1 /* $NetBSD: nfs_serv.c,v 1.156 2011/02/05 13:35:11 yamt 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 1.156 2011/02/05 13:35:11 yamt 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 nfsm_reply(NFSX_POSTOPATTR(v3)); 467 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 468 return (0); 469 } 470 471 nqsrv_getl(ndp->ni_startdir, ND_READ); 472 pathbuf_destroy(nd.ni_pathbuf); 473 if (ipb != NULL) { 474 pathbuf_destroy(ipb); 475 } 476 vp = ndp->ni_vp; 477 error = nfsrv_composefh(vp, &nsfh, v3); 478 if (!error) 479 error = VOP_GETATTR(vp, &va, cred); 480 vput(vp); 481 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) + 482 NFSX_POSTOPATTR(v3)); 483 if (error) { 484 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 485 return (0); 486 } 487 nfsm_srvfhtom(&nsfh, v3); 488 if (v3) { 489 nfsm_srvpostop_attr(0, &va); 490 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 491 } else { 492 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 493 nfsm_srvfillattr(&va, fp); 494 } 495 nfsm_srvdone; 496 } 497 498 /* 499 * nfs readlink service 500 */ 501 int 502 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 503 { 504 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 505 struct mbuf *nam = nfsd->nd_nam; 506 char *dpos = nfsd->nd_dpos; 507 kauth_cred_t cred = nfsd->nd_cr; 508 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 509 struct iovec *ivp = iv; 510 struct mbuf *mp; 511 u_int32_t *tl; 512 int32_t t1; 513 char *bpos; 514 int error = 0, rdonly, cache = 0, i, padlen, getret; 515 uint32_t len; 516 int v3 = (nfsd->nd_flag & ND_NFSV3); 517 char *cp2; 518 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 519 struct vnode *vp; 520 struct vattr attr; 521 nfsrvfh_t nsfh; 522 struct uio io, *uiop = &io; 523 u_quad_t frev; 524 525 nfsm_srvmtofh(&nsfh); 526 len = 0; 527 i = 0; 528 while (len < NFS_MAXPATHLEN) { 529 mp = m_get(M_WAIT, MT_DATA); 530 MCLAIM(mp, &nfs_mowner); 531 m_clget(mp, M_WAIT); 532 mp->m_len = NFSMSIZ(mp); 533 if (len == 0) 534 mp3 = mp2 = mp; 535 else { 536 mp2->m_next = mp; 537 mp2 = mp; 538 } 539 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 540 mp->m_len = NFS_MAXPATHLEN-len; 541 len = NFS_MAXPATHLEN; 542 } else 543 len += mp->m_len; 544 ivp->iov_base = mtod(mp, void *); 545 ivp->iov_len = mp->m_len; 546 i++; 547 ivp++; 548 } 549 uiop->uio_iov = iv; 550 uiop->uio_iovcnt = i; 551 uiop->uio_offset = 0; 552 uiop->uio_resid = len; 553 uiop->uio_rw = UIO_READ; 554 UIO_SETUP_SYSSPACE(uiop); 555 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 556 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 557 if (error) { 558 m_freem(mp3); 559 nfsm_reply(2 * NFSX_UNSIGNED); 560 nfsm_srvpostop_attr(1, (struct vattr *)0); 561 return (0); 562 } 563 if (vp->v_type != VLNK) { 564 if (v3) 565 error = EINVAL; 566 else 567 error = ENXIO; 568 goto out; 569 } 570 nqsrv_getl(vp, ND_READ); 571 error = VOP_READLINK(vp, uiop, cred); 572 out: 573 getret = VOP_GETATTR(vp, &attr, cred); 574 vput(vp); 575 if (error) 576 m_freem(mp3); 577 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 578 if (v3) { 579 nfsm_srvpostop_attr(getret, &attr); 580 if (error) 581 return (0); 582 } 583 len -= uiop->uio_resid; 584 padlen = nfsm_padlen(len); 585 if (uiop->uio_resid || padlen) 586 nfs_zeropad(mp3, uiop->uio_resid, padlen); 587 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 588 *tl = txdr_unsigned(len); 589 mb->m_next = mp3; 590 nfsm_srvdone; 591 } 592 593 /* 594 * nfs read service 595 */ 596 int 597 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 598 { 599 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 600 struct mbuf *nam = nfsd->nd_nam; 601 char *dpos = nfsd->nd_dpos; 602 kauth_cred_t cred = nfsd->nd_cr; 603 struct mbuf *m; 604 struct nfs_fattr *fp; 605 u_int32_t *tl; 606 int32_t t1; 607 int i; 608 char *bpos; 609 int error = 0, rdonly, cache = 0, getret; 610 int v3 = (nfsd->nd_flag & ND_NFSV3); 611 uint32_t reqlen, len, cnt, left; 612 int padlen; 613 char *cp2; 614 struct mbuf *mb, *mreq; 615 struct vnode *vp; 616 nfsrvfh_t nsfh; 617 struct uio io, *uiop = &io; 618 struct vattr va; 619 off_t off; 620 u_quad_t frev; 621 622 nfsm_srvmtofh(&nsfh); 623 if (v3) { 624 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 625 off = fxdr_hyper(tl); 626 } else { 627 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 628 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 629 } 630 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 631 reqlen = fxdr_unsigned(uint32_t, *tl); 632 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 633 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 634 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 635 if (error) { 636 nfsm_reply(2 * NFSX_UNSIGNED); 637 nfsm_srvpostop_attr(1, (struct vattr *)0); 638 return (0); 639 } 640 if (vp->v_type != VREG) { 641 if (v3) 642 error = EINVAL; 643 else 644 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 645 } 646 if (!error) { 647 nqsrv_getl(vp, ND_READ); 648 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 649 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 650 } 651 getret = VOP_GETATTR(vp, &va, cred); 652 if (!error) 653 error = getret; 654 if (error) { 655 vput(vp); 656 nfsm_reply(NFSX_POSTOPATTR(v3)); 657 nfsm_srvpostop_attr(getret, &va); 658 return (0); 659 } 660 if (off >= va.va_size) 661 cnt = 0; 662 else if ((off + reqlen) > va.va_size) 663 cnt = va.va_size - off; 664 else 665 cnt = reqlen; 666 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 667 if (v3) { 668 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 669 *tl++ = nfs_true; 670 fp = (struct nfs_fattr *)tl; 671 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 672 } else { 673 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 674 fp = (struct nfs_fattr *)tl; 675 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 676 } 677 len = left = cnt; 678 if (cnt > 0) { 679 if (nfsd_use_loan) { 680 struct vm_page **pgpp; 681 voff_t pgoff = trunc_page(off); 682 int npages; 683 vaddr_t lva; 684 685 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 686 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 687 688 /* allocate kva for mbuf data */ 689 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 690 if (lva == 0) { 691 /* fall back to VOP_READ */ 692 goto loan_fail; 693 } 694 695 /* allocate mbuf */ 696 m = m_get(M_WAIT, MT_DATA); 697 MCLAIM(m, &nfs_mowner); 698 pgpp = m->m_ext.ext_pgs; 699 700 /* loan pages */ 701 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 702 pgpp); 703 if (error) { 704 sokvafree(lva, npages << PAGE_SHIFT); 705 m_free(m); 706 if (error == EBUSY) 707 goto loan_fail; 708 goto read_error; 709 } 710 711 /* associate kva to mbuf */ 712 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 713 cnt, M_MBUF, soloanfree, slp->ns_so); 714 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 715 m->m_len = cnt; 716 717 /* map pages */ 718 for (i = 0; i < npages; i++) { 719 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 720 VM_PROT_READ, 0); 721 lva += PAGE_SIZE; 722 } 723 724 pmap_update(pmap_kernel()); 725 726 mb->m_next = m; 727 mb = m; 728 error = 0; 729 uiop->uio_resid = 0; 730 } else { 731 struct iovec *iv; 732 struct iovec *iv2; 733 struct mbuf *m2; 734 int siz; 735 loan_fail: 736 /* 737 * Generate the mbuf list with the uio_iov ref. to it. 738 */ 739 i = 0; 740 m = m2 = mb; 741 while (left > 0) { 742 siz = min(M_TRAILINGSPACE(m), left); 743 if (siz > 0) { 744 left -= siz; 745 i++; 746 } 747 if (left > 0) { 748 m = m_get(M_WAIT, MT_DATA); 749 MCLAIM(m, &nfs_mowner); 750 m_clget(m, M_WAIT); 751 m->m_len = 0; 752 m2->m_next = m; 753 m2 = m; 754 } 755 } 756 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 757 uiop->uio_iov = iv2 = iv; 758 m = mb; 759 left = cnt; 760 i = 0; 761 while (left > 0) { 762 if (m == NULL) 763 panic("nfsrv_read iov"); 764 siz = min(M_TRAILINGSPACE(m), left); 765 if (siz > 0) { 766 iv->iov_base = mtod(m, char *) + 767 m->m_len; 768 iv->iov_len = siz; 769 m->m_len += siz; 770 left -= siz; 771 iv++; 772 i++; 773 } 774 m = m->m_next; 775 } 776 uiop->uio_iovcnt = i; 777 uiop->uio_offset = off; 778 uiop->uio_resid = cnt; 779 uiop->uio_rw = UIO_READ; 780 UIO_SETUP_SYSSPACE(uiop); 781 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 782 free((void *)iv2, M_TEMP); 783 } 784 read_error: 785 if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){ 786 if (!error) 787 error = getret; 788 m_freem(mreq); 789 vput(vp); 790 nfsm_reply(NFSX_POSTOPATTR(v3)); 791 nfsm_srvpostop_attr(getret, &va); 792 return (0); 793 } 794 } else { 795 uiop->uio_resid = 0; 796 } 797 vput(vp); 798 nfsm_srvfillattr(&va, fp); 799 len -= uiop->uio_resid; 800 padlen = nfsm_padlen(len); 801 if (uiop->uio_resid || padlen) 802 nfs_zeropad(mb, uiop->uio_resid, padlen); 803 if (v3) { 804 /* count */ 805 *tl++ = txdr_unsigned(len); 806 /* eof */ 807 if (off + len >= va.va_size) 808 *tl++ = nfs_true; 809 else 810 *tl++ = nfs_false; 811 } 812 *tl = txdr_unsigned(len); 813 nfsm_srvdone; 814 } 815 816 /* 817 * nfs write service 818 */ 819 int 820 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 821 { 822 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 823 struct mbuf *nam = nfsd->nd_nam; 824 char *dpos = nfsd->nd_dpos; 825 kauth_cred_t cred = nfsd->nd_cr; 826 struct iovec *ivp; 827 int i, cnt; 828 struct mbuf *mp; 829 struct nfs_fattr *fp; 830 struct iovec *iv; 831 struct vattr va, forat; 832 u_int32_t *tl; 833 int32_t t1; 834 char *bpos; 835 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 836 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 837 int stable = NFSV3WRITE_FILESYNC; 838 int v3 = (nfsd->nd_flag & ND_NFSV3); 839 char *cp2; 840 struct mbuf *mb, *mreq; 841 struct vnode *vp; 842 nfsrvfh_t nsfh; 843 struct uio io, *uiop = &io; 844 off_t off; 845 u_quad_t frev; 846 847 if (mrep == NULL) { 848 *mrq = NULL; 849 return (0); 850 } 851 nfsm_srvmtofh(&nsfh); 852 if (v3) { 853 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 854 off = fxdr_hyper(tl); 855 tl += 3; 856 stable = fxdr_unsigned(int, *tl++); 857 } else { 858 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 859 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 860 tl += 2; 861 } 862 retlen = len = fxdr_unsigned(int32_t, *tl); 863 cnt = i = 0; 864 865 /* 866 * For NFS Version 2, it is not obvious what a write of zero length 867 * should do, but I might as well be consistent with Version 3, 868 * which is to return ok so long as there are no permission problems. 