nfs_vnops.c revision 1.302
1 /* $NetBSD: nfs_vnops.c,v 1.302 2014/01/17 10:55:02 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 1.302 2014/01/17 10:55:02 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 } 1553 VTONFS(dvp)->n_flag |= NMODIFIED; 1554 if (!wccflag) 1555 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1556 return (error); 1557 } 1558 1559 /* 1560 * nfs mknod vop 1561 * just call nfs_mknodrpc() to do the work. 1562 */ 1563 /* ARGSUSED */ 1564 int 1565 nfs_mknod(void *v) 1566 { 1567 struct vop_mknod_v2_args /* { 1568 struct vnode *a_dvp; 1569 struct vnode **a_vpp; 1570 struct componentname *a_cnp; 1571 struct vattr *a_vap; 1572 } */ *ap = v; 1573 struct vnode *dvp = ap->a_dvp; 1574 struct componentname *cnp = ap->a_cnp; 1575 int error; 1576 1577 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1578 VN_KNOTE(dvp, NOTE_WRITE); 1579 if (error == 0 || error == EEXIST) 1580 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1581 return (error); 1582 } 1583 1584 /* 1585 * nfs file create call 1586 */ 1587 int 1588 nfs_create(void *v) 1589 { 1590 struct vop_create_v2_args /* { 1591 struct vnode *a_dvp; 1592 struct vnode **a_vpp; 1593 struct componentname *a_cnp; 1594 struct vattr *a_vap; 1595 } */ *ap = v; 1596 struct vnode *dvp = ap->a_dvp; 1597 struct vattr *vap = ap->a_vap; 1598 struct componentname *cnp = ap->a_cnp; 1599 struct nfsv2_sattr *sp; 1600 u_int32_t *tl; 1601 char *cp; 1602 int32_t t1, t2; 1603 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1604 struct vnode *newvp = (struct vnode *)0; 1605 char *bpos, *dpos, *cp2; 1606 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1607 struct mbuf *mreq, *mrep, *md, *mb; 1608 const int v3 = NFS_ISV3(dvp); 1609 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1610 1611 /* 1612 * Oops, not for me.. 1613 */ 1614 if (vap->va_type == VSOCK) 1615 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1616 1617 KASSERT(vap->va_type == VREG); 1618 1619 #ifdef VA_EXCLUSIVE 1620 if (vap->va_vaflags & VA_EXCLUSIVE) { 1621 excl_mode = NFSV3CREATE_EXCLUSIVE; 1622 } 1623 #endif 1624 again: 1625 error = 0; 1626 nfsstats.rpccnt[NFSPROC_CREATE]++; 1627 dnp = VTONFS(dvp); 1628 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1629 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1630 nfsm_fhtom(dnp, v3); 1631 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1632 #ifndef NFS_V2_ONLY 1633 if (v3) { 1634 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1635 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1636 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1637 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1638 *tl++ = cprng_fast32(); 1639 *tl = cprng_fast32(); 1640 } else { 1641 *tl = txdr_unsigned(excl_mode); 1642 nfsm_v3attrbuild(vap, false); 1643 } 1644 } else 1645 #endif 1646 { 1647 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1648 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1649 sp->sa_uid = nfs_xdrneg1; 1650 sp->sa_gid = nfs_xdrneg1; 1651 sp->sa_size = 0; 1652 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1653 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1654 } 1655 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1656 if (!error) { 1657 nfsm_mtofh(dvp, newvp, v3, gotvp); 1658 if (!gotvp) { 1659 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1660 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1661 if (!error) 1662 newvp = NFSTOV(np); 1663 } 1664 } 1665 #ifndef NFS_V2_ONLY 1666 if (v3) 1667 nfsm_wcc_data(dvp, wccflag, 0, !error); 1668 #endif 1669 nfsm_reqdone; 1670 if (error) { 1671 /* 1672 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1673 */ 1674 if (v3 && error == ENOTSUP) { 1675 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1676 excl_mode = NFSV3CREATE_GUARDED; 1677 goto again; 1678 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1679 excl_mode = NFSV3CREATE_UNCHECKED; 1680 goto again; 1681 } 1682 } 1683 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1684 struct timespec ts; 1685 1686 getnanotime(&ts); 1687 1688 /* 1689 * make sure that we'll update timestamps as 1690 * most server implementations use them to store 1691 * the create verifier. 1692 * 1693 * XXX it's better to use TOSERVER always. 1694 */ 1695 1696 if (vap->va_atime.tv_sec == VNOVAL) 1697 vap->va_atime = ts; 1698 if (vap->va_mtime.tv_sec == VNOVAL) 1699 vap->va_mtime = ts; 1700 1701 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1702 } 1703 if (error == 0) { 1704 if (cnp->cn_flags & MAKEENTRY) 1705 nfs_cache_enter(dvp, newvp, cnp); 1706 else 1707 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1708 *ap->a_vpp = newvp; 1709 } else { 1710 if (newvp) 1711 vput(newvp); 1712 if (error == EEXIST) 1713 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1714 } 1715 VTONFS(dvp)->n_flag |= NMODIFIED; 1716 if (!wccflag) 1717 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1718 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1719 return (error); 1720 } 1721 1722 /* 1723 * nfs file remove call 1724 * To try and make nfs semantics closer to ufs semantics, a file that has 1725 * other processes using the vnode is renamed instead of removed and then 1726 * removed later on the last close. 1727 * - If v_usecount > 1 1728 * If a rename is not already in the works 1729 * call nfs_sillyrename() to set it up 1730 * else 1731 * do the remove rpc 1732 */ 1733 int 1734 nfs_remove(void *v) 1735 { 1736 struct vop_remove_args /* { 1737 struct vnodeop_desc *a_desc; 1738 struct vnode * a_dvp; 1739 struct vnode * a_vp; 1740 struct componentname * a_cnp; 1741 } */ *ap = v; 1742 struct vnode *vp = ap->a_vp; 1743 struct vnode *dvp = ap->a_dvp; 1744 struct componentname *cnp = ap->a_cnp; 1745 struct nfsnode *np = VTONFS(vp); 1746 int error = 0; 1747 struct vattr vattr; 1748 1749 #ifndef DIAGNOSTIC 1750 if (vp->v_usecount < 1) 1751 panic("nfs_remove: bad v_usecount"); 1752 #endif 1753 if (vp->v_type == VDIR) 1754 error = EPERM; 1755 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1756 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1757 vattr.va_nlink > 1)) { 1758 /* 1759 * Purge the name cache so that the chance of a lookup for 1760 * the name succeeding while the remove is in progress is 1761 * minimized. Without node locking it can still happen, such 1762 * that an I/O op returns ESTALE, but since you get this if 1763 * another host removes the file.. 1764 */ 1765 cache_purge(vp); 1766 /* 1767 * throw away biocache buffers, mainly to avoid 1768 * unnecessary delayed writes later. 