1 /* $NetBSD: nfs_serv.c,v 1.99 2005/12/11 12:25:16 christos 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 1.99 2005/12/11 12:25:16 christos 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 74 #include <uvm/uvm.h> 75 76 #include <nfs/nfsproto.h> 77 #include <nfs/rpcv2.h> 78 #include <nfs/nfs.h> 79 #include <nfs/xdr_subs.h> 80 #include <nfs/nfsm_subs.h> 81 #include <nfs/nqnfs.h> 82 #include <nfs/nfs_var.h> 83 84 /* Global vars */ 85 extern u_int32_t nfs_xdrneg1; 86 extern u_int32_t nfs_false, nfs_true; 87 extern enum vtype nv3tov_type[8]; 88 extern struct nfsstats nfsstats; 89 extern const nfstype nfsv2_type[9]; 90 extern const nfstype nfsv3_type[9]; 91 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000; 92 int nfsd_use_loan = 1; /* use page-loan for READ OP */ 93 94 /* 95 * nfs v3 access service 96 */ 97 int 98 nfsrv3_access(nfsd, slp, lwp, mrq) 99 struct nfsrv_descript *nfsd; 100 struct nfssvc_sock *slp; 101 struct lwp *lwp; 102 struct mbuf **mrq; 103 { 104 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 105 struct mbuf *nam = nfsd->nd_nam; 106 caddr_t dpos = nfsd->nd_dpos; 107 struct ucred *cred = &nfsd->nd_cr; 108 struct vnode *vp; 109 nfsfh_t nfh; 110 fhandle_t *fhp; 111 u_int32_t *tl; 112 int32_t t1; 113 caddr_t bpos; 114 int error = 0, rdonly, cache = 0, getret; 115 char *cp2; 116 struct mbuf *mb, *mreq; 117 struct vattr va; 118 u_long inmode, testmode, outmode; 119 u_quad_t frev; 120 121 fhp = &nfh.fh_generic; 122 nfsm_srvmtofh(fhp); 123 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 124 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, 125 (nfsd->nd_flag & ND_KERBAUTH), FALSE); 126 if (error) { 127 nfsm_reply(NFSX_UNSIGNED); 128 nfsm_srvpostop_attr(1, (struct vattr *)0); 129 return (0); 130 } 131 inmode = fxdr_unsigned(u_int32_t, *tl); 132 outmode = 0; 133 if ((inmode & NFSV3ACCESS_READ) && 134 nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0) 135 outmode |= NFSV3ACCESS_READ; 136 if (vp->v_type != VDIR) { 137 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 138 if (testmode && 139 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 140 outmode |= testmode; 141 if ((inmode & NFSV3ACCESS_EXECUTE) && 142 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 143 outmode |= NFSV3ACCESS_EXECUTE; 144 } else { 145 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 146 NFSV3ACCESS_DELETE); 147 if (testmode && 148 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 149 outmode |= testmode; 150 if ((inmode & NFSV3ACCESS_LOOKUP) && 151 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 152 outmode |= NFSV3ACCESS_LOOKUP; 153 } 154 getret = VOP_GETATTR(vp, &va, cred, lwp); 155 vput(vp); 156 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 157 nfsm_srvpostop_attr(getret, &va); 158 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 159 *tl = txdr_unsigned(outmode); 160 nfsm_srvdone; 161 } 162 163 /* 164 * nfs getattr service 165 */ 166 int 167 nfsrv_getattr(nfsd, slp, lwp, mrq) 168 struct nfsrv_descript *nfsd; 169 struct nfssvc_sock *slp; 170 struct lwp *lwp; 171 struct mbuf **mrq; 172 { 173 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 174 struct mbuf *nam = nfsd->nd_nam; 175 caddr_t dpos = nfsd->nd_dpos; 176 struct ucred *cred = &nfsd->nd_cr; 177 struct nfs_fattr *fp; 178 struct vattr va; 179 struct vnode *vp; 180 nfsfh_t nfh; 181 fhandle_t *fhp; 182 u_int32_t *tl; 183 int32_t t1; 184 caddr_t bpos; 185 int error = 0, rdonly, cache; 186 char *cp2; 187 struct mbuf *mb, *mreq; 188 u_quad_t frev; 189 190 fhp = &nfh.fh_generic; 191 nfsm_srvmtofh(fhp); 192 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, 193 (nfsd->nd_flag & ND_KERBAUTH), FALSE); 194 if (error) { 195 nfsm_reply(0); 196 return (0); 197 } 198 nqsrv_getl(vp, ND_READ); 199 error = VOP_GETATTR(vp, &va, cred, lwp); 200 vput(vp); 201 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 202 if (error) 203 return (0); 204 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 205 nfsm_srvfillattr(&va, fp); 206 nfsm_srvdone; 207 } 208 209 /* 210 * nfs setattr service 211 */ 212 int 213 nfsrv_setattr(nfsd, slp, lwp, mrq) 214 struct nfsrv_descript *nfsd; 215 struct nfssvc_sock *slp; 216 struct lwp *lwp; 217 struct mbuf **mrq; 218 { 219 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 220 struct mbuf *nam = nfsd->nd_nam; 221 caddr_t dpos = nfsd->nd_dpos; 222 struct ucred *cred = &nfsd->nd_cr; 223 struct vattr va, preat; 224 struct nfsv2_sattr *sp; 225 struct nfs_fattr *fp; 226 struct vnode *vp; 227 nfsfh_t nfh; 228 fhandle_t *fhp; 229 u_int32_t *tl; 230 int32_t t1; 231 caddr_t bpos; 232 int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1; 233 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0; 234 char *cp2; 235 struct mbuf *mb, *mreq; 236 u_quad_t frev; 237 struct timespec guard; 238 struct mount *mp = NULL; 239 240 fhp = &nfh.fh_generic; 241 nfsm_srvmtofh(fhp); 242 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 243 return (ESTALE); 244 vn_start_write(NULL, &mp, V_WAIT); 245 VATTR_NULL(&va); 246 if (v3) { 247 nfsm_srvsattr(&va); 248 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 249 gcheck = fxdr_unsigned(int, *tl); 250 if (gcheck) { 251 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 252 fxdr_nfsv3time(tl, &guard); 253 } 254 } else { 255 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 256 /* 257 * Nah nah nah nah na nah 258 * There is a bug in the Sun client that puts 0xffff in the mode 259 * field of sattr when it should put in 0xffffffff. The u_short 260 * doesn't sign extend. 261 * --> check the low order 2 bytes for 0xffff 262 */ 263 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 264 va.va_mode = nfstov_mode(sp->sa_mode); 265 if (sp->sa_uid != nfs_xdrneg1) 266 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 267 if (sp->sa_gid != nfs_xdrneg1) 268 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 269 if (sp->sa_size != nfs_xdrneg1) 270 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 271 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 272 #ifdef notyet 273 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 274 #else 275 va.va_atime.tv_sec = 276 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 277 va.va_atime.tv_nsec = 0; 278 #endif 279 } 280 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 281 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 282 283 } 284 285 /* 286 * Now that we have all the fields, lets do it. 287 */ 288 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, 289 (nfsd->nd_flag & ND_KERBAUTH), FALSE); 290 if (error) { 291 nfsm_reply(2 * NFSX_UNSIGNED); 292 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 293 vn_finished_write(mp, 0); 294 return (0); 295 } 296 nqsrv_getl(vp, ND_WRITE); 297 if (v3) { 298 error = preat_ret = VOP_GETATTR(vp, &preat, cred, lwp); 299 if (!error && gcheck && 300 (preat.va_ctime.tv_sec != guard.tv_sec || 301 preat.va_ctime.tv_nsec != guard.tv_nsec)) 302 error = NFSERR_NOT_SYNC; 303 if (error) { 304 vput(vp); 305 nfsm_reply(NFSX_WCCDATA(v3)); 306 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 307 vn_finished_write(mp, 0); 308 return (0); 309 } 310 } 311 312 /* 313 * If the size is being changed write acces is required, otherwise 314 * just check for a read only file system. 315 */ 316 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 317 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 318 error = EROFS; 319 goto out; 320 } 321 } else { 322 if (vp->v_type == VDIR) { 323 error = EISDIR; 324 goto out; 325 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 326 lwp, 0)) != 0) 327 goto out; 328 } 329 error = VOP_SETATTR(vp, &va, cred, lwp); 330 postat_ret = VOP_GETATTR(vp, &va, cred, lwp); 331 if (!error) 332 error = postat_ret; 333 out: 334 vput(vp); 335 vn_finished_write(mp, 0); 336 nfsm_reply(NFSX_WCCORFATTR(v3)); 337 if (v3) { 338 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 339 return (0); 340 } else { 341 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 342 nfsm_srvfillattr(&va, fp); 343 } 344 nfsm_srvdone; 345 } 346 347 /* 348 * nfs lookup rpc 349 */ 350 int 351 nfsrv_lookup(nfsd, slp, lwp, mrq) 352 struct nfsrv_descript *nfsd; 353 struct nfssvc_sock *slp; 354 struct lwp *lwp; 355 struct mbuf **mrq; 356 { 357 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 358 struct mbuf *nam = nfsd->nd_nam; 359 caddr_t dpos = nfsd->nd_dpos; 360 struct ucred *cred = &nfsd->nd_cr; 361 struct nfs_fattr *fp; 362 struct nameidata nd, ind, *ndp = &nd; 363 struct vnode *vp, *dirp; 364 nfsfh_t nfh; 365 fhandle_t *fhp; 366 caddr_t cp; 367 u_int32_t *tl; 368 int32_t t1; 369 caddr_t bpos; 370 int error = 0, cache, dirattr_ret = 1; 371 uint32_t len; 372 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 373 char *cp2; 374 struct mbuf *mb, *mreq; 375 struct vattr va, dirattr; 376 u_quad_t frev; 377 378 fhp = &nfh.fh_generic; 379 nfsm_srvmtofh(fhp); 380 nfsm_srvnamesiz(len); 381 382 pubflag = nfs_ispublicfh(fhp); 383 384 nd.ni_cnd.cn_cred = cred; 385 nd.ni_cnd.cn_nameiop = LOOKUP; 386 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART; 387 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 388 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 389 390 if (!error && pubflag) { 391 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) { 392 /* 393 * Setup call to lookup() to see if we can find 394 * the index file. Arguably, this doesn't belong 395 * in a kernel.. Ugh. 396 */ 397 ind = nd; 398 VOP_UNLOCK(nd.ni_vp, 0); 399 ind.ni_pathlen = strlen(nfs_pub.np_index); 400 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf = 401 nfs_pub.np_index; 402 ind.ni_startdir = nd.ni_vp; 403 VREF(ind.ni_startdir); 404 error = lookup(&ind); 405 if (!error) { 406 /* 407 * Found an index file. Get rid of 408 * the old references. 409 */ 410 if (dirp) 411 vrele(dirp); 412 dirp = nd.ni_vp; 413 vrele(nd.ni_startdir); 414 ndp = &ind; 415 } else 416 error = 0; 417 } 418 /* 419 * If the public filehandle was used, check that this lookup 420 * didn't result in a filehandle outside the publicly exported 421 * filesystem. 