nfs_serv.c revision 1.100.4.2
1 /* $NetBSD: nfs_serv.c,v 1.100.4.2 2006/04/22 11:40:15 simonb 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 1.100.4.2 2006/04/22 11:40:15 simonb 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 timeval now; 1004 struct iovec *ivp; 1005 struct mbuf *mp; 1006 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1007 struct nfs_fattr *fp; 1008 int i = 0; 1009 struct iovec *iov; 1010 struct nfsrvw_delayhash *wpp; 1011 struct ucred *cred; 1012 struct vattr va, forat; 1013 u_int32_t *tl; 1014 int32_t t1; 1015 caddr_t bpos, dpos; 1016 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1017 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing; 1018 char *cp2; 1019 struct mbuf *mb, *mreq, *mrep, *md; 1020 struct vnode *vp; 1021 struct uio io, *uiop = &io; 1022 u_quad_t frev, cur_usec; 1023 struct mount *mntp = NULL; 1024 1025 *mrq = NULL; 1026 if (*ndp) { 1027 nfsd = *ndp; 1028 *ndp = NULL; 1029 mrep = nfsd->nd_mrep; 1030 md = nfsd->nd_md; 1031 dpos = nfsd->nd_dpos; 1032 cred = &nfsd->nd_cr; 1033 v3 = (nfsd->nd_flag & ND_NFSV3); 1034 LIST_INIT(&nfsd->nd_coalesce); 1035 nfsd->nd_mreq = NULL; 1036 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1037 getmicrotime(&now); 1038 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1039 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1040 1041 /* 1042 * Now, get the write header.. 1043 */ 1044 nfsm_srvmtofh(&nfsd->nd_fh); 1045 if (v3) { 1046 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1047 nfsd->nd_off = fxdr_hyper(tl); 1048 tl += 3; 1049 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1050 } else { 1051 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1052 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1053 tl += 2; 1054 } 1055 len = fxdr_unsigned(int32_t, *tl); 1056 nfsd->nd_len = len; 1057 nfsd->nd_eoff = nfsd->nd_off + len; 1058 1059 /* 1060 * Trim the header out of the mbuf list and trim off any trailing 1061 * junk so that the mbuf list has only the write data. 1062 */ 1063 zeroing = 1; 1064 i = 0; 1065 mp = mrep; 1066 while (mp) { 1067 if (mp == md) { 1068 zeroing = 0; 1069 adjust = dpos - mtod(mp, caddr_t); 1070 mp->m_len -= adjust; 1071 if (mp->m_len > 0 && adjust > 0) 1072 NFSMADV(mp, adjust); 1073 } 1074 if (zeroing) 1075 mp->m_len = 0; 1076 else { 1077 i += mp->m_len; 1078 if (i > len) { 1079 mp->m_len -= (i - len); 1080 zeroing = 1; 1081 } 1082 } 1083 mp = mp->m_next; 1084 } 1085 if (len > NFS_MAXDATA || len < 0 || i < len) { 1086 nfsmout: 1087 m_freem(mrep); 1088 error = EIO; 1089 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1090 if (v3) 1091 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1092 nfsd->nd_mreq = mreq; 1093 nfsd->nd_mrep = NULL; 1094 nfsd->nd_time = 0; 1095 } 1096 1097 /* 1098 * Add this entry to the hash and time queues. 1099 */ 1100 s = splsoftclock(); 1101 owp = NULL; 1102 wp = LIST_FIRST(&slp->ns_tq); 1103 while (wp && wp->nd_time < nfsd->nd_time) { 1104 owp = wp; 1105 wp = LIST_NEXT(wp, nd_tq); 1106 } 1107 if (owp) { 1108 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1109 } else { 1110 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1111 } 1112 if (nfsd->nd_mrep) { 1113 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data); 1114 owp = NULL; 1115 wp = LIST_FIRST(wpp); 1116 while (wp && 1117 memcmp(&nfsd->nd_fh, &wp->nd_fh, NFSX_V3FH)) { 1118 owp = wp; 1119 wp = LIST_NEXT(wp, nd_hash); 1120 } 1121 while (wp && wp->nd_off < nfsd->nd_off && 1122 !memcmp(&nfsd->nd_fh, &wp->nd_fh, NFSX_V3FH)) { 1123 owp = wp; 1124 wp = LIST_NEXT(wp, nd_hash); 1125 } 1126 if (owp) { 1127 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1128 1129 /* 1130 * Search the hash list for overlapping entries and 1131 * coalesce. 1132 */ 1133 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1134 wp = LIST_NEXT(nfsd, nd_hash); 1135 if (NFSW_SAMECRED(owp, nfsd)) 1136 nfsrvw_coalesce(owp, nfsd); 1137 } 1138 } else { 1139 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1140 } 1141 } 1142 splx(s); 1143 } 1144 1145 /* 1146 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1147 * and generate the associated reply mbuf list(s). 1148 */ 1149 loop1: 1150 getmicrotime(&now); 1151 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1152 s = splsoftclock(); 1153 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1154 owp = LIST_NEXT(nfsd, nd_tq); 1155 if (nfsd->nd_time > cur_usec) 1156 break; 1157 if (nfsd->nd_mreq) 1158 continue; 1159 LIST_REMOVE(nfsd, nd_tq); 1160 LIST_REMOVE(nfsd, nd_hash); 1161 splx(s); 1162 mrep = nfsd->nd_mrep; 1163 nfsd->nd_mrep = NULL; 1164 cred = &nfsd->nd_cr; 1165 v3 = (nfsd->nd_flag & ND_NFSV3); 1166 forat_ret = aftat_ret = 1; 1167 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1168 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1169 FALSE); 1170 if (!error) { 1171 if (v3) 1172 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 1173 if (vp->v_type != VREG) { 1174 if (v3) 1175 error = EINVAL; 1176 else 1177 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1178 } 1179 } else 1180 vp = NULL; 1181 if (!error) { 1182 nqsrv_getl(vp, ND_WRITE); 1183 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1184 } 1185 1186 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1187 ioflags = IO_NODELOCKED; 1188 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1189 ioflags = (IO_SYNC | IO_NODELOCKED); 1190 else 1191 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1192 uiop->uio_rw = UIO_WRITE; 1193 uiop->uio_offset = nfsd->nd_off; 1194 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1195 UIO_SETUP_SYSSPACE(uiop); 1196 if (uiop->uio_resid > 0) { 1197 mp = mrep; 1198 i = 0; 1199 while (mp) { 1200 if (mp->m_len > 0) 1201 i++; 1202 mp = mp->m_next; 1203 } 1204 uiop->uio_iovcnt = i; 1205 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1206 uiop->uio_iov = ivp = iov; 1207 mp = mrep; 1208 while (mp) { 1209 if (mp->m_len > 0) { 1210 ivp->iov_base = mtod(mp, caddr_t); 1211 ivp->iov_len = mp->m_len; 1212 ivp++; 1213 } 1214 mp = mp->m_next; 1215 } 1216 if (!error) { 1217 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) { 1218 VOP_UNLOCK(vp, 0); 1219 vn_start_write(NULL, &mntp, V_WAIT); 1220 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1221 } 1222 if (!error) { 1223 error = VOP_WRITE(vp, uiop, ioflags, cred); 1224 nfsstats.srvvop_writes++; 1225 vn_finished_write(mntp, 0); 1226 } 1227 } 1228 free((caddr_t)iov, M_TEMP); 1229 } 1230 m_freem(mrep); 1231 if (vp) { 1232 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 1233 vput(vp); 1234 } 1235 1236 /* 1237 * Loop around generating replies for all write rpcs that have 1238 * now been completed. 1239 */ 1240 swp = nfsd; 1241 do { 1242 if (error) { 1243 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1244 if (v3) { 1245 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1246 } 1247 } else { 1248 nfsm_writereply(NFSX_PREOPATTR(v3) + 1249 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1250 NFSX_WRITEVERF(v3), v3); 1251 if (v3) { 1252 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1253 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1254 *tl++ = txdr_unsigned(nfsd->nd_len); 1255 *tl++ = txdr_unsigned(swp->nd_stable); 1256 /* 1257 * Actually, there is no need to txdr these fields, 1258 * but it may make the values more human readable, 1259 * for debugging purposes. 1260 */ 1261 *tl++ = txdr_unsigned(boottime.tv_sec); 1262 *tl = txdr_unsigned(boottime.tv_usec); 1263 } else { 1264 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1265 nfsm_srvfillattr(&va, fp); 1266 } 1267 } 1268 nfsd->nd_mreq = mreq; 1269 if (nfsd->nd_mrep) 1270 panic("nfsrv_write: nd_mrep not free"); 1271 1272 /* 1273 * Done. Put it at the head of the timer queue so that 1274 * the final phase can return the reply. 1275 */ 1276 s = splsoftclock(); 1277 if (nfsd != swp) { 1278 nfsd->nd_time = 0; 1279 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1280 } 1281 nfsd = LIST_FIRST(&swp->nd_coalesce); 1282 if (nfsd) { 1283 LIST_REMOVE(nfsd, nd_tq); 1284 } 1285 splx(s); 1286 } while (nfsd); 1287 s = splsoftclock(); 1288 swp->nd_time = 0; 1289 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1290 splx(s); 1291 goto loop1; 1292 } 1293 splx(s); 1294 1295 /* 1296 * Search for a reply to return. 1297 */ 1298 s = splsoftclock(); 1299 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1300 if (nfsd->nd_mreq) { 1301 LIST_REMOVE(nfsd, nd_tq); 1302 *mrq = nfsd->nd_mreq; 1303 *ndp = nfsd; 1304 break; 1305 } 1306 } 1307 splx(s); 1308 return (0); 1309 } 1310 1311 /* 1312 * Coalesce the write request nfsd into owp. To do this we must: 1313 * - remove nfsd from the queues 1314 * - merge nfsd->nd_mrep into owp->nd_mrep 1315 * - update the nd_eoff and nd_stable for owp 1316 * - put nfsd on owp's nd_coalesce list 1317 * NB: Must be called at splsoftclock(). 