nfs_vnops.c revision 1.192
1 /* $NetBSD: nfs_vnops.c,v 1.192 2004/05/06 21:58:17 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 1.192 2004/05/06 21:58:17 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 /* 921 * because nfsv3 has the same CREATE semantics as ours, 922 * we don't have to perform LOOKUPs beforehand. 923 * 924 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 925 * XXX although we have no way to know if O_EXCL is requested or not. 926 */ 927 928 if (v3 && cnp->cn_nameiop == CREATE && (flags & ISLASTCN)) { 929 KASSERT(lockparent); 930 cnp->cn_flags |= SAVENAME; 931 return (EJUSTRETURN); 932 } 933 934 error = 0; 935 newvp = NULLVP; 936 nfsstats.lookupcache_misses++; 937 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 938 len = cnp->cn_namelen; 939 nfsm_reqhead(np, NFSPROC_LOOKUP, 940 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 941 nfsm_fhtom(np, v3); 942 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 943 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred); 944 if (error) { 945 nfsm_postop_attr(dvp, attrflag, 0); 946 m_freem(mrep); 947 goto nfsmout; 948 } 949 nfsm_getfh(fhp, fhsize, v3); 950 951 /* 952 * Handle RENAME case... 953 */ 954 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) { 955 if (NFS_CMPFH(np, fhp, fhsize)) { 956 m_freem(mrep); 957 return (EISDIR); 958 } 959 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 960 if (error) { 961 m_freem(mrep); 962 return error; 963 } 964 newvp = NFSTOV(np); 965 if (v3) { 966 nfsm_postop_attr(newvp, attrflag, 0); 967 nfsm_postop_attr(dvp, attrflag, 0); 968 } else 969 nfsm_loadattr(newvp, (struct vattr *)0, 0); 970 *vpp = newvp; 971 m_freem(mrep); 972 cnp->cn_flags |= SAVENAME; 973 if (!lockparent) { 974 VOP_UNLOCK(dvp, 0); 975 cnp->cn_flags |= PDIRUNLOCK; 976 } 977 goto validate; 978 } 979 980 /* 981 * The postop attr handling is duplicated for each if case, 982 * because it should be done while dvp is locked (unlocking 983 * dvp is different for each case). 984 */ 985 986 if (NFS_CMPFH(np, fhp, fhsize)) { 987 /* 988 * "." lookup 989 */ 990 VREF(dvp); 991 newvp = dvp; 992 if (v3) { 993 nfsm_postop_attr(newvp, attrflag, 0); 994 nfsm_postop_attr(dvp, attrflag, 0); 995 } else 996 nfsm_loadattr(newvp, (struct vattr *)0, 0); 997 } else if (flags & ISDOTDOT) { 998 /* 999 * ".." lookup 1000 */ 1001 VOP_UNLOCK(dvp, 0); 1002 cnp->cn_flags |= PDIRUNLOCK; 1003 1004 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1005 if (error) { 1006 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 1007 cnp->cn_flags &= ~PDIRUNLOCK; 1008 m_freem(mrep); 1009 return error; 1010 } 1011 newvp = NFSTOV(np); 1012 1013 if (v3) { 1014 nfsm_postop_attr(newvp, attrflag, 0); 1015 nfsm_postop_attr(dvp, attrflag, 0); 1016 } else 1017 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1018 1019 if (lockparent && (flags & ISLASTCN)) { 1020 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) { 1021 m_freem(mrep); 1022 vput(newvp); 1023 return error; 1024 } 1025 cnp->cn_flags &= ~PDIRUNLOCK; 1026 } 1027 } else { 1028 /* 1029 * Other lookups. 1030 */ 1031 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1032 if (error) { 1033 m_freem(mrep); 1034 return error; 1035 } 1036 newvp = NFSTOV(np); 1037 if (v3) { 1038 nfsm_postop_attr(newvp, attrflag, 0); 1039 nfsm_postop_attr(dvp, attrflag, 0); 1040 } else 1041 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1042 if (!lockparent || !(flags & ISLASTCN)) { 1043 VOP_UNLOCK(dvp, 0); 1044 cnp->cn_flags |= PDIRUNLOCK; 1045 } 1046 } 1047 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1048 cnp->cn_flags |= SAVENAME; 1049 if ((cnp->cn_flags & MAKEENTRY) && 1050 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1051 nfs_cache_enter(dvp, newvp, cnp); 1052 } 1053 *vpp = newvp; 1054 nfsm_reqdone; 1055 if (error) { 1056 /* 1057 * We get here only because of errors returned by 1058 * the RPC. Otherwise we'll have returned above 1059 * (the nfsm_* macros will jump to nfsm_reqdone 1060 * on error). 1061 */ 1062 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1063 cnp->cn_nameiop != CREATE) { 1064 nfs_cache_enter(dvp, NULL, cnp); 1065 } 1066 if (newvp != NULLVP) { 1067 vrele(newvp); 1068 if (newvp != dvp) 1069 VOP_UNLOCK(newvp, 0); 1070 } 1071 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1072 (flags & ISLASTCN) && error == ENOENT) { 1073 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1074 error = EROFS; 1075 else 1076 error = EJUSTRETURN; 1077 } 1078 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1079 cnp->cn_flags |= SAVENAME; 1080 *vpp = NULL; 1081 return error; 1082 } 1083 1084 validate: 1085 /* 1086 * make sure we have valid type and size. 1087 */ 1088 1089 newvp = *vpp; 1090 if (newvp->v_type == VNON) { 1091 struct vattr vattr; /* dummy */ 1092 1093 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1094 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc); 1095 if (error) { 1096 vput(newvp); 1097 *vpp = NULL; 1098 } 1099 } 1100 1101 return error; 1102 } 1103 1104 /* 1105 * nfs read call. 1106 * Just call nfs_bioread() to do the work. 1107 */ 1108 int 1109 nfs_read(v) 1110 void *v; 1111 { 1112 struct vop_read_args /* { 1113 struct vnode *a_vp; 1114 struct uio *a_uio; 1115 int a_ioflag; 1116 struct ucred *a_cred; 1117 } */ *ap = v; 1118 struct vnode *vp = ap->a_vp; 1119 1120 if (vp->v_type != VREG) 1121 return (EPERM); 1122 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1123 } 1124 1125 /* 1126 * nfs readlink call 1127 */ 1128 int 1129 nfs_readlink(v) 1130 void *v; 1131 { 1132 struct vop_readlink_args /* { 1133 struct vnode *a_vp; 1134 struct uio *a_uio; 1135 struct ucred *a_cred; 1136 } */ *ap = v; 1137 struct vnode *vp = ap->a_vp; 1138 1139 if (vp->v_type != VLNK) 1140 return (EPERM); 1141 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1142 } 1143 1144 /* 1145 * Do a readlink rpc. 1146 * Called by nfs_doio() from below the buffer cache. 1147 */ 1148 int 1149 nfs_readlinkrpc(vp, uiop, cred) 1150 struct vnode *vp; 1151 struct uio *uiop; 1152 struct ucred *cred; 1153 { 1154 u_int32_t *tl; 1155 caddr_t cp; 1156 int32_t t1, t2; 1157 caddr_t bpos, dpos, cp2; 1158 int error = 0, attrflag; 1159 uint32_t len; 1160 struct mbuf *mreq, *mrep, *md, *mb; 1161 const int v3 = NFS_ISV3(vp); 1162 struct nfsnode *np = VTONFS(vp); 1163 1164 nfsstats.rpccnt[NFSPROC_READLINK]++; 1165 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1166 nfsm_fhtom(np, v3); 1167 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred); 1168 if (v3) 1169 nfsm_postop_attr(vp, attrflag, 0); 1170 if (!error) { 1171 if (v3) { 1172 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1173 len = fxdr_unsigned(uint32_t, *tl); 1174 if (len > MAXPATHLEN) { 1175 /* 1176 * this pathname is too long for us. 1177 */ 1178 m_freem(mrep); 1179 /* Solaris returns EINVAL. should we follow? */ 1180 error = ENAMETOOLONG; 1181 goto nfsmout; 1182 } 1183 } else { 1184 nfsm_strsiz(len, NFS_MAXPATHLEN); 1185 } 1186 nfsm_mtouio(uiop, len); 1187 } 1188 nfsm_reqdone; 1189 return (error); 1190 } 1191 1192 /* 1193 * nfs read rpc call 1194 * Ditto above 1195 */ 1196 int 1197 nfs_readrpc(vp, uiop) 1198 struct vnode *vp; 1199 struct uio *uiop; 1200 { 1201 u_int32_t *tl; 1202 caddr_t cp; 1203 int32_t t1, t2; 1204 caddr_t bpos, dpos, cp2; 1205 struct mbuf *mreq, *mrep, *md, *mb; 1206 struct nfsmount *nmp; 1207 int error = 0, len, retlen, tsiz, eof, attrflag; 1208 const int v3 = NFS_ISV3(vp); 1209 struct nfsnode *np = VTONFS(vp); 1210 1211 #ifndef nolint 1212 eof = 0; 1213 #endif 1214 nmp = VFSTONFS(vp->v_mount); 1215 tsiz = uiop->uio_resid; 1216 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1217 return (EFBIG); 1218 while (tsiz > 0) { 1219 nfsstats.rpccnt[NFSPROC_READ]++; 1220 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1221 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1222 nfsm_fhtom(np, v3); 1223 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1224 if (v3) { 1225 txdr_hyper(uiop->uio_offset, tl); 1226 *(tl + 2) = txdr_unsigned(len); 1227 } else { 1228 *tl++ = txdr_unsigned(uiop->uio_offset); 1229 *tl++ = txdr_unsigned(len); 1230 *tl = 0; 1231 } 1232 nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred); 1233 if (v3) { 1234 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1235 if (error) { 1236 m_freem(mrep); 1237 goto nfsmout; 1238 } 1239 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1240 eof = fxdr_unsigned(int, *(tl + 1)); 1241 } else 1242 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1243 nfsm_strsiz(retlen, nmp->nm_rsize); 1244 nfsm_mtouio(uiop, retlen); 1245 m_freem(mrep); 1246 tsiz -= retlen; 1247 if (v3) { 1248 if (eof || retlen == 0) 1249 tsiz = 0; 1250 } else if (retlen < len) 1251 tsiz = 0; 1252 } 1253 nfsmout: 1254 return (error); 1255 } 1256 1257 struct nfs_writerpc_context { 1258 struct simplelock nwc_slock; 1259 volatile int nwc_mbufcount; 1260 }; 1261 1262 /* 1263 * free mbuf used to refer protected pages while write rpc call. 1264 * called at splvm. 