1 /* $NetBSD: nfs_serv.c,v 1.159 2011/04/18 00:38:33 dholland Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * nfs version 2 and 3 server calls to vnode ops 39 * - these routines generally have 3 phases 40 * 1 - break down and validate rpc request in mbuf list 41 * 2 - do the vnode ops for the request 42 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c) 43 * 3 - build the rpc reply in an mbuf list 44 * nb: 45 * - do not mix the phases, since the nfsm_?? macros can return failures 46 * on a bad rpc or similar and do not do any vrele() or vput()'s 47 * 48 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs 49 * error number iff error != 0 whereas 50 * returning an error from the server function implies a fatal error 51 * such as a badly constructed rpc request that should be dropped without 52 * a reply. 53 * For Version 3, nfsm_reply() does not return for the error case, since 54 * most version 3 rpcs return more than the status for error cases. 55 */ 56 57 #include <sys/cdefs.h> 58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.159 2011/04/18 00:38:33 dholland Exp $"); 59 60 #include <sys/param.h> 61 #include <sys/systm.h> 62 #include <sys/proc.h> 63 #include <sys/file.h> 64 #include <sys/namei.h> 65 #include <sys/vnode.h> 66 #include <sys/mount.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/mbuf.h> 70 #include <sys/dirent.h> 71 #include <sys/stat.h> 72 #include <sys/kernel.h> 73 #include <sys/hash.h> 74 #include <sys/kauth.h> 75 #include <sys/module.h> 76 #include <sys/syscall.h> 77 #include <sys/syscallargs.h> 78 #include <sys/syscallvar.h> 79 80 #include <uvm/uvm.h> 81 82 #include <nfs/nfsproto.h> 83 #include <nfs/rpcv2.h> 84 #include <nfs/nfs.h> 85 #include <nfs/xdr_subs.h> 86 #include <nfs/nfsm_subs.h> 87 #include <nfs/nfs_var.h> 88 89 MODULE(MODULE_CLASS_MISC, nfsserver, "nfs"); 90 91 /* Global vars */ 92 extern u_int32_t nfs_xdrneg1; 93 extern u_int32_t nfs_false, nfs_true; 94 extern const enum vtype nv3tov_type[8]; 95 extern struct nfsstats nfsstats; 96 extern const nfstype nfsv2_type[9]; 97 extern const nfstype nfsv3_type[9]; 98 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000; 99 bool nfsd_use_loan = true; /* use page-loan for READ OP */ 100 101 #define nqsrv_getl(vp, rw) /* nothing */ 102 103 static const struct syscall_package nfsserver_syscalls[] = { 104 { SYS_nfssvc, 0, (sy_call_t *)sys_nfssvc }, 105 { 0, 0, NULL }, 106 }; 107 108 static int 109 nfsserver_modcmd(modcmd_t cmd, void *arg) 110 { 111 extern struct vfs_hooks nfs_export_hooks; /* XXX */ 112 int error; 113 114 switch (cmd) { 115 case MODULE_CMD_INIT: 116 error = syscall_establish(NULL, nfsserver_syscalls); 117 if (error != 0) { 118 return error; 119 } 120 nfs_init(); /* XXX for monolithic kernel */ 121 netexport_init(); 122 nfsrv_initcache(); /* Init the server request cache */ 123 nfsrv_init(0); /* Init server data structures */ 124 vfs_hooks_attach(&nfs_export_hooks); 125 nfs_timer_srvinit(nfsrv_timer); 126 return 0; 127 case MODULE_CMD_FINI: 128 error = syscall_disestablish(NULL, nfsserver_syscalls); 129 if (error != 0) { 130 return error; 131 } 132 /* 133 * Kill export list before detaching VFS hooks, so we 134 * we don't leak state due to a concurrent umount(). 135 */ 136 netexport_fini(); 137 vfs_hooks_detach(&nfs_export_hooks); 138 139 /* Kill timer before server goes away. */ 140 nfs_timer_srvfini(); 141 nfsrv_fini(); 142 143 /* Server uses server cache, so kill cache last. */ 144 nfsrv_finicache(); 145 return 0; 146 default: 147 return ENOTTY; 148 } 149 } 150 151 /* 152 * nfs v3 access service 153 */ 154 int 155 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 156 { 157 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 158 struct mbuf *nam = nfsd->nd_nam; 159 char *dpos = nfsd->nd_dpos; 160 kauth_cred_t cred = nfsd->nd_cr; 161 struct vnode *vp; 162 nfsrvfh_t nsfh; 163 u_int32_t *tl; 164 int32_t t1; 165 char *bpos; 166 int error = 0, rdonly, cache = 0, getret; 167 char *cp2; 168 struct mbuf *mb, *mreq; 169 struct vattr va; 170 u_long inmode, testmode, outmode; 171 u_quad_t frev; 172 173 nfsm_srvmtofh(&nsfh); 174 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 175 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 176 (nfsd->nd_flag & ND_KERBAUTH), false); 177 if (error) { 178 nfsm_reply(NFSX_UNSIGNED); 179 nfsm_srvpostop_attr(1, (struct vattr *)0); 180 return (0); 181 } 182 inmode = fxdr_unsigned(u_int32_t, *tl); 183 outmode = 0; 184 if ((inmode & NFSV3ACCESS_READ) && 185 nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0) 186 outmode |= NFSV3ACCESS_READ; 187 if (vp->v_type != VDIR) { 188 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 189 if (testmode && 190 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 191 outmode |= testmode; 192 if ((inmode & NFSV3ACCESS_EXECUTE) && 193 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 194 outmode |= NFSV3ACCESS_EXECUTE; 195 } else { 196 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 197 NFSV3ACCESS_DELETE); 198 if (testmode && 199 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 200 outmode |= testmode; 201 if ((inmode & NFSV3ACCESS_LOOKUP) && 202 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 203 outmode |= NFSV3ACCESS_LOOKUP; 204 } 205 getret = VOP_GETATTR(vp, &va, cred); 206 vput(vp); 207 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 208 nfsm_srvpostop_attr(getret, &va); 209 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 210 *tl = txdr_unsigned(outmode); 211 nfsm_srvdone; 212 } 213 214 /* 215 * nfs getattr service 216 */ 217 int 218 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 219 { 220 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 221 struct mbuf *nam = nfsd->nd_nam; 222 char *dpos = nfsd->nd_dpos; 223 kauth_cred_t cred = nfsd->nd_cr; 224 struct nfs_fattr *fp; 225 struct vattr va; 226 struct vnode *vp; 227 nfsrvfh_t nsfh; 228 u_int32_t *tl; 229 int32_t t1; 230 char *bpos; 231 int error = 0, rdonly, cache = 0; 232 char *cp2; 233 struct mbuf *mb, *mreq; 234 u_quad_t frev; 235 236 nfsm_srvmtofh(&nsfh); 237 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 238 (nfsd->nd_flag & ND_KERBAUTH), false); 239 if (error) { 240 nfsm_reply(0); 241 return (0); 242 } 243 nqsrv_getl(vp, ND_READ); 244 error = VOP_GETATTR(vp, &va, cred); 245 vput(vp); 246 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 247 if (error) 248 return (0); 249 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 250 nfsm_srvfillattr(&va, fp); 251 nfsm_srvdone; 252 } 253 254 /* 255 * nfs setattr service 256 */ 257 int 258 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 259 { 260 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 261 struct mbuf *nam = nfsd->nd_nam; 262 char *dpos = nfsd->nd_dpos; 263 kauth_cred_t cred = nfsd->nd_cr; 264 struct vattr va, preat; 265 struct nfsv2_sattr *sp; 266 struct nfs_fattr *fp; 267 struct vnode *vp; 268 nfsrvfh_t nsfh; 269 u_int32_t *tl; 270 int32_t t1; 271 char *bpos; 272 int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1; 273 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0; 274 char *cp2; 275 struct mbuf *mb, *mreq; 276 u_quad_t frev; 277 struct timespec guard; 278 279 memset(&guard, 0, sizeof guard); /* XXX gcc */ 280 281 nfsm_srvmtofh(&nsfh); 282 vattr_null(&va); 283 if (v3) { 284 nfsm_srvsattr(&va); 285 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 286 gcheck = fxdr_unsigned(int, *tl); 287 if (gcheck) { 288 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 289 fxdr_nfsv3time(tl, &guard); 290 } 291 } else { 292 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 293 /* 294 * Nah nah nah nah na nah 295 * There is a bug in the Sun client that puts 0xffff in the mode 296 * field of sattr when it should put in 0xffffffff. The u_short 297 * doesn't sign extend. 298 * --> check the low order 2 bytes for 0xffff 299 */ 300 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 301 va.va_mode = nfstov_mode(sp->sa_mode); 302 if (sp->sa_uid != nfs_xdrneg1) 303 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 304 if (sp->sa_gid != nfs_xdrneg1) 305 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 306 if (sp->sa_size != nfs_xdrneg1) 307 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 308 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 309 #ifdef notyet 310 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 311 #else 312 va.va_atime.tv_sec = 313 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 314 va.va_atime.tv_nsec = 0; 315 #endif 316 } 317 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 318 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 319 320 } 321 322 /* 323 * Now that we have all the fields, lets do it. 324 */ 325 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 326 (nfsd->nd_flag & ND_KERBAUTH), false); 327 if (error) { 328 nfsm_reply(2 * NFSX_UNSIGNED); 329 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 330 return (0); 331 } 332 nqsrv_getl(vp, ND_WRITE); 333 if (v3) { 334 error = preat_ret = VOP_GETATTR(vp, &preat, cred); 335 if (!error && gcheck && 336 (preat.va_ctime.tv_sec != guard.tv_sec || 337 preat.va_ctime.tv_nsec != guard.tv_nsec)) 338 error = NFSERR_NOT_SYNC; 339 if (error) { 340 vput(vp); 341 nfsm_reply(NFSX_WCCDATA(v3)); 342 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 343 return (0); 344 } 345 } 346 347 /* 348 * If the size is being changed write acces is required, otherwise 349 * just check for a read only file system. 350 */ 351 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 352 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 353 error = EROFS; 354 goto out; 355 } 356 } else { 357 if (vp->v_type == VDIR) { 358 error = EISDIR; 359 goto out; 360 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 361 lwp, 0)) != 0) 362 goto out; 363 } 364 error = VOP_SETATTR(vp, &va, cred); 365 postat_ret = VOP_GETATTR(vp, &va, cred); 366 if (!error) 367 error = postat_ret; 368 out: 369 vput(vp); 370 nfsm_reply(NFSX_WCCORFATTR(v3)); 371 if (v3) { 372 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 373 return (0); 374 } else { 375 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 376 nfsm_srvfillattr(&va, fp); 377 } 378 nfsm_srvdone; 379 } 380 381 /* 382 * nfs lookup rpc 383 */ 384 int 385 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 386 { 387 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 388 struct mbuf *nam = nfsd->nd_nam; 389 char *dpos = nfsd->nd_dpos; 390 kauth_cred_t cred = nfsd->nd_cr; 391 struct nfs_fattr *fp; 392 struct nameidata nd, ind, *ndp = &nd; 393 struct pathbuf *ipb = NULL; 394 struct vnode *vp, *dirp; 395 nfsrvfh_t nsfh; 396 char *cp; 397 u_int32_t *tl; 398 int32_t t1; 399 char *bpos; 400 int error = 0, cache = 0, dirattr_ret = 1; 401 uint32_t len; 402 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 403 char *cp2; 404 struct mbuf *mb, *mreq; 405 struct vattr va, dirattr; 406 u_quad_t frev; 407 408 nfsm_srvmtofh(&nsfh); 409 nfsm_srvnamesiz(len); 410 411 pubflag = nfs_ispublicfh(&nsfh); 412 413 nd.ni_cnd.cn_cred = cred; 414 nd.ni_cnd.cn_nameiop = LOOKUP; 415 nd.ni_cnd.cn_flags = LOCKLEAF; 416 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 417 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 418 419 if (!error && pubflag) { 420 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL && 421 (ipb = pathbuf_create(nfs_pub.np_index)) != NULL) { 422 /* 423 * Setup call to lookup() to see if we can find 424 * the index file. Arguably, this doesn't belong 425 * in a kernel.. Ugh. 426 */ 427 ind = nd; 428 VOP_UNLOCK(nd.ni_vp); 429 ind.ni_pathbuf = ipb; 430 431 error = lookup_for_nfsd_index(&ind, nd.ni_vp); 432 if (!error) { 433 /* 434 * Found an index file. Get rid of 435 * the old references. 436 */ 437 if (dirp) 438 vrele(dirp); 439 dirp = nd.ni_vp; 440 ndp = &ind; 441 } else 442 error = 0; 443 } 444 /* 445 * If the public filehandle was used, check that this lookup 446 * didn't result in a filehandle outside the publicly exported 447 * filesystem. 