nfs_serv.c revision 1.105
1 /* $NetBSD: nfs_serv.c,v 1.105 2006/04/15 01:39:15 christos Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * nfs version 2 and 3 server calls to vnode ops 39 * - these routines generally have 3 phases 40 * 1 - break down and validate rpc request in mbuf list 41 * 2 - do the vnode ops for the request 42 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c) 43 * 3 - build the rpc reply in an mbuf list 44 * nb: 45 * - do not mix the phases, since the nfsm_?? macros can return failures 46 * on a bad rpc or similar and do not do any vrele() or vput()'s 47 * 48 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs 49 * error number iff error != 0 whereas 50 * returning an error from the server function implies a fatal error 51 * such as a badly constructed rpc request that should be dropped without 52 * a reply. 53 * For Version 3, nfsm_reply() does not return for the error case, since 54 * most version 3 rpcs return more than the status for error cases. 55 */ 56 57 #include <sys/cdefs.h> 58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.105 2006/04/15 01:39:15 christos Exp $"); 59 60 #include <sys/param.h> 61 #include <sys/systm.h> 62 #include <sys/proc.h> 63 #include <sys/file.h> 64 #include <sys/namei.h> 65 #include <sys/vnode.h> 66 #include <sys/mount.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/mbuf.h> 70 #include <sys/dirent.h> 71 #include <sys/stat.h> 72 #include <sys/kernel.h> 73 74 #include <uvm/uvm.h> 75 76 #include <nfs/nfsproto.h> 77 #include <nfs/rpcv2.h> 78 #include <nfs/nfs.h> 79 #include <nfs/xdr_subs.h> 80 #include <nfs/nfsm_subs.h> 81 #include <nfs/nqnfs.h> 82 #include <nfs/nfs_var.h> 83 84 /* Global vars */ 85 extern u_int32_t nfs_xdrneg1; 86 extern u_int32_t nfs_false, nfs_true; 87 extern enum vtype nv3tov_type[8]; 88 extern struct nfsstats nfsstats; 89 extern const nfstype nfsv2_type[9]; 90 extern const nfstype nfsv3_type[9]; 91 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000; 92 int nfsd_use_loan = 1; /* use page-loan for READ OP */ 93 94 /* 95 * nfs v3 access service 96 */ 97 int 98 nfsrv3_access(nfsd, slp, lwp, mrq) 99 struct nfsrv_descript *nfsd; 100 struct nfssvc_sock *slp; 101 struct lwp *lwp; 102 struct mbuf **mrq; 103 { 104 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 105 struct mbuf *nam = nfsd->nd_nam; 106 caddr_t dpos = nfsd->nd_dpos; 107 struct ucred *cred = &nfsd->nd_cr; 108 struct vnode *vp; 109 nfsfh_t nfh; 110 fhandle_t *fhp; 111 u_int32_t *tl; 112 int32_t t1; 113 caddr_t bpos; 114 int error = 0, rdonly, cache = 0, getret; 115 char *cp2; 116 struct mbuf *mb, *mreq; 117 struct vattr va; 118 u_long inmode, testmode, outmode; 119 u_quad_t frev; 120 121 fhp = &nfh.fh_generic; 122 nfsm_srvmtofh(fhp); 123 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 124 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, 125 (nfsd->nd_flag & ND_KERBAUTH), FALSE); 126 if (error) { 127 nfsm_reply(NFSX_UNSIGNED); 128 nfsm_srvpostop_attr(1, (struct vattr *)0); 129 return (0); 130 } 131 inmode = fxdr_unsigned(u_int32_t, *tl); 132 outmode = 0; 133 if ((inmode & NFSV3ACCESS_READ) && 134 nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0) 135 outmode |= NFSV3ACCESS_READ; 136 if (vp->v_type != VDIR) { 137 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 138 if (testmode && 139 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 140 outmode |= testmode; 141 if ((inmode & NFSV3ACCESS_EXECUTE) && 142 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 143 outmode |= NFSV3ACCESS_EXECUTE; 144 } else { 145 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 146 NFSV3ACCESS_DELETE); 147 if (testmode && 148 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 149 outmode |= testmode; 150 if ((inmode & NFSV3ACCESS_LOOKUP) && 151 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 152 outmode |= NFSV3ACCESS_LOOKUP; 153 } 154 getret = VOP_GETATTR(vp, &va, cred, lwp); 155 vput(vp); 156 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 157 nfsm_srvpostop_attr(getret, &va); 158 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 159 *tl = txdr_unsigned(outmode); 160 nfsm_srvdone; 161 } 162 163 /* 164 * nfs getattr service 165 */ 166 int 167 nfsrv_getattr(nfsd, slp, lwp, mrq) 168 struct nfsrv_descript *nfsd; 169 struct nfssvc_sock *slp; 170 struct lwp *lwp; 171 struct mbuf **mrq; 172 { 173 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 174 struct mbuf *nam = nfsd->nd_nam; 175 caddr_t dpos = nfsd->nd_dpos; 176 struct ucred *cred = &nfsd->nd_cr; 177 struct nfs_fattr *fp; 178 struct vattr va; 179 struct vnode *vp; 180 nfsfh_t nfh; 181 fhandle_t *fhp; 182 u_int32_t *tl; 183 int32_t t1; 184 caddr_t bpos; 185 int error = 0, rdonly, cache = 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 fhp = &nfh.fh_generic; 241 nfsm_srvmtofh(fhp); 242 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 243 return (ESTALE); 244 vn_start_write(NULL, &mp, V_WAIT); 245 VATTR_NULL(&va); 246 if (v3) { 247 nfsm_srvsattr(&va); 248 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 249 gcheck = fxdr_unsigned(int, *tl); 250 if (gcheck) { 251 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 252 fxdr_nfsv3time(tl, &guard); 253 } 254 } else { 255 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 256 /* 257 * Nah nah nah nah na nah 258 * There is a bug in the Sun client that puts 0xffff in the mode 259 * field of sattr when it should put in 0xffffffff. The u_short 260 * doesn't sign extend. 261 * --> check the low order 2 bytes for 0xffff 262 */ 263 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 264 va.va_mode = nfstov_mode(sp->sa_mode); 265 if (sp->sa_uid != nfs_xdrneg1) 266 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 267 if (sp->sa_gid != nfs_xdrneg1) 268 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 269 if (sp->sa_size != nfs_xdrneg1) 270 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 271 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 272 #ifdef notyet 273 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 274 #else 275 va.va_atime.tv_sec = 276 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 277 va.va_atime.tv_nsec = 0; 278 #endif 279 } 280 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 281 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 282 283 } 284 285 /* 286 * Now that we have all the fields, lets do it. 287 */ 288 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, 289 (nfsd->nd_flag & ND_KERBAUTH), FALSE); 290 if (error) { 291 nfsm_reply(2 * NFSX_UNSIGNED); 292 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 293 vn_finished_write(mp, 0); 294 return (0); 295 } 296 nqsrv_getl(vp, ND_WRITE); 297 if (v3) { 298 error = preat_ret = VOP_GETATTR(vp, &preat, cred, lwp); 299 if (!error && gcheck && 300 (preat.va_ctime.tv_sec != guard.tv_sec || 301 preat.va_ctime.tv_nsec != guard.tv_nsec)) 302 error = NFSERR_NOT_SYNC; 303 if (error) { 304 vput(vp); 305 nfsm_reply(NFSX_WCCDATA(v3)); 306 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 307 vn_finished_write(mp, 0); 308 return (0); 309 } 310 } 311 312 /* 313 * If the size is being changed write acces is required, otherwise 314 * just check for a read only file system. 315 */ 316 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 317 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 318 error = EROFS; 319 goto out; 320 } 321 } else { 322 if (vp->v_type == VDIR) { 323 error = EISDIR; 324 goto out; 325 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 326 lwp, 0)) != 0) 327 goto out; 328 } 329 error = VOP_SETATTR(vp, &va, cred, lwp); 330 postat_ret = VOP_GETATTR(vp, &va, cred, lwp); 331 if (!error) 332 error = postat_ret; 333 out: 334 vput(vp); 335 vn_finished_write(mp, 0); 336 nfsm_reply(NFSX_WCCORFATTR(v3)); 337 if (v3) { 338 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 339 return (0); 340 } else { 341 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 342 nfsm_srvfillattr(&va, fp); 343 } 344 nfsm_srvdone; 345 } 346 347 /* 348 * nfs lookup rpc 349 */ 350 int 351 nfsrv_lookup(nfsd, slp, lwp, mrq) 352 struct nfsrv_descript *nfsd; 353 struct nfssvc_sock *slp; 354 struct lwp *lwp; 355 struct mbuf **mrq; 356 { 357 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 358 struct mbuf *nam = nfsd->nd_nam; 359 caddr_t dpos = nfsd->nd_dpos; 360 struct ucred *cred = &nfsd->nd_cr; 361 struct nfs_fattr *fp; 362 struct nameidata nd, ind, *ndp = &nd; 363 struct vnode *vp, *dirp; 364 nfsfh_t nfh; 365 fhandle_t *fhp; 366 caddr_t cp; 367 u_int32_t *tl; 368 int32_t t1; 369 caddr_t bpos; 370 int error = 0, cache = 0, dirattr_ret = 1; 371 uint32_t len; 372 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 373 char *cp2; 374 struct mbuf *mb, *mreq; 375 struct vattr va, dirattr; 376 u_quad_t frev; 377 378 fhp = &nfh.fh_generic; 379 nfsm_srvmtofh(fhp); 380 nfsm_srvnamesiz(len); 381 382 pubflag = nfs_ispublicfh(fhp); 383 384 nd.ni_cnd.cn_cred = cred; 385 nd.ni_cnd.cn_nameiop = LOOKUP; 386 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART; 387 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 388 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 389 390 if (!error && pubflag) { 391 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) { 392 /* 393 * Setup call to lookup() to see if we can find 394 * the index file. Arguably, this doesn't belong 395 * in a kernel.. Ugh. 396 */ 397 ind = nd; 398 VOP_UNLOCK(nd.ni_vp, 0); 399 ind.ni_pathlen = strlen(nfs_pub.np_index); 400 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf = 401 nfs_pub.np_index; 402 ind.ni_startdir = nd.ni_vp; 403 VREF(ind.ni_startdir); 404 error = lookup(&ind); 405 if (!error) { 406 /* 407 * Found an index file. Get rid of 408 * the old references. 409 */ 410 if (dirp) 411 vrele(dirp); 412 dirp = nd.ni_vp; 413 vrele(nd.ni_startdir); 414 ndp = &ind; 415 } else 416 error = 0; 417 } 418 /* 419 * If the public filehandle was used, check that this lookup 420 * didn't result in a filehandle outside the publicly exported 421 * filesystem. 422 */ 423 424 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 425 vput(nd.ni_vp); 426 error = EPERM; 427 } 428 } 429 430 if (dirp) { 431 if (v3) 432 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred, lwp); 433 vrele(dirp); 434 } 435 436 if (error) { 437 nfsm_reply(NFSX_POSTOPATTR(v3)); 438 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 439 return (0); 440 } 441 442 nqsrv_getl(ndp->ni_startdir, ND_READ); 443 vrele(ndp->ni_startdir); 444 PNBUF_PUT(nd.ni_cnd.cn_pnbuf); 445 vp = ndp->ni_vp; 446 memset((caddr_t)fhp, 0, sizeof(nfh)); 447 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 448 error = VFS_VPTOFH(vp, &fhp->fh_fid); 449 if (!error) 450 error = VOP_GETATTR(vp, &va, cred, lwp); 451 vput(vp); 452 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 453 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3)); 454 if (error) { 455 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 456 return (0); 457 } 458 nfsm_srvfhtom(fhp, v3); 459 if (v3) { 460 nfsm_srvpostop_attr(0, &va); 461 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 462 } else { 463 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 464 nfsm_srvfillattr(&va, fp); 465 } 466 nfsm_srvdone; 467 } 468 469 /* 470 * nfs readlink service 471 */ 472 int 473 nfsrv_readlink(nfsd, slp, lwp, mrq) 474 struct nfsrv_descript *nfsd; 475 struct nfssvc_sock *slp; 476 struct lwp *lwp; 477 struct mbuf **mrq; 478 { 479 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 480 struct mbuf *nam = nfsd->nd_nam; 481 caddr_t dpos = nfsd->nd_dpos; 482 struct ucred *cred = &nfsd->nd_cr; 483 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 484 struct iovec *ivp = iv; 485 struct mbuf *mp; 486 u_int32_t *tl; 487 int32_t t1; 488 caddr_t bpos; 489 int error = 0, rdonly, cache = 0, i, padlen, getret; 490 uint32_t len; 491 int v3 = (nfsd->nd_flag & ND_NFSV3); 492 char *cp2; 493 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 