nfs_vnops.c revision 1.266.10.5
1 /* $NetBSD: nfs_vnops.c,v 1.266.10.5 2009/07/18 14:53:26 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_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 1.266.10.5 2009/07/18 14:53:26 yamt 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 72 #include <uvm/uvm_extern.h> 73 #include <uvm/uvm.h> 74 75 #include <miscfs/fifofs/fifo.h> 76 #include <miscfs/genfs/genfs.h> 77 #include <miscfs/genfs/genfs_node.h> 78 #include <miscfs/specfs/specdev.h> 79 80 #include <nfs/rpcv2.h> 81 #include <nfs/nfsproto.h> 82 #include <nfs/nfs.h> 83 #include <nfs/nfsnode.h> 84 #include <nfs/nfsmount.h> 85 #include <nfs/xdr_subs.h> 86 #include <nfs/nfsm_subs.h> 87 #include <nfs/nfs_var.h> 88 89 #include <net/if.h> 90 #include <netinet/in.h> 91 #include <netinet/in_var.h> 92 93 /* 94 * Global vfs data structures for nfs 95 */ 96 int (**nfsv2_vnodeop_p)(void *); 97 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 98 { &vop_default_desc, vn_default_error }, 99 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 100 { &vop_create_desc, nfs_create }, /* create */ 101 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 102 { &vop_open_desc, nfs_open }, /* open */ 103 { &vop_close_desc, nfs_close }, /* close */ 104 { &vop_access_desc, nfs_access }, /* access */ 105 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 106 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 107 { &vop_read_desc, nfs_read }, /* read */ 108 { &vop_write_desc, nfs_write }, /* write */ 109 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 110 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 111 { &vop_poll_desc, nfs_poll }, /* poll */ 112 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 113 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 114 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 115 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 116 { &vop_seek_desc, nfs_seek }, /* seek */ 117 { &vop_remove_desc, nfs_remove }, /* remove */ 118 { &vop_link_desc, nfs_link }, /* link */ 119 { &vop_rename_desc, nfs_rename }, /* rename */ 120 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 121 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 122 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 123 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 124 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 125 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 126 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 127 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 128 { &vop_lock_desc, nfs_lock }, /* lock */ 129 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 130 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 131 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 132 { &vop_print_desc, nfs_print }, /* print */ 133 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 134 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 135 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 136 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 137 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 138 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 139 { NULL, NULL } 140 }; 141 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 142 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 143 144 /* 145 * Special device vnode ops 146 */ 147 int (**spec_nfsv2nodeop_p)(void *); 148 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 149 { &vop_default_desc, vn_default_error }, 150 { &vop_lookup_desc, spec_lookup }, /* lookup */ 151 { &vop_create_desc, spec_create }, /* create */ 152 { &vop_mknod_desc, spec_mknod }, /* mknod */ 153 { &vop_open_desc, spec_open }, /* open */ 154 { &vop_close_desc, nfsspec_close }, /* close */ 155 { &vop_access_desc, nfsspec_access }, /* access */ 156 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 157 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 158 { &vop_read_desc, nfsspec_read }, /* read */ 159 { &vop_write_desc, nfsspec_write }, /* write */ 160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 162 { &vop_poll_desc, spec_poll }, /* poll */ 163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 164 { &vop_revoke_desc, spec_revoke }, /* revoke */ 165 { &vop_mmap_desc, spec_mmap }, /* mmap */ 166 { &vop_fsync_desc, spec_fsync }, /* fsync */ 167 { &vop_seek_desc, spec_seek }, /* seek */ 168 { &vop_remove_desc, spec_remove }, /* remove */ 169 { &vop_link_desc, spec_link }, /* link */ 170 { &vop_rename_desc, spec_rename }, /* rename */ 171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 173 { &vop_symlink_desc, spec_symlink }, /* symlink */ 174 { &vop_readdir_desc, spec_readdir }, /* readdir */ 175 { &vop_readlink_desc, spec_readlink }, /* readlink */ 176 { &vop_abortop_desc, spec_abortop }, /* abortop */ 177 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 179 { &vop_lock_desc, nfs_lock }, /* lock */ 180 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 181 { &vop_bmap_desc, spec_bmap }, /* bmap */ 182 { &vop_strategy_desc, spec_strategy }, /* strategy */ 183 { &vop_print_desc, nfs_print }, /* print */ 184 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, spec_advlock }, /* advlock */ 187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 188 { &vop_getpages_desc, spec_getpages }, /* getpages */ 189 { &vop_putpages_desc, spec_putpages }, /* putpages */ 190 { NULL, NULL } 191 }; 192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 194 195 int (**fifo_nfsv2nodeop_p)(void *); 196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 197 { &vop_default_desc, vn_default_error }, 198 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 199 { &vop_create_desc, fifo_create }, /* create */ 200 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 201 { &vop_open_desc, fifo_open }, /* open */ 202 { &vop_close_desc, nfsfifo_close }, /* close */ 203 { &vop_access_desc, nfsspec_access }, /* access */ 204 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 205 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 206 { &vop_read_desc, nfsfifo_read }, /* read */ 207 { &vop_write_desc, nfsfifo_write }, /* write */ 208 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 209 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 210 { &vop_poll_desc, fifo_poll }, /* poll */ 211 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 212 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 213 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 214 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 215 { &vop_seek_desc, fifo_seek }, /* seek */ 216 { &vop_remove_desc, fifo_remove }, /* remove */ 217 { &vop_link_desc, fifo_link }, /* link */ 218 { &vop_rename_desc, fifo_rename }, /* rename */ 219 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 220 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 221 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 222 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 223 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 224 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 225 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 226 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 227 { &vop_lock_desc, nfs_lock }, /* lock */ 228 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 229 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 230 { &vop_strategy_desc, genfs_badop }, /* strategy */ 231 { &vop_print_desc, nfs_print }, /* print */ 232 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 233 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 234 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 235 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 236 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 237 { NULL, NULL } 238 }; 239 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 240 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 241 242 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 243 size_t, kauth_cred_t, struct lwp *); 244 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 245 246 /* 247 * Global variables 248 */ 249 extern u_int32_t nfs_true, nfs_false; 250 extern u_int32_t nfs_xdrneg1; 251 extern const nfstype nfsv3_type[9]; 252 253 int nfs_numasync = 0; 254 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 255 #define UIO_ADVANCE(uio, siz) \ 256 (void)((uio)->uio_resid -= (siz), \ 257 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 258 (uio)->uio_iov->iov_len -= (siz)) 259 260 static void nfs_cache_enter(struct vnode *, struct vnode *, 261 struct componentname *); 262 263 static void 264 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 265 struct componentname *cnp) 266 { 267 struct nfsnode *dnp = VTONFS(dvp); 268 269 if (vp != NULL) { 270 struct nfsnode *np = VTONFS(vp); 271 272 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 273 } 274 275 if (!timespecisset(&dnp->n_nctime)) 276 dnp->n_nctime = dnp->n_vattr->va_mtime; 277 278 cache_enter(dvp, vp, cnp); 279 } 280 281 /* 282 * nfs null call from vfs. 283 */ 284 int 285 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 286 { 287 char *bpos, *dpos; 288 int error = 0; 289 struct mbuf *mreq, *mrep, *md, *mb; 290 struct nfsnode *np = VTONFS(vp); 291 292 nfsm_reqhead(np, NFSPROC_NULL, 0); 293 nfsm_request(np, NFSPROC_NULL, l, cred); 294 nfsm_reqdone; 295 return (error); 296 } 297 298 /* 299 * nfs access vnode op. 300 * For nfs version 2, just return ok. File accesses may fail later. 301 * For nfs version 3, use the access rpc to check accessibility. If file modes 302 * are changed on the server, accesses might still fail later. 303 */ 304 int 305 nfs_access(void *v) 306 { 307 struct vop_access_args /* { 308 struct vnode *a_vp; 309 int a_mode; 310 kauth_cred_t a_cred; 311 } */ *ap = v; 312 struct vnode *vp = ap->a_vp; 313 #ifndef NFS_V2_ONLY 314 u_int32_t *tl; 315 char *cp; 316 int32_t t1, t2; 317 char *bpos, *dpos, *cp2; 318 int error = 0, attrflag; 319 struct mbuf *mreq, *mrep, *md, *mb; 320 u_int32_t mode, rmode; 321 const int v3 = NFS_ISV3(vp); 322 #endif 323 int cachevalid; 324 struct nfsnode *np = VTONFS(vp); 325 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 326 uid_t uid = kauth_cred_geteuid(ap->a_cred); 327 int result; 328 329 mutex_enter(&np->n_attrlock); 330 cachevalid = np->n_accstamp != -1 && 331 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 332 np->n_accuid == uid; 333 334 /* 335 * Check access cache first. If this request has been made for this 336 * uid shortly before, use the cached result. 337 */ 338 if (cachevalid) { 339 if (!np->n_accerror) { 340 if ((np->n_accmode & ap->a_mode) == ap->a_mode) { 341 result = np->n_accerror; 342 mutex_exit(&np->n_attrlock); 343 return result; 344 } 345 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) { 346 result = np->n_accerror; 347 mutex_exit(&np->n_attrlock); 348 return result; 349 } 350 } 351 mutex_exit(&np->n_attrlock); 352 353 #ifndef NFS_V2_ONLY 354 /* 355 * For nfs v3, do an access rpc, otherwise you are stuck emulating 356 * ufs_access() locally using the vattr. This may not be correct, 357 * since the server may apply other access criteria such as 358 * client uid-->server uid mapping that we do not know about, but 359 * this is better than just returning anything that is lying about 360 * in the cache. 361 */ 362 if (v3) { 363 nfsstats.rpccnt[NFSPROC_ACCESS]++; 364 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 365 nfsm_fhtom(np, v3); 366 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 367 if (ap->a_mode & VREAD) 368 mode = NFSV3ACCESS_READ; 369 else 370 mode = 0; 371 if (vp->v_type != VDIR) { 372 if (ap->a_mode & VWRITE) 373 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 374 if (ap->a_mode & VEXEC) 375 mode |= NFSV3ACCESS_EXECUTE; 376 } else { 377 if (ap->a_mode & VWRITE) 378 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 379 NFSV3ACCESS_DELETE); 380 if (ap->a_mode & VEXEC) 381 mode |= NFSV3ACCESS_LOOKUP; 382 } 383 *tl = txdr_unsigned(mode); 384 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 385 nfsm_postop_attr(vp, attrflag, 0); 386 if (!error) { 387 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 388 rmode = fxdr_unsigned(u_int32_t, *tl); 389 /* 390 * The NFS V3 spec does not clarify whether or not 391 * the returned access bits can be a superset of 392 * the ones requested, so... 393 */ 394 if ((rmode & mode) != mode) 395 error = EACCES; 396 } 397 nfsm_reqdone; 398 } else 399 #endif 400 return (nfsspec_access(ap)); 401 #ifndef NFS_V2_ONLY 402 /* 403 * Disallow write attempts on filesystems mounted read-only; 404 * unless the file is a socket, fifo, or a block or character 405 * device resident on the filesystem. 