nfs_vnops.c revision 1.303
1 /* $NetBSD: nfs_vnops.c,v 1.303 2014/01/23 10:13:57 hannken 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 1.303 2014/01/23 10:13:57 hannken Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/proc.h> 51 #include <sys/kernel.h> 52 #include <sys/systm.h> 53 #include <sys/resourcevar.h> 54 #include <sys/mount.h> 55 #include <sys/buf.h> 56 #include <sys/condvar.h> 57 #include <sys/disk.h> 58 #include <sys/malloc.h> 59 #include <sys/kmem.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/vnode.h> 64 #include <sys/dirent.h> 65 #include <sys/fcntl.h> 66 #include <sys/hash.h> 67 #include <sys/lockf.h> 68 #include <sys/stat.h> 69 #include <sys/unistd.h> 70 #include <sys/kauth.h> 71 #include <sys/cprng.h> 72 73 #include <uvm/uvm_extern.h> 74 #include <uvm/uvm.h> 75 76 #include <miscfs/fifofs/fifo.h> 77 #include <miscfs/genfs/genfs.h> 78 #include <miscfs/genfs/genfs_node.h> 79 #include <miscfs/specfs/specdev.h> 80 81 #include <nfs/rpcv2.h> 82 #include <nfs/nfsproto.h> 83 #include <nfs/nfs.h> 84 #include <nfs/nfsnode.h> 85 #include <nfs/nfsmount.h> 86 #include <nfs/xdr_subs.h> 87 #include <nfs/nfsm_subs.h> 88 #include <nfs/nfs_var.h> 89 90 #include <net/if.h> 91 #include <netinet/in.h> 92 #include <netinet/in_var.h> 93 94 /* 95 * Global vfs data structures for nfs 96 */ 97 int (**nfsv2_vnodeop_p)(void *); 98 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 99 { &vop_default_desc, vn_default_error }, 100 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 101 { &vop_create_desc, nfs_create }, /* create */ 102 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 103 { &vop_open_desc, nfs_open }, /* open */ 104 { &vop_close_desc, nfs_close }, /* close */ 105 { &vop_access_desc, nfs_access }, /* access */ 106 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 107 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 108 { &vop_read_desc, nfs_read }, /* read */ 109 { &vop_write_desc, nfs_write }, /* write */ 110 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 111 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 112 { &vop_poll_desc, nfs_poll }, /* poll */ 113 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 114 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 115 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 116 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 117 { &vop_seek_desc, nfs_seek }, /* seek */ 118 { &vop_remove_desc, nfs_remove }, /* remove */ 119 { &vop_link_desc, nfs_link }, /* link */ 120 { &vop_rename_desc, nfs_rename }, /* rename */ 121 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 122 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 123 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 124 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 125 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 126 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 127 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 128 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 129 { &vop_lock_desc, nfs_lock }, /* lock */ 130 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 131 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 132 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 133 { &vop_print_desc, nfs_print }, /* print */ 134 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 135 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 136 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 137 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 138 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 139 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 140 { NULL, NULL } 141 }; 142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 144 145 /* 146 * Special device vnode ops 147 */ 148 int (**spec_nfsv2nodeop_p)(void *); 149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 150 { &vop_default_desc, vn_default_error }, 151 { &vop_lookup_desc, spec_lookup }, /* lookup */ 152 { &vop_create_desc, spec_create }, /* create */ 153 { &vop_mknod_desc, spec_mknod }, /* mknod */ 154 { &vop_open_desc, spec_open }, /* open */ 155 { &vop_close_desc, nfsspec_close }, /* close */ 156 { &vop_access_desc, nfsspec_access }, /* access */ 157 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 158 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 159 { &vop_read_desc, nfsspec_read }, /* read */ 160 { &vop_write_desc, nfsspec_write }, /* write */ 161 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 162 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 163 { &vop_poll_desc, spec_poll }, /* poll */ 164 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 165 { &vop_revoke_desc, spec_revoke }, /* revoke */ 166 { &vop_mmap_desc, spec_mmap }, /* mmap */ 167 { &vop_fsync_desc, spec_fsync }, /* fsync */ 168 { &vop_seek_desc, spec_seek }, /* seek */ 169 { &vop_remove_desc, spec_remove }, /* remove */ 170 { &vop_link_desc, spec_link }, /* link */ 171 { &vop_rename_desc, spec_rename }, /* rename */ 172 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 173 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 174 { &vop_symlink_desc, spec_symlink }, /* symlink */ 175 { &vop_readdir_desc, spec_readdir }, /* readdir */ 176 { &vop_readlink_desc, spec_readlink }, /* readlink */ 177 { &vop_abortop_desc, spec_abortop }, /* abortop */ 178 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 179 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 180 { &vop_lock_desc, nfs_lock }, /* lock */ 181 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 182 { &vop_bmap_desc, spec_bmap }, /* bmap */ 183 { &vop_strategy_desc, spec_strategy }, /* strategy */ 184 { &vop_print_desc, nfs_print }, /* print */ 185 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 186 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 187 { &vop_advlock_desc, spec_advlock }, /* advlock */ 188 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 189 { &vop_getpages_desc, spec_getpages }, /* getpages */ 190 { &vop_putpages_desc, spec_putpages }, /* putpages */ 191 { NULL, NULL } 192 }; 193 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 194 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 195 196 int (**fifo_nfsv2nodeop_p)(void *); 197 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 198 { &vop_default_desc, vn_default_error }, 199 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 200 { &vop_create_desc, vn_fifo_bypass }, /* create */ 201 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 202 { &vop_open_desc, vn_fifo_bypass }, /* open */ 203 { &vop_close_desc, nfsfifo_close }, /* close */ 204 { &vop_access_desc, nfsspec_access }, /* access */ 205 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 206 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 207 { &vop_read_desc, nfsfifo_read }, /* read */ 208 { &vop_write_desc, nfsfifo_write }, /* write */ 209 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 210 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 211 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 212 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 213 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 214 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 215 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 216 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 217 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 218 { &vop_link_desc, vn_fifo_bypass }, /* link */ 219 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 220 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 221 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 222 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 223 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 224 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 225 { &vop_abortop_desc, vn_fifo_bypass }, /* 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, vn_fifo_bypass }, /* 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, vn_fifo_bypass }, /* pathconf */ 235 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 236 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 237 { &vop_putpages_desc, vn_fifo_bypass }, /* 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 ((cnp->cn_flags & MAKEENTRY) == 0) { 271 return; 272 } 273 if (vp != NULL) { 274 struct nfsnode *np = VTONFS(vp); 275 276 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 277 } 278 279 if (!timespecisset(&dnp->n_nctime)) 280 dnp->n_nctime = dnp->n_vattr->va_mtime; 281 282 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags); 283 } 284 285 /* 286 * nfs null call from vfs. 287 */ 288 int 289 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 290 { 291 char *bpos, *dpos; 292 int error = 0; 293 struct mbuf *mreq, *mrep, *md, *mb __unused; 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(void *v) 310 { 311 struct vop_access_args /* { 312 struct vnode *a_vp; 313 int a_mode; 314 kauth_cred_t a_cred; 315 } */ *ap = v; 316 struct vnode *vp = ap->a_vp; 317 #ifndef NFS_V2_ONLY 318 u_int32_t *tl; 319 char *cp; 320 int32_t t1, t2; 321 char *bpos, *dpos, *cp2; 322 int error = 0, attrflag; 323 struct mbuf *mreq, *mrep, *md, *mb; 324 u_int32_t mode, rmode; 325 const int v3 = NFS_ISV3(vp); 326 #endif 327 int cachevalid; 328 struct nfsnode *np = VTONFS(vp); 329 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 330 331 cachevalid = (np->n_accstamp != -1 && 332 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 333 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 334 335 /* 336 * Check access cache first. If this request has been made for this 337 * uid shortly before, use the cached result. 338 */ 339 if (cachevalid) { 340 if (!np->n_accerror) { 341 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 342 return np->n_accerror; 343 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 344 return np->n_accerror; 345 } 346 347 #ifndef NFS_V2_ONLY 348 /* 349 * For nfs v3, do an access rpc, otherwise you are stuck emulating 350 * ufs_access() locally using the vattr. This may not be correct, 351 * since the server may apply other access criteria such as 352 * client uid-->server uid mapping that we do not know about, but 353 * this is better than just returning anything that is lying about 354 * in the cache. 355 */ 356 if (v3) { 357 nfsstats.rpccnt[NFSPROC_ACCESS]++; 358 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 359 nfsm_fhtom(np, v3); 360 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 361 if (ap->a_mode & VREAD) 362 mode = NFSV3ACCESS_READ; 363 else 364 mode = 0; 365 if (vp->v_type != VDIR) { 366 if (ap->a_mode & VWRITE) 367 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 368 if (ap->a_mode & VEXEC) 369 mode |= NFSV3ACCESS_EXECUTE; 370 } else { 371 if (ap->a_mode & VWRITE) 372 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 373 NFSV3ACCESS_DELETE); 374 if (ap->a_mode & VEXEC) 375 mode |= NFSV3ACCESS_LOOKUP; 376 } 377 *tl = txdr_unsigned(mode); 378 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 379 nfsm_postop_attr(vp, attrflag, 0); 380 if (!error) { 381 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 382 rmode = fxdr_unsigned(u_int32_t, *tl); 383 /* 384 * The NFS V3 spec does not clarify whether or not 385 * the returned access bits can be a superset of 386 * the ones requested, so... 387 */ 388 if ((rmode & mode) != mode) 389 error = EACCES; 390 } 391 nfsm_reqdone; 392 } else 393 #endif 394 return (nfsspec_access(ap)); 395 #ifndef NFS_V2_ONLY 396 /* 397 * Disallow write attempts on filesystems mounted read-only; 398 * unless the file is a socket, fifo, or a block or character 399 * device resident on the filesystem. 