nfs_vnops.c revision 1.253
1 /* $NetBSD: nfs_vnops.c,v 1.253 2007/04/29 08:41:10 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 1.253 2007/04/29 08:41:10 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 struct lwp *a_l; 317 } */ *ap = v; 318 struct vnode *vp = ap->a_vp; 319 #ifndef NFS_V2_ONLY 320 u_int32_t *tl; 321 char *cp; 322 int32_t t1, t2; 323 char *bpos, *dpos, *cp2; 324 int error = 0, attrflag; 325 struct mbuf *mreq, *mrep, *md, *mb; 326 u_int32_t mode, rmode; 327 const int v3 = NFS_ISV3(vp); 328 #endif 329 int cachevalid; 330 struct nfsnode *np = VTONFS(vp); 331 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 332 333 cachevalid = (np->n_accstamp != -1 && 334 (time_uptime - np->n_accstamp) < NFS_ATTRTIMEO(nmp, np) && 335 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 336 337 /* 338 * Check access cache first. If this request has been made for this 339 * uid shortly before, use the cached result. 340 */ 341 if (cachevalid) { 342 if (!np->n_accerror) { 343 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 344 return np->n_accerror; 345 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 346 return np->n_accerror; 347 } 348 349 #ifndef NFS_V2_ONLY 350 /* 351 * For nfs v3, do an access rpc, otherwise you are stuck emulating 352 * ufs_access() locally using the vattr. This may not be correct, 353 * since the server may apply other access criteria such as 354 * client uid-->server uid mapping that we do not know about, but 355 * this is better than just returning anything that is lying about 356 * in the cache. 357 */ 358 if (v3) { 359 nfsstats.rpccnt[NFSPROC_ACCESS]++; 360 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 361 nfsm_fhtom(np, v3); 362 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 363 if (ap->a_mode & VREAD) 364 mode = NFSV3ACCESS_READ; 365 else 366 mode = 0; 367 if (vp->v_type != VDIR) { 368 if (ap->a_mode & VWRITE) 369 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 370 if (ap->a_mode & VEXEC) 371 mode |= NFSV3ACCESS_EXECUTE; 372 } else { 373 if (ap->a_mode & VWRITE) 374 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 375 NFSV3ACCESS_DELETE); 376 if (ap->a_mode & VEXEC) 377 mode |= NFSV3ACCESS_LOOKUP; 378 } 379 *tl = txdr_unsigned(mode); 380 nfsm_request(np, NFSPROC_ACCESS, ap->a_l, ap->a_cred); 381 nfsm_postop_attr(vp, attrflag, 0); 382 if (!error) { 383 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 384 rmode = fxdr_unsigned(u_int32_t, *tl); 385 /* 386 * The NFS V3 spec does not clarify whether or not 387 * the returned access bits can be a superset of 388 * the ones requested, so... 389 */ 390 if ((rmode & mode) != mode) 391 error = EACCES; 392 } 393 nfsm_reqdone; 394 } else 395 #endif 396 return (nfsspec_access(ap)); 397 #ifndef NFS_V2_ONLY 398 /* 399 * Disallow write attempts on filesystems mounted read-only; 400 * unless the file is a socket, fifo, or a block or character 401 * device resident on the filesystem. 402 */ 403 if (!error && (ap->a_mode & VWRITE) && 404 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 405 switch (vp->v_type) { 406 case VREG: 407 case VDIR: 408 case VLNK: 409 error = EROFS; 410 default: 411 break; 412 } 413 } 414 415 if (!error || error == EACCES) { 416 /* 417 * If we got the same result as for a previous, 418 * different request, OR it in. Don't update 419 * the timestamp in that case. 420 */ 421 if (cachevalid && np->n_accstamp != -1 && 422 error == np->n_accerror) { 423 if (!error) 424 np->n_accmode |= ap->a_mode; 425 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 426 np->n_accmode = ap->a_mode; 427 } else { 428 np->n_accstamp = time_uptime; 429 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 430 np->n_accmode = ap->a_mode; 431 np->n_accerror = error; 432 } 433 } 434 435 return (error); 436 #endif 437 } 438 439 /* 440 * nfs open vnode op 441 * Check to see if the type is ok 442 * and that deletion is not in progress. 443 * For paged in text files, you will need to flush the page cache 444 * if consistency is lost. 445 */ 446 /* ARGSUSED */ 447 int 448 nfs_open(v) 449 void *v; 450 { 451 struct vop_open_args /* { 452 struct vnode *a_vp; 453 int a_mode; 454 kauth_cred_t a_cred; 455 struct lwp *a_l; 456 } */ *ap = v; 457 struct vnode *vp = ap->a_vp; 458 struct nfsnode *np = VTONFS(vp); 459 int error; 460 461 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 462 return (EACCES); 463 } 464 465 if (ap->a_mode & FREAD) { 466 if (np->n_rcred != NULL) 467 kauth_cred_free(np->n_rcred); 468 np->n_rcred = ap->a_cred; 469 kauth_cred_hold(np->n_rcred); 470 } 471 if (ap->a_mode & FWRITE) { 472 if (np->n_wcred != NULL) 473 kauth_cred_free(np->n_wcred); 474 np->n_wcred = ap->a_cred; 475 kauth_cred_hold(np->n_wcred); 476 } 477 478 error = nfs_flushstalebuf(vp, ap->a_cred, ap->a_l, 0); 479 if (error) 480 return error; 481 482 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 483 484 return (0); 485 } 486 487 /* 488 * nfs close vnode op 489 * What an NFS client should do upon close after writing is a debatable issue. 490 * Most NFS clients push delayed writes to the server upon close, basically for 491 * two reasons: 492 * 1 - So that any write errors may be reported back to the client process 493 * doing the close system call. By far the two most likely errors are 494 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 495 * 2 - To put a worst case upper bound on cache inconsistency between 496 * multiple clients for the file. 497 * There is also a consistency problem for Version 2 of the protocol w.r.t. 498 * not being able to tell if other clients are writing a file concurrently, 499 * since there is no way of knowing if the changed modify time in the reply 500 * is only due to the write for this client. 501 * (NFS Version 3 provides weak cache consistency data in the reply that 502 * should be sufficient to detect and handle this case.) 503 * 504 * The current code does the following: 505 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 506 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 507 * or commit them (this satisfies 1 and 2 except for the 508 * case where the server crashes after this close but 509 * before the commit RPC, which is felt to be "good 510 * enough". Changing the last argument to nfs_flush() to 511 * a 1 would force a commit operation, if it is felt a 512 * commit is necessary now. 513 */ 514 /* ARGSUSED */ 515 int 516 nfs_close(v) 517 void *v; 518 { 519 struct vop_close_args /* { 520 struct vnodeop_desc *a_desc; 521 struct vnode *a_vp; 522 int a_fflag; 523 kauth_cred_t a_cred; 524 struct lwp *a_l; 525 } */ *ap = v; 526 struct vnode *vp = ap->a_vp; 527 struct nfsnode *np = VTONFS(vp); 528 int error = 0; 529 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 530 531 if (vp->v_type == VREG) { 532 if (np->n_flag & NMODIFIED) { 533 #ifndef NFS_V2_ONLY 534 if (NFS_ISV3(vp)) { 535 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_l, 0); 536 np->n_flag &= ~NMODIFIED; 537 } else 538 #endif 539 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_l, 1); 540 NFS_INVALIDATE_ATTRCACHE(np); 541 } 542 if (np->n_flag & NWRITEERR) { 543 np->n_flag &= ~NWRITEERR; 544 error = np->n_error; 545 } 546 } 547 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 548 return (error); 549 } 550 551 /* 552 * nfs getattr call from vfs. 553 */ 554 int 555 nfs_getattr(v) 556 void *v; 557 { 558 struct vop_getattr_args /* { 559 struct vnode *a_vp; 560 struct vattr *a_vap; 561 kauth_cred_t a_cred; 562 struct lwp *a_l; 563 } */ *ap = v; 564 struct vnode *vp = ap->a_vp; 565 struct nfsnode *np = VTONFS(vp); 566 char *cp; 567 u_int32_t *tl; 568 int32_t t1, t2; 569 char *bpos, *dpos; 570 int error = 0; 571 struct mbuf *mreq, *mrep, *md, *mb; 572 const int v3 = NFS_ISV3(vp); 573 574 /* 575 * Update local times for special files. 576 */ 577 if (np->n_flag & (NACC | NUPD)) 578 np->n_flag |= NCHG; 579 580 /* 581 * if we have delayed truncation, do it now. 582 */ 583 nfs_delayedtruncate(vp); 584 585 /* 586 * First look in the cache. 587 */ 588 if (nfs_getattrcache(vp, ap->a_vap) == 0) 589 return (0); 590 nfsstats.rpccnt[NFSPROC_GETATTR]++; 591 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 592 nfsm_fhtom(np, v3); 593 nfsm_request(np, NFSPROC_GETATTR, ap->a_l, ap->a_cred); 594 if (!error) { 595 nfsm_loadattr(vp, ap->a_vap, 0); 596 if (vp->v_type == VDIR && 597 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 598 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 599 } 600 nfsm_reqdone; 601 return (error); 602 } 603 604 /* 605 * nfs setattr call. 606 */ 607 int 608 nfs_setattr(v) 609 void *v; 610 { 611 struct vop_setattr_args /* { 612 struct vnodeop_desc *a_desc; 613 struct vnode *a_vp; 614 struct vattr *a_vap; 615 kauth_cred_t a_cred; 616 struct lwp *a_l; 617 } */ *ap = v; 618 struct vnode *vp = ap->a_vp; 619 struct nfsnode *np = VTONFS(vp); 620 struct vattr *vap = ap->a_vap; 621 int error = 0; 622 u_quad_t tsize = 0; 623 624 /* 625 * Setting of flags is not supported. 626 */ 627 if (vap->va_flags != VNOVAL) 628 return (EOPNOTSUPP); 629 630 /* 631 * Disallow write attempts if the filesystem is mounted read-only. 632 */ 633 if ((vap->va_uid != (uid_t)VNOVAL || 634 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 635 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 636 (vp->v_mount->mnt_flag & MNT_RDONLY)) 637 return (EROFS); 638 if (vap->va_size != VNOVAL) { 639 switch (vp->v_type) { 640 case VDIR: 641 return (EISDIR); 642 case VCHR: 643 case VBLK: 644 case VSOCK: 645 case VFIFO: 646 if (vap->va_mtime.tv_sec == VNOVAL && 647 vap->va_atime.tv_sec == VNOVAL && 648 vap->va_mode == (mode_t)VNOVAL && 649 vap->va_uid == (uid_t)VNOVAL && 650 vap->va_gid == (gid_t)VNOVAL) 651 return (0); 652 vap->va_size = VNOVAL; 653 break; 654 default: 655 /* 656 * Disallow write attempts if the filesystem is 657 * mounted read-only. 658 */ 659 if (vp->v_mount->mnt_flag & MNT_RDONLY) 660 return (EROFS); 661 genfs_node_wrlock(vp); 662 uvm_vnp_setsize(vp, vap->va_size); 663 tsize = np->n_size; 664 np->n_size = vap->va_size; 665 if (vap->va_size == 0) 666 error = nfs_vinvalbuf(vp, 0, 667 ap->a_cred, ap->a_l, 1); 668 else 669 error = nfs_vinvalbuf(vp, V_SAVE, 670 ap->a_cred, ap->a_l, 1); 671 if (error) { 672 uvm_vnp_setsize(vp, tsize); 673 genfs_node_unlock(vp); 674 return (error); 675 } 676 np->n_vattr->va_size = vap->va_size; 677 } 678 } else { 679 /* 680 * flush files before setattr because a later write of 681 * cached data might change timestamps or reset sugid bits 682 */ 683 if ((vap->va_mtime.tv_sec != VNOVAL || 684 vap->va_atime.tv_sec != VNOVAL || 685 vap->va_mode != VNOVAL) && 686 vp->v_type == VREG && 687 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 688 ap->a_l, 1)) == EINTR) 689 return (error); 690 } 691 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_l); 692 if (vap->va_size != VNOVAL) { 693 if (error) { 694 np->n_size = np->n_vattr->va_size = tsize; 695 uvm_vnp_setsize(vp, np->n_size); 696 } 697 genfs_node_unlock(vp); 698 } 699 VN_KNOTE(vp, NOTE_ATTRIB); 700 return (error); 701 } 702 703 /* 704 * Do an nfs setattr rpc. 705 */ 706 int 707 nfs_setattrrpc(vp, vap, cred, l) 708 struct vnode *vp; 709 struct vattr *vap; 710 kauth_cred_t cred; 711 struct lwp *l; 712 { 713 struct nfsv2_sattr *sp; 714 char *cp; 715 int32_t t1, t2; 716 char *bpos, *dpos; 717 u_int32_t *tl; 718 int error = 0; 719 struct mbuf *mreq, *mrep, *md, *mb; 720 const int v3 = NFS_ISV3(vp); 721 struct nfsnode *np = VTONFS(vp); 722 #ifndef NFS_V2_ONLY 723 int wccflag = NFSV3_WCCRATTR; 724 char *cp2; 725 #endif 726 727 nfsstats.rpccnt[NFSPROC_SETATTR]++; 728 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 729 nfsm_fhtom(np, v3); 730 #ifndef NFS_V2_ONLY 731 if (v3) { 732 nfsm_v3attrbuild(vap, true); 733 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 734 *tl = nfs_false; 735 } else { 736 #endif 737 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 738 if (vap->va_mode == (mode_t)VNOVAL) 739 sp->sa_mode = nfs_xdrneg1; 740 else 741 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 742 if (vap->va_uid == (uid_t)VNOVAL) 743 sp->sa_uid = nfs_xdrneg1; 744 else 745 sp->sa_uid = txdr_unsigned(vap->va_uid); 746 if (vap->va_gid == (gid_t)VNOVAL) 747 sp->sa_gid = nfs_xdrneg1; 748 else 749 sp->sa_gid = txdr_unsigned(vap->va_gid); 750 sp->sa_size = txdr_unsigned(vap->va_size); 751 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 752 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 753 #ifndef NFS_V2_ONLY 754 } 755 #endif 756 nfsm_request(np, NFSPROC_SETATTR, l, cred); 757 #ifndef NFS_V2_ONLY 758 if (v3) { 759 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 760 } else 761 #endif 762 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 763 nfsm_reqdone; 764 return (error); 765 } 766 767 /* 768 * nfs lookup call, one step at a time... 769 * First look in cache 770 * If not found, unlock the directory nfsnode and do the rpc 771 * 772 * This code is full of lock/unlock statements and checks, because 773 * we continue after cache_lookup has finished (we need to check 774 * with the attr cache and do an rpc if it has timed out). This means 775 * that the locking effects of cache_lookup have to be taken into 776 * account. 777 */ 778 int 779 nfs_lookup(v) 780 void *v; 781 { 782 struct vop_lookup_args /* { 783 struct vnodeop_desc *a_desc; 784 struct vnode *a_dvp; 785 struct vnode **a_vpp; 786 struct componentname *a_cnp; 787 } */ *ap = v; 788 struct componentname *cnp = ap->a_cnp; 789 struct vnode *dvp = ap->a_dvp; 790 struct vnode **vpp = ap->a_vpp; 791 int flags; 792 struct vnode *newvp; 793 u_int32_t *tl; 794 char *cp; 795 int32_t t1, t2; 796 char *bpos, *dpos, *cp2; 797 struct mbuf *mreq, *mrep, *md, *mb; 798 long len; 799 nfsfh_t *fhp; 800 struct nfsnode *np; 801 int error = 0, attrflag, fhsize; 802 const int v3 = NFS_ISV3(dvp); 803 804 flags = cnp->cn_flags; 805 806 *vpp = NULLVP; 807 newvp = NULLVP; 808 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 809 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 810 return (EROFS); 811 if (dvp->v_type != VDIR) 812 return (ENOTDIR); 813 814 /* 815 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 816 */ 817 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 818 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp); 819 if (error) 820 return error; 821 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 822 return EISDIR; 823 VREF(dvp); 824 *vpp = dvp; 825 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 826 cnp->cn_flags |= SAVENAME; 827 return 0; 828 } 829 830 np = VTONFS(dvp); 831 832 /* 833 * Before tediously performing a linear scan of the directory, 834 * check the name cache to see if the directory/name pair 835 * we are looking for is known already. 836 * If the directory/name pair is found in the name cache, 837 * we have to ensure the directory has not changed from 838 * the time the cache entry has been created. If it has, 839 * the cache entry has to be ignored. 840 */ 841 error = cache_lookup_raw(dvp, vpp, cnp); 842 KASSERT(dvp != *vpp); 843 if (error >= 0) { 844 struct vattr vattr; 845 int err2; 846 847 if (error && error != ENOENT) { 848 *vpp = NULLVP; 849 return error; 850 } 851 852 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp); 853 if (err2 != 0) { 854 if (error == 0) 855 vrele(*vpp); 856 *vpp = NULLVP; 857 return err2; 858 } 859 860 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred, 861 cnp->cn_lwp) || timespeccmp(&vattr.va_mtime, 862 &VTONFS(dvp)->n_nctime, !=)) { 863 if (error == 0) { 864 vrele(*vpp); 865 *vpp = NULLVP; 866 } 867 cache_purge1(dvp, NULL, PURGE_CHILDREN); 868 timespecclear(&np->n_nctime); 869 goto dorpc; 870 } 871 872 if (error == ENOENT) { 873 goto noentry; 874 } 875 876 newvp = *vpp; 877 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp) 878 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 879 nfsstats.lookupcache_hits++; 880 if ((flags & ISDOTDOT) != 0) { 881 VOP_UNLOCK(dvp, 0); 882 } 883 error = vn_lock(newvp, LK_EXCLUSIVE); 884 if ((flags & ISDOTDOT) != 0) { 885 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 886 } 887 if (error) { 888 /* newvp has been revoked. */ 889 vrele(newvp); 890 *vpp = NULL; 891 return error; 892 } 893 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 894 cnp->cn_flags |= SAVENAME; 895 KASSERT(newvp->v_type != VNON); 896 return (0); 897 } 898 cache_purge1(newvp, NULL, PURGE_PARENTS); 899 vrele(newvp); 900 *vpp = NULLVP; 901 } 902 dorpc: 903 #if 0 904 /* 905 * because nfsv3 has the same CREATE semantics as ours, 906 * we don't have to perform LOOKUPs beforehand. 907 * 908 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 909 * XXX although we have no way to know if O_EXCL is requested or not. 910 */ 911 912 if (v3 && cnp->cn_nameiop == CREATE && 913 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 914 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 915 cnp->cn_flags |= SAVENAME; 916 return (EJUSTRETURN); 917 } 918 #endif /* 0 */ 919 920 error = 0; 921 newvp = NULLVP; 922 nfsstats.lookupcache_misses++; 923 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 924 len = cnp->cn_namelen; 925 nfsm_reqhead(np, NFSPROC_LOOKUP, 926 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 927 nfsm_fhtom(np, v3); 928 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 929 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_lwp, cnp->cn_cred); 930 if (error) { 931 nfsm_postop_attr(dvp, attrflag, 0); 932 m_freem(mrep); 933 goto nfsmout; 934 } 935 nfsm_getfh(fhp, fhsize, v3); 936 937 /* 938 * Handle RENAME case... 939 */ 940 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 941 if (NFS_CMPFH(np, fhp, fhsize)) { 942 m_freem(mrep); 943 return (EISDIR); 944 } 945 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 946 if (error) { 947 m_freem(mrep); 948 return error; 949 } 950 newvp = NFSTOV(np); 951 #ifndef NFS_V2_ONLY 952 if (v3) { 953 nfsm_postop_attr(newvp, attrflag, 0); 954 nfsm_postop_attr(dvp, attrflag, 0); 955 } else 956 #endif 957 nfsm_loadattr(newvp, (struct vattr *)0, 0); 958 *vpp = newvp; 959 m_freem(mrep); 960 cnp->cn_flags |= SAVENAME; 961 goto validate; 962 } 963 964 /* 965 * The postop attr handling is duplicated for each if case, 966 * because it should be done while dvp is locked (unlocking 967 * dvp is different for each case). 968 */ 969 970 if (NFS_CMPFH(np, fhp, fhsize)) { 971 /* 972 * "." lookup 973 */ 974 VREF(dvp); 975 newvp = dvp; 976 #ifndef NFS_V2_ONLY 977 if (v3) { 978 nfsm_postop_attr(newvp, attrflag, 0); 979 nfsm_postop_attr(dvp, attrflag, 0); 980 } else 981 #endif 982 nfsm_loadattr(newvp, (struct vattr *)0, 0); 983 } else if (flags & ISDOTDOT) { 984 /* 985 * ".." lookup 986 */ 987 VOP_UNLOCK(dvp, 0); 988 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 989 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 990 if (error) { 991 m_freem(mrep); 992 return error; 993 } 994 newvp = NFSTOV(np); 995 996 #ifndef NFS_V2_ONLY 997 if (v3) { 998 nfsm_postop_attr(newvp, attrflag, 0); 999 nfsm_postop_attr(dvp, attrflag, 0); 1000 } else 1001 #endif 1002 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1003 } else { 1004 /* 1005 * Other lookups. 1006 */ 1007 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1008 if (error) { 1009 m_freem(mrep); 1010 return error; 1011 } 1012 newvp = NFSTOV(np); 1013 #ifndef NFS_V2_ONLY 1014 if (v3) { 1015 nfsm_postop_attr(newvp, attrflag, 0); 1016 nfsm_postop_attr(dvp, attrflag, 0); 1017 } else 1018 #endif 1019 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1020 } 1021 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1022 cnp->cn_flags |= SAVENAME; 1023 if ((cnp->cn_flags & MAKEENTRY) && 1024 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1025 nfs_cache_enter(dvp, newvp, cnp); 1026 } 1027 *vpp = newvp; 1028 nfsm_reqdone; 1029 if (error) { 1030 /* 1031 * We get here only because of errors returned by 1032 * the RPC. Otherwise we'll have returned above 1033 * (the nfsm_* macros will jump to nfsm_reqdone 1034 * on error). 1035 */ 1036 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1037 cnp->cn_nameiop != CREATE) { 1038 nfs_cache_enter(dvp, NULL, cnp); 1039 } 1040 if (newvp != NULLVP) { 1041 if (newvp == dvp) { 1042 vrele(newvp); 1043 } else { 1044 vput(newvp); 1045 } 1046 } 1047 noentry: 1048 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1049 (flags & ISLASTCN) && error == ENOENT) { 1050 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1051 error = EROFS; 1052 } else { 1053 error = EJUSTRETURN; 1054 cnp->cn_flags |= SAVENAME; 1055 } 1056 } 1057 *vpp = NULL; 1058 return error; 1059 } 1060 1061 validate: 1062 /* 1063 * make sure we have valid type and size. 1064 */ 1065 1066 newvp = *vpp; 1067 if (newvp->v_type == VNON) { 1068 struct vattr vattr; /* dummy */ 1069 1070 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1071 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_lwp); 1072 if (error) { 1073 vput(newvp); 1074 *vpp = NULL; 1075 } 1076 } 1077 1078 return error; 1079 } 1080 1081 /* 1082 * nfs read call. 1083 * Just call nfs_bioread() to do the work. 1084 */ 1085 int 1086 nfs_read(v) 1087 void *v; 1088 { 1089 struct vop_read_args /* { 1090 struct vnode *a_vp; 1091 struct uio *a_uio; 1092 int a_ioflag; 1093 kauth_cred_t a_cred; 1094 } */ *ap = v; 1095 struct vnode *vp = ap->a_vp; 1096 1097 if (vp->v_type != VREG) 1098 return EISDIR; 1099 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1100 } 1101 1102 /* 1103 * nfs readlink call 1104 */ 1105 int 1106 nfs_readlink(v) 1107 void *v; 1108 { 1109 struct vop_readlink_args /* { 1110 struct vnode *a_vp; 1111 struct uio *a_uio; 1112 kauth_cred_t a_cred; 1113 } */ *ap = v; 1114 struct vnode *vp = ap->a_vp; 1115 struct nfsnode *np = VTONFS(vp); 1116 1117 if (vp->v_type != VLNK) 1118 return (EPERM); 1119 1120 if (np->n_rcred != NULL) { 1121 kauth_cred_free(np->n_rcred); 1122 } 1123 np->n_rcred = ap->a_cred; 1124 kauth_cred_hold(np->n_rcred); 1125 1126 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1127 } 1128 1129 /* 1130 * Do a readlink rpc. 1131 * Called by nfs_doio() from below the buffer cache. 