37
38/*
39 * nfs version 2 and 3 server calls to vnode ops
40 * - these routines generally have 3 phases
41 * 1 - break down and validate rpc request in mbuf list
42 * 2 - do the vnode ops for the request
43 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
44 * 3 - build the rpc reply in an mbuf list
45 * nb:
46 * - do not mix the phases, since the nfsm_?? macros can return failures
47 * on a bad rpc or similar and do not do any vrele() or vput()'s
48 *
49 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
50 * error number iff error != 0 whereas
51 * returning an error from the server function implies a fatal error
52 * such as a badly constructed rpc request that should be dropped without
53 * a reply.
54 * For nfsm_reply(), the case where error == EBADRPC is treated
55 * specially; after constructing a reply, it does an immediate
56 * `goto nfsmout' to avoid getting any V3 post-op status appended.
57 *
58 * Other notes:
59 * Warning: always pay careful attention to resource cleanup on return
60 * and note that nfsm_*() macros can terminate a procedure on certain
61 * errors.
62 *
63 * lookup() and namei()
64 * may return garbage in various structural fields/return elements
65 * if an error is returned, and may garbage up nd.ni_dvp even if no
66 * error is returned and you did not request LOCKPARENT or WANTPARENT.
67 *
68 * We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
69 * buffer has been freed or not.
70 */
71
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/proc.h>
75#include <sys/namei.h>
76#include <sys/unistd.h>
77#include <sys/vnode.h>
78#include <sys/mount.h>
79#include <sys/socket.h>
80#include <sys/socketvar.h>
81#include <sys/malloc.h>
82#include <sys/mbuf.h>
83#include <sys/priv.h>
84#include <sys/dirent.h>
85#include <sys/stat.h>
86#include <sys/kernel.h>
87#include <sys/sysctl.h>
88#include <sys/bio.h>
89#include <sys/buf.h>
90
91#include <vm/vm.h>
92#include <vm/vm_extern.h>
93#include <vm/vm_object.h>
94
95#include <nfs/nfsproto.h>
96#include <nfs/rpcv2.h>
97#include <nfsserver/nfs.h>
98#include <nfs/xdr_subs.h>
99#include <nfsserver/nfsm_subs.h>
100
101#ifdef NFSRV_DEBUG
102#define nfsdbprintf(info) printf info
103#else
104#define nfsdbprintf(info)
105#endif
106
107#define MAX_COMMIT_COUNT (1024 * 1024)
108
109#define NUM_HEURISTIC 1017
110#define NHUSE_INIT 64
111#define NHUSE_INC 16
112#define NHUSE_MAX 2048
113
114static struct nfsheur {
115 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
116 off_t nh_nextr; /* next offset for sequential detection */
117 int nh_use; /* use count for selection */
118 int nh_seqcount; /* heuristic */
119} nfsheur[NUM_HEURISTIC];
120
121/* Global vars */
122
123int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
124int nfsrvw_procrastinate_v3 = 0;
125
126static struct timeval nfsver = { 0 };
127
128SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server");
129
130static int nfs_async;
131static int nfs_commit_blks;
132static int nfs_commit_miss;
133SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
134SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
135SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
136
137struct nfsrvstats nfsrvstats;
138SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RW,
139 &nfsrvstats, nfsrvstats, "S,nfsrvstats");
140
141static int nfsrv_access(struct vnode *, int, struct ucred *, int,
142 struct thread *, int);
143static void nfsrvw_coalesce(struct nfsrv_descript *,
144 struct nfsrv_descript *);
145
146/*
147 * Clear nameidata fields that are tested in nsfmout cleanup code prior
148 * to using first nfsm macro (that might jump to the cleanup code).
149 */
150
151static __inline void
152ndclear(struct nameidata *nd)
153{
154
155 nd->ni_cnd.cn_flags = 0;
156 nd->ni_vp = NULL;
157 nd->ni_dvp = NULL;
158 nd->ni_startdir = NULL;
159}
160
161/*
162 * Takes two vfslocked integers and returns with at most one
163 * reference to giant. The return value indicates whether giant
164 * is held by either lock. This simplifies nfsrv ops by allowing
165 * them to track only one vfslocked var.
166 */
167static __inline int
168nfsrv_lockedpair(int vfs1, int vfs2)
169{
170
171 if (vfs1 && vfs2)
172 VFS_UNLOCK_GIANT(vfs2);
173
174 return (vfs1 | vfs2);
175}
176
177static __inline int
178nfsrv_lockedpair_nd(int vfs1, struct nameidata *nd)
179{
180 int vfs2;
181
182 vfs2 = NDHASGIANT(nd);
183
184 return nfsrv_lockedpair(vfs1, vfs2);
185}
186
187/*
188 * nfs v3 access service
189 */
190int
191nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
192 struct thread *td, struct mbuf **mrq)
193{
194 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
195 struct sockaddr *nam = nfsd->nd_nam;
196 caddr_t dpos = nfsd->nd_dpos;
197 struct ucred *cred = nfsd->nd_cr;
198 struct vnode *vp = NULL;
199 nfsfh_t nfh;
200 fhandle_t *fhp;
201 u_int32_t *tl;
202 caddr_t bpos;
203 int error = 0, rdonly, getret;
204 struct mbuf *mb, *mreq;
205 struct vattr vattr, *vap = &vattr;
206 u_long testmode, nfsmode;
207 int v3 = (nfsd->nd_flag & ND_NFSV3);
208 int vfslocked;
209
210 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
211 if (!v3)
212 panic("nfsrv3_access: v3 proc called on a v2 connection");
213 fhp = &nfh.fh_generic;
214 nfsm_srvmtofh(fhp);
215 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
216 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
217 nam, &rdonly, TRUE);
218 if (error) {
219 nfsm_reply(NFSX_UNSIGNED);
220 nfsm_srvpostop_attr(1, NULL);
221 error = 0;
222 goto nfsmout;
223 }
224 nfsmode = fxdr_unsigned(u_int32_t, *tl);
225 if ((nfsmode & NFSV3ACCESS_READ) &&
226 nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
227 nfsmode &= ~NFSV3ACCESS_READ;
228 if (vp->v_type == VDIR)
229 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
230 NFSV3ACCESS_DELETE);
231 else
232 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
233 if ((nfsmode & testmode) &&
234 nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
235 nfsmode &= ~testmode;
236 if (vp->v_type == VDIR)
237 testmode = NFSV3ACCESS_LOOKUP;
238 else
239 testmode = NFSV3ACCESS_EXECUTE;
240 if ((nfsmode & testmode) &&
241 nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
242 nfsmode &= ~testmode;
243 getret = VOP_GETATTR(vp, vap, cred, td);
244 vput(vp);
245 vp = NULL;
246 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
247 nfsm_srvpostop_attr(getret, vap);
248 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
249 *tl = txdr_unsigned(nfsmode);
250nfsmout:
251 if (vp)
252 vput(vp);
253 VFS_UNLOCK_GIANT(vfslocked);
254 return(error);
255}
256
257/*
258 * nfs getattr service
259 */
260int
261nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
262 struct thread *td, struct mbuf **mrq)
263{
264 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
265 struct sockaddr *nam = nfsd->nd_nam;
266 caddr_t dpos = nfsd->nd_dpos;
267 struct ucred *cred = nfsd->nd_cr;
268 struct nfs_fattr *fp;
269 struct vattr va;
270 struct vattr *vap = &va;
271 struct vnode *vp = NULL;
272 nfsfh_t nfh;
273 fhandle_t *fhp;
274 caddr_t bpos;
275 int error = 0, rdonly;
276 struct mbuf *mb, *mreq;
277 int vfslocked;
278
279 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
280 vfslocked = 0;
281 fhp = &nfh.fh_generic;
282 nfsm_srvmtofh(fhp);
283 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp, nam,
284 &rdonly, TRUE);
285 if (error) {
286 nfsm_reply(0);
287 error = 0;
288 goto nfsmout;
289 }
290 error = VOP_GETATTR(vp, vap, cred, td);
291 vput(vp);
292 vp = NULL;
293 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
294 if (error) {
295 error = 0;
296 goto nfsmout;
297 }
298 fp = nfsm_build(struct nfs_fattr *,
299 NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
300 nfsm_srvfillattr(vap, fp);
301 /* fall through */
302
303nfsmout:
304 if (vp)
305 vput(vp);
306 VFS_UNLOCK_GIANT(vfslocked);
307 return(error);
308}
309
310/*
311 * nfs setattr service
312 */
313int
314nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
315 struct thread *td, struct mbuf **mrq)
316{
317 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
318 struct sockaddr *nam = nfsd->nd_nam;
319 caddr_t dpos = nfsd->nd_dpos;
320 struct ucred *cred = nfsd->nd_cr;
321 struct vattr va, preat;
322 struct vattr *vap = &va;
323 struct nfsv2_sattr *sp;
324 struct nfs_fattr *fp;
325 struct vnode *vp = NULL;
326 nfsfh_t nfh;
327 fhandle_t *fhp;
328 u_int32_t *tl;
329 caddr_t bpos;
330 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
331 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
332 struct mbuf *mb, *mreq;
333 struct timespec guard = { 0, 0 };
334 struct mount *mp = NULL;
335 int tvfslocked;
336 int vfslocked;
337
338 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
339 vfslocked = 0;
340 fhp = &nfh.fh_generic;
341 nfsm_srvmtofh(fhp);
342 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
343 error = ESTALE;
344 goto out;
345 }
346 vfslocked = VFS_LOCK_GIANT(mp);
347 (void) vn_start_write(NULL, &mp, V_WAIT);
348 vfs_rel(mp); /* The write holds a ref. */
349 VATTR_NULL(vap);
350 if (v3) {
351 nfsm_srvsattr(vap);
352 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
353 gcheck = fxdr_unsigned(int, *tl);
354 if (gcheck) {
355 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
356 fxdr_nfsv3time(tl, &guard);
357 }
358 } else {
359 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
360 /*
361 * Nah nah nah nah na nah
362 * There is a bug in the Sun client that puts 0xffff in the mode
363 * field of sattr when it should put in 0xffffffff. The u_short
364 * doesn't sign extend.
365 * --> check the low order 2 bytes for 0xffff
366 */
367 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
368 vap->va_mode = nfstov_mode(sp->sa_mode);
369 if (sp->sa_uid != nfsrv_nfs_xdrneg1)
370 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
371 if (sp->sa_gid != nfsrv_nfs_xdrneg1)
372 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
373 if (sp->sa_size != nfsrv_nfs_xdrneg1)
374 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
375 if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) {
376#ifdef notyet
377 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
378#else
379 vap->va_atime.tv_sec =
380 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
381 vap->va_atime.tv_nsec = 0;
382#endif
383 }
384 if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1)
385 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
386
387 }
388
389 /*
390 * Now that we have all the fields, lets do it.
391 */
392 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
393 nam, &rdonly, TRUE);
394 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
395 if (error) {
396 nfsm_reply(2 * NFSX_UNSIGNED);
397 if (v3)
398 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
399 error = 0;
400 goto nfsmout;
401 }
402
403 /*
404 * vp now an active resource, pay careful attention to cleanup
405 */
406 if (v3) {
407 error = preat_ret = VOP_GETATTR(vp, &preat, cred, td);
408 if (!error && gcheck &&
409 (preat.va_ctime.tv_sec != guard.tv_sec ||
410 preat.va_ctime.tv_nsec != guard.tv_nsec))
411 error = NFSERR_NOT_SYNC;
412 if (error) {
413 vput(vp);
414 vp = NULL;
415 nfsm_reply(NFSX_WCCDATA(v3));
416 if (v3)
417 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
418 error = 0;
419 goto nfsmout;
420 }
421 }
422
423 /*
424 * If the size is being changed write acces is required, otherwise
425 * just check for a read only filesystem.
426 */
427 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
428 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
429 error = EROFS;
430 goto out;
431 }
432 } else {
433 if (vp->v_type == VDIR) {
434 error = EISDIR;
435 goto out;
436 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
437 td, 0)) != 0)
438 goto out;
439 }
440 error = VOP_SETATTR(vp, vap, cred, td);
441 postat_ret = VOP_GETATTR(vp, vap, cred, td);
442 if (!error)
443 error = postat_ret;
444out:
445 if (vp != NULL)
446 vput(vp);
447
448 vp = NULL;
449 nfsm_reply(NFSX_WCCORFATTR(v3));
450 if (v3) {
451 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
452 } else if (!error) {
453 /* v2 non-error case. */
454 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
455 nfsm_srvfillattr(vap, fp);
456 }
457 error = 0;
458 /* fall through */
459
460nfsmout:
461 if (vp)
462 vput(vp);
463 vn_finished_write(mp);
464 VFS_UNLOCK_GIANT(vfslocked);
465 return(error);
466}
467
468/*
469 * nfs lookup rpc
470 */
471int
472nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
473 struct thread *td, struct mbuf **mrq)
474{
475 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
476 struct sockaddr *nam = nfsd->nd_nam;
477 caddr_t dpos = nfsd->nd_dpos;
478 struct ucred *cred = nfsd->nd_cr;
479 struct nfs_fattr *fp;
480 struct nameidata nd, ind, *ndp = &nd;
481 struct vnode *vp, *dirp = NULL;
482 nfsfh_t nfh;
483 fhandle_t *fhp;
484 caddr_t bpos;
485 int error = 0, len, dirattr_ret = 1;
486 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
487 struct mbuf *mb, *mreq;
488 struct vattr va, dirattr, *vap = &va;
489 int tvfslocked;
490 int vfslocked;
491
492 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
493 ndclear(&nd);
494 vfslocked = 0;
495
496 fhp = &nfh.fh_generic;
497 nfsm_srvmtofh(fhp);
498 nfsm_srvnamesiz(len);
499
500 pubflag = nfs_ispublicfh(fhp);
501
502 nd.ni_cnd.cn_cred = cred;
503 nd.ni_cnd.cn_nameiop = LOOKUP;
504 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART | MPSAFE;
505 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
506 &dirp, v3, &dirattr, &dirattr_ret, td, pubflag);
507 vfslocked = NDHASGIANT(&nd);
508
509 /*
510 * namei failure, only dirp to cleanup. Clear out garbarge from
511 * structure in case macros jump to nfsmout.
512 */
513
514 if (error) {
515 if (dirp) {
516 vrele(dirp);
517 dirp = NULL;
518 }
519 nfsm_reply(NFSX_POSTOPATTR(v3));
520 if (v3)
521 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
522 error = 0;
523 goto nfsmout;
524 }
525
526 /*
527 * Locate index file for public filehandle
528 *
529 * error is 0 on entry and 0 on exit from this block.
530 */
531
532 if (pubflag) {
533 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
534 /*
535 * Setup call to lookup() to see if we can find
536 * the index file. Arguably, this doesn't belong
537 * in a kernel.. Ugh. If an error occurs, do not
538 * try to install an index file and then clear the
539 * error.
540 *
541 * When we replace nd with ind and redirect ndp,
542 * maintenance of ni_startdir and ni_vp shift to
543 * ind and we have to clean them up in the old nd.
544 * However, the cnd resource continues to be maintained
545 * via the original nd. Confused? You aren't alone!
546 */
547 ind = nd;
548 VOP_UNLOCK(nd.ni_vp, 0);
549 ind.ni_pathlen = strlen(nfs_pub.np_index);
550 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
551 nfs_pub.np_index;
552 ind.ni_startdir = nd.ni_vp;
553 VREF(ind.ni_startdir);
554 ind.ni_cnd.cn_flags &= ~GIANTHELD;
555 tvfslocked = VFS_LOCK_GIANT(ind.ni_startdir->v_mount);
556 if (tvfslocked)
557 nd.ni_cnd.cn_flags |= GIANTHELD;
558 error = lookup(&ind);
559 ind.ni_dvp = NULL;
560 vfslocked = nfsrv_lockedpair_nd(vfslocked, &ind);
561 ind.ni_cnd.cn_flags &= ~GIANTHELD;
562
563 if (error == 0) {
564 /*
565 * Found an index file. Get rid of
566 * the old references. transfer nd.ni_vp'
567 */
568 if (dirp)
569 vrele(dirp);
570 dirp = nd.ni_vp;
571 nd.ni_vp = NULL;
572 vrele(nd.ni_startdir);
573 nd.ni_startdir = NULL;
574 ndp = &ind;
575 }
576 error = 0;
577 }
578 /*
579 * If the public filehandle was used, check that this lookup
580 * didn't result in a filehandle outside the publicly exported
581 * filesystem. We clear the poor vp here to avoid lockups due
582 * to NFS I/O.
583 */
584
585 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
586 vput(nd.ni_vp);
587 nd.ni_vp = NULL;
588 error = EPERM;
589 }
590 }
591
592 /*
593 * Resources at this point:
594 * ndp->ni_vp may not be NULL
595 */
596
597 if (error) {
598 nfsm_reply(NFSX_POSTOPATTR(v3));
599 if (v3)
600 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
601 error = 0;
602 goto nfsmout;
603 }
604
605 /*
606 * Get underlying attribute, then release remaining resources ( for
607 * the same potential blocking reason ) and reply.
608 */
609 vp = ndp->ni_vp;
610 bzero((caddr_t)fhp, sizeof(nfh));
611 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
612 error = VOP_VPTOFH(vp, &fhp->fh_fid);
613 if (!error)
614 error = VOP_GETATTR(vp, vap, cred, td);
615
616 vput(vp);
617 vrele(ndp->ni_startdir);
618 vrele(dirp);
619 ndp->ni_vp = NULL;
620 ndp->ni_startdir = NULL;
621 dirp = NULL;
622 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
623 if (error) {
624 if (v3)
625 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
626 error = 0;
627 goto nfsmout;
628 }
629 nfsm_srvfhtom(fhp, v3);
630 if (v3) {
631 nfsm_srvpostop_attr(0, vap);
632 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
633 } else {
634 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
635 nfsm_srvfillattr(vap, fp);
636 }
637
638nfsmout:
639 if (ndp->ni_vp || dirp || ndp->ni_startdir) {
640 if (ndp->ni_vp)
641 vput(ndp->ni_vp);
642 if (dirp)
643 vrele(dirp);
644 if (ndp->ni_startdir)
645 vrele(ndp->ni_startdir);
646 }
647 NDFREE(&nd, NDF_ONLY_PNBUF);
648 VFS_UNLOCK_GIANT(vfslocked);
649 return (error);
650}
651
652/*
653 * nfs readlink service
654 */
655int
656nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
657 struct thread *td, struct mbuf **mrq)
658{
659 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
660 struct sockaddr *nam = nfsd->nd_nam;
661 caddr_t dpos = nfsd->nd_dpos;
662 struct ucred *cred = nfsd->nd_cr;
663 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
664 struct iovec *ivp = iv;
665 struct mbuf *mp;
666 u_int32_t *tl;
667 caddr_t bpos;
668 int error = 0, rdonly, i, tlen, len, getret;
669 int v3 = (nfsd->nd_flag & ND_NFSV3);
670 struct mbuf *mb, *mp3, *nmp, *mreq;
671 struct vnode *vp = NULL;
672 struct vattr attr;
673 nfsfh_t nfh;
674 fhandle_t *fhp;
675 struct uio io, *uiop = &io;
676 int vfslocked;
677
678 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
679 vfslocked = 0;
680#ifndef nolint
681 mp = NULL;
682#endif
683 mp3 = NULL;
684 fhp = &nfh.fh_generic;
685 nfsm_srvmtofh(fhp);
686 len = 0;
687 i = 0;
688 while (len < NFS_MAXPATHLEN) {
| 37
38/*
39 * nfs version 2 and 3 server calls to vnode ops
40 * - these routines generally have 3 phases
41 * 1 - break down and validate rpc request in mbuf list
42 * 2 - do the vnode ops for the request
43 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
44 * 3 - build the rpc reply in an mbuf list
45 * nb:
46 * - do not mix the phases, since the nfsm_?? macros can return failures
47 * on a bad rpc or similar and do not do any vrele() or vput()'s
48 *
49 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
50 * error number iff error != 0 whereas
51 * returning an error from the server function implies a fatal error
52 * such as a badly constructed rpc request that should be dropped without
53 * a reply.
54 * For nfsm_reply(), the case where error == EBADRPC is treated
55 * specially; after constructing a reply, it does an immediate
56 * `goto nfsmout' to avoid getting any V3 post-op status appended.
57 *
58 * Other notes:
59 * Warning: always pay careful attention to resource cleanup on return
60 * and note that nfsm_*() macros can terminate a procedure on certain
61 * errors.
62 *
63 * lookup() and namei()
64 * may return garbage in various structural fields/return elements
65 * if an error is returned, and may garbage up nd.ni_dvp even if no
66 * error is returned and you did not request LOCKPARENT or WANTPARENT.
67 *
68 * We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
69 * buffer has been freed or not.
70 */
71
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/proc.h>
75#include <sys/namei.h>
76#include <sys/unistd.h>
77#include <sys/vnode.h>
78#include <sys/mount.h>
79#include <sys/socket.h>
80#include <sys/socketvar.h>
81#include <sys/malloc.h>
82#include <sys/mbuf.h>
83#include <sys/priv.h>
84#include <sys/dirent.h>
85#include <sys/stat.h>
86#include <sys/kernel.h>
87#include <sys/sysctl.h>
88#include <sys/bio.h>
89#include <sys/buf.h>
90
91#include <vm/vm.h>
92#include <vm/vm_extern.h>
93#include <vm/vm_object.h>
94
95#include <nfs/nfsproto.h>
96#include <nfs/rpcv2.h>
97#include <nfsserver/nfs.h>
98#include <nfs/xdr_subs.h>
99#include <nfsserver/nfsm_subs.h>
100
101#ifdef NFSRV_DEBUG
102#define nfsdbprintf(info) printf info
103#else
104#define nfsdbprintf(info)
105#endif
106
107#define MAX_COMMIT_COUNT (1024 * 1024)
108
109#define NUM_HEURISTIC 1017
110#define NHUSE_INIT 64
111#define NHUSE_INC 16
112#define NHUSE_MAX 2048
113
114static struct nfsheur {
115 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
116 off_t nh_nextr; /* next offset for sequential detection */
117 int nh_use; /* use count for selection */
118 int nh_seqcount; /* heuristic */
119} nfsheur[NUM_HEURISTIC];
120
121/* Global vars */
122
123int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
124int nfsrvw_procrastinate_v3 = 0;
125
126static struct timeval nfsver = { 0 };
127
128SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server");
129
130static int nfs_async;
131static int nfs_commit_blks;
132static int nfs_commit_miss;
133SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
134SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
135SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
136
137struct nfsrvstats nfsrvstats;
138SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RW,
139 &nfsrvstats, nfsrvstats, "S,nfsrvstats");
140
141static int nfsrv_access(struct vnode *, int, struct ucred *, int,
142 struct thread *, int);
143static void nfsrvw_coalesce(struct nfsrv_descript *,
144 struct nfsrv_descript *);
145
146/*
147 * Clear nameidata fields that are tested in nsfmout cleanup code prior
148 * to using first nfsm macro (that might jump to the cleanup code).
149 */
150
151static __inline void
152ndclear(struct nameidata *nd)
153{
154
155 nd->ni_cnd.cn_flags = 0;
156 nd->ni_vp = NULL;
157 nd->ni_dvp = NULL;
158 nd->ni_startdir = NULL;
159}
160
161/*
162 * Takes two vfslocked integers and returns with at most one
163 * reference to giant. The return value indicates whether giant
164 * is held by either lock. This simplifies nfsrv ops by allowing
165 * them to track only one vfslocked var.
166 */
167static __inline int
168nfsrv_lockedpair(int vfs1, int vfs2)
169{
170
171 if (vfs1 && vfs2)
172 VFS_UNLOCK_GIANT(vfs2);
173
174 return (vfs1 | vfs2);
175}
176
177static __inline int
178nfsrv_lockedpair_nd(int vfs1, struct nameidata *nd)
179{
180 int vfs2;
181
182 vfs2 = NDHASGIANT(nd);
183
184 return nfsrv_lockedpair(vfs1, vfs2);
185}
186
187/*
188 * nfs v3 access service
189 */
190int
191nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
192 struct thread *td, struct mbuf **mrq)
193{
194 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
195 struct sockaddr *nam = nfsd->nd_nam;
196 caddr_t dpos = nfsd->nd_dpos;
197 struct ucred *cred = nfsd->nd_cr;
198 struct vnode *vp = NULL;
199 nfsfh_t nfh;
200 fhandle_t *fhp;
201 u_int32_t *tl;
202 caddr_t bpos;
203 int error = 0, rdonly, getret;
204 struct mbuf *mb, *mreq;
205 struct vattr vattr, *vap = &vattr;
206 u_long testmode, nfsmode;
207 int v3 = (nfsd->nd_flag & ND_NFSV3);
208 int vfslocked;
209
210 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
211 if (!v3)
212 panic("nfsrv3_access: v3 proc called on a v2 connection");
213 fhp = &nfh.fh_generic;
214 nfsm_srvmtofh(fhp);
215 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
216 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
217 nam, &rdonly, TRUE);
218 if (error) {
219 nfsm_reply(NFSX_UNSIGNED);
220 nfsm_srvpostop_attr(1, NULL);
221 error = 0;
222 goto nfsmout;
223 }
224 nfsmode = fxdr_unsigned(u_int32_t, *tl);
225 if ((nfsmode & NFSV3ACCESS_READ) &&
226 nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
227 nfsmode &= ~NFSV3ACCESS_READ;
228 if (vp->v_type == VDIR)
229 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
230 NFSV3ACCESS_DELETE);
231 else
232 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
233 if ((nfsmode & testmode) &&
234 nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
235 nfsmode &= ~testmode;
236 if (vp->v_type == VDIR)
237 testmode = NFSV3ACCESS_LOOKUP;
238 else
239 testmode = NFSV3ACCESS_EXECUTE;
240 if ((nfsmode & testmode) &&
241 nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
242 nfsmode &= ~testmode;
243 getret = VOP_GETATTR(vp, vap, cred, td);
244 vput(vp);
245 vp = NULL;
246 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
247 nfsm_srvpostop_attr(getret, vap);
248 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
249 *tl = txdr_unsigned(nfsmode);
250nfsmout:
251 if (vp)
252 vput(vp);
253 VFS_UNLOCK_GIANT(vfslocked);
254 return(error);
255}
256
257/*
258 * nfs getattr service
259 */
260int
261nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
262 struct thread *td, struct mbuf **mrq)
263{
264 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
265 struct sockaddr *nam = nfsd->nd_nam;
266 caddr_t dpos = nfsd->nd_dpos;
267 struct ucred *cred = nfsd->nd_cr;
268 struct nfs_fattr *fp;
269 struct vattr va;
270 struct vattr *vap = &va;
271 struct vnode *vp = NULL;
272 nfsfh_t nfh;
273 fhandle_t *fhp;
274 caddr_t bpos;
275 int error = 0, rdonly;
276 struct mbuf *mb, *mreq;
277 int vfslocked;
278
279 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
280 vfslocked = 0;
281 fhp = &nfh.fh_generic;
282 nfsm_srvmtofh(fhp);
283 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp, nam,
284 &rdonly, TRUE);
285 if (error) {
286 nfsm_reply(0);
287 error = 0;
288 goto nfsmout;
289 }
290 error = VOP_GETATTR(vp, vap, cred, td);
291 vput(vp);
292 vp = NULL;
293 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
294 if (error) {
295 error = 0;
296 goto nfsmout;
297 }
298 fp = nfsm_build(struct nfs_fattr *,
299 NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
300 nfsm_srvfillattr(vap, fp);
301 /* fall through */
302
303nfsmout:
304 if (vp)
305 vput(vp);
306 VFS_UNLOCK_GIANT(vfslocked);
307 return(error);
308}
309
310/*
311 * nfs setattr service
312 */
313int
314nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
315 struct thread *td, struct mbuf **mrq)
316{
317 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
318 struct sockaddr *nam = nfsd->nd_nam;
319 caddr_t dpos = nfsd->nd_dpos;
320 struct ucred *cred = nfsd->nd_cr;
321 struct vattr va, preat;
322 struct vattr *vap = &va;
323 struct nfsv2_sattr *sp;
324 struct nfs_fattr *fp;
325 struct vnode *vp = NULL;
326 nfsfh_t nfh;
327 fhandle_t *fhp;
328 u_int32_t *tl;
329 caddr_t bpos;
330 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
331 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
332 struct mbuf *mb, *mreq;
333 struct timespec guard = { 0, 0 };
334 struct mount *mp = NULL;
335 int tvfslocked;
336 int vfslocked;
337
338 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
339 vfslocked = 0;
340 fhp = &nfh.fh_generic;
341 nfsm_srvmtofh(fhp);
342 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
343 error = ESTALE;
344 goto out;
345 }
346 vfslocked = VFS_LOCK_GIANT(mp);
347 (void) vn_start_write(NULL, &mp, V_WAIT);
348 vfs_rel(mp); /* The write holds a ref. */
349 VATTR_NULL(vap);
350 if (v3) {
351 nfsm_srvsattr(vap);
352 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
353 gcheck = fxdr_unsigned(int, *tl);
354 if (gcheck) {
355 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
356 fxdr_nfsv3time(tl, &guard);
357 }
358 } else {
359 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
360 /*
361 * Nah nah nah nah na nah
362 * There is a bug in the Sun client that puts 0xffff in the mode
363 * field of sattr when it should put in 0xffffffff. The u_short
364 * doesn't sign extend.
365 * --> check the low order 2 bytes for 0xffff
366 */
367 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
368 vap->va_mode = nfstov_mode(sp->sa_mode);
369 if (sp->sa_uid != nfsrv_nfs_xdrneg1)
370 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
371 if (sp->sa_gid != nfsrv_nfs_xdrneg1)
372 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
373 if (sp->sa_size != nfsrv_nfs_xdrneg1)
374 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
375 if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) {
376#ifdef notyet
377 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
378#else
379 vap->va_atime.tv_sec =
380 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
381 vap->va_atime.tv_nsec = 0;
382#endif
383 }
384 if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1)
385 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
386
387 }
388
389 /*
390 * Now that we have all the fields, lets do it.
391 */
392 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
393 nam, &rdonly, TRUE);
394 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
395 if (error) {
396 nfsm_reply(2 * NFSX_UNSIGNED);
397 if (v3)
398 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
399 error = 0;
400 goto nfsmout;
401 }
402
403 /*
404 * vp now an active resource, pay careful attention to cleanup
405 */
406 if (v3) {
407 error = preat_ret = VOP_GETATTR(vp, &preat, cred, td);
408 if (!error && gcheck &&
409 (preat.va_ctime.tv_sec != guard.tv_sec ||
410 preat.va_ctime.tv_nsec != guard.tv_nsec))
411 error = NFSERR_NOT_SYNC;
412 if (error) {
413 vput(vp);
414 vp = NULL;
415 nfsm_reply(NFSX_WCCDATA(v3));
416 if (v3)
417 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
418 error = 0;
419 goto nfsmout;
420 }
421 }
422
423 /*
424 * If the size is being changed write acces is required, otherwise
425 * just check for a read only filesystem.