494 struct vnode *vp; 495 struct vattr attr; 496 nfsfh_t nfh; 497 fhandle_t *fhp; 498 struct uio io, *uiop = &io; 499 u_quad_t frev; 500 501 fhp = &nfh.fh_generic; 502 nfsm_srvmtofh(fhp); 503 len = 0; 504 i = 0; 505 while (len < NFS_MAXPATHLEN) { 506 mp = m_get(M_WAIT, MT_DATA); 507 MCLAIM(mp, &nfs_mowner); 508 m_clget(mp, M_WAIT); 509 mp->m_len = NFSMSIZ(mp); 510 if (len == 0) 511 mp3 = mp2 = mp; 512 else { 513 mp2->m_next = mp; 514 mp2 = mp; 515 } 516 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 517 mp->m_len = NFS_MAXPATHLEN-len; 518 len = NFS_MAXPATHLEN; 519 } else 520 len += mp->m_len; 521 ivp->iov_base = mtod(mp, caddr_t); 522 ivp->iov_len = mp->m_len; 523 i++; 524 ivp++; 525 } 526 uiop->uio_iov = iv; 527 uiop->uio_iovcnt = i; 528 uiop->uio_offset = 0; 529 uiop->uio_resid = len; 530 uiop->uio_rw = UIO_READ; 531 UIO_SETUP_SYSSPACE(uiop); 532 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 533 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 534 if (error) { 535 m_freem(mp3); 536 nfsm_reply(2 * NFSX_UNSIGNED); 537 nfsm_srvpostop_attr(1, (struct vattr *)0); 538 return (0); 539 } 540 if (vp->v_type != VLNK) { 541 if (v3) 542 error = EINVAL; 543 else 544 error = ENXIO; 545 goto out; 546 } 547 nqsrv_getl(vp, ND_READ); 548 error = VOP_READLINK(vp, uiop, cred); 549 out: 550 getret = VOP_GETATTR(vp, &attr, cred, lwp); 551 vput(vp); 552 if (error) 553 m_freem(mp3); 554 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 555 if (v3) { 556 nfsm_srvpostop_attr(getret, &attr); 557 if (error) 558 return (0); 559 } 560 len -= uiop->uio_resid; 561 padlen = nfsm_padlen(len); 562 if (uiop->uio_resid || padlen) 563 nfs_zeropad(mp3, uiop->uio_resid, padlen); 564 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 565 *tl = txdr_unsigned(len); 566 mb->m_next = mp3; 567 nfsm_srvdone; 568 } 569 570 /* 571 * nfs read service 572 */ 573 int 574 nfsrv_read(nfsd, slp, lwp, mrq) 575 struct nfsrv_descript *nfsd; 576 struct nfssvc_sock *slp; 577 struct lwp *lwp; 578 struct mbuf **mrq; 579 { 580 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 581 struct mbuf *nam = nfsd->nd_nam; 582 caddr_t dpos = nfsd->nd_dpos; 583 struct ucred *cred = &nfsd->nd_cr; 584 struct mbuf *m; 585 struct nfs_fattr *fp; 586 u_int32_t *tl; 587 int32_t t1; 588 int i; 589 caddr_t bpos; 590 int error = 0, rdonly, cache = 0, getret; 591 int v3 = (nfsd->nd_flag & ND_NFSV3); 592 uint32_t reqlen, len, cnt, left; 593 int padlen; 594 char *cp2; 595 struct mbuf *mb, *mreq; 596 struct vnode *vp; 597 nfsfh_t nfh; 598 fhandle_t *fhp; 599 struct uio io, *uiop = &io; 600 struct vattr va; 601 off_t off; 602 u_quad_t frev; 603 604 fhp = &nfh.fh_generic; 605 nfsm_srvmtofh(fhp); 606 if (v3) { 607 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 608 off = fxdr_hyper(tl); 609 } else { 610 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 611 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 612 } 613 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 614 reqlen = fxdr_unsigned(uint32_t, *tl); 615 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 616 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 617 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 618 if (error) { 619 nfsm_reply(2 * NFSX_UNSIGNED); 620 nfsm_srvpostop_attr(1, (struct vattr *)0); 621 return (0); 622 } 623 if (vp->v_type != VREG) { 624 if (v3) 625 error = EINVAL; 626 else 627 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 628 } 629 if (!error) { 630 nqsrv_getl(vp, ND_READ); 631 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 632 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 633 } 634 getret = VOP_GETATTR(vp, &va, cred, lwp); 635 if (!error) 636 error = getret; 637 if (error) { 638 vput(vp); 639 nfsm_reply(NFSX_POSTOPATTR(v3)); 640 nfsm_srvpostop_attr(getret, &va); 641 return (0); 642 } 643 if (off >= va.va_size) 644 cnt = 0; 645 else if ((off + reqlen) > va.va_size) 646 cnt = va.va_size - off; 647 else 648 cnt = reqlen; 649 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 650 if (v3) { 651 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 652 *tl++ = nfs_true; 653 fp = (struct nfs_fattr *)tl; 654 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 655 } else { 656 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 657 fp = (struct nfs_fattr *)tl; 658 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 659 } 660 len = left = cnt; 661 if (cnt > 0) { 662 if (nfsd_use_loan) { 663 struct vm_page **pgpp; 664 voff_t pgoff = trunc_page(off); 665 int npages; 666 vaddr_t lva; 667 668 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 669 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 670 671 /* allocate kva for mbuf data */ 672 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 673 if (lva == 0) { 674 /* fall back to VOP_READ */ 675 goto loan_fail; 676 } 677 678 /* allocate mbuf */ 679 m = m_get(M_WAIT, MT_DATA); 680 MCLAIM(m, &nfs_mowner); 681 pgpp = m->m_ext.ext_pgs; 682 683 /* loan pages */ 684 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 685 pgpp); 686 if (error) { 687 sokvafree(lva, npages << PAGE_SHIFT); 688 m_free(m); 689 if (error == EBUSY) 690 goto loan_fail; 691 goto read_error; 692 } 693 694 /* associate kva to mbuf */ 695 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 696 cnt, M_MBUF, soloanfree, slp->ns_so); 697 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 698 m->m_len = cnt; 699 700 /* map pages */ 701 for (i = 0; i < npages; i++) { 702 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 703 VM_PROT_READ); 704 lva += PAGE_SIZE; 705 } 706 707 pmap_update(pmap_kernel()); 708 709 mb->m_next = m; 710 mb = m; 711 error = 0; 712 uiop->uio_resid = 0; 713 } else { 714 struct iovec *iv; 715 struct iovec *iv2; 716 struct mbuf *m2; 717 int siz; 718 loan_fail: 719 /* 720 * Generate the mbuf list with the uio_iov ref. to it. 721 */ 722 i = 0; 723 m = m2 = mb; 724 while (left > 0) { 725 siz = min(M_TRAILINGSPACE(m), left); 726 if (siz > 0) { 727 left -= siz; 728 i++; 729 } 730 if (left > 0) { 731 m = m_get(M_WAIT, MT_DATA); 732 MCLAIM(m, &nfs_mowner); 733 m_clget(m, M_WAIT); 734 m->m_len = 0; 735 m2->m_next = m; 736 m2 = m; 737 } 738 } 739 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 740 uiop->uio_iov = iv2 = iv; 741 m = mb; 742 left = cnt; 743 i = 0; 744 while (left > 0) { 745 if (m == NULL) 746 panic("nfsrv_read iov"); 747 siz = min(M_TRAILINGSPACE(m), left); 748 if (siz > 0) { 749 iv->iov_base = mtod(m, caddr_t) + 750 m->m_len; 751 iv->iov_len = siz; 752 m->m_len += siz; 753 left -= siz; 754 iv++; 755 i++; 756 } 757 m = m->m_next; 758 } 759 uiop->uio_iovcnt = i; 760 uiop->uio_offset = off; 761 uiop->uio_resid = cnt; 762 uiop->uio_rw = UIO_READ; 763 UIO_SETUP_SYSSPACE(uiop); 764 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 765 free((caddr_t)iv2, M_TEMP); 766 } 767 read_error: 768 if (error || (getret = VOP_GETATTR(vp, &va, cred, lwp)) != 0){ 769 if (!error) 770 error = getret; 771 m_freem(mreq); 772 vput(vp); 773 nfsm_reply(NFSX_POSTOPATTR(v3)); 774 nfsm_srvpostop_attr(getret, &va); 775 return (0); 776 } 777 } else { 778 uiop->uio_resid = 0; 779 } 780 vput(vp); 781 nfsm_srvfillattr(&va, fp); 782 len -= uiop->uio_resid; 783 padlen = nfsm_padlen(len); 784 if (uiop->uio_resid || padlen) 785 nfs_zeropad(mb, uiop->uio_resid, padlen); 786 if (v3) { 787 /* count */ 788 *tl++ = txdr_unsigned(len); 789 /* eof */ 790 if (off + len >= va.va_size) 791 *tl++ = nfs_true; 792 else 793 *tl++ = nfs_false; 794 } 795 *tl = txdr_unsigned(len); 796 nfsm_srvdone; 797 } 798 799 /* 800 * nfs write service 801 */ 802 int 803 nfsrv_write(nfsd, slp, lwp, mrq) 804 struct nfsrv_descript *nfsd; 805 struct nfssvc_sock *slp; 806 struct lwp *lwp; 807 struct mbuf **mrq; 808 { 809 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 810 struct mbuf *nam = nfsd->nd_nam; 811 caddr_t dpos = nfsd->nd_dpos; 812 struct ucred *cred = &nfsd->nd_cr; 813 struct iovec *ivp; 814 int i, cnt; 815 struct mbuf *mp; 816 struct nfs_fattr *fp; 817 struct iovec *iv; 818 struct vattr va, forat; 819 u_int32_t *tl; 820 int32_t t1; 821 caddr_t bpos; 822 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 823 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 824 int stable = NFSV3WRITE_FILESYNC; 825 int v3 = (nfsd->nd_flag & ND_NFSV3); 826 char *cp2; 827 struct mbuf *mb, *mreq; 828 struct vnode *vp; 829 nfsfh_t nfh; 830 fhandle_t *fhp; 831 struct uio io, *uiop = &io; 832 off_t off; 833 u_quad_t frev; 834 struct mount *mntp = NULL; 835 836 if (mrep == NULL) { 837 *mrq = NULL; 838 return (0); 839 } 840 fhp = &nfh.fh_generic; 841 nfsm_srvmtofh(fhp); 842 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 843 return (ESTALE); 844 vn_start_write(NULL, &mntp, V_WAIT); 845 if (v3) { 846 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 847 off = fxdr_hyper(tl); 848 tl += 3; 849 stable = fxdr_unsigned(int, *tl++); 850 } else { 851 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 852 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 853 tl += 2; 854 } 855 retlen = len = fxdr_unsigned(int32_t, *tl); 856 cnt = i = 0; 857 858 /* 859 * For NFS Version 2, it is not obvious what a write of zero length 860 * should do, but I might as well be consistent with Version 3, 861 * which is to return ok so long as there are no permission problems. 862 */ 863 if (len > 0) { 864 zeroing = 1; 865 mp = mrep; 866 while (mp) { 867 if (mp == md) { 868 zeroing = 0; 869 adjust = dpos - mtod(mp, caddr_t); 870 mp->m_len -= adjust; 871 if (mp->m_len > 0 && adjust > 0) 872 NFSMADV(mp, adjust); 873 } 874 if (zeroing) 875 mp->m_len = 0; 876 else if (mp->m_len > 0) { 877 i += mp->m_len; 878 if (i > len) { 879 mp->m_len -= (i - len); 880 zeroing = 1; 881 } 882 if (mp->m_len > 0) 883 cnt++; 884 } 885 mp = mp->m_next; 886 } 887 } 888 if (len > NFS_MAXDATA || len < 0 || i < len) { 889 error = EIO; 890 nfsm_reply(2 * NFSX_UNSIGNED); 891 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 892 vn_finished_write(mntp, 0); 893 return (0); 894 } 895 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 896 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 897 if (error) { 898 nfsm_reply(2 * NFSX_UNSIGNED); 899 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 900 vn_finished_write(mntp, 0); 901 return (0); 902 } 903 if (v3) 904 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 905 if (vp->v_type != VREG) { 906 if (v3) 907 error = EINVAL; 908 else 909 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 910 } 911 if (!error) { 912 nqsrv_getl(vp, ND_WRITE); 913 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 914 } 915 if (error) { 916 vput(vp); 917 nfsm_reply(NFSX_WCCDATA(v3)); 918 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 919 vn_finished_write(mntp, 0); 920 return (0); 921 } 922 923 if (len > 0) { 924 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 925 uiop->uio_iov = iv = ivp; 926 uiop->uio_iovcnt = cnt; 927 mp = mrep; 928 while (mp) { 929 if (mp->m_len > 0) { 930 ivp->iov_base = mtod(mp, caddr_t); 931 ivp->iov_len = mp->m_len; 932 ivp++; 933 } 934 mp = mp->m_next; 935 } 936 937 /* 938 * XXX 939 * The IO_METASYNC flag indicates that all metadata (and not 940 * just enough to ensure data integrity) must be written to 941 * stable storage synchronously. 942 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 943 */ 944 if (stable == NFSV3WRITE_UNSTABLE) 945 ioflags = IO_NODELOCKED; 946 else if (stable == NFSV3WRITE_DATASYNC) 947 ioflags = (IO_SYNC | IO_NODELOCKED); 948 else 949 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 950 uiop->uio_resid = len; 951 uiop->uio_rw = UIO_WRITE; 952 uiop->uio_offset = off; 953 UIO_SETUP_SYSSPACE(uiop); 954 error = VOP_WRITE(vp, uiop, ioflags, cred); 955 nfsstats.srvvop_writes++; 956 free(iv, M_TEMP); 957 } 958 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 959 vput(vp); 960 vn_finished_write(mntp, 0); 961 if (!error) 962 error = aftat_ret; 963 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 964 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 965 if (v3) { 966 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 967 if (error) 968 return (0); 969 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 970 *tl++ = txdr_unsigned(retlen); 971 if (stable == NFSV3WRITE_UNSTABLE) 972 *tl++ = txdr_unsigned(stable); 973 else 974 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 975 /* 976 * Actually, there is no need to txdr these fields, 977 * but it may make the values more human readable, 978 * for debugging purposes. 