nfs_serv.c revision 1.108
1 /* $NetBSD: nfs_serv.c,v 1.108 2006/05/10 21:53:19 mrg 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 1.108 2006/05/10 21:53:19 mrg 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 = 0; 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 = 0, 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 memset(&guard, 0, sizeof guard); /* XXX gcc */ 241 242 fhp = &nfh.fh_generic; 243 nfsm_srvmtofh(fhp); 244 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 245 return (ESTALE); 246 vn_start_write(NULL, &mp, V_WAIT); 247 VATTR_NULL(&va); 248 if (v3) { 249 nfsm_srvsattr(&va); 250 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 251 gcheck = fxdr_unsigned(int, *tl); 252 if (gcheck) { 253 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 254 fxdr_nfsv3time(tl, &guard); 255 } 256 } else { 257 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 258 /* 259 * Nah nah nah nah na nah 260 * There is a bug in the Sun client that puts 0xffff in the mode 261 * field of sattr when it should put in 0xffffffff. The u_short 262 * doesn't sign extend. 263 * --> check the low order 2 bytes for 0xffff 264 */ 265 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 266 va.va_mode = nfstov_mode(sp->sa_mode); 267 if (sp->sa_uid != nfs_xdrneg1) 268 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 269 if (sp->sa_gid != nfs_xdrneg1) 270 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 271 if (sp->sa_size != nfs_xdrneg1) 272 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 273 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 274 #ifdef notyet 275 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 276 #else 277 va.va_atime.tv_sec = 278 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 279 va.va_atime.tv_nsec = 0; 280 #endif 281 } 282 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 283 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 284 285 } 286 287 /* 288 * Now that we have all the fields, lets do it. 289 */ 290 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, 291 (nfsd->nd_flag & ND_KERBAUTH), FALSE); 292 if (error) { 293 nfsm_reply(2 * NFSX_UNSIGNED); 294 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 295 vn_finished_write(mp, 0); 296 return (0); 297 } 298 nqsrv_getl(vp, ND_WRITE); 299 if (v3) { 300 error = preat_ret = VOP_GETATTR(vp, &preat, cred, lwp); 301 if (!error && gcheck && 302 (preat.va_ctime.tv_sec != guard.tv_sec || 303 preat.va_ctime.tv_nsec != guard.tv_nsec)) 304 error = NFSERR_NOT_SYNC; 305 if (error) { 306 vput(vp); 307 nfsm_reply(NFSX_WCCDATA(v3)); 308 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 309 vn_finished_write(mp, 0); 310 return (0); 311 } 312 } 313 314 /* 315 * If the size is being changed write acces is required, otherwise 316 * just check for a read only file system. 317 */ 318 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 319 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 320 error = EROFS; 321 goto out; 322 } 323 } else { 324 if (vp->v_type == VDIR) { 325 error = EISDIR; 326 goto out; 327 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 328 lwp, 0)) != 0) 329 goto out; 330 } 331 error = VOP_SETATTR(vp, &va, cred, lwp); 332 postat_ret = VOP_GETATTR(vp, &va, cred, lwp); 333 if (!error) 334 error = postat_ret; 335 out: 336 vput(vp); 337 vn_finished_write(mp, 0); 338 nfsm_reply(NFSX_WCCORFATTR(v3)); 339 if (v3) { 340 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 341 return (0); 342 } else { 343 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 344 nfsm_srvfillattr(&va, fp); 345 } 346 nfsm_srvdone; 347 } 348 349 /* 350 * nfs lookup rpc 351 */ 352 int 353 nfsrv_lookup(nfsd, slp, lwp, mrq) 354 struct nfsrv_descript *nfsd; 355 struct nfssvc_sock *slp; 356 struct lwp *lwp; 357 struct mbuf **mrq; 358 { 359 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 360 struct mbuf *nam = nfsd->nd_nam; 361 caddr_t dpos = nfsd->nd_dpos; 362 struct ucred *cred = &nfsd->nd_cr; 363 struct nfs_fattr *fp; 364 struct nameidata nd, ind, *ndp = &nd; 365 struct vnode *vp, *dirp; 366 nfsfh_t nfh; 367 fhandle_t *fhp; 368 caddr_t cp; 369 u_int32_t *tl; 370 int32_t t1; 371 caddr_t bpos; 372 int error = 0, cache = 0, dirattr_ret = 1; 373 uint32_t len; 374 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 375 char *cp2; 376 struct mbuf *mb, *mreq; 377 struct vattr va, dirattr; 378 u_quad_t frev; 379 380 fhp = &nfh.fh_generic; 381 nfsm_srvmtofh(fhp); 382 nfsm_srvnamesiz(len); 383 384 pubflag = nfs_ispublicfh(fhp); 385 386 nd.ni_cnd.cn_cred = cred; 387 nd.ni_cnd.cn_nameiop = LOOKUP; 388 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART; 389 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 390 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 391 392 if (!error && pubflag) { 393 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) { 394 /* 395 * Setup call to lookup() to see if we can find 396 * the index file. Arguably, this doesn't belong 397 * in a kernel.. Ugh. 398 */ 399 ind = nd; 400 VOP_UNLOCK(nd.ni_vp, 0); 401 ind.ni_pathlen = strlen(nfs_pub.np_index); 402 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf = 403 nfs_pub.np_index; 404 ind.ni_startdir = nd.ni_vp; 405 VREF(ind.ni_startdir); 406 error = lookup(&ind); 407 if (!error) { 408 /* 409 * Found an index file. Get rid of 410 * the old references. 411 */ 412 if (dirp) 413 vrele(dirp); 414 dirp = nd.ni_vp; 415 vrele(nd.ni_startdir); 416 ndp = &ind; 417 } else 418 error = 0; 419 } 420 /* 421 * If the public filehandle was used, check that this lookup 422 * didn't result in a filehandle outside the publicly exported 423 * filesystem. 424 */ 425 426 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 427 vput(nd.ni_vp); 428 error = EPERM; 429 } 430 } 431 432 if (dirp) { 433 if (v3) 434 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred, lwp); 435 vrele(dirp); 436 } 437 438 if (error) { 439 nfsm_reply(NFSX_POSTOPATTR(v3)); 440 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 441 return (0); 442 } 443 444 nqsrv_getl(ndp->ni_startdir, ND_READ); 445 vrele(ndp->ni_startdir); 446 PNBUF_PUT(nd.ni_cnd.cn_pnbuf); 447 vp = ndp->ni_vp; 448 memset((caddr_t)fhp, 0, sizeof(nfh)); 449 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 450 error = VFS_VPTOFH(vp, &fhp->fh_fid); 451 if (!error) 452 error = VOP_GETATTR(vp, &va, cred, lwp); 453 vput(vp); 454 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 455 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3)); 456 if (error) { 457 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 458 return (0); 459 } 460 nfsm_srvfhtom(fhp, v3); 461 if (v3) { 462 nfsm_srvpostop_attr(0, &va); 463 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 464 } else { 465 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 466 nfsm_srvfillattr(&va, fp); 467 } 468 nfsm_srvdone; 469 } 470 471 /* 472 * nfs readlink service 473 */ 474 int 475 nfsrv_readlink(nfsd, slp, lwp, mrq) 476 struct nfsrv_descript *nfsd; 477 struct nfssvc_sock *slp; 478 struct lwp *lwp; 479 struct mbuf **mrq; 480 { 481 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 482 struct mbuf *nam = nfsd->nd_nam; 483 caddr_t dpos = nfsd->nd_dpos; 484 struct ucred *cred = &nfsd->nd_cr; 485 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 486 struct iovec *ivp = iv; 487 struct mbuf *mp; 488 u_int32_t *tl; 489 int32_t t1; 490 caddr_t bpos; 491 int error = 0, rdonly, cache = 0, i, padlen, getret; 492 uint32_t len; 493 int v3 = (nfsd->nd_flag & ND_NFSV3); 494 char *cp2; 495 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 496 struct vnode *vp; 497 struct vattr attr; 498 nfsfh_t nfh; 499 fhandle_t *fhp; 500 struct uio io, *uiop = &io; 501 u_quad_t frev; 502 503 fhp = &nfh.fh_generic; 504 nfsm_srvmtofh(fhp); 505 len = 0; 506 i = 0; 507 while (len < NFS_MAXPATHLEN) { 508 mp = m_get(M_WAIT, MT_DATA); 509 MCLAIM(mp, &nfs_mowner); 510 m_clget(mp, M_WAIT); 511 mp->m_len = NFSMSIZ(mp); 512 if (len == 0) 513 mp3 = mp2 = mp; 514 else { 515 mp2->m_next = mp; 516 mp2 = mp; 517 } 518 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 519 mp->m_len = NFS_MAXPATHLEN-len; 520 len = NFS_MAXPATHLEN; 521 } else 522 len += mp->m_len; 523 ivp->iov_base = mtod(mp, caddr_t); 524 ivp->iov_len = mp->m_len; 525 i++; 526 ivp++; 527 } 528 uiop->uio_iov = iv; 529 uiop->uio_iovcnt = i; 530 uiop->uio_offset = 0; 531 uiop->uio_resid = len; 532 uiop->uio_rw = UIO_READ; 533 UIO_SETUP_SYSSPACE(uiop); 534 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 535 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 536 if (error) { 537 m_freem(mp3); 538 nfsm_reply(2 * NFSX_UNSIGNED); 539 nfsm_srvpostop_attr(1, (struct vattr *)0); 540 return (0); 541 } 542 if (vp->v_type != VLNK) { 543 if (v3) 544 error = EINVAL; 545 else 546 error = ENXIO; 547 goto out; 548 } 549 nqsrv_getl(vp, ND_READ); 550 error = VOP_READLINK(vp, uiop, cred); 551 out: 552 getret = VOP_GETATTR(vp, &attr, cred, lwp); 553 vput(vp); 554 if (error) 555 m_freem(mp3); 556 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 557 if (v3) { 558 nfsm_srvpostop_attr(getret, &attr); 559 if (error) 560 return (0); 561 } 562 len -= uiop->uio_resid; 563 padlen = nfsm_padlen(len); 564 if (uiop->uio_resid || padlen) 565 nfs_zeropad(mp3, uiop->uio_resid, padlen); 566 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 567 *tl = txdr_unsigned(len); 568 mb->m_next = mp3; 569 nfsm_srvdone; 570 } 571 572 /* 573 * nfs read service 574 */ 575 int 576 nfsrv_read(nfsd, slp, lwp, mrq) 577 struct nfsrv_descript *nfsd; 578 struct nfssvc_sock *slp; 579 struct lwp *lwp; 580 struct mbuf **mrq; 581 { 582 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 583 struct mbuf *nam = nfsd->nd_nam; 584 caddr_t dpos = nfsd->nd_dpos; 585 struct ucred *cred = &nfsd->nd_cr; 586 struct mbuf *m; 587 struct nfs_fattr *fp; 588 u_int32_t *tl; 589 int32_t t1; 590 int i; 591 caddr_t bpos; 592 int error = 0, rdonly, cache = 0, getret; 593 int v3 = (nfsd->nd_flag & ND_NFSV3); 594 uint32_t reqlen, len, cnt, left; 595 int padlen; 596 char *cp2; 597 struct mbuf *mb, *mreq; 598 struct vnode *vp; 599 nfsfh_t nfh; 600 