nfs_vnops.c revision 1.262
1 /* $NetBSD: nfs_vnops.c,v 1.262 2007/12/17 16:04:31 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 1.262 2007/12/17 16:04:31 yamt Exp $"); 43 44 #include "opt_inet.h" 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 48 #include <sys/param.h> 49 #include <sys/proc.h> 50 #include <sys/kernel.h> 51 #include <sys/systm.h> 52 #include <sys/resourcevar.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/condvar.h> 56 #include <sys/disk.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.h> 59 #include <sys/mutex.h> 60 #include <sys/namei.h> 61 #include <sys/vnode.h> 62 #include <sys/dirent.h> 63 #include <sys/fcntl.h> 64 #include <sys/hash.h> 65 #include <sys/lockf.h> 66 #include <sys/stat.h> 67 #include <sys/unistd.h> 68 #include <sys/kauth.h> 69 70 #include <uvm/uvm_extern.h> 71 #include <uvm/uvm.h> 72 73 #include <miscfs/fifofs/fifo.h> 74 #include <miscfs/genfs/genfs.h> 75 #include <miscfs/genfs/genfs_node.h> 76 #include <miscfs/specfs/specdev.h> 77 78 #include <nfs/rpcv2.h> 79 #include <nfs/nfsproto.h> 80 #include <nfs/nfs.h> 81 #include <nfs/nfsnode.h> 82 #include <nfs/nfsmount.h> 83 #include <nfs/xdr_subs.h> 84 #include <nfs/nfsm_subs.h> 85 #include <nfs/nfs_var.h> 86 87 #include <net/if.h> 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 91 /* 92 * Global vfs data structures for nfs 93 */ 94 int (**nfsv2_vnodeop_p) __P((void *)); 95 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 96 { &vop_default_desc, vn_default_error }, 97 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 98 { &vop_create_desc, nfs_create }, /* create */ 99 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 100 { &vop_open_desc, nfs_open }, /* open */ 101 { &vop_close_desc, nfs_close }, /* close */ 102 { &vop_access_desc, nfs_access }, /* access */ 103 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 104 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 105 { &vop_read_desc, nfs_read }, /* read */ 106 { &vop_write_desc, nfs_write }, /* write */ 107 { &vop_lease_desc, nfs_lease_check }, /* lease */ 108 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 109 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 110 { &vop_poll_desc, nfs_poll }, /* poll */ 111 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 112 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 113 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 114 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 115 { &vop_seek_desc, nfs_seek }, /* seek */ 116 { &vop_remove_desc, nfs_remove }, /* remove */ 117 { &vop_link_desc, nfs_link }, /* link */ 118 { &vop_rename_desc, nfs_rename }, /* rename */ 119 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 120 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 121 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 122 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 123 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 124 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 125 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 126 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 127 { &vop_lock_desc, nfs_lock }, /* lock */ 128 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 129 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 130 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 131 { &vop_print_desc, nfs_print }, /* print */ 132 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 133 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 134 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 135 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 136 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 137 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 138 { NULL, NULL } 139 }; 140 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 141 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 142 143 /* 144 * Special device vnode ops 145 */ 146 int (**spec_nfsv2nodeop_p) __P((void *)); 147 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 148 { &vop_default_desc, vn_default_error }, 149 { &vop_lookup_desc, spec_lookup }, /* lookup */ 150 { &vop_create_desc, spec_create }, /* create */ 151 { &vop_mknod_desc, spec_mknod }, /* mknod */ 152 { &vop_open_desc, spec_open }, /* open */ 153 { &vop_close_desc, nfsspec_close }, /* close */ 154 { &vop_access_desc, nfsspec_access }, /* access */ 155 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 156 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 157 { &vop_read_desc, nfsspec_read }, /* read */ 158 { &vop_write_desc, nfsspec_write }, /* write */ 159 { &vop_lease_desc, spec_lease_check }, /* lease */ 160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 162 { &vop_poll_desc, spec_poll }, /* poll */ 163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 164 { &vop_revoke_desc, spec_revoke }, /* revoke */ 165 { &vop_mmap_desc, spec_mmap }, /* mmap */ 166 { &vop_fsync_desc, spec_fsync }, /* fsync */ 167 { &vop_seek_desc, spec_seek }, /* seek */ 168 { &vop_remove_desc, spec_remove }, /* remove */ 169 { &vop_link_desc, spec_link }, /* link */ 170 { &vop_rename_desc, spec_rename }, /* rename */ 171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 173 { &vop_symlink_desc, spec_symlink }, /* symlink */ 174 { &vop_readdir_desc, spec_readdir }, /* readdir */ 175 { &vop_readlink_desc, spec_readlink }, /* readlink */ 176 { &vop_abortop_desc, spec_abortop }, /* abortop */ 177 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 179 { &vop_lock_desc, nfs_lock }, /* lock */ 180 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 181 { &vop_bmap_desc, spec_bmap }, /* bmap */ 182 { &vop_strategy_desc, spec_strategy }, /* strategy */ 183 { &vop_print_desc, nfs_print }, /* print */ 184 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, spec_advlock }, /* advlock */ 187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 188 { &vop_getpages_desc, spec_getpages }, /* getpages */ 189 { &vop_putpages_desc, spec_putpages }, /* putpages */ 190 { NULL, NULL } 191 }; 192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 194 195 int (**fifo_nfsv2nodeop_p) __P((void *)); 196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 197 { &vop_default_desc, vn_default_error }, 198 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 199 { &vop_create_desc, fifo_create }, /* create */ 200 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 201 { &vop_open_desc, fifo_open }, /* open */ 202 { &vop_close_desc, nfsfifo_close }, /* close */ 203 { &vop_access_desc, nfsspec_access }, /* access */ 204 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 205 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 206 { &vop_read_desc, nfsfifo_read }, /* read */ 207 { &vop_write_desc, nfsfifo_write }, /* write */ 208 { &vop_lease_desc, fifo_lease_check }, /* lease */ 209 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 210 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 211 { &vop_poll_desc, fifo_poll }, /* poll */ 212 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 213 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 214 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 215 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 216 { &vop_seek_desc, fifo_seek }, /* seek */ 217 { &vop_remove_desc, fifo_remove }, /* remove */ 218 { &vop_link_desc, fifo_link }, /* link */ 219 { &vop_rename_desc, fifo_rename }, /* rename */ 220 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 221 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 222 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 223 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 224 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 225 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 226 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 227 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 228 { &vop_lock_desc, nfs_lock }, /* lock */ 229 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 230 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 231 { &vop_strategy_desc, genfs_badop }, /* strategy */ 232 { &vop_print_desc, nfs_print }, /* print */ 233 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 234 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 235 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 236 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 237 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 238 { NULL, NULL } 239 }; 240 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 241 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 242 243 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 244 size_t, kauth_cred_t, struct lwp *); 245 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 246 247 /* 248 * Global variables 249 */ 250 extern u_int32_t nfs_true, nfs_false; 251 extern u_int32_t nfs_xdrneg1; 252 extern const nfstype nfsv3_type[9]; 253 254 int nfs_numasync = 0; 255 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 256 #define UIO_ADVANCE(uio, siz) \ 257 (void)((uio)->uio_resid -= (siz), \ 258 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 259 (uio)->uio_iov->iov_len -= (siz)) 260 261 static void nfs_cache_enter(struct vnode *, struct vnode *, 262 struct componentname *); 263 264 static void 265 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 266 struct componentname *cnp) 267 { 268 struct nfsnode *dnp = VTONFS(dvp); 269 270 if (vp != NULL) { 271 struct nfsnode *np = VTONFS(vp); 272 273 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 274 } 275 276 if (!timespecisset(&dnp->n_nctime)) 277 dnp->n_nctime = dnp->n_vattr->va_mtime; 278 279 cache_enter(dvp, vp, cnp); 280 } 281 282 /* 283 * nfs null call from vfs. 284 */ 285 int 286 nfs_null(vp, cred, l) 287 struct vnode *vp; 288 kauth_cred_t cred; 289 struct lwp *l; 290 { 291 char *bpos, *dpos; 292 int error = 0; 293 struct mbuf *mreq, *mrep, *md, *mb; 294 struct nfsnode *np = VTONFS(vp); 295 296 nfsm_reqhead(np, NFSPROC_NULL, 0); 297 nfsm_request(np, NFSPROC_NULL, l, cred); 298 nfsm_reqdone; 299 return (error); 300 } 301 302 /* 303 * nfs access vnode op. 304 * For nfs version 2, just return ok. File accesses may fail later. 305 * For nfs version 3, use the access rpc to check accessibility. If file modes 306 * are changed on the server, accesses might still fail later. 307 */ 308 int 309 nfs_access(v) 310 void *v; 311 { 312 struct vop_access_args /* { 313 struct vnode *a_vp; 314 int a_mode; 315 kauth_cred_t a_cred; 316 } */ *ap = v; 317 struct vnode *vp = ap->a_vp; 318 #ifndef NFS_V2_ONLY 319 u_int32_t *tl; 320 char *cp; 321 int32_t t1, t2; 322 char *bpos, *dpos, *cp2; 323 int error = 0, attrflag; 324 struct mbuf *mreq, *mrep, *md, *mb; 325 u_int32_t mode, rmode; 326 const int v3 = NFS_ISV3(vp); 327 #endif 328 int cachevalid; 329 struct nfsnode *np = VTONFS(vp); 330 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 331 332 cachevalid = (np->n_accstamp != -1 && 333 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 334 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 335 336 /* 337 * Check access cache first. If this request has been made for this 338 * uid shortly before, use the cached result. 339 */ 340 if (cachevalid) { 341 if (!np->n_accerror) { 342 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 343 return np->n_accerror; 344 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 345 return np->n_accerror; 346 } 347 348 #ifndef NFS_V2_ONLY 349 /* 350 * For nfs v3, do an access rpc, otherwise you are stuck emulating 351 * ufs_access() locally using the vattr. This may not be correct, 352 * since the server may apply other access criteria such as 353 * client uid-->server uid mapping that we do not know about, but 354 * this is better than just returning anything that is lying about 355 * in the cache. 356 */ 357 if (v3) { 358 nfsstats.rpccnt[NFSPROC_ACCESS]++; 359 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 360 nfsm_fhtom(np, v3); 361 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 362 if (ap->a_mode & VREAD) 363 mode = NFSV3ACCESS_READ; 364 else 365 mode = 0; 366 if (vp->v_type != VDIR) { 367 if (ap->a_mode & VWRITE) 368 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 369 if (ap->a_mode & VEXEC) 370 mode |= NFSV3ACCESS_EXECUTE; 371 } else { 372 if (ap->a_mode & VWRITE) 373 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 374 NFSV3ACCESS_DELETE); 375 if (ap->a_mode & VEXEC) 376 mode |= NFSV3ACCESS_LOOKUP; 377 } 378 *tl = txdr_unsigned(mode); 379 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 380 nfsm_postop_attr(vp, attrflag, 0); 381 if (!error) { 382 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 383 rmode = fxdr_unsigned(u_int32_t, *tl); 384 /* 385 * The NFS V3 spec does not clarify whether or not 386 * the returned access bits can be a superset of 387 * the ones requested, so... 388 */ 389 if ((rmode & mode) != mode) 390 error = EACCES; 391 } 392 nfsm_reqdone; 393 } else 394 #endif 395 return (nfsspec_access(ap)); 396 #ifndef NFS_V2_ONLY 397 /* 398 * Disallow write attempts on filesystems mounted read-only; 399 * unless the file is a socket, fifo, or a block or character 400 * device resident on the filesystem. 