400 */ 401 if (!error && (ap->a_mode & VWRITE) && 402 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 403 switch (vp->v_type) { 404 case VREG: 405 case VDIR: 406 case VLNK: 407 error = EROFS; 408 default: 409 break; 410 } 411 } 412 413 if (!error || error == EACCES) { 414 /* 415 * If we got the same result as for a previous, 416 * different request, OR it in. Don't update 417 * the timestamp in that case. 418 */ 419 if (cachevalid && np->n_accstamp != -1 && 420 error == np->n_accerror) { 421 if (!error) 422 np->n_accmode |= ap->a_mode; 423 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 424 np->n_accmode = ap->a_mode; 425 } else { 426 np->n_accstamp = time_uptime; 427 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 428 np->n_accmode = ap->a_mode; 429 np->n_accerror = error; 430 } 431 } 432 433 return (error); 434 #endif 435 } 436 437 /* 438 * nfs open vnode op 439 * Check to see if the type is ok 440 * and that deletion is not in progress. 441 * For paged in text files, you will need to flush the page cache 442 * if consistency is lost. 443 */ 444 /* ARGSUSED */ 445 int 446 nfs_open(void *v) 447 { 448 struct vop_open_args /* { 449 struct vnode *a_vp; 450 int a_mode; 451 kauth_cred_t a_cred; 452 } */ *ap = v; 453 struct vnode *vp = ap->a_vp; 454 struct nfsnode *np = VTONFS(vp); 455 int error; 456 457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 458 return (EACCES); 459 } 460 461 if (ap->a_mode & FREAD) { 462 if (np->n_rcred != NULL) 463 kauth_cred_free(np->n_rcred); 464 np->n_rcred = ap->a_cred; 465 kauth_cred_hold(np->n_rcred); 466 } 467 if (ap->a_mode & FWRITE) { 468 if (np->n_wcred != NULL) 469 kauth_cred_free(np->n_wcred); 470 np->n_wcred = ap->a_cred; 471 kauth_cred_hold(np->n_wcred); 472 } 473 474 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 475 if (error) 476 return error; 477 478 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 479 480 return (0); 481 } 482 483 /* 484 * nfs close vnode op 485 * What an NFS client should do upon close after writing is a debatable issue. 486 * Most NFS clients push delayed writes to the server upon close, basically for 487 * two reasons: 488 * 1 - So that any write errors may be reported back to the client process 489 * doing the close system call. By far the two most likely errors are 490 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 491 * 2 - To put a worst case upper bound on cache inconsistency between 492 * multiple clients for the file. 493 * There is also a consistency problem for Version 2 of the protocol w.r.t. 494 * not being able to tell if other clients are writing a file concurrently, 495 * since there is no way of knowing if the changed modify time in the reply 496 * is only due to the write for this client. 497 * (NFS Version 3 provides weak cache consistency data in the reply that 498 * should be sufficient to detect and handle this case.) 499 * 500 * The current code does the following: 501 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 502 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 503 * or commit them (this satisfies 1 and 2 except for the 504 * case where the server crashes after this close but 505 * before the commit RPC, which is felt to be "good 506 * enough". Changing the last argument to nfs_flush() to 507 * a 1 would force a commit operation, if it is felt a 508 * commit is necessary now. 509 */ 510 /* ARGSUSED */ 511 int 512 nfs_close(void *v) 513 { 514 struct vop_close_args /* { 515 struct vnodeop_desc *a_desc; 516 struct vnode *a_vp; 517 int a_fflag; 518 kauth_cred_t a_cred; 519 } */ *ap = v; 520 struct vnode *vp = ap->a_vp; 521 struct nfsnode *np = VTONFS(vp); 522 int error = 0; 523 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 524 525 if (vp->v_type == VREG) { 526 if (np->n_flag & NMODIFIED) { 527 #ifndef NFS_V2_ONLY 528 if (NFS_ISV3(vp)) { 529 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 530 np->n_flag &= ~NMODIFIED; 531 } else 532 #endif 533 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 534 NFS_INVALIDATE_ATTRCACHE(np); 535 } 536 if (np->n_flag & NWRITEERR) { 537 np->n_flag &= ~NWRITEERR; 538 error = np->n_error; 539 } 540 } 541 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 542 return (error); 543 } 544 545 /* 546 * nfs getattr call from vfs. 547 */ 548 int 549 nfs_getattr(void *v) 550 { 551 struct vop_getattr_args /* { 552 struct vnode *a_vp; 553 struct vattr *a_vap; 554 kauth_cred_t a_cred; 555 } */ *ap = v; 556 struct vnode *vp = ap->a_vp; 557 struct nfsnode *np = VTONFS(vp); 558 char *cp; 559 u_int32_t *tl; 560 int32_t t1, t2; 561 char *bpos, *dpos; 562 int error = 0; 563 struct mbuf *mreq, *mrep, *md, *mb; 564 const int v3 = NFS_ISV3(vp); 565 566 /* 567 * Update local times for special files. 568 */ 569 if (np->n_flag & (NACC | NUPD)) 570 np->n_flag |= NCHG; 571 572 /* 573 * if we have delayed truncation, do it now. 574 */ 575 nfs_delayedtruncate(vp); 576 577 /* 578 * First look in the cache. 579 */ 580 if (nfs_getattrcache(vp, ap->a_vap) == 0) 581 return (0); 582 nfsstats.rpccnt[NFSPROC_GETATTR]++; 583 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 584 nfsm_fhtom(np, v3); 585 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 586 if (!error) { 587 nfsm_loadattr(vp, ap->a_vap, 0); 588 if (vp->v_type == VDIR && 589 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 590 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 591 } 592 nfsm_reqdone; 593 return (error); 594 } 595 596 /* 597 * nfs setattr call. 598 */ 599 int 600 nfs_setattr(void *v) 601 { 602 struct vop_setattr_args /* { 603 struct vnodeop_desc *a_desc; 604 struct vnode *a_vp; 605 struct vattr *a_vap; 606 kauth_cred_t a_cred; 607 } */ *ap = v; 608 struct vnode *vp = ap->a_vp; 609 struct nfsnode *np = VTONFS(vp); 610 struct vattr *vap = ap->a_vap; 611 int error = 0; 612 u_quad_t tsize = 0; 613 614 /* 615 * Setting of flags is not supported. 616 */ 617 if (vap->va_flags != VNOVAL) 618 return (EOPNOTSUPP); 619 620 /* 621 * Disallow write attempts if the filesystem is mounted read-only. 622 */ 623 if ((vap->va_uid != (uid_t)VNOVAL || 624 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 625 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 626 (vp->v_mount->mnt_flag & MNT_RDONLY)) 627 return (EROFS); 628 if (vap->va_size != VNOVAL) { 629 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 630 return EFBIG; 631 } 632 switch (vp->v_type) { 633 case VDIR: 634 return (EISDIR); 635 case VCHR: 636 case VBLK: 637 case VSOCK: 638 case VFIFO: 639 if (vap->va_mtime.tv_sec == VNOVAL && 640 vap->va_atime.tv_sec == VNOVAL && 641 vap->va_mode == (mode_t)VNOVAL && 642 vap->va_uid == (uid_t)VNOVAL && 643 vap->va_gid == (gid_t)VNOVAL) 644 return (0); 645 vap->va_size = VNOVAL; 646 break; 647 default: 648 /* 649 * Disallow write attempts if the filesystem is 650 * mounted read-only. 651 */ 652 if (vp->v_mount->mnt_flag & MNT_RDONLY) 653 return (EROFS); 654 genfs_node_wrlock(vp); 655 uvm_vnp_setsize(vp, vap->va_size); 656 tsize = np->n_size; 657 np->n_size = vap->va_size; 658 if (vap->va_size == 0) 659 error = nfs_vinvalbuf(vp, 0, 660 ap->a_cred, curlwp, 1); 661 else 662 error = nfs_vinvalbuf(vp, V_SAVE, 663 ap->a_cred, curlwp, 1); 664 if (error) { 665 uvm_vnp_setsize(vp, tsize); 666 genfs_node_unlock(vp); 667 return (error); 668 } 669 np->n_vattr->va_size = vap->va_size; 670 } 671 } else { 672 /* 673 * flush files before setattr because a later write of 674 * cached data might change timestamps or reset sugid bits 675 */ 676 if ((vap->va_mtime.tv_sec != VNOVAL || 677 vap->va_atime.tv_sec != VNOVAL || 678 vap->va_mode != VNOVAL) && 679 vp->v_type == VREG && 680 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 681 curlwp, 1)) == EINTR) 682 return (error); 683 } 684 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 685 if (vap->va_size != VNOVAL) { 686 if (error) { 687 np->n_size = np->n_vattr->va_size = tsize; 688 uvm_vnp_setsize(vp, np->n_size); 689 } 690 genfs_node_unlock(vp); 691 } 692 VN_KNOTE(vp, NOTE_ATTRIB); 693 return (error); 694 } 695 696 /* 697 * Do an nfs setattr rpc. 698 */ 699 int 700 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 701 { 702 struct nfsv2_sattr *sp; 703 char *cp; 704 int32_t t1, t2; 705 char *bpos, *dpos; 706 u_int32_t *tl; 707 int error = 0; 708 struct mbuf *mreq, *mrep, *md, *mb; 709 const int v3 = NFS_ISV3(vp); 710 struct nfsnode *np = VTONFS(vp); 711 #ifndef NFS_V2_ONLY 712 int wccflag = NFSV3_WCCRATTR; 713 char *cp2; 714 #endif 715 716 nfsstats.rpccnt[NFSPROC_SETATTR]++; 717 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 718 nfsm_fhtom(np, v3); 719 #ifndef NFS_V2_ONLY 720 if (v3) { 721 nfsm_v3attrbuild(vap, true); 722 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 723 *tl = nfs_false; 724 } else { 725 #endif 726 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 727 if (vap->va_mode == (mode_t)VNOVAL) 728 sp->sa_mode = nfs_xdrneg1; 729 else 730 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 731 if (vap->va_uid == (uid_t)VNOVAL) 732 sp->sa_uid = nfs_xdrneg1; 733 else 734 sp->sa_uid = txdr_unsigned(vap->va_uid); 735 if (vap->va_gid == (gid_t)VNOVAL) 736 sp->sa_gid = nfs_xdrneg1; 737 else 738 sp->sa_gid = txdr_unsigned(vap->va_gid); 739 sp->sa_size = txdr_unsigned(vap->va_size); 740 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 741 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 742 #ifndef NFS_V2_ONLY 743 } 744 #endif 745 nfsm_request(np, NFSPROC_SETATTR, l, cred); 746 #ifndef NFS_V2_ONLY 747 if (v3) { 748 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 749 } else 750 #endif 751 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 752 nfsm_reqdone; 753 return (error); 754 } 755 756 /* 757 * nfs lookup call, one step at a time... 758 * First look in cache 759 * If not found, do the rpc. 760 */ 761 int 762 nfs_lookup(void *v) 763 { 764 struct vop_lookup_args /* { 765 struct vnodeop_desc *a_desc; 766 struct vnode *a_dvp; 767 struct vnode **a_vpp; 768 struct componentname *a_cnp; 769 } */ *ap = v; 770 struct componentname *cnp = ap->a_cnp; 771 struct vnode *dvp = ap->a_dvp; 772 struct vnode **vpp = ap->a_vpp; 773 int flags; 774 struct vnode *newvp; 775 u_int32_t *tl; 776 char *cp; 777 int32_t t1, t2; 778 char *bpos, *dpos, *cp2; 779 struct mbuf *mreq, *mrep, *md, *mb; 780 long len; 781 nfsfh_t *fhp; 782 struct nfsnode *np; 783 int cachefound; 784 int error = 0, attrflag, fhsize; 785 const int v3 = NFS_ISV3(dvp); 786 787 flags = cnp->cn_flags; 788 789 *vpp = NULLVP; 790 newvp = NULLVP; 791 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 792 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 793 return (EROFS); 794 if (dvp->v_type != VDIR) 795 return (ENOTDIR); 796 797 /* 798 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 799 */ 800 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 801 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 802 if (error) 803 return error; 804 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 805 return EISDIR; 806 vref(dvp); 807 *vpp = dvp; 808 return 0; 809 } 810 811 np = VTONFS(dvp); 812 813 /* 814 * Before performing an RPC, check the name cache to see if 815 * the directory/name pair we are looking for is known already. 816 * If the directory/name pair is found in the name cache, 817 * we have to ensure the directory has not changed from 818 * the time the cache entry has been created. If it has, 819 * the cache entry has to be ignored. 820 */ 821 cachefound = cache_lookup_raw(dvp, cnp->cn_nameptr, cnp->cn_namelen, 822 cnp->cn_flags, NULL, vpp); 823 KASSERT(dvp != *vpp); 824 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 825 if (cachefound) { 826 struct vattr vattr; 827 828 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 829 if (error != 0) { 830 if (*vpp != NULLVP) 831 vrele(*vpp); 832 *vpp = NULLVP; 833 return error; 834 } 835 836 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 837 || timespeccmp(&vattr.va_mtime, 838 &VTONFS(dvp)->n_nctime, !=)) { 839 if (*vpp != NULLVP) { 840 vrele(*vpp); 841 *vpp = NULLVP; 842 } 843 cache_purge1(dvp, NULL, 0, PURGE_CHILDREN); 844 timespecclear(&np->n_nctime); 845 goto dorpc; 846 } 847 848 if (*vpp == NULLVP) { 849 /* namecache gave us a negative result */ 850 error = ENOENT; 851 goto noentry; 852 } 853 854 /* 855 * investigate the vnode returned by cache_lookup_raw. 856 * if it isn't appropriate, do an rpc. 857 */ 858 newvp = *vpp; 859 if ((flags & ISDOTDOT) != 0) { 860 VOP_UNLOCK(dvp); 861 } 862 error = vn_lock(newvp, LK_EXCLUSIVE); 863 if ((flags & ISDOTDOT) != 0) { 864 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 865 } 866 if (error != 0) { 867 /* newvp has been reclaimed. */ 868 vrele(newvp); 869 *vpp = NULLVP; 870 goto dorpc; 871 } 872 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 873 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 874 nfsstats.lookupcache_hits++; 875 KASSERT(newvp->v_type != VNON); 876 return (0); 877 } 878 cache_purge1(newvp, NULL, 0, PURGE_PARENTS); 879 vput(newvp); 880 *vpp = NULLVP; 881 } 882 dorpc: 883 #if 0 884 /* 885 * because nfsv3 has the same CREATE semantics as ours, 886 * we don't have to perform LOOKUPs beforehand. 