nfs_serv.c revision 1.104
1 /* $NetBSD: nfs_serv.c,v 1.104 2006/04/15 01:37:46 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 1.104 2006/04/15 01:37:46 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 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1560 vrele(dirp); 1561 } 1562 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1563 if (v3) { 1564 if (!error) { 1565 nfsm_srvpostop_fh(fhp); 1566 nfsm_srvpostop_attr(0, &va); 1567 } 1568 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1569 } else { 1570 nfsm_srvfhtom(fhp, v3); 1571 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1572 nfsm_srvfillattr(&va, fp); 1573 } 1574 vn_finished_write(mp, 0); 1575 return (0); 1576 nfsmout: 1577 if (dirp) 1578 vrele(dirp); 1579 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1580 if (nd.ni_dvp == nd.ni_vp) 1581 vrele(nd.ni_dvp); 1582 else 1583 vput(nd.ni_dvp); 1584 if (nd.ni_vp) 1585 vput(nd.ni_vp); 1586 vn_finished_write(mp, 0); 1587 return (error); 1588 } 1589 1590 /* 1591 * nfs v3 mknod service 1592 */ 1593 int 1594 nfsrv_mknod(nfsd, slp, lwp, mrq) 1595 struct nfsrv_descript *nfsd; 1596 struct nfssvc_sock *slp; 1597 struct lwp *lwp; 1598 struct mbuf **mrq; 1599 { 1600 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1601 struct mbuf *nam = nfsd->nd_nam; 1602 caddr_t dpos = nfsd->nd_dpos; 1603 struct ucred *cred = &nfsd->nd_cr; 1604 struct vattr va, dirfor, diraft; 1605 u_int32_t *tl; 1606 struct nameidata nd; 1607 int32_t t1; 1608 caddr_t bpos; 1609 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1610 u_int32_t major, minor; 1611 enum vtype vtyp; 1612 char *cp2; 1613 struct mbuf *mb, *mreq; 1614 struct vnode *vp, *dirp = (struct vnode *)0; 1615 nfsfh_t nfh; 1616 fhandle_t *fhp; 1617 u_quad_t frev; 1618 struct mount *mp = NULL; 1619 1620 nd.ni_cnd.cn_nameiop = 0; 1621 fhp = &nfh.fh_generic; 1622 nfsm_srvmtofh(fhp); 1623 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1624 return (ESTALE); 1625 vn_start_write(NULL, &mp, V_WAIT); 1626 nfsm_srvnamesiz(len); 1627 nd.ni_cnd.cn_cred = cred; 1628 nd.ni_cnd.cn_nameiop = CREATE; 1629 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1630 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1631 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1632 if (dirp) 1633 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1634 if (error) { 1635 nfsm_reply(NFSX_WCCDATA(1)); 1636 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1637 if (dirp) 1638 vrele(dirp); 1639 vn_finished_write(mp, 0); 1640 return (0); 1641 } 1642 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1643 vtyp = nfsv3tov_type(*tl); 1644 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1645 error = NFSERR_BADTYPE; 1646 goto abort; 1647 } 1648 VATTR_NULL(&va); 1649 va.va_mode = 0; 1650 nfsm_srvsattr(&va); 1651 if (vtyp == VCHR || vtyp == VBLK) { 1652 dev_t rdev; 1653 1654 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1655 major = fxdr_unsigned(u_int32_t, *tl++); 1656 minor = fxdr_unsigned(u_int32_t, *tl); 1657 rdev = makedev(major, minor); 1658 if (major(rdev) != major || minor(rdev) != minor) { 1659 error = EINVAL; 1660 goto abort; 1661 } 1662 va.va_rdev = rdev; 1663 } 1664 1665 /* 1666 * Iff doesn't exist, create it. 1667 */ 1668 if (nd.ni_vp) { 1669 error = EEXIST; 1670 abort: 1671 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1672 if (nd.ni_dvp == nd.ni_vp) 1673 vrele(nd.ni_dvp); 1674 else 1675 vput(nd.ni_dvp); 1676 if (nd.ni_vp) 1677 vput(nd.ni_vp); 1678 goto out; 1679 } 1680 va.va_type = vtyp; 1681 if (vtyp == VSOCK) { 1682 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1683 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1684 } else { 1685 if (va.va_type != VFIFO && 1686 (error = suser(cred, (u_short *)0))) { 1687 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1688 vput(nd.ni_dvp); 1689 goto out; 1690 } 1691 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1692 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1693 if (error) { 1694 goto out; 1695 } 1696 if (error) 1697 goto out; 1698 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1699 vput(nd.ni_vp); 1700 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1701 error = EINVAL; 1702 } 1703 } 1704 out: 1705 vp = nd.ni_vp; 1706 if (!error) { 1707 memset((caddr_t)fhp, 0, sizeof(nfh)); 1708 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1709 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1710 if (!error) 1711 error = VOP_GETATTR(vp, &va, cred, lwp); 1712 vput(vp); 1713 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1714 } 1715 if (dirp) { 1716 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1717 vrele(dirp); 1718 } 1719 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1)); 1720 if (!error) { 1721 nfsm_srvpostop_fh(fhp); 1722 nfsm_srvpostop_attr(0, &va); 1723 } 1724 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1725 vn_finished_write(mp, 0); 1726 return (0); 1727 nfsmout: 1728 if (dirp) 1729 vrele(dirp); 1730 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1731 if (nd.ni_dvp == nd.ni_vp) 1732 vrele(nd.ni_dvp); 1733 else 1734 vput(nd.ni_dvp); 1735 if (nd.ni_vp) 1736 vput(nd.ni_vp); 1737 vn_finished_write(mp, 0); 1738 return (error); 1739 } 1740 1741 /* 1742 * nfs remove service 1743 */ 1744 int 1745 nfsrv_remove(nfsd, slp, lwp, mrq) 1746 struct nfsrv_descript *nfsd; 1747 struct nfssvc_sock *slp; 1748 struct lwp *lwp; 1749 struct mbuf **mrq; 1750 { 1751 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1752 struct mbuf *nam = nfsd->nd_nam; 1753 caddr_t dpos = nfsd->nd_dpos; 1754 struct ucred *cred = &nfsd->nd_cr; 1755 struct nameidata nd; 1756 u_int32_t *tl; 1757 int32_t t1; 1758 caddr_t bpos; 1759 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1760 int v3 = (nfsd->nd_flag & ND_NFSV3); 1761 char *cp2; 1762 struct mbuf *mb, *mreq; 1763 struct vnode *vp, *dirp; 1764 struct vattr dirfor, diraft; 1765 nfsfh_t nfh; 1766 fhandle_t *fhp; 1767 u_quad_t frev; 1768 struct mount *mp = NULL; 1769 1770 #ifndef nolint 1771 vp = (struct vnode *)0; 1772 #endif 1773 fhp = &nfh.fh_generic; 1774 nfsm_srvmtofh(fhp); 1775 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1776 return (ESTALE); 1777 vn_start_write(NULL, &mp, V_WAIT); 1778 nfsm_srvnamesiz(len); 1779 nd.ni_cnd.cn_cred = cred; 1780 nd.ni_cnd.cn_nameiop = DELETE; 1781 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1782 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1783 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1784 if (dirp) { 1785 if (v3) 1786 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1787 else 1788 vrele(dirp); 1789 } 1790 if (!error) { 1791 vp = nd.ni_vp; 1792 if (vp->v_type == VDIR && 1793 (error = suser(cred, (u_short *)0)) != 0) 1794 goto out; 1795 /* 1796 * The root of a mounted filesystem cannot be deleted. 