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>
50 #include <sys/sysctl.h>
51 #include <nlm/nlm_prot.h>
54 FEATURE(nfsd, "NFSv4 server");
56 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
57 extern int nfsrv_useacl;
58 extern int newnfs_numnfsd;
59 extern struct mount nfsv4root_mnt;
60 extern struct nfsrv_stablefirst nfsrv_stablefirst;
61 extern void (*nfsd_call_servertimer)(void);
62 extern SVCPOOL *nfsrvd_pool;
63 extern struct nfsv4lock nfsd_suspend_lock;
64 extern struct nfsclienthashhead *nfsclienthash;
65 extern struct nfslockhashhead *nfslockhash;
66 extern struct nfssessionhash *nfssessionhash;
67 extern int nfsrv_sessionhashsize;
68 extern struct nfsstatsv1 nfsstatsv1;
69 extern struct nfslayouthash *nfslayouthash;
70 extern int nfsrv_layouthashsize;
71 extern struct mtx nfsrv_dslock_mtx;
72 extern int nfs_pnfsiothreads;
73 extern struct nfsdontlisthead nfsrv_dontlisthead;
74 extern volatile int nfsrv_dontlistlen;
75 extern volatile int nfsrv_devidcnt;
76 extern int nfsrv_maxpnfsmirror;
77 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
80 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
81 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
82 struct mtx nfsrc_udpmtx;
83 struct mtx nfs_v4root_mutex;
84 struct mtx nfsrv_dontlistlock_mtx;
85 struct mtx nfsrv_recalllock_mtx;
86 struct nfsrvfh nfs_rootfh, nfs_pubfh;
87 int nfs_pubfhset = 0, nfs_rootfhset = 0;
88 struct proc *nfsd_master_proc = NULL;
89 int nfsd_debuglevel = 0;
90 static pid_t nfsd_master_pid = (pid_t)-1;
91 static char nfsd_master_comm[MAXCOMLEN + 1];
92 static struct timeval nfsd_master_start;
93 static uint32_t nfsv4_sysid = 0;
94 static fhandle_t zerofh;
96 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
99 int nfsrv_enable_crossmntpt = 1;
100 static int nfs_commit_blks;
101 static int nfs_commit_miss;
102 extern int nfsrv_issuedelegs;
103 extern int nfsrv_dolocallocks;
104 extern int nfsd_enable_stringtouid;
105 extern struct nfsdevicehead nfsrv_devidhead;
107 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
109 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
110 int *, char *, fhandle_t *);
111 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
113 static int nfsrv_proxyds(struct nfsrv_descript *, struct vnode *, off_t, int,
114 struct ucred *, struct thread *, int, struct mbuf **, char *,
115 struct mbuf **, struct nfsvattr *, struct acl *);
116 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
117 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
118 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
119 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
120 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
122 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
123 struct vnode *, struct nfsmount **, int, struct acl *, int *);
124 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
125 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
126 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
127 struct vnode *, struct nfsmount *, struct nfsvattr *);
128 static int nfsrv_putfhname(fhandle_t *, char *);
129 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
130 struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
131 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
132 struct vnode *, NFSPROC_T *);
133 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
134 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
136 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
138 int nfs_pnfsio(task_fn_t *, void *);
140 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server");
141 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
142 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
143 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
145 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
147 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
148 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
149 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
150 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
151 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
152 0, "Debug level for NFS server");
153 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
154 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
155 static int nfsrv_pnfsgetdsattr = 1;
156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
157 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
160 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
162 * The dsN subdirectories for the increased values must have been created
163 * on all DS servers before this increase is done.
165 u_int nfsrv_dsdirsize = 20;
167 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
169 int error, newdsdirsize;
171 newdsdirsize = nfsrv_dsdirsize;
172 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
173 if (error != 0 || req->newptr == NULL)
175 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
178 nfsrv_dsdirsize = newdsdirsize;
181 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, CTLTYPE_UINT | CTLFLAG_RW, 0,
182 sizeof(nfsrv_dsdirsize), sysctl_dsdirsize, "IU",
183 "Number of dsN subdirs on the DS servers");
185 #define MAX_REORDERED_RPC 16
186 #define NUM_HEURISTIC 1031
187 #define NHUSE_INIT 64
189 #define NHUSE_MAX 2048
191 static struct nfsheur {
192 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
193 off_t nh_nextoff; /* next offset for sequential detection */
194 int nh_use; /* use count for selection */
195 int nh_seqcount; /* heuristic */
196 } nfsheur[NUM_HEURISTIC];
200 * Heuristic to detect sequential operation.
202 static struct nfsheur *
203 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
208 /* Locate best candidate. */
210 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
213 if (nfsheur[hi].nh_vp == vp) {
217 if (nfsheur[hi].nh_use > 0)
218 --nfsheur[hi].nh_use;
219 hi = (hi + 1) % NUM_HEURISTIC;
220 if (nfsheur[hi].nh_use < nh->nh_use)
224 /* Initialize hint if this is a new file. */
225 if (nh->nh_vp != vp) {
227 nh->nh_nextoff = uio->uio_offset;
228 nh->nh_use = NHUSE_INIT;
229 if (uio->uio_offset == 0)
235 /* Calculate heuristic. */
236 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
237 uio->uio_offset == nh->nh_nextoff) {
238 /* See comments in vfs_vnops.c:sequential_heuristic(). */
239 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
240 if (nh->nh_seqcount > IO_SEQMAX)
241 nh->nh_seqcount = IO_SEQMAX;
242 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
243 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
244 /* Probably a reordered RPC, leave seqcount alone. */
245 } else if (nh->nh_seqcount > 1) {
246 nh->nh_seqcount /= 2;
250 nh->nh_use += NHUSE_INC;
251 if (nh->nh_use > NHUSE_MAX)
252 nh->nh_use = NHUSE_MAX;
257 * Get attributes into nfsvattr structure.
260 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
261 struct nfsrv_descript *nd, struct thread *p, int vpislocked,
262 nfsattrbit_t *attrbitp)
264 int error, gotattr, lockedit = 0;
267 if (vpislocked == 0) {
269 * When vpislocked == 0, the vnode is either exclusively
270 * locked by this thread or not locked by this thread.
271 * As such, shared lock it, if not exclusively locked.
273 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
275 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
280 * Acquire the Change, Size and TimeModify attributes, as required.
281 * This needs to be done for regular files if:
282 * - non-NFSv4 RPCs or
283 * - when attrbitp == NULL or
284 * - an NFSv4 RPC with any of the above attributes in attrbitp.
285 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
286 * these attributes. nfsrv_proxyds() will return an error if the
287 * server is not a pNFS one.
290 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
291 (nd->nd_flag & ND_NFSV4) == 0 ||
292 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
293 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
294 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
295 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY))) {
296 error = nfsrv_proxyds(nd, vp, 0, 0, nd->nd_cred, p,
297 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL);
302 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
307 * If we got the Change, Size and Modify Time from the DS,
311 nvap->na_atime = na.na_atime;
312 nvap->na_mtime = na.na_mtime;
313 nvap->na_filerev = na.na_filerev;
314 nvap->na_size = na.na_size;
316 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
317 error, (uintmax_t)na.na_filerev);
324 * Get a file handle for a vnode.
327 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
331 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
332 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
333 error = VOP_VPTOFH(vp, &fhp->fh_fid);
340 * Perform access checking for vnodes obtained from file handles that would
341 * refer to files already opened by a Unix client. You cannot just use
342 * vn_writechk() and VOP_ACCESSX() for two reasons.
343 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
345 * 2 - The owner is to be given access irrespective of mode bits for some
346 * operations, so that processes that chmod after opening a file don't
350 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
351 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
352 u_int32_t *supportedtypep)
355 int error = 0, getret = 0;
357 if (vpislocked == 0) {
358 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
363 if (accmode & VWRITE) {
364 /* Just vn_writechk() changed to check rdonly */
366 * Disallow write attempts on read-only file systems;
367 * unless the file is a socket or a block or character
368 * device resident on the file system.
370 if (NFSVNO_EXRDONLY(exp) ||
371 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
372 switch (vp->v_type) {
382 * If there's shared text associated with
383 * the inode, try to free it up once. If
384 * we fail, we can't allow writing.
386 if (VOP_IS_TEXT(vp) && error == 0)
396 * Should the override still be applied when ACLs are enabled?
398 error = VOP_ACCESSX(vp, accmode, cred, p);
399 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
401 * Try again with VEXPLICIT_DENY, to see if the test for
402 * deletion is supported.
404 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
406 if (vp->v_type == VDIR) {
407 accmode &= ~(VDELETE | VDELETE_CHILD);
409 error = VOP_ACCESSX(vp, accmode, cred, p);
410 } else if (supportedtypep != NULL) {
411 *supportedtypep &= ~NFSACCESS_DELETE;
417 * Allow certain operations for the owner (reads and writes
418 * on files that are already open).
420 if (override != NFSACCCHK_NOOVERRIDE &&
421 (error == EPERM || error == EACCES)) {
422 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
424 else if (override & NFSACCCHK_ALLOWOWNER) {
425 getret = VOP_GETATTR(vp, &vattr, cred);
426 if (getret == 0 && cred->cr_uid == vattr.va_uid)
439 * Set attribute(s) vnop.
442 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
443 struct thread *p, struct nfsexstuff *exp)
445 u_quad_t savsize = 0;
449 * If this is an exported file system and a pNFS service is running,
450 * don't VOP_SETATTR() of size for the MDS file system.
454 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
455 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
456 nvap->na_vattr.va_size > 0) {
457 savsize = nvap->na_vattr.va_size;
458 nvap->na_vattr.va_size = VNOVAL;
459 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
460 nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
461 nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
462 nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
463 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
469 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
471 nvap->na_vattr.va_size = savsize;
472 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
473 nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
474 nvap->na_vattr.va_size != VNOVAL ||
475 nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
476 nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
477 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
478 /* For a pNFS server, set the attributes on the DS file. */
479 error = nfsrv_proxyds(NULL, vp, 0, 0, cred, p, NFSPROC_SETATTR,
480 NULL, NULL, NULL, nvap, NULL);
489 * Set up nameidata for a lookup() call and do it.
492 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
493 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
494 struct vnode **retdirp)
496 struct componentname *cnp = &ndp->ni_cnd;
500 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
505 cnp->cn_nameptr = cnp->cn_pnbuf;
508 * Extract and set starting directory.
510 if (dp->v_type != VDIR) {
515 nfsvno_relpathbuf(ndp);
523 if (NFSVNO_EXRDONLY(exp))
524 cnp->cn_flags |= RDONLY;
525 ndp->ni_segflg = UIO_SYSSPACE;
527 if (nd->nd_flag & ND_PUBLOOKUP) {
529 if (cnp->cn_pnbuf[0] == '/') {
532 * Check for degenerate pathnames here, since lookup()
535 for (i = 1; i < ndp->ni_pathlen; i++)
536 if (cnp->cn_pnbuf[i] != '/')
538 if (i == ndp->ni_pathlen) {
539 error = NFSERR_ACCES;
545 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
546 (nd->nd_flag & ND_NFSV4) == 0) {
548 * Only cross mount points for NFSv4 when doing a
549 * mount while traversing the file system above
550 * the mount point, unless nfsrv_enable_crossmntpt is set.
552 cnp->cn_flags |= NOCROSSMOUNT;
556 * Initialize for scan, set ni_startdir and bump ref on dp again
557 * because lookup() will dereference ni_startdir.
561 ndp->ni_startdir = dp;
562 ndp->ni_rootdir = rootvnode;
563 ndp->ni_topdir = NULL;
566 cnp->cn_flags |= LOCKLEAF;
568 cnp->cn_nameptr = cnp->cn_pnbuf;
570 * Call lookup() to do the real work. If an error occurs,
571 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
572 * we do not have to dereference anything before returning.
573 * In either case ni_startdir will be dereferenced and NULLed
581 * Check for encountering a symbolic link. Trivial
582 * termination occurs if no symlink encountered.
584 if ((cnp->cn_flags & ISSYMLINK) == 0) {
585 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
586 nfsvno_relpathbuf(ndp);
587 if (ndp->ni_vp && !lockleaf)
588 NFSVOPUNLOCK(ndp->ni_vp, 0);
595 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
596 NFSVOPUNLOCK(ndp->ni_dvp, 0);
597 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
602 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
606 if (ndp->ni_pathlen > 1)
607 cp = uma_zalloc(namei_zone, M_WAITOK);
611 aiov.iov_len = MAXPATHLEN;
612 auio.uio_iov = &aiov;
615 auio.uio_rw = UIO_READ;
616 auio.uio_segflg = UIO_SYSSPACE;
618 auio.uio_resid = MAXPATHLEN;
619 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
622 if (ndp->ni_pathlen > 1)
623 uma_zfree(namei_zone, cp);
629 linklen = MAXPATHLEN - auio.uio_resid;
634 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
635 error = ENAMETOOLONG;
640 * Adjust or replace path
642 if (ndp->ni_pathlen > 1) {
643 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
644 uma_zfree(namei_zone, cnp->cn_pnbuf);
647 cnp->cn_pnbuf[linklen] = '\0';
648 ndp->ni_pathlen += linklen;
651 * Cleanup refs for next loop and check if root directory
652 * should replace current directory. Normally ni_dvp
653 * becomes the new base directory and is cleaned up when
654 * we loop. Explicitly null pointers after invalidation
655 * to clarify operation.
660 if (cnp->cn_pnbuf[0] == '/') {
662 ndp->ni_dvp = ndp->ni_rootdir;
665 ndp->ni_startdir = ndp->ni_dvp;
669 cnp->cn_flags &= ~LOCKLEAF;
673 nfsvno_relpathbuf(ndp);
676 ndp->ni_startdir = NULL;
677 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
682 NFSEXITCODE2(error, nd);
687 * Set up a pathname buffer and return a pointer to it and, optionally
688 * set a hash pointer.
691 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
693 struct componentname *cnp = &ndp->ni_cnd;
695 cnp->cn_flags |= (NOMACCHECK | HASBUF);
696 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
699 *bufpp = cnp->cn_pnbuf;
703 * Release the above path buffer, if not released by nfsvno_namei().
706 nfsvno_relpathbuf(struct nameidata *ndp)
709 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
711 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
712 ndp->ni_cnd.cn_flags &= ~HASBUF;
716 * Readlink vnode op into an mbuf list.
719 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
720 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
722 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
723 struct iovec *ivp = iv;
724 struct uio io, *uiop = &io;
725 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
726 int i, len, tlen, error = 0;
730 while (len < NFS_MAXPATHLEN) {
732 MCLGET(mp, M_WAITOK);
733 mp->m_len = M_SIZE(mp);
740 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
741 mp->m_len = NFS_MAXPATHLEN - len;
742 len = NFS_MAXPATHLEN;
746 ivp->iov_base = mtod(mp, caddr_t);
747 ivp->iov_len = mp->m_len;
752 uiop->uio_iovcnt = i;
753 uiop->uio_offset = 0;
754 uiop->uio_resid = len;
755 uiop->uio_rw = UIO_READ;
756 uiop->uio_segflg = UIO_SYSSPACE;
758 error = VOP_READLINK(vp, uiop, cred);
764 if (uiop->uio_resid > 0) {
765 len -= uiop->uio_resid;
766 tlen = NFSM_RNDUP(len);
767 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
779 * Read vnode op call into mbuf list.
