies monotonically
2020 while (cpos < cend && ncookies > 0 &&
2021 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2022 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2023 cpos += dp->d_reclen;
2024 dp = (struct dirent *)cpos;
2028 if (cpos >= cend || ncookies == 0) {
2036 * dirlen is the size of the reply, including all XDR and must
2037 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2038 * if the XDR should be included in "count", but to be safe, we do.
2039 * (Include the two booleans at the end of the reply in dirlen now.)
2041 if (nd->nd_flag & ND_NFSV3) {
2042 nfsrv_postopattr(nd, getret, &at);
2043 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2044 txdr_hyper(at.na_filerev, tl);
2045 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2047 dirlen = 2 * NFSX_UNSIGNED;
2050 /* Loop through the records and build reply */
2051 while (cpos < cend && ncookies > 0) {
2052 nlen = dp->d_namlen;
2053 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2054 nlen <= NFS_MAXNAMLEN) {
2055 if (nd->nd_flag & ND_NFSV3)
2056 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2058 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2065 * Build the directory record xdr from
2068 if (nd->nd_flag & ND_NFSV3) {
2069 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2070 *tl++ = newnfs_true;
2073 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2074 *tl++ = newnfs_true;
2076 *tl = txdr_unsigned(dp->d_fileno);
2077 (void) nfsm_strtom(nd, dp->d_name, nlen);
2078 if (nd->nd_flag & ND_NFSV3) {
2079 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2082 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2083 *tl = txdr_unsigned(*cookiep);
2085 cpos += dp->d_reclen;
2086 dp = (struct dirent *)cpos;
2092 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2093 *tl++ = newnfs_false;
2099 free(cookies, M_TEMP);
2102 NFSEXITCODE2(0, nd);
2106 NFSEXITCODE2(error, nd);
2111 * Readdirplus for V3 and Readdir for V4.
2114 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2115 struct vnode *vp, struct nfsexstuff *exp)
2120 char *cpos, *cend, *rbuf;
2123 struct nfsvattr nva, at, *nvap = &nva;
2124 struct mbuf *mb0, *mb1;
2125 struct nfsreferral *refp;
2126 int nlen, r, error = 0, getret = 1, usevget = 1;
2127 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2128 caddr_t bpos0, bpos1;
2129 u_int64_t off, toff, verf;
2130 u_long *cookies = NULL, *cookiep;
2131 nfsattrbit_t attrbits, rderrbits, savbits;
2134 struct componentname cn;
2135 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2136 struct mount *mp, *new_mp;
2137 uint64_t mounted_on_fileno;
2138 struct thread *p = curthread;
2140 if (nd->nd_repstat) {
2141 nfsrv_postopattr(nd, getret, &at);
2144 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2145 off = fxdr_hyper(tl);
2148 verf = fxdr_hyper(tl);
2150 siz = fxdr_unsigned(int, *tl++);
2151 cnt = fxdr_unsigned(int, *tl);
2154 * Use the server's maximum data transfer size as the upper bound
2157 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2158 cnt = NFS_SRVMAXDATA(nd);
2161 * siz is a "hint" of how much directory information (name, fileid,
2162 * cookie) should be in the reply. At least one client "hints" 0,
2163 * so I set it to cnt for that case. I also round it up to the
2164 * next multiple of DIRBLKSIZ.
2165 * Since the size of a Readdirplus directory entry reply will always
2166 * be greater than a directory entry returned by VOP_READDIR(), it
2167 * does not make sense to read more than NFS_SRVMAXDATA() via
2172 else if (siz > NFS_SRVMAXDATA(nd))
2173 siz = NFS_SRVMAXDATA(nd);
2174 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2176 if (nd->nd_flag & ND_NFSV4) {
2177 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2180 NFSSET_ATTRBIT(&savbits, &attrbits);
2181 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2182 NFSZERO_ATTRBIT(&rderrbits);
2183 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2185 NFSZERO_ATTRBIT(&attrbits);
2188 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2190 if (!nd->nd_repstat) {
2191 if (off && verf != at.na_filerev) {
2193 * va_filerev is not sufficient as a cookie verifier,
2194 * since it is not supposed to change when entries are
2195 * removed/added unless that offset cookies returned to
2196 * the client are no longer valid.
2198 if (nd->nd_flag & ND_NFSV4) {
2199 nd->nd_repstat = NFSERR_NOTSAME;
2201 nd->nd_repstat = NFSERR_BAD_COOKIE;
2206 if (!nd->nd_repstat && vp->v_type != VDIR)
2207 nd->nd_repstat = NFSERR_NOTDIR;
2208 if (!nd->nd_repstat && cnt == 0)
2209 nd->nd_repstat = NFSERR_TOOSMALL;
2210 if (!nd->nd_repstat)
2211 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2212 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2213 NFSACCCHK_VPISLOCKED, NULL);
2214 if (nd->nd_repstat) {
2216 if (nd->nd_flag & ND_NFSV3)
2217 nfsrv_postopattr(nd, getret, &at);
2220 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2221 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2223 rbuf = malloc(siz, M_TEMP, M_WAITOK);
2227 free(cookies, M_TEMP);
2235 io.uio_offset = (off_t)off;
2237 io.uio_segflg = UIO_SYSSPACE;
2238 io.uio_rw = UIO_READ;
2240 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2242 off = (u_int64_t)io.uio_offset;
2244 siz -= io.uio_resid;
2246 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2248 if (!cookies && !nd->nd_repstat)
2249 nd->nd_repstat = NFSERR_PERM;
2250 if (!nd->nd_repstat)
2251 nd->nd_repstat = getret;
2252 if (nd->nd_repstat) {
2255 free(cookies, M_TEMP);
2257 if (nd->nd_flag & ND_NFSV3)
2258 nfsrv_postopattr(nd, getret, &at);
2262 * If nothing read, return eof
2267 if (nd->nd_flag & ND_NFSV3)
2268 nfsrv_postopattr(nd, getret, &at);
2269 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2270 txdr_hyper(at.na_filerev, tl);
2272 *tl++ = newnfs_false;
2274 free(cookies, M_TEMP);
2280 * Check for degenerate cases of nothing useful read.
2281 * If so go try again
2285 dp = (struct dirent *)cpos;
2289 * For some reason FreeBSD's ufs_readdir() chooses to back the
2290 * directory offset up to a block boundary, so it is necessary to
2291 * skip over the records that precede the requested offset. This
2292 * requires the assumption that file offset cookies monotonically
2295 while (cpos < cend && ncookies > 0 &&
2296 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2297 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2298 ((nd->nd_flag & ND_NFSV4) &&
2299 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2300 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2301 cpos += dp->d_reclen;
2302 dp = (struct dirent *)cpos;
2306 if (cpos >= cend || ncookies == 0) {
2313 * Busy the file system so that the mount point won't go away
2314 * and, as such, VFS_VGET() can be used safely.
2319 nd->nd_repstat = vfs_busy(mp, 0);
2321 if (nd->nd_repstat != 0) {
2323 free(cookies, M_TEMP);
2325 if (nd->nd_flag & ND_NFSV3)
2326 nfsrv_postopattr(nd, getret, &at);
2331 * Check to see if entries in this directory can be safely acquired
2332 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2333 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2334 * automount of the snapshot directory that is required will
2336 * This needs to be done here for NFSv4, since NFSv4 never does
2337 * a VFS_VGET() for "." or "..".
2340 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2341 if (r == EOPNOTSUPP) {
2343 cn.cn_nameiop = LOOKUP;
2344 cn.cn_lkflags = LK_SHARED | LK_RETRY;
2345 cn.cn_cred = nd->nd_cred;
2352 * Save this position, in case there is an error before one entry
2356 bpos0 = nd->nd_bpos;
2359 * Fill in the first part of the reply.
2360 * dirlen is the reply length in bytes and cannot exceed cnt.
2361 * (Include the two booleans at the end of the reply in dirlen now,
2362 * so we recognize when we have exceeded cnt.)
2364 if (nd->nd_flag & ND_NFSV3) {
2365 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2366 nfsrv_postopattr(nd, getret, &at);
2368 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2370 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2371 txdr_hyper(at.na_filerev, tl);
2374 * Save this position, in case there is an empty reply needed.
2377 bpos1 = nd->nd_bpos;
2379 /* Loop through the records and build reply */
2381 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2382 nlen = dp->d_namlen;
2383 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2384 nlen <= NFS_MAXNAMLEN &&
2385 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2386 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2387 || (nlen == 1 && dp->d_name[0] != '.'))) {
2389 * Save the current position in the reply, in case
2390 * this entry exceeds cnt.
2393 bpos1 = nd->nd_bpos;
2396 * For readdir_and_lookup get the vnode using
2405 mounted_on_fileno = (uint64_t)dp->d_fileno;
2406 if ((nd->nd_flag & ND_NFSV3) ||
2407 NFSNONZERO_ATTRBIT(&savbits)) {
2408 if (nd->nd_flag & ND_NFSV4)
2409 refp = nfsv4root_getreferral(NULL,
2413 r = VFS_VGET(mp, dp->d_fileno,
2417 if (r == EOPNOTSUPP) {
2420 cn.cn_nameiop = LOOKUP;
2428 cn.cn_nameptr = dp->d_name;
2429 cn.cn_namelen = nlen;
2430 cn.cn_flags = ISLASTCN |
2431 NOFOLLOW | LOCKLEAF;
2433 dp->d_name[0] == '.' &&
2434 dp->d_name[1] == '.')
2437 if (NFSVOPLOCK(vp, LK_SHARED)
2439 nd->nd_repstat = EPERM;
2442 if ((vp->v_vflag & VV_ROOT) != 0
2443 && (cn.cn_flags & ISDOTDOT)
2449 r = VOP_LOOKUP(vp, &nvp,
2457 * For NFSv4, check to see if nvp is
2458 * a mount point and get the mount
2459 * point vnode, as required.
2462 nfsrv_enable_crossmntpt != 0 &&
2463 (nd->nd_flag & ND_NFSV4) != 0 &&
2464 nvp->v_type == VDIR &&
2465 nvp->v_mountedhere != NULL) {
2466 new_mp = nvp->v_mountedhere;
2467 r = vfs_busy(new_mp, 0);
2471 r = VFS_ROOT(new_mp,
2481 * If we failed to look up the entry, then it
2482 * has become invalid, most likely removed.
2489 KASSERT(refp != NULL || nvp != NULL,
2490 ("%s: undetected lookup error", __func__));
2493 ((nd->nd_flag & ND_NFSV3) ||
2494 NFSNONZERO_ATTRBIT(&attrbits))) {
2495 r = nfsvno_getfh(nvp, &nfh, p);
2497 r = nfsvno_getattr(nvp, nvap, nd, p,
2499 if (r == 0 && is_zfs == 1 &&
2500 nfsrv_enable_crossmntpt != 0 &&
2501 (nd->nd_flag & ND_NFSV4) != 0 &&
2502 nvp->v_type == VDIR &&
2503 vp->v_mount != nvp->v_mount) {
2505 * For a ZFS snapshot, there is a
2506 * pseudo mount that does not set
2507 * v_mountedhere, so it needs to
2508 * be detected via a different
2513 new_mp = nvp->v_mount;
2518 * If we failed to get attributes of the entry,
2519 * then just skip it for NFSv3 (the traditional
2520 * behavior in the old NFS server).
2521 * For NFSv4 the behavior is controlled by
2522 * RDATTRERROR: we either ignore the error or
2524 * Note that RDATTRERROR is never set for NFSv3.