979 */ 980 *tl++ = txdr_unsigned(boottime.tv_sec); 981 *tl = txdr_unsigned(boottime.tv_usec); 982 } else { 983 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 984 nfsm_srvfillattr(&va, fp); 985 } 986 nfsm_srvdone; 987 } 988 989 /* 990 * NFS write service with write gathering support. Called when 991 * nfsrvw_procrastinate > 0. 992 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 993 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 994 * Jan. 1994. 995 */ 996 int 997 nfsrv_writegather(ndp, slp, lwp, mrq) 998 struct nfsrv_descript **ndp; 999 struct nfssvc_sock *slp; 1000 struct lwp *lwp; 1001 struct mbuf **mrq; 1002 { 1003 struct iovec *ivp; 1004 struct mbuf *mp; 1005 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1006 struct nfs_fattr *fp; 1007 int i = 0; 1008 struct iovec *iov; 1009 struct nfsrvw_delayhash *wpp; 1010 struct ucred *cred; 1011 struct vattr va, forat; 1012 u_int32_t *tl; 1013 int32_t t1; 1014 caddr_t bpos, dpos; 1015 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1016 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing; 1017 char *cp2; 1018 struct mbuf *mb, *mreq, *mrep, *md; 1019 struct vnode *vp; 1020 struct uio io, *uiop = &io; 1021 u_quad_t frev, cur_usec; 1022 struct mount *mntp = NULL; 1023 1024 *mrq = NULL; 1025 if (*ndp) { 1026 nfsd = *ndp; 1027 *ndp = NULL; 1028 mrep = nfsd->nd_mrep; 1029 md = nfsd->nd_md; 1030 dpos = nfsd->nd_dpos; 1031 cred = &nfsd->nd_cr; 1032 v3 = (nfsd->nd_flag & ND_NFSV3); 1033 LIST_INIT(&nfsd->nd_coalesce); 1034 nfsd->nd_mreq = NULL; 1035 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1036 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1037 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1038 1039 /* 1040 * Now, get the write header.. 1041 */ 1042 nfsm_srvmtofh(&nfsd->nd_fh); 1043 if (v3) { 1044 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1045 nfsd->nd_off = fxdr_hyper(tl); 1046 tl += 3; 1047 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1048 } else { 1049 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1050 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1051 tl += 2; 1052 } 1053 len = fxdr_unsigned(int32_t, *tl); 1054 nfsd->nd_len = len; 1055 nfsd->nd_eoff = nfsd->nd_off + len; 1056 1057 /* 1058 * Trim the header out of the mbuf list and trim off any trailing 1059 * junk so that the mbuf list has only the write data. 1060 */ 1061 zeroing = 1; 1062 i = 0; 1063 mp = mrep; 1064 while (mp) { 1065 if (mp == md) { 1066 zeroing = 0; 1067 adjust = dpos - mtod(mp, caddr_t); 1068 mp->m_len -= adjust; 1069 if (mp->m_len > 0 && adjust > 0) 1070 NFSMADV(mp, adjust); 1071 } 1072 if (zeroing) 1073 mp->m_len = 0; 1074 else { 1075 i += mp->m_len; 1076 if (i > len) { 1077 mp->m_len -= (i - len); 1078 zeroing = 1; 1079 } 1080 } 1081 mp = mp->m_next; 1082 } 1083 if (len > NFS_MAXDATA || len < 0 || i < len) { 1084 nfsmout: 1085 m_freem(mrep); 1086 error = EIO; 1087 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1088 if (v3) 1089 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1090 nfsd->nd_mreq = mreq; 1091 nfsd->nd_mrep = NULL; 1092 nfsd->nd_time = 0; 1093 } 1094 1095 /* 1096 * Add this entry to the hash and time queues. 1097 */ 1098 s = splsoftclock(); 1099 owp = NULL; 1100 wp = LIST_FIRST(&slp->ns_tq); 1101 while (wp && wp->nd_time < nfsd->nd_time) { 1102 owp = wp; 1103 wp = LIST_NEXT(wp, nd_tq); 1104 } 1105 if (owp) { 1106 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1107 } else { 1108 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1109 } 1110 if (nfsd->nd_mrep) { 1111 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data); 1112 owp = NULL; 1113 wp = LIST_FIRST(wpp); 1114 while (wp && 1115 memcmp(&nfsd->nd_fh, &wp->nd_fh, NFSX_V3FH)) { 1116 owp = wp; 1117 wp = LIST_NEXT(wp, nd_hash); 1118 } 1119 while (wp && wp->nd_off < nfsd->nd_off && 1120 !memcmp(&nfsd->nd_fh, &wp->nd_fh, NFSX_V3FH)) { 1121 owp = wp; 1122 wp = LIST_NEXT(wp, nd_hash); 1123 } 1124 if (owp) { 1125 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1126 1127 /* 1128 * Search the hash list for overlapping entries and 1129 * coalesce. 1130 */ 1131 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1132 wp = LIST_NEXT(nfsd, nd_hash); 1133 if (NFSW_SAMECRED(owp, nfsd)) 1134 nfsrvw_coalesce(owp, nfsd); 1135 } 1136 } else { 1137 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1138 } 1139 } 1140 splx(s); 1141 } 1142 1143 /* 1144 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1145 * and generate the associated reply mbuf list(s). 1146 */ 1147 loop1: 1148 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec; 1149 s = splsoftclock(); 1150 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1151 owp = LIST_NEXT(nfsd, nd_tq); 1152 if (nfsd->nd_time > cur_usec) 1153 break; 1154 if (nfsd->nd_mreq) 1155 continue; 1156 LIST_REMOVE(nfsd, nd_tq); 1157 LIST_REMOVE(nfsd, nd_hash); 1158 splx(s); 1159 mrep = nfsd->nd_mrep; 1160 nfsd->nd_mrep = NULL; 1161 cred = &nfsd->nd_cr; 1162 v3 = (nfsd->nd_flag & ND_NFSV3); 1163 forat_ret = aftat_ret = 1; 1164 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1165 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1166 FALSE); 1167 if (!error) { 1168 if (v3) 1169 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 1170 if (vp->v_type != VREG) { 1171 if (v3) 1172 error = EINVAL; 1173 else 1174 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1175 } 1176 } else 1177 vp = NULL; 1178 if (!error) { 1179 nqsrv_getl(vp, ND_WRITE); 1180 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1181 } 1182 1183 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1184 ioflags = IO_NODELOCKED; 1185 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1186 ioflags = (IO_SYNC | IO_NODELOCKED); 1187 else 1188 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1189 uiop->uio_rw = UIO_WRITE; 1190 uiop->uio_offset = nfsd->nd_off; 1191 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1192 UIO_SETUP_SYSSPACE(uiop); 1193 if (uiop->uio_resid > 0) { 1194 mp = mrep; 1195 i = 0; 1196 while (mp) { 1197 if (mp->m_len > 0) 1198 i++; 1199 mp = mp->m_next; 1200 } 1201 uiop->uio_iovcnt = i; 1202 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1203 uiop->uio_iov = ivp = iov; 1204 mp = mrep; 1205 while (mp) { 1206 if (mp->m_len > 0) { 1207 ivp->iov_base = mtod(mp, caddr_t); 1208 ivp->iov_len = mp->m_len; 1209 ivp++; 1210 } 1211 mp = mp->m_next; 1212 } 1213 if (!error) { 1214 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) { 1215 VOP_UNLOCK(vp, 0); 1216 vn_start_write(NULL, &mntp, V_WAIT); 1217 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1218 } 1219 if (!error) { 1220 error = VOP_WRITE(vp, uiop, ioflags, cred); 1221 nfsstats.srvvop_writes++; 1222 vn_finished_write(mntp, 0); 1223 } 1224 } 1225 free((caddr_t)iov, M_TEMP); 1226 } 1227 m_freem(mrep); 1228 if (vp) { 1229 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 1230 vput(vp); 1231 } 1232 1233 /* 1234 * Loop around generating replies for all write rpcs that have 1235 * now been completed. 1236 */ 1237 swp = nfsd; 1238 do { 1239 if (error) { 1240 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1241 if (v3) { 1242 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1243 } 1244 } else { 1245 nfsm_writereply(NFSX_PREOPATTR(v3) + 1246 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1247 NFSX_WRITEVERF(v3), v3); 1248 if (v3) { 1249 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1250 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1251 *tl++ = txdr_unsigned(nfsd->nd_len); 1252 *tl++ = txdr_unsigned(swp->nd_stable); 1253 /* 1254 * Actually, there is no need to txdr these fields, 1255 * but it may make the values more human readable, 1256 * for debugging purposes. 1257 */ 1258 *tl++ = txdr_unsigned(boottime.tv_sec); 1259 *tl = txdr_unsigned(boottime.tv_usec); 1260 } else { 1261 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1262 nfsm_srvfillattr(&va, fp); 1263 } 1264 } 1265 nfsd->nd_mreq = mreq; 1266 if (nfsd->nd_mrep) 1267 panic("nfsrv_write: nd_mrep not free"); 1268 1269 /* 1270 * Done. Put it at the head of the timer queue so that 1271 * the final phase can return the reply. 1272 */ 1273 s = splsoftclock(); 1274 if (nfsd != swp) { 1275 nfsd->nd_time = 0; 1276 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1277 } 1278 nfsd = LIST_FIRST(&swp->nd_coalesce); 1279 if (nfsd) { 1280 LIST_REMOVE(nfsd, nd_tq); 1281 } 1282 splx(s); 1283 } while (nfsd); 1284 s = splsoftclock(); 1285 swp->nd_time = 0; 1286 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1287 splx(s); 1288 goto loop1; 1289 } 1290 splx(s); 1291 1292 /* 1293 * Search for a reply to return. 1294 */ 1295 s = splsoftclock(); 1296 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1297 if (nfsd->nd_mreq) { 1298 LIST_REMOVE(nfsd, nd_tq); 1299 *mrq = nfsd->nd_mreq; 1300 *ndp = nfsd; 1301 break; 1302 } 1303 } 1304 splx(s); 1305 return (0); 1306 } 1307 1308 /* 1309 * Coalesce the write request nfsd into owp. To do this we must: 1310 * - remove nfsd from the queues 1311 * - merge nfsd->nd_mrep into owp->nd_mrep 1312 * - update the nd_eoff and nd_stable for owp 1313 * - put nfsd on owp's nd_coalesce list 1314 * NB: Must be called at splsoftclock(). 1315 */ 1316 void 1317 nfsrvw_coalesce(owp, nfsd) 1318 struct nfsrv_descript *owp; 1319 struct nfsrv_descript *nfsd; 1320 { 1321 int overlap; 1322 struct mbuf *mp; 1323 struct nfsrv_descript *m; 1324 1325 LIST_REMOVE(nfsd, nd_hash); 1326 LIST_REMOVE(nfsd, nd_tq); 1327 if (owp->nd_eoff < nfsd->nd_eoff) { 1328 overlap = owp->nd_eoff - nfsd->nd_off; 1329 if (overlap < 0) 1330 panic("nfsrv_coalesce: bad off"); 1331 if (overlap > 0) 1332 m_adj(nfsd->nd_mrep, overlap); 1333 mp = owp->nd_mrep; 1334 while (mp->m_next) 1335 mp = mp->m_next; 1336 mp->m_next = nfsd->nd_mrep; 1337 owp->nd_eoff = nfsd->nd_eoff; 1338 } else 1339 m_freem(nfsd->nd_mrep); 1340 nfsd->nd_mrep = NULL; 1341 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1342 owp->nd_stable = NFSV3WRITE_FILESYNC; 1343 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1344 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1345 owp->nd_stable = NFSV3WRITE_DATASYNC; 1346 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1347 /* 1348 * nfsd might hold coalesce elements! Move them to owp. 1349 * Otherwise, requests may be lost and clients will be stuck. 