401 */ 402 if (!error && (ap->a_mode & VWRITE) && 403 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 404 switch (vp->v_type) { 405 case VREG: 406 case VDIR: 407 case VLNK: 408 error = EROFS; 409 default: 410 break; 411 } 412 } 413 414 if (!error || error == EACCES) { 415 /* 416 * If we got the same result as for a previous, 417 * different request, OR it in. Don't update 418 * the timestamp in that case. 419 */ 420 if (cachevalid && np->n_accstamp != -1 && 421 error == np->n_accerror) { 422 if (!error) 423 np->n_accmode |= ap->a_mode; 424 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 425 np->n_accmode = ap->a_mode; 426 } else { 427 np->n_accstamp = time_uptime; 428 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 429 np->n_accmode = ap->a_mode; 430 np->n_accerror = error; 431 } 432 } 433 434 return (error); 435 #endif 436 } 437 438 /* 439 * nfs open vnode op 440 * Check to see if the type is ok 441 * and that deletion is not in progress. 442 * For paged in text files, you will need to flush the page cache 443 * if consistency is lost. 444 */ 445 /* ARGSUSED */ 446 int 447 nfs_open(v) 448 void *v; 449 { 450 struct vop_open_args /* { 451 struct vnode *a_vp; 452 int a_mode; 453 kauth_cred_t a_cred; 454 } */ *ap = v; 455 struct vnode *vp = ap->a_vp; 456 struct nfsnode *np = VTONFS(vp); 457 int error; 458 459 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 460 return (EACCES); 461 } 462 463 if (ap->a_mode & FREAD) { 464 if (np->n_rcred != NULL) 465 kauth_cred_free(np->n_rcred); 466 np->n_rcred = ap->a_cred; 467 kauth_cred_hold(np->n_rcred); 468 } 469 if (ap->a_mode & FWRITE) { 470 if (np->n_wcred != NULL) 471 kauth_cred_free(np->n_wcred); 472 np->n_wcred = ap->a_cred; 473 kauth_cred_hold(np->n_wcred); 474 } 475 476 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 477 if (error) 478 return error; 479 480 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 481 482 return (0); 483 } 484 485 /* 486 * nfs close vnode op 487 * What an NFS client should do upon close after writing is a debatable issue. 488 * Most NFS clients push delayed writes to the server upon close, basically for 489 * two reasons: 490 * 1 - So that any write errors may be reported back to the client process 491 * doing the close system call. By far the two most likely errors are 492 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 493 * 2 - To put a worst case upper bound on cache inconsistency between 494 * multiple clients for the file. 495 * There is also a consistency problem for Version 2 of the protocol w.r.t. 496 * not being able to tell if other clients are writing a file concurrently, 497 * since there is no way of knowing if the changed modify time in the reply 498 * is only due to the write for this client. 499 * (NFS Version 3 provides weak cache consistency data in the reply that 500 * should be sufficient to detect and handle this case.) 501 * 502 * The current code does the following: 503 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 504 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 505 * or commit them (this satisfies 1 and 2 except for the 506 * case where the server crashes after this close but 507 * before the commit RPC, which is felt to be "good 508 * enough". Changing the last argument to nfs_flush() to 509 * a 1 would force a commit operation, if it is felt a 510 * commit is necessary now. 511 */ 512 /* ARGSUSED */ 513 int 514 nfs_close(v) 515 void *v; 516 { 517 struct vop_close_args /* { 518 struct vnodeop_desc *a_desc; 519 struct vnode *a_vp; 520 int a_fflag; 521 kauth_cred_t a_cred; 522 } */ *ap = v; 523 struct vnode *vp = ap->a_vp; 524 struct nfsnode *np = VTONFS(vp); 525 int error = 0; 526 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 527 528 if (vp->v_type == VREG) { 529 if (np->n_flag & NMODIFIED) { 530 #ifndef NFS_V2_ONLY 531 if (NFS_ISV3(vp)) { 532 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 533 np->n_flag &= ~NMODIFIED; 534 } else 535 #endif 536 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 537 NFS_INVALIDATE_ATTRCACHE(np); 538 } 539 if (np->n_flag & NWRITEERR) { 540 np->n_flag &= ~NWRITEERR; 541 error = np->n_error; 542 } 543 } 544 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 545 return (error); 546 } 547 548 /* 549 * nfs getattr call from vfs. 550 */ 551 int 552 nfs_getattr(v) 553 void *v; 554 { 555 struct vop_getattr_args /* { 556 struct vnode *a_vp; 557 struct vattr *a_vap; 558 kauth_cred_t a_cred; 559 } */ *ap = v; 560 struct vnode *vp = ap->a_vp; 561 struct nfsnode *np = VTONFS(vp); 562 char *cp; 563 u_int32_t *tl; 564 int32_t t1, t2; 565 char *bpos, *dpos; 566 int error = 0; 567 struct mbuf *mreq, *mrep, *md, *mb; 568 const int v3 = NFS_ISV3(vp); 569 570 /* 571 * Update local times for special files. 572 */ 573 if (np->n_flag & (NACC | NUPD)) 574 np->n_flag |= NCHG; 575 576 /* 577 * if we have delayed truncation, do it now. 578 */ 579 nfs_delayedtruncate(vp); 580 581 /* 582 * First look in the cache. 583 */ 584 if (nfs_getattrcache(vp, ap->a_vap) == 0) 585 return (0); 586 nfsstats.rpccnt[NFSPROC_GETATTR]++; 587 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 588 nfsm_fhtom(np, v3); 589 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 590 if (!error) { 591 nfsm_loadattr(vp, ap->a_vap, 0); 592 if (vp->v_type == VDIR && 593 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 594 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 595 } 596 nfsm_reqdone; 597 return (error); 598 } 599 600 /* 601 * nfs setattr call. 602 */ 603 int 604 nfs_setattr(v) 605 void *v; 606 { 607 struct vop_setattr_args /* { 608 struct vnodeop_desc *a_desc; 609 struct vnode *a_vp; 610 struct vattr *a_vap; 611 kauth_cred_t a_cred; 612 } */ *ap = v; 613 struct vnode *vp = ap->a_vp; 614 struct nfsnode *np = VTONFS(vp); 615 struct vattr *vap = ap->a_vap; 616 int error = 0; 617 u_quad_t tsize = 0; 618 619 /* 620 * Setting of flags is not supported. 621 */ 622 if (vap->va_flags != VNOVAL) 623 return (EOPNOTSUPP); 624 625 /* 626 * Disallow write attempts if the filesystem is mounted read-only. 627 */ 628 if ((vap->va_uid != (uid_t)VNOVAL || 629 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 630 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 631 (vp->v_mount->mnt_flag & MNT_RDONLY)) 632 return (EROFS); 633 if (vap->va_size != VNOVAL) { 634 switch (vp->v_type) { 635 case VDIR: 636 return (EISDIR); 637 case VCHR: 638 case VBLK: 639 case VSOCK: 640 case VFIFO: 641 if (vap->va_mtime.tv_sec == VNOVAL && 642 vap->va_atime.tv_sec == VNOVAL && 643 vap->va_mode == (mode_t)VNOVAL && 644 vap->va_uid == (uid_t)VNOVAL && 645 vap->va_gid == (gid_t)VNOVAL) 646 return (0); 647 vap->va_size = VNOVAL; 648 break; 649 default: 650 /* 651 * Disallow write attempts if the filesystem is 652 * mounted read-only. 653 */ 654 if (vp->v_mount->mnt_flag & MNT_RDONLY) 655 return (EROFS); 656 genfs_node_wrlock(vp); 657 uvm_vnp_setsize(vp, vap->va_size); 658 tsize = np->n_size; 659 np->n_size = vap->va_size; 660 if (vap->va_size == 0) 661 error = nfs_vinvalbuf(vp, 0, 662 ap->a_cred, curlwp, 1); 663 else 664 error = nfs_vinvalbuf(vp, V_SAVE, 665 ap->a_cred, curlwp, 1); 666 if (error) { 667 uvm_vnp_setsize(vp, tsize); 668 genfs_node_unlock(vp); 669 return (error); 670 } 671 np->n_vattr->va_size = vap->va_size; 672 } 673 } else { 674 /* 675 * flush files before setattr because a later write of 676 * cached data might change timestamps or reset sugid bits 677 */ 678 if ((vap->va_mtime.tv_sec != VNOVAL || 679 vap->va_atime.tv_sec != VNOVAL || 680 vap->va_mode != VNOVAL) && 681 vp->v_type == VREG && 682 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 683 curlwp, 1)) == EINTR) 684 return (error); 685 } 686 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 687 if (vap->va_size != VNOVAL) { 688 if (error) { 689 np->n_size = np->n_vattr->va_size = tsize; 690 uvm_vnp_setsize(vp, np->n_size); 691 } 692 genfs_node_unlock(vp); 693 } 694 VN_KNOTE(vp, NOTE_ATTRIB); 695 return (error); 696 } 697 698 /* 699 * Do an nfs setattr rpc. 700 */ 701 int 702 nfs_setattrrpc(vp, vap, cred, l) 703 struct vnode *vp; 704 struct vattr *vap; 705 kauth_cred_t cred; 706 struct lwp *l; 707 { 708 struct nfsv2_sattr *sp; 709 char *cp; 710 int32_t t1, t2; 711 char *bpos, *dpos; 712 u_int32_t *tl; 713 int error = 0; 714 struct mbuf *mreq, *mrep, *md, *mb; 715 const int v3 = NFS_ISV3(vp); 716 struct nfsnode *np = VTONFS(vp); 717 #ifndef NFS_V2_ONLY 718 int wccflag = NFSV3_WCCRATTR; 719 char *cp2; 720 #endif 721 722 nfsstats.rpccnt[NFSPROC_SETATTR]++; 723 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 724 nfsm_fhtom(np, v3); 725 #ifndef NFS_V2_ONLY 726 if (v3) { 727 nfsm_v3attrbuild(vap, true); 728 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 729 *tl = nfs_false; 730 } else { 731 #endif 732 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 733 if (vap->va_mode == (mode_t)VNOVAL) 734 sp->sa_mode = nfs_xdrneg1; 735 else 736 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 737 if (vap->va_uid == (uid_t)VNOVAL) 738 sp->sa_uid = nfs_xdrneg1; 739 else 740 sp->sa_uid = txdr_unsigned(vap->va_uid); 741 if (vap->va_gid == (gid_t)VNOVAL) 742 sp->sa_gid = nfs_xdrneg1; 743 else 744 sp->sa_gid = txdr_unsigned(vap->va_gid); 745 sp->sa_size = txdr_unsigned(vap->va_size); 746 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 747 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 748 #ifndef NFS_V2_ONLY 749 } 750 #endif 751 nfsm_request(np, NFSPROC_SETATTR, l, cred); 752 #ifndef NFS_V2_ONLY 753 if (v3) { 754 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 755 } else 756 #endif 757 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 758 nfsm_reqdone; 759 return (error); 760 } 761 762 /* 763 * nfs lookup call, one step at a time... 764 * First look in cache 765 * If not found, unlock the directory nfsnode and do the rpc 766 * 767 * This code is full of lock/unlock statements and checks, because 768 * we continue after cache_lookup has finished (we need to check 769 * with the attr cache and do an rpc if it has timed out). This means 770 * that the locking effects of cache_lookup have to be taken into 771 * account. 772 */ 773 int 774 nfs_lookup(v) 775 void *v; 776 { 777 struct vop_lookup_args /* { 778 struct vnodeop_desc *a_desc; 779 struct vnode *a_dvp; 780 struct vnode **a_vpp; 781 struct componentname *a_cnp; 782 } */ *ap = v; 783 struct componentname *cnp = ap->a_cnp; 784 struct vnode *dvp = ap->a_dvp; 785 struct vnode **vpp = ap->a_vpp; 786 int flags; 787 struct vnode *newvp; 788 u_int32_t *tl; 789 char *cp; 790 int32_t t1, t2; 791 char *bpos, *dpos, *cp2; 792 struct mbuf *mreq, *mrep, *md, *mb; 793 long len; 794 nfsfh_t *fhp; 795 struct nfsnode *np; 796 int error = 0, attrflag, fhsize; 797 const int v3 = NFS_ISV3(dvp); 798 799 flags = cnp->cn_flags; 800 801 *vpp = NULLVP; 802 newvp = NULLVP; 803 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 804 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 805 return (EROFS); 806 if (dvp->v_type != VDIR) 807 return (ENOTDIR); 808 809 /* 810 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 811 */ 812 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 813 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 814 if (error) 815 return error; 816 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 817 return EISDIR; 818 VREF(dvp); 819 *vpp = dvp; 820 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 821 cnp->cn_flags |= SAVENAME; 822 return 0; 823 } 824 825 np = VTONFS(dvp); 826 827 /* 828 * Before tediously performing a linear scan of the directory, 829 * check the name cache to see if the directory/name pair 830 * we are looking for is known already. 831 * If the directory/name pair is found in the name cache, 832 * we have to ensure the directory has not changed from 833 * the time the cache entry has been created. If it has, 834 * the cache entry has to be ignored. 