nfs_serv.c revision 1.127
1 /* $NetBSD: nfs_serv.c,v 1.127 2007/03/05 23:18:50 yamt 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 1.127 2007/03/05 23:18:50 yamt 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 #include <sys/hash.h> 74 #include <sys/kauth.h> 75 76 #include <uvm/uvm.h> 77 78 #include <nfs/nfsproto.h> 79 #include <nfs/rpcv2.h> 80 #include <nfs/nfs.h> 81 #include <nfs/xdr_subs.h> 82 #include <nfs/nfsm_subs.h> 83 #include <nfs/nfs_var.h> 84 85 /* Global vars */ 86 extern u_int32_t nfs_xdrneg1; 87 extern u_int32_t nfs_false, nfs_true; 88 extern enum vtype nv3tov_type[8]; 89 extern struct nfsstats nfsstats; 90 extern const nfstype nfsv2_type[9]; 91 extern const nfstype nfsv3_type[9]; 92 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000; 93 int nfsd_use_loan = 1; /* use page-loan for READ OP */ 94 95 #define nqsrv_getl(vp, rw) /* nothing */ 96 97 /* 98 * nfs v3 access service 99 */ 100 int 101 nfsrv3_access(nfsd, slp, lwp, mrq) 102 struct nfsrv_descript *nfsd; 103 struct nfssvc_sock *slp; 104 struct lwp *lwp; 105 struct mbuf **mrq; 106 { 107 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 108 struct mbuf *nam = nfsd->nd_nam; 109 char *dpos = nfsd->nd_dpos; 110 kauth_cred_t cred = nfsd->nd_cr; 111 struct vnode *vp; 112 nfsrvfh_t nsfh; 113 u_int32_t *tl; 114 int32_t t1; 115 char *bpos; 116 int error = 0, rdonly, cache = 0, getret; 117 char *cp2; 118 struct mbuf *mb, *mreq; 119 struct vattr va; 120 u_long inmode, testmode, outmode; 121 u_quad_t frev; 122 123 nfsm_srvmtofh(&nsfh); 124 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 125 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 126 (nfsd->nd_flag & ND_KERBAUTH), false); 127 if (error) { 128 nfsm_reply(NFSX_UNSIGNED); 129 nfsm_srvpostop_attr(1, (struct vattr *)0); 130 return (0); 131 } 132 inmode = fxdr_unsigned(u_int32_t, *tl); 133 outmode = 0; 134 if ((inmode & NFSV3ACCESS_READ) && 135 nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0) 136 outmode |= NFSV3ACCESS_READ; 137 if (vp->v_type != VDIR) { 138 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 139 if (testmode && 140 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 141 outmode |= testmode; 142 if ((inmode & NFSV3ACCESS_EXECUTE) && 143 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 144 outmode |= NFSV3ACCESS_EXECUTE; 145 } else { 146 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 147 NFSV3ACCESS_DELETE); 148 if (testmode && 149 nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0) 150 outmode |= testmode; 151 if ((inmode & NFSV3ACCESS_LOOKUP) && 152 nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0) 153 outmode |= NFSV3ACCESS_LOOKUP; 154 } 155 getret = VOP_GETATTR(vp, &va, cred, lwp); 156 vput(vp); 157 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED); 158 nfsm_srvpostop_attr(getret, &va); 159 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 160 *tl = txdr_unsigned(outmode); 161 nfsm_srvdone; 162 } 163 164 /* 165 * nfs getattr service 166 */ 167 int 168 nfsrv_getattr(nfsd, slp, lwp, mrq) 169 struct nfsrv_descript *nfsd; 170 struct nfssvc_sock *slp; 171 struct lwp *lwp; 172 struct mbuf **mrq; 173 { 174 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 175 struct mbuf *nam = nfsd->nd_nam; 176 char *dpos = nfsd->nd_dpos; 177 kauth_cred_t cred = nfsd->nd_cr; 178 struct nfs_fattr *fp; 179 struct vattr va; 180 struct vnode *vp; 181 nfsrvfh_t nsfh; 182 u_int32_t *tl; 183 int32_t t1; 184 char *bpos; 185 int error = 0, rdonly, cache = 0; 186 char *cp2; 187 struct mbuf *mb, *mreq; 188 u_quad_t frev; 189 190 nfsm_srvmtofh(&nsfh); 191 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 192 (nfsd->nd_flag & ND_KERBAUTH), false); 193 if (error) { 194 nfsm_reply(0); 195 return (0); 196 } 197 nqsrv_getl(vp, ND_READ); 198 error = VOP_GETATTR(vp, &va, cred, lwp); 199 vput(vp); 200 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 201 if (error) 202 return (0); 203 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3)); 204 nfsm_srvfillattr(&va, fp); 205 nfsm_srvdone; 206 } 207 208 /* 209 * nfs setattr service 210 */ 211 int 212 nfsrv_setattr(nfsd, slp, lwp, mrq) 213 struct nfsrv_descript *nfsd; 214 struct nfssvc_sock *slp; 215 struct lwp *lwp; 216 struct mbuf **mrq; 217 { 218 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 219 struct mbuf *nam = nfsd->nd_nam; 220 char *dpos = nfsd->nd_dpos; 221 kauth_cred_t cred = nfsd->nd_cr; 222 struct vattr va, preat; 223 struct nfsv2_sattr *sp; 224 struct nfs_fattr *fp; 225 struct vnode *vp; 226 nfsrvfh_t nsfh; 227 u_int32_t *tl; 228 int32_t t1; 229 char *bpos; 230 int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1; 231 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0; 232 char *cp2; 233 struct mbuf *mb, *mreq; 234 u_quad_t frev; 235 struct timespec guard; 236 struct mount *mp = NULL; 237 238 memset(&guard, 0, sizeof guard); /* XXX gcc */ 239 240 nfsm_srvmtofh(&nsfh); 241 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 242 return (ESTALE); 243 vn_start_write(NULL, &mp, V_WAIT); 244 VATTR_NULL(&va); 245 if (v3) { 246 nfsm_srvsattr(&va); 247 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 248 gcheck = fxdr_unsigned(int, *tl); 249 if (gcheck) { 250 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 251 fxdr_nfsv3time(tl, &guard); 252 } 253 } else { 254 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 255 /* 256 * Nah nah nah nah na nah 257 * There is a bug in the Sun client that puts 0xffff in the mode 258 * field of sattr when it should put in 0xffffffff. The u_short 259 * doesn't sign extend. 260 * --> check the low order 2 bytes for 0xffff 261 */ 262 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 263 va.va_mode = nfstov_mode(sp->sa_mode); 264 if (sp->sa_uid != nfs_xdrneg1) 265 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid); 266 if (sp->sa_gid != nfs_xdrneg1) 267 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid); 268 if (sp->sa_size != nfs_xdrneg1) 269 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size); 270 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) { 271 #ifdef notyet 272 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime); 273 #else 274 va.va_atime.tv_sec = 275 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 276 va.va_atime.tv_nsec = 0; 277 #endif 278 } 279 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1) 280 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime); 281 282 } 283 284 /* 285 * Now that we have all the fields, lets do it. 286 */ 287 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly, 288 (nfsd->nd_flag & ND_KERBAUTH), false); 289 if (error) { 290 nfsm_reply(2 * NFSX_UNSIGNED); 291 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 292 vn_finished_write(mp, 0); 293 return (0); 294 } 295 nqsrv_getl(vp, ND_WRITE); 296 if (v3) { 297 error = preat_ret = VOP_GETATTR(vp, &preat, cred, lwp); 298 if (!error && gcheck && 299 (preat.va_ctime.tv_sec != guard.tv_sec || 300 preat.va_ctime.tv_nsec != guard.tv_nsec)) 301 error = NFSERR_NOT_SYNC; 302 if (error) { 303 vput(vp); 304 nfsm_reply(NFSX_WCCDATA(v3)); 305 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 306 vn_finished_write(mp, 0); 307 return (0); 308 } 309 } 310 311 /* 312 * If the size is being changed write acces is required, otherwise 313 * just check for a read only file system. 314 */ 315 if (va.va_size == ((u_quad_t)((quad_t) -1))) { 316 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 317 error = EROFS; 318 goto out; 319 } 320 } else { 321 if (vp->v_type == VDIR) { 322 error = EISDIR; 323 goto out; 324 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly, 325 lwp, 0)) != 0) 326 goto out; 327 } 328 error = VOP_SETATTR(vp, &va, cred, lwp); 329 postat_ret = VOP_GETATTR(vp, &va, cred, lwp); 330 if (!error) 331 error = postat_ret; 332 out: 333 vput(vp); 334 vn_finished_write(mp, 0); 335 nfsm_reply(NFSX_WCCORFATTR(v3)); 336 if (v3) { 337 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va); 338 return (0); 339 } else { 340 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 341 nfsm_srvfillattr(&va, fp); 342 } 343 nfsm_srvdone; 344 } 345 346 /* 347 * nfs lookup rpc 348 */ 349 int 350 nfsrv_lookup(nfsd, slp, lwp, mrq) 351 struct nfsrv_descript *nfsd; 352 struct nfssvc_sock *slp; 353 struct lwp *lwp; 354 struct mbuf **mrq; 355 { 356 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 357 struct mbuf *nam = nfsd->nd_nam; 358 char *dpos = nfsd->nd_dpos; 359 kauth_cred_t cred = nfsd->nd_cr; 360 struct nfs_fattr *fp; 361 struct nameidata nd, ind, *ndp = &nd; 362 struct vnode *vp, *dirp; 363 nfsrvfh_t nsfh; 364 char *cp; 365 u_int32_t *tl; 366 int32_t t1; 367 char *bpos; 368 int error = 0, cache = 0, dirattr_ret = 1; 369 uint32_t len; 370 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag; 371 char *cp2; 372 struct mbuf *mb, *mreq; 373 struct vattr va, dirattr; 374 u_quad_t frev; 375 376 nfsm_srvmtofh(&nsfh); 377 nfsm_srvnamesiz(len); 378 379 pubflag = nfs_ispublicfh(&nsfh); 380 381 nd.ni_cnd.cn_cred = cred; 382 nd.ni_cnd.cn_nameiop = LOOKUP; 383 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART; 384 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 385 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag); 386 387 if (!error && pubflag) { 388 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) { 389 /* 390 * Setup call to lookup() to see if we can find 391 * the index file. Arguably, this doesn't belong 392 * in a kernel.. Ugh. 393 */ 394 ind = nd; 395 VOP_UNLOCK(nd.ni_vp, 0); 396 ind.ni_pathlen = strlen(nfs_pub.np_index); 397 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf = 398 nfs_pub.np_index; 399 ind.ni_startdir = nd.ni_vp; 400 VREF(ind.ni_startdir); 401 error = lookup(&ind); 402 if (!error) { 403 /* 404 * Found an index file. Get rid of 405 * the old references. 406 */ 407 if (dirp) 408 vrele(dirp); 409 dirp = nd.ni_vp; 410 vrele(nd.ni_startdir); 411 ndp = &ind; 412 } else 413 error = 0; 414 } 415 /* 416 * If the public filehandle was used, check that this lookup 417 * didn't result in a filehandle outside the publicly exported 418 * filesystem. 