426 */
427 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
428 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
429 error = EROFS;
430 goto out;
431 }
432 } else {
433 if (vp->v_type == VDIR) {
434 error = EISDIR;
435 goto out;
436 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
437 td, 0)) != 0)
438 goto out;
439 }
440 error = VOP_SETATTR(vp, vap, cred, td);
441 postat_ret = VOP_GETATTR(vp, vap, cred, td);
442 if (!error)
443 error = postat_ret;
444out:
445 if (vp != NULL)
446 vput(vp);
447
448 vp = NULL;
449 nfsm_reply(NFSX_WCCORFATTR(v3));
450 if (v3) {
451 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
452 } else if (!error) {
453 /* v2 non-error case. */
454 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
455 nfsm_srvfillattr(vap, fp);
456 }
457 error = 0;
458 /* fall through */
459
460nfsmout:
461 if (vp)
462 vput(vp);
463 vn_finished_write(mp);
464 VFS_UNLOCK_GIANT(vfslocked);
465 return(error);
466}
467
468/*
469 * nfs lookup rpc
470 */
471int
472nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
473 struct thread *td, struct mbuf **mrq)
474{
475 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
476 struct sockaddr *nam = nfsd->nd_nam;
477 caddr_t dpos = nfsd->nd_dpos;
478 struct ucred *cred = nfsd->nd_cr;
479 struct nfs_fattr *fp;
480 struct nameidata nd, ind, *ndp = &nd;
481 struct vnode *vp, *dirp = NULL;
482 nfsfh_t nfh;
483 fhandle_t *fhp;
484 caddr_t bpos;
485 int error = 0, len, dirattr_ret = 1;
486 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
487 struct mbuf *mb, *mreq;
488 struct vattr va, dirattr, *vap = &va;
489 int tvfslocked;
490 int vfslocked;
491
492 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
493 ndclear(&nd);
494 vfslocked = 0;
495
496 fhp = &nfh.fh_generic;
497 nfsm_srvmtofh(fhp);
498 nfsm_srvnamesiz(len);
499
500 pubflag = nfs_ispublicfh(fhp);
501
502 nd.ni_cnd.cn_cred = cred;
503 nd.ni_cnd.cn_nameiop = LOOKUP;
504 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART | MPSAFE;
505 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
506 &dirp, v3, &dirattr, &dirattr_ret, td, pubflag);
507 vfslocked = NDHASGIANT(&nd);
508
509 /*
510 * namei failure, only dirp to cleanup. Clear out garbarge from
511 * structure in case macros jump to nfsmout.
512 */
513
514 if (error) {
515 if (dirp) {
516 vrele(dirp);
517 dirp = NULL;
518 }
519 nfsm_reply(NFSX_POSTOPATTR(v3));
520 if (v3)
521 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
522 error = 0;
523 goto nfsmout;
524 }
525
526 /*
527 * Locate index file for public filehandle
528 *
529 * error is 0 on entry and 0 on exit from this block.
530 */
531
532 if (pubflag) {
533 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
534 /*
535 * Setup call to lookup() to see if we can find
536 * the index file. Arguably, this doesn't belong
537 * in a kernel.. Ugh. If an error occurs, do not
538 * try to install an index file and then clear the
539 * error.
540 *
541 * When we replace nd with ind and redirect ndp,
542 * maintenance of ni_startdir and ni_vp shift to
543 * ind and we have to clean them up in the old nd.
544 * However, the cnd resource continues to be maintained
545 * via the original nd. Confused? You aren't alone!
546 */
547 ind = nd;
548 VOP_UNLOCK(nd.ni_vp, 0);
549 ind.ni_pathlen = strlen(nfs_pub.np_index);
550 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
551 nfs_pub.np_index;
552 ind.ni_startdir = nd.ni_vp;
553 VREF(ind.ni_startdir);
554 ind.ni_cnd.cn_flags &= ~GIANTHELD;
555 tvfslocked = VFS_LOCK_GIANT(ind.ni_startdir->v_mount);
556 if (tvfslocked)
557 nd.ni_cnd.cn_flags |= GIANTHELD;
558 error = lookup(&ind);
559 ind.ni_dvp = NULL;
560 vfslocked = nfsrv_lockedpair_nd(vfslocked, &ind);
561 ind.ni_cnd.cn_flags &= ~GIANTHELD;
562
563 if (error == 0) {
564 /*
565 * Found an index file. Get rid of
566 * the old references. transfer nd.ni_vp'
567 */
568 if (dirp)
569 vrele(dirp);
570 dirp = nd.ni_vp;
571 nd.ni_vp = NULL;
572 vrele(nd.ni_startdir);
573 nd.ni_startdir = NULL;
574 ndp = &ind;
575 }
576 error = 0;
577 }
578 /*
579 * If the public filehandle was used, check that this lookup
580 * didn't result in a filehandle outside the publicly exported
581 * filesystem. We clear the poor vp here to avoid lockups due
582 * to NFS I/O.
583 */
584
585 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
586 vput(nd.ni_vp);
587 nd.ni_vp = NULL;
588 error = EPERM;
589 }
590 }
591
592 /*
593 * Resources at this point:
594 * ndp->ni_vp may not be NULL
595 */
596
597 if (error) {
598 nfsm_reply(NFSX_POSTOPATTR(v3));
599 if (v3)
600 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
601 error = 0;
602 goto nfsmout;
603 }
604
605 /*
606 * Get underlying attribute, then release remaining resources ( for
607 * the same potential blocking reason ) and reply.
608 */
609 vp = ndp->ni_vp;
610 bzero((caddr_t)fhp, sizeof(nfh));
611 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
612 error = VOP_VPTOFH(vp, &fhp->fh_fid);
613 if (!error)
614 error = VOP_GETATTR(vp, vap, cred, td);
615
616 vput(vp);
617 vrele(ndp->ni_startdir);
618 vrele(dirp);
619 ndp->ni_vp = NULL;
620 ndp->ni_startdir = NULL;
621 dirp = NULL;
622 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
623 if (error) {
624 if (v3)
625 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
626 error = 0;
627 goto nfsmout;
628 }
629 nfsm_srvfhtom(fhp, v3);
630 if (v3) {
631 nfsm_srvpostop_attr(0, vap);
632 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
633 } else {
634 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
635 nfsm_srvfillattr(vap, fp);
636 }
637
638nfsmout:
639 if (ndp->ni_vp || dirp || ndp->ni_startdir) {
640 if (ndp->ni_vp)
641 vput(ndp->ni_vp);
642 if (dirp)
643 vrele(dirp);
644 if (ndp->ni_startdir)
645 vrele(ndp->ni_startdir);
646 }
647 NDFREE(&nd, NDF_ONLY_PNBUF);
648 VFS_UNLOCK_GIANT(vfslocked);
649 return (error);
650}
651
652/*
653 * nfs readlink service
654 */
655int
656nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
657 struct thread *td, struct mbuf **mrq)
658{
659 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
660 struct sockaddr *nam = nfsd->nd_nam;
661 caddr_t dpos = nfsd->nd_dpos;
662 struct ucred *cred = nfsd->nd_cr;
663 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
664 struct iovec *ivp = iv;
665 struct mbuf *mp;
666 u_int32_t *tl;
667 caddr_t bpos;
668 int error = 0, rdonly, i, tlen, len, getret;
669 int v3 = (nfsd->nd_flag & ND_NFSV3);
670 struct mbuf *mb, *mp3, *nmp, *mreq;
671 struct vnode *vp = NULL;
672 struct vattr attr;
673 nfsfh_t nfh;
674 fhandle_t *fhp;
675 struct uio io, *uiop = &io;
676 int vfslocked;
677
678 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
679 vfslocked = 0;
680#ifndef nolint
681 mp = NULL;
682#endif
683 mp3 = NULL;
684 fhp = &nfh.fh_generic;
685 nfsm_srvmtofh(fhp);
686 len = 0;
687 i = 0;
688 while (len < NFS_MAXPATHLEN) {
|
940 m->m_len = 0;
941 m2->m_next = m;
942 m2 = m;
943 }
944 }
945 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
946 M_TEMP, M_WAITOK);
947 uiop->uio_iov = iv2 = iv;
948 m = mb;
949 left = len;
950 i = 0;
951 while (left > 0) {
952 if (m == NULL)
953 panic("nfsrv_read iov");
954 siz = min(M_TRAILINGSPACE(m), left);
955 if (siz > 0) {
956 iv->iov_base = mtod(m, caddr_t) + m->m_len;
957 iv->iov_len = siz;
958 m->m_len += siz;
959 left -= siz;
960 iv++;
961 i++;
962 }
963 m = m->m_next;
964 }
965 uiop->uio_iovcnt = i;
966 uiop->uio_offset = off;
967 uiop->uio_resid = len;
968 uiop->uio_rw = UIO_READ;
969 uiop->uio_segflg = UIO_SYSSPACE;
970 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
971 off = uiop->uio_offset;
972 nh->nh_nextr = off;
973 FREE((caddr_t)iv2, M_TEMP);
974 if (error || (getret = VOP_GETATTR(vp, vap, cred, td))) {
975 if (!error)
976 error = getret;
977 m_freem(mreq);
978 vput(vp);
979 vp = NULL;
980 nfsm_reply(NFSX_POSTOPATTR(v3));
981 if (v3)
982 nfsm_srvpostop_attr(getret, vap);
983 error = 0;
984 goto nfsmout;
985 }
986 } else
987 uiop->uio_resid = 0;
988 vput(vp);
989 vp = NULL;
990 nfsm_srvfillattr(vap, fp);
991 tlen = len - uiop->uio_resid;
992 cnt = cnt < tlen ? cnt : tlen;
993 tlen = nfsm_rndup(cnt);
994 if (len != tlen || tlen != cnt)
995 nfsm_adj(mb, len - tlen, tlen - cnt);
996 if (v3) {
997 *tl++ = txdr_unsigned(cnt);
998 if (cnt < reqlen)
999 *tl++ = nfsrv_nfs_true;
1000 else
1001 *tl++ = nfsrv_nfs_false;
1002 }
1003 *tl = txdr_unsigned(cnt);
1004nfsmout:
1005 if (vp)
1006 vput(vp);
1007 VFS_UNLOCK_GIANT(vfslocked);
1008 return(error);
1009}
1010
1011/*
1012 * nfs write service
1013 */
1014int
1015nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1016 struct thread *td, struct mbuf **mrq)
1017{
1018 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1019 struct sockaddr *nam = nfsd->nd_nam;
1020 caddr_t dpos = nfsd->nd_dpos;
1021 struct ucred *cred = nfsd->nd_cr;
1022 struct iovec *ivp;
1023 int i, cnt;
1024 struct mbuf *mp;
1025 struct nfs_fattr *fp;
1026 struct iovec *iv;
1027 struct vattr va, forat;
1028 struct vattr *vap = &va;
1029 u_int32_t *tl;
1030 caddr_t bpos;
1031 int error = 0, rdonly, len, forat_ret = 1;
1032 int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust;
1033 int stable = NFSV3WRITE_FILESYNC;
1034 int v3 = (nfsd->nd_flag & ND_NFSV3);
1035 struct mbuf *mb, *mreq;
1036 struct vnode *vp = NULL;
1037 nfsfh_t nfh;
1038 fhandle_t *fhp;
1039 struct uio io, *uiop = &io;
1040 off_t off;
1041 struct mount *mntp = NULL;
1042 int tvfslocked;
1043 int vfslocked;
1044
1045 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1046 vfslocked = 0;
1047 if (mrep == NULL) {
1048 *mrq = NULL;
1049 error = 0;
1050 goto nfsmout;
1051 }
1052 fhp = &nfh.fh_generic;
1053 nfsm_srvmtofh(fhp);
1054 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1055 error = ESTALE;
1056 goto ereply;
1057 }
1058 vfslocked = VFS_LOCK_GIANT(mntp);
1059 (void) vn_start_write(NULL, &mntp, V_WAIT);
1060 vfs_rel(mntp); /* The write holds a ref. */
1061 if (v3) {
1062 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1063 off = fxdr_hyper(tl);
1064 tl += 3;
1065 stable = fxdr_unsigned(int, *tl++);
1066 } else {
1067 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1068 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1069 tl += 2;
1070 if (nfs_async)
1071 stable = NFSV3WRITE_UNSTABLE;
1072 }
1073 retlen = len = fxdr_unsigned(int32_t, *tl);
1074 cnt = i = 0;
1075
1076 /*
1077 * For NFS Version 2, it is not obvious what a write of zero length
1078 * should do, but I might as well be consistent with Version 3,
1079 * which is to return ok so long as there are no permission problems.
1080 */
1081 if (len > 0) {
1082 zeroing = 1;
1083 mp = mrep;
1084 while (mp) {
1085 if (mp == md) {
1086 zeroing = 0;
1087 adjust = dpos - mtod(mp, caddr_t);
1088 mp->m_len -= adjust;
1089 if (mp->m_len > 0 && adjust > 0)
1090 mp->m_data += adjust;
1091 }
1092 if (zeroing)
1093 mp->m_len = 0;
1094 else if (mp->m_len > 0) {
1095 i += mp->m_len;
1096 if (i > len) {
1097 mp->m_len -= (i - len);
1098 zeroing = 1;
1099 }
1100 if (mp->m_len > 0)
1101 cnt++;
1102 }
1103 mp = mp->m_next;
1104 }
1105 }
1106 if (len > NFS_MAXDATA || len < 0 || i < len) {
1107 error = EIO;
1108 nfsm_reply(2 * NFSX_UNSIGNED);
1109 if (v3)
1110 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1111 error = 0;
1112 goto nfsmout;
1113 }
1114 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
1115 nam, &rdonly, TRUE);
1116 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
1117 if (error) {
1118 vp = NULL;
1119 nfsm_reply(2 * NFSX_UNSIGNED);
1120 if (v3)
1121 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1122 error = 0;
1123 goto nfsmout;
1124 }
1125 if (v3)
1126 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1127 if (vp->v_type != VREG) {
1128 if (v3)
1129 error = EINVAL;
1130 else
1131 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1132 }
1133 if (!error)
1134 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1135 if (error) {
1136 vput(vp);
1137 vp = NULL;
1138 nfsm_reply(NFSX_WCCDATA(v3));
1139 if (v3)
1140 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1141 error = 0;
1142 goto nfsmout;
1143 }
1144
1145 if (len > 0) {
1146 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1147 M_WAITOK);
1148 uiop->uio_iov = iv = ivp;
1149 uiop->uio_iovcnt = cnt;
1150 mp = mrep;
1151 while (mp) {
1152 if (mp->m_len > 0) {
1153 ivp->iov_base = mtod(mp, caddr_t);
1154 ivp->iov_len = mp->m_len;
1155 ivp++;
1156 }
1157 mp = mp->m_next;
1158 }
1159
1160 /*
1161 * XXX
1162 * The IO_METASYNC flag indicates that all metadata (and not just
1163 * enough to ensure data integrity) mus be written to stable storage
1164 * synchronously.
1165 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1166 */
1167 if (stable == NFSV3WRITE_UNSTABLE)
1168 ioflags = IO_NODELOCKED;
1169 else if (stable == NFSV3WRITE_DATASYNC)
1170 ioflags = (IO_SYNC | IO_NODELOCKED);
1171 else
1172 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1173 uiop->uio_resid = len;
1174 uiop->uio_rw = UIO_WRITE;
1175 uiop->uio_segflg = UIO_SYSSPACE;
1176 uiop->uio_td = NULL;
1177 uiop->uio_offset = off;
1178 error = VOP_WRITE(vp, uiop, ioflags, cred);
1179 /* XXXRW: unlocked write. */
1180 nfsrvstats.srvvop_writes++;
1181 FREE((caddr_t)iv, M_TEMP);
1182 }
1183 aftat_ret = VOP_GETATTR(vp, vap, cred, td);
1184 vput(vp);
1185 vp = NULL;
1186 if (!error)
1187 error = aftat_ret;
1188ereply:
1189 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1190 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1191 if (v3) {
1192 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1193 if (error) {
1194 error = 0;
1195 goto nfsmout;
1196 }
1197 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1198 *tl++ = txdr_unsigned(retlen);
1199 /*
1200 * If nfs_async is set, then pretend the write was FILESYNC.
1201 */
1202 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1203 *tl++ = txdr_unsigned(stable);
1204 else
1205 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1206 /*
1207 * Actually, there is no need to txdr these fields,
1208 * but it may make the values more human readable,
1209 * for debugging purposes.
1210 */
1211 if (nfsver.tv_sec == 0)
1212 nfsver = boottime;
1213 *tl++ = txdr_unsigned(nfsver.tv_sec);
1214 *tl = txdr_unsigned(nfsver.tv_usec);
1215 } else if (!error) {
1216 /* v2 non-error case. */
1217 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1218 nfsm_srvfillattr(vap, fp);
1219 }
1220 error = 0;
1221nfsmout:
1222 if (vp)
1223 vput(vp);
1224 vn_finished_write(mntp);
1225 VFS_UNLOCK_GIANT(vfslocked);
1226 return(error);
1227}
1228
1229/*
1230 * For the purposes of write gathering, we must decide if the credential
1231 * associated with two pending requests have equivilent privileges. Since
1232 * NFS only uses a subset of the BSD ucred -- the effective uid and group
1233 * IDs -- we have a compare routine that checks only the relevant fields.
1234 */
1235static int
1236nfsrv_samecred(struct ucred *cr1, struct ucred *cr2)
1237{
1238 int i;
1239
1240 if (cr1->cr_uid != cr2->cr_uid)
1241 return (0);
1242 if (cr1->cr_ngroups != cr2->cr_ngroups)
1243 return (0);
1244 for (i = 0; i < cr1->cr_ngroups; i++) {
1245 if (cr1->cr_groups[i] != cr2->cr_groups[i])
1246 return (0);
1247 }
1248 return (1);
1249}
1250
1251/*
1252 * NFS write service with write gathering support. Called when
1253 * nfsrvw_procrastinate > 0.
1254 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1255 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1256 * Jan. 1994.
1257 */
1258int
1259nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1260 struct thread *td, struct mbuf **mrq)
1261{
1262 struct iovec *ivp;
1263 struct mbuf *mp;
1264 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1265 struct nfs_fattr *fp;
1266 int i;
1267 struct iovec *iov;
1268 struct nfsrvw_delayhash *wpp;
1269 struct ucred *cred;
1270 struct vattr va, forat;
1271 u_int32_t *tl;
1272 caddr_t bpos, dpos;
1273 int error = 0, rdonly, len, forat_ret = 1;
1274 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1275 struct mbuf *mb, *mreq, *mrep, *md;
1276 struct vnode *vp = NULL;
1277 struct uio io, *uiop = &io;
1278 u_quad_t cur_usec;
1279 struct mount *mntp = NULL;
1280 int mvfslocked;
1281 int vfslocked;
1282
1283 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1284#ifndef nolint
1285 i = 0;
1286 len = 0;
1287#endif
1288 *mrq = NULL;
1289 if (*ndp) {
1290 nfsd = *ndp;
1291 *ndp = NULL;
1292 mrep = nfsd->nd_mrep;
1293 md = nfsd->nd_md;
1294 dpos = nfsd->nd_dpos;
1295 cred = nfsd->nd_cr;
1296 v3 = (nfsd->nd_flag & ND_NFSV3);
1297 LIST_INIT(&nfsd->nd_coalesce);
1298 nfsd->nd_mreq = NULL;
1299 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1300 cur_usec = nfs_curusec();
1301 nfsd->nd_time = cur_usec +
1302 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1303
1304 /*
1305 * Now, get the write header..
1306 */
1307 nfsm_srvmtofh(&nfsd->nd_fh);
1308 if (v3) {
1309 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1310 nfsd->nd_off = fxdr_hyper(tl);
1311 tl += 3;
1312 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1313 } else {
1314 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1315 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1316 tl += 2;
1317 if (nfs_async)
1318 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1319 }
1320 len = fxdr_unsigned(int32_t, *tl);
1321 nfsd->nd_len = len;
1322 nfsd->nd_eoff = nfsd->nd_off + len;
1323
1324 /*
1325 * Trim the header out of the mbuf list and trim off any trailing
1326 * junk so that the mbuf list has only the write data.
1327 */
1328 zeroing = 1;
1329 i = 0;
1330 mp = mrep;
1331 while (mp) {
1332 if (mp == md) {
1333 zeroing = 0;
1334 adjust = dpos - mtod(mp, caddr_t);
1335 mp->m_len -= adjust;
1336 if (mp->m_len > 0 && adjust > 0)
1337 mp->m_data += adjust;
1338 }
1339 if (zeroing)
1340 mp->m_len = 0;
1341 else {
1342 i += mp->m_len;
1343 if (i > len) {
1344 mp->m_len -= (i - len);
1345 zeroing = 1;
1346 }
1347 }
1348 mp = mp->m_next;
1349 }
1350 if (len > NFS_MAXDATA || len < 0 || i < len) {
1351nfsmout:
1352 m_freem(mrep);
1353 error = EIO;
1354 nfsm_writereply(2 * NFSX_UNSIGNED);
1355 if (v3)
1356 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1357 nfsd->nd_mreq = mreq;
1358 nfsd->nd_mrep = NULL;
1359 nfsd->nd_time = 0;
1360 }
1361
1362 /*
1363 * Add this entry to the hash and time queues.
1364 */
1365 s = splsoftclock();
1366 owp = NULL;
1367 wp = LIST_FIRST(&slp->ns_tq);
1368 while (wp && wp->nd_time < nfsd->nd_time) {
1369 owp = wp;
1370 wp = LIST_NEXT(wp, nd_tq);
1371 }
1372 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1373 if (owp) {
1374 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1375 } else {
1376 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1377 }
1378 if (nfsd->nd_mrep) {
1379 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1380 owp = NULL;
1381 wp = LIST_FIRST(wpp);
1382 while (wp &&
1383 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)){
1384 owp = wp;
1385 wp = LIST_NEXT(wp, nd_hash);
1386 }
1387 while (wp && wp->nd_off < nfsd->nd_off &&
1388 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1389 owp = wp;
1390 wp = LIST_NEXT(wp, nd_hash);
1391 }
1392 if (owp) {
1393 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1394
1395 /*
1396 * Search the hash list for overlapping entries and
1397 * coalesce.
1398 */
1399 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1400 wp = LIST_NEXT(nfsd, nd_hash);
1401 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1402 nfsrvw_coalesce(owp, nfsd);
1403 }
1404 } else {
1405 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1406 }
1407 }
1408 splx(s);
1409 }
1410
1411 /*
1412 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1413 * and generate the associated reply mbuf list(s).
1414 */
1415loop1:
1416 cur_usec = nfs_curusec();
1417 s = splsoftclock();
1418 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1419 owp = LIST_NEXT(nfsd, nd_tq);
1420 if (nfsd->nd_time > cur_usec)
1421 break;
1422 if (nfsd->nd_mreq)
1423 continue;
1424 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1425 LIST_REMOVE(nfsd, nd_tq);
1426 LIST_REMOVE(nfsd, nd_hash);
1427 splx(s);
1428 mrep = nfsd->nd_mrep;
1429 nfsd->nd_mrep = NULL;
1430 cred = nfsd->nd_cr;
1431 v3 = (nfsd->nd_flag & ND_NFSV3);
1432 forat_ret = aftat_ret = 1;
1433 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, &vfslocked, cred,
1434 slp, nfsd->nd_nam, &rdonly, TRUE);
1435 if (!error) {
1436 if (v3)
1437 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1438 if (vp->v_type != VREG) {
1439 if (v3)
1440 error = EINVAL;
1441 else
1442 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1443 }
1444 } else {
1445 vp = NULL;
1446 }
1447 if (!error)
1448 error = nfsrv_access(vp, VWRITE, cred, rdonly,
1449 td, 1);
1450 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1451 ioflags = IO_NODELOCKED;
1452 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1453 ioflags = (IO_SYNC | IO_NODELOCKED);
1454 else
1455 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1456 uiop->uio_rw = UIO_WRITE;
1457 uiop->uio_segflg = UIO_SYSSPACE;
1458 uiop->uio_td = NULL;
1459 uiop->uio_offset = nfsd->nd_off;
1460 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1461 if (uiop->uio_resid > 0) {
1462 mp = mrep;
1463 i = 0;
1464 while (mp) {
1465 if (mp->m_len > 0)
1466 i++;
1467 mp = mp->m_next;
1468 }
1469 uiop->uio_iovcnt = i;
1470 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1471 M_TEMP, M_WAITOK);
1472 uiop->uio_iov = ivp = iov;
1473 mp = mrep;
1474 while (mp) {
1475 if (mp->m_len > 0) {
1476 ivp->iov_base = mtod(mp, caddr_t);
1477 ivp->iov_len = mp->m_len;
1478 ivp++;
1479 }
1480 mp = mp->m_next;
1481 }
1482 mvfslocked = 0;
1483 if (!error) {
1484 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) {
1485 VOP_UNLOCK(vp, 0);
1486 error = vn_start_write(NULL, &mntp, V_WAIT);
1487 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1488 }
1489 mvfslocked = VFS_LOCK_GIANT(mntp);
1490 }
1491 if (!error) {
1492 error = VOP_WRITE(vp, uiop, ioflags, cred);
1493 /* XXXRW: unlocked write. */
1494 nfsrvstats.srvvop_writes++;
1495 vn_finished_write(mntp);
1496 }
1497 VFS_UNLOCK_GIANT(mvfslocked);
1498 FREE((caddr_t)iov, M_TEMP);
1499 }
1500 m_freem(mrep);
1501 if (vp) {
1502 aftat_ret = VOP_GETATTR(vp, &va, cred, td);
1503 vput(vp);
1504 vp = NULL;
1505 }
1506 VFS_UNLOCK_GIANT(vfslocked);
1507 /*
1508 * Loop around generating replies for all write rpcs that have
1509 * now been completed.
1510 */
1511 swp = nfsd;
1512 do {
1513 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1514 if (error) {
1515 nfsm_writereply(NFSX_WCCDATA(v3));
1516 if (v3) {
1517 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1518 }
1519 } else {
1520 nfsm_writereply(NFSX_PREOPATTR(v3) +
1521 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1522 NFSX_WRITEVERF(v3));
1523 if (v3) {
1524 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1525 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1526 *tl++ = txdr_unsigned(nfsd->nd_len);
1527 *tl++ = txdr_unsigned(swp->nd_stable);
1528 /*
1529 * Actually, there is no need to txdr these fields,
1530 * but it may make the values more human readable,
1531 * for debugging purposes.
1532 */
1533 if (nfsver.tv_sec == 0)
1534 nfsver = boottime;
1535 *tl++ = txdr_unsigned(nfsver.tv_sec);
1536 *tl = txdr_unsigned(nfsver.tv_usec);
1537 } else {
1538 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1539 nfsm_srvfillattr(&va, fp);
1540 }
1541 }
1542 nfsd->nd_mreq = mreq;
1543 if (nfsd->nd_mrep)
1544 panic("nfsrv_write: nd_mrep not free");
1545
1546 /*
1547 * Done. Put it at the head of the timer queue so that
1548 * the final phase can return the reply.
1549 */
1550 s = splsoftclock();
1551 if (nfsd != swp) {
1552 nfsd->nd_time = 0;
1553 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1554 }
1555 nfsd = LIST_FIRST(&swp->nd_coalesce);
1556 if (nfsd) {
1557 LIST_REMOVE(nfsd, nd_tq);
1558 }
1559 splx(s);
1560 } while (nfsd);
1561 s = splsoftclock();
1562 swp->nd_time = 0;
1563 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1564 splx(s);
1565 goto loop1;
1566 }
1567 splx(s);
1568
1569 /*
1570 * Search for a reply to return.
1571 */
1572 s = splsoftclock();
1573 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq)
1574 if (nfsd->nd_mreq) {
1575 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1576 LIST_REMOVE(nfsd, nd_tq);
1577 *mrq = nfsd->nd_mreq;
1578 *ndp = nfsd;
1579 break;
1580 }
1581 splx(s);
1582 return (0);
1583}
1584
1585/*
1586 * Coalesce the write request nfsd into owp. To do this we must:
1587 * - remove nfsd from the queues
1588 * - merge nfsd->nd_mrep into owp->nd_mrep
1589 * - update the nd_eoff and nd_stable for owp
1590 * - put nfsd on owp's nd_coalesce list
1591 * NB: Must be called at splsoftclock().
1592 */
1593static void
1594nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1595{
1596 int overlap;
1597 struct mbuf *mp;
1598 struct nfsrv_descript *p;
1599
1600 NFS_DPF(WG, ("C%03x-%03x",
1601 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1602 LIST_REMOVE(nfsd, nd_hash);
1603 LIST_REMOVE(nfsd, nd_tq);
1604 if (owp->nd_eoff < nfsd->nd_eoff) {
1605 overlap = owp->nd_eoff - nfsd->nd_off;
1606 if (overlap < 0)
1607 panic("nfsrv_coalesce: bad off");
1608 if (overlap > 0)
1609 m_adj(nfsd->nd_mrep, overlap);
1610 mp = owp->nd_mrep;
1611 while (mp->m_next)
1612 mp = mp->m_next;
1613 mp->m_next = nfsd->nd_mrep;
1614 owp->nd_eoff = nfsd->nd_eoff;
1615 } else
1616 m_freem(nfsd->nd_mrep);
1617 nfsd->nd_mrep = NULL;
1618 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1619 owp->nd_stable = NFSV3WRITE_FILESYNC;
1620 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1621 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1622 owp->nd_stable = NFSV3WRITE_DATASYNC;
1623 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1624
1625 /*
1626 * If nfsd had anything else coalesced into it, transfer them
1627 * to owp, otherwise their replies will never get sent.
1628 */
1629 for (p = LIST_FIRST(&nfsd->nd_coalesce); p;
1630 p = LIST_FIRST(&nfsd->nd_coalesce)) {
1631 LIST_REMOVE(p, nd_tq);
1632 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1633 }
1634}
1635
1636/*
1637 * nfs create service
1638 * now does a truncate to 0 length via. setattr if it already exists
1639 */
1640int
1641nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1642 struct thread *td, struct mbuf **mrq)
1643{
1644 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1645 struct sockaddr *nam = nfsd->nd_nam;
1646 caddr_t dpos = nfsd->nd_dpos;
1647 struct ucred *cred = nfsd->nd_cr;
1648 struct nfs_fattr *fp;
1649 struct vattr va, dirfor, diraft;
1650 struct vattr *vap = &va;
1651 struct nfsv2_sattr *sp;
1652 u_int32_t *tl;
1653 struct nameidata nd;
1654 caddr_t bpos;
1655 int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1656 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1657 caddr_t cp;
1658 struct mbuf *mb, *mreq;
1659 struct vnode *dirp = NULL;
1660 nfsfh_t nfh;
1661 fhandle_t *fhp;
1662 u_quad_t tempsize;
1663 u_char cverf[NFSX_V3CREATEVERF];
1664 struct mount *mp = NULL;
1665 int tvfslocked;
1666 int vfslocked;
1667
1668 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1669 vfslocked = 0;
1670#ifndef nolint
1671 rdev = 0;
1672#endif
1673 ndclear(&nd);
1674
1675 fhp = &nfh.fh_generic;
1676 nfsm_srvmtofh(fhp);
1677 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1678 error = ESTALE;
1679 goto ereply;
1680 }
1681 vfslocked = VFS_LOCK_GIANT(mp);
1682 (void) vn_start_write(NULL, &mp, V_WAIT);
1683 vfs_rel(mp); /* The write holds a ref. */
1684 nfsm_srvnamesiz(len);
1685
1686 nd.ni_cnd.cn_cred = cred;
1687 nd.ni_cnd.cn_nameiop = CREATE;
1688 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1689
1690 /*
1691 * Call namei and do initial cleanup to get a few things
1692 * out of the way. If we get an initial error we cleanup
1693 * and return here to avoid special-casing the invalid nd
1694 * structure through the rest of the case. dirp may be
1695 * set even if an error occurs, but the nd structure will not
1696 * be valid at all if an error occurs so we have to invalidate it
1697 * prior to calling nfsm_reply ( which might goto nfsmout ).
