nfs_vnops.c revision 1.194
1 /* $NetBSD: nfs_vnops.c,v 1.194 2004/05/08 21:32:34 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 1.194 2004/05/08 21:32:34 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 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 930 KASSERT(lockparent); 931 cnp->cn_flags |= SAVENAME; 932 return (EJUSTRETURN); 933 } 934 935 error = 0; 936 newvp = NULLVP; 937 nfsstats.lookupcache_misses++; 938 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 939 len = cnp->cn_namelen; 940 nfsm_reqhead(np, NFSPROC_LOOKUP, 941 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 942 nfsm_fhtom(np, v3); 943 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 944 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred); 945 if (error) { 946 nfsm_postop_attr(dvp, attrflag, 0); 947 m_freem(mrep); 948 goto nfsmout; 949 } 950 nfsm_getfh(fhp, fhsize, v3); 951 952 /* 953 * Handle RENAME case... 954 */ 955 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) { 956 if (NFS_CMPFH(np, fhp, fhsize)) { 957 m_freem(mrep); 958 return (EISDIR); 959 } 960 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 961 if (error) { 962 m_freem(mrep); 963 return error; 964 } 965 newvp = NFSTOV(np); 966 if (v3) { 967 nfsm_postop_attr(newvp, attrflag, 0); 968 nfsm_postop_attr(dvp, attrflag, 0); 969 } else 970 nfsm_loadattr(newvp, (struct vattr *)0, 0); 971 *vpp = newvp; 972 m_freem(mrep); 973 cnp->cn_flags |= SAVENAME; 974 if (!lockparent) { 975 VOP_UNLOCK(dvp, 0); 976 cnp->cn_flags |= PDIRUNLOCK; 977 } 978 goto validate; 979 } 980 981 /* 982 * The postop attr handling is duplicated for each if case, 983 * because it should be done while dvp is locked (unlocking 984 * dvp is different for each case). 985 */ 986 987 if (NFS_CMPFH(np, fhp, fhsize)) { 988 /* 989 * "." lookup 990 */ 991 VREF(dvp); 992 newvp = dvp; 993 if (v3) { 994 nfsm_postop_attr(newvp, attrflag, 0); 995 nfsm_postop_attr(dvp, attrflag, 0); 996 } else 997 nfsm_loadattr(newvp, (struct vattr *)0, 0); 998 } else if (flags & ISDOTDOT) { 999 /* 1000 * ".." lookup 1001 */ 1002 VOP_UNLOCK(dvp, 0); 1003 cnp->cn_flags |= PDIRUNLOCK; 1004 1005 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1006 if (error) { 1007 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 1008 cnp->cn_flags &= ~PDIRUNLOCK; 1009 m_freem(mrep); 1010 return error; 1011 } 1012 newvp = NFSTOV(np); 1013 1014 if (v3) { 1015 nfsm_postop_attr(newvp, attrflag, 0); 1016 nfsm_postop_attr(dvp, attrflag, 0); 1017 } else 1018 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1019 1020 if (lockparent && (flags & ISLASTCN)) { 1021 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) { 1022 m_freem(mrep); 1023 vput(newvp); 1024 return error; 1025 } 1026 cnp->cn_flags &= ~PDIRUNLOCK; 1027 } 1028 } else { 1029 /* 1030 * Other lookups. 1031 */ 1032 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1033 if (error) { 1034 m_freem(mrep); 1035 return error; 1036 } 1037 newvp = NFSTOV(np); 1038 if (v3) { 1039 nfsm_postop_attr(newvp, attrflag, 0); 1040 nfsm_postop_attr(dvp, attrflag, 0); 1041 } else 1042 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1043 if (!lockparent || !(flags & ISLASTCN)) { 1044 VOP_UNLOCK(dvp, 0); 1045 cnp->cn_flags |= PDIRUNLOCK; 1046 } 1047 } 1048 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1049 cnp->cn_flags |= SAVENAME; 1050 if ((cnp->cn_flags & MAKEENTRY) && 1051 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1052 nfs_cache_enter(dvp, newvp, cnp); 1053 } 1054 *vpp = newvp; 1055 nfsm_reqdone; 1056 if (error) { 1057 /* 1058 * We get here only because of errors returned by 1059 * the RPC. Otherwise we'll have returned above 1060 * (the nfsm_* macros will jump to nfsm_reqdone 1061 * on error). 1062 */ 1063 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1064 cnp->cn_nameiop != CREATE) { 1065 nfs_cache_enter(dvp, NULL, cnp); 1066 } 1067 if (newvp != NULLVP) { 1068 vrele(newvp); 1069 if (newvp != dvp) 1070 VOP_UNLOCK(newvp, 0); 1071 } 1072 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1073 (flags & ISLASTCN) && error == ENOENT) { 1074 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1075 error = EROFS; 1076 else 1077 error = EJUSTRETURN; 1078 } 1079 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1080 cnp->cn_flags |= SAVENAME; 1081 *vpp = NULL; 1082 return error; 1083 } 1084 1085 validate: 1086 /* 1087 * make sure we have valid type and size. 1088 */ 1089 1090 newvp = *vpp; 1091 if (newvp->v_type == VNON) { 1092 struct vattr vattr; /* dummy */ 1093 1094 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1095 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc); 1096 if (error) { 1097 vput(newvp); 1098 *vpp = NULL; 1099 } 1100 } 1101 1102 return error; 1103 } 1104 1105 /* 1106 * nfs read call. 1107 * Just call nfs_bioread() to do the work. 1108 */ 1109 int 1110 nfs_read(v) 1111 void *v; 1112 { 1113 struct vop_read_args /* { 1114 struct vnode *a_vp; 1115 struct uio *a_uio; 1116 int a_ioflag; 1117 struct ucred *a_cred; 1118 } */ *ap = v; 1119 struct vnode *vp = ap->a_vp; 1120 1121 if (vp->v_type != VREG) 1122 return (EPERM); 1123 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1124 } 1125 1126 /* 1127 * nfs readlink call 1128 */ 1129 int 1130 nfs_readlink(v) 1131 void *v; 1132 { 1133 struct vop_readlink_args /* { 1134 struct vnode *a_vp; 1135 struct uio *a_uio; 1136 struct ucred *a_cred; 1137 } */ *ap = v; 1138 struct vnode *vp = ap->a_vp; 1139 1140 if (vp->v_type != VLNK) 1141 return (EPERM); 1142 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1143 } 1144 1145 /* 1146 * Do a readlink rpc. 1147 * Called by nfs_doio() from below the buffer cache. 1148 */ 1149 int 1150 nfs_readlinkrpc(vp, uiop, cred) 1151 struct vnode *vp; 1152 struct uio *uiop; 1153 struct ucred *cred; 1154 { 1155 u_int32_t *tl; 1156 caddr_t cp; 1157 int32_t t1, t2; 1158 caddr_t bpos, dpos, cp2; 1159 int error = 0, attrflag; 1160 uint32_t len; 1161 struct mbuf *mreq, *mrep, *md, *mb; 1162 const int v3 = NFS_ISV3(vp); 1163 struct nfsnode *np = VTONFS(vp); 1164 1165 nfsstats.rpccnt[NFSPROC_READLINK]++; 1166 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1167 nfsm_fhtom(np, v3); 1168 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred); 1169 if (v3) 1170 nfsm_postop_attr(vp, attrflag, 0); 1171 if (!error) { 1172 if (v3) { 1173 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1174 len = fxdr_unsigned(uint32_t, *tl); 1175 if (len > MAXPATHLEN) { 1176 /* 1177 * this pathname is too long for us. 1178 */ 1179 m_freem(mrep); 1180 /* Solaris returns EINVAL. should we follow? */ 1181 error = ENAMETOOLONG; 1182 goto nfsmout; 1183 } 1184 } else { 1185 nfsm_strsiz(len, NFS_MAXPATHLEN); 1186 } 1187 nfsm_mtouio(uiop, len); 1188 } 1189 nfsm_reqdone; 1190 return (error); 1191 } 1192 1193 /* 1194 * nfs read rpc call 1195 * Ditto above 1196 */ 1197 int 1198 nfs_readrpc(vp, uiop) 1199 struct vnode *vp; 1200 struct uio *uiop; 1201 { 1202 u_int32_t *tl; 1203 caddr_t cp; 1204 int32_t t1, t2; 1205 caddr_t bpos, dpos, cp2; 1206 struct mbuf *mreq, *mrep, *md, *mb; 1207 struct nfsmount *nmp; 1208 int error = 0, len, retlen, tsiz, eof, attrflag; 1209 const int v3 = NFS_ISV3(vp); 1210 struct nfsnode *np = VTONFS(vp); 1211 1212 #ifndef nolint 1213 eof = 0; 1214 #endif 1215 nmp = VFSTONFS(vp->v_mount); 1216 tsiz = uiop->uio_resid; 1217 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1218 return (EFBIG); 1219 while (tsiz > 0) { 1220 nfsstats.rpccnt[NFSPROC_READ]++; 1221 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1222 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1223 nfsm_fhtom(np, v3); 1224 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1225 if (v3) { 1226 txdr_hyper(uiop->uio_offset, tl); 1227 *(tl + 2) = txdr_unsigned(len); 1228 } else { 1229 *tl++ = txdr_unsigned(uiop->uio_offset); 1230 *tl++ = txdr_unsigned(len); 1231 *tl = 0; 1232 } 1233 nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred); 1234 if (v3) { 1235 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1236 if (error) { 1237 m_freem(mrep); 1238 goto nfsmout; 1239 } 1240 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1241 eof = fxdr_unsigned(int, *(tl + 1)); 1242 } else 1243 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1244 nfsm_strsiz(retlen, nmp->nm_rsize); 1245 nfsm_mtouio(uiop, retlen); 1246 m_freem(mrep); 1247 tsiz -= retlen; 1248 if (v3) { 1249 if (eof || retlen == 0) 1250 tsiz = 0; 1251 } else if (retlen < len) 1252 tsiz = 0; 1253 } 1254 nfsmout: 1255 return (error); 1256 } 1257 1258 struct nfs_writerpc_context { 1259 struct simplelock nwc_slock; 1260 volatile int nwc_mbufcount; 1261 }; 1262 1263 /* 1264 * free mbuf used to refer protected pages while write rpc call. 1265 * called at splvm. 