Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/nfsserver/nfs_serv.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 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 <nfsserver/nfs.h> 97 #include <nfs/xdr_subs.h> 98 #include <nfsserver/nfsm_subs.h> 99 100 #ifdef NFSRV_DEBUG 101 #define nfsdbprintf(info) printf info 102 #else 103 #define nfsdbprintf(info) 104 #endif 105 106 #define MAX_COMMIT_COUNT (1024 * 1024) 107 108 #define MAX_REORDERED_RPC 16 109 #define NUM_HEURISTIC 1031 110 #define NHUSE_INIT 64 111 #define NHUSE_INC 16 112 #define NHUSE_MAX 2048 113 114 static struct nfsheur { 115 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 116 off_t nh_nextoff; /* 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 */ 122 123 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000; 124 int nfsrvw_procrastinate_v3 = 0; 125 126 static struct timeval nfsver = { 0 }; 127 128 SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server"); 129 130 static int nfs_async; 131 static int nfs_commit_blks; 132 static int nfs_commit_miss; 133 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, 134 "Tell client that writes were synced even though they were not"); 135 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, 136 "Number of completed commits"); 137 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, ""); 138 139 struct nfsrvstats nfsrvstats; 140 SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RW, 141 &nfsrvstats, nfsrvstats, "S,nfsrvstats"); 142 143 static int nfsrv_access(struct vnode *, accmode_t, struct ucred *, 144 int, int); 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 */ 150 151 static __inline void 152 ndclear(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 */ 167 static __inline int 168 nfsrv_lockedpair(int vfs1, int vfs2) 169 { 170 171 if (vfs1 && vfs2) 172 VFS_UNLOCK_GIANT(vfs2); 173 174 return (vfs1 | vfs2); 175 } 176 177 static __inline int 178 nfsrv_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 * Heuristic to detect sequential operation. 189 */ 190 static struct nfsheur * 191 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 192 { 193 struct nfsheur *nh; 194 int hi, try; 195 196 /* Locate best candidate. */ 197 try = 32; 198 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 199 nh = &nfsheur[hi]; 200 while (try--) { 201 if (nfsheur[hi].nh_vp == vp) { 202 nh = &nfsheur[hi]; 203 break; 204 } 205 if (nfsheur[hi].nh_use > 0) 206 --nfsheur[hi].nh_use; 207 hi = (hi + 1) % NUM_HEURISTIC; 208 if (nfsheur[hi].nh_use < nh->nh_use) 209 nh = &nfsheur[hi]; 210 } 211 212 /* Initialize hint if this is a new file. */ 213 if (nh->nh_vp != vp) { 214 nh->nh_vp = vp; 215 nh->nh_nextoff = uio->uio_offset; 216 nh->nh_use = NHUSE_INIT; 217 if (uio->uio_offset == 0) 218 nh->nh_seqcount = 4; 219 else 220 nh->nh_seqcount = 1; 221 } 222 223 /* Calculate heuristic. */ 224 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 225 uio->uio_offset == nh->nh_nextoff) { 226 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 227 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 228 if (nh->nh_seqcount > IO_SEQMAX) 229 nh->nh_seqcount = IO_SEQMAX; 230 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 231 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 232 /* Probably a reordered RPC, leave seqcount alone. */ 233 } else if (nh->nh_seqcount > 1) { 234 nh->nh_seqcount /= 2; 235 } else { 236 nh->nh_seqcount = 0; 237 } 238 nh->nh_use += NHUSE_INC; 239 if (nh->nh_use > NHUSE_MAX) 240 nh->nh_use = NHUSE_MAX; 241 return (nh); 242 } 243 244 /* 245 * nfs v3 access service 246 */ 247 int 248 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 249 struct mbuf **mrq) 250 { 251 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 252 struct sockaddr *nam = nfsd->nd_nam; 253 caddr_t dpos = nfsd->nd_dpos; 254 struct ucred *cred = nfsd->nd_cr; 255 struct vnode *vp = NULL; 256 nfsfh_t nfh; 257 fhandle_t *fhp; 258 u_int32_t *tl; 259 caddr_t bpos; 260 int error = 0, rdonly, getret; 261 struct mbuf *mb, *mreq; 262 struct vattr vattr, *vap = &vattr; 263 u_long testmode, nfsmode; 264 int v3 = (nfsd->nd_flag & ND_NFSV3); 265 int vfslocked; 266 267 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 268 if (!v3) 269 panic("nfsrv3_access: v3 proc called on a v2 connection"); 270 vfslocked = 0; 271 fhp = &nfh.fh_generic; 272 nfsm_srvmtofh(fhp); 273 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED); 274 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 275 if (error) { 276 nfsm_reply(NFSX_UNSIGNED); 277 nfsm_srvpostop_attr(1, NULL); 278 error = 0; 279 goto nfsmout; 280 } 281 nfsmode = fxdr_unsigned(u_int32_t, *tl); 282 if ((nfsmode & NFSV3ACCESS_READ) && 283 nfsrv_access(vp, VREAD, cred, rdonly, 0)) 284 nfsmode &= ~NFSV3ACCESS_READ; 285 if (vp->v_type == VDIR) 286 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 287 NFSV3ACCESS_DELETE); 288 else 289 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 290 if ((nfsmode & testmode) && 291 nfsrv_access(vp, VWRITE, cred, rdonly, 0)) 292 nfsmode &= ~testmode; 293 if (vp->v_type == VDIR) 294 testmode = NFSV3ACCESS_LOOKUP; 295 else 296 testmode = NFSV3ACCESS_EXECUTE; 297 if ((nfsmode & testmode) && 298 nfsrv_access(vp, VEXEC, cred, rdonly, 0)) 299 nfsmode &= ~testmode; 300 getret = VOP_GETATTR(vp, vap, cred); 301 vput(vp); 302 vp = NULL; 303 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 304 nfsm_srvpostop_attr(getret, vap); 305 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 306 *tl = txdr_unsigned(nfsmode); 307 nfsmout: 308 if (vp) 309 vput(vp); 310 VFS_UNLOCK_GIANT(vfslocked); 311 return(error); 312 } 313 314 /* 315 * nfs getattr service 316 */ 317 int 318 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 319 struct mbuf **mrq) 320 { 321 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 322 struct sockaddr *nam = nfsd->nd_nam; 323 caddr_t dpos = nfsd->nd_dpos; 324 struct ucred *cred = nfsd->nd_cr; 325 struct nfs_fattr *fp; 326 struct vattr va; 327 struct vattr *vap = &va; 328 struct vnode *vp = NULL; 329 nfsfh_t nfh; 330 fhandle_t *fhp; 331 caddr_t bpos; 332 int error = 0, rdonly; 333 struct mbuf *mb, *mreq; 334 int vfslocked; 335 336 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 337 vfslocked = 0; 338 fhp = &nfh.fh_generic; 339 nfsm_srvmtofh(fhp); 340 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 341 if (error) { 342 nfsm_reply(0); 343 error = 0; 344 goto nfsmout; 345 } 346 error = VOP_GETATTR(vp, vap, cred); 347 vput(vp); 348 vp = NULL; 349 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 350 if (error) { 351 error = 0; 352 goto nfsmout; 353 } 354 fp = nfsm_build(struct nfs_fattr *, 355 NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 356 nfsm_srvfillattr(vap, fp); 357 /* fall through */ 358 359 nfsmout: 360 if (vp) 361 vput(vp); 362 VFS_UNLOCK_GIANT(vfslocked); 363 return(error); 364 } 365 366 /* 367 * nfs setattr service 368 */ 369 int 370 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 371 struct mbuf **mrq) 372 { 373 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 374 struct sockaddr *nam = nfsd->nd_nam; 375 caddr_t dpos = nfsd->nd_dpos; 376 struct ucred *cred = nfsd->nd_cr; 377 struct vattr va, preat; 378 struct vattr *vap = &va; 379 struct nfsv2_sattr *sp; 380 struct nfs_fattr *fp; 381 struct vnode *vp = NULL; 382 nfsfh_t nfh; 383 fhandle_t *fhp; 384 u_int32_t *tl; 385 caddr_t bpos; 386 int error = 0, rdonly, preat_ret = 1, postat_ret = 1; 387 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0; 388 struct mbuf *mb, *mreq; 389 struct timespec guard = { 0, 0 }; 390 struct mount *mp = NULL; 391 int tvfslocked; 392 int vfslocked; 393 394 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 395 vfslocked = 0; 396 fhp = &nfh.fh_generic; 397 nfsm_srvmtofh(fhp); 398 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 399 error = ESTALE; 400 goto out; 401 } 402 vfslocked = VFS_LOCK_GIANT(mp); 403 (void) vn_start_write(NULL, &mp, V_WAIT); 404 vfs_rel(mp); /* The write holds a ref. */ 405 VATTR_NULL(vap); 406 if (v3) { 407 nfsm_srvsattr(vap); 408 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED); 409 gcheck = fxdr_unsigned(int, *tl); 410 if (gcheck) { 411 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED); 412 fxdr_nfsv3time(tl, &guard); 413 } 414 } else { 415 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR); 416 /* 417 * Nah nah nah nah na nah 418 * There is a bug in the Sun client that puts 0xffff in the mode 419 * field of sattr when it should put in 0xffffffff. The u_short 420 * doesn't sign extend. 421 * --> check the low order 2 bytes for 0xffff 422 */ 423 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 424 vap->va_mode = nfstov_mode(sp->sa_mode); 425 if (sp->sa_uid != nfsrv_nfs_xdrneg1) 426 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 427 if (sp->sa_gid != nfsrv_nfs_xdrneg1) 428 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 429 if (sp->sa_size != nfsrv_nfs_xdrneg1) 430 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 431 if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) { 432 #ifdef notyet 433 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime); 434 #else 435 vap->va_atime.tv_sec = 436 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec); 437 vap->va_atime.tv_nsec = 0; 438 #endif 439 } 440 if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1) 441 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime); 442 443 } 444 445 /* 446 * Now that we have all the fields, lets do it. 447 */ 448 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly); 449 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked); 450 if (error) { 451 nfsm_reply(2 * NFSX_UNSIGNED); 452 if (v3) 453 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap); 454 error = 0; 455 goto nfsmout; 456 } 457 458 /* 459 * vp now an active resource, pay careful attention to cleanup 460 */ 461 if (v3) { 462 error = preat_ret = VOP_GETATTR(vp, &preat, cred); 463 if (!error && gcheck && 464 (preat.va_ctime.tv_sec != guard.tv_sec || 465 preat.va_ctime.tv_nsec != guard.tv_nsec)) 466 error = NFSERR_NOT_SYNC; 467 if (error) { 468 vput(vp); 469 vp = NULL; 470 nfsm_reply(NFSX_WCCDATA(v3)); 471 if (v3) 472 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap); 473 error = 0; 474 goto nfsmout; 475 } 476 } 477 478 /* 479 * If the size is being changed write acces is required, otherwise 480 * just check for a read only filesystem. 481 */ 482 if (vap->va_size == ((u_quad_t)((quad_t) -1))) { 483 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 484 error = EROFS; 485 goto out; 486 } 487 } else { 488 if (vp->v_type == VDIR) { 489 error = EISDIR; 490 goto out; 491 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 492 0)) != 0) 493 goto out; 494 } 495 error = VOP_SETATTR(vp, vap, cred); 496 postat_ret = VOP_GETATTR(vp, vap, cred); 497 if (!error) 498 error = postat_ret; 499 out: 500 if (vp != NULL) 501 vput(vp); 502 503 vp = NULL; 504 nfsm_reply(NFSX_WCCORFATTR(v3)); 505 if (v3) { 506 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap); 507 } else if (!error) { 508 /* v2 non-error case. */ 509 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR); 510 nfsm_srvfillattr(vap, fp); 511 } 512 error = 0; 513 /* fall through */ 514 515 nfsmout: 516 if (vp) 517 vput(vp); 518 vn_finished_write(mp); 519 VFS_UNLOCK_GIANT(vfslocked); 520 return(error); 521 } 522 523 /* 524 * nfs lookup rpc 525 */ 526 int 527 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 528 struct mbuf **mrq) 529 { 530 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 531 struct sockaddr *nam = nfsd->nd_nam; 532 caddr_t dpos = nfsd->nd_dpos; 533 struct ucred *cred = nfsd->nd_cr; 534 struct nfs_fattr *fp; 535 struct nameidata nd, ind, *ndp = &nd; 536 struct vnode *vp, *dirp = NULL; 537 nfsfh_t nfh; 538 fhandle_t *fhp; 539 caddr_t bpos; 540 int error = 0, len, dirattr_ret = 1; 541 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 542 struct mbuf *mb, *mreq; 543 struct vattr va, dirattr, *vap = &va; 544 int tvfslocked; 545 int vfslocked; 546 547 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 548 ndclear(&nd); 549 vfslocked = 0; 550 551 fhp = &nfh.fh_generic; 552 nfsm_srvmtofh(fhp); 553 nfsm_srvnamesiz(len); 554 555 pubflag = nfs_ispublicfh(fhp); 556 557 nd.ni_cnd.cn_cred = cred; 558 nd.ni_cnd.cn_nameiop = LOOKUP; 559 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART | MPSAFE; 560 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 561 &dirp, v3, &dirattr, &dirattr_ret, pubflag); 562 vfslocked = NDHASGIANT(&nd); 563 564 /* 565 * namei failure, only dirp to cleanup. Clear out garbarge from 566 * structure in case macros jump to nfsmout. 567 */ 568 569 if (error) { 570 if (dirp) { 571 vrele(dirp); 572 dirp = NULL; 573 } 574 nfsm_reply(NFSX_POSTOPATTR(v3)); 575 if (v3) 576 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 577 error = 0; 578 goto nfsmout; 579 } 580 581 /* 582 * Locate index file for public filehandle 583 * 584 * error is 0 on entry and 0 on exit from this block. 