1698 */
1699 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1700 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1701 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1702 if (dirp && !v3) {
1703 vrele(dirp);
1704 dirp = NULL;
1705 }
1706 if (error) {
1707 nfsm_reply(NFSX_WCCDATA(v3));
1708 if (v3)
1709 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1710 error = 0;
1711 goto nfsmout;
1712 }
1713
1714 /*
1715 * No error. Continue. State:
1716 *
1717 * startdir is valid ( we release this immediately )
1718 * dirp may be valid
1719 * nd.ni_vp may be valid
1720 * nd.ni_dvp is valid
1721 *
1722 * The error state is set through the code and we may also do some
1723 * opportunistic releasing of vnodes to avoid holding locks through
1724 * NFS I/O. The cleanup at the end is a catch-all
1725 */
1726
1727 VATTR_NULL(vap);
1728 if (v3) {
1729 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1730 how = fxdr_unsigned(int, *tl);
1731 switch (how) {
1732 case NFSV3CREATE_GUARDED:
1733 if (nd.ni_vp) {
1734 error = EEXIST;
1735 break;
1736 }
1737 /* fall through */
1738 case NFSV3CREATE_UNCHECKED:
1739 nfsm_srvsattr(vap);
1740 break;
1741 case NFSV3CREATE_EXCLUSIVE:
1742 cp = nfsm_dissect_nonblock(caddr_t, NFSX_V3CREATEVERF);
1743 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1744 exclusive_flag = 1;
1745 break;
1746 };
1747 vap->va_type = VREG;
1748 } else {
1749 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
1750 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1751 if (vap->va_type == VNON)
1752 vap->va_type = VREG;
1753 vap->va_mode = nfstov_mode(sp->sa_mode);
1754 switch (vap->va_type) {
1755 case VREG:
1756 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1757 if (tsize != -1)
1758 vap->va_size = (u_quad_t)tsize;
1759 break;
1760 case VCHR:
1761 case VBLK:
1762 case VFIFO:
1763 rdev = fxdr_unsigned(long, sp->sa_size);
1764 break;
1765 default:
1766 break;
1767 };
1768 }
1769
1770 /*
1771 * Iff doesn't exist, create it
1772 * otherwise just truncate to 0 length
1773 * should I set the mode too ?
1774 *
1775 * The only possible error we can have at this point is EEXIST.
1776 * nd.ni_vp will also be non-NULL in that case.
1777 */
1778 if (nd.ni_vp == NULL) {
1779 if (vap->va_mode == (mode_t)VNOVAL)
1780 vap->va_mode = 0;
1781 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1782 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1783 if (error)
1784 NDFREE(&nd, NDF_ONLY_PNBUF);
1785 else {
1786 if (exclusive_flag) {
1787 exclusive_flag = 0;
1788 VATTR_NULL(vap);
1789 bcopy(cverf, (caddr_t)&vap->va_atime,
1790 NFSX_V3CREATEVERF);
1791 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1792 td);
1793 }
1794 }
1795 } else if (vap->va_type == VCHR || vap->va_type == VBLK ||
1796 vap->va_type == VFIFO) {
1797 /*
1798 * NFSv2-specific code for creating device nodes
1799 * and fifos.
1800 *
1801 * Handle SysV FIFO node special cases. All other
1802 * devices require super user to access.
1803 */
1804 if (vap->va_type == VCHR && rdev == 0xffffffff)
1805 vap->va_type = VFIFO;
1806 if (vap->va_type != VFIFO &&
1807 (error = suser_cred(cred, 0))) {
1808 goto ereply;
1809 }
1810 vap->va_rdev = rdev;
1811 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1812 if (error) {
1813 NDFREE(&nd, NDF_ONLY_PNBUF);
1814 goto ereply;
1815 }
1816 vput(nd.ni_vp);
1817 nd.ni_vp = NULL;
1818
1819 /*
1820 * release dvp prior to lookup
1821 */
1822 vput(nd.ni_dvp);
1823 nd.ni_dvp = NULL;
1824 /*
1825 * Setup for lookup.
1826 *
1827 * Even though LOCKPARENT was cleared, ni_dvp may
1828 * be garbage.
1829 */
1830 nd.ni_cnd.cn_nameiop = LOOKUP;
1831 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1832 nd.ni_cnd.cn_thread = td;
1833 nd.ni_cnd.cn_cred = cred;
1834 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
1835 if (tvfslocked)
1836 nd.ni_cnd.cn_flags |= GIANTHELD;
1837 error = lookup(&nd);
1838 nd.ni_dvp = NULL;
1839 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1840 nd.ni_cnd.cn_flags &= ~GIANTHELD;
1841 if (error)
1842 goto ereply;
1843
1844 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1845 error = EINVAL;
1846 goto ereply;
1847 }
1848 } else {
1849 error = ENXIO;
1850 }
1851 } else {
1852 if (vap->va_size != -1) {
1853 error = nfsrv_access(nd.ni_vp, VWRITE,
1854 cred, (nd.ni_cnd.cn_flags & RDONLY), td, 0);
1855 if (!error) {
1856 tempsize = vap->va_size;
1857 VATTR_NULL(vap);
1858 vap->va_size = tempsize;
1859 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1860 td);
1861 }
1862 }
1863 }
1864
1865 if (!error) {
1866 bzero((caddr_t)fhp, sizeof(nfh));
1867 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1868 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1869 if (!error)
1870 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
1871 }
1872 if (v3) {
1873 if (exclusive_flag && !error &&
1874 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1875 error = EEXIST;
1876 if (dirp == nd.ni_dvp)
1877 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1878 else {
1879 /* Drop the other locks to avoid deadlock. */
1880 if (nd.ni_dvp) {
1881 if (nd.ni_dvp == nd.ni_vp)
1882 vrele(nd.ni_dvp);
1883 else
1884 vput(nd.ni_dvp);
1885 }
1886 if (nd.ni_vp)
1887 vput(nd.ni_vp);
1888 nd.ni_dvp = NULL;
1889 nd.ni_vp = NULL;
1890
1891 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1892 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1893 VOP_UNLOCK(dirp, 0);
1894 }
1895 }
1896ereply:
1897 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1898 if (v3) {
1899 if (!error) {
1900 nfsm_srvpostop_fh(fhp);
1901 nfsm_srvpostop_attr(0, vap);
1902 }
1903 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1904 } else if (!error) {
1905 /* v2 non-error case. */
1906 nfsm_srvfhtom(fhp, v3);
1907 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1908 nfsm_srvfillattr(vap, fp);
1909 }
1910 error = 0;
1911
1912nfsmout:
1913 if (nd.ni_dvp) {
1914 if (nd.ni_dvp == nd.ni_vp)
1915 vrele(nd.ni_dvp);
1916 else
1917 vput(nd.ni_dvp);
1918 }
1919 if (nd.ni_vp)
1920 vput(nd.ni_vp);
1921 if (nd.ni_startdir) {
1922 vrele(nd.ni_startdir);
1923 nd.ni_startdir = NULL;
1924 }
1925 if (dirp)
1926 vrele(dirp);
1927 NDFREE(&nd, NDF_ONLY_PNBUF);
1928 vn_finished_write(mp);
1929 VFS_UNLOCK_GIANT(vfslocked);
1930 return (error);
1931}
1932
1933/*
1934 * nfs v3 mknod service
1935 */
1936int
1937nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1938 struct thread *td, struct mbuf **mrq)
1939{
1940 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1941 struct sockaddr *nam = nfsd->nd_nam;
1942 caddr_t dpos = nfsd->nd_dpos;
1943 struct ucred *cred = nfsd->nd_cr;
1944 struct vattr va, dirfor, diraft;
1945 struct vattr *vap = &va;
1946 u_int32_t *tl;
1947 struct nameidata nd;
1948 caddr_t bpos;
1949 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1950 u_int32_t major, minor;
1951 enum vtype vtyp;
1952 struct mbuf *mb, *mreq;
1953 struct vnode *vp, *dirp = NULL;
1954 nfsfh_t nfh;
1955 fhandle_t *fhp;
1956 struct mount *mp = NULL;
1957 int v3 = (nfsd->nd_flag & ND_NFSV3);
1958 int tvfslocked;
1959 int vfslocked;
1960
1961 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1962 vfslocked = 0;
1963 if (!v3)
1964 panic("nfsrv_mknod: v3 proc called on a v2 connection");
1965 ndclear(&nd);
1966
1967 fhp = &nfh.fh_generic;
1968 nfsm_srvmtofh(fhp);
1969 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1970 error = ESTALE;
1971 goto ereply;
1972 }
1973 vfslocked = VFS_LOCK_GIANT(mp);
1974 (void) vn_start_write(NULL, &mp, V_WAIT);
1975 vfs_rel(mp); /* The write holds a ref. */
1976 nfsm_srvnamesiz(len);
1977
1978 nd.ni_cnd.cn_cred = cred;
1979 nd.ni_cnd.cn_nameiop = CREATE;
1980 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1981
1982 /*
1983 * Handle nfs_namei() call. If an error occurs, the nd structure
1984 * is not valid. However, nfsm_*() routines may still jump to
1985 * nfsmout.
1986 */
1987
1988 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1989 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1990 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1991 if (error) {
1992 nfsm_reply(NFSX_WCCDATA(1));
1993 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1994 error = 0;
1995 goto nfsmout;
1996 }
1997 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1998 vtyp = nfsv3tov_type(*tl);
1999 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
2000 error = NFSERR_BADTYPE;
2001 goto out;
2002 }
2003 VATTR_NULL(vap);
2004 nfsm_srvsattr(vap);
2005 if (vtyp == VCHR || vtyp == VBLK) {
2006 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
2007 major = fxdr_unsigned(u_int32_t, *tl++);
2008 minor = fxdr_unsigned(u_int32_t, *tl);
2009 vap->va_rdev = makedev(major, minor);
2010 }
2011
2012 /*
2013 * Iff doesn't exist, create it.
2014 */
2015 if (nd.ni_vp) {
2016 error = EEXIST;
2017 goto out;
2018 }
2019 vap->va_type = vtyp;
2020 if (vap->va_mode == (mode_t)VNOVAL)
2021 vap->va_mode = 0;
2022 if (vtyp == VSOCK) {
2023 vrele(nd.ni_startdir);
2024 nd.ni_startdir = NULL;
2025 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2026 if (error)
2027 NDFREE(&nd, NDF_ONLY_PNBUF);
2028 } else {
2029 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
2030 goto out;
2031 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2032 if (error) {
2033 NDFREE(&nd, NDF_ONLY_PNBUF);
2034 goto out;
2035 }
2036 vput(nd.ni_vp);
2037 nd.ni_vp = NULL;
2038
2039 /*
2040 * Release dvp prior to lookup
2041 */
2042 vput(nd.ni_dvp);
2043 nd.ni_dvp = NULL;
2044
2045 nd.ni_cnd.cn_nameiop = LOOKUP;
2046 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
2047 nd.ni_cnd.cn_thread = td;
2048 nd.ni_cnd.cn_cred = td->td_ucred;
2049 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2050 if (tvfslocked)
2051 nd.ni_cnd.cn_flags |= GIANTHELD;
2052 error = lookup(&nd);
2053 nd.ni_dvp = NULL;
2054 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2055 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2056
2057 if (error)
2058 goto out;
2059 if (nd.ni_cnd.cn_flags & ISSYMLINK)
2060 error = EINVAL;
2061 }
2062
2063 /*
2064 * send response, cleanup, return.
2065 */
2066out:
2067 vp = nd.ni_vp;
2068 if (!error) {
2069 bzero((caddr_t)fhp, sizeof(nfh));
2070 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2071 error = VOP_VPTOFH(vp, &fhp->fh_fid);
2072 if (!error)
2073 error = VOP_GETATTR(vp, vap, cred, td);
2074 }
2075 if (nd.ni_dvp) {
2076 if (nd.ni_dvp == nd.ni_vp)
2077 vrele(nd.ni_dvp);
2078 else
2079 vput(nd.ni_dvp);
2080 nd.ni_dvp = NULL;
2081 }
2082 if (vp) {
2083 vput(vp);
2084 vp = NULL;
2085 nd.ni_vp = NULL;
2086 }
2087 if (nd.ni_startdir) {
2088 vrele(nd.ni_startdir);
2089 nd.ni_startdir = NULL;
2090 }
2091 NDFREE(&nd, NDF_ONLY_PNBUF);
2092 if (dirp) {
2093 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2094 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2095 VOP_UNLOCK(dirp, 0);
2096 }
2097ereply:
2098 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
2099 if (v3) {
2100 if (!error) {
2101 nfsm_srvpostop_fh(fhp);
2102 nfsm_srvpostop_attr(0, vap);
2103 }
2104 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2105 }
2106 vn_finished_write(mp);
2107 VFS_UNLOCK_GIANT(vfslocked);
2108 return (0);
2109nfsmout:
2110 if (nd.ni_dvp) {
2111 if (nd.ni_dvp == nd.ni_vp)
2112 vrele(nd.ni_dvp);
2113 else
2114 vput(nd.ni_dvp);
2115 }
2116 if (nd.ni_vp)
2117 vput(nd.ni_vp);
2118 if (dirp)
2119 vrele(dirp);
2120 if (nd.ni_startdir)
2121 vrele(nd.ni_startdir);
2122 NDFREE(&nd, NDF_ONLY_PNBUF);
2123 vn_finished_write(mp);
2124 VFS_UNLOCK_GIANT(vfslocked);
2125 return (error);
2126}
2127
2128/*
2129 * nfs remove service
2130 */
2131int
2132nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2133 struct thread *td, struct mbuf **mrq)
2134{
2135 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2136 struct sockaddr *nam = nfsd->nd_nam;
2137 caddr_t dpos = nfsd->nd_dpos;
2138 struct ucred *cred = nfsd->nd_cr;
2139 struct nameidata nd;
2140 caddr_t bpos;
2141 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2142 int v3 = (nfsd->nd_flag & ND_NFSV3);
2143 struct mbuf *mb, *mreq;
2144 struct vnode *dirp;
2145 struct vattr dirfor, diraft;
2146 nfsfh_t nfh;
2147 fhandle_t *fhp;
2148 struct mount *mp = NULL;
2149 int vfslocked, vfslocked1;
2150
2151 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2152 ndclear(&nd);
2153 vfslocked = 0;
2154
2155 fhp = &nfh.fh_generic;
2156 nfsm_srvmtofh(fhp);
2157 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2158 error = ESTALE;
2159 goto ereply;
2160 }
2161 vfslocked = VFS_LOCK_GIANT(mp);
2162 (void) vn_start_write(NULL, &mp, V_WAIT);
2163 vfs_rel(mp); /* The write holds a ref. */
2164 nfsm_srvnamesiz(len);
2165
2166 nd.ni_cnd.cn_cred = cred;
2167 nd.ni_cnd.cn_nameiop = DELETE;
2168 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
2169 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2170 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2171 vfslocked1 = NDHASGIANT(&nd);
2172 if (vfslocked && vfslocked1)
2173 VFS_UNLOCK_GIANT(vfslocked1);
2174 if (vfslocked || vfslocked1)
2175 vfslocked = 1;
2176 if (dirp && !v3) {
2177 vrele(dirp);
2178 dirp = NULL;
2179 }
2180 if (error == 0) {
2181 if (nd.ni_vp->v_type == VDIR) {
2182 error = EPERM; /* POSIX */
2183 goto out;
2184 }
2185 /*
2186 * The root of a mounted filesystem cannot be deleted.
2187 */
2188 if (nd.ni_vp->v_vflag & VV_ROOT) {
2189 error = EBUSY;
2190 goto out;
2191 }
2192out:
2193 if (!error) {
2194 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2195 NDFREE(&nd, NDF_ONLY_PNBUF);
2196 }
2197 }
2198 if (dirp && v3) {
2199 if (dirp == nd.ni_dvp)
2200 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2201 else {
2202 /* Drop the other locks to avoid deadlock. */
2203 if (nd.ni_dvp) {
2204 if (nd.ni_dvp == nd.ni_vp)
2205 vrele(nd.ni_dvp);
2206 else
2207 vput(nd.ni_dvp);
2208 }
2209 if (nd.ni_vp)
2210 vput(nd.ni_vp);
2211 nd.ni_dvp = NULL;
2212 nd.ni_vp = NULL;
2213
2214 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2215 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2216 VOP_UNLOCK(dirp, 0);
2217 }
2218 vrele(dirp);
2219 dirp = NULL;
2220 }
2221ereply:
2222 nfsm_reply(NFSX_WCCDATA(v3));
2223 if (v3) {
2224 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2225 error = 0;
2226 }
2227nfsmout:
2228 NDFREE(&nd, NDF_ONLY_PNBUF);
2229 if (nd.ni_dvp) {
2230 if (nd.ni_dvp == nd.ni_vp)
2231 vrele(nd.ni_dvp);
2232 else
2233 vput(nd.ni_dvp);
2234 }
2235 if (nd.ni_vp)
2236 vput(nd.ni_vp);
2237 vn_finished_write(mp);
2238 VFS_UNLOCK_GIANT(vfslocked);
2239 return(error);
2240}
2241
2242/*
2243 * nfs rename service
2244 */
2245int
2246nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2247 struct thread *td, struct mbuf **mrq)
2248{
2249 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2250 struct sockaddr *nam = nfsd->nd_nam;
2251 caddr_t dpos = nfsd->nd_dpos;
2252 struct ucred *cred = nfsd->nd_cr;
2253 caddr_t bpos;
2254 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2255 int tdirfor_ret = 1, tdiraft_ret = 1;
2256 int v3 = (nfsd->nd_flag & ND_NFSV3);
2257 struct mbuf *mb, *mreq;
2258 struct nameidata fromnd, tond;
2259 struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL;
2260 struct vnode *tdirp = NULL;
2261 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2262 nfsfh_t fnfh, tnfh;
2263 fhandle_t *ffhp, *tfhp;
2264 uid_t saved_uid;
2265 struct mount *mp = NULL;
2266 int vfslocked;
2267
2268 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2269 vfslocked = 0;
2270#ifndef nolint
2271 fvp = NULL;
2272#endif
2273 ffhp = &fnfh.fh_generic;
2274 tfhp = &tnfh.fh_generic;
2275
2276 /*
2277 * Clear fields incase goto nfsmout occurs from macro.
2278 */
2279
2280 ndclear(&fromnd);
2281 ndclear(&tond);
2282
2283 nfsm_srvmtofh(ffhp);
2284 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
2285 error = ESTALE;
2286 goto out1;
2287 }
2288 vfslocked = VFS_LOCK_GIANT(mp);
2289 (void) vn_start_write(NULL, &mp, V_WAIT);
2290 vfs_rel(mp); /* The write holds a ref. */
2291 nfsm_srvnamesiz(len);
2292 /*
2293 * Remember our original uid so that we can reset cr_uid before
2294 * the second nfs_namei() call, in case it is remapped.
2295 */
2296 saved_uid = cred->cr_uid;
2297 fromnd.ni_cnd.cn_cred = cred;
2298 fromnd.ni_cnd.cn_nameiop = DELETE;
2299 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART | MPSAFE;
2300 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
2301 &dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, td, FALSE);
2302 vfslocked = nfsrv_lockedpair_nd(vfslocked, &fromnd);
2303 if (fdirp && !v3) {
2304 vrele(fdirp);
2305 fdirp = NULL;
2306 }
2307 if (error) {
2308 nfsm_reply(2 * NFSX_WCCDATA(v3));
2309 if (v3) {
2310 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2311 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2312 }
2313 error = 0;
2314 goto nfsmout;
2315 }
2316 fvp = fromnd.ni_vp;
2317 nfsm_srvmtofh(tfhp);
2318 nfsm_srvnamesiz(len2);
2319 cred->cr_uid = saved_uid;
2320 tond.ni_cnd.cn_cred = cred;
2321 tond.ni_cnd.cn_nameiop = RENAME;
2322 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | MPSAFE;
2323 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
2324 &dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, td, FALSE);
2325 vfslocked = nfsrv_lockedpair_nd(vfslocked, &tond);
2326 if (tdirp && !v3) {
2327 vrele(tdirp);
2328 tdirp = NULL;
2329 }
2330 if (error)
2331 goto out1;
2332
2333 tdvp = tond.ni_dvp;
2334 tvp = tond.ni_vp;
2335 if (tvp != NULL) {
2336 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2337 if (v3)
2338 error = EEXIST;
2339 else
2340 error = EISDIR;
2341 goto out;
2342 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2343 if (v3)
2344 error = EEXIST;
2345 else
2346 error = ENOTDIR;
2347 goto out;
2348 }
2349 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2350 if (v3)
2351 error = EXDEV;
2352 else
2353 error = ENOTEMPTY;
2354 goto out;
2355 }
2356 }
2357 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2358 if (v3)
2359 error = EXDEV;
2360 else
2361 error = ENOTEMPTY;
2362 goto out;
2363 }
2364 if (fvp->v_mount != tdvp->v_mount) {
2365 if (v3)
2366 error = EXDEV;
2367 else
2368 error = ENOTEMPTY;
2369 goto out;
2370 }
2371 if (fvp == tdvp) {
2372 if (v3)
2373 error = EINVAL;
2374 else
2375 error = ENOTEMPTY;
2376 }
2377 /*
2378 * If source is the same as the destination (that is the
2379 * same vnode with the same name in the same directory),
2380 * then there is nothing to do.
2381 */
2382 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2383 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2384 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2385 fromnd.ni_cnd.cn_namelen))
2386 error = -1;
2387out:
2388 if (!error) {
2389 /*
2390 * The VOP_RENAME function releases all vnode references &
2391 * locks prior to returning so we need to clear the pointers
2392 * to bypass cleanup code later on.
2393 */
2394 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2395 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2396 fromnd.ni_dvp = NULL;
2397 fromnd.ni_vp = NULL;
2398 tond.ni_dvp = NULL;
2399 tond.ni_vp = NULL;
2400 if (error) {
2401 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2402 NDFREE(&tond, NDF_ONLY_PNBUF);
2403 }
2404 } else {
2405 if (error == -1)
2406 error = 0;
2407 }
2408 /* fall through */
2409out1:
2410 nfsm_reply(2 * NFSX_WCCDATA(v3));
2411 if (v3) {
2412 /* Release existing locks to prevent deadlock. */
2413 if (tond.ni_dvp) {
2414 if (tond.ni_dvp == tond.ni_vp)
2415 vrele(tond.ni_dvp);
2416 else
2417 vput(tond.ni_dvp);
2418 }
2419 if (tond.ni_vp)
2420 vput(tond.ni_vp);
2421 tond.ni_dvp = NULL;
2422 tond.ni_vp = NULL;
2423
2424 if (fdirp) {
2425 vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY);
2426 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, td);
2427 VOP_UNLOCK(fdirp, 0);
2428 }
2429 if (tdirp) {
2430 vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY);
2431 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, td);
2432 VOP_UNLOCK(tdirp, 0);
2433 }
2434 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2435 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2436 }
2437 error = 0;
2438 /* fall through */
2439
2440nfsmout:
2441 /*
2442 * Clear out tond related fields
2443 */
2444 if (tond.ni_dvp) {
2445 if (tond.ni_dvp == tond.ni_vp)
2446 vrele(tond.ni_dvp);
2447 else
2448 vput(tond.ni_dvp);
2449 }
2450 if (tond.ni_vp)
2451 vput(tond.ni_vp);
2452 if (tdirp)
2453 vrele(tdirp);
2454 if (tond.ni_startdir)
2455 vrele(tond.ni_startdir);
2456 NDFREE(&tond, NDF_ONLY_PNBUF);
2457 /*
2458 * Clear out fromnd related fields
2459 */
2460 if (fdirp)
2461 vrele(fdirp);
2462 if (fromnd.ni_startdir)
2463 vrele(fromnd.ni_startdir);
2464 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2465 if (fromnd.ni_dvp)
2466 vrele(fromnd.ni_dvp);
2467 if (fromnd.ni_vp)
2468 vrele(fromnd.ni_vp);
2469
2470 vn_finished_write(mp);
2471 VFS_UNLOCK_GIANT(vfslocked);
2472 return (error);
2473}
2474
2475/*
2476 * nfs link service
2477 */
2478int
2479nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2480 struct thread *td, struct mbuf **mrq)
2481{
2482 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2483 struct sockaddr *nam = nfsd->nd_nam;
2484 caddr_t dpos = nfsd->nd_dpos;
2485 struct ucred *cred = nfsd->nd_cr;
2486 struct nameidata nd;
2487 caddr_t bpos;
2488 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2489 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2490 struct mbuf *mb, *mreq;
2491 struct vnode *vp = NULL, *xp, *dirp = NULL;
2492 struct vattr dirfor, diraft, at;
2493 nfsfh_t nfh, dnfh;
2494 fhandle_t *fhp, *dfhp;
2495 struct mount *mp = NULL;
2496 int tvfslocked;
2497 int vfslocked;
2498
2499 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2500 ndclear(&nd);
2501 vfslocked = 0;
2502
2503 fhp = &nfh.fh_generic;
2504 dfhp = &dnfh.fh_generic;
2505 nfsm_srvmtofh(fhp);
2506 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2507 error = ESTALE;
2508 goto ereply;
2509 }
2510 vfslocked = VFS_LOCK_GIANT(mp);
2511 (void) vn_start_write(NULL, &mp, V_WAIT);
2512 vfs_rel(mp); /* The write holds a ref. */
2513 nfsm_srvmtofh(dfhp);
2514 nfsm_srvnamesiz(len);
2515
2516 error = nfsrv_fhtovp(fhp, TRUE, &vp, &tvfslocked, cred, slp,
2517 nam, &rdonly, TRUE);
2518 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
2519 if (error) {
2520 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2521 if (v3) {
2522 nfsm_srvpostop_attr(getret, &at);
2523 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2524 }
2525 vp = NULL;
2526 error = 0;
2527 goto nfsmout;
2528 }
2529 if (v3)
2530 getret = VOP_GETATTR(vp, &at, cred, td);
2531 if (vp->v_type == VDIR) {
2532 error = EPERM; /* POSIX */
2533 goto out1;
2534 }
2535 VOP_UNLOCK(vp, 0);
2536 nd.ni_cnd.cn_cred = cred;
2537 nd.ni_cnd.cn_nameiop = CREATE;
2538 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE | MPSAFE;
2539 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2540 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2541 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2542 if (dirp && !v3) {
2543 vrele(dirp);
2544 dirp = NULL;
2545 }
2546 if (error) {
2547 vrele(vp);
2548 vp = NULL;
2549 goto out2;
2550 }
2551 xp = nd.ni_vp;
2552 if (xp != NULL) {
2553 error = EEXIST;
2554 vrele(vp);
2555 vp = NULL;
2556 goto out2;
2557 }
2558 xp = nd.ni_dvp;
2559 if (vp->v_mount != xp->v_mount) {
2560 error = EXDEV;
2561 vrele(vp);
2562 vp = NULL;
2563 goto out2;
2564 }
2565 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2566 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2567 NDFREE(&nd, NDF_ONLY_PNBUF);
2568 /* fall through */
2569
2570out1:
2571 if (v3)
2572 getret = VOP_GETATTR(vp, &at, cred, td);
2573out2:
2574 if (dirp) {
2575 if (dirp == nd.ni_dvp)
2576 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2577 else {
2578 /* Release existing locks to prevent deadlock. */
2579 if (nd.ni_dvp) {
2580 if (nd.ni_dvp == nd.ni_vp)
2581 vrele(nd.ni_dvp);
2582 else
2583 vput(nd.ni_dvp);
2584 }
2585 if (nd.ni_vp)
2586 vrele(nd.ni_vp);
2587 nd.ni_dvp = NULL;
2588 nd.ni_vp = NULL;
2589
2590 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2591 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2592 VOP_UNLOCK(dirp, 0);
2593 }
2594 }
2595ereply:
2596 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2597 if (v3) {
2598 nfsm_srvpostop_attr(getret, &at);
2599 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2600 error = 0;
2601 }
2602 /* fall through */
2603
2604nfsmout:
2605 NDFREE(&nd, NDF_ONLY_PNBUF);
2606 if (vp)
2607 vput(vp);
2608 if (nd.ni_dvp) {
2609 if (nd.ni_dvp == nd.ni_vp)
2610 vrele(nd.ni_dvp);
2611 else
2612 vput(nd.ni_dvp);
2613 }
2614 if (dirp)
2615 vrele(dirp);
2616 if (nd.ni_vp)
2617 vrele(nd.ni_vp);
2618 vn_finished_write(mp);
2619 VFS_UNLOCK_GIANT(vfslocked);
2620 return(error);
2621}
2622
2623/*
2624 * nfs symbolic link service
2625 */
2626int
2627nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2628 struct thread *td, struct mbuf **mrq)
2629{
2630 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2631 struct sockaddr *nam = nfsd->nd_nam;
2632 caddr_t dpos = nfsd->nd_dpos;
2633 struct ucred *cred = nfsd->nd_cr;
2634 struct vattr va, dirfor, diraft;
2635 struct nameidata nd;
2636 struct vattr *vap = &va;
2637 struct nfsv2_sattr *sp;
2638 char *bpos, *pathcp = NULL;
2639 struct uio io;
2640 struct iovec iv;
2641 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2642 int v3 = (nfsd->nd_flag & ND_NFSV3);
2643 struct mbuf *mb, *mreq;
2644 struct vnode *dirp = NULL;
2645 nfsfh_t nfh;
2646 fhandle_t *fhp;
2647 struct mount *mp = NULL;
2648 int tvfslocked;
2649 int vfslocked;
2650
2651 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2652 ndclear(&nd);
2653 vfslocked = 0;
2654
2655 fhp = &nfh.fh_generic;
2656 nfsm_srvmtofh(fhp);
2657 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2658 error = ESTALE;
2659 goto out;
2660 }
2661 vfslocked = VFS_LOCK_GIANT(mp);
2662 (void) vn_start_write(NULL, &mp, V_WAIT);
2663 vfs_rel(mp); /* The write holds a ref. */
2664 nfsm_srvnamesiz(len);
2665 nd.ni_cnd.cn_cred = cred;
2666 nd.ni_cnd.cn_nameiop = CREATE;
2667 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART | MPSAFE;
2668 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2669 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2670 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2671 if (error == 0) {
2672 VATTR_NULL(vap);
2673 if (v3)
2674 nfsm_srvsattr(vap);
2675 nfsm_srvpathsiz(len2);
2676 }
2677 if (dirp && !v3) {
2678 vrele(dirp);
2679 dirp = NULL;
2680 }
2681 if (error)
2682 goto out;
2683 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2684 iv.iov_base = pathcp;
2685 iv.iov_len = len2;
2686 io.uio_resid = len2;
2687 io.uio_offset = 0;
2688 io.uio_iov = &iv;
2689 io.uio_iovcnt = 1;
2690 io.uio_segflg = UIO_SYSSPACE;
2691 io.uio_rw = UIO_READ;
2692 io.uio_td = NULL;
2693 nfsm_mtouio(&io, len2);
2694 if (!v3) {
2695 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
2696 vap->va_mode = nfstov_mode(sp->sa_mode);
2697 }
2698 *(pathcp + len2) = '\0';
2699 if (nd.ni_vp) {
2700 error = EEXIST;
2701 goto out;
2702 }
2703
2704 /*
2705 * issue symlink op. SAVESTART is set so the underlying path component
2706 * is only freed by the VOP if an error occurs.