869 */ 870 if (len > 0) { 871 zeroing = 1; 872 mp = mrep; 873 while (mp) { 874 if (mp == md) { 875 zeroing = 0; 876 adjust = dpos - mtod(mp, char *); 877 mp->m_len -= adjust; 878 if (mp->m_len > 0 && adjust > 0) 879 NFSMADV(mp, adjust); 880 } 881 if (zeroing) 882 mp->m_len = 0; 883 else if (mp->m_len > 0) { 884 i += mp->m_len; 885 if (i > len) { 886 mp->m_len -= (i - len); 887 zeroing = 1; 888 } 889 if (mp->m_len > 0) 890 cnt++; 891 } 892 mp = mp->m_next; 893 } 894 } 895 if (len > NFS_MAXDATA || len < 0 || i < len) { 896 error = EIO; 897 nfsm_reply(2 * NFSX_UNSIGNED); 898 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 899 return (0); 900 } 901 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 902 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 903 if (error) { 904 nfsm_reply(2 * NFSX_UNSIGNED); 905 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 906 return (0); 907 } 908 if (v3) 909 forat_ret = VOP_GETATTR(vp, &forat, cred); 910 if (vp->v_type != VREG) { 911 if (v3) 912 error = EINVAL; 913 else 914 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 915 } 916 if (!error) { 917 nqsrv_getl(vp, ND_WRITE); 918 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 919 } 920 if (error) { 921 vput(vp); 922 nfsm_reply(NFSX_WCCDATA(v3)); 923 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 924 return (0); 925 } 926 927 if (len > 0) { 928 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 929 uiop->uio_iov = iv = ivp; 930 uiop->uio_iovcnt = cnt; 931 mp = mrep; 932 while (mp) { 933 if (mp->m_len > 0) { 934 ivp->iov_base = mtod(mp, void *); 935 ivp->iov_len = mp->m_len; 936 ivp++; 937 } 938 mp = mp->m_next; 939 } 940 941 /* 942 * XXX 943 * The IO_METASYNC flag indicates that all metadata (and not 944 * just enough to ensure data integrity) must be written to 945 * stable storage synchronously. 946 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 947 */ 948 if (stable == NFSV3WRITE_UNSTABLE) 949 ioflags = IO_NODELOCKED; 950 else if (stable == NFSV3WRITE_DATASYNC) 951 ioflags = (IO_SYNC | IO_NODELOCKED); 952 else 953 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 954 uiop->uio_resid = len; 955 uiop->uio_rw = UIO_WRITE; 956 uiop->uio_offset = off; 957 UIO_SETUP_SYSSPACE(uiop); 958 error = VOP_WRITE(vp, uiop, ioflags, cred); 959 nfsstats.srvvop_writes++; 960 free(iv, M_TEMP); 961 } 962 aftat_ret = VOP_GETATTR(vp, &va, cred); 963 vput(vp); 964 if (!error) 965 error = aftat_ret; 966 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 967 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 968 if (v3) { 969 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 970 if (error) 971 return (0); 972 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 973 *tl++ = txdr_unsigned(retlen); 974 if (stable == NFSV3WRITE_UNSTABLE) 975 *tl++ = txdr_unsigned(stable); 976 else 977 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 978 /* 979 * Actually, there is no need to txdr these fields, 980 * but it may make the values more human readable, 981 * for debugging purposes. 982 */ 983 *tl++ = txdr_unsigned(boottime.tv_sec); 984 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 985 } else { 986 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 987 nfsm_srvfillattr(&va, fp); 988 } 989 nfsm_srvdone; 990 } 991 992 /* 993 * XXX elad: the original NFSW_SAMECRED() macro also made sure the 994 * two nd_flag fields of the descriptors contained 995 * ND_KERBAUTH. 996 */ 997 static int 998 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2) 999 { 1000 int i, do_ngroups; 1001 1002 if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2)) 1003 return (0); 1004 if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2)) 1005 return (0); 1006 do_ngroups = kauth_cred_ngroups(cred1); 1007 for (i = 0; i < do_ngroups; i++) 1008 if (kauth_cred_group(cred1, i) != 1009 kauth_cred_group(cred2, i)) 1010 return (0); 1011 1012 return (1); 1013 } 1014 1015 static struct nfsrvw_delayhash * 1016 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh) 1017 { 1018 uint32_t hash; 1019 1020 hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh), 1021 HASH32_BUF_INIT); 1022 return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ]; 1023 } 1024 1025 /* 1026 * NFS write service with write gathering support. Called when 1027 * nfsrvw_procrastinate > 0. 1028 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 1029 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 1030 * Jan. 1994. 1031 */ 1032 int 1033 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1034 { 1035 struct timeval now; 1036 struct iovec *ivp; 1037 struct mbuf *mp; 1038 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1039 struct nfs_fattr *fp; 1040 int i = 0; 1041 struct iovec *iov; 1042 struct nfsrvw_delayhash *wpp; 1043 kauth_cred_t cred; 1044 struct vattr va, forat; 1045 u_int32_t *tl; 1046 int32_t t1; 1047 char *bpos, *dpos; 1048 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1049 int ioflags, aftat_ret = 1, adjust, v3, zeroing; 1050 char *cp2; 1051 struct mbuf *mb, *mreq, *mrep, *md; 1052 struct vnode *vp; 1053 struct uio io, *uiop = &io; 1054 u_quad_t frev, cur_usec; 1055 1056 *mrq = NULL; 1057 if (*ndp) { 1058 nfsd = *ndp; 1059 *ndp = NULL; 1060 mrep = nfsd->nd_mrep; 1061 md = nfsd->nd_md; 1062 dpos = nfsd->nd_dpos; 1063 cred = nfsd->nd_cr; 1064 v3 = (nfsd->nd_flag & ND_NFSV3); 1065 LIST_INIT(&nfsd->nd_coalesce); 1066 nfsd->nd_mreq = NULL; 1067 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1068 getmicrotime(&now); 1069 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1070 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1071 1072 /* 1073 * Now, get the write header.. 1074 */ 1075 nfsm_srvmtofh(&nfsd->nd_fh); 1076 if (v3) { 1077 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1078 nfsd->nd_off = fxdr_hyper(tl); 1079 tl += 3; 1080 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1081 } else { 1082 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1083 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1084 tl += 2; 1085 } 1086 len = fxdr_unsigned(int32_t, *tl); 1087 nfsd->nd_len = len; 1088 nfsd->nd_eoff = nfsd->nd_off + len; 1089 1090 /* 1091 * Trim the header out of the mbuf list and trim off any trailing 1092 * junk so that the mbuf list has only the write data. 1093 */ 1094 zeroing = 1; 1095 i = 0; 1096 mp = mrep; 1097 while (mp) { 1098 if (mp == md) { 1099 zeroing = 0; 1100 adjust = dpos - mtod(mp, char *); 1101 mp->m_len -= adjust; 1102 if (mp->m_len > 0 && adjust > 0) 1103 NFSMADV(mp, adjust); 1104 } 1105 if (zeroing) 1106 mp->m_len = 0; 1107 else { 1108 i += mp->m_len; 1109 if (i > len) { 1110 mp->m_len -= (i - len); 1111 zeroing = 1; 1112 } 1113 } 1114 mp = mp->m_next; 1115 } 1116 if (len > NFS_MAXDATA || len < 0 || i < len) { 1117 nfsmout: 1118 m_freem(mrep); 1119 error = EIO; 1120 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1121 if (v3) 1122 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1123 nfsd->nd_mreq = mreq; 1124 nfsd->nd_mrep = NULL; 1125 nfsd->nd_time = 0; 1126 } 1127 1128 /* 1129 * Add this entry to the hash and time queues. 1130 */ 1131 owp = NULL; 1132 mutex_enter(&nfsd_lock); 1133 wp = LIST_FIRST(&slp->ns_tq); 1134 while (wp && wp->nd_time < nfsd->nd_time) { 1135 owp = wp; 1136 wp = LIST_NEXT(wp, nd_tq); 1137 } 1138 if (owp) { 1139 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1140 } else { 1141 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1142 } 1143 if (nfsd->nd_mrep) { 1144 wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh); 1145 owp = NULL; 1146 wp = LIST_FIRST(wpp); 1147 while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1148 owp = wp; 1149 wp = LIST_NEXT(wp, nd_hash); 1150 } 1151 while (wp && wp->nd_off < nfsd->nd_off && 1152 !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1153 owp = wp; 1154 wp = LIST_NEXT(wp, nd_hash); 1155 } 1156 if (owp) { 1157 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1158 1159 /* 1160 * Search the hash list for overlapping entries and 1161 * coalesce. 1162 */ 1163 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1164 wp = LIST_NEXT(nfsd, nd_hash); 1165 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr)) 1166 nfsrvw_coalesce(owp, nfsd); 1167 } 1168 } else { 1169 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1170 } 1171 } 1172 mutex_exit(&nfsd_lock); 1173 } 1174 1175 /* 1176 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1177 * and generate the associated reply mbuf list(s). 1178 */ 1179 loop1: 1180 getmicrotime(&now); 1181 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1182 mutex_enter(&nfsd_lock); 1183 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1184 owp = LIST_NEXT(nfsd, nd_tq); 1185 if (nfsd->nd_time > cur_usec) 1186 break; 1187 if (nfsd->nd_mreq) 1188 continue; 1189 LIST_REMOVE(nfsd, nd_tq); 1190 LIST_REMOVE(nfsd, nd_hash); 1191 mutex_exit(&nfsd_lock); 1192 1193 mrep = nfsd->nd_mrep; 1194 nfsd->nd_mrep = NULL; 1195 cred = nfsd->nd_cr; 1196 v3 = (nfsd->nd_flag & ND_NFSV3); 1197 forat_ret = aftat_ret = 1; 1198 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1199 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1200 false); 1201 if (!error) { 1202 if (v3) 1203 forat_ret = VOP_GETATTR(vp, &forat, cred); 1204 if (vp->v_type != VREG) { 1205 if (v3) 1206 error = EINVAL; 1207 else 1208 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1209 } 1210 } else 1211 vp = NULL; 1212 if (!error) { 1213 nqsrv_getl(vp, ND_WRITE); 1214 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1215 } 1216 1217 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1218 ioflags = IO_NODELOCKED; 1219 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1220 ioflags = (IO_SYNC | IO_NODELOCKED); 1221 else 1222 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1223 uiop->uio_rw = UIO_WRITE; 1224 uiop->uio_offset = nfsd->nd_off; 1225 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1226 UIO_SETUP_SYSSPACE(uiop); 1227 if (uiop->uio_resid > 0) { 1228 mp = mrep; 1229 i = 0; 1230 while (mp) { 1231 if (mp->m_len > 0) 1232 i++; 1233 mp = mp->m_next; 1234 } 1235 uiop->uio_iovcnt = i; 1236 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1237 uiop->uio_iov = ivp = iov; 1238 mp = mrep; 1239 while (mp) { 1240 if (mp->m_len > 0) { 1241 ivp->iov_base = mtod(mp, void *); 1242 ivp->iov_len = mp->m_len; 1243 ivp++; 1244 } 1245 mp = mp->m_next; 1246 } 1247 if (!error) { 1248 error = VOP_WRITE(vp, uiop, ioflags, cred); 1249 nfsstats.srvvop_writes++; 1250 } 1251 free((void *)iov, M_TEMP); 1252 } 1253 m_freem(mrep); 1254 if (vp) { 1255 aftat_ret = VOP_GETATTR(vp, &va, cred); 1256 vput(vp); 1257 } 1258 1259 /* 1260 * Loop around generating replies for all write rpcs that have 1261 * now been completed. 