782 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
783 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
789 int error = 0, len, left, siz, tlen, ioflag = 0;
790 struct mbuf *m2 = NULL, *m3;
791 struct uio io, *uiop = &io;
795 * Attempt to read from a DS file. A return of ENOENT implies
796 * there is no DS file to read.
798 error = nfsrv_proxyds(NULL, vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
799 NULL, mpendp, NULL, NULL);
803 len = left = NFSM_RNDUP(cnt);
806 * Generate the mbuf list with the uio_iov ref. to it.
813 siz = min(M_TRAILINGSPACE(m), left);
822 iv = malloc(i * sizeof (struct iovec),
824 uiop->uio_iov = iv2 = iv;
830 panic("nfsvno_read iov");
831 siz = min(M_TRAILINGSPACE(m), left);
833 iv->iov_base = mtod(m, caddr_t) + m->m_len;
842 uiop->uio_iovcnt = i;
843 uiop->uio_offset = off;
844 uiop->uio_resid = len;
845 uiop->uio_rw = UIO_READ;
846 uiop->uio_segflg = UIO_SYSSPACE;
848 nh = nfsrv_sequential_heuristic(uiop, vp);
849 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
850 /* XXX KDM make this more systematic? */
851 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
852 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
859 nh->nh_nextoff = uiop->uio_offset;
860 tlen = len - uiop->uio_resid;
861 cnt = cnt < tlen ? cnt : tlen;
862 tlen = NFSM_RNDUP(cnt);
866 } else if (len != tlen || tlen != cnt)
867 nfsrv_adj(m3, len - tlen, tlen - cnt);
877 * Write vnode op from an mbuf list.
880 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int *stable,
881 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
887 struct uio io, *uiop = &io;
891 * Attempt to write to a DS file. A return of ENOENT implies
892 * there is no DS file to write.
894 error = nfsrv_proxyds(NULL, vp, off, retlen, cred, p, NFSPROC_WRITEDS,
895 &mp, cp, NULL, NULL, NULL);
896 if (error != ENOENT) {
897 *stable = NFSWRITE_FILESYNC;
901 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
903 uiop->uio_iov = iv = ivp;
904 uiop->uio_iovcnt = cnt;
905 i = mtod(mp, caddr_t) + mp->m_len - cp;
909 panic("nfsvno_write");
920 cp = mtod(mp, caddr_t);
924 if (*stable == NFSWRITE_UNSTABLE)
925 ioflags = IO_NODELOCKED;
927 ioflags = (IO_SYNC | IO_NODELOCKED);
928 uiop->uio_resid = retlen;
929 uiop->uio_rw = UIO_WRITE;
930 uiop->uio_segflg = UIO_SYSSPACE;
932 uiop->uio_offset = off;
933 nh = nfsrv_sequential_heuristic(uiop, vp);
934 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
935 /* XXX KDM make this more systematic? */
936 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
937 error = VOP_WRITE(vp, uiop, ioflags, cred);
939 nh->nh_nextoff = uiop->uio_offset;
947 * Common code for creating a regular file (plus special files for V2).
950 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
951 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
952 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
956 struct thread *p = curthread;
958 error = nd->nd_repstat;
959 if (!error && ndp->ni_vp == NULL) {
960 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
961 vrele(ndp->ni_startdir);
962 error = VOP_CREATE(ndp->ni_dvp,
963 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
964 /* For a pNFS server, create the data file on a DS. */
965 if (error == 0 && nvap->na_type == VREG) {
967 * Create a data file on a DS for a pNFS server.
968 * This function just returns if not
969 * running a pNFS DS or the creation fails.
971 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
975 nfsvno_relpathbuf(ndp);
977 if (*exclusive_flagp) {
978 *exclusive_flagp = 0;
979 NFSVNO_ATTRINIT(nvap);
980 nvap->na_atime.tv_sec = cverf[0];
981 nvap->na_atime.tv_nsec = cverf[1];
982 error = VOP_SETATTR(ndp->ni_vp,
983 &nvap->na_vattr, nd->nd_cred);
987 error = NFSERR_NOTSUPP;
992 * NFS V2 Only. nfsrvd_mknod() does this for V3.
993 * (This implies, just get out on an error.)
995 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
996 nvap->na_type == VFIFO) {
997 if (nvap->na_type == VCHR && rdev == 0xffffffff)
998 nvap->na_type = VFIFO;
999 if (nvap->na_type != VFIFO &&
1000 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1001 vrele(ndp->ni_startdir);
1002 nfsvno_relpathbuf(ndp);
1006 nvap->na_rdev = rdev;
1007 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1008 &ndp->ni_cnd, &nvap->na_vattr);
1010 nfsvno_relpathbuf(ndp);
1011 vrele(ndp->ni_startdir);
1015 vrele(ndp->ni_startdir);
1016 nfsvno_relpathbuf(ndp);
1024 * Handle cases where error is already set and/or
1026 * 1 - clean up the lookup
1027 * 2 - iff !error and na_size set, truncate it
1029 vrele(ndp->ni_startdir);
1030 nfsvno_relpathbuf(ndp);
1032 if (ndp->ni_dvp == *vpp)
1036 if (!error && nvap->na_size != VNOVAL) {
1037 error = nfsvno_accchk(*vpp, VWRITE,
1038 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1039 NFSACCCHK_VPISLOCKED, NULL);
1041 tempsize = nvap->na_size;
1042 NFSVNO_ATTRINIT(nvap);
1043 nvap->na_size = tempsize;
1044 error = VOP_SETATTR(*vpp,
1045 &nvap->na_vattr, nd->nd_cred);
1058 * Do a mknod vnode op.
1061 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1067 vtyp = nvap->na_type;
1069 * Iff doesn't exist, create it.
1072 vrele(ndp->ni_startdir);
1073 nfsvno_relpathbuf(ndp);
1079 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1080 vrele(ndp->ni_startdir);
1081 nfsvno_relpathbuf(ndp);
1083 error = NFSERR_BADTYPE;
1086 if (vtyp == VSOCK) {
1087 vrele(ndp->ni_startdir);
1088 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1089 &ndp->ni_cnd, &nvap->na_vattr);
1091 nfsvno_relpathbuf(ndp);
1093 if (nvap->na_type != VFIFO &&
1094 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1095 vrele(ndp->ni_startdir);
1096 nfsvno_relpathbuf(ndp);
1100 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1101 &ndp->ni_cnd, &nvap->na_vattr);
1103 nfsvno_relpathbuf(ndp);
1104 vrele(ndp->ni_startdir);
1106 * Since VOP_MKNOD returns the ni_vp, I can't
1107 * see any reason to do the lookup.
1120 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1121 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1125 if (ndp->ni_vp != NULL) {
1126 if (ndp->ni_dvp == ndp->ni_vp)
1131 nfsvno_relpathbuf(ndp);
1135 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1138 nfsvno_relpathbuf(ndp);
1149 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1150 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1151 struct nfsexstuff *exp)
1156 vrele(ndp->ni_startdir);
1157 nfsvno_relpathbuf(ndp);
1158 if (ndp->ni_dvp == ndp->ni_vp)
1167 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1168 &nvap->na_vattr, pathcp);
1170 vrele(ndp->ni_startdir);
1171 nfsvno_relpathbuf(ndp);
1173 * Although FreeBSD still had the lookup code in
1174 * it for 7/current, there doesn't seem to be any
1175 * point, since VOP_SYMLINK() returns the ni_vp.
1176 * Just vput it for v2.
1178 if (!not_v2 && !error)
1187 * Parse symbolic link arguments.
1188 * This function has an ugly side effect. It will malloc() an area for
1189 * the symlink and set iov_base to point to it, only if it succeeds.
1190 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1194 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1195 struct thread *p, char **pathcpp, int *lenp)
1198 char *pathcp = NULL;
1200 struct nfsv2_sattr *sp;
1204 if ((nd->nd_flag & ND_NFSV3) &&
1205 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1207 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1208 len = fxdr_unsigned(int, *tl);
1209 if (len > NFS_MAXPATHLEN || len <= 0) {
1213 pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1214 error = nfsrv_mtostr(nd, pathcp, len);
1217 if (nd->nd_flag & ND_NFSV2) {
1218 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1219 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1223 NFSEXITCODE2(0, nd);
1227 free(pathcp, M_TEMP);
1228 NFSEXITCODE2(error, nd);
1233 * Remove a non-directory object.
1236 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1237 struct thread *p, struct nfsexstuff *exp)
1239 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1240 int error = 0, mirrorcnt;
1241 char fname[PNFS_FILENAME_LEN + 1];
1246 if (vp->v_type == VDIR)
1247 error = NFSERR_ISDIR;
1249 error = nfsrv_checkremove(vp, 1, p);
1251 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1253 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1254 if (error == 0 && dsdvp[0] != NULL)
1255 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1256 if (ndp->ni_dvp == vp)
1261 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1262 nfsvno_relpathbuf(ndp);
1268 * Remove a directory.
1271 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1272 struct thread *p, struct nfsexstuff *exp)
1278 if (vp->v_type != VDIR) {
1283 * No rmdir "." please.
1285 if (ndp->ni_dvp == vp) {
1290 * The root of a mounted filesystem cannot be deleted.
1292 if (vp->v_vflag & VV_ROOT)
1296 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1297 if (ndp->ni_dvp == vp)
1302 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1303 nfsvno_relpathbuf(ndp);
1312 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1313 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1315 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1316 int error = 0, mirrorcnt;
1317 char fname[PNFS_FILENAME_LEN + 1];
1321 fvp = fromndp->ni_vp;
1323 vrele(fromndp->ni_dvp);
1328 tdvp = tondp->ni_dvp;
1331 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1332 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1334 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1335 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1338 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1339 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1344 * A rename to '.' or '..' results in a prematurely
1345 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1348 if ((tondp->ni_cnd.cn_namelen == 1 &&
1349 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1350 (tondp->ni_cnd.cn_namelen == 2 &&
1351 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1352 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1357 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1358 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1361 if (fvp->v_mount != tdvp->v_mount) {
1362 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1366 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1371 * If source and destination are the same, there is nothing to
1372 * do. Set error to -1 to indicate this.
1377 if (ndflag & ND_NFSV4) {
1378 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1379 error = nfsrv_checkremove(fvp, 0, p);
1380 NFSVOPUNLOCK(fvp, 0);
1384 error = nfsrv_checkremove(tvp, 1, p);
1387 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1388 * that the NFSv4 client won't be confused by the rename.
1389 * Since nfsd_recalldelegation() can only be called on an
1390 * unlocked vnode at this point and fvp is the file that will
1391 * still exist after the rename, just do fvp.
1393 nfsd_recalldelegation(fvp, p);
1395 if (error == 0 && tvp != NULL) {
1396 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1397 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1398 " dsdvp=%p\n", dsdvp[0]);
1402 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1403 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1412 vrele(fromndp->ni_dvp);
1419 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1420 * if the rename succeeded, the DS file for the tvp needs to be
1423 if (error == 0 && dsdvp[0] != NULL) {
1424 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1425 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1428 vrele(tondp->ni_startdir);
1429 nfsvno_relpathbuf(tondp);
1431 vrele(fromndp->ni_startdir);
1432 nfsvno_relpathbuf(fromndp);
1441 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1442 struct thread *p, struct nfsexstuff *exp)
1452 if (vp->v_mount != xp->v_mount)
1456 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1457 if ((vp->v_iflag & VI_DOOMED) == 0)
1458 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1461 if (ndp->ni_dvp == vp)
1465 NFSVOPUNLOCK(vp, 0);
1467 if (ndp->ni_dvp == ndp->ni_vp)
1474 nfsvno_relpathbuf(ndp);
1480 * Do the fsync() appropriate for the commit.
1483 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1489 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1490 * file is done. At this time VOP_FSYNC does not accept offset and
1491 * byte count parameters so call VOP_FSYNC the whole file for now.
1492 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1493 * File systems that do not use the buffer cache (as indicated
1494 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1496 if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1497 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1499 * Give up and do the whole thing
1502 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1503 VM_OBJECT_WLOCK(vp->v_object);
1504 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1505 VM_OBJECT_WUNLOCK(vp->v_object);
1507 error = VOP_FSYNC(vp, MNT_WAIT, td);
1510 * Locate and synchronously write any buffers that fall
1511 * into the requested range. Note: we are assuming that
1512 * f_iosize is a power of 2.
1514 int iosize = vp->v_mount->mnt_stat.f_iosize;
1515 int iomask = iosize - 1;
1520 * Align to iosize boundary, super-align to page boundary.
1523 cnt += off & iomask;
1524 off &= ~(u_quad_t)iomask;
1526 if (off & PAGE_MASK) {
1527 cnt += off & PAGE_MASK;
1528 off &= ~(u_quad_t)PAGE_MASK;
1530 lblkno = off / iosize;
1533 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1534 VM_OBJECT_WLOCK(vp->v_object);
1535 vm_object_page_clean(vp->v_object, off, off + cnt,
1537 VM_OBJECT_WUNLOCK(vp->v_object);
1546 * If we have a buffer and it is marked B_DELWRI we
1547 * have to lock and write it. Otherwise the prior
1548 * write is assumed to have already been committed.
1550 * gbincore() can return invalid buffers now so we
1551 * have to check that bit as well (though B_DELWRI
1552 * should not be set if B_INVAL is set there could be
1553 * a race here since we haven't locked the buffer).
1555 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1556 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1557 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1559 continue; /* retry */
1561 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1564 bp->b_flags &= ~B_ASYNC;
1587 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1593 if (nfsrv_devidcnt > 0) {
1594 /* For a pNFS service, get the DS numbers. */
1595 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1596 error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1602 error = VFS_STATFS(vp->v_mount, sf);
1605 sf->f_blocks = tsf->f_blocks;
1606 sf->f_bavail = tsf->f_bavail;
1607 sf->f_bfree = tsf->f_bfree;
1608 sf->f_bsize = tsf->f_bsize;
1611 * Since NFS handles these values as unsigned on the
1612 * wire, there is no way to represent negative values,
1613 * so set them to 0. Without this, they will appear
1614 * to be very large positive values for clients like
1617 if (sf->f_bavail < 0)
1619 if (sf->f_ffree < 0)
1628 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1629 * must handle nfsrv_opencheck() calls after any other access checks.