2707 */
2708 if (vap->va_mode == (mode_t)VNOVAL)
2709 vap->va_mode = 0;
2710 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2711 if (error)
2712 NDFREE(&nd, NDF_ONLY_PNBUF);
2713 else
2714 vput(nd.ni_vp);
2715 nd.ni_vp = NULL;
2716 /*
2717 * releases directory prior to potential lookup op.
2718 */
2719 vput(nd.ni_dvp);
2720 nd.ni_dvp = NULL;
2721
2722 if (error == 0) {
2723 if (v3) {
2724 /*
2725 * Issue lookup. Leave SAVESTART set so we can easily free
2726 * the name buffer later on.
2727 *
2728 * since LOCKPARENT is not set, ni_dvp will be garbage on
2729 * return whether an error occurs or not.
2730 */
2731 nd.ni_cnd.cn_nameiop = LOOKUP;
2732 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2733 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2734 nd.ni_cnd.cn_thread = td;
2735 nd.ni_cnd.cn_cred = cred;
2736 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2737 if (tvfslocked)
2738 nd.ni_cnd.cn_flags |= GIANTHELD;
2739 error = lookup(&nd);
2740 nd.ni_dvp = NULL;
2741 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2742 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2743
2744 if (error == 0) {
2745 bzero((caddr_t)fhp, sizeof(nfh));
2746 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2747 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2748 if (!error)
2749 error = VOP_GETATTR(nd.ni_vp, vap, cred,
2750 td);
2751 vput(nd.ni_vp);
2752 nd.ni_vp = NULL;
2753 }
2754 }
2755 }
2756out:
2757 /*
2758 * These releases aren't strictly required, does even doing them
2759 * make any sense? XXX can nfsm_reply() block?
2760 */
2761 if (pathcp) {
2762 FREE(pathcp, M_TEMP);
2763 pathcp = NULL;
2764 }
2765 if (dirp) {
2766 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2767 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2768 VOP_UNLOCK(dirp, 0);
2769 }
2770 if (nd.ni_startdir) {
2771 vrele(nd.ni_startdir);
2772 nd.ni_startdir = NULL;
2773 }
2774 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2775 if (v3) {
2776 if (!error) {
2777 nfsm_srvpostop_fh(fhp);
2778 nfsm_srvpostop_attr(0, vap);
2779 }
2780 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2781 }
2782 error = 0;
2783 /* fall through */
2784
2785nfsmout:
2786 NDFREE(&nd, NDF_ONLY_PNBUF);
2787 if (nd.ni_dvp) {
2788 if (nd.ni_dvp == nd.ni_vp)
2789 vrele(nd.ni_dvp);
2790 else
2791 vput(nd.ni_dvp);
2792 }
2793 if (nd.ni_vp)
2794 vrele(nd.ni_vp);
2795 if (nd.ni_startdir)
2796 vrele(nd.ni_startdir);
2797 if (dirp)
2798 vrele(dirp);
2799 if (pathcp)
2800 FREE(pathcp, M_TEMP);
2801
2802 vn_finished_write(mp);
2803 VFS_UNLOCK_GIANT(vfslocked);
2804 return (error);
2805}
2806
2807/*
2808 * nfs mkdir service
2809 */
2810int
2811nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2812 struct thread *td, struct mbuf **mrq)
2813{
2814 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2815 struct sockaddr *nam = nfsd->nd_nam;
2816 caddr_t dpos = nfsd->nd_dpos;
2817 struct ucred *cred = nfsd->nd_cr;
2818 struct vattr va, dirfor, diraft;
2819 struct vattr *vap = &va;
2820 struct nfs_fattr *fp;
2821 struct nameidata nd;
2822 u_int32_t *tl;
2823 caddr_t bpos;
2824 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2825 int v3 = (nfsd->nd_flag & ND_NFSV3);
2826 struct mbuf *mb, *mreq;
2827 struct vnode *dirp = NULL;
2828 int vpexcl = 0;
2829 nfsfh_t nfh;
2830 fhandle_t *fhp;
2831 struct mount *mp = NULL;
2832 int vfslocked;
2833
2834 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2835 ndclear(&nd);
2836 vfslocked = 0;
2837
2838 fhp = &nfh.fh_generic;
2839 nfsm_srvmtofh(fhp);
2840 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2841 error = ESTALE;
2842 goto out;
2843 }
2844 vfslocked = VFS_LOCK_GIANT(mp);
2845 (void) vn_start_write(NULL, &mp, V_WAIT);
2846 vfs_rel(mp); /* The write holds a ref. */
2847 nfsm_srvnamesiz(len);
2848 nd.ni_cnd.cn_cred = cred;
2849 nd.ni_cnd.cn_nameiop = CREATE;
2850 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE;
2851
2852 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2853 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2854 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2855 if (dirp && !v3) {
2856 vrele(dirp);
2857 dirp = NULL;
2858 }
2859 if (error) {
2860 nfsm_reply(NFSX_WCCDATA(v3));
2861 if (v3)
2862 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2863 error = 0;
2864 goto nfsmout;
2865 }
2866 VATTR_NULL(vap);
2867 if (v3) {
2868 nfsm_srvsattr(vap);
2869 } else {
2870 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2871 vap->va_mode = nfstov_mode(*tl++);
2872 }
2873
2874 /*
2875 * At this point nd.ni_dvp is referenced and exclusively locked and
2876 * nd.ni_vp, if it exists, is referenced but not locked.
2877 */
2878
2879 vap->va_type = VDIR;
2880 if (nd.ni_vp != NULL) {
2881 NDFREE(&nd, NDF_ONLY_PNBUF);
2882 error = EEXIST;
2883 goto out;
2884 }
2885
2886 /*
2887 * Issue mkdir op. Since SAVESTART is not set, the pathname
2888 * component is freed by the VOP call. This will fill-in
2889 * nd.ni_vp, reference, and exclusively lock it.
2890 */
2891 if (vap->va_mode == (mode_t)VNOVAL)
2892 vap->va_mode = 0;
2893 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2894 NDFREE(&nd, NDF_ONLY_PNBUF);
2895 vpexcl = 1;
2896
2897 vput(nd.ni_dvp);
2898 nd.ni_dvp = NULL;
2899
2900 if (!error) {
2901 bzero((caddr_t)fhp, sizeof(nfh));
2902 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2903 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2904 if (!error)
2905 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
2906 }
2907out:
2908 if (dirp) {
2909 if (dirp == nd.ni_dvp) {
2910 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2911 } else {
2912 /* Release existing locks to prevent deadlock. */
2913 if (nd.ni_dvp) {
2914 NDFREE(&nd, NDF_ONLY_PNBUF);
2915 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2916 vrele(nd.ni_dvp);
2917 else
2918 vput(nd.ni_dvp);
2919 }
2920 if (nd.ni_vp) {
2921 if (vpexcl)
2922 vput(nd.ni_vp);
2923 else
2924 vrele(nd.ni_vp);
2925 }
2926 nd.ni_dvp = NULL;
2927 nd.ni_vp = NULL;
2928 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2929 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2930 VOP_UNLOCK(dirp, 0);
2931 }
2932 }
2933 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2934 if (v3) {
2935 if (!error) {
2936 nfsm_srvpostop_fh(fhp);
2937 nfsm_srvpostop_attr(0, vap);
2938 }
2939 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2940 } else if (!error) {
2941 /* v2 non-error case. */
2942 nfsm_srvfhtom(fhp, v3);
2943 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2944 nfsm_srvfillattr(vap, fp);
2945 }
2946 error = 0;
2947 /* fall through */
2948
2949nfsmout:
2950 if (nd.ni_dvp) {
2951 NDFREE(&nd, NDF_ONLY_PNBUF);
2952 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2953 vrele(nd.ni_dvp);
2954 else
2955 vput(nd.ni_dvp);
2956 }
2957 if (nd.ni_vp) {
2958 if (vpexcl)
2959 vput(nd.ni_vp);
2960 else
2961 vrele(nd.ni_vp);
2962 }
2963 if (dirp)
2964 vrele(dirp);
2965 vn_finished_write(mp);
2966 VFS_UNLOCK_GIANT(vfslocked);
2967 return (error);
2968}
2969
2970/*
2971 * nfs rmdir service
2972 */
2973int
2974nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2975 struct thread *td, struct mbuf **mrq)
2976{
2977 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2978 struct sockaddr *nam = nfsd->nd_nam;
2979 caddr_t dpos = nfsd->nd_dpos;
2980 struct ucred *cred = nfsd->nd_cr;
2981 caddr_t bpos;
2982 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2983 int v3 = (nfsd->nd_flag & ND_NFSV3);
2984 struct mbuf *mb, *mreq;
2985 struct vnode *vp, *dirp = NULL;
2986 struct vattr dirfor, diraft;
2987 nfsfh_t nfh;
2988 fhandle_t *fhp;
2989 struct nameidata nd;
2990 struct mount *mp = NULL;
2991 int vfslocked;
2992
2993 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2994 ndclear(&nd);
2995 vfslocked = 0;
2996
2997 fhp = &nfh.fh_generic;
2998 nfsm_srvmtofh(fhp);
2999 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3000 error = ESTALE;
3001 goto out;
3002 }
3003 vfslocked = VFS_LOCK_GIANT(mp);
3004 (void) vn_start_write(NULL, &mp, V_WAIT);
3005 vfs_rel(mp); /* The write holds a ref. */
3006 nfsm_srvnamesiz(len);
3007 nd.ni_cnd.cn_cred = cred;
3008 nd.ni_cnd.cn_nameiop = DELETE;
3009 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
3010 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
3011 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
3012 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
3013 if (dirp && !v3) {
3014 vrele(dirp);
3015 dirp = NULL;
3016 }
3017 if (error) {
3018 nfsm_reply(NFSX_WCCDATA(v3));
3019 if (v3)
3020 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3021 error = 0;
3022 goto nfsmout;
3023 }
3024 vp = nd.ni_vp;
3025 if (vp->v_type != VDIR) {
3026 error = ENOTDIR;
3027 goto out;
3028 }
3029 /*
3030 * No rmdir "." please.
3031 */
3032 if (nd.ni_dvp == vp) {
3033 error = EINVAL;
3034 goto out;
3035 }
3036 /*
3037 * The root of a mounted filesystem cannot be deleted.
3038 */
3039 if (vp->v_vflag & VV_ROOT)
3040 error = EBUSY;
3041out:
3042 /*
3043 * Issue or abort op. Since SAVESTART is not set, path name
3044 * component is freed by the VOP after either.
3045 */
3046 if (!error)
3047 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
3048 NDFREE(&nd, NDF_ONLY_PNBUF);
3049
3050 if (dirp) {
3051 if (dirp == nd.ni_dvp)
3052 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
3053 else {
3054 /* Release existing locks to prevent deadlock. */
3055 if (nd.ni_dvp) {
3056 if (nd.ni_dvp == nd.ni_vp)
3057 vrele(nd.ni_dvp);
3058 else
3059 vput(nd.ni_dvp);
3060 }
3061 if (nd.ni_vp)
3062 vput(nd.ni_vp);
3063 nd.ni_dvp = NULL;
3064 nd.ni_vp = NULL;
3065 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
3066 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
3067 VOP_UNLOCK(dirp, 0);
3068 }
3069 }
3070 nfsm_reply(NFSX_WCCDATA(v3));
3071 error = 0;
3072 if (v3)
3073 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3074 /* fall through */
3075
3076nfsmout:
3077 NDFREE(&nd, NDF_ONLY_PNBUF);
3078 if (nd.ni_dvp) {
3079 if (nd.ni_dvp == nd.ni_vp)
3080 vrele(nd.ni_dvp);
3081 else
3082 vput(nd.ni_dvp);
3083 }
3084 if (nd.ni_vp)
3085 vput(nd.ni_vp);
3086 if (dirp)
3087 vrele(dirp);
3088
3089 vn_finished_write(mp);
3090 VFS_UNLOCK_GIANT(vfslocked);
3091 return(error);
3092}
3093
3094/*
3095 * nfs readdir service
3096 * - mallocs what it thinks is enough to read
3097 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
3098 * - calls VOP_READDIR()
3099 * - loops around building the reply
3100 * if the output generated exceeds count break out of loop
3101 * The nfsm_clget macro is used here so that the reply will be packed
3102 * tightly in mbuf clusters.
3103 * - it only knows that it has encountered eof when the VOP_READDIR()
3104 * reads nothing
3105 * - as such one readdir rpc will return eof false although you are there
3106 * and then the next will return eof
3107 * - it trims out records with d_fileno == 0
3108 * this doesn't matter for Unix clients, but they might confuse clients
3109 * for other os'.
3110 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
3111 * than requested, but this may not apply to all filesystems. For
3112 * example, client NFS does not { although it is never remote mounted
3113 * anyhow }
3114 * The alternate call nfsrv_readdirplus() does lookups as well.
3115 * PS: The NFS protocol spec. does not clarify what the "count" byte
3116 * argument is a count of.. just name strings and file id's or the
3117 * entire reply rpc or ...
3118 * I tried just file name and id sizes and it confused the Sun client,
3119 * so I am using the full rpc size now. The "paranoia.." comment refers
3120 * to including the status longwords that are not a part of the dir.
3121 * "entry" structures, but are in the rpc.
3122 */
3123struct flrep {
3124 nfsuint64 fl_off;
3125 u_int32_t fl_postopok;
3126 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
3127 u_int32_t fl_fhok;
3128 u_int32_t fl_fhsize;
3129 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
3130};
3131
3132int
3133nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3134 struct thread *td, struct mbuf **mrq)
3135{
3136 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3137 struct sockaddr *nam = nfsd->nd_nam;
3138 caddr_t dpos = nfsd->nd_dpos;
3139 struct ucred *cred = nfsd->nd_cr;
3140 char *bp, *be;
3141 struct mbuf *mp;
3142 struct dirent *dp;
3143 caddr_t cp;
3144 u_int32_t *tl;
3145 caddr_t bpos;
3146 struct mbuf *mb, *mreq;
3147 char *cpos, *cend, *rbuf;
3148 struct vnode *vp = NULL;
3149 struct vattr at;
3150 nfsfh_t nfh;
3151 fhandle_t *fhp;
3152 struct uio io;
3153 struct iovec iv;
3154 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3155 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
3156 int v3 = (nfsd->nd_flag & ND_NFSV3);
3157 u_quad_t off, toff, verf;
3158 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3159 int vfslocked;
3160
3161 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3162 vfslocked = 0;
3163 fhp = &nfh.fh_generic;
3164 nfsm_srvmtofh(fhp);
3165 if (v3) {
3166 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
3167 toff = fxdr_hyper(tl);
3168 tl += 2;
3169 verf = fxdr_hyper(tl);
3170 tl += 2;
3171 } else {
3172 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
3173 toff = fxdr_unsigned(u_quad_t, *tl++);
3174 verf = 0; /* shut up gcc */
3175 }
3176 off = toff;
3177 cnt = fxdr_unsigned(int, *tl);
3178 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3179 xfer = NFS_SRVMAXDATA(nfsd);
3180 if (cnt > xfer)
3181 cnt = xfer;
3182 if (siz > xfer)
3183 siz = xfer;
3184 fullsiz = siz;
3185 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3186 nam, &rdonly, TRUE);
3187 if (!error && vp->v_type != VDIR) {
3188 error = ENOTDIR;
3189 vput(vp);
3190 vp = NULL;
3191 }
3192 if (error) {
3193 nfsm_reply(NFSX_UNSIGNED);
3194 if (v3)
3195 nfsm_srvpostop_attr(getret, &at);
3196 error = 0;
3197 goto nfsmout;
3198 }
3199
3200 /*
3201 * Obtain lock on vnode for this section of the code
3202 */
3203 if (v3) {
3204 error = getret = VOP_GETATTR(vp, &at, cred, td);
3205#if 0
3206 /*
3207 * XXX This check may be too strict for Solaris 2.5 clients.
3208 */
3209 if (!error && toff && verf && verf != at.va_filerev)
3210 error = NFSERR_BAD_COOKIE;
3211#endif
3212 }
3213 if (!error)
3214 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3215 if (error) {
3216 vput(vp);
3217 vp = NULL;
3218 nfsm_reply(NFSX_POSTOPATTR(v3));
3219 if (v3)
3220 nfsm_srvpostop_attr(getret, &at);
3221 error = 0;
3222 goto nfsmout;
3223 }
3224 VOP_UNLOCK(vp, 0);
3225
3226 /*
3227 * end section. Allocate rbuf and continue
3228 */
3229 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3230again:
3231 iv.iov_base = rbuf;
3232 iv.iov_len = fullsiz;
3233 io.uio_iov = &iv;
3234 io.uio_iovcnt = 1;
3235 io.uio_offset = (off_t)off;
3236 io.uio_resid = fullsiz;
3237 io.uio_segflg = UIO_SYSSPACE;
3238 io.uio_rw = UIO_READ;
3239 io.uio_td = NULL;
3240 eofflag = 0;
3241 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3242 if (cookies) {
3243 free((caddr_t)cookies, M_TEMP);
3244 cookies = NULL;
3245 }
3246 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3247 off = (off_t)io.uio_offset;
3248 if (!cookies && !error)
3249 error = NFSERR_PERM;
3250 if (v3) {
3251 getret = VOP_GETATTR(vp, &at, cred, td);
3252 if (!error)
3253 error = getret;
3254 }
3255 VOP_UNLOCK(vp, 0);
3256 if (error) {
3257 vrele(vp);
3258 vp = NULL;
3259 free((caddr_t)rbuf, M_TEMP);
3260 if (cookies)
3261 free((caddr_t)cookies, M_TEMP);
3262 nfsm_reply(NFSX_POSTOPATTR(v3));
3263 if (v3)
3264 nfsm_srvpostop_attr(getret, &at);
3265 error = 0;
3266 goto nfsmout;
3267 }
3268 if (io.uio_resid) {
3269 siz -= io.uio_resid;
3270
3271 /*
3272 * If nothing read, return eof
3273 * rpc reply
3274 */
3275 if (siz == 0) {
3276 vrele(vp);
3277 vp = NULL;
3278 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
3279 2 * NFSX_UNSIGNED);
3280 if (v3) {
3281 nfsm_srvpostop_attr(getret, &at);
3282 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3283 txdr_hyper(at.va_filerev, tl);
3284 tl += 2;
3285 } else
3286 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3287 *tl++ = nfsrv_nfs_false;
3288 *tl = nfsrv_nfs_true;
3289 FREE((caddr_t)rbuf, M_TEMP);
3290 FREE((caddr_t)cookies, M_TEMP);
3291 error = 0;
3292 goto nfsmout;
3293 }
3294 }
3295
3296 /*
3297 * Check for degenerate cases of nothing useful read.
3298 * If so go try again
3299 */
3300 cpos = rbuf;
3301 cend = rbuf + siz;
3302 dp = (struct dirent *)cpos;
3303 cookiep = cookies;
3304 /*
3305 * For some reason FreeBSD's ufs_readdir() chooses to back the
3306 * directory offset up to a block boundary, so it is necessary to
3307 * skip over the records that precede the requested offset. This
3308 * requires the assumption that file offset cookies monotonically
3309 * increase.
3310 */
3311 while (cpos < cend && ncookies > 0 &&
3312 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3313 ((u_quad_t)(*cookiep)) <= toff)) {
3314 cpos += dp->d_reclen;
3315 dp = (struct dirent *)cpos;
3316 cookiep++;
3317 ncookies--;
3318 }
3319 if (cpos >= cend || ncookies == 0) {
3320 toff = off;
3321 siz = fullsiz;
3322 goto again;
3323 }
3324
3325 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
3326 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3327 if (v3) {
3328 nfsm_srvpostop_attr(getret, &at);
3329 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3330 txdr_hyper(at.va_filerev, tl);
3331 }
3332 mp = mb;
3333 bp = bpos;
3334 be = bp + M_TRAILINGSPACE(mp);
3335
3336 /* Loop through the records and build reply */
3337 while (cpos < cend && ncookies > 0) {
3338 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3339 nlen = dp->d_namlen;
3340 rem = nfsm_rndup(nlen) - nlen;
3341 len += (4 * NFSX_UNSIGNED + nlen + rem);
3342 if (v3)
3343 len += 2 * NFSX_UNSIGNED;
3344 if (len > cnt) {
3345 eofflag = 0;
3346 break;
3347 }
3348 /*
3349 * Build the directory record xdr from
3350 * the dirent entry.
3351 */
3352 nfsm_clget;
3353 *tl = nfsrv_nfs_true;
3354 bp += NFSX_UNSIGNED;
3355 if (v3) {
3356 nfsm_clget;
3357 *tl = 0;
3358 bp += NFSX_UNSIGNED;
3359 }
3360 nfsm_clget;
3361 *tl = txdr_unsigned(dp->d_fileno);
3362 bp += NFSX_UNSIGNED;
3363 nfsm_clget;
3364 *tl = txdr_unsigned(nlen);
3365 bp += NFSX_UNSIGNED;
3366
3367 /* And loop around copying the name */
3368 xfer = nlen;
3369 cp = dp->d_name;
3370 while (xfer > 0) {
3371 nfsm_clget;
3372 if ((bp+xfer) > be)
3373 tsiz = be-bp;
3374 else
3375 tsiz = xfer;
3376 bcopy(cp, bp, tsiz);
3377 bp += tsiz;
3378 xfer -= tsiz;
3379 if (xfer > 0)
3380 cp += tsiz;
3381 }
3382 /* And null pad to an int32_t boundary. */
3383 for (i = 0; i < rem; i++)
3384 *bp++ = '\0';
3385 nfsm_clget;
3386
3387 /* Finish off the record */
3388 if (v3) {
3389 *tl = 0;
3390 bp += NFSX_UNSIGNED;
3391 nfsm_clget;
3392 }
3393 *tl = txdr_unsigned(*cookiep);
3394 bp += NFSX_UNSIGNED;
3395 }
3396 cpos += dp->d_reclen;
3397 dp = (struct dirent *)cpos;
3398 cookiep++;
3399 ncookies--;
3400 }
3401 vrele(vp);
3402 vp = NULL;
3403 nfsm_clget;
3404 *tl = nfsrv_nfs_false;
3405 bp += NFSX_UNSIGNED;
3406 nfsm_clget;
3407 if (eofflag)
3408 *tl = nfsrv_nfs_true;
3409 else
3410 *tl = nfsrv_nfs_false;
3411 bp += NFSX_UNSIGNED;
3412 if (mp != mb) {
3413 if (bp < be)
3414 mp->m_len = bp - mtod(mp, caddr_t);
3415 } else
3416 mp->m_len += bp - bpos;
3417 FREE((caddr_t)rbuf, M_TEMP);
3418 FREE((caddr_t)cookies, M_TEMP);
3419
3420nfsmout:
3421 if (vp)
3422 vrele(vp);
3423 VFS_UNLOCK_GIANT(vfslocked);
3424 return(error);
3425}
3426
3427int
3428nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3429 struct thread *td, struct mbuf **mrq)
3430{
3431 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3432 struct sockaddr *nam = nfsd->nd_nam;
3433 caddr_t dpos = nfsd->nd_dpos;
3434 struct ucred *cred = nfsd->nd_cr;
3435 char *bp, *be;
3436 struct mbuf *mp;
3437 struct dirent *dp;
3438 caddr_t cp;
3439 u_int32_t *tl;
3440 caddr_t bpos;
3441 struct mbuf *mb, *mreq;
3442 char *cpos, *cend, *rbuf;
3443 struct vnode *vp = NULL, *nvp;
3444 struct flrep fl;
3445 nfsfh_t nfh;
3446 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3447 struct uio io;
3448 struct iovec iv;
3449 struct vattr va, at, *vap = &va;
3450 struct nfs_fattr *fp;
3451 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3452 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3453 u_quad_t off, toff, verf;
3454 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3455 int v3 = (nfsd->nd_flag & ND_NFSV3);
3456 int vfslocked;
3457
3458 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3459 vfslocked = 0;
3460 if (!v3)
3461 panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
3462 fhp = &nfh.fh_generic;
3463 nfsm_srvmtofh(fhp);
3464 tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED);
3465 toff = fxdr_hyper(tl);
3466 tl += 2;
3467 verf = fxdr_hyper(tl);
3468 tl += 2;
3469 siz = fxdr_unsigned(int, *tl++);
3470 cnt = fxdr_unsigned(int, *tl);
3471 off = toff;
3472 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3473 xfer = NFS_SRVMAXDATA(nfsd);
3474 if (cnt > xfer)
3475 cnt = xfer;
3476 if (siz > xfer)
3477 siz = xfer;
3478 fullsiz = siz;
3479 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3480 nam, &rdonly, TRUE);
3481 if (!error && vp->v_type != VDIR) {
3482 error = ENOTDIR;
3483 vput(vp);
3484 vp = NULL;
3485 }
3486 if (error) {
3487 nfsm_reply(NFSX_UNSIGNED);
3488 nfsm_srvpostop_attr(getret, &at);
3489 error = 0;
3490 goto nfsmout;
3491 }
3492 error = getret = VOP_GETATTR(vp, &at, cred, td);
3493#if 0
3494 /*
3495 * XXX This check may be too strict for Solaris 2.5 clients.
3496 */
3497 if (!error && toff && verf && verf != at.va_filerev)
3498 error = NFSERR_BAD_COOKIE;
3499#endif
3500 if (!error)
3501 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3502 if (error) {
3503 vput(vp);
3504 vp = NULL;
3505 nfsm_reply(NFSX_V3POSTOPATTR);
3506 nfsm_srvpostop_attr(getret, &at);
3507 error = 0;
3508 goto nfsmout;
3509 }
3510 VOP_UNLOCK(vp, 0);
3511 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3512again:
3513 iv.iov_base = rbuf;
3514 iv.iov_len = fullsiz;
3515 io.uio_iov = &iv;
3516 io.uio_iovcnt = 1;
3517 io.uio_offset = (off_t)off;
3518 io.uio_resid = fullsiz;
3519 io.uio_segflg = UIO_SYSSPACE;
3520 io.uio_rw = UIO_READ;
3521 io.uio_td = NULL;
3522 eofflag = 0;
3523 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3524 if (cookies) {
3525 free((caddr_t)cookies, M_TEMP);
3526 cookies = NULL;
3527 }
3528 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3529 off = (u_quad_t)io.uio_offset;
3530 getret = VOP_GETATTR(vp, &at, cred, td);
3531 VOP_UNLOCK(vp, 0);
3532 if (!cookies && !error)
3533 error = NFSERR_PERM;
3534 if (!error)
3535 error = getret;
3536 if (error) {
3537 vrele(vp);
3538 vp = NULL;
3539 if (cookies)
3540 free((caddr_t)cookies, M_TEMP);
3541 free((caddr_t)rbuf, M_TEMP);
3542 nfsm_reply(NFSX_V3POSTOPATTR);
3543 nfsm_srvpostop_attr(getret, &at);
3544 error = 0;
3545 goto nfsmout;
3546 }
3547 if (io.uio_resid) {
3548 siz -= io.uio_resid;
3549
3550 /*
3551 * If nothing read, return eof
3552 * rpc reply
3553 */
3554 if (siz == 0) {
3555 vrele(vp);
3556 vp = NULL;
3557 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3558 2 * NFSX_UNSIGNED);
3559 nfsm_srvpostop_attr(getret, &at);
3560 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3561 txdr_hyper(at.va_filerev, tl);
3562 tl += 2;
3563 *tl++ = nfsrv_nfs_false;
3564 *tl = nfsrv_nfs_true;
3565 FREE((caddr_t)cookies, M_TEMP);
3566 FREE((caddr_t)rbuf, M_TEMP);
3567 error = 0;
3568 goto nfsmout;
3569 }
3570 }
3571
3572 /*
3573 * Check for degenerate cases of nothing useful read.
3574 * If so go try again
3575 */
3576 cpos = rbuf;
3577 cend = rbuf + siz;
3578 dp = (struct dirent *)cpos;
3579 cookiep = cookies;
3580 /*
3581 * For some reason FreeBSD's ufs_readdir() chooses to back the
3582 * directory offset up to a block boundary, so it is necessary to
3583 * skip over the records that precede the requested offset. This
3584 * requires the assumption that file offset cookies monotonically
3585 * increase.
3586 */
3587 while (cpos < cend && ncookies > 0 &&
3588 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3589 ((u_quad_t)(*cookiep)) <= toff)) {
3590 cpos += dp->d_reclen;
3591 dp = (struct dirent *)cpos;
3592 cookiep++;
3593 ncookies--;
3594 }
3595 if (cpos >= cend || ncookies == 0) {
3596 toff = off;
3597 siz = fullsiz;
3598 goto again;
3599 }
3600
3601 /*
3602 * Probe one of the directory entries to see if the filesystem
3603 * supports VGET.
3604 */
3605 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE, &nvp) ==
3606 EOPNOTSUPP) {
3607 error = NFSERR_NOTSUPP;
3608 vrele(vp);
3609 vp = NULL;
3610 free((caddr_t)cookies, M_TEMP);
3611 free((caddr_t)rbuf, M_TEMP);
3612 nfsm_reply(NFSX_V3POSTOPATTR);
3613 nfsm_srvpostop_attr(getret, &at);
3614 error = 0;
3615 goto nfsmout;
3616 }
3617 vput(nvp);
3618 nvp = NULL;
3619
3620 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3621 2 * NFSX_UNSIGNED;
3622 nfsm_reply(cnt);
3623 nfsm_srvpostop_attr(getret, &at);
3624 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3625 txdr_hyper(at.va_filerev, tl);
3626 mp = mb;
3627 bp = bpos;
3628 be = bp + M_TRAILINGSPACE(mp);
3629
3630 /* Loop through the records and build reply */
3631 while (cpos < cend && ncookies > 0) {
3632 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3633 nlen = dp->d_namlen;
3634 rem = nfsm_rndup(nlen)-nlen;
3635
3636 /*
3637 * For readdir_and_lookup get the vnode using
3638 * the file number.
3639 */
3640 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE,
3641 &nvp))
3642 goto invalid;
3643 bzero((caddr_t)nfhp, NFSX_V3FH);
3644 nfhp->fh_fsid =
3645 nvp->v_mount->mnt_stat.f_fsid;
3646 /*
3647 * XXXRW: Assert the mountpoints are the same so that
3648 * we know that acquiring Giant based on the
3649 * directory is the right thing for the child.
3650 */
3651 KASSERT(nvp->v_mount == vp->v_mount,
3652 ("nfsrv_readdirplus: nvp mount != vp mount"));
3653 if (VOP_VPTOFH(nvp, &nfhp->fh_fid)) {
3654 vput(nvp);
3655 nvp = NULL;
3656 goto invalid;
3657 }
3658 if (VOP_GETATTR(nvp, vap, cred, td)) {
3659 vput(nvp);
3660 nvp = NULL;
3661 goto invalid;
3662 }
3663 vput(nvp);
3664 nvp = NULL;
3665
3666 /*
3667 * If either the dircount or maxcount will be
3668 * exceeded, get out now. Both of these lengths
3669 * are calculated conservatively, including all
3670 * XDR overheads.
3671 */
3672 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3673 NFSX_V3POSTOPATTR);
3674 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3675 if (len > cnt || dirlen > fullsiz) {
3676 eofflag = 0;
3677 break;
3678 }
3679
3680 /*
3681 * Build the directory record xdr from
3682 * the dirent entry.