1262 */ 1263 swp = nfsd; 1264 do { 1265 if (error) { 1266 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1267 if (v3) { 1268 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1269 } 1270 } else { 1271 nfsm_writereply(NFSX_PREOPATTR(v3) + 1272 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1273 NFSX_WRITEVERF(v3), v3); 1274 if (v3) { 1275 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1276 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1277 *tl++ = txdr_unsigned(nfsd->nd_len); 1278 *tl++ = txdr_unsigned(swp->nd_stable); 1279 /* 1280 * Actually, there is no need to txdr these fields, 1281 * but it may make the values more human readable, 1282 * for debugging purposes. 1283 */ 1284 *tl++ = txdr_unsigned(boottime.tv_sec); 1285 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 1286 } else { 1287 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1288 nfsm_srvfillattr(&va, fp); 1289 } 1290 } 1291 nfsd->nd_mreq = mreq; 1292 if (nfsd->nd_mrep) 1293 panic("nfsrv_write: nd_mrep not free"); 1294 1295 /* 1296 * Done. Put it at the head of the timer queue so that 1297 * the final phase can return the reply. 1298 */ 1299 mutex_enter(&nfsd_lock); 1300 if (nfsd != swp) { 1301 nfsd->nd_time = 0; 1302 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1303 } 1304 nfsd = LIST_FIRST(&swp->nd_coalesce); 1305 if (nfsd) { 1306 LIST_REMOVE(nfsd, nd_tq); 1307 } 1308 mutex_exit(&nfsd_lock); 1309 } while (nfsd); 1310 swp->nd_time = 0; 1311 1312 mutex_enter(&nfsd_lock); 1313 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1314 mutex_exit(&nfsd_lock); 1315 goto loop1; 1316 } 1317 mutex_exit(&nfsd_lock); 1318 nfs_timer_start(); 1319 1320 /* 1321 * Search for a reply to return. 1322 */ 1323 mutex_enter(&nfsd_lock); 1324 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1325 if (nfsd->nd_mreq) { 1326 LIST_REMOVE(nfsd, nd_tq); 1327 *mrq = nfsd->nd_mreq; 1328 *ndp = nfsd; 1329 break; 1330 } 1331 } 1332 mutex_exit(&nfsd_lock); 1333 return (0); 1334 } 1335 1336 /* 1337 * Coalesce the write request nfsd into owp. To do this we must: 1338 * - remove nfsd from the queues 1339 * - merge nfsd->nd_mrep into owp->nd_mrep 1340 * - update the nd_eoff and nd_stable for owp 1341 * - put nfsd on owp's nd_coalesce list 1342 * NB: Must be called at splsoftclock(). 1343 */ 1344 void 1345 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd) 1346 { 1347 int overlap; 1348 struct mbuf *mp; 1349 struct nfsrv_descript *m; 1350 1351 KASSERT(mutex_owned(&nfsd_lock)); 1352 1353 LIST_REMOVE(nfsd, nd_hash); 1354 LIST_REMOVE(nfsd, nd_tq); 1355 if (owp->nd_eoff < nfsd->nd_eoff) { 1356 overlap = owp->nd_eoff - nfsd->nd_off; 1357 if (overlap < 0) 1358 panic("nfsrv_coalesce: bad off"); 1359 if (overlap > 0) 1360 m_adj(nfsd->nd_mrep, overlap); 1361 mp = owp->nd_mrep; 1362 while (mp->m_next) 1363 mp = mp->m_next; 1364 mp->m_next = nfsd->nd_mrep; 1365 owp->nd_eoff = nfsd->nd_eoff; 1366 } else 1367 m_freem(nfsd->nd_mrep); 1368 nfsd->nd_mrep = NULL; 1369 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1370 owp->nd_stable = NFSV3WRITE_FILESYNC; 1371 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1372 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1373 owp->nd_stable = NFSV3WRITE_DATASYNC; 1374 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1375 /* 1376 * nfsd might hold coalesce elements! Move them to owp. 1377 * Otherwise, requests may be lost and clients will be stuck. 1378 */ 1379 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1380 LIST_REMOVE(m, nd_tq); 1381 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1382 } 1383 } 1384 1385 /* 1386 * nfs create service 1387 * now does a truncate to 0 length via. setattr if it already exists 1388 */ 1389 int 1390 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1391 { 1392 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1393 struct mbuf *nam = nfsd->nd_nam; 1394 char *dpos = nfsd->nd_dpos; 1395 kauth_cred_t cred = nfsd->nd_cr; 1396 struct nfs_fattr *fp; 1397 struct vattr va, dirfor, diraft; 1398 struct nfsv2_sattr *sp; 1399 u_int32_t *tl; 1400 struct nameidata nd; 1401 char *cp; 1402 int32_t t1; 1403 char *bpos; 1404 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1405 int rdev = 0, abort = 0; 1406 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1407 char *cp2; 1408 struct mbuf *mb, *mreq; 1409 struct vnode *vp = NULL, *dirp = NULL; 1410 nfsrvfh_t nsfh; 1411 u_quad_t frev, tempsize; 1412 u_char cverf[NFSX_V3CREATEVERF]; 1413 1414 nd.ni_cnd.cn_nameiop = 0; 1415 nfsm_srvmtofh(&nsfh); 1416 nfsm_srvnamesiz(len); 1417 nd.ni_cnd.cn_cred = cred; 1418 nd.ni_cnd.cn_nameiop = CREATE; 1419 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1420 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1421 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1422 if (dirp && v3) { 1423 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1424 } 1425 if (error) { 1426 nfsm_reply(NFSX_WCCDATA(v3)); 1427 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1428 if (dirp) 1429 vrele(dirp); 1430 if (nd.ni_pathbuf != NULL) { 1431 pathbuf_destroy(nd.ni_pathbuf); 1432 nd.ni_pathbuf = NULL; 1433 } 1434 return (0); 1435 } 1436 abort = 1; 1437 vattr_null(&va); 1438 if (v3) { 1439 va.va_mode = 0; 1440 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1441 how = fxdr_unsigned(int, *tl); 1442 switch (how) { 1443 case NFSV3CREATE_GUARDED: 1444 if (nd.ni_vp) { 1445 error = EEXIST; 1446 break; 1447 } 1448 case NFSV3CREATE_UNCHECKED: 1449 nfsm_srvsattr(&va); 1450 break; 1451 case NFSV3CREATE_EXCLUSIVE: 1452 nfsm_dissect(cp, void *, NFSX_V3CREATEVERF); 1453 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1454 exclusive_flag = 1; 1455 break; 1456 }; 1457 va.va_type = VREG; 1458 } else { 1459 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1460 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1461 if (va.va_type == VNON) 1462 va.va_type = VREG; 1463 va.va_mode = nfstov_mode(sp->sa_mode); 1464 switch (va.va_type) { 1465 case VREG: 1466 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1467 if (tsize != -1) 1468 va.va_size = (u_quad_t)tsize; 1469 break; 1470 case VCHR: 1471 case VBLK: 1472 case VFIFO: 1473 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1474 break; 1475 default: 1476 break; 1477 }; 1478 } 1479 1480 /* 1481 * Iff doesn't exist, create it 1482 * otherwise just truncate to 0 length 1483 * should I set the mode too ?? 1484 */ 1485 if (nd.ni_vp == NULL) { 1486 if (va.va_type == VREG || va.va_type == VSOCK) { 1487 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1488 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1489 if (!error) { 1490 if (exclusive_flag) { 1491 exclusive_flag = 0; 1492 vattr_null(&va); 1493 /* 1494 * XXX 1495 * assuming NFSX_V3CREATEVERF 1496 * == sizeof(nfstime3) 1497 */ 1498 fxdr_nfsv3time(cverf, &va.va_atime); 1499 error = VOP_SETATTR(nd.ni_vp, &va, 1500 cred); 1501 } 1502 } 1503 } else if (va.va_type == VCHR || va.va_type == VBLK || 1504 va.va_type == VFIFO) { 1505 if (va.va_type == VCHR && rdev == 0xffffffff) 1506 va.va_type = VFIFO; 1507 if (va.va_type != VFIFO && 1508 (error = kauth_authorize_system(cred, 1509 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1510 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1511 vput(nd.ni_dvp); 1512 abort = 0; 1513 nfsm_reply(0); 1514 if (nd.ni_pathbuf != NULL) { 1515 pathbuf_destroy(nd.ni_pathbuf); 1516 nd.ni_pathbuf = NULL; 1517 } 1518 return (error); 1519 } else 1520 va.va_rdev = (dev_t)rdev; 1521 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1522 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1523 &va); 1524 if (error) { 1525 nfsm_reply(0); 1526 } 1527 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1528 vput(nd.ni_vp); 1529 vrele(nd.ni_dvp); 1530 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1531 if (nd.ni_pathbuf != NULL) { 1532 pathbuf_destroy(nd.ni_pathbuf); 1533 nd.ni_pathbuf = NULL; 1534 } 1535 error = EINVAL; 1536 abort = 0; 1537 nfsm_reply(0); 1538 } 1539 } else { 1540 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1541 if (nd.ni_pathbuf != NULL) { 1542 pathbuf_destroy(nd.ni_pathbuf); 1543 nd.ni_pathbuf = NULL; 1544 } 1545 vput(nd.ni_dvp); 1546 error = ENXIO; 1547 abort = 0; 1548 } 1549 vp = nd.ni_vp; 1550 } else { 1551 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1552 if (nd.ni_pathbuf != NULL) { 1553 pathbuf_destroy(nd.ni_pathbuf); 1554 nd.ni_pathbuf = NULL; 1555 } 1556 vp = nd.ni_vp; 1557 if (nd.ni_dvp == vp) 1558 vrele(nd.ni_dvp); 1559 else 1560 vput(nd.ni_dvp); 1561 abort = 0; 1562 if (!error && va.va_size != -1) { 1563 error = nfsrv_access(vp, VWRITE, cred, 1564 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1565 if (!error) { 1566 nqsrv_getl(vp, ND_WRITE); 1567 tempsize = va.va_size; 1568 vattr_null(&va); 1569 va.va_size = tempsize; 1570 error = VOP_SETATTR(vp, &va, cred); 1571 } 1572 } 1573 if (error) 1574 vput(vp); 1575 } 1576 if (!error) { 1577 error = nfsrv_composefh(vp, &nsfh, v3); 1578 if (!error) 1579 error = VOP_GETATTR(vp, &va, cred); 1580 vput(vp); 1581 } 1582 if (v3) { 1583 if (exclusive_flag && !error) { 1584 /* 1585 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1586 */ 1587 char oldverf[NFSX_V3CREATEVERF]; 1588 1589 txdr_nfsv3time(&va.va_atime, oldverf); 1590 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1591 error = EEXIST; 1592 } 1593 if (dirp) { 1594 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1595 } 1596 } 1597 if (dirp) { 1598 vrele(dirp); 1599 dirp = NULL; 1600 } 1601 if (nd.ni_pathbuf != NULL) { 1602 pathbuf_destroy(nd.ni_pathbuf); 1603 nd.ni_pathbuf = NULL; 1604 } 1605 abort = 0; 1606 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1607 if (v3) { 1608 if (!error) { 1609 nfsm_srvpostop_fh(&nsfh); 1610 nfsm_srvpostop_attr(0, &va); 1611 } 1612 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1613 } else { 1614 nfsm_srvfhtom(&nsfh, v3); 1615 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1616 nfsm_srvfillattr(&va, fp); 1617 } 1618 return (0); 1619 nfsmout: 1620 if (dirp) 1621 vrele(dirp); 1622 if (abort) { 1623 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1624 if (nd.ni_dvp == nd.ni_vp) 