1318 */ 1319 void 1320 nfsrvw_coalesce(owp, nfsd) 1321 struct nfsrv_descript *owp; 1322 struct nfsrv_descript *nfsd; 1323 { 1324 int overlap; 1325 struct mbuf *mp; 1326 struct nfsrv_descript *m; 1327 1328 LIST_REMOVE(nfsd, nd_hash); 1329 LIST_REMOVE(nfsd, nd_tq); 1330 if (owp->nd_eoff < nfsd->nd_eoff) { 1331 overlap = owp->nd_eoff - nfsd->nd_off; 1332 if (overlap < 0) 1333 panic("nfsrv_coalesce: bad off"); 1334 if (overlap > 0) 1335 m_adj(nfsd->nd_mrep, overlap); 1336 mp = owp->nd_mrep; 1337 while (mp->m_next) 1338 mp = mp->m_next; 1339 mp->m_next = nfsd->nd_mrep; 1340 owp->nd_eoff = nfsd->nd_eoff; 1341 } else 1342 m_freem(nfsd->nd_mrep); 1343 nfsd->nd_mrep = NULL; 1344 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1345 owp->nd_stable = NFSV3WRITE_FILESYNC; 1346 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1347 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1348 owp->nd_stable = NFSV3WRITE_DATASYNC; 1349 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1350 /* 1351 * nfsd might hold coalesce elements! Move them to owp. 1352 * Otherwise, requests may be lost and clients will be stuck. 1353 */ 1354 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1355 LIST_REMOVE(m, nd_tq); 1356 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1357 } 1358 } 1359 1360 /* 1361 * nfs create service 1362 * now does a truncate to 0 length via. setattr if it already exists 1363 */ 1364 int 1365 nfsrv_create(nfsd, slp, lwp, mrq) 1366 struct nfsrv_descript *nfsd; 1367 struct nfssvc_sock *slp; 1368 struct lwp *lwp; 1369 struct mbuf **mrq; 1370 { 1371 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1372 struct mbuf *nam = nfsd->nd_nam; 1373 caddr_t dpos = nfsd->nd_dpos; 1374 struct ucred *cred = &nfsd->nd_cr; 1375 struct nfs_fattr *fp; 1376 struct vattr va, dirfor, diraft; 1377 struct nfsv2_sattr *sp; 1378 u_int32_t *tl; 1379 struct nameidata nd; 1380 caddr_t cp; 1381 int32_t t1; 1382 caddr_t bpos; 1383 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1384 int rdev = 0; 1385 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1386 char *cp2; 1387 struct mbuf *mb, *mreq; 1388 struct vnode *vp = NULL, *dirp = NULL; 1389 nfsfh_t nfh; 1390 fhandle_t *fhp; 1391 u_quad_t frev, tempsize; 1392 u_char cverf[NFSX_V3CREATEVERF]; 1393 struct mount *mp = NULL; 1394 1395 nd.ni_cnd.cn_nameiop = 0; 1396 fhp = &nfh.fh_generic; 1397 nfsm_srvmtofh(fhp); 1398 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1399 return (ESTALE); 1400 vn_start_write(NULL, &mp, V_WAIT); 1401 nfsm_srvnamesiz(len); 1402 nd.ni_cnd.cn_cred = cred; 1403 nd.ni_cnd.cn_nameiop = CREATE; 1404 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1405 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1406 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1407 if (dirp) { 1408 if (v3) 1409 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1410 else { 1411 vrele(dirp); 1412 dirp = (struct vnode *)0; 1413 } 1414 } 1415 if (error) { 1416 nfsm_reply(NFSX_WCCDATA(v3)); 1417 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1418 if (dirp) 1419 vrele(dirp); 1420 vn_finished_write(mp, 0); 1421 return (0); 1422 } 1423 VATTR_NULL(&va); 1424 if (v3) { 1425 va.va_mode = 0; 1426 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1427 how = fxdr_unsigned(int, *tl); 1428 switch (how) { 1429 case NFSV3CREATE_GUARDED: 1430 if (nd.ni_vp) { 1431 error = EEXIST; 1432 break; 1433 } 1434 case NFSV3CREATE_UNCHECKED: 1435 nfsm_srvsattr(&va); 1436 break; 1437 case NFSV3CREATE_EXCLUSIVE: 1438 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF); 1439 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1440 exclusive_flag = 1; 1441 break; 1442 }; 1443 va.va_type = VREG; 1444 } else { 1445 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1446 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1447 if (va.va_type == VNON) 1448 va.va_type = VREG; 1449 va.va_mode = nfstov_mode(sp->sa_mode); 1450 switch (va.va_type) { 1451 case VREG: 1452 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1453 if (tsize != -1) 1454 va.va_size = (u_quad_t)tsize; 1455 break; 1456 case VCHR: 1457 case VBLK: 1458 case VFIFO: 1459 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1460 break; 1461 default: 1462 break; 1463 }; 1464 } 1465 1466 /* 1467 * Iff doesn't exist, create it 1468 * otherwise just truncate to 0 length 1469 * should I set the mode too ?? 1470 */ 1471 if (nd.ni_vp == NULL) { 1472 if (va.va_type == VREG || va.va_type == VSOCK) { 1473 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1474 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1475 if (!error) { 1476 if (exclusive_flag) { 1477 exclusive_flag = 0; 1478 VATTR_NULL(&va); 1479 /* 1480 * XXX 1481 * assuming NFSX_V3CREATEVERF 1482 * == sizeof(nfstime3) 1483 */ 1484 fxdr_nfsv3time(cverf, &va.va_atime); 1485 error = VOP_SETATTR(nd.ni_vp, &va, cred, 1486 lwp); 1487 } 1488 } 1489 } else if (va.va_type == VCHR || va.va_type == VBLK || 1490 va.va_type == VFIFO) { 1491 if (va.va_type == VCHR && rdev == 0xffffffff) 1492 va.va_type = VFIFO; 1493 if (va.va_type != VFIFO && 1494 (error = suser(cred, (u_short *)0))) { 1495 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1496 vput(nd.ni_dvp); 1497 nfsm_reply(0); 1498 vn_finished_write(mp, 0); 1499 return (error); 1500 } else 1501 va.va_rdev = (dev_t)rdev; 1502 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1503 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1504 &va); 1505 if (error) { 1506 nfsm_reply(0); 1507 } 1508 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1509 vrele(nd.ni_dvp); 1510 vput(nd.ni_vp); 1511 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1512 error = EINVAL; 1513 nfsm_reply(0); 1514 } 1515 } else { 1516 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1517 vput(nd.ni_dvp); 1518 error = ENXIO; 1519 } 1520 vp = nd.ni_vp; 1521 } else { 1522 vp = nd.ni_vp; 1523 if (nd.ni_dvp == vp) 1524 vrele(nd.ni_dvp); 1525 else 1526 vput(nd.ni_dvp); 1527 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1528 if (!error && va.va_size != -1) { 1529 error = nfsrv_access(vp, VWRITE, cred, 1530 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1531 if (!error) { 1532 nqsrv_getl(vp, ND_WRITE); 1533 tempsize = va.va_size; 1534 VATTR_NULL(&va); 1535 va.va_size = tempsize; 1536 error = VOP_SETATTR(vp, &va, cred, lwp); 1537 } 1538 } 1539 if (error) 1540 vput(vp); 1541 } 1542 if (!error) { 1543 memset((caddr_t)fhp, 0, sizeof(nfh)); 1544 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1545 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1546 if (!error) 1547 error = VOP_GETATTR(vp, &va, cred, lwp); 1548 vput(vp); 1549 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1550 } 1551 if (v3) { 1552 if (exclusive_flag && !error) { 1553 /* 1554 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1555 */ 1556 char oldverf[NFSX_V3CREATEVERF]; 1557 1558 txdr_nfsv3time(&va.va_atime, oldverf); 1559 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1560 error = EEXIST; 1561 } 1562 if (dirp) { 1563 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1564 vrele(dirp); 1565 } 1566 } 1567 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1568 if (v3) { 1569 if (!error) { 1570 nfsm_srvpostop_fh(fhp); 1571 nfsm_srvpostop_attr(0, &va); 1572 } 1573 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1574 } else { 1575 nfsm_srvfhtom(fhp, v3); 1576 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1577 nfsm_srvfillattr(&va, fp); 1578 } 1579 vn_finished_write(mp, 0); 1580 return (0); 1581 nfsmout: 1582 if (dirp) 1583 vrele(dirp); 1584 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1585 if (nd.ni_dvp == nd.ni_vp) 1586 vrele(nd.ni_dvp); 1587 else 1588 vput(nd.ni_dvp); 1589 if (nd.ni_vp) 1590 vput(nd.ni_vp); 1591 vn_finished_write(mp, 0); 1592 return (error); 1593 } 1594 1595 /* 1596 * nfs v3 mknod service 1597 */ 1598 int 1599 nfsrv_mknod(nfsd, slp, lwp, mrq) 1600 struct nfsrv_descript *nfsd; 1601 struct nfssvc_sock *slp; 1602 struct lwp *lwp; 1603 struct mbuf **mrq; 1604 { 1605 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1606 struct mbuf *nam = nfsd->nd_nam; 1607 caddr_t dpos = nfsd->nd_dpos; 1608 struct ucred *cred = &nfsd->nd_cr; 1609 struct vattr va, dirfor, diraft; 1610 u_int32_t *tl; 1611 struct nameidata nd; 1612 int32_t t1; 1613 caddr_t bpos; 1614 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1615 u_int32_t major, minor; 1616 enum vtype vtyp; 1617 char *cp2; 1618 struct mbuf *mb, *mreq; 1619 struct vnode *vp, *dirp = (struct vnode *)0; 1620 nfsfh_t nfh; 1621 fhandle_t *fhp; 1622 u_quad_t frev; 1623 struct mount *mp = NULL; 1624 1625 nd.ni_cnd.cn_nameiop = 0; 1626 fhp = &nfh.fh_generic; 1627 nfsm_srvmtofh(fhp); 1628 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1629 return (ESTALE); 1630 