3683 */
3684 fp = (struct nfs_fattr *)&fl.fl_fattr;
3685 nfsm_srvfillattr(vap, fp);
3686 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3687 fl.fl_fhok = nfsrv_nfs_true;
3688 fl.fl_postopok = nfsrv_nfs_true;
3689 fl.fl_off.nfsuquad[0] = 0;
3690 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3691
3692 nfsm_clget;
3693 *tl = nfsrv_nfs_true;
3694 bp += NFSX_UNSIGNED;
3695 nfsm_clget;
3696 *tl = 0;
3697 bp += NFSX_UNSIGNED;
3698 nfsm_clget;
3699 *tl = txdr_unsigned(dp->d_fileno);
3700 bp += NFSX_UNSIGNED;
3701 nfsm_clget;
3702 *tl = txdr_unsigned(nlen);
3703 bp += NFSX_UNSIGNED;
3704
3705 /* And loop around copying the name */
3706 xfer = nlen;
3707 cp = dp->d_name;
3708 while (xfer > 0) {
3709 nfsm_clget;
3710 if ((bp + xfer) > be)
3711 tsiz = be - bp;
3712 else
3713 tsiz = xfer;
3714 bcopy(cp, bp, tsiz);
3715 bp += tsiz;
3716 xfer -= tsiz;
3717 if (xfer > 0)
3718 cp += tsiz;
3719 }
3720 /* And null pad to an int32_t boundary. */
3721 for (i = 0; i < rem; i++)
3722 *bp++ = '\0';
3723
3724 /*
3725 * Now copy the flrep structure out.
3726 */
3727 xfer = sizeof (struct flrep);
3728 cp = (caddr_t)&fl;
3729 while (xfer > 0) {
3730 nfsm_clget;
3731 if ((bp + xfer) > be)
3732 tsiz = be - bp;
3733 else
3734 tsiz = xfer;
3735 bcopy(cp, bp, tsiz);
3736 bp += tsiz;
3737 xfer -= tsiz;
3738 if (xfer > 0)
3739 cp += tsiz;
3740 }
3741 }
3742invalid:
3743 cpos += dp->d_reclen;
3744 dp = (struct dirent *)cpos;
3745 cookiep++;
3746 ncookies--;
3747 }
3748 vrele(vp);
3749 vp = NULL;
3750 nfsm_clget;
3751 *tl = nfsrv_nfs_false;
3752 bp += NFSX_UNSIGNED;
3753 nfsm_clget;
3754 if (eofflag)
3755 *tl = nfsrv_nfs_true;
3756 else
3757 *tl = nfsrv_nfs_false;
3758 bp += NFSX_UNSIGNED;
3759 if (mp != mb) {
3760 if (bp < be)
3761 mp->m_len = bp - mtod(mp, caddr_t);
3762 } else
3763 mp->m_len += bp - bpos;
3764 FREE((caddr_t)cookies, M_TEMP);
3765 FREE((caddr_t)rbuf, M_TEMP);
3766nfsmout:
3767 if (vp)
3768 vrele(vp);
3769 VFS_UNLOCK_GIANT(vfslocked);
3770 return(error);
3771}
3772
3773/*
3774 * nfs commit service
3775 */
3776int
3777nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3778 struct thread *td, struct mbuf **mrq)
3779{
3780 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3781 struct sockaddr *nam = nfsd->nd_nam;
3782 caddr_t dpos = nfsd->nd_dpos;
3783 struct ucred *cred = nfsd->nd_cr;
3784 struct vattr bfor, aft;
3785 struct vnode *vp = NULL;
3786 nfsfh_t nfh;
3787 fhandle_t *fhp;
3788 u_int32_t *tl;
3789 caddr_t bpos;
3790 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3791 struct mbuf *mb, *mreq;
3792 u_quad_t off;
3793 struct mount *mp = NULL;
3794 int v3 = (nfsd->nd_flag & ND_NFSV3);
3795 int tvfslocked;
3796 int vfslocked;
3797
3798 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3799 vfslocked = 0;
3800 if (!v3)
3801 panic("nfsrv_commit: v3 proc called on a v2 connection");
3802 fhp = &nfh.fh_generic;
3803 nfsm_srvmtofh(fhp);
3804 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3805 error = ESTALE;
3806 goto ereply;
3807 }
3808 vfslocked = VFS_LOCK_GIANT(mp);
3809 (void) vn_start_write(NULL, &mp, V_WAIT);
3810 vfs_rel(mp); /* The write holds a ref. */
3811 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
3812
3813 /*
3814 * XXX At this time VOP_FSYNC() does not accept offset and byte
3815 * count parameters, so these arguments are useless (someday maybe).
3816 */
3817 off = fxdr_hyper(tl);
3818 tl += 2;
3819 cnt = fxdr_unsigned(int, *tl);
3820 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
3821 nam, &rdonly, TRUE);
3822 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
3823 if (error) {
3824 nfsm_reply(2 * NFSX_UNSIGNED);
3825 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3826 error = 0;
3827 goto nfsmout;
3828 }
3829 for_ret = VOP_GETATTR(vp, &bfor, cred, td);
3830
3831 if (cnt > MAX_COMMIT_COUNT) {
3832 /*
3833 * Give up and do the whole thing
3834 */
3835 if (vp->v_object &&
3836 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3837 VM_OBJECT_LOCK(vp->v_object);
3838 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3839 VM_OBJECT_UNLOCK(vp->v_object);
3840 }
3841 error = VOP_FSYNC(vp, MNT_WAIT, td);
3842 } else {
3843 /*
3844 * Locate and synchronously write any buffers that fall
3845 * into the requested range. Note: we are assuming that
3846 * f_iosize is a power of 2.
3847 */
3848 int iosize = vp->v_mount->mnt_stat.f_iosize;
3849 int iomask = iosize - 1;
3850 struct bufobj *bo;
3851 daddr_t lblkno;
3852
3853 /*
3854 * Align to iosize boundry, super-align to page boundry.
3855 */
3856 if (off & iomask) {
3857 cnt += off & iomask;
3858 off &= ~(u_quad_t)iomask;
3859 }
3860 if (off & PAGE_MASK) {
3861 cnt += off & PAGE_MASK;
3862 off &= ~(u_quad_t)PAGE_MASK;
3863 }
3864 lblkno = off / iosize;
3865
3866 if (vp->v_object &&
3867 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3868 VM_OBJECT_LOCK(vp->v_object);
3869 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3870 VM_OBJECT_UNLOCK(vp->v_object);
3871 }
3872
3873 bo = &vp->v_bufobj;
3874 BO_LOCK(bo);
3875 while (cnt > 0) {
3876 struct buf *bp;
3877
3878 /*
3879 * If we have a buffer and it is marked B_DELWRI we
3880 * have to lock and write it. Otherwise the prior
3881 * write is assumed to have already been committed.
3882 *
3883 * gbincore() can return invalid buffers now so we
3884 * have to check that bit as well (though B_DELWRI
3885 * should not be set if B_INVAL is set there could be
3886 * a race here since we haven't locked the buffer).
3887 */
3888 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
3889 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
3890 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
3891 BO_LOCK(bo);
3892 continue; /* retry */
3893 }
3894 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
3895 B_DELWRI) {
3896 bremfree(bp);
3897 bp->b_flags &= ~B_ASYNC;
3898 bwrite(bp);
3899 ++nfs_commit_miss;
3900 } else
3901 BUF_UNLOCK(bp);
3902 BO_LOCK(bo);
3903 }
3904 ++nfs_commit_blks;
3905 if (cnt < iosize)
3906 break;
3907 cnt -= iosize;
3908 ++lblkno;
3909 }
3910 BO_UNLOCK(bo);
3911 }
3912
3913 aft_ret = VOP_GETATTR(vp, &aft, cred, td);
3914 vput(vp);
3915 vp = NULL;
3916ereply:
3917 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3918 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3919 if (!error) {
3920 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3921 if (nfsver.tv_sec == 0)
3922 nfsver = boottime;
3923 *tl++ = txdr_unsigned(nfsver.tv_sec);
3924 *tl = txdr_unsigned(nfsver.tv_usec);
3925 } else {
3926 error = 0;
3927 }
3928nfsmout:
3929 if (vp)
3930 vput(vp);
3931 vn_finished_write(mp);
3932 VFS_UNLOCK_GIANT(vfslocked);
3933 return(error);
3934}
3935
3936/*
3937 * nfs statfs service
3938 */
3939int
3940nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3941 struct thread *td, struct mbuf **mrq)
3942{
3943 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3944 struct sockaddr *nam = nfsd->nd_nam;
3945 caddr_t dpos = nfsd->nd_dpos;
3946 struct ucred *cred = nfsd->nd_cr;
3947 struct statfs *sf;
3948 struct nfs_statfs *sfp;
3949 caddr_t bpos;
3950 int error = 0, rdonly, getret = 1;
3951 int v3 = (nfsd->nd_flag & ND_NFSV3);
3952 struct mbuf *mb, *mreq;
3953 struct vnode *vp = NULL;
3954 struct vattr at;
3955 nfsfh_t nfh;
3956 fhandle_t *fhp;
3957 struct statfs statfs;
3958 u_quad_t tval;
3959 int vfslocked;
3960
3961 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3962 vfslocked = 0;
3963 fhp = &nfh.fh_generic;
3964 nfsm_srvmtofh(fhp);
3965 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3966 nam, &rdonly, TRUE);
3967 if (error) {
3968 nfsm_reply(NFSX_UNSIGNED);
3969 if (v3)
3970 nfsm_srvpostop_attr(getret, &at);
3971 error = 0;
3972 goto nfsmout;
3973 }
3974 sf = &statfs;
3975 error = VFS_STATFS(vp->v_mount, sf, td);
3976 getret = VOP_GETATTR(vp, &at, cred, td);
3977 vput(vp);
3978 vp = NULL;
3979 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3980 if (v3)
3981 nfsm_srvpostop_attr(getret, &at);
3982 if (error) {
3983 error = 0;
3984 goto nfsmout;
3985 }
3986 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3987 if (v3) {
3988 tval = (u_quad_t)sf->f_blocks;
3989 tval *= (u_quad_t)sf->f_bsize;
3990 txdr_hyper(tval, &sfp->sf_tbytes);
3991 tval = (u_quad_t)sf->f_bfree;
3992 tval *= (u_quad_t)sf->f_bsize;
3993 txdr_hyper(tval, &sfp->sf_fbytes);
3994 /*
3995 * Don't send negative values for available space,
3996 * since this field is unsigned in the NFS protocol.
3997 * Otherwise, the client would see absurdly high
3998 * numbers for free space.
3999 */
4000 if (sf->f_bavail < 0)
4001 tval = 0;
4002 else
4003 tval = (u_quad_t)sf->f_bavail;
4004 tval *= (u_quad_t)sf->f_bsize;
4005 txdr_hyper(tval, &sfp->sf_abytes);
4006 sfp->sf_tfiles.nfsuquad[0] = 0;
4007 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
4008 sfp->sf_ffiles.nfsuquad[0] = 0;
4009 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4010 sfp->sf_afiles.nfsuquad[0] = 0;
4011 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4012 sfp->sf_invarsec = 0;
4013 } else {
4014 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
4015 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
4016 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
4017 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
4018 if (sf->f_bavail < 0)
4019 sfp->sf_bavail = 0;
4020 else
4021 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
4022 }
4023nfsmout:
4024 if (vp)
4025 vput(vp);
4026 VFS_UNLOCK_GIANT(vfslocked);
4027 return(error);
4028}
4029
4030/*
4031 * nfs fsinfo service
4032 */
4033int
4034nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4035 struct thread *td, struct mbuf **mrq)
4036{
4037 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4038 struct sockaddr *nam = nfsd->nd_nam;
4039 caddr_t dpos = nfsd->nd_dpos;
4040 struct ucred *cred = nfsd->nd_cr;
4041 struct nfsv3_fsinfo *sip;
4042 caddr_t bpos;
4043 int error = 0, rdonly, getret = 1, pref;
4044 struct mbuf *mb, *mreq;
4045 struct vnode *vp = NULL;
4046 struct vattr at;
4047 nfsfh_t nfh;
4048 fhandle_t *fhp;
4049 u_quad_t maxfsize;
4050 struct statfs sb;
4051 int v3 = (nfsd->nd_flag & ND_NFSV3);
4052 int vfslocked;
4053
4054 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4055 if (!v3)
4056 panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
4057 fhp = &nfh.fh_generic;
4058 vfslocked = 0;
4059 nfsm_srvmtofh(fhp);
4060 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4061 nam, &rdonly, TRUE);
4062 if (error) {
4063 nfsm_reply(NFSX_UNSIGNED);
4064 nfsm_srvpostop_attr(getret, &at);
4065 error = 0;
4066 goto nfsmout;
4067 }
4068
4069 /* XXX Try to make a guess on the max file size. */
4070 VFS_STATFS(vp->v_mount, &sb, td);
4071 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
4072
4073 getret = VOP_GETATTR(vp, &at, cred, td);
4074 vput(vp);
4075 vp = NULL;
4076 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
4077 nfsm_srvpostop_attr(getret, &at);
4078 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
4079
4080 /*
4081 * XXX
4082 * There should be filesystem VFS OP(s) to get this information.
4083 * For now, assume ufs.
4084 */
4085 if (slp->ns_so->so_type == SOCK_DGRAM)
4086 pref = NFS_MAXDGRAMDATA;
4087 else
4088 pref = NFS_MAXDATA;
4089 sip->fs_rtmax = txdr_unsigned(pref);
4090 sip->fs_rtpref = txdr_unsigned(pref);
4091 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
4092 sip->fs_wtmax = txdr_unsigned(pref);
4093 sip->fs_wtpref = txdr_unsigned(pref);
4094 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
4095 sip->fs_dtpref = txdr_unsigned(pref);
4096 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
4097 sip->fs_timedelta.nfsv3_sec = 0;
4098 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
4099 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
4100 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
4101 NFSV3FSINFO_CANSETTIME);
4102nfsmout:
4103 if (vp)
4104 vput(vp);
4105 VFS_UNLOCK_GIANT(vfslocked);
4106 return(error);
4107}
4108
4109/*
4110 * nfs pathconf service
4111 */
4112int
4113nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4114 struct thread *td, struct mbuf **mrq)
4115{
4116 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4117 struct sockaddr *nam = nfsd->nd_nam;
4118 caddr_t dpos = nfsd->nd_dpos;
4119 struct ucred *cred = nfsd->nd_cr;
4120 struct nfsv3_pathconf *pc;
4121 caddr_t bpos;
4122 int error = 0, rdonly, getret = 1;
4123 register_t linkmax, namemax, chownres, notrunc;
4124 struct mbuf *mb, *mreq;
4125 struct vnode *vp = NULL;
4126 struct vattr at;
4127 nfsfh_t nfh;
4128 fhandle_t *fhp;
4129 int v3 = (nfsd->nd_flag & ND_NFSV3);
4130 int vfslocked;
4131
4132 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4133 if (!v3)
4134 panic("nfsrv_pathconf: v3 proc called on a v2 connection");
4135 fhp = &nfh.fh_generic;
4136 nfsm_srvmtofh(fhp);
4137 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4138 nam, &rdonly, TRUE);
4139 if (error) {
4140 nfsm_reply(NFSX_UNSIGNED);
4141 nfsm_srvpostop_attr(getret, &at);
4142 error = 0;
4143 goto nfsmout;
4144 }
4145 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
4146 if (!error)
4147 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
4148 if (!error)
4149 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
4150 if (!error)
4151 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
4152 getret = VOP_GETATTR(vp, &at, cred, td);
4153 vput(vp);
4154 vp = NULL;
4155 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
4156 nfsm_srvpostop_attr(getret, &at);
4157 if (error) {
4158 error = 0;
4159 goto nfsmout;
4160 }
4161 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
4162
4163 pc->pc_linkmax = txdr_unsigned(linkmax);
4164 pc->pc_namemax = txdr_unsigned(namemax);
4165 pc->pc_notrunc = txdr_unsigned(notrunc);
4166 pc->pc_chownrestricted = txdr_unsigned(chownres);
4167
4168 /*
4169 * These should probably be supported by VOP_PATHCONF(), but
4170 * until msdosfs is exportable (why would you want to?), the
4171 * Unix defaults should be ok.
4172 */
4173 pc->pc_caseinsensitive = nfsrv_nfs_false;
4174 pc->pc_casepreserving = nfsrv_nfs_true;
4175nfsmout:
4176 if (vp)
4177 vput(vp);
4178 VFS_UNLOCK_GIANT(vfslocked);
4179 return(error);
4180}
4181
4182/*
4183 * Null operation, used by clients to ping server
4184 */
4185/* ARGSUSED */
4186int
4187nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4188 struct thread *td, struct mbuf **mrq)
4189{
4190 struct mbuf *mrep = nfsd->nd_mrep;
4191 caddr_t bpos;
4192 int error = NFSERR_RETVOID;
4193 struct mbuf *mb, *mreq;
4194
4195 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4196 nfsm_reply(0);
4197nfsmout:
4198 return (error);
4199}
4200
4201/*
4202 * No operation, used for obsolete procedures
4203 */
4204/* ARGSUSED */
4205int
4206nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4207 struct thread *td, struct mbuf **mrq)
4208{
4209 struct mbuf *mrep = nfsd->nd_mrep;
4210 caddr_t bpos;
4211 int error;
4212 struct mbuf *mb, *mreq;
4213
4214 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4215 if (nfsd->nd_repstat)
4216 error = nfsd->nd_repstat;
4217 else
4218 error = EPROCUNAVAIL;
4219 nfsm_reply(0);
4220 error = 0;
4221nfsmout:
4222 return (error);
4223}
4224
4225/*
4226 * Perform access checking for vnodes obtained from file handles that would
4227 * refer to files already opened by a Unix client. You cannot just use
4228 * vn_writechk() and VOP_ACCESS() for two reasons.
4229 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
4230 * case.
4231 * 2 - The owner is to be given access irrespective of mode bits for some
4232 * operations, so that processes that chmod after opening a file don't
4233 * break. I don't like this because it opens a security hole, but since
4234 * the nfs server opens a security hole the size of a barn door anyhow,
4235 * what the heck.
4236 *
4237 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
4238 * will return EPERM instead of EACCESS. EPERM is always an error.
4239 */
4240static int
4241nfsrv_access(struct vnode *vp, int flags, struct ucred *cred,
4242 int rdonly, struct thread *td, int override)
4243{
4244 struct vattr vattr;
4245 int error;
4246
4247 VFS_ASSERT_GIANT(vp->v_mount);
4248
4249 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4250
4251 if (flags & VWRITE) {
4252 /* Just vn_writechk() changed to check rdonly */
4253 /*
4254 * Disallow write attempts on read-only filesystems;
4255 * unless the file is a socket or a block or character
4256 * device resident on the filesystem.
4257 */
4258 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
4259 switch (vp->v_type) {
4260 case VREG:
4261 case VDIR:
4262 case VLNK:
4263 return (EROFS);
4264 default:
4265 break;
4266 }
4267 }
4268 /*
4269 * If there's shared text associated with
4270 * the inode, we can't allow writing.
4271 */
4272 if (vp->v_vflag & VV_TEXT)
4273 return (ETXTBSY);
4274 }
4275
4276 error = VOP_GETATTR(vp, &vattr, cred, td);
4277 if (error)
4278 return (error);
4279 error = VOP_ACCESS(vp, flags, cred, td);
4280 /*
4281 * Allow certain operations for the owner (reads and writes
4282 * on files that are already open).
4283 */
4284 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
4285 error = 0;
4286 return (error);
4287}
| 940 m->m_len = 0;
941 m2->m_next = m;
942 m2 = m;
943 }
944 }
945 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
946 M_TEMP, M_WAITOK);
947 uiop->uio_iov = iv2 = iv;
948 m = mb;
949 left = len;
950 i = 0;
951 while (left > 0) {
952 if (m == NULL)
953 panic("nfsrv_read iov");
954 siz = min(M_TRAILINGSPACE(m), left);
955 if (siz > 0) {
956 iv->iov_base = mtod(m, caddr_t) + m->m_len;
957 iv->iov_len = siz;
958 m->m_len += siz;
959 left -= siz;
960 iv++;
961 i++;
962 }
963 m = m->m_next;
964 }
965 uiop->uio_iovcnt = i;
966 uiop->uio_offset = off;
967 uiop->uio_resid = len;
968 uiop->uio_rw = UIO_READ;
969 uiop->uio_segflg = UIO_SYSSPACE;
970 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
971 off = uiop->uio_offset;
972 nh->nh_nextr = off;
973 FREE((caddr_t)iv2, M_TEMP);
974 if (error || (getret = VOP_GETATTR(vp, vap, cred, td))) {
975 if (!error)
976 error = getret;
977 m_freem(mreq);
978 vput(vp);
979 vp = NULL;
980 nfsm_reply(NFSX_POSTOPATTR(v3));
981 if (v3)
982 nfsm_srvpostop_attr(getret, vap);
983 error = 0;
984 goto nfsmout;
985 }
986 } else
987 uiop->uio_resid = 0;
988 vput(vp);
989 vp = NULL;
990 nfsm_srvfillattr(vap, fp);
991 tlen = len - uiop->uio_resid;
992 cnt = cnt < tlen ? cnt : tlen;
993 tlen = nfsm_rndup(cnt);
994 if (len != tlen || tlen != cnt)
995 nfsm_adj(mb, len - tlen, tlen - cnt);
996 if (v3) {
997 *tl++ = txdr_unsigned(cnt);
998 if (cnt < reqlen)
999 *tl++ = nfsrv_nfs_true;
1000 else
1001 *tl++ = nfsrv_nfs_false;
1002 }
1003 *tl = txdr_unsigned(cnt);
1004nfsmout:
1005 if (vp)
1006 vput(vp);
1007 VFS_UNLOCK_GIANT(vfslocked);
1008 return(error);
1009}
1010
1011/*
1012 * nfs write service
1013 */
1014int
1015nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1016 struct thread *td, struct mbuf **mrq)
1017{
1018 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1019 struct sockaddr *nam = nfsd->nd_nam;
1020 caddr_t dpos = nfsd->nd_dpos;
1021 struct ucred *cred = nfsd->nd_cr;
1022 struct iovec *ivp;
1023 int i, cnt;
1024 struct mbuf *mp;
1025 struct nfs_fattr *fp;
1026 struct iovec *iv;
1027 struct vattr va, forat;
1028 struct vattr *vap = &va;
1029 u_int32_t *tl;
1030 caddr_t bpos;
1031 int error = 0, rdonly, len, forat_ret = 1;
1032 int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust;
1033 int stable = NFSV3WRITE_FILESYNC;
1034 int v3 = (nfsd->nd_flag & ND_NFSV3);
1035 struct mbuf *mb, *mreq;
1036 struct vnode *vp = NULL;
1037 nfsfh_t nfh;
1038 fhandle_t *fhp;
1039 struct uio io, *uiop = &io;
1040 off_t off;
1041 struct mount *mntp = NULL;
1042 int tvfslocked;
1043 int vfslocked;
1044
1045 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1046 vfslocked = 0;
1047 if (mrep == NULL) {
1048 *mrq = NULL;
1049 error = 0;
1050 goto nfsmout;
1051 }
1052 fhp = &nfh.fh_generic;
1053 nfsm_srvmtofh(fhp);
1054 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1055 error = ESTALE;
1056 goto ereply;
1057 }
1058 vfslocked = VFS_LOCK_GIANT(mntp);
1059 (void) vn_start_write(NULL, &mntp, V_WAIT);
1060 vfs_rel(mntp); /* The write holds a ref. */
1061 if (v3) {
1062 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1063 off = fxdr_hyper(tl);
1064 tl += 3;
1065 stable = fxdr_unsigned(int, *tl++);
1066 } else {
1067 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1068 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1069 tl += 2;
1070 if (nfs_async)
1071 stable = NFSV3WRITE_UNSTABLE;
1072 }
1073 retlen = len = fxdr_unsigned(int32_t, *tl);
1074 cnt = i = 0;
1075
1076 /*
1077 * For NFS Version 2, it is not obvious what a write of zero length
1078 * should do, but I might as well be consistent with Version 3,
1079 * which is to return ok so long as there are no permission problems.
1080 */
1081 if (len > 0) {
1082 zeroing = 1;
1083 mp = mrep;
1084 while (mp) {
1085 if (mp == md) {
1086 zeroing = 0;
1087 adjust = dpos - mtod(mp, caddr_t);
1088 mp->m_len -= adjust;
1089 if (mp->m_len > 0 && adjust > 0)
1090 mp->m_data += adjust;
1091 }
1092 if (zeroing)
1093 mp->m_len = 0;
1094 else if (mp->m_len > 0) {
1095 i += mp->m_len;
1096 if (i > len) {
1097 mp->m_len -= (i - len);
1098 zeroing = 1;
1099 }
1100 if (mp->m_len > 0)
1101 cnt++;
1102 }
1103 mp = mp->m_next;
1104 }
1105 }
1106 if (len > NFS_MAXDATA || len < 0 || i < len) {
1107 error = EIO;
1108 nfsm_reply(2 * NFSX_UNSIGNED);
1109 if (v3)
1110 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1111 error = 0;
1112 goto nfsmout;
1113 }
1114 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
1115 nam, &rdonly, TRUE);
1116 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
1117 if (error) {
1118 vp = NULL;
1119 nfsm_reply(2 * NFSX_UNSIGNED);
1120 if (v3)
1121 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1122 error = 0;
1123 goto nfsmout;
1124 }
1125 if (v3)
1126 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1127 if (vp->v_type != VREG) {
1128 if (v3)
1129 error = EINVAL;
1130 else
1131 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1132 }
1133 if (!error)
1134 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1135 if (error) {
1136 vput(vp);
1137 vp = NULL;
1138 nfsm_reply(NFSX_WCCDATA(v3));
1139 if (v3)
1140 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1141 error = 0;
1142 goto nfsmout;
1143 }
1144
1145 if (len > 0) {
1146 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1147 M_WAITOK);
1148 uiop->uio_iov = iv = ivp;
1149 uiop->uio_iovcnt = cnt;
1150 mp = mrep;
1151 while (mp) {
1152 if (mp->m_len > 0) {
1153 ivp->iov_base = mtod(mp, caddr_t);
1154 ivp->iov_len = mp->m_len;
1155 ivp++;
1156 }
1157 mp = mp->m_next;
1158 }
1159
1160 /*
1161 * XXX
1162 * The IO_METASYNC flag indicates that all metadata (and not just
1163 * enough to ensure data integrity) mus be written to stable storage
1164 * synchronously.
1165 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1166 */
1167 if (stable == NFSV3WRITE_UNSTABLE)
1168 ioflags = IO_NODELOCKED;
1169 else if (stable == NFSV3WRITE_DATASYNC)
1170 ioflags = (IO_SYNC | IO_NODELOCKED);
1171 else
1172 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1173 uiop->uio_resid = len;
1174 uiop->uio_rw = UIO_WRITE;
1175 uiop->uio_segflg = UIO_SYSSPACE;
1176 uiop->uio_td = NULL;
1177 uiop->uio_offset = off;
1178 error = VOP_WRITE(vp, uiop, ioflags, cred);
1179 /* XXXRW: unlocked write. */
1180 nfsrvstats.srvvop_writes++;
1181 FREE((caddr_t)iv, M_TEMP);
1182 }
1183 aftat_ret = VOP_GETATTR(vp, vap, cred, td);
1184 vput(vp);
1185 vp = NULL;
1186 if (!error)
1187 error = aftat_ret;
1188ereply:
1189 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1190 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1191 if (v3) {
1192 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1193 if (error) {
1194 error = 0;
1195 goto nfsmout;
1196 }
1197 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1198 *tl++ = txdr_unsigned(retlen);
1199 /*
1200 * If nfs_async is set, then pretend the write was FILESYNC.
1201 */
1202 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1203 *tl++ = txdr_unsigned(stable);
1204 else
1205 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1206 /*
1207 * Actually, there is no need to txdr these fields,
1208 * but it may make the values more human readable,
1209 * for debugging purposes.
1210 */
1211 if (nfsver.tv_sec == 0)
1212 nfsver = boottime;
1213 *tl++ = txdr_unsigned(nfsver.tv_sec);
1214 *tl = txdr_unsigned(nfsver.tv_usec);
1215 } else if (!error) {
1216 /* v2 non-error case. */
1217 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1218 nfsm_srvfillattr(vap, fp);
1219 }
1220 error = 0;
1221nfsmout:
1222 if (vp)
1223 vput(vp);
1224 vn_finished_write(mntp);
1225 VFS_UNLOCK_GIANT(vfslocked);
1226 return(error);
1227}
1228
1229/*
1230 * For the purposes of write gathering, we must decide if the credential
1231 * associated with two pending requests have equivilent privileges. Since
1232 * NFS only uses a subset of the BSD ucred -- the effective uid and group
1233 * IDs -- we have a compare routine that checks only the relevant fields.
1234 */
1235static int
1236nfsrv_samecred(struct ucred *cr1, struct ucred *cr2)
1237{
1238 int i;
1239
1240 if (cr1->cr_uid != cr2->cr_uid)
1241 return (0);
1242 if (cr1->cr_ngroups != cr2->cr_ngroups)
1243 return (0);
1244 for (i = 0; i < cr1->cr_ngroups; i++) {
1245 if (cr1->cr_groups[i] != cr2->cr_groups[i])
1246 return (0);
1247 }
1248 return (1);
1249}
1250
1251/*
1252 * NFS write service with write gathering support. Called when
1253 * nfsrvw_procrastinate > 0.
1254 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1255 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1256 * Jan. 1994.
1257 */
1258int
1259nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1260 struct thread *td, struct mbuf **mrq)
1261{
1262 struct iovec *ivp;
1263 struct mbuf *mp;
1264 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1265 struct nfs_fattr *fp;
1266 int i;
1267 struct iovec *iov;
1268 struct nfsrvw_delayhash *wpp;
1269 struct ucred *cred;
1270 struct vattr va, forat;
1271 u_int32_t *tl;
1272 caddr_t bpos, dpos;
1273 int error = 0, rdonly, len, forat_ret = 1;
1274 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1275 struct mbuf *mb, *mreq, *mrep, *md;
1276 struct vnode *vp = NULL;
1277 struct uio io, *uiop = &io;
1278 u_quad_t cur_usec;
1279 struct mount *mntp = NULL;
1280 int mvfslocked;
1281 int vfslocked;
1282
1283 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1284#ifndef nolint
1285 i = 0;
1286 len = 0;
1287#endif
1288 *mrq = NULL;
1289 if (*ndp) {
1290 nfsd = *ndp;
1291 *ndp = NULL;
1292 mrep = nfsd->nd_mrep;
1293 md = nfsd->nd_md;
1294 dpos = nfsd->nd_dpos;
1295 cred = nfsd->nd_cr;
1296 v3 = (nfsd->nd_flag & ND_NFSV3);
1297 LIST_INIT(&nfsd->nd_coalesce);
1298 nfsd->nd_mreq = NULL;
1299 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1300 cur_usec = nfs_curusec();
1301 nfsd->nd_time = cur_usec +
1302 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1303
1304 /*
1305 * Now, get the write header..
1306 */
1307 nfsm_srvmtofh(&nfsd->nd_fh);
1308 if (v3) {
1309 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1310 nfsd->nd_off = fxdr_hyper(tl);
1311 tl += 3;
1312 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1313 } else {
1314 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1315 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1316 tl += 2;
1317 if (nfs_async)
1318 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1319 }
1320 len = fxdr_unsigned(int32_t, *tl);
1321 nfsd->nd_len = len;
1322 nfsd->nd_eoff = nfsd->nd_off + len;
1323
1324 /*
1325 * Trim the header out of the mbuf list and trim off any trailing
1326 * junk so that the mbuf list has only the write data.
