nfs_vnops.c revision 1.191
1 /* $NetBSD: nfs_vnops.c,v 1.191 2004/04/05 10:44:09 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 1.191 2004/04/05 10:44:09 yamt Exp $"); 43 44 #include "opt_nfs.h" 45 #include "opt_uvmhist.h" 46 47 #include <sys/param.h> 48 #include <sys/proc.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/vnode.h> 59 #include <sys/dirent.h> 60 #include <sys/fcntl.h> 61 #include <sys/hash.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/unistd.h> 65 66 #include <uvm/uvm_extern.h> 67 #include <uvm/uvm.h> 68 69 #include <miscfs/fifofs/fifo.h> 70 #include <miscfs/genfs/genfs.h> 71 #include <miscfs/specfs/specdev.h> 72 73 #include <nfs/rpcv2.h> 74 #include <nfs/nfsproto.h> 75 #include <nfs/nfs.h> 76 #include <nfs/nfsnode.h> 77 #include <nfs/nfsmount.h> 78 #include <nfs/xdr_subs.h> 79 #include <nfs/nfsm_subs.h> 80 #include <nfs/nqnfs.h> 81 #include <nfs/nfs_var.h> 82 83 #include <net/if.h> 84 #include <netinet/in.h> 85 #include <netinet/in_var.h> 86 87 /* 88 * Global vfs data structures for nfs 89 */ 90 int (**nfsv2_vnodeop_p) __P((void *)); 91 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 92 { &vop_default_desc, vn_default_error }, 93 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 94 { &vop_create_desc, nfs_create }, /* create */ 95 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 96 { &vop_open_desc, nfs_open }, /* open */ 97 { &vop_close_desc, nfs_close }, /* close */ 98 { &vop_access_desc, nfs_access }, /* access */ 99 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 100 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 101 { &vop_read_desc, nfs_read }, /* read */ 102 { &vop_write_desc, nfs_write }, /* write */ 103 { &vop_lease_desc, nfs_lease_check }, /* lease */ 104 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 105 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 106 { &vop_poll_desc, nfs_poll }, /* poll */ 107 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 108 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 109 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 110 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 111 { &vop_seek_desc, nfs_seek }, /* seek */ 112 { &vop_remove_desc, nfs_remove }, /* remove */ 113 { &vop_link_desc, nfs_link }, /* link */ 114 { &vop_rename_desc, nfs_rename }, /* rename */ 115 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 116 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 117 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 118 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 119 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 120 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 121 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 122 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 123 { &vop_lock_desc, nfs_lock }, /* lock */ 124 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 125 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 126 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 127 { &vop_print_desc, nfs_print }, /* print */ 128 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 129 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 130 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 131 { &vop_blkatoff_desc, nfs_blkatoff }, /* blkatoff */ 132 { &vop_valloc_desc, nfs_valloc }, /* valloc */ 133 { &vop_reallocblks_desc, nfs_reallocblks }, /* reallocblks */ 134 { &vop_vfree_desc, nfs_vfree }, /* vfree */ 135 { &vop_truncate_desc, nfs_truncate }, /* truncate */ 136 { &vop_update_desc, nfs_update }, /* update */ 137 { &vop_bwrite_desc, nfs_bwrite }, /* bwrite */ 138 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 139 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 140 { NULL, NULL } 141 }; 142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 144 145 /* 146 * Special device vnode ops 147 */ 148 int (**spec_nfsv2nodeop_p) __P((void *)); 149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 150 { &vop_default_desc, vn_default_error }, 151 { &vop_lookup_desc, spec_lookup }, /* lookup */ 152 { &vop_create_desc, spec_create }, /* create */ 153 { &vop_mknod_desc, spec_mknod }, /* mknod */ 154 { &vop_open_desc, spec_open }, /* open */ 155 { &vop_close_desc, nfsspec_close }, /* close */ 156 { &vop_access_desc, nfsspec_access }, /* access */ 157 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 158 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 159 { &vop_read_desc, nfsspec_read }, /* read */ 160 { &vop_write_desc, nfsspec_write }, /* write */ 161 { &vop_lease_desc, spec_lease_check }, /* lease */ 162 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 163 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 164 { &vop_poll_desc, spec_poll }, /* poll */ 165 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 166 { &vop_revoke_desc, spec_revoke }, /* revoke */ 167 { &vop_mmap_desc, spec_mmap }, /* mmap */ 168 { &vop_fsync_desc, spec_fsync }, /* fsync */ 169 { &vop_seek_desc, spec_seek }, /* seek */ 170 { &vop_remove_desc, spec_remove }, /* remove */ 171 { &vop_link_desc, spec_link }, /* link */ 172 { &vop_rename_desc, spec_rename }, /* rename */ 173 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 174 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 175 { &vop_symlink_desc, spec_symlink }, /* symlink */ 176 { &vop_readdir_desc, spec_readdir }, /* readdir */ 177 { &vop_readlink_desc, spec_readlink }, /* readlink */ 178 { &vop_abortop_desc, spec_abortop }, /* abortop */ 179 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 180 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 181 { &vop_lock_desc, nfs_lock }, /* lock */ 182 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 183 { &vop_bmap_desc, spec_bmap }, /* bmap */ 184 { &vop_strategy_desc, spec_strategy }, /* strategy */ 185 { &vop_print_desc, nfs_print }, /* print */ 186 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 187 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 188 { &vop_advlock_desc, spec_advlock }, /* advlock */ 189 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 190 { &vop_valloc_desc, spec_valloc }, /* valloc */ 191 { &vop_reallocblks_desc, spec_reallocblks }, /* reallocblks */ 192 { &vop_vfree_desc, spec_vfree }, /* vfree */ 193 { &vop_truncate_desc, spec_truncate }, /* truncate */ 194 { &vop_update_desc, nfs_update }, /* update */ 195 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 196 { &vop_getpages_desc, spec_getpages }, /* getpages */ 197 { &vop_putpages_desc, spec_putpages }, /* putpages */ 198 { NULL, NULL } 199 }; 200 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 201 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 202 203 int (**fifo_nfsv2nodeop_p) __P((void *)); 204 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 205 { &vop_default_desc, vn_default_error }, 206 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 207 { &vop_create_desc, fifo_create }, /* create */ 208 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 209 { &vop_open_desc, fifo_open }, /* open */ 210 { &vop_close_desc, nfsfifo_close }, /* close */ 211 { &vop_access_desc, nfsspec_access }, /* access */ 212 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 213 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 214 { &vop_read_desc, nfsfifo_read }, /* read */ 215 { &vop_write_desc, nfsfifo_write }, /* write */ 216 { &vop_lease_desc, fifo_lease_check }, /* lease */ 217 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 218 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 219 { &vop_poll_desc, fifo_poll }, /* poll */ 220 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 221 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 222 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 223 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 224 { &vop_seek_desc, fifo_seek }, /* seek */ 225 { &vop_remove_desc, fifo_remove }, /* remove */ 226 { &vop_link_desc, fifo_link }, /* link */ 227 { &vop_rename_desc, fifo_rename }, /* rename */ 228 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 229 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 230 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 231 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 232 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 233 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 234 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 235 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 236 { &vop_lock_desc, nfs_lock }, /* lock */ 237 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 238 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 239 { &vop_strategy_desc, genfs_badop }, /* strategy */ 240 { &vop_print_desc, nfs_print }, /* print */ 241 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 242 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 243 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 244 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 245 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 246 { &vop_reallocblks_desc, fifo_reallocblks }, /* reallocblks */ 247 { &vop_vfree_desc, fifo_vfree }, /* vfree */ 248 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 249 { &vop_update_desc, nfs_update }, /* update */ 250 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 251 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 252 { NULL, NULL } 253 }; 254 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 255 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 256 257 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *); 258 259 /* 260 * Global variables 261 */ 262 extern u_int32_t nfs_true, nfs_false; 263 extern u_int32_t nfs_xdrneg1; 264 extern const nfstype nfsv3_type[9]; 265 266 int nfs_numasync = 0; 267 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 268 269 static void nfs_cache_enter(struct vnode *, struct vnode *, 270 struct componentname *); 271 272 static void 273 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 274 struct componentname *cnp) 275 { 276 277 if (vp != NULL) { 278 struct nfsnode *np = VTONFS(vp); 279 280 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 281 } else { 282 struct nfsnode *dnp = VTONFS(dvp); 283 284 if (!timespecisset(&dnp->n_nctime)) 285 dnp->n_nctime = dnp->n_vattr->va_mtime; 286 } 287 288 cache_enter(dvp, vp, cnp); 289 } 290 291 /* 292 * nfs null call from vfs. 293 */ 294 int 295 nfs_null(vp, cred, procp) 296 struct vnode *vp; 297 struct ucred *cred; 298 struct proc *procp; 299 { 300 caddr_t bpos, dpos; 301 int error = 0; 302 struct mbuf *mreq, *mrep, *md, *mb; 303 struct nfsnode *np = VTONFS(vp); 304 305 nfsm_reqhead(np, NFSPROC_NULL, 0); 306 nfsm_request(np, NFSPROC_NULL, procp, cred); 307 nfsm_reqdone; 308 return (error); 309 } 310 311 /* 312 * nfs access vnode op. 313 * For nfs version 2, just return ok. File accesses may fail later. 314 * For nfs version 3, use the access rpc to check accessibility. If file modes 315 * are changed on the server, accesses might still fail later. 316 */ 317 int 318 nfs_access(v) 319 void *v; 320 { 321 struct vop_access_args /* { 322 struct vnode *a_vp; 323 int a_mode; 324 struct ucred *a_cred; 325 struct proc *a_p; 326 } */ *ap = v; 327 struct vnode *vp = ap->a_vp; 328 u_int32_t *tl; 329 caddr_t cp; 330 int32_t t1, t2; 331 caddr_t bpos, dpos, cp2; 332 int error = 0, attrflag, cachevalid; 333 struct mbuf *mreq, *mrep, *md, *mb; 334 u_int32_t mode, rmode; 335 const int v3 = NFS_ISV3(vp); 336 struct nfsnode *np = VTONFS(vp); 337 338 cachevalid = (np->n_accstamp != -1 && 339 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) && 340 np->n_accuid == ap->a_cred->cr_uid); 341 342 /* 343 * Check access cache first. If this request has been made for this 344 * uid shortly before, use the cached result. 345 */ 346 if (cachevalid) { 347 if (!np->n_accerror) { 348 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 349 return np->n_accerror; 350 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 351 return np->n_accerror; 352 } 353 354 /* 355 * For nfs v3, do an access rpc, otherwise you are stuck emulating 356 * ufs_access() locally using the vattr. This may not be correct, 357 * since the server may apply other access criteria such as 358 * client uid-->server uid mapping that we do not know about, but 359 * this is better than just returning anything that is lying about 360 * in the cache. 