448 */ 449 450 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 451 vput(nd.ni_vp); 452 error = EPERM; 453 } 454 } 455 456 if (dirp) { 457 if (v3) 458 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred); 459 vrele(dirp); 460 } 461 462 if (error) { 463 if (nd.ni_pathbuf != NULL) { 464 pathbuf_destroy(nd.ni_pathbuf); 465 } 466 if (ipb != NULL) { 467 pathbuf_destroy(ipb); 468 } 469 nfsm_reply(NFSX_POSTOPATTR(v3)); 470 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 471 return (0); 472 } 473 474 nqsrv_getl(ndp->ni_startdir, ND_READ); 475 pathbuf_destroy(nd.ni_pathbuf); 476 if (ipb != NULL) { 477 pathbuf_destroy(ipb); 478 } 479 vp = ndp->ni_vp; 480 error = nfsrv_composefh(vp, &nsfh, v3); 481 if (!error) 482 error = VOP_GETATTR(vp, &va, cred); 483 vput(vp); 484 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) + 485 NFSX_POSTOPATTR(v3)); 486 if (error) { 487 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 488 return (0); 489 } 490 nfsm_srvfhtom(&nsfh, v3); 491 if (v3) { 492 nfsm_srvpostop_attr(0, &va); 493 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 494 } else { 495 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 496 nfsm_srvfillattr(&va, fp); 497 } 498 nfsm_srvdone; 499 } 500 501 /* 502 * nfs readlink service 503 */ 504 int 505 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 506 { 507 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 508 struct mbuf *nam = nfsd->nd_nam; 509 char *dpos = nfsd->nd_dpos; 510 kauth_cred_t cred = nfsd->nd_cr; 511 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 512 struct iovec *ivp = iv; 513 struct mbuf *mp; 514 u_int32_t *tl; 515 int32_t t1; 516 char *bpos; 517 int error = 0, rdonly, cache = 0, i, padlen, getret; 518 uint32_t len; 519 int v3 = (nfsd->nd_flag & ND_NFSV3); 520 char *cp2; 521 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 522 struct vnode *vp; 523 struct vattr attr; 524 nfsrvfh_t nsfh; 525 struct uio io, *uiop = &io; 526 u_quad_t frev; 527 528 nfsm_srvmtofh(&nsfh); 529 len = 0; 530 i = 0; 531 while (len < NFS_MAXPATHLEN) { 532 mp = m_get(M_WAIT, MT_DATA); 533 MCLAIM(mp, &nfs_mowner); 534 m_clget(mp, M_WAIT); 535 mp->m_len = NFSMSIZ(mp); 536 if (len == 0) 537 mp3 = mp2 = mp; 538 else { 539 mp2->m_next = mp; 540 mp2 = mp; 541 } 542 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 543 mp->m_len = NFS_MAXPATHLEN-len; 544 len = NFS_MAXPATHLEN; 545 } else 546 len += mp->m_len; 547 ivp->iov_base = mtod(mp, void *); 548 ivp->iov_len = mp->m_len; 549 i++; 550 ivp++; 551 } 552 uiop->uio_iov = iv; 553 uiop->uio_iovcnt = i; 554 uiop->uio_offset = 0; 555 uiop->uio_resid = len; 556 uiop->uio_rw = UIO_READ; 557 UIO_SETUP_SYSSPACE(uiop); 558 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 559 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 560 if (error) { 561 m_freem(mp3); 562 nfsm_reply(2 * NFSX_UNSIGNED); 563 nfsm_srvpostop_attr(1, (struct vattr *)0); 564 return (0); 565 } 566 if (vp->v_type != VLNK) { 567 if (v3) 568 error = EINVAL; 569 else 570 error = ENXIO; 571 goto out; 572 } 573 nqsrv_getl(vp, ND_READ); 574 error = VOP_READLINK(vp, uiop, cred); 575 out: 576 getret = VOP_GETATTR(vp, &attr, cred); 577 vput(vp); 578 if (error) 579 m_freem(mp3); 580 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 581 if (v3) { 582 nfsm_srvpostop_attr(getret, &attr); 583 if (error) 584 return (0); 585 } 586 len -= uiop->uio_resid; 587 padlen = nfsm_padlen(len); 588 if (uiop->uio_resid || padlen) 589 nfs_zeropad(mp3, uiop->uio_resid, padlen); 590 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 591 *tl = txdr_unsigned(len); 592 mb->m_next = mp3; 593 nfsm_srvdone; 594 } 595 596 /* 597 * nfs read service 598 */ 599 int 600 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 601 { 602 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 603 struct mbuf *nam = nfsd->nd_nam; 604 char *dpos = nfsd->nd_dpos; 605 kauth_cred_t cred = nfsd->nd_cr; 606 struct mbuf *m; 607 struct nfs_fattr *fp; 608 u_int32_t *tl; 609 int32_t t1; 610 int i; 611 char *bpos; 612 int error = 0, rdonly, cache = 0, getret; 613 int v3 = (nfsd->nd_flag & ND_NFSV3); 614 uint32_t reqlen, len, cnt, left; 615 int padlen; 616 char *cp2; 617 struct mbuf *mb, *mreq; 618 struct vnode *vp; 619 nfsrvfh_t nsfh; 620 struct uio io, *uiop = &io; 621 struct vattr va; 622 off_t off; 623 u_quad_t frev; 624 625 nfsm_srvmtofh(&nsfh); 626 if (v3) { 627 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 628 off = fxdr_hyper(tl); 629 } else { 630 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 631 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 632 } 633 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 634 reqlen = fxdr_unsigned(uint32_t, *tl); 635 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 636 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 637 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 638 if (error) { 639 nfsm_reply(2 * NFSX_UNSIGNED); 640 nfsm_srvpostop_attr(1, (struct vattr *)0); 641 return (0); 642 } 643 if (vp->v_type != VREG) { 644 if (v3) 645 error = EINVAL; 646 else 647 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 648 } 649 if (!error) { 650 nqsrv_getl(vp, ND_READ); 651 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 652 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 653 } 654 getret = VOP_GETATTR(vp, &va, cred); 655 if (!error) 656 error = getret; 657 if (error) { 658 vput(vp); 659 nfsm_reply(NFSX_POSTOPATTR(v3)); 660 nfsm_srvpostop_attr(getret, &va); 661 return (0); 662 } 663 if (off >= va.va_size) 664 cnt = 0; 665 else if ((off + reqlen) > va.va_size) 666 cnt = va.va_size - off; 667 else 668 cnt = reqlen; 669 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 670 if (v3) { 671 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 672 *tl++ = nfs_true; 673 fp = (struct nfs_fattr *)tl; 674 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 675 } else { 676 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 677 fp = (struct nfs_fattr *)tl; 678 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 679 } 680 len = left = cnt; 681 if (cnt > 0) { 682 if (nfsd_use_loan) { 683 struct vm_page **pgpp; 684 voff_t pgoff = trunc_page(off); 685 int npages; 686 vaddr_t lva; 687 688 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 689 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 690 691 /* allocate kva for mbuf data */ 692 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 693 if (lva == 0) { 694 /* fall back to VOP_READ */ 695 goto loan_fail; 696 } 697 698 /* allocate mbuf */ 699 m = m_get(M_WAIT, MT_DATA); 700 MCLAIM(m, &nfs_mowner); 701 pgpp = m->m_ext.ext_pgs; 702 703 /* loan pages */ 704 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 705 pgpp); 706 if (error) { 707 sokvafree(lva, npages << PAGE_SHIFT); 708 m_free(m); 709 if (error == EBUSY) 710 goto loan_fail; 711 goto read_error; 712 } 713 714 /* associate kva to mbuf */ 715 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 716 cnt, M_MBUF, soloanfree, slp->ns_so); 717 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 718 m->m_len = cnt; 719 720 /* map pages */ 721 for (i = 0; i < npages; i++) { 722 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 723 VM_PROT_READ, 0); 724 lva += PAGE_SIZE; 725 } 726 727 pmap_update(pmap_kernel()); 728 729 mb->m_next = m; 730 mb = m; 731 error = 0; 732 uiop->uio_resid = 0; 733 } else { 734 struct iovec *iv; 735 struct iovec *iv2; 736 struct mbuf *m2; 737 int siz; 738 loan_fail: 739 /* 740 * Generate the mbuf list with the uio_iov ref. to it. 741 */ 742 i = 0; 743 m = m2 = mb; 744 while (left > 0) { 745 siz = min(M_TRAILINGSPACE(m), left); 746 if (siz > 0) { 747 left -= siz; 748 i++; 749 } 750 if (left > 0) { 751 m = m_get(M_WAIT, MT_DATA); 752 MCLAIM(m, &nfs_mowner); 753 m_clget(m, M_WAIT); 754 m->m_len = 0; 755 m2->m_next = m; 756 m2 = m; 757 } 758 } 759 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 760 uiop->uio_iov = iv2 = iv; 761 m = mb; 762 left = cnt; 763 i = 0; 764 while (left > 0) { 765 if (m == NULL) 766 panic("nfsrv_read iov"); 767 siz = min(M_TRAILINGSPACE(m), left); 768 if (siz > 0) { 769 iv->iov_base = mtod(m, char *) + 770 m->m_len; 771 iv->iov_len = siz; 772 m->m_len += siz; 773 left -= siz; 774 iv++; 775 i++; 776 } 777 m = m->m_next; 778 } 779 uiop->uio_iovcnt = i; 780 uiop->uio_offset = off; 781 uiop->uio_resid = cnt; 782 uiop->uio_rw = UIO_READ; 783 UIO_SETUP_SYSSPACE(uiop); 784 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 785 free((void *)iv2, M_TEMP); 786 } 787 read_error: 788 if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){ 789 if (!error) 790 error = getret; 791 m_freem(mreq); 792 vput(vp); 793 nfsm_reply(NFSX_POSTOPATTR(v3)); 794 nfsm_srvpostop_attr(getret, &va); 795 return (0); 796 } 797 } else { 798 uiop->uio_resid = 0; 799 } 800 vput(vp); 801 nfsm_srvfillattr(&va, fp); 802 len -= uiop->uio_resid; 803 padlen = nfsm_padlen(len); 804 if (uiop->uio_resid || padlen) 805 nfs_zeropad(mb, uiop->uio_resid, padlen); 806 if (v3) { 807 /* count */ 808 *tl++ = txdr_unsigned(len); 809 /* eof */ 810 if (off + len >= va.va_size) 811 *tl++ = nfs_true; 812 else 813 *tl++ = nfs_false; 814 } 815 *tl = txdr_unsigned(len); 816 nfsm_srvdone; 817 } 818 819 /* 820 * nfs write service 821 */ 822 int 823 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 824 { 825 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 826 struct mbuf *nam = nfsd->nd_nam; 827 char *dpos = nfsd->nd_dpos; 828 kauth_cred_t cred = nfsd->nd_cr; 829 struct iovec *ivp; 830 int i, cnt; 831 struct mbuf *mp; 832 struct nfs_fattr *fp; 833 struct iovec *iv; 834 struct vattr va, forat; 835 u_int32_t *tl; 836 int32_t t1; 837 char *bpos; 838 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 839 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 840 int stable = NFSV3WRITE_FILESYNC; 841 int v3 = (nfsd->nd_flag & ND_NFSV3); 842 char *cp2; 843 struct mbuf *mb, *mreq; 844 struct vnode *vp; 845 nfsrvfh_t nsfh; 846 struct uio io, *uiop = &io; 847 off_t off; 848 u_quad_t frev; 849 850 if (mrep == NULL) { 851 *mrq = NULL; 852 return (0); 853 } 854 nfsm_srvmtofh(&nsfh); 855 if (v3) { 856 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 857 off = fxdr_hyper(tl); 858 tl += 3; 859 stable = fxdr_unsigned(int, *tl++); 860 } else { 861 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 862 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 863 tl += 2; 864 } 865 retlen = len = fxdr_unsigned(int32_t, *tl); 866 cnt = i = 0; 867 868 /* 869 * For NFS Version 2, it is not obvious what a write of zero length 870 * should do, but I might as well be consistent with Version 3, 871 * which is to return ok so long as there are no permission problems. 872 */ 873 if (len > 0) { 874 zeroing = 1; 875 mp = mrep; 876 while (mp) { 877 if (mp == md) { 878 zeroing = 0; 879 adjust = dpos - mtod(mp, char *); 880 mp->m_len -= adjust; 881 if (mp->m_len > 0 && adjust > 0) 882 NFSMADV(mp, adjust); 883 } 884 if (zeroing) 885 mp->m_len = 0; 886 else if (mp->m_len > 0) { 887 i += mp->m_len; 888 if (i > len) { 889 mp->m_len -= (i - len); 890 zeroing = 1; 891 } 892 if (mp->m_len > 0) 893 cnt++; 894 } 895 mp = mp->m_next; 896 } 897 } 898 if (len > NFS_MAXDATA || len < 0 || i < len) { 899 error = EIO; 900 nfsm_reply(2 * NFSX_UNSIGNED); 901 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 902 return (0); 903 } 904 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 905 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 906 if (error) { 907 nfsm_reply(2 * NFSX_UNSIGNED); 908 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 909 return (0); 910 } 911 if (v3) 912 forat_ret = VOP_GETATTR(vp, &forat, cred); 913 if (vp->v_type != VREG) { 914 if (v3) 915 error = EINVAL; 916 else 917 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 918 } 919 if (!error) { 920 nqsrv_getl(vp, ND_WRITE); 921 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 922 } 923 if (error) { 924 vput(vp); 925 nfsm_reply(NFSX_WCCDATA(v3)); 926 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 927 return (0); 928 } 929 930 if (len > 0) { 931 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 932 uiop->uio_iov = iv = ivp; 933 uiop->uio_iovcnt = cnt; 934 mp = mrep; 935 while (mp) { 936 if (mp->m_len > 0) { 937 ivp->iov_base = mtod(mp, void *); 938 ivp->iov_len = mp->m_len; 939 ivp++; 940 } 941 mp = mp->m_next; 942 } 943 944 /* 945 * XXX 946 * The IO_METASYNC flag indicates that all metadata (and not 947 * just enough to ensure data integrity) must be written to 948 * stable storage synchronously. 