1327 */
1328 zeroing = 1;
1329 i = 0;
1330 mp = mrep;
1331 while (mp) {
1332 if (mp == md) {
1333 zeroing = 0;
1334 adjust = dpos - mtod(mp, caddr_t);
1335 mp->m_len -= adjust;
1336 if (mp->m_len > 0 && adjust > 0)
1337 mp->m_data += adjust;
1338 }
1339 if (zeroing)
1340 mp->m_len = 0;
1341 else {
1342 i += mp->m_len;
1343 if (i > len) {
1344 mp->m_len -= (i - len);
1345 zeroing = 1;
1346 }
1347 }
1348 mp = mp->m_next;
1349 }
1350 if (len > NFS_MAXDATA || len < 0 || i < len) {
1351nfsmout:
1352 m_freem(mrep);
1353 error = EIO;
1354 nfsm_writereply(2 * NFSX_UNSIGNED);
1355 if (v3)
1356 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1357 nfsd->nd_mreq = mreq;
1358 nfsd->nd_mrep = NULL;
1359 nfsd->nd_time = 0;
1360 }
1361
1362 /*
1363 * Add this entry to the hash and time queues.
1364 */
1365 s = splsoftclock();
1366 owp = NULL;
1367 wp = LIST_FIRST(&slp->ns_tq);
1368 while (wp && wp->nd_time < nfsd->nd_time) {
1369 owp = wp;
1370 wp = LIST_NEXT(wp, nd_tq);
1371 }
1372 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1373 if (owp) {
1374 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1375 } else {
1376 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1377 }
1378 if (nfsd->nd_mrep) {
1379 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1380 owp = NULL;
1381 wp = LIST_FIRST(wpp);
1382 while (wp &&
1383 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)){
1384 owp = wp;
1385 wp = LIST_NEXT(wp, nd_hash);
1386 }
1387 while (wp && wp->nd_off < nfsd->nd_off &&
1388 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1389 owp = wp;
1390 wp = LIST_NEXT(wp, nd_hash);
1391 }
1392 if (owp) {
1393 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1394
1395 /*
1396 * Search the hash list for overlapping entries and
1397 * coalesce.
1398 */
1399 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1400 wp = LIST_NEXT(nfsd, nd_hash);
1401 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1402 nfsrvw_coalesce(owp, nfsd);
1403 }
1404 } else {
1405 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1406 }
1407 }
1408 splx(s);
1409 }
1410
1411 /*
1412 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1413 * and generate the associated reply mbuf list(s).
1414 */
1415loop1:
1416 cur_usec = nfs_curusec();
1417 s = splsoftclock();
1418 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1419 owp = LIST_NEXT(nfsd, nd_tq);
1420 if (nfsd->nd_time > cur_usec)
1421 break;
1422 if (nfsd->nd_mreq)
1423 continue;
1424 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1425 LIST_REMOVE(nfsd, nd_tq);
1426 LIST_REMOVE(nfsd, nd_hash);
1427 splx(s);
1428 mrep = nfsd->nd_mrep;
1429 nfsd->nd_mrep = NULL;
1430 cred = nfsd->nd_cr;
1431 v3 = (nfsd->nd_flag & ND_NFSV3);
1432 forat_ret = aftat_ret = 1;
1433 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, &vfslocked, cred,
1434 slp, nfsd->nd_nam, &rdonly, TRUE);
1435 if (!error) {
1436 if (v3)
1437 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1438 if (vp->v_type != VREG) {
1439 if (v3)
1440 error = EINVAL;
1441 else
1442 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1443 }
1444 } else {
1445 vp = NULL;
1446 }
1447 if (!error)
1448 error = nfsrv_access(vp, VWRITE, cred, rdonly,
1449 td, 1);
1450 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1451 ioflags = IO_NODELOCKED;
1452 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1453 ioflags = (IO_SYNC | IO_NODELOCKED);
1454 else
1455 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1456 uiop->uio_rw = UIO_WRITE;
1457 uiop->uio_segflg = UIO_SYSSPACE;
1458 uiop->uio_td = NULL;
1459 uiop->uio_offset = nfsd->nd_off;
1460 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1461 if (uiop->uio_resid > 0) {
1462 mp = mrep;
1463 i = 0;
1464 while (mp) {
1465 if (mp->m_len > 0)
1466 i++;
1467 mp = mp->m_next;
1468 }
1469 uiop->uio_iovcnt = i;
1470 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1471 M_TEMP, M_WAITOK);
1472 uiop->uio_iov = ivp = iov;
1473 mp = mrep;
1474 while (mp) {
1475 if (mp->m_len > 0) {
1476 ivp->iov_base = mtod(mp, caddr_t);
1477 ivp->iov_len = mp->m_len;
1478 ivp++;
1479 }
1480 mp = mp->m_next;
1481 }
1482 mvfslocked = 0;
1483 if (!error) {
1484 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) {
1485 VOP_UNLOCK(vp, 0);
1486 error = vn_start_write(NULL, &mntp, V_WAIT);
1487 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1488 }
1489 mvfslocked = VFS_LOCK_GIANT(mntp);
1490 }
1491 if (!error) {
1492 error = VOP_WRITE(vp, uiop, ioflags, cred);
1493 /* XXXRW: unlocked write. */
1494 nfsrvstats.srvvop_writes++;
1495 vn_finished_write(mntp);
1496 }
1497 VFS_UNLOCK_GIANT(mvfslocked);
1498 FREE((caddr_t)iov, M_TEMP);
1499 }
1500 m_freem(mrep);
1501 if (vp) {
1502 aftat_ret = VOP_GETATTR(vp, &va, cred, td);
1503 vput(vp);
1504 vp = NULL;
1505 }
1506 VFS_UNLOCK_GIANT(vfslocked);
1507 /*
1508 * Loop around generating replies for all write rpcs that have
1509 * now been completed.
1510 */
1511 swp = nfsd;
1512 do {
1513 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1514 if (error) {
1515 nfsm_writereply(NFSX_WCCDATA(v3));
1516 if (v3) {
1517 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1518 }
1519 } else {
1520 nfsm_writereply(NFSX_PREOPATTR(v3) +
1521 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1522 NFSX_WRITEVERF(v3));
1523 if (v3) {
1524 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1525 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1526 *tl++ = txdr_unsigned(nfsd->nd_len);
1527 *tl++ = txdr_unsigned(swp->nd_stable);
1528 /*
1529 * Actually, there is no need to txdr these fields,
1530 * but it may make the values more human readable,
1531 * for debugging purposes.
1532 */
1533 if (nfsver.tv_sec == 0)
1534 nfsver = boottime;
1535 *tl++ = txdr_unsigned(nfsver.tv_sec);
1536 *tl = txdr_unsigned(nfsver.tv_usec);
1537 } else {
1538 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1539 nfsm_srvfillattr(&va, fp);
1540 }
1541 }
1542 nfsd->nd_mreq = mreq;
1543 if (nfsd->nd_mrep)
1544 panic("nfsrv_write: nd_mrep not free");
1545
1546 /*
1547 * Done. Put it at the head of the timer queue so that
1548 * the final phase can return the reply.
1549 */
1550 s = splsoftclock();
1551 if (nfsd != swp) {
1552 nfsd->nd_time = 0;
1553 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1554 }
1555 nfsd = LIST_FIRST(&swp->nd_coalesce);
1556 if (nfsd) {
1557 LIST_REMOVE(nfsd, nd_tq);
1558 }
1559 splx(s);
1560 } while (nfsd);
1561 s = splsoftclock();
1562 swp->nd_time = 0;
1563 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1564 splx(s);
1565 goto loop1;
1566 }
1567 splx(s);
1568
1569 /*
1570 * Search for a reply to return.
1571 */
1572 s = splsoftclock();
1573 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq)
1574 if (nfsd->nd_mreq) {
1575 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1576 LIST_REMOVE(nfsd, nd_tq);
1577 *mrq = nfsd->nd_mreq;
1578 *ndp = nfsd;
1579 break;
1580 }
1581 splx(s);
1582 return (0);
1583}
1584
1585/*
1586 * Coalesce the write request nfsd into owp. To do this we must:
1587 * - remove nfsd from the queues
1588 * - merge nfsd->nd_mrep into owp->nd_mrep
1589 * - update the nd_eoff and nd_stable for owp
1590 * - put nfsd on owp's nd_coalesce list
1591 * NB: Must be called at splsoftclock().
1592 */
1593static void
1594nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1595{
1596 int overlap;
1597 struct mbuf *mp;
1598 struct nfsrv_descript *p;
1599
1600 NFS_DPF(WG, ("C%03x-%03x",
1601 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1602 LIST_REMOVE(nfsd, nd_hash);
1603 LIST_REMOVE(nfsd, nd_tq);
1604 if (owp->nd_eoff < nfsd->nd_eoff) {
1605 overlap = owp->nd_eoff - nfsd->nd_off;
1606 if (overlap < 0)
1607 panic("nfsrv_coalesce: bad off");
1608 if (overlap > 0)
1609 m_adj(nfsd->nd_mrep, overlap);
1610 mp = owp->nd_mrep;
1611 while (mp->m_next)
1612 mp = mp->m_next;
1613 mp->m_next = nfsd->nd_mrep;
1614 owp->nd_eoff = nfsd->nd_eoff;
1615 } else
1616 m_freem(nfsd->nd_mrep);
1617 nfsd->nd_mrep = NULL;
1618 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1619 owp->nd_stable = NFSV3WRITE_FILESYNC;
1620 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1621 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1622 owp->nd_stable = NFSV3WRITE_DATASYNC;
1623 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1624
1625 /*
1626 * If nfsd had anything else coalesced into it, transfer them
1627 * to owp, otherwise their replies will never get sent.
1628 */
1629 for (p = LIST_FIRST(&nfsd->nd_coalesce); p;
1630 p = LIST_FIRST(&nfsd->nd_coalesce)) {
1631 LIST_REMOVE(p, nd_tq);
1632 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1633 }
1634}
1635
1636/*
1637 * nfs create service
1638 * now does a truncate to 0 length via. setattr if it already exists
1639 */
1640int
1641nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1642 struct thread *td, struct mbuf **mrq)
1643{
1644 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1645 struct sockaddr *nam = nfsd->nd_nam;
1646 caddr_t dpos = nfsd->nd_dpos;
1647 struct ucred *cred = nfsd->nd_cr;
1648 struct nfs_fattr *fp;
1649 struct vattr va, dirfor, diraft;
1650 struct vattr *vap = &va;
1651 struct nfsv2_sattr *sp;
1652 u_int32_t *tl;
1653 struct nameidata nd;
1654 caddr_t bpos;
1655 int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1656 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1657 caddr_t cp;
1658 struct mbuf *mb, *mreq;
1659 struct vnode *dirp = NULL;
1660 nfsfh_t nfh;
1661 fhandle_t *fhp;
1662 u_quad_t tempsize;
1663 u_char cverf[NFSX_V3CREATEVERF];
1664 struct mount *mp = NULL;
1665 int tvfslocked;
1666 int vfslocked;
1667
1668 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1669 vfslocked = 0;
1670#ifndef nolint
1671 rdev = 0;
1672#endif
1673 ndclear(&nd);
1674
1675 fhp = &nfh.fh_generic;
1676 nfsm_srvmtofh(fhp);
1677 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1678 error = ESTALE;
1679 goto ereply;
1680 }
1681 vfslocked = VFS_LOCK_GIANT(mp);
1682 (void) vn_start_write(NULL, &mp, V_WAIT);
1683 vfs_rel(mp); /* The write holds a ref. */
1684 nfsm_srvnamesiz(len);
1685
1686 nd.ni_cnd.cn_cred = cred;
1687 nd.ni_cnd.cn_nameiop = CREATE;
1688 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1689
1690 /*
1691 * Call namei and do initial cleanup to get a few things
1692 * out of the way. If we get an initial error we cleanup
1693 * and return here to avoid special-casing the invalid nd
1694 * structure through the rest of the case. dirp may be
1695 * set even if an error occurs, but the nd structure will not
1696 * be valid at all if an error occurs so we have to invalidate it
1697 * prior to calling nfsm_reply ( which might goto nfsmout ).
1698 */
1699 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1700 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1701 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1702 if (dirp && !v3) {
1703 vrele(dirp);
1704 dirp = NULL;
1705 }
1706 if (error) {
1707 nfsm_reply(NFSX_WCCDATA(v3));
1708 if (v3)
1709 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1710 error = 0;
1711 goto nfsmout;
1712 }
1713
1714 /*
1715 * No error. Continue. State:
1716 *
1717 * startdir is valid ( we release this immediately )
1718 * dirp may be valid
1719 * nd.ni_vp may be valid
1720 * nd.ni_dvp is valid
1721 *
1722 * The error state is set through the code and we may also do some
1723 * opportunistic releasing of vnodes to avoid holding locks through
1724 * NFS I/O. The cleanup at the end is a catch-all
1725 */
1726
1727 VATTR_NULL(vap);
1728 if (v3) {
1729 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1730 how = fxdr_unsigned(int, *tl);
1731 switch (how) {
1732 case NFSV3CREATE_GUARDED:
1733 if (nd.ni_vp) {
1734 error = EEXIST;
1735 break;
1736 }
1737 /* fall through */
1738 case NFSV3CREATE_UNCHECKED:
1739 nfsm_srvsattr(vap);
1740 break;
1741 case NFSV3CREATE_EXCLUSIVE:
1742 cp = nfsm_dissect_nonblock(caddr_t, NFSX_V3CREATEVERF);
1743 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1744 exclusive_flag = 1;
1745 break;
1746 };
1747 vap->va_type = VREG;
1748 } else {
1749 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
1750 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1751 if (vap->va_type == VNON)
1752 vap->va_type = VREG;
1753 vap->va_mode = nfstov_mode(sp->sa_mode);
1754 switch (vap->va_type) {
1755 case VREG:
1756 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1757 if (tsize != -1)
1758 vap->va_size = (u_quad_t)tsize;
1759 break;
1760 case VCHR:
1761 case VBLK:
1762 case VFIFO:
1763 rdev = fxdr_unsigned(long, sp->sa_size);
1764 break;
1765 default:
1766 break;
1767 };
1768 }
1769
1770 /*
1771 * Iff doesn't exist, create it
1772 * otherwise just truncate to 0 length
1773 * should I set the mode too ?
1774 *
1775 * The only possible error we can have at this point is EEXIST.
1776 * nd.ni_vp will also be non-NULL in that case.
1777 */
1778 if (nd.ni_vp == NULL) {
1779 if (vap->va_mode == (mode_t)VNOVAL)
1780 vap->va_mode = 0;
1781 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1782 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1783 if (error)
1784 NDFREE(&nd, NDF_ONLY_PNBUF);
1785 else {
1786 if (exclusive_flag) {
1787 exclusive_flag = 0;
1788 VATTR_NULL(vap);
1789 bcopy(cverf, (caddr_t)&vap->va_atime,
1790 NFSX_V3CREATEVERF);
1791 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1792 td);
1793 }
1794 }
1795 } else if (vap->va_type == VCHR || vap->va_type == VBLK ||
1796 vap->va_type == VFIFO) {
1797 /*
1798 * NFSv2-specific code for creating device nodes
1799 * and fifos.
1800 *
1801 * Handle SysV FIFO node special cases. All other
1802 * devices require super user to access.
1803 */
1804 if (vap->va_type == VCHR && rdev == 0xffffffff)
1805 vap->va_type = VFIFO;
1806 if (vap->va_type != VFIFO &&
1807 (error = suser_cred(cred, 0))) {
1808 goto ereply;
1809 }
1810 vap->va_rdev = rdev;
1811 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1812 if (error) {
1813 NDFREE(&nd, NDF_ONLY_PNBUF);
1814 goto ereply;
1815 }
1816 vput(nd.ni_vp);
1817 nd.ni_vp = NULL;
1818
1819 /*
1820 * release dvp prior to lookup
1821 */
1822 vput(nd.ni_dvp);
1823 nd.ni_dvp = NULL;
1824 /*
1825 * Setup for lookup.
1826 *
1827 * Even though LOCKPARENT was cleared, ni_dvp may
1828 * be garbage.
1829 */
1830 nd.ni_cnd.cn_nameiop = LOOKUP;
1831 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1832 nd.ni_cnd.cn_thread = td;
1833 nd.ni_cnd.cn_cred = cred;
1834 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
1835 if (tvfslocked)
1836 nd.ni_cnd.cn_flags |= GIANTHELD;
1837 error = lookup(&nd);
1838 nd.ni_dvp = NULL;
1839 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1840 nd.ni_cnd.cn_flags &= ~GIANTHELD;
1841 if (error)
1842 goto ereply;
1843
1844 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1845 error = EINVAL;
1846 goto ereply;
1847 }
1848 } else {
1849 error = ENXIO;
1850 }
1851 } else {
1852 if (vap->va_size != -1) {
1853 error = nfsrv_access(nd.ni_vp, VWRITE,
1854 cred, (nd.ni_cnd.cn_flags & RDONLY), td, 0);
1855 if (!error) {
1856 tempsize = vap->va_size;
1857 VATTR_NULL(vap);
1858 vap->va_size = tempsize;
1859 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1860 td);
1861 }
1862 }
1863 }
1864
1865 if (!error) {
1866 bzero((caddr_t)fhp, sizeof(nfh));
1867 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1868 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1869 if (!error)
1870 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
1871 }
1872 if (v3) {
1873 if (exclusive_flag && !error &&
1874 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1875 error = EEXIST;
1876 if (dirp == nd.ni_dvp)
1877 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1878 else {
1879 /* Drop the other locks to avoid deadlock. */
1880 if (nd.ni_dvp) {
1881 if (nd.ni_dvp == nd.ni_vp)
1882 vrele(nd.ni_dvp);
1883 else
1884 vput(nd.ni_dvp);
1885 }
1886 if (nd.ni_vp)
1887 vput(nd.ni_vp);
1888 nd.ni_dvp = NULL;
1889 nd.ni_vp = NULL;
1890
1891 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1892 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1893 VOP_UNLOCK(dirp, 0);
1894 }
1895 }
1896ereply:
1897 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1898 if (v3) {
1899 if (!error) {
1900 nfsm_srvpostop_fh(fhp);
1901 nfsm_srvpostop_attr(0, vap);
1902 }
1903 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1904 } else if (!error) {
1905 /* v2 non-error case. */
1906 nfsm_srvfhtom(fhp, v3);
1907 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1908 nfsm_srvfillattr(vap, fp);
1909 }
1910 error = 0;
1911
1912nfsmout:
1913 if (nd.ni_dvp) {
1914 if (nd.ni_dvp == nd.ni_vp)
1915 vrele(nd.ni_dvp);
1916 else
1917 vput(nd.ni_dvp);
1918 }
1919 if (nd.ni_vp)
1920 vput(nd.ni_vp);
1921 if (nd.ni_startdir) {
1922 vrele(nd.ni_startdir);
1923 nd.ni_startdir = NULL;
1924 }
1925 if (dirp)
1926 vrele(dirp);
1927 NDFREE(&nd, NDF_ONLY_PNBUF);
1928 vn_finished_write(mp);
1929 VFS_UNLOCK_GIANT(vfslocked);
1930 return (error);
1931}
1932
1933/*
1934 * nfs v3 mknod service
1935 */
1936int
1937nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1938 struct thread *td, struct mbuf **mrq)
1939{
1940 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1941 struct sockaddr *nam = nfsd->nd_nam;
1942 caddr_t dpos = nfsd->nd_dpos;
1943 struct ucred *cred = nfsd->nd_cr;
1944 struct vattr va, dirfor, diraft;
1945 struct vattr *vap = &va;
1946 u_int32_t *tl;
1947 struct nameidata nd;
1948 caddr_t bpos;
1949 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1950 u_int32_t major, minor;
1951 enum vtype vtyp;
1952 struct mbuf *mb, *mreq;
1953 struct vnode *vp, *dirp = NULL;
1954 nfsfh_t nfh;
1955 fhandle_t *fhp;
1956 struct mount *mp = NULL;
1957 int v3 = (nfsd->nd_flag & ND_NFSV3);
1958 int tvfslocked;
1959 int vfslocked;
1960
1961 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1962 vfslocked = 0;
1963 if (!v3)
1964 panic("nfsrv_mknod: v3 proc called on a v2 connection");
1965 ndclear(&nd);
1966
1967 fhp = &nfh.fh_generic;
1968 nfsm_srvmtofh(fhp);
1969 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1970 error = ESTALE;
1971 goto ereply;
1972 }
1973 vfslocked = VFS_LOCK_GIANT(mp);
1974 (void) vn_start_write(NULL, &mp, V_WAIT);
1975 vfs_rel(mp); /* The write holds a ref. */
1976 nfsm_srvnamesiz(len);
1977
1978 nd.ni_cnd.cn_cred = cred;
1979 nd.ni_cnd.cn_nameiop = CREATE;
1980 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1981
1982 /*
1983 * Handle nfs_namei() call. If an error occurs, the nd structure
1984 * is not valid. However, nfsm_*() routines may still jump to
1985 * nfsmout.
1986 */
1987
1988 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1989 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1990 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1991 if (error) {
1992 nfsm_reply(NFSX_WCCDATA(1));
1993 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1994 error = 0;
1995 goto nfsmout;
1996 }
1997 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1998 vtyp = nfsv3tov_type(*tl);
1999 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
2000 error = NFSERR_BADTYPE;
2001 goto out;
2002 }
2003 VATTR_NULL(vap);
2004 nfsm_srvsattr(vap);
2005 if (vtyp == VCHR || vtyp == VBLK) {
2006 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
2007 major = fxdr_unsigned(u_int32_t, *tl++);
2008 minor = fxdr_unsigned(u_int32_t, *tl);
2009 vap->va_rdev = makedev(major, minor);
2010 }
2011
2012 /*
2013 * Iff doesn't exist, create it.
2014 */
2015 if (nd.ni_vp) {
2016 error = EEXIST;
2017 goto out;
2018 }
2019 vap->va_type = vtyp;
2020 if (vap->va_mode == (mode_t)VNOVAL)
2021 vap->va_mode = 0;
2022 if (vtyp == VSOCK) {
2023 vrele(nd.ni_startdir);
2024 nd.ni_startdir = NULL;
2025 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2026 if (error)
2027 NDFREE(&nd, NDF_ONLY_PNBUF);
2028 } else {
2029 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
2030 goto out;
2031 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2032 if (error) {
2033 NDFREE(&nd, NDF_ONLY_PNBUF);
2034 goto out;
2035 }
2036 vput(nd.ni_vp);
2037 nd.ni_vp = NULL;
2038
2039 /*
2040 * Release dvp prior to lookup
2041 */
2042 vput(nd.ni_dvp);
2043 nd.ni_dvp = NULL;
2044
2045 nd.ni_cnd.cn_nameiop = LOOKUP;
2046 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
2047 nd.ni_cnd.cn_thread = td;
2048 nd.ni_cnd.cn_cred = td->td_ucred;
2049 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2050 if (tvfslocked)
2051 nd.ni_cnd.cn_flags |= GIANTHELD;
2052 error = lookup(&nd);
2053 nd.ni_dvp = NULL;
2054 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2055 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2056
2057 if (error)
2058 goto out;
2059 if (nd.ni_cnd.cn_flags & ISSYMLINK)
2060 error = EINVAL;
2061 }
2062
2063 /*
2064 * send response, cleanup, return.
2065 */
2066out:
2067 vp = nd.ni_vp;
2068 if (!error) {
2069 bzero((caddr_t)fhp, sizeof(nfh));
2070 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2071 error = VOP_VPTOFH(vp, &fhp->fh_fid);
2072 if (!error)
2073 error = VOP_GETATTR(vp, vap, cred, td);
2074 }
2075 if (nd.ni_dvp) {
2076 if (nd.ni_dvp == nd.ni_vp)
2077 vrele(nd.ni_dvp);
2078 else
2079 vput(nd.ni_dvp);
2080 nd.ni_dvp = NULL;
2081 }
2082 if (vp) {
2083 vput(vp);
2084 vp = NULL;
2085 nd.ni_vp = NULL;
2086 }
2087 if (nd.ni_startdir) {
2088 vrele(nd.ni_startdir);
2089 nd.ni_startdir = NULL;
2090 }
2091 NDFREE(&nd, NDF_ONLY_PNBUF);
2092 if (dirp) {
2093 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2094 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2095 VOP_UNLOCK(dirp, 0);
2096 }
2097ereply:
2098 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
2099 if (v3) {
2100 if (!error) {
2101 nfsm_srvpostop_fh(fhp);
2102 nfsm_srvpostop_attr(0, vap);
2103 }
2104 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2105 }
2106 vn_finished_write(mp);
2107 VFS_UNLOCK_GIANT(vfslocked);
2108 return (0);
2109nfsmout:
2110 if (nd.ni_dvp) {
2111 if (nd.ni_dvp == nd.ni_vp)
2112 vrele(nd.ni_dvp);
2113 else
2114 vput(nd.ni_dvp);
2115 }
2116 if (nd.ni_vp)
2117 vput(nd.ni_vp);
2118 if (dirp)
2119 vrele(dirp);
2120 if (nd.ni_startdir)
2121 vrele(nd.ni_startdir);
2122 NDFREE(&nd, NDF_ONLY_PNBUF);
2123 vn_finished_write(mp);
2124 VFS_UNLOCK_GIANT(vfslocked);
2125 return (error);
2126}
2127
2128/*
2129 * nfs remove service
2130 */
2131int
2132nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2133 struct thread *td, struct mbuf **mrq)
2134{
2135 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2136 struct sockaddr *nam = nfsd->nd_nam;
2137 caddr_t dpos = nfsd->nd_dpos;
2138 struct ucred *cred = nfsd->nd_cr;
2139 struct nameidata nd;
2140 caddr_t bpos;
2141 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2142 int v3 = (nfsd->nd_flag & ND_NFSV3);
2143 struct mbuf *mb, *mreq;
2144 struct vnode *dirp;
2145 struct vattr dirfor, diraft;
2146 nfsfh_t nfh;
2147 fhandle_t *fhp;
2148 struct mount *mp = NULL;
2149 int vfslocked, vfslocked1;
2150
2151 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2152 ndclear(&nd);
2153 vfslocked = 0;
2154
2155 fhp = &nfh.fh_generic;
2156 nfsm_srvmtofh(fhp);
2157 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2158 error = ESTALE;
2159 goto ereply;
2160 }
2161 vfslocked = VFS_LOCK_GIANT(mp);
2162 (void) vn_start_write(NULL, &mp, V_WAIT);
2163 vfs_rel(mp); /* The write holds a ref. */
2164 nfsm_srvnamesiz(len);
2165
2166 nd.ni_cnd.cn_cred = cred;
2167 nd.ni_cnd.cn_nameiop = DELETE;
2168 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
2169 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2170 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2171 vfslocked1 = NDHASGIANT(&nd);
2172 if (vfslocked && vfslocked1)
2173 VFS_UNLOCK_GIANT(vfslocked1);
2174 if (vfslocked || vfslocked1)
2175 vfslocked = 1;
2176 if (dirp && !v3) {
2177 vrele(dirp);
2178 dirp = NULL;
2179 }
2180 if (error == 0) {
2181 if (nd.ni_vp->v_type == VDIR) {
2182 error = EPERM; /* POSIX */
2183 goto out;
2184 }
2185 /*
2186 * The root of a mounted filesystem cannot be deleted.
2187 */
2188 if (nd.ni_vp->v_vflag & VV_ROOT) {
2189 error = EBUSY;
2190 goto out;
2191 }
2192out:
2193 if (!error) {
2194 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2195 NDFREE(&nd, NDF_ONLY_PNBUF);
2196 }
2197 }
2198 if (dirp && v3) {
2199 if (dirp == nd.ni_dvp)
2200 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2201 else {
2202 /* Drop the other locks to avoid deadlock. */
2203 if (nd.ni_dvp) {
2204 if (nd.ni_dvp == nd.ni_vp)
2205 vrele(nd.ni_dvp);
2206 else
2207 vput(nd.ni_dvp);
2208 }
2209 if (nd.ni_vp)
2210 vput(nd.ni_vp);
2211 nd.ni_dvp = NULL;
2212 nd.ni_vp = NULL;
2213
2214 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2215 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2216 VOP_UNLOCK(dirp, 0);
2217 }
2218 vrele(dirp);
2219 dirp = NULL;
2220 }
2221ereply:
2222 nfsm_reply(NFSX_WCCDATA(v3));
2223 if (v3) {
2224 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2225 error = 0;
2226 }
2227nfsmout:
2228 NDFREE(&nd, NDF_ONLY_PNBUF);
2229 if (nd.ni_dvp) {
2230 if (nd.ni_dvp == nd.ni_vp)
2231 vrele(nd.ni_dvp);
2232 else
2233 vput(nd.ni_dvp);
2234 }
2235 if (nd.ni_vp)
2236 vput(nd.ni_vp);
2237 vn_finished_write(mp);
2238 VFS_UNLOCK_GIANT(vfslocked);
2239 return(error);
2240}
2241
2242/*
2243 * nfs rename service
2244 */
2245int
2246nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2247 struct thread *td, struct mbuf **mrq)
2248{
2249 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2250 struct sockaddr *nam = nfsd->nd_nam;
2251 caddr_t dpos = nfsd->nd_dpos;
2252 struct ucred *cred = nfsd->nd_cr;
2253 caddr_t bpos;
2254 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2255 int tdirfor_ret = 1, tdiraft_ret = 1;
2256 int v3 = (nfsd->nd_flag & ND_NFSV3);
2257 struct mbuf *mb, *mreq;
2258 struct nameidata fromnd, tond;
2259 struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL;
2260 struct vnode *tdirp = NULL;
2261 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2262 nfsfh_t fnfh, tnfh;
2263 fhandle_t *ffhp, *tfhp;
2264 uid_t saved_uid;
2265 struct mount *mp = NULL;
2266 int vfslocked;
2267
2268 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2269 vfslocked = 0;
2270#ifndef nolint
2271 fvp = NULL;
2272#endif
2273 ffhp = &fnfh.fh_generic;
2274 tfhp = &tnfh.fh_generic;
2275
2276 /*
2277 * Clear fields incase goto nfsmout occurs from macro.
2278 */
2279
2280 ndclear(&fromnd);
2281 ndclear(&tond);
2282
2283 nfsm_srvmtofh(ffhp);
2284 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
2285 error = ESTALE;
2286 goto out1;
2287 }
2288 vfslocked = VFS_LOCK_GIANT(mp);
2289 (void) vn_start_write(NULL, &mp, V_WAIT);
2290 vfs_rel(mp); /* The write holds a ref. */
2291 nfsm_srvnamesiz(len);
2292 /*
2293 * Remember our original uid so that we can reset cr_uid before
2294 * the second nfs_namei() call, in case it is remapped.