361 */ 362 if (v3) { 363 nfsstats.rpccnt[NFSPROC_ACCESS]++; 364 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 365 nfsm_fhtom(np, v3); 366 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 367 if (ap->a_mode & VREAD) 368 mode = NFSV3ACCESS_READ; 369 else 370 mode = 0; 371 if (vp->v_type != VDIR) { 372 if (ap->a_mode & VWRITE) 373 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 374 if (ap->a_mode & VEXEC) 375 mode |= NFSV3ACCESS_EXECUTE; 376 } else { 377 if (ap->a_mode & VWRITE) 378 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 379 NFSV3ACCESS_DELETE); 380 if (ap->a_mode & VEXEC) 381 mode |= NFSV3ACCESS_LOOKUP; 382 } 383 *tl = txdr_unsigned(mode); 384 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred); 385 nfsm_postop_attr(vp, attrflag, 0); 386 if (!error) { 387 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 388 rmode = fxdr_unsigned(u_int32_t, *tl); 389 /* 390 * The NFS V3 spec does not clarify whether or not 391 * the returned access bits can be a superset of 392 * the ones requested, so... 393 */ 394 if ((rmode & mode) != mode) 395 error = EACCES; 396 } 397 nfsm_reqdone; 398 } else 399 return (nfsspec_access(ap)); 400 /* 401 * Disallow write attempts on filesystems mounted read-only; 402 * unless the file is a socket, fifo, or a block or character 403 * device resident on the filesystem. 404 */ 405 if (!error && (ap->a_mode & VWRITE) && 406 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 407 switch (vp->v_type) { 408 case VREG: 409 case VDIR: 410 case VLNK: 411 error = EROFS; 412 default: 413 break; 414 } 415 } 416 417 if (!error || error == EACCES) { 418 /* 419 * If we got the same result as for a previous, 420 * different request, OR it in. Don't update 421 * the timestamp in that case. 422 */ 423 if (cachevalid && np->n_accstamp != -1 && 424 error == np->n_accerror) { 425 if (!error) 426 np->n_accmode |= ap->a_mode; 427 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 428 np->n_accmode = ap->a_mode; 429 } else { 430 np->n_accstamp = time.tv_sec; 431 np->n_accuid = ap->a_cred->cr_uid; 432 np->n_accmode = ap->a_mode; 433 np->n_accerror = error; 434 } 435 } 436 437 return (error); 438 } 439 440 /* 441 * nfs open vnode op 442 * Check to see if the type is ok 443 * and that deletion is not in progress. 444 * For paged in text files, you will need to flush the page cache 445 * if consistency is lost. 446 */ 447 /* ARGSUSED */ 448 int 449 nfs_open(v) 450 void *v; 451 { 452 struct vop_open_args /* { 453 struct vnode *a_vp; 454 int a_mode; 455 struct ucred *a_cred; 456 struct proc *a_p; 457 } */ *ap = v; 458 struct vnode *vp = ap->a_vp; 459 struct nfsnode *np = VTONFS(vp); 460 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 461 struct vattr vattr; 462 int error; 463 464 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 465 return (EACCES); 466 } 467 468 if (ap->a_mode & FREAD) { 469 if (np->n_rcred != NULL) 470 crfree(np->n_rcred); 471 np->n_rcred = ap->a_cred; 472 crhold(np->n_rcred); 473 } 474 if (ap->a_mode & FWRITE) { 475 if (np->n_wcred != NULL) 476 crfree(np->n_wcred); 477 np->n_wcred = ap->a_cred; 478 crhold(np->n_wcred); 479 } 480 481 #ifndef NFS_V2_ONLY 482 /* 483 * Get a valid lease. If cached data is stale, flush it. 484 */ 485 if (nmp->nm_flag & NFSMNT_NQNFS) { 486 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 487 do { 488 error = nqnfs_getlease(vp, ND_READ, ap->a_cred, 489 ap->a_p); 490 } while (error == NQNFS_EXPIRED); 491 if (error) 492 return (error); 493 if (np->n_lrev != np->n_brev || 494 (np->n_flag & NQNFSNONCACHE)) { 495 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 496 ap->a_p, 1)) == EINTR) 497 return (error); 498 np->n_brev = np->n_lrev; 499 } 500 } 501 } else 502 #endif 503 { 504 if (np->n_flag & NMODIFIED) { 505 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 506 ap->a_p, 1)) == EINTR) 507 return (error); 508 NFS_INVALIDATE_ATTRCACHE(np); 509 if (vp->v_type == VDIR) { 510 nfs_invaldircache(vp, 0); 511 np->n_direofoffset = 0; 512 } 513 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p); 514 if (error) 515 return (error); 516 np->n_mtime = vattr.va_mtime; 517 } else { 518 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p); 519 if (error) 520 return (error); 521 if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) { 522 if (vp->v_type == VDIR) { 523 nfs_invaldircache(vp, 0); 524 np->n_direofoffset = 0; 525 } 526 if ((error = nfs_vinvalbuf(vp, V_SAVE, 527 ap->a_cred, ap->a_p, 1)) == EINTR) 528 return (error); 529 np->n_mtime = vattr.va_mtime; 530 } 531 } 532 } 533 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) 534 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 535 return (0); 536 } 537 538 /* 539 * nfs close vnode op 540 * What an NFS client should do upon close after writing is a debatable issue. 541 * Most NFS clients push delayed writes to the server upon close, basically for 542 * two reasons: 543 * 1 - So that any write errors may be reported back to the client process 544 * doing the close system call. By far the two most likely errors are 545 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 546 * 2 - To put a worst case upper bound on cache inconsistency between 547 * multiple clients for the file. 548 * There is also a consistency problem for Version 2 of the protocol w.r.t. 549 * not being able to tell if other clients are writing a file concurrently, 550 * since there is no way of knowing if the changed modify time in the reply 551 * is only due to the write for this client. 552 * (NFS Version 3 provides weak cache consistency data in the reply that 553 * should be sufficient to detect and handle this case.) 554 * 555 * The current code does the following: 556 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 557 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 558 * or commit them (this satisfies 1 and 2 except for the 559 * case where the server crashes after this close but 560 * before the commit RPC, which is felt to be "good 561 * enough". Changing the last argument to nfs_flush() to 562 * a 1 would force a commit operation, if it is felt a 563 * commit is necessary now. 564 * for NQNFS - do nothing now, since 2 is dealt with via leases and 565 * 1 should be dealt with via an fsync() system call for 566 * cases where write errors are important. 567 */ 568 /* ARGSUSED */ 569 int 570 nfs_close(v) 571 void *v; 572 { 573 struct vop_close_args /* { 574 struct vnodeop_desc *a_desc; 575 struct vnode *a_vp; 576 int a_fflag; 577 struct ucred *a_cred; 578 struct proc *a_p; 579 } */ *ap = v; 580 struct vnode *vp = ap->a_vp; 581 struct nfsnode *np = VTONFS(vp); 582 int error = 0; 583 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 584 585 if (vp->v_type == VREG) { 586 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 587 (np->n_flag & NMODIFIED)) { 588 if (NFS_ISV3(vp)) { 589 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0); 590 np->n_flag &= ~NMODIFIED; 591 } else 592 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1); 593 NFS_INVALIDATE_ATTRCACHE(np); 594 } 595 if (np->n_flag & NWRITEERR) { 596 np->n_flag &= ~NWRITEERR; 597 error = np->n_error; 598 } 599 } 600 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 601 return (error); 602 } 603 604 /* 605 * nfs getattr call from vfs. 606 */ 607 int 608 nfs_getattr(v) 609 void *v; 610 { 611 struct vop_getattr_args /* { 612 struct vnode *a_vp; 613 struct vattr *a_vap; 614 struct ucred *a_cred; 615 struct proc *a_p; 616 } */ *ap = v; 617 struct vnode *vp = ap->a_vp; 618 struct nfsnode *np = VTONFS(vp); 619 caddr_t cp; 620 u_int32_t *tl; 621 int32_t t1, t2; 622 caddr_t bpos, dpos; 623 int error = 0; 624 struct mbuf *mreq, *mrep, *md, *mb; 625 const int v3 = NFS_ISV3(vp); 626 627 /* 628 * Update local times for special files. 629 */ 630 if (np->n_flag & (NACC | NUPD)) 631 np->n_flag |= NCHG; 632 633 /* 634 * if we have delayed truncation, do it now. 635 */ 636 nfs_delayedtruncate(vp); 637 638 /* 639 * First look in the cache. 640 */ 641 if (nfs_getattrcache(vp, ap->a_vap) == 0) 642 return (0); 643 nfsstats.rpccnt[NFSPROC_GETATTR]++; 644 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 645 nfsm_fhtom(np, v3); 646 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred); 647 if (!error) { 648 nfsm_loadattr(vp, ap->a_vap, 0); 649 if (vp->v_type == VDIR && 650 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 651 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 652 } 653 nfsm_reqdone; 654 return (error); 655 } 656 657 /* 658 * nfs setattr call. 659 */ 660 int 661 nfs_setattr(v) 662 void *v; 663 { 664 struct vop_setattr_args /* { 665 struct vnodeop_desc *a_desc; 666 struct vnode *a_vp; 667 struct vattr *a_vap; 668 struct ucred *a_cred; 669 struct proc *a_p; 670 } */ *ap = v; 671 struct vnode *vp = ap->a_vp; 672 struct nfsnode *np = VTONFS(vp); 673 struct vattr *vap = ap->a_vap; 674 int error = 0; 675 u_quad_t tsize = 0; 676 677 /* 678 * Setting of flags is not supported. 679 */ 680 if (vap->va_flags != VNOVAL) 681 return (EOPNOTSUPP); 682 683 /* 684 * Disallow write attempts if the filesystem is mounted read-only. 685 */ 686 if ((vap->va_uid != (uid_t)VNOVAL || 687 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 688 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 689 (vp->v_mount->mnt_flag & MNT_RDONLY)) 690 return (EROFS); 691 if (vap->va_size != VNOVAL) { 692 switch (vp->v_type) { 693 case VDIR: 694 return (EISDIR); 695 case VCHR: 696 case VBLK: 697 case VSOCK: 698 case VFIFO: 699 if (vap->va_mtime.tv_sec == VNOVAL && 700 vap->va_atime.tv_sec == VNOVAL && 701 vap->va_mode == (mode_t)VNOVAL && 702 vap->va_uid == (uid_t)VNOVAL && 703 vap->va_gid == (gid_t)VNOVAL) 704 return (0); 705 vap->va_size = VNOVAL; 706 break; 707 default: 708 /* 709 * Disallow write attempts if the filesystem is 710 * mounted read-only. 711 */ 712 if (vp->v_mount->mnt_flag & MNT_RDONLY) 713 return (EROFS); 714 uvm_vnp_setsize(vp, vap->va_size); 715 tsize = np->n_size; 716 np->n_size = vap->va_size; 717 if (vap->va_size == 0) 718 error = nfs_vinvalbuf(vp, 0, 719 ap->a_cred, ap->a_p, 1); 720 else 721 error = nfs_vinvalbuf(vp, V_SAVE, 722 ap->a_cred, ap->a_p, 1); 723 if (error) { 724 uvm_vnp_setsize(vp, tsize); 725 return (error); 726 } 727 np->n_vattr->va_size = vap->va_size; 728 } 729 } else if ((vap->va_mtime.tv_sec != VNOVAL || 730 vap->va_atime.tv_sec != VNOVAL) && 731 vp->v_type == VREG && 732 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 733 ap->a_p, 1)) == EINTR) 734 return (error); 735 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p); 736 if (error && vap->va_size != VNOVAL) { 737 np->n_size = np->n_vattr->va_size = tsize; 738 uvm_vnp_setsize(vp, np->n_size); 739 } 740 VN_KNOTE(vp, NOTE_ATTRIB); 741 return (error); 742 } 743 744 /* 745 * Do an nfs setattr rpc. 746 */ 747 int 748 nfs_setattrrpc(vp, vap, cred, procp) 749 struct vnode *vp; 750 struct vattr *vap; 751 struct ucred *cred; 752 struct proc *procp; 753 { 754 struct nfsv2_sattr *sp; 755 caddr_t cp; 756 int32_t t1, t2; 757 caddr_t bpos, dpos, cp2; 758 u_int32_t *tl; 759 int error = 0, wccflag = NFSV3_WCCRATTR; 760 struct mbuf *mreq, *mrep, *md, *mb; 761 const int v3 = NFS_ISV3(vp); 762 struct nfsnode *np = VTONFS(vp); 763 764 nfsstats.rpccnt[NFSPROC_SETATTR]++; 765 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 766 nfsm_fhtom(np, v3); 767 if (v3) { 768 nfsm_v3attrbuild(vap, TRUE); 769 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 770 *tl = nfs_false; 771 } else { 772 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 773 if (vap->va_mode == (mode_t)VNOVAL) 774 sp->sa_mode = nfs_xdrneg1; 775 else 776 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 777 if (vap->va_uid == (uid_t)VNOVAL) 778 sp->sa_uid = nfs_xdrneg1; 779 else 780 sp->sa_uid = txdr_unsigned(vap->va_uid); 781 if (vap->va_gid == (gid_t)VNOVAL) 782 sp->sa_gid = nfs_xdrneg1; 783 else 784 sp->sa_gid = txdr_unsigned(vap->va_gid); 785 sp->sa_size = txdr_unsigned(vap->va_size); 786 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 787 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 788 } 789 nfsm_request(np, NFSPROC_SETATTR, procp, cred); 790 if (v3) { 791 nfsm_wcc_data(vp, wccflag, 0); 792 } else 793 nfsm_loadattr(vp, (struct vattr *)0, 0); 794 nfsm_reqdone; 795 return (error); 796 } 797 798 /* 799 * nfs lookup call, one step at a time... 800 * First look in cache 801 * If not found, unlock the directory nfsnode and do the rpc 802 * 803 * This code is full of lock/unlock statements and checks, because 804 * we continue after cache_lookup has finished (we need to check 805 * with the attr cache and do an rpc if it has timed out). This means 806 * that the locking effects of cache_lookup have to be taken into 807 * account. 