419 */ 420 421 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) { 422 vput(nd.ni_vp); 423 error = EPERM; 424 } 425 } 426 427 if (dirp) { 428 if (v3) 429 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred, lwp); 430 vrele(dirp); 431 } 432 433 if (error) { 434 nfsm_reply(NFSX_POSTOPATTR(v3)); 435 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 436 return (0); 437 } 438 439 nqsrv_getl(ndp->ni_startdir, ND_READ); 440 PNBUF_PUT(nd.ni_cnd.cn_pnbuf); 441 vp = ndp->ni_vp; 442 error = nfsrv_composefh(vp, &nsfh, v3); 443 if (!error) 444 error = VOP_GETATTR(vp, &va, cred, lwp); 445 vput(vp); 446 vrele(ndp->ni_startdir); 447 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) + 448 NFSX_POSTOPATTR(v3)); 449 if (error) { 450 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 451 return (0); 452 } 453 nfsm_srvfhtom(&nsfh, v3); 454 if (v3) { 455 nfsm_srvpostop_attr(0, &va); 456 nfsm_srvpostop_attr(dirattr_ret, &dirattr); 457 } else { 458 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 459 nfsm_srvfillattr(&va, fp); 460 } 461 nfsm_srvdone; 462 } 463 464 /* 465 * nfs readlink service 466 */ 467 int 468 nfsrv_readlink(nfsd, slp, lwp, mrq) 469 struct nfsrv_descript *nfsd; 470 struct nfssvc_sock *slp; 471 struct lwp *lwp; 472 struct mbuf **mrq; 473 { 474 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 475 struct mbuf *nam = nfsd->nd_nam; 476 char *dpos = nfsd->nd_dpos; 477 kauth_cred_t cred = nfsd->nd_cr; 478 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 479 struct iovec *ivp = iv; 480 struct mbuf *mp; 481 u_int32_t *tl; 482 int32_t t1; 483 char *bpos; 484 int error = 0, rdonly, cache = 0, i, padlen, getret; 485 uint32_t len; 486 int v3 = (nfsd->nd_flag & ND_NFSV3); 487 char *cp2; 488 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq; 489 struct vnode *vp; 490 struct vattr attr; 491 nfsrvfh_t nsfh; 492 struct uio io, *uiop = &io; 493 u_quad_t frev; 494 495 nfsm_srvmtofh(&nsfh); 496 len = 0; 497 i = 0; 498 while (len < NFS_MAXPATHLEN) { 499 mp = m_get(M_WAIT, MT_DATA); 500 MCLAIM(mp, &nfs_mowner); 501 m_clget(mp, M_WAIT); 502 mp->m_len = NFSMSIZ(mp); 503 if (len == 0) 504 mp3 = mp2 = mp; 505 else { 506 mp2->m_next = mp; 507 mp2 = mp; 508 } 509 if ((len+mp->m_len) > NFS_MAXPATHLEN) { 510 mp->m_len = NFS_MAXPATHLEN-len; 511 len = NFS_MAXPATHLEN; 512 } else 513 len += mp->m_len; 514 ivp->iov_base = mtod(mp, void *); 515 ivp->iov_len = mp->m_len; 516 i++; 517 ivp++; 518 } 519 uiop->uio_iov = iv; 520 uiop->uio_iovcnt = i; 521 uiop->uio_offset = 0; 522 uiop->uio_resid = len; 523 uiop->uio_rw = UIO_READ; 524 UIO_SETUP_SYSSPACE(uiop); 525 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 526 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 527 if (error) { 528 m_freem(mp3); 529 nfsm_reply(2 * NFSX_UNSIGNED); 530 nfsm_srvpostop_attr(1, (struct vattr *)0); 531 return (0); 532 } 533 if (vp->v_type != VLNK) { 534 if (v3) 535 error = EINVAL; 536 else 537 error = ENXIO; 538 goto out; 539 } 540 nqsrv_getl(vp, ND_READ); 541 error = VOP_READLINK(vp, uiop, cred); 542 out: 543 getret = VOP_GETATTR(vp, &attr, cred, lwp); 544 vput(vp); 545 if (error) 546 m_freem(mp3); 547 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED); 548 if (v3) { 549 nfsm_srvpostop_attr(getret, &attr); 550 if (error) 551 return (0); 552 } 553 len -= uiop->uio_resid; 554 padlen = nfsm_padlen(len); 555 if (uiop->uio_resid || padlen) 556 nfs_zeropad(mp3, uiop->uio_resid, padlen); 557 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 558 *tl = txdr_unsigned(len); 559 mb->m_next = mp3; 560 nfsm_srvdone; 561 } 562 563 /* 564 * nfs read service 565 */ 566 int 567 nfsrv_read(nfsd, slp, lwp, mrq) 568 struct nfsrv_descript *nfsd; 569 struct nfssvc_sock *slp; 570 struct lwp *lwp; 571 struct mbuf **mrq; 572 { 573 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 574 struct mbuf *nam = nfsd->nd_nam; 575 char *dpos = nfsd->nd_dpos; 576 kauth_cred_t cred = nfsd->nd_cr; 577 struct mbuf *m; 578 struct nfs_fattr *fp; 579 u_int32_t *tl; 580 int32_t t1; 581 int i; 582 char *bpos; 583 int error = 0, rdonly, cache = 0, getret; 584 int v3 = (nfsd->nd_flag & ND_NFSV3); 585 uint32_t reqlen, len, cnt, left; 586 int padlen; 587 char *cp2; 588 struct mbuf *mb, *mreq; 589 struct vnode *vp; 590 nfsrvfh_t nsfh; 591 struct uio io, *uiop = &io; 592 struct vattr va; 593 off_t off; 594 u_quad_t frev; 595 596 nfsm_srvmtofh(&nsfh); 597 if (v3) { 598 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 599 off = fxdr_hyper(tl); 600 } else { 601 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 602 off = (off_t)fxdr_unsigned(u_int32_t, *tl); 603 } 604 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 605 reqlen = fxdr_unsigned(uint32_t, *tl); 606 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd)); 607 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 608 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 609 if (error) { 610 nfsm_reply(2 * NFSX_UNSIGNED); 611 nfsm_srvpostop_attr(1, (struct vattr *)0); 612 return (0); 613 } 614 if (vp->v_type != VREG) { 615 if (v3) 616 error = EINVAL; 617 else 618 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 619 } 620 if (!error) { 621 nqsrv_getl(vp, ND_READ); 622 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0) 623 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1); 624 } 625 getret = VOP_GETATTR(vp, &va, cred, lwp); 626 if (!error) 627 error = getret; 628 if (error) { 629 vput(vp); 630 nfsm_reply(NFSX_POSTOPATTR(v3)); 631 nfsm_srvpostop_attr(getret, &va); 632 return (0); 633 } 634 if (off >= va.va_size) 635 cnt = 0; 636 else if ((off + reqlen) > va.va_size) 637 cnt = va.va_size - off; 638 else 639 cnt = reqlen; 640 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt)); 641 if (v3) { 642 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED); 643 *tl++ = nfs_true; 644 fp = (struct nfs_fattr *)tl; 645 tl += (NFSX_V3FATTR / sizeof (u_int32_t)); 646 } else { 647 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED); 648 fp = (struct nfs_fattr *)tl; 649 tl += (NFSX_V2FATTR / sizeof (u_int32_t)); 650 } 651 len = left = cnt; 652 if (cnt > 0) { 653 if (nfsd_use_loan) { 654 struct vm_page **pgpp; 655 voff_t pgoff = trunc_page(off); 656 int npages; 657 vaddr_t lva; 658 659 npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT; 660 KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */ 661 662 /* allocate kva for mbuf data */ 663 lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so); 664 if (lva == 0) { 665 /* fall back to VOP_READ */ 666 goto loan_fail; 667 } 668 669 /* allocate mbuf */ 670 m = m_get(M_WAIT, MT_DATA); 671 MCLAIM(m, &nfs_mowner); 672 pgpp = m->m_ext.ext_pgs; 673 674 /* loan pages */ 675 error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages, 676 pgpp); 677 if (error) { 678 sokvafree(lva, npages << PAGE_SHIFT); 679 m_free(m); 680 if (error == EBUSY) 681 goto loan_fail; 682 goto read_error; 683 } 684 685 /* associate kva to mbuf */ 686 MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)), 687 cnt, M_MBUF, soloanfree, slp->ns_so); 688 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 689 m->m_len = cnt; 690 691 /* map pages */ 692 for (i = 0; i < npages; i++) { 693 pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]), 694 VM_PROT_READ); 695 lva += PAGE_SIZE; 696 } 697 698 pmap_update(pmap_kernel()); 699 700 mb->m_next = m; 701 mb = m; 702 error = 0; 703 uiop->uio_resid = 0; 704 } else { 705 struct iovec *iv; 706 struct iovec *iv2; 707 struct mbuf *m2; 708 int siz; 709 loan_fail: 710 /* 711 * Generate the mbuf list with the uio_iov ref. to it. 712 */ 713 i = 0; 714 m = m2 = mb; 715 while (left > 0) { 716 siz = min(M_TRAILINGSPACE(m), left); 717 if (siz > 0) { 718 left -= siz; 719 i++; 720 } 721 if (left > 0) { 722 m = m_get(M_WAIT, MT_DATA); 723 MCLAIM(m, &nfs_mowner); 724 m_clget(m, M_WAIT); 725 m->m_len = 0; 726 m2->m_next = m; 727 m2 = m; 728 } 729 } 730 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK); 731 uiop->uio_iov = iv2 = iv; 732 m = mb; 733 left = cnt; 734 i = 0; 735 while (left > 0) { 736 if (m == NULL) 737 panic("nfsrv_read iov"); 738 siz = min(M_TRAILINGSPACE(m), left); 739 if (siz > 0) { 740 iv->iov_base = mtod(m, char *) + 741 m->m_len; 742 iv->iov_len = siz; 743 m->m_len += siz; 744 left -= siz; 745 iv++; 746 i++; 747 } 748 m = m->m_next; 749 } 750 uiop->uio_iovcnt = i; 751 uiop->uio_offset = off; 752 uiop->uio_resid = cnt; 753 uiop->uio_rw = UIO_READ; 754 UIO_SETUP_SYSSPACE(uiop); 755 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred); 756 free((void *)iv2, M_TEMP); 757 } 758 read_error: 759 if (error || (getret = VOP_GETATTR(vp, &va, cred, lwp)) != 0){ 760 if (!error) 761 error = getret; 762 m_freem(mreq); 763 vput(vp); 764 nfsm_reply(NFSX_POSTOPATTR(v3)); 765 nfsm_srvpostop_attr(getret, &va); 766 return (0); 767 } 768 } else { 769 uiop->uio_resid = 0; 770 } 771 vput(vp); 772 nfsm_srvfillattr(&va, fp); 773 len -= uiop->uio_resid; 774 padlen = nfsm_padlen(len); 775 if (uiop->uio_resid || padlen) 776 nfs_zeropad(mb, uiop->uio_resid, padlen); 777 if (v3) { 778 /* count */ 779 *tl++ = txdr_unsigned(len); 780 /* eof */ 781 if (off + len >= va.va_size) 782 *tl++ = nfs_true; 783 else 784 *tl++ = nfs_false; 785 } 786 *tl = txdr_unsigned(len); 787 nfsm_srvdone; 788 } 789 790 /* 791 * nfs write service 792 */ 793 int 794 nfsrv_write(nfsd, slp, lwp, mrq) 795 struct nfsrv_descript *nfsd; 796 struct nfssvc_sock *slp; 797 struct lwp *lwp; 798 struct mbuf **mrq; 799 { 800 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 801 struct mbuf *nam = nfsd->nd_nam; 802 char *dpos = nfsd->nd_dpos; 803 kauth_cred_t cred = nfsd->nd_cr; 804 struct iovec *ivp; 805 int i, cnt; 806 struct mbuf *mp; 807 struct nfs_fattr *fp; 808 struct iovec *iv; 809 struct vattr va, forat; 810 u_int32_t *tl; 811 int32_t t1; 812 char *bpos; 813 int error = 0, rdonly, cache = 0, len, forat_ret = 1; 814 int ioflags, aftat_ret = 1, retlen, zeroing, adjust; 815 int stable = NFSV3WRITE_FILESYNC; 816 int v3 = (nfsd->nd_flag & ND_NFSV3); 817 char *cp2; 818 struct mbuf *mb, *mreq; 819 struct vnode *vp; 820 nfsrvfh_t nsfh; 821 struct uio io, *uiop = &io; 822 off_t off; 823 u_quad_t frev; 824 struct mount *mntp = NULL; 825 826 if (mrep == NULL) { 827 *mrq = NULL; 828 return (0); 829 } 830 nfsm_srvmtofh(&nsfh); 831 if ((mntp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 832 return (ESTALE); 833 vn_start_write(NULL, &mntp, V_WAIT); 834 if (v3) { 835 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 836 off = fxdr_hyper(tl); 837 tl += 3; 838 stable = fxdr_unsigned(int, *tl++); 839 } else { 840 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 841 off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 842 tl += 2; 843 } 844 retlen = len = fxdr_unsigned(int32_t, *tl); 845 cnt = i = 0; 846 847 /* 848 * For NFS Version 2, it is not obvious what a write of zero length 849 * should do, but I might as well be consistent with Version 3, 850 * which is to return ok so long as there are no permission problems. 851 */ 852 if (len > 0) { 853 zeroing = 1; 854 mp = mrep; 855 while (mp) { 856 if (mp == md) { 857 zeroing = 0; 858 adjust = dpos - mtod(mp, char *); 859 mp->m_len -= adjust; 860 if (mp->m_len > 0 && adjust > 0) 861 NFSMADV(mp, adjust); 862 } 863 if (zeroing) 864 mp->m_len = 0; 865 else if (mp->m_len > 0) { 866 i += mp->m_len; 867 if (i > len) { 868 mp->m_len -= (i - len); 869 zeroing = 1; 870 } 871 if (mp->m_len > 0) 872 cnt++; 873 } 874 mp = mp->m_next; 875 } 876 } 877 if (len > NFS_MAXDATA || len < 0 || i < len) { 878 error = EIO; 879 nfsm_reply(2 * NFSX_UNSIGNED); 880 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 881 vn_finished_write(mntp, 0); 882 return (0); 883 } 884 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 885 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 886 if (error) { 887 nfsm_reply(2 * NFSX_UNSIGNED); 888 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 889 vn_finished_write(mntp, 0); 890 return (0); 891 } 892 if (v3) 893 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 894 if (vp->v_type != VREG) { 895 if (v3) 896 error = EINVAL; 897 else 898 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 899 } 900 if (!error) { 901 nqsrv_getl(vp, ND_WRITE); 902 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 903 } 904 if (error) { 905 vput(vp); 906 nfsm_reply(NFSX_WCCDATA(v3)); 907 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 908 vn_finished_write(mntp, 0); 909 return (0); 910 } 911 912 if (len > 0) { 913 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); 914 uiop->uio_iov = iv = ivp; 915 uiop->uio_iovcnt = cnt; 916 mp = mrep; 917 while (mp) { 918 if (mp->m_len > 0) { 919 ivp->iov_base = mtod(mp, void *); 920 ivp->iov_len = mp->m_len; 921 ivp++; 922 } 923 mp = mp->m_next; 924 } 925 926 /* 927 * XXX 928 * The IO_METASYNC flag indicates that all metadata (and not 929 * just enough to ensure data integrity) must be written to 930 * stable storage synchronously. 931 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.) 932 */ 933 if (stable == NFSV3WRITE_UNSTABLE) 934 ioflags = IO_NODELOCKED; 935 else if (stable == NFSV3WRITE_DATASYNC) 936 ioflags = (IO_SYNC | IO_NODELOCKED); 937 else 938 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 939 uiop->uio_resid = len; 940 uiop->uio_rw = UIO_WRITE; 941 uiop->uio_offset = off; 942 UIO_SETUP_SYSSPACE(uiop); 943 error = VOP_WRITE(vp, uiop, ioflags, cred); 944 nfsstats.srvvop_writes++; 945 free(iv, M_TEMP); 946 } 947 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 948 vput(vp); 949 vn_finished_write(mntp, 0); 950 if (!error) 951 error = aftat_ret; 952 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) + 953 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3)); 954 if (v3) { 955 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 956 if (error) 957 return (0); 958 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 959 *tl++ = txdr_unsigned(retlen); 960 if (stable == NFSV3WRITE_UNSTABLE) 961 *tl++ = txdr_unsigned(stable); 962 else 963 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC); 964 /* 965 * Actually, there is no need to txdr these fields, 966 * but it may make the values more human readable, 967 * for debugging purposes. 