1 /* $Id: nfs_vnops.c,v 1.45 2003/11/05 14:59:02 rees Exp $ */
5 * the regents of the university of michigan
8 * permission is granted to use, copy, create derivative works and redistribute
9 * this software and such derivative works for any purpose, so long as the name
10 * of the university of michigan is not used in any advertising or publicity
11 * pertaining to the use or distribution of this software without specific,
12 * written prior authorization. if the above copyright notice or any other
13 * identification of the university of michigan is included in any copy of any
14 * portion of this software, then the disclaimer below must also be included.
16 * this software is provided as is, without representation from the university
17 * of michigan as to its fitness for any purpose, and without warranty by the
18 * university of michigan of any kind, either express or implied, including
19 * without limitation the implied warranties of merchantability and fitness for
20 * a particular purpose. the regents of the university of michigan shall not be
21 * liable for any damages, including special, indirect, incidental, or
22 * consequential damages, with respect to any claim arising out of or in
23 * connection with the use of the software, even if it has been or is hereafter
24 * advised of the possibility of such damages.
28 * Copyright (c) 1989, 1993
29 * The Regents of the University of California. All rights reserved.
31 * This code is derived from software contributed to Berkeley by
32 * Rick Macklem at The University of Guelph.
34 * Redistribution and use in source and binary forms, with or without
35 * modification, are permitted provided that the following conditions
37 * 1. Redistributions of source code must retain the above copyright
38 * notice, this list of conditions and the following disclaimer.
39 * 2. Redistributions in binary form must reproduce the above copyright
40 * notice, this list of conditions and the following disclaimer in the
41 * documentation and/or other materials provided with the distribution.
42 * 4. Neither the name of the University nor the names of its contributors
43 * may be used to endorse or promote products derived from this software
44 * without specific prior written permission.
46 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
47 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
48 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
49 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
50 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
51 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
52 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
53 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
54 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
55 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
65 * vnode op calls for Sun NFS version 2 and 3
70 #include <sys/param.h>
71 #include <sys/kernel.h>
72 #include <sys/systm.h>
73 #include <sys/resourcevar.h>
75 #include <sys/mount.h>
78 #include <sys/malloc.h>
80 #include <sys/namei.h>
81 #include <sys/socket.h>
82 #include <sys/vnode.h>
83 #include <sys/dirent.h>
84 #include <sys/fcntl.h>
85 #include <sys/lockf.h>
87 #include <sys/sysctl.h>
88 #include <sys/lockmgr.h>
89 #include <sys/signalvar.h>
92 #include <vm/vm_extern.h>
94 #include <fs/fifofs/fifo.h>
96 #include <rpc/rpcclnt.h>
98 #include <nfs/rpcv2.h>
99 #include <nfs/nfsproto.h>
100 #include <nfsclient/nfs.h>
101 #include <nfs4client/nfs4.h>
102 #include <nfsclient/nfsnode.h>
103 #include <nfsclient/nfsmount.h>
104 #include <nfsclient/nfs_lock.h>
105 #include <nfs/xdr_subs.h>
106 #include <nfsclient/nfsm_subs.h>
109 #include <netinet/in.h>
110 #include <netinet/in_var.h>
113 #include <nfs4client/nfs4m_subs.h>
114 #include <nfs4client/nfs4_vn.h>
121 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
122 * calls are not in getblk() and brelse() so that they would not be necessary
126 #define vfs_busy_pages(bp, f)
129 static int nfs4_flush(struct vnode *, int, struct thread *,
131 static int nfs4_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
132 static int nfs4_closerpc(struct vnode *, struct ucred *, int);
134 static vop_lookup_t nfs4_lookup;
135 static vop_create_t nfs4_create;
136 static vop_mknod_t nfs4_mknod;
137 static vop_open_t nfs4_open;
138 static vop_close_t nfs4_close;
139 static vop_access_t nfs4_access;
140 static vop_getattr_t nfs4_getattr;
141 static vop_setattr_t nfs4_setattr;
142 static vop_read_t nfs4_read;
143 static vop_fsync_t nfs4_fsync;
144 static vop_remove_t nfs4_remove;
145 static vop_link_t nfs4_link;
146 static vop_rename_t nfs4_rename;
147 static vop_mkdir_t nfs4_mkdir;
148 static vop_rmdir_t nfs4_rmdir;
149 static vop_symlink_t nfs4_symlink;
150 static vop_readdir_t nfs4_readdir;
151 static vop_strategy_t nfs4_strategy;
152 static int nfs4_lookitup(struct vnode *, const char *, int,
153 struct ucred *, struct thread *, struct nfsnode **);
154 static int nfs4_sillyrename(struct vnode *, struct vnode *,
155 struct componentname *);
156 static vop_readlink_t nfs4_readlink;
157 static vop_print_t nfs4_print;
158 static vop_advlock_t nfs4_advlock;
159 static vop_advlockasync_t nfs4_advlockasync;
162 * Global vfs data structures for nfs
164 struct vop_vector nfs4_vnodeops = {
165 .vop_default = &default_vnodeops,
166 .vop_access = nfs4_access,
167 .vop_advlock = nfs4_advlock,
168 .vop_advlockasync = nfs4_advlockasync,
169 .vop_close = nfs4_close,
170 .vop_create = nfs4_create,
171 .vop_fsync = nfs4_fsync,
172 .vop_getattr = nfs4_getattr,
173 .vop_getpages = nfs_getpages,
174 .vop_putpages = nfs_putpages,
175 .vop_inactive = nfs_inactive,
176 .vop_lease = VOP_NULL,
177 .vop_link = nfs4_link,
178 .vop_lookup = nfs4_lookup,
179 .vop_mkdir = nfs4_mkdir,
180 .vop_mknod = nfs4_mknod,
181 .vop_open = nfs4_open,
182 .vop_print = nfs4_print,
183 .vop_read = nfs4_read,
184 .vop_readdir = nfs4_readdir,
185 .vop_readlink = nfs4_readlink,
186 .vop_reclaim = nfs_reclaim,
187 .vop_remove = nfs4_remove,
188 .vop_rename = nfs4_rename,
189 .vop_rmdir = nfs4_rmdir,
190 .vop_setattr = nfs4_setattr,
191 .vop_strategy = nfs4_strategy,
192 .vop_symlink = nfs4_symlink,
193 .vop_write = nfs_write,
196 static int nfs4_removerpc(struct vnode *dvp, const char *name, int namelen,
197 struct ucred *cred, struct thread *td);
198 static int nfs4_renamerpc(struct vnode *fdvp, const char *fnameptr,
199 int fnamelen, struct vnode *tdvp,
200 const char *tnameptr, int tnamelen,
201 struct ucred *cred, struct thread *td);
202 static int nfs4_renameit(struct vnode *sdvp, struct componentname *scnp,
203 struct sillyrename *sp);
204 static int nfs4_openrpc(struct vnode *, struct vnode **,
205 struct componentname *, int, struct vattr *);
206 static int nfs4_open_confirm(struct vnode *vp, struct nfs4_compound *cpp,
207 struct nfs4_oparg_open *openap,
208 struct nfs4_oparg_getfh *gfh,
209 struct ucred *cred, struct thread *td);
210 static int nfs4_createrpc(struct vnode *, struct vnode **,
211 struct componentname *, nfstype,
212 struct vattr *, char *);
217 struct nfs4_lowner nfs4_masterlowner;
219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
221 SYSCTL_DECL(_vfs_nfs4);
223 static int nfs4_access_cache_timeout = NFS_MAXATTRTIMO;
224 SYSCTL_INT(_vfs_nfs4, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
225 &nfs4_access_cache_timeout, 0, "NFS ACCESS cache timeout");
228 static int nfsv3_commit_on_close = 0;
229 SYSCTL_INT(_vfs_nfs4, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
230 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
232 SYSCTL_INT(_vfs_nfs4, OID_AUTO, access_cache_hits, CTLFLAG_RD,
233 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
235 SYSCTL_INT(_vfs_nfs4, OID_AUTO, access_cache_misses, CTLFLAG_RD,
236 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
239 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
240 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
241 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
243 nfs4_v3_access_otw(struct vnode *vp, int wmode, struct thread *td,
248 int error = 0, attrflag;
252 struct mbuf *mreq, *mrep = NULL, *md, *mb;
255 struct nfsnode *np = VTONFS(vp);
257 nfsstats.rpccnt[NFSPROC_ACCESS]++;
258 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
260 bpos = mtod(mb, caddr_t);
262 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
263 *tl = txdr_unsigned(wmode);
264 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
265 nfsm_postop_attr(vp, attrflag);
267 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
268 rmode = fxdr_unsigned(u_int32_t, *tl);
270 np->n_modeuid = cred->cr_uid;
271 np->n_modestamp = time_second;
279 * nfs access vnode op.