585 */ 586 587 if (pubflag) { 588 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) { 589 /* 590 * Setup call to lookup() to see if we can find 591 * the index file. Arguably, this doesn't belong 592 * in a kernel.. Ugh. If an error occurs, do not 593 * try to install an index file and then clear the 594 * error. 595 * 596 * When we replace nd with ind and redirect ndp, 597 * maintenance of ni_startdir and ni_vp shift to 598 * ind and we have to clean them up in the old nd. 599 * However, the cnd resource continues to be maintained 600 * via the original nd. Confused? You aren't alone! 601 */ 602 ind = nd; 603 VOP_UNLOCK(nd.ni_vp, 0); 604 ind.ni_pathlen = strlen(nfs_pub.np_index); 605 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf = 606 nfs_pub.np_index; 607 ind.ni_startdir = nd.ni_vp; 608 VREF(ind.ni_startdir); 609 ind.ni_cnd.cn_flags &= ~GIANTHELD; 610 tvfslocked = VFS_LOCK_GIANT(ind.ni_startdir->v_mount); 611 if (tvfslocked) 612 nd.ni_cnd.cn_flags |= GIANTHELD; 613 error = lookup(&ind); 614 ind.ni_dvp = NULL; 615 vfslocked = nfsrv_lockedpair_nd(vfslocked, &ind); 616 ind.ni_cnd.cn_flags &= ~GIANTHELD; 617 618 if (error == 0) { 619 /* 620 * Found an index file. Get rid of 621 * the old references. transfer nd.ni_vp' 622 */ 623 if (dirp) 624 vrele(dirp); 625 dirp = nd.ni_vp; 626 nd.ni_vp = NULL; 627 vrele(nd.ni_startdir); 628 nd.ni_startdir = NULL; 629 ndp = &ind; 630 } 631 error = 0; 632 } 633 /* 634 * If the public filehandle was used, check that this lookup 635 * didn't result in a filehandle outside the publicly exported 636 * filesystem. We clear the poor vp here to avoid lockups due 637 * to NFS I/O. 638 */ 639 640 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) { 641 vput(nd.ni_vp); 642 nd.ni_vp = NULL; 643 error = EPERM; 644 } 645 } 646 647 /* 648 * Resources at this point: 649 * ndp->ni_vp may not be NULL 650 */ 651 652 if (error) { 653 nfsm_reply(NFSX_POSTOPATTR(v3)); 654 if (v3) 655 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 656 error = 0; 657 goto nfsmout; 658 } 659 660 /* 661 * Get underlying attribute, then release remaining resources ( for 662 * the same potential blocking reason ) and reply. 663 */ 664 vp = ndp->ni_vp; 665 bzero((caddr_t)fhp, sizeof(nfh)); 666 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 667 error = VOP_VPTOFH(vp, &fhp->fh_fid); 668 if (!error) 669 error = VOP_GETATTR(vp, vap, cred); 670 671 vput(vp); 672 vrele(ndp->ni_startdir); 673 vrele(dirp); 674 ndp->ni_vp = NULL; 675 ndp->ni_startdir = NULL; 676 dirp = NULL; 677 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3)); 678 if (error) { 679 if (v3) 680 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 681 error = 0; 682 goto nfsmout; 683 } 684 nfsm_srvfhtom(fhp, v3); 685 if (v3) { 686 nfsm_srvpostop_attr(0, vap); 687 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 688 } else { 689 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR); 690 nfsm_srvfillattr(vap, fp); 691 } 692 693 nfsmout: 694 if (ndp->ni_vp || dirp || ndp->ni_startdir) { 695 if (ndp->ni_vp) 696 vput(ndp->ni_vp); 697 if (dirp) 698 vrele(dirp); 699 if (ndp->ni_startdir) 700 vrele(ndp->ni_startdir); 701 } 702 NDFREE(&nd, NDF_ONLY_PNBUF); 703 VFS_UNLOCK_GIANT(vfslocked); 704 return (error); 705 } 706 707 /* 708 * nfs readlink service 709 */ 710 int 711 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 712 struct mbuf **mrq) 713 { 714 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 715 struct sockaddr *nam = nfsd->nd_nam; 716 caddr_t dpos = nfsd->nd_dpos; 717 struct ucred *cred = nfsd->nd_cr; 718 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 719 struct iovec *ivp = iv; 720 struct mbuf *mp; 721 u_int32_t *tl; 722 caddr_t bpos; 723 int error = 0, rdonly, i, tlen, len, getret; 724 int v3 = (nfsd->nd_flag & ND_NFSV3); 725 struct mbuf *mb, *mp3, *nmp, *mreq; 726 struct vnode *vp = NULL; 727 struct vattr attr; 728 nfsfh_t nfh; 729 fhandle_t *fhp; 730 struct uio io, *uiop = &io; 731 int vfslocked; 732 733 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 734 vfslocked = 0; 735 #ifndef nolint 736 mp = NULL; 737 #endif 738 mp3 = NULL; 739 fhp = &nfh.fh_generic; 740 nfsm_srvmtofh(fhp); 741 len = 0; 742 i = 0; 743 while (len < NFS_MAXPATHLEN) { 744 MGET(nmp, M_WAIT, MT_DATA); 745 MCLGET(nmp, M_WAIT); 746 nmp->m_len = NFSMSIZ(nmp); 747 if (len == 0) 748 mp3 = mp = nmp; 749 else { 750 mp->m_next = nmp; 751 mp = nmp; 752 } 753 if ((len + mp->m_len) > NFS_MAXPATHLEN) { 754 mp->m_len = NFS_MAXPATHLEN - len; 755 len = NFS_MAXPATHLEN; 756 } else 757 len += mp->m_len; 758 ivp->iov_base = mtod(mp, caddr_t); 759 ivp->iov_len = mp->m_len; 760 i++; 761 ivp++; 762 } 763 uiop->uio_iov = iv; 764 uiop->uio_iovcnt = i; 765 uiop->uio_offset = 0; 766 uiop->uio_resid = len; 767 uiop->uio_rw = UIO_READ; 768 uiop->uio_segflg = UIO_SYSSPACE; 769 uiop->uio_td = NULL; 770 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 771 if (error) { 772 nfsm_reply(2 * NFSX_UNSIGNED); 773 if (v3) 774 nfsm_srvpostop_attr(1, NULL); 775 error = 0; 776 goto nfsmout; 777 } 778 if (vp->v_type != VLNK) { 779 if (v3) 780 error = EINVAL; 781 else 782 error = ENXIO; 783 } else 784 error = VOP_READLINK(vp, uiop, cred); 785 getret = VOP_GETATTR(vp, &attr, cred); 786 vput(vp); 787 vp = NULL; 788 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 789 if (v3) 790 nfsm_srvpostop_attr(getret, &attr); 791 if (error) { 792 error = 0; 793 goto nfsmout; 794 } 795 if (uiop->uio_resid > 0) { 796 len -= uiop->uio_resid; 797 tlen = nfsm_rndup(len); 798 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len); 799 } 800 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); 801 *tl = txdr_unsigned(len); 802 mb->m_next = mp3; 803 mp3 = NULL; 804 nfsmout: 805 if (mp3) 806 m_freem(mp3); 807 if (vp) 808 vput(vp); 809 VFS_UNLOCK_GIANT(vfslocked); 810 return(error); 811 } 812 813 /* 814 * nfs read service 815 */ 816 int 817 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 818 struct mbuf **mrq) 819 { 820 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 821 struct sockaddr *nam = nfsd->nd_nam; 822 caddr_t dpos = nfsd->nd_dpos; 823 struct ucred *cred = nfsd->nd_cr; 824 struct iovec *iv; 825 struct iovec *iv2; 826 struct mbuf *m; 827 struct nfs_fattr *fp; 828 u_int32_t *tl; 829 int i; 830 caddr_t bpos; 831 int error = 0, rdonly, cnt, len, left, siz, tlen, getret; 832 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen; 833 struct mbuf *mb, *mreq; 834 struct mbuf *m2; 835 struct vnode *vp = NULL; 836 nfsfh_t nfh; 837 fhandle_t *fhp; 838 struct uio io, *uiop = &io; 839 struct vattr va, *vap = &va; 840 struct nfsheur *nh; 841 off_t off; 842 int ioflag = 0; 843 int vfslocked; 844 845 846 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 847 vfslocked = 0; 848 fhp = &nfh.fh_generic; 849 nfsm_srvmtofh(fhp); 850 if (v3) { 851 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED); 852 off = fxdr_hyper(tl); 853 } else { 854 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED); 855 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 856 } 857 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd)); 858 859 /* 860 * Reference vp. If an error occurs, vp will be invalid, but we 861 * have to NULL it just in case. The macros might goto nfsmout 862 * as well. 863 */ 864 865 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 866 if (error) { 867 vp = NULL; 868 nfsm_reply(2 * NFSX_UNSIGNED); 869 if (v3) 870 nfsm_srvpostop_attr(1, NULL); 871 error = 0; 872 goto nfsmout; 873 } 874 875 if (vp->v_type != VREG) { 876 if (v3) 877 error = EINVAL; 878 else 879 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 880 } 881 if (!error) { 882 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, 1)) != 0) 883 error = nfsrv_access(vp, VEXEC, cred, rdonly, 1); 884 } 885 getret = VOP_GETATTR(vp, vap, cred); 886 if (!error) 887 error = getret; 888 if (error) { 889 vput(vp); 890 vp = NULL; 891 nfsm_reply(NFSX_POSTOPATTR(v3)); 892 if (v3) 893 nfsm_srvpostop_attr(getret, vap); 894 error = 0; 895 goto nfsmout; 896 } 897 898 /* 899 * Calculate byte count to read 900 */ 901 if (off >= vap->va_size) 902 cnt = 0; 903 else if ((off + reqlen) > vap->va_size) 904 cnt = vap->va_size - off; 905 else 906 cnt = reqlen; 907 908 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 909 if (v3) { 910 tl = nfsm_build(u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 911 *tl++ = nfsrv_nfs_true; 912 fp = (struct nfs_fattr *)tl; 913 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 914 } else { 915 tl = nfsm_build(u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 916 fp = (struct nfs_fattr *)tl; 917 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 918 } 919 len = left = nfsm_rndup(cnt); 920 if (cnt > 0) { 921 /* 922 * Generate the mbuf list with the uio_iov ref. to it. 923 */ 924 i = 0; 925 m = m2 = mb; 926 while (left > 0) { 927 siz = min(M_TRAILINGSPACE(m), left); 928 if (siz > 0) { 929 left -= siz; 930 i++; 931 } 932 if (left > 0) { 933 MGET(m, M_WAIT, MT_DATA); 934 MCLGET(m, M_WAIT); 935 m->m_len = 0; 936 m2->m_next = m; 937 m2 = m; 938 } 939 } 940 iv = malloc(i * sizeof (struct iovec), 941 M_TEMP, M_WAITOK); 942 uiop->uio_iov = iv2 = iv; 943 m = mb; 944 left = len; 945 i = 0; 946 while (left > 0) { 947 if (m == NULL) 948 panic("nfsrv_read iov"); 949 siz = min(M_TRAILINGSPACE(m), left); 950 if (siz > 0) { 951 iv->iov_base = mtod(m, caddr_t) + m->m_len; 952 iv->iov_len = siz; 953 m->m_len += siz; 954 left -= siz; 955 iv++; 956 i++; 957 } 958 m = m->m_next; 959 } 960 uiop->uio_iovcnt = i; 961 uiop->uio_offset = off; 962 uiop->uio_resid = len; 963 uiop->uio_rw = UIO_READ; 964 uiop->uio_segflg = UIO_SYSSPACE; 965 nh = nfsrv_sequential_heuristic(uiop, vp); 966 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 967 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 968 if (error == 0) 969 nh->nh_nextoff = uiop->uio_offset; 970 free((caddr_t)iv2, M_TEMP); 971 if (error || (getret = VOP_GETATTR(vp, vap, cred))) { 972 if (!error) 973 error = getret; 974 m_freem(mreq); 975 vput(vp); 976 vp = NULL; 977 nfsm_reply(NFSX_POSTOPATTR(v3)); 978 if (v3) 979 nfsm_srvpostop_attr(getret, vap); 980 error = 0; 981 goto nfsmout; 982 } 983 } else 984 uiop->uio_resid = 0; 985 vput(vp); 986 vp = NULL; 987 nfsm_srvfillattr(vap, fp); 988 tlen = len - uiop->uio_resid; 989 cnt = cnt < tlen ? cnt : tlen; 990 tlen = nfsm_rndup(cnt); 991 if (len != tlen || tlen != cnt) 992 nfsm_adj(mb, len - tlen, tlen - cnt); 993 if (v3) { 994 *tl++ = txdr_unsigned(cnt); 995 if (cnt < reqlen) 996 *tl++ = nfsrv_nfs_true; 997 else 998 *tl++ = nfsrv_nfs_false; 999 } 1000 *tl = txdr_unsigned(cnt); 1001 nfsmout: 1002 if (vp) 1003 vput(vp); 1004 VFS_UNLOCK_GIANT(vfslocked); 1005 return(error); 1006 } 1007 1008 /* 1009 * nfs write service 1010 */ 1011 int 1012 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 1013 struct mbuf **mrq) 1014 { 1015 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1016 struct sockaddr *nam = nfsd->nd_nam; 1017 caddr_t dpos = nfsd->nd_dpos; 1018 struct ucred *cred = nfsd->nd_cr; 1019 struct iovec *ivp; 1020 int i, cnt; 1021 struct mbuf *mp; 1022 struct nfs_fattr *fp; 1023 struct iovec *iv; 1024 struct vattr va, forat; 1025 struct vattr *vap = &va; 1026 u_int32_t *tl; 1027 caddr_t bpos; 1028 int error = 0, rdonly, len, forat_ret = 1; 1029 int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust; 1030 int stable = NFSV3WRITE_FILESYNC; 1031 int v3 = (nfsd->nd_flag & ND_NFSV3); 1032 struct mbuf *mb, *mreq; 1033 struct vnode *vp = NULL; 1034 struct nfsheur *nh; 1035 nfsfh_t nfh; 1036 fhandle_t *fhp; 1037 struct uio io, *uiop = &io; 1038 off_t off; 1039 struct mount *mntp = NULL; 1040 int tvfslocked; 1041 int vfslocked; 1042 1043 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 1044 vfslocked = 0; 1045 if (mrep == NULL) { 1046 *mrq = NULL; 1047 error = 0; 1048 goto nfsmout; 1049 } 1050 fhp = &nfh.fh_generic; 1051 nfsm_srvmtofh(fhp); 1052 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 1053 error = ESTALE; 1054 goto ereply; 1055 } 1056 vfslocked = VFS_LOCK_GIANT(mntp); 1057 (void) vn_start_write(NULL, &mntp, V_WAIT); 1058 vfs_rel(mntp); /* The write holds a ref. */ 1059 if (v3) { 1060 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED); 1061 off = fxdr_hyper(tl); 1062 tl += 3; 1063 stable = fxdr_unsigned(int, *tl++); 1064 } else { 1065 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED); 1066 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1067 tl += 2; 1068 if (nfs_async) 1069 stable = NFSV3WRITE_UNSTABLE; 1070 } 1071 retlen = len = fxdr_unsigned(int32_t, *tl); 1072 cnt = i = 0; 1073 1074 /* 1075 * For NFS Version 2, it is not obvious what a write of zero length 1076 * should do, but I might as well be consistent with Version 3, 1077 * which is to return ok so long as there are no permission problems. 