422 */ 423 424 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 425 vput(nd.ni_vp); 426 error = EPERM; 427 } 428 } 429 430 if (dirp) { 431 if (v3) 432 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred, lwp); 433 vrele(dirp); 434 } 435 436 if (error) { 437 nfsm_reply(NFSX_POSTOPATTR(v3)); 438 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 439 return (0); 440 } 441 442 nqsrv_getl(ndp->ni_startdir, ND_READ); 443 vrele(ndp->ni_startdir); 444 PNBUF_PUT(nd.ni_cnd.cn_pnbuf); 445 vp = ndp->ni_vp; 446 memset((caddr_t)fhp, 0, sizeof(nfh)); 447 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 448 error = VFS_VPTOFH(vp, &fhp->fh_fid); 449 if (!error) 450 error = VOP_GETATTR(vp, &va, cred, lwp); 451 vput(vp); 452 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3)); 453 if (error) { 454 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 455 return (0); 456 } 457 nfsm_srvfhtom(fhp, v3); 458 if (v3) { 459 nfsm_srvpostop_attr(0, &va); 460 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 461 } else { 462 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 463 nfsm_srvfillattr(&va, fp); 464 } 465 nfsm_srvdone; 466 } 467 468 /* 469 * nfs readlink service 470 */ 471 int 472 nfsrv_readlink(nfsd, slp, lwp, mrq) 473 struct nfsrv_descript *nfsd; 474 struct nfssvc_sock *slp; 475 struct lwp *lwp; 476 struct mbuf **mrq; 477 { 478 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 479 struct mbuf *nam = nfsd->nd_nam; 480 caddr_t dpos = nfsd->nd_dpos; 481 struct ucred *cred = &nfsd->nd_cr; 482 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 483 struct iovec *ivp = iv; 484 struct mbuf *mp; 485 u_int32_t *tl; 486 int32_t t1; 487 caddr_t bpos; 488 int error = 0, rdonly, cache, i, padlen, getret; 489 uint32_t len; 490 int v3 = (nfsd->nd_flag & ND_NFSV3); 491 char *cp2; 492 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 493 struct vnode *vp; 494 struct vattr attr; 495 nfsfh_t nfh; 496 fhandle_t *fhp; 497 struct uio io, *uiop = &io; 498 u_quad_t frev; 499 500 fhp = &nfh.fh_generic; 501 nfsm_srvmtofh(fhp); 502 len = 0; 503 i = 0; 504 while (len < NFS_MAXPATHLEN) { 505 mp = m_get(M_WAIT, MT_DATA); 506 MCLAIM(mp, &nfs_mowner); 507 m_clget(mp, M_WAIT); 508 mp->m_len = NFSMSIZ(mp); 509 if (len == 0) 510 mp3 = mp2 = mp; 511 else { 512 mp2->m_next = mp; 513 mp2 = mp; 514 } 515 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 516 mp->m_len = NFS_MAXPATHLEN-len; 517 len = NFS_MAXPATHLEN; 518 } else 519 len += mp->m_len; 520 ivp->iov_base = mtod(mp, caddr_t); 521 ivp->iov_len = mp->m_len; 522 i++; 523 ivp++; 524 } 525 uiop->uio_iov = iv; 526 uiop->uio_iovcnt = i; 527 uiop->uio_offset = 0; 528 uiop->uio_resid = len; 529 uiop->uio_rw = UIO_READ; 530 uiop->uio_segflg = UIO_SYSSPACE; 531 uiop->uio_lwp = NULL; 532 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 533 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 534 if (error) { 535 m_freem(mp3); 536 nfsm_reply(2 * NFSX_UNSIGNED); 537 nfsm_srvpostop_attr(1, (struct vattr *)0); 538 return (0); 539 } 540 if (vp->v_type != VLNK) { 541 if (v3) 542 error = EINVAL; 543 else 544 error = ENXIO; 545 goto out; 546 } 547 nqsrv_getl(vp, ND_READ); 548 error = VOP_READLINK(vp, uiop, cred); 549 out: 550 getret = VOP_GETATTR(vp, &attr, cred, lwp); 551 vput(vp); 552 if (error) 553 m_freem(mp3); 554 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 555 if (v3) { 556 nfsm_srvpostop_attr(getret, &attr); 557 if (error) 558 return (0); 559 } 560 len -= uiop->uio_resid; 561 padlen = nfsm_padlen(len); 562 if (uiop->uio_resid || padlen) 563 nfs_zeropad(mp3, uiop->uio_resid, padlen); 564 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 565 *tl = txdr_unsigned(len); 566 mb->m_next = mp3; 567 nfsm_srvdone; 568 } 569 570 /* 571 * nfs read service 572 */ 573 int 574 nfsrv_read(nfsd, slp, lwp, mrq) 575 struct nfsrv_descript *nfsd; 576 struct nfssvc_sock *slp; 577 struct lwp *lwp; 578 struct mbuf **mrq; 579 { 580 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 581 struct mbuf *nam = nfsd->nd_nam; 582 caddr_t dpos = nfsd->nd_dpos; 583 struct ucred *cred = &nfsd->nd_cr; 584 struct mbuf *m; 585 struct nfs_fattr *fp; 586 u_int32_t *tl; 587 int32_t t1; 588 int i; 589 caddr_t bpos; 590 int error = 0, rdonly, cache, getret; 591 int v3 = (nfsd->nd_flag & ND_NFSV3); 592 uint32_t reqlen, len, cnt, left; 593 int padlen; 594 char *cp2; 595 struct mbuf *mb, *mreq; 596 struct vnode *vp; 597 nfsfh_t nfh; 598 fhandle_t *fhp; 599 struct uio io, *uiop = &io; 600 struct vattr va; 601 off_t off; 602 u_quad_t frev; 603 604 fhp = &nfh.fh_generic; 605 nfsm_srvmtofh(fhp); 606 if (v3) { 607 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 608 off = fxdr_hyper(tl); 609 } else { 610 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 611 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 612 } 613 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 614 reqlen = fxdr_unsigned(uint32_t, *tl); 615 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 616 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 617 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 618 if (error) { 619 nfsm_reply(2 * NFSX_UNSIGNED); 620 nfsm_srvpostop_attr(1, (struct vattr *)0); 621 return (0); 622 } 623 if (vp->v_type != VREG) { 624 if (v3) 625 error = EINVAL; 626 else 627 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 628 } 629 if (!error) { 630 nqsrv_getl(vp, ND_READ); 631 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 632 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 633 } 634 getret = VOP_GETATTR(vp, &va, cred, lwp); 635 if (!error) 636 error = getret; 637 if (error) { 638 vput(vp); 639 nfsm_reply(NFSX_POSTOPATTR(v3)); 640 nfsm_srvpostop_attr(getret, &va); 641 return (0); 642 } 643 if (off >= va.va_size) 644 cnt = 0; 645 else if ((off + reqlen) > va.va_size) 646 cnt = va.va_size - off; 647 else 648 cnt = reqlen; 649 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 650 if (v3) { 651 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 652 *tl++ = nfs_true; 653 fp = (struct nfs_fattr *)tl; 654 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 655 } else { 656 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 657 fp = (struct nfs_fattr *)tl; 658 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 659 } 660 len = left = cnt; 661 if (cnt > 0) { 662 if (nfsd_use_loan) { 663 struct vm_page **pgpp; 664 voff_t pgoff = trunc_page(off); 665 int npages; 666 vaddr_t lva; 667 668 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 669 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 670 671 /* allocate kva for mbuf data */ 672 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 673 if (lva == 0) { 674 /* fall back to VOP_READ */ 675 goto loan_fail; 676 } 677 678 /* allocate mbuf */ 679 m = m_get(M_WAIT, MT_DATA); 680 MCLAIM(m, &nfs_mowner); 681 pgpp = m->m_ext.ext_pgs; 682 683 /* loan pages */ 684 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 685 pgpp); 686 if (error) { 687 sokvafree(lva, npages << PAGE_SHIFT); 688 m_free(m); 689 if (error == EBUSY) 690 goto loan_fail; 691 goto read_error; 692 } 693 694 /* associate kva to mbuf */ 695 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 696 cnt, M_MBUF, soloanfree, slp->ns_so); 697 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 698 m->m_len = cnt; 699 700 /* map pages */ 701 for (i = 0; i < npages; i++) { 702 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 703 VM_PROT_READ); 704 lva += PAGE_SIZE; 705 } 706 707 pmap_update(pmap_kernel()); 708 709 mb->m_next = m; 710 mb = m; 711 error = 0; 712 uiop->uio_resid = 0; 713 } else { 714 struct iovec *iv; 715 struct iovec *iv2; 716 struct mbuf *m2; 717 int siz; 718 loan_fail: 719 /* 720 * Generate the mbuf list with the uio_iov ref. to it. 721 */ 722 i = 0; 723 m = m2 = mb; 724 while (left > 0) { 725 siz = min(M_TRAILINGSPACE(m), left); 726 if (siz > 0) { 727 left -= siz; 728 i++; 729 } 730 if (left > 0) { 731 m = m_get(M_WAIT, MT_DATA); 732 MCLAIM(m, &nfs_mowner); 733 m_clget(m, M_WAIT); 734 m->m_len = 0; 735 m2->m_next = m; 736 m2 = m; 737 } 738 } 739 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 740 uiop->uio_iov = iv2 = iv; 741 m = mb; 742 left = cnt; 743 i = 0; 744 while (left > 0) { 745 if (m == NULL) 746 panic("nfsrv_read iov"); 747 siz = min(M_TRAILINGSPACE(m), left); 748 if (siz > 0) { 749 iv->iov_base = mtod(m, caddr_t) + 750 m->m_len; 751 iv->iov_len = siz; 752 m->m_len += siz; 753 left -= siz; 754 iv++; 755 i++; 756 } 757 m = m->m_next; 758 } 759 uiop->uio_iovcnt = i; 760 uiop->uio_offset = off; 761 uiop->uio_resid = cnt; 762 uiop->uio_rw = UIO_READ; 763 uiop->uio_segflg = UIO_SYSSPACE; 764 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 765 free((caddr_t)iv2, M_TEMP); 766 } 767 read_error: 768 if (error || (getret = VOP_GETATTR(vp, &va, cred, lwp)) != 0){ 769 if (!error) 770 error = getret; 771 m_freem(mreq); 772 vput(vp); 773 nfsm_reply(NFSX_POSTOPATTR(v3)); 774 nfsm_srvpostop_attr(getret, &va); 775 return (0); 776 } 777 } else { 778 uiop->uio_resid = 0; 779 } 780 vput(vp); 781 nfsm_srvfillattr(&va, fp); 782 len -= uiop->uio_resid; 783 padlen = nfsm_padlen(len); 784 if (uiop->uio_resid || padlen) 785 nfs_zeropad(mb, uiop->uio_resid, padlen); 786 if (v3) { 787 /* count */ 788 *tl++ = txdr_unsigned(len); 789 /* eof */ 790 if (off + len >= va.va_size) 791 *tl++ = nfs_true; 792 else 793 *tl++ = nfs_false; 794 } 795 *tl = txdr_unsigned(len); 796 nfsm_srvdone; 797 } 798 799 /* 800 * nfs write service 801 */ 802 int 803 nfsrv_write(nfsd, slp, lwp, mrq) 804 struct nfsrv_descript *nfsd; 805 struct nfssvc_sock *slp; 806 struct lwp *lwp; 807 struct mbuf **mrq; 808 { 809 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 810 struct mbuf *nam = nfsd->nd_nam; 811 caddr_t dpos = nfsd->nd_dpos; 812 struct ucred *cred = &nfsd->nd_cr; 813 struct iovec *ivp; 814 int i, cnt; 815 struct mbuf *mp; 816 struct nfs_fattr *fp; 817 struct iovec *iv; 818 struct vattr va, forat; 819 u_int32_t *tl; 820 int32_t t1; 821 caddr_t bpos; 822 int error = 0, rdonly, cache, len, forat_ret = 1; 823 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 824 int stable = NFSV3WRITE_FILESYNC; 825 int v3 = (nfsd->nd_flag & ND_NFSV3); 826 char *cp2; 827 struct mbuf *mb, *mreq; 828 struct vnode *vp; 829 nfsfh_t nfh; 830 fhandle_t *fhp; 831 struct uio io, *uiop = &io; 832 off_t off; 833 u_quad_t frev; 834 struct mount *mntp = NULL; 835 836 if (mrep == NULL) { 837 *mrq = NULL; 838 return (0); 839 } 840 fhp = &nfh.fh_generic; 841 nfsm_srvmtofh(fhp); 842 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 843 return (ESTALE); 844 vn_start_write(NULL, &mntp, V_WAIT); 845 if (v3) { 846 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 847 off = fxdr_hyper(tl); 848 tl += 3; 849 stable = fxdr_unsigned(int, *tl++); 850 } else { 851 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 852 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 853 tl += 2; 854 } 855 retlen = len = fxdr_unsigned(int32_t, *tl); 856 cnt = i = 0; 857 858 /* 859 * For NFS Version 2, it is not obvious what a write of zero length 860 * should do, but I might as well be consistent with Version 3, 861 * which is to return ok so long as there are no permission problems. 862 */ 863 if (len > 0) { 864 zeroing = 1; 865 mp = mrep; 866 while (mp) { 867 if (mp == md) { 868 zeroing = 0; 869 adjust = dpos - mtod(mp, caddr_t); 870 mp->m_len -= adjust; 871 if (mp->m_len > 0 && adjust > 0) 872 NFSMADV(mp, adjust); 873 } 874 if (zeroing) 875 mp->m_len = 0; 876 else if (mp->m_len > 0) { 877 i += mp->m_len; 878 if (i > len) { 879 mp->m_len -= (i - len); 880 zeroing = 1; 881 } 882 if (mp->m_len > 0) 883 cnt++; 884 } 885 mp = mp->m_next; 886 } 887 } 888 if (len > NFS_MAXDATA || len < 0 || i < len) { 889 error = EIO; 890 nfsm_reply(2 * NFSX_UNSIGNED); 891 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 892 vn_finished_write(mntp, 0); 893 return (0); 894 } 895 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 896 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 897 if (error) { 898 nfsm_reply(2 * NFSX_UNSIGNED); 899 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 900 vn_finished_write(mntp, 0); 901 return (0); 902 } 903 if (v3) 904 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 905 if (vp->v_type != VREG) { 906 if (v3) 907 error = EINVAL; 908 else 909 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 910 } 911 if (!error) { 912 nqsrv_getl(vp, ND_WRITE); 913 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 914 } 915 if (error) { 916 vput(vp); 917 nfsm_reply(NFSX_WCCDATA(v3)); 918 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 919 vn_finished_write(mntp, 0); 920 return (0); 921 } 922 923 if (len > 0) { 924 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 925 uiop->uio_iov = iv = ivp; 926 uiop->uio_iovcnt = cnt; 927 mp = mrep; 928 while (mp) { 929 if (mp->m_len > 0) { 930 ivp->iov_base = mtod(mp, caddr_t); 931 ivp->iov_len = mp->m_len; 932 ivp++; 933 } 934 mp = mp->m_next; 935 } 936 937 /* 938 * XXX 939 * The IO_METASYNC flag indicates that all metadata (and not 940 * just enough to ensure data integrity) must be written to 941 * stable storage synchronously. 