280 * For nfs version 2, just return ok. File accesses may fail later.
281 * For nfs version 3, use the access rpc to check accessibility. If file modes
282 * are changed on the server, accesses might still fail later.
285 nfs4_access(struct vop_access_args *ap)
287 struct vnode *vp = ap->a_vp;
289 u_int32_t mode, wmode;
290 int v3 = NFS_ISV3(vp); /* v3 \in v4 */
291 struct nfsnode *np = VTONFS(vp);
293 struct mbuf *mreq, *mrep = NULL, *md, *mb;
294 struct nfs4_compound cp;
295 struct nfs4_oparg_access acc;
296 struct thread *td = ap->a_td;
297 struct ucred *cred = ap->a_cred;
300 * Disallow write attempts on filesystems mounted read-only;
301 * unless the file is a socket, fifo, or a block or character
302 * device resident on the filesystem.
304 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
305 switch (vp->v_type) {
315 * For nfs v3, check to see if we have done this recently, and if
316 * so return our cached result instead of making an ACCESS call.
317 * If not, do an access rpc, otherwise you are stuck emulating
318 * ufs_access() locally using the vattr. This may not be correct,
319 * since the server may apply other access criteria such as
320 * client uid-->server uid mapping that we do not know about.
322 /* XXX Disable this for now; needs fixing of _access_otw() */
324 if (ap->a_mode & VREAD)
325 mode = NFSV3ACCESS_READ;
328 if (vp->v_type != VDIR) {
329 if (ap->a_mode & VWRITE)
330 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
331 if (ap->a_mode & VEXEC)
332 mode |= NFSV3ACCESS_EXECUTE;
334 if (ap->a_mode & VWRITE)
335 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
337 if (ap->a_mode & VEXEC)
338 mode |= NFSV3ACCESS_LOOKUP;
340 /* XXX safety belt, only make blanket request if caching */
341 if (nfs4_access_cache_timeout > 0) {
342 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
343 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
344 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
350 * Does our cached result allow us to give a definite yes to
353 if (time_second < np->n_modestamp + nfs4_access_cache_timeout &&
354 ap->a_cred->cr_uid == np->n_modeuid &&
355 (np->n_mode & mode) == mode) {
356 nfsstats.accesscache_hits++;
359 * Either a no, or a don't know. Go to the wire.
361 nfsstats.accesscache_misses++;
362 error = nfs4_v3_access_otw(vp, wmode, ap->a_td,
365 if ((np->n_mode & mode) != mode)
372 /* XXX use generic access code here? */
373 mode = ap->a_mode & VREAD ? NFSV4ACCESS_READ : 0;
374 if (vp->v_type == VDIR) {
375 if (ap->a_mode & VWRITE)
376 mode |= NFSV4ACCESS_MODIFY | NFSV4ACCESS_EXTEND | NFSV4ACCESS_DELETE;
377 if (ap->a_mode & VEXEC)
378 mode |= NFSV4ACCESS_LOOKUP;
380 if (ap->a_mode & VWRITE)
381 mode |= NFSV4ACCESS_MODIFY | NFSV4ACCESS_EXTEND;
382 if (ap->a_mode & VEXEC)
383 mode |= NFSV4ACCESS_EXECUTE;
386 nfs_v4initcompound(&cp);
389 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
391 bpos = mtod(mb, caddr_t);
393 nfsm_v4build_compound(&cp, "nfs4_access()");
394 nfsm_v4build_putfh(&cp, vp);
395 nfsm_v4build_access(&cp, &acc);
396 nfsm_v4build_finalize(&cp);
398 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
402 nfsm_v4dissect_compound(&cp);
403 nfsm_v4dissect_putfh(&cp);
404 nfsm_v4dissect_access(&cp, &acc);
406 if ((acc.rmode & mode) != mode)
410 error = nfs_v4postop(&cp, error);
419 nfs4_openrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
420 int flags, struct vattr *vap)
422 struct vnode *vp = *vpp;
423 struct nfs4_oparg_getattr getattr;
424 struct nfs4_oparg_getfh getfh;
425 struct nfs4_oparg_open opena;
426 struct nfs4_compound cp;
429 struct mbuf *mreq, *mrep = NULL, *md, *mb;
430 struct ucred *cred = cnp->cn_cred;
431 struct thread *td = cnp->cn_thread;
432 struct nfs4_fctx xfc, *fcp;
436 /* Create a new file */
439 bzero(fcp, sizeof(*fcp));
442 fcp = flags & FWRITE ? &np->n_wfc : &np->n_rfc;
446 * Since we are currently only one lockowner; we only open the
447 * file once each for reading and writing.
449 if (fcp->refcnt++ != 0) {
451 /*printf("not opening %s\n", np->n_name != NULL ? np->n_name : "");*/
455 fcp->lop = &nfs4_masterlowner;
458 nfs_v4initcompound(&cp);
459 cp.nmp = VFSTONFS(dvp->v_mount);
461 opena.ctype = NCLNULL;
464 opena.fcp = fcp; /* For lockowner */
467 getattr.bm = &nfsv4_getattrbm;
469 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
471 bpos = mtod(mb, caddr_t);
473 nfsm_v4build_compound(&cp, "nfs4_openrpc()");
474 nfsm_v4build_putfh(&cp, dvp);
475 nfsm_v4build_open(&cp, &opena);
476 nfsm_v4build_getattr(&cp, &getattr);
477 nfsm_v4build_getfh(&cp, &getfh);
478 nfsm_v4build_finalize(&cp);
480 nfsm_request(vp != NULL ? vp : dvp, NFSV4PROC_COMPOUND, td, cred);
484 nfsm_v4dissect_compound(&cp);
485 nfsm_v4dissect_putfh(&cp);
486 nfsm_v4dissect_open(&cp, &opena);
487 nfsm_v4dissect_getattr(&cp, &getattr);
488 nfsm_v4dissect_getfh(&cp, &getfh);
490 error = nfs_v4postop(&cp, error);
492 if (opena.rflags & NFSV4OPENRES_CONFIRM) {
493 error = nfs4_open_confirm(vp ? vp : dvp, &cp, &opena, &getfh, cred, td);
500 error = nfs_nget(dvp->v_mount, &getfh.fh_val,
501 getfh.fh_len, &np, LK_EXCLUSIVE);
507 np->n_namelen = cnp->cn_namelen; /* XXX memory leaks on these; track! */
508 if (np->n_name != NULL)
509 FREE(np->n_name, M_NFSREQ);
510 MALLOC(np->n_name, u_char *, np->n_namelen + 1, M_NFSREQ, M_WAITOK);
511 bcopy(cnp->cn_nameptr, np->n_name, np->n_namelen);
512 np->n_name[np->n_namelen] = '\0';
518 /*printf("opened new file %s\n", np->n_name);*/
520 nfs4_vnop_loadattrcache(vp, &getattr.fa, NULL);
523 /*printf("openend \"old\" %s\n", np->n_name != NULL ? np->n_name : "");*/
525 if (flags & O_TRUNC && np->n_size != 0) {
530 error = nfs4_setattrrpc(vp, &va, cnp->cn_cred);
543 nfs4_open_confirm(struct vnode *vp, struct nfs4_compound *cpp,
544 struct nfs4_oparg_open *openap, struct nfs4_oparg_getfh *gfh,
545 struct ucred *cred, struct thread *td)
549 struct mbuf *mreq, *mrep = NULL, *md, *mb;
551 nfs_v4initcompound(cpp);
552 cpp->nmp = VFSTONFS(vp->v_mount);
554 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
556 bpos = mtod(mb, caddr_t);
558 nfsm_v4build_compound(cpp, "nfs4_open_confirm()");
559 nfsm_v4build_putfh_nv(cpp, gfh);
560 nfsm_v4build_open_confirm(cpp, openap);
561 nfsm_v4build_finalize(cpp);
563 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
567 nfsm_v4dissect_compound(cpp);
568 nfsm_v4dissect_putfh(cpp);
569 nfsm_v4dissect_open_confirm(cpp, openap);
572 error = nfs_v4postop(cpp, error);
583 * Check to see if the type is ok
584 * and that deletion is not in progress.
585 * For paged in text files, you will need to flush the page cache
586 * if consistency is lost.