1078 */ 1079 if (len > 0) { 1080 zeroing = 1; 1081 mp = mrep; 1082 while (mp) { 1083 if (mp == md) { 1084 zeroing = 0; 1085 adjust = dpos - mtod(mp, caddr_t); 1086 mp->m_len -= adjust; 1087 if (mp->m_len > 0 && adjust > 0) 1088 mp->m_data += adjust; 1089 } 1090 if (zeroing) 1091 mp->m_len = 0; 1092 else if (mp->m_len > 0) { 1093 i += mp->m_len; 1094 if (i > len) { 1095 mp->m_len -= (i - len); 1096 zeroing = 1; 1097 } 1098 if (mp->m_len > 0) 1099 cnt++; 1100 } 1101 mp = mp->m_next; 1102 } 1103 } 1104 if (len > NFS_MAXDATA || len < 0 || i < len) { 1105 error = EIO; 1106 nfsm_reply(2 * NFSX_UNSIGNED); 1107 if (v3) 1108 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap); 1109 error = 0; 1110 goto nfsmout; 1111 } 1112 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly); 1113 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked); 1114 if (error) { 1115 vp = NULL; 1116 nfsm_reply(2 * NFSX_UNSIGNED); 1117 if (v3) 1118 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap); 1119 error = 0; 1120 goto nfsmout; 1121 } 1122 if (v3) 1123 forat_ret = VOP_GETATTR(vp, &forat, cred); 1124 if (vp->v_type != VREG) { 1125 if (v3) 1126 error = EINVAL; 1127 else 1128 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1129 } 1130 if (!error) 1131 error = nfsrv_access(vp, VWRITE, cred, rdonly, 1); 1132 if (error) { 1133 vput(vp); 1134 vp = NULL; 1135 nfsm_reply(NFSX_WCCDATA(v3)); 1136 if (v3) 1137 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap); 1138 error = 0; 1139 goto nfsmout; 1140 } 1141 1142 if (len > 0) { 1143 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1144 M_WAITOK); 1145 uiop->uio_iov = iv = ivp; 1146 uiop->uio_iovcnt = cnt; 1147 mp = mrep; 1148 while (mp) { 1149 if (mp->m_len > 0) { 1150 ivp->iov_base = mtod(mp, caddr_t); 1151 ivp->iov_len = mp->m_len; 1152 ivp++; 1153 } 1154 mp = mp->m_next; 1155 } 1156 1157 /* 1158 * XXX 1159 * The IO_METASYNC flag indicates that all metadata (and not just 1160 * enough to ensure data integrity) mus be written to stable storage 1161 * synchronously. 1162 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 1163 */ 1164 if (stable == NFSV3WRITE_UNSTABLE) 1165 ioflags = IO_NODELOCKED; 1166 else if (stable == NFSV3WRITE_DATASYNC) 1167 ioflags = (IO_SYNC | IO_NODELOCKED); 1168 else 1169 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1170 uiop->uio_resid = len; 1171 uiop->uio_rw = UIO_WRITE; 1172 uiop->uio_segflg = UIO_SYSSPACE; 1173 uiop->uio_td = NULL; 1174 uiop->uio_offset = off; 1175 nh = nfsrv_sequential_heuristic(uiop, vp); 1176 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1177 error = VOP_WRITE(vp, uiop, ioflags, cred); 1178 if (error == 0) 1179 nh->nh_nextoff = uiop->uio_offset; 1180 /* Unlocked write. */ 1181 nfsrvstats.srvvop_writes++; 1182 free((caddr_t)iv, M_TEMP); 1183 } 1184 aftat_ret = VOP_GETATTR(vp, vap, cred); 1185 vput(vp); 1186 vp = NULL; 1187 if (!error) 1188 error = aftat_ret; 1189 ereply: 1190 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 1191 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 1192 if (v3) { 1193 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap); 1194 if (error) { 1195 error = 0; 1196 goto nfsmout; 1197 } 1198 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); 1199 *tl++ = txdr_unsigned(retlen); 1200 /* 1201 * If nfs_async is set, then pretend the write was FILESYNC. 1202 */ 1203 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async) 1204 *tl++ = txdr_unsigned(stable); 1205 else 1206 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 1207 /* 1208 * Actually, there is no need to txdr these fields, 1209 * but it may make the values more human readable, 1210 * for debugging purposes. 1211 */ 1212 if (nfsver.tv_sec == 0) 1213 nfsver = boottime; 1214 *tl++ = txdr_unsigned(nfsver.tv_sec); 1215 *tl = txdr_unsigned(nfsver.tv_usec); 1216 } else if (!error) { 1217 /* v2 non-error case. */ 1218 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR); 1219 nfsm_srvfillattr(vap, fp); 1220 } 1221 error = 0; 1222 nfsmout: 1223 if (vp) 1224 vput(vp); 1225 vn_finished_write(mntp); 1226 VFS_UNLOCK_GIANT(vfslocked); 1227 return(error); 1228 } 1229 1230 /* 1231 * nfs create service 1232 * now does a truncate to 0 length via. setattr if it already exists 1233 */ 1234 int 1235 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 1236 struct mbuf **mrq) 1237 { 1238 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1239 struct sockaddr *nam = nfsd->nd_nam; 1240 caddr_t dpos = nfsd->nd_dpos; 1241 struct ucred *cred = nfsd->nd_cr; 1242 struct nfs_fattr *fp; 1243 struct vattr va, dirfor, diraft; 1244 struct vattr *vap = &va; 1245 struct nfsv2_sattr *sp; 1246 u_int32_t *tl; 1247 struct nameidata nd; 1248 caddr_t bpos; 1249 int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1250 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1251 struct mbuf *mb, *mreq; 1252 struct vnode *dirp = NULL; 1253 nfsfh_t nfh; 1254 fhandle_t *fhp; 1255 u_quad_t tempsize; 1256 struct timespec cverf; 1257 struct mount *mp = NULL; 1258 int tvfslocked; 1259 int vfslocked; 1260 1261 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 1262 vfslocked = 0; 1263 #ifndef nolint 1264 rdev = 0; 1265 #endif 1266 ndclear(&nd); 1267 1268 fhp = &nfh.fh_generic; 1269 nfsm_srvmtofh(fhp); 1270 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 1271 error = ESTALE; 1272 goto ereply; 1273 } 1274 vfslocked = VFS_LOCK_GIANT(mp); 1275 (void) vn_start_write(NULL, &mp, V_WAIT); 1276 vfs_rel(mp); /* The write holds a ref. */ 1277 nfsm_srvnamesiz(len); 1278 1279 nd.ni_cnd.cn_cred = cred; 1280 nd.ni_cnd.cn_nameiop = CREATE; 1281 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE; 1282 1283 /* 1284 * Call namei and do initial cleanup to get a few things 1285 * out of the way. If we get an initial error we cleanup 1286 * and return here to avoid special-casing the invalid nd 1287 * structure through the rest of the case. dirp may be 1288 * set even if an error occurs, but the nd structure will not 1289 * be valid at all if an error occurs so we have to invalidate it 1290 * prior to calling nfsm_reply ( which might goto nfsmout ). 1291 */ 1292 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 1293 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 1294 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 1295 if (dirp && !v3) { 1296 vrele(dirp); 1297 dirp = NULL; 1298 } 1299 if (error) { 1300 nfsm_reply(NFSX_WCCDATA(v3)); 1301 if (v3) 1302 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1303 error = 0; 1304 goto nfsmout; 1305 } 1306 1307 /* 1308 * No error. Continue. State: 1309 * 1310 * startdir is valid ( we release this immediately ) 1311 * dirp may be valid 1312 * nd.ni_vp may be valid 1313 * nd.ni_dvp is valid 1314 * 1315 * The error state is set through the code and we may also do some 1316 * opportunistic releasing of vnodes to avoid holding locks through 1317 * NFS I/O. The cleanup at the end is a catch-all 1318 */ 1319 1320 VATTR_NULL(vap); 1321 if (v3) { 1322 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED); 1323 how = fxdr_unsigned(int, *tl); 1324 switch (how) { 1325 case NFSV3CREATE_GUARDED: 1326 if (nd.ni_vp) { 1327 error = EEXIST; 1328 break; 1329 } 1330 /* fall through */ 1331 case NFSV3CREATE_UNCHECKED: 1332 nfsm_srvsattr(vap); 1333 break; 1334 case NFSV3CREATE_EXCLUSIVE: 1335 tl = nfsm_dissect_nonblock(u_int32_t *, 1336 NFSX_V3CREATEVERF); 1337 /* Unique bytes, endianness is not important. */ 1338 cverf.tv_sec = (int32_t)tl[0]; 1339 cverf.tv_nsec = tl[1]; 1340 exclusive_flag = 1; 1341 break; 1342 }; 1343 vap->va_type = VREG; 1344 } else { 1345 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR); 1346 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1347 if (vap->va_type == VNON) 1348 vap->va_type = VREG; 1349 vap->va_mode = nfstov_mode(sp->sa_mode); 1350 switch (vap->va_type) { 1351 case VREG: 1352 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1353 if (tsize != -1) 1354 vap->va_size = (u_quad_t)tsize; 1355 break; 1356 case VCHR: 1357 case VBLK: 1358 case VFIFO: 1359 rdev = fxdr_unsigned(long, sp->sa_size); 1360 break; 1361 default: 1362 break; 1363 }; 1364 } 1365 1366 /* 1367 * Iff doesn't exist, create it 1368 * otherwise just truncate to 0 length 1369 * should I set the mode too ? 1370 * 1371 * The only possible error we can have at this point is EEXIST. 1372 * nd.ni_vp will also be non-NULL in that case. 1373 */ 1374 if (nd.ni_vp == NULL) { 1375 if (vap->va_mode == (mode_t)VNOVAL) 1376 vap->va_mode = 0; 1377 if (vap->va_type == VREG || vap->va_type == VSOCK) { 1378 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap); 1379 if (error) 1380 NDFREE(&nd, NDF_ONLY_PNBUF); 1381 else { 1382 if (exclusive_flag) { 1383 exclusive_flag = 0; 1384 VATTR_NULL(vap); 1385 vap->va_atime = cverf; 1386 error = VOP_SETATTR(nd.ni_vp, vap, 1387 cred); 1388 } 1389 } 1390 } else if (vap->va_type == VCHR || vap->va_type == VBLK || 1391 vap->va_type == VFIFO) { 1392 /* 1393 * NFSv2-specific code for creating device nodes 1394 * and fifos. 1395 * 1396 * Handle SysV FIFO node special cases. All other 1397 * devices require super user to access. 1398 */ 1399 if (vap->va_type == VCHR && rdev == 0xffffffff) 1400 vap->va_type = VFIFO; 1401 if (vap->va_type != VFIFO && 1402 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 1403 0))) { 1404 goto ereply; 1405 } 1406 vap->va_rdev = rdev; 1407 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap); 1408 if (error) { 1409 NDFREE(&nd, NDF_ONLY_PNBUF); 1410 goto ereply; 1411 } 1412 vput(nd.ni_vp); 1413 nd.ni_vp = NULL; 1414 1415 /* 1416 * release dvp prior to lookup 1417 */ 1418 vput(nd.ni_dvp); 1419 nd.ni_dvp = NULL; 1420 /* 1421 * Setup for lookup. 1422 * 1423 * Even though LOCKPARENT was cleared, ni_dvp may 1424 * be garbage. 1425 */ 1426 nd.ni_cnd.cn_nameiop = LOOKUP; 1427 nd.ni_cnd.cn_flags &= ~(LOCKPARENT); 1428 nd.ni_cnd.cn_thread = curthread; 1429 nd.ni_cnd.cn_cred = cred; 1430 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount); 1431 if (tvfslocked) 1432 nd.ni_cnd.cn_flags |= GIANTHELD; 1433 error = lookup(&nd); 1434 nd.ni_dvp = NULL; 1435 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 1436 nd.ni_cnd.cn_flags &= ~GIANTHELD; 1437 if (error) 1438 goto ereply; 1439 1440 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1441 error = EINVAL; 1442 goto ereply; 1443 } 1444 } else { 1445 error = ENXIO; 1446 } 1447 } else { 1448 if (vap->va_size != -1) { 1449 error = nfsrv_access(nd.ni_vp, VWRITE, 1450 cred, (nd.ni_cnd.cn_flags & RDONLY), 0); 1451 if (!error) { 1452 tempsize = vap->va_size; 1453 VATTR_NULL(vap); 1454 vap->va_size = tempsize; 1455 error = VOP_SETATTR(nd.ni_vp, vap, cred); 1456 } 1457 } 1458 } 1459 1460 if (!error) { 1461 bzero((caddr_t)fhp, sizeof(nfh)); 1462 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid; 1463 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid); 1464 if (!error) 1465 error = VOP_GETATTR(nd.ni_vp, vap, cred); 1466 } 1467 if (v3) { 1468 if (exclusive_flag && !error && 1469 bcmp(&cverf, &vap->va_atime, sizeof (cverf))) 1470 error = EEXIST; 1471 if (dirp == nd.ni_dvp) 1472 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1473 else { 1474 /* Drop the other locks to avoid deadlock. */ 1475 if (nd.ni_dvp) { 1476 if (nd.ni_dvp == nd.ni_vp) 1477 vrele(nd.ni_dvp); 1478 else 1479 vput(nd.ni_dvp); 1480 } 1481 if (nd.ni_vp) 1482 vput(nd.ni_vp); 1483 nd.ni_dvp = NULL; 1484 nd.ni_vp = NULL; 1485 1486 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 1487 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1488 VOP_UNLOCK(dirp, 0); 1489 } 1490 } 1491 ereply: 1492 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1493 if (v3) { 1494 if (!error) { 1495 nfsm_srvpostop_fh(fhp); 1496 nfsm_srvpostop_attr(0, vap); 1497 } 1498 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1499 } else if (!error) { 1500 /* v2 non-error case. */ 1501 nfsm_srvfhtom(fhp, v3); 1502 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR); 1503 nfsm_srvfillattr(vap, fp); 1504 } 1505 error = 0; 1506 1507 nfsmout: 1508 if (nd.ni_dvp) { 1509 if (nd.ni_dvp == nd.ni_vp) 1510 vrele(nd.ni_dvp); 1511 else 1512 vput(nd.ni_dvp); 1513 } 1514 if (nd.ni_vp) 1515 vput(nd.ni_vp); 1516 if (nd.ni_startdir) { 1517 vrele(nd.ni_startdir); 1518 nd.ni_startdir = NULL; 1519 } 1520 if (dirp) 1521 vrele(dirp); 1522 NDFREE(&nd, NDF_ONLY_PNBUF); 1523 vn_finished_write(mp); 1524 VFS_UNLOCK_GIANT(vfslocked); 1525 return (error); 1526 } 1527 1528 /* 1529 * nfs v3 mknod service 1530 */ 1531 int 1532 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 1533 struct mbuf **mrq) 1534 { 1535 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1536 struct sockaddr *nam = nfsd->nd_nam; 1537 caddr_t dpos = nfsd->nd_dpos; 1538 struct ucred *cred = nfsd->nd_cr; 1539 struct vattr va, dirfor, diraft; 1540 struct vattr *vap = &va; 1541 struct thread *td = curthread; 1542 u_int32_t *tl; 1543 struct nameidata nd; 1544 caddr_t bpos; 1545 int error = 0, len, dirfor_ret = 1, diraft_ret = 1; 1546 u_int32_t major, minor; 1547 enum vtype vtyp; 1548 struct mbuf *mb, *mreq; 1549 struct vnode *vp, *dirp = NULL; 1550 nfsfh_t nfh; 1551 fhandle_t *fhp; 1552 struct mount *mp = NULL; 1553 int v3 = (nfsd->nd_flag & ND_NFSV3); 1554 int tvfslocked; 1555 int vfslocked; 1556 1557 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 1558 vfslocked = 0; 1559 if (!v3) 1560 panic("nfsrv_mknod: v3 proc called on a v2 connection"); 1561 ndclear(&nd); 1562 1563 fhp = &nfh.fh_generic; 1564 nfsm_srvmtofh(fhp); 1565 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 1566 error = ESTALE; 1567 goto ereply; 1568 } 1569 vfslocked = VFS_LOCK_GIANT(mp); 1570 (void) vn_start_write(NULL, &mp, V_WAIT); 1571 vfs_rel(mp); /* The write holds a ref. */ 1572 nfsm_srvnamesiz(len); 1573 1574 nd.ni_cnd.cn_cred = cred; 1575 nd.ni_cnd.cn_nameiop = CREATE; 1576 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE; 1577 1578 /* 1579 * Handle nfs_namei() call. If an error occurs, the nd structure 1580 * is not valid. However, nfsm_*() routines may still jump to 1581 * nfsmout. 