vn_start_write(NULL, &mp, V_WAIT); 1631 nfsm_srvnamesiz(len); 1632 nd.ni_cnd.cn_cred = cred; 1633 nd.ni_cnd.cn_nameiop = CREATE; 1634 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1635 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1636 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1637 if (dirp) 1638 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1639 if (error) { 1640 nfsm_reply(NFSX_WCCDATA(1)); 1641 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1642 if (dirp) 1643 vrele(dirp); 1644 vn_finished_write(mp, 0); 1645 return (0); 1646 } 1647 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1648 vtyp = nfsv3tov_type(*tl); 1649 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1650 error = NFSERR_BADTYPE; 1651 goto abort; 1652 } 1653 VATTR_NULL(&va); 1654 va.va_mode = 0; 1655 nfsm_srvsattr(&va); 1656 if (vtyp == VCHR || vtyp == VBLK) { 1657 dev_t rdev; 1658 1659 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1660 major = fxdr_unsigned(u_int32_t, *tl++); 1661 minor = fxdr_unsigned(u_int32_t, *tl); 1662 rdev = makedev(major, minor); 1663 if (major(rdev) != major || minor(rdev) != minor) { 1664 error = EINVAL; 1665 goto abort; 1666 } 1667 va.va_rdev = rdev; 1668 } 1669 1670 /* 1671 * Iff doesn't exist, create it. 1672 */ 1673 if (nd.ni_vp) { 1674 error = EEXIST; 1675 abort: 1676 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1677 if (nd.ni_dvp == nd.ni_vp) 1678 vrele(nd.ni_dvp); 1679 else 1680 vput(nd.ni_dvp); 1681 if (nd.ni_vp) 1682 vput(nd.ni_vp); 1683 goto out; 1684 } 1685 va.va_type = vtyp; 1686 if (vtyp == VSOCK) { 1687 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1688 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1689 } else { 1690 if (va.va_type != VFIFO && 1691 (error = suser(cred, (u_short *)0))) { 1692 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1693 vput(nd.ni_dvp); 1694 goto out; 1695 } 1696 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1697 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1698 if (error) 1699 goto out; 1700 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1701 vput(nd.ni_vp); 1702 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1703 error = EINVAL; 1704 } 1705 } 1706 out: 1707 vp = nd.ni_vp; 1708 if (!error) { 1709 memset((caddr_t)fhp, 0, sizeof(nfh)); 1710 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 1711 error = VFS_VPTOFH(vp, &fhp->fh_fid); 1712 if (!error) 1713 error = VOP_GETATTR(vp, &va, cred, lwp); 1714 vput(vp); 1715 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 1716 } 1717 if (dirp) { 1718 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1719 vrele(dirp); 1720 } 1721 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1)); 1722 if (!error) { 1723 nfsm_srvpostop_fh(fhp); 1724 nfsm_srvpostop_attr(0, &va); 1725 } 1726 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1727 vn_finished_write(mp, 0); 1728 return (0); 1729 nfsmout: 1730 if (dirp) 1731 vrele(dirp); 1732 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1733 if (nd.ni_dvp == nd.ni_vp) 1734 vrele(nd.ni_dvp); 1735 else 1736 vput(nd.ni_dvp); 1737 if (nd.ni_vp) 1738 vput(nd.ni_vp); 1739 vn_finished_write(mp, 0); 1740 return (error); 1741 } 1742 1743 /* 1744 * nfs remove service 1745 */ 1746 int 1747 nfsrv_remove(nfsd, slp, lwp, mrq) 1748 struct nfsrv_descript *nfsd; 1749 struct nfssvc_sock *slp; 1750 struct lwp *lwp; 1751 struct mbuf **mrq; 1752 { 1753 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1754 struct mbuf *nam = nfsd->nd_nam; 1755 caddr_t dpos = nfsd->nd_dpos; 1756 struct ucred *cred = &nfsd->nd_cr; 1757 struct nameidata nd; 1758 u_int32_t *tl; 1759 int32_t t1; 1760 caddr_t bpos; 1761 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1762 int v3 = (nfsd->nd_flag & ND_NFSV3); 1763 char *cp2; 1764 struct mbuf *mb, *mreq; 1765 struct vnode *vp, *dirp; 1766 struct vattr dirfor, diraft; 1767 nfsfh_t nfh; 1768 fhandle_t *fhp; 1769 u_quad_t frev; 1770 struct mount *mp = NULL; 1771 1772 #ifndef nolint 1773 vp = (struct vnode *)0; 1774 #endif 1775 fhp = &nfh.fh_generic; 1776 nfsm_srvmtofh(fhp); 1777 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 1778 return (ESTALE); 1779 vn_start_write(NULL, &mp, V_WAIT); 1780 nfsm_srvnamesiz(len); 1781 nd.ni_cnd.cn_cred = cred; 1782 nd.ni_cnd.cn_nameiop = DELETE; 1783 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1784 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 1785 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1786 if (dirp) { 1787 if (v3) 1788 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1789 else 1790 vrele(dirp); 1791 } 1792 if (!error) { 1793 vp = nd.ni_vp; 1794 if (vp->v_type == VDIR && 1795 (error = suser(cred, (u_short *)0)) != 0) 1796 goto out; 1797 /* 1798 * The root of a mounted filesystem cannot be deleted. 1799 */ 1800 if (vp->v_flag & VROOT) { 1801 error = EBUSY; 1802 goto out; 1803 } 1804 out: 1805 if (!error) { 1806 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1807 nqsrv_getl(vp, ND_WRITE); 1808 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1809 } else { 1810 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1811 if (nd.ni_dvp == vp) 1812 vrele(nd.ni_dvp); 1813 else 1814 vput(nd.ni_dvp); 1815 vput(vp); 1816 } 1817 } 1818 if (dirp && v3) { 1819 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1820 vrele(dirp); 1821 } 1822 nfsm_reply(NFSX_WCCDATA(v3)); 1823 if (v3) { 1824 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1825 vn_finished_write(mp, 0); 1826 return (0); 1827 } 1828 vn_finished_write(mp, 0); 1829 nfsm_srvdone; 1830 } 1831 1832 /* 1833 * nfs rename service 1834 */ 1835 int 1836 nfsrv_rename(nfsd, slp, lwp, mrq) 1837 struct nfsrv_descript *nfsd; 1838 struct nfssvc_sock *slp; 1839 struct lwp *lwp; 1840 struct mbuf **mrq; 1841 { 1842 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1843 struct mbuf *nam = nfsd->nd_nam; 1844 caddr_t dpos = nfsd->nd_dpos; 1845 struct ucred *cred = &nfsd->nd_cr; 1846 u_int32_t *tl; 1847 int32_t t1; 1848 caddr_t bpos; 1849 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1850 uint32_t len, len2; 1851 int tdirfor_ret = 1, tdiraft_ret = 1; 1852 int v3 = (nfsd->nd_flag & ND_NFSV3); 1853 char *cp2; 1854 struct mbuf *mb, *mreq; 1855 struct nameidata fromnd, tond; 1856 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0; 1857 struct vnode *tdirp = (struct vnode *)0; 1858 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1859 nfsfh_t fnfh, tnfh; 1860 fhandle_t *ffhp, *tfhp; 1861 u_quad_t frev; 1862 uid_t saved_uid; 1863 struct mount *mp = NULL; 1864 1865 #ifndef nolint 1866 fvp = (struct vnode *)0; 1867 #endif 1868 ffhp = &fnfh.fh_generic; 1869 tfhp = &tnfh.fh_generic; 1870 fromnd.ni_cnd.cn_nameiop = 0; 1871 tond.ni_cnd.cn_nameiop = 0; 1872 nfsm_srvmtofh(ffhp); 1873 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) 1874 return (ESTALE); 1875 vn_start_write(NULL, &mp, V_WAIT); 1876 nfsm_srvnamesiz(len); 1877 /* 1878 * Remember our original uid so that we can reset cr_uid before 1879 * the second nfs_namei() call, in case it is remapped. 1880 */ 1881 saved_uid = cred->cr_uid; 1882 fromnd.ni_cnd.cn_cred = cred; 1883 fromnd.ni_cnd.cn_nameiop = DELETE; 1884 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART; 1885 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md, 1886 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1887 if (fdirp) { 1888 if (v3) 1889 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp); 1890 else { 1891 vrele(fdirp); 1892 fdirp = (struct vnode *)0; 1893 } 1894 } 1895 if (error) { 1896 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1897 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1898 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1899 if (fdirp) 1900 vrele(fdirp); 1901 vn_finished_write(mp, 0); 1902 return (0); 1903 } 1904 fvp = fromnd.ni_vp; 1905 nfsm_srvmtofh(tfhp); 1906 if (v3) { 1907 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1908 len2 = fxdr_unsigned(uint32_t, *tl); 1909 /* len2 will be checked by nfs_namei */ 1910 } 1911 else { 1912 /* NFSv2 */ 1913 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1914 } 1915 cred->cr_uid = saved_uid; 1916 tond.ni_cnd.cn_cred = cred; 1917 tond.ni_cnd.cn_nameiop = RENAME; 1918 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1919 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md, 1920 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 1921 if (tdirp) { 1922 if (v3) 1923 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp); 1924 else { 1925 vrele(tdirp); 1926 tdirp = (struct vnode *)0; 1927 } 1928 } 1929 if (error) { 1930 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1931 vrele(fromnd.ni_dvp); 1932 vrele(fvp); 1933 goto out1; 1934 } 1935 tdvp = tond.ni_dvp; 1936 tvp = tond.ni_vp; 1937 if (tvp != NULL) { 1938 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1939 