1350 */ 1351 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1352 LIST_REMOVE(m, nd_tq); 1353 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1354 } 1355 } 1356 1357 /* 1358 * nfs create service 1359 * now does a truncate to 0 length via. setattr if it already exists 1360 */ 1361 int 1362 nfsrv_create(nfsd, slp, lwp, mrq) 1363 struct nfsrv_descript *nfsd; 1364 struct nfssvc_sock *slp; 1365 struct lwp *lwp; 1366 struct mbuf **mrq; 1367 { 1368 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1369 struct mbuf *nam = nfsd->nd_nam; 1370 caddr_t dpos = nfsd->nd_dpos; 1371 struct ucred *cred = &nfsd->nd_cr; 1372 struct nfs_fattr *fp; 1373 struct vattr va, dirfor, diraft; 1374 struct nfsv2_sattr *sp; 1375 u_int32_t *tl; 1376 struct nameidata nd; 1377 caddr_t cp; 1378 int32_t t1; 1379 caddr_t bpos; 1380 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1381 int rdev = 0; 1382 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1383 char *cp2; 1384 struct mbuf *mb, *mreq; 1385 struct vnode *vp = NULL, *dirp = NULL; 1386 nfsfh_t nfh; 1387 fhandle_t *fhp; 1388 u_quad_t frev, tempsize; 1389 u_char cverf[NFSX_V3CREATEVERF]; 1390 struct mount *mp = NULL; 1391 1392 nd.ni_cnd.cn_nameiop = 0; 1393 fhp = &nfh.fh_generic; 1394 nfsm_srvmtofh(fhp); 1395 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1396 return (ESTALE); 1397 vn_start_write(NULL, &mp, V_WAIT); 1398 nfsm_srvnamesiz(len); 1399 nd.ni_cnd.cn_cred = cred; 1400 nd.ni_cnd.cn_nameiop = CREATE; 1401 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1402 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1403 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1404 if (dirp) { 1405 if (v3) 1406 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1407 else { 1408 vrele(dirp); 1409 dirp = (struct vnode *)0; 1410 } 1411 } 1412 if (error) { 1413 nfsm_reply(NFSX_WCCDATA(v3)); 1414 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1415 if (dirp) 1416 vrele(dirp); 1417 vn_finished_write(mp, 0); 1418 return (0); 1419 } 1420 VATTR_NULL(&va); 1421 if (v3) { 1422 va.va_mode = 0; 1423 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1424 how = fxdr_unsigned(int, *tl); 1425 switch (how) { 1426 case NFSV3CREATE_GUARDED: 1427 if (nd.ni_vp) { 1428 error = EEXIST; 1429 break; 1430 } 1431 case NFSV3CREATE_UNCHECKED: 1432 nfsm_srvsattr(&va); 1433 break; 1434 case NFSV3CREATE_EXCLUSIVE: 1435 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF); 1436 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1437 exclusive_flag = 1; 1438 break; 1439 }; 1440 va.va_type = VREG; 1441 } else { 1442 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1443 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1444 if (va.va_type == VNON) 1445 va.va_type = VREG; 1446 va.va_mode = nfstov_mode(sp->sa_mode); 1447 switch (va.va_type) { 1448 case VREG: 1449 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1450 if (tsize != -1) 1451 va.va_size = (u_quad_t)tsize; 1452 break; 1453 case VCHR: 1454 case VBLK: 1455 case VFIFO: 1456 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1457 break; 1458 default: 1459 break; 1460 }; 1461 } 1462 1463 /* 1464 * Iff doesn't exist, create it 1465 * otherwise just truncate to 0 length 1466 * should I set the mode too ?? 1467 */ 1468 if (nd.ni_vp == NULL) { 1469 if (va.va_type == VREG || va.va_type == VSOCK) { 1470 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1471 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1472 if (!error) { 1473 if (exclusive_flag) { 1474 exclusive_flag = 0; 1475 VATTR_NULL(&va); 1476 /* 1477 * XXX 1478 * assuming NFSX_V3CREATEVERF 1479 * == sizeof(nfstime3) 1480 */ 1481 fxdr_nfsv3time(cverf, &va.va_atime); 1482 error = VOP_SETATTR(nd.ni_vp, &va, cred, 1483 lwp); 1484 } 1485 } 1486 } else if (va.va_type == VCHR || va.va_type == VBLK || 1487 va.va_type == VFIFO) { 1488 if (va.va_type == VCHR && rdev == 0xffffffff) 1489 va.va_type = VFIFO; 1490 if (va.va_type != VFIFO && 1491 (error = suser(cred, (u_short *)0))) { 1492 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1493 vput(nd.ni_dvp); 1494 nfsm_reply(0); 1495 vn_finished_write(mp, 0); 1496 return (error); 1497 } else 1498 va.va_rdev = (dev_t)rdev; 1499 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1500 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1501 &va); 1502 if (error) { 1503 nfsm_reply(0); 1504 } 1505 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1506 vrele(nd.ni_dvp); 1507 vput(nd.ni_vp); 1508 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1509 error = EINVAL; 1510 nfsm_reply(0); 1511 } 1512 } else { 1513 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1514 vput(nd.ni_dvp); 1515 error = ENXIO; 1516 } 1517 vp = nd.ni_vp; 1518 } else { 1519 vp = nd.ni_vp; 1520 if (nd.ni_dvp == vp) 1521 vrele(nd.ni_dvp); 1522 else 1523 vput(nd.ni_dvp); 1524 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1525 if (!error && va.va_size != -1) { 1526 error = nfsrv_access(vp, VWRITE, cred, 1527 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1528 if (!error) { 1529 nqsrv_getl(vp, ND_WRITE); 1530 tempsize = va.va_size; 1531 VATTR_NULL(&va); 1532 va.va_size = tempsize; 1533 error = VOP_SETATTR(vp, &va, cred, lwp); 1534 } 1535 } 1536 if (error) 1537 vput(vp); 1538 } 1539 if (!error) { 1540 memset((caddr_t)fhp, 0, sizeof(nfh)); 1541 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1542 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1543 if (!error) 1544 error = VOP_GETATTR(vp, &va, cred, lwp); 1545 vput(vp); 1546 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1547 } 1548 if (v3) { 1549 if (exclusive_flag && !error) { 1550 /* 1551 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1552 */ 1553 char oldverf[NFSX_V3CREATEVERF]; 1554 1555 txdr_nfsv3time(&va.va_atime, oldverf); 1556 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1557 error = EEXIST; 1558 } 1559 if (dirp) { 1560 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1561 vrele(dirp); 1562 } 1563 } 1564 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1565 if (v3) { 1566 if (!error) { 1567 nfsm_srvpostop_fh(fhp); 1568 nfsm_srvpostop_attr(0, &va); 1569 } 1570 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1571 } else { 1572 nfsm_srvfhtom(fhp, v3); 1573 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1574 nfsm_srvfillattr(&va, fp); 1575 } 1576 vn_finished_write(mp, 0); 1577 return (0); 1578 nfsmout: 1579 if (dirp) 1580 vrele(dirp); 1581 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1582 if (nd.ni_dvp == nd.ni_vp) 1583 vrele(nd.ni_dvp); 1584 else 1585 vput(nd.ni_dvp); 1586 if (nd.ni_vp) 1587 vput(nd.ni_vp); 1588 vn_finished_write(mp, 0); 1589 return (error); 1590 } 1591 1592 /* 1593 * nfs v3 mknod service 1594 */ 1595 int 1596 nfsrv_mknod(nfsd, slp, lwp, mrq) 1597 struct nfsrv_descript *nfsd; 1598 struct nfssvc_sock *slp; 1599 struct lwp *lwp; 1600 struct mbuf **mrq; 1601 { 1602 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1603 struct mbuf *nam = nfsd->nd_nam; 1604 caddr_t dpos = nfsd->nd_dpos; 1605 struct ucred *cred = &nfsd->nd_cr; 1606 struct vattr va, dirfor, diraft; 1607 u_int32_t *tl; 1608 struct nameidata nd; 1609 int32_t t1; 1610 caddr_t bpos; 1611 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1612 u_int32_t major, minor; 1613 enum vtype vtyp; 1614 char *cp2; 1615 struct mbuf *mb, *mreq; 1616 struct vnode *vp, *dirp = (struct vnode *)0; 1617 nfsfh_t nfh; 1618 fhandle_t *fhp; 1619 u_quad_t frev; 1620 struct mount *mp = NULL; 1621 1622 nd.ni_cnd.cn_nameiop = 0; 1623 fhp = &nfh.fh_generic; 1624 nfsm_srvmtofh(fhp); 1625 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1626 return (ESTALE); 1627 vn_start_write(NULL, &mp, V_WAIT); 1628 nfsm_srvnamesiz(len); 1629 nd.ni_cnd.cn_cred = cred; 1630 nd.ni_cnd.cn_nameiop = CREATE; 1631 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1632 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1633 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1634 if (dirp) 1635 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1636 if (error) { 1637 nfsm_reply(NFSX_WCCDATA(1)); 1638 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1639 if (dirp) 1640 vrele(dirp); 1641 vn_finished_write(mp, 0); 1642 return (0); 1643 } 1644 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1645 vtyp = nfsv3tov_type(*tl); 1646 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1647 error = NFSERR_BADTYPE; 1648 goto abort; 1649 } 1650 VATTR_NULL(&va); 1651 va.va_mode = 0; 1652 nfsm_srvsattr(&va); 1653 if (vtyp == VCHR || vtyp == VBLK) { 1654 dev_t rdev; 1655 1656 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1657 major = fxdr_unsigned(u_int32_t, *tl++); 1658 minor = fxdr_unsigned(u_int32_t, *tl); 1659 rdev = makedev(major, minor); 1660 if (major(rdev) != major || minor(rdev) != minor) { 1661 error = EINVAL; 1662 goto abort; 1663 } 1664 va.va_rdev = rdev; 1665 } 1666 1667 /* 1668 * Iff doesn't exist, create it. 1669 */ 1670 if (nd.ni_vp) { 1671 error = EEXIST; 1672 abort: 1673 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1674 if (nd.ni_dvp == nd.ni_vp) 1675 vrele(nd.ni_dvp); 1676 else 1677 vput(nd.ni_dvp); 1678 if (nd.ni_vp) 1679 vput(nd.ni_vp); 1680 goto out; 1681 } 1682 va.va_type = vtyp; 1683 if (vtyp == VSOCK) { 1684 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1685 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1686 } else { 1687 if (va.va_type != VFIFO && 1688 (error = suser(cred, (u_short *)0))) { 1689 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1690 vput(nd.ni_dvp); 1691 goto out; 1692 } 1693 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1694 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1695 if (error) { 1696 goto out; 1697 } 1698 if (error) 1699 goto out; 1700 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1701 vput(nd.ni_vp); 1702 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1703 error = EINVAL; 1704 } 1705 } 1706 out: 1707 vp = nd.ni_vp; 1708 if (!error) { 1709 memset((caddr_t)fhp, 0, sizeof(nfh)); 1710 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1711 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1712 if (!error) 1713 error = VOP_GETATTR(vp, &va, cred, lwp); 1714 vput(vp); 1715 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1716 } 1717 if (dirp) { 1718 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1719 vrele(dirp); 1720 } 1721 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1)); 1722 if (!error) { 1723 nfsm_srvpostop_fh(fhp); 1724 nfsm_srvpostop_attr(0, &va); 1725 } 1726 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1727 vn_finished_write(mp, 0); 1728 return (0); 1729 nfsmout: 1730 if (dirp) 1731 vrele(dirp); 1732 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1733 if (nd.ni_dvp == nd.ni_vp) 1734 vrele(nd.ni_dvp); 1735 else 1736 vput(nd.ni_dvp); 1737 if (nd.ni_vp) 1738 vput(nd.ni_vp); 1739 vn_finished_write(mp, 0); 1740 return (error); 1741 } 1742 1743 /* 1744 * nfs remove service 1745 */ 1746 int 1747 nfsrv_remove(nfsd, slp, lwp, mrq) 1748 struct nfsrv_descript *nfsd; 1749 struct nfssvc_sock *slp; 1750 struct lwp *lwp; 1751 struct mbuf **mrq; 1752 { 1753 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1754 struct mbuf *nam = nfsd->nd_nam; 1755 caddr_t dpos = nfsd->nd_dpos; 1756 struct ucred *cred = &nfsd->nd_cr; 1757 struct nameidata nd; 1758 u_int32_t *tl; 1759 int32_t t1; 1760 caddr_t bpos; 1761 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1762 int v3 = (nfsd->nd_flag & ND_NFSV3); 1763 char *cp2; 1764 struct mbuf *mb, *mreq; 1765 struct vnode *vp, *dirp; 1766 struct vattr dirfor, diraft; 1767 nfsfh_t nfh; 1768 fhandle_t *fhp; 1769 u_quad_t frev; 1770 struct mount *mp = NULL; 1771 1772 #ifndef nolint 1773 vp = (struct vnode *)0; 1774 #endif 1775 fhp = &nfh.fh_generic; 1776 nfsm_srvmtofh(fhp); 1777 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1778 return (ESTALE); 1779 vn_start_write(NULL, &mp, V_WAIT); 1780 nfsm_srvnamesiz(len); 1781 nd.ni_cnd.cn_cred = cred; 1782 nd.ni_cnd.cn_nameiop = DELETE; 1783 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1784 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1785 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1786 if (dirp) { 1787 if (v3) 1788 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1789 else 1790 vrele(dirp); 1791 } 1792 if (!error) { 1793 vp = nd.ni_vp; 1794 if (vp->v_type == VDIR && 1795 (error = suser(cred, (u_short *)0)) != 0) 1796 goto out; 1797 /* 1798 * The root of a mounted filesystem cannot be deleted. 