1132 */ 1133 int 1134 nfs_readlinkrpc(vp, uiop, cred) 1135 struct vnode *vp; 1136 struct uio *uiop; 1137 kauth_cred_t cred; 1138 { 1139 u_int32_t *tl; 1140 char *cp; 1141 int32_t t1, t2; 1142 char *bpos, *dpos, *cp2; 1143 int error = 0; 1144 uint32_t len; 1145 struct mbuf *mreq, *mrep, *md, *mb; 1146 const int v3 = NFS_ISV3(vp); 1147 struct nfsnode *np = VTONFS(vp); 1148 #ifndef NFS_V2_ONLY 1149 int attrflag; 1150 #endif 1151 1152 nfsstats.rpccnt[NFSPROC_READLINK]++; 1153 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1154 nfsm_fhtom(np, v3); 1155 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1156 #ifndef NFS_V2_ONLY 1157 if (v3) 1158 nfsm_postop_attr(vp, attrflag, 0); 1159 #endif 1160 if (!error) { 1161 #ifndef NFS_V2_ONLY 1162 if (v3) { 1163 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1164 len = fxdr_unsigned(uint32_t, *tl); 1165 if (len > MAXPATHLEN) { 1166 /* 1167 * this pathname is too long for us. 1168 */ 1169 m_freem(mrep); 1170 /* Solaris returns EINVAL. should we follow? */ 1171 error = ENAMETOOLONG; 1172 goto nfsmout; 1173 } 1174 } else 1175 #endif 1176 { 1177 nfsm_strsiz(len, NFS_MAXPATHLEN); 1178 } 1179 nfsm_mtouio(uiop, len); 1180 } 1181 nfsm_reqdone; 1182 return (error); 1183 } 1184 1185 /* 1186 * nfs read rpc call 1187 * Ditto above 1188 */ 1189 int 1190 nfs_readrpc(vp, uiop) 1191 struct vnode *vp; 1192 struct uio *uiop; 1193 { 1194 u_int32_t *tl; 1195 char *cp; 1196 int32_t t1, t2; 1197 char *bpos, *dpos, *cp2; 1198 struct mbuf *mreq, *mrep, *md, *mb; 1199 struct nfsmount *nmp; 1200 int error = 0, len, retlen, tsiz, eof, byte_count; 1201 const int v3 = NFS_ISV3(vp); 1202 struct nfsnode *np = VTONFS(vp); 1203 #ifndef NFS_V2_ONLY 1204 int attrflag; 1205 #endif 1206 1207 #ifndef nolint 1208 eof = 0; 1209 #endif 1210 nmp = VFSTONFS(vp->v_mount); 1211 tsiz = uiop->uio_resid; 1212 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1213 return (EFBIG); 1214 iostat_busy(nmp->nm_stats); 1215 byte_count = 0; /* count bytes actually transferred */ 1216 while (tsiz > 0) { 1217 nfsstats.rpccnt[NFSPROC_READ]++; 1218 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1219 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1220 nfsm_fhtom(np, v3); 1221 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1222 #ifndef NFS_V2_ONLY 1223 if (v3) { 1224 txdr_hyper(uiop->uio_offset, tl); 1225 *(tl + 2) = txdr_unsigned(len); 1226 } else 1227 #endif 1228 { 1229 *tl++ = txdr_unsigned(uiop->uio_offset); 1230 *tl++ = txdr_unsigned(len); 1231 *tl = 0; 1232 } 1233 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1234 #ifndef NFS_V2_ONLY 1235 if (v3) { 1236 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1237 if (error) { 1238 m_freem(mrep); 1239 goto nfsmout; 1240 } 1241 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1242 eof = fxdr_unsigned(int, *(tl + 1)); 1243 } else 1244 #endif 1245 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1246 nfsm_strsiz(retlen, nmp->nm_rsize); 1247 nfsm_mtouio(uiop, retlen); 1248 m_freem(mrep); 1249 tsiz -= retlen; 1250 byte_count += retlen; 1251 #ifndef NFS_V2_ONLY 1252 if (v3) { 1253 if (eof || retlen == 0) 1254 tsiz = 0; 1255 } else 1256 #endif 1257 if (retlen < len) 1258 tsiz = 0; 1259 } 1260 nfsmout: 1261 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1262 return (error); 1263 } 1264 1265 struct nfs_writerpc_context { 1266 kmutex_t nwc_lock; 1267 kcondvar_t nwc_cv; 1268 int nwc_mbufcount; 1269 }; 1270 1271 /* 1272 * free mbuf used to refer protected pages while write rpc call. 1273 * called at splvm. 1274 */ 1275 static void 1276 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1277 { 1278 struct nfs_writerpc_context *ctx = arg; 1279 1280 KASSERT(m != NULL); 1281 KASSERT(ctx != NULL); 1282 pool_cache_put(&mbpool_cache, m); 1283 mutex_enter(&ctx->nwc_lock); 1284 if (--ctx->nwc_mbufcount == 0) { 1285 cv_signal(&ctx->nwc_cv); 1286 } 1287 mutex_exit(&ctx->nwc_lock); 1288 } 1289 1290 /* 1291 * nfs write call 1292 */ 1293 int 1294 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1295 struct vnode *vp; 1296 struct uio *uiop; 1297 int *iomode; 1298 bool pageprotected; 1299 bool *stalewriteverfp; 1300 { 1301 u_int32_t *tl; 1302 char *cp; 1303 int32_t t1, t2; 1304 char *bpos, *dpos; 1305 struct mbuf *mreq, *mrep, *md, *mb; 1306 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1307 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1308 const int v3 = NFS_ISV3(vp); 1309 int committed = NFSV3WRITE_FILESYNC; 1310 struct nfsnode *np = VTONFS(vp); 1311 struct nfs_writerpc_context ctx; 1312 int byte_count; 1313 struct lwp *l = NULL; 1314 size_t origresid; 1315 #ifndef NFS_V2_ONLY 1316 char *cp2; 1317 int rlen, commit; 1318 #endif 1319 1320 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1321 cv_init(&ctx.nwc_cv, "nfsmblk"); 1322 ctx.nwc_mbufcount = 1; 1323 1324 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1325 panic("writerpc readonly vp %p", vp); 1326 } 1327 1328 #ifdef DIAGNOSTIC 1329 if (uiop->uio_iovcnt != 1) 1330 panic("nfs: writerpc iovcnt > 1"); 1331 #endif 1332 tsiz = uiop->uio_resid; 1333 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1334 return (EFBIG); 1335 if (pageprotected) { 1336 l = curlwp; 1337 PHOLD(l); 1338 } 1339 retry: 1340 origresid = uiop->uio_resid; 1341 KASSERT(origresid == uiop->uio_iov->iov_len); 1342 iostat_busy(nmp->nm_stats); 1343 byte_count = 0; /* count of bytes actually written */ 1344 while (tsiz > 0) { 1345 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1346 uint32_t backup; 1347 bool stalewriteverf = false; 1348 1349 nfsstats.rpccnt[NFSPROC_WRITE]++; 1350 len = min(tsiz, nmp->nm_wsize); 1351 datalen = pageprotected ? 0 : nfsm_rndup(len); 1352 nfsm_reqhead(np, NFSPROC_WRITE, 1353 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1354 nfsm_fhtom(np, v3); 1355 #ifndef NFS_V2_ONLY 1356 if (v3) { 1357 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1358 txdr_hyper(uiop->uio_offset, tl); 1359 tl += 2; 1360 *tl++ = txdr_unsigned(len); 1361 *tl++ = txdr_unsigned(*iomode); 1362 *tl = txdr_unsigned(len); 1363 } else 1364 #endif 1365 { 1366 u_int32_t x; 1367 1368 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1369 /* Set both "begin" and "current" to non-garbage. */ 1370 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1371 *tl++ = x; /* "begin offset" */ 1372 *tl++ = x; /* "current offset" */ 1373 x = txdr_unsigned(len); 1374 *tl++ = x; /* total to this offset */ 1375 *tl = x; /* size of this write */ 1376 1377 } 1378 if (pageprotected) { 1379 /* 1380 * since we know pages can't be modified during i/o, 1381 * no need to copy them for us. 1382 */ 1383 struct mbuf *m; 1384 struct iovec *iovp = uiop->uio_iov; 1385 1386 m = m_get(M_WAIT, MT_DATA); 1387 MCLAIM(m, &nfs_mowner); 1388 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1389 nfs_writerpc_extfree, &ctx); 1390 m->m_flags |= M_EXT_ROMAP; 1391 m->m_len = len; 1392 mb->m_next = m; 1393 /* 1394 * no need to maintain mb and bpos here 1395 * because no one care them later. 1396 */ 1397 #if 0 1398 mb = m; 1399 bpos = mtod(void *, mb) + mb->m_len; 1400 #endif 1401 UIO_ADVANCE(uiop, len); 1402 uiop->uio_offset += len; 1403 mutex_enter(&ctx.nwc_lock); 1404 ctx.nwc_mbufcount++; 1405 mutex_exit(&ctx.nwc_lock); 1406 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1407 } else { 1408 nfsm_uiotom(uiop, len); 1409 } 1410 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1411 #ifndef NFS_V2_ONLY 1412 if (v3) { 1413 wccflag = NFSV3_WCCCHK; 1414 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1415 if (!error) { 1416 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1417 + NFSX_V3WRITEVERF); 1418 rlen = fxdr_unsigned(int, *tl++); 1419 if (rlen == 0) { 1420 error = NFSERR_IO; 1421 m_freem(mrep); 1422 break; 1423 } else if (rlen < len) { 1424 backup = len - rlen; 1425 UIO_ADVANCE(uiop, -backup); 1426 uiop->uio_offset -= backup; 1427 len = rlen; 1428 } 1429 commit = fxdr_unsigned(int, *tl++); 1430 1431 /* 1432 * Return the lowest committment level 1433 * obtained by any of the RPCs. 1434 */ 1435 if (committed == NFSV3WRITE_FILESYNC) 1436 committed = commit; 1437 else if (committed == NFSV3WRITE_DATASYNC && 1438 commit == NFSV3WRITE_UNSTABLE) 1439 committed = commit; 1440 simple_lock(&nmp->nm_slock); 1441 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1442 memcpy(nmp->nm_writeverf, tl, 1443 NFSX_V3WRITEVERF); 1444 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1445 } else if ((nmp->nm_iflag & 1446 NFSMNT_STALEWRITEVERF) || 1447 memcmp(tl, nmp->nm_writeverf, 1448 NFSX_V3WRITEVERF)) { 1449 memcpy(nmp->nm_writeverf, tl, 1450 NFSX_V3WRITEVERF); 1451 /* 1452 * note NFSMNT_STALEWRITEVERF 1453 * if we're the first thread to 1454 * notice it. 1455 */ 1456 if ((nmp->nm_iflag & 1457 NFSMNT_STALEWRITEVERF) == 0) { 1458 stalewriteverf = true; 1459 nmp->nm_iflag |= 1460 NFSMNT_STALEWRITEVERF; 1461 } 1462 } 1463 simple_unlock(&nmp->nm_slock); 1464 } 1465 } else 1466 #endif 1467 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1468 if (wccflag) 1469 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1470 m_freem(mrep); 1471 if (error) 1472 break; 1473 tsiz -= len; 1474 byte_count += len; 1475 if (stalewriteverf) { 1476 *stalewriteverfp = true; 1477 stalewriteverf = false; 1478 if (committed == NFSV3WRITE_UNSTABLE && 1479 len != origresid) { 1480 /* 1481 * if our write requests weren't atomic but 1482 * unstable, datas in previous iterations 1483 * might have already been lost now. 1484 * then, we should resend them to nfsd. 1485 */ 1486 backup = origresid - tsiz; 1487 UIO_ADVANCE(uiop, -backup); 1488 uiop->uio_offset -= backup; 1489 tsiz = origresid; 1490 goto retry; 1491 } 1492 } 1493 } 1494 nfsmout: 1495 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1496 if (pageprotected) { 1497 /* 1498 * wait until mbufs go away. 1499 * retransmitted mbufs can survive longer than rpc requests 1500 * themselves. 1501 */ 1502 mutex_enter(&ctx.nwc_lock); 1503 ctx.nwc_mbufcount--; 1504 while (ctx.nwc_mbufcount > 0) { 1505 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1506 } 1507 mutex_exit(&ctx.nwc_lock); 1508 PRELE(l); 1509 } 1510 *iomode = committed; 1511 if (error) 1512 uiop->uio_resid = tsiz; 1513 return (error); 1514 } 1515 1516 /* 1517 * nfs mknod rpc 1518 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1519 * mode set to specify the file type and the size field for rdev. 1520 */ 1521 int 1522 nfs_mknodrpc(dvp, vpp, cnp, vap) 1523 struct vnode *dvp; 1524 struct vnode **vpp; 1525 struct componentname *cnp; 1526 struct vattr *vap; 1527 { 1528 struct nfsv2_sattr *sp; 1529 u_int32_t *tl; 1530 char *cp; 1531 int32_t t1, t2; 1532 struct vnode *newvp = (struct vnode *)0; 1533 struct nfsnode *dnp, *np; 1534 char *cp2; 1535 char *bpos, *dpos; 1536 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1537 struct mbuf *mreq, *mrep, *md, *mb; 1538 u_int32_t rdev; 1539 const int v3 = NFS_ISV3(dvp); 1540 1541 if (vap->va_type == VCHR || vap->va_type == VBLK) 1542 rdev = txdr_unsigned(vap->va_rdev); 1543 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1544 rdev = nfs_xdrneg1; 1545 else { 1546 VOP_ABORTOP(dvp, cnp); 1547 vput(dvp); 1548 return (EOPNOTSUPP); 1549 } 1550 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1551 dnp = VTONFS(dvp); 1552 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1553 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1554 nfsm_fhtom(dnp, v3); 1555 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1556 #ifndef NFS_V2_ONLY 1557 if (v3) { 1558 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1559 *tl++ = vtonfsv3_type(vap->va_type); 1560 nfsm_v3attrbuild(vap, false); 1561 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1562 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1563 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1564 *tl = txdr_unsigned(minor(vap->va_rdev)); 1565 } 1566 } else 1567 #endif 1568 { 1569 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1570 