1582 */ 1583 1584 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 1585 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 1586 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 1587 if (error) { 1588 nfsm_reply(NFSX_WCCDATA(1)); 1589 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1590 error = 0; 1591 goto nfsmout; 1592 } 1593 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED); 1594 vtyp = nfsv3tov_type(*tl); 1595 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1596 error = NFSERR_BADTYPE; 1597 goto out; 1598 } 1599 VATTR_NULL(vap); 1600 nfsm_srvsattr(vap); 1601 if (vtyp == VCHR || vtyp == VBLK) { 1602 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED); 1603 major = fxdr_unsigned(u_int32_t, *tl++); 1604 minor = fxdr_unsigned(u_int32_t, *tl); 1605 vap->va_rdev = makedev(major, minor); 1606 } 1607 1608 /* 1609 * Iff doesn't exist, create it. 1610 */ 1611 if (nd.ni_vp) { 1612 error = EEXIST; 1613 goto out; 1614 } 1615 vap->va_type = vtyp; 1616 if (vap->va_mode == (mode_t)VNOVAL) 1617 vap->va_mode = 0; 1618 if (vtyp == VSOCK) { 1619 vrele(nd.ni_startdir); 1620 nd.ni_startdir = NULL; 1621 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap); 1622 if (error) 1623 NDFREE(&nd, NDF_ONLY_PNBUF); 1624 } else { 1625 if (vtyp != VFIFO && (error = priv_check_cred(cred, 1626 PRIV_VFS_MKNOD_DEV, 0))) 1627 goto out; 1628 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap); 1629 if (error) { 1630 NDFREE(&nd, NDF_ONLY_PNBUF); 1631 goto out; 1632 } 1633 vput(nd.ni_vp); 1634 nd.ni_vp = NULL; 1635 1636 /* 1637 * Release dvp prior to lookup 1638 */ 1639 vput(nd.ni_dvp); 1640 nd.ni_dvp = NULL; 1641 1642 nd.ni_cnd.cn_nameiop = LOOKUP; 1643 nd.ni_cnd.cn_flags &= ~(LOCKPARENT); 1644 nd.ni_cnd.cn_thread = td; 1645 nd.ni_cnd.cn_cred = td->td_ucred; 1646 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount); 1647 if (tvfslocked) 1648 nd.ni_cnd.cn_flags |= GIANTHELD; 1649 error = lookup(&nd); 1650 nd.ni_dvp = NULL; 1651 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 1652 nd.ni_cnd.cn_flags &= ~GIANTHELD; 1653 1654 if (error) 1655 goto out; 1656 if (nd.ni_cnd.cn_flags & ISSYMLINK) 1657 error = EINVAL; 1658 } 1659 1660 /* 1661 * send response, cleanup, return. 1662 */ 1663 out: 1664 vp = nd.ni_vp; 1665 if (!error) { 1666 bzero((caddr_t)fhp, sizeof(nfh)); 1667 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 1668 error = VOP_VPTOFH(vp, &fhp->fh_fid); 1669 if (!error) 1670 error = VOP_GETATTR(vp, vap, cred); 1671 } 1672 if (nd.ni_dvp) { 1673 if (nd.ni_dvp == nd.ni_vp) 1674 vrele(nd.ni_dvp); 1675 else 1676 vput(nd.ni_dvp); 1677 nd.ni_dvp = NULL; 1678 } 1679 if (vp) { 1680 vput(vp); 1681 vp = NULL; 1682 nd.ni_vp = NULL; 1683 } 1684 if (nd.ni_startdir) { 1685 vrele(nd.ni_startdir); 1686 nd.ni_startdir = NULL; 1687 } 1688 NDFREE(&nd, NDF_ONLY_PNBUF); 1689 if (dirp) { 1690 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 1691 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1692 vput(dirp); 1693 } 1694 ereply: 1695 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1)); 1696 if (v3) { 1697 if (!error) { 1698 nfsm_srvpostop_fh(fhp); 1699 nfsm_srvpostop_attr(0, vap); 1700 } 1701 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1702 } 1703 vn_finished_write(mp); 1704 VFS_UNLOCK_GIANT(vfslocked); 1705 return (0); 1706 nfsmout: 1707 if (nd.ni_dvp) { 1708 if (nd.ni_dvp == nd.ni_vp) 1709 vrele(nd.ni_dvp); 1710 else 1711 vput(nd.ni_dvp); 1712 } 1713 if (nd.ni_vp) 1714 vput(nd.ni_vp); 1715 if (dirp) 1716 vrele(dirp); 1717 if (nd.ni_startdir) 1718 vrele(nd.ni_startdir); 1719 NDFREE(&nd, NDF_ONLY_PNBUF); 1720 vn_finished_write(mp); 1721 VFS_UNLOCK_GIANT(vfslocked); 1722 return (error); 1723 } 1724 1725 /* 1726 * nfs remove service 1727 */ 1728 int 1729 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 1730 struct mbuf **mrq) 1731 { 1732 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1733 struct sockaddr *nam = nfsd->nd_nam; 1734 caddr_t dpos = nfsd->nd_dpos; 1735 struct ucred *cred = nfsd->nd_cr; 1736 struct nameidata nd; 1737 caddr_t bpos; 1738 int error = 0, len, dirfor_ret = 1, diraft_ret = 1; 1739 int v3 = (nfsd->nd_flag & ND_NFSV3); 1740 struct mbuf *mb, *mreq; 1741 struct vnode *dirp; 1742 struct vattr dirfor, diraft; 1743 nfsfh_t nfh; 1744 fhandle_t *fhp; 1745 struct mount *mp = NULL; 1746 int vfslocked; 1747 1748 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 1749 ndclear(&nd); 1750 vfslocked = 0; 1751 1752 fhp = &nfh.fh_generic; 1753 nfsm_srvmtofh(fhp); 1754 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 1755 error = ESTALE; 1756 goto ereply; 1757 } 1758 vfslocked = VFS_LOCK_GIANT(mp); 1759 (void) vn_start_write(NULL, &mp, V_WAIT); 1760 vfs_rel(mp); /* The write holds a ref. */ 1761 nfsm_srvnamesiz(len); 1762 1763 nd.ni_cnd.cn_cred = cred; 1764 nd.ni_cnd.cn_nameiop = DELETE; 1765 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE; 1766 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 1767 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 1768 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 1769 if (dirp && !v3) { 1770 vrele(dirp); 1771 dirp = NULL; 1772 } 1773 if (error == 0) { 1774 if (nd.ni_vp->v_type == VDIR) { 1775 error = EPERM; /* POSIX */ 1776 goto out; 1777 } 1778 /* 1779 * The root of a mounted filesystem cannot be deleted. 1780 */ 1781 if (nd.ni_vp->v_vflag & VV_ROOT) { 1782 error = EBUSY; 1783 goto out; 1784 } 1785 out: 1786 if (!error) { 1787 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1788 NDFREE(&nd, NDF_ONLY_PNBUF); 1789 } 1790 } 1791 if (dirp && v3) { 1792 if (dirp == nd.ni_dvp) 1793 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1794 else { 1795 /* Drop the other locks to avoid deadlock. */ 1796 if (nd.ni_dvp) { 1797 if (nd.ni_dvp == nd.ni_vp) 1798 vrele(nd.ni_dvp); 1799 else 1800 vput(nd.ni_dvp); 1801 } 1802 if (nd.ni_vp) 1803 vput(nd.ni_vp); 1804 nd.ni_dvp = NULL; 1805 nd.ni_vp = NULL; 1806 1807 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 1808 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1809 VOP_UNLOCK(dirp, 0); 1810 } 1811 vrele(dirp); 1812 dirp = NULL; 1813 } 1814 ereply: 1815 nfsm_reply(NFSX_WCCDATA(v3)); 1816 if (v3) 1817 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1818 error = 0; 1819 nfsmout: 1820 NDFREE(&nd, NDF_ONLY_PNBUF); 1821 if (nd.ni_dvp) { 1822 if (nd.ni_dvp == nd.ni_vp) 1823 vrele(nd.ni_dvp); 1824 else 1825 vput(nd.ni_dvp); 1826 } 1827 if (nd.ni_vp) 1828 vput(nd.ni_vp); 1829 vn_finished_write(mp); 1830 VFS_UNLOCK_GIANT(vfslocked); 1831 return(error); 1832 } 1833 1834 /* 1835 * nfs rename service 1836 */ 1837 int 1838 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 1839 struct mbuf **mrq) 1840 { 1841 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1842 struct sockaddr *nam = nfsd->nd_nam; 1843 caddr_t dpos = nfsd->nd_dpos; 1844 struct ucred *cred = nfsd->nd_cr; 1845 caddr_t bpos; 1846 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1; 1847 int tdirfor_ret = 1, tdiraft_ret = 1; 1848 int v3 = (nfsd->nd_flag & ND_NFSV3); 1849 struct mbuf *mb, *mreq; 1850 struct nameidata fromnd, tond; 1851 struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL; 1852 struct vnode *tdirp = NULL; 1853 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1854 nfsfh_t fnfh, tnfh; 1855 fhandle_t *ffhp, *tfhp; 1856 uid_t saved_uid; 1857 struct mount *mp = NULL; 1858 int vfslocked; 1859 1860 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 1861 vfslocked = 0; 1862 #ifndef nolint 1863 fvp = NULL; 1864 #endif 1865 ffhp = &fnfh.fh_generic; 1866 tfhp = &tnfh.fh_generic; 1867 1868 /* 1869 * Clear fields incase goto nfsmout occurs from macro. 1870 */ 1871 1872 ndclear(&fromnd); 1873 ndclear(&tond); 1874 1875 nfsm_srvmtofh(ffhp); 1876 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) { 1877 error = ESTALE; 1878 goto out1; 1879 } 1880 vfslocked = VFS_LOCK_GIANT(mp); 1881 (void) vn_start_write(NULL, &mp, V_WAIT); 1882 vfs_rel(mp); /* The write holds a ref. */ 1883 nfsm_srvnamesiz(len); 1884 /* 1885 * Remember our original uid so that we can reset cr_uid before 1886 * the second nfs_namei() call, in case it is remapped. 1887 */ 1888 saved_uid = cred->cr_uid; 1889 fromnd.ni_cnd.cn_cred = cred; 1890 fromnd.ni_cnd.cn_nameiop = DELETE; 1891 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART | MPSAFE; 1892 error = nfs_namei(&fromnd, nfsd, ffhp, len, slp, nam, &md, 1893 &dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, FALSE); 1894 vfslocked = nfsrv_lockedpair_nd(vfslocked, &fromnd); 1895 if (fdirp && !v3) { 1896 vrele(fdirp); 1897 fdirp = NULL; 1898 } 1899 if (error) { 1900 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1901 if (v3) { 1902 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1903 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1904 } 1905 error = 0; 1906 goto nfsmout; 1907 } 1908 fvp = fromnd.ni_vp; 1909 nfsm_srvmtofh(tfhp); 1910 nfsm_srvnamesiz(len2); 1911 cred->cr_uid = saved_uid; 1912 tond.ni_cnd.cn_cred = cred; 1913 tond.ni_cnd.cn_nameiop = RENAME; 1914 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | MPSAFE; 1915 error = nfs_namei(&tond, nfsd, tfhp, len2, slp, nam, &md, 1916 &dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, FALSE); 1917 vfslocked = nfsrv_lockedpair_nd(vfslocked, &tond); 1918 if (tdirp && !v3) { 1919 vrele(tdirp); 1920 tdirp = NULL; 1921 } 1922 if (error) 1923 goto out1; 1924 1925 tdvp = tond.ni_dvp; 1926 tvp = tond.ni_vp; 1927 if (tvp != NULL) { 1928 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1929 if (v3) 1930 error = EEXIST; 1931 else 1932 error = EISDIR; 1933 goto out; 1934 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1935 if (v3) 1936 error = EEXIST; 1937 else 1938 error = ENOTDIR; 1939 goto out; 1940 } 1941 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1942 if (v3) 1943 error = EXDEV; 1944 else 1945 error = ENOTEMPTY; 1946 goto out; 1947 } 1948 } 1949 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1950 if (v3) 1951 error = EXDEV; 1952 else 1953 error = ENOTEMPTY; 1954 goto out; 1955 } 1956 if (fvp->v_mount != tdvp->v_mount) { 1957 if (v3) 1958 error = EXDEV; 1959 else 1960 error = ENOTEMPTY; 1961 goto out; 1962 } 1963 if (fvp == tdvp) { 1964 if (v3) 1965 error = EINVAL; 1966 else 1967 error = ENOTEMPTY; 1968 } 1969 /* 1970 * If source is the same as the destination (that is the 1971 * same vnode with the same name in the same directory), 1972 * then there is nothing to do. 1973 */ 1974 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1975 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1976 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1977 fromnd.ni_cnd.cn_namelen)) 1978 error = -1; 1979 out: 1980 if (!error) { 1981 /* 1982 * The VOP_RENAME function releases all vnode references & 1983 * locks prior to returning so we need to clear the pointers 1984 * to bypass cleanup code later on. 1985 */ 1986 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1987 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1988 fromnd.ni_dvp = NULL; 1989 fromnd.ni_vp = NULL; 1990 tond.ni_dvp = NULL; 1991 tond.ni_vp = NULL; 1992 if (error) { 1993 NDFREE(&fromnd, NDF_ONLY_PNBUF); 1994 NDFREE(&tond, NDF_ONLY_PNBUF); 1995 } 1996 } else { 1997 if (error == -1) 1998 error = 0; 1999 } 2000 /* fall through */ 2001 out1: 2002 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2003 if (v3) { 2004 /* Release existing locks to prevent deadlock. */ 2005 if (tond.ni_dvp) { 2006 if (tond.ni_dvp == tond.ni_vp) 2007 vrele(tond.ni_dvp); 2008 else 2009 vput(tond.ni_dvp); 2010 } 2011 if (tond.ni_vp) 2012 vput(tond.ni_vp); 2013 tond.ni_dvp = NULL; 2014 tond.ni_vp = NULL; 2015 2016 if (fdirp) { 2017 vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY); 2018 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred); 2019 VOP_UNLOCK(fdirp, 0); 2020 } 2021 if (tdirp) { 2022 vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY); 2023 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred); 2024 VOP_UNLOCK(tdirp, 0); 2025 } 2026 