nfs_serv.c revision 1.157
1 /* $NetBSD: nfs_serv.c,v 1.157 2011/03/19 01:34:24 dholland Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * nfs version 2 and 3 server calls to vnode ops 39 * - these routines generally have 3 phases 40 * 1 - break down and validate rpc request in mbuf list 41 * 2 - do the vnode ops for the request 42 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c) 43 * 3 - build the rpc reply in an mbuf list 44 * nb: 45 * - do not mix the phases, since the nfsm_?? macros can return failures 46 * on a bad rpc or similar and do not do any vrele() or vput()'s 47 * 48 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs 49 * error number iff error != 0 whereas 50 * returning an error from the server function implies a fatal error 51 * such as a badly constructed rpc request that should be dropped without 52 * a reply. 53 * For Version 3, nfsm_reply() does not return for the error case, since 54 * most version 3 rpcs return more than the status for error cases. 55 */ 56 57 #include <sys/cdefs.h> 58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.157 2011/03/19 01:34:24 dholland Exp $"); 59 60 #include <sys/param.h> 61 #include <sys/systm.h> 62 #include <sys/proc.h> 63 #include <sys/file.h> 64 #include <sys/namei.h> 65 #include <sys/vnode.h> 66 #include <sys/mount.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/mbuf.h> 70 #include <sys/dirent.h> 71 #include <sys/stat.h> 72 #include <sys/kernel.h> 73 #include <sys/hash.h> 74 #include <sys/kauth.h> 75 #include <sys/module.h> 76 #include <sys/syscall.h> 77 #include <sys/syscallargs.h> 78 #include <sys/syscallvar.h> 79 80 #include <uvm/uvm.h> 81 82 #include <nfs/nfsproto.h> 83 #include <nfs/rpcv2.h> 84 #include <nfs/nfs.h> 85 #include <nfs/xdr_subs.h> 86 #include <nfs/nfsm_subs.h> 87 #include <nfs/nfs_var.h> 88 89 MODULE(MODULE_CLASS_MISC, nfsserver, "nfs"); 90 91 /* Global vars */ 92 extern u_int32_t nfs_xdrneg1; 93 extern u_int32_t nfs_false, nfs_true; 94 extern const enum vtype nv3tov_type[8]; 95 extern struct nfsstats nfsstats; 96 extern const nfstype nfsv2_type[9]; 97 extern const nfstype nfsv3_type[9]; 98 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000; 99 bool nfsd_use_loan = true; /* use page-loan for READ OP */ 100 101 #define nqsrv_getl(vp, rw) /* nothing */ 102 103 static const struct syscall_package nfsserver_syscalls[] = { 104 { SYS_nfssvc, 0, (sy_call_t *)sys_nfssvc }, 105 { 0, 0, NULL }, 106 }; 107 108 static int 109 nfsserver_modcmd(modcmd_t cmd, void *arg) 110 { 111 extern struct vfs_hooks nfs_export_hooks; /* XXX */ 112 int error; 113 114 switch (cmd) { 115 case MODULE_CMD_INIT: 116 error = syscall_establish(NULL, nfsserver_syscalls); 117 if (error != 0) { 118 return error; 119 } 120 nfs_init(); /* XXX for monolithic kernel */ 121 netexport_init(); 122 nfsrv_initcache(); /* Init the server request cache */ 123 nfsrv_init(0); /* Init server data structures */ 124 vfs_hooks_attach(&nfs_export_hooks); 125 nfs_timer_srvinit(nfsrv_timer); 126 return 0; 127 case MODULE_CMD_FINI: 128 error = syscall_disestablish(NULL, nfsserver_syscalls); 129 if (error != 0) { 130 return error; 131 } 132 /* 133 * Kill export list before detaching VFS hooks, so we 134 * we don't leak state due to a concurrent umount(). 135 */ 136 netexport_fini(); 137 vfs_hooks_detach(&nfs_export_hooks); 138 139 /* Kill timer before server goes away. */ 140 nfs_timer_srvfini(); 141 nfsrv_fini(); 142 143 /* Server uses server cache, so kill cache last. */ 144 nfsrv_finicache(); 145 return 0; 146 default: 147 return ENOTTY; 148 } 149 } 150 151 /* 152 * nfs v3 access service 153 */ 154 int 155 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 156 { 157 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 158 struct mbuf *nam = nfsd->nd_nam; 159 char *dpos = nfsd->nd_dpos; 160 kauth_cred_t cred = nfsd->nd_cr; 161 struct vnode *vp; 162 nfsrvfh_t nsfh; 163 u_int32_t *tl; 164 int32_t t1; 165 char *bpos; 166 int error = 0, rdonly, cache = 0, getret; 167 char *cp2; 168 struct mbuf *mb, *mreq; 169 struct vattr va; 170 u_long inmode, testmode, outmode; 171 u_quad_t frev; 172 173 nfsm_srvmtofh(&nsfh); 174 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 175 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 176 (nfsd->nd_flag & ND_KERBAUTH), false); 177 if (error) { 178 nfsm_reply(NFSX_UNSIGNED); 179 nfsm_srvpostop_attr(1, (struct vattr *)0); 180 return (0); 181 } 182 inmode = fxdr_unsigned(u_int32_t, *tl); 183 outmode = 0; 184 if ((inmode & NFSV3ACCESS_READ) && 185 nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0) 186 outmode |= NFSV3ACCESS_READ; 187 if (vp->v_type != VDIR) { 188 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 189 if (testmode && 190 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 191 outmode |= testmode; 192 if ((inmode & NFSV3ACCESS_EXECUTE) && 193 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 194 outmode |= NFSV3ACCESS_EXECUTE; 195 } else { 196 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 197 NFSV3ACCESS_DELETE); 198 if (testmode && 199 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 200 outmode |= testmode; 201 if ((inmode & NFSV3ACCESS_LOOKUP) && 202 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 203 outmode |= NFSV3ACCESS_LOOKUP; 204 } 205 getret = VOP_GETATTR(vp, &va, cred); 206 vput(vp); 207 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 208 nfsm_srvpostop_attr(getret, &va); 209 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 210 *tl = txdr_unsigned(outmode); 211 nfsm_srvdone; 212 } 213 214 /* 215 * nfs getattr service 216 */ 217 int 218 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 219 { 220 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 221 struct mbuf *nam = nfsd->nd_nam; 222 char *dpos = nfsd->nd_dpos; 223 kauth_cred_t cred = nfsd->nd_cr; 224 struct nfs_fattr *fp; 225 struct vattr va; 226 struct vnode *vp; 227 nfsrvfh_t nsfh; 228 u_int32_t *tl; 229 int32_t t1; 230 char *bpos; 231 int error = 0, rdonly, cache = 0; 232 char *cp2; 233 struct mbuf *mb, *mreq; 234 u_quad_t frev; 235 236 nfsm_srvmtofh(&nsfh); 237 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 238 (nfsd->nd_flag & ND_KERBAUTH), false); 239 if (error) { 240 nfsm_reply(0); 241 return (0); 242 } 243 nqsrv_getl(vp, ND_READ); 244 error = VOP_GETATTR(vp, &va, cred); 245 vput(vp); 246 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 247 if (error) 248 return (0); 249 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 250 nfsm_srvfillattr(&va, fp); 251 nfsm_srvdone; 252 } 253 254 /* 255 * nfs setattr service 256 */ 257 int 258 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 259 { 260 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 261 struct mbuf *nam = nfsd->nd_nam; 262 char *dpos = nfsd->nd_dpos; 263 kauth_cred_t cred = nfsd->nd_cr; 264 struct vattr va, preat; 265 struct nfsv2_sattr *sp; 266 struct nfs_fattr *fp; 267 struct vnode *vp; 268 nfsrvfh_t nsfh; 269 u_int32_t *tl; 270 int32_t t1; 271 char *bpos; 272 int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1; 273 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0; 274 char *cp2; 275 struct mbuf *mb, *mreq; 276 u_quad_t frev; 277 struct timespec guard; 278 279 memset(&guard, 0, sizeof guard); /* XXX gcc */ 280 281 nfsm_srvmtofh(&nsfh); 282 vattr_null(&va); 283 if (v3) { 284 nfsm_srvsattr(&va); 285 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 286 gcheck = fxdr_unsigned(int, *tl); 287 if (gcheck) { 288 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 289 fxdr_nfsv3time(tl, &guard); 290 } 291 } else { 292 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 293 /* 294 * Nah nah nah nah na nah 295 * There is a bug in the Sun client that puts 0xffff in the mode 296 * field of sattr when it should put in 0xffffffff. The u_short 297 * doesn't sign extend. 298 * --> check the low order 2 bytes for 0xffff 299 */ 300 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 301 va.va_mode = nfstov_mode(sp->sa_mode); 302 if (sp->sa_uid != nfs_xdrneg1) 303 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 304 if (sp->sa_gid != nfs_xdrneg1) 305 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 306 if (sp->sa_size != nfs_xdrneg1) 307 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 308 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 309 #ifdef notyet 310 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 311 #else 312 va.va_atime.tv_sec = 313 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 314 va.va_atime.tv_nsec = 0; 315 #endif 316 } 317 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 318 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 319 320 } 321 322 /* 323 * Now that we have all the fields, lets do it. 324 */ 325 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 326 (nfsd->nd_flag & ND_KERBAUTH), false); 327 if (error) { 328 nfsm_reply(2 * NFSX_UNSIGNED); 329 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 330 return (0); 331 } 332 nqsrv_getl(vp, ND_WRITE); 333 if (v3) { 334 error = preat_ret = VOP_GETATTR(vp, &preat, cred); 335 if (!error && gcheck && 336 (preat.va_ctime.tv_sec != guard.tv_sec || 337 preat.va_ctime.tv_nsec != guard.tv_nsec)) 338 error = NFSERR_NOT_SYNC; 339 if (error) { 340 vput(vp); 341 nfsm_reply(NFSX_WCCDATA(v3)); 342 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 343 return (0); 344 } 345 } 346 347 /* 348 * If the size is being changed write acces is required, otherwise 349 * just check for a read only file system. 350 */ 351 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 352 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 353 error = EROFS; 354 goto out; 355 } 356 } else { 357 if (vp->v_type == VDIR) { 358 error = EISDIR; 359 goto out; 360 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 361 lwp, 0)) != 0) 362 goto out; 363 } 364 error = VOP_SETATTR(vp, &va, cred); 365 postat_ret = VOP_GETATTR(vp, &va, cred); 366 if (!error) 367 error = postat_ret; 368 out: 369 vput(vp); 370 nfsm_reply(NFSX_WCCORFATTR(v3)); 371 if (v3) { 372 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 373 return (0); 374 } else { 375 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 376 nfsm_srvfillattr(&va, fp); 377 } 378 nfsm_srvdone; 379 } 380 381 /* 382 * nfs lookup rpc 383 */ 384 int 385 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 386 { 387 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 388 struct mbuf *nam = nfsd->nd_nam; 389 char *dpos = nfsd->nd_dpos; 390 kauth_cred_t cred = nfsd->nd_cr; 391 struct nfs_fattr *fp; 392 struct nameidata nd, ind, *ndp = &nd; 393 struct pathbuf *ipb = NULL; 394 struct vnode *vp, *dirp; 395 nfsrvfh_t nsfh; 396 char *cp; 397 u_int32_t *tl; 398 int32_t t1; 399 char *bpos; 400 int error = 0, cache = 0, dirattr_ret = 1; 401 uint32_t len; 402 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 403 char *cp2; 404 struct mbuf *mb, *mreq; 405 struct vattr va, dirattr; 406 u_quad_t frev; 407 408 nfsm_srvmtofh(&nsfh); 409 nfsm_srvnamesiz(len); 410 411 pubflag = nfs_ispublicfh(&nsfh); 412 413 nd.ni_cnd.cn_cred = cred; 414 nd.ni_cnd.cn_nameiop = LOOKUP; 415 nd.ni_cnd.cn_flags = LOCKLEAF; 416 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 417 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 418 419 if (!error && pubflag) { 420 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL && 421 (ipb = pathbuf_create(nfs_pub.np_index)) != NULL) { 422 /* 423 * Setup call to lookup() to see if we can find 424 * the index file. Arguably, this doesn't belong 425 * in a kernel.. Ugh. 426 */ 427 ind = nd; 428 VOP_UNLOCK(nd.ni_vp); 429 ind.ni_pathbuf = ipb; 430 ind.ni_pathlen = strlen(nfs_pub.np_index); 431 ind.ni_pnbuf = NULL; 432 ind.ni_cnd.cn_nameptr = NULL; 433 434 error = lookup_for_nfsd_index(&ind, nd.ni_vp); 435 if (!error) { 436 /* 437 * Found an index file. Get rid of 438 * the old references. 439 */ 440 if (dirp) 441 vrele(dirp); 442 dirp = nd.ni_vp; 443 ndp = &ind; 444 } else 445 error = 0; 446 } 447 /* 448 * If the public filehandle was used, check that this lookup 449 * didn't result in a filehandle outside the publicly exported 450 * filesystem. 