949 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 950 */ 951 if (stable == NFSV3WRITE_UNSTABLE) 952 ioflags = IO_NODELOCKED; 953 else if (stable == NFSV3WRITE_DATASYNC) 954 ioflags = (IO_SYNC | IO_NODELOCKED); 955 else 956 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 957 uiop->uio_resid = len; 958 uiop->uio_rw = UIO_WRITE; 959 uiop->uio_offset = off; 960 UIO_SETUP_SYSSPACE(uiop); 961 error = VOP_WRITE(vp, uiop, ioflags, cred); 962 nfsstats.srvvop_writes++; 963 free(iv, M_TEMP); 964 } 965 aftat_ret = VOP_GETATTR(vp, &va, cred); 966 vput(vp); 967 if (!error) 968 error = aftat_ret; 969 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 970 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 971 if (v3) { 972 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 973 if (error) 974 return (0); 975 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 976 *tl++ = txdr_unsigned(retlen); 977 if (stable == NFSV3WRITE_UNSTABLE) 978 *tl++ = txdr_unsigned(stable); 979 else 980 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 981 /* 982 * Actually, there is no need to txdr these fields, 983 * but it may make the values more human readable, 984 * for debugging purposes. 985 */ 986 *tl++ = txdr_unsigned(boottime.tv_sec); 987 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 988 } else { 989 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 990 nfsm_srvfillattr(&va, fp); 991 } 992 nfsm_srvdone; 993 } 994 995 /* 996 * XXX elad: the original NFSW_SAMECRED() macro also made sure the 997 * two nd_flag fields of the descriptors contained 998 * ND_KERBAUTH. 999 */ 1000 static int 1001 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2) 1002 { 1003 int i, do_ngroups; 1004 1005 if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2)) 1006 return (0); 1007 if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2)) 1008 return (0); 1009 do_ngroups = kauth_cred_ngroups(cred1); 1010 for (i = 0; i < do_ngroups; i++) 1011 if (kauth_cred_group(cred1, i) != 1012 kauth_cred_group(cred2, i)) 1013 return (0); 1014 1015 return (1); 1016 } 1017 1018 static struct nfsrvw_delayhash * 1019 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh) 1020 { 1021 uint32_t hash; 1022 1023 hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh), 1024 HASH32_BUF_INIT); 1025 return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ]; 1026 } 1027 1028 /* 1029 * NFS write service with write gathering support. Called when 1030 * nfsrvw_procrastinate > 0. 1031 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 1032 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 1033 * Jan. 1994. 1034 */ 1035 int 1036 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1037 { 1038 struct timeval now; 1039 struct iovec *ivp; 1040 struct mbuf *mp; 1041 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1042 struct nfs_fattr *fp; 1043 int i = 0; 1044 struct iovec *iov; 1045 struct nfsrvw_delayhash *wpp; 1046 kauth_cred_t cred; 1047 struct vattr va, forat; 1048 u_int32_t *tl; 1049 int32_t t1; 1050 char *bpos, *dpos; 1051 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1052 int ioflags, aftat_ret = 1, adjust, v3, zeroing; 1053 char *cp2; 1054 struct mbuf *mb, *mreq, *mrep, *md; 1055 struct vnode *vp; 1056 struct uio io, *uiop = &io; 1057 u_quad_t frev, cur_usec; 1058 1059 *mrq = NULL; 1060 if (*ndp) { 1061 nfsd = *ndp; 1062 *ndp = NULL; 1063 mrep = nfsd->nd_mrep; 1064 md = nfsd->nd_md; 1065 dpos = nfsd->nd_dpos; 1066 cred = nfsd->nd_cr; 1067 v3 = (nfsd->nd_flag & ND_NFSV3); 1068 LIST_INIT(&nfsd->nd_coalesce); 1069 nfsd->nd_mreq = NULL; 1070 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1071 getmicrotime(&now); 1072 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1073 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1074 1075 /* 1076 * Now, get the write header.. 1077 */ 1078 nfsm_srvmtofh(&nfsd->nd_fh); 1079 if (v3) { 1080 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1081 nfsd->nd_off = fxdr_hyper(tl); 1082 tl += 3; 1083 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1084 } else { 1085 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1086 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1087 tl += 2; 1088 } 1089 len = fxdr_unsigned(int32_t, *tl); 1090 nfsd->nd_len = len; 1091 nfsd->nd_eoff = nfsd->nd_off + len; 1092 1093 /* 1094 * Trim the header out of the mbuf list and trim off any trailing 1095 * junk so that the mbuf list has only the write data. 1096 */ 1097 zeroing = 1; 1098 i = 0; 1099 mp = mrep; 1100 while (mp) { 1101 if (mp == md) { 1102 zeroing = 0; 1103 adjust = dpos - mtod(mp, char *); 1104 mp->m_len -= adjust; 1105 if (mp->m_len > 0 && adjust > 0) 1106 NFSMADV(mp, adjust); 1107 } 1108 if (zeroing) 1109 mp->m_len = 0; 1110 else { 1111 i += mp->m_len; 1112 if (i > len) { 1113 mp->m_len -= (i - len); 1114 zeroing = 1; 1115 } 1116 } 1117 mp = mp->m_next; 1118 } 1119 if (len > NFS_MAXDATA || len < 0 || i < len) { 1120 nfsmout: 1121 m_freem(mrep); 1122 error = EIO; 1123 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1124 if (v3) 1125 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1126 nfsd->nd_mreq = mreq; 1127 nfsd->nd_mrep = NULL; 1128 nfsd->nd_time = 0; 1129 } 1130 1131 /* 1132 * Add this entry to the hash and time queues. 1133 */ 1134 owp = NULL; 1135 mutex_enter(&nfsd_lock); 1136 wp = LIST_FIRST(&slp->ns_tq); 1137 while (wp && wp->nd_time < nfsd->nd_time) { 1138 owp = wp; 1139 wp = LIST_NEXT(wp, nd_tq); 1140 } 1141 if (owp) { 1142 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1143 } else { 1144 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1145 } 1146 if (nfsd->nd_mrep) { 1147 wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh); 1148 owp = NULL; 1149 wp = LIST_FIRST(wpp); 1150 while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1151 owp = wp; 1152 wp = LIST_NEXT(wp, nd_hash); 1153 } 1154 while (wp && wp->nd_off < nfsd->nd_off && 1155 !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1156 owp = wp; 1157 wp = LIST_NEXT(wp, nd_hash); 1158 } 1159 if (owp) { 1160 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1161 1162 /* 1163 * Search the hash list for overlapping entries and 1164 * coalesce. 1165 */ 1166 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1167 wp = LIST_NEXT(nfsd, nd_hash); 1168 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr)) 1169 nfsrvw_coalesce(owp, nfsd); 1170 } 1171 } else { 1172 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1173 } 1174 } 1175 mutex_exit(&nfsd_lock); 1176 } 1177 1178 /* 1179 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1180 * and generate the associated reply mbuf list(s). 1181 */ 1182 loop1: 1183 getmicrotime(&now); 1184 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1185 mutex_enter(&nfsd_lock); 1186 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1187 owp = LIST_NEXT(nfsd, nd_tq); 1188 if (nfsd->nd_time > cur_usec) 1189 break; 1190 if (nfsd->nd_mreq) 1191 continue; 1192 LIST_REMOVE(nfsd, nd_tq); 1193 LIST_REMOVE(nfsd, nd_hash); 1194 mutex_exit(&nfsd_lock); 1195 1196 mrep = nfsd->nd_mrep; 1197 nfsd->nd_mrep = NULL; 1198 cred = nfsd->nd_cr; 1199 v3 = (nfsd->nd_flag & ND_NFSV3); 1200 forat_ret = aftat_ret = 1; 1201 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1202 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1203 false); 1204 if (!error) { 1205 if (v3) 1206 forat_ret = VOP_GETATTR(vp, &forat, cred); 1207 if (vp->v_type != VREG) { 1208 if (v3) 1209 error = EINVAL; 1210 else 1211 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1212 } 1213 } else 1214 vp = NULL; 1215 if (!error) { 1216 nqsrv_getl(vp, ND_WRITE); 1217 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1218 } 1219 1220 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1221 ioflags = IO_NODELOCKED; 1222 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1223 ioflags = (IO_SYNC | IO_NODELOCKED); 1224 else 1225 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1226 uiop->uio_rw = UIO_WRITE; 1227 uiop->uio_offset = nfsd->nd_off; 1228 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1229 UIO_SETUP_SYSSPACE(uiop); 1230 if (uiop->uio_resid > 0) { 1231 mp = mrep; 1232 i = 0; 1233 while (mp) { 1234 if (mp->m_len > 0) 1235 i++; 1236 mp = mp->m_next; 1237 } 1238 uiop->uio_iovcnt = i; 1239 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1240 uiop->uio_iov = ivp = iov; 1241 mp = mrep; 1242 while (mp) { 1243 if (mp->m_len > 0) { 1244 ivp->iov_base = mtod(mp, void *); 1245 ivp->iov_len = mp->m_len; 1246 ivp++; 1247 } 1248 mp = mp->m_next; 1249 } 1250 if (!error) { 1251 error = VOP_WRITE(vp, uiop, ioflags, cred); 1252 nfsstats.srvvop_writes++; 1253 } 1254 free((void *)iov, M_TEMP); 1255 } 1256 m_freem(mrep); 1257 if (vp) { 1258 aftat_ret = VOP_GETATTR(vp, &va, cred); 1259 vput(vp); 1260 } 1261 1262 /* 1263 * Loop around generating replies for all write rpcs that have 1264 * now been completed. 1265 */ 1266 swp = nfsd; 1267 do { 1268 if (error) { 1269 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1270 if (v3) { 1271 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1272 } 1273 } else { 1274 nfsm_writereply(NFSX_PREOPATTR(v3) + 1275 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1276 NFSX_WRITEVERF(v3), v3); 1277 if (v3) { 1278 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1279 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1280 *tl++ = txdr_unsigned(nfsd->nd_len); 1281 *tl++ = txdr_unsigned(swp->nd_stable); 1282 /* 1283 * Actually, there is no need to txdr these fields, 1284 * but it may make the values more human readable, 1285 * for debugging purposes. 1286 */ 1287 *tl++ = txdr_unsigned(boottime.tv_sec); 1288 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 1289 } else { 1290 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1291 nfsm_srvfillattr(&va, fp); 1292 } 1293 } 1294 nfsd->nd_mreq = mreq; 1295 if (nfsd->nd_mrep) 1296 panic("nfsrv_write: nd_mrep not free"); 1297 1298 /* 1299 * Done. Put it at the head of the timer queue so that 1300 * the final phase can return the reply. 1301 */ 1302 mutex_enter(&nfsd_lock); 1303 if (nfsd != swp) { 1304 nfsd->nd_time = 0; 1305 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1306 } 1307 nfsd = LIST_FIRST(&swp->nd_coalesce); 1308 if (nfsd) { 1309 LIST_REMOVE(nfsd, nd_tq); 1310 } 1311 mutex_exit(&nfsd_lock); 1312 } while (nfsd); 1313 swp->nd_time = 0; 1314 1315 mutex_enter(&nfsd_lock); 1316 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1317 mutex_exit(&nfsd_lock); 1318 goto loop1; 1319 } 1320 mutex_exit(&nfsd_lock); 1321 nfs_timer_start(); 1322 1323 /* 1324 * Search for a reply to return. 1325 */ 1326 mutex_enter(&nfsd_lock); 1327 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1328 if (nfsd->nd_mreq) { 1329 LIST_REMOVE(nfsd, nd_tq); 1330 *mrq = nfsd->nd_mreq; 1331 *ndp = nfsd; 1332 break; 1333 } 1334 } 1335 mutex_exit(&nfsd_lock); 1336 return (0); 1337 } 1338 1339 /* 1340 * Coalesce the write request nfsd into owp. To do this we must: 1341 * - remove nfsd from the queues 1342 * - merge nfsd->nd_mrep into owp->nd_mrep 1343 * - update the nd_eoff and nd_stable for owp 1344 * - put nfsd on owp's nd_coalesce list 1345 * NB: Must be called at splsoftclock(). 