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1571 sp->sa_uid = nfs_xdrneg1; 1572 sp->sa_gid = nfs_xdrneg1; 1573 sp->sa_size = rdev; 1574 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1575 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1576 } 1577 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_lwp, cnp->cn_cred); 1578 if (!error) { 1579 nfsm_mtofh(dvp, newvp, v3, gotvp); 1580 if (!gotvp) { 1581 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1582 cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np); 1583 if (!error) 1584 newvp = NFSTOV(np); 1585 } 1586 } 1587 #ifndef NFS_V2_ONLY 1588 if (v3) 1589 nfsm_wcc_data(dvp, wccflag, 0, !error); 1590 #endif 1591 nfsm_reqdone; 1592 if (error) { 1593 if (newvp) 1594 vput(newvp); 1595 } else { 1596 if (cnp->cn_flags & MAKEENTRY) 1597 nfs_cache_enter(dvp, newvp, cnp); 1598 *vpp = newvp; 1599 } 1600 PNBUF_PUT(cnp->cn_pnbuf); 1601 VTONFS(dvp)->n_flag |= NMODIFIED; 1602 if (!wccflag) 1603 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1604 vput(dvp); 1605 return (error); 1606 } 1607 1608 /* 1609 * nfs mknod vop 1610 * just call nfs_mknodrpc() to do the work. 1611 */ 1612 /* ARGSUSED */ 1613 int 1614 nfs_mknod(v) 1615 void *v; 1616 { 1617 struct vop_mknod_args /* { 1618 struct vnode *a_dvp; 1619 struct vnode **a_vpp; 1620 struct componentname *a_cnp; 1621 struct vattr *a_vap; 1622 } */ *ap = v; 1623 struct vnode *dvp = ap->a_dvp; 1624 struct componentname *cnp = ap->a_cnp; 1625 int error; 1626 1627 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1628 VN_KNOTE(dvp, NOTE_WRITE); 1629 if (error == 0 || error == EEXIST) 1630 cache_purge1(dvp, cnp, 0); 1631 return (error); 1632 } 1633 1634 #ifndef NFS_V2_ONLY 1635 static u_long create_verf; 1636 #endif 1637 /* 1638 * nfs file create call 1639 */ 1640 int 1641 nfs_create(v) 1642 void *v; 1643 { 1644 struct vop_create_args /* { 1645 struct vnode *a_dvp; 1646 struct vnode **a_vpp; 1647 struct componentname *a_cnp; 1648 struct vattr *a_vap; 1649 } */ *ap = v; 1650 struct vnode *dvp = ap->a_dvp; 1651 struct vattr *vap = ap->a_vap; 1652 struct componentname *cnp = ap->a_cnp; 1653 struct nfsv2_sattr *sp; 1654 u_int32_t *tl; 1655 char *cp; 1656 int32_t t1, t2; 1657 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1658 struct vnode *newvp = (struct vnode *)0; 1659 char *bpos, *dpos, *cp2; 1660 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1661 struct mbuf *mreq, *mrep, *md, *mb; 1662 const int v3 = NFS_ISV3(dvp); 1663 1664 /* 1665 * Oops, not for me.. 1666 */ 1667 if (vap->va_type == VSOCK) 1668 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1669 1670 KASSERT(vap->va_type == VREG); 1671 1672 #ifdef VA_EXCLUSIVE 1673 if (vap->va_vaflags & VA_EXCLUSIVE) 1674 fmode |= O_EXCL; 1675 #endif 1676 again: 1677 error = 0; 1678 nfsstats.rpccnt[NFSPROC_CREATE]++; 1679 dnp = VTONFS(dvp); 1680 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1681 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1682 nfsm_fhtom(dnp, v3); 1683 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1684 #ifndef NFS_V2_ONLY 1685 if (v3) { 1686 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1687 if (fmode & O_EXCL) { 1688 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1689 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1690 #ifdef INET 1691 if (TAILQ_FIRST(&in_ifaddrhead)) 1692 *tl++ = TAILQ_FIRST(&in_ifaddrhead)-> 1693 ia_addr.sin_addr.s_addr; 1694 else 1695 *tl++ = create_verf; 1696 #else 1697 *tl++ = create_verf; 1698 #endif 1699 *tl = ++create_verf; 1700 } else { 1701 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1702 nfsm_v3attrbuild(vap, false); 1703 } 1704 } else 1705 #endif 1706 { 1707 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1708 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1709 sp->sa_uid = nfs_xdrneg1; 1710 sp->sa_gid = nfs_xdrneg1; 1711 sp->sa_size = 0; 1712 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1713 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1714 } 1715 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_lwp, cnp->cn_cred); 1716 if (!error) { 1717 nfsm_mtofh(dvp, newvp, v3, gotvp); 1718 if (!gotvp) { 1719 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1720 cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp, &np); 1721 if (!error) 1722 newvp = NFSTOV(np); 1723 } 1724 } 1725 #ifndef NFS_V2_ONLY 1726 if (v3) 1727 nfsm_wcc_data(dvp, wccflag, 0, !error); 1728 #endif 1729 nfsm_reqdone; 1730 if (error) { 1731 /* 1732 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1733 */ 1734 if (v3 && (fmode & O_EXCL) && error == ENOTSUP) { 1735 fmode &= ~O_EXCL; 1736 goto again; 1737 } 1738 } else if (v3 && (fmode & O_EXCL)) { 1739 struct timespec ts; 1740 1741 getnanotime(&ts); 1742 1743 /* 1744 * make sure that we'll update timestamps as 1745 * most server implementations use them to store 1746 * the create verifier. 1747 * 1748 * XXX it's better to use TOSERVER always. 1749 */ 1750 1751 if (vap->va_atime.tv_sec == VNOVAL) 1752 vap->va_atime = ts; 1753 if (vap->va_mtime.tv_sec == VNOVAL) 1754 vap->va_mtime = ts; 1755 1756 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_lwp); 1757 } 1758 if (error == 0) { 1759 if (cnp->cn_flags & MAKEENTRY) 1760 nfs_cache_enter(dvp, newvp, cnp); 1761 else 1762 cache_purge1(dvp, cnp, 0); 1763 *ap->a_vpp = newvp; 1764 } else { 1765 if (newvp) 1766 vput(newvp); 1767 if (error == EEXIST) 1768 cache_purge1(dvp, cnp, 0); 1769 } 1770 PNBUF_PUT(cnp->cn_pnbuf); 1771 VTONFS(dvp)->n_flag |= NMODIFIED; 1772 if (!wccflag) 1773 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1774 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1775 vput(dvp); 1776 return (error); 1777 } 1778 1779 /* 1780 * nfs file remove call 1781 * To try and make nfs semantics closer to ufs semantics, a file that has 1782 * other processes using the vnode is renamed instead of removed and then 1783 * removed later on the last close. 1784 * - If v_usecount > 1 1785 * If a rename is not already in the works 1786 * call nfs_sillyrename() to set it up 1787 * else 1788 * do the remove rpc 1789 */ 1790 int 1791 nfs_remove(v) 1792 void *v; 1793 { 1794 struct vop_remove_args /* { 1795 struct vnodeop_desc *a_desc; 1796 struct vnode * a_dvp; 1797 struct vnode * a_vp; 1798 struct componentname * a_cnp; 1799 } */ *ap = v; 1800 struct vnode *vp = ap->a_vp; 1801 struct vnode *dvp = ap->a_dvp; 1802 struct componentname *cnp = ap->a_cnp; 1803 struct nfsnode *np = VTONFS(vp); 1804 int error = 0; 1805 struct vattr vattr; 1806 1807 #ifndef DIAGNOSTIC 1808 if ((cnp->cn_flags & HASBUF) == 0) 1809 panic("nfs_remove: no name"); 1810 if (vp->v_usecount < 1) 1811 panic("nfs_remove: bad v_usecount"); 1812 #endif 1813 if (vp->v_type == VDIR) 1814 error = EPERM; 1815 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1816 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_lwp) == 0 && 1817 vattr.va_nlink > 1)) { 1818 /* 1819 * Purge the name cache so that the chance of a lookup for 1820 * the name succeeding while the remove is in progress is 1821 * minimized. Without node locking it can still happen, such 1822 * that an I/O op returns ESTALE, but since you get this if 1823 * another host removes the file.. 1824 */ 1825 cache_purge(vp); 1826 /* 1827 * throw away biocache buffers, mainly to avoid 1828 * unnecessary delayed writes later. 1829 */ 1830 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_lwp, 1); 1831 /* Do the rpc */ 1832 if (error != EINTR) 1833 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1834 cnp->cn_namelen, cnp->cn_cred, cnp->cn_lwp); 1835 } else if (!np->n_sillyrename) 1836 error = nfs_sillyrename(dvp, vp, cnp, false); 1837 PNBUF_PUT(cnp->cn_pnbuf); 1838 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1839 vattr.va_nlink == 1) { 1840 np->n_flag |= NREMOVED; 1841 } 1842 NFS_INVALIDATE_ATTRCACHE(np); 1843 VN_KNOTE(vp, NOTE_DELETE); 1844 VN_KNOTE(dvp, NOTE_WRITE); 1845 if (dvp == vp) 1846 vrele(vp); 1847 else 1848 vput(vp); 1849 vput(dvp); 1850 return (error); 1851 } 1852 1853 /* 1854 * nfs file remove rpc called from nfs_inactive 1855 */ 1856 int 1857 nfs_removeit(sp) 1858 struct sillyrename *sp; 1859 { 1860 1861 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1862 (struct lwp *)0)); 1863 } 1864 1865 /* 1866 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1867 */ 1868 int 1869 nfs_removerpc(dvp, name, namelen, cred, l) 1870 struct vnode *dvp; 1871 const char *name; 1872 int namelen; 1873 kauth_cred_t cred; 1874 struct lwp *l; 1875 { 1876 u_int32_t *tl; 1877 char *cp; 1878 #ifndef NFS_V2_ONLY 1879 int32_t t1; 1880 char *cp2; 1881 #endif 1882 int32_t t2; 1883 char *bpos, *dpos; 1884 int error = 0, wccflag = NFSV3_WCCRATTR; 1885 struct mbuf *mreq, *mrep, *md, *mb; 1886 const int v3 = NFS_ISV3(dvp); 1887 int rexmit = 0; 1888 struct nfsnode *dnp = VTONFS(dvp); 1889 1890 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1891 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1892 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1893 nfsm_fhtom(dnp, v3); 1894 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1895 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1896 #ifndef NFS_V2_ONLY 1897 if (v3) 1898 nfsm_wcc_data(dvp, wccflag, 0, !error); 1899 #endif 1900 nfsm_reqdone; 1901 VTONFS(dvp)->n_flag |= NMODIFIED; 1902 if (!wccflag) 1903 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1904 /* 1905 * Kludge City: If the first reply to the remove rpc is lost.. 1906 * the reply to the retransmitted request will be ENOENT 1907 * since the file was in fact removed 1908 * Therefore, we cheat and return success. 1909 */ 1910 if (rexmit && error == ENOENT) 1911 error = 0; 1912 return (error); 1913 } 1914 1915 /* 1916 * nfs file rename call 1917 */ 1918 int 1919 nfs_rename(v) 1920 void *v; 1921 { 1922 struct vop_rename_args /* { 1923 struct vnode *a_fdvp; 1924 struct vnode *a_fvp; 1925 struct componentname *a_fcnp; 1926 struct vnode *a_tdvp; 1927 struct vnode *a_tvp; 1928 struct componentname *a_tcnp; 1929 } */ *ap = v; 1930 struct vnode *fvp = ap->a_fvp; 1931 struct vnode *tvp = ap->a_tvp; 1932 struct vnode *fdvp = ap->a_fdvp; 1933 struct vnode *tdvp = ap->a_tdvp; 1934 struct componentname *tcnp = ap->a_tcnp; 1935 struct componentname *fcnp = ap->a_fcnp; 1936 int error; 1937 1938 #ifndef DIAGNOSTIC 1939 if ((tcnp->cn_flags & HASBUF) == 0 || 1940 (fcnp->cn_flags & HASBUF) == 0) 1941 panic("nfs_rename: no name"); 1942 #endif 1943 /* Check for cross-device rename */ 1944 if ((fvp->v_mount != tdvp->v_mount) || 1945 (tvp && (fvp->v_mount != tvp->v_mount))) { 1946 error = EXDEV; 1947 goto out; 1948 } 1949 1950 /* 1951 * If the tvp exists and is in use, sillyrename it before doing the 1952 * rename of the new file over it. 1953 * 1954 * Have sillyrename use link instead of rename if possible, 1955 * so that we don't lose the file if the rename fails, and so 1956 * that there's no window when the "to" file doesn't exist. 