451 */ 452 453 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 454 vput(nd.ni_vp); 455 error = EPERM; 456 } 457 } 458 459 if (dirp) { 460 if (v3) 461 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred); 462 vrele(dirp); 463 } 464 465 if (error) { 466 if (nd.ni_pathbuf != NULL) { 467 pathbuf_destroy(nd.ni_pathbuf); 468 } 469 if (ipb != NULL) { 470 pathbuf_destroy(ipb); 471 } 472 nfsm_reply(NFSX_POSTOPATTR(v3)); 473 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 474 return (0); 475 } 476 477 nqsrv_getl(ndp->ni_startdir, ND_READ); 478 pathbuf_destroy(nd.ni_pathbuf); 479 if (ipb != NULL) { 480 pathbuf_destroy(ipb); 481 } 482 vp = ndp->ni_vp; 483 error = nfsrv_composefh(vp, &nsfh, v3); 484 if (!error) 485 error = VOP_GETATTR(vp, &va, cred); 486 vput(vp); 487 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) + 488 NFSX_POSTOPATTR(v3)); 489 if (error) { 490 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 491 return (0); 492 } 493 nfsm_srvfhtom(&nsfh, v3); 494 if (v3) { 495 nfsm_srvpostop_attr(0, &va); 496 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 497 } else { 498 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 499 nfsm_srvfillattr(&va, fp); 500 } 501 nfsm_srvdone; 502 } 503 504 /* 505 * nfs readlink service 506 */ 507 int 508 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 509 { 510 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 511 struct mbuf *nam = nfsd->nd_nam; 512 char *dpos = nfsd->nd_dpos; 513 kauth_cred_t cred = nfsd->nd_cr; 514 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 515 struct iovec *ivp = iv; 516 struct mbuf *mp; 517 u_int32_t *tl; 518 int32_t t1; 519 char *bpos; 520 int error = 0, rdonly, cache = 0, i, padlen, getret; 521 uint32_t len; 522 int v3 = (nfsd->nd_flag & ND_NFSV3); 523 char *cp2; 524 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 525 struct vnode *vp; 526 struct vattr attr; 527 nfsrvfh_t nsfh; 528 struct uio io, *uiop = &io; 529 u_quad_t frev; 530 531 nfsm_srvmtofh(&nsfh); 532 len = 0; 533 i = 0; 534 while (len < NFS_MAXPATHLEN) { 535 mp = m_get(M_WAIT, MT_DATA); 536 MCLAIM(mp, &nfs_mowner); 537 m_clget(mp, M_WAIT); 538 mp->m_len = NFSMSIZ(mp); 539 if (len == 0) 540 mp3 = mp2 = mp; 541 else { 542 mp2->m_next = mp; 543 mp2 = mp; 544 } 545 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 546 mp->m_len = NFS_MAXPATHLEN-len; 547 len = NFS_MAXPATHLEN; 548 } else 549 len += mp->m_len; 550 ivp->iov_base = mtod(mp, void *); 551 ivp->iov_len = mp->m_len; 552 i++; 553 ivp++; 554 } 555 uiop->uio_iov = iv; 556 uiop->uio_iovcnt = i; 557 uiop->uio_offset = 0; 558 uiop->uio_resid = len; 559 uiop->uio_rw = UIO_READ; 560 UIO_SETUP_SYSSPACE(uiop); 561 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 562 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 563 if (error) { 564 m_freem(mp3); 565 nfsm_reply(2 * NFSX_UNSIGNED); 566 nfsm_srvpostop_attr(1, (struct vattr *)0); 567 return (0); 568 } 569 if (vp->v_type != VLNK) { 570 if (v3) 571 error = EINVAL; 572 else 573 error = ENXIO; 574 goto out; 575 } 576 nqsrv_getl(vp, ND_READ); 577 error = VOP_READLINK(vp, uiop, cred); 578 out: 579 getret = VOP_GETATTR(vp, &attr, cred); 580 vput(vp); 581 if (error) 582 m_freem(mp3); 583 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 584 if (v3) { 585 nfsm_srvpostop_attr(getret, &attr); 586 if (error) 587 return (0); 588 } 589 len -= uiop->uio_resid; 590 padlen = nfsm_padlen(len); 591 if (uiop->uio_resid || padlen) 592 nfs_zeropad(mp3, uiop->uio_resid, padlen); 593 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 594 *tl = txdr_unsigned(len); 595 mb->m_next = mp3; 596 nfsm_srvdone; 597 } 598 599 /* 600 * nfs read service 601 */ 602 int 603 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 604 { 605 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 606 struct mbuf *nam = nfsd->nd_nam; 607 char *dpos = nfsd->nd_dpos; 608 kauth_cred_t cred = nfsd->nd_cr; 609 struct mbuf *m; 610 struct nfs_fattr *fp; 611 u_int32_t *tl; 612 int32_t t1; 613 int i; 614 char *bpos; 615 int error = 0, rdonly, cache = 0, getret; 616 int v3 = (nfsd->nd_flag & ND_NFSV3); 617 uint32_t reqlen, len, cnt, left; 618 int padlen; 619 char *cp2; 620 struct mbuf *mb, *mreq; 621 struct vnode *vp; 622 nfsrvfh_t nsfh; 623 struct uio io, *uiop = &io; 624 struct vattr va; 625 off_t off; 626 u_quad_t frev; 627 628 nfsm_srvmtofh(&nsfh); 629 if (v3) { 630 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 631 off = fxdr_hyper(tl); 632 } else { 633 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 634 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 635 } 636 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 637 reqlen = fxdr_unsigned(uint32_t, *tl); 638 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 639 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 640 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 641 if (error) { 642 nfsm_reply(2 * NFSX_UNSIGNED); 643 nfsm_srvpostop_attr(1, (struct vattr *)0); 644 return (0); 645 } 646 if (vp->v_type != VREG) { 647 if (v3) 648 error = EINVAL; 649 else 650 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 651 } 652 if (!error) { 653 nqsrv_getl(vp, ND_READ); 654 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 655 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 656 } 657 getret = VOP_GETATTR(vp, &va, cred); 658 if (!error) 659 error = getret; 660 if (error) { 661 vput(vp); 662 nfsm_reply(NFSX_POSTOPATTR(v3)); 663 nfsm_srvpostop_attr(getret, &va); 664 return (0); 665 } 666 if (off >= va.va_size) 667 cnt = 0; 668 else if ((off + reqlen) > va.va_size) 669 cnt = va.va_size - off; 670 else 671 cnt = reqlen; 672 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 673 if (v3) { 674 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 675 *tl++ = nfs_true; 676 fp = (struct nfs_fattr *)tl; 677 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 678 } else { 679 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 680 fp = (struct nfs_fattr *)tl; 681 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 682 } 683 len = left = cnt; 684 if (cnt > 0) { 685 if (nfsd_use_loan) { 686 struct vm_page **pgpp; 687 voff_t pgoff = trunc_page(off); 688 int npages; 689 vaddr_t lva; 690 691 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 692 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 693 694 /* allocate kva for mbuf data */ 695 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 696 if (lva == 0) { 697 /* fall back to VOP_READ */ 698 goto loan_fail; 699 } 700 701 /* allocate mbuf */ 702 m = m_get(M_WAIT, MT_DATA); 703 MCLAIM(m, &nfs_mowner); 704 pgpp = m->m_ext.ext_pgs; 705 706 /* loan pages */ 707 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 708 pgpp); 709 if (error) { 710 sokvafree(lva, npages << PAGE_SHIFT); 711 m_free(m); 712 if (error == EBUSY) 713 goto loan_fail; 714 goto read_error; 715 } 716 717 /* associate kva to mbuf */ 718 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 719 cnt, M_MBUF, soloanfree, slp->ns_so); 720 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 721 m->m_len = cnt; 722 723 /* map pages */ 724 for (i = 0; i < npages; i++) { 725 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 726 VM_PROT_READ, 0); 727 lva += PAGE_SIZE; 728 } 729 730 pmap_update(pmap_kernel()); 731 732 mb->m_next = m; 733 mb = m; 734 error = 0; 735 uiop->uio_resid = 0; 736 } else { 737 struct iovec *iv; 738 struct iovec *iv2; 739 struct mbuf *m2; 740 int siz; 741 loan_fail: 742 /* 743 * Generate the mbuf list with the uio_iov ref. to it. 744 */ 745 i = 0; 746 m = m2 = mb; 747 while (left > 0) { 748 siz = min(M_TRAILINGSPACE(m), left); 749 if (siz > 0) { 750 left -= siz; 751 i++; 752 } 753 if (left > 0) { 754 m = m_get(M_WAIT, MT_DATA); 755 MCLAIM(m, &nfs_mowner); 756 m_clget(m, M_WAIT); 757 m->m_len = 0; 758 m2->m_next = m; 759 m2 = m; 760 } 761 } 762 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 763 uiop->uio_iov = iv2 = iv; 764 m = mb; 765 left = cnt; 766 i = 0; 767 while (left > 0) { 768 if (m == NULL) 769 panic("nfsrv_read iov"); 770 siz = min(M_TRAILINGSPACE(m), left); 771 if (siz > 0) { 772 iv->iov_base = mtod(m, char *) + 773 m->m_len; 774 iv->iov_len = siz; 775 m->m_len += siz; 776 left -= siz; 777 iv++; 778 i++; 779 } 780 m = m->m_next; 781 } 782 uiop->uio_iovcnt = i; 783 uiop->uio_offset = off; 784 uiop->uio_resid = cnt; 785 uiop->uio_rw = UIO_READ; 786 UIO_SETUP_SYSSPACE(uiop); 787 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 788 free((void *)iv2, M_TEMP); 789 } 790 read_error: 791 if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){ 792 if (!error) 793 error = getret; 794 m_freem(mreq); 795 vput(vp); 796 nfsm_reply(NFSX_POSTOPATTR(v3)); 797 nfsm_srvpostop_attr(getret, &va); 798 return (0); 799 } 800 } else { 801 uiop->uio_resid = 0; 802 } 803 vput(vp); 804 nfsm_srvfillattr(&va, fp); 805 len -= uiop->uio_resid; 806 padlen = nfsm_padlen(len); 807 if (uiop->uio_resid || padlen) 808 nfs_zeropad(mb, uiop->uio_resid, padlen); 809 if (v3) { 810 /* count */ 811 *tl++ = txdr_unsigned(len); 812 /* eof */ 813 if (off + len >= va.va_size) 814 *tl++ = nfs_true; 815 else 816 *tl++ = nfs_false; 817 } 818 *tl = txdr_unsigned(len); 819 nfsm_srvdone; 820 } 821 822 /* 823 * nfs write service 824 */ 825 int 826 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 827 { 828 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 829 struct mbuf *nam = nfsd->nd_nam; 830 char *dpos = nfsd->nd_dpos; 831 kauth_cred_t cred = nfsd->nd_cr; 832 struct iovec *ivp; 833 int i, cnt; 834 struct mbuf *mp; 835 struct nfs_fattr *fp; 836 struct iovec *iv; 837 struct vattr va, forat; 838 u_int32_t *tl; 839 int32_t t1; 840 char *bpos; 841 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 842 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 843 int stable = NFSV3WRITE_FILESYNC; 844 int v3 = (nfsd->nd_flag & ND_NFSV3); 845 char *cp2; 846 struct mbuf *mb, *mreq; 847 struct vnode *vp; 848 nfsrvfh_t nsfh; 849 struct uio io, *uiop = &io; 850 off_t off; 851 u_quad_t frev; 852 853 if (mrep == NULL) { 854 *mrq = NULL; 855 return (0); 856 } 857 nfsm_srvmtofh(&nsfh); 858 if (v3) { 859 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 860 off = fxdr_hyper(tl); 861 tl += 3; 862 stable = fxdr_unsigned(int, *tl++); 863 } else { 864 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 865 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 866 tl += 2; 867 } 868 retlen = len = fxdr_unsigned(int32_t, *tl); 869 cnt = i = 0; 870 871 /* 872 * For NFS Version 2, it is not obvious what a write of zero length 873 * should do, but I might as well be consistent with Version 3, 874 * which is to return ok so long as there are no permission problems. 875 */ 876 if (len > 0) { 877 zeroing = 1; 878 mp = mrep; 879 while (mp) { 880 if (mp == md) { 881 zeroing = 0; 882 adjust = dpos - mtod(mp, char *); 883 mp->m_len -= adjust; 884 if (mp->m_len > 0 && adjust > 0) 885 NFSMADV(mp, adjust); 886 } 887 if (zeroing) 888 mp->m_len = 0; 889 else if (mp->m_len > 0) { 890 i += mp->m_len; 891 if (i > len) { 892 mp->m_len -= (i - len); 893 zeroing = 1; 894 } 895 if (mp->m_len > 0) 896 cnt++; 897 } 898 mp = mp->m_next; 899 } 900 } 901 if (len > NFS_MAXDATA || len < 0 || i < len) { 902 error = EIO; 903 nfsm_reply(2 * NFSX_UNSIGNED); 904 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 905 return (0); 906 } 907 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 908 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 909 if (error) { 910 nfsm_reply(2 * NFSX_UNSIGNED); 911 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 912 return (0); 913 } 914 if (v3) 915 forat_ret = VOP_GETATTR(vp, &forat, cred); 916 if (vp->v_type != VREG) { 917 if (v3) 918 error = EINVAL; 919 else 920 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 921 } 922 if (!error) { 923 nqsrv_getl(vp, ND_WRITE); 924 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 925 } 926 if (error) { 927 vput(vp); 928 nfsm_reply(NFSX_WCCDATA(v3)); 929 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 930 return (0); 931 } 932 933 if (len > 0) { 934 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 935 uiop->uio_iov = iv = ivp; 936 uiop->uio_iovcnt = cnt; 937 mp = mrep; 938 while (mp) { 939 if (mp->m_len > 0) { 940 ivp->iov_base = mtod(mp, void *); 941 ivp->iov_len = mp->m_len; 942 ivp++; 943 } 944 mp = mp->m_next; 945 } 946 947 /* 948 * XXX 949 * The IO_METASYNC flag indicates that all metadata (and not 950 * just enough to ensure data integrity) must be written to 951 * stable storage synchronously. 