1625 vrele(nd.ni_dvp); 1626 else 1627 vput(nd.ni_dvp); 1628 if (nd.ni_vp) 1629 vput(nd.ni_vp); 1630 } 1631 if (nd.ni_pathbuf != NULL) { 1632 pathbuf_destroy(nd.ni_pathbuf); 1633 nd.ni_pathbuf = NULL; 1634 } 1635 return (error); 1636 } 1637 1638 /* 1639 * nfs v3 mknod service 1640 */ 1641 int 1642 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1643 { 1644 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1645 struct mbuf *nam = nfsd->nd_nam; 1646 char *dpos = nfsd->nd_dpos; 1647 kauth_cred_t cred = nfsd->nd_cr; 1648 struct vattr va, dirfor, diraft; 1649 u_int32_t *tl; 1650 struct nameidata nd; 1651 int32_t t1; 1652 char *bpos; 1653 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1654 int abort = 0; 1655 u_int32_t major, minor; 1656 enum vtype vtyp; 1657 char *cp2; 1658 struct mbuf *mb, *mreq; 1659 struct vnode *vp, *dirp = (struct vnode *)0; 1660 nfsrvfh_t nsfh; 1661 u_quad_t frev; 1662 1663 nd.ni_cnd.cn_nameiop = 0; 1664 nfsm_srvmtofh(&nsfh); 1665 nfsm_srvnamesiz(len); 1666 nd.ni_cnd.cn_cred = cred; 1667 nd.ni_cnd.cn_nameiop = CREATE; 1668 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1669 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1670 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1671 if (dirp) 1672 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1673 if (error) { 1674 nfsm_reply(NFSX_WCCDATA(1)); 1675 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1676 if (dirp) 1677 vrele(dirp); 1678 if (nd.ni_pathbuf != NULL) { 1679 pathbuf_destroy(nd.ni_pathbuf); 1680 nd.ni_pathbuf = NULL; 1681 } 1682 return (0); 1683 } 1684 abort = 1; 1685 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1686 vtyp = nfsv3tov_type(*tl); 1687 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1688 error = NFSERR_BADTYPE; 1689 goto abort; 1690 } 1691 vattr_null(&va); 1692 va.va_mode = 0; 1693 nfsm_srvsattr(&va); 1694 if (vtyp == VCHR || vtyp == VBLK) { 1695 dev_t rdev; 1696 1697 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1698 major = fxdr_unsigned(u_int32_t, *tl++); 1699 minor = fxdr_unsigned(u_int32_t, *tl); 1700 rdev = makedev(major, minor); 1701 if (major(rdev) != major || minor(rdev) != minor) { 1702 error = EINVAL; 1703 goto abort; 1704 } 1705 va.va_rdev = rdev; 1706 } 1707 1708 /* 1709 * Iff doesn't exist, create it. 1710 */ 1711 if (nd.ni_vp) { 1712 error = EEXIST; 1713 abort: 1714 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1715 if (nd.ni_dvp == nd.ni_vp) 1716 vrele(nd.ni_dvp); 1717 else 1718 vput(nd.ni_dvp); 1719 if (nd.ni_vp) 1720 vput(nd.ni_vp); 1721 if (nd.ni_pathbuf != NULL) { 1722 pathbuf_destroy(nd.ni_pathbuf); 1723 nd.ni_pathbuf = NULL; 1724 } 1725 goto out; 1726 } 1727 va.va_type = vtyp; 1728 if (vtyp == VSOCK) { 1729 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1730 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1731 } else { 1732 if (va.va_type != VFIFO && 1733 (error = kauth_authorize_system(cred, 1734 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1735 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1736 vput(nd.ni_dvp); 1737 goto out; 1738 } 1739 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1740 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1741 if (error) 1742 goto out; 1743 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1744 vput(nd.ni_vp); 1745 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1746 error = EINVAL; 1747 } 1748 } 1749 out: 1750 vp = nd.ni_vp; 1751 if (!error) { 1752 error = nfsrv_composefh(vp, &nsfh, true); 1753 if (!error) 1754 error = VOP_GETATTR(vp, &va, cred); 1755 vput(vp); 1756 } 1757 if (dirp) { 1758 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1759 vrele(dirp); 1760 dirp = NULL; 1761 } 1762 if (nd.ni_pathbuf != NULL) { 1763 pathbuf_destroy(nd.ni_pathbuf); 1764 nd.ni_pathbuf = NULL; 1765 } 1766 abort = 0; 1767 nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) + 1768 NFSX_WCCDATA(1)); 1769 if (!error) { 1770 nfsm_srvpostop_fh(&nsfh); 1771 nfsm_srvpostop_attr(0, &va); 1772 } 1773 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1774 return (0); 1775 nfsmout: 1776 if (abort) { 1777 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1778 if (nd.ni_dvp == nd.ni_vp) 1779 vrele(nd.ni_dvp); 1780 else 1781 vput(nd.ni_dvp); 1782 if (nd.ni_vp) 1783 vput(nd.ni_vp); 1784 } 1785 if (nd.ni_pathbuf != NULL) { 1786 pathbuf_destroy(nd.ni_pathbuf); 1787 nd.ni_pathbuf = NULL; 1788 } 1789 if (dirp) 1790 vrele(dirp); 1791 return (error); 1792 } 1793 1794 /* 1795 * nfs remove service 1796 */ 1797 int 1798 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1799 { 1800 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1801 struct mbuf *nam = nfsd->nd_nam; 1802 char *dpos = nfsd->nd_dpos; 1803 kauth_cred_t cred = nfsd->nd_cr; 1804 struct nameidata nd; 1805 u_int32_t *tl; 1806 int32_t t1; 1807 char *bpos; 1808 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1809 int v3 = (nfsd->nd_flag & ND_NFSV3); 1810 char *cp2; 1811 struct mbuf *mb, *mreq; 1812 struct vnode *vp, *dirp; 1813 struct vattr dirfor, diraft; 1814 nfsrvfh_t nsfh; 1815 u_quad_t frev; 1816 1817 #ifndef nolint 1818 vp = (struct vnode *)0; 1819 #endif 1820 nfsm_srvmtofh(&nsfh); 1821 nfsm_srvnamesiz(len); 1822 nd.ni_cnd.cn_cred = cred; 1823 nd.ni_cnd.cn_nameiop = DELETE; 1824 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1825 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1826 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1827 if (dirp && v3) { 1828 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1829 } 1830 if (!error) { 1831 vp = nd.ni_vp; 1832 if (vp->v_type == VDIR) { 1833 error = EPERM; 1834 goto out; 1835 } 1836 /* 1837 * The root of a mounted filesystem cannot be deleted. 1838 */ 1839 if (vp->v_vflag & VV_ROOT) { 1840 error = EBUSY; 1841 } 1842 out: 1843 if (!error) { 1844 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1845 nqsrv_getl(vp, ND_WRITE); 1846 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1847 } else { 1848 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1849 if (nd.ni_dvp == vp) 1850 vrele(nd.ni_dvp); 1851 else 1852 vput(nd.ni_dvp); 1853 vput(vp); 1854 } 1855 } 1856 if (nd.ni_pathbuf != NULL) { 1857 pathbuf_destroy(nd.ni_pathbuf); 1858 nd.ni_pathbuf = NULL; 1859 } 1860 if (dirp) { 1861 if (v3) { 1862 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1863 } 1864 vrele(dirp); 1865 } 1866 nfsm_reply(NFSX_WCCDATA(v3)); 1867 if (v3) { 1868 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1869 return (0); 1870 } 1871 nfsm_srvdone; 1872 } 1873 1874 /* 1875 * nfs rename service 1876 */ 1877 int 1878 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1879 { 1880 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1881 struct mbuf *nam = nfsd->nd_nam; 1882 char *dpos = nfsd->nd_dpos; 1883 kauth_cred_t cred = nfsd->nd_cr; 1884 u_int32_t *tl; 1885 int32_t t1; 1886 char *bpos; 1887 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1888 uint32_t len, len2; 1889 int tdirfor_ret = 1, tdiraft_ret = 1; 1890 int v3 = (nfsd->nd_flag & ND_NFSV3); 1891 char *cp2; 1892 struct mbuf *mb, *mreq; 1893 struct nameidata fromnd, tond; 1894 struct vnode *fvp, *tvp, *tdvp; 1895 struct vnode *fdirp = NULL, *tdirp = NULL; 1896 struct mount *localfs = NULL; 1897 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1898 nfsrvfh_t fnsfh, tnsfh; 1899 u_quad_t frev; 1900 uid_t saved_uid; 1901 1902 #ifndef nolint 1903 fvp = (struct vnode *)0; 1904 #endif 1905 fromnd.ni_cnd.cn_nameiop = 0; 1906 tond.ni_cnd.cn_nameiop = 0; 1907 nfsm_srvmtofh(&fnsfh); 1908 nfsm_srvnamesiz(len); 1909 /* 1910 * Remember our original uid so that we can reset cr_uid before 1911 * the second nfs_namei() call, in case it is remapped. 1912 */ 1913 saved_uid = kauth_cred_geteuid(cred); 1914 fromnd.ni_cnd.cn_cred = cred; 1915 fromnd.ni_cnd.cn_nameiop = DELETE; 1916 fromnd.ni_cnd.cn_flags = LOCKPARENT | INRENAME; 1917 error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md, 1918 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1919 if (fdirp && v3) { 1920 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred); 1921 } 1922 if (error) { 1923 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1924 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1925 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1926 if (fdirp) 1927 vrele(fdirp); 1928 pathbuf_destroy(fromnd.ni_pathbuf); 1929 return (0); 1930 } 1931 if (fromnd.ni_dvp != fromnd.ni_vp) { 1932 VOP_UNLOCK(fromnd.ni_dvp); 1933 } 1934 fvp = fromnd.ni_vp; 1935 1936 localfs = fvp->v_mount; 1937 error = VFS_RENAMELOCK_ENTER(localfs); 1938 if (error) { 1939 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1940 vrele(fromnd.ni_dvp); 1941 vrele(fvp); 1942 goto out1; 1943 } 1944 1945 /* Copied, regrettably, from vfs_syscalls.c (q.v.) */ 1946 vrele(fvp); 1947 if ((fromnd.ni_cnd.cn_namelen == 1 && 1948 fromnd.ni_cnd.cn_nameptr[0] == '.') || 1949 (fromnd.ni_cnd.cn_namelen == 2 && 1950 fromnd.ni_cnd.cn_nameptr[0] == '.' && 1951 fromnd.ni_cnd.cn_nameptr[1] == '.')) { 1952 error = EINVAL; 1953 VFS_RENAMELOCK_EXIT(localfs); 1954 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1955 vrele(fromnd.ni_dvp); 1956 goto out1; 1957 } 1958 vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY); 1959 error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0); 1960 if (error) { 1961 VOP_UNLOCK(fromnd.ni_dvp); 1962 VFS_RENAMELOCK_EXIT(localfs); 1963 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1964 vrele(fromnd.ni_dvp); 1965 goto out1; 1966 } 1967 VOP_UNLOCK(fromnd.ni_vp); 1968 if (fromnd.ni_dvp != fromnd.ni_vp) 1969 VOP_UNLOCK(fromnd.ni_dvp); 1970 fvp = fromnd.ni_vp; 1971 1972 nfsm_srvmtofh(&tnsfh); 1973 if (v3) { 1974 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1975 len2 = fxdr_unsigned(uint32_t, *tl); 1976 /* len2 will be checked by nfs_namei */ 1977 } 1978 else { 1979 /* NFSv2 */ 1980 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1981 } 1982 kauth_cred_seteuid(cred, saved_uid); 1983 tond.ni_cnd.cn_cred = cred; 1984 tond.ni_cnd.cn_nameiop = RENAME; 1985 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | INRENAME; 1986 error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md, 1987 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1988 if (tdirp && v3) { 1989 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred); 1990 } 1991 if (error) { 1992 VFS_RENAMELOCK_EXIT(localfs); 1993 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1994 vrele(fromnd.ni_dvp); 1995 