1957 */ 1958 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1959 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1960 VN_KNOTE(tvp, NOTE_DELETE); 1961 vput(tvp); 1962 tvp = NULL; 1963 } 1964 1965 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1966 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1967 tcnp->cn_lwp); 1968 1969 VN_KNOTE(fdvp, NOTE_WRITE); 1970 VN_KNOTE(tdvp, NOTE_WRITE); 1971 if (error == 0 || error == EEXIST) { 1972 if (fvp->v_type == VDIR) 1973 cache_purge(fvp); 1974 else 1975 cache_purge1(fdvp, fcnp, 0); 1976 if (tvp != NULL && tvp->v_type == VDIR) 1977 cache_purge(tvp); 1978 else 1979 cache_purge1(tdvp, tcnp, 0); 1980 } 1981 out: 1982 if (tdvp == tvp) 1983 vrele(tdvp); 1984 else 1985 vput(tdvp); 1986 if (tvp) 1987 vput(tvp); 1988 vrele(fdvp); 1989 vrele(fvp); 1990 return (error); 1991 } 1992 1993 /* 1994 * nfs file rename rpc called from nfs_remove() above 1995 */ 1996 int 1997 nfs_renameit(sdvp, scnp, sp) 1998 struct vnode *sdvp; 1999 struct componentname *scnp; 2000 struct sillyrename *sp; 2001 { 2002 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 2003 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_lwp)); 2004 } 2005 2006 /* 2007 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2008 */ 2009 int 2010 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l) 2011 struct vnode *fdvp; 2012 const char *fnameptr; 2013 int fnamelen; 2014 struct vnode *tdvp; 2015 const char *tnameptr; 2016 int tnamelen; 2017 kauth_cred_t cred; 2018 struct lwp *l; 2019 { 2020 u_int32_t *tl; 2021 char *cp; 2022 #ifndef NFS_V2_ONLY 2023 int32_t t1; 2024 char *cp2; 2025 #endif 2026 int32_t t2; 2027 char *bpos, *dpos; 2028 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 2029 struct mbuf *mreq, *mrep, *md, *mb; 2030 const int v3 = NFS_ISV3(fdvp); 2031 int rexmit = 0; 2032 struct nfsnode *fdnp = VTONFS(fdvp); 2033 2034 nfsstats.rpccnt[NFSPROC_RENAME]++; 2035 nfsm_reqhead(fdnp, NFSPROC_RENAME, 2036 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 2037 nfsm_rndup(tnamelen)); 2038 nfsm_fhtom(fdnp, v3); 2039 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 2040 nfsm_fhtom(VTONFS(tdvp), v3); 2041 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 2042 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 2043 #ifndef NFS_V2_ONLY 2044 if (v3) { 2045 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 2046 nfsm_wcc_data(tdvp, twccflag, 0, !error); 2047 } 2048 #endif 2049 nfsm_reqdone; 2050 VTONFS(fdvp)->n_flag |= NMODIFIED; 2051 VTONFS(tdvp)->n_flag |= NMODIFIED; 2052 if (!fwccflag) 2053 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 2054 if (!twccflag) 2055 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 2056 /* 2057 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2058 */ 2059 if (rexmit && error == ENOENT) 2060 error = 0; 2061 return (error); 2062 } 2063 2064 /* 2065 * NFS link RPC, called from nfs_link. 2066 * Assumes dvp and vp locked, and leaves them that way. 2067 */ 2068 2069 static int 2070 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2071 size_t namelen, kauth_cred_t cred, struct lwp *l) 2072 { 2073 u_int32_t *tl; 2074 char *cp; 2075 #ifndef NFS_V2_ONLY 2076 int32_t t1; 2077 char *cp2; 2078 #endif 2079 int32_t t2; 2080 char *bpos, *dpos; 2081 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2082 struct mbuf *mreq, *mrep, *md, *mb; 2083 const int v3 = NFS_ISV3(dvp); 2084 int rexmit = 0; 2085 struct nfsnode *np = VTONFS(vp); 2086 2087 nfsstats.rpccnt[NFSPROC_LINK]++; 2088 nfsm_reqhead(np, NFSPROC_LINK, 2089 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2090 nfsm_fhtom(np, v3); 2091 nfsm_fhtom(VTONFS(dvp), v3); 2092 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2093 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2094 #ifndef NFS_V2_ONLY 2095 if (v3) { 2096 nfsm_postop_attr(vp, attrflag, 0); 2097 nfsm_wcc_data(dvp, wccflag, 0, !error); 2098 } 2099 #endif 2100 nfsm_reqdone; 2101 2102 VTONFS(dvp)->n_flag |= NMODIFIED; 2103 if (!attrflag) 2104 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2105 if (!wccflag) 2106 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2107 2108 /* 2109 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2110 */ 2111 if (rexmit && error == EEXIST) 2112 error = 0; 2113 2114 return error; 2115 } 2116 2117 /* 2118 * nfs hard link create call 2119 */ 2120 int 2121 nfs_link(v) 2122 void *v; 2123 { 2124 struct vop_link_args /* { 2125 struct vnode *a_dvp; 2126 struct vnode *a_vp; 2127 struct componentname *a_cnp; 2128 } */ *ap = v; 2129 struct vnode *vp = ap->a_vp; 2130 struct vnode *dvp = ap->a_dvp; 2131 struct componentname *cnp = ap->a_cnp; 2132 int error = 0; 2133 2134 if (dvp->v_mount != vp->v_mount) { 2135 VOP_ABORTOP(dvp, cnp); 2136 vput(dvp); 2137 return (EXDEV); 2138 } 2139 if (dvp != vp) { 2140 error = vn_lock(vp, LK_EXCLUSIVE); 2141 if (error != 0) { 2142 VOP_ABORTOP(dvp, cnp); 2143 vput(dvp); 2144 return error; 2145 } 2146 } 2147 2148 /* 2149 * Push all writes to the server, so that the attribute cache 2150 * doesn't get "out of sync" with the server. 2151 * XXX There should be a better way! 2152 */ 2153 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_lwp); 2154 2155 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2156 cnp->cn_cred, cnp->cn_lwp); 2157 2158 if (error == 0) 2159 cache_purge1(dvp, cnp, 0); 2160 PNBUF_PUT(cnp->cn_pnbuf); 2161 if (dvp != vp) 2162 VOP_UNLOCK(vp, 0); 2163 VN_KNOTE(vp, NOTE_LINK); 2164 VN_KNOTE(dvp, NOTE_WRITE); 2165 vput(dvp); 2166 return (error); 2167 } 2168 2169 /* 2170 * nfs symbolic link create call 2171 */ 2172 int 2173 nfs_symlink(v) 2174 void *v; 2175 { 2176 struct vop_symlink_args /* { 2177 struct vnode *a_dvp; 2178 struct vnode **a_vpp; 2179 struct componentname *a_cnp; 2180 struct vattr *a_vap; 2181 char *a_target; 2182 } */ *ap = v; 2183 struct vnode *dvp = ap->a_dvp; 2184 struct vattr *vap = ap->a_vap; 2185 struct componentname *cnp = ap->a_cnp; 2186 struct nfsv2_sattr *sp; 2187 u_int32_t *tl; 2188 char *cp; 2189 int32_t t1, t2; 2190 char *bpos, *dpos, *cp2; 2191 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2192 struct mbuf *mreq, *mrep, *md, *mb; 2193 struct vnode *newvp = (struct vnode *)0; 2194 const int v3 = NFS_ISV3(dvp); 2195 int rexmit = 0; 2196 struct nfsnode *dnp = VTONFS(dvp); 2197 2198 *ap->a_vpp = NULL; 2199 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2200 slen = strlen(ap->a_target); 2201 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2202 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2203 nfsm_fhtom(dnp, v3); 2204 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2205 #ifndef NFS_V2_ONlY 2206 if (v3) 2207 nfsm_v3attrbuild(vap, false); 2208 #endif 2209 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2210 #ifndef NFS_V2_ONlY 2211 if (!v3) { 2212 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2213 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2214 sp->sa_uid = nfs_xdrneg1; 2215 sp->sa_gid = nfs_xdrneg1; 2216 sp->sa_size = nfs_xdrneg1; 2217 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2218 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2219 } 2220 #endif 2221 nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_lwp, cnp->cn_cred, 2222 &rexmit); 2223 #ifndef NFS_V2_ONlY 2224 if (v3) { 2225 if (!error) 2226 nfsm_mtofh(dvp, newvp, v3, gotvp); 2227 nfsm_wcc_data(dvp, wccflag, 0, !error); 2228 } 2229 #endif 2230 nfsm_reqdone; 2231 /* 2232 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2233 */ 2234 if (rexmit && error == EEXIST) 2235 error = 0; 2236 if (error == 0 || error == EEXIST) 2237 cache_purge1(dvp, cnp, 0); 2238 if (error == 0 && newvp == NULL) { 2239 struct nfsnode *np = NULL; 2240 2241 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2242 cnp->cn_cred, cnp->cn_lwp, &np); 2243 if (error == 0) 2244 newvp = NFSTOV(np); 2245 } 2246 if (error) { 2247 if (newvp != NULL) 2248 vput(newvp); 2249 } else { 2250 *ap->a_vpp = newvp; 2251 } 2252 PNBUF_PUT(cnp->cn_pnbuf); 2253 VTONFS(dvp)->n_flag |= NMODIFIED; 2254 if (!wccflag) 2255 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2256 VN_KNOTE(dvp, NOTE_WRITE); 2257 vput(dvp); 2258 return (error); 2259 } 2260 2261 /* 2262 * nfs make dir call 2263 */ 2264 int 2265 nfs_mkdir(v) 2266 void *v; 2267 { 2268 struct vop_mkdir_args /* { 2269 struct vnode *a_dvp; 2270 struct vnode **a_vpp; 2271 struct componentname *a_cnp; 2272 struct vattr *a_vap; 2273 } */ *ap = v; 2274 struct vnode *dvp = ap->a_dvp; 2275 struct vattr *vap = ap->a_vap; 2276 struct componentname *cnp = ap->a_cnp; 2277 struct nfsv2_sattr *sp; 2278 u_int32_t *tl; 2279 char *cp; 2280 int32_t t1, t2; 2281 int len; 2282 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2283 struct vnode *newvp = (struct vnode *)0; 2284 char *bpos, *dpos, *cp2; 2285 int error = 0, wccflag = NFSV3_WCCRATTR; 2286 int gotvp = 0; 2287 int rexmit = 0; 2288 struct mbuf *mreq, *mrep, *md, *mb; 2289 const int v3 = NFS_ISV3(dvp); 2290 2291 len = cnp->cn_namelen; 2292 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2293 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2294 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2295 nfsm_fhtom(dnp, v3); 2296 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2297 #ifndef NFS_V2_ONLY 2298 if (v3) { 2299 nfsm_v3attrbuild(vap, false); 2300 } else 2301 #endif 2302 { 2303 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2304 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2305 sp->sa_uid = nfs_xdrneg1; 2306 sp->sa_gid = nfs_xdrneg1; 2307 sp->sa_size = nfs_xdrneg1; 2308 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2309 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2310 } 2311 nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit); 2312 if (!error) 2313 nfsm_mtofh(dvp, newvp, v3, gotvp); 2314 if (v3) 2315 nfsm_wcc_data(dvp, wccflag, 0, !error); 2316 nfsm_reqdone; 2317 VTONFS(dvp)->n_flag |= NMODIFIED; 2318 if (!wccflag) 2319 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2320 /* 2321 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2322 * if we can succeed in looking up the directory. 