968 */ 969 *tl++ = txdr_unsigned(boottime.tv_sec); 970 *tl = txdr_unsigned(boottime.tv_usec); 971 } else { 972 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 973 nfsm_srvfillattr(&va, fp); 974 } 975 nfsm_srvdone; 976 } 977 978 /* 979 * XXX elad: the original NFSW_SAMECRED() macro also made sure the 980 * two nd_flag fields of the descriptors contained 981 * ND_KERBAUTH. 982 */ 983 static int 984 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2) 985 { 986 int i, do_ngroups; 987 988 if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2)) 989 return (0); 990 if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2)) 991 return (0); 992 do_ngroups = kauth_cred_ngroups(cred1); 993 for (i = 0; i < do_ngroups; i++) 994 if (kauth_cred_group(cred1, i) != 995 kauth_cred_group(cred2, i)) 996 return (0); 997 998 return (1); 999 } 1000 1001 static struct nfsrvw_delayhash * 1002 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh) 1003 { 1004 uint32_t hash; 1005 1006 hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh), 1007 HASH32_BUF_INIT); 1008 return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ]; 1009 } 1010 1011 /* 1012 * NFS write service with write gathering support. Called when 1013 * nfsrvw_procrastinate > 0. 1014 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server", 1015 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco, 1016 * Jan. 1994. 1017 */ 1018 int 1019 nfsrv_writegather(ndp, slp, lwp, mrq) 1020 struct nfsrv_descript **ndp; 1021 struct nfssvc_sock *slp; 1022 struct lwp *lwp; 1023 struct mbuf **mrq; 1024 { 1025 struct timeval now; 1026 struct iovec *ivp; 1027 struct mbuf *mp; 1028 struct nfsrv_descript *wp, *nfsd, *owp, *swp; 1029 struct nfs_fattr *fp; 1030 int i = 0; 1031 struct iovec *iov; 1032 struct nfsrvw_delayhash *wpp; 1033 kauth_cred_t cred; 1034 struct vattr va, forat; 1035 u_int32_t *tl; 1036 int32_t t1; 1037 char *bpos, *dpos; 1038 int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1; 1039 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing; 1040 char *cp2; 1041 struct mbuf *mb, *mreq, *mrep, *md; 1042 struct vnode *vp; 1043 struct uio io, *uiop = &io; 1044 u_quad_t frev, cur_usec; 1045 struct mount *mntp = NULL; 1046 1047 *mrq = NULL; 1048 if (*ndp) { 1049 nfsd = *ndp; 1050 *ndp = NULL; 1051 mrep = nfsd->nd_mrep; 1052 md = nfsd->nd_md; 1053 dpos = nfsd->nd_dpos; 1054 cred = nfsd->nd_cr; 1055 v3 = (nfsd->nd_flag & ND_NFSV3); 1056 LIST_INIT(&nfsd->nd_coalesce); 1057 nfsd->nd_mreq = NULL; 1058 nfsd->nd_stable = NFSV3WRITE_FILESYNC; 1059 getmicrotime(&now); 1060 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1061 nfsd->nd_time = cur_usec + nfsrvw_procrastinate; 1062 1063 /* 1064 * Now, get the write header.. 1065 */ 1066 nfsm_srvmtofh(&nfsd->nd_fh); 1067 if (v3) { 1068 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1069 nfsd->nd_off = fxdr_hyper(tl); 1070 tl += 3; 1071 nfsd->nd_stable = fxdr_unsigned(int, *tl++); 1072 } else { 1073 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1074 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl); 1075 tl += 2; 1076 } 1077 len = fxdr_unsigned(int32_t, *tl); 1078 nfsd->nd_len = len; 1079 nfsd->nd_eoff = nfsd->nd_off + len; 1080 1081 /* 1082 * Trim the header out of the mbuf list and trim off any trailing 1083 * junk so that the mbuf list has only the write data. 1084 */ 1085 zeroing = 1; 1086 i = 0; 1087 mp = mrep; 1088 while (mp) { 1089 if (mp == md) { 1090 zeroing = 0; 1091 adjust = dpos - mtod(mp, char *); 1092 mp->m_len -= adjust; 1093 if (mp->m_len > 0 && adjust > 0) 1094 NFSMADV(mp, adjust); 1095 } 1096 if (zeroing) 1097 mp->m_len = 0; 1098 else { 1099 i += mp->m_len; 1100 if (i > len) { 1101 mp->m_len -= (i - len); 1102 zeroing = 1; 1103 } 1104 } 1105 mp = mp->m_next; 1106 } 1107 if (len > NFS_MAXDATA || len < 0 || i < len) { 1108 nfsmout: 1109 m_freem(mrep); 1110 error = EIO; 1111 nfsm_writereply(2 * NFSX_UNSIGNED, v3); 1112 if (v3) 1113 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1114 nfsd->nd_mreq = mreq; 1115 nfsd->nd_mrep = NULL; 1116 nfsd->nd_time = 0; 1117 } 1118 1119 /* 1120 * Add this entry to the hash and time queues. 1121 */ 1122 s = splsoftclock(); 1123 owp = NULL; 1124 wp = LIST_FIRST(&slp->ns_tq); 1125 while (wp && wp->nd_time < nfsd->nd_time) { 1126 owp = wp; 1127 wp = LIST_NEXT(wp, nd_tq); 1128 } 1129 if (owp) { 1130 LIST_INSERT_AFTER(owp, nfsd, nd_tq); 1131 } else { 1132 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1133 } 1134 if (nfsd->nd_mrep) { 1135 wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh); 1136 owp = NULL; 1137 wp = LIST_FIRST(wpp); 1138 while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1139 owp = wp; 1140 wp = LIST_NEXT(wp, nd_hash); 1141 } 1142 while (wp && wp->nd_off < nfsd->nd_off && 1143 !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) { 1144 owp = wp; 1145 wp = LIST_NEXT(wp, nd_hash); 1146 } 1147 if (owp) { 1148 LIST_INSERT_AFTER(owp, nfsd, nd_hash); 1149 1150 /* 1151 * Search the hash list for overlapping entries and 1152 * coalesce. 1153 */ 1154 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) { 1155 wp = LIST_NEXT(nfsd, nd_hash); 1156 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr)) 1157 nfsrvw_coalesce(owp, nfsd); 1158 } 1159 } else { 1160 LIST_INSERT_HEAD(wpp, nfsd, nd_hash); 1161 } 1162 } 1163 splx(s); 1164 } 1165 1166 /* 1167 * Now, do VOP_WRITE()s for any one(s) that need to be done now 1168 * and generate the associated reply mbuf list(s). 1169 */ 1170 loop1: 1171 getmicrotime(&now); 1172 cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; 1173 s = splsoftclock(); 1174 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) { 1175 owp = LIST_NEXT(nfsd, nd_tq); 1176 if (nfsd->nd_time > cur_usec) 1177 break; 1178 if (nfsd->nd_mreq) 1179 continue; 1180 LIST_REMOVE(nfsd, nd_tq); 1181 LIST_REMOVE(nfsd, nd_hash); 1182 splx(s); 1183 mrep = nfsd->nd_mrep; 1184 nfsd->nd_mrep = NULL; 1185 cred = nfsd->nd_cr; 1186 v3 = (nfsd->nd_flag & ND_NFSV3); 1187 forat_ret = aftat_ret = 1; 1188 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp, 1189 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), 1190 false); 1191 if (!error) { 1192 if (v3) 1193 forat_ret = VOP_GETATTR(vp, &forat, cred, lwp); 1194 if (vp->v_type != VREG) { 1195 if (v3) 1196 error = EINVAL; 1197 else 1198 error = (vp->v_type == VDIR) ? EISDIR : EACCES; 1199 } 1200 } else 1201 vp = NULL; 1202 if (!error) { 1203 nqsrv_getl(vp, ND_WRITE); 1204 error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1); 1205 } 1206 1207 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE) 1208 ioflags = IO_NODELOCKED; 1209 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC) 1210 ioflags = (IO_SYNC | IO_NODELOCKED); 1211 else 1212 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED); 1213 uiop->uio_rw = UIO_WRITE; 1214 uiop->uio_offset = nfsd->nd_off; 1215 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off; 1216 UIO_SETUP_SYSSPACE(uiop); 1217 if (uiop->uio_resid > 0) { 1218 mp = mrep; 1219 i = 0; 1220 while (mp) { 1221 if (mp->m_len > 0) 1222 i++; 1223 mp = mp->m_next; 1224 } 1225 uiop->uio_iovcnt = i; 1226 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1227 uiop->uio_iov = ivp = iov; 1228 mp = mrep; 1229 while (mp) { 1230 if (mp->m_len > 0) { 1231 ivp->iov_base = mtod(mp, void *); 1232 ivp->iov_len = mp->m_len; 1233 ivp++; 1234 } 1235 mp = mp->m_next; 1236 } 1237 if (!error) { 1238 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) { 1239 VOP_UNLOCK(vp, 0); 1240 vn_start_write(NULL, &mntp, V_WAIT); 1241 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1242 } 1243 if (!error) { 1244 error = VOP_WRITE(vp, uiop, ioflags, cred); 1245 nfsstats.srvvop_writes++; 1246 vn_finished_write(mntp, 0); 1247 } 1248 } 1249 free((void *)iov, M_TEMP); 1250 } 1251 m_freem(mrep); 1252 if (vp) { 1253 aftat_ret = VOP_GETATTR(vp, &va, cred, lwp); 1254 vput(vp); 1255 } 1256 1257 /* 1258 * Loop around generating replies for all write rpcs that have 1259 * now been completed. 1260 */ 1261 swp = nfsd; 1262 do { 1263 if (error) { 1264 nfsm_writereply(NFSX_WCCDATA(v3), v3); 1265 if (v3) { 1266 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1267 } 1268 } else { 1269 nfsm_writereply(NFSX_PREOPATTR(v3) + 1270 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED + 1271 NFSX_WRITEVERF(v3), v3); 1272 if (v3) { 1273 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va); 1274 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1275 *tl++ = txdr_unsigned(nfsd->nd_len); 1276 *tl++ = txdr_unsigned(swp->nd_stable); 1277 /* 1278 * Actually, there is no need to txdr these fields, 1279 * but it may make the values more human readable, 1280 * for debugging purposes. 1281 */ 1282 *tl++ = txdr_unsigned(boottime.tv_sec); 1283 *tl = txdr_unsigned(boottime.tv_usec); 1284 } else { 1285 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1286 nfsm_srvfillattr(&va, fp); 1287 } 1288 } 1289 nfsd->nd_mreq = mreq; 1290 if (nfsd->nd_mrep) 1291 panic("nfsrv_write: nd_mrep not free"); 1292 1293 /* 1294 * Done. Put it at the head of the timer queue so that 1295 * the final phase can return the reply. 1296 */ 1297 s = splsoftclock(); 1298 if (nfsd != swp) { 1299 nfsd->nd_time = 0; 1300 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq); 1301 } 1302 nfsd = LIST_FIRST(&swp->nd_coalesce); 1303 if (nfsd) { 1304 LIST_REMOVE(nfsd, nd_tq); 1305 } 1306 splx(s); 1307 } while (nfsd); 1308 s = splsoftclock(); 1309 swp->nd_time = 0; 1310 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq); 1311 splx(s); 1312 goto loop1; 1313 } 1314 splx(s); 1315 1316 /* 1317 * Search for a reply to return. 1318 */ 1319 s = splsoftclock(); 1320 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) { 1321 if (nfsd->nd_mreq) { 1322 LIST_REMOVE(nfsd, nd_tq); 1323 *mrq = nfsd->nd_mreq; 1324 *ndp = nfsd; 1325 break; 1326 } 1327 } 1328 splx(s); 1329 return (0); 1330 } 1331 1332 /* 1333 * Coalesce the write request nfsd into owp. To do this we must: 1334 * - remove nfsd from the queues 1335 * - merge nfsd->nd_mrep into owp->nd_mrep 1336 * - update the nd_eoff and nd_stable for owp 1337 * - put nfsd on owp's nd_coalesce list 1338 * NB: Must be called at splsoftclock(). 1339 */ 1340 void 1341 nfsrvw_coalesce(owp, nfsd) 1342 struct nfsrv_descript *owp; 1343 struct nfsrv_descript *nfsd; 1344 { 1345 int overlap; 1346 struct mbuf *mp; 1347 struct nfsrv_descript *m; 1348 1349 LIST_REMOVE(nfsd, nd_hash); 1350 LIST_REMOVE(nfsd, nd_tq); 1351 if (owp->nd_eoff < nfsd->nd_eoff) { 1352 overlap = owp->nd_eoff - nfsd->nd_off; 1353 if (overlap < 0) 1354 panic("nfsrv_coalesce: bad off"); 1355 if (overlap > 0) 1356 m_adj(nfsd->nd_mrep, overlap); 1357 mp = owp->nd_mrep; 1358 while (mp->m_next) 1359 mp = mp->m_next; 1360 mp->m_next = nfsd->nd_mrep; 1361 owp->nd_eoff = nfsd->nd_eoff; 1362 } else 1363 m_freem(nfsd->nd_mrep); 1364 nfsd->nd_mrep = NULL; 1365 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC) 1366 owp->nd_stable = NFSV3WRITE_FILESYNC; 1367 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC && 1368 owp->nd_stable == NFSV3WRITE_UNSTABLE) 1369 owp->nd_stable = NFSV3WRITE_DATASYNC; 1370 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq); 1371 /* 1372 * nfsd might hold coalesce elements! Move them to owp. 1373 * Otherwise, requests may be lost and clients will be stuck. 