nfs_serv.c revision 1.158
1 /* $NetBSD: nfs_serv.c,v 1.158 2011/04/11 01:33:05 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 1.158 2011/04/11 01:33:05 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 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1531 vput(nd.ni_vp); 1532 vrele(nd.ni_dvp); 1533 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1534 if (nd.ni_pathbuf != NULL) { 1535 pathbuf_destroy(nd.ni_pathbuf); 1536 nd.ni_pathbuf = NULL; 1537 } 1538 error = EINVAL; 1539 abort = 0; 1540 nfsm_reply(0); 1541 } 1542 } else { 1543 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1544 if (nd.ni_pathbuf != NULL) { 1545 pathbuf_destroy(nd.ni_pathbuf); 1546 nd.ni_pathbuf = NULL; 1547 } 1548 vput(nd.ni_dvp); 1549 error = ENXIO; 1550 abort = 0; 1551 } 1552 vp = nd.ni_vp; 1553 } else { 1554 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1555 if (nd.ni_pathbuf != NULL) { 1556 pathbuf_destroy(nd.ni_pathbuf); 1557 nd.ni_pathbuf = NULL; 1558 } 1559 vp = nd.ni_vp; 1560 if (nd.ni_dvp == vp) 1561 vrele(nd.ni_dvp); 1562 else 1563 vput(nd.ni_dvp); 1564 abort = 0; 1565 if (!error && va.va_size != -1) { 1566 error = nfsrv_access(vp, VWRITE, cred, 1567 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1568 if (!error) { 1569 nqsrv_getl(vp, ND_WRITE); 1570 tempsize = va.va_size; 1571 vattr_null(&va); 1572 va.va_size = tempsize; 1573 error = VOP_SETATTR(vp, &va, cred); 1574 } 1575 } 1576 if (error) 1577 vput(vp); 1578 } 1579 if (!error) { 1580 error = nfsrv_composefh(vp, &nsfh, v3); 1581 if (!error) 1582 error = VOP_GETATTR(vp, &va, cred); 1583 vput(vp); 1584 } 1585 if (v3) { 1586 if (exclusive_flag && !error) { 1587 /* 1588 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1589 */ 1590 char oldverf[NFSX_V3CREATEVERF]; 1591 1592 txdr_nfsv3time(&va.va_atime, oldverf); 1593 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1594 error = EEXIST; 1595 } 1596 if (dirp) { 1597 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1598 } 1599 } 1600 if (dirp) { 1601 vrele(dirp); 1602 dirp = NULL; 1603 } 1604 if (nd.ni_pathbuf != NULL) { 1605 pathbuf_destroy(nd.ni_pathbuf); 1606 nd.ni_pathbuf = NULL; 1607 } 1608 abort = 0; 1609 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1610 if (v3) { 1611 if (!error) { 1612 nfsm_srvpostop_fh(&nsfh); 1613 nfsm_srvpostop_attr(0, &va); 1614 } 1615 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1616 } else { 1617 nfsm_srvfhtom(&nsfh, v3); 1618 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1619 nfsm_srvfillattr(&va, fp); 1620 } 1621 return (0); 1622 nfsmout: 1623 if (dirp) 1624 vrele(dirp); 1625 if (abort) { 1626 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1627 if (nd.ni_dvp == nd.ni_vp) 1628 vrele(nd.ni_dvp); 1629 else 1630 vput(nd.ni_dvp); 1631 if (nd.ni_vp) 1632 vput(nd.ni_vp); 1633 } 1634 if (nd.ni_pathbuf != NULL) { 1635 pathbuf_destroy(nd.ni_pathbuf); 1636 nd.ni_pathbuf = NULL; 1637 } 1638 return (error); 1639 } 1640 1641 /* 1642 * nfs v3 mknod service 1643 */ 1644 int 1645 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1646 { 1647 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1648 struct mbuf *nam = nfsd->nd_nam; 1649 char *dpos = nfsd->nd_dpos; 1650 kauth_cred_t cred = nfsd->nd_cr; 1651 struct vattr va, dirfor, diraft; 1652 u_int32_t *tl; 1653 struct nameidata nd; 1654 int32_t t1; 1655 char *bpos; 1656 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1657 int abort = 0; 1658 u_int32_t major, minor; 1659 enum vtype vtyp; 1660 char *cp2; 1661 struct mbuf *mb, *mreq; 1662 struct vnode *vp, *dirp = (struct vnode *)0; 1663 nfsrvfh_t nsfh; 1664 u_quad_t frev; 1665 1666 nd.ni_cnd.cn_nameiop = 0; 1667 nfsm_srvmtofh(&nsfh); 1668 nfsm_srvnamesiz(len); 1669 nd.ni_cnd.cn_cred = cred; 1670 nd.ni_cnd.cn_nameiop = CREATE; 1671 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1672 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1673 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1674 if (dirp) 1675 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1676 if (error) { 1677 nfsm_reply(NFSX_WCCDATA(1)); 1678 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1679 if (dirp) 1680 vrele(dirp); 1681 if (nd.ni_pathbuf != NULL) { 1682 pathbuf_destroy(nd.ni_pathbuf); 1683 nd.ni_pathbuf = NULL; 1684 } 1685 return (0); 1686 } 1687 abort = 1; 1688 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1689 vtyp = nfsv3tov_type(*tl); 1690 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1691 error = NFSERR_BADTYPE; 1692 goto abort; 1693 } 1694 vattr_null(&va); 1695 va.va_mode = 0; 1696 nfsm_srvsattr(&va); 1697 if (vtyp == VCHR || vtyp == VBLK) { 1698 dev_t rdev; 1699 1700 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1701 major = fxdr_unsigned(u_int32_t, *tl++); 1702 minor = fxdr_unsigned(u_int32_t, *tl); 1703 rdev = makedev(major, minor); 1704 if (major(rdev) != major || minor(rdev) != minor) { 1705 error = EINVAL; 1706 goto abort; 1707 } 1708 va.va_rdev = rdev; 1709 } 1710 1711 /* 1712 * Iff doesn't exist, create it. 1713 */ 1714 if (nd.ni_vp) { 1715 error = EEXIST; 1716 abort: 1717 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1718 if (nd.ni_dvp == nd.ni_vp) 1719 vrele(nd.ni_dvp); 1720 else 1721 vput(nd.ni_dvp); 1722 if (nd.ni_vp) 1723 vput(nd.ni_vp); 1724 if (nd.ni_pathbuf != NULL) { 1725 pathbuf_destroy(nd.ni_pathbuf); 1726 nd.ni_pathbuf = NULL; 1727 } 1728 goto out; 1729 } 1730 va.va_type = vtyp; 1731 if (vtyp == VSOCK) { 1732 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1733 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1734 } else { 1735 if (va.va_type != VFIFO && 1736 (error = kauth_authorize_system(cred, 1737 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1738 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1739 vput(nd.ni_dvp); 1740 goto out; 1741 } 1742 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1743 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1744 if (error) 1745 goto out; 1746 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1747 vput(nd.ni_vp); 1748 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1749 error = EINVAL; 1750 } 1751 } 1752 out: 1753 vp = nd.ni_vp; 1754 if (!error) { 1755 error = nfsrv_composefh(vp, &nsfh, true); 1756 if (!error) 1757 error = VOP_GETATTR(vp, &va, cred); 1758 vput(vp); 1759 } 1760 if (dirp) { 1761 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1762 vrele(dirp); 1763 dirp = NULL; 1764 } 1765 if (nd.ni_pathbuf != NULL) { 1766 pathbuf_destroy(nd.ni_pathbuf); 1767 nd.ni_pathbuf = NULL; 1768 } 1769 abort = 0; 1770 nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) + 1771 NFSX_WCCDATA(1)); 1772 if (!error) { 1773 nfsm_srvpostop_fh(&nsfh); 1774 nfsm_srvpostop_attr(0, &va); 1775 } 1776 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1777 return (0); 1778 nfsmout: 1779 if (abort) { 1780 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1781 if (nd.ni_dvp == nd.ni_vp) 1782 vrele(nd.ni_dvp); 1783 else 1784 vput(nd.ni_dvp); 1785 if (nd.ni_vp) 1786 vput(nd.ni_vp); 1787 } 1788 if (nd.ni_pathbuf != NULL) { 1789 pathbuf_destroy(nd.ni_pathbuf); 1790 nd.ni_pathbuf = NULL; 1791 } 1792 if (dirp) 1793 vrele(dirp); 1794 return (error); 1795 } 1796 1797 /* 1798 * nfs remove service 1799 */ 1800 int 1801 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1802 { 1803 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1804 struct mbuf *nam = nfsd->nd_nam; 1805 char *dpos = nfsd->nd_dpos; 1806 kauth_cred_t cred = nfsd->nd_cr; 1807 struct nameidata nd; 1808 u_int32_t *tl; 1809 int32_t t1; 1810 char *bpos; 1811 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1812 int v3 = (nfsd->nd_flag & ND_NFSV3); 1813 char *cp2; 1814 struct mbuf *mb, *mreq; 1815 struct vnode *vp, *dirp; 1816 struct vattr dirfor, diraft; 1817 nfsrvfh_t nsfh; 1818 u_quad_t frev; 1819 1820 #ifndef nolint 1821 vp = (struct vnode *)0; 1822 #endif 1823 nfsm_srvmtofh(&nsfh); 1824 nfsm_srvnamesiz(len); 1825 nd.ni_cnd.cn_cred = cred; 1826 nd.ni_cnd.cn_nameiop = DELETE; 1827 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1828 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1829 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1830 if (dirp && v3) { 1831 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1832 } 1833 if (!error) { 1834 vp = nd.ni_vp; 1835 if (vp->v_type == VDIR) { 1836 error = EPERM; 1837 goto out; 1838 } 1839 /* 1840 * The root of a mounted filesystem cannot be deleted. 1841 */ 1842 if (vp->v_vflag & VV_ROOT) { 1843 error = EBUSY; 1844 } 1845 out: 1846 if (!error) { 1847 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1848 nqsrv_getl(vp, ND_WRITE); 1849 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1850 } else { 1851 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1852 if (nd.ni_dvp == vp) 1853 vrele(nd.ni_dvp); 1854 else 1855 vput(nd.ni_dvp); 1856 vput(vp); 1857 } 1858 } 1859 if (nd.ni_pathbuf != NULL) { 1860 pathbuf_destroy(nd.ni_pathbuf); 1861 nd.ni_pathbuf = NULL; 1862 } 1863 if (dirp) { 1864 if (v3) { 1865 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1866 } 1867 vrele(dirp); 1868 } 1869 nfsm_reply(NFSX_WCCDATA(v3)); 1870 if (v3) { 1871 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1872 return (0); 1873 } 1874 nfsm_srvdone; 1875 } 1876 1877 /* 1878 * nfs rename service 1879 */ 1880 int 1881 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1882 { 1883 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1884 struct mbuf *nam = nfsd->nd_nam; 1885 char *dpos = nfsd->nd_dpos; 1886 kauth_cred_t cred = nfsd->nd_cr; 1887 u_int32_t *tl; 1888 int32_t t1; 1889 char *bpos; 1890 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1891 uint32_t len, len2; 1892 int tdirfor_ret = 1, tdiraft_ret = 1; 1893 int v3 = (nfsd->nd_flag & ND_NFSV3); 1894 char *cp2; 1895 struct mbuf *mb, *mreq; 1896 struct nameidata fromnd, tond; 1897 struct vnode *fvp, *tvp, *tdvp; 1898 struct vnode *fdirp = NULL, *tdirp = NULL; 1899 struct mount *localfs = NULL; 1900 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1901 nfsrvfh_t fnsfh, tnsfh; 1902 u_quad_t frev; 1903 uid_t saved_uid; 1904 1905 #ifndef nolint 1906 fvp = (struct vnode *)0; 1907 #endif 1908 fromnd.ni_cnd.cn_nameiop = 0; 1909 tond.ni_cnd.cn_nameiop = 0; 1910 nfsm_srvmtofh(&fnsfh); 1911 nfsm_srvnamesiz(len); 1912 /* 1913 * Remember our original uid so that we can reset cr_uid before 1914 * the second nfs_namei() call, in case it is remapped. 1915 */ 1916 saved_uid = kauth_cred_geteuid(cred); 1917 fromnd.ni_cnd.cn_cred = cred; 1918 fromnd.ni_cnd.cn_nameiop = DELETE; 1919 fromnd.ni_cnd.cn_flags = LOCKPARENT | INRENAME; 1920 error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md, 1921 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1922 if (fdirp && v3) { 1923 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred); 1924 } 1925 if (error) { 1926 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1927 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1928 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1929 if (fdirp) 1930 vrele(fdirp); 1931 pathbuf_destroy(fromnd.ni_pathbuf); 1932 return (0); 1933 } 1934 if (fromnd.ni_dvp != fromnd.ni_vp) { 1935 VOP_UNLOCK(fromnd.ni_dvp); 1936 } 1937 fvp = fromnd.ni_vp; 1938 1939 localfs = fvp->v_mount; 1940 error = VFS_RENAMELOCK_ENTER(localfs); 1941 if (error) { 1942 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1943 vrele(fromnd.ni_dvp); 1944 vrele(fvp); 1945 goto out1; 1946 } 1947 1948 /* Copied, regrettably, from vfs_syscalls.c (q.v.) */ 1949 vrele(fvp); 1950 if ((fromnd.ni_cnd.cn_namelen == 1 && 1951 fromnd.ni_cnd.cn_nameptr[0] == '.') || 1952 (fromnd.ni_cnd.cn_namelen == 2 && 1953 fromnd.ni_cnd.cn_nameptr[0] == '.' && 1954 fromnd.ni_cnd.cn_nameptr[1] == '.')) { 1955 error = EINVAL; 1956 VFS_RENAMELOCK_EXIT(localfs); 1957 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1958 vrele(fromnd.ni_dvp); 1959 goto out1; 1960 } 1961 vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY); 1962 error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0); 1963 if (error) { 1964 VOP_UNLOCK(fromnd.ni_dvp); 1965 VFS_RENAMELOCK_EXIT(localfs); 1966 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1967 vrele(fromnd.ni_dvp); 1968 goto out1; 1969 } 1970 VOP_UNLOCK(fromnd.ni_vp); 1971 if (fromnd.ni_dvp != fromnd.ni_vp) 1972 VOP_UNLOCK(fromnd.ni_dvp); 1973 fvp = fromnd.ni_vp; 1974 1975 nfsm_srvmtofh(&tnsfh); 1976 if (v3) { 1977 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1978 len2 = fxdr_unsigned(uint32_t, *tl); 1979 /* len2 will be checked by nfs_namei */ 1980 } 1981 else { 1982 /* NFSv2 */ 1983 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1984 } 1985 kauth_cred_seteuid(cred, saved_uid); 1986 tond.ni_cnd.cn_cred = cred; 1987 tond.ni_cnd.cn_nameiop = RENAME; 1988 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | INRENAME; 1989 error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md, 1990 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1991 if (tdirp && v3) { 1992 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred); 1993 } 1994 if (error) { 1995 VFS_RENAMELOCK_EXIT(localfs); 1996 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1997 vrele(fromnd.ni_dvp); 1998 vrele(fvp); 1999 goto out1; 2000 } 2001 tdvp = tond.ni_dvp; 2002 tvp = tond.ni_vp; 2003 if (tvp != NULL) { 2004 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 2005 if (v3) 2006 error = EEXIST; 2007 else 2008 error = EISDIR; 2009 goto out; 2010 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 2011 if (v3) 2012 error = EEXIST; 2013 else 2014 error = ENOTDIR; 2015 goto out; 2016 } 2017 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 2018 if (v3) 2019 error = EXDEV; 2020 else 2021 error = ENOTEMPTY; 2022 goto out; 2023 } 2024 } 2025 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 