1346 */ 1347 void 1348 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd) 1349 { 1350 int overlap; 1351 struct mbuf *mp; 1352 struct nfsrv_descript *m; 1353 1354 KASSERT(mutex_owned(&nfsd_lock)); 1355 1356 LIST_REMOVE(nfsd, nd_hash); 1357 LIST_REMOVE(nfsd, nd_tq); 1358 if (owp->nd_eoff < nfsd->nd_eoff) { 1359 overlap = owp->nd_eoff - nfsd->nd_off; 1360 if (overlap < 0) 1361 panic("nfsrv_coalesce: bad off"); 1362 if (overlap > 0) 1363 m_adj(nfsd->nd_mrep, overlap); 1364 mp = owp->nd_mrep; 1365 while (mp->m_next) 1366 mp = mp->m_next; 1367 mp->m_next = nfsd->nd_mrep; 1368 owp->nd_eoff = nfsd->nd_eoff; 1369 } else 1370 m_freem(nfsd->nd_mrep); 1371 nfsd->nd_mrep = NULL; 1372 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1373 owp->nd_stable = NFSV3WRITE_FILESYNC; 1374 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1375 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1376 owp->nd_stable = NFSV3WRITE_DATASYNC; 1377 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1378 /* 1379 * nfsd might hold coalesce elements! Move them to owp. 1380 * Otherwise, requests may be lost and clients will be stuck. 1381 */ 1382 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1383 LIST_REMOVE(m, nd_tq); 1384 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1385 } 1386 } 1387 1388 /* 1389 * nfs create service 1390 * now does a truncate to 0 length via. setattr if it already exists 1391 */ 1392 int 1393 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1394 { 1395 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1396 struct mbuf *nam = nfsd->nd_nam; 1397 char *dpos = nfsd->nd_dpos; 1398 kauth_cred_t cred = nfsd->nd_cr; 1399 struct nfs_fattr *fp; 1400 struct vattr va, dirfor, diraft; 1401 struct nfsv2_sattr *sp; 1402 u_int32_t *tl; 1403 struct nameidata nd; 1404 char *cp; 1405 int32_t t1; 1406 char *bpos; 1407 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1408 int rdev = 0, abort = 0; 1409 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1410 char *cp2; 1411 struct mbuf *mb, *mreq; 1412 struct vnode *vp = NULL, *dirp = NULL; 1413 nfsrvfh_t nsfh; 1414 u_quad_t frev, tempsize; 1415 u_char cverf[NFSX_V3CREATEVERF]; 1416 1417 nd.ni_cnd.cn_nameiop = 0; 1418 nfsm_srvmtofh(&nsfh); 1419 nfsm_srvnamesiz(len); 1420 nd.ni_cnd.cn_cred = cred; 1421 nd.ni_cnd.cn_nameiop = CREATE; 1422 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1423 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1424 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1425 if (dirp && v3) { 1426 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1427 } 1428 if (error) { 1429 nfsm_reply(NFSX_WCCDATA(v3)); 1430 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1431 if (dirp) 1432 vrele(dirp); 1433 if (nd.ni_pathbuf != NULL) { 1434 pathbuf_destroy(nd.ni_pathbuf); 1435 nd.ni_pathbuf = NULL; 1436 } 1437 return (0); 1438 } 1439 abort = 1; 1440 vattr_null(&va); 1441 if (v3) { 1442 va.va_mode = 0; 1443 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1444 how = fxdr_unsigned(int, *tl); 1445 switch (how) { 1446 case NFSV3CREATE_GUARDED: 1447 if (nd.ni_vp) { 1448 error = EEXIST; 1449 break; 1450 } 1451 case NFSV3CREATE_UNCHECKED: 1452 nfsm_srvsattr(&va); 1453 break; 1454 case NFSV3CREATE_EXCLUSIVE: 1455 nfsm_dissect(cp, void *, NFSX_V3CREATEVERF); 1456 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1457 exclusive_flag = 1; 1458 break; 1459 }; 1460 va.va_type = VREG; 1461 } else { 1462 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1463 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1464 if (va.va_type == VNON) 1465 va.va_type = VREG; 1466 va.va_mode = nfstov_mode(sp->sa_mode); 1467 switch (va.va_type) { 1468 case VREG: 1469 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1470 if (tsize != -1) 1471 va.va_size = (u_quad_t)tsize; 1472 break; 1473 case VCHR: 1474 case VBLK: 1475 case VFIFO: 1476 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1477 break; 1478 default: 1479 break; 1480 }; 1481 } 1482 1483 /* 1484 * Iff doesn't exist, create it 1485 * otherwise just truncate to 0 length 1486 * should I set the mode too ?? 1487 */ 1488 if (nd.ni_vp == NULL) { 1489 if (va.va_type == VREG || va.va_type == VSOCK) { 1490 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1491 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1492 if (!error) { 1493 if (exclusive_flag) { 1494 exclusive_flag = 0; 1495 vattr_null(&va); 1496 /* 1497 * XXX 1498 * assuming NFSX_V3CREATEVERF 1499 * == sizeof(nfstime3) 1500 */ 1501 fxdr_nfsv3time(cverf, &va.va_atime); 1502 error = VOP_SETATTR(nd.ni_vp, &va, 1503 cred); 1504 } 1505 } 1506 } else if (va.va_type == VCHR || va.va_type == VBLK || 1507 va.va_type == VFIFO) { 1508 if (va.va_type == VCHR && rdev == 0xffffffff) 1509 va.va_type = VFIFO; 1510 if (va.va_type != VFIFO && 1511 (error = kauth_authorize_system(cred, 1512 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1513 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1514 vput(nd.ni_dvp); 1515 abort = 0; 1516 nfsm_reply(0); 1517 if (nd.ni_pathbuf != NULL) { 1518 pathbuf_destroy(nd.ni_pathbuf); 1519 nd.ni_pathbuf = NULL; 1520 } 1521 return (error); 1522 } else 1523 va.va_rdev = (dev_t)rdev; 1524 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1525 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1526 &va); 1527 if (error) { 1528 nfsm_reply(0); 1529 } 1530 } else { 1531 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1532 if (nd.ni_pathbuf != NULL) { 1533 pathbuf_destroy(nd.ni_pathbuf); 1534 nd.ni_pathbuf = NULL; 1535 } 1536 vput(nd.ni_dvp); 1537 error = ENXIO; 1538 abort = 0; 1539 } 1540 vp = nd.ni_vp; 1541 } else { 1542 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1543 if (nd.ni_pathbuf != NULL) { 1544 pathbuf_destroy(nd.ni_pathbuf); 1545 nd.ni_pathbuf = NULL; 1546 } 1547 vp = nd.ni_vp; 1548 if (nd.ni_dvp == vp) 1549 vrele(nd.ni_dvp); 1550 else 1551 vput(nd.ni_dvp); 1552 abort = 0; 1553 if (!error && va.va_size != -1) { 1554 error = nfsrv_access(vp, VWRITE, cred, 1555 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1556 if (!error) { 1557 nqsrv_getl(vp, ND_WRITE); 1558 tempsize = va.va_size; 1559 vattr_null(&va); 1560 va.va_size = tempsize; 1561 error = VOP_SETATTR(vp, &va, cred); 1562 } 1563 } 1564 if (error) 1565 vput(vp); 1566 } 1567 if (!error) { 1568 error = nfsrv_composefh(vp, &nsfh, v3); 1569 if (!error) 1570 error = VOP_GETATTR(vp, &va, cred); 1571 vput(vp); 1572 } 1573 if (v3) { 1574 if (exclusive_flag && !error) { 1575 /* 1576 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1577 */ 1578 char oldverf[NFSX_V3CREATEVERF]; 1579 1580 txdr_nfsv3time(&va.va_atime, oldverf); 1581 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1582 error = EEXIST; 1583 } 1584 if (dirp) { 1585 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1586 } 1587 } 1588 if (dirp) { 1589 vrele(dirp); 1590 dirp = NULL; 1591 } 1592 if (nd.ni_pathbuf != NULL) { 1593 pathbuf_destroy(nd.ni_pathbuf); 1594 nd.ni_pathbuf = NULL; 1595 } 1596 abort = 0; 1597 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1598 if (v3) { 1599 if (!error) { 1600 nfsm_srvpostop_fh(&nsfh); 1601 nfsm_srvpostop_attr(0, &va); 1602 } 1603 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1604 } else { 1605 nfsm_srvfhtom(&nsfh, v3); 1606 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1607 nfsm_srvfillattr(&va, fp); 1608 } 1609 return (0); 1610 nfsmout: 1611 if (dirp) 1612 vrele(dirp); 1613 if (abort) { 1614 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1615 if (nd.ni_dvp == nd.ni_vp) 1616 vrele(nd.ni_dvp); 1617 else 1618 vput(nd.ni_dvp); 1619 if (nd.ni_vp) 1620 vput(nd.ni_vp); 1621 } 1622 if (nd.ni_pathbuf != NULL) { 1623 pathbuf_destroy(nd.ni_pathbuf); 1624 nd.ni_pathbuf = NULL; 1625 } 1626 return (error); 1627 } 1628 1629 /* 1630 * nfs v3 mknod service 1631 */ 1632 int 1633 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1634 { 1635 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1636 struct mbuf *nam = nfsd->nd_nam; 1637 char *dpos = nfsd->nd_dpos; 1638 kauth_cred_t cred = nfsd->nd_cr; 1639 struct vattr va, dirfor, diraft; 1640 u_int32_t *tl; 1641 struct nameidata nd; 1642 int32_t t1; 1643 char *bpos; 1644 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1645 int abort = 0; 1646 u_int32_t major, minor; 1647 enum vtype vtyp; 1648 char *cp2; 1649 struct mbuf *mb, *mreq; 1650 struct vnode *vp, *dirp = (struct vnode *)0; 1651 nfsrvfh_t nsfh; 1652 u_quad_t frev; 1653 1654 nd.ni_cnd.cn_nameiop = 0; 1655 nfsm_srvmtofh(&nsfh); 1656 nfsm_srvnamesiz(len); 1657 nd.ni_cnd.cn_cred = cred; 1658 nd.ni_cnd.cn_nameiop = CREATE; 1659 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1660 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1661 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1662 if (dirp) 1663 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1664 if (error) { 1665 nfsm_reply(NFSX_WCCDATA(1)); 1666 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1667 if (dirp) 1668 vrele(dirp); 1669 if (nd.ni_pathbuf != NULL) { 1670 pathbuf_destroy(nd.ni_pathbuf); 1671 nd.ni_pathbuf = NULL; 1672 } 1673 return (0); 1674 } 1675 abort = 1; 1676 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1677 vtyp = nfsv3tov_type(*tl); 1678 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1679 error = NFSERR_BADTYPE; 1680 goto abort; 1681 } 1682 vattr_null(&va); 1683 va.va_mode = 0; 1684 nfsm_srvsattr(&va); 1685 if (vtyp == VCHR || vtyp == VBLK) { 1686 dev_t rdev; 1687 1688 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1689 major = fxdr_unsigned(u_int32_t, *tl++); 1690 minor = fxdr_unsigned(u_int32_t, *tl); 1691 rdev = makedev(major, minor); 1692 if (major(rdev) != major || minor(rdev) != minor) { 1693 error = EINVAL; 1694 goto abort; 1695 } 1696 va.va_rdev = rdev; 1697 } 1698 1699 /* 1700 * Iff doesn't exist, create it. 1701 */ 1702 if (nd.ni_vp) { 1703 error = EEXIST; 1704 abort: 1705 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1706 if (nd.ni_dvp == nd.ni_vp) 1707 vrele(nd.ni_dvp); 1708 else 1709 vput(nd.ni_dvp); 1710 if (nd.ni_vp) 1711 vput(nd.ni_vp); 1712 if (nd.ni_pathbuf != NULL) { 1713 pathbuf_destroy(nd.ni_pathbuf); 1714 nd.ni_pathbuf = NULL; 1715 } 1716 goto out; 1717 } 1718 va.va_type = vtyp; 1719 if (vtyp == VSOCK) { 1720 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1721 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1722 } else { 1723 if (va.va_type != VFIFO && 1724 (error = kauth_authorize_system(cred, 1725 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1726 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1727 vput(nd.ni_dvp); 1728 goto out; 1729 } 1730 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1731 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1732 if (error) 1733 goto out; 1734 } 1735 out: 1736 vp = nd.ni_vp; 1737 if (!error) { 1738 error = nfsrv_composefh(vp, &nsfh, true); 1739 if (!error) 1740 error = VOP_GETATTR(vp, &va, cred); 1741 vput(vp); 1742 } 1743 if (dirp) { 1744 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1745 vrele(dirp); 1746 dirp = NULL; 1747 } 1748 if (nd.ni_pathbuf != NULL) { 1749 pathbuf_destroy(nd.ni_pathbuf); 1750 nd.ni_pathbuf = NULL; 1751 } 1752 abort = 0; 1753 nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) + 1754 NFSX_WCCDATA(1)); 1755 if (!error) { 1756 nfsm_srvpostop_fh(&nsfh); 1757 nfsm_srvpostop_attr(0, &va); 1758 } 1759 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1760 return (0); 1761 nfsmout: 1762 if (abort) { 1763 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1764 if (nd.ni_dvp == nd.ni_vp) 1765 vrele(nd.ni_dvp); 1766 else 1767 vput(nd.ni_dvp); 1768 if (nd.ni_vp) 1769 vput(nd.ni_vp); 1770 } 1771 if (nd.ni_pathbuf != NULL) { 1772 pathbuf_destroy(nd.ni_pathbuf); 1773 nd.ni_pathbuf = NULL; 1774 } 1775 if (dirp) 1776 vrele(dirp); 1777 return (error); 1778 } 1779 1780 /* 1781 * nfs remove service 1782 */ 1783 int 1784 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1785 { 1786 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1787 struct mbuf *nam = nfsd->nd_nam; 1788 char *dpos = nfsd->nd_dpos; 1789 kauth_cred_t cred = nfsd->nd_cr; 1790 struct nameidata nd; 1791 u_int32_t *tl; 1792 int32_t t1; 1793 char *bpos; 1794 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1795 int v3 = (nfsd->nd_flag & ND_NFSV3); 1796 char *cp2; 1797 struct mbuf *mb, *mreq; 1798 struct vnode *vp, *dirp; 1799 struct vattr dirfor, diraft; 1800 nfsrvfh_t nsfh; 1801 u_quad_t frev; 1802 1803 #ifndef nolint 1804 vp = (struct vnode *)0; 1805 #endif 1806 nfsm_srvmtofh(&nsfh); 1807 nfsm_srvnamesiz(len); 1808 nd.ni_cnd.cn_cred = cred; 1809 nd.ni_cnd.cn_nameiop = DELETE; 1810 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1811 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1812 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1813 if (dirp && v3) { 1814 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1815 } 1816 if (!error) { 1817 vp = nd.ni_vp; 1818 if (vp->v_type == VDIR) { 1819 error = EPERM; 1820 goto out; 1821 } 1822 /* 1823 * The root of a mounted filesystem cannot be deleted. 