590 nfs4_open(struct vop_open_args *ap)
592 struct vnode *vp = ap->a_vp;
593 struct nfsnode *np = VTONFS(vp);
594 enum vtype vtype = vp->v_type;
595 int mode = ap->a_mode;
596 struct componentname cn;
599 if (vtype != VDIR && vtype != VLNK) {
601 printf("open eacces vtyp=%d\n", vp->v_type);
608 if (np->n_flag & NCREATED) {
609 np->n_flag &= ~NCREATED;
613 cn.cn_nameptr = np->n_name;
614 cn.cn_namelen = np->n_namelen;
615 cn.cn_cred = ap->a_cred;
616 cn.cn_thread = ap->a_td;
618 return (nfs4_openrpc(np->n_dvp, &vp, &cn, mode, NULL));
622 nfs4_closerpc(struct vnode *vp, struct ucred *cred, int flags)
626 struct mbuf *mreq, *mrep = NULL, *md, *mb;
628 struct nfs4_fctx *fcp;
629 struct nfs4_compound cp;
630 struct nfsnode *np = VTONFS(vp);
633 fcp = flags & FWRITE ? &np->n_wfc : &np->n_rfc;
635 nfs_v4initcompound(&cp);
637 if (--fcp->refcnt != 0)
640 /*printf("closing %s\n", np->n_name != NULL ? np->n_name : "");*/
644 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
646 bpos = mtod(mb, caddr_t);
648 nfsm_v4build_compound(&cp, "nfs4_closerpc()");
649 nfsm_v4build_putfh(&cp, vp);
650 nfsm_v4build_close(&cp, fcp);
651 nfsm_v4build_finalize(&cp);
653 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
657 nfsm_v4dissect_compound(&cp);
658 nfsm_v4dissect_putfh(&cp);
659 nfsm_v4dissect_close(&cp, fcp);
662 error = nfs_v4postop(&cp, error);
672 * play it safe for now (see comments in v2/v3 nfs_close regarding dirty buffers)
676 nfs4_close(struct vop_close_args *ap)
678 struct vnode *vp = ap->a_vp;
679 struct nfsnode *np = VTONFS(vp);
682 if (vp->v_type != VREG)
685 if (np->n_flag & NMODIFIED) {
686 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
690 error = nfs4_closerpc(vp, ap->a_cred, ap->a_fflag);
692 if (!error && np->n_flag & NWRITEERR) {
693 np->n_flag &= ~NWRITEERR;
700 * nfs getattr call from vfs.
703 nfs4_getattr(struct vop_getattr_args *ap)
705 struct vnode *vp = ap->a_vp;
706 struct nfsnode *np = VTONFS(vp);
709 struct mbuf *mreq, *mrep = NULL, *md, *mb;
710 struct nfs4_oparg_getattr ga;
711 struct nfs4_compound cp;
714 * Update local times for special files.
716 if (np->n_flag & (NACC | NUPD))
719 * First look in the cache.
721 if (nfs_getattrcache(vp, ap->a_vap) == 0)
724 nfsstats.rpccnt[NFSPROC_GETATTR]++;
726 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, NFSX_FH(1));
728 bpos = mtod(mb, caddr_t);
730 ga.bm = &nfsv4_getattrbm;
731 nfs_v4initcompound(&cp);
733 nfsm_v4build_compound(&cp, "nfs4_getattr()");
734 nfsm_v4build_putfh(&cp, vp);
735 nfsm_v4build_getattr(&cp, &ga);
736 nfsm_v4build_finalize(&cp);
738 nfsm_request(vp, NFSV4PROC_COMPOUND, curthread, ap->a_cred);
742 nfsm_v4dissect_compound(&cp);
743 nfsm_v4dissect_putfh(&cp);
744 nfsm_v4dissect_getattr(&cp, &ga);
746 nfs4_vnop_loadattrcache(vp, &ga.fa, ap->a_vap);
749 error = nfs_v4postop(&cp, error);
760 nfs4_setattr(struct vop_setattr_args *ap)
762 struct vnode *vp = ap->a_vp;
763 struct nfsnode *np = VTONFS(vp);
764 struct vattr *vap = ap->a_vap;
765 struct thread *td = curthread;
774 * Setting of flags is not supported.
776 if (vap->va_flags != VNOVAL)
780 * Disallow write attempts if the filesystem is mounted read-only.
782 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
783 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
784 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
785 (vp->v_mount->mnt_flag & MNT_RDONLY))
787 if (vap->va_size != VNOVAL) {
788 switch (vp->v_type) {
795 if (vap->va_mtime.tv_sec == VNOVAL &&
796 vap->va_atime.tv_sec == VNOVAL &&
797 vap->va_mode == (mode_t)VNOVAL &&
798 vap->va_uid == (uid_t)VNOVAL &&
799 vap->va_gid == (gid_t)VNOVAL)
801 vap->va_size = VNOVAL;
805 * Disallow write attempts if the filesystem is
808 if (vp->v_mount->mnt_flag & MNT_RDONLY)
812 * We run vnode_pager_setsize() early (why?),
813 * we must set np->n_size now to avoid vinvalbuf
814 * V_SAVE races that might setsize a lower
819 error = nfs_meta_setsize(vp, ap->a_cred, td,
822 if (np->n_flag & NMODIFIED) {
823 if (vap->va_size == 0)
824 error = nfs_vinvalbuf(vp, 0, td, 1);
826 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
828 vnode_pager_setsize(vp, np->n_size);
833 * np->n_size has already been set to vap->va_size
834 * in nfs_meta_setsize(). We must set it again since
835 * nfs_loadattrcache() could be called through
836 * nfs_meta_setsize() and could modify np->n_size.
838 np->n_vattr.va_size = np->n_size = vap->va_size;
840 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
841 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
842 vp->v_type == VREG &&
843 (error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) == EINTR)
846 if (vap->va_size != VNOVAL && np->n_wfc.refcnt == 0) {
847 /* Have to open the file before we can truncate it */
848 struct componentname cn;
850 cn.cn_nameptr = np->n_name;
851 cn.cn_namelen = np->n_namelen;
852 cn.cn_cred = ap->a_cred;
854 error = nfs4_openrpc(np->n_dvp, &vp, &cn, FWRITE, NULL);
857 np->n_flag |= NTRUNCATE;
860 error = nfs4_setattrrpc(vp, vap, ap->a_cred);
861 if (error && vap->va_size != VNOVAL) {
862 np->n_size = np->n_vattr.va_size = tsize;
863 vnode_pager_setsize(vp, np->n_size);
869 * Do an nfs setattr rpc.
872 nfs4_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
876 struct mbuf *mreq, *mrep = NULL, *md, *mb;
878 struct nfs4_compound cp;
879 struct nfs4_oparg_getattr ga;
880 struct nfsnode *np = VTONFS(vp);
881 struct nfs4_fctx *fcp;
884 nfsstats.rpccnt[NFSPROC_SETATTR]++;
885 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
887 bpos = mtod(mb, caddr_t);
889 ga.bm = &nfsv4_getattrbm;
890 fcp = (vap->va_size != VNOVAL) ? &np->n_wfc : NULL;
891 nfs_v4initcompound(&cp);
893 nfsm_v4build_compound(&cp, "nfs4_setattrrpc");
894 nfsm_v4build_putfh(&cp, vp);
895 nfsm_v4build_setattr(&cp, vap, fcp);
896 nfsm_v4build_getattr(&cp, &ga);
897 nfsm_v4build_finalize(&cp);
899 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
903 nfsm_v4dissect_compound(&cp);
904 nfsm_v4dissect_putfh(&cp);
905 nfsm_v4dissect_setattr(&cp);
906 nfsm_v4dissect_getattr(&cp, &ga);
908 nfs4_vnop_loadattrcache(vp, &ga.fa, NULL);
910 /* TODO: do the settatr and close in a single compound rpc */
911 if (np->n_flag & NTRUNCATE) {
912 error = nfs4_closerpc(vp, cred, FWRITE);
913 np->n_flag &= ~NTRUNCATE;
917 error = nfs_v4postop(&cp, error);
926 * nfs lookup call, one step at a time...