2026 if (v3) 2027 error = EXDEV; 2028 else 2029 error = ENOTEMPTY; 2030 goto out; 2031 } 2032 if (fvp->v_mount != tdvp->v_mount) { 2033 if (v3) 2034 error = EXDEV; 2035 else 2036 error = ENOTEMPTY; 2037 goto out; 2038 } 2039 if (fvp == tdvp) { 2040 if (v3) 2041 error = EINVAL; 2042 else 2043 error = ENOTEMPTY; 2044 } 2045 /* 2046 * If source is the same as the destination (that is the 2047 * same vnode with the same name in the same directory), 2048 * then there is nothing to do. 2049 */ 2050 if (fvp == tvp && fromnd.ni_dvp == tdvp && 2051 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 2052 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 2053 fromnd.ni_cnd.cn_namelen)) 2054 error = -1; 2055 out: 2056 if (!error) { 2057 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 2058 nqsrv_getl(tdvp, ND_WRITE); 2059 if (tvp) { 2060 nqsrv_getl(tvp, ND_WRITE); 2061 } 2062 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 2063 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 2064 VFS_RENAMELOCK_EXIT(localfs); 2065 } else { 2066 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 2067 if (tdvp == tvp) 2068 vrele(tdvp); 2069 else 2070 vput(tdvp); 2071 if (tvp) 2072 vput(tvp); 2073 VFS_RENAMELOCK_EXIT(localfs); 2074 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2075 vrele(fromnd.ni_dvp); 2076 vrele(fvp); 2077 if (error == -1) 2078 error = 0; 2079 } 2080 pathbuf_destroy(tond.ni_pathbuf); 2081 tond.ni_cnd.cn_nameiop = 0; 2082 out1: 2083 if (fdirp) { 2084 if (v3) { 2085 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred); 2086 } 2087 vrele(fdirp); 2088 fdirp = NULL; 2089 } 2090 if (tdirp) { 2091 if (v3) { 2092 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred); 2093 } 2094 vrele(tdirp); 2095 tdirp = NULL; 2096 } 2097 pathbuf_destroy(fromnd.ni_pathbuf); 2098 fromnd.ni_cnd.cn_nameiop = 0; 2099 localfs = NULL; 2100 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2101 if (v3) { 2102 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2103 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2104 } 2105 return (0); 2106 2107 nfsmout: 2108 if (fdirp) 2109 vrele(fdirp); 2110 #ifdef notdef 2111 if (tdirp) 2112 vrele(tdirp); 2113 #endif 2114 if (tond.ni_cnd.cn_nameiop) { 2115 pathbuf_destroy(tond.ni_pathbuf); 2116 } 2117 if (localfs) { 2118 VFS_RENAMELOCK_EXIT(localfs); 2119 } 2120 if (fromnd.ni_cnd.cn_nameiop) { 2121 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2122 pathbuf_destroy(fromnd.ni_pathbuf); 2123 vrele(fromnd.ni_dvp); 2124 vrele(fvp); 2125 } 2126 return (error); 2127 } 2128 2129 /* 2130 * nfs link service 2131 */ 2132 int 2133 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2134 { 2135 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2136 struct mbuf *nam = nfsd->nd_nam; 2137 char *dpos = nfsd->nd_dpos; 2138 kauth_cred_t cred = nfsd->nd_cr; 2139 struct nameidata nd; 2140 u_int32_t *tl; 2141 int32_t t1; 2142 char *bpos; 2143 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2144 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2145 char *cp2; 2146 struct mbuf *mb, *mreq; 2147 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2148 struct vattr dirfor, diraft, at; 2149 nfsrvfh_t nsfh, dnsfh; 2150 u_quad_t frev; 2151 2152 nfsm_srvmtofh(&nsfh); 2153 nfsm_srvmtofh(&dnsfh); 2154 nfsm_srvnamesiz(len); 2155 error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam, 2156 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2157 if (error) { 2158 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2159 nfsm_srvpostop_attr(getret, &at); 2160 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2161 return (0); 2162 } 2163 if (vp->v_type == VDIR) { 2164 error = EPERM; 2165 goto out1; 2166 } 2167 nd.ni_cnd.cn_cred = cred; 2168 nd.ni_cnd.cn_nameiop = CREATE; 2169 nd.ni_cnd.cn_flags = LOCKPARENT; 2170 error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos, 2171 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2172 if (dirp && v3) { 2173 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2174 } 2175 if (error) 2176 goto out1; 2177 xp = nd.ni_vp; 2178 if (xp != NULL) { 2179 error = EEXIST; 2180 goto out; 2181 } 2182 xp = nd.ni_dvp; 2183 if (vp->v_mount != xp->v_mount) 2184 error = EXDEV; 2185 out: 2186 if (!error) { 2187 nqsrv_getl(vp, ND_WRITE); 2188 nqsrv_getl(xp, ND_WRITE); 2189 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2190 } else { 2191 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2192 if (nd.ni_dvp == nd.ni_vp) 2193 vrele(nd.ni_dvp); 2194 else 2195 vput(nd.ni_dvp); 2196 if (nd.ni_vp) 2197 vrele(nd.ni_vp); 2198 } 2199 out1: 2200 if (v3) 2201 getret = VOP_GETATTR(vp, &at, cred); 2202 if (dirp) { 2203 if (v3) { 2204 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2205 } 2206 vrele(dirp); 2207 } 2208 vrele(vp); 2209 pathbuf_destroy(nd.ni_pathbuf); 2210 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2211 if (v3) { 2212 nfsm_srvpostop_attr(getret, &at); 2213 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2214 return (0); 2215 } 2216 nfsm_srvdone; 2217 } 2218 2219 /* 2220 * nfs symbolic link service 2221 */ 2222 int 2223 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2224 { 2225 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2226 struct mbuf *nam = nfsd->nd_nam; 2227 char *dpos = nfsd->nd_dpos; 2228 kauth_cred_t cred = nfsd->nd_cr; 2229 struct vattr va, dirfor, diraft; 2230 struct nameidata nd; 2231 u_int32_t *tl; 2232 int32_t t1; 2233 struct nfsv2_sattr *sp; 2234 char *bpos, *pathcp = NULL, *cp2; 2235 struct uio io; 2236 struct iovec iv; 2237 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0; 2238 uint32_t len, len2; 2239 int v3 = (nfsd->nd_flag & ND_NFSV3); 2240 struct mbuf *mb, *mreq; 2241 struct vnode *dirp = (struct vnode *)0; 2242 nfsrvfh_t nsfh; 2243 u_quad_t frev; 2244 2245 nd.ni_cnd.cn_nameiop = 0; 2246 nfsm_srvmtofh(&nsfh); 2247 nfsm_srvnamesiz(len); 2248 nd.ni_cnd.cn_cred = cred; 2249 nd.ni_cnd.cn_nameiop = CREATE; 2250 nd.ni_cnd.cn_flags = LOCKPARENT; 2251 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2252 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2253 if (dirp && v3) { 2254 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2255 } 2256 if (error) 2257 goto out; 2258 abort = 1; 2259 vattr_null(&va); 2260 va.va_type = VLNK; 2261 if (v3) { 2262 va.va_mode = 0; 2263 nfsm_srvsattr(&va); 2264 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2265 len2 = fxdr_unsigned(uint32_t, *tl); 2266 if (len2 > PATH_MAX) { 2267 /* XXX should check _PC_NO_TRUNC */ 2268 error = ENAMETOOLONG; 2269 goto abortop; 2270 } 2271 } 2272 else { 2273 /* NFSv2 */ 2274 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2275 } 2276 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2277 iv.iov_base = pathcp; 2278 iv.iov_len = len2; 2279 io.uio_resid = len2; 2280 io.uio_offset = 0; 2281 io.uio_iov = &iv; 2282 io.uio_iovcnt = 1; 