if (v3) 1940 error = EEXIST; 1941 else 1942 error = EISDIR; 1943 goto out; 1944 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1945 if (v3) 1946 error = EEXIST; 1947 else 1948 error = ENOTDIR; 1949 goto out; 1950 } 1951 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1952 if (v3) 1953 error = EXDEV; 1954 else 1955 error = ENOTEMPTY; 1956 goto out; 1957 } 1958 } 1959 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1960 if (v3) 1961 error = EXDEV; 1962 else 1963 error = ENOTEMPTY; 1964 goto out; 1965 } 1966 if (fvp->v_mount != tdvp->v_mount) { 1967 if (v3) 1968 error = EXDEV; 1969 else 1970 error = ENOTEMPTY; 1971 goto out; 1972 } 1973 if (fvp == tdvp) { 1974 if (v3) 1975 error = EINVAL; 1976 else 1977 error = ENOTEMPTY; 1978 } 1979 /* 1980 * If source is the same as the destination (that is the 1981 * same vnode with the same name in the same directory), 1982 * then there is nothing to do. 1983 */ 1984 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1985 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1986 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1987 fromnd.ni_cnd.cn_namelen)) 1988 error = -1; 1989 out: 1990 if (!error) { 1991 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1992 nqsrv_getl(tdvp, ND_WRITE); 1993 if (tvp) { 1994 nqsrv_getl(tvp, ND_WRITE); 1995 } 1996 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1997 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1998 } else { 1999 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 2000 if (tdvp == tvp) 2001 vrele(tdvp); 2002 else 2003 vput(tdvp); 2004 if (tvp) 2005 vput(tvp); 2006 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2007 vrele(fromnd.ni_dvp); 2008 vrele(fvp); 2009 if (error == -1) 2010 error = 0; 2011 } 2012 vrele(tond.ni_startdir); 2013 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2014 out1: 2015 if (fdirp) { 2016 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp); 2017 vrele(fdirp); 2018 } 2019 if (tdirp) { 2020 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp); 2021 vrele(tdirp); 2022 } 2023 vrele(fromnd.ni_startdir); 2024 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2025 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2026 if (v3) { 2027 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2028 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2029 } 2030 vn_finished_write(mp, 0); 2031 return (0); 2032 2033 nfsmout: 2034 if (fdirp) 2035 vrele(fdirp); 2036 #ifdef notdef 2037 if (tdirp) 2038 vrele(tdirp); 2039 #endif 2040 if (tond.ni_cnd.cn_nameiop) { 2041 vrele(tond.ni_startdir); 2042 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2043 } 2044 if (fromnd.ni_cnd.cn_nameiop) { 2045 vrele(fromnd.ni_startdir); 2046 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2047 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2048 vrele(fromnd.ni_dvp); 2049 vrele(fvp); 2050 } 2051 return (error); 2052 } 2053 2054 /* 2055 * nfs link service 2056 */ 2057 int 2058 nfsrv_link(nfsd, slp, lwp, mrq) 2059 struct nfsrv_descript *nfsd; 2060 struct nfssvc_sock *slp; 2061 struct lwp *lwp; 2062 struct mbuf **mrq; 2063 { 2064 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2065 struct mbuf *nam = nfsd->nd_nam; 2066 caddr_t dpos = nfsd->nd_dpos; 2067 struct ucred *cred = &nfsd->nd_cr; 2068 struct nameidata nd; 2069 u_int32_t *tl; 2070 int32_t t1; 2071 caddr_t bpos; 2072 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2073 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2074 char *cp2; 2075 struct mbuf *mb, *mreq; 2076 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2077 struct vattr dirfor, diraft, at; 2078 nfsfh_t nfh, dnfh; 2079 fhandle_t *fhp, *dfhp; 2080 u_quad_t frev; 2081 struct mount *mp = NULL; 2082 2083 fhp = &nfh.fh_generic; 2084 dfhp = &dnfh.fh_generic; 2085 nfsm_srvmtofh(fhp); 2086 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2087 return (ESTALE); 2088 vn_start_write(NULL, &mp, V_WAIT); 2089 nfsm_srvmtofh(dfhp); 2090 nfsm_srvnamesiz(len); 2091 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam, 2092 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2093 if (error) { 2094 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2095 nfsm_srvpostop_attr(getret, &at); 2096 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2097 vn_finished_write(mp, 0); 2098 return (0); 2099 } 2100 if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0) 2101 goto out1; 2102 nd.ni_cnd.cn_cred = cred; 2103 nd.ni_cnd.cn_nameiop = CREATE; 2104 nd.ni_cnd.cn_flags = LOCKPARENT; 2105 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos, 2106 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2107 if (dirp) { 2108 if (v3) 2109 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2110 else { 2111 vrele(dirp); 2112 dirp = (struct vnode *)0; 2113 } 2114 } 2115 if (error) 2116 goto out1; 2117 xp = nd.ni_vp; 2118 if (xp != NULL) { 2119 error = EEXIST; 2120 goto out; 2121 } 2122 xp = nd.ni_dvp; 2123 if (vp->v_mount != xp->v_mount) 2124 error = EXDEV; 2125 out: 2126 if (!error) { 2127 nqsrv_getl(vp, ND_WRITE); 2128 nqsrv_getl(xp, ND_WRITE); 2129 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2130 } else { 2131 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2132 if (nd.ni_dvp == nd.ni_vp) 2133 vrele(nd.ni_dvp); 2134 else 2135 vput(nd.ni_dvp); 2136 if (nd.ni_vp) 2137 vrele(nd.ni_vp); 2138 } 2139 out1: 2140 if (v3) 2141 getret = VOP_GETATTR(vp, &at, cred, lwp); 2142 if (dirp) { 2143 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2144 vrele(dirp); 2145 } 2146 vrele(vp); 2147 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2148 if (v3) { 2149 nfsm_srvpostop_attr(getret, &at); 2150 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2151 vn_finished_write(mp, 0); 2152 return (0); 2153 } 2154 vn_finished_write(mp, 0); 2155 nfsm_srvdone; 2156 } 2157 2158 /* 2159 * nfs symbolic link service 2160 */ 2161 int 2162 nfsrv_symlink(nfsd, slp, lwp, mrq) 2163 struct nfsrv_descript *nfsd; 2164 struct nfssvc_sock *slp; 2165 struct lwp *lwp; 2166 struct mbuf **mrq; 2167 { 2168 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2169 struct mbuf *nam = nfsd->nd_nam; 2170 caddr_t dpos = nfsd->nd_dpos; 2171 struct ucred *cred = &nfsd->nd_cr; 2172 struct vattr va, dirfor, diraft; 2173 struct nameidata nd; 2174 u_int32_t *tl; 2175 int32_t t1; 2176 struct nfsv2_sattr *sp; 2177 char *bpos, *pathcp = NULL, *cp2; 2178 struct uio io; 2179 struct iovec iv; 2180 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1; 2181 uint32_t len, len2; 2182 int v3 = (nfsd->nd_flag & ND_NFSV3); 2183 struct mbuf *mb, *mreq; 2184 struct vnode *dirp = (struct vnode *)0; 2185 nfsfh_t nfh; 2186 fhandle_t *fhp; 2187 u_quad_t frev; 2188 struct mount *mp = NULL; 2189 2190 nd.ni_cnd.cn_nameiop = 0; 2191 fhp = &nfh.fh_generic; 2192 nfsm_srvmtofh(fhp); 2193 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2194 return (ESTALE); 2195 vn_start_write(NULL, &mp, V_WAIT); 2196 nfsm_srvnamesiz(len); 2197 nd.ni_cnd.cn_cred = cred; 2198 nd.ni_cnd.cn_nameiop = CREATE; 2199 nd.ni_cnd.cn_flags = LOCKPARENT; 2200 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2201 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2202 if (dirp) { 2203 if (v3) 2204 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2205 else { 2206 vrele(dirp); 2207 dirp = (struct vnode *)0; 2208 } 2209 } 2210 if (error) 2211 goto out; 2212 VATTR_NULL(&va); 2213 va.va_type = VLNK; 2214 if (v3) { 2215 va.va_mode = 0; 2216 nfsm_srvsattr(&va); 2217 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2218 len2 = fxdr_unsigned(uint32_t, *tl); 2219 if (len2 > PATH_MAX) { 2220 /* XXX should check _PC_NO_TRUNC */ 2221 error = ENAMETOOLONG; 2222 goto abortop; 2223 } 2224 } 2225 else { 2226 /* NFSv2 */ 2227 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2228 } 2229 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2230 iv.iov_base = pathcp; 2231 iv.iov_len = len2; 2232 io.uio_resid = len2; 2233 io.uio_offset = 0; 2234 io.uio_iov = &iv; 2235 io.uio_iovcnt = 1; 2236 io.uio_rw = UIO_READ; 2237 UIO_SETUP_SYSSPACE(&io); 2238 nfsm_mtouio(&io, len2); 2239 if (!v3) { 2240 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2241 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2242 } 2243 *(pathcp + len2) = '\0'; 2244 if (nd.ni_vp) { 2245 error = EEXIST; 2246 abortop: 2247 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2248 if (nd.ni_dvp == nd.ni_vp) 2249 vrele(nd.ni_dvp); 2250 else 2251 vput(nd.ni_dvp); 2252 if (nd.ni_vp) 2253 vrele(nd.ni_vp); 2254 goto out; 2255 } 2256 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2257 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2258 if (!error) { 2259 if (v3) { 2260 memset((caddr_t)fhp, 0, sizeof(nfh)); 2261 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsidx; 2262 