1 /* $NetBSD: nfs_vnops.c,v 1.258 2007/11/07 00:23:39 ad 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_vnops.c 8.19 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * vnode op calls for Sun NFS version 2 and 3 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.258 2007/11/07 00:23:39 ad Exp $"); 43 44 #include "opt_inet.h" 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 48 #include <sys/param.h> 49 #include <sys/proc.h> 50 #include <sys/kernel.h> 51 #include <sys/systm.h> 52 #include <sys/resourcevar.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/condvar.h> 56 #include <sys/disk.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.h> 59 #include <sys/mutex.h> 60 #include <sys/namei.h> 61 #include <sys/vnode.h> 62 #include <sys/dirent.h> 63 #include <sys/fcntl.h> 64 #include <sys/hash.h> 65 #include <sys/lockf.h> 66 #include <sys/stat.h> 67 #include <sys/unistd.h> 68 #include <sys/kauth.h> 69 70 #include <uvm/uvm_extern.h> 71 #include <uvm/uvm.h> 72 73 #include <miscfs/fifofs/fifo.h> 74 #include <miscfs/genfs/genfs.h> 75 #include <miscfs/genfs/genfs_node.h> 76 #include <miscfs/specfs/specdev.h> 77 78 #include <nfs/rpcv2.h> 79 #include <nfs/nfsproto.h> 80 #include <nfs/nfs.h> 81 #include <nfs/nfsnode.h> 82 #include <nfs/nfsmount.h> 83 #include <nfs/xdr_subs.h> 84 #include <nfs/nfsm_subs.h> 85 #include <nfs/nfs_var.h> 86 87 #include <net/if.h> 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 91 /* 92 * Global vfs data structures for nfs 93 */ 94 int (**nfsv2_vnodeop_p) __P((void *)); 95 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 96 { &vop_default_desc, vn_default_error }, 97 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 98 { &vop_create_desc, nfs_create }, /* create */ 99 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 100 { &vop_open_desc, nfs_open }, /* open */ 101 { &vop_close_desc, nfs_close }, /* close */ 102 { &vop_access_desc, nfs_access }, /* access */ 103 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 104 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 105 { &vop_read_desc, nfs_read }, /* read */ 106 { &vop_write_desc, nfs_write }, /* write */ 107 { &vop_lease_desc, nfs_lease_check }, /* lease */ 108 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 109 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 110 { &vop_poll_desc, nfs_poll }, /* poll */ 111 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 112 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 113 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 114 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 115 { &vop_seek_desc, nfs_seek }, /* seek */ 116 { &vop_remove_desc, nfs_remove }, /* remove */ 117 { &vop_link_desc, nfs_link }, /* link */ 118 { &vop_rename_desc, nfs_rename }, /* rename */ 119 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 120 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 121 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 122 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 123 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 124 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 125 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 126 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 127 { &vop_lock_desc, nfs_lock }, /* lock */ 128 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 129 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 130 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 131 { &vop_print_desc, nfs_print }, /* print */ 132 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 133 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 134 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 135 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 136 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 137 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 138 { NULL, NULL } 139 }; 140 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 141 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 142 143 /* 144 * Special device vnode ops 145 */ 146 int (**spec_nfsv2nodeop_p) __P((void *)); 147 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 148 { &vop_default_desc, vn_default_error }, 149 { &vop_lookup_desc, spec_lookup }, /* lookup */ 150 { &vop_create_desc, spec_create }, /* create */ 151 { &vop_mknod_desc, spec_mknod }, /* mknod */ 152 { &vop_open_desc, spec_open }, /* open */ 153 { &vop_close_desc, nfsspec_close }, /* close */ 154 { &vop_access_desc, nfsspec_access }, /* access */ 155 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 156 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 157 { &vop_read_desc, nfsspec_read }, /* read */ 158 { &vop_write_desc, nfsspec_write }, /* write */ 159 { &vop_lease_desc, spec_lease_check }, /* lease */ 160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 162 { &vop_poll_desc, spec_poll }, /* poll */ 163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 164 { &vop_revoke_desc, spec_revoke }, /* revoke */ 165 { &vop_mmap_desc, spec_mmap }, /* mmap */ 166 { &vop_fsync_desc, spec_fsync }, /* fsync */ 167 { &vop_seek_desc, spec_seek }, /* seek */ 168 { &vop_remove_desc, spec_remove }, /* remove */ 169 { &vop_link_desc, spec_link }, /* link */ 170 { &vop_rename_desc, spec_rename }, /* rename */ 171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 173 { &vop_symlink_desc, spec_symlink }, /* symlink */ 174 { &vop_readdir_desc, spec_readdir }, /* readdir */ 175 { &vop_readlink_desc, spec_readlink }, /* readlink */ 176 { &vop_abortop_desc, spec_abortop }, /* abortop */ 177 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 179 { &vop_lock_desc, nfs_lock }, /* lock */ 180 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 181 { &vop_bmap_desc, spec_bmap }, /* bmap */ 182 { &vop_strategy_desc, spec_strategy }, /* strategy */ 183 { &vop_print_desc, nfs_print }, /* print */ 184 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, spec_advlock }, /* advlock */ 187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 188 { &vop_getpages_desc, spec_getpages }, /* getpages */ 189 { &vop_putpages_desc, spec_putpages }, /* putpages */ 190 { NULL, NULL } 191 }; 192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 194 195 int (**fifo_nfsv2nodeop_p) __P((void *)); 196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 197 { &vop_default_desc, vn_default_error }, 198 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 199 { &vop_create_desc, fifo_create }, /* create */ 200 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 201 { &vop_open_desc, fifo_open }, /* open */ 202 { &vop_close_desc, nfsfifo_close }, /* close */ 203 { &vop_access_desc, nfsspec_access }, /* access */ 204 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 205 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 206 { &vop_read_desc, nfsfifo_read }, /* read */ 207 { &vop_write_desc, nfsfifo_write }, /* write */ 208 { &vop_lease_desc, fifo_lease_check }, /* lease */ 209 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 210 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 211 { &vop_poll_desc, fifo_poll }, /* poll */ 212 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 213 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 214 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 215 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 216 { &vop_seek_desc, fifo_seek }, /* seek */ 217 { &vop_remove_desc, fifo_remove }, /* remove */ 218 { &vop_link_desc, fifo_link }, /* link */ 219 { &vop_rename_desc, fifo_rename }, /* rename */ 220 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 221 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 222 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 223 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 224 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 225 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 226 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 227 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 228 { &vop_lock_desc, nfs_lock }, /* lock */ 229 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 230 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 231 { &vop_strategy_desc, genfs_badop }, /* strategy */ 232 { &vop_print_desc, nfs_print }, /* print */ 233 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 234 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 235 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 236 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 237 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 238 { NULL, NULL } 239 }; 240 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 241 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 242 243 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 244 size_t, kauth_cred_t, struct lwp *); 245 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 246 247 /* 248 * Global variables 249 */ 250 extern u_int32_t nfs_true, nfs_false; 251 extern u_int32_t nfs_xdrneg1; 252 extern const nfstype nfsv3_type[9]; 253 254 int nfs_numasync = 0; 255 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 256 #define UIO_ADVANCE(uio, siz) \ 257 (void)((uio)->uio_resid -= (siz), \ 258 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 259 (uio)->uio_iov->iov_len -= (siz)) 260 261 static void nfs_cache_enter(struct vnode *, struct vnode *, 262 struct componentname *); 263 264 static void 265 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 266 struct componentname *cnp) 267 { 268 struct nfsnode *dnp = VTONFS(dvp); 269 270 if (vp != NULL) { 271 struct nfsnode *np = VTONFS(vp); 272 273 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 274 } 275 276 if (!timespecisset(&dnp->n_nctime)) 277 dnp->n_nctime = dnp->n_vattr->va_mtime; 278 279 cache_enter(dvp, vp, cnp); 280 } 281 282 /* 283 * nfs null call from vfs. 284 */ 285 int 286 nfs_null(vp, cred, l) 287 struct vnode *vp; 288 kauth_cred_t cred; 289 struct lwp *l; 290 { 291 char *bpos, *dpos; 292 int error = 0; 293 struct mbuf *mreq, *mrep, *md, *mb; 294 struct nfsnode *np = VTONFS(vp); 295 296 nfsm_reqhead(np, NFSPROC_NULL, 0); 297 nfsm_request(np, NFSPROC_NULL, l, cred); 298 nfsm_reqdone; 299 return (error); 300 } 301 302 /* 303 * nfs access vnode op. 304 * For nfs version 2, just return ok. File accesses may fail later. 305 * For nfs version 3, use the access rpc to check accessibility. If file modes 306 * are changed on the server, accesses might still fail later. 307 */ 308 int 309 nfs_access(v) 310 void *v; 311 { 312 struct vop_access_args /* { 313 struct vnode *a_vp; 314 int a_mode; 315 kauth_cred_t a_cred; 316 struct lwp *a_l; 317 } */ *ap = v; 318 struct vnode *vp = ap->a_vp; 319 #ifndef NFS_V2_ONLY 320 u_int32_t *tl; 321 char *cp; 322 int32_t t1, t2; 323 char *bpos, *dpos, *cp2; 324 int error = 0, attrflag; 325 struct mbuf *mreq, *mrep, *md, *mb; 326 u_int32_t mode, rmode; 327 const int v3 = NFS_ISV3(vp); 328 #endif 329 int cachevalid; 330 struct nfsnode *np = VTONFS(vp); 331 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 332 333 cachevalid = (np->n_accstamp != -1 && 334 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 335 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 336 337 /* 338 * Check access cache first. If this request has been made for this 339 * uid shortly before, use the cached result. 340 */ 341 if (cachevalid) { 342 if (!np->n_accerror) { 343 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 344 return np->n_accerror; 345 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 346 return np->n_accerror; 347 } 348 349 #ifndef NFS_V2_ONLY 350 /* 351 * For nfs v3, do an access rpc, otherwise you are stuck emulating 352 * ufs_access() locally using the vattr. This may not be correct, 353 * since the server may apply other access criteria such as 354 * client uid-->server uid mapping that we do not know about, but 355 * this is better than just returning anything that is lying about 356 * in the cache. 