vrele(fvp); 1996 goto out1; 1997 } 1998 tdvp = tond.ni_dvp; 1999 tvp = tond.ni_vp; 2000 if (tvp != NULL) { 2001 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 2002 if (v3) 2003 error = EEXIST; 2004 else 2005 error = EISDIR; 2006 goto out; 2007 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 2008 if (v3) 2009 error = EEXIST; 2010 else 2011 error = ENOTDIR; 2012 goto out; 2013 } 2014 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 2015 if (v3) 2016 error = EXDEV; 2017 else 2018 error = ENOTEMPTY; 2019 goto out; 2020 } 2021 } 2022 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 2023 if (v3) 2024 error = EXDEV; 2025 else 2026 error = ENOTEMPTY; 2027 goto out; 2028 } 2029 if (fvp->v_mount != tdvp->v_mount) { 2030 if (v3) 2031 error = EXDEV; 2032 else 2033 error = ENOTEMPTY; 2034 goto out; 2035 } 2036 if (fvp == tdvp) { 2037 if (v3) 2038 error = EINVAL; 2039 else 2040 error = ENOTEMPTY; 2041 } 2042 /* 2043 * If source is the same as the destination (that is the 2044 * same vnode with the same name in the same directory), 2045 * then there is nothing to do. 2046 */ 2047 if (fvp == tvp && fromnd.ni_dvp == tdvp && 2048 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 2049 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 2050 fromnd.ni_cnd.cn_namelen)) 2051 error = -1; 2052 out: 2053 if (!error) { 2054 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 2055 nqsrv_getl(tdvp, ND_WRITE); 2056 if (tvp) { 2057 nqsrv_getl(tvp, ND_WRITE); 2058 } 2059 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 2060 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 2061 VFS_RENAMELOCK_EXIT(localfs); 2062 } else { 2063 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 2064 if (tdvp == tvp) 2065 vrele(tdvp); 2066 else 2067 vput(tdvp); 2068 if (tvp) 2069 vput(tvp); 2070 VFS_RENAMELOCK_EXIT(localfs); 2071 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2072 vrele(fromnd.ni_dvp); 2073 vrele(fvp); 2074 if (error == -1) 2075 error = 0; 2076 } 2077 pathbuf_destroy(tond.ni_pathbuf); 2078 tond.ni_cnd.cn_nameiop = 0; 2079 out1: 2080 if (fdirp) { 2081 if (v3) { 2082 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred); 2083 } 2084 vrele(fdirp); 2085 fdirp = NULL; 2086 } 2087 if (tdirp) { 2088 if (v3) { 2089 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred); 2090 } 2091 vrele(tdirp); 2092 tdirp = NULL; 2093 } 2094 pathbuf_destroy(fromnd.ni_pathbuf); 2095 fromnd.ni_cnd.cn_nameiop = 0; 2096 localfs = NULL; 2097 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2098 if (v3) { 2099 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2100 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2101 } 2102 return (0); 2103 2104 nfsmout: 2105 if (fdirp) 2106 vrele(fdirp); 2107 #ifdef notdef 2108 if (tdirp) 2109 vrele(tdirp); 2110 #endif 2111 if (tond.ni_cnd.cn_nameiop) { 2112 pathbuf_destroy(tond.ni_pathbuf); 2113 } 2114 if (localfs) { 2115 VFS_RENAMELOCK_EXIT(localfs); 2116 } 2117 if (fromnd.ni_cnd.cn_nameiop) { 2118 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2119 pathbuf_destroy(fromnd.ni_pathbuf); 2120 vrele(fromnd.ni_dvp); 2121 vrele(fvp); 2122 } 2123 return (error); 2124 } 2125 2126 /* 2127 * nfs link service 2128 */ 2129 int 2130 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2131 { 2132 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2133 struct mbuf *nam = nfsd->nd_nam; 2134 char *dpos = nfsd->nd_dpos; 2135 kauth_cred_t cred = nfsd->nd_cr; 2136 struct nameidata nd; 2137 u_int32_t *tl; 2138 int32_t t1; 2139 char *bpos; 2140 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2141 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2142 char *cp2; 2143 struct mbuf *mb, *mreq; 2144 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2145 struct vattr dirfor, diraft, at; 2146 nfsrvfh_t nsfh, dnsfh; 2147 u_quad_t frev; 2148 2149 nfsm_srvmtofh(&nsfh); 2150 nfsm_srvmtofh(&dnsfh); 2151 nfsm_srvnamesiz(len); 2152 error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam, 2153 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2154 if (error) { 2155 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2156 nfsm_srvpostop_attr(getret, &at); 2157 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2158 return (0); 2159 } 2160 if (vp->v_type == VDIR) { 2161 error = EPERM; 2162 goto out1; 2163 } 2164 nd.ni_cnd.cn_cred = cred; 2165 nd.ni_cnd.cn_nameiop = CREATE; 2166 nd.ni_cnd.cn_flags = LOCKPARENT; 2167 error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos, 2168 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2169 if (dirp && v3) { 2170 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2171 } 2172 if (error) 2173 goto out1; 2174 xp = nd.ni_vp; 2175 if (xp != NULL) { 2176 error = EEXIST; 2177 goto out; 2178 } 2179 xp = nd.ni_dvp; 2180 if (vp->v_mount != xp->v_mount) 2181 error = EXDEV; 2182 out: 2183 if (!error) { 2184 nqsrv_getl(vp, ND_WRITE); 2185 nqsrv_getl(xp, ND_WRITE); 2186 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2187 } else { 2188 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2189 if (nd.ni_dvp == nd.ni_vp) 2190 vrele(nd.ni_dvp); 2191 else 2192 vput(nd.ni_dvp); 2193 if (nd.ni_vp) 2194 vrele(nd.ni_vp); 2195 } 2196 out1: 2197 if (v3) 2198 getret = VOP_GETATTR(vp, &at, cred); 2199 if (dirp) { 2200 if (v3) { 2201 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2202 } 2203 vrele(dirp); 2204 } 2205 vrele(vp); 2206 pathbuf_destroy(nd.ni_pathbuf); 2207 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2208 if (v3) { 2209 nfsm_srvpostop_attr(getret, &at); 2210 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2211 return (0); 2212 } 2213 nfsm_srvdone; 2214 } 2215 2216 /* 2217 * nfs symbolic link service 2218 */ 2219 int 2220 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2221 { 2222 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2223 struct mbuf *nam = nfsd->nd_nam; 2224 char *dpos = nfsd->nd_dpos; 2225 kauth_cred_t cred = nfsd->nd_cr; 2226 struct vattr va, dirfor, diraft; 2227 struct nameidata nd; 2228 u_int32_t *tl; 2229 int32_t t1; 2230 struct nfsv2_sattr *sp; 2231 char *bpos, *pathcp = NULL, *cp2; 2232 struct uio io; 2233 struct iovec iv; 2234 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0; 2235 uint32_t len, len2; 2236 int v3 = (nfsd->nd_flag & ND_NFSV3); 2237 struct mbuf *mb, *mreq; 2238 struct vnode *dirp = (struct vnode *)0; 2239 nfsrvfh_t nsfh; 2240 u_quad_t frev; 2241 2242 nd.ni_cnd.cn_nameiop = 0; 2243 nfsm_srvmtofh(&nsfh); 2244 nfsm_srvnamesiz(len); 2245 nd.ni_cnd.cn_cred = cred; 2246 nd.ni_cnd.cn_nameiop = CREATE; 2247 nd.ni_cnd.cn_flags = LOCKPARENT; 2248 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2249 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2250 if (dirp && v3) { 2251 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2252 } 2253 if (error) 2254 goto out; 2255 abort = 1; 2256 vattr_null(&va); 2257 va.va_type = VLNK; 2258 if (v3) { 2259 va.va_mode = 0; 2260 nfsm_srvsattr(&va); 2261 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2262 len2 = fxdr_unsigned(uint32_t, *tl); 2263 if (len2 > PATH_MAX) { 2264 /* XXX should check _PC_NO_TRUNC */ 2265 error = ENAMETOOLONG; 2266 goto abortop; 2267 } 2268 } 2269 else { 2270 /* NFSv2 */ 2271 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2272 } 2273 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2274 iv.iov_base = pathcp; 2275 iv.iov_len = len2; 2276 io.uio_resid = len2; 2277 io.uio_offset = 0; 2278 io.uio_iov = &iv; 2279 io.uio_iovcnt = 1; 2280 io.uio_rw = UIO_READ; 2281 UIO_SETUP_SYSSPACE(&io); 2282 nfsm_mtouio(&io, len2); 2283 if (!v3) { 2284 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2285 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2286 } 2287 *(pathcp + len2) = '\0'; 2288 if (nd.ni_vp) { 2289 error = EEXIST; 2290 abortop: 2291 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2292 if (nd.ni_dvp == nd.ni_vp) 2293 vrele(nd.ni_dvp); 2294 else 2295 vput(nd.ni_dvp); 2296 if (nd.ni_vp) 2297 vrele(nd.ni_vp); 2298 goto out; 2299 } 2300 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2301 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2302 if (!error) { 2303 if (v3) { 2304 error = nfsrv_composefh(nd.ni_vp, &nsfh, v3); 2305 if (!error) 2306 error = VOP_GETATTR(nd.ni_vp, &va, cred); 2307 vput(nd.ni_vp); 2308 } else { 2309 vput(nd.ni_vp); 2310 } 2311 } 2312 out: 2313 if (pathcp) 2314 free(pathcp, M_TEMP); 2315 if (dirp) { 2316 if (v3) { 2317 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2318 } 2319 vrele(dirp); 2320 dirp = NULL; 2321 } 2322 pathbuf_destroy(nd.ni_pathbuf); 2323 abort = 0; 2324 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2325 NFSX_WCCDATA(v3)); 2326 if (v3) { 2327 if (!error) { 2328 nfsm_srvpostop_fh(&nsfh); 2329 nfsm_srvpostop_attr(0, &va); 2330 } 2331 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2332 } 2333 return (0); 2334 nfsmout: 2335 if (abort) { 2336 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2337 if (nd.ni_dvp == nd.ni_vp) 2338 vrele(nd.ni_dvp); 2339 else 2340 vput(nd.ni_dvp); 2341 if (nd.ni_vp) 2342 vrele(nd.ni_vp); 2343 pathbuf_destroy(nd.ni_pathbuf); 2344 } 2345 if (dirp) 2346 vrele(dirp); 2347 if (pathcp) 2348 free(pathcp, M_TEMP); 2349 return (error); 2350 } 2351 2352 /* 2353 * nfs mkdir service 2354 */ 2355 int 2356 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2357 { 2358 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2359 struct mbuf *nam = nfsd->nd_nam; 2360 char *dpos = nfsd->nd_dpos; 2361 kauth_cred_t cred = nfsd->nd_cr; 2362 struct vattr va, dirfor, diraft; 2363 struct nfs_fattr *fp; 2364 struct nameidata nd; 2365 char *cp; 2366 u_int32_t *tl; 2367 int32_t t1; 2368 char *bpos; 2369 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2370 int abort = 0; 2371 int v3 = (nfsd->nd_flag & ND_NFSV3); 2372 char *cp2; 2373 struct mbuf *mb, *mreq; 2374 struct vnode *vp, *dirp = (struct vnode *)0; 2375 nfsrvfh_t nsfh; 2376 u_quad_t frev; 2377 2378 nfsm_srvmtofh(&nsfh); 2379 nfsm_srvnamesiz(len); 2380 nd.ni_cnd.cn_cred = cred; 2381 nd.ni_cnd.cn_nameiop = CREATE; 2382 nd.ni_cnd.cn_flags = LOCKPARENT; 2383 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2384 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2385 if (dirp && v3) { 2386 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2387 } 2388 if (error) { 2389 if (nd.ni_pathbuf != NULL) { 2390 pathbuf_destroy(nd.ni_pathbuf); 2391 } 2392 nfsm_reply(NFSX_WCCDATA(v3)); 2393 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2394 