2283 io.uio_rw = UIO_READ; 2284 UIO_SETUP_SYSSPACE(&io); 2285 nfsm_mtouio(&io, len2); 2286 if (!v3) { 2287 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2288 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2289 } 2290 *(pathcp + len2) = '\0'; 2291 if (nd.ni_vp) { 2292 error = EEXIST; 2293 abortop: 2294 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2295 if (nd.ni_dvp == nd.ni_vp) 2296 vrele(nd.ni_dvp); 2297 else 2298 vput(nd.ni_dvp); 2299 if (nd.ni_vp) 2300 vrele(nd.ni_vp); 2301 goto out; 2302 } 2303 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2304 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2305 if (!error) { 2306 if (v3) { 2307 error = nfsrv_composefh(nd.ni_vp, &nsfh, v3); 2308 if (!error) 2309 error = VOP_GETATTR(nd.ni_vp, &va, cred); 2310 vput(nd.ni_vp); 2311 } else { 2312 vput(nd.ni_vp); 2313 } 2314 } 2315 out: 2316 if (pathcp) 2317 free(pathcp, M_TEMP); 2318 if (dirp) { 2319 if (v3) { 2320 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2321 } 2322 vrele(dirp); 2323 dirp = NULL; 2324 } 2325 pathbuf_destroy(nd.ni_pathbuf); 2326 abort = 0; 2327 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2328 NFSX_WCCDATA(v3)); 2329 if (v3) { 2330 if (!error) { 2331 nfsm_srvpostop_fh(&nsfh); 2332 nfsm_srvpostop_attr(0, &va); 2333 } 2334 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2335 } 2336 return (0); 2337 nfsmout: 2338 if (abort) { 2339 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2340 if (nd.ni_dvp == nd.ni_vp) 2341 vrele(nd.ni_dvp); 2342 else 2343 vput(nd.ni_dvp); 2344 if (nd.ni_vp) 2345 vrele(nd.ni_vp); 2346 pathbuf_destroy(nd.ni_pathbuf); 2347 } 2348 if (dirp) 2349 vrele(dirp); 2350 if (pathcp) 2351 free(pathcp, M_TEMP); 2352 return (error); 2353 } 2354 2355 /* 2356 * nfs mkdir service 2357 */ 2358 int 2359 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2360 { 2361 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2362 struct mbuf *nam = nfsd->nd_nam; 2363 char *dpos = nfsd->nd_dpos; 2364 kauth_cred_t cred = nfsd->nd_cr; 2365 struct vattr va, dirfor, diraft; 2366 struct nfs_fattr *fp; 2367 struct nameidata nd; 2368 char *cp; 2369 u_int32_t *tl; 2370 int32_t t1; 2371 char *bpos; 2372 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2373 int abort = 0; 2374 int v3 = (nfsd->nd_flag & ND_NFSV3); 2375 char *cp2; 2376 struct mbuf *mb, *mreq; 2377 struct vnode *vp, *dirp = (struct vnode *)0; 2378 nfsrvfh_t nsfh; 2379 u_quad_t frev; 2380 2381 nfsm_srvmtofh(&nsfh); 2382 nfsm_srvnamesiz(len); 2383 nd.ni_cnd.cn_cred = cred; 2384 nd.ni_cnd.cn_nameiop = CREATE; 2385 nd.ni_cnd.cn_flags = LOCKPARENT; 2386 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2387 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2388 if (dirp && v3) { 2389 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2390 } 2391 if (error) { 2392 if (nd.ni_pathbuf != NULL) { 2393 pathbuf_destroy(nd.ni_pathbuf); 2394 } 2395 nfsm_reply(NFSX_WCCDATA(v3)); 2396 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2397 if (dirp) 2398 vrele(dirp); 2399 return (0); 2400 } 2401 abort = 1; 2402 vattr_null(&va); 2403 if (v3) { 2404 va.va_mode = 0; 2405 nfsm_srvsattr(&va); 2406 } else { 2407 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2408 va.va_mode = nfstov_mode(*tl++); 2409 } 2410 va.va_type = VDIR; 2411 vp = nd.ni_vp; 2412 if (vp != NULL) { 2413 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2414 if (nd.ni_dvp == vp) 2415 vrele(nd.ni_dvp); 2416 else 2417 vput(nd.ni_dvp); 2418 vrele(vp); 2419 error = EEXIST; 2420 goto out; 2421 } 2422 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2423 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2424 if (!error) { 2425 vp = nd.ni_vp; 2426 error = nfsrv_composefh(vp, &nsfh, v3); 2427 if (!error) 2428 error = VOP_GETATTR(vp, &va, cred); 2429 vput(vp); 2430 } 2431 out: 2432 if (dirp) { 2433 if (v3) { 2434 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2435 } 2436 vrele(dirp); 2437 dirp = NULL; 2438 } 2439 pathbuf_destroy(nd.ni_pathbuf); 2440 abort = 0; 2441 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2442 NFSX_WCCDATA(v3)); 2443 if (v3) { 2444 if (!error) { 2445 nfsm_srvpostop_fh(&nsfh); 2446 nfsm_srvpostop_attr(0, &va); 2447 } 2448 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2449 } else { 2450 nfsm_srvfhtom(&nsfh, v3); 2451 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2452 nfsm_srvfillattr(&va, fp); 2453 } 2454 return (0); 2455 nfsmout: 2456 if (abort) { 2457 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2458 if (nd.ni_dvp == nd.ni_vp) 2459 vrele(nd.ni_dvp); 2460 else 2461 vput(nd.ni_dvp); 2462 if (nd.ni_vp) 2463 vrele(nd.ni_vp); 2464 pathbuf_destroy(nd.ni_pathbuf); 2465 } 2466 if (dirp) 2467 vrele(dirp); 2468 return (error); 2469 } 2470 2471 /* 2472 * nfs rmdir service 2473 */ 2474 int 2475 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2476 { 2477 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2478 struct mbuf *nam = nfsd->nd_nam; 2479 char *dpos = nfsd->nd_dpos; 2480 kauth_cred_t cred = nfsd->nd_cr; 2481 u_int32_t *tl; 2482 int32_t t1; 2483 char *bpos; 2484 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2485 int v3 = (nfsd->nd_flag & ND_NFSV3); 2486 char *cp2; 2487 struct mbuf *mb, *mreq; 2488 struct vnode *vp, *dirp = (struct vnode *)0; 2489 struct vattr dirfor, diraft; 2490 nfsrvfh_t nsfh; 2491 struct nameidata nd; 2492 u_quad_t frev; 2493 2494 nfsm_srvmtofh(&nsfh); 2495 nfsm_srvnamesiz(len); 2496 nd.ni_cnd.cn_cred = cred; 2497 nd.ni_cnd.cn_nameiop = DELETE; 2498 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2499 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2500 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2501 if (dirp && v3) { 2502 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2503 } 2504 if (error) { 2505 if (nd.ni_pathbuf != NULL) { 2506 pathbuf_destroy(nd.ni_pathbuf); 2507 } 2508 nfsm_reply(NFSX_WCCDATA(v3)); 2509 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2510 if (dirp) 2511 vrele(dirp); 2512 return (0); 2513 } 2514 vp = nd.ni_vp; 2515 if (vp->v_type != VDIR) { 2516 error = ENOTDIR; 2517 goto out; 2518 } 2519 /* 2520 * No rmdir "." please. 2521 */ 2522 if (nd.ni_dvp == vp) { 2523 error = EINVAL; 2524 goto out; 2525 } 2526 /* 2527 * The root of a mounted filesystem cannot be deleted. 2528 */ 2529 if (vp->v_vflag & VV_ROOT) 2530 error = EBUSY; 2531 out: 2532 if (!error) { 2533 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2534 nqsrv_getl(vp, ND_WRITE); 2535 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2536 } else { 2537 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2538 if (nd.ni_dvp == nd.ni_vp) 2539 vrele(nd.ni_dvp); 2540 else 2541 vput(nd.ni_dvp); 2542 vput(vp); 2543 } 2544 pathbuf_destroy(nd.ni_pathbuf); 2545 if (dirp) { 2546 if (v3) { 2547 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2548 } 2549 vrele(dirp); 2550 } 2551 nfsm_reply(NFSX_WCCDATA(v3)); 2552 if (v3) { 2553 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2554 return (0); 2555 } 2556 nfsm_srvdone; 2557 } 2558 2559 /* 2560 * nfs readdir service 2561 * - mallocs what it thinks is enough to read 2562 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2563 * - calls VOP_READDIR() 2564 * - loops around building the reply 2565 * if the output generated exceeds count break out of loop 2566 * The nfsm_clget macro is used here so that the reply will be packed 2567 * tightly in mbuf clusters. 