357 */ 358 if (v3) { 359 nfsstats.rpccnt[NFSPROC_ACCESS]++; 360 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 361 nfsm_fhtom(np, v3); 362 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 363 if (ap->a_mode & VREAD) 364 mode = NFSV3ACCESS_READ; 365 else 366 mode = 0; 367 if (vp->v_type != VDIR) { 368 if (ap->a_mode & VWRITE) 369 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 370 if (ap->a_mode & VEXEC) 371 mode |= NFSV3ACCESS_EXECUTE; 372 } else { 373 if (ap->a_mode & VWRITE) 374 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 375 NFSV3ACCESS_DELETE); 376 if (ap->a_mode & VEXEC) 377 mode |= NFSV3ACCESS_LOOKUP; 378 } 379 *tl = txdr_unsigned(mode); 380 nfsm_request(np, NFSPROC_ACCESS, ap->a_l, ap->a_cred); 381 nfsm_postop_attr(vp, attrflag, 0); 382 if (!error) { 383 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 384 rmode = fxdr_unsigned(u_int32_t, *tl); 385 /* 386 * The NFS V3 spec does not clarify whether or not 387 * the returned access bits can be a superset of 388 * the ones requested, so... 389 */ 390 if ((rmode & mode) != mode) 391 error = EACCES; 392 } 393 nfsm_reqdone; 394 } else 395 #endif 396 return (nfsspec_access(ap)); 397 #ifndef NFS_V2_ONLY 398 /* 399 * Disallow write attempts on filesystems mounted read-only; 400 * unless the file is a socket, fifo, or a block or character 401 * device resident on the filesystem. 402 */ 403 if (!error && (ap->a_mode & VWRITE) && 404 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 405 switch (vp->v_type) { 406 case VREG: 407 case VDIR: 408 case VLNK: 409 error = EROFS; 410 default: 411 break; 412 } 413 } 414 415 if (!error || error == EACCES) { 416 /* 417 * If we got the same result as for a previous, 418 * different request, OR it in. Don't update 419 * the timestamp in that case. 420 */ 421 if (cachevalid && np->n_accstamp != -1 && 422 error == np->n_accerror) { 423 if (!error) 424 np->n_accmode |= ap->a_mode; 425 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 426 np->n_accmode = ap->a_mode; 427 } else { 428 np->n_accstamp = time_uptime; 429 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 430 np->n_accmode = ap->a_mode; 431 np->n_accerror = error; 432 } 433 } 434 435 return (error); 436 #endif 437 } 438 439 /* 440 * nfs open vnode op 441 * Check to see if the type is ok 442 * and that deletion is not in progress. 443 * For paged in text files, you will need to flush the page cache 444 * if consistency is lost. 445 */ 446 /* ARGSUSED */ 447 int 448 nfs_open(v) 449 void *v; 450 { 451 struct vop_open_args /* { 452 struct vnode *a_vp; 453 int a_mode; 454 kauth_cred_t a_cred; 455 struct lwp *a_l; 456 } */ *ap = v; 457 struct vnode *vp = ap->a_vp; 458 struct nfsnode *np = VTONFS(vp); 459 int error; 460 461 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 462 return (EACCES); 463 } 464 465 if (ap->a_mode & FREAD) { 466 if (np->n_rcred != NULL) 467 kauth_cred_free(np->n_rcred); 468 np->n_rcred = ap->a_cred; 469 kauth_cred_hold(np->n_rcred); 470 } 471 if (ap->a_mode & FWRITE) { 472 if (np->n_wcred != NULL) 473 kauth_cred_free(np->n_wcred); 474 np->n_wcred = ap->a_cred; 475 kauth_cred_hold(np->n_wcred); 476 } 477 478 error = nfs_flushstalebuf(vp, ap->a_cred, ap->a_l, 0); 479 if (error) 480 return error; 481 482 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 483 484 return (0); 485 } 486 487 /* 488 * nfs close vnode op 489 * What an NFS client should do upon close after writing is a debatable issue. 490 * Most NFS clients push delayed writes to the server upon close, basically for 491 * two reasons: 492 * 1 - So that any write errors may be reported back to the client process 493 * doing the close system call. By far the two most likely errors are 494 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 495 * 2 - To put a worst case upper bound on cache inconsistency between 496 * multiple clients for the file. 497 * There is also a consistency problem for Version 2 of the protocol w.r.t. 498 * not being able to tell if other clients are writing a file concurrently, 499 * since there is no way of knowing if the changed modify time in the reply 500 * is only due to the write for this client. 501 * (NFS Version 3 provides weak cache consistency data in the reply that 502 * should be sufficient to detect and handle this case.) 503 * 504 * The current code does the following: 505 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 506 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 507 * or commit them (this satisfies 1 and 2 except for the 508 * case where the server crashes after this close but 509 * before the commit RPC, which is felt to be "good 510 * enough". Changing the last argument to nfs_flush() to 511 * a 1 would force a commit operation, if it is felt a 512 * commit is necessary now. 513 */ 514 /* ARGSUSED */ 515 int 516 nfs_close(v) 517 void *v; 518 { 519 struct vop_close_args /* { 520 struct vnodeop_desc *a_desc; 521 struct vnode *a_vp; 522 int a_fflag; 523 kauth_cred_t a_cred; 524 struct lwp *a_l; 525 } */ *ap = v; 526 struct vnode *vp = ap->a_vp; 527 struct nfsnode *np = VTONFS(vp); 528 int error = 0; 529 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 530 531 if (vp->v_type == VREG) { 532 if (np->n_flag & NMODIFIED) { 533 #ifndef NFS_V2_ONLY 534 if (NFS_ISV3(vp)) { 535 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_l, 0); 536 np->n_flag &= ~NMODIFIED; 537 } else 538 #endif 539 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 1); 540 NFS_INVALIDATE_ATTRCACHE(np); 541 } 542 if (np->n_flag & NWRITEERR) { 543 np->n_flag &= ~NWRITEERR; 544 error = np->n_error; 545 } 546 } 547 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 548 return (error); 549 } 550 551 /* 552 * nfs getattr call from vfs. 553 */ 554 int 555 nfs_getattr(v) 556 void *v; 557 { 558 struct vop_getattr_args /* { 559 struct vnode *a_vp; 560 struct vattr *a_vap; 561 kauth_cred_t a_cred; 562 struct lwp *a_l; 563 } */ *ap = v; 564 struct vnode *vp = ap->a_vp; 565 struct nfsnode *np = VTONFS(vp); 566 char *cp; 567 u_int32_t *tl; 568 int32_t t1, t2; 569 char *bpos, *dpos; 570 int error = 0; 571 struct mbuf *mreq, *mrep, *md, *mb; 572 const int v3 = NFS_ISV3(vp); 573 574 /* 575 * Update local times for special files. 576 */ 577 if (np->n_flag & (NACC | NUPD)) 578 np->n_flag |= NCHG; 579 580 /* 581 * if we have delayed truncation, do it now. 582 */ 583 nfs_delayedtruncate(vp); 584 585 /* 586 * First look in the cache. 587 */ 588 if (nfs_getattrcache(vp, ap->a_vap) == 0) 589 return (0); 590 nfsstats.rpccnt[NFSPROC_GETATTR]++; 591 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 592 nfsm_fhtom(np, v3); 593 nfsm_request(np, NFSPROC_GETATTR, ap->a_l, ap->a_cred); 594 if (!error) { 595 nfsm_loadattr(vp, ap->a_vap, 0); 596 if (vp->v_type == VDIR && 597 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 598 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 599 } 600 nfsm_reqdone; 601 return (error); 602 } 603 604 /* 605 * nfs setattr call. 606 */ 607 int 608 nfs_setattr(v) 609 void *v; 610 { 611 struct vop_setattr_args /* { 612 struct vnodeop_desc *a_desc; 613 struct vnode *a_vp; 614 struct vattr *a_vap; 615 kauth_cred_t a_cred; 616 struct lwp *a_l; 617 } */ *ap = v; 618 struct vnode *vp = ap->a_vp; 619 struct nfsnode *np = VTONFS(vp); 620 struct vattr *vap = ap->a_vap; 621 int error = 0; 622 u_quad_t tsize = 0; 623 624 /* 625 * Setting of flags is not supported. 626 */ 627 if (vap->va_flags != VNOVAL) 628 return (EOPNOTSUPP); 629 630 /* 631 * Disallow write attempts if the filesystem is mounted read-only. 632 */ 633 if ((vap->va_uid != (uid_t)VNOVAL || 634 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 635 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 636 (vp->v_mount->mnt_flag & MNT_RDONLY)) 637 return (EROFS); 638 if (vap->va_size != VNOVAL) { 639 switch (vp->v_type) { 640 case VDIR: 641 return (EISDIR); 642 case VCHR: 643 case VBLK: 644 case VSOCK: 645 case VFIFO: 646 if (vap->va_mtime.tv_sec == VNOVAL && 647 vap->va_atime.tv_sec == VNOVAL && 648 vap->va_mode == (mode_t)VNOVAL && 649 vap->va_uid == (uid_t)VNOVAL && 650 vap->va_gid == (gid_t)VNOVAL) 651 return (0); 652 vap->va_size = VNOVAL; 653 break; 654 default: 655 /* 656 * Disallow write attempts if the filesystem is 657 * mounted read-only. 658 */ 659 if (vp->v_mount->mnt_flag & MNT_RDONLY) 660 return (EROFS); 661 genfs_node_wrlock(vp); 662 uvm_vnp_setsize(vp, vap->va_size); 663 tsize = np->n_size; 664 np->n_size = vap->va_size; 665 if (vap->va_size == 0) 666 error = nfs_vinvalbuf(vp, 0, 667 ap->a_cred, ap->a_l, 1); 668 else 669 error = nfs_vinvalbuf(vp, V_SAVE, 670 ap->a_cred, ap->a_l, 1); 671 if (error) { 672 uvm_vnp_setsize(vp, tsize); 673 genfs_node_unlock(vp); 674 return (error); 675 } 676 np->n_vattr->va_size = vap->va_size; 677 } 678 } else { 679 /* 680 * flush files before setattr because a later write of 681 * cached data might change timestamps or reset sugid bits 682 */ 683 if ((vap->va_mtime.tv_sec != VNOVAL || 684 vap->va_atime.tv_sec != VNOVAL || 685 vap->va_mode != VNOVAL) && 686 vp->v_type == VREG && 687 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 688 ap->a_l, 1)) == EINTR) 689 return (error); 690 } 691 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_l); 692 if (vap->va_size != VNOVAL) { 693 if (error) { 694 np->n_size = np->n_vattr->va_size = tsize; 695 uvm_vnp_setsize(vp, np->n_size); 696 } 697 genfs_node_unlock(vp); 698 } 699 VN_KNOTE(vp, NOTE_ATTRIB); 700 return (error); 701 } 702 703 /* 704 * Do an nfs setattr rpc. 705 */ 706 int 707 nfs_setattrrpc(vp, vap, cred, l) 708 struct vnode *vp; 709 struct vattr *vap; 710 kauth_cred_t cred; 711 struct lwp *l; 712 { 713 struct nfsv2_sattr *sp; 714 char *cp; 715 int32_t t1, t2; 716 char *bpos, *dpos; 717 u_int32_t *tl; 718 int error = 0; 719 struct mbuf *mreq, *mrep, *md, *mb; 720 const int v3 = NFS_ISV3(vp); 721 struct nfsnode *np = VTONFS(vp); 722 #ifndef NFS_V2_ONLY 723 int wccflag = NFSV3_WCCRATTR; 724 char *cp2; 725 #endif 726 727 nfsstats.rpccnt[NFSPROC_SETATTR]++; 728 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 729 nfsm_fhtom(np, v3); 730 #ifndef NFS_V2_ONLY 731 if (v3) { 732 nfsm_v3attrbuild(vap, true); 733 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 734 *tl = nfs_false; 735 } else { 736 #endif 737 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 738 if (vap->va_mode == (mode_t)VNOVAL) 739 sp->sa_mode = nfs_xdrneg1; 740 else 741 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 742 if (vap->va_uid == (uid_t)VNOVAL) 743 sp->sa_uid = nfs_xdrneg1; 744 else 745 sp->sa_uid = txdr_unsigned(vap->va_uid); 746 if (vap->va_gid == (gid_t)VNOVAL) 747 sp->sa_gid = nfs_xdrneg1; 748 else 749 sp->sa_gid = txdr_unsigned(vap->va_gid); 750 sp->sa_size = txdr_unsigned(vap->va_size); 751 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 752 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 753 #ifndef NFS_V2_ONLY 754 } 755 #endif 756 nfsm_request(np, NFSPROC_SETATTR, l, cred); 757 #ifndef NFS_V2_ONLY 758 if (v3) { 759 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 760 } else 761 #endif 762 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 763 nfsm_reqdone; 764 return (error); 765 } 766 767 /* 768 * nfs lookup call, one step at a time... 769 * First look in cache 770 * If not found, unlock the directory nfsnode and do the rpc 771 * 772 * This code is full of lock/unlock statements and checks, because 773 * we continue after cache_lookup has finished (we need to check 774 * with the attr cache and do an rpc if it has timed out). This means 775 * that the locking effects of cache_lookup have to be taken into 776 * account. 777 */ 778 int 779 nfs_lookup(v) 780 void *v; 781 { 782 struct vop_lookup_args /* { 783 struct vnodeop_desc *a_desc; 784 struct vnode *a_dvp; 785 struct vnode **a_vpp; 786 struct componentname *a_cnp; 787 } */ *ap = v; 788 struct componentname *cnp = ap->a_cnp; 789 struct vnode *dvp = ap->a_dvp; 790 struct vnode **vpp = ap->a_vpp; 791 int flags; 792 struct vnode *newvp; 793 u_int32_t *tl; 794 char *cp; 795 int32_t t1, t2; 796 char *bpos, *dpos, *cp2; 797 struct mbuf *mreq, *mrep, *md, *mb; 798 long len; 799 nfsfh_t *fhp; 800 struct nfsnode *np; 801 int error = 0, attrflag, fhsize; 802 const int v3 = NFS_ISV3(dvp); 803 804 flags = cnp->cn_flags; 805 806 *vpp = NULLVP; 807 newvp = NULLVP; 808 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 809 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 810 return (EROFS); 811 if (dvp->v_type != VDIR) 812 return (ENOTDIR); 813 814 /* 815 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 816 */ 817 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 818 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp); 819 if (error) 820 return error; 821 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 822 return EISDIR; 823 VREF(dvp); 824 *vpp = dvp; 825 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 826 cnp->cn_flags |= SAVENAME; 827 return 0; 828 } 829 830 np = VTONFS(dvp); 831 832 /* 833 * Before tediously performing a linear scan of the directory, 834 * check the name cache to see if the directory/name pair 835 * we are looking for is known already. 836 * If the directory/name pair is found in the name cache, 837 * we have to ensure the directory has not changed from 838 * the time the cache entry has been created. If it has, 839 * the cache entry has to be ignored. 