942 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 943 */ 944 if (stable == NFSV3WRITE_UNSTABLE) 945 ioflags = IO_NODELOCKED; 946 else if (stable == NFSV3WRITE_DATASYNC) 947 ioflags = (IO_SYNC | IO_NODELOCKED); 948 else 949 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 950 uiop->uio_resid = len; 951 uiop->uio_rw = UIO_WRITE; 952 uiop->uio_segflg = UIO_SYSSPACE; 953 uiop->uio_lwp = NULL; 954 uiop->uio_offset = off; 955 error = VOP_WRITE(vp, uiop, ioflags, cred); 956 nfsstats.srvvop_writes++; 957 free(iv, M_TEMP); 958 } 959 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 960 vput(vp); 961 vn_finished_write(mntp, 0); 962 if (!error) 963 error = aftat_ret; 964 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 965 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 966 if (v3) { 967 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 968 if (error) 969 return (0); 970 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 971 *tl++ = txdr_unsigned(retlen); 972 if (stable == NFSV3WRITE_UNSTABLE) 973 *tl++ = txdr_unsigned(stable); 974 else 975 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 976 /* 977 * Actually, there is no need to txdr these fields, 978 * but it may make the values more human readable, 979 * for debugging purposes. 980 */ 981 *tl++ = txdr_unsigned(boottime.tv_sec); 982 *tl = txdr_unsigned(boottime.tv_usec); 983 } else { 984 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 985 nfsm_srvfillattr(&va, fp); 986 } 987 nfsm_srvdone; 988 } 989 990 /* 991 * NFS write service with write gathering support. Called when 992 * nfsrvw_procrastinate > 0. 993 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 994 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 995 * Jan. 1994. 996 */ 997 int 998 nfsrv_writegather(ndp, slp, lwp, mrq) 999 struct nfsrv_descript **ndp; 1000 struct nfssvc_sock *slp; 1001 struct lwp *lwp; 1002 struct mbuf **mrq; 1003 { 1004 struct iovec *ivp; 1005 struct mbuf *mp; 1006 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1007 struct nfs_fattr *fp; 1008 int i = 0; 1009 struct iovec *iov; 1010 struct nfsrvw_delayhash *wpp; 1011 struct ucred *cred; 1012 struct vattr va, forat; 1013 u_int32_t *tl; 1014 int32_t t1; 1015 caddr_t bpos, dpos; 1016 int error = 0, rdonly, cache, len = 0, forat_ret = 1; 1017 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing; 1018 char *cp2; 1019 struct mbuf *mb, *mreq, *mrep, *md; 1020 struct vnode *vp; 1021 struct uio io, *uiop = &io; 1022 u_quad_t frev, cur_usec; 1023 struct mount *mntp = NULL; 1024 1025 *mrq = NULL; 1026 if (*ndp) { 1027 nfsd = *ndp; 1028 *ndp = NULL; 1029 mrep = nfsd->nd_mrep; 1030 md = nfsd->nd_md; 1031 dpos = nfsd->nd_dpos; 1032 cred = &nfsd->nd_cr; 1033 v3 = (nfsd->nd_flag & ND_NFSV3); 1034 LIST_INIT(&nfsd->nd_coalesce); 1035 nfsd->nd_mreq = NULL; 1036 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1037 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1038 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1039 1040 /* 1041 * Now, get the write header.. 1042 */ 1043 nfsm_srvmtofh(&nfsd->nd_fh); 1044 if (v3) { 1045 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1046 nfsd->nd_off = fxdr_hyper(tl); 1047 tl += 3; 1048 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1049 } else { 1050 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1051 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1052 tl += 2; 1053 } 1054 len = fxdr_unsigned(int32_t, *tl); 1055 nfsd->nd_len = len; 1056 nfsd->nd_eoff = nfsd->nd_off + len; 1057 1058 /* 1059 * Trim the header out of the mbuf list and trim off any trailing 1060 * junk so that the mbuf list has only the write data. 1061 */ 1062 zeroing = 1; 1063 i = 0; 1064 mp = mrep; 1065 while (mp) { 1066 if (mp == md) { 1067 zeroing = 0; 1068 adjust = dpos - mtod(mp, caddr_t); 1069 mp->m_len -= adjust; 1070 if (mp->m_len > 0 && adjust > 0) 1071 NFSMADV(mp, adjust); 1072 } 1073 if (zeroing) 1074 mp->m_len = 0; 1075 else { 1076 i += mp->m_len; 1077 if (i > len) { 1078 mp->m_len -= (i - len); 1079 zeroing = 1; 1080 } 1081 } 1082 mp = mp->m_next; 1083 } 1084 if (len > NFS_MAXDATA || len < 0 || i < len) { 1085 nfsmout: 1086 m_freem(mrep); 1087 error = EIO; 1088 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1089 if (v3) 1090 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1091 nfsd->nd_mreq = mreq; 1092 nfsd->nd_mrep = NULL; 1093 nfsd->nd_time = 0; 1094 } 1095 1096 /* 1097 * Add this entry to the hash and time queues. 1098 */ 1099 s = splsoftclock(); 1100 owp = NULL; 1101 wp = LIST_FIRST(&slp->ns_tq); 1102 while (wp && wp->nd_time < nfsd->nd_time) { 1103 owp = wp; 1104 wp = LIST_NEXT(wp, nd_tq); 1105 } 1106 if (owp) { 1107 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1108 } else { 1109 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1110 } 1111 if (nfsd->nd_mrep) { 1112 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data); 1113 owp = NULL; 1114 wp = LIST_FIRST(wpp); 1115 while (wp && 1116 memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) { 1117 owp = wp; 1118 wp = LIST_NEXT(wp, nd_hash); 1119 } 1120 while (wp && wp->nd_off < nfsd->nd_off && 1121 !memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) { 1122 owp = wp; 1123 wp = LIST_NEXT(wp, nd_hash); 1124 } 1125 if (owp) { 1126 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1127 1128 /* 1129 * Search the hash list for overlapping entries and 1130 * coalesce. 1131 */ 1132 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1133 wp = LIST_NEXT(nfsd, nd_hash); 1134 if (NFSW_SAMECRED(owp, nfsd)) 1135 nfsrvw_coalesce(owp, nfsd); 1136 } 1137 } else { 1138 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1139 } 1140 } 1141 splx(s); 1142 } 1143 1144 /* 1145 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1146 * and generate the associated reply mbuf list(s). 1147 */ 1148 loop1: 1149 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1150 s = splsoftclock(); 1151 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1152 owp = LIST_NEXT(nfsd, nd_tq); 1153 if (nfsd->nd_time > cur_usec) 1154 break; 1155 if (nfsd->nd_mreq) 1156 continue; 1157 LIST_REMOVE(nfsd, nd_tq); 1158 LIST_REMOVE(nfsd, nd_hash); 1159 splx(s); 1160 mrep = nfsd->nd_mrep; 1161 nfsd->nd_mrep = NULL; 1162 cred = &nfsd->nd_cr; 1163 v3 = (nfsd->nd_flag & ND_NFSV3); 1164 forat_ret = aftat_ret = 1; 1165 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1166 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1167 FALSE); 1168 if (!error) { 1169 if (v3) 1170 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 1171 if (vp->v_type != VREG) { 1172 if (v3) 1173 error = EINVAL; 1174 else 1175 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1176 } 1177 } else 1178 vp = NULL; 1179 if (!error) { 1180 nqsrv_getl(vp, ND_WRITE); 1181 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1182 } 1183 1184 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1185 ioflags = IO_NODELOCKED; 1186 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1187 ioflags = (IO_SYNC | IO_NODELOCKED); 1188 else 1189 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1190 uiop->uio_rw = UIO_WRITE; 1191 uiop->uio_segflg = UIO_SYSSPACE; 1192 uiop->uio_lwp = NULL; 1193 uiop->uio_offset = nfsd->nd_off; 1194 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1195 if (uiop->uio_resid > 0) { 1196 mp = mrep; 1197 i = 0; 1198 while (mp) { 1199 if (mp->m_len > 0) 1200 i++; 1201 mp = mp->m_next; 1202 } 1203 uiop->uio_iovcnt = i; 1204 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1205 uiop->uio_iov = ivp = iov; 1206 mp = mrep; 1207 while (mp) { 1208 if (mp->m_len > 0) { 1209 ivp->iov_base = mtod(mp, caddr_t); 1210 ivp->iov_len = mp->m_len; 1211 ivp++; 1212 } 1213 mp = mp->m_next; 1214 } 1215 if (!error) { 1216 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) { 1217 VOP_UNLOCK(vp, 0); 1218 vn_start_write(NULL, &mntp, V_WAIT); 1219 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1220 } 1221 if (!error) { 1222 error = VOP_WRITE(vp, uiop, ioflags, cred); 1223 nfsstats.srvvop_writes++; 1224 vn_finished_write(mntp, 0); 1225 } 1226 } 1227 free((caddr_t)iov, M_TEMP); 1228 } 1229 m_freem(mrep); 1230 if (vp) { 1231 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 1232 vput(vp); 1233 } 1234 1235 /* 1236 * Loop around generating replies for all write rpcs that have 1237 * now been completed. 1238 */ 1239 swp = nfsd; 1240 do { 1241 if (error) { 1242 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1243 if (v3) { 1244 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1245 } 1246 } else { 1247 nfsm_writereply(NFSX_PREOPATTR(v3) + 1248 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1249 NFSX_WRITEVERF(v3), v3); 1250 if (v3) { 1251 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1252 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1253 *tl++ = txdr_unsigned(nfsd->nd_len); 1254 *tl++ = txdr_unsigned(swp->nd_stable); 1255 /* 1256 * Actually, there is no need to txdr these fields, 1257 * but it may make the values more human readable, 1258 * for debugging purposes. 1259 */ 1260 *tl++ = txdr_unsigned(boottime.tv_sec); 1261 *tl = txdr_unsigned(boottime.tv_usec); 1262 } else { 1263 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1264 nfsm_srvfillattr(&va, fp); 1265 } 1266 } 1267 nfsd->nd_mreq = mreq; 1268 if (nfsd->nd_mrep) 1269 panic("nfsrv_write: nd_mrep not free"); 1270 1271 /* 1272 * Done. Put it at the head of the timer queue so that 1273 * the final phase can return the reply. 1274 */ 1275 s = splsoftclock(); 1276 if (nfsd != swp) { 1277 nfsd->nd_time = 0; 1278 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1279 } 1280 nfsd = LIST_FIRST(&swp->nd_coalesce); 1281 if (nfsd) { 1282 LIST_REMOVE(nfsd, nd_tq); 1283 } 1284 splx(s); 1285 } while (nfsd); 1286 s = splsoftclock(); 1287 swp->nd_time = 0; 1288 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1289 splx(s); 1290 goto loop1; 1291 } 1292 splx(s); 1293 1294 /* 1295 * Search for a reply to return. 1296 */ 1297 s = splsoftclock(); 1298 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1299 if (nfsd->nd_mreq) { 1300 LIST_REMOVE(nfsd, nd_tq); 1301 *mrq = nfsd->nd_mreq; 1302 *ndp = nfsd; 1303 break; 1304 } 1305 } 1306 splx(s); 1307 return (0); 1308 } 1309 1310 /* 1311 * Coalesce the write request nfsd into owp. To do this we must: 1312 * - remove nfsd from the queues 1313 * - merge nfsd->nd_mrep into owp->nd_mrep 1314 * - update the nd_eoff and nd_stable for owp 1315 * - put nfsd on owp's nd_coalesce list 1316 * NB: Must be called at splsoftclock(). 1317 */ 1318 void 1319 nfsrvw_coalesce(owp, nfsd) 1320 struct nfsrv_descript *owp; 1321 struct nfsrv_descript *nfsd; 1322 { 1323 int overlap; 1324 struct mbuf *mp; 1325 struct nfsrv_descript *m; 1326 1327 LIST_REMOVE(nfsd, nd_hash); 1328 LIST_REMOVE(nfsd, nd_tq); 1329 if (owp->nd_eoff < nfsd->nd_eoff) { 1330 overlap = owp->nd_eoff - nfsd->nd_off; 1331 if (overlap < 0) 1332 panic("nfsrv_coalesce: bad off"); 1333 if (overlap > 0) 1334 m_adj(nfsd->nd_mrep, overlap); 1335 mp = owp->nd_mrep; 1336 while (mp->m_next) 1337 mp = mp->m_next; 1338 mp->m_next = nfsd->nd_mrep; 1339 owp->nd_eoff = nfsd->nd_eoff; 1340 } else 1341 m_freem(nfsd->nd_mrep); 1342 nfsd->nd_mrep = NULL; 1343 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1344 owp->nd_stable = NFSV3WRITE_FILESYNC; 1345 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1346 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1347 owp->nd_stable = NFSV3WRITE_DATASYNC; 1348 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1349 /* 1350 * nfsd might hold coalesce elements! Move them to owp. 1351 * Otherwise, requests may be lost and clients will be stuck. 