1769 */ 1770 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1771 /* Do the rpc */ 1772 if (error != EINTR) 1773 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1774 cnp->cn_namelen, cnp->cn_cred, curlwp); 1775 } else if (!np->n_sillyrename) 1776 error = nfs_sillyrename(dvp, vp, cnp, false); 1777 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1778 vattr.va_nlink == 1) { 1779 np->n_flag |= NREMOVED; 1780 } 1781 NFS_INVALIDATE_ATTRCACHE(np); 1782 VN_KNOTE(vp, NOTE_DELETE); 1783 VN_KNOTE(dvp, NOTE_WRITE); 1784 if (dvp == vp) 1785 vrele(vp); 1786 else 1787 vput(vp); 1788 vput(dvp); 1789 return (error); 1790 } 1791 1792 /* 1793 * nfs file remove rpc called from nfs_inactive 1794 */ 1795 int 1796 nfs_removeit(struct sillyrename *sp) 1797 { 1798 1799 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1800 (struct lwp *)0)); 1801 } 1802 1803 /* 1804 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1805 */ 1806 int 1807 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1808 { 1809 u_int32_t *tl; 1810 char *cp; 1811 #ifndef NFS_V2_ONLY 1812 int32_t t1; 1813 char *cp2; 1814 #endif 1815 int32_t t2; 1816 char *bpos, *dpos; 1817 int error = 0, wccflag = NFSV3_WCCRATTR; 1818 struct mbuf *mreq, *mrep, *md, *mb; 1819 const int v3 = NFS_ISV3(dvp); 1820 int rexmit = 0; 1821 struct nfsnode *dnp = VTONFS(dvp); 1822 1823 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1824 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1825 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1826 nfsm_fhtom(dnp, v3); 1827 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1828 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1829 #ifndef NFS_V2_ONLY 1830 if (v3) 1831 nfsm_wcc_data(dvp, wccflag, 0, !error); 1832 #endif 1833 nfsm_reqdone; 1834 VTONFS(dvp)->n_flag |= NMODIFIED; 1835 if (!wccflag) 1836 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1837 /* 1838 * Kludge City: If the first reply to the remove rpc is lost.. 1839 * the reply to the retransmitted request will be ENOENT 1840 * since the file was in fact removed 1841 * Therefore, we cheat and return success. 1842 */ 1843 if (rexmit && error == ENOENT) 1844 error = 0; 1845 return (error); 1846 } 1847 1848 /* 1849 * nfs file rename call 1850 */ 1851 int 1852 nfs_rename(void *v) 1853 { 1854 struct vop_rename_args /* { 1855 struct vnode *a_fdvp; 1856 struct vnode *a_fvp; 1857 struct componentname *a_fcnp; 1858 struct vnode *a_tdvp; 1859 struct vnode *a_tvp; 1860 struct componentname *a_tcnp; 1861 } */ *ap = v; 1862 struct vnode *fvp = ap->a_fvp; 1863 struct vnode *tvp = ap->a_tvp; 1864 struct vnode *fdvp = ap->a_fdvp; 1865 struct vnode *tdvp = ap->a_tdvp; 1866 struct componentname *tcnp = ap->a_tcnp; 1867 struct componentname *fcnp = ap->a_fcnp; 1868 int error; 1869 1870 /* Check for cross-device rename */ 1871 if ((fvp->v_mount != tdvp->v_mount) || 1872 (tvp && (fvp->v_mount != tvp->v_mount))) { 1873 error = EXDEV; 1874 goto out; 1875 } 1876 1877 /* 1878 * If the tvp exists and is in use, sillyrename it before doing the 1879 * rename of the new file over it. 1880 * 1881 * Have sillyrename use link instead of rename if possible, 1882 * so that we don't lose the file if the rename fails, and so 1883 * that there's no window when the "to" file doesn't exist. 1884 */ 1885 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1886 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1887 VN_KNOTE(tvp, NOTE_DELETE); 1888 vput(tvp); 1889 tvp = NULL; 1890 } 1891 1892 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1893 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1894 curlwp); 1895 1896 VN_KNOTE(fdvp, NOTE_WRITE); 1897 VN_KNOTE(tdvp, NOTE_WRITE); 1898 if (error == 0 || error == EEXIST) { 1899 if (fvp->v_type == VDIR) 1900 cache_purge(fvp); 1901 else 1902 cache_purge1(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1903 0); 1904 if (tvp != NULL && tvp->v_type == VDIR) 1905 cache_purge(tvp); 1906 else 1907 cache_purge1(tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, 1908 0); 1909 } 1910 out: 1911 if (tdvp == tvp) 1912 vrele(tdvp); 1913 else 1914 vput(tdvp); 1915 if (tvp) 1916 vput(tvp); 1917 vrele(fdvp); 1918 vrele(fvp); 1919 return (error); 1920 } 1921 1922 /* 1923 * nfs file rename rpc called from nfs_remove() above 1924 */ 1925 int 1926 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1927 { 1928 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1929 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1930 } 1931 1932 /* 1933 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1934 */ 1935 int 1936 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) 1937 { 1938 u_int32_t *tl; 1939 char *cp; 1940 #ifndef NFS_V2_ONLY 1941 int32_t t1; 1942 char *cp2; 1943 #endif 1944 int32_t t2; 1945 char *bpos, *dpos; 1946 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1947 struct mbuf *mreq, *mrep, *md, *mb; 1948 const int v3 = NFS_ISV3(fdvp); 1949 int rexmit = 0; 1950 struct nfsnode *fdnp = VTONFS(fdvp); 1951 1952 nfsstats.rpccnt[NFSPROC_RENAME]++; 1953 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1954 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1955 nfsm_rndup(tnamelen)); 1956 nfsm_fhtom(fdnp, v3); 1957 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1958 nfsm_fhtom(VTONFS(tdvp), v3); 1959 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1960 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1961 #ifndef NFS_V2_ONLY 1962 if (v3) { 1963 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1964 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1965 } 1966 #endif 1967 nfsm_reqdone; 1968 VTONFS(fdvp)->n_flag |= NMODIFIED; 1969 VTONFS(tdvp)->n_flag |= NMODIFIED; 1970 if (!fwccflag) 1971 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1972 if (!twccflag) 1973 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1974 /* 1975 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1976 */ 1977 if (rexmit && error == ENOENT) 1978 error = 0; 1979 return (error); 1980 } 1981 1982 /* 1983 * NFS link RPC, called from nfs_link. 1984 * Assumes dvp and vp locked, and leaves them that way. 1985 */ 1986 1987 static int 1988 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 1989 size_t namelen, kauth_cred_t cred, struct lwp *l) 1990 { 1991 u_int32_t *tl; 1992 char *cp; 1993 #ifndef NFS_V2_ONLY 1994 int32_t t1; 1995 char *cp2; 1996 #endif 1997 int32_t t2; 1998 char *bpos, *dpos; 1999 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2000 struct mbuf *mreq, *mrep, *md, *mb; 2001 const int v3 = NFS_ISV3(dvp); 2002 int rexmit = 0; 2003 struct nfsnode *np = VTONFS(vp); 2004 2005 nfsstats.rpccnt[NFSPROC_LINK]++; 2006 nfsm_reqhead(np, NFSPROC_LINK, 2007 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2008 nfsm_fhtom(np, v3); 2009 nfsm_fhtom(VTONFS(dvp), v3); 2010 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2011 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2012 #ifndef NFS_V2_ONLY 2013 if (v3) { 2014 nfsm_postop_attr(vp, attrflag, 0); 2015 nfsm_wcc_data(dvp, wccflag, 0, !error); 2016 } 2017 #endif 2018 nfsm_reqdone; 2019 2020 VTONFS(dvp)->n_flag |= NMODIFIED; 2021 if (!attrflag) 2022 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2023 if (!wccflag) 2024 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2025 2026 /* 2027 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2028 */ 2029 if (rexmit && error == EEXIST) 2030 error = 0; 2031 2032 return error; 2033 } 2034 2035 /* 2036 * nfs hard link create call 2037 */ 2038 int 2039 nfs_link(void *v) 2040 { 2041 struct vop_link_args /* { 2042 struct vnode *a_dvp; 2043 struct vnode *a_vp; 2044 struct componentname *a_cnp; 2045 } */ *ap = v; 2046 struct vnode *vp = ap->a_vp; 2047 struct vnode *dvp = ap->a_dvp; 2048 struct componentname *cnp = ap->a_cnp; 2049 int error = 0; 2050 2051 error = vn_lock(vp, LK_EXCLUSIVE); 2052 if (error != 0) { 2053 VOP_ABORTOP(dvp, cnp); 2054 vput(dvp); 2055 return error; 2056 } 2057 2058 /* 2059 * Push all writes to the server, so that the attribute cache 2060 * doesn't get "out of sync" with the server. 