808 */ 809 int 810 nfs_lookup(v) 811 void *v; 812 { 813 struct vop_lookup_args /* { 814 struct vnodeop_desc *a_desc; 815 struct vnode *a_dvp; 816 struct vnode **a_vpp; 817 struct componentname *a_cnp; 818 } */ *ap = v; 819 struct componentname *cnp = ap->a_cnp; 820 struct vnode *dvp = ap->a_dvp; 821 struct vnode **vpp = ap->a_vpp; 822 int flags; 823 struct vnode *newvp; 824 u_int32_t *tl; 825 caddr_t cp; 826 int32_t t1, t2; 827 caddr_t bpos, dpos, cp2; 828 struct mbuf *mreq, *mrep, *md, *mb; 829 long len; 830 nfsfh_t *fhp; 831 struct nfsnode *np; 832 int lockparent, wantparent, error = 0, attrflag, fhsize; 833 const int v3 = NFS_ISV3(dvp); 834 835 cnp->cn_flags &= ~PDIRUNLOCK; 836 flags = cnp->cn_flags; 837 838 *vpp = NULLVP; 839 newvp = NULLVP; 840 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 841 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 842 return (EROFS); 843 if (dvp->v_type != VDIR) 844 return (ENOTDIR); 845 846 lockparent = flags & LOCKPARENT; 847 wantparent = flags & (LOCKPARENT|WANTPARENT); 848 np = VTONFS(dvp); 849 850 /* 851 * Before tediously performing a linear scan of the directory, 852 * check the name cache to see if the directory/name pair 853 * we are looking for is known already. 854 * If the directory/name pair is found in the name cache, 855 * we have to ensure the directory has not changed from 856 * the time the cache entry has been created. If it has, 857 * the cache entry has to be ignored. 858 */ 859 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) { 860 struct vattr vattr; 861 int err2; 862 863 if (error && error != ENOENT) { 864 *vpp = NULLVP; 865 return error; 866 } 867 868 if (cnp->cn_flags & PDIRUNLOCK) { 869 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 870 if (err2 != 0) { 871 *vpp = NULLVP; 872 return err2; 873 } 874 cnp->cn_flags &= ~PDIRUNLOCK; 875 } 876 877 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc); 878 if (err2 != 0) { 879 if (error == 0) { 880 if (*vpp != dvp) 881 vput(*vpp); 882 else 883 vrele(*vpp); 884 } 885 *vpp = NULLVP; 886 return err2; 887 } 888 889 if (error == ENOENT) { 890 if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred, 891 cnp->cn_proc) && timespeccmp(&vattr.va_mtime, 892 &VTONFS(dvp)->n_nctime, ==)) 893 return ENOENT; 894 cache_purge1(dvp, NULL, PURGE_CHILDREN); 895 timespecclear(&np->n_nctime); 896 goto dorpc; 897 } 898 899 newvp = *vpp; 900 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc) 901 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) 902 { 903 nfsstats.lookupcache_hits++; 904 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 905 cnp->cn_flags |= SAVENAME; 906 if ((!lockparent || !(flags & ISLASTCN)) && 907 newvp != dvp) 908 VOP_UNLOCK(dvp, 0); 909 KASSERT(newvp->v_type != VNON); 910 return (0); 911 } 912 cache_purge1(newvp, NULL, PURGE_PARENTS); 913 if (newvp != dvp) 914 vput(newvp); 915 else 916 vrele(newvp); 917 *vpp = NULLVP; 918 } 919 dorpc: 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_proc, 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 && wantparent && (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 if (v3) { 952 nfsm_postop_attr(newvp, attrflag, 0); 953 nfsm_postop_attr(dvp, attrflag, 0); 954 } else 955 nfsm_loadattr(newvp, (struct vattr *)0, 0); 956 *vpp = newvp; 957 m_freem(mrep); 958 cnp->cn_flags |= SAVENAME; 959 if (!lockparent) { 960 VOP_UNLOCK(dvp, 0); 961 cnp->cn_flags |= PDIRUNLOCK; 962 } 963 goto validate; 964 } 965 966 /* 967 * The postop attr handling is duplicated for each if case, 968 * because it should be done while dvp is locked (unlocking 969 * dvp is different for each case). 970 */ 971 972 if (NFS_CMPFH(np, fhp, fhsize)) { 973 /* 974 * "." lookup 975 */ 976 VREF(dvp); 977 newvp = dvp; 978 if (v3) { 979 nfsm_postop_attr(newvp, attrflag, 0); 980 nfsm_postop_attr(dvp, attrflag, 0); 981 } else 982 nfsm_loadattr(newvp, (struct vattr *)0, 0); 983 } else if (flags & ISDOTDOT) { 984 /* 985 * ".." lookup 986 */ 987 VOP_UNLOCK(dvp, 0); 988 cnp->cn_flags |= PDIRUNLOCK; 989 990 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 991 if (error) { 992 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 993 cnp->cn_flags &= ~PDIRUNLOCK; 994 m_freem(mrep); 995 return error; 996 } 997 newvp = NFSTOV(np); 998 999 if (v3) { 1000 nfsm_postop_attr(newvp, attrflag, 0); 1001 nfsm_postop_attr(dvp, attrflag, 0); 1002 } else 1003 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1004 1005 if (lockparent && (flags & ISLASTCN)) { 1006 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) { 1007 m_freem(mrep); 1008 vput(newvp); 1009 return error; 1010 } 1011 cnp->cn_flags &= ~PDIRUNLOCK; 1012 } 1013 } else { 1014 /* 1015 * Other lookups. 1016 */ 1017 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1018 if (error) { 1019 m_freem(mrep); 1020 return error; 1021 } 1022 newvp = NFSTOV(np); 1023 if (v3) { 1024 nfsm_postop_attr(newvp, attrflag, 0); 1025 nfsm_postop_attr(dvp, attrflag, 0); 1026 } else 1027 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1028 if (!lockparent || !(flags & ISLASTCN)) { 1029 VOP_UNLOCK(dvp, 0); 1030 cnp->cn_flags |= PDIRUNLOCK; 1031 } 1032 } 1033 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1034 cnp->cn_flags |= SAVENAME; 1035 if ((cnp->cn_flags & MAKEENTRY) && 1036 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1037 nfs_cache_enter(dvp, newvp, cnp); 1038 } 1039 *vpp = newvp; 1040 nfsm_reqdone; 1041 if (error) { 1042 /* 1043 * We get here only because of errors returned by 1044 * the RPC. Otherwise we'll have returned above 1045 * (the nfsm_* macros will jump to nfsm_reqdone 1046 * on error). 1047 */ 1048 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1049 cnp->cn_nameiop != CREATE) { 1050 nfs_cache_enter(dvp, NULL, cnp); 1051 } 1052 if (newvp != NULLVP) { 1053 vrele(newvp); 1054 if (newvp != dvp) 1055 VOP_UNLOCK(newvp, 0); 1056 } 1057 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1058 (flags & ISLASTCN) && error == ENOENT) { 1059 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1060 error = EROFS; 1061 else 1062 error = EJUSTRETURN; 1063 } 1064 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1065 cnp->cn_flags |= SAVENAME; 1066 *vpp = NULL; 1067 return error; 1068 } 1069 1070 validate: 1071 /* 1072 * make sure we have valid type and size. 1073 */ 1074 1075 newvp = *vpp; 1076 if (newvp->v_type == VNON) { 1077 struct vattr vattr; /* dummy */ 1078 1079 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1080 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc); 1081 if (error) { 1082 vput(newvp); 1083 *vpp = NULL; 1084 } 1085 } 1086 1087 return error; 1088 } 1089 1090 /* 1091 * nfs read call. 1092 * Just call nfs_bioread() to do the work. 1093 */ 1094 int 1095 nfs_read(v) 1096 void *v; 1097 { 1098 struct vop_read_args /* { 1099 struct vnode *a_vp; 1100 struct uio *a_uio; 1101 int a_ioflag; 1102 struct ucred *a_cred; 1103 } */ *ap = v; 1104 struct vnode *vp = ap->a_vp; 1105 1106 if (vp->v_type != VREG) 1107 return (EPERM); 1108 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1109 } 1110 1111 /* 1112 * nfs readlink call 1113 */ 1114 int 1115 nfs_readlink(v) 1116 void *v; 1117 { 1118 struct vop_readlink_args /* { 1119 struct vnode *a_vp; 1120 struct uio *a_uio; 1121 struct ucred *a_cred; 1122 } */ *ap = v; 1123 struct vnode *vp = ap->a_vp; 1124 1125 if (vp->v_type != VLNK) 1126 return (EPERM); 1127 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1128 } 1129 1130 /* 1131 * Do a readlink rpc. 1132 * Called by nfs_doio() from below the buffer cache. 1133 */ 1134 int 1135 nfs_readlinkrpc(vp, uiop, cred) 1136 struct vnode *vp; 1137 struct uio *uiop; 1138 struct ucred *cred; 1139 { 1140 u_int32_t *tl; 1141 caddr_t cp; 1142 int32_t t1, t2; 1143 caddr_t bpos, dpos, cp2; 1144 int error = 0, attrflag; 1145 uint32_t len; 1146 struct mbuf *mreq, *mrep, *md, *mb; 1147 const int v3 = NFS_ISV3(vp); 1148 struct nfsnode *np = VTONFS(vp); 1149 1150 nfsstats.rpccnt[NFSPROC_READLINK]++; 1151 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1152 nfsm_fhtom(np, v3); 1153 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred); 1154 if (v3) 1155 nfsm_postop_attr(vp, attrflag, 0); 1156 if (!error) { 1157 if (v3) { 1158 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1159 len = fxdr_unsigned(uint32_t, *tl); 1160 if (len > MAXPATHLEN) { 1161 /* 1162 * this pathname is too long for us. 1163 */ 1164 m_freem(mrep); 1165 /* Solaris returns EINVAL. should we follow? */ 1166 error = ENAMETOOLONG; 1167 goto nfsmout; 1168 } 1169 } else { 1170 nfsm_strsiz(len, NFS_MAXPATHLEN); 1171 } 1172 nfsm_mtouio(uiop, len); 1173 } 1174 nfsm_reqdone; 1175 return (error); 1176 } 1177 1178 /* 1179 * nfs read rpc call 1180 * Ditto above 1181 */ 1182 int 1183 nfs_readrpc(vp, uiop) 1184 struct vnode *vp; 1185 struct uio *uiop; 1186 { 1187 u_int32_t *tl; 1188 caddr_t cp; 1189 int32_t t1, t2; 1190 caddr_t bpos, dpos, cp2; 1191 struct mbuf *mreq, *mrep, *md, *mb; 1192 struct nfsmount *nmp; 1193 int error = 0, len, retlen, tsiz, eof, attrflag; 1194 const int v3 = NFS_ISV3(vp); 1195 struct nfsnode *np = VTONFS(vp); 1196 1197 #ifndef nolint 1198 eof = 0; 1199 #endif 1200 nmp = VFSTONFS(vp->v_mount); 1201 tsiz = uiop->uio_resid; 1202 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1203 return (EFBIG); 1204 while (tsiz > 0) { 1205 nfsstats.rpccnt[NFSPROC_READ]++; 1206 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1207 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1208 nfsm_fhtom(np, v3); 1209 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1210 if (v3) { 1211 txdr_hyper(uiop->uio_offset, tl); 1212 *(tl + 2) = txdr_unsigned(len); 1213 } else { 1214 *tl++ = txdr_unsigned(uiop->uio_offset); 1215 *tl++ = txdr_unsigned(len); 1216 *tl = 0; 1217 } 1218 nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred); 1219 if (v3) { 1220 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1221 if (error) { 1222 m_freem(mrep); 1223 goto nfsmout; 1224 } 1225 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1226 eof = fxdr_unsigned(int, *(tl + 1)); 1227 } else 1228 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1229 nfsm_strsiz(retlen, nmp->nm_rsize); 1230 nfsm_mtouio(uiop, retlen); 1231 m_freem(mrep); 1232 tsiz -= retlen; 1233 if (v3) { 1234 if (eof || retlen == 0) 1235 tsiz = 0; 1236 } else if (retlen < len) 1237 tsiz = 0; 1238 } 1239 nfsmout: 1240 return (error); 1241 } 1242 1243 struct nfs_writerpc_context { 1244 struct simplelock nwc_slock; 1245 volatile int nwc_mbufcount; 1246 }; 1247 1248 /* 1249 * free mbuf used to refer protected pages while write rpc call. 1250 * called at splvm. 1251 */ 1252 static void 1253 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 1254 { 1255 struct nfs_writerpc_context *ctx = arg; 1256 1257 KASSERT(m != NULL); 1258 KASSERT(ctx != NULL); 1259 pool_cache_put(&mbpool_cache, m); 1260 simple_lock(&ctx->nwc_slock); 1261 if (--ctx->nwc_mbufcount == 0) { 1262 wakeup(ctx); 1263 } 1264 simple_unlock(&ctx->nwc_slock); 1265 } 1266 1267 /* 1268 * nfs write call 1269 */ 1270 int 1271 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1272 struct vnode *vp; 1273 struct uio *uiop; 1274 int *iomode; 1275 boolean_t pageprotected; 1276 boolean_t *stalewriteverfp; 1277 { 1278 u_int32_t *tl; 1279 caddr_t cp; 1280 int32_t t1, t2; 1281 caddr_t bpos, dpos, cp2; 1282 struct mbuf *mreq, *mrep, *md, *mb; 1283 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1284 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1285 const int v3 = NFS_ISV3(vp); 1286 int committed = NFSV3WRITE_FILESYNC; 1287 struct nfsnode *np = VTONFS(vp); 1288 struct nfs_writerpc_context ctx; 1289 int s; 1290 struct lwp *l = NULL; 1291 size_t origresid; 1292 1293 simple_lock_init(&ctx.nwc_slock); 1294 ctx.nwc_mbufcount = 1; 1295 1296 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1297 panic("writerpc readonly vp %p", vp); 1298 } 1299 1300 #ifdef DIAGNOSTIC 1301 if (uiop->uio_iovcnt != 1) 1302 panic("nfs: writerpc iovcnt > 1"); 1303 #endif 1304 tsiz = uiop->uio_resid; 1305 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1306 return (EFBIG); 1307 if (pageprotected) { 1308 l = curlwp; 1309 PHOLD(l); 1310 } 1311 retry: 1312 origresid = uiop->uio_resid; 1313 KASSERT(origresid == uiop->uio_iov->iov_len); 1314 while (tsiz > 0) { 1315 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1316 uint32_t backup; 1317 boolean_t stalewriteverf = FALSE; 1318 1319 nfsstats.rpccnt[NFSPROC_WRITE]++; 1320 len = min(tsiz, nmp->nm_wsize); 1321 datalen = pageprotected ? 