1352 */ 1353 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1354 LIST_REMOVE(m, nd_tq); 1355 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1356 } 1357 } 1358 1359 /* 1360 * nfs create service 1361 * now does a truncate to 0 length via. setattr if it already exists 1362 */ 1363 int 1364 nfsrv_create(nfsd, slp, lwp, mrq) 1365 struct nfsrv_descript *nfsd; 1366 struct nfssvc_sock *slp; 1367 struct lwp *lwp; 1368 struct mbuf **mrq; 1369 { 1370 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1371 struct mbuf *nam = nfsd->nd_nam; 1372 caddr_t dpos = nfsd->nd_dpos; 1373 struct ucred *cred = &nfsd->nd_cr; 1374 struct nfs_fattr *fp; 1375 struct vattr va, dirfor, diraft; 1376 struct nfsv2_sattr *sp; 1377 u_int32_t *tl; 1378 struct nameidata nd; 1379 caddr_t cp; 1380 int32_t t1; 1381 caddr_t bpos; 1382 int error = 0, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1383 int rdev = 0; 1384 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1385 char *cp2; 1386 struct mbuf *mb, *mreq; 1387 struct vnode *vp = NULL, *dirp = NULL; 1388 nfsfh_t nfh; 1389 fhandle_t *fhp; 1390 u_quad_t frev, tempsize; 1391 u_char cverf[NFSX_V3CREATEVERF]; 1392 struct mount *mp = NULL; 1393 1394 nd.ni_cnd.cn_nameiop = 0; 1395 fhp = &nfh.fh_generic; 1396 nfsm_srvmtofh(fhp); 1397 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1398 return (ESTALE); 1399 vn_start_write(NULL, &mp, V_WAIT); 1400 nfsm_srvnamesiz(len); 1401 nd.ni_cnd.cn_cred = cred; 1402 nd.ni_cnd.cn_nameiop = CREATE; 1403 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1404 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1405 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1406 if (dirp) { 1407 if (v3) 1408 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1409 else { 1410 vrele(dirp); 1411 dirp = (struct vnode *)0; 1412 } 1413 } 1414 if (error) { 1415 nfsm_reply(NFSX_WCCDATA(v3)); 1416 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1417 if (dirp) 1418 vrele(dirp); 1419 vn_finished_write(mp, 0); 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, caddr_t, 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, cred, 1485 lwp); 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 = suser(cred, (u_short *)0))) { 1494 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1495 vput(nd.ni_dvp); 1496 nfsm_reply(0); 1497 vn_finished_write(mp, 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 vrele(nd.ni_dvp); 1509 vput(nd.ni_vp); 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 vp = nd.ni_vp; 1522 if (nd.ni_dvp == vp) 1523 vrele(nd.ni_dvp); 1524 else 1525 vput(nd.ni_dvp); 1526 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 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, lwp); 1536 } 1537 } 1538 if (error) 1539 vput(vp); 1540 } 1541 if (!error) { 1542 memset((caddr_t)fhp, 0, sizeof(nfh)); 1543 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1544 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1545 if (!error) 1546 error = VOP_GETATTR(vp, &va, cred, lwp); 1547 vput(vp); 1548 } 1549 if (v3) { 1550 if (exclusive_flag && !error) { 1551 /* 1552 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1553 */ 1554 char oldverf[NFSX_V3CREATEVERF]; 1555 1556 txdr_nfsv3time(&va.va_atime, oldverf); 1557 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1558 error = EEXIST; 1559 } 1560 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1561 vrele(dirp); 1562 } 1563 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1564 if (v3) { 1565 if (!error) { 1566 nfsm_srvpostop_fh(fhp); 1567 nfsm_srvpostop_attr(0, &va); 1568 } 1569 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1570 } else { 1571 nfsm_srvfhtom(fhp, v3); 1572 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1573 nfsm_srvfillattr(&va, fp); 1574 } 1575 vn_finished_write(mp, 0); 1576 return (0); 1577 nfsmout: 1578 if (dirp) 1579 vrele(dirp); 1580 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1581 if (nd.ni_dvp == nd.ni_vp) 1582 vrele(nd.ni_dvp); 1583 else 1584 vput(nd.ni_dvp); 1585 if (nd.ni_vp) 1586 vput(nd.ni_vp); 1587 vn_finished_write(mp, 0); 1588 return (error); 1589 } 1590 1591 /* 1592 * nfs v3 mknod service 1593 */ 1594 int 1595 nfsrv_mknod(nfsd, slp, lwp, mrq) 1596 struct nfsrv_descript *nfsd; 1597 struct nfssvc_sock *slp; 1598 struct lwp *lwp; 1599 struct mbuf **mrq; 1600 { 1601 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1602 struct mbuf *nam = nfsd->nd_nam; 1603 caddr_t dpos = nfsd->nd_dpos; 1604 struct ucred *cred = &nfsd->nd_cr; 1605 struct vattr va, dirfor, diraft; 1606 u_int32_t *tl; 1607 struct nameidata nd; 1608 int32_t t1; 1609 caddr_t bpos; 1610 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1; 1611 u_int32_t major, minor; 1612 enum vtype vtyp; 1613 char *cp2; 1614 struct mbuf *mb, *mreq; 1615 struct vnode *vp, *dirp = (struct vnode *)0; 1616 nfsfh_t nfh; 1617 fhandle_t *fhp; 1618 u_quad_t frev; 1619 struct mount *mp = NULL; 1620 1621 nd.ni_cnd.cn_nameiop = 0; 1622 fhp = &nfh.fh_generic; 1623 nfsm_srvmtofh(fhp); 1624 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1625 return (ESTALE); 1626 vn_start_write(NULL, &mp, V_WAIT); 1627 nfsm_srvnamesiz(len); 1628 nd.ni_cnd.cn_cred = cred; 1629 nd.ni_cnd.cn_nameiop = CREATE; 1630 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1631 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1632 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1633 if (dirp) 1634 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1635 if (error) { 1636 nfsm_reply(NFSX_WCCDATA(1)); 1637 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1638 if (dirp) 1639 vrele(dirp); 1640 vn_finished_write(mp, 0); 1641 return (0); 1642 } 1643 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1644 vtyp = nfsv3tov_type(*tl); 1645 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1646 error = NFSERR_BADTYPE; 1647 goto abort; 1648 } 1649 VATTR_NULL(&va); 1650 va.va_mode = 0; 1651 nfsm_srvsattr(&va); 1652 if (vtyp == VCHR || vtyp == VBLK) { 1653 dev_t rdev; 1654 1655 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1656 major = fxdr_unsigned(u_int32_t, *tl++); 1657 minor = fxdr_unsigned(u_int32_t, *tl); 1658 rdev = makedev(major, minor); 1659 if (major(rdev) != major || minor(rdev) != minor) { 1660 error = EINVAL; 1661 goto abort; 1662 } 1663 va.va_rdev = rdev; 1664 } 1665 1666 /* 1667 * Iff doesn't exist, create it. 1668 */ 1669 if (nd.ni_vp) { 1670 error = EEXIST; 1671 abort: 1672 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1673 if (nd.ni_dvp == nd.ni_vp) 1674 vrele(nd.ni_dvp); 1675 else 1676 vput(nd.ni_dvp); 1677 if (nd.ni_vp) 1678 vput(nd.ni_vp); 1679 goto out; 1680 } 1681 va.va_type = vtyp; 1682 if (vtyp == VSOCK) { 1683 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1684 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1685 } else { 1686 if (va.va_type != VFIFO && 1687 (error = suser(cred, (u_short *)0))) { 1688 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1689 vput(nd.ni_dvp); 1690 goto out; 1691 } 1692 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1693 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1694 if (error) { 1695 goto out; 1696 } 1697 if (error) 1698 goto out; 1699 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1700 vput(nd.ni_vp); 1701 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1702 error = EINVAL; 1703 } 1704 } 1705 out: 1706 vp = nd.ni_vp; 1707 if (!error) { 1708 memset((caddr_t)fhp, 0, sizeof(nfh)); 1709 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1710 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1711 if (!error) 1712 error = VOP_GETATTR(vp, &va, cred, lwp); 1713 vput(vp); 1714 } 1715 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1716 vrele(dirp); 1717 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1)); 1718 if (!error) { 1719 nfsm_srvpostop_fh(fhp); 1720 nfsm_srvpostop_attr(0, &va); 1721 } 1722 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1723 vn_finished_write(mp, 0); 1724 return (0); 1725 nfsmout: 1726 if (dirp) 1727 vrele(dirp); 1728 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1729 if (nd.ni_dvp == nd.ni_vp) 1730 vrele(nd.ni_dvp); 1731 else 1732 vput(nd.ni_dvp); 1733 if (nd.ni_vp) 1734 vput(nd.ni_vp); 1735 vn_finished_write(mp, 0); 1736 return (error); 1737 } 1738 1739 /* 1740 * nfs remove service 1741 */ 1742 int 1743 nfsrv_remove(nfsd, slp, lwp, mrq) 1744 struct nfsrv_descript *nfsd; 1745 struct nfssvc_sock *slp; 1746 struct lwp *lwp; 1747 struct mbuf **mrq; 1748 { 1749 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1750 struct mbuf *nam = nfsd->nd_nam; 1751 caddr_t dpos = nfsd->nd_dpos; 1752 struct ucred *cred = &nfsd->nd_cr; 1753 struct nameidata nd; 1754 u_int32_t *tl; 1755 int32_t t1; 1756 caddr_t bpos; 1757 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1; 1758 int v3 = (nfsd->nd_flag & ND_NFSV3); 1759 char *cp2; 1760 struct mbuf *mb, *mreq; 1761 struct vnode *vp, *dirp; 1762 struct vattr dirfor, diraft; 1763 nfsfh_t nfh; 1764 fhandle_t *fhp; 1765 u_quad_t frev; 1766 struct mount *mp = NULL; 1767 1768 #ifndef nolint 1769 vp = (struct vnode *)0; 1770 #endif 1771 fhp = &nfh.fh_generic; 1772 nfsm_srvmtofh(fhp); 1773 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1774 return (ESTALE); 1775 vn_start_write(NULL, &mp, V_WAIT); 1776 nfsm_srvnamesiz(len); 1777 nd.ni_cnd.cn_cred = cred; 1778 nd.ni_cnd.cn_nameiop = DELETE; 1779 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1780 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1781 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1782 if (dirp) { 1783 if (v3) 1784 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1785 else 1786 vrele(dirp); 1787 } 1788 if (!error) { 1789 vp = nd.ni_vp; 1790 if (vp->v_type == VDIR && 1791 (error = suser(cred, (u_short *)0)) != 0) 1792 goto out; 1793 /* 1794 * The root of a mounted filesystem cannot be deleted. 1795 */ 1796 if (vp->v_flag & VROOT) { 1797 error = EBUSY; 1798 goto out; 1799 } 1800 out: 1801 if (!error) { 1802 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1803 nqsrv_getl(vp, ND_WRITE); 1804 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1805 } else { 1806 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1807 if (nd.ni_dvp == vp) 1808 vrele(nd.ni_dvp); 1809 else 1810 vput(nd.ni_dvp); 1811 vput(vp); 1812 } 1813 } 1814 if (dirp && v3) { 1815 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1816 vrele(dirp); 1817 } 1818 nfsm_reply(NFSX_WCCDATA(v3)); 1819 if (v3) { 1820 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1821 vn_finished_write(mp, 0); 1822 return (0); 1823 } 1824 vn_finished_write(mp, 0); 1825 nfsm_srvdone; 1826 } 1827 1828 /* 1829 * nfs rename service 1830 */ 1831 int 1832 nfsrv_rename(nfsd, slp, lwp, mrq) 1833 struct nfsrv_descript *nfsd; 1834 struct nfssvc_sock *slp; 1835 struct lwp *lwp; 1836 struct mbuf **mrq; 1837 { 1838 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1839 struct mbuf *nam = nfsd->nd_nam; 1840 caddr_t dpos = nfsd->nd_dpos; 1841 struct ucred *cred = &nfsd->nd_cr; 1842 u_int32_t *tl; 1843 int32_t t1; 1844 caddr_t bpos; 1845 int error = 0, cache, fdirfor_ret = 1, fdiraft_ret = 1; 1846 uint32_t len, len2; 1847 int tdirfor_ret = 1, tdiraft_ret = 1; 1848 int v3 = (nfsd->nd_flag & ND_NFSV3); 1849 char *cp2; 1850 struct mbuf *mb, *mreq; 1851 struct nameidata fromnd, tond; 1852 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0; 1853 struct vnode *tdirp = (struct vnode *)0; 1854 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1855 nfsfh_t fnfh, tnfh; 1856 fhandle_t *ffhp, *tfhp; 1857 u_quad_t frev; 1858 uid_t saved_uid; 1859 struct mount *mp = NULL; 1860 1861 #ifndef nolint 1862 fvp = (struct vnode *)0; 1863 #endif 1864 ffhp = &fnfh.fh_generic; 1865 tfhp = &tnfh.fh_generic; 1866 fromnd.ni_cnd.cn_nameiop = 0; 1867 tond.ni_cnd.cn_nameiop = 0; 1868 nfsm_srvmtofh(ffhp); 1869 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) 1870 return (ESTALE); 1871 vn_start_write(NULL, &mp, V_WAIT); 1872 nfsm_srvnamesiz(len); 1873 /* 1874 * Remember our original uid so that we can reset cr_uid before 1875 * the second nfs_namei() call, in case it is remapped. 