406 */ 407 if (!error && (ap->a_mode & VWRITE) && 408 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 409 switch (vp->v_type) { 410 case VREG: 411 case VDIR: 412 case VLNK: 413 error = EROFS; 414 default: 415 break; 416 } 417 } 418 419 if (!error || error == EACCES) { 420 mutex_enter(&np->n_attrlock); 421 /* 422 * If we got the same result as for a previous, 423 * different request, OR it in. Don't update 424 * the timestamp in that case. 425 */ 426 if (cachevalid && np->n_accstamp != -1 && 427 error == np->n_accerror) { 428 if (!error) 429 np->n_accmode |= ap->a_mode; 430 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 431 np->n_accmode = ap->a_mode; 432 } else { 433 np->n_accstamp = time_uptime; 434 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 435 np->n_accmode = ap->a_mode; 436 np->n_accerror = error; 437 } 438 mutex_exit(&np->n_attrlock); 439 } 440 441 return (error); 442 #endif 443 } 444 445 /* 446 * nfs open vnode op 447 * Check to see if the type is ok 448 * and that deletion is not in progress. 449 * For paged in text files, you will need to flush the page cache 450 * if consistency is lost. 451 */ 452 /* ARGSUSED */ 453 int 454 nfs_open(void *v) 455 { 456 struct vop_open_args /* { 457 struct vnode *a_vp; 458 int a_mode; 459 kauth_cred_t a_cred; 460 } */ *ap = v; 461 struct vnode *vp = ap->a_vp; 462 struct nfsnode *np = VTONFS(vp); 463 int error; 464 465 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 466 return (EACCES); 467 } 468 469 if (ap->a_mode & FREAD) { 470 if (np->n_rcred != NULL) 471 kauth_cred_free(np->n_rcred); 472 np->n_rcred = ap->a_cred; 473 kauth_cred_hold(np->n_rcred); 474 } 475 if (ap->a_mode & FWRITE) { 476 if (np->n_wcred != NULL) 477 kauth_cred_free(np->n_wcred); 478 np->n_wcred = ap->a_cred; 479 kauth_cred_hold(np->n_wcred); 480 } 481 482 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 483 if (error) 484 return error; 485 486 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 487 488 return (0); 489 } 490 491 /* 492 * nfs close vnode op 493 * What an NFS client should do upon close after writing is a debatable issue. 494 * Most NFS clients push delayed writes to the server upon close, basically for 495 * two reasons: 496 * 1 - So that any write errors may be reported back to the client process 497 * doing the close system call. By far the two most likely errors are 498 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 499 * 2 - To put a worst case upper bound on cache inconsistency between 500 * multiple clients for the file. 501 * There is also a consistency problem for Version 2 of the protocol w.r.t. 502 * not being able to tell if other clients are writing a file concurrently, 503 * since there is no way of knowing if the changed modify time in the reply 504 * is only due to the write for this client. 505 * (NFS Version 3 provides weak cache consistency data in the reply that 506 * should be sufficient to detect and handle this case.) 507 * 508 * The current code does the following: 509 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 510 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 511 * or commit them (this satisfies 1 and 2 except for the 512 * case where the server crashes after this close but 513 * before the commit RPC, which is felt to be "good 514 * enough". Changing the last argument to nfs_flush() to 515 * a 1 would force a commit operation, if it is felt a 516 * commit is necessary now. 517 */ 518 /* ARGSUSED */ 519 int 520 nfs_close(void *v) 521 { 522 struct vop_close_args /* { 523 struct vnodeop_desc *a_desc; 524 struct vnode *a_vp; 525 int a_fflag; 526 kauth_cred_t a_cred; 527 } */ *ap = v; 528 struct vnode *vp = ap->a_vp; 529 struct nfsnode *np = VTONFS(vp); 530 int error = 0; 531 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 532 533 if (vp->v_type == VREG) { 534 if (np->n_flag & NMODIFIED) { 535 #ifndef NFS_V2_ONLY 536 if (NFS_ISV3(vp)) { 537 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 538 np->n_flag &= ~NMODIFIED; 539 } else 540 #endif 541 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 542 NFS_INVALIDATE_ATTRCACHE(np); 543 } 544 if (np->n_flag & NWRITEERR) { 545 np->n_flag &= ~NWRITEERR; 546 error = np->n_error; 547 } 548 } 549 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 550 return (error); 551 } 552 553 /* 554 * nfs getattr call from vfs. 555 */ 556 int 557 nfs_getattr(void *v) 558 { 559 struct vop_getattr_args /* { 560 struct vnode *a_vp; 561 struct vattr *a_vap; 562 kauth_cred_t a_cred; 563 } */ *ap = v; 564 struct vnode *vp = ap->a_vp; 565 struct nfsnode *np = VTONFS(vp); 566 char *cp; 567 u_int32_t *tl; 568 int32_t t1, t2; 569 char *bpos, *dpos; 570 int error = 0; 571 struct mbuf *mreq, *mrep, *md, *mb; 572 const int v3 = NFS_ISV3(vp); 573 574 KASSERT(VOP_ISLOCKED(vp)); 575 576 /* 577 * Update local times for special files. 578 */ 579 if (np->n_flag & (NACC | NUPD)) 580 np->n_flag |= NCHG; 581 582 /* 583 * if we have delayed truncation, do it now. 584 */ 585 nfs_delayedtruncate(vp); 586 587 /* 588 * First look in the cache. 589 */ 590 if (nfs_getattrcache(vp, ap->a_vap) == 0) 591 return (0); 592 nfsstats.rpccnt[NFSPROC_GETATTR]++; 593 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 594 nfsm_fhtom(np, v3); 595 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 596 if (!error) { 597 nfsm_loadattr(vp, ap->a_vap, 0); 598 if (vp->v_type == VDIR && 599 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 600 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 601 } 602 nfsm_reqdone; 603 return (error); 604 } 605 606 /* 607 * nfs setattr call. 608 */ 609 int 610 nfs_setattr(void *v) 611 { 612 struct vop_setattr_args /* { 613 struct vnodeop_desc *a_desc; 614 struct vnode *a_vp; 615 struct vattr *a_vap; 616 kauth_cred_t a_cred; 617 } */ *ap = v; 618 struct vnode *vp = ap->a_vp; 619 struct nfsnode *np = VTONFS(vp); 620 struct vattr *vap = ap->a_vap; 621 int error = 0; 622 u_quad_t tsize = 0; 623 624 /* 625 * Setting of flags is not supported. 626 */ 627 if (vap->va_flags != VNOVAL) 628 return (EOPNOTSUPP); 629 630 /* 631 * Disallow write attempts if the filesystem is mounted read-only. 632 */ 633 if ((vap->va_uid != (uid_t)VNOVAL || 634 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 635 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 636 (vp->v_mount->mnt_flag & MNT_RDONLY)) 637 return (EROFS); 638 if (vap->va_size != VNOVAL) { 639 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 640 return EFBIG; 641 } 642 switch (vp->v_type) { 643 case VDIR: 644 return (EISDIR); 645 case VCHR: 646 case VBLK: 647 case VSOCK: 648 case VFIFO: 649 if (vap->va_mtime.tv_sec == VNOVAL && 650 vap->va_atime.tv_sec == VNOVAL && 651 vap->va_mode == (mode_t)VNOVAL && 652 vap->va_uid == (uid_t)VNOVAL && 653 vap->va_gid == (gid_t)VNOVAL) 654 return (0); 655 vap->va_size = VNOVAL; 656 break; 657 default: 658 /* 659 * Disallow write attempts if the filesystem is 660 * mounted read-only. 661 */ 662 if (vp->v_mount->mnt_flag & MNT_RDONLY) 663 return (EROFS); 664 genfs_node_wrlock(vp); 665 uvm_vnp_setsize(vp, vap->va_size); 666 tsize = np->n_size; 667 np->n_size = vap->va_size; 668 if (vap->va_size == 0) 669 error = nfs_vinvalbuf(vp, 0, 670 ap->a_cred, curlwp, 1); 671 else 672 error = nfs_vinvalbuf(vp, V_SAVE, 673 ap->a_cred, curlwp, 1); 674 if (error) { 675 uvm_vnp_setsize(vp, tsize); 676 genfs_node_unlock(vp); 677 return (error); 678 } 679 np->n_vattr->va_size = vap->va_size; 680 } 681 } else { 682 /* 683 * flush files before setattr because a later write of 684 * cached data might change timestamps or reset sugid bits 685 */ 686 if ((vap->va_mtime.tv_sec != VNOVAL || 687 vap->va_atime.tv_sec != VNOVAL || 688 vap->va_mode != VNOVAL) && 689 vp->v_type == VREG && 690 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 691 curlwp, 1)) == EINTR) 692 return (error); 693 } 694 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 695 if (vap->va_size != VNOVAL) { 696 if (error) { 697 np->n_size = np->n_vattr->va_size = tsize; 698 uvm_vnp_setsize(vp, np->n_size); 699 } 700 genfs_node_unlock(vp); 701 } 702 VN_KNOTE(vp, NOTE_ATTRIB); 703 return (error); 704 } 705 706 /* 707 * Do an nfs setattr rpc. 708 */ 709 int 710 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 711 { 712 struct nfsv2_sattr *sp; 713 char *cp; 714 int32_t t1, t2; 715 char *bpos, *dpos; 716 u_int32_t *tl; 717 int error = 0; 718 struct mbuf *mreq, *mrep, *md, *mb; 719 const int v3 = NFS_ISV3(vp); 720 struct nfsnode *np = VTONFS(vp); 721 #ifndef NFS_V2_ONLY 722 int wccflag = NFSV3_WCCRATTR; 723 char *cp2; 724 #endif 725 726 nfsstats.rpccnt[NFSPROC_SETATTR]++; 727 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 728 nfsm_fhtom(np, v3); 729 #ifndef NFS_V2_ONLY 730 if (v3) { 731 nfsm_v3attrbuild(vap, true); 732 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 733 *tl = nfs_false; 734 } else { 735 #endif 736 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 737 if (vap->va_mode == (mode_t)VNOVAL) 738 sp->sa_mode = nfs_xdrneg1; 739 else 740 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 741 if (vap->va_uid == (uid_t)VNOVAL) 742 sp->sa_uid = nfs_xdrneg1; 743 else 744 sp->sa_uid = txdr_unsigned(vap->va_uid); 745 if (vap->va_gid == (gid_t)VNOVAL) 746 sp->sa_gid = nfs_xdrneg1; 747 else 748 sp->sa_gid = txdr_unsigned(vap->va_gid); 749 sp->sa_size = txdr_unsigned(vap->va_size); 750 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 751 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 752 #ifndef NFS_V2_ONLY 753 } 754 #endif 755 nfsm_request(np, NFSPROC_SETATTR, l, cred); 756 #ifndef NFS_V2_ONLY 757 if (v3) { 758 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 759 } else 760 #endif 761 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 762 nfsm_reqdone; 763 return (error); 764 } 765 766 /* 767 * nfs lookup call, one step at a time... 768 * First look in cache 769 * If not found, do the rpc. 770 */ 771 int 772 nfs_lookup(void *v) 773 { 774 struct vop_lookup_args /* { 775 struct vnodeop_desc *a_desc; 776 struct vnode *a_dvp; 777 struct vnode **a_vpp; 778 struct componentname *a_cnp; 779 } */ *ap = v; 780 struct componentname *cnp = ap->a_cnp; 781 struct vnode *dvp = ap->a_dvp; 782 struct vnode **vpp = ap->a_vpp; 783 int flags; 784 struct vnode *newvp; 785 u_int32_t *tl; 786 char *cp; 787 int32_t t1, t2; 788 char *bpos, *dpos, *cp2; 789 struct mbuf *mreq, *mrep, *md, *mb; 790 long len; 791 nfsfh_t *fhp; 792 struct nfsnode *np; 793 int error = 0, attrflag, fhsize; 794 const int v3 = NFS_ISV3(dvp); 795 796 flags = cnp->cn_flags; 797 798 *vpp = NULLVP; 799 newvp = NULLVP; 800 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 801 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 802 return (EROFS); 803 if (dvp->v_type != VDIR) 804 return (ENOTDIR); 805 806 /* 807 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 808 */ 809 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 810 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 811 if (error) 812 return error; 813 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 814 return EISDIR; 815 VREF(dvp); 816 *vpp = dvp; 817 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 818 cnp->cn_flags |= SAVENAME; 819 return 0; 820 } 821 822 np = VTONFS(dvp); 823 824 /* 825 * Before performing an RPC, check the name cache to see if 826 * the directory/name pair we are looking for is known already. 827 * If the directory/name pair is found in the name cache, 828 * we have to ensure the directory has not changed from 829 * the time the cache entry has been created. If it has, 830 * the cache entry has to be ignored. 831 */ 832 error = cache_lookup_raw(dvp, vpp, cnp); 833 KASSERT(dvp != *vpp); 834 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 835 if (error >= 0) { 836 struct vattr vattr; 837 int err2; 838 839 if (error && error != ENOENT) { 840 *vpp = NULLVP; 841 return error; 842 } 843 844 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 845 if (err2 != 0) { 846 if (error == 0) 847 vrele(*vpp); 848 *vpp = NULLVP; 849 return err2; 850 } 851 852 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 853 || timespeccmp(&vattr.va_mtime, 854 &VTONFS(dvp)->n_nctime, !