2527 if (!NFSISSET_ATTRBIT(&attrbits,
2528 NFSATTRBIT_RDATTRERROR)) {
2530 if (needs_unbusy != 0)
2532 if ((nd->nd_flag & ND_NFSV3))
2541 * Build the directory record xdr
2543 if (nd->nd_flag & ND_NFSV3) {
2544 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2545 *tl++ = newnfs_true;
2547 *tl = txdr_unsigned(dp->d_fileno);
2548 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2549 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2551 *tl = txdr_unsigned(*cookiep);
2552 nfsrv_postopattr(nd, 0, nvap);
2553 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2554 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2558 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2559 *tl++ = newnfs_true;
2561 *tl = txdr_unsigned(*cookiep);
2562 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2564 supports_nfsv4acls =
2565 nfs_supportsnfsv4acls(nvp);
2568 supports_nfsv4acls = 0;
2570 dirlen += nfsrv_putreferralattr(nd,
2573 if (nd->nd_repstat) {
2576 if (needs_unbusy != 0)
2581 dirlen += nfsvno_fillattr(nd, new_mp,
2582 nvp, nvap, &nfh, r, &rderrbits,
2583 nd->nd_cred, p, isdgram, 0,
2584 supports_nfsv4acls, at_root,
2587 dirlen += nfsvno_fillattr(nd, new_mp,
2588 nvp, nvap, &nfh, r, &attrbits,
2589 nd->nd_cred, p, isdgram, 0,
2590 supports_nfsv4acls, at_root,
2595 dirlen += (3 * NFSX_UNSIGNED);
2597 if (needs_unbusy != 0)
2603 cpos += dp->d_reclen;
2604 dp = (struct dirent *)cpos;
2612 * If dirlen > cnt, we must strip off the last entry. If that
2613 * results in an empty reply, report NFSERR_TOOSMALL.
2615 if (dirlen > cnt || nd->nd_repstat) {
2616 if (!nd->nd_repstat && entrycnt == 0)
2617 nd->nd_repstat = NFSERR_TOOSMALL;
2618 if (nd->nd_repstat) {
2619 newnfs_trimtrailing(nd, mb0, bpos0);
2620 if (nd->nd_flag & ND_NFSV3)
2621 nfsrv_postopattr(nd, getret, &at);
2623 newnfs_trimtrailing(nd, mb1, bpos1);
2625 } else if (cpos < cend)
2627 if (!nd->nd_repstat) {
2628 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2629 *tl++ = newnfs_false;
2635 free(cookies, M_TEMP);
2639 NFSEXITCODE2(0, nd);
2643 NFSEXITCODE2(error, nd);
2648 * Get the settable attributes out of the mbuf list.
2649 * (Return 0 or EBADRPC)
2652 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2653 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2656 struct nfsv2_sattr *sp;
2657 int error = 0, toclient = 0;
2659 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2661 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2663 * Some old clients didn't fill in the high order 16bits.
2664 * --> check the low order 2 bytes for 0xffff
2666 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2667 nvap->na_mode = nfstov_mode(sp->sa_mode);
2668 if (sp->sa_uid != newnfs_xdrneg1)
2669 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2670 if (sp->sa_gid != newnfs_xdrneg1)
2671 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2672 if (sp->sa_size != newnfs_xdrneg1)
2673 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2674 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2676 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2678 nvap->na_atime.tv_sec =
2679 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2680 nvap->na_atime.tv_nsec = 0;
2683 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2684 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2687 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2688 if (*tl == newnfs_true) {
2689 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2690 nvap->na_mode = nfstov_mode(*tl);
2692 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2693 if (*tl == newnfs_true) {
2694 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2695 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2697 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2698 if (*tl == newnfs_true) {
2699 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2700 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2702 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2703 if (*tl == newnfs_true) {
2704 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2705 nvap->na_size = fxdr_hyper(tl);
2707 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2708 switch (fxdr_unsigned(int, *tl)) {
2709 case NFSV3SATTRTIME_TOCLIENT:
2710 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2711 fxdr_nfsv3time(tl, &nvap->na_atime);
2714 case NFSV3SATTRTIME_TOSERVER:
2715 vfs_timestamp(&nvap->na_atime);
2716 nvap->na_vaflags |= VA_UTIMES_NULL;
2719 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2720 switch (fxdr_unsigned(int, *tl)) {
2721 case NFSV3SATTRTIME_TOCLIENT:
2722 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2723 fxdr_nfsv3time(tl, &nvap->na_mtime);
2724 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2726 case NFSV3SATTRTIME_TOSERVER:
2727 vfs_timestamp(&nvap->na_mtime);
2729 nvap->na_vaflags |= VA_UTIMES_NULL;
2734 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2737 NFSEXITCODE2(error, nd);
2742 * Handle the setable attributes for V4.
2743 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2746 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2747 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2752 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2753 int moderet, toclient = 0;
2754 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2757 u_short mode, mask; /* Same type as va_mode. */
2760 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2763 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2764 attrsize = fxdr_unsigned(int, *tl);
2767 * Loop around getting the setable attributes. If an unsupported
2768 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2771 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2772 bitpos = NFSATTRBIT_MAX;
2777 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2778 if (attrsum > attrsize) {
2779 error = NFSERR_BADXDR;
2782 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2784 case NFSATTRBIT_SIZE:
2785 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2786 if (vp != NULL && vp->v_type != VREG) {
2787 error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2791 nvap->na_size = fxdr_hyper(tl);
2792 attrsum += NFSX_HYPER;
2794 case NFSATTRBIT_ACL:
2795 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2799 if (aceerr && !nd->nd_repstat)
2800 nd->nd_repstat = aceerr;
2803 case NFSATTRBIT_ARCHIVE:
2804 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2805 if (!nd->nd_repstat)
2806 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2807 attrsum += NFSX_UNSIGNED;
2809 case NFSATTRBIT_HIDDEN:
2810 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2811 if (!nd->nd_repstat)
2812 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2813 attrsum += NFSX_UNSIGNED;
2815 case NFSATTRBIT_MIMETYPE:
2816 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2817 i = fxdr_unsigned(int, *tl);
2818 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2821 if (!nd->nd_repstat)
2822 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2823 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2825 case NFSATTRBIT_MODE:
2826 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */
2827 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2828 nvap->na_mode = nfstov_mode(*tl);
2829 attrsum += NFSX_UNSIGNED;
2831 case NFSATTRBIT_OWNER:
2832 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2833 j = fxdr_unsigned(int, *tl);
2835 error = NFSERR_BADXDR;
2838 if (j > NFSV4_SMALLSTR)
2839 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2842 error = nfsrv_mtostr(nd, cp, j);
2844 if (j > NFSV4_SMALLSTR)
2845 free(cp, M_NFSSTRING);
2848 if (!nd->nd_repstat) {
2849 nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
2851 if (!nd->nd_repstat)
2854 if (j > NFSV4_SMALLSTR)
2855 free(cp, M_NFSSTRING);
2856 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2858 case NFSATTRBIT_OWNERGROUP:
2859 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2860 j = fxdr_unsigned(int, *tl);
2862 error = NFSERR_BADXDR;
2865 if (j > NFSV4_SMALLSTR)
2866 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2869 error = nfsrv_mtostr(nd, cp, j);
2871 if (j > NFSV4_SMALLSTR)
2872 free(cp, M_NFSSTRING);
2875 if (!nd->nd_repstat) {
2876 nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
2878 if (!nd->nd_repstat)
2881 if (j > NFSV4_SMALLSTR)
2882 free(cp, M_NFSSTRING);
2883 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2885 case NFSATTRBIT_SYSTEM:
2886 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2887 if (!nd->nd_repstat)
2888 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2889 attrsum += NFSX_UNSIGNED;
2891 case NFSATTRBIT_TIMEACCESSSET:
2892 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2893 attrsum += NFSX_UNSIGNED;
2894 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2895 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2896 fxdr_nfsv4time(tl, &nvap->na_atime);
2898 attrsum += NFSX_V4TIME;
2900 vfs_timestamp(&nvap->na_atime);
2901 nvap->na_vaflags |= VA_UTIMES_NULL;
2904 case NFSATTRBIT_TIMEBACKUP:
2905 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2906 if (!nd->nd_repstat)
2907 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2908 attrsum += NFSX_V4TIME;
2910 case NFSATTRBIT_TIMECREATE:
2911 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2912 if (!nd->nd_repstat)
2913 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2914 attrsum += NFSX_V4TIME;
2916 case NFSATTRBIT_TIMEMODIFYSET:
2917 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2918 attrsum += NFSX_UNSIGNED;
2919 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2920 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2921 fxdr_nfsv4time(tl, &nvap->na_mtime);
2922 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2923 attrsum += NFSX_V4TIME;
2925 vfs_timestamp(&nvap->na_mtime);
2927 nvap->na_vaflags |= VA_UTIMES_NULL;
2930 case NFSATTRBIT_MODESETMASKED:
2931 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
2932 mode = fxdr_unsigned(u_short, *tl++);
2933 mask = fxdr_unsigned(u_short, *tl);
2935 * vp == NULL implies an Open/Create operation.
2936 * This attribute can only be used for Setattr and
2937 * only for NFSv4.1 or higher.
2938 * If moderet != 0, a mode attribute has also been
2939 * specified and this attribute cannot be done in the
2940 * same Setattr operation.
2942 if ((nd->nd_flag & ND_NFSV41) == 0)
2943 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2944 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
2946 nd->nd_repstat = NFSERR_INVAL;
2947 else if (moderet == 0)
2948 moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
2950 nvap->na_mode = (mode & mask) |
2951 (va.va_mode & ~mask);
2953 nd->nd_repstat = moderet;
2954 attrsum += 2 * NFSX_UNSIGNED;
2957 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2959 * set bitpos so we drop out of the loop.
2961 bitpos = NFSATTRBIT_MAX;
2967 * some clients pad the attrlist, so we need to skip over the
2970 if (attrsum > attrsize) {
2971 error = NFSERR_BADXDR;
2973 attrsize = NFSM_RNDUP(attrsize);
2974 if (attrsum < attrsize)
2975 error = nfsm_advance(nd, attrsize - attrsum, -1);
2978 NFSEXITCODE2(error, nd);
2983 * Check/setup export credentials.
2986 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2987 struct ucred *credanon)
2992 * Check/setup credentials.
2994 if (nd->nd_flag & ND_GSS)
2995 exp->nes_exflag &= ~MNT_EXPORTANON;
2998 * Check to see if the operation is allowed for this security flavor.
2999 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
3000 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
3001 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
3003 if (nfsvno_testexp(nd, exp) &&
3004 nd->nd_procnum != NFSV4OP_SECINFO &&
3005 nd->nd_procnum != NFSPROC_FSINFO) {
3006 if (nd->nd_flag & ND_NFSV4)
3007 error = NFSERR_WRONGSEC;
3009 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3014 * Check to see if the file system is exported V4 only.
3016 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3017 error = NFSERR_PROGNOTV4;
3022 * Now, map the user credentials.
3023 * (Note that ND_AUTHNONE will only be set for an NFSv3
3024 * Fsinfo RPC. If set for anything else, this code might need
3027 if (NFSVNO_EXPORTED(exp)) {
3028 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3029 NFSVNO_EXPORTANON(exp) ||
3030 (nd->nd_flag & ND_AUTHNONE) != 0) {
3031 nd->nd_cred->cr_uid = credanon->cr_uid;
3032 nd->nd_cred->cr_gid = credanon->cr_gid;
3033 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3034 credanon->cr_groups);
3035 } else if ((nd->nd_flag & ND_GSS) == 0) {
3037 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3038 * if there is a replacement credential with a group
3039 * list set up by "nfsuserd -manage-gids".