1824 */ 1825 if (vp->v_vflag & VV_ROOT) { 1826 error = EBUSY; 1827 } 1828 out: 1829 if (!error) { 1830 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1831 nqsrv_getl(vp, ND_WRITE); 1832 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1833 } else { 1834 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1835 if (nd.ni_dvp == vp) 1836 vrele(nd.ni_dvp); 1837 else 1838 vput(nd.ni_dvp); 1839 vput(vp); 1840 } 1841 } 1842 if (nd.ni_pathbuf != NULL) { 1843 pathbuf_destroy(nd.ni_pathbuf); 1844 nd.ni_pathbuf = NULL; 1845 } 1846 if (dirp) { 1847 if (v3) { 1848 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1849 } 1850 vrele(dirp); 1851 } 1852 nfsm_reply(NFSX_WCCDATA(v3)); 1853 if (v3) { 1854 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1855 return (0); 1856 } 1857 nfsm_srvdone; 1858 } 1859 1860 /* 1861 * nfs rename service 1862 */ 1863 int 1864 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1865 { 1866 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1867 struct mbuf *nam = nfsd->nd_nam; 1868 char *dpos = nfsd->nd_dpos; 1869 kauth_cred_t cred = nfsd->nd_cr; 1870 u_int32_t *tl; 1871 int32_t t1; 1872 char *bpos; 1873 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1874 uint32_t len, len2; 1875 int tdirfor_ret = 1, tdiraft_ret = 1; 1876 int v3 = (nfsd->nd_flag & ND_NFSV3); 1877 char *cp2; 1878 struct mbuf *mb, *mreq; 1879 struct nameidata fromnd, tond; 1880 struct vnode *fvp, *tvp, *tdvp; 1881 struct vnode *fdirp = NULL, *tdirp = NULL; 1882 struct mount *localfs = NULL; 1883 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1884 nfsrvfh_t fnsfh, tnsfh; 1885 u_quad_t frev; 1886 uid_t saved_uid; 1887 1888 #ifndef nolint 1889 fvp = (struct vnode *)0; 1890 #endif 1891 fromnd.ni_cnd.cn_nameiop = 0; 1892 tond.ni_cnd.cn_nameiop = 0; 1893 nfsm_srvmtofh(&fnsfh); 1894 nfsm_srvnamesiz(len); 1895 /* 1896 * Remember our original uid so that we can reset cr_uid before 1897 * the second nfs_namei() call, in case it is remapped. 1898 */ 1899 saved_uid = kauth_cred_geteuid(cred); 1900 fromnd.ni_cnd.cn_cred = cred; 1901 fromnd.ni_cnd.cn_nameiop = DELETE; 1902 fromnd.ni_cnd.cn_flags = LOCKPARENT | INRENAME; 1903 error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md, 1904 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1905 if (fdirp && v3) { 1906 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred); 1907 } 1908 if (error) { 1909 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1910 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1911 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1912 if (fdirp) 1913 vrele(fdirp); 1914 pathbuf_destroy(fromnd.ni_pathbuf); 1915 return (0); 1916 } 1917 if (fromnd.ni_dvp != fromnd.ni_vp) { 1918 VOP_UNLOCK(fromnd.ni_dvp); 1919 } 1920 fvp = fromnd.ni_vp; 1921 1922 localfs = fvp->v_mount; 1923 error = VFS_RENAMELOCK_ENTER(localfs); 1924 if (error) { 1925 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1926 vrele(fromnd.ni_dvp); 1927 vrele(fvp); 1928 goto out1; 1929 } 1930 1931 /* Copied, regrettably, from vfs_syscalls.c (q.v.) */ 1932 vrele(fvp); 1933 if ((fromnd.ni_cnd.cn_namelen == 1 && 1934 fromnd.ni_cnd.cn_nameptr[0] == '.') || 1935 (fromnd.ni_cnd.cn_namelen == 2 && 1936 fromnd.ni_cnd.cn_nameptr[0] == '.' && 1937 fromnd.ni_cnd.cn_nameptr[1] == '.')) { 1938 error = EINVAL; 1939 VFS_RENAMELOCK_EXIT(localfs); 1940 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1941 vrele(fromnd.ni_dvp); 1942 goto out1; 1943 } 1944 vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY); 1945 error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0); 1946 if (error) { 1947 VOP_UNLOCK(fromnd.ni_dvp); 1948 VFS_RENAMELOCK_EXIT(localfs); 1949 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1950 vrele(fromnd.ni_dvp); 1951 goto out1; 1952 } 1953 VOP_UNLOCK(fromnd.ni_vp); 1954 if (fromnd.ni_dvp != fromnd.ni_vp) 1955 VOP_UNLOCK(fromnd.ni_dvp); 1956 fvp = fromnd.ni_vp; 1957 1958 nfsm_srvmtofh(&tnsfh); 1959 if (v3) { 1960 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1961 len2 = fxdr_unsigned(uint32_t, *tl); 1962 /* len2 will be checked by nfs_namei */ 1963 } 1964 else { 1965 /* NFSv2 */ 1966 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1967 } 1968 kauth_cred_seteuid(cred, saved_uid); 1969 tond.ni_cnd.cn_cred = cred; 1970 tond.ni_cnd.cn_nameiop = RENAME; 1971 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | INRENAME; 1972 error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md, 1973 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1974 if (tdirp && v3) { 1975 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred); 1976 } 1977 if (error) { 1978 VFS_RENAMELOCK_EXIT(localfs); 1979 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1980 vrele(fromnd.ni_dvp); 1981 vrele(fvp); 1982 goto out1; 1983 } 1984 tdvp = tond.ni_dvp; 1985 tvp = tond.ni_vp; 1986 if (tvp != NULL) { 1987 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1988 if (v3) 1989 error = EEXIST; 1990 else 1991 error = EISDIR; 1992 goto out; 1993 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1994 if (v3) 1995 error = EEXIST; 1996 else 1997 error = ENOTDIR; 1998 goto out; 1999 } 2000 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 2001 if (v3) 2002 error = EXDEV; 2003 else 2004 error = ENOTEMPTY; 2005 goto out; 2006 } 2007 } 2008 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 2009 if (v3) 2010 error = EXDEV; 2011 else 2012 error = ENOTEMPTY; 2013 goto out; 2014 } 2015 if (fvp->v_mount != tdvp->v_mount) { 2016 if (v3) 2017 error = EXDEV; 2018 else 2019 error = ENOTEMPTY; 2020 goto out; 2021 } 2022 if (fvp == tdvp) { 2023 if (v3) 2024 error = EINVAL; 2025 else 2026 error = ENOTEMPTY; 2027 } 2028 /* 2029 * If source is the same as the destination (that is the 2030 * same vnode with the same name in the same directory), 2031 * then there is nothing to do. 2032 */ 2033 if (fvp == tvp && fromnd.ni_dvp == tdvp && 2034 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 2035 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 2036 fromnd.ni_cnd.cn_namelen)) 2037 error = -1; 2038 out: 2039 if (!error) { 2040 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 2041 nqsrv_getl(tdvp, ND_WRITE); 2042 if (tvp) { 2043 nqsrv_getl(tvp, ND_WRITE); 2044 } 2045 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 2046 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 2047 VFS_RENAMELOCK_EXIT(localfs); 2048 } else { 2049 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 2050 if (tdvp == tvp) 2051 vrele(tdvp); 2052 else 2053 vput(tdvp); 2054 if (tvp) 2055 vput(tvp); 2056 VFS_RENAMELOCK_EXIT(localfs); 2057 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2058 vrele(fromnd.ni_dvp); 2059 vrele(fvp); 2060 if (error == -1) 2061 error = 0; 2062 } 2063 pathbuf_destroy(tond.ni_pathbuf); 2064 tond.ni_cnd.cn_nameiop = 0; 2065 out1: 2066 if (fdirp) { 2067 if (v3) { 2068 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred); 2069 } 2070 vrele(fdirp); 2071 fdirp = NULL; 2072 } 2073 if (tdirp) { 2074 if (v3) { 2075 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred); 2076 } 2077 vrele(tdirp); 2078 tdirp = NULL; 2079 } 2080 pathbuf_destroy(fromnd.ni_pathbuf); 2081 fromnd.ni_cnd.cn_nameiop = 0; 2082 localfs = NULL; 2083 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2084 if (v3) { 2085 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2086 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2087 } 2088 return (0); 2089 2090 nfsmout: 2091 if (fdirp) 2092 vrele(fdirp); 2093 #ifdef notdef 2094 if (tdirp) 2095 vrele(tdirp); 2096 #endif 2097 if (tond.ni_cnd.cn_nameiop) { 2098 pathbuf_destroy(tond.ni_pathbuf); 2099 } 2100 if (localfs) { 2101 VFS_RENAMELOCK_EXIT(localfs); 2102 } 2103 if (fromnd.ni_cnd.cn_nameiop) { 2104 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2105 pathbuf_destroy(fromnd.ni_pathbuf); 2106 vrele(fromnd.ni_dvp); 2107 vrele(fvp); 2108 } 2109 return (error); 2110 } 2111 2112 /* 2113 * nfs link service 2114 */ 2115 int 2116 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2117 { 2118 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2119 struct mbuf *nam = nfsd->nd_nam; 2120 char *dpos = nfsd->nd_dpos; 2121 kauth_cred_t cred = nfsd->nd_cr; 2122 struct nameidata nd; 2123 u_int32_t *tl; 2124 int32_t t1; 2125 char *bpos; 2126 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2127 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2128 char *cp2; 2129 struct mbuf *mb, *mreq; 2130 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2131 struct vattr dirfor, diraft, at; 2132 nfsrvfh_t nsfh, dnsfh; 2133 u_quad_t frev; 2134 2135 nfsm_srvmtofh(&nsfh); 2136 nfsm_srvmtofh(&dnsfh); 2137 nfsm_srvnamesiz(len); 2138 error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam, 2139 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2140 if (error) { 2141 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2142 nfsm_srvpostop_attr(getret, &at); 2143 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2144 return (0); 2145 } 2146 if (vp->v_type == VDIR) { 2147 error = EPERM; 2148 goto out1; 2149 } 2150 nd.ni_cnd.cn_cred = cred; 2151 nd.ni_cnd.cn_nameiop = CREATE; 2152 nd.ni_cnd.cn_flags = LOCKPARENT; 2153 error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos, 2154 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2155 if (dirp && v3) { 2156 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2157 } 2158 if (error) 2159 goto out1; 2160 xp = nd.ni_vp; 2161 if (xp != NULL) { 2162 error = EEXIST; 2163 goto out; 2164 } 2165 xp = nd.ni_dvp; 2166 if (vp->v_mount != xp->v_mount) 2167 error = EXDEV; 2168 out: 2169 if (!error) { 2170 nqsrv_getl(vp, ND_WRITE); 2171 nqsrv_getl(xp, ND_WRITE); 2172 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2173 } else { 2174 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2175 if (nd.ni_dvp == nd.ni_vp) 2176 vrele(nd.ni_dvp); 2177 else 2178 vput(nd.ni_dvp); 2179 if (nd.ni_vp) 2180 vrele(nd.ni_vp); 2181 } 2182 out1: 2183 if (v3) 2184 getret = VOP_GETATTR(vp, &at, cred); 2185 if (dirp) { 2186 if (v3) { 2187 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2188 } 2189 vrele(dirp); 2190 } 2191 vrele(vp); 2192 pathbuf_destroy(nd.ni_pathbuf); 2193 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2194 if (v3) { 2195 nfsm_srvpostop_attr(getret, &at); 2196 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2197 return (0); 2198 } 2199 nfsm_srvdone; 2200 } 2201 2202 /* 2203 * nfs symbolic link service 2204 */ 2205 int 2206 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2207 { 2208 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2209 struct mbuf *nam = nfsd->nd_nam; 2210 char *dpos = nfsd->nd_dpos; 2211 kauth_cred_t cred = nfsd->nd_cr; 2212 struct vattr va, dirfor, diraft; 2213 struct nameidata nd; 2214 u_int32_t *tl; 2215 int32_t t1; 2216 struct nfsv2_sattr *sp; 