952 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 953 */ 954 if (stable == NFSV3WRITE_UNSTABLE) 955 ioflags = IO_NODELOCKED; 956 else if (stable == NFSV3WRITE_DATASYNC) 957 ioflags = (IO_SYNC | IO_NODELOCKED); 958 else 959 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 960 uiop->uio_resid = len; 961 uiop->uio_rw = UIO_WRITE; 962 uiop->uio_offset = off; 963 UIO_SETUP_SYSSPACE(uiop); 964 error = VOP_WRITE(vp, uiop, ioflags, cred); 965 nfsstats.srvvop_writes++; 966 free(iv, M_TEMP); 967 } 968 aftat_ret = VOP_GETATTR(vp, &va, cred); 969 vput(vp); 970 if (!error) 971 error = aftat_ret; 972 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 973 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 974 if (v3) { 975 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 976 if (error) 977 return (0); 978 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 979 *tl++ = txdr_unsigned(retlen); 980 if (stable == NFSV3WRITE_UNSTABLE) 981 *tl++ = txdr_unsigned(stable); 982 else 983 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 984 /* 985 * Actually, there is no need to txdr these fields, 986 * but it may make the values more human readable, 987 * for debugging purposes. 988 */ 989 *tl++ = txdr_unsigned(boottime.tv_sec); 990 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 991 } else { 992 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 993 nfsm_srvfillattr(&va, fp); 994 } 995 nfsm_srvdone; 996 } 997 998 /* 999 * XXX elad: the original NFSW_SAMECRED() macro also made sure the 1000 * two nd_flag fields of the descriptors contained 1001 * ND_KERBAUTH. 1002 */ 1003 static int 1004 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2) 1005 { 1006 int i, do_ngroups; 1007 1008 if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2)) 1009 return (0); 1010 if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2)) 1011 return (0); 1012 do_ngroups = kauth_cred_ngroups(cred1); 1013 for (i = 0; i < do_ngroups; i++) 1014 if (kauth_cred_group(cred1, i) != 1015 kauth_cred_group(cred2, i)) 1016 return (0); 1017 1018 return (1); 1019 } 1020 1021 static struct nfsrvw_delayhash * 1022 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh) 1023 { 1024 uint32_t hash; 1025 1026 hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh), 1027 HASH32_BUF_INIT); 1028 return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ]; 1029 } 1030 1031 /* 1032 * NFS write service with write gathering support. Called when 1033 * nfsrvw_procrastinate > 0. 1034 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 1035 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 1036 * Jan. 1994. 1037 */ 1038 int 1039 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1040 { 1041 struct timeval now; 1042 struct iovec *ivp; 1043 struct mbuf *mp; 1044 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1045 struct nfs_fattr *fp; 1046 int i = 0; 1047 struct iovec *iov; 1048 struct nfsrvw_delayhash *wpp; 1049 kauth_cred_t cred; 1050 struct vattr va, forat; 1051 u_int32_t *tl; 1052 int32_t t1; 1053 char *bpos, *dpos; 1054 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1055 int ioflags, aftat_ret = 1, adjust, v3, zeroing; 1056 char *cp2; 1057 struct mbuf *mb, *mreq, *mrep, *md; 1058 struct vnode *vp; 1059 struct uio io, *uiop = &io; 1060 u_quad_t frev, cur_usec; 1061 1062 *mrq = NULL; 1063 if (*ndp) { 1064 nfsd = *ndp; 1065 *ndp = NULL; 1066 mrep = nfsd->nd_mrep; 1067 md = nfsd->nd_md; 1068 dpos = nfsd->nd_dpos; 1069 cred = nfsd->nd_cr; 1070 v3 = (nfsd->nd_flag & ND_NFSV3); 1071 LIST_INIT(&nfsd->nd_coalesce); 1072 nfsd->nd_mreq = NULL; 1073 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1074 getmicrotime(&now); 1075 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1076 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1077 1078 /* 1079 * Now, get the write header.. 1080 */ 1081 nfsm_srvmtofh(&nfsd->nd_fh); 1082 if (v3) { 1083 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1084 nfsd->nd_off = fxdr_hyper(tl); 1085 tl += 3; 1086 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1087 } else { 1088 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1089 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1090 tl += 2; 1091 } 1092 len = fxdr_unsigned(int32_t, *tl); 1093 nfsd->nd_len = len; 1094 nfsd->nd_eoff = nfsd->nd_off + len; 1095 1096 /* 1097 * Trim the header out of the mbuf list and trim off any trailing 1098 * junk so that the mbuf list has only the write data. 1099 */ 1100 zeroing = 1; 1101 i = 0; 1102 mp = mrep; 1103 while (mp) { 1104 if (mp == md) { 1105 zeroing = 0; 1106 adjust = dpos - mtod(mp, char *); 1107 mp->m_len -= adjust; 1108 if (mp->m_len > 0 && adjust > 0) 1109 NFSMADV(mp, adjust); 1110 } 1111 if (zeroing) 1112 mp->m_len = 0; 1113 else { 1114 i += mp->m_len; 1115 if (i > len) { 1116 mp->m_len -= (i - len); 1117 zeroing = 1; 1118 } 1119 } 1120 mp = mp->m_next; 1121 } 1122 if (len > NFS_MAXDATA || len < 0 || i < len) { 1123 nfsmout: 1124 m_freem(mrep); 1125 error = EIO; 1126 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1127 if (v3) 1128 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1129 nfsd->nd_mreq = mreq; 1130 nfsd->nd_mrep = NULL; 1131 nfsd->nd_time = 0; 1132 } 1133 1134 /* 1135 * Add this entry to the hash and time queues. 1136 */ 1137 owp = NULL; 1138 mutex_enter(&nfsd_lock); 1139 wp = LIST_FIRST(&slp->ns_tq); 1140 while (wp && wp->nd_time < nfsd->nd_time) { 1141 owp = wp; 1142 wp = LIST_NEXT(wp, nd_tq); 1143 } 1144 if (owp) { 1145 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1146 } else { 1147 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1148 } 1149 if (nfsd->nd_mrep) { 1150 wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh); 1151 owp = NULL; 1152 wp = LIST_FIRST(wpp); 1153 while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1154 owp = wp; 1155 wp = LIST_NEXT(wp, nd_hash); 1156 } 1157 while (wp && wp->nd_off < nfsd->nd_off && 1158 !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1159 owp = wp; 1160 wp = LIST_NEXT(wp, nd_hash); 1161 } 1162 if (owp) { 1163 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1164 1165 /* 1166 * Search the hash list for overlapping entries and 1167 * coalesce. 1168 */ 1169 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1170 wp = LIST_NEXT(nfsd, nd_hash); 1171 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr)) 1172 nfsrvw_coalesce(owp, nfsd); 1173 } 1174 } else { 1175 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1176 } 1177 } 1178 mutex_exit(&nfsd_lock); 1179 } 1180 1181 /* 1182 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1183 * and generate the associated reply mbuf list(s). 1184 */ 1185 loop1: 1186 getmicrotime(&now); 1187 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1188 mutex_enter(&nfsd_lock); 1189 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1190 owp = LIST_NEXT(nfsd, nd_tq); 1191 if (nfsd->nd_time > cur_usec) 1192 break; 1193 if (nfsd->nd_mreq) 1194 continue; 1195 LIST_REMOVE(nfsd, nd_tq); 1196 LIST_REMOVE(nfsd, nd_hash); 1197 mutex_exit(&nfsd_lock); 1198 1199 mrep = nfsd->nd_mrep; 1200 nfsd->nd_mrep = NULL; 1201 cred = nfsd->nd_cr; 1202 v3 = (nfsd->nd_flag & ND_NFSV3); 1203 forat_ret = aftat_ret = 1; 1204 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1205 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1206 false); 1207 if (!error) { 1208 if (v3) 1209 forat_ret = VOP_GETATTR(vp, &forat, cred); 1210 if (vp->v_type != VREG) { 1211 if (v3) 1212 error = EINVAL; 1213 else 1214 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1215 } 1216 } else 1217 vp = NULL; 1218 if (!error) { 1219 nqsrv_getl(vp, ND_WRITE); 1220 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1221 } 1222 1223 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1224 ioflags = IO_NODELOCKED; 1225 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1226 ioflags = (IO_SYNC | IO_NODELOCKED); 1227 else 1228 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1229 uiop->uio_rw = UIO_WRITE; 1230 uiop->uio_offset = nfsd->nd_off; 1231 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1232 UIO_SETUP_SYSSPACE(uiop); 1233 if (uiop->uio_resid > 0) { 1234 mp = mrep; 1235 i = 0; 1236 while (mp) { 1237 if (mp->m_len > 0) 1238 i++; 1239 mp = mp->m_next; 1240 } 1241 uiop->uio_iovcnt = i; 1242 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1243 uiop->uio_iov = ivp = iov; 1244 mp = mrep; 1245 while (mp) { 1246 if (mp->m_len > 0) { 1247 ivp->iov_base = mtod(mp, void *); 1248 ivp->iov_len = mp->m_len; 1249 ivp++; 1250 } 1251 mp = mp->m_next; 1252 } 1253 if (!error) { 1254 error = VOP_WRITE(vp, uiop, ioflags, cred); 1255 nfsstats.srvvop_writes++; 1256 } 1257 free((void *)iov, M_TEMP); 1258 } 1259 m_freem(mrep); 1260 if (vp) { 1261 aftat_ret = VOP_GETATTR(vp, &va, cred); 1262 vput(vp); 1263 } 1264 1265 /* 1266 * Loop around generating replies for all write rpcs that have 1267 * now been completed. 1268 */ 1269 swp = nfsd; 1270 do { 1271 if (error) { 1272 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1273 if (v3) { 1274 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1275 } 1276 } else { 1277 nfsm_writereply(NFSX_PREOPATTR(v3) + 1278 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1279 NFSX_WRITEVERF(v3), v3); 1280 if (v3) { 1281 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1282 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1283 *tl++ = txdr_unsigned(nfsd->nd_len); 1284 *tl++ = txdr_unsigned(swp->nd_stable); 1285 /* 1286 * Actually, there is no need to txdr these fields, 1287 * but it may make the values more human readable, 1288 * for debugging purposes. 1289 */ 1290 *tl++ = txdr_unsigned(boottime.tv_sec); 1291 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 1292 } else { 1293 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1294 nfsm_srvfillattr(&va, fp); 1295 } 1296 } 1297 nfsd->nd_mreq = mreq; 1298 if (nfsd->nd_mrep) 1299 panic("nfsrv_write: nd_mrep not free"); 1300 1301 /* 1302 * Done. Put it at the head of the timer queue so that 1303 * the final phase can return the reply. 1304 */ 1305 mutex_enter(&nfsd_lock); 1306 if (nfsd != swp) { 1307 nfsd->nd_time = 0; 1308 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1309 } 1310 nfsd = LIST_FIRST(&swp->nd_coalesce); 1311 if (nfsd) { 1312 LIST_REMOVE(nfsd, nd_tq); 1313 } 1314 mutex_exit(&nfsd_lock); 1315 } while (nfsd); 1316 swp->nd_time = 0; 1317 1318 mutex_enter(&nfsd_lock); 1319 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1320 mutex_exit(&nfsd_lock); 1321 goto loop1; 1322 } 1323 mutex_exit(&nfsd_lock); 1324 nfs_timer_start(); 1325 1326 /* 1327 * Search for a reply to return. 1328 */ 1329 mutex_enter(&nfsd_lock); 1330 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1331 if (nfsd->nd_mreq) { 1332 LIST_REMOVE(nfsd, nd_tq); 1333 *mrq = nfsd->nd_mreq; 1334 *ndp = nfsd; 1335 break; 1336 } 1337 } 1338 mutex_exit(&nfsd_lock); 1339 return (0); 1340 } 1341 1342 /* 1343 * Coalesce the write request nfsd into owp. To do this we must: 1344 * - remove nfsd from the queues 1345 * - merge nfsd->nd_mrep into owp->nd_mrep 1346 * - update the nd_eoff and nd_stable for owp 1347 * - put nfsd on owp's nd_coalesce list 1348 * NB: Must be called at splsoftclock(). 