if (dirp) 2395 vrele(dirp); 2396 return (0); 2397 } 2398 abort = 1; 2399 vattr_null(&va); 2400 if (v3) { 2401 va.va_mode = 0; 2402 nfsm_srvsattr(&va); 2403 } else { 2404 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2405 va.va_mode = nfstov_mode(*tl++); 2406 } 2407 va.va_type = VDIR; 2408 vp = nd.ni_vp; 2409 if (vp != NULL) { 2410 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2411 if (nd.ni_dvp == vp) 2412 vrele(nd.ni_dvp); 2413 else 2414 vput(nd.ni_dvp); 2415 vrele(vp); 2416 error = EEXIST; 2417 goto out; 2418 } 2419 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2420 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2421 if (!error) { 2422 vp = nd.ni_vp; 2423 error = nfsrv_composefh(vp, &nsfh, v3); 2424 if (!error) 2425 error = VOP_GETATTR(vp, &va, cred); 2426 vput(vp); 2427 } 2428 out: 2429 if (dirp) { 2430 if (v3) { 2431 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2432 } 2433 vrele(dirp); 2434 dirp = NULL; 2435 } 2436 pathbuf_destroy(nd.ni_pathbuf); 2437 abort = 0; 2438 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2439 NFSX_WCCDATA(v3)); 2440 if (v3) { 2441 if (!error) { 2442 nfsm_srvpostop_fh(&nsfh); 2443 nfsm_srvpostop_attr(0, &va); 2444 } 2445 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2446 } else { 2447 nfsm_srvfhtom(&nsfh, v3); 2448 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2449 nfsm_srvfillattr(&va, fp); 2450 } 2451 return (0); 2452 nfsmout: 2453 if (abort) { 2454 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2455 if (nd.ni_dvp == nd.ni_vp) 2456 vrele(nd.ni_dvp); 2457 else 2458 vput(nd.ni_dvp); 2459 if (nd.ni_vp) 2460 vrele(nd.ni_vp); 2461 pathbuf_destroy(nd.ni_pathbuf); 2462 } 2463 if (dirp) 2464 vrele(dirp); 2465 return (error); 2466 } 2467 2468 /* 2469 * nfs rmdir service 2470 */ 2471 int 2472 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2473 { 2474 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2475 struct mbuf *nam = nfsd->nd_nam; 2476 char *dpos = nfsd->nd_dpos; 2477 kauth_cred_t cred = nfsd->nd_cr; 2478 u_int32_t *tl; 2479 int32_t t1; 2480 char *bpos; 2481 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2482 int v3 = (nfsd->nd_flag & ND_NFSV3); 2483 char *cp2; 2484 struct mbuf *mb, *mreq; 2485 struct vnode *vp, *dirp = (struct vnode *)0; 2486 struct vattr dirfor, diraft; 2487 nfsrvfh_t nsfh; 2488 struct nameidata nd; 2489 u_quad_t frev; 2490 2491 nfsm_srvmtofh(&nsfh); 2492 nfsm_srvnamesiz(len); 2493 nd.ni_cnd.cn_cred = cred; 2494 nd.ni_cnd.cn_nameiop = DELETE; 2495 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2496 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2497 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2498 if (dirp && v3) { 2499 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2500 } 2501 if (error) { 2502 if (nd.ni_pathbuf != NULL) { 2503 pathbuf_destroy(nd.ni_pathbuf); 2504 } 2505 nfsm_reply(NFSX_WCCDATA(v3)); 2506 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2507 if (dirp) 2508 vrele(dirp); 2509 return (0); 2510 } 2511 vp = nd.ni_vp; 2512 if (vp->v_type != VDIR) { 2513 error = ENOTDIR; 2514 goto out; 2515 } 2516 /* 2517 * No rmdir "." please. 2518 */ 2519 if (nd.ni_dvp == vp) { 2520 error = EINVAL; 2521 goto out; 2522 } 2523 /* 2524 * The root of a mounted filesystem cannot be deleted. 2525 */ 2526 if (vp->v_vflag & VV_ROOT) 2527 error = EBUSY; 2528 out: 2529 if (!error) { 2530 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2531 nqsrv_getl(vp, ND_WRITE); 2532 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2533 } else { 2534 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2535 if (nd.ni_dvp == nd.ni_vp) 2536 vrele(nd.ni_dvp); 2537 else 2538 vput(nd.ni_dvp); 2539 vput(vp); 2540 } 2541 pathbuf_destroy(nd.ni_pathbuf); 2542 if (dirp) { 2543 if (v3) { 2544 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2545 } 2546 vrele(dirp); 2547 } 2548 nfsm_reply(NFSX_WCCDATA(v3)); 2549 if (v3) { 2550 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2551 return (0); 2552 } 2553 nfsm_srvdone; 2554 } 2555 2556 /* 2557 * nfs readdir service 2558 * - mallocs what it thinks is enough to read 2559 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2560 * - calls VOP_READDIR() 2561 * - loops around building the reply 2562 * if the output generated exceeds count break out of loop 2563 * The nfsm_clget macro is used here so that the reply will be packed 2564 * tightly in mbuf clusters. 2565 * - it only knows that it has encountered eof when the VOP_READDIR() 2566 * reads nothing 2567 * - as such one readdir rpc will return eof false although you are there 2568 * and then the next will return eof 2569 * - it trims out records with d_fileno == 0 2570 * this doesn't matter for Unix clients, but they might confuse clients 2571 * for other os'. 2572 * - it trims out records with d_type == DT_WHT 2573 * these cannot be seen through NFS (unless we extend the protocol) 2574 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2575 * than requested, but this may not apply to all filesystems. For 2576 * example, client NFS does not { although it is never remote mounted 2577 * anyhow } 2578 * The alternate call nfsrv_readdirplus() does lookups as well. 2579 * PS: The NFS protocol spec. does not clarify what the "count" byte 2580 * argument is a count of.. just name strings and file id's or the 2581 * entire reply rpc or ... 2582 * I tried just file name and id sizes and it confused the Sun client, 2583 * so I am using the full rpc size now. The "paranoia.." comment refers 2584 * to including the status longwords that are not a part of the dir. 2585 * "entry" structures, but are in the rpc. 2586 */ 2587 2588 #define NFS_SRVDIRBLKSIZ 1024 2589 2590 struct flrep { 2591 nfsuint64 fl_off; 2592 u_int32_t fl_postopok; 2593 struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */ 2594 u_int32_t fl_fhok; 2595 u_int32_t fl_fhsize; 2596 /* handle comes here, filled in dynamically */ 2597 }; 2598 2599 int 2600 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2601 { 2602 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2603 struct mbuf *nam = nfsd->nd_nam; 2604 char *dpos = nfsd->nd_dpos; 2605 kauth_cred_t cred = nfsd->nd_cr; 2606 char *bp, *be; 2607 struct mbuf *mp; 2608 struct dirent *dp; 2609 char *cp; 2610 u_int32_t *tl; 2611 int32_t t1; 2612 char *bpos; 2613 struct mbuf *mb, *mreq, *mp2; 2614 char *cpos, *cend, *cp2, *rbuf; 2615 struct vnode *vp; 2616 struct vattr at; 2617 nfsrvfh_t nsfh; 2618 struct uio io; 2619 struct iovec iv; 2620 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2621 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2622 int v3 = (nfsd->nd_flag & ND_NFSV3); 2623 u_quad_t frev, off, toff, verf; 2624 off_t *cookies = NULL, *cookiep; 2625 nfsuint64 jar; 2626 2627 nfsm_srvmtofh(&nsfh); 2628 if (v3) { 2629 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2630 toff = fxdr_hyper(tl); 2631 tl += 2; 2632 verf = fxdr_hyper(tl); 2633 tl += 2; 2634 } else { 2635 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2636 toff = fxdr_unsigned(u_quad_t, *tl++); 2637 } 2638 off = toff; 2639 cnt = fxdr_unsigned(int, *tl); 2640 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2641 xfer = NFS_SRVMAXDATA(nfsd); 2642 if (siz > xfer) 2643 siz = xfer; 2644 fullsiz = siz; 2645 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2646 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2647 if (!error && vp->v_type != VDIR) { 2648 error = ENOTDIR; 2649 vput(vp); 2650 } 2651 if (error) { 2652 nfsm_reply(NFSX_UNSIGNED); 2653 nfsm_srvpostop_attr(getret, &at); 2654 return (0); 2655 } 2656 nqsrv_getl(vp, ND_READ); 2657 if (v3) { 2658 error = getret = VOP_GETATTR(vp, &at, cred); 2659 #ifdef NFS3_STRICTVERF 2660 /* 2661 * XXX This check is too strict for Solaris 2.5 clients. 2662 */ 2663 if (!error && toff && verf != at.va_filerev) 2664 error = NFSERR_BAD_COOKIE; 2665 #endif 2666 } 2667 if (!error) 2668 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2669 if (error) { 2670 vput(vp); 2671 nfsm_reply(NFSX_POSTOPATTR(v3)); 2672 nfsm_srvpostop_attr(getret, &at); 2673 return (0); 2674 } 2675 VOP_UNLOCK(vp); 2676 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2677 again: 2678 iv.iov_base = rbuf; 2679 iv.iov_len = fullsiz; 2680 io.uio_iov = &iv; 2681 io.uio_iovcnt = 1; 2682 io.uio_offset = (off_t)off; 2683 io.uio_resid = fullsiz; 2684 io.uio_rw = UIO_READ; 2685 UIO_SETUP_SYSSPACE(&io); 2686 eofflag = 0; 2687 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2688 2689 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2690 2691 off = (off_t)io.uio_offset; 2692 if (!cookies && !error) 2693 error = NFSERR_PERM; 2694 if (v3) { 2695 getret = VOP_GETATTR(vp, &at, cred); 2696 if (!error) 2697 error = getret; 2698 } 2699 2700 VOP_UNLOCK(vp); 2701 if (error) { 2702 vrele(vp); 2703 free((void *)rbuf, M_TEMP); 2704 if (cookies) 2705 free((void *)cookies, M_TEMP); 2706 nfsm_reply(NFSX_POSTOPATTR(v3)); 2707 nfsm_srvpostop_attr(getret, &at); 2708 return (0); 2709 } 2710 if (io.uio_resid) { 2711 siz -= io.uio_resid; 2712 2713 /* 2714 * If nothing read, return eof 2715 * rpc reply 2716 */ 2717 if (siz == 0) { 2718 vrele(vp); 2719 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2720 2 * NFSX_UNSIGNED); 2721 if (v3) { 2722 nfsm_srvpostop_attr(getret, &at); 2723 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2724 txdr_hyper(at.va_filerev, tl); 2725 tl += 2; 2726 } else 2727 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2728 *tl++ = nfs_false; 2729 *tl = nfs_true; 2730 free((void *)rbuf, M_TEMP); 2731 free((void *)cookies, M_TEMP); 2732 return (0); 2733 } 2734 } 2735 2736 /* 2737 * Check for degenerate cases of nothing useful read. 2738 * If so go try again 2739 */ 2740 cpos = rbuf; 2741 cend = rbuf + siz; 2742 dp = (struct dirent *)cpos; 2743 cookiep = cookies; 2744 2745 while (cpos < cend && ncookies > 0 && 2746 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2747 cpos += dp->d_reclen; 2748 dp = (struct dirent *)cpos; 2749 cookiep++; 2750 ncookies--; 2751 } 2752 if (cpos >= cend || ncookies == 0) { 2753 toff = off; 2754 siz = fullsiz; 2755 free(cookies, M_TEMP); 2756 cookies = NULL; 2757 goto again; 2758 } 2759 2760 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2761 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2762 if (v3) { 2763 nfsm_srvpostop_attr(getret, &at); 2764 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2765 txdr_hyper(at.va_filerev, tl); 2766 } 2767 mp = mp2 = mb; 2768 bp = bpos; 2769 be = bp + M_TRAILINGSPACE(mp); 2770 2771 /* Loop through the records and build reply */ 2772 while (cpos < cend && ncookies > 0) { 2773 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2774 nlen = dp->d_namlen; 2775 rem = nfsm_rndup(nlen)-nlen; 2776 len += (4 * NFSX_UNSIGNED + nlen + rem); 2777 if (v3) 2778 len += 2 * NFSX_UNSIGNED; 2779 if (len > cnt) { 2780 eofflag = 0; 2781 break; 2782 } 2783 /* 2784 * Build the directory record xdr from 2785 * the dirent entry. 