835 */ 836 error = cache_lookup_raw(dvp, vpp, cnp); 837 KASSERT(dvp != *vpp); 838 if (error >= 0) { 839 struct vattr vattr; 840 int err2; 841 842 if (error && error != ENOENT) { 843 *vpp = NULLVP; 844 return error; 845 } 846 847 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 848 if (err2 != 0) { 849 if (error == 0) 850 vrele(*vpp); 851 *vpp = NULLVP; 852 return err2; 853 } 854 855 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 856 || timespeccmp(&vattr.va_mtime, 857 &VTONFS(dvp)->n_nctime, !=)) { 858 if (error == 0) { 859 vrele(*vpp); 860 *vpp = NULLVP; 861 } 862 cache_purge1(dvp, NULL, PURGE_CHILDREN); 863 timespecclear(&np->n_nctime); 864 goto dorpc; 865 } 866 867 if (error == ENOENT) { 868 goto noentry; 869 } 870 871 newvp = *vpp; 872 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 873 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 874 nfsstats.lookupcache_hits++; 875 if ((flags & ISDOTDOT) != 0) { 876 VOP_UNLOCK(dvp, 0); 877 } 878 error = vn_lock(newvp, LK_EXCLUSIVE); 879 if ((flags & ISDOTDOT) != 0) { 880 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 881 } 882 if (error) { 883 /* newvp has been revoked. */ 884 vrele(newvp); 885 *vpp = NULL; 886 return error; 887 } 888 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 889 cnp->cn_flags |= SAVENAME; 890 KASSERT(newvp->v_type != VNON); 891 return (0); 892 } 893 cache_purge1(newvp, NULL, PURGE_PARENTS); 894 vrele(newvp); 895 *vpp = NULLVP; 896 } 897 dorpc: 898 #if 0 899 /* 900 * because nfsv3 has the same CREATE semantics as ours, 901 * we don't have to perform LOOKUPs beforehand. 902 * 903 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 904 * XXX although we have no way to know if O_EXCL is requested or not. 905 */ 906 907 if (v3 && cnp->cn_nameiop == CREATE && 908 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 909 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 910 cnp->cn_flags |= SAVENAME; 911 return (EJUSTRETURN); 912 } 913 #endif /* 0 */ 914 915 error = 0; 916 newvp = NULLVP; 917 nfsstats.lookupcache_misses++; 918 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 919 len = cnp->cn_namelen; 920 nfsm_reqhead(np, NFSPROC_LOOKUP, 921 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 922 nfsm_fhtom(np, v3); 923 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 924 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 925 if (error) { 926 nfsm_postop_attr(dvp, attrflag, 0); 927 m_freem(mrep); 928 goto nfsmout; 929 } 930 nfsm_getfh(fhp, fhsize, v3); 931 932 /* 933 * Handle RENAME case... 934 */ 935 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 936 if (NFS_CMPFH(np, fhp, fhsize)) { 937 m_freem(mrep); 938 return (EISDIR); 939 } 940 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 941 if (error) { 942 m_freem(mrep); 943 return error; 944 } 945 newvp = NFSTOV(np); 946 #ifndef NFS_V2_ONLY 947 if (v3) { 948 nfsm_postop_attr(newvp, attrflag, 0); 949 nfsm_postop_attr(dvp, attrflag, 0); 950 } else 951 #endif 952 nfsm_loadattr(newvp, (struct vattr *)0, 0); 953 *vpp = newvp; 954 m_freem(mrep); 955 cnp->cn_flags |= SAVENAME; 956 goto validate; 957 } 958 959 /* 960 * The postop attr handling is duplicated for each if case, 961 * because it should be done while dvp is locked (unlocking 962 * dvp is different for each case). 963 */ 964 965 if (NFS_CMPFH(np, fhp, fhsize)) { 966 /* 967 * "." lookup 968 */ 969 VREF(dvp); 970 newvp = dvp; 971 #ifndef NFS_V2_ONLY 972 if (v3) { 973 nfsm_postop_attr(newvp, attrflag, 0); 974 nfsm_postop_attr(dvp, attrflag, 0); 975 } else 976 #endif 977 nfsm_loadattr(newvp, (struct vattr *)0, 0); 978 } else if (flags & ISDOTDOT) { 979 /* 980 * ".." lookup 981 */ 982 VOP_UNLOCK(dvp, 0); 983 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 984 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 985 if (error) { 986 m_freem(mrep); 987 return error; 988 } 989 newvp = NFSTOV(np); 990 991 #ifndef NFS_V2_ONLY 992 if (v3) { 993 nfsm_postop_attr(newvp, attrflag, 0); 994 nfsm_postop_attr(dvp, attrflag, 0); 995 } else 996 #endif 997 nfsm_loadattr(newvp, (struct vattr *)0, 0); 998 } else { 999 /* 1000 * Other lookups. 1001 */ 1002 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1003 if (error) { 1004 m_freem(mrep); 1005 return error; 1006 } 1007 newvp = NFSTOV(np); 1008 #ifndef NFS_V2_ONLY 1009 if (v3) { 1010 nfsm_postop_attr(newvp, attrflag, 0); 1011 nfsm_postop_attr(dvp, attrflag, 0); 1012 } else 1013 #endif 1014 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1015 } 1016 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1017 cnp->cn_flags |= SAVENAME; 1018 if ((cnp->cn_flags & MAKEENTRY) && 1019 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1020 nfs_cache_enter(dvp, newvp, cnp); 1021 } 1022 *vpp = newvp; 1023 nfsm_reqdone; 1024 if (error) { 1025 /* 1026 * We get here only because of errors returned by 1027 * the RPC. Otherwise we'll have returned above 1028 * (the nfsm_* macros will jump to nfsm_reqdone 1029 * on error). 1030 */ 1031 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1032 cnp->cn_nameiop != CREATE) { 1033 nfs_cache_enter(dvp, NULL, cnp); 1034 } 1035 if (newvp != NULLVP) { 1036 if (newvp == dvp) { 1037 vrele(newvp); 1038 } else { 1039 vput(newvp); 1040 } 1041 } 1042 noentry: 1043 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1044 (flags & ISLASTCN) && error == ENOENT) { 1045 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1046 error = EROFS; 1047 } else { 1048 error = EJUSTRETURN; 1049 cnp->cn_flags |= SAVENAME; 1050 } 1051 } 1052 *vpp = NULL; 1053 return error; 1054 } 1055 1056 validate: 1057 /* 1058 * make sure we have valid type and size. 1059 */ 1060 1061 newvp = *vpp; 1062 if (newvp->v_type == VNON) { 1063 struct vattr vattr; /* dummy */ 1064 1065 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1066 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1067 if (error) { 1068 vput(newvp); 1069 *vpp = NULL; 1070 } 1071 } 1072 1073 return error; 1074 } 1075 1076 /* 1077 * nfs read call. 1078 * Just call nfs_bioread() to do the work. 1079 */ 1080 int 1081 nfs_read(v) 1082 void *v; 1083 { 1084 struct vop_read_args /* { 1085 struct vnode *a_vp; 1086 struct uio *a_uio; 1087 int a_ioflag; 1088 kauth_cred_t a_cred; 1089 } */ *ap = v; 1090 struct vnode *vp = ap->a_vp; 1091 1092 if (vp->v_type != VREG) 1093 return EISDIR; 1094 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1095 } 1096 1097 /* 1098 * nfs readlink call 1099 */ 1100 int 1101 nfs_readlink(v) 1102 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(vp, uiop, cred) 1130 struct vnode *vp; 1131 struct uio *uiop; 1132 kauth_cred_t cred; 1133 { 1134 u_int32_t *tl; 1135 char *cp; 1136 int32_t t1, t2; 1137 char *bpos, *dpos, *cp2; 1138 int error = 0; 1139 uint32_t len; 1140 struct mbuf *mreq, *mrep, *md, *mb; 1141 const int v3 = NFS_ISV3(vp); 1142 struct nfsnode *np = VTONFS(vp); 1143 #ifndef NFS_V2_ONLY 1144 int attrflag; 1145 #endif 1146 1147 nfsstats.rpccnt[NFSPROC_READLINK]++; 1148 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1149 nfsm_fhtom(np, v3); 1150 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1151 #ifndef NFS_V2_ONLY 1152 if (v3) 1153 nfsm_postop_attr(vp, attrflag, 0); 1154 #endif 1155 if (!error) { 1156 #ifndef NFS_V2_ONLY 1157 if (v3) { 1158 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1159 len = fxdr_unsigned(uint32_t, *tl); 1160 if (len > MAXPATHLEN) { 1161 /* 1162 * this pathname is too long for us. 1163 */ 1164 m_freem(mrep); 1165 /* Solaris returns EINVAL. should we follow? */ 1166 error = ENAMETOOLONG; 1167 goto nfsmout; 1168 } 1169 } else 1170 #endif 1171 { 1172 nfsm_strsiz(len, NFS_MAXPATHLEN); 1173 } 1174 nfsm_mtouio(uiop, len); 1175 } 1176 nfsm_reqdone; 1177 return (error); 1178 } 1179 1180 /* 1181 * nfs read rpc call 1182 * Ditto above 1183 */ 1184 int 1185 nfs_readrpc(vp, uiop) 1186 struct vnode *vp; 1187 struct uio *uiop; 1188 { 1189 u_int32_t *tl; 1190 char *cp; 1191 int32_t t1, t2; 1192 char *bpos, *dpos, *cp2; 1193 struct mbuf *mreq, *mrep, *md, *mb; 1194 struct nfsmount *nmp; 1195 int error = 0, len, retlen, tsiz, eof, byte_count; 1196 const int v3 = NFS_ISV3(vp); 1197 struct nfsnode *np = VTONFS(vp); 1198 #ifndef NFS_V2_ONLY 1199 int attrflag; 1200 #endif 1201 1202 #ifndef nolint 1203 eof = 0; 1204 #endif 1205 nmp = VFSTONFS(vp->v_mount); 1206 tsiz = uiop->uio_resid; 1207 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1208 return (EFBIG); 1209 iostat_busy(nmp->nm_stats); 1210 byte_count = 0; /* count bytes actually transferred */ 1211 while (tsiz > 0) { 1212 nfsstats.rpccnt[NFSPROC_READ]++; 1213 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1214 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1215 nfsm_fhtom(np, v3); 1216 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1217 #ifndef NFS_V2_ONLY 1218 if (v3) { 1219 txdr_hyper(uiop->uio_offset, tl); 1220 *(tl + 2) = txdr_unsigned(len); 1221 } else 1222 #endif 1223 { 1224 *tl++ = txdr_unsigned(uiop->uio_offset); 1225 *tl++ = txdr_unsigned(len); 1226 *tl = 0; 1227 } 1228 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1229 #ifndef NFS_V2_ONLY 1230 if (v3) { 1231 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1232 if (error) { 1233 m_freem(mrep); 1234 goto nfsmout; 1235 } 1236 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1237 eof = fxdr_unsigned(int, *(tl + 1)); 1238 } else 1239 #endif 1240 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1241 nfsm_strsiz(retlen, nmp->nm_rsize); 1242 nfsm_mtouio(uiop, retlen); 1243 m_freem(mrep); 1244 tsiz -= retlen; 1245 byte_count += retlen; 1246 #ifndef NFS_V2_ONLY 1247 if (v3) { 1248 if (eof || retlen == 0) 1249 tsiz = 0; 1250 } else 1251 #endif 1252 if (retlen < len) 1253 tsiz = 0; 1254 } 1255 nfsmout: 1256 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1257 return (error); 1258 } 1259 1260 struct nfs_writerpc_context { 1261 kmutex_t nwc_lock; 1262 kcondvar_t nwc_cv; 1263 int nwc_mbufcount; 1264 }; 1265 1266 /* 1267 * free mbuf used to refer protected pages while write rpc call. 1268 * called at splvm. 1269 */ 1270 static void 1271 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1272 { 1273 struct nfs_writerpc_context *ctx = arg; 1274 1275 KASSERT(m != NULL); 1276 KASSERT(ctx != NULL); 1277 pool_cache_put(mb_cache, m); 1278 mutex_enter(&ctx->nwc_lock); 1279 if (--ctx->nwc_mbufcount == 0) { 1280 cv_signal(&ctx->nwc_cv); 1281 } 1282 mutex_exit(&ctx->nwc_lock); 1283 } 1284 1285 /* 1286 * nfs write call 1287 */ 1288 int 1289 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1290 struct vnode *vp; 1291 struct uio *uiop; 1292 int *iomode; 1293 bool pageprotected; 1294 bool *stalewriteverfp; 1295 { 1296 u_int32_t *tl; 1297 char *cp; 1298 int32_t t1, t2; 1299 char *bpos, *dpos; 1300 struct mbuf *mreq, *mrep, *md, *mb; 1301 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1302 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1303 const int v3 = NFS_ISV3(vp); 1304 int committed = NFSV3WRITE_FILESYNC; 1305 struct nfsnode *np = VTONFS(vp); 1306 struct nfs_writerpc_context ctx; 1307 int byte_count; 1308 struct lwp *l = NULL; 1309 size_t origresid; 1310 #ifndef NFS_V2_ONLY 1311 char *cp2; 1312 int rlen, commit; 1313 #endif 1314 1315 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1316 cv_init(&ctx.nwc_cv, "nfsmblk"); 1317 ctx.nwc_mbufcount = 1; 1318 1319 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1320 panic("writerpc readonly vp %p", vp); 1321 } 1322 1323 #ifdef DIAGNOSTIC 1324 if (uiop->uio_iovcnt != 1) 1325 panic("nfs: writerpc iovcnt > 1"); 1326 #endif 1327 tsiz = uiop->uio_resid; 1328 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1329 return (EFBIG); 1330 if (pageprotected) { 1331 l = curlwp; 1332 uvm_lwp_hold(l); 1333 } 1334 retry: 1335 origresid = uiop->uio_resid; 1336 KASSERT(origresid == uiop->uio_iov->iov_len); 1337 iostat_busy(nmp->nm_stats); 1338 byte_count = 0; /* count of bytes actually written */ 1339 while (tsiz > 0) { 1340 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1341 uint32_t backup; 1342 bool stalewriteverf = false; 1343 1344 nfsstats.rpccnt[NFSPROC_WRITE]++; 1345 len = min(tsiz, nmp->nm_wsize); 1346 datalen = pageprotected ? 