2217 char *bpos, *pathcp = NULL, *cp2; 2218 struct uio io; 2219 struct iovec iv; 2220 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0; 2221 uint32_t len, len2; 2222 int v3 = (nfsd->nd_flag & ND_NFSV3); 2223 struct mbuf *mb, *mreq; 2224 struct vnode *dirp = (struct vnode *)0; 2225 nfsrvfh_t nsfh; 2226 u_quad_t frev; 2227 2228 nd.ni_cnd.cn_nameiop = 0; 2229 nfsm_srvmtofh(&nsfh); 2230 nfsm_srvnamesiz(len); 2231 nd.ni_cnd.cn_cred = cred; 2232 nd.ni_cnd.cn_nameiop = CREATE; 2233 nd.ni_cnd.cn_flags = LOCKPARENT; 2234 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2235 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2236 if (dirp && v3) { 2237 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2238 } 2239 if (error) 2240 goto out; 2241 abort = 1; 2242 vattr_null(&va); 2243 va.va_type = VLNK; 2244 if (v3) { 2245 va.va_mode = 0; 2246 nfsm_srvsattr(&va); 2247 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2248 len2 = fxdr_unsigned(uint32_t, *tl); 2249 if (len2 > PATH_MAX) { 2250 /* XXX should check _PC_NO_TRUNC */ 2251 error = ENAMETOOLONG; 2252 goto abortop; 2253 } 2254 } 2255 else { 2256 /* NFSv2 */ 2257 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2258 } 2259 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2260 iv.iov_base = pathcp; 2261 iv.iov_len = len2; 2262 io.uio_resid = len2; 2263 io.uio_offset = 0; 2264 io.uio_iov = &iv; 2265 io.uio_iovcnt = 1; 2266 io.uio_rw = UIO_READ; 2267 UIO_SETUP_SYSSPACE(&io); 2268 nfsm_mtouio(&io, len2); 2269 if (!v3) { 2270 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2271 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2272 } 2273 *(pathcp + len2) = '\0'; 2274 if (nd.ni_vp) { 2275 error = EEXIST; 2276 abortop: 2277 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2278 if (nd.ni_dvp == nd.ni_vp) 2279 vrele(nd.ni_dvp); 2280 else 2281 vput(nd.ni_dvp); 2282 if (nd.ni_vp) 2283 vrele(nd.ni_vp); 2284 goto out; 2285 } 2286 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2287 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2288 if (!error) { 2289 if (v3) { 2290 error = nfsrv_composefh(nd.ni_vp, &nsfh, v3); 2291 if (!error) 2292 error = VOP_GETATTR(nd.ni_vp, &va, cred); 2293 vput(nd.ni_vp); 2294 } else { 2295 vput(nd.ni_vp); 2296 } 2297 } 2298 out: 2299 if (pathcp) 2300 free(pathcp, M_TEMP); 2301 if (dirp) { 2302 if (v3) { 2303 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2304 } 2305 vrele(dirp); 2306 dirp = NULL; 2307 } 2308 pathbuf_destroy(nd.ni_pathbuf); 2309 abort = 0; 2310 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2311 NFSX_WCCDATA(v3)); 2312 if (v3) { 2313 if (!error) { 2314 nfsm_srvpostop_fh(&nsfh); 2315 nfsm_srvpostop_attr(0, &va); 2316 } 2317 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2318 } 2319 return (0); 2320 nfsmout: 2321 if (abort) { 2322 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2323 if (nd.ni_dvp == nd.ni_vp) 2324 vrele(nd.ni_dvp); 2325 else 2326 vput(nd.ni_dvp); 2327 if (nd.ni_vp) 2328 vrele(nd.ni_vp); 2329 pathbuf_destroy(nd.ni_pathbuf); 2330 } 2331 if (dirp) 2332 vrele(dirp); 2333 if (pathcp) 2334 free(pathcp, M_TEMP); 2335 return (error); 2336 } 2337 2338 /* 2339 * nfs mkdir service 2340 */ 2341 int 2342 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2343 { 2344 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2345 struct mbuf *nam = nfsd->nd_nam; 2346 char *dpos = nfsd->nd_dpos; 2347 kauth_cred_t cred = nfsd->nd_cr; 2348 struct vattr va, dirfor, diraft; 2349 struct nfs_fattr *fp; 2350 struct nameidata nd; 2351 char *cp; 2352 u_int32_t *tl; 2353 int32_t t1; 2354 char *bpos; 2355 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2356 int abort = 0; 2357 int v3 = (nfsd->nd_flag & ND_NFSV3); 2358 char *cp2; 2359 struct mbuf *mb, *mreq; 2360 struct vnode *vp, *dirp = (struct vnode *)0; 2361 nfsrvfh_t nsfh; 2362 u_quad_t frev; 2363 2364 nfsm_srvmtofh(&nsfh); 2365 nfsm_srvnamesiz(len); 2366 nd.ni_cnd.cn_cred = cred; 2367 nd.ni_cnd.cn_nameiop = CREATE; 2368 nd.ni_cnd.cn_flags = LOCKPARENT; 2369 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2370 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2371 if (dirp && v3) { 2372 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2373 } 2374 if (error) { 2375 if (nd.ni_pathbuf != NULL) { 2376 pathbuf_destroy(nd.ni_pathbuf); 2377 } 2378 nfsm_reply(NFSX_WCCDATA(v3)); 2379 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2380 if (dirp) 2381 vrele(dirp); 2382 return (0); 2383 } 2384 abort = 1; 2385 vattr_null(&va); 2386 if (v3) { 2387 va.va_mode = 0; 2388 nfsm_srvsattr(&va); 2389 } else { 2390 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2391 va.va_mode = nfstov_mode(*tl++); 2392 } 2393 va.va_type = VDIR; 2394 vp = nd.ni_vp; 2395 if (vp != NULL) { 2396 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2397 if (nd.ni_dvp == vp) 2398 vrele(nd.ni_dvp); 2399 else 2400 vput(nd.ni_dvp); 2401 vrele(vp); 2402 error = EEXIST; 2403 goto out; 2404 } 2405 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2406 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2407 if (!error) { 2408 vp = nd.ni_vp; 2409 error = nfsrv_composefh(vp, &nsfh, v3); 2410 if (!error) 2411 error = VOP_GETATTR(vp, &va, cred); 2412 vput(vp); 2413 } 2414 out: 2415 if (dirp) { 2416 if (v3) { 2417 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2418 } 2419 vrele(dirp); 2420 dirp = NULL; 2421 } 2422 pathbuf_destroy(nd.ni_pathbuf); 2423 abort = 0; 2424 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2425 NFSX_WCCDATA(v3)); 2426 if (v3) { 2427 if (!error) { 2428 nfsm_srvpostop_fh(&nsfh); 2429 nfsm_srvpostop_attr(0, &va); 2430 } 2431 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2432 } else { 2433 nfsm_srvfhtom(&nsfh, v3); 2434 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2435 nfsm_srvfillattr(&va, fp); 2436 } 2437 return (0); 2438 nfsmout: 2439 if (abort) { 2440 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2441 if (nd.ni_dvp == nd.ni_vp) 2442 vrele(nd.ni_dvp); 2443 else 2444 vput(nd.ni_dvp); 2445 if (nd.ni_vp) 2446 vrele(nd.ni_vp); 2447 pathbuf_destroy(nd.ni_pathbuf); 2448 } 2449 if (dirp) 2450 vrele(dirp); 2451 return (error); 2452 } 2453 2454 /* 2455 * nfs rmdir service 2456 */ 2457 int 2458 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2459 { 2460 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2461 struct mbuf *nam = nfsd->nd_nam; 2462 char *dpos = nfsd->nd_dpos; 2463 kauth_cred_t cred = nfsd->nd_cr; 2464 u_int32_t *tl; 2465 int32_t t1; 2466 char *bpos; 2467 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2468 int v3 = (nfsd->nd_flag & ND_NFSV3); 2469 char *cp2; 2470 struct mbuf *mb, *mreq; 2471 struct vnode *vp, *dirp = (struct vnode *)0; 2472 struct vattr dirfor, diraft; 2473 nfsrvfh_t nsfh; 2474 struct nameidata nd; 2475 u_quad_t frev; 2476 2477 nfsm_srvmtofh(&nsfh); 2478 nfsm_srvnamesiz(len); 2479 nd.ni_cnd.cn_cred = cred; 2480 nd.ni_cnd.cn_nameiop = DELETE; 2481 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2482 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2483 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2484 if (dirp && v3) { 2485 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2486 } 2487 if (error) { 2488 if (nd.ni_pathbuf != NULL) { 2489 pathbuf_destroy(nd.ni_pathbuf); 2490 } 2491 nfsm_reply(NFSX_WCCDATA(v3)); 2492 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2493 if (dirp) 2494 vrele(dirp); 2495 return (0); 2496 } 2497 vp = nd.ni_vp; 2498 if (vp->v_type != VDIR) { 2499 error = ENOTDIR; 2500 goto out; 2501 } 2502 /* 2503 * No rmdir "." please. 2504 */ 2505 if (nd.ni_dvp == vp) { 2506 error = EINVAL; 2507 goto out; 2508 } 2509 /* 2510 * The root of a mounted filesystem cannot be deleted. 2511 */ 2512 if (vp->v_vflag & VV_ROOT) 2513 error = EBUSY; 2514 out: 2515 if (!error) { 2516 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2517 nqsrv_getl(vp, ND_WRITE); 2518 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2519 } else { 2520 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2521 if (nd.ni_dvp == nd.ni_vp) 2522 vrele(nd.ni_dvp); 2523 else 2524 vput(nd.ni_dvp); 2525 vput(vp); 2526 } 2527 pathbuf_destroy(nd.ni_pathbuf); 2528 if (dirp) { 2529 if (v3) { 2530 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2531 } 2532 vrele(dirp); 2533 } 2534 nfsm_reply(NFSX_WCCDATA(v3)); 2535 if (v3) { 2536 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2537 return (0); 2538 } 2539 nfsm_srvdone; 2540 } 2541 2542 /* 2543 * nfs readdir service 2544 * - mallocs what it thinks is enough to read 2545 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2546 * - calls VOP_READDIR() 2547 * - loops around building the reply 2548 * if the output generated exceeds count break out of loop 2549 * The nfsm_clget macro is used here so that the reply will be packed 2550 * tightly in mbuf clusters. 2551 * - it only knows that it has encountered eof when the VOP_READDIR() 2552 * reads nothing 2553 * - as such one readdir rpc will return eof false although you are there 2554 * and then the next will return eof 2555 * - it trims out records with d_fileno == 0 2556 * this doesn't matter for Unix clients, but they might confuse clients 2557 * for other os'. 2558 * - it trims out records with d_type == DT_WHT 2559 * these cannot be seen through NFS (unless we extend the protocol) 2560 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2561 * than requested, but this may not apply to all filesystems. For 2562 * example, client NFS does not { although it is never remote mounted 2563 * anyhow } 2564 * The alternate call nfsrv_readdirplus() does lookups as well. 2565 * PS: The NFS protocol spec. does not clarify what the "count" byte 2566 * argument is a count of.. just name strings and file id's or the 2567 * entire reply rpc or ... 2568 * I tried just file name and id sizes and it confused the Sun client, 2569 * so I am using the full rpc size now. The "paranoia.." comment refers 2570 * to including the status longwords that are not a part of the dir. 2571 * "entry" structures, but are in the rpc. 2572 */ 2573 2574 #define NFS_SRVDIRBLKSIZ 1024 2575 2576 struct flrep { 2577 nfsuint64 fl_off; 2578 u_int32_t fl_postopok; 2579 struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */ 2580 u_int32_t fl_fhok; 2581 u_int32_t fl_fhsize; 2582 /* handle comes here, filled in dynamically */ 2583 }; 2584 2585 int 2586 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2587 { 2588 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2589 struct mbuf *nam = nfsd->nd_nam; 2590 char *dpos = nfsd->nd_dpos; 2591 kauth_cred_t cred = nfsd->nd_cr; 2592 char *bp, *be; 2593 struct mbuf *mp; 2594 struct dirent *dp; 2595 char *cp; 2596 u_int32_t *tl; 2597 int32_t t1; 2598 char *bpos; 2599 struct mbuf *mb, *mreq, *mp2; 2600 char *cpos, *cend, *cp2, *rbuf; 2601 struct vnode *vp; 2602 struct vattr at; 2603 nfsrvfh_t nsfh; 2604 struct uio io; 2605 struct iovec iv; 2606 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2607 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2608 int v3 = (nfsd->nd_flag & ND_NFSV3); 2609 u_quad_t frev, off, toff, verf; 2610 off_t *cookies = NULL, *cookiep; 2611 nfsuint64 jar; 2612 2613 nfsm_srvmtofh(&nsfh); 2614 if (v3) { 2615 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2616 toff = fxdr_hyper(tl); 2617 tl += 2; 2618 verf = fxdr_hyper(tl); 2619 tl += 2; 2620 } else { 2621 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2622 toff = fxdr_unsigned(u_quad_t, *tl++); 2623 } 2624 off = toff; 2625 cnt = fxdr_unsigned(int, *tl); 2626 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2627 xfer = NFS_SRVMAXDATA(nfsd); 2628 if (siz > xfer) 2629 siz = xfer; 2630 fullsiz = siz; 2631 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2632 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2633 if (!error && vp->v_type != VDIR) { 2634 error = ENOTDIR; 2635 vput(vp); 2636 } 2637 if (error) { 2638 nfsm_reply(NFSX_UNSIGNED); 2639 nfsm_srvpostop_attr(getret, &at); 2640 return (0); 2641 } 2642 nqsrv_getl(vp, ND_READ); 2643 if (v3) { 2644 error = getret = VOP_GETATTR(vp, &at, cred); 2645 #ifdef NFS3_STRICTVERF 2646 /* 2647 * XXX This check is too strict for Solaris 2.5 clients. 