1265 */ 1266 static void 1267 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 1268 { 1269 struct nfs_writerpc_context *ctx = arg; 1270 1271 KASSERT(m != NULL); 1272 KASSERT(ctx != NULL); 1273 pool_cache_put(&mbpool_cache, m); 1274 simple_lock(&ctx->nwc_slock); 1275 if (--ctx->nwc_mbufcount == 0) { 1276 wakeup(ctx); 1277 } 1278 simple_unlock(&ctx->nwc_slock); 1279 } 1280 1281 /* 1282 * nfs write call 1283 */ 1284 int 1285 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1286 struct vnode *vp; 1287 struct uio *uiop; 1288 int *iomode; 1289 boolean_t pageprotected; 1290 boolean_t *stalewriteverfp; 1291 { 1292 u_int32_t *tl; 1293 caddr_t cp; 1294 int32_t t1, t2; 1295 caddr_t bpos, dpos, cp2; 1296 struct mbuf *mreq, *mrep, *md, *mb; 1297 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1298 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1299 const int v3 = NFS_ISV3(vp); 1300 int committed = NFSV3WRITE_FILESYNC; 1301 struct nfsnode *np = VTONFS(vp); 1302 struct nfs_writerpc_context ctx; 1303 int s; 1304 struct lwp *l = NULL; 1305 size_t origresid; 1306 1307 simple_lock_init(&ctx.nwc_slock); 1308 ctx.nwc_mbufcount = 1; 1309 1310 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1311 panic("writerpc readonly vp %p", vp); 1312 } 1313 1314 #ifdef DIAGNOSTIC 1315 if (uiop->uio_iovcnt != 1) 1316 panic("nfs: writerpc iovcnt > 1"); 1317 #endif 1318 tsiz = uiop->uio_resid; 1319 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1320 return (EFBIG); 1321 if (pageprotected) { 1322 l = curlwp; 1323 PHOLD(l); 1324 } 1325 retry: 1326 origresid = uiop->uio_resid; 1327 KASSERT(origresid == uiop->uio_iov->iov_len); 1328 while (tsiz > 0) { 1329 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1330 uint32_t backup; 1331 boolean_t stalewriteverf = FALSE; 1332 1333 nfsstats.rpccnt[NFSPROC_WRITE]++; 1334 len = min(tsiz, nmp->nm_wsize); 1335 datalen = pageprotected ? 0 : nfsm_rndup(len); 1336 nfsm_reqhead(np, NFSPROC_WRITE, 1337 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1338 nfsm_fhtom(np, v3); 1339 if (v3) { 1340 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1341 txdr_hyper(uiop->uio_offset, tl); 1342 tl += 2; 1343 *tl++ = txdr_unsigned(len); 1344 *tl++ = txdr_unsigned(*iomode); 1345 *tl = txdr_unsigned(len); 1346 } else { 1347 u_int32_t x; 1348 1349 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1350 /* Set both "begin" and "current" to non-garbage. */ 1351 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1352 *tl++ = x; /* "begin offset" */ 1353 *tl++ = x; /* "current offset" */ 1354 x = txdr_unsigned(len); 1355 *tl++ = x; /* total to this offset */ 1356 *tl = x; /* size of this write */ 1357 1358 } 1359 if (pageprotected) { 1360 /* 1361 * since we know pages can't be modified during i/o, 1362 * no need to copy them for us. 1363 */ 1364 struct mbuf *m; 1365 struct iovec *iovp = uiop->uio_iov; 1366 1367 m = m_get(M_WAIT, MT_DATA); 1368 MCLAIM(m, &nfs_mowner); 1369 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1370 nfs_writerpc_extfree, &ctx); 1371 m->m_flags |= M_EXT_ROMAP; 1372 m->m_len = len; 1373 mb->m_next = m; 1374 /* 1375 * no need to maintain mb and bpos here 1376 * because no one care them later. 1377 */ 1378 #if 0 1379 mb = m; 1380 bpos = mtod(caddr_t, mb) + mb->m_len; 1381 #endif 1382 iovp->iov_base = (char *)iovp->iov_base + len; 1383 iovp->iov_len -= len; 1384 uiop->uio_offset += len; 1385 uiop->uio_resid -= len; 1386 s = splvm(); 1387 simple_lock(&ctx.nwc_slock); 1388 ctx.nwc_mbufcount++; 1389 simple_unlock(&ctx.nwc_slock); 1390 splx(s); 1391 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1392 } else { 1393 nfsm_uiotom(uiop, len); 1394 } 1395 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred); 1396 if (v3) { 1397 wccflag = NFSV3_WCCCHK; 1398 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC); 1399 if (!error) { 1400 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1401 + NFSX_V3WRITEVERF); 1402 rlen = fxdr_unsigned(int, *tl++); 1403 if (rlen == 0) { 1404 error = NFSERR_IO; 1405 m_freem(mrep); 1406 break; 1407 } else if (rlen < len) { 1408 backup = len - rlen; 1409 uiop->uio_iov->iov_base = 1410 (caddr_t)uiop->uio_iov->iov_base - 1411 backup; 1412 uiop->uio_iov->iov_len += backup; 1413 uiop->uio_offset -= backup; 1414 uiop->uio_resid += backup; 1415 len = rlen; 1416 } 1417 commit = fxdr_unsigned(int, *tl++); 1418 1419 /* 1420 * Return the lowest committment level 1421 * obtained by any of the RPCs. 1422 */ 1423 if (committed == NFSV3WRITE_FILESYNC) 1424 committed = commit; 1425 else if (committed == NFSV3WRITE_DATASYNC && 1426 commit == NFSV3WRITE_UNSTABLE) 1427 committed = commit; 1428 simple_lock(&nmp->nm_slock); 1429 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1430 memcpy(nmp->nm_writeverf, tl, 1431 NFSX_V3WRITEVERF); 1432 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1433 } else if ((nmp->nm_iflag & 1434 NFSMNT_STALEWRITEVERF) || 1435 memcmp(tl, nmp->nm_writeverf, 1436 NFSX_V3WRITEVERF)) { 1437 memcpy(nmp->nm_writeverf, tl, 1438 NFSX_V3WRITEVERF); 1439 /* 1440 * note NFSMNT_STALEWRITEVERF 1441 * if we're the first thread to 1442 * notice it. 1443 */ 1444 if ((nmp->nm_iflag & 1445 NFSMNT_STALEWRITEVERF) == 0) { 1446 stalewriteverf = TRUE; 1447 nmp->nm_iflag |= 1448 NFSMNT_STALEWRITEVERF; 1449 } 1450 } 1451 simple_unlock(&nmp->nm_slock); 1452 } 1453 } else 1454 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1455 if (wccflag) 1456 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1457 m_freem(mrep); 1458 if (error) 1459 break; 1460 tsiz -= len; 1461 if (stalewriteverf) { 1462 *stalewriteverfp = TRUE; 1463 stalewriteverf = FALSE; 1464 if (committed == NFSV3WRITE_UNSTABLE && 1465 len != origresid) { 1466 /* 1467 * if our write requests weren't atomic but 1468 * unstable, datas in previous iterations 1469 * might have already been lost now. 1470 * then, we should resend them to nfsd. 1471 */ 1472 backup = origresid - tsiz; 1473 uiop->uio_iov->iov_base = 1474 (caddr_t)uiop->uio_iov->iov_base - backup; 1475 uiop->uio_iov->iov_len += backup; 1476 uiop->uio_offset -= backup; 1477 uiop->uio_resid += backup; 1478 tsiz = origresid; 1479 goto retry; 1480 } 1481 } 1482 } 1483 if (pageprotected) { 1484 /* 1485 * wait until mbufs go away. 1486 * retransmitted mbufs can survive longer than rpc requests 1487 * themselves. 1488 */ 1489 s = splvm(); 1490 simple_lock(&ctx.nwc_slock); 1491 ctx.nwc_mbufcount--; 1492 while (ctx.nwc_mbufcount > 0) { 1493 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock); 1494 } 1495 simple_unlock(&ctx.nwc_slock); 1496 splx(s); 1497 PRELE(l); 1498 } 1499 nfsmout: 1500 *iomode = committed; 1501 if (error) 1502 uiop->uio_resid = tsiz; 1503 return (error); 1504 } 1505 1506 /* 1507 * nfs mknod rpc 1508 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1509 * mode set to specify the file type and the size field for rdev. 1510 */ 1511 int 1512 nfs_mknodrpc(dvp, vpp, cnp, vap) 1513 struct vnode *dvp; 1514 struct vnode **vpp; 1515 struct componentname *cnp; 1516 struct vattr *vap; 1517 { 1518 struct nfsv2_sattr *sp; 1519 u_int32_t *tl; 1520 caddr_t cp; 1521 int32_t t1, t2; 1522 struct vnode *newvp = (struct vnode *)0; 1523 struct nfsnode *dnp, *np; 1524 char *cp2; 1525 caddr_t bpos, dpos; 1526 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1527 struct mbuf *mreq, *mrep, *md, *mb; 1528 u_int32_t rdev; 1529 const int v3 = NFS_ISV3(dvp); 1530 1531 if (vap->va_type == VCHR || vap->va_type == VBLK) 1532 rdev = txdr_unsigned(vap->va_rdev); 1533 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1534 rdev = nfs_xdrneg1; 1535 else { 1536 VOP_ABORTOP(dvp, cnp); 1537 vput(dvp); 1538 return (EOPNOTSUPP); 1539 } 1540 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1541 dnp = VTONFS(dvp); 1542 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1543 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1544 nfsm_fhtom(dnp, v3); 1545 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1546 if (v3) { 1547 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1548 *tl++ = vtonfsv3_type(vap->va_type); 1549 nfsm_v3attrbuild(vap, FALSE); 1550 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1551 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1552 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1553 *tl = txdr_unsigned(minor(vap->va_rdev)); 1554 } 1555 } else { 1556 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1557 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1558 sp->sa_uid = nfs_xdrneg1; 1559 sp->sa_gid = nfs_xdrneg1; 1560 sp->sa_size = rdev; 1561 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1562 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1563 } 1564 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred); 1565 if (!error) { 1566 nfsm_mtofh(dvp, newvp, v3, gotvp); 1567 if (!gotvp) { 1568 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1569 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1570 if (!error) 1571 newvp = NFSTOV(np); 1572 } 1573 } 1574 if (v3) 1575 nfsm_wcc_data(dvp, wccflag, 0); 1576 nfsm_reqdone; 1577 if (error) { 1578 if (newvp) 1579 vput(newvp); 1580 } else { 1581 if (cnp->cn_flags & MAKEENTRY) 1582 nfs_cache_enter(dvp, newvp, cnp); 1583 *vpp = newvp; 1584 } 1585 PNBUF_PUT(cnp->cn_pnbuf); 1586 VTONFS(dvp)->n_flag |= NMODIFIED; 1587 if (!wccflag) 1588 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1589 vput(dvp); 1590 return (error); 1591 } 1592 1593 /* 1594 * nfs mknod vop 1595 * just call nfs_mknodrpc() to do the work. 1596 */ 1597 /* ARGSUSED */ 1598 int 1599 nfs_mknod(v) 1600 void *v; 1601 { 1602 struct vop_mknod_args /* { 1603 struct vnode *a_dvp; 1604 struct vnode **a_vpp; 1605 struct componentname *a_cnp; 1606 struct vattr *a_vap; 1607 } */ *ap = v; 1608 int error; 1609 1610 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1611 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1612 return (error); 1613 } 1614 1615 static u_long create_verf; 1616 /* 1617 * nfs file create call 1618 */ 1619 int 1620 nfs_create(v) 1621 void *v; 1622 { 1623 struct vop_create_args /* { 1624 struct vnode *a_dvp; 1625 struct vnode **a_vpp; 1626 struct componentname *a_cnp; 1627 struct vattr *a_vap; 1628 } */ *ap = v; 1629 struct vnode *dvp = ap->a_dvp; 1630 struct vattr *vap = ap->a_vap; 1631 struct componentname *cnp = ap->a_cnp; 1632 struct nfsv2_sattr *sp; 1633 u_int32_t *tl; 1634 caddr_t cp; 1635 int32_t t1, t2; 1636 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1637 struct vnode *newvp = (struct vnode *)0; 1638 caddr_t bpos, dpos, cp2; 1639 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1640 struct mbuf *mreq, *mrep, *md, *mb; 1641 const int v3 = NFS_ISV3(dvp); 1642 1643 /* 1644 * Oops, not for me.. 1645 */ 1646 if (vap->va_type == VSOCK) 1647 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1648 1649 #ifdef VA_EXCLUSIVE 1650 if (vap->va_vaflags & VA_EXCLUSIVE) 1651 fmode |= O_EXCL; 1652 #endif 1653 again: 1654 error = 0; 1655 nfsstats.rpccnt[NFSPROC_CREATE]++; 1656 dnp = VTONFS(dvp); 1657 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1658 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1659 nfsm_fhtom(dnp, v3); 1660 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1661 if (v3) { 1662 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1663 if (fmode & O_EXCL) { 1664 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1665 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1666 #ifdef INET 1667 if (TAILQ_FIRST(&in_ifaddr)) 1668 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr; 1669 else 1670 *tl++ = create_verf; 1671 #else 1672 *tl++ = create_verf; 1673 #endif 1674 *tl = ++create_verf; 1675 } else { 1676 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1677 nfsm_v3attrbuild(vap, FALSE); 1678 } 1679 } else { 1680 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1681 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1682 sp->sa_uid = nfs_xdrneg1; 1683 sp->sa_gid = nfs_xdrneg1; 1684 sp->sa_size = 0; 1685 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1686 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1687 } 1688 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred); 1689 if (!error) { 1690 nfsm_mtofh(dvp, newvp, v3, gotvp); 1691 if (!gotvp) { 1692 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1693 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1694 if (!error) 1695 newvp = NFSTOV(np); 1696 } 1697 } 1698 if (v3) 1699 nfsm_wcc_data(dvp, wccflag, 0); 1700 nfsm_reqdone; 1701 if (error) { 1702 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1703 fmode &= ~O_EXCL; 1704 goto again; 1705 } 1706 if (newvp) 1707 vput(newvp); 1708 } else if (v3 && (fmode & O_EXCL)) 1709 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc); 1710 if (!error) { 1711 if (cnp->cn_flags & MAKEENTRY) 1712 nfs_cache_enter(dvp, newvp, cnp); 1713 *ap->a_vpp = newvp; 1714 } 1715 PNBUF_PUT(cnp->cn_pnbuf); 1716 VTONFS(dvp)->n_flag |= NMODIFIED; 1717 if (!wccflag) 1718 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1719 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1720 vput(dvp); 1721 return (error); 1722 } 1723 1724 /* 1725 * nfs file remove call 1726 * To try and make nfs semantics closer to ufs semantics, a file that has 1727 * other processes using the vnode is renamed instead of removed and then 1728 * removed later on the last close. 1729 * - If v_usecount > 1 1730 * If a rename is not already in the works 1731 * call nfs_sillyrename() to set it up 1732 * else 1733 * do the remove rpc 1734 */ 1735 int 1736 nfs_remove(v) 1737 void *v; 1738 { 1739 struct vop_remove_args /* { 1740 struct vnodeop_desc *a_desc; 1741 struct vnode * a_dvp; 1742 struct vnode * a_vp; 1743 struct componentname * a_cnp; 1744 } */ *ap = v; 1745 struct vnode *vp = ap->a_vp; 1746 struct vnode *dvp = ap->a_dvp; 1747 struct componentname *cnp = ap->a_cnp; 1748 struct nfsnode *np = VTONFS(vp); 1749 int error = 0; 1750 struct vattr vattr; 1751 1752 #ifndef DIAGNOSTIC 1753 if ((cnp->cn_flags & HASBUF) == 0) 1754 panic("nfs_remove: no name"); 1755 if (vp->v_usecount < 1) 1756 panic("nfs_remove: bad v_usecount"); 1757 #endif 1758 if (vp->v_type == VDIR) 1759 error = EPERM; 1760 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1761 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 && 1762 vattr.va_nlink > 1)) { 1763 /* 1764 * Purge the name cache so that the chance of a lookup for 1765 * the name succeeding while the remove is in progress is 1766 * minimized. Without node locking it can still happen, such 1767 * that an I/O op returns ESTALE, but since you get this if 1768 * another host removes the file.. 1769 */ 1770 cache_purge(vp); 1771 /* 1772 * throw away biocache buffers, mainly to avoid 1773 * unnecessary delayed writes later. 1774 */ 1775 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1); 1776 /* Do the rpc */ 1777 if (error != EINTR) 1778 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1779 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc); 1780 /* 1781 * Kludge City: If the first reply to the remove rpc is lost.. 1782 * the reply to the retransmitted request will be ENOENT 1783 * since the file was in fact removed 1784 * Therefore, we cheat and return success. 1785 */ 1786 if (error == ENOENT) 1787 error = 0; 1788 } else if (!np->n_sillyrename) 1789 error = nfs_sillyrename(dvp, vp, cnp); 1790 PNBUF_PUT(cnp->cn_pnbuf); 1791 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1792 vattr.va_nlink == 1) { 1793 np->n_flag |= NREMOVED; 1794 } 1795 NFS_INVALIDATE_ATTRCACHE(np); 1796 VN_KNOTE(vp, NOTE_DELETE); 1797 VN_KNOTE(dvp, NOTE_WRITE); 1798 if (dvp == vp) 1799 vrele(vp); 1800 else 1801 vput(vp); 1802 vput(dvp); 1803 return (error); 1804 } 1805 1806 /* 1807 * nfs file remove rpc called from nfs_inactive 1808 */ 1809 int 1810 nfs_removeit(sp) 1811 struct sillyrename *sp; 1812 { 1813 1814 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1815 (struct proc *)0)); 1816 } 1817 1818 /* 1819 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1820 */ 1821 int 1822 nfs_removerpc(dvp, name, namelen, cred, proc) 1823 struct vnode *dvp; 1824 const char *name; 1825 int namelen; 1826 struct ucred *cred; 1827 struct proc *proc; 1828 { 1829 u_int32_t *tl; 1830 caddr_t cp; 1831 int32_t t1, t2; 1832 caddr_t bpos, dpos, cp2; 1833 int error = 0, wccflag = NFSV3_WCCRATTR; 1834 struct mbuf *mreq, *mrep, *md, *mb; 1835 const int v3 = NFS_ISV3(dvp); 1836 struct nfsnode *dnp = VTONFS(dvp); 1837 1838 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1839 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1840 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1841 nfsm_fhtom(dnp, v3); 1842 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1843 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred); 1844 if (v3) 1845 nfsm_wcc_data(dvp, wccflag, 0); 1846 nfsm_reqdone; 1847 VTONFS(dvp)->n_flag |= NMODIFIED; 1848 if (!wccflag) 1849 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1850 return (error); 1851 } 1852 1853 /* 1854 * nfs file rename call 1855 */ 1856 int 1857 nfs_rename(v) 1858 void *v; 1859 { 1860 struct vop_rename_args /* { 1861 struct vnode *a_fdvp; 1862 struct vnode *a_fvp; 1863 struct componentname *a_fcnp; 1864 struct vnode *a_tdvp; 1865 struct vnode *a_tvp; 1866 struct componentname *a_tcnp; 1867 } */ *ap = v; 1868 struct vnode *fvp = ap->a_fvp; 1869 struct vnode *tvp = ap->a_tvp; 1870 struct vnode *fdvp = ap->a_fdvp; 1871 struct vnode *tdvp = ap->a_tdvp; 1872 struct componentname *tcnp = ap->a_tcnp; 1873 struct componentname *fcnp = ap->a_fcnp; 1874 int error; 1875 1876 #ifndef DIAGNOSTIC 1877 if ((tcnp->cn_flags & HASBUF) == 0 || 1878 (fcnp->cn_flags & HASBUF) == 0) 1879 panic("nfs_rename: no name"); 1880 #endif 1881 /* Check for cross-device rename */ 1882 if ((fvp->v_mount != tdvp->v_mount) || 1883 (tvp && (fvp->v_mount != tvp->v_mount))) { 1884 error = EXDEV; 1885 goto out; 1886 } 1887 1888 /* 1889 * If the tvp exists and is in use, sillyrename it before doing the 1890 * rename of the new file over it. 1891 */ 1892 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1893 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1894 VN_KNOTE(tvp, NOTE_DELETE); 1895 vput(tvp); 1896 tvp = NULL; 