1876 */ 1877 saved_uid = cred->cr_uid; 1878 fromnd.ni_cnd.cn_cred = cred; 1879 fromnd.ni_cnd.cn_nameiop = DELETE; 1880 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART; 1881 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md, 1882 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1883 if (fdirp) { 1884 if (v3) 1885 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp); 1886 else { 1887 vrele(fdirp); 1888 fdirp = (struct vnode *)0; 1889 } 1890 } 1891 if (error) { 1892 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1893 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1894 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1895 if (fdirp) 1896 vrele(fdirp); 1897 vn_finished_write(mp, 0); 1898 return (0); 1899 } 1900 fvp = fromnd.ni_vp; 1901 nfsm_srvmtofh(tfhp); 1902 if (v3) { 1903 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1904 len2 = fxdr_unsigned(uint32_t, *tl); 1905 /* len2 will be checked by nfs_namei */ 1906 } 1907 else { 1908 /* NFSv2 */ 1909 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1910 } 1911 cred->cr_uid = saved_uid; 1912 tond.ni_cnd.cn_cred = cred; 1913 tond.ni_cnd.cn_nameiop = RENAME; 1914 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1915 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md, 1916 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1917 if (tdirp) { 1918 if (v3) 1919 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp); 1920 else { 1921 vrele(tdirp); 1922 tdirp = (struct vnode *)0; 1923 } 1924 } 1925 if (error) { 1926 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1927 vrele(fromnd.ni_dvp); 1928 vrele(fvp); 1929 goto out1; 1930 } 1931 tdvp = tond.ni_dvp; 1932 tvp = tond.ni_vp; 1933 if (tvp != NULL) { 1934 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1935 if (v3) 1936 error = EEXIST; 1937 else 1938 error = EISDIR; 1939 goto out; 1940 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1941 if (v3) 1942 error = EEXIST; 1943 else 1944 error = ENOTDIR; 1945 goto out; 1946 } 1947 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1948 if (v3) 1949 error = EXDEV; 1950 else 1951 error = ENOTEMPTY; 1952 goto out; 1953 } 1954 } 1955 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1956 if (v3) 1957 error = EXDEV; 1958 else 1959 error = ENOTEMPTY; 1960 goto out; 1961 } 1962 if (fvp->v_mount != tdvp->v_mount) { 1963 if (v3) 1964 error = EXDEV; 1965 else 1966 error = ENOTEMPTY; 1967 goto out; 1968 } 1969 if (fvp == tdvp) { 1970 if (v3) 1971 error = EINVAL; 1972 else 1973 error = ENOTEMPTY; 1974 } 1975 /* 1976 * If source is the same as the destination (that is the 1977 * same vnode with the same name in the same directory), 1978 * then there is nothing to do. 1979 */ 1980 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1981 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1982 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1983 fromnd.ni_cnd.cn_namelen)) 1984 error = -1; 1985 out: 1986 if (!error) { 1987 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1988 nqsrv_getl(tdvp, ND_WRITE); 1989 if (tvp) { 1990 nqsrv_getl(tvp, ND_WRITE); 1991 } 1992 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1993 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1994 } else { 1995 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 1996 if (tdvp == tvp) 1997 vrele(tdvp); 1998 else 1999 vput(tdvp); 2000 if (tvp) 2001 vput(tvp); 2002 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2003 vrele(fromnd.ni_dvp); 2004 vrele(fvp); 2005 if (error == -1) 2006 error = 0; 2007 } 2008 vrele(tond.ni_startdir); 2009 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2010 out1: 2011 if (fdirp) { 2012 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp); 2013 vrele(fdirp); 2014 } 2015 if (tdirp) { 2016 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp); 2017 vrele(tdirp); 2018 } 2019 vrele(fromnd.ni_startdir); 2020 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2021 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2022 if (v3) { 2023 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2024 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2025 } 2026 vn_finished_write(mp, 0); 2027 return (0); 2028 2029 nfsmout: 2030 if (fdirp) 2031 vrele(fdirp); 2032 if (tdirp) 2033 vrele(tdirp); 2034 if (tond.ni_cnd.cn_nameiop) { 2035 vrele(tond.ni_startdir); 2036 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2037 } 2038 if (fromnd.ni_cnd.cn_nameiop) { 2039 vrele(fromnd.ni_startdir); 2040 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2041 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2042 vrele(fromnd.ni_dvp); 2043 vrele(fvp); 2044 } 2045 return (error); 2046 } 2047 2048 /* 2049 * nfs link service 2050 */ 2051 int 2052 nfsrv_link(nfsd, slp, lwp, mrq) 2053 struct nfsrv_descript *nfsd; 2054 struct nfssvc_sock *slp; 2055 struct lwp *lwp; 2056 struct mbuf **mrq; 2057 { 2058 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2059 struct mbuf *nam = nfsd->nd_nam; 2060 caddr_t dpos = nfsd->nd_dpos; 2061 struct ucred *cred = &nfsd->nd_cr; 2062 struct nameidata nd; 2063 u_int32_t *tl; 2064 int32_t t1; 2065 caddr_t bpos; 2066 int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1; 2067 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2068 char *cp2; 2069 struct mbuf *mb, *mreq; 2070 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2071 struct vattr dirfor, diraft, at; 2072 nfsfh_t nfh, dnfh; 2073 fhandle_t *fhp, *dfhp; 2074 u_quad_t frev; 2075 struct mount *mp = NULL; 2076 2077 fhp = &nfh.fh_generic; 2078 dfhp = &dnfh.fh_generic; 2079 nfsm_srvmtofh(fhp); 2080 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2081 return (ESTALE); 2082 vn_start_write(NULL, &mp, V_WAIT); 2083 nfsm_srvmtofh(dfhp); 2084 nfsm_srvnamesiz(len); 2085 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam, 2086 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2087 if (error) { 2088 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2089 nfsm_srvpostop_attr(getret, &at); 2090 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2091 vn_finished_write(mp, 0); 2092 return (0); 2093 } 2094 if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0) 2095 goto out1; 2096 nd.ni_cnd.cn_cred = cred; 2097 nd.ni_cnd.cn_nameiop = CREATE; 2098 nd.ni_cnd.cn_flags = LOCKPARENT; 2099 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos, 2100 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2101 if (dirp) { 2102 if (v3) 2103 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2104 else { 2105 vrele(dirp); 2106 dirp = (struct vnode *)0; 2107 } 2108 } 2109 if (error) 2110 goto out1; 2111 xp = nd.ni_vp; 2112 if (xp != NULL) { 2113 error = EEXIST; 2114 goto out; 2115 } 2116 xp = nd.ni_dvp; 2117 if (vp->v_mount != xp->v_mount) 2118 error = EXDEV; 2119 out: 2120 if (!error) { 2121 nqsrv_getl(vp, ND_WRITE); 2122 nqsrv_getl(xp, ND_WRITE); 2123 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2124 } else { 2125 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2126 if (nd.ni_dvp == nd.ni_vp) 2127 vrele(nd.ni_dvp); 2128 else 2129 vput(nd.ni_dvp); 2130 if (nd.ni_vp) 2131 vrele(nd.ni_vp); 2132 } 2133 out1: 2134 if (v3) 2135 getret = VOP_GETATTR(vp, &at, cred, lwp); 2136 if (dirp) { 2137 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2138 vrele(dirp); 2139 } 2140 vrele(vp); 2141 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2142 if (v3) { 2143 nfsm_srvpostop_attr(getret, &at); 2144 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2145 vn_finished_write(mp, 0); 2146 return (0); 2147 } 2148 vn_finished_write(mp, 0); 2149 nfsm_srvdone; 2150 } 2151 2152 /* 2153 * nfs symbolic link service 2154 */ 2155 int 2156 nfsrv_symlink(nfsd, slp, lwp, mrq) 2157 struct nfsrv_descript *nfsd; 2158 struct nfssvc_sock *slp; 2159 struct lwp *lwp; 2160 struct mbuf **mrq; 2161 { 2162 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2163 struct mbuf *nam = nfsd->nd_nam; 2164 caddr_t dpos = nfsd->nd_dpos; 2165 struct ucred *cred = &nfsd->nd_cr; 2166 struct vattr va, dirfor, diraft; 2167 struct nameidata nd; 2168 u_int32_t *tl; 2169 int32_t t1; 2170 struct nfsv2_sattr *sp; 2171 char *bpos, *pathcp = NULL, *cp2; 2172 struct uio io; 2173 struct iovec iv; 2174 int error = 0, cache, dirfor_ret = 1, diraft_ret = 1; 2175 uint32_t len, len2; 2176 int v3 = (nfsd->nd_flag & ND_NFSV3); 2177 struct mbuf *mb, *mreq; 2178 struct vnode *dirp = (struct vnode *)0; 2179 nfsfh_t nfh; 2180 fhandle_t *fhp; 2181 u_quad_t frev; 2182 struct mount *mp = NULL; 2183 2184 nd.ni_cnd.cn_nameiop = 0; 2185 fhp = &nfh.fh_generic; 2186 nfsm_srvmtofh(fhp); 2187 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2188 return (ESTALE); 2189 vn_start_write(NULL, &mp, V_WAIT); 2190 nfsm_srvnamesiz(len); 2191 nd.ni_cnd.cn_cred = cred; 2192 nd.ni_cnd.cn_nameiop = CREATE; 2193 nd.ni_cnd.cn_flags = LOCKPARENT; 2194 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2195 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2196 if (dirp) { 2197 if (v3) 2198 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2199 else { 2200 vrele(dirp); 2201 dirp = (struct vnode *)0; 2202 } 2203 } 2204 if (error) 2205 goto out; 2206 VATTR_NULL(&va); 2207 va.va_type = VLNK; 2208 if (v3) { 2209 va.va_mode = 0; 2210 nfsm_srvsattr(&va); 2211 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2212 len2 = fxdr_unsigned(uint32_t, *tl); 2213 if (len2 > PATH_MAX) { 2214 /* XXX should check _PC_NO_TRUNC */ 2215 error = ENAMETOOLONG; 2216 goto abortop; 2217 } 2218 } 2219 else { 2220 /* NFSv2 */ 2221 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2222 } 2223 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2224 iv.iov_base = pathcp; 2225 iv.iov_len = len2; 2226 io.uio_resid = len2; 2227 io.uio_offset = 0; 2228 io.uio_iov = &iv; 2229 io.uio_iovcnt = 1; 2230 io.uio_segflg = UIO_SYSSPACE; 2231 io.uio_rw = UIO_READ; 2232 io.uio_lwp = NULL; 2233 nfsm_mtouio(&io, len2); 2234 if (!v3) { 2235 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2236 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2237 } 2238 *(pathcp + len2) = '\0'; 2239 if (nd.ni_vp) { 2240 error = EEXIST; 2241 abortop: 2242 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2243 if (nd.ni_dvp == nd.ni_vp) 2244 vrele(nd.ni_dvp); 2245 else 2246 vput(nd.ni_dvp); 2247 if (nd.ni_vp) 2248 vrele(nd.ni_vp); 2249 goto out; 2250 } 2251 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2252 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2253 if (!error) { 2254 if (v3) { 2255 memset((caddr_t)fhp, 0, sizeof(nfh)); 2256 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsidx; 2257 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2258 if (!error) 2259 error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp); 2260 vput(nd.ni_vp); 2261 } else { 2262 vput(nd.ni_vp); 2263 } 2264 } 2265 out: 2266 if (pathcp) 2267 free(pathcp, M_TEMP); 2268 if (dirp) { 2269 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2270 vrele(dirp); 2271 } 2272 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2273 if (v3) { 2274 if (!error) { 2275 nfsm_srvpostop_fh(fhp); 