887 * 888 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 889 * XXX although we have no way to know if O_EXCL is requested or not. 890 */ 891 892 if (v3 && cnp->cn_nameiop == CREATE && 893 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 894 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 895 return (EJUSTRETURN); 896 } 897 #endif /* 0 */ 898 899 error = 0; 900 newvp = NULLVP; 901 nfsstats.lookupcache_misses++; 902 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 903 len = cnp->cn_namelen; 904 nfsm_reqhead(np, NFSPROC_LOOKUP, 905 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 906 nfsm_fhtom(np, v3); 907 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 908 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 909 if (error) { 910 nfsm_postop_attr(dvp, attrflag, 0); 911 m_freem(mrep); 912 goto nfsmout; 913 } 914 nfsm_getfh(fhp, fhsize, v3); 915 916 /* 917 * Handle RENAME case... 918 */ 919 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 920 if (NFS_CMPFH(np, fhp, fhsize)) { 921 m_freem(mrep); 922 return (EISDIR); 923 } 924 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 925 if (error) { 926 m_freem(mrep); 927 return error; 928 } 929 newvp = NFSTOV(np); 930 #ifndef NFS_V2_ONLY 931 if (v3) { 932 nfsm_postop_attr(newvp, attrflag, 0); 933 nfsm_postop_attr(dvp, attrflag, 0); 934 } else 935 #endif 936 nfsm_loadattr(newvp, (struct vattr *)0, 0); 937 *vpp = newvp; 938 m_freem(mrep); 939 goto validate; 940 } 941 942 /* 943 * The postop attr handling is duplicated for each if case, 944 * because it should be done while dvp is locked (unlocking 945 * dvp is different for each case). 946 */ 947 948 if (NFS_CMPFH(np, fhp, fhsize)) { 949 /* 950 * as we handle "." lookup locally, this should be 951 * a broken server. 952 */ 953 vref(dvp); 954 newvp = dvp; 955 #ifndef NFS_V2_ONLY 956 if (v3) { 957 nfsm_postop_attr(newvp, attrflag, 0); 958 nfsm_postop_attr(dvp, attrflag, 0); 959 } else 960 #endif 961 nfsm_loadattr(newvp, (struct vattr *)0, 0); 962 } else if (flags & ISDOTDOT) { 963 /* 964 * ".." lookup 965 */ 966 VOP_UNLOCK(dvp); 967 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 968 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 969 if (error) { 970 m_freem(mrep); 971 return error; 972 } 973 newvp = NFSTOV(np); 974 975 #ifndef NFS_V2_ONLY 976 if (v3) { 977 nfsm_postop_attr(newvp, attrflag, 0); 978 nfsm_postop_attr(dvp, attrflag, 0); 979 } else 980 #endif 981 nfsm_loadattr(newvp, (struct vattr *)0, 0); 982 } else { 983 /* 984 * Other lookups. 985 */ 986 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 987 if (error) { 988 m_freem(mrep); 989 return error; 990 } 991 newvp = NFSTOV(np); 992 #ifndef NFS_V2_ONLY 993 if (v3) { 994 nfsm_postop_attr(newvp, attrflag, 0); 995 nfsm_postop_attr(dvp, attrflag, 0); 996 } else 997 #endif 998 nfsm_loadattr(newvp, (struct vattr *)0, 0); 999 } 1000 if (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) { 1001 nfs_cache_enter(dvp, newvp, cnp); 1002 } 1003 *vpp = newvp; 1004 nfsm_reqdone; 1005 if (error) { 1006 /* 1007 * We get here only because of errors returned by 1008 * the RPC. Otherwise we'll have returned above 1009 * (the nfsm_* macros will jump to nfsm_reqdone 1010 * on error). 1011 */ 1012 if (error == ENOENT && cnp->cn_nameiop != CREATE) { 1013 nfs_cache_enter(dvp, NULL, cnp); 1014 } 1015 if (newvp != NULLVP) { 1016 if (newvp == dvp) { 1017 vrele(newvp); 1018 } else { 1019 vput(newvp); 1020 } 1021 } 1022 noentry: 1023 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1024 (flags & ISLASTCN) && error == ENOENT) { 1025 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1026 error = EROFS; 1027 } else { 1028 error = EJUSTRETURN; 1029 } 1030 } 1031 *vpp = NULL; 1032 return error; 1033 } 1034 1035 validate: 1036 /* 1037 * make sure we have valid type and size. 1038 */ 1039 1040 newvp = *vpp; 1041 if (newvp->v_type == VNON) { 1042 struct vattr vattr; /* dummy */ 1043 1044 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1045 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1046 if (error) { 1047 vput(newvp); 1048 *vpp = NULL; 1049 } 1050 } 1051 1052 return error; 1053 } 1054 1055 /* 1056 * nfs read call. 1057 * Just call nfs_bioread() to do the work. 1058 */ 1059 int 1060 nfs_read(void *v) 1061 { 1062 struct vop_read_args /* { 1063 struct vnode *a_vp; 1064 struct uio *a_uio; 1065 int a_ioflag; 1066 kauth_cred_t a_cred; 1067 } */ *ap = v; 1068 struct vnode *vp = ap->a_vp; 1069 1070 if (vp->v_type != VREG) 1071 return EISDIR; 1072 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1073 } 1074 1075 /* 1076 * nfs readlink call 1077 */ 1078 int 1079 nfs_readlink(void *v) 1080 { 1081 struct vop_readlink_args /* { 1082 struct vnode *a_vp; 1083 struct uio *a_uio; 1084 kauth_cred_t a_cred; 1085 } */ *ap = v; 1086 struct vnode *vp = ap->a_vp; 1087 struct nfsnode *np = VTONFS(vp); 1088 1089 if (vp->v_type != VLNK) 1090 return (EPERM); 1091 1092 if (np->n_rcred != NULL) { 1093 kauth_cred_free(np->n_rcred); 1094 } 1095 np->n_rcred = ap->a_cred; 1096 kauth_cred_hold(np->n_rcred); 1097 1098 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1099 } 1100 1101 /* 1102 * Do a readlink rpc. 1103 * Called by nfs_doio() from below the buffer cache. 1104 */ 1105 int 1106 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1107 { 1108 u_int32_t *tl; 1109 char *cp; 1110 int32_t t1, t2; 1111 char *bpos, *dpos, *cp2; 1112 int error = 0; 1113 uint32_t len; 1114 struct mbuf *mreq, *mrep, *md, *mb; 1115 const int v3 = NFS_ISV3(vp); 1116 struct nfsnode *np = VTONFS(vp); 1117 #ifndef NFS_V2_ONLY 1118 int attrflag; 1119 #endif 1120 1121 nfsstats.rpccnt[NFSPROC_READLINK]++; 1122 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1123 nfsm_fhtom(np, v3); 1124 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1125 #ifndef NFS_V2_ONLY 1126 if (v3) 1127 nfsm_postop_attr(vp, attrflag, 0); 1128 #endif 1129 if (!error) { 1130 #ifndef NFS_V2_ONLY 1131 if (v3) { 1132 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1133 len = fxdr_unsigned(uint32_t, *tl); 1134 if (len > NFS_MAXPATHLEN) { 1135 /* 1136 * this pathname is too long for us. 1137 */ 1138 m_freem(mrep); 1139 /* Solaris returns EINVAL. should we follow? */ 1140 error = ENAMETOOLONG; 1141 goto nfsmout; 1142 } 1143 } else 1144 #endif 1145 { 1146 nfsm_strsiz(len, NFS_MAXPATHLEN); 1147 } 1148 nfsm_mtouio(uiop, len); 1149 } 1150 nfsm_reqdone; 1151 return (error); 1152 } 1153 1154 /* 1155 * nfs read rpc call 1156 * Ditto above 1157 */ 1158 int 1159 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1160 { 1161 u_int32_t *tl; 1162 char *cp; 1163 int32_t t1, t2; 1164 char *bpos, *dpos, *cp2; 1165 struct mbuf *mreq, *mrep, *md, *mb; 1166 struct nfsmount *nmp; 1167 int error = 0, len, retlen, tsiz, eof __unused, byte_count; 1168 const int v3 = NFS_ISV3(vp); 1169 struct nfsnode *np = VTONFS(vp); 1170 #ifndef NFS_V2_ONLY 1171 int attrflag; 1172 #endif 1173 1174 #ifndef nolint 1175 eof = 0; 1176 #endif 1177 nmp = VFSTONFS(vp->v_mount); 1178 tsiz = uiop->uio_resid; 1179 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1180 return (EFBIG); 1181 iostat_busy(nmp->nm_stats); 1182 byte_count = 0; /* count bytes actually transferred */ 1183 while (tsiz > 0) { 1184 nfsstats.rpccnt[NFSPROC_READ]++; 1185 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1186 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1187 nfsm_fhtom(np, v3); 1188 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1189 #ifndef NFS_V2_ONLY 1190 if (v3) { 1191 txdr_hyper(uiop->uio_offset, tl); 1192 *(tl + 2) = txdr_unsigned(len); 1193 } else 1194 #endif 1195 { 1196 *tl++ = txdr_unsigned(uiop->uio_offset); 1197 *tl++ = txdr_unsigned(len); 1198 *tl = 0; 1199 } 1200 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1201 #ifndef NFS_V2_ONLY 1202 if (v3) { 1203 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1204 if (error) { 1205 m_freem(mrep); 1206 goto nfsmout; 1207 } 1208 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1209 eof = fxdr_unsigned(int, *(tl + 1)); 1210 } else 1211 #endif 1212 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1213 nfsm_strsiz(retlen, nmp->nm_rsize); 1214 nfsm_mtouio(uiop, retlen); 1215 m_freem(mrep); 1216 tsiz -= retlen; 1217 byte_count += retlen; 1218 #ifndef NFS_V2_ONLY 1219 if (v3) { 1220 if (eof || retlen == 0) 1221 tsiz = 0; 1222 } else 1223 #endif 1224 if (retlen < len) 1225 tsiz = 0; 1226 } 1227 nfsmout: 1228 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1229 return (error); 1230 } 1231 1232 struct nfs_writerpc_context { 1233 kmutex_t nwc_lock; 1234 kcondvar_t nwc_cv; 1235 int nwc_mbufcount; 1236 }; 1237 1238 /* 1239 * free mbuf used to refer protected pages while write rpc call. 1240 * called at splvm. 1241 */ 1242 static void 1243 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1244 { 1245 struct nfs_writerpc_context *ctx = arg; 1246 1247 KASSERT(m != NULL); 1248 KASSERT(ctx != NULL); 1249 pool_cache_put(mb_cache, m); 1250 mutex_enter(&ctx->nwc_lock); 1251 if (--ctx->nwc_mbufcount == 0) { 1252 cv_signal(&ctx->nwc_cv); 1253 } 1254 mutex_exit(&ctx->nwc_lock); 1255 } 1256 1257 /* 1258 * nfs write call 1259 */ 1260 int 1261 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1262 { 1263 u_int32_t *tl; 1264 char *cp; 1265 int32_t t1, t2; 1266 char *bpos, *dpos; 1267 struct mbuf *mreq, *mrep, *md, *mb; 1268 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1269 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1270 const int v3 = NFS_ISV3(vp); 1271 int committed = NFSV3WRITE_FILESYNC; 1272 struct nfsnode *np = VTONFS(vp); 1273 struct nfs_writerpc_context ctx; 1274 int byte_count; 1275 size_t origresid; 1276 #ifndef NFS_V2_ONLY 1277 char *cp2; 1278 int rlen, commit; 1279 #endif 1280 1281 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1282 panic("writerpc readonly vp %p", vp); 1283 } 1284 1285 #ifdef DIAGNOSTIC 1286 if (uiop->uio_iovcnt != 1) 1287 panic("nfs: writerpc iovcnt > 1"); 1288 #endif 1289 tsiz = uiop->uio_resid; 1290 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1291 return EFBIG; 1292 1293 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1294 cv_init(&ctx.nwc_cv, "nfsmblk"); 1295 ctx.nwc_mbufcount = 1; 1296 1297 retry: 1298 origresid = uiop->uio_resid; 1299 KASSERT(origresid == uiop->uio_iov->iov_len); 1300 iostat_busy(nmp->nm_stats); 1301 byte_count = 0; /* count of bytes actually written */ 1302 while (tsiz > 0) { 1303 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1304 uint32_t backup; 1305 bool stalewriteverf = false; 1306 1307 nfsstats.rpccnt[NFSPROC_WRITE]++; 1308 len = min(tsiz, nmp->nm_wsize); 1309 datalen = pageprotected ? 0 : nfsm_rndup(len); 1310 nfsm_reqhead(np, NFSPROC_WRITE, 1311 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1312 nfsm_fhtom(np, v3); 1313 #ifndef NFS_V2_ONLY 1314 if (v3) { 1315 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1316 txdr_hyper(uiop->uio_offset, tl); 1317 tl += 2; 1318 *tl++ = txdr_unsigned(len); 1319 *tl++ = txdr_unsigned(*iomode); 1320 *tl = txdr_unsigned(len); 1321 } else 1322 #endif 1323 { 1324 u_int32_t x; 1325 1326 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1327 /* Set both "begin" and "current" to non-garbage. */ 1328 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1329 *tl++ = x; /* "begin offset" */ 1330 *tl++ = x; /* "current offset" */ 1331 x = txdr_unsigned(len); 1332 *tl++ = x; /* total to this offset */ 1333 *tl = x; /* size of this write */ 1334 1335 } 1336 if (pageprotected) { 1337 /* 1338 * since we know pages can't be modified during i/o, 1339 * no need to copy them for us. 1340 */ 1341 struct mbuf *m; 1342 struct iovec *iovp = uiop->uio_iov; 1343 1344 m = m_get(M_WAIT, MT_DATA); 1345 MCLAIM(m, &nfs_mowner); 1346 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1347 nfs_writerpc_extfree, &ctx); 1348 m->m_flags |= M_EXT_ROMAP; 1349 m->m_len = len; 1350 mb->m_next = m; 1351 /* 1352 * no need to maintain mb and bpos here 1353 * because no one care them later. 