1 /* $NetBSD: nfs_serv.c,v 1.143 2009/03/14 15:36:24 dsl 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 1.143 2009/03/14 15:36:24 dsl 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 krwlock_t netexport_lock; /* XXX */ 112 extern struct vfs_hooks nfs_export_hooks; /* XXX */ 113 int error; 114 115 switch (cmd) { 116 case MODULE_CMD_INIT: 117 error = syscall_establish(NULL, nfsserver_syscalls); 118 if (error != 0) { 119 return error; 120 } 121 nfs_init(); /* XXX for monolithic kernel */ 122 rw_init(&netexport_lock); 123 nfsrv_init(0); /* Init server data structures */ 124 nfsrv_initcache(); /* Init the server request cache */ 125 vfs_hooks_attach(&nfs_export_hooks); 126 nfs_timer_srvinit(nfsrv_timer); 127 return 0; 128 case MODULE_CMD_FINI: 129 error = syscall_disestablish(NULL, nfsserver_syscalls); 130 if (error != 0) { 131 return error; 132 } 133 nfs_timer_srvfini(); 134 vfs_hooks_detach(&nfs_export_hooks); 135 nfsrv_finicache(); 136 nfsrv_fini(); 137 rw_destroy(&netexport_lock); 138 return 0; 139 default: 140 return ENOTTY; 141 } 142 } 143 144 /* 145 * nfs v3 access service 146 */ 147 int 148 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 149 { 150 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 151 struct mbuf *nam = nfsd->nd_nam; 152 char *dpos = nfsd->nd_dpos; 153 kauth_cred_t cred = nfsd->nd_cr; 154 struct vnode *vp; 155 nfsrvfh_t nsfh; 156 u_int32_t *tl; 157 int32_t t1; 158 char *bpos; 159 int error = 0, rdonly, cache = 0, getret; 160 char *cp2; 161 struct mbuf *mb, *mreq; 162 struct vattr va; 163 u_long inmode, testmode, outmode; 164 u_quad_t frev; 165 166 nfsm_srvmtofh(&nsfh); 167 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 168 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 169 (nfsd->nd_flag & ND_KERBAUTH), false); 170 if (error) { 171 nfsm_reply(NFSX_UNSIGNED); 172 nfsm_srvpostop_attr(1, (struct vattr *)0); 173 return (0); 174 } 175 inmode = fxdr_unsigned(u_int32_t, *tl); 176 outmode = 0; 177 if ((inmode & NFSV3ACCESS_READ) && 178 nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0) 179 outmode |= NFSV3ACCESS_READ; 180 if (vp->v_type != VDIR) { 181 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 182 if (testmode && 183 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 184 outmode |= testmode; 185 if ((inmode & NFSV3ACCESS_EXECUTE) && 186 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 187 outmode |= NFSV3ACCESS_EXECUTE; 188 } else { 189 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 190 NFSV3ACCESS_DELETE); 191 if (testmode && 192 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 193 outmode |= testmode; 194 if ((inmode & NFSV3ACCESS_LOOKUP) && 195 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 196 outmode |= NFSV3ACCESS_LOOKUP; 197 } 198 getret = VOP_GETATTR(vp, &va, cred); 199 vput(vp); 200 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 201 nfsm_srvpostop_attr(getret, &va); 202 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 203 *tl = txdr_unsigned(outmode); 204 nfsm_srvdone; 205 } 206 207 /* 208 * nfs getattr service 209 */ 210 int 211 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 212 { 213 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 214 struct mbuf *nam = nfsd->nd_nam; 215 char *dpos = nfsd->nd_dpos; 216 kauth_cred_t cred = nfsd->nd_cr; 217 struct nfs_fattr *fp; 218 struct vattr va; 219 struct vnode *vp; 220 nfsrvfh_t nsfh; 221 u_int32_t *tl; 222 int32_t t1; 223 char *bpos; 224 int error = 0, rdonly, cache = 0; 225 char *cp2; 226 struct mbuf *mb, *mreq; 227 u_quad_t frev; 228 229 nfsm_srvmtofh(&nsfh); 230 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 231 (nfsd->nd_flag & ND_KERBAUTH), false); 232 if (error) { 233 nfsm_reply(0); 234 return (0); 235 } 236 nqsrv_getl(vp, ND_READ); 237 error = VOP_GETATTR(vp, &va, cred); 238 vput(vp); 239 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 240 if (error) 241 return (0); 242 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 243 nfsm_srvfillattr(&va, fp); 244 nfsm_srvdone; 245 } 246 247 /* 248 * nfs setattr service 249 */ 250 int 251 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 252 { 253 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 254 struct mbuf *nam = nfsd->nd_nam; 255 char *dpos = nfsd->nd_dpos; 256 kauth_cred_t cred = nfsd->nd_cr; 257 struct vattr va, preat; 258 struct nfsv2_sattr *sp; 259 struct nfs_fattr *fp; 260 struct vnode *vp; 261 nfsrvfh_t nsfh; 262 u_int32_t *tl; 263 int32_t t1; 264 char *bpos; 265 int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1; 266 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0; 267 char *cp2; 268 struct mbuf *mb, *mreq; 269 u_quad_t frev; 270 struct timespec guard; 271 272 memset(&guard, 0, sizeof guard); /* XXX gcc */ 273 274 nfsm_srvmtofh(&nsfh); 275 VATTR_NULL(&va); 276 if (v3) { 277 nfsm_srvsattr(&va); 278 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 279 gcheck = fxdr_unsigned(int, *tl); 280 if (gcheck) { 281 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 282 fxdr_nfsv3time(tl, &guard); 283 } 284 } else { 285 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 286 /* 287 * Nah nah nah nah na nah 288 * There is a bug in the Sun client that puts 0xffff in the mode 289 * field of sattr when it should put in 0xffffffff. The u_short 290 * doesn't sign extend. 291 * --> check the low order 2 bytes for 0xffff 292 */ 293 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 294 va.va_mode = nfstov_mode(sp->sa_mode); 295 if (sp->sa_uid != nfs_xdrneg1) 296 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 297 if (sp->sa_gid != nfs_xdrneg1) 298 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 299 if (sp->sa_size != nfs_xdrneg1) 300 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 301 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 302 #ifdef notyet 303 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 304 #else 305 va.va_atime.tv_sec = 306 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 307 va.va_atime.tv_nsec = 0; 308 #endif 309 } 310 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 311 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 312 313 } 314 315 /* 316 * Now that we have all the fields, lets do it. 317 */ 318 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 319 (nfsd->nd_flag & ND_KERBAUTH), false); 320 if (error) { 321 nfsm_reply(2 * NFSX_UNSIGNED); 322 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 323 return (0); 324 } 325 nqsrv_getl(vp, ND_WRITE); 326 if (v3) { 327 error = preat_ret = VOP_GETATTR(vp, &preat, cred); 328 if (!error && gcheck && 329 (preat.va_ctime.tv_sec != guard.tv_sec || 330 preat.va_ctime.tv_nsec != guard.tv_nsec)) 331 error = NFSERR_NOT_SYNC; 332 if (error) { 333 vput(vp); 334 nfsm_reply(NFSX_WCCDATA(v3)); 335 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 336 return (0); 337 } 338 } 339 340 /* 341 * If the size is being changed write acces is required, otherwise 342 * just check for a read only file system. 343 */ 344 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 345 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 346 error = EROFS; 347 goto out; 348 } 349 } else { 350 if (vp->v_type == VDIR) { 351 error = EISDIR; 352 goto out; 353 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 354 lwp, 0)) != 0) 355 goto out; 356 } 357 error = VOP_SETATTR(vp, &va, cred); 358 postat_ret = VOP_GETATTR(vp, &va, cred); 359 if (!error) 360 error = postat_ret; 361 out: 362 vput(vp); 363 nfsm_reply(NFSX_WCCORFATTR(v3)); 364 if (v3) { 365 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 366 return (0); 367 } else { 368 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 369 nfsm_srvfillattr(&va, fp); 370 } 371 nfsm_srvdone; 372 } 373 374 /* 375 * nfs lookup rpc 376 */ 377 int 378 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 379 { 380 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 381 struct mbuf *nam = nfsd->nd_nam; 382 char *dpos = nfsd->nd_dpos; 383 kauth_cred_t cred = nfsd->nd_cr; 384 struct nfs_fattr *fp; 385 struct nameidata nd, ind, *ndp = &nd; 386 struct vnode *vp, *dirp; 387 nfsrvfh_t nsfh; 388 char *cp; 389 u_int32_t *tl; 390 int32_t t1; 391 char *bpos; 392 int error = 0, cache = 0, dirattr_ret = 1; 393 uint32_t len; 394 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 395 char *cp2; 396 struct mbuf *mb, *mreq; 397 struct vattr va, dirattr; 398 u_quad_t frev; 399 400 nfsm_srvmtofh(&nsfh); 401 nfsm_srvnamesiz(len); 402 403 pubflag = nfs_ispublicfh(&nsfh); 404 405 nd.ni_cnd.cn_cred = cred; 406 nd.ni_cnd.cn_nameiop = LOOKUP; 407 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART; 408 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 409 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 410 411 if (!error && pubflag) { 412 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) { 413 /* 414 * Setup call to lookup() to see if we can find 415 * the index file. Arguably, this doesn't belong 416 * in a kernel.. Ugh. 417 */ 418 ind = nd; 419 VOP_UNLOCK(nd.ni_vp, 0); 420 ind.ni_pathlen = strlen(nfs_pub.np_index); 421 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf = 422 nfs_pub.np_index; 423 ind.ni_startdir = nd.ni_vp; 424 VREF(ind.ni_startdir); 425 error = lookup(&ind); 426 if (!error) { 427 /* 428 * Found an index file. Get rid of 429 * the old references. 430 */ 431 if (dirp) 432 vrele(dirp); 433 dirp = nd.ni_vp; 434 vrele(nd.ni_startdir); 435 ndp = &ind; 436 } else 437 error = 0; 438 } 439 /* 440 * If the public filehandle was used, check that this lookup 441 * didn't result in a filehandle outside the publicly exported 442 * filesystem. 443 */ 444 445 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 446 vput(nd.ni_vp); 447 error = EPERM; 448 } 449 } 450 451 if (dirp) { 452 if (v3) 453 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred); 454 vrele(dirp); 455 } 456 457 if (error) { 458 nfsm_reply(NFSX_POSTOPATTR(v3)); 459 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 460 return (0); 461 } 462 463 nqsrv_getl(ndp->ni_startdir, ND_READ); 464 PNBUF_PUT(nd.ni_cnd.cn_pnbuf); 465 vp = ndp->ni_vp; 466 error = nfsrv_composefh(vp, &nsfh, v3); 467 if (!error) 468 error = VOP_GETATTR(vp, &va, cred); 469 vput(vp); 470 vrele(ndp->ni_startdir); 471 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) + 472 NFSX_POSTOPATTR(v3)); 473 if (error) { 474 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 475 return (0); 476 } 477 nfsm_srvfhtom(&nsfh, v3); 478 if (v3) { 479 nfsm_srvpostop_attr(0, &va); 480 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 481 } else { 482 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 483 nfsm_srvfillattr(&va, fp); 484 } 485 nfsm_srvdone; 486 } 487 488 /* 489 * nfs readlink service 490 */ 491 int 492 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 493 { 494 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 495 struct mbuf *nam = nfsd->nd_nam; 496 char *dpos = nfsd->nd_dpos; 497 kauth_cred_t cred = nfsd->nd_cr; 498 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 499 struct iovec *ivp = iv; 500 struct mbuf *mp; 501 u_int32_t *tl; 502 int32_t t1; 503 char *bpos; 504 int error = 0, rdonly, cache = 0, i, padlen, getret; 505 uint32_t len; 506 int v3 = (nfsd->nd_flag & ND_NFSV3); 507 char *cp2; 508 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 509 struct vnode *vp; 510 struct vattr attr; 511 nfsrvfh_t nsfh; 512 struct uio io, *uiop = &io; 513 u_quad_t frev; 