fhandle_t *fhp; 601 struct uio io, *uiop = &io; 602 struct vattr va; 603 off_t off; 604 u_quad_t frev; 605 606 fhp = &nfh.fh_generic; 607 nfsm_srvmtofh(fhp); 608 if (v3) { 609 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 610 off = fxdr_hyper(tl); 611 } else { 612 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 613 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 614 } 615 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 616 reqlen = fxdr_unsigned(uint32_t, *tl); 617 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 618 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 619 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 620 if (error) { 621 nfsm_reply(2 * NFSX_UNSIGNED); 622 nfsm_srvpostop_attr(1, (struct vattr *)0); 623 return (0); 624 } 625 if (vp->v_type != VREG) { 626 if (v3) 627 error = EINVAL; 628 else 629 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 630 } 631 if (!error) { 632 nqsrv_getl(vp, ND_READ); 633 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 634 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 635 } 636 getret = VOP_GETATTR(vp, &va, cred, lwp); 637 if (!error) 638 error = getret; 639 if (error) { 640 vput(vp); 641 nfsm_reply(NFSX_POSTOPATTR(v3)); 642 nfsm_srvpostop_attr(getret, &va); 643 return (0); 644 } 645 if (off >= va.va_size) 646 cnt = 0; 647 else if ((off + reqlen) > va.va_size) 648 cnt = va.va_size - off; 649 else 650 cnt = reqlen; 651 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 652 if (v3) { 653 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 654 *tl++ = nfs_true; 655 fp = (struct nfs_fattr *)tl; 656 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 657 } else { 658 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 659 fp = (struct nfs_fattr *)tl; 660 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 661 } 662 len = left = cnt; 663 if (cnt > 0) { 664 if (nfsd_use_loan) { 665 struct vm_page **pgpp; 666 voff_t pgoff = trunc_page(off); 667 int npages; 668 vaddr_t lva; 669 670 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 671 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 672 673 /* allocate kva for mbuf data */ 674 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 675 if (lva == 0) { 676 /* fall back to VOP_READ */ 677 goto loan_fail; 678 } 679 680 /* allocate mbuf */ 681 m = m_get(M_WAIT, MT_DATA); 682 MCLAIM(m, &nfs_mowner); 683 pgpp = m->m_ext.ext_pgs; 684 685 /* loan pages */ 686 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 687 pgpp); 688 if (error) { 689 sokvafree(lva, npages << PAGE_SHIFT); 690 m_free(m); 691 if (error == EBUSY) 692 goto loan_fail; 693 goto read_error; 694 } 695 696 /* associate kva to mbuf */ 697 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 698 cnt, M_MBUF, soloanfree, slp->ns_so); 699 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 700 m->m_len = cnt; 701 702 /* map pages */ 703 for (i = 0; i < npages; i++) { 704 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 705 VM_PROT_READ); 706 lva += PAGE_SIZE; 707 } 708 709 pmap_update(pmap_kernel()); 710 711 mb->m_next = m; 712 mb = m; 713 error = 0; 714 uiop->uio_resid = 0; 715 } else { 716 struct iovec *iv; 717 struct iovec *iv2; 718 struct mbuf *m2; 719 int siz; 720 loan_fail: 721 /* 722 * Generate the mbuf list with the uio_iov ref. to it. 723 */ 724 i = 0; 725 m = m2 = mb; 726 while (left > 0) { 727 siz = min(M_TRAILINGSPACE(m), left); 728 if (siz > 0) { 729 left -= siz; 730 i++; 731 } 732 if (left > 0) { 733 m = m_get(M_WAIT, MT_DATA); 734 MCLAIM(m, &nfs_mowner); 735 m_clget(m, M_WAIT); 736 m->m_len = 0; 737 m2->m_next = m; 738 m2 = m; 739 } 740 } 741 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 742 uiop->uio_iov = iv2 = iv; 743 m = mb; 744 left = cnt; 745 i = 0; 746 while (left > 0) { 747 if (m == NULL) 748 panic("nfsrv_read iov"); 749 siz = min(M_TRAILINGSPACE(m), left); 750 if (siz > 0) { 751 iv->iov_base = mtod(m, caddr_t) + 752 m->m_len; 753 iv->iov_len = siz; 754 m->m_len += siz; 755 left -= siz; 756 iv++; 757 i++; 758 } 759 m = m->m_next; 760 } 761 uiop->uio_iovcnt = i; 762 uiop->uio_offset = off; 763 uiop->uio_resid = cnt; 764 uiop->uio_rw = UIO_READ; 765 UIO_SETUP_SYSSPACE(uiop); 766 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 767 free((caddr_t)iv2, M_TEMP); 768 } 769 read_error: 770 if (error || (getret = VOP_GETATTR(vp, &va, cred, lwp)) != 0){ 771 if (!error) 772 error = getret; 773 m_freem(mreq); 774 vput(vp); 775 nfsm_reply(NFSX_POSTOPATTR(v3)); 776 nfsm_srvpostop_attr(getret, &va); 777 return (0); 778 } 779 } else { 780 uiop->uio_resid = 0; 781 } 782 vput(vp); 783 nfsm_srvfillattr(&va, fp); 784 len -= uiop->uio_resid; 785 padlen = nfsm_padlen(len); 786 if (uiop->uio_resid || padlen) 787 nfs_zeropad(mb, uiop->uio_resid, padlen); 788 if (v3) { 789 /* count */ 790 *tl++ = txdr_unsigned(len); 791 /* eof */ 792 if (off + len >= va.va_size) 793 *tl++ = nfs_true; 794 else 795 *tl++ = nfs_false; 796 } 797 *tl = txdr_unsigned(len); 798 nfsm_srvdone; 799 } 800 801 /* 802 * nfs write service 803 */ 804 int 805 nfsrv_write(nfsd, slp, lwp, mrq) 806 struct nfsrv_descript *nfsd; 807 struct nfssvc_sock *slp; 808 struct lwp *lwp; 809 struct mbuf **mrq; 810 { 811 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 812 struct mbuf *nam = nfsd->nd_nam; 813 caddr_t dpos = nfsd->nd_dpos; 814 struct ucred *cred = &nfsd->nd_cr; 815 struct iovec *ivp; 816 int i, cnt; 817 struct mbuf *mp; 818 struct nfs_fattr *fp; 819 struct iovec *iv; 820 struct vattr va, forat; 821 u_int32_t *tl; 822 int32_t t1; 823 caddr_t bpos; 824 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 825 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 826 int stable = NFSV3WRITE_FILESYNC; 827 int v3 = (nfsd->nd_flag & ND_NFSV3); 828 char *cp2; 829 struct mbuf *mb, *mreq; 830 struct vnode *vp; 831 nfsfh_t nfh; 832 fhandle_t *fhp; 833 struct uio io, *uiop = &io; 834 off_t off; 835 u_quad_t frev; 836 struct mount *mntp = NULL; 837 838 if (mrep == NULL) { 839 *mrq = NULL; 840 return (0); 841 } 842 fhp = &nfh.fh_generic; 843 nfsm_srvmtofh(fhp); 844 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 845 return (ESTALE); 846 vn_start_write(NULL, &mntp, V_WAIT); 847 if (v3) { 848 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 849 off = fxdr_hyper(tl); 850 tl += 3; 851 stable = fxdr_unsigned(int, *tl++); 852 } else { 853 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 854 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 855 tl += 2; 856 } 857 retlen = len = fxdr_unsigned(int32_t, *tl); 858 cnt = i = 0; 859 860 /* 861 * For NFS Version 2, it is not obvious what a write of zero length 862 * should do, but I might as well be consistent with Version 3, 863 * which is to return ok so long as there are no permission problems. 864 */ 865 if (len > 0) { 866 zeroing = 1; 867 mp = mrep; 868 while (mp) { 869 if (mp == md) { 870 zeroing = 0; 871 adjust = dpos - mtod(mp, caddr_t); 872 mp->m_len -= adjust; 873 if (mp->m_len > 0 && adjust > 0) 874 NFSMADV(mp, adjust); 875 } 876 if (zeroing) 877 mp->m_len = 0; 878 else if (mp->m_len > 0) { 879 i += mp->m_len; 880 if (i > len) { 881 mp->m_len -= (i - len); 882 zeroing = 1; 883 } 884 if (mp->m_len > 0) 885 cnt++; 886 } 887 mp = mp->m_next; 888 } 889 } 890 if (len > NFS_MAXDATA || len < 0 || i < len) { 891 error = EIO; 892 nfsm_reply(2 * NFSX_UNSIGNED); 893 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 894 vn_finished_write(mntp, 0); 895 return (0); 896 } 897 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 898 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 899 if (error) { 900 nfsm_reply(2 * NFSX_UNSIGNED); 901 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 902 vn_finished_write(mntp, 0); 903 return (0); 904 } 905 if (v3) 906 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 907 if (vp->v_type != VREG) { 908 if (v3) 909 error = EINVAL; 910 else 911 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 912 } 913 if (!error) { 914 nqsrv_getl(vp, ND_WRITE); 915 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 916 } 917 if (error) { 918 vput(vp); 919 nfsm_reply(NFSX_WCCDATA(v3)); 920 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 921 vn_finished_write(mntp, 0); 922 return (0); 923 } 924 925 if (len > 0) { 926 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 927 uiop->uio_iov = iv = ivp; 928 uiop->uio_iovcnt = cnt; 929 mp = mrep; 930 while (mp) { 931 if (mp->m_len > 0) { 932 ivp->iov_base = mtod(mp, caddr_t); 933 ivp->iov_len = mp->m_len; 934 ivp++; 935 } 936 mp = mp->m_next; 937 } 938 939 /* 940 * XXX 941 * The IO_METASYNC flag indicates that all metadata (and not 942 * just enough to ensure data integrity) must be written to 943 * stable storage synchronously. 944 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 945 */ 946 if (stable == NFSV3WRITE_UNSTABLE) 947 ioflags = IO_NODELOCKED; 948 else if (stable == NFSV3WRITE_DATASYNC) 949 ioflags = (IO_SYNC | IO_NODELOCKED); 950 else 951 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 952 uiop->uio_resid = len; 953 uiop->uio_rw = UIO_WRITE; 954 uiop->uio_offset = off; 955 UIO_SETUP_SYSSPACE(uiop); 956 error = VOP_WRITE(vp, uiop, ioflags, cred); 957 nfsstats.srvvop_writes++; 958 free(iv, M_TEMP); 959 } 960 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 961 vput(vp); 962 vn_finished_write(mntp, 0); 963 if (!error) 964 error = aftat_ret; 965 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 966 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 967 if (v3) { 968 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 969 if (error) 970 return (0); 971 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 972 *tl++ = txdr_unsigned(retlen); 973 if (stable == NFSV3WRITE_UNSTABLE) 974 *tl++ = txdr_unsigned(stable); 975 else 976 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 977 /* 978 * Actually, there is no need to txdr these fields, 979 * but it may make the values more human readable, 980 * for debugging purposes. 