927 * First look in cache
928 * If not found, unlock the directory nfsnode and do the rpc
931 nfs4_lookup(struct vop_lookup_args *ap)
933 struct componentname *cnp = ap->a_cnp;
934 struct vnode *dvp = ap->a_dvp;
935 struct vnode **vpp = ap->a_vpp;
936 int isdot, flags = cnp->cn_flags;
938 struct nfsmount *nmp;
940 struct mbuf *mreq, *mrep = NULL, *md, *mb;
944 int error = 0, fhsize;
945 struct thread *td = cnp->cn_thread;
946 struct nfs4_compound cp;
947 struct nfs4_oparg_getattr ga, dga;
948 struct nfs4_oparg_lookup l;
949 struct nfs4_oparg_getfh gfh;
952 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
953 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
955 if (dvp->v_type != VDIR)
957 nmp = VFSTONFS(dvp->v_mount);
960 isdot = cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.';
962 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
966 if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
970 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
971 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
972 nfsstats.lookupcache_hits++;
973 if (cnp->cn_nameiop != LOOKUP &&
975 cnp->cn_flags |= SAVENAME;
987 nfsstats.lookupcache_misses++;
988 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
990 len = cnp->cn_namelen;
991 mreq = nfsm_reqhead(NULL, NFSV4PROC_COMPOUND, 0);
993 bpos = mtod(mb, caddr_t);
995 ga.bm = &nfsv4_getattrbm;
996 dga.bm = &nfsv4_getattrbm;
997 nfs_v4initcompound(&cp);
999 nfsm_v4build_compound(&cp, "nfs4_lookup()");
1000 nfsm_v4build_putfh(&cp, dvp);
1001 nfsm_v4build_getattr(&cp, &dga);
1002 if (flags & ISDOTDOT)
1003 nfsm_v4build_lookupp(&cp);
1005 l.name = cnp->cn_nameptr;
1007 nfsm_v4build_lookup(&cp, &l);
1009 nfsm_v4build_getattr(&cp, &ga);
1010 nfsm_v4build_getfh(&cp, &gfh);
1011 nfsm_v4build_finalize(&cp);
1013 nfsm_request(dvp, NFSV4PROC_COMPOUND, cnp->cn_thread, cnp->cn_cred);
1017 nfsm_v4dissect_compound(&cp);
1018 nfsm_v4dissect_putfh(&cp);
1019 nfsm_v4dissect_getattr(&cp, &dga);
1020 if (flags & ISDOTDOT)
1021 nfsm_v4dissect_lookupp(&cp);
1023 nfsm_v4dissect_lookup(&cp);
1024 nfsm_v4dissect_getattr(&cp, &ga);
1025 nfsm_v4dissect_getfh(&cp, &gfh);
1027 nfs4_vnop_loadattrcache(dvp, &dga.fa, NULL);
1029 fhsize = gfh.fh_len;
1032 * Handle RENAME case...
1034 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1035 if (NFS_CMPFH(np, fhp, fhsize))
1038 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
1044 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1047 cnp->cn_flags |= SAVENAME;
1051 if (flags & ISDOTDOT) {
1054 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
1055 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1060 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1061 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1065 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
1070 /* Fill in np used by open. */
1072 np->n_namelen = cnp->cn_namelen;
1073 if (np->n_name != NULL)
1074 FREE(np->n_name, M_NFSREQ);
1075 MALLOC(np->n_name, u_char *, np->n_namelen + 1, M_NFSREQ, M_WAITOK);
1076 bcopy(cnp->cn_nameptr, np->n_name, np->n_namelen);
1077 np->n_name[np->n_namelen] = '\0';
1079 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1082 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1083 cnp->cn_flags |= SAVENAME;
1084 if ((cnp->cn_flags & MAKEENTRY) &&
1085 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1086 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
1087 cache_enter(dvp, newvp, cnp);
1092 error = nfs_v4postop(&cp, error);
1095 if (newvp != NULLVP) {
1099 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1100 (flags & ISLASTCN) && error == ENOENT) {
1101 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1104 error = EJUSTRETURN;
1106 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1107 cnp->cn_flags |= SAVENAME;
1115 * Just call nfs_bioread() to do the work.
1118 nfs4_read(struct vop_read_args *ap)
1120 struct vnode *vp = ap->a_vp;
1122 switch (vp->v_type) {
1124 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1128 return (EOPNOTSUPP);
1136 nfs4_readlink(struct vop_readlink_args *ap)
1138 struct vnode *vp = ap->a_vp;
1140 if (vp->v_type != VLNK)
1142 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1146 * Do a readlink rpc.
1147 * Called by nfs_doio() from below the buffer cache.
1150 nfs4_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1154 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1155 struct nfs4_compound cp;
1157 nfsstats.rpccnt[NFSPROC_READLINK]++;
1159 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1161 bpos = mtod(mb, caddr_t);
1163 nfs_v4initcompound(&cp);
1165 nfsm_v4build_compound(&cp, "nfs4_readlinkrpc()");
1166 nfsm_v4build_putfh(&cp, vp);
1167 nfsm_v4build_readlink(&cp);
1168 nfsm_v4build_finalize(&cp);
1170 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
1174 nfsm_v4dissect_compound(&cp);
1175 nfsm_v4dissect_putfh(&cp);
1176 nfsm_v4dissect_readlink(&cp, uiop);
1179 error = nfs_v4postop(&cp, error);
1191 nfs4_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1194 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1195 struct nfsmount *nmp;
1196 int error = 0, len, tsiz;
1197 struct nfs4_compound cp;
1198 struct nfs4_oparg_read read;
1199 struct nfsnode *np = VTONFS(vp);
1201 nmp = VFSTONFS(vp->v_mount);
1202 tsiz = uiop->uio_resid;
1203 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1210 read.fcp = np->n_rfc.refcnt > 0 ? &np->n_rfc : &np->n_wfc;
1213 nfsstats.rpccnt[NFSPROC_READ]++;
1214 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1216 read.off = uiop->uio_offset;
1218 nfs_v4initcompound(&cp);
1220 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1222 bpos = mtod(mb, caddr_t);
1224 nfsm_v4build_compound(&cp, "nfs4_readrpc()");
1225 nfsm_v4build_putfh(&cp, vp);
1226 nfsm_v4build_read(&cp, &read);
1227 nfsm_v4build_finalize(&cp);
1229 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
1231 error = nfs_v4postop(&cp, error);
1235 nfsm_v4dissect_compound(&cp);
1236 nfsm_v4dissect_putfh(&cp);
1237 nfsm_v4dissect_read(&cp, &read);
1239 if (read.eof || read.retlen == 0)
1242 tsiz -= read.retlen;
1244 error = nfs_v4postop(&cp, error);
1260 nfs4_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1261 int *iomode, int *must_commit)
1265 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1266 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1267 int error = 0, len, tsiz, wccflag = 1, rlen;
1268 struct nfs4_compound cp;
1269 struct nfs4_oparg_write write;
1270 nfsv4stablehow commit, committed = NSHFILESYNC;
1272 struct nfsnode *np = VTONFS(vp);
1275 if (uiop->uio_iovcnt != 1)
1276 panic("nfs: writerpc iovcnt > 1");
1279 tsiz = uiop->uio_resid;
1280 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1286 write.stable = (nfsv4stablehow)*iomode;
1288 write.fcp = &np->n_wfc;
1291 nfsstats.rpccnt[NFSPROC_WRITE]++;
1292 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1294 write.off = uiop->uio_offset;
1296 nfs_v4initcompound(&cp);
1298 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1300 bpos = mtod(mb, caddr_t);
1302 nfsm_v4build_compound(&cp, "nfs4_writerpc()");
1303 nfsm_v4build_putfh(&cp, vp);
1304 nfsm_v4build_write(&cp, &write);
1305 nfsm_v4build_finalize(&cp);
1307 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
1309 error = nfs_v4postop(&cp, error);
1313 nfsm_v4dissect_compound(&cp);
1314 nfsm_v4dissect_putfh(&cp);
1315 nfsm_v4dissect_write(&cp, &write);
1317 rlen = write.retlen;
1321 } else if (rlen < len) {
1322 backup = len - rlen;
1323 uiop->uio_iov->iov_base =
1324 (char *)uiop->uio_iov->iov_base - backup;
1325 uiop->uio_iov->iov_len += backup;
1326 uiop->uio_offset -= backup;
1327 uiop->uio_resid += backup;
1331 commit = write.committed;
1333 if (committed == NSHFILESYNC ||
1334 (committed = NSHDATASYNC && commit == NSHUNSTABLE))
1337 verf = (caddr_t)write.wverf;
1339 if ((nmp->nm_flag & NFSSTA_HASWRITEVERF) == 0) {
1340 bcopy(verf, nmp->nm_verf, NFSX_V4VERF);
1341 nmp->nm_flag |= NFSMNT_HASWRITEVERF;
1342 } else if (bcmp(verf, nmp->nm_verf, NFSX_V4VERF)) {
1344 bcopy(verf, nmp->nm_verf, NFSX_V4VERF);