514 515 nfsm_srvmtofh(&nsfh); 516 len = 0; 517 i = 0; 518 while (len < NFS_MAXPATHLEN) { 519 mp = m_get(M_WAIT, MT_DATA); 520 MCLAIM(mp, &nfs_mowner); 521 m_clget(mp, M_WAIT); 522 mp->m_len = NFSMSIZ(mp); 523 if (len == 0) 524 mp3 = mp2 = mp; 525 else { 526 mp2->m_next = mp; 527 mp2 = mp; 528 } 529 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 530 mp->m_len = NFS_MAXPATHLEN-len; 531 len = NFS_MAXPATHLEN; 532 } else 533 len += mp->m_len; 534 ivp->iov_base = mtod(mp, void *); 535 ivp->iov_len = mp->m_len; 536 i++; 537 ivp++; 538 } 539 uiop->uio_iov = iv; 540 uiop->uio_iovcnt = i; 541 uiop->uio_offset = 0; 542 uiop->uio_resid = len; 543 uiop->uio_rw = UIO_READ; 544 UIO_SETUP_SYSSPACE(uiop); 545 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 546 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 547 if (error) { 548 m_freem(mp3); 549 nfsm_reply(2 * NFSX_UNSIGNED); 550 nfsm_srvpostop_attr(1, (struct vattr *)0); 551 return (0); 552 } 553 if (vp->v_type != VLNK) { 554 if (v3) 555 error = EINVAL; 556 else 557 error = ENXIO; 558 goto out; 559 } 560 nqsrv_getl(vp, ND_READ); 561 error = VOP_READLINK(vp, uiop, cred); 562 out: 563 getret = VOP_GETATTR(vp, &attr, cred); 564 vput(vp); 565 if (error) 566 m_freem(mp3); 567 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 568 if (v3) { 569 nfsm_srvpostop_attr(getret, &attr); 570 if (error) 571 return (0); 572 } 573 len -= uiop->uio_resid; 574 padlen = nfsm_padlen(len); 575 if (uiop->uio_resid || padlen) 576 nfs_zeropad(mp3, uiop->uio_resid, padlen); 577 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 578 *tl = txdr_unsigned(len); 579 mb->m_next = mp3; 580 nfsm_srvdone; 581 } 582 583 /* 584 * nfs read service 585 */ 586 int 587 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 588 { 589 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 590 struct mbuf *nam = nfsd->nd_nam; 591 char *dpos = nfsd->nd_dpos; 592 kauth_cred_t cred = nfsd->nd_cr; 593 struct mbuf *m; 594 struct nfs_fattr *fp; 595 u_int32_t *tl; 596 int32_t t1; 597 int i; 598 char *bpos; 599 int error = 0, rdonly, cache = 0, getret; 600 int v3 = (nfsd->nd_flag & ND_NFSV3); 601 uint32_t reqlen, len, cnt, left; 602 int padlen; 603 char *cp2; 604 struct mbuf *mb, *mreq; 605 struct vnode *vp; 606 nfsrvfh_t nsfh; 607 struct uio io, *uiop = &io; 608 struct vattr va; 609 off_t off; 610 u_quad_t frev; 611 612 nfsm_srvmtofh(&nsfh); 613 if (v3) { 614 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 615 off = fxdr_hyper(tl); 616 } else { 617 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 618 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 619 } 620 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 621 reqlen = fxdr_unsigned(uint32_t, *tl); 622 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 623 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 624 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 625 if (error) { 626 nfsm_reply(2 * NFSX_UNSIGNED); 627 nfsm_srvpostop_attr(1, (struct vattr *)0); 628 return (0); 629 } 630 if (vp->v_type != VREG) { 631 if (v3) 632 error = EINVAL; 633 else 634 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 635 } 636 if (!error) { 637 nqsrv_getl(vp, ND_READ); 638 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 639 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 640 } 641 getret = VOP_GETATTR(vp, &va, cred); 642 if (!error) 643 error = getret; 644 if (error) { 645 vput(vp); 646 nfsm_reply(NFSX_POSTOPATTR(v3)); 647 nfsm_srvpostop_attr(getret, &va); 648 return (0); 649 } 650 if (off >= va.va_size) 651 cnt = 0; 652 else if ((off + reqlen) > va.va_size) 653 cnt = va.va_size - off; 654 else 655 cnt = reqlen; 656 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 657 if (v3) { 658 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 659 *tl++ = nfs_true; 660 fp = (struct nfs_fattr *)tl; 661 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 662 } else { 663 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 664 fp = (struct nfs_fattr *)tl; 665 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 666 } 667 len = left = cnt; 668 if (cnt > 0) { 669 if (nfsd_use_loan) { 670 struct vm_page **pgpp; 671 voff_t pgoff = trunc_page(off); 672 int npages; 673 vaddr_t lva; 674 675 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 676 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 677 678 /* allocate kva for mbuf data */ 679 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 680 if (lva == 0) { 681 /* fall back to VOP_READ */ 682 goto loan_fail; 683 } 684 685 /* allocate mbuf */ 686 m = m_get(M_WAIT, MT_DATA); 687 MCLAIM(m, &nfs_mowner); 688 pgpp = m->m_ext.ext_pgs; 689 690 /* loan pages */ 691 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 692 pgpp); 693 if (error) { 694 sokvafree(lva, npages << PAGE_SHIFT); 695 m_free(m); 696 if (error == EBUSY) 697 goto loan_fail; 698 goto read_error; 699 } 700 701 /* associate kva to mbuf */ 702 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 703 cnt, M_MBUF, soloanfree, slp->ns_so); 704 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 705 m->m_len = cnt; 706 707 /* map pages */ 708 for (i = 0; i < npages; i++) { 709 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 710 VM_PROT_READ); 711 lva += PAGE_SIZE; 712 } 713 714 pmap_update(pmap_kernel()); 715 716 mb->m_next = m; 717 mb = m; 718 error = 0; 719 uiop->uio_resid = 0; 720 } else { 721 struct iovec *iv; 722 struct iovec *iv2; 723 struct mbuf *m2; 724 int siz; 725 loan_fail: 726 /* 727 * Generate the mbuf list with the uio_iov ref. to it. 728 */ 729 i = 0; 730 m = m2 = mb; 731 while (left > 0) { 732 siz = min(M_TRAILINGSPACE(m), left); 733 if (siz > 0) { 734 left -= siz; 735 i++; 736 } 737 if (left > 0) { 738 m = m_get(M_WAIT, MT_DATA); 739 MCLAIM(m, &nfs_mowner); 740 m_clget(m, M_WAIT); 741 m->m_len = 0; 742 m2->m_next = m; 743 m2 = m; 744 } 745 } 746 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 747 uiop->uio_iov = iv2 = iv; 748 m = mb; 749 left = cnt; 750 i = 0; 751 while (left > 0) { 752 if (m == NULL) 753 panic("nfsrv_read iov"); 754 siz = min(M_TRAILINGSPACE(m), left); 755 if (siz > 0) { 756 iv->iov_base = mtod(m, char *) + 757 m->m_len; 758 iv->iov_len = siz; 759 m->m_len += siz; 760 left -= siz; 761 iv++; 762 i++; 763 } 764 m = m->m_next; 765 } 766 uiop->uio_iovcnt = i; 767 uiop->uio_offset = off; 768 uiop->uio_resid = cnt; 769 uiop->uio_rw = UIO_READ; 770 UIO_SETUP_SYSSPACE(uiop); 771 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 772 free((void *)iv2, M_TEMP); 773 } 774 read_error: 775 if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){ 776 if (!error) 777 error = getret; 778 m_freem(mreq); 779 vput(vp); 780 nfsm_reply(NFSX_POSTOPATTR(v3)); 781 nfsm_srvpostop_attr(getret, &va); 782 return (0); 783 } 784 } else { 785 uiop->uio_resid = 0; 786 } 787 vput(vp); 788 nfsm_srvfillattr(&va, fp); 789 len -= uiop->uio_resid; 790 padlen = nfsm_padlen(len); 791 if (uiop->uio_resid || padlen) 792 nfs_zeropad(mb, uiop->uio_resid, padlen); 793 if (v3) { 794 /* count */ 795 *tl++ = txdr_unsigned(len); 796 /* eof */ 797 if (off + len >= va.va_size) 798 *tl++ = nfs_true; 799 else 800 *tl++ = nfs_false; 801 } 802 *tl = txdr_unsigned(len); 803 nfsm_srvdone; 804 } 805 806 /* 807 * nfs write service 808 */ 809 int 810 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 811 { 812 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 813 struct mbuf *nam = nfsd->nd_nam; 814 char *dpos = nfsd->nd_dpos; 815 kauth_cred_t cred = nfsd->nd_cr; 816 struct iovec *ivp; 817 int i, cnt; 818 struct mbuf *mp; 819 struct nfs_fattr *fp; 820 struct iovec *iv; 821 struct vattr va, forat; 822 u_int32_t *tl; 823 int32_t t1; 824 char *bpos; 825 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 826 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 827 int stable = NFSV3WRITE_FILESYNC; 828 int v3 = (nfsd->nd_flag & ND_NFSV3); 829 char *cp2; 830 struct mbuf *mb, *mreq; 831 struct vnode *vp; 832 nfsrvfh_t nsfh; 833 struct uio io, *uiop = &io; 834 off_t off; 835 u_quad_t frev; 836 837 if (mrep == NULL) { 838 *mrq = NULL; 839 return (0); 840 } 841 nfsm_srvmtofh(&nsfh); 842 if (v3) { 843 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 844 off = fxdr_hyper(tl); 845 tl += 3; 846 stable = fxdr_unsigned(int, *tl++); 847 } else { 848 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 849 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 850 tl += 2; 851 } 852 retlen = len = fxdr_unsigned(int32_t, *tl); 853 cnt = i = 0; 854 855 /* 856 * For NFS Version 2, it is not obvious what a write of zero length 857 * should do, but I might as well be consistent with Version 3, 858 * which is to return ok so long as there are no permission problems. 859 */ 860 if (len > 0) { 861 zeroing = 1; 862 mp = mrep; 863 while (mp) { 864 if (mp == md) { 865 zeroing = 0; 866 adjust = dpos - mtod(mp, char *); 867 mp->m_len -= adjust; 868 if (mp->m_len > 0 && adjust > 0) 869 NFSMADV(mp, adjust); 870 } 871 if (zeroing) 872 mp->m_len = 0; 873 else if (mp->m_len > 0) { 874 i += mp->m_len; 875 if (i > len) { 876 mp->m_len -= (i - len); 877 zeroing = 1; 878 } 879 if (mp->m_len > 0) 880 cnt++; 881 } 882 mp = mp->m_next; 883 } 884 } 885 if (len > NFS_MAXDATA || len < 0 || i < len) { 886 error = EIO; 887 nfsm_reply(2 * NFSX_UNSIGNED); 888 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 889 return (0); 890 } 891 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 892 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 893 if (error) { 894 nfsm_reply(2 * NFSX_UNSIGNED); 895 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 896 return (0); 897 } 898 if (v3) 899 forat_ret = VOP_GETATTR(vp, &forat, cred); 900 if (vp->v_type != VREG) { 901 if (v3) 902 error = EINVAL; 903 else 904 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 905 } 906 if (!error) { 907 nqsrv_getl(vp, ND_WRITE); 908 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 909 } 910 if (error) { 911 vput(vp); 912 nfsm_reply(NFSX_WCCDATA(v3)); 913 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 914 return (0); 915 } 916 917 if (len > 0) { 918 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 919 uiop->uio_iov = iv = ivp; 920 uiop->uio_iovcnt = cnt; 921 mp = mrep; 922 while (mp) { 923 if (mp->m_len > 0) { 924 ivp->iov_base = mtod(mp, void *); 925 ivp->iov_len = mp->m_len; 926 ivp++; 927 } 928 mp = mp->m_next; 929 } 930 931 /* 932 * XXX 933 * The IO_METASYNC flag indicates that all metadata (and not 934 * just enough to ensure data integrity) must be written to 935 * stable storage synchronously. 936 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 937 */ 938 if (stable == NFSV3WRITE_UNSTABLE) 939 ioflags = IO_NODELOCKED; 940 else if (stable == NFSV3WRITE_DATASYNC) 941 ioflags = (IO_SYNC | IO_NODELOCKED); 942 else 943 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 944 uiop->uio_resid = len; 945 uiop->uio_rw = UIO_WRITE; 946 uiop->uio_offset = off; 947 UIO_SETUP_SYSSPACE(uiop); 948 error = VOP_WRITE(vp, uiop, ioflags, cred); 949 nfsstats.srvvop_writes++; 950 free(iv, M_TEMP); 951 } 952 aftat_ret = VOP_GETATTR(vp, &va, cred); 953 vput(vp); 954 if (!error) 955 error = aftat_ret; 956 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 957 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 958 if (v3) { 959 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 960 if (error) 961 return (0); 962 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 963 *tl++ = txdr_unsigned(retlen); 964 if (stable == NFSV3WRITE_UNSTABLE) 965 *tl++ = txdr_unsigned(stable); 966 else 967 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 968 /* 969 * Actually, there is no need to txdr these fields, 970 * but it may make the values more human readable, 971 * for debugging purposes. 