2061 * XXX There should be a better way! 2062 */ 2063 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2064 2065 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2066 cnp->cn_cred, curlwp); 2067 2068 if (error == 0) { 2069 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2070 } 2071 VOP_UNLOCK(vp); 2072 VN_KNOTE(vp, NOTE_LINK); 2073 VN_KNOTE(dvp, NOTE_WRITE); 2074 vput(dvp); 2075 return (error); 2076 } 2077 2078 /* 2079 * nfs symbolic link create call 2080 */ 2081 int 2082 nfs_symlink(void *v) 2083 { 2084 struct vop_symlink_v2_args /* { 2085 struct vnode *a_dvp; 2086 struct vnode **a_vpp; 2087 struct componentname *a_cnp; 2088 struct vattr *a_vap; 2089 char *a_target; 2090 } */ *ap = v; 2091 struct vnode *dvp = ap->a_dvp; 2092 struct vattr *vap = ap->a_vap; 2093 struct componentname *cnp = ap->a_cnp; 2094 struct nfsv2_sattr *sp; 2095 u_int32_t *tl; 2096 char *cp; 2097 int32_t t1, t2; 2098 char *bpos, *dpos, *cp2; 2099 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2100 struct mbuf *mreq, *mrep, *md, *mb; 2101 struct vnode *newvp = (struct vnode *)0; 2102 const int v3 = NFS_ISV3(dvp); 2103 int rexmit = 0; 2104 struct nfsnode *dnp = VTONFS(dvp); 2105 2106 *ap->a_vpp = NULL; 2107 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2108 slen = strlen(ap->a_target); 2109 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2110 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2111 nfsm_fhtom(dnp, v3); 2112 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2113 #ifndef NFS_V2_ONlY 2114 if (v3) 2115 nfsm_v3attrbuild(vap, false); 2116 #endif 2117 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2118 #ifndef NFS_V2_ONlY 2119 if (!v3) { 2120 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2121 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2122 sp->sa_uid = nfs_xdrneg1; 2123 sp->sa_gid = nfs_xdrneg1; 2124 sp->sa_size = nfs_xdrneg1; 2125 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2126 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2127 } 2128 #endif 2129 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2130 &rexmit); 2131 #ifndef NFS_V2_ONlY 2132 if (v3) { 2133 if (!error) 2134 nfsm_mtofh(dvp, newvp, v3, gotvp); 2135 nfsm_wcc_data(dvp, wccflag, 0, !error); 2136 } 2137 #endif 2138 nfsm_reqdone; 2139 /* 2140 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2141 */ 2142 if (rexmit && error == EEXIST) 2143 error = 0; 2144 if (error == 0 || error == EEXIST) 2145 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2146 if (error == 0 && newvp == NULL) { 2147 struct nfsnode *np = NULL; 2148 2149 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2150 cnp->cn_cred, curlwp, &np); 2151 if (error == 0) 2152 newvp = NFSTOV(np); 2153 } 2154 if (error) { 2155 if (newvp != NULL) 2156 vput(newvp); 2157 } else { 2158 *ap->a_vpp = newvp; 2159 } 2160 VTONFS(dvp)->n_flag |= NMODIFIED; 2161 if (!wccflag) 2162 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2163 VN_KNOTE(dvp, NOTE_WRITE); 2164 return (error); 2165 } 2166 2167 /* 2168 * nfs make dir call 2169 */ 2170 int 2171 nfs_mkdir(void *v) 2172 { 2173 struct vop_mkdir_v2_args /* { 2174 struct vnode *a_dvp; 2175 struct vnode **a_vpp; 2176 struct componentname *a_cnp; 2177 struct vattr *a_vap; 2178 } */ *ap = v; 2179 struct vnode *dvp = ap->a_dvp; 2180 struct vattr *vap = ap->a_vap; 2181 struct componentname *cnp = ap->a_cnp; 2182 struct nfsv2_sattr *sp; 2183 u_int32_t *tl; 2184 char *cp; 2185 int32_t t1, t2; 2186 int len; 2187 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2188 struct vnode *newvp = (struct vnode *)0; 2189 char *bpos, *dpos, *cp2; 2190 int error = 0, wccflag = NFSV3_WCCRATTR; 2191 int gotvp = 0; 2192 int rexmit = 0; 2193 struct mbuf *mreq, *mrep, *md, *mb; 2194 const int v3 = NFS_ISV3(dvp); 2195 2196 len = cnp->cn_namelen; 2197 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2198 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2199 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2200 nfsm_fhtom(dnp, v3); 2201 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2202 #ifndef NFS_V2_ONLY 2203 if (v3) { 2204 nfsm_v3attrbuild(vap, false); 2205 } else 2206 #endif 2207 { 2208 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2209 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2210 sp->sa_uid = nfs_xdrneg1; 2211 sp->sa_gid = nfs_xdrneg1; 2212 sp->sa_size = nfs_xdrneg1; 2213 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2214 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2215 } 2216 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2217 if (!error) 2218 nfsm_mtofh(dvp, newvp, v3, gotvp); 2219 if (v3) 2220 nfsm_wcc_data(dvp, wccflag, 0, !error); 2221 nfsm_reqdone; 2222 VTONFS(dvp)->n_flag |= NMODIFIED; 2223 if (!wccflag) 2224 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2225 /* 2226 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2227 * if we can succeed in looking up the directory. 