1349 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1351 error = nfs_v4postop(&cp, error);
1362 *iomode = committed;
1364 uiop->uio_resid = tsiz;
1370 nfs4_mknod(struct vop_mknod_args *ap)
1372 struct vattr *vap = ap->a_vap;
1373 struct vnode *newvp = NULL;
1376 error = nfs4_createrpc(ap->a_dvp, &newvp,
1377 ap->a_cnp, (nfstype)vap->va_type, vap, NULL);
1379 /* XXX - is this actually referenced here? */
1389 nfs4_createrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1390 nfstype ftype, struct vattr *vap, char *linktarget)
1392 struct nfsnode *dnp = VTONFS(dvp);
1393 struct nfsnode *np = NULL;
1394 struct vnode *newvp = NULL;
1395 struct nfs4_compound cp;
1396 struct nfs4_oparg_create c;
1397 struct nfs4_oparg_getattr ga;
1398 struct nfs4_oparg_getfh gfh;
1400 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1403 nfsstats.rpccnt[NFSPROC_CREATE]++;
1405 mreq = nfsm_reqhead(dvp, NFSV4PROC_COMPOUND, 0);
1407 bpos = mtod(mb, caddr_t);
1409 bzero(&c, sizeof(c));
1410 bzero(&ga, sizeof(ga));
1414 c.linktext = linktarget;
1415 c.name = cnp->cn_nameptr;
1416 c.namelen = cnp->cn_namelen;
1418 ga.bm = &nfsv4_getattrbm;
1419 nfs_v4initcompound(&cp);
1421 nfsm_v4build_compound(&cp, "nfs4_createrpc()");
1422 nfsm_v4build_putfh(&cp, dvp);
1423 nfsm_v4build_create(&cp, &c);
1424 nfsm_v4build_getattr(&cp, &ga);
1425 nfsm_v4build_getfh(&cp, &gfh);
1426 nfsm_v4build_finalize(&cp);
1428 nfsm_request(dvp, NFSV4PROC_COMPOUND, cnp->cn_thread, cnp->cn_cred);
1432 nfsm_v4dissect_compound(&cp);
1433 nfsm_v4dissect_putfh(&cp);
1434 nfsm_v4dissect_create(&cp, &c);
1435 nfsm_v4dissect_getattr(&cp, &ga);
1436 nfsm_v4dissect_getfh(&cp, &gfh);
1438 error = nfs_nget(dvp->v_mount, &gfh.fh_val, gfh.fh_len, &np, LK_EXCLUSIVE);
1443 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1445 if (cnp->cn_flags & MAKEENTRY)
1446 cache_enter(dvp, newvp, cnp);
1448 dnp->n_flag |= NMODIFIED;
1449 dnp->n_attrstamp = 0;
1452 error = nfs_v4postop(&cp, error);
1458 /*FREE(cnp->cn_pnbuf, M_NAMEI);*/
1459 if (error != 0 && newvp != NULL)
1461 else if (error == 0)
1468 nfs4_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1469 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1470 struct ucred *cred, struct thread *td)
1473 struct nfsnode *fnp = VTONFS(fdvp), *tnp = VTONFS(tdvp);
1475 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1476 struct nfs4_compound cp;
1477 struct nfs4_oparg_rename r;
1480 nfsstats.rpccnt[NFSPROC_RENAME]++;
1483 r.fnamelen = fnamelen;
1485 r.tnamelen = tnamelen;
1486 nfs_v4initcompound(&cp);
1488 mreq = nfsm_reqhead(fdvp, NFSV4PROC_COMPOUND, 0);
1490 bpos = mtod(mb, caddr_t);
1492 nfsm_v4build_compound(&cp, "nfs4_renamerpc()");
1493 nfsm_v4build_putfh(&cp, fdvp);
1494 nfsm_v4build_savefh(&cp);
1495 nfsm_v4build_putfh(&cp, tdvp);
1496 nfsm_v4build_rename(&cp, &r);
1497 nfsm_v4build_finalize(&cp);
1499 nfsm_request(fdvp, NFSV4PROC_COMPOUND, td, cred);
1503 nfsm_v4dissect_compound(&cp);
1504 nfsm_v4dissect_putfh(&cp);
1505 nfsm_v4dissect_savefh(&cp);
1506 nfsm_v4dissect_putfh(&cp);
1507 nfsm_v4dissect_rename(&cp);
1509 /* XXX should this always be performed? */
1510 fnp->n_flag |= NMODIFIED;
1511 tnp->n_flag |= NMODIFIED;
1512 fnp->n_attrstamp = tnp->n_attrstamp = 0;
1515 error = nfs_v4postop(&cp, error);
1524 * nfs file create call
1527 nfs4_create(struct vop_create_args *ap)
1529 struct vnode *dvp = ap->a_dvp;
1530 struct vattr *vap = ap->a_vap;
1531 struct nfsnode *dnp = VTONFS(dvp);
1532 struct componentname *cnp = ap->a_cnp;
1533 struct vnode *newvp = NULL;
1534 int error = 0, fmode = (O_CREAT | FREAD | FWRITE);
1537 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1540 if (vap->va_vaflags & VA_EXCLUSIVE)
1543 error = nfs4_openrpc(dvp, &newvp, cnp, fmode, vap);
1547 VTONFS(newvp)->n_flag |= NCREATED;
1549 if (cnp->cn_flags & MAKEENTRY)
1550 cache_enter(dvp, newvp, cnp);
1554 dnp->n_flag |= NMODIFIED;
1555 dnp->n_attrstamp = 0; /* XXX; wccflag */
1562 * nfs file remove call
1563 * To try and make nfs semantics closer to ufs semantics, a file that has
1564 * other processes using the vnode is renamed instead of removed and then
1565 * removed later on the last close.
1566 * - If v_usecount > 1
1567 * If a rename is not already in the works
1568 * call nfs4_sillyrename() to set it up
1573 nfs4_remove(struct vop_remove_args *ap)
1575 struct vnode *vp = ap->a_vp;
1576 struct vnode *dvp = ap->a_dvp;
1577 struct componentname *cnp = ap->a_cnp;
1578 struct nfsnode *np = VTONFS(vp);
1583 if ((cnp->cn_flags & HASBUF) == 0)
1584 panic("nfs4_remove: no name");
1585 if (vrefcnt(vp) < 1)
1586 panic("nfs4_remove: bad v_usecount");
1588 if (vp->v_type == VDIR)
1590 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1591 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1593 * Purge the name cache so that the chance of a lookup for
1594 * the name succeeding while the remove is in progress is
1595 * minimized. Without node locking it can still happen, such
1596 * that an I/O op returns ESTALE, but since you get this if
1597 * another host removes the file..
1601 * throw away biocache buffers, mainly to avoid
1602 * unnecessary delayed writes later.
1604 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1607 error = nfs4_removerpc(dvp, cnp->cn_nameptr,
1608 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1610 * Kludge City: If the first reply to the remove rpc is lost..
1611 * the reply to the retransmitted request will be ENOENT
1612 * since the file was in fact removed
1613 * Therefore, we cheat and return success.
1615 if (error == ENOENT)
1617 } else if (!np->n_sillyrename)
1618 error = nfs4_sillyrename(dvp, vp, cnp);
1619 np->n_attrstamp = 0;
1624 * nfs file remove rpc called from nfs_inactive
1627 nfs4_removeit(struct sillyrename *sp)
1630 * Make sure that the directory vnode is still valid.
1631 * XXX we should lock sp->s_dvp here.
1633 if (sp->s_dvp->v_type == VBAD)
1635 return (nfs4_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1640 * Nfs remove rpc, called from nfs4_remove() and nfs4_removeit().
1643 nfs4_removerpc(struct vnode *dvp, const char *name, int namelen,
1644 struct ucred *cred, struct thread *td)
1648 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1649 struct nfs4_compound cp;
1651 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1653 mreq = nfsm_reqhead(dvp, NFSV4PROC_COMPOUND, 0);
1655 bpos = mtod(mb, caddr_t);
1657 nfs_v4initcompound(&cp);
1659 nfsm_v4build_compound(&cp, "nfs4_removerpc()");
1660 nfsm_v4build_putfh(&cp, dvp);
1661 nfsm_v4build_remove(&cp, name, namelen);
1662 nfsm_v4build_finalize(&cp);
1664 nfsm_request(dvp, NFSV4PROC_COMPOUND, td, cred);
1668 nfsm_v4dissect_compound(&cp);
1669 nfsm_v4dissect_putfh(&cp);
1670 nfsm_v4dissect_remove(&cp);
1673 error = nfs_v4postop(&cp, error);
1678 VTONFS(dvp)->n_flag |= NMODIFIED;
1679 VTONFS(dvp)->n_attrstamp = 0; /* XXX wccflag */
1685 * nfs file rename call
1688 nfs4_rename(struct vop_rename_args *ap)
1690 struct vnode *fvp = ap->a_fvp;
1691 struct vnode *tvp = ap->a_tvp;
1692 struct vnode *fdvp = ap->a_fdvp;
1693 struct vnode *tdvp = ap->a_tdvp;
1694 struct componentname *tcnp = ap->a_tcnp;
1695 struct componentname *fcnp = ap->a_fcnp;
1699 if ((tcnp->cn_flags & HASBUF) == 0 ||
1700 (fcnp->cn_flags & HASBUF) == 0)
1701 panic("nfs4_rename: no name");
1703 /* Check for cross-device rename */
1704 if ((fvp->v_mount != tdvp->v_mount) ||
1705 (tvp && (fvp->v_mount != tvp->v_mount))) {
1711 printf("nfs4_rename: fvp == tvp (can't happen)\n");
1715 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1719 * We have to flush B_DELWRI data prior to renaming
1720 * the file. If we don't, the delayed-write buffers
1721 * can be flushed out later after the file has gone stale
1722 * under NFSV3. NFSV2 does not have this problem because
1723 * ( as far as I can tell ) it flushes dirty buffers more
1726 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1729 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1732 * If the tvp exists and is in use, sillyrename it before doing the
1733 * rename of the new file over it.