1266 */ 1267 static void 1268 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 1269 { 1270 struct nfs_writerpc_context *ctx = arg; 1271 1272 KASSERT(m != NULL); 1273 KASSERT(ctx != NULL); 1274 pool_cache_put(&mbpool_cache, m); 1275 simple_lock(&ctx->nwc_slock); 1276 if (--ctx->nwc_mbufcount == 0) { 1277 wakeup(ctx); 1278 } 1279 simple_unlock(&ctx->nwc_slock); 1280 } 1281 1282 /* 1283 * nfs write call 1284 */ 1285 int 1286 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1287 struct vnode *vp; 1288 struct uio *uiop; 1289 int *iomode; 1290 boolean_t pageprotected; 1291 boolean_t *stalewriteverfp; 1292 { 1293 u_int32_t *tl; 1294 caddr_t cp; 1295 int32_t t1, t2; 1296 caddr_t bpos, dpos, cp2; 1297 struct mbuf *mreq, *mrep, *md, *mb; 1298 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1299 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1300 const int v3 = NFS_ISV3(vp); 1301 int committed = NFSV3WRITE_FILESYNC; 1302 struct nfsnode *np = VTONFS(vp); 1303 struct nfs_writerpc_context ctx; 1304 int s; 1305 struct lwp *l = NULL; 1306 size_t origresid; 1307 1308 simple_lock_init(&ctx.nwc_slock); 1309 ctx.nwc_mbufcount = 1; 1310 1311 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1312 panic("writerpc readonly vp %p", vp); 1313 } 1314 1315 #ifdef DIAGNOSTIC 1316 if (uiop->uio_iovcnt != 1) 1317 panic("nfs: writerpc iovcnt > 1"); 1318 #endif 1319 tsiz = uiop->uio_resid; 1320 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1321 return (EFBIG); 1322 if (pageprotected) { 1323 l = curlwp; 1324 PHOLD(l); 1325 } 1326 retry: 1327 origresid = uiop->uio_resid; 1328 KASSERT(origresid == uiop->uio_iov->iov_len); 1329 while (tsiz > 0) { 1330 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1331 uint32_t backup; 1332 boolean_t stalewriteverf = FALSE; 1333 1334 nfsstats.rpccnt[NFSPROC_WRITE]++; 1335 len = min(tsiz, nmp->nm_wsize); 1336 datalen = pageprotected ? 0 : nfsm_rndup(len); 1337 nfsm_reqhead(np, NFSPROC_WRITE, 1338 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1339 nfsm_fhtom(np, v3); 1340 if (v3) { 1341 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1342 txdr_hyper(uiop->uio_offset, tl); 1343 tl += 2; 1344 *tl++ = txdr_unsigned(len); 1345 *tl++ = txdr_unsigned(*iomode); 1346 *tl = txdr_unsigned(len); 1347 } else { 1348 u_int32_t x; 1349 1350 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1351 /* Set both "begin" and "current" to non-garbage. */ 1352 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1353 *tl++ = x; /* "begin offset" */ 1354 *tl++ = x; /* "current offset" */ 1355 x = txdr_unsigned(len); 1356 *tl++ = x; /* total to this offset */ 1357 *tl = x; /* size of this write */ 1358 1359 } 1360 if (pageprotected) { 1361 /* 1362 * since we know pages can't be modified during i/o, 1363 * no need to copy them for us. 1364 */ 1365 struct mbuf *m; 1366 struct iovec *iovp = uiop->uio_iov; 1367 1368 m = m_get(M_WAIT, MT_DATA); 1369 MCLAIM(m, &nfs_mowner); 1370 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1371 nfs_writerpc_extfree, &ctx); 1372 m->m_flags |= M_EXT_ROMAP; 1373 m->m_len = len; 1374 mb->m_next = m; 1375 /* 1376 * no need to maintain mb and bpos here 1377 * because no one care them later. 1378 */ 1379 #if 0 1380 mb = m; 1381 bpos = mtod(caddr_t, mb) + mb->m_len; 1382 #endif 1383 iovp->iov_base = (char *)iovp->iov_base + len; 1384 iovp->iov_len -= len; 1385 uiop->uio_offset += len; 1386 uiop->uio_resid -= len; 1387 s = splvm(); 1388 simple_lock(&ctx.nwc_slock); 1389 ctx.nwc_mbufcount++; 1390 simple_unlock(&ctx.nwc_slock); 1391 splx(s); 1392 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1393 } else { 1394 nfsm_uiotom(uiop, len); 1395 } 1396 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred); 1397 if (v3) { 1398 wccflag = NFSV3_WCCCHK; 1399 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC); 1400 if (!error) { 1401 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1402 + NFSX_V3WRITEVERF); 1403 rlen = fxdr_unsigned(int, *tl++); 1404 if (rlen == 0) { 1405 error = NFSERR_IO; 1406 m_freem(mrep); 1407 break; 1408 } else if (rlen < len) { 1409 backup = len - rlen; 1410 uiop->uio_iov->iov_base = 1411 (caddr_t)uiop->uio_iov->iov_base - 1412 backup; 1413 uiop->uio_iov->iov_len += backup; 1414 uiop->uio_offset -= backup; 1415 uiop->uio_resid += backup; 1416 len = rlen; 1417 } 1418 commit = fxdr_unsigned(int, *tl++); 1419 1420 /* 1421 * Return the lowest committment level 1422 * obtained by any of the RPCs. 1423 */ 1424 if (committed == NFSV3WRITE_FILESYNC) 1425 committed = commit; 1426 else if (committed == NFSV3WRITE_DATASYNC && 1427 commit == NFSV3WRITE_UNSTABLE) 1428 committed = commit; 1429 simple_lock(&nmp->nm_slock); 1430 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1431 memcpy(nmp->nm_writeverf, tl, 1432 NFSX_V3WRITEVERF); 1433 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1434 } else if ((nmp->nm_iflag & 1435 NFSMNT_STALEWRITEVERF) || 1436 memcmp(tl, nmp->nm_writeverf, 1437 NFSX_V3WRITEVERF)) { 1438 memcpy(nmp->nm_writeverf, tl, 1439 NFSX_V3WRITEVERF); 1440 /* 1441 * note NFSMNT_STALEWRITEVERF 1442 * if we're the first thread to 1443 * notice it. 1444 */ 1445 if ((nmp->nm_iflag & 1446 NFSMNT_STALEWRITEVERF) == 0) { 1447 stalewriteverf = TRUE; 1448 nmp->nm_iflag |= 1449 NFSMNT_STALEWRITEVERF; 1450 } 1451 } 1452 simple_unlock(&nmp->nm_slock); 1453 } 1454 } else 1455 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1456 if (wccflag) 1457 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1458 m_freem(mrep); 1459 if (error) 1460 break; 1461 tsiz -= len; 1462 if (stalewriteverf) { 1463 *stalewriteverfp = TRUE; 1464 stalewriteverf = FALSE; 1465 if (committed == NFSV3WRITE_UNSTABLE && 1466 len != origresid) { 1467 /* 1468 * if our write requests weren't atomic but 1469 * unstable, datas in previous iterations 1470 * might have already been lost now. 1471 * then, we should resend them to nfsd. 1472 */ 1473 backup = origresid - tsiz; 1474 uiop->uio_iov->iov_base = 1475 (caddr_t)uiop->uio_iov->iov_base - backup; 1476 uiop->uio_iov->iov_len += backup; 1477 uiop->uio_offset -= backup; 1478 uiop->uio_resid += backup; 1479 tsiz = origresid; 1480 goto retry; 1481 } 1482 } 1483 } 1484 if (pageprotected) { 1485 /* 1486 * wait until mbufs go away. 1487 * retransmitted mbufs can survive longer than rpc requests 1488 * themselves. 1489 */ 1490 s = splvm(); 1491 simple_lock(&ctx.nwc_slock); 1492 ctx.nwc_mbufcount--; 1493 while (ctx.nwc_mbufcount > 0) { 1494 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock); 1495 } 1496 simple_unlock(&ctx.nwc_slock); 1497 splx(s); 1498 PRELE(l); 1499 } 1500 nfsmout: 1501 *iomode = committed; 1502 if (error) 1503 uiop->uio_resid = tsiz; 1504 return (error); 1505 } 1506 1507 /* 1508 * nfs mknod rpc 1509 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1510 * mode set to specify the file type and the size field for rdev. 1511 */ 1512 int 1513 nfs_mknodrpc(dvp, vpp, cnp, vap) 1514 struct vnode *dvp; 1515 struct vnode **vpp; 1516 struct componentname *cnp; 1517 struct vattr *vap; 1518 { 1519 struct nfsv2_sattr *sp; 1520 u_int32_t *tl; 1521 caddr_t cp; 1522 int32_t t1, t2; 1523 struct vnode *newvp = (struct vnode *)0; 1524 struct nfsnode *dnp, *np; 1525 char *cp2; 1526 caddr_t bpos, dpos; 1527 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1528 struct mbuf *mreq, *mrep, *md, *mb; 1529 u_int32_t rdev; 1530 const int v3 = NFS_ISV3(dvp); 1531 1532 if (vap->va_type == VCHR || vap->va_type == VBLK) 1533 rdev = txdr_unsigned(vap->va_rdev); 1534 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1535 rdev = nfs_xdrneg1; 1536 else { 1537 VOP_ABORTOP(dvp, cnp); 1538 vput(dvp); 1539 return (EOPNOTSUPP); 1540 } 1541 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1542 dnp = VTONFS(dvp); 1543 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1544 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1545 nfsm_fhtom(dnp, v3); 1546 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1547 if (v3) { 1548 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1549 *tl++ = vtonfsv3_type(vap->va_type); 1550 nfsm_v3attrbuild(vap, FALSE); 1551 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1552 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1553 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1554 *tl = txdr_unsigned(minor(vap->va_rdev)); 1555 } 1556 } else { 1557 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1558 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1559 sp->sa_uid = nfs_xdrneg1; 1560 sp->sa_gid = nfs_xdrneg1; 1561 sp->sa_size = rdev; 1562 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1563 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1564 } 1565 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred); 1566 if (!error) { 1567 nfsm_mtofh(dvp, newvp, v3, gotvp); 1568 if (!gotvp) { 1569 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1570 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1571 if (!error) 1572 newvp = NFSTOV(np); 1573 } 1574 } 1575 if (v3) 1576 nfsm_wcc_data(dvp, wccflag, 0); 1577 nfsm_reqdone; 1578 if (error) { 1579 if (newvp) 1580 vput(newvp); 1581 } else { 1582 if (cnp->cn_flags & MAKEENTRY) 1583 nfs_cache_enter(dvp, newvp, cnp); 1584 *vpp = newvp; 1585 } 1586 PNBUF_PUT(cnp->cn_pnbuf); 1587 VTONFS(dvp)->n_flag |= NMODIFIED; 1588 if (!wccflag) 1589 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1590 vput(dvp); 1591 return (error); 1592 } 1593 1594 /* 1595 * nfs mknod vop 1596 * just call nfs_mknodrpc() to do the work. 1597 */ 1598 /* ARGSUSED */ 1599 int 1600 nfs_mknod(v) 1601 void *v; 1602 { 1603 struct vop_mknod_args /* { 1604 struct vnode *a_dvp; 1605 struct vnode **a_vpp; 1606 struct componentname *a_cnp; 1607 struct vattr *a_vap; 1608 } */ *ap = v; 1609 int error; 1610 1611 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1612 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1613 return (error); 1614 } 1615 1616 static u_long create_verf; 1617 /* 1618 * nfs file create call 1619 */ 1620 int 1621 nfs_create(v) 1622 void *v; 1623 { 1624 struct vop_create_args /* { 1625 struct vnode *a_dvp; 1626 struct vnode **a_vpp; 1627 struct componentname *a_cnp; 1628 struct vattr *a_vap; 1629 } */ *ap = v; 1630 struct vnode *dvp = ap->a_dvp; 1631 struct vattr *vap = ap->a_vap; 1632 struct componentname *cnp = ap->a_cnp; 1633 struct nfsv2_sattr *sp; 1634 u_int32_t *tl; 1635 caddr_t cp; 1636 int32_t t1, t2; 1637 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1638 struct vnode *newvp = (struct vnode *)0; 1639 caddr_t bpos, dpos, cp2; 1640 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1641 struct mbuf *mreq, *mrep, *md, *mb; 1642 const int v3 = NFS_ISV3(dvp); 1643 1644 /* 1645 * Oops, not for me.. 1646 */ 1647 if (vap->va_type == VSOCK) 1648 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1649 1650 #ifdef VA_EXCLUSIVE 1651 if (vap->va_vaflags & VA_EXCLUSIVE) 1652 fmode |= O_EXCL; 1653 #endif 1654 again: 1655 error = 0; 1656 nfsstats.rpccnt[NFSPROC_CREATE]++; 1657 dnp = VTONFS(dvp); 1658 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1659 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1660 nfsm_fhtom(dnp, v3); 1661 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1662 if (v3) { 1663 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1664 if (fmode & O_EXCL) { 1665 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1666 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1667 #ifdef INET 1668 if (TAILQ_FIRST(&in_ifaddr)) 1669 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr; 1670 else 1671 *tl++ = create_verf; 1672 #else 1673 *tl++ = create_verf; 1674 #endif 1675 *tl = ++create_verf; 1676 } else { 1677 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1678 nfsm_v3attrbuild(vap, FALSE); 1679 } 1680 } else { 1681 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1682 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1683 sp->sa_uid = nfs_xdrneg1; 1684 sp->sa_gid = nfs_xdrneg1; 1685 sp->sa_size = 0; 1686 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1687 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1688 } 1689 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred); 1690 if (!error) { 1691 nfsm_mtofh(dvp, newvp, v3, gotvp); 1692 if (!gotvp) { 1693 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1694 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1695 if (!error) 1696 newvp = NFSTOV(np); 1697 } 1698 } 1699 if (v3) 1700 nfsm_wcc_data(dvp, wccflag, 0); 1701 nfsm_reqdone; 1702 if (error) { 1703 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1704 fmode &= ~O_EXCL; 1705 goto again; 1706 } 1707 if (newvp) 1708 vput(newvp); 1709 } else if (v3 && (fmode & O_EXCL)) 1710 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc); 1711 if (!error) { 1712 if (cnp->cn_flags & MAKEENTRY) 1713 nfs_cache_enter(dvp, newvp, cnp); 1714 *ap->a_vpp = newvp; 1715 } 1716 PNBUF_PUT(cnp->cn_pnbuf); 1717 VTONFS(dvp)->n_flag |= NMODIFIED; 1718 if (!wccflag) 1719 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1720 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1721 vput(dvp); 1722 return (error); 1723 } 1724 1725 /* 1726 * nfs file remove call 1727 * To try and make nfs semantics closer to ufs semantics, a file that has 1728 * other processes using the vnode is renamed instead of removed and then 1729 * removed later on the last close. 1730 * - If v_usecount > 1 1731 * If a rename is not already in the works 1732 * call nfs_sillyrename() to set it up 1733 * else 1734 * do the remove rpc 1735 */ 1736 int 1737 nfs_remove(v) 1738 void *v; 1739 { 1740 struct vop_remove_args /* { 1741 struct vnodeop_desc *a_desc; 1742 struct vnode * a_dvp; 1743 struct vnode * a_vp; 1744 struct componentname * a_cnp; 1745 } */ *ap = v; 1746 struct vnode *vp = ap->a_vp; 1747 struct vnode *dvp = ap->a_dvp; 1748 struct componentname *cnp = ap->a_cnp; 1749 struct nfsnode *np = VTONFS(vp); 1750 int error = 0; 1751 struct vattr vattr; 1752 1753 #ifndef DIAGNOSTIC 1754 if ((cnp->cn_flags & HASBUF) == 0) 1755 panic("nfs_remove: no name"); 1756 if (vp->v_usecount < 1) 1757 panic("nfs_remove: bad v_usecount"); 1758 #endif 1759 if (vp->v_type == VDIR) 1760 error = EPERM; 1761 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1762 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 && 1763 vattr.va_nlink > 1)) { 1764 /* 1765 * Purge the name cache so that the chance of a lookup for 1766 * the name succeeding while the remove is in progress is 1767 * minimized. Without node locking it can still happen, such 1768 * that an I/O op returns ESTALE, but since you get this if 1769 * another host removes the file.. 1770 */ 1771 cache_purge(vp); 1772 /* 1773 * throw away biocache buffers, mainly to avoid 1774 * unnecessary delayed writes later. 1775 */ 1776 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1); 1777 /* Do the rpc */ 1778 if (error != EINTR) 1779 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1780 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc); 1781 /* 1782 * Kludge City: If the first reply to the remove rpc is lost.. 1783 * the reply to the retransmitted request will be ENOENT 1784 * since the file was in fact removed 1785 * Therefore, we cheat and return success. 1786 */ 1787 if (error == ENOENT) 1788 error = 0; 1789 } else if (!np->n_sillyrename) 1790 error = nfs_sillyrename(dvp, vp, cnp); 1791 PNBUF_PUT(cnp->cn_pnbuf); 1792 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1793 vattr.va_nlink == 1) { 1794 np->n_flag |= NREMOVED; 1795 } 1796 NFS_INVALIDATE_ATTRCACHE(np); 1797 VN_KNOTE(vp, NOTE_DELETE); 1798 VN_KNOTE(dvp, NOTE_WRITE); 1799 if (dvp == vp) 1800 vrele(vp); 1801 else 1802 vput(vp); 1803 vput(dvp); 1804 return (error); 1805 } 1806 1807 /* 1808 * nfs file remove rpc called from nfs_inactive 1809 */ 1810 int 1811 nfs_removeit(sp) 1812 struct sillyrename *sp; 1813 { 1814 1815 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1816 (struct proc *)0)); 1817 } 1818 1819 /* 1820 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1821 */ 1822 int 1823 nfs_removerpc(dvp, name, namelen, cred, proc) 1824 struct vnode *dvp; 1825 const char *name; 1826 int namelen; 1827 struct ucred *cred; 1828 struct proc *proc; 1829 { 1830 u_int32_t *tl; 1831 caddr_t cp; 1832 int32_t t1, t2; 1833 caddr_t bpos, dpos, cp2; 1834 int error = 0, wccflag = NFSV3_WCCRATTR; 1835 struct mbuf *mreq, *mrep, *md, *mb; 1836 const int v3 = NFS_ISV3(dvp); 1837 struct nfsnode *dnp = VTONFS(dvp); 1838 1839 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1840 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1841 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1842 nfsm_fhtom(dnp, v3); 1843 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1844 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred); 1845 if (v3) 1846 nfsm_wcc_data(dvp, wccflag, 0); 1847 nfsm_reqdone; 1848 VTONFS(dvp)->n_flag |= NMODIFIED; 1849 if (!wccflag) 1850 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1851 return (error); 1852 } 1853 1854 /* 1855 * nfs file rename call 1856 */ 1857 int 1858 nfs_rename(v) 1859 void *v; 1860 { 1861 struct vop_rename_args /* { 1862 struct vnode *a_fdvp; 1863 struct vnode *a_fvp; 1864 struct componentname *a_fcnp; 1865 struct vnode *a_tdvp; 1866 struct vnode *a_tvp; 1867 struct componentname *a_tcnp; 1868 } */ *ap = v; 1869 struct vnode *fvp = ap->a_fvp; 1870 struct vnode *tvp = ap->a_tvp; 1871 struct vnode *fdvp = ap->a_fdvp; 1872 struct vnode *tdvp = ap->a_tdvp; 1873 struct componentname *tcnp = ap->a_tcnp; 1874 struct componentname *fcnp = ap->a_fcnp; 1875 int error; 1876 1877 #ifndef DIAGNOSTIC 1878 if ((tcnp->cn_flags & HASBUF) == 0 || 1879 (fcnp->cn_flags & HASBUF) == 0) 1880 panic("nfs_rename: no name"); 1881 #endif 1882 /* Check for cross-device rename */ 1883 if ((fvp->v_mount != tdvp->v_mount) || 1884 (tvp && (fvp->v_mount != tvp->v_mount))) { 1885 error = EXDEV; 1886 goto out; 1887 } 1888 1889 /* 1890 * If the tvp exists and is in use, sillyrename it before doing the 1891 * rename of the new file over it. 1892 */ 1893 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1894 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1895 VN_KNOTE(tvp, NOTE_DELETE); 1896 vput(tvp); 1897 tvp = NULL; 1898 } 1899 1900 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1901 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1902 tcnp->cn_proc); 1903 1904 VN_KNOTE(fdvp, NOTE_WRITE); 1905 VN_KNOTE(tdvp, NOTE_WRITE); 1906 if (fvp->v_type == VDIR) { 1907 if (tvp != NULL && tvp->v_type == VDIR) 1908 cache_purge(tdvp); 1909 cache_purge(fdvp); 1910 } 1911 out: 1912 if (tdvp == tvp) 1913 vrele(tdvp); 1914 else 1915 vput(tdvp); 1916 if (tvp) 1917 vput(tvp); 1918 vrele(fdvp); 1919 vrele(fvp); 1920 /* 1921 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1922 */ 1923 if (error == ENOENT) 1924 error = 0; 1925 return (error); 1926 } 1927 1928 /* 1929 * nfs file rename rpc called from nfs_remove() above 1930 */ 1931 int 1932 nfs_renameit(sdvp, scnp, sp) 1933 struct vnode *sdvp; 1934 struct componentname *scnp; 1935 struct sillyrename *sp; 1936 { 1937 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1938 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc)); 1939 } 1940 1941 /* 1942 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1943 */ 1944 int 1945 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc) 1946 struct vnode *fdvp; 1947 const char *fnameptr; 1948 int fnamelen; 1949 struct vnode *tdvp; 1950 const char *tnameptr; 1951 int tnamelen; 1952 struct ucred *cred; 1953 struct proc *proc; 1954 { 1955 u_int32_t *tl; 1956 caddr_t cp; 1957 int32_t t1, t2; 1958 caddr_t bpos, dpos, cp2; 1959 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1960 struct mbuf *mreq, *mrep, *md, *mb; 1961 const int v3 = NFS_ISV3(fdvp); 1962 struct nfsnode *fdnp = VTONFS(fdvp); 1963 1964 nfsstats.rpccnt[NFSPROC_RENAME]++; 1965 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1966 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1967 nfsm_rndup(tnamelen)); 1968 nfsm_fhtom(fdnp, v3); 1969 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1970 nfsm_fhtom(VTONFS(tdvp), v3); 1971 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1972 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred); 1973 if (v3) { 1974 nfsm_wcc_data(fdvp, fwccflag, 0); 1975 nfsm_wcc_data(tdvp, twccflag, 0); 1976 } 1977 nfsm_reqdone; 1978 VTONFS(fdvp)->n_flag |= NMODIFIED; 1979 VTONFS(tdvp)->n_flag |= NMODIFIED; 1980 if (!fwccflag) 1981 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1982 if (!twccflag) 1983 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1984 return (error); 1985 } 1986 1987 /* 1988 * nfs hard link create call 1989 */ 1990 int 1991 nfs_link(v) 1992 void *v; 1993 { 1994 struct vop_link_args /* { 1995 struct vnode *a_dvp; 1996 struct vnode *a_vp; 1997 struct componentname *a_cnp; 1998 } */ *ap = v; 1999 struct vnode *vp = ap->a_vp; 2000 struct vnode *dvp = ap->a_dvp; 2001 struct componentname *cnp = ap->a_cnp; 2002 u_int32_t *tl; 2003 caddr_t cp; 2004 int32_t t1, t2; 2005 caddr_t bpos, dpos, cp2; 2006 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2007 struct mbuf *mreq, *mrep, *md, *mb; 2008 /* XXX Should be const and initialised? */ 2009 int v3; 2010 struct nfsnode *np; 2011 2012 if (dvp->v_mount != vp->v_mount) { 2013 VOP_ABORTOP(dvp, cnp); 2014 vput(dvp); 2015 return (EXDEV); 2016 } 2017 if (dvp != vp) { 2018 error = vn_lock(vp, LK_EXCLUSIVE); 2019 if (error != 0) { 2020 VOP_ABORTOP(dvp, cnp); 2021 vput(dvp); 2022 return error; 2023 } 2024 } 2025 2026 /* 2027 * Push all writes to the server, so that the attribute cache 2028 * doesn't get "out of sync" with the server. 2029 * XXX There should be a better way! 2030 */ 2031 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc); 2032 2033 v3 = NFS_ISV3(vp); 2034 nfsstats.rpccnt[NFSPROC_LINK]++; 2035 np = VTONFS(vp); 2036 nfsm_reqhead(np, NFSPROC_LINK, 2037 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2038 nfsm_fhtom(np, v3); 2039 nfsm_fhtom(VTONFS(dvp), v3); 2040 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2041 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred); 2042 if (v3) { 2043 nfsm_postop_attr(vp, attrflag, 0); 2044 nfsm_wcc_data(dvp, wccflag, 0); 2045 } 2046 nfsm_reqdone; 2047 PNBUF_PUT(cnp->cn_pnbuf); 2048 VTONFS(dvp)->n_flag |= NMODIFIED; 2049 if (!attrflag) 2050 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2051 if (!wccflag) 2052 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2053 if (dvp != vp) 2054 VOP_UNLOCK(vp, 0); 2055 VN_KNOTE(vp, NOTE_LINK); 2056 VN_KNOTE(dvp, NOTE_WRITE); 2057 vput(dvp); 2058 /* 2059 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2060 */ 2061 if (error == EEXIST) 2062 error = 0; 2063 return (error); 2064 } 2065 2066 /* 2067 * nfs symbolic link create call 2068 */ 2069 int 2070 nfs_symlink(v) 2071 void *v; 2072 { 2073 struct vop_symlink_args /* { 2074 struct vnode *a_dvp; 2075 struct vnode **a_vpp; 2076 struct componentname *a_cnp; 2077 struct vattr *a_vap; 2078 char *a_target; 2079 } */ *ap = v; 2080 struct vnode *dvp = ap->a_dvp; 2081 struct vattr *vap = ap->a_vap; 2082 struct componentname *cnp = ap->a_cnp; 2083 struct nfsv2_sattr *sp; 2084 u_int32_t *tl; 2085 caddr_t cp; 2086 int32_t t1, t2; 2087 caddr_t bpos, dpos, cp2; 2088 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2089 struct mbuf *mreq, *mrep, *md, *mb; 2090 struct vnode *newvp = (struct vnode *)0; 2091 const int v3 = NFS_ISV3(dvp); 2092 struct nfsnode *dnp = VTONFS(dvp); 2093 2094 *ap->a_vpp = NULL; 2095 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2096 slen = strlen(ap->a_target); 2097 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2098 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2099 nfsm_fhtom(dnp, v3); 2100 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2101 if (v3) 2102 nfsm_v3attrbuild(vap, FALSE); 2103 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2104 if (!v3) { 2105 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2106 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2107 sp->sa_uid = nfs_xdrneg1; 2108 sp->sa_gid = nfs_xdrneg1; 2109 sp->sa_size = nfs_xdrneg1; 2110 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2111 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2112 } 2113 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred); 2114 if (v3) { 2115 if (!error) 2116 nfsm_mtofh(dvp, newvp, v3, gotvp); 2117 nfsm_wcc_data(dvp, wccflag, 0); 2118 } 2119 nfsm_reqdone; 2120 /* 2121 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2122 */ 2123 if (error == EEXIST) 2124 error = 0; 2125 if (error == 0 && newvp == NULL) { 2126 struct nfsnode *np = NULL; 2127 2128 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2129 cnp->cn_cred, cnp->cn_proc, &np); 2130 if (error == 0) 2131 newvp = NFSTOV(np); 2132 } 2133 if (error) { 2134 if (newvp != NULL) 2135 vput(newvp); 2136 } else { 2137 *ap->a_vpp = newvp; 2138 } 2139 PNBUF_PUT(cnp->cn_pnbuf); 2140 VTONFS(dvp)->n_flag |= NMODIFIED; 2141 if (!wccflag) 2142 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2143 VN_KNOTE(dvp, NOTE_WRITE); 2144 vput(dvp); 2145 return (error); 2146 } 2147 2148 /* 2149 * nfs make dir call 2150 */ 2151 int 2152 nfs_mkdir(v) 2153 void *v; 2154 { 2155 struct vop_mkdir_args /* { 2156 struct vnode *a_dvp; 2157 struct vnode **a_vpp; 2158 struct componentname *a_cnp; 2159 struct vattr *a_vap; 2160 } */ *ap = v; 2161 struct vnode *dvp = ap->a_dvp; 2162 struct vattr *vap = ap->a_vap; 2163 struct componentname *cnp = ap->a_cnp; 2164 struct nfsv2_sattr *sp; 2165 u_int32_t *tl; 2166 caddr_t cp; 2167 int32_t t1, t2; 2168 int len; 2169 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2170 struct vnode *newvp = (struct vnode *)0; 2171 caddr_t bpos, dpos, cp2; 2172 int error = 0, wccflag = NFSV3_WCCRATTR; 2173 int gotvp = 0; 2174 struct mbuf *mreq, *mrep, *md, *mb; 2175 const int v3 = NFS_ISV3(dvp); 2176 2177 len = cnp->cn_namelen; 2178 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2179 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2180 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2181 nfsm_fhtom(dnp, v3); 2182 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2183 if (v3) { 2184 nfsm_v3attrbuild(vap, FALSE); 2185 } else { 2186 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2187 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2188 sp->sa_uid = nfs_xdrneg1; 2189 sp->sa_gid = nfs_xdrneg1; 2190 sp->sa_size = nfs_xdrneg1; 2191 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2192 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2193 } 2194 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred); 2195 if (!error) 2196 nfsm_mtofh(dvp, newvp, v3, gotvp); 2197 if (v3) 2198 nfsm_wcc_data(dvp, wccflag, 0); 2199 nfsm_reqdone; 2200 VTONFS(dvp)->n_flag |= NMODIFIED; 2201 if (!wccflag) 2202 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2203 /* 2204 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2205 * if we can succeed in looking up the directory. 2206 */ 2207 if (error == EEXIST || (!error && !gotvp)) { 2208 if (newvp) { 2209 vput(newvp); 2210 newvp = (struct vnode *)0; 2211 } 2212 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2213 cnp->cn_proc, &np); 2214 if (!error) { 2215 newvp = NFSTOV(np); 2216 if (newvp->v_type != VDIR) 2217 error = EEXIST; 2218 } 2219 } 2220 if (error) { 2221 if (newvp) 2222 vput(newvp); 2223 } else { 2224 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2225 if (cnp->cn_flags & MAKEENTRY) 2226 nfs_cache_enter(dvp, newvp, cnp); 2227 *ap->a_vpp = newvp; 2228 } 2229 PNBUF_PUT(cnp->cn_pnbuf); 2230 vput(dvp); 2231 return (error); 2232 } 2233 2234 /* 2235 * nfs remove directory call 2236 */ 2237 int 2238 nfs_rmdir(v) 2239 void *v; 2240 { 2241 struct vop_rmdir_args /* { 2242 struct vnode *a_dvp; 2243 struct vnode *a_vp; 2244 struct componentname *a_cnp; 2245 } */ *ap = v; 2246 struct vnode *vp = ap->a_vp; 2247 struct vnode *dvp = ap->a_dvp; 2248 struct componentname *cnp = ap->a_cnp; 2249 u_int32_t *tl; 2250 caddr_t cp; 2251 int32_t t1, t2; 2252 caddr_t bpos, dpos, cp2; 2253 int error = 0, wccflag = NFSV3_WCCRATTR; 2254 struct mbuf *mreq, *mrep, *md, *mb; 2255 const int v3 = NFS_ISV3(dvp); 2256 struct nfsnode *dnp; 2257 2258 if (dvp == vp) { 2259 vrele(dvp); 2260 vput(dvp); 2261 PNBUF_PUT(cnp->cn_pnbuf); 2262 return (EINVAL); 2263 } 2264 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2265 dnp = VTONFS(dvp); 2266 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2267 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2268 nfsm_fhtom(dnp, v3); 2269 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2270 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred); 2271 if (v3) 2272 nfsm_wcc_data(dvp, wccflag, 0); 2273 nfsm_reqdone; 2274 PNBUF_PUT(cnp->cn_pnbuf); 2275 VTONFS(dvp)->n_flag |= NMODIFIED; 2276 if (!wccflag) 2277 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2278 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2279 VN_KNOTE(vp, NOTE_DELETE); 2280 cache_purge(vp); 2281 vput(vp); 2282 vput(dvp); 2283 /* 2284 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2285 */ 2286 if (error == ENOENT) 2287 error = 0; 2288 return (error); 2289 } 2290 2291 /* 2292 * nfs readdir call 2293 */ 2294 int 2295 nfs_readdir(v) 2296 void *v; 2297 { 2298 struct vop_readdir_args /* { 2299 struct vnode *a_vp; 2300 struct uio *a_uio; 2301 struct ucred *a_cred; 2302 int *a_eofflag; 2303 off_t **a_cookies; 2304 int *a_ncookies; 2305 } */ *ap = v; 2306 struct vnode *vp = ap->a_vp; 2307 struct uio *uio = ap->a_uio; 2308 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2309 char *base = uio->uio_iov->iov_base; 2310 int tresid, error; 2311 size_t count, lost; 2312 struct dirent *dp; 2313 off_t *cookies = NULL; 2314 int ncookies = 0, nc; 2315 2316 if (vp->v_type != VDIR) 2317 return (EPERM); 2318 2319 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2320 count = uio->uio_resid - lost; 2321 if (count <= 0) 2322 return (EINVAL); 2323 2324 /* 2325 * Call nfs_bioread() to do the real work. 2326 */ 2327 tresid = uio->uio_resid = count; 2328 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2329 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2330 2331 if (!error && ap->a_cookies) { 2332 ncookies = count / 16; 2333 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2334 *ap->a_cookies = cookies; 2335 } 2336 2337 if (!error && uio->uio_resid == tresid) { 2338 uio->uio_resid += lost; 2339 nfsstats.direofcache_misses++; 2340 if (ap->a_cookies) 2341 *ap->a_ncookies = 0; 2342 *ap->a_eofflag = 1; 2343 return (0); 2344 } 2345 2346 if (!error && ap->a_cookies) { 2347 /* 2348 * Only the NFS server and emulations use cookies, and they 2349 * load the directory block into system space, so we can 2350 * just look at it directly. 2351 */ 2352 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2353 panic("nfs_readdir: lost in space"); 2354 for (nc = 0; ncookies-- && 2355 base < (char *)uio->uio_iov->iov_base; nc++){ 2356 dp = (struct dirent *) base; 2357 if (dp->d_reclen == 0) 2358 break; 2359 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2360 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2361 else 2362 *(cookies++) = NFS_GETCOOKIE(dp); 2363 base += dp->d_reclen; 2364 } 2365 uio->uio_resid += 2366 ((caddr_t)uio->uio_iov->iov_base - base); 2367 uio->uio_iov->iov_len += 2368 ((caddr_t)uio->uio_iov->iov_base - base); 2369 uio->uio_iov->iov_base = base; 2370 *ap->a_ncookies = nc; 2371 } 2372 2373 uio->uio_resid += lost; 2374 *ap->a_eofflag = 0; 2375 return (error); 2376 } 2377 2378 /* 2379 * Readdir rpc call. 2380 * Called from below the buffer cache by nfs_doio(). 2381 */ 2382 int 2383 nfs_readdirrpc(vp, uiop, cred) 2384 struct vnode *vp; 2385 struct uio *uiop; 2386 struct ucred *cred; 2387 { 2388 int len, left; 2389 struct dirent *dp = NULL; 2390 u_int32_t *tl; 2391 caddr_t cp; 2392 int32_t t1, t2; 2393 caddr_t bpos, dpos, cp2; 2394 struct mbuf *mreq, *mrep, *md, *mb; 2395 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2396 struct nfsnode *dnp = VTONFS(vp); 2397 u_quad_t fileno; 2398 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2399 int attrflag, nrpcs = 0, reclen; 2400 const int v3 = NFS_ISV3(vp); 2401 2402 #ifdef DIAGNOSTIC 2403 /* 2404 * Should be called from buffer cache, so only amount of 2405 * NFS_DIRBLKSIZ will be requested. 2406 */ 2407 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2408 panic("nfs readdirrpc bad uio"); 2409 #endif 2410 2411 /* 2412 * Loop around doing readdir rpc's of size nm_readdirsize 2413 * truncated to a multiple of NFS_DIRFRAGSIZ. 2414 * The stopping criteria is EOF or buffer full. 2415 */ 2416 while (more_dirs && bigenough) { 2417 /* 2418 * Heuristic: don't bother to do another RPC to further 2419 * fill up this block if there is not much room left. (< 50% 2420 * of the readdir RPC size). This wastes some buffer space 2421 * but can save up to 50% in RPC calls. 