981 */ 982 *tl++ = txdr_unsigned(boottime.tv_sec); 983 *tl = txdr_unsigned(boottime.tv_usec); 984 } else { 985 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 986 nfsm_srvfillattr(&va, fp); 987 } 988 nfsm_srvdone; 989 } 990 991 /* 992 * NFS write service with write gathering support. Called when 993 * nfsrvw_procrastinate > 0. 994 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 995 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 996 * Jan. 1994. 997 */ 998 int 999 nfsrv_writegather(ndp, slp, lwp, mrq) 1000 struct nfsrv_descript **ndp; 1001 struct nfssvc_sock *slp; 1002 struct lwp *lwp; 1003 struct mbuf **mrq; 1004 { 1005 struct iovec *ivp; 1006 struct mbuf *mp; 1007 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1008 struct nfs_fattr *fp; 1009 int i = 0; 1010 struct iovec *iov; 1011 struct nfsrvw_delayhash *wpp; 1012 struct ucred *cred; 1013 struct vattr va, forat; 1014 u_int32_t *tl; 1015 int32_t t1; 1016 caddr_t bpos, dpos; 1017 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1018 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing; 1019 char *cp2; 1020 struct mbuf *mb, *mreq, *mrep, *md; 1021 struct vnode *vp; 1022 struct uio io, *uiop = &io; 1023 u_quad_t frev, cur_usec; 1024 struct mount *mntp = NULL; 1025 1026 *mrq = NULL; 1027 if (*ndp) { 1028 nfsd = *ndp; 1029 *ndp = NULL; 1030 mrep = nfsd->nd_mrep; 1031 md = nfsd->nd_md; 1032 dpos = nfsd->nd_dpos; 1033 cred = &nfsd->nd_cr; 1034 v3 = (nfsd->nd_flag & ND_NFSV3); 1035 LIST_INIT(&nfsd->nd_coalesce); 1036 nfsd->nd_mreq = NULL; 1037 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1038 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1039 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1040 1041 /* 1042 * Now, get the write header.. 1043 */ 1044 nfsm_srvmtofh(&nfsd->nd_fh); 1045 if (v3) { 1046 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1047 nfsd->nd_off = fxdr_hyper(tl); 1048 tl += 3; 1049 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1050 } else { 1051 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1052 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1053 tl += 2; 1054 } 1055 len = fxdr_unsigned(int32_t, *tl); 1056 nfsd->nd_len = len; 1057 nfsd->nd_eoff = nfsd->nd_off + len; 1058 1059 /* 1060 * Trim the header out of the mbuf list and trim off any trailing 1061 * junk so that the mbuf list has only the write data. 1062 */ 1063 zeroing = 1; 1064 i = 0; 1065 mp = mrep; 1066 while (mp) { 1067 if (mp == md) { 1068 zeroing = 0; 1069 adjust = dpos - mtod(mp, caddr_t); 1070 mp->m_len -= adjust; 1071 if (mp->m_len > 0 && adjust > 0) 1072 NFSMADV(mp, adjust); 1073 } 1074 if (zeroing) 1075 mp->m_len = 0; 1076 else { 1077 i += mp->m_len; 1078 if (i > len) { 1079 mp->m_len -= (i - len); 1080 zeroing = 1; 1081 } 1082 } 1083 mp = mp->m_next; 1084 } 1085 if (len > NFS_MAXDATA || len < 0 || i < len) { 1086 nfsmout: 1087 m_freem(mrep); 1088 error = EIO; 1089 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1090 if (v3) 1091 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1092 nfsd->nd_mreq = mreq; 1093 nfsd->nd_mrep = NULL; 1094 nfsd->nd_time = 0; 1095 } 1096 1097 /* 1098 * Add this entry to the hash and time queues. 1099 */ 1100 s = splsoftclock(); 1101 owp = NULL; 1102 wp = LIST_FIRST(&slp->ns_tq); 1103 while (wp && wp->nd_time < nfsd->nd_time) { 1104 owp = wp; 1105 wp = LIST_NEXT(wp, nd_tq); 1106 } 1107 if (owp) { 1108 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1109 } else { 1110 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1111 } 1112 if (nfsd->nd_mrep) { 1113 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data); 1114 owp = NULL; 1115 wp = LIST_FIRST(wpp); 1116 while (wp && 1117 memcmp(&nfsd->nd_fh, &wp->nd_fh, NFSX_V3FH)) { 1118 owp = wp; 1119 wp = LIST_NEXT(wp, nd_hash); 1120 } 1121 while (wp && wp->nd_off < nfsd->nd_off && 1122 !memcmp(&nfsd->nd_fh, &wp->nd_fh, NFSX_V3FH)) { 1123 owp = wp; 1124 wp = LIST_NEXT(wp, nd_hash); 1125 } 1126 if (owp) { 1127 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1128 1129 /* 1130 * Search the hash list for overlapping entries and 1131 * coalesce. 1132 */ 1133 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1134 wp = LIST_NEXT(nfsd, nd_hash); 1135 if (NFSW_SAMECRED(owp, nfsd)) 1136 nfsrvw_coalesce(owp, nfsd); 1137 } 1138 } else { 1139 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1140 } 1141 } 1142 splx(s); 1143 } 1144 1145 /* 1146 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1147 * and generate the associated reply mbuf list(s). 1148 */ 1149 loop1: 1150 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1151 s = splsoftclock(); 1152 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1153 owp = LIST_NEXT(nfsd, nd_tq); 1154 if (nfsd->nd_time > cur_usec) 1155 break; 1156 if (nfsd->nd_mreq) 1157 continue; 1158 LIST_REMOVE(nfsd, nd_tq); 1159 LIST_REMOVE(nfsd, nd_hash); 1160 splx(s); 1161 mrep = nfsd->nd_mrep; 1162 nfsd->nd_mrep = NULL; 1163 cred = &nfsd->nd_cr; 1164 v3 = (nfsd->nd_flag & ND_NFSV3); 1165 forat_ret = aftat_ret = 1; 1166 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1167 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1168 FALSE); 1169 if (!error) { 1170 if (v3) 1171 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 1172 if (vp->v_type != VREG) { 1173 if (v3) 1174 error = EINVAL; 1175 else 1176 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1177 } 1178 } else 1179 vp = NULL; 1180 if (!error) { 1181 nqsrv_getl(vp, ND_WRITE); 1182 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1183 } 1184 1185 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1186 ioflags = IO_NODELOCKED; 1187 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1188 ioflags = (IO_SYNC | IO_NODELOCKED); 1189 else 1190 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1191 uiop->uio_rw = UIO_WRITE; 1192 uiop->uio_offset = nfsd->nd_off; 1193 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1194 UIO_SETUP_SYSSPACE(uiop); 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 = 0, 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 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1549 } 1550 if (v3) { 1551 if (exclusive_flag && !error) { 1552 /* 1553 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1554 */ 1555 char oldverf[NFSX_V3CREATEVERF]; 1556 1557 txdr_nfsv3time(&va.va_atime, oldverf); 1558 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1559 error = EEXIST; 1560 } 1561 if (dirp) { 1562 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1563 vrele(dirp); 1564 } 1565 } 1566 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1567 if (v3) { 1568 if (!error) { 1569 nfsm_srvpostop_fh(fhp); 1570 nfsm_srvpostop_attr(0, &va); 1571 } 1572 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1573 } else { 1574 nfsm_srvfhtom(fhp, v3); 1575 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1576 nfsm_srvfillattr(&va, fp); 1577 } 1578 vn_finished_write(mp, 0); 1579 return (0); 1580 nfsmout: 1581 if (dirp) 1582 vrele(dirp); 1583 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1584 if (nd.ni_dvp == nd.ni_vp) 1585 vrele(nd.ni_dvp); 1586 else 1587 vput(nd.ni_dvp); 1588 if (nd.ni_vp) 1589 vput(nd.ni_vp); 1590 vn_finished_write(mp, 0); 1591 return (error); 1592 } 1593 1594 /* 1595 * nfs v3 mknod service 1596 */ 1597 int 1598 nfsrv_mknod(nfsd, slp, lwp, mrq) 1599 struct nfsrv_descript *nfsd; 1600 struct nfssvc_sock *slp; 1601 struct lwp *lwp; 1602 struct mbuf **mrq; 1603 { 1604 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1605 struct mbuf *nam = nfsd->nd_nam; 1606 caddr_t dpos = nfsd->nd_dpos; 1607 struct ucred *cred = &nfsd->nd_cr; 1608 struct vattr va, dirfor, diraft; 1609 u_int32_t *tl; 1610 struct nameidata nd; 1611 int32_t t1; 1612 caddr_t bpos; 1613 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1614 u_int32_t major, minor; 1615 enum vtype vtyp; 1616 char *cp2; 1617 struct mbuf *mb, *mreq; 1618 struct vnode *vp, *dirp = (struct vnode *)0; 1619 nfsfh_t nfh; 1620 fhandle_t *fhp; 1621 u_quad_t frev; 1622 struct mount *mp = NULL; 1623 1624 nd.ni_cnd.cn_nameiop = 0; 1625 fhp = &nfh.fh_generic; 1626 nfsm_srvmtofh(fhp); 1627 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1628 return (ESTALE); 1629 vn_start_write(NULL, &mp, V_WAIT); 1630 nfsm_srvnamesiz(len); 1631 nd.ni_cnd.cn_cred = cred; 1632 nd.ni_cnd.cn_nameiop = CREATE; 1633 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1634 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1635 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1636 if (dirp) 1637 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1638 if (error) { 1639 nfsm_reply(NFSX_WCCDATA(1)); 1640 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1641 if (dirp) 1642 vrele(dirp); 1643 vn_finished_write(mp, 0); 1644 return (0); 1645 } 1646 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1647 vtyp = nfsv3tov_type(*tl); 1648 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1649 error = NFSERR_BADTYPE; 1650 goto abort; 1651 } 1652 VATTR_NULL(&va); 1653 va.va_mode = 0; 1654 nfsm_srvsattr(&va); 1655 if (vtyp == VCHR || vtyp == VBLK) { 1656 dev_t rdev; 1657 1658 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1659 major = fxdr_unsigned(u_int32_t, *tl++); 1660 minor = fxdr_unsigned(u_int32_t, *tl); 1661 rdev = makedev(major, minor); 1662 if (major(rdev) != major || minor(rdev) != minor) { 1663 error = EINVAL; 1664 goto abort; 1665 } 1666 va.va_rdev = rdev; 1667 } 1668 1669 /* 1670 * Iff doesn't exist, create it. 1671 */ 1672 if (nd.ni_vp) { 1673 error = EEXIST; 1674 abort: 1675 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1676 if (nd.ni_dvp == nd.ni_vp) 1677 vrele(nd.ni_dvp); 1678 else 1679 vput(nd.ni_dvp); 1680 if (nd.ni_vp) 1681 vput(nd.ni_vp); 1682 goto out; 1683 } 1684 va.va_type = vtyp; 1685 if (vtyp == VSOCK) { 1686 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1687 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1688 } else { 1689 if (va.va_type != VFIFO && 1690 (error = suser(cred, (u_short *)0))) { 1691 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1692 vput(nd.ni_dvp); 1693 goto out; 1694 } 1695 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1696 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 