=)) { 855 if (error == 0) { 856 vrele(*vpp); 857 *vpp = NULLVP; 858 } 859 cache_purge1(dvp, NULL, PURGE_CHILDREN); 860 timespecclear(&np->n_nctime); 861 goto dorpc; 862 } 863 864 if (error == ENOENT) { 865 goto noentry; 866 } 867 868 /* 869 * investigate the vnode returned by cache_lookup_raw. 870 * if it isn't appropriate, do an rpc. 871 */ 872 newvp = *vpp; 873 if ((flags & ISDOTDOT) != 0) { 874 VOP_UNLOCK(dvp, 0); 875 } 876 error = vn_lock(newvp, LK_EXCLUSIVE); 877 if ((flags & ISDOTDOT) != 0) { 878 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 879 } 880 if (error != 0) { 881 /* newvp has been reclaimed. */ 882 vrele(newvp); 883 *vpp = NULLVP; 884 goto dorpc; 885 } 886 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 887 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 888 nfsstats.lookupcache_hits++; 889 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 890 cnp->cn_flags |= SAVENAME; 891 KASSERT(newvp->v_type != VNON); 892 return (0); 893 } 894 cache_purge1(newvp, NULL, PURGE_PARENTS); 895 vput(newvp); 896 *vpp = NULLVP; 897 } 898 dorpc: 899 #if 0 900 /* 901 * because nfsv3 has the same CREATE semantics as ours, 902 * we don't have to perform LOOKUPs beforehand. 903 * 904 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 905 * XXX although we have no way to know if O_EXCL is requested or not. 906 */ 907 908 if (v3 && cnp->cn_nameiop == CREATE && 909 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 910 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 911 cnp->cn_flags |= SAVENAME; 912 return (EJUSTRETURN); 913 } 914 #endif /* 0 */ 915 916 error = 0; 917 newvp = NULLVP; 918 nfsstats.lookupcache_misses++; 919 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 920 len = cnp->cn_namelen; 921 nfsm_reqhead(np, NFSPROC_LOOKUP, 922 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 923 nfsm_fhtom(np, v3); 924 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 925 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 926 if (error) { 927 nfsm_postop_attr(dvp, attrflag, 0); 928 m_freem(mrep); 929 goto nfsmout; 930 } 931 nfsm_getfh(fhp, fhsize, v3); 932 933 /* 934 * Handle RENAME case... 935 */ 936 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 937 if (NFS_CMPFH(np, fhp, fhsize)) { 938 m_freem(mrep); 939 return (EISDIR); 940 } 941 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 942 if (error) { 943 m_freem(mrep); 944 return error; 945 } 946 newvp = NFSTOV(np); 947 #ifndef NFS_V2_ONLY 948 if (v3) { 949 nfsm_postop_attr(newvp, attrflag, 0); 950 nfsm_postop_attr(dvp, attrflag, 0); 951 } else 952 #endif 953 nfsm_loadattr(newvp, (struct vattr *)0, 0); 954 *vpp = newvp; 955 m_freem(mrep); 956 cnp->cn_flags |= SAVENAME; 957 goto validate; 958 } 959 960 /* 961 * The postop attr handling is duplicated for each if case, 962 * because it should be done while dvp is locked (unlocking 963 * dvp is different for each case). 964 */ 965 966 if (NFS_CMPFH(np, fhp, fhsize)) { 967 /* 968 * as we handle "." lookup locally, this should be 969 * a broken server. 970 */ 971 VREF(dvp); 972 newvp = dvp; 973 #ifndef NFS_V2_ONLY 974 if (v3) { 975 nfsm_postop_attr(newvp, attrflag, 0); 976 nfsm_postop_attr(dvp, attrflag, 0); 977 } else 978 #endif 979 nfsm_loadattr(newvp, (struct vattr *)0, 0); 980 } else if (flags & ISDOTDOT) { 981 /* 982 * ".." lookup 983 */ 984 VOP_UNLOCK(dvp, 0); 985 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 986 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 987 if (error) { 988 m_freem(mrep); 989 return error; 990 } 991 newvp = NFSTOV(np); 992 993 #ifndef NFS_V2_ONLY 994 if (v3) { 995 nfsm_postop_attr(newvp, attrflag, 0); 996 nfsm_postop_attr(dvp, attrflag, 0); 997 } else 998 #endif 999 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1000 } else { 1001 /* 1002 * Other lookups. 1003 */ 1004 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1005 if (error) { 1006 m_freem(mrep); 1007 return error; 1008 } 1009 newvp = NFSTOV(np); 1010 #ifndef NFS_V2_ONLY 1011 if (v3) { 1012 nfsm_postop_attr(newvp, attrflag, 0); 1013 nfsm_postop_attr(dvp, attrflag, 0); 1014 } else 1015 #endif 1016 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1017 } 1018 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1019 cnp->cn_flags |= SAVENAME; 1020 if ((cnp->cn_flags & MAKEENTRY) && 1021 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1022 nfs_cache_enter(dvp, newvp, cnp); 1023 } 1024 *vpp = newvp; 1025 nfsm_reqdone; 1026 if (error) { 1027 /* 1028 * We get here only because of errors returned by 1029 * the RPC. Otherwise we'll have returned above 1030 * (the nfsm_* macros will jump to nfsm_reqdone 1031 * on error). 1032 */ 1033 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1034 cnp->cn_nameiop != CREATE) { 1035 nfs_cache_enter(dvp, NULL, cnp); 1036 } 1037 if (newvp != NULLVP) { 1038 if (newvp == dvp) { 1039 vrele(newvp); 1040 } else { 1041 vput(newvp); 1042 } 1043 } 1044 noentry: 1045 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1046 (flags & ISLASTCN) && error == ENOENT) { 1047 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1048 error = EROFS; 1049 } else { 1050 error = EJUSTRETURN; 1051 cnp->cn_flags |= SAVENAME; 1052 } 1053 } 1054 *vpp = NULL; 1055 return error; 1056 } 1057 1058 validate: 1059 /* 1060 * make sure we have valid type and size. 1061 */ 1062 1063 newvp = *vpp; 1064 if (newvp->v_type == VNON) { 1065 struct vattr vattr; /* dummy */ 1066 1067 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1068 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1069 if (error) { 1070 vput(newvp); 1071 *vpp = NULL; 1072 } 1073 } 1074 1075 return error; 1076 } 1077 1078 /* 1079 * nfs read call. 1080 * Just call nfs_bioread() to do the work. 1081 */ 1082 int 1083 nfs_read(void *v) 1084 { 1085 struct vop_read_args /* { 1086 struct vnode *a_vp; 1087 struct uio *a_uio; 1088 int a_ioflag; 1089 kauth_cred_t a_cred; 1090 } */ *ap = v; 1091 struct vnode *vp = ap->a_vp; 1092 1093 if (vp->v_type != VREG) 1094 return EISDIR; 1095 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1096 } 1097 1098 /* 1099 * nfs readlink call 1100 */ 1101 int 1102 nfs_readlink(void *v) 1103 { 1104 struct vop_readlink_args /* { 1105 struct vnode *a_vp; 1106 struct uio *a_uio; 1107 kauth_cred_t a_cred; 1108 } */ *ap = v; 1109 struct vnode *vp = ap->a_vp; 1110 struct nfsnode *np = VTONFS(vp); 1111 1112 if (vp->v_type != VLNK) 1113 return (EPERM); 1114 1115 if (np->n_rcred != NULL) { 1116 kauth_cred_free(np->n_rcred); 1117 } 1118 np->n_rcred = ap->a_cred; 1119 kauth_cred_hold(np->n_rcred); 1120 1121 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1122 } 1123 1124 /* 1125 * Do a readlink rpc. 1126 * Called by nfs_doio() from below the buffer cache. 1127 */ 1128 int 1129 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1130 { 1131 u_int32_t *tl; 1132 char *cp; 1133 int32_t t1, t2; 1134 char *bpos, *dpos, *cp2; 1135 int error = 0; 1136 uint32_t len; 1137 struct mbuf *mreq, *mrep, *md, *mb; 1138 const int v3 = NFS_ISV3(vp); 1139 struct nfsnode *np = VTONFS(vp); 1140 #ifndef NFS_V2_ONLY 1141 int attrflag; 1142 #endif 1143 1144 nfsstats.rpccnt[NFSPROC_READLINK]++; 1145 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1146 nfsm_fhtom(np, v3); 1147 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1148 #ifndef NFS_V2_ONLY 1149 if (v3) 1150 nfsm_postop_attr(vp, attrflag, 0); 1151 #endif 1152 if (!error) { 1153 #ifndef NFS_V2_ONLY 1154 if (v3) { 1155 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1156 len = fxdr_unsigned(uint32_t, *tl); 1157 if (len > MAXPATHLEN) { 1158 /* 1159 * this pathname is too long for us. 1160 */ 1161 m_freem(mrep); 1162 /* Solaris returns EINVAL. should we follow? */ 1163 error = ENAMETOOLONG; 1164 goto nfsmout; 1165 } 1166 } else 1167 #endif 1168 { 1169 nfsm_strsiz(len, NFS_MAXPATHLEN); 1170 } 1171 nfsm_mtouio(uiop, len); 1172 } 1173 nfsm_reqdone; 1174 return (error); 1175 } 1176 1177 /* 1178 * nfs read rpc call 1179 * Ditto above 1180 */ 1181 int 1182 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1183 { 1184 u_int32_t *tl; 1185 char *cp; 1186 int32_t t1, t2; 1187 char *bpos, *dpos, *cp2; 1188 struct mbuf *mreq, *mrep, *md, *mb; 1189 struct nfsmount *nmp; 1190 int error = 0, len, retlen, tsiz, eof, byte_count; 1191 const int v3 = NFS_ISV3(vp); 1192 struct nfsnode *np = VTONFS(vp); 1193 #ifndef NFS_V2_ONLY 1194 int attrflag; 1195 #endif 1196 1197 #ifndef nolint 1198 eof = 0; 1199 #endif 1200 nmp = VFSTONFS(vp->v_mount); 1201 tsiz = uiop->uio_resid; 1202 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1203 return (EFBIG); 1204 mutex_enter(&nmp->nm_lock); 1205 iostat_busy(nmp->nm_stats); 1206 mutex_exit(&nmp->nm_lock); 1207 byte_count = 0; /* count bytes actually transferred */ 1208 while (tsiz > 0) { 1209 nfsstats.rpccnt[NFSPROC_READ]++; 1210 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1211 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1212 nfsm_fhtom(np, v3); 1213 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1214 #ifndef NFS_V2_ONLY 1215 if (v3) { 1216 txdr_hyper(uiop->uio_offset, tl); 1217 *(tl + 2) = txdr_unsigned(len); 1218 } else 1219 #endif 1220 { 1221 *tl++ = txdr_unsigned(uiop->uio_offset); 1222 *tl++ = txdr_unsigned(len); 1223 *tl = 0; 1224 } 1225 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1226 #ifndef NFS_V2_ONLY 1227 if (v3) { 1228 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1229 if (error) { 1230 m_freem(mrep); 1231 goto nfsmout; 1232 } 1233 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1234 eof = fxdr_unsigned(int, *(tl + 1)); 1235 } else 1236 #endif 1237 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1238 nfsm_strsiz(retlen, nmp->nm_rsize); 1239 nfsm_mtouio(uiop, retlen); 1240 m_freem(mrep); 1241 tsiz -= retlen; 1242 byte_count += retlen; 1243 #ifndef NFS_V2_ONLY 1244 if (v3) { 1245 if (eof || retlen == 0) 1246 tsiz = 0; 1247 } else 1248 #endif 1249 if (retlen < len) 1250 tsiz = 0; 1251 } 1252 nfsmout: 1253 mutex_enter(&nmp->nm_lock); 1254 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1255 mutex_exit(&nmp->nm_lock); 1256 return (error); 1257 } 1258 1259 struct nfs_writerpc_context { 1260 kmutex_t nwc_lock; 1261 kcondvar_t nwc_cv; 1262 int nwc_mbufcount; 1263 }; 1264 1265 /* 1266 * free mbuf used to refer protected pages while write rpc call. 1267 * called at splvm. 1268 */ 1269 static void 1270 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1271 { 1272 struct nfs_writerpc_context *ctx = arg; 1273 1274 KASSERT(m != NULL); 1275 KASSERT(ctx != NULL); 1276 pool_cache_put(mb_cache, m); 1277 mutex_enter(&ctx->nwc_lock); 1278 if (--ctx->nwc_mbufcount == 0) { 1279 cv_signal(&ctx->nwc_cv); 1280 } 1281 mutex_exit(&ctx->nwc_lock); 1282 } 1283 1284 /* 1285 * nfs write call 1286 */ 1287 int 1288 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1289 { 1290 u_int32_t *tl; 1291 char *cp; 1292 int32_t t1, t2; 1293 char *bpos, *dpos; 1294 struct mbuf *mreq, *mrep, *md, *mb; 1295 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1296 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1297 const int v3 = NFS_ISV3(vp); 1298 int committed = NFSV3WRITE_FILESYNC; 1299 struct nfsnode *np = VTONFS(vp); 1300 struct nfs_writerpc_context ctx; 1301 int byte_count; 1302 struct lwp *l = NULL; 1303 size_t origresid; 1304 #ifndef NFS_V2_ONLY 1305 char *cp2; 1306 int rlen, commit; 1307 #endif 1308 1309 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1310 cv_init(&ctx.nwc_cv, "nfsmblk"); 1311 ctx.nwc_mbufcount = 1; 1312 1313 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1314 panic("writerpc readonly vp %p", vp); 1315 } 1316 1317 #ifdef DIAGNOSTIC 1318 if (uiop->uio_iovcnt != 1) 1319 panic("nfs: writerpc iovcnt > 1"); 1320 #endif 1321 tsiz = uiop->uio_resid; 1322 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1323 return (EFBIG); 1324 if (pageprotected) { 1325 l = curlwp; 1326 uvm_lwp_hold(l); 1327 } 1328 retry: 1329 origresid = uiop->uio_resid; 1330 KASSERT(origresid == uiop->uio_iov->iov_len); 1331 mutex_enter(&nmp->nm_lock); 1332 iostat_busy(nmp->nm_stats); 1333 mutex_exit(&nmp->nm_lock); 1334 byte_count = 0; /* count of bytes actually written */ 1335 while (tsiz > 0) { 1336 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1337 uint32_t backup; 1338 bool stalewriteverf = false; 1339 1340 nfsstats.rpccnt[NFSPROC_WRITE]++; 1341 len = min(tsiz, nmp->nm_wsize); 1342 datalen = pageprotected ? 