1374 */ 1375 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) { 1376 LIST_REMOVE(m, nd_tq); 1377 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq); 1378 } 1379 } 1380 1381 /* 1382 * nfs create service 1383 * now does a truncate to 0 length via. setattr if it already exists 1384 */ 1385 int 1386 nfsrv_create(nfsd, slp, lwp, mrq) 1387 struct nfsrv_descript *nfsd; 1388 struct nfssvc_sock *slp; 1389 struct lwp *lwp; 1390 struct mbuf **mrq; 1391 { 1392 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1393 struct mbuf *nam = nfsd->nd_nam; 1394 char *dpos = nfsd->nd_dpos; 1395 kauth_cred_t cred = nfsd->nd_cr; 1396 struct nfs_fattr *fp; 1397 struct vattr va, dirfor, diraft; 1398 struct nfsv2_sattr *sp; 1399 u_int32_t *tl; 1400 struct nameidata nd; 1401 char *cp; 1402 int32_t t1; 1403 char *bpos; 1404 int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1; 1405 int rdev = 0; 1406 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0; 1407 char *cp2; 1408 struct mbuf *mb, *mreq; 1409 struct vnode *vp = NULL, *dirp = NULL; 1410 nfsrvfh_t nsfh; 1411 u_quad_t frev, tempsize; 1412 u_char cverf[NFSX_V3CREATEVERF]; 1413 struct mount *mp = NULL; 1414 1415 nd.ni_cnd.cn_nameiop = 0; 1416 nfsm_srvmtofh(&nsfh); 1417 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 1418 return (ESTALE); 1419 vn_start_write(NULL, &mp, V_WAIT); 1420 nfsm_srvnamesiz(len); 1421 nd.ni_cnd.cn_cred = cred; 1422 nd.ni_cnd.cn_nameiop = CREATE; 1423 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1424 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1425 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1426 if (dirp && v3) { 1427 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1428 } 1429 if (error) { 1430 nfsm_reply(NFSX_WCCDATA(v3)); 1431 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1432 if (dirp) 1433 vrele(dirp); 1434 vn_finished_write(mp, 0); 1435 return (0); 1436 } 1437 VATTR_NULL(&va); 1438 if (v3) { 1439 va.va_mode = 0; 1440 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1441 how = fxdr_unsigned(int, *tl); 1442 switch (how) { 1443 case NFSV3CREATE_GUARDED: 1444 if (nd.ni_vp) { 1445 error = EEXIST; 1446 break; 1447 } 1448 case NFSV3CREATE_UNCHECKED: 1449 nfsm_srvsattr(&va); 1450 break; 1451 case NFSV3CREATE_EXCLUSIVE: 1452 nfsm_dissect(cp, void *, NFSX_V3CREATEVERF); 1453 memcpy(cverf, cp, NFSX_V3CREATEVERF); 1454 exclusive_flag = 1; 1455 break; 1456 }; 1457 va.va_type = VREG; 1458 } else { 1459 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1460 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode)); 1461 if (va.va_type == VNON) 1462 va.va_type = VREG; 1463 va.va_mode = nfstov_mode(sp->sa_mode); 1464 switch (va.va_type) { 1465 case VREG: 1466 tsize = fxdr_unsigned(int32_t, sp->sa_size); 1467 if (tsize != -1) 1468 va.va_size = (u_quad_t)tsize; 1469 break; 1470 case VCHR: 1471 case VBLK: 1472 case VFIFO: 1473 rdev = fxdr_unsigned(int32_t, sp->sa_size); 1474 break; 1475 default: 1476 break; 1477 }; 1478 } 1479 1480 /* 1481 * Iff doesn't exist, create it 1482 * otherwise just truncate to 0 length 1483 * should I set the mode too ?? 1484 */ 1485 if (nd.ni_vp == NULL) { 1486 if (va.va_type == VREG || va.va_type == VSOCK) { 1487 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1488 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1489 if (!error) { 1490 if (exclusive_flag) { 1491 exclusive_flag = 0; 1492 VATTR_NULL(&va); 1493 /* 1494 * XXX 1495 * assuming NFSX_V3CREATEVERF 1496 * == sizeof(nfstime3) 1497 */ 1498 fxdr_nfsv3time(cverf, &va.va_atime); 1499 error = VOP_SETATTR(nd.ni_vp, &va, cred, 1500 lwp); 1501 } 1502 } 1503 } else if (va.va_type == VCHR || va.va_type == VBLK || 1504 va.va_type == VFIFO) { 1505 if (va.va_type == VCHR && rdev == 0xffffffff) 1506 va.va_type = VFIFO; 1507 if (va.va_type != VFIFO && 1508 (error = kauth_authorize_generic(cred, 1509 KAUTH_GENERIC_ISSUSER, NULL))) { 1510 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1511 vput(nd.ni_dvp); 1512 nfsm_reply(0); 1513 vn_finished_write(mp, 0); 1514 return (error); 1515 } else 1516 va.va_rdev = (dev_t)rdev; 1517 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1518 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, 1519 &va); 1520 if (error) { 1521 nfsm_reply(0); 1522 } 1523 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1524 vput(nd.ni_vp); 1525 vrele(nd.ni_dvp); 1526 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1527 error = EINVAL; 1528 nfsm_reply(0); 1529 } 1530 } else { 1531 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1532 vput(nd.ni_dvp); 1533 error = ENXIO; 1534 } 1535 vp = nd.ni_vp; 1536 } else { 1537 vp = nd.ni_vp; 1538 if (nd.ni_dvp == vp) 1539 vrele(nd.ni_dvp); 1540 else 1541 vput(nd.ni_dvp); 1542 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1543 if (!error && va.va_size != -1) { 1544 error = nfsrv_access(vp, VWRITE, cred, 1545 (nd.ni_cnd.cn_flags & RDONLY), lwp, 0); 1546 if (!error) { 1547 nqsrv_getl(vp, ND_WRITE); 1548 tempsize = va.va_size; 1549 VATTR_NULL(&va); 1550 va.va_size = tempsize; 1551 error = VOP_SETATTR(vp, &va, cred, lwp); 1552 } 1553 } 1554 if (error) 1555 vput(vp); 1556 } 1557 if (!error) { 1558 error = nfsrv_composefh(vp, &nsfh, v3); 1559 if (!error) 1560 error = VOP_GETATTR(vp, &va, cred, lwp); 1561 vput(vp); 1562 } 1563 if (v3) { 1564 if (exclusive_flag && !error) { 1565 /* 1566 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3) 1567 */ 1568 char oldverf[NFSX_V3CREATEVERF]; 1569 1570 txdr_nfsv3time(&va.va_atime, oldverf); 1571 if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF)) 1572 error = EEXIST; 1573 } 1574 if (dirp) { 1575 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1576 } 1577 } 1578 if (dirp) { 1579 vrele(dirp); 1580 } 1581 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3)); 1582 if (v3) { 1583 if (!error) { 1584 nfsm_srvpostop_fh(&nsfh); 1585 nfsm_srvpostop_attr(0, &va); 1586 } 1587 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1588 } else { 1589 nfsm_srvfhtom(&nsfh, v3); 1590 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 1591 nfsm_srvfillattr(&va, fp); 1592 } 1593 vn_finished_write(mp, 0); 1594 return (0); 1595 nfsmout: 1596 if (dirp) 1597 vrele(dirp); 1598 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1599 if (nd.ni_dvp == nd.ni_vp) 1600 vrele(nd.ni_dvp); 1601 else 1602 vput(nd.ni_dvp); 1603 if (nd.ni_vp) 1604 vput(nd.ni_vp); 1605 vn_finished_write(mp, 0); 1606 return (error); 1607 } 1608 1609 /* 1610 * nfs v3 mknod service 1611 */ 1612 int 1613 nfsrv_mknod(nfsd, slp, lwp, mrq) 1614 struct nfsrv_descript *nfsd; 1615 struct nfssvc_sock *slp; 1616 struct lwp *lwp; 1617 struct mbuf **mrq; 1618 { 1619 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1620 struct mbuf *nam = nfsd->nd_nam; 1621 char *dpos = nfsd->nd_dpos; 1622 kauth_cred_t cred = nfsd->nd_cr; 1623 struct vattr va, dirfor, diraft; 1624 u_int32_t *tl; 1625 struct nameidata nd; 1626 int32_t t1; 1627 char *bpos; 1628 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1629 u_int32_t major, minor; 1630 enum vtype vtyp; 1631 char *cp2; 1632 struct mbuf *mb, *mreq; 1633 struct vnode *vp, *dirp = (struct vnode *)0; 1634 nfsrvfh_t nsfh; 1635 u_quad_t frev; 1636 struct mount *mp = NULL; 1637 1638 nd.ni_cnd.cn_nameiop = 0; 1639 nfsm_srvmtofh(&nsfh); 1640 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 1641 return (ESTALE); 1642 vn_start_write(NULL, &mp, V_WAIT); 1643 nfsm_srvnamesiz(len); 1644 nd.ni_cnd.cn_cred = cred; 1645 nd.ni_cnd.cn_nameiop = CREATE; 1646 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1647 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1648 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1649 if (dirp) 1650 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1651 if (error) { 1652 nfsm_reply(NFSX_WCCDATA(1)); 1653 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1654 if (dirp) 1655 vrele(dirp); 1656 vn_finished_write(mp, 0); 1657 return (0); 1658 } 1659 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 1660 vtyp = nfsv3tov_type(*tl); 1661 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1662 error = NFSERR_BADTYPE; 1663 goto abort; 1664 } 1665 VATTR_NULL(&va); 1666 va.va_mode = 0; 1667 nfsm_srvsattr(&va); 1668 if (vtyp == VCHR || vtyp == VBLK) { 1669 dev_t rdev; 1670 1671 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1672 major = fxdr_unsigned(u_int32_t, *tl++); 1673 minor = fxdr_unsigned(u_int32_t, *tl); 1674 rdev = makedev(major, minor); 1675 if (major(rdev) != major || minor(rdev) != minor) { 1676 error = EINVAL; 1677 goto abort; 1678 } 1679 va.va_rdev = rdev; 1680 } 1681 1682 /* 1683 * Iff doesn't exist, create it. 1684 */ 1685 if (nd.ni_vp) { 1686 error = EEXIST; 1687 abort: 1688 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1689 if (nd.ni_dvp == nd.ni_vp) 1690 vrele(nd.ni_dvp); 1691 else 1692 vput(nd.ni_dvp); 1693 if (nd.ni_vp) 1694 vput(nd.ni_vp); 1695 goto out; 1696 } 1697 va.va_type = vtyp; 1698 if (vtyp == VSOCK) { 1699 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1700 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1701 } else { 1702 if (va.va_type != VFIFO && 1703 (error = kauth_authorize_generic(cred, 1704 KAUTH_GENERIC_ISSUSER, NULL))) { 1705 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1706 vput(nd.ni_dvp); 1707 goto out; 1708 } 1709 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1710 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 1711 if (error) 1712 goto out; 1713 if (nd.ni_cnd.cn_flags & ISSYMLINK) { 1714 vput(nd.ni_vp); 1715 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1716 error = EINVAL; 1717 } 1718 } 1719 out: 1720 vp = nd.ni_vp; 1721 if (!error) { 1722 error = nfsrv_composefh(vp, &nsfh, true); 1723 if (!error) 1724 error = VOP_GETATTR(vp, &va, cred, lwp); 1725 vput(vp); 1726 } 1727 if (dirp) { 1728 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1729 vrele(dirp); 1730 } 1731 nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) + 1732 NFSX_WCCDATA(1)); 1733 if (!error) { 1734 nfsm_srvpostop_fh(&nsfh); 1735 nfsm_srvpostop_attr(0, &va); 1736 } 1737 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1738 vn_finished_write(mp, 0); 1739 return (0); 1740 nfsmout: 1741 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1742 if (nd.ni_dvp == nd.ni_vp) 1743 vrele(nd.ni_dvp); 1744 else 1745 vput(nd.ni_dvp); 1746 if (nd.ni_vp) 1747 vput(nd.ni_vp); 1748 if (dirp) 1749 vrele(dirp); 1750 vn_finished_write(mp, 0); 1751 return (error); 1752 } 1753 1754 /* 1755 * nfs remove service 1756 */ 1757 int 1758 nfsrv_remove(nfsd, slp, lwp, mrq) 1759 struct nfsrv_descript *nfsd; 1760 struct nfssvc_sock *slp; 1761 struct lwp *lwp; 1762 struct mbuf **mrq; 1763 { 1764 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1765 struct mbuf *nam = nfsd->nd_nam; 1766 char *dpos = nfsd->nd_dpos; 1767 kauth_cred_t cred = nfsd->nd_cr; 1768 struct nameidata nd; 1769 u_int32_t *tl; 1770 int32_t t1; 1771 char *bpos; 1772 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 1773 int v3 = (nfsd->nd_flag & ND_NFSV3); 1774 char *cp2; 1775 struct mbuf *mb, *mreq; 1776 struct vnode *vp, *dirp; 1777 struct vattr dirfor, diraft; 1778 nfsrvfh_t nsfh; 1779 u_quad_t frev; 1780 struct mount *mp = NULL; 1781 1782 #ifndef nolint 1783 vp = (struct vnode *)0; 1784 #endif 1785 nfsm_srvmtofh(&nsfh); 1786 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 1787 return (ESTALE); 1788 vn_start_write(NULL, &mp, V_WAIT); 1789 nfsm_srvnamesiz(len); 1790 nd.ni_cnd.cn_cred = cred; 1791 nd.ni_cnd.cn_nameiop = DELETE; 1792 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 1793 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 1794 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1795 if (dirp && v3) { 1796 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 1797 } 1798 if (!error) { 1799 vp = nd.ni_vp; 1800 if (vp->v_type == VDIR && 1801 (error = kauth_authorize_generic(cred, 1802 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1803 goto out; 1804 /* 1805 * The root of a mounted filesystem cannot be deleted. 