3040 * If there is no replacement, nfsrv_getgrpscred()
3041 * simply returns its argument.
3043 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3048 NFSEXITCODE2(error, nd);
3056 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3057 struct ucred **credp)
3061 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3062 &exp->nes_numsecflavor, exp->nes_secflavors);
3064 if (nfs_rootfhset) {
3065 exp->nes_exflag = 0;
3066 exp->nes_numsecflavor = 0;
3069 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3071 printf("nfsvno_checkexp: numsecflavors out of range\n");
3072 exp->nes_numsecflavor = 0;
3080 * Get a vnode for a file handle and export stuff.
3083 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3084 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3085 struct ucred **credp)
3090 exp->nes_numsecflavor = 0;
3091 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3093 /* Make sure the server replies ESTALE to the client. */
3095 if (nam && !error) {
3096 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3097 &exp->nes_numsecflavor, exp->nes_secflavors);
3099 if (nfs_rootfhset) {
3100 exp->nes_exflag = 0;
3101 exp->nes_numsecflavor = 0;
3106 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3108 printf("nfsvno_fhtovp: numsecflavors out of range\n");
3109 exp->nes_numsecflavor = 0;
3119 * nfsd_fhtovp() - convert a fh to a vnode ptr
3120 * - look up fsid in mount list (if not found ret error)
3121 * - get vp and export rights by calling nfsvno_fhtovp()
3122 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3124 * - if mpp != NULL, return the mount point so that it can
3125 * be used for vn_finished_write() by the caller
3128 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3129 struct vnode **vpp, struct nfsexstuff *exp,
3130 struct mount **mpp, int startwrite)
3133 struct ucred *credanon;
3136 fhp = (fhandle_t *)nfp->nfsrvfh_data;
3138 * Check for the special case of the nfsv4root_fh.
3140 mp = vfs_busyfs(&fhp->fh_fsid);
3145 nd->nd_repstat = ESTALE;
3150 vn_start_write(NULL, mpp, V_WAIT);
3151 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3152 lktype = LK_EXCLUSIVE;
3154 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3159 * For NFSv4 without a pseudo root fs, unexported file handles
3160 * can be returned, so that Lookup works everywhere.
3162 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3163 !(nd->nd_flag & ND_NFSV4)) {
3165 nd->nd_repstat = EACCES;
3169 * Personally, I've never seen any point in requiring a
3170 * reserved port#, since only in the rare case where the
3171 * clients are all boxes with secure system privileges,
3172 * does it provide any enhanced security, but... some people
3173 * believe it to be useful and keep putting this code back in.
3174 * (There is also some "security checker" out there that
3175 * complains if the nfs server doesn't enforce this.)
3176 * However, note the following:
3177 * RFC3530 (NFSv4) specifies that a reserved port# not be
3179 * RFC2623 recommends that, if a reserved port# is checked for,
3180 * that there be a way to turn that off--> ifdef'd.
3182 #ifdef NFS_REQRSVPORT
3183 if (!nd->nd_repstat) {
3184 struct sockaddr_in *saddr;
3185 struct sockaddr_in6 *saddr6;
3187 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3188 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3189 if (!(nd->nd_flag & ND_NFSV4) &&
3190 ((saddr->sin_family == AF_INET &&
3191 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3192 (saddr6->sin6_family == AF_INET6 &&
3193 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3195 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3198 #endif /* NFS_REQRSVPORT */
3201 * Check/setup credentials.
3203 if (!nd->nd_repstat) {
3204 nd->nd_saveduid = nd->nd_cred->cr_uid;
3205 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
3209 if (credanon != NULL)
3211 if (nd->nd_repstat) {
3213 vn_finished_write(mp);
3220 NFSEXITCODE2(0, nd);
3227 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3229 struct filedesc *fdp;
3233 fdp = p->td_proc->p_fd;
3234 if (fd < 0 || fd >= fdp->fd_nfiles ||
3235 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3247 * Called from nfssvc() to update the exports list. Just call
3248 * vfs_export(). This has to be done, since the v4 root fake fs isn't
3249 * in the mount list.
3252 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3254 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3256 struct nameidata nd;
3259 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3260 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3262 else if (error == 0) {
3263 if (nfsexargp->fspec == NULL) {
3268 * If fspec != NULL, this is the v4root path.
3270 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3271 nfsexargp->fspec, p);
3272 if ((error = namei(&nd)) != 0)
3274 error = nfsvno_getfh(nd.ni_vp, &fh, p);
3277 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3278 NFSBCOPY((caddr_t)&fh,
3279 nfs_rootfh.nfsrvfh_data,
3280 sizeof (fhandle_t));
3291 * This function needs to test to see if the system is near its limit
3292 * for memory allocation via malloc() or mget() and return True iff
3293 * either of these resources are near their limit.
3294 * XXX (For now, this is just a stub.)
3296 int nfsrv_testmalloclimit = 0;
3298 nfsrv_mallocmget_limit(void)
3300 static int printmesg = 0;
3301 static int testval = 1;
3303 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3304 if ((printmesg++ % 100) == 0)
3305 printf("nfsd: malloc/mget near limit\n");
3312 * BSD specific initialization of a mount point.
3317 static int inited = 0;
3322 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3323 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3324 TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3325 nfsv4root_mnt.mnt_export = NULL;
3326 TAILQ_INIT(&nfsv4root_opt);
3327 TAILQ_INIT(&nfsv4root_newopt);
3328 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3329 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3330 nfsv4root_mnt.mnt_nvnodelistsize = 0;
3331 nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3335 * Get a vnode for a file handle, without checking exports, etc.
3338 nfsvno_getvp(fhandle_t *fhp)
3344 mp = vfs_busyfs(&fhp->fh_fsid);
3347 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3355 * Do a local VOP_ADVLOCK().
3358 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3359 u_int64_t end, struct thread *td)
3365 if (nfsrv_dolocallocks == 0)
3367 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3369 fl.l_whence = SEEK_SET;
3371 fl.l_start = (off_t)first;
3372 if (end == NFS64BITSSET) {
3376 fl.l_len = (off_t)tlen;
3379 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3380 * values for all calls, so that all locks will be held by the
3381 * nfsd server. (The nfsd server handles conflicts between the
3383 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3384 * bytes, so it can't be put in l_sysid.
3386 if (nfsv4_sysid == 0)
3387 nfsv4_sysid = nlm_acquire_next_sysid();
3388 fl.l_pid = (pid_t)0;
3389 fl.l_sysid = (int)nfsv4_sysid;
3391 if (ftype == F_UNLCK)
3392 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3393 (F_POSIX | F_REMOTE));
3395 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3396 (F_POSIX | F_REMOTE));
3404 * Check the nfsv4 root exports.
3407 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3409 struct ucred *credanon;
3410 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3413 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3414 &credanon, &numsecflavor, secflavors);
3416 error = NFSERR_PROGUNAVAIL;
3419 if (credanon != NULL)
3421 for (i = 0; i < numsecflavor; i++) {
3422 if (secflavors[i] == AUTH_SYS)
3423 nd->nd_flag |= ND_EXAUTHSYS;
3424 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3425 nd->nd_flag |= ND_EXGSS;
3426 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3427 nd->nd_flag |= ND_EXGSSINTEGRITY;
3428 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3429 nd->nd_flag |= ND_EXGSSPRIVACY;
3438 * Nfs server pseudo system call for the nfsd's
3444 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3447 struct nfsd_addsock_args sockarg;
3448 struct nfsd_nfsd_args nfsdarg;
3449 struct nfsd_nfsd_oargs onfsdarg;
3450 struct nfsd_pnfsd_args pnfsdarg;
3451 struct vnode *vp, *nvp, *curdvp;
3452 struct pnfsdsfile *pf;
3453 struct nfsdevice *ds, *fds;
3454 cap_rights_t rights;
3455 int buflen, error, ret;
3456 char *buf, *cp, *cp2, *cp3;
3457 char fname[PNFS_FILENAME_LEN + 1];
3459 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3460 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3464 * Since we don't know what rights might be required,
3465 * pretend that we need them all. It is better to be too
3466 * careful than too reckless.