2228 */ 2229 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2230 if (newvp) { 2231 vput(newvp); 2232 newvp = (struct vnode *)0; 2233 } 2234 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2235 curlwp, &np); 2236 if (!error) { 2237 newvp = NFSTOV(np); 2238 if (newvp->v_type != VDIR || newvp == dvp) 2239 error = EEXIST; 2240 } 2241 } 2242 if (error) { 2243 if (newvp) { 2244 if (dvp != newvp) 2245 vput(newvp); 2246 else 2247 vrele(newvp); 2248 } 2249 } else { 2250 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2251 nfs_cache_enter(dvp, newvp, cnp); 2252 *ap->a_vpp = newvp; 2253 } 2254 return (error); 2255 } 2256 2257 /* 2258 * nfs remove directory call 2259 */ 2260 int 2261 nfs_rmdir(void *v) 2262 { 2263 struct vop_rmdir_args /* { 2264 struct vnode *a_dvp; 2265 struct vnode *a_vp; 2266 struct componentname *a_cnp; 2267 } */ *ap = v; 2268 struct vnode *vp = ap->a_vp; 2269 struct vnode *dvp = ap->a_dvp; 2270 struct componentname *cnp = ap->a_cnp; 2271 u_int32_t *tl; 2272 char *cp; 2273 #ifndef NFS_V2_ONLY 2274 int32_t t1; 2275 char *cp2; 2276 #endif 2277 int32_t t2; 2278 char *bpos, *dpos; 2279 int error = 0, wccflag = NFSV3_WCCRATTR; 2280 int rexmit = 0; 2281 struct mbuf *mreq, *mrep, *md, *mb; 2282 const int v3 = NFS_ISV3(dvp); 2283 struct nfsnode *dnp; 2284 2285 if (dvp == vp) { 2286 vrele(dvp); 2287 vput(dvp); 2288 return (EINVAL); 2289 } 2290 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2291 dnp = VTONFS(dvp); 2292 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2293 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2294 nfsm_fhtom(dnp, v3); 2295 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2296 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2297 #ifndef NFS_V2_ONLY 2298 if (v3) 2299 nfsm_wcc_data(dvp, wccflag, 0, !error); 2300 #endif 2301 nfsm_reqdone; 2302 VTONFS(dvp)->n_flag |= NMODIFIED; 2303 if (!wccflag) 2304 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2305 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2306 VN_KNOTE(vp, NOTE_DELETE); 2307 cache_purge(vp); 2308 vput(vp); 2309 vput(dvp); 2310 /* 2311 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2312 */ 2313 if (rexmit && error == ENOENT) 2314 error = 0; 2315 return (error); 2316 } 2317 2318 /* 2319 * nfs readdir call 2320 */ 2321 int 2322 nfs_readdir(void *v) 2323 { 2324 struct vop_readdir_args /* { 2325 struct vnode *a_vp; 2326 struct uio *a_uio; 2327 kauth_cred_t a_cred; 2328 int *a_eofflag; 2329 off_t **a_cookies; 2330 int *a_ncookies; 2331 } */ *ap = v; 2332 struct vnode *vp = ap->a_vp; 2333 struct uio *uio = ap->a_uio; 2334 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2335 char *base = uio->uio_iov->iov_base; 2336 int tresid, error; 2337 size_t count, lost; 2338 struct dirent *dp; 2339 off_t *cookies = NULL; 2340 int ncookies = 0, nc; 2341 2342 if (vp->v_type != VDIR) 2343 return (EPERM); 2344 2345 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2346 count = uio->uio_resid - lost; 2347 if (count <= 0) 2348 return (EINVAL); 2349 2350 /* 2351 * Call nfs_bioread() to do the real work. 2352 */ 2353 tresid = uio->uio_resid = count; 2354 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2355 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2356 2357 if (!error && ap->a_cookies) { 2358 ncookies = count / 16; 2359 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2360 *ap->a_cookies = cookies; 2361 } 2362 2363 if (!error && uio->uio_resid == tresid) { 2364 uio->uio_resid += lost; 2365 nfsstats.direofcache_misses++; 2366 if (ap->a_cookies) 2367 *ap->a_ncookies = 0; 2368 *ap->a_eofflag = 1; 2369 return (0); 2370 } 2371 2372 if (!error && ap->a_cookies) { 2373 /* 2374 * Only the NFS server and emulations use cookies, and they 2375 * load the directory block into system space, so we can 2376 * just look at it directly. 2377 */ 2378 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2379 uio->uio_iovcnt != 1) 2380 panic("nfs_readdir: lost in space"); 2381 for (nc = 0; ncookies-- && 2382 base < (char *)uio->uio_iov->iov_base; nc++){ 2383 dp = (struct dirent *) base; 2384 if (dp->d_reclen == 0) 2385 break; 2386 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2387 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2388 else 2389 *(cookies++) = NFS_GETCOOKIE(dp); 2390 base += dp->d_reclen; 2391 } 2392 uio->uio_resid += 2393 ((char *)uio->uio_iov->iov_base - base); 2394 uio->uio_iov->iov_len += 2395 ((char *)uio->uio_iov->iov_base - base); 2396 uio->uio_iov->iov_base = base; 2397 *ap->a_ncookies = nc; 2398 } 2399 2400 uio->uio_resid += lost; 2401 *ap->a_eofflag = 0; 2402 return (error); 2403 } 2404 2405 /* 2406 * Readdir rpc call. 2407 * Called from below the buffer cache by nfs_doio(). 2408 */ 2409 int 2410 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2411 { 2412 int len, left; 2413 struct dirent *dp = NULL; 2414 u_int32_t *tl; 2415 char *cp; 2416 int32_t t1, t2; 2417 char *bpos, *dpos, *cp2; 2418 struct mbuf *mreq, *mrep, *md, *mb; 2419 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2420 struct nfsnode *dnp = VTONFS(vp); 2421 u_quad_t fileno; 2422 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2423 #ifndef NFS_V2_ONLY 2424 int attrflag; 2425 #endif 2426 int nrpcs = 0, reclen; 2427 const int v3 = NFS_ISV3(vp); 2428 2429 #ifdef DIAGNOSTIC 2430 /* 2431 * Should be called from buffer cache, so only amount of 2432 * NFS_DIRBLKSIZ will be requested. 2433 */ 2434 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2435 panic("nfs readdirrpc bad uio"); 2436 #endif 2437 2438 /* 2439 * Loop around doing readdir rpc's of size nm_readdirsize 2440 * truncated to a multiple of NFS_DIRFRAGSIZ. 2441 * The stopping criteria is EOF or buffer full. 2442 */ 2443 while (more_dirs && bigenough) { 2444 /* 2445 * Heuristic: don't bother to do another RPC to further 2446 * fill up this block if there is not much room left. (< 50% 2447 * of the readdir RPC size). This wastes some buffer space 2448 * but can save up to 50% in RPC calls. 2449 */ 2450 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2451 bigenough = 0; 2452 break; 2453 } 2454 nfsstats.rpccnt[NFSPROC_READDIR]++; 2455 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2456 NFSX_READDIR(v3)); 2457 nfsm_fhtom(dnp, v3); 2458 #ifndef NFS_V2_ONLY 2459 if (v3) { 2460 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2461 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2462 txdr_swapcookie3(uiop->uio_offset, tl); 2463 } else { 2464 txdr_cookie3(uiop->uio_offset, tl); 2465 } 2466 tl += 2; 2467 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2468 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2469 } else 2470 #endif 2471 { 2472 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2473 *tl++ = txdr_unsigned(uiop->uio_offset); 2474 } 2475 *tl = txdr_unsigned(nmp->nm_readdirsize); 2476 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2477 nrpcs++; 2478 #ifndef NFS_V2_ONLY 2479 if (v3) { 2480 nfsm_postop_attr(vp, attrflag, 0); 2481 if (!error) { 2482 nfsm_dissect(tl, u_int32_t *, 2483 2 * NFSX_UNSIGNED); 2484 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2485 dnp->n_cookieverf.nfsuquad[1] = *tl; 2486 } else { 2487 m_freem(mrep); 2488 goto nfsmout; 2489 } 2490 } 2491 #endif 2492 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2493 more_dirs = fxdr_unsigned(int, *tl); 2494 2495 /* loop thru the dir entries, doctoring them to 4bsd form */ 2496 while (more_dirs && bigenough) { 2497 #ifndef NFS_V2_ONLY 2498 if (v3) { 2499 nfsm_dissect(tl, u_int32_t *, 2500 3 * NFSX_UNSIGNED); 2501 fileno = fxdr_hyper(tl); 2502 len = fxdr_unsigned(int, *(tl + 2)); 2503 } else 2504 #endif 2505 { 2506 nfsm_dissect(tl, u_int32_t *, 2507 2 * NFSX_UNSIGNED); 2508 fileno = fxdr_unsigned(u_quad_t, *tl++); 2509 len = fxdr_unsigned(int, *tl); 2510 } 2511 if (len <= 0 || len > NFS_MAXNAMLEN) { 2512 error = EBADRPC; 2513 m_freem(mrep); 2514 goto nfsmout; 2515 } 2516 /* for cookie stashing */ 2517 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2518 left = NFS_DIRFRAGSIZ - blksiz; 2519 if (reclen > left) { 2520 memset(uiop->uio_iov->iov_base, 0, left); 2521 dp->d_reclen += left; 2522 UIO_ADVANCE(uiop, left); 2523 blksiz = 0; 2524 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2525 } 2526 if (reclen > uiop->uio_resid) 2527 bigenough = 0; 2528 if (bigenough) { 2529 int tlen; 2530 2531 dp = (struct dirent *)uiop->uio_iov->iov_base; 2532 dp->d_fileno = fileno; 2533 dp->d_namlen = len; 2534 dp->d_reclen = reclen; 2535 dp->d_type = DT_UNKNOWN; 2536 blksiz += reclen; 2537 if (blksiz == NFS_DIRFRAGSIZ) 2538 blksiz = 0; 2539 UIO_ADVANCE(uiop, DIRHDSIZ); 2540 nfsm_mtouio(uiop, len); 2541 tlen = reclen - (DIRHDSIZ + len); 2542 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2543 UIO_ADVANCE(uiop, tlen); 2544 } else 2545 nfsm_adv(nfsm_rndup(len)); 2546 #ifndef NFS_V2_ONLY 2547 if (v3) { 2548 nfsm_dissect(tl, u_int32_t *, 2549 3 * NFSX_UNSIGNED); 2550 } else 2551 #endif 2552 { 2553 nfsm_dissect(tl, u_int32_t *, 2554 2 * NFSX_UNSIGNED); 2555 } 2556 if (bigenough) { 2557 #ifndef NFS_V2_ONLY 2558 if (v3) { 2559 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2560 uiop->uio_offset = 2561 fxdr_swapcookie3(tl); 2562 else 2563 uiop->uio_offset = 2564 fxdr_cookie3(tl); 2565 } 2566 else 2567 #endif 2568 { 2569 uiop->uio_offset = 2570 fxdr_unsigned(off_t, *tl); 2571 } 2572 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2573 } 2574 if (v3) 2575 tl += 2; 2576 else 2577 tl++; 2578 more_dirs = fxdr_unsigned(int, *tl); 2579 } 2580 /* 2581 * If at end of rpc data, get the eof boolean 2582 */ 2583 if (!more_dirs) { 2584 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2585 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2586 2587 /* 2588 * kludge: if we got no entries, treat it as EOF. 2589 * some server sometimes send a reply without any 2590 * entries or EOF. 2591 * although it might mean the server has very long name, 2592 * we can't handle such entries anyway. 2593 */ 2594 2595 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2596 more_dirs = 0; 2597 } 2598 m_freem(mrep); 2599 } 2600 /* 2601 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2602 * by increasing d_reclen for the last record. 2603 */ 2604 if (blksiz > 0) { 2605 left = NFS_DIRFRAGSIZ - blksiz; 2606 memset(uiop->uio_iov->iov_base, 0, left); 2607 dp->d_reclen += left; 2608 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2609 UIO_ADVANCE(uiop, left); 2610 } 2611 2612 /* 2613 * We are now either at the end of the directory or have filled the 2614 * block. 2615 */ 2616 if (bigenough) { 2617 dnp->n_direofoffset = uiop->uio_offset; 2618 dnp->n_flag |= NEOFVALID; 2619 } 2620 nfsmout: 2621 return (error); 2622 } 2623 2624 #ifndef NFS_V2_ONLY 2625 /* 2626 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2627 */ 2628 int 2629 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2630 { 2631 int len, left; 2632 struct dirent *dp = NULL; 2633 u_int32_t *tl; 2634 char *cp; 2635 int32_t t1, t2; 2636 struct vnode *newvp; 2637 char *bpos, *dpos, *cp2; 2638 struct mbuf *mreq, *mrep, *md, *mb; 2639 struct nameidata nami, *ndp = &nami; 2640 struct componentname *cnp = &ndp->ni_cnd; 2641 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2642 struct nfsnode *dnp = VTONFS(vp), *np; 2643 nfsfh_t *fhp; 2644 u_quad_t fileno; 2645 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2646 int attrflag, fhsize, nrpcs = 0, reclen; 2647 struct nfs_fattr fattr, *fp; 2648 2649 #ifdef DIAGNOSTIC 2650 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2651 panic("nfs readdirplusrpc bad uio"); 2652 #endif 2653 ndp->ni_dvp = vp; 2654 newvp = NULLVP; 2655 2656 /* 2657 * Loop around doing readdir rpc's of size nm_readdirsize 2658 * truncated to a multiple of NFS_DIRFRAGSIZ. 2659 * The stopping criteria is EOF or buffer full. 2660 */ 2661 while (more_dirs && bigenough) { 2662 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2663 bigenough = 0; 2664 break; 2665 } 2666 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2667 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2668 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2669 nfsm_fhtom(dnp, 1); 2670 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2671 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2672 txdr_swapcookie3(uiop->uio_offset, tl); 2673 } else { 2674 txdr_cookie3(uiop->uio_offset, tl); 2675 } 2676 tl += 2; 2677 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2678 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2679 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2680 *tl = txdr_unsigned(nmp->nm_rsize); 2681 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2682 nfsm_postop_attr(vp, attrflag, 0); 2683 if (error) { 2684 m_freem(mrep); 2685 goto nfsmout; 2686 } 2687 nrpcs++; 2688 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2689 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2690 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2691 more_dirs = fxdr_unsigned(int, *tl); 2692 2693 /* loop thru the dir entries, doctoring them to 4bsd form */ 2694 while (more_dirs && bigenough) { 2695 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2696 fileno = fxdr_hyper(tl); 2697 len = fxdr_unsigned(int, *(tl + 2)); 2698 if (len <= 0 || len > NFS_MAXNAMLEN) { 2699 error = EBADRPC; 2700 m_freem(mrep); 2701 goto nfsmout; 2702 } 2703 /* for cookie stashing */ 2704 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2705 left = NFS_DIRFRAGSIZ - blksiz; 2706 if (reclen > left) { 2707 /* 2708 * DIRFRAGSIZ is aligned, no need to align 2709 * again here. 2710 */ 2711 memset(uiop->uio_iov->iov_base, 0, left); 2712 dp->d_reclen += left; 2713 UIO_ADVANCE(uiop, left); 2714 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2715 blksiz = 0; 2716 } 2717 if (reclen > uiop->uio_resid) 2718 bigenough = 0; 2719 if (bigenough) { 2720 int tlen; 2721 2722 dp = (struct dirent *)uiop->uio_iov->iov_base; 2723 dp->d_fileno = fileno; 2724 dp->d_namlen = len; 2725 dp->d_reclen = reclen; 2726 dp->d_type = DT_UNKNOWN; 2727 blksiz += reclen; 2728 if (blksiz == NFS_DIRFRAGSIZ) 2729 blksiz = 0; 2730 UIO_ADVANCE(uiop, DIRHDSIZ); 2731 nfsm_mtouio(uiop, len); 2732 tlen = reclen - (DIRHDSIZ + len); 2733 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2734 UIO_ADVANCE(uiop, tlen); 2735 cnp->cn_nameptr = dp->d_name; 2736 cnp->cn_namelen = dp->d_namlen; 2737 } else 2738 nfsm_adv(nfsm_rndup(len)); 2739 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2740 if (bigenough) { 2741 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2742 