0 : nfsm_rndup(len); 1343 nfsm_reqhead(np, NFSPROC_WRITE, 1344 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1345 nfsm_fhtom(np, v3); 1346 #ifndef NFS_V2_ONLY 1347 if (v3) { 1348 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1349 txdr_hyper(uiop->uio_offset, tl); 1350 tl += 2; 1351 *tl++ = txdr_unsigned(len); 1352 *tl++ = txdr_unsigned(*iomode); 1353 *tl = txdr_unsigned(len); 1354 } else 1355 #endif 1356 { 1357 u_int32_t x; 1358 1359 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1360 /* Set both "begin" and "current" to non-garbage. */ 1361 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1362 *tl++ = x; /* "begin offset" */ 1363 *tl++ = x; /* "current offset" */ 1364 x = txdr_unsigned(len); 1365 *tl++ = x; /* total to this offset */ 1366 *tl = x; /* size of this write */ 1367 1368 } 1369 if (pageprotected) { 1370 /* 1371 * since we know pages can't be modified during i/o, 1372 * no need to copy them for us. 1373 */ 1374 struct mbuf *m; 1375 struct iovec *iovp = uiop->uio_iov; 1376 1377 m = m_get(M_WAIT, MT_DATA); 1378 MCLAIM(m, &nfs_mowner); 1379 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1380 nfs_writerpc_extfree, &ctx); 1381 m->m_flags |= M_EXT_ROMAP; 1382 m->m_len = len; 1383 mb->m_next = m; 1384 /* 1385 * no need to maintain mb and bpos here 1386 * because no one care them later. 1387 */ 1388 #if 0 1389 mb = m; 1390 bpos = mtod(void *, mb) + mb->m_len; 1391 #endif 1392 UIO_ADVANCE(uiop, len); 1393 uiop->uio_offset += len; 1394 mutex_enter(&ctx.nwc_lock); 1395 ctx.nwc_mbufcount++; 1396 mutex_exit(&ctx.nwc_lock); 1397 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1398 } else { 1399 nfsm_uiotom(uiop, len); 1400 } 1401 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1402 #ifndef NFS_V2_ONLY 1403 if (v3) { 1404 wccflag = NFSV3_WCCCHK; 1405 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1406 if (!error) { 1407 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1408 + NFSX_V3WRITEVERF); 1409 rlen = fxdr_unsigned(int, *tl++); 1410 if (rlen == 0) { 1411 error = NFSERR_IO; 1412 m_freem(mrep); 1413 break; 1414 } else if (rlen < len) { 1415 backup = len - rlen; 1416 UIO_ADVANCE(uiop, -backup); 1417 uiop->uio_offset -= backup; 1418 len = rlen; 1419 } 1420 commit = fxdr_unsigned(int, *tl++); 1421 1422 /* 1423 * Return the lowest committment level 1424 * obtained by any of the RPCs. 1425 */ 1426 if (committed == NFSV3WRITE_FILESYNC) 1427 committed = commit; 1428 else if (committed == NFSV3WRITE_DATASYNC && 1429 commit == NFSV3WRITE_UNSTABLE) 1430 committed = commit; 1431 mutex_enter(&nmp->nm_lock); 1432 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1433 memcpy(nmp->nm_writeverf, tl, 1434 NFSX_V3WRITEVERF); 1435 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1436 } else if ((nmp->nm_iflag & 1437 NFSMNT_STALEWRITEVERF) || 1438 memcmp(tl, nmp->nm_writeverf, 1439 NFSX_V3WRITEVERF)) { 1440 memcpy(nmp->nm_writeverf, tl, 1441 NFSX_V3WRITEVERF); 1442 /* 1443 * note NFSMNT_STALEWRITEVERF 1444 * if we're the first thread to 1445 * notice it. 1446 */ 1447 if ((nmp->nm_iflag & 1448 NFSMNT_STALEWRITEVERF) == 0) { 1449 stalewriteverf = true; 1450 nmp->nm_iflag |= 1451 NFSMNT_STALEWRITEVERF; 1452 } 1453 } 1454 mutex_exit(&nmp->nm_lock); 1455 } 1456 } else 1457 #endif 1458 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1459 if (wccflag) 1460 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1461 m_freem(mrep); 1462 if (error) 1463 break; 1464 tsiz -= len; 1465 byte_count += len; 1466 if (stalewriteverf) { 1467 *stalewriteverfp = true; 1468 stalewriteverf = false; 1469 if (committed == NFSV3WRITE_UNSTABLE && 1470 len != origresid) { 1471 /* 1472 * if our write requests weren't atomic but 1473 * unstable, datas in previous iterations 1474 * might have already been lost now. 1475 * then, we should resend them to nfsd. 1476 */ 1477 backup = origresid - tsiz; 1478 UIO_ADVANCE(uiop, -backup); 1479 uiop->uio_offset -= backup; 1480 tsiz = origresid; 1481 goto retry; 1482 } 1483 } 1484 } 1485 nfsmout: 1486 mutex_enter(&nmp->nm_lock); 1487 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1488 mutex_exit(&nmp->nm_lock); 1489 if (pageprotected) { 1490 /* 1491 * wait until mbufs go away. 1492 * retransmitted mbufs can survive longer than rpc requests 1493 * themselves. 1494 */ 1495 mutex_enter(&ctx.nwc_lock); 1496 ctx.nwc_mbufcount--; 1497 while (ctx.nwc_mbufcount > 0) { 1498 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1499 } 1500 mutex_exit(&ctx.nwc_lock); 1501 uvm_lwp_rele(l); 1502 } 1503 mutex_destroy(&ctx.nwc_lock); 1504 cv_destroy(&ctx.nwc_cv); 1505 *iomode = committed; 1506 if (error) 1507 uiop->uio_resid = tsiz; 1508 return (error); 1509 } 1510 1511 /* 1512 * nfs mknod rpc 1513 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1514 * mode set to specify the file type and the size field for rdev. 1515 */ 1516 int 1517 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1518 { 1519 struct nfsv2_sattr *sp; 1520 u_int32_t *tl; 1521 char *cp; 1522 int32_t t1, t2; 1523 struct vnode *newvp = (struct vnode *)0; 1524 struct nfsnode *dnp, *np; 1525 char *cp2; 1526 char *bpos, *dpos; 1527 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1528 struct mbuf *mreq, *mrep, *md, *mb; 1529 u_int32_t rdev; 1530 const int v3 = NFS_ISV3(dvp); 1531 1532 if (vap->va_type == VCHR || vap->va_type == VBLK) 1533 rdev = txdr_unsigned(vap->va_rdev); 1534 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1535 rdev = nfs_xdrneg1; 1536 else { 1537 VOP_ABORTOP(dvp, cnp); 1538 vput(dvp); 1539 return (EOPNOTSUPP); 1540 } 1541 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1542 dnp = VTONFS(dvp); 1543 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1544 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1545 nfsm_fhtom(dnp, v3); 1546 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1547 #ifndef NFS_V2_ONLY 1548 if (v3) { 1549 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1550 *tl++ = vtonfsv3_type(vap->va_type); 1551 nfsm_v3attrbuild(vap, false); 1552 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1553 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1554 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1555 *tl = txdr_unsigned(minor(vap->va_rdev)); 1556 } 1557 } else 1558 #endif 1559 { 1560 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1561 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1562 sp->sa_uid = nfs_xdrneg1; 1563 sp->sa_gid = nfs_xdrneg1; 1564 sp->sa_size = rdev; 1565 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1566 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1567 } 1568 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1569 if (!error) { 1570 nfsm_mtofh(dvp, newvp, v3, gotvp); 1571 if (!gotvp) { 1572 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1573 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1574 if (!error) 1575 newvp = NFSTOV(np); 1576 } 1577 } 1578 #ifndef NFS_V2_ONLY 1579 if (v3) 1580 nfsm_wcc_data(dvp, wccflag, 0, !error); 1581 #endif 1582 nfsm_reqdone; 1583 if (error) { 1584 if (newvp) 1585 vput(newvp); 1586 } else { 1587 if (cnp->cn_flags & MAKEENTRY) 1588 nfs_cache_enter(dvp, newvp, cnp); 1589 *vpp = newvp; 1590 } 1591 PNBUF_PUT(cnp->cn_pnbuf); 1592 VTONFS(dvp)->n_flag |= NMODIFIED; 1593 if (!wccflag) 1594 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1595 vput(dvp); 1596 return (error); 1597 } 1598 1599 /* 1600 * nfs mknod vop 1601 * just call nfs_mknodrpc() to do the work. 1602 */ 1603 /* ARGSUSED */ 1604 int 1605 nfs_mknod(void *v) 1606 { 1607 struct vop_mknod_args /* { 1608 struct vnode *a_dvp; 1609 struct vnode **a_vpp; 1610 struct componentname *a_cnp; 1611 struct vattr *a_vap; 1612 } */ *ap = v; 1613 struct vnode *dvp = ap->a_dvp; 1614 struct componentname *cnp = ap->a_cnp; 1615 int error; 1616 1617 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1618 VN_KNOTE(dvp, NOTE_WRITE); 1619 if (error == 0 || error == EEXIST) 1620 cache_purge1(dvp, cnp, 0); 1621 return (error); 1622 } 1623 1624 /* 1625 * nfs file create call 1626 */ 1627 int 1628 nfs_create(void *v) 1629 { 1630 struct vop_create_args /* { 1631 struct vnode *a_dvp; 1632 struct vnode **a_vpp; 1633 struct componentname *a_cnp; 1634 struct vattr *a_vap; 1635 } */ *ap = v; 1636 struct vnode *dvp = ap->a_dvp; 1637 struct vattr *vap = ap->a_vap; 1638 struct componentname *cnp = ap->a_cnp; 1639 struct nfsv2_sattr *sp; 1640 u_int32_t *tl; 1641 char *cp; 1642 int32_t t1, t2; 1643 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1644 struct vnode *newvp = (struct vnode *)0; 1645 char *bpos, *dpos, *cp2; 1646 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1647 struct mbuf *mreq, *mrep, *md, *mb; 1648 const int v3 = NFS_ISV3(dvp); 1649 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1650 1651 /* 1652 * Oops, not for me.. 1653 */ 1654 if (vap->va_type == VSOCK) 1655 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1656 1657 KASSERT(vap->va_type == VREG); 1658 1659 #ifdef VA_EXCLUSIVE 1660 if (vap->va_vaflags & VA_EXCLUSIVE) { 1661 excl_mode = NFSV3CREATE_EXCLUSIVE; 1662 } 1663 #endif 1664 again: 1665 error = 0; 1666 nfsstats.rpccnt[NFSPROC_CREATE]++; 1667 dnp = VTONFS(dvp); 1668 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1669 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1670 nfsm_fhtom(dnp, v3); 1671 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1672 #ifndef NFS_V2_ONLY 1673 if (v3) { 1674 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1675 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1676 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1677 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1678 *tl++ = arc4random(); 1679 *tl = arc4random(); 1680 } else { 1681 *tl = txdr_unsigned(excl_mode); 1682 nfsm_v3attrbuild(vap, false); 1683 } 1684 } else 1685 #endif 1686 { 1687 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1688 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1689 sp->sa_uid = nfs_xdrneg1; 1690 sp->sa_gid = nfs_xdrneg1; 1691 sp->sa_size = 0; 1692 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1693 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1694 } 1695 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1696 if (!error) { 1697 nfsm_mtofh(dvp, newvp, v3, gotvp); 1698 if (!gotvp) { 1699 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1700 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1701 if (!error) 1702 newvp = NFSTOV(np); 1703 } 1704 } 1705 #ifndef NFS_V2_ONLY 1706 if (v3) 1707 nfsm_wcc_data(dvp, wccflag, 0, !error); 1708 #endif 1709 nfsm_reqdone; 1710 if (error) { 1711 /* 1712 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1713 */ 1714 if (v3 && error == ENOTSUP) { 1715 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1716 excl_mode = NFSV3CREATE_GUARDED; 1717 goto again; 1718 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1719 excl_mode = NFSV3CREATE_UNCHECKED; 1720 goto again; 1721 } 1722 } 1723 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1724 struct timespec ts; 1725 1726 getnanotime(&ts); 1727 1728 /* 1729 * make sure that we'll update timestamps as 1730 * most server implementations use them to store 1731 * the create verifier. 1732 * 1733 * XXX it's better to use TOSERVER always. 1734 */ 1735 1736 if (vap->va_atime.tv_sec == VNOVAL) 1737 vap->va_atime = ts; 1738 if (vap->va_mtime.tv_sec == VNOVAL) 1739 vap->va_mtime = ts; 1740 1741 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1742 } 1743 if (error == 0) { 1744 if (cnp->cn_flags & MAKEENTRY) 1745 nfs_cache_enter(dvp, newvp, cnp); 1746 else 1747 cache_purge1(dvp, cnp, 0); 1748 *ap->a_vpp = newvp; 1749 } else { 1750 if (newvp) 1751 vput(newvp); 1752 if (error == EEXIST) 1753 cache_purge1(dvp, cnp, 0); 1754 } 1755 PNBUF_PUT(cnp->cn_pnbuf); 1756 VTONFS(dvp)->n_flag |= NMODIFIED; 1757 if (!wccflag) 1758 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1759 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1760 vput(dvp); 1761 return (error); 1762 } 1763 1764 /* 1765 * nfs file remove call 1766 * To try and make nfs semantics closer to ufs semantics, a file that has 1767 * other processes using the vnode is renamed instead of removed and then 1768 * removed later on the last close. 