0 : nfsm_rndup(len); 1347 nfsm_reqhead(np, NFSPROC_WRITE, 1348 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1349 nfsm_fhtom(np, v3); 1350 #ifndef NFS_V2_ONLY 1351 if (v3) { 1352 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1353 txdr_hyper(uiop->uio_offset, tl); 1354 tl += 2; 1355 *tl++ = txdr_unsigned(len); 1356 *tl++ = txdr_unsigned(*iomode); 1357 *tl = txdr_unsigned(len); 1358 } else 1359 #endif 1360 { 1361 u_int32_t x; 1362 1363 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1364 /* Set both "begin" and "current" to non-garbage. */ 1365 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1366 *tl++ = x; /* "begin offset" */ 1367 *tl++ = x; /* "current offset" */ 1368 x = txdr_unsigned(len); 1369 *tl++ = x; /* total to this offset */ 1370 *tl = x; /* size of this write */ 1371 1372 } 1373 if (pageprotected) { 1374 /* 1375 * since we know pages can't be modified during i/o, 1376 * no need to copy them for us. 1377 */ 1378 struct mbuf *m; 1379 struct iovec *iovp = uiop->uio_iov; 1380 1381 m = m_get(M_WAIT, MT_DATA); 1382 MCLAIM(m, &nfs_mowner); 1383 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1384 nfs_writerpc_extfree, &ctx); 1385 m->m_flags |= M_EXT_ROMAP; 1386 m->m_len = len; 1387 mb->m_next = m; 1388 /* 1389 * no need to maintain mb and bpos here 1390 * because no one care them later. 1391 */ 1392 #if 0 1393 mb = m; 1394 bpos = mtod(void *, mb) + mb->m_len; 1395 #endif 1396 UIO_ADVANCE(uiop, len); 1397 uiop->uio_offset += len; 1398 mutex_enter(&ctx.nwc_lock); 1399 ctx.nwc_mbufcount++; 1400 mutex_exit(&ctx.nwc_lock); 1401 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1402 } else { 1403 nfsm_uiotom(uiop, len); 1404 } 1405 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1406 #ifndef NFS_V2_ONLY 1407 if (v3) { 1408 wccflag = NFSV3_WCCCHK; 1409 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1410 if (!error) { 1411 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1412 + NFSX_V3WRITEVERF); 1413 rlen = fxdr_unsigned(int, *tl++); 1414 if (rlen == 0) { 1415 error = NFSERR_IO; 1416 m_freem(mrep); 1417 break; 1418 } else if (rlen < len) { 1419 backup = len - rlen; 1420 UIO_ADVANCE(uiop, -backup); 1421 uiop->uio_offset -= backup; 1422 len = rlen; 1423 } 1424 commit = fxdr_unsigned(int, *tl++); 1425 1426 /* 1427 * Return the lowest committment level 1428 * obtained by any of the RPCs. 1429 */ 1430 if (committed == NFSV3WRITE_FILESYNC) 1431 committed = commit; 1432 else if (committed == NFSV3WRITE_DATASYNC && 1433 commit == NFSV3WRITE_UNSTABLE) 1434 committed = commit; 1435 mutex_enter(&nmp->nm_lock); 1436 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1437 memcpy(nmp->nm_writeverf, tl, 1438 NFSX_V3WRITEVERF); 1439 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1440 } else if ((nmp->nm_iflag & 1441 NFSMNT_STALEWRITEVERF) || 1442 memcmp(tl, nmp->nm_writeverf, 1443 NFSX_V3WRITEVERF)) { 1444 memcpy(nmp->nm_writeverf, tl, 1445 NFSX_V3WRITEVERF); 1446 /* 1447 * note NFSMNT_STALEWRITEVERF 1448 * if we're the first thread to 1449 * notice it. 1450 */ 1451 if ((nmp->nm_iflag & 1452 NFSMNT_STALEWRITEVERF) == 0) { 1453 stalewriteverf = true; 1454 nmp->nm_iflag |= 1455 NFSMNT_STALEWRITEVERF; 1456 } 1457 } 1458 mutex_exit(&nmp->nm_lock); 1459 } 1460 } else 1461 #endif 1462 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1463 if (wccflag) 1464 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1465 m_freem(mrep); 1466 if (error) 1467 break; 1468 tsiz -= len; 1469 byte_count += len; 1470 if (stalewriteverf) { 1471 *stalewriteverfp = true; 1472 stalewriteverf = false; 1473 if (committed == NFSV3WRITE_UNSTABLE && 1474 len != origresid) { 1475 /* 1476 * if our write requests weren't atomic but 1477 * unstable, datas in previous iterations 1478 * might have already been lost now. 1479 * then, we should resend them to nfsd. 1480 */ 1481 backup = origresid - tsiz; 1482 UIO_ADVANCE(uiop, -backup); 1483 uiop->uio_offset -= backup; 1484 tsiz = origresid; 1485 goto retry; 1486 } 1487 } 1488 } 1489 nfsmout: 1490 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1491 if (pageprotected) { 1492 /* 1493 * wait until mbufs go away. 1494 * retransmitted mbufs can survive longer than rpc requests 1495 * themselves. 1496 */ 1497 mutex_enter(&ctx.nwc_lock); 1498 ctx.nwc_mbufcount--; 1499 while (ctx.nwc_mbufcount > 0) { 1500 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1501 } 1502 mutex_exit(&ctx.nwc_lock); 1503 uvm_lwp_rele(l); 1504 } 1505 mutex_destroy(&ctx.nwc_lock); 1506 cv_destroy(&ctx.nwc_cv); 1507 *iomode = committed; 1508 if (error) 1509 uiop->uio_resid = tsiz; 1510 return (error); 1511 } 1512 1513 /* 1514 * nfs mknod rpc 1515 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1516 * mode set to specify the file type and the size field for rdev. 1517 */ 1518 int 1519 nfs_mknodrpc(dvp, vpp, cnp, vap) 1520 struct vnode *dvp; 1521 struct vnode **vpp; 1522 struct componentname *cnp; 1523 struct vattr *vap; 1524 { 1525 struct nfsv2_sattr *sp; 1526 u_int32_t *tl; 1527 char *cp; 1528 int32_t t1, t2; 1529 struct vnode *newvp = (struct vnode *)0; 1530 struct nfsnode *dnp, *np; 1531 char *cp2; 1532 char *bpos, *dpos; 1533 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1534 struct mbuf *mreq, *mrep, *md, *mb; 1535 u_int32_t rdev; 1536 const int v3 = NFS_ISV3(dvp); 1537 1538 if (vap->va_type == VCHR || vap->va_type == VBLK) 1539 rdev = txdr_unsigned(vap->va_rdev); 1540 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1541 rdev = nfs_xdrneg1; 1542 else { 1543 VOP_ABORTOP(dvp, cnp); 1544 vput(dvp); 1545 return (EOPNOTSUPP); 1546 } 1547 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1548 dnp = VTONFS(dvp); 1549 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1550 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1551 nfsm_fhtom(dnp, v3); 1552 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1553 #ifndef NFS_V2_ONLY 1554 if (v3) { 1555 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1556 *tl++ = vtonfsv3_type(vap->va_type); 1557 nfsm_v3attrbuild(vap, false); 1558 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1559 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1560 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1561 *tl = txdr_unsigned(minor(vap->va_rdev)); 1562 } 1563 } else 1564 #endif 1565 { 1566 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1567 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1568 sp->sa_uid = nfs_xdrneg1; 1569 sp->sa_gid = nfs_xdrneg1; 1570 sp->sa_size = rdev; 1571 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1572 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1573 } 1574 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1575 if (!error) { 1576 nfsm_mtofh(dvp, newvp, v3, gotvp); 1577 if (!gotvp) { 1578 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1579 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1580 if (!error) 1581 newvp = NFSTOV(np); 1582 } 1583 } 1584 #ifndef NFS_V2_ONLY 1585 if (v3) 1586 nfsm_wcc_data(dvp, wccflag, 0, !error); 1587 #endif 1588 nfsm_reqdone; 1589 if (error) { 1590 if (newvp) 1591 vput(newvp); 1592 } else { 1593 if (cnp->cn_flags & MAKEENTRY) 1594 nfs_cache_enter(dvp, newvp, cnp); 1595 *vpp = newvp; 1596 } 1597 PNBUF_PUT(cnp->cn_pnbuf); 1598 VTONFS(dvp)->n_flag |= NMODIFIED; 1599 if (!wccflag) 1600 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1601 vput(dvp); 1602 return (error); 1603 } 1604 1605 /* 1606 * nfs mknod vop 1607 * just call nfs_mknodrpc() to do the work. 1608 */ 1609 /* ARGSUSED */ 1610 int 1611 nfs_mknod(v) 1612 void *v; 1613 { 1614 struct vop_mknod_args /* { 1615 struct vnode *a_dvp; 1616 struct vnode **a_vpp; 1617 struct componentname *a_cnp; 1618 struct vattr *a_vap; 1619 } */ *ap = v; 1620 struct vnode *dvp = ap->a_dvp; 1621 struct componentname *cnp = ap->a_cnp; 1622 int error; 1623 1624 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1625 VN_KNOTE(dvp, NOTE_WRITE); 1626 if (error == 0 || error == EEXIST) 1627 cache_purge1(dvp, cnp, 0); 1628 return (error); 1629 } 1630 1631 #ifndef NFS_V2_ONLY 1632 static u_long create_verf; 1633 #endif 1634 /* 1635 * nfs file create call 1636 */ 1637 int 1638 nfs_create(v) 1639 void *v; 1640 { 1641 struct vop_create_args /* { 1642 struct vnode *a_dvp; 1643 struct vnode **a_vpp; 1644 struct componentname *a_cnp; 1645 struct vattr *a_vap; 1646 } */ *ap = v; 1647 struct vnode *dvp = ap->a_dvp; 1648 struct vattr *vap = ap->a_vap; 1649 struct componentname *cnp = ap->a_cnp; 1650 struct nfsv2_sattr *sp; 1651 u_int32_t *tl; 1652 char *cp; 1653 int32_t t1, t2; 1654 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1655 struct vnode *newvp = (struct vnode *)0; 1656 char *bpos, *dpos, *cp2; 1657 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1658 struct mbuf *mreq, *mrep, *md, *mb; 1659 const int v3 = NFS_ISV3(dvp); 1660 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1661 1662 /* 1663 * Oops, not for me.. 1664 */ 1665 if (vap->va_type == VSOCK) 1666 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1667 1668 KASSERT(vap->va_type == VREG); 1669 1670 #ifdef VA_EXCLUSIVE 1671 if (vap->va_vaflags & VA_EXCLUSIVE) { 1672 excl_mode = NFSV3CREATE_EXCLUSIVE; 1673 } 1674 #endif 1675 again: 1676 error = 0; 1677 nfsstats.rpccnt[NFSPROC_CREATE]++; 1678 dnp = VTONFS(dvp); 1679 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1680 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1681 nfsm_fhtom(dnp, v3); 1682 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1683 #ifndef NFS_V2_ONLY 1684 if (v3) { 1685 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1686 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1687 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1688 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1689 #ifdef INET 1690 if (TAILQ_FIRST(&in_ifaddrhead)) 1691 *tl++ = TAILQ_FIRST(&in_ifaddrhead)-> 1692 ia_addr.sin_addr.s_addr; 1693 else 1694 *tl++ = create_verf; 1695 #else 1696 *tl++ = create_verf; 1697 #endif 1698 *tl = ++create_verf; 1699 } else { 1700 *tl = txdr_unsigned(excl_mode); 1701 nfsm_v3attrbuild(vap, false); 1702 } 1703 } else 1704 #endif 1705 { 1706 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1707 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1708 sp->sa_uid = nfs_xdrneg1; 1709 sp->sa_gid = nfs_xdrneg1; 1710 sp->sa_size = 0; 1711 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1712 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1713 } 1714 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1715 if (!error) { 1716 nfsm_mtofh(dvp, newvp, v3, gotvp); 1717 if (!gotvp) { 1718 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1719 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1720 if (!error) 1721 newvp = NFSTOV(np); 1722 } 1723 } 1724 #ifndef NFS_V2_ONLY 1725 if (v3) 1726 nfsm_wcc_data(dvp, wccflag, 0, !error); 1727 #endif 1728 nfsm_reqdone; 1729 if (error) { 1730 /* 1731 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1732 */ 1733 if (v3 && error == ENOTSUP) { 1734 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1735 excl_mode = NFSV3CREATE_GUARDED; 1736 goto again; 1737 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1738 excl_mode = NFSV3CREATE_UNCHECKED; 1739 goto again; 1740 } 1741 } 1742 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1743 struct timespec ts; 1744 1745 getnanotime(&ts); 1746 1747 /* 1748 * make sure that we'll update timestamps as 1749 * most server implementations use them to store 1750 * the create verifier. 1751 * 1752 * XXX it's better to use TOSERVER always. 1753 */ 1754 1755 if (vap->va_atime.tv_sec == VNOVAL) 1756 vap->va_atime = ts; 1757 if (vap->va_mtime.tv_sec == VNOVAL) 1758 vap->va_mtime = ts; 1759 1760 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1761 } 1762 if (error == 0) { 1763 if (cnp->cn_flags & MAKEENTRY) 1764 nfs_cache_enter(dvp, newvp, cnp); 1765 else 1766 cache_purge1(dvp, cnp, 0); 1767 *ap->a_vpp = newvp; 1768 } else { 1769 if (newvp) 1770 vput(newvp); 1771 if (error == EEXIST) 1772 cache_purge1(dvp, cnp, 0); 1773 } 1774 PNBUF_PUT(cnp->cn_pnbuf); 1775 VTONFS(dvp)->n_flag |= NMODIFIED; 1776 if (!wccflag) 1777 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1778 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1779 vput(dvp); 1780 return (error); 1781 } 1782 1783 /* 1784 * nfs file remove call 1785 * To try and make nfs semantics closer to ufs semantics, a file that has 1786 * other processes using the vnode is renamed instead of removed and then 1787 * removed later on the last close. 