3468 error = fget(td, sockarg.sock,
3469 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3472 if (fp->f_type != DTYPE_SOCKET) {
3477 error = nfsrvd_addsock(fp);
3479 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3480 if (uap->argp == NULL) {
3484 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3485 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3487 nfsdarg.principal = onfsdarg.principal;
3488 nfsdarg.minthreads = onfsdarg.minthreads;
3489 nfsdarg.maxthreads = onfsdarg.maxthreads;
3490 nfsdarg.version = 1;
3491 nfsdarg.addr = NULL;
3492 nfsdarg.addrlen = 0;
3493 nfsdarg.dnshost = NULL;
3494 nfsdarg.dnshostlen = 0;
3495 nfsdarg.dspath = NULL;
3496 nfsdarg.dspathlen = 0;
3497 nfsdarg.mdspath = NULL;
3498 nfsdarg.mdspathlen = 0;
3499 nfsdarg.mirrorcnt = 1;
3502 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3505 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3506 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3507 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3508 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3509 nfsdarg.mirrorcnt >= 1 &&
3510 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3511 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3512 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3513 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3514 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3515 nfsdarg.dspathlen, nfsdarg.dnshostlen,
3516 nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3517 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3518 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3523 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */
3525 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3526 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3528 free(nfsdarg.addr, M_TEMP);
3532 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */
3533 nfsdarg.dnshost = cp;
3534 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3535 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3537 free(nfsdarg.addr, M_TEMP);
3538 free(nfsdarg.dnshost, M_TEMP);
3542 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */
3543 nfsdarg.dspath = cp;
3544 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3545 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3547 free(nfsdarg.addr, M_TEMP);
3548 free(nfsdarg.dnshost, M_TEMP);
3549 free(nfsdarg.dspath, M_TEMP);
3553 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */
3554 nfsdarg.mdspath = cp;
3556 nfsdarg.addr = NULL;
3557 nfsdarg.addrlen = 0;
3558 nfsdarg.dnshost = NULL;
3559 nfsdarg.dnshostlen = 0;
3560 nfsdarg.dspath = NULL;
3561 nfsdarg.dspathlen = 0;
3562 nfsdarg.mdspath = NULL;
3563 nfsdarg.mdspathlen = 0;
3564 nfsdarg.mirrorcnt = 1;
3566 error = nfsrvd_nfsd(td, &nfsdarg);
3567 free(nfsdarg.addr, M_TEMP);
3568 free(nfsdarg.dnshost, M_TEMP);
3569 free(nfsdarg.dspath, M_TEMP);
3570 free(nfsdarg.mdspath, M_TEMP);
3571 } else if (uap->flag & NFSSVC_PNFSDS) {
3572 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3573 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3574 pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3575 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3576 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3579 error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3581 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3582 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3583 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3584 buf = malloc(buflen, M_TEMP, M_WAITOK);
3585 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3587 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3588 if (error == 0 && pnfsdarg.dspath != NULL) {
3589 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3590 error = copyinstr(pnfsdarg.dspath, cp2,
3591 PATH_MAX + 1, NULL);
3592 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3596 if (error == 0 && pnfsdarg.curdspath != NULL) {
3597 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3598 error = copyinstr(pnfsdarg.curdspath, cp3,
3599 PATH_MAX + 1, NULL);
3600 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3607 error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3608 &buflen, fname, td, &vp, &nvp, &pf, &ds,
3610 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3612 if (pf->dsf_dir >= nfsrv_dsdirsize) {
3613 printf("copymr: dsdir out of range\n");
3616 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3617 error = nfsrv_copymr(vp, nvp,
3618 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3619 (struct pnfsdsfile *)buf,
3620 buflen / sizeof(*pf), td->td_ucred, td);
3623 if (fds != NULL && error == 0) {
3624 curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3625 ret = vn_lock(curdvp, LK_EXCLUSIVE);
3627 nfsrv_dsremove(curdvp, fname,
3629 NFSVOPUNLOCK(curdvp);
3632 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3640 error = nfssvc_srvcall(td, uap, td->td_ucred);
3649 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3651 struct nfsex_args export;
3652 struct nfsex_oldargs oexp;
3653 struct file *fp = NULL;
3654 int stablefd, i, len;
3655 struct nfsd_clid adminrevoke;
3656 struct nfsd_dumplist dumplist;
3657 struct nfsd_dumpclients *dumpclients;
3658 struct nfsd_dumplocklist dumplocklist;
3659 struct nfsd_dumplocks *dumplocks;
3660 struct nameidata nd;
3662 int error = EINVAL, igotlock;
3665 static int suspend_nfsd = 0;
3667 if (uap->flag & NFSSVC_PUBLICFH) {
3668 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3669 sizeof (fhandle_t));
3670 error = copyin(uap->argp,
3671 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3674 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3675 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3676 error = copyin(uap->argp,(caddr_t)&export,
3677 sizeof (struct nfsex_args));
3680 if (export.export.ex_ngroups > NGROUPS_MAX ||
3681 export.export.ex_ngroups < 0)
3683 else if (export.export.ex_ngroups > 0) {
3684 grps = malloc(export.export.ex_ngroups *
3685 sizeof(gid_t), M_TEMP, M_WAITOK);
3686 error = copyin(export.export.ex_groups, grps,
3687 export.export.ex_ngroups * sizeof(gid_t));
3688 export.export.ex_groups = grps;
3690 export.export.ex_groups = NULL;
3692 error = nfsrv_v4rootexport(&export, cred, p);
3695 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3696 NFSSVC_V4ROOTEXPORT) {
3697 error = copyin(uap->argp,(caddr_t)&oexp,
3698 sizeof (struct nfsex_oldargs));
3700 memset(&export.export, 0, sizeof(export.export));
3701 export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3702 export.export.ex_root = oexp.export.ex_root;
3703 export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3704 export.export.ex_ngroups =
3705 oexp.export.ex_anon.cr_ngroups;
3706 export.export.ex_groups = NULL;
3707 if (export.export.ex_ngroups > XU_NGROUPS ||
3708 export.export.ex_ngroups < 0)
3710 else if (export.export.ex_ngroups > 0) {
3711 export.export.ex_groups = malloc(
3712 export.export.ex_ngroups * sizeof(gid_t),
3714 for (i = 0; i < export.export.ex_ngroups; i++)
3715 export.export.ex_groups[i] =
3716 oexp.export.ex_anon.cr_groups[i];
3718 export.export.ex_addr = oexp.export.ex_addr;
3719 export.export.ex_addrlen = oexp.export.ex_addrlen;
3720 export.export.ex_mask = oexp.export.ex_mask;
3721 export.export.ex_masklen = oexp.export.ex_masklen;
3722 export.export.ex_indexfile = oexp.export.ex_indexfile;
3723 export.export.ex_numsecflavors =
3724 oexp.export.ex_numsecflavors;
3725 if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3726 export.export.ex_numsecflavors < 0)
3729 for (i = 0; i < export.export.ex_numsecflavors;
3731 export.export.ex_secflavors[i] =
3732 oexp.export.ex_secflavors[i];
3734 export.fspec = oexp.fspec;
3736 error = nfsrv_v4rootexport(&export, cred, p);
3737 free(export.export.ex_groups, M_TEMP);
3739 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3742 } else if (uap->flag & NFSSVC_STABLERESTART) {
3743 error = copyin(uap->argp, (caddr_t)&stablefd,
3746 error = fp_getfvp(p, stablefd, &fp, &vp);
3747 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3749 if (!error && newnfs_numnfsd != 0)
3752 nfsrv_stablefirst.nsf_fp = fp;
3753 nfsrv_setupstable(p);
3755 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3756 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3757 sizeof (struct nfsd_clid));
3759 error = nfsrv_adminrevoke(&adminrevoke, p);
3760 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3761 error = copyin(uap->argp, (caddr_t)&dumplist,
3762 sizeof (struct nfsd_dumplist));
3763 if (!error && (dumplist.ndl_size < 1 ||
3764 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3767 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3768 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3769 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3770 error = copyout(dumpclients, dumplist.ndl_list, len);
3771 free(dumpclients, M_TEMP);
3773 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3774 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3775 sizeof (struct nfsd_dumplocklist));
3776 if (!error && (dumplocklist.ndllck_size < 1 ||
3777 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3780 error = nfsrv_lookupfilename(&nd,
3781 dumplocklist.ndllck_fname, p);
3783 len = sizeof (struct nfsd_dumplocks) *
3784 dumplocklist.ndllck_size;
3785 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3786 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3787 dumplocklist.ndllck_size, p);
3789 error = copyout(dumplocks, dumplocklist.ndllck_list,
3791 free(dumplocks, M_TEMP);
3793 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3796 nfsd_master_pid = procp->p_pid;
3797 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3798 nfsd_master_start = procp->p_stats->p_start;
3799 nfsd_master_proc = procp;
3801 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3802 NFSLOCKV4ROOTMUTEX();
3803 if (suspend_nfsd == 0) {
3804 /* Lock out all nfsd threads */
3806 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3807 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3808 } while (igotlock == 0 && suspend_nfsd == 0);
3811 NFSUNLOCKV4ROOTMUTEX();
3813 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3814 NFSLOCKV4ROOTMUTEX();
3815 if (suspend_nfsd != 0) {
3816 nfsv4_unlock(&nfsd_suspend_lock, 0);
3819 NFSUNLOCKV4ROOTMUTEX();
3829 * Returns 0 if ok, 1 otherwise.
3832 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3837 * This seems odd, but allow the case where the security flavor
3838 * list is empty. This happens when NFSv4 is traversing non-exported
3839 * file systems. Exported file systems should always have a non-empty
3840 * security flavor list.
3842 if (exp->nes_numsecflavor == 0)
3845 for (i = 0; i < exp->nes_numsecflavor; i++) {
3847 * The tests for privacy and integrity must be first,
3848 * since ND_GSS is set for everything but AUTH_SYS.
3850 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3851 (nd->nd_flag & ND_GSSPRIVACY))
3853 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3854 (nd->nd_flag & ND_GSSINTEGRITY))
3856 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3857 (nd->nd_flag & ND_GSS))
3859 if (exp->nes_secflavors[i] == AUTH_SYS &&
3860 (nd->nd_flag & ND_GSS) == 0)
3867 * Calculate a hash value for the fid in a file handle.
3870 nfsrv_hashfh(fhandle_t *fhp)
3874 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3879 * Calculate a hash value for the sessionid.
3882 nfsrv_hashsessionid(uint8_t *sessionid)
3886 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3891 * Signal the userland master nfsd to backup the stable restart file.
3894 nfsrv_backupstable(void)
3898 if (nfsd_master_proc != NULL) {
3899 procp = pfind(nfsd_master_pid);
3900 /* Try to make sure it is the correct process. */
3901 if (procp == nfsd_master_proc &&
3902 procp->p_stats->p_start.tv_sec ==
3903 nfsd_master_start.tv_sec &&
3904 procp->p_stats->p_start.tv_usec ==
3905 nfsd_master_start.tv_usec &&
3906 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3907 kern_psignal(procp, SIGUSR2);
3909 nfsd_master_proc = NULL;
3917 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
3918 * The arguments are in a structure, so that they can be passed through
3919 * taskqueue for a kernel process to execute this function.
3921 struct nfsrvdscreate {
3925 struct ucred *tcred;
3928 struct pnfsdsfile *pf;
3932 struct vattr createva;
3936 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
3937 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
3938 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
3941 struct nameidata named;
3946 struct nfsmount *nmp;
3949 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
3950 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
3951 nfsvno_setpathbuf(&named, &bufp, &hashp);
3952 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
3953 named.ni_cnd.cn_thread = p;
3954 named.ni_cnd.cn_nameptr = bufp;
3955 if (fnamep != NULL) {
3956 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
3957 named.ni_cnd.cn_namelen = strlen(bufp);
3959 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
3960 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
3962 /* Create the date file in the DS mount. */
3963 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
3965 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
3968 /* Set the ownership of the file. */
3969 error = VOP_SETATTR(nvp, nvap, tcred);
3970 NFSD_DEBUG(4, "nfsrv_dscreate:"
3971 " setattr-uid=%d\n", error);
3976 printf("pNFS: pnfscreate failed=%d\n", error);
3978 printf("pNFS: pnfscreate vnlock=%d\n", error);
3981 nmp = VFSTONFS(nvp->v_mount);
3982 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
3983 != 0 || nmp->nm_nam->sa_len > sizeof(
3984 struct sockaddr_in6) ||
3985 np->n_fhp->nfh_len != NFSX_MYFH) {
3986 printf("Bad DS file: fstype=%s salen=%d"
3988 nvp->v_mount->mnt_vfc->vfc_name,
3989 nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
3993 /* Set extattrs for the DS on the MDS file. */
3996 error = VOP_GETATTR(nvp, &va, tcred);
3998 dsa->dsa_filerev = va.va_filerev;
3999 dsa->dsa_size = va.va_size;
4000 dsa->dsa_atime = va.va_atime;
4001 dsa->dsa_mtime = va.va_mtime;
4002 dsa->dsa_bytes = va.va_bytes;
4006 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4008 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4009 nmp->nm_nam->sa_len);
4010 NFSBCOPY(named.ni_cnd.cn_nameptr,
4012 sizeof(pf->dsf_filename));
4015 printf("pNFS: pnfscreate can't get DS"
4016 " attr=%d\n", error);
4017 if (nvpp != NULL && error == 0)
4022 nfsvno_relpathbuf(&named);
4027 * Start up the thread that will execute nfsrv_dscreate().
4030 start_dscreate(void *arg, int pending)
4032 struct nfsrvdscreate *dsc;
4034 dsc = (struct nfsrvdscreate *)arg;
4035 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4036 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4038 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4042 * Create a pNFS data file on the Data Server(s).
4045 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4048 struct nfsrvdscreate *dsc, *tdsc = NULL;
4049 struct nfsdevice *ds, *tds, *fds;
4051 struct pnfsdsfile *pf, *tpf;
4052 struct pnfsdsattr dsattr;
4054 struct vnode *dvp[NFSDEV_MAXMIRRORS];
4055 struct nfsmount *nmp;
4060 struct ucred *tcred;
4061 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4064 /* Get a DS server directory in a round-robin order. */
4070 * Search for the first entry that handles this MDS fs, but use the
4071 * first entry for all MDS fs's otherwise.