1897 } 1898 1899 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1900 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1901 tcnp->cn_proc); 1902 1903 VN_KNOTE(fdvp, NOTE_WRITE); 1904 VN_KNOTE(tdvp, NOTE_WRITE); 1905 if (fvp->v_type == VDIR) { 1906 if (tvp != NULL && tvp->v_type == VDIR) 1907 cache_purge(tdvp); 1908 cache_purge(fdvp); 1909 } 1910 out: 1911 if (tdvp == tvp) 1912 vrele(tdvp); 1913 else 1914 vput(tdvp); 1915 if (tvp) 1916 vput(tvp); 1917 vrele(fdvp); 1918 vrele(fvp); 1919 /* 1920 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1921 */ 1922 if (error == ENOENT) 1923 error = 0; 1924 return (error); 1925 } 1926 1927 /* 1928 * nfs file rename rpc called from nfs_remove() above 1929 */ 1930 int 1931 nfs_renameit(sdvp, scnp, sp) 1932 struct vnode *sdvp; 1933 struct componentname *scnp; 1934 struct sillyrename *sp; 1935 { 1936 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1937 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc)); 1938 } 1939 1940 /* 1941 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1942 */ 1943 int 1944 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc) 1945 struct vnode *fdvp; 1946 const char *fnameptr; 1947 int fnamelen; 1948 struct vnode *tdvp; 1949 const char *tnameptr; 1950 int tnamelen; 1951 struct ucred *cred; 1952 struct proc *proc; 1953 { 1954 u_int32_t *tl; 1955 caddr_t cp; 1956 int32_t t1, t2; 1957 caddr_t bpos, dpos, cp2; 1958 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1959 struct mbuf *mreq, *mrep, *md, *mb; 1960 const int v3 = NFS_ISV3(fdvp); 1961 struct nfsnode *fdnp = VTONFS(fdvp); 1962 1963 nfsstats.rpccnt[NFSPROC_RENAME]++; 1964 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1965 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1966 nfsm_rndup(tnamelen)); 1967 nfsm_fhtom(fdnp, v3); 1968 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1969 nfsm_fhtom(VTONFS(tdvp), v3); 1970 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1971 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred); 1972 if (v3) { 1973 nfsm_wcc_data(fdvp, fwccflag, 0); 1974 nfsm_wcc_data(tdvp, twccflag, 0); 1975 } 1976 nfsm_reqdone; 1977 VTONFS(fdvp)->n_flag |= NMODIFIED; 1978 VTONFS(tdvp)->n_flag |= NMODIFIED; 1979 if (!fwccflag) 1980 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1981 if (!twccflag) 1982 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1983 return (error); 1984 } 1985 1986 /* 1987 * nfs hard link create call 1988 */ 1989 int 1990 nfs_link(v) 1991 void *v; 1992 { 1993 struct vop_link_args /* { 1994 struct vnode *a_dvp; 1995 struct vnode *a_vp; 1996 struct componentname *a_cnp; 1997 } */ *ap = v; 1998 struct vnode *vp = ap->a_vp; 1999 struct vnode *dvp = ap->a_dvp; 2000 struct componentname *cnp = ap->a_cnp; 2001 u_int32_t *tl; 2002 caddr_t cp; 2003 int32_t t1, t2; 2004 caddr_t bpos, dpos, cp2; 2005 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2006 struct mbuf *mreq, *mrep, *md, *mb; 2007 /* XXX Should be const and initialised? */ 2008 int v3; 2009 struct nfsnode *np; 2010 2011 if (dvp->v_mount != vp->v_mount) { 2012 VOP_ABORTOP(dvp, cnp); 2013 vput(dvp); 2014 return (EXDEV); 2015 } 2016 if (dvp != vp) { 2017 error = vn_lock(vp, LK_EXCLUSIVE); 2018 if (error != 0) { 2019 VOP_ABORTOP(dvp, cnp); 2020 vput(dvp); 2021 return error; 2022 } 2023 } 2024 2025 /* 2026 * Push all writes to the server, so that the attribute cache 2027 * doesn't get "out of sync" with the server. 2028 * XXX There should be a better way! 2029 */ 2030 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc); 2031 2032 v3 = NFS_ISV3(vp); 2033 nfsstats.rpccnt[NFSPROC_LINK]++; 2034 np = VTONFS(vp); 2035 nfsm_reqhead(np, NFSPROC_LINK, 2036 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2037 nfsm_fhtom(np, v3); 2038 nfsm_fhtom(VTONFS(dvp), v3); 2039 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2040 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred); 2041 if (v3) { 2042 nfsm_postop_attr(vp, attrflag, 0); 2043 nfsm_wcc_data(dvp, wccflag, 0); 2044 } 2045 nfsm_reqdone; 2046 PNBUF_PUT(cnp->cn_pnbuf); 2047 VTONFS(dvp)->n_flag |= NMODIFIED; 2048 if (!attrflag) 2049 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2050 if (!wccflag) 2051 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2052 if (dvp != vp) 2053 VOP_UNLOCK(vp, 0); 2054 VN_KNOTE(vp, NOTE_LINK); 2055 VN_KNOTE(dvp, NOTE_WRITE); 2056 vput(dvp); 2057 /* 2058 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2059 */ 2060 if (error == EEXIST) 2061 error = 0; 2062 return (error); 2063 } 2064 2065 /* 2066 * nfs symbolic link create call 2067 */ 2068 int 2069 nfs_symlink(v) 2070 void *v; 2071 { 2072 struct vop_symlink_args /* { 2073 struct vnode *a_dvp; 2074 struct vnode **a_vpp; 2075 struct componentname *a_cnp; 2076 struct vattr *a_vap; 2077 char *a_target; 2078 } */ *ap = v; 2079 struct vnode *dvp = ap->a_dvp; 2080 struct vattr *vap = ap->a_vap; 2081 struct componentname *cnp = ap->a_cnp; 2082 struct nfsv2_sattr *sp; 2083 u_int32_t *tl; 2084 caddr_t cp; 2085 int32_t t1, t2; 2086 caddr_t bpos, dpos, cp2; 2087 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2088 struct mbuf *mreq, *mrep, *md, *mb; 2089 struct vnode *newvp = (struct vnode *)0; 2090 const int v3 = NFS_ISV3(dvp); 2091 struct nfsnode *dnp = VTONFS(dvp); 2092 2093 *ap->a_vpp = NULL; 2094 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2095 slen = strlen(ap->a_target); 2096 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2097 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2098 nfsm_fhtom(dnp, v3); 2099 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2100 if (v3) 2101 nfsm_v3attrbuild(vap, FALSE); 2102 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2103 if (!v3) { 2104 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2105 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2106 sp->sa_uid = nfs_xdrneg1; 2107 sp->sa_gid = nfs_xdrneg1; 2108 sp->sa_size = nfs_xdrneg1; 2109 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2110 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2111 } 2112 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred); 2113 if (v3) { 2114 if (!error) 2115 nfsm_mtofh(dvp, newvp, v3, gotvp); 2116 nfsm_wcc_data(dvp, wccflag, 0); 2117 } 2118 nfsm_reqdone; 2119 /* 2120 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2121 */ 2122 if (error == EEXIST) 2123 error = 0; 2124 if (error == 0 && newvp == NULL) { 2125 struct nfsnode *np = NULL; 2126 2127 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2128 cnp->cn_cred, cnp->cn_proc, &np); 2129 if (error == 0) 2130 newvp = NFSTOV(np); 2131 } 2132 if (error) { 2133 if (newvp != NULL) 2134 vput(newvp); 2135 } else { 2136 *ap->a_vpp = newvp; 2137 } 2138 PNBUF_PUT(cnp->cn_pnbuf); 2139 VTONFS(dvp)->n_flag |= NMODIFIED; 2140 if (!wccflag) 2141 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2142 VN_KNOTE(dvp, NOTE_WRITE); 2143 vput(dvp); 2144 return (error); 2145 } 2146 2147 /* 2148 * nfs make dir call 2149 */ 2150 int 2151 nfs_mkdir(v) 2152 void *v; 2153 { 2154 struct vop_mkdir_args /* { 2155 struct vnode *a_dvp; 2156 struct vnode **a_vpp; 2157 struct componentname *a_cnp; 2158 struct vattr *a_vap; 2159 } */ *ap = v; 2160 struct vnode *dvp = ap->a_dvp; 2161 struct vattr *vap = ap->a_vap; 2162 struct componentname *cnp = ap->a_cnp; 2163 struct nfsv2_sattr *sp; 2164 u_int32_t *tl; 2165 caddr_t cp; 2166 int32_t t1, t2; 2167 int len; 2168 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2169 struct vnode *newvp = (struct vnode *)0; 2170 caddr_t bpos, dpos, cp2; 2171 int error = 0, wccflag = NFSV3_WCCRATTR; 2172 int gotvp = 0; 2173 struct mbuf *mreq, *mrep, *md, *mb; 2174 const int v3 = NFS_ISV3(dvp); 2175 2176 len = cnp->cn_namelen; 2177 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2178 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2179 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2180 nfsm_fhtom(dnp, v3); 2181 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2182 if (v3) { 2183 nfsm_v3attrbuild(vap, FALSE); 2184 } else { 2185 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2186 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2187 sp->sa_uid = nfs_xdrneg1; 2188 sp->sa_gid = nfs_xdrneg1; 2189 sp->sa_size = nfs_xdrneg1; 2190 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2191 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2192 } 2193 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred); 2194 if (!error) 2195 nfsm_mtofh(dvp, newvp, v3, gotvp); 2196 if (v3) 2197 nfsm_wcc_data(dvp, wccflag, 0); 2198 nfsm_reqdone; 2199 VTONFS(dvp)->n_flag |= NMODIFIED; 2200 if (!wccflag) 2201 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2202 /* 2203 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2204 * if we can succeed in looking up the directory. 