1769 * - If v_usecount > 1 1770 * If a rename is not already in the works 1771 * call nfs_sillyrename() to set it up 1772 * else 1773 * do the remove rpc 1774 */ 1775 int 1776 nfs_remove(void *v) 1777 { 1778 struct vop_remove_args /* { 1779 struct vnodeop_desc *a_desc; 1780 struct vnode * a_dvp; 1781 struct vnode * a_vp; 1782 struct componentname * a_cnp; 1783 } */ *ap = v; 1784 struct vnode *vp = ap->a_vp; 1785 struct vnode *dvp = ap->a_dvp; 1786 struct componentname *cnp = ap->a_cnp; 1787 struct nfsnode *np = VTONFS(vp); 1788 int error = 0; 1789 struct vattr vattr; 1790 1791 #ifndef DIAGNOSTIC 1792 if ((cnp->cn_flags & HASBUF) == 0) 1793 panic("nfs_remove: no name"); 1794 if (vp->v_usecount < 1) 1795 panic("nfs_remove: bad v_usecount"); 1796 #endif 1797 if (vp->v_type == VDIR) 1798 error = EPERM; 1799 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1800 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1801 vattr.va_nlink > 1)) { 1802 /* 1803 * Purge the name cache so that the chance of a lookup for 1804 * the name succeeding while the remove is in progress is 1805 * minimized. Without node locking it can still happen, such 1806 * that an I/O op returns ESTALE, but since you get this if 1807 * another host removes the file.. 1808 */ 1809 cache_purge(vp); 1810 /* 1811 * throw away biocache buffers, mainly to avoid 1812 * unnecessary delayed writes later. 1813 */ 1814 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1815 /* Do the rpc */ 1816 if (error != EINTR) 1817 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1818 cnp->cn_namelen, cnp->cn_cred, curlwp); 1819 } else if (!np->n_sillyrename) 1820 error = nfs_sillyrename(dvp, vp, cnp, false); 1821 PNBUF_PUT(cnp->cn_pnbuf); 1822 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1823 vattr.va_nlink == 1) { 1824 np->n_flag |= NREMOVED; 1825 } 1826 NFS_INVALIDATE_ATTRCACHE(np); 1827 VN_KNOTE(vp, NOTE_DELETE); 1828 VN_KNOTE(dvp, NOTE_WRITE); 1829 if (dvp == vp) 1830 vrele(vp); 1831 else 1832 vput(vp); 1833 vput(dvp); 1834 return (error); 1835 } 1836 1837 /* 1838 * nfs file remove rpc called from nfs_inactive 1839 */ 1840 int 1841 nfs_removeit(struct sillyrename *sp) 1842 { 1843 1844 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1845 (struct lwp *)0)); 1846 } 1847 1848 /* 1849 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1850 */ 1851 int 1852 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1853 { 1854 u_int32_t *tl; 1855 char *cp; 1856 #ifndef NFS_V2_ONLY 1857 int32_t t1; 1858 char *cp2; 1859 #endif 1860 int32_t t2; 1861 char *bpos, *dpos; 1862 int error = 0, wccflag = NFSV3_WCCRATTR; 1863 struct mbuf *mreq, *mrep, *md, *mb; 1864 const int v3 = NFS_ISV3(dvp); 1865 int rexmit = 0; 1866 struct nfsnode *dnp = VTONFS(dvp); 1867 1868 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1869 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1870 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1871 nfsm_fhtom(dnp, v3); 1872 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1873 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1874 #ifndef NFS_V2_ONLY 1875 if (v3) 1876 nfsm_wcc_data(dvp, wccflag, 0, !error); 1877 #endif 1878 nfsm_reqdone; 1879 VTONFS(dvp)->n_flag |= NMODIFIED; 1880 if (!wccflag) 1881 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1882 /* 1883 * Kludge City: If the first reply to the remove rpc is lost.. 1884 * the reply to the retransmitted request will be ENOENT 1885 * since the file was in fact removed 1886 * Therefore, we cheat and return success. 1887 */ 1888 if (rexmit && error == ENOENT) 1889 error = 0; 1890 return (error); 1891 } 1892 1893 /* 1894 * nfs file rename call 1895 */ 1896 int 1897 nfs_rename(void *v) 1898 { 1899 struct vop_rename_args /* { 1900 struct vnode *a_fdvp; 1901 struct vnode *a_fvp; 1902 struct componentname *a_fcnp; 1903 struct vnode *a_tdvp; 1904 struct vnode *a_tvp; 1905 struct componentname *a_tcnp; 1906 } */ *ap = v; 1907 struct vnode *fvp = ap->a_fvp; 1908 struct vnode *tvp = ap->a_tvp; 1909 struct vnode *fdvp = ap->a_fdvp; 1910 struct vnode *tdvp = ap->a_tdvp; 1911 struct componentname *tcnp = ap->a_tcnp; 1912 struct componentname *fcnp = ap->a_fcnp; 1913 int error; 1914 1915 #ifndef DIAGNOSTIC 1916 if ((tcnp->cn_flags & HASBUF) == 0 || 1917 (fcnp->cn_flags & HASBUF) == 0) 1918 panic("nfs_rename: no name"); 1919 #endif 1920 /* Check for cross-device rename */ 1921 if ((fvp->v_mount != tdvp->v_mount) || 1922 (tvp && (fvp->v_mount != tvp->v_mount))) { 1923 error = EXDEV; 1924 goto out; 1925 } 1926 1927 /* 1928 * If the tvp exists and is in use, sillyrename it before doing the 1929 * rename of the new file over it. 1930 * 1931 * Have sillyrename use link instead of rename if possible, 1932 * so that we don't lose the file if the rename fails, and so 1933 * that there's no window when the "to" file doesn't exist. 1934 */ 1935 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1936 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1937 VN_KNOTE(tvp, NOTE_DELETE); 1938 vput(tvp); 1939 tvp = NULL; 1940 } 1941 1942 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1943 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1944 curlwp); 1945 1946 VN_KNOTE(fdvp, NOTE_WRITE); 1947 VN_KNOTE(tdvp, NOTE_WRITE); 1948 if (error == 0 || error == EEXIST) { 1949 if (fvp->v_type == VDIR) 1950 cache_purge(fvp); 1951 else 1952 cache_purge1(fdvp, fcnp, 0); 1953 if (tvp != NULL && tvp->v_type == VDIR) 1954 cache_purge(tvp); 1955 else 1956 cache_purge1(tdvp, tcnp, 0); 1957 } 1958 out: 1959 if (tdvp == tvp) 1960 vrele(tdvp); 1961 else 1962 vput(tdvp); 1963 if (tvp) 1964 vput(tvp); 1965 vrele(fdvp); 1966 vrele(fvp); 1967 return (error); 1968 } 1969 1970 /* 1971 * nfs file rename rpc called from nfs_remove() above 1972 */ 1973 int 1974 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1975 { 1976 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1977 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1978 } 1979 1980 /* 1981 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1982 */ 1983 int 1984 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) 1985 { 1986 u_int32_t *tl; 1987 char *cp; 1988 #ifndef NFS_V2_ONLY 1989 int32_t t1; 1990 char *cp2; 1991 #endif 1992 int32_t t2; 1993 char *bpos, *dpos; 1994 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1995 struct mbuf *mreq, *mrep, *md, *mb; 1996 const int v3 = NFS_ISV3(fdvp); 1997 int rexmit = 0; 1998 struct nfsnode *fdnp = VTONFS(fdvp); 1999 2000 nfsstats.rpccnt[NFSPROC_RENAME]++; 2001 nfsm_reqhead(fdnp, NFSPROC_RENAME, 2002 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 2003 nfsm_rndup(tnamelen)); 2004 nfsm_fhtom(fdnp, v3); 2005 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 2006 nfsm_fhtom(VTONFS(tdvp), v3); 2007 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 2008 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 2009 #ifndef NFS_V2_ONLY 2010 if (v3) { 2011 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 2012 nfsm_wcc_data(tdvp, twccflag, 0, !error); 2013 } 2014 #endif 2015 nfsm_reqdone; 2016 VTONFS(fdvp)->n_flag |= NMODIFIED; 2017 VTONFS(tdvp)->n_flag |= NMODIFIED; 2018 if (!fwccflag) 2019 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 2020 if (!twccflag) 2021 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 2022 /* 2023 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2024 */ 2025 if (rexmit && error == ENOENT) 2026 error = 0; 2027 return (error); 2028 } 2029 2030 /* 2031 * NFS link RPC, called from nfs_link. 2032 * Assumes dvp and vp locked, and leaves them that way. 2033 */ 2034 2035 static int 2036 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2037 size_t namelen, kauth_cred_t cred, struct lwp *l) 2038 { 2039 u_int32_t *tl; 2040 char *cp; 2041 #ifndef NFS_V2_ONLY 2042 int32_t t1; 2043 char *cp2; 2044 #endif 2045 int32_t t2; 2046 char *bpos, *dpos; 2047 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2048 struct mbuf *mreq, *mrep, *md, *mb; 2049 const int v3 = NFS_ISV3(dvp); 2050 int rexmit = 0; 2051 struct nfsnode *np = VTONFS(vp); 2052 2053 nfsstats.rpccnt[NFSPROC_LINK]++; 2054 nfsm_reqhead(np, NFSPROC_LINK, 2055 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2056 nfsm_fhtom(np, v3); 2057 nfsm_fhtom(VTONFS(dvp), v3); 2058 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2059 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2060 #ifndef NFS_V2_ONLY 2061 if (v3) { 2062 nfsm_postop_attr(vp, attrflag, 0); 2063 nfsm_wcc_data(dvp, wccflag, 0, !error); 2064 } 2065 #endif 2066 nfsm_reqdone; 2067 2068 VTONFS(dvp)->n_flag |= NMODIFIED; 2069 if (!attrflag) 2070 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2071 if (!wccflag) 2072 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2073 2074 /* 2075 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2076 */ 2077 if (rexmit && error == EEXIST) 2078 error = 0; 2079 2080 return error; 2081 } 2082 2083 /* 2084 * nfs hard link create call 2085 */ 2086 int 2087 nfs_link(void *v) 2088 { 2089 struct vop_link_args /* { 2090 struct vnode *a_dvp; 2091 struct vnode *a_vp; 2092 struct componentname *a_cnp; 2093 } */ *ap = v; 2094 struct vnode *vp = ap->a_vp; 2095 struct vnode *dvp = ap->a_dvp; 2096 struct componentname *cnp = ap->a_cnp; 2097 int error = 0; 2098 2099 if (dvp->v_mount != vp->v_mount) { 2100 VOP_ABORTOP(dvp, cnp); 2101 vput(dvp); 2102 return (EXDEV); 2103 } 2104 if (dvp != vp) { 2105 error = vn_lock(vp, LK_EXCLUSIVE); 2106 if (error != 0) { 2107 VOP_ABORTOP(dvp, cnp); 2108 vput(dvp); 2109 return error; 2110 } 2111 } 2112 2113 /* 2114 * Push all writes to the server, so that the attribute cache 2115 * doesn't get "out of sync" with the server. 2116 * XXX There should be a better way! 2117 */ 2118 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2119 2120 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2121 cnp->cn_cred, curlwp); 2122 2123 if (error == 0) 2124 cache_purge1(dvp, cnp, 0); 2125 PNBUF_PUT(cnp->cn_pnbuf); 2126 if (dvp != vp) 2127 VOP_UNLOCK(vp, 0); 2128 VN_KNOTE(vp, NOTE_LINK); 2129 VN_KNOTE(dvp, NOTE_WRITE); 2130 vput(dvp); 2131 return (error); 2132 } 2133 2134 /* 2135 * nfs symbolic link create call 2136 */ 2137 int 2138 nfs_symlink(void *v) 2139 { 2140 struct vop_symlink_args /* { 2141 struct vnode *a_dvp; 2142 struct vnode **a_vpp; 2143 struct componentname *a_cnp; 2144 struct vattr *a_vap; 2145 char *a_target; 2146 } */ *ap = v; 2147 struct vnode *dvp = ap->a_dvp; 2148 struct vattr *vap = ap->a_vap; 2149 struct componentname *cnp = ap->a_cnp; 2150 struct nfsv2_sattr *sp; 2151 u_int32_t *tl; 2152 char *cp; 2153 int32_t t1, t2; 2154 char *bpos, *dpos, *cp2; 2155 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2156 struct mbuf *mreq, *mrep, *md, *mb; 2157 struct vnode *newvp = (struct vnode *)0; 2158 const int v3 = NFS_ISV3(dvp); 2159 int rexmit = 0; 2160 struct nfsnode *dnp = VTONFS(dvp); 2161 2162 *ap->a_vpp = NULL; 2163 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2164 slen = strlen(ap->a_target); 2165 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2166 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2167 nfsm_fhtom(dnp, v3); 2168 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2169 #ifndef NFS_V2_ONlY 2170 if (v3) 2171 nfsm_v3attrbuild(vap, false); 2172 #endif 2173 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2174 #ifndef NFS_V2_ONlY 2175 if (!v3) { 2176 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2177 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2178 sp->sa_uid = nfs_xdrneg1; 2179 sp->sa_gid = nfs_xdrneg1; 2180 sp->sa_size = nfs_xdrneg1; 2181 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2182 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2183 } 2184 #endif 2185 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2186 &rexmit); 2187 #ifndef NFS_V2_ONlY 2188 if (v3) { 2189 if (!error) 2190 nfsm_mtofh(dvp, newvp, v3, gotvp); 2191 nfsm_wcc_data(dvp, wccflag, 0, !error); 2192 } 2193 #endif 2194 nfsm_reqdone; 2195 /* 2196 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2197 */ 2198 if (rexmit && error == EEXIST) 2199 error = 0; 2200 if (error == 0 || error == EEXIST) 2201 cache_purge1(dvp, cnp, 0); 2202 if (error == 0 && newvp == NULL) { 2203 struct nfsnode *np = NULL; 2204 2205 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2206 cnp->cn_cred, curlwp, &np); 2207 if (error == 0) 2208 newvp = NFSTOV(np); 2209 } 2210 if (error) { 2211 if (newvp != NULL) 2212 vput(newvp); 2213 } else { 2214 *ap->a_vpp = newvp; 2215 } 2216 PNBUF_PUT(cnp->cn_pnbuf); 2217 VTONFS(dvp)->n_flag |= NMODIFIED; 2218 if (!wccflag) 2219 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2220 VN_KNOTE(dvp, NOTE_WRITE); 2221 vput(dvp); 2222 return (error); 2223 } 2224 2225 /* 2226 * nfs make dir call 2227 */ 2228 int 2229 nfs_mkdir(void *v) 2230 { 2231 struct vop_mkdir_args /* { 2232 struct vnode *a_dvp; 2233 struct vnode **a_vpp; 2234 struct componentname *a_cnp; 2235 struct vattr *a_vap; 2236 } */ *ap = v; 2237 struct vnode *dvp = ap->a_dvp; 2238 struct vattr *vap = ap->a_vap; 2239 struct componentname *cnp = ap->a_cnp; 2240 struct nfsv2_sattr *sp; 2241 u_int32_t *tl; 2242 char *cp; 2243 int32_t t1, t2; 2244 int len; 2245 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2246 struct vnode *newvp = (struct vnode *)0; 2247 char *bpos, *dpos, *cp2; 2248 int error = 0, wccflag = NFSV3_WCCRATTR; 2249 int gotvp = 0; 2250 int rexmit = 0; 2251 struct mbuf *mreq, *mrep, *md, *mb; 2252 const int v3 = NFS_ISV3(dvp); 2253 2254 len = cnp->cn_namelen; 2255 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2256 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2257 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2258 nfsm_fhtom(dnp, v3); 2259 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2260 #ifndef NFS_V2_ONLY 2261 if (v3) { 2262 nfsm_v3attrbuild(vap, false); 2263 } else 2264 #endif 2265 { 2266 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2267 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2268 sp->sa_uid = nfs_xdrneg1; 2269 sp->sa_gid = nfs_xdrneg1; 2270 sp->sa_size = nfs_xdrneg1; 2271 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2272 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2273 } 2274 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2275 if (!error) 2276 nfsm_mtofh(dvp, newvp, v3, gotvp); 2277 if (v3) 2278 nfsm_wcc_data(dvp, wccflag, 0, !error); 2279 nfsm_reqdone; 2280 VTONFS(dvp)->n_flag |= NMODIFIED; 2281 if (!wccflag) 2282 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2283 /* 2284 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2285 * if we can succeed in looking up the directory. 2286 */ 2287 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2288 if (newvp) { 2289 vput(newvp); 2290 newvp = (struct vnode *)0; 2291 } 2292 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2293 curlwp, &np); 2294 if (!error) { 2295 newvp = NFSTOV(np); 2296 if (newvp->v_type != VDIR || newvp == dvp) 2297 error = EEXIST; 2298 } 2299 } 2300 if (error) { 2301 if (newvp) { 2302 if (dvp != newvp) 2303 vput(newvp); 2304 else 2305 vrele(newvp); 2306 } 2307 } else { 2308 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2309 if (cnp->cn_flags & MAKEENTRY) 2310 nfs_cache_enter(dvp, newvp, cnp); 2311 *ap->a_vpp = newvp; 2312 } 2313 PNBUF_PUT(cnp->cn_pnbuf); 2314 vput(dvp); 2315 return (error); 2316 } 2317 2318 /* 2319 * nfs remove directory call 2320 */ 2321 int 2322 nfs_rmdir(void *v) 2323 { 2324 struct vop_rmdir_args /* { 2325 struct vnode *a_dvp; 2326 struct vnode *a_vp; 2327 struct componentname *a_cnp; 2328 } */ *ap = v; 2329 struct vnode *vp = ap->a_vp; 2330 struct vnode *dvp = ap->a_dvp; 2331 struct componentname *cnp = ap->a_cnp; 2332 u_int32_t *tl; 2333 char *cp; 2334 #ifndef NFS_V2_ONLY 2335 int32_t t1; 2336 char *cp2; 2337 #endif 2338 int32_t t2; 2339 char *bpos, *dpos; 2340 int error = 0, wccflag = NFSV3_WCCRATTR; 2341 int rexmit = 0; 2342 struct mbuf *mreq, *mrep, *md, *mb; 2343 const int v3 = NFS_ISV3(dvp); 2344 struct nfsnode *dnp; 2345 2346 if (dvp == vp) { 2347 vrele(dvp); 2348 vput(dvp); 2349 PNBUF_PUT(cnp->cn_pnbuf); 2350 return (EINVAL); 2351 } 2352 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2353 dnp = VTONFS(dvp); 2354 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2355 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2356 nfsm_fhtom(dnp, v3); 2357 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2358 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2359 #ifndef NFS_V2_ONLY 2360 if (v3) 2361 nfsm_wcc_data(dvp, wccflag, 0, !error); 2362 #endif 2363 nfsm_reqdone; 2364 PNBUF_PUT(cnp->cn_pnbuf); 2365 VTONFS(dvp)->n_flag |= NMODIFIED; 2366 if (!wccflag) 2367 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2368 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2369 VN_KNOTE(vp, NOTE_DELETE); 2370 cache_purge(vp); 2371 vput(vp); 2372 vput(dvp); 2373 /* 2374 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2375 */ 2376 if (rexmit && error == ENOENT) 2377 error = 0; 2378 return (error); 2379 } 2380 2381 /* 2382 * nfs readdir call 2383 */ 2384 int 2385 nfs_readdir(void *v) 2386 { 2387 struct vop_readdir_args /* { 2388 struct vnode *a_vp; 2389 struct uio *a_uio; 2390 kauth_cred_t a_cred; 2391 int *a_eofflag; 2392 off_t **a_cookies; 2393 int *a_ncookies; 2394 } */ *ap = v; 2395 struct vnode *vp = ap->a_vp; 2396 struct uio *uio = ap->a_uio; 2397 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2398 char *base = uio->uio_iov->iov_base; 2399 int tresid, error; 2400 size_t count, lost; 2401 struct dirent *dp; 2402 off_t *cookies = NULL; 2403 int ncookies = 0, nc; 2404 2405 if (vp->v_type != VDIR) 2406 return (EPERM); 2407 2408 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2409 count = uio->uio_resid - lost; 2410 if (count <= 0) 2411 return (EINVAL); 2412 2413 /* 2414 * Call nfs_bioread() to do the real work. 2415 */ 2416 tresid = uio->uio_resid = count; 2417 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2418 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2419 2420 if (!error && ap->a_cookies) { 2421 ncookies = count / 16; 2422 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2423 *ap->a_cookies = cookies; 2424 } 2425 2426 if (!error && uio->uio_resid == tresid) { 2427 uio->uio_resid += lost; 2428 nfsstats.direofcache_misses++; 2429 if (ap->a_cookies) 2430 *ap->a_ncookies = 0; 2431 *ap->a_eofflag = 1; 2432 return (0); 2433 } 2434 2435 if (!error && ap->a_cookies) { 2436 /* 2437 * Only the NFS server and emulations use cookies, and they 2438 * load the directory block into system space, so we can 2439 * just look at it directly. 2440 */ 2441 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2442 uio->uio_iovcnt != 1) 2443 panic("nfs_readdir: lost in space"); 2444 for (nc = 0; ncookies-- && 2445 base < (char *)uio->uio_iov->iov_base; nc++){ 2446 dp = (struct dirent *) base; 2447 if (dp->d_reclen == 0) 2448 break; 2449 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2450 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2451 else 2452 *(cookies++) = NFS_GETCOOKIE(dp); 2453 base += dp->d_reclen; 2454 } 2455 uio->uio_resid += 2456 ((char *)uio->uio_iov->iov_base - base); 2457 uio->uio_iov->iov_len += 2458 ((char *)uio->uio_iov->iov_base - base); 2459 uio->uio_iov->iov_base = base; 2460 *ap->a_ncookies = nc; 2461 } 2462 2463 uio->uio_resid += lost; 2464 *ap->a_eofflag = 0; 2465 return (error); 2466 } 2467 2468 /* 2469 * Readdir rpc call. 2470 * Called from below the buffer cache by nfs_doio(). 2471 */ 2472 int 2473 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2474 { 2475 int len, left; 2476 struct dirent *dp = NULL; 2477 u_int32_t *tl; 2478 char *cp; 2479 int32_t t1, t2; 2480 char *bpos, *dpos, *cp2; 2481 struct mbuf *mreq, *mrep, *md, *mb; 2482 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2483 struct nfsnode *dnp = VTONFS(vp); 2484 u_quad_t fileno; 2485 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2486 #ifndef NFS_V2_ONLY 2487 int attrflag; 2488 #endif 2489 int nrpcs = 0, reclen; 2490 const int v3 = NFS_ISV3(vp); 2491 2492 #ifdef DIAGNOSTIC 2493 /* 2494 * Should be called from buffer cache, so only amount of 2495 * NFS_DIRBLKSIZ will be requested. 2496 */ 2497 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2498 panic("nfs readdirrpc bad uio"); 2499 #endif 2500 2501 /* 2502 * Loop around doing readdir rpc's of size nm_readdirsize 2503 * truncated to a multiple of NFS_DIRFRAGSIZ. 2504 * The stopping criteria is EOF or buffer full. 2505 */ 2506 while (more_dirs && bigenough) { 2507 /* 2508 * Heuristic: don't bother to do another RPC to further 2509 * fill up this block if there is not much room left. (< 50% 2510 * of the readdir RPC size). This wastes some buffer space 2511 * but can save up to 50% in RPC calls. 2512 */ 2513 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2514 bigenough = 0; 2515 break; 2516 } 2517 nfsstats.rpccnt[NFSPROC_READDIR]++; 2518 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2519 NFSX_READDIR(v3)); 2520 nfsm_fhtom(dnp, v3); 2521 #ifndef NFS_V2_ONLY 2522 if (v3) { 2523 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2524 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2525 txdr_swapcookie3(uiop->uio_offset, tl); 2526 } else { 2527 txdr_cookie3(uiop->uio_offset, tl); 2528 } 2529 tl += 2; 2530 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2531 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2532 } else 2533 #endif 2534 { 2535 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2536 *tl++ = txdr_unsigned(uiop->uio_offset); 2537 } 2538 *tl = txdr_unsigned(nmp->nm_readdirsize); 2539 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2540 nrpcs++; 2541 #ifndef NFS_V2_ONLY 2542 if (v3) { 2543 nfsm_postop_attr(vp, attrflag, 0); 2544 if (!error) { 2545 nfsm_dissect(tl, u_int32_t *, 2546 2 * NFSX_UNSIGNED); 2547 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2548 dnp->n_cookieverf.nfsuquad[1] = *tl; 2549 } else { 2550 m_freem(mrep); 2551 goto nfsmout; 2552 } 2553 } 2554 #endif 2555 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2556 more_dirs = fxdr_unsigned(int, *tl); 2557 2558 /* loop thru the dir entries, doctoring them to 4bsd form */ 2559 while (more_dirs && bigenough) { 2560 #ifndef NFS_V2_ONLY 2561 if (v3) { 2562 nfsm_dissect(tl, u_int32_t *, 2563 3 * NFSX_UNSIGNED); 2564 fileno = fxdr_hyper(tl); 2565 len = fxdr_unsigned(int, *(tl + 2)); 2566 } else 2567 #endif 2568 { 2569 nfsm_dissect(tl, u_int32_t *, 2570 2 * NFSX_UNSIGNED); 2571 fileno = fxdr_unsigned(u_quad_t, *tl++); 2572 len = fxdr_unsigned(int, *tl); 2573 } 2574 if (len <= 0 || len > NFS_MAXNAMLEN) { 2575 error = EBADRPC; 2576 m_freem(mrep); 2577 goto nfsmout; 2578 } 2579 /* for cookie stashing */ 2580 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2581 left = NFS_DIRFRAGSIZ - blksiz; 2582 if (reclen > left) { 2583 memset(uiop->uio_iov->iov_base, 0, left); 2584 dp->d_reclen += left; 2585 UIO_ADVANCE(uiop, left); 2586 blksiz = 0; 2587 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2588 } 2589 if (reclen > uiop->uio_resid) 2590 bigenough = 0; 2591 if (bigenough) { 2592 int tlen; 2593 2594 dp = (struct dirent *)uiop->uio_iov->iov_base; 2595 dp->d_fileno = fileno; 2596 dp->d_namlen = len; 2597 dp->d_reclen = reclen; 2598 dp->d_type = DT_UNKNOWN; 2599 blksiz += reclen; 2600 if (blksiz == NFS_DIRFRAGSIZ) 2601 blksiz = 0; 2602 UIO_ADVANCE(uiop, DIRHDSIZ); 2603 nfsm_mtouio(uiop, len); 2604 tlen = reclen - (DIRHDSIZ + len); 2605 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2606 UIO_ADVANCE(uiop, tlen); 2607 } else 2608 nfsm_adv(nfsm_rndup(len)); 2609 #ifndef NFS_V2_ONLY 2610 if (v3) { 2611 nfsm_dissect(tl, u_int32_t *, 2612 3 * NFSX_UNSIGNED); 2613 } else 2614 #endif 2615 { 2616 nfsm_dissect(tl, u_int32_t *, 2617 2 * NFSX_UNSIGNED); 2618 } 2619 if (bigenough) { 2620 #ifndef NFS_V2_ONLY 2621 if (v3) { 2622 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2623 uiop->uio_offset = 2624 fxdr_swapcookie3(tl); 2625 else 2626 uiop->uio_offset = 2627 fxdr_cookie3(tl); 2628 } 2629 else 2630 #endif 2631 { 2632 uiop->uio_offset = 2633 fxdr_unsigned(off_t, *tl); 2634 } 2635 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2636 } 2637 if (v3) 2638 tl += 2; 2639 else 2640 tl++; 2641 more_dirs = fxdr_unsigned(int, *tl); 2642 } 2643 /* 2644 * If at end of rpc data, get the eof boolean 2645 */ 2646 if (!more_dirs) { 2647 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2648 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2649 2650 /* 2651 * kludge: if we got no entries, treat it as EOF. 2652 * some server sometimes send a reply without any 2653 * entries or EOF. 2654 * although it might mean the server has very long name, 2655 * we can't handle such entries anyway. 2656 */ 2657 2658 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2659 more_dirs = 0; 2660 } 2661 m_freem(mrep); 2662 } 2663 /* 2664 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2665 * by increasing d_reclen for the last record. 