1806 */ 1807 if (vp->v_flag & VROOT) { 1808 error = EBUSY; 1809 goto out; 1810 } 1811 out: 1812 if (!error) { 1813 nqsrv_getl(nd.ni_dvp, ND_WRITE); 1814 nqsrv_getl(vp, ND_WRITE); 1815 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 1816 } else { 1817 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 1818 if (nd.ni_dvp == vp) 1819 vrele(nd.ni_dvp); 1820 else 1821 vput(nd.ni_dvp); 1822 vput(vp); 1823 } 1824 } 1825 if (dirp) { 1826 if (v3) { 1827 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 1828 } 1829 vrele(dirp); 1830 } 1831 nfsm_reply(NFSX_WCCDATA(v3)); 1832 if (v3) { 1833 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 1834 vn_finished_write(mp, 0); 1835 return (0); 1836 } 1837 vn_finished_write(mp, 0); 1838 nfsm_srvdone; 1839 } 1840 1841 /* 1842 * nfs rename service 1843 */ 1844 int 1845 nfsrv_rename(nfsd, slp, lwp, mrq) 1846 struct nfsrv_descript *nfsd; 1847 struct nfssvc_sock *slp; 1848 struct lwp *lwp; 1849 struct mbuf **mrq; 1850 { 1851 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 1852 struct mbuf *nam = nfsd->nd_nam; 1853 char *dpos = nfsd->nd_dpos; 1854 kauth_cred_t cred = nfsd->nd_cr; 1855 u_int32_t *tl; 1856 int32_t t1; 1857 char *bpos; 1858 int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1; 1859 uint32_t len, len2; 1860 int tdirfor_ret = 1, tdiraft_ret = 1; 1861 int v3 = (nfsd->nd_flag & ND_NFSV3); 1862 char *cp2; 1863 struct mbuf *mb, *mreq; 1864 struct nameidata fromnd, tond; 1865 struct vnode *fvp, *tvp, *tdvp; 1866 struct vnode *fdirp = NULL, *tdirp = NULL; 1867 struct vattr fdirfor, fdiraft, tdirfor, tdiraft; 1868 nfsrvfh_t fnsfh, tnsfh; 1869 u_quad_t frev; 1870 uid_t saved_uid; 1871 struct mount *mp = NULL; 1872 1873 #ifndef nolint 1874 fvp = (struct vnode *)0; 1875 #endif 1876 fromnd.ni_cnd.cn_nameiop = 0; 1877 tond.ni_cnd.cn_nameiop = 0; 1878 nfsm_srvmtofh(&fnsfh); 1879 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&fnsfh)->fh_fsid)) == NULL) 1880 return (ESTALE); 1881 vn_start_write(NULL, &mp, V_WAIT); 1882 nfsm_srvnamesiz(len); 1883 /* 1884 * Remember our original uid so that we can reset cr_uid before 1885 * the second nfs_namei() call, in case it is remapped. 1886 */ 1887 saved_uid = kauth_cred_geteuid(cred); 1888 fromnd.ni_cnd.cn_cred = cred; 1889 fromnd.ni_cnd.cn_nameiop = DELETE; 1890 fromnd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART; 1891 error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md, 1892 &dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1893 if (fdirp && v3) { 1894 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp); 1895 } 1896 if (error) { 1897 nfsm_reply(2 * NFSX_WCCDATA(v3)); 1898 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 1899 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 1900 if (fdirp) 1901 vrele(fdirp); 1902 vn_finished_write(mp, 0); 1903 return (0); 1904 } 1905 if (fromnd.ni_dvp != fromnd.ni_vp) { 1906 VOP_UNLOCK(fromnd.ni_dvp, 0); 1907 } 1908 fvp = fromnd.ni_vp; 1909 nfsm_srvmtofh(&tnsfh); 1910 if (v3) { 1911 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1912 len2 = fxdr_unsigned(uint32_t, *tl); 1913 /* len2 will be checked by nfs_namei */ 1914 } 1915 else { 1916 /* NFSv2 */ 1917 nfsm_strsiz(len2, NFS_MAXNAMLEN); 1918 } 1919 kauth_cred_seteuid(cred, saved_uid); 1920 tond.ni_cnd.cn_cred = cred; 1921 tond.ni_cnd.cn_nameiop = RENAME; 1922 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART; 1923 error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md, 1924 &dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 1925 if (tdirp && v3) { 1926 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp); 1927 } 1928 if (error) { 1929 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 1930 vrele(fromnd.ni_dvp); 1931 vrele(fvp); 1932 goto out1; 1933 } 1934 tdvp = tond.ni_dvp; 1935 tvp = tond.ni_vp; 1936 if (tvp != NULL) { 1937 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1938 if (v3) 1939 error = EEXIST; 1940 else 1941 error = EISDIR; 1942 goto out; 1943 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1944 if (v3) 1945 error = EEXIST; 1946 else 1947 error = ENOTDIR; 1948 goto out; 1949 } 1950 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1951 if (v3) 1952 error = EXDEV; 1953 else 1954 error = ENOTEMPTY; 1955 goto out; 1956 } 1957 } 1958 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1959 if (v3) 1960 error = EXDEV; 1961 else 1962 error = ENOTEMPTY; 1963 goto out; 1964 } 1965 if (fvp->v_mount != tdvp->v_mount) { 1966 if (v3) 1967 error = EXDEV; 1968 else 1969 error = ENOTEMPTY; 1970 goto out; 1971 } 1972 if (fvp == tdvp) { 1973 if (v3) 1974 error = EINVAL; 1975 else 1976 error = ENOTEMPTY; 1977 } 1978 /* 1979 * If source is the same as the destination (that is the 1980 * same vnode with the same name in the same directory), 1981 * then there is nothing to do. 1982 */ 1983 if (fvp == tvp && fromnd.ni_dvp == tdvp && 1984 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen && 1985 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr, 1986 fromnd.ni_cnd.cn_namelen)) 1987 error = -1; 1988 out: 1989 if (!error) { 1990 nqsrv_getl(fromnd.ni_dvp, ND_WRITE); 1991 nqsrv_getl(tdvp, ND_WRITE); 1992 if (tvp) { 1993 nqsrv_getl(tvp, ND_WRITE); 1994 } 1995 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, 1996 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); 1997 } else { 1998 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd); 1999 if (tdvp == tvp) 2000 vrele(tdvp); 2001 else 2002 vput(tdvp); 2003 if (tvp) 2004 vput(tvp); 2005 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2006 vrele(fromnd.ni_dvp); 2007 vrele(fvp); 2008 if (error == -1) 2009 error = 0; 2010 } 2011 vrele(tond.ni_startdir); 2012 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2013 out1: 2014 if (fdirp) { 2015 if (v3) { 2016 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp); 2017 } 2018 vrele(fdirp); 2019 } 2020 if (tdirp) { 2021 if (v3) { 2022 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp); 2023 } 2024 vrele(tdirp); 2025 } 2026 vrele(fromnd.ni_startdir); 2027 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2028 nfsm_reply(2 * NFSX_WCCDATA(v3)); 2029 if (v3) { 2030 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft); 2031 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft); 2032 } 2033 vn_finished_write(mp, 0); 2034 return (0); 2035 2036 nfsmout: 2037 if (fdirp) 2038 vrele(fdirp); 2039 #ifdef notdef 2040 if (tdirp) 2041 vrele(tdirp); 2042 #endif 2043 if (tond.ni_cnd.cn_nameiop) { 2044 vrele(tond.ni_startdir); 2045 PNBUF_PUT(tond.ni_cnd.cn_pnbuf); 2046 } 2047 if (fromnd.ni_cnd.cn_nameiop) { 2048 vrele(fromnd.ni_startdir); 2049 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf); 2050 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd); 2051 vrele(fromnd.ni_dvp); 2052 vrele(fvp); 2053 } 2054 return (error); 2055 } 2056 2057 /* 2058 * nfs link service 2059 */ 2060 int 2061 nfsrv_link(nfsd, slp, lwp, mrq) 2062 struct nfsrv_descript *nfsd; 2063 struct nfssvc_sock *slp; 2064 struct lwp *lwp; 2065 struct mbuf **mrq; 2066 { 2067 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2068 struct mbuf *nam = nfsd->nd_nam; 2069 char *dpos = nfsd->nd_dpos; 2070 kauth_cred_t cred = nfsd->nd_cr; 2071 struct nameidata nd; 2072 u_int32_t *tl; 2073 int32_t t1; 2074 char *bpos; 2075 int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2076 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3); 2077 char *cp2; 2078 struct mbuf *mb, *mreq; 2079 struct vnode *vp, *xp, *dirp = (struct vnode *)0; 2080 struct vattr dirfor, diraft, at; 2081 nfsrvfh_t nsfh, dnsfh; 2082 u_quad_t frev; 2083 struct mount *mp = NULL; 2084 2085 nfsm_srvmtofh(&nsfh); 2086 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 2087 return (ESTALE); 2088 vn_start_write(NULL, &mp, V_WAIT); 2089 nfsm_srvmtofh(&dnsfh); 2090 nfsm_srvnamesiz(len); 2091 error = nfsrv_fhtovp(&nsfh, 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 = kauth_authorize_generic(cred, 2101 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 2102 goto out1; 2103 nd.ni_cnd.cn_cred = cred; 2104 nd.ni_cnd.cn_nameiop = CREATE; 2105 nd.ni_cnd.cn_flags = LOCKPARENT; 2106 error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos, 2107 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2108 if (dirp && v3) { 2109 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2110 } 2111 if (error) 2112 goto out1; 2113 xp = nd.ni_vp; 2114 if (xp != NULL) { 2115 error = EEXIST; 2116 goto out; 2117 } 2118 xp = nd.ni_dvp; 2119 if (vp->v_mount != xp->v_mount) 2120 error = EXDEV; 2121 out: 2122 if (!error) { 2123 nqsrv_getl(vp, ND_WRITE); 2124 nqsrv_getl(xp, ND_WRITE); 2125 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); 2126 } else { 2127 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2128 if (nd.ni_dvp == nd.ni_vp) 2129 vrele(nd.ni_dvp); 2130 else 2131 vput(nd.ni_dvp); 2132 if (nd.ni_vp) 2133 vrele(nd.ni_vp); 2134 } 2135 out1: 2136 if (v3) 2137 getret = VOP_GETATTR(vp, &at, cred, lwp); 2138 if (dirp) { 2139 if (v3) { 2140 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2141 } 2142 vrele(dirp); 2143 } 2144 vrele(vp); 2145 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3)); 2146 if (v3) { 2147 nfsm_srvpostop_attr(getret, &at); 2148 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2149 vn_finished_write(mp, 0); 2150 return (0); 2151 } 2152 vn_finished_write(mp, 0); 2153 nfsm_srvdone; 2154 } 2155 2156 /* 2157 * nfs symbolic link service 2158 */ 2159 int 2160 nfsrv_symlink(nfsd, slp, lwp, mrq) 2161 struct nfsrv_descript *nfsd; 2162 struct nfssvc_sock *slp; 2163 struct lwp *lwp; 2164 struct mbuf **mrq; 2165 { 2166 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2167 struct mbuf *nam = nfsd->nd_nam; 2168 char *dpos = nfsd->nd_dpos; 2169 kauth_cred_t cred = nfsd->nd_cr; 2170 struct vattr va, dirfor, diraft; 2171 struct nameidata nd; 2172 u_int32_t *tl; 2173 int32_t t1; 2174 struct nfsv2_sattr *sp; 2175 char *bpos, *pathcp = NULL, *cp2; 2176 struct uio io; 2177 struct iovec iv; 2178 int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1; 2179 uint32_t len, len2; 2180 int v3 = (nfsd->nd_flag & ND_NFSV3); 2181 struct mbuf *mb, *mreq; 2182 struct vnode *dirp = (struct vnode *)0; 2183 nfsrvfh_t nsfh; 2184 u_quad_t frev; 2185 struct mount *mp = NULL; 2186 2187 nd.ni_cnd.cn_nameiop = 0; 2188 nfsm_srvmtofh(&nsfh); 2189 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 2190 return (ESTALE); 2191 vn_start_write(NULL, &mp, V_WAIT); 2192 nfsm_srvnamesiz(len); 2193 nd.ni_cnd.cn_cred = cred; 2194 nd.ni_cnd.cn_nameiop = CREATE; 2195 nd.ni_cnd.cn_flags = LOCKPARENT; 2196 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2197 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2198 if (dirp && v3) { 2199 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2200 } 2201 if (error) 2202 goto out; 2203 VATTR_NULL(&va); 2204 va.va_type = VLNK; 2205 if (v3) { 2206 va.va_mode = 0; 2207 nfsm_srvsattr(&va); 2208 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 2209 len2 = fxdr_unsigned(uint32_t, *tl); 2210 if (len2 > PATH_MAX) { 2211 /* XXX should check _PC_NO_TRUNC */ 2212 error = ENAMETOOLONG; 2213 goto abortop; 2214 } 2215 } 2216 else { 2217 /* NFSv2 */ 2218 nfsm_strsiz(len2, NFS_MAXPATHLEN); 2219 } 2220 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK); 2221 iv.iov_base = pathcp; 2222 iv.iov_len = len2; 2223 io.uio_resid = len2; 2224 io.uio_offset = 0; 2225 io.uio_iov = &iv; 2226 io.uio_iovcnt = 1; 2227 io.uio_rw = UIO_READ; 2228 UIO_SETUP_SYSSPACE(&io); 2229 nfsm_mtouio(&io, len2); 2230 if (!v3) { 2231 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2232 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 