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2027 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2028 } 2029 error = 0; 2030 /* fall through */ 2031 2032 nfsmout: 2033 /* 2034 * Clear out tond related fields 2035 */ 2036 if (tond.ni_dvp) { 2037 if (tond.ni_dvp == tond.ni_vp) 2038 vrele(tond.ni_dvp); 2039 else 2040 vput(tond.ni_dvp); 2041 } 2042 if (tond.ni_vp) 2043 vput(tond.ni_vp); 2044 if (tdirp) 2045 vrele(tdirp); 2046 if (tond.ni_startdir) 2047 vrele(tond.ni_startdir); 2048 NDFREE(&tond, NDF_ONLY_PNBUF); 2049 /* 2050 * Clear out fromnd related fields 2051 */ 2052 if (fdirp) 2053 vrele(fdirp); 2054 if (fromnd.ni_startdir) 2055 vrele(fromnd.ni_startdir); 2056 NDFREE(&fromnd, NDF_ONLY_PNBUF); 2057 if (fromnd.ni_dvp) 2058 vrele(fromnd.ni_dvp); 2059 if (fromnd.ni_vp) 2060 vrele(fromnd.ni_vp); 2061 2062 vn_finished_write(mp); 2063 VFS_UNLOCK_GIANT(vfslocked); 2064 return (error); 2065 } 2066 2067 /* 2068 * nfs link service 2069 */ 2070 int 2071 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 2072 struct mbuf **mrq) 2073 { 2074 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2075 struct sockaddr *nam = nfsd->nd_nam; 2076 caddr_t dpos = nfsd->nd_dpos; 2077 struct ucred *cred = nfsd->nd_cr; 2078 struct nameidata nd; 2079 caddr_t bpos; 2080 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1; 2081 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2082 struct mbuf *mb, *mreq; 2083 struct vnode *vp = NULL, *xp, *dirp = NULL; 2084 struct vattr dirfor, diraft, at; 2085 nfsfh_t nfh, dnfh; 2086 fhandle_t *fhp, *dfhp; 2087 struct mount *mp = NULL; 2088 int tvfslocked; 2089 int vfslocked; 2090 2091 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 2092 ndclear(&nd); 2093 vfslocked = 0; 2094 2095 fhp = &nfh.fh_generic; 2096 dfhp = &dnfh.fh_generic; 2097 nfsm_srvmtofh(fhp); 2098 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 2099 error = ESTALE; 2100 goto ereply; 2101 } 2102 vfslocked = VFS_LOCK_GIANT(mp); 2103 (void) vn_start_write(NULL, &mp, V_WAIT); 2104 vfs_rel(mp); /* The write holds a ref. */ 2105 nfsm_srvmtofh(dfhp); 2106 nfsm_srvnamesiz(len); 2107 2108 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly); 2109 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked); 2110 if (error) { 2111 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2112 if (v3) { 2113 nfsm_srvpostop_attr(getret, &at); 2114 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2115 } 2116 vp = NULL; 2117 error = 0; 2118 goto nfsmout; 2119 } 2120 if (v3) 2121 getret = VOP_GETATTR(vp, &at, cred); 2122 if (vp->v_type == VDIR) { 2123 error = EPERM; /* POSIX */ 2124 goto out1; 2125 } 2126 VOP_UNLOCK(vp, 0); 2127 nd.ni_cnd.cn_cred = cred; 2128 nd.ni_cnd.cn_nameiop = CREATE; 2129 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE | MPSAFE; 2130 error = nfs_namei(&nd, nfsd, dfhp, len, slp, nam, &md, &dpos, 2131 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 2132 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 2133 if (dirp && !v3) { 2134 vrele(dirp); 2135 dirp = NULL; 2136 } 2137 if (error) { 2138 vrele(vp); 2139 vp = NULL; 2140 goto out2; 2141 } 2142 xp = nd.ni_vp; 2143 if (xp != NULL) { 2144 error = EEXIST; 2145 vrele(vp); 2146 vp = NULL; 2147 goto out2; 2148 } 2149 xp = nd.ni_dvp; 2150 if (vp->v_mount != xp->v_mount) { 2151 error = EXDEV; 2152 vrele(vp); 2153 vp = NULL; 2154 goto out2; 2155 } 2156 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2157 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2158 NDFREE(&nd, NDF_ONLY_PNBUF); 2159 /* fall through */ 2160 2161 out1: 2162 if (v3) 2163 getret = VOP_GETATTR(vp, &at, cred); 2164 out2: 2165 if (dirp) { 2166 if (dirp == nd.ni_dvp) 2167 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2168 else { 2169 /* Release existing locks to prevent deadlock. */ 2170 if (nd.ni_dvp) { 2171 if (nd.ni_dvp == nd.ni_vp) 2172 vrele(nd.ni_dvp); 2173 else 2174 vput(nd.ni_dvp); 2175 } 2176 if (nd.ni_vp) 2177 vrele(nd.ni_vp); 2178 nd.ni_dvp = NULL; 2179 nd.ni_vp = NULL; 2180 2181 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 2182 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2183 VOP_UNLOCK(dirp, 0); 2184 } 2185 } 2186 ereply: 2187 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2188 if (v3) { 2189 nfsm_srvpostop_attr(getret, &at); 2190 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2191 } 2192 error = 0; 2193 /* fall through */ 2194 2195 nfsmout: 2196 NDFREE(&nd, NDF_ONLY_PNBUF); 2197 if (vp) 2198 vput(vp); 2199 if (nd.ni_dvp) { 2200 if (nd.ni_dvp == nd.ni_vp) 2201 vrele(nd.ni_dvp); 2202 else 2203 vput(nd.ni_dvp); 2204 } 2205 if (dirp) 2206 vrele(dirp); 2207 if (nd.ni_vp) 2208 vrele(nd.ni_vp); 2209 vn_finished_write(mp); 2210 VFS_UNLOCK_GIANT(vfslocked); 2211 return(error); 2212 } 2213 2214 /* 2215 * nfs symbolic link service 2216 */ 2217 int 2218 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 2219 struct mbuf **mrq) 2220 { 2221 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2222 struct sockaddr *nam = nfsd->nd_nam; 2223 caddr_t dpos = nfsd->nd_dpos; 2224 struct ucred *cred = nfsd->nd_cr; 2225 struct vattr va, dirfor, diraft; 2226 struct nameidata nd; 2227 struct vattr *vap = &va; 2228 struct nfsv2_sattr *sp; 2229 char *bpos, *pathcp = NULL; 2230 struct uio io; 2231 struct iovec iv; 2232 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1; 2233 int v3 = (nfsd->nd_flag & ND_NFSV3); 2234 struct mbuf *mb, *mreq; 2235 struct vnode *dirp = NULL; 2236 nfsfh_t nfh; 2237 fhandle_t *fhp; 2238 struct mount *mp = NULL; 2239 int tvfslocked; 2240 int vfslocked; 2241 2242 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 2243 ndclear(&nd); 2244 vfslocked = 0; 2245 2246 fhp = &nfh.fh_generic; 2247 nfsm_srvmtofh(fhp); 2248 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 2249 error = ESTALE; 2250 goto out; 2251 } 2252 vfslocked = VFS_LOCK_GIANT(mp); 2253 (void) vn_start_write(NULL, &mp, V_WAIT); 2254 vfs_rel(mp); /* The write holds a ref. */ 2255 nfsm_srvnamesiz(len); 2256 nd.ni_cnd.cn_cred = cred; 2257 nd.ni_cnd.cn_nameiop = CREATE; 2258 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART | MPSAFE; 2259 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 2260 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 2261 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 2262 if (error == 0) { 2263 VATTR_NULL(vap); 2264 if (v3) 2265 nfsm_srvsattr(vap); 2266 nfsm_srvpathsiz(len2); 2267 } 2268 if (dirp && !v3) { 2269 vrele(dirp); 2270 dirp = NULL; 2271 } 2272 if (error) 2273 goto out; 2274 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2275 iv.iov_base = pathcp; 2276 iv.iov_len = len2; 2277 io.uio_resid = len2; 2278 io.uio_offset = 0; 2279 io.uio_iov = &iv; 2280 io.uio_iovcnt = 1; 2281 io.uio_segflg = UIO_SYSSPACE; 2282 io.uio_rw = UIO_READ; 2283 io.uio_td = NULL; 2284 nfsm_mtouio(&io, len2); 2285 if (!v3) { 2286 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR); 2287 vap->va_mode = nfstov_mode(sp->sa_mode); 2288 } 2289 *(pathcp + len2) = '\0'; 2290 if (nd.ni_vp) { 2291 error = EEXIST; 2292 goto out; 2293 } 2294 2295 /* 2296 * issue symlink op. SAVESTART is set so the underlying path component 2297 * is only freed by the VOP if an error occurs. 2298 */ 2299 if (vap->va_mode == (mode_t)VNOVAL) 2300 vap->va_mode = 0; 2301 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp); 2302 if (error) 2303 NDFREE(&nd, NDF_ONLY_PNBUF); 2304 else 2305 vput(nd.ni_vp); 2306 nd.ni_vp = NULL; 2307 /* 2308 * releases directory prior to potential lookup op. 2309 */ 2310 vput(nd.ni_dvp); 2311 nd.ni_dvp = NULL; 2312 2313 if (error == 0) { 2314 if (v3) { 2315 /* 2316 * Issue lookup. Leave SAVESTART set so we can easily free 2317 * the name buffer later on. 2318 * 2319 * since LOCKPARENT is not set, ni_dvp will be garbage on 2320 * return whether an error occurs or not. 2321 */ 2322 nd.ni_cnd.cn_nameiop = LOOKUP; 2323 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW); 2324 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF); 2325 nd.ni_cnd.cn_thread = curthread; 2326 nd.ni_cnd.cn_cred = cred; 2327 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount); 2328 if (tvfslocked) 2329 nd.ni_cnd.cn_flags |= GIANTHELD; 2330 error = lookup(&nd); 2331 nd.ni_dvp = NULL; 2332 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 2333 nd.ni_cnd.cn_flags &= ~GIANTHELD; 2334 2335 if (error == 0) { 2336 bzero((caddr_t)fhp, sizeof(nfh)); 2337 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid; 2338 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2339 if (!error) 2340 error = VOP_GETATTR(nd.ni_vp, vap, cred); 2341 vput(nd.ni_vp); 2342 nd.ni_vp = NULL; 2343 } 2344 } 2345 } 2346 out: 2347 /* 2348 * These releases aren't strictly required, does even doing them 2349 * make any sense? XXX can nfsm_reply() block? 2350 */ 2351 if (pathcp) { 2352 free(pathcp, M_TEMP); 2353 pathcp = NULL; 2354 } 2355 if (dirp) { 2356 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 2357 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2358 VOP_UNLOCK(dirp, 0); 2359 } 2360 if (nd.ni_startdir) { 2361 vrele(nd.ni_startdir); 2362 nd.ni_startdir = NULL; 2363 } 2364 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2365 if (v3) { 2366 if (!error) { 2367 nfsm_srvpostop_fh(fhp); 2368 nfsm_srvpostop_attr(0, vap); 2369 } 2370 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2371 } 2372 error = 0; 2373 /* fall through */ 2374 2375 nfsmout: 2376 NDFREE(&nd, NDF_ONLY_PNBUF); 2377 if (nd.ni_dvp) { 2378 if (nd.ni_dvp == nd.ni_vp) 2379 vrele(nd.ni_dvp); 2380 else 2381 vput(nd.ni_dvp); 2382 } 2383 if (nd.ni_vp) 2384 vrele(nd.ni_vp); 2385 if (nd.ni_startdir) 2386 vrele(nd.ni_startdir); 2387 if (dirp) 2388 vrele(dirp); 2389 if (pathcp) 2390 free(pathcp, M_TEMP); 2391 2392 vn_finished_write(mp); 2393 VFS_UNLOCK_GIANT(vfslocked); 2394 return (error); 2395 } 2396 2397 /* 2398 * nfs mkdir service 2399 */ 2400 int 2401 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 2402 struct mbuf **mrq) 2403 { 2404 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2405 struct sockaddr *nam = nfsd->nd_nam; 2406 caddr_t dpos = nfsd->nd_dpos; 2407 struct ucred *cred = nfsd->nd_cr; 2408 struct vattr va, dirfor, diraft; 2409 struct vattr *vap = &va; 2410 struct nfs_fattr *fp; 2411 struct nameidata nd; 2412 u_int32_t *tl; 2413 caddr_t bpos; 2414 int error = 0, len, dirfor_ret = 1, diraft_ret = 1; 2415 int v3 = (nfsd->nd_flag & ND_NFSV3); 2416 struct mbuf *mb, *mreq; 2417 struct vnode *dirp = NULL; 2418 int vpexcl = 0; 2419 nfsfh_t nfh; 2420 fhandle_t *fhp; 2421 struct mount *mp = NULL; 2422 int vfslocked; 2423 2424 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 2425 ndclear(&nd); 2426 vfslocked = 0; 2427 2428 fhp = &nfh.fh_generic; 2429 nfsm_srvmtofh(fhp); 2430 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 2431 error = ESTALE; 2432 goto out; 2433 } 2434 vfslocked = VFS_LOCK_GIANT(mp); 2435 (void) vn_start_write(NULL, &mp, V_WAIT); 2436 vfs_rel(mp); /* The write holds a ref. */ 2437 nfsm_srvnamesiz(len); 2438 nd.ni_cnd.cn_cred = cred; 2439 nd.ni_cnd.cn_nameiop = CREATE; 2440 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE; 2441 2442 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 2443 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 2444 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 2445 if (dirp && !v3) { 2446 vrele(dirp); 2447 dirp = NULL; 2448 } 2449 if (error) { 2450 nfsm_reply(NFSX_WCCDATA(v3)); 2451 if (v3) 2452 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2453 error = 0; 2454 goto nfsmout; 2455 } 2456 VATTR_NULL(vap); 2457 if (v3) { 2458 nfsm_srvsattr(vap); 2459 } else { 2460 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED); 2461 vap->va_mode = nfstov_mode(*tl++); 2462 } 2463 2464 /* 2465 * At this point nd.ni_dvp is referenced and exclusively locked and 2466 * nd.ni_vp, if it exists, is referenced but not locked. 