2295 */
2296 saved_uid = cred->cr_uid;
2297 fromnd.ni_cnd.cn_cred = cred;
2298 fromnd.ni_cnd.cn_nameiop = DELETE;
2299 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART | MPSAFE;
2300 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
2301 &dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, td, FALSE);
2302 vfslocked = nfsrv_lockedpair_nd(vfslocked, &fromnd);
2303 if (fdirp && !v3) {
2304 vrele(fdirp);
2305 fdirp = NULL;
2306 }
2307 if (error) {
2308 nfsm_reply(2 * NFSX_WCCDATA(v3));
2309 if (v3) {
2310 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2311 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2312 }
2313 error = 0;
2314 goto nfsmout;
2315 }
2316 fvp = fromnd.ni_vp;
2317 nfsm_srvmtofh(tfhp);
2318 nfsm_srvnamesiz(len2);
2319 cred->cr_uid = saved_uid;
2320 tond.ni_cnd.cn_cred = cred;
2321 tond.ni_cnd.cn_nameiop = RENAME;
2322 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | MPSAFE;
2323 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
2324 &dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, td, FALSE);
2325 vfslocked = nfsrv_lockedpair_nd(vfslocked, &tond);
2326 if (tdirp && !v3) {
2327 vrele(tdirp);
2328 tdirp = NULL;
2329 }
2330 if (error)
2331 goto out1;
2332
2333 tdvp = tond.ni_dvp;
2334 tvp = tond.ni_vp;
2335 if (tvp != NULL) {
2336 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2337 if (v3)
2338 error = EEXIST;
2339 else
2340 error = EISDIR;
2341 goto out;
2342 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2343 if (v3)
2344 error = EEXIST;
2345 else
2346 error = ENOTDIR;
2347 goto out;
2348 }
2349 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2350 if (v3)
2351 error = EXDEV;
2352 else
2353 error = ENOTEMPTY;
2354 goto out;
2355 }
2356 }
2357 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2358 if (v3)
2359 error = EXDEV;
2360 else
2361 error = ENOTEMPTY;
2362 goto out;
2363 }
2364 if (fvp->v_mount != tdvp->v_mount) {
2365 if (v3)
2366 error = EXDEV;
2367 else
2368 error = ENOTEMPTY;
2369 goto out;
2370 }
2371 if (fvp == tdvp) {
2372 if (v3)
2373 error = EINVAL;
2374 else
2375 error = ENOTEMPTY;
2376 }
2377 /*
2378 * If source is the same as the destination (that is the
2379 * same vnode with the same name in the same directory),
2380 * then there is nothing to do.
2381 */
2382 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2383 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2384 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2385 fromnd.ni_cnd.cn_namelen))
2386 error = -1;
2387out:
2388 if (!error) {
2389 /*
2390 * The VOP_RENAME function releases all vnode references &
2391 * locks prior to returning so we need to clear the pointers
2392 * to bypass cleanup code later on.
2393 */
2394 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2395 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2396 fromnd.ni_dvp = NULL;
2397 fromnd.ni_vp = NULL;
2398 tond.ni_dvp = NULL;
2399 tond.ni_vp = NULL;
2400 if (error) {
2401 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2402 NDFREE(&tond, NDF_ONLY_PNBUF);
2403 }
2404 } else {
2405 if (error == -1)
2406 error = 0;
2407 }
2408 /* fall through */
2409out1:
2410 nfsm_reply(2 * NFSX_WCCDATA(v3));
2411 if (v3) {
2412 /* Release existing locks to prevent deadlock. */
2413 if (tond.ni_dvp) {
2414 if (tond.ni_dvp == tond.ni_vp)
2415 vrele(tond.ni_dvp);
2416 else
2417 vput(tond.ni_dvp);
2418 }
2419 if (tond.ni_vp)
2420 vput(tond.ni_vp);
2421 tond.ni_dvp = NULL;
2422 tond.ni_vp = NULL;
2423
2424 if (fdirp) {
2425 vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY);
2426 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, td);
2427 VOP_UNLOCK(fdirp, 0);
2428 }
2429 if (tdirp) {
2430 vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY);
2431 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, td);
2432 VOP_UNLOCK(tdirp, 0);
2433 }
2434 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2435 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2436 }
2437 error = 0;
2438 /* fall through */
2439
2440nfsmout:
2441 /*
2442 * Clear out tond related fields
2443 */
2444 if (tond.ni_dvp) {
2445 if (tond.ni_dvp == tond.ni_vp)
2446 vrele(tond.ni_dvp);
2447 else
2448 vput(tond.ni_dvp);
2449 }
2450 if (tond.ni_vp)
2451 vput(tond.ni_vp);
2452 if (tdirp)
2453 vrele(tdirp);
2454 if (tond.ni_startdir)
2455 vrele(tond.ni_startdir);
2456 NDFREE(&tond, NDF_ONLY_PNBUF);
2457 /*
2458 * Clear out fromnd related fields
2459 */
2460 if (fdirp)
2461 vrele(fdirp);
2462 if (fromnd.ni_startdir)
2463 vrele(fromnd.ni_startdir);
2464 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2465 if (fromnd.ni_dvp)
2466 vrele(fromnd.ni_dvp);
2467 if (fromnd.ni_vp)
2468 vrele(fromnd.ni_vp);
2469
2470 vn_finished_write(mp);
2471 VFS_UNLOCK_GIANT(vfslocked);
2472 return (error);
2473}
2474
2475/*
2476 * nfs link service
2477 */
2478int
2479nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2480 struct thread *td, struct mbuf **mrq)
2481{
2482 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2483 struct sockaddr *nam = nfsd->nd_nam;
2484 caddr_t dpos = nfsd->nd_dpos;
2485 struct ucred *cred = nfsd->nd_cr;
2486 struct nameidata nd;
2487 caddr_t bpos;
2488 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2489 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2490 struct mbuf *mb, *mreq;
2491 struct vnode *vp = NULL, *xp, *dirp = NULL;
2492 struct vattr dirfor, diraft, at;
2493 nfsfh_t nfh, dnfh;
2494 fhandle_t *fhp, *dfhp;
2495 struct mount *mp = NULL;
2496 int tvfslocked;
2497 int vfslocked;
2498
2499 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2500 ndclear(&nd);
2501 vfslocked = 0;
2502
2503 fhp = &nfh.fh_generic;
2504 dfhp = &dnfh.fh_generic;
2505 nfsm_srvmtofh(fhp);
2506 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2507 error = ESTALE;
2508 goto ereply;
2509 }
2510 vfslocked = VFS_LOCK_GIANT(mp);
2511 (void) vn_start_write(NULL, &mp, V_WAIT);
2512 vfs_rel(mp); /* The write holds a ref. */
2513 nfsm_srvmtofh(dfhp);
2514 nfsm_srvnamesiz(len);
2515
2516 error = nfsrv_fhtovp(fhp, TRUE, &vp, &tvfslocked, cred, slp,
2517 nam, &rdonly, TRUE);
2518 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
2519 if (error) {
2520 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2521 if (v3) {
2522 nfsm_srvpostop_attr(getret, &at);
2523 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2524 }
2525 vp = NULL;
2526 error = 0;
2527 goto nfsmout;
2528 }
2529 if (v3)
2530 getret = VOP_GETATTR(vp, &at, cred, td);
2531 if (vp->v_type == VDIR) {
2532 error = EPERM; /* POSIX */
2533 goto out1;
2534 }
2535 VOP_UNLOCK(vp, 0);
2536 nd.ni_cnd.cn_cred = cred;
2537 nd.ni_cnd.cn_nameiop = CREATE;
2538 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE | MPSAFE;
2539 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2540 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2541 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2542 if (dirp && !v3) {
2543 vrele(dirp);
2544 dirp = NULL;
2545 }
2546 if (error) {
2547 vrele(vp);
2548 vp = NULL;
2549 goto out2;
2550 }
2551 xp = nd.ni_vp;
2552 if (xp != NULL) {
2553 error = EEXIST;
2554 vrele(vp);
2555 vp = NULL;
2556 goto out2;
2557 }
2558 xp = nd.ni_dvp;
2559 if (vp->v_mount != xp->v_mount) {
2560 error = EXDEV;
2561 vrele(vp);
2562 vp = NULL;
2563 goto out2;
2564 }
2565 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2566 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2567 NDFREE(&nd, NDF_ONLY_PNBUF);
2568 /* fall through */
2569
2570out1:
2571 if (v3)
2572 getret = VOP_GETATTR(vp, &at, cred, td);
2573out2:
2574 if (dirp) {
2575 if (dirp == nd.ni_dvp)
2576 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2577 else {
2578 /* Release existing locks to prevent deadlock. */
2579 if (nd.ni_dvp) {
2580 if (nd.ni_dvp == nd.ni_vp)
2581 vrele(nd.ni_dvp);
2582 else
2583 vput(nd.ni_dvp);
2584 }
2585 if (nd.ni_vp)
2586 vrele(nd.ni_vp);
2587 nd.ni_dvp = NULL;
2588 nd.ni_vp = NULL;
2589
2590 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2591 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2592 VOP_UNLOCK(dirp, 0);
2593 }
2594 }
2595ereply:
2596 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2597 if (v3) {
2598 nfsm_srvpostop_attr(getret, &at);
2599 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2600 error = 0;
2601 }
2602 /* fall through */
2603
2604nfsmout:
2605 NDFREE(&nd, NDF_ONLY_PNBUF);
2606 if (vp)
2607 vput(vp);
2608 if (nd.ni_dvp) {
2609 if (nd.ni_dvp == nd.ni_vp)
2610 vrele(nd.ni_dvp);
2611 else
2612 vput(nd.ni_dvp);
2613 }
2614 if (dirp)
2615 vrele(dirp);
2616 if (nd.ni_vp)
2617 vrele(nd.ni_vp);
2618 vn_finished_write(mp);
2619 VFS_UNLOCK_GIANT(vfslocked);
2620 return(error);
2621}
2622
2623/*
2624 * nfs symbolic link service
2625 */
2626int
2627nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2628 struct thread *td, struct mbuf **mrq)
2629{
2630 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2631 struct sockaddr *nam = nfsd->nd_nam;
2632 caddr_t dpos = nfsd->nd_dpos;
2633 struct ucred *cred = nfsd->nd_cr;
2634 struct vattr va, dirfor, diraft;
2635 struct nameidata nd;
2636 struct vattr *vap = &va;
2637 struct nfsv2_sattr *sp;
2638 char *bpos, *pathcp = NULL;
2639 struct uio io;
2640 struct iovec iv;
2641 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2642 int v3 = (nfsd->nd_flag & ND_NFSV3);
2643 struct mbuf *mb, *mreq;
2644 struct vnode *dirp = NULL;
2645 nfsfh_t nfh;
2646 fhandle_t *fhp;
2647 struct mount *mp = NULL;
2648 int tvfslocked;
2649 int vfslocked;
2650
2651 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2652 ndclear(&nd);
2653 vfslocked = 0;
2654
2655 fhp = &nfh.fh_generic;
2656 nfsm_srvmtofh(fhp);
2657 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2658 error = ESTALE;
2659 goto out;
2660 }
2661 vfslocked = VFS_LOCK_GIANT(mp);
2662 (void) vn_start_write(NULL, &mp, V_WAIT);
2663 vfs_rel(mp); /* The write holds a ref. */
2664 nfsm_srvnamesiz(len);
2665 nd.ni_cnd.cn_cred = cred;
2666 nd.ni_cnd.cn_nameiop = CREATE;
2667 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART | MPSAFE;
2668 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2669 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2670 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2671 if (error == 0) {
2672 VATTR_NULL(vap);
2673 if (v3)
2674 nfsm_srvsattr(vap);
2675 nfsm_srvpathsiz(len2);
2676 }
2677 if (dirp && !v3) {
2678 vrele(dirp);
2679 dirp = NULL;
2680 }
2681 if (error)
2682 goto out;
2683 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2684 iv.iov_base = pathcp;
2685 iv.iov_len = len2;
2686 io.uio_resid = len2;
2687 io.uio_offset = 0;
2688 io.uio_iov = &iv;
2689 io.uio_iovcnt = 1;
2690 io.uio_segflg = UIO_SYSSPACE;
2691 io.uio_rw = UIO_READ;
2692 io.uio_td = NULL;
2693 nfsm_mtouio(&io, len2);
2694 if (!v3) {
2695 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
2696 vap->va_mode = nfstov_mode(sp->sa_mode);
2697 }
2698 *(pathcp + len2) = '\0';
2699 if (nd.ni_vp) {
2700 error = EEXIST;
2701 goto out;
2702 }
2703
2704 /*
2705 * issue symlink op. SAVESTART is set so the underlying path component
2706 * is only freed by the VOP if an error occurs.
2707 */
2708 if (vap->va_mode == (mode_t)VNOVAL)
2709 vap->va_mode = 0;
2710 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2711 if (error)
2712 NDFREE(&nd, NDF_ONLY_PNBUF);
2713 else
2714 vput(nd.ni_vp);
2715 nd.ni_vp = NULL;
2716 /*
2717 * releases directory prior to potential lookup op.
2718 */
2719 vput(nd.ni_dvp);
2720 nd.ni_dvp = NULL;
2721
2722 if (error == 0) {
2723 if (v3) {
2724 /*
2725 * Issue lookup. Leave SAVESTART set so we can easily free
2726 * the name buffer later on.
2727 *
2728 * since LOCKPARENT is not set, ni_dvp will be garbage on
2729 * return whether an error occurs or not.
2730 */
2731 nd.ni_cnd.cn_nameiop = LOOKUP;
2732 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2733 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2734 nd.ni_cnd.cn_thread = td;
2735 nd.ni_cnd.cn_cred = cred;
2736 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2737 if (tvfslocked)
2738 nd.ni_cnd.cn_flags |= GIANTHELD;
2739 error = lookup(&nd);
2740 nd.ni_dvp = NULL;
2741 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2742 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2743
2744 if (error == 0) {
2745 bzero((caddr_t)fhp, sizeof(nfh));
2746 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2747 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2748 if (!error)
2749 error = VOP_GETATTR(nd.ni_vp, vap, cred,
2750 td);
2751 vput(nd.ni_vp);
2752 nd.ni_vp = NULL;
2753 }
2754 }
2755 }
2756out:
2757 /*
2758 * These releases aren't strictly required, does even doing them
2759 * make any sense? XXX can nfsm_reply() block?
2760 */
2761 if (pathcp) {
2762 FREE(pathcp, M_TEMP);
2763 pathcp = NULL;
2764 }
2765 if (dirp) {
2766 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2767 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2768 VOP_UNLOCK(dirp, 0);
2769 }
2770 if (nd.ni_startdir) {
2771 vrele(nd.ni_startdir);
2772 nd.ni_startdir = NULL;
2773 }
2774 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2775 if (v3) {
2776 if (!error) {
2777 nfsm_srvpostop_fh(fhp);
2778 nfsm_srvpostop_attr(0, vap);
2779 }
2780 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2781 }
2782 error = 0;
2783 /* fall through */
2784
2785nfsmout:
2786 NDFREE(&nd, NDF_ONLY_PNBUF);
2787 if (nd.ni_dvp) {
2788 if (nd.ni_dvp == nd.ni_vp)
2789 vrele(nd.ni_dvp);
2790 else
2791 vput(nd.ni_dvp);
2792 }
2793 if (nd.ni_vp)
2794 vrele(nd.ni_vp);
2795 if (nd.ni_startdir)
2796 vrele(nd.ni_startdir);
2797 if (dirp)
2798 vrele(dirp);
2799 if (pathcp)
2800 FREE(pathcp, M_TEMP);
2801
2802 vn_finished_write(mp);
2803 VFS_UNLOCK_GIANT(vfslocked);
2804 return (error);
2805}
2806
2807/*
2808 * nfs mkdir service
2809 */
2810int
2811nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2812 struct thread *td, struct mbuf **mrq)
2813{
2814 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2815 struct sockaddr *nam = nfsd->nd_nam;
2816 caddr_t dpos = nfsd->nd_dpos;
2817 struct ucred *cred = nfsd->nd_cr;
2818 struct vattr va, dirfor, diraft;
2819 struct vattr *vap = &va;
2820 struct nfs_fattr *fp;
2821 struct nameidata nd;
2822 u_int32_t *tl;
2823 caddr_t bpos;
2824 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2825 int v3 = (nfsd->nd_flag & ND_NFSV3);
2826 struct mbuf *mb, *mreq;
2827 struct vnode *dirp = NULL;
2828 int vpexcl = 0;
2829 nfsfh_t nfh;
2830 fhandle_t *fhp;
2831 struct mount *mp = NULL;
2832 int vfslocked;
2833
2834 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2835 ndclear(&nd);
2836 vfslocked = 0;
2837
2838 fhp = &nfh.fh_generic;
2839 nfsm_srvmtofh(fhp);
2840 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2841 error = ESTALE;
2842 goto out;
2843 }
2844 vfslocked = VFS_LOCK_GIANT(mp);
2845 (void) vn_start_write(NULL, &mp, V_WAIT);
2846 vfs_rel(mp); /* The write holds a ref. */
2847 nfsm_srvnamesiz(len);
2848 nd.ni_cnd.cn_cred = cred;
2849 nd.ni_cnd.cn_nameiop = CREATE;
2850 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE;
2851
2852 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2853 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2854 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2855 if (dirp && !v3) {
2856 vrele(dirp);
2857 dirp = NULL;
2858 }
2859 if (error) {
2860 nfsm_reply(NFSX_WCCDATA(v3));
2861 if (v3)
2862 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2863 error = 0;
2864 goto nfsmout;
2865 }
2866 VATTR_NULL(vap);
2867 if (v3) {
2868 nfsm_srvsattr(vap);
2869 } else {
2870 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2871 vap->va_mode = nfstov_mode(*tl++);
2872 }
2873
2874 /*
2875 * At this point nd.ni_dvp is referenced and exclusively locked and
2876 * nd.ni_vp, if it exists, is referenced but not locked.
2877 */
2878
2879 vap->va_type = VDIR;
2880 if (nd.ni_vp != NULL) {
2881 NDFREE(&nd, NDF_ONLY_PNBUF);
2882 error = EEXIST;
2883 goto out;
2884 }
2885
2886 /*
2887 * Issue mkdir op. Since SAVESTART is not set, the pathname
2888 * component is freed by the VOP call. This will fill-in
2889 * nd.ni_vp, reference, and exclusively lock it.
2890 */
2891 if (vap->va_mode == (mode_t)VNOVAL)
2892 vap->va_mode = 0;
2893 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2894 NDFREE(&nd, NDF_ONLY_PNBUF);
2895 vpexcl = 1;
2896
2897 vput(nd.ni_dvp);
2898 nd.ni_dvp = NULL;
2899
2900 if (!error) {
2901 bzero((caddr_t)fhp, sizeof(nfh));
2902 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2903 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2904 if (!error)
2905 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
2906 }
2907out:
2908 if (dirp) {
2909 if (dirp == nd.ni_dvp) {
2910 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2911 } else {
2912 /* Release existing locks to prevent deadlock. */
2913 if (nd.ni_dvp) {
2914 NDFREE(&nd, NDF_ONLY_PNBUF);
2915 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2916 vrele(nd.ni_dvp);
2917 else
2918 vput(nd.ni_dvp);
2919 }
2920 if (nd.ni_vp) {
2921 if (vpexcl)
2922 vput(nd.ni_vp);
2923 else
2924 vrele(nd.ni_vp);
2925 }
2926 nd.ni_dvp = NULL;
2927 nd.ni_vp = NULL;
2928 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2929 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2930 VOP_UNLOCK(dirp, 0);
2931 }
2932 }
2933 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2934 if (v3) {
2935 if (!error) {
2936 nfsm_srvpostop_fh(fhp);
2937 nfsm_srvpostop_attr(0, vap);
2938 }
2939 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2940 } else if (!error) {
2941 /* v2 non-error case. */
2942 nfsm_srvfhtom(fhp, v3);
2943 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2944 nfsm_srvfillattr(vap, fp);
2945 }
2946 error = 0;
2947 /* fall through */
2948
2949nfsmout:
2950 if (nd.ni_dvp) {
2951 NDFREE(&nd, NDF_ONLY_PNBUF);
2952 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2953 vrele(nd.ni_dvp);
2954 else
2955 vput(nd.ni_dvp);
2956 }
2957 if (nd.ni_vp) {
2958 if (vpexcl)
2959 vput(nd.ni_vp);
2960 else
2961 vrele(nd.ni_vp);
2962 }
2963 if (dirp)
2964 vrele(dirp);
2965 vn_finished_write(mp);
2966 VFS_UNLOCK_GIANT(vfslocked);
2967 return (error);
2968}
2969
2970/*
2971 * nfs rmdir service
2972 */
2973int
2974nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2975 struct thread *td, struct mbuf **mrq)
2976{
2977 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2978 struct sockaddr *nam = nfsd->nd_nam;
2979 caddr_t dpos = nfsd->nd_dpos;
2980 struct ucred *cred = nfsd->nd_cr;
2981 caddr_t bpos;
2982 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2983 int v3 = (nfsd->nd_flag & ND_NFSV3);
2984 struct mbuf *mb, *mreq;
2985 struct vnode *vp, *dirp = NULL;
2986 struct vattr dirfor, diraft;
2987 nfsfh_t nfh;
2988 fhandle_t *fhp;
2989 struct nameidata nd;
2990 struct mount *mp = NULL;
2991 int vfslocked;
2992
2993 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2994 ndclear(&nd);
2995 vfslocked = 0;
2996
2997 fhp = &nfh.fh_generic;
2998 nfsm_srvmtofh(fhp);
2999 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3000 error = ESTALE;
3001 goto out;
3002 }
3003 vfslocked = VFS_LOCK_GIANT(mp);
3004 (void) vn_start_write(NULL, &mp, V_WAIT);
3005 vfs_rel(mp); /* The write holds a ref. */
3006 nfsm_srvnamesiz(len);
3007 nd.ni_cnd.cn_cred = cred;
3008 nd.ni_cnd.cn_nameiop = DELETE;
3009 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
3010 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
3011 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
3012 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
3013 if (dirp && !v3) {
3014 vrele(dirp);
3015 dirp = NULL;
3016 }
3017 if (error) {
3018 nfsm_reply(NFSX_WCCDATA(v3));
3019 if (v3)
3020 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3021 error = 0;
3022 goto nfsmout;
3023 }
3024 vp = nd.ni_vp;
3025 if (vp->v_type != VDIR) {
3026 error = ENOTDIR;
3027 goto out;
3028 }
3029 /*
3030 * No rmdir "." please.
3031 */
3032 if (nd.ni_dvp == vp) {
3033 error = EINVAL;
3034 goto out;
3035 }
3036 /*
3037 * The root of a mounted filesystem cannot be deleted.
3038 */
3039 if (vp->v_vflag & VV_ROOT)
3040 error = EBUSY;
3041out:
3042 /*
3043 * Issue or abort op. Since SAVESTART is not set, path name
3044 * component is freed by the VOP after either.
3045 */
3046 if (!error)
3047 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
3048 NDFREE(&nd, NDF_ONLY_PNBUF);
3049
3050 if (dirp) {
3051 if (dirp == nd.ni_dvp)
3052 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
3053 else {
3054 /* Release existing locks to prevent deadlock. */
3055 if (nd.ni_dvp) {
3056 if (nd.ni_dvp == nd.ni_vp)
3057 vrele(nd.ni_dvp);
3058 else
3059 vput(nd.ni_dvp);
3060 }
3061 if (nd.ni_vp)
3062 vput(nd.ni_vp);
3063 nd.ni_dvp = NULL;
3064 nd.ni_vp = NULL;
3065 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
3066 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
3067 VOP_UNLOCK(dirp, 0);
3068 }
3069 }
3070 nfsm_reply(NFSX_WCCDATA(v3));
3071 error = 0;
3072 if (v3)
3073 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3074 /* fall through */
3075
3076nfsmout:
3077 NDFREE(&nd, NDF_ONLY_PNBUF);
3078 if (nd.ni_dvp) {
3079 if (nd.ni_dvp == nd.ni_vp)
3080 vrele(nd.ni_dvp);
3081 else
3082 vput(nd.ni_dvp);
3083 }
3084 if (nd.ni_vp)
3085 vput(nd.ni_vp);
3086 if (dirp)
3087 vrele(dirp);
3088
3089 vn_finished_write(mp);
3090 VFS_UNLOCK_GIANT(vfslocked);
3091 return(error);
3092}
3093
3094/*
3095 * nfs readdir service
3096 * - mallocs what it thinks is enough to read
3097 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
3098 * - calls VOP_READDIR()
3099 * - loops around building the reply
3100 * if the output generated exceeds count break out of loop
3101 * The nfsm_clget macro is used here so that the reply will be packed
3102 * tightly in mbuf clusters.
3103 * - it only knows that it has encountered eof when the VOP_READDIR()
3104 * reads nothing
3105 * - as such one readdir rpc will return eof false although you are there
3106 * and then the next will return eof
3107 * - it trims out records with d_fileno == 0
3108 * this doesn't matter for Unix clients, but they might confuse clients
3109 * for other os'.
3110 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
3111 * than requested, but this may not apply to all filesystems. For
3112 * example, client NFS does not { although it is never remote mounted
3113 * anyhow }
3114 * The alternate call nfsrv_readdirplus() does lookups as well.
3115 * PS: The NFS protocol spec. does not clarify what the "count" byte
3116 * argument is a count of.. just name strings and file id's or the
3117 * entire reply rpc or ...
3118 * I tried just file name and id sizes and it confused the Sun client,
3119 * so I am using the full rpc size now. The "paranoia.." comment refers
3120 * to including the status longwords that are not a part of the dir.
3121 * "entry" structures, but are in the rpc.
3122 */
3123struct flrep {
3124 nfsuint64 fl_off;
3125 u_int32_t fl_postopok;
3126 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
3127 u_int32_t fl_fhok;
3128 u_int32_t fl_fhsize;
3129 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
3130};
3131
3132int
3133nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3134 struct thread *td, struct mbuf **mrq)
3135{
3136 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3137 struct sockaddr *nam = nfsd->nd_nam;
3138 caddr_t dpos = nfsd->nd_dpos;
3139 struct ucred *cred = nfsd->nd_cr;
3140 char *bp, *be;
3141 struct mbuf *mp;
3142 struct dirent *dp;
3143 caddr_t cp;
3144 u_int32_t *tl;
3145 caddr_t bpos;
3146 struct mbuf *mb, *mreq;
3147 char *cpos, *cend, *rbuf;
3148 struct vnode *vp = NULL;
3149 struct vattr at;
3150 nfsfh_t nfh;
3151 fhandle_t *fhp;
3152 struct uio io;
3153 struct iovec iv;
3154 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3155 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
3156 int v3 = (nfsd->nd_flag & ND_NFSV3);
3157 u_quad_t off, toff, verf;
3158 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3159 int vfslocked;
3160
3161 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3162 vfslocked = 0;
3163 fhp = &nfh.fh_generic;
3164 nfsm_srvmtofh(fhp);
3165 if (v3) {
3166 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
3167 toff = fxdr_hyper(tl);
3168 tl += 2;
3169 verf = fxdr_hyper(tl);
3170 tl += 2;
3171 } else {
3172 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
3173 toff = fxdr_unsigned(u_quad_t, *tl++);
3174 verf = 0; /* shut up gcc */
3175 }
3176 off = toff;
3177 cnt = fxdr_unsigned(int, *tl);
3178 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3179 xfer = NFS_SRVMAXDATA(nfsd);
3180 if (cnt > xfer)
3181 cnt = xfer;
3182 if (siz > xfer)
3183 siz = xfer;
3184 fullsiz = siz;
3185 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3186 nam, &rdonly, TRUE);
3187 if (!error && vp->v_type != VDIR) {
3188 error = ENOTDIR;
3189 vput(vp);
3190 vp = NULL;
3191 }
3192 if (error) {
3193 nfsm_reply(NFSX_UNSIGNED);
3194 if (v3)
3195 nfsm_srvpostop_attr(getret, &at);
3196 error = 0;
3197 goto nfsmout;
3198 }
3199
3200 /*
3201 * Obtain lock on vnode for this section of the code
3202 */
3203 if (v3) {
3204 error = getret = VOP_GETATTR(vp, &at, cred, td);
3205#if 0
3206 /*
3207 * XXX This check may be too strict for Solaris 2.5 clients.
3208 */
3209 if (!error && toff && verf && verf != at.va_filerev)
3210 error = NFSERR_BAD_COOKIE;
3211#endif
3212 }
3213 if (!error)
3214 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3215 if (error) {
3216 vput(vp);
3217 vp = NULL;
3218 nfsm_reply(NFSX_POSTOPATTR(v3));
3219 if (v3)
3220 nfsm_srvpostop_attr(getret, &at);
3221 error = 0;
3222 goto nfsmout;
3223 }
3224 VOP_UNLOCK(vp, 0);
3225
3226 /*
3227 * end section. Allocate rbuf and continue
3228 */
3229 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3230again:
3231 iv.iov_base = rbuf;
3232 iv.iov_len = fullsiz;
3233 io.uio_iov = &iv;
3234 io.uio_iovcnt = 1;
3235 io.uio_offset = (off_t)off;
3236 io.uio_resid = fullsiz;
3237 io.uio_segflg = UIO_SYSSPACE;
3238 io.uio_rw = UIO_READ;
3239 io.uio_td = NULL;
3240 eofflag = 0;
3241 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3242 if (cookies) {
3243 free((caddr_t)cookies, M_TEMP);
3244 cookies = NULL;
3245 }
3246 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3247 off = (off_t)io.uio_offset;
3248 if (!cookies && !error)
3249 error = NFSERR_PERM;
3250 if (v3) {
3251 getret = VOP_GETATTR(vp, &at, cred, td);
3252 if (!error)
3253 error = getret;
3254 }
3255 VOP_UNLOCK(vp, 0);
3256 if (error) {
3257 vrele(vp);
3258 vp = NULL;
3259 free((caddr_t)rbuf, M_TEMP);
3260 if (cookies)
3261 free((caddr_t)cookies, M_TEMP);
3262 nfsm_reply(NFSX_POSTOPATTR(v3));
3263 if (v3)
3264 nfsm_srvpostop_attr(getret, &at);
3265 error = 0;
3266 goto nfsmout;
3267 }
3268 if (io.uio_resid) {
3269 siz -= io.uio_resid;
3270
3271 /*
3272 * If nothing read, return eof
3273 * rpc reply
3274 */
3275 if (siz == 0) {
3276 vrele(vp);
3277 vp = NULL;
3278 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
3279 2 * NFSX_UNSIGNED);
3280 if (v3) {
3281 nfsm_srvpostop_attr(getret, &at);
3282 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3283 txdr_hyper(at.va_filerev, tl);
3284 tl += 2;
3285 } else
3286 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3287 *tl++ = nfsrv_nfs_false;
3288 *tl = nfsrv_nfs_true;
3289 FREE((caddr_t)rbuf, M_TEMP);
3290 FREE((caddr_t)cookies, M_TEMP);
3291 error = 0;
3292 goto nfsmout;
3293 }
3294 }
3295
3296 /*
3297 * Check for degenerate cases of nothing useful read.
3298 * If so go try again
3299 */
3300 cpos = rbuf;
3301 cend = rbuf + siz;
3302 dp = (struct dirent *)cpos;
3303 cookiep = cookies;
3304 /*
3305 * For some reason FreeBSD's ufs_readdir() chooses to back the
3306 * directory offset up to a block boundary, so it is necessary to
3307 * skip over the records that precede the requested offset. This
3308 * requires the assumption that file offset cookies monotonically
3309 * increase.