2323 */ 2324 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2325 if (newvp) { 2326 vput(newvp); 2327 newvp = (struct vnode *)0; 2328 } 2329 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2330 cnp->cn_lwp, &np); 2331 if (!error) { 2332 newvp = NFSTOV(np); 2333 if (newvp->v_type != VDIR || newvp == dvp) 2334 error = EEXIST; 2335 } 2336 } 2337 if (error) { 2338 if (newvp) { 2339 if (dvp != newvp) 2340 vput(newvp); 2341 else 2342 vrele(newvp); 2343 } 2344 } else { 2345 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2346 if (cnp->cn_flags & MAKEENTRY) 2347 nfs_cache_enter(dvp, newvp, cnp); 2348 *ap->a_vpp = newvp; 2349 } 2350 PNBUF_PUT(cnp->cn_pnbuf); 2351 vput(dvp); 2352 return (error); 2353 } 2354 2355 /* 2356 * nfs remove directory call 2357 */ 2358 int 2359 nfs_rmdir(v) 2360 void *v; 2361 { 2362 struct vop_rmdir_args /* { 2363 struct vnode *a_dvp; 2364 struct vnode *a_vp; 2365 struct componentname *a_cnp; 2366 } */ *ap = v; 2367 struct vnode *vp = ap->a_vp; 2368 struct vnode *dvp = ap->a_dvp; 2369 struct componentname *cnp = ap->a_cnp; 2370 u_int32_t *tl; 2371 char *cp; 2372 #ifndef NFS_V2_ONLY 2373 int32_t t1; 2374 char *cp2; 2375 #endif 2376 int32_t t2; 2377 char *bpos, *dpos; 2378 int error = 0, wccflag = NFSV3_WCCRATTR; 2379 int rexmit = 0; 2380 struct mbuf *mreq, *mrep, *md, *mb; 2381 const int v3 = NFS_ISV3(dvp); 2382 struct nfsnode *dnp; 2383 2384 if (dvp == vp) { 2385 vrele(dvp); 2386 vput(dvp); 2387 PNBUF_PUT(cnp->cn_pnbuf); 2388 return (EINVAL); 2389 } 2390 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2391 dnp = VTONFS(dvp); 2392 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2393 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2394 nfsm_fhtom(dnp, v3); 2395 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2396 nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_lwp, cnp->cn_cred, &rexmit); 2397 #ifndef NFS_V2_ONLY 2398 if (v3) 2399 nfsm_wcc_data(dvp, wccflag, 0, !error); 2400 #endif 2401 nfsm_reqdone; 2402 PNBUF_PUT(cnp->cn_pnbuf); 2403 VTONFS(dvp)->n_flag |= NMODIFIED; 2404 if (!wccflag) 2405 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2406 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2407 VN_KNOTE(vp, NOTE_DELETE); 2408 cache_purge(vp); 2409 vput(vp); 2410 vput(dvp); 2411 /* 2412 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2413 */ 2414 if (rexmit && error == ENOENT) 2415 error = 0; 2416 return (error); 2417 } 2418 2419 /* 2420 * nfs readdir call 2421 */ 2422 int 2423 nfs_readdir(v) 2424 void *v; 2425 { 2426 struct vop_readdir_args /* { 2427 struct vnode *a_vp; 2428 struct uio *a_uio; 2429 kauth_cred_t a_cred; 2430 int *a_eofflag; 2431 off_t **a_cookies; 2432 int *a_ncookies; 2433 } */ *ap = v; 2434 struct vnode *vp = ap->a_vp; 2435 struct uio *uio = ap->a_uio; 2436 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2437 char *base = uio->uio_iov->iov_base; 2438 int tresid, error; 2439 size_t count, lost; 2440 struct dirent *dp; 2441 off_t *cookies = NULL; 2442 int ncookies = 0, nc; 2443 2444 if (vp->v_type != VDIR) 2445 return (EPERM); 2446 2447 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2448 count = uio->uio_resid - lost; 2449 if (count <= 0) 2450 return (EINVAL); 2451 2452 /* 2453 * Call nfs_bioread() to do the real work. 2454 */ 2455 tresid = uio->uio_resid = count; 2456 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2457 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2458 2459 if (!error && ap->a_cookies) { 2460 ncookies = count / 16; 2461 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2462 *ap->a_cookies = cookies; 2463 } 2464 2465 if (!error && uio->uio_resid == tresid) { 2466 uio->uio_resid += lost; 2467 nfsstats.direofcache_misses++; 2468 if (ap->a_cookies) 2469 *ap->a_ncookies = 0; 2470 *ap->a_eofflag = 1; 2471 return (0); 2472 } 2473 2474 if (!error && ap->a_cookies) { 2475 /* 2476 * Only the NFS server and emulations use cookies, and they 2477 * load the directory block into system space, so we can 2478 * just look at it directly. 2479 */ 2480 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2481 uio->uio_iovcnt != 1) 2482 panic("nfs_readdir: lost in space"); 2483 for (nc = 0; ncookies-- && 2484 base < (char *)uio->uio_iov->iov_base; nc++){ 2485 dp = (struct dirent *) base; 2486 if (dp->d_reclen == 0) 2487 break; 2488 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2489 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2490 else 2491 *(cookies++) = NFS_GETCOOKIE(dp); 2492 base += dp->d_reclen; 2493 } 2494 uio->uio_resid += 2495 ((char *)uio->uio_iov->iov_base - base); 2496 uio->uio_iov->iov_len += 2497 ((char *)uio->uio_iov->iov_base - base); 2498 uio->uio_iov->iov_base = base; 2499 *ap->a_ncookies = nc; 2500 } 2501 2502 uio->uio_resid += lost; 2503 *ap->a_eofflag = 0; 2504 return (error); 2505 } 2506 2507 /* 2508 * Readdir rpc call. 2509 * Called from below the buffer cache by nfs_doio(). 2510 */ 2511 int 2512 nfs_readdirrpc(vp, uiop, cred) 2513 struct vnode *vp; 2514 struct uio *uiop; 2515 kauth_cred_t cred; 2516 { 2517 int len, left; 2518 struct dirent *dp = NULL; 2519 u_int32_t *tl; 2520 char *cp; 2521 int32_t t1, t2; 2522 char *bpos, *dpos, *cp2; 2523 struct mbuf *mreq, *mrep, *md, *mb; 2524 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2525 struct nfsnode *dnp = VTONFS(vp); 2526 u_quad_t fileno; 2527 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2528 #ifndef NFS_V2_ONLY 2529 int attrflag; 2530 #endif 2531 int nrpcs = 0, reclen; 2532 const int v3 = NFS_ISV3(vp); 2533 2534 #ifdef DIAGNOSTIC 2535 /* 2536 * Should be called from buffer cache, so only amount of 2537 * NFS_DIRBLKSIZ will be requested. 2538 */ 2539 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2540 panic("nfs readdirrpc bad uio"); 2541 #endif 2542 2543 /* 2544 * Loop around doing readdir rpc's of size nm_readdirsize 2545 * truncated to a multiple of NFS_DIRFRAGSIZ. 2546 * The stopping criteria is EOF or buffer full. 2547 */ 2548 while (more_dirs && bigenough) { 2549 /* 2550 * Heuristic: don't bother to do another RPC to further 2551 * fill up this block if there is not much room left. (< 50% 2552 * of the readdir RPC size). This wastes some buffer space 2553 * but can save up to 50% in RPC calls. 2554 */ 2555 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2556 bigenough = 0; 2557 break; 2558 } 2559 nfsstats.rpccnt[NFSPROC_READDIR]++; 2560 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2561 NFSX_READDIR(v3)); 2562 nfsm_fhtom(dnp, v3); 2563 #ifndef NFS_V2_ONLY 2564 if (v3) { 2565 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2566 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2567 txdr_swapcookie3(uiop->uio_offset, tl); 2568 } else { 2569 txdr_cookie3(uiop->uio_offset, tl); 2570 } 2571 tl += 2; 2572 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2573 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2574 } else 2575 #endif 2576 { 2577 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2578 *tl++ = txdr_unsigned(uiop->uio_offset); 2579 } 2580 *tl = txdr_unsigned(nmp->nm_readdirsize); 2581 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2582 nrpcs++; 2583 #ifndef NFS_V2_ONLY 2584 if (v3) { 2585 nfsm_postop_attr(vp, attrflag, 0); 2586 if (!error) { 2587 nfsm_dissect(tl, u_int32_t *, 2588 2 * NFSX_UNSIGNED); 2589 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2590 dnp->n_cookieverf.nfsuquad[1] = *tl; 2591 } else { 2592 m_freem(mrep); 2593 goto nfsmout; 2594 } 2595 } 2596 #endif 2597 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2598 more_dirs = fxdr_unsigned(int, *tl); 2599 2600 /* loop thru the dir entries, doctoring them to 4bsd form */ 2601 while (more_dirs && bigenough) { 2602 #ifndef NFS_V2_ONLY 2603 if (v3) { 2604 nfsm_dissect(tl, u_int32_t *, 2605 3 * NFSX_UNSIGNED); 2606 fileno = fxdr_hyper(tl); 2607 len = fxdr_unsigned(int, *(tl + 2)); 2608 } else 2609 #endif 2610 { 2611 nfsm_dissect(tl, u_int32_t *, 2612 2 * NFSX_UNSIGNED); 2613 fileno = fxdr_unsigned(u_quad_t, *tl++); 2614 len = fxdr_unsigned(int, *tl); 2615 } 2616 if (len <= 0 || len > NFS_MAXNAMLEN) { 2617 error = EBADRPC; 2618 m_freem(mrep); 2619 goto nfsmout; 2620 } 2621 /* for cookie stashing */ 2622 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2623 left = NFS_DIRFRAGSIZ - blksiz; 2624 if (reclen > left) { 2625 memset(uiop->uio_iov->iov_base, 0, left); 2626 dp->d_reclen += left; 2627 UIO_ADVANCE(uiop, left); 2628 blksiz = 0; 2629 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2630 } 2631 if (reclen > uiop->uio_resid) 2632 bigenough = 0; 2633 if (bigenough) { 2634 int tlen; 2635 2636 dp = (struct dirent *)uiop->uio_iov->iov_base; 2637 dp->d_fileno = fileno; 2638 dp->d_namlen = len; 2639 dp->d_reclen = reclen; 2640 dp->d_type = DT_UNKNOWN; 2641 blksiz += reclen; 2642 if (blksiz == NFS_DIRFRAGSIZ) 2643 blksiz = 0; 2644 UIO_ADVANCE(uiop, DIRHDSIZ); 2645 nfsm_mtouio(uiop, len); 2646 tlen = reclen - (DIRHDSIZ + len); 2647 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2648 UIO_ADVANCE(uiop, tlen); 2649 } else 2650 nfsm_adv(nfsm_rndup(len)); 2651 #ifndef NFS_V2_ONLY 2652 if (v3) { 2653 nfsm_dissect(tl, u_int32_t *, 2654 3 * NFSX_UNSIGNED); 2655 } else 2656 #endif 2657 { 2658 nfsm_dissect(tl, u_int32_t *, 2659 2 * NFSX_UNSIGNED); 2660 } 2661 if (bigenough) { 2662 #ifndef NFS_V2_ONLY 2663 if (v3) { 2664 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2665 uiop->uio_offset = 2666 fxdr_swapcookie3(tl); 2667 else 2668 uiop->uio_offset = 2669 fxdr_cookie3(tl); 2670 } 2671 else 2672 #endif 2673 { 2674 uiop->uio_offset = 2675 fxdr_unsigned(off_t, *tl); 2676 } 2677 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2678 } 2679 if (v3) 2680 tl += 2; 2681 else 2682 tl++; 2683 more_dirs = fxdr_unsigned(int, *tl); 2684 } 2685 /* 2686 * If at end of rpc data, get the eof boolean 2687 */ 2688 if (!more_dirs) { 2689 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2690 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2691 2692 /* 2693 * kludge: if we got no entries, treat it as EOF. 2694 * some server sometimes send a reply without any 2695 * entries or EOF. 2696 * although it might mean the server has very long name, 2697 * we can't handle such entries anyway. 2698 */ 2699 2700 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2701 more_dirs = 0; 2702 } 2703 m_freem(mrep); 2704 } 2705 /* 2706 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2707 * by increasing d_reclen for the last record. 2708 */ 2709 if (blksiz > 0) { 2710 left = NFS_DIRFRAGSIZ - blksiz; 2711 memset(uiop->uio_iov->iov_base, 0, left); 2712 dp->d_reclen += left; 2713 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2714 UIO_ADVANCE(uiop, left); 2715 } 2716 2717 /* 2718 * We are now either at the end of the directory or have filled the 2719 * block. 2720 */ 2721 if (bigenough) { 2722 dnp->n_direofoffset = uiop->uio_offset; 2723 dnp->n_flag |= NEOFVALID; 2724 } 2725 nfsmout: 2726 return (error); 2727 } 2728 2729 #ifndef NFS_V2_ONLY 2730 /* 2731 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2732 */ 2733 int 2734 nfs_readdirplusrpc(vp, uiop, cred) 2735 struct vnode *vp; 2736 struct uio *uiop; 2737 kauth_cred_t cred; 2738 { 2739 int len, left; 2740 struct dirent *dp = NULL; 2741 u_int32_t *tl; 2742 char *cp; 2743 int32_t t1, t2; 2744 struct vnode *newvp; 2745 char *bpos, *dpos, *cp2; 2746 struct mbuf *mreq, *mrep, *md, *mb; 2747 struct nameidata nami, *ndp = &nami; 2748 struct componentname *cnp = &ndp->ni_cnd; 2749 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2750 struct nfsnode *dnp = VTONFS(vp), *np; 2751 nfsfh_t *fhp; 2752 u_quad_t fileno; 2753 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2754 int attrflag, fhsize, nrpcs = 0, reclen; 2755 struct nfs_fattr fattr, *fp; 2756 2757 #ifdef DIAGNOSTIC 2758 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2759 panic("nfs readdirplusrpc bad uio"); 2760 #endif 2761 ndp->ni_dvp = vp; 2762 newvp = NULLVP; 2763 2764 /* 2765 * Loop around doing readdir rpc's of size nm_readdirsize 2766 * truncated to a multiple of NFS_DIRFRAGSIZ. 