972 */ 973 *tl++ = txdr_unsigned(boottime.tv_sec); 974 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 975 } else { 976 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 977 nfsm_srvfillattr(&va, fp); 978 } 979 nfsm_srvdone; 980 } 981 982 /* 983 * XXX elad: the original NFSW_SAMECRED() macro also made sure the 984 * two nd_flag fields of the descriptors contained 985 * ND_KERBAUTH. 986 */ 987 static int 988 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2) 989 { 990 int i, do_ngroups; 991 992 if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2)) 993 return (0); 994 if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2)) 995 return (0); 996 do_ngroups = kauth_cred_ngroups(cred1); 997 for (i = 0; i < do_ngroups; i++) 998 if (kauth_cred_group(cred1, i) != 999 kauth_cred_group(cred2, i)) 1000 return (0); 1001 1002 return (1); 1003 } 1004 1005 static struct nfsrvw_delayhash * 1006 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh) 1007 { 1008 uint32_t hash; 1009 1010 hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh), 1011 HASH32_BUF_INIT); 1012 return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ]; 1013 } 1014 1015 /* 1016 * NFS write service with write gathering support. Called when 1017 * nfsrvw_procrastinate > 0. 1018 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 1019 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 1020 * Jan. 1994. 1021 */ 1022 int 1023 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1024 { 1025 struct timeval now; 1026 struct iovec *ivp; 1027 struct mbuf *mp; 1028 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1029 struct nfs_fattr *fp; 1030 int i = 0; 1031 struct iovec *iov; 1032 struct nfsrvw_delayhash *wpp; 1033 kauth_cred_t cred; 1034 struct vattr va, forat; 1035 u_int32_t *tl; 1036 int32_t t1; 1037 char *bpos, *dpos; 1038 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1039 int ioflags, aftat_ret = 1, adjust, v3, zeroing; 1040 char *cp2; 1041 struct mbuf *mb, *mreq, *mrep, *md; 1042 struct vnode *vp; 1043 struct uio io, *uiop = &io; 1044 u_quad_t frev, cur_usec; 1045 1046 *mrq = NULL; 1047 if (*ndp) { 1048 nfsd = *ndp; 1049 *ndp = NULL; 1050 mrep = nfsd->nd_mrep; 1051 md = nfsd->nd_md; 1052 dpos = nfsd->nd_dpos; 1053 cred = nfsd->nd_cr; 1054 v3 = (nfsd->nd_flag & ND_NFSV3); 1055 LIST_INIT(&nfsd->nd_coalesce); 1056 nfsd->nd_mreq = NULL; 1057 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1058 getmicrotime(&now); 1059 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1060 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1061 1062 /* 1063 * Now, get the write header.. 1064 */ 1065 nfsm_srvmtofh(&nfsd->nd_fh); 1066 if (v3) { 1067 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1068 nfsd->nd_off = fxdr_hyper(tl); 1069 tl += 3; 1070 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1071 } else { 1072 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1073 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1074 tl += 2; 1075 } 1076 len = fxdr_unsigned(int32_t, *tl); 1077 nfsd->nd_len = len; 1078 nfsd->nd_eoff = nfsd->nd_off + len; 1079 1080 /* 1081 * Trim the header out of the mbuf list and trim off any trailing 1082 * junk so that the mbuf list has only the write data. 1083 */ 1084 zeroing = 1; 1085 i = 0; 1086 mp = mrep; 1087 while (mp) { 1088 if (mp == md) { 1089 zeroing = 0; 1090 adjust = dpos - mtod(mp, char *); 1091 mp->m_len -= adjust; 1092 if (mp->m_len > 0 && adjust > 0) 1093 NFSMADV(mp, adjust); 1094 } 1095 if (zeroing) 1096 mp->m_len = 0; 1097 else { 1098 i += mp->m_len; 1099 if (i > len) { 1100 mp->m_len -= (i - len); 1101 zeroing = 1; 1102 } 1103 } 1104 mp = mp->m_next; 1105 } 1106 if (len > NFS_MAXDATA || len < 0 || i < len) { 1107 nfsmout: 1108 m_freem(mrep); 1109 error = EIO; 1110 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1111 if (v3) 1112 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1113 nfsd->nd_mreq = mreq; 1114 nfsd->nd_mrep = NULL; 1115 nfsd->nd_time = 0; 1116 } 1117 1118 /* 1119 * Add this entry to the hash and time queues. 1120 */ 1121 owp = NULL; 1122 mutex_enter(&nfsd_lock); 1123 wp = LIST_FIRST(&slp->ns_tq); 1124 while (wp && wp->nd_time < nfsd->nd_time) { 1125 owp = wp; 1126 wp = LIST_NEXT(wp, nd_tq); 1127 } 1128 if (owp) { 1129 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1130 } else { 1131 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1132 } 1133 if (nfsd->nd_mrep) { 1134 wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh); 1135 owp = NULL; 1136 wp = LIST_FIRST(wpp); 1137 while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1138 owp = wp; 1139 wp = LIST_NEXT(wp, nd_hash); 1140 } 1141 while (wp && wp->nd_off < nfsd->nd_off && 1142 !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1143 owp = wp; 1144 wp = LIST_NEXT(wp, nd_hash); 1145 } 1146 if (owp) { 1147 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1148 1149 /* 1150 * Search the hash list for overlapping entries and 1151 * coalesce. 1152 */ 1153 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1154 wp = LIST_NEXT(nfsd, nd_hash); 1155 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr)) 1156 nfsrvw_coalesce(owp, nfsd); 1157 } 1158 } else { 1159 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1160 } 1161 } 1162 mutex_exit(&nfsd_lock); 1163 } 1164 1165 /* 1166 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1167 * and generate the associated reply mbuf list(s). 1168 */ 1169 loop1: 1170 getmicrotime(&now); 1171 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1172 mutex_enter(&nfsd_lock); 1173 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1174 owp = LIST_NEXT(nfsd, nd_tq); 1175 if (nfsd->nd_time > cur_usec) 1176 break; 1177 if (nfsd->nd_mreq) 1178 continue; 1179 LIST_REMOVE(nfsd, nd_tq); 1180 LIST_REMOVE(nfsd, nd_hash); 1181 mutex_exit(&nfsd_lock); 1182 1183 mrep = nfsd->nd_mrep; 1184 nfsd->nd_mrep = NULL; 1185 cred = nfsd->nd_cr; 1186 v3 = (nfsd->nd_flag & ND_NFSV3); 1187 forat_ret = aftat_ret = 1; 1188 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1189 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1190 false); 1191 if (!error) { 1192 if (v3) 1193 forat_ret = VOP_GETATTR(vp, &forat, cred); 1194 if (vp->v_type != VREG) { 1195 if (v3) 1196 error = EINVAL; 1197 else 1198 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1199 } 1200 } else 1201 vp = NULL; 1202 if (!error) { 1203 nqsrv_getl(vp, ND_WRITE); 1204 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1205 } 1206 1207 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1208 ioflags = IO_NODELOCKED; 1209 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1210 ioflags = (IO_SYNC | IO_NODELOCKED); 1211 else 1212 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1213 uiop->uio_rw = UIO_WRITE; 1214 uiop->uio_offset = nfsd->nd_off; 1215 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1216 UIO_SETUP_SYSSPACE(uiop); 1217 if (uiop->uio_resid > 0) { 1218 mp = mrep; 1219 i = 0; 1220 while (mp) { 1221 if (mp->m_len > 0) 1222 i++; 1223 mp = mp->m_next; 1224 } 1225 uiop->uio_iovcnt = i; 1226 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1227 uiop->uio_iov = ivp = iov; 1228 mp = mrep; 1229 while (mp) { 1230 if (mp->m_len > 0) { 1231 ivp->iov_base = mtod(mp, void *); 1232 ivp->iov_len = mp->m_len; 1233 ivp++; 1234 } 1235 mp = mp->m_next; 1236 } 1237 if (!error) { 1238 error = VOP_WRITE(vp, uiop, ioflags, cred); 1239 nfsstats.srvvop_writes++; 1240 } 1241 free((void *)iov, M_TEMP); 1242 } 1243 m_freem(mrep); 1244 if (vp) { 1245 aftat_ret = VOP_GETATTR(vp, &va, cred); 1246 vput(vp); 1247 } 1248 1249 /* 1250 * Loop around generating replies for all write rpcs that have 1251 * now been completed. 1252 */ 1253 swp = nfsd; 1254 do { 1255 if (error) { 1256 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1257 if (v3) { 1258 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1259 } 1260 } else { 1261 nfsm_writereply(NFSX_PREOPATTR(v3) + 1262 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1263 NFSX_WRITEVERF(v3), v3); 1264 if (v3) { 1265 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1266 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1267 *tl++ = txdr_unsigned(nfsd->nd_len); 1268 *tl++ = txdr_unsigned(swp->nd_stable); 1269 /* 1270 * Actually, there is no need to txdr these fields, 1271 * but it may make the values more human readable, 1272 * for debugging purposes. 1273 */ 1274 *tl++ = txdr_unsigned(boottime.tv_sec); 1275 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 1276 } else { 1277 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1278 nfsm_srvfillattr(&va, fp); 1279 } 1280 } 1281 nfsd->nd_mreq = mreq; 1282 if (nfsd->nd_mrep) 1283 panic("nfsrv_write: nd_mrep not free"); 1284 1285 /* 1286 * Done. Put it at the head of the timer queue so that 1287 * the final phase can return the reply. 1288 */ 1289 mutex_enter(&nfsd_lock); 1290 if (nfsd != swp) { 1291 nfsd->nd_time = 0; 1292 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1293 } 1294 nfsd = LIST_FIRST(&swp->nd_coalesce); 1295 if (nfsd) { 1296 LIST_REMOVE(nfsd, nd_tq); 1297 } 1298 mutex_exit(&nfsd_lock); 1299 } while (nfsd); 1300 swp->nd_time = 0; 1301 1302 mutex_enter(&nfsd_lock); 1303 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1304 mutex_exit(&nfsd_lock); 1305 goto loop1; 1306 } 1307 mutex_exit(&nfsd_lock); 1308 nfs_timer_start(); 1309 1310 /* 1311 * Search for a reply to return. 1312 */ 1313 mutex_enter(&nfsd_lock); 1314 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1315 if (nfsd->nd_mreq) { 1316 LIST_REMOVE(nfsd, nd_tq); 1317 *mrq = nfsd->nd_mreq; 1318 *ndp = nfsd; 1319 break; 1320 } 1321 } 1322 mutex_exit(&nfsd_lock); 1323 return (0); 1324 } 1325 1326 /* 1327 * Coalesce the write request nfsd into owp. To do this we must: 1328 * - remove nfsd from the queues 1329 * - merge nfsd->nd_mrep into owp->nd_mrep 1330 * - update the nd_eoff and nd_stable for owp 1331 * - put nfsd on owp's nd_coalesce list 1332 * NB: Must be called at splsoftclock(). 1333 */ 1334 void 1335 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd) 1336 { 1337 int overlap; 1338 struct mbuf *mp; 1339 struct nfsrv_descript *m; 1340 1341 KASSERT(mutex_owned(&nfsd_lock)); 1342 1343 LIST_REMOVE(nfsd, nd_hash); 1344 LIST_REMOVE(nfsd, nd_tq); 1345 if (owp->nd_eoff < nfsd->nd_eoff) { 1346 overlap = owp->nd_eoff - nfsd->nd_off; 1347 if (overlap < 0) 1348 panic("nfsrv_coalesce: bad off"); 1349 if (overlap > 0) 1350 m_adj(nfsd->nd_mrep, overlap); 1351 mp = owp->nd_mrep; 1352 while (mp->m_next) 1353 mp = mp->m_next; 1354 mp->m_next = nfsd->nd_mrep; 1355 owp->nd_eoff = nfsd->nd_eoff; 1356 } else 1357 m_freem(nfsd->nd_mrep); 1358 nfsd->nd_mrep = NULL; 1359 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1360 owp->nd_stable = NFSV3WRITE_FILESYNC; 1361 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1362 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1363 owp->nd_stable = NFSV3WRITE_DATASYNC; 1364 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1365 /* 1366 * nfsd might hold coalesce elements! Move them to owp. 1367 * Otherwise, requests may be lost and clients will be stuck. 