2467 */ 2468 2469 vap->va_type = VDIR; 2470 if (nd.ni_vp != NULL) { 2471 NDFREE(&nd, NDF_ONLY_PNBUF); 2472 error = EEXIST; 2473 goto out; 2474 } 2475 2476 /* 2477 * Issue mkdir op. Since SAVESTART is not set, the pathname 2478 * component is freed by the VOP call. This will fill-in 2479 * nd.ni_vp, reference, and exclusively lock it. 2480 */ 2481 if (vap->va_mode == (mode_t)VNOVAL) 2482 vap->va_mode = 0; 2483 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap); 2484 NDFREE(&nd, NDF_ONLY_PNBUF); 2485 vpexcl = 1; 2486 2487 vput(nd.ni_dvp); 2488 nd.ni_dvp = NULL; 2489 2490 if (!error) { 2491 bzero((caddr_t)fhp, sizeof(nfh)); 2492 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid; 2493 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2494 if (!error) 2495 error = VOP_GETATTR(nd.ni_vp, vap, cred); 2496 } 2497 out: 2498 if (dirp) { 2499 if (dirp == nd.ni_dvp) { 2500 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2501 } else { 2502 /* Release existing locks to prevent deadlock. */ 2503 if (nd.ni_dvp) { 2504 NDFREE(&nd, NDF_ONLY_PNBUF); 2505 if (nd.ni_dvp == nd.ni_vp && vpexcl) 2506 vrele(nd.ni_dvp); 2507 else 2508 vput(nd.ni_dvp); 2509 } 2510 if (nd.ni_vp) { 2511 if (vpexcl) 2512 vput(nd.ni_vp); 2513 else 2514 vrele(nd.ni_vp); 2515 } 2516 nd.ni_dvp = NULL; 2517 nd.ni_vp = NULL; 2518 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 2519 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2520 VOP_UNLOCK(dirp, 0); 2521 } 2522 } 2523 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2524 if (v3) { 2525 if (!error) { 2526 nfsm_srvpostop_fh(fhp); 2527 nfsm_srvpostop_attr(0, vap); 2528 } 2529 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2530 } else if (!error) { 2531 /* v2 non-error case. */ 2532 nfsm_srvfhtom(fhp, v3); 2533 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR); 2534 nfsm_srvfillattr(vap, fp); 2535 } 2536 error = 0; 2537 /* fall through */ 2538 2539 nfsmout: 2540 if (nd.ni_dvp) { 2541 NDFREE(&nd, NDF_ONLY_PNBUF); 2542 if (nd.ni_dvp == nd.ni_vp && vpexcl) 2543 vrele(nd.ni_dvp); 2544 else 2545 vput(nd.ni_dvp); 2546 } 2547 if (nd.ni_vp) { 2548 if (vpexcl) 2549 vput(nd.ni_vp); 2550 else 2551 vrele(nd.ni_vp); 2552 } 2553 if (dirp) 2554 vrele(dirp); 2555 vn_finished_write(mp); 2556 VFS_UNLOCK_GIANT(vfslocked); 2557 return (error); 2558 } 2559 2560 /* 2561 * nfs rmdir service 2562 */ 2563 int 2564 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 2565 struct mbuf **mrq) 2566 { 2567 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2568 struct sockaddr *nam = nfsd->nd_nam; 2569 caddr_t dpos = nfsd->nd_dpos; 2570 struct ucred *cred = nfsd->nd_cr; 2571 caddr_t bpos; 2572 int error = 0, len, dirfor_ret = 1, diraft_ret = 1; 2573 int v3 = (nfsd->nd_flag & ND_NFSV3); 2574 struct mbuf *mb, *mreq; 2575 struct vnode *vp, *dirp = NULL; 2576 struct vattr dirfor, diraft; 2577 nfsfh_t nfh; 2578 fhandle_t *fhp; 2579 struct nameidata nd; 2580 struct mount *mp = NULL; 2581 int vfslocked; 2582 2583 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 2584 ndclear(&nd); 2585 vfslocked = 0; 2586 2587 fhp = &nfh.fh_generic; 2588 nfsm_srvmtofh(fhp); 2589 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 2590 error = ESTALE; 2591 goto out; 2592 } 2593 vfslocked = VFS_LOCK_GIANT(mp); 2594 (void) vn_start_write(NULL, &mp, V_WAIT); 2595 vfs_rel(mp); /* The write holds a ref. */ 2596 nfsm_srvnamesiz(len); 2597 nd.ni_cnd.cn_cred = cred; 2598 nd.ni_cnd.cn_nameiop = DELETE; 2599 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE; 2600 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos, 2601 &dirp, v3, &dirfor, &dirfor_ret, FALSE); 2602 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd); 2603 if (dirp && !v3) { 2604 vrele(dirp); 2605 dirp = NULL; 2606 } 2607 if (error) { 2608 nfsm_reply(NFSX_WCCDATA(v3)); 2609 if (v3) 2610 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2611 error = 0; 2612 goto nfsmout; 2613 } 2614 vp = nd.ni_vp; 2615 if (vp->v_type != VDIR) { 2616 error = ENOTDIR; 2617 goto out; 2618 } 2619 /* 2620 * No rmdir "." please. 2621 */ 2622 if (nd.ni_dvp == vp) { 2623 error = EINVAL; 2624 goto out; 2625 } 2626 /* 2627 * The root of a mounted filesystem cannot be deleted. 2628 */ 2629 if (vp->v_vflag & VV_ROOT) 2630 error = EBUSY; 2631 out: 2632 /* 2633 * Issue or abort op. Since SAVESTART is not set, path name 2634 * component is freed by the VOP after either. 2635 */ 2636 if (!error) 2637 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2638 NDFREE(&nd, NDF_ONLY_PNBUF); 2639 2640 if (dirp) { 2641 if (dirp == nd.ni_dvp) 2642 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2643 else { 2644 /* Release existing locks to prevent deadlock. */ 2645 if (nd.ni_dvp) { 2646 if (nd.ni_dvp == nd.ni_vp) 2647 vrele(nd.ni_dvp); 2648 else 2649 vput(nd.ni_dvp); 2650 } 2651 if (nd.ni_vp) 2652 vput(nd.ni_vp); 2653 nd.ni_dvp = NULL; 2654 nd.ni_vp = NULL; 2655 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY); 2656 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2657 VOP_UNLOCK(dirp, 0); 2658 } 2659 } 2660 nfsm_reply(NFSX_WCCDATA(v3)); 2661 error = 0; 2662 if (v3) 2663 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2664 /* fall through */ 2665 2666 nfsmout: 2667 NDFREE(&nd, NDF_ONLY_PNBUF); 2668 if (nd.ni_dvp) { 2669 if (nd.ni_dvp == nd.ni_vp) 2670 vrele(nd.ni_dvp); 2671 else 2672 vput(nd.ni_dvp); 2673 } 2674 if (nd.ni_vp) 2675 vput(nd.ni_vp); 2676 if (dirp) 2677 vrele(dirp); 2678 2679 vn_finished_write(mp); 2680 VFS_UNLOCK_GIANT(vfslocked); 2681 return(error); 2682 } 2683 2684 /* 2685 * nfs readdir service 2686 * - mallocs what it thinks is enough to read 2687 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR 2688 * - calls VOP_READDIR() 2689 * - loops around building the reply 2690 * if the output generated exceeds count break out of loop 2691 * The nfsm_clget macro is used here so that the reply will be packed 2692 * tightly in mbuf clusters. 2693 * - it only knows that it has encountered eof when the VOP_READDIR() 2694 * reads nothing 2695 * - as such one readdir rpc will return eof false although you are there 2696 * and then the next will return eof 2697 * - it trims out records with d_fileno == 0 2698 * this doesn't matter for Unix clients, but they might confuse clients 2699 * for other os'. 2700 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2701 * than requested, but this may not apply to all filesystems. For 2702 * example, client NFS does not { although it is never remote mounted 2703 * anyhow } 2704 * The alternate call nfsrv_readdirplus() does lookups as well. 2705 * PS: The NFS protocol spec. does not clarify what the "count" byte 2706 * argument is a count of.. just name strings and file id's or the 2707 * entire reply rpc or ... 2708 * I tried just file name and id sizes and it confused the Sun client, 2709 * so I am using the full rpc size now. The "paranoia.." comment refers 2710 * to including the status longwords that are not a part of the dir. 2711 * "entry" structures, but are in the rpc. 2712 */ 2713 struct flrep { 2714 nfsuint64 fl_off; 2715 u_int32_t fl_postopok; 2716 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2717 u_int32_t fl_fhok; 2718 u_int32_t fl_fhsize; 2719 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)]; 2720 }; 2721 2722 int 2723 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 2724 struct mbuf **mrq) 2725 { 2726 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2727 struct sockaddr *nam = nfsd->nd_nam; 2728 caddr_t dpos = nfsd->nd_dpos; 2729 struct ucred *cred = nfsd->nd_cr; 2730 char *bp, *be; 2731 struct mbuf *mp; 2732 struct dirent *dp; 2733 caddr_t cp; 2734 u_int32_t *tl; 2735 caddr_t bpos; 2736 struct mbuf *mb, *mreq; 2737 char *cpos, *cend, *rbuf; 2738 struct vnode *vp = NULL; 2739 struct vattr at; 2740 nfsfh_t nfh; 2741 fhandle_t *fhp; 2742 struct uio io; 2743 struct iovec iv; 2744 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2745 int siz, cnt, fullsiz, eofflag, rdonly, ncookies; 2746 int v3 = (nfsd->nd_flag & ND_NFSV3); 2747 u_quad_t off, toff, verf; 2748 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */ 2749 int vfslocked, not_zfs; 2750 2751 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 2752 vfslocked = 0; 2753 fhp = &nfh.fh_generic; 2754 nfsm_srvmtofh(fhp); 2755 if (v3) { 2756 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED); 2757 toff = fxdr_hyper(tl); 2758 tl += 2; 2759 verf = fxdr_hyper(tl); 2760 tl += 2; 2761 } else { 2762 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED); 2763 toff = fxdr_unsigned(u_quad_t, *tl++); 2764 verf = 0; /* shut up gcc */ 2765 } 2766 off = toff; 2767 cnt = fxdr_unsigned(int, *tl); 2768 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2769 xfer = NFS_SRVMAXDATA(nfsd); 2770 if (cnt > xfer) 2771 cnt = xfer; 2772 if (siz > xfer) 2773 siz = xfer; 2774 fullsiz = siz; 2775 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 2776 if (!error && vp->v_type != VDIR) { 2777 error = ENOTDIR; 2778 vput(vp); 2779 vp = NULL; 2780 } 2781 if (error) { 2782 nfsm_reply(NFSX_UNSIGNED); 2783 if (v3) 2784 nfsm_srvpostop_attr(getret, &at); 2785 error = 0; 2786 goto nfsmout; 2787 } 2788 2789 /* 2790 * Obtain lock on vnode for this section of the code 2791 */ 2792 if (v3) { 2793 error = getret = VOP_GETATTR(vp, &at, cred); 2794 #if 0 2795 /* 2796 * XXX This check may be too strict for Solaris 2.5 clients. 2797 */ 2798 if (!error && toff && verf && verf != at.va_filerev) 2799 error = NFSERR_BAD_COOKIE; 2800 #endif 2801 } 2802 if (!error) 2803 error = nfsrv_access(vp, VEXEC, cred, rdonly, 0); 2804 if (error) { 2805 vput(vp); 2806 vp = NULL; 2807 nfsm_reply(NFSX_POSTOPATTR(v3)); 2808 if (v3) 2809 nfsm_srvpostop_attr(getret, &at); 2810 error = 0; 2811 goto nfsmout; 2812 } 2813 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") != 0; 2814 VOP_UNLOCK(vp, 0); 2815 2816 /* 2817 * end section. Allocate rbuf and continue 2818 */ 2819 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2820 again: 2821 iv.iov_base = rbuf; 2822 iv.iov_len = fullsiz; 2823 io.uio_iov = &iv; 2824 io.uio_iovcnt = 1; 2825 io.uio_offset = (off_t)off; 2826 io.uio_resid = fullsiz; 2827 io.uio_segflg = UIO_SYSSPACE; 2828 io.uio_rw = UIO_READ; 2829 io.uio_td = NULL; 2830 eofflag = 0; 2831 if (cookies) { 2832 free((caddr_t)cookies, M_TEMP); 2833 cookies = NULL; 2834 } 2835 vn_lock(vp, LK_SHARED | LK_RETRY); 2836 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies); 2837 off = (off_t)io.uio_offset; 2838 if (!cookies && !error) 2839 error = NFSERR_PERM; 2840 if (v3) { 2841 getret = VOP_GETATTR(vp, &at, cred); 2842 if (!error) 2843 error = getret; 2844 } 2845 VOP_UNLOCK(vp, 0); 2846 if (error) { 2847 vrele(vp); 2848 vp = NULL; 2849 free((caddr_t)rbuf, M_TEMP); 2850 if (cookies) 2851 free((caddr_t)cookies, M_TEMP); 2852 nfsm_reply(NFSX_POSTOPATTR(v3)); 2853 if (v3) 2854 nfsm_srvpostop_attr(getret, &at); 2855 error = 0; 2856 goto nfsmout; 2857 } 2858 if (io.uio_resid) { 2859 siz -= io.uio_resid; 2860 2861 /* 2862 * If nothing read, return eof 2863 * rpc reply 2864 */ 2865 if (siz == 0) { 2866 vrele(vp); 2867 vp = NULL; 2868 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2869 2 * NFSX_UNSIGNED); 2870 if (v3) { 2871 nfsm_srvpostop_attr(getret, &at); 2872 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); 2873 txdr_hyper(at.va_filerev, tl); 2874 tl += 2; 2875 } else 2876 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 2877 *tl++ = nfsrv_nfs_false; 2878 *tl = nfsrv_nfs_true; 2879 free((caddr_t)rbuf, M_TEMP); 2880 free((caddr_t)cookies, M_TEMP); 2881 error = 0; 2882 goto nfsmout; 2883 } 2884 } 2885 2886 /* 2887 * Check for degenerate cases of nothing useful read. 2888 * If so go try again 2889 */ 2890 cpos = rbuf; 2891 cend = rbuf + siz; 2892 dp = (struct dirent *)cpos; 2893 cookiep = cookies; 2894 /* 2895 * For some reason FreeBSD's ufs_readdir() chooses to back the 2896 * directory offset up to a block boundary, so it is necessary to 2897 * skip over the records that precede the requested offset. This 2898 * requires the assumption that file offset cookies monotonically 2899 * increase. 2900 * Since the offset cookies don't monotonically increase for ZFS, 2901 * this is not done when ZFS is the file system. 2902 */ 2903 while (cpos < cend && ncookies > 0 && 2904 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2905 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) { 2906 cpos += dp->d_reclen; 2907 dp = (struct dirent *)cpos; 2908 cookiep++; 2909 ncookies--; 2910 } 2911 if (cpos >= cend || ncookies == 0) { 2912 toff = off; 2913 siz = fullsiz; 2914 goto again; 2915 } 2916 2917 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2918 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2919 if (v3) { 2920 nfsm_srvpostop_attr(getret, &at); 2921 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 2922 txdr_hyper(at.va_filerev, tl); 2923 } 2924 mp = mb; 2925 bp = bpos; 2926 be = bp + M_TRAILINGSPACE(mp); 2927 2928 /* Loop through the records and build reply */ 2929 while (cpos < cend && ncookies > 0) { 2930 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2931 nlen = dp->d_namlen; 2932 rem = nfsm_rndup(nlen) - nlen; 2933 len += (4 * NFSX_UNSIGNED + nlen + rem); 2934 if (v3) 2935 len += 2 * NFSX_UNSIGNED; 2936 if (len > cnt) { 2937 eofflag = 0; 2938 break; 2939 } 2940 /* 2941 * Build the directory record xdr from 2942 * the dirent entry. 