2786 */ 2787 nfsm_clget; 2788 *tl = nfs_true; 2789 bp += NFSX_UNSIGNED; 2790 if (v3) { 2791 nfsm_clget; 2792 *tl = txdr_unsigned(dp->d_fileno >> 32); 2793 bp += NFSX_UNSIGNED; 2794 } 2795 nfsm_clget; 2796 *tl = txdr_unsigned(dp->d_fileno); 2797 bp += NFSX_UNSIGNED; 2798 nfsm_clget; 2799 *tl = txdr_unsigned(nlen); 2800 bp += NFSX_UNSIGNED; 2801 2802 /* And loop around copying the name */ 2803 xfer = nlen; 2804 cp = dp->d_name; 2805 while (xfer > 0) { 2806 nfsm_clget; 2807 if ((bp+xfer) > be) 2808 tsiz = be-bp; 2809 else 2810 tsiz = xfer; 2811 memcpy(bp, cp, tsiz); 2812 bp += tsiz; 2813 xfer -= tsiz; 2814 if (xfer > 0) 2815 cp += tsiz; 2816 } 2817 /* And null pad to an int32_t boundary */ 2818 for (i = 0; i < rem; i++) 2819 *bp++ = '\0'; 2820 nfsm_clget; 2821 2822 /* Finish off the record */ 2823 txdr_hyper(*cookiep, &jar); 2824 if (v3) { 2825 *tl = jar.nfsuquad[0]; 2826 bp += NFSX_UNSIGNED; 2827 nfsm_clget; 2828 } 2829 *tl = jar.nfsuquad[1]; 2830 bp += NFSX_UNSIGNED; 2831 } 2832 cpos += dp->d_reclen; 2833 dp = (struct dirent *)cpos; 2834 cookiep++; 2835 ncookies--; 2836 } 2837 vrele(vp); 2838 nfsm_clget; 2839 *tl = nfs_false; 2840 bp += NFSX_UNSIGNED; 2841 nfsm_clget; 2842 if (eofflag) 2843 *tl = nfs_true; 2844 else 2845 *tl = nfs_false; 2846 bp += NFSX_UNSIGNED; 2847 if (mp != mb) { 2848 if (bp < be) 2849 mp->m_len = bp - mtod(mp, char *); 2850 } else 2851 mp->m_len += bp - bpos; 2852 free((void *)rbuf, M_TEMP); 2853 free((void *)cookies, M_TEMP); 2854 nfsm_srvdone; 2855 } 2856 2857 int 2858 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2859 { 2860 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2861 struct mbuf *nam = nfsd->nd_nam; 2862 char *dpos = nfsd->nd_dpos; 2863 kauth_cred_t cred = nfsd->nd_cr; 2864 char *bp, *be; 2865 struct mbuf *mp; 2866 struct dirent *dp; 2867 char *cp; 2868 u_int32_t *tl; 2869 int32_t t1; 2870 char *bpos; 2871 struct mbuf *mb, *mreq, *mp2; 2872 char *cpos, *cend, *cp2, *rbuf; 2873 struct vnode *vp, *nvp; 2874 struct flrep fl; 2875 nfsrvfh_t nsfh; 2876 struct uio io; 2877 struct iovec iv; 2878 struct vattr va, at, *vap = &va; 2879 struct nfs_fattr *fp; 2880 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2881 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2882 u_quad_t frev, off, toff, verf; 2883 off_t *cookies = NULL, *cookiep; 2884 2885 nfsm_srvmtofh(&nsfh); 2886 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2887 toff = fxdr_hyper(tl); 2888 tl += 2; 2889 verf = fxdr_hyper(tl); 2890 tl += 2; 2891 siz = fxdr_unsigned(int, *tl++); 2892 cnt = fxdr_unsigned(int, *tl); 2893 off = toff; 2894 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2895 xfer = NFS_SRVMAXDATA(nfsd); 2896 if (siz > xfer) 2897 siz = xfer; 2898 fullsiz = siz; 2899 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2900 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2901 if (!error && vp->v_type != VDIR) { 2902 error = ENOTDIR; 2903 vput(vp); 2904 } 2905 if (error) { 2906 nfsm_reply(NFSX_UNSIGNED); 2907 nfsm_srvpostop_attr(getret, &at); 2908 return (0); 2909 } 2910 error = getret = VOP_GETATTR(vp, &at, cred); 2911 #ifdef NFS3_STRICTVERF 2912 /* 2913 * XXX This check is too strict for Solaris 2.5 clients. 2914 */ 2915 if (!error && toff && verf != at.va_filerev) 2916 error = NFSERR_BAD_COOKIE; 2917 #endif 2918 if (!error) { 2919 nqsrv_getl(vp, ND_READ); 2920 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2921 } 2922 if (error) { 2923 vput(vp); 2924 nfsm_reply(NFSX_V3POSTOPATTR); 2925 nfsm_srvpostop_attr(getret, &at); 2926 return (0); 2927 } 2928 VOP_UNLOCK(vp); 2929 2930 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2931 again: 2932 iv.iov_base = rbuf; 2933 iv.iov_len = fullsiz; 2934 io.uio_iov = &iv; 2935 io.uio_iovcnt = 1; 2936 io.uio_offset = (off_t)off; 2937 io.uio_resid = fullsiz; 2938 io.uio_rw = UIO_READ; 2939 UIO_SETUP_SYSSPACE(&io); 2940 eofflag = 0; 2941 2942 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2943 2944 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2945 2946 off = (u_quad_t)io.uio_offset; 2947 getret = VOP_GETATTR(vp, &at, cred); 2948 2949 VOP_UNLOCK(vp); 2950 2951 /* 2952 * If the VGET operation doesn't work for this filesystem, 2953 * we can't support readdirplus. Returning NOTSUPP should 2954 * make clients fall back to plain readdir. 2955 * There's no need to check for VPTOFH as well, we wouldn't 2956 * even be here otherwise. 2957 */ 2958 if (!getret) { 2959 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2960 getret = (getret == EOPNOTSUPP) ? 2961 NFSERR_NOTSUPP : NFSERR_IO; 2962 else 2963 vput(nvp); 2964 } 2965 2966 if (!cookies && !error) 2967 error = NFSERR_PERM; 2968 if (!error) 2969 error = getret; 2970 if (error) { 2971 vrele(vp); 2972 if (cookies) 2973 free((void *)cookies, M_TEMP); 2974 free((void *)rbuf, M_TEMP); 2975 nfsm_reply(NFSX_V3POSTOPATTR); 2976 nfsm_srvpostop_attr(getret, &at); 2977 return (0); 2978 } 2979 if (io.uio_resid) { 2980 siz -= io.uio_resid; 2981 2982 /* 2983 * If nothing read, return eof 2984 * rpc reply 2985 */ 2986 if (siz == 0) { 2987 vrele(vp); 2988 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2989 2 * NFSX_UNSIGNED); 2990 nfsm_srvpostop_attr(getret, &at); 2991 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2992 txdr_hyper(at.va_filerev, tl); 2993 tl += 2; 2994 *tl++ = nfs_false; 2995 *tl = nfs_true; 2996 free((void *)cookies, M_TEMP); 2997 free((void *)rbuf, M_TEMP); 2998 return (0); 2999 } 3000 } 3001 3002 /* 3003 * Check for degenerate cases of nothing useful read. 3004 * If so go try again 3005 */ 3006 cpos = rbuf; 3007 cend = rbuf + siz; 3008 dp = (struct dirent *)cpos; 3009 cookiep = cookies; 3010 3011 while (cpos < cend && ncookies > 0 && 3012 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 3013 cpos += dp->d_reclen; 3014 dp = (struct dirent *)cpos; 3015 cookiep++; 3016 ncookies--; 3017 } 3018 if (cpos >= cend || ncookies == 0) { 3019 toff = off; 3020 siz = fullsiz; 3021 free(cookies, M_TEMP); 3022 cookies = NULL; 3023 goto again; 3024 } 3025 3026 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3027 nfsm_reply(cnt); 3028 nfsm_srvpostop_attr(getret, &at); 3029 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3030 txdr_hyper(at.va_filerev, tl); 3031 mp = mp2 = mb; 3032 bp = bpos; 3033 be = bp + M_TRAILINGSPACE(mp); 3034 3035 /* Loop through the records and build reply */ 3036 while (cpos < cend && ncookies > 0) { 3037 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3038 nfsrvfh_t nnsfh; 3039 3040 nlen = dp->d_namlen; 3041 rem = nfsm_rndup(nlen)-nlen; 3042 3043 /* 3044 * For readdir_and_lookup get the vnode using 3045 * the file number. 3046 */ 3047 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3048 goto invalid; 3049 if (nfsrv_composefh(nvp, &nnsfh, true)) { 3050 vput(nvp); 3051 goto invalid; 3052 } 3053 if (VOP_GETATTR(nvp, vap, cred)) { 3054 vput(nvp); 3055 goto invalid; 3056 } 3057 vput(nvp); 3058 3059 /* 3060 * If either the dircount or maxcount will be 3061 * exceeded, get out now. Both of these lengths 3062 * are calculated conservatively, including all 3063 * XDR overheads. 3064 */ 3065 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3066 NFSX_V3POSTOPATTR); 3067 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3068 if (len > cnt || dirlen > fullsiz) { 3069 eofflag = 0; 3070 break; 3071 } 3072 3073 /* 3074 * Build the directory record xdr from 3075 * the dirent entry. 3076 */ 3077 fp = (struct nfs_fattr *)&fl.fl_fattr; 3078 nfsm_srvfillattr(vap, fp); 3079 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3080 fl.fl_fhok = nfs_true; 3081 fl.fl_postopok = nfs_true; 3082 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3083 3084 nfsm_clget; 3085 *tl = nfs_true; 3086 bp += NFSX_UNSIGNED; 3087 nfsm_clget; 3088 *tl = txdr_unsigned(dp->d_fileno >> 32); 3089 bp += NFSX_UNSIGNED; 3090 nfsm_clget; 3091 *tl = txdr_unsigned(dp->d_fileno); 3092 bp += NFSX_UNSIGNED; 3093 nfsm_clget; 3094 *tl = txdr_unsigned(nlen); 3095 bp += NFSX_UNSIGNED; 3096 3097 /* And loop around copying the name */ 3098 xfer = nlen; 3099 cp = dp->d_name; 3100 while (xfer > 0) { 3101 nfsm_clget; 3102 if ((bp + xfer) > be) 3103 tsiz = be - bp; 3104 else 3105 tsiz = xfer; 3106 memcpy(bp, cp, tsiz); 3107 bp += tsiz; 3108 xfer -= tsiz; 3109 if (xfer > 0) 3110 cp += tsiz; 3111 } 3112 /* And null pad to an int32_t boundary */ 3113 for (i = 0; i < rem; i++) 3114 *bp++ = '\0'; 3115 3116 /* 3117 * Now copy the flrep structure out. 3118 */ 3119 xfer = sizeof(struct flrep); 3120 cp = (void *)&fl; 3121 while (xfer > 0) { 3122 nfsm_clget; 3123 if ((bp + xfer) > be) 3124 tsiz = be - bp; 3125 else 3126 tsiz = xfer; 3127 memcpy(bp, cp, tsiz); 3128 bp += tsiz; 3129 xfer -= tsiz; 3130 if (xfer > 0) 3131 cp += tsiz; 3132 } 3133 3134 /* 3135 * ... and filehandle. 