1368 */ 1369 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1370 LIST_REMOVE(m, nd_tq); 1371 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1372 } 1373 } 1374 1375 /* 1376 * nfs create service 1377 * now does a truncate to 0 length via. setattr if it already exists 1378 */ 1379 int 1380 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1381 { 1382 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1383 struct mbuf *nam = nfsd->nd_nam; 1384 char *dpos = nfsd->nd_dpos; 1385 kauth_cred_t cred = nfsd->nd_cr; 1386 struct nfs_fattr *fp; 1387 struct vattr va, dirfor, diraft; 1388 struct nfsv2_sattr *sp; 1389 u_int32_t *tl; 1390 struct nameidata nd; 1391 char *cp; 1392 int32_t t1; 1393 char *bpos; 1394 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1395 int rdev = 0; 1396 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1397 char *cp2; 1398 struct mbuf *mb, *mreq; 1399 struct vnode *vp = NULL, *dirp = NULL; 1400 nfsrvfh_t nsfh; 1401 u_quad_t frev, tempsize; 1402 u_char cverf[NFSX_V3CREATEVERF]; 1403 1404 nd.ni_cnd.cn_nameiop = 0; 1405 nfsm_srvmtofh(&nsfh); 1406 nfsm_srvnamesiz(len); 1407 nd.ni_cnd.cn_cred = cred; 1408 nd.ni_cnd.cn_nameiop = CREATE; 1409 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1410 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1411 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1412 if (dirp && v3) { 1413 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1414 } 1415 if (error) { 1416 nfsm_reply(NFSX_WCCDATA(v3)); 1417 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1418 if (dirp) 1419 vrele(dirp); 1420 return (0); 1421 } 1422 VATTR_NULL(&va); 1423 if (v3) { 1424 va.va_mode = 0; 1425 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1426 how = fxdr_unsigned(int, *tl); 1427 switch (how) { 1428 case NFSV3CREATE_GUARDED: 1429 if (nd.ni_vp) { 1430 error = EEXIST; 1431 break; 1432 } 1433 case NFSV3CREATE_UNCHECKED: 1434 nfsm_srvsattr(&va); 1435 break; 1436 case NFSV3CREATE_EXCLUSIVE: 1437 nfsm_dissect(cp, void *, NFSX_V3CREATEVERF); 1438 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1439 exclusive_flag = 1; 1440 break; 1441 }; 1442 va.va_type = VREG; 1443 } else { 1444 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1445 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1446 if (va.va_type == VNON) 1447 va.va_type = VREG; 1448 va.va_mode = nfstov_mode(sp->sa_mode); 1449 switch (va.va_type) { 1450 case VREG: 1451 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1452 if (tsize != -1) 1453 va.va_size = (u_quad_t)tsize; 1454 break; 1455 case VCHR: 1456 case VBLK: 1457 case VFIFO: 1458 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1459 break; 1460 default: 1461 break; 1462 }; 1463 } 1464 1465 /* 1466 * Iff doesn't exist, create it 1467 * otherwise just truncate to 0 length 1468 * should I set the mode too ?? 1469 */ 1470 if (nd.ni_vp == NULL) { 1471 if (va.va_type == VREG || va.va_type == VSOCK) { 1472 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1473 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1474 if (!error) { 1475 if (exclusive_flag) { 1476 exclusive_flag = 0; 1477 VATTR_NULL(&va); 1478 /* 1479 * XXX 1480 * assuming NFSX_V3CREATEVERF 1481 * == sizeof(nfstime3) 1482 */ 1483 fxdr_nfsv3time(cverf, &va.va_atime); 1484 error = VOP_SETATTR(nd.ni_vp, &va, 1485 cred); 1486 } 1487 } 1488 } else if (va.va_type == VCHR || va.va_type == VBLK || 1489 va.va_type == VFIFO) { 1490 if (va.va_type == VCHR && rdev == 0xffffffff) 1491 va.va_type = VFIFO; 1492 if (va.va_type != VFIFO && 1493 (error = kauth_authorize_system(cred, 1494 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1495 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1496 vput(nd.ni_dvp); 1497 nfsm_reply(0); 1498 return (error); 1499 } else 1500 va.va_rdev = (dev_t)rdev; 1501 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1502 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1503 &va); 1504 if (error) { 1505 nfsm_reply(0); 1506 } 1507 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1508 vput(nd.ni_vp); 1509 vrele(nd.ni_dvp); 1510 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1511 error = EINVAL; 1512 nfsm_reply(0); 1513 } 1514 } else { 1515 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1516 vput(nd.ni_dvp); 1517 error = ENXIO; 1518 } 1519 vp = nd.ni_vp; 1520 } else { 1521 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1522 vp = nd.ni_vp; 1523 if (nd.ni_dvp == vp) 1524 vrele(nd.ni_dvp); 1525 else 1526 vput(nd.ni_dvp); 1527 if (!error && va.va_size != -1) { 1528 error = nfsrv_access(vp, VWRITE, cred, 1529 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1530 if (!error) { 1531 nqsrv_getl(vp, ND_WRITE); 1532 tempsize = va.va_size; 1533 VATTR_NULL(&va); 1534 va.va_size = tempsize; 1535 error = VOP_SETATTR(vp, &va, cred); 1536 } 1537 } 1538 if (error) 1539 vput(vp); 1540 } 1541 if (!error) { 1542 error = nfsrv_composefh(vp, &nsfh, v3); 1543 if (!error) 1544 error = VOP_GETATTR(vp, &va, cred); 1545 vput(vp); 1546 } 1547 if (v3) { 1548 if (exclusive_flag && !error) { 1549 /* 1550 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1551 */ 1552 char oldverf[NFSX_V3CREATEVERF]; 1553 1554 txdr_nfsv3time(&va.va_atime, oldverf); 1555 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1556 error = EEXIST; 1557 } 1558 if (dirp) { 1559 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1560 } 1561 } 1562 if (dirp) { 1563 vrele(dirp); 1564 } 1565 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1566 if (v3) { 1567 if (!error) { 1568 nfsm_srvpostop_fh(&nsfh); 1569 nfsm_srvpostop_attr(0, &va); 1570 } 1571 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1572 } else { 1573 nfsm_srvfhtom(&nsfh, v3); 1574 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1575 nfsm_srvfillattr(&va, fp); 1576 } 1577 return (0); 1578 nfsmout: 1579 if (dirp) 1580 vrele(dirp); 1581 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1582 if (nd.ni_dvp == nd.ni_vp) 1583 vrele(nd.ni_dvp); 1584 else 1585 vput(nd.ni_dvp); 1586 if (nd.ni_vp) 1587 vput(nd.ni_vp); 1588 return (error); 1589 } 1590 1591 /* 1592 * nfs v3 mknod service 1593 */ 1594 int 1595 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1596 { 1597 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1598 struct mbuf *nam = nfsd->nd_nam; 1599 char *dpos = nfsd->nd_dpos; 1600 kauth_cred_t cred = nfsd->nd_cr; 1601 struct vattr va, dirfor, diraft; 1602 u_int32_t *tl; 1603 struct nameidata nd; 1604 int32_t t1; 1605 char *bpos; 1606 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1607 u_int32_t major, minor; 1608 enum vtype vtyp; 1609 char *cp2; 1610 struct mbuf *mb, *mreq; 1611 struct vnode *vp, *dirp = (struct vnode *)0; 1612 nfsrvfh_t nsfh; 1613 u_quad_t frev; 1614 1615 nd.ni_cnd.cn_nameiop = 0; 1616 nfsm_srvmtofh(&nsfh); 1617 nfsm_srvnamesiz(len); 1618 nd.ni_cnd.cn_cred = cred; 1619 nd.ni_cnd.cn_nameiop = CREATE; 1620 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1621 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1622 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1623 if (dirp) 1624 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1625 if (error) { 1626 nfsm_reply(NFSX_WCCDATA(1)); 1627 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1628 if (dirp) 1629 vrele(dirp); 1630 return (0); 1631 } 1632 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1633 vtyp = nfsv3tov_type(*tl); 1634 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1635 error = NFSERR_BADTYPE; 1636 goto abort; 1637 } 1638 VATTR_NULL(&va); 1639 va.va_mode = 0; 1640 nfsm_srvsattr(&va); 1641 if (vtyp == VCHR || vtyp == VBLK) { 1642 dev_t rdev; 1643 1644 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1645 major = fxdr_unsigned(u_int32_t, *tl++); 1646 minor = fxdr_unsigned(u_int32_t, *tl); 1647 rdev = makedev(major, minor); 1648 if (major(rdev) != major || minor(rdev) != minor) { 1649 error = EINVAL; 1650 goto abort; 1651 } 1652 va.va_rdev = rdev; 1653 } 1654 1655 /* 1656 * Iff doesn't exist, create it. 1657 */ 1658 if (nd.ni_vp) { 1659 error = EEXIST; 1660 abort: 1661 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1662 if (nd.ni_dvp == nd.ni_vp) 1663 vrele(nd.ni_dvp); 1664 else 1665 vput(nd.ni_dvp); 1666 if (nd.ni_vp) 1667 vput(nd.ni_vp); 1668 goto out; 1669 } 1670 va.va_type = vtyp; 1671 if (vtyp == VSOCK) { 1672 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1673 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1674 } else { 1675 if (va.va_type != VFIFO && 1676 (error = kauth_authorize_system(cred, 1677 KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) { 1678 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1679 vput(nd.ni_dvp); 1680 goto out; 1681 } 1682 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1683 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1684 if (error) 1685 goto out; 1686 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1687 vput(nd.ni_vp); 1688 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1689 error = EINVAL; 1690 } 1691 } 1692 out: 1693 vp = nd.ni_vp; 1694 if (!error) { 1695 error = nfsrv_composefh(vp, &nsfh, true); 1696 if (!error) 1697 error = VOP_GETATTR(vp, &va, cred); 1698 vput(vp); 1699 } 1700 if (dirp) { 1701 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1702 vrele(dirp); 1703 } 1704 nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) + 1705 NFSX_WCCDATA(1)); 1706 if (!error) { 1707 nfsm_srvpostop_fh(&nsfh); 1708 nfsm_srvpostop_attr(0, &va); 1709 } 1710 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1711 return (0); 1712 nfsmout: 1713 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1714 if (nd.ni_dvp == nd.ni_vp) 1715 vrele(nd.ni_dvp); 1716 else 1717 vput(nd.ni_dvp); 1718 if (nd.ni_vp) 1719 vput(nd.ni_vp); 1720 if (dirp) 1721 vrele(dirp); 1722 return (error); 1723 } 1724 1725 /* 1726 * nfs remove service 1727 */ 1728 int 1729 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1730 { 1731 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1732 struct mbuf *nam = nfsd->nd_nam; 1733 char *dpos = nfsd->nd_dpos; 1734 kauth_cred_t cred = nfsd->nd_cr; 1735 struct nameidata nd; 1736 u_int32_t *tl; 1737 int32_t t1; 1738 char *bpos; 1739 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1740 int v3 = (nfsd->nd_flag & ND_NFSV3); 1741 char *cp2; 1742 struct mbuf *mb, *mreq; 1743 struct vnode *vp, *dirp; 1744 struct vattr dirfor, diraft; 1745 nfsrvfh_t nsfh; 1746 u_quad_t frev; 1747 1748 #ifndef nolint 1749 vp = (struct vnode *)0; 1750 #endif 1751 nfsm_srvmtofh(&nsfh); 1752 nfsm_srvnamesiz(len); 1753 nd.ni_cnd.cn_cred = cred; 1754 nd.ni_cnd.cn_nameiop = DELETE; 1755 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1756 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1757 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1758 if (dirp && v3) { 1759 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 1760 } 1761 if (!error) { 1762 vp = nd.ni_vp; 1763 if (vp->v_type == VDIR) { 1764 error = EPERM; 1765 goto out; 1766 } 1767 /* 1768 * The root of a mounted filesystem cannot be deleted. 