1788 * - If v_usecount > 1 1789 * If a rename is not already in the works 1790 * call nfs_sillyrename() to set it up 1791 * else 1792 * do the remove rpc 1793 */ 1794 int 1795 nfs_remove(v) 1796 void *v; 1797 { 1798 struct vop_remove_args /* { 1799 struct vnodeop_desc *a_desc; 1800 struct vnode * a_dvp; 1801 struct vnode * a_vp; 1802 struct componentname * a_cnp; 1803 } */ *ap = v; 1804 struct vnode *vp = ap->a_vp; 1805 struct vnode *dvp = ap->a_dvp; 1806 struct componentname *cnp = ap->a_cnp; 1807 struct nfsnode *np = VTONFS(vp); 1808 int error = 0; 1809 struct vattr vattr; 1810 1811 #ifndef DIAGNOSTIC 1812 if ((cnp->cn_flags & HASBUF) == 0) 1813 panic("nfs_remove: no name"); 1814 if (vp->v_usecount < 1) 1815 panic("nfs_remove: bad v_usecount"); 1816 #endif 1817 if (vp->v_type == VDIR) 1818 error = EPERM; 1819 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1820 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1821 vattr.va_nlink > 1)) { 1822 /* 1823 * Purge the name cache so that the chance of a lookup for 1824 * the name succeeding while the remove is in progress is 1825 * minimized. Without node locking it can still happen, such 1826 * that an I/O op returns ESTALE, but since you get this if 1827 * another host removes the file.. 1828 */ 1829 cache_purge(vp); 1830 /* 1831 * throw away biocache buffers, mainly to avoid 1832 * unnecessary delayed writes later. 1833 */ 1834 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1835 /* Do the rpc */ 1836 if (error != EINTR) 1837 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1838 cnp->cn_namelen, cnp->cn_cred, curlwp); 1839 } else if (!np->n_sillyrename) 1840 error = nfs_sillyrename(dvp, vp, cnp, false); 1841 PNBUF_PUT(cnp->cn_pnbuf); 1842 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1843 vattr.va_nlink == 1) { 1844 np->n_flag |= NREMOVED; 1845 } 1846 NFS_INVALIDATE_ATTRCACHE(np); 1847 VN_KNOTE(vp, NOTE_DELETE); 1848 VN_KNOTE(dvp, NOTE_WRITE); 1849 if (dvp == vp) 1850 vrele(vp); 1851 else 1852 vput(vp); 1853 vput(dvp); 1854 return (error); 1855 } 1856 1857 /* 1858 * nfs file remove rpc called from nfs_inactive 1859 */ 1860 int 1861 nfs_removeit(sp) 1862 struct sillyrename *sp; 1863 { 1864 1865 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1866 (struct lwp *)0)); 1867 } 1868 1869 /* 1870 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1871 */ 1872 int 1873 nfs_removerpc(dvp, name, namelen, cred, l) 1874 struct vnode *dvp; 1875 const char *name; 1876 int namelen; 1877 kauth_cred_t cred; 1878 struct lwp *l; 1879 { 1880 u_int32_t *tl; 1881 char *cp; 1882 #ifndef NFS_V2_ONLY 1883 int32_t t1; 1884 char *cp2; 1885 #endif 1886 int32_t t2; 1887 char *bpos, *dpos; 1888 int error = 0, wccflag = NFSV3_WCCRATTR; 1889 struct mbuf *mreq, *mrep, *md, *mb; 1890 const int v3 = NFS_ISV3(dvp); 1891 int rexmit = 0; 1892 struct nfsnode *dnp = VTONFS(dvp); 1893 1894 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1895 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1896 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1897 nfsm_fhtom(dnp, v3); 1898 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1899 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1900 #ifndef NFS_V2_ONLY 1901 if (v3) 1902 nfsm_wcc_data(dvp, wccflag, 0, !error); 1903 #endif 1904 nfsm_reqdone; 1905 VTONFS(dvp)->n_flag |= NMODIFIED; 1906 if (!wccflag) 1907 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1908 /* 1909 * Kludge City: If the first reply to the remove rpc is lost.. 1910 * the reply to the retransmitted request will be ENOENT 1911 * since the file was in fact removed 1912 * Therefore, we cheat and return success. 1913 */ 1914 if (rexmit && error == ENOENT) 1915 error = 0; 1916 return (error); 1917 } 1918 1919 /* 1920 * nfs file rename call 1921 */ 1922 int 1923 nfs_rename(v) 1924 void *v; 1925 { 1926 struct vop_rename_args /* { 1927 struct vnode *a_fdvp; 1928 struct vnode *a_fvp; 1929 struct componentname *a_fcnp; 1930 struct vnode *a_tdvp; 1931 struct vnode *a_tvp; 1932 struct componentname *a_tcnp; 1933 } */ *ap = v; 1934 struct vnode *fvp = ap->a_fvp; 1935 struct vnode *tvp = ap->a_tvp; 1936 struct vnode *fdvp = ap->a_fdvp; 1937 struct vnode *tdvp = ap->a_tdvp; 1938 struct componentname *tcnp = ap->a_tcnp; 1939 struct componentname *fcnp = ap->a_fcnp; 1940 int error; 1941 1942 #ifndef DIAGNOSTIC 1943 if ((tcnp->cn_flags & HASBUF) == 0 || 1944 (fcnp->cn_flags & HASBUF) == 0) 1945 panic("nfs_rename: no name"); 1946 #endif 1947 /* Check for cross-device rename */ 1948 if ((fvp->v_mount != tdvp->v_mount) || 1949 (tvp && (fvp->v_mount != tvp->v_mount))) { 1950 error = EXDEV; 1951 goto out; 1952 } 1953 1954 /* 1955 * If the tvp exists and is in use, sillyrename it before doing the 1956 * rename of the new file over it. 1957 * 1958 * Have sillyrename use link instead of rename if possible, 1959 * so that we don't lose the file if the rename fails, and so 1960 * that there's no window when the "to" file doesn't exist. 1961 */ 1962 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1963 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1964 VN_KNOTE(tvp, NOTE_DELETE); 1965 vput(tvp); 1966 tvp = NULL; 1967 } 1968 1969 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1970 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1971 curlwp); 1972 1973 VN_KNOTE(fdvp, NOTE_WRITE); 1974 VN_KNOTE(tdvp, NOTE_WRITE); 1975 if (error == 0 || error == EEXIST) { 1976 if (fvp->v_type == VDIR) 1977 cache_purge(fvp); 1978 else 1979 cache_purge1(fdvp, fcnp, 0); 1980 if (tvp != NULL && tvp->v_type == VDIR) 1981 cache_purge(tvp); 1982 else 1983 cache_purge1(tdvp, tcnp, 0); 1984 } 1985 out: 1986 if (tdvp == tvp) 1987 vrele(tdvp); 1988 else 1989 vput(tdvp); 1990 if (tvp) 1991 vput(tvp); 1992 vrele(fdvp); 1993 vrele(fvp); 1994 return (error); 1995 } 1996 1997 /* 1998 * nfs file rename rpc called from nfs_remove() above 1999 */ 2000 int 2001 nfs_renameit(sdvp, scnp, sp) 2002 struct vnode *sdvp; 2003 struct componentname *scnp; 2004 struct sillyrename *sp; 2005 { 2006 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 2007 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 2008 } 2009 2010 /* 2011 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2012 */ 2013 int 2014 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l) 2015 struct vnode *fdvp; 2016 const char *fnameptr; 2017 int fnamelen; 2018 struct vnode *tdvp; 2019 const char *tnameptr; 2020 int tnamelen; 2021 kauth_cred_t cred; 2022 struct lwp *l; 2023 { 2024 u_int32_t *tl; 2025 char *cp; 2026 #ifndef NFS_V2_ONLY 2027 int32_t t1; 2028 char *cp2; 2029 #endif 2030 int32_t t2; 2031 char *bpos, *dpos; 2032 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 2033 struct mbuf *mreq, *mrep, *md, *mb; 2034 const int v3 = NFS_ISV3(fdvp); 2035 int rexmit = 0; 2036 struct nfsnode *fdnp = VTONFS(fdvp); 2037 2038 nfsstats.rpccnt[NFSPROC_RENAME]++; 2039 nfsm_reqhead(fdnp, NFSPROC_RENAME, 2040 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 2041 nfsm_rndup(tnamelen)); 2042 nfsm_fhtom(fdnp, v3); 2043 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 2044 nfsm_fhtom(VTONFS(tdvp), v3); 2045 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 2046 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 2047 #ifndef NFS_V2_ONLY 2048 if (v3) { 2049 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 2050 nfsm_wcc_data(tdvp, twccflag, 0, !error); 2051 } 2052 #endif 2053 nfsm_reqdone; 2054 VTONFS(fdvp)->n_flag |= NMODIFIED; 2055 VTONFS(tdvp)->n_flag |= NMODIFIED; 2056 if (!fwccflag) 2057 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 2058 if (!twccflag) 2059 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 2060 /* 2061 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2062 */ 2063 if (rexmit && error == ENOENT) 2064 error = 0; 2065 return (error); 2066 } 2067 2068 /* 2069 * NFS link RPC, called from nfs_link. 2070 * Assumes dvp and vp locked, and leaves them that way. 2071 */ 2072 2073 static int 2074 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2075 size_t namelen, kauth_cred_t cred, struct lwp *l) 2076 { 2077 u_int32_t *tl; 2078 char *cp; 2079 #ifndef NFS_V2_ONLY 2080 int32_t t1; 2081 char *cp2; 2082 #endif 2083 int32_t t2; 2084 char *bpos, *dpos; 2085 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2086 struct mbuf *mreq, *mrep, *md, *mb; 2087 const int v3 = NFS_ISV3(dvp); 2088 int rexmit = 0; 2089 struct nfsnode *np = VTONFS(vp); 2090 2091 nfsstats.rpccnt[NFSPROC_LINK]++; 2092 nfsm_reqhead(np, NFSPROC_LINK, 2093 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2094 nfsm_fhtom(np, v3); 2095 nfsm_fhtom(VTONFS(dvp), v3); 2096 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2097 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2098 #ifndef NFS_V2_ONLY 2099 if (v3) { 2100 nfsm_postop_attr(vp, attrflag, 0); 2101 nfsm_wcc_data(dvp, wccflag, 0, !error); 2102 } 2103 #endif 2104 nfsm_reqdone; 2105 2106 VTONFS(dvp)->n_flag |= NMODIFIED; 2107 if (!attrflag) 2108 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2109 if (!wccflag) 2110 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2111 2112 /* 2113 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2114 */ 2115 if (rexmit && error == EEXIST) 2116 error = 0; 2117 2118 return error; 2119 } 2120 2121 /* 2122 * nfs hard link create call 2123 */ 2124 int 2125 nfs_link(v) 2126 void *v; 2127 { 2128 struct vop_link_args /* { 2129 struct vnode *a_dvp; 2130 struct vnode *a_vp; 2131 struct componentname *a_cnp; 2132 } */ *ap = v; 2133 struct vnode *vp = ap->a_vp; 2134 struct vnode *dvp = ap->a_dvp; 2135 struct componentname *cnp = ap->a_cnp; 2136 int error = 0; 2137 2138 if (dvp->v_mount != vp->v_mount) { 2139 VOP_ABORTOP(dvp, cnp); 2140 vput(dvp); 2141 return (EXDEV); 2142 } 2143 if (dvp != vp) { 2144 error = vn_lock(vp, LK_EXCLUSIVE); 2145 if (error != 0) { 2146 VOP_ABORTOP(dvp, cnp); 2147 vput(dvp); 2148 return error; 2149 } 2150 } 2151 2152 /* 2153 * Push all writes to the server, so that the attribute cache 2154 * doesn't get "out of sync" with the server. 2155 * XXX There should be a better way! 2156 */ 2157 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2158 2159 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2160 cnp->cn_cred, curlwp); 2161 2162 if (error == 0) 2163 cache_purge1(dvp, cnp, 0); 2164 PNBUF_PUT(cnp->cn_pnbuf); 2165 if (dvp != vp) 2166 VOP_UNLOCK(vp, 0); 2167 VN_KNOTE(vp, NOTE_LINK); 2168 VN_KNOTE(dvp, NOTE_WRITE); 2169 vput(dvp); 2170 return (error); 2171 } 2172 2173 /* 2174 * nfs symbolic link create call 2175 */ 2176 int 2177 nfs_symlink(v) 2178 void *v; 2179 { 2180 struct vop_symlink_args /* { 2181 struct vnode *a_dvp; 2182 struct vnode **a_vpp; 2183 struct componentname *a_cnp; 2184 struct vattr *a_vap; 2185 char *a_target; 2186 } */ *ap = v; 2187 struct vnode *dvp = ap->a_dvp; 2188 struct vattr *vap = ap->a_vap; 2189 struct componentname *cnp = ap->a_cnp; 2190 struct nfsv2_sattr *sp; 2191 u_int32_t *tl; 2192 char *cp; 2193 int32_t t1, t2; 2194 char *bpos, *dpos, *cp2; 2195 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2196 struct mbuf *mreq, *mrep, *md, *mb; 2197 struct vnode *newvp = (struct vnode *)0; 2198 const int v3 = NFS_ISV3(dvp); 2199 int rexmit = 0; 2200 struct nfsnode *dnp = VTONFS(dvp); 2201 2202 *ap->a_vpp = NULL; 2203 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2204 slen = strlen(ap->a_target); 2205 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2206 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2207 nfsm_fhtom(dnp, v3); 2208 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2209 #ifndef NFS_V2_ONlY 2210 if (v3) 2211 nfsm_v3attrbuild(vap, false); 2212 #endif 2213 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2214 #ifndef NFS_V2_ONlY 2215 if (!v3) { 2216 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2217 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2218 sp->sa_uid = nfs_xdrneg1; 2219 sp->sa_gid = nfs_xdrneg1; 2220 sp->sa_size = nfs_xdrneg1; 2221 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2222 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2223 } 2224 #endif 2225 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2226 &rexmit); 2227 #ifndef NFS_V2_ONlY 2228 if (v3) { 2229 if (!error) 2230 nfsm_mtofh(dvp, newvp, v3, gotvp); 2231 nfsm_wcc_data(dvp, wccflag, 0, !error); 2232 } 2233 #endif 2234 nfsm_reqdone; 2235 /* 2236 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2237 */ 2238 if (rexmit && error == EEXIST) 2239 error = 0; 2240 if (error == 0 || error == EEXIST) 2241 cache_purge1(dvp, cnp, 0); 2242 if (error == 0 && newvp == NULL) { 2243 struct nfsnode *np = NULL; 2244 2245 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2246 cnp->cn_cred, curlwp, &np); 2247 if (error == 0) 2248 newvp = NFSTOV(np); 2249 } 2250 if (error) { 2251 if (newvp != NULL) 2252 vput(newvp); 2253 } else { 2254 *ap->a_vpp = newvp; 2255 } 2256 PNBUF_PUT(cnp->cn_pnbuf); 2257 VTONFS(dvp)->n_flag |= NMODIFIED; 2258 if (!wccflag) 2259 