840 */ 841 error = cache_lookup_raw(dvp, vpp, cnp); 842 KASSERT(dvp != *vpp); 843 if (error >= 0) { 844 struct vattr vattr; 845 int err2; 846 847 if (error && error != ENOENT) { 848 *vpp = NULLVP; 849 return error; 850 } 851 852 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp); 853 if (err2 != 0) { 854 if (error == 0) 855 vrele(*vpp); 856 *vpp = NULLVP; 857 return err2; 858 } 859 860 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred, 861 cnp->cn_lwp) || timespeccmp(&vattr.va_mtime, 862 &VTONFS(dvp)->n_nctime, !=)) { 863 if (error == 0) { 864 vrele(*vpp); 865 *vpp = NULLVP; 866 } 867 cache_purge1(dvp, NULL, PURGE_CHILDREN); 868 timespecclear(&np->n_nctime); 869 goto dorpc; 870 } 871 872 if (error == ENOENT) { 873 goto noentry; 874 } 875 876 newvp = *vpp; 877 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp) 878 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 879 nfsstats.lookupcache_hits++; 880 if ((flags & ISDOTDOT) != 0) { 881 VOP_UNLOCK(dvp, 0); 882 } 883 error = vn_lock(newvp, LK_EXCLUSIVE); 884 if ((flags & ISDOTDOT) != 0) { 885 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 886 } 887 if (error) { 888 /* newvp has been revoked. */ 889 vrele(newvp); 890 *vpp = NULL; 891 return error; 892 } 893 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 894 cnp->cn_flags |= SAVENAME; 895 KASSERT(newvp->v_type != VNON); 896 return (0); 897 } 898 cache_purge1(newvp, NULL, PURGE_PARENTS); 899 vrele(newvp); 900 *vpp = NULLVP; 901 } 902 dorpc: 903 #if 0 904 /* 905 * because nfsv3 has the same CREATE semantics as ours, 906 * we don't have to perform LOOKUPs beforehand. 907 * 908 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 909 * XXX although we have no way to know if O_EXCL is requested or not. 910 */ 911 912 if (v3 && cnp->cn_nameiop == CREATE && 913 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 914 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 915 cnp->cn_flags |= SAVENAME; 916 return (EJUSTRETURN); 917 } 918 #endif /* 0 */ 919 920 error = 0; 921 newvp = NULLVP; 922 nfsstats.lookupcache_misses++; 923 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 924 len = cnp->cn_namelen; 925 nfsm_reqhead(np, NFSPROC_LOOKUP, 926 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 927 nfsm_fhtom(np, v3); 928 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 929 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_lwp, cnp->cn_cred); 930 if (error) { 931 nfsm_postop_attr(dvp, attrflag, 0); 932 m_freem(mrep); 933 goto nfsmout; 934 } 935 nfsm_getfh(fhp, fhsize, v3); 936 937 /* 938 * Handle RENAME case... 939 */ 940 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 941 if (NFS_CMPFH(np, fhp, fhsize)) { 942 m_freem(mrep); 943 return (EISDIR); 944 } 945 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 946 if (error) { 947 m_freem(mrep); 948 return error; 949 } 950 newvp = NFSTOV(np); 951 #ifndef NFS_V2_ONLY 952 if (v3) { 953 nfsm_postop_attr(newvp, attrflag, 0); 954 nfsm_postop_attr(dvp, attrflag, 0); 955 } else 956 #endif 957 nfsm_loadattr(newvp, (struct vattr *)0, 0); 958 *vpp = newvp; 959 m_freem(mrep); 960 cnp->cn_flags |= SAVENAME; 961 goto validate; 962 } 963 964 /* 965 * The postop attr handling is duplicated for each if case, 966 * because it should be done while dvp is locked (unlocking 967 * dvp is different for each case). 968 */ 969 970 if (NFS_CMPFH(np, fhp, fhsize)) { 971 /* 972 * "." lookup 973 */ 974 VREF(dvp); 975 newvp = dvp; 976 #ifndef NFS_V2_ONLY 977 if (v3) { 978 nfsm_postop_attr(newvp, attrflag, 0); 979 nfsm_postop_attr(dvp, attrflag, 0); 980 } else 981 #endif 982 nfsm_loadattr(newvp, (struct vattr *)0, 0); 983 } else if (flags & ISDOTDOT) { 984 /* 985 * ".." lookup 986 */ 987 VOP_UNLOCK(dvp, 0); 988 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 989 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 990 if (error) { 991 m_freem(mrep); 992 return error; 993 } 994 newvp = NFSTOV(np); 995 996 #ifndef NFS_V2_ONLY 997 if (v3) { 998 nfsm_postop_attr(newvp, attrflag, 0); 999 nfsm_postop_attr(dvp, attrflag, 0); 1000 } else 1001 #endif 1002 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1003 } else { 1004 /* 1005 * Other lookups. 1006 */ 1007 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1008 if (error) { 1009 m_freem(mrep); 1010 return error; 1011 } 1012 newvp = NFSTOV(np); 1013 #ifndef NFS_V2_ONLY 1014 if (v3) { 1015 nfsm_postop_attr(newvp, attrflag, 0); 1016 nfsm_postop_attr(dvp, attrflag, 0); 1017 } else 1018 #endif 1019 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1020 } 1021 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1022 cnp->cn_flags |= SAVENAME; 1023 if ((cnp->cn_flags & MAKEENTRY) && 1024 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1025 nfs_cache_enter(dvp, newvp, cnp); 1026 } 1027 *vpp = newvp; 1028 nfsm_reqdone; 1029 if (error) { 1030 /* 1031 * We get here only because of errors returned by 1032 * the RPC. Otherwise we'll have returned above 1033 * (the nfsm_* macros will jump to nfsm_reqdone 1034 * on error). 1035 */ 1036 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1037 cnp->cn_nameiop != CREATE) { 1038 nfs_cache_enter(dvp, NULL, cnp); 1039 } 1040 if (newvp != NULLVP) { 1041 if (newvp == dvp) { 1042 vrele(newvp); 1043 } else { 1044 vput(newvp); 1045 } 1046 } 1047 noentry: 1048 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1049 (flags & ISLASTCN) && error == ENOENT) { 1050 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1051 error = EROFS; 1052 } else { 1053 error = EJUSTRETURN; 1054 cnp->cn_flags |= SAVENAME; 1055 } 1056 } 1057 *vpp = NULL; 1058 return error; 1059 } 1060 1061 validate: 1062 /* 1063 * make sure we have valid type and size. 1064 */ 1065 1066 newvp = *vpp; 1067 if (newvp->v_type == VNON) { 1068 struct vattr vattr; /* dummy */ 1069 1070 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1071 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp); 1072 if (error) { 1073 vput(newvp); 1074 *vpp = NULL; 1075 } 1076 } 1077 1078 return error; 1079 } 1080 1081 /* 1082 * nfs read call. 1083 * Just call nfs_bioread() to do the work. 1084 */ 1085 int 1086 nfs_read(v) 1087 void *v; 1088 { 1089 struct vop_read_args /* { 1090 struct vnode *a_vp; 1091 struct uio *a_uio; 1092 int a_ioflag; 1093 kauth_cred_t a_cred; 1094 } */ *ap = v; 1095 struct vnode *vp = ap->a_vp; 1096 1097 if (vp->v_type != VREG) 1098 return EISDIR; 1099 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1100 } 1101 1102 /* 1103 * nfs readlink call 1104 */ 1105 int 1106 nfs_readlink(v) 1107 void *v; 1108 { 1109 struct vop_readlink_args /* { 1110 struct vnode *a_vp; 1111 struct uio *a_uio; 1112 kauth_cred_t a_cred; 1113 } */ *ap = v; 1114 struct vnode *vp = ap->a_vp; 1115 struct nfsnode *np = VTONFS(vp); 1116 1117 if (vp->v_type != VLNK) 1118 return (EPERM); 1119 1120 if (np->n_rcred != NULL) { 1121 kauth_cred_free(np->n_rcred); 1122 } 1123 np->n_rcred = ap->a_cred; 1124 kauth_cred_hold(np->n_rcred); 1125 1126 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1127 } 1128 1129 /* 1130 * Do a readlink rpc. 1131 * Called by nfs_doio() from below the buffer cache. 1132 */ 1133 int 1134 nfs_readlinkrpc(vp, uiop, cred) 1135 struct vnode *vp; 1136 struct uio *uiop; 1137 kauth_cred_t cred; 1138 { 1139 u_int32_t *tl; 1140 char *cp; 1141 int32_t t1, t2; 1142 char *bpos, *dpos, *cp2; 1143 int error = 0; 1144 uint32_t len; 1145 struct mbuf *mreq, *mrep, *md, *mb; 1146 const int v3 = NFS_ISV3(vp); 1147 struct nfsnode *np = VTONFS(vp); 1148 #ifndef NFS_V2_ONLY 1149 int attrflag; 1150 #endif 1151 1152 nfsstats.rpccnt[NFSPROC_READLINK]++; 1153 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1154 nfsm_fhtom(np, v3); 1155 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1156 #ifndef NFS_V2_ONLY 1157 if (v3) 1158 nfsm_postop_attr(vp, attrflag, 0); 1159 #endif 1160 if (!error) { 1161 #ifndef NFS_V2_ONLY 1162 if (v3) { 1163 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1164 len = fxdr_unsigned(uint32_t, *tl); 1165 if (len > MAXPATHLEN) { 1166 /* 1167 * this pathname is too long for us. 1168 */ 1169 m_freem(mrep); 1170 /* Solaris returns EINVAL. should we follow? */ 1171 error = ENAMETOOLONG; 1172 goto nfsmout; 1173 } 1174 } else 1175 #endif 1176 { 1177 nfsm_strsiz(len, NFS_MAXPATHLEN); 1178 } 1179 nfsm_mtouio(uiop, len); 1180 } 1181 nfsm_reqdone; 1182 return (error); 1183 } 1184 1185 /* 1186 * nfs read rpc call 1187 * Ditto above 1188 */ 1189 int 1190 nfs_readrpc(vp, uiop) 1191 struct vnode *vp; 1192 struct uio *uiop; 1193 { 1194 u_int32_t *tl; 1195 char *cp; 1196 int32_t t1, t2; 1197 char *bpos, *dpos, *cp2; 1198 struct mbuf *mreq, *mrep, *md, *mb; 1199 struct nfsmount *nmp; 1200 int error = 0, len, retlen, tsiz, eof, byte_count; 1201 const int v3 = NFS_ISV3(vp); 1202 struct nfsnode *np = VTONFS(vp); 1203 #ifndef NFS_V2_ONLY 1204 int attrflag; 1205 #endif 1206 1207 #ifndef nolint 1208 eof = 0; 1209 #endif 1210 nmp = VFSTONFS(vp->v_mount); 1211 tsiz = uiop->uio_resid; 1212 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1213 return (EFBIG); 1214 iostat_busy(nmp->nm_stats); 1215 byte_count = 0; /* count bytes actually transferred */ 1216 while (tsiz > 0) { 1217 nfsstats.rpccnt[NFSPROC_READ]++; 1218 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1219 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1220 nfsm_fhtom(np, v3); 1221 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1222 #ifndef NFS_V2_ONLY 1223 if (v3) { 1224 txdr_hyper(uiop->uio_offset, tl); 1225 *(tl + 2) = txdr_unsigned(len); 1226 } else 1227 #endif 1228 { 1229 *tl++ = txdr_unsigned(uiop->uio_offset); 1230 *tl++ = txdr_unsigned(len); 1231 *tl = 0; 1232 } 1233 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1234 #ifndef NFS_V2_ONLY 1235 if (v3) { 1236 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1237 if (error) { 1238 m_freem(mrep); 1239 goto nfsmout; 1240 } 1241 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1242 eof = fxdr_unsigned(int, *(tl + 1)); 1243 } else 1244 #endif 1245 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1246 nfsm_strsiz(retlen, nmp->nm_rsize); 1247 nfsm_mtouio(uiop, retlen); 1248 m_freem(mrep); 1249 tsiz -= retlen; 1250 byte_count += retlen; 1251 #ifndef NFS_V2_ONLY 1252 if (v3) { 1253 if (eof || retlen == 0) 1254 tsiz = 0; 1255 } else 1256 #endif 1257 if (retlen < len) 1258 tsiz = 0; 1259 } 1260 nfsmout: 1261 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1262 return (error); 1263 } 1264 1265 struct nfs_writerpc_context { 1266 kmutex_t nwc_lock; 1267 kcondvar_t nwc_cv; 1268 int nwc_mbufcount; 1269 }; 1270 1271 /* 1272 * free mbuf used to refer protected pages while write rpc call. 1273 * called at splvm. 1274 */ 1275 static void 1276 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1277 { 1278 struct nfs_writerpc_context *ctx = arg; 1279 1280 KASSERT(m != NULL); 1281 KASSERT(ctx != NULL); 1282 pool_cache_put(mb_cache, m); 1283 mutex_enter(&ctx->nwc_lock); 1284 if (--ctx->nwc_mbufcount == 0) { 1285 cv_signal(&ctx->nwc_cv); 1286 } 1287 mutex_exit(&ctx->nwc_lock); 1288 } 1289 1290 /* 1291 * nfs write call 1292 */ 1293 int 1294 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1295 struct vnode *vp; 1296 struct uio *uiop; 1297 int *iomode; 1298 bool pageprotected; 1299 bool *stalewriteverfp; 1300 { 1301 u_int32_t *tl; 1302 char *cp; 1303 int32_t t1, t2; 1304 char *bpos, *dpos; 1305 struct mbuf *mreq, *mrep, *md, *mb; 1306 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1307 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1308 const int v3 = NFS_ISV3(vp); 1309 int committed = NFSV3WRITE_FILESYNC; 1310 struct nfsnode *np = VTONFS(vp); 1311 struct nfs_writerpc_context ctx; 1312 int byte_count; 1313 struct lwp *l = NULL; 1314 size_t origresid; 1315 #ifndef NFS_V2_ONLY 1316 char *cp2; 1317 int rlen, commit; 1318 #endif 1319 1320 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1321 cv_init(&ctx.nwc_cv, "nfsmblk"); 1322 ctx.nwc_mbufcount = 1; 1323 1324 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1325 panic("writerpc readonly vp %p", vp); 1326 } 1327 1328 #ifdef DIAGNOSTIC 1329 if (uiop->uio_iovcnt != 1) 1330 panic("nfs: writerpc iovcnt > 1"); 1331 #endif 1332 tsiz = uiop->uio_resid; 1333 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1334 return (EFBIG); 1335 if (pageprotected) { 1336 l = curlwp; 1337 uvm_lwp_hold(l); 1338 } 1339 retry: 1340 origresid = uiop->uio_resid; 1341 KASSERT(origresid == uiop->uio_iov->iov_len); 1342 iostat_busy(nmp->nm_stats); 1343 byte_count = 0; /* count of bytes actually written */ 1344 while (tsiz > 0) { 1345 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1346 uint32_t backup; 1347 bool stalewriteverf = false; 1348 1349 nfsstats.rpccnt[NFSPROC_WRITE]++; 1350 len = min(tsiz, nmp->nm_wsize); 1351 datalen = pageprotected ? 