1799 */ 1800 if (vp->v_flag & VROOT) { 1801 error = EBUSY; 1802 goto out; 1803 } 1804 out: 1805 if (!error) { 1806 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1807 nqsrv_getl(vp, ND_WRITE); 1808 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1809 } else { 1810 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1811 if (nd.ni_dvp == vp) 1812 vrele(nd.ni_dvp); 1813 else 1814 vput(nd.ni_dvp); 1815 vput(vp); 1816 } 1817 } 1818 if (dirp && v3) { 1819 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1820 vrele(dirp); 1821 } 1822 nfsm_reply(NFSX_WCCDATA(v3)); 1823 if (v3) { 1824 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1825 vn_finished_write(mp, 0); 1826 return (0); 1827 } 1828 vn_finished_write(mp, 0); 1829 nfsm_srvdone; 1830 } 1831 1832 /* 1833 * nfs rename service 1834 */ 1835 int 1836 nfsrv_rename(nfsd, slp, lwp, mrq) 1837 struct nfsrv_descript *nfsd; 1838 struct nfssvc_sock *slp; 1839 struct lwp *lwp; 1840 struct mbuf **mrq; 1841 { 1842 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1843 struct mbuf *nam = nfsd->nd_nam; 1844 caddr_t dpos = nfsd->nd_dpos; 1845 struct ucred *cred = &nfsd->nd_cr; 1846 u_int32_t *tl; 1847 int32_t t1; 1848 caddr_t bpos; 1849 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1850 uint32_t len, len2; 1851 int tdirfor_ret = 1, tdiraft_ret = 1; 1852 int v3 = (nfsd->nd_flag & ND_NFSV3); 1853 char *cp2; 1854 struct mbuf *mb, *mreq; 1855 struct nameidata fromnd, tond; 1856 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0; 1857 struct vnode *tdirp = (struct vnode *)0; 1858 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1859 nfsfh_t fnfh, tnfh; 1860 fhandle_t *ffhp, *tfhp; 1861 u_quad_t frev; 1862 uid_t saved_uid; 1863 struct mount *mp = NULL; 1864 1865 #ifndef nolint 1866 fvp = (struct vnode *)0; 1867 #endif 1868 ffhp = &fnfh.fh_generic; 1869 tfhp = &tnfh.fh_generic; 1870 fromnd.ni_cnd.cn_nameiop = 0; 1871 tond.ni_cnd.cn_nameiop = 0; 1872 nfsm_srvmtofh(ffhp); 1873 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) 1874 return (ESTALE); 1875 vn_start_write(NULL, &mp, V_WAIT); 1876 nfsm_srvnamesiz(len); 1877 /* 1878 * Remember our original uid so that we can reset cr_uid before 1879 * the second nfs_namei() call, in case it is remapped. 1880 */ 1881 saved_uid = cred->cr_uid; 1882 fromnd.ni_cnd.cn_cred = cred; 1883 fromnd.ni_cnd.cn_nameiop = DELETE; 1884 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART; 1885 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md, 1886 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1887 if (fdirp) { 1888 if (v3) 1889 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp); 1890 else { 1891 vrele(fdirp); 1892 fdirp = (struct vnode *)0; 1893 } 1894 } 1895 if (error) { 1896 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1897 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1898 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1899 if (fdirp) 1900 vrele(fdirp); 1901 vn_finished_write(mp, 0); 1902 return (0); 1903 } 1904 fvp = fromnd.ni_vp; 1905 nfsm_srvmtofh(tfhp); 1906 if (v3) { 1907 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1908 len2 = fxdr_unsigned(uint32_t, *tl); 1909 /* len2 will be checked by nfs_namei */ 1910 } 1911 else { 1912 /* NFSv2 */ 1913 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1914 } 1915 cred->cr_uid = saved_uid; 1916 tond.ni_cnd.cn_cred = cred; 1917 tond.ni_cnd.cn_nameiop = RENAME; 1918 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1919 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md, 1920 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1921 if (tdirp) { 1922 if (v3) 1923 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp); 1924 else { 1925 vrele(tdirp); 1926 tdirp = (struct vnode *)0; 1927 } 1928 } 1929 if (error) { 1930 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1931 vrele(fromnd.ni_dvp); 1932 vrele(fvp); 1933 goto out1; 1934 } 1935 tdvp = tond.ni_dvp; 1936 tvp = tond.ni_vp; 1937 if (tvp != NULL) { 1938 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1939 if (v3) 1940 error = EEXIST; 1941 else 1942 error = EISDIR; 1943 goto out; 1944 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1945 if (v3) 1946 error = EEXIST; 1947 else 1948 error = ENOTDIR; 1949 goto out; 1950 } 1951 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1952 if (v3) 1953 error = EXDEV; 1954 else 1955 error = ENOTEMPTY; 1956 goto out; 1957 } 1958 } 1959 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1960 if (v3) 1961 error = EXDEV; 1962 else 1963 error = ENOTEMPTY; 1964 goto out; 1965 } 1966 if (fvp->v_mount != tdvp->v_mount) { 1967 if (v3) 1968 error = EXDEV; 1969 else 1970 error = ENOTEMPTY; 1971 goto out; 1972 } 1973 if (fvp == tdvp) { 1974 if (v3) 1975 error = EINVAL; 1976 else 1977 error = ENOTEMPTY; 1978 } 1979 /* 1980 * If source is the same as the destination (that is the 1981 * same vnode with the same name in the same directory), 1982 * then there is nothing to do. 1983 */ 1984 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1985 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1986 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1987 fromnd.ni_cnd.cn_namelen)) 1988 error = -1; 1989 out: 1990 if (!error) { 1991 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1992 nqsrv_getl(tdvp, ND_WRITE); 1993 if (tvp) { 1994 nqsrv_getl(tvp, ND_WRITE); 1995 } 1996 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1997 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1998 } else { 1999 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 2000 if (tdvp == tvp) 2001 vrele(tdvp); 2002 else 2003 vput(tdvp); 2004 if (tvp) 2005 vput(tvp); 2006 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2007 vrele(fromnd.ni_dvp); 2008 vrele(fvp); 2009 if (error == -1) 2010 error = 0; 2011 } 2012 vrele(tond.ni_startdir); 2013 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2014 out1: 2015 if (fdirp) { 2016 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp); 2017 vrele(fdirp); 2018 } 2019 if (tdirp) { 2020 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp); 2021 vrele(tdirp); 2022 } 2023 vrele(fromnd.ni_startdir); 2024 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2025 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2026 if (v3) { 2027 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2028 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2029 } 2030 vn_finished_write(mp, 0); 2031 return (0); 2032 2033 nfsmout: 2034 if (fdirp) 2035 vrele(fdirp); 2036 if (tdirp) 2037 vrele(tdirp); 2038 if (tond.ni_cnd.cn_nameiop) { 2039 vrele(tond.ni_startdir); 2040 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2041 } 2042 if (fromnd.ni_cnd.cn_nameiop) { 2043 vrele(fromnd.ni_startdir); 2044 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2045 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2046 vrele(fromnd.ni_dvp); 2047 vrele(fvp); 2048 } 2049 return (error); 2050 } 2051 2052 /* 2053 * nfs link service 2054 */ 2055 int 2056 nfsrv_link(nfsd, slp, lwp, mrq) 2057 struct nfsrv_descript *nfsd; 2058 struct nfssvc_sock *slp; 2059 struct lwp *lwp; 2060 struct mbuf **mrq; 2061 { 2062 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2063 struct mbuf *nam = nfsd->nd_nam; 2064 caddr_t dpos = nfsd->nd_dpos; 2065 struct ucred *cred = &nfsd->nd_cr; 2066 struct nameidata nd; 2067 u_int32_t *tl; 2068 int32_t t1; 2069 caddr_t bpos; 2070 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2071 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2072 char *cp2; 2073 struct mbuf *mb, *mreq; 2074 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2075 struct vattr dirfor, diraft, at; 2076 nfsfh_t nfh, dnfh; 2077 fhandle_t *fhp, *dfhp; 2078 u_quad_t frev; 2079 struct mount *mp = NULL; 2080 2081 fhp = &nfh.fh_generic; 2082 dfhp = &dnfh.fh_generic; 2083 nfsm_srvmtofh(fhp); 2084 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2085 return (ESTALE); 2086 vn_start_write(NULL, &mp, V_WAIT); 2087 nfsm_srvmtofh(dfhp); 2088 nfsm_srvnamesiz(len); 2089 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam, 2090 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2091 if (error) { 2092 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2093 nfsm_srvpostop_attr(getret, &at); 2094 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2095 vn_finished_write(mp, 0); 2096 return (0); 2097 } 2098 if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0) 2099 goto out1; 2100 nd.ni_cnd.cn_cred = cred; 2101 nd.ni_cnd.cn_nameiop = CREATE; 2102 nd.ni_cnd.cn_flags = LOCKPARENT; 2103 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos, 2104 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2105 if (dirp) { 2106 if (v3) 2107 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2108 else { 2109 vrele(dirp); 2110 dirp = (struct vnode *)0; 2111 } 2112 } 2113 if (error) 2114 goto out1; 2115 xp = nd.ni_vp; 2116 if (xp != NULL) { 2117 error = EEXIST; 2118 goto out; 2119 } 2120 xp = nd.ni_dvp; 2121 if (vp->v_mount != xp->v_mount) 2122 error = EXDEV; 2123 out: 2124 if (!error) { 2125 nqsrv_getl(vp, ND_WRITE); 2126 nqsrv_getl(xp, ND_WRITE); 2127 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2128 } else { 2129 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2130 if (nd.ni_dvp == nd.ni_vp) 2131 vrele(nd.ni_dvp); 2132 else 2133 vput(nd.ni_dvp); 2134 if (nd.ni_vp) 2135 vrele(nd.ni_vp); 2136 } 2137 out1: 2138 if (v3) 2139 getret = VOP_GETATTR(vp, &at, cred, lwp); 2140 if (dirp) { 2141 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2142 vrele(dirp); 2143 } 2144 vrele(vp); 2145 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2146 if (v3) { 2147 nfsm_srvpostop_attr(getret, &at); 2148 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2149 vn_finished_write(mp, 0); 2150 return (0); 2151 } 2152 vn_finished_write(mp, 0); 2153 nfsm_srvdone; 2154 } 2155 2156 /* 2157 * nfs symbolic link service 2158 */ 2159 int 2160 nfsrv_symlink(nfsd, slp, lwp, mrq) 2161 struct nfsrv_descript *nfsd; 2162 struct nfssvc_sock *slp; 2163 struct lwp *lwp; 2164 struct mbuf **mrq; 2165 { 2166 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2167 struct mbuf *nam = nfsd->nd_nam; 2168 caddr_t dpos = nfsd->nd_dpos; 2169 struct ucred *cred = &nfsd->nd_cr; 2170 struct vattr va, dirfor, diraft; 2171 struct nameidata nd; 2172 u_int32_t *tl; 2173 int32_t t1; 2174 struct nfsv2_sattr *sp; 2175 char *bpos, *pathcp = NULL, *cp2; 2176 struct uio io; 2177 struct iovec iv; 2178 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1; 2179 uint32_t len, len2; 2180 int v3 = (nfsd->nd_flag & ND_NFSV3); 2181 struct mbuf *mb, *mreq; 2182 struct vnode *dirp = (struct vnode *)0; 2183 nfsfh_t nfh; 2184 fhandle_t *fhp; 2185 u_quad_t frev; 2186 struct mount *mp = NULL; 2187 2188 nd.ni_cnd.cn_nameiop = 0; 2189 fhp = &nfh.fh_generic; 2190 nfsm_srvmtofh(fhp); 2191 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2192 return (ESTALE); 2193 vn_start_write(NULL, &mp, V_WAIT); 2194 nfsm_srvnamesiz(len); 2195 nd.ni_cnd.cn_cred = cred; 2196 nd.ni_cnd.cn_nameiop = CREATE; 2197 nd.ni_cnd.cn_flags = LOCKPARENT; 2198 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2199 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2200 if (dirp) { 2201 if (v3) 2202 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2203 else { 2204 vrele(dirp); 2205 dirp = (struct vnode *)0; 2206 } 2207 } 2208 if (error) 2209 goto out; 2210 VATTR_NULL(&va); 2211 va.va_type = VLNK; 2212 if (v3) { 2213 va.va_mode = 0; 2214 nfsm_srvsattr(&va); 2215 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2216 len2 = fxdr_unsigned(uint32_t, *tl); 2217 if (len2 > PATH_MAX) { 2218 /* XXX should check _PC_NO_TRUNC */ 2219 error = ENAMETOOLONG; 2220 goto abortop; 2221 } 2222 } 2223 else { 2224 /* NFSv2 */ 2225 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2226 } 2227 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2228 iv.iov_base = pathcp; 2229 iv.iov_len = len2; 2230 io.uio_resid = len2; 2231 io.uio_offset = 0; 2232 io.uio_iov = &iv; 2233 io.uio_iovcnt = 1; 2234 io.uio_rw = UIO_READ; 2235 UIO_SETUP_SYSSPACE(&io); 2236 nfsm_mtouio(&io, len2); 2237 if (!v3) { 2238 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2239 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2240 } 2241 *(pathcp + len2) = '\0'; 2242 if (nd.ni_vp) { 2243 error = EEXIST; 