2422 */ 2423 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2424 bigenough = 0; 2425 break; 2426 } 2427 nfsstats.rpccnt[NFSPROC_READDIR]++; 2428 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2429 NFSX_READDIR(v3)); 2430 nfsm_fhtom(dnp, v3); 2431 if (v3) { 2432 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2433 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2434 txdr_swapcookie3(uiop->uio_offset, tl); 2435 } else { 2436 txdr_cookie3(uiop->uio_offset, tl); 2437 } 2438 tl += 2; 2439 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2440 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2441 } else { 2442 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2443 *tl++ = txdr_unsigned(uiop->uio_offset); 2444 } 2445 *tl = txdr_unsigned(nmp->nm_readdirsize); 2446 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred); 2447 nrpcs++; 2448 if (v3) { 2449 nfsm_postop_attr(vp, attrflag, 0); 2450 if (!error) { 2451 nfsm_dissect(tl, u_int32_t *, 2452 2 * NFSX_UNSIGNED); 2453 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2454 dnp->n_cookieverf.nfsuquad[1] = *tl; 2455 } else { 2456 m_freem(mrep); 2457 goto nfsmout; 2458 } 2459 } 2460 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2461 more_dirs = fxdr_unsigned(int, *tl); 2462 2463 /* loop thru the dir entries, doctoring them to 4bsd form */ 2464 while (more_dirs && bigenough) { 2465 if (v3) { 2466 nfsm_dissect(tl, u_int32_t *, 2467 3 * NFSX_UNSIGNED); 2468 fileno = fxdr_hyper(tl); 2469 len = fxdr_unsigned(int, *(tl + 2)); 2470 } else { 2471 nfsm_dissect(tl, u_int32_t *, 2472 2 * NFSX_UNSIGNED); 2473 fileno = fxdr_unsigned(u_quad_t, *tl++); 2474 len = fxdr_unsigned(int, *tl); 2475 } 2476 if (len <= 0 || len > NFS_MAXNAMLEN) { 2477 error = EBADRPC; 2478 m_freem(mrep); 2479 goto nfsmout; 2480 } 2481 tlen = nfsm_rndup(len); 2482 if (tlen == len) 2483 tlen += 4; /* To ensure null termination */ 2484 tlen += sizeof (off_t) + sizeof (int); 2485 reclen = ALIGN(tlen + DIRHDSIZ); 2486 tlen = reclen - DIRHDSIZ; 2487 left = NFS_DIRFRAGSIZ - blksiz; 2488 if (reclen > left) { 2489 dp->d_reclen += left; 2490 uiop->uio_iov->iov_base = 2491 (caddr_t)uiop->uio_iov->iov_base + left; 2492 uiop->uio_iov->iov_len -= left; 2493 uiop->uio_resid -= left; 2494 blksiz = 0; 2495 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2496 } 2497 if (reclen > uiop->uio_resid) 2498 bigenough = 0; 2499 if (bigenough) { 2500 dp = (struct dirent *)uiop->uio_iov->iov_base; 2501 dp->d_fileno = (int)fileno; 2502 dp->d_namlen = len; 2503 dp->d_reclen = reclen; 2504 dp->d_type = DT_UNKNOWN; 2505 blksiz += dp->d_reclen; 2506 if (blksiz == NFS_DIRFRAGSIZ) 2507 blksiz = 0; 2508 uiop->uio_resid -= DIRHDSIZ; 2509 uiop->uio_iov->iov_base = 2510 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ; 2511 uiop->uio_iov->iov_len -= DIRHDSIZ; 2512 nfsm_mtouio(uiop, len); 2513 cp = uiop->uio_iov->iov_base; 2514 tlen -= len; 2515 *cp = '\0'; /* null terminate */ 2516 uiop->uio_iov->iov_base = 2517 (caddr_t)uiop->uio_iov->iov_base + tlen; 2518 uiop->uio_iov->iov_len -= tlen; 2519 uiop->uio_resid -= tlen; 2520 } else 2521 nfsm_adv(nfsm_rndup(len)); 2522 if (v3) { 2523 nfsm_dissect(tl, u_int32_t *, 2524 3 * NFSX_UNSIGNED); 2525 } else { 2526 nfsm_dissect(tl, u_int32_t *, 2527 2 * NFSX_UNSIGNED); 2528 } 2529 if (bigenough) { 2530 if (v3) { 2531 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2532 uiop->uio_offset = 2533 fxdr_swapcookie3(tl); 2534 else 2535 uiop->uio_offset = 2536 fxdr_cookie3(tl); 2537 } 2538 else { 2539 uiop->uio_offset = 2540 fxdr_unsigned(off_t, *tl); 2541 } 2542 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2543 } 2544 if (v3) 2545 tl += 2; 2546 else 2547 tl++; 2548 more_dirs = fxdr_unsigned(int, *tl); 2549 } 2550 /* 2551 * If at end of rpc data, get the eof boolean 2552 */ 2553 if (!more_dirs) { 2554 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2555 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2556 } 2557 m_freem(mrep); 2558 } 2559 /* 2560 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2561 * by increasing d_reclen for the last record. 2562 */ 2563 if (blksiz > 0) { 2564 left = NFS_DIRFRAGSIZ - blksiz; 2565 dp->d_reclen += left; 2566 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2567 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2568 left; 2569 uiop->uio_iov->iov_len -= left; 2570 uiop->uio_resid -= left; 2571 } 2572 2573 /* 2574 * We are now either at the end of the directory or have filled the 2575 * block. 2576 */ 2577 if (bigenough) 2578 dnp->n_direofoffset = uiop->uio_offset; 2579 nfsmout: 2580 return (error); 2581 } 2582 2583 /* 2584 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2585 */ 2586 int 2587 nfs_readdirplusrpc(vp, uiop, cred) 2588 struct vnode *vp; 2589 struct uio *uiop; 2590 struct ucred *cred; 2591 { 2592 int len, left; 2593 struct dirent *dp = NULL; 2594 u_int32_t *tl; 2595 caddr_t cp; 2596 int32_t t1, t2; 2597 struct vnode *newvp; 2598 caddr_t bpos, dpos, cp2; 2599 struct mbuf *mreq, *mrep, *md, *mb; 2600 struct nameidata nami, *ndp = &nami; 2601 struct componentname *cnp = &ndp->ni_cnd; 2602 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2603 struct nfsnode *dnp = VTONFS(vp), *np; 2604 nfsfh_t *fhp; 2605 u_quad_t fileno; 2606 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2607 int attrflag, fhsize, nrpcs = 0, reclen; 2608 struct nfs_fattr fattr, *fp; 2609 2610 #ifdef DIAGNOSTIC 2611 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2612 panic("nfs readdirplusrpc bad uio"); 2613 #endif 2614 ndp->ni_dvp = vp; 2615 newvp = NULLVP; 2616 2617 /* 2618 * Loop around doing readdir rpc's of size nm_readdirsize 2619 * truncated to a multiple of NFS_DIRFRAGSIZ. 2620 * The stopping criteria is EOF or buffer full. 2621 */ 2622 while (more_dirs && bigenough) { 2623 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2624 bigenough = 0; 2625 break; 2626 } 2627 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2628 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2629 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2630 nfsm_fhtom(dnp, 1); 2631 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2632 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2633 txdr_swapcookie3(uiop->uio_offset, tl); 2634 } else { 2635 txdr_cookie3(uiop->uio_offset, tl); 2636 } 2637 tl += 2; 2638 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2639 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2640 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2641 *tl = txdr_unsigned(nmp->nm_rsize); 2642 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred); 2643 nfsm_postop_attr(vp, attrflag, 0); 2644 if (error) { 2645 m_freem(mrep); 2646 goto nfsmout; 2647 } 2648 nrpcs++; 2649 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2650 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2651 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2652 more_dirs = fxdr_unsigned(int, *tl); 2653 2654 /* loop thru the dir entries, doctoring them to 4bsd form */ 2655 while (more_dirs && bigenough) { 2656 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2657 fileno = fxdr_hyper(tl); 2658 len = fxdr_unsigned(int, *(tl + 2)); 2659 if (len <= 0 || len > NFS_MAXNAMLEN) { 2660 error = EBADRPC; 2661 m_freem(mrep); 2662 goto nfsmout; 2663 } 2664 tlen = nfsm_rndup(len); 2665 if (tlen == len) 2666 tlen += 4; /* To ensure null termination*/ 2667 tlen += sizeof (off_t) + sizeof (int); 2668 reclen = ALIGN(tlen + DIRHDSIZ); 2669 tlen = reclen - DIRHDSIZ; 2670 left = NFS_DIRFRAGSIZ - blksiz; 2671 if (reclen > left) { 2672 /* 2673 * DIRFRAGSIZ is aligned, no need to align 2674 * again here. 2675 */ 2676 dp->d_reclen += left; 2677 uiop->uio_iov->iov_base = 2678 (caddr_t)uiop->uio_iov->iov_base + left; 2679 uiop->uio_iov->iov_len -= left; 2680 uiop->uio_resid -= left; 2681 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2682 blksiz = 0; 2683 } 2684 if (reclen > uiop->uio_resid) 2685 bigenough = 0; 2686 if (bigenough) { 2687 dp = (struct dirent *)uiop->uio_iov->iov_base; 2688 dp->d_fileno = (int)fileno; 2689 dp->d_namlen = len; 2690 dp->d_reclen = reclen; 2691 dp->d_type = DT_UNKNOWN; 2692 blksiz += dp->d_reclen; 2693 if (blksiz == NFS_DIRFRAGSIZ) 2694 blksiz = 0; 2695 uiop->uio_resid -= DIRHDSIZ; 2696 uiop->uio_iov->iov_base = 2697 (caddr_t)uiop->uio_iov->iov_base + 2698 DIRHDSIZ; 2699 uiop->uio_iov->iov_len -= DIRHDSIZ; 2700 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2701 cnp->cn_namelen = len; 2702 nfsm_mtouio(uiop, len); 2703 cp = uiop->uio_iov->iov_base; 2704 tlen -= len; 2705 *cp = '\0'; 2706 uiop->uio_iov->iov_base = 2707 (caddr_t)uiop->uio_iov->iov_base + tlen; 2708 uiop->uio_iov->iov_len -= tlen; 2709 uiop->uio_resid -= tlen; 2710 } else 2711 nfsm_adv(nfsm_rndup(len)); 2712 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2713 if (bigenough) { 2714 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2715 uiop->uio_offset = 2716 fxdr_swapcookie3(tl); 2717 else 2718 uiop->uio_offset = 2719 fxdr_cookie3(tl); 2720 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2721 } 2722 tl += 2; 2723 2724 /* 2725 * Since the attributes are before the file handle 2726 * (sigh), we must skip over the attributes and then 2727 * come back and get them. 2728 */ 2729 attrflag = fxdr_unsigned(int, *tl); 2730 if (attrflag) { 2731 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2732 