0 : nfsm_rndup(len); 1352 nfsm_reqhead(np, NFSPROC_WRITE, 1353 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1354 nfsm_fhtom(np, v3); 1355 #ifndef NFS_V2_ONLY 1356 if (v3) { 1357 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1358 txdr_hyper(uiop->uio_offset, tl); 1359 tl += 2; 1360 *tl++ = txdr_unsigned(len); 1361 *tl++ = txdr_unsigned(*iomode); 1362 *tl = txdr_unsigned(len); 1363 } else 1364 #endif 1365 { 1366 u_int32_t x; 1367 1368 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1369 /* Set both "begin" and "current" to non-garbage. */ 1370 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1371 *tl++ = x; /* "begin offset" */ 1372 *tl++ = x; /* "current offset" */ 1373 x = txdr_unsigned(len); 1374 *tl++ = x; /* total to this offset */ 1375 *tl = x; /* size of this write */ 1376 1377 } 1378 if (pageprotected) { 1379 /* 1380 * since we know pages can't be modified during i/o, 1381 * no need to copy them for us. 1382 */ 1383 struct mbuf *m; 1384 struct iovec *iovp = uiop->uio_iov; 1385 1386 m = m_get(M_WAIT, MT_DATA); 1387 MCLAIM(m, &nfs_mowner); 1388 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1389 nfs_writerpc_extfree, &ctx); 1390 m->m_flags |= M_EXT_ROMAP; 1391 m->m_len = len; 1392 mb->m_next = m; 1393 /* 1394 * no need to maintain mb and bpos here 1395 * because no one care them later. 1396 */ 1397 #if 0 1398 mb = m; 1399 bpos = mtod(void *, mb) + mb->m_len; 1400 #endif 1401 UIO_ADVANCE(uiop, len); 1402 uiop->uio_offset += len; 1403 mutex_enter(&ctx.nwc_lock); 1404 ctx.nwc_mbufcount++; 1405 mutex_exit(&ctx.nwc_lock); 1406 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1407 } else { 1408 nfsm_uiotom(uiop, len); 1409 } 1410 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1411 #ifndef NFS_V2_ONLY 1412 if (v3) { 1413 wccflag = NFSV3_WCCCHK; 1414 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1415 if (!error) { 1416 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1417 + NFSX_V3WRITEVERF); 1418 rlen = fxdr_unsigned(int, *tl++); 1419 if (rlen == 0) { 1420 error = NFSERR_IO; 1421 m_freem(mrep); 1422 break; 1423 } else if (rlen < len) { 1424 backup = len - rlen; 1425 UIO_ADVANCE(uiop, -backup); 1426 uiop->uio_offset -= backup; 1427 len = rlen; 1428 } 1429 commit = fxdr_unsigned(int, *tl++); 1430 1431 /* 1432 * Return the lowest committment level 1433 * obtained by any of the RPCs. 1434 */ 1435 if (committed == NFSV3WRITE_FILESYNC) 1436 committed = commit; 1437 else if (committed == NFSV3WRITE_DATASYNC && 1438 commit == NFSV3WRITE_UNSTABLE) 1439 committed = commit; 1440 mutex_enter(&nmp->nm_lock); 1441 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1442 memcpy(nmp->nm_writeverf, tl, 1443 NFSX_V3WRITEVERF); 1444 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1445 } else if ((nmp->nm_iflag & 1446 NFSMNT_STALEWRITEVERF) || 1447 memcmp(tl, nmp->nm_writeverf, 1448 NFSX_V3WRITEVERF)) { 1449 memcpy(nmp->nm_writeverf, tl, 1450 NFSX_V3WRITEVERF); 1451 /* 1452 * note NFSMNT_STALEWRITEVERF 1453 * if we're the first thread to 1454 * notice it. 1455 */ 1456 if ((nmp->nm_iflag & 1457 NFSMNT_STALEWRITEVERF) == 0) { 1458 stalewriteverf = true; 1459 nmp->nm_iflag |= 1460 NFSMNT_STALEWRITEVERF; 1461 } 1462 } 1463 mutex_exit(&nmp->nm_lock); 1464 } 1465 } else 1466 #endif 1467 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1468 if (wccflag) 1469 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1470 m_freem(mrep); 1471 if (error) 1472 break; 1473 tsiz -= len; 1474 byte_count += len; 1475 if (stalewriteverf) { 1476 *stalewriteverfp = true; 1477 stalewriteverf = false; 1478 if (committed == NFSV3WRITE_UNSTABLE && 1479 len != origresid) { 1480 /* 1481 * if our write requests weren't atomic but 1482 * unstable, datas in previous iterations 1483 * might have already been lost now. 1484 * then, we should resend them to nfsd. 1485 */ 1486 backup = origresid - tsiz; 1487 UIO_ADVANCE(uiop, -backup); 1488 uiop->uio_offset -= backup; 1489 tsiz = origresid; 1490 goto retry; 1491 } 1492 } 1493 } 1494 nfsmout: 1495 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1496 if (pageprotected) { 1497 /* 1498 * wait until mbufs go away. 1499 * retransmitted mbufs can survive longer than rpc requests 1500 * themselves. 1501 */ 1502 mutex_enter(&ctx.nwc_lock); 1503 ctx.nwc_mbufcount--; 1504 while (ctx.nwc_mbufcount > 0) { 1505 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1506 } 1507 mutex_exit(&ctx.nwc_lock); 1508 uvm_lwp_rele(l); 1509 } 1510 mutex_destroy(&ctx.nwc_lock); 1511 cv_destroy(&ctx.nwc_cv); 1512 *iomode = committed; 1513 if (error) 1514 uiop->uio_resid = tsiz; 1515 return (error); 1516 } 1517 1518 /* 1519 * nfs mknod rpc 1520 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1521 * mode set to specify the file type and the size field for rdev. 1522 */ 1523 int 1524 nfs_mknodrpc(dvp, vpp, cnp, vap) 1525 struct vnode *dvp; 1526 struct vnode **vpp; 1527 struct componentname *cnp; 1528 struct vattr *vap; 1529 { 1530 struct nfsv2_sattr *sp; 1531 u_int32_t *tl; 1532 char *cp; 1533 int32_t t1, t2; 1534 struct vnode *newvp = (struct vnode *)0; 1535 struct nfsnode *dnp, *np; 1536 char *cp2; 1537 char *bpos, *dpos; 1538 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1539 struct mbuf *mreq, *mrep, *md, *mb; 1540 u_int32_t rdev; 1541 const int v3 = NFS_ISV3(dvp); 1542 1543 if (vap->va_type == VCHR || vap->va_type == VBLK) 1544 rdev = txdr_unsigned(vap->va_rdev); 1545 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1546 rdev = nfs_xdrneg1; 1547 else { 1548 VOP_ABORTOP(dvp, cnp); 1549 vput(dvp); 1550 return (EOPNOTSUPP); 1551 } 1552 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1553 dnp = VTONFS(dvp); 1554 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1555 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1556 nfsm_fhtom(dnp, v3); 1557 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1558 #ifndef NFS_V2_ONLY 1559 if (v3) { 1560 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1561 *tl++ = vtonfsv3_type(vap->va_type); 1562 nfsm_v3attrbuild(vap, false); 1563 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1564 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1565 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1566 *tl = txdr_unsigned(minor(vap->va_rdev)); 1567 } 1568 } else 1569 #endif 1570 { 1571 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1572 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1573 sp->sa_uid = nfs_xdrneg1; 1574 sp->sa_gid = nfs_xdrneg1; 1575 sp->sa_size = rdev; 1576 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1577 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1578 } 1579 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_lwp, cnp->cn_cred); 1580 if (!error) { 1581 nfsm_mtofh(dvp, newvp, v3, gotvp); 1582 if (!gotvp) { 1583 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1584 cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np); 1585 if (!error) 1586 newvp = NFSTOV(np); 1587 } 1588 } 1589 #ifndef NFS_V2_ONLY 1590 if (v3) 1591 nfsm_wcc_data(dvp, wccflag, 0, !error); 1592 #endif 1593 nfsm_reqdone; 1594 if (error) { 1595 if (newvp) 1596 vput(newvp); 1597 } else { 1598 if (cnp->cn_flags & MAKEENTRY) 1599 nfs_cache_enter(dvp, newvp, cnp); 1600 *vpp = newvp; 1601 } 1602 PNBUF_PUT(cnp->cn_pnbuf); 1603 VTONFS(dvp)->n_flag |= NMODIFIED; 1604 if (!wccflag) 1605 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1606 vput(dvp); 1607 return (error); 1608 } 1609 1610 /* 1611 * nfs mknod vop 1612 * just call nfs_mknodrpc() to do the work. 1613 */ 1614 /* ARGSUSED */ 1615 int 1616 nfs_mknod(v) 1617 void *v; 1618 { 1619 struct vop_mknod_args /* { 1620 struct vnode *a_dvp; 1621 struct vnode **a_vpp; 1622 struct componentname *a_cnp; 1623 struct vattr *a_vap; 1624 } */ *ap = v; 1625 struct vnode *dvp = ap->a_dvp; 1626 struct componentname *cnp = ap->a_cnp; 1627 int error; 1628 1629 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1630 VN_KNOTE(dvp, NOTE_WRITE); 1631 if (error == 0 || error == EEXIST) 1632 cache_purge1(dvp, cnp, 0); 1633 return (error); 1634 } 1635 1636 #ifndef NFS_V2_ONLY 1637 static u_long create_verf; 1638 #endif 1639 /* 1640 * nfs file create call 1641 */ 1642 int 1643 nfs_create(v) 1644 void *v; 1645 { 1646 struct vop_create_args /* { 1647 struct vnode *a_dvp; 1648 struct vnode **a_vpp; 1649 struct componentname *a_cnp; 1650 struct vattr *a_vap; 1651 } */ *ap = v; 1652 struct vnode *dvp = ap->a_dvp; 1653 struct vattr *vap = ap->a_vap; 1654 struct componentname *cnp = ap->a_cnp; 1655 struct nfsv2_sattr *sp; 1656 u_int32_t *tl; 1657 char *cp; 1658 int32_t t1, t2; 1659 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1660 struct vnode *newvp = (struct vnode *)0; 1661 char *bpos, *dpos, *cp2; 1662 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1663 struct mbuf *mreq, *mrep, *md, *mb; 1664 const int v3 = NFS_ISV3(dvp); 1665 1666 /* 1667 * Oops, not for me.. 1668 */ 1669 if (vap->va_type == VSOCK) 1670 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1671 1672 KASSERT(vap->va_type == VREG); 1673 1674 #ifdef VA_EXCLUSIVE 1675 if (vap->va_vaflags & VA_EXCLUSIVE) 1676 fmode |= O_EXCL; 1677 #endif 1678 again: 1679 error = 0; 1680 nfsstats.rpccnt[NFSPROC_CREATE]++; 1681 dnp = VTONFS(dvp); 1682 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1683 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1684 nfsm_fhtom(dnp, v3); 1685 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1686 #ifndef NFS_V2_ONLY 1687 if (v3) { 1688 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1689 if (fmode & O_EXCL) { 1690 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1691 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1692 #ifdef INET 1693 if (TAILQ_FIRST(&in_ifaddrhead)) 1694 *tl++ = TAILQ_FIRST(&in_ifaddrhead)-> 1695 ia_addr.sin_addr.s_addr; 1696 else 1697 *tl++ = create_verf; 1698 #else 1699 *tl++ = create_verf; 1700 #endif 1701 *tl = ++create_verf; 1702 } else { 1703 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1704 nfsm_v3attrbuild(vap, false); 1705 } 1706 } else 1707 #endif 1708 { 1709 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1710 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1711 sp->sa_uid = nfs_xdrneg1; 1712 sp->sa_gid = nfs_xdrneg1; 1713 sp->sa_size = 0; 1714 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1715 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1716 } 1717 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_lwp, cnp->cn_cred); 1718 if (!error) { 1719 nfsm_mtofh(dvp, newvp, v3, gotvp); 1720 if (!gotvp) { 1721 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1722 cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np); 1723 if (!error) 1724 newvp = NFSTOV(np); 1725 } 1726 } 1727 #ifndef NFS_V2_ONLY 1728 if (v3) 1729 nfsm_wcc_data(dvp, wccflag, 0, !error); 1730 #endif 1731 nfsm_reqdone; 1732 if (error) { 1733 /* 1734 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1735 */ 1736 if (v3 && (fmode & O_EXCL) && error == ENOTSUP) { 1737 fmode &= ~O_EXCL; 1738 goto again; 1739 } 1740 } else if (v3 && (fmode & O_EXCL)) { 1741 struct timespec ts; 1742 1743 getnanotime(&ts); 1744 1745 /* 1746 * make sure that we'll update timestamps as 1747 * most server implementations use them to store 1748 * the create verifier. 1749 * 1750 * XXX it's better to use TOSERVER always. 1751 */ 1752 1753 if (vap->va_atime.tv_sec == VNOVAL) 1754 vap->va_atime = ts; 1755 if (vap->va_mtime.tv_sec == VNOVAL) 1756 vap->va_mtime = ts; 1757 1758 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_lwp); 1759 } 1760 if (error == 0) { 1761 if (cnp->cn_flags & MAKEENTRY) 1762 nfs_cache_enter(dvp, newvp, cnp); 1763 else 1764 cache_purge1(dvp, cnp, 0); 1765 *ap->a_vpp = newvp; 1766 } else { 1767 if (newvp) 1768 vput(newvp); 1769 if (error == EEXIST) 1770 cache_purge1(dvp, cnp, 0); 1771 } 1772 PNBUF_PUT(cnp->cn_pnbuf); 1773 VTONFS(dvp)->n_flag |= NMODIFIED; 1774 if (!wccflag) 1775 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1776 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1777 vput(dvp); 1778 return (error); 1779 } 1780 1781 /* 1782 * nfs file remove call 1783 * To try and make nfs semantics closer to ufs semantics, a file that has 1784 * other processes using the vnode is renamed instead of removed and then 1785 * removed later on the last close. 