2244 abortop: 2245 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2246 if (nd.ni_dvp == nd.ni_vp) 2247 vrele(nd.ni_dvp); 2248 else 2249 vput(nd.ni_dvp); 2250 if (nd.ni_vp) 2251 vrele(nd.ni_vp); 2252 goto out; 2253 } 2254 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2255 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2256 if (!error) { 2257 if (v3) { 2258 memset((caddr_t)fhp, 0, sizeof(nfh)); 2259 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsidx; 2260 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2261 if (!error) 2262 error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp); 2263 vput(nd.ni_vp); 2264 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2265 } else { 2266 vput(nd.ni_vp); 2267 } 2268 } 2269 out: 2270 if (pathcp) 2271 free(pathcp, M_TEMP); 2272 if (dirp) { 2273 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2274 vrele(dirp); 2275 } 2276 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2277 if (v3) { 2278 if (!error) { 2279 nfsm_srvpostop_fh(fhp); 2280 nfsm_srvpostop_attr(0, &va); 2281 } 2282 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2283 } 2284 vn_finished_write(mp, 0); 2285 return (0); 2286 nfsmout: 2287 if (dirp) 2288 vrele(dirp); 2289 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2290 if (nd.ni_dvp == nd.ni_vp) 2291 vrele(nd.ni_dvp); 2292 else 2293 vput(nd.ni_dvp); 2294 if (nd.ni_vp) 2295 vrele(nd.ni_vp); 2296 if (pathcp) 2297 free(pathcp, M_TEMP); 2298 vn_finished_write(mp, 0); 2299 return (error); 2300 } 2301 2302 /* 2303 * nfs mkdir service 2304 */ 2305 int 2306 nfsrv_mkdir(nfsd, slp, lwp, mrq) 2307 struct nfsrv_descript *nfsd; 2308 struct nfssvc_sock *slp; 2309 struct lwp *lwp; 2310 struct mbuf **mrq; 2311 { 2312 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2313 struct mbuf *nam = nfsd->nd_nam; 2314 caddr_t dpos = nfsd->nd_dpos; 2315 struct ucred *cred = &nfsd->nd_cr; 2316 struct vattr va, dirfor, diraft; 2317 struct nfs_fattr *fp; 2318 struct nameidata nd; 2319 caddr_t cp; 2320 u_int32_t *tl; 2321 int32_t t1; 2322 caddr_t bpos; 2323 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2324 int v3 = (nfsd->nd_flag & ND_NFSV3); 2325 char *cp2; 2326 struct mbuf *mb, *mreq; 2327 struct vnode *vp, *dirp = (struct vnode *)0; 2328 nfsfh_t nfh; 2329 fhandle_t *fhp; 2330 u_quad_t frev; 2331 struct mount *mp = NULL; 2332 2333 fhp = &nfh.fh_generic; 2334 nfsm_srvmtofh(fhp); 2335 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2336 return (ESTALE); 2337 vn_start_write(NULL, &mp, V_WAIT); 2338 nfsm_srvnamesiz(len); 2339 nd.ni_cnd.cn_cred = cred; 2340 nd.ni_cnd.cn_nameiop = CREATE; 2341 nd.ni_cnd.cn_flags = LOCKPARENT; 2342 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2343 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2344 if (dirp) { 2345 if (v3) 2346 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2347 else { 2348 vrele(dirp); 2349 dirp = (struct vnode *)0; 2350 } 2351 } 2352 if (error) { 2353 nfsm_reply(NFSX_WCCDATA(v3)); 2354 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2355 if (dirp) 2356 vrele(dirp); 2357 vn_finished_write(mp, 0); 2358 return (0); 2359 } 2360 VATTR_NULL(&va); 2361 if (v3) { 2362 va.va_mode = 0; 2363 nfsm_srvsattr(&va); 2364 } else { 2365 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2366 va.va_mode = nfstov_mode(*tl++); 2367 } 2368 va.va_type = VDIR; 2369 vp = nd.ni_vp; 2370 if (vp != NULL) { 2371 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2372 if (nd.ni_dvp == vp) 2373 vrele(nd.ni_dvp); 2374 else 2375 vput(nd.ni_dvp); 2376 vrele(vp); 2377 error = EEXIST; 2378 goto out; 2379 } 2380 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2381 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2382 if (!error) { 2383 vp = nd.ni_vp; 2384 memset((caddr_t)fhp, 0, sizeof(nfh)); 2385 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 2386 error = VFS_VPTOFH(vp, &fhp->fh_fid); 2387 if (!error) 2388 error = VOP_GETATTR(vp, &va, cred, lwp); 2389 vput(vp); 2390 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2391 } 2392 out: 2393 if (dirp) { 2394 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2395 vrele(dirp); 2396 } 2397 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2398 if (v3) { 2399 if (!error) { 2400 nfsm_srvpostop_fh(fhp); 2401 nfsm_srvpostop_attr(0, &va); 2402 } 2403 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2404 } else { 2405 nfsm_srvfhtom(fhp, v3); 2406 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2407 nfsm_srvfillattr(&va, fp); 2408 } 2409 vn_finished_write(mp, 0); 2410 return (0); 2411 nfsmout: 2412 if (dirp) 2413 vrele(dirp); 2414 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2415 if (nd.ni_dvp == nd.ni_vp) 2416 vrele(nd.ni_dvp); 2417 else 2418 vput(nd.ni_dvp); 2419 if (nd.ni_vp) 2420 vrele(nd.ni_vp); 2421 vn_finished_write(mp, 0); 2422 return (error); 2423 } 2424 2425 /* 2426 * nfs rmdir service 2427 */ 2428 int 2429 nfsrv_rmdir(nfsd, slp, lwp, mrq) 2430 struct nfsrv_descript *nfsd; 2431 struct nfssvc_sock *slp; 2432 struct lwp *lwp; 2433 struct mbuf **mrq; 2434 { 2435 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2436 struct mbuf *nam = nfsd->nd_nam; 2437 caddr_t dpos = nfsd->nd_dpos; 2438 struct ucred *cred = &nfsd->nd_cr; 2439 u_int32_t *tl; 2440 int32_t t1; 2441 caddr_t bpos; 2442 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2443 int v3 = (nfsd->nd_flag & ND_NFSV3); 2444 char *cp2; 2445 struct mbuf *mb, *mreq; 2446 struct vnode *vp, *dirp = (struct vnode *)0; 2447 struct vattr dirfor, diraft; 2448 nfsfh_t nfh; 2449 fhandle_t *fhp; 2450 struct nameidata nd; 2451 u_quad_t frev; 2452 struct mount *mp = NULL; 2453 2454 fhp = &nfh.fh_generic; 2455 nfsm_srvmtofh(fhp); 2456 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2457 return (ESTALE); 2458 vn_start_write(NULL, &mp, V_WAIT); 2459 nfsm_srvnamesiz(len); 2460 nd.ni_cnd.cn_cred = cred; 2461 nd.ni_cnd.cn_nameiop = DELETE; 2462 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2463 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2464 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2465 if (dirp) { 2466 if (v3) 2467 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2468 else { 2469 vrele(dirp); 2470 dirp = (struct vnode *)0; 2471 } 2472 } 2473 if (error) { 2474 nfsm_reply(NFSX_WCCDATA(v3)); 2475 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2476 if (dirp) 2477 vrele(dirp); 2478 vn_finished_write(mp, 0); 2479 return (0); 2480 } 2481 vp = nd.ni_vp; 2482 if (vp->v_type != VDIR) { 2483 error = ENOTDIR; 2484 goto out; 2485 } 2486 /* 2487 * No rmdir "." please. 2488 */ 2489 if (nd.ni_dvp == vp) { 2490 error = EINVAL; 2491 goto out; 2492 } 2493 /* 2494 * The root of a mounted filesystem cannot be deleted. 2495 */ 2496 if (vp->v_flag & VROOT) 2497 error = EBUSY; 2498 out: 2499 if (!error) { 2500 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2501 nqsrv_getl(vp, ND_WRITE); 2502 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2503 } else { 2504 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2505 if (nd.ni_dvp == nd.ni_vp) 2506 vrele(nd.ni_dvp); 2507 else 2508 vput(nd.ni_dvp); 2509 vput(vp); 2510 } 2511 if (dirp) { 2512 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2513 vrele(dirp); 2514 } 2515 nfsm_reply(NFSX_WCCDATA(v3)); 2516 if (v3) { 2517 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2518 vn_finished_write(mp, 0); 2519 return (0); 2520 } 2521 vn_finished_write(mp, 0); 2522 nfsm_srvdone; 2523 } 2524 2525 /* 2526 * nfs readdir service 2527 * - mallocs what it thinks is enough to read 2528 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2529 * - calls VOP_READDIR() 2530 * - loops around building the reply 2531 * if the output generated exceeds count break out of loop 2532 * The nfsm_clget macro is used here so that the reply will be packed 2533 * tightly in mbuf clusters. 2534 * - it only knows that it has encountered eof when the VOP_READDIR() 2535 * reads nothing 2536 * - as such one readdir rpc will return eof false although you are there 2537 * and then the next will return eof 2538 * - it trims out records with d_fileno == 0 2539 * this doesn't matter for Unix clients, but they might confuse clients 2540 * for other os'. 2541 * - it trims out records with d_type == DT_WHT 2542 * these cannot be seen through NFS (unless we extend the protocol) 2543 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2544 * than requested, but this may not apply to all filesystems. For 2545 * example, client NFS does not { although it is never remote mounted 2546 * anyhow } 2547 * The alternate call nfsrv_readdirplus() does lookups as well. 2548 * PS: The NFS protocol spec. does not clarify what the "count" byte 2549 * argument is a count of.. just name strings and file id's or the 2550 * entire reply rpc or ... 2551 * I tried just file name and id sizes and it confused the Sun client, 2552 * so I am using the full rpc size now. The "paranoia.." comment refers 2553 * to including the status longwords that are not a part of the dir. 2554 * "entry" structures, but are in the rpc. 2555 */ 2556 2557 #define NFS_SRVDIRBLKSIZ 1024 2558 2559 struct flrep { 2560 nfsuint64 fl_off; 2561 u_int32_t fl_postopok; 2562 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2563 u_int32_t fl_fhok; 2564 u_int32_t fl_fhsize; 2565 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)]; 2566 }; 2567 2568 int 2569 nfsrv_readdir(nfsd, slp, lwp, mrq) 2570 struct nfsrv_descript *nfsd; 2571 struct nfssvc_sock *slp; 2572 struct lwp *lwp; 2573 struct mbuf **mrq; 2574 { 2575 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2576 struct mbuf *nam = nfsd->nd_nam; 2577 caddr_t dpos = nfsd->nd_dpos; 2578 struct ucred *cred = &nfsd->nd_cr; 2579 char *bp, *be; 2580 struct mbuf *mp; 2581 struct dirent *dp; 2582 caddr_t cp; 2583 u_int32_t *tl; 2584 int32_t t1; 2585 caddr_t bpos; 2586 struct mbuf *mb, *mreq, *mp2; 2587 char *cpos, *cend, *cp2, *rbuf; 2588 struct vnode *vp; 2589 struct vattr at; 2590 nfsfh_t nfh; 2591 fhandle_t *fhp; 2592 struct uio io; 2593 struct iovec iv; 2594 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2595 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2596 int v3 = (nfsd->nd_flag & ND_NFSV3); 2597 u_quad_t frev, off, toff, verf; 2598 off_t *cookies = NULL, *cookiep; 2599 nfsuint64 jar; 2600 2601 fhp = &nfh.fh_generic; 2602 nfsm_srvmtofh(fhp); 2603 if (v3) { 2604 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2605 toff = fxdr_hyper(tl); 2606 tl += 2; 2607 verf = fxdr_hyper(tl); 2608 tl += 2; 2609 } else { 2610 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2611 toff = fxdr_unsigned(u_quad_t, *tl++); 2612 } 2613 off = toff; 2614 cnt = fxdr_unsigned(int, *tl); 2615 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2616 xfer = NFS_SRVMAXDATA(nfsd); 2617 if (siz > xfer) 2618 siz = xfer; 2619 fullsiz = siz; 2620 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2621 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2622 if (!error && vp->v_type != VDIR) { 2623 error = ENOTDIR; 2624 vput(vp); 2625 } 2626 if (error) { 2627 nfsm_reply(NFSX_UNSIGNED); 2628 nfsm_srvpostop_attr(getret, &at); 2629 return (0); 2630 } 2631 nqsrv_getl(vp, ND_READ); 2632 if (v3) { 2633 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2634 #ifdef NFS3_STRICTVERF 2635 /* 2636 * XXX This check is too strict for Solaris 2.5 clients. 