1349 */ 1350 void 1351 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd) 1352 { 1353 int overlap; 1354 struct mbuf *mp; 1355 struct nfsrv_descript *m; 1356 1357 KASSERT(mutex_owned(&nfsd_lock)); 1358 1359 LIST_REMOVE(nfsd, nd_hash); 1360 LIST_REMOVE(nfsd, nd_tq); 1361 if (owp->nd_eoff < nfsd->nd_eoff) { 1362 overlap = owp->nd_eoff - nfsd->nd_off; 1363 if (overlap < 0) 1364 panic("nfsrv_coalesce: bad off"); 1365 if (overlap > 0) 1366 m_adj(nfsd->nd_mrep, overlap); 1367 mp = owp->nd_mrep; 1368 while (mp->m_next) 1369 mp = mp->m_next; 1370 mp->m_next = nfsd->nd_mrep; 1371 owp->nd_eoff = nfsd->nd_eoff; 1372 } else 1373 m_freem(nfsd->nd_mrep); 1374 nfsd->nd_mrep = NULL; 1375 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1376 owp->nd_stable = NFSV3WRITE_FILESYNC; 1377 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1378 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1379 owp->nd_stable = NFSV3WRITE_DATASYNC; 1380 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1381 /* 1382 * nfsd might hold coalesce elements! Move them to owp. 1383 * Otherwise, requests may be lost and clients will be stuck. 1384 */ 1385 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1386 LIST_REMOVE(m, nd_tq); 1387 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1388 } 1389 } 1390 1391 /* 1392 * nfs create service 1393 * now does a truncate to 0 length via. setattr if it already exists 1394 */ 1395 int 1396 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1397 { 1398 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1399 struct mbuf *nam = nfsd->nd_nam; 1400 char *dpos = nfsd->nd_dpos; 1401 kauth_cred_t cred = nfsd->nd_cr; 1402 struct nfs_fattr *fp; 1403 struct vattr va, dirfor, diraft; 1404 struct nfsv2_sattr *sp; 1405 u_int32_t *tl; 1406 struct nameidata nd; 1407 char *cp; 1408 int32_t t1; 1409 char *bpos; 1410 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1411 int rdev = 0, abort = 0; 1412 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1413 char *cp2; 1414 struct mbuf *mb, *mreq; 1415 struct vnode *vp = NULL, *dirp = NULL; 1416 nfsrvfh_t nsfh; 1417 u_quad_t frev, tempsize; 1418 u_char cverf[NFSX_V3CREATEVERF]; 1419 1420 nd.ni_cnd.cn_nameiop = 0; 1421 nfsm_srvmtofh(&nsfh); 1422 nfsm_srvnamesiz(len); 1423 nd.ni_cnd.cn_cred = cred; 1424 nd.ni_cnd.cn_nameiop = CREATE; 1425 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1426 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1427 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1428 if (dirp && v3) { 1429 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1430 } 1431 if (error) { 1432 nfsm_reply(NFSX_WCCDATA(v3)); 1433 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1434 if (dirp) 1435 vrele(dirp); 1436 if (nd.ni_pathbuf != NULL) { 1437 pathbuf_destroy(nd.ni_pathbuf); 1438 nd.ni_pathbuf = NULL; 1439 } 1440 return (0); 1441 } 1442 abort = 1; 1443 vattr_null(&va); 1444 if (v3) { 1445 va.va_mode = 0; 1446 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1447 how = fxdr_unsigned(int, *tl); 1448 switch (how) { 1449 case NFSV3CREATE_GUARDED: 1450 if (nd.ni_vp) { 1451 error = EEXIST; 1452 break; 1453 } 1454 case NFSV3CREATE_UNCHECKED: 1455 nfsm_srvsattr(&va); 1456 break; 1457 case NFSV3CREATE_EXCLUSIVE: 1458 nfsm_dissect(cp, void *, NFSX_V3CREATEVERF); 1459 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1460 exclusive_flag = 1; 1461 break; 1462 }; 1463 va.va_type = VREG; 1464 } else { 1465 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1466 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1467 if (va.va_type == VNON) 1468 va.va_type = VREG; 1469 va.va_mode = nfstov_mode(sp->sa_mode); 1470 switch (va.va_type) { 1471 case VREG: 1472 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1473 if (tsize != -1) 1474 va.va_size = (u_quad_t)tsize; 1475 break; 1476 case VCHR: 1477 case VBLK: 1478 case VFIFO: 1479 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1480 break; 1481 default: 1482 break; 1483 }; 1484 } 1485 1486 /* 1487 * Iff doesn't exist, create it 1488 * otherwise just truncate to 0 length 1489 * should I set the mode too ?? 1490 */ 1491 if (nd.ni_vp == NULL) { 1492 if (va.va_type == VREG || va.va_type == VSOCK) { 1493 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1494 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1495 if (!error) { 1496 if (exclusive_flag) { 1497 exclusive_flag = 0; 1498 vattr_null(&va); 1499 /* 1500 * XXX 1501 * assuming NFSX_V3CREATEVERF 1502 * == sizeof(nfstime3) 1503 */ 1504 fxdr_nfsv3time(cverf, &va.va_atime); 1505 error = VOP_SETATTR(nd.ni_vp, &va, 1506 cred); 1507 } 1508 } 1509 } else if (va.va_type == VCHR || va.va_type == VBLK || 1510 va.va_type == VFIFO) { 1511 if (va.va_type == VCHR && rdev == 0xffffffff) 1512 va.va_type = VFIFO; 1513 if (va.va_type != VFIFO && 1514 (error = kauth_authorize_system(cred, 1515 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1516 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1517 vput(nd.ni_dvp); 1518 abort = 0; 1519 nfsm_reply(0); 1520 if (nd.ni_pathbuf != NULL) { 1521 pathbuf_destroy(nd.ni_pathbuf); 1522 nd.ni_pathbuf = NULL; 1523 } 1524 return (error); 1525 } else 1526 va.va_rdev = (dev_t)rdev; 1527 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1528 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1529 &va); 1530 if (error) { 1531 nfsm_reply(0); 1532 } 1533 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1534 vput(nd.ni_vp); 1535 vrele(nd.ni_dvp); 1536 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1537 if (nd.ni_pathbuf != NULL) { 1538 pathbuf_destroy(nd.ni_pathbuf); 1539 nd.ni_pathbuf = NULL; 1540 } 1541 error = EINVAL; 1542 abort = 0; 1543 nfsm_reply(0); 1544 } 1545 } else { 1546 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1547 if (nd.ni_pathbuf != NULL) { 1548 pathbuf_destroy(nd.ni_pathbuf); 1549 nd.ni_pathbuf = NULL; 1550 } 1551 vput(nd.ni_dvp); 1552 error = ENXIO; 1553 abort = 0; 1554 } 1555 vp = nd.ni_vp; 1556 } else { 1557 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1558 if (nd.ni_pathbuf != NULL) { 1559 pathbuf_destroy(nd.ni_pathbuf); 1560 nd.ni_pathbuf = NULL; 1561 } 1562 vp = nd.ni_vp; 1563 if (nd.ni_dvp == vp) 1564 vrele(nd.ni_dvp); 1565 else 1566 vput(nd.ni_dvp); 1567 abort = 0; 1568 if (!error && va.va_size != -1) { 1569 error = nfsrv_access(vp, VWRITE, cred, 1570 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1571 if (!error) { 1572 nqsrv_getl(vp, ND_WRITE); 1573 tempsize = va.va_size; 1574 vattr_null(&va); 1575 va.va_size = tempsize; 1576 error = VOP_SETATTR(vp, &va, cred); 1577 } 1578 } 1579 if (error) 1580 vput(vp); 1581 } 1582 if (!error) { 1583 error = nfsrv_composefh(vp, &nsfh, v3); 1584 if (!error) 1585 error = VOP_GETATTR(vp, &va, cred); 1586 vput(vp); 1587 } 1588 if (v3) { 1589 if (exclusive_flag && !error) { 1590 /* 1591 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1592 */ 1593 char oldverf[NFSX_V3CREATEVERF]; 1594 1595 txdr_nfsv3time(&va.va_atime, oldverf); 1596 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1597 error = EEXIST; 1598 } 1599 if (dirp) { 1600 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1601 } 1602 } 1603 if (dirp) { 1604 vrele(dirp); 1605 dirp = NULL; 1606 } 1607 if (nd.ni_pathbuf != NULL) { 1608 pathbuf_destroy(nd.ni_pathbuf); 1609 nd.ni_pathbuf = NULL; 1610 } 1611 abort = 0; 1612 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1613 if (v3) { 1614 if (!error) { 1615 nfsm_srvpostop_fh(&nsfh); 1616 nfsm_srvpostop_attr(0, &va); 1617 } 1618 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1619 } else { 1620 nfsm_srvfhtom(&nsfh, v3); 1621 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1622 nfsm_srvfillattr(&va, fp); 1623 } 1624 return (0); 1625 nfsmout: 1626 if (dirp) 1627 vrele(dirp); 1628 if (abort) { 1629 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1630 if (nd.ni_dvp == nd.ni_vp) 1631 vrele(nd.ni_dvp); 1632 else 1633 vput(nd.ni_dvp); 1634 if (nd.ni_vp) 1635 vput(nd.ni_vp); 1636 } 1637 if (nd.ni_pathbuf != NULL) { 1638 pathbuf_destroy(nd.ni_pathbuf); 1639 nd.ni_pathbuf = NULL; 1640 } 1641 return (error); 1642 } 1643 1644 /* 1645 * nfs v3 mknod service 1646 */ 1647 int 1648 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1649 { 1650 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1651 struct mbuf *nam = nfsd->nd_nam; 1652 char *dpos = nfsd->nd_dpos; 1653 kauth_cred_t cred = nfsd->nd_cr; 1654 struct vattr va, dirfor, diraft; 1655 u_int32_t *tl; 1656 struct nameidata nd; 1657 int32_t t1; 1658 char *bpos; 1659 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1660 int abort = 0; 1661 u_int32_t major, minor; 1662 enum vtype vtyp; 1663 char *cp2; 1664 struct mbuf *mb, *mreq; 1665 struct vnode *vp, *dirp = (struct vnode *)0; 1666 nfsrvfh_t nsfh; 1667 u_quad_t frev; 1668 1669 nd.ni_cnd.cn_nameiop = 0; 1670 nfsm_srvmtofh(&nsfh); 1671 nfsm_srvnamesiz(len); 1672 nd.ni_cnd.cn_cred = cred; 1673 nd.ni_cnd.cn_nameiop = CREATE; 1674 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1675 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1676 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1677 if (dirp) 1678 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1679 if (error) { 1680 nfsm_reply(NFSX_WCCDATA(1)); 1681 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1682 if (dirp) 1683 vrele(dirp); 1684 if (nd.ni_pathbuf != NULL) { 1685 pathbuf_destroy(nd.ni_pathbuf); 1686 nd.ni_pathbuf = NULL; 1687 } 1688 return (0); 1689 } 1690 abort = 1; 1691 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1692 vtyp = nfsv3tov_type(*tl); 1693 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1694 error = NFSERR_BADTYPE; 1695 goto abort; 1696 } 1697 vattr_null(&va); 1698 va.va_mode = 0; 1699 nfsm_srvsattr(&va); 1700 if (vtyp == VCHR || vtyp == VBLK) { 1701 dev_t rdev; 1702 1703 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1704 major = fxdr_unsigned(u_int32_t, *tl++); 1705 minor = fxdr_unsigned(u_int32_t, *tl); 1706 rdev = makedev(major, minor); 1707 if (major(rdev) != major || minor(rdev) != minor) { 1708 error = EINVAL; 1709 goto abort; 1710 } 1711 va.va_rdev = rdev; 1712 } 1713 1714 /* 1715 * Iff doesn't exist, create it. 1716 */ 1717 if (nd.ni_vp) { 1718 error = EEXIST; 1719 abort: 1720 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1721 if (nd.ni_dvp == nd.ni_vp) 1722 vrele(nd.ni_dvp); 1723 else 1724 vput(nd.ni_dvp); 1725 if (nd.ni_vp) 1726 vput(nd.ni_vp); 1727 if (nd.ni_pathbuf != NULL) { 1728 pathbuf_destroy(nd.ni_pathbuf); 1729 nd.ni_pathbuf = NULL; 1730 } 1731 goto out; 1732 } 1733 va.va_type = vtyp; 1734 if (vtyp == VSOCK) { 1735 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1736 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1737 } else { 1738 if (va.va_type != VFIFO && 1739 (error = kauth_authorize_system(cred, 1740 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1741 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1742 vput(nd.ni_dvp); 1743 goto out; 1744 } 1745 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1746 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1747 if (error) 1748 goto out; 1749 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1750 vput(nd.ni_vp); 1751 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1752 error = EINVAL; 1753 } 1754 } 1755 out: 1756 vp = nd.ni_vp; 1757 if (!error) { 1758 error = nfsrv_composefh(vp, &nsfh, true); 1759 if (!error) 1760 error = VOP_GETATTR(vp, &va, cred); 1761 vput(vp); 1762 } 1763 if (dirp) { 1764 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1765 vrele(dirp); 1766 dirp = NULL; 1767 } 1768 if (nd.ni_pathbuf != NULL) { 1769 pathbuf_destroy(nd.ni_pathbuf); 1770 nd.ni_pathbuf = NULL; 1771 } 1772 abort = 0; 1773 nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) + 1774 NFSX_WCCDATA(1)); 1775 if (!error) { 1776 nfsm_srvpostop_fh(&nsfh); 1777 nfsm_srvpostop_attr(0, &va); 1778 } 1779 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1780 return (0); 1781 nfsmout: 1782 if (abort) { 1783 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1784 if (nd.ni_dvp == nd.ni_vp) 1785 vrele(nd.ni_dvp); 1786 else 1787 vput(nd.ni_dvp); 1788 if (nd.ni_vp) 1789 vput(nd.ni_vp); 1790 } 1791 if (nd.ni_pathbuf != NULL) { 1792 pathbuf_destroy(nd.ni_pathbuf); 1793 nd.ni_pathbuf = NULL; 1794 } 1795 if (dirp) 1796 vrele(dirp); 1797 return (error); 1798 } 1799 1800 /* 1801 * nfs remove service 1802 */ 1803 int 1804 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1805 { 1806 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1807 struct mbuf *nam = nfsd->nd_nam; 1808 char *dpos = nfsd->nd_dpos; 1809 kauth_cred_t cred = nfsd->nd_cr; 1810 struct nameidata nd; 1811 u_int32_t *tl; 1812 int32_t t1; 1813 char *bpos; 1814 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1815 int v3 = (nfsd->nd_flag & ND_NFSV3); 1816 char *cp2; 1817 struct mbuf *mb, *mreq; 1818 struct vnode *vp, *dirp; 1819 struct vattr dirfor, diraft; 1820 nfsrvfh_t nsfh; 1821 u_quad_t frev; 1822 1823 #ifndef nolint 1824 vp = (struct vnode *)0; 1825 #endif 1826 nfsm_srvmtofh(&nsfh); 1827 nfsm_srvnamesiz(len); 1828 nd.ni_cnd.cn_cred = cred; 1829 nd.ni_cnd.cn_nameiop = DELETE; 1830 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1831 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1832 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1833 if (dirp && v3) { 1834 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1835 } 1836 if (!error) { 1837 vp = nd.ni_vp; 1838 if (vp->v_type == VDIR) { 1839 error = EPERM; 1840 goto out; 1841 } 1842 /* 1843 * The root of a mounted filesystem cannot be deleted. 