1354 */ 1355 #if 0 1356 mb = m; 1357 bpos = mtod(void *, mb) + mb->m_len; 1358 #endif 1359 UIO_ADVANCE(uiop, len); 1360 uiop->uio_offset += len; 1361 mutex_enter(&ctx.nwc_lock); 1362 ctx.nwc_mbufcount++; 1363 mutex_exit(&ctx.nwc_lock); 1364 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1365 } else { 1366 nfsm_uiotom(uiop, len); 1367 } 1368 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1369 #ifndef NFS_V2_ONLY 1370 if (v3) { 1371 wccflag = NFSV3_WCCCHK; 1372 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1373 if (!error) { 1374 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1375 + NFSX_V3WRITEVERF); 1376 rlen = fxdr_unsigned(int, *tl++); 1377 if (rlen == 0) { 1378 error = NFSERR_IO; 1379 m_freem(mrep); 1380 break; 1381 } else if (rlen < len) { 1382 backup = len - rlen; 1383 UIO_ADVANCE(uiop, -backup); 1384 uiop->uio_offset -= backup; 1385 len = rlen; 1386 } 1387 commit = fxdr_unsigned(int, *tl++); 1388 1389 /* 1390 * Return the lowest committment level 1391 * obtained by any of the RPCs. 1392 */ 1393 if (committed == NFSV3WRITE_FILESYNC) 1394 committed = commit; 1395 else if (committed == NFSV3WRITE_DATASYNC && 1396 commit == NFSV3WRITE_UNSTABLE) 1397 committed = commit; 1398 mutex_enter(&nmp->nm_lock); 1399 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1400 memcpy(nmp->nm_writeverf, tl, 1401 NFSX_V3WRITEVERF); 1402 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1403 } else if ((nmp->nm_iflag & 1404 NFSMNT_STALEWRITEVERF) || 1405 memcmp(tl, nmp->nm_writeverf, 1406 NFSX_V3WRITEVERF)) { 1407 memcpy(nmp->nm_writeverf, tl, 1408 NFSX_V3WRITEVERF); 1409 /* 1410 * note NFSMNT_STALEWRITEVERF 1411 * if we're the first thread to 1412 * notice it. 1413 */ 1414 if ((nmp->nm_iflag & 1415 NFSMNT_STALEWRITEVERF) == 0) { 1416 stalewriteverf = true; 1417 nmp->nm_iflag |= 1418 NFSMNT_STALEWRITEVERF; 1419 } 1420 } 1421 mutex_exit(&nmp->nm_lock); 1422 } 1423 } else 1424 #endif 1425 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1426 if (wccflag) 1427 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1428 m_freem(mrep); 1429 if (error) 1430 break; 1431 tsiz -= len; 1432 byte_count += len; 1433 if (stalewriteverf) { 1434 *stalewriteverfp = true; 1435 stalewriteverf = false; 1436 if (committed == NFSV3WRITE_UNSTABLE && 1437 len != origresid) { 1438 /* 1439 * if our write requests weren't atomic but 1440 * unstable, datas in previous iterations 1441 * might have already been lost now. 1442 * then, we should resend them to nfsd. 1443 */ 1444 backup = origresid - tsiz; 1445 UIO_ADVANCE(uiop, -backup); 1446 uiop->uio_offset -= backup; 1447 tsiz = origresid; 1448 goto retry; 1449 } 1450 } 1451 } 1452 nfsmout: 1453 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1454 if (pageprotected) { 1455 /* 1456 * wait until mbufs go away. 1457 * retransmitted mbufs can survive longer than rpc requests 1458 * themselves. 1459 */ 1460 mutex_enter(&ctx.nwc_lock); 1461 ctx.nwc_mbufcount--; 1462 while (ctx.nwc_mbufcount > 0) { 1463 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1464 } 1465 mutex_exit(&ctx.nwc_lock); 1466 } 1467 mutex_destroy(&ctx.nwc_lock); 1468 cv_destroy(&ctx.nwc_cv); 1469 *iomode = committed; 1470 if (error) 1471 uiop->uio_resid = tsiz; 1472 return (error); 1473 } 1474 1475 /* 1476 * nfs mknod rpc 1477 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1478 * mode set to specify the file type and the size field for rdev. 1479 */ 1480 int 1481 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1482 { 1483 struct nfsv2_sattr *sp; 1484 u_int32_t *tl; 1485 char *cp; 1486 int32_t t1, t2; 1487 struct vnode *newvp = (struct vnode *)0; 1488 struct nfsnode *dnp, *np; 1489 char *cp2; 1490 char *bpos, *dpos; 1491 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1492 struct mbuf *mreq, *mrep, *md, *mb; 1493 u_int32_t rdev; 1494 const int v3 = NFS_ISV3(dvp); 1495 1496 if (vap->va_type == VCHR || vap->va_type == VBLK) 1497 rdev = txdr_unsigned(vap->va_rdev); 1498 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1499 rdev = nfs_xdrneg1; 1500 else { 1501 VOP_ABORTOP(dvp, cnp); 1502 return (EOPNOTSUPP); 1503 } 1504 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1505 dnp = VTONFS(dvp); 1506 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1507 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1508 nfsm_fhtom(dnp, v3); 1509 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1510 #ifndef NFS_V2_ONLY 1511 if (v3) { 1512 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1513 *tl++ = vtonfsv3_type(vap->va_type); 1514 nfsm_v3attrbuild(vap, false); 1515 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1516 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1517 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1518 *tl = txdr_unsigned(minor(vap->va_rdev)); 1519 } 1520 } else 1521 #endif 1522 { 1523 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1524 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1525 sp->sa_uid = nfs_xdrneg1; 1526 sp->sa_gid = nfs_xdrneg1; 1527 sp->sa_size = rdev; 1528 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1529 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1530 } 1531 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1532 if (!error) { 1533 nfsm_mtofh(dvp, newvp, v3, gotvp); 1534 if (!gotvp) { 1535 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1536 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1537 if (!error) 1538 newvp = NFSTOV(np); 1539 } 1540 } 1541 #ifndef NFS_V2_ONLY 1542 if (v3) 1543 nfsm_wcc_data(dvp, wccflag, 0, !error); 1544 #endif 1545 nfsm_reqdone; 1546 if (error) { 1547 if (newvp) 1548 vput(newvp); 1549 } else { 1550 nfs_cache_enter(dvp, newvp, cnp); 1551 *vpp = newvp; 1552 VOP_UNLOCK(newvp); 1553 } 1554 VTONFS(dvp)->n_flag |= NMODIFIED; 1555 if (!wccflag) 1556 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1557 return (error); 1558 } 1559 1560 /* 1561 * nfs mknod vop 1562 * just call nfs_mknodrpc() to do the work. 1563 */ 1564 /* ARGSUSED */ 1565 int 1566 nfs_mknod(void *v) 1567 { 1568 struct vop_mknod_v3_args /* { 1569 struct vnode *a_dvp; 1570 struct vnode **a_vpp; 1571 struct componentname *a_cnp; 1572 struct vattr *a_vap; 1573 } */ *ap = v; 1574 struct vnode *dvp = ap->a_dvp; 1575 struct componentname *cnp = ap->a_cnp; 1576 int error; 1577 1578 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1579 VN_KNOTE(dvp, NOTE_WRITE); 1580 if (error == 0 || error == EEXIST) 1581 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1582 return (error); 1583 } 1584 1585 /* 1586 * nfs file create call 1587 */ 1588 int 1589 nfs_create(void *v) 1590 { 1591 struct vop_create_v3_args /* { 1592 struct vnode *a_dvp; 1593 struct vnode **a_vpp; 1594 struct componentname *a_cnp; 1595 struct vattr *a_vap; 1596 } */ *ap = v; 1597 struct vnode *dvp = ap->a_dvp; 1598 struct vattr *vap = ap->a_vap; 1599 struct componentname *cnp = ap->a_cnp; 1600 struct nfsv2_sattr *sp; 1601 u_int32_t *tl; 1602 char *cp; 1603 int32_t t1, t2; 1604 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1605 struct vnode *newvp = (struct vnode *)0; 1606 char *bpos, *dpos, *cp2; 1607 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1608 struct mbuf *mreq, *mrep, *md, *mb; 1609 const int v3 = NFS_ISV3(dvp); 1610 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1611 1612 /* 1613 * Oops, not for me.. 1614 */ 1615 if (vap->va_type == VSOCK) 1616 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1617 1618 KASSERT(vap->va_type == VREG); 1619 1620 #ifdef VA_EXCLUSIVE 1621 if (vap->va_vaflags & VA_EXCLUSIVE) { 1622 excl_mode = NFSV3CREATE_EXCLUSIVE; 1623 } 1624 #endif 1625 again: 1626 error = 0; 1627 nfsstats.rpccnt[NFSPROC_CREATE]++; 1628 dnp = VTONFS(dvp); 1629 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1630 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1631 nfsm_fhtom(dnp, v3); 1632 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1633 #ifndef NFS_V2_ONLY 1634 if (v3) { 1635 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1636 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1637 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1638 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1639 *tl++ = cprng_fast32(); 1640 *tl = cprng_fast32(); 1641 } else { 1642 *tl = txdr_unsigned(excl_mode); 1643 nfsm_v3attrbuild(vap, false); 1644 } 1645 } else 1646 #endif 1647 { 1648 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1649 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1650 sp->sa_uid = nfs_xdrneg1; 1651 sp->sa_gid = nfs_xdrneg1; 1652 sp->sa_size = 0; 1653 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1654 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1655 } 1656 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1657 if (!error) { 1658 nfsm_mtofh(dvp, newvp, v3, gotvp); 1659 if (!gotvp) { 1660 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1661 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1662 if (!error) 1663 newvp = NFSTOV(np); 1664 } 1665 } 1666 #ifndef NFS_V2_ONLY 1667 if (v3) 1668 nfsm_wcc_data(dvp, wccflag, 0, !error); 1669 #endif 1670 nfsm_reqdone; 1671 if (error) { 1672 /* 1673 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1674 */ 1675 if (v3 && error == ENOTSUP) { 1676 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1677 excl_mode = NFSV3CREATE_GUARDED; 1678 goto again; 1679 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1680 excl_mode = NFSV3CREATE_UNCHECKED; 1681 goto again; 1682 } 1683 } 1684 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1685 struct timespec ts; 1686 1687 getnanotime(&ts); 1688 1689 /* 1690 * make sure that we'll update timestamps as 1691 * most server implementations use them to store 1692 * the create verifier. 1693 * 1694 * XXX it's better to use TOSERVER always. 1695 */ 1696 1697 if (vap->va_atime.tv_sec == VNOVAL) 1698 vap->va_atime = ts; 1699 if (vap->va_mtime.tv_sec == VNOVAL) 1700 vap->va_mtime = ts; 1701 1702 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1703 } 1704 if (error == 0) { 1705 if (cnp->cn_flags & MAKEENTRY) 1706 nfs_cache_enter(dvp, newvp, cnp); 1707 else 1708 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1709 *ap->a_vpp = newvp; 1710 VOP_UNLOCK(newvp); 1711 } else { 1712 if (newvp) 1713 vput(newvp); 1714 if (error == EEXIST) 1715 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1716 } 1717 VTONFS(dvp)->n_flag |= NMODIFIED; 1718 if (!wccflag) 1719 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1720 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1721 return (error); 1722 } 1723 1724 /* 1725 * nfs file remove call 1726 * To try and make nfs semantics closer to ufs semantics, a file that has 1727 * other processes using the vnode is renamed instead of removed and then 1728 * removed later on the last close. 1729 * - If v_usecount > 1 1730 * If a rename is not already in the works 1731 * call nfs_sillyrename() to set it up 1732 * else 1733 * do the remove rpc 1734 */ 1735 int 1736 nfs_remove(void *v) 1737 { 1738 struct vop_remove_args /* { 1739 struct vnodeop_desc *a_desc; 1740 struct vnode * a_dvp; 1741 struct vnode * a_vp; 1742 struct componentname * a_cnp; 1743 } */ *ap = v; 1744 struct vnode *vp = ap->a_vp; 1745 struct vnode *dvp = ap->a_dvp; 1746 struct componentname *cnp = ap->a_cnp; 1747 struct nfsnode *np = VTONFS(vp); 1748 int error = 0; 1749 struct vattr vattr; 1750 1751 #ifndef DIAGNOSTIC 1752 if (vp->v_usecount < 1) 1753 panic("nfs_remove: bad v_usecount"); 1754 #endif 1755 if (vp->v_type == VDIR) 1756 error = EPERM; 1757 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1758 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1759 vattr.va_nlink > 1)) { 1760 /* 1761 * Purge the name cache so that the chance of a lookup for 1762 * the name succeeding while the remove is in progress is 1763 * minimized. Without node locking it can still happen, such 1764 * that an I/O op returns ESTALE, but since you get this if 1765 * another host removes the file.. 1766 */ 1767 cache_purge(vp); 1768 /* 1769 * throw away biocache buffers, mainly to avoid 1770 * unnecessary delayed writes later. 