2637 */ 2638 if (!error && toff && verf != at.va_filerev) 2639 error = NFSERR_BAD_COOKIE; 2640 #endif 2641 } 2642 if (!error) 2643 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2644 if (error) { 2645 vput(vp); 2646 nfsm_reply(NFSX_POSTOPATTR(v3)); 2647 nfsm_srvpostop_attr(getret, &at); 2648 return (0); 2649 } 2650 VOP_UNLOCK(vp, 0); 2651 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2652 again: 2653 iv.iov_base = rbuf; 2654 iv.iov_len = fullsiz; 2655 io.uio_iov = &iv; 2656 io.uio_iovcnt = 1; 2657 io.uio_offset = (off_t)off; 2658 io.uio_resid = fullsiz; 2659 io.uio_rw = UIO_READ; 2660 UIO_SETUP_SYSSPACE(&io); 2661 eofflag = 0; 2662 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2663 2664 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2665 2666 off = (off_t)io.uio_offset; 2667 if (!cookies && !error) 2668 error = NFSERR_PERM; 2669 if (v3) { 2670 getret = VOP_GETATTR(vp, &at, cred, lwp); 2671 if (!error) 2672 error = getret; 2673 } 2674 2675 VOP_UNLOCK(vp, 0); 2676 if (error) { 2677 vrele(vp); 2678 free((caddr_t)rbuf, M_TEMP); 2679 if (cookies) 2680 free((caddr_t)cookies, M_TEMP); 2681 nfsm_reply(NFSX_POSTOPATTR(v3)); 2682 nfsm_srvpostop_attr(getret, &at); 2683 return (0); 2684 } 2685 if (io.uio_resid) { 2686 siz -= io.uio_resid; 2687 2688 /* 2689 * If nothing read, return eof 2690 * rpc reply 2691 */ 2692 if (siz == 0) { 2693 vrele(vp); 2694 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2695 2 * NFSX_UNSIGNED); 2696 if (v3) { 2697 nfsm_srvpostop_attr(getret, &at); 2698 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2699 txdr_hyper(at.va_filerev, tl); 2700 tl += 2; 2701 } else 2702 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2703 *tl++ = nfs_false; 2704 *tl = nfs_true; 2705 free((caddr_t)rbuf, M_TEMP); 2706 free((caddr_t)cookies, M_TEMP); 2707 return (0); 2708 } 2709 } 2710 2711 /* 2712 * Check for degenerate cases of nothing useful read. 2713 * If so go try again 2714 */ 2715 cpos = rbuf; 2716 cend = rbuf + siz; 2717 dp = (struct dirent *)cpos; 2718 cookiep = cookies; 2719 2720 while (cpos < cend && ncookies > 0 && 2721 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2722 cpos += dp->d_reclen; 2723 dp = (struct dirent *)cpos; 2724 cookiep++; 2725 ncookies--; 2726 } 2727 if (cpos >= cend || ncookies == 0) { 2728 toff = off; 2729 siz = fullsiz; 2730 goto again; 2731 } 2732 2733 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2734 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2735 if (v3) { 2736 nfsm_srvpostop_attr(getret, &at); 2737 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2738 txdr_hyper(at.va_filerev, tl); 2739 } 2740 mp = mp2 = mb; 2741 bp = bpos; 2742 be = bp + M_TRAILINGSPACE(mp); 2743 2744 /* Loop through the records and build reply */ 2745 while (cpos < cend && ncookies > 0) { 2746 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2747 nlen = dp->d_namlen; 2748 rem = nfsm_rndup(nlen)-nlen; 2749 len += (4 * NFSX_UNSIGNED + nlen + rem); 2750 if (v3) 2751 len += 2 * NFSX_UNSIGNED; 2752 if (len > cnt) { 2753 eofflag = 0; 2754 break; 2755 } 2756 /* 2757 * Build the directory record xdr from 2758 * the dirent entry. 2759 */ 2760 nfsm_clget; 2761 *tl = nfs_true; 2762 bp += NFSX_UNSIGNED; 2763 if (v3) { 2764 nfsm_clget; 2765 *tl = txdr_unsigned(dp->d_fileno >> 32); 2766 bp += NFSX_UNSIGNED; 2767 } 2768 nfsm_clget; 2769 *tl = txdr_unsigned(dp->d_fileno); 2770 bp += NFSX_UNSIGNED; 2771 nfsm_clget; 2772 *tl = txdr_unsigned(nlen); 2773 bp += NFSX_UNSIGNED; 2774 2775 /* And loop around copying the name */ 2776 xfer = nlen; 2777 cp = dp->d_name; 2778 while (xfer > 0) { 2779 nfsm_clget; 2780 if ((bp+xfer) > be) 2781 tsiz = be-bp; 2782 else 2783 tsiz = xfer; 2784 memcpy(bp, cp, tsiz); 2785 bp += tsiz; 2786 xfer -= tsiz; 2787 if (xfer > 0) 2788 cp += tsiz; 2789 } 2790 /* And null pad to an int32_t boundary */ 2791 for (i = 0; i < rem; i++) 2792 *bp++ = '\0'; 2793 nfsm_clget; 2794 2795 /* Finish off the record */ 2796 txdr_hyper(*cookiep, &jar); 2797 if (v3) { 2798 *tl = jar.nfsuquad[0]; 2799 bp += NFSX_UNSIGNED; 2800 nfsm_clget; 2801 } 2802 *tl = jar.nfsuquad[1]; 2803 bp += NFSX_UNSIGNED; 2804 } 2805 cpos += dp->d_reclen; 2806 dp = (struct dirent *)cpos; 2807 cookiep++; 2808 ncookies--; 2809 } 2810 vrele(vp); 2811 nfsm_clget; 2812 *tl = nfs_false; 2813 bp += NFSX_UNSIGNED; 2814 nfsm_clget; 2815 if (eofflag) 2816 *tl = nfs_true; 2817 else 2818 *tl = nfs_false; 2819 bp += NFSX_UNSIGNED; 2820 if (mp != mb) { 2821 if (bp < be) 2822 mp->m_len = bp - mtod(mp, caddr_t); 2823 } else 2824 mp->m_len += bp - bpos; 2825 free((caddr_t)rbuf, M_TEMP); 2826 free((caddr_t)cookies, M_TEMP); 2827 nfsm_srvdone; 2828 } 2829 2830 int 2831 nfsrv_readdirplus(nfsd, slp, lwp, mrq) 2832 struct nfsrv_descript *nfsd; 2833 struct nfssvc_sock *slp; 2834 struct lwp *lwp; 2835 struct mbuf **mrq; 2836 { 2837 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2838 struct mbuf *nam = nfsd->nd_nam; 2839 caddr_t dpos = nfsd->nd_dpos; 2840 struct ucred *cred = &nfsd->nd_cr; 2841 char *bp, *be; 2842 struct mbuf *mp; 2843 struct dirent *dp; 2844 caddr_t cp; 2845 u_int32_t *tl; 2846 int32_t t1; 2847 caddr_t bpos; 2848 struct mbuf *mb, *mreq, *mp2; 2849 char *cpos, *cend, *cp2, *rbuf; 2850 struct vnode *vp, *nvp; 2851 struct flrep fl; 2852 nfsfh_t nfh; 2853 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh; 2854 struct uio io; 2855 struct iovec iv; 2856 struct vattr va, at, *vap = &va; 2857 struct nfs_fattr *fp; 2858 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2859 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2860 u_quad_t frev, off, toff, verf; 2861 off_t *cookies = NULL, *cookiep; 2862 2863 fhp = &nfh.fh_generic; 2864 nfsm_srvmtofh(fhp); 2865 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2866 toff = fxdr_hyper(tl); 2867 tl += 2; 2868 verf = fxdr_hyper(tl); 2869 tl += 2; 2870 siz = fxdr_unsigned(int, *tl++); 2871 cnt = fxdr_unsigned(int, *tl); 2872 off = toff; 2873 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2874 xfer = NFS_SRVMAXDATA(nfsd); 2875 if (siz > xfer) 2876 siz = xfer; 2877 fullsiz = siz; 2878 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2879 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2880 if (!error && vp->v_type != VDIR) { 2881 error = ENOTDIR; 2882 vput(vp); 2883 } 2884 if (error) { 2885 nfsm_reply(NFSX_UNSIGNED); 2886 nfsm_srvpostop_attr(getret, &at); 2887 return (0); 2888 } 2889 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2890 #ifdef NFS3_STRICTVERF 2891 /* 2892 * XXX This check is too strict for Solaris 2.5 clients. 2893 */ 2894 if (!error && toff && verf != at.va_filerev) 2895 error = NFSERR_BAD_COOKIE; 2896 #endif 2897 if (!error) { 2898 nqsrv_getl(vp, ND_READ); 2899 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2900 } 2901 if (error) { 2902 vput(vp); 2903 nfsm_reply(NFSX_V3POSTOPATTR); 2904 nfsm_srvpostop_attr(getret, &at); 2905 return (0); 2906 } 2907 VOP_UNLOCK(vp, 0); 2908 2909 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2910 again: 2911 iv.iov_base = rbuf; 2912 iv.iov_len = fullsiz; 2913 io.uio_iov = &iv; 2914 io.uio_iovcnt = 1; 2915 io.uio_offset = (off_t)off; 2916 io.uio_resid = fullsiz; 2917 io.uio_rw = UIO_READ; 2918 UIO_SETUP_SYSSPACE(&io); 2919 eofflag = 0; 2920 2921 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2922 2923 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2924 2925 off = (u_quad_t)io.uio_offset; 2926 getret = VOP_GETATTR(vp, &at, cred, lwp); 2927 2928 VOP_UNLOCK(vp, 0); 2929 2930 /* 2931 * If the VGET operation doesn't work for this filesystem, 2932 * we can't support readdirplus. Returning NOTSUPP should 2933 * make clients fall back to plain readdir. 2934 * There's no need to check for VPTOFH as well, we wouldn't 2935 * even be here otherwise. 2936 */ 2937 if (!getret) { 2938 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2939 getret = (getret == EOPNOTSUPP) ? 2940 NFSERR_NOTSUPP : NFSERR_IO; 2941 else 2942 vput(nvp); 2943 } 2944 2945 if (!cookies && !error) 2946 error = NFSERR_PERM; 2947 if (!error) 2948 error = getret; 2949 if (error) { 2950 vrele(vp); 2951 if (cookies) 2952 free((caddr_t)cookies, M_TEMP); 2953 free((caddr_t)rbuf, M_TEMP); 2954 nfsm_reply(NFSX_V3POSTOPATTR); 2955 nfsm_srvpostop_attr(getret, &at); 2956 return (0); 2957 } 2958 if (io.uio_resid) { 2959 siz -= io.uio_resid; 2960 2961 /* 2962 * If nothing read, return eof 2963 * rpc reply 2964 */ 2965 if (siz == 0) { 2966 vrele(vp); 2967 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2968 2 * NFSX_UNSIGNED); 2969 nfsm_srvpostop_attr(getret, &at); 2970 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2971 txdr_hyper(at.va_filerev, tl); 2972 tl += 2; 2973 *tl++ = nfs_false; 2974 *tl = nfs_true; 2975 free((caddr_t)cookies, M_TEMP); 2976 free((caddr_t)rbuf, M_TEMP); 2977 return (0); 2978 } 2979 } 2980 2981 /* 2982 * Check for degenerate cases of nothing useful read. 2983 * If so go try again 2984 */ 2985 cpos = rbuf; 2986 cend = rbuf + siz; 2987 dp = (struct dirent *)cpos; 2988 cookiep = cookies; 2989 2990 while (cpos < cend && ncookies > 0 && 2991 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2992 cpos += dp->d_reclen; 2993 dp = (struct dirent *)cpos; 2994 cookiep++; 2995 ncookies--; 2996 } 2997 if (cpos >= cend || ncookies == 0) { 2998 toff = off; 2999 siz = fullsiz; 3000 goto again; 3001 } 3002 3003 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3004 nfsm_reply(cnt); 3005 nfsm_srvpostop_attr(getret, &at); 3006 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3007 txdr_hyper(at.va_filerev, tl); 3008 mp = mp2 = mb; 3009 bp = bpos; 3010 be = bp + M_TRAILINGSPACE(mp); 3011 3012 /* Loop through the records and build reply */ 3013 while (cpos < cend && ncookies > 0) { 3014 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3015 nlen = dp->d_namlen; 3016 rem = nfsm_rndup(nlen)-nlen; 3017 3018 /* 3019 * For readdir_and_lookup get the vnode using 3020 * the file number. 3021 */ 3022 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3023 goto invalid; 3024 memset((caddr_t)nfhp, 0, NFSX_V3FH); 3025 nfhp->fh_fsid = 3026 nvp->v_mount->mnt_stat.f_fsidx; 3027 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) { 3028 vput(nvp); 3029 goto invalid; 3030 } 3031 if (VOP_GETATTR(nvp, vap, cred, lwp)) { 3032 vput(nvp); 3033 goto invalid; 3034 } 3035 vput(nvp); 3036 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 3037 3038 /* 3039 * If either the dircount or maxcount will be 3040 * exceeded, get out now. Both of these lengths 3041 * are calculated conservatively, including all 3042 * XDR overheads. 3043 */ 3044 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3045 NFSX_V3POSTOPATTR); 3046 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3047 if (len > cnt || dirlen > fullsiz) { 3048 eofflag = 0; 3049 break; 3050 } 3051 3052 /* 3053 * Build the directory record xdr from 3054 * the dirent entry. 