1844 */ 1845 if (vp->v_vflag & VV_ROOT) { 1846 error = EBUSY; 1847 } 1848 out: 1849 if (!error) { 1850 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1851 nqsrv_getl(vp, ND_WRITE); 1852 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1853 } else { 1854 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1855 if (nd.ni_dvp == vp) 1856 vrele(nd.ni_dvp); 1857 else 1858 vput(nd.ni_dvp); 1859 vput(vp); 1860 } 1861 } 1862 if (nd.ni_pathbuf != NULL) { 1863 pathbuf_destroy(nd.ni_pathbuf); 1864 nd.ni_pathbuf = NULL; 1865 } 1866 if (dirp) { 1867 if (v3) { 1868 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1869 } 1870 vrele(dirp); 1871 } 1872 nfsm_reply(NFSX_WCCDATA(v3)); 1873 if (v3) { 1874 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1875 return (0); 1876 } 1877 nfsm_srvdone; 1878 } 1879 1880 /* 1881 * nfs rename service 1882 */ 1883 int 1884 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1885 { 1886 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1887 struct mbuf *nam = nfsd->nd_nam; 1888 char *dpos = nfsd->nd_dpos; 1889 kauth_cred_t cred = nfsd->nd_cr; 1890 u_int32_t *tl; 1891 int32_t t1; 1892 char *bpos; 1893 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1894 uint32_t len, len2; 1895 int tdirfor_ret = 1, tdiraft_ret = 1; 1896 int v3 = (nfsd->nd_flag & ND_NFSV3); 1897 char *cp2; 1898 struct mbuf *mb, *mreq; 1899 struct nameidata fromnd, tond; 1900 struct vnode *fvp, *tvp, *tdvp; 1901 struct vnode *fdirp = NULL, *tdirp = NULL; 1902 struct mount *localfs = NULL; 1903 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1904 nfsrvfh_t fnsfh, tnsfh; 1905 u_quad_t frev; 1906 uid_t saved_uid; 1907 1908 #ifndef nolint 1909 fvp = (struct vnode *)0; 1910 #endif 1911 fromnd.ni_cnd.cn_nameiop = 0; 1912 tond.ni_cnd.cn_nameiop = 0; 1913 nfsm_srvmtofh(&fnsfh); 1914 nfsm_srvnamesiz(len); 1915 /* 1916 * Remember our original uid so that we can reset cr_uid before 1917 * the second nfs_namei() call, in case it is remapped. 1918 */ 1919 saved_uid = kauth_cred_geteuid(cred); 1920 fromnd.ni_cnd.cn_cred = cred; 1921 fromnd.ni_cnd.cn_nameiop = DELETE; 1922 fromnd.ni_cnd.cn_flags = LOCKPARENT | INRENAME; 1923 error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md, 1924 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1925 if (fdirp && v3) { 1926 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred); 1927 } 1928 if (error) { 1929 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1930 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1931 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1932 if (fdirp) 1933 vrele(fdirp); 1934 pathbuf_destroy(fromnd.ni_pathbuf); 1935 return (0); 1936 } 1937 if (fromnd.ni_dvp != fromnd.ni_vp) { 1938 VOP_UNLOCK(fromnd.ni_dvp); 1939 } 1940 fvp = fromnd.ni_vp; 1941 1942 localfs = fvp->v_mount; 1943 error = VFS_RENAMELOCK_ENTER(localfs); 1944 if (error) { 1945 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1946 vrele(fromnd.ni_dvp); 1947 vrele(fvp); 1948 goto out1; 1949 } 1950 1951 /* Copied, regrettably, from vfs_syscalls.c (q.v.) */ 1952 vrele(fvp); 1953 if ((fromnd.ni_cnd.cn_namelen == 1 && 1954 fromnd.ni_cnd.cn_nameptr[0] == '.') || 1955 (fromnd.ni_cnd.cn_namelen == 2 && 1956 fromnd.ni_cnd.cn_nameptr[0] == '.' && 1957 fromnd.ni_cnd.cn_nameptr[1] == '.')) { 1958 error = EINVAL; 1959 VFS_RENAMELOCK_EXIT(localfs); 1960 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1961 vrele(fromnd.ni_dvp); 1962 goto out1; 1963 } 1964 vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY); 1965 error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0); 1966 if (error) { 1967 VOP_UNLOCK(fromnd.ni_dvp); 1968 VFS_RENAMELOCK_EXIT(localfs); 1969 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1970 vrele(fromnd.ni_dvp); 1971 goto out1; 1972 } 1973 VOP_UNLOCK(fromnd.ni_vp); 1974 if (fromnd.ni_dvp != fromnd.ni_vp) 1975 VOP_UNLOCK(fromnd.ni_dvp); 1976 fvp = fromnd.ni_vp; 1977 1978 nfsm_srvmtofh(&tnsfh); 1979 if (v3) { 1980 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1981 len2 = fxdr_unsigned(uint32_t, *tl); 1982 /* len2 will be checked by nfs_namei */ 1983 } 1984 else { 1985 /* NFSv2 */ 1986 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1987 } 1988 kauth_cred_seteuid(cred, saved_uid); 1989 tond.ni_cnd.cn_cred = cred; 1990 tond.ni_cnd.cn_nameiop = RENAME; 1991 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | INRENAME; 1992 error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md, 1993 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1994 if (tdirp && v3) { 1995 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred); 1996 } 1997 if (error) { 1998 VFS_RENAMELOCK_EXIT(localfs); 1999 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2000 vrele(fromnd.ni_dvp); 2001 vrele(fvp); 2002 goto out1; 2003 } 2004 tdvp = tond.ni_dvp; 2005 tvp = tond.ni_vp; 2006 if (tvp != NULL) { 2007 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 2008 if (v3) 2009 error = EEXIST; 2010 else 2011 error = EISDIR; 2012 goto out; 2013 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 2014 if (v3) 2015 error = EEXIST; 2016 else 2017 error = ENOTDIR; 2018 goto out; 2019 } 2020 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 2021 if (v3) 2022 error = EXDEV; 2023 else 2024 error = ENOTEMPTY; 2025 goto out; 2026 } 2027 } 2028 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 2029 if (v3) 2030 error = EXDEV; 2031 else 2032 error = ENOTEMPTY; 2033 goto out; 2034 } 2035 if (fvp->v_mount != tdvp->v_mount) { 2036 if (v3) 2037 error = EXDEV; 2038 else 2039 error = ENOTEMPTY; 2040 goto out; 2041 } 2042 if (fvp == tdvp) { 2043 if (v3) 2044 error = EINVAL; 2045 else 2046 error = ENOTEMPTY; 2047 } 2048 /* 2049 * If source is the same as the destination (that is the 2050 * same vnode with the same name in the same directory), 2051 * then there is nothing to do. 2052 */ 2053 if (fvp == tvp && fromnd.ni_dvp == tdvp && 2054 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 2055 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 2056 fromnd.ni_cnd.cn_namelen)) 2057 error = -1; 2058 out: 2059 if (!error) { 2060 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 2061 nqsrv_getl(tdvp, ND_WRITE); 2062 if (tvp) { 2063 nqsrv_getl(tvp, ND_WRITE); 2064 } 2065 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 2066 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 2067 VFS_RENAMELOCK_EXIT(localfs); 2068 } else { 2069 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 2070 if (tdvp == tvp) 2071 vrele(tdvp); 2072 else 2073 vput(tdvp); 2074 if (tvp) 2075 vput(tvp); 2076 VFS_RENAMELOCK_EXIT(localfs); 2077 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2078 vrele(fromnd.ni_dvp); 2079 vrele(fvp); 2080 if (error == -1) 2081 error = 0; 2082 } 2083 pathbuf_destroy(tond.ni_pathbuf); 2084 tond.ni_cnd.cn_nameiop = 0; 2085 out1: 2086 if (fdirp) { 2087 if (v3) { 2088 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred); 2089 } 2090 vrele(fdirp); 2091 fdirp = NULL; 2092 } 2093 if (tdirp) { 2094 if (v3) { 2095 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred); 2096 } 2097 vrele(tdirp); 2098 tdirp = NULL; 2099 } 2100 pathbuf_destroy(fromnd.ni_pathbuf); 2101 fromnd.ni_cnd.cn_nameiop = 0; 2102 localfs = NULL; 2103 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2104 if (v3) { 2105 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2106 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2107 } 2108 return (0); 2109 2110 nfsmout: 2111 if (fdirp) 2112 vrele(fdirp); 2113 #ifdef notdef 2114 if (tdirp) 2115 vrele(tdirp); 2116 #endif 2117 if (tond.ni_cnd.cn_nameiop) { 2118 pathbuf_destroy(tond.ni_pathbuf); 2119 } 2120 if (localfs) { 2121 VFS_RENAMELOCK_EXIT(localfs); 2122 } 2123 if (fromnd.ni_cnd.cn_nameiop) { 2124 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2125 pathbuf_destroy(fromnd.ni_pathbuf); 2126 vrele(fromnd.ni_dvp); 2127 vrele(fvp); 2128 } 2129 return (error); 2130 } 2131 2132 /* 2133 * nfs link service 2134 */ 2135 int 2136 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2137 { 2138 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2139 struct mbuf *nam = nfsd->nd_nam; 2140 char *dpos = nfsd->nd_dpos; 2141 kauth_cred_t cred = nfsd->nd_cr; 2142 struct nameidata nd; 2143 u_int32_t *tl; 2144 int32_t t1; 2145 char *bpos; 2146 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2147 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2148 char *cp2; 2149 struct mbuf *mb, *mreq; 2150 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2151 struct vattr dirfor, diraft, at; 2152 nfsrvfh_t nsfh, dnsfh; 2153 u_quad_t frev; 2154 2155 nfsm_srvmtofh(&nsfh); 2156 nfsm_srvmtofh(&dnsfh); 2157 nfsm_srvnamesiz(len); 2158 error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam, 2159 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2160 if (error) { 2161 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2162 nfsm_srvpostop_attr(getret, &at); 2163 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2164 return (0); 2165 } 2166 if (vp->v_type == VDIR) { 2167 error = EPERM; 2168 goto out1; 2169 } 2170 nd.ni_cnd.cn_cred = cred; 2171 nd.ni_cnd.cn_nameiop = CREATE; 2172 nd.ni_cnd.cn_flags = LOCKPARENT; 2173 error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos, 2174 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2175 if (dirp && v3) { 2176 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2177 } 2178 if (error) 2179 goto out1; 2180 xp = nd.ni_vp; 2181 if (xp != NULL) { 2182 error = EEXIST; 2183 goto out; 2184 } 2185 xp = nd.ni_dvp; 2186 if (vp->v_mount != xp->v_mount) 2187 error = EXDEV; 2188 out: 2189 if (!error) { 2190 nqsrv_getl(vp, ND_WRITE); 2191 nqsrv_getl(xp, ND_WRITE); 2192 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2193 } else { 2194 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2195 if (nd.ni_dvp == nd.ni_vp) 2196 vrele(nd.ni_dvp); 2197 else 2198 vput(nd.ni_dvp); 2199 if (nd.ni_vp) 2200 vrele(nd.ni_vp); 2201 } 2202 out1: 2203 if (v3) 2204 getret = VOP_GETATTR(vp, &at, cred); 2205 if (dirp) { 2206 if (v3) { 2207 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2208 } 2209 vrele(dirp); 2210 } 2211 vrele(vp); 2212 pathbuf_destroy(nd.ni_pathbuf); 2213 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2214 if (v3) { 2215 nfsm_srvpostop_attr(getret, &at); 2216 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2217 return (0); 2218 } 2219 nfsm_srvdone; 2220 } 2221 2222 /* 2223 * nfs symbolic link service 2224 */ 2225 int 2226 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2227 { 2228 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2229 struct mbuf *nam = nfsd->nd_nam; 