uiop->uio_offset = 2743 fxdr_swapcookie3(tl); 2744 else 2745 uiop->uio_offset = 2746 fxdr_cookie3(tl); 2747 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2748 } 2749 tl += 2; 2750 2751 /* 2752 * Since the attributes are before the file handle 2753 * (sigh), we must skip over the attributes and then 2754 * come back and get them. 2755 */ 2756 attrflag = fxdr_unsigned(int, *tl); 2757 if (attrflag) { 2758 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2759 memcpy(&fattr, fp, NFSX_V3FATTR); 2760 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2761 doit = fxdr_unsigned(int, *tl); 2762 if (doit) { 2763 nfsm_getfh(fhp, fhsize, 1); 2764 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2765 vref(vp); 2766 newvp = vp; 2767 np = dnp; 2768 } else { 2769 error = nfs_nget1(vp->v_mount, fhp, 2770 fhsize, &np, LK_NOWAIT); 2771 if (!error) 2772 newvp = NFSTOV(np); 2773 } 2774 if (!error) { 2775 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2776 if (bigenough) { 2777 dp->d_type = 2778 IFTODT(VTTOIF(np->n_vattr->va_type)); 2779 if (cnp->cn_namelen <= NCHNAMLEN) { 2780 ndp->ni_vp = newvp; 2781 nfs_cache_enter(ndp->ni_dvp, 2782 ndp->ni_vp, cnp); 2783 } 2784 } 2785 } 2786 error = 0; 2787 } 2788 } else { 2789 /* Just skip over the file handle */ 2790 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2791 i = fxdr_unsigned(int, *tl); 2792 nfsm_adv(nfsm_rndup(i)); 2793 } 2794 if (newvp != NULLVP) { 2795 if (newvp == vp) 2796 vrele(newvp); 2797 else 2798 vput(newvp); 2799 newvp = NULLVP; 2800 } 2801 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2802 more_dirs = fxdr_unsigned(int, *tl); 2803 } 2804 /* 2805 * If at end of rpc data, get the eof boolean 2806 */ 2807 if (!more_dirs) { 2808 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2809 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2810 2811 /* 2812 * kludge: see a comment in nfs_readdirrpc. 2813 */ 2814 2815 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2816 more_dirs = 0; 2817 } 2818 m_freem(mrep); 2819 } 2820 /* 2821 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2822 * by increasing d_reclen for the last record. 2823 */ 2824 if (blksiz > 0) { 2825 left = NFS_DIRFRAGSIZ - blksiz; 2826 memset(uiop->uio_iov->iov_base, 0, left); 2827 dp->d_reclen += left; 2828 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2829 UIO_ADVANCE(uiop, left); 2830 } 2831 2832 /* 2833 * We are now either at the end of the directory or have filled the 2834 * block. 2835 */ 2836 if (bigenough) { 2837 dnp->n_direofoffset = uiop->uio_offset; 2838 dnp->n_flag |= NEOFVALID; 2839 } 2840 nfsmout: 2841 if (newvp != NULLVP) { 2842 if(newvp == vp) 2843 vrele(newvp); 2844 else 2845 vput(newvp); 2846 } 2847 return (error); 2848 } 2849 #endif 2850 2851 /* 2852 * Silly rename. To make the NFS filesystem that is stateless look a little 2853 * more like the "ufs" a remove of an active vnode is translated to a rename 2854 * to a funny looking filename that is removed by nfs_inactive on the 2855 * nfsnode. There is the potential for another process on a different client 2856 * to create the same funny name between the nfs_lookitup() fails and the 2857 * nfs_rename() completes, but... 2858 */ 2859 int 2860 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2861 { 2862 struct sillyrename *sp; 2863 struct nfsnode *np; 2864 int error; 2865 pid_t pid; 2866 2867 cache_purge(dvp); 2868 np = VTONFS(vp); 2869 #ifndef DIAGNOSTIC 2870 if (vp->v_type == VDIR) 2871 panic("nfs: sillyrename dir"); 2872 #endif 2873 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2874 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2875 sp->s_dvp = dvp; 2876 vref(dvp); 2877 2878 /* Fudge together a funny name */ 2879 pid = curlwp->l_proc->p_pid; 2880 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2881 sp->s_namlen = 12; 2882 sp->s_name[8] = hexdigits[pid & 0xf]; 2883 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2884 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2885 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2886 2887 /* Try lookitups until we get one that isn't there */ 2888 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2889 curlwp, (struct nfsnode **)0) == 0) { 2890 sp->s_name[4]++; 2891 if (sp->s_name[4] > 'z') { 2892 error = EINVAL; 2893 goto bad; 2894 } 2895 } 2896 if (dolink) { 2897 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2898 sp->s_cred, curlwp); 2899 /* 2900 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2901 */ 2902 if (error == ENOTSUP) { 2903 error = nfs_renameit(dvp, cnp, sp); 2904 } 2905 } else { 2906 error = nfs_renameit(dvp, cnp, sp); 2907 } 2908 if (error) 2909 goto bad; 2910 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2911 curlwp, &np); 2912 np->n_sillyrename = sp; 2913 return (0); 2914 bad: 2915 vrele(sp->s_dvp); 2916 kauth_cred_free(sp->s_cred); 2917 kmem_free(sp, sizeof(*sp)); 2918 return (error); 2919 } 2920 2921 /* 2922 * Look up a file name and optionally either update the file handle or 2923 * allocate an nfsnode, depending on the value of npp. 2924 * npp == NULL --> just do the lookup 2925 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2926 * handled too 2927 * *npp != NULL --> update the file handle in the vnode 2928 */ 2929 int 2930 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2931 { 2932 u_int32_t *tl; 2933 char *cp; 2934 int32_t t1, t2; 2935 struct vnode *newvp = (struct vnode *)0; 2936 struct nfsnode *np, *dnp = VTONFS(dvp); 2937 char *bpos, *dpos, *cp2; 2938 int error = 0, fhlen; 2939 #ifndef NFS_V2_ONLY 2940 int attrflag; 2941 #endif 2942 struct mbuf *mreq, *mrep, *md, *mb; 2943 nfsfh_t *nfhp; 2944 const int v3 = NFS_ISV3(dvp); 2945 2946 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2947 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2948 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2949 nfsm_fhtom(dnp, v3); 2950 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2951 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2952 if (npp && !error) { 2953 nfsm_getfh(nfhp, fhlen, v3); 2954 if (*npp) { 2955 np = *npp; 2956 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2957 kmem_free(np->n_fhp, np->n_fhsize); 2958 np->n_fhp = &np->n_fh; 2959 } 2960 #if NFS_SMALLFH < NFSX_V3FHMAX 2961 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2962 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2963 #endif 2964 memcpy(np->n_fhp, nfhp, fhlen); 2965 np->n_fhsize = fhlen; 2966 newvp = NFSTOV(np); 2967 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2968 vref(dvp); 2969 newvp = dvp; 2970 np = dnp; 2971 } else { 2972 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2973 if (error) { 2974 m_freem(mrep); 2975 return (error); 2976 } 2977 newvp = NFSTOV(np); 2978 } 2979 #ifndef NFS_V2_ONLY 2980 if (v3) { 2981 nfsm_postop_attr(newvp, attrflag, 