1786 * - If v_usecount > 1 1787 * If a rename is not already in the works 1788 * call nfs_sillyrename() to set it up 1789 * else 1790 * do the remove rpc 1791 */ 1792 int 1793 nfs_remove(v) 1794 void *v; 1795 { 1796 struct vop_remove_args /* { 1797 struct vnodeop_desc *a_desc; 1798 struct vnode * a_dvp; 1799 struct vnode * a_vp; 1800 struct componentname * a_cnp; 1801 } */ *ap = v; 1802 struct vnode *vp = ap->a_vp; 1803 struct vnode *dvp = ap->a_dvp; 1804 struct componentname *cnp = ap->a_cnp; 1805 struct nfsnode *np = VTONFS(vp); 1806 int error = 0; 1807 struct vattr vattr; 1808 1809 #ifndef DIAGNOSTIC 1810 if ((cnp->cn_flags & HASBUF) == 0) 1811 panic("nfs_remove: no name"); 1812 if (vp->v_usecount < 1) 1813 panic("nfs_remove: bad v_usecount"); 1814 #endif 1815 if (vp->v_type == VDIR) 1816 error = EPERM; 1817 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1818 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_lwp) == 0 && 1819 vattr.va_nlink > 1)) { 1820 /* 1821 * Purge the name cache so that the chance of a lookup for 1822 * the name succeeding while the remove is in progress is 1823 * minimized. Without node locking it can still happen, such 1824 * that an I/O op returns ESTALE, but since you get this if 1825 * another host removes the file.. 1826 */ 1827 cache_purge(vp); 1828 /* 1829 * throw away biocache buffers, mainly to avoid 1830 * unnecessary delayed writes later. 1831 */ 1832 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_lwp, 1); 1833 /* Do the rpc */ 1834 if (error != EINTR) 1835 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1836 cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp); 1837 } else if (!np->n_sillyrename) 1838 error = nfs_sillyrename(dvp, vp, cnp, false); 1839 PNBUF_PUT(cnp->cn_pnbuf); 1840 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1841 vattr.va_nlink == 1) { 1842 np->n_flag |= NREMOVED; 1843 } 1844 NFS_INVALIDATE_ATTRCACHE(np); 1845 VN_KNOTE(vp, NOTE_DELETE); 1846 VN_KNOTE(dvp, NOTE_WRITE); 1847 if (dvp == vp) 1848 vrele(vp); 1849 else 1850 vput(vp); 1851 vput(dvp); 1852 return (error); 1853 } 1854 1855 /* 1856 * nfs file remove rpc called from nfs_inactive 1857 */ 1858 int 1859 nfs_removeit(sp) 1860 struct sillyrename *sp; 1861 { 1862 1863 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1864 (struct lwp *)0)); 1865 } 1866 1867 /* 1868 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1869 */ 1870 int 1871 nfs_removerpc(dvp, name, namelen, cred, l) 1872 struct vnode *dvp; 1873 const char *name; 1874 int namelen; 1875 kauth_cred_t cred; 1876 struct lwp *l; 1877 { 1878 u_int32_t *tl; 1879 char *cp; 1880 #ifndef NFS_V2_ONLY 1881 int32_t t1; 1882 char *cp2; 1883 #endif 1884 int32_t t2; 1885 char *bpos, *dpos; 1886 int error = 0, wccflag = NFSV3_WCCRATTR; 1887 struct mbuf *mreq, *mrep, *md, *mb; 1888 const int v3 = NFS_ISV3(dvp); 1889 int rexmit = 0; 1890 struct nfsnode *dnp = VTONFS(dvp); 1891 1892 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1893 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1894 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1895 nfsm_fhtom(dnp, v3); 1896 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1897 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1898 #ifndef NFS_V2_ONLY 1899 if (v3) 1900 nfsm_wcc_data(dvp, wccflag, 0, !error); 1901 #endif 1902 nfsm_reqdone; 1903 VTONFS(dvp)->n_flag |= NMODIFIED; 1904 if (!wccflag) 1905 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1906 /* 1907 * Kludge City: If the first reply to the remove rpc is lost.. 1908 * the reply to the retransmitted request will be ENOENT 1909 * since the file was in fact removed 1910 * Therefore, we cheat and return success. 1911 */ 1912 if (rexmit && error == ENOENT) 1913 error = 0; 1914 return (error); 1915 } 1916 1917 /* 1918 * nfs file rename call 1919 */ 1920 int 1921 nfs_rename(v) 1922 void *v; 1923 { 1924 struct vop_rename_args /* { 1925 struct vnode *a_fdvp; 1926 struct vnode *a_fvp; 1927 struct componentname *a_fcnp; 1928 struct vnode *a_tdvp; 1929 struct vnode *a_tvp; 1930 struct componentname *a_tcnp; 1931 } */ *ap = v; 1932 struct vnode *fvp = ap->a_fvp; 1933 struct vnode *tvp = ap->a_tvp; 1934 struct vnode *fdvp = ap->a_fdvp; 1935 struct vnode *tdvp = ap->a_tdvp; 1936 struct componentname *tcnp = ap->a_tcnp; 1937 struct componentname *fcnp = ap->a_fcnp; 1938 int error; 1939 1940 #ifndef DIAGNOSTIC 1941 if ((tcnp->cn_flags & HASBUF) == 0 || 1942 (fcnp->cn_flags & HASBUF) == 0) 1943 panic("nfs_rename: no name"); 1944 #endif 1945 /* Check for cross-device rename */ 1946 if ((fvp->v_mount != tdvp->v_mount) || 1947 (tvp && (fvp->v_mount != tvp->v_mount))) { 1948 error = EXDEV; 1949 goto out; 1950 } 1951 1952 /* 1953 * If the tvp exists and is in use, sillyrename it before doing the 1954 * rename of the new file over it. 1955 * 1956 * Have sillyrename use link instead of rename if possible, 1957 * so that we don't lose the file if the rename fails, and so 1958 * that there's no window when the "to" file doesn't exist. 1959 */ 1960 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1961 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1962 VN_KNOTE(tvp, NOTE_DELETE); 1963 vput(tvp); 1964 tvp = NULL; 1965 } 1966 1967 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1968 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1969 tcnp->cn_lwp); 1970 1971 VN_KNOTE(fdvp, NOTE_WRITE); 1972 VN_KNOTE(tdvp, NOTE_WRITE); 1973 if (error == 0 || error == EEXIST) { 1974 if (fvp->v_type == VDIR) 1975 cache_purge(fvp); 1976 else 1977 cache_purge1(fdvp, fcnp, 0); 1978 if (tvp != NULL && tvp->v_type == VDIR) 1979 cache_purge(tvp); 1980 else 1981 cache_purge1(tdvp, tcnp, 0); 1982 } 1983 out: 1984 if (tdvp == tvp) 1985 vrele(tdvp); 1986 else 1987 vput(tdvp); 1988 if (tvp) 1989 vput(tvp); 1990 vrele(fdvp); 1991 vrele(fvp); 1992 return (error); 1993 } 1994 1995 /* 1996 * nfs file rename rpc called from nfs_remove() above 1997 */ 1998 int 1999 nfs_renameit(sdvp, scnp, sp) 2000 struct vnode *sdvp; 2001 struct componentname *scnp; 2002 struct sillyrename *sp; 2003 { 2004 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 2005 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_lwp)); 2006 } 2007 2008 /* 2009 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2010 */ 2011 int 2012 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l) 2013 struct vnode *fdvp; 2014 const char *fnameptr; 2015 int fnamelen; 2016 struct vnode *tdvp; 2017 const char *tnameptr; 2018 int tnamelen; 2019 kauth_cred_t cred; 2020 struct lwp *l; 2021 { 2022 u_int32_t *tl; 2023 char *cp; 2024 #ifndef NFS_V2_ONLY 2025 int32_t t1; 2026 char *cp2; 2027 #endif 2028 int32_t t2; 2029 char *bpos, *dpos; 2030 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 2031 struct mbuf *mreq, *mrep, *md, *mb; 2032 const int v3 = NFS_ISV3(fdvp); 2033 int rexmit = 0; 2034 struct nfsnode *fdnp = VTONFS(fdvp); 2035 2036 nfsstats.rpccnt[NFSPROC_RENAME]++; 2037 nfsm_reqhead(fdnp, NFSPROC_RENAME, 2038 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 2039 nfsm_rndup(tnamelen)); 2040 nfsm_fhtom(fdnp, v3); 2041 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 2042 nfsm_fhtom(VTONFS(tdvp), v3); 2043 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 2044 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 2045 #ifndef NFS_V2_ONLY 2046 if (v3) { 2047 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 2048 nfsm_wcc_data(tdvp, twccflag, 0, !error); 2049 } 2050 #endif 2051 nfsm_reqdone; 2052 VTONFS(fdvp)->n_flag |= NMODIFIED; 2053 VTONFS(tdvp)->n_flag |= NMODIFIED; 2054 if (!fwccflag) 2055 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 2056 if (!twccflag) 2057 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 2058 /* 2059 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2060 */ 2061 if (rexmit && error == ENOENT) 2062 error = 0; 2063 return (error); 2064 } 2065 2066 /* 2067 * NFS link RPC, called from nfs_link. 2068 * Assumes dvp and vp locked, and leaves them that way. 2069 */ 2070 2071 static int 2072 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2073 size_t namelen, kauth_cred_t cred, struct lwp *l) 2074 { 2075 u_int32_t *tl; 2076 char *cp; 2077 #ifndef NFS_V2_ONLY 2078 int32_t t1; 2079 char *cp2; 2080 #endif 2081 int32_t t2; 2082 char *bpos, *dpos; 2083 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2084 struct mbuf *mreq, *mrep, *md, *mb; 2085 const int v3 = NFS_ISV3(dvp); 2086 int rexmit = 0; 2087 struct nfsnode *np = VTONFS(vp); 2088 2089 nfsstats.rpccnt[NFSPROC_LINK]++; 2090 nfsm_reqhead(np, NFSPROC_LINK, 2091 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2092 nfsm_fhtom(np, v3); 2093 nfsm_fhtom(VTONFS(dvp), v3); 2094 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2095 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2096 #ifndef NFS_V2_ONLY 2097 if (v3) { 2098 nfsm_postop_attr(vp, attrflag, 0); 2099 nfsm_wcc_data(dvp, wccflag, 0, !error); 2100 } 2101 #endif 2102 nfsm_reqdone; 2103 2104 VTONFS(dvp)->n_flag |= NMODIFIED; 2105 if (!attrflag) 2106 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2107 if (!wccflag) 2108 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2109 2110 /* 2111 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2112 */ 2113 if (rexmit && error == EEXIST) 2114 error = 0; 2115 2116 return error; 2117 } 2118 2119 /* 2120 * nfs hard link create call 2121 */ 2122 int 2123 nfs_link(v) 2124 void *v; 2125 { 2126 struct vop_link_args /* { 2127 struct vnode *a_dvp; 2128 struct vnode *a_vp; 2129 struct componentname *a_cnp; 2130 } */ *ap = v; 2131 struct vnode *vp = ap->a_vp; 2132 struct vnode *dvp = ap->a_dvp; 2133 struct componentname *cnp = ap->a_cnp; 2134 int error = 0; 2135 2136 if (dvp->v_mount != vp->v_mount) { 2137 VOP_ABORTOP(dvp, cnp); 2138 vput(dvp); 2139 return (EXDEV); 2140 } 2141 if (dvp != vp) { 2142 error = vn_lock(vp, LK_EXCLUSIVE); 2143 if (error != 0) { 2144 VOP_ABORTOP(dvp, cnp); 2145 vput(dvp); 2146 return error; 2147 } 2148 } 2149 2150 /* 2151 * Push all writes to the server, so that the attribute cache 2152 * doesn't get "out of sync" with the server. 2153 * XXX There should be a better way! 2154 */ 2155 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_lwp); 2156 2157 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2158 cnp->cn_cred, cnp->cn_lwp); 2159 2160 if (error == 0) 2161 cache_purge1(dvp, cnp, 0); 2162 PNBUF_PUT(cnp->cn_pnbuf); 2163 if (dvp != vp) 2164 VOP_UNLOCK(vp, 0); 2165 VN_KNOTE(vp, NOTE_LINK); 2166 VN_KNOTE(dvp, NOTE_WRITE); 2167 vput(dvp); 2168 return (error); 2169 } 2170 2171 /* 2172 * nfs symbolic link create call 2173 */ 2174 int 2175 nfs_symlink(v) 2176 void *v; 2177 { 2178 struct vop_symlink_args /* { 2179 struct vnode *a_dvp; 2180 struct vnode **a_vpp; 2181 struct componentname *a_cnp; 2182 struct vattr *a_vap; 2183 char *a_target; 2184 } */ *ap = v; 2185 struct vnode *dvp = ap->a_dvp; 2186 struct vattr *vap = ap->a_vap; 2187 struct componentname *cnp = ap->a_cnp; 2188 struct nfsv2_sattr *sp; 2189 u_int32_t *tl; 2190 char *cp; 2191 int32_t t1, t2; 2192 char *bpos, *dpos, *cp2; 2193 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2194 struct mbuf *mreq, *mrep, *md, *mb; 2195 struct vnode *newvp = (struct vnode *)0; 2196 const int v3 = NFS_ISV3(dvp); 2197 int rexmit = 0; 2198 struct nfsnode *dnp = VTONFS(dvp); 2199 2200 *ap->a_vpp = NULL; 2201 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2202 slen = strlen(ap->a_target); 2203 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2204 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2205 nfsm_fhtom(dnp, v3); 2206 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2207 #ifndef NFS_V2_ONlY 2208 if (v3) 2209 nfsm_v3attrbuild(vap, false); 2210 #endif 2211 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2212 #ifndef NFS_V2_ONlY 2213 if (!v3) { 2214 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2215 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2216 sp->sa_uid = nfs_xdrneg1; 2217 sp->sa_gid = nfs_xdrneg1; 2218 sp->sa_size = nfs_xdrneg1; 2219 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2220 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2221 } 2222 #endif 2223 nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_lwp, cnp->cn_cred, 2224 &rexmit); 2225 #ifndef NFS_V2_ONlY 2226 if (v3) { 2227 if (!error) 2228 nfsm_mtofh(dvp, newvp, v3, gotvp); 2229 nfsm_wcc_data(dvp, wccflag, 0, !error); 2230 } 2231 #endif 2232 nfsm_reqdone; 2233 /* 2234 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2235 */ 2236 if (rexmit && error == EEXIST) 2237 error = 0; 2238 if (error == 0 || error == EEXIST) 2239 cache_purge1(dvp, cnp, 0); 2240 if (error == 0 && newvp == NULL) { 2241 struct nfsnode *np = NULL; 2242 2243 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2244 cnp->cn_cred, cnp->cn_lwp, &np); 