2666 */ 2667 if (blksiz > 0) { 2668 left = NFS_DIRFRAGSIZ - blksiz; 2669 memset(uiop->uio_iov->iov_base, 0, left); 2670 dp->d_reclen += left; 2671 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2672 UIO_ADVANCE(uiop, left); 2673 } 2674 2675 /* 2676 * We are now either at the end of the directory or have filled the 2677 * block. 2678 */ 2679 if (bigenough) { 2680 dnp->n_direofoffset = uiop->uio_offset; 2681 dnp->n_flag |= NEOFVALID; 2682 } 2683 nfsmout: 2684 return (error); 2685 } 2686 2687 #ifndef NFS_V2_ONLY 2688 /* 2689 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2690 */ 2691 int 2692 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2693 { 2694 int len, left; 2695 struct dirent *dp = NULL; 2696 u_int32_t *tl; 2697 char *cp; 2698 int32_t t1, t2; 2699 struct vnode *newvp; 2700 char *bpos, *dpos, *cp2; 2701 struct mbuf *mreq, *mrep, *md, *mb; 2702 struct nameidata nami, *ndp = &nami; 2703 struct componentname *cnp = &ndp->ni_cnd; 2704 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2705 struct nfsnode *dnp = VTONFS(vp), *np; 2706 nfsfh_t *fhp; 2707 u_quad_t fileno; 2708 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2709 int attrflag, fhsize, nrpcs = 0, reclen; 2710 struct nfs_fattr fattr, *fp; 2711 2712 #ifdef DIAGNOSTIC 2713 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2714 panic("nfs readdirplusrpc bad uio"); 2715 #endif 2716 ndp->ni_dvp = vp; 2717 newvp = NULLVP; 2718 2719 /* 2720 * Loop around doing readdir rpc's of size nm_readdirsize 2721 * truncated to a multiple of NFS_DIRFRAGSIZ. 2722 * The stopping criteria is EOF or buffer full. 2723 */ 2724 while (more_dirs && bigenough) { 2725 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2726 bigenough = 0; 2727 break; 2728 } 2729 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2730 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2731 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2732 nfsm_fhtom(dnp, 1); 2733 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2734 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2735 txdr_swapcookie3(uiop->uio_offset, tl); 2736 } else { 2737 txdr_cookie3(uiop->uio_offset, tl); 2738 } 2739 tl += 2; 2740 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2741 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2742 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2743 *tl = txdr_unsigned(nmp->nm_rsize); 2744 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2745 nfsm_postop_attr(vp, attrflag, 0); 2746 if (error) { 2747 m_freem(mrep); 2748 goto nfsmout; 2749 } 2750 nrpcs++; 2751 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2752 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2753 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2754 more_dirs = fxdr_unsigned(int, *tl); 2755 2756 /* loop thru the dir entries, doctoring them to 4bsd form */ 2757 while (more_dirs && bigenough) { 2758 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2759 fileno = fxdr_hyper(tl); 2760 len = fxdr_unsigned(int, *(tl + 2)); 2761 if (len <= 0 || len > NFS_MAXNAMLEN) { 2762 error = EBADRPC; 2763 m_freem(mrep); 2764 goto nfsmout; 2765 } 2766 /* for cookie stashing */ 2767 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2768 left = NFS_DIRFRAGSIZ - blksiz; 2769 if (reclen > left) { 2770 /* 2771 * DIRFRAGSIZ is aligned, no need to align 2772 * again here. 2773 */ 2774 memset(uiop->uio_iov->iov_base, 0, left); 2775 dp->d_reclen += left; 2776 UIO_ADVANCE(uiop, left); 2777 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2778 blksiz = 0; 2779 } 2780 if (reclen > uiop->uio_resid) 2781 bigenough = 0; 2782 if (bigenough) { 2783 int tlen; 2784 2785 dp = (struct dirent *)uiop->uio_iov->iov_base; 2786 dp->d_fileno = fileno; 2787 dp->d_namlen = len; 2788 dp->d_reclen = reclen; 2789 dp->d_type = DT_UNKNOWN; 2790 blksiz += reclen; 2791 if (blksiz == NFS_DIRFRAGSIZ) 2792 blksiz = 0; 2793 UIO_ADVANCE(uiop, DIRHDSIZ); 2794 nfsm_mtouio(uiop, len); 2795 tlen = reclen - (DIRHDSIZ + len); 2796 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2797 UIO_ADVANCE(uiop, tlen); 2798 cnp->cn_nameptr = dp->d_name; 2799 cnp->cn_namelen = dp->d_namlen; 2800 } else 2801 nfsm_adv(nfsm_rndup(len)); 2802 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2803 if (bigenough) { 2804 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2805 uiop->uio_offset = 2806 fxdr_swapcookie3(tl); 2807 else 2808 uiop->uio_offset = 2809 fxdr_cookie3(tl); 2810 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2811 } 2812 tl += 2; 2813 2814 /* 2815 * Since the attributes are before the file handle 2816 * (sigh), we must skip over the attributes and then 2817 * come back and get them. 2818 */ 2819 attrflag = fxdr_unsigned(int, *tl); 2820 if (attrflag) { 2821 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2822 memcpy(&fattr, fp, NFSX_V3FATTR); 2823 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2824 doit = fxdr_unsigned(int, *tl); 2825 if (doit) { 2826 nfsm_getfh(fhp, fhsize, 1); 2827 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2828 VREF(vp); 2829 newvp = vp; 2830 np = dnp; 2831 } else { 2832 error = nfs_nget1(vp->v_mount, fhp, 2833 fhsize, &np, LK_NOWAIT); 2834 if (!error) 2835 newvp = NFSTOV(np); 2836 } 2837 if (!error) { 2838 const char *xcp; 2839 2840 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2841 if (bigenough) { 2842 dp->d_type = 2843 IFTODT(VTTOIF(np->n_vattr->va_type)); 2844 if (cnp->cn_namelen <= NCHNAMLEN) { 2845 ndp->ni_vp = newvp; 2846 xcp = cnp->cn_nameptr + 2847 cnp->cn_namelen; 2848 cnp->cn_hash = 2849 namei_hash(cnp->cn_nameptr, &xcp); 2850 nfs_cache_enter(ndp->ni_dvp, 2851 ndp->ni_vp, cnp); 2852 } 2853 } 2854 } 2855 error = 0; 2856 } 2857 } else { 2858 /* Just skip over the file handle */ 2859 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2860 i = fxdr_unsigned(int, *tl); 2861 nfsm_adv(nfsm_rndup(i)); 2862 } 2863 if (newvp != NULLVP) { 2864 if (newvp == vp) 2865 vrele(newvp); 2866 else 2867 vput(newvp); 2868 newvp = NULLVP; 2869 } 2870 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2871 more_dirs = fxdr_unsigned(int, *tl); 2872 } 2873 /* 2874 * If at end of rpc data, get the eof boolean 2875 */ 2876 if (!more_dirs) { 2877 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2878 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2879 2880 /* 2881 * kludge: see a comment in nfs_readdirrpc. 2882 */ 2883 2884 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2885 more_dirs = 0; 2886 } 2887 m_freem(mrep); 2888 } 2889 /* 2890 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2891 * by increasing d_reclen for the last record. 2892 */ 2893 if (blksiz > 0) { 2894 left = NFS_DIRFRAGSIZ - blksiz; 2895 memset(uiop->uio_iov->iov_base, 0, left); 2896 dp->d_reclen += left; 2897 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2898 UIO_ADVANCE(uiop, left); 2899 } 2900 2901 /* 2902 * We are now either at the end of the directory or have filled the 2903 * block. 2904 */ 2905 if (bigenough) { 2906 dnp->n_direofoffset = uiop->uio_offset; 2907 dnp->n_flag |= NEOFVALID; 2908 } 2909 nfsmout: 2910 if (newvp != NULLVP) { 2911 if(newvp == vp) 2912 vrele(newvp); 2913 else 2914 vput(newvp); 2915 } 2916 return (error); 2917 } 2918 #endif 2919 2920 /* 2921 * Silly rename. To make the NFS filesystem that is stateless look a little 2922 * more like the "ufs" a remove of an active vnode is translated to a rename 2923 * to a funny looking filename that is removed by nfs_inactive on the 2924 * nfsnode. There is the potential for another process on a different client 2925 * to create the same funny name between the nfs_lookitup() fails and the 2926 * nfs_rename() completes, but... 2927 */ 2928 int 2929 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2930 { 2931 struct sillyrename *sp; 2932 struct nfsnode *np; 2933 int error; 2934 pid_t pid; 2935 2936 cache_purge(dvp); 2937 np = VTONFS(vp); 2938 #ifndef DIAGNOSTIC 2939 if (vp->v_type == VDIR) 2940 panic("nfs: sillyrename dir"); 2941 #endif 2942 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2943 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2944 sp->s_dvp = dvp; 2945 VREF(dvp); 2946 2947 /* Fudge together a funny name */ 2948 pid = curlwp->l_proc->p_pid; 2949 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2950 sp->s_namlen = 12; 2951 sp->s_name[8] = hexdigits[pid & 0xf]; 2952 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2953 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2954 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2955 2956 /* Try lookitups until we get one that isn't there */ 2957 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2958 curlwp, (struct nfsnode **)0) == 0) { 2959 sp->s_name[4]++; 2960 if (sp->s_name[4] > 'z') { 2961 error = EINVAL; 2962 goto bad; 2963 } 2964 } 2965 if (dolink) { 2966 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2967 sp->s_cred, curlwp); 2968 /* 2969 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2970 */ 2971 if (error == ENOTSUP) { 2972 error = nfs_renameit(dvp, cnp, sp); 2973 } 2974 } else { 2975 error = nfs_renameit(dvp, cnp, sp); 2976 } 2977 if (error) 2978 goto bad; 2979 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2980 curlwp, &np); 2981 np->n_sillyrename = sp; 2982 return (0); 2983 bad: 2984 vrele(sp->s_dvp); 2985 kauth_cred_free(sp->s_cred); 2986 kmem_free(sp, sizeof(*sp)); 2987 return (error); 2988 } 2989 2990 /* 2991 * Look up a file name and optionally either update the file handle or 2992 * allocate an nfsnode, depending on the value of npp. 2993 * npp == NULL --> just do the lookup 2994 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2995 * handled too 2996 * *npp != NULL --> update the file handle in the vnode 2997 */ 2998 int 2999 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 3000 { 3001 u_int32_t *tl; 3002 char *cp; 3003 int32_t t1, t2; 3004 struct vnode *newvp = (struct vnode *)0; 3005 struct nfsnode *np, *dnp = VTONFS(dvp); 3006 char *bpos, *dpos, *cp2; 3007 int error = 0, fhlen; 3008 #ifndef NFS_V2_ONLY 3009 int attrflag; 3010 #endif 3011 struct mbuf *mreq, *mrep, *md, *mb; 3012 nfsfh_t *nfhp; 3013 const int v3 = NFS_ISV3(dvp); 3014 3015 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 3016 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 3017 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 3018 nfsm_fhtom(dnp, v3); 3019 nfsm_strtom(name, len, NFS_MAXNAMLEN); 3020 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 3021 if (npp && !error) { 3022 nfsm_getfh(nfhp, fhlen, v3); 3023 if (*npp) { 3024 np = *npp; 3025 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3026 kmem_free(np->n_fhp, np->n_fhsize); 3027 np->n_fhp = &np->n_fh; 3028 } 3029 #if NFS_SMALLFH < NFSX_V3FHMAX 3030 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 3031 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 3032 #endif 3033 memcpy(np->n_fhp, nfhp, fhlen); 3034 np->n_fhsize = fhlen; 3035 newvp = NFSTOV(np); 3036 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3037 VREF(dvp); 3038 newvp = dvp; 3039 np = dnp; 3040 } else { 3041 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 3042 if (error) { 3043 m_freem(mrep); 3044 return (error); 3045 } 3046 newvp = NFSTOV(np); 3047 } 3048 #ifndef NFS_V2_ONLY 3049 if (v3) { 3050 nfsm_postop_attr(newvp, attrflag, 0); 3051 if (!attrflag && *npp == NULL) { 3052 m_freem(mrep); 3053 vput(newvp); 3054 return (ENOENT); 3055 } 3056 } else 3057 #endif 3058 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3059 } 3060 nfsm_reqdone; 3061 if (npp && *npp == NULL) { 3062 if (error) { 3063 if (newvp) 3064 vput(newvp); 3065 } else 3066 *npp = np; 3067 } 3068 return (error); 3069 } 3070 3071 #ifndef NFS_V2_ONLY 3072 /* 3073 * Nfs Version 3 commit rpc 3074 */ 3075 int 3076 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3077 { 3078 char *cp; 3079 u_int32_t *tl; 3080 int32_t t1, t2; 3081 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3082 char *bpos, *dpos, *cp2; 3083 int error = 0, wccflag = NFSV3_WCCRATTR; 3084 struct mbuf *mreq, *mrep, *md, *mb; 3085 struct nfsnode *np; 3086 3087 KASSERT(NFS_ISV3(vp)); 3088 3089 #ifdef NFS_DEBUG_COMMIT 3090 printf("commit %lu - %lu\n", (unsigned long)offset, 3091 (unsigned long)(offset + cnt)); 3092 #endif 3093 3094 mutex_enter(&nmp->nm_lock); 3095 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3096 mutex_exit(&nmp->nm_lock); 3097 return (0); 3098 } 3099 mutex_exit(&nmp->nm_lock); 3100 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3101 np = VTONFS(vp); 3102 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3103 nfsm_fhtom(np, 1); 3104 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3105 txdr_hyper(offset, tl); 3106 tl += 2; 3107 *tl = txdr_unsigned(cnt); 3108 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3109 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3110 if (!error) { 3111 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3112 mutex_enter(&nmp->nm_lock); 3113 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3114 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3115 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3116 error = NFSERR_STALEWRITEVERF; 3117 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3118 } 3119 mutex_exit(&nmp->nm_lock); 3120 } 3121 nfsm_reqdone; 3122 return (error); 3123 } 3124 #endif 3125 3126 /* 3127 * Kludge City.. 3128 * - make nfs_bmap() essentially a no-op that does no translation 3129 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3130 * (Maybe I could use the process's page mapping, but I was concerned that 3131 * Kernel Write might not be enabled and also figured copyout() would do 3132 * a lot more work than memcpy() and also it currently happens in the 3133 * context of the swapper process (2). 