0); 2982 if (!attrflag && *npp == NULL) { 2983 m_freem(mrep); 2984 vput(newvp); 2985 return (ENOENT); 2986 } 2987 } else 2988 #endif 2989 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2990 } 2991 nfsm_reqdone; 2992 if (npp && *npp == NULL) { 2993 if (error) { 2994 if (newvp) 2995 vput(newvp); 2996 } else 2997 *npp = np; 2998 } 2999 return (error); 3000 } 3001 3002 #ifndef NFS_V2_ONLY 3003 /* 3004 * Nfs Version 3 commit rpc 3005 */ 3006 int 3007 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3008 { 3009 char *cp; 3010 u_int32_t *tl; 3011 int32_t t1, t2; 3012 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3013 char *bpos, *dpos, *cp2; 3014 int error = 0, wccflag = NFSV3_WCCRATTR; 3015 struct mbuf *mreq, *mrep, *md, *mb; 3016 struct nfsnode *np; 3017 3018 KASSERT(NFS_ISV3(vp)); 3019 3020 #ifdef NFS_DEBUG_COMMIT 3021 printf("commit %lu - %lu\n", (unsigned long)offset, 3022 (unsigned long)(offset + cnt)); 3023 #endif 3024 3025 mutex_enter(&nmp->nm_lock); 3026 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3027 mutex_exit(&nmp->nm_lock); 3028 return (0); 3029 } 3030 mutex_exit(&nmp->nm_lock); 3031 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3032 np = VTONFS(vp); 3033 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3034 nfsm_fhtom(np, 1); 3035 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3036 txdr_hyper(offset, tl); 3037 tl += 2; 3038 *tl = txdr_unsigned(cnt); 3039 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3040 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3041 if (!error) { 3042 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3043 mutex_enter(&nmp->nm_lock); 3044 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3045 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3046 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3047 error = NFSERR_STALEWRITEVERF; 3048 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3049 } 3050 mutex_exit(&nmp->nm_lock); 3051 } 3052 nfsm_reqdone; 3053 return (error); 3054 } 3055 #endif 3056 3057 /* 3058 * Kludge City.. 3059 * - make nfs_bmap() essentially a no-op that does no translation 3060 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3061 * (Maybe I could use the process's page mapping, but I was concerned that 3062 * Kernel Write might not be enabled and also figured copyout() would do 3063 * a lot more work than memcpy() and also it currently happens in the 3064 * context of the swapper process (2). 3065 */ 3066 int 3067 nfs_bmap(void *v) 3068 { 3069 struct vop_bmap_args /* { 3070 struct vnode *a_vp; 3071 daddr_t a_bn; 3072 struct vnode **a_vpp; 3073 daddr_t *a_bnp; 3074 int *a_runp; 3075 } */ *ap = v; 3076 struct vnode *vp = ap->a_vp; 3077 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3078 3079 if (ap->a_vpp != NULL) 3080 *ap->a_vpp = vp; 3081 if (ap->a_bnp != NULL) 3082 *ap->a_bnp = ap->a_bn << bshift; 3083 if (ap->a_runp != NULL) 3084 *ap->a_runp = 1024 * 1024; /* XXX */ 3085 return (0); 3086 } 3087 3088 /* 3089 * Strategy routine. 3090 * For async requests when nfsiod(s) are running, queue the request by 3091 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3092 * request. 3093 */ 3094 int 3095 nfs_strategy(void *v) 3096 { 3097 struct vop_strategy_args *ap = v; 3098 struct buf *bp = ap->a_bp; 3099 int error = 0; 3100 3101 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3102 panic("nfs physio/async"); 3103 3104 /* 3105 * If the op is asynchronous and an i/o daemon is waiting 3106 * queue the request, wake it up and wait for completion 3107 * otherwise just do it ourselves. 3108 */ 3109 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3110 error = nfs_doio(bp); 3111 return (error); 3112 } 3113 3114 /* 3115 * fsync vnode op. Just call nfs_flush() with commit == 1. 3116 */ 3117 /* ARGSUSED */ 3118 int 3119 nfs_fsync(void *v) 3120 { 3121 struct vop_fsync_args /* { 3122 struct vnodeop_desc *a_desc; 3123 struct vnode * a_vp; 3124 kauth_cred_t a_cred; 3125 int a_flags; 3126 off_t offlo; 3127 off_t offhi; 3128 struct lwp * a_l; 3129 } */ *ap = v; 3130 3131 struct vnode *vp = ap->a_vp; 3132 3133 if (vp->v_type != VREG) 3134 return 0; 3135 3136 return (nfs_flush(vp, ap->a_cred, 3137 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3138 } 3139 3140 /* 3141 * Flush all the data associated with a vnode. 3142 */ 3143 int 3144 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3145 int commit) 3146 { 3147 struct nfsnode *np = VTONFS(vp); 3148 int error; 3149 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3150 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3151 3152 mutex_enter(vp->v_interlock); 3153 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3154 if (np->n_flag & NWRITEERR) { 3155 error = np->n_error; 3156 np->n_flag &= ~NWRITEERR; 3157 } 3158 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3159 return (error); 3160 } 3161 3162 /* 3163 * Return POSIX pathconf information applicable to nfs. 3164 * 3165 * N.B. The NFS V2 protocol doesn't support this RPC. 3166 */ 3167 /* ARGSUSED */ 3168 int 3169 nfs_pathconf(void *v) 3170 { 3171 struct vop_pathconf_args /* { 3172 struct vnode *a_vp; 3173 int a_name; 3174 register_t *a_retval; 3175 } */ *ap = v; 3176 struct nfsv3_pathconf *pcp; 3177 struct vnode *vp = ap->a_vp; 3178 struct mbuf *mreq, *mrep, *md, *mb; 3179 int32_t t1, t2; 3180 u_int32_t *tl; 3181 char *bpos, *dpos, *cp, *cp2; 3182 int error = 0, attrflag; 3183 #ifndef NFS_V2_ONLY 3184 struct nfsmount *nmp; 3185 unsigned int l; 3186 u_int64_t maxsize; 3187 #endif 3188 const int v3 = NFS_ISV3(vp); 3189 struct nfsnode *np = VTONFS(vp); 3190 3191 switch (ap->a_name) { 3192 /* Names that can be resolved locally. */ 3193 case _PC_PIPE_BUF: 3194 *ap->a_retval = PIPE_BUF; 3195 break; 3196 case _PC_SYNC_IO: 3197 *ap->a_retval = 1; 3198 break; 3199 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3200 case _PC_LINK_MAX: 3201 case _PC_NAME_MAX: 3202 case _PC_CHOWN_RESTRICTED: 3203 case _PC_NO_TRUNC: 3204 if (!v3) { 3205 error = EINVAL; 3206 break; 3207 } 3208 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3209 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3210 nfsm_fhtom(np, 1); 3211 nfsm_request(np, NFSPROC_PATHCONF, 3212 curlwp, curlwp->l_cred); /* XXX */ 3213 nfsm_postop_attr(vp, attrflag, 0); 3214 if (!error) { 3215 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3216 NFSX_V3PATHCONF); 3217 switch (ap->a_name) { 3218 case _PC_LINK_MAX: 3219 *ap->a_retval = 3220 fxdr_unsigned(register_t, pcp->pc_linkmax); 3221 break; 3222 case _PC_NAME_MAX: 3223 *ap->a_retval = 3224 fxdr_unsigned(register_t, pcp->pc_namemax); 3225 break; 3226 case _PC_CHOWN_RESTRICTED: 3227 *ap->a_retval = 3228 (pcp->pc_chownrestricted == nfs_true); 3229 break; 3230 case _PC_NO_TRUNC: 3231 *ap->a_retval = 3232 (pcp->pc_notrunc == nfs_true); 3233 break; 3234 } 3235 } 3236 nfsm_reqdone; 3237 break; 3238 case _PC_FILESIZEBITS: 3239 #ifndef NFS_V2_ONLY 3240 if (v3) { 3241 nmp = VFSTONFS(vp->v_mount); 3242 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3243 if ((error = nfs_fsinfo(nmp, vp, 3244 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3245 break; 3246 for (l = 0, maxsize = nmp->nm_maxfilesize; 3247 (maxsize >> l) > 0; l++) 3248 ; 3249 *ap->a_retval = l + 1; 3250 } else 3251 #endif 3252 { 3253 *ap->a_retval = 32; /* NFS V2 limitation */ 3254 } 3255 break; 3256 default: 3257 error = EINVAL; 3258 break; 3259 } 3260 3261 return (error); 3262 } 3263 3264 /* 3265 * NFS advisory byte-level locks. 