0 : nfsm_rndup(len); 1322 nfsm_reqhead(np, NFSPROC_WRITE, 1323 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1324 nfsm_fhtom(np, v3); 1325 if (v3) { 1326 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1327 txdr_hyper(uiop->uio_offset, tl); 1328 tl += 2; 1329 *tl++ = txdr_unsigned(len); 1330 *tl++ = txdr_unsigned(*iomode); 1331 *tl = txdr_unsigned(len); 1332 } else { 1333 u_int32_t x; 1334 1335 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1336 /* Set both "begin" and "current" to non-garbage. */ 1337 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1338 *tl++ = x; /* "begin offset" */ 1339 *tl++ = x; /* "current offset" */ 1340 x = txdr_unsigned(len); 1341 *tl++ = x; /* total to this offset */ 1342 *tl = x; /* size of this write */ 1343 1344 } 1345 if (pageprotected) { 1346 /* 1347 * since we know pages can't be modified during i/o, 1348 * no need to copy them for us. 1349 */ 1350 struct mbuf *m; 1351 struct iovec *iovp = uiop->uio_iov; 1352 1353 m = m_get(M_WAIT, MT_DATA); 1354 MCLAIM(m, &nfs_mowner); 1355 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1356 nfs_writerpc_extfree, &ctx); 1357 m->m_flags |= M_EXT_ROMAP; 1358 m->m_len = len; 1359 mb->m_next = m; 1360 /* 1361 * no need to maintain mb and bpos here 1362 * because no one care them later. 1363 */ 1364 #if 0 1365 mb = m; 1366 bpos = mtod(caddr_t, mb) + mb->m_len; 1367 #endif 1368 iovp->iov_base = (char *)iovp->iov_base + len; 1369 iovp->iov_len -= len; 1370 uiop->uio_offset += len; 1371 uiop->uio_resid -= len; 1372 s = splvm(); 1373 simple_lock(&ctx.nwc_slock); 1374 ctx.nwc_mbufcount++; 1375 simple_unlock(&ctx.nwc_slock); 1376 splx(s); 1377 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1378 } else { 1379 nfsm_uiotom(uiop, len); 1380 } 1381 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred); 1382 if (v3) { 1383 wccflag = NFSV3_WCCCHK; 1384 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC); 1385 if (!error) { 1386 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1387 + NFSX_V3WRITEVERF); 1388 rlen = fxdr_unsigned(int, *tl++); 1389 if (rlen == 0) { 1390 error = NFSERR_IO; 1391 m_freem(mrep); 1392 break; 1393 } else if (rlen < len) { 1394 backup = len - rlen; 1395 uiop->uio_iov->iov_base = 1396 (caddr_t)uiop->uio_iov->iov_base - 1397 backup; 1398 uiop->uio_iov->iov_len += backup; 1399 uiop->uio_offset -= backup; 1400 uiop->uio_resid += backup; 1401 len = rlen; 1402 } 1403 commit = fxdr_unsigned(int, *tl++); 1404 1405 /* 1406 * Return the lowest committment level 1407 * obtained by any of the RPCs. 1408 */ 1409 if (committed == NFSV3WRITE_FILESYNC) 1410 committed = commit; 1411 else if (committed == NFSV3WRITE_DATASYNC && 1412 commit == NFSV3WRITE_UNSTABLE) 1413 committed = commit; 1414 simple_lock(&nmp->nm_slock); 1415 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1416 memcpy(nmp->nm_writeverf, tl, 1417 NFSX_V3WRITEVERF); 1418 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1419 } else if ((nmp->nm_iflag & 1420 NFSMNT_STALEWRITEVERF) || 1421 memcmp(tl, nmp->nm_writeverf, 1422 NFSX_V3WRITEVERF)) { 1423 memcpy(nmp->nm_writeverf, tl, 1424 NFSX_V3WRITEVERF); 1425 /* 1426 * note NFSMNT_STALEWRITEVERF 1427 * if we're the first thread to 1428 * notice it. 1429 */ 1430 if ((nmp->nm_iflag & 1431 NFSMNT_STALEWRITEVERF) == 0) { 1432 stalewriteverf = TRUE; 1433 nmp->nm_iflag |= 1434 NFSMNT_STALEWRITEVERF; 1435 } 1436 } 1437 simple_unlock(&nmp->nm_slock); 1438 } 1439 } else 1440 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1441 if (wccflag) 1442 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1443 m_freem(mrep); 1444 if (error) 1445 break; 1446 tsiz -= len; 1447 if (stalewriteverf) { 1448 *stalewriteverfp = TRUE; 1449 stalewriteverf = FALSE; 1450 if (committed == NFSV3WRITE_UNSTABLE && 1451 len != origresid) { 1452 /* 1453 * if our write requests weren't atomic but 1454 * unstable, datas in previous iterations 1455 * might have already been lost now. 1456 * then, we should resend them to nfsd. 1457 */ 1458 backup = origresid - tsiz; 1459 uiop->uio_iov->iov_base = 1460 (caddr_t)uiop->uio_iov->iov_base - backup; 1461 uiop->uio_iov->iov_len += backup; 1462 uiop->uio_offset -= backup; 1463 uiop->uio_resid += backup; 1464 tsiz = origresid; 1465 goto retry; 1466 } 1467 } 1468 } 1469 if (pageprotected) { 1470 /* 1471 * wait until mbufs go away. 1472 * retransmitted mbufs can survive longer than rpc requests 1473 * themselves. 1474 */ 1475 s = splvm(); 1476 simple_lock(&ctx.nwc_slock); 1477 ctx.nwc_mbufcount--; 1478 while (ctx.nwc_mbufcount > 0) { 1479 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock); 1480 } 1481 simple_unlock(&ctx.nwc_slock); 1482 splx(s); 1483 PRELE(l); 1484 } 1485 nfsmout: 1486 *iomode = committed; 1487 if (error) 1488 uiop->uio_resid = tsiz; 1489 return (error); 1490 } 1491 1492 /* 1493 * nfs mknod rpc 1494 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1495 * mode set to specify the file type and the size field for rdev. 1496 */ 1497 int 1498 nfs_mknodrpc(dvp, vpp, cnp, vap) 1499 struct vnode *dvp; 1500 struct vnode **vpp; 1501 struct componentname *cnp; 1502 struct vattr *vap; 1503 { 1504 struct nfsv2_sattr *sp; 1505 u_int32_t *tl; 1506 caddr_t cp; 1507 int32_t t1, t2; 1508 struct vnode *newvp = (struct vnode *)0; 1509 struct nfsnode *dnp, *np; 1510 char *cp2; 1511 caddr_t bpos, dpos; 1512 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1513 struct mbuf *mreq, *mrep, *md, *mb; 1514 u_int32_t rdev; 1515 const int v3 = NFS_ISV3(dvp); 1516 1517 if (vap->va_type == VCHR || vap->va_type == VBLK) 1518 rdev = txdr_unsigned(vap->va_rdev); 1519 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1520 rdev = nfs_xdrneg1; 1521 else { 1522 VOP_ABORTOP(dvp, cnp); 1523 vput(dvp); 1524 return (EOPNOTSUPP); 1525 } 1526 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1527 dnp = VTONFS(dvp); 1528 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1529 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1530 nfsm_fhtom(dnp, v3); 1531 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1532 if (v3) { 1533 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1534 *tl++ = vtonfsv3_type(vap->va_type); 1535 nfsm_v3attrbuild(vap, FALSE); 1536 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1537 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1538 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1539 *tl = txdr_unsigned(minor(vap->va_rdev)); 1540 } 1541 } else { 1542 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1543 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1544 sp->sa_uid = nfs_xdrneg1; 1545 sp->sa_gid = nfs_xdrneg1; 1546 sp->sa_size = rdev; 1547 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1548 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1549 } 1550 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred); 1551 if (!error) { 1552 nfsm_mtofh(dvp, newvp, v3, gotvp); 1553 if (!gotvp) { 1554 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1555 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1556 if (!error) 1557 newvp = NFSTOV(np); 1558 } 1559 } 1560 if (v3) 1561 nfsm_wcc_data(dvp, wccflag, 0); 1562 nfsm_reqdone; 1563 if (error) { 1564 if (newvp) 1565 vput(newvp); 1566 } else { 1567 if (cnp->cn_flags & MAKEENTRY) 1568 nfs_cache_enter(dvp, newvp, cnp); 1569 *vpp = newvp; 1570 } 1571 PNBUF_PUT(cnp->cn_pnbuf); 1572 VTONFS(dvp)->n_flag |= NMODIFIED; 1573 if (!wccflag) 1574 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1575 vput(dvp); 1576 return (error); 1577 } 1578 1579 /* 1580 * nfs mknod vop 1581 * just call nfs_mknodrpc() to do the work. 1582 */ 1583 /* ARGSUSED */ 1584 int 1585 nfs_mknod(v) 1586 void *v; 1587 { 1588 struct vop_mknod_args /* { 1589 struct vnode *a_dvp; 1590 struct vnode **a_vpp; 1591 struct componentname *a_cnp; 1592 struct vattr *a_vap; 1593 } */ *ap = v; 1594 int error; 1595 1596 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1597 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1598 return (error); 1599 } 1600 1601 static u_long create_verf; 1602 /* 1603 * nfs file create call 1604 */ 1605 int 1606 nfs_create(v) 1607 void *v; 1608 { 1609 struct vop_create_args /* { 1610 struct vnode *a_dvp; 1611 struct vnode **a_vpp; 1612 struct componentname *a_cnp; 1613 struct vattr *a_vap; 1614 } */ *ap = v; 1615 struct vnode *dvp = ap->a_dvp; 1616 struct vattr *vap = ap->a_vap; 1617 struct componentname *cnp = ap->a_cnp; 1618 struct nfsv2_sattr *sp; 1619 u_int32_t *tl; 1620 caddr_t cp; 1621 int32_t t1, t2; 1622 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1623 struct vnode *newvp = (struct vnode *)0; 1624 caddr_t bpos, dpos, cp2; 1625 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1626 struct mbuf *mreq, *mrep, *md, *mb; 1627 const int v3 = NFS_ISV3(dvp); 1628 1629 /* 1630 * Oops, not for me.. 1631 */ 1632 if (vap->va_type == VSOCK) 1633 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1634 1635 #ifdef VA_EXCLUSIVE 1636 if (vap->va_vaflags & VA_EXCLUSIVE) 1637 fmode |= O_EXCL; 1638 #endif 1639 again: 1640 error = 0; 1641 nfsstats.rpccnt[NFSPROC_CREATE]++; 1642 dnp = VTONFS(dvp); 1643 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1644 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1645 nfsm_fhtom(dnp, v3); 1646 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1647 if (v3) { 1648 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1649 if (fmode & O_EXCL) { 1650 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1651 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1652 #ifdef INET 1653 if (TAILQ_FIRST(&in_ifaddr)) 1654 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr; 1655 else 1656 *tl++ = create_verf; 1657 #else 1658 *tl++ = create_verf; 1659 #endif 1660 *tl = ++create_verf; 1661 } else { 1662 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1663 nfsm_v3attrbuild(vap, FALSE); 1664 } 1665 } else { 1666 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1667 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1668 sp->sa_uid = nfs_xdrneg1; 1669 sp->sa_gid = nfs_xdrneg1; 1670 sp->sa_size = 0; 1671 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1672 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1673 } 1674 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred); 1675 if (!error) { 1676 nfsm_mtofh(dvp, newvp, v3, gotvp); 1677 if (!gotvp) { 1678 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1679 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1680 if (!error) 1681 newvp = NFSTOV(np); 1682 } 1683 } 1684 if (v3) 1685 nfsm_wcc_data(dvp, wccflag, 0); 1686 nfsm_reqdone; 1687 if (error) { 1688 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1689 fmode &= ~O_EXCL; 1690 goto again; 1691 } 1692 if (newvp) 1693 vput(newvp); 1694 } else if (v3 && (fmode & O_EXCL)) 1695 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc); 1696 if (!error) { 1697 if (cnp->cn_flags & MAKEENTRY) 1698 nfs_cache_enter(dvp, newvp, cnp); 1699 *ap->a_vpp = newvp; 1700 } 1701 PNBUF_PUT(cnp->cn_pnbuf); 1702 VTONFS(dvp)->n_flag |= NMODIFIED; 1703 if (!wccflag) 1704 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1705 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1706 vput(dvp); 1707 return (error); 1708 } 1709 1710 /* 1711 * nfs file remove call 1712 * To try and make nfs semantics closer to ufs semantics, a file that has 1713 * other processes using the vnode is renamed instead of removed and then 1714 * removed later on the last close. 