2276 nfsm_srvpostop_attr(0, &va); 2277 } 2278 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2279 } 2280 vn_finished_write(mp, 0); 2281 return (0); 2282 nfsmout: 2283 if (dirp) 2284 vrele(dirp); 2285 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2286 if (nd.ni_dvp == nd.ni_vp) 2287 vrele(nd.ni_dvp); 2288 else 2289 vput(nd.ni_dvp); 2290 if (nd.ni_vp) 2291 vrele(nd.ni_vp); 2292 if (pathcp) 2293 free(pathcp, M_TEMP); 2294 vn_finished_write(mp, 0); 2295 return (error); 2296 } 2297 2298 /* 2299 * nfs mkdir service 2300 */ 2301 int 2302 nfsrv_mkdir(nfsd, slp, lwp, mrq) 2303 struct nfsrv_descript *nfsd; 2304 struct nfssvc_sock *slp; 2305 struct lwp *lwp; 2306 struct mbuf **mrq; 2307 { 2308 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2309 struct mbuf *nam = nfsd->nd_nam; 2310 caddr_t dpos = nfsd->nd_dpos; 2311 struct ucred *cred = &nfsd->nd_cr; 2312 struct vattr va, dirfor, diraft; 2313 struct nfs_fattr *fp; 2314 struct nameidata nd; 2315 caddr_t cp; 2316 u_int32_t *tl; 2317 int32_t t1; 2318 caddr_t bpos; 2319 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1; 2320 int v3 = (nfsd->nd_flag & ND_NFSV3); 2321 char *cp2; 2322 struct mbuf *mb, *mreq; 2323 struct vnode *vp, *dirp = (struct vnode *)0; 2324 nfsfh_t nfh; 2325 fhandle_t *fhp; 2326 u_quad_t frev; 2327 struct mount *mp = NULL; 2328 2329 fhp = &nfh.fh_generic; 2330 nfsm_srvmtofh(fhp); 2331 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2332 return (ESTALE); 2333 vn_start_write(NULL, &mp, V_WAIT); 2334 nfsm_srvnamesiz(len); 2335 nd.ni_cnd.cn_cred = cred; 2336 nd.ni_cnd.cn_nameiop = CREATE; 2337 nd.ni_cnd.cn_flags = LOCKPARENT; 2338 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2339 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2340 if (dirp) { 2341 if (v3) 2342 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2343 else { 2344 vrele(dirp); 2345 dirp = (struct vnode *)0; 2346 } 2347 } 2348 if (error) { 2349 nfsm_reply(NFSX_WCCDATA(v3)); 2350 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2351 if (dirp) 2352 vrele(dirp); 2353 vn_finished_write(mp, 0); 2354 return (0); 2355 } 2356 VATTR_NULL(&va); 2357 if (v3) { 2358 va.va_mode = 0; 2359 nfsm_srvsattr(&va); 2360 } else { 2361 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2362 va.va_mode = nfstov_mode(*tl++); 2363 } 2364 va.va_type = VDIR; 2365 vp = nd.ni_vp; 2366 if (vp != NULL) { 2367 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2368 if (nd.ni_dvp == vp) 2369 vrele(nd.ni_dvp); 2370 else 2371 vput(nd.ni_dvp); 2372 vrele(vp); 2373 error = EEXIST; 2374 goto out; 2375 } 2376 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2377 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2378 if (!error) { 2379 vp = nd.ni_vp; 2380 memset((caddr_t)fhp, 0, sizeof(nfh)); 2381 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 2382 error = VFS_VPTOFH(vp, &fhp->fh_fid); 2383 if (!error) 2384 error = VOP_GETATTR(vp, &va, cred, lwp); 2385 vput(vp); 2386 } 2387 out: 2388 if (dirp) { 2389 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2390 vrele(dirp); 2391 } 2392 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2393 if (v3) { 2394 if (!error) { 2395 nfsm_srvpostop_fh(fhp); 2396 nfsm_srvpostop_attr(0, &va); 2397 } 2398 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2399 } else { 2400 nfsm_srvfhtom(fhp, v3); 2401 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2402 nfsm_srvfillattr(&va, fp); 2403 } 2404 vn_finished_write(mp, 0); 2405 return (0); 2406 nfsmout: 2407 if (dirp) 2408 vrele(dirp); 2409 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2410 if (nd.ni_dvp == nd.ni_vp) 2411 vrele(nd.ni_dvp); 2412 else 2413 vput(nd.ni_dvp); 2414 if (nd.ni_vp) 2415 vrele(nd.ni_vp); 2416 vn_finished_write(mp, 0); 2417 return (error); 2418 } 2419 2420 /* 2421 * nfs rmdir service 2422 */ 2423 int 2424 nfsrv_rmdir(nfsd, slp, lwp, mrq) 2425 struct nfsrv_descript *nfsd; 2426 struct nfssvc_sock *slp; 2427 struct lwp *lwp; 2428 struct mbuf **mrq; 2429 { 2430 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2431 struct mbuf *nam = nfsd->nd_nam; 2432 caddr_t dpos = nfsd->nd_dpos; 2433 struct ucred *cred = &nfsd->nd_cr; 2434 u_int32_t *tl; 2435 int32_t t1; 2436 caddr_t bpos; 2437 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1; 2438 int v3 = (nfsd->nd_flag & ND_NFSV3); 2439 char *cp2; 2440 struct mbuf *mb, *mreq; 2441 struct vnode *vp, *dirp = (struct vnode *)0; 2442 struct vattr dirfor, diraft; 2443 nfsfh_t nfh; 2444 fhandle_t *fhp; 2445 struct nameidata nd; 2446 u_quad_t frev; 2447 struct mount *mp = NULL; 2448 2449 fhp = &nfh.fh_generic; 2450 nfsm_srvmtofh(fhp); 2451 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2452 return (ESTALE); 2453 vn_start_write(NULL, &mp, V_WAIT); 2454 nfsm_srvnamesiz(len); 2455 nd.ni_cnd.cn_cred = cred; 2456 nd.ni_cnd.cn_nameiop = DELETE; 2457 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2458 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2459 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2460 if (dirp) { 2461 if (v3) 2462 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2463 else { 2464 vrele(dirp); 2465 dirp = (struct vnode *)0; 2466 } 2467 } 2468 if (error) { 2469 nfsm_reply(NFSX_WCCDATA(v3)); 2470 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2471 if (dirp) 2472 vrele(dirp); 2473 vn_finished_write(mp, 0); 2474 return (0); 2475 } 2476 vp = nd.ni_vp; 2477 if (vp->v_type != VDIR) { 2478 error = ENOTDIR; 2479 goto out; 2480 } 2481 /* 2482 * No rmdir "." please. 2483 */ 2484 if (nd.ni_dvp == vp) { 2485 error = EINVAL; 2486 goto out; 2487 } 2488 /* 2489 * The root of a mounted filesystem cannot be deleted. 2490 */ 2491 if (vp->v_flag & VROOT) 2492 error = EBUSY; 2493 out: 2494 if (!error) { 2495 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2496 nqsrv_getl(vp, ND_WRITE); 2497 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2498 } else { 2499 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2500 if (nd.ni_dvp == nd.ni_vp) 2501 vrele(nd.ni_dvp); 2502 else 2503 vput(nd.ni_dvp); 2504 vput(vp); 2505 } 2506 if (dirp) { 2507 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2508 vrele(dirp); 2509 } 2510 nfsm_reply(NFSX_WCCDATA(v3)); 2511 if (v3) { 2512 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2513 vn_finished_write(mp, 0); 2514 return (0); 2515 } 2516 vn_finished_write(mp, 0); 2517 nfsm_srvdone; 2518 } 2519 2520 /* 2521 * nfs readdir service 2522 * - mallocs what it thinks is enough to read 2523 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2524 * - calls VOP_READDIR() 2525 * - loops around building the reply 2526 * if the output generated exceeds count break out of loop 2527 * The nfsm_clget macro is used here so that the reply will be packed 2528 * tightly in mbuf clusters. 2529 * - it only knows that it has encountered eof when the VOP_READDIR() 2530 * reads nothing 2531 * - as such one readdir rpc will return eof false although you are there 2532 * and then the next will return eof 2533 * - it trims out records with d_fileno == 0 2534 * this doesn't matter for Unix clients, but they might confuse clients 2535 * for other os'. 2536 * - it trims out records with d_type == DT_WHT 2537 * these cannot be seen through NFS (unless we extend the protocol) 2538 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2539 * than requested, but this may not apply to all filesystems. For 2540 * example, client NFS does not { although it is never remote mounted 2541 * anyhow } 2542 * The alternate call nfsrv_readdirplus() does lookups as well. 2543 * PS: The NFS protocol spec. does not clarify what the "count" byte 2544 * argument is a count of.. just name strings and file id's or the 2545 * entire reply rpc or ... 2546 * I tried just file name and id sizes and it confused the Sun client, 2547 * so I am using the full rpc size now. The "paranoia.." comment refers 2548 * to including the status longwords that are not a part of the dir. 2549 * "entry" structures, but are in the rpc. 2550 */ 2551 2552 #define NFS_SRVDIRBLKSIZ 1024 2553 2554 struct flrep { 2555 nfsuint64 fl_off; 2556 u_int32_t fl_postopok; 2557 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2558 u_int32_t fl_fhok; 2559 u_int32_t fl_fhsize; 2560 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)]; 2561 }; 2562 2563 int 2564 nfsrv_readdir(nfsd, slp, lwp, mrq) 2565 struct nfsrv_descript *nfsd; 2566 struct nfssvc_sock *slp; 2567 struct lwp *lwp; 2568 struct mbuf **mrq; 2569 { 2570 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2571 struct mbuf *nam = nfsd->nd_nam; 2572 caddr_t dpos = nfsd->nd_dpos; 2573 struct ucred *cred = &nfsd->nd_cr; 2574 char *bp, *be; 2575 struct mbuf *mp; 2576 struct dirent *dp; 2577 caddr_t cp; 2578 u_int32_t *tl; 2579 int32_t t1; 2580 caddr_t bpos; 2581 struct mbuf *mb, *mreq, *mp2; 2582 char *cpos, *cend, *cp2, *rbuf; 2583 struct vnode *vp; 2584 struct vattr at; 2585 nfsfh_t nfh; 2586 fhandle_t *fhp; 2587 struct uio io; 2588 struct iovec iv; 2589 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2590 int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies; 2591 int v3 = (nfsd->nd_flag & ND_NFSV3); 2592 u_quad_t frev, off, toff, verf; 2593 off_t *cookies = NULL, *cookiep; 2594 nfsuint64 jar; 2595 2596 fhp = &nfh.fh_generic; 2597 nfsm_srvmtofh(fhp); 2598 if (v3) { 2599 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2600 toff = fxdr_hyper(tl); 2601 tl += 2; 2602 verf = fxdr_hyper(tl); 2603 tl += 2; 2604 } else { 2605 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2606 toff = fxdr_unsigned(u_quad_t, *tl++); 2607 } 2608 off = toff; 2609 cnt = fxdr_unsigned(int, *tl); 2610 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2611 xfer = NFS_SRVMAXDATA(nfsd); 2612 if (siz > xfer) 2613 siz = xfer; 2614 fullsiz = siz; 2615 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2616 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2617 if (!error && vp->v_type != VDIR) { 2618 error = ENOTDIR; 2619 vput(vp); 2620 } 2621 if (error) { 2622 nfsm_reply(NFSX_UNSIGNED); 2623 nfsm_srvpostop_attr(getret, &at); 2624 return (0); 2625 } 2626 nqsrv_getl(vp, ND_READ); 2627 if (v3) { 2628 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2629 #ifdef NFS3_STRICTVERF 2630 /* 2631 * XXX This check is too strict for Solaris 2.5 clients. 