1797 */ 1798 if (vp->v_flag & VROOT) { 1799 error = EBUSY; 1800 goto out; 1801 } 1802 out: 1803 if (!error) { 1804 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1805 nqsrv_getl(vp, ND_WRITE); 1806 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1807 } else { 1808 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1809 if (nd.ni_dvp == vp) 1810 vrele(nd.ni_dvp); 1811 else 1812 vput(nd.ni_dvp); 1813 vput(vp); 1814 } 1815 } 1816 if (dirp && v3) { 1817 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1818 vrele(dirp); 1819 } 1820 nfsm_reply(NFSX_WCCDATA(v3)); 1821 if (v3) { 1822 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1823 vn_finished_write(mp, 0); 1824 return (0); 1825 } 1826 vn_finished_write(mp, 0); 1827 nfsm_srvdone; 1828 } 1829 1830 /* 1831 * nfs rename service 1832 */ 1833 int 1834 nfsrv_rename(nfsd, slp, lwp, mrq) 1835 struct nfsrv_descript *nfsd; 1836 struct nfssvc_sock *slp; 1837 struct lwp *lwp; 1838 struct mbuf **mrq; 1839 { 1840 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1841 struct mbuf *nam = nfsd->nd_nam; 1842 caddr_t dpos = nfsd->nd_dpos; 1843 struct ucred *cred = &nfsd->nd_cr; 1844 u_int32_t *tl; 1845 int32_t t1; 1846 caddr_t bpos; 1847 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1848 uint32_t len, len2; 1849 int tdirfor_ret = 1, tdiraft_ret = 1; 1850 int v3 = (nfsd->nd_flag & ND_NFSV3); 1851 char *cp2; 1852 struct mbuf *mb, *mreq; 1853 struct nameidata fromnd, tond; 1854 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0; 1855 struct vnode *tdirp = (struct vnode *)0; 1856 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1857 nfsfh_t fnfh, tnfh; 1858 fhandle_t *ffhp, *tfhp; 1859 u_quad_t frev; 1860 uid_t saved_uid; 1861 struct mount *mp = NULL; 1862 1863 #ifndef nolint 1864 fvp = (struct vnode *)0; 1865 #endif 1866 ffhp = &fnfh.fh_generic; 1867 tfhp = &tnfh.fh_generic; 1868 fromnd.ni_cnd.cn_nameiop = 0; 1869 tond.ni_cnd.cn_nameiop = 0; 1870 nfsm_srvmtofh(ffhp); 1871 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) 1872 return (ESTALE); 1873 vn_start_write(NULL, &mp, V_WAIT); 1874 nfsm_srvnamesiz(len); 1875 /* 1876 * Remember our original uid so that we can reset cr_uid before 1877 * the second nfs_namei() call, in case it is remapped. 1878 */ 1879 saved_uid = cred->cr_uid; 1880 fromnd.ni_cnd.cn_cred = cred; 1881 fromnd.ni_cnd.cn_nameiop = DELETE; 1882 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART; 1883 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md, 1884 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1885 if (fdirp) { 1886 if (v3) 1887 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp); 1888 else { 1889 vrele(fdirp); 1890 fdirp = (struct vnode *)0; 1891 } 1892 } 1893 if (error) { 1894 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1895 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1896 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1897 if (fdirp) 1898 vrele(fdirp); 1899 vn_finished_write(mp, 0); 1900 return (0); 1901 } 1902 fvp = fromnd.ni_vp; 1903 nfsm_srvmtofh(tfhp); 1904 if (v3) { 1905 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1906 len2 = fxdr_unsigned(uint32_t, *tl); 1907 /* len2 will be checked by nfs_namei */ 1908 } 1909 else { 1910 /* NFSv2 */ 1911 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1912 } 1913 cred->cr_uid = saved_uid; 1914 tond.ni_cnd.cn_cred = cred; 1915 tond.ni_cnd.cn_nameiop = RENAME; 1916 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1917 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md, 1918 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1919 if (tdirp) { 1920 if (v3) 1921 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp); 1922 else { 1923 vrele(tdirp); 1924 tdirp = (struct vnode *)0; 1925 } 1926 } 1927 if (error) { 1928 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1929 vrele(fromnd.ni_dvp); 1930 vrele(fvp); 1931 goto out1; 1932 } 1933 tdvp = tond.ni_dvp; 1934 tvp = tond.ni_vp; 1935 if (tvp != NULL) { 1936 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1937 if (v3) 1938 error = EEXIST; 1939 else 1940 error = EISDIR; 1941 goto out; 1942 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1943 if (v3) 1944 error = EEXIST; 1945 else 1946 error = ENOTDIR; 1947 goto out; 1948 } 1949 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1950 if (v3) 1951 error = EXDEV; 1952 else 1953 error = ENOTEMPTY; 1954 goto out; 1955 } 1956 } 1957 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1958 if (v3) 1959 error = EXDEV; 1960 else 1961 error = ENOTEMPTY; 1962 goto out; 1963 } 1964 if (fvp->v_mount != tdvp->v_mount) { 1965 if (v3) 1966 error = EXDEV; 1967 else 1968 error = ENOTEMPTY; 1969 goto out; 1970 } 1971 if (fvp == tdvp) { 1972 if (v3) 1973 error = EINVAL; 1974 else 1975 error = ENOTEMPTY; 1976 } 1977 /* 1978 * If source is the same as the destination (that is the 1979 * same vnode with the same name in the same directory), 1980 * then there is nothing to do. 1981 */ 1982 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1983 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1984 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1985 fromnd.ni_cnd.cn_namelen)) 1986 error = -1; 1987 out: 1988 if (!error) { 1989 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1990 nqsrv_getl(tdvp, ND_WRITE); 1991 if (tvp) { 1992 nqsrv_getl(tvp, ND_WRITE); 1993 } 1994 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1995 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1996 } else { 1997 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 1998 if (tdvp == tvp) 1999 vrele(tdvp); 2000 else 2001 vput(tdvp); 2002 if (tvp) 2003 vput(tvp); 2004 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2005 vrele(fromnd.ni_dvp); 2006 vrele(fvp); 2007 if (error == -1) 2008 error = 0; 2009 } 2010 vrele(tond.ni_startdir); 2011 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2012 out1: 2013 if (fdirp) { 2014 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp); 2015 vrele(fdirp); 2016 } 2017 if (tdirp) { 2018 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp); 2019 vrele(tdirp); 2020 } 2021 vrele(fromnd.ni_startdir); 2022 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2023 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2024 if (v3) { 2025 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2026 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2027 } 2028 vn_finished_write(mp, 0); 2029 return (0); 2030 2031 nfsmout: 2032 if (fdirp) 2033 vrele(fdirp); 2034 if (tdirp) 2035 vrele(tdirp); 2036 if (tond.ni_cnd.cn_nameiop) { 2037 vrele(tond.ni_startdir); 2038 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2039 } 2040 if (fromnd.ni_cnd.cn_nameiop) { 2041 vrele(fromnd.ni_startdir); 2042 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2043 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2044 vrele(fromnd.ni_dvp); 2045 vrele(fvp); 2046 } 2047 return (error); 2048 } 2049 2050 /* 2051 * nfs link service 2052 */ 2053 int 2054 nfsrv_link(nfsd, slp, lwp, mrq) 2055 struct nfsrv_descript *nfsd; 2056 struct nfssvc_sock *slp; 2057 struct lwp *lwp; 2058 struct mbuf **mrq; 2059 { 2060 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2061 struct mbuf *nam = nfsd->nd_nam; 2062 caddr_t dpos = nfsd->nd_dpos; 2063 struct ucred *cred = &nfsd->nd_cr; 2064 struct nameidata nd; 2065 u_int32_t *tl; 2066 int32_t t1; 2067 caddr_t bpos; 2068 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2069 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2070 char *cp2; 2071 struct mbuf *mb, *mreq; 2072 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2073 struct vattr dirfor, diraft, at; 2074 nfsfh_t nfh, dnfh; 2075 fhandle_t *fhp, *dfhp; 2076 u_quad_t frev; 2077 struct mount *mp = NULL; 2078 2079 fhp = &nfh.fh_generic; 2080 dfhp = &dnfh.fh_generic; 2081 nfsm_srvmtofh(fhp); 2082 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2083 return (ESTALE); 2084 vn_start_write(NULL, &mp, V_WAIT); 2085 nfsm_srvmtofh(dfhp); 2086 nfsm_srvnamesiz(len); 2087 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam, 2088 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2089 if (error) { 2090 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2091 nfsm_srvpostop_attr(getret, &at); 2092 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2093 vn_finished_write(mp, 0); 2094 return (0); 2095 } 2096 if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0) 2097 goto out1; 2098 nd.ni_cnd.cn_cred = cred; 2099 nd.ni_cnd.cn_nameiop = CREATE; 2100 nd.ni_cnd.cn_flags = LOCKPARENT; 2101 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos, 2102 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2103 if (dirp) { 2104 if (v3) 2105 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2106 else { 2107 vrele(dirp); 2108 dirp = (struct vnode *)0; 2109 } 2110 } 2111 if (error) 2112 goto out1; 2113 xp = nd.ni_vp; 2114 if (xp != NULL) { 2115 error = EEXIST; 2116 goto out; 2117 } 2118 xp = nd.ni_dvp; 2119 if (vp->v_mount != xp->v_mount) 2120 error = EXDEV; 2121 out: 2122 if (!error) { 2123 nqsrv_getl(vp, ND_WRITE); 2124 nqsrv_getl(xp, ND_WRITE); 2125 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2126 } else { 2127 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2128 if (nd.ni_dvp == nd.ni_vp) 2129 vrele(nd.ni_dvp); 2130 else 2131 vput(nd.ni_dvp); 2132 if (nd.ni_vp) 2133 vrele(nd.ni_vp); 2134 } 2135 out1: 2136 if (v3) 2137 getret = VOP_GETATTR(vp, &at, cred, lwp); 2138 if (dirp) { 2139 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2140 vrele(dirp); 2141 } 2142 vrele(vp); 2143 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2144 if (v3) { 2145 nfsm_srvpostop_attr(getret, &at); 2146 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2147 vn_finished_write(mp, 0); 2148 return (0); 2149 } 2150 vn_finished_write(mp, 0); 2151 nfsm_srvdone; 2152 } 2153 2154 /* 2155 * nfs symbolic link service 2156 */ 2157 int 2158 nfsrv_symlink(nfsd, slp, lwp, mrq) 2159 struct nfsrv_descript *nfsd; 2160 struct nfssvc_sock *slp; 2161 struct lwp *lwp; 2162 struct mbuf **mrq; 2163 { 2164 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2165 struct mbuf *nam = nfsd->nd_nam; 2166 caddr_t dpos = nfsd->nd_dpos; 2167 struct ucred *cred = &nfsd->nd_cr; 2168 struct vattr va, dirfor, diraft; 2169 struct nameidata nd; 2170 u_int32_t *tl; 2171 int32_t t1; 2172 struct nfsv2_sattr *sp; 2173 char *bpos, *pathcp = NULL, *cp2; 2174 struct uio io; 2175 struct iovec iv; 2176 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1; 2177 uint32_t len, len2; 2178 int v3 = (nfsd->nd_flag & ND_NFSV3); 2179 struct mbuf *mb, *mreq; 2180 struct vnode *dirp = (struct vnode *)0; 2181 nfsfh_t nfh; 2182 fhandle_t *fhp; 2183 u_quad_t frev; 2184 struct mount *mp = NULL; 2185 2186 nd.ni_cnd.cn_nameiop = 0; 2187 fhp = &nfh.fh_generic; 2188 nfsm_srvmtofh(fhp); 2189 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2190 return (ESTALE); 2191 vn_start_write(NULL, &mp, V_WAIT); 2192 nfsm_srvnamesiz(len); 2193 nd.ni_cnd.cn_cred = cred; 2194 nd.ni_cnd.cn_nameiop = CREATE; 2195 nd.ni_cnd.cn_flags = LOCKPARENT; 2196 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2197 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2198 if (dirp) { 2199 if (v3) 2200 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2201 else { 2202 vrele(dirp); 2203 dirp = (struct vnode *)0; 2204 } 2205 } 2206 if (error) 2207 goto out; 2208 VATTR_NULL(&va); 2209 va.va_type = VLNK; 2210 if (v3) { 2211 va.va_mode = 0; 2212 nfsm_srvsattr(&va); 2213 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2214 len2 = fxdr_unsigned(uint32_t, *tl); 2215 if (len2 > PATH_MAX) { 2216 /* XXX should check _PC_NO_TRUNC */ 2217 error = ENAMETOOLONG; 2218 goto abortop; 2219 } 2220 } 2221 else { 2222 /* NFSv2 */ 2223 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2224 } 2225 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2226 iv.iov_base = pathcp; 2227 iv.iov_len = len2; 2228 io.uio_resid = len2; 2229 io.uio_offset = 0; 2230 io.uio_iov = &iv; 2231 io.uio_iovcnt = 1; 2232 io.uio_rw = UIO_READ; 2233 UIO_SETUP_SYSSPACE(&io); 2234 nfsm_mtouio(&io, len2); 2235 if (!v3) { 2236 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2237 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2238 } 2239 *(pathcp + len2) = '\0'; 2240 if (nd.ni_vp) { 2241 error = EEXIST; 2242 abortop: 2243 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2244 if (nd.ni_dvp == nd.ni_vp) 2245 vrele(nd.ni_dvp); 2246 else 2247 vput(nd.ni_dvp); 2248 if (nd.ni_vp) 2249 vrele(nd.ni_vp); 2250 goto out; 2251 } 2252 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2253 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2254 if (!error) { 2255 if (v3) { 2256 memset((caddr_t)fhp, 0, sizeof(nfh)); 2257 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsidx; 2258 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2259 if (!error) 2260 error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp); 2261 vput(nd.ni_vp); 2262 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2263 } else { 2264 vput(nd.ni_vp); 2265 } 2266 } 2267 out: 2268 if (pathcp) 2269 free(pathcp, M_TEMP); 2270 if (dirp) { 2271 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2272 vrele(dirp); 2273 } 2274 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2275 if (v3) { 2276 if (!error) { 2277 nfsm_srvpostop_fh(fhp); 2278 nfsm_srvpostop_attr(0, &va); 2279 } 2280 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2281 } 2282 vn_finished_write(mp, 0); 2283 return (0); 2284 nfsmout: 2285 if (dirp) 2286 vrele(dirp); 2287 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2288 if (nd.ni_dvp == nd.ni_vp) 2289 vrele(nd.ni_dvp); 2290 else 2291 vput(nd.ni_dvp); 2292 if (nd.ni_vp) 2293 vrele(nd.ni_vp); 2294 if (pathcp) 2295 free(pathcp, M_TEMP); 