2230 char *dpos = nfsd->nd_dpos; 2231 kauth_cred_t cred = nfsd->nd_cr; 2232 struct vattr va, dirfor, diraft; 2233 struct nameidata nd; 2234 u_int32_t *tl; 2235 int32_t t1; 2236 struct nfsv2_sattr *sp; 2237 char *bpos, *pathcp = NULL, *cp2; 2238 struct uio io; 2239 struct iovec iv; 2240 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0; 2241 uint32_t len, len2; 2242 int v3 = (nfsd->nd_flag & ND_NFSV3); 2243 struct mbuf *mb, *mreq; 2244 struct vnode *dirp = (struct vnode *)0; 2245 nfsrvfh_t nsfh; 2246 u_quad_t frev; 2247 2248 nd.ni_cnd.cn_nameiop = 0; 2249 nfsm_srvmtofh(&nsfh); 2250 nfsm_srvnamesiz(len); 2251 nd.ni_cnd.cn_cred = cred; 2252 nd.ni_cnd.cn_nameiop = CREATE; 2253 nd.ni_cnd.cn_flags = LOCKPARENT; 2254 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2255 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2256 if (dirp && v3) { 2257 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2258 } 2259 if (error) 2260 goto out; 2261 abort = 1; 2262 vattr_null(&va); 2263 va.va_type = VLNK; 2264 if (v3) { 2265 va.va_mode = 0; 2266 nfsm_srvsattr(&va); 2267 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2268 len2 = fxdr_unsigned(uint32_t, *tl); 2269 if (len2 > PATH_MAX) { 2270 /* XXX should check _PC_NO_TRUNC */ 2271 error = ENAMETOOLONG; 2272 goto abortop; 2273 } 2274 } 2275 else { 2276 /* NFSv2 */ 2277 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2278 } 2279 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2280 iv.iov_base = pathcp; 2281 iv.iov_len = len2; 2282 io.uio_resid = len2; 2283 io.uio_offset = 0; 2284 io.uio_iov = &iv; 2285 io.uio_iovcnt = 1; 2286 io.uio_rw = UIO_READ; 2287 UIO_SETUP_SYSSPACE(&io); 2288 nfsm_mtouio(&io, len2); 2289 if (!v3) { 2290 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2291 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2292 } 2293 *(pathcp + len2) = '\0'; 2294 if (nd.ni_vp) { 2295 error = EEXIST; 2296 abortop: 2297 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2298 if (nd.ni_dvp == nd.ni_vp) 2299 vrele(nd.ni_dvp); 2300 else 2301 vput(nd.ni_dvp); 2302 if (nd.ni_vp) 2303 vrele(nd.ni_vp); 2304 goto out; 2305 } 2306 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2307 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2308 if (!error) { 2309 if (v3) { 2310 error = nfsrv_composefh(nd.ni_vp, &nsfh, v3); 2311 if (!error) 2312 error = VOP_GETATTR(nd.ni_vp, &va, cred); 2313 vput(nd.ni_vp); 2314 } else { 2315 vput(nd.ni_vp); 2316 } 2317 } 2318 out: 2319 if (pathcp) 2320 free(pathcp, M_TEMP); 2321 if (dirp) { 2322 if (v3) { 2323 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2324 } 2325 vrele(dirp); 2326 dirp = NULL; 2327 } 2328 pathbuf_destroy(nd.ni_pathbuf); 2329 abort = 0; 2330 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2331 NFSX_WCCDATA(v3)); 2332 if (v3) { 2333 if (!error) { 2334 nfsm_srvpostop_fh(&nsfh); 2335 nfsm_srvpostop_attr(0, &va); 2336 } 2337 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2338 } 2339 return (0); 2340 nfsmout: 2341 if (abort) { 2342 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2343 if (nd.ni_dvp == nd.ni_vp) 2344 vrele(nd.ni_dvp); 2345 else 2346 vput(nd.ni_dvp); 2347 if (nd.ni_vp) 2348 vrele(nd.ni_vp); 2349 pathbuf_destroy(nd.ni_pathbuf); 2350 } 2351 if (dirp) 2352 vrele(dirp); 2353 if (pathcp) 2354 free(pathcp, M_TEMP); 2355 return (error); 2356 } 2357 2358 /* 2359 * nfs mkdir service 2360 */ 2361 int 2362 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2363 { 2364 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2365 struct mbuf *nam = nfsd->nd_nam; 2366 char *dpos = nfsd->nd_dpos; 2367 kauth_cred_t cred = nfsd->nd_cr; 2368 struct vattr va, dirfor, diraft; 2369 struct nfs_fattr *fp; 2370 struct nameidata nd; 2371 char *cp; 2372 u_int32_t *tl; 2373 int32_t t1; 2374 char *bpos; 2375 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2376 int abort = 0; 2377 int v3 = (nfsd->nd_flag & ND_NFSV3); 2378 char *cp2; 2379 struct mbuf *mb, *mreq; 2380 struct vnode *vp, *dirp = (struct vnode *)0; 2381 nfsrvfh_t nsfh; 2382 u_quad_t frev; 2383 2384 nfsm_srvmtofh(&nsfh); 2385 nfsm_srvnamesiz(len); 2386 nd.ni_cnd.cn_cred = cred; 2387 nd.ni_cnd.cn_nameiop = CREATE; 2388 nd.ni_cnd.cn_flags = LOCKPARENT; 2389 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2390 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2391 if (dirp && v3) { 2392 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2393 } 2394 if (error) { 2395 if (nd.ni_pathbuf != NULL) { 2396 pathbuf_destroy(nd.ni_pathbuf); 2397 } 2398 nfsm_reply(NFSX_WCCDATA(v3)); 2399 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2400 if (dirp) 2401 vrele(dirp); 2402 return (0); 2403 } 2404 abort = 1; 2405 vattr_null(&va); 2406 if (v3) { 2407 va.va_mode = 0; 2408 nfsm_srvsattr(&va); 2409 } else { 2410 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2411 va.va_mode = nfstov_mode(*tl++); 2412 } 2413 va.va_type = VDIR; 2414 vp = nd.ni_vp; 2415 if (vp != NULL) { 2416 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2417 if (nd.ni_dvp == vp) 2418 vrele(nd.ni_dvp); 2419 else 2420 vput(nd.ni_dvp); 2421 vrele(vp); 2422 error = EEXIST; 2423 goto out; 2424 } 2425 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2426 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2427 if (!error) { 2428 vp = nd.ni_vp; 2429 error = nfsrv_composefh(vp, &nsfh, v3); 2430 if (!error) 2431 error = VOP_GETATTR(vp, &va, cred); 2432 vput(vp); 2433 } 2434 out: 2435 if (dirp) { 2436 if (v3) { 2437 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2438 } 2439 vrele(dirp); 2440 dirp = NULL; 2441 } 2442 pathbuf_destroy(nd.ni_pathbuf); 2443 abort = 0; 2444 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2445 NFSX_WCCDATA(v3)); 2446 if (v3) { 2447 if (!error) { 2448 nfsm_srvpostop_fh(&nsfh); 2449 nfsm_srvpostop_attr(0, &va); 2450 } 2451 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2452 } else { 2453 nfsm_srvfhtom(&nsfh, v3); 2454 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2455 nfsm_srvfillattr(&va, fp); 2456 } 2457 return (0); 2458 nfsmout: 2459 if (abort) { 2460 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2461 if (nd.ni_dvp == nd.ni_vp) 2462 vrele(nd.ni_dvp); 2463 else 2464 vput(nd.ni_dvp); 2465 if (nd.ni_vp) 2466 vrele(nd.ni_vp); 2467 pathbuf_destroy(nd.ni_pathbuf); 2468 } 2469 if (dirp) 2470 vrele(dirp); 2471 return (error); 2472 } 2473 2474 /* 2475 * nfs rmdir service 2476 */ 2477 int 2478 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2479 { 2480 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2481 struct mbuf *nam = nfsd->nd_nam; 2482 char *dpos = nfsd->nd_dpos; 2483 kauth_cred_t cred = nfsd->nd_cr; 2484 u_int32_t *tl; 2485 int32_t t1; 2486 char *bpos; 2487 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2488 int v3 = (nfsd->nd_flag & ND_NFSV3); 2489 char *cp2; 2490 struct mbuf *mb, *mreq; 2491 struct vnode *vp, *dirp = (struct vnode *)0; 2492 struct vattr dirfor, diraft; 2493 nfsrvfh_t nsfh; 2494 struct nameidata nd; 2495 u_quad_t frev; 2496 2497 nfsm_srvmtofh(&nsfh); 2498 nfsm_srvnamesiz(len); 2499 nd.ni_cnd.cn_cred = cred; 2500 nd.ni_cnd.cn_nameiop = DELETE; 2501 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2502 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2503 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2504 if (dirp && v3) { 2505 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2506 } 2507 if (error) { 2508 if (nd.ni_pathbuf != NULL) { 2509 pathbuf_destroy(nd.ni_pathbuf); 2510 } 2511 nfsm_reply(NFSX_WCCDATA(v3)); 2512 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2513 if (dirp) 2514 vrele(dirp); 2515 return (0); 2516 } 2517 vp = nd.ni_vp; 2518 if (vp->v_type != VDIR) { 2519 error = ENOTDIR; 2520 goto out; 2521 } 2522 /* 2523 * No rmdir "." please. 2524 */ 2525 if (nd.ni_dvp == vp) { 2526 error = EINVAL; 2527 goto out; 2528 } 2529 /* 2530 * The root of a mounted filesystem cannot be deleted. 2531 */ 2532 if (vp->v_vflag & VV_ROOT) 2533 error = EBUSY; 2534 out: 2535 if (!error) { 2536 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2537 nqsrv_getl(vp, ND_WRITE); 2538 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2539 } else { 2540 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2541 if (nd.ni_dvp == nd.ni_vp) 2542 vrele(nd.ni_dvp); 2543 else 2544 vput(nd.ni_dvp); 2545 vput(vp); 2546 } 2547 pathbuf_destroy(nd.ni_pathbuf); 2548 if (dirp) { 2549 if (v3) { 2550 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2551 } 2552 vrele(dirp); 2553 } 2554 nfsm_reply(NFSX_WCCDATA(v3)); 2555 if (v3) { 2556 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2557 return (0); 2558 } 2559 nfsm_srvdone; 2560 } 2561 2562 /* 2563 * nfs readdir service 2564 * - mallocs what it thinks is enough to read 2565 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2566 * - calls VOP_READDIR() 2567 * - loops around building the reply 2568 * if the output generated exceeds count break out of loop 2569 * The nfsm_clget macro is used here so that the reply will be packed 2570 * tightly in mbuf clusters. 2571 * - it only knows that it has encountered eof when the VOP_READDIR() 2572 * reads nothing 2573 * - as such one readdir rpc will return eof false although you are there 2574 * and then the next will return eof 2575 * - it trims out records with d_fileno == 0 2576 * this doesn't matter for Unix clients, but they might confuse clients 2577 * for other os'. 2578 * - it trims out records with d_type == DT_WHT 2579 * these cannot be seen through NFS (unless we extend the protocol) 2580 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2581 * than requested, but this may not apply to all filesystems. For 2582 * example, client NFS does not { although it is never remote mounted 2583 * anyhow } 2584 * The alternate call nfsrv_readdirplus() does lookups as well. 2585 * PS: The NFS protocol spec. does not clarify what the "count" byte 2586 * argument is a count of.. just name strings and file id's or the 2587 * entire reply rpc or ... 2588 * I tried just file name and id sizes and it confused the Sun client, 2589 * so I am using the full rpc size now. The "paranoia.." comment refers 2590 * to including the status longwords that are not a part of the dir. 2591 * "entry" structures, but are in the rpc. 2592 */ 2593 2594 #define NFS_SRVDIRBLKSIZ 1024 2595 2596 struct flrep { 2597 nfsuint64 fl_off; 2598 u_int32_t fl_postopok; 2599 struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */ 2600 u_int32_t fl_fhok; 2601 u_int32_t fl_fhsize; 2602 /* handle comes here, filled in dynamically */ 2603 }; 2604 2605 int 2606 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2607 { 2608 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2609 struct mbuf *nam = nfsd->nd_nam; 2610 char *dpos = nfsd->nd_dpos; 2611 kauth_cred_t cred = nfsd->nd_cr; 2612 char *bp, *be; 2613 struct mbuf *mp; 2614 struct dirent *dp; 2615 char *cp; 2616 u_int32_t *tl; 2617 int32_t t1; 2618 char *bpos; 2619 struct mbuf *mb, *mreq, *mp2; 2620 char *cpos, *cend, *cp2, *rbuf; 2621 struct vnode *vp; 2622 struct vattr at; 2623 nfsrvfh_t nsfh; 2624 struct uio io; 2625 struct iovec iv; 2626 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2627 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2628 int v3 = (nfsd->nd_flag & ND_NFSV3); 2629 u_quad_t frev, off, toff, verf; 2630 off_t *cookies = NULL, *cookiep; 2631 nfsuint64 jar; 2632 2633 nfsm_srvmtofh(&nsfh); 2634 if (v3) { 2635 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2636 toff = fxdr_hyper(tl); 2637 tl += 2; 2638 verf = fxdr_hyper(tl); 2639 tl += 2; 2640 } else { 2641 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2642 toff = fxdr_unsigned(u_quad_t, *tl++); 2643 } 2644 off = toff; 2645 cnt = fxdr_unsigned(int, *tl); 2646 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2647 xfer = NFS_SRVMAXDATA(nfsd); 2648 if (siz > xfer) 2649 siz = xfer; 2650 fullsiz = siz; 2651 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2652 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2653 if (!error && vp->v_type != VDIR) { 2654 error = ENOTDIR; 2655 vput(vp); 2656 } 2657 if (error) { 2658 nfsm_reply(NFSX_UNSIGNED); 2659 nfsm_srvpostop_attr(getret, &at); 2660 return (0); 2661 } 2662 nqsrv_getl(vp, ND_READ); 2663 if (v3) { 2664 error = getret = VOP_GETATTR(vp, &at, cred); 2665 #ifdef NFS3_STRICTVERF 2666 /* 2667 * XXX This check is too strict for Solaris 2.5 clients. 