2205 */ 2206 if (error == EEXIST || (!error && !gotvp)) { 2207 if (newvp) { 2208 vput(newvp); 2209 newvp = (struct vnode *)0; 2210 } 2211 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2212 cnp->cn_proc, &np); 2213 if (!error) { 2214 newvp = NFSTOV(np); 2215 if (newvp->v_type != VDIR) 2216 error = EEXIST; 2217 } 2218 } 2219 if (error) { 2220 if (newvp) 2221 vput(newvp); 2222 } else { 2223 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2224 if (cnp->cn_flags & MAKEENTRY) 2225 nfs_cache_enter(dvp, newvp, cnp); 2226 *ap->a_vpp = newvp; 2227 } 2228 PNBUF_PUT(cnp->cn_pnbuf); 2229 vput(dvp); 2230 return (error); 2231 } 2232 2233 /* 2234 * nfs remove directory call 2235 */ 2236 int 2237 nfs_rmdir(v) 2238 void *v; 2239 { 2240 struct vop_rmdir_args /* { 2241 struct vnode *a_dvp; 2242 struct vnode *a_vp; 2243 struct componentname *a_cnp; 2244 } */ *ap = v; 2245 struct vnode *vp = ap->a_vp; 2246 struct vnode *dvp = ap->a_dvp; 2247 struct componentname *cnp = ap->a_cnp; 2248 u_int32_t *tl; 2249 caddr_t cp; 2250 int32_t t1, t2; 2251 caddr_t bpos, dpos, cp2; 2252 int error = 0, wccflag = NFSV3_WCCRATTR; 2253 struct mbuf *mreq, *mrep, *md, *mb; 2254 const int v3 = NFS_ISV3(dvp); 2255 struct nfsnode *dnp; 2256 2257 if (dvp == vp) { 2258 vrele(dvp); 2259 vput(dvp); 2260 PNBUF_PUT(cnp->cn_pnbuf); 2261 return (EINVAL); 2262 } 2263 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2264 dnp = VTONFS(dvp); 2265 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2266 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2267 nfsm_fhtom(dnp, v3); 2268 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2269 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred); 2270 if (v3) 2271 nfsm_wcc_data(dvp, wccflag, 0); 2272 nfsm_reqdone; 2273 PNBUF_PUT(cnp->cn_pnbuf); 2274 VTONFS(dvp)->n_flag |= NMODIFIED; 2275 if (!wccflag) 2276 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2277 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2278 VN_KNOTE(vp, NOTE_DELETE); 2279 cache_purge(vp); 2280 vput(vp); 2281 vput(dvp); 2282 /* 2283 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2284 */ 2285 if (error == ENOENT) 2286 error = 0; 2287 return (error); 2288 } 2289 2290 /* 2291 * nfs readdir call 2292 */ 2293 int 2294 nfs_readdir(v) 2295 void *v; 2296 { 2297 struct vop_readdir_args /* { 2298 struct vnode *a_vp; 2299 struct uio *a_uio; 2300 struct ucred *a_cred; 2301 int *a_eofflag; 2302 off_t **a_cookies; 2303 int *a_ncookies; 2304 } */ *ap = v; 2305 struct vnode *vp = ap->a_vp; 2306 struct uio *uio = ap->a_uio; 2307 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2308 char *base = uio->uio_iov->iov_base; 2309 int tresid, error; 2310 size_t count, lost; 2311 struct dirent *dp; 2312 off_t *cookies = NULL; 2313 int ncookies = 0, nc; 2314 2315 if (vp->v_type != VDIR) 2316 return (EPERM); 2317 2318 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2319 count = uio->uio_resid - lost; 2320 if (count <= 0) 2321 return (EINVAL); 2322 2323 /* 2324 * Call nfs_bioread() to do the real work. 2325 */ 2326 tresid = uio->uio_resid = count; 2327 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2328 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2329 2330 if (!error && ap->a_cookies) { 2331 ncookies = count / 16; 2332 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2333 *ap->a_cookies = cookies; 2334 } 2335 2336 if (!error && uio->uio_resid == tresid) { 2337 uio->uio_resid += lost; 2338 nfsstats.direofcache_misses++; 2339 if (ap->a_cookies) 2340 *ap->a_ncookies = 0; 2341 *ap->a_eofflag = 1; 2342 return (0); 2343 } 2344 2345 if (!error && ap->a_cookies) { 2346 /* 2347 * Only the NFS server and emulations use cookies, and they 2348 * load the directory block into system space, so we can 2349 * just look at it directly. 2350 */ 2351 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2352 panic("nfs_readdir: lost in space"); 2353 for (nc = 0; ncookies-- && 2354 base < (char *)uio->uio_iov->iov_base; nc++){ 2355 dp = (struct dirent *) base; 2356 if (dp->d_reclen == 0) 2357 break; 2358 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2359 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2360 else 2361 *(cookies++) = NFS_GETCOOKIE(dp); 2362 base += dp->d_reclen; 2363 } 2364 uio->uio_resid += 2365 ((caddr_t)uio->uio_iov->iov_base - base); 2366 uio->uio_iov->iov_len += 2367 ((caddr_t)uio->uio_iov->iov_base - base); 2368 uio->uio_iov->iov_base = base; 2369 *ap->a_ncookies = nc; 2370 } 2371 2372 uio->uio_resid += lost; 2373 *ap->a_eofflag = 0; 2374 return (error); 2375 } 2376 2377 /* 2378 * Readdir rpc call. 2379 * Called from below the buffer cache by nfs_doio(). 2380 */ 2381 int 2382 nfs_readdirrpc(vp, uiop, cred) 2383 struct vnode *vp; 2384 struct uio *uiop; 2385 struct ucred *cred; 2386 { 2387 int len, left; 2388 struct dirent *dp = NULL; 2389 u_int32_t *tl; 2390 caddr_t cp; 2391 int32_t t1, t2; 2392 caddr_t bpos, dpos, cp2; 2393 struct mbuf *mreq, *mrep, *md, *mb; 2394 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2395 struct nfsnode *dnp = VTONFS(vp); 2396 u_quad_t fileno; 2397 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2398 int attrflag, nrpcs = 0, reclen; 2399 const int v3 = NFS_ISV3(vp); 2400 2401 #ifdef DIAGNOSTIC 2402 /* 2403 * Should be called from buffer cache, so only amount of 2404 * NFS_DIRBLKSIZ will be requested. 2405 */ 2406 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2407 panic("nfs readdirrpc bad uio"); 2408 #endif 2409 2410 /* 2411 * Loop around doing readdir rpc's of size nm_readdirsize 2412 * truncated to a multiple of NFS_DIRFRAGSIZ. 2413 * The stopping criteria is EOF or buffer full. 2414 */ 2415 while (more_dirs && bigenough) { 2416 /* 2417 * Heuristic: don't bother to do another RPC to further 2418 * fill up this block if there is not much room left. (< 50% 2419 * of the readdir RPC size). This wastes some buffer space 2420 * but can save up to 50% in RPC calls. 2421 */ 2422 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2423 bigenough = 0; 2424 break; 2425 } 2426 nfsstats.rpccnt[NFSPROC_READDIR]++; 2427 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2428 NFSX_READDIR(v3)); 2429 nfsm_fhtom(dnp, v3); 2430 if (v3) { 2431 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2432 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2433 txdr_swapcookie3(uiop->uio_offset, tl); 2434 } else { 2435 txdr_cookie3(uiop->uio_offset, tl); 2436 } 2437 tl += 2; 2438 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2439 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2440 } else { 2441 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2442 *tl++ = txdr_unsigned(uiop->uio_offset); 2443 } 2444 *tl = txdr_unsigned(nmp->nm_readdirsize); 2445 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred); 2446 nrpcs++; 2447 if (v3) { 2448 nfsm_postop_attr(vp, attrflag, 0); 2449 if (!error) { 2450 nfsm_dissect(tl, u_int32_t *, 2451 2 * NFSX_UNSIGNED); 2452 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2453 dnp->n_cookieverf.nfsuquad[1] = *tl; 2454 } else { 2455 m_freem(mrep); 2456 goto nfsmout; 2457 } 2458 } 2459 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2460 more_dirs = fxdr_unsigned(int, *tl); 2461 2462 /* loop thru the dir entries, doctoring them to 4bsd form */ 2463 while (more_dirs && bigenough) { 2464 if (v3) { 2465 nfsm_dissect(tl, u_int32_t *, 2466 3 * NFSX_UNSIGNED); 2467 fileno = fxdr_hyper(tl); 2468 len = fxdr_unsigned(int, *(tl + 2)); 2469 } else { 2470 nfsm_dissect(tl, u_int32_t *, 2471 2 * NFSX_UNSIGNED); 2472 fileno = fxdr_unsigned(u_quad_t, *tl++); 2473 len = fxdr_unsigned(int, *tl); 2474 } 2475 if (len <= 0 || len > NFS_MAXNAMLEN) { 2476 error = EBADRPC; 2477 m_freem(mrep); 2478 goto nfsmout; 2479 } 2480 tlen = nfsm_rndup(len); 2481 if (tlen == len) 2482 tlen += 4; /* To ensure null termination */ 2483 tlen += sizeof (off_t) + sizeof (int); 2484 reclen = ALIGN(tlen + DIRHDSIZ); 2485 tlen = reclen - DIRHDSIZ; 2486 left = NFS_DIRFRAGSIZ - blksiz; 2487 if (reclen > left) { 2488 dp->d_reclen += left; 2489 uiop->uio_iov->iov_base = 2490 (caddr_t)uiop->uio_iov->iov_base + left; 2491 uiop->uio_iov->iov_len -= left; 2492 uiop->uio_resid -= left; 2493 blksiz = 0; 2494 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2495 } 2496 if (reclen > uiop->uio_resid) 2497 bigenough = 0; 2498 if (bigenough) { 2499 dp = (struct dirent *)uiop->uio_iov->iov_base; 2500 dp->d_fileno = (int)fileno; 2501 dp->d_namlen = len; 2502 dp->d_reclen = reclen; 2503 dp->d_type = DT_UNKNOWN; 2504 blksiz += dp->d_reclen; 2505 if (blksiz == NFS_DIRFRAGSIZ) 2506 blksiz = 0; 2507 uiop->uio_resid -= DIRHDSIZ; 2508 uiop->uio_iov->iov_base = 2509 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ; 2510 uiop->uio_iov->iov_len -= DIRHDSIZ; 2511 nfsm_mtouio(uiop, len); 2512 cp = uiop->uio_iov->iov_base; 2513 tlen -= len; 2514 *cp = '\0'; /* null terminate */ 2515 uiop->uio_iov->iov_base = 2516 (caddr_t)uiop->uio_iov->iov_base + tlen; 2517 uiop->uio_iov->iov_len -= tlen; 2518 uiop->uio_resid -= tlen; 2519 } else 2520 nfsm_adv(nfsm_rndup(len)); 2521 if (v3) { 2522 nfsm_dissect(tl, u_int32_t *, 2523 3 * NFSX_UNSIGNED); 2524 } else { 2525 nfsm_dissect(tl, u_int32_t *, 2526 2 * NFSX_UNSIGNED); 2527 } 2528 if (bigenough) { 2529 if (v3) { 2530 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2531 uiop->uio_offset = 2532 fxdr_swapcookie3(tl); 2533 else 2534 uiop->uio_offset = 2535 fxdr_cookie3(tl); 2536 } 2537 else { 2538 uiop->uio_offset = 2539 fxdr_unsigned(off_t, *tl); 2540 } 2541 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2542 } 2543 if (v3) 2544 tl += 2; 2545 else 2546 tl++; 2547 more_dirs = fxdr_unsigned(int, *tl); 2548 } 2549 /* 2550 * If at end of rpc data, get the eof boolean 2551 */ 2552 if (!more_dirs) { 2553 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2554 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2555 } 2556 m_freem(mrep); 2557 } 2558 /* 2559 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2560 * by increasing d_reclen for the last record. 