2296 vn_finished_write(mp, 0); 2297 return (error); 2298 } 2299 2300 /* 2301 * nfs mkdir service 2302 */ 2303 int 2304 nfsrv_mkdir(nfsd, slp, lwp, mrq) 2305 struct nfsrv_descript *nfsd; 2306 struct nfssvc_sock *slp; 2307 struct lwp *lwp; 2308 struct mbuf **mrq; 2309 { 2310 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2311 struct mbuf *nam = nfsd->nd_nam; 2312 caddr_t dpos = nfsd->nd_dpos; 2313 struct ucred *cred = &nfsd->nd_cr; 2314 struct vattr va, dirfor, diraft; 2315 struct nfs_fattr *fp; 2316 struct nameidata nd; 2317 caddr_t cp; 2318 u_int32_t *tl; 2319 int32_t t1; 2320 caddr_t bpos; 2321 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2322 int v3 = (nfsd->nd_flag & ND_NFSV3); 2323 char *cp2; 2324 struct mbuf *mb, *mreq; 2325 struct vnode *vp, *dirp = (struct vnode *)0; 2326 nfsfh_t nfh; 2327 fhandle_t *fhp; 2328 u_quad_t frev; 2329 struct mount *mp = NULL; 2330 2331 fhp = &nfh.fh_generic; 2332 nfsm_srvmtofh(fhp); 2333 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2334 return (ESTALE); 2335 vn_start_write(NULL, &mp, V_WAIT); 2336 nfsm_srvnamesiz(len); 2337 nd.ni_cnd.cn_cred = cred; 2338 nd.ni_cnd.cn_nameiop = CREATE; 2339 nd.ni_cnd.cn_flags = LOCKPARENT; 2340 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2341 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2342 if (dirp) { 2343 if (v3) 2344 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2345 else { 2346 vrele(dirp); 2347 dirp = (struct vnode *)0; 2348 } 2349 } 2350 if (error) { 2351 nfsm_reply(NFSX_WCCDATA(v3)); 2352 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2353 if (dirp) 2354 vrele(dirp); 2355 vn_finished_write(mp, 0); 2356 return (0); 2357 } 2358 VATTR_NULL(&va); 2359 if (v3) { 2360 va.va_mode = 0; 2361 nfsm_srvsattr(&va); 2362 } else { 2363 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2364 va.va_mode = nfstov_mode(*tl++); 2365 } 2366 va.va_type = VDIR; 2367 vp = nd.ni_vp; 2368 if (vp != NULL) { 2369 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2370 if (nd.ni_dvp == vp) 2371 vrele(nd.ni_dvp); 2372 else 2373 vput(nd.ni_dvp); 2374 vrele(vp); 2375 error = EEXIST; 2376 goto out; 2377 } 2378 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2379 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2380 if (!error) { 2381 vp = nd.ni_vp; 2382 memset((caddr_t)fhp, 0, sizeof(nfh)); 2383 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 2384 error = VFS_VPTOFH(vp, &fhp->fh_fid); 2385 if (!error) 2386 error = VOP_GETATTR(vp, &va, cred, lwp); 2387 vput(vp); 2388 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2389 } 2390 out: 2391 if (dirp) { 2392 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2393 vrele(dirp); 2394 } 2395 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2396 if (v3) { 2397 if (!error) { 2398 nfsm_srvpostop_fh(fhp); 2399 nfsm_srvpostop_attr(0, &va); 2400 } 2401 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2402 } else { 2403 nfsm_srvfhtom(fhp, v3); 2404 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2405 nfsm_srvfillattr(&va, fp); 2406 } 2407 vn_finished_write(mp, 0); 2408 return (0); 2409 nfsmout: 2410 if (dirp) 2411 vrele(dirp); 2412 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2413 if (nd.ni_dvp == nd.ni_vp) 2414 vrele(nd.ni_dvp); 2415 else 2416 vput(nd.ni_dvp); 2417 if (nd.ni_vp) 2418 vrele(nd.ni_vp); 2419 vn_finished_write(mp, 0); 2420 return (error); 2421 } 2422 2423 /* 2424 * nfs rmdir service 2425 */ 2426 int 2427 nfsrv_rmdir(nfsd, slp, lwp, mrq) 2428 struct nfsrv_descript *nfsd; 2429 struct nfssvc_sock *slp; 2430 struct lwp *lwp; 2431 struct mbuf **mrq; 2432 { 2433 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2434 struct mbuf *nam = nfsd->nd_nam; 2435 caddr_t dpos = nfsd->nd_dpos; 2436 struct ucred *cred = &nfsd->nd_cr; 2437 u_int32_t *tl; 2438 int32_t t1; 2439 caddr_t bpos; 2440 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2441 int v3 = (nfsd->nd_flag & ND_NFSV3); 2442 char *cp2; 2443 struct mbuf *mb, *mreq; 2444 struct vnode *vp, *dirp = (struct vnode *)0; 2445 struct vattr dirfor, diraft; 2446 nfsfh_t nfh; 2447 fhandle_t *fhp; 2448 struct nameidata nd; 2449 u_quad_t frev; 2450 struct mount *mp = NULL; 2451 2452 fhp = &nfh.fh_generic; 2453 nfsm_srvmtofh(fhp); 2454 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2455 return (ESTALE); 2456 vn_start_write(NULL, &mp, V_WAIT); 2457 nfsm_srvnamesiz(len); 2458 nd.ni_cnd.cn_cred = cred; 2459 nd.ni_cnd.cn_nameiop = DELETE; 2460 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2461 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2462 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2463 if (dirp) { 2464 if (v3) 2465 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2466 else { 2467 vrele(dirp); 2468 dirp = (struct vnode *)0; 2469 } 2470 } 2471 if (error) { 2472 nfsm_reply(NFSX_WCCDATA(v3)); 2473 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2474 if (dirp) 2475 vrele(dirp); 2476 vn_finished_write(mp, 0); 2477 return (0); 2478 } 2479 vp = nd.ni_vp; 2480 if (vp->v_type != VDIR) { 2481 error = ENOTDIR; 2482 goto out; 2483 } 2484 /* 2485 * No rmdir "." please. 2486 */ 2487 if (nd.ni_dvp == vp) { 2488 error = EINVAL; 2489 goto out; 2490 } 2491 /* 2492 * The root of a mounted filesystem cannot be deleted. 2493 */ 2494 if (vp->v_flag & VROOT) 2495 error = EBUSY; 2496 out: 2497 if (!error) { 2498 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2499 nqsrv_getl(vp, ND_WRITE); 2500 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2501 } else { 2502 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2503 if (nd.ni_dvp == nd.ni_vp) 2504 vrele(nd.ni_dvp); 2505 else 2506 vput(nd.ni_dvp); 2507 vput(vp); 2508 } 2509 if (dirp) { 2510 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2511 vrele(dirp); 2512 } 2513 nfsm_reply(NFSX_WCCDATA(v3)); 2514 if (v3) { 2515 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2516 vn_finished_write(mp, 0); 2517 return (0); 2518 } 2519 vn_finished_write(mp, 0); 2520 nfsm_srvdone; 2521 } 2522 2523 /* 2524 * nfs readdir service 2525 * - mallocs what it thinks is enough to read 2526 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2527 * - calls VOP_READDIR() 2528 * - loops around building the reply 2529 * if the output generated exceeds count break out of loop 2530 * The nfsm_clget macro is used here so that the reply will be packed 2531 * tightly in mbuf clusters. 2532 * - it only knows that it has encountered eof when the VOP_READDIR() 2533 * reads nothing 2534 * - as such one readdir rpc will return eof false although you are there 2535 * and then the next will return eof 2536 * - it trims out records with d_fileno == 0 2537 * this doesn't matter for Unix clients, but they might confuse clients 2538 * for other os'. 