4073 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4074 if (tds->nfsdev_nmp != NULL) {
4075 if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4077 else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4078 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4086 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4089 i = dsdir[0] = ds->nfsdev_nextdir;
4090 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4091 dvp[0] = ds->nfsdev_dsdir[i];
4092 tds = TAILQ_NEXT(ds, nfsdev_list);
4093 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4094 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4095 if (tds->nfsdev_nmp != NULL &&
4096 ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4097 (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4098 fsidcmp(&mp->mnt_stat.f_fsid,
4099 &tds->nfsdev_mdsfsid) == 0))) {
4100 dsdir[mirrorcnt] = i;
4101 dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4103 if (mirrorcnt >= nfsrv_maxpnfsmirror)
4108 /* Put at end of list to implement round-robin usage. */
4109 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4110 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4114 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4116 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4119 error = nfsvno_getfh(vp, &fh, p);
4121 error = VOP_GETATTR(vp, &va, cred);
4123 /* Set the attributes for "vp" to Setattr the DS vp. */
4126 vamode = va.va_mode;
4130 va.va_mode = vamode;
4133 printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4135 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4137 /* Make data file name based on FH. */
4138 tcred = newnfs_getcred();
4141 * Create the file on each DS mirror, using kernel process(es) for the
4142 * additional mirrors.
4145 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4146 tpf->dsf_dir = dsdir[i];
4147 tdsc->tcred = tcred;
4150 tdsc->createva = *vap;
4151 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4158 if (nfs_pnfsiothreads != 0) {
4159 ret = nfs_pnfsio(start_dscreate, tdsc);
4160 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4163 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4164 NULL, tcred, p, NULL);
4166 KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4168 if (failpos == -1 && nfsds_failerr(ret))
4176 tpf->dsf_dir = dsdir[mirrorcnt - 1];
4177 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4178 &dsattr, NULL, tcred, p, NULL);
4179 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4180 failpos = mirrorcnt - 1;
4184 timo = hz / 50; /* Wait for 20msec. */
4187 /* Wait for kernel task(s) to complete. */
4188 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4189 while (tdsc->inprog != 0 && tdsc->done == 0)
4190 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4191 if (tdsc->err != 0) {
4192 if (failpos == -1 && nfsds_failerr(tdsc->err))
4194 else if (error == 0)
4200 * If failpos has been set, that mirror has failed, so it needs
4204 nmp = VFSTONFS(dvp[failpos]->v_mount);
4206 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4207 NFSMNTP_CANCELRPCS)) == 0) {
4208 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4210 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4211 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4214 nfsrv_killrpcs(nmp);
4216 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4224 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4226 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4227 mirrorcnt, nfsrv_maxpnfsmirror);
4229 * For all mirrors that couldn't be created, fill in the
4230 * *pf structure, but with an IP address == 0.0.0.0.
4232 tpf = pf + mirrorcnt;
4233 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4235 tpf->dsf_sin.sin_family = AF_INET;
4236 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4237 tpf->dsf_sin.sin_addr.s_addr = 0;
4238 tpf->dsf_sin.sin_port = 0;
4241 error = vn_extattr_set(vp, IO_NODELOCKED,
4242 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4243 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4245 error = vn_extattr_set(vp, IO_NODELOCKED,
4246 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4247 sizeof(dsattr), (char *)&dsattr, p);
4249 printf("pNFS: pnfscreate setextattr=%d\n",
4252 printf("pNFS: pnfscreate=%d\n", error);
4258 * Get the information needed to remove the pNFS Data Server file from the
4259 * Metadata file. Upon success, ddvp is set non-NULL to the locked
4260 * DS directory vnode. The caller must unlock *ddvp when done with it.
4263 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4264 int *mirrorcntp, char *fname, fhandle_t *fhp)
4267 struct ucred *tcred;
4272 /* If not an exported regular file or not a pNFS server, just return. */
4273 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4274 nfsrv_devidcnt == 0)
4277 /* Check to see if this is the last hard link. */
4278 tcred = newnfs_getcred();
4279 error = VOP_GETATTR(vp, &va, tcred);
4282 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4285 if (va.va_nlink > 1)
4288 error = nfsvno_getfh(vp, fhp, p);
4290 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4295 buf = malloc(buflen, M_TEMP, M_WAITOK);
4296 /* Get the directory vnode for the DS mount and the file handle. */
4297 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4298 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4301 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4305 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4306 * The arguments are in a structure, so that they can be passed through
4307 * taskqueue for a kernel process to execute this function.
4309 struct nfsrvdsremove {
4313 struct ucred *tcred;
4317 char fname[PNFS_FILENAME_LEN + 1];
4321 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4324 struct nameidata named;
4330 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4333 named.ni_cnd.cn_nameiop = DELETE;
4334 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4335 named.ni_cnd.cn_cred = tcred;
4336 named.ni_cnd.cn_thread = p;
4337 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4338 nfsvno_setpathbuf(&named, &bufp, &hashp);
4339 named.ni_cnd.cn_nameptr = bufp;
4340 named.ni_cnd.cn_namelen = strlen(fname);
4341 strlcpy(bufp, fname, NAME_MAX);
4342 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4343 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4344 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4346 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4350 nfsvno_relpathbuf(&named);
4352 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4357 * Start up the thread that will execute nfsrv_dsremove().
4360 start_dsremove(void *arg, int pending)
4362 struct nfsrvdsremove *dsrm;
4364 dsrm = (struct nfsrvdsremove *)arg;
4365 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4368 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4372 * Remove a pNFS data file from a Data Server.
4373 * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4374 * removed to set up the dvp and fill in the FH.
4377 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4380 struct ucred *tcred;
4381 struct nfsrvdsremove *dsrm, *tdsrm;
4382 struct nfsdevice *ds;
4383 struct nfsmount *nmp;
4384 int failpos, i, ret, timo;
4386 tcred = newnfs_getcred();
4389 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4391 * Remove the file on each DS mirror, using kernel process(es) for the
4392 * additional mirrors.
4395 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4396 tdsrm->tcred = tcred;
4398 tdsrm->dvp = dvp[i];
4399 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4404 if (nfs_pnfsiothreads != 0) {
4405 ret = nfs_pnfsio(start_dsremove, tdsrm);
4406 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4409 ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4410 if (failpos == -1 && nfsds_failerr(ret))
4414 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4415 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4416 failpos = mirrorcnt - 1;
4417 timo = hz / 50; /* Wait for 20msec. */
4420 /* Wait for kernel task(s) to complete. */
4421 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4422 while (tdsrm->inprog != 0 && tdsrm->done == 0)
4423 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4424 if (failpos == -1 && nfsds_failerr(tdsrm->err))
4429 * If failpos has been set, that mirror has failed, so it needs
4433 nmp = VFSTONFS(dvp[failpos]->v_mount);
4435 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4436 NFSMNTP_CANCELRPCS)) == 0) {
4437 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4439 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4440 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4443 nfsrv_killrpcs(nmp);
4445 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4451 /* Get rid all layouts for the file. */
4452 nfsrv_freefilelayouts(fhp);
4459 * Generate a file name based on the file handle and put it in *bufp.
4460 * Return the number of bytes generated.
4463 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4467 const uint8_t *hexdigits = "0123456789abcdef";
4469 cp = (uint8_t *)fhp;
4470 for (i = 0; i < sizeof(*fhp); i++) {
4471 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4472 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4479 * Update the Metadata file's attributes from the DS file when a Read/Write
4480 * layout is returned.
4481 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4482 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4485 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4487 struct ucred *tcred;
4490 /* Do this as root so that it won't fail with EACCES. */
4491 tcred = newnfs_getcred();
4492 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4493 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4499 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4502 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4507 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4508 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4513 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4514 struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4515 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4516 off_t *offp, int content, bool *eofp)
4518 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4519 fhandle_t fh[NFSDEV_MAXMIRRORS];
4520 struct vnode *dvp[NFSDEV_MAXMIRRORS];
4521 struct nfsdevice *ds;
4522 struct pnfsdsattr dsattr;
4523 struct opnfsdsattr odsattr;
4525 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4527 NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4529 * If not a regular file, not exported or not a pNFS server,
4530 * just return ENOENT.
4532 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4533 nfsrv_devidcnt == 0)
4537 buf = malloc(buflen, M_TEMP, M_WAITOK);
4541 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4542 * from the MetaData file's extended attribute.
4544 if (ioproc == NFSPROC_GETATTR) {
4545 error = vn_extattr_get(vp, IO_NODELOCKED,
4546 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4549 if (buflen == sizeof(odsattr)) {
4550 NFSBCOPY(buf, &odsattr, buflen);
4551 nap->na_filerev = odsattr.dsa_filerev;
4552 nap->na_size = odsattr.dsa_size;
4553 nap->na_atime = odsattr.dsa_atime;
4554 nap->na_mtime = odsattr.dsa_mtime;
4556 * Fake na_bytes by rounding up na_size.
4557 * Since we don't know the block size, just
4558 * use BLKDEV_IOSIZE.
4560 nap->na_bytes = (odsattr.dsa_size +
4561 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4562 } else if (buflen == sizeof(dsattr)) {
4563 NFSBCOPY(buf, &dsattr, buflen);
4564 nap->na_filerev = dsattr.dsa_filerev;
4565 nap->na_size = dsattr.dsa_size;
4566 nap->na_atime = dsattr.dsa_atime;
4567 nap->na_mtime = dsattr.dsa_mtime;
4568 nap->na_bytes = dsattr.dsa_bytes;
4574 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4575 * returns 0, just return now. nfsrv_checkdsattr()
4576 * returns 0 if there is no Read/Write layout
4577 * plus either an Open/Write_access or Write
4578 * delegation issued to a client for the file.
4580 if (nfsrv_pnfsgetdsattr == 0 ||
4581 nfsrv_checkdsattr(vp, p) == 0) {
4588 * Clear ENOATTR so the code below will attempt to do a
4589 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4590 * the extended attribute.
4592 if (error == ENOATTR)
4601 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4603 printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4604 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4605 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4608 for (i = 0; i < mirrorcnt; i++)
4609 nmp[i] = VFSTONFS(dvp[i]->v_mount);
4611 printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4613 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4616 if (origmircnt == -1)
4617 origmircnt = mirrorcnt;
4619 * If failpos is set to a mirror#, then that mirror has
4620 * failed and will be disabled. For Read, Getattr and Seek, the
4621 * function only tries one mirror, so if that mirror has
4622 * failed, it will need to be retried. As such, increment
4623 * tryitagain for these cases.
4624 * For Write, Setattr and Setacl, the function tries all
4625 * mirrors and will not return an error for the case where
4626 * one mirror has failed. For these cases, the functioning
4627 * mirror(s) will have been modified, so a retry isn't
4628 * necessary. These functions will set failpos for the
4631 if (ioproc == NFSPROC_READDS) {
4632 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4634 if (nfsds_failerr(error) && mirrorcnt > 1) {
4636 * Setting failpos will cause the mirror
4637 * to be disabled and then a retry of this
4644 } else if (ioproc == NFSPROC_WRITEDS)
4645 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4646 &nmp[0], mirrorcnt, mpp, cp, &failpos);
4647 else if (ioproc == NFSPROC_SETATTR)
4648 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4649 mirrorcnt, nap, &failpos);
4650 else if (ioproc == NFSPROC_SETACL)
4651 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4652 mirrorcnt, aclp, &failpos);
4653 else if (ioproc == NFSPROC_SEEKDS) {
4654 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4656 if (nfsds_failerr(error) && mirrorcnt > 1) {
4658 * Setting failpos will cause the mirror
4659 * to be disabled and then a retry of this
4666 } else if (ioproc == NFSPROC_ALLOCATE)
4667 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4668 &nmp[0], mirrorcnt, &failpos);
4670 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4671 vp, nmp[mirrorcnt - 1], nap);
4672 if (nfsds_failerr(error) && mirrorcnt > 1) {
4674 * Setting failpos will cause the mirror
4675 * to be disabled and then a retry of this
4676 * getattr is required.