1632 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1633 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1634 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1635 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1636 struct nfsexstuff *exp, struct vnode **vpp)
1638 struct vnode *vp = NULL;
1640 struct nfsexstuff nes;
1641 struct thread *p = curthread;
1643 if (ndp->ni_vp == NULL)
1644 nd->nd_repstat = nfsrv_opencheck(clientid,
1645 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1646 if (!nd->nd_repstat) {
1647 if (ndp->ni_vp == NULL) {
1648 vrele(ndp->ni_startdir);
1649 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1650 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1651 /* For a pNFS server, create the data file on a DS. */
1652 if (nd->nd_repstat == 0) {
1654 * Create a data file on a DS for a pNFS server.
1655 * This function just returns if not
1656 * running a pNFS DS or the creation fails.
1658 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1662 nfsvno_relpathbuf(ndp);
1663 if (!nd->nd_repstat) {
1664 if (*exclusive_flagp) {
1665 *exclusive_flagp = 0;
1666 NFSVNO_ATTRINIT(nvap);
1667 nvap->na_atime.tv_sec = cverf[0];
1668 nvap->na_atime.tv_nsec = cverf[1];
1669 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1670 &nvap->na_vattr, cred);
1671 if (nd->nd_repstat != 0) {
1674 nd->nd_repstat = NFSERR_NOTSUPP;
1676 NFSSETBIT_ATTRBIT(attrbitp,
1677 NFSATTRBIT_TIMEACCESS);
1679 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1680 aclp, p, attrbitp, exp);
1685 if (ndp->ni_startdir)
1686 vrele(ndp->ni_startdir);
1687 nfsvno_relpathbuf(ndp);
1689 if (create == NFSV4OPEN_CREATE) {
1690 if (ndp->ni_dvp == vp)
1695 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1696 if (ndp->ni_cnd.cn_flags & RDONLY)
1697 NFSVNO_SETEXRDONLY(&nes);
1699 NFSVNO_EXINIT(&nes);
1700 nd->nd_repstat = nfsvno_accchk(vp,
1701 VWRITE, cred, &nes, p,
1702 NFSACCCHK_NOOVERRIDE,
1703 NFSACCCHK_VPISLOCKED, NULL);
1704 nd->nd_repstat = nfsrv_opencheck(clientid,
1705 stateidp, stp, vp, nd, p, nd->nd_repstat);
1706 if (!nd->nd_repstat) {
1707 tempsize = nvap->na_size;
1708 NFSVNO_ATTRINIT(nvap);
1709 nvap->na_size = tempsize;
1710 nd->nd_repstat = VOP_SETATTR(vp,
1711 &nvap->na_vattr, cred);
1713 } else if (vp->v_type == VREG) {
1714 nd->nd_repstat = nfsrv_opencheck(clientid,
1715 stateidp, stp, vp, nd, p, nd->nd_repstat);
1719 if (ndp->ni_cnd.cn_flags & HASBUF)
1720 nfsvno_relpathbuf(ndp);
1721 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1722 vrele(ndp->ni_startdir);
1723 if (ndp->ni_dvp == ndp->ni_vp)
1733 NFSEXITCODE2(0, nd);
1737 * Updates the file rev and sets the mtime and ctime
1738 * to the current clock time, returning the va_filerev and va_Xtime
1740 * Return ESTALE to indicate the vnode is VI_DOOMED.
1743 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1744 struct nfsrv_descript *nd, struct thread *p)
1749 vfs_timestamp(&va.va_mtime);
1750 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1751 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1752 if ((vp->v_iflag & VI_DOOMED) != 0)
1755 (void) VOP_SETATTR(vp, &va, nd->nd_cred);
1756 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1761 * Glue routine to nfsv4_fillattr().
1764 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1765 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1766 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1767 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1773 if (nfsrv_devidcnt > 0 &&
1774 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1775 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1776 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1777 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1778 error = nfsrv_pnfsstatfs(sf, mp);
1784 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1785 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1786 mounted_on_fileno, sf);
1788 NFSEXITCODE2(0, nd);
1792 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1794 * nfs readdir service
1795 * - mallocs what it thinks is enough to read
1796 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1797 * - calls VOP_READDIR()
1798 * - loops around building the reply
1799 * if the output generated exceeds count break out of loop
1800 * The NFSM_CLGET macro is used here so that the reply will be packed
1801 * tightly in mbuf clusters.
1802 * - it trims out records with d_fileno == 0
1803 * this doesn't matter for Unix clients, but they might confuse clients
1805 * - it trims out records with d_type == DT_WHT
1806 * these cannot be seen through NFS (unless we extend the protocol)
1807 * The alternate call nfsrvd_readdirplus() does lookups as well.
1808 * PS: The NFS protocol spec. does not clarify what the "count" byte
1809 * argument is a count of.. just name strings and file id's or the
1810 * entire reply rpc or ...
1811 * I tried just file name and id sizes and it confused the Sun client,
1812 * so I am using the full rpc size now. The "paranoia.." comment refers
1813 * to including the status longwords that are not a part of the dir.
1814 * "entry" structures, but are in the rpc.
1817 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1818 struct vnode *vp, struct nfsexstuff *exp)
1823 char *cpos, *cend, *rbuf;
1825 int nlen, error = 0, getret = 1;
1826 int siz, cnt, fullsiz, eofflag, ncookies;
1827 u_int64_t off, toff, verf __unused;
1828 u_long *cookies = NULL, *cookiep;
1832 struct thread *p = curthread;
1834 if (nd->nd_repstat) {
1835 nfsrv_postopattr(nd, getret, &at);
1838 if (nd->nd_flag & ND_NFSV2) {
1839 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1840 off = fxdr_unsigned(u_quad_t, *tl++);
1842 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1843 off = fxdr_hyper(tl);
1845 verf = fxdr_hyper(tl);
1849 cnt = fxdr_unsigned(int, *tl);
1850 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1851 cnt = NFS_SRVMAXDATA(nd);
1852 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1854 if (nd->nd_flag & ND_NFSV3) {
1855 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
1859 * va_filerev is not sufficient as a cookie verifier,
1860 * since it is not supposed to change when entries are
1861 * removed/added unless that offset cookies returned to
1862 * the client are no longer valid.
1864 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1865 nd->nd_repstat = NFSERR_BAD_COOKIE;
1868 if (!nd->nd_repstat && vp->v_type != VDIR)
1869 nd->nd_repstat = NFSERR_NOTDIR;
1870 if (nd->nd_repstat == 0 && cnt == 0) {
1871 if (nd->nd_flag & ND_NFSV2)
1872 /* NFSv2 does not have NFSERR_TOOSMALL */
1873 nd->nd_repstat = EPERM;
1875 nd->nd_repstat = NFSERR_TOOSMALL;
1877 if (!nd->nd_repstat)
1878 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1879 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1880 NFSACCCHK_VPISLOCKED, NULL);
1881 if (nd->nd_repstat) {
1883 if (nd->nd_flag & ND_NFSV3)
1884 nfsrv_postopattr(nd, getret, &at);
1887 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1888 rbuf = malloc(siz, M_TEMP, M_WAITOK);
1892 free(cookies, M_TEMP);
1900 io.uio_offset = (off_t)off;
1902 io.uio_segflg = UIO_SYSSPACE;
1903 io.uio_rw = UIO_READ;
1905 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1907 off = (u_int64_t)io.uio_offset;
1909 siz -= io.uio_resid;
1911 if (!cookies && !nd->nd_repstat)
1912 nd->nd_repstat = NFSERR_PERM;
1913 if (nd->nd_flag & ND_NFSV3) {
1914 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
1915 if (!nd->nd_repstat)
1916 nd->nd_repstat = getret;
1920 * Handles the failed cases. nd->nd_repstat == 0 past here.
1922 if (nd->nd_repstat) {
1926 free(cookies, M_TEMP);
1927 if (nd->nd_flag & ND_NFSV3)
1928 nfsrv_postopattr(nd, getret, &at);
1932 * If nothing read, return eof
1937 if (nd->nd_flag & ND_NFSV2) {
1938 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1940 nfsrv_postopattr(nd, getret, &at);
1941 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1942 txdr_hyper(at.na_filerev, tl);
1945 *tl++ = newnfs_false;
1948 free(cookies, M_TEMP);
1953 * Check for degenerate cases of nothing useful read.
1954 * If so go try again
1958 dp = (struct dirent *)cpos;
1962 * For some reason FreeBSD's ufs_readdir() chooses to back the
1963 * directory offset up to a block boundary, so it is necessary to
1964 * skip over the records that precede the requested offset. This
1965 * requires the assumption that file offset cookies monotonically
1968 while (cpos < cend && ncookies > 0 &&
1969 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1970 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1971 cpos += dp->d_reclen;
1972 dp = (struct dirent *)cpos;
1976 if (cpos >= cend || ncookies == 0) {
1984 * dirlen is the size of the reply, including all XDR and must
1985 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1986 * if the XDR should be included in "count", but to be safe, we do.
1987 * (Include the two booleans at the end of the reply in dirlen now.)
1989 if (nd->nd_flag & ND_NFSV3) {
1990 nfsrv_postopattr(nd, getret, &at);
1991 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1992 txdr_hyper(at.na_filerev, tl);
1993 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1995 dirlen = 2 * NFSX_UNSIGNED;
1998 /* Loop through the records and build reply */
1999 while (cpos < cend && ncookies > 0) {
2000 nlen = dp->d_namlen;
2001 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2002 nlen <= NFS_MAXNAMLEN) {
2003 if (nd->nd_flag & ND_NFSV3)
2004 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2006 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2013 * Build the directory record xdr from
2016 if (nd->nd_flag & ND_NFSV3) {
2017 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2018 *tl++ = newnfs_true;
2021 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2022 *tl++ = newnfs_true;
2024 *tl = txdr_unsigned(dp->d_fileno);
2025 (void) nfsm_strtom(nd, dp->d_name, nlen);
2026 if (nd->nd_flag & ND_NFSV3) {
2027 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2030 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2031 *tl = txdr_unsigned(*cookiep);
2033 cpos += dp->d_reclen;
2034 dp = (struct dirent *)cpos;
2040 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2041 *tl++ = newnfs_false;
2047 free(cookies, M_TEMP);
2050 NFSEXITCODE2(0, nd);
2054 NFSEXITCODE2(error, nd);
2059 * Readdirplus for V3 and Readdir for V4.
2062 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2063 struct vnode *vp, struct nfsexstuff *exp)
2068 char *cpos, *cend, *rbuf;
2071 struct nfsvattr nva, at, *nvap = &nva;
2072 struct mbuf *mb0, *mb1;
2073 struct nfsreferral *refp;
2074 int nlen, r, error = 0, getret = 1, usevget = 1;
2075 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2076 caddr_t bpos0, bpos1;
2077 u_int64_t off, toff, verf;
2078 u_long *cookies = NULL, *cookiep;
2079 nfsattrbit_t attrbits, rderrbits, savbits;
2082 struct componentname cn;
2083 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2084 struct mount *mp, *new_mp;
2085 uint64_t mounted_on_fileno;
2086 struct thread *p = curthread;
2088 if (nd->nd_repstat) {
2089 nfsrv_postopattr(nd, getret, &at);
2092 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2093 off = fxdr_hyper(tl);
2096 verf = fxdr_hyper(tl);
2098 siz = fxdr_unsigned(int, *tl++);
2099 cnt = fxdr_unsigned(int, *tl);
2102 * Use the server's maximum data transfer size as the upper bound
2105 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2106 cnt = NFS_SRVMAXDATA(nd);
2109 * siz is a "hint" of how much directory information (name, fileid,
2110 * cookie) should be in the reply. At least one client "hints" 0,
2111 * so I set it to cnt for that case. I also round it up to the
2112 * next multiple of DIRBLKSIZ.
2113 * Since the size of a Readdirplus directory entry reply will always
2114 * be greater than a directory entry returned by VOP_READDIR(), it
2115 * does not make sense to read more than NFS_SRVMAXDATA() via
2120 else if (siz > NFS_SRVMAXDATA(nd))
2121 siz = NFS_SRVMAXDATA(nd);
2122 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2124 if (nd->nd_flag & ND_NFSV4) {
2125 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2128 NFSSET_ATTRBIT(&savbits, &attrbits);
2129 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
2130 NFSZERO_ATTRBIT(&rderrbits);
2131 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2133 NFSZERO_ATTRBIT(&attrbits);
2136 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2138 if (!nd->nd_repstat) {
2139 if (off && verf != at.na_filerev) {
2141 * va_filerev is not sufficient as a cookie verifier,
2142 * since it is not supposed to change when entries are
2143 * removed/added unless that offset cookies returned to
2144 * the client are no longer valid.
2146 if (nd->nd_flag & ND_NFSV4) {
2147 nd->nd_repstat = NFSERR_NOTSAME;
2149 nd->nd_repstat = NFSERR_BAD_COOKIE;
2154 if (!nd->nd_repstat && vp->v_type != VDIR)
2155 nd->nd_repstat = NFSERR_NOTDIR;
2156 if (!nd->nd_repstat && cnt == 0)
2157 nd->nd_repstat = NFSERR_TOOSMALL;
2158 if (!nd->nd_repstat)
2159 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2160 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2161 NFSACCCHK_VPISLOCKED, NULL);
2162 if (nd->nd_repstat) {
2164 if (nd->nd_flag & ND_NFSV3)
2165 nfsrv_postopattr(nd, getret, &at);
2168 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2169 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2171 rbuf = malloc(siz, M_TEMP, M_WAITOK);
2175 free(cookies, M_TEMP);
2183 io.uio_offset = (off_t)off;
2185 io.uio_segflg = UIO_SYSSPACE;
2186 io.uio_rw = UIO_READ;
2188 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2190 off = (u_int64_t)io.uio_offset;
2192 siz -= io.uio_resid;
2194 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2196 if (!cookies && !nd->nd_repstat)
2197 nd->nd_repstat = NFSERR_PERM;
2198 if (!nd->nd_repstat)
2199 nd->nd_repstat = getret;
2200 if (nd->nd_repstat) {
2203 free(cookies, M_TEMP);
2205 if (nd->nd_flag & ND_NFSV3)
2206 nfsrv_postopattr(nd, getret, &at);
2210 * If nothing read, return eof
2215 if (nd->nd_flag & ND_NFSV3)
2216 nfsrv_postopattr(nd, getret, &at);
2217 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2218 txdr_hyper(at.na_filerev, tl);
2220 *tl++ = newnfs_false;
2222 free(cookies, M_TEMP);
2228 * Check for degenerate cases of nothing useful read.
2229 * If so go try again
2233 dp = (struct dirent *)cpos;
2237 * For some reason FreeBSD's ufs_readdir() chooses to back the
2238 * directory offset up to a block boundary, so it is necessary to
2239 * skip over the records that precede the requested offset. This
2240 * requires the assumption that file offset cookies monotonically
2243 while (cpos < cend && ncookies > 0 &&
2244 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2245 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2246 ((nd->nd_flag & ND_NFSV4) &&
2247 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2248 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2249 cpos += dp->d_reclen;
2250 dp = (struct dirent *)cpos;
2254 if (cpos >= cend || ncookies == 0) {
2261 * Busy the file system so that the mount point won't go away
2262 * and, as such, VFS_VGET() can be used safely.