1769 */ 1770 if (vp->v_vflag & VV_ROOT) { 1771 error = EBUSY; 1772 } 1773 out: 1774 if (!error) { 1775 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1776 nqsrv_getl(vp, ND_WRITE); 1777 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1778 } else { 1779 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1780 if (nd.ni_dvp == vp) 1781 vrele(nd.ni_dvp); 1782 else 1783 vput(nd.ni_dvp); 1784 vput(vp); 1785 } 1786 } 1787 if (dirp) { 1788 if (v3) { 1789 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 1790 } 1791 vrele(dirp); 1792 } 1793 nfsm_reply(NFSX_WCCDATA(v3)); 1794 if (v3) { 1795 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1796 return (0); 1797 } 1798 nfsm_srvdone; 1799 } 1800 1801 /* 1802 * nfs rename service 1803 */ 1804 int 1805 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 1806 { 1807 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1808 struct mbuf *nam = nfsd->nd_nam; 1809 char *dpos = nfsd->nd_dpos; 1810 kauth_cred_t cred = nfsd->nd_cr; 1811 u_int32_t *tl; 1812 int32_t t1; 1813 char *bpos; 1814 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1815 uint32_t len, len2; 1816 int tdirfor_ret = 1, tdiraft_ret = 1; 1817 int v3 = (nfsd->nd_flag & ND_NFSV3); 1818 char *cp2; 1819 struct mbuf *mb, *mreq; 1820 struct nameidata fromnd, tond; 1821 struct vnode *fvp, *tvp, *tdvp; 1822 struct vnode *fdirp = NULL, *tdirp = NULL; 1823 struct mount *localfs = NULL; 1824 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1825 nfsrvfh_t fnsfh, tnsfh; 1826 u_quad_t frev; 1827 uid_t saved_uid; 1828 uint32_t saveflag; 1829 1830 #ifndef nolint 1831 fvp = (struct vnode *)0; 1832 #endif 1833 fromnd.ni_cnd.cn_nameiop = 0; 1834 tond.ni_cnd.cn_nameiop = 0; 1835 nfsm_srvmtofh(&fnsfh); 1836 nfsm_srvnamesiz(len); 1837 /* 1838 * Remember our original uid so that we can reset cr_uid before 1839 * the second nfs_namei() call, in case it is remapped. 1840 */ 1841 saved_uid = kauth_cred_geteuid(cred); 1842 fromnd.ni_cnd.cn_cred = cred; 1843 fromnd.ni_cnd.cn_nameiop = DELETE; 1844 fromnd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART; 1845 error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md, 1846 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1847 if (fdirp && v3) { 1848 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred); 1849 } 1850 if (error) { 1851 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1852 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1853 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1854 if (fdirp) 1855 vrele(fdirp); 1856 return (0); 1857 } 1858 if (fromnd.ni_dvp != fromnd.ni_vp) { 1859 VOP_UNLOCK(fromnd.ni_dvp, 0); 1860 } 1861 fvp = fromnd.ni_vp; 1862 1863 localfs = fvp->v_mount; 1864 error = VFS_RENAMELOCK_ENTER(localfs); 1865 if (error) { 1866 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1867 vrele(fromnd.ni_dvp); 1868 vrele(fvp); 1869 goto out1; 1870 } 1871 1872 /* Copied, regrettably, from vfs_syscalls.c (q.v.) */ 1873 vrele(fvp); 1874 if ((fromnd.ni_cnd.cn_namelen == 1 && 1875 fromnd.ni_cnd.cn_nameptr[0] == '.') || 1876 (fromnd.ni_cnd.cn_namelen == 2 && 1877 fromnd.ni_cnd.cn_nameptr[0] == '.' && 1878 fromnd.ni_cnd.cn_nameptr[1] == '.')) { 1879 error = EINVAL; 1880 VFS_RENAMELOCK_EXIT(localfs); 1881 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1882 vrele(fromnd.ni_dvp); 1883 goto out1; 1884 } 1885 saveflag = fromnd.ni_cnd.cn_flags & SAVESTART; 1886 fromnd.ni_cnd.cn_flags &= ~SAVESTART; 1887 vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY); 1888 error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd); 1889 fromnd.ni_cnd.cn_flags |= saveflag; 1890 if (error) { 1891 VOP_UNLOCK(fromnd.ni_dvp, 0); 1892 VFS_RENAMELOCK_EXIT(localfs); 1893 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1894 vrele(fromnd.ni_dvp); 1895 goto out1; 1896 } 1897 VOP_UNLOCK(fromnd.ni_vp, 0); 1898 if (fromnd.ni_dvp != fromnd.ni_vp) 1899 VOP_UNLOCK(fromnd.ni_dvp, 0); 1900 fvp = fromnd.ni_vp; 1901 1902 nfsm_srvmtofh(&tnsfh); 1903 if (v3) { 1904 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1905 len2 = fxdr_unsigned(uint32_t, *tl); 1906 /* len2 will be checked by nfs_namei */ 1907 } 1908 else { 1909 /* NFSv2 */ 1910 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1911 } 1912 kauth_cred_seteuid(cred, saved_uid); 1913 tond.ni_cnd.cn_cred = cred; 1914 tond.ni_cnd.cn_nameiop = RENAME; 1915 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1916 error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md, 1917 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1918 if (tdirp && v3) { 1919 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred); 1920 } 1921 if (error) { 1922 VFS_RENAMELOCK_EXIT(localfs); 1923 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1924 vrele(fromnd.ni_dvp); 1925 vrele(fvp); 1926 goto out1; 1927 } 1928 tdvp = tond.ni_dvp; 1929 tvp = tond.ni_vp; 1930 if (tvp != NULL) { 1931 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1932 if (v3) 1933 error = EEXIST; 1934 else 1935 error = EISDIR; 1936 goto out; 1937 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1938 if (v3) 1939 error = EEXIST; 1940 else 1941 error = ENOTDIR; 1942 goto out; 1943 } 1944 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1945 if (v3) 1946 error = EXDEV; 1947 else 1948 error = ENOTEMPTY; 1949 goto out; 1950 } 1951 } 1952 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1953 if (v3) 1954 error = EXDEV; 1955 else 1956 error = ENOTEMPTY; 1957 goto out; 1958 } 1959 if (fvp->v_mount != tdvp->v_mount) { 1960 if (v3) 1961 error = EXDEV; 1962 else 1963 error = ENOTEMPTY; 1964 goto out; 1965 } 1966 if (fvp == tdvp) { 1967 if (v3) 1968 error = EINVAL; 1969 else 1970 error = ENOTEMPTY; 1971 } 1972 /* 1973 * If source is the same as the destination (that is the 1974 * same vnode with the same name in the same directory), 1975 * then there is nothing to do. 1976 */ 1977 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1978 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1979 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1980 fromnd.ni_cnd.cn_namelen)) 1981 error = -1; 1982 out: 1983 if (!error) { 1984 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1985 nqsrv_getl(tdvp, ND_WRITE); 1986 if (tvp) { 1987 nqsrv_getl(tvp, ND_WRITE); 1988 } 1989 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1990 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1991 VFS_RENAMELOCK_EXIT(localfs); 1992 } else { 1993 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 1994 if (tdvp == tvp) 1995 vrele(tdvp); 1996 else 1997 vput(tdvp); 1998 if (tvp) 1999 vput(tvp); 2000 VFS_RENAMELOCK_EXIT(localfs); 2001 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2002 vrele(fromnd.ni_dvp); 2003 vrele(fvp); 2004 if (error == -1) 2005 error = 0; 2006 } 2007 vrele(tond.ni_startdir); 2008 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2009 out1: 2010 if (fdirp) { 2011 if (v3) { 2012 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred); 2013 } 2014 vrele(fdirp); 2015 } 2016 if (tdirp) { 2017 if (v3) { 2018 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred); 2019 } 2020 vrele(tdirp); 2021 } 2022 vrele(fromnd.ni_startdir); 2023 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2024 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2025 if (v3) { 2026 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2027 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2028 } 2029 return (0); 2030 2031 nfsmout: 2032 if (fdirp) 2033 vrele(fdirp); 2034 #ifdef notdef 2035 if (tdirp) 2036 vrele(tdirp); 2037 #endif 2038 if (tond.ni_cnd.cn_nameiop) { 2039 vrele(tond.ni_startdir); 2040 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2041 } 2042 if (localfs) { 2043 VFS_RENAMELOCK_EXIT(localfs); 2044 } 2045 if (fromnd.ni_cnd.cn_nameiop) { 2046 vrele(fromnd.ni_startdir); 2047 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2048 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2049 vrele(fromnd.ni_dvp); 2050 vrele(fvp); 2051 } 2052 return (error); 2053 } 2054 2055 /* 2056 * nfs link service 2057 */ 2058 int 2059 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2060 { 2061 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2062 struct mbuf *nam = nfsd->nd_nam; 2063 char *dpos = nfsd->nd_dpos; 2064 kauth_cred_t cred = nfsd->nd_cr; 2065 struct nameidata nd; 2066 u_int32_t *tl; 2067 int32_t t1; 2068 char *bpos; 2069 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2070 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2071 char *cp2; 2072 struct mbuf *mb, *mreq; 2073 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2074 struct vattr dirfor, diraft, at; 2075 nfsrvfh_t nsfh, dnsfh; 2076 u_quad_t frev; 2077 2078 nfsm_srvmtofh(&nsfh); 2079 nfsm_srvmtofh(&dnsfh); 2080 nfsm_srvnamesiz(len); 2081 error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam, 2082 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2083 if (error) { 2084 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2085 nfsm_srvpostop_attr(getret, &at); 2086 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2087 return (0); 2088 } 2089 if (vp->v_type == VDIR) { 2090 error = EPERM; 2091 goto out1; 2092 } 2093 nd.ni_cnd.cn_cred = cred; 2094 nd.ni_cnd.cn_nameiop = CREATE; 2095 nd.ni_cnd.cn_flags = LOCKPARENT; 2096 error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos, 2097 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2098 if (dirp && v3) { 2099 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2100 } 2101 if (error) 2102 goto out1; 2103 xp = nd.ni_vp; 2104 if (xp != NULL) { 2105 error = EEXIST; 2106 goto out; 2107 } 2108 xp = nd.ni_dvp; 2109 if (vp->v_mount != xp->v_mount) 2110 error = EXDEV; 2111 out: 2112 if (!error) { 2113 nqsrv_getl(vp, ND_WRITE); 2114 nqsrv_getl(xp, ND_WRITE); 2115 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2116 } else { 2117 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2118 if (nd.ni_dvp == nd.ni_vp) 2119 vrele(nd.ni_dvp); 2120 else 2121 vput(nd.ni_dvp); 2122 if (nd.ni_vp) 2123 vrele(nd.ni_vp); 2124 } 2125 out1: 2126 if (v3) 2127 getret = VOP_GETATTR(vp, &at, cred); 2128 if (dirp) { 2129 if (v3) { 2130 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2131 } 2132 vrele(dirp); 2133 } 2134 vrele(vp); 2135 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2136 if (v3) { 2137 nfsm_srvpostop_attr(getret, &at); 2138 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2139 return (0); 2140 } 2141 nfsm_srvdone; 2142 } 2143 2144 /* 2145 * nfs symbolic link service 2146 */ 2147 int 2148 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2149 { 2150 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2151 struct mbuf *nam = nfsd->nd_nam; 2152 char *dpos = nfsd->nd_dpos; 2153 kauth_cred_t cred = nfsd->nd_cr; 2154 struct vattr va, dirfor, diraft; 2155 struct nameidata nd; 2156 u_int32_t *tl; 2157 int32_t t1; 2158 struct nfsv2_sattr *sp; 2159 char *bpos, *pathcp = NULL, *cp2; 2160 struct uio io; 2161 struct iovec iv; 2162 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1; 2163 uint32_t len, len2; 2164 int v3 = (nfsd->nd_flag & ND_NFSV3); 2165 struct mbuf *mb, *mreq; 2166 struct vnode *dirp = (struct vnode *)0; 2167 nfsrvfh_t nsfh; 2168 u_quad_t frev; 2169 2170 nd.ni_cnd.cn_nameiop = 0; 2171 nfsm_srvmtofh(&nsfh); 2172 nfsm_srvnamesiz(len); 2173 nd.ni_cnd.cn_cred = cred; 2174 nd.ni_cnd.cn_nameiop = CREATE; 2175 nd.ni_cnd.cn_flags = LOCKPARENT; 2176 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2177 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2178 if (dirp && v3) { 2179 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2180 } 2181 if (error) 2182 goto out; 2183 VATTR_NULL(&va); 2184 va.va_type = VLNK; 2185 if (v3) { 2186 va.va_mode = 0; 2187 nfsm_srvsattr(&va); 2188 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2189 len2 = fxdr_unsigned(uint32_t, *tl); 2190 if (len2 > PATH_MAX) { 2191 /* XXX should check _PC_NO_TRUNC */ 2192 error = ENAMETOOLONG; 2193 goto abortop; 2194 } 2195 } 2196 else { 2197 /* NFSv2 */ 2198 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2199 } 2200 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2201 iv.iov_base = pathcp; 2202 iv.iov_len = len2; 2203 io.uio_resid = len2; 2204 io.uio_offset = 0; 2205 io.uio_iov = &iv; 2206 io.uio_iovcnt = 1; 2207 io.uio_rw = UIO_READ; 2208 UIO_SETUP_SYSSPACE(&io); 2209 nfsm_mtouio(&io, len2); 2210 if (!v3) { 2211 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2212 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2213 } 2214 *(pathcp + len2) = '\0'; 2215 if (nd.ni_vp) { 2216 error = EEXIST; 2217 abortop: 2218 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2219 if (nd.ni_dvp == nd.ni_vp) 2220 vrele(nd.ni_dvp); 2221 else 2222 vput(nd.ni_dvp); 2223 if (nd.ni_vp) 2224 vrele(nd.ni_vp); 2225 goto out; 2226 } 2227 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2228 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2229 if (!error) { 2230 if (v3) { 2231 error = nfsrv_composefh(nd.ni_vp, &nsfh, v3); 2232 if (!error) 2233 error = VOP_GETATTR(nd.ni_vp, &va, cred); 2234 vput(nd.ni_vp); 2235 } else { 2236 vput(nd.ni_vp); 2237 } 2238 } 2239 out: 2240 if (pathcp) 2241 free(pathcp, M_TEMP); 2242 if (dirp) { 2243 if (v3) { 2244 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2245 } 2246 vrele(dirp); 2247 } 2248 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2249 NFSX_WCCDATA(v3)); 2250 if (v3) { 2251 if (!error) { 2252 nfsm_srvpostop_fh(&nsfh); 2253 nfsm_srvpostop_attr(0, &va); 2254 } 2255 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2256 } 2257 return (0); 2258 nfsmout: 2259 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2260 if (nd.ni_dvp == nd.ni_vp) 2261 vrele(nd.ni_dvp); 2262 else 2263 vput(nd.ni_dvp); 2264 if (nd.ni_vp) 2265 vrele(nd.ni_vp); 2266 if (dirp) 2267 vrele(dirp); 2268 if (pathcp) 2269 free(pathcp, M_TEMP); 2270 return (error); 2271 } 2272 2273 /* 2274 * nfs mkdir service 2275 */ 2276 int 2277 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2278 { 2279 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2280 struct mbuf *nam = nfsd->nd_nam; 2281 char *dpos = nfsd->nd_dpos; 2282 kauth_cred_t cred = nfsd->nd_cr; 2283 struct vattr va, dirfor, diraft; 2284 struct nfs_fattr *fp; 2285 struct nameidata nd; 2286 char *cp; 2287 u_int32_t *tl; 2288 int32_t t1; 2289 char *bpos; 2290 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2291 int v3 = (nfsd->nd_flag & ND_NFSV3); 2292 char *cp2; 2293 struct mbuf *mb, *mreq; 2294 struct vnode *vp, *dirp = (struct vnode *)0; 2295 nfsrvfh_t nsfh; 2296 u_quad_t frev; 2297 2298 nfsm_srvmtofh(&nsfh); 2299 nfsm_srvnamesiz(len); 2300 nd.ni_cnd.cn_cred = cred; 2301 nd.ni_cnd.cn_nameiop = CREATE; 2302 nd.ni_cnd.cn_flags = LOCKPARENT; 2303 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2304 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2305 if (dirp && v3) { 2306 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2307 } 2308 if (error) { 2309 nfsm_reply(NFSX_WCCDATA(v3)); 2310 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2311 if (dirp) 2312 vrele(dirp); 2313 return (0); 2314 } 2315 VATTR_NULL(&va); 2316 if (v3) { 2317 va.va_mode = 0; 2318 nfsm_srvsattr(&va); 2319 } else { 2320 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2321 va.va_mode = nfstov_mode(*tl++); 2322 } 2323 va.va_type = VDIR; 2324 vp = nd.ni_vp; 2325 if (vp != NULL) { 2326 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2327 if (nd.ni_dvp == vp) 2328 vrele(nd.ni_dvp); 2329 else 2330 vput(nd.ni_dvp); 2331 vrele(vp); 2332 error = EEXIST; 2333 goto out; 2334 } 2335 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2336 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2337 if (!error) { 2338 vp = nd.ni_vp; 2339 error = nfsrv_composefh(vp, &nsfh, v3); 2340 if (!error) 2341 error = VOP_GETATTR(vp, &va, cred); 2342 vput(vp); 2343 } 2344 out: 2345 if (dirp) { 2346 if (v3) { 2347 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2348 } 2349 vrele(dirp); 2350 } 2351 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2352 NFSX_WCCDATA(v3)); 2353 if (v3) { 2354 if (!error) { 2355 nfsm_srvpostop_fh(&nsfh); 2356 nfsm_srvpostop_attr(0, &va); 2357 } 2358 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2359 } else { 2360 nfsm_srvfhtom(&nsfh, v3); 2361 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2362 nfsm_srvfillattr(&va, fp); 2363 } 2364 return (0); 2365 nfsmout: 2366 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2367 if (nd.ni_dvp == nd.ni_vp) 2368 vrele(nd.ni_dvp); 2369 else 2370 vput(nd.ni_dvp); 2371 if (nd.ni_vp) 2372 vrele(nd.ni_vp); 2373 if (dirp) 2374 vrele(dirp); 2375 return (error); 2376 } 2377 2378 /* 2379 * nfs rmdir service 2380 */ 2381 int 2382 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2383 { 2384 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2385 struct mbuf *nam = nfsd->nd_nam; 2386 char *dpos = nfsd->nd_dpos; 2387 kauth_cred_t cred = nfsd->nd_cr; 2388 u_int32_t *tl; 2389 int32_t t1; 2390 char *bpos; 2391 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2392 int v3 = (nfsd->nd_flag & ND_NFSV3); 2393 char *cp2; 2394 struct mbuf *mb, *mreq; 2395 struct vnode *vp, *dirp = (struct vnode *)0; 2396 struct vattr dirfor, diraft; 2397 nfsrvfh_t nsfh; 2398 struct nameidata nd; 2399 u_quad_t frev; 2400 2401 nfsm_srvmtofh(&nsfh); 2402 nfsm_srvnamesiz(len); 2403 nd.ni_cnd.cn_cred = cred; 2404 nd.ni_cnd.cn_nameiop = DELETE; 2405 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2406 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2407 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2408 if (dirp && v3) { 2409 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred); 2410 } 2411 if (error) { 2412 nfsm_reply(NFSX_WCCDATA(v3)); 2413 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2414 if (dirp) 2415 vrele(dirp); 2416 return (0); 2417 } 2418 vp = nd.ni_vp; 2419 if (vp->v_type != VDIR) { 2420 error = ENOTDIR; 2421 goto out; 2422 } 2423 /* 2424 * No rmdir "." please. 2425 */ 2426 if (nd.ni_dvp == vp) { 2427 error = EINVAL; 2428 goto out; 2429 } 2430 /* 2431 * The root of a mounted filesystem cannot be deleted. 2432 */ 2433 if (vp->v_vflag & VV_ROOT) 2434 error = EBUSY; 2435 out: 2436 if (!error) { 2437 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2438 nqsrv_getl(vp, ND_WRITE); 2439 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2440 } else { 2441 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2442 if (nd.ni_dvp == nd.ni_vp) 2443 vrele(nd.ni_dvp); 2444 else 2445 vput(nd.ni_dvp); 2446 vput(vp); 2447 } 2448 if (dirp) { 2449 if (v3) { 2450 diraft_ret = VOP_GETATTR(dirp, &diraft, cred); 2451 } 2452 vrele(dirp); 2453 } 2454 nfsm_reply(NFSX_WCCDATA(v3)); 2455 if (v3) { 2456 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2457 return (0); 2458 } 2459 nfsm_srvdone; 2460 } 2461 2462 /* 2463 * nfs readdir service 2464 * - mallocs what it thinks is enough to read 2465 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2466 * - calls VOP_READDIR() 2467 * - loops around building the reply 2468 * if the output generated exceeds count break out of loop 2469 * The nfsm_clget macro is used here so that the reply will be packed 2470 * tightly in mbuf clusters. 2471 * - it only knows that it has encountered eof when the VOP_READDIR() 2472 * reads nothing 2473 * - as such one readdir rpc will return eof false although you are there 2474 * and then the next will return eof 2475 * - it trims out records with d_fileno == 0 2476 * this doesn't matter for Unix clients, but they might confuse clients 2477 * for other os'. 2478 * - it trims out records with d_type == DT_WHT 2479 * these cannot be seen through NFS (unless we extend the protocol) 2480 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2481 * than requested, but this may not apply to all filesystems. For 2482 * example, client NFS does not { although it is never remote mounted 2483 * anyhow } 2484 * The alternate call nfsrv_readdirplus() does lookups as well. 2485 * PS: The NFS protocol spec. does not clarify what the "count" byte 2486 * argument is a count of.. just name strings and file id's or the 2487 * entire reply rpc or ... 2488 * I tried just file name and id sizes and it confused the Sun client, 2489 * so I am using the full rpc size now. The "paranoia.." comment refers 2490 * to including the status longwords that are not a part of the dir. 2491 * "entry" structures, but are in the rpc. 2492 */ 2493 2494 #define NFS_SRVDIRBLKSIZ 1024 2495 2496 struct flrep { 2497 nfsuint64 fl_off; 2498 u_int32_t fl_postopok; 2499 struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */ 2500 u_int32_t fl_fhok; 2501 u_int32_t fl_fhsize; 2502 /* handle comes here, filled in dynamically */ 2503 }; 2504 2505 int 2506 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2507 { 2508 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2509 struct mbuf *nam = nfsd->nd_nam; 2510 char *dpos = nfsd->nd_dpos; 2511 kauth_cred_t cred = nfsd->nd_cr; 2512 char *bp, *be; 2513 struct mbuf *mp; 2514 struct dirent *dp; 2515 char *cp; 2516 u_int32_t *tl; 2517 int32_t t1; 2518 char *bpos; 2519 struct mbuf *mb, *mreq, *mp2; 2520 char *cpos, *cend, *cp2, *rbuf; 2521 struct vnode *vp; 2522 struct vattr at; 2523 nfsrvfh_t nsfh; 2524 struct uio io; 2525 struct iovec iv; 2526 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2527 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2528 int v3 = (nfsd->nd_flag & ND_NFSV3); 2529 u_quad_t frev, off, toff, verf; 2530 off_t *cookies = NULL, *cookiep; 2531 nfsuint64 jar; 2532 2533 nfsm_srvmtofh(&nsfh); 2534 if (v3) { 2535 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2536 toff = fxdr_hyper(tl); 2537 tl += 2; 2538 verf = fxdr_hyper(tl); 2539 tl += 2; 2540 } else { 2541 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2542 toff = fxdr_unsigned(u_quad_t, *tl++); 2543 } 2544 off = toff; 2545 cnt = fxdr_unsigned(int, *tl); 2546 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2547 xfer = NFS_SRVMAXDATA(nfsd); 2548 if (siz > xfer) 2549 siz = xfer; 2550 fullsiz = siz; 2551 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2552 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2553 if (!error && vp->v_type != VDIR) { 2554 error = ENOTDIR; 2555 vput(vp); 2556 } 2557 if (error) { 2558 nfsm_reply(NFSX_UNSIGNED); 2559 nfsm_srvpostop_attr(getret, &at); 2560 return (0); 2561 } 2562 nqsrv_getl(vp, ND_READ); 2563 if (v3) { 2564 error = getret = VOP_GETATTR(vp, &at, cred); 2565 #ifdef NFS3_STRICTVERF 2566 /* 2567 * XXX This check is too strict for Solaris 2.5 clients. 