1734 * XXX Can't sillyrename a directory.
1736 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1737 tvp->v_type != VDIR && !nfs4_sillyrename(tdvp, tvp, tcnp)) {
1742 error = nfs4_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1743 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1746 if (fvp->v_type == VDIR) {
1747 if (tvp != NULL && tvp->v_type == VDIR)
1762 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1764 if (error == ENOENT)
1770 * nfs file rename rpc called from nfs4_remove() above
1773 nfs4_renameit(struct vnode *sdvp, struct componentname *scnp,
1774 struct sillyrename *sp)
1776 return (nfs4_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1777 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1781 * nfs hard link create call
1784 nfs4_link(struct vop_link_args *ap)
1786 struct vnode *vp = ap->a_vp;
1787 struct vnode *tdvp = ap->a_tdvp;
1788 struct componentname *cnp = ap->a_cnp;
1791 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1792 struct nfs4_compound cp;
1793 struct nfs4_oparg_link l;
1795 if (vp->v_mount != tdvp->v_mount) {
1800 * Push all writes to the server, so that the attribute cache
1801 * doesn't get "out of sync" with the server.
1802 * XXX There should be a better way!
1804 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1806 nfsstats.rpccnt[NFSPROC_LINK]++;
1808 l.name = cnp->cn_nameptr;
1809 l.namelen = cnp->cn_namelen;
1810 nfs_v4initcompound(&cp);
1812 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1814 bpos = mtod(mb, caddr_t);
1816 nfsm_v4build_compound(&cp, "nfs4_link()");
1817 nfsm_v4build_putfh(&cp, vp);
1818 nfsm_v4build_savefh(&cp);
1819 nfsm_v4build_putfh(&cp, tdvp);
1820 nfsm_v4build_link(&cp, &l);
1821 nfsm_v4build_finalize(&cp);
1823 nfsm_request(vp, NFSV4PROC_COMPOUND, cnp->cn_thread, cnp->cn_cred);
1827 nfsm_v4dissect_compound(&cp);
1828 nfsm_v4dissect_putfh(&cp);
1829 nfsm_v4dissect_savefh(&cp);
1830 nfsm_v4dissect_putfh(&cp);
1831 nfsm_v4dissect_link(&cp);
1833 VTONFS(tdvp)->n_flag |= NMODIFIED;
1834 VTONFS(vp)->n_attrstamp = 0;
1835 VTONFS(tdvp)->n_attrstamp = 0;
1838 error = nfs_v4postop(&cp, error);
1847 * nfs symbolic link create call
1850 nfs4_symlink(struct vop_symlink_args *ap)
1852 struct vnode *dvp = ap->a_dvp;
1854 struct vnode *newvp = NULL;
1856 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1858 error = nfs4_createrpc(ap->a_dvp, &newvp, ap->a_cnp, NFLNK,
1859 ap->a_vap, ap->a_target);
1861 if (error != 0 && newvp != NULL)
1863 else if (error == 0)
1866 VTONFS(dvp)->n_flag |= NMODIFIED;
1867 VTONFS(dvp)->n_attrstamp = 0; /* XXX wccflags */
1876 nfs4_mkdir(struct vop_mkdir_args *ap)
1878 return (nfs4_createrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, NFDIR,
1883 * nfs remove directory call
1886 nfs4_rmdir(struct vop_rmdir_args *ap)
1888 struct vnode *vp = ap->a_vp;
1889 struct vnode *dvp = ap->a_dvp;
1890 struct nfsnode *dnp = VTONFS(dvp);
1891 struct componentname *cnp = ap->a_cnp;
1897 error = (nfs4_removerpc(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred,
1902 dnp->n_flag |= NMODIFIED;
1903 dnp->n_attrstamp = 0;
1914 nfs4_readdir(struct vop_readdir_args *ap)
1916 struct vnode *vp = ap->a_vp;
1917 struct nfsnode *np = VTONFS(vp);
1918 struct uio *uio = ap->a_uio;
1922 if (vp->v_type != VDIR)
1925 * First, check for hit on the EOF offset cache
1927 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1928 (np->n_flag & NMODIFIED) == 0) {
1929 if (!VOP_GETATTR(vp, &vattr, ap->a_cred) &&
1930 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
1931 nfsstats.direofcache_hits++;
1937 * Call nfs_bioread() to do the real work.
1939 tresid = uio->uio_resid;
1940 error = nfs_bioread(vp, uio, 0, ap->a_cred);
1942 if (!error && uio->uio_resid == tresid)
1943 nfsstats.direofcache_misses++;
1947 static u_char fty_to_dty[] = {
1948 DT_UNKNOWN, /* NFNON */
1954 DT_SOCK, /* NFSOCK */
1955 DT_FIFO, /* NFFIFO */
1956 DT_UNKNOWN, /* NFATTRDIT */
1957 DT_UNKNOWN, /* NFNAMEDATTR */
1958 DT_UNKNOWN, /* NFBAD */
1963 * Called from below the buffer cache by nfs_doio().
1966 nfs4_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1969 struct dirent *dp = NULL;
1974 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1976 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1977 struct nfsnode *dnp = VTONFS(vp);
1978 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
1979 struct nfs4_compound cp;
1980 struct nfs4_oparg_readdir readdir;
1981 struct nfsv4_fattr fattr;
1985 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
1986 (uiop->uio_resid & (DIRBLKSIZ - 1)))
1987 panic("nfs readdirrpc bad uio");
1991 * If there is no cookie, assume directory was stale.
1993 cookiep = nfs4_getcookie(dnp, uiop->uio_offset, 0);
1997 return (NFSERR_BAD_COOKIE);
1999 /* Generate fake entries for "." and ".." */
2000 while (cookie < 2 && bigenough) {
2004 if (len > uiop->uio_resid) {
2008 dp = (struct dirent *)uiop->uio_iov->iov_base;
2010 dp->d_namlen = cookie;
2012 dp->d_type = DT_DIR;
2014 dp->d_fileno = dnp->n_vattr.va_fileid; /* XXX has problems with pynfs virtualhandles */
2016 dp->d_fileno = dnp->n_dvp != NULL ?
2017 VTONFS(dnp->n_dvp)->n_vattr.va_fileid : cookie;
2026 if (blksiz == DIRBLKSIZ)
2028 uiop->uio_offset += len;
2029 uiop->uio_resid -= len;
2030 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + len;
2031 uiop->uio_iov->iov_len -= len;
2037 /* This is sort of ugly, to prevent v4postop() from acting weird */
2038 bzero(&cp, sizeof(cp));
2041 * Loop around doing readdir rpc's of size nm_readdirsize
2042 * truncated to a multiple of DIRBLKSIZ.
2043 * The stopping criteria is EOF or buffer full.
2046 * XXX this is sort of ugly for nfsv4; we don't maintain the
2047 * strict abstraction, but do the decoding inline. that's ok.