2632 */ 2633 if (!error && toff && verf != at.va_filerev) 2634 error = NFSERR_BAD_COOKIE; 2635 #endif 2636 } 2637 if (!error) 2638 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2639 if (error) { 2640 vput(vp); 2641 nfsm_reply(NFSX_POSTOPATTR(v3)); 2642 nfsm_srvpostop_attr(getret, &at); 2643 return (0); 2644 } 2645 VOP_UNLOCK(vp, 0); 2646 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2647 again: 2648 iv.iov_base = rbuf; 2649 iv.iov_len = fullsiz; 2650 io.uio_iov = &iv; 2651 io.uio_iovcnt = 1; 2652 io.uio_offset = (off_t)off; 2653 io.uio_resid = fullsiz; 2654 io.uio_segflg = UIO_SYSSPACE; 2655 io.uio_rw = UIO_READ; 2656 io.uio_lwp = NULL; 2657 eofflag = 0; 2658 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2659 2660 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2661 2662 off = (off_t)io.uio_offset; 2663 if (!cookies && !error) 2664 error = NFSERR_PERM; 2665 if (v3) { 2666 getret = VOP_GETATTR(vp, &at, cred, lwp); 2667 if (!error) 2668 error = getret; 2669 } 2670 2671 VOP_UNLOCK(vp, 0); 2672 if (error) { 2673 vrele(vp); 2674 free((caddr_t)rbuf, M_TEMP); 2675 if (cookies) 2676 free((caddr_t)cookies, M_TEMP); 2677 nfsm_reply(NFSX_POSTOPATTR(v3)); 2678 nfsm_srvpostop_attr(getret, &at); 2679 return (0); 2680 } 2681 if (io.uio_resid) { 2682 siz -= io.uio_resid; 2683 2684 /* 2685 * If nothing read, return eof 2686 * rpc reply 2687 */ 2688 if (siz == 0) { 2689 vrele(vp); 2690 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2691 2 * NFSX_UNSIGNED); 2692 if (v3) { 2693 nfsm_srvpostop_attr(getret, &at); 2694 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2695 txdr_hyper(at.va_filerev, tl); 2696 tl += 2; 2697 } else 2698 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2699 *tl++ = nfs_false; 2700 *tl = nfs_true; 2701 free((caddr_t)rbuf, M_TEMP); 2702 free((caddr_t)cookies, M_TEMP); 2703 return (0); 2704 } 2705 } 2706 2707 /* 2708 * Check for degenerate cases of nothing useful read. 2709 * If so go try again 2710 */ 2711 cpos = rbuf; 2712 cend = rbuf + siz; 2713 dp = (struct dirent *)cpos; 2714 cookiep = cookies; 2715 2716 while (cpos < cend && ncookies > 0 && 2717 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2718 cpos += dp->d_reclen; 2719 dp = (struct dirent *)cpos; 2720 cookiep++; 2721 ncookies--; 2722 } 2723 if (cpos >= cend || ncookies == 0) { 2724 toff = off; 2725 siz = fullsiz; 2726 goto again; 2727 } 2728 2729 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2730 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2731 if (v3) { 2732 nfsm_srvpostop_attr(getret, &at); 2733 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2734 txdr_hyper(at.va_filerev, tl); 2735 } 2736 mp = mp2 = mb; 2737 bp = bpos; 2738 be = bp + M_TRAILINGSPACE(mp); 2739 2740 /* Loop through the records and build reply */ 2741 while (cpos < cend && ncookies > 0) { 2742 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2743 nlen = dp->d_namlen; 2744 rem = nfsm_rndup(nlen)-nlen; 2745 len += (4 * NFSX_UNSIGNED + nlen + rem); 2746 if (v3) 2747 len += 2 * NFSX_UNSIGNED; 2748 if (len > cnt) { 2749 eofflag = 0; 2750 break; 2751 } 2752 /* 2753 * Build the directory record xdr from 2754 * the dirent entry. 2755 */ 2756 nfsm_clget; 2757 *tl = nfs_true; 2758 bp += NFSX_UNSIGNED; 2759 if (v3) { 2760 nfsm_clget; 2761 *tl = txdr_unsigned(dp->d_fileno >> 32); 2762 bp += NFSX_UNSIGNED; 2763 } 2764 nfsm_clget; 2765 *tl = txdr_unsigned(dp->d_fileno); 2766 bp += NFSX_UNSIGNED; 2767 nfsm_clget; 2768 *tl = txdr_unsigned(nlen); 2769 bp += NFSX_UNSIGNED; 2770 2771 /* And loop around copying the name */ 2772 xfer = nlen; 2773 cp = dp->d_name; 2774 while (xfer > 0) { 2775 nfsm_clget; 2776 if ((bp+xfer) > be) 2777 tsiz = be-bp; 2778 else 2779 tsiz = xfer; 2780 memcpy(bp, cp, tsiz); 2781 bp += tsiz; 2782 xfer -= tsiz; 2783 if (xfer > 0) 2784 cp += tsiz; 2785 } 2786 /* And null pad to an int32_t boundary */ 2787 for (i = 0; i < rem; i++) 2788 *bp++ = '\0'; 2789 nfsm_clget; 2790 2791 /* Finish off the record */ 2792 txdr_hyper(*cookiep, &jar); 2793 if (v3) { 2794 *tl = jar.nfsuquad[0]; 2795 bp += NFSX_UNSIGNED; 2796 nfsm_clget; 2797 } 2798 *tl = jar.nfsuquad[1]; 2799 bp += NFSX_UNSIGNED; 2800 } 2801 cpos += dp->d_reclen; 2802 dp = (struct dirent *)cpos; 2803 cookiep++; 2804 ncookies--; 2805 } 2806 vrele(vp); 2807 nfsm_clget; 2808 *tl = nfs_false; 2809 bp += NFSX_UNSIGNED; 2810 nfsm_clget; 2811 if (eofflag) 2812 *tl = nfs_true; 2813 else 2814 *tl = nfs_false; 2815 bp += NFSX_UNSIGNED; 2816 if (mp != mb) { 2817 if (bp < be) 2818 mp->m_len = bp - mtod(mp, caddr_t); 2819 } else 2820 mp->m_len += bp - bpos; 2821 free((caddr_t)rbuf, M_TEMP); 2822 free((caddr_t)cookies, M_TEMP); 2823 nfsm_srvdone; 2824 } 2825 2826 int 2827 nfsrv_readdirplus(nfsd, slp, lwp, mrq) 2828 struct nfsrv_descript *nfsd; 2829 struct nfssvc_sock *slp; 2830 struct lwp *lwp; 2831 struct mbuf **mrq; 2832 { 2833 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2834 struct mbuf *nam = nfsd->nd_nam; 2835 caddr_t dpos = nfsd->nd_dpos; 2836 struct ucred *cred = &nfsd->nd_cr; 2837 char *bp, *be; 2838 struct mbuf *mp; 2839 struct dirent *dp; 2840 caddr_t cp; 2841 u_int32_t *tl; 2842 int32_t t1; 2843 caddr_t bpos; 2844 struct mbuf *mb, *mreq, *mp2; 2845 char *cpos, *cend, *cp2, *rbuf; 2846 struct vnode *vp, *nvp; 2847 struct flrep fl; 2848 nfsfh_t nfh; 2849 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh; 2850 struct uio io; 2851 struct iovec iv; 2852 struct vattr va, at, *vap = &va; 2853 struct nfs_fattr *fp; 2854 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2855 int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies; 2856 u_quad_t frev, off, toff, verf; 2857 off_t *cookies = NULL, *cookiep; 2858 2859 fhp = &nfh.fh_generic; 2860 nfsm_srvmtofh(fhp); 2861 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2862 toff = fxdr_hyper(tl); 2863 tl += 2; 2864 verf = fxdr_hyper(tl); 2865 tl += 2; 2866 siz = fxdr_unsigned(int, *tl++); 2867 cnt = fxdr_unsigned(int, *tl); 2868 off = toff; 2869 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2870 xfer = NFS_SRVMAXDATA(nfsd); 2871 if (siz > xfer) 2872 siz = xfer; 2873 fullsiz = siz; 2874 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2875 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2876 if (!error && vp->v_type != VDIR) { 2877 error = ENOTDIR; 2878 vput(vp); 2879 } 2880 if (error) { 2881 nfsm_reply(NFSX_UNSIGNED); 2882 nfsm_srvpostop_attr(getret, &at); 2883 return (0); 2884 } 2885 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2886 #ifdef NFS3_STRICTVERF 2887 /* 2888 * XXX This check is too strict for Solaris 2.5 clients. 2889 */ 2890 if (!error && toff && verf != at.va_filerev) 2891 error = NFSERR_BAD_COOKIE; 2892 #endif 2893 if (!error) { 2894 nqsrv_getl(vp, ND_READ); 2895 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2896 } 2897 if (error) { 2898 vput(vp); 2899 nfsm_reply(NFSX_V3POSTOPATTR); 2900 nfsm_srvpostop_attr(getret, &at); 2901 return (0); 2902 } 2903 VOP_UNLOCK(vp, 0); 2904 2905 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2906 again: 2907 iv.iov_base = rbuf; 2908 iv.iov_len = fullsiz; 2909 io.uio_iov = &iv; 2910 io.uio_iovcnt = 1; 2911 io.uio_offset = (off_t)off; 2912 io.uio_resid = fullsiz; 2913 io.uio_segflg = UIO_SYSSPACE; 2914 io.uio_rw = UIO_READ; 2915 io.uio_lwp = NULL; 2916 eofflag = 0; 2917 2918 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2919 2920 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2921 2922 off = (u_quad_t)io.uio_offset; 2923 getret = VOP_GETATTR(vp, &at, cred, lwp); 2924 2925 VOP_UNLOCK(vp, 0); 2926 2927 /* 2928 * If the VGET operation doesn't work for this filesystem, 2929 * we can't support readdirplus. Returning NOTSUPP should 2930 * make clients fall back to plain readdir. 2931 * There's no need to check for VPTOFH as well, we wouldn't 2932 * even be here otherwise. 2933 */ 2934 if (!getret) { 2935 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2936 getret = (getret == EOPNOTSUPP) ? 2937 NFSERR_NOTSUPP : NFSERR_IO; 2938 else 2939 vput(nvp); 2940 } 2941 2942 if (!cookies && !error) 2943 error = NFSERR_PERM; 2944 if (!error) 2945 error = getret; 2946 if (error) { 2947 vrele(vp); 2948 if (cookies) 2949 free((caddr_t)cookies, M_TEMP); 2950 free((caddr_t)rbuf, M_TEMP); 2951 nfsm_reply(NFSX_V3POSTOPATTR); 2952 nfsm_srvpostop_attr(getret, &at); 2953 return (0); 2954 } 2955 if (io.uio_resid) { 2956 siz -= io.uio_resid; 2957 2958 /* 2959 * If nothing read, return eof 2960 * rpc reply 2961 */ 2962 if (siz == 0) { 2963 vrele(vp); 2964 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2965 2 * NFSX_UNSIGNED); 2966 nfsm_srvpostop_attr(getret, &at); 2967 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2968 txdr_hyper(at.va_filerev, tl); 2969 tl += 2; 2970 *tl++ = nfs_false; 2971 *tl = nfs_true; 2972 free((caddr_t)cookies, M_TEMP); 2973 free((caddr_t)rbuf, M_TEMP); 2974 return (0); 2975 } 2976 } 2977 2978 /* 2979 * Check for degenerate cases of nothing useful read. 2980 * If so go try again 2981 */ 2982 cpos = rbuf; 2983 cend = rbuf + siz; 2984 dp = (struct dirent *)cpos; 2985 cookiep = cookies; 2986 2987 while (cpos < cend && ncookies > 0 && 2988 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2989 cpos += dp->d_reclen; 2990 dp = (struct dirent *)cpos; 2991 cookiep++; 2992 ncookies--; 2993 } 2994 if (cpos >= cend || ncookies == 0) { 2995 toff = off; 2996 siz = fullsiz; 2997 goto again; 2998 } 2999 3000 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3001 nfsm_reply(cnt); 3002 nfsm_srvpostop_attr(getret, &at); 3003 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3004 txdr_hyper(at.va_filerev, tl); 3005 mp = mp2 = mb; 3006 bp = bpos; 3007 be = bp + M_TRAILINGSPACE(mp); 3008 3009 /* Loop through the records and build reply */ 3010 while (cpos < cend && ncookies > 0) { 3011 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3012 nlen = dp->d_namlen; 3013 rem = nfsm_rndup(nlen)-nlen; 3014 3015 /* 3016 * For readdir_and_lookup get the vnode using 3017 * the file number. 3018 */ 3019 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3020 goto invalid; 3021 memset((caddr_t)nfhp, 0, NFSX_V3FH); 3022 nfhp->fh_fsid = 3023 nvp->v_mount->mnt_stat.f_fsidx; 3024 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) { 3025 vput(nvp); 3026 goto invalid; 3027 } 3028 if (VOP_GETATTR(nvp, vap, cred, lwp)) { 3029 vput(nvp); 3030 goto invalid; 3031 } 3032 vput(nvp); 3033 3034 /* 3035 * If either the dircount or maxcount will be 3036 * exceeded, get out now. Both of these lengths 3037 * are calculated conservatively, including all 3038 * XDR overheads. 3039 */ 3040 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3041 NFSX_V3POSTOPATTR); 3042 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3043 if (len > cnt || dirlen > fullsiz) { 3044 eofflag = 0; 3045 break; 3046 } 3047 3048 /* 3049 * Build the directory record xdr from 3050 * the dirent entry. 3051 */ 3052 fp = (struct nfs_fattr *)&fl.fl_fattr; 3053 nfsm_srvfillattr(vap, fp); 3054 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3055 fl.fl_fhok = nfs_true; 3056 fl.fl_postopok = nfs_true; 3057 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3058 3059 nfsm_clget; 3060 *tl = nfs_true; 3061 bp += NFSX_UNSIGNED; 3062 nfsm_clget; 3063 *tl = txdr_unsigned(dp->d_fileno >> 32); 3064 bp += NFSX_UNSIGNED; 3065 nfsm_clget; 3066 *tl = txdr_unsigned(dp->d_fileno); 3067 bp += NFSX_UNSIGNED; 3068 nfsm_clget; 3069 *tl = txdr_unsigned(nlen); 3070 bp += NFSX_UNSIGNED; 3071 3072 /* And loop around copying the name */ 3073 xfer = nlen; 3074 cp = dp->d_name; 3075 while (xfer > 0) { 3076 nfsm_clget; 3077 if ((bp + xfer) > be) 3078 tsiz = be - bp; 3079 else 3080 tsiz = xfer; 3081 memcpy(bp, cp, tsiz); 3082 bp += tsiz; 3083 xfer -= tsiz; 3084 if (xfer > 0) 3085 cp += tsiz; 3086 } 3087 /* And null pad to an int32_t boundary */ 3088 for (i = 0; i < rem; i++) 3089 *bp++ = '\0'; 3090 3091 /* 3092 * Now copy the flrep structure out. 3093 */ 3094 xfer = sizeof (struct flrep); 3095 cp = (caddr_t)&fl; 3096 while (xfer > 0) { 3097 nfsm_clget; 3098 if ((bp + xfer) > be) 3099 tsiz = be - bp; 3100 else 3101 tsiz = xfer; 3102 memcpy(bp, cp, tsiz); 3103 bp += tsiz; 3104 xfer -= tsiz; 3105 if (xfer > 0) 3106 cp += tsiz; 3107 } 3108 } 3109 invalid: 3110 cpos += dp->d_reclen; 3111 dp = (struct dirent *)cpos; 3112 cookiep++; 3113 ncookies--; 3114 } 3115 vrele(vp); 3116 nfsm_clget; 3117 *tl = nfs_false; 3118 bp += NFSX_UNSIGNED; 3119 nfsm_clget; 3120 if (eofflag) 3121 *tl = nfs_true; 3122 else 3123 *tl = nfs_false; 3124 bp += NFSX_UNSIGNED; 3125 if (mp != mb) { 3126 if (bp < be) 3127 mp->m_len = bp - mtod(mp, caddr_t); 3128 } else 3129 mp->m_len += bp - bpos; 3130 free((caddr_t)cookies, M_TEMP); 3131 free((caddr_t)rbuf, M_TEMP); 3132 nfsm_srvdone; 3133 } 3134 3135 /* 3136 * nfs commit service 3137 */ 3138 int 3139 nfsrv_commit(nfsd, slp, lwp, mrq) 3140 struct nfsrv_descript *nfsd; 3141 struct nfssvc_sock *slp; 3142 struct lwp *lwp; 3143 struct mbuf **mrq; 3144 { 3145 