2266 NFSVOPUNLOCK(vp, 0);
2267 nd->nd_repstat = vfs_busy(mp, 0);
2269 if (nd->nd_repstat != 0) {
2271 free(cookies, M_TEMP);
2273 if (nd->nd_flag & ND_NFSV3)
2274 nfsrv_postopattr(nd, getret, &at);
2279 * Check to see if entries in this directory can be safely acquired
2280 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2281 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2282 * automount of the snapshot directory that is required will
2284 * This needs to be done here for NFSv4, since NFSv4 never does
2285 * a VFS_VGET() for "." or "..".
2288 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2289 if (r == EOPNOTSUPP) {
2291 cn.cn_nameiop = LOOKUP;
2292 cn.cn_lkflags = LK_SHARED | LK_RETRY;
2293 cn.cn_cred = nd->nd_cred;
2300 * Save this position, in case there is an error before one entry
2304 bpos0 = nd->nd_bpos;
2307 * Fill in the first part of the reply.
2308 * dirlen is the reply length in bytes and cannot exceed cnt.
2309 * (Include the two booleans at the end of the reply in dirlen now,
2310 * so we recognize when we have exceeded cnt.)
2312 if (nd->nd_flag & ND_NFSV3) {
2313 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2314 nfsrv_postopattr(nd, getret, &at);
2316 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2318 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2319 txdr_hyper(at.na_filerev, tl);
2322 * Save this position, in case there is an empty reply needed.
2325 bpos1 = nd->nd_bpos;
2327 /* Loop through the records and build reply */
2329 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2330 nlen = dp->d_namlen;
2331 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2332 nlen <= NFS_MAXNAMLEN &&
2333 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2334 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2335 || (nlen == 1 && dp->d_name[0] != '.'))) {
2337 * Save the current position in the reply, in case
2338 * this entry exceeds cnt.
2341 bpos1 = nd->nd_bpos;
2344 * For readdir_and_lookup get the vnode using
2353 mounted_on_fileno = (uint64_t)dp->d_fileno;
2354 if ((nd->nd_flag & ND_NFSV3) ||
2355 NFSNONZERO_ATTRBIT(&savbits)) {
2356 if (nd->nd_flag & ND_NFSV4)
2357 refp = nfsv4root_getreferral(NULL,
2361 r = VFS_VGET(mp, dp->d_fileno,
2365 if (r == EOPNOTSUPP) {
2368 cn.cn_nameiop = LOOKUP;
2376 cn.cn_nameptr = dp->d_name;
2377 cn.cn_namelen = nlen;
2378 cn.cn_flags = ISLASTCN |
2379 NOFOLLOW | LOCKLEAF;
2381 dp->d_name[0] == '.' &&
2382 dp->d_name[1] == '.')
2385 if (NFSVOPLOCK(vp, LK_SHARED)
2387 nd->nd_repstat = EPERM;
2390 if ((vp->v_vflag & VV_ROOT) != 0
2391 && (cn.cn_flags & ISDOTDOT)
2397 r = VOP_LOOKUP(vp, &nvp,
2406 * For NFSv4, check to see if nvp is
2407 * a mount point and get the mount
2408 * point vnode, as required.
2411 nfsrv_enable_crossmntpt != 0 &&
2412 (nd->nd_flag & ND_NFSV4) != 0 &&
2413 nvp->v_type == VDIR &&
2414 nvp->v_mountedhere != NULL) {
2415 new_mp = nvp->v_mountedhere;
2416 r = vfs_busy(new_mp, 0);
2420 r = VFS_ROOT(new_mp,
2430 * If we failed to look up the entry, then it
2431 * has become invalid, most likely removed.
2438 KASSERT(refp != NULL || nvp != NULL,
2439 ("%s: undetected lookup error", __func__));
2442 ((nd->nd_flag & ND_NFSV3) ||
2443 NFSNONZERO_ATTRBIT(&attrbits))) {
2444 r = nfsvno_getfh(nvp, &nfh, p);
2446 r = nfsvno_getattr(nvp, nvap, nd, p,
2448 if (r == 0 && is_zfs == 1 &&
2449 nfsrv_enable_crossmntpt != 0 &&
2450 (nd->nd_flag & ND_NFSV4) != 0 &&
2451 nvp->v_type == VDIR &&
2452 vp->v_mount != nvp->v_mount) {
2454 * For a ZFS snapshot, there is a
2455 * pseudo mount that does not set
2456 * v_mountedhere, so it needs to
2457 * be detected via a different
2462 new_mp = nvp->v_mount;
2467 * If we failed to get attributes of the entry,
2468 * then just skip it for NFSv3 (the traditional
2469 * behavior in the old NFS server).
2470 * For NFSv4 the behavior is controlled by
2471 * RDATTRERROR: we either ignore the error or
2473 * Note that RDATTRERROR is never set for NFSv3.
2476 if (!NFSISSET_ATTRBIT(&attrbits,
2477 NFSATTRBIT_RDATTRERROR)) {
2479 if (needs_unbusy != 0)
2481 if ((nd->nd_flag & ND_NFSV3))
2490 * Build the directory record xdr
2492 if (nd->nd_flag & ND_NFSV3) {
2493 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2494 *tl++ = newnfs_true;
2496 *tl = txdr_unsigned(dp->d_fileno);
2497 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2498 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2500 *tl = txdr_unsigned(*cookiep);
2501 nfsrv_postopattr(nd, 0, nvap);
2502 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2503 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2507 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2508 *tl++ = newnfs_true;
2510 *tl = txdr_unsigned(*cookiep);
2511 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2513 supports_nfsv4acls =
2514 nfs_supportsnfsv4acls(nvp);
2515 NFSVOPUNLOCK(nvp, 0);
2517 supports_nfsv4acls = 0;
2519 dirlen += nfsrv_putreferralattr(nd,
2522 if (nd->nd_repstat) {
2525 if (needs_unbusy != 0)
2530 dirlen += nfsvno_fillattr(nd, new_mp,
2531 nvp, nvap, &nfh, r, &rderrbits,
2532 nd->nd_cred, p, isdgram, 0,
2533 supports_nfsv4acls, at_root,
2536 dirlen += nfsvno_fillattr(nd, new_mp,
2537 nvp, nvap, &nfh, r, &attrbits,
2538 nd->nd_cred, p, isdgram, 0,
2539 supports_nfsv4acls, at_root,
2544 dirlen += (3 * NFSX_UNSIGNED);
2546 if (needs_unbusy != 0)
2552 cpos += dp->d_reclen;
2553 dp = (struct dirent *)cpos;
2561 * If dirlen > cnt, we must strip off the last entry. If that
2562 * results in an empty reply, report NFSERR_TOOSMALL.
2564 if (dirlen > cnt || nd->nd_repstat) {
2565 if (!nd->nd_repstat && entrycnt == 0)
2566 nd->nd_repstat = NFSERR_TOOSMALL;
2567 if (nd->nd_repstat) {
2568 newnfs_trimtrailing(nd, mb0, bpos0);
2569 if (nd->nd_flag & ND_NFSV3)
2570 nfsrv_postopattr(nd, getret, &at);
2572 newnfs_trimtrailing(nd, mb1, bpos1);
2574 } else if (cpos < cend)
2576 if (!nd->nd_repstat) {
2577 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2578 *tl++ = newnfs_false;
2584 free(cookies, M_TEMP);
2588 NFSEXITCODE2(0, nd);
2592 NFSEXITCODE2(error, nd);
2597 * Get the settable attributes out of the mbuf list.
2598 * (Return 0 or EBADRPC)
2601 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2602 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2605 struct nfsv2_sattr *sp;
2606 int error = 0, toclient = 0;
2608 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2610 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2612 * Some old clients didn't fill in the high order 16bits.
2613 * --> check the low order 2 bytes for 0xffff
2615 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2616 nvap->na_mode = nfstov_mode(sp->sa_mode);
2617 if (sp->sa_uid != newnfs_xdrneg1)
2618 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2619 if (sp->sa_gid != newnfs_xdrneg1)
2620 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2621 if (sp->sa_size != newnfs_xdrneg1)
2622 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2623 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2625 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2627 nvap->na_atime.tv_sec =
2628 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2629 nvap->na_atime.tv_nsec = 0;
2632 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2633 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2636 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2637 if (*tl == newnfs_true) {
2638 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2639 nvap->na_mode = nfstov_mode(*tl);
2641 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2642 if (*tl == newnfs_true) {
2643 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2644 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2646 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2647 if (*tl == newnfs_true) {
2648 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2649 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2651 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2652 if (*tl == newnfs_true) {
2653 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2654 nvap->na_size = fxdr_hyper(tl);
2656 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2657 switch (fxdr_unsigned(int, *tl)) {
2658 case NFSV3SATTRTIME_TOCLIENT:
2659 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2660 fxdr_nfsv3time(tl, &nvap->na_atime);
2663 case NFSV3SATTRTIME_TOSERVER:
2664 vfs_timestamp(&nvap->na_atime);
2665 nvap->na_vaflags |= VA_UTIMES_NULL;
2668 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2669 switch (fxdr_unsigned(int, *tl)) {
2670 case NFSV3SATTRTIME_TOCLIENT:
2671 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2672 fxdr_nfsv3time(tl, &nvap->na_mtime);
2673 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2675 case NFSV3SATTRTIME_TOSERVER:
2676 vfs_timestamp(&nvap->na_mtime);
2678 nvap->na_vaflags |= VA_UTIMES_NULL;
2683 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2686 NFSEXITCODE2(error, nd);
2691 * Handle the setable attributes for V4.
2692 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2695 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2696 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2701 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2703 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2707 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2710 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2711 attrsize = fxdr_unsigned(int, *tl);
2714 * Loop around getting the setable attributes. If an unsupported
2715 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2718 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2719 bitpos = NFSATTRBIT_MAX;
2723 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2724 if (attrsum > attrsize) {
2725 error = NFSERR_BADXDR;
2728 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2730 case NFSATTRBIT_SIZE:
2731 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2732 if (vp != NULL && vp->v_type != VREG) {
2733 error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2737 nvap->na_size = fxdr_hyper(tl);
2738 attrsum += NFSX_HYPER;
2740 case NFSATTRBIT_ACL:
2741 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2745 if (aceerr && !nd->nd_repstat)
2746 nd->nd_repstat = aceerr;
2749 case NFSATTRBIT_ARCHIVE:
2750 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2751 if (!nd->nd_repstat)
2752 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2753 attrsum += NFSX_UNSIGNED;
2755 case NFSATTRBIT_HIDDEN:
2756 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2757 if (!nd->nd_repstat)
2758 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2759 attrsum += NFSX_UNSIGNED;
2761 case NFSATTRBIT_MIMETYPE:
2762 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2763 i = fxdr_unsigned(int, *tl);
2764 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2767 if (!nd->nd_repstat)
2768 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2769 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2771 case NFSATTRBIT_MODE:
2772 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2773 nvap->na_mode = nfstov_mode(*tl);
2774 attrsum += NFSX_UNSIGNED;
2776 case NFSATTRBIT_OWNER:
2777 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2778 j = fxdr_unsigned(int, *tl);
2780 error = NFSERR_BADXDR;
2783 if (j > NFSV4_SMALLSTR)
2784 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2787 error = nfsrv_mtostr(nd, cp, j);
2789 if (j > NFSV4_SMALLSTR)
2790 free(cp, M_NFSSTRING);
2793 if (!nd->nd_repstat) {
2794 nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
2796 if (!nd->nd_repstat)
2799 if (j > NFSV4_SMALLSTR)
2800 free(cp, M_NFSSTRING);
2801 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2803 case NFSATTRBIT_OWNERGROUP:
2804 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2805 j = fxdr_unsigned(int, *tl);
2807 error = NFSERR_BADXDR;
2810 if (j > NFSV4_SMALLSTR)
2811 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2814 error = nfsrv_mtostr(nd, cp, j);
2816 if (j > NFSV4_SMALLSTR)
2817 free(cp, M_NFSSTRING);
2820 if (!nd->nd_repstat) {
2821 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2823 if (!nd->nd_repstat)
2826 if (j > NFSV4_SMALLSTR)
2827 free(cp, M_NFSSTRING);
2828 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2830 case NFSATTRBIT_SYSTEM:
2831 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2832 if (!nd->nd_repstat)
2833 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2834 attrsum += NFSX_UNSIGNED;
2836 case NFSATTRBIT_TIMEACCESSSET:
2837 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2838 attrsum += NFSX_UNSIGNED;
2839 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2840 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2841 fxdr_nfsv4time(tl, &nvap->na_atime);
2843 attrsum += NFSX_V4TIME;
2845 vfs_timestamp(&nvap->na_atime);
2846 nvap->na_vaflags |= VA_UTIMES_NULL;
2849 case NFSATTRBIT_TIMEBACKUP:
2850 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2851 if (!nd->nd_repstat)
2852 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2853 attrsum += NFSX_V4TIME;
2855 case NFSATTRBIT_TIMECREATE:
2856 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2857 if (!nd->nd_repstat)
2858 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2859 attrsum += NFSX_V4TIME;
2861 case NFSATTRBIT_TIMEMODIFYSET:
2862 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2863 attrsum += NFSX_UNSIGNED;
2864 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2865 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2866 fxdr_nfsv4time(tl, &nvap->na_mtime);
2867 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2868 attrsum += NFSX_V4TIME;
2870 vfs_timestamp(&nvap->na_mtime);
2872 nvap->na_vaflags |= VA_UTIMES_NULL;
2876 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2878 * set bitpos so we drop out of the loop.
2880 bitpos = NFSATTRBIT_MAX;
2886 * some clients pad the attrlist, so we need to skip over the
2889 if (attrsum > attrsize) {
2890 error = NFSERR_BADXDR;
2892 attrsize = NFSM_RNDUP(attrsize);
2893 if (attrsum < attrsize)
2894 error = nfsm_advance(nd, attrsize - attrsum, -1);
2897 NFSEXITCODE2(error, nd);
2902 * Check/setup export credentials.
2905 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2906 struct ucred *credanon)
2911 * Check/setup credentials.
2913 if (nd->nd_flag & ND_GSS)
2914 exp->nes_exflag &= ~MNT_EXPORTANON;
2917 * Check to see if the operation is allowed for this security flavor.
2918 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2919 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2920 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2922 if (nfsvno_testexp(nd, exp) &&
2923 nd->nd_procnum != NFSV4OP_SECINFO &&
2924 nd->nd_procnum != NFSPROC_FSINFO) {
2925 if (nd->nd_flag & ND_NFSV4)
2926 error = NFSERR_WRONGSEC;
2928 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2933 * Check to see if the file system is exported V4 only.