3310 */
3311 while (cpos < cend && ncookies > 0 &&
3312 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3313 ((u_quad_t)(*cookiep)) <= toff)) {
3314 cpos += dp->d_reclen;
3315 dp = (struct dirent *)cpos;
3316 cookiep++;
3317 ncookies--;
3318 }
3319 if (cpos >= cend || ncookies == 0) {
3320 toff = off;
3321 siz = fullsiz;
3322 goto again;
3323 }
3324
3325 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
3326 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3327 if (v3) {
3328 nfsm_srvpostop_attr(getret, &at);
3329 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3330 txdr_hyper(at.va_filerev, tl);
3331 }
3332 mp = mb;
3333 bp = bpos;
3334 be = bp + M_TRAILINGSPACE(mp);
3335
3336 /* Loop through the records and build reply */
3337 while (cpos < cend && ncookies > 0) {
3338 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3339 nlen = dp->d_namlen;
3340 rem = nfsm_rndup(nlen) - nlen;
3341 len += (4 * NFSX_UNSIGNED + nlen + rem);
3342 if (v3)
3343 len += 2 * NFSX_UNSIGNED;
3344 if (len > cnt) {
3345 eofflag = 0;
3346 break;
3347 }
3348 /*
3349 * Build the directory record xdr from
3350 * the dirent entry.
3351 */
3352 nfsm_clget;
3353 *tl = nfsrv_nfs_true;
3354 bp += NFSX_UNSIGNED;
3355 if (v3) {
3356 nfsm_clget;
3357 *tl = 0;
3358 bp += NFSX_UNSIGNED;
3359 }
3360 nfsm_clget;
3361 *tl = txdr_unsigned(dp->d_fileno);
3362 bp += NFSX_UNSIGNED;
3363 nfsm_clget;
3364 *tl = txdr_unsigned(nlen);
3365 bp += NFSX_UNSIGNED;
3366
3367 /* And loop around copying the name */
3368 xfer = nlen;
3369 cp = dp->d_name;
3370 while (xfer > 0) {
3371 nfsm_clget;
3372 if ((bp+xfer) > be)
3373 tsiz = be-bp;
3374 else
3375 tsiz = xfer;
3376 bcopy(cp, bp, tsiz);
3377 bp += tsiz;
3378 xfer -= tsiz;
3379 if (xfer > 0)
3380 cp += tsiz;
3381 }
3382 /* And null pad to an int32_t boundary. */
3383 for (i = 0; i < rem; i++)
3384 *bp++ = '\0';
3385 nfsm_clget;
3386
3387 /* Finish off the record */
3388 if (v3) {
3389 *tl = 0;
3390 bp += NFSX_UNSIGNED;
3391 nfsm_clget;
3392 }
3393 *tl = txdr_unsigned(*cookiep);
3394 bp += NFSX_UNSIGNED;
3395 }
3396 cpos += dp->d_reclen;
3397 dp = (struct dirent *)cpos;
3398 cookiep++;
3399 ncookies--;
3400 }
3401 vrele(vp);
3402 vp = NULL;
3403 nfsm_clget;
3404 *tl = nfsrv_nfs_false;
3405 bp += NFSX_UNSIGNED;
3406 nfsm_clget;
3407 if (eofflag)
3408 *tl = nfsrv_nfs_true;
3409 else
3410 *tl = nfsrv_nfs_false;
3411 bp += NFSX_UNSIGNED;
3412 if (mp != mb) {
3413 if (bp < be)
3414 mp->m_len = bp - mtod(mp, caddr_t);
3415 } else
3416 mp->m_len += bp - bpos;
3417 FREE((caddr_t)rbuf, M_TEMP);
3418 FREE((caddr_t)cookies, M_TEMP);
3419
3420nfsmout:
3421 if (vp)
3422 vrele(vp);
3423 VFS_UNLOCK_GIANT(vfslocked);
3424 return(error);
3425}
3426
3427int
3428nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3429 struct thread *td, struct mbuf **mrq)
3430{
3431 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3432 struct sockaddr *nam = nfsd->nd_nam;
3433 caddr_t dpos = nfsd->nd_dpos;
3434 struct ucred *cred = nfsd->nd_cr;
3435 char *bp, *be;
3436 struct mbuf *mp;
3437 struct dirent *dp;
3438 caddr_t cp;
3439 u_int32_t *tl;
3440 caddr_t bpos;
3441 struct mbuf *mb, *mreq;
3442 char *cpos, *cend, *rbuf;
3443 struct vnode *vp = NULL, *nvp;
3444 struct flrep fl;
3445 nfsfh_t nfh;
3446 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3447 struct uio io;
3448 struct iovec iv;
3449 struct vattr va, at, *vap = &va;
3450 struct nfs_fattr *fp;
3451 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3452 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3453 u_quad_t off, toff, verf;
3454 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3455 int v3 = (nfsd->nd_flag & ND_NFSV3);
3456 int vfslocked;
3457
3458 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3459 vfslocked = 0;
3460 if (!v3)
3461 panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
3462 fhp = &nfh.fh_generic;
3463 nfsm_srvmtofh(fhp);
3464 tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED);
3465 toff = fxdr_hyper(tl);
3466 tl += 2;
3467 verf = fxdr_hyper(tl);
3468 tl += 2;
3469 siz = fxdr_unsigned(int, *tl++);
3470 cnt = fxdr_unsigned(int, *tl);
3471 off = toff;
3472 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3473 xfer = NFS_SRVMAXDATA(nfsd);
3474 if (cnt > xfer)
3475 cnt = xfer;
3476 if (siz > xfer)
3477 siz = xfer;
3478 fullsiz = siz;
3479 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3480 nam, &rdonly, TRUE);
3481 if (!error && vp->v_type != VDIR) {
3482 error = ENOTDIR;
3483 vput(vp);
3484 vp = NULL;
3485 }
3486 if (error) {
3487 nfsm_reply(NFSX_UNSIGNED);
3488 nfsm_srvpostop_attr(getret, &at);
3489 error = 0;
3490 goto nfsmout;
3491 }
3492 error = getret = VOP_GETATTR(vp, &at, cred, td);
3493#if 0
3494 /*
3495 * XXX This check may be too strict for Solaris 2.5 clients.
3496 */
3497 if (!error && toff && verf && verf != at.va_filerev)
3498 error = NFSERR_BAD_COOKIE;
3499#endif
3500 if (!error)
3501 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3502 if (error) {
3503 vput(vp);
3504 vp = NULL;
3505 nfsm_reply(NFSX_V3POSTOPATTR);
3506 nfsm_srvpostop_attr(getret, &at);
3507 error = 0;
3508 goto nfsmout;
3509 }
3510 VOP_UNLOCK(vp, 0);
3511 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3512again:
3513 iv.iov_base = rbuf;
3514 iv.iov_len = fullsiz;
3515 io.uio_iov = &iv;
3516 io.uio_iovcnt = 1;
3517 io.uio_offset = (off_t)off;
3518 io.uio_resid = fullsiz;
3519 io.uio_segflg = UIO_SYSSPACE;
3520 io.uio_rw = UIO_READ;
3521 io.uio_td = NULL;
3522 eofflag = 0;
3523 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3524 if (cookies) {
3525 free((caddr_t)cookies, M_TEMP);
3526 cookies = NULL;
3527 }
3528 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3529 off = (u_quad_t)io.uio_offset;
3530 getret = VOP_GETATTR(vp, &at, cred, td);
3531 VOP_UNLOCK(vp, 0);
3532 if (!cookies && !error)
3533 error = NFSERR_PERM;
3534 if (!error)
3535 error = getret;
3536 if (error) {
3537 vrele(vp);
3538 vp = NULL;
3539 if (cookies)
3540 free((caddr_t)cookies, M_TEMP);
3541 free((caddr_t)rbuf, M_TEMP);
3542 nfsm_reply(NFSX_V3POSTOPATTR);
3543 nfsm_srvpostop_attr(getret, &at);
3544 error = 0;
3545 goto nfsmout;
3546 }
3547 if (io.uio_resid) {
3548 siz -= io.uio_resid;
3549
3550 /*
3551 * If nothing read, return eof
3552 * rpc reply
3553 */
3554 if (siz == 0) {
3555 vrele(vp);
3556 vp = NULL;
3557 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3558 2 * NFSX_UNSIGNED);
3559 nfsm_srvpostop_attr(getret, &at);
3560 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3561 txdr_hyper(at.va_filerev, tl);
3562 tl += 2;
3563 *tl++ = nfsrv_nfs_false;
3564 *tl = nfsrv_nfs_true;
3565 FREE((caddr_t)cookies, M_TEMP);
3566 FREE((caddr_t)rbuf, M_TEMP);
3567 error = 0;
3568 goto nfsmout;
3569 }
3570 }
3571
3572 /*
3573 * Check for degenerate cases of nothing useful read.
3574 * If so go try again
3575 */
3576 cpos = rbuf;
3577 cend = rbuf + siz;
3578 dp = (struct dirent *)cpos;
3579 cookiep = cookies;
3580 /*
3581 * For some reason FreeBSD's ufs_readdir() chooses to back the
3582 * directory offset up to a block boundary, so it is necessary to
3583 * skip over the records that precede the requested offset. This
3584 * requires the assumption that file offset cookies monotonically
3585 * increase.
3586 */
3587 while (cpos < cend && ncookies > 0 &&
3588 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3589 ((u_quad_t)(*cookiep)) <= toff)) {
3590 cpos += dp->d_reclen;
3591 dp = (struct dirent *)cpos;
3592 cookiep++;
3593 ncookies--;
3594 }
3595 if (cpos >= cend || ncookies == 0) {
3596 toff = off;
3597 siz = fullsiz;
3598 goto again;
3599 }
3600
3601 /*
3602 * Probe one of the directory entries to see if the filesystem
3603 * supports VGET.
3604 */
3605 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE, &nvp) ==
3606 EOPNOTSUPP) {
3607 error = NFSERR_NOTSUPP;
3608 vrele(vp);
3609 vp = NULL;
3610 free((caddr_t)cookies, M_TEMP);
3611 free((caddr_t)rbuf, M_TEMP);
3612 nfsm_reply(NFSX_V3POSTOPATTR);
3613 nfsm_srvpostop_attr(getret, &at);
3614 error = 0;
3615 goto nfsmout;
3616 }
3617 vput(nvp);
3618 nvp = NULL;
3619
3620 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3621 2 * NFSX_UNSIGNED;
3622 nfsm_reply(cnt);
3623 nfsm_srvpostop_attr(getret, &at);
3624 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3625 txdr_hyper(at.va_filerev, tl);
3626 mp = mb;
3627 bp = bpos;
3628 be = bp + M_TRAILINGSPACE(mp);
3629
3630 /* Loop through the records and build reply */
3631 while (cpos < cend && ncookies > 0) {
3632 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3633 nlen = dp->d_namlen;
3634 rem = nfsm_rndup(nlen)-nlen;
3635
3636 /*
3637 * For readdir_and_lookup get the vnode using
3638 * the file number.
3639 */
3640 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE,
3641 &nvp))
3642 goto invalid;
3643 bzero((caddr_t)nfhp, NFSX_V3FH);
3644 nfhp->fh_fsid =
3645 nvp->v_mount->mnt_stat.f_fsid;
3646 /*
3647 * XXXRW: Assert the mountpoints are the same so that
3648 * we know that acquiring Giant based on the
3649 * directory is the right thing for the child.
3650 */
3651 KASSERT(nvp->v_mount == vp->v_mount,
3652 ("nfsrv_readdirplus: nvp mount != vp mount"));
3653 if (VOP_VPTOFH(nvp, &nfhp->fh_fid)) {
3654 vput(nvp);
3655 nvp = NULL;
3656 goto invalid;
3657 }
3658 if (VOP_GETATTR(nvp, vap, cred, td)) {
3659 vput(nvp);
3660 nvp = NULL;
3661 goto invalid;
3662 }
3663 vput(nvp);
3664 nvp = NULL;
3665
3666 /*
3667 * If either the dircount or maxcount will be
3668 * exceeded, get out now. Both of these lengths
3669 * are calculated conservatively, including all
3670 * XDR overheads.
3671 */
3672 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3673 NFSX_V3POSTOPATTR);
3674 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3675 if (len > cnt || dirlen > fullsiz) {
3676 eofflag = 0;
3677 break;
3678 }
3679
3680 /*
3681 * Build the directory record xdr from
3682 * the dirent entry.
3683 */
3684 fp = (struct nfs_fattr *)&fl.fl_fattr;
3685 nfsm_srvfillattr(vap, fp);
3686 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3687 fl.fl_fhok = nfsrv_nfs_true;
3688 fl.fl_postopok = nfsrv_nfs_true;
3689 fl.fl_off.nfsuquad[0] = 0;
3690 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3691
3692 nfsm_clget;
3693 *tl = nfsrv_nfs_true;
3694 bp += NFSX_UNSIGNED;
3695 nfsm_clget;
3696 *tl = 0;
3697 bp += NFSX_UNSIGNED;
3698 nfsm_clget;
3699 *tl = txdr_unsigned(dp->d_fileno);
3700 bp += NFSX_UNSIGNED;
3701 nfsm_clget;
3702 *tl = txdr_unsigned(nlen);
3703 bp += NFSX_UNSIGNED;
3704
3705 /* And loop around copying the name */
3706 xfer = nlen;
3707 cp = dp->d_name;
3708 while (xfer > 0) {
3709 nfsm_clget;
3710 if ((bp + xfer) > be)
3711 tsiz = be - bp;
3712 else
3713 tsiz = xfer;
3714 bcopy(cp, bp, tsiz);
3715 bp += tsiz;
3716 xfer -= tsiz;
3717 if (xfer > 0)
3718 cp += tsiz;
3719 }
3720 /* And null pad to an int32_t boundary. */
3721 for (i = 0; i < rem; i++)
3722 *bp++ = '\0';
3723
3724 /*
3725 * Now copy the flrep structure out.
3726 */
3727 xfer = sizeof (struct flrep);
3728 cp = (caddr_t)&fl;
3729 while (xfer > 0) {
3730 nfsm_clget;
3731 if ((bp + xfer) > be)
3732 tsiz = be - bp;
3733 else
3734 tsiz = xfer;
3735 bcopy(cp, bp, tsiz);
3736 bp += tsiz;
3737 xfer -= tsiz;
3738 if (xfer > 0)
3739 cp += tsiz;
3740 }
3741 }
3742invalid:
3743 cpos += dp->d_reclen;
3744 dp = (struct dirent *)cpos;
3745 cookiep++;
3746 ncookies--;
3747 }
3748 vrele(vp);
3749 vp = NULL;
3750 nfsm_clget;
3751 *tl = nfsrv_nfs_false;
3752 bp += NFSX_UNSIGNED;
3753 nfsm_clget;
3754 if (eofflag)
3755 *tl = nfsrv_nfs_true;
3756 else
3757 *tl = nfsrv_nfs_false;
3758 bp += NFSX_UNSIGNED;
3759 if (mp != mb) {
3760 if (bp < be)
3761 mp->m_len = bp - mtod(mp, caddr_t);
3762 } else
3763 mp->m_len += bp - bpos;
3764 FREE((caddr_t)cookies, M_TEMP);
3765 FREE((caddr_t)rbuf, M_TEMP);
3766nfsmout:
3767 if (vp)
3768 vrele(vp);
3769 VFS_UNLOCK_GIANT(vfslocked);
3770 return(error);
3771}
3772
3773/*
3774 * nfs commit service
3775 */
3776int
3777nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3778 struct thread *td, struct mbuf **mrq)
3779{
3780 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3781 struct sockaddr *nam = nfsd->nd_nam;
3782 caddr_t dpos = nfsd->nd_dpos;
3783 struct ucred *cred = nfsd->nd_cr;
3784 struct vattr bfor, aft;
3785 struct vnode *vp = NULL;
3786 nfsfh_t nfh;
3787 fhandle_t *fhp;
3788 u_int32_t *tl;
3789 caddr_t bpos;
3790 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3791 struct mbuf *mb, *mreq;
3792 u_quad_t off;
3793 struct mount *mp = NULL;
3794 int v3 = (nfsd->nd_flag & ND_NFSV3);
3795 int tvfslocked;
3796 int vfslocked;
3797
3798 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3799 vfslocked = 0;
3800 if (!v3)
3801 panic("nfsrv_commit: v3 proc called on a v2 connection");
3802 fhp = &nfh.fh_generic;
3803 nfsm_srvmtofh(fhp);
3804 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3805 error = ESTALE;
3806 goto ereply;
3807 }
3808 vfslocked = VFS_LOCK_GIANT(mp);
3809 (void) vn_start_write(NULL, &mp, V_WAIT);
3810 vfs_rel(mp); /* The write holds a ref. */
3811 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
3812
3813 /*
3814 * XXX At this time VOP_FSYNC() does not accept offset and byte
3815 * count parameters, so these arguments are useless (someday maybe).
3816 */
3817 off = fxdr_hyper(tl);
3818 tl += 2;
3819 cnt = fxdr_unsigned(int, *tl);
3820 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
3821 nam, &rdonly, TRUE);
3822 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
3823 if (error) {
3824 nfsm_reply(2 * NFSX_UNSIGNED);
3825 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3826 error = 0;
3827 goto nfsmout;
3828 }
3829 for_ret = VOP_GETATTR(vp, &bfor, cred, td);
3830
3831 if (cnt > MAX_COMMIT_COUNT) {
3832 /*
3833 * Give up and do the whole thing
3834 */
3835 if (vp->v_object &&
3836 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3837 VM_OBJECT_LOCK(vp->v_object);
3838 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3839 VM_OBJECT_UNLOCK(vp->v_object);
3840 }
3841 error = VOP_FSYNC(vp, MNT_WAIT, td);
3842 } else {
3843 /*
3844 * Locate and synchronously write any buffers that fall
3845 * into the requested range. Note: we are assuming that
3846 * f_iosize is a power of 2.
3847 */
3848 int iosize = vp->v_mount->mnt_stat.f_iosize;
3849 int iomask = iosize - 1;
3850 struct bufobj *bo;
3851 daddr_t lblkno;
3852
3853 /*
3854 * Align to iosize boundry, super-align to page boundry.
3855 */
3856 if (off & iomask) {
3857 cnt += off & iomask;
3858 off &= ~(u_quad_t)iomask;
3859 }
3860 if (off & PAGE_MASK) {
3861 cnt += off & PAGE_MASK;
3862 off &= ~(u_quad_t)PAGE_MASK;
3863 }
3864 lblkno = off / iosize;
3865
3866 if (vp->v_object &&
3867 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3868 VM_OBJECT_LOCK(vp->v_object);
3869 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3870 VM_OBJECT_UNLOCK(vp->v_object);
3871 }
3872
3873 bo = &vp->v_bufobj;
3874 BO_LOCK(bo);
3875 while (cnt > 0) {
3876 struct buf *bp;
3877
3878 /*
3879 * If we have a buffer and it is marked B_DELWRI we
3880 * have to lock and write it. Otherwise the prior
3881 * write is assumed to have already been committed.
3882 *
3883 * gbincore() can return invalid buffers now so we
3884 * have to check that bit as well (though B_DELWRI
3885 * should not be set if B_INVAL is set there could be
3886 * a race here since we haven't locked the buffer).
3887 */
3888 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
3889 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
3890 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
3891 BO_LOCK(bo);
3892 continue; /* retry */
3893 }
3894 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
3895 B_DELWRI) {
3896 bremfree(bp);
3897 bp->b_flags &= ~B_ASYNC;
3898 bwrite(bp);
3899 ++nfs_commit_miss;
3900 } else
3901 BUF_UNLOCK(bp);
3902 BO_LOCK(bo);
3903 }
3904 ++nfs_commit_blks;
3905 if (cnt < iosize)
3906 break;
3907 cnt -= iosize;
3908 ++lblkno;
3909 }
3910 BO_UNLOCK(bo);
3911 }
3912
3913 aft_ret = VOP_GETATTR(vp, &aft, cred, td);
3914 vput(vp);
3915 vp = NULL;
3916ereply:
3917 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3918 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3919 if (!error) {
3920 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3921 if (nfsver.tv_sec == 0)
3922 nfsver = boottime;
3923 *tl++ = txdr_unsigned(nfsver.tv_sec);
3924 *tl = txdr_unsigned(nfsver.tv_usec);
3925 } else {
3926 error = 0;
3927 }
3928nfsmout:
3929 if (vp)
3930 vput(vp);
3931 vn_finished_write(mp);
3932 VFS_UNLOCK_GIANT(vfslocked);
3933 return(error);
3934}
3935
3936/*
3937 * nfs statfs service
3938 */
3939int
3940nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3941 struct thread *td, struct mbuf **mrq)
3942{
3943 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3944 struct sockaddr *nam = nfsd->nd_nam;
3945 caddr_t dpos = nfsd->nd_dpos;
3946 struct ucred *cred = nfsd->nd_cr;
3947 struct statfs *sf;
3948 struct nfs_statfs *sfp;
3949 caddr_t bpos;
3950 int error = 0, rdonly, getret = 1;
3951 int v3 = (nfsd->nd_flag & ND_NFSV3);
3952 struct mbuf *mb, *mreq;
3953 struct vnode *vp = NULL;
3954 struct vattr at;
3955 nfsfh_t nfh;
3956 fhandle_t *fhp;
3957 struct statfs statfs;
3958 u_quad_t tval;
3959 int vfslocked;
3960
3961 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3962 vfslocked = 0;
3963 fhp = &nfh.fh_generic;
3964 nfsm_srvmtofh(fhp);
3965 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3966 nam, &rdonly, TRUE);
3967 if (error) {
3968 nfsm_reply(NFSX_UNSIGNED);
3969 if (v3)
3970 nfsm_srvpostop_attr(getret, &at);
3971 error = 0;
3972 goto nfsmout;
3973 }
3974 sf = &statfs;
3975 error = VFS_STATFS(vp->v_mount, sf, td);
3976 getret = VOP_GETATTR(vp, &at, cred, td);
3977 vput(vp);
3978 vp = NULL;
3979 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3980 if (v3)
3981 nfsm_srvpostop_attr(getret, &at);
3982 if (error) {
3983 error = 0;
3984 goto nfsmout;
3985 }
3986 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3987 if (v3) {
3988 tval = (u_quad_t)sf->f_blocks;
3989 tval *= (u_quad_t)sf->f_bsize;
3990 txdr_hyper(tval, &sfp->sf_tbytes);
3991 tval = (u_quad_t)sf->f_bfree;
3992 tval *= (u_quad_t)sf->f_bsize;
3993 txdr_hyper(tval, &sfp->sf_fbytes);
3994 /*
3995 * Don't send negative values for available space,
3996 * since this field is unsigned in the NFS protocol.
3997 * Otherwise, the client would see absurdly high
3998 * numbers for free space.
3999 */
4000 if (sf->f_bavail < 0)
4001 tval = 0;
4002 else
4003 tval = (u_quad_t)sf->f_bavail;
4004 tval *= (u_quad_t)sf->f_bsize;
4005 txdr_hyper(tval, &sfp->sf_abytes);
4006 sfp->sf_tfiles.nfsuquad[0] = 0;
4007 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
4008 sfp->sf_ffiles.nfsuquad[0] = 0;
4009 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4010 sfp->sf_afiles.nfsuquad[0] = 0;
4011 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4012 sfp->sf_invarsec = 0;
4013 } else {
4014 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
4015 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
4016 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
4017 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
4018 if (sf->f_bavail < 0)
4019 sfp->sf_bavail = 0;
4020 else
4021 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
4022 }
4023nfsmout:
4024 if (vp)
4025 vput(vp);
4026 VFS_UNLOCK_GIANT(vfslocked);
4027 return(error);
4028}
4029
4030/*
4031 * nfs fsinfo service
4032 */
4033int
4034nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4035 struct thread *td, struct mbuf **mrq)
4036{
4037 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4038 struct sockaddr *nam = nfsd->nd_nam;
4039 caddr_t dpos = nfsd->nd_dpos;
4040 struct ucred *cred = nfsd->nd_cr;
4041 struct nfsv3_fsinfo *sip;
4042 caddr_t bpos;
4043 int error = 0, rdonly, getret = 1, pref;
4044 struct mbuf *mb, *mreq;
4045 struct vnode *vp = NULL;
4046 struct vattr at;
4047 nfsfh_t nfh;
4048 fhandle_t *fhp;
4049 u_quad_t maxfsize;
4050 struct statfs sb;
4051 int v3 = (nfsd->nd_flag & ND_NFSV3);
4052 int vfslocked;
4053
4054 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4055 if (!v3)
4056 panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
4057 fhp = &nfh.fh_generic;
4058 vfslocked = 0;
4059 nfsm_srvmtofh(fhp);
4060 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4061 nam, &rdonly, TRUE);
4062 if (error) {
4063 nfsm_reply(NFSX_UNSIGNED);
4064 nfsm_srvpostop_attr(getret, &at);
4065 error = 0;
4066 goto nfsmout;
4067 }
4068
4069 /* XXX Try to make a guess on the max file size. */
4070 VFS_STATFS(vp->v_mount, &sb, td);
4071 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
4072
4073 getret = VOP_GETATTR(vp, &at, cred, td);
4074 vput(vp);
4075 vp = NULL;
4076 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
4077 nfsm_srvpostop_attr(getret, &at);
4078 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
4079
4080 /*
4081 * XXX
4082 * There should be filesystem VFS OP(s) to get this information.
4083 * For now, assume ufs.
4084 */
4085 if (slp->ns_so->so_type == SOCK_DGRAM)
4086 pref = NFS_MAXDGRAMDATA;
4087 else
4088 pref = NFS_MAXDATA;
4089 sip->fs_rtmax = txdr_unsigned(pref);
4090 sip->fs_rtpref = txdr_unsigned(pref);
4091 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
4092 sip->fs_wtmax = txdr_unsigned(pref);
4093 sip->fs_wtpref = txdr_unsigned(pref);
4094 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
4095 sip->fs_dtpref = txdr_unsigned(pref);
4096 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
4097 sip->fs_timedelta.nfsv3_sec = 0;
4098 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
4099 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
4100 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
4101 NFSV3FSINFO_CANSETTIME);
4102nfsmout:
4103 if (vp)
4104 vput(vp);
4105 VFS_UNLOCK_GIANT(vfslocked);
4106 return(error);
4107}
4108
4109/*
4110 * nfs pathconf service
4111 */
4112int
4113nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4114 struct thread *td, struct mbuf **mrq)
4115{
4116 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4117 struct sockaddr *nam = nfsd->nd_nam;
4118 caddr_t dpos = nfsd->nd_dpos;
4119 struct ucred *cred = nfsd->nd_cr;
4120 struct nfsv3_pathconf *pc;
4121 caddr_t bpos;
4122 int error = 0, rdonly, getret = 1;
4123 register_t linkmax, namemax, chownres, notrunc;
4124 struct mbuf *mb, *mreq;
4125 struct vnode *vp = NULL;
4126 struct vattr at;
4127 nfsfh_t nfh;
4128 fhandle_t *fhp;
4129 int v3 = (nfsd->nd_flag & ND_NFSV3);
4130 int vfslocked;
4131
4132 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4133 if (!v3)
4134 panic("nfsrv_pathconf: v3 proc called on a v2 connection");
4135 fhp = &nfh.fh_generic;
4136 nfsm_srvmtofh(fhp);
4137 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4138 nam, &rdonly, TRUE);
4139 if (error) {
4140 nfsm_reply(NFSX_UNSIGNED);
4141 nfsm_srvpostop_attr(getret, &at);
4142 error = 0;
4143 goto nfsmout;
4144 }
4145 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
4146 if (!error)
4147 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
4148 if (!error)
4149 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
4150 if (!error)
4151 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
4152 getret = VOP_GETATTR(vp, &at, cred, td);
4153 vput(vp);
4154 vp = NULL;
4155 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
4156 nfsm_srvpostop_attr(getret, &at);
4157 if (error) {
4158 error = 0;
4159 goto nfsmout;
4160 }
4161 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
4162
4163 pc->pc_linkmax = txdr_unsigned(linkmax);
4164 pc->pc_namemax = txdr_unsigned(namemax);
4165 pc->pc_notrunc = txdr_unsigned(notrunc);
4166 pc->pc_chownrestricted = txdr_unsigned(chownres);
4167
4168 /*
4169 * These should probably be supported by VOP_PATHCONF(), but
4170 * until msdosfs is exportable (why would you want to?), the
4171 * Unix defaults should be ok.
4172 */
4173 pc->pc_caseinsensitive = nfsrv_nfs_false;
4174 pc->pc_casepreserving = nfsrv_nfs_true;
4175nfsmout:
4176 if (vp)
4177 vput(vp);
4178 VFS_UNLOCK_GIANT(vfslocked);
4179 return(error);
4180}
4181
4182/*
4183 * Null operation, used by clients to ping server
4184 */
4185/* ARGSUSED */
4186int
4187nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4188 struct thread *td, struct mbuf **mrq)
4189{
4190 struct mbuf *mrep = nfsd->nd_mrep;
4191 caddr_t bpos;
4192 int error = NFSERR_RETVOID;
4193 struct mbuf *mb, *mreq;
4194
4195 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4196 nfsm_reply(0);
4197nfsmout:
4198 return (error);
4199}
4200
4201/*
4202 * No operation, used for obsolete procedures
4203 */
4204/* ARGSUSED */
4205int
4206nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4207 struct thread *td, struct mbuf **mrq)
4208{
4209 struct mbuf *mrep = nfsd->nd_mrep;
4210 caddr_t bpos;
4211 int error;
4212 struct mbuf *mb, *mreq;
4213
4214 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4215 if (nfsd->nd_repstat)
4216 error = nfsd->nd_repstat;
4217 else
4218 error = EPROCUNAVAIL;
4219 nfsm_reply(0);
4220 error = 0;
4221nfsmout:
4222 return (error);
4223}
4224
4225/*
4226 * Perform access checking for vnodes obtained from file handles that would
4227 * refer to files already opened by a Unix client. You cannot just use
4228 * vn_writechk() and VOP_ACCESS() for two reasons.
4229 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
4230 * case.
4231 * 2 - The owner is to be given access irrespective of mode bits for some
4232 * operations, so that processes that chmod after opening a file don't
4233 * break. I don't like this because it opens a security hole, but since
4234 * the nfs server opens a security hole the size of a barn door anyhow,
4235 * what the heck.
4236 *
4237 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
4238 * will return EPERM instead of EACCESS. EPERM is always an error.
4239 */
4240static int
4241nfsrv_access(struct vnode *vp, int flags, struct ucred *cred,
4242 int rdonly, struct thread *td, int override)
4243{
4244 struct vattr vattr;
4245 int error;
4246
4247 VFS_ASSERT_GIANT(vp->v_mount);
4248
4249 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4250
4251 if (flags & VWRITE) {
4252 /* Just vn_writechk() changed to check rdonly */
4253 /*
4254 * Disallow write attempts on read-only filesystems;
4255 * unless the file is a socket or a block or character
4256 * device resident on the filesystem.
4257 */
4258 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
4259 switch (vp->v_type) {
4260 case VREG:
4261 case VDIR:
4262 case VLNK:
4263 return (EROFS);
4264 default:
4265 break;
4266 }
4267 }
4268 /*
4269 * If there's shared text associated with
4270 * the inode, we can't allow writing.
4271 */
4272 if (vp->v_vflag & VV_TEXT)
4273 return (ETXTBSY);
4274 }
4275
4276 error = VOP_GETATTR(vp, &vattr, cred, td);
4277 if (error)
4278 return (error);
4279 error = VOP_ACCESS(vp, flags, cred, td);
4280 /*
4281 * Allow certain operations for the owner (reads and writes
4282 * on files that are already open).
4283 */
4284 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
4285 error = 0;
4286 return (error);
4287}
|