2233 } 2234 *(pathcp + len2) = '\0'; 2235 if (nd.ni_vp) { 2236 error = EEXIST; 2237 abortop: 2238 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2239 if (nd.ni_dvp == nd.ni_vp) 2240 vrele(nd.ni_dvp); 2241 else 2242 vput(nd.ni_dvp); 2243 if (nd.ni_vp) 2244 vrele(nd.ni_vp); 2245 goto out; 2246 } 2247 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2248 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp); 2249 if (!error) { 2250 if (v3) { 2251 error = nfsrv_composefh(nd.ni_vp, &nsfh, v3); 2252 if (!error) 2253 error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp); 2254 vput(nd.ni_vp); 2255 } else { 2256 vput(nd.ni_vp); 2257 } 2258 } 2259 out: 2260 if (pathcp) 2261 free(pathcp, M_TEMP); 2262 if (dirp) { 2263 if (v3) { 2264 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2265 } 2266 vrele(dirp); 2267 } 2268 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2269 NFSX_WCCDATA(v3)); 2270 if (v3) { 2271 if (!error) { 2272 nfsm_srvpostop_fh(&nsfh); 2273 nfsm_srvpostop_attr(0, &va); 2274 } 2275 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2276 } 2277 vn_finished_write(mp, 0); 2278 return (0); 2279 nfsmout: 2280 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2281 if (nd.ni_dvp == nd.ni_vp) 2282 vrele(nd.ni_dvp); 2283 else 2284 vput(nd.ni_dvp); 2285 if (nd.ni_vp) 2286 vrele(nd.ni_vp); 2287 if (dirp) 2288 vrele(dirp); 2289 if (pathcp) 2290 free(pathcp, M_TEMP); 2291 vn_finished_write(mp, 0); 2292 return (error); 2293 } 2294 2295 /* 2296 * nfs mkdir service 2297 */ 2298 int 2299 nfsrv_mkdir(nfsd, slp, lwp, mrq) 2300 struct nfsrv_descript *nfsd; 2301 struct nfssvc_sock *slp; 2302 struct lwp *lwp; 2303 struct mbuf **mrq; 2304 { 2305 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2306 struct mbuf *nam = nfsd->nd_nam; 2307 char *dpos = nfsd->nd_dpos; 2308 kauth_cred_t cred = nfsd->nd_cr; 2309 struct vattr va, dirfor, diraft; 2310 struct nfs_fattr *fp; 2311 struct nameidata nd; 2312 char *cp; 2313 u_int32_t *tl; 2314 int32_t t1; 2315 char *bpos; 2316 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2317 int v3 = (nfsd->nd_flag & ND_NFSV3); 2318 char *cp2; 2319 struct mbuf *mb, *mreq; 2320 struct vnode *vp, *dirp = (struct vnode *)0; 2321 nfsrvfh_t nsfh; 2322 u_quad_t frev; 2323 struct mount *mp = NULL; 2324 2325 nfsm_srvmtofh(&nsfh); 2326 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 2327 return (ESTALE); 2328 vn_start_write(NULL, &mp, V_WAIT); 2329 nfsm_srvnamesiz(len); 2330 nd.ni_cnd.cn_cred = cred; 2331 nd.ni_cnd.cn_nameiop = CREATE; 2332 nd.ni_cnd.cn_flags = LOCKPARENT; 2333 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2334 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2335 if (dirp && v3) { 2336 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2337 } 2338 if (error) { 2339 nfsm_reply(NFSX_WCCDATA(v3)); 2340 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2341 if (dirp) 2342 vrele(dirp); 2343 vn_finished_write(mp, 0); 2344 return (0); 2345 } 2346 VATTR_NULL(&va); 2347 if (v3) { 2348 va.va_mode = 0; 2349 nfsm_srvsattr(&va); 2350 } else { 2351 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2352 va.va_mode = nfstov_mode(*tl++); 2353 } 2354 va.va_type = VDIR; 2355 vp = nd.ni_vp; 2356 if (vp != NULL) { 2357 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2358 if (nd.ni_dvp == vp) 2359 vrele(nd.ni_dvp); 2360 else 2361 vput(nd.ni_dvp); 2362 vrele(vp); 2363 error = EEXIST; 2364 goto out; 2365 } 2366 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2367 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va); 2368 if (!error) { 2369 vp = nd.ni_vp; 2370 error = nfsrv_composefh(vp, &nsfh, v3); 2371 if (!error) 2372 error = VOP_GETATTR(vp, &va, cred, lwp); 2373 vput(vp); 2374 } 2375 out: 2376 if (dirp) { 2377 if (v3) { 2378 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2379 } 2380 vrele(dirp); 2381 } 2382 nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) + 2383 NFSX_WCCDATA(v3)); 2384 if (v3) { 2385 if (!error) { 2386 nfsm_srvpostop_fh(&nsfh); 2387 nfsm_srvpostop_attr(0, &va); 2388 } 2389 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2390 } else { 2391 nfsm_srvfhtom(&nsfh, v3); 2392 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR); 2393 nfsm_srvfillattr(&va, fp); 2394 } 2395 vn_finished_write(mp, 0); 2396 return (0); 2397 nfsmout: 2398 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2399 if (nd.ni_dvp == nd.ni_vp) 2400 vrele(nd.ni_dvp); 2401 else 2402 vput(nd.ni_dvp); 2403 if (nd.ni_vp) 2404 vrele(nd.ni_vp); 2405 if (dirp) 2406 vrele(dirp); 2407 vn_finished_write(mp, 0); 2408 return (error); 2409 } 2410 2411 /* 2412 * nfs rmdir service 2413 */ 2414 int 2415 nfsrv_rmdir(nfsd, slp, lwp, mrq) 2416 struct nfsrv_descript *nfsd; 2417 struct nfssvc_sock *slp; 2418 struct lwp *lwp; 2419 struct mbuf **mrq; 2420 { 2421 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2422 struct mbuf *nam = nfsd->nd_nam; 2423 char *dpos = nfsd->nd_dpos; 2424 kauth_cred_t cred = nfsd->nd_cr; 2425 u_int32_t *tl; 2426 int32_t t1; 2427 char *bpos; 2428 int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1; 2429 int v3 = (nfsd->nd_flag & ND_NFSV3); 2430 char *cp2; 2431 struct mbuf *mb, *mreq; 2432 struct vnode *vp, *dirp = (struct vnode *)0; 2433 struct vattr dirfor, diraft; 2434 nfsrvfh_t nsfh; 2435 struct nameidata nd; 2436 u_quad_t frev; 2437 struct mount *mp = NULL; 2438 2439 nfsm_srvmtofh(&nsfh); 2440 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 2441 return (ESTALE); 2442 vn_start_write(NULL, &mp, V_WAIT); 2443 nfsm_srvnamesiz(len); 2444 nd.ni_cnd.cn_cred = cred; 2445 nd.ni_cnd.cn_nameiop = DELETE; 2446 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF; 2447 error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos, 2448 &dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false); 2449 if (dirp && v3) { 2450 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp); 2451 } 2452 if (error) { 2453 nfsm_reply(NFSX_WCCDATA(v3)); 2454 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2455 if (dirp) 2456 vrele(dirp); 2457 vn_finished_write(mp, 0); 2458 return (0); 2459 } 2460 vp = nd.ni_vp; 2461 if (vp->v_type != VDIR) { 2462 error = ENOTDIR; 2463 goto out; 2464 } 2465 /* 2466 * No rmdir "." please. 2467 */ 2468 if (nd.ni_dvp == vp) { 2469 error = EINVAL; 2470 goto out; 2471 } 2472 /* 2473 * The root of a mounted filesystem cannot be deleted. 2474 */ 2475 if (vp->v_flag & VROOT) 2476 error = EBUSY; 2477 out: 2478 if (!error) { 2479 nqsrv_getl(nd.ni_dvp, ND_WRITE); 2480 nqsrv_getl(vp, ND_WRITE); 2481 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); 2482 } else { 2483 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 2484 if (nd.ni_dvp == nd.ni_vp) 2485 vrele(nd.ni_dvp); 2486 else 2487 vput(nd.ni_dvp); 2488 vput(vp); 2489 } 2490 if (dirp) { 2491 if (v3) { 2492 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp); 2493 } 2494 vrele(dirp); 2495 } 2496 nfsm_reply(NFSX_WCCDATA(v3)); 2497 if (v3) { 2498 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft); 2499 vn_finished_write(mp, 0); 2500 return (0); 2501 } 2502 vn_finished_write(mp, 0); 2503 nfsm_srvdone; 2504 } 2505 2506 /* 2507 * nfs readdir service 2508 * - mallocs what it thinks is enough to read 2509 * count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR 2510 * - calls VOP_READDIR() 2511 * - loops around building the reply 2512 * if the output generated exceeds count break out of loop 2513 * The nfsm_clget macro is used here so that the reply will be packed 2514 * tightly in mbuf clusters. 2515 * - it only knows that it has encountered eof when the VOP_READDIR() 2516 * reads nothing 2517 * - as such one readdir rpc will return eof false although you are there 2518 * and then the next will return eof 2519 * - it trims out records with d_fileno == 0 2520 * this doesn't matter for Unix clients, but they might confuse clients 2521 * for other os'. 2522 * - it trims out records with d_type == DT_WHT 2523 * these cannot be seen through NFS (unless we extend the protocol) 2524 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less 2525 * than requested, but this may not apply to all filesystems. For 2526 * example, client NFS does not { although it is never remote mounted 2527 * anyhow } 2528 * The alternate call nfsrv_readdirplus() does lookups as well. 2529 * PS: The NFS protocol spec. does not clarify what the "count" byte 2530 * argument is a count of.. just name strings and file id's or the 2531 * entire reply rpc or ... 2532 * I tried just file name and id sizes and it confused the Sun client, 2533 * so I am using the full rpc size now. The "paranoia.." comment refers 2534 * to including the status longwords that are not a part of the dir. 2535 * "entry" structures, but are in the rpc. 2536 */ 2537 2538 #define NFS_SRVDIRBLKSIZ 1024 2539 2540 struct flrep { 2541 nfsuint64 fl_off; 2542 u_int32_t fl_postopok; 2543 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)]; 2544 u_int32_t fl_fhok; 2545 u_int32_t fl_fhsize; 2546 }; 2547 2548 int 2549 nfsrv_readdir(nfsd, slp, lwp, mrq) 2550 struct nfsrv_descript *nfsd; 2551 struct nfssvc_sock *slp; 2552 struct lwp *lwp; 2553 struct mbuf **mrq; 2554 { 2555 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2556 struct mbuf *nam = nfsd->nd_nam; 2557 char *dpos = nfsd->nd_dpos; 2558 kauth_cred_t cred = nfsd->nd_cr; 2559 char *bp, *be; 2560 struct mbuf *mp; 2561 struct dirent *dp; 2562 char *cp; 2563 u_int32_t *tl; 2564 int32_t t1; 2565 char *bpos; 2566 struct mbuf *mb, *mreq, *mp2; 2567 char *cpos, *cend, *cp2, *rbuf; 2568 struct vnode *vp; 2569 struct vattr at; 2570 nfsrvfh_t nsfh; 2571 struct uio io; 2572 struct iovec iv; 2573 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2574 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies; 2575 int v3 = (nfsd->nd_flag & ND_NFSV3); 2576 u_quad_t frev, off, toff, verf; 2577 off_t *cookies = NULL, *cookiep; 2578 nfsuint64 jar; 2579 2580 nfsm_srvmtofh(&nsfh); 2581 if (v3) { 2582 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2583 toff = fxdr_hyper(tl); 2584 tl += 2; 2585 verf = fxdr_hyper(tl); 2586 tl += 2; 2587 } else { 2588 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2589 toff = fxdr_unsigned(u_quad_t, *tl++); 2590 } 2591 off = toff; 2592 cnt = fxdr_unsigned(int, *tl); 2593 siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2594 xfer = NFS_SRVMAXDATA(nfsd); 2595 if (siz > xfer) 2596 siz = xfer; 2597 fullsiz = siz; 2598 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2599 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2600 if (!error && vp->v_type != VDIR) { 2601 error = ENOTDIR; 2602 vput(vp); 2603 } 2604 if (error) { 2605 nfsm_reply(NFSX_UNSIGNED); 2606 nfsm_srvpostop_attr(getret, &at); 2607 return (0); 2608 } 2609 nqsrv_getl(vp, ND_READ); 2610 if (v3) { 2611 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2612 #ifdef NFS3_STRICTVERF 2613 /* 2614 * XXX This check is too strict for Solaris 2.5 clients. 