2935 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2936 error = NFSERR_PROGNOTV4;
2941 * Now, map the user credentials.
2942 * (Note that ND_AUTHNONE will only be set for an NFSv3
2943 * Fsinfo RPC. If set for anything else, this code might need
2946 if (NFSVNO_EXPORTED(exp)) {
2947 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2948 NFSVNO_EXPORTANON(exp) ||
2949 (nd->nd_flag & ND_AUTHNONE) != 0) {
2950 nd->nd_cred->cr_uid = credanon->cr_uid;
2951 nd->nd_cred->cr_gid = credanon->cr_gid;
2952 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2953 credanon->cr_groups);
2954 } else if ((nd->nd_flag & ND_GSS) == 0) {
2956 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2957 * if there is a replacement credential with a group
2958 * list set up by "nfsuserd -manage-gids".
2959 * If there is no replacement, nfsrv_getgrpscred()
2960 * simply returns its argument.
2962 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2967 NFSEXITCODE2(error, nd);
2975 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2976 struct ucred **credp)
2978 int i, error, *secflavors;
2980 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2981 &exp->nes_numsecflavor, &secflavors);
2983 if (nfs_rootfhset) {
2984 exp->nes_exflag = 0;
2985 exp->nes_numsecflavor = 0;
2989 /* Copy the security flavors. */
2990 for (i = 0; i < exp->nes_numsecflavor; i++)
2991 exp->nes_secflavors[i] = secflavors[i];
2998 * Get a vnode for a file handle and export stuff.
3001 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3002 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3003 struct ucred **credp)
3005 int i, error, *secflavors;
3008 exp->nes_numsecflavor = 0;
3009 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3011 /* Make sure the server replies ESTALE to the client. */
3013 if (nam && !error) {
3014 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3015 &exp->nes_numsecflavor, &secflavors);
3017 if (nfs_rootfhset) {
3018 exp->nes_exflag = 0;
3019 exp->nes_numsecflavor = 0;
3025 /* Copy the security flavors. */
3026 for (i = 0; i < exp->nes_numsecflavor; i++)
3027 exp->nes_secflavors[i] = secflavors[i];
3035 * nfsd_fhtovp() - convert a fh to a vnode ptr
3036 * - look up fsid in mount list (if not found ret error)
3037 * - get vp and export rights by calling nfsvno_fhtovp()
3038 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3040 * - if mpp != NULL, return the mount point so that it can
3041 * be used for vn_finished_write() by the caller
3044 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3045 struct vnode **vpp, struct nfsexstuff *exp,
3046 struct mount **mpp, int startwrite)
3049 struct ucred *credanon;
3052 fhp = (fhandle_t *)nfp->nfsrvfh_data;
3054 * Check for the special case of the nfsv4root_fh.
3056 mp = vfs_busyfs(&fhp->fh_fsid);
3061 nd->nd_repstat = ESTALE;
3066 vn_start_write(NULL, mpp, V_WAIT);
3067 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3068 lktype = LK_EXCLUSIVE;
3070 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3075 * For NFSv4 without a pseudo root fs, unexported file handles
3076 * can be returned, so that Lookup works everywhere.
3078 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3079 !(nd->nd_flag & ND_NFSV4)) {
3081 nd->nd_repstat = EACCES;
3085 * Personally, I've never seen any point in requiring a
3086 * reserved port#, since only in the rare case where the
3087 * clients are all boxes with secure system privileges,
3088 * does it provide any enhanced security, but... some people
3089 * believe it to be useful and keep putting this code back in.
3090 * (There is also some "security checker" out there that
3091 * complains if the nfs server doesn't enforce this.)
3092 * However, note the following:
3093 * RFC3530 (NFSv4) specifies that a reserved port# not be
3095 * RFC2623 recommends that, if a reserved port# is checked for,
3096 * that there be a way to turn that off--> ifdef'd.
3098 #ifdef NFS_REQRSVPORT
3099 if (!nd->nd_repstat) {
3100 struct sockaddr_in *saddr;
3101 struct sockaddr_in6 *saddr6;
3103 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3104 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3105 if (!(nd->nd_flag & ND_NFSV4) &&
3106 ((saddr->sin_family == AF_INET &&
3107 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3108 (saddr6->sin6_family == AF_INET6 &&
3109 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3111 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3114 #endif /* NFS_REQRSVPORT */
3117 * Check/setup credentials.
3119 if (!nd->nd_repstat) {
3120 nd->nd_saveduid = nd->nd_cred->cr_uid;
3121 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
3125 if (credanon != NULL)
3127 if (nd->nd_repstat) {
3129 vn_finished_write(mp);
3136 NFSEXITCODE2(0, nd);
3143 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3145 struct filedesc *fdp;
3149 fdp = p->td_proc->p_fd;
3150 if (fd < 0 || fd >= fdp->fd_nfiles ||
3151 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3163 * Called from nfssvc() to update the exports list. Just call
3164 * vfs_export(). This has to be done, since the v4 root fake fs isn't
3165 * in the mount list.
3168 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3170 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3172 struct nameidata nd;
3175 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3176 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3178 else if (error == 0) {
3179 if (nfsexargp->fspec == NULL) {
3184 * If fspec != NULL, this is the v4root path.
3186 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3187 nfsexargp->fspec, p);
3188 if ((error = namei(&nd)) != 0)
3190 error = nfsvno_getfh(nd.ni_vp, &fh, p);
3193 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3194 NFSBCOPY((caddr_t)&fh,
3195 nfs_rootfh.nfsrvfh_data,
3196 sizeof (fhandle_t));
3207 * This function needs to test to see if the system is near its limit
3208 * for memory allocation via malloc() or mget() and return True iff
3209 * either of these resources are near their limit.
3210 * XXX (For now, this is just a stub.)
3212 int nfsrv_testmalloclimit = 0;
3214 nfsrv_mallocmget_limit(void)
3216 static int printmesg = 0;
3217 static int testval = 1;
3219 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3220 if ((printmesg++ % 100) == 0)
3221 printf("nfsd: malloc/mget near limit\n");
3228 * BSD specific initialization of a mount point.
3233 static int inited = 0;
3238 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3239 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3240 TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
3241 nfsv4root_mnt.mnt_export = NULL;
3242 TAILQ_INIT(&nfsv4root_opt);
3243 TAILQ_INIT(&nfsv4root_newopt);
3244 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3245 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3246 nfsv4root_mnt.mnt_nvnodelistsize = 0;
3247 nfsv4root_mnt.mnt_activevnodelistsize = 0;
3251 * Get a vnode for a file handle, without checking exports, etc.
3254 nfsvno_getvp(fhandle_t *fhp)
3260 mp = vfs_busyfs(&fhp->fh_fsid);
3263 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3271 * Do a local VOP_ADVLOCK().
3274 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3275 u_int64_t end, struct thread *td)
3281 if (nfsrv_dolocallocks == 0)
3283 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3285 fl.l_whence = SEEK_SET;
3287 fl.l_start = (off_t)first;
3288 if (end == NFS64BITSSET) {
3292 fl.l_len = (off_t)tlen;
3295 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3296 * values for all calls, so that all locks will be held by the
3297 * nfsd server. (The nfsd server handles conflicts between the
3299 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3300 * bytes, so it can't be put in l_sysid.
3302 if (nfsv4_sysid == 0)
3303 nfsv4_sysid = nlm_acquire_next_sysid();
3304 fl.l_pid = (pid_t)0;
3305 fl.l_sysid = (int)nfsv4_sysid;
3307 if (ftype == F_UNLCK)
3308 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3309 (F_POSIX | F_REMOTE));
3311 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3312 (F_POSIX | F_REMOTE));
3320 * Check the nfsv4 root exports.
3323 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3325 struct ucred *credanon;
3326 int exflags, error = 0, numsecflavor, *secflavors, i;
3328 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3329 &credanon, &numsecflavor, &secflavors);
3331 error = NFSERR_PROGUNAVAIL;
3334 if (credanon != NULL)
3336 for (i = 0; i < numsecflavor; i++) {
3337 if (secflavors[i] == AUTH_SYS)
3338 nd->nd_flag |= ND_EXAUTHSYS;
3339 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3340 nd->nd_flag |= ND_EXGSS;
3341 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3342 nd->nd_flag |= ND_EXGSSINTEGRITY;
3343 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3344 nd->nd_flag |= ND_EXGSSPRIVACY;
3353 * Nfs server pseudo system call for the nfsd's
3359 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3362 struct nfsd_addsock_args sockarg;
3363 struct nfsd_nfsd_args nfsdarg;
3364 struct nfsd_nfsd_oargs onfsdarg;
3365 struct nfsd_pnfsd_args pnfsdarg;
3366 struct vnode *vp, *nvp, *curdvp;
3367 struct pnfsdsfile *pf;
3368 struct nfsdevice *ds, *fds;
3369 cap_rights_t rights;
3370 int buflen, error, ret;
3371 char *buf, *cp, *cp2, *cp3;
3372 char fname[PNFS_FILENAME_LEN + 1];
3374 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3375 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3379 * Since we don't know what rights might be required,
3380 * pretend that we need them all. It is better to be too
3381 * careful than too reckless.
3383 error = fget(td, sockarg.sock,
3384 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3387 if (fp->f_type != DTYPE_SOCKET) {
3392 error = nfsrvd_addsock(fp);
3394 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3395 if (uap->argp == NULL) {
3399 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3400 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3402 nfsdarg.principal = onfsdarg.principal;
3403 nfsdarg.minthreads = onfsdarg.minthreads;
3404 nfsdarg.maxthreads = onfsdarg.maxthreads;
3405 nfsdarg.version = 1;
3406 nfsdarg.addr = NULL;
3407 nfsdarg.addrlen = 0;
3408 nfsdarg.dnshost = NULL;
3409 nfsdarg.dnshostlen = 0;
3410 nfsdarg.dspath = NULL;
3411 nfsdarg.dspathlen = 0;
3412 nfsdarg.mdspath = NULL;
3413 nfsdarg.mdspathlen = 0;
3414 nfsdarg.mirrorcnt = 1;
3417 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3420 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3421 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3422 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3423 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3424 nfsdarg.mirrorcnt >= 1 &&
3425 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3426 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3427 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3428 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3429 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3430 nfsdarg.dspathlen, nfsdarg.dnshostlen,
3431 nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3432 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3433 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3438 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */
3440 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3441 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3443 free(nfsdarg.addr, M_TEMP);
3447 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */
3448 nfsdarg.dnshost = cp;
3449 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3450 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3452 free(nfsdarg.addr, M_TEMP);
3453 free(nfsdarg.dnshost, M_TEMP);
3457 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */
3458 nfsdarg.dspath = cp;
3459 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3460 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3462 free(nfsdarg.addr, M_TEMP);
3463 free(nfsdarg.dnshost, M_TEMP);
3464 free(nfsdarg.dspath, M_TEMP);
3468 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */
3469 nfsdarg.mdspath = cp;
3471 nfsdarg.addr = NULL;
3472 nfsdarg.addrlen = 0;
3473 nfsdarg.dnshost = NULL;
3474 nfsdarg.dnshostlen = 0;
3475 nfsdarg.dspath = NULL;
3476 nfsdarg.dspathlen = 0;
3477 nfsdarg.mdspath = NULL;
3478 nfsdarg.mdspathlen = 0;
3479 nfsdarg.mirrorcnt = 1;
3481 error = nfsrvd_nfsd(td, &nfsdarg);
3482 free(nfsdarg.addr, M_TEMP);
3483 free(nfsdarg.dnshost, M_TEMP);
3484 free(nfsdarg.dspath, M_TEMP);
3485 free(nfsdarg.mdspath, M_TEMP);
3486 } else if (uap->flag & NFSSVC_PNFSDS) {
3487 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3488 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3489 pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3490 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3491 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3494 error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3496 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3497 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3498 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3499 buf = malloc(buflen, M_TEMP, M_WAITOK);
3500 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3502 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3503 if (error == 0 && pnfsdarg.dspath != NULL) {
3504 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3505 error = copyinstr(pnfsdarg.dspath, cp2,
3506 PATH_MAX + 1, NULL);
3507 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3511 if (error == 0 && pnfsdarg.curdspath != NULL) {
3512 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3513 error = copyinstr(pnfsdarg.curdspath, cp3,
3514 PATH_MAX + 1, NULL);
3515 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3522 error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3523 &buflen, fname, td, &vp, &nvp, &pf, &ds,
3525 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3527 if (pf->dsf_dir >= nfsrv_dsdirsize) {
3528 printf("copymr: dsdir out of range\n");
3531 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3532 error = nfsrv_copymr(vp, nvp,
3533 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3534 (struct pnfsdsfile *)buf,
3535 buflen / sizeof(*pf), td->td_ucred, td);
3538 if (fds != NULL && error == 0) {
3539 curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3540 ret = vn_lock(curdvp, LK_EXCLUSIVE);
3542 nfsrv_dsremove(curdvp, fname,
3544 NFSVOPUNLOCK(curdvp, 0);
3547 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3555 error = nfssvc_srvcall(td, uap, td->td_ucred);
3564 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3566 struct nfsex_args export;
3567 struct file *fp = NULL;
3569 struct nfsd_clid adminrevoke;
3570 struct nfsd_dumplist dumplist;
3571 struct nfsd_dumpclients *dumpclients;
3572 struct nfsd_dumplocklist dumplocklist;
3573 struct nfsd_dumplocks *dumplocks;
3574 struct nameidata nd;
3576 int error = EINVAL, igotlock;
3578 static int suspend_nfsd = 0;
3580 if (uap->flag & NFSSVC_PUBLICFH) {
3581 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3582 sizeof (fhandle_t));
3583 error = copyin(uap->argp,
3584 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3587 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3588 error = copyin(uap->argp,(caddr_t)&export,
3589 sizeof (struct nfsex_args));
3591 error = nfsrv_v4rootexport(&export, cred, p);
3592 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3595 } else if (uap->flag & NFSSVC_STABLERESTART) {
3596 error = copyin(uap->argp, (caddr_t)&stablefd,
3599 error = fp_getfvp(p, stablefd, &fp, &vp);
3600 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3602 if (!error && newnfs_numnfsd != 0)
3605 nfsrv_stablefirst.nsf_fp = fp;
3606 nfsrv_setupstable(p);
3608 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3609 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3610 sizeof (struct nfsd_clid));
3612 error = nfsrv_adminrevoke(&adminrevoke, p);
3613 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3614 error = copyin(uap->argp, (caddr_t)&dumplist,
3615 sizeof (struct nfsd_dumplist));
3616 if (!error && (dumplist.ndl_size < 1 ||
3617 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3620 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3621 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3622 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3623 error = copyout(dumpclients,
3624 CAST_USER_ADDR_T(dumplist.ndl_list), len);
3625 free(dumpclients, M_TEMP);
3627 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3628 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3629 sizeof (struct nfsd_dumplocklist));
3630 if (!error && (dumplocklist.ndllck_size < 1 ||
3631 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3634 error = nfsrv_lookupfilename(&nd,
3635 dumplocklist.ndllck_fname, p);
3637 len = sizeof (struct nfsd_dumplocks) *
3638 dumplocklist.ndllck_size;
3639 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3640 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3641 dumplocklist.ndllck_size, p);
3643 error = copyout(dumplocks,
3644 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3645 free(dumplocks, M_TEMP);
3647 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3650 nfsd_master_pid = procp->p_pid;
3651 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3652 nfsd_master_start = procp->p_stats->p_start;
3653 nfsd_master_proc = procp;
3655 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3656 NFSLOCKV4ROOTMUTEX();
3657 if (suspend_nfsd == 0) {
3658 /* Lock out all nfsd threads */
3660 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3661 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3662 } while (igotlock == 0 && suspend_nfsd == 0);
3665 NFSUNLOCKV4ROOTMUTEX();
3667 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3668 NFSLOCKV4ROOTMUTEX();
3669 if (suspend_nfsd != 0) {
3670 nfsv4_unlock(&nfsd_suspend_lock, 0);
3673 NFSUNLOCKV4ROOTMUTEX();
3683 * Returns 0 if ok, 1 otherwise.