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid); 2263 if (!error) 2264 error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp); 2265 vput(nd.ni_vp); 2266 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2267 } else { 2268 vput(nd.ni_vp); 2269 } 2270 } 2271 out: 2272 if (pathcp) 2273 free(pathcp, M_TEMP); 2274 if (dirp) { 2275 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2276 vrele(dirp); 2277 } 2278 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2279 if (v3) { 2280 if (!error) { 2281 nfsm_srvpostop_fh(fhp); 2282 nfsm_srvpostop_attr(0, &va); 2283 } 2284 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2285 } 2286 vn_finished_write(mp, 0); 2287 return (0); 2288 nfsmout: 2289 if (dirp) 2290 vrele(dirp); 2291 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2292 if (nd.ni_dvp == nd.ni_vp) 2293 vrele(nd.ni_dvp); 2294 else 2295 vput(nd.ni_dvp); 2296 if (nd.ni_vp) 2297 vrele(nd.ni_vp); 2298 if (pathcp) 2299 free(pathcp, M_TEMP); 2300 vn_finished_write(mp, 0); 2301 return (error); 2302 } 2303 2304 /* 2305 * nfs mkdir service 2306 */ 2307 int 2308 nfsrv_mkdir(nfsd, slp, lwp, mrq) 2309 struct nfsrv_descript *nfsd; 2310 struct nfssvc_sock *slp; 2311 struct lwp *lwp; 2312 struct mbuf **mrq; 2313 { 2314 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2315 struct mbuf *nam = nfsd->nd_nam; 2316 caddr_t dpos = nfsd->nd_dpos; 2317 struct ucred *cred = &nfsd->nd_cr; 2318 struct vattr va, dirfor, diraft; 2319 struct nfs_fattr *fp; 2320 struct nameidata nd; 2321 caddr_t cp; 2322 u_int32_t *tl; 2323 int32_t t1; 2324 caddr_t bpos; 2325 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2326 int v3 = (nfsd->nd_flag & ND_NFSV3); 2327 char *cp2; 2328 struct mbuf *mb, *mreq; 2329 struct vnode *vp, *dirp = (struct vnode *)0; 2330 nfsfh_t nfh; 2331 fhandle_t *fhp; 2332 u_quad_t frev; 2333 struct mount *mp = NULL; 2334 2335 fhp = &nfh.fh_generic; 2336 nfsm_srvmtofh(fhp); 2337 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2338 return (ESTALE); 2339 vn_start_write(NULL, &mp, V_WAIT); 2340 nfsm_srvnamesiz(len); 2341 nd.ni_cnd.cn_cred = cred; 2342 nd.ni_cnd.cn_nameiop = CREATE; 2343 nd.ni_cnd.cn_flags = LOCKPARENT; 2344 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2345 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2346 if (dirp) { 2347 if (v3) 2348 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2349 else { 2350 vrele(dirp); 2351 dirp = (struct vnode *)0; 2352 } 2353 } 2354 if (error) { 2355 nfsm_reply(NFSX_WCCDATA(v3)); 2356 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2357 if (dirp) 2358 vrele(dirp); 2359 vn_finished_write(mp, 0); 2360 return (0); 2361 } 2362 VATTR_NULL(&va); 2363 if (v3) { 2364 va.va_mode = 0; 2365 nfsm_srvsattr(&va); 2366 } else { 2367 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2368 va.va_mode = nfstov_mode(*tl++); 2369 } 2370 va.va_type = VDIR; 2371 vp = nd.ni_vp; 2372 if (vp != NULL) { 2373 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2374 if (nd.ni_dvp == vp) 2375 vrele(nd.ni_dvp); 2376 else 2377 vput(nd.ni_dvp); 2378 vrele(vp); 2379 error = EEXIST; 2380 goto out; 2381 } 2382 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2383 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2384 if (!error) { 2385 vp = nd.ni_vp; 2386 memset((caddr_t)fhp, 0, sizeof(nfh)); 2387 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsidx; 2388 error = VFS_VPTOFH(vp, &fhp->fh_fid); 2389 if (!error) 2390 error = VOP_GETATTR(vp, &va, cred, lwp); 2391 vput(vp); 2392 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 2393 } 2394 out: 2395 if (dirp) { 2396 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2397 vrele(dirp); 2398 } 2399 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2400 if (v3) { 2401 if (!error) { 2402 nfsm_srvpostop_fh(fhp); 2403 nfsm_srvpostop_attr(0, &va); 2404 } 2405 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2406 } else { 2407 nfsm_srvfhtom(fhp, v3); 2408 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2409 nfsm_srvfillattr(&va, fp); 2410 } 2411 vn_finished_write(mp, 0); 2412 return (0); 2413 nfsmout: 2414 if (dirp) 2415 vrele(dirp); 2416 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2417 if (nd.ni_dvp == nd.ni_vp) 2418 vrele(nd.ni_dvp); 2419 else 2420 vput(nd.ni_dvp); 2421 if (nd.ni_vp) 2422 vrele(nd.ni_vp); 2423 vn_finished_write(mp, 0); 2424 return (error); 2425 } 2426 2427 /* 2428 * nfs rmdir service 2429 */ 2430 int 2431 nfsrv_rmdir(nfsd, slp, lwp, mrq) 2432 struct nfsrv_descript *nfsd; 2433 struct nfssvc_sock *slp; 2434 struct lwp *lwp; 2435 struct mbuf **mrq; 2436 { 2437 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2438 struct mbuf *nam = nfsd->nd_nam; 2439 caddr_t dpos = nfsd->nd_dpos; 2440 struct ucred *cred = &nfsd->nd_cr; 2441 u_int32_t *tl; 2442 int32_t t1; 2443 caddr_t bpos; 2444 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2445 int v3 = (nfsd->nd_flag & ND_NFSV3); 2446 char *cp2; 2447 struct mbuf *mb, *mreq; 2448 struct vnode *vp, *dirp = (struct vnode *)0; 2449 struct vattr dirfor, diraft; 2450 nfsfh_t nfh; 2451 fhandle_t *fhp; 2452 struct nameidata nd; 2453 u_quad_t frev; 2454 struct mount *mp = NULL; 2455 2456 fhp = &nfh.fh_generic; 2457 nfsm_srvmtofh(fhp); 2458 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 2459 return (ESTALE); 2460 vn_start_write(NULL, &mp, V_WAIT); 2461 nfsm_srvnamesiz(len); 2462 nd.ni_cnd.cn_cred = cred; 2463 nd.ni_cnd.cn_nameiop = DELETE; 2464 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2465 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos, 2466 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2467 if (dirp) { 2468 if (v3) 2469 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2470 else { 2471 vrele(dirp); 2472 dirp = (struct vnode *)0; 2473 } 2474 } 2475 if (error) { 2476 nfsm_reply(NFSX_WCCDATA(v3)); 2477 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2478 if (dirp) 2479 vrele(dirp); 2480 vn_finished_write(mp, 0); 2481 return (0); 2482 } 2483 vp = nd.ni_vp; 2484 if (vp->v_type != VDIR) { 2485 error = ENOTDIR; 2486 goto out; 2487 } 2488 /* 2489 * No rmdir "." please. 2490 */ 2491 if (nd.ni_dvp == vp) { 2492 error = EINVAL; 2493 goto out; 2494 } 2495 /* 2496 * The root of a mounted filesystem cannot be deleted. 2497 */ 2498 if (vp->v_flag & VROOT) 2499 error = EBUSY; 2500 out: 2501 if (!error) { 2502 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2503 nqsrv_getl(vp, ND_WRITE); 2504 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2505 } else { 2506 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2507 if (nd.ni_dvp == nd.ni_vp) 2508 vrele(nd.ni_dvp); 2509 else 2510 vput(nd.ni_dvp); 2511 vput(vp); 2512 } 2513 if (dirp) { 2514 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2515 vrele(dirp); 2516 } 2517 nfsm_reply(NFSX_WCCDATA(v3)); 2518 if (v3) { 2519 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2520 vn_finished_write(mp, 0); 2521 return (0); 2522 } 2523 vn_finished_write(mp, 0); 2524 nfsm_srvdone; 2525 } 2526 2527 /* 2528 * nfs readdir service 2529 * - mallocs what it thinks is enough to read 2530 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2531 * - calls VOP_READDIR() 2532 * - loops around building the reply 2533 * if the output generated exceeds count break out of loop 2534 * The nfsm_clget macro is used here so that the reply will be packed 2535 * tightly in mbuf clusters. 2536 * - it only knows that it has encountered eof when the VOP_READDIR() 2537 * reads nothing 2538 * - as such one readdir rpc will return eof false although you are there 2539 * and then the next will return eof 2540 * - it trims out records with d_fileno == 0 2541 * this doesn't matter for Unix clients, but they might confuse clients 2542 * for other os'. 2543 * - it trims out records with d_type == DT_WHT 2544 * these cannot be seen through NFS (unless we extend the protocol) 2545 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2546 * than requested, but this may not apply to all filesystems. For 2547 * example, client NFS does not { although it is never remote mounted 2548 * anyhow } 2549 * The alternate call nfsrv_readdirplus() does lookups as well. 2550 * PS: The NFS protocol spec. does not clarify what the "count" byte 2551 * argument is a count of.. just name strings and file id's or the 2552 * entire reply rpc or ... 