2668 */ 2669 if (!error && toff && verf != at.va_filerev) 2670 error = NFSERR_BAD_COOKIE; 2671 #endif 2672 } 2673 if (!error) 2674 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2675 if (error) { 2676 vput(vp); 2677 nfsm_reply(NFSX_POSTOPATTR(v3)); 2678 nfsm_srvpostop_attr(getret, &at); 2679 return (0); 2680 } 2681 VOP_UNLOCK(vp); 2682 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2683 again: 2684 iv.iov_base = rbuf; 2685 iv.iov_len = fullsiz; 2686 io.uio_iov = &iv; 2687 io.uio_iovcnt = 1; 2688 io.uio_offset = (off_t)off; 2689 io.uio_resid = fullsiz; 2690 io.uio_rw = UIO_READ; 2691 UIO_SETUP_SYSSPACE(&io); 2692 eofflag = 0; 2693 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2694 2695 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2696 2697 off = (off_t)io.uio_offset; 2698 if (!cookies && !error) 2699 error = NFSERR_PERM; 2700 if (v3) { 2701 getret = VOP_GETATTR(vp, &at, cred); 2702 if (!error) 2703 error = getret; 2704 } 2705 2706 VOP_UNLOCK(vp); 2707 if (error) { 2708 vrele(vp); 2709 free((void *)rbuf, M_TEMP); 2710 if (cookies) 2711 free((void *)cookies, M_TEMP); 2712 nfsm_reply(NFSX_POSTOPATTR(v3)); 2713 nfsm_srvpostop_attr(getret, &at); 2714 return (0); 2715 } 2716 if (io.uio_resid) { 2717 siz -= io.uio_resid; 2718 2719 /* 2720 * If nothing read, return eof 2721 * rpc reply 2722 */ 2723 if (siz == 0) { 2724 vrele(vp); 2725 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2726 2 * NFSX_UNSIGNED); 2727 if (v3) { 2728 nfsm_srvpostop_attr(getret, &at); 2729 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2730 txdr_hyper(at.va_filerev, tl); 2731 tl += 2; 2732 } else 2733 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2734 *tl++ = nfs_false; 2735 *tl = nfs_true; 2736 free((void *)rbuf, M_TEMP); 2737 free((void *)cookies, M_TEMP); 2738 return (0); 2739 } 2740 } 2741 2742 /* 2743 * Check for degenerate cases of nothing useful read. 2744 * If so go try again 2745 */ 2746 cpos = rbuf; 2747 cend = rbuf + siz; 2748 dp = (struct dirent *)cpos; 2749 cookiep = cookies; 2750 2751 while (cpos < cend && ncookies > 0 && 2752 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2753 cpos += dp->d_reclen; 2754 dp = (struct dirent *)cpos; 2755 cookiep++; 2756 ncookies--; 2757 } 2758 if (cpos >= cend || ncookies == 0) { 2759 toff = off; 2760 siz = fullsiz; 2761 free(cookies, M_TEMP); 2762 cookies = NULL; 2763 goto again; 2764 } 2765 2766 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2767 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2768 if (v3) { 2769 nfsm_srvpostop_attr(getret, &at); 2770 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2771 txdr_hyper(at.va_filerev, tl); 2772 } 2773 mp = mp2 = mb; 2774 bp = bpos; 2775 be = bp + M_TRAILINGSPACE(mp); 2776 2777 /* Loop through the records and build reply */ 2778 while (cpos < cend && ncookies > 0) { 2779 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2780 nlen = dp->d_namlen; 2781 rem = nfsm_rndup(nlen)-nlen; 2782 len += (4 * NFSX_UNSIGNED + nlen + rem); 2783 if (v3) 2784 len += 2 * NFSX_UNSIGNED; 2785 if (len > cnt) { 2786 eofflag = 0; 2787 break; 2788 } 2789 /* 2790 * Build the directory record xdr from 2791 * the dirent entry. 2792 */ 2793 nfsm_clget; 2794 *tl = nfs_true; 2795 bp += NFSX_UNSIGNED; 2796 if (v3) { 2797 nfsm_clget; 2798 *tl = txdr_unsigned(dp->d_fileno >> 32); 2799 bp += NFSX_UNSIGNED; 2800 } 2801 nfsm_clget; 2802 *tl = txdr_unsigned(dp->d_fileno); 2803 bp += NFSX_UNSIGNED; 2804 nfsm_clget; 2805 *tl = txdr_unsigned(nlen); 2806 bp += NFSX_UNSIGNED; 2807 2808 /* And loop around copying the name */ 2809 xfer = nlen; 2810 cp = dp->d_name; 2811 while (xfer > 0) { 2812 nfsm_clget; 2813 if ((bp+xfer) > be) 2814 tsiz = be-bp; 2815 else 2816 tsiz = xfer; 2817 memcpy(bp, cp, tsiz); 2818 bp += tsiz; 2819 xfer -= tsiz; 2820 if (xfer > 0) 2821 cp += tsiz; 2822 } 2823 /* And null pad to an int32_t boundary */ 2824 for (i = 0; i < rem; i++) 2825 *bp++ = '\0'; 2826 nfsm_clget; 2827 2828 /* Finish off the record */ 2829 txdr_hyper(*cookiep, &jar); 2830 if (v3) { 2831 *tl = jar.nfsuquad[0]; 2832 bp += NFSX_UNSIGNED; 2833 nfsm_clget; 2834 } 2835 *tl = jar.nfsuquad[1]; 2836 bp += NFSX_UNSIGNED; 2837 } 2838 cpos += dp->d_reclen; 2839 dp = (struct dirent *)cpos; 2840 cookiep++; 2841 ncookies--; 2842 } 2843 vrele(vp); 2844 nfsm_clget; 2845 *tl = nfs_false; 2846 bp += NFSX_UNSIGNED; 2847 nfsm_clget; 2848 if (eofflag) 2849 *tl = nfs_true; 2850 else 2851 *tl = nfs_false; 2852 bp += NFSX_UNSIGNED; 2853 if (mp != mb) { 2854 if (bp < be) 2855 mp->m_len = bp - mtod(mp, char *); 2856 } else 2857 mp->m_len += bp - bpos; 2858 free((void *)rbuf, M_TEMP); 2859 free((void *)cookies, M_TEMP); 2860 nfsm_srvdone; 2861 } 2862 2863 int 2864 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2865 { 2866 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2867 struct mbuf *nam = nfsd->nd_nam; 2868 char *dpos = nfsd->nd_dpos; 2869 kauth_cred_t cred = nfsd->nd_cr; 2870 char *bp, *be; 2871 struct mbuf *mp; 2872 struct dirent *dp; 2873 char *cp; 2874 u_int32_t *tl; 2875 int32_t t1; 2876 char *bpos; 2877 struct mbuf *mb, *mreq, *mp2; 2878 char *cpos, *cend, *cp2, *rbuf; 2879 struct vnode *vp, *nvp; 2880 struct flrep fl; 2881 nfsrvfh_t nsfh; 2882 struct uio io; 2883 struct iovec iv; 2884 struct vattr va, at, *vap = &va; 2885 struct nfs_fattr *fp; 2886 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2887 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2888 u_quad_t frev, off, toff, verf; 2889 off_t *cookies = NULL, *cookiep; 2890 2891 nfsm_srvmtofh(&nsfh); 2892 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2893 toff = fxdr_hyper(tl); 2894 tl += 2; 2895 verf = fxdr_hyper(tl); 2896 tl += 2; 2897 siz = fxdr_unsigned(int, *tl++); 2898 cnt = fxdr_unsigned(int, *tl); 2899 off = toff; 2900 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2901 xfer = NFS_SRVMAXDATA(nfsd); 2902 if (siz > xfer) 2903 siz = xfer; 2904 fullsiz = siz; 2905 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2906 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2907 if (!error && vp->v_type != VDIR) { 2908 error = ENOTDIR; 2909 vput(vp); 2910 } 2911 if (error) { 2912 nfsm_reply(NFSX_UNSIGNED); 2913 nfsm_srvpostop_attr(getret, &at); 2914 return (0); 2915 } 2916 error = getret = VOP_GETATTR(vp, &at, cred); 2917 #ifdef NFS3_STRICTVERF 2918 /* 2919 * XXX This check is too strict for Solaris 2.5 clients. 2920 */ 2921 if (!error && toff && verf != at.va_filerev) 2922 error = NFSERR_BAD_COOKIE; 2923 #endif 2924 if (!error) { 2925 nqsrv_getl(vp, ND_READ); 2926 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2927 } 2928 if (error) { 2929 vput(vp); 2930 nfsm_reply(NFSX_V3POSTOPATTR); 2931 nfsm_srvpostop_attr(getret, &at); 2932 return (0); 2933 } 2934 VOP_UNLOCK(vp); 2935 2936 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2937 again: 2938 iv.iov_base = rbuf; 2939 iv.iov_len = fullsiz; 2940 io.uio_iov = &iv; 2941 io.uio_iovcnt = 1; 2942 io.uio_offset = (off_t)off; 2943 io.uio_resid = fullsiz; 2944 io.uio_rw = UIO_READ; 2945 UIO_SETUP_SYSSPACE(&io); 2946 eofflag = 0; 2947 2948 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2949 2950 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2951 2952 off = (u_quad_t)io.uio_offset; 2953 getret = VOP_GETATTR(vp, &at, cred); 2954 2955 VOP_UNLOCK(vp); 2956 2957 /* 2958 * If the VGET operation doesn't work for this filesystem, 2959 * we can't support readdirplus. Returning NOTSUPP should 2960 * make clients fall back to plain readdir. 2961 * There's no need to check for VPTOFH as well, we wouldn't 2962 * even be here otherwise. 2963 */ 2964 if (!getret) { 2965 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2966 getret = (getret == EOPNOTSUPP) ? 2967 NFSERR_NOTSUPP : NFSERR_IO; 2968 else 2969 vput(nvp); 2970 } 2971 2972 if (!cookies && !error) 2973 error = NFSERR_PERM; 2974 if (!error) 2975 error = getret; 2976 if (error) { 2977 vrele(vp); 2978 if (cookies) 2979 free((void *)cookies, M_TEMP); 2980 free((void *)rbuf, M_TEMP); 2981 nfsm_reply(NFSX_V3POSTOPATTR); 2982 nfsm_srvpostop_attr(getret, &at); 2983 return (0); 2984 } 2985 if (io.uio_resid) { 2986 siz -= io.uio_resid; 2987 2988 /* 2989 * If nothing read, return eof 2990 * rpc reply 2991 */ 2992 if (siz == 0) { 2993 vrele(vp); 2994 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2995 2 * NFSX_UNSIGNED); 2996 nfsm_srvpostop_attr(getret, &at); 2997 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2998 txdr_hyper(at.va_filerev, tl); 2999 tl += 2; 3000 *tl++ = nfs_false; 3001 *tl = nfs_true; 3002 free((void *)cookies, M_TEMP); 3003 free((void *)rbuf, M_TEMP); 3004 return (0); 3005 } 3006 } 3007 3008 /* 3009 * Check for degenerate cases of nothing useful read. 3010 * If so go try again 3011 */ 3012 cpos = rbuf; 3013 cend = rbuf + siz; 3014 dp = (struct dirent *)cpos; 3015 cookiep = cookies; 3016 3017 while (cpos < cend && ncookies > 0 && 3018 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 3019 cpos += dp->d_reclen; 3020 dp = (struct dirent *)cpos; 3021 cookiep++; 3022 ncookies--; 3023 } 3024 if (cpos >= cend || ncookies == 0) { 3025 toff = off; 3026 siz = fullsiz; 3027 free(cookies, M_TEMP); 3028 cookies = NULL; 3029 goto again; 3030 } 3031 3032 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3033 nfsm_reply(cnt); 3034 nfsm_srvpostop_attr(getret, &at); 3035 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3036 txdr_hyper(at.va_filerev, tl); 3037 mp = mp2 = mb; 3038 bp = bpos; 3039 be = bp + M_TRAILINGSPACE(mp); 3040 3041 /* Loop through the records and build reply */ 3042 while (cpos < cend && ncookies > 0) { 3043 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3044 nfsrvfh_t nnsfh; 3045 3046 nlen = dp->d_namlen; 3047 rem = nfsm_rndup(nlen)-nlen; 3048 3049 /* 3050 * For readdir_and_lookup get the vnode using 3051 * the file number. 3052 */ 3053 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3054 goto invalid; 3055 if (nfsrv_composefh(nvp, &nnsfh, true)) { 3056 vput(nvp); 3057 goto invalid; 3058 } 3059 if (VOP_GETATTR(nvp, vap, cred)) { 3060 vput(nvp); 3061 goto invalid; 3062 } 3063 vput(nvp); 3064 3065 /* 3066 * If either the dircount or maxcount will be 3067 * exceeded, get out now. Both of these lengths 3068 * are calculated conservatively, including all 3069 * XDR overheads. 3070 */ 3071 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3072 NFSX_V3POSTOPATTR); 3073 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3074 if (len > cnt || dirlen > fullsiz) { 3075 eofflag = 0; 3076 break; 3077 } 3078 3079 /* 3080 * Build the directory record xdr from 3081 * the dirent entry. 3082 */ 3083 fp = (struct nfs_fattr *)&fl.fl_fattr; 3084 nfsm_srvfillattr(vap, fp); 3085 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3086 fl.fl_fhok = nfs_true; 3087 fl.fl_postopok = nfs_true; 3088 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3089 3090 nfsm_clget; 3091 *tl = nfs_true; 3092 bp += NFSX_UNSIGNED; 3093 nfsm_clget; 3094 *tl = txdr_unsigned(dp->d_fileno >> 32); 3095 bp += NFSX_UNSIGNED; 3096 nfsm_clget; 3097 *tl = txdr_unsigned(dp->d_fileno); 3098 bp += NFSX_UNSIGNED; 3099 nfsm_clget; 3100 *tl = txdr_unsigned(nlen); 3101 bp += NFSX_UNSIGNED; 3102 3103 /* And loop around copying the name */ 3104 xfer = nlen; 3105 cp = dp->d_name; 3106 while (xfer > 0) { 3107 nfsm_clget; 3108 if ((bp + xfer) > be) 3109 tsiz = be - bp; 3110 else 3111 tsiz = xfer; 3112 memcpy(bp, cp, tsiz); 3113 bp += tsiz; 3114 xfer -= tsiz; 3115 if (xfer > 0) 3116 cp += tsiz; 3117 } 3118 /* And null pad to an int32_t boundary */ 3119 for (i = 0; i < rem; i++) 3120 *bp++ = '\0'; 3121 3122 /* 3123 * Now copy the flrep structure out. 3124 */ 3125 xfer = sizeof(struct flrep); 3126 cp = (void *)&fl; 3127 while (xfer > 0) { 3128 nfsm_clget; 3129 if ((bp + xfer) > be) 3130 tsiz = be - bp; 3131 else 3132 tsiz = xfer; 3133 memcpy(bp, cp, tsiz); 3134 bp += tsiz; 3135 xfer -= tsiz; 3136 if (xfer > 0) 3137 cp += tsiz; 3138 } 3139 3140 /* 3141 * ... and filehandle. 