2561 */ 2562 if (blksiz > 0) { 2563 left = NFS_DIRFRAGSIZ - blksiz; 2564 dp->d_reclen += left; 2565 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2566 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2567 left; 2568 uiop->uio_iov->iov_len -= left; 2569 uiop->uio_resid -= left; 2570 } 2571 2572 /* 2573 * We are now either at the end of the directory or have filled the 2574 * block. 2575 */ 2576 if (bigenough) 2577 dnp->n_direofoffset = uiop->uio_offset; 2578 nfsmout: 2579 return (error); 2580 } 2581 2582 /* 2583 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2584 */ 2585 int 2586 nfs_readdirplusrpc(vp, uiop, cred) 2587 struct vnode *vp; 2588 struct uio *uiop; 2589 struct ucred *cred; 2590 { 2591 int len, left; 2592 struct dirent *dp = NULL; 2593 u_int32_t *tl; 2594 caddr_t cp; 2595 int32_t t1, t2; 2596 struct vnode *newvp; 2597 caddr_t bpos, dpos, cp2; 2598 struct mbuf *mreq, *mrep, *md, *mb; 2599 struct nameidata nami, *ndp = &nami; 2600 struct componentname *cnp = &ndp->ni_cnd; 2601 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2602 struct nfsnode *dnp = VTONFS(vp), *np; 2603 nfsfh_t *fhp; 2604 u_quad_t fileno; 2605 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2606 int attrflag, fhsize, nrpcs = 0, reclen; 2607 struct nfs_fattr fattr, *fp; 2608 2609 #ifdef DIAGNOSTIC 2610 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2611 panic("nfs readdirplusrpc bad uio"); 2612 #endif 2613 ndp->ni_dvp = vp; 2614 newvp = NULLVP; 2615 2616 /* 2617 * Loop around doing readdir rpc's of size nm_readdirsize 2618 * truncated to a multiple of NFS_DIRFRAGSIZ. 2619 * The stopping criteria is EOF or buffer full. 2620 */ 2621 while (more_dirs && bigenough) { 2622 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2623 bigenough = 0; 2624 break; 2625 } 2626 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2627 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2628 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2629 nfsm_fhtom(dnp, 1); 2630 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2631 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2632 txdr_swapcookie3(uiop->uio_offset, tl); 2633 } else { 2634 txdr_cookie3(uiop->uio_offset, tl); 2635 } 2636 tl += 2; 2637 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2638 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2639 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2640 *tl = txdr_unsigned(nmp->nm_rsize); 2641 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred); 2642 nfsm_postop_attr(vp, attrflag, 0); 2643 if (error) { 2644 m_freem(mrep); 2645 goto nfsmout; 2646 } 2647 nrpcs++; 2648 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2649 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2650 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2651 more_dirs = fxdr_unsigned(int, *tl); 2652 2653 /* loop thru the dir entries, doctoring them to 4bsd form */ 2654 while (more_dirs && bigenough) { 2655 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2656 fileno = fxdr_hyper(tl); 2657 len = fxdr_unsigned(int, *(tl + 2)); 2658 if (len <= 0 || len > NFS_MAXNAMLEN) { 2659 error = EBADRPC; 2660 m_freem(mrep); 2661 goto nfsmout; 2662 } 2663 tlen = nfsm_rndup(len); 2664 if (tlen == len) 2665 tlen += 4; /* To ensure null termination*/ 2666 tlen += sizeof (off_t) + sizeof (int); 2667 reclen = ALIGN(tlen + DIRHDSIZ); 2668 tlen = reclen - DIRHDSIZ; 2669 left = NFS_DIRFRAGSIZ - blksiz; 2670 if (reclen > left) { 2671 /* 2672 * DIRFRAGSIZ is aligned, no need to align 2673 * again here. 2674 */ 2675 dp->d_reclen += left; 2676 uiop->uio_iov->iov_base = 2677 (caddr_t)uiop->uio_iov->iov_base + left; 2678 uiop->uio_iov->iov_len -= left; 2679 uiop->uio_resid -= left; 2680 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2681 blksiz = 0; 2682 } 2683 if (reclen > uiop->uio_resid) 2684 bigenough = 0; 2685 if (bigenough) { 2686 dp = (struct dirent *)uiop->uio_iov->iov_base; 2687 dp->d_fileno = (int)fileno; 2688 dp->d_namlen = len; 2689 dp->d_reclen = reclen; 2690 dp->d_type = DT_UNKNOWN; 2691 blksiz += dp->d_reclen; 2692 if (blksiz == NFS_DIRFRAGSIZ) 2693 blksiz = 0; 2694 uiop->uio_resid -= DIRHDSIZ; 2695 uiop->uio_iov->iov_base = 2696 (caddr_t)uiop->uio_iov->iov_base + 2697 DIRHDSIZ; 2698 uiop->uio_iov->iov_len -= DIRHDSIZ; 2699 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2700 cnp->cn_namelen = len; 2701 nfsm_mtouio(uiop, len); 2702 cp = uiop->uio_iov->iov_base; 2703 tlen -= len; 2704 *cp = '\0'; 2705 uiop->uio_iov->iov_base = 2706 (caddr_t)uiop->uio_iov->iov_base + tlen; 2707 uiop->uio_iov->iov_len -= tlen; 2708 uiop->uio_resid -= tlen; 2709 } else 2710 nfsm_adv(nfsm_rndup(len)); 2711 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2712 if (bigenough) { 2713 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2714 uiop->uio_offset = 2715 fxdr_swapcookie3(tl); 2716 else 2717 uiop->uio_offset = 2718 fxdr_cookie3(tl); 2719 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2720 } 2721 tl += 2; 2722 2723 /* 2724 * Since the attributes are before the file handle 2725 * (sigh), we must skip over the attributes and then 2726 * come back and get them. 2727 */ 2728 attrflag = fxdr_unsigned(int, *tl); 2729 if (attrflag) { 2730 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2731 memcpy(&fattr, fp, NFSX_V3FATTR); 2732 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2733 doit = fxdr_unsigned(int, *tl); 2734 if (doit) { 2735 nfsm_getfh(fhp, fhsize, 1); 2736 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2737 VREF(vp); 2738 newvp = vp; 2739 np = dnp; 2740 } else { 2741 error = nfs_nget1(vp->v_mount, fhp, 2742 fhsize, &np, LK_NOWAIT); 2743 if (!error) 2744 newvp = NFSTOV(np); 2745 } 2746 if (!error) { 2747 const char *cp; 2748 2749 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2750 dp->d_type = 2751 IFTODT(VTTOIF(np->n_vattr->va_type)); 2752 ndp->ni_vp = newvp; 2753 cp = cnp->cn_nameptr + cnp->cn_namelen; 2754 cnp->cn_hash = 2755 namei_hash(cnp->cn_nameptr, &cp); 2756 if (cnp->cn_namelen <= NCHNAMLEN) { 2757 cache_purge1(ndp->ni_dvp, cnp, 0); 2758 nfs_cache_enter(ndp->ni_dvp, ndp->ni_vp, 2759 cnp); 2760 } 2761 } 2762 } 2763 } else { 2764 /* Just skip over the file handle */ 2765 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2766 i = fxdr_unsigned(int, *tl); 2767 nfsm_adv(nfsm_rndup(i)); 2768 } 2769 if (newvp != NULLVP) { 2770 if (newvp == vp) 2771 vrele(newvp); 2772 else 2773 vput(newvp); 2774 newvp = NULLVP; 2775 } 2776 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2777 more_dirs = fxdr_unsigned(int, *tl); 2778 } 2779 /* 2780 * If at end of rpc data, get the eof boolean 2781 */ 2782 if (!more_dirs) { 2783 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2784 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2785 } 2786 m_freem(mrep); 2787 } 2788 /* 2789 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2790 * by increasing d_reclen for the last record. 2791 */ 2792 if (blksiz > 0) { 2793 left = NFS_DIRFRAGSIZ - blksiz; 2794 dp->d_reclen += left; 2795 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2796 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2797 left; 2798 uiop->uio_iov->iov_len -= left; 2799 uiop->uio_resid -= left; 2800 } 2801 2802 /* 2803 * We are now either at the end of the directory or have filled the 2804 * block. 