4678 failpos = mirrorcnt - 1;
4685 failnmp = nmp[failpos];
4686 NFSLOCKMNT(failnmp);
4687 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4688 NFSMNTP_CANCELRPCS)) == 0) {
4689 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4690 NFSUNLOCKMNT(failnmp);
4691 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4693 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4696 nfsrv_killrpcs(failnmp);
4697 NFSLOCKMNT(failnmp);
4698 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4701 NFSUNLOCKMNT(failnmp);
4703 for (i = 0; i < mirrorcnt; i++)
4704 NFSVOPUNLOCK(dvp[i]);
4705 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4707 /* Try the Read/Getattr again if a mirror was deleted. */
4708 if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4711 /* Return ENOENT for any Extended Attribute error. */
4715 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4720 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4722 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4723 * to be checked. If it points to a NULL nmp, then it returns
4724 * a suitable destination.
4725 * curnmp - If non-NULL, it is the source mount for the copy.
4728 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4729 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4730 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4731 struct nfsmount *curnmp, int *ippos, int *dsdirp)
4733 struct vnode *dvp, *nvp = NULL, **tdvpp;
4735 struct nfsmount *nmp, *newnmp;
4736 struct sockaddr *sad;
4737 struct sockaddr_in *sin;
4738 struct nfsdevice *ds, *tds, *fndds;
4739 struct pnfsdsfile *pf;
4741 int error, fhiszero, fnd, gotone, i, mirrorcnt;
4743 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4752 if (newnmpp != NULL)
4757 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4758 "pnfsd.dsfile", buflenp, buf, p);
4759 mirrorcnt = *buflenp / sizeof(*pf);
4760 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4761 *buflenp != sizeof(*pf) * mirrorcnt))
4764 pf = (struct pnfsdsfile *)buf;
4765 /* If curnmp != NULL, check for a match in the mirror list. */
4766 if (curnmp != NULL && error == 0) {
4768 for (i = 0; i < mirrorcnt; i++, pf++) {
4769 sad = (struct sockaddr *)&pf->dsf_sin;
4770 if (nfsaddr2_match(sad, curnmp->nm_nam)) {
4782 pf = (struct pnfsdsfile *)buf;
4783 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
4785 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
4787 sad = (struct sockaddr *)&pf->dsf_sin;
4789 dsdir = pf->dsf_dir;
4790 if (dsdir >= nfsrv_dsdirsize) {
4791 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
4793 } else if (nvpp != NULL && newnmp != NULL &&
4794 nfsaddr2_match(sad, newnmp->nm_nam))
4797 if (ippos != NULL && curnmp == NULL &&
4798 sad->sa_family == AF_INET &&
4799 sin->sin_addr.s_addr == 0)
4801 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
4803 /* Use the socket address to find the mount point. */
4806 /* Find a match for the IP address. */
4807 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
4808 if (ds->nfsdev_nmp != NULL) {
4809 dvp = ds->nfsdev_dvp;
4810 nmp = VFSTONFS(dvp->v_mount);
4811 if (nmp != ds->nfsdev_nmp)
4812 printf("different2 nmp %p %p\n",
4813 nmp, ds->nfsdev_nmp);
4814 if (nfsaddr2_match(sad, nmp->nm_nam)) {
4820 if (fndds != NULL && newnmpp != NULL &&
4822 /* Search for a place to make a mirror copy. */
4823 TAILQ_FOREACH(tds, &nfsrv_devidhead,
4825 if (tds->nfsdev_nmp != NULL &&
4827 ((tds->nfsdev_mdsisset == 0 &&
4828 fndds->nfsdev_mdsisset == 0) ||
4829 (tds->nfsdev_mdsisset != 0 &&
4830 fndds->nfsdev_mdsisset != 0 &&
4831 fsidcmp(&tds->nfsdev_mdsfsid,
4832 &mp->mnt_stat.f_fsid) == 0))) {
4833 *newnmpp = tds->nfsdev_nmp;
4839 * Move this entry to the end of the
4840 * list, so it won't be selected as
4841 * easily the next time.
4843 TAILQ_REMOVE(&nfsrv_devidhead, tds,
4845 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
4850 if (fndds != NULL) {
4851 dvp = fndds->nfsdev_dsdir[dsdir];
4852 if (lktype != 0 || fhiszero != 0 ||
4853 (nvpp != NULL && *nvpp == NULL)) {
4855 error = vn_lock(dvp,
4857 else if (lktype != 0)
4858 error = vn_lock(dvp, lktype);
4860 error = vn_lock(dvp, LK_SHARED);
4862 * If the file handle is all 0's, try to
4863 * do a Lookup against the DS to acquire
4865 * If dvpp == NULL or the Lookup fails,
4866 * unlock dvp after the call.
4868 if (error == 0 && (fhiszero != 0 ||
4869 (nvpp != NULL && *nvpp == NULL))) {
4870 error = nfsrv_pnfslookupds(vp,
4886 if (error != 0 || lktype == 0)
4892 NFSD_DEBUG(4, "gotone=%d\n", gotone);
4893 if (devid != NULL) {
4894 NFSBCOPY(fndds->nfsdev_deviceid,
4895 devid, NFSX_V4DEVICEID);
4896 devid += NFSX_V4DEVICEID;
4901 NFSBCOPY(&pf->dsf_fh, fhp++,
4903 if (fnamep != NULL && gotone == 1)
4906 sizeof(pf->dsf_filename));
4908 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
4913 if (error == 0 && gotone == 0)
4916 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
4919 *mirrorcntp = gotone;
4921 if (gotone > 0 && dvpp != NULL) {
4923 * If the error didn't occur on the first one and
4924 * dvpp != NULL, the one(s) prior to the failure will
4925 * have locked dvp's that need to be unlocked.
4927 for (i = 0; i < gotone; i++) {
4928 NFSVOPUNLOCK(*dvpp);
4933 * If it found the vnode to be copied from before a failure,
4934 * it needs to be vput()'d.
4936 if (nvpp != NULL && *nvpp != NULL) {
4945 * Set the extended attribute for the Change attribute.
4948 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4950 struct pnfsdsattr dsattr;
4953 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
4954 dsattr.dsa_filerev = nap->na_filerev;
4955 dsattr.dsa_size = nap->na_size;
4956 dsattr.dsa_atime = nap->na_atime;
4957 dsattr.dsa_mtime = nap->na_mtime;
4958 dsattr.dsa_bytes = nap->na_bytes;
4959 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4960 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
4962 printf("pNFS: setextattr=%d\n", error);
4967 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
4968 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
4971 struct nfsrv_descript *nd;
4973 struct mbuf *m, *m2;
4974 int error = 0, retlen, tlen, trimlen;
4976 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
4977 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4980 * Use a stateid where other is an alternating 01010 pattern and
4981 * seqid is 0xffffffff. This value is not defined as special by
4982 * the RFC and is used by the FreeBSD NFS server to indicate an
4983 * MDS->DS proxy operation.
4985 st.other[0] = 0x55555555;
4986 st.other[1] = 0x55555555;
4987 st.other[2] = 0x55555555;
4988 st.seqid = 0xffffffff;
4989 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
4991 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
4992 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
4993 txdr_hyper(off, tl);
4994 *(tl + 2) = txdr_unsigned(len);
4995 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
4996 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5001 if (nd->nd_repstat == 0) {
5002 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5003 NFSM_STRSIZ(retlen, len);
5005 /* Trim off the pre-data XDR from the mbuf chain. */
5007 while (m != NULL && m != nd->nd_md) {
5008 if (m->m_next == nd->nd_md) {
5010 m_freem(nd->nd_mrep);
5011 nd->nd_mrep = m = nd->nd_md;
5016 printf("nfsrv_readdsrpc: busted mbuf list\n");
5022 * Now, adjust first mbuf so that any XDR before the
5023 * read data is skipped over.
5025 trimlen = nd->nd_dpos - mtod(m, char *);
5027 m->m_len -= trimlen;
5028 NFSM_DATAP(m, trimlen);
5032 * Truncate the mbuf chain at retlen bytes of data,
5033 * plus XDR padding that brings the length up to a
5036 tlen = NFSM_RNDUP(retlen);
5038 if (m->m_len >= tlen) {
5048 } while (m != NULL);
5050 printf("nfsrv_readdsrpc: busted mbuf list\n");
5059 error = nd->nd_repstat;
5061 /* If nd->nd_mrep is already NULL, this is a no-op. */
5062 m_freem(nd->nd_mrep);
5064 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5069 * Do a write RPC on a DS data file, using this structure for the arguments,
5070 * so that this function can be executed by a separate kernel process.
5072 struct nfsrvwritedsdorpc {
5079 struct nfsmount *nmp;
5087 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5088 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5091 struct nfsrv_descript *nd;
5092 nfsattrbit_t attrbits;
5094 int commit, error, retlen;
5096 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5097 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5098 sizeof(fhandle_t), NULL, NULL, 0, 0);
5101 * Use a stateid where other is an alternating 01010 pattern and
5102 * seqid is 0xffffffff. This value is not defined as special by
5103 * the RFC and is used by the FreeBSD NFS server to indicate an
5104 * MDS->DS proxy operation.
5106 st.other[0] = 0x55555555;
5107 st.other[1] = 0x55555555;
5108 st.other[2] = 0x55555555;
5109 st.seqid = 0xffffffff;
5110 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5111 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5112 txdr_hyper(off, tl);
5115 * Do all writes FileSync, since the server doesn't hold onto dirty
5116 * buffers. Since clients should be accessing the DS servers directly
5117 * using the pNFS layouts, this just needs to work correctly as a
5120 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5121 *tl = txdr_unsigned(len);
5122 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5124 /* Put data in mbuf chain. */
5125 nd->nd_mb->m_next = m;
5127 /* Set nd_mb and nd_bpos to end of data. */
5128 while (m->m_next != NULL)
5131 nfsm_set(nd, m->m_len);
5132 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5134 /* Do a Getattr for the attributes that change upon writing. */
5135 NFSZERO_ATTRBIT(&attrbits);
5136 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5137 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5138 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5139 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5140 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5141 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5142 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5143 (void) nfsrv_putattrbit(nd, &attrbits);
5144 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5145 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5150 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5151 /* Get rid of weak cache consistency data for now. */
5152 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5153 (ND_NFSV4 | ND_V4WCCATTR)) {
5154 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5155 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5156 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5160 * Get rid of Op# and status for next op.
5162 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5164 nd->nd_flag |= ND_NOMOREDATA;
5166 if (nd->nd_repstat == 0) {
5167 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5168 retlen = fxdr_unsigned(int, *tl++);
5169 commit = fxdr_unsigned(int, *tl);
5170 if (commit != NFSWRITE_FILESYNC)
5172 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5173 retlen, commit, error);
5175 error = nd->nd_repstat;
5176 /* We have no use for the Write Verifier since we use FileSync. */
5179 * Get the Change, Size, Access Time and Modify Time attributes and set
5180 * on the Metadata file, so its attributes will be what the file's
5181 * would be if it had been written.