3686 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3691 * This seems odd, but allow the case where the security flavor
3692 * list is empty. This happens when NFSv4 is traversing non-exported
3693 * file systems. Exported file systems should always have a non-empty
3694 * security flavor list.
3696 if (exp->nes_numsecflavor == 0)
3699 for (i = 0; i < exp->nes_numsecflavor; i++) {
3701 * The tests for privacy and integrity must be first,
3702 * since ND_GSS is set for everything but AUTH_SYS.
3704 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3705 (nd->nd_flag & ND_GSSPRIVACY))
3707 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3708 (nd->nd_flag & ND_GSSINTEGRITY))
3710 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3711 (nd->nd_flag & ND_GSS))
3713 if (exp->nes_secflavors[i] == AUTH_SYS &&
3714 (nd->nd_flag & ND_GSS) == 0)
3721 * Calculate a hash value for the fid in a file handle.
3724 nfsrv_hashfh(fhandle_t *fhp)
3728 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3733 * Calculate a hash value for the sessionid.
3736 nfsrv_hashsessionid(uint8_t *sessionid)
3740 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3745 * Signal the userland master nfsd to backup the stable restart file.
3748 nfsrv_backupstable(void)
3752 if (nfsd_master_proc != NULL) {
3753 procp = pfind(nfsd_master_pid);
3754 /* Try to make sure it is the correct process. */
3755 if (procp == nfsd_master_proc &&
3756 procp->p_stats->p_start.tv_sec ==
3757 nfsd_master_start.tv_sec &&
3758 procp->p_stats->p_start.tv_usec ==
3759 nfsd_master_start.tv_usec &&
3760 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3761 kern_psignal(procp, SIGUSR2);
3763 nfsd_master_proc = NULL;
3771 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
3772 * The arguments are in a structure, so that they can be passed through
3773 * taskqueue for a kernel process to execute this function.
3775 struct nfsrvdscreate {
3779 struct ucred *tcred;
3782 struct pnfsdsfile *pf;
3786 struct vattr createva;
3790 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
3791 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
3792 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
3795 struct nameidata named;
3800 struct nfsmount *nmp;
3803 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
3804 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
3805 nfsvno_setpathbuf(&named, &bufp, &hashp);
3806 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
3807 named.ni_cnd.cn_thread = p;
3808 named.ni_cnd.cn_nameptr = bufp;
3809 if (fnamep != NULL) {
3810 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
3811 named.ni_cnd.cn_namelen = strlen(bufp);
3813 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
3814 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
3816 /* Create the date file in the DS mount. */
3817 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
3819 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
3820 NFSVOPUNLOCK(dvp, 0);
3822 /* Set the ownership of the file. */
3823 error = VOP_SETATTR(nvp, nvap, tcred);
3824 NFSD_DEBUG(4, "nfsrv_dscreate:"
3825 " setattr-uid=%d\n", error);
3830 printf("pNFS: pnfscreate failed=%d\n", error);
3832 printf("pNFS: pnfscreate vnlock=%d\n", error);
3835 nmp = VFSTONFS(nvp->v_mount);
3836 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
3837 != 0 || nmp->nm_nam->sa_len > sizeof(
3838 struct sockaddr_in6) ||
3839 np->n_fhp->nfh_len != NFSX_MYFH) {
3840 printf("Bad DS file: fstype=%s salen=%d"
3842 nvp->v_mount->mnt_vfc->vfc_name,
3843 nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
3847 /* Set extattrs for the DS on the MDS file. */
3850 error = VOP_GETATTR(nvp, &va, tcred);
3852 dsa->dsa_filerev = va.va_filerev;
3853 dsa->dsa_size = va.va_size;
3854 dsa->dsa_atime = va.va_atime;
3855 dsa->dsa_mtime = va.va_mtime;
3859 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
3861 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
3862 nmp->nm_nam->sa_len);
3863 NFSBCOPY(named.ni_cnd.cn_nameptr,
3865 sizeof(pf->dsf_filename));
3868 printf("pNFS: pnfscreate can't get DS"
3869 " attr=%d\n", error);
3870 if (nvpp != NULL && error == 0)
3875 nfsvno_relpathbuf(&named);
3880 * Start up the thread that will execute nfsrv_dscreate().
3883 start_dscreate(void *arg, int pending)
3885 struct nfsrvdscreate *dsc;
3887 dsc = (struct nfsrvdscreate *)arg;
3888 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
3889 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
3891 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
3895 * Create a pNFS data file on the Data Server(s).
3898 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
3901 struct nfsrvdscreate *dsc, *tdsc;
3902 struct nfsdevice *ds, *tds, *fds;
3904 struct pnfsdsfile *pf, *tpf;
3905 struct pnfsdsattr dsattr;
3907 struct vnode *dvp[NFSDEV_MAXMIRRORS];
3908 struct nfsmount *nmp;
3913 struct ucred *tcred;
3914 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
3917 /* Get a DS server directory in a round-robin order. */
3923 * Search for the first entry that handles this MDS fs, but use the
3924 * first entry for all MDS fs's otherwise.
3926 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
3927 if (tds->nfsdev_nmp != NULL) {
3928 if (tds->nfsdev_mdsisset == 0 && ds == NULL)
3930 else if (tds->nfsdev_mdsisset != 0 &&
3931 mp->mnt_stat.f_fsid.val[0] ==
3932 tds->nfsdev_mdsfsid.val[0] &&
3933 mp->mnt_stat.f_fsid.val[1] ==
3934 tds->nfsdev_mdsfsid.val[1]) {
3942 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
3945 i = dsdir[0] = ds->nfsdev_nextdir;
3946 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
3947 dvp[0] = ds->nfsdev_dsdir[i];
3948 tds = TAILQ_NEXT(ds, nfsdev_list);
3949 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
3950 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
3951 if (tds->nfsdev_nmp != NULL &&
3952 ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
3953 (tds->nfsdev_mdsisset != 0 && fds != NULL &&
3954 mp->mnt_stat.f_fsid.val[0] ==
3955 tds->nfsdev_mdsfsid.val[0] &&
3956 mp->mnt_stat.f_fsid.val[1] ==
3957 tds->nfsdev_mdsfsid.val[1]))) {
3958 dsdir[mirrorcnt] = i;
3959 dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
3961 if (mirrorcnt >= nfsrv_maxpnfsmirror)
3966 /* Put at end of list to implement round-robin usage. */
3967 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
3968 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
3972 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
3974 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
3977 error = nfsvno_getfh(vp, &fh, p);
3979 error = VOP_GETATTR(vp, &va, cred);
3981 /* Set the attributes for "vp" to Setattr the DS vp. */
3984 vamode = va.va_mode;
3988 va.va_mode = vamode;
3991 printf("pNFS: pnfscreate getfh+attr=%d\n", error);
3993 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
3995 /* Make data file name based on FH. */
3996 tcred = newnfs_getcred();
3999 * Create the file on each DS mirror, using kernel process(es) for the
4000 * additional mirrors.
4003 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4004 tpf->dsf_dir = dsdir[i];
4005 tdsc->tcred = tcred;
4008 tdsc->createva = *vap;
4009 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4016 if (nfs_pnfsiothreads != 0) {
4017 ret = nfs_pnfsio(start_dscreate, tdsc);
4018 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4021 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4022 NULL, tcred, p, NULL);
4024 KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4026 if (failpos == -1 && nfsds_failerr(ret))
4034 tpf->dsf_dir = dsdir[mirrorcnt - 1];
4035 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4036 &dsattr, NULL, tcred, p, NULL);
4037 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4038 failpos = mirrorcnt - 1;
4042 timo = hz / 50; /* Wait for 20msec. */
4045 /* Wait for kernel task(s) to complete. */
4046 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4047 while (tdsc->inprog != 0 && tdsc->done == 0)
4048 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4049 if (tdsc->err != 0) {
4050 if (failpos == -1 && nfsds_failerr(tdsc->err))
4052 else if (error == 0)
4058 * If failpos has been set, that mirror has failed, so it needs
4062 nmp = VFSTONFS(dvp[failpos]->v_mount);
4064 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4065 NFSMNTP_CANCELRPCS)) == 0) {
4066 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4068 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4069 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4072 nfsrv_killrpcs(nmp);
4074 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4082 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4084 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4085 mirrorcnt, nfsrv_maxpnfsmirror);
4087 * For all mirrors that couldn't be created, fill in the
4088 * *pf structure, but with an IP address == 0.0.0.0.
4090 tpf = pf + mirrorcnt;
4091 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4093 tpf->dsf_sin.sin_family = AF_INET;
4094 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4095 tpf->dsf_sin.sin_addr.s_addr = 0;
4096 tpf->dsf_sin.sin_port = 0;
4099 error = vn_extattr_set(vp, IO_NODELOCKED,
4100 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4101 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4103 error = vn_extattr_set(vp, IO_NODELOCKED,
4104 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4105 sizeof(dsattr), (char *)&dsattr, p);
4107 printf("pNFS: pnfscreate setextattr=%d\n",
4110 printf("pNFS: pnfscreate=%d\n", error);
4116 * Get the information needed to remove the pNFS Data Server file from the
4117 * Metadata file. Upon success, ddvp is set non-NULL to the locked
4118 * DS directory vnode. The caller must unlock *ddvp when done with it.
4121 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4122 int *mirrorcntp, char *fname, fhandle_t *fhp)
4125 struct ucred *tcred;
4130 /* If not an exported regular file or not a pNFS server, just return. */
4131 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4132 nfsrv_devidcnt == 0)
4135 /* Check to see if this is the last hard link. */
4136 tcred = newnfs_getcred();
4137 error = VOP_GETATTR(vp, &va, tcred);
4140 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4143 if (va.va_nlink > 1)
4146 error = nfsvno_getfh(vp, fhp, p);
4148 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4153 buf = malloc(buflen, M_TEMP, M_WAITOK);
4154 /* Get the directory vnode for the DS mount and the file handle. */
4155 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4156 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4159 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4163 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4164 * The arguments are in a structure, so that they can be passed through
4165 * taskqueue for a kernel process to execute this function.
4167 struct nfsrvdsremove {
4171 struct ucred *tcred;
4175 char fname[PNFS_FILENAME_LEN + 1];
4179 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4182 struct nameidata named;
4188 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4191 named.ni_cnd.cn_nameiop = DELETE;
4192 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4193 named.ni_cnd.cn_cred = tcred;
4194 named.ni_cnd.cn_thread = p;
4195 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4196 nfsvno_setpathbuf(&named, &bufp, &hashp);
4197 named.ni_cnd.cn_nameptr = bufp;
4198 named.ni_cnd.cn_namelen = strlen(fname);
4199 strlcpy(bufp, fname, NAME_MAX);
4200 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4201 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4202 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4204 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4207 NFSVOPUNLOCK(dvp, 0);
4208 nfsvno_relpathbuf(&named);
4210 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4215 * Start up the thread that will execute nfsrv_dsremove().
4218 start_dsremove(void *arg, int pending)
4220 struct nfsrvdsremove *dsrm;
4222 dsrm = (struct nfsrvdsremove *)arg;
4223 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4226 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4230 * Remove a pNFS data file from a Data Server.
4231 * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4232 * removed to set up the dvp and fill in the FH.
4235 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4238 struct ucred *tcred;
4239 struct nfsrvdsremove *dsrm, *tdsrm;
4240 struct nfsdevice *ds;
4241 struct nfsmount *nmp;
4242 int failpos, i, ret, timo;
4244 tcred = newnfs_getcred();
4247 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4249 * Remove the file on each DS mirror, using kernel process(es) for the
4250 * additional mirrors.
4253 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4254 tdsrm->tcred = tcred;
4256 tdsrm->dvp = dvp[i];
4257 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4262 if (nfs_pnfsiothreads != 0) {
4263 ret = nfs_pnfsio(start_dsremove, tdsrm);
4264 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4267 ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4268 if (failpos == -1 && nfsds_failerr(ret))
4272 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4273 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4274 failpos = mirrorcnt - 1;
4275 timo = hz / 50; /* Wait for 20msec. */
4278 /* Wait for kernel task(s) to complete. */
4279 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4280 while (tdsrm->inprog != 0 && tdsrm->done == 0)
4281 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4282 if (failpos == -1 && nfsds_failerr(tdsrm->err))
4287 * If failpos has been set, that mirror has failed, so it needs
4291 nmp = VFSTONFS(dvp[failpos]->v_mount);
4293 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4294 NFSMNTP_CANCELRPCS)) == 0) {
4295 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4297 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4298 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4301 nfsrv_killrpcs(nmp);
4303 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4309 /* Get rid all layouts for the file. */
4310 nfsrv_freefilelayouts(fhp);
4317 * Generate a file name based on the file handle and put it in *bufp.
4318 * Return the number of bytes generated.
4321 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4325 const uint8_t *hexdigits = "0123456789abcdef";
4327 cp = (uint8_t *)fhp;
4328 for (i = 0; i < sizeof(*fhp); i++) {
4329 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4330 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4337 * Update the Metadata file's attributes from the DS file when a Read/Write
4338 * layout is returned.
4339 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4340 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4343 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4345 struct ucred *tcred;
4348 /* Do this as root so that it won't fail with EACCES. */
4349 tcred = newnfs_getcred();
4350 error = nfsrv_proxyds(NULL, vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4351 NULL, NULL, NULL, nap, NULL);
4357 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4360 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4365 error = nfsrv_proxyds(NULL, vp, 0, 0, cred, p, NFSPROC_SETACL,
4366 NULL, NULL, NULL, NULL, aclp);
4371 nfsrv_proxyds(struct nfsrv_descript *nd, struct vnode *vp, off_t off, int cnt,
4372 struct ucred *cred, struct thread *p, int ioproc, struct mbuf **mpp,
4373 char *cp, struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp)
4375 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4376 fhandle_t fh[NFSDEV_MAXMIRRORS];
4377 struct vnode *dvp[NFSDEV_MAXMIRRORS];
4378 struct nfsdevice *ds;
4379 struct pnfsdsattr dsattr;
4381 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4383 NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4385 * If not a regular file, not exported or not a pNFS server,
4386 * just return ENOENT.