2568 * - it only knows that it has encountered eof when the VOP_READDIR() 2569 * reads nothing 2570 * - as such one readdir rpc will return eof false although you are there 2571 * and then the next will return eof 2572 * - it trims out records with d_fileno == 0 2573 * this doesn't matter for Unix clients, but they might confuse clients 2574 * for other os'. 2575 * - it trims out records with d_type == DT_WHT 2576 * these cannot be seen through NFS (unless we extend the protocol) 2577 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2578 * than requested, but this may not apply to all filesystems. For 2579 * example, client NFS does not { although it is never remote mounted 2580 * anyhow } 2581 * The alternate call nfsrv_readdirplus() does lookups as well. 2582 * PS: The NFS protocol spec. does not clarify what the "count" byte 2583 * argument is a count of.. just name strings and file id's or the 2584 * entire reply rpc or ... 2585 * I tried just file name and id sizes and it confused the Sun client, 2586 * so I am using the full rpc size now. The "paranoia.." comment refers 2587 * to including the status longwords that are not a part of the dir. 2588 * "entry" structures, but are in the rpc. 2589 */ 2590 2591 #define NFS_SRVDIRBLKSIZ 1024 2592 2593 struct flrep { 2594 nfsuint64 fl_off; 2595 u_int32_t fl_postopok; 2596 struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */ 2597 u_int32_t fl_fhok; 2598 u_int32_t fl_fhsize; 2599 /* handle comes here, filled in dynamically */ 2600 }; 2601 2602 int 2603 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2604 { 2605 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2606 struct mbuf *nam = nfsd->nd_nam; 2607 char *dpos = nfsd->nd_dpos; 2608 kauth_cred_t cred = nfsd->nd_cr; 2609 char *bp, *be; 2610 struct mbuf *mp; 2611 struct dirent *dp; 2612 char *cp; 2613 u_int32_t *tl; 2614 int32_t t1; 2615 char *bpos; 2616 struct mbuf *mb, *mreq, *mp2; 2617 char *cpos, *cend, *cp2, *rbuf; 2618 struct vnode *vp; 2619 struct vattr at; 2620 nfsrvfh_t nsfh; 2621 struct uio io; 2622 struct iovec iv; 2623 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2624 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2625 int v3 = (nfsd->nd_flag & ND_NFSV3); 2626 u_quad_t frev, off, toff, verf; 2627 off_t *cookies = NULL, *cookiep; 2628 nfsuint64 jar; 2629 2630 nfsm_srvmtofh(&nsfh); 2631 if (v3) { 2632 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2633 toff = fxdr_hyper(tl); 2634 tl += 2; 2635 verf = fxdr_hyper(tl); 2636 tl += 2; 2637 } else { 2638 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2639 toff = fxdr_unsigned(u_quad_t, *tl++); 2640 } 2641 off = toff; 2642 cnt = fxdr_unsigned(int, *tl); 2643 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2644 xfer = NFS_SRVMAXDATA(nfsd); 2645 if (siz > xfer) 2646 siz = xfer; 2647 fullsiz = siz; 2648 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2649 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2650 if (!error && vp->v_type != VDIR) { 2651 error = ENOTDIR; 2652 vput(vp); 2653 } 2654 if (error) { 2655 nfsm_reply(NFSX_UNSIGNED); 2656 nfsm_srvpostop_attr(getret, &at); 2657 return (0); 2658 } 2659 nqsrv_getl(vp, ND_READ); 2660 if (v3) { 2661 error = getret = VOP_GETATTR(vp, &at, cred); 2662 #ifdef NFS3_STRICTVERF 2663 /* 2664 * XXX This check is too strict for Solaris 2.5 clients. 2665 */ 2666 if (!error && toff && verf != at.va_filerev) 2667 error = NFSERR_BAD_COOKIE; 2668 #endif 2669 } 2670 if (!error) 2671 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2672 if (error) { 2673 vput(vp); 2674 nfsm_reply(NFSX_POSTOPATTR(v3)); 2675 nfsm_srvpostop_attr(getret, &at); 2676 return (0); 2677 } 2678 VOP_UNLOCK(vp); 2679 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2680 again: 2681 iv.iov_base = rbuf; 2682 iv.iov_len = fullsiz; 2683 io.uio_iov = &iv; 2684 io.uio_iovcnt = 1; 2685 io.uio_offset = (off_t)off; 2686 io.uio_resid = fullsiz; 2687 io.uio_rw = UIO_READ; 2688 UIO_SETUP_SYSSPACE(&io); 2689 eofflag = 0; 2690 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2691 2692 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2693 2694 off = (off_t)io.uio_offset; 2695 if (!cookies && !error) 2696 error = NFSERR_PERM; 2697 if (v3) { 2698 getret = VOP_GETATTR(vp, &at, cred); 2699 if (!error) 2700 error = getret; 2701 } 2702 2703 VOP_UNLOCK(vp); 2704 if (error) { 2705 vrele(vp); 2706 free((void *)rbuf, M_TEMP); 2707 if (cookies) 2708 free((void *)cookies, M_TEMP); 2709 nfsm_reply(NFSX_POSTOPATTR(v3)); 2710 nfsm_srvpostop_attr(getret, &at); 2711 return (0); 2712 } 2713 if (io.uio_resid) { 2714 siz -= io.uio_resid; 2715 2716 /* 2717 * If nothing read, return eof 2718 * rpc reply 2719 */ 2720 if (siz == 0) { 2721 vrele(vp); 2722 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2723 2 * NFSX_UNSIGNED); 2724 if (v3) { 2725 nfsm_srvpostop_attr(getret, &at); 2726 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2727 txdr_hyper(at.va_filerev, tl); 2728 tl += 2; 2729 } else 2730 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2731 *tl++ = nfs_false; 2732 *tl = nfs_true; 2733 free((void *)rbuf, M_TEMP); 2734 free((void *)cookies, M_TEMP); 2735 return (0); 2736 } 2737 } 2738 2739 /* 2740 * Check for degenerate cases of nothing useful read. 2741 * If so go try again 2742 */ 2743 cpos = rbuf; 2744 cend = rbuf + siz; 2745 dp = (struct dirent *)cpos; 2746 cookiep = cookies; 2747 2748 while (cpos < cend && ncookies > 0 && 2749 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2750 cpos += dp->d_reclen; 2751 dp = (struct dirent *)cpos; 2752 cookiep++; 2753 ncookies--; 2754 } 2755 if (cpos >= cend || ncookies == 0) { 2756 toff = off; 2757 siz = fullsiz; 2758 free(cookies, M_TEMP); 2759 cookies = NULL; 2760 goto again; 2761 } 2762 2763 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2764 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2765 if (v3) { 2766 nfsm_srvpostop_attr(getret, &at); 2767 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2768 txdr_hyper(at.va_filerev, tl); 2769 } 2770 mp = mp2 = mb; 2771 bp = bpos; 2772 be = bp + M_TRAILINGSPACE(mp); 2773 2774 /* Loop through the records and build reply */ 2775 while (cpos < cend && ncookies > 0) { 2776 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2777 nlen = dp->d_namlen; 2778 rem = nfsm_rndup(nlen)-nlen; 2779 len += (4 * NFSX_UNSIGNED + nlen + rem); 2780 if (v3) 2781 len += 2 * NFSX_UNSIGNED; 2782 if (len > cnt) { 2783 eofflag = 0; 2784 break; 2785 } 2786 /* 2787 * Build the directory record xdr from 2788 * the dirent entry. 