3134 */ 3135 int 3136 nfs_bmap(void *v) 3137 { 3138 struct vop_bmap_args /* { 3139 struct vnode *a_vp; 3140 daddr_t a_bn; 3141 struct vnode **a_vpp; 3142 daddr_t *a_bnp; 3143 int *a_runp; 3144 } */ *ap = v; 3145 struct vnode *vp = ap->a_vp; 3146 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3147 3148 if (ap->a_vpp != NULL) 3149 *ap->a_vpp = vp; 3150 if (ap->a_bnp != NULL) 3151 *ap->a_bnp = ap->a_bn << bshift; 3152 if (ap->a_runp != NULL) 3153 *ap->a_runp = 1024 * 1024; /* XXX */ 3154 return (0); 3155 } 3156 3157 /* 3158 * Strategy routine. 3159 * For async requests when nfsiod(s) are running, queue the request by 3160 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3161 * request. 3162 */ 3163 int 3164 nfs_strategy(void *v) 3165 { 3166 struct vop_strategy_args *ap = v; 3167 struct buf *bp = ap->a_bp; 3168 int error = 0; 3169 3170 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3171 panic("nfs physio/async"); 3172 3173 /* 3174 * If the op is asynchronous and an i/o daemon is waiting 3175 * queue the request, wake it up and wait for completion 3176 * otherwise just do it ourselves. 3177 */ 3178 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3179 error = nfs_doio(bp); 3180 return (error); 3181 } 3182 3183 /* 3184 * fsync vnode op. Just call nfs_flush() with commit == 1. 3185 */ 3186 /* ARGSUSED */ 3187 int 3188 nfs_fsync(void *v) 3189 { 3190 struct vop_fsync_args /* { 3191 struct vnodeop_desc *a_desc; 3192 struct vnode * a_vp; 3193 kauth_cred_t a_cred; 3194 int a_flags; 3195 off_t offlo; 3196 off_t offhi; 3197 struct lwp * a_l; 3198 } */ *ap = v; 3199 3200 struct vnode *vp = ap->a_vp; 3201 3202 if (vp->v_type != VREG) 3203 return 0; 3204 3205 return (nfs_flush(vp, ap->a_cred, 3206 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3207 } 3208 3209 /* 3210 * Flush all the data associated with a vnode. 3211 */ 3212 int 3213 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3214 int commit) 3215 { 3216 struct nfsnode *np = VTONFS(vp); 3217 int error; 3218 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3219 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3220 3221 mutex_enter(&vp->v_interlock); 3222 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3223 if (np->n_flag & NWRITEERR) { 3224 error = np->n_error; 3225 np->n_flag &= ~NWRITEERR; 3226 } 3227 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3228 return (error); 3229 } 3230 3231 /* 3232 * Return POSIX pathconf information applicable to nfs. 3233 * 3234 * N.B. The NFS V2 protocol doesn't support this RPC. 3235 */ 3236 /* ARGSUSED */ 3237 int 3238 nfs_pathconf(void *v) 3239 { 3240 struct vop_pathconf_args /* { 3241 struct vnode *a_vp; 3242 int a_name; 3243 register_t *a_retval; 3244 } */ *ap = v; 3245 struct nfsv3_pathconf *pcp; 3246 struct vnode *vp = ap->a_vp; 3247 struct mbuf *mreq, *mrep, *md, *mb; 3248 int32_t t1, t2; 3249 u_int32_t *tl; 3250 char *bpos, *dpos, *cp, *cp2; 3251 int error = 0, attrflag; 3252 #ifndef NFS_V2_ONLY 3253 struct nfsmount *nmp; 3254 unsigned int l; 3255 u_int64_t maxsize; 3256 #endif 3257 const int v3 = NFS_ISV3(vp); 3258 struct nfsnode *np = VTONFS(vp); 3259 3260 switch (ap->a_name) { 3261 /* Names that can be resolved locally. */ 3262 case _PC_PIPE_BUF: 3263 *ap->a_retval = PIPE_BUF; 3264 break; 3265 case _PC_SYNC_IO: 3266 *ap->a_retval = 1; 3267 break; 3268 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3269 case _PC_LINK_MAX: 3270 case _PC_NAME_MAX: 3271 case _PC_CHOWN_RESTRICTED: 3272 case _PC_NO_TRUNC: 3273 if (!v3) { 3274 error = EINVAL; 3275 break; 3276 } 3277 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3278 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3279 nfsm_fhtom(np, 1); 3280 nfsm_request(np, NFSPROC_PATHCONF, 3281 curlwp, curlwp->l_cred); /* XXX */ 3282 nfsm_postop_attr(vp, attrflag, 0); 3283 if (!error) { 3284 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3285 NFSX_V3PATHCONF); 3286 switch (ap->a_name) { 3287 case _PC_LINK_MAX: 3288 *ap->a_retval = 3289 fxdr_unsigned(register_t, pcp->pc_linkmax); 3290 break; 3291 case _PC_NAME_MAX: 3292 *ap->a_retval = 3293 fxdr_unsigned(register_t, pcp->pc_namemax); 3294 break; 3295 case _PC_CHOWN_RESTRICTED: 3296 *ap->a_retval = 3297 (pcp->pc_chownrestricted == nfs_true); 3298 break; 3299 case _PC_NO_TRUNC: 3300 *ap->a_retval = 3301 (pcp->pc_notrunc == nfs_true); 3302 break; 3303 } 3304 } 3305 nfsm_reqdone; 3306 break; 3307 case _PC_FILESIZEBITS: 3308 #ifndef NFS_V2_ONLY 3309 if (v3) { 3310 nmp = VFSTONFS(vp->v_mount); 3311 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3312 if ((error = nfs_fsinfo(nmp, vp, 3313 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3314 break; 3315 for (l = 0, maxsize = nmp->nm_maxfilesize; 3316 (maxsize >> l) > 0; l++) 3317 ; 3318 *ap->a_retval = l + 1; 3319 } else 3320 #endif 3321 { 3322 *ap->a_retval = 32; /* NFS V2 limitation */ 3323 } 3324 break; 3325 default: 3326 error = EINVAL; 3327 break; 3328 } 3329 3330 return (error); 3331 } 3332 3333 /* 3334 * NFS advisory byte-level locks. 3335 */ 3336 int 3337 nfs_advlock(void *v) 3338 { 3339 struct vop_advlock_args /* { 3340 struct vnode *a_vp; 3341 void *a_id; 3342 int a_op; 3343 struct flock *a_fl; 3344 int a_flags; 3345 } */ *ap = v; 3346 struct nfsnode *np = VTONFS(ap->a_vp); 3347 3348 return lf_advlock(ap, &np->n_lockf, np->n_size); 3349 } 3350 3351 /* 3352 * Print out the contents of an nfsnode. 3353 */ 3354 int 3355 nfs_print(void *v) 3356 { 3357 struct vop_print_args /* { 3358 struct vnode *a_vp; 3359 } */ *ap = v; 3360 struct vnode *vp = ap->a_vp; 3361 struct nfsnode *np = VTONFS(vp); 3362 3363 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3364 (unsigned long long)np->n_vattr->va_fileid, 3365 (unsigned long long)np->n_vattr->va_fsid); 3366 if (vp->v_type == VFIFO) 3367 fifo_printinfo(vp); 3368 printf("\n"); 3369 return (0); 3370 } 3371 3372 /* 3373 * nfs unlock wrapper. 3374 */ 3375 int 3376 nfs_unlock(void *v) 3377 { 3378 struct vop_unlock_args /* { 3379 struct vnode *a_vp; 3380 int a_flags; 3381 } */ *ap = v; 3382 struct vnode *vp = ap->a_vp; 3383 3384 /* 3385 * VOP_UNLOCK can be called by nfs_loadattrcache 3386 * with v_data == 0. 3387 */ 3388 if (VTONFS(vp)) { 3389 nfs_delayedtruncate(vp); 3390 } 3391 3392 return genfs_unlock(v); 3393 } 3394 3395 /* 3396 * nfs special file access vnode op. 3397 * Essentially just get vattr and then imitate iaccess() since the device is 3398 * local to the client. 3399 */ 3400 int 3401 nfsspec_access(void *v) 3402 { 3403 struct vop_access_args /* { 3404 struct vnode *a_vp; 3405 int a_mode; 3406 kauth_cred_t a_cred; 3407 struct lwp *a_l; 3408 } */ *ap = v; 3409 struct vattr va; 3410 struct vnode *vp = ap->a_vp; 3411 int error; 3412 3413 error = VOP_GETATTR(vp, &va, ap->a_cred); 3414 if (error) 3415 return (error); 3416 3417 /* 3418 * Disallow write attempts on filesystems mounted read-only; 3419 * unless the file is a socket, fifo, or a block or character 3420 * device resident on the filesystem. 3421 */ 3422 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3423 switch (vp->v_type) { 3424 case VREG: 3425 case VDIR: 3426 case VLNK: 3427 return (EROFS); 3428 default: 3429 break; 3430 } 3431 } 3432 3433 return (genfs_can_access(va.va_type, va.va_mode, 3434 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3435 } 3436 3437 /* 3438 * Read wrapper for special devices. 3439 */ 3440 int 3441 nfsspec_read(void *v) 3442 { 3443 struct vop_read_args /* { 3444 struct vnode *a_vp; 3445 struct uio *a_uio; 3446 int a_ioflag; 3447 kauth_cred_t a_cred; 3448 } */ *ap = v; 3449 struct nfsnode *np = VTONFS(ap->a_vp); 3450 3451 /* 3452 * Set access flag. 3453 */ 3454 np->n_flag |= NACC; 3455 getnanotime(&np->n_atim); 3456 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3457 } 3458 3459 /* 3460 * Write wrapper for special devices. 3461 */ 3462 int 3463 nfsspec_write(void *v) 3464 { 3465 struct vop_write_args /* { 3466 struct vnode *a_vp; 3467 struct uio *a_uio; 3468 int a_ioflag; 3469 kauth_cred_t a_cred; 3470 } */ *ap = v; 3471 struct nfsnode *np = VTONFS(ap->a_vp); 3472 3473 /* 3474 * Set update flag. 3475 */ 3476 np->n_flag |= NUPD; 3477 getnanotime(&np->n_mtim); 3478 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3479 } 3480 3481 /* 3482 * Close wrapper for special devices. 3483 * 3484 * Update the times on the nfsnode then do device close. 3485 */ 3486 int 3487 nfsspec_close(void *v) 3488 { 3489 struct vop_close_args /* { 3490 struct vnode *a_vp; 3491 int a_fflag; 3492 kauth_cred_t a_cred; 3493 struct lwp *a_l; 3494 } */ *ap = v; 3495 struct vnode *vp = ap->a_vp; 3496 struct nfsnode *np = VTONFS(vp); 3497 struct vattr vattr; 3498 3499 if (np->n_flag & (NACC | NUPD)) { 3500 np->n_flag |= NCHG; 3501 if (vp->v_usecount == 1 && 3502 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3503 VATTR_NULL(&vattr); 3504 if (np->n_flag & NACC) 3505 vattr.va_atime = np->n_atim; 3506 if (np->n_flag & NUPD) 3507 vattr.va_mtime = np->n_mtim; 3508 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3509 } 3510 } 3511 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3512 } 3513 3514 /* 3515 * Read wrapper for fifos. 3516 */ 3517 int 3518 nfsfifo_read(void *v) 3519 { 3520 struct vop_read_args /* { 3521 struct vnode *a_vp; 3522 struct uio *a_uio; 3523 int a_ioflag; 3524 kauth_cred_t a_cred; 3525 } */ *ap = v; 3526 struct nfsnode *np = VTONFS(ap->a_vp); 3527 3528 /* 3529 * Set access flag. 3530 */ 3531 np->n_flag |= NACC; 3532 getnanotime(&np->n_atim); 3533 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3534 } 3535 3536 /* 3537 * Write wrapper for fifos. 3538 */ 3539 int 3540 nfsfifo_write(void *v) 3541 { 3542 struct vop_write_args /* { 3543 struct vnode *a_vp; 3544 struct uio *a_uio; 3545 int a_ioflag; 3546 kauth_cred_t a_cred; 3547 } */ *ap = v; 3548 struct nfsnode *np = VTONFS(ap->a_vp); 3549 3550 /* 3551 * Set update flag. 3552 */ 3553 np->n_flag |= NUPD; 3554 getnanotime(&np->n_mtim); 3555 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3556 } 3557 3558 /* 3559 * Close wrapper for fifos. 3560 * 3561 * Update the times on the nfsnode then do fifo close. 3562 */ 3563 int 3564 nfsfifo_close(void *v) 3565 { 3566 struct vop_close_args /* { 3567 struct vnode *a_vp; 3568 int a_fflag; 3569 kauth_cred_t a_cred; 3570 struct lwp *a_l; 3571 } */ *ap = v; 3572 struct vnode *vp = ap->a_vp; 3573 struct nfsnode *np = VTONFS(vp); 3574 struct vattr vattr; 3575 3576 if (np->n_flag & (NACC | NUPD)) { 3577 struct timespec ts; 3578 3579 getnanotime(&ts); 3580 if (np->n_flag & NACC) 3581 np->n_atim = ts; 3582 if (np->n_flag & NUPD) 3583 np->n_mtim = ts; 3584 np->n_flag |= NCHG; 3585 if (vp->v_usecount == 1 && 3586 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3587 VATTR_NULL(&vattr); 3588 if (np->n_flag & NACC) 3589 vattr.va_atime = np->n_atim; 3590 if (np->n_flag & NUPD) 3591 vattr.va_mtime = np->n_mtim; 3592 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3593 } 3594 } 3595 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3596 } 3597
Indexes created Tue Oct 14 15:09:51 GMT 2025