2539 * - it trims out records with d_type == DT_WHT 2540 * these cannot be seen through NFS (unless we extend the protocol) 2541 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2542 * than requested, but this may not apply to all filesystems. For 2543 * example, client NFS does not { although it is never remote mounted 2544 * anyhow } 2545 * The alternate call nfsrv_readdirplus() does lookups as well. 2546 * PS: The NFS protocol spec. does not clarify what the "count" byte 2547 * argument is a count of.. just name strings and file id's or the 2548 * entire reply rpc or ... 2549 * I tried just file name and id sizes and it confused the Sun client, 2550 * so I am using the full rpc size now. The "paranoia.." comment refers 2551 * to including the status longwords that are not a part of the dir. 2552 * "entry" structures, but are in the rpc. 2553 */ 2554 2555 #define NFS_SRVDIRBLKSIZ 1024 2556 2557 struct flrep { 2558 nfsuint64 fl_off; 2559 u_int32_t fl_postopok; 2560 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2561 u_int32_t fl_fhok; 2562 u_int32_t fl_fhsize; 2563 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)]; 2564 }; 2565 2566 int 2567 nfsrv_readdir(nfsd, slp, lwp, mrq) 2568 struct nfsrv_descript *nfsd; 2569 struct nfssvc_sock *slp; 2570 struct lwp *lwp; 2571 struct mbuf **mrq; 2572 { 2573 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2574 struct mbuf *nam = nfsd->nd_nam; 2575 caddr_t dpos = nfsd->nd_dpos; 2576 struct ucred *cred = &nfsd->nd_cr; 2577 char *bp, *be; 2578 struct mbuf *mp; 2579 struct dirent *dp; 2580 caddr_t cp; 2581 u_int32_t *tl; 2582 int32_t t1; 2583 caddr_t bpos; 2584 struct mbuf *mb, *mreq, *mp2; 2585 char *cpos, *cend, *cp2, *rbuf; 2586 struct vnode *vp; 2587 struct vattr at; 2588 nfsfh_t nfh; 2589 fhandle_t *fhp; 2590 struct uio io; 2591 struct iovec iv; 2592 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2593 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2594 int v3 = (nfsd->nd_flag & ND_NFSV3); 2595 u_quad_t frev, off, toff, verf; 2596 off_t *cookies = NULL, *cookiep; 2597 nfsuint64 jar; 2598 2599 fhp = &nfh.fh_generic; 2600 nfsm_srvmtofh(fhp); 2601 if (v3) { 2602 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2603 toff = fxdr_hyper(tl); 2604 tl += 2; 2605 verf = fxdr_hyper(tl); 2606 tl += 2; 2607 } else { 2608 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2609 toff = fxdr_unsigned(u_quad_t, *tl++); 2610 } 2611 off = toff; 2612 cnt = fxdr_unsigned(int, *tl); 2613 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2614 xfer = NFS_SRVMAXDATA(nfsd); 2615 if (siz > xfer) 2616 siz = xfer; 2617 fullsiz = siz; 2618 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2619 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2620 if (!error && vp->v_type != VDIR) { 2621 error = ENOTDIR; 2622 vput(vp); 2623 } 2624 if (error) { 2625 nfsm_reply(NFSX_UNSIGNED); 2626 nfsm_srvpostop_attr(getret, &at); 2627 return (0); 2628 } 2629 nqsrv_getl(vp, ND_READ); 2630 if (v3) { 2631 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2632 #ifdef NFS3_STRICTVERF 2633 /* 2634 * XXX This check is too strict for Solaris 2.5 clients. 2635 */ 2636 if (!error && toff && verf != at.va_filerev) 2637 error = NFSERR_BAD_COOKIE; 2638 #endif 2639 } 2640 if (!error) 2641 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2642 if (error) { 2643 vput(vp); 2644 nfsm_reply(NFSX_POSTOPATTR(v3)); 2645 nfsm_srvpostop_attr(getret, &at); 2646 return (0); 2647 } 2648 VOP_UNLOCK(vp, 0); 2649 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2650 again: 2651 iv.iov_base = rbuf; 2652 iv.iov_len = fullsiz; 2653 io.uio_iov = &iv; 2654 io.uio_iovcnt = 1; 2655 io.uio_offset = (off_t)off; 2656 io.uio_resid = fullsiz; 2657 io.uio_rw = UIO_READ; 2658 UIO_SETUP_SYSSPACE(&io); 2659 eofflag = 0; 2660 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2661 2662 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2663 2664 off = (off_t)io.uio_offset; 2665 if (!cookies && !error) 2666 error = NFSERR_PERM; 2667 if (v3) { 2668 getret = VOP_GETATTR(vp, &at, cred, lwp); 2669 if (!error) 2670 error = getret; 2671 } 2672 2673 VOP_UNLOCK(vp, 0); 2674 if (error) { 2675 vrele(vp); 2676 free((caddr_t)rbuf, M_TEMP); 2677 if (cookies) 2678 free((caddr_t)cookies, M_TEMP); 2679 nfsm_reply(NFSX_POSTOPATTR(v3)); 2680 nfsm_srvpostop_attr(getret, &at); 2681 return (0); 2682 } 2683 if (io.uio_resid) { 2684 siz -= io.uio_resid; 2685 2686 /* 2687 * If nothing read, return eof 2688 * rpc reply 2689 */ 2690 if (siz == 0) { 2691 vrele(vp); 2692 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2693 2 * NFSX_UNSIGNED); 2694 if (v3) { 2695 nfsm_srvpostop_attr(getret, &at); 2696 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2697 txdr_hyper(at.va_filerev, tl); 2698 tl += 2; 2699 } else 2700 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2701 *tl++ = nfs_false; 2702 *tl = nfs_true; 2703 free((caddr_t)rbuf, M_TEMP); 2704 free((caddr_t)cookies, M_TEMP); 2705 return (0); 2706 } 2707 } 2708 2709 /* 2710 * Check for degenerate cases of nothing useful read. 2711 * If so go try again 2712 */ 2713 cpos = rbuf; 2714 cend = rbuf + siz; 2715 dp = (struct dirent *)cpos; 2716 cookiep = cookies; 2717 2718 while (cpos < cend && ncookies > 0 && 2719 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2720 cpos += dp->d_reclen; 2721 dp = (struct dirent *)cpos; 2722 cookiep++; 2723 ncookies--; 2724 } 2725 if (cpos >= cend || ncookies == 0) { 2726 toff = off; 2727 siz = fullsiz; 2728 goto again; 2729 } 2730 2731 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2732 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2733 if (v3) { 2734 nfsm_srvpostop_attr(getret, &at); 2735 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2736 txdr_hyper(at.va_filerev, tl); 2737 } 2738 mp = mp2 = mb; 2739 bp = bpos; 2740 be = bp + M_TRAILINGSPACE(mp); 2741 2742 /* Loop through the records and build reply */ 2743 while (cpos < cend && ncookies > 0) { 2744 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2745 nlen = dp->d_namlen; 2746 rem = nfsm_rndup(nlen)-nlen; 2747 len += (4 * NFSX_UNSIGNED + nlen + rem); 2748 if (v3) 2749 len += 2 * NFSX_UNSIGNED; 2750 if (len > cnt) { 2751 eofflag = 0; 2752 break; 2753 } 2754 /* 2755 * Build the directory record xdr from 2756 * the dirent entry. 