3266 */ 3267 int 3268 nfs_advlock(void *v) 3269 { 3270 struct vop_advlock_args /* { 3271 struct vnode *a_vp; 3272 void *a_id; 3273 int a_op; 3274 struct flock *a_fl; 3275 int a_flags; 3276 } */ *ap = v; 3277 struct nfsnode *np = VTONFS(ap->a_vp); 3278 3279 return lf_advlock(ap, &np->n_lockf, np->n_size); 3280 } 3281 3282 /* 3283 * Print out the contents of an nfsnode. 3284 */ 3285 int 3286 nfs_print(void *v) 3287 { 3288 struct vop_print_args /* { 3289 struct vnode *a_vp; 3290 } */ *ap = v; 3291 struct vnode *vp = ap->a_vp; 3292 struct nfsnode *np = VTONFS(vp); 3293 3294 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3295 (unsigned long long)np->n_vattr->va_fileid, 3296 (unsigned long long)np->n_vattr->va_fsid); 3297 if (vp->v_type == VFIFO) 3298 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3299 printf("\n"); 3300 return (0); 3301 } 3302 3303 /* 3304 * nfs unlock wrapper. 3305 */ 3306 int 3307 nfs_unlock(void *v) 3308 { 3309 struct vop_unlock_args /* { 3310 struct vnode *a_vp; 3311 int a_flags; 3312 } */ *ap = v; 3313 struct vnode *vp = ap->a_vp; 3314 3315 /* 3316 * VOP_UNLOCK can be called by nfs_loadattrcache 3317 * with v_data == 0. 3318 */ 3319 if (VTONFS(vp)) { 3320 nfs_delayedtruncate(vp); 3321 } 3322 3323 return genfs_unlock(v); 3324 } 3325 3326 /* 3327 * nfs special file access vnode op. 3328 * Essentially just get vattr and then imitate iaccess() since the device is 3329 * local to the client. 3330 */ 3331 int 3332 nfsspec_access(void *v) 3333 { 3334 struct vop_access_args /* { 3335 struct vnode *a_vp; 3336 int a_mode; 3337 kauth_cred_t a_cred; 3338 struct lwp *a_l; 3339 } */ *ap = v; 3340 struct vattr va; 3341 struct vnode *vp = ap->a_vp; 3342 int error; 3343 3344 error = VOP_GETATTR(vp, &va, ap->a_cred); 3345 if (error) 3346 return (error); 3347 3348 /* 3349 * Disallow write attempts on filesystems mounted read-only; 3350 * unless the file is a socket, fifo, or a block or character 3351 * device resident on the filesystem. 3352 */ 3353 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3354 switch (vp->v_type) { 3355 case VREG: 3356 case VDIR: 3357 case VLNK: 3358 return (EROFS); 3359 default: 3360 break; 3361 } 3362 } 3363 3364 return kauth_authorize_vnode(ap->a_cred, KAUTH_ACCESS_ACTION(ap->a_mode, 3365 va.va_type, va.va_mode), vp, NULL, genfs_can_access(va.va_type, 3366 va.va_mode, va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3367 } 3368 3369 /* 3370 * Read wrapper for special devices. 3371 */ 3372 int 3373 nfsspec_read(void *v) 3374 { 3375 struct vop_read_args /* { 3376 struct vnode *a_vp; 3377 struct uio *a_uio; 3378 int a_ioflag; 3379 kauth_cred_t a_cred; 3380 } */ *ap = v; 3381 struct nfsnode *np = VTONFS(ap->a_vp); 3382 3383 /* 3384 * Set access flag. 3385 */ 3386 np->n_flag |= NACC; 3387 getnanotime(&np->n_atim); 3388 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3389 } 3390 3391 /* 3392 * Write wrapper for special devices. 3393 */ 3394 int 3395 nfsspec_write(void *v) 3396 { 3397 struct vop_write_args /* { 3398 struct vnode *a_vp; 3399 struct uio *a_uio; 3400 int a_ioflag; 3401 kauth_cred_t a_cred; 3402 } */ *ap = v; 3403 struct nfsnode *np = VTONFS(ap->a_vp); 3404 3405 /* 3406 * Set update flag. 3407 */ 3408 np->n_flag |= NUPD; 3409 getnanotime(&np->n_mtim); 3410 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3411 } 3412 3413 /* 3414 * Close wrapper for special devices. 3415 * 3416 * Update the times on the nfsnode then do device close. 3417 */ 3418 int 3419 nfsspec_close(void *v) 3420 { 3421 struct vop_close_args /* { 3422 struct vnode *a_vp; 3423 int a_fflag; 3424 kauth_cred_t a_cred; 3425 struct lwp *a_l; 3426 } */ *ap = v; 3427 struct vnode *vp = ap->a_vp; 3428 struct nfsnode *np = VTONFS(vp); 3429 struct vattr vattr; 3430 3431 if (np->n_flag & (NACC | NUPD)) { 3432 np->n_flag |= NCHG; 3433 if (vp->v_usecount == 1 && 3434 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3435 vattr_null(&vattr); 3436 if (np->n_flag & NACC) 3437 vattr.va_atime = np->n_atim; 3438 if (np->n_flag & NUPD) 3439 vattr.va_mtime = np->n_mtim; 3440 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3441 } 3442 } 3443 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3444 } 3445 3446 /* 3447 * Read wrapper for fifos. 3448 */ 3449 int 3450 nfsfifo_read(void *v) 3451 { 3452 struct vop_read_args /* { 3453 struct vnode *a_vp; 3454 struct uio *a_uio; 3455 int a_ioflag; 3456 kauth_cred_t a_cred; 3457 } */ *ap = v; 3458 struct nfsnode *np = VTONFS(ap->a_vp); 3459 3460 /* 3461 * Set access flag. 3462 */ 3463 np->n_flag |= NACC; 3464 getnanotime(&np->n_atim); 3465 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3466 } 3467 3468 /* 3469 * Write wrapper for fifos. 3470 */ 3471 int 3472 nfsfifo_write(void *v) 3473 { 3474 struct vop_write_args /* { 3475 struct vnode *a_vp; 3476 struct uio *a_uio; 3477 int a_ioflag; 3478 kauth_cred_t a_cred; 3479 } */ *ap = v; 3480 struct nfsnode *np = VTONFS(ap->a_vp); 3481 3482 /* 3483 * Set update flag. 3484 */ 3485 np->n_flag |= NUPD; 3486 getnanotime(&np->n_mtim); 3487 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3488 } 3489 3490 /* 3491 * Close wrapper for fifos. 3492 * 3493 * Update the times on the nfsnode then do fifo close. 3494 */ 3495 int 3496 nfsfifo_close(void *v) 3497 { 3498 struct vop_close_args /* { 3499 struct vnode *a_vp; 3500 int a_fflag; 3501 kauth_cred_t a_cred; 3502 struct lwp *a_l; 3503 } */ *ap = v; 3504 struct vnode *vp = ap->a_vp; 3505 struct nfsnode *np = VTONFS(vp); 3506 struct vattr vattr; 3507 3508 if (np->n_flag & (NACC | NUPD)) { 3509 struct timespec ts; 3510 3511 getnanotime(&ts); 3512 if (np->n_flag & NACC) 3513 np->n_atim = ts; 3514 if (np->n_flag & NUPD) 3515 np->n_mtim = ts; 3516 np->n_flag |= NCHG; 3517 if (vp->v_usecount == 1 && 3518 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3519 vattr_null(&vattr); 3520 if (np->n_flag & NACC) 3521 vattr.va_atime = np->n_atim; 3522 if (np->n_flag & NUPD) 3523 vattr.va_mtime = np->n_mtim; 3524 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3525 } 3526 } 3527 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3528 } 3529
Indexes created Fri Oct 17 00:09:41 GMT 2025