2789 */ 2790 nfsm_clget; 2791 *tl = nfs_true; 2792 bp += NFSX_UNSIGNED; 2793 if (v3) { 2794 nfsm_clget; 2795 *tl = txdr_unsigned(dp->d_fileno >> 32); 2796 bp += NFSX_UNSIGNED; 2797 } 2798 nfsm_clget; 2799 *tl = txdr_unsigned(dp->d_fileno); 2800 bp += NFSX_UNSIGNED; 2801 nfsm_clget; 2802 *tl = txdr_unsigned(nlen); 2803 bp += NFSX_UNSIGNED; 2804 2805 /* And loop around copying the name */ 2806 xfer = nlen; 2807 cp = dp->d_name; 2808 while (xfer > 0) { 2809 nfsm_clget; 2810 if ((bp+xfer) > be) 2811 tsiz = be-bp; 2812 else 2813 tsiz = xfer; 2814 memcpy(bp, cp, tsiz); 2815 bp += tsiz; 2816 xfer -= tsiz; 2817 if (xfer > 0) 2818 cp += tsiz; 2819 } 2820 /* And null pad to an int32_t boundary */ 2821 for (i = 0; i < rem; i++) 2822 *bp++ = '\0'; 2823 nfsm_clget; 2824 2825 /* Finish off the record */ 2826 txdr_hyper(*cookiep, &jar); 2827 if (v3) { 2828 *tl = jar.nfsuquad[0]; 2829 bp += NFSX_UNSIGNED; 2830 nfsm_clget; 2831 } 2832 *tl = jar.nfsuquad[1]; 2833 bp += NFSX_UNSIGNED; 2834 } 2835 cpos += dp->d_reclen; 2836 dp = (struct dirent *)cpos; 2837 cookiep++; 2838 ncookies--; 2839 } 2840 vrele(vp); 2841 nfsm_clget; 2842 *tl = nfs_false; 2843 bp += NFSX_UNSIGNED; 2844 nfsm_clget; 2845 if (eofflag) 2846 *tl = nfs_true; 2847 else 2848 *tl = nfs_false; 2849 bp += NFSX_UNSIGNED; 2850 if (mp != mb) { 2851 if (bp < be) 2852 mp->m_len = bp - mtod(mp, char *); 2853 } else 2854 mp->m_len += bp - bpos; 2855 free((void *)rbuf, M_TEMP); 2856 free((void *)cookies, M_TEMP); 2857 nfsm_srvdone; 2858 } 2859 2860 int 2861 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2862 { 2863 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2864 struct mbuf *nam = nfsd->nd_nam; 2865 char *dpos = nfsd->nd_dpos; 2866 kauth_cred_t cred = nfsd->nd_cr; 2867 char *bp, *be; 2868 struct mbuf *mp; 2869 struct dirent *dp; 2870 char *cp; 2871 u_int32_t *tl; 2872 int32_t t1; 2873 char *bpos; 2874 struct mbuf *mb, *mreq, *mp2; 2875 char *cpos, *cend, *cp2, *rbuf; 2876 struct vnode *vp, *nvp; 2877 struct flrep fl; 2878 nfsrvfh_t nsfh; 2879 struct uio io; 2880 struct iovec iv; 2881 struct vattr va, at, *vap = &va; 2882 struct nfs_fattr *fp; 2883 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2884 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2885 u_quad_t frev, off, toff, verf; 2886 off_t *cookies = NULL, *cookiep; 2887 2888 nfsm_srvmtofh(&nsfh); 2889 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2890 toff = fxdr_hyper(tl); 2891 tl += 2; 2892 verf = fxdr_hyper(tl); 2893 tl += 2; 2894 siz = fxdr_unsigned(int, *tl++); 2895 cnt = fxdr_unsigned(int, *tl); 2896 off = toff; 2897 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2898 xfer = NFS_SRVMAXDATA(nfsd); 2899 if (siz > xfer) 2900 siz = xfer; 2901 fullsiz = siz; 2902 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2903 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2904 if (!error && vp->v_type != VDIR) { 2905 error = ENOTDIR; 2906 vput(vp); 2907 } 2908 if (error) { 2909 nfsm_reply(NFSX_UNSIGNED); 2910 nfsm_srvpostop_attr(getret, &at); 2911 return (0); 2912 } 2913 error = getret = VOP_GETATTR(vp, &at, cred); 2914 #ifdef NFS3_STRICTVERF 2915 /* 2916 * XXX This check is too strict for Solaris 2.5 clients. 2917 */ 2918 if (!error && toff && verf != at.va_filerev) 2919 error = NFSERR_BAD_COOKIE; 2920 #endif 2921 if (!error) { 2922 nqsrv_getl(vp, ND_READ); 2923 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2924 } 2925 if (error) { 2926 vput(vp); 2927 nfsm_reply(NFSX_V3POSTOPATTR); 2928 nfsm_srvpostop_attr(getret, &at); 2929 return (0); 2930 } 2931 VOP_UNLOCK(vp); 2932 2933 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2934 again: 2935 iv.iov_base = rbuf; 2936 iv.iov_len = fullsiz; 2937 io.uio_iov = &iv; 2938 io.uio_iovcnt = 1; 2939 io.uio_offset = (off_t)off; 2940 io.uio_resid = fullsiz; 2941 io.uio_rw = UIO_READ; 2942 UIO_SETUP_SYSSPACE(&io); 2943 eofflag = 0; 2944 2945 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2946 2947 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2948 2949 off = (u_quad_t)io.uio_offset; 2950 getret = VOP_GETATTR(vp, &at, cred); 2951 2952 VOP_UNLOCK(vp); 2953 2954 /* 2955 * If the VGET operation doesn't work for this filesystem, 2956 * we can't support readdirplus. Returning NOTSUPP should 2957 * make clients fall back to plain readdir. 2958 * There's no need to check for VPTOFH as well, we wouldn't 2959 * even be here otherwise. 2960 */ 2961 if (!getret) { 2962 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2963 getret = (getret == EOPNOTSUPP) ? 2964 NFSERR_NOTSUPP : NFSERR_IO; 2965 else 2966 vput(nvp); 2967 } 2968 2969 if (!cookies && !error) 2970 error = NFSERR_PERM; 2971 if (!error) 2972 error = getret; 2973 if (error) { 2974 vrele(vp); 2975 if (cookies) 2976 free((void *)cookies, M_TEMP); 2977 free((void *)rbuf, M_TEMP); 2978 nfsm_reply(NFSX_V3POSTOPATTR); 2979 nfsm_srvpostop_attr(getret, &at); 2980 return (0); 2981 } 2982 if (io.uio_resid) { 2983 siz -= io.uio_resid; 2984 2985 /* 2986 * If nothing read, return eof 2987 * rpc reply 2988 */ 2989 if (siz == 0) { 2990 vrele(vp); 2991 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2992 2 * NFSX_UNSIGNED); 2993 nfsm_srvpostop_attr(getret, &at); 2994 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2995 txdr_hyper(at.va_filerev, tl); 2996 tl += 2; 2997 *tl++ = nfs_false; 2998 *tl = nfs_true; 2999 free((void *)cookies, M_TEMP); 3000 free((void *)rbuf, M_TEMP); 3001 return (0); 3002 } 3003 } 3004 3005 /* 3006 * Check for degenerate cases of nothing useful read. 3007 * If so go try again 3008 */ 3009 cpos = rbuf; 3010 cend = rbuf + siz; 3011 dp = (struct dirent *)cpos; 3012 cookiep = cookies; 3013 3014 while (cpos < cend && ncookies > 0 && 3015 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 3016 cpos += dp->d_reclen; 3017 dp = (struct dirent *)cpos; 3018 cookiep++; 3019 ncookies--; 3020 } 3021 if (cpos >= cend || ncookies == 0) { 3022 toff = off; 3023 siz = fullsiz; 3024 free(cookies, M_TEMP); 3025 cookies = NULL; 3026 goto again; 3027 } 3028 3029 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3030 nfsm_reply(cnt); 3031 nfsm_srvpostop_attr(getret, &at); 3032 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3033 txdr_hyper(at.va_filerev, tl); 3034 mp = mp2 = mb; 3035 bp = bpos; 3036 be = bp + M_TRAILINGSPACE(mp); 3037 3038 /* Loop through the records and build reply */ 3039 while (cpos < cend && ncookies > 0) { 3040 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3041 nfsrvfh_t nnsfh; 3042 3043 nlen = dp->d_namlen; 3044 rem = nfsm_rndup(nlen)-nlen; 3045 3046 /* 3047 * For readdir_and_lookup get the vnode using 3048 * the file number. 