2767 * The stopping criteria is EOF or buffer full. 2768 */ 2769 while (more_dirs && bigenough) { 2770 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2771 bigenough = 0; 2772 break; 2773 } 2774 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2775 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2776 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2777 nfsm_fhtom(dnp, 1); 2778 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2779 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2780 txdr_swapcookie3(uiop->uio_offset, tl); 2781 } else { 2782 txdr_cookie3(uiop->uio_offset, tl); 2783 } 2784 tl += 2; 2785 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2786 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2787 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2788 *tl = txdr_unsigned(nmp->nm_rsize); 2789 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2790 nfsm_postop_attr(vp, attrflag, 0); 2791 if (error) { 2792 m_freem(mrep); 2793 goto nfsmout; 2794 } 2795 nrpcs++; 2796 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2797 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2798 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2799 more_dirs = fxdr_unsigned(int, *tl); 2800 2801 /* loop thru the dir entries, doctoring them to 4bsd form */ 2802 while (more_dirs && bigenough) { 2803 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2804 fileno = fxdr_hyper(tl); 2805 len = fxdr_unsigned(int, *(tl + 2)); 2806 if (len <= 0 || len > NFS_MAXNAMLEN) { 2807 error = EBADRPC; 2808 m_freem(mrep); 2809 goto nfsmout; 2810 } 2811 /* for cookie stashing */ 2812 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2813 left = NFS_DIRFRAGSIZ - blksiz; 2814 if (reclen > left) { 2815 /* 2816 * DIRFRAGSIZ is aligned, no need to align 2817 * again here. 2818 */ 2819 memset(uiop->uio_iov->iov_base, 0, left); 2820 dp->d_reclen += left; 2821 UIO_ADVANCE(uiop, left); 2822 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2823 blksiz = 0; 2824 } 2825 if (reclen > uiop->uio_resid) 2826 bigenough = 0; 2827 if (bigenough) { 2828 int tlen; 2829 2830 dp = (struct dirent *)uiop->uio_iov->iov_base; 2831 dp->d_fileno = fileno; 2832 dp->d_namlen = len; 2833 dp->d_reclen = reclen; 2834 dp->d_type = DT_UNKNOWN; 2835 blksiz += reclen; 2836 if (blksiz == NFS_DIRFRAGSIZ) 2837 blksiz = 0; 2838 UIO_ADVANCE(uiop, DIRHDSIZ); 2839 nfsm_mtouio(uiop, len); 2840 tlen = reclen - (DIRHDSIZ + len); 2841 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2842 UIO_ADVANCE(uiop, tlen); 2843 cnp->cn_nameptr = dp->d_name; 2844 cnp->cn_namelen = dp->d_namlen; 2845 } else 2846 nfsm_adv(nfsm_rndup(len)); 2847 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2848 if (bigenough) { 2849 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2850 uiop->uio_offset = 2851 fxdr_swapcookie3(tl); 2852 else 2853 uiop->uio_offset = 2854 fxdr_cookie3(tl); 2855 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2856 } 2857 tl += 2; 2858 2859 /* 2860 * Since the attributes are before the file handle 2861 * (sigh), we must skip over the attributes and then 2862 * come back and get them. 2863 */ 2864 attrflag = fxdr_unsigned(int, *tl); 2865 if (attrflag) { 2866 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2867 memcpy(&fattr, fp, NFSX_V3FATTR); 2868 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2869 doit = fxdr_unsigned(int, *tl); 2870 if (doit) { 2871 nfsm_getfh(fhp, fhsize, 1); 2872 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2873 VREF(vp); 2874 newvp = vp; 2875 np = dnp; 2876 } else { 2877 error = nfs_nget1(vp->v_mount, fhp, 2878 fhsize, &np, LK_NOWAIT); 2879 if (!error) 2880 newvp = NFSTOV(np); 2881 } 2882 if (!error) { 2883 const char *xcp; 2884 2885 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2886 if (bigenough) { 2887 dp->d_type = 2888 IFTODT(VTTOIF(np->n_vattr->va_type)); 2889 if (cnp->cn_namelen <= NCHNAMLEN) { 2890 ndp->ni_vp = newvp; 2891 xcp = cnp->cn_nameptr + 2892 cnp->cn_namelen; 2893 cnp->cn_hash = 2894 namei_hash(cnp->cn_nameptr, &xcp); 2895 nfs_cache_enter(ndp->ni_dvp, 2896 ndp->ni_vp, cnp); 2897 } 2898 } 2899 } 2900 error = 0; 2901 } 2902 } else { 2903 /* Just skip over the file handle */ 2904 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2905 i = fxdr_unsigned(int, *tl); 2906 nfsm_adv(nfsm_rndup(i)); 2907 } 2908 if (newvp != NULLVP) { 2909 if (newvp == vp) 2910 vrele(newvp); 2911 else 2912 vput(newvp); 2913 newvp = NULLVP; 2914 } 2915 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2916 more_dirs = fxdr_unsigned(int, *tl); 2917 } 2918 /* 2919 * If at end of rpc data, get the eof boolean 2920 */ 2921 if (!more_dirs) { 2922 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2923 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2924 2925 /* 2926 * kludge: see a comment in nfs_readdirrpc. 2927 */ 2928 2929 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2930 more_dirs = 0; 2931 } 2932 m_freem(mrep); 2933 } 2934 /* 2935 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2936 * by increasing d_reclen for the last record. 2937 */ 2938 if (blksiz > 0) { 2939 left = NFS_DIRFRAGSIZ - blksiz; 2940 memset(uiop->uio_iov->iov_base, 0, left); 2941 dp->d_reclen += left; 2942 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2943 UIO_ADVANCE(uiop, left); 2944 } 2945 2946 /* 2947 * We are now either at the end of the directory or have filled the 2948 * block. 2949 */ 2950 if (bigenough) { 2951 dnp->n_direofoffset = uiop->uio_offset; 2952 dnp->n_flag |= NEOFVALID; 2953 } 2954 nfsmout: 2955 if (newvp != NULLVP) { 2956 if(newvp == vp) 2957 vrele(newvp); 2958 else 2959 vput(newvp); 2960 } 2961 return (error); 2962 } 2963 #endif 2964 2965 /* 2966 * Silly rename. To make the NFS filesystem that is stateless look a little 2967 * more like the "ufs" a remove of an active vnode is translated to a rename 2968 * to a funny looking filename that is removed by nfs_inactive on the 2969 * nfsnode. There is the potential for another process on a different client 2970 * to create the same funny name between the nfs_lookitup() fails and the 2971 * nfs_rename() completes, but... 2972 */ 2973 int 2974 nfs_sillyrename(dvp, vp, cnp, dolink) 2975 struct vnode *dvp, *vp; 2976 struct componentname *cnp; 2977 bool dolink; 2978 { 2979 struct sillyrename *sp; 2980 struct nfsnode *np; 2981 int error; 2982 short pid; 2983 2984 cache_purge(dvp); 2985 np = VTONFS(vp); 2986 #ifndef DIAGNOSTIC 2987 if (vp->v_type == VDIR) 2988 panic("nfs: sillyrename dir"); 2989 #endif 2990 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2991 M_NFSREQ, M_WAITOK); 2992 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2993 sp->s_dvp = dvp; 2994 VREF(dvp); 2995 2996 /* Fudge together a funny name */ 2997 pid = cnp->cn_lwp->l_proc->p_pid; 2998 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2999 sp->s_namlen = 12; 3000 sp->s_name[8] = hexdigits[pid & 0xf]; 3001 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 3002 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 3003 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 3004 3005 /* Try lookitups until we get one that isn't there */ 3006 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3007 cnp->cn_lwp, (struct nfsnode **)0) == 0) { 3008 sp->s_name[4]++; 3009 if (sp->s_name[4] > 'z') { 3010 error = EINVAL; 3011 goto bad; 3012 } 3013 } 3014 if (dolink) { 3015 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 3016 sp->s_cred, cnp->cn_lwp); 3017 /* 3018 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 3019 */ 3020 if (error == ENOTSUP) { 3021 error = nfs_renameit(dvp, cnp, sp); 3022 } 3023 } else { 3024 error = nfs_renameit(dvp, cnp, sp); 3025 } 3026 if (error) 3027 goto bad; 3028 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3029 cnp->cn_lwp, &np); 3030 np->n_sillyrename = sp; 3031 return (0); 3032 bad: 3033 vrele(sp->s_dvp); 3034 kauth_cred_free(sp->s_cred); 3035 free((void *)sp, M_NFSREQ); 3036 return (error); 3037 } 3038 3039 /* 3040 * Look up a file name and optionally either update the file handle or 3041 * allocate an nfsnode, depending on the value of npp. 3042 * npp == NULL --> just do the lookup 3043 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 3044 * handled too 3045 * *npp != NULL --> update the file handle in the vnode 3046 */ 3047 int 3048 nfs_lookitup(dvp, name, len, cred, l, npp) 3049 struct vnode *dvp; 3050 const char *name; 3051 int len; 3052 kauth_cred_t cred; 3053 struct lwp *l; 3054 struct nfsnode **npp; 3055 { 3056 u_int32_t *tl; 3057 char *cp; 3058 int32_t t1, t2; 3059 struct vnode *newvp = (struct vnode *)0; 3060 struct nfsnode *np, *dnp = VTONFS(dvp); 3061 char *bpos, *dpos, *cp2; 3062 int error = 0, fhlen; 3063 #ifndef NFS_V2_ONLY 3064 int attrflag; 3065 #endif 3066 struct mbuf *mreq, *mrep, *md, *mb; 3067 nfsfh_t *nfhp; 3068 const int v3 = NFS_ISV3(dvp); 3069 3070 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 3071 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 3072 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 3073 nfsm_fhtom(dnp, v3); 3074 nfsm_strtom(name, len, NFS_MAXNAMLEN); 3075 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 3076 if (npp && !error) { 3077 nfsm_getfh(nfhp, fhlen, v3); 3078 if (*npp) { 3079 np = *npp; 3080 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3081 free((void *)np->n_fhp, M_NFSBIGFH); 3082 np->n_fhp = &np->n_fh; 3083 } 3084 #if NFS_SMALLFH < NFSX_V3FHMAX 3085 else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 3086 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 3087 #endif 3088 memcpy((void *)np->n_fhp, (void *)nfhp, fhlen); 3089 np->n_fhsize = fhlen; 3090 newvp = NFSTOV(np); 3091 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3092 VREF(dvp); 3093 newvp = dvp; 3094 np = dnp; 3095 } else { 3096 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 3097 if (error) { 3098 m_freem(mrep); 3099 return (error); 3100 } 3101 newvp = NFSTOV(np); 3102 } 3103 #ifndef NFS_V2_ONLY 3104 if (v3) { 3105 nfsm_postop_attr(newvp, attrflag, 0); 3106 if (!attrflag && *npp == NULL) { 3107 m_freem(mrep); 3108 vput(newvp); 3109 return (ENOENT); 3110 } 3111 } else 3112 #endif 3113 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3114 } 3115 nfsm_reqdone; 3116 if (npp && *npp == NULL) { 3117 if (error) { 3118 if (newvp) 3119 vput(newvp); 3120 } else 3121 *npp = np; 3122 } 3123 return (error); 3124 } 3125 3126 #ifndef NFS_V2_ONLY 3127 /* 3128 * Nfs Version 3 commit rpc 3129 */ 3130 int 3131 nfs_commit(vp, offset, cnt, l) 3132 struct vnode *vp; 3133 off_t offset; 3134 uint32_t cnt; 3135 struct lwp *l; 3136 { 3137 char *cp; 3138 u_int32_t *tl; 3139 int32_t t1, t2; 3140 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3141 char *bpos, *dpos, *cp2; 3142 int error = 0, wccflag = NFSV3_WCCRATTR; 3143 struct mbuf *mreq, *mrep, *md, *mb; 3144 struct nfsnode *np; 3145 3146 KASSERT(NFS_ISV3(vp)); 3147 3148 #ifdef NFS_DEBUG_COMMIT 3149 printf("commit %lu - %lu\n", (unsigned long)offset, 3150 (unsigned long)(offset + cnt)); 3151 #endif 3152 3153 simple_lock(&nmp->nm_slock); 3154 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3155 simple_unlock(&nmp->nm_slock); 3156 return (0); 3157 } 3158 simple_unlock(&nmp->nm_slock); 3159 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3160 np = VTONFS(vp); 3161 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3162 nfsm_fhtom(np, 1); 3163 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3164 txdr_hyper(offset, tl); 3165 tl += 2; 3166 *tl = txdr_unsigned(cnt); 3167 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3168 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3169 if (!error) { 3170 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3171 simple_lock(&nmp->nm_slock); 3172 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3173 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3174 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3175 error = NFSERR_STALEWRITEVERF; 3176 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3177 } 3178 simple_unlock(&nmp->nm_slock); 3179 } 3180 nfsm_reqdone; 3181 return (error); 3182 } 3183 #endif 3184 3185 /* 3186 * Kludge City.. 3187 * - make nfs_bmap() essentially a no-op that does no translation 3188 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3189 * (Maybe I could use the process's page mapping, but I was concerned that 3190 * Kernel Write might not be enabled and also figured copyout() would do 3191 * a lot more work than memcpy() and also it currently happens in the 3192 * context of the swapper process (2). 