4388 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4389 nfsrv_devidcnt == 0)
4393 buf = malloc(buflen, M_TEMP, M_WAITOK);
4397 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4398 * from the MetaData file's extended attribute.
4400 if (ioproc == NFSPROC_GETATTR) {
4401 error = vn_extattr_get(vp, IO_NODELOCKED,
4402 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4404 if (error == 0 && buflen != sizeof(dsattr))
4407 NFSBCOPY(buf, &dsattr, buflen);
4408 nap->na_filerev = dsattr.dsa_filerev;
4409 nap->na_size = dsattr.dsa_size;
4410 nap->na_atime = dsattr.dsa_atime;
4411 nap->na_mtime = dsattr.dsa_mtime;
4414 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4415 * returns 0, just return now. nfsrv_checkdsattr()
4416 * returns 0 if there is no Read/Write layout
4417 * plus either an Open/Write_access or Write
4418 * delegation issued to a client for the file.
4420 if (nfsrv_pnfsgetdsattr == 0 ||
4421 nfsrv_checkdsattr(nd, vp, p) == 0) {
4428 * Clear ENOATTR so the code below will attempt to do a
4429 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4430 * the extended attribute.
4432 if (error == ENOATTR)
4441 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4443 printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4444 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4445 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4448 for (i = 0; i < mirrorcnt; i++)
4449 nmp[i] = VFSTONFS(dvp[i]->v_mount);
4451 printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4453 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4456 if (origmircnt == -1)
4457 origmircnt = mirrorcnt;
4459 * If failpos is set to a mirror#, then that mirror has
4460 * failed and will be disabled. For Read and Getattr, the
4461 * function only tries one mirror, so if that mirror has
4462 * failed, it will need to be retried. As such, increment
4463 * tryitagain for these cases.
4464 * For Write, Setattr and Setacl, the function tries all
4465 * mirrors and will not return an error for the case where
4466 * one mirror has failed. For these cases, the functioning
4467 * mirror(s) will have been modified, so a retry isn't
4468 * necessary. These functions will set failpos for the
4471 if (ioproc == NFSPROC_READDS) {
4472 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4474 if (nfsds_failerr(error) && mirrorcnt > 1) {
4476 * Setting failpos will cause the mirror
4477 * to be disabled and then a retry of this
4484 } else if (ioproc == NFSPROC_WRITEDS)
4485 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4486 &nmp[0], mirrorcnt, mpp, cp, &failpos);
4487 else if (ioproc == NFSPROC_SETATTR)
4488 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4489 mirrorcnt, nap, &failpos);
4490 else if (ioproc == NFSPROC_SETACL)
4491 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4492 mirrorcnt, aclp, &failpos);
4494 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4495 vp, nmp[mirrorcnt - 1], nap);
4496 if (nfsds_failerr(error) && mirrorcnt > 1) {
4498 * Setting failpos will cause the mirror
4499 * to be disabled and then a retry of this
4500 * getattr is required.
4502 failpos = mirrorcnt - 1;
4509 failnmp = nmp[failpos];
4510 NFSLOCKMNT(failnmp);
4511 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4512 NFSMNTP_CANCELRPCS)) == 0) {
4513 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4514 NFSUNLOCKMNT(failnmp);
4515 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4517 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4520 nfsrv_killrpcs(failnmp);
4521 NFSLOCKMNT(failnmp);
4522 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4525 NFSUNLOCKMNT(failnmp);
4527 for (i = 0; i < mirrorcnt; i++)
4528 NFSVOPUNLOCK(dvp[i], 0);
4529 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4531 /* Try the Read/Getattr again if a mirror was deleted. */
4532 if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4535 /* Return ENOENT for any Extended Attribute error. */
4539 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4544 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4546 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4547 * to be checked. If it points to a NULL nmp, then it returns
4548 * a suitable destination.
4549 * curnmp - If non-NULL, it is the source mount for the copy.
4552 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4553 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4554 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4555 struct nfsmount *curnmp, int *ippos, int *dsdirp)
4557 struct vnode *dvp, *nvp, **tdvpp;
4559 struct nfsmount *nmp, *newnmp;
4560 struct sockaddr *sad;
4561 struct sockaddr_in *sin;
4562 struct nfsdevice *ds, *tds, *fndds;
4563 struct pnfsdsfile *pf;
4565 int error, fhiszero, fnd, gotone, i, mirrorcnt;
4567 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4576 if (newnmpp != NULL)
4581 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4582 "pnfsd.dsfile", buflenp, buf, p);
4583 mirrorcnt = *buflenp / sizeof(*pf);
4584 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4585 *buflenp != sizeof(*pf) * mirrorcnt))
4588 pf = (struct pnfsdsfile *)buf;
4589 /* If curnmp != NULL, check for a match in the mirror list. */
4590 if (curnmp != NULL && error == 0) {
4592 for (i = 0; i < mirrorcnt; i++, pf++) {
4593 sad = (struct sockaddr *)&pf->dsf_sin;
4594 if (nfsaddr2_match(sad, curnmp->nm_nam)) {
4606 pf = (struct pnfsdsfile *)buf;
4607 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
4609 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
4611 sad = (struct sockaddr *)&pf->dsf_sin;
4613 dsdir = pf->dsf_dir;
4614 if (dsdir >= nfsrv_dsdirsize) {
4615 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
4617 } else if (nvpp != NULL && newnmp != NULL &&
4618 nfsaddr2_match(sad, newnmp->nm_nam))
4621 if (ippos != NULL && curnmp == NULL &&
4622 sad->sa_family == AF_INET &&
4623 sin->sin_addr.s_addr == 0)
4625 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
4627 /* Use the socket address to find the mount point. */
4630 /* Find a match for the IP address. */
4631 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
4632 if (ds->nfsdev_nmp != NULL) {
4633 dvp = ds->nfsdev_dvp;
4634 nmp = VFSTONFS(dvp->v_mount);
4635 if (nmp != ds->nfsdev_nmp)
4636 printf("different2 nmp %p %p\n",
4637 nmp, ds->nfsdev_nmp);
4638 if (nfsaddr2_match(sad, nmp->nm_nam)) {
4644 if (fndds != NULL && newnmpp != NULL &&
4646 /* Search for a place to make a mirror copy. */
4647 TAILQ_FOREACH(tds, &nfsrv_devidhead,
4649 if (tds->nfsdev_nmp != NULL &&
4651 ((tds->nfsdev_mdsisset == 0 &&
4652 fndds->nfsdev_mdsisset == 0) ||
4653 (tds->nfsdev_mdsisset != 0 &&
4654 fndds->nfsdev_mdsisset != 0 &&
4655 tds->nfsdev_mdsfsid.val[0] ==
4656 mp->mnt_stat.f_fsid.val[0] &&
4657 tds->nfsdev_mdsfsid.val[1] ==
4658 mp->mnt_stat.f_fsid.val[1]))) {
4659 *newnmpp = tds->nfsdev_nmp;
4665 * Move this entry to the end of the
4666 * list, so it won't be selected as
4667 * easily the next time.
4669 TAILQ_REMOVE(&nfsrv_devidhead, tds,
4671 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
4676 if (fndds != NULL) {
4677 dvp = fndds->nfsdev_dsdir[dsdir];
4678 if (lktype != 0 || fhiszero != 0 ||
4679 (nvpp != NULL && *nvpp == NULL)) {
4681 error = vn_lock(dvp,
4683 else if (lktype != 0)
4684 error = vn_lock(dvp, lktype);
4686 error = vn_lock(dvp, LK_SHARED);
4688 * If the file handle is all 0's, try to
4689 * do a Lookup against the DS to acquire
4691 * If dvpp == NULL or the Lookup fails,
4692 * unlock dvp after the call.
4694 if (error == 0 && (fhiszero != 0 ||
4695 (nvpp != NULL && *nvpp == NULL))) {
4696 error = nfsrv_pnfslookupds(vp,
4712 if (error != 0 || lktype == 0)
4713 NFSVOPUNLOCK(dvp, 0);
4718 NFSD_DEBUG(4, "gotone=%d\n", gotone);
4719 if (devid != NULL) {
4720 NFSBCOPY(fndds->nfsdev_deviceid,
4721 devid, NFSX_V4DEVICEID);
4722 devid += NFSX_V4DEVICEID;
4727 NFSBCOPY(&pf->dsf_fh, fhp++,
4729 if (fnamep != NULL && gotone == 1)
4732 sizeof(pf->dsf_filename));
4734 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
4739 if (error == 0 && gotone == 0)
4742 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
4745 *mirrorcntp = gotone;
4747 if (gotone > 0 && dvpp != NULL) {
4749 * If the error didn't occur on the first one and
4750 * dvpp != NULL, the one(s) prior to the failure will
4751 * have locked dvp's that need to be unlocked.
4753 for (i = 0; i < gotone; i++) {
4754 NFSVOPUNLOCK(*dvpp, 0);
4759 * If it found the vnode to be copied from before a failure,
4760 * it needs to be vput()'d.
4762 if (nvpp != NULL && *nvpp != NULL) {
4771 * Set the extended attribute for the Change attribute.
4774 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4776 struct pnfsdsattr dsattr;
4779 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
4780 dsattr.dsa_filerev = nap->na_filerev;
4781 dsattr.dsa_size = nap->na_size;
4782 dsattr.dsa_atime = nap->na_atime;
4783 dsattr.dsa_mtime = nap->na_mtime;
4784 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4785 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
4787 printf("pNFS: setextattr=%d\n", error);
4792 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
4793 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
4796 struct nfsrv_descript *nd;
4798 struct mbuf *m, *m2;
4799 int error = 0, retlen, tlen, trimlen;
4801 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
4802 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4805 * Use a stateid where other is an alternating 01010 pattern and
4806 * seqid is 0xffffffff. This value is not defined as special by
4807 * the RFC and is used by the FreeBSD NFS server to indicate an
4808 * MDS->DS proxy operation.
4810 st.other[0] = 0x55555555;
4811 st.other[1] = 0x55555555;
4812 st.other[2] = 0x55555555;
4813 st.seqid = 0xffffffff;
4814 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
4816 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
4817 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
4818 txdr_hyper(off, tl);
4819 *(tl + 2) = txdr_unsigned(len);
4820 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
4821 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
4826 if (nd->nd_repstat == 0) {
4827 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4828 NFSM_STRSIZ(retlen, len);
4830 /* Trim off the pre-data XDR from the mbuf chain. */
4832 while (m != NULL && m != nd->nd_md) {
4833 if (m->m_next == nd->nd_md) {
4835 m_freem(nd->nd_mrep);
4836 nd->nd_mrep = m = nd->nd_md;
4841 printf("nfsrv_readdsrpc: busted mbuf list\n");
4847 * Now, adjust first mbuf so that any XDR before the
4848 * read data is skipped over.
4850 trimlen = nd->nd_dpos - mtod(m, char *);
4852 m->m_len -= trimlen;
4853 NFSM_DATAP(m, trimlen);
4857 * Truncate the mbuf chain at retlen bytes of data,
4858 * plus XDR padding that brings the length up to a
4861 tlen = NFSM_RNDUP(retlen);
4863 if (m->m_len >= tlen) {
4873 } while (m != NULL);
4875 printf("nfsrv_readdsrpc: busted mbuf list\n");
4884 error = nd->nd_repstat;
4886 /* If nd->nd_mrep is already NULL, this is a no-op. */
4887 m_freem(nd->nd_mrep);
4889 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
4894 * Do a write RPC on a DS data file, using this structure for the arguments,
4895 * so that this function can be executed by a separate kernel process.
4897 struct nfsrvwritedsdorpc {
4904 struct nfsmount *nmp;
4912 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
4913 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
4916 struct nfsrv_descript *nd;
4917 nfsattrbit_t attrbits;
4919 int commit, error, retlen;
4921 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4922 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
4923 sizeof(fhandle_t), NULL, NULL, 0, 0);
4926 * Use a stateid where other is an alternating 01010 pattern and
4927 * seqid is 0xffffffff. This value is not defined as special by
4928 * the RFC and is used by the FreeBSD NFS server to indicate an
4929 * MDS->DS proxy operation.
4931 st.other[0] = 0x55555555;
4932 st.other[1] = 0x55555555;
4933 st.other[2] = 0x55555555;
4934 st.seqid = 0xffffffff;
4935 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
4936 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
4937 txdr_hyper(off, tl);
4940 * Do all writes FileSync, since the server doesn't hold onto dirty
4941 * buffers. Since clients should be accessing the DS servers directly
4942 * using the pNFS layouts, this just needs to work correctly as a
4945 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
4946 *tl = txdr_unsigned(len);
4947 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
4949 /* Put data in mbuf chain. */
4950 nd->nd_mb->m_next = m;
4952 /* Set nd_mb and nd_bpos to end of data. */
4953 while (m->m_next != NULL)
4956 nd->nd_bpos = mtod(m, char *) + m->m_len;
4957 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
4959 /* Do a Getattr for Size, Change and Modify Time. */
4960 NFSZERO_ATTRBIT(&attrbits);
4961 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
4962 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
4963 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
4964 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
4965 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4966 *tl = txdr_unsigned(NFSV4OP_GETATTR);
4967 (void) nfsrv_putattrbit(nd, &attrbits);
4968 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
4969 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
4974 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
4975 /* Get rid of weak cache consistency data for now. */
4976 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
4977 (ND_NFSV4 | ND_V4WCCATTR)) {
4978 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
4979 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
4980 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
4984 * Get rid of Op# and status for next op.
4986 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
4988 nd->nd_flag |= ND_NOMOREDATA;
4990 if (nd->nd_repstat == 0) {
4991 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
4992 retlen = fxdr_unsigned(int, *tl++);
4993 commit = fxdr_unsigned(int, *tl);
4994 if (commit != NFSWRITE_FILESYNC)
4996 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
4997 retlen, commit, error);
4999 error = nd->nd_repstat;
5000 /* We have no use for the Write Verifier since we use FileSync. */
5003 * Get the Change, Size, Access Time and Modify Time attributes and set
5004 * on the Metadata file, so its attributes will be what the file's
5005 * would be if it had been written.
5008 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5009 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5010 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5012 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5014 m_freem(nd->nd_mrep);
5016 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5021 * Start up the thread that will execute nfsrv_writedsdorpc().