1771 */ 1772 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1773 /* Do the rpc */ 1774 if (error != EINTR) 1775 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1776 cnp->cn_namelen, cnp->cn_cred, curlwp); 1777 } else if (!np->n_sillyrename) 1778 error = nfs_sillyrename(dvp, vp, cnp, false); 1779 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1780 vattr.va_nlink == 1) { 1781 np->n_flag |= NREMOVED; 1782 } 1783 NFS_INVALIDATE_ATTRCACHE(np); 1784 VN_KNOTE(vp, NOTE_DELETE); 1785 VN_KNOTE(dvp, NOTE_WRITE); 1786 if (dvp == vp) 1787 vrele(vp); 1788 else 1789 vput(vp); 1790 vput(dvp); 1791 return (error); 1792 } 1793 1794 /* 1795 * nfs file remove rpc called from nfs_inactive 1796 */ 1797 int 1798 nfs_removeit(struct sillyrename *sp) 1799 { 1800 1801 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1802 (struct lwp *)0)); 1803 } 1804 1805 /* 1806 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1807 */ 1808 int 1809 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1810 { 1811 u_int32_t *tl; 1812 char *cp; 1813 #ifndef NFS_V2_ONLY 1814 int32_t t1; 1815 char *cp2; 1816 #endif 1817 int32_t t2; 1818 char *bpos, *dpos; 1819 int error = 0, wccflag = NFSV3_WCCRATTR; 1820 struct mbuf *mreq, *mrep, *md, *mb; 1821 const int v3 = NFS_ISV3(dvp); 1822 int rexmit = 0; 1823 struct nfsnode *dnp = VTONFS(dvp); 1824 1825 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1826 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1827 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1828 nfsm_fhtom(dnp, v3); 1829 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1830 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1831 #ifndef NFS_V2_ONLY 1832 if (v3) 1833 nfsm_wcc_data(dvp, wccflag, 0, !error); 1834 #endif 1835 nfsm_reqdone; 1836 VTONFS(dvp)->n_flag |= NMODIFIED; 1837 if (!wccflag) 1838 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1839 /* 1840 * Kludge City: If the first reply to the remove rpc is lost.. 1841 * the reply to the retransmitted request will be ENOENT 1842 * since the file was in fact removed 1843 * Therefore, we cheat and return success. 1844 */ 1845 if (rexmit && error == ENOENT) 1846 error = 0; 1847 return (error); 1848 } 1849 1850 /* 1851 * nfs file rename call 1852 */ 1853 int 1854 nfs_rename(void *v) 1855 { 1856 struct vop_rename_args /* { 1857 struct vnode *a_fdvp; 1858 struct vnode *a_fvp; 1859 struct componentname *a_fcnp; 1860 struct vnode *a_tdvp; 1861 struct vnode *a_tvp; 1862 struct componentname *a_tcnp; 1863 } */ *ap = v; 1864 struct vnode *fvp = ap->a_fvp; 1865 struct vnode *tvp = ap->a_tvp; 1866 struct vnode *fdvp = ap->a_fdvp; 1867 struct vnode *tdvp = ap->a_tdvp; 1868 struct componentname *tcnp = ap->a_tcnp; 1869 struct componentname *fcnp = ap->a_fcnp; 1870 int error; 1871 1872 /* Check for cross-device rename */ 1873 if ((fvp->v_mount != tdvp->v_mount) || 1874 (tvp && (fvp->v_mount != tvp->v_mount))) { 1875 error = EXDEV; 1876 goto out; 1877 } 1878 1879 /* 1880 * If the tvp exists and is in use, sillyrename it before doing the 1881 * rename of the new file over it. 1882 * 1883 * Have sillyrename use link instead of rename if possible, 1884 * so that we don't lose the file if the rename fails, and so 1885 * that there's no window when the "to" file doesn't exist. 1886 */ 1887 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1888 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1889 VN_KNOTE(tvp, NOTE_DELETE); 1890 vput(tvp); 1891 tvp = NULL; 1892 } 1893 1894 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1895 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1896 curlwp); 1897 1898 VN_KNOTE(fdvp, NOTE_WRITE); 1899 VN_KNOTE(tdvp, NOTE_WRITE); 1900 if (error == 0 || error == EEXIST) { 1901 if (fvp->v_type == VDIR) 1902 cache_purge(fvp); 1903 else 1904 cache_purge1(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1905 0); 1906 if (tvp != NULL && tvp->v_type == VDIR) 1907 cache_purge(tvp); 1908 else 1909 cache_purge1(tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, 1910 0); 1911 } 1912 out: 1913 if (tdvp == tvp) 1914 vrele(tdvp); 1915 else 1916 vput(tdvp); 1917 if (tvp) 1918 vput(tvp); 1919 vrele(fdvp); 1920 vrele(fvp); 1921 return (error); 1922 } 1923 1924 /* 1925 * nfs file rename rpc called from nfs_remove() above 1926 */ 1927 int 1928 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1929 { 1930 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1931 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1932 } 1933 1934 /* 1935 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1936 */ 1937 int 1938 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, struct vnode *tdvp, const char *tnameptr, int tnamelen, kauth_cred_t cred, struct lwp *l) 1939 { 1940 u_int32_t *tl; 1941 char *cp; 1942 #ifndef NFS_V2_ONLY 1943 int32_t t1; 1944 char *cp2; 1945 #endif 1946 int32_t t2; 1947 char *bpos, *dpos; 1948 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1949 struct mbuf *mreq, *mrep, *md, *mb; 1950 const int v3 = NFS_ISV3(fdvp); 1951 int rexmit = 0; 1952 struct nfsnode *fdnp = VTONFS(fdvp); 1953 1954 nfsstats.rpccnt[NFSPROC_RENAME]++; 1955 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1956 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1957 nfsm_rndup(tnamelen)); 1958 nfsm_fhtom(fdnp, v3); 1959 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1960 nfsm_fhtom(VTONFS(tdvp), v3); 1961 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1962 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1963 #ifndef NFS_V2_ONLY 1964 if (v3) { 1965 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1966 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1967 } 1968 #endif 1969 nfsm_reqdone; 1970 VTONFS(fdvp)->n_flag |= NMODIFIED; 1971 VTONFS(tdvp)->n_flag |= NMODIFIED; 1972 if (!fwccflag) 1973 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1974 if (!twccflag) 1975 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1976 /* 1977 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1978 */ 1979 if (rexmit && error == ENOENT) 1980 error = 0; 1981 return (error); 1982 } 1983 1984 /* 1985 * NFS link RPC, called from nfs_link. 1986 * Assumes dvp and vp locked, and leaves them that way. 1987 */ 1988 1989 static int 1990 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 1991 size_t namelen, kauth_cred_t cred, struct lwp *l) 1992 { 1993 u_int32_t *tl; 1994 char *cp; 1995 #ifndef NFS_V2_ONLY 1996 int32_t t1; 1997 char *cp2; 1998 #endif 1999 int32_t t2; 2000 char *bpos, *dpos; 2001 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2002 struct mbuf *mreq, *mrep, *md, *mb; 2003 const int v3 = NFS_ISV3(dvp); 2004 int rexmit = 0; 2005 struct nfsnode *np = VTONFS(vp); 2006 2007 nfsstats.rpccnt[NFSPROC_LINK]++; 2008 nfsm_reqhead(np, NFSPROC_LINK, 2009 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2010 nfsm_fhtom(np, v3); 2011 nfsm_fhtom(VTONFS(dvp), v3); 2012 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2013 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2014 #ifndef NFS_V2_ONLY 2015 if (v3) { 2016 nfsm_postop_attr(vp, attrflag, 0); 2017 nfsm_wcc_data(dvp, wccflag, 0, !error); 2018 } 2019 #endif 2020 nfsm_reqdone; 2021 2022 VTONFS(dvp)->n_flag |= NMODIFIED; 2023 if (!attrflag) 2024 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2025 if (!wccflag) 2026 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2027 2028 /* 2029 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2030 */ 2031 if (rexmit && error == EEXIST) 2032 error = 0; 2033 2034 return error; 2035 } 2036 2037 /* 2038 * nfs hard link create call 2039 */ 2040 int 2041 nfs_link(void *v) 2042 { 2043 struct vop_link_args /* { 2044 struct vnode *a_dvp; 2045 struct vnode *a_vp; 2046 struct componentname *a_cnp; 2047 } */ *ap = v; 2048 struct vnode *vp = ap->a_vp; 2049 struct vnode *dvp = ap->a_dvp; 2050 struct componentname *cnp = ap->a_cnp; 2051 int error = 0; 2052 2053 error = vn_lock(vp, LK_EXCLUSIVE); 2054 if (error != 0) { 2055 VOP_ABORTOP(dvp, cnp); 2056 vput(dvp); 2057 return error; 2058 } 2059 2060 /* 2061 * Push all writes to the server, so that the attribute cache 2062 * doesn't get "out of sync" with the server. 2063 * XXX There should be a better way! 2064 */ 2065 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2066 2067 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2068 cnp->cn_cred, curlwp); 2069 2070 if (error == 0) { 2071 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2072 } 2073 VOP_UNLOCK(vp); 2074 VN_KNOTE(vp, NOTE_LINK); 2075 VN_KNOTE(dvp, NOTE_WRITE); 2076 vput(dvp); 2077 return (error); 2078 } 2079 2080 /* 2081 * nfs symbolic link create call 2082 */ 2083 int 2084 nfs_symlink(void *v) 2085 { 2086 struct vop_symlink_v3_args /* { 2087 struct vnode *a_dvp; 2088 struct vnode **a_vpp; 2089 struct componentname *a_cnp; 2090 struct vattr *a_vap; 2091 char *a_target; 2092 } */ *ap = v; 2093 struct vnode *dvp = ap->a_dvp; 2094 struct vattr *vap = ap->a_vap; 2095 struct componentname *cnp = ap->a_cnp; 2096 struct nfsv2_sattr *sp; 2097 u_int32_t *tl; 2098 char *cp; 2099 int32_t t1, t2; 2100 char *bpos, *dpos, *cp2; 2101 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2102 struct mbuf *mreq, *mrep, *md, *mb; 2103 struct vnode *newvp = (struct vnode *)0; 2104 const int v3 = NFS_ISV3(dvp); 2105 int rexmit = 0; 2106 struct nfsnode *dnp = VTONFS(dvp); 2107 2108 *ap->a_vpp = NULL; 2109 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2110 slen = strlen(ap->a_target); 2111 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2112 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2113 nfsm_fhtom(dnp, v3); 2114 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2115 #ifndef NFS_V2_ONlY 2116 if (v3) 2117 nfsm_v3attrbuild(vap, false); 2118 #endif 2119 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2120 #ifndef NFS_V2_ONlY 2121 if (!v3) { 2122 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2123 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2124 sp->sa_uid = nfs_xdrneg1; 2125 sp->sa_gid = nfs_xdrneg1; 2126 sp->sa_size = nfs_xdrneg1; 2127 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2128 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2129 } 2130 #endif 2131 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2132 &rexmit); 2133 #ifndef NFS_V2_ONlY 2134 if (v3) { 2135 if (!error) 2136 nfsm_mtofh(dvp, newvp, v3, gotvp); 2137 nfsm_wcc_data(dvp, wccflag, 0, !error); 2138 } 2139 #endif 2140 nfsm_reqdone; 2141 /* 2142 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2143 */ 2144 if (rexmit && error == EEXIST) 2145 error = 0; 2146 if (error == 0 || error == EEXIST) 2147 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2148 if (error == 0 && newvp == NULL) { 2149 struct nfsnode *np = NULL; 2150 2151 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2152 cnp->cn_cred, curlwp, &np); 2153 if (error == 0) 2154 newvp = NFSTOV(np); 2155 } 2156 if (error) { 2157 if (newvp != NULL) 2158 vput(newvp); 2159 } else { 2160 *ap->a_vpp = newvp; 2161 VOP_UNLOCK(newvp); 2162 } 2163 VTONFS(dvp)->n_flag |= NMODIFIED; 2164 if (!wccflag) 2165 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2166 VN_KNOTE(dvp, NOTE_WRITE); 2167 return (error); 2168 } 2169 2170 /* 2171 * nfs make dir call 2172 */ 2173 int 2174 nfs_mkdir(void *v) 2175 { 2176 struct vop_mkdir_v3_args /* { 2177 struct vnode *a_dvp; 2178 struct vnode **a_vpp; 2179 struct componentname *a_cnp; 2180 struct vattr *a_vap; 2181 } */ *ap = v; 2182 struct vnode *dvp = ap->a_dvp; 2183 struct vattr *vap = ap->a_vap; 2184 struct componentname *cnp = ap->a_cnp; 2185 struct nfsv2_sattr *sp; 2186 u_int32_t *tl; 2187 char *cp; 2188 int32_t t1, t2; 2189 int len; 2190 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2191 struct vnode *newvp = (struct vnode *)0; 2192 char *bpos, *dpos, *cp2; 2193 int error = 0, wccflag = NFSV3_WCCRATTR; 2194 int gotvp = 0; 2195 int rexmit = 0; 2196 struct mbuf *mreq, *mrep, *md, *mb; 2197 const int v3 = NFS_ISV3(dvp); 2198 2199 len = cnp->cn_namelen; 2200 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2201 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2202 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2203 nfsm_fhtom(dnp, v3); 2204 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2205 #ifndef NFS_V2_ONLY 2206 if (v3) { 2207 nfsm_v3attrbuild(vap, false); 2208 } else 2209 #endif 2210 { 2211 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2212 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2213 sp->sa_uid = nfs_xdrneg1; 2214 sp->sa_gid = nfs_xdrneg1; 2215 sp->sa_size = nfs_xdrneg1; 2216 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2217 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2218 } 2219 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2220 if (!error) 2221 nfsm_mtofh(dvp, newvp, v3, gotvp); 2222 if (v3) 2223 nfsm_wcc_data(dvp, wccflag, 0, !error); 2224 nfsm_reqdone; 2225 VTONFS(dvp)->n_flag |= NMODIFIED; 2226 if (!wccflag) 2227 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2228 /* 2229 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2230 * if we can succeed in looking up the directory. 