2648 */ 2649 if (!error && toff && verf != at.va_filerev) 2650 error = NFSERR_BAD_COOKIE; 2651 #endif 2652 } 2653 if (!error) 2654 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2655 if (error) { 2656 vput(vp); 2657 nfsm_reply(NFSX_POSTOPATTR(v3)); 2658 nfsm_srvpostop_attr(getret, &at); 2659 return (0); 2660 } 2661 VOP_UNLOCK(vp); 2662 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2663 again: 2664 iv.iov_base = rbuf; 2665 iv.iov_len = fullsiz; 2666 io.uio_iov = &iv; 2667 io.uio_iovcnt = 1; 2668 io.uio_offset = (off_t)off; 2669 io.uio_resid = fullsiz; 2670 io.uio_rw = UIO_READ; 2671 UIO_SETUP_SYSSPACE(&io); 2672 eofflag = 0; 2673 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2674 2675 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2676 2677 off = (off_t)io.uio_offset; 2678 if (!cookies && !error) 2679 error = NFSERR_PERM; 2680 if (v3) { 2681 getret = VOP_GETATTR(vp, &at, cred); 2682 if (!error) 2683 error = getret; 2684 } 2685 2686 VOP_UNLOCK(vp); 2687 if (error) { 2688 vrele(vp); 2689 free((void *)rbuf, M_TEMP); 2690 if (cookies) 2691 free((void *)cookies, M_TEMP); 2692 nfsm_reply(NFSX_POSTOPATTR(v3)); 2693 nfsm_srvpostop_attr(getret, &at); 2694 return (0); 2695 } 2696 if (io.uio_resid) { 2697 siz -= io.uio_resid; 2698 2699 /* 2700 * If nothing read, return eof 2701 * rpc reply 2702 */ 2703 if (siz == 0) { 2704 vrele(vp); 2705 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2706 2 * NFSX_UNSIGNED); 2707 if (v3) { 2708 nfsm_srvpostop_attr(getret, &at); 2709 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2710 txdr_hyper(at.va_filerev, tl); 2711 tl += 2; 2712 } else 2713 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2714 *tl++ = nfs_false; 2715 *tl = nfs_true; 2716 free((void *)rbuf, M_TEMP); 2717 free((void *)cookies, M_TEMP); 2718 return (0); 2719 } 2720 } 2721 2722 /* 2723 * Check for degenerate cases of nothing useful read. 2724 * If so go try again 2725 */ 2726 cpos = rbuf; 2727 cend = rbuf + siz; 2728 dp = (struct dirent *)cpos; 2729 cookiep = cookies; 2730 2731 while (cpos < cend && ncookies > 0 && 2732 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2733 cpos += dp->d_reclen; 2734 dp = (struct dirent *)cpos; 2735 cookiep++; 2736 ncookies--; 2737 } 2738 if (cpos >= cend || ncookies == 0) { 2739 toff = off; 2740 siz = fullsiz; 2741 free(cookies, M_TEMP); 2742 cookies = NULL; 2743 goto again; 2744 } 2745 2746 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2747 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2748 if (v3) { 2749 nfsm_srvpostop_attr(getret, &at); 2750 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2751 txdr_hyper(at.va_filerev, tl); 2752 } 2753 mp = mp2 = mb; 2754 bp = bpos; 2755 be = bp + M_TRAILINGSPACE(mp); 2756 2757 /* Loop through the records and build reply */ 2758 while (cpos < cend && ncookies > 0) { 2759 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2760 nlen = dp->d_namlen; 2761 rem = nfsm_rndup(nlen)-nlen; 2762 len += (4 * NFSX_UNSIGNED + nlen + rem); 2763 if (v3) 2764 len += 2 * NFSX_UNSIGNED; 2765 if (len > cnt) { 2766 eofflag = 0; 2767 break; 2768 } 2769 /* 2770 * Build the directory record xdr from 2771 * the dirent entry. 2772 */ 2773 nfsm_clget; 2774 *tl = nfs_true; 2775 bp += NFSX_UNSIGNED; 2776 if (v3) { 2777 nfsm_clget; 2778 *tl = txdr_unsigned(dp->d_fileno >> 32); 2779 bp += NFSX_UNSIGNED; 2780 } 2781 nfsm_clget; 2782 *tl = txdr_unsigned(dp->d_fileno); 2783 bp += NFSX_UNSIGNED; 2784 nfsm_clget; 2785 *tl = txdr_unsigned(nlen); 2786 bp += NFSX_UNSIGNED; 2787 2788 /* And loop around copying the name */ 2789 xfer = nlen; 2790 cp = dp->d_name; 2791 while (xfer > 0) { 2792 nfsm_clget; 2793 if ((bp+xfer) > be) 2794 tsiz = be-bp; 2795 else 2796 tsiz = xfer; 2797 memcpy(bp, cp, tsiz); 2798 bp += tsiz; 2799 xfer -= tsiz; 2800 if (xfer > 0) 2801 cp += tsiz; 2802 } 2803 /* And null pad to an int32_t boundary */ 2804 for (i = 0; i < rem; i++) 2805 *bp++ = '\0'; 2806 nfsm_clget; 2807 2808 /* Finish off the record */ 2809 txdr_hyper(*cookiep, &jar); 2810 if (v3) { 2811 *tl = jar.nfsuquad[0]; 2812 bp += NFSX_UNSIGNED; 2813 nfsm_clget; 2814 } 2815 *tl = jar.nfsuquad[1]; 2816 bp += NFSX_UNSIGNED; 2817 } 2818 cpos += dp->d_reclen; 2819 dp = (struct dirent *)cpos; 2820 cookiep++; 2821 ncookies--; 2822 } 2823 vrele(vp); 2824 nfsm_clget; 2825 *tl = nfs_false; 2826 bp += NFSX_UNSIGNED; 2827 nfsm_clget; 2828 if (eofflag) 2829 *tl = nfs_true; 2830 else 2831 *tl = nfs_false; 2832 bp += NFSX_UNSIGNED; 2833 if (mp != mb) { 2834 if (bp < be) 2835 mp->m_len = bp - mtod(mp, char *); 2836 } else 2837 mp->m_len += bp - bpos; 2838 free((void *)rbuf, M_TEMP); 2839 free((void *)cookies, M_TEMP); 2840 nfsm_srvdone; 2841 } 2842 2843 int 2844 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2845 { 2846 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2847 struct mbuf *nam = nfsd->nd_nam; 2848 char *dpos = nfsd->nd_dpos; 2849 kauth_cred_t cred = nfsd->nd_cr; 2850 char *bp, *be; 2851 struct mbuf *mp; 2852 struct dirent *dp; 2853 char *cp; 2854 u_int32_t *tl; 2855 int32_t t1; 2856 char *bpos; 2857 struct mbuf *mb, *mreq, *mp2; 2858 char *cpos, *cend, *cp2, *rbuf; 2859 struct vnode *vp, *nvp; 2860 struct flrep fl; 2861 nfsrvfh_t nsfh; 2862 struct uio io; 2863 struct iovec iv; 2864 struct vattr va, at, *vap = &va; 2865 struct nfs_fattr *fp; 2866 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2867 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2868 u_quad_t frev, off, toff, verf; 2869 off_t *cookies = NULL, *cookiep; 2870 2871 nfsm_srvmtofh(&nsfh); 2872 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2873 toff = fxdr_hyper(tl); 2874 tl += 2; 2875 verf = fxdr_hyper(tl); 2876 tl += 2; 2877 siz = fxdr_unsigned(int, *tl++); 2878 cnt = fxdr_unsigned(int, *tl); 2879 off = toff; 2880 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2881 xfer = NFS_SRVMAXDATA(nfsd); 2882 if (siz > xfer) 2883 siz = xfer; 2884 fullsiz = siz; 2885 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2886 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2887 if (!error && vp->v_type != VDIR) { 2888 error = ENOTDIR; 2889 vput(vp); 2890 } 2891 if (error) { 2892 nfsm_reply(NFSX_UNSIGNED); 2893 nfsm_srvpostop_attr(getret, &at); 2894 return (0); 2895 } 2896 error = getret = VOP_GETATTR(vp, &at, cred); 2897 #ifdef NFS3_STRICTVERF 2898 /* 2899 * XXX This check is too strict for Solaris 2.5 clients. 2900 */ 2901 if (!error && toff && verf != at.va_filerev) 2902 error = NFSERR_BAD_COOKIE; 2903 #endif 2904 if (!error) { 2905 nqsrv_getl(vp, ND_READ); 2906 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2907 } 2908 if (error) { 2909 vput(vp); 2910 nfsm_reply(NFSX_V3POSTOPATTR); 2911 nfsm_srvpostop_attr(getret, &at); 2912 return (0); 2913 } 2914 VOP_UNLOCK(vp); 2915 2916 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2917 again: 2918 iv.iov_base = rbuf; 2919 iv.iov_len = fullsiz; 2920 io.uio_iov = &iv; 2921 io.uio_iovcnt = 1; 2922 io.uio_offset = (off_t)off; 2923 io.uio_resid = fullsiz; 2924 io.uio_rw = UIO_READ; 2925 UIO_SETUP_SYSSPACE(&io); 2926 eofflag = 0; 2927 2928 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2929 2930 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2931 2932 off = (u_quad_t)io.uio_offset; 2933 getret = VOP_GETATTR(vp, &at, cred); 2934 2935 VOP_UNLOCK(vp); 2936 2937 /* 2938 * If the VGET operation doesn't work for this filesystem, 2939 * we can't support readdirplus. Returning NOTSUPP should 2940 * make clients fall back to plain readdir. 2941 * There's no need to check for VPTOFH as well, we wouldn't 2942 * even be here otherwise. 2943 */ 2944 if (!getret) { 2945 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2946 getret = (getret == EOPNOTSUPP) ? 2947 NFSERR_NOTSUPP : NFSERR_IO; 2948 else 2949 vput(nvp); 2950 } 2951 2952 if (!cookies && !error) 2953 error = NFSERR_PERM; 2954 if (!error) 2955 error = getret; 2956 if (error) { 2957 vrele(vp); 2958 if (cookies) 2959 free((void *)cookies, M_TEMP); 2960 free((void *)rbuf, M_TEMP); 2961 nfsm_reply(NFSX_V3POSTOPATTR); 2962 nfsm_srvpostop_attr(getret, &at); 2963 return (0); 2964 } 2965 if (io.uio_resid) { 2966 siz -= io.uio_resid; 2967 2968 /* 2969 * If nothing read, return eof 2970 * rpc reply 2971 */ 2972 if (siz == 0) { 2973 vrele(vp); 2974 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2975 2 * NFSX_UNSIGNED); 2976 nfsm_srvpostop_attr(getret, &at); 2977 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2978 txdr_hyper(at.va_filerev, tl); 2979 tl += 2; 2980 *tl++ = nfs_false; 2981 *tl = nfs_true; 2982 free((void *)cookies, M_TEMP); 2983 free((void *)rbuf, M_TEMP); 2984 return (0); 2985 } 2986 } 2987 2988 /* 2989 * Check for degenerate cases of nothing useful read. 2990 * If so go try again 2991 */ 2992 cpos = rbuf; 2993 cend = rbuf + siz; 2994 dp = (struct dirent *)cpos; 2995 cookiep = cookies; 2996 2997 while (cpos < cend && ncookies > 0 && 2998 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2999 cpos += dp->d_reclen; 3000 dp = (struct dirent *)cpos; 3001 cookiep++; 3002 ncookies--; 3003 } 3004 if (cpos >= cend || ncookies == 0) { 3005 toff = off; 3006 siz = fullsiz; 3007 free(cookies, M_TEMP); 3008 cookies = NULL; 3009 goto again; 3010 } 3011 3012 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3013 nfsm_reply(cnt); 3014 nfsm_srvpostop_attr(getret, &at); 3015 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3016 txdr_hyper(at.va_filerev, tl); 3017 mp = mp2 = mb; 3018 bp = bpos; 3019 be = bp + M_TRAILINGSPACE(mp); 3020 3021 /* Loop through the records and build reply */ 3022 while (cpos < cend && ncookies > 0) { 3023 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3024 nfsrvfh_t nnsfh; 3025 3026 nlen = dp->d_namlen; 3027 rem = nfsm_rndup(nlen)-nlen; 3028 3029 /* 3030 * For readdir_and_lookup get the vnode using 3031 * the file number. 3032 */ 3033 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3034 goto invalid; 3035 if (nfsrv_composefh(nvp, &nnsfh, true)) { 3036 vput(nvp); 3037 goto invalid; 3038 } 3039 if (VOP_GETATTR(nvp, vap, cred)) { 3040 vput(nvp); 3041 goto invalid; 3042 } 3043 vput(nvp); 3044 3045 /* 3046 * If either the dircount or maxcount will be 3047 * exceeded, get out now. Both of these lengths 3048 * are calculated conservatively, including all 3049 * XDR overheads. 3050 */ 3051 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3052 NFSX_V3POSTOPATTR); 3053 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3054 if (len > cnt || dirlen > fullsiz) { 3055 eofflag = 0; 3056 break; 3057 } 3058 3059 /* 3060 * Build the directory record xdr from 3061 * the dirent entry. 3062 */ 3063 fp = (struct nfs_fattr *)&fl.fl_fattr; 3064 nfsm_srvfillattr(vap, fp); 3065 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3066 fl.fl_fhok = nfs_true; 3067 fl.fl_postopok = nfs_true; 3068 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3069 3070 nfsm_clget; 3071 *tl = nfs_true; 3072 bp += NFSX_UNSIGNED; 3073 nfsm_clget; 3074 *tl = txdr_unsigned(dp->d_fileno >> 32); 3075 bp += NFSX_UNSIGNED; 3076 nfsm_clget; 3077 *tl = txdr_unsigned(dp->d_fileno); 3078 bp += NFSX_UNSIGNED; 3079 nfsm_clget; 3080 *tl = txdr_unsigned(nlen); 3081 bp += NFSX_UNSIGNED; 3082 3083 /* And loop around copying the name */ 3084 xfer = nlen; 3085 cp = dp->d_name; 3086 while (xfer > 0) { 3087 nfsm_clget; 3088 if ((bp + xfer) > be) 3089 tsiz = be - bp; 3090 else 3091 tsiz = xfer; 3092 memcpy(bp, cp, tsiz); 3093 bp += tsiz; 3094 xfer -= tsiz; 3095 if (xfer > 0) 3096 cp += tsiz; 3097 } 3098 /* And null pad to an int32_t boundary */ 3099 for (i = 0; i < rem; i++) 3100 *bp++ = '\0'; 3101 3102 /* 3103 * Now copy the flrep structure out. 3104 */ 3105 xfer = sizeof(struct flrep); 3106 cp = (void *)&fl; 3107 while (xfer > 0) { 3108 nfsm_clget; 3109 if ((bp + xfer) > be) 3110 tsiz = be - bp; 3111 else 3112 tsiz = xfer; 3113 memcpy(bp, cp, tsiz); 3114 bp += tsiz; 3115 xfer -= tsiz; 3116 if (xfer > 0) 3117 cp += tsiz; 3118 } 3119 3120 /* 3121 * ... and filehandle. 