5184 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5185 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5186 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5188 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5190 m_freem(nd->nd_mrep);
5192 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5197 * Start up the thread that will execute nfsrv_writedsdorpc().
5200 start_writedsdorpc(void *arg, int pending)
5202 struct nfsrvwritedsdorpc *drpc;
5204 drpc = (struct nfsrvwritedsdorpc *)arg;
5205 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5206 drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5208 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5212 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5213 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5214 struct mbuf **mpp, char *cp, int *failposp)
5216 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5219 int error, i, offs, ret, timo;
5221 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5222 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5225 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5228 /* Calculate offset in mbuf chain that data starts. */
5229 offs = cp - mtod(*mpp, char *);
5230 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5233 * Do the write RPC for every DS, using a separate kernel process
5234 * for every DS except the last one.
5237 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5239 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5247 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5249 if (nfs_pnfsiothreads != 0) {
5250 ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5251 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5255 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5257 if (nfsds_failerr(ret) && *failposp == -1)
5259 else if (error == 0 && ret != 0)
5265 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5266 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5267 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5268 *failposp = mirrorcnt - 1;
5269 else if (error == 0 && ret != 0)
5272 error = nfsrv_setextattr(vp, &na, p);
5273 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5275 timo = hz / 50; /* Wait for 20msec. */
5278 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5279 /* Wait for RPCs on separate threads to complete. */
5280 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5281 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5282 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5284 else if (error == 0 && tdrpc->err != 0)
5292 * Do a allocate RPC on a DS data file, using this structure for the arguments,
5293 * so that this function can be executed by a separate kernel process.
5295 struct nfsrvallocatedsdorpc {
5302 struct nfsmount *nmp;
5309 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5310 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5313 struct nfsrv_descript *nd;
5314 nfsattrbit_t attrbits;
5318 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5319 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5320 sizeof(fhandle_t), NULL, NULL, 0, 0);
5323 * Use a stateid where other is an alternating 01010 pattern and
5324 * seqid is 0xffffffff. This value is not defined as special by
5325 * the RFC and is used by the FreeBSD NFS server to indicate an
5326 * MDS->DS proxy operation.
5328 st.other[0] = 0x55555555;
5329 st.other[1] = 0x55555555;
5330 st.other[2] = 0x55555555;
5331 st.seqid = 0xffffffff;
5332 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5333 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5334 txdr_hyper(off, tl); tl += 2;
5335 txdr_hyper(len, tl); tl += 2;
5336 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5338 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5339 NFSGETATTR_ATTRBIT(&attrbits);
5340 nfsrv_putattrbit(nd, &attrbits);
5341 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5342 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5347 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5349 if (nd->nd_repstat == 0) {
5350 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5351 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5352 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5354 error = nd->nd_repstat;
5355 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5357 m_freem(nd->nd_mrep);
5359 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5364 * Start up the thread that will execute nfsrv_allocatedsdorpc().
5367 start_allocatedsdorpc(void *arg, int pending)
5369 struct nfsrvallocatedsdorpc *drpc;
5371 drpc = (struct nfsrvallocatedsdorpc *)arg;
5372 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5373 drpc->len, NULL, drpc->cred, drpc->p);
5375 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5379 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5380 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5383 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5385 int error, i, ret, timo;
5387 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5390 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5394 * Do the allocate RPC for every DS, using a separate kernel process
5395 * for every DS except the last one.
5398 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5400 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5409 if (nfs_pnfsiothreads != 0) {
5410 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5411 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5415 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5417 if (nfsds_failerr(ret) && *failposp == -1)
5419 else if (error == 0 && ret != 0)
5425 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5426 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5427 *failposp = mirrorcnt - 1;
5428 else if (error == 0 && ret != 0)
5431 error = nfsrv_setextattr(vp, &na, p);
5432 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5434 timo = hz / 50; /* Wait for 20msec. */
5437 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5438 /* Wait for RPCs on separate threads to complete. */
5439 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5440 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5441 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5443 else if (error == 0 && tdrpc->err != 0)
5451 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5452 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5453 struct nfsvattr *dsnap)
5456 struct nfsrv_descript *nd;
5458 nfsattrbit_t attrbits;
5461 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5462 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5464 * Use a stateid where other is an alternating 01010 pattern and
5465 * seqid is 0xffffffff. This value is not defined as special by
5466 * the RFC and is used by the FreeBSD NFS server to indicate an
5467 * MDS->DS proxy operation.
5469 st.other[0] = 0x55555555;
5470 st.other[1] = 0x55555555;
5471 st.other[2] = 0x55555555;
5472 st.seqid = 0xffffffff;
5473 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5475 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5476 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5478 /* Do a Getattr for the attributes that change due to writing. */
5479 NFSZERO_ATTRBIT(&attrbits);
5480 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5481 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5482 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5483 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5484 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5485 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5486 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5487 (void) nfsrv_putattrbit(nd, &attrbits);
5488 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5489 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5494 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5496 /* Get rid of weak cache consistency data for now. */
5497 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5498 (ND_NFSV4 | ND_V4WCCATTR)) {
5499 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5500 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5501 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5505 * Get rid of Op# and status for next op.
5507 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5509 nd->nd_flag |= ND_NOMOREDATA;
5511 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5514 if (nd->nd_repstat != 0)
5515 error = nd->nd_repstat;
5517 * Get the Change, Size, Access Time and Modify Time attributes and set
5518 * on the Metadata file, so its attributes will be what the file's
5519 * would be if it had been written.
5522 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5523 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5524 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5526 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5528 m_freem(nd->nd_mrep);
5530 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5534 struct nfsrvsetattrdsdorpc {
5539 struct nfsmount *nmp;
5544 struct nfsvattr dsna;
5549 * Start up the thread that will execute nfsrv_setattrdsdorpc().
5552 start_setattrdsdorpc(void *arg, int pending)
5554 struct nfsrvsetattrdsdorpc *drpc;
5556 drpc = (struct nfsrvsetattrdsdorpc *)arg;
5557 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5558 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5563 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5564 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5565 struct nfsvattr *nap, int *failposp)
5567 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5569 int error, i, ret, timo;
5571 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5574 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5578 * Do the setattr RPC for every DS, using a separate kernel process
5579 * for every DS except the last one.
5582 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5585 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5593 if (nfs_pnfsiothreads != 0) {
5594 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5595 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5599 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5601 if (nfsds_failerr(ret) && *failposp == -1)
5603 else if (error == 0 && ret != 0)
5609 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
5610 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5611 *failposp = mirrorcnt - 1;
5612 else if (error == 0 && ret != 0)
5615 error = nfsrv_setextattr(vp, &na, p);
5616 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
5618 timo = hz / 50; /* Wait for 20msec. */
5621 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5622 /* Wait for RPCs on separate threads to complete. */
5623 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5624 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
5625 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5627 else if (error == 0 && tdrpc->err != 0)
5635 * Do a Setattr of an NFSv4 ACL on the DS file.
5638 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5639 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
5641 struct nfsrv_descript *nd;
5643 nfsattrbit_t attrbits;
5646 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
5647 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5649 * Use a stateid where other is an alternating 01010 pattern and
5650 * seqid is 0xffffffff. This value is not defined as special by
5651 * the RFC and is used by the FreeBSD NFS server to indicate an
5652 * MDS->DS proxy operation.
5654 st.other[0] = 0x55555555;
5655 st.other[1] = 0x55555555;
5656 st.other[2] = 0x55555555;
5657 st.seqid = 0xffffffff;
5658 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5660 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5661 NFSZERO_ATTRBIT(&attrbits);
5662 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
5664 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
5665 * so passing in the metadata "vp" will be ok, since it is of
5666 * the same type (VREG).
5668 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
5669 NULL, 0, 0, 0, 0, 0, NULL);
5670 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5671 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5676 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
5678 error = nd->nd_repstat;
5679 m_freem(nd->nd_mrep);
5684 struct nfsrvsetacldsdorpc {
5689 struct nfsmount *nmp;
5698 * Start up the thread that will execute nfsrv_setacldsdorpc().
5701 start_setacldsdorpc(void *arg, int pending)
5703 struct nfsrvsetacldsdorpc *drpc;
5705 drpc = (struct nfsrvsetacldsdorpc *)arg;
5706 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
5707 drpc->vp, drpc->nmp, drpc->aclp);
5712 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5713 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
5716 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
5717 int error, i, ret, timo;
5719 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
5722 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5726 * Do the setattr RPC for every DS, using a separate kernel process
5727 * for every DS except the last one.
5730 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5733 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5741 if (nfs_pnfsiothreads != 0) {
5742 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
5743 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
5747 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
5749 if (nfsds_failerr(ret) && *failposp == -1)
5751 else if (error == 0 && ret != 0)
5757 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
5758 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5759 *failposp = mirrorcnt - 1;
5760 else if (error == 0 && ret != 0)
5762 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
5764 timo = hz / 50; /* Wait for 20msec. */
5767 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5768 /* Wait for RPCs on separate threads to complete. */
5769 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5770 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
5771 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5773 else if (error == 0 && tdrpc->err != 0)
5781 * Getattr call to the DS for the attributes that change due to writing.
5784 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5785 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
5787 struct nfsrv_descript *nd;
5789 nfsattrbit_t attrbits;
5791 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
5792 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5793 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
5794 sizeof(fhandle_t), NULL, NULL, 0, 0);
5795 NFSZERO_ATTRBIT(&attrbits);
5796 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5797 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5798 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5799 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5800 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5801 (void) nfsrv_putattrbit(nd, &attrbits);
5802 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5803 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5808 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
5810 if (nd->nd_repstat == 0) {
5811 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
5812 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
5815 * We can only save the updated values in the extended
5816 * attribute if the vp is exclusively locked.
5817 * This should happen when any of the following operations
5818 * occur on the vnode:
5819 * Close, Delegreturn, LayoutCommit, LayoutReturn
5820 * As such, the updated extended attribute should get saved
5821 * before nfsrv_checkdsattr() returns 0 and allows the cached
5822 * attributes to be returned without calling this function.
5824 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
5825 error = nfsrv_setextattr(vp, nap, p);
5826 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
5830 error = nd->nd_repstat;
5831 m_freem(nd->nd_mrep);
5833 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
5838 * Seek call to a DS.
5841 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
5842 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
5845 struct nfsrv_descript *nd;
5849 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
5851 * Use a stateid where other is an alternating 01010 pattern and
5852 * seqid is 0xffffffff. This value is not defined as special by
5853 * the RFC and is used by the FreeBSD NFS server to indicate an
5854 * MDS->DS proxy operation.
5856 st.other[0] = 0x55555555;
5857 st.other[1] = 0x55555555;
5858 st.other[2] = 0x55555555;
5859 st.seqid = 0xffffffff;
5860 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5861 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
5862 sizeof(fhandle_t), NULL, NULL, 0, 0);
5863 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5864 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
5865 txdr_hyper(*offp, tl); tl += 2;
5866 *tl = txdr_unsigned(content);
5867 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5868 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5873 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
5874 if (nd->nd_repstat == 0) {
5875 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
5876 if (*tl++ == newnfs_true)
5880 *offp = fxdr_hyper(tl);
5882 error = nd->nd_repstat;
5884 m_freem(nd->nd_mrep);
5886 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
5891 * Get the device id and file handle for a DS file.