2615 */ 2616 if (!error && toff && verf != at.va_filerev) 2617 error = NFSERR_BAD_COOKIE; 2618 #endif 2619 } 2620 if (!error) 2621 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2622 if (error) { 2623 vput(vp); 2624 nfsm_reply(NFSX_POSTOPATTR(v3)); 2625 nfsm_srvpostop_attr(getret, &at); 2626 return (0); 2627 } 2628 VOP_UNLOCK(vp, 0); 2629 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2630 again: 2631 iv.iov_base = rbuf; 2632 iv.iov_len = fullsiz; 2633 io.uio_iov = &iv; 2634 io.uio_iovcnt = 1; 2635 io.uio_offset = (off_t)off; 2636 io.uio_resid = fullsiz; 2637 io.uio_rw = UIO_READ; 2638 UIO_SETUP_SYSSPACE(&io); 2639 eofflag = 0; 2640 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2641 2642 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2643 2644 off = (off_t)io.uio_offset; 2645 if (!cookies && !error) 2646 error = NFSERR_PERM; 2647 if (v3) { 2648 getret = VOP_GETATTR(vp, &at, cred, lwp); 2649 if (!error) 2650 error = getret; 2651 } 2652 2653 VOP_UNLOCK(vp, 0); 2654 if (error) { 2655 vrele(vp); 2656 free((void *)rbuf, M_TEMP); 2657 if (cookies) 2658 free((void *)cookies, M_TEMP); 2659 nfsm_reply(NFSX_POSTOPATTR(v3)); 2660 nfsm_srvpostop_attr(getret, &at); 2661 return (0); 2662 } 2663 if (io.uio_resid) { 2664 siz -= io.uio_resid; 2665 2666 /* 2667 * If nothing read, return eof 2668 * rpc reply 2669 */ 2670 if (siz == 0) { 2671 vrele(vp); 2672 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + 2673 2 * NFSX_UNSIGNED); 2674 if (v3) { 2675 nfsm_srvpostop_attr(getret, &at); 2676 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2677 txdr_hyper(at.va_filerev, tl); 2678 tl += 2; 2679 } else 2680 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2681 *tl++ = nfs_false; 2682 *tl = nfs_true; 2683 free((void *)rbuf, M_TEMP); 2684 free((void *)cookies, M_TEMP); 2685 return (0); 2686 } 2687 } 2688 2689 /* 2690 * Check for degenerate cases of nothing useful read. 2691 * If so go try again 2692 */ 2693 cpos = rbuf; 2694 cend = rbuf + siz; 2695 dp = (struct dirent *)cpos; 2696 cookiep = cookies; 2697 2698 while (cpos < cend && ncookies > 0 && 2699 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2700 cpos += dp->d_reclen; 2701 dp = (struct dirent *)cpos; 2702 cookiep++; 2703 ncookies--; 2704 } 2705 if (cpos >= cend || ncookies == 0) { 2706 toff = off; 2707 siz = fullsiz; 2708 free(cookies, M_TEMP); 2709 cookies = NULL; 2710 goto again; 2711 } 2712 2713 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */ 2714 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz); 2715 if (v3) { 2716 nfsm_srvpostop_attr(getret, &at); 2717 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2718 txdr_hyper(at.va_filerev, tl); 2719 } 2720 mp = mp2 = mb; 2721 bp = bpos; 2722 be = bp + M_TRAILINGSPACE(mp); 2723 2724 /* Loop through the records and build reply */ 2725 while (cpos < cend && ncookies > 0) { 2726 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2727 nlen = dp->d_namlen; 2728 rem = nfsm_rndup(nlen)-nlen; 2729 len += (4 * NFSX_UNSIGNED + nlen + rem); 2730 if (v3) 2731 len += 2 * NFSX_UNSIGNED; 2732 if (len > cnt) { 2733 eofflag = 0; 2734 break; 2735 } 2736 /* 2737 * Build the directory record xdr from 2738 * the dirent entry. 2739 */ 2740 nfsm_clget; 2741 *tl = nfs_true; 2742 bp += NFSX_UNSIGNED; 2743 if (v3) { 2744 nfsm_clget; 2745 *tl = txdr_unsigned(dp->d_fileno >> 32); 2746 bp += NFSX_UNSIGNED; 2747 } 2748 nfsm_clget; 2749 *tl = txdr_unsigned(dp->d_fileno); 2750 bp += NFSX_UNSIGNED; 2751 nfsm_clget; 2752 *tl = txdr_unsigned(nlen); 2753 bp += NFSX_UNSIGNED; 2754 2755 /* And loop around copying the name */ 2756 xfer = nlen; 2757 cp = dp->d_name; 2758 while (xfer > 0) { 2759 nfsm_clget; 2760 if ((bp+xfer) > be) 2761 tsiz = be-bp; 2762 else 2763 tsiz = xfer; 2764 memcpy(bp, cp, tsiz); 2765 bp += tsiz; 2766 xfer -= tsiz; 2767 if (xfer > 0) 2768 cp += tsiz; 2769 } 2770 /* And null pad to an int32_t boundary */ 2771 for (i = 0; i < rem; i++) 2772 *bp++ = '\0'; 2773 nfsm_clget; 2774 2775 /* Finish off the record */ 2776 txdr_hyper(*cookiep, &jar); 2777 if (v3) { 2778 *tl = jar.nfsuquad[0]; 2779 bp += NFSX_UNSIGNED; 2780 nfsm_clget; 2781 } 2782 *tl = jar.nfsuquad[1]; 2783 bp += NFSX_UNSIGNED; 2784 } 2785 cpos += dp->d_reclen; 2786 dp = (struct dirent *)cpos; 2787 cookiep++; 2788 ncookies--; 2789 } 2790 vrele(vp); 2791 nfsm_clget; 2792 *tl = nfs_false; 2793 bp += NFSX_UNSIGNED; 2794 nfsm_clget; 2795 if (eofflag) 2796 *tl = nfs_true; 2797 else 2798 *tl = nfs_false; 2799 bp += NFSX_UNSIGNED; 2800 if (mp != mb) { 2801 if (bp < be) 2802 mp->m_len = bp - mtod(mp, char *); 2803 } else 2804 mp->m_len += bp - bpos; 2805 free((void *)rbuf, M_TEMP); 2806 free((void *)cookies, M_TEMP); 2807 nfsm_srvdone; 2808 } 2809 2810 int 2811 nfsrv_readdirplus(nfsd, slp, lwp, mrq) 2812 struct nfsrv_descript *nfsd; 2813 struct nfssvc_sock *slp; 2814 struct lwp *lwp; 2815 struct mbuf **mrq; 2816 { 2817 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 2818 struct mbuf *nam = nfsd->nd_nam; 2819 char *dpos = nfsd->nd_dpos; 2820 kauth_cred_t cred = nfsd->nd_cr; 2821 char *bp, *be; 2822 struct mbuf *mp; 2823 struct dirent *dp; 2824 char *cp; 2825 u_int32_t *tl; 2826 int32_t t1; 2827 char *bpos; 2828 struct mbuf *mb, *mreq, *mp2; 2829 char *cpos, *cend, *cp2, *rbuf; 2830 struct vnode *vp, *nvp; 2831 struct flrep fl; 2832 nfsrvfh_t nsfh; 2833 struct uio io; 2834 struct iovec iv; 2835 struct vattr va, at, *vap = &va; 2836 struct nfs_fattr *fp; 2837 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1; 2838 int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies; 2839 u_quad_t frev, off, toff, verf; 2840 off_t *cookies = NULL, *cookiep; 2841 2842 nfsm_srvmtofh(&nsfh); 2843 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2844 toff = fxdr_hyper(tl); 2845 tl += 2; 2846 verf = fxdr_hyper(tl); 2847 tl += 2; 2848 siz = fxdr_unsigned(int, *tl++); 2849 cnt = fxdr_unsigned(int, *tl); 2850 off = toff; 2851 siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1)); 2852 xfer = NFS_SRVMAXDATA(nfsd); 2853 if (siz > xfer) 2854 siz = xfer; 2855 fullsiz = siz; 2856 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 2857 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 2858 if (!error && vp->v_type != VDIR) { 2859 error = ENOTDIR; 2860 vput(vp); 2861 } 2862 if (error) { 2863 nfsm_reply(NFSX_UNSIGNED); 2864 nfsm_srvpostop_attr(getret, &at); 2865 return (0); 2866 } 2867 error = getret = VOP_GETATTR(vp, &at, cred, lwp); 2868 #ifdef NFS3_STRICTVERF 2869 /* 2870 * XXX This check is too strict for Solaris 2.5 clients. 2871 */ 2872 if (!error && toff && verf != at.va_filerev) 2873 error = NFSERR_BAD_COOKIE; 2874 #endif 2875 if (!error) { 2876 nqsrv_getl(vp, ND_READ); 2877 error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0); 2878 } 2879 if (error) { 2880 vput(vp); 2881 nfsm_reply(NFSX_V3POSTOPATTR); 2882 nfsm_srvpostop_attr(getret, &at); 2883 return (0); 2884 } 2885 VOP_UNLOCK(vp, 0); 2886 2887 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2888 again: 2889 iv.iov_base = rbuf; 2890 iv.iov_len = fullsiz; 2891 io.uio_iov = &iv; 2892 io.uio_iovcnt = 1; 2893 io.uio_offset = (off_t)off; 2894 io.uio_resid = fullsiz; 2895 io.uio_rw = UIO_READ; 2896 UIO_SETUP_SYSSPACE(&io); 2897 eofflag = 0; 2898 2899 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2900 2901 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies); 2902 2903 off = (u_quad_t)io.uio_offset; 2904 getret = VOP_GETATTR(vp, &at, cred, lwp); 2905 2906 VOP_UNLOCK(vp, 0); 2907 2908 /* 2909 * If the VGET operation doesn't work for this filesystem, 2910 * we can't support readdirplus. Returning NOTSUPP should 2911 * make clients fall back to plain readdir. 2912 * There's no need to check for VPTOFH as well, we wouldn't 2913 * even be here otherwise. 2914 */ 2915 if (!getret) { 2916 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp))) 2917 getret = (getret == EOPNOTSUPP) ? 2918 NFSERR_NOTSUPP : NFSERR_IO; 2919 else 2920 vput(nvp); 2921 } 2922 2923 if (!cookies && !error) 2924 error = NFSERR_PERM; 2925 if (!error) 2926 error = getret; 2927 if (error) { 2928 vrele(vp); 2929 if (cookies) 2930 free((void *)cookies, M_TEMP); 2931 free((void *)rbuf, M_TEMP); 2932 nfsm_reply(NFSX_V3POSTOPATTR); 2933 nfsm_srvpostop_attr(getret, &at); 2934 return (0); 2935 } 2936 if (io.uio_resid) { 2937 siz -= io.uio_resid; 2938 2939 /* 2940 * If nothing read, return eof 2941 * rpc reply 2942 */ 2943 if (siz == 0) { 2944 vrele(vp); 2945 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2946 2 * NFSX_UNSIGNED); 2947 nfsm_srvpostop_attr(getret, &at); 2948 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2949 txdr_hyper(at.va_filerev, tl); 2950 tl += 2; 2951 *tl++ = nfs_false; 2952 *tl = nfs_true; 2953 free((void *)cookies, M_TEMP); 2954 free((void *)rbuf, M_TEMP); 2955 return (0); 2956 } 2957 } 2958 2959 /* 2960 * Check for degenerate cases of nothing useful read. 2961 * If so go try again 2962 */ 2963 cpos = rbuf; 2964 cend = rbuf + siz; 2965 dp = (struct dirent *)cpos; 2966 cookiep = cookies; 2967 2968 while (cpos < cend && ncookies > 0 && 2969 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) { 2970 cpos += dp->d_reclen; 2971 dp = (struct dirent *)cpos; 2972 cookiep++; 2973 ncookies--; 2974 } 2975 if (cpos >= cend || ncookies == 0) { 2976 toff = off; 2977 siz = fullsiz; 2978 free(cookies, M_TEMP); 2979 cookies = NULL; 2980 goto again; 2981 } 2982 2983 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED; 2984 nfsm_reply(cnt); 2985 nfsm_srvpostop_attr(getret, &at); 2986 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2987 txdr_hyper(at.va_filerev, tl); 2988 mp = mp2 = mb; 2989 bp = bpos; 2990 be = bp + M_TRAILINGSPACE(mp); 2991 2992 /* Loop through the records and build reply */ 2993 while (cpos < cend && ncookies > 0) { 2994 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) { 2995 nfsrvfh_t nnsfh; 2996 2997 nlen = dp->d_namlen; 2998 rem = nfsm_rndup(nlen)-nlen; 2999 3000 /* 3001 * For readdir_and_lookup get the vnode using 3002 * the file number. 3003 */ 3004 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp)) 3005 goto invalid; 3006 if (nfsrv_composefh(nvp, &nnsfh, true)) { 3007 vput(nvp); 3008 goto invalid; 3009 } 3010 if (VOP_GETATTR(nvp, vap, cred, lwp)) { 3011 vput(nvp); 3012 goto invalid; 3013 } 3014 vput(nvp); 3015 3016 /* 3017 * If either the dircount or maxcount will be 3018 * exceeded, get out now. Both of these lengths 3019 * are calculated conservatively, including all 3020 * XDR overheads. 3021 */ 3022 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH + 3023 NFSX_V3POSTOPATTR); 3024 dirlen += (6 * NFSX_UNSIGNED + nlen + rem); 3025 if (len > cnt || dirlen > fullsiz) { 3026 eofflag = 0; 3027 break; 3028 } 3029 3030 /* 3031 * Build the directory record xdr from 3032 * the dirent entry. 3033 */ 3034 fp = (struct nfs_fattr *)&fl.fl_fattr; 3035 nfsm_srvfillattr(vap, fp); 3036 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH); 3037 fl.fl_fhok = nfs_true; 3038 fl.fl_postopok = nfs_true; 3039 txdr_hyper(*cookiep, fl.fl_off.nfsuquad); 3040 3041 nfsm_clget; 3042 *tl = nfs_true; 3043 bp += NFSX_UNSIGNED; 3044 nfsm_clget; 3045 *tl = txdr_unsigned(dp->d_fileno >> 32); 3046 bp += NFSX_UNSIGNED; 3047 nfsm_clget; 3048 *tl = txdr_unsigned(dp->d_fileno); 3049 bp += NFSX_UNSIGNED; 3050 nfsm_clget; 3051 *tl = txdr_unsigned(nlen); 3052 bp += NFSX_UNSIGNED; 3053 3054 /* And loop around copying the name */ 3055 xfer = nlen; 3056 cp = dp->d_name; 3057 while (xfer > 0) { 3058 nfsm_clget; 3059 if ((bp + xfer) > be) 3060 tsiz = be - bp; 3061 else 3062 tsiz = xfer; 3063 memcpy(bp, cp, tsiz); 3064 bp += tsiz; 3065 xfer -= tsiz; 3066 if (xfer > 0) 3067 cp += tsiz; 3068 } 3069 /* And null pad to an int32_t boundary */ 3070 for (i = 0; i < rem; i++) 3071 *bp++ = '\0'; 3072 3073 /* 3074 * Now copy the flrep structure out. 3075 */ 3076 xfer = sizeof(struct flrep); 3077 cp = (void *)&fl; 3078 while (xfer > 0) { 3079 nfsm_clget; 3080 if ((bp + xfer) > be) 3081 tsiz = be - bp; 3082 else 3083 tsiz = xfer; 3084 memcpy(bp, cp, tsiz); 3085 bp += tsiz; 3086 xfer -= tsiz; 3087 if (xfer > 0) 3088 cp += tsiz; 3089 } 3090 3091 /* 3092 * ... and filehandle. 