2553 * I tried just file name and id sizes and it confused the Sun client, 2554 * so I am using the full rpc size now. The "paranoia.." comment refers 2555 * to including the status longwords that are not a part of the dir. 2556 * "entry" structures, but are in the rpc. 2557 */ 2558 2559 #define NFS_SRVDIRBLKSIZ 1024 2560 2561 struct flrep { 2562 nfsuint64 fl_off; 2563 u_int32_t fl_postopok; 2564 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2565 u_int32_t fl_fhok; 2566 u_int32_t fl_fhsize; 2567 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)]; 2568 }; 2569 2570 int 2571 nfsrv_readdir(nfsd, slp, lwp, mrq) 2572 struct nfsrv_descript *nfsd; 2573 struct nfssvc_sock *slp; 2574 struct lwp *lwp; 2575 struct mbuf **mrq; 2576 { 2577 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2578 struct mbuf *nam = nfsd->nd_nam; 2579 caddr_t dpos = nfsd->nd_dpos; 2580 struct ucred *cred = &nfsd->nd_cr; 2581 char *bp, *be; 2582 struct mbuf *mp; 2583 struct dirent *dp; 2584 caddr_t cp; 2585 u_int32_t *tl; 2586 int32_t t1; 2587 caddr_t bpos; 2588 struct mbuf *mb, *mreq, *mp2; 2589 char *cpos, *cend, *cp2, *rbuf; 2590 struct vnode *vp; 2591 struct vattr at; 2592 nfsfh_t nfh; 2593 fhandle_t *fhp; 2594 struct uio io; 2595 struct iovec iv; 2596 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2597 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2598 int v3 = (nfsd->nd_flag & ND_NFSV3); 2599 u_quad_t frev, off, toff, verf; 2600 off_t *cookies = NULL, *cookiep; 2601 nfsuint64 jar; 2602 2603 fhp = &nfh.fh_generic; 2604 nfsm_srvmtofh(fhp); 2605 if (v3) { 2606 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2607 toff = fxdr_hyper(tl); 2608 tl += 2; 2609 verf = fxdr_hyper(tl); 2610 tl += 2; 2611 } else { 2612 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2613 toff = fxdr_unsigned(u_quad_t, *tl++); 2614 } 2615 off = toff; 2616 cnt = fxdr_unsigned(int, *tl); 2617 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2618 xfer = NFS_SRVMAXDATA(nfsd); 2619 if (siz > xfer) 2620 siz = xfer; 2621 fullsiz = siz; 2622 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2623 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2624 if (!error && vp->v_type != VDIR) { 2625 error = ENOTDIR; 2626 vput(vp); 2627 } 2628 if (error) { 2629 nfsm_reply(NFSX_UNSIGNED); 2630 nfsm_srvpostop_attr(getret, &at); 2631 return (0); 2632 } 2633 nqsrv_getl(vp, ND_READ); 2634 if (v3) { 2635 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2636 #ifdef NFS3_STRICTVERF 2637 /* 2638 * XXX This check is too strict for Solaris 2.5 clients. 2639 */ 2640 if (!error && toff && verf != at.va_filerev) 2641 error = NFSERR_BAD_COOKIE; 2642 #endif 2643 } 2644 if (!error) 2645 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2646 if (error) { 2647 vput(vp); 2648 nfsm_reply(NFSX_POSTOPATTR(v3)); 2649 nfsm_srvpostop_attr(getret, &at); 2650 return (0); 2651 } 2652 VOP_UNLOCK(vp, 0); 2653 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2654 again: 2655 iv.iov_base = rbuf; 2656 iv.iov_len = fullsiz; 2657 io.uio_iov = &iv; 2658 io.uio_iovcnt = 1; 2659 io.uio_offset = (off_t)off; 2660 io.uio_resid = fullsiz; 2661 io.uio_rw = UIO_READ; 2662 UIO_SETUP_SYSSPACE(&io); 2663 eofflag = 0; 2664 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2665 2666 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2667 2668 off = (off_t)io.uio_offset; 2669 if (!cookies && !error) 2670 error = NFSERR_PERM; 2671 if (v3) { 2672 getret = VOP_GETATTR(vp, &at, cred, lwp); 2673 if (!error) 2674 error = getret; 2675 } 2676 2677 VOP_UNLOCK(vp, 0); 2678 if (error) { 2679 vrele(vp); 2680 free((caddr_t)rbuf, M_TEMP); 2681 if (cookies) 2682 free((caddr_t)cookies, M_TEMP); 2683 nfsm_reply(NFSX_POSTOPATTR(v3)); 2684 nfsm_srvpostop_attr(getret, &at); 2685 return (0); 2686 } 2687 if (io.uio_resid) { 2688 siz -= io.uio_resid; 2689 2690 /* 2691 * If nothing read, return eof 2692 * rpc reply 2693 */ 2694 if (siz == 0) { 2695 vrele(vp); 2696 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2697 2 * NFSX_UNSIGNED); 2698 if (v3) { 2699 nfsm_srvpostop_attr(getret, &at); 2700 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2701 txdr_hyper(at.va_filerev, tl); 2702 tl += 2; 2703 } else 2704 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2705 *tl++ = nfs_false; 2706 *tl = nfs_true; 2707 free((caddr_t)rbuf, M_TEMP); 2708 free((caddr_t)cookies, M_TEMP); 2709 return (0); 2710 } 2711 } 2712 2713 /* 2714 * Check for degenerate cases of nothing useful read. 2715 * If so go try again 2716 */ 2717 cpos = rbuf; 2718 cend = rbuf + siz; 2719 dp = (struct dirent *)cpos; 2720 cookiep = cookies; 2721 2722 while (cpos < cend && ncookies > 0 && 2723 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2724 cpos += dp->d_reclen; 2725 dp = (struct dirent *)cpos; 2726 cookiep++; 2727 ncookies--; 2728 } 2729 if (cpos >= cend || ncookies == 0) { 2730 toff = off; 2731 siz = fullsiz; 2732 goto again; 2733 } 2734 2735 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2736 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2737 if (v3) { 2738 nfsm_srvpostop_attr(getret, &at); 2739 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2740 txdr_hyper(at.va_filerev, tl); 2741 } 2742 mp = mp2 = mb; 2743 bp = bpos; 2744 be = bp + M_TRAILINGSPACE(mp); 2745 2746 /* Loop through the records and build reply */ 2747 while (cpos < cend && ncookies > 0) { 2748 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2749 nlen = dp->d_namlen; 2750 rem = nfsm_rndup(nlen)-nlen; 2751 len += (4 * NFSX_UNSIGNED + nlen + rem); 2752 if (v3) 2753 len += 2 * NFSX_UNSIGNED; 2754 if (len > cnt) { 2755 eofflag = 0; 2756 break; 2757 } 2758 /* 2759 * Build the directory record xdr from 2760 * the dirent entry. 2761 */ 2762 nfsm_clget; 2763 *tl = nfs_true; 2764 bp += NFSX_UNSIGNED; 2765 if (v3) { 2766 nfsm_clget; 2767 *tl = txdr_unsigned(dp->d_fileno >> 32); 2768 bp += NFSX_UNSIGNED; 2769 } 2770 nfsm_clget; 2771 *tl = txdr_unsigned(dp->d_fileno); 2772 bp += NFSX_UNSIGNED; 2773 nfsm_clget; 2774 *tl = txdr_unsigned(nlen); 2775 bp += NFSX_UNSIGNED; 2776 2777 /* And loop around copying the name */ 2778 xfer = nlen; 2779 cp = dp->d_name; 2780 while (xfer > 0) { 2781 nfsm_clget; 2782 if ((bp+xfer) > be) 2783 tsiz = be-bp; 2784 else 2785 tsiz = xfer; 2786 memcpy(bp, cp, tsiz); 2787 bp += tsiz; 2788 xfer -= tsiz; 2789 if (xfer > 0) 2790 cp += tsiz; 2791 } 2792 /* And null pad to an int32_t boundary */ 2793 for (i = 0; i < rem; i++) 2794 *bp++ = '\0'; 2795 nfsm_clget; 2796 2797 /* Finish off the record */ 2798 txdr_hyper(*cookiep, &jar); 2799 if (v3) { 2800 *tl = jar.nfsuquad[0]; 2801 bp += NFSX_UNSIGNED; 2802 nfsm_clget; 2803 } 2804 *tl = jar.nfsuquad[1]; 2805 bp += NFSX_UNSIGNED; 2806 } 2807 cpos += dp->d_reclen; 2808 dp = (struct dirent *)cpos; 2809 cookiep++; 2810 ncookies--; 2811 } 2812 vrele(vp); 2813 nfsm_clget; 2814 *tl = nfs_false; 2815 bp += NFSX_UNSIGNED; 2816 nfsm_clget; 2817 if (eofflag) 2818 *tl = nfs_true; 2819 else 2820 *tl = nfs_false; 2821 bp += NFSX_UNSIGNED; 2822 if (mp != mb) { 2823 if (bp < be) 2824 mp->m_len = bp - mtod(mp, caddr_t); 2825 } else 2826 mp->m_len += bp - bpos; 2827 free((caddr_t)rbuf, M_TEMP); 2828 free((caddr_t)cookies, M_TEMP); 2829 nfsm_srvdone; 2830 } 2831 2832 int 2833 nfsrv_readdirplus(nfsd, slp, lwp, mrq) 2834 struct nfsrv_descript *nfsd; 2835 struct nfssvc_sock *slp; 2836 struct lwp *lwp; 2837 struct mbuf **mrq; 2838 { 2839 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2840 struct mbuf *nam = nfsd->nd_nam; 2841 caddr_t dpos = nfsd->nd_dpos; 2842 struct ucred *cred = &nfsd->nd_cr; 2843 char *bp, *be; 2844 struct mbuf *mp; 2845 struct dirent *dp; 2846 caddr_t cp; 2847 u_int32_t *tl; 2848 int32_t t1; 2849 caddr_t bpos; 2850 struct mbuf *mb, *mreq, *mp2; 2851 char *cpos, *cend, *cp2, *rbuf; 2852 struct vnode *vp, *nvp; 2853 struct flrep fl; 2854 nfsfh_t nfh; 2855 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh; 2856 struct uio io; 2857 struct iovec iv; 2858 struct vattr va, at, *vap = &va; 2859 struct nfs_fattr *fp; 2860 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2861 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2862 u_quad_t frev, off, toff, verf; 2863 off_t *cookies = NULL, *cookiep; 2864 2865 fhp = &nfh.fh_generic; 2866 nfsm_srvmtofh(fhp); 2867 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2868 toff = fxdr_hyper(tl); 2869 tl += 2; 2870 verf = fxdr_hyper(tl); 2871 tl += 2; 2872 siz = fxdr_unsigned(int, *tl++); 2873 cnt = fxdr_unsigned(int, *tl); 2874 off = toff; 2875 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2876 xfer = NFS_SRVMAXDATA(nfsd); 2877 if (siz > xfer) 2878 siz = xfer; 2879 fullsiz = siz; 2880 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 2881 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 2882 if (!error && vp->v_type != VDIR) { 2883 error = ENOTDIR; 2884 vput(vp); 2885 } 2886 if (error) { 2887 nfsm_reply(NFSX_UNSIGNED); 2888 nfsm_srvpostop_attr(getret, &at); 2889 return (0); 2890 } 2891 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2892 #ifdef NFS3_STRICTVERF 2893 /* 2894 * XXX This check is too strict for Solaris 2.5 clients. 