1715 * - If v_usecount > 1 1716 * If a rename is not already in the works 1717 * call nfs_sillyrename() to set it up 1718 * else 1719 * do the remove rpc 1720 */ 1721 int 1722 nfs_remove(v) 1723 void *v; 1724 { 1725 struct vop_remove_args /* { 1726 struct vnodeop_desc *a_desc; 1727 struct vnode * a_dvp; 1728 struct vnode * a_vp; 1729 struct componentname * a_cnp; 1730 } */ *ap = v; 1731 struct vnode *vp = ap->a_vp; 1732 struct vnode *dvp = ap->a_dvp; 1733 struct componentname *cnp = ap->a_cnp; 1734 struct nfsnode *np = VTONFS(vp); 1735 int error = 0; 1736 struct vattr vattr; 1737 1738 #ifndef DIAGNOSTIC 1739 if ((cnp->cn_flags & HASBUF) == 0) 1740 panic("nfs_remove: no name"); 1741 if (vp->v_usecount < 1) 1742 panic("nfs_remove: bad v_usecount"); 1743 #endif 1744 if (vp->v_type == VDIR) 1745 error = EPERM; 1746 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1747 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 && 1748 vattr.va_nlink > 1)) { 1749 /* 1750 * Purge the name cache so that the chance of a lookup for 1751 * the name succeeding while the remove is in progress is 1752 * minimized. Without node locking it can still happen, such 1753 * that an I/O op returns ESTALE, but since you get this if 1754 * another host removes the file.. 1755 */ 1756 cache_purge(vp); 1757 /* 1758 * throw away biocache buffers, mainly to avoid 1759 * unnecessary delayed writes later. 1760 */ 1761 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1); 1762 /* Do the rpc */ 1763 if (error != EINTR) 1764 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1765 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc); 1766 /* 1767 * Kludge City: If the first reply to the remove rpc is lost.. 1768 * the reply to the retransmitted request will be ENOENT 1769 * since the file was in fact removed 1770 * Therefore, we cheat and return success. 1771 */ 1772 if (error == ENOENT) 1773 error = 0; 1774 } else if (!np->n_sillyrename) 1775 error = nfs_sillyrename(dvp, vp, cnp); 1776 PNBUF_PUT(cnp->cn_pnbuf); 1777 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1778 vattr.va_nlink == 1) { 1779 np->n_flag |= NREMOVED; 1780 } 1781 NFS_INVALIDATE_ATTRCACHE(np); 1782 VN_KNOTE(vp, NOTE_DELETE); 1783 VN_KNOTE(dvp, NOTE_WRITE); 1784 if (dvp == vp) 1785 vrele(vp); 1786 else 1787 vput(vp); 1788 vput(dvp); 1789 return (error); 1790 } 1791 1792 /* 1793 * nfs file remove rpc called from nfs_inactive 1794 */ 1795 int 1796 nfs_removeit(sp) 1797 struct sillyrename *sp; 1798 { 1799 1800 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1801 (struct proc *)0)); 1802 } 1803 1804 /* 1805 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1806 */ 1807 int 1808 nfs_removerpc(dvp, name, namelen, cred, proc) 1809 struct vnode *dvp; 1810 const char *name; 1811 int namelen; 1812 struct ucred *cred; 1813 struct proc *proc; 1814 { 1815 u_int32_t *tl; 1816 caddr_t cp; 1817 int32_t t1, t2; 1818 caddr_t bpos, dpos, cp2; 1819 int error = 0, wccflag = NFSV3_WCCRATTR; 1820 struct mbuf *mreq, *mrep, *md, *mb; 1821 const int v3 = NFS_ISV3(dvp); 1822 struct nfsnode *dnp = VTONFS(dvp); 1823 1824 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1825 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1826 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1827 nfsm_fhtom(dnp, v3); 1828 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1829 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred); 1830 if (v3) 1831 nfsm_wcc_data(dvp, wccflag, 0); 1832 nfsm_reqdone; 1833 VTONFS(dvp)->n_flag |= NMODIFIED; 1834 if (!wccflag) 1835 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1836 return (error); 1837 } 1838 1839 /* 1840 * nfs file rename call 1841 */ 1842 int 1843 nfs_rename(v) 1844 void *v; 1845 { 1846 struct vop_rename_args /* { 1847 struct vnode *a_fdvp; 1848 struct vnode *a_fvp; 1849 struct componentname *a_fcnp; 1850 struct vnode *a_tdvp; 1851 struct vnode *a_tvp; 1852 struct componentname *a_tcnp; 1853 } */ *ap = v; 1854 struct vnode *fvp = ap->a_fvp; 1855 struct vnode *tvp = ap->a_tvp; 1856 struct vnode *fdvp = ap->a_fdvp; 1857 struct vnode *tdvp = ap->a_tdvp; 1858 struct componentname *tcnp = ap->a_tcnp; 1859 struct componentname *fcnp = ap->a_fcnp; 1860 int error; 1861 1862 #ifndef DIAGNOSTIC 1863 if ((tcnp->cn_flags & HASBUF) == 0 || 1864 (fcnp->cn_flags & HASBUF) == 0) 1865 panic("nfs_rename: no name"); 1866 #endif 1867 /* Check for cross-device rename */ 1868 if ((fvp->v_mount != tdvp->v_mount) || 1869 (tvp && (fvp->v_mount != tvp->v_mount))) { 1870 error = EXDEV; 1871 goto out; 1872 } 1873 1874 /* 1875 * If the tvp exists and is in use, sillyrename it before doing the 1876 * rename of the new file over it. 1877 */ 1878 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1879 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1880 VN_KNOTE(tvp, NOTE_DELETE); 1881 vput(tvp); 1882 tvp = NULL; 1883 } 1884 1885 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1886 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1887 tcnp->cn_proc); 1888 1889 VN_KNOTE(fdvp, NOTE_WRITE); 1890 VN_KNOTE(tdvp, NOTE_WRITE); 1891 if (fvp->v_type == VDIR) { 1892 if (tvp != NULL && tvp->v_type == VDIR) 1893 cache_purge(tdvp); 1894 cache_purge(fdvp); 1895 } 1896 out: 1897 if (tdvp == tvp) 1898 vrele(tdvp); 1899 else 1900 vput(tdvp); 1901 if (tvp) 1902 vput(tvp); 1903 vrele(fdvp); 1904 vrele(fvp); 1905 /* 1906 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1907 */ 1908 if (error == ENOENT) 1909 error = 0; 1910 return (error); 1911 } 1912 1913 /* 1914 * nfs file rename rpc called from nfs_remove() above 1915 */ 1916 int 1917 nfs_renameit(sdvp, scnp, sp) 1918 struct vnode *sdvp; 1919 struct componentname *scnp; 1920 struct sillyrename *sp; 1921 { 1922 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1923 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc)); 1924 } 1925 1926 /* 1927 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1928 */ 1929 int 1930 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc) 1931 struct vnode *fdvp; 1932 const char *fnameptr; 1933 int fnamelen; 1934 struct vnode *tdvp; 1935 const char *tnameptr; 1936 int tnamelen; 1937 struct ucred *cred; 1938 struct proc *proc; 1939 { 1940 u_int32_t *tl; 1941 caddr_t cp; 1942 int32_t t1, t2; 1943 caddr_t bpos, dpos, cp2; 1944 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1945 struct mbuf *mreq, *mrep, *md, *mb; 1946 const int v3 = NFS_ISV3(fdvp); 1947 struct nfsnode *fdnp = VTONFS(fdvp); 1948 1949 nfsstats.rpccnt[NFSPROC_RENAME]++; 1950 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1951 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1952 nfsm_rndup(tnamelen)); 1953 nfsm_fhtom(fdnp, v3); 1954 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1955 nfsm_fhtom(VTONFS(tdvp), v3); 1956 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1957 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred); 1958 if (v3) { 1959 nfsm_wcc_data(fdvp, fwccflag, 0); 1960 nfsm_wcc_data(tdvp, twccflag, 0); 1961 } 1962 nfsm_reqdone; 1963 VTONFS(fdvp)->n_flag |= NMODIFIED; 1964 VTONFS(tdvp)->n_flag |= NMODIFIED; 1965 if (!fwccflag) 1966 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1967 if (!twccflag) 1968 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1969 return (error); 1970 } 1971 1972 /* 1973 * nfs hard link create call 1974 */ 1975 int 1976 nfs_link(v) 1977 void *v; 1978 { 1979 struct vop_link_args /* { 1980 struct vnode *a_dvp; 1981 struct vnode *a_vp; 1982 struct componentname *a_cnp; 1983 } */ *ap = v; 1984 struct vnode *vp = ap->a_vp; 1985 struct vnode *dvp = ap->a_dvp; 1986 struct componentname *cnp = ap->a_cnp; 1987 u_int32_t *tl; 1988 caddr_t cp; 1989 int32_t t1, t2; 1990 caddr_t bpos, dpos, cp2; 1991 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1992 struct mbuf *mreq, *mrep, *md, *mb; 1993 /* XXX Should be const and initialised? */ 1994 int v3; 1995 struct nfsnode *np; 1996 1997 if (dvp->v_mount != vp->v_mount) { 1998 VOP_ABORTOP(dvp, cnp); 1999 vput(dvp); 2000 return (EXDEV); 2001 } 2002 if (dvp != vp) { 2003 error = vn_lock(vp, LK_EXCLUSIVE); 2004 if (error != 0) { 2005 VOP_ABORTOP(dvp, cnp); 2006 vput(dvp); 2007 return error; 2008 } 2009 } 2010 2011 /* 2012 * Push all writes to the server, so that the attribute cache 2013 * doesn't get "out of sync" with the server. 2014 * XXX There should be a better way! 2015 */ 2016 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc); 2017 2018 v3 = NFS_ISV3(vp); 2019 nfsstats.rpccnt[NFSPROC_LINK]++; 2020 np = VTONFS(vp); 2021 nfsm_reqhead(np, NFSPROC_LINK, 2022 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2023 nfsm_fhtom(np, v3); 2024 nfsm_fhtom(VTONFS(dvp), v3); 2025 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2026 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred); 2027 if (v3) { 2028 nfsm_postop_attr(vp, attrflag, 0); 2029 nfsm_wcc_data(dvp, wccflag, 0); 2030 } 2031 nfsm_reqdone; 2032 PNBUF_PUT(cnp->cn_pnbuf); 2033 VTONFS(dvp)->n_flag |= NMODIFIED; 2034 if (!attrflag) 2035 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2036 if (!wccflag) 2037 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2038 if (dvp != vp) 2039 VOP_UNLOCK(vp, 0); 2040 VN_KNOTE(vp, NOTE_LINK); 2041 VN_KNOTE(dvp, NOTE_WRITE); 2042 vput(dvp); 2043 /* 2044 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2045 */ 2046 if (error == EEXIST) 2047 error = 0; 2048 return (error); 2049 } 2050 2051 /* 2052 * nfs symbolic link create call 2053 */ 2054 int 2055 nfs_symlink(v) 2056 void *v; 2057 { 2058 struct vop_symlink_args /* { 2059 struct vnode *a_dvp; 2060 struct vnode **a_vpp; 2061 struct componentname *a_cnp; 2062 struct vattr *a_vap; 2063 char *a_target; 2064 } */ *ap = v; 2065 struct vnode *dvp = ap->a_dvp; 2066 struct vattr *vap = ap->a_vap; 2067 struct componentname *cnp = ap->a_cnp; 2068 struct nfsv2_sattr *sp; 2069 u_int32_t *tl; 2070 caddr_t cp; 2071 int32_t t1, t2; 2072 caddr_t bpos, dpos, cp2; 2073 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2074 struct mbuf *mreq, *mrep, *md, *mb; 2075 struct vnode *newvp = (struct vnode *)0; 2076 const int v3 = NFS_ISV3(dvp); 2077 struct nfsnode *dnp = VTONFS(dvp); 2078 2079 *ap->a_vpp = NULL; 2080 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2081 slen = strlen(ap->a_target); 2082 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2083 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2084 nfsm_fhtom(dnp, v3); 2085 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2086 if (v3) 2087 nfsm_v3attrbuild(vap, FALSE); 2088 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2089 if (!v3) { 2090 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2091 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2092 sp->sa_uid = nfs_xdrneg1; 2093 sp->sa_gid = nfs_xdrneg1; 2094 sp->sa_size = nfs_xdrneg1; 2095 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2096 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2097 } 2098 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred); 2099 if (v3) { 2100 if (!error) 2101 nfsm_mtofh(dvp, newvp, v3, gotvp); 2102 nfsm_wcc_data(dvp, wccflag, 0); 2103 } 2104 nfsm_reqdone; 2105 /* 2106 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2107 */ 2108 if (error == EEXIST) 2109 error = 0; 2110 if (error == 0 && newvp == NULL) { 2111 struct nfsnode *np = NULL; 2112 2113 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2114 cnp->cn_cred, cnp->cn_proc, &np); 2115 if (error == 0) 2116 newvp = NFSTOV(np); 2117 } 2118 if (error) { 2119 if (newvp != NULL) 2120 vput(newvp); 2121 } else { 2122 *ap->a_vpp = newvp; 2123 } 2124 PNBUF_PUT(cnp->cn_pnbuf); 2125 VTONFS(dvp)->n_flag |= NMODIFIED; 2126 if (!wccflag) 2127 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2128 VN_KNOTE(dvp, NOTE_WRITE); 2129 vput(dvp); 2130 return (error); 2131 } 2132 2133 /* 2134 * nfs make dir call 2135 */ 2136 int 2137 nfs_mkdir(v) 2138 void *v; 2139 { 2140 struct vop_mkdir_args /* { 2141 struct vnode *a_dvp; 2142 struct vnode **a_vpp; 2143 struct componentname *a_cnp; 2144 struct vattr *a_vap; 2145 } */ *ap = v; 2146 struct vnode *dvp = ap->a_dvp; 2147 struct vattr *vap = ap->a_vap; 2148 struct componentname *cnp = ap->a_cnp; 2149 struct nfsv2_sattr *sp; 2150 u_int32_t *tl; 2151 caddr_t cp; 2152 int32_t t1, t2; 2153 int len; 2154 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2155 struct vnode *newvp = (struct vnode *)0; 2156 caddr_t bpos, dpos, cp2; 2157 int error = 0, wccflag = NFSV3_WCCRATTR; 2158 int gotvp = 0; 2159 struct mbuf *mreq, *mrep, *md, *mb; 2160 const int v3 = NFS_ISV3(dvp); 2161 2162 len = cnp->cn_namelen; 2163 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2164 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2165 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2166 nfsm_fhtom(dnp, v3); 2167 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2168 if (v3) { 2169 nfsm_v3attrbuild(vap, FALSE); 2170 } else { 2171 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2172 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2173 sp->sa_uid = nfs_xdrneg1; 2174 sp->sa_gid = nfs_xdrneg1; 2175 sp->sa_size = nfs_xdrneg1; 2176 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2177 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2178 } 2179 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred); 2180 if (!error) 2181 nfsm_mtofh(dvp, newvp, v3, gotvp); 2182 if (v3) 2183 nfsm_wcc_data(dvp, wccflag, 0); 2184 nfsm_reqdone; 2185 VTONFS(dvp)->n_flag |= NMODIFIED; 2186 if (!wccflag) 2187 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2188 /* 2189 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2190 * if we can succeed in looking up the directory. 