2757 */ 2758 nfsm_clget; 2759 *tl = nfs_true; 2760 bp += NFSX_UNSIGNED; 2761 if (v3) { 2762 nfsm_clget; 2763 *tl = txdr_unsigned(dp->d_fileno >> 32); 2764 bp += NFSX_UNSIGNED; 2765 } 2766 nfsm_clget; 2767 *tl = txdr_unsigned(dp->d_fileno); 2768 bp += NFSX_UNSIGNED; 2769 nfsm_clget; 2770 *tl = txdr_unsigned(nlen); 2771 bp += NFSX_UNSIGNED; 2772 2773 /* And loop around copying the name */ 2774 xfer = nlen; 2775 cp = dp->d_name; 2776 while (xfer > 0) { 2777 nfsm_clget; 2778 if ((bp+xfer) > be) 2779 tsiz = be-bp; 2780 else 2781 tsiz = xfer; 2782 memcpy(bp, cp, tsiz); 2783 bp += tsiz; 2784 xfer -= tsiz; 2785 if (xfer > 0) 2786 cp += tsiz; 2787 } 2788 /* And null pad to an int32_t boundary */ 2789 for (i = 0; i < rem; i++) 2790 *bp++ = '\0'; 2791 nfsm_clget; 2792 2793 /* Finish off the record */ 2794 txdr_hyper(*cookiep, &jar); 2795 if (v3) { 2796 *tl = jar.nfsuquad[0]; 2797 bp += NFSX_UNSIGNED; 2798 nfsm_clget; 2799 } 2800 *tl = jar.nfsuquad[1]; 2801 bp += NFSX_UNSIGNED; 2802 } 2803 cpos += dp->d_reclen; 2804 dp = (struct dirent *)cpos; 2805 cookiep++; 2806 ncookies--; 2807 } 2808 vrele(vp); 2809 nfsm_clget; 2810 *tl = nfs_false; 2811 bp += NFSX_UNSIGNED; 2812 nfsm_clget; 2813 if (eofflag) 2814 *tl = nfs_true; 2815 else 2816 *tl = nfs_false; 2817 bp += NFSX_UNSIGNED; 2818 if (mp != mb) { 2819 if (bp < be) 2820 mp->m_len = bp - mtod(mp, caddr_t); 2821 } else 2822 mp->m_len += bp - bpos; 2823 free((caddr_t)rbuf, M_TEMP); 2824 free((caddr_t)cookies, M_TEMP); 2825 nfsm_srvdone; 2826 } 2827 2828 int 2829 nfsrv_readdirplus(nfsd, slp, lwp, mrq) 2830 struct nfsrv_descript *nfsd; 2831 struct nfssvc_sock *slp; 2832 struct lwp *lwp; 2833 struct mbuf **mrq; 2834 { 2835 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2836 struct mbuf *nam = nfsd->nd_nam; 2837 caddr_t dpos = nfsd->nd_dpos; 2838 struct ucred *cred = &nfsd->nd_cr; 2839 char *bp, *be; 2840 struct mbuf *mp; 2841 struct dirent *dp; 2842 caddr_t cp; 2843 u_int32_t *tl; 2844 int32_t t1; 2845 caddr_t bpos; 2846 struct mbuf *mb, *mreq, *mp2; 2847 char *cpos, *cend, *cp2, *rbuf; 2848 struct vnode *vp, *nvp; 2849 struct flrep fl; 2850 nfsfh_t nfh; 2851 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh; 2852 struct uio io; 2853 struct iovec iv; 2854 struct vattr va, at, *vap = &va; 2855 struct nfs_fattr *fp; 2856 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2857 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2858 u_quad_t frev, off, toff, verf; 2859 off_t *cookies = NULL, *cookiep; 2860 2861 fhp = &nfh.fh_generic; 2862 nfsm_srvmtofh(fhp); 2863 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2864 toff = fxdr_hyper(tl); 2865 tl += 2; 2866 verf = fxdr_hyper(tl); 2867 tl += 2; 2868 siz = fxdr_unsigned(int, *tl++); 2869 cnt = fxdr_unsigned(int, *tl); 2870 off = toff; 2871 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2872 xfer = NFS_SRVMAXDATA(nfsd); 2873 if (siz > xfer) 2874 siz = xfer; 2875 fullsiz = siz; 2876 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2877 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2878 if (!error && vp->v_type != VDIR) { 2879 error = ENOTDIR; 2880 vput(vp); 2881 } 2882 if (error) { 2883 nfsm_reply(NFSX_UNSIGNED); 2884 nfsm_srvpostop_attr(getret, &at); 2885 return (0); 2886 } 2887 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2888 #ifdef NFS3_STRICTVERF 2889 /* 2890 * XXX This check is too strict for Solaris 2.5 clients. 2891 */ 2892 if (!error && toff && verf != at.va_filerev) 2893 error = NFSERR_BAD_COOKIE; 2894 #endif 2895 if (!error) { 2896 nqsrv_getl(vp, ND_READ); 2897 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2898 } 2899 if (error) { 2900 vput(vp); 2901 nfsm_reply(NFSX_V3POSTOPATTR); 2902 nfsm_srvpostop_attr(getret, &at); 2903 return (0); 2904 } 2905 VOP_UNLOCK(vp, 0); 2906 2907 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2908 again: 2909 iv.iov_base = rbuf; 2910 iv.iov_len = fullsiz; 2911 io.uio_iov = &iv; 2912 io.uio_iovcnt = 1; 2913 io.uio_offset = (off_t)off; 2914 io.uio_resid = fullsiz; 2915 io.uio_rw = UIO_READ; 2916 UIO_SETUP_SYSSPACE(&io); 2917 eofflag = 0; 2918 2919 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2920 2921 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2922 2923 off = (u_quad_t)io.uio_offset; 2924 getret = VOP_GETATTR(vp, &at, cred, lwp); 2925 2926 VOP_UNLOCK(vp, 0); 2927 2928 /* 2929 * If the VGET operation doesn't work for this filesystem, 2930 * we can't support readdirplus. Returning NOTSUPP should 2931 * make clients fall back to plain readdir. 2932 * There's no need to check for VPTOFH as well, we wouldn't 2933 * even be here otherwise. 2934 */ 2935 if (!getret) { 2936 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2937 getret = (getret == EOPNOTSUPP) ? 2938 NFSERR_NOTSUPP : NFSERR_IO; 2939 else 2940 vput(nvp); 2941 } 2942 2943 if (!cookies && !error) 2944 error = NFSERR_PERM; 2945 if (!error) 2946 error = getret; 2947 if (error) { 2948 vrele(vp); 2949 if (cookies) 2950 free((caddr_t)cookies, M_TEMP); 2951 free((caddr_t)rbuf, M_TEMP); 2952 nfsm_reply(NFSX_V3POSTOPATTR); 2953 nfsm_srvpostop_attr(getret, &at); 2954 return (0); 2955 } 2956 if (io.uio_resid) { 2957 siz -= io.uio_resid; 2958 2959 /* 2960 * If nothing read, return eof 2961 * rpc reply 2962 */ 2963 if (siz == 0) { 2964 vrele(vp); 2965 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2966 2 * NFSX_UNSIGNED); 2967 nfsm_srvpostop_attr(getret, &at); 2968 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2969 txdr_hyper(at.va_filerev, tl); 2970 tl += 2; 2971 *tl++ = nfs_false; 2972 *tl = nfs_true; 2973 free((caddr_t)cookies, M_TEMP); 2974 free((caddr_t)rbuf, M_TEMP); 2975 return (0); 2976 } 2977 } 2978 2979 /* 2980 * Check for degenerate cases of nothing useful read. 2981 * If so go try again 2982 */ 2983 cpos = rbuf; 2984 cend = rbuf + siz; 2985 dp = (struct dirent *)cpos; 2986 cookiep = cookies; 2987 2988 while (cpos < cend && ncookies > 0 && 2989 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2990 cpos += dp->d_reclen; 2991 dp = (struct dirent *)cpos; 2992 cookiep++; 2993 ncookies--; 2994 } 2995 if (cpos >= cend || ncookies == 0) { 2996 toff = off; 2997 siz = fullsiz; 2998 goto again; 2999 } 3000 3001 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3002 nfsm_reply(cnt); 3003 nfsm_srvpostop_attr(getret, &at); 3004 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3005 txdr_hyper(at.va_filerev, tl); 3006 mp = mp2 = mb; 3007 bp = bpos; 3008 be = bp + M_TRAILINGSPACE(mp); 3009 3010 /* Loop through the records and build reply */ 3011 while (cpos < cend && ncookies > 0) { 3012 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3013 nlen = dp->d_namlen; 3014 rem = nfsm_rndup(nlen)-nlen; 3015 3016 /* 3017 * For readdir_and_lookup get the vnode using 3018 * the file number. 3019 */ 3020 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3021 goto invalid; 3022 memset((caddr_t)nfhp, 0, NFSX_V3FH); 3023 nfhp->fh_fsid = 3024 nvp->v_mount->mnt_stat.f_fsidx; 3025 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) { 3026 vput(nvp); 3027 goto invalid; 3028 } 3029 if (VOP_GETATTR(nvp, vap, cred, lwp)) { 3030 vput(nvp); 3031 goto invalid; 3032 } 3033 vput(nvp); 3034 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 3035 3036 /* 3037 * If either the dircount or maxcount will be 3038 * exceeded, get out now. Both of these lengths 3039 * are calculated conservatively, including all 3040 * XDR overheads. 