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3146 struct mbuf *nam = nfsd->nd_nam; 3147 caddr_t dpos = nfsd->nd_dpos; 3148 struct ucred *cred = &nfsd->nd_cr; 3149 struct vattr bfor, aft; 3150 struct vnode *vp; 3151 nfsfh_t nfh; 3152 fhandle_t *fhp; 3153 u_int32_t *tl; 3154 int32_t t1; 3155 caddr_t bpos; 3156 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache; 3157 uint32_t cnt; 3158 char *cp2; 3159 struct mbuf *mb, *mreq; 3160 u_quad_t frev, off, end; 3161 struct mount *mp = NULL; 3162 3163 #ifndef nolint 3164 cache = 0; 3165 #endif 3166 fhp = &nfh.fh_generic; 3167 nfsm_srvmtofh(fhp); 3168 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 3169 return (ESTALE); 3170 vn_start_write(NULL, &mp, V_WAIT); 3171 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3172 3173 off = fxdr_hyper(tl); 3174 tl += 2; 3175 cnt = fxdr_unsigned(uint32_t, *tl); 3176 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3177 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3178 if (error) { 3179 nfsm_reply(2 * NFSX_UNSIGNED); 3180 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3181 vn_finished_write(mp, 0); 3182 return (0); 3183 } 3184 for_ret = VOP_GETATTR(vp, &bfor, cred, lwp); 3185 end = (cnt > 0) ? off + cnt : vp->v_size; 3186 if (end < off || end > vp->v_size) 3187 end = vp->v_size; 3188 if (off < vp->v_size) 3189 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp); 3190 /* else error == 0, from nfsrv_fhtovp() */ 3191 aft_ret = VOP_GETATTR(vp, &aft, cred, lwp); 3192 vput(vp); 3193 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3194 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3195 if (!error) { 3196 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3197 *tl++ = txdr_unsigned(boottime.tv_sec); 3198 *tl = txdr_unsigned(boottime.tv_usec); 3199 } else { 3200 vn_finished_write(mp, 0); 3201 return (0); 3202 } 3203 vn_finished_write(mp, 0); 3204 nfsm_srvdone; 3205 } 3206 3207 /* 3208 * nfs statfs service 3209 */ 3210 int 3211 nfsrv_statfs(nfsd, slp, lwp, mrq) 3212 struct nfsrv_descript *nfsd; 3213 struct nfssvc_sock *slp; 3214 struct lwp *lwp; 3215 struct mbuf **mrq; 3216 { 3217 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3218 struct mbuf *nam = nfsd->nd_nam; 3219 caddr_t dpos = nfsd->nd_dpos; 3220 struct ucred *cred = &nfsd->nd_cr; 3221 struct statvfs *sf; 3222 struct nfs_statfs *sfp; 3223 u_int32_t *tl; 3224 int32_t t1; 3225 caddr_t bpos; 3226 int error = 0, rdonly, cache, getret = 1; 3227 int v3 = (nfsd->nd_flag & ND_NFSV3); 3228 char *cp2; 3229 struct mbuf *mb, *mreq; 3230 struct vnode *vp; 3231 struct vattr at; 3232 nfsfh_t nfh; 3233 fhandle_t *fhp; 3234 struct statvfs statvfs; 3235 u_quad_t frev, tval; 3236 3237 #ifndef nolint 3238 cache = 0; 3239 #endif 3240 fhp = &nfh.fh_generic; 3241 nfsm_srvmtofh(fhp); 3242 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3243 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3244 if (error) { 3245 nfsm_reply(NFSX_UNSIGNED); 3246 nfsm_srvpostop_attr(getret, &at); 3247 return (0); 3248 } 3249 sf = &statvfs; 3250 error = VFS_STATVFS(vp->v_mount, sf, lwp); 3251 getret = VOP_GETATTR(vp, &at, cred, lwp); 3252 vput(vp); 3253 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3254 if (v3) 3255 nfsm_srvpostop_attr(getret, &at); 3256 if (error) 3257 return (0); 3258 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3259 if (v3) { 3260 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3261 txdr_hyper(tval, &sfp->sf_tbytes); 3262 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3263 txdr_hyper(tval, &sfp->sf_fbytes); 3264 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3265 txdr_hyper(tval, &sfp->sf_abytes); 3266 tval = (u_quad_t)sf->f_files; 3267 txdr_hyper(tval, &sfp->sf_tfiles); 3268 tval = (u_quad_t)sf->f_ffree; 3269 txdr_hyper(tval, &sfp->sf_ffiles); 3270 txdr_hyper(tval, &sfp->sf_afiles); 3271 sfp->sf_invarsec = 0; 3272 } else { 3273 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3274 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3275 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3276 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3277 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3278 } 3279 nfsm_srvdone; 3280 } 3281 3282 /* 3283 * nfs fsinfo service 3284 */ 3285 int 3286 nfsrv_fsinfo(nfsd, slp, lwp, mrq) 3287 struct nfsrv_descript *nfsd; 3288 struct nfssvc_sock *slp; 3289 struct lwp *lwp; 3290 struct mbuf **mrq; 3291 { 3292 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3293 struct mbuf *nam = nfsd->nd_nam; 3294 caddr_t dpos = nfsd->nd_dpos; 3295 struct ucred *cred = &nfsd->nd_cr; 3296 u_int32_t *tl; 3297 struct nfsv3_fsinfo *sip; 3298 int32_t t1; 3299 caddr_t bpos; 3300 int error = 0, rdonly, cache, getret = 1; 3301 uint32_t maxdata; 3302 char *cp2; 3303 struct mbuf *mb, *mreq; 3304 struct vnode *vp; 3305 struct vattr at; 3306 nfsfh_t nfh; 3307 fhandle_t *fhp; 3308 u_quad_t frev, maxfsize; 3309 struct statvfs sb; 3310 3311 #ifndef nolint 3312 cache = 0; 3313 #endif 3314 fhp = &nfh.fh_generic; 3315 nfsm_srvmtofh(fhp); 3316 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3317 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3318 if (error) { 3319 nfsm_reply(NFSX_UNSIGNED); 3320 nfsm_srvpostop_attr(getret, &at); 3321 return (0); 3322 } 3323 3324 /* XXX Try to make a guess on the max file size. */ 3325 VFS_STATVFS(vp->v_mount, &sb, (struct lwp *)0); 3326 maxfsize = (u_quad_t)0x80000000 * sb.f_frsize - 1; 3327 3328 getret = VOP_GETATTR(vp, &at, cred, lwp); 3329 vput(vp); 3330 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3331 nfsm_srvpostop_attr(getret, &at); 3332 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3333 3334 /* 3335 * XXX 3336 * There should be file system VFS OP(s) to get this information. 3337 * For now, assume ufs. 3338 */ 3339 if (slp->ns_so->so_type == SOCK_DGRAM) 3340 maxdata = NFS_MAXDGRAMDATA; 3341 else 3342 maxdata = NFS_MAXDATA; 3343 sip->fs_rtmax = txdr_unsigned(maxdata); 3344 sip->fs_rtpref = txdr_unsigned(maxdata); 3345 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3346 sip->fs_wtmax = txdr_unsigned(maxdata); 3347 sip->fs_wtpref = txdr_unsigned(maxdata); 3348 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3349 sip->fs_dtpref = txdr_unsigned(maxdata); 3350 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3351 sip->fs_timedelta.nfsv3_sec = 0; 3352 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3353 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3354 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3355 NFSV3FSINFO_CANSETTIME); 3356 nfsm_srvdone; 3357 } 3358 3359 /* 3360 * nfs pathconf service 3361 */ 3362 int 3363 nfsrv_pathconf(nfsd, slp, lwp, mrq) 3364 struct nfsrv_descript *nfsd; 3365 struct nfssvc_sock *slp; 3366 struct lwp *lwp; 3367 struct mbuf **mrq; 3368 { 3369 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3370 struct mbuf *nam = nfsd->nd_nam; 3371 caddr_t dpos = nfsd->nd_dpos; 3372 struct ucred *cred = &nfsd->nd_cr; 3373 u_int32_t *tl; 3374 struct nfsv3_pathconf *pc; 3375 int32_t t1; 3376 caddr_t bpos; 3377 int error = 0, rdonly, cache, getret = 1; 3378 register_t linkmax, namemax, chownres, notrunc; 3379 char *cp2; 3380 struct mbuf *mb, *mreq; 3381 struct vnode *vp; 3382 struct vattr at; 3383 nfsfh_t nfh; 3384 fhandle_t *fhp; 3385 u_quad_t frev; 3386 3387 #ifndef nolint 3388 cache = 0; 3389 #endif 3390 fhp = &nfh.fh_generic; 3391 nfsm_srvmtofh(fhp); 3392 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3393 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3394 if (error) { 3395 nfsm_reply(NFSX_UNSIGNED); 3396 nfsm_srvpostop_attr(getret, &at); 3397 return (0); 3398 } 3399 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3400 if (!error) 3401 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3402 if (!error) 3403 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3404 if (!error) 3405 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3406 getret = VOP_GETATTR(vp, &at, cred, lwp); 3407 vput(vp); 3408 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3409 nfsm_srvpostop_attr(getret, &at); 3410 if (error) 3411 return (0); 3412 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3413 3414 pc->pc_linkmax = txdr_unsigned(linkmax); 3415 pc->pc_namemax = txdr_unsigned(namemax); 3416 pc->pc_notrunc = txdr_unsigned(notrunc); 3417 pc->pc_chownrestricted = txdr_unsigned(chownres); 3418 3419 /* 3420 * These should probably be supported by VOP_PATHCONF(), but 3421 * until msdosfs is exportable (why would you want to?), the 3422 * Unix defaults should be ok. 3423 */ 3424 pc->pc_caseinsensitive = nfs_false; 3425 pc->pc_casepreserving = nfs_true; 3426 nfsm_srvdone; 3427 } 3428 3429 /* 3430 * Null operation, used by clients to ping server 3431 */ 3432 /* ARGSUSED */ 3433 int 3434 nfsrv_null(nfsd, slp, lwp, mrq) 3435 struct nfsrv_descript *nfsd; 3436 struct nfssvc_sock *slp; 3437 struct lwp *lwp; 3438 struct mbuf **mrq; 3439 { 3440 struct mbuf *mrep = nfsd->nd_mrep; 3441 caddr_t bpos; 3442 int error = NFSERR_RETVOID, cache = 0; 3443 struct mbuf *mb, *mreq; 3444 u_quad_t frev; 3445 3446 nfsm_reply(0); 3447 return (0); 3448 } 3449 3450 /* 3451 * No operation, used for obsolete procedures 3452 */ 3453 /* ARGSUSED */ 3454 int 3455 nfsrv_noop(nfsd, slp, lwp, mrq) 3456 struct nfsrv_descript *nfsd; 3457 struct nfssvc_sock *slp; 3458 struct lwp *lwp; 3459 struct mbuf **mrq; 3460 { 3461 struct mbuf *mrep = nfsd->nd_mrep; 3462 caddr_t bpos; 3463 int error, cache = 0; 3464 struct mbuf *mb, *mreq; 3465 u_quad_t frev; 3466 3467 if (nfsd->nd_repstat) 3468 error = nfsd->nd_repstat; 3469 else 3470 error = EPROCUNAVAIL; 3471 nfsm_reply(0); 3472 return (0); 3473 } 3474 3475 /* 3476 * Perform access checking for vnodes obtained from file handles that would 3477 * refer to files already opened by a Unix client. You cannot just use 3478 * vn_writechk() and VOP_ACCESS() for two reasons. 3479 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3480 * 2 - The owner is to be given access irrespective of mode bits for some 3481 * operations, so that processes that chmod after opening a file don't 3482 * break. I don't like this because it opens a security hole, but since 3483 * the nfs server opens a security hole the size of a barn door anyhow, 3484 * what the heck. 3485 * 3486 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3487 * will return EPERM instead of EACCESS. EPERM is always an error. 3488 */ 3489 int 3490 nfsrv_access(vp, flags, cred, rdonly, lwp, override) 3491 struct vnode *vp; 3492 int flags; 3493 struct ucred *cred; 3494 int rdonly; 3495 struct lwp *lwp; 3496 { 3497 struct vattr vattr; 3498 int error; 3499 if (flags & VWRITE) { 3500 /* Just vn_writechk() changed to check rdonly */ 3501 /* 3502 * Disallow write attempts on read-only file systems; 3503 * unless the file is a socket or a block or character 3504 * device resident on the file system. 3505 */ 3506 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3507 switch (vp->v_type) { 3508 case VREG: 3509 case VDIR: 3510 case VLNK: 3511 return (EROFS); 3512 default: 3513 break; 3514 } 3515 } 3516 3517 /* 3518 * If the vnode is in use as a process's text, 3519 * we can't allow writing. 3520 */ 3521 if (vp->v_flag & VTEXT) 3522 return (ETXTBSY); 3523 } 3524 error = VOP_GETATTR(vp, &vattr, cred, lwp); 3525 if (error) 3526 return (error); 3527 error = VOP_ACCESS(vp, flags, cred, lwp); 3528 /* 3529 * Allow certain operations for the owner (reads and writes 3530 * on files that are already open). 3531 */ 3532 if (override && error == EACCES && cred->cr_uid == vattr.va_uid) 3533 error = 0; 3534 return error; 3535 } 3536