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 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1715 } 1716 if (dirp) { 1717 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1718 vrele(dirp); 1719 } 1720 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1)); 1721 if (!error) { 1722 nfsm_srvpostop_fh(fhp); 1723 nfsm_srvpostop_attr(0, &va); 1724 } 1725 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1726 vn_finished_write(mp, 0); 1727 return (0); 1728 nfsmout: 1729 if (dirp) 1730 vrele(dirp); 1731 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1732 if (nd.ni_dvp == nd.ni_vp) 1733 vrele(nd.ni_dvp); 1734 else 1735 vput(nd.ni_dvp); 1736 if (nd.ni_vp) 1737 vput(nd.ni_vp); 1738 vn_finished_write(mp, 0); 1739 return (error); 1740 } 1741 1742 /* 1743 * nfs remove service 1744 */ 1745 int 1746 nfsrv_remove(nfsd, slp, lwp, mrq) 1747 struct nfsrv_descript *nfsd; 1748 struct nfssvc_sock *slp; 1749 struct lwp *lwp; 1750 struct mbuf **mrq; 1751 { 1752 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1753 struct mbuf *nam = nfsd->nd_nam; 1754 caddr_t dpos = nfsd->nd_dpos; 1755 struct ucred *cred = &nfsd->nd_cr; 1756 struct nameidata nd; 1757 u_int32_t *tl; 1758 int32_t t1; 1759 caddr_t bpos; 1760 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1761 int v3 = (nfsd->nd_flag & ND_NFSV3); 1762 char *cp2; 1763 struct mbuf *mb, *mreq; 1764 struct vnode *vp, *dirp; 1765 struct vattr dirfor, diraft; 1766 nfsfh_t nfh; 1767 fhandle_t *fhp; 1768 u_quad_t frev; 1769 struct mount *mp = NULL; 1770 1771 #ifndef nolint 1772 vp = (struct vnode *)0; 1773 #endif 1774 fhp = &nfh.fh_generic; 1775 nfsm_srvmtofh(fhp); 1776 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1777 return (ESTALE); 1778 vn_start_write(NULL, &mp, V_WAIT); 1779 nfsm_srvnamesiz(len); 1780 nd.ni_cnd.cn_cred = cred; 1781 nd.ni_cnd.cn_nameiop = DELETE; 1782 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1783 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1784 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1785 if (dirp) { 1786 if (v3) 1787 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1788 else 1789 vrele(dirp); 1790 } 1791 if (!error) { 1792 vp = nd.ni_vp; 1793 if (vp->v_type == VDIR && 1794 (error = suser(cred, (u_short *)0)) != 0) 1795 goto out; 1796 /* 1797 * The root of a mounted filesystem cannot be deleted. 1798 */ 1799 if (vp->v_flag & VROOT) { 1800 error = EBUSY; 1801 goto out; 1802 } 1803 out: 1804 if (!error) { 1805 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1806 nqsrv_getl(vp, ND_WRITE); 1807 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1808 } else { 1809 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1810 if (nd.ni_dvp == vp) 1811 vrele(nd.ni_dvp); 1812 else 1813 vput(nd.ni_dvp); 1814 vput(vp); 1815 } 1816 } 1817 if (dirp && v3) { 1818 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1819 vrele(dirp); 1820 } 1821 nfsm_reply(NFSX_WCCDATA(v3)); 1822 if (v3) { 1823 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1824 vn_finished_write(mp, 0); 1825 return (0); 1826 } 1827 vn_finished_write(mp, 0); 1828 nfsm_srvdone; 1829 } 1830 1831 /* 1832 * nfs rename service 1833 */ 1834 int 1835 nfsrv_rename(nfsd, slp, lwp, mrq) 1836 struct nfsrv_descript *nfsd; 1837 struct nfssvc_sock *slp; 1838 struct lwp *lwp; 1839 struct mbuf **mrq; 1840 { 1841 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1842 struct mbuf *nam = nfsd->nd_nam; 1843 caddr_t dpos = nfsd->nd_dpos; 1844 struct ucred *cred = &nfsd->nd_cr; 1845 u_int32_t *tl; 1846 int32_t t1; 1847 caddr_t bpos; 1848 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1849 uint32_t len, len2; 1850 int tdirfor_ret = 1, tdiraft_ret = 1; 1851 int v3 = (nfsd->nd_flag & ND_NFSV3); 1852 char *cp2; 1853 struct mbuf *mb, *mreq; 1854 struct nameidata fromnd, tond; 1855 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0; 1856 struct vnode *tdirp = (struct vnode *)0; 1857 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1858 nfsfh_t fnfh, tnfh; 1859 fhandle_t *ffhp, *tfhp; 1860 u_quad_t frev; 1861 uid_t saved_uid; 1862 struct mount *mp = NULL; 1863 1864 #ifndef nolint 1865 fvp = (struct vnode *)0; 1866 #endif 1867 ffhp = &fnfh.fh_generic; 1868 tfhp = &tnfh.fh_generic; 1869 fromnd.ni_cnd.cn_nameiop = 0; 1870 tond.ni_cnd.cn_nameiop = 0; 1871 nfsm_srvmtofh(ffhp); 1872 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) 1873 return (ESTALE); 1874 vn_start_write(NULL, &mp, V_WAIT); 1875 nfsm_srvnamesiz(len); 1876 /* 1877 * Remember our original uid so that we can reset cr_uid before 1878 * the second nfs_namei() call, in case it is remapped. 1879 */ 1880 saved_uid = cred->cr_uid; 1881 fromnd.ni_cnd.cn_cred = cred; 1882 fromnd.ni_cnd.cn_nameiop = DELETE; 1883 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART; 1884 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md, 1885 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1886 if (fdirp) { 1887 if (v3) 1888 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp); 1889 else { 1890 vrele(fdirp); 1891 fdirp = (struct vnode *)0; 1892 } 1893 } 1894 if (error) { 1895 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1896 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1897 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1898 if (fdirp) 1899 vrele(fdirp); 1900 vn_finished_write(mp, 0); 1901 return (0); 1902 } 1903 fvp = fromnd.ni_vp; 1904 nfsm_srvmtofh(tfhp); 1905 if (v3) { 1906 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1907 len2 = fxdr_unsigned(uint32_t, *tl); 1908 /* len2 will be checked by nfs_namei */ 1909 } 1910 else { 1911 /* NFSv2 */ 1912 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1913 } 1914 cred->cr_uid = saved_uid; 1915 tond.ni_cnd.cn_cred = cred; 1916 tond.ni_cnd.cn_nameiop = RENAME; 1917 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1918 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md, 1919 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1920 if (tdirp) { 1921 if (v3) 1922 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp); 1923 else { 1924 vrele(tdirp); 1925 tdirp = (struct vnode *)0; 1926 } 1927 } 1928 if (error) { 1929 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1930 vrele(fromnd.ni_dvp); 1931 vrele(fvp); 1932 goto out1; 1933 } 1934 tdvp = tond.ni_dvp; 1935 tvp = tond.ni_vp; 1936 if (tvp != NULL) { 1937 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1938 if (v3) 1939 error = EEXIST; 1940 else 1941 error = EISDIR; 1942 goto out; 1943 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1944 if (v3) 1945 error = EEXIST; 1946 else 1947 error = ENOTDIR; 1948 goto out; 1949 } 1950 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1951 if (v3) 1952 error = EXDEV; 1953 else 1954 error = ENOTEMPTY; 1955 goto out; 1956 } 1957 } 1958 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1959 if (v3) 1960 error = EXDEV; 1961 else 1962 error = ENOTEMPTY; 1963 goto out; 1964 } 1965 if (fvp->v_mount != tdvp->v_mount) { 1966 if (v3) 1967 error = EXDEV; 1968 else 1969 error = ENOTEMPTY; 1970 goto out; 1971 } 1972 if (fvp == tdvp) { 1973 if (v3) 1974 error = EINVAL; 1975 else 1976 error = ENOTEMPTY; 1977 } 1978 /* 1979 * If source is the same as the destination (that is the 1980 * same vnode with the same name in the same directory), 1981 * then there is nothing to do. 1982 */ 1983 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1984 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1985 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1986 fromnd.ni_cnd.cn_namelen)) 1987 error = -1; 1988 out: 1989 if (!error) { 1990 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1991 nqsrv_getl(tdvp, ND_WRITE); 1992 if (tvp) { 1993 nqsrv_getl(tvp, ND_WRITE); 1994 } 1995 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1996 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1997 } else { 1998 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 1999 if (tdvp == tvp) 2000 vrele(tdvp); 2001 else 2002 vput(tdvp); 2003 if (tvp) 2004 vput(tvp); 2005 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2006 vrele(fromnd.ni_dvp); 2007 vrele(fvp); 2008 if (error == -1) 2009 error = 0; 2010 } 2011 vrele(tond.ni_startdir); 2012 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2013 out1: 2014 if (fdirp) { 2015 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp); 2016 vrele(fdirp); 2017 } 2018 if (tdirp) { 2019 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp); 2020 vrele(tdirp); 2021 } 2022 vrele(fromnd.ni_startdir); 2023 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2024 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2025 if (v3) { 2026 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2027 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2028 } 2029 vn_finished_write(mp, 0); 2030 return (0); 2031 2032 nfsmout: 2033 if (fdirp) 2034 vrele(fdirp); 2035 #ifdef notdef 2036 if (tdirp) 2037 vrele(tdirp); 2038 #endif 2039 if (tond.ni_cnd.cn_nameiop) { 2040 vrele(tond.ni_startdir); 2041 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2042 } 2043 if (fromnd.ni_cnd.cn_nameiop) { 2044 vrele(fromnd.ni_startdir); 2045 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2046 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2047 vrele(fromnd.ni_dvp); 2048 vrele(fvp); 2049 } 2050 return (error); 2051 } 2052 2053 /* 2054 * nfs link service 2055 */ 2056 int 2057 nfsrv_link(nfsd, slp, lwp, mrq) 2058 struct nfsrv_descript *nfsd; 2059 struct nfssvc_sock *slp; 2060 struct lwp *lwp; 2061 struct mbuf **mrq; 2062 { 2063 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2064 struct mbuf *nam = nfsd->nd_nam; 2065 caddr_t dpos = nfsd->nd_dpos; 2066 struct ucred *cred = &nfsd->nd_cr; 2067 struct nameidata nd; 2068 u_int32_t *tl; 2069 int32_t t1; 2070 caddr_t bpos; 2071 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2072 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2073 char *cp2; 2074 struct mbuf *mb, *mreq; 2075 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2076 struct vattr dirfor, diraft, at; 2077 nfsfh_t nfh, dnfh; 2078 fhandle_t *fhp, *dfhp; 2079 