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2260 VN_KNOTE(dvp, NOTE_WRITE); 2261 vput(dvp); 2262 return (error); 2263 } 2264 2265 /* 2266 * nfs make dir call 2267 */ 2268 int 2269 nfs_mkdir(v) 2270 void *v; 2271 { 2272 struct vop_mkdir_args /* { 2273 struct vnode *a_dvp; 2274 struct vnode **a_vpp; 2275 struct componentname *a_cnp; 2276 struct vattr *a_vap; 2277 } */ *ap = v; 2278 struct vnode *dvp = ap->a_dvp; 2279 struct vattr *vap = ap->a_vap; 2280 struct componentname *cnp = ap->a_cnp; 2281 struct nfsv2_sattr *sp; 2282 u_int32_t *tl; 2283 char *cp; 2284 int32_t t1, t2; 2285 int len; 2286 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2287 struct vnode *newvp = (struct vnode *)0; 2288 char *bpos, *dpos, *cp2; 2289 int error = 0, wccflag = NFSV3_WCCRATTR; 2290 int gotvp = 0; 2291 int rexmit = 0; 2292 struct mbuf *mreq, *mrep, *md, *mb; 2293 const int v3 = NFS_ISV3(dvp); 2294 2295 len = cnp->cn_namelen; 2296 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2297 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2298 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2299 nfsm_fhtom(dnp, v3); 2300 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2301 #ifndef NFS_V2_ONLY 2302 if (v3) { 2303 nfsm_v3attrbuild(vap, false); 2304 } else 2305 #endif 2306 { 2307 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2308 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2309 sp->sa_uid = nfs_xdrneg1; 2310 sp->sa_gid = nfs_xdrneg1; 2311 sp->sa_size = nfs_xdrneg1; 2312 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2313 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2314 } 2315 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2316 if (!error) 2317 nfsm_mtofh(dvp, newvp, v3, gotvp); 2318 if (v3) 2319 nfsm_wcc_data(dvp, wccflag, 0, !error); 2320 nfsm_reqdone; 2321 VTONFS(dvp)->n_flag |= NMODIFIED; 2322 if (!wccflag) 2323 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2324 /* 2325 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2326 * if we can succeed in looking up the directory. 2327 */ 2328 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2329 if (newvp) { 2330 vput(newvp); 2331 newvp = (struct vnode *)0; 2332 } 2333 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2334 curlwp, &np); 2335 if (!error) { 2336 newvp = NFSTOV(np); 2337 if (newvp->v_type != VDIR || newvp == dvp) 2338 error = EEXIST; 2339 } 2340 } 2341 if (error) { 2342 if (newvp) { 2343 if (dvp != newvp) 2344 vput(newvp); 2345 else 2346 vrele(newvp); 2347 } 2348 } else { 2349 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2350 if (cnp->cn_flags & MAKEENTRY) 2351 nfs_cache_enter(dvp, newvp, cnp); 2352 *ap->a_vpp = newvp; 2353 } 2354 PNBUF_PUT(cnp->cn_pnbuf); 2355 vput(dvp); 2356 return (error); 2357 } 2358 2359 /* 2360 * nfs remove directory call 2361 */ 2362 int 2363 nfs_rmdir(v) 2364 void *v; 2365 { 2366 struct vop_rmdir_args /* { 2367 struct vnode *a_dvp; 2368 struct vnode *a_vp; 2369 struct componentname *a_cnp; 2370 } */ *ap = v; 2371 struct vnode *vp = ap->a_vp; 2372 struct vnode *dvp = ap->a_dvp; 2373 struct componentname *cnp = ap->a_cnp; 2374 u_int32_t *tl; 2375 char *cp; 2376 #ifndef NFS_V2_ONLY 2377 int32_t t1; 2378 char *cp2; 2379 #endif 2380 int32_t t2; 2381 char *bpos, *dpos; 2382 int error = 0, wccflag = NFSV3_WCCRATTR; 2383 int rexmit = 0; 2384 struct mbuf *mreq, *mrep, *md, *mb; 2385 const int v3 = NFS_ISV3(dvp); 2386 struct nfsnode *dnp; 2387 2388 if (dvp == vp) { 2389 vrele(dvp); 2390 vput(dvp); 2391 PNBUF_PUT(cnp->cn_pnbuf); 2392 return (EINVAL); 2393 } 2394 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2395 dnp = VTONFS(dvp); 2396 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2397 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2398 nfsm_fhtom(dnp, v3); 2399 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2400 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2401 #ifndef NFS_V2_ONLY 2402 if (v3) 2403 nfsm_wcc_data(dvp, wccflag, 0, !error); 2404 #endif 2405 nfsm_reqdone; 2406 PNBUF_PUT(cnp->cn_pnbuf); 2407 VTONFS(dvp)->n_flag |= NMODIFIED; 2408 if (!wccflag) 2409 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2410 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2411 VN_KNOTE(vp, NOTE_DELETE); 2412 cache_purge(vp); 2413 vput(vp); 2414 vput(dvp); 2415 /* 2416 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2417 */ 2418 if (rexmit && error == ENOENT) 2419 error = 0; 2420 return (error); 2421 } 2422 2423 /* 2424 * nfs readdir call 2425 */ 2426 int 2427 nfs_readdir(v) 2428 void *v; 2429 { 2430 struct vop_readdir_args /* { 2431 struct vnode *a_vp; 2432 struct uio *a_uio; 2433 kauth_cred_t a_cred; 2434 int *a_eofflag; 2435 off_t **a_cookies; 2436 int *a_ncookies; 2437 } */ *ap = v; 2438 struct vnode *vp = ap->a_vp; 2439 struct uio *uio = ap->a_uio; 2440 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2441 char *base = uio->uio_iov->iov_base; 2442 int tresid, error; 2443 size_t count, lost; 2444 struct dirent *dp; 2445 off_t *cookies = NULL; 2446 int ncookies = 0, nc; 2447 2448 if (vp->v_type != VDIR) 2449 return (EPERM); 2450 2451 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2452 count = uio->uio_resid - lost; 2453 if (count <= 0) 2454 return (EINVAL); 2455 2456 /* 2457 * Call nfs_bioread() to do the real work. 2458 */ 2459 tresid = uio->uio_resid = count; 2460 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2461 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2462 2463 if (!error && ap->a_cookies) { 2464 ncookies = count / 16; 2465 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2466 *ap->a_cookies = cookies; 2467 } 2468 2469 if (!error && uio->uio_resid == tresid) { 2470 uio->uio_resid += lost; 2471 nfsstats.direofcache_misses++; 2472 if (ap->a_cookies) 2473 *ap->a_ncookies = 0; 2474 *ap->a_eofflag = 1; 2475 return (0); 2476 } 2477 2478 if (!error && ap->a_cookies) { 2479 /* 2480 * Only the NFS server and emulations use cookies, and they 2481 * load the directory block into system space, so we can 2482 * just look at it directly. 2483 */ 2484 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2485 uio->uio_iovcnt != 1) 2486 panic("nfs_readdir: lost in space"); 2487 for (nc = 0; ncookies-- && 2488 base < (char *)uio->uio_iov->iov_base; nc++){ 2489 dp = (struct dirent *) base; 2490 if (dp->d_reclen == 0) 2491 break; 2492 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2493 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2494 else 2495 *(cookies++) = NFS_GETCOOKIE(dp); 2496 base += dp->d_reclen; 2497 } 2498 uio->uio_resid += 2499 ((char *)uio->uio_iov->iov_base - base); 2500 uio->uio_iov->iov_len += 2501 ((char *)uio->uio_iov->iov_base - base); 2502 uio->uio_iov->iov_base = base; 2503 *ap->a_ncookies = nc; 2504 } 2505 2506 uio->uio_resid += lost; 2507 *ap->a_eofflag = 0; 2508 return (error); 2509 } 2510 2511 /* 2512 * Readdir rpc call. 2513 * Called from below the buffer cache by nfs_doio(). 2514 */ 2515 int 2516 nfs_readdirrpc(vp, uiop, cred) 2517 struct vnode *vp; 2518 struct uio *uiop; 2519 kauth_cred_t cred; 2520 { 2521 int len, left; 2522 struct dirent *dp = NULL; 2523 u_int32_t *tl; 2524 char *cp; 2525 int32_t t1, t2; 2526 char *bpos, *dpos, *cp2; 2527 struct mbuf *mreq, *mrep, *md, *mb; 2528 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2529 struct nfsnode *dnp = VTONFS(vp); 2530 u_quad_t fileno; 2531 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2532 #ifndef NFS_V2_ONLY 2533 int attrflag; 2534 #endif 2535 int nrpcs = 0, reclen; 2536 const int v3 = NFS_ISV3(vp); 2537 2538 #ifdef DIAGNOSTIC 2539 /* 2540 * Should be called from buffer cache, so only amount of 2541 * NFS_DIRBLKSIZ will be requested. 2542 */ 2543 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2544 panic("nfs readdirrpc bad uio"); 2545 #endif 2546 2547 /* 2548 * Loop around doing readdir rpc's of size nm_readdirsize 2549 * truncated to a multiple of NFS_DIRFRAGSIZ. 2550 * The stopping criteria is EOF or buffer full. 2551 */ 2552 while (more_dirs && bigenough) { 2553 /* 2554 * Heuristic: don't bother to do another RPC to further 2555 * fill up this block if there is not much room left. (< 50% 2556 * of the readdir RPC size). This wastes some buffer space 2557 * but can save up to 50% in RPC calls. 2558 */ 2559 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2560 bigenough = 0; 2561 break; 2562 } 2563 nfsstats.rpccnt[NFSPROC_READDIR]++; 2564 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2565 NFSX_READDIR(v3)); 2566 nfsm_fhtom(dnp, v3); 2567 #ifndef NFS_V2_ONLY 2568 if (v3) { 2569 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2570 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2571 txdr_swapcookie3(uiop->uio_offset, tl); 2572 } else { 2573 txdr_cookie3(uiop->uio_offset, tl); 2574 } 2575 tl += 2; 2576 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2577 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2578 } else 2579 #endif 2580 { 2581 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2582 *tl++ = txdr_unsigned(uiop->uio_offset); 2583 } 2584 *tl = txdr_unsigned(nmp->nm_readdirsize); 2585 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2586 nrpcs++; 2587 #ifndef NFS_V2_ONLY 2588 if (v3) { 2589 nfsm_postop_attr(vp, attrflag, 0); 2590 if (!error) { 2591 nfsm_dissect(tl, u_int32_t *, 2592 2 * NFSX_UNSIGNED); 2593 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2594 dnp->n_cookieverf.nfsuquad[1] = *tl; 2595 } else { 2596 m_freem(mrep); 2597 goto nfsmout; 2598 } 2599 } 2600 #endif 2601 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2602 more_dirs = fxdr_unsigned(int, *tl); 2603 2604 /* loop thru the dir entries, doctoring them to 4bsd form */ 2605 while (more_dirs && bigenough) { 2606 #ifndef NFS_V2_ONLY 2607 if (v3) { 2608 nfsm_dissect(tl, u_int32_t *, 2609 3 * NFSX_UNSIGNED); 2610 fileno = fxdr_hyper(tl); 2611 len = fxdr_unsigned(int, *(tl + 2)); 2612 } else 2613 #endif 2614 { 2615 nfsm_dissect(tl, u_int32_t *, 2616 2 * NFSX_UNSIGNED); 2617 fileno = fxdr_unsigned(u_quad_t, *tl++); 2618 len = fxdr_unsigned(int, *tl); 2619 } 2620 if (len <= 0 || len > NFS_MAXNAMLEN) { 2621 error = EBADRPC; 2622 m_freem(mrep); 2623 goto nfsmout; 2624 } 2625 /* for cookie stashing */ 2626 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2627 left = NFS_DIRFRAGSIZ - blksiz; 2628 if (reclen > left) { 2629 memset(uiop->uio_iov->iov_base, 0, left); 2630 dp->d_reclen += left; 2631 UIO_ADVANCE(uiop, left); 2632 blksiz = 0; 2633 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2634 } 2635 if (reclen > uiop->uio_resid) 2636 bigenough = 0; 2637 if (bigenough) { 2638 int tlen; 2639 2640 dp = (struct dirent *)uiop->uio_iov->iov_base; 2641 dp->d_fileno = fileno; 2642 dp->d_namlen = len; 2643 dp->d_reclen = reclen; 2644 dp->d_type = DT_UNKNOWN; 2645 blksiz += reclen; 2646 if (blksiz == NFS_DIRFRAGSIZ) 2647 blksiz = 0; 2648 UIO_ADVANCE(uiop, DIRHDSIZ); 2649 nfsm_mtouio(uiop, len); 2650 tlen = reclen - (DIRHDSIZ + len); 2651 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2652 UIO_ADVANCE(uiop, tlen); 2653 } else 2654 nfsm_adv(nfsm_rndup(len)); 2655 #ifndef NFS_V2_ONLY 2656 if (v3) { 2657 nfsm_dissect(tl, u_int32_t *, 2658 3 * NFSX_UNSIGNED); 2659 } else 2660 #endif 2661 { 2662 nfsm_dissect(tl, u_int32_t *, 2663 2 * NFSX_UNSIGNED); 2664 } 2665 if (bigenough) { 2666 #ifndef NFS_V2_ONLY 2667 if (v3) { 2668 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2669 uiop->uio_offset = 2670 fxdr_swapcookie3(tl); 2671 else 2672 uiop->uio_offset = 2673 fxdr_cookie3(tl); 2674 } 2675 else 2676 #endif 2677 { 2678 uiop->uio_offset = 2679 fxdr_unsigned(off_t, *tl); 2680 } 2681 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2682 } 2683 if (v3) 2684 tl += 2; 2685 else 2686 tl++; 2687 more_dirs = fxdr_unsigned(int, *tl); 2688 } 2689 /* 2690 * If at end of rpc data, get the eof boolean 2691 */ 2692 if (!more_dirs) { 2693 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2694 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2695 2696 /* 2697 * kludge: if we got no entries, treat it as EOF. 2698 * some server sometimes send a reply without any 2699 * entries or EOF. 2700 * although it might mean the server has very long name, 2701 * we can't handle such entries anyway. 2702 */ 2703 2704 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2705 more_dirs = 0; 2706 } 2707 m_freem(mrep); 2708 } 2709 /* 2710 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2711 * by increasing d_reclen for the last record. 