2049 while (more_dirs && bigenough) {
2050 nfsstats.rpccnt[NFSPROC_READDIR]++;
2052 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
2054 bpos = mtod(mb, caddr_t);
2056 readdir.cnt = nmp->nm_readdirsize;
2057 readdir.cookie = cookie;
2058 readdir.bm = &nfsv4_readdirbm;
2060 bzero(&readdir.verf, sizeof(readdir.verf));
2062 bcopy(&dnp->n_cookieverf, &readdir.verf,
2063 sizeof(readdir.verf));
2065 nfs_v4initcompound(&cp);
2067 nfsm_v4build_compound(&cp, "nfs4_readdirrpc()");
2068 nfsm_v4build_putfh(&cp, vp);
2069 nfsm_v4build_readdir(&cp, &readdir);
2070 nfsm_v4build_finalize(&cp);
2072 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
2076 nfsm_v4dissect_compound(&cp);
2077 nfsm_v4dissect_putfh(&cp);
2080 * XXX - Readdir gets handled inline like in
2081 * NFSv{2,3}. This is a nasty inconsistency and
2085 tl = nfsm_dissect(uint32_t *, 5 * NFSX_UNSIGNED);
2086 if (fxdr_unsigned(uint32_t, *tl++) != NFSV4OP_READDIR) {
2090 if (fxdr_unsigned(uint32_t, *tl++) != 0) {
2095 bcopy(tl, &dnp->n_cookieverf, NFSX_V4VERF);
2097 more_dirs = fxdr_unsigned(int, *tl++);
2099 /* loop thru the dir entries, doctoring them to 4bsd form */
2100 while (more_dirs && bigenough) {
2101 tl = nfsm_dissect(uint32_t *, 3 * NFSX_UNSIGNED);
2102 cookie = fxdr_hyper(tl);
2104 /* XXX cookie sanity check */
2105 len = fxdr_unsigned(int, *tl++);
2106 if (len <= 0 || len > NFS_MAXNAMLEN) {
2110 tlen = nfsm_rndup(len);
2112 tlen += 4; /* To ensure null termination */
2113 left = DIRBLKSIZ - blksiz;
2114 if ((tlen + DIRHDSIZ) > left) {
2115 dp->d_reclen += left;
2116 uiop->uio_iov->iov_base =
2117 (char *)uiop->uio_iov->iov_base + left;
2118 uiop->uio_iov->iov_len -= left;
2119 uiop->uio_offset += left;
2120 uiop->uio_resid -= left;
2123 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2126 dp = (struct dirent *)uiop->uio_iov->iov_base;
2129 dp->d_reclen = tlen + DIRHDSIZ;
2131 blksiz += dp->d_reclen;
2132 if (blksiz == DIRBLKSIZ)
2134 uiop->uio_offset += DIRHDSIZ;
2135 uiop->uio_resid -= DIRHDSIZ;
2136 uiop->uio_iov->iov_base =
2137 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2138 uiop->uio_iov->iov_len -= DIRHDSIZ;
2141 nfsm_mtouio(uiop, len);
2142 p = uiop->uio_iov->iov_base;
2144 *p = '\0'; /* null terminate */
2145 /* printf("nfs4_readdirrpc: name: \"%s\" cookie %d\n",
2146 p - len, (int) cookie);*/
2147 uiop->uio_iov->iov_base =
2148 (char *)uiop->uio_iov->iov_base + tlen;
2149 uiop->uio_iov->iov_len -= tlen;
2150 uiop->uio_offset += tlen;
2151 uiop->uio_resid -= tlen;
2153 /* Copy attributes */
2154 nfsm_v4dissect_attrs(&fattr);
2156 dp->d_fileno = nfs_v4fileid4_to_fileid(
2157 fattr.fa4_valid & FA4V_FILEID &&
2159 fattr.fa4_fileid : cookie);
2161 fty = (u_int)fattr.fa4_type;
2162 dp->d_type = fattr.fa4_valid & FA4V_TYPE &&
2163 (fty < sizeof(fty_to_dty)) ?
2164 fty_to_dty[fty] : DT_UNKNOWN;
2166 nfsm_adv(nfsm_rndup(len));
2168 tl = nfsm_dissect(uint32_t *, NFSX_UNSIGNED);
2169 more_dirs = fxdr_unsigned(int, *tl++);
2172 * If at end of rpc data, get the eof boolean
2175 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2176 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2179 error = nfs_v4postop(&cp, error);
2185 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2186 * by increasing d_reclen for the last record.
2189 left = DIRBLKSIZ - blksiz;
2190 dp->d_reclen += left;
2191 uiop->uio_iov->iov_base =
2192 (char *)uiop->uio_iov->iov_base + left;
2193 uiop->uio_iov->iov_len -= left;
2194 uiop->uio_offset += left;
2195 uiop->uio_resid -= left;
2199 * We are now either at the end of the directory or have filled the
2203 dnp->n_direofoffset = uiop->uio_offset;
2205 if (uiop->uio_resid > 0)
2206 printf("EEK! readdirrpc resid > 0\n");
2207 cookiep = nfs4_getcookie(dnp, uiop->uio_offset, 1);
2217 * Silly rename. To make the NFS filesystem that is stateless look a little
2218 * more like the "ufs" a remove of an active vnode is translated to a rename
2219 * to a funny looking filename that is removed by nfs_inactive on the
2220 * nfsnode. There is the potential for another process on a different client
2221 * to create the same funny name between the nfs_lookitup() fails and the
2222 * nfs_rename() completes, but...
2225 nfs4_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2227 struct sillyrename *sp;
2235 if (vp->v_type == VDIR)
2236 panic("nfs: sillyrename dir");
2238 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2239 M_NFSREQ, M_WAITOK);
2240 sp->s_cred = crhold(cnp->cn_cred);
2242 sp->s_removeit = nfs4_removeit;
2245 /* Fudge together a funny name */
2246 pid = cnp->cn_thread->td_proc->p_pid;
2247 sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
2249 /* Try lookitups until we get one that isn't there */
2250 while (nfs4_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2251 cnp->cn_thread, NULL) == 0) {
2253 if (sp->s_name[4] > 'z') {
2258 error = nfs4_renameit(dvp, cnp, sp);
2261 error = nfs4_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2262 cnp->cn_thread, &np);
2263 np->n_sillyrename = sp;
2268 free((caddr_t)sp, M_NFSREQ);
2273 * Look up a file name and optionally either update the file handle or
2274 * allocate an nfsnode, depending on the value of npp.
2275 * npp == NULL --> just do the lookup
2276 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2278 * *npp != NULL --> update the file handle in the vnode
2281 nfs4_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2282 struct thread *td, struct nfsnode **npp)
2284 struct vnode *newvp = NULL;
2285 struct nfsnode *np, *dnp = VTONFS(dvp);
2287 int error = 0, fhlen;
2288 struct mbuf *mreq, *mrep = NULL, *md, *mb;
2290 struct nfs4_compound cp;
2291 struct nfs4_oparg_lookup l;
2292 struct nfs4_oparg_getfh gfh;
2293 struct nfs4_oparg_getattr ga;
2295 nfsstats.rpccnt[NFSPROC_RENAME]++;
2297 mreq = nfsm_reqhead(dvp, NFSV4PROC_COMPOUND, 0);
2299 bpos = mtod(mb, caddr_t);
2304 nfs_v4initcompound(&cp);
2306 ga.bm = &nfsv4_getattrbm;
2308 nfsm_v4build_compound(&cp, "nfs4_renamerpc()");
2309 nfsm_v4build_putfh(&cp, dvp);
2310 nfsm_v4build_lookup(&cp, &l);
2311 nfsm_v4build_getfh(&cp, &gfh);
2312 nfsm_v4build_getattr(&cp, &ga);
2314 nfsm_request(dvp, NFSV4PROC_COMPOUND, td, cred);
2318 nfsm_v4dissect_compound(&cp);
2319 nfsm_v4dissect_putfh(&cp);
2320 nfsm_v4dissect_lookup(&cp);
2321 nfsm_v4dissect_getfh(&cp, &gfh);
2322 nfsm_v4dissect_getattr(&cp, &ga);
2324 if (npp != NULL && error == 0) {
2330 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2331 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2332 np->n_fhp = &np->n_fh;
2333 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2334 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2335 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2336 np->n_fhsize = fhlen;
2338 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2342 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2352 np->n_namelen = len;
2353 if (np->n_name != NULL)
2354 FREE(np->n_name, M_NFSREQ);
2355 MALLOC(np->n_name, u_char *,
2356 np->n_namelen + 1, M_NFSREQ, M_WAITOK);
2357 memcpy(np->n_name, name, len);
2358 np->n_name[len] = '\0';
2360 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
2364 error = nfs_v4postop(&cp, error);
2368 if (npp && *npp == NULL) {
2385 * Nfs Version 3 commit rpc
2388 nfs4_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2391 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2394 struct mbuf *mreq, *mrep = NULL, *md, *mb;
2395 struct nfs4_compound cp;
2396 struct nfs4_oparg_commit commit;
2398 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2400 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2402 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
2404 bpos = mtod(mb, caddr_t);
2406 commit.start = offset;
2409 nfs_v4initcompound(&cp);
2411 nfsm_v4build_compound(&cp, "nfs4_commit()");
2412 nfsm_v4build_putfh(&cp, vp);
2413 nfsm_v4build_commit(&cp, &commit);
2414 nfsm_v4build_finalize(&cp);
2416 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
2420 nfsm_v4dissect_compound(&cp);
2421 nfsm_v4dissect_putfh(&cp);
2422 nfsm_v4dissect_commit(&cp, &commit);
2425 /* nfsm_wcc_data(vp, wccflag);*/
2426 if (bcmp(nmp->nm_verf, commit.verf, NFSX_V4VERF)) {
2427 bcopy(commit.verf, nmp->nm_verf, NFSX_V4VERF);
2428 error = NFSERR_STALEWRITEVERF;
2432 error = nfs_v4postop(&cp, error);
2441 * For async requests when nfsiod(s) are running, queue the request by
2442 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2446 nfs4_strategy(struct vop_strategy_args *ap)
2448 struct buf *bp = ap->a_bp;
2452 KASSERT(!(bp->b_flags & B_DONE),
2453 ("nfs4_strategy: buffer %p unexpectedly marked B_DONE", bp));
2454 BUF_ASSERT_HELD(bp);
2456 if (bp->b_iocmd == BIO_READ)
2462 * If the op is asynchronous and an i/o daemon is waiting
2463 * queue the request, wake it up and wait for completion
2464 * otherwise just do it ourselves.