2568 */ 2569 if (!error && toff && verf != at.va_filerev) 2570 error = NFSERR_BAD_COOKIE; 2571 #endif 2572 } 2573 if (!error) 2574 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2575 if (error) { 2576 vput(vp); 2577 nfsm_reply(NFSX_POSTOPATTR(v3)); 2578 nfsm_srvpostop_attr(getret, &at); 2579 return (0); 2580 } 2581 VOP_UNLOCK(vp, 0); 2582 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2583 again: 2584 iv.iov_base = rbuf; 2585 iv.iov_len = fullsiz; 2586 io.uio_iov = &iv; 2587 io.uio_iovcnt = 1; 2588 io.uio_offset = (off_t)off; 2589 io.uio_resid = fullsiz; 2590 io.uio_rw = UIO_READ; 2591 UIO_SETUP_SYSSPACE(&io); 2592 eofflag = 0; 2593 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2594 2595 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2596 2597 off = (off_t)io.uio_offset; 2598 if (!cookies && !error) 2599 error = NFSERR_PERM; 2600 if (v3) { 2601 getret = VOP_GETATTR(vp, &at, cred); 2602 if (!error) 2603 error = getret; 2604 } 2605 2606 VOP_UNLOCK(vp, 0); 2607 if (error) { 2608 vrele(vp); 2609 free((void *)rbuf, M_TEMP); 2610 if (cookies) 2611 free((void *)cookies, M_TEMP); 2612 nfsm_reply(NFSX_POSTOPATTR(v3)); 2613 nfsm_srvpostop_attr(getret, &at); 2614 return (0); 2615 } 2616 if (io.uio_resid) { 2617 siz -= io.uio_resid; 2618 2619 /* 2620 * If nothing read, return eof 2621 * rpc reply 2622 */ 2623 if (siz == 0) { 2624 vrele(vp); 2625 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2626 2 * NFSX_UNSIGNED); 2627 if (v3) { 2628 nfsm_srvpostop_attr(getret, &at); 2629 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2630 txdr_hyper(at.va_filerev, tl); 2631 tl += 2; 2632 } else 2633 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2634 *tl++ = nfs_false; 2635 *tl = nfs_true; 2636 free((void *)rbuf, M_TEMP); 2637 free((void *)cookies, M_TEMP); 2638 return (0); 2639 } 2640 } 2641 2642 /* 2643 * Check for degenerate cases of nothing useful read. 2644 * If so go try again 2645 */ 2646 cpos = rbuf; 2647 cend = rbuf + siz; 2648 dp = (struct dirent *)cpos; 2649 cookiep = cookies; 2650 2651 while (cpos < cend && ncookies > 0 && 2652 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2653 cpos += dp->d_reclen; 2654 dp = (struct dirent *)cpos; 2655 cookiep++; 2656 ncookies--; 2657 } 2658 if (cpos >= cend || ncookies == 0) { 2659 toff = off; 2660 siz = fullsiz; 2661 free(cookies, M_TEMP); 2662 cookies = NULL; 2663 goto again; 2664 } 2665 2666 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2667 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2668 if (v3) { 2669 nfsm_srvpostop_attr(getret, &at); 2670 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2671 txdr_hyper(at.va_filerev, tl); 2672 } 2673 mp = mp2 = mb; 2674 bp = bpos; 2675 be = bp + M_TRAILINGSPACE(mp); 2676 2677 /* Loop through the records and build reply */ 2678 while (cpos < cend && ncookies > 0) { 2679 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2680 nlen = dp->d_namlen; 2681 rem = nfsm_rndup(nlen)-nlen; 2682 len += (4 * NFSX_UNSIGNED + nlen + rem); 2683 if (v3) 2684 len += 2 * NFSX_UNSIGNED; 2685 if (len > cnt) { 2686 eofflag = 0; 2687 break; 2688 } 2689 /* 2690 * Build the directory record xdr from 2691 * the dirent entry. 2692 */ 2693 nfsm_clget; 2694 *tl = nfs_true; 2695 bp += NFSX_UNSIGNED; 2696 if (v3) { 2697 nfsm_clget; 2698 *tl = txdr_unsigned(dp->d_fileno >> 32); 2699 bp += NFSX_UNSIGNED; 2700 } 2701 nfsm_clget; 2702 *tl = txdr_unsigned(dp->d_fileno); 2703 bp += NFSX_UNSIGNED; 2704 nfsm_clget; 2705 *tl = txdr_unsigned(nlen); 2706 bp += NFSX_UNSIGNED; 2707 2708 /* And loop around copying the name */ 2709 xfer = nlen; 2710 cp = dp->d_name; 2711 while (xfer > 0) { 2712 nfsm_clget; 2713 if ((bp+xfer) > be) 2714 tsiz = be-bp; 2715 else 2716 tsiz = xfer; 2717 memcpy(bp, cp, tsiz); 2718 bp += tsiz; 2719 xfer -= tsiz; 2720 if (xfer > 0) 2721 cp += tsiz; 2722 } 2723 /* And null pad to an int32_t boundary */ 2724 for (i = 0; i < rem; i++) 2725 *bp++ = '\0'; 2726 nfsm_clget; 2727 2728 /* Finish off the record */ 2729 txdr_hyper(*cookiep, &jar); 2730 if (v3) { 2731 *tl = jar.nfsuquad[0]; 2732 bp += NFSX_UNSIGNED; 2733 nfsm_clget; 2734 } 2735 *tl = jar.nfsuquad[1]; 2736 bp += NFSX_UNSIGNED; 2737 } 2738 cpos += dp->d_reclen; 2739 dp = (struct dirent *)cpos; 2740 cookiep++; 2741 ncookies--; 2742 } 2743 vrele(vp); 2744 nfsm_clget; 2745 *tl = nfs_false; 2746 bp += NFSX_UNSIGNED; 2747 nfsm_clget; 2748 if (eofflag) 2749 *tl = nfs_true; 2750 else 2751 *tl = nfs_false; 2752 bp += NFSX_UNSIGNED; 2753 if (mp != mb) { 2754 if (bp < be) 2755 mp->m_len = bp - mtod(mp, char *); 2756 } else 2757 mp->m_len += bp - bpos; 2758 free((void *)rbuf, M_TEMP); 2759 free((void *)cookies, M_TEMP); 2760 nfsm_srvdone; 2761 } 2762 2763 int 2764 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 2765 { 2766 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2767 struct mbuf *nam = nfsd->nd_nam; 2768 char *dpos = nfsd->nd_dpos; 2769 kauth_cred_t cred = nfsd->nd_cr; 2770 char *bp, *be; 2771 struct mbuf *mp; 2772 struct dirent *dp; 2773 char *cp; 2774 u_int32_t *tl; 2775 int32_t t1; 2776 char *bpos; 2777 struct mbuf *mb, *mreq, *mp2; 2778 char *cpos, *cend, *cp2, *rbuf; 2779 struct vnode *vp, *nvp; 2780 struct flrep fl; 2781 nfsrvfh_t nsfh; 2782 struct uio io; 2783 struct iovec iv; 2784 struct vattr va, at, *vap = &va; 2785 struct nfs_fattr *fp; 2786 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2787 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2788 u_quad_t frev, off, toff, verf; 2789 off_t *cookies = NULL, *cookiep; 2790 2791 nfsm_srvmtofh(&nsfh); 2792 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2793 toff = fxdr_hyper(tl); 2794 tl += 2; 2795 verf = fxdr_hyper(tl); 2796 tl += 2; 2797 siz = fxdr_unsigned(int, *tl++); 2798 cnt = fxdr_unsigned(int, *tl); 2799 off = toff; 2800 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2801 xfer = NFS_SRVMAXDATA(nfsd); 2802 if (siz > xfer) 2803 siz = xfer; 2804 fullsiz = siz; 2805 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2806 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2807 if (!error && vp->v_type != VDIR) { 2808 error = ENOTDIR; 2809 vput(vp); 2810 } 2811 if (error) { 2812 nfsm_reply(NFSX_UNSIGNED); 2813 nfsm_srvpostop_attr(getret, &at); 2814 return (0); 2815 } 2816 error = getret = VOP_GETATTR(vp, &at, cred); 2817 #ifdef NFS3_STRICTVERF 2818 /* 2819 * XXX This check is too strict for Solaris 2.5 clients. 2820 */ 2821 if (!error && toff && verf != at.va_filerev) 2822 error = NFSERR_BAD_COOKIE; 2823 #endif 2824 if (!error) { 2825 nqsrv_getl(vp, ND_READ); 2826 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2827 } 2828 if (error) { 2829 vput(vp); 2830 nfsm_reply(NFSX_V3POSTOPATTR); 2831 nfsm_srvpostop_attr(getret, &at); 2832 return (0); 2833 } 2834 VOP_UNLOCK(vp, 0); 2835 2836 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2837 again: 2838 iv.iov_base = rbuf; 2839 iv.iov_len = fullsiz; 2840 io.uio_iov = &iv; 2841 io.uio_iovcnt = 1; 2842 io.uio_offset = (off_t)off; 2843 io.uio_resid = fullsiz; 2844 io.uio_rw = UIO_READ; 2845 UIO_SETUP_SYSSPACE(&io); 2846 eofflag = 0; 2847 2848 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2849 2850 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2851 2852 off = (u_quad_t)io.uio_offset; 2853 getret = VOP_GETATTR(vp, &at, cred); 2854 2855 VOP_UNLOCK(vp, 0); 2856 2857 /* 2858 * If the VGET operation doesn't work for this filesystem, 2859 * we can't support readdirplus. Returning NOTSUPP should 2860 * make clients fall back to plain readdir. 2861 * There's no need to check for VPTOFH as well, we wouldn't 2862 * even be here otherwise. 2863 */ 2864 if (!getret) { 2865 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2866 getret = (getret == EOPNOTSUPP) ? 2867 NFSERR_NOTSUPP : NFSERR_IO; 2868 else 2869 vput(nvp); 2870 } 2871 2872 if (!cookies && !error) 2873 error = NFSERR_PERM; 2874 if (!error) 2875 error = getret; 2876 if (error) { 2877 vrele(vp); 2878 if (cookies) 2879 free((void *)cookies, M_TEMP); 2880 free((void *)rbuf, M_TEMP); 2881 nfsm_reply(NFSX_V3POSTOPATTR); 2882 nfsm_srvpostop_attr(getret, &at); 2883 return (0); 2884 } 2885 if (io.uio_resid) { 2886 siz -= io.uio_resid; 2887 2888 /* 2889 * If nothing read, return eof 2890 * rpc reply 2891 */ 2892 if (siz == 0) { 2893 vrele(vp); 2894 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2895 2 * NFSX_UNSIGNED); 2896 nfsm_srvpostop_attr(getret, &at); 2897 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2898 txdr_hyper(at.va_filerev, tl); 2899 tl += 2; 2900 *tl++ = nfs_false; 2901 *tl = nfs_true; 2902 free((void *)cookies, M_TEMP); 2903 free((void *)rbuf, M_TEMP); 2904 return (0); 2905 } 2906 } 2907 2908 /* 2909 * Check for degenerate cases of nothing useful read. 2910 * If so go try again 2911 */ 2912 cpos = rbuf; 2913 cend = rbuf + siz; 2914 dp = (struct dirent *)cpos; 2915 cookiep = cookies; 2916 2917 while (cpos < cend && ncookies > 0 && 2918 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2919 cpos += dp->d_reclen; 2920 dp = (struct dirent *)cpos; 2921 cookiep++; 2922 ncookies--; 2923 } 2924 if (cpos >= cend || ncookies == 0) { 2925 toff = off; 2926 siz = fullsiz; 2927 free(cookies, M_TEMP); 2928 cookies = NULL; 2929 goto again; 2930 } 2931 2932 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 2933 nfsm_reply(cnt); 2934 nfsm_srvpostop_attr(getret, &at); 2935 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2936 txdr_hyper(at.va_filerev, tl); 2937 mp = mp2 = mb; 2938 bp = bpos; 2939 be = bp + M_TRAILINGSPACE(mp); 2940 2941 /* Loop through the records and build reply */ 2942 while (cpos < cend && ncookies > 0) { 2943 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2944 nfsrvfh_t nnsfh; 2945 2946 nlen = dp->d_namlen; 2947 rem = nfsm_rndup(nlen)-nlen; 2948 2949 /* 2950 * For readdir_and_lookup get the vnode using 2951 * the file number. 2952 */ 2953 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 2954 goto invalid; 2955 if (nfsrv_composefh(nvp, &nnsfh, true)) { 2956 vput(nvp); 2957 goto invalid; 2958 } 2959 if (VOP_GETATTR(nvp, vap, cred)) { 2960 vput(nvp); 2961 goto invalid; 2962 } 2963 vput(nvp); 2964 2965 /* 2966 * If either the dircount or maxcount will be 2967 * exceeded, get out now. Both of these lengths 2968 * are calculated conservatively, including all 2969 * XDR overheads. 2970 */ 2971 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 2972 NFSX_V3POSTOPATTR); 2973 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 2974 if (len > cnt || dirlen > fullsiz) { 2975 eofflag = 0; 2976 break; 2977 } 2978 2979 /* 2980 * Build the directory record xdr from 2981 * the dirent entry. 2982 */ 2983 fp = (struct nfs_fattr *)&fl.fl_fattr; 2984 nfsm_srvfillattr(vap, fp); 2985 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 2986 fl.fl_fhok = nfs_true; 2987 fl.fl_postopok = nfs_true; 2988 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 2989 2990 nfsm_clget; 2991 *tl = nfs_true; 2992 bp += NFSX_UNSIGNED; 2993 nfsm_clget; 2994 *tl = txdr_unsigned(dp->d_fileno >> 32); 2995 bp += NFSX_UNSIGNED; 2996 nfsm_clget; 2997 *tl = txdr_unsigned(dp->d_fileno); 2998 bp += NFSX_UNSIGNED; 2999 nfsm_clget; 3000 *tl = txdr_unsigned(nlen); 3001 bp += NFSX_UNSIGNED; 3002 3003 /* And loop around copying the name */ 3004 xfer = nlen; 3005 cp = dp->d_name; 3006 while (xfer > 0) { 3007 nfsm_clget; 3008 if ((bp + xfer) > be) 3009 tsiz = be - bp; 3010 else 3011 tsiz = xfer; 3012 memcpy(bp, cp, tsiz); 3013 bp += tsiz; 3014 xfer -= tsiz; 3015 if (xfer > 0) 3016 cp += tsiz; 3017 } 3018 /* And null pad to an int32_t boundary */ 3019 for (i = 0; i < rem; i++) 3020 *bp++ = '\0'; 3021 3022 /* 3023 * Now copy the flrep structure out. 3024 */ 3025 xfer = sizeof(struct flrep); 3026 cp = (void *)&fl; 3027 while (xfer > 0) { 3028 nfsm_clget; 3029 if ((bp + xfer) > be) 3030 tsiz = be - bp; 3031 else 3032 tsiz = xfer; 3033 memcpy(bp, cp, tsiz); 3034 bp += tsiz; 3035 xfer -= tsiz; 3036 if (xfer > 0) 3037 cp += tsiz; 3038 } 3039 3040 /* 3041 * ... and filehandle. 