2245 if (error == 0) 2246 newvp = NFSTOV(np); 2247 } 2248 if (error) { 2249 if (newvp != NULL) 2250 vput(newvp); 2251 } else { 2252 *ap->a_vpp = newvp; 2253 } 2254 PNBUF_PUT(cnp->cn_pnbuf); 2255 VTONFS(dvp)->n_flag |= NMODIFIED; 2256 if (!wccflag) 2257 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2258 VN_KNOTE(dvp, NOTE_WRITE); 2259 vput(dvp); 2260 return (error); 2261 } 2262 2263 /* 2264 * nfs make dir call 2265 */ 2266 int 2267 nfs_mkdir(v) 2268 void *v; 2269 { 2270 struct vop_mkdir_args /* { 2271 struct vnode *a_dvp; 2272 struct vnode **a_vpp; 2273 struct componentname *a_cnp; 2274 struct vattr *a_vap; 2275 } */ *ap = v; 2276 struct vnode *dvp = ap->a_dvp; 2277 struct vattr *vap = ap->a_vap; 2278 struct componentname *cnp = ap->a_cnp; 2279 struct nfsv2_sattr *sp; 2280 u_int32_t *tl; 2281 char *cp; 2282 int32_t t1, t2; 2283 int len; 2284 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2285 struct vnode *newvp = (struct vnode *)0; 2286 char *bpos, *dpos, *cp2; 2287 int error = 0, wccflag = NFSV3_WCCRATTR; 2288 int gotvp = 0; 2289 int rexmit = 0; 2290 struct mbuf *mreq, *mrep, *md, *mb; 2291 const int v3 = NFS_ISV3(dvp); 2292 2293 len = cnp->cn_namelen; 2294 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2295 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2296 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2297 nfsm_fhtom(dnp, v3); 2298 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2299 #ifndef NFS_V2_ONLY 2300 if (v3) { 2301 nfsm_v3attrbuild(vap, false); 2302 } else 2303 #endif 2304 { 2305 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2306 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2307 sp->sa_uid = nfs_xdrneg1; 2308 sp->sa_gid = nfs_xdrneg1; 2309 sp->sa_size = nfs_xdrneg1; 2310 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2311 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2312 } 2313 nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit); 2314 if (!error) 2315 nfsm_mtofh(dvp, newvp, v3, gotvp); 2316 if (v3) 2317 nfsm_wcc_data(dvp, wccflag, 0, !error); 2318 nfsm_reqdone; 2319 VTONFS(dvp)->n_flag |= NMODIFIED; 2320 if (!wccflag) 2321 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2322 /* 2323 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2324 * if we can succeed in looking up the directory. 2325 */ 2326 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2327 if (newvp) { 2328 vput(newvp); 2329 newvp = (struct vnode *)0; 2330 } 2331 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2332 cnp->cn_lwp, &np); 2333 if (!error) { 2334 newvp = NFSTOV(np); 2335 if (newvp->v_type != VDIR || newvp == dvp) 2336 error = EEXIST; 2337 } 2338 } 2339 if (error) { 2340 if (newvp) { 2341 if (dvp != newvp) 2342 vput(newvp); 2343 else 2344 vrele(newvp); 2345 } 2346 } else { 2347 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2348 if (cnp->cn_flags & MAKEENTRY) 2349 nfs_cache_enter(dvp, newvp, cnp); 2350 *ap->a_vpp = newvp; 2351 } 2352 PNBUF_PUT(cnp->cn_pnbuf); 2353 vput(dvp); 2354 return (error); 2355 } 2356 2357 /* 2358 * nfs remove directory call 2359 */ 2360 int 2361 nfs_rmdir(v) 2362 void *v; 2363 { 2364 struct vop_rmdir_args /* { 2365 struct vnode *a_dvp; 2366 struct vnode *a_vp; 2367 struct componentname *a_cnp; 2368 } */ *ap = v; 2369 struct vnode *vp = ap->a_vp; 2370 struct vnode *dvp = ap->a_dvp; 2371 struct componentname *cnp = ap->a_cnp; 2372 u_int32_t *tl; 2373 char *cp; 2374 #ifndef NFS_V2_ONLY 2375 int32_t t1; 2376 char *cp2; 2377 #endif 2378 int32_t t2; 2379 char *bpos, *dpos; 2380 int error = 0, wccflag = NFSV3_WCCRATTR; 2381 int rexmit = 0; 2382 struct mbuf *mreq, *mrep, *md, *mb; 2383 const int v3 = NFS_ISV3(dvp); 2384 struct nfsnode *dnp; 2385 2386 if (dvp == vp) { 2387 vrele(dvp); 2388 vput(dvp); 2389 PNBUF_PUT(cnp->cn_pnbuf); 2390 return (EINVAL); 2391 } 2392 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2393 dnp = VTONFS(dvp); 2394 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2395 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2396 nfsm_fhtom(dnp, v3); 2397 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2398 nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit); 2399 #ifndef NFS_V2_ONLY 2400 if (v3) 2401 nfsm_wcc_data(dvp, wccflag, 0, !error); 2402 #endif 2403 nfsm_reqdone; 2404 PNBUF_PUT(cnp->cn_pnbuf); 2405 VTONFS(dvp)->n_flag |= NMODIFIED; 2406 if (!wccflag) 2407 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2408 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2409 VN_KNOTE(vp, NOTE_DELETE); 2410 cache_purge(vp); 2411 vput(vp); 2412 vput(dvp); 2413 /* 2414 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2415 */ 2416 if (rexmit && error == ENOENT) 2417 error = 0; 2418 return (error); 2419 } 2420 2421 /* 2422 * nfs readdir call 2423 */ 2424 int 2425 nfs_readdir(v) 2426 void *v; 2427 { 2428 struct vop_readdir_args /* { 2429 struct vnode *a_vp; 2430 struct uio *a_uio; 2431 kauth_cred_t a_cred; 2432 int *a_eofflag; 2433 off_t **a_cookies; 2434 int *a_ncookies; 2435 } */ *ap = v; 2436 struct vnode *vp = ap->a_vp; 2437 struct uio *uio = ap->a_uio; 2438 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2439 char *base = uio->uio_iov->iov_base; 2440 int tresid, error; 2441 size_t count, lost; 2442 struct dirent *dp; 2443 off_t *cookies = NULL; 2444 int ncookies = 0, nc; 2445 2446 if (vp->v_type != VDIR) 2447 return (EPERM); 2448 2449 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2450 count = uio->uio_resid - lost; 2451 if (count <= 0) 2452 return (EINVAL); 2453 2454 /* 2455 * Call nfs_bioread() to do the real work. 2456 */ 2457 tresid = uio->uio_resid = count; 2458 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2459 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2460 2461 if (!error && ap->a_cookies) { 2462 ncookies = count / 16; 2463 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2464 *ap->a_cookies = cookies; 2465 } 2466 2467 if (!error && uio->uio_resid == tresid) { 2468 uio->uio_resid += lost; 2469 nfsstats.direofcache_misses++; 2470 if (ap->a_cookies) 2471 *ap->a_ncookies = 0; 2472 *ap->a_eofflag = 1; 2473 return (0); 2474 } 2475 2476 if (!error && ap->a_cookies) { 2477 /* 2478 * Only the NFS server and emulations use cookies, and they 2479 * load the directory block into system space, so we can 2480 * just look at it directly. 2481 */ 2482 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2483 uio->uio_iovcnt != 1) 2484 panic("nfs_readdir: lost in space"); 2485 for (nc = 0; ncookies-- && 2486 base < (char *)uio->uio_iov->iov_base; nc++){ 2487 dp = (struct dirent *) base; 2488 if (dp->d_reclen == 0) 2489 break; 2490 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2491 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2492 else 2493 *(cookies++) = NFS_GETCOOKIE(dp); 2494 base += dp->d_reclen; 2495 } 2496 uio->uio_resid += 2497 ((char *)uio->uio_iov->iov_base - base); 2498 uio->uio_iov->iov_len += 2499 ((char *)uio->uio_iov->iov_base - base); 2500 uio->uio_iov->iov_base = base; 2501 *ap->a_ncookies = nc; 2502 } 2503 2504 uio->uio_resid += lost; 2505 *ap->a_eofflag = 0; 2506 return (error); 2507 } 2508 2509 /* 2510 * Readdir rpc call. 2511 * Called from below the buffer cache by nfs_doio(). 2512 */ 2513 int 2514 nfs_readdirrpc(vp, uiop, cred) 2515 struct vnode *vp; 2516 struct uio *uiop; 2517 kauth_cred_t cred; 2518 { 2519 int len, left; 2520 struct dirent *dp = NULL; 2521 u_int32_t *tl; 2522 char *cp; 2523 int32_t t1, t2; 2524 char *bpos, *dpos, *cp2; 2525 struct mbuf *mreq, *mrep, *md, *mb; 2526 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2527 struct nfsnode *dnp = VTONFS(vp); 2528 u_quad_t fileno; 2529 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2530 #ifndef NFS_V2_ONLY 2531 int attrflag; 2532 #endif 2533 int nrpcs = 0, reclen; 2534 const int v3 = NFS_ISV3(vp); 2535 2536 #ifdef DIAGNOSTIC 2537 /* 2538 * Should be called from buffer cache, so only amount of 2539 * NFS_DIRBLKSIZ will be requested. 2540 */ 2541 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2542 panic("nfs readdirrpc bad uio"); 2543 #endif 2544 2545 /* 2546 * Loop around doing readdir rpc's of size nm_readdirsize 2547 * truncated to a multiple of NFS_DIRFRAGSIZ. 2548 * The stopping criteria is EOF or buffer full. 2549 */ 2550 while (more_dirs && bigenough) { 2551 /* 2552 * Heuristic: don't bother to do another RPC to further 2553 * fill up this block if there is not much room left. (< 50% 2554 * of the readdir RPC size). This wastes some buffer space 2555 * but can save up to 50% in RPC calls. 2556 */ 2557 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2558 bigenough = 0; 2559 break; 2560 } 2561 nfsstats.rpccnt[NFSPROC_READDIR]++; 2562 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2563 NFSX_READDIR(v3)); 2564 nfsm_fhtom(dnp, v3); 2565 #ifndef NFS_V2_ONLY 2566 if (v3) { 2567 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2568 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2569 txdr_swapcookie3(uiop->uio_offset, tl); 2570 } else { 2571 txdr_cookie3(uiop->uio_offset, tl); 2572 } 2573 tl += 2; 2574 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2575 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2576 } else 2577 #endif 2578 { 2579 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2580 *tl++ = txdr_unsigned(uiop->uio_offset); 2581 } 2582 *tl = txdr_unsigned(nmp->nm_readdirsize); 2583 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2584 nrpcs++; 2585 #ifndef NFS_V2_ONLY 2586 if (v3) { 2587 nfsm_postop_attr(vp, attrflag, 0); 2588 if (!error) { 2589 nfsm_dissect(tl, u_int32_t *, 2590 2 * NFSX_UNSIGNED); 2591 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2592 dnp->n_cookieverf.nfsuquad[1] = *tl; 2593 } else { 2594 m_freem(mrep); 2595 goto nfsmout; 2596 } 2597 } 2598 #endif 2599 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2600 more_dirs = fxdr_unsigned(int, *tl); 2601 2602 /* loop thru the dir entries, doctoring them to 4bsd form */ 2603 while (more_dirs && bigenough) { 2604 #ifndef NFS_V2_ONLY 2605 if (v3) { 2606 nfsm_dissect(tl, u_int32_t *, 2607 3 * NFSX_UNSIGNED); 2608 fileno = fxdr_hyper(tl); 2609 len = fxdr_unsigned(int, *(tl + 2)); 2610 } else 2611 #endif 2612 { 2613 nfsm_dissect(tl, u_int32_t *, 2614 2 * NFSX_UNSIGNED); 2615 fileno = fxdr_unsigned(u_quad_t, *tl++); 2616 len = fxdr_unsigned(int, *tl); 2617 } 2618 if (len <= 0 || len > NFS_MAXNAMLEN) { 2619 error = EBADRPC; 2620 m_freem(mrep); 2621 goto nfsmout; 2622 } 2623 /* for cookie stashing */ 2624 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2625 left = NFS_DIRFRAGSIZ - blksiz; 2626 if (reclen > left) { 2627 memset(uiop->uio_iov->iov_base, 0, left); 2628 dp->d_reclen += left; 2629 UIO_ADVANCE(uiop, left); 2630 blksiz = 0; 2631 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2632 } 2633 if (reclen > uiop->uio_resid) 2634 bigenough = 0; 2635 if (bigenough) { 2636 int tlen; 2637 2638 dp = (struct dirent *)uiop->uio_iov->iov_base; 2639 dp->d_fileno = fileno; 2640 dp->d_namlen = len; 2641 dp->d_reclen = reclen; 2642 dp->d_type = DT_UNKNOWN; 2643 blksiz += reclen; 2644 if (blksiz == NFS_DIRFRAGSIZ) 2645 blksiz = 0; 2646 UIO_ADVANCE(uiop, DIRHDSIZ); 2647 nfsm_mtouio(uiop, len); 2648 tlen = reclen - (DIRHDSIZ + len); 2649 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2650 UIO_ADVANCE(uiop, tlen); 2651 } else 2652 nfsm_adv(nfsm_rndup(len)); 2653 #ifndef NFS_V2_ONLY 2654 if (v3) { 2655 nfsm_dissect(tl, u_int32_t *, 2656 3 * NFSX_UNSIGNED); 2657 } else 2658 #endif 2659 { 2660 nfsm_dissect(tl, u_int32_t *, 2661 2 * NFSX_UNSIGNED); 2662 } 2663 if (bigenough) { 2664 #ifndef NFS_V2_ONLY 2665 if (v3) { 2666 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2667 uiop->uio_offset = 2668 fxdr_swapcookie3(tl); 2669 else 2670 uiop->uio_offset = 2671 fxdr_cookie3(tl); 2672 } 2673 else 2674 #endif 2675 { 2676 uiop->uio_offset = 2677 fxdr_unsigned(off_t, *tl); 2678 } 2679 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2680 } 2681 if (v3) 2682 tl += 2; 2683 else 2684 tl++; 2685 more_dirs = fxdr_unsigned(int, *tl); 2686 } 2687 /* 2688 * If at end of rpc data, get the eof boolean 2689 */ 2690 if (!more_dirs) { 2691 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2692 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2693 2694 /* 2695 * kludge: if we got no entries, treat it as EOF. 2696 * some server sometimes send a reply without any 2697 * entries or EOF. 2698 * although it might mean the server has very long name, 2699 * we can't handle such entries anyway. 2700 */ 2701 2702 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2703 more_dirs = 0; 2704 } 2705 m_freem(mrep); 2706 } 2707 /* 2708 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2709 * by increasing d_reclen for the last record. 