2712 */ 2713 if (blksiz > 0) { 2714 left = NFS_DIRFRAGSIZ - blksiz; 2715 memset(uiop->uio_iov->iov_base, 0, left); 2716 dp->d_reclen += left; 2717 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2718 UIO_ADVANCE(uiop, left); 2719 } 2720 2721 /* 2722 * We are now either at the end of the directory or have filled the 2723 * block. 2724 */ 2725 if (bigenough) { 2726 dnp->n_direofoffset = uiop->uio_offset; 2727 dnp->n_flag |= NEOFVALID; 2728 } 2729 nfsmout: 2730 return (error); 2731 } 2732 2733 #ifndef NFS_V2_ONLY 2734 /* 2735 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2736 */ 2737 int 2738 nfs_readdirplusrpc(vp, uiop, cred) 2739 struct vnode *vp; 2740 struct uio *uiop; 2741 kauth_cred_t cred; 2742 { 2743 int len, left; 2744 struct dirent *dp = NULL; 2745 u_int32_t *tl; 2746 char *cp; 2747 int32_t t1, t2; 2748 struct vnode *newvp; 2749 char *bpos, *dpos, *cp2; 2750 struct mbuf *mreq, *mrep, *md, *mb; 2751 struct nameidata nami, *ndp = &nami; 2752 struct componentname *cnp = &ndp->ni_cnd; 2753 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2754 struct nfsnode *dnp = VTONFS(vp), *np; 2755 nfsfh_t *fhp; 2756 u_quad_t fileno; 2757 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2758 int attrflag, fhsize, nrpcs = 0, reclen; 2759 struct nfs_fattr fattr, *fp; 2760 2761 #ifdef DIAGNOSTIC 2762 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2763 panic("nfs readdirplusrpc bad uio"); 2764 #endif 2765 ndp->ni_dvp = vp; 2766 newvp = NULLVP; 2767 2768 /* 2769 * Loop around doing readdir rpc's of size nm_readdirsize 2770 * truncated to a multiple of NFS_DIRFRAGSIZ. 2771 * The stopping criteria is EOF or buffer full. 2772 */ 2773 while (more_dirs && bigenough) { 2774 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2775 bigenough = 0; 2776 break; 2777 } 2778 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2779 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2780 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2781 nfsm_fhtom(dnp, 1); 2782 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2783 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2784 txdr_swapcookie3(uiop->uio_offset, tl); 2785 } else { 2786 txdr_cookie3(uiop->uio_offset, tl); 2787 } 2788 tl += 2; 2789 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2790 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2791 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2792 *tl = txdr_unsigned(nmp->nm_rsize); 2793 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2794 nfsm_postop_attr(vp, attrflag, 0); 2795 if (error) { 2796 m_freem(mrep); 2797 goto nfsmout; 2798 } 2799 nrpcs++; 2800 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2801 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2802 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2803 more_dirs = fxdr_unsigned(int, *tl); 2804 2805 /* loop thru the dir entries, doctoring them to 4bsd form */ 2806 while (more_dirs && bigenough) { 2807 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2808 fileno = fxdr_hyper(tl); 2809 len = fxdr_unsigned(int, *(tl + 2)); 2810 if (len <= 0 || len > NFS_MAXNAMLEN) { 2811 error = EBADRPC; 2812 m_freem(mrep); 2813 goto nfsmout; 2814 } 2815 /* for cookie stashing */ 2816 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2817 left = NFS_DIRFRAGSIZ - blksiz; 2818 if (reclen > left) { 2819 /* 2820 * DIRFRAGSIZ is aligned, no need to align 2821 * again here. 2822 */ 2823 memset(uiop->uio_iov->iov_base, 0, left); 2824 dp->d_reclen += left; 2825 UIO_ADVANCE(uiop, left); 2826 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2827 blksiz = 0; 2828 } 2829 if (reclen > uiop->uio_resid) 2830 bigenough = 0; 2831 if (bigenough) { 2832 int tlen; 2833 2834 dp = (struct dirent *)uiop->uio_iov->iov_base; 2835 dp->d_fileno = fileno; 2836 dp->d_namlen = len; 2837 dp->d_reclen = reclen; 2838 dp->d_type = DT_UNKNOWN; 2839 blksiz += reclen; 2840 if (blksiz == NFS_DIRFRAGSIZ) 2841 blksiz = 0; 2842 UIO_ADVANCE(uiop, DIRHDSIZ); 2843 nfsm_mtouio(uiop, len); 2844 tlen = reclen - (DIRHDSIZ + len); 2845 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2846 UIO_ADVANCE(uiop, tlen); 2847 cnp->cn_nameptr = dp->d_name; 2848 cnp->cn_namelen = dp->d_namlen; 2849 } else 2850 nfsm_adv(nfsm_rndup(len)); 2851 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2852 if (bigenough) { 2853 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2854 uiop->uio_offset = 2855 fxdr_swapcookie3(tl); 2856 else 2857 uiop->uio_offset = 2858 fxdr_cookie3(tl); 2859 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2860 } 2861 tl += 2; 2862 2863 /* 2864 * Since the attributes are before the file handle 2865 * (sigh), we must skip over the attributes and then 2866 * come back and get them. 2867 */ 2868 attrflag = fxdr_unsigned(int, *tl); 2869 if (attrflag) { 2870 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2871 memcpy(&fattr, fp, NFSX_V3FATTR); 2872 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2873 doit = fxdr_unsigned(int, *tl); 2874 if (doit) { 2875 nfsm_getfh(fhp, fhsize, 1); 2876 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2877 VREF(vp); 2878 newvp = vp; 2879 np = dnp; 2880 } else { 2881 error = nfs_nget1(vp->v_mount, fhp, 2882 fhsize, &np, LK_NOWAIT); 2883 if (!error) 2884 newvp = NFSTOV(np); 2885 } 2886 if (!error) { 2887 const char *xcp; 2888 2889 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2890 if (bigenough) { 2891 dp->d_type = 2892 IFTODT(VTTOIF(np->n_vattr->va_type)); 2893 if (cnp->cn_namelen <= NCHNAMLEN) { 2894 ndp->ni_vp = newvp; 2895 xcp = cnp->cn_nameptr + 2896 cnp->cn_namelen; 2897 cnp->cn_hash = 2898 namei_hash(cnp->cn_nameptr, &xcp); 2899 nfs_cache_enter(ndp->ni_dvp, 2900 ndp->ni_vp, cnp); 2901 } 2902 } 2903 } 2904 error = 0; 2905 } 2906 } else { 2907 /* Just skip over the file handle */ 2908 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2909 i = fxdr_unsigned(int, *tl); 2910 nfsm_adv(nfsm_rndup(i)); 2911 } 2912 if (newvp != NULLVP) { 2913 if (newvp == vp) 2914 vrele(newvp); 2915 else 2916 vput(newvp); 2917 newvp = NULLVP; 2918 } 2919 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2920 more_dirs = fxdr_unsigned(int, *tl); 2921 } 2922 /* 2923 * If at end of rpc data, get the eof boolean 2924 */ 2925 if (!more_dirs) { 2926 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2927 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2928 2929 /* 2930 * kludge: see a comment in nfs_readdirrpc. 2931 */ 2932 2933 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2934 more_dirs = 0; 2935 } 2936 m_freem(mrep); 2937 } 2938 /* 2939 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2940 * by increasing d_reclen for the last record. 2941 */ 2942 if (blksiz > 0) { 2943 left = NFS_DIRFRAGSIZ - blksiz; 2944 memset(uiop->uio_iov->iov_base, 0, left); 2945 dp->d_reclen += left; 2946 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2947 UIO_ADVANCE(uiop, left); 2948 } 2949 2950 /* 2951 * We are now either at the end of the directory or have filled the 2952 * block. 2953 */ 2954 if (bigenough) { 2955 dnp->n_direofoffset = uiop->uio_offset; 2956 dnp->n_flag |= NEOFVALID; 2957 } 2958 nfsmout: 2959 if (newvp != NULLVP) { 2960 if(newvp == vp) 2961 vrele(newvp); 2962 else 2963 vput(newvp); 2964 } 2965 return (error); 2966 } 2967 #endif 2968 2969 /* 2970 * Silly rename. To make the NFS filesystem that is stateless look a little 2971 * more like the "ufs" a remove of an active vnode is translated to a rename 2972 * to a funny looking filename that is removed by nfs_inactive on the 2973 * nfsnode. There is the potential for another process on a different client 2974 * to create the same funny name between the nfs_lookitup() fails and the 2975 * nfs_rename() completes, but... 2976 */ 2977 int 2978 nfs_sillyrename(dvp, vp, cnp, dolink) 2979 struct vnode *dvp, *vp; 2980 struct componentname *cnp; 2981 bool dolink; 2982 { 2983 struct sillyrename *sp; 2984 struct nfsnode *np; 2985 int error; 2986 short pid; 2987 2988 cache_purge(dvp); 2989 np = VTONFS(vp); 2990 #ifndef DIAGNOSTIC 2991 if (vp->v_type == VDIR) 2992 panic("nfs: sillyrename dir"); 2993 #endif 2994 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2995 M_NFSREQ, M_WAITOK); 2996 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2997 sp->s_dvp = dvp; 2998 VREF(dvp); 2999 3000 /* Fudge together a funny name */ 3001 pid = curlwp->l_proc->p_pid; 3002 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 3003 sp->s_namlen = 12; 3004 sp->s_name[8] = hexdigits[pid & 0xf]; 3005 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 3006 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 3007 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 3008 3009 /* Try lookitups until we get one that isn't there */ 3010 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3011 curlwp, (struct nfsnode **)0) == 0) { 3012 sp->s_name[4]++; 3013 if (sp->s_name[4] > 'z') { 3014 error = EINVAL; 3015 goto bad; 3016 } 3017 } 3018 if (dolink) { 3019 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 3020 sp->s_cred, curlwp); 3021 /* 3022 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 3023 */ 3024 if (error == ENOTSUP) { 3025 error = nfs_renameit(dvp, cnp, sp); 3026 } 3027 } else { 3028 error = nfs_renameit(dvp, cnp, sp); 3029 } 3030 if (error) 3031 goto bad; 3032 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3033 curlwp, &np); 3034 np->n_sillyrename = sp; 3035 return (0); 3036 bad: 3037 vrele(sp->s_dvp); 3038 kauth_cred_free(sp->s_cred); 3039 free((void *)sp, M_NFSREQ); 3040 return (error); 3041 } 3042 3043 /* 3044 * Look up a file name and optionally either update the file handle or 3045 * allocate an nfsnode, depending on the value of npp. 3046 * npp == NULL --> just do the lookup 3047 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 3048 * handled too 3049 * *npp != NULL --> update the file handle in the vnode 3050 */ 3051 int 3052 nfs_lookitup(dvp, name, len, cred, l, npp) 3053 struct vnode *dvp; 3054 const char *name; 3055 int len; 3056 kauth_cred_t cred; 3057 struct lwp *l; 3058 struct nfsnode **npp; 3059 { 3060 u_int32_t *tl; 3061 char *cp; 3062 int32_t t1, t2; 3063 struct vnode *newvp = (struct vnode *)0; 3064 struct nfsnode *np, *dnp = VTONFS(dvp); 3065 char *bpos, *dpos, *cp2; 3066 int error = 0, fhlen; 3067 #ifndef NFS_V2_ONLY 3068 int attrflag; 3069 #endif 3070 struct mbuf *mreq, *mrep, *md, *mb; 3071 nfsfh_t *nfhp; 3072 const int v3 = NFS_ISV3(dvp); 3073 3074 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 3075 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 3076 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 3077 nfsm_fhtom(dnp, v3); 3078 nfsm_strtom(name, len, NFS_MAXNAMLEN); 3079 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 3080 if (npp && !error) { 3081 nfsm_getfh(nfhp, fhlen, v3); 3082 if (*npp) { 3083 np = *npp; 3084 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3085 free((void *)np->n_fhp, M_NFSBIGFH); 3086 np->n_fhp = &np->n_fh; 3087 } 3088 #if NFS_SMALLFH < NFSX_V3FHMAX 3089 else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 3090 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 3091 #endif 3092 memcpy((void *)np->n_fhp, (void *)nfhp, fhlen); 3093 np->n_fhsize = fhlen; 3094 newvp = NFSTOV(np); 3095 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3096 VREF(dvp); 3097 newvp = dvp; 3098 np = dnp; 3099 } else { 3100 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 3101 if (error) { 3102 m_freem(mrep); 3103 return (error); 3104 } 3105 newvp = NFSTOV(np); 3106 } 3107 #ifndef NFS_V2_ONLY 3108 if (v3) { 3109 nfsm_postop_attr(newvp, attrflag, 0); 3110 if (!attrflag && *npp == NULL) { 3111 m_freem(mrep); 3112 vput(newvp); 3113 return (ENOENT); 3114 } 3115 } else 3116 #endif 3117 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3118 } 3119 nfsm_reqdone; 3120 if (npp && *npp == NULL) { 3121 if (error) { 3122 if (newvp) 3123 vput(newvp); 3124 } else 3125 *npp = np; 3126 } 3127 return (error); 3128 } 3129 3130 #ifndef NFS_V2_ONLY 3131 /* 3132 * Nfs Version 3 commit rpc 3133 */ 3134 int 3135 nfs_commit(vp, offset, cnt, l) 3136 struct vnode *vp; 3137 off_t offset; 3138 uint32_t cnt; 3139 struct lwp *l; 3140 { 3141 char *cp; 3142 u_int32_t *tl; 3143 int32_t t1, t2; 3144 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3145 char *bpos, *dpos, *cp2; 3146 int error = 0, wccflag = NFSV3_WCCRATTR; 3147 struct mbuf *mreq, *mrep, *md, *mb; 3148 struct nfsnode *np; 3149 3150 KASSERT(NFS_ISV3(vp)); 3151 3152 #ifdef NFS_DEBUG_COMMIT 3153 printf("commit %lu - %lu\n", (unsigned long)offset, 3154 (unsigned long)(offset + cnt)); 3155 #endif 3156 3157 mutex_enter(&nmp->nm_lock); 3158 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3159 mutex_exit(&nmp->nm_lock); 3160 return (0); 3161 } 3162 mutex_exit(&nmp->nm_lock); 3163 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3164 np = VTONFS(vp); 3165 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3166 nfsm_fhtom(np, 1); 3167 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3168 txdr_hyper(offset, tl); 3169 tl += 2; 3170 *tl = txdr_unsigned(cnt); 3171 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3172 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3173 if (!error) { 3174 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3175 mutex_enter(&nmp->nm_lock); 3176 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3177 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3178 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3179 error = NFSERR_STALEWRITEVERF; 3180 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3181 } 3182 mutex_exit(&nmp->nm_lock); 3183 } 3184 nfsm_reqdone; 3185 return (error); 3186 } 3187 #endif 3188 3189 /* 3190 * Kludge City.. 3191 * - make nfs_bmap() essentially a no-op that does no translation 3192 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3193 * (Maybe I could use the process's page mapping, but I was concerned that 3194 * Kernel Write might not be enabled and also figured copyout() would do 3195 * a lot more work than memcpy() and also it currently happens in the 3196 * context of the swapper process (2). 