2191 */ 2192 if (error == EEXIST || (!error && !gotvp)) { 2193 if (newvp) { 2194 vput(newvp); 2195 newvp = (struct vnode *)0; 2196 } 2197 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2198 cnp->cn_proc, &np); 2199 if (!error) { 2200 newvp = NFSTOV(np); 2201 if (newvp->v_type != VDIR) 2202 error = EEXIST; 2203 } 2204 } 2205 if (error) { 2206 if (newvp) 2207 vput(newvp); 2208 } else { 2209 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2210 if (cnp->cn_flags & MAKEENTRY) 2211 nfs_cache_enter(dvp, newvp, cnp); 2212 *ap->a_vpp = newvp; 2213 } 2214 PNBUF_PUT(cnp->cn_pnbuf); 2215 vput(dvp); 2216 return (error); 2217 } 2218 2219 /* 2220 * nfs remove directory call 2221 */ 2222 int 2223 nfs_rmdir(v) 2224 void *v; 2225 { 2226 struct vop_rmdir_args /* { 2227 struct vnode *a_dvp; 2228 struct vnode *a_vp; 2229 struct componentname *a_cnp; 2230 } */ *ap = v; 2231 struct vnode *vp = ap->a_vp; 2232 struct vnode *dvp = ap->a_dvp; 2233 struct componentname *cnp = ap->a_cnp; 2234 u_int32_t *tl; 2235 caddr_t cp; 2236 int32_t t1, t2; 2237 caddr_t bpos, dpos, cp2; 2238 int error = 0, wccflag = NFSV3_WCCRATTR; 2239 struct mbuf *mreq, *mrep, *md, *mb; 2240 const int v3 = NFS_ISV3(dvp); 2241 struct nfsnode *dnp; 2242 2243 if (dvp == vp) { 2244 vrele(dvp); 2245 vput(dvp); 2246 PNBUF_PUT(cnp->cn_pnbuf); 2247 return (EINVAL); 2248 } 2249 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2250 dnp = VTONFS(dvp); 2251 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2252 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2253 nfsm_fhtom(dnp, v3); 2254 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2255 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred); 2256 if (v3) 2257 nfsm_wcc_data(dvp, wccflag, 0); 2258 nfsm_reqdone; 2259 PNBUF_PUT(cnp->cn_pnbuf); 2260 VTONFS(dvp)->n_flag |= NMODIFIED; 2261 if (!wccflag) 2262 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2263 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2264 VN_KNOTE(vp, NOTE_DELETE); 2265 cache_purge(vp); 2266 vput(vp); 2267 vput(dvp); 2268 /* 2269 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2270 */ 2271 if (error == ENOENT) 2272 error = 0; 2273 return (error); 2274 } 2275 2276 /* 2277 * nfs readdir call 2278 */ 2279 int 2280 nfs_readdir(v) 2281 void *v; 2282 { 2283 struct vop_readdir_args /* { 2284 struct vnode *a_vp; 2285 struct uio *a_uio; 2286 struct ucred *a_cred; 2287 int *a_eofflag; 2288 off_t **a_cookies; 2289 int *a_ncookies; 2290 } */ *ap = v; 2291 struct vnode *vp = ap->a_vp; 2292 struct uio *uio = ap->a_uio; 2293 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2294 char *base = uio->uio_iov->iov_base; 2295 int tresid, error; 2296 size_t count, lost; 2297 struct dirent *dp; 2298 off_t *cookies = NULL; 2299 int ncookies = 0, nc; 2300 2301 if (vp->v_type != VDIR) 2302 return (EPERM); 2303 2304 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2305 count = uio->uio_resid - lost; 2306 if (count <= 0) 2307 return (EINVAL); 2308 2309 /* 2310 * Call nfs_bioread() to do the real work. 2311 */ 2312 tresid = uio->uio_resid = count; 2313 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2314 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2315 2316 if (!error && ap->a_cookies) { 2317 ncookies = count / 16; 2318 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2319 *ap->a_cookies = cookies; 2320 } 2321 2322 if (!error && uio->uio_resid == tresid) { 2323 uio->uio_resid += lost; 2324 nfsstats.direofcache_misses++; 2325 if (ap->a_cookies) 2326 *ap->a_ncookies = 0; 2327 *ap->a_eofflag = 1; 2328 return (0); 2329 } 2330 2331 if (!error && ap->a_cookies) { 2332 /* 2333 * Only the NFS server and emulations use cookies, and they 2334 * load the directory block into system space, so we can 2335 * just look at it directly. 2336 */ 2337 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2338 panic("nfs_readdir: lost in space"); 2339 for (nc = 0; ncookies-- && 2340 base < (char *)uio->uio_iov->iov_base; nc++){ 2341 dp = (struct dirent *) base; 2342 if (dp->d_reclen == 0) 2343 break; 2344 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2345 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2346 else 2347 *(cookies++) = NFS_GETCOOKIE(dp); 2348 base += dp->d_reclen; 2349 } 2350 uio->uio_resid += 2351 ((caddr_t)uio->uio_iov->iov_base - base); 2352 uio->uio_iov->iov_len += 2353 ((caddr_t)uio->uio_iov->iov_base - base); 2354 uio->uio_iov->iov_base = base; 2355 *ap->a_ncookies = nc; 2356 } 2357 2358 uio->uio_resid += lost; 2359 *ap->a_eofflag = 0; 2360 return (error); 2361 } 2362 2363 /* 2364 * Readdir rpc call. 2365 * Called from below the buffer cache by nfs_doio(). 2366 */ 2367 int 2368 nfs_readdirrpc(vp, uiop, cred) 2369 struct vnode *vp; 2370 struct uio *uiop; 2371 struct ucred *cred; 2372 { 2373 int len, left; 2374 struct dirent *dp = NULL; 2375 u_int32_t *tl; 2376 caddr_t cp; 2377 int32_t t1, t2; 2378 caddr_t bpos, dpos, cp2; 2379 struct mbuf *mreq, *mrep, *md, *mb; 2380 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2381 struct nfsnode *dnp = VTONFS(vp); 2382 u_quad_t fileno; 2383 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2384 int attrflag, nrpcs = 0, reclen; 2385 const int v3 = NFS_ISV3(vp); 2386 2387 #ifdef DIAGNOSTIC 2388 /* 2389 * Should be called from buffer cache, so only amount of 2390 * NFS_DIRBLKSIZ will be requested. 2391 */ 2392 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2393 panic("nfs readdirrpc bad uio"); 2394 #endif 2395 2396 /* 2397 * Loop around doing readdir rpc's of size nm_readdirsize 2398 * truncated to a multiple of NFS_DIRFRAGSIZ. 2399 * The stopping criteria is EOF or buffer full. 2400 */ 2401 while (more_dirs && bigenough) { 2402 /* 2403 * Heuristic: don't bother to do another RPC to further 2404 * fill up this block if there is not much room left. (< 50% 2405 * of the readdir RPC size). This wastes some buffer space 2406 * but can save up to 50% in RPC calls. 2407 */ 2408 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2409 bigenough = 0; 2410 break; 2411 } 2412 nfsstats.rpccnt[NFSPROC_READDIR]++; 2413 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2414 NFSX_READDIR(v3)); 2415 nfsm_fhtom(dnp, v3); 2416 if (v3) { 2417 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2418 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2419 txdr_swapcookie3(uiop->uio_offset, tl); 2420 } else { 2421 txdr_cookie3(uiop->uio_offset, tl); 2422 } 2423 tl += 2; 2424 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2425 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2426 } else { 2427 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2428 *tl++ = txdr_unsigned(uiop->uio_offset); 2429 } 2430 *tl = txdr_unsigned(nmp->nm_readdirsize); 2431 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred); 2432 nrpcs++; 2433 if (v3) { 2434 nfsm_postop_attr(vp, attrflag, 0); 2435 if (!error) { 2436 nfsm_dissect(tl, u_int32_t *, 2437 2 * NFSX_UNSIGNED); 2438 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2439 dnp->n_cookieverf.nfsuquad[1] = *tl; 2440 } else { 2441 m_freem(mrep); 2442 goto nfsmout; 2443 } 2444 } 2445 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2446 more_dirs = fxdr_unsigned(int, *tl); 2447 2448 /* loop thru the dir entries, doctoring them to 4bsd form */ 2449 while (more_dirs && bigenough) { 2450 if (v3) { 2451 nfsm_dissect(tl, u_int32_t *, 2452 3 * NFSX_UNSIGNED); 2453 fileno = fxdr_hyper(tl); 2454 len = fxdr_unsigned(int, *(tl + 2)); 2455 } else { 2456 nfsm_dissect(tl, u_int32_t *, 2457 2 * NFSX_UNSIGNED); 2458 fileno = fxdr_unsigned(u_quad_t, *tl++); 2459 len = fxdr_unsigned(int, *tl); 2460 } 2461 if (len <= 0 || len > NFS_MAXNAMLEN) { 2462 error = EBADRPC; 2463 m_freem(mrep); 2464 goto nfsmout; 2465 } 2466 tlen = nfsm_rndup(len); 2467 if (tlen == len) 2468 tlen += 4; /* To ensure null termination */ 2469 tlen += sizeof (off_t) + sizeof (int); 2470 reclen = ALIGN(tlen + DIRHDSIZ); 2471 tlen = reclen - DIRHDSIZ; 2472 left = NFS_DIRFRAGSIZ - blksiz; 2473 if (reclen > left) { 2474 dp->d_reclen += left; 2475 uiop->uio_iov->iov_base = 2476 (caddr_t)uiop->uio_iov->iov_base + left; 2477 uiop->uio_iov->iov_len -= left; 2478 uiop->uio_resid -= left; 2479 blksiz = 0; 2480 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2481 } 2482 if (reclen > uiop->uio_resid) 2483 bigenough = 0; 2484 if (bigenough) { 2485 dp = (struct dirent *)uiop->uio_iov->iov_base; 2486 dp->d_fileno = (int)fileno; 2487 dp->d_namlen = len; 2488 dp->d_reclen = reclen; 2489 dp->d_type = DT_UNKNOWN; 2490 blksiz += dp->d_reclen; 2491 if (blksiz == NFS_DIRFRAGSIZ) 2492 blksiz = 0; 2493 uiop->uio_resid -= DIRHDSIZ; 2494 uiop->uio_iov->iov_base = 2495 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ; 2496 uiop->uio_iov->iov_len -= DIRHDSIZ; 2497 nfsm_mtouio(uiop, len); 2498 cp = uiop->uio_iov->iov_base; 2499 tlen -= len; 2500 *cp = '\0'; /* null terminate */ 2501 uiop->uio_iov->iov_base = 2502 (caddr_t)uiop->uio_iov->iov_base + tlen; 2503 uiop->uio_iov->iov_len -= tlen; 2504 uiop->uio_resid -= tlen; 2505 } else 2506 nfsm_adv(nfsm_rndup(len)); 2507 if (v3) { 2508 nfsm_dissect(tl, u_int32_t *, 2509 3 * NFSX_UNSIGNED); 2510 } else { 2511 nfsm_dissect(tl, u_int32_t *, 2512 2 * NFSX_UNSIGNED); 2513 } 2514 if (bigenough) { 2515 if (v3) { 2516 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2517 uiop->uio_offset = 2518 fxdr_swapcookie3(tl); 2519 else 2520 uiop->uio_offset = 2521 fxdr_cookie3(tl); 2522 } 2523 else { 2524 uiop->uio_offset = 2525 fxdr_unsigned(off_t, *tl); 2526 } 2527 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2528 } 2529 if (v3) 2530 tl += 2; 2531 else 2532 tl++; 2533 more_dirs = fxdr_unsigned(int, *tl); 2534 } 2535 /* 2536 * If at end of rpc data, get the eof boolean 2537 */ 2538 if (!more_dirs) { 2539 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2540 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2541 } 2542 m_freem(mrep); 2543 } 2544 /* 2545 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2546 * by increasing d_reclen for the last record. 2547 */ 2548 if (blksiz > 0) { 2549 left = NFS_DIRFRAGSIZ - blksiz; 2550 dp->d_reclen += left; 2551 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2552 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2553 left; 2554 uiop->uio_iov->iov_len -= left; 2555 uiop->uio_resid -= left; 2556 } 2557 2558 /* 2559 * We are now either at the end of the directory or have filled the 2560 * block. 2561 */ 2562 if (bigenough) 2563 dnp->n_direofoffset = uiop->uio_offset; 2564 nfsmout: 2565 return (error); 2566 } 2567 2568 /* 2569 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2570 */ 2571 int 2572 nfs_readdirplusrpc(vp, uiop, cred) 2573 struct vnode *vp; 2574 struct uio *uiop; 2575 struct ucred *cred; 2576 { 2577 int len, left; 2578 struct dirent *dp = NULL; 2579 u_int32_t *tl; 2580 caddr_t cp; 2581 int32_t t1, t2; 2582 struct vnode *newvp; 2583 caddr_t bpos, dpos, cp2; 2584 struct mbuf *mreq, *mrep, *md, *mb; 2585 struct nameidata nami, *ndp = &nami; 2586 struct componentname *cnp = &ndp->ni_cnd; 2587 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2588 struct nfsnode *dnp = VTONFS(vp), *np; 2589 nfsfh_t *fhp; 2590 u_quad_t fileno; 2591 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2592 int attrflag, fhsize, nrpcs = 0, reclen; 2593 struct nfs_fattr fattr, *fp; 2594 2595 #ifdef DIAGNOSTIC 2596 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2597 panic("nfs readdirplusrpc bad uio"); 2598 #endif 2599 ndp->ni_dvp = vp; 2600 newvp = NULLVP; 2601 2602 /* 2603 * Loop around doing readdir rpc's of size nm_readdirsize 2604 * truncated to a multiple of NFS_DIRFRAGSIZ. 