u_quad_t frev; 2080 struct mount *mp = NULL; 2081 2082 fhp = &nfh.fh_generic; 2083 dfhp = &dnfh.fh_generic; 2084 nfsm_srvmtofh(fhp); 2085 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2086 return (ESTALE); 2087 vn_start_write(NULL, &mp, V_WAIT); 2088 nfsm_srvmtofh(dfhp); 2089 nfsm_srvnamesiz(len); 2090 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam, 2091 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2092 if (error) { 2093 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2094 nfsm_srvpostop_attr(getret, &at); 2095 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2096 vn_finished_write(mp, 0); 2097 return (0); 2098 } 2099 if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0) 2100 goto out1; 2101 nd.ni_cnd.cn_cred = cred; 2102 nd.ni_cnd.cn_nameiop = CREATE; 2103 nd.ni_cnd.cn_flags = LOCKPARENT; 2104 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos, 2105 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2106 if (dirp) { 2107 if (v3) 2108 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2109 else { 2110 vrele(dirp); 2111 dirp = (struct vnode *)0; 2112 } 2113 } 2114 if (error) 2115 goto out1; 2116 xp = nd.ni_vp; 2117 if (xp != NULL) { 2118 error = EEXIST; 2119 goto out; 2120 } 2121 xp = nd.ni_dvp; 2122 if (vp->v_mount != xp->v_mount) 2123 error = EXDEV; 2124 out: 2125 if (!error) { 2126 nqsrv_getl(vp, ND_WRITE); 2127 nqsrv_getl(xp, ND_WRITE); 2128 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2129 } else { 2130 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2131 if (nd.ni_dvp == nd.ni_vp) 2132 vrele(nd.ni_dvp); 2133 else 2134 vput(nd.ni_dvp); 2135 if (nd.ni_vp) 2136 vrele(nd.ni_vp); 2137 } 2138 out1: 2139 if (v3) 2140 getret = VOP_GETATTR(vp, &at, cred, lwp); 2141 if (dirp) { 2142 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2143 vrele(dirp); 2144 } 2145 vrele(vp); 2146 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2147 if (v3) { 2148 nfsm_srvpostop_attr(getret, &at); 2149 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2150 vn_finished_write(mp, 0); 2151 return (0); 2152 } 2153 vn_finished_write(mp, 0); 2154 nfsm_srvdone; 2155 } 2156 2157 /* 2158 * nfs symbolic link service 2159 */ 2160 int 2161 nfsrv_symlink(nfsd, slp, lwp, mrq) 2162 struct nfsrv_descript *nfsd; 2163 struct nfssvc_sock *slp; 2164 struct lwp *lwp; 2165 struct mbuf **mrq; 2166 { 2167 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2168 struct mbuf *nam = nfsd->nd_nam; 2169 caddr_t dpos = nfsd->nd_dpos; 2170 struct ucred *cred = &nfsd->nd_cr; 2171 struct vattr va, dirfor, diraft; 2172 struct nameidata nd; 2173 u_int32_t *tl; 2174 int32_t t1; 2175 struct nfsv2_sattr *sp; 2176 char *bpos, *pathcp = NULL, *cp2; 2177 struct uio io; 2178 struct iovec iv; 2179 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1; 2180 uint32_t len, len2; 2181 int v3 = (nfsd->nd_flag & ND_NFSV3); 2182 struct mbuf *mb, *mreq; 2183 struct vnode *dirp = (struct vnode *)0; 2184 nfsfh_t nfh; 2185 fhandle_t *fhp; 2186 u_quad_t frev; 2187 struct mount *mp = NULL; 2188 2189 nd.ni_cnd.cn_nameiop = 0; 2190 fhp = &nfh.fh_generic; 2191 nfsm_srvmtofh(fhp); 2192 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2193 return (ESTALE); 2194 vn_start_write(NULL, &mp, V_WAIT); 2195 nfsm_srvnamesiz(len); 2196 nd.ni_cnd.cn_cred = cred; 2197 nd.ni_cnd.cn_nameiop = CREATE; 2198 nd.ni_cnd.cn_flags = LOCKPARENT; 2199 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2200 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2201 if (dirp) { 2202 if (v3) 2203 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2204 else { 2205 vrele(dirp); 2206 dirp = (struct vnode *)0; 2207 } 2208 } 2209 if (error) 2210 goto out; 2211 VATTR_NULL(&va); 2212 va.va_type = VLNK; 2213 if (v3) { 2214 va.va_mode = 0; 2215 nfsm_srvsattr(&va); 2216 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2217 len2 = fxdr_unsigned(uint32_t, *tl); 2218 if (len2 > PATH_MAX) { 2219 /* XXX should check _PC_NO_TRUNC */ 2220 error = ENAMETOOLONG; 2221 goto abortop; 2222 } 2223 } 2224 else { 2225 /* NFSv2 */ 2226 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2227 } 2228 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2229 iv.iov_base = pathcp; 2230 iv.iov_len = len2; 2231 io.uio_resid = len2; 2232 io.uio_offset = 0; 2233 io.uio_iov = &iv; 2234 io.uio_iovcnt = 1; 2235 io.uio_rw = UIO_READ; 2236 UIO_SETUP_SYSSPACE(&io); 2237 nfsm_mtouio(&io, len2); 2238 if (!v3) { 2239 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2240 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2241 } 2242 *(pathcp + len2) = '\0'; 2243 if (nd.ni_vp) { 2244 error = EEXIST; 2245 abortop: 2246 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2247 if (nd.ni_dvp == nd.ni_vp) 2248 vrele(nd.ni_dvp); 2249 else 2250 vput(nd.ni_dvp); 2251 if (nd.ni_vp) 2252 vrele(nd.ni_vp); 2253 goto out; 2254 } 2255 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2256 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2257 if (!error) { 2258 if (v3) { 2259 memset((caddr_t)fhp, 0, sizeof(nfh)); 2260 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsidx; 2261 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2262 if (!error) 2263 error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp); 2264 vput(nd.ni_vp); 2265 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2266 } else { 2267 vput(nd.ni_vp); 2268 } 2269 } 2270 out: 2271 if (pathcp) 2272 free(pathcp, M_TEMP); 2273 if (dirp) { 2274 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2275 vrele(dirp); 2276 } 2277 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2278 if (v3) { 2279 if (!error) { 2280 nfsm_srvpostop_fh(fhp); 2281 nfsm_srvpostop_attr(0, &va); 2282 } 2283 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2284 } 2285 vn_finished_write(mp, 0); 2286 return (0); 2287 nfsmout: 2288 if (dirp) 2289 vrele(dirp); 2290 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2291 if (nd.ni_dvp == nd.ni_vp) 2292 vrele(nd.ni_dvp); 2293 else 2294 vput(nd.ni_dvp); 2295 if (nd.ni_vp) 2296 vrele(nd.ni_vp); 2297 if (pathcp) 2298 free(pathcp, M_TEMP); 2299 vn_finished_write(mp, 0); 2300 return (error); 2301 } 2302 2303 /* 2304 * nfs mkdir service 2305 */ 2306 int 2307 nfsrv_mkdir(nfsd, slp, lwp, mrq) 2308 struct nfsrv_descript *nfsd; 2309 struct nfssvc_sock *slp; 2310 struct lwp *lwp; 2311 struct mbuf **mrq; 2312 { 2313 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2314 struct mbuf *nam = nfsd->nd_nam; 2315 caddr_t dpos = nfsd->nd_dpos; 2316 struct ucred *cred = &nfsd->nd_cr; 2317 struct vattr va, dirfor, diraft; 2318 struct nfs_fattr *fp; 2319 struct nameidata nd; 2320 caddr_t cp; 2321 u_int32_t *tl; 2322 int32_t t1; 2323 caddr_t bpos; 2324 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2325 int v3 = (nfsd->nd_flag & ND_NFSV3); 2326 char *cp2; 2327 struct mbuf *mb, *mreq; 2328 struct vnode *vp, *dirp = (struct vnode *)0; 2329 nfsfh_t nfh; 2330 fhandle_t *fhp; 2331 u_quad_t frev; 2332 struct mount *mp = NULL; 2333 2334 fhp = &nfh.fh_generic; 2335 nfsm_srvmtofh(fhp); 2336 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2337 return (ESTALE); 2338 vn_start_write(NULL, &mp, V_WAIT); 2339 nfsm_srvnamesiz(len); 2340 nd.ni_cnd.cn_cred = cred; 2341 nd.ni_cnd.cn_nameiop = CREATE; 2342 nd.ni_cnd.cn_flags = LOCKPARENT; 2343 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2344 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2345 if (dirp) { 2346 if (v3) 2347 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2348 else { 2349 vrele(dirp); 2350 dirp = (struct vnode *)0; 2351 } 2352 } 2353 if (error) { 2354 nfsm_reply(NFSX_WCCDATA(v3)); 2355 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2356 if (dirp) 2357 vrele(dirp); 2358 vn_finished_write(mp, 0); 2359 return (0); 2360 } 2361 VATTR_NULL(&va); 2362 if (v3) { 2363 va.va_mode = 0; 2364 nfsm_srvsattr(&va); 2365 } else { 2366 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2367 va.va_mode = nfstov_mode(*tl++); 2368 } 2369 va.va_type = VDIR; 2370 vp = nd.ni_vp; 2371 if (vp != NULL) { 2372 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2373 if (nd.ni_dvp == vp) 2374 vrele(nd.ni_dvp); 2375 else 2376 vput(nd.ni_dvp); 2377 vrele(vp); 2378 error = EEXIST; 2379 goto out; 2380 } 2381 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2382 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2383 if (!error) { 2384 vp = nd.ni_vp; 2385 memset((caddr_t)fhp, 0, sizeof(nfh)); 2386 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 2387 error = VFS_VPTOFH(vp, &fhp->fh_fid); 2388 if (!error) 2389 error = VOP_GETATTR(vp, &va, cred, lwp); 2390 vput(vp); 2391 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2392 } 2393 out: 2394 if (dirp) { 2395 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2396 vrele(dirp); 2397 } 2398 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2399 if (v3) { 2400 if (!error) { 2401 nfsm_srvpostop_fh(fhp); 2402 nfsm_srvpostop_attr(0, &va); 2403 } 2404 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2405 } else { 2406 nfsm_srvfhtom(fhp, v3); 2407 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2408 nfsm_srvfillattr(&va, fp); 2409 } 2410 vn_finished_write(mp, 0); 2411 return (0); 2412 nfsmout: 2413 if (dirp) 2414 vrele(dirp); 2415 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2416 if (nd.ni_dvp == nd.ni_vp) 2417 vrele(nd.ni_dvp); 2418 else 2419 vput(nd.ni_dvp); 2420 if (nd.ni_vp) 2421 vrele(nd.ni_vp); 2422 vn_finished_write(mp, 0); 2423 return (error); 2424 } 2425 2426 /* 2427 * nfs rmdir service 2428 */ 2429 int 2430 nfsrv_rmdir(nfsd, slp, lwp, mrq) 2431 struct nfsrv_descript *nfsd; 2432 struct nfssvc_sock *slp; 2433 struct lwp *lwp; 2434 struct mbuf **mrq; 2435 { 2436 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2437 struct mbuf *nam = nfsd->nd_nam; 2438 caddr_t dpos = nfsd->nd_dpos; 2439 struct ucred *cred = &nfsd->nd_cr; 2440 u_int32_t *tl; 2441 int32_t t1; 2442 caddr_t bpos; 2443 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2444 int v3 = (nfsd->nd_flag & ND_NFSV3); 2445 char *cp2; 2446 struct mbuf *mb, *mreq; 2447 struct vnode *vp, *dirp = (struct vnode *)0; 2448 struct vattr dirfor, diraft; 2449 nfsfh_t nfh; 2450 fhandle_t *fhp; 2451 struct nameidata nd; 2452 u_quad_t frev; 2453 struct mount *mp = NULL; 2454 2455 fhp = &nfh.fh_generic; 2456 nfsm_srvmtofh(fhp); 2457 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2458 return (ESTALE); 2459 vn_start_write(NULL, &mp, V_WAIT); 2460 nfsm_srvnamesiz(len); 2461 nd.ni_cnd.cn_cred = cred; 2462 nd.ni_cnd.cn_nameiop = DELETE; 2463 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2464 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2465 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2466 if (dirp) { 2467 if (v3) 2468 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2469 else { 2470 vrele(dirp); 2471 dirp = (struct vnode *)0; 2472 } 2473 } 2474 if (error) { 2475 nfsm_reply(NFSX_WCCDATA(v3)); 2476 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2477 if (dirp) 2478 vrele(dirp); 2479 vn_finished_write(mp, 0); 2480 return (0); 2481 } 2482 vp = nd.ni_vp; 2483 if (vp->v_type != VDIR) { 2484 error = ENOTDIR; 2485 goto out; 2486 } 2487 /* 2488 * No rmdir "." please. 