2710 */ 2711 if (blksiz > 0) { 2712 left = NFS_DIRFRAGSIZ - blksiz; 2713 memset(uiop->uio_iov->iov_base, 0, left); 2714 dp->d_reclen += left; 2715 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2716 UIO_ADVANCE(uiop, left); 2717 } 2718 2719 /* 2720 * We are now either at the end of the directory or have filled the 2721 * block. 2722 */ 2723 if (bigenough) { 2724 dnp->n_direofoffset = uiop->uio_offset; 2725 dnp->n_flag |= NEOFVALID; 2726 } 2727 nfsmout: 2728 return (error); 2729 } 2730 2731 #ifndef NFS_V2_ONLY 2732 /* 2733 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2734 */ 2735 int 2736 nfs_readdirplusrpc(vp, uiop, cred) 2737 struct vnode *vp; 2738 struct uio *uiop; 2739 kauth_cred_t cred; 2740 { 2741 int len, left; 2742 struct dirent *dp = NULL; 2743 u_int32_t *tl; 2744 char *cp; 2745 int32_t t1, t2; 2746 struct vnode *newvp; 2747 char *bpos, *dpos, *cp2; 2748 struct mbuf *mreq, *mrep, *md, *mb; 2749 struct nameidata nami, *ndp = &nami; 2750 struct componentname *cnp = &ndp->ni_cnd; 2751 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2752 struct nfsnode *dnp = VTONFS(vp), *np; 2753 nfsfh_t *fhp; 2754 u_quad_t fileno; 2755 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2756 int attrflag, fhsize, nrpcs = 0, reclen; 2757 struct nfs_fattr fattr, *fp; 2758 2759 #ifdef DIAGNOSTIC 2760 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2761 panic("nfs readdirplusrpc bad uio"); 2762 #endif 2763 ndp->ni_dvp = vp; 2764 newvp = NULLVP; 2765 2766 /* 2767 * Loop around doing readdir rpc's of size nm_readdirsize 2768 * truncated to a multiple of NFS_DIRFRAGSIZ. 2769 * The stopping criteria is EOF or buffer full. 2770 */ 2771 while (more_dirs && bigenough) { 2772 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2773 bigenough = 0; 2774 break; 2775 } 2776 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2777 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2778 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2779 nfsm_fhtom(dnp, 1); 2780 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2781 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2782 txdr_swapcookie3(uiop->uio_offset, tl); 2783 } else { 2784 txdr_cookie3(uiop->uio_offset, tl); 2785 } 2786 tl += 2; 2787 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2788 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2789 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2790 *tl = txdr_unsigned(nmp->nm_rsize); 2791 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2792 nfsm_postop_attr(vp, attrflag, 0); 2793 if (error) { 2794 m_freem(mrep); 2795 goto nfsmout; 2796 } 2797 nrpcs++; 2798 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2799 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2800 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2801 more_dirs = fxdr_unsigned(int, *tl); 2802 2803 /* loop thru the dir entries, doctoring them to 4bsd form */ 2804 while (more_dirs && bigenough) { 2805 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2806 fileno = fxdr_hyper(tl); 2807 len = fxdr_unsigned(int, *(tl + 2)); 2808 if (len <= 0 || len > NFS_MAXNAMLEN) { 2809 error = EBADRPC; 2810 m_freem(mrep); 2811 goto nfsmout; 2812 } 2813 /* for cookie stashing */ 2814 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2815 left = NFS_DIRFRAGSIZ - blksiz; 2816 if (reclen > left) { 2817 /* 2818 * DIRFRAGSIZ is aligned, no need to align 2819 * again here. 2820 */ 2821 memset(uiop->uio_iov->iov_base, 0, left); 2822 dp->d_reclen += left; 2823 UIO_ADVANCE(uiop, left); 2824 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2825 blksiz = 0; 2826 } 2827 if (reclen > uiop->uio_resid) 2828 bigenough = 0; 2829 if (bigenough) { 2830 int tlen; 2831 2832 dp = (struct dirent *)uiop->uio_iov->iov_base; 2833 dp->d_fileno = fileno; 2834 dp->d_namlen = len; 2835 dp->d_reclen = reclen; 2836 dp->d_type = DT_UNKNOWN; 2837 blksiz += reclen; 2838 if (blksiz == NFS_DIRFRAGSIZ) 2839 blksiz = 0; 2840 UIO_ADVANCE(uiop, DIRHDSIZ); 2841 nfsm_mtouio(uiop, len); 2842 tlen = reclen - (DIRHDSIZ + len); 2843 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2844 UIO_ADVANCE(uiop, tlen); 2845 cnp->cn_nameptr = dp->d_name; 2846 cnp->cn_namelen = dp->d_namlen; 2847 } else 2848 nfsm_adv(nfsm_rndup(len)); 2849 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2850 if (bigenough) { 2851 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2852 uiop->uio_offset = 2853 fxdr_swapcookie3(tl); 2854 else 2855 uiop->uio_offset = 2856 fxdr_cookie3(tl); 2857 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2858 } 2859 tl += 2; 2860 2861 /* 2862 * Since the attributes are before the file handle 2863 * (sigh), we must skip over the attributes and then 2864 * come back and get them. 2865 */ 2866 attrflag = fxdr_unsigned(int, *tl); 2867 if (attrflag) { 2868 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2869 memcpy(&fattr, fp, NFSX_V3FATTR); 2870 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2871 doit = fxdr_unsigned(int, *tl); 2872 if (doit) { 2873 nfsm_getfh(fhp, fhsize, 1); 2874 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2875 VREF(vp); 2876 newvp = vp; 2877 np = dnp; 2878 } else { 2879 error = nfs_nget1(vp->v_mount, fhp, 2880 fhsize, &np, LK_NOWAIT); 2881 if (!error) 2882 newvp = NFSTOV(np); 2883 } 2884 if (!error) { 2885 const char *xcp; 2886 2887 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2888 if (bigenough) { 2889 dp->d_type = 2890 IFTODT(VTTOIF(np->n_vattr->va_type)); 2891 if (cnp->cn_namelen <= NCHNAMLEN) { 2892 ndp->ni_vp = newvp; 2893 xcp = cnp->cn_nameptr + 2894 cnp->cn_namelen; 2895 cnp->cn_hash = 2896 namei_hash(cnp->cn_nameptr, &xcp); 2897 nfs_cache_enter(ndp->ni_dvp, 2898 ndp->ni_vp, cnp); 2899 } 2900 } 2901 } 2902 error = 0; 2903 } 2904 } else { 2905 /* Just skip over the file handle */ 2906 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2907 i = fxdr_unsigned(int, *tl); 2908 nfsm_adv(nfsm_rndup(i)); 2909 } 2910 if (newvp != NULLVP) { 2911 if (newvp == vp) 2912 vrele(newvp); 2913 else 2914 vput(newvp); 2915 newvp = NULLVP; 2916 } 2917 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2918 more_dirs = fxdr_unsigned(int, *tl); 2919 } 2920 /* 2921 * If at end of rpc data, get the eof boolean 2922 */ 2923 if (!more_dirs) { 2924 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2925 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2926 2927 /* 2928 * kludge: see a comment in nfs_readdirrpc. 2929 */ 2930 2931 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2932 more_dirs = 0; 2933 } 2934 m_freem(mrep); 2935 } 2936 /* 2937 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2938 * by increasing d_reclen for the last record. 2939 */ 2940 if (blksiz > 0) { 2941 left = NFS_DIRFRAGSIZ - blksiz; 2942 memset(uiop->uio_iov->iov_base, 0, left); 2943 dp->d_reclen += left; 2944 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2945 UIO_ADVANCE(uiop, left); 2946 } 2947 2948 /* 2949 * We are now either at the end of the directory or have filled the 2950 * block. 2951 */ 2952 if (bigenough) { 2953 dnp->n_direofoffset = uiop->uio_offset; 2954 dnp->n_flag |= NEOFVALID; 2955 } 2956 nfsmout: 2957 if (newvp != NULLVP) { 2958 if(newvp == vp) 2959 vrele(newvp); 2960 else 2961 vput(newvp); 2962 } 2963 return (error); 2964 } 2965 #endif 2966 2967 /* 2968 * Silly rename. To make the NFS filesystem that is stateless look a little 2969 * more like the "ufs" a remove of an active vnode is translated to a rename 2970 * to a funny looking filename that is removed by nfs_inactive on the 2971 * nfsnode. There is the potential for another process on a different client 2972 * to create the same funny name between the nfs_lookitup() fails and the 2973 * nfs_rename() completes, but... 2974 */ 2975 int 2976 nfs_sillyrename(dvp, vp, cnp, dolink) 2977 struct vnode *dvp, *vp; 2978 struct componentname *cnp; 2979 bool dolink; 2980 { 2981 struct sillyrename *sp; 2982 struct nfsnode *np; 2983 int error; 2984 short pid; 2985 2986 cache_purge(dvp); 2987 np = VTONFS(vp); 2988 #ifndef DIAGNOSTIC 2989 if (vp->v_type == VDIR) 2990 panic("nfs: sillyrename dir"); 2991 #endif 2992 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2993 M_NFSREQ, M_WAITOK); 2994 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2995 sp->s_dvp = dvp; 2996 VREF(dvp); 2997 2998 /* Fudge together a funny name */ 2999 pid = cnp->cn_lwp->l_proc->p_pid; 3000 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 3001 sp->s_namlen = 12; 3002 sp->s_name[8] = hexdigits[pid & 0xf]; 3003 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 3004 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 3005 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 3006 3007 /* Try lookitups until we get one that isn't there */ 3008 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3009 cnp->cn_lwp, (struct nfsnode **)0) == 0) { 3010 sp->s_name[4]++; 3011 if (sp->s_name[4] > 'z') { 3012 error = EINVAL; 3013 goto bad; 3014 } 3015 } 3016 if (dolink) { 3017 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 3018 sp->s_cred, cnp->cn_lwp); 3019 /* 3020 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 3021 */ 3022 if (error == ENOTSUP) { 3023 error = nfs_renameit(dvp, cnp, sp); 3024 } 3025 } else { 3026 error = nfs_renameit(dvp, cnp, sp); 3027 } 3028 if (error) 3029 goto bad; 3030 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3031 cnp->cn_lwp, &np); 3032 np->n_sillyrename = sp; 3033 return (0); 3034 bad: 3035 vrele(sp->s_dvp); 3036 kauth_cred_free(sp->s_cred); 3037 free((void *)sp, M_NFSREQ); 3038 return (error); 3039 } 3040 3041 /* 3042 * Look up a file name and optionally either update the file handle or 3043 * allocate an nfsnode, depending on the value of npp. 3044 * npp == NULL --> just do the lookup 3045 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 3046 * handled too 3047 * *npp != NULL --> update the file handle in the vnode 3048 */ 3049 int 3050 nfs_lookitup(dvp, name, len, cred, l, npp) 3051 struct vnode *dvp; 3052 const char *name; 3053 int len; 3054 kauth_cred_t cred; 3055 struct lwp *l; 3056 struct nfsnode **npp; 3057 { 3058 u_int32_t *tl; 3059 char *cp; 3060 int32_t t1, t2; 3061 struct vnode *newvp = (struct vnode *)0; 3062 struct nfsnode *np, *dnp = VTONFS(dvp); 3063 char *bpos, *dpos, *cp2; 3064 int error = 0, fhlen; 3065 #ifndef NFS_V2_ONLY 3066 int attrflag; 3067 #endif 3068 struct mbuf *mreq, *mrep, *md, *mb; 3069 nfsfh_t *nfhp; 3070 const int v3 = NFS_ISV3(dvp); 3071 3072 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 3073 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 3074 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 3075 nfsm_fhtom(dnp, v3); 3076 nfsm_strtom(name, len, NFS_MAXNAMLEN); 3077 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 3078 if (npp && !error) { 3079 nfsm_getfh(nfhp, fhlen, v3); 3080 if (*npp) { 3081 np = *npp; 3082 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3083 free((void *)np->n_fhp, M_NFSBIGFH); 3084 np->n_fhp = &np->n_fh; 3085 } 3086 #if NFS_SMALLFH < NFSX_V3FHMAX 3087 else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 3088 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 3089 #endif 3090 memcpy((void *)np->n_fhp, (void *)nfhp, fhlen); 3091 np->n_fhsize = fhlen; 3092 newvp = NFSTOV(np); 3093 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3094 VREF(dvp); 3095 newvp = dvp; 3096 np = dnp; 3097 } else { 3098 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 3099 if (error) { 3100 m_freem(mrep); 3101 return (error); 3102 } 3103 newvp = NFSTOV(np); 3104 } 3105 #ifndef NFS_V2_ONLY 3106 if (v3) { 3107 nfsm_postop_attr(newvp, attrflag, 0); 3108 if (!attrflag && *npp == NULL) { 3109 m_freem(mrep); 3110 vput(newvp); 3111 return (ENOENT); 3112 } 3113 } else 3114 #endif 3115 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3116 } 3117 nfsm_reqdone; 3118 if (npp && *npp == NULL) { 3119 if (error) { 3120 if (newvp) 3121 vput(newvp); 3122 } else 3123 *npp = np; 3124 } 3125 return (error); 3126 } 3127 3128 #ifndef NFS_V2_ONLY 3129 /* 3130 * Nfs Version 3 commit rpc 3131 */ 3132 int 3133 nfs_commit(vp, offset, cnt, l) 3134 struct vnode *vp; 3135 off_t offset; 3136 uint32_t cnt; 3137 struct lwp *l; 3138 { 3139 char *cp; 3140 u_int32_t *tl; 3141 int32_t t1, t2; 3142 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3143 char *bpos, *dpos, *cp2; 3144 int error = 0, wccflag = NFSV3_WCCRATTR; 3145 struct mbuf *mreq, *mrep, *md, *mb; 3146 struct nfsnode *np; 3147 3148 KASSERT(NFS_ISV3(vp)); 3149 3150 #ifdef NFS_DEBUG_COMMIT 3151 printf("commit %lu - %lu\n", (unsigned long)offset, 3152 (unsigned long)(offset + cnt)); 3153 #endif 3154 3155 mutex_enter(&nmp->nm_lock); 3156 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3157 mutex_exit(&nmp->nm_lock); 3158 return (0); 3159 } 3160 mutex_exit(&nmp->nm_lock); 3161 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3162 np = VTONFS(vp); 3163 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3164 nfsm_fhtom(np, 1); 3165 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3166 txdr_hyper(offset, tl); 3167 tl += 2; 3168 *tl = txdr_unsigned(cnt); 3169 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3170 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3171 if (!error) { 3172 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3173 mutex_enter(&nmp->nm_lock); 3174 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3175 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3176 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3177 error = NFSERR_STALEWRITEVERF; 3178 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3179 } 3180 mutex_exit(&nmp->nm_lock); 3181 } 3182 nfsm_reqdone; 3183 return (error); 3184 } 3185 #endif 3186 3187 /* 3188 * Kludge City.. 3189 * - make nfs_bmap() essentially a no-op that does no translation 3190 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3191 * (Maybe I could use the process's page mapping, but I was concerned that 3192 * Kernel Write might not be enabled and also figured copyout() would do 3193 * a lot more work than memcpy() and also it currently happens in the 3194 * context of the swapper process (2). 