2605 * The stopping criteria is EOF or buffer full. 2606 */ 2607 while (more_dirs && bigenough) { 2608 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2609 bigenough = 0; 2610 break; 2611 } 2612 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2613 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2614 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2615 nfsm_fhtom(dnp, 1); 2616 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2617 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2618 txdr_swapcookie3(uiop->uio_offset, tl); 2619 } else { 2620 txdr_cookie3(uiop->uio_offset, tl); 2621 } 2622 tl += 2; 2623 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2624 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2625 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2626 *tl = txdr_unsigned(nmp->nm_rsize); 2627 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred); 2628 nfsm_postop_attr(vp, attrflag, 0); 2629 if (error) { 2630 m_freem(mrep); 2631 goto nfsmout; 2632 } 2633 nrpcs++; 2634 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2635 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2636 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2637 more_dirs = fxdr_unsigned(int, *tl); 2638 2639 /* loop thru the dir entries, doctoring them to 4bsd form */ 2640 while (more_dirs && bigenough) { 2641 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2642 fileno = fxdr_hyper(tl); 2643 len = fxdr_unsigned(int, *(tl + 2)); 2644 if (len <= 0 || len > NFS_MAXNAMLEN) { 2645 error = EBADRPC; 2646 m_freem(mrep); 2647 goto nfsmout; 2648 } 2649 tlen = nfsm_rndup(len); 2650 if (tlen == len) 2651 tlen += 4; /* To ensure null termination*/ 2652 tlen += sizeof (off_t) + sizeof (int); 2653 reclen = ALIGN(tlen + DIRHDSIZ); 2654 tlen = reclen - DIRHDSIZ; 2655 left = NFS_DIRFRAGSIZ - blksiz; 2656 if (reclen > left) { 2657 /* 2658 * DIRFRAGSIZ is aligned, no need to align 2659 * again here. 2660 */ 2661 dp->d_reclen += left; 2662 uiop->uio_iov->iov_base = 2663 (caddr_t)uiop->uio_iov->iov_base + left; 2664 uiop->uio_iov->iov_len -= left; 2665 uiop->uio_resid -= left; 2666 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2667 blksiz = 0; 2668 } 2669 if (reclen > uiop->uio_resid) 2670 bigenough = 0; 2671 if (bigenough) { 2672 dp = (struct dirent *)uiop->uio_iov->iov_base; 2673 dp->d_fileno = (int)fileno; 2674 dp->d_namlen = len; 2675 dp->d_reclen = reclen; 2676 dp->d_type = DT_UNKNOWN; 2677 blksiz += dp->d_reclen; 2678 if (blksiz == NFS_DIRFRAGSIZ) 2679 blksiz = 0; 2680 uiop->uio_resid -= DIRHDSIZ; 2681 uiop->uio_iov->iov_base = 2682 (caddr_t)uiop->uio_iov->iov_base + 2683 DIRHDSIZ; 2684 uiop->uio_iov->iov_len -= DIRHDSIZ; 2685 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2686 cnp->cn_namelen = len; 2687 nfsm_mtouio(uiop, len); 2688 cp = uiop->uio_iov->iov_base; 2689 tlen -= len; 2690 *cp = '\0'; 2691 uiop->uio_iov->iov_base = 2692 (caddr_t)uiop->uio_iov->iov_base + tlen; 2693 uiop->uio_iov->iov_len -= tlen; 2694 uiop->uio_resid -= tlen; 2695 } else 2696 nfsm_adv(nfsm_rndup(len)); 2697 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2698 if (bigenough) { 2699 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2700 uiop->uio_offset = 2701 fxdr_swapcookie3(tl); 2702 else 2703 uiop->uio_offset = 2704 fxdr_cookie3(tl); 2705 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2706 } 2707 tl += 2; 2708 2709 /* 2710 * Since the attributes are before the file handle 2711 * (sigh), we must skip over the attributes and then 2712 * come back and get them. 2713 */ 2714 attrflag = fxdr_unsigned(int, *tl); 2715 if (attrflag) { 2716 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2717 memcpy(&fattr, fp, NFSX_V3FATTR); 2718 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2719 doit = fxdr_unsigned(int, *tl); 2720 if (doit) { 2721 nfsm_getfh(fhp, fhsize, 1); 2722 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2723 VREF(vp); 2724 newvp = vp; 2725 np = dnp; 2726 } else { 2727 error = nfs_nget1(vp->v_mount, fhp, 2728 fhsize, &np, LK_NOWAIT); 2729 if (!error) 2730 newvp = NFSTOV(np); 2731 } 2732 if (!error) { 2733 const char *cp; 2734 2735 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2736 dp->d_type = 2737 IFTODT(VTTOIF(np->n_vattr->va_type)); 2738 ndp->ni_vp = newvp; 2739 cp = cnp->cn_nameptr + cnp->cn_namelen; 2740 cnp->cn_hash = 2741 namei_hash(cnp->cn_nameptr, &cp); 2742 if (cnp->cn_namelen <= NCHNAMLEN) { 2743 cache_purge1(ndp->ni_dvp, cnp, 0); 2744 nfs_cache_enter(ndp->ni_dvp, ndp->ni_vp, 2745 cnp); 2746 } 2747 } 2748 } 2749 } else { 2750 /* Just skip over the file handle */ 2751 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2752 i = fxdr_unsigned(int, *tl); 2753 nfsm_adv(nfsm_rndup(i)); 2754 } 2755 if (newvp != NULLVP) { 2756 if (newvp == vp) 2757 vrele(newvp); 2758 else 2759 vput(newvp); 2760 newvp = NULLVP; 2761 } 2762 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2763 more_dirs = fxdr_unsigned(int, *tl); 2764 } 2765 /* 2766 * If at end of rpc data, get the eof boolean 2767 */ 2768 if (!more_dirs) { 2769 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2770 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2771 } 2772 m_freem(mrep); 2773 } 2774 /* 2775 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2776 * by increasing d_reclen for the last record. 2777 */ 2778 if (blksiz > 0) { 2779 left = NFS_DIRFRAGSIZ - blksiz; 2780 dp->d_reclen += left; 2781 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2782 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2783 left; 2784 uiop->uio_iov->iov_len -= left; 2785 uiop->uio_resid -= left; 2786 } 2787 2788 /* 2789 * We are now either at the end of the directory or have filled the 2790 * block. 2791 */ 2792 if (bigenough) 2793 dnp->n_direofoffset = uiop->uio_offset; 2794 nfsmout: 2795 if (newvp != NULLVP) { 2796 if(newvp == vp) 2797 vrele(newvp); 2798 else 2799 vput(newvp); 2800 } 2801 return (error); 2802 } 2803 static char hextoasc[] = "0123456789abcdef"; 2804 2805 /* 2806 * Silly rename. To make the NFS filesystem that is stateless look a little 2807 * more like the "ufs" a remove of an active vnode is translated to a rename 2808 * to a funny looking filename that is removed by nfs_inactive on the 2809 * nfsnode. There is the potential for another process on a different client 2810 * to create the same funny name between the nfs_lookitup() fails and the 2811 * nfs_rename() completes, but... 2812 */ 2813 int 2814 nfs_sillyrename(dvp, vp, cnp) 2815 struct vnode *dvp, *vp; 2816 struct componentname *cnp; 2817 { 2818 struct sillyrename *sp; 2819 struct nfsnode *np; 2820 int error; 2821 short pid; 2822 2823 cache_purge(dvp); 2824 np = VTONFS(vp); 2825 #ifndef DIAGNOSTIC 2826 if (vp->v_type == VDIR) 2827 panic("nfs: sillyrename dir"); 2828 #endif 2829 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2830 M_NFSREQ, M_WAITOK); 2831 sp->s_cred = crdup(cnp->cn_cred); 2832 sp->s_dvp = dvp; 2833 VREF(dvp); 2834 2835 /* Fudge together a funny name */ 2836 pid = cnp->cn_proc->p_pid; 2837 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2838 sp->s_namlen = 12; 2839 sp->s_name[8] = hextoasc[pid & 0xf]; 2840 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf]; 2841 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf]; 2842 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf]; 2843 2844 /* Try lookitups until we get one that isn't there */ 2845 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2846 cnp->cn_proc, (struct nfsnode **)0) == 0) { 2847 sp->s_name[4]++; 2848 if (sp->s_name[4] > 'z') { 2849 error = EINVAL; 2850 goto bad; 2851 } 2852 } 2853 error = nfs_renameit(dvp, cnp, sp); 2854 if (error) 2855 goto bad; 2856 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2857 cnp->cn_proc, &np); 2858 np->n_sillyrename = sp; 2859 return (0); 2860 bad: 2861 vrele(sp->s_dvp); 2862 crfree(sp->s_cred); 2863 free((caddr_t)sp, M_NFSREQ); 2864 return (error); 2865 } 2866 2867 /* 2868 * Look up a file name and optionally either update the file handle or 2869 * allocate an nfsnode, depending on the value of npp. 2870 * npp == NULL --> just do the lookup 2871 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2872 * handled too 2873 * *npp != NULL --> update the file handle in the vnode 2874 */ 2875 int 2876 nfs_lookitup(dvp, name, len, cred, procp, npp) 2877 struct vnode *dvp; 2878 const char *name; 2879 int len; 2880 struct ucred *cred; 2881 struct proc *procp; 2882 struct nfsnode **npp; 2883 { 2884 u_int32_t *tl; 2885 caddr_t cp; 2886 int32_t t1, t2; 2887 struct vnode *newvp = (struct vnode *)0; 2888 struct nfsnode *np, *dnp = VTONFS(dvp); 2889 caddr_t bpos, dpos, cp2; 2890 int error = 0, fhlen, attrflag; 2891 struct mbuf *mreq, *mrep, *md, *mb; 2892 nfsfh_t *nfhp; 2893 const int v3 = NFS_ISV3(dvp); 2894 2895 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2896 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2897 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2898 nfsm_fhtom(dnp, v3); 2899 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2900 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred); 2901 if (npp && !error) { 2902 nfsm_getfh(nfhp, fhlen, v3); 2903 if (*npp) { 2904 np = *npp; 2905 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2906 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2907 np->n_fhp = &np->n_fh; 2908 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2909 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2910 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen); 2911 np->n_fhsize = fhlen; 2912 newvp = NFSTOV(np); 2913 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2914 VREF(dvp); 2915 newvp = dvp; 2916 } else { 2917 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2918 if (error) { 2919 m_freem(mrep); 2920 return (error); 2921 } 2922 newvp = NFSTOV(np); 2923 } 2924 if (v3) { 2925 nfsm_postop_attr(newvp, attrflag, 0); 2926 if (!attrflag && *npp == NULL) { 2927 m_freem(mrep); 2928 vput(newvp); 2929 return (ENOENT); 2930 } 2931 } else 2932 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2933 } 2934 nfsm_reqdone; 2935 if (npp && *npp == NULL) { 2936 if (error) { 2937 if (newvp) 2938 vput(newvp); 2939 } else 2940 *npp = np; 2941 } 2942 return (error); 2943 } 2944 2945 /* 2946 * Nfs Version 3 commit rpc 2947 */ 2948 int 2949 nfs_commit(vp, offset, cnt, procp) 2950 struct vnode *vp; 2951 off_t offset; 2952 uint32_t cnt; 2953 struct proc *procp; 2954 { 2955 caddr_t cp; 2956 u_int32_t *tl; 2957 int32_t t1, t2; 2958 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2959 caddr_t bpos, dpos, cp2; 2960 int error = 0, wccflag = NFSV3_WCCRATTR; 2961 struct mbuf *mreq, *mrep, *md, *mb; 2962 struct nfsnode *np; 2963 2964 KASSERT(NFS_ISV3(vp)); 2965 2966 #ifdef NFS_DEBUG_COMMIT 2967 printf("commit %lu - %lu\n", (unsigned long)offset, 2968 (unsigned long)(offset + cnt)); 2969 #endif 2970 2971 simple_lock(&nmp->nm_slock); 2972 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 2973 simple_unlock(&nmp->nm_slock); 2974 return (0); 2975 } 2976 simple_unlock(&nmp->nm_slock); 2977 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2978 np = VTONFS(vp); 2979 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 2980 nfsm_fhtom(np, 1); 2981 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2982 txdr_hyper(offset, tl); 2983 tl += 2; 2984 *tl = txdr_unsigned(cnt); 2985 nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred); 2986 nfsm_wcc_data(vp, wccflag, 0); 2987 if (!error) { 2988 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2989 simple_lock(&nmp->nm_slock); 2990 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 2991 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 2992 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 2993 error = NFSERR_STALEWRITEVERF; 2994 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 2995 } 2996 simple_unlock(&nmp->nm_slock); 2997 } 2998 nfsm_reqdone; 2999 return (error); 3000 } 3001 3002 /* 3003 * Kludge City.. 3004 * - make nfs_bmap() essentially a no-op that does no translation 3005 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3006 * (Maybe I could use the process's page mapping, but I was concerned that 3007 * Kernel Write might not be enabled and also figured copyout() would do 3008 * a lot more work than memcpy() and also it currently happens in the 3009 * context of the swapper process (2). 