3049 */ 3050 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3051 goto invalid; 3052 if (nfsrv_composefh(nvp, &nnsfh, true)) { 3053 vput(nvp); 3054 goto invalid; 3055 } 3056 if (VOP_GETATTR(nvp, vap, cred)) { 3057 vput(nvp); 3058 goto invalid; 3059 } 3060 vput(nvp); 3061 3062 /* 3063 * If either the dircount or maxcount will be 3064 * exceeded, get out now. Both of these lengths 3065 * are calculated conservatively, including all 3066 * XDR overheads. 3067 */ 3068 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3069 NFSX_V3POSTOPATTR); 3070 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3071 if (len > cnt || dirlen > fullsiz) { 3072 eofflag = 0; 3073 break; 3074 } 3075 3076 /* 3077 * Build the directory record xdr from 3078 * the dirent entry. 3079 */ 3080 fp = (struct nfs_fattr *)&fl.fl_fattr; 3081 nfsm_srvfillattr(vap, fp); 3082 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3083 fl.fl_fhok = nfs_true; 3084 fl.fl_postopok = nfs_true; 3085 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3086 3087 nfsm_clget; 3088 *tl = nfs_true; 3089 bp += NFSX_UNSIGNED; 3090 nfsm_clget; 3091 *tl = txdr_unsigned(dp->d_fileno >> 32); 3092 bp += NFSX_UNSIGNED; 3093 nfsm_clget; 3094 *tl = txdr_unsigned(dp->d_fileno); 3095 bp += NFSX_UNSIGNED; 3096 nfsm_clget; 3097 *tl = txdr_unsigned(nlen); 3098 bp += NFSX_UNSIGNED; 3099 3100 /* And loop around copying the name */ 3101 xfer = nlen; 3102 cp = dp->d_name; 3103 while (xfer > 0) { 3104 nfsm_clget; 3105 if ((bp + xfer) > be) 3106 tsiz = be - bp; 3107 else 3108 tsiz = xfer; 3109 memcpy(bp, cp, tsiz); 3110 bp += tsiz; 3111 xfer -= tsiz; 3112 if (xfer > 0) 3113 cp += tsiz; 3114 } 3115 /* And null pad to an int32_t boundary */ 3116 for (i = 0; i < rem; i++) 3117 *bp++ = '\0'; 3118 3119 /* 3120 * Now copy the flrep structure out. 3121 */ 3122 xfer = sizeof(struct flrep); 3123 cp = (void *)&fl; 3124 while (xfer > 0) { 3125 nfsm_clget; 3126 if ((bp + xfer) > be) 3127 tsiz = be - bp; 3128 else 3129 tsiz = xfer; 3130 memcpy(bp, cp, tsiz); 3131 bp += tsiz; 3132 xfer -= tsiz; 3133 if (xfer > 0) 3134 cp += tsiz; 3135 } 3136 3137 /* 3138 * ... and filehandle. 3139 */ 3140 xfer = NFSRVFH_SIZE(&nnsfh); 3141 cp = NFSRVFH_DATA(&nnsfh); 3142 while (xfer > 0) { 3143 nfsm_clget; 3144 if ((bp + xfer) > be) 3145 tsiz = be - bp; 3146 else 3147 tsiz = xfer; 3148 memcpy(bp, cp, tsiz); 3149 bp += tsiz; 3150 xfer -= tsiz; 3151 if (xfer > 0) 3152 cp += tsiz; 3153 } 3154 } 3155 invalid: 3156 cpos += dp->d_reclen; 3157 dp = (struct dirent *)cpos; 3158 cookiep++; 3159 ncookies--; 3160 } 3161 vrele(vp); 3162 nfsm_clget; 3163 *tl = nfs_false; 3164 bp += NFSX_UNSIGNED; 3165 nfsm_clget; 3166 if (eofflag) 3167 *tl = nfs_true; 3168 else 3169 *tl = nfs_false; 3170 bp += NFSX_UNSIGNED; 3171 if (mp != mb) { 3172 if (bp < be) 3173 mp->m_len = bp - mtod(mp, char *); 3174 } else 3175 mp->m_len += bp - bpos; 3176 free((void *)cookies, M_TEMP); 3177 free((void *)rbuf, M_TEMP); 3178 nfsm_srvdone; 3179 } 3180 3181 /* 3182 * nfs commit service 3183 */ 3184 int 3185 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3186 { 3187 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3188 struct mbuf *nam = nfsd->nd_nam; 3189 char *dpos = nfsd->nd_dpos; 3190 kauth_cred_t cred = nfsd->nd_cr; 3191 struct vattr bfor, aft; 3192 struct vnode *vp; 3193 nfsrvfh_t nsfh; 3194 u_int32_t *tl; 3195 int32_t t1; 3196 char *bpos; 3197 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3198 uint32_t cnt; 3199 char *cp2; 3200 struct mbuf *mb, *mreq; 3201 u_quad_t frev, off, end; 3202 3203 nfsm_srvmtofh(&nsfh); 3204 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3205 3206 off = fxdr_hyper(tl); 3207 tl += 2; 3208 cnt = fxdr_unsigned(uint32_t, *tl); 3209 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3210 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3211 if (error) { 3212 nfsm_reply(2 * NFSX_UNSIGNED); 3213 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3214 return (0); 3215 } 3216 for_ret = VOP_GETATTR(vp, &bfor, cred); 3217 end = (cnt > 0) ? off + cnt : vp->v_size; 3218 if (end < off || end > vp->v_size) 3219 end = vp->v_size; 3220 if (off < vp->v_size) 3221 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end); 3222 /* else error == 0, from nfsrv_fhtovp() */ 3223 aft_ret = VOP_GETATTR(vp, &aft, cred); 3224 vput(vp); 3225 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3226 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3227 if (!error) { 3228 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3229 *tl++ = txdr_unsigned(boottime.tv_sec); 3230 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 3231 } else { 3232 return (0); 3233 } 3234 nfsm_srvdone; 3235 } 3236 3237 /* 3238 * nfs statfs service 3239 */ 3240 int 3241 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3242 { 3243 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3244 struct mbuf *nam = nfsd->nd_nam; 3245 char *dpos = nfsd->nd_dpos; 3246 kauth_cred_t cred = nfsd->nd_cr; 3247 struct statvfs *sf = NULL; 3248 struct nfs_statfs *sfp; 3249 u_int32_t *tl; 3250 int32_t t1; 3251 char *bpos; 3252 int error = 0, rdonly, cache = 0, getret = 1; 3253 int v3 = (nfsd->nd_flag & ND_NFSV3); 3254 char *cp2; 3255 struct mbuf *mb, *mreq; 3256 struct vnode *vp; 3257 struct vattr at; 3258 nfsrvfh_t nsfh; 3259 u_quad_t frev, tval; 3260 3261 nfsm_srvmtofh(&nsfh); 3262 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3263 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3264 if (error) { 3265 nfsm_reply(NFSX_UNSIGNED); 3266 nfsm_srvpostop_attr(getret, &at); 3267 return (0); 3268 } 3269 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK); 3270 error = VFS_STATVFS(vp->v_mount, sf); 3271 getret = VOP_GETATTR(vp, &at, cred); 3272 vput(vp); 3273 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3274 if (v3) 3275 nfsm_srvpostop_attr(getret, &at); 3276 if (error) { 3277 free(sf, M_TEMP); 3278 return (0); 3279 } 3280 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3281 if (v3) { 3282 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3283 txdr_hyper(tval, &sfp->sf_tbytes); 3284 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3285 txdr_hyper(tval, &sfp->sf_fbytes); 3286 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3287 txdr_hyper(tval, &sfp->sf_abytes); 3288 tval = (u_quad_t)sf->f_files; 3289 txdr_hyper(tval, &sfp->sf_tfiles); 3290 tval = (u_quad_t)sf->f_ffree; 3291 txdr_hyper(tval, &sfp->sf_ffiles); 3292 txdr_hyper(tval, &sfp->sf_afiles); 3293 sfp->sf_invarsec = 0; 3294 } else { 3295 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3296 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3297 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3298 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3299 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3300 } 3301 nfsmout: 3302 if (sf) 3303 free(sf, M_TEMP); 3304 return error; 3305 } 3306 3307 /* 3308 * nfs fsinfo service 3309 */ 3310 int 3311 