3093 */ 3094 xfer = NFSRVFH_SIZE(&nnsfh); 3095 cp = NFSRVFH_DATA(&nnsfh); 3096 while (xfer > 0) { 3097 nfsm_clget; 3098 if ((bp + xfer) > be) 3099 tsiz = be - bp; 3100 else 3101 tsiz = xfer; 3102 memcpy(bp, cp, tsiz); 3103 bp += tsiz; 3104 xfer -= tsiz; 3105 if (xfer > 0) 3106 cp += tsiz; 3107 } 3108 } 3109 invalid: 3110 cpos += dp->d_reclen; 3111 dp = (struct dirent *)cpos; 3112 cookiep++; 3113 ncookies--; 3114 } 3115 vrele(vp); 3116 nfsm_clget; 3117 *tl = nfs_false; 3118 bp += NFSX_UNSIGNED; 3119 nfsm_clget; 3120 if (eofflag) 3121 *tl = nfs_true; 3122 else 3123 *tl = nfs_false; 3124 bp += NFSX_UNSIGNED; 3125 if (mp != mb) { 3126 if (bp < be) 3127 mp->m_len = bp - mtod(mp, char *); 3128 } else 3129 mp->m_len += bp - bpos; 3130 free((void *)cookies, M_TEMP); 3131 free((void *)rbuf, M_TEMP); 3132 nfsm_srvdone; 3133 } 3134 3135 /* 3136 * nfs commit service 3137 */ 3138 int 3139 nfsrv_commit(nfsd, slp, lwp, mrq) 3140 struct nfsrv_descript *nfsd; 3141 struct nfssvc_sock *slp; 3142 struct lwp *lwp; 3143 struct mbuf **mrq; 3144 { 3145 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3146 struct mbuf *nam = nfsd->nd_nam; 3147 char *dpos = nfsd->nd_dpos; 3148 kauth_cred_t cred = nfsd->nd_cr; 3149 struct vattr bfor, aft; 3150 struct vnode *vp; 3151 nfsrvfh_t nsfh; 3152 u_int32_t *tl; 3153 int32_t t1; 3154 char *bpos; 3155 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0; 3156 uint32_t cnt; 3157 char *cp2; 3158 struct mbuf *mb, *mreq; 3159 u_quad_t frev, off, end; 3160 struct mount *mp = NULL; 3161 3162 nfsm_srvmtofh(&nsfh); 3163 if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL) 3164 return (ESTALE); 3165 vn_start_write(NULL, &mp, V_WAIT); 3166 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3167 3168 off = fxdr_hyper(tl); 3169 tl += 2; 3170 cnt = fxdr_unsigned(uint32_t, *tl); 3171 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3172 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3173 if (error) { 3174 nfsm_reply(2 * NFSX_UNSIGNED); 3175 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3176 vn_finished_write(mp, 0); 3177 return (0); 3178 } 3179 for_ret = VOP_GETATTR(vp, &bfor, cred, lwp); 3180 end = (cnt > 0) ? off + cnt : vp->v_size; 3181 if (end < off || end > vp->v_size) 3182 end = vp->v_size; 3183 if (off < vp->v_size) 3184 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp); 3185 /* else error == 0, from nfsrv_fhtovp() */ 3186 aft_ret = VOP_GETATTR(vp, &aft, cred, lwp); 3187 vput(vp); 3188 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF); 3189 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft); 3190 if (!error) { 3191 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF); 3192 *tl++ = txdr_unsigned(boottime.tv_sec); 3193 *tl = txdr_unsigned(boottime.tv_usec); 3194 } else { 3195 vn_finished_write(mp, 0); 3196 return (0); 3197 } 3198 vn_finished_write(mp, 0); 3199 nfsm_srvdone; 3200 } 3201 3202 /* 3203 * nfs statfs service 3204 */ 3205 int 3206 nfsrv_statfs(nfsd, slp, lwp, mrq) 3207 struct nfsrv_descript *nfsd; 3208 struct nfssvc_sock *slp; 3209 struct lwp *lwp; 3210 struct mbuf **mrq; 3211 { 3212 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3213 struct mbuf *nam = nfsd->nd_nam; 3214 char *dpos = nfsd->nd_dpos; 3215 kauth_cred_t cred = nfsd->nd_cr; 3216 struct statvfs *sf = NULL; 3217 struct nfs_statfs *sfp; 3218 u_int32_t *tl; 3219 int32_t t1; 3220 char *bpos; 3221 int error = 0, rdonly, cache = 0, getret = 1; 3222 int v3 = (nfsd->nd_flag & ND_NFSV3); 3223 char *cp2; 3224 struct mbuf *mb, *mreq; 3225 struct vnode *vp; 3226 struct vattr at; 3227 nfsrvfh_t nsfh; 3228 u_quad_t frev, tval; 3229 3230 nfsm_srvmtofh(&nsfh); 3231 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3232 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3233 if (error) { 3234 nfsm_reply(NFSX_UNSIGNED); 3235 nfsm_srvpostop_attr(getret, &at); 3236 return (0); 3237 } 3238 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK); 3239 error = VFS_STATVFS(vp->v_mount, sf, lwp); 3240 getret = VOP_GETATTR(vp, &at, cred, lwp); 3241 vput(vp); 3242 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3)); 3243 if (v3) 3244 nfsm_srvpostop_attr(getret, &at); 3245 if (error) { 3246 free(sf, M_TEMP); 3247 return (0); 3248 } 3249 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 3250 if (v3) { 3251 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize); 3252 txdr_hyper(tval, &sfp->sf_tbytes); 3253 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize); 3254 txdr_hyper(tval, &sfp->sf_fbytes); 3255 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize); 3256 txdr_hyper(tval, &sfp->sf_abytes); 3257 tval = (u_quad_t)sf->f_files; 3258 txdr_hyper(tval, &sfp->sf_tfiles); 3259 tval = (u_quad_t)sf->f_ffree; 3260 txdr_hyper(tval, &sfp->sf_ffiles); 3261 txdr_hyper(tval, &sfp->sf_afiles); 3262 sfp->sf_invarsec = 0; 3263 } else { 3264 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA); 3265 sfp->sf_bsize = txdr_unsigned(sf->f_frsize); 3266 sfp->sf_blocks = txdr_unsigned(sf->f_blocks); 3267 sfp->sf_bfree = txdr_unsigned(sf->f_bfree); 3268 sfp->sf_bavail = txdr_unsigned(sf->f_bavail); 3269 } 3270 nfsmout: 3271 if (sf) 3272 free(sf, M_TEMP); 3273 return error; 3274 } 3275 3276 /* 3277 * nfs fsinfo service 3278 */ 3279 int 3280 nfsrv_fsinfo(nfsd, slp, lwp, mrq) 3281 struct nfsrv_descript *nfsd; 3282 struct nfssvc_sock *slp; 3283 struct lwp *lwp; 3284 struct mbuf **mrq; 3285 { 3286 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3287 struct mbuf *nam = nfsd->nd_nam; 3288 char *dpos = nfsd->nd_dpos; 3289 kauth_cred_t cred = nfsd->nd_cr; 3290 u_int32_t *tl; 3291 struct nfsv3_fsinfo *sip; 3292 int32_t t1; 3293 char *bpos; 3294 int error = 0, rdonly, cache = 0, getret = 1; 3295 uint32_t maxdata; 3296 char *cp2; 3297 struct mbuf *mb, *mreq; 3298 struct vnode *vp; 3299 struct vattr at; 3300 nfsrvfh_t nsfh; 3301 u_quad_t frev, maxfsize; 3302 struct statvfs *sb; 3303 3304 nfsm_srvmtofh(&nsfh); 3305 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3306 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3307 if (error) { 3308 nfsm_reply(NFSX_UNSIGNED); 3309 nfsm_srvpostop_attr(getret, &at); 3310 return (0); 3311 } 3312 3313 /* XXX Try to make a guess on the max file size. */ 3314 sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK); 3315 VFS_STATVFS(vp->v_mount, sb, (struct lwp *)0); 3316 maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1; 3317 free(sb, M_TEMP); 3318 3319 getret = VOP_GETATTR(vp, &at, cred, lwp); 3320 vput(vp); 3321 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO); 3322 nfsm_srvpostop_attr(getret, &at); 3323 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 3324 3325 /* 3326 * XXX 3327 * There should be file system VFS OP(s) to get this information. 3328 * For now, assume ufs. 3329 */ 3330 if (slp->ns_so->so_type == SOCK_DGRAM) 3331 maxdata = NFS_MAXDGRAMDATA; 3332 else 3333 maxdata = NFS_MAXDATA; 3334 sip->fs_rtmax = txdr_unsigned(maxdata); 3335 sip->fs_rtpref = txdr_unsigned(maxdata); 3336 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE); 3337 sip->fs_wtmax = txdr_unsigned(maxdata); 3338 sip->fs_wtpref = txdr_unsigned(maxdata); 3339 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE); 3340 sip->fs_dtpref = txdr_unsigned(maxdata); 3341 txdr_hyper(maxfsize, &sip->fs_maxfilesize); 3342 sip->fs_timedelta.nfsv3_sec = 0; 3343 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1); 3344 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK | 3345 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | 3346 NFSV3FSINFO_CANSETTIME); 3347 nfsm_srvdone; 3348 } 3349 3350 /* 3351 * nfs pathconf service 3352 */ 3353 int 3354 nfsrv_pathconf(nfsd, slp, lwp, mrq) 3355 struct nfsrv_descript *nfsd; 3356 struct nfssvc_sock *slp; 3357 struct lwp *lwp; 3358 struct mbuf **mrq; 3359 { 3360 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md; 3361 struct mbuf *nam = nfsd->nd_nam; 3362 char *dpos = nfsd->nd_dpos; 3363 kauth_cred_t cred = nfsd->nd_cr; 3364 u_int32_t *tl; 3365 struct nfsv3_pathconf *pc; 3366 int32_t t1; 3367 char *bpos; 3368 int error = 0, rdonly, cache = 0, getret = 1; 3369 register_t linkmax, namemax, chownres, notrunc; 3370 char *cp2; 3371 struct mbuf *mb, *mreq; 3372 struct vnode *vp; 3373 struct vattr at; 3374 nfsrvfh_t nsfh; 3375 u_quad_t frev; 3376 3377 nfsm_srvmtofh(&nsfh); 3378 error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, 3379 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false); 3380 if (error) { 3381 nfsm_reply(NFSX_UNSIGNED); 3382 nfsm_srvpostop_attr(getret, &at); 3383 return (0); 3384 } 3385 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax); 3386 if (!error) 3387 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax); 3388 if (!error) 3389 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres); 3390 if (!error) 3391 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc); 3392 getret = VOP_GETATTR(vp, &at, cred, lwp); 3393 vput(vp); 3394 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF); 3395 nfsm_srvpostop_attr(getret, &at); 3396 if (error) 3397 return (0); 3398 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF); 3399 3400 pc->pc_linkmax = txdr_unsigned(linkmax); 3401 pc->pc_namemax = txdr_unsigned(namemax); 3402 pc->pc_notrunc = txdr_unsigned(notrunc); 3403 pc->pc_chownrestricted = txdr_unsigned(chownres); 3404 3405 /* 3406 * These should probably be supported by VOP_PATHCONF(), but 3407 * until msdosfs is exportable (why would you want to?), the 3408 * Unix defaults should be ok. 3409 */ 3410 pc->pc_caseinsensitive = nfs_false; 3411 pc->pc_casepreserving = nfs_true; 3412 nfsm_srvdone; 3413 } 3414 3415 /* 3416 * Null operation, used by clients to ping server 3417 */ 3418 /* ARGSUSED */ 3419 int 3420 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3421 struct lwp *lwp, struct mbuf **mrq) 3422 { 3423 struct mbuf *mrep = nfsd->nd_mrep; 3424 char *bpos; 3425 int error = NFSERR_RETVOID, cache = 0; 3426 struct mbuf *mb, *mreq; 3427 u_quad_t frev; 3428 3429 nfsm_reply(0); 3430 return (0); 3431 } 3432 3433 /* 3434 * No operation, used for obsolete procedures 3435 */ 3436 /* ARGSUSED */ 3437 int 3438 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, 3439 struct lwp *lwp, struct mbuf **mrq) 3440 { 3441 struct mbuf *mrep = nfsd->nd_mrep; 3442 char *bpos; 3443 int error, cache = 0; 3444 struct mbuf *mb, *mreq; 3445 u_quad_t frev; 3446 3447 if (nfsd->nd_repstat) 3448 error = nfsd->nd_repstat; 3449 else 3450 error = EPROCUNAVAIL; 3451 nfsm_reply(0); 3452 return (0); 3453 } 3454 3455 /* 3456 * Perform access checking for vnodes obtained from file handles that would 3457 * refer to files already opened by a Unix client. You cannot just use 3458 * vn_writechk() and VOP_ACCESS() for two reasons. 3459 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case 3460 * 2 - The owner is to be given access irrespective of mode bits for some 3461 * operations, so that processes that chmod after opening a file don't 3462 * break. I don't like this because it opens a security hole, but since 3463 * the nfs server opens a security hole the size of a barn door anyhow, 3464 * what the heck. 3465 * 3466 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS() 3467 * will return EPERM instead of EACCESS. EPERM is always an error. 3468 */ 3469 int 3470 nfsrv_access(vp, flags, cred, rdonly, lwp, override) 3471 struct vnode *vp; 3472 int flags; 3473 kauth_cred_t cred; 3474 int rdonly; 3475 struct lwp *lwp; 3476 int override; 3477 { 3478 struct vattr vattr; 3479 int error; 3480 if (flags & VWRITE) { 3481 /* Just vn_writechk() changed to check rdonly */ 3482 /* 3483 * Disallow write attempts on read-only file systems; 3484 * unless the file is a socket or a block or character 3485 * device resident on the file system. 3486 */ 3487 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3488 switch (vp->v_type) { 3489 case VREG: 3490 case VDIR: 3491 case VLNK: 3492 return (EROFS); 3493 default: 3494 break; 3495 } 3496 } 3497 3498 /* 3499 * If the vnode is in use as a process's text, 3500 * we can't allow writing. 3501 */ 3502 if (vp->v_flag & VTEXT) 3503 return (ETXTBSY); 3504 } 3505 error = VOP_GETATTR(vp, &vattr, cred, lwp); 3506 if (error) 3507 return (error); 3508 error = VOP_ACCESS(vp, flags, cred, lwp); 3509 /* 3510 * Allow certain operations for the owner (reads and writes 3511 * on files that are already open). 3512 */ 3513 if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid) 3514 error = 0; 3515 return error; 3516 } 3517
Indexes created Mon Sep 22 10:09:38 GMT 2025