3122 */ 3123 xfer = NFSRVFH_SIZE(&nnsfh); 3124 cp = NFSRVFH_DATA(&nnsfh); 3125 while (xfer > 0) { 3126 nfsm_clget; 3127 if ((bp + xfer) > be) 3128 tsiz = be - bp; 3129 else 3130 tsiz = xfer; 3131 memcpy(bp, cp, tsiz); 3132 bp += tsiz; 3133 xfer -= tsiz; 3134 if (xfer > 0) 3135 cp += tsiz; 3136 } 3137 } 3138 invalid: 3139 cpos += dp->d_reclen; 3140 dp = (struct dirent *)cpos; 3141 cookiep++; 3142 ncookies--; 3143 } 3144 vrele(vp); 3145 nfsm_clget; 3146 *tl = nfs_false; 3147 bp += NFSX_UNSIGNED; 3148 nfsm_clget; 3149 if (eofflag) 3150 *tl = nfs_true; 3151 else 3152 *tl = nfs_false; 3153 bp += NFSX_UNSIGNED; 3154 if (mp != mb) { 3155 if (bp < be) 3156 mp->m_len = bp - mtod(mp, char *); 3157 } else 3158 mp->m_len += bp - bpos; 3159 free((void *)cookies, M_TEMP); 3160 free((void *)rbuf, M_TEMP); 3161 nfsm_srvdone; 3162 } 3163 3164 /* 3165 * nfs commit service 3166 */ 3167 int 3168 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3169 { 3170 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3171 struct mbuf *nam = nfsd->nd_nam; 3172 char *dpos = nfsd->nd_dpos; 3173 kauth_cred_t cred = nfsd->nd_cr; 3174 struct vattr bfor, aft; 3175 struct vnode *vp; 3176 nfsrvfh_t nsfh; 3177 u_int32_t *tl; 3178 int32_t t1; 3179 char *bpos; 3180 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3181 uint32_t cnt; 3182 char *cp2; 3183 struct mbuf *mb, *mreq; 3184 u_quad_t frev, off, end; 3185 3186 nfsm_srvmtofh(&nsfh); 3187 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3188 3189 off = fxdr_hyper(tl); 3190 tl += 2; 3191 cnt = fxdr_unsigned(uint32_t, *tl); 3192 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3193 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3194 if (error) { 3195 nfsm_reply(2 * NFSX_UNSIGNED); 3196 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3197 return (0); 3198 } 3199 for_ret = VOP_GETATTR(vp, &bfor, cred); 3200 end = (cnt > 0) ? off + cnt : vp->v_size; 3201 if (end < off || end > vp->v_size) 3202 end = vp->v_size; 3203 if (off < vp->v_size) 3204 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end); 3205 /* else error == 0, from nfsrv_fhtovp() */ 3206 aft_ret = VOP_GETATTR(vp, &aft, cred); 3207 vput(vp); 3208 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3209 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3210 if (!error) { 3211 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3212 *tl++ = txdr_unsigned(boottime.tv_sec); 3213 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 3214 } else { 3215 return (0); 3216 } 3217 nfsm_srvdone; 3218 } 3219 3220 /* 3221 * nfs statfs service 3222 */ 3223 int 3224 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3225 { 3226 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3227 struct mbuf *nam = nfsd->nd_nam; 3228 char *dpos = nfsd->nd_dpos; 3229 kauth_cred_t cred = nfsd->nd_cr; 3230 struct statvfs *sf = NULL; 3231 struct nfs_statfs *sfp; 3232 u_int32_t *tl; 3233 int32_t t1; 3234 char *bpos; 3235 int error = 0, rdonly, cache = 0, getret = 1; 3236 int v3 = (nfsd->nd_flag & ND_NFSV3); 3237 char *cp2; 3238 struct mbuf *mb, *mreq; 3239 struct vnode *vp; 3240 struct vattr at; 3241 nfsrvfh_t nsfh; 3242 u_quad_t frev, tval; 3243 3244 nfsm_srvmtofh(&nsfh); 3245 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3246 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3247 if (error) { 3248 nfsm_reply(NFSX_UNSIGNED); 3249 nfsm_srvpostop_attr(getret, &at); 3250 return (0); 3251 } 3252 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK); 3253 error = VFS_STATVFS(vp->v_mount, sf); 3254 getret = VOP_GETATTR(vp, &at, cred); 3255 vput(vp); 3256 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3257 if (v3) 3258 nfsm_srvpostop_attr(getret, &at); 3259 if (error) { 3260 free(sf, M_TEMP); 3261 return (0); 3262 } 3263 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3264 if (v3) { 3265 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3266 txdr_hyper(tval, &sfp->sf_tbytes); 3267 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3268 txdr_hyper(tval, &sfp->sf_fbytes); 3269 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3270 txdr_hyper(tval, &sfp->sf_abytes); 3271 tval = (u_quad_t)sf->f_files; 3272 txdr_hyper(tval, &sfp->sf_tfiles); 3273 tval = (u_quad_t)sf->f_ffree; 3274 txdr_hyper(tval, &sfp->sf_ffiles); 3275 txdr_hyper(tval, &sfp->sf_afiles); 3276 sfp->sf_invarsec = 0; 3277 } else { 3278 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3279 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3280 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3281 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3282 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3283 } 3284 nfsmout: 3285 if (sf) 3286 free(sf, M_TEMP); 3287 return error; 3288 } 3289 3290 /* 3291 * nfs fsinfo service 3292 */ 3293 int 3294 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3295 { 3296 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3297 struct mbuf *nam = nfsd->nd_nam; 3298 char *dpos = nfsd->nd_dpos; 3299 kauth_cred_t cred = nfsd->nd_cr; 3300 u_int32_t *tl; 3301 struct nfsv3_fsinfo *sip; 3302 int32_t t1; 3303 char *bpos; 3304 int error = 0, rdonly, cache = 0, getret = 1; 3305 uint32_t maxdata; 3306 char *cp2; 3307 struct mbuf *mb, *mreq; 3308 struct vnode *vp; 3309 struct vattr at; 3310 nfsrvfh_t nsfh; 3311 u_quad_t frev, maxfsize; 3312 struct statvfs *sb; 3313 3314 nfsm_srvmtofh(&nsfh); 3315 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3316 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3317 if (error) { 3318 nfsm_reply(NFSX_UNSIGNED); 3319 nfsm_srvpostop_attr(getret, &at); 3320 return (0); 3321 } 3322 3323 /* XXX Try to make a guess on the max file size. */ 3324 sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK); 3325 VFS_STATVFS(vp->v_mount, sb); 3326 maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1; 3327 free(sb, M_TEMP); 3328 3329 getret = VOP_GETATTR(vp, &at, cred); 3330 vput(vp); 3331 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3332 nfsm_srvpostop_attr(getret, &at); 3333 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3334 3335 /* 3336 * XXX 3337 * There should be file system VFS OP(s) to get this information. 3338 * For now, assume ufs. 3339 */ 3340 if (slp->ns_so->so_type == SOCK_DGRAM) 3341 maxdata = NFS_MAXDGRAMDATA; 3342 else 3343 maxdata = NFS_MAXDATA; 3344 sip->fs_rtmax = txdr_unsigned(maxdata); 3345 sip->fs_rtpref = txdr_unsigned(maxdata); 3346 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3347 sip->fs_wtmax = txdr_unsigned(maxdata); 3348 sip->fs_wtpref = txdr_unsigned(maxdata); 3349 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3350 sip->fs_dtpref = txdr_unsigned(maxdata); 3351 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3352 sip->fs_timedelta.nfsv3_sec = 0; 3353 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3354 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3355 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3356 NFSV3FSINFO_CANSETTIME); 3357 nfsm_srvdone; 3358 } 3359 3360 /* 3361 * nfs pathconf service 3362 */ 3363 int 3364 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3365 { 3366 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3367 struct mbuf *nam = nfsd->nd_nam; 3368 char *dpos = nfsd->nd_dpos; 3369 kauth_cred_t cred = nfsd->nd_cr; 3370 u_int32_t *tl; 3371 struct nfsv3_pathconf *pc; 3372 int32_t t1; 3373 char *bpos; 3374 int error = 0, rdonly, cache = 0, getret = 1; 3375 register_t linkmax, namemax, chownres, notrunc; 3376 char *cp2; 3377 struct mbuf *mb, *mreq; 3378 struct vnode *vp; 3379 struct vattr at; 3380 nfsrvfh_t nsfh; 3381 u_quad_t frev; 3382 3383 nfsm_srvmtofh(&nsfh); 3384 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3385 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3386 if (error) { 3387 nfsm_reply(NFSX_UNSIGNED); 3388 nfsm_srvpostop_attr(getret, &at); 3389 return (0); 3390 } 3391 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3392 if (!error) 3393 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3394 if (!error) 3395 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3396 if (!error) 3397 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3398 getret = VOP_GETATTR(vp, &at, cred); 3399 vput(vp); 3400 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3401 nfsm_srvpostop_attr(getret, &at); 3402 if (error) 3403 return (0); 3404 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3405 3406 pc->pc_linkmax = txdr_unsigned(linkmax); 3407 pc->pc_namemax = txdr_unsigned(namemax); 3408 pc->pc_notrunc = txdr_unsigned(notrunc); 3409 pc->pc_chownrestricted = txdr_unsigned(chownres); 3410 3411 /* 3412 * These should probably be supported by VOP_PATHCONF(), but 3413 * until msdosfs is exportable (why would you want to?), the 3414 * Unix defaults should be ok. 3415 */ 3416 pc->pc_caseinsensitive = nfs_false; 3417 pc->pc_casepreserving = nfs_true; 3418 nfsm_srvdone; 3419 } 3420 3421 /* 3422 * Null operation, used by clients to ping server 3423 */ 3424 /* ARGSUSED */ 3425 int 3426 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3427 struct lwp *lwp, struct mbuf **mrq) 3428 { 3429 struct mbuf *mrep = nfsd->nd_mrep; 3430 char *bpos; 3431 int error = NFSERR_RETVOID, cache = 0; 3432 struct mbuf *mb, *mreq; 3433 u_quad_t frev; 3434 3435 nfsm_reply(0); 3436 nfsmout: 3437 return (0); 3438 } 3439 3440 /* 3441 * No operation, used for obsolete procedures 3442 */ 3443 /* ARGSUSED */ 3444 int 3445 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3446 struct lwp *lwp, struct mbuf **mrq) 3447 { 3448 struct mbuf *mrep = nfsd->nd_mrep; 3449 char *bpos; 3450 int error, cache = 0; 3451 struct mbuf *mb, *mreq; 3452 u_quad_t frev; 3453 3454 if (nfsd->nd_repstat) 3455 error = nfsd->nd_repstat; 3456 else 3457 error = EPROCUNAVAIL; 3458 nfsm_reply(0); 3459 nfsmout: 3460 return (0); 3461 } 3462 3463 /* 3464 * Perform access checking for vnodes obtained from file handles that would 3465 * refer to files already opened by a Unix client. You cannot just use 3466 * vn_writechk() and VOP_ACCESS() for two reasons. 3467 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3468 * 2 - The owner is to be given access irrespective of mode bits for some 3469 * operations, so that processes that chmod after opening a file don't 3470 * break. I don't like this because it opens a security hole, but since 3471 * the nfs server opens a security hole the size of a barn door anyhow, 3472 * what the heck. 3473 * 3474 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3475 * will return EPERM instead of EACCES. EPERM is always an error. 3476 */ 3477 int 3478 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override) 3479 { 3480 struct vattr vattr; 3481 int error; 3482 if (flags & VWRITE) { 3483 /* Just vn_writechk() changed to check rdonly */ 3484 /* 3485 * Disallow write attempts on read-only file systems; 3486 * unless the file is a socket or a block or character 3487 * device resident on the file system. 3488 */ 3489 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3490 switch (vp->v_type) { 3491 case VREG: 3492 case VDIR: 3493 case VLNK: 3494 return (EROFS); 3495 default: 3496 break; 3497 } 3498 } 3499 3500 /* 3501 * If the vnode is in use as a process's text, 3502 * we can't allow writing. 3503 */ 3504 if (vp->v_iflag & VI_TEXT) 3505 return (ETXTBSY); 3506 } 3507 error = VOP_GETATTR(vp, &vattr, cred); 3508 if (error) 3509 return (error); 3510 error = VOP_ACCESS(vp, flags, cred); 3511 /* 3512 * Allow certain operations for the owner (reads and writes 3513 * on files that are already open). 3514 */ 3515 if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid) 3516 error = 0; 3517 return error; 3518 } 3519