3055 */ 3056 fp = (struct nfs_fattr *)&fl.fl_fattr; 3057 nfsm_srvfillattr(vap, fp); 3058 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3059 fl.fl_fhok = nfs_true; 3060 fl.fl_postopok = nfs_true; 3061 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3062 3063 nfsm_clget; 3064 *tl = nfs_true; 3065 bp += NFSX_UNSIGNED; 3066 nfsm_clget; 3067 *tl = txdr_unsigned(dp->d_fileno >> 32); 3068 bp += NFSX_UNSIGNED; 3069 nfsm_clget; 3070 *tl = txdr_unsigned(dp->d_fileno); 3071 bp += NFSX_UNSIGNED; 3072 nfsm_clget; 3073 *tl = txdr_unsigned(nlen); 3074 bp += NFSX_UNSIGNED; 3075 3076 /* And loop around copying the name */ 3077 xfer = nlen; 3078 cp = dp->d_name; 3079 while (xfer > 0) { 3080 nfsm_clget; 3081 if ((bp + xfer) > be) 3082 tsiz = be - bp; 3083 else 3084 tsiz = xfer; 3085 memcpy(bp, cp, tsiz); 3086 bp += tsiz; 3087 xfer -= tsiz; 3088 if (xfer > 0) 3089 cp += tsiz; 3090 } 3091 /* And null pad to an int32_t boundary */ 3092 for (i = 0; i < rem; i++) 3093 *bp++ = '\0'; 3094 3095 /* 3096 * Now copy the flrep structure out. 3097 */ 3098 xfer = sizeof (struct flrep); 3099 cp = (caddr_t)&fl; 3100 while (xfer > 0) { 3101 nfsm_clget; 3102 if ((bp + xfer) > be) 3103 tsiz = be - bp; 3104 else 3105 tsiz = xfer; 3106 memcpy(bp, cp, tsiz); 3107 bp += tsiz; 3108 xfer -= tsiz; 3109 if (xfer > 0) 3110 cp += tsiz; 3111 } 3112 } 3113 invalid: 3114 cpos += dp->d_reclen; 3115 dp = (struct dirent *)cpos; 3116 cookiep++; 3117 ncookies--; 3118 } 3119 vrele(vp); 3120 nfsm_clget; 3121 *tl = nfs_false; 3122 bp += NFSX_UNSIGNED; 3123 nfsm_clget; 3124 if (eofflag) 3125 *tl = nfs_true; 3126 else 3127 *tl = nfs_false; 3128 bp += NFSX_UNSIGNED; 3129 if (mp != mb) { 3130 if (bp < be) 3131 mp->m_len = bp - mtod(mp, caddr_t); 3132 } else 3133 mp->m_len += bp - bpos; 3134 free((caddr_t)cookies, M_TEMP); 3135 free((caddr_t)rbuf, M_TEMP); 3136 nfsm_srvdone; 3137 } 3138 3139 /* 3140 * nfs commit service 3141 */ 3142 int 3143 nfsrv_commit(nfsd, slp, lwp, mrq) 3144 struct nfsrv_descript *nfsd; 3145 struct nfssvc_sock *slp; 3146 struct lwp *lwp; 3147 struct mbuf **mrq; 3148 { 3149 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3150 struct mbuf *nam = nfsd->nd_nam; 3151 caddr_t dpos = nfsd->nd_dpos; 3152 struct ucred *cred = &nfsd->nd_cr; 3153 struct vattr bfor, aft; 3154 struct vnode *vp; 3155 nfsfh_t nfh; 3156 fhandle_t *fhp; 3157 u_int32_t *tl; 3158 int32_t t1; 3159 caddr_t bpos; 3160 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3161 uint32_t cnt; 3162 char *cp2; 3163 struct mbuf *mb, *mreq; 3164 u_quad_t frev, off, end; 3165 struct mount *mp = NULL; 3166 3167 fhp = &nfh.fh_generic; 3168 nfsm_srvmtofh(fhp); 3169 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 3170 return (ESTALE); 3171 vn_start_write(NULL, &mp, V_WAIT); 3172 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3173 3174 off = fxdr_hyper(tl); 3175 tl += 2; 3176 cnt = fxdr_unsigned(uint32_t, *tl); 3177 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3178 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3179 if (error) { 3180 nfsm_reply(2 * NFSX_UNSIGNED); 3181 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3182 vn_finished_write(mp, 0); 3183 return (0); 3184 } 3185 for_ret = VOP_GETATTR(vp, &bfor, cred, lwp); 3186 end = (cnt > 0) ? off + cnt : vp->v_size; 3187 if (end < off || end > vp->v_size) 3188 end = vp->v_size; 3189 if (off < vp->v_size) 3190 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp); 3191 /* else error == 0, from nfsrv_fhtovp() */ 3192 aft_ret = VOP_GETATTR(vp, &aft, cred, lwp); 3193 vput(vp); 3194 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3195 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3196 if (!error) { 3197 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3198 *tl++ = txdr_unsigned(boottime.tv_sec); 3199 *tl = txdr_unsigned(boottime.tv_usec); 3200 } else { 3201 vn_finished_write(mp, 0); 3202 return (0); 3203 } 3204 vn_finished_write(mp, 0); 3205 nfsm_srvdone; 3206 } 3207 3208 /* 3209 * nfs statfs service 3210 */ 3211 int 3212 nfsrv_statfs(nfsd, slp, lwp, mrq) 3213 struct nfsrv_descript *nfsd; 3214 struct nfssvc_sock *slp; 3215 struct lwp *lwp; 3216 struct mbuf **mrq; 3217 { 3218 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3219 struct mbuf *nam = nfsd->nd_nam; 3220 caddr_t dpos = nfsd->nd_dpos; 3221 struct ucred *cred = &nfsd->nd_cr; 3222 struct statvfs *sf; 3223 struct nfs_statfs *sfp; 3224 u_int32_t *tl; 3225 int32_t t1; 3226 caddr_t bpos; 3227 int error = 0, rdonly, cache = 0, getret = 1; 3228 int v3 = (nfsd->nd_flag & ND_NFSV3); 3229 char *cp2; 3230 struct mbuf *mb, *mreq; 3231 struct vnode *vp; 3232 struct vattr at; 3233 nfsfh_t nfh; 3234 fhandle_t *fhp; 3235 struct statvfs statvfs; 3236 u_quad_t frev, tval; 3237 3238 fhp = &nfh.fh_generic; 3239 nfsm_srvmtofh(fhp); 3240 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3241 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3242 if (error) { 3243 nfsm_reply(NFSX_UNSIGNED); 3244 nfsm_srvpostop_attr(getret, &at); 3245 return (0); 3246 } 3247 sf = &statvfs; 3248 error = VFS_STATVFS(vp->v_mount, sf, lwp); 3249 getret = VOP_GETATTR(vp, &at, cred, lwp); 3250 vput(vp); 3251 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3252 if (v3) 3253 nfsm_srvpostop_attr(getret, &at); 3254 if (error) 3255 return (0); 3256 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3257 if (v3) { 3258 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3259 txdr_hyper(tval, &sfp->sf_tbytes); 3260 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3261 txdr_hyper(tval, &sfp->sf_fbytes); 3262 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3263 txdr_hyper(tval, &sfp->sf_abytes); 3264 tval = (u_quad_t)sf->f_files; 3265 txdr_hyper(tval, &sfp->sf_tfiles); 3266 tval = (u_quad_t)sf->f_ffree; 3267 txdr_hyper(tval, &sfp->sf_ffiles); 3268 txdr_hyper(tval, &sfp->sf_afiles); 3269 sfp->sf_invarsec = 0; 3270 } else { 3271 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3272 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3273 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3274 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3275 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3276 } 3277 nfsm_srvdone; 3278 } 3279 3280 /* 3281 * nfs fsinfo service 3282 */ 3283 int 3284 nfsrv_fsinfo(nfsd, slp, lwp, mrq) 3285 struct nfsrv_descript *nfsd; 3286 struct nfssvc_sock *slp; 3287 struct lwp *lwp; 3288 struct mbuf **mrq; 3289 { 3290 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3291 struct mbuf *nam = nfsd->nd_nam; 3292 caddr_t dpos = nfsd->nd_dpos; 3293 struct ucred *cred = &nfsd->nd_cr; 3294 u_int32_t *tl; 3295 struct nfsv3_fsinfo *sip; 3296 int32_t t1; 3297 caddr_t bpos; 3298 int error = 0, rdonly, cache = 0, getret = 1; 3299 uint32_t maxdata; 3300 char *cp2; 3301 struct mbuf *mb, *mreq; 3302 struct vnode *vp; 3303 struct vattr at; 3304 nfsfh_t nfh; 3305 fhandle_t *fhp; 3306 u_quad_t frev, maxfsize; 3307 struct statvfs sb; 3308 3309 fhp = &nfh.fh_generic; 3310 nfsm_srvmtofh(fhp); 3311 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3312 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3313 if (error) { 3314 nfsm_reply(NFSX_UNSIGNED); 3315 nfsm_srvpostop_attr(getret, &at); 3316 return (0); 3317 } 3318 3319 /* XXX Try to make a guess on the max file size. */ 3320 VFS_STATVFS(vp->v_mount, &sb, (struct lwp *)0); 3321 maxfsize = (u_quad_t)0x80000000 * sb.f_frsize - 1; 3322 3323 getret = VOP_GETATTR(vp, &at, cred, lwp); 3324 vput(vp); 3325 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3326 nfsm_srvpostop_attr(getret, &at); 3327 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3328 3329 /* 3330 * XXX 3331 * There should be file system VFS OP(s) to get this information. 3332 * For now, assume ufs. 3333 */ 3334 if (slp->ns_so->so_type == SOCK_DGRAM) 3335 maxdata = NFS_MAXDGRAMDATA; 3336 else 3337 maxdata = NFS_MAXDATA; 3338 sip->fs_rtmax = txdr_unsigned(maxdata); 3339 sip->fs_rtpref = txdr_unsigned(maxdata); 3340 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3341 sip->fs_wtmax = txdr_unsigned(maxdata); 3342 sip->fs_wtpref = txdr_unsigned(maxdata); 3343 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3344 sip->fs_dtpref = txdr_unsigned(maxdata); 3345 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3346 sip->fs_timedelta.nfsv3_sec = 0; 3347 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3348 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3349 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3350 NFSV3FSINFO_CANSETTIME); 3351 nfsm_srvdone; 3352 } 3353 3354 /* 3355 * nfs pathconf service 3356 */ 3357 int 3358 nfsrv_pathconf(nfsd, slp, lwp, mrq) 3359 struct nfsrv_descript *nfsd; 3360 struct nfssvc_sock *slp; 3361 struct lwp *lwp; 3362 struct mbuf **mrq; 3363 { 3364 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3365 struct mbuf *nam = nfsd->nd_nam; 3366 caddr_t dpos = nfsd->nd_dpos; 3367 struct ucred *cred = &nfsd->nd_cr; 3368 u_int32_t *tl; 3369 struct nfsv3_pathconf *pc; 3370 int32_t t1; 3371 caddr_t bpos; 3372 int error = 0, rdonly, cache = 0, getret = 1; 3373 register_t linkmax, namemax, chownres, notrunc; 3374 char *cp2; 3375 struct mbuf *mb, *mreq; 3376 struct vnode *vp; 3377 struct vattr at; 3378 nfsfh_t nfh; 3379 fhandle_t *fhp; 3380 u_quad_t frev; 3381 3382 fhp = &nfh.fh_generic; 3383 nfsm_srvmtofh(fhp); 3384 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3385 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3386 if (error) { 3387 nfsm_reply(NFSX_UNSIGNED); 3388 nfsm_srvpostop_attr(getret, &at); 3389 return (0); 3390 } 3391 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3392 if (!error) 3393 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3394 if (!error) 3395 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3396 if (!error) 3397 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3398 getret = VOP_GETATTR(vp, &at, cred, lwp); 3399 vput(vp); 3400 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3401 nfsm_srvpostop_attr(getret, &at); 3402 if (error) 3403 return (0); 3404 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3405 3406 pc->pc_linkmax = txdr_unsigned(linkmax); 3407 pc->pc_namemax = txdr_unsigned(namemax); 3408 pc->pc_notrunc = txdr_unsigned(notrunc); 3409 pc->pc_chownrestricted = txdr_unsigned(chownres); 3410 3411 /* 3412 * These should probably be supported by VOP_PATHCONF(), but 3413 * until msdosfs is exportable (why would you want to?), the 3414 * Unix defaults should be ok. 3415 */ 3416 pc->pc_caseinsensitive = nfs_false; 3417 pc->pc_casepreserving = nfs_true; 3418 nfsm_srvdone; 3419 } 3420 3421 /* 3422 * Null operation, used by clients to ping server 3423 */ 3424 /* ARGSUSED */ 3425 int 3426 nfsrv_null(nfsd, slp, lwp, mrq) 3427 struct nfsrv_descript *nfsd; 3428 struct nfssvc_sock *slp; 3429 struct lwp *lwp; 3430 struct mbuf **mrq; 3431 { 3432 struct mbuf *mrep = nfsd->nd_mrep; 3433 caddr_t bpos; 3434 int error = NFSERR_RETVOID, cache = 0; 3435 struct mbuf *mb, *mreq; 3436 u_quad_t frev; 3437 3438 nfsm_reply(0); 3439 return (0); 3440 } 3441 3442 /* 3443 * No operation, used for obsolete procedures 3444 */ 3445 /* ARGSUSED */ 3446 int 3447 nfsrv_noop(nfsd, slp, lwp, mrq) 3448 struct nfsrv_descript *nfsd; 3449 struct nfssvc_sock *slp; 3450 struct lwp *lwp; 3451 struct mbuf **mrq; 3452 { 3453 struct mbuf *mrep = nfsd->nd_mrep; 3454 caddr_t bpos; 3455 int error, cache = 0; 3456 struct mbuf *mb, *mreq; 3457 u_quad_t frev; 3458 3459 if (nfsd->nd_repstat) 3460 error = nfsd->nd_repstat; 3461 else 3462 error = EPROCUNAVAIL; 3463 nfsm_reply(0); 3464 return (0); 3465 } 3466 3467 /* 3468 * Perform access checking for vnodes obtained from file handles that would 3469 * refer to files already opened by a Unix client. You cannot just use 3470 * vn_writechk() and VOP_ACCESS() for two reasons. 3471 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3472 * 2 - The owner is to be given access irrespective of mode bits for some 3473 * operations, so that processes that chmod after opening a file don't 3474 * break. I don't like this because it opens a security hole, but since 3475 * the nfs server opens a security hole the size of a barn door anyhow, 3476 * what the heck. 3477 * 3478 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3479 * will return EPERM instead of EACCESS. EPERM is always an error. 3480 */ 3481 int 3482 nfsrv_access(vp, flags, cred, rdonly, lwp, override) 3483 struct vnode *vp; 3484 int flags; 3485 struct ucred *cred; 3486 int rdonly; 3487 struct lwp *lwp; 3488 { 3489 struct vattr vattr; 3490 int error; 3491 if (flags & VWRITE) { 3492 /* Just vn_writechk() changed to check rdonly */ 3493 /* 3494 * Disallow write attempts on read-only file systems; 3495 * unless the file is a socket or a block or character 3496 * device resident on the file system. 3497 */ 3498 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3499 switch (vp->v_type) { 3500 case VREG: 3501 case VDIR: 3502 case VLNK: 3503 return (EROFS); 3504 default: 3505 break; 3506 } 3507 } 3508 3509 /* 3510 * If the vnode is in use as a process's text, 3511 * we can't allow writing. 3512 */ 3513 if (vp->v_flag & VTEXT) 3514 return (ETXTBSY); 3515 } 3516 error = VOP_GETATTR(vp, &vattr, cred, lwp); 3517 if (error) 3518 return (error); 3519 error = VOP_ACCESS(vp, flags, cred, lwp); 3520 /* 3521 * Allow certain operations for the owner (reads and writes 3522 * on files that are already open). 3523 */ 3524 if (override && error == EACCES && cred->cr_uid == vattr.va_uid) 3525 error = 0; 3526 return error; 3527 } 3528
Indexes created Wed Sep 24 05:09:52 GMT 2025