3042 */ 3043 xfer = NFSRVFH_SIZE(&nnsfh); 3044 cp = NFSRVFH_DATA(&nnsfh); 3045 while (xfer > 0) { 3046 nfsm_clget; 3047 if ((bp + xfer) > be) 3048 tsiz = be - bp; 3049 else 3050 tsiz = xfer; 3051 memcpy(bp, cp, tsiz); 3052 bp += tsiz; 3053 xfer -= tsiz; 3054 if (xfer > 0) 3055 cp += tsiz; 3056 } 3057 } 3058 invalid: 3059 cpos += dp->d_reclen; 3060 dp = (struct dirent *)cpos; 3061 cookiep++; 3062 ncookies--; 3063 } 3064 vrele(vp); 3065 nfsm_clget; 3066 *tl = nfs_false; 3067 bp += NFSX_UNSIGNED; 3068 nfsm_clget; 3069 if (eofflag) 3070 *tl = nfs_true; 3071 else 3072 *tl = nfs_false; 3073 bp += NFSX_UNSIGNED; 3074 if (mp != mb) { 3075 if (bp < be) 3076 mp->m_len = bp - mtod(mp, char *); 3077 } else 3078 mp->m_len += bp - bpos; 3079 free((void *)cookies, M_TEMP); 3080 free((void *)rbuf, M_TEMP); 3081 nfsm_srvdone; 3082 } 3083 3084 /* 3085 * nfs commit service 3086 */ 3087 int 3088 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3089 { 3090 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3091 struct mbuf *nam = nfsd->nd_nam; 3092 char *dpos = nfsd->nd_dpos; 3093 kauth_cred_t cred = nfsd->nd_cr; 3094 struct vattr bfor, aft; 3095 struct vnode *vp; 3096 nfsrvfh_t nsfh; 3097 u_int32_t *tl; 3098 int32_t t1; 3099 char *bpos; 3100 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3101 uint32_t cnt; 3102 char *cp2; 3103 struct mbuf *mb, *mreq; 3104 u_quad_t frev, off, end; 3105 3106 nfsm_srvmtofh(&nsfh); 3107 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3108 3109 off = fxdr_hyper(tl); 3110 tl += 2; 3111 cnt = fxdr_unsigned(uint32_t, *tl); 3112 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3113 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3114 if (error) { 3115 nfsm_reply(2 * NFSX_UNSIGNED); 3116 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3117 return (0); 3118 } 3119 for_ret = VOP_GETATTR(vp, &bfor, cred); 3120 end = (cnt > 0) ? off + cnt : vp->v_size; 3121 if (end < off || end > vp->v_size) 3122 end = vp->v_size; 3123 if (off < vp->v_size) 3124 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end); 3125 /* else error == 0, from nfsrv_fhtovp() */ 3126 aft_ret = VOP_GETATTR(vp, &aft, cred); 3127 vput(vp); 3128 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3129 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3130 if (!error) { 3131 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3132 *tl++ = txdr_unsigned(boottime.tv_sec); 3133 *tl = txdr_unsigned(boottime.tv_nsec / 1000); 3134 } else { 3135 return (0); 3136 } 3137 nfsm_srvdone; 3138 } 3139 3140 /* 3141 * nfs statfs service 3142 */ 3143 int 3144 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3145 { 3146 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3147 struct mbuf *nam = nfsd->nd_nam; 3148 char *dpos = nfsd->nd_dpos; 3149 kauth_cred_t cred = nfsd->nd_cr; 3150 struct statvfs *sf = NULL; 3151 struct nfs_statfs *sfp; 3152 u_int32_t *tl; 3153 int32_t t1; 3154 char *bpos; 3155 int error = 0, rdonly, cache = 0, getret = 1; 3156 int v3 = (nfsd->nd_flag & ND_NFSV3); 3157 char *cp2; 3158 struct mbuf *mb, *mreq; 3159 struct vnode *vp; 3160 struct vattr at; 3161 nfsrvfh_t nsfh; 3162 u_quad_t frev, tval; 3163 3164 nfsm_srvmtofh(&nsfh); 3165 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3166 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3167 if (error) { 3168 nfsm_reply(NFSX_UNSIGNED); 3169 nfsm_srvpostop_attr(getret, &at); 3170 return (0); 3171 } 3172 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK); 3173 error = VFS_STATVFS(vp->v_mount, sf); 3174 getret = VOP_GETATTR(vp, &at, cred); 3175 vput(vp); 3176 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3177 if (v3) 3178 nfsm_srvpostop_attr(getret, &at); 3179 if (error) { 3180 free(sf, M_TEMP); 3181 return (0); 3182 } 3183 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3184 if (v3) { 3185 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3186 txdr_hyper(tval, &sfp->sf_tbytes); 3187 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3188 txdr_hyper(tval, &sfp->sf_fbytes); 3189 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3190 txdr_hyper(tval, &sfp->sf_abytes); 3191 tval = (u_quad_t)sf->f_files; 3192 txdr_hyper(tval, &sfp->sf_tfiles); 3193 tval = (u_quad_t)sf->f_ffree; 3194 txdr_hyper(tval, &sfp->sf_ffiles); 3195 txdr_hyper(tval, &sfp->sf_afiles); 3196 sfp->sf_invarsec = 0; 3197 } else { 3198 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3199 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3200 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3201 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3202 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3203 } 3204 nfsmout: 3205 if (sf) 3206 free(sf, M_TEMP); 3207 return error; 3208 } 3209 3210 /* 3211 * nfs fsinfo service 3212 */ 3213 int 3214 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3215 { 3216 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3217 struct mbuf *nam = nfsd->nd_nam; 3218 char *dpos = nfsd->nd_dpos; 3219 kauth_cred_t cred = nfsd->nd_cr; 3220 u_int32_t *tl; 3221 struct nfsv3_fsinfo *sip; 3222 int32_t t1; 3223 char *bpos; 3224 int error = 0, rdonly, cache = 0, getret = 1; 3225 uint32_t maxdata; 3226 char *cp2; 3227 struct mbuf *mb, *mreq; 3228 struct vnode *vp; 3229 struct vattr at; 3230 nfsrvfh_t nsfh; 3231 u_quad_t frev, maxfsize; 3232 struct statvfs *sb; 3233 3234 nfsm_srvmtofh(&nsfh); 3235 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3236 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3237 if (error) { 3238 nfsm_reply(NFSX_UNSIGNED); 3239 nfsm_srvpostop_attr(getret, &at); 3240 return (0); 3241 } 3242 3243 /* XXX Try to make a guess on the max file size. */ 3244 sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK); 3245 VFS_STATVFS(vp->v_mount, sb); 3246 maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1; 3247 free(sb, M_TEMP); 3248 3249 getret = VOP_GETATTR(vp, &at, cred); 3250 vput(vp); 3251 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3252 nfsm_srvpostop_attr(getret, &at); 3253 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3254 3255 /* 3256 * XXX 3257 * There should be file system VFS OP(s) to get this information. 3258 * For now, assume ufs. 3259 */ 3260 if (slp->ns_so->so_type == SOCK_DGRAM) 3261 maxdata = NFS_MAXDGRAMDATA; 3262 else 3263 maxdata = NFS_MAXDATA; 3264 sip->fs_rtmax = txdr_unsigned(maxdata); 3265 sip->fs_rtpref = txdr_unsigned(maxdata); 3266 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3267 sip->fs_wtmax = txdr_unsigned(maxdata); 3268 sip->fs_wtpref = txdr_unsigned(maxdata); 3269 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3270 sip->fs_dtpref = txdr_unsigned(maxdata); 3271 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3272 sip->fs_timedelta.nfsv3_sec = 0; 3273 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3274 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3275 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3276 NFSV3FSINFO_CANSETTIME); 3277 nfsm_srvdone; 3278 } 3279 3280 /* 3281 * nfs pathconf service 3282 */ 3283 int 3284 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq) 3285 { 3286 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3287 struct mbuf *nam = nfsd->nd_nam; 3288 char *dpos = nfsd->nd_dpos; 3289 kauth_cred_t cred = nfsd->nd_cr; 3290 u_int32_t *tl; 3291 struct nfsv3_pathconf *pc; 3292 int32_t t1; 3293 char *bpos; 3294 int error = 0, rdonly, cache = 0, getret = 1; 3295 register_t linkmax, namemax, chownres, notrunc; 3296 char *cp2; 3297 struct mbuf *mb, *mreq; 3298 struct vnode *vp; 3299 struct vattr at; 3300 nfsrvfh_t nsfh; 3301 u_quad_t frev; 3302 3303 nfsm_srvmtofh(&nsfh); 3304 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3305 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3306 if (error) { 3307 nfsm_reply(NFSX_UNSIGNED); 3308 nfsm_srvpostop_attr(getret, &at); 3309 return (0); 3310 } 3311 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3312 if (!error) 3313 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3314 if (!error) 3315 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3316 if (!error) 3317 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3318 getret = VOP_GETATTR(vp, &at, cred); 3319 vput(vp); 3320 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3321 nfsm_srvpostop_attr(getret, &at); 3322 if (error) 3323 return (0); 3324 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3325 3326 pc->pc_linkmax = txdr_unsigned(linkmax); 3327 pc->pc_namemax = txdr_unsigned(namemax); 3328 pc->pc_notrunc = txdr_unsigned(notrunc); 3329 pc->pc_chownrestricted = txdr_unsigned(chownres); 3330 3331 /* 3332 * These should probably be supported by VOP_PATHCONF(), but 3333 * until msdosfs is exportable (why would you want to?), the 3334 * Unix defaults should be ok. 3335 */ 3336 pc->pc_caseinsensitive = nfs_false; 3337 pc->pc_casepreserving = nfs_true; 3338 nfsm_srvdone; 3339 } 3340 3341 /* 3342 * Null operation, used by clients to ping server 3343 */ 3344 /* ARGSUSED */ 3345 int 3346 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3347 struct lwp *lwp, struct mbuf **mrq) 3348 { 3349 struct mbuf *mrep = nfsd->nd_mrep; 3350 char *bpos; 3351 int error = NFSERR_RETVOID, cache = 0; 3352 struct mbuf *mb, *mreq; 3353 u_quad_t frev; 3354 3355 nfsm_reply(0); 3356 return (0); 3357 } 3358 3359 /* 3360 * No operation, used for obsolete procedures 3361 */ 3362 /* ARGSUSED */ 3363 int 3364 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3365 struct lwp *lwp, struct mbuf **mrq) 3366 { 3367 struct mbuf *mrep = nfsd->nd_mrep; 3368 char *bpos; 3369 int error, cache = 0; 3370 struct mbuf *mb, *mreq; 3371 u_quad_t frev; 3372 3373 if (nfsd->nd_repstat) 3374 error = nfsd->nd_repstat; 3375 else 3376 error = EPROCUNAVAIL; 3377 nfsm_reply(0); 3378 return (0); 3379 } 3380 3381 /* 3382 * Perform access checking for vnodes obtained from file handles that would 3383 * refer to files already opened by a Unix client. You cannot just use 3384 * vn_writechk() and VOP_ACCESS() for two reasons. 3385 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3386 * 2 - The owner is to be given access irrespective of mode bits for some 3387 * operations, so that processes that chmod after opening a file don't 3388 * break. I don't like this because it opens a security hole, but since 3389 * the nfs server opens a security hole the size of a barn door anyhow, 3390 * what the heck. 3391 * 3392 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3393 * will return EPERM instead of EACCESS. EPERM is always an error. 3394 */ 3395 int 3396 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override) 3397 { 3398 struct vattr vattr; 3399 int error; 3400 if (flags & VWRITE) { 3401 /* Just vn_writechk() changed to check rdonly */ 3402 /* 3403 * Disallow write attempts on read-only file systems; 3404 * unless the file is a socket or a block or character 3405 * device resident on the file system. 3406 */ 3407 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3408 switch (vp->v_type) { 3409 case VREG: 3410 case VDIR: 3411 case VLNK: 3412 return (EROFS); 3413 default: 3414 break; 3415 } 3416 } 3417 3418 /* 3419 * If the vnode is in use as a process's text, 3420 * we can't allow writing. 3421 */ 3422 if (vp->v_iflag & VI_TEXT) 3423 return (ETXTBSY); 3424 } 3425 error = VOP_GETATTR(vp, &vattr, cred); 3426 if (error) 3427 return (error); 3428 error = VOP_ACCESS(vp, flags, cred); 3429 /* 3430 * Allow certain operations for the owner (reads and writes 3431 * on files that are already open). 3432 */ 3433 if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid) 3434 error = 0; 3435 return error; 3436 } 3437