2805 */ 2806 if (bigenough) 2807 dnp->n_direofoffset = uiop->uio_offset; 2808 nfsmout: 2809 if (newvp != NULLVP) { 2810 if(newvp == vp) 2811 vrele(newvp); 2812 else 2813 vput(newvp); 2814 } 2815 return (error); 2816 } 2817 static char hextoasc[] = "0123456789abcdef"; 2818 2819 /* 2820 * Silly rename. To make the NFS filesystem that is stateless look a little 2821 * more like the "ufs" a remove of an active vnode is translated to a rename 2822 * to a funny looking filename that is removed by nfs_inactive on the 2823 * nfsnode. There is the potential for another process on a different client 2824 * to create the same funny name between the nfs_lookitup() fails and the 2825 * nfs_rename() completes, but... 2826 */ 2827 int 2828 nfs_sillyrename(dvp, vp, cnp) 2829 struct vnode *dvp, *vp; 2830 struct componentname *cnp; 2831 { 2832 struct sillyrename *sp; 2833 struct nfsnode *np; 2834 int error; 2835 short pid; 2836 2837 cache_purge(dvp); 2838 np = VTONFS(vp); 2839 #ifndef DIAGNOSTIC 2840 if (vp->v_type == VDIR) 2841 panic("nfs: sillyrename dir"); 2842 #endif 2843 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2844 M_NFSREQ, M_WAITOK); 2845 sp->s_cred = crdup(cnp->cn_cred); 2846 sp->s_dvp = dvp; 2847 VREF(dvp); 2848 2849 /* Fudge together a funny name */ 2850 pid = cnp->cn_proc->p_pid; 2851 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2852 sp->s_namlen = 12; 2853 sp->s_name[8] = hextoasc[pid & 0xf]; 2854 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf]; 2855 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf]; 2856 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf]; 2857 2858 /* Try lookitups until we get one that isn't there */ 2859 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2860 cnp->cn_proc, (struct nfsnode **)0) == 0) { 2861 sp->s_name[4]++; 2862 if (sp->s_name[4] > 'z') { 2863 error = EINVAL; 2864 goto bad; 2865 } 2866 } 2867 error = nfs_renameit(dvp, cnp, sp); 2868 if (error) 2869 goto bad; 2870 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2871 cnp->cn_proc, &np); 2872 np->n_sillyrename = sp; 2873 return (0); 2874 bad: 2875 vrele(sp->s_dvp); 2876 crfree(sp->s_cred); 2877 free((caddr_t)sp, M_NFSREQ); 2878 return (error); 2879 } 2880 2881 /* 2882 * Look up a file name and optionally either update the file handle or 2883 * allocate an nfsnode, depending on the value of npp. 2884 * npp == NULL --> just do the lookup 2885 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2886 * handled too 2887 * *npp != NULL --> update the file handle in the vnode 2888 */ 2889 int 2890 nfs_lookitup(dvp, name, len, cred, procp, npp) 2891 struct vnode *dvp; 2892 const char *name; 2893 int len; 2894 struct ucred *cred; 2895 struct proc *procp; 2896 struct nfsnode **npp; 2897 { 2898 u_int32_t *tl; 2899 caddr_t cp; 2900 int32_t t1, t2; 2901 struct vnode *newvp = (struct vnode *)0; 2902 struct nfsnode *np, *dnp = VTONFS(dvp); 2903 caddr_t bpos, dpos, cp2; 2904 int error = 0, fhlen, attrflag; 2905 struct mbuf *mreq, *mrep, *md, *mb; 2906 nfsfh_t *nfhp; 2907 const int v3 = NFS_ISV3(dvp); 2908 2909 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2910 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2911 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2912 nfsm_fhtom(dnp, v3); 2913 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2914 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred); 2915 if (npp && !error) { 2916 nfsm_getfh(nfhp, fhlen, v3); 2917 if (*npp) { 2918 np = *npp; 2919 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2920 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2921 np->n_fhp = &np->n_fh; 2922 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2923 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2924 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen); 2925 np->n_fhsize = fhlen; 2926 newvp = NFSTOV(np); 2927 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2928 VREF(dvp); 2929 newvp = dvp; 2930 } else { 2931 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2932 if (error) { 2933 m_freem(mrep); 2934 return (error); 2935 } 2936 newvp = NFSTOV(np); 2937 } 2938 if (v3) { 2939 nfsm_postop_attr(newvp, attrflag, 0); 2940 if (!attrflag && *npp == NULL) { 2941 m_freem(mrep); 2942 vput(newvp); 2943 return (ENOENT); 2944 } 2945 } else 2946 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2947 } 2948 nfsm_reqdone; 2949 if (npp && *npp == NULL) { 2950 if (error) { 2951 if (newvp) 2952 vput(newvp); 2953 } else 2954 *npp = np; 2955 } 2956 return (error); 2957 } 2958 2959 /* 2960 * Nfs Version 3 commit rpc 2961 */ 2962 int 2963 nfs_commit(vp, offset, cnt, procp) 2964 struct vnode *vp; 2965 off_t offset; 2966 uint32_t cnt; 2967 struct proc *procp; 2968 { 2969 caddr_t cp; 2970 u_int32_t *tl; 2971 int32_t t1, t2; 2972 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2973 caddr_t bpos, dpos, cp2; 2974 int error = 0, wccflag = NFSV3_WCCRATTR; 2975 struct mbuf *mreq, *mrep, *md, *mb; 2976 struct nfsnode *np; 2977 2978 KASSERT(NFS_ISV3(vp)); 2979 2980 #ifdef NFS_DEBUG_COMMIT 2981 printf("commit %lu - %lu\n", (unsigned long)offset, 2982 (unsigned long)(offset + cnt)); 2983 #endif 2984 2985 simple_lock(&nmp->nm_slock); 2986 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 2987 simple_unlock(&nmp->nm_slock); 2988 return (0); 2989 } 2990 simple_unlock(&nmp->nm_slock); 2991 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2992 np = VTONFS(vp); 2993 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 2994 nfsm_fhtom(np, 1); 2995 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2996 txdr_hyper(offset, tl); 2997 tl += 2; 2998 *tl = txdr_unsigned(cnt); 2999 nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred); 3000 nfsm_wcc_data(vp, wccflag, 0); 3001 if (!error) { 3002 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3003 simple_lock(&nmp->nm_slock); 3004 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3005 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3006 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3007 error = NFSERR_STALEWRITEVERF; 3008 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3009 } 3010 simple_unlock(&nmp->nm_slock); 3011 } 3012 nfsm_reqdone; 3013 return (error); 3014 } 3015 3016 /* 3017 * Kludge City.. 3018 * - make nfs_bmap() essentially a no-op that does no translation 3019 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3020 * (Maybe I could use the process's page mapping, but I was concerned that 3021 * Kernel Write might not be enabled and also figured copyout() would do 3022 * a lot more work than memcpy() and also it currently happens in the 3023 * context of the swapper process (2). 3024 */ 3025 int 3026 nfs_bmap(v) 3027 void *v; 3028 { 3029 struct vop_bmap_args /* { 3030 struct vnode *a_vp; 3031 daddr_t a_bn; 3032 struct vnode **a_vpp; 3033 daddr_t *a_bnp; 3034 int *a_runp; 3035 } */ *ap = v; 3036 struct vnode *vp = ap->a_vp; 3037 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3038 3039 if (ap->a_vpp != NULL) 3040 *ap->a_vpp = vp; 3041 if (ap->a_bnp != NULL) 3042 *ap->a_bnp = ap->a_bn << bshift; 3043 if (ap->a_runp != NULL) 3044 *ap->a_runp = 1024 * 1024; /* XXX */ 3045 return (0); 3046 } 3047 3048 /* 3049 * Strategy routine. 3050 * For async requests when nfsiod(s) are running, queue the request by 3051 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3052 * request. 3053 */ 3054 int 3055 nfs_strategy(v) 3056 void *v; 3057 { 3058 struct vop_strategy_args *ap = v; 3059 struct buf *bp = ap->a_bp; 3060 struct proc *p; 3061 int error = 0; 3062 3063 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3064 panic("nfs physio/async"); 3065 if (bp->b_flags & B_ASYNC) 3066 p = NULL; 3067 else 3068 p = curproc; /* XXX */ 3069 3070 /* 3071 * If the op is asynchronous and an i/o daemon is waiting 3072 * queue the request, wake it up and wait for completion 3073 * otherwise just do it ourselves. 3074 */ 3075 3076 if ((bp->b_flags & B_ASYNC) == 0 || 3077 nfs_asyncio(bp)) 3078 error = nfs_doio(bp, p); 3079 return (error); 3080 } 3081 3082 /* 3083 * fsync vnode op. Just call nfs_flush() with commit == 1. 3084 */ 3085 /* ARGSUSED */ 3086 int 3087 nfs_fsync(v) 3088 void *v; 3089 { 3090 struct vop_fsync_args /* { 3091 struct vnodeop_desc *a_desc; 3092 struct vnode * a_vp; 3093 struct ucred * a_cred; 3094 int a_flags; 3095 off_t offlo; 3096 off_t offhi; 3097 struct proc * a_p; 3098 } */ *ap = v; 3099 3100 struct vnode *vp = ap->a_vp; 3101 3102 if (vp->v_type != VREG) 3103 return 0; 3104 3105 return (nfs_flush(vp, ap->a_cred, 3106 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1)); 3107 } 3108 3109 /* 3110 * Flush all the data associated with a vnode. 3111 */ 3112 int 3113 nfs_flush(vp, cred, waitfor, p, commit) 3114 struct vnode *vp; 3115 struct ucred *cred; 3116 int waitfor; 3117 struct proc *p; 3118 int commit; 3119 { 3120 struct nfsnode *np = VTONFS(vp); 3121 int error; 3122 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3123 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3124 3125 simple_lock(&vp->v_interlock); 3126 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3127 if (np->n_flag & NWRITEERR) { 3128 error = np->n_error; 3129 np->n_flag &= ~NWRITEERR; 3130 } 3131 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3132 return (error); 3133 } 3134 3135 /* 3136 * Return POSIX pathconf information applicable to nfs. 3137 * 3138 * N.B. The NFS V2 protocol doesn't support this RPC. 3139 */ 3140 /* ARGSUSED */ 3141 int 3142 nfs_pathconf(v) 3143 void *v; 3144 { 3145 struct vop_pathconf_args /* { 3146 struct vnode *a_vp; 3147 int a_name; 3148 register_t *a_retval; 3149 } */ *ap = v; 3150 struct nfsv3_pathconf *pcp; 3151 struct vnode *vp = ap->a_vp; 3152 struct mbuf *mreq, *mrep, *md, *mb; 3153 int32_t t1, t2; 3154 u_int32_t *tl; 3155 caddr_t bpos, dpos, cp, cp2; 3156 int error = 0, attrflag; 3157 #ifndef NFS_V2_ONLY 3158 struct nfsmount *nmp; 3159 unsigned int l; 3160 u_int64_t maxsize; 3161 #endif 3162 const int v3 = NFS_ISV3(vp); 3163 struct nfsnode *np = VTONFS(vp); 3164 3165 switch (ap->a_name) { 3166 /* Names that can be resolved locally. */ 3167 case _PC_PIPE_BUF: 3168 *ap->a_retval = PIPE_BUF; 3169 break; 3170 case _PC_SYNC_IO: 3171 *ap->a_retval = 1; 3172 break; 3173 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3174 case _PC_LINK_MAX: 3175 case _PC_NAME_MAX: 3176 case _PC_CHOWN_RESTRICTED: 3177 case _PC_NO_TRUNC: 3178 if (!v3) { 3179 error = EINVAL; 3180 break; 3181 } 3182 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3183 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3184 nfsm_fhtom(np, 1); 3185 nfsm_request(np, NFSPROC_PATHCONF, 3186 curproc, curproc->p_ucred); /* XXX */ 3187 nfsm_postop_attr(vp, attrflag, 0); 3188 if (!error) { 3189 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3190 NFSX_V3PATHCONF); 3191 switch (ap->a_name) { 3192 case _PC_LINK_MAX: 3193 *ap->a_retval = 3194 fxdr_unsigned(register_t, pcp->pc_linkmax); 3195 break; 3196 case _PC_NAME_MAX: 3197 *ap->a_retval = 3198 fxdr_unsigned(register_t, pcp->pc_namemax); 3199 break; 3200 case _PC_CHOWN_RESTRICTED: 3201 *ap->a_retval = 3202 (pcp->pc_chownrestricted == nfs_true); 3203 break; 3204 case _PC_NO_TRUNC: 3205 *ap->a_retval = 3206 (pcp->pc_notrunc == nfs_true); 3207 break; 3208 } 3209 } 3210 nfsm_reqdone; 3211 break; 3212 case _PC_FILESIZEBITS: 3213 #ifndef NFS_V2_ONLY 3214 if (v3) { 3215 nmp = VFSTONFS(vp->v_mount); 3216 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3217 if ((error = nfs_fsinfo(nmp, vp, 3218 curproc->p_ucred, curproc)) != 0) /* XXX */ 3219 break; 3220 for (l = 0, maxsize = nmp->nm_maxfilesize; 3221 (maxsize >> l) > 0; l++) 3222 ; 3223 *ap->a_retval = l + 1; 3224 } else 3225 #endif 3226 { 3227 *ap->a_retval = 32; /* NFS V2 limitation */ 3228 } 3229 break; 3230 default: 3231 error = EINVAL; 3232 break; 3233 } 3234 3235 return (error); 3236 } 3237 3238 /* 3239 * NFS advisory byte-level locks. 3240 */ 3241 int 3242 nfs_advlock(v) 3243 void *v; 3244 { 3245 struct vop_advlock_args /* { 3246 struct vnode *a_vp; 3247 caddr_t a_id; 3248 int a_op; 3249 struct flock *a_fl; 3250 int a_flags; 3251 } */ *ap = v; 3252 struct nfsnode *np = VTONFS(ap->a_vp); 3253 3254 return lf_advlock(ap, &np->n_lockf, np->n_size); 3255 } 3256 3257 /* 3258 * Print out the contents of an nfsnode. 3259 */ 3260 int 3261 nfs_print(v) 3262 void *v; 3263 { 3264 struct vop_print_args /* { 3265 struct vnode *a_vp; 3266 } */ *ap = v; 3267 struct vnode *vp = ap->a_vp; 3268 struct nfsnode *np = VTONFS(vp); 3269 3270 printf("tag VT_NFS, fileid %ld fsid 0x%lx", 3271 np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3272 if (vp->v_type == VFIFO) 3273 fifo_printinfo(vp); 3274 printf("\n"); 3275 return (0); 3276 } 3277 3278 /* 3279 * NFS file truncation. 3280 */ 3281 int 3282 nfs_truncate(v) 3283 void *v; 3284 { 3285 #if 0 3286 struct vop_truncate_args /* { 3287 struct vnode *a_vp; 3288 off_t a_length; 3289 int a_flags; 3290 struct ucred *a_cred; 3291 struct proc *a_p; 3292 } */ *ap = v; 3293 #endif 3294 3295 /* Use nfs_setattr */ 3296 return (EOPNOTSUPP); 3297 } 3298 3299 /* 3300 * NFS update. 3301 */ 3302 int 3303 nfs_update(v) 3304 void *v; 3305 #if 0 3306 struct vop_update_args /* { 3307 struct vnode *a_vp; 3308 struct timespec *a_ta; 3309 struct timespec *a_tm; 3310 int a_waitfor; 3311 } */ *ap = v; 3312 #endif 3313 { 3314 3315 /* Use nfs_setattr */ 3316 return (EOPNOTSUPP); 3317 } 3318 3319 /* 3320 * Just call bwrite(). 3321 */ 3322 int 3323 nfs_bwrite(v) 3324 void *v; 3325 { 3326 struct vop_bwrite_args /* { 3327 struct vnode *a_bp; 3328 } */ *ap = v; 3329 3330 return (bwrite(ap->a_bp)); 3331 } 3332 3333 /* 3334 * nfs unlock wrapper. 3335 */ 3336 int 3337 nfs_unlock(void *v) 3338 { 3339 struct vop_unlock_args /* { 3340 struct vnode *a_vp; 3341 int a_flags; 3342 } */ *ap = v; 3343 struct vnode *vp = ap->a_vp; 3344 3345 /* 3346 * VOP_UNLOCK can be called by nfs_loadattrcache 3347 * with v_data == 0. 3348 */ 3349 if (VTONFS(vp)) { 3350 nfs_delayedtruncate(vp); 3351 } 3352 3353 return genfs_unlock(v); 3354 } 3355 3356 /* 3357 * nfs special file access vnode op. 3358 * Essentially just get vattr and then imitate iaccess() since the device is 3359 * local to the client. 3360 */ 3361 int 3362 nfsspec_access(v) 3363 void *v; 3364 { 3365 struct vop_access_args /* { 3366 struct vnode *a_vp; 3367 int a_mode; 3368 struct ucred *a_cred; 3369 struct proc *a_p; 3370 } */ *ap = v; 3371 struct vattr va; 3372 struct vnode *vp = ap->a_vp; 3373 int error; 3374 3375 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p); 3376 if (error) 3377 return (error); 3378 3379 /* 3380 * Disallow write attempts on filesystems mounted read-only; 3381 * unless the file is a socket, fifo, or a block or character 3382 * device resident on the filesystem. 3383 */ 3384 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3385 switch (vp->v_type) { 3386 case VREG: 3387 case VDIR: 3388 case VLNK: 3389 return (EROFS); 3390 default: 3391 break; 3392 } 3393 } 3394 3395 return (vaccess(va.va_type, va.va_mode, 3396 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3397 } 3398 3399 /* 3400 * Read wrapper for special devices. 3401 */ 3402 int 3403 nfsspec_read(v) 3404 void *v; 3405 { 3406 struct vop_read_args /* { 3407 struct vnode *a_vp; 3408 struct uio *a_uio; 3409 int a_ioflag; 3410 struct ucred *a_cred; 3411 } */ *ap = v; 3412 struct nfsnode *np = VTONFS(ap->a_vp); 3413 3414 /* 3415 * Set access flag. 3416 */ 3417 np->n_flag |= NACC; 3418 np->n_atim.tv_sec = time.tv_sec; 3419 np->n_atim.tv_nsec = time.tv_usec * 1000; 3420 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3421 } 3422 3423 /* 3424 * Write wrapper for special devices. 3425 */ 3426 int 3427 nfsspec_write(v) 3428 void *v; 3429 { 3430 struct vop_write_args /* { 3431 struct vnode *a_vp; 3432 struct uio *a_uio; 3433 int a_ioflag; 3434 struct ucred *a_cred; 3435 } */ *ap = v; 3436 struct nfsnode *np = VTONFS(ap->a_vp); 3437 3438 /* 3439 * Set update flag. 3440 */ 3441 np->n_flag |= NUPD; 3442 np->n_mtim.tv_sec = time.tv_sec; 3443 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3444 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3445 } 3446 3447 /* 3448 * Close wrapper for special devices. 3449 * 3450 * Update the times on the nfsnode then do device close. 3451 */ 3452 int 3453 nfsspec_close(v) 3454 void *v; 3455 { 3456 struct vop_close_args /* { 3457 struct vnode *a_vp; 3458 int a_fflag; 3459 struct ucred *a_cred; 3460 struct proc *a_p; 3461 } */ *ap = v; 3462 struct vnode *vp = ap->a_vp; 3463 struct nfsnode *np = VTONFS(vp); 3464 struct vattr vattr; 3465 3466 if (np->n_flag & (NACC | NUPD)) { 3467 np->n_flag |= NCHG; 3468 if (vp->v_usecount == 1 && 3469 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3470 VATTR_NULL(&vattr); 3471 if (np->n_flag & NACC) 3472 vattr.va_atime = np->n_atim; 3473 if (np->n_flag & NUPD) 3474 vattr.va_mtime = np->n_mtim; 3475 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3476 } 3477 } 3478 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3479 } 3480 3481 /* 3482 * Read wrapper for fifos. 3483 */ 3484 int 3485 nfsfifo_read(v) 3486 void *v; 3487 { 3488 struct vop_read_args /* { 3489 struct vnode *a_vp; 3490 struct uio *a_uio; 3491 int a_ioflag; 3492 struct ucred *a_cred; 3493 } */ *ap = v; 3494 struct nfsnode *np = VTONFS(ap->a_vp); 3495 3496 /* 3497 * Set access flag. 3498 */ 3499 np->n_flag |= NACC; 3500 np->n_atim.tv_sec = time.tv_sec; 3501 np->n_atim.tv_nsec = time.tv_usec * 1000; 3502 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3503 } 3504 3505 /* 3506 * Write wrapper for fifos. 3507 */ 3508 int 3509 nfsfifo_write(v) 3510 void *v; 3511 { 3512 struct vop_write_args /* { 3513 struct vnode *a_vp; 3514 struct uio *a_uio; 3515 int a_ioflag; 3516 struct ucred *a_cred; 3517 } */ *ap = v; 3518 struct nfsnode *np = VTONFS(ap->a_vp); 3519 3520 /* 3521 * Set update flag. 3522 */ 3523 np->n_flag |= NUPD; 3524 np->n_mtim.tv_sec = time.tv_sec; 3525 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3526 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3527 } 3528 3529 /* 3530 * Close wrapper for fifos. 3531 * 3532 * Update the times on the nfsnode then do fifo close. 3533 */ 3534 int 3535 nfsfifo_close(v) 3536 void *v; 3537 { 3538 struct vop_close_args /* { 3539 struct vnode *a_vp; 3540 int a_fflag; 3541 struct ucred *a_cred; 3542 struct proc *a_p; 3543 } */ *ap = v; 3544 struct vnode *vp = ap->a_vp; 3545 struct nfsnode *np = VTONFS(vp); 3546 struct vattr vattr; 3547 3548 if (np->n_flag & (NACC | NUPD)) { 3549 if (np->n_flag & NACC) { 3550 np->n_atim.tv_sec = time.tv_sec; 3551 np->n_atim.tv_nsec = time.tv_usec * 1000; 3552 } 3553 if (np->n_flag & NUPD) { 3554 np->n_mtim.tv_sec = time.tv_sec; 3555 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3556 } 3557 np->n_flag |= NCHG; 3558 if (vp->v_usecount == 1 && 3559 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3560 VATTR_NULL(&vattr); 3561 if (np->n_flag & NACC) 3562 vattr.va_atime = np->n_atim; 3563 if (np->n_flag & NUPD) 3564 vattr.va_mtime = np->n_mtim; 3565 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3566 } 3567 } 3568 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3569 } 3570
Indexes created Wed Oct 15 16:09:53 GMT 2025