3193 */ 3194 int 3195 nfs_bmap(v) 3196 void *v; 3197 { 3198 struct vop_bmap_args /* { 3199 struct vnode *a_vp; 3200 daddr_t a_bn; 3201 struct vnode **a_vpp; 3202 daddr_t *a_bnp; 3203 int *a_runp; 3204 } */ *ap = v; 3205 struct vnode *vp = ap->a_vp; 3206 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3207 3208 if (ap->a_vpp != NULL) 3209 *ap->a_vpp = vp; 3210 if (ap->a_bnp != NULL) 3211 *ap->a_bnp = ap->a_bn << bshift; 3212 if (ap->a_runp != NULL) 3213 *ap->a_runp = 1024 * 1024; /* XXX */ 3214 return (0); 3215 } 3216 3217 /* 3218 * Strategy routine. 3219 * For async requests when nfsiod(s) are running, queue the request by 3220 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3221 * request. 3222 */ 3223 int 3224 nfs_strategy(v) 3225 void *v; 3226 { 3227 struct vop_strategy_args *ap = v; 3228 struct buf *bp = ap->a_bp; 3229 int error = 0; 3230 3231 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3232 panic("nfs physio/async"); 3233 3234 /* 3235 * If the op is asynchronous and an i/o daemon is waiting 3236 * queue the request, wake it up and wait for completion 3237 * otherwise just do it ourselves. 3238 */ 3239 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3240 error = nfs_doio(bp); 3241 return (error); 3242 } 3243 3244 /* 3245 * fsync vnode op. Just call nfs_flush() with commit == 1. 3246 */ 3247 /* ARGSUSED */ 3248 int 3249 nfs_fsync(v) 3250 void *v; 3251 { 3252 struct vop_fsync_args /* { 3253 struct vnodeop_desc *a_desc; 3254 struct vnode * a_vp; 3255 kauth_cred_t a_cred; 3256 int a_flags; 3257 off_t offlo; 3258 off_t offhi; 3259 struct lwp * a_l; 3260 } */ *ap = v; 3261 3262 struct vnode *vp = ap->a_vp; 3263 3264 if (vp->v_type != VREG) 3265 return 0; 3266 3267 return (nfs_flush(vp, ap->a_cred, 3268 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_l, 1)); 3269 } 3270 3271 /* 3272 * Flush all the data associated with a vnode. 3273 */ 3274 int 3275 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3276 int commit) 3277 { 3278 struct nfsnode *np = VTONFS(vp); 3279 int error; 3280 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3281 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3282 3283 simple_lock(&vp->v_interlock); 3284 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3285 if (np->n_flag & NWRITEERR) { 3286 error = np->n_error; 3287 np->n_flag &= ~NWRITEERR; 3288 } 3289 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3290 return (error); 3291 } 3292 3293 /* 3294 * Return POSIX pathconf information applicable to nfs. 3295 * 3296 * N.B. The NFS V2 protocol doesn't support this RPC. 3297 */ 3298 /* ARGSUSED */ 3299 int 3300 nfs_pathconf(v) 3301 void *v; 3302 { 3303 struct vop_pathconf_args /* { 3304 struct vnode *a_vp; 3305 int a_name; 3306 register_t *a_retval; 3307 } */ *ap = v; 3308 struct nfsv3_pathconf *pcp; 3309 struct vnode *vp = ap->a_vp; 3310 struct mbuf *mreq, *mrep, *md, *mb; 3311 int32_t t1, t2; 3312 u_int32_t *tl; 3313 char *bpos, *dpos, *cp, *cp2; 3314 int error = 0, attrflag; 3315 #ifndef NFS_V2_ONLY 3316 struct nfsmount *nmp; 3317 unsigned int l; 3318 u_int64_t maxsize; 3319 #endif 3320 const int v3 = NFS_ISV3(vp); 3321 struct nfsnode *np = VTONFS(vp); 3322 3323 switch (ap->a_name) { 3324 /* Names that can be resolved locally. */ 3325 case _PC_PIPE_BUF: 3326 *ap->a_retval = PIPE_BUF; 3327 break; 3328 case _PC_SYNC_IO: 3329 *ap->a_retval = 1; 3330 break; 3331 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3332 case _PC_LINK_MAX: 3333 case _PC_NAME_MAX: 3334 case _PC_CHOWN_RESTRICTED: 3335 case _PC_NO_TRUNC: 3336 if (!v3) { 3337 error = EINVAL; 3338 break; 3339 } 3340 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3341 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3342 nfsm_fhtom(np, 1); 3343 nfsm_request(np, NFSPROC_PATHCONF, 3344 curlwp, curlwp->l_cred); /* XXX */ 3345 nfsm_postop_attr(vp, attrflag, 0); 3346 if (!error) { 3347 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3348 NFSX_V3PATHCONF); 3349 switch (ap->a_name) { 3350 case _PC_LINK_MAX: 3351 *ap->a_retval = 3352 fxdr_unsigned(register_t, pcp->pc_linkmax); 3353 break; 3354 case _PC_NAME_MAX: 3355 *ap->a_retval = 3356 fxdr_unsigned(register_t, pcp->pc_namemax); 3357 break; 3358 case _PC_CHOWN_RESTRICTED: 3359 *ap->a_retval = 3360 (pcp->pc_chownrestricted == nfs_true); 3361 break; 3362 case _PC_NO_TRUNC: 3363 *ap->a_retval = 3364 (pcp->pc_notrunc == nfs_true); 3365 break; 3366 } 3367 } 3368 nfsm_reqdone; 3369 break; 3370 case _PC_FILESIZEBITS: 3371 #ifndef NFS_V2_ONLY 3372 if (v3) { 3373 nmp = VFSTONFS(vp->v_mount); 3374 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3375 if ((error = nfs_fsinfo(nmp, vp, 3376 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3377 break; 3378 for (l = 0, maxsize = nmp->nm_maxfilesize; 3379 (maxsize >> l) > 0; l++) 3380 ; 3381 *ap->a_retval = l + 1; 3382 } else 3383 #endif 3384 { 3385 *ap->a_retval = 32; /* NFS V2 limitation */ 3386 } 3387 break; 3388 default: 3389 error = EINVAL; 3390 break; 3391 } 3392 3393 return (error); 3394 } 3395 3396 /* 3397 * NFS advisory byte-level locks. 3398 */ 3399 int 3400 nfs_advlock(v) 3401 void *v; 3402 { 3403 struct vop_advlock_args /* { 3404 struct vnode *a_vp; 3405 void *a_id; 3406 int a_op; 3407 struct flock *a_fl; 3408 int a_flags; 3409 } */ *ap = v; 3410 struct nfsnode *np = VTONFS(ap->a_vp); 3411 3412 return lf_advlock(ap, &np->n_lockf, np->n_size); 3413 } 3414 3415 /* 3416 * Print out the contents of an nfsnode. 3417 */ 3418 int 3419 nfs_print(v) 3420 void *v; 3421 { 3422 struct vop_print_args /* { 3423 struct vnode *a_vp; 3424 } */ *ap = v; 3425 struct vnode *vp = ap->a_vp; 3426 struct nfsnode *np = VTONFS(vp); 3427 3428 printf("tag VT_NFS, fileid %lld fsid 0x%lx", 3429 (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3430 if (vp->v_type == VFIFO) 3431 fifo_printinfo(vp); 3432 printf("\n"); 3433 return (0); 3434 } 3435 3436 /* 3437 * nfs unlock wrapper. 3438 */ 3439 int 3440 nfs_unlock(void *v) 3441 { 3442 struct vop_unlock_args /* { 3443 struct vnode *a_vp; 3444 int a_flags; 3445 } */ *ap = v; 3446 struct vnode *vp = ap->a_vp; 3447 3448 /* 3449 * VOP_UNLOCK can be called by nfs_loadattrcache 3450 * with v_data == 0. 3451 */ 3452 if (VTONFS(vp)) { 3453 nfs_delayedtruncate(vp); 3454 } 3455 3456 return genfs_unlock(v); 3457 } 3458 3459 /* 3460 * nfs special file access vnode op. 3461 * Essentially just get vattr and then imitate iaccess() since the device is 3462 * local to the client. 3463 */ 3464 int 3465 nfsspec_access(v) 3466 void *v; 3467 { 3468 struct vop_access_args /* { 3469 struct vnode *a_vp; 3470 int a_mode; 3471 kauth_cred_t a_cred; 3472 struct lwp *a_l; 3473 } */ *ap = v; 3474 struct vattr va; 3475 struct vnode *vp = ap->a_vp; 3476 int error; 3477 3478 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_l); 3479 if (error) 3480 return (error); 3481 3482 /* 3483 * Disallow write attempts on filesystems mounted read-only; 3484 * unless the file is a socket, fifo, or a block or character 3485 * device resident on the filesystem. 3486 */ 3487 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3488 switch (vp->v_type) { 3489 case VREG: 3490 case VDIR: 3491 case VLNK: 3492 return (EROFS); 3493 default: 3494 break; 3495 } 3496 } 3497 3498 return (vaccess(va.va_type, va.va_mode, 3499 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3500 } 3501 3502 /* 3503 * Read wrapper for special devices. 3504 */ 3505 int 3506 nfsspec_read(v) 3507 void *v; 3508 { 3509 struct vop_read_args /* { 3510 struct vnode *a_vp; 3511 struct uio *a_uio; 3512 int a_ioflag; 3513 kauth_cred_t a_cred; 3514 } */ *ap = v; 3515 struct nfsnode *np = VTONFS(ap->a_vp); 3516 3517 /* 3518 * Set access flag. 3519 */ 3520 np->n_flag |= NACC; 3521 getnanotime(&np->n_atim); 3522 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3523 } 3524 3525 /* 3526 * Write wrapper for special devices. 3527 */ 3528 int 3529 nfsspec_write(v) 3530 void *v; 3531 { 3532 struct vop_write_args /* { 3533 struct vnode *a_vp; 3534 struct uio *a_uio; 3535 int a_ioflag; 3536 kauth_cred_t a_cred; 3537 } */ *ap = v; 3538 struct nfsnode *np = VTONFS(ap->a_vp); 3539 3540 /* 3541 * Set update flag. 3542 */ 3543 np->n_flag |= NUPD; 3544 getnanotime(&np->n_mtim); 3545 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3546 } 3547 3548 /* 3549 * Close wrapper for special devices. 3550 * 3551 * Update the times on the nfsnode then do device close. 3552 */ 3553 int 3554 nfsspec_close(v) 3555 void *v; 3556 { 3557 struct vop_close_args /* { 3558 struct vnode *a_vp; 3559 int a_fflag; 3560 kauth_cred_t a_cred; 3561 struct lwp *a_l; 3562 } */ *ap = v; 3563 struct vnode *vp = ap->a_vp; 3564 struct nfsnode *np = VTONFS(vp); 3565 struct vattr vattr; 3566 3567 if (np->n_flag & (NACC | NUPD)) { 3568 np->n_flag |= NCHG; 3569 if (vp->v_usecount == 1 && 3570 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3571 VATTR_NULL(&vattr); 3572 if (np->n_flag & NACC) 3573 vattr.va_atime = np->n_atim; 3574 if (np->n_flag & NUPD) 3575 vattr.va_mtime = np->n_mtim; 3576 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l); 3577 } 3578 } 3579 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3580 } 3581 3582 /* 3583 * Read wrapper for fifos. 3584 */ 3585 int 3586 nfsfifo_read(v) 3587 void *v; 3588 { 3589 struct vop_read_args /* { 3590 struct vnode *a_vp; 3591 struct uio *a_uio; 3592 int a_ioflag; 3593 kauth_cred_t a_cred; 3594 } */ *ap = v; 3595 struct nfsnode *np = VTONFS(ap->a_vp); 3596 3597 /* 3598 * Set access flag. 3599 */ 3600 np->n_flag |= NACC; 3601 getnanotime(&np->n_atim); 3602 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3603 } 3604 3605 /* 3606 * Write wrapper for fifos. 3607 */ 3608 int 3609 nfsfifo_write(v) 3610 void *v; 3611 { 3612 struct vop_write_args /* { 3613 struct vnode *a_vp; 3614 struct uio *a_uio; 3615 int a_ioflag; 3616 kauth_cred_t a_cred; 3617 } */ *ap = v; 3618 struct nfsnode *np = VTONFS(ap->a_vp); 3619 3620 /* 3621 * Set update flag. 3622 */ 3623 np->n_flag |= NUPD; 3624 getnanotime(&np->n_mtim); 3625 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3626 } 3627 3628 /* 3629 * Close wrapper for fifos. 3630 * 3631 * Update the times on the nfsnode then do fifo close. 3632 */ 3633 int 3634 nfsfifo_close(v) 3635 void *v; 3636 { 3637 struct vop_close_args /* { 3638 struct vnode *a_vp; 3639 int a_fflag; 3640 kauth_cred_t a_cred; 3641 struct lwp *a_l; 3642 } */ *ap = v; 3643 struct vnode *vp = ap->a_vp; 3644 struct nfsnode *np = VTONFS(vp); 3645 struct vattr vattr; 3646 3647 if (np->n_flag & (NACC | NUPD)) { 3648 struct timespec ts; 3649 3650 getnanotime(&ts); 3651 if (np->n_flag & NACC) 3652 np->n_atim = ts; 3653 if (np->n_flag & NUPD) 3654 np->n_mtim = ts; 3655 np->n_flag |= NCHG; 3656 if (vp->v_usecount == 1 && 3657 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3658 VATTR_NULL(&vattr); 3659 if (np->n_flag & NACC) 3660 vattr.va_atime = np->n_atim; 3661 if (np->n_flag & NUPD) 3662 vattr.va_mtime = np->n_mtim; 3663 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_l); 3664 } 3665 } 3666 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3667 } 3668
Indexes created Wed Sep 24 15:09:51 GMT 2025