3197 */ 3198 int 3199 nfs_bmap(v) 3200 void *v; 3201 { 3202 struct vop_bmap_args /* { 3203 struct vnode *a_vp; 3204 daddr_t a_bn; 3205 struct vnode **a_vpp; 3206 daddr_t *a_bnp; 3207 int *a_runp; 3208 } */ *ap = v; 3209 struct vnode *vp = ap->a_vp; 3210 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3211 3212 if (ap->a_vpp != NULL) 3213 *ap->a_vpp = vp; 3214 if (ap->a_bnp != NULL) 3215 *ap->a_bnp = ap->a_bn << bshift; 3216 if (ap->a_runp != NULL) 3217 *ap->a_runp = 1024 * 1024; /* XXX */ 3218 return (0); 3219 } 3220 3221 /* 3222 * Strategy routine. 3223 * For async requests when nfsiod(s) are running, queue the request by 3224 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3225 * request. 3226 */ 3227 int 3228 nfs_strategy(v) 3229 void *v; 3230 { 3231 struct vop_strategy_args *ap = v; 3232 struct buf *bp = ap->a_bp; 3233 int error = 0; 3234 3235 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3236 panic("nfs physio/async"); 3237 3238 /* 3239 * If the op is asynchronous and an i/o daemon is waiting 3240 * queue the request, wake it up and wait for completion 3241 * otherwise just do it ourselves. 3242 */ 3243 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3244 error = nfs_doio(bp); 3245 return (error); 3246 } 3247 3248 /* 3249 * fsync vnode op. Just call nfs_flush() with commit == 1. 3250 */ 3251 /* ARGSUSED */ 3252 int 3253 nfs_fsync(v) 3254 void *v; 3255 { 3256 struct vop_fsync_args /* { 3257 struct vnodeop_desc *a_desc; 3258 struct vnode * a_vp; 3259 kauth_cred_t a_cred; 3260 int a_flags; 3261 off_t offlo; 3262 off_t offhi; 3263 struct lwp * a_l; 3264 } */ *ap = v; 3265 3266 struct vnode *vp = ap->a_vp; 3267 3268 if (vp->v_type != VREG) 3269 return 0; 3270 3271 return (nfs_flush(vp, ap->a_cred, 3272 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3273 } 3274 3275 /* 3276 * Flush all the data associated with a vnode. 3277 */ 3278 int 3279 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3280 int commit) 3281 { 3282 struct nfsnode *np = VTONFS(vp); 3283 int error; 3284 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3285 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3286 3287 simple_lock(&vp->v_interlock); 3288 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3289 if (np->n_flag & NWRITEERR) { 3290 error = np->n_error; 3291 np->n_flag &= ~NWRITEERR; 3292 } 3293 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3294 return (error); 3295 } 3296 3297 /* 3298 * Return POSIX pathconf information applicable to nfs. 3299 * 3300 * N.B. The NFS V2 protocol doesn't support this RPC. 3301 */ 3302 /* ARGSUSED */ 3303 int 3304 nfs_pathconf(v) 3305 void *v; 3306 { 3307 struct vop_pathconf_args /* { 3308 struct vnode *a_vp; 3309 int a_name; 3310 register_t *a_retval; 3311 } */ *ap = v; 3312 struct nfsv3_pathconf *pcp; 3313 struct vnode *vp = ap->a_vp; 3314 struct mbuf *mreq, *mrep, *md, *mb; 3315 int32_t t1, t2; 3316 u_int32_t *tl; 3317 char *bpos, *dpos, *cp, *cp2; 3318 int error = 0, attrflag; 3319 #ifndef NFS_V2_ONLY 3320 struct nfsmount *nmp; 3321 unsigned int l; 3322 u_int64_t maxsize; 3323 #endif 3324 const int v3 = NFS_ISV3(vp); 3325 struct nfsnode *np = VTONFS(vp); 3326 3327 switch (ap->a_name) { 3328 /* Names that can be resolved locally. */ 3329 case _PC_PIPE_BUF: 3330 *ap->a_retval = PIPE_BUF; 3331 break; 3332 case _PC_SYNC_IO: 3333 *ap->a_retval = 1; 3334 break; 3335 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3336 case _PC_LINK_MAX: 3337 case _PC_NAME_MAX: 3338 case _PC_CHOWN_RESTRICTED: 3339 case _PC_NO_TRUNC: 3340 if (!v3) { 3341 error = EINVAL; 3342 break; 3343 } 3344 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3345 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3346 nfsm_fhtom(np, 1); 3347 nfsm_request(np, NFSPROC_PATHCONF, 3348 curlwp, curlwp->l_cred); /* XXX */ 3349 nfsm_postop_attr(vp, attrflag, 0); 3350 if (!error) { 3351 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3352 NFSX_V3PATHCONF); 3353 switch (ap->a_name) { 3354 case _PC_LINK_MAX: 3355 *ap->a_retval = 3356 fxdr_unsigned(register_t, pcp->pc_linkmax); 3357 break; 3358 case _PC_NAME_MAX: 3359 *ap->a_retval = 3360 fxdr_unsigned(register_t, pcp->pc_namemax); 3361 break; 3362 case _PC_CHOWN_RESTRICTED: 3363 *ap->a_retval = 3364 (pcp->pc_chownrestricted == nfs_true); 3365 break; 3366 case _PC_NO_TRUNC: 3367 *ap->a_retval = 3368 (pcp->pc_notrunc == nfs_true); 3369 break; 3370 } 3371 } 3372 nfsm_reqdone; 3373 break; 3374 case _PC_FILESIZEBITS: 3375 #ifndef NFS_V2_ONLY 3376 if (v3) { 3377 nmp = VFSTONFS(vp->v_mount); 3378 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3379 if ((error = nfs_fsinfo(nmp, vp, 3380 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3381 break; 3382 for (l = 0, maxsize = nmp->nm_maxfilesize; 3383 (maxsize >> l) > 0; l++) 3384 ; 3385 *ap->a_retval = l + 1; 3386 } else 3387 #endif 3388 { 3389 *ap->a_retval = 32; /* NFS V2 limitation */ 3390 } 3391 break; 3392 default: 3393 error = EINVAL; 3394 break; 3395 } 3396 3397 return (error); 3398 } 3399 3400 /* 3401 * NFS advisory byte-level locks. 3402 */ 3403 int 3404 nfs_advlock(v) 3405 void *v; 3406 { 3407 struct vop_advlock_args /* { 3408 struct vnode *a_vp; 3409 void *a_id; 3410 int a_op; 3411 struct flock *a_fl; 3412 int a_flags; 3413 } */ *ap = v; 3414 struct nfsnode *np = VTONFS(ap->a_vp); 3415 3416 return lf_advlock(ap, &np->n_lockf, np->n_size); 3417 } 3418 3419 /* 3420 * Print out the contents of an nfsnode. 3421 */ 3422 int 3423 nfs_print(v) 3424 void *v; 3425 { 3426 struct vop_print_args /* { 3427 struct vnode *a_vp; 3428 } */ *ap = v; 3429 struct vnode *vp = ap->a_vp; 3430 struct nfsnode *np = VTONFS(vp); 3431 3432 printf("tag VT_NFS, fileid %lld fsid 0x%lx", 3433 (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3434 if (vp->v_type == VFIFO) 3435 fifo_printinfo(vp); 3436 printf("\n"); 3437 return (0); 3438 } 3439 3440 /* 3441 * nfs unlock wrapper. 3442 */ 3443 int 3444 nfs_unlock(void *v) 3445 { 3446 struct vop_unlock_args /* { 3447 struct vnode *a_vp; 3448 int a_flags; 3449 } */ *ap = v; 3450 struct vnode *vp = ap->a_vp; 3451 3452 /* 3453 * VOP_UNLOCK can be called by nfs_loadattrcache 3454 * with v_data == 0. 3455 */ 3456 if (VTONFS(vp)) { 3457 nfs_delayedtruncate(vp); 3458 } 3459 3460 return genfs_unlock(v); 3461 } 3462 3463 /* 3464 * nfs special file access vnode op. 3465 * Essentially just get vattr and then imitate iaccess() since the device is 3466 * local to the client. 3467 */ 3468 int 3469 nfsspec_access(v) 3470 void *v; 3471 { 3472 struct vop_access_args /* { 3473 struct vnode *a_vp; 3474 int a_mode; 3475 kauth_cred_t a_cred; 3476 struct lwp *a_l; 3477 } */ *ap = v; 3478 struct vattr va; 3479 struct vnode *vp = ap->a_vp; 3480 int error; 3481 3482 error = VOP_GETATTR(vp, &va, ap->a_cred); 3483 if (error) 3484 return (error); 3485 3486 /* 3487 * Disallow write attempts on filesystems mounted read-only; 3488 * unless the file is a socket, fifo, or a block or character 3489 * device resident on the filesystem. 3490 */ 3491 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3492 switch (vp->v_type) { 3493 case VREG: 3494 case VDIR: 3495 case VLNK: 3496 return (EROFS); 3497 default: 3498 break; 3499 } 3500 } 3501 3502 return (vaccess(va.va_type, va.va_mode, 3503 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3504 } 3505 3506 /* 3507 * Read wrapper for special devices. 3508 */ 3509 int 3510 nfsspec_read(v) 3511 void *v; 3512 { 3513 struct vop_read_args /* { 3514 struct vnode *a_vp; 3515 struct uio *a_uio; 3516 int a_ioflag; 3517 kauth_cred_t a_cred; 3518 } */ *ap = v; 3519 struct nfsnode *np = VTONFS(ap->a_vp); 3520 3521 /* 3522 * Set access flag. 3523 */ 3524 np->n_flag |= NACC; 3525 getnanotime(&np->n_atim); 3526 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3527 } 3528 3529 /* 3530 * Write wrapper for special devices. 3531 */ 3532 int 3533 nfsspec_write(v) 3534 void *v; 3535 { 3536 struct vop_write_args /* { 3537 struct vnode *a_vp; 3538 struct uio *a_uio; 3539 int a_ioflag; 3540 kauth_cred_t a_cred; 3541 } */ *ap = v; 3542 struct nfsnode *np = VTONFS(ap->a_vp); 3543 3544 /* 3545 * Set update flag. 3546 */ 3547 np->n_flag |= NUPD; 3548 getnanotime(&np->n_mtim); 3549 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3550 } 3551 3552 /* 3553 * Close wrapper for special devices. 3554 * 3555 * Update the times on the nfsnode then do device close. 3556 */ 3557 int 3558 nfsspec_close(v) 3559 void *v; 3560 { 3561 struct vop_close_args /* { 3562 struct vnode *a_vp; 3563 int a_fflag; 3564 kauth_cred_t a_cred; 3565 struct lwp *a_l; 3566 } */ *ap = v; 3567 struct vnode *vp = ap->a_vp; 3568 struct nfsnode *np = VTONFS(vp); 3569 struct vattr vattr; 3570 3571 if (np->n_flag & (NACC | NUPD)) { 3572 np->n_flag |= NCHG; 3573 if (vp->v_usecount == 1 && 3574 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3575 VATTR_NULL(&vattr); 3576 if (np->n_flag & NACC) 3577 vattr.va_atime = np->n_atim; 3578 if (np->n_flag & NUPD) 3579 vattr.va_mtime = np->n_mtim; 3580 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3581 } 3582 } 3583 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3584 } 3585 3586 /* 3587 * Read wrapper for fifos. 3588 */ 3589 int 3590 nfsfifo_read(v) 3591 void *v; 3592 { 3593 struct vop_read_args /* { 3594 struct vnode *a_vp; 3595 struct uio *a_uio; 3596 int a_ioflag; 3597 kauth_cred_t a_cred; 3598 } */ *ap = v; 3599 struct nfsnode *np = VTONFS(ap->a_vp); 3600 3601 /* 3602 * Set access flag. 3603 */ 3604 np->n_flag |= NACC; 3605 getnanotime(&np->n_atim); 3606 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3607 } 3608 3609 /* 3610 * Write wrapper for fifos. 3611 */ 3612 int 3613 nfsfifo_write(v) 3614 void *v; 3615 { 3616 struct vop_write_args /* { 3617 struct vnode *a_vp; 3618 struct uio *a_uio; 3619 int a_ioflag; 3620 kauth_cred_t a_cred; 3621 } */ *ap = v; 3622 struct nfsnode *np = VTONFS(ap->a_vp); 3623 3624 /* 3625 * Set update flag. 3626 */ 3627 np->n_flag |= NUPD; 3628 getnanotime(&np->n_mtim); 3629 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3630 } 3631 3632 /* 3633 * Close wrapper for fifos. 3634 * 3635 * Update the times on the nfsnode then do fifo close. 3636 */ 3637 int 3638 nfsfifo_close(v) 3639 void *v; 3640 { 3641 struct vop_close_args /* { 3642 struct vnode *a_vp; 3643 int a_fflag; 3644 kauth_cred_t a_cred; 3645 struct lwp *a_l; 3646 } */ *ap = v; 3647 struct vnode *vp = ap->a_vp; 3648 struct nfsnode *np = VTONFS(vp); 3649 struct vattr vattr; 3650 3651 if (np->n_flag & (NACC | NUPD)) { 3652 struct timespec ts; 3653 3654 getnanotime(&ts); 3655 if (np->n_flag & NACC) 3656 np->n_atim = ts; 3657 if (np->n_flag & NUPD) 3658 np->n_mtim = ts; 3659 np->n_flag |= NCHG; 3660 if (vp->v_usecount == 1 && 3661 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3662 VATTR_NULL(&vattr); 3663 if (np->n_flag & NACC) 3664 vattr.va_atime = np->n_atim; 3665 if (np->n_flag & NUPD) 3666 vattr.va_mtime = np->n_mtim; 3667 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3668 } 3669 } 3670 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3671 } 3672
Indexes created Fri Sep 26 08:10:20 GMT 2025