2489 */ 2490 if (nd.ni_dvp == vp) { 2491 error = EINVAL; 2492 goto out; 2493 } 2494 /* 2495 * The root of a mounted filesystem cannot be deleted. 2496 */ 2497 if (vp->v_flag & VROOT) 2498 error = EBUSY; 2499 out: 2500 if (!error) { 2501 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2502 nqsrv_getl(vp, ND_WRITE); 2503 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2504 } else { 2505 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2506 if (nd.ni_dvp == nd.ni_vp) 2507 vrele(nd.ni_dvp); 2508 else 2509 vput(nd.ni_dvp); 2510 vput(vp); 2511 } 2512 if (dirp) { 2513 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2514 vrele(dirp); 2515 } 2516 nfsm_reply(NFSX_WCCDATA(v3)); 2517 if (v3) { 2518 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2519 vn_finished_write(mp, 0); 2520 return (0); 2521 } 2522 vn_finished_write(mp, 0); 2523 nfsm_srvdone; 2524 } 2525 2526 /* 2527 * nfs readdir service 2528 * - mallocs what it thinks is enough to read 2529 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2530 * - calls VOP_READDIR() 2531 * - loops around building the reply 2532 * if the output generated exceeds count break out of loop 2533 * The nfsm_clget macro is used here so that the reply will be packed 2534 * tightly in mbuf clusters. 2535 * - it only knows that it has encountered eof when the VOP_READDIR() 2536 * reads nothing 2537 * - as such one readdir rpc will return eof false although you are there 2538 * and then the next will return eof 2539 * - it trims out records with d_fileno == 0 2540 * this doesn't matter for Unix clients, but they might confuse clients 2541 * for other os'. 2542 * - it trims out records with d_type == DT_WHT 2543 * these cannot be seen through NFS (unless we extend the protocol) 2544 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2545 * than requested, but this may not apply to all filesystems. For 2546 * example, client NFS does not { although it is never remote mounted 2547 * anyhow } 2548 * The alternate call nfsrv_readdirplus() does lookups as well. 2549 * PS: The NFS protocol spec. does not clarify what the "count" byte 2550 * argument is a count of.. just name strings and file id's or the 2551 * entire reply rpc or ... 2552 * I tried just file name and id sizes and it confused the Sun client, 2553 * so I am using the full rpc size now. The "paranoia.." comment refers 2554 * to including the status longwords that are not a part of the dir. 2555 * "entry" structures, but are in the rpc. 2556 */ 2557 2558 #define NFS_SRVDIRBLKSIZ 1024 2559 2560 struct flrep { 2561 nfsuint64 fl_off; 2562 u_int32_t fl_postopok; 2563 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2564 u_int32_t fl_fhok; 2565 u_int32_t fl_fhsize; 2566 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)]; 2567 }; 2568 2569 int 2570 nfsrv_readdir(nfsd, slp, lwp, mrq) 2571 struct nfsrv_descript *nfsd; 2572 struct nfssvc_sock *slp; 2573 struct lwp *lwp; 2574 struct mbuf **mrq; 2575 { 2576 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2577 struct mbuf *nam = nfsd->nd_nam; 2578 caddr_t dpos = nfsd->nd_dpos; 2579 struct ucred *cred = &nfsd->nd_cr; 2580 char *bp, *be; 2581 struct mbuf *mp; 2582 struct dirent *dp; 2583 caddr_t cp; 2584 u_int32_t *tl; 2585 int32_t t1; 2586 caddr_t bpos; 2587 struct mbuf *mb, *mreq, *mp2; 2588 char *cpos, *cend, *cp2, *rbuf; 2589 struct vnode *vp; 2590 struct vattr at; 2591 nfsfh_t nfh; 2592 fhandle_t *fhp; 2593 struct uio io; 2594 struct iovec iv; 2595 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2596 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2597 int v3 = (nfsd->nd_flag & ND_NFSV3); 2598 u_quad_t frev, off, toff, verf; 2599 off_t *cookies = NULL, *cookiep; 2600 nfsuint64 jar; 2601 2602 fhp = &nfh.fh_generic; 2603 nfsm_srvmtofh(fhp); 2604 if (v3) { 2605 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2606 toff = fxdr_hyper(tl); 2607 tl += 2; 2608 verf = fxdr_hyper(tl); 2609 tl += 2; 2610 } else { 2611 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2612 toff = fxdr_unsigned(u_quad_t, *tl++); 2613 } 2614 off = toff; 2615 cnt = fxdr_unsigned(int, *tl); 2616 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2617 xfer = NFS_SRVMAXDATA(nfsd); 2618 if (siz > xfer) 2619 siz = xfer; 2620 fullsiz = siz; 2621 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2622 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2623 if (!error && vp->v_type != VDIR) { 2624 error = ENOTDIR; 2625 vput(vp); 2626 } 2627 if (error) { 2628 nfsm_reply(NFSX_UNSIGNED); 2629 nfsm_srvpostop_attr(getret, &at); 2630 return (0); 2631 } 2632 nqsrv_getl(vp, ND_READ); 2633 if (v3) { 2634 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2635 #ifdef NFS3_STRICTVERF 2636 /* 2637 * XXX This check is too strict for Solaris 2.5 clients. 2638 */ 2639 if (!error && toff && verf != at.va_filerev) 2640 error = NFSERR_BAD_COOKIE; 2641 #endif 2642 } 2643 if (!error) 2644 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2645 if (error) { 2646 vput(vp); 2647 nfsm_reply(NFSX_POSTOPATTR(v3)); 2648 nfsm_srvpostop_attr(getret, &at); 2649 return (0); 2650 } 2651 VOP_UNLOCK(vp, 0); 2652 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2653 again: 2654 iv.iov_base = rbuf; 2655 iv.iov_len = fullsiz; 2656 io.uio_iov = &iv; 2657 io.uio_iovcnt = 1; 2658 io.uio_offset = (off_t)off; 2659 io.uio_resid = fullsiz; 2660 io.uio_rw = UIO_READ; 2661 UIO_SETUP_SYSSPACE(&io); 2662 eofflag = 0; 2663 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2664 2665 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2666 2667 off = (off_t)io.uio_offset; 2668 if (!cookies && !error) 2669 error = NFSERR_PERM; 2670 if (v3) { 2671 getret = VOP_GETATTR(vp, &at, cred, lwp); 2672 if (!error) 2673 error = getret; 2674 } 2675 2676 VOP_UNLOCK(vp, 0); 2677 if (error) { 2678 vrele(vp); 2679 free((caddr_t)rbuf, M_TEMP); 2680 if (cookies) 2681 free((caddr_t)cookies, M_TEMP); 2682 nfsm_reply(NFSX_POSTOPATTR(v3)); 2683 nfsm_srvpostop_attr(getret, &at); 2684 return (0); 2685 } 2686 if (io.uio_resid) { 2687 siz -= io.uio_resid; 2688 2689 /* 2690 * If nothing read, return eof 2691 * rpc reply 2692 */ 2693 if (siz == 0) { 2694 vrele(vp); 2695 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2696 2 * NFSX_UNSIGNED); 2697 if (v3) { 2698 nfsm_srvpostop_attr(getret, &at); 2699 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2700 txdr_hyper(at.va_filerev, tl); 2701 tl += 2; 2702 } else 2703 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2704 *tl++ = nfs_false; 2705 *tl = nfs_true; 2706 free((caddr_t)rbuf, M_TEMP); 2707 free((caddr_t)cookies, M_TEMP); 2708 return (0); 2709 } 2710 } 2711 2712 /* 2713 * Check for degenerate cases of nothing useful read. 2714 * If so go try again 2715 */ 2716 cpos = rbuf; 2717 cend = rbuf + siz; 2718 dp = (struct dirent *)cpos; 2719 cookiep = cookies; 2720 2721 while (cpos < cend && ncookies > 0 && 2722 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2723 cpos += dp->d_reclen; 2724 dp = (struct dirent *)cpos; 2725 cookiep++; 2726 ncookies--; 2727 } 2728 if (cpos >= cend || ncookies == 0) { 2729 toff = off; 2730 siz = fullsiz; 2731 goto again; 2732 } 2733 2734 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2735 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2736 if (v3) { 2737 nfsm_srvpostop_attr(getret, &at); 2738 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2739 txdr_hyper(at.va_filerev, tl); 2740 } 2741 mp = mp2 = mb; 2742 bp = bpos; 2743 be = bp + M_TRAILINGSPACE(mp); 2744 2745 /* Loop through the records and build reply */ 2746 while (cpos < cend && ncookies > 0) { 2747 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2748 nlen = dp->d_namlen; 2749 rem = nfsm_rndup(nlen)-nlen; 2750 len += (4 * NFSX_UNSIGNED + nlen + rem); 2751 if (v3) 2752 len += 2 * NFSX_UNSIGNED; 2753 if (len > cnt) { 2754 eofflag = 0; 2755 break; 2756 } 2757 /* 2758 * Build the directory record xdr from 2759 * the dirent entry. 