memcpy(&fattr, fp, NFSX_V3FATTR); 2733 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2734 doit = fxdr_unsigned(int, *tl); 2735 if (doit) { 2736 nfsm_getfh(fhp, fhsize, 1); 2737 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2738 VREF(vp); 2739 newvp = vp; 2740 np = dnp; 2741 } else { 2742 error = nfs_nget1(vp->v_mount, fhp, 2743 fhsize, &np, LK_NOWAIT); 2744 if (!error) 2745 newvp = NFSTOV(np); 2746 } 2747 if (!error) { 2748 const char *cp; 2749 2750 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2751 dp->d_type = 2752 IFTODT(VTTOIF(np->n_vattr->va_type)); 2753 ndp->ni_vp = newvp; 2754 cp = cnp->cn_nameptr + cnp->cn_namelen; 2755 cnp->cn_hash = 2756 namei_hash(cnp->cn_nameptr, &cp); 2757 if (cnp->cn_namelen <= NCHNAMLEN) { 2758 cache_purge1(ndp->ni_dvp, cnp, 0); 2759 nfs_cache_enter(ndp->ni_dvp, ndp->ni_vp, 2760 cnp); 2761 } 2762 } 2763 } 2764 } else { 2765 /* Just skip over the file handle */ 2766 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2767 i = fxdr_unsigned(int, *tl); 2768 nfsm_adv(nfsm_rndup(i)); 2769 } 2770 if (newvp != NULLVP) { 2771 if (newvp == vp) 2772 vrele(newvp); 2773 else 2774 vput(newvp); 2775 newvp = NULLVP; 2776 } 2777 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2778 more_dirs = fxdr_unsigned(int, *tl); 2779 } 2780 /* 2781 * If at end of rpc data, get the eof boolean 2782 */ 2783 if (!more_dirs) { 2784 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2785 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2786 } 2787 m_freem(mrep); 2788 } 2789 /* 2790 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2791 * by increasing d_reclen for the last record. 2792 */ 2793 if (blksiz > 0) { 2794 left = NFS_DIRFRAGSIZ - blksiz; 2795 dp->d_reclen += left; 2796 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2797 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2798 left; 2799 uiop->uio_iov->iov_len -= left; 2800 uiop->uio_resid -= left; 2801 } 2802 2803 /* 2804 * We are now either at the end of the directory or have filled the 2805 * block. 2806 */ 2807 if (bigenough) 2808 dnp->n_direofoffset = uiop->uio_offset; 2809 nfsmout: 2810 if (newvp != NULLVP) { 2811 if(newvp == vp) 2812 vrele(newvp); 2813 else 2814 vput(newvp); 2815 } 2816 return (error); 2817 } 2818 static char hextoasc[] = "0123456789abcdef"; 2819 2820 /* 2821 * Silly rename. To make the NFS filesystem that is stateless look a little 2822 * more like the "ufs" a remove of an active vnode is translated to a rename 2823 * to a funny looking filename that is removed by nfs_inactive on the 2824 * nfsnode. There is the potential for another process on a different client 2825 * to create the same funny name between the nfs_lookitup() fails and the 2826 * nfs_rename() completes, but... 2827 */ 2828 int 2829 nfs_sillyrename(dvp, vp, cnp) 2830 struct vnode *dvp, *vp; 2831 struct componentname *cnp; 2832 { 2833 struct sillyrename *sp; 2834 struct nfsnode *np; 2835 int error; 2836 short pid; 2837 2838 cache_purge(dvp); 2839 np = VTONFS(vp); 2840 #ifndef DIAGNOSTIC 2841 if (vp->v_type == VDIR) 2842 panic("nfs: sillyrename dir"); 2843 #endif 2844 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2845 M_NFSREQ, M_WAITOK); 2846 sp->s_cred = crdup(cnp->cn_cred); 2847 sp->s_dvp = dvp; 2848 VREF(dvp); 2849 2850 /* Fudge together a funny name */ 2851 pid = cnp->cn_proc->p_pid; 2852 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2853 sp->s_namlen = 12; 2854 sp->s_name[8] = hextoasc[pid & 0xf]; 2855 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf]; 2856 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf]; 2857 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf]; 2858 2859 /* Try lookitups until we get one that isn't there */ 2860 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2861 cnp->cn_proc, (struct nfsnode **)0) == 0) { 2862 sp->s_name[4]++; 2863 if (sp->s_name[4] > 'z') { 2864 error = EINVAL; 2865 goto bad; 2866 } 2867 } 2868 error = nfs_renameit(dvp, cnp, sp); 2869 if (error) 2870 goto bad; 2871 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2872 cnp->cn_proc, &np); 2873 np->n_sillyrename = sp; 2874 return (0); 2875 bad: 2876 vrele(sp->s_dvp); 2877 crfree(sp->s_cred); 2878 free((caddr_t)sp, M_NFSREQ); 2879 return (error); 2880 } 2881 2882 /* 2883 * Look up a file name and optionally either update the file handle or 2884 * allocate an nfsnode, depending on the value of npp. 2885 * npp == NULL --> just do the lookup 2886 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2887 * handled too 2888 * *npp != NULL --> update the file handle in the vnode 2889 */ 2890 int 2891 nfs_lookitup(dvp, name, len, cred, procp, npp) 2892 struct vnode *dvp; 2893 const char *name; 2894 int len; 2895 struct ucred *cred; 2896 struct proc *procp; 2897 struct nfsnode **npp; 2898 { 2899 u_int32_t *tl; 2900 caddr_t cp; 2901 int32_t t1, t2; 2902 struct vnode *newvp = (struct vnode *)0; 2903 struct nfsnode *np, *dnp = VTONFS(dvp); 2904 caddr_t bpos, dpos, cp2; 2905 int error = 0, fhlen, attrflag; 2906 struct mbuf *mreq, *mrep, *md, *mb; 2907 nfsfh_t *nfhp; 2908 const int v3 = NFS_ISV3(dvp); 2909 2910 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2911 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2912 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2913 nfsm_fhtom(dnp, v3); 2914 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2915 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred); 2916 if (npp && !error) { 2917 nfsm_getfh(nfhp, fhlen, v3); 2918 if (*npp) { 2919 np = *npp; 2920 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2921 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2922 np->n_fhp = &np->n_fh; 2923 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2924 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2925 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen); 2926 np->n_fhsize = fhlen; 2927 newvp = NFSTOV(np); 2928 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2929 VREF(dvp); 2930 newvp = dvp; 2931 np = dnp; 2932 } else { 2933 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2934 if (error) { 2935 m_freem(mrep); 2936 return (error); 2937 } 2938 newvp = NFSTOV(np); 2939 } 2940 if (v3) { 2941 nfsm_postop_attr(newvp, attrflag, 0); 2942 if (!attrflag && *npp == NULL) { 2943 m_freem(mrep); 2944 vput(newvp); 2945 return (ENOENT); 2946 } 2947 } else 2948 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2949 } 2950 nfsm_reqdone; 2951 if (npp && *npp == NULL) { 2952 if (error) { 2953 if (newvp) 2954 vput(newvp); 2955 } else 2956 *npp = np; 2957 } 2958 return (error); 2959 } 2960 2961 /* 2962 * Nfs Version 3 commit rpc 2963 */ 2964 int 2965 nfs_commit(vp, offset, cnt, procp) 2966 struct vnode *vp; 2967 off_t offset; 2968 uint32_t cnt; 2969 struct proc *procp; 2970 { 2971 caddr_t cp; 2972 u_int32_t *tl; 2973 int32_t t1, t2; 2974 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2975 caddr_t bpos, dpos, cp2; 2976 int error = 0, wccflag = NFSV3_WCCRATTR; 2977 struct mbuf *mreq, *mrep, *md, *mb; 2978 struct nfsnode *np; 2979 2980 KASSERT(NFS_ISV3(vp)); 2981 2982 #ifdef NFS_DEBUG_COMMIT 2983 printf("commit %lu - %lu\n", (unsigned long)offset, 2984 (unsigned long)(offset + cnt)); 2985 #endif 2986 2987 simple_lock(&nmp->nm_slock); 2988 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 2989 simple_unlock(&nmp->nm_slock); 2990 return (0); 2991 } 2992 simple_unlock(&nmp->nm_slock); 2993 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2994 np = VTONFS(vp); 2995 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 2996 nfsm_fhtom(np, 1); 2997 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2998 txdr_hyper(offset, tl); 2999 tl += 2; 3000 *tl = txdr_unsigned(cnt); 3001 nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred); 3002 nfsm_wcc_data(vp, wccflag, 0); 3003 if (!error) { 3004 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3005 simple_lock(&nmp->nm_slock); 3006 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3007 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3008 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3009 error = NFSERR_STALEWRITEVERF; 3010 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3011 } 3012 simple_unlock(&nmp->nm_slock); 3013 } 3014 nfsm_reqdone; 3015 return (error); 3016 } 3017 3018 /* 3019 * Kludge City.. 3020 * - make nfs_bmap() essentially a no-op that does no translation 3021 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3022 * (Maybe I could use the process's page mapping, but I was concerned that 3023 * Kernel Write might not be enabled and also figured copyout() would do 3024 * a lot more work than memcpy() and also it currently happens in the 3025 * context of the swapper process (2). 