3010 */ 3011 int 3012 nfs_bmap(v) 3013 void *v; 3014 { 3015 struct vop_bmap_args /* { 3016 struct vnode *a_vp; 3017 daddr_t a_bn; 3018 struct vnode **a_vpp; 3019 daddr_t *a_bnp; 3020 int *a_runp; 3021 } */ *ap = v; 3022 struct vnode *vp = ap->a_vp; 3023 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3024 3025 if (ap->a_vpp != NULL) 3026 *ap->a_vpp = vp; 3027 if (ap->a_bnp != NULL) 3028 *ap->a_bnp = ap->a_bn << bshift; 3029 if (ap->a_runp != NULL) 3030 *ap->a_runp = 1024 * 1024; /* XXX */ 3031 return (0); 3032 } 3033 3034 /* 3035 * Strategy routine. 3036 * For async requests when nfsiod(s) are running, queue the request by 3037 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3038 * request. 3039 */ 3040 int 3041 nfs_strategy(v) 3042 void *v; 3043 { 3044 struct vop_strategy_args *ap = v; 3045 struct buf *bp = ap->a_bp; 3046 struct proc *p; 3047 int error = 0; 3048 3049 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3050 panic("nfs physio/async"); 3051 if (bp->b_flags & B_ASYNC) 3052 p = NULL; 3053 else 3054 p = curproc; /* XXX */ 3055 3056 /* 3057 * If the op is asynchronous and an i/o daemon is waiting 3058 * queue the request, wake it up and wait for completion 3059 * otherwise just do it ourselves. 3060 */ 3061 3062 if ((bp->b_flags & B_ASYNC) == 0 || 3063 nfs_asyncio(bp)) 3064 error = nfs_doio(bp, p); 3065 return (error); 3066 } 3067 3068 /* 3069 * fsync vnode op. Just call nfs_flush() with commit == 1. 3070 */ 3071 /* ARGSUSED */ 3072 int 3073 nfs_fsync(v) 3074 void *v; 3075 { 3076 struct vop_fsync_args /* { 3077 struct vnodeop_desc *a_desc; 3078 struct vnode * a_vp; 3079 struct ucred * a_cred; 3080 int a_flags; 3081 off_t offlo; 3082 off_t offhi; 3083 struct proc * a_p; 3084 } */ *ap = v; 3085 3086 struct vnode *vp = ap->a_vp; 3087 3088 if (vp->v_type != VREG) 3089 return 0; 3090 3091 return (nfs_flush(vp, ap->a_cred, 3092 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1)); 3093 } 3094 3095 /* 3096 * Flush all the data associated with a vnode. 3097 */ 3098 int 3099 nfs_flush(vp, cred, waitfor, p, commit) 3100 struct vnode *vp; 3101 struct ucred *cred; 3102 int waitfor; 3103 struct proc *p; 3104 int commit; 3105 { 3106 struct nfsnode *np = VTONFS(vp); 3107 int error; 3108 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3109 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3110 3111 simple_lock(&vp->v_interlock); 3112 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3113 if (np->n_flag & NWRITEERR) { 3114 error = np->n_error; 3115 np->n_flag &= ~NWRITEERR; 3116 } 3117 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3118 return (error); 3119 } 3120 3121 /* 3122 * Return POSIX pathconf information applicable to nfs. 3123 * 3124 * N.B. The NFS V2 protocol doesn't support this RPC. 3125 */ 3126 /* ARGSUSED */ 3127 int 3128 nfs_pathconf(v) 3129 void *v; 3130 { 3131 struct vop_pathconf_args /* { 3132 struct vnode *a_vp; 3133 int a_name; 3134 register_t *a_retval; 3135 } */ *ap = v; 3136 struct nfsv3_pathconf *pcp; 3137 struct vnode *vp = ap->a_vp; 3138 struct mbuf *mreq, *mrep, *md, *mb; 3139 int32_t t1, t2; 3140 u_int32_t *tl; 3141 caddr_t bpos, dpos, cp, cp2; 3142 int error = 0, attrflag; 3143 #ifndef NFS_V2_ONLY 3144 struct nfsmount *nmp; 3145 unsigned int l; 3146 u_int64_t maxsize; 3147 #endif 3148 const int v3 = NFS_ISV3(vp); 3149 struct nfsnode *np = VTONFS(vp); 3150 3151 switch (ap->a_name) { 3152 /* Names that can be resolved locally. */ 3153 case _PC_PIPE_BUF: 3154 *ap->a_retval = PIPE_BUF; 3155 break; 3156 case _PC_SYNC_IO: 3157 *ap->a_retval = 1; 3158 break; 3159 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3160 case _PC_LINK_MAX: 3161 case _PC_NAME_MAX: 3162 case _PC_CHOWN_RESTRICTED: 3163 case _PC_NO_TRUNC: 3164 if (!v3) { 3165 error = EINVAL; 3166 break; 3167 } 3168 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3169 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3170 nfsm_fhtom(np, 1); 3171 nfsm_request(np, NFSPROC_PATHCONF, 3172 curproc, curproc->p_ucred); /* XXX */ 3173 nfsm_postop_attr(vp, attrflag, 0); 3174 if (!error) { 3175 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3176 NFSX_V3PATHCONF); 3177 switch (ap->a_name) { 3178 case _PC_LINK_MAX: 3179 *ap->a_retval = 3180 fxdr_unsigned(register_t, pcp->pc_linkmax); 3181 break; 3182 case _PC_NAME_MAX: 3183 *ap->a_retval = 3184 fxdr_unsigned(register_t, pcp->pc_namemax); 3185 break; 3186 case _PC_CHOWN_RESTRICTED: 3187 *ap->a_retval = 3188 (pcp->pc_chownrestricted == nfs_true); 3189 break; 3190 case _PC_NO_TRUNC: 3191 *ap->a_retval = 3192 (pcp->pc_notrunc == nfs_true); 3193 break; 3194 } 3195 } 3196 nfsm_reqdone; 3197 break; 3198 case _PC_FILESIZEBITS: 3199 #ifndef NFS_V2_ONLY 3200 if (v3) { 3201 nmp = VFSTONFS(vp->v_mount); 3202 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3203 if ((error = nfs_fsinfo(nmp, vp, 3204 curproc->p_ucred, curproc)) != 0) /* XXX */ 3205 break; 3206 for (l = 0, maxsize = nmp->nm_maxfilesize; 3207 (maxsize >> l) > 0; l++) 3208 ; 3209 *ap->a_retval = l + 1; 3210 } else 3211 #endif 3212 { 3213 *ap->a_retval = 32; /* NFS V2 limitation */ 3214 } 3215 break; 3216 default: 3217 error = EINVAL; 3218 break; 3219 } 3220 3221 return (error); 3222 } 3223 3224 /* 3225 * NFS advisory byte-level locks. 3226 */ 3227 int 3228 nfs_advlock(v) 3229 void *v; 3230 { 3231 struct vop_advlock_args /* { 3232 struct vnode *a_vp; 3233 caddr_t a_id; 3234 int a_op; 3235 struct flock *a_fl; 3236 int a_flags; 3237 } */ *ap = v; 3238 struct nfsnode *np = VTONFS(ap->a_vp); 3239 3240 return lf_advlock(ap, &np->n_lockf, np->n_size); 3241 } 3242 3243 /* 3244 * Print out the contents of an nfsnode. 3245 */ 3246 int 3247 nfs_print(v) 3248 void *v; 3249 { 3250 struct vop_print_args /* { 3251 struct vnode *a_vp; 3252 } */ *ap = v; 3253 struct vnode *vp = ap->a_vp; 3254 struct nfsnode *np = VTONFS(vp); 3255 3256 printf("tag VT_NFS, fileid %ld fsid 0x%lx", 3257 np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3258 if (vp->v_type == VFIFO) 3259 fifo_printinfo(vp); 3260 printf("\n"); 3261 return (0); 3262 } 3263 3264 /* 3265 * NFS file truncation. 3266 */ 3267 int 3268 nfs_truncate(v) 3269 void *v; 3270 { 3271 #if 0 3272 struct vop_truncate_args /* { 3273 struct vnode *a_vp; 3274 off_t a_length; 3275 int a_flags; 3276 struct ucred *a_cred; 3277 struct proc *a_p; 3278 } */ *ap = v; 3279 #endif 3280 3281 /* Use nfs_setattr */ 3282 return (EOPNOTSUPP); 3283 } 3284 3285 /* 3286 * NFS update. 3287 */ 3288 int 3289 nfs_update(v) 3290 void *v; 3291 #if 0 3292 struct vop_update_args /* { 3293 struct vnode *a_vp; 3294 struct timespec *a_ta; 3295 struct timespec *a_tm; 3296 int a_waitfor; 3297 } */ *ap = v; 3298 #endif 3299 { 3300 3301 /* Use nfs_setattr */ 3302 return (EOPNOTSUPP); 3303 } 3304 3305 /* 3306 * Just call bwrite(). 3307 */ 3308 int 3309 nfs_bwrite(v) 3310 void *v; 3311 { 3312 struct vop_bwrite_args /* { 3313 struct vnode *a_bp; 3314 } */ *ap = v; 3315 3316 return (bwrite(ap->a_bp)); 3317 } 3318 3319 /* 3320 * nfs unlock wrapper. 3321 */ 3322 int 3323 nfs_unlock(void *v) 3324 { 3325 struct vop_unlock_args /* { 3326 struct vnode *a_vp; 3327 int a_flags; 3328 } */ *ap = v; 3329 struct vnode *vp = ap->a_vp; 3330 3331 /* 3332 * VOP_UNLOCK can be called by nfs_loadattrcache 3333 * with v_data == 0. 3334 */ 3335 if (VTONFS(vp)) { 3336 nfs_delayedtruncate(vp); 3337 } 3338 3339 return genfs_unlock(v); 3340 } 3341 3342 /* 3343 * nfs special file access vnode op. 3344 * Essentially just get vattr and then imitate iaccess() since the device is 3345 * local to the client. 3346 */ 3347 int 3348 nfsspec_access(v) 3349 void *v; 3350 { 3351 struct vop_access_args /* { 3352 struct vnode *a_vp; 3353 int a_mode; 3354 struct ucred *a_cred; 3355 struct proc *a_p; 3356 } */ *ap = v; 3357 struct vattr va; 3358 struct vnode *vp = ap->a_vp; 3359 int error; 3360 3361 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p); 3362 if (error) 3363 return (error); 3364 3365 /* 3366 * Disallow write attempts on filesystems mounted read-only; 3367 * unless the file is a socket, fifo, or a block or character 3368 * device resident on the filesystem. 3369 */ 3370 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3371 switch (vp->v_type) { 3372 case VREG: 3373 case VDIR: 3374 case VLNK: 3375 return (EROFS); 3376 default: 3377 break; 3378 } 3379 } 3380 3381 return (vaccess(va.va_type, va.va_mode, 3382 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3383 } 3384 3385 /* 3386 * Read wrapper for special devices. 3387 */ 3388 int 3389 nfsspec_read(v) 3390 void *v; 3391 { 3392 struct vop_read_args /* { 3393 struct vnode *a_vp; 3394 struct uio *a_uio; 3395 int a_ioflag; 3396 struct ucred *a_cred; 3397 } */ *ap = v; 3398 struct nfsnode *np = VTONFS(ap->a_vp); 3399 3400 /* 3401 * Set access flag. 3402 */ 3403 np->n_flag |= NACC; 3404 np->n_atim.tv_sec = time.tv_sec; 3405 np->n_atim.tv_nsec = time.tv_usec * 1000; 3406 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3407 } 3408 3409 /* 3410 * Write wrapper for special devices. 3411 */ 3412 int 3413 nfsspec_write(v) 3414 void *v; 3415 { 3416 struct vop_write_args /* { 3417 struct vnode *a_vp; 3418 struct uio *a_uio; 3419 int a_ioflag; 3420 struct ucred *a_cred; 3421 } */ *ap = v; 3422 struct nfsnode *np = VTONFS(ap->a_vp); 3423 3424 /* 3425 * Set update flag. 3426 */ 3427 np->n_flag |= NUPD; 3428 np->n_mtim.tv_sec = time.tv_sec; 3429 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3430 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3431 } 3432 3433 /* 3434 * Close wrapper for special devices. 3435 * 3436 * Update the times on the nfsnode then do device close. 3437 */ 3438 int 3439 nfsspec_close(v) 3440 void *v; 3441 { 3442 struct vop_close_args /* { 3443 struct vnode *a_vp; 3444 int a_fflag; 3445 struct ucred *a_cred; 3446 struct proc *a_p; 3447 } */ *ap = v; 3448 struct vnode *vp = ap->a_vp; 3449 struct nfsnode *np = VTONFS(vp); 3450 struct vattr vattr; 3451 3452 if (np->n_flag & (NACC | NUPD)) { 3453 np->n_flag |= NCHG; 3454 if (vp->v_usecount == 1 && 3455 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3456 VATTR_NULL(&vattr); 3457 if (np->n_flag & NACC) 3458 vattr.va_atime = np->n_atim; 3459 if (np->n_flag & NUPD) 3460 vattr.va_mtime = np->n_mtim; 3461 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3462 } 3463 } 3464 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3465 } 3466 3467 /* 3468 * Read wrapper for fifos. 3469 */ 3470 int 3471 nfsfifo_read(v) 3472 void *v; 3473 { 3474 struct vop_read_args /* { 3475 struct vnode *a_vp; 3476 struct uio *a_uio; 3477 int a_ioflag; 3478 struct ucred *a_cred; 3479 } */ *ap = v; 3480 struct nfsnode *np = VTONFS(ap->a_vp); 3481 3482 /* 3483 * Set access flag. 3484 */ 3485 np->n_flag |= NACC; 3486 np->n_atim.tv_sec = time.tv_sec; 3487 np->n_atim.tv_nsec = time.tv_usec * 1000; 3488 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3489 } 3490 3491 /* 3492 * Write wrapper for fifos. 3493 */ 3494 int 3495 nfsfifo_write(v) 3496 void *v; 3497 { 3498 struct vop_write_args /* { 3499 struct vnode *a_vp; 3500 struct uio *a_uio; 3501 int a_ioflag; 3502 struct ucred *a_cred; 3503 } */ *ap = v; 3504 struct nfsnode *np = VTONFS(ap->a_vp); 3505 3506 /* 3507 * Set update flag. 3508 */ 3509 np->n_flag |= NUPD; 3510 np->n_mtim.tv_sec = time.tv_sec; 3511 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3512 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3513 } 3514 3515 /* 3516 * Close wrapper for fifos. 3517 * 3518 * Update the times on the nfsnode then do fifo close. 3519 */ 3520 int 3521 nfsfifo_close(v) 3522 void *v; 3523 { 3524 struct vop_close_args /* { 3525 struct vnode *a_vp; 3526 int a_fflag; 3527 struct ucred *a_cred; 3528 struct proc *a_p; 3529 } */ *ap = v; 3530 struct vnode *vp = ap->a_vp; 3531 struct nfsnode *np = VTONFS(vp); 3532 struct vattr vattr; 3533 3534 if (np->n_flag & (NACC | NUPD)) { 3535 if (np->n_flag & NACC) { 3536 np->n_atim.tv_sec = time.tv_sec; 3537 np->n_atim.tv_nsec = time.tv_usec * 1000; 3538 } 3539 if (np->n_flag & NUPD) { 3540 np->n_mtim.tv_sec = time.tv_sec; 3541 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3542 } 3543 np->n_flag |= NCHG; 3544 if (vp->v_usecount == 1 && 3545 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3546 VATTR_NULL(&vattr); 3547 if (np->n_flag & NACC) 3548 vattr.va_atime = np->n_atim; 3549 if (np->n_flag & NUPD) 3550 vattr.va_mtime = np->n_mtim; 3551 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3552 } 3553 } 3554 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3555 } 3556
Indexes created Mon Sep 22 13:09:51 GMT 2025