2760 */ 2761 nfsm_clget; 2762 *tl = nfs_true; 2763 bp += NFSX_UNSIGNED; 2764 if (v3) { 2765 nfsm_clget; 2766 *tl = txdr_unsigned(dp->d_fileno >> 32); 2767 bp += NFSX_UNSIGNED; 2768 } 2769 nfsm_clget; 2770 *tl = txdr_unsigned(dp->d_fileno); 2771 bp += NFSX_UNSIGNED; 2772 nfsm_clget; 2773 *tl = txdr_unsigned(nlen); 2774 bp += NFSX_UNSIGNED; 2775 2776 /* And loop around copying the name */ 2777 xfer = nlen; 2778 cp = dp->d_name; 2779 while (xfer > 0) { 2780 nfsm_clget; 2781 if ((bp+xfer) > be) 2782 tsiz = be-bp; 2783 else 2784 tsiz = xfer; 2785 memcpy(bp, cp, tsiz); 2786 bp += tsiz; 2787 xfer -= tsiz; 2788 if (xfer > 0) 2789 cp += tsiz; 2790 } 2791 /* And null pad to an int32_t boundary */ 2792 for (i = 0; i < rem; i++) 2793 *bp++ = '\0'; 2794 nfsm_clget; 2795 2796 /* Finish off the record */ 2797 txdr_hyper(*cookiep, &jar); 2798 if (v3) { 2799 *tl = jar.nfsuquad[0]; 2800 bp += NFSX_UNSIGNED; 2801 nfsm_clget; 2802 } 2803 *tl = jar.nfsuquad[1]; 2804 bp += NFSX_UNSIGNED; 2805 } 2806 cpos += dp->d_reclen; 2807 dp = (struct dirent *)cpos; 2808 cookiep++; 2809 ncookies--; 2810 } 2811 vrele(vp); 2812 nfsm_clget; 2813 *tl = nfs_false; 2814 bp += NFSX_UNSIGNED; 2815 nfsm_clget; 2816 if (eofflag) 2817 *tl = nfs_true; 2818 else 2819 *tl = nfs_false; 2820 bp += NFSX_UNSIGNED; 2821 if (mp != mb) { 2822 if (bp < be) 2823 mp->m_len = bp - mtod(mp, caddr_t); 2824 } else 2825 mp->m_len += bp - bpos; 2826 free((caddr_t)rbuf, M_TEMP); 2827 free((caddr_t)cookies, M_TEMP); 2828 nfsm_srvdone; 2829 } 2830 2831 int 2832 nfsrv_readdirplus(nfsd, slp, lwp, mrq) 2833 struct nfsrv_descript *nfsd; 2834 struct nfssvc_sock *slp; 2835 struct lwp *lwp; 2836 struct mbuf **mrq; 2837 { 2838 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2839 struct mbuf *nam = nfsd->nd_nam; 2840 caddr_t dpos = nfsd->nd_dpos; 2841 struct ucred *cred = &nfsd->nd_cr; 2842 char *bp, *be; 2843 struct mbuf *mp; 2844 struct dirent *dp; 2845 caddr_t cp; 2846 u_int32_t *tl; 2847 int32_t t1; 2848 caddr_t bpos; 2849 struct mbuf *mb, *mreq, *mp2; 2850 char *cpos, *cend, *cp2, *rbuf; 2851 struct vnode *vp, *nvp; 2852 struct flrep fl; 2853 nfsfh_t nfh; 2854 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh; 2855 struct uio io; 2856 struct iovec iv; 2857 struct vattr va, at, *vap = &va; 2858 struct nfs_fattr *fp; 2859 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2860 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2861 u_quad_t frev, off, toff, verf; 2862 off_t *cookies = NULL, *cookiep; 2863 2864 fhp = &nfh.fh_generic; 2865 nfsm_srvmtofh(fhp); 2866 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2867 toff = fxdr_hyper(tl); 2868 tl += 2; 2869 verf = fxdr_hyper(tl); 2870 tl += 2; 2871 siz = fxdr_unsigned(int, *tl++); 2872 cnt = fxdr_unsigned(int, *tl); 2873 off = toff; 2874 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2875 xfer = NFS_SRVMAXDATA(nfsd); 2876 if (siz > xfer) 2877 siz = xfer; 2878 fullsiz = siz; 2879 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2880 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2881 if (!error && vp->v_type != VDIR) { 2882 error = ENOTDIR; 2883 vput(vp); 2884 } 2885 if (error) { 2886 nfsm_reply(NFSX_UNSIGNED); 2887 nfsm_srvpostop_attr(getret, &at); 2888 return (0); 2889 } 2890 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2891 #ifdef NFS3_STRICTVERF 2892 /* 2893 * XXX This check is too strict for Solaris 2.5 clients. 2894 */ 2895 if (!error && toff && verf != at.va_filerev) 2896 error = NFSERR_BAD_COOKIE; 2897 #endif 2898 if (!error) { 2899 nqsrv_getl(vp, ND_READ); 2900 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2901 } 2902 if (error) { 2903 vput(vp); 2904 nfsm_reply(NFSX_V3POSTOPATTR); 2905 nfsm_srvpostop_attr(getret, &at); 2906 return (0); 2907 } 2908 VOP_UNLOCK(vp, 0); 2909 2910 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2911 again: 2912 iv.iov_base = rbuf; 2913 iv.iov_len = fullsiz; 2914 io.uio_iov = &iv; 2915 io.uio_iovcnt = 1; 2916 io.uio_offset = (off_t)off; 2917 io.uio_resid = fullsiz; 2918 io.uio_rw = UIO_READ; 2919 UIO_SETUP_SYSSPACE(&io); 2920 eofflag = 0; 2921 2922 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2923 2924 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2925 2926 off = (u_quad_t)io.uio_offset; 2927 getret = VOP_GETATTR(vp, &at, cred, lwp); 2928 2929 VOP_UNLOCK(vp, 0); 2930 2931 /* 2932 * If the VGET operation doesn't work for this filesystem, 2933 * we can't support readdirplus. Returning NOTSUPP should 2934 * make clients fall back to plain readdir. 2935 * There's no need to check for VPTOFH as well, we wouldn't 2936 * even be here otherwise. 2937 */ 2938 if (!getret) { 2939 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2940 getret = (getret == EOPNOTSUPP) ? 2941 NFSERR_NOTSUPP : NFSERR_IO; 2942 else 2943 vput(nvp); 2944 } 2945 2946 if (!cookies && !error) 2947 error = NFSERR_PERM; 2948 if (!error) 2949 error = getret; 2950 if (error) { 2951 vrele(vp); 2952 if (cookies) 2953 free((caddr_t)cookies, M_TEMP); 2954 free((caddr_t)rbuf, M_TEMP); 2955 nfsm_reply(NFSX_V3POSTOPATTR); 2956 nfsm_srvpostop_attr(getret, &at); 2957 return (0); 2958 } 2959 if (io.uio_resid) { 2960 siz -= io.uio_resid; 2961 2962 /* 2963 * If nothing read, return eof 2964 * rpc reply 2965 */ 2966 if (siz == 0) { 2967 vrele(vp); 2968 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2969 2 * NFSX_UNSIGNED); 2970 nfsm_srvpostop_attr(getret, &at); 2971 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2972 txdr_hyper(at.va_filerev, tl); 2973 tl += 2; 2974 *tl++ = nfs_false; 2975 *tl = nfs_true; 2976 free((caddr_t)cookies, M_TEMP); 2977 free((caddr_t)rbuf, M_TEMP); 2978 return (0); 2979 } 2980 } 2981 2982 /* 2983 * Check for degenerate cases of nothing useful read. 2984 * If so go try again 2985 */ 2986 cpos = rbuf; 2987 cend = rbuf + siz; 2988 dp = (struct dirent *)cpos; 2989 cookiep = cookies; 2990 2991 while (cpos < cend && ncookies > 0 && 2992 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2993 cpos += dp->d_reclen; 2994 dp = (struct dirent *)cpos; 2995 cookiep++; 2996 ncookies--; 2997 } 2998 if (cpos >= cend || ncookies == 0) { 2999 toff = off; 3000 siz = fullsiz; 3001 goto again; 3002 } 3003 3004 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3005 nfsm_reply(cnt); 3006 nfsm_srvpostop_attr(getret, &at); 3007 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3008 txdr_hyper(at.va_filerev, tl); 3009 mp = mp2 = mb; 3010 bp = bpos; 3011 be = bp + M_TRAILINGSPACE(mp); 3012 3013 /* Loop through the records and build reply */ 3014 while (cpos < cend && ncookies > 0) { 3015 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3016 nlen = dp->d_namlen; 3017 rem = nfsm_rndup(nlen)-nlen; 3018 3019 /* 3020 * For readdir_and_lookup get the vnode using 3021 * the file number. 3022 */ 3023 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3024 goto invalid; 3025 memset((caddr_t)nfhp, 0, NFSX_V3FH); 3026 nfhp->fh_fsid = 3027 nvp->v_mount->mnt_stat.f_fsidx; 3028 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) { 3029 vput(nvp); 3030 goto invalid; 3031 } 3032 if (VOP_GETATTR(nvp, vap, cred, lwp)) { 3033 vput(nvp); 3034 goto invalid; 3035 } 3036 vput(nvp); 3037 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 3038 3039 /* 3040 * If either the dircount or maxcount will be 3041 * exceeded, get out now. Both of these lengths 3042 * are calculated conservatively, including all 3043 * XDR overheads. 3044 */ 3045 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3046 NFSX_V3POSTOPATTR); 3047 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3048 if (len > cnt || dirlen > fullsiz) { 3049 eofflag = 0; 3050 break; 3051 } 3052 3053 /* 3054 * Build the directory record xdr from 3055 * the dirent entry. 3056 */ 3057 fp = (struct nfs_fattr *)&fl.fl_fattr; 3058 nfsm_srvfillattr(vap, fp); 3059 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3060 fl.fl_fhok = nfs_true; 3061 fl.fl_postopok = nfs_true; 3062 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3063 3064 nfsm_clget; 3065 *tl = nfs_true; 3066 bp += NFSX_UNSIGNED; 3067 nfsm_clget; 3068 *tl = txdr_unsigned(dp->d_fileno >> 32); 3069 bp += NFSX_UNSIGNED; 3070 nfsm_clget; 3071 *tl = txdr_unsigned(dp->d_fileno); 3072 bp += NFSX_UNSIGNED; 3073 nfsm_clget; 3074 *tl = txdr_unsigned(nlen); 3075 bp += NFSX_UNSIGNED; 3076 3077 /* And loop around copying the name */ 3078 xfer = nlen; 3079 cp = dp->d_name; 3080 while (xfer > 0) { 3081 nfsm_clget; 3082 if ((bp + xfer) > be) 3083 tsiz = be - bp; 3084 else 3085 tsiz = xfer; 3086 memcpy(bp, cp, tsiz); 3087 bp += tsiz; 3088 xfer -= tsiz; 3089 if (xfer > 0) 3090 cp += tsiz; 3091 } 3092 /* And null pad to an int32_t boundary */ 3093 for (i = 0; i < rem; i++) 3094 *bp++ = '\0'; 3095 3096 /* 3097 * Now copy the flrep structure out. 3098 */ 3099 xfer = sizeof (struct flrep); 3100 cp = (caddr_t)&fl; 3101 while (xfer > 0) { 3102 nfsm_clget; 3103 if ((bp + xfer) > be) 3104 tsiz = be - bp; 3105 else 3106 tsiz = xfer; 3107 memcpy(bp, cp, tsiz); 3108 bp += tsiz; 3109 xfer -= tsiz; 3110 if (xfer > 0) 3111 cp += tsiz; 3112 } 3113 } 3114 invalid: 3115 cpos += dp->d_reclen; 3116 dp = (struct dirent *)cpos; 3117 cookiep++; 3118 ncookies--; 3119 } 3120 vrele(vp); 3121 nfsm_clget; 3122 *tl = nfs_false; 3123 bp += NFSX_UNSIGNED; 3124 nfsm_clget; 3125 if (eofflag) 3126 *tl = nfs_true; 3127 else 3128 *tl = nfs_false; 3129 bp += NFSX_UNSIGNED; 3130 if (mp != mb) { 3131 if (bp < be) 3132 mp->m_len = bp - mtod(mp, caddr_t); 3133 } else 3134 mp->m_len += bp - bpos; 3135 free((caddr_t)cookies, M_TEMP); 3136 free((caddr_t)rbuf, M_TEMP); 3137 nfsm_srvdone; 3138 } 3139 3140 /* 3141 * nfs commit service 3142 */ 3143 int 3144 nfsrv_commit(nfsd, slp, lwp, mrq) 3145 struct nfsrv_descript *nfsd; 3146 struct nfssvc_sock *slp; 3147 struct lwp *lwp; 3148 struct mbuf **mrq; 3149 { 3150 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3151 struct mbuf *nam = nfsd->nd_nam; 3152 caddr_t dpos = nfsd->nd_dpos; 3153 struct ucred *cred = &nfsd->nd_cr; 3154 struct vattr bfor, aft; 3155 struct vnode *vp; 3156 nfsfh_t nfh; 3157 fhandle_t *fhp; 3158 u_int32_t *tl; 3159 int32_t t1; 3160 caddr_t bpos; 3161 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3162 uint32_t cnt; 3163 char *cp2; 3164 struct mbuf *mb, *mreq; 3165 u_quad_t frev, off, end; 3166 struct mount *mp = NULL; 3167 3168 fhp = &nfh.fh_generic; 3169 nfsm_srvmtofh(fhp); 3170 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 3171 return (ESTALE); 3172 vn_start_write(NULL, &mp, V_WAIT); 3173 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3174 3175 off = fxdr_hyper(tl); 3176 tl += 2; 3177 cnt = fxdr_unsigned(uint32_t, *tl); 3178 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3179 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3180 if (error) { 3181 nfsm_reply(2 * NFSX_UNSIGNED); 3182 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3183 vn_finished_write(mp, 0); 3184 return (0); 3185 } 3186 for_ret = VOP_GETATTR(vp, &bfor, cred, lwp); 3187 end = (cnt > 0) ? off + cnt : vp->v_size; 3188 if (end < off || end > vp->v_size) 3189 end = vp->v_size; 3190 if (off < vp->v_size) 3191 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp); 3192 /* else error == 0, from