3041 */ 3042 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3043 NFSX_V3POSTOPATTR); 3044 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3045 if (len > cnt || dirlen > fullsiz) { 3046 eofflag = 0; 3047 break; 3048 } 3049 3050 /* 3051 * Build the directory record xdr from 3052 * the dirent entry. 3053 */ 3054 fp = (struct nfs_fattr *)&fl.fl_fattr; 3055 nfsm_srvfillattr(vap, fp); 3056 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3057 fl.fl_fhok = nfs_true; 3058 fl.fl_postopok = nfs_true; 3059 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3060 3061 nfsm_clget; 3062 *tl = nfs_true; 3063 bp += NFSX_UNSIGNED; 3064 nfsm_clget; 3065 *tl = txdr_unsigned(dp->d_fileno >> 32); 3066 bp += NFSX_UNSIGNED; 3067 nfsm_clget; 3068 *tl = txdr_unsigned(dp->d_fileno); 3069 bp += NFSX_UNSIGNED; 3070 nfsm_clget; 3071 *tl = txdr_unsigned(nlen); 3072 bp += NFSX_UNSIGNED; 3073 3074 /* And loop around copying the name */ 3075 xfer = nlen; 3076 cp = dp->d_name; 3077 while (xfer > 0) { 3078 nfsm_clget; 3079 if ((bp + xfer) > be) 3080 tsiz = be - bp; 3081 else 3082 tsiz = xfer; 3083 memcpy(bp, cp, tsiz); 3084 bp += tsiz; 3085 xfer -= tsiz; 3086 if (xfer > 0) 3087 cp += tsiz; 3088 } 3089 /* And null pad to an int32_t boundary */ 3090 for (i = 0; i < rem; i++) 3091 *bp++ = '\0'; 3092 3093 /* 3094 * Now copy the flrep structure out. 3095 */ 3096 xfer = sizeof (struct flrep); 3097 cp = (caddr_t)&fl; 3098 while (xfer > 0) { 3099 nfsm_clget; 3100 if ((bp + xfer) > be) 3101 tsiz = be - bp; 3102 else 3103 tsiz = xfer; 3104 memcpy(bp, cp, tsiz); 3105 bp += tsiz; 3106 xfer -= tsiz; 3107 if (xfer > 0) 3108 cp += tsiz; 3109 } 3110 } 3111 invalid: 3112 cpos += dp->d_reclen; 3113 dp = (struct dirent *)cpos; 3114 cookiep++; 3115 ncookies--; 3116 } 3117 vrele(vp); 3118 nfsm_clget; 3119 *tl = nfs_false; 3120 bp += NFSX_UNSIGNED; 3121 nfsm_clget; 3122 if (eofflag) 3123 *tl = nfs_true; 3124 else 3125 *tl = nfs_false; 3126 bp += NFSX_UNSIGNED; 3127 if (mp != mb) { 3128 if (bp < be) 3129 mp->m_len = bp - mtod(mp, caddr_t); 3130 } else 3131 mp->m_len += bp - bpos; 3132 free((caddr_t)cookies, M_TEMP); 3133 free((caddr_t)rbuf, M_TEMP); 3134 nfsm_srvdone; 3135 } 3136 3137 /* 3138 * nfs commit service 3139 */ 3140 int 3141 nfsrv_commit(nfsd, slp, lwp, mrq) 3142 struct nfsrv_descript *nfsd; 3143 struct nfssvc_sock *slp; 3144 struct lwp *lwp; 3145 struct mbuf **mrq; 3146 { 3147 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3148 struct mbuf *nam = nfsd->nd_nam; 3149 caddr_t dpos = nfsd->nd_dpos; 3150 struct ucred *cred = &nfsd->nd_cr; 3151 struct vattr bfor, aft; 3152 struct vnode *vp; 3153 nfsfh_t nfh; 3154 fhandle_t *fhp; 3155 u_int32_t *tl; 3156 int32_t t1; 3157 caddr_t bpos; 3158 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3159 uint32_t cnt; 3160 char *cp2; 3161 struct mbuf *mb, *mreq; 3162 u_quad_t frev, off, end; 3163 struct mount *mp = NULL; 3164 3165 fhp = &nfh.fh_generic; 3166 nfsm_srvmtofh(fhp); 3167 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 3168 return (ESTALE); 3169 vn_start_write(NULL, &mp, V_WAIT); 3170 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3171 3172 off = fxdr_hyper(tl); 3173 tl += 2; 3174 cnt = fxdr_unsigned(uint32_t, *tl); 3175 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3176 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3177 if (error) { 3178 nfsm_reply(2 * NFSX_UNSIGNED); 3179 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3180 vn_finished_write(mp, 0); 3181 return (0); 3182 } 3183 for_ret = VOP_GETATTR(vp, &bfor, cred, lwp); 3184 end = (cnt > 0) ? off + cnt : vp->v_size; 3185 if (end < off || end > vp->v_size) 3186 end = vp->v_size; 3187 if (off < vp->v_size) 3188 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp); 3189 /* else error == 0, from nfsrv_fhtovp() */ 3190 aft_ret = VOP_GETATTR(vp, &aft, cred, lwp); 3191 vput(vp); 3192 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3193 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3194 if (!error) { 3195 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3196 *tl++ = txdr_unsigned(boottime.tv_sec); 3197 *tl = txdr_unsigned(boottime.tv_usec); 3198 } else { 3199 vn_finished_write(mp, 0); 3200 return (0); 3201 } 3202 vn_finished_write(mp, 0); 3203 nfsm_srvdone; 3204 } 3205 3206 /* 3207 * nfs statfs service 3208 */ 3209 int 3210 nfsrv_statfs(nfsd, slp, lwp, mrq) 3211 struct nfsrv_descript *nfsd; 3212 struct nfssvc_sock *slp; 3213 struct lwp *lwp; 3214 struct mbuf **mrq; 3215 { 3216 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3217 struct mbuf *nam = nfsd->nd_nam; 3218 caddr_t dpos = nfsd->nd_dpos; 3219 struct ucred *cred = &nfsd->nd_cr; 3220 struct statvfs *sf; 3221 struct nfs_statfs *sfp; 3222 u_int32_t *tl; 3223 int32_t t1; 3224 caddr_t bpos; 3225 int error = 0, rdonly, cache = 0, getret = 1; 3226 int v3 = (nfsd->nd_flag & ND_NFSV3); 3227 char *cp2; 3228 struct mbuf *mb, *mreq; 3229 struct vnode *vp; 3230 struct vattr at; 3231 nfsfh_t nfh; 3232 fhandle_t *fhp; 3233 struct statvfs statvfs; 3234 u_quad_t frev, tval; 3235 3236 fhp = &nfh.fh_generic; 3237 nfsm_srvmtofh(fhp); 3238 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3239 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3240 if (error) { 3241 nfsm_reply(NFSX_UNSIGNED); 3242 nfsm_srvpostop_attr(getret, &at); 3243 return (0); 3244 } 3245 sf = &statvfs; 3246 error = VFS_STATVFS(vp->v_mount, sf, lwp); 3247 getret = VOP_GETATTR(vp, &at, cred, lwp); 3248 vput(vp); 3249 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3250 if (v3) 3251 nfsm_srvpostop_attr(getret, &at); 3252 if (error) 3253 return (0); 3254 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3255 if (v3) { 3256 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3257 txdr_hyper(tval, &sfp->sf_tbytes); 3258 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3259 txdr_hyper(tval, &sfp->sf_fbytes); 3260 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3261 txdr_hyper(tval, &sfp->sf_abytes); 3262 tval = (u_quad_t)sf->f_files; 3263 txdr_hyper(tval, &sfp->sf_tfiles); 3264 tval = (u_quad_t)sf->f_ffree; 3265 txdr_hyper(tval, &sfp->sf_ffiles); 3266 txdr_hyper(tval, &sfp->sf_afiles); 3267 sfp->sf_invarsec = 0; 3268 } else { 3269 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3270 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3271 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3272 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3273 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3274 } 3275 nfsm_srvdone; 3276 } 3277 3278 /* 3279 * nfs fsinfo service 3280 */ 3281 int 3282 nfsrv_fsinfo(nfsd, slp, lwp, mrq) 3283 struct nfsrv_descript *nfsd; 3284 struct nfssvc_sock *slp; 3285 struct lwp *lwp; 3286 struct mbuf **mrq; 3287 { 3288 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3289 struct mbuf *nam = nfsd->nd_nam; 3290 caddr_t dpos = nfsd->nd_dpos; 3291 struct ucred *cred = &nfsd->nd_cr; 3292 u_int32_t *tl; 3293 struct nfsv3_fsinfo *sip; 3294 int32_t t1; 3295 caddr_t bpos; 3296 int error = 0, rdonly, cache = 0, getret = 1; 3297 uint32_t maxdata; 3298 char *cp2; 3299 struct mbuf *mb, *mreq; 3300 struct vnode *vp; 3301 struct vattr at; 3302 nfsfh_t nfh; 3303 fhandle_t *fhp; 3304 u_quad_t frev, maxfsize; 3305 struct statvfs sb; 3306 3307 fhp = &nfh.fh_generic; 3308 nfsm_srvmtofh(fhp); 3309 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3310 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3311 if (error) { 3312 nfsm_reply(NFSX_UNSIGNED); 3313 nfsm_srvpostop_attr(getret, &at); 3314 return (0); 3315 } 3316 3317 /* XXX Try to make a guess on the max file size. */ 3318 VFS_STATVFS(vp->v_mount, &sb, (struct lwp *)0); 3319 maxfsize = (u_quad_t)0x80000000 * sb.f_frsize - 1; 3320 3321 getret = VOP_GETATTR(vp, &at, cred, lwp); 3322 vput(vp); 3323 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3324 nfsm_srvpostop_attr(getret, &at); 3325 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3326 3327 /* 3328 * XXX 3329 * There should be file system VFS OP(s) to get this information. 