2943 */ 2944 nfsm_clget; 2945 *tl = nfsrv_nfs_true; 2946 bp += NFSX_UNSIGNED; 2947 if (v3) { 2948 nfsm_clget; 2949 *tl = 0; 2950 bp += NFSX_UNSIGNED; 2951 } 2952 nfsm_clget; 2953 *tl = txdr_unsigned(dp->d_fileno); 2954 bp += NFSX_UNSIGNED; 2955 nfsm_clget; 2956 *tl = txdr_unsigned(nlen); 2957 bp += NFSX_UNSIGNED; 2958 2959 /* And loop around copying the name */ 2960 xfer = nlen; 2961 cp = dp->d_name; 2962 while (xfer > 0) { 2963 nfsm_clget; 2964 if ((bp+xfer) > be) 2965 tsiz = be-bp; 2966 else 2967 tsiz = xfer; 2968 bcopy(cp, bp, tsiz); 2969 bp += tsiz; 2970 xfer -= tsiz; 2971 if (xfer > 0) 2972 cp += tsiz; 2973 } 2974 /* And null pad to an int32_t boundary. */ 2975 for (i = 0; i < rem; i++) 2976 *bp++ = '\0'; 2977 nfsm_clget; 2978 2979 /* Finish off the record */ 2980 if (v3) { 2981 *tl = 0; 2982 bp += NFSX_UNSIGNED; 2983 nfsm_clget; 2984 } 2985 *tl = txdr_unsigned(*cookiep); 2986 bp += NFSX_UNSIGNED; 2987 } 2988 cpos += dp->d_reclen; 2989 dp = (struct dirent *)cpos; 2990 cookiep++; 2991 ncookies--; 2992 } 2993 vrele(vp); 2994 vp = NULL; 2995 nfsm_clget; 2996 *tl = nfsrv_nfs_false; 2997 bp += NFSX_UNSIGNED; 2998 nfsm_clget; 2999 if (eofflag) 3000 *tl = nfsrv_nfs_true; 3001 else 3002 *tl = nfsrv_nfs_false; 3003 bp += NFSX_UNSIGNED; 3004 if (mp != mb) { 3005 if (bp < be) 3006 mp->m_len = bp - mtod(mp, caddr_t); 3007 } else 3008 mp->m_len += bp - bpos; 3009 free((caddr_t)rbuf, M_TEMP); 3010 free((caddr_t)cookies, M_TEMP); 3011 3012 nfsmout: 3013 if (vp) 3014 vrele(vp); 3015 VFS_UNLOCK_GIANT(vfslocked); 3016 return(error); 3017 } 3018 3019 int 3020 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3021 struct mbuf **mrq) 3022 { 3023 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3024 struct sockaddr *nam = nfsd->nd_nam; 3025 caddr_t dpos = nfsd->nd_dpos; 3026 struct ucred *cred = nfsd->nd_cr; 3027 char *bp, *be; 3028 struct mbuf *mp; 3029 struct dirent *dp; 3030 caddr_t cp; 3031 u_int32_t *tl; 3032 caddr_t bpos; 3033 struct mbuf *mb, *mreq; 3034 char *cpos, *cend, *rbuf; 3035 struct vnode *vp = NULL, *nvp; 3036 struct flrep fl; 3037 nfsfh_t nfh; 3038 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh; 3039 struct uio io; 3040 struct iovec iv; 3041 struct vattr va, at, *vap = &va; 3042 struct nfs_fattr *fp; 3043 int len, nlen, rem, xfer, tsiz, i, error = 0, error1, getret = 1; 3044 int vp_locked; 3045 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies; 3046 u_quad_t off, toff, verf; 3047 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */ 3048 int v3 = (nfsd->nd_flag & ND_NFSV3); 3049 int usevget = 1, vfslocked; 3050 struct componentname cn; 3051 struct mount *mntp = NULL; 3052 int not_zfs; 3053 3054 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3055 vfslocked = 0; 3056 vp_locked = 0; 3057 if (!v3) 3058 panic("nfsrv_readdirplus: v3 proc called on a v2 connection"); 3059 fhp = &nfh.fh_generic; 3060 nfsm_srvmtofh(fhp); 3061 tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED); 3062 toff = fxdr_hyper(tl); 3063 tl += 2; 3064 verf = fxdr_hyper(tl); 3065 tl += 2; 3066 siz = fxdr_unsigned(int, *tl++); 3067 cnt = fxdr_unsigned(int, *tl); 3068 off = toff; 3069 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 3070 xfer = NFS_SRVMAXDATA(nfsd); 3071 if (cnt > xfer) 3072 cnt = xfer; 3073 if (siz > xfer) 3074 siz = xfer; 3075 fullsiz = siz; 3076 error = nfsrv_fhtovp(fhp, NFSRV_FLAG_BUSY, &vp, &vfslocked, nfsd, slp, 3077 nam, &rdonly); 3078 if (!error) { 3079 vp_locked = 1; 3080 mntp = vp->v_mount; 3081 if (vp->v_type != VDIR) { 3082 error = ENOTDIR; 3083 vput(vp); 3084 vp = NULL; 3085 vp_locked = 0; 3086 } 3087 } 3088 if (error) { 3089 nfsm_reply(NFSX_UNSIGNED); 3090 nfsm_srvpostop_attr(getret, &at); 3091 error = 0; 3092 goto nfsmout; 3093 } 3094 error = getret = VOP_GETATTR(vp, &at, cred); 3095 #if 0 3096 /* 3097 * XXX This check may be too strict for Solaris 2.5 clients. 3098 */ 3099 if (!error && toff && verf && verf != at.va_filerev) 3100 error = NFSERR_BAD_COOKIE; 3101 #endif 3102 if (!error) 3103 error = nfsrv_access(vp, VEXEC, cred, rdonly, 0); 3104 if (error) { 3105 vput(vp); 3106 vp_locked = 0; 3107 vp = NULL; 3108 nfsm_reply(NFSX_V3POSTOPATTR); 3109 nfsm_srvpostop_attr(getret, &at); 3110 error = 0; 3111 goto nfsmout; 3112 } 3113 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") != 0; 3114 VOP_UNLOCK(vp, 0); 3115 vp_locked = 0; 3116 rbuf = malloc(siz, M_TEMP, M_WAITOK); 3117 again: 3118 iv.iov_base = rbuf; 3119 iv.iov_len = fullsiz; 3120 io.uio_iov = &iv; 3121 io.uio_iovcnt = 1; 3122 io.uio_offset = (off_t)off; 3123 io.uio_resid = fullsiz; 3124 io.uio_segflg = UIO_SYSSPACE; 3125 io.uio_rw = UIO_READ; 3126 io.uio_td = NULL; 3127 eofflag = 0; 3128 vp_locked = 1; 3129 if (cookies) { 3130 free((caddr_t)cookies, M_TEMP); 3131 cookies = NULL; 3132 } 3133 vn_lock(vp, LK_SHARED | LK_RETRY); 3134 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies); 3135 off = (u_quad_t)io.uio_offset; 3136 getret = VOP_GETATTR(vp, &at, cred); 3137 VOP_UNLOCK(vp, 0); 3138 vp_locked = 0; 3139 if (!cookies && !error) 3140 error = NFSERR_PERM; 3141 if (!error) 3142 error = getret; 3143 if (error) { 3144 vrele(vp); 3145 vp = NULL; 3146 if (cookies) 3147 free((caddr_t)cookies, M_TEMP); 3148 free((caddr_t)rbuf, M_TEMP); 3149 nfsm_reply(NFSX_V3POSTOPATTR); 3150 nfsm_srvpostop_attr(getret, &at); 3151 error = 0; 3152 goto nfsmout; 3153 } 3154 if (io.uio_resid) { 3155 siz -= io.uio_resid; 3156 3157 /* 3158 * If nothing read, return eof 3159 * rpc reply 3160 */ 3161 if (siz == 0) { 3162 vrele(vp); 3163 vp = NULL; 3164 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 3165 2 * NFSX_UNSIGNED); 3166 nfsm_srvpostop_attr(getret, &at); 3167 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); 3168 txdr_hyper(at.va_filerev, tl); 3169 tl += 2; 3170 *tl++ = nfsrv_nfs_false; 3171 *tl = nfsrv_nfs_true; 3172 free((caddr_t)cookies, M_TEMP); 3173 free((caddr_t)rbuf, M_TEMP); 3174 error = 0; 3175 goto nfsmout; 3176 } 3177 } 3178 3179 /* 3180 * Check for degenerate cases of nothing useful read. 3181 * If so go try again 3182 */ 3183 cpos = rbuf; 3184 cend = rbuf + siz; 3185 dp = (struct dirent *)cpos; 3186 cookiep = cookies; 3187 /* 3188 * For some reason FreeBSD's ufs_readdir() chooses to back the 3189 * directory offset up to a block boundary, so it is necessary to 3190 * skip over the records that precede the requested offset. This 3191 * requires the assumption that file offset cookies monotonically 3192 * increase. 3193 * Since the offset cookies don't monotonically increase for ZFS, 3194 * this is not done when ZFS is the file system. 3195 */ 3196 while (cpos < cend && ncookies > 0 && 3197 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 3198 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) { 3199 cpos += dp->d_reclen; 3200 dp = (struct dirent *)cpos; 3201 cookiep++; 3202 ncookies--; 3203 } 3204 if (cpos >= cend || ncookies == 0) { 3205 toff = off; 3206 siz = fullsiz; 3207 goto again; 3208 } 3209 3210 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 3211 2 * NFSX_UNSIGNED; 3212 nfsm_reply(cnt); 3213 nfsm_srvpostop_attr(getret, &at); 3214 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); 3215 txdr_hyper(at.va_filerev, tl); 3216 mp = mb; 3217 bp = bpos; 3218 be = bp + M_TRAILINGSPACE(mp); 3219 3220 /* Loop through the records and build reply */ 3221 while (cpos < cend && ncookies > 0) { 3222 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3223 nlen = dp->d_namlen; 3224 rem = nfsm_rndup(nlen)-nlen; 3225 3226 if (usevget) { 3227 /* 3228 * For readdir_and_lookup get the vnode using 3229 * the file number. 3230 */ 3231 error = VFS_VGET(mntp, dp->d_fileno, LK_SHARED, 3232 &nvp); 3233 if (error != 0 && error != EOPNOTSUPP) { 3234 error = 0; 3235 goto invalid; 3236 } else if (error == EOPNOTSUPP) { 3237 /* 3238 * VFS_VGET() not supported? 3239 * Let's switch to VOP_LOOKUP(). 3240 */ 3241 error = 0; 3242 usevget = 0; 3243 cn.cn_nameiop = LOOKUP; 3244 cn.cn_flags = ISLASTCN | NOFOLLOW | \ 3245 LOCKSHARED | LOCKLEAF | MPSAFE; 3246 cn.cn_lkflags = LK_SHARED | LK_RETRY; 3247 cn.cn_cred = cred; 3248 cn.cn_thread = curthread; 3249 } 3250 } 3251 if (!usevget) { 3252 cn.cn_nameptr = dp->d_name; 3253 cn.cn_namelen = dp->d_namlen; 3254 if (dp->d_namlen == 2 && 3255 dp->d_name[0] == '.' && 3256 dp->d_name[1] == '.') { 3257 cn.cn_flags |= ISDOTDOT; 3258 } else { 3259 cn.cn_flags &= ~ISDOTDOT; 3260 } 3261 if (!vp_locked) { 3262 vn_lock(vp, LK_SHARED | LK_RETRY); 3263 vp_locked = 1; 3264 } 3265 if ((vp->v_vflag & VV_ROOT) != 0 && 3266 (cn.cn_flags & ISDOTDOT) != 0) { 3267 vref(vp); 3268 nvp = vp; 3269 } else if (VOP_LOOKUP(vp, &nvp, &cn) != 0) 3270 goto invalid; 3271 } 3272 3273 bzero((caddr_t)nfhp, NFSX_V3FH); 3274 nfhp->fh_fsid = nvp->v_mount->mnt_stat.f_fsid; 3275 if ((error1 = VOP_VPTOFH(nvp, &nfhp->fh_fid)) == 0) 3276 error1 = VOP_GETATTR(nvp, vap, cred); 3277 if (!usevget && vp == nvp) 3278 vunref(nvp); 3279 else 3280 vput(nvp); 3281 nvp = NULL; 3282 if (error1 != 0) 3283 goto invalid; 3284 3285 /* 3286 * If either the dircount or maxcount will be 3287 * exceeded, get out now. Both of these lengths 3288 * are calculated conservatively, including all 3289 * XDR overheads. 3290 */ 3291 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3292 NFSX_V3POSTOPATTR); 3293 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3294 if (len > cnt || dirlen > fullsiz) { 3295 eofflag = 0; 3296 break; 3297 } 3298 3299 /* 3300 * Build the directory record xdr from 3301 * the dirent entry. 3302 */ 3303 fp = (struct nfs_fattr *)&fl.fl_fattr; 3304 nfsm_srvfillattr(vap, fp); 3305 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3306 fl.fl_fhok = nfsrv_nfs_true; 3307 fl.fl_postopok = nfsrv_nfs_true; 3308 fl.fl_off.nfsuquad[0] = 0; 3309 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep); 3310 3311 nfsm_clget; 3312 *tl = nfsrv_nfs_true; 3313 bp += NFSX_UNSIGNED; 3314 nfsm_clget; 3315 *tl = 0; 3316 bp += NFSX_UNSIGNED; 3317 nfsm_clget; 3318 *tl = txdr_unsigned(dp->d_fileno); 3319 bp += NFSX_UNSIGNED; 3320 nfsm_clget; 3321 *tl = txdr_unsigned(nlen); 3322 bp += NFSX_UNSIGNED; 3323 3324 /* And loop around copying the name */ 3325 xfer = nlen; 3326 cp = dp->d_name; 3327 while (xfer > 0) { 3328 nfsm_clget; 3329 if ((bp + xfer) > be) 3330 tsiz = be - bp; 3331 else 3332 tsiz = xfer; 3333 bcopy(cp, bp, tsiz); 3334 bp += tsiz; 3335 xfer -= tsiz; 3336 if (xfer > 0) 3337 cp += tsiz; 3338 } 3339 /* And null pad to an int32_t boundary. */ 3340 for (i = 0; i < rem; i++) 3341 *bp++ = '\0'; 3342 3343 /* 3344 * Now copy the flrep structure out. 3345 */ 3346 xfer = sizeof (struct flrep); 3347 cp = (caddr_t)&fl; 3348 while (xfer > 0) { 3349 nfsm_clget; 3350 if ((bp + xfer) > be) 3351 tsiz = be - bp; 3352 else 3353 tsiz = xfer; 3354 bcopy(cp, bp, tsiz); 3355 bp += tsiz; 3356 xfer -= tsiz; 3357 if (xfer > 0) 3358 cp += tsiz; 3359 } 3360 } 3361 invalid: 3362 cpos += dp->d_reclen; 3363 dp = (struct dirent *)cpos; 3364 cookiep++; 3365 ncookies--; 3366 } 3367 if (!usevget && vp_locked) 3368 vput(vp); 3369 else 3370 vrele(vp); 3371 vp = NULL; 3372 nfsm_clget; 3373 *tl = nfsrv_nfs_false; 3374 bp += NFSX_UNSIGNED; 3375 nfsm_clget; 3376 if (eofflag) 3377 *tl = nfsrv_nfs_true; 3378 else 3379 *tl = nfsrv_nfs_false; 3380 bp += NFSX_UNSIGNED; 3381 if (mp != mb) { 3382 if (bp < be) 3383 mp->m_len = bp - mtod(mp, caddr_t); 3384 } else 3385 mp->m_len += bp - bpos; 3386 free((caddr_t)cookies, M_TEMP); 3387 free((caddr_t)rbuf, M_TEMP); 3388 nfsmout: 3389 if (vp) 3390 vrele(vp); 3391 if (mntp) 3392 vfs_unbusy(mntp); 3393 VFS_UNLOCK_GIANT(vfslocked); 3394 return(error); 3395 } 3396 3397 /* 3398 * nfs commit service 3399 */ 3400 int 3401 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3402 struct mbuf **mrq) 3403 { 3404 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3405 struct sockaddr *nam = nfsd->nd_nam; 3406 caddr_t dpos = nfsd->nd_dpos; 3407 struct ucred *cred = nfsd->nd_cr; 3408 struct vattr bfor, aft; 3409 struct vnode *vp = NULL; 3410 nfsfh_t nfh; 3411 fhandle_t *fhp; 3412 u_int32_t *tl; 3413 caddr_t bpos; 3414 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt; 3415 struct mbuf *mb, *mreq; 3416 u_quad_t off; 3417 struct mount *mp = NULL; 3418 int v3 = (nfsd->nd_flag & ND_NFSV3); 3419 int tvfslocked; 3420 int vfslocked; 3421 3422 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3423 vfslocked = 0; 3424 if (!v3) 3425 panic("nfsrv_commit: v3 proc called on a v2 connection"); 3426 fhp = &nfh.fh_generic; 3427 nfsm_srvmtofh(fhp); 3428 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) { 3429 error = ESTALE; 3430 goto ereply; 3431 } 3432 vfslocked = VFS_LOCK_GIANT(mp); 3433 (void) vn_start_write(NULL, &mp, V_WAIT); 3434 vfs_rel(mp); /* The write holds a ref. */ 3435 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED); 3436 3437 /* 3438 * XXX At this time VOP_FSYNC() does not accept offset and byte 3439 * count parameters, so these arguments are useless (someday maybe). 3440 */ 3441 off = fxdr_hyper(tl); 3442 tl += 2; 3443 cnt = fxdr_unsigned(int, *tl); 3444 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly); 3445 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked); 3446 if (error) { 3447 nfsm_reply(2 * NFSX_UNSIGNED); 3448 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3449 error = 0; 3450 goto nfsmout; 3451 } 3452 for_ret = VOP_GETATTR(vp, &bfor, cred); 3453 3454 if (cnt > MAX_COMMIT_COUNT) { 3455 /* 3456 * Give up and do the whole thing 3457 */ 3458 if (vp->v_object && 3459 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 3460 VM_OBJECT_LOCK(vp->v_object); 3461 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 3462 VM_OBJECT_UNLOCK(vp->v_object); 3463 } 3464 error = VOP_FSYNC(vp, MNT_WAIT, curthread); 3465 } else { 3466 /* 3467 * Locate and synchronously write any buffers that fall 3468 * into the requested range. Note: we are assuming that 3469 * f_iosize is a power of 2. 