3136 */ 3137 xfer = NFSRVFH_SIZE(&nnsfh); 3138 cp = NFSRVFH_DATA(&nnsfh); 3139 while (xfer > 0) { 3140 nfsm_clget; 3141 if ((bp + xfer) > be) 3142 tsiz = be - bp; 3143 else 3144 tsiz = xfer; 3145 memcpy(bp, cp, tsiz); 3146 bp += tsiz; 3147 xfer -= tsiz; 3148 if (xfer > 0) 3149 cp += tsiz; 3150 } 3151 } 3152 invalid: 3153 cpos += dp->d_reclen; 3154 dp = (struct dirent *)cpos; 3155 cookiep++; 3156 ncookies--; 3157 } 3158 vrele(vp); 3159 nfsm_clget; 3160 *tl = nfs_false; 3161 bp += NFSX_UNSIGNED; 3162 nfsm_clget; 3163 if (eofflag) 3164 *tl = nfs_true; 3165 else 3166 *tl = nfs_false; 3167 bp += NFSX_UNSIGNED; 3168 if (mp != mb) { 3169 if (bp < be) 3170 mp->m_len = bp - mtod(mp, char *); 3171 } else 3172 mp->m_len += bp - bpos; 3173 free((void *)cookies, M_TEMP); 3174 free((void *)rbuf, M_TEMP); 3175 nfsm_srvdone; 3176 } 3177 3178 /* 3179 * nfs commit service 3180 */ 3181 int 3182 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3183 { 3184 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3185 struct mbuf *nam = nfsd->nd_nam; 3186 char *dpos = nfsd->nd_dpos; 3187 kauth_cred_t cred = nfsd->nd_cr; 3188 struct vattr bfor, aft; 3189 struct vnode *vp; 3190 nfsrvfh_t nsfh; 3191 u_int32_t *tl; 3192 int32_t t1; 3193 char *bpos; 3194 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3195 uint32_t cnt; 3196 char *cp2; 3197 struct mbuf *mb, *mreq; 3198 u_quad_t frev, off, end; 3199 3200 nfsm_srvmtofh(&nsfh); 3201 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3202 3203 off = fxdr_hyper(tl); 3204 tl += 2; 3205 cnt = fxdr_unsigned(uint32_t, *tl); 3206 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3207 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3208 if (error) { 3209 nfsm_reply(2 * NFSX_UNSIGNED); 3210 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3211 return (0); 3212 } 3213 for_ret = VOP_GETATTR(vp, &bfor, cred); 3214 end = (cnt > 0) ? off + cnt : vp->v_size; 3215 if (end < off || end > vp->v_size) 3216 end = vp->v_size; 3217 if (off < vp->v_size) 3218 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end); 3219 /* else error == 0, from nfsrv_fhtovp() */ 3220 aft_ret = VOP_GETATTR(vp, &aft, cred); 3221 vput(vp); 3222 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3223 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3224 if (!error) { 3225 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3226 *tl++ = txdr_unsigned(boottime.tv_sec); 3227 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 3228 } else { 3229 return (0); 3230 } 3231 nfsm_srvdone; 3232 } 3233 3234 /* 3235 * nfs statfs service 3236 */ 3237 int 3238 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3239 { 3240 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3241 struct mbuf *nam = nfsd->nd_nam; 3242 char *dpos = nfsd->nd_dpos; 3243 kauth_cred_t cred = nfsd->nd_cr; 3244 struct statvfs *sf = NULL; 3245 struct nfs_statfs *sfp; 3246 u_int32_t *tl; 3247 int32_t t1; 3248 char *bpos; 3249 int error = 0, rdonly, cache = 0, getret = 1; 3250 int v3 = (nfsd->nd_flag & ND_NFSV3); 3251 char *cp2; 3252 struct mbuf *mb, *mreq; 3253 struct vnode *vp; 3254 struct vattr at; 3255 nfsrvfh_t nsfh; 3256 u_quad_t frev, tval; 3257 3258 nfsm_srvmtofh(&nsfh); 3259 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3260 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3261 if (error) { 3262 nfsm_reply(NFSX_UNSIGNED); 3263 nfsm_srvpostop_attr(getret, &at); 3264 return (0); 3265 } 3266 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK); 3267 error = VFS_STATVFS(vp->v_mount, sf); 3268 getret = VOP_GETATTR(vp, &at, cred); 3269 vput(vp); 3270 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3271 if (v3) 3272 nfsm_srvpostop_attr(getret, &at); 3273 if (error) { 3274 free(sf, M_TEMP); 3275 return (0); 3276 } 3277 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3278 if (v3) { 3279 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3280 txdr_hyper(tval, &sfp->sf_tbytes); 3281 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3282 txdr_hyper(tval, &sfp->sf_fbytes); 3283 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3284 txdr_hyper(tval, &sfp->sf_abytes); 3285 tval = (u_quad_t)sf->f_files; 3286 txdr_hyper(tval, &sfp->sf_tfiles); 3287 tval = (u_quad_t)sf->f_ffree; 3288 txdr_hyper(tval, &sfp->sf_ffiles); 3289 txdr_hyper(tval, &sfp->sf_afiles); 3290 sfp->sf_invarsec = 0; 3291 } else { 3292 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3293 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3294 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3295 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3296 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3297 } 3298 nfsmout: 3299 if (sf) 3300 free(sf, M_TEMP); 3301 return error; 3302 } 3303 3304 /* 3305 * nfs fsinfo service 3306 */ 3307 int 3308 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3309 { 3310 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3311 struct mbuf *nam = nfsd->nd_nam; 3312 char *dpos = nfsd->nd_dpos; 3313 kauth_cred_t cred = nfsd->nd_cr; 3314 u_int32_t *tl; 3315 struct nfsv3_fsinfo *sip; 3316 int32_t t1; 3317 char *bpos; 3318 int error = 0, rdonly, cache = 0, getret = 1; 3319 uint32_t maxdata; 3320 char *cp2; 3321 struct mbuf *mb, *mreq; 3322 struct vnode *vp; 3323 struct vattr at; 3324 nfsrvfh_t nsfh; 3325 u_quad_t frev, maxfsize; 3326 struct statvfs *sb; 3327 3328 nfsm_srvmtofh(&nsfh); 3329 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3330 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3331 if (error) { 3332 nfsm_reply(NFSX_UNSIGNED); 3333 nfsm_srvpostop_attr(getret, &at); 3334 return (0); 3335 } 3336 3337 /* XXX Try to make a guess on the max file size. */ 3338 sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK); 3339 VFS_STATVFS(vp->v_mount, sb); 3340 maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1; 3341 free(sb, M_TEMP); 3342 3343 getret = VOP_GETATTR(vp, &at, cred); 3344 vput(vp); 3345 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3346 nfsm_srvpostop_attr(getret, &at); 3347 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3348 3349 /* 3350 * XXX 3351 * There should be file system VFS OP(s) to get this information. 3352 * For now, assume ufs. 3353 */ 3354 if (slp->ns_so->so_type == SOCK_DGRAM) 3355 maxdata = NFS_MAXDGRAMDATA; 3356 else 3357 maxdata = NFS_MAXDATA; 3358 sip->fs_rtmax = txdr_unsigned(maxdata); 3359 sip->fs_rtpref = txdr_unsigned(maxdata); 3360 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3361 sip->fs_wtmax = txdr_unsigned(maxdata); 3362 sip->fs_wtpref = txdr_unsigned(maxdata); 3363 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3364 sip->fs_dtpref = txdr_unsigned(maxdata); 3365 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3366 sip->fs_timedelta.nfsv3_sec = 0; 3367 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3368 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3369 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3370 NFSV3FSINFO_CANSETTIME); 3371 nfsm_srvdone; 3372 } 3373 3374 /* 3375 * nfs pathconf service 3376 */ 3377 int 3378 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3379 { 3380 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3381 struct mbuf *nam = nfsd->nd_nam; 3382 char *dpos = nfsd->nd_dpos; 3383 kauth_cred_t cred = nfsd->nd_cr; 3384 u_int32_t *tl; 3385 struct nfsv3_pathconf *pc; 3386 int32_t t1; 3387 char *bpos; 3388 int error = 0, rdonly, cache = 0, getret = 1; 3389 register_t linkmax, namemax, chownres, notrunc; 3390 char *cp2; 3391 struct mbuf *mb, *mreq; 3392 struct vnode *vp; 3393 struct vattr at; 3394 nfsrvfh_t nsfh; 3395 u_quad_t frev; 3396 3397 nfsm_srvmtofh(&nsfh); 3398 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3399 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3400 if (error) { 3401 nfsm_reply(NFSX_UNSIGNED); 3402 nfsm_srvpostop_attr(getret, &at); 3403 return (0); 3404 } 3405 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3406 if (!error) 3407 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3408 if (!error) 3409 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3410 if (!error) 3411 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3412 getret = VOP_GETATTR(vp, &at, cred); 3413 vput(vp); 3414 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3415 nfsm_srvpostop_attr(getret, &at); 3416 if (error) 3417 return (0); 3418 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3419 3420 pc->pc_linkmax = txdr_unsigned(linkmax); 3421 pc->pc_namemax = txdr_unsigned(namemax); 3422 pc->pc_notrunc = txdr_unsigned(notrunc); 3423 pc->pc_chownrestricted = txdr_unsigned(chownres); 3424 3425 /* 3426 * These should probably be supported by VOP_PATHCONF(), but 3427 * until msdosfs is exportable (why would you want to?), the 3428 * Unix defaults should be ok. 3429 */ 3430 pc->pc_caseinsensitive = nfs_false; 3431 pc->pc_casepreserving = nfs_true; 3432 nfsm_srvdone; 3433 } 3434 3435 /* 3436 * Null operation, used by clients to ping server 3437 */ 3438 /* ARGSUSED */ 3439 int 3440 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3441 struct lwp *lwp, struct mbuf **mrq) 3442 { 3443 struct mbuf *mrep = nfsd->nd_mrep; 3444 char *bpos; 3445 int error = NFSERR_RETVOID, cache = 0; 3446 struct mbuf *mb, *mreq; 3447 u_quad_t frev; 3448 3449 nfsm_reply(0); 3450 nfsmout: 3451 return (0); 3452 } 3453 3454 /* 3455 * No operation, used for obsolete procedures 3456 */ 3457 /* ARGSUSED */ 3458 int 3459 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3460 struct lwp *lwp, struct mbuf **mrq) 3461 { 3462 struct mbuf *mrep = nfsd->nd_mrep; 3463 char *bpos; 3464 int error, cache = 0; 3465 struct mbuf *mb, *mreq; 3466 u_quad_t frev; 3467 3468 if (nfsd->nd_repstat) 3469 error = nfsd->nd_repstat; 3470 else 3471 error = EPROCUNAVAIL; 3472 nfsm_reply(0); 3473 nfsmout: 3474 return (0); 3475 } 3476 3477 /* 3478 * Perform access checking for vnodes obtained from file handles that would 3479 * refer to files already opened by a Unix client. You cannot just use 3480 * vn_writechk() and VOP_ACCESS() for two reasons. 3481 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3482 * 2 - The owner is to be given access irrespective of mode bits for some 3483 * operations, so that processes that chmod after opening a file don't 3484 * break. I don't like this because it opens a security hole, but since 3485 * the nfs server opens a security hole the size of a barn door anyhow, 3486 * what the heck. 3487 * 3488 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3489 * will return EPERM instead of EACCES. EPERM is always an error. 3490 */ 3491 int 3492 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override) 3493 { 3494 struct vattr vattr; 3495 int error; 3496 if (flags & VWRITE) { 3497 /* Just vn_writechk() changed to check rdonly */ 3498 /* 3499 * Disallow write attempts on read-only file systems; 3500 * unless the file is a socket or a block or character 3501 * device resident on the file system. 3502 */ 3503 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3504 switch (vp->v_type) { 3505 case VREG: 3506 case VDIR: 3507 case VLNK: 3508 return (EROFS); 3509 default: 3510 break; 3511 } 3512 } 3513 3514 /* 3515 * If the vnode is in use as a process's text, 3516 * we can't allow writing. 3517 */ 3518 if (vp->v_iflag & VI_TEXT) 3519 return (ETXTBSY); 3520 } 3521 error = VOP_GETATTR(vp, &vattr, cred); 3522 if (error) 3523 return (error); 3524 error = VOP_ACCESS(vp, flags, cred); 3525 /* 3526 * Allow certain operations for the owner (reads and writes 3527 * on files that are already open). 3528 */ 3529 if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid) 3530 error = 0; 3531 return error; 3532 } 3533