3330 * For now, assume ufs. 3331 */ 3332 if (slp->ns_so->so_type == SOCK_DGRAM) 3333 maxdata = NFS_MAXDGRAMDATA; 3334 else 3335 maxdata = NFS_MAXDATA; 3336 sip->fs_rtmax = txdr_unsigned(maxdata); 3337 sip->fs_rtpref = txdr_unsigned(maxdata); 3338 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3339 sip->fs_wtmax = txdr_unsigned(maxdata); 3340 sip->fs_wtpref = txdr_unsigned(maxdata); 3341 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3342 sip->fs_dtpref = txdr_unsigned(maxdata); 3343 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3344 sip->fs_timedelta.nfsv3_sec = 0; 3345 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3346 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3347 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3348 NFSV3FSINFO_CANSETTIME); 3349 nfsm_srvdone; 3350 } 3351 3352 /* 3353 * nfs pathconf service 3354 */ 3355 int 3356 nfsrv_pathconf(nfsd, slp, lwp, mrq) 3357 struct nfsrv_descript *nfsd; 3358 struct nfssvc_sock *slp; 3359 struct lwp *lwp; 3360 struct mbuf **mrq; 3361 { 3362 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3363 struct mbuf *nam = nfsd->nd_nam; 3364 caddr_t dpos = nfsd->nd_dpos; 3365 struct ucred *cred = &nfsd->nd_cr; 3366 u_int32_t *tl; 3367 struct nfsv3_pathconf *pc; 3368 int32_t t1; 3369 caddr_t bpos; 3370 int error = 0, rdonly, cache = 0, getret = 1; 3371 register_t linkmax, namemax, chownres, notrunc; 3372 char *cp2; 3373 struct mbuf *mb, *mreq; 3374 struct vnode *vp; 3375 struct vattr at; 3376 nfsfh_t nfh; 3377 fhandle_t *fhp; 3378 u_quad_t frev; 3379 3380 fhp = &nfh.fh_generic; 3381 nfsm_srvmtofh(fhp); 3382 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3383 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3384 if (error) { 3385 nfsm_reply(NFSX_UNSIGNED); 3386 nfsm_srvpostop_attr(getret, &at); 3387 return (0); 3388 } 3389 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3390 if (!error) 3391 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3392 if (!error) 3393 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3394 if (!error) 3395 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3396 getret = VOP_GETATTR(vp, &at, cred, lwp); 3397 vput(vp); 3398 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3399 nfsm_srvpostop_attr(getret, &at); 3400 if (error) 3401 return (0); 3402 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3403 3404 pc->pc_linkmax = txdr_unsigned(linkmax); 3405 pc->pc_namemax = txdr_unsigned(namemax); 3406 pc->pc_notrunc = txdr_unsigned(notrunc); 3407 pc->pc_chownrestricted = txdr_unsigned(chownres); 3408 3409 /* 3410 * These should probably be supported by VOP_PATHCONF(), but 3411 * until msdosfs is exportable (why would you want to?), the 3412 * Unix defaults should be ok. 3413 */ 3414 pc->pc_caseinsensitive = nfs_false; 3415 pc->pc_casepreserving = nfs_true; 3416 nfsm_srvdone; 3417 } 3418 3419 /* 3420 * Null operation, used by clients to ping server 3421 */ 3422 /* ARGSUSED */ 3423 int 3424 nfsrv_null(nfsd, slp, lwp, mrq) 3425 struct nfsrv_descript *nfsd; 3426 struct nfssvc_sock *slp; 3427 struct lwp *lwp; 3428 struct mbuf **mrq; 3429 { 3430 struct mbuf *mrep = nfsd->nd_mrep; 3431 caddr_t bpos; 3432 int error = NFSERR_RETVOID, cache = 0; 3433 struct mbuf *mb, *mreq; 3434 u_quad_t frev; 3435 3436 nfsm_reply(0); 3437 return (0); 3438 } 3439 3440 /* 3441 * No operation, used for obsolete procedures 3442 */ 3443 /* ARGSUSED */ 3444 int 3445 nfsrv_noop(nfsd, slp, lwp, mrq) 3446 struct nfsrv_descript *nfsd; 3447 struct nfssvc_sock *slp; 3448 struct lwp *lwp; 3449 struct mbuf **mrq; 3450 { 3451 struct mbuf *mrep = nfsd->nd_mrep; 3452 caddr_t bpos; 3453 int error, cache = 0; 3454 struct mbuf *mb, *mreq; 3455 u_quad_t frev; 3456 3457 if (nfsd->nd_repstat) 3458 error = nfsd->nd_repstat; 3459 else 3460 error = EPROCUNAVAIL; 3461 nfsm_reply(0); 3462 return (0); 3463 } 3464 3465 /* 3466 * Perform access checking for vnodes obtained from file handles that would 3467 * refer to files already opened by a Unix client. You cannot just use 3468 * vn_writechk() and VOP_ACCESS() for two reasons. 3469 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3470 * 2 - The owner is to be given access irrespective of mode bits for some 3471 * operations, so that processes that chmod after opening a file don't 3472 * break. I don't like this because it opens a security hole, but since 3473 * the nfs server opens a security hole the size of a barn door anyhow, 3474 * what the heck. 3475 * 3476 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3477 * will return EPERM instead of EACCESS. EPERM is always an error. 3478 */ 3479 int 3480 nfsrv_access(vp, flags, cred, rdonly, lwp, override) 3481 struct vnode *vp; 3482 int flags; 3483 struct ucred *cred; 3484 int rdonly; 3485 struct lwp *lwp; 3486 { 3487 struct vattr vattr; 3488 int error; 3489 if (flags & VWRITE) { 3490 /* Just vn_writechk() changed to check rdonly */ 3491 /* 3492 * Disallow write attempts on read-only file systems; 3493 * unless the file is a socket or a block or character 3494 * device resident on the file system. 3495 */ 3496 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3497 switch (vp->v_type) { 3498 case VREG: 3499 case VDIR: 3500 case VLNK: 3501 return (EROFS); 3502 default: 3503 break; 3504 } 3505 } 3506 3507 /* 3508 * If the vnode is in use as a process's text, 3509 * we can't allow writing. 3510 */ 3511 if (vp->v_flag & VTEXT) 3512 return (ETXTBSY); 3513 } 3514 error = VOP_GETATTR(vp, &vattr, cred, lwp); 3515 if (error) 3516 return (error); 3517 error = VOP_ACCESS(vp, flags, cred, lwp); 3518 /* 3519 * Allow certain operations for the owner (reads and writes 3520 * on files that are already open). 3521 */ 3522 if (override && error == EACCES && cred->cr_uid == vattr.va_uid) 3523 error = 0; 3524 return error; 3525 } 3526
Indexes created Tue Sep 23 02:09:52 GMT 2025