3142 */ 3143 xfer = NFSRVFH_SIZE(&nnsfh); 3144 cp = NFSRVFH_DATA(&nnsfh); 3145 while (xfer > 0) { 3146 nfsm_clget; 3147 if ((bp + xfer) > be) 3148 tsiz = be - bp; 3149 else 3150 tsiz = xfer; 3151 memcpy(bp, cp, tsiz); 3152 bp += tsiz; 3153 xfer -= tsiz; 3154 if (xfer > 0) 3155 cp += tsiz; 3156 } 3157 } 3158 invalid: 3159 cpos += dp->d_reclen; 3160 dp = (struct dirent *)cpos; 3161 cookiep++; 3162 ncookies--; 3163 } 3164 vrele(vp); 3165 nfsm_clget; 3166 *tl = nfs_false; 3167 bp += NFSX_UNSIGNED; 3168 nfsm_clget; 3169 if (eofflag) 3170 *tl = nfs_true; 3171 else 3172 *tl = nfs_false; 3173 bp += NFSX_UNSIGNED; 3174 if (mp != mb) { 3175 if (bp < be) 3176 mp->m_len = bp - mtod(mp, char *); 3177 } else 3178 mp->m_len += bp - bpos; 3179 free((void *)cookies, M_TEMP); 3180 free((void *)rbuf, M_TEMP); 3181 nfsm_srvdone; 3182 } 3183 3184 /* 3185 * nfs commit service 3186 */ 3187 int 3188 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3189 { 3190 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3191 struct mbuf *nam = nfsd->nd_nam; 3192 char *dpos = nfsd->nd_dpos; 3193 kauth_cred_t cred = nfsd->nd_cr; 3194 struct vattr bfor, aft; 3195 struct vnode *vp; 3196 nfsrvfh_t nsfh; 3197 u_int32_t *tl; 3198 int32_t t1; 3199 char *bpos; 3200 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3201 uint32_t cnt; 3202 char *cp2; 3203 struct mbuf *mb, *mreq; 3204 u_quad_t frev, off, end; 3205 3206 nfsm_srvmtofh(&nsfh); 3207 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3208 3209 off = fxdr_hyper(tl); 3210 tl += 2; 3211 cnt = fxdr_unsigned(uint32_t, *tl); 3212 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3213 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3214 if (error) { 3215 nfsm_reply(2 * NFSX_UNSIGNED); 3216 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3217 return (0); 3218 } 3219 for_ret = VOP_GETATTR(vp, &bfor, cred); 3220 end = (cnt > 0) ? off + cnt : vp->v_size; 3221 if (end < off || end > vp->v_size) 3222 end = vp->v_size; 3223 if (off < vp->v_size) 3224 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end); 3225 /* else error == 0, from nfsrv_fhtovp() */ 3226 aft_ret = VOP_GETATTR(vp, &aft, cred); 3227 vput(vp); 3228 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3229 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3230 if (!error) { 3231 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3232 *tl++ = txdr_unsigned(boottime.tv_sec); 3233 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 3234 } else { 3235 return (0); 3236 } 3237 nfsm_srvdone; 3238 } 3239 3240 /* 3241 * nfs statfs service 3242 */ 3243 int 3244 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3245 { 3246 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3247 struct mbuf *nam = nfsd->nd_nam; 3248 char *dpos = nfsd->nd_dpos; 3249 kauth_cred_t cred = nfsd->nd_cr; 3250 struct statvfs *sf = NULL; 3251 struct nfs_statfs *sfp; 3252 u_int32_t *tl; 3253 int32_t t1; 3254 char *bpos; 3255 int error = 0, rdonly, cache = 0, getret = 1; 3256 int v3 = (nfsd->nd_flag & ND_NFSV3); 3257 char *cp2; 3258 struct mbuf *mb, *mreq; 3259 struct vnode *vp; 3260 struct vattr at; 3261 nfsrvfh_t nsfh; 3262 u_quad_t frev, tval; 3263 3264 nfsm_srvmtofh(&nsfh); 3265 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3266 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3267 if (error) { 3268 nfsm_reply(NFSX_UNSIGNED); 3269 nfsm_srvpostop_attr(getret, &at); 3270 return (0); 3271 } 3272 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK); 3273 error = VFS_STATVFS(vp->v_mount, sf); 3274 getret = VOP_GETATTR(vp, &at, cred); 3275 vput(vp); 3276 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3277 if (v3) 3278 nfsm_srvpostop_attr(getret, &at); 3279 if (error) { 3280 free(sf, M_TEMP); 3281 return (0); 3282 } 3283 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3284 if (v3) { 3285 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3286 txdr_hyper(tval, &sfp->sf_tbytes); 3287 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3288 txdr_hyper(tval, &sfp->sf_fbytes); 3289 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3290 txdr_hyper(tval, &sfp->sf_abytes); 3291 tval = (u_quad_t)sf->f_files; 3292 txdr_hyper(tval, &sfp->sf_tfiles); 3293 tval = (u_quad_t)sf->f_ffree; 3294 txdr_hyper(tval, &sfp->sf_ffiles); 3295 txdr_hyper(tval, &sfp->sf_afiles); 3296 sfp->sf_invarsec = 0; 3297 } else { 3298 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3299 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3300 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3301 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3302 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3303 } 3304 nfsmout: 3305 if (sf) 3306 free(sf, M_TEMP); 3307 return error; 3308 } 3309 3310 /* 3311 * nfs fsinfo service 3312 */ 3313 int 3314 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3315 { 3316 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3317 struct mbuf *nam = nfsd->nd_nam; 3318 char *dpos = nfsd->nd_dpos; 3319 kauth_cred_t cred = nfsd->nd_cr; 3320 u_int32_t *tl; 3321 struct nfsv3_fsinfo *sip; 3322 int32_t t1; 3323 char *bpos; 3324 int error = 0, rdonly, cache = 0, getret = 1; 3325 uint32_t maxdata; 3326 char *cp2; 3327 struct mbuf *mb, *mreq; 3328 struct vnode *vp; 3329 struct vattr at; 3330 nfsrvfh_t nsfh; 3331 u_quad_t frev, maxfsize; 3332 struct statvfs *sb; 3333 3334 nfsm_srvmtofh(&nsfh); 3335 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3336 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3337 if (error) { 3338 nfsm_reply(NFSX_UNSIGNED); 3339 nfsm_srvpostop_attr(getret, &at); 3340 return (0); 3341 } 3342 3343 /* XXX Try to make a guess on the max file size. */ 3344 sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK); 3345 VFS_STATVFS(vp->v_mount, sb); 3346 maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1; 3347 free(sb, M_TEMP); 3348 3349 getret = VOP_GETATTR(vp, &at, cred); 3350 vput(vp); 3351 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3352 nfsm_srvpostop_attr(getret, &at); 3353 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3354 3355 /* 3356 * XXX 3357 * There should be file system VFS OP(s) to get this information. 3358 * For now, assume ufs. 3359 */ 3360 if (slp->ns_so->so_type == SOCK_DGRAM) 3361 maxdata = NFS_MAXDGRAMDATA; 3362 else 3363 maxdata = NFS_MAXDATA; 3364 sip->fs_rtmax = txdr_unsigned(maxdata); 3365 sip->fs_rtpref = txdr_unsigned(maxdata); 3366 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3367 sip->fs_wtmax = txdr_unsigned(maxdata); 3368 sip->fs_wtpref = txdr_unsigned(maxdata); 3369 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3370 sip->fs_dtpref = txdr_unsigned(maxdata); 3371 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3372 sip->fs_timedelta.nfsv3_sec = 0; 3373 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3374 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3375 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3376 NFSV3FSINFO_CANSETTIME); 3377 nfsm_srvdone; 3378 } 3379 3380 /* 3381 * nfs pathconf service 3382 */ 3383 int 3384 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3385 { 3386 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3387 struct mbuf *nam = nfsd->nd_nam; 3388 char *dpos = nfsd->nd_dpos; 3389 kauth_cred_t cred = nfsd->nd_cr; 3390 u_int32_t *tl; 3391 struct nfsv3_pathconf *pc; 3392 int32_t t1; 3393 char *bpos; 3394 int error = 0, rdonly, cache = 0, getret = 1; 3395 register_t linkmax, namemax, chownres, notrunc; 3396 char *cp2; 3397 struct mbuf *mb, *mreq; 3398 struct vnode *vp; 3399 struct vattr at; 3400 nfsrvfh_t nsfh; 3401 u_quad_t frev; 3402 3403 nfsm_srvmtofh(&nsfh); 3404 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3405 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3406 if (error) { 3407 nfsm_reply(NFSX_UNSIGNED); 3408 nfsm_srvpostop_attr(getret, &at); 3409 return (0); 3410 } 3411 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3412 if (!error) 3413 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3414 if (!error) 3415 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3416 if (!error) 3417 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3418 getret = VOP_GETATTR(vp, &at, cred); 3419 vput(vp); 3420 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3421 nfsm_srvpostop_attr(getret, &at); 3422 if (error) 3423 return (0); 3424 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3425 3426 pc->pc_linkmax = txdr_unsigned(linkmax); 3427 pc->pc_namemax = txdr_unsigned(namemax); 3428 pc->pc_notrunc = txdr_unsigned(notrunc); 3429 pc->pc_chownrestricted = txdr_unsigned(chownres); 3430 3431 /* 3432 * These should probably be supported by VOP_PATHCONF(), but 3433 * until msdosfs is exportable (why would you want to?), the 3434 * Unix defaults should be ok. 3435 */ 3436 pc->pc_caseinsensitive = nfs_false; 3437 pc->pc_casepreserving = nfs_true; 3438 nfsm_srvdone; 3439 } 3440 3441 /* 3442 * Null operation, used by clients to ping server 3443 */ 3444 /* ARGSUSED */ 3445 int 3446 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3447 struct lwp *lwp, struct mbuf **mrq) 3448 { 3449 struct mbuf *mrep = nfsd->nd_mrep; 3450 char *bpos; 3451 int error = NFSERR_RETVOID, cache = 0; 3452 struct mbuf *mb, *mreq; 3453 u_quad_t frev; 3454 3455 nfsm_reply(0); 3456 nfsmout: 3457 return (0); 3458 } 3459 3460 /* 3461 * No operation, used for obsolete procedures 3462 */ 3463 /* ARGSUSED */ 3464 int 3465 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3466 struct lwp *lwp, struct mbuf **mrq) 3467 { 3468 struct mbuf *mrep = nfsd->nd_mrep; 3469 char *bpos; 3470 int error, cache = 0; 3471 struct mbuf *mb, *mreq; 3472 u_quad_t frev; 3473 3474 if (nfsd->nd_repstat) 3475 error = nfsd->nd_repstat; 3476 else 3477 error = EPROCUNAVAIL; 3478 nfsm_reply(0); 3479 nfsmout: 3480 return (0); 3481 } 3482 3483 /* 3484 * Perform access checking for vnodes obtained from file handles that would 3485 * refer to files already opened by a Unix client. You cannot just use 3486 * vn_writechk() and VOP_ACCESS() for two reasons. 3487 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3488 * 2 - The owner is to be given access irrespective of mode bits for some 3489 * operations, so that processes that chmod after opening a file don't 3490 * break. I don't like this because it opens a security hole, but since 3491 * the nfs server opens a security hole the size of a barn door anyhow, 3492 * what the heck. 3493 * 3494 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3495 * will return EPERM instead of EACCES. EPERM is always an error. 3496 */ 3497 int 3498 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override) 3499 { 3500 struct vattr vattr; 3501 int error; 3502 if (flags & VWRITE) { 3503 /* Just vn_writechk() changed to check rdonly */ 3504 /* 3505 * Disallow write attempts on read-only file systems; 3506 * unless the file is a socket or a block or character 3507 * device resident on the file system. 3508 */ 3509 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3510 switch (vp->v_type) { 3511 case VREG: 3512 case VDIR: 3513 case VLNK: 3514 return (EROFS); 3515 default: 3516 break; 3517 } 3518 } 3519 3520 /* 3521 * If the vnode is in use as a process's text, 3522 * we can't allow writing. 3523 */ 3524 if (vp->v_iflag & VI_TEXT) 3525 return (ETXTBSY); 3526 } 3527 error = VOP_GETATTR(vp, &vattr, cred); 3528 if (error) 3529 return (error); 3530 error = VOP_ACCESS(vp, flags, cred); 3531 /* 3532 * Allow certain operations for the owner (reads and writes 3533 * on files that are already open). 3534 */ 3535 if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid) 3536 error = 0; 3537 return error; 3538 } 3539
Indexes created Wed Sep 24 05:09:52 GMT 2025