3195 */ 3196 int 3197 nfs_bmap(v) 3198 void *v; 3199 { 3200 struct vop_bmap_args /* { 3201 struct vnode *a_vp; 3202 daddr_t a_bn; 3203 struct vnode **a_vpp; 3204 daddr_t *a_bnp; 3205 int *a_runp; 3206 } */ *ap = v; 3207 struct vnode *vp = ap->a_vp; 3208 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3209 3210 if (ap->a_vpp != NULL) 3211 *ap->a_vpp = vp; 3212 if (ap->a_bnp != NULL) 3213 *ap->a_bnp = ap->a_bn << bshift; 3214 if (ap->a_runp != NULL) 3215 *ap->a_runp = 1024 * 1024; /* XXX */ 3216 return (0); 3217 } 3218 3219 /* 3220 * Strategy routine. 3221 * For async requests when nfsiod(s) are running, queue the request by 3222 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3223 * request. 3224 */ 3225 int 3226 nfs_strategy(v) 3227 void *v; 3228 { 3229 struct vop_strategy_args *ap = v; 3230 struct buf *bp = ap->a_bp; 3231 int error = 0; 3232 3233 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3234 panic("nfs physio/async"); 3235 3236 /* 3237 * If the op is asynchronous and an i/o daemon is waiting 3238 * queue the request, wake it up and wait for completion 3239 * otherwise just do it ourselves. 3240 */ 3241 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3242 error = nfs_doio(bp); 3243 return (error); 3244 } 3245 3246 /* 3247 * fsync vnode op. Just call nfs_flush() with commit == 1. 3248 */ 3249 /* ARGSUSED */ 3250 int 3251 nfs_fsync(v) 3252 void *v; 3253 { 3254 struct vop_fsync_args /* { 3255 struct vnodeop_desc *a_desc; 3256 struct vnode * a_vp; 3257 kauth_cred_t a_cred; 3258 int a_flags; 3259 off_t offlo; 3260 off_t offhi; 3261 struct lwp * a_l; 3262 } */ *ap = v; 3263 3264 struct vnode *vp = ap->a_vp; 3265 3266 if (vp->v_type != VREG) 3267 return 0; 3268 3269 return (nfs_flush(vp, ap->a_cred, 3270 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_l, 1)); 3271 } 3272 3273 /* 3274 * Flush all the data associated with a vnode. 3275 */ 3276 int 3277 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3278 int commit) 3279 { 3280 struct nfsnode *np = VTONFS(vp); 3281 int error; 3282 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3283 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3284 3285 simple_lock(&vp->v_interlock); 3286 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3287 if (np->n_flag & NWRITEERR) { 3288 error = np->n_error; 3289 np->n_flag &= ~NWRITEERR; 3290 } 3291 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3292 return (error); 3293 } 3294 3295 /* 3296 * Return POSIX pathconf information applicable to nfs. 3297 * 3298 * N.B. The NFS V2 protocol doesn't support this RPC. 3299 */ 3300 /* ARGSUSED */ 3301 int 3302 nfs_pathconf(v) 3303 void *v; 3304 { 3305 struct vop_pathconf_args /* { 3306 struct vnode *a_vp; 3307 int a_name; 3308 register_t *a_retval; 3309 } */ *ap = v; 3310 struct nfsv3_pathconf *pcp; 3311 struct vnode *vp = ap->a_vp; 3312 struct mbuf *mreq, *mrep, *md, *mb; 3313 int32_t t1, t2; 3314 u_int32_t *tl; 3315 char *bpos, *dpos, *cp, *cp2; 3316 int error = 0, attrflag; 3317 #ifndef NFS_V2_ONLY 3318 struct nfsmount *nmp; 3319 unsigned int l; 3320 u_int64_t maxsize; 3321 #endif 3322 const int v3 = NFS_ISV3(vp); 3323 struct nfsnode *np = VTONFS(vp); 3324 3325 switch (ap->a_name) { 3326 /* Names that can be resolved locally. */ 3327 case _PC_PIPE_BUF: 3328 *ap->a_retval = PIPE_BUF; 3329 break; 3330 case _PC_SYNC_IO: 3331 *ap->a_retval = 1; 3332 break; 3333 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3334 case _PC_LINK_MAX: 3335 case _PC_NAME_MAX: 3336 case _PC_CHOWN_RESTRICTED: 3337 case _PC_NO_TRUNC: 3338 if (!v3) { 3339 error = EINVAL; 3340 break; 3341 } 3342 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3343 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3344 nfsm_fhtom(np, 1); 3345 nfsm_request(np, NFSPROC_PATHCONF, 3346 curlwp, curlwp->l_cred); /* XXX */ 3347 nfsm_postop_attr(vp, attrflag, 0); 3348 if (!error) { 3349 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3350 NFSX_V3PATHCONF); 3351 switch (ap->a_name) { 3352 case _PC_LINK_MAX: 3353 *ap->a_retval = 3354 fxdr_unsigned(register_t, pcp->pc_linkmax); 3355 break; 3356 case _PC_NAME_MAX: 3357 *ap->a_retval = 3358 fxdr_unsigned(register_t, pcp->pc_namemax); 3359 break; 3360 case _PC_CHOWN_RESTRICTED: 3361 *ap->a_retval = 3362 (pcp->pc_chownrestricted == nfs_true); 3363 break; 3364 case _PC_NO_TRUNC: 3365 *ap->a_retval = 3366 (pcp->pc_notrunc == nfs_true); 3367 break; 3368 } 3369 } 3370 nfsm_reqdone; 3371 break; 3372 case _PC_FILESIZEBITS: 3373 #ifndef NFS_V2_ONLY 3374 if (v3) { 3375 nmp = VFSTONFS(vp->v_mount); 3376 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3377 if ((error = nfs_fsinfo(nmp, vp, 3378 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3379 break; 3380 for (l = 0, maxsize = nmp->nm_maxfilesize; 3381 (maxsize >> l) > 0; l++) 3382 ; 3383 *ap->a_retval = l + 1; 3384 } else 3385 #endif 3386 { 3387 *ap->a_retval = 32; /* NFS V2 limitation */ 3388 } 3389 break; 3390 default: 3391 error = EINVAL; 3392 break; 3393 } 3394 3395 return (error); 3396 } 3397 3398 /* 3399 * NFS advisory byte-level locks. 3400 */ 3401 int 3402 nfs_advlock(v) 3403 void *v; 3404 { 3405 struct vop_advlock_args /* { 3406 struct vnode *a_vp; 3407 void *a_id; 3408 int a_op; 3409 struct flock *a_fl; 3410 int a_flags; 3411 } */ *ap = v; 3412 struct nfsnode *np = VTONFS(ap->a_vp); 3413 3414 return lf_advlock(ap, &np->n_lockf, np->n_size); 3415 } 3416 3417 /* 3418 * Print out the contents of an nfsnode. 3419 */ 3420 int 3421 nfs_print(v) 3422 void *v; 3423 { 3424 struct vop_print_args /* { 3425 struct vnode *a_vp; 3426 } */ *ap = v; 3427 struct vnode *vp = ap->a_vp; 3428 struct nfsnode *np = VTONFS(vp); 3429 3430 printf("tag VT_NFS, fileid %lld fsid 0x%lx", 3431 (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3432 if (vp->v_type == VFIFO) 3433 fifo_printinfo(vp); 3434 printf("\n"); 3435 return (0); 3436 } 3437 3438 /* 3439 * nfs unlock wrapper. 3440 */ 3441 int 3442 nfs_unlock(void *v) 3443 { 3444 struct vop_unlock_args /* { 3445 struct vnode *a_vp; 3446 int a_flags; 3447 } */ *ap = v; 3448 struct vnode *vp = ap->a_vp; 3449 3450 /* 3451 * VOP_UNLOCK can be called by nfs_loadattrcache 3452 * with v_data == 0. 3453 */ 3454 if (VTONFS(vp)) { 3455 nfs_delayedtruncate(vp); 3456 } 3457 3458 return genfs_unlock(v); 3459 } 3460 3461 /* 3462 * nfs special file access vnode op. 3463 * Essentially just get vattr and then imitate iaccess() since the device is 3464 * local to the client. 3465 */ 3466 int 3467 nfsspec_access(v) 3468 void *v; 3469 { 3470 struct vop_access_args /* { 3471 struct vnode *a_vp; 3472 int a_mode; 3473 kauth_cred_t a_cred; 3474 struct lwp *a_l; 3475 } */ *ap = v; 3476 struct vattr va; 3477 struct vnode *vp = ap->a_vp; 3478 int error; 3479 3480 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_l); 3481 if (error) 3482 return (error); 3483 3484 /* 3485 * Disallow write attempts on filesystems mounted read-only; 3486 * unless the file is a socket, fifo, or a block or character 3487 * device resident on the filesystem. 3488 */ 3489 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3490 switch (vp->v_type) { 3491 case VREG: 3492 case VDIR: 3493 case VLNK: 3494 return (EROFS); 3495 default: 3496 break; 3497 } 3498 } 3499 3500 return (vaccess(va.va_type, va.va_mode, 3501 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3502 } 3503 3504 /* 3505 * Read wrapper for special devices. 3506 */ 3507 int 3508 nfsspec_read(v) 3509 void *v; 3510 { 3511 struct vop_read_args /* { 3512 struct vnode *a_vp; 3513 struct uio *a_uio; 3514 int a_ioflag; 3515 kauth_cred_t a_cred; 3516 } */ *ap = v; 3517 struct nfsnode *np = VTONFS(ap->a_vp); 3518 3519 /* 3520 * Set access flag. 3521 */ 3522 np->n_flag |= NACC; 3523 getnanotime(&np->n_atim); 3524 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3525 } 3526 3527 /* 3528 * Write wrapper for special devices. 3529 */ 3530 int 3531 nfsspec_write(v) 3532 void *v; 3533 { 3534 struct vop_write_args /* { 3535 struct vnode *a_vp; 3536 struct uio *a_uio; 3537 int a_ioflag; 3538 kauth_cred_t a_cred; 3539 } */ *ap = v; 3540 struct nfsnode *np = VTONFS(ap->a_vp); 3541 3542 /* 3543 * Set update flag. 3544 */ 3545 np->n_flag |= NUPD; 3546 getnanotime(&np->n_mtim); 3547 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3548 } 3549 3550 /* 3551 * Close wrapper for special devices. 3552 * 3553 * Update the times on the nfsnode then do device close. 3554 */ 3555 int 3556 nfsspec_close(v) 3557 void *v; 3558 { 3559 struct vop_close_args /* { 3560 struct vnode *a_vp; 3561 int a_fflag; 3562 kauth_cred_t a_cred; 3563 struct lwp *a_l; 3564 } */ *ap = v; 3565 struct vnode *vp = ap->a_vp; 3566 struct nfsnode *np = VTONFS(vp); 3567 struct vattr vattr; 3568 3569 if (np->n_flag & (NACC | NUPD)) { 3570 np->n_flag |= NCHG; 3571 if (vp->v_usecount == 1 && 3572 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3573 VATTR_NULL(&vattr); 3574 if (np->n_flag & NACC) 3575 vattr.va_atime = np->n_atim; 3576 if (np->n_flag & NUPD) 3577 vattr.va_mtime = np->n_mtim; 3578 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l); 3579 } 3580 } 3581 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3582 } 3583 3584 /* 3585 * Read wrapper for fifos. 3586 */ 3587 int 3588 nfsfifo_read(v) 3589 void *v; 3590 { 3591 struct vop_read_args /* { 3592 struct vnode *a_vp; 3593 struct uio *a_uio; 3594 int a_ioflag; 3595 kauth_cred_t a_cred; 3596 } */ *ap = v; 3597 struct nfsnode *np = VTONFS(ap->a_vp); 3598 3599 /* 3600 * Set access flag. 3601 */ 3602 np->n_flag |= NACC; 3603 getnanotime(&np->n_atim); 3604 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3605 } 3606 3607 /* 3608 * Write wrapper for fifos. 3609 */ 3610 int 3611 nfsfifo_write(v) 3612 void *v; 3613 { 3614 struct vop_write_args /* { 3615 struct vnode *a_vp; 3616 struct uio *a_uio; 3617 int a_ioflag; 3618 kauth_cred_t a_cred; 3619 } */ *ap = v; 3620 struct nfsnode *np = VTONFS(ap->a_vp); 3621 3622 /* 3623 * Set update flag. 3624 */ 3625 np->n_flag |= NUPD; 3626 getnanotime(&np->n_mtim); 3627 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3628 } 3629 3630 /* 3631 * Close wrapper for fifos. 3632 * 3633 * Update the times on the nfsnode then do fifo close. 3634 */ 3635 int 3636 nfsfifo_close(v) 3637 void *v; 3638 { 3639 struct vop_close_args /* { 3640 struct vnode *a_vp; 3641 int a_fflag; 3642 kauth_cred_t a_cred; 3643 struct lwp *a_l; 3644 } */ *ap = v; 3645 struct vnode *vp = ap->a_vp; 3646 struct nfsnode *np = VTONFS(vp); 3647 struct vattr vattr; 3648 3649 if (np->n_flag & (NACC | NUPD)) { 3650 struct timespec ts; 3651 3652 getnanotime(&ts); 3653 if (np->n_flag & NACC) 3654 np->n_atim = ts; 3655 if (np->n_flag & NUPD) 3656 np->n_mtim = ts; 3657 np->n_flag |= NCHG; 3658 if (vp->v_usecount == 1 && 3659 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3660 VATTR_NULL(&vattr); 3661 if (np->n_flag & NACC) 3662 vattr.va_atime = np->n_atim; 3663 if (np->n_flag & NUPD) 3664 vattr.va_mtime = np->n_mtim; 3665 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l); 3666 } 3667 } 3668 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3669 } 3670