2895 */ 2896 if (!error && toff && verf != at.va_filerev) 2897 error = NFSERR_BAD_COOKIE; 2898 #endif 2899 if (!error) { 2900 nqsrv_getl(vp, ND_READ); 2901 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2902 } 2903 if (error) { 2904 vput(vp); 2905 nfsm_reply(NFSX_V3POSTOPATTR); 2906 nfsm_srvpostop_attr(getret, &at); 2907 return (0); 2908 } 2909 VOP_UNLOCK(vp, 0); 2910 2911 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2912 again: 2913 iv.iov_base = rbuf; 2914 iv.iov_len = fullsiz; 2915 io.uio_iov = &iv; 2916 io.uio_iovcnt = 1; 2917 io.uio_offset = (off_t)off; 2918 io.uio_resid = fullsiz; 2919 io.uio_rw = UIO_READ; 2920 UIO_SETUP_SYSSPACE(&io); 2921 eofflag = 0; 2922 2923 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2924 2925 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2926 2927 off = (u_quad_t)io.uio_offset; 2928 getret = VOP_GETATTR(vp, &at, cred, lwp); 2929 2930 VOP_UNLOCK(vp, 0); 2931 2932 /* 2933 * If the VGET operation doesn't work for this filesystem, 2934 * we can't support readdirplus. Returning NOTSUPP should 2935 * make clients fall back to plain readdir. 2936 * There's no need to check for VPTOFH as well, we wouldn't 2937 * even be here otherwise. 2938 */ 2939 if (!getret) { 2940 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2941 getret = (getret == EOPNOTSUPP) ? 2942 NFSERR_NOTSUPP : NFSERR_IO; 2943 else 2944 vput(nvp); 2945 } 2946 2947 if (!cookies && !error) 2948 error = NFSERR_PERM; 2949 if (!error) 2950 error = getret; 2951 if (error) { 2952 vrele(vp); 2953 if (cookies) 2954 free((caddr_t)cookies, M_TEMP); 2955 free((caddr_t)rbuf, M_TEMP); 2956 nfsm_reply(NFSX_V3POSTOPATTR); 2957 nfsm_srvpostop_attr(getret, &at); 2958 return (0); 2959 } 2960 if (io.uio_resid) { 2961 siz -= io.uio_resid; 2962 2963 /* 2964 * If nothing read, return eof 2965 * rpc reply 2966 */ 2967 if (siz == 0) { 2968 vrele(vp); 2969 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2970 2 * NFSX_UNSIGNED); 2971 nfsm_srvpostop_attr(getret, &at); 2972 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2973 txdr_hyper(at.va_filerev, tl); 2974 tl += 2; 2975 *tl++ = nfs_false; 2976 *tl = nfs_true; 2977 free((caddr_t)cookies, M_TEMP); 2978 free((caddr_t)rbuf, M_TEMP); 2979 return (0); 2980 } 2981 } 2982 2983 /* 2984 * Check for degenerate cases of nothing useful read. 2985 * If so go try again 2986 */ 2987 cpos = rbuf; 2988 cend = rbuf + siz; 2989 dp = (struct dirent *)cpos; 2990 cookiep = cookies; 2991 2992 while (cpos < cend && ncookies > 0 && 2993 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2994 cpos += dp->d_reclen; 2995 dp = (struct dirent *)cpos; 2996 cookiep++; 2997 ncookies--; 2998 } 2999 if (cpos >= cend || ncookies == 0) { 3000 toff = off; 3001 siz = fullsiz; 3002 goto again; 3003 } 3004 3005 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 3006 nfsm_reply(cnt); 3007 nfsm_srvpostop_attr(getret, &at); 3008 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3009 txdr_hyper(at.va_filerev, tl); 3010 mp = mp2 = mb; 3011 bp = bpos; 3012 be = bp + M_TRAILINGSPACE(mp); 3013 3014 /* Loop through the records and build reply */ 3015 while (cpos < cend && ncookies > 0) { 3016 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 3017 nlen = dp->d_namlen; 3018 rem = nfsm_rndup(nlen)-nlen; 3019 3020 /* 3021 * For readdir_and_lookup get the vnode using 3022 * the file number. 3023 */ 3024 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3025 goto invalid; 3026 memset((caddr_t)nfhp, 0, NFSX_V3FH); 3027 nfhp->fh_fsid = 3028 nvp->v_mount->mnt_stat.f_fsidx; 3029 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) { 3030 vput(nvp); 3031 goto invalid; 3032 } 3033 if (VOP_GETATTR(nvp, vap, cred, lwp)) { 3034 vput(nvp); 3035 goto invalid; 3036 } 3037 vput(nvp); 3038 KASSERT(fhp->fh_fid.fid_len <= _VFS_MAXFIDSZ); 3039 3040 /* 3041 * If either the dircount or maxcount will be 3042 * exceeded, get out now. Both of these lengths 3043 * are calculated conservatively, including all 3044 * XDR overheads. 3045 */ 3046 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3047 NFSX_V3POSTOPATTR); 3048 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3049 if (len > cnt || dirlen > fullsiz) { 3050 eofflag = 0; 3051 break; 3052 } 3053 3054 /* 3055 * Build the directory record xdr from 3056 * the dirent entry. 3057 */ 3058 fp = (struct nfs_fattr *)&fl.fl_fattr; 3059 nfsm_srvfillattr(vap, fp); 3060 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3061 fl.fl_fhok = nfs_true; 3062 fl.fl_postopok = nfs_true; 3063 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3064 3065 nfsm_clget; 3066 *tl = nfs_true; 3067 bp += NFSX_UNSIGNED; 3068 nfsm_clget; 3069 *tl = txdr_unsigned(dp->d_fileno >> 32); 3070 bp += NFSX_UNSIGNED; 3071 nfsm_clget; 3072 *tl = txdr_unsigned(dp->d_fileno); 3073 bp += NFSX_UNSIGNED; 3074 nfsm_clget; 3075 *tl = txdr_unsigned(nlen); 3076 bp += NFSX_UNSIGNED; 3077 3078 /* And loop around copying the name */ 3079 xfer = nlen; 3080 cp = dp->d_name; 3081 while (xfer > 0) { 3082 nfsm_clget; 3083 if ((bp + xfer) > be) 3084 tsiz = be - bp; 3085 else 3086 tsiz = xfer; 3087 memcpy(bp, cp, tsiz); 3088 bp += tsiz; 3089 xfer -= tsiz; 3090 if (xfer > 0) 3091 cp += tsiz; 3092 } 3093 /* And null pad to an int32_t boundary */ 3094 for (i = 0; i < rem; i++) 3095 *bp++ = '\0'; 3096 3097 /* 3098 * Now copy the flrep structure out. 3099 */ 3100 xfer = sizeof (struct flrep); 3101 cp = (caddr_t)&fl; 3102 while (xfer > 0) { 3103 nfsm_clget; 3104 if ((bp + xfer) > be) 3105 tsiz = be - bp; 3106 else 3107 tsiz = xfer; 3108 memcpy(bp, cp, tsiz); 3109 bp += tsiz; 3110 xfer -= tsiz; 3111 if (xfer > 0) 3112 cp += tsiz; 3113 } 3114 } 3115 invalid: 3116 cpos += dp->d_reclen; 3117 dp = (struct dirent *)cpos; 3118 cookiep++; 3119 ncookies--; 3120 } 3121 vrele(vp); 3122 nfsm_clget; 3123 *tl = nfs_false; 3124 bp += NFSX_UNSIGNED; 3125 nfsm_clget; 3126 if (eofflag) 3127 *tl = nfs_true; 3128 else 3129 *tl = nfs_false; 3130 bp += NFSX_UNSIGNED; 3131 if (mp != mb) { 3132 if (bp < be) 3133 mp->m_len = bp - mtod(mp, caddr_t); 3134 } else 3135 mp->m_len += bp - bpos; 3136 free((caddr_t)cookies, M_TEMP); 3137 free((caddr_t)rbuf, M_TEMP); 3138 nfsm_srvdone; 3139 } 3140 3141 /* 3142 * nfs commit service 3143 */ 3144 int 3145 nfsrv_commit(nfsd, slp, lwp, mrq) 3146 struct nfsrv_descript *nfsd; 3147 struct nfssvc_sock *slp; 3148 struct lwp *lwp; 3149 struct mbuf **mrq; 3150 { 3151 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3152 struct mbuf *nam = nfsd->nd_nam; 3153 caddr_t dpos = nfsd->nd_dpos; 3154 struct ucred *cred = &nfsd->nd_cr; 3155 struct vattr bfor, aft; 3156 struct vnode *vp; 3157 nfsfh_t nfh; 3158 fhandle_t *fhp; 3159 u_int32_t *tl; 3160 int32_t t1; 3161 caddr_t bpos; 3162 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3163 uint32_t cnt; 3164 char *cp2; 3165 struct mbuf *mb, *mreq; 3166 u_quad_t frev, off, end; 3167 struct mount *mp = NULL; 3168 3169 fhp = &nfh.fh_generic; 3170 nfsm_srvmtofh(fhp); 3171 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) 3172 return (ESTALE); 3173 vn_start_write(NULL, &mp, V_WAIT); 3174 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3175 3176 off = fxdr_hyper(tl); 3177 tl += 2; 3178 cnt = fxdr_unsigned(uint32_t, *tl); 3179 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3180 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3181 if (error) { 3182 nfsm_reply(2 * NFSX_UNSIGNED); 3183 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3184 vn_finished_write(mp, 0); 3185 return (0); 3186 } 3187 for_ret = VOP_GETATTR(vp, &bfor, cred, lwp); 3188 end = (cnt > 0) ? off + cnt : vp->v_size; 3189 if (end < off || end > vp->v_size) 3190 end = vp->v_size; 3191 if (off < vp->v_size) 3192 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp); 3193 /* else error == 0, from nfsrv_fhtovp() */ 3194 aft_ret = VOP_GETATTR(vp, &aft, cred, lwp); 3195 vput(vp); 3196 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3197 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3198 if (!error) { 3199 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3200 *tl++ = txdr_unsigned(boottime.tv_sec); 3201 *tl = txdr_unsigned(boottime.tv_usec); 3202 } else { 3203 vn_finished_write(mp, 0); 3204 return (0); 3205 } 3206 vn_finished_write(mp, 0); 3207 nfsm_srvdone; 3208 } 3209 3210 /* 3211 * nfs statfs service 3212 */ 3213 int 3214 nfsrv_statfs(nfsd, slp, lwp, mrq) 3215 struct nfsrv_descript *nfsd; 3216 struct nfssvc_sock *slp; 3217 struct lwp *lwp; 3218 struct mbuf **mrq; 3219 { 3220 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3221 struct mbuf *nam = nfsd->nd_nam; 3222 caddr_t dpos = nfsd->nd_dpos; 3223 struct ucred *cred = &nfsd->nd_cr; 3224 struct statvfs *sf; 3225 struct nfs_statfs *sfp; 3226 u_int32_t *tl; 3227 int32_t t1; 3228 caddr_t bpos; 3229 int error = 0, rdonly, cache = 0, getret = 1; 3230 int v3 = (nfsd->nd_flag & ND_NFSV3); 3231 char *cp2; 3232 struct mbuf *mb, *mreq; 3233 struct vnode *vp; 3234 struct vattr at; 3235 nfsfh_t nfh; 3236 fhandle_t *fhp; 3237 struct statvfs statvfs; 3238 u_quad_t frev, tval; 3239 3240 fhp = &nfh.fh_generic; 3241 nfsm_srvmtofh(fhp); 3242 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3243 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3244 if (error) { 3245 nfsm_reply(NFSX_UNSIGNED); 3246 nfsm_srvpostop_attr(getret, &at); 3247 return (0); 3248 } 3249 sf = &statvfs; 3250 error = VFS_STATVFS(vp->v_mount, sf, lwp); 3251 getret = VOP_GETATTR(vp, &at, cred, lwp); 3252 vput(vp); 3253 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3254 if (v3) 3255 nfsm_srvpostop_attr(getret, &at); 3256 if (error) 3257 return (0); 3258 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3259 if (v3) { 3260 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3261 txdr_hyper(tval, &sfp->sf_tbytes); 3262 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3263 txdr_hyper(tval, &sfp->sf_fbytes); 3264 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3265 txdr_hyper(tval, &sfp->sf_abytes); 3266 tval = (u_quad_t)sf->f_files; 3267 txdr_hyper(tval, &sfp->sf_tfiles); 3268 tval = (u_quad_t)sf->f_ffree; 3269 txdr_hyper(tval, &sfp->sf_ffiles); 3270 txdr_hyper(tval, &sfp->sf_afiles); 3271 sfp->sf_invarsec = 0; 3272 } else { 3273 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3274 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3275 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3276 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3277 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3278 } 3279 nfsm_srvdone; 3280 } 3281 3282 /* 3283 * nfs fsinfo service 3284 */ 3285 int 3286 nfsrv_fsinfo(nfsd, slp, lwp, mrq) 3287 struct nfsrv_descript *nfsd; 3288 struct nfssvc_sock *slp; 3289 struct lwp *lwp; 3290 struct mbuf **mrq; 3291 { 3292 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3293 struct mbuf *nam = nfsd->nd_nam; 3294 caddr_t dpos = nfsd->nd_dpos; 3295 struct ucred *cred = &nfsd->nd_cr; 3296 u_int32_t *tl; 3297 struct nfsv3_fsinfo *sip; 3298 int32_t t1; 3299 caddr_t bpos; 3300 int error = 0, rdonly, cache = 0, getret = 1; 3301 uint32_t maxdata; 3302 char *cp2; 3303 struct mbuf *mb, *mreq; 3304 struct vnode *vp; 3305 struct vattr at; 3306 nfsfh_t nfh; 3307 fhandle_t *fhp; 3308 u_quad_t frev, maxfsize; 3309 struct statvfs sb; 3310 3311 fhp = &nfh.fh_generic; 3312 nfsm_srvmtofh(fhp); 3313 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3314 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3315 if (error) { 3316 nfsm_reply(NFSX_UNSIGNED); 3317 nfsm_srvpostop_attr(getret, &at); 3318 return (0); 3319 } 3320 3321 /* XXX Try to make a guess on the max file size. */ 3322 VFS_STATVFS(vp->v_mount, &sb, (struct lwp *)0); 3323 maxfsize = (u_quad_t)0x80000000 * sb.f_frsize - 1; 3324 3325 getret = VOP_GETATTR(vp, &at, cred, lwp); 3326 vput(vp); 3327 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3328 nfsm_srvpostop_attr(getret, &at); 3329 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3330 3331 /* 3332 * XXX 3333 * There should be file system VFS OP(s) to get this information. 3334 * For now, assume ufs. 3335 */ 3336 if (slp->ns_so->so_type == SOCK_DGRAM) 3337 maxdata = NFS_MAXDGRAMDATA; 3338 else 3339 maxdata = NFS_MAXDATA; 3340 sip->fs_rtmax = txdr_unsigned(maxdata); 3341 sip->fs_rtpref = txdr_unsigned(maxdata); 3342 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3343 sip->fs_wtmax = txdr_unsigned(maxdata); 3344 sip->fs_wtpref = txdr_unsigned(maxdata); 3345 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3346 sip->fs_dtpref = txdr_unsigned(maxdata); 3347 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3348 sip->fs_timedelta.nfsv3_sec = 0; 3349 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3350 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3351 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3352 NFSV3FSINFO_CANSETTIME); 3353 nfsm_srvdone; 3354 } 3355 3356 /* 3357 * nfs pathconf service 3358 */ 3359 int 3360 nfsrv_pathconf(nfsd, slp, lwp, mrq) 3361 struct nfsrv_descript *nfsd; 3362 struct nfssvc_sock *slp; 3363 struct lwp *lwp; 3364 struct mbuf **mrq; 3365 { 3366 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3367 struct mbuf *nam = nfsd->nd_nam; 3368 caddr_t dpos = nfsd->nd_dpos; 3369 struct ucred *cred = &nfsd->nd_cr; 3370 u_int32_t *tl; 3371 struct nfsv3_pathconf *pc; 3372 int32_t t1; 3373 caddr_t bpos; 3374 int error = 0, rdonly, cache = 0, getret = 1; 3375 register_t linkmax, namemax, chownres, notrunc; 3376 char *cp2; 3377 struct mbuf *mb, *mreq; 3378 struct vnode *vp; 3379 struct vattr at; 3380 nfsfh_t nfh; 3381 fhandle_t *fhp; 3382 u_quad_t frev; 3383 3384 fhp = &nfh.fh_generic; 3385 nfsm_srvmtofh(fhp); 3386 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, 3387 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE); 3388 if (error) { 3389 nfsm_reply(NFSX_UNSIGNED); 3390 nfsm_srvpostop_attr(getret, &at); 3391 return (0); 3392 } 3393 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3394 if (!error) 3395 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3396 if (!error) 3397 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3398 if (!error) 3399 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3400 getret = VOP_GETATTR(vp, &at, cred, lwp); 3401 vput(vp); 3402 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3403 nfsm_srvpostop_attr(getret, &at); 3404 if (error) 3405 return (0); 3406 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3407 3408 pc->pc_linkmax = txdr_unsigned(linkmax); 3409 pc->pc_namemax = txdr_unsigned(namemax); 3410 pc->pc_notrunc = txdr_unsigned(notrunc); 3411 pc->pc_chownrestricted = txdr_unsigned(chownres); 3412 3413 /* 3414 * These should probably be supported by VOP_PATHCONF(), but 3415 * until msdosfs is exportable (why would you want to?), the 3416 * Unix defaults should be ok. 3417 */ 3418 pc->pc_caseinsensitive = nfs_false; 3419 pc->pc_casepreserving = nfs_true; 3420 nfsm_srvdone; 3421 } 3422 3423 /* 3424 * Null operation, used by clients to ping server 3425 */ 3426 /* ARGSUSED */ 3427 int 3428 nfsrv_null(nfsd, slp, lwp, mrq) 3429 struct nfsrv_descript *nfsd; 3430 struct nfssvc_sock *slp; 3431 struct lwp *lwp; 3432 struct mbuf **mrq; 3433 { 3434 struct mbuf *mrep = nfsd->nd_mrep; 3435 caddr_t bpos; 3436 int error = NFSERR_RETVOID, cache = 0; 3437 struct mbuf *mb, *mreq; 3438 u_quad_t frev; 3439 3440 nfsm_reply(0); 3441 return (0); 3442 } 3443 3444 /* 3445 * No operation, used for obsolete procedures 3446 */ 3447 /* ARGSUSED */ 3448 int 3449 nfsrv_noop(nfsd, slp, lwp, mrq) 3450 struct nfsrv_descript *nfsd; 3451 struct nfssvc_sock *slp; 3452 struct lwp *lwp; 3453 struct mbuf **mrq; 3454 { 3455 struct mbuf *mrep = nfsd->nd_mrep; 3456 caddr_t bpos; 3457 int error, cache = 0; 3458 struct mbuf *mb, *mreq; 3459 u_quad_t frev; 3460 3461 if (nfsd->nd_repstat) 3462 error = nfsd->nd_repstat; 3463 else 3464 error = EPROCUNAVAIL; 3465 nfsm_reply(0); 3466 return (0); 3467 } 3468 3469 /* 3470 * Perform access checking for vnodes obtained from file handles that would 3471 * refer to files already opened by a Unix client. You cannot just use 3472 * vn_writechk() and VOP_ACCESS() for two reasons. 3473 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3474 * 2 - The owner is to be given access irrespective of mode bits for some 3475 * operations, so that processes that chmod after opening a file don't 3476 * break. I don't like this because it opens a security hole, but since 3477 * the nfs server opens a security hole the size of a barn door anyhow, 3478 * what the heck. 3479 * 3480 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3481 * will return EPERM instead of EACCESS. EPERM is always an error. 3482 */ 3483 int 3484 nfsrv_access(vp, flags, cred, rdonly, lwp, override) 3485 struct vnode *vp; 3486 int flags; 3487 struct ucred *cred; 3488 int rdonly; 3489 struct lwp *lwp; 3490 { 3491 struct vattr vattr; 3492 int error; 3493 if (flags & VWRITE) { 3494 /* Just vn_writechk() changed to check rdonly */ 3495 /* 3496 * Disallow write attempts on read-only file systems; 3497 * unless the file is a socket or a block or character 3498 * device resident on the file system. 3499 */ 3500 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3501 switch (vp->v_type) { 3502 case VREG: 3503 case VDIR: 3504 case VLNK: 3505 return (EROFS); 3506 default: 3507 break; 3508 } 3509 } 3510 3511 /* 3512 * If the vnode is in use as a process's text, 3513 * we can't allow writing. 3514 */ 3515 if (vp->v_flag & VTEXT) 3516 return (ETXTBSY); 3517 } 3518 error = VOP_GETATTR(vp, &vattr, cred, lwp); 3519 if (error) 3520 return (error); 3521 error = VOP_ACCESS(vp, flags, cred, lwp); 3522 /* 3523 * Allow certain operations for the owner (reads and writes 3524 * on files that are already open). 3525 */ 3526 if (override && error == EACCES && cred->cr_uid == vattr.va_uid) 3527 error = 0; 3528 return error; 3529 } 3530
Indexes created Thu Oct 02 14:10:14 GMT 2025