2466 if ((bp->b_flags & B_ASYNC) == 0 ||
2467 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2468 error = nfs_doio(ap->a_vp, bp, cr, curthread);
2473 * fsync vnode op. Just call nfs4_flush() with commit == 1.
2477 nfs4_fsync(struct vop_fsync_args *ap)
2479 return (nfs4_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2483 * Flush all the blocks associated with a vnode.
2484 * Walk through the buffer pool and push any dirty pages
2485 * associated with the vnode.
2488 nfs4_flush(struct vnode *vp, int waitfor, struct thread *td,
2491 struct nfsnode *np = VTONFS(vp);
2496 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2497 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2499 u_quad_t off, endoff, toff;
2500 struct ucred* wcred = NULL;
2501 struct buf **bvec = NULL;
2502 #ifndef NFS_COMMITBVECSIZ
2503 #define NFS_COMMITBVECSIZ 20
2505 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2506 int bvecsize = 0, bveccount;
2509 if (nmp->nm_flag & NFSMNT_INT)
2514 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2515 * server, but nas not been committed to stable storage on the server
2516 * yet. On the first pass, the byte range is worked out and the commit
2517 * rpc is done. On the second pass, nfs_writebp() is called to do the
2524 if (NFS_ISV3(vp) && commit) {
2525 if (bvec != NULL && bvec != bvec_on_stack)
2528 * Count up how many buffers waiting for a commit.
2532 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2533 if (!BUF_ISLOCKED(bp) &&
2534 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2535 == (B_DELWRI | B_NEEDCOMMIT))
2539 * Allocate space to remember the list of bufs to commit. It is
2540 * important to use M_NOWAIT here to avoid a race with nfs4_write.
2541 * If we can't get memory (for whatever reason), we will end up
2542 * committing the buffers one-by-one in the loop below.
2544 if (bveccount > NFS_COMMITBVECSIZ) {
2546 * Release the vnode interlock to avoid a lock
2550 bvec = (struct buf **)
2551 malloc(bveccount * sizeof(struct buf *),
2555 bvec = bvec_on_stack;
2556 bvecsize = NFS_COMMITBVECSIZ;
2558 bvecsize = bveccount;
2560 bvec = bvec_on_stack;
2561 bvecsize = NFS_COMMITBVECSIZ;
2563 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2564 if (bvecpos >= bvecsize)
2566 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2567 nbp = TAILQ_NEXT(bp, b_bobufs);
2570 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2571 (B_DELWRI | B_NEEDCOMMIT)) {
2573 nbp = TAILQ_NEXT(bp, b_bobufs);
2579 * Work out if all buffers are using the same cred
2580 * so we can deal with them all with one commit.
2582 * NOTE: we are not clearing B_DONE here, so we have
2583 * to do it later on in this routine if we intend to
2584 * initiate I/O on the bp.
2586 * Note: to avoid loopback deadlocks, we do not
2587 * assign b_runningbufspace.
2590 wcred = bp->b_wcred;
2591 else if (wcred != bp->b_wcred)
2593 vfs_busy_pages(bp, 1);
2597 * bp is protected by being locked, but nbp is not
2598 * and vfs_busy_pages() may sleep. We have to
2601 nbp = TAILQ_NEXT(bp, b_bobufs);
2604 * A list of these buffers is kept so that the
2605 * second loop knows which buffers have actually
2606 * been committed. This is necessary, since there
2607 * may be a race between the commit rpc and new
2608 * uncommitted writes on the file.
2610 bvec[bvecpos++] = bp;
2611 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2615 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2623 * Commit data on the server, as required.
2624 * If all bufs are using the same wcred, then use that with
2625 * one call for all of them, otherwise commit each one
2628 if (wcred != NOCRED)
2629 retv = nfs4_commit(vp, off, (int)(endoff - off),
2633 for (i = 0; i < bvecpos; i++) {
2636 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2638 size = (u_quad_t)(bp->b_dirtyend
2640 retv = nfs4_commit(vp, off, (int)size,
2646 if (retv == NFSERR_STALEWRITEVERF)
2647 nfs_clearcommit(vp->v_mount);
2650 * Now, either mark the blocks I/O done or mark the
2651 * blocks dirty, depending on whether the commit
2654 for (i = 0; i < bvecpos; i++) {
2656 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2659 * Error, leave B_DELWRI intact
2661 vfs_unbusy_pages(bp);
2665 * Success, remove B_DELWRI ( bundirty() ).
2667 * b_dirtyoff/b_dirtyend seem to be NFS
2668 * specific. We should probably move that
2669 * into bundirty(). XXX
2672 bp->b_flags |= B_ASYNC;
2674 bp->b_flags &= ~B_DONE;
2675 bp->b_ioflags &= ~BIO_ERROR;
2676 bp->b_dirtyoff = bp->b_dirtyend = 0;
2683 * Start/do any write(s) that are required.
2687 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2688 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2689 if (waitfor != MNT_WAIT || passone)
2692 error = BUF_TIMELOCK(bp,
2693 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2694 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2696 panic("nfs4_fsync: inconsistent lock");
2697 if (error == ENOLCK) {
2701 if (nfs4_sigintr(nmp, NULL, td)) {
2705 if (slpflag == PCATCH) {
2711 if ((bp->b_flags & B_DELWRI) == 0)
2712 panic("nfs4_fsync: not dirty");
2713 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2719 if (passone || !commit)
2720 bp->b_flags |= B_ASYNC;
2722 bp->b_flags |= B_ASYNC;
2731 if (waitfor == MNT_WAIT) {
2732 while (bo->bo_numoutput) {
2733 error = bufobj_wwait(bo, slpflag, slptimeo);
2736 if (nfs4_sigintr(nmp, NULL, td)) {
2740 if (slpflag == PCATCH) {
2747 if (vp->v_bufobj.bo_dirty.bv_cnt > 0 && commit) {
2753 if (np->n_flag & NWRITEERR) {
2754 error = np->n_error;
2755 np->n_flag &= ~NWRITEERR;
2758 if (bvec != NULL && bvec != bvec_on_stack)
2764 * NFS advisory byte-level locks.
2767 nfs4_advlock(struct vop_advlock_args *ap)
2769 struct vnode *vp = ap->a_vp;
2775 error = vn_lock(vp, LK_SHARED);
2778 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2779 size = VTONFS(vp)->n_size;
2781 error = lf_advlock(ap, &(vp->v_lockf), size);
2783 error = nfs_dolock(ap);
2788 * NFS advisory byte-level locks.
2791 nfs4_advlockasync(struct vop_advlockasync_args *ap)
2793 struct vnode *vp = ap->a_vp;
2799 error = vn_lock(vp, LK_SHARED);
2802 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2803 size = VTONFS(vp)->n_size;
2805 error = lf_advlockasync(ap, &(vp->v_lockf), size);
2814 * Print out the contents of an nfsnode.
2817 nfs4_print(struct vop_print_args *ap)
2819 struct vnode *vp = ap->a_vp;
2820 struct nfsnode *np = VTONFS(vp);
2822 printf("\tfileid %ld fsid 0x%x",
2823 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2824 if (vp->v_type == VFIFO)
2831 * This is the "real" nfs::bwrite(struct buf*).
2832 * We set B_CACHE if this is a VMIO buffer.
2835 nfs4_writebp(struct buf *bp, int force __unused, struct thread *td)
2838 int oldflags = bp->b_flags;
2844 BUF_ASSERT_HELD(bp);
2846 if (bp->b_flags & B_INVAL) {
2851 bp->b_flags |= B_CACHE;
2854 * Undirty the bp. We will redirty it later if the I/O fails.
2859 bp->b_flags &= ~B_DONE;
2860 bp->b_ioflags &= ~BIO_ERROR;
2861 bp->b_iocmd = BIO_WRITE;
2863 bufobj_wref(bp->b_bufobj);
2864 curthread->td_ru.ru_oublock++;
2868 * Note: to avoid loopback deadlocks, we do not
2869 * assign b_runningbufspace.
2871 vfs_busy_pages(bp, 1);
2874 bp->b_iooffset = dbtob(bp->b_blkno);
2877 if( (oldflags & B_ASYNC) == 0) {
2878 int rtval = bufwait(bp);
2880 if (oldflags & B_DELWRI) {
2894 * Just call nfs_writebp() with the force argument set to 1.
2896 * NOTE: B_DONE may or may not be set in a_bp on call.
2899 nfs4_bwrite(struct buf *bp)
2902 return (nfs4_writebp(bp, 1, curthread));
2905 struct buf_ops buf_ops_nfs4 = {
2906 .bop_name = "buf_ops_nfs4",
2907 .bop_write = nfs4_bwrite,
2908 .bop_strategy = bufstrategy,
2909 .bop_sync = bufsync,
2910 .bop_bdflush = bufbdflush,