2231 */ 2232 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2233 if (newvp) { 2234 vput(newvp); 2235 newvp = (struct vnode *)0; 2236 } 2237 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2238 curlwp, &np); 2239 if (!error) { 2240 newvp = NFSTOV(np); 2241 if (newvp->v_type != VDIR || newvp == dvp) 2242 error = EEXIST; 2243 } 2244 } 2245 if (error) { 2246 if (newvp) { 2247 if (dvp != newvp) 2248 vput(newvp); 2249 else 2250 vrele(newvp); 2251 } 2252 } else { 2253 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2254 nfs_cache_enter(dvp, newvp, cnp); 2255 *ap->a_vpp = newvp; 2256 VOP_UNLOCK(newvp); 2257 } 2258 return (error); 2259 } 2260 2261 /* 2262 * nfs remove directory call 2263 */ 2264 int 2265 nfs_rmdir(void *v) 2266 { 2267 struct vop_rmdir_args /* { 2268 struct vnode *a_dvp; 2269 struct vnode *a_vp; 2270 struct componentname *a_cnp; 2271 } */ *ap = v; 2272 struct vnode *vp = ap->a_vp; 2273 struct vnode *dvp = ap->a_dvp; 2274 struct componentname *cnp = ap->a_cnp; 2275 u_int32_t *tl; 2276 char *cp; 2277 #ifndef NFS_V2_ONLY 2278 int32_t t1; 2279 char *cp2; 2280 #endif 2281 int32_t t2; 2282 char *bpos, *dpos; 2283 int error = 0, wccflag = NFSV3_WCCRATTR; 2284 int rexmit = 0; 2285 struct mbuf *mreq, *mrep, *md, *mb; 2286 const int v3 = NFS_ISV3(dvp); 2287 struct nfsnode *dnp; 2288 2289 if (dvp == vp) { 2290 vrele(dvp); 2291 vput(dvp); 2292 return (EINVAL); 2293 } 2294 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2295 dnp = VTONFS(dvp); 2296 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2297 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2298 nfsm_fhtom(dnp, v3); 2299 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2300 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2301 #ifndef NFS_V2_ONLY 2302 if (v3) 2303 nfsm_wcc_data(dvp, wccflag, 0, !error); 2304 #endif 2305 nfsm_reqdone; 2306 VTONFS(dvp)->n_flag |= NMODIFIED; 2307 if (!wccflag) 2308 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2309 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2310 VN_KNOTE(vp, NOTE_DELETE); 2311 cache_purge(vp); 2312 vput(vp); 2313 vput(dvp); 2314 /* 2315 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2316 */ 2317 if (rexmit && error == ENOENT) 2318 error = 0; 2319 return (error); 2320 } 2321 2322 /* 2323 * nfs readdir call 2324 */ 2325 int 2326 nfs_readdir(void *v) 2327 { 2328 struct vop_readdir_args /* { 2329 struct vnode *a_vp; 2330 struct uio *a_uio; 2331 kauth_cred_t a_cred; 2332 int *a_eofflag; 2333 off_t **a_cookies; 2334 int *a_ncookies; 2335 } */ *ap = v; 2336 struct vnode *vp = ap->a_vp; 2337 struct uio *uio = ap->a_uio; 2338 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2339 char *base = uio->uio_iov->iov_base; 2340 int tresid, error; 2341 size_t count, lost; 2342 struct dirent *dp; 2343 off_t *cookies = NULL; 2344 int ncookies = 0, nc; 2345 2346 if (vp->v_type != VDIR) 2347 return (EPERM); 2348 2349 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2350 count = uio->uio_resid - lost; 2351 if (count <= 0) 2352 return (EINVAL); 2353 2354 /* 2355 * Call nfs_bioread() to do the real work. 2356 */ 2357 tresid = uio->uio_resid = count; 2358 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2359 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2360 2361 if (!error && ap->a_cookies) { 2362 ncookies = count / 16; 2363 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2364 *ap->a_cookies = cookies; 2365 } 2366 2367 if (!error && uio->uio_resid == tresid) { 2368 uio->uio_resid += lost; 2369 nfsstats.direofcache_misses++; 2370 if (ap->a_cookies) 2371 *ap->a_ncookies = 0; 2372 *ap->a_eofflag = 1; 2373 return (0); 2374 } 2375 2376 if (!error && ap->a_cookies) { 2377 /* 2378 * Only the NFS server and emulations use cookies, and they 2379 * load the directory block into system space, so we can 2380 * just look at it directly. 2381 */ 2382 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2383 uio->uio_iovcnt != 1) 2384 panic("nfs_readdir: lost in space"); 2385 for (nc = 0; ncookies-- && 2386 base < (char *)uio->uio_iov->iov_base; nc++){ 2387 dp = (struct dirent *) base; 2388 if (dp->d_reclen == 0) 2389 break; 2390 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2391 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2392 else 2393 *(cookies++) = NFS_GETCOOKIE(dp); 2394 base += dp->d_reclen; 2395 } 2396 uio->uio_resid += 2397 ((char *)uio->uio_iov->iov_base - base); 2398 uio->uio_iov->iov_len += 2399 ((char *)uio->uio_iov->iov_base - base); 2400 uio->uio_iov->iov_base = base; 2401 *ap->a_ncookies = nc; 2402 } 2403 2404 uio->uio_resid += lost; 2405 *ap->a_eofflag = 0; 2406 return (error); 2407 } 2408 2409 /* 2410 * Readdir rpc call. 2411 * Called from below the buffer cache by nfs_doio(). 2412 */ 2413 int 2414 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2415 { 2416 int len, left; 2417 struct dirent *dp = NULL; 2418 u_int32_t *tl; 2419 char *cp; 2420 int32_t t1, t2; 2421 char *bpos, *dpos, *cp2; 2422 struct mbuf *mreq, *mrep, *md, *mb; 2423 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2424 struct nfsnode *dnp = VTONFS(vp); 2425 u_quad_t fileno; 2426 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2427 #ifndef NFS_V2_ONLY 2428 int attrflag; 2429 #endif 2430 int nrpcs = 0, reclen; 2431 const int v3 = NFS_ISV3(vp); 2432 2433 #ifdef DIAGNOSTIC 2434 /* 2435 * Should be called from buffer cache, so only amount of 2436 * NFS_DIRBLKSIZ will be requested. 2437 */ 2438 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2439 panic("nfs readdirrpc bad uio"); 2440 #endif 2441 2442 /* 2443 * Loop around doing readdir rpc's of size nm_readdirsize 2444 * truncated to a multiple of NFS_DIRFRAGSIZ. 2445 * The stopping criteria is EOF or buffer full. 2446 */ 2447 while (more_dirs && bigenough) { 2448 /* 2449 * Heuristic: don't bother to do another RPC to further 2450 * fill up this block if there is not much room left. (< 50% 2451 * of the readdir RPC size). This wastes some buffer space 2452 * but can save up to 50% in RPC calls. 2453 */ 2454 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2455 bigenough = 0; 2456 break; 2457 } 2458 nfsstats.rpccnt[NFSPROC_READDIR]++; 2459 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2460 NFSX_READDIR(v3)); 2461 nfsm_fhtom(dnp, v3); 2462 #ifndef NFS_V2_ONLY 2463 if (v3) { 2464 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2465 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2466 txdr_swapcookie3(uiop->uio_offset, tl); 2467 } else { 2468 txdr_cookie3(uiop->uio_offset, tl); 2469 } 2470 tl += 2; 2471 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2472 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2473 } else 2474 #endif 2475 { 2476 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2477 *tl++ = txdr_unsigned(uiop->uio_offset); 2478 } 2479 *tl = txdr_unsigned(nmp->nm_readdirsize); 2480 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2481 nrpcs++; 2482 #ifndef NFS_V2_ONLY 2483 if (v3) { 2484 nfsm_postop_attr(vp, attrflag, 0); 2485 if (!error) { 2486 nfsm_dissect(tl, u_int32_t *, 2487 2 * NFSX_UNSIGNED); 2488 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2489 dnp->n_cookieverf.nfsuquad[1] = *tl; 2490 } else { 2491 m_freem(mrep); 2492 goto nfsmout; 2493 } 2494 } 2495 #endif 2496 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2497 more_dirs = fxdr_unsigned(int, *tl); 2498 2499 /* loop thru the dir entries, doctoring them to 4bsd form */ 2500 while (more_dirs && bigenough) { 2501 #ifndef NFS_V2_ONLY 2502 if (v3) { 2503 nfsm_dissect(tl, u_int32_t *, 2504 3 * NFSX_UNSIGNED); 2505 fileno = fxdr_hyper(tl); 2506 len = fxdr_unsigned(int, *(tl + 2)); 2507 } else 2508 #endif 2509 { 2510 nfsm_dissect(tl, u_int32_t *, 2511 2 * NFSX_UNSIGNED); 2512 fileno = fxdr_unsigned(u_quad_t, *tl++); 2513 len = fxdr_unsigned(int, *tl); 2514 } 2515 if (len <= 0 || len > NFS_MAXNAMLEN) { 2516 error = EBADRPC; 2517 m_freem(mrep); 2518 goto nfsmout; 2519 } 2520 /* for cookie stashing */ 2521 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2522 left = NFS_DIRFRAGSIZ - blksiz; 2523 if (reclen > left) { 2524 memset(uiop->uio_iov->iov_base, 0, left); 2525 dp->d_reclen += left; 2526 UIO_ADVANCE(uiop, left); 2527 blksiz = 0; 2528 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2529 } 2530 if (reclen > uiop->uio_resid) 2531 bigenough = 0; 2532 if (bigenough) { 2533 int tlen; 2534 2535 dp = (struct dirent *)uiop->uio_iov->iov_base; 2536 dp->d_fileno = fileno; 2537 dp->d_namlen = len; 2538 dp->d_reclen = reclen; 2539 dp->d_type = DT_UNKNOWN; 2540 blksiz += reclen; 2541 if (blksiz == NFS_DIRFRAGSIZ) 2542 blksiz = 0; 2543 UIO_ADVANCE(uiop, DIRHDSIZ); 2544 nfsm_mtouio(uiop, len); 2545 tlen = reclen - (DIRHDSIZ + len); 2546 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2547 UIO_ADVANCE(uiop, tlen); 2548 } else 2549 nfsm_adv(nfsm_rndup(len)); 2550 #ifndef NFS_V2_ONLY 2551 if (v3) { 2552 nfsm_dissect(tl, u_int32_t *, 2553 3 * NFSX_UNSIGNED); 2554 } else 2555 #endif 2556 { 2557 nfsm_dissect(tl, u_int32_t *, 2558 2 * NFSX_UNSIGNED); 2559 } 2560 if (bigenough) { 2561 #ifndef NFS_V2_ONLY 2562 if (v3) { 2563 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2564 uiop->uio_offset = 2565 fxdr_swapcookie3(tl); 2566 else 2567 uiop->uio_offset = 2568 fxdr_cookie3(tl); 2569 } 2570 else 2571 #endif 2572 { 2573 uiop->uio_offset = 2574 fxdr_unsigned(off_t, *tl); 2575 } 2576 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2577 } 2578 if (v3) 2579 tl += 2; 2580 else 2581 tl++; 2582 more_dirs = fxdr_unsigned(int, *tl); 2583 } 2584 /* 2585 * If at end of rpc data, get the eof boolean 2586 */ 2587 if (!more_dirs) { 2588 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2589 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2590 2591 /* 2592 * kludge: if we got no entries, treat it as EOF. 2593 * some server sometimes send a reply without any 2594 * entries or EOF. 2595 * although it might mean the server has very long name, 2596 * we can't handle such entries anyway. 2597 */ 2598 2599 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2600 more_dirs = 0; 2601 } 2602 m_freem(mrep); 2603 } 2604 /* 2605 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2606 * by increasing d_reclen for the last record. 2607 */ 2608 if (blksiz > 0) { 2609 left = NFS_DIRFRAGSIZ - blksiz; 2610 memset(uiop->uio_iov->iov_base, 0, left); 2611 dp->d_reclen += left; 2612 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2613 UIO_ADVANCE(uiop, left); 2614 } 2615 2616 /* 2617 * We are now either at the end of the directory or have filled the 2618 * block. 