5024 start_writedsdorpc(void *arg, int pending)
5026 struct nfsrvwritedsdorpc *drpc;
5028 drpc = (struct nfsrvwritedsdorpc *)arg;
5029 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5030 drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5032 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5036 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5037 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5038 struct mbuf **mpp, char *cp, int *failposp)
5040 struct nfsrvwritedsdorpc *drpc, *tdrpc;
5043 int error, i, offs, ret, timo;
5045 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5046 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5049 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5052 /* Calculate offset in mbuf chain that data starts. */
5053 offs = cp - mtod(*mpp, char *);
5054 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5057 * Do the write RPC for every DS, using a separate kernel process
5058 * for every DS except the last one.
5061 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5063 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5071 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5073 if (nfs_pnfsiothreads != 0) {
5074 ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5075 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5079 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5081 if (nfsds_failerr(ret) && *failposp == -1)
5083 else if (error == 0 && ret != 0)
5089 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5090 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5091 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5092 *failposp = mirrorcnt - 1;
5093 else if (error == 0 && ret != 0)
5096 error = nfsrv_setextattr(vp, &na, p);
5097 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5099 timo = hz / 50; /* Wait for 20msec. */
5102 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5103 /* Wait for RPCs on separate threads to complete. */
5104 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5105 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5106 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5108 else if (error == 0 && tdrpc->err != 0)
5116 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5117 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5118 struct nfsvattr *dsnap)
5121 struct nfsrv_descript *nd;
5123 nfsattrbit_t attrbits;
5126 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5127 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5129 * Use a stateid where other is an alternating 01010 pattern and
5130 * seqid is 0xffffffff. This value is not defined as special by
5131 * the RFC and is used by the FreeBSD NFS server to indicate an
5132 * MDS->DS proxy operation.
5134 st.other[0] = 0x55555555;
5135 st.other[1] = 0x55555555;
5136 st.other[2] = 0x55555555;
5137 st.seqid = 0xffffffff;
5138 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5140 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5141 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5143 /* Do a Getattr for Size, Change, Access Time and Modify Time. */
5144 NFSZERO_ATTRBIT(&attrbits);
5145 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5146 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5147 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5148 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5149 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5150 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5151 (void) nfsrv_putattrbit(nd, &attrbits);
5152 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5153 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5158 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5160 /* Get rid of weak cache consistency data for now. */
5161 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5162 (ND_NFSV4 | ND_V4WCCATTR)) {
5163 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5164 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5165 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5169 * Get rid of Op# and status for next op.
5171 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5173 nd->nd_flag |= ND_NOMOREDATA;
5175 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5178 if (nd->nd_repstat != 0)
5179 error = nd->nd_repstat;
5181 * Get the Change, Size, Access Time and Modify Time attributes and set
5182 * on the Metadata file, so its attributes will be what the file's
5183 * would be if it had been written.
5186 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5187 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5188 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5190 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5192 m_freem(nd->nd_mrep);
5194 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5198 struct nfsrvsetattrdsdorpc {
5203 struct nfsmount *nmp;
5208 struct nfsvattr dsna;
5213 * Start up the thread that will execute nfsrv_setattrdsdorpc().
5216 start_setattrdsdorpc(void *arg, int pending)
5218 struct nfsrvsetattrdsdorpc *drpc;
5220 drpc = (struct nfsrvsetattrdsdorpc *)arg;
5221 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5222 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5227 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5228 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5229 struct nfsvattr *nap, int *failposp)
5231 struct nfsrvsetattrdsdorpc *drpc, *tdrpc;
5233 int error, i, ret, timo;
5235 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5238 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5242 * Do the setattr RPC for every DS, using a separate kernel process
5243 * for every DS except the last one.
5246 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5249 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5257 if (nfs_pnfsiothreads != 0) {
5258 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5259 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5263 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5265 if (nfsds_failerr(ret) && *failposp == -1)
5267 else if (error == 0 && ret != 0)
5273 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
5274 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5275 *failposp = mirrorcnt - 1;
5276 else if (error == 0 && ret != 0)
5279 error = nfsrv_setextattr(vp, &na, p);
5280 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
5282 timo = hz / 50; /* Wait for 20msec. */
5285 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5286 /* Wait for RPCs on separate threads to complete. */
5287 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5288 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
5289 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5291 else if (error == 0 && tdrpc->err != 0)
5299 * Do a Setattr of an NFSv4 ACL on the DS file.
5302 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5303 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
5305 struct nfsrv_descript *nd;
5307 nfsattrbit_t attrbits;
5310 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
5311 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5313 * Use a stateid where other is an alternating 01010 pattern and
5314 * seqid is 0xffffffff. This value is not defined as special by
5315 * the RFC and is used by the FreeBSD NFS server to indicate an
5316 * MDS->DS proxy operation.
5318 st.other[0] = 0x55555555;
5319 st.other[1] = 0x55555555;
5320 st.other[2] = 0x55555555;
5321 st.seqid = 0xffffffff;
5322 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5324 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5325 NFSZERO_ATTRBIT(&attrbits);
5326 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
5328 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
5329 * so passing in the metadata "vp" will be ok, since it is of
5330 * the same type (VREG).
5332 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
5333 NULL, 0, 0, 0, 0, 0, NULL);
5334 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5335 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5340 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
5342 error = nd->nd_repstat;
5343 m_freem(nd->nd_mrep);
5348 struct nfsrvsetacldsdorpc {
5353 struct nfsmount *nmp;
5362 * Start up the thread that will execute nfsrv_setacldsdorpc().
5365 start_setacldsdorpc(void *arg, int pending)
5367 struct nfsrvsetacldsdorpc *drpc;
5369 drpc = (struct nfsrvsetacldsdorpc *)arg;
5370 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
5371 drpc->vp, drpc->nmp, drpc->aclp);
5376 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5377 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
5380 struct nfsrvsetacldsdorpc *drpc, *tdrpc;
5381 int error, i, ret, timo;
5383 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
5386 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5390 * Do the setattr RPC for every DS, using a separate kernel process
5391 * for every DS except the last one.
5394 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5397 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5405 if (nfs_pnfsiothreads != 0) {
5406 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
5407 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
5411 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
5413 if (nfsds_failerr(ret) && *failposp == -1)
5415 else if (error == 0 && ret != 0)
5421 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
5422 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5423 *failposp = mirrorcnt - 1;
5424 else if (error == 0 && ret != 0)
5426 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
5428 timo = hz / 50; /* Wait for 20msec. */
5431 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5432 /* Wait for RPCs on separate threads to complete. */
5433 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5434 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
5435 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5437 else if (error == 0 && tdrpc->err != 0)
5445 * Getattr call to the DS for the Modify, Size and Change attributes.
5448 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5449 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
5451 struct nfsrv_descript *nd;
5453 nfsattrbit_t attrbits;
5455 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
5456 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5457 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
5458 sizeof(fhandle_t), NULL, NULL, 0, 0);
5459 NFSZERO_ATTRBIT(&attrbits);
5460 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5461 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5462 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5463 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5464 (void) nfsrv_putattrbit(nd, &attrbits);
5465 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5466 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5471 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
5473 if (nd->nd_repstat == 0) {
5474 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
5475 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
5478 * We can only save the updated values in the extended
5479 * attribute if the vp is exclusively locked.
5480 * This should happen when any of the following operations
5481 * occur on the vnode:
5482 * Close, Delegreturn, LayoutCommit, LayoutReturn
5483 * As such, the updated extended attribute should get saved
5484 * before nfsrv_checkdsattr() returns 0 and allows the cached
5485 * attributes to be returned without calling this function.
5487 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
5488 error = nfsrv_setextattr(vp, nap, p);
5489 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
5493 error = nd->nd_repstat;
5494 m_freem(nd->nd_mrep);
5496 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
5501 * Get the device id and file handle for a DS file.
5504 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
5505 fhandle_t *fhp, char *devid)
5511 buf = malloc(buflen, M_TEMP, M_WAITOK);
5512 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
5513 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
5519 * Do a Lookup against the DS for the filename.
5522 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
5523 struct vnode **nvpp, NFSPROC_T *p)
5525 struct nameidata named;
5526 struct ucred *tcred;
5532 tcred = newnfs_getcred();
5533 named.ni_cnd.cn_nameiop = LOOKUP;
5534 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
5535 named.ni_cnd.cn_cred = tcred;
5536 named.ni_cnd.cn_thread = p;
5537 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
5538 nfsvno_setpathbuf(&named, &bufp, &hashp);
5539 named.ni_cnd.cn_nameptr = bufp;
5540 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
5541 strlcpy(bufp, pf->dsf_filename, NAME_MAX);
5542 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
5543 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
5544 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
5546 nfsvno_relpathbuf(&named);
5549 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
5554 * Set the file handle to the correct one.
5557 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
5558 char *fnamep, struct vnode *nvp, NFSPROC_T *p)
5564 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
5566 * We can only do a vn_set_extattr() if the vnode is exclusively
5567 * locked and vn_start_write() has been done. If devid != NULL or
5568 * fnamep != NULL or the vnode is shared locked, vn_start_write()
5569 * may not have been done.
5570 * If not done now, it will be done on a future call.
5572 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
5574 ret = vn_extattr_set(vp, IO_NODELOCKED,
5575 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
5577 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
5581 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point
5582 * when the DS has failed.
5585 nfsrv_killrpcs(struct nfsmount *nmp)
5589 * Call newnfs_nmcancelreqs() to cause
5590 * any RPCs in progress on the mount point to
5592 * This will cause any process waiting for an
5593 * RPC to complete while holding a vnode lock
5594 * on the mounted-on vnode (such as "df" or
5595 * a non-forced "umount") to fail.
5596 * This will unlock the mounted-on vnode so
5597 * a forced dismount can succeed.
5598 * The NFSMNTP_CANCELRPCS flag should be set when this function is
5601 newnfs_nmcancelreqs(nmp);
5605 * Sum up the statfs info for each of the DSs, so that the client will
5606 * receive the total for all DSs.
5609 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
5612 struct nfsdevice *ds;
5613 struct vnode **dvpp, **tdvpp, *dvp;
5615 int cnt, error = 0, i;
5617 if (nfsrv_devidcnt <= 0)
5619 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
5620 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
5622 /* Get an array of the dvps for the DSs. */
5626 /* First, search for matches for same file system. */
5627 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5628 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
5629 ds->nfsdev_mdsfsid.val[0] == mp->mnt_stat.f_fsid.val[0] &&
5630 ds->nfsdev_mdsfsid.val[1] == mp->mnt_stat.f_fsid.val[1]) {
5631 if (++i > nfsrv_devidcnt)
5633 *tdvpp++ = ds->nfsdev_dvp;
5637 * If no matches for same file system, total all servers not assigned
5641 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5642 if (ds->nfsdev_nmp != NULL &&
5643 ds->nfsdev_mdsisset == 0) {
5644 if (++i > nfsrv_devidcnt)
5646 *tdvpp++ = ds->nfsdev_dvp;
5653 /* Do a VFS_STATFS() for each of the DSs and sum them up. */
5655 for (i = 0; i < cnt && error == 0; i++) {
5657 error = VFS_STATFS(dvp->v_mount, tsf);
5659 if (sf->f_bsize == 0) {
5660 if (tsf->f_bsize > 0)
5661 sf->f_bsize = tsf->f_bsize;
5665 if (tsf->f_blocks > 0) {
5666 if (sf->f_bsize != tsf->f_bsize) {
5667 tot = tsf->f_blocks * tsf->f_bsize;
5668 sf->f_blocks += (tot / sf->f_bsize);
5670 sf->f_blocks += tsf->f_blocks;
5672 if (tsf->f_bfree > 0) {
5673 if (sf->f_bsize != tsf->f_bsize) {
5674 tot = tsf->f_bfree * tsf->f_bsize;
5675 sf->f_bfree += (tot / sf->f_bsize);
5677 sf->f_bfree += tsf->f_bfree;
5679 if (tsf->f_bavail > 0) {
5680 if (sf->f_bsize != tsf->f_bsize) {
5681 tot = tsf->f_bavail * tsf->f_bsize;
5682 sf->f_bavail += (tot / sf->f_bsize);
5684 sf->f_bavail += tsf->f_bavail;
5697 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
5701 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
5702 error = NFSERR_ATTRNOTSUPP;
5706 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
5707 * Make sure it has enough room for that - splitting every entry
5708 * into two and appending "canonical six" entries at the end.
5709 * Cribbed out of kern/vfs_acl.c - Rick M.
5711 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
5712 error = NFSERR_ATTRNOTSUPP;
5715 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
5717 error = nfsrv_dssetacl(vp, aclp, cred, p);
5718 if (error == ENOENT)
5727 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
5730 * Called once to initialize data structures...
5733 nfsd_modevent(module_t mod, int type, void *data)
5736 static int loaded = 0;
5743 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
5744 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
5746 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
5749 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
5750 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
5751 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
5752 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
5753 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
5754 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
5761 #ifdef VV_DISABLEDELEG
5762 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
5763 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
5765 nfsd_call_servertimer = nfsrv_servertimer;
5766 nfsd_call_nfsd = nfssvc_nfsd;
5771 if (newnfs_numnfsd != 0) {
5776 #ifdef VV_DISABLEDELEG
5777 vn_deleg_ops.vndeleg_recall = NULL;
5778 vn_deleg_ops.vndeleg_disable = NULL;
5780 nfsd_call_servertimer = NULL;
5781 nfsd_call_nfsd = NULL;
5783 /* Clean out all NFSv4 state. */
5784 nfsrv_throwawayallstate(curthread);
5786 /* Clean the NFS server reply cache */
5787 nfsrvd_cleancache();
5789 /* Free up the krpc server pool. */
5790 if (nfsrvd_pool != NULL)
5791 svcpool_destroy(nfsrvd_pool);
5793 /* and get rid of the locks */
5794 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
5795 mtx_destroy(&nfsrchash_table[i].mtx);
5796 mtx_destroy(&nfsrcahash_table[i].mtx);
5798 mtx_destroy(&nfsrc_udpmtx);
5799 mtx_destroy(&nfs_v4root_mutex);
5800 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
5801 mtx_destroy(&nfsrv_dontlistlock_mtx);
5802 mtx_destroy(&nfsrv_recalllock_mtx);
5803 for (i = 0; i < nfsrv_sessionhashsize; i++)
5804 mtx_destroy(&nfssessionhash[i].mtx);
5805 if (nfslayouthash != NULL) {
5806 for (i = 0; i < nfsrv_layouthashsize; i++)
5807 mtx_destroy(&nfslayouthash[i].mtx);
5808 free(nfslayouthash, M_NFSDSESSION);
5810 lockdestroy(&nfsv4root_mnt.mnt_explock);
5811 free(nfsclienthash, M_NFSDCLIENT);
5812 free(nfslockhash, M_NFSDLOCKFILE);
5813 free(nfssessionhash, M_NFSDSESSION);
5825 static moduledata_t nfsd_mod = {
5830 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
5832 /* So that loader and kldload(2) can find us, wherever we are.. */
5833 MODULE_VERSION(nfsd, 1);
5834 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
5835 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
5836 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
5837 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
5838 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);