nfsrv_fhtovp() */ 3193 aft_ret = VOP_GETATTR(vp, &aft, cred, lwp); 3194 vput(vp); 3195 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3196 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3197 if (!error) { 3198 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3199 *tl++ = txdr_unsigned(boottime.tv_sec); 3200 *tl = txdr_unsigned(boottime.tv_usec); 3201 } else { 3202 vn_finished_write(mp, 0); 3203 return (0); 3204 } 3205 vn_finished_write(mp, 0); 3206 nfsm_srvdone; 3207 } 3208 3209 /* 3210 * nfs statfs service 3211 */ 3212 int 3213 nfsrv_statfs(nfsd, slp, lwp, mrq) 3214 struct nfsrv_descript *nfsd; 3215 struct nfssvc_sock *slp; 3216 struct lwp *lwp; 3217 struct mbuf **mrq; 3218 { 3219 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3220 struct mbuf *nam = nfsd->nd_nam; 3221 caddr_t dpos = nfsd->nd_dpos; 3222 struct ucred *cred = &nfsd->nd_cr; 3223 struct statvfs *sf; 3224 struct nfs_statfs *sfp; 3225 u_int32_t *tl; 3226 int32_t t1; 3227 caddr_t bpos; 3228 int error = 0, rdonly, cache = 0, getret = 1; 3229 int v3 = (nfsd->nd_flag & ND_NFSV3); 3230 char *cp2; 3231 struct mbuf *mb, *mreq; 3232 struct vnode *vp; 3233 struct vattr at; 3234 nfsfh_t nfh; 3235 fhandle_t *fhp; 3236 struct statvfs statvfs; 3237 u_quad_t frev, tval; 3238 3239 fhp = &nfh.fh_generic; 3240 nfsm_srvmtofh(fhp); 3241 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3242 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3243 if (error) { 3244 nfsm_reply(NFSX_UNSIGNED); 3245 nfsm_srvpostop_attr(getret, &at); 3246 return (0); 3247 } 3248 sf = &statvfs; 3249 error = VFS_STATVFS(vp->v_mount, sf, lwp); 3250 getret = VOP_GETATTR(vp, &at, cred, lwp); 3251 vput(vp); 3252 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3253 if (v3) 3254 nfsm_srvpostop_attr(getret, &at); 3255 if (error) 3256 return (0); 3257 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3258 if (v3) { 3259 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3260 txdr_hyper(tval, &sfp->sf_tbytes); 3261 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3262 txdr_hyper(tval, &sfp->sf_fbytes); 3263 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3264 txdr_hyper(tval, &sfp->sf_abytes); 3265 tval = (u_quad_t)sf->f_files; 3266 txdr_hyper(tval, &sfp->sf_tfiles); 3267 tval = (u_quad_t)sf->f_ffree; 3268 txdr_hyper(tval, &sfp->sf_ffiles); 3269 txdr_hyper(tval, &sfp->sf_afiles); 3270 sfp->sf_invarsec = 0; 3271 } else { 3272 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3273 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3274 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3275 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3276 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3277 } 3278 nfsm_srvdone; 3279 } 3280 3281 /* 3282 * nfs fsinfo service 3283 */ 3284 int 3285 nfsrv_fsinfo(nfsd, slp, lwp, mrq) 3286 struct nfsrv_descript *nfsd; 3287 struct nfssvc_sock *slp; 3288 struct lwp *lwp; 3289 struct mbuf **mrq; 3290 { 3291 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3292 struct mbuf *nam = nfsd->nd_nam; 3293 caddr_t dpos = nfsd->nd_dpos; 3294 struct ucred *cred = &nfsd->nd_cr; 3295 u_int32_t *tl; 3296 struct nfsv3_fsinfo *sip; 3297 int32_t t1; 3298 caddr_t bpos; 3299 int error = 0, rdonly, cache = 0, getret = 1; 3300 uint32_t maxdata; 3301 char *cp2; 3302 struct mbuf *mb, *mreq; 3303 struct vnode *vp; 3304 struct vattr at; 3305 nfsfh_t nfh; 3306 fhandle_t *fhp; 3307 u_quad_t frev, maxfsize; 3308 struct statvfs sb; 3309 3310 fhp = &nfh.fh_generic; 3311 nfsm_srvmtofh(fhp); 3312 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3313 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3314 if (error) { 3315 nfsm_reply(NFSX_UNSIGNED); 3316 nfsm_srvpostop_attr(getret, &at); 3317 return (0); 3318 } 3319 3320 /* XXX Try to make a guess on the max file size. */ 3321 VFS_STATVFS(vp->v_mount, &sb, (struct lwp *)0); 3322 maxfsize = (u_quad_t)0x80000000 * sb.f_frsize - 1; 3323 3324 getret = VOP_GETATTR(vp, &at, cred, lwp); 3325 vput(vp); 3326 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3327 nfsm_srvpostop_attr(getret, &at); 3328 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3329 3330 /* 3331 * XXX 3332 * There should be file system VFS OP(s) to get this information. 3333 * For now, assume ufs. 3334 */ 3335 if (slp->ns_so->so_type == SOCK_DGRAM) 3336 maxdata = NFS_MAXDGRAMDATA; 3337 else 3338 maxdata = NFS_MAXDATA; 3339 sip->fs_rtmax = txdr_unsigned(maxdata); 3340 sip->fs_rtpref = txdr_unsigned(maxdata); 3341 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3342 sip->fs_wtmax = txdr_unsigned(maxdata); 3343 sip->fs_wtpref = txdr_unsigned(maxdata); 3344 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3345 sip->fs_dtpref = txdr_unsigned(maxdata); 3346 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3347 sip->fs_timedelta.nfsv3_sec = 0; 3348 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3349 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3350 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3351 NFSV3FSINFO_CANSETTIME); 3352 nfsm_srvdone; 3353 } 3354 3355 /* 3356 * nfs pathconf service 3357 */ 3358 int 3359 nfsrv_pathconf(nfsd, slp, lwp, mrq) 3360 struct nfsrv_descript *nfsd; 3361 struct nfssvc_sock *slp; 3362 struct lwp *lwp; 3363 struct mbuf **mrq; 3364 { 3365 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3366 struct mbuf *nam = nfsd->nd_nam; 3367 caddr_t dpos = nfsd->nd_dpos; 3368 struct ucred *cred = &nfsd->nd_cr; 3369 u_int32_t *tl; 3370 struct nfsv3_pathconf *pc; 3371 int32_t t1; 3372 caddr_t bpos; 3373 int error = 0, rdonly, cache = 0, getret = 1; 3374 register_t linkmax, namemax, chownres, notrunc; 3375 char *cp2; 3376 struct mbuf *mb, *mreq; 3377 struct vnode *vp; 3378 struct vattr at; 3379 nfsfh_t nfh; 3380 fhandle_t *fhp; 3381 u_quad_t frev; 3382 3383 fhp = &nfh.fh_generic; 3384 nfsm_srvmtofh(fhp); 3385 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3386 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3387 if (error) { 3388 nfsm_reply(NFSX_UNSIGNED); 3389 nfsm_srvpostop_attr(getret, &at); 3390 return (0); 3391 } 3392 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3393 if (!error) 3394 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3395 if (!error) 3396 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3397 if (!error) 3398 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3399 getret = VOP_GETATTR(vp, &at, cred, lwp); 3400 vput(vp); 3401 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3402 nfsm_srvpostop_attr(getret, &at); 3403 if (error) 3404 return (0); 3405 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3406 3407 pc->pc_linkmax = txdr_unsigned(linkmax); 3408 pc->pc_namemax = txdr_unsigned(namemax); 3409 pc->pc_notrunc = txdr_unsigned(notrunc); 3410 pc->pc_chownrestricted = txdr_unsigned(chownres); 3411 3412 /* 3413 * These should probably be supported by VOP_PATHCONF(), but 3414 * until msdosfs is exportable (why would you want to?), the 3415 * Unix defaults should be ok. 3416 */ 3417 pc->pc_caseinsensitive = nfs_false; 3418 pc->pc_casepreserving = nfs_true; 3419 nfsm_srvdone; 3420 } 3421 3422 /* 3423 * Null operation, used by clients to ping server 3424 */ 3425 /* ARGSUSED */ 3426 int 3427 nfsrv_null(nfsd, slp, lwp, mrq) 3428 struct nfsrv_descript *nfsd; 3429 struct nfssvc_sock *slp; 3430 struct lwp *lwp; 3431 struct mbuf **mrq; 3432 { 3433 struct mbuf *mrep = nfsd->nd_mrep; 3434 caddr_t bpos; 3435 int error = NFSERR_RETVOID, cache = 0; 3436 struct mbuf *mb, *mreq; 3437 u_quad_t frev; 3438 3439 nfsm_reply(0); 3440 return (0); 3441 } 3442 3443 /* 3444 * No operation, used for obsolete procedures 3445 */ 3446 /* ARGSUSED */ 3447 int 3448 nfsrv_noop(nfsd, slp, lwp, mrq) 3449 struct nfsrv_descript *nfsd; 3450 struct nfssvc_sock *slp; 3451 struct lwp *lwp; 3452 struct mbuf **mrq; 3453 { 3454 struct mbuf *mrep = nfsd->nd_mrep; 3455 caddr_t bpos; 3456 int error, cache = 0; 3457 struct mbuf *mb, *mreq; 3458 u_quad_t frev; 3459 3460 if (nfsd->nd_repstat) 3461 error = nfsd->nd_repstat; 3462 else 3463 error = EPROCUNAVAIL; 3464 nfsm_reply(0); 3465 return (0); 3466 } 3467 3468 /* 3469 * Perform access checking for vnodes obtained from file handles that would 3470 * refer to files already opened by a Unix client. You cannot just use 3471 * vn_writechk() and VOP_ACCESS() for two reasons. 3472 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3473 * 2 - The owner is to be given access irrespective of mode bits for some 3474 * operations, so that processes that chmod after opening a file don't 3475 * break. I don't like this because it opens a security hole, but since 3476 * the nfs server opens a security hole the size of a barn door anyhow, 3477 * what the heck. 3478 * 3479 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3480 * will return EPERM instead of EACCESS. EPERM is always an error. 3481 */ 3482 int 3483 nfsrv_access(vp, flags, cred, rdonly, lwp, override) 3484 struct vnode *vp; 3485 int flags; 3486 struct ucred *cred; 3487 int rdonly; 3488 struct lwp *lwp; 3489 { 3490 struct vattr vattr; 3491 int error; 3492 if (flags & VWRITE) { 3493 /* Just vn_writechk() changed to check rdonly */ 3494 /* 3495 * Disallow write attempts on read-only file systems; 3496 * unless the file is a socket or a block or character 3497 * device resident on the file system. 3498 */ 3499 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3500 switch (vp->v_type) { 3501 case VREG: 3502 case VDIR: 3503 case VLNK: 3504 return (EROFS); 3505 default: 3506 break; 3507 } 3508 } 3509 3510 /* 3511 * If the vnode is in use as a process's text, 3512 * we can't allow writing. 3513 */ 3514 if (vp->v_flag & VTEXT) 3515 return (ETXTBSY); 3516 } 3517 error = VOP_GETATTR(vp, &vattr, cred, lwp); 3518 if (error) 3519 return (error); 3520 error = VOP_ACCESS(vp, flags, cred, lwp); 3521 /* 3522 * Allow certain operations for the owner (reads and writes 3523 * on files that are already open). 3524 */ 3525 if (override && error == EACCES && cred->cr_uid == vattr.va_uid) 3526 error = 0; 3527 return error; 3528 } 3529
Indexes created Tue Sep 23 21:09:50 GMT 2025