2619 */ 2620 if (bigenough) { 2621 dnp->n_direofoffset = uiop->uio_offset; 2622 dnp->n_flag |= NEOFVALID; 2623 } 2624 nfsmout: 2625 return (error); 2626 } 2627 2628 #ifndef NFS_V2_ONLY 2629 /* 2630 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2631 */ 2632 int 2633 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2634 { 2635 int len, left; 2636 struct dirent *dp = NULL; 2637 u_int32_t *tl; 2638 char *cp; 2639 int32_t t1, t2; 2640 struct vnode *newvp; 2641 char *bpos, *dpos, *cp2; 2642 struct mbuf *mreq, *mrep, *md, *mb; 2643 struct nameidata nami, *ndp = &nami; 2644 struct componentname *cnp = &ndp->ni_cnd; 2645 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2646 struct nfsnode *dnp = VTONFS(vp), *np; 2647 nfsfh_t *fhp; 2648 u_quad_t fileno; 2649 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2650 int attrflag, fhsize, nrpcs = 0, reclen; 2651 struct nfs_fattr fattr, *fp; 2652 2653 #ifdef DIAGNOSTIC 2654 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2655 panic("nfs readdirplusrpc bad uio"); 2656 #endif 2657 ndp->ni_dvp = vp; 2658 newvp = NULLVP; 2659 2660 /* 2661 * Loop around doing readdir rpc's of size nm_readdirsize 2662 * truncated to a multiple of NFS_DIRFRAGSIZ. 2663 * The stopping criteria is EOF or buffer full. 2664 */ 2665 while (more_dirs && bigenough) { 2666 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2667 bigenough = 0; 2668 break; 2669 } 2670 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2671 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2672 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2673 nfsm_fhtom(dnp, 1); 2674 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2675 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2676 txdr_swapcookie3(uiop->uio_offset, tl); 2677 } else { 2678 txdr_cookie3(uiop->uio_offset, tl); 2679 } 2680 tl += 2; 2681 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2682 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2683 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2684 *tl = txdr_unsigned(nmp->nm_rsize); 2685 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2686 nfsm_postop_attr(vp, attrflag, 0); 2687 if (error) { 2688 m_freem(mrep); 2689 goto nfsmout; 2690 } 2691 nrpcs++; 2692 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2693 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2694 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2695 more_dirs = fxdr_unsigned(int, *tl); 2696 2697 /* loop thru the dir entries, doctoring them to 4bsd form */ 2698 while (more_dirs && bigenough) { 2699 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2700 fileno = fxdr_hyper(tl); 2701 len = fxdr_unsigned(int, *(tl + 2)); 2702 if (len <= 0 || len > NFS_MAXNAMLEN) { 2703 error = EBADRPC; 2704 m_freem(mrep); 2705 goto nfsmout; 2706 } 2707 /* for cookie stashing */ 2708 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2709 left = NFS_DIRFRAGSIZ - blksiz; 2710 if (reclen > left) { 2711 /* 2712 * DIRFRAGSIZ is aligned, no need to align 2713 * again here. 2714 */ 2715 memset(uiop->uio_iov->iov_base, 0, left); 2716 dp->d_reclen += left; 2717 UIO_ADVANCE(uiop, left); 2718 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2719 blksiz = 0; 2720 } 2721 if (reclen > uiop->uio_resid) 2722 bigenough = 0; 2723 if (bigenough) { 2724 int tlen; 2725 2726 dp = (struct dirent *)uiop->uio_iov->iov_base; 2727 dp->d_fileno = fileno; 2728 dp->d_namlen = len; 2729 dp->d_reclen = reclen; 2730 dp->d_type = DT_UNKNOWN; 2731 blksiz += reclen; 2732 if (blksiz == NFS_DIRFRAGSIZ) 2733 blksiz = 0; 2734 UIO_ADVANCE(uiop, DIRHDSIZ); 2735 nfsm_mtouio(uiop, len); 2736 tlen = reclen - (DIRHDSIZ + len); 2737 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2738 UIO_ADVANCE(uiop, tlen); 2739 cnp->cn_nameptr = dp->d_name; 2740 cnp->cn_namelen = dp->d_namlen; 2741 } else 2742 nfsm_adv(nfsm_rndup(len)); 2743 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2744 if (bigenough) { 2745 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2746 uiop->uio_offset = 2747 fxdr_swapcookie3(tl); 2748 else 2749 uiop->uio_offset = 2750 fxdr_cookie3(tl); 2751 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2752 } 2753 tl += 2; 2754 2755 /* 2756 * Since the attributes are before the file handle 2757 * (sigh), we must skip over the attributes and then 2758 * come back and get them. 2759 */ 2760 attrflag = fxdr_unsigned(int, *tl); 2761 if (attrflag) { 2762 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2763 memcpy(&fattr, fp, NFSX_V3FATTR); 2764 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2765 doit = fxdr_unsigned(int, *tl); 2766 if (doit) { 2767 nfsm_getfh(fhp, fhsize, 1); 2768 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2769 vref(vp); 2770 newvp = vp; 2771 np = dnp; 2772 } else { 2773 error = nfs_nget1(vp->v_mount, fhp, 2774 fhsize, &np, LK_NOWAIT); 2775 if (!error) 2776 newvp = NFSTOV(np); 2777 } 2778 if (!error) { 2779 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2780 if (bigenough) { 2781 dp->d_type = 2782 IFTODT(VTTOIF(np->n_vattr->va_type)); 2783 if (cnp->cn_namelen <= NCHNAMLEN) { 2784 ndp->ni_vp = newvp; 2785 nfs_cache_enter(ndp->ni_dvp, 2786 ndp->ni_vp, cnp); 2787 } 2788 } 2789 } 2790 error = 0; 2791 } 2792 } else { 2793 /* Just skip over the file handle */ 2794 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2795 i = fxdr_unsigned(int, *tl); 2796 nfsm_adv(nfsm_rndup(i)); 2797 } 2798 if (newvp != NULLVP) { 2799 if (newvp == vp) 2800 vrele(newvp); 2801 else 2802 vput(newvp); 2803 newvp = NULLVP; 2804 } 2805 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2806 more_dirs = fxdr_unsigned(int, *tl); 2807 } 2808 /* 2809 * If at end of rpc data, get the eof boolean 2810 */ 2811 if (!more_dirs) { 2812 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2813 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2814 2815 /* 2816 * kludge: see a comment in nfs_readdirrpc. 2817 */ 2818 2819 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2820 more_dirs = 0; 2821 } 2822 m_freem(mrep); 2823 } 2824 /* 2825 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2826 * by increasing d_reclen for the last record. 2827 */ 2828 if (blksiz > 0) { 2829 left = NFS_DIRFRAGSIZ - blksiz; 2830 memset(uiop->uio_iov->iov_base, 0, left); 2831 dp->d_reclen += left; 2832 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2833 UIO_ADVANCE(uiop, left); 2834 } 2835 2836 /* 2837 * We are now either at the end of the directory or have filled the 2838 * block. 2839 */ 2840 if (bigenough) { 2841 dnp->n_direofoffset = uiop->uio_offset; 2842 dnp->n_flag |= NEOFVALID; 2843 } 2844 nfsmout: 2845 if (newvp != NULLVP) { 2846 if(newvp == vp) 2847 vrele(newvp); 2848 else 2849 vput(newvp); 2850 } 2851 return (error); 2852 } 2853 #endif 2854 2855 /* 2856 * Silly rename. To make the NFS filesystem that is stateless look a little 2857 * more like the "ufs" a remove of an active vnode is translated to a rename 2858 * to a funny looking filename that is removed by nfs_inactive on the 2859 * nfsnode. There is the potential for another process on a different client 2860 * to create the same funny name between the nfs_lookitup() fails and the 2861 * nfs_rename() completes, but... 2862 */ 2863 int 2864 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2865 { 2866 struct sillyrename *sp; 2867 struct nfsnode *np; 2868 int error; 2869 pid_t pid; 2870 2871 cache_purge(dvp); 2872 np = VTONFS(vp); 2873 #ifndef DIAGNOSTIC 2874 if (vp->v_type == VDIR) 2875 panic("nfs: sillyrename dir"); 2876 #endif 2877 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2878 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2879 sp->s_dvp = dvp; 2880 vref(dvp); 2881 2882 /* Fudge together a funny name */ 2883 pid = curlwp->l_proc->p_pid; 2884 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2885 sp->s_namlen = 12; 2886 sp->s_name[8] = hexdigits[pid & 0xf]; 2887 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2888 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2889 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2890 2891 /* Try lookitups until we get one that isn't there */ 2892 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2893 curlwp, (struct nfsnode **)0) == 0) { 2894 sp->s_name[4]++; 2895 if (sp->s_name[4] > 'z') { 2896 error = EINVAL; 2897 goto bad; 2898 } 2899 } 2900 if (dolink) { 2901 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2902 sp->s_cred, curlwp); 2903 /* 2904 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2905 */ 2906 if (error == ENOTSUP) { 2907 error = nfs_renameit(dvp, cnp, sp); 2908 } 2909 } else { 2910 error = nfs_renameit(dvp, cnp, sp); 2911 } 2912 if (error) 2913 goto bad; 2914 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2915 curlwp, &np); 2916 np->n_sillyrename = sp; 2917 return (0); 2918 bad: 2919 vrele(sp->s_dvp); 2920 kauth_cred_free(sp->s_cred); 2921 kmem_free(sp, sizeof(*sp)); 2922 return (error); 2923 } 2924 2925 /* 2926 * Look up a file name and optionally either update the file handle or 2927 * allocate an nfsnode, depending on the value of npp. 2928 * npp == NULL --> just do the lookup 2929 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2930 * handled too 2931 * *npp != NULL --> update the file handle in the vnode 2932 */ 2933 int 2934 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2935 { 2936 u_int32_t *tl; 2937 char *cp; 2938 int32_t t1, t2; 2939 struct vnode *newvp = (struct vnode *)0; 2940 struct nfsnode *np, *dnp = VTONFS(dvp); 2941 char *bpos, *dpos, *cp2; 2942 int error = 0, fhlen; 2943 #ifndef NFS_V2_ONLY 2944 int attrflag; 2945 #endif 2946 struct mbuf *mreq, *mrep, *md, *mb; 2947 nfsfh_t *nfhp; 2948 const int v3 = NFS_ISV3(dvp); 2949 2950 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2951 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2952 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2953 nfsm_fhtom(dnp, v3); 2954 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2955 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2956 if (npp && !error) { 2957 nfsm_getfh(nfhp, fhlen, v3); 2958 if (*npp) { 2959 np = *npp; 2960 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2961 kmem_free(np->n_fhp, np->n_fhsize); 2962 np->n_fhp = &np->n_fh; 2963 } 2964 #if NFS_SMALLFH < NFSX_V3FHMAX 2965 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2966 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2967 #endif 2968 memcpy(np->n_fhp, nfhp, fhlen); 2969 np->n_fhsize = fhlen; 2970 newvp = NFSTOV(np); 2971 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2972 vref(dvp); 2973 newvp = dvp; 2974 np = dnp; 2975 } else { 2976 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2977 if (error) { 2978 m_freem(mrep); 2979 return (error); 2980 } 2981 newvp = NFSTOV(np); 2982 } 2983 #ifndef NFS_V2_ONLY 2984 if (v3) { 2985 nfsm_postop_attr(newvp, attrflag, 0); 2986 if (!attrflag && *npp == NULL) { 2987 m_freem(mrep); 2988 vput(newvp); 2989 return (ENOENT); 2990 } 2991 } else 2992 #endif 2993 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2994 } 2995 nfsm_reqdone; 2996 if (npp && *npp == NULL) { 2997 if (error) { 2998 if (newvp) 2999 vput(newvp); 3000 } else 3001 *npp = np; 3002 } 3003 return (error); 3004 } 3005 3006 #ifndef NFS_V2_ONLY 3007 /* 3008 * Nfs Version 3 commit rpc 3009 */ 3010 int 3011 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3012 { 3013 char *cp; 3014 u_int32_t *tl; 3015 int32_t t1, t2; 3016 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3017 char *bpos, *dpos, *cp2; 3018 int error = 0, wccflag = NFSV3_WCCRATTR; 3019 struct mbuf *mreq, *mrep, *md, *mb; 3020 struct nfsnode *np; 3021 3022 KASSERT(NFS_ISV3(vp)); 3023 3024 #ifdef NFS_DEBUG_COMMIT 3025 printf("commit %lu - %lu\n", (unsigned long)offset, 3026 (unsigned long)(offset + cnt)); 3027 #endif 3028 3029 mutex_enter(&nmp->nm_lock); 3030 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3031 mutex_exit(&nmp->nm_lock); 3032 return (0); 3033 } 3034 mutex_exit(&nmp->nm_lock); 3035 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3036 np = VTONFS(vp); 3037 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3038 nfsm_fhtom(np, 1); 3039 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3040 txdr_hyper(offset, tl); 3041 tl += 2; 3042 *tl = txdr_unsigned(cnt); 3043 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3044 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3045 if (!error) { 3046 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3047 mutex_enter(&nmp->nm_lock); 3048 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3049 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3050 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3051 error = NFSERR_STALEWRITEVERF; 3052 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3053 } 3054 mutex_exit(&nmp->nm_lock); 3055 } 3056 nfsm_reqdone; 3057 return (error); 3058 } 3059 #endif 3060 3061 /* 3062 * Kludge City.. 3063 * - make nfs_bmap() essentially a no-op that does no translation 3064 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3065 * (Maybe I could use the process's page mapping, but I was concerned that 3066 * Kernel Write might not be enabled and also figured copyout() would do 3067 * a lot more work than memcpy() and also it currently happens in the 3068 * context of the swapper process (2). 