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3312 { 3313 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3314 struct mbuf *nam = nfsd->nd_nam; 3315 char *dpos = nfsd->nd_dpos; 3316 kauth_cred_t cred = nfsd->nd_cr; 3317 u_int32_t *tl; 3318 struct nfsv3_fsinfo *sip; 3319 int32_t t1; 3320 char *bpos; 3321 int error = 0, rdonly, cache = 0, getret = 1; 3322 uint32_t maxdata; 3323 char *cp2; 3324 struct mbuf *mb, *mreq; 3325 struct vnode *vp; 3326 struct vattr at; 3327 nfsrvfh_t nsfh; 3328 u_quad_t frev, maxfsize; 3329 struct statvfs *sb; 3330 3331 nfsm_srvmtofh(&nsfh); 3332 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3333 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3334 if (error) { 3335 nfsm_reply(NFSX_UNSIGNED); 3336 nfsm_srvpostop_attr(getret, &at); 3337 return (0); 3338 } 3339 3340 /* XXX Try to make a guess on the max file size. */ 3341 sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK); 3342 VFS_STATVFS(vp->v_mount, sb); 3343 maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1; 3344 free(sb, M_TEMP); 3345 3346 getret = VOP_GETATTR(vp, &at, cred); 3347 vput(vp); 3348 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3349 nfsm_srvpostop_attr(getret, &at); 3350 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3351 3352 /* 3353 * XXX 3354 * There should be file system VFS OP(s) to get this information. 3355 * For now, assume ufs. 3356 */ 3357 if (slp->ns_so->so_type == SOCK_DGRAM) 3358 maxdata = NFS_MAXDGRAMDATA; 3359 else 3360 maxdata = NFS_MAXDATA; 3361 sip->fs_rtmax = txdr_unsigned(maxdata); 3362 sip->fs_rtpref = txdr_unsigned(maxdata); 3363 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3364 sip->fs_wtmax = txdr_unsigned(maxdata); 3365 sip->fs_wtpref = txdr_unsigned(maxdata); 3366 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3367 sip->fs_dtpref = txdr_unsigned(maxdata); 3368 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3369 sip->fs_timedelta.nfsv3_sec = 0; 3370 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3371 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3372 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3373 NFSV3FSINFO_CANSETTIME); 3374 nfsm_srvdone; 3375 } 3376 3377 /* 3378 * nfs pathconf service 3379 */ 3380 int 3381 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3382 { 3383 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3384 struct mbuf *nam = nfsd->nd_nam; 3385 char *dpos = nfsd->nd_dpos; 3386 kauth_cred_t cred = nfsd->nd_cr; 3387 u_int32_t *tl; 3388 struct nfsv3_pathconf *pc; 3389 int32_t t1; 3390 char *bpos; 3391 int error = 0, rdonly, cache = 0, getret = 1; 3392 register_t linkmax, namemax, chownres, notrunc; 3393 char *cp2; 3394 struct mbuf *mb, *mreq; 3395 struct vnode *vp; 3396 struct vattr at; 3397 nfsrvfh_t nsfh; 3398 u_quad_t frev; 3399 3400 nfsm_srvmtofh(&nsfh); 3401 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3402 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3403 if (error) { 3404 nfsm_reply(NFSX_UNSIGNED); 3405 nfsm_srvpostop_attr(getret, &at); 3406 return (0); 3407 } 3408 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3409 if (!error) 3410 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3411 if (!error) 3412 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3413 if (!error) 3414 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3415 getret = VOP_GETATTR(vp, &at, cred); 3416 vput(vp); 3417 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3418 nfsm_srvpostop_attr(getret, &at); 3419 if (error) 3420 return (0); 3421 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3422 3423 pc->pc_linkmax = txdr_unsigned(linkmax); 3424 pc->pc_namemax = txdr_unsigned(namemax); 3425 pc->pc_notrunc = txdr_unsigned(notrunc); 3426 pc->pc_chownrestricted = txdr_unsigned(chownres); 3427 3428 /* 3429 * These should probably be supported by VOP_PATHCONF(), but 3430 * until msdosfs is exportable (why would you want to?), the 3431 * Unix defaults should be ok. 3432 */ 3433 pc->pc_caseinsensitive = nfs_false; 3434 pc->pc_casepreserving = nfs_true; 3435 nfsm_srvdone; 3436 } 3437 3438 /* 3439 * Null operation, used by clients to ping server 3440 */ 3441 /* ARGSUSED */ 3442 int 3443 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3444 struct lwp *lwp, struct mbuf **mrq) 3445 { 3446 struct mbuf *mrep = nfsd->nd_mrep; 3447 char *bpos; 3448 int error = NFSERR_RETVOID, cache = 0; 3449 struct mbuf *mb, *mreq; 3450 u_quad_t frev; 3451 3452 nfsm_reply(0); 3453 nfsmout: 3454 return (0); 3455 } 3456 3457 /* 3458 * No operation, used for obsolete procedures 3459 */ 3460 /* ARGSUSED */ 3461 int 3462 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3463 struct lwp *lwp, struct mbuf **mrq) 3464 { 3465 struct mbuf *mrep = nfsd->nd_mrep; 3466 char *bpos; 3467 int error, cache = 0; 3468 struct mbuf *mb, *mreq; 3469 u_quad_t frev; 3470 3471 if (nfsd->nd_repstat) 3472 error = nfsd->nd_repstat; 3473 else 3474 error = EPROCUNAVAIL; 3475 nfsm_reply(0); 3476 nfsmout: 3477 return (0); 3478 } 3479 3480 /* 3481 * Perform access checking for vnodes obtained from file handles that would 3482 * refer to files already opened by a Unix client. You cannot just use 3483 * vn_writechk() and VOP_ACCESS() for two reasons. 3484 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3485 * 2 - The owner is to be given access irrespective of mode bits for some 3486 * operations, so that processes that chmod after opening a file don't 3487 * break. I don't like this because it opens a security hole, but since 3488 * the nfs server opens a security hole the size of a barn door anyhow, 3489 * what the heck. 3490 * 3491 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3492 * will return EPERM instead of EACCES. EPERM is always an error. 3493 */ 3494 int 3495 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override) 3496 { 3497 struct vattr vattr; 3498 int error; 3499 if (flags & VWRITE) { 3500 /* Just vn_writechk() changed to check rdonly */ 3501 /* 3502 * Disallow write attempts on read-only file systems; 3503 * unless the file is a socket or a block or character 3504 * device resident on the file system. 3505 */ 3506 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3507 switch (vp->v_type) { 3508 case VREG: 3509 case VDIR: 3510 case VLNK: 3511 return (EROFS); 3512 default: 3513 break; 3514 } 3515 } 3516 3517 /* 3518 * If the vnode is in use as a process's text, 3519 * we can't allow writing. 3520 */ 3521 if (vp->v_iflag & VI_TEXT) 3522 return (ETXTBSY); 3523 } 3524 error = VOP_GETATTR(vp, &vattr, cred); 3525 if (error) 3526 return (error); 3527 error = VOP_ACCESS(vp, flags, cred); 3528 /* 3529 * Allow certain operations for the owner (reads and writes 3530 * on files that are already open). 3531 */ 3532 if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid) 3533 error = 0; 3534 return error; 3535 } 3536
Indexes created Thu Oct 02 14:10:14 GMT 2025