5894 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
5895 fhandle_t *fhp, char *devid)
5901 buf = malloc(buflen, M_TEMP, M_WAITOK);
5902 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
5903 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
5909 * Do a Lookup against the DS for the filename.
5912 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
5913 struct vnode **nvpp, NFSPROC_T *p)
5915 struct nameidata named;
5916 struct ucred *tcred;
5922 tcred = newnfs_getcred();
5923 named.ni_cnd.cn_nameiop = LOOKUP;
5924 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
5925 named.ni_cnd.cn_cred = tcred;
5926 named.ni_cnd.cn_thread = p;
5927 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
5928 nfsvno_setpathbuf(&named, &bufp, &hashp);
5929 named.ni_cnd.cn_nameptr = bufp;
5930 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
5931 strlcpy(bufp, pf->dsf_filename, NAME_MAX);
5932 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
5933 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
5934 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
5936 nfsvno_relpathbuf(&named);
5939 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
5944 * Set the file handle to the correct one.
5947 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
5948 char *fnamep, struct vnode *nvp, NFSPROC_T *p)
5954 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
5956 * We can only do a vn_set_extattr() if the vnode is exclusively
5957 * locked and vn_start_write() has been done. If devid != NULL or
5958 * fnamep != NULL or the vnode is shared locked, vn_start_write()
5959 * may not have been done.
5960 * If not done now, it will be done on a future call.
5962 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
5964 ret = vn_extattr_set(vp, IO_NODELOCKED,
5965 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
5967 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
5971 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point
5972 * when the DS has failed.
5975 nfsrv_killrpcs(struct nfsmount *nmp)
5979 * Call newnfs_nmcancelreqs() to cause
5980 * any RPCs in progress on the mount point to
5982 * This will cause any process waiting for an
5983 * RPC to complete while holding a vnode lock
5984 * on the mounted-on vnode (such as "df" or
5985 * a non-forced "umount") to fail.
5986 * This will unlock the mounted-on vnode so
5987 * a forced dismount can succeed.
5988 * The NFSMNTP_CANCELRPCS flag should be set when this function is
5991 newnfs_nmcancelreqs(nmp);
5995 * Sum up the statfs info for each of the DSs, so that the client will
5996 * receive the total for all DSs.
5999 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6002 struct nfsdevice *ds;
6003 struct vnode **dvpp, **tdvpp, *dvp;
6005 int cnt, error = 0, i;
6007 if (nfsrv_devidcnt <= 0)
6009 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6010 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6012 /* Get an array of the dvps for the DSs. */
6016 /* First, search for matches for same file system. */
6017 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6018 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6019 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6020 if (++i > nfsrv_devidcnt)
6022 *tdvpp++ = ds->nfsdev_dvp;
6026 * If no matches for same file system, total all servers not assigned
6030 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6031 if (ds->nfsdev_nmp != NULL &&
6032 ds->nfsdev_mdsisset == 0) {
6033 if (++i > nfsrv_devidcnt)
6035 *tdvpp++ = ds->nfsdev_dvp;
6042 /* Do a VFS_STATFS() for each of the DSs and sum them up. */
6044 for (i = 0; i < cnt && error == 0; i++) {
6046 error = VFS_STATFS(dvp->v_mount, tsf);
6048 if (sf->f_bsize == 0) {
6049 if (tsf->f_bsize > 0)
6050 sf->f_bsize = tsf->f_bsize;
6054 if (tsf->f_blocks > 0) {
6055 if (sf->f_bsize != tsf->f_bsize) {
6056 tot = tsf->f_blocks * tsf->f_bsize;
6057 sf->f_blocks += (tot / sf->f_bsize);
6059 sf->f_blocks += tsf->f_blocks;
6061 if (tsf->f_bfree > 0) {
6062 if (sf->f_bsize != tsf->f_bsize) {
6063 tot = tsf->f_bfree * tsf->f_bsize;
6064 sf->f_bfree += (tot / sf->f_bsize);
6066 sf->f_bfree += tsf->f_bfree;
6068 if (tsf->f_bavail > 0) {
6069 if (sf->f_bsize != tsf->f_bsize) {
6070 tot = tsf->f_bavail * tsf->f_bsize;
6071 sf->f_bavail += (tot / sf->f_bsize);
6073 sf->f_bavail += tsf->f_bavail;
6086 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6090 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6091 error = NFSERR_ATTRNOTSUPP;
6095 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6096 * Make sure it has enough room for that - splitting every entry
6097 * into two and appending "canonical six" entries at the end.
6098 * Cribbed out of kern/vfs_acl.c - Rick M.
6100 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6101 error = NFSERR_ATTRNOTSUPP;
6104 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6106 error = nfsrv_dssetacl(vp, aclp, cred, p);
6107 if (error == ENOENT)
6117 * Seek vnode op call (actually it is a VOP_IOCTL()).
6118 * This function is called with the vnode locked, but unlocks and vrele()s
6119 * the vp before returning.
6122 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6123 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6128 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6130 * Attempt to seek on a DS file. A return of ENOENT implies
6131 * there is no DS file to seek on.
6133 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6134 NULL, NULL, NULL, NULL, offp, content, eofp);
6135 if (error != ENOENT) {
6141 * Do the VOP_IOCTL() call. For the case where *offp == file_size,
6142 * VOP_IOCTL() will return ENXIO. However, the correct reply for
6143 * NFSv4.2 is *eofp == true and error == 0 for this case.
6146 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6148 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6149 /* Handle the cases where we might be at EOF. */
6150 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6151 if (ret == 0 && *offp == at.na_size) {
6155 if (ret != 0 && error == 0)
6164 * Allocate vnode op call.
6167 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6172 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6174 * Attempt to allocate on a DS file. A return of ENOENT implies
6175 * there is no DS file to allocate on.
6177 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6178 NULL, NULL, NULL, NULL, &len, 0, NULL);
6179 if (error != ENOENT)
6184 * Do the actual VOP_ALLOCATE(), looping a reasonable number of
6185 * times to achieve completion.
6188 while (error == 0 && len > 0 && trycnt++ < 20)
6189 error = VOP_ALLOCATE(vp, &off, &len);
6190 if (error == 0 && len > 0)
6197 * Get Extended Atribute vnode op into an mbuf list.
6200 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6201 struct ucred *cred, struct thread *p, struct mbuf **mpp,
6202 struct mbuf **mpendp, int *lenp)
6205 struct uio io, *uiop = &io;
6206 struct mbuf *m, *m2;
6207 int alen, error, len, tlen;
6210 /* First, find out the size of the extended attribute. */
6211 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6214 return (NFSERR_NOXATTR);
6215 if (siz > maxresp - NFS_MAXXDR)
6216 return (NFSERR_XATTR2BIG);
6218 tlen = NFSM_RNDUP(len);
6220 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, &iv);
6223 uiop->uio_iovcnt = 0;
6224 uiop->uio_iov = iv = NULL;
6227 uiop->uio_offset = 0;
6228 uiop->uio_resid = tlen;
6229 uiop->uio_rw = UIO_READ;
6230 uiop->uio_segflg = UIO_SYSSPACE;
6233 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6240 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6244 if (uiop->uio_resid > 0) {
6246 len = tlen - uiop->uio_resid;
6247 tlen = NFSM_RNDUP(len);
6249 printf("nfsvno_getxattr: weird size read\n");
6250 nfsrv_adj(m, alen - tlen, tlen - len);
6268 * Set Extended attribute vnode op from an mbuf list.
6271 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6272 char *cp, struct ucred *cred, struct thread *p)
6275 struct uio uio, *uiop = &uio;
6280 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6286 uiop->uio_rw = UIO_WRITE;
6287 uiop->uio_segflg = UIO_SYSSPACE;
6289 uiop->uio_offset = 0;
6290 uiop->uio_resid = len;
6292 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6294 uiop->uio_iovcnt = cnt;
6296 uiop->uio_iov = iv = NULL;
6297 uiop->uio_iovcnt = 0;
6300 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6311 * Remove Extended attribute vnode op.
6314 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6315 struct ucred *cred, struct thread *p)
6320 * Get rid of any delegations. I am not sure why this is required,
6321 * but RFC-8276 says so.
6323 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6327 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6333 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6334 if (error == EOPNOTSUPP)
6335 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6343 * List Extended Atribute vnode op into an mbuf list.
6346 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6347 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6355 /* First, find out the size of the extended attribute. */
6356 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6359 return (NFSERR_NOXATTR);
6360 if (siz <= cookie) {
6365 if (siz > cookie + *lenp) {
6366 siz = cookie + *lenp;
6370 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6371 if (siz > 10 * 1024 * 1024) {
6372 error = NFSERR_XATTR2BIG;
6375 *bufp = malloc(siz, M_TEMP, M_WAITOK);
6376 iv.iov_base = *bufp;
6382 io.uio_rw = UIO_READ;
6383 io.uio_segflg = UIO_SYSSPACE;
6386 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6391 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6395 if (io.uio_resid > 0)
6396 siz -= io.uio_resid;
6401 free(*bufp, M_TEMP);
6408 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
6411 * Called once to initialize data structures...
6414 nfsd_modevent(module_t mod, int type, void *data)
6417 static int loaded = 0;
6424 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6425 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
6427 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
6430 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
6431 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
6432 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
6433 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
6434 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
6435 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
6442 #ifdef VV_DISABLEDELEG
6443 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
6444 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
6446 nfsd_call_servertimer = nfsrv_servertimer;
6447 nfsd_call_nfsd = nfssvc_nfsd;
6452 if (newnfs_numnfsd != 0) {
6457 #ifdef VV_DISABLEDELEG
6458 vn_deleg_ops.vndeleg_recall = NULL;
6459 vn_deleg_ops.vndeleg_disable = NULL;
6461 nfsd_call_servertimer = NULL;
6462 nfsd_call_nfsd = NULL;
6464 /* Clean out all NFSv4 state. */
6465 nfsrv_throwawayallstate(curthread);
6467 /* Clean the NFS server reply cache */
6468 nfsrvd_cleancache();
6470 /* Free up the krpc server pool. */
6471 if (nfsrvd_pool != NULL)
6472 svcpool_destroy(nfsrvd_pool);
6474 /* and get rid of the locks */
6475 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6476 mtx_destroy(&nfsrchash_table[i].mtx);
6477 mtx_destroy(&nfsrcahash_table[i].mtx);
6479 mtx_destroy(&nfsrc_udpmtx);
6480 mtx_destroy(&nfs_v4root_mutex);
6481 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
6482 mtx_destroy(&nfsrv_dontlistlock_mtx);
6483 mtx_destroy(&nfsrv_recalllock_mtx);
6484 for (i = 0; i < nfsrv_sessionhashsize; i++)
6485 mtx_destroy(&nfssessionhash[i].mtx);
6486 if (nfslayouthash != NULL) {
6487 for (i = 0; i < nfsrv_layouthashsize; i++)
6488 mtx_destroy(&nfslayouthash[i].mtx);
6489 free(nfslayouthash, M_NFSDSESSION);
6491 lockdestroy(&nfsv4root_mnt.mnt_explock);
6492 free(nfsclienthash, M_NFSDCLIENT);
6493 free(nfslockhash, M_NFSDLOCKFILE);
6494 free(nfssessionhash, M_NFSDSESSION);
6506 static moduledata_t nfsd_mod = {
6511 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
6513 /* So that loader and kldload(2) can find us, wherever we are.. */
6514 MODULE_VERSION(nfsd, 1);
6515 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
6516 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
6517 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
6518 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);