3069 */ 3070 int 3071 nfs_bmap(void *v) 3072 { 3073 struct vop_bmap_args /* { 3074 struct vnode *a_vp; 3075 daddr_t a_bn; 3076 struct vnode **a_vpp; 3077 daddr_t *a_bnp; 3078 int *a_runp; 3079 } */ *ap = v; 3080 struct vnode *vp = ap->a_vp; 3081 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3082 3083 if (ap->a_vpp != NULL) 3084 *ap->a_vpp = vp; 3085 if (ap->a_bnp != NULL) 3086 *ap->a_bnp = ap->a_bn << bshift; 3087 if (ap->a_runp != NULL) 3088 *ap->a_runp = 1024 * 1024; /* XXX */ 3089 return (0); 3090 } 3091 3092 /* 3093 * Strategy routine. 3094 * For async requests when nfsiod(s) are running, queue the request by 3095 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3096 * request. 3097 */ 3098 int 3099 nfs_strategy(void *v) 3100 { 3101 struct vop_strategy_args *ap = v; 3102 struct buf *bp = ap->a_bp; 3103 int error = 0; 3104 3105 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3106 panic("nfs physio/async"); 3107 3108 /* 3109 * If the op is asynchronous and an i/o daemon is waiting 3110 * queue the request, wake it up and wait for completion 3111 * otherwise just do it ourselves. 3112 */ 3113 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3114 error = nfs_doio(bp); 3115 return (error); 3116 } 3117 3118 /* 3119 * fsync vnode op. Just call nfs_flush() with commit == 1. 3120 */ 3121 /* ARGSUSED */ 3122 int 3123 nfs_fsync(void *v) 3124 { 3125 struct vop_fsync_args /* { 3126 struct vnodeop_desc *a_desc; 3127 struct vnode * a_vp; 3128 kauth_cred_t a_cred; 3129 int a_flags; 3130 off_t offlo; 3131 off_t offhi; 3132 struct lwp * a_l; 3133 } */ *ap = v; 3134 3135 struct vnode *vp = ap->a_vp; 3136 3137 if (vp->v_type != VREG) 3138 return 0; 3139 3140 return (nfs_flush(vp, ap->a_cred, 3141 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3142 } 3143 3144 /* 3145 * Flush all the data associated with a vnode. 3146 */ 3147 int 3148 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3149 int commit) 3150 { 3151 struct nfsnode *np = VTONFS(vp); 3152 int error; 3153 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3154 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3155 3156 mutex_enter(vp->v_interlock); 3157 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3158 if (np->n_flag & NWRITEERR) { 3159 error = np->n_error; 3160 np->n_flag &= ~NWRITEERR; 3161 } 3162 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3163 return (error); 3164 } 3165 3166 /* 3167 * Return POSIX pathconf information applicable to nfs. 3168 * 3169 * N.B. The NFS V2 protocol doesn't support this RPC. 3170 */ 3171 /* ARGSUSED */ 3172 int 3173 nfs_pathconf(void *v) 3174 { 3175 struct vop_pathconf_args /* { 3176 struct vnode *a_vp; 3177 int a_name; 3178 register_t *a_retval; 3179 } */ *ap = v; 3180 struct nfsv3_pathconf *pcp; 3181 struct vnode *vp = ap->a_vp; 3182 struct mbuf *mreq, *mrep, *md, *mb; 3183 int32_t t1, t2; 3184 u_int32_t *tl; 3185 char *bpos, *dpos, *cp, *cp2; 3186 int error = 0, attrflag; 3187 #ifndef NFS_V2_ONLY 3188 struct nfsmount *nmp; 3189 unsigned int l; 3190 u_int64_t maxsize; 3191 #endif 3192 const int v3 = NFS_ISV3(vp); 3193 struct nfsnode *np = VTONFS(vp); 3194 3195 switch (ap->a_name) { 3196 /* Names that can be resolved locally. */ 3197 case _PC_PIPE_BUF: 3198 *ap->a_retval = PIPE_BUF; 3199 break; 3200 case _PC_SYNC_IO: 3201 *ap->a_retval = 1; 3202 break; 3203 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3204 case _PC_LINK_MAX: 3205 case _PC_NAME_MAX: 3206 case _PC_CHOWN_RESTRICTED: 3207 case _PC_NO_TRUNC: 3208 if (!v3) { 3209 error = EINVAL; 3210 break; 3211 } 3212 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3213 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3214 nfsm_fhtom(np, 1); 3215 nfsm_request(np, NFSPROC_PATHCONF, 3216 curlwp, curlwp->l_cred); /* XXX */ 3217 nfsm_postop_attr(vp, attrflag, 0); 3218 if (!error) { 3219 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3220 NFSX_V3PATHCONF); 3221 switch (ap->a_name) { 3222 case _PC_LINK_MAX: 3223 *ap->a_retval = 3224 fxdr_unsigned(register_t, pcp->pc_linkmax); 3225 break; 3226 case _PC_NAME_MAX: 3227 *ap->a_retval = 3228 fxdr_unsigned(register_t, pcp->pc_namemax); 3229 break; 3230 case _PC_CHOWN_RESTRICTED: 3231 *ap->a_retval = 3232 (pcp->pc_chownrestricted == nfs_true); 3233 break; 3234 case _PC_NO_TRUNC: 3235 *ap->a_retval = 3236 (pcp->pc_notrunc == nfs_true); 3237 break; 3238 } 3239 } 3240 nfsm_reqdone; 3241 break; 3242 case _PC_FILESIZEBITS: 3243 #ifndef NFS_V2_ONLY 3244 if (v3) { 3245 nmp = VFSTONFS(vp->v_mount); 3246 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3247 if ((error = nfs_fsinfo(nmp, vp, 3248 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3249 break; 3250 for (l = 0, maxsize = nmp->nm_maxfilesize; 3251 (maxsize >> l) > 0; l++) 3252 ; 3253 *ap->a_retval = l + 1; 3254 } else 3255 #endif 3256 { 3257 *ap->a_retval = 32; /* NFS V2 limitation */ 3258 } 3259 break; 3260 default: 3261 error = EINVAL; 3262 break; 3263 } 3264 3265 return (error); 3266 } 3267 3268 /* 3269 * NFS advisory byte-level locks. 3270 */ 3271 int 3272 nfs_advlock(void *v) 3273 { 3274 struct vop_advlock_args /* { 3275 struct vnode *a_vp; 3276 void *a_id; 3277 int a_op; 3278 struct flock *a_fl; 3279 int a_flags; 3280 } */ *ap = v; 3281 struct nfsnode *np = VTONFS(ap->a_vp); 3282 3283 return lf_advlock(ap, &np->n_lockf, np->n_size); 3284 } 3285 3286 /* 3287 * Print out the contents of an nfsnode. 3288 */ 3289 int 3290 nfs_print(void *v) 3291 { 3292 struct vop_print_args /* { 3293 struct vnode *a_vp; 3294 } */ *ap = v; 3295 struct vnode *vp = ap->a_vp; 3296 struct nfsnode *np = VTONFS(vp); 3297 3298 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3299 (unsigned long long)np->n_vattr->va_fileid, 3300 (unsigned long long)np->n_vattr->va_fsid); 3301 if (vp->v_type == VFIFO) 3302 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3303 printf("\n"); 3304 return (0); 3305 } 3306 3307 /* 3308 * nfs unlock wrapper. 3309 */ 3310 int 3311 nfs_unlock(void *v) 3312 { 3313 struct vop_unlock_args /* { 3314 struct vnode *a_vp; 3315 int a_flags; 3316 } */ *ap = v; 3317 struct vnode *vp = ap->a_vp; 3318 3319 /* 3320 * VOP_UNLOCK can be called by nfs_loadattrcache 3321 * with v_data == 0. 3322 */ 3323 if (VTONFS(vp)) { 3324 nfs_delayedtruncate(vp); 3325 } 3326 3327 return genfs_unlock(v); 3328 } 3329 3330 /* 3331 * nfs special file access vnode op. 3332 * Essentially just get vattr and then imitate iaccess() since the device is 3333 * local to the client. 3334 */ 3335 int 3336 nfsspec_access(void *v) 3337 { 3338 struct vop_access_args /* { 3339 struct vnode *a_vp; 3340 int a_mode; 3341 kauth_cred_t a_cred; 3342 struct lwp *a_l; 3343 } */ *ap = v; 3344 struct vattr va; 3345 struct vnode *vp = ap->a_vp; 3346 int error; 3347 3348 error = VOP_GETATTR(vp, &va, ap->a_cred); 3349 if (error) 3350 return (error); 3351 3352 /* 3353 * Disallow write attempts on filesystems mounted read-only; 3354 * unless the file is a socket, fifo, or a block or character 3355 * device resident on the filesystem. 3356 */ 3357 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3358 switch (vp->v_type) { 3359 case VREG: 3360 case VDIR: 3361 case VLNK: 3362 return (EROFS); 3363 default: 3364 break; 3365 } 3366 } 3367 3368 return kauth_authorize_vnode(ap->a_cred, KAUTH_ACCESS_ACTION(ap->a_mode, 3369 va.va_type, va.va_mode), vp, NULL, genfs_can_access(va.va_type, 3370 va.va_mode, va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3371 } 3372 3373 /* 3374 * Read wrapper for special devices. 3375 */ 3376 int 3377 nfsspec_read(void *v) 3378 { 3379 struct vop_read_args /* { 3380 struct vnode *a_vp; 3381 struct uio *a_uio; 3382 int a_ioflag; 3383 kauth_cred_t a_cred; 3384 } */ *ap = v; 3385 struct nfsnode *np = VTONFS(ap->a_vp); 3386 3387 /* 3388 * Set access flag. 3389 */ 3390 np->n_flag |= NACC; 3391 getnanotime(&np->n_atim); 3392 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3393 } 3394 3395 /* 3396 * Write wrapper for special devices. 3397 */ 3398 int 3399 nfsspec_write(void *v) 3400 { 3401 struct vop_write_args /* { 3402 struct vnode *a_vp; 3403 struct uio *a_uio; 3404 int a_ioflag; 3405 kauth_cred_t a_cred; 3406 } */ *ap = v; 3407 struct nfsnode *np = VTONFS(ap->a_vp); 3408 3409 /* 3410 * Set update flag. 3411 */ 3412 np->n_flag |= NUPD; 3413 getnanotime(&np->n_mtim); 3414 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3415 } 3416 3417 /* 3418 * Close wrapper for special devices. 3419 * 3420 * Update the times on the nfsnode then do device close. 3421 */ 3422 int 3423 nfsspec_close(void *v) 3424 { 3425 struct vop_close_args /* { 3426 struct vnode *a_vp; 3427 int a_fflag; 3428 kauth_cred_t a_cred; 3429 struct lwp *a_l; 3430 } */ *ap = v; 3431 struct vnode *vp = ap->a_vp; 3432 struct nfsnode *np = VTONFS(vp); 3433 struct vattr vattr; 3434 3435 if (np->n_flag & (NACC | NUPD)) { 3436 np->n_flag |= NCHG; 3437 if (vp->v_usecount == 1 && 3438 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3439 vattr_null(&vattr); 3440 if (np->n_flag & NACC) 3441 vattr.va_atime = np->n_atim; 3442 if (np->n_flag & NUPD) 3443 vattr.va_mtime = np->n_mtim; 3444 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3445 } 3446 } 3447 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3448 } 3449 3450 /* 3451 * Read wrapper for fifos. 3452 */ 3453 int 3454 nfsfifo_read(void *v) 3455 { 3456 struct vop_read_args /* { 3457 struct vnode *a_vp; 3458 struct uio *a_uio; 3459 int a_ioflag; 3460 kauth_cred_t a_cred; 3461 } */ *ap = v; 3462 struct nfsnode *np = VTONFS(ap->a_vp); 3463 3464 /* 3465 * Set access flag. 3466 */ 3467 np->n_flag |= NACC; 3468 getnanotime(&np->n_atim); 3469 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3470 } 3471 3472 /* 3473 * Write wrapper for fifos. 3474 */ 3475 int 3476 nfsfifo_write(void *v) 3477 { 3478 struct vop_write_args /* { 3479 struct vnode *a_vp; 3480 struct uio *a_uio; 3481 int a_ioflag; 3482 kauth_cred_t a_cred; 3483 } */ *ap = v; 3484 struct nfsnode *np = VTONFS(ap->a_vp); 3485 3486 /* 3487 * Set update flag. 3488 */ 3489 np->n_flag |= NUPD; 3490 getnanotime(&np->n_mtim); 3491 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3492 } 3493 3494 /* 3495 * Close wrapper for fifos. 3496 * 3497 * Update the times on the nfsnode then do fifo close. 3498 */ 3499 int 3500 nfsfifo_close(void *v) 3501 { 3502 struct vop_close_args /* { 3503 struct vnode *a_vp; 3504 int a_fflag; 3505 kauth_cred_t a_cred; 3506 struct lwp *a_l; 3507 } */ *ap = v; 3508 struct vnode *vp = ap->a_vp; 3509 struct nfsnode *np = VTONFS(vp); 3510 struct vattr vattr; 3511 3512 if (np->n_flag & (NACC | NUPD)) { 3513 struct timespec ts; 3514 3515 getnanotime(&ts); 3516 if (np->n_flag & NACC) 3517 np->n_atim = ts; 3518 if (np->n_flag & NUPD) 3519 np->n_mtim = ts; 3520 np->n_flag |= NCHG; 3521 if (vp->v_usecount == 1 && 3522 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3523 vattr_null(&vattr); 3524 if (np->n_flag & NACC) 3525 vattr.va_atime = np->n_atim; 3526 if (np->n_flag & NUPD) 3527 vattr.va_mtime = np->n_mtim; 3528 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3529 } 3530 } 3531 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3532 } 3533
Indexes created Thu Sep 25 16:09:42 GMT 2025