3470 */ 3471 int iosize = vp->v_mount->mnt_stat.f_iosize; 3472 int iomask = iosize - 1; 3473 struct bufobj *bo; 3474 daddr_t lblkno; 3475 3476 /* 3477 * Align to iosize boundry, super-align to page boundry. 3478 */ 3479 if (off & iomask) { 3480 cnt += off & iomask; 3481 off &= ~(u_quad_t)iomask; 3482 } 3483 if (off & PAGE_MASK) { 3484 cnt += off & PAGE_MASK; 3485 off &= ~(u_quad_t)PAGE_MASK; 3486 } 3487 lblkno = off / iosize; 3488 3489 if (vp->v_object && 3490 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 3491 VM_OBJECT_LOCK(vp->v_object); 3492 vm_object_page_clean(vp->v_object, OFF_TO_IDX(off), 3493 OFF_TO_IDX(off + cnt + PAGE_MASK), OBJPC_SYNC); 3494 VM_OBJECT_UNLOCK(vp->v_object); 3495 } 3496 3497 bo = &vp->v_bufobj; 3498 BO_LOCK(bo); 3499 while (cnt > 0) { 3500 struct buf *bp; 3501 3502 /* 3503 * If we have a buffer and it is marked B_DELWRI we 3504 * have to lock and write it. Otherwise the prior 3505 * write is assumed to have already been committed. 3506 * 3507 * gbincore() can return invalid buffers now so we 3508 * have to check that bit as well (though B_DELWRI 3509 * should not be set if B_INVAL is set there could be 3510 * a race here since we haven't locked the buffer). 3511 */ 3512 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 3513 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 3514 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) { 3515 BO_LOCK(bo); 3516 continue; /* retry */ 3517 } 3518 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 3519 B_DELWRI) { 3520 bremfree(bp); 3521 bp->b_flags &= ~B_ASYNC; 3522 bwrite(bp); 3523 ++nfs_commit_miss; 3524 } else 3525 BUF_UNLOCK(bp); 3526 BO_LOCK(bo); 3527 } 3528 ++nfs_commit_blks; 3529 if (cnt < iosize) 3530 break; 3531 cnt -= iosize; 3532 ++lblkno; 3533 } 3534 BO_UNLOCK(bo); 3535 } 3536 3537 aft_ret = VOP_GETATTR(vp, &aft, cred); 3538 vput(vp); 3539 vp = NULL; 3540 ereply: 3541 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3542 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3543 if (!error) { 3544 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF); 3545 if (nfsver.tv_sec == 0) 3546 nfsver = boottime; 3547 *tl++ = txdr_unsigned(nfsver.tv_sec); 3548 *tl = txdr_unsigned(nfsver.tv_usec); 3549 } else { 3550 error = 0; 3551 } 3552 nfsmout: 3553 if (vp) 3554 vput(vp); 3555 vn_finished_write(mp); 3556 VFS_UNLOCK_GIANT(vfslocked); 3557 return(error); 3558 } 3559 3560 /* 3561 * nfs statfs service 3562 */ 3563 int 3564 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3565 struct mbuf **mrq) 3566 { 3567 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3568 struct sockaddr *nam = nfsd->nd_nam; 3569 caddr_t dpos = nfsd->nd_dpos; 3570 struct ucred *cred = nfsd->nd_cr; 3571 struct statfs *sf; 3572 struct nfs_statfs *sfp; 3573 caddr_t bpos; 3574 int error = 0, rdonly, getret = 1; 3575 int v3 = (nfsd->nd_flag & ND_NFSV3); 3576 struct mbuf *mb, *mreq; 3577 struct vnode *vp = NULL; 3578 struct vattr at; 3579 nfsfh_t nfh; 3580 fhandle_t *fhp; 3581 struct statfs statfs; 3582 u_quad_t tval; 3583 int vfslocked; 3584 3585 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3586 vfslocked = 0; 3587 fhp = &nfh.fh_generic; 3588 nfsm_srvmtofh(fhp); 3589 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 3590 if (error) { 3591 nfsm_reply(NFSX_UNSIGNED); 3592 if (v3) 3593 nfsm_srvpostop_attr(getret, &at); 3594 error = 0; 3595 goto nfsmout; 3596 } 3597 sf = &statfs; 3598 error = VFS_STATFS(vp->v_mount, sf); 3599 getret = VOP_GETATTR(vp, &at, cred); 3600 vput(vp); 3601 vp = NULL; 3602 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3603 if (v3) 3604 nfsm_srvpostop_attr(getret, &at); 3605 if (error) { 3606 error = 0; 3607 goto nfsmout; 3608 } 3609 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3)); 3610 if (v3) { 3611 tval = (u_quad_t)sf->f_blocks; 3612 tval *= (u_quad_t)sf->f_bsize; 3613 txdr_hyper(tval, &sfp->sf_tbytes); 3614 tval = (u_quad_t)sf->f_bfree; 3615 tval *= (u_quad_t)sf->f_bsize; 3616 txdr_hyper(tval, &sfp->sf_fbytes); 3617 /* 3618 * Don't send negative values for available space, 3619 * since this field is unsigned in the NFS protocol. 3620 * Otherwise, the client would see absurdly high 3621 * numbers for free space. 3622 */ 3623 if (sf->f_bavail < 0) 3624 tval = 0; 3625 else 3626 tval = (u_quad_t)sf->f_bavail; 3627 tval *= (u_quad_t)sf->f_bsize; 3628 txdr_hyper(tval, &sfp->sf_abytes); 3629 sfp->sf_tfiles.nfsuquad[0] = 0; 3630 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files); 3631 sfp->sf_ffiles.nfsuquad[0] = 0; 3632 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree); 3633 sfp->sf_afiles.nfsuquad[0] = 0; 3634 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree); 3635 sfp->sf_invarsec = 0; 3636 } else { 3637 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3638 sfp->sf_bsize = txdr_unsigned(sf->f_bsize); 3639 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3640 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3641 if (sf->f_bavail < 0) 3642 sfp->sf_bavail = 0; 3643 else 3644 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3645 } 3646 nfsmout: 3647 if (vp) 3648 vput(vp); 3649 VFS_UNLOCK_GIANT(vfslocked); 3650 return(error); 3651 } 3652 3653 /* 3654 * nfs fsinfo service 3655 */ 3656 int 3657 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3658 struct mbuf **mrq) 3659 { 3660 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3661 struct sockaddr *nam = nfsd->nd_nam; 3662 caddr_t dpos = nfsd->nd_dpos; 3663 struct ucred *cred = nfsd->nd_cr; 3664 struct nfsv3_fsinfo *sip; 3665 caddr_t bpos; 3666 int error = 0, rdonly, getret = 1, pref; 3667 struct mbuf *mb, *mreq; 3668 struct vnode *vp = NULL; 3669 struct vattr at; 3670 nfsfh_t nfh; 3671 fhandle_t *fhp; 3672 u_quad_t maxfsize; 3673 struct statfs sb; 3674 int v3 = (nfsd->nd_flag & ND_NFSV3); 3675 int vfslocked; 3676 3677 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3678 if (!v3) 3679 panic("nfsrv_fsinfo: v3 proc called on a v2 connection"); 3680 fhp = &nfh.fh_generic; 3681 vfslocked = 0; 3682 nfsm_srvmtofh(fhp); 3683 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 3684 if (error) { 3685 nfsm_reply(NFSX_UNSIGNED); 3686 nfsm_srvpostop_attr(getret, &at); 3687 error = 0; 3688 goto nfsmout; 3689 } 3690 3691 /* XXX Try to make a guess on the max file size. */ 3692 VFS_STATFS(vp->v_mount, &sb); 3693 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1; 3694 3695 getret = VOP_GETATTR(vp, &at, cred); 3696 vput(vp); 3697 vp = NULL; 3698 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3699 nfsm_srvpostop_attr(getret, &at); 3700 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3701 3702 /* 3703 * XXX 3704 * There should be filesystem VFS OP(s) to get this information. 3705 * For now, assume ufs. 3706 */ 3707 pref = NFS_SRVMAXDATA(nfsd); 3708 sip->fs_rtmax = txdr_unsigned(pref); 3709 sip->fs_rtpref = txdr_unsigned(pref); 3710 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3711 sip->fs_wtmax = txdr_unsigned(pref); 3712 sip->fs_wtpref = txdr_unsigned(pref); 3713 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3714 sip->fs_dtpref = txdr_unsigned(pref); 3715 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3716 sip->fs_timedelta.nfsv3_sec = 0; 3717 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3718 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3719 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3720 NFSV3FSINFO_CANSETTIME); 3721 nfsmout: 3722 if (vp) 3723 vput(vp); 3724 VFS_UNLOCK_GIANT(vfslocked); 3725 return(error); 3726 } 3727 3728 /* 3729 * nfs pathconf service 3730 */ 3731 int 3732 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3733 struct mbuf **mrq) 3734 { 3735 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3736 struct sockaddr *nam = nfsd->nd_nam; 3737 caddr_t dpos = nfsd->nd_dpos; 3738 struct ucred *cred = nfsd->nd_cr; 3739 struct nfsv3_pathconf *pc; 3740 caddr_t bpos; 3741 int error = 0, rdonly, getret = 1; 3742 register_t linkmax, namemax, chownres, notrunc; 3743 struct mbuf *mb, *mreq; 3744 struct vnode *vp = NULL; 3745 struct vattr at; 3746 nfsfh_t nfh; 3747 fhandle_t *fhp; 3748 int v3 = (nfsd->nd_flag & ND_NFSV3); 3749 int vfslocked; 3750 3751 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3752 if (!v3) 3753 panic("nfsrv_pathconf: v3 proc called on a v2 connection"); 3754 vfslocked = 0; 3755 fhp = &nfh.fh_generic; 3756 nfsm_srvmtofh(fhp); 3757 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly); 3758 if (error) { 3759 nfsm_reply(NFSX_UNSIGNED); 3760 nfsm_srvpostop_attr(getret, &at); 3761 error = 0; 3762 goto nfsmout; 3763 } 3764 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3765 if (!error) 3766 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3767 if (!error) 3768 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3769 if (!error) 3770 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3771 getret = VOP_GETATTR(vp, &at, cred); 3772 vput(vp); 3773 vp = NULL; 3774 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3775 nfsm_srvpostop_attr(getret, &at); 3776 if (error) { 3777 error = 0; 3778 goto nfsmout; 3779 } 3780 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3781 3782 pc->pc_linkmax = txdr_unsigned(linkmax); 3783 pc->pc_namemax = txdr_unsigned(namemax); 3784 pc->pc_notrunc = txdr_unsigned(notrunc); 3785 pc->pc_chownrestricted = txdr_unsigned(chownres); 3786 3787 /* 3788 * These should probably be supported by VOP_PATHCONF(), but 3789 * until msdosfs is exportable (why would you want to?), the 3790 * Unix defaults should be ok. 3791 */ 3792 pc->pc_caseinsensitive = nfsrv_nfs_false; 3793 pc->pc_casepreserving = nfsrv_nfs_true; 3794 nfsmout: 3795 if (vp) 3796 vput(vp); 3797 VFS_UNLOCK_GIANT(vfslocked); 3798 return(error); 3799 } 3800 3801 /* 3802 * Null operation, used by clients to ping server 3803 */ 3804 /* ARGSUSED */ 3805 int 3806 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3807 struct mbuf **mrq) 3808 { 3809 struct mbuf *mrep = nfsd->nd_mrep; 3810 caddr_t bpos; 3811 int error = NFSERR_RETVOID; 3812 struct mbuf *mb, *mreq; 3813 3814 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3815 nfsm_reply(0); 3816 nfsmout: 3817 return (error); 3818 } 3819 3820 /* 3821 * No operation, used for obsolete procedures 3822 */ 3823 /* ARGSUSED */ 3824 int 3825 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3826 struct mbuf **mrq) 3827 { 3828 struct mbuf *mrep = nfsd->nd_mrep; 3829 caddr_t bpos; 3830 int error; 3831 struct mbuf *mb, *mreq; 3832 3833 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3834 if (nfsd->nd_repstat) 3835 error = nfsd->nd_repstat; 3836 else 3837 error = EPROCUNAVAIL; 3838 nfsm_reply(0); 3839 error = 0; 3840 nfsmout: 3841 return (error); 3842 } 3843 3844 /* 3845 * Perform access checking for vnodes obtained from file handles that would 3846 * refer to files already opened by a Unix client. You cannot just use 3847 * vn_writechk() and VOP_ACCESS() for two reasons. 3848 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 3849 * case. 3850 * 2 - The owner is to be given access irrespective of mode bits for some 3851 * operations, so that processes that chmod after opening a file don't 3852 * break. I don't like this because it opens a security hole, but since 3853 * the nfs server opens a security hole the size of a barn door anyhow, 3854 * what the heck. 3855 * 3856 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3857 * will return EPERM instead of EACCES. EPERM is always an error. 3858 */ 3859 static int 3860 nfsrv_access(struct vnode *vp, accmode_t accmode, struct ucred *cred, 3861 int rdonly, int override) 3862 { 3863 struct vattr vattr; 3864 int error; 3865 3866 VFS_ASSERT_GIANT(vp->v_mount); 3867 3868 nfsdbprintf(("%s %d\n", __FILE__, __LINE__)); 3869 3870 if (accmode & VWRITE) { 3871 /* Just vn_writechk() changed to check rdonly */ 3872 /* 3873 * Disallow write attempts on read-only filesystems; 3874 * unless the file is a socket or a block or character 3875 * device resident on the filesystem. 3876 */ 3877 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3878 switch (vp->v_type) { 3879 case VREG: 3880 case VDIR: 3881 case VLNK: 3882 return (EROFS); 3883 default: 3884 break; 3885 } 3886 } 3887 /* 3888 * If there's shared text associated with 3889 * the inode, we can't allow writing. 3890 */ 3891 if (vp->v_vflag & VV_TEXT) 3892 return (ETXTBSY); 3893 } 3894 3895 error = VOP_GETATTR(vp, &vattr, cred); 3896 if (error) 3897 return (error); 3898 error = VOP_ACCESS(vp, accmode, cred, curthread); 3899 /* 3900 * Allow certain operations for the owner (reads and writes 3901 * on files that are already open). 3902 */ 3903 if (override && error == EACCES && cred->cr_uid == vattr.va_uid) 3904 error = 0; 3905 return (error); 3906 }
Cache object: 50abcc5f3b2ef7f693ba61d5b57c8b52
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]