3026 */ 3027 int 3028 nfs_bmap(v) 3029 void *v; 3030 { 3031 struct vop_bmap_args /* { 3032 struct vnode *a_vp; 3033 daddr_t a_bn; 3034 struct vnode **a_vpp; 3035 daddr_t *a_bnp; 3036 int *a_runp; 3037 } */ *ap = v; 3038 struct vnode *vp = ap->a_vp; 3039 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3040 3041 if (ap->a_vpp != NULL) 3042 *ap->a_vpp = vp; 3043 if (ap->a_bnp != NULL) 3044 *ap->a_bnp = ap->a_bn << bshift; 3045 if (ap->a_runp != NULL) 3046 *ap->a_runp = 1024 * 1024; /* XXX */ 3047 return (0); 3048 } 3049 3050 /* 3051 * Strategy routine. 3052 * For async requests when nfsiod(s) are running, queue the request by 3053 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3054 * request. 3055 */ 3056 int 3057 nfs_strategy(v) 3058 void *v; 3059 { 3060 struct vop_strategy_args *ap = v; 3061 struct buf *bp = ap->a_bp; 3062 struct proc *p; 3063 int error = 0; 3064 3065 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3066 panic("nfs physio/async"); 3067 if (bp->b_flags & B_ASYNC) 3068 p = NULL; 3069 else 3070 p = curproc; /* XXX */ 3071 3072 /* 3073 * If the op is asynchronous and an i/o daemon is waiting 3074 * queue the request, wake it up and wait for completion 3075 * otherwise just do it ourselves. 3076 */ 3077 3078 if ((bp->b_flags & B_ASYNC) == 0 || 3079 nfs_asyncio(bp)) 3080 error = nfs_doio(bp, p); 3081 return (error); 3082 } 3083 3084 /* 3085 * fsync vnode op. Just call nfs_flush() with commit == 1. 3086 */ 3087 /* ARGSUSED */ 3088 int 3089 nfs_fsync(v) 3090 void *v; 3091 { 3092 struct vop_fsync_args /* { 3093 struct vnodeop_desc *a_desc; 3094 struct vnode * a_vp; 3095 struct ucred * a_cred; 3096 int a_flags; 3097 off_t offlo; 3098 off_t offhi; 3099 struct proc * a_p; 3100 } */ *ap = v; 3101 3102 struct vnode *vp = ap->a_vp; 3103 3104 if (vp->v_type != VREG) 3105 return 0; 3106 3107 return (nfs_flush(vp, ap->a_cred, 3108 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1)); 3109 } 3110 3111 /* 3112 * Flush all the data associated with a vnode. 3113 */ 3114 int 3115 nfs_flush(vp, cred, waitfor, p, commit) 3116 struct vnode *vp; 3117 struct ucred *cred; 3118 int waitfor; 3119 struct proc *p; 3120 int commit; 3121 { 3122 struct nfsnode *np = VTONFS(vp); 3123 int error; 3124 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3125 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3126 3127 simple_lock(&vp->v_interlock); 3128 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3129 if (np->n_flag & NWRITEERR) { 3130 error = np->n_error; 3131 np->n_flag &= ~NWRITEERR; 3132 } 3133 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3134 return (error); 3135 } 3136 3137 /* 3138 * Return POSIX pathconf information applicable to nfs. 3139 * 3140 * N.B. The NFS V2 protocol doesn't support this RPC. 3141 */ 3142 /* ARGSUSED */ 3143 int 3144 nfs_pathconf(v) 3145 void *v; 3146 { 3147 struct vop_pathconf_args /* { 3148 struct vnode *a_vp; 3149 int a_name; 3150 register_t *a_retval; 3151 } */ *ap = v; 3152 struct nfsv3_pathconf *pcp; 3153 struct vnode *vp = ap->a_vp; 3154 struct mbuf *mreq, *mrep, *md, *mb; 3155 int32_t t1, t2; 3156 u_int32_t *tl; 3157 caddr_t bpos, dpos, cp, cp2; 3158 int error = 0, attrflag; 3159 #ifndef NFS_V2_ONLY 3160 struct nfsmount *nmp; 3161 unsigned int l; 3162 u_int64_t maxsize; 3163 #endif 3164 const int v3 = NFS_ISV3(vp); 3165 struct nfsnode *np = VTONFS(vp); 3166 3167 switch (ap->a_name) { 3168 /* Names that can be resolved locally. */ 3169 case _PC_PIPE_BUF: 3170 *ap->a_retval = PIPE_BUF; 3171 break; 3172 case _PC_SYNC_IO: 3173 *ap->a_retval = 1; 3174 break; 3175 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3176 case _PC_LINK_MAX: 3177 case _PC_NAME_MAX: 3178 case _PC_CHOWN_RESTRICTED: 3179 case _PC_NO_TRUNC: 3180 if (!v3) { 3181 error = EINVAL; 3182 break; 3183 } 3184 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3185 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3186 nfsm_fhtom(np, 1); 3187 nfsm_request(np, NFSPROC_PATHCONF, 3188 curproc, curproc->p_ucred); /* XXX */ 3189 nfsm_postop_attr(vp, attrflag, 0); 3190 if (!error) { 3191 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3192 NFSX_V3PATHCONF); 3193 switch (ap->a_name) { 3194 case _PC_LINK_MAX: 3195 *ap->a_retval = 3196 fxdr_unsigned(register_t, pcp->pc_linkmax); 3197 break; 3198 case _PC_NAME_MAX: 3199 *ap->a_retval = 3200 fxdr_unsigned(register_t, pcp->pc_namemax); 3201 break; 3202 case _PC_CHOWN_RESTRICTED: 3203 *ap->a_retval = 3204 (pcp->pc_chownrestricted == nfs_true); 3205 break; 3206 case _PC_NO_TRUNC: 3207 *ap->a_retval = 3208 (pcp->pc_notrunc == nfs_true); 3209 break; 3210 } 3211 } 3212 nfsm_reqdone; 3213 break; 3214 case _PC_FILESIZEBITS: 3215 #ifndef NFS_V2_ONLY 3216 if (v3) { 3217 nmp = VFSTONFS(vp->v_mount); 3218 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3219 if ((error = nfs_fsinfo(nmp, vp, 3220 curproc->p_ucred, curproc)) != 0) /* XXX */ 3221 break; 3222 for (l = 0, maxsize = nmp->nm_maxfilesize; 3223 (maxsize >> l) > 0; l++) 3224 ; 3225 *ap->a_retval = l + 1; 3226 } else 3227 #endif 3228 { 3229 *ap->a_retval = 32; /* NFS V2 limitation */ 3230 } 3231 break; 3232 default: 3233 error = EINVAL; 3234 break; 3235 } 3236 3237 return (error); 3238 } 3239 3240 /* 3241 * NFS advisory byte-level locks. 3242 */ 3243 int 3244 nfs_advlock(v) 3245 void *v; 3246 { 3247 struct vop_advlock_args /* { 3248 struct vnode *a_vp; 3249 caddr_t a_id; 3250 int a_op; 3251 struct flock *a_fl; 3252 int a_flags; 3253 } */ *ap = v; 3254 struct nfsnode *np = VTONFS(ap->a_vp); 3255 3256 return lf_advlock(ap, &np->n_lockf, np->n_size); 3257 } 3258 3259 /* 3260 * Print out the contents of an nfsnode. 3261 */ 3262 int 3263 nfs_print(v) 3264 void *v; 3265 { 3266 struct vop_print_args /* { 3267 struct vnode *a_vp; 3268 } */ *ap = v; 3269 struct vnode *vp = ap->a_vp; 3270 struct nfsnode *np = VTONFS(vp); 3271 3272 printf("tag VT_NFS, fileid %ld fsid 0x%lx", 3273 np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3274 if (vp->v_type == VFIFO) 3275 fifo_printinfo(vp); 3276 printf("\n"); 3277 return (0); 3278 } 3279 3280 /* 3281 * NFS file truncation. 3282 */ 3283 int 3284 nfs_truncate(v) 3285 void *v; 3286 { 3287 #if 0 3288 struct vop_truncate_args /* { 3289 struct vnode *a_vp; 3290 off_t a_length; 3291 int a_flags; 3292 struct ucred *a_cred; 3293 struct proc *a_p; 3294 } */ *ap = v; 3295 #endif 3296 3297 /* Use nfs_setattr */ 3298 return (EOPNOTSUPP); 3299 } 3300 3301 /* 3302 * NFS update. 3303 */ 3304 int 3305 nfs_update(v) 3306 void *v; 3307 #if 0 3308 struct vop_update_args /* { 3309 struct vnode *a_vp; 3310 struct timespec *a_ta; 3311 struct timespec *a_tm; 3312 int a_waitfor; 3313 } */ *ap = v; 3314 #endif 3315 { 3316 3317 /* Use nfs_setattr */ 3318 return (EOPNOTSUPP); 3319 } 3320 3321 /* 3322 * Just call bwrite(). 3323 */ 3324 int 3325 nfs_bwrite(v) 3326 void *v; 3327 { 3328 struct vop_bwrite_args /* { 3329 struct vnode *a_bp; 3330 } */ *ap = v; 3331 3332 return (bwrite(ap->a_bp)); 3333 } 3334 3335 /* 3336 * nfs unlock wrapper. 3337 */ 3338 int 3339 nfs_unlock(void *v) 3340 { 3341 struct vop_unlock_args /* { 3342 struct vnode *a_vp; 3343 int a_flags; 3344 } */ *ap = v; 3345 struct vnode *vp = ap->a_vp; 3346 3347 /* 3348 * VOP_UNLOCK can be called by nfs_loadattrcache 3349 * with v_data == 0. 3350 */ 3351 if (VTONFS(vp)) { 3352 nfs_delayedtruncate(vp); 3353 } 3354 3355 return genfs_unlock(v); 3356 } 3357 3358 /* 3359 * nfs special file access vnode op. 3360 * Essentially just get vattr and then imitate iaccess() since the device is 3361 * local to the client. 3362 */ 3363 int 3364 nfsspec_access(v) 3365 void *v; 3366 { 3367 struct vop_access_args /* { 3368 struct vnode *a_vp; 3369 int a_mode; 3370 struct ucred *a_cred; 3371 struct proc *a_p; 3372 } */ *ap = v; 3373 struct vattr va; 3374 struct vnode *vp = ap->a_vp; 3375 int error; 3376 3377 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p); 3378 if (error) 3379 return (error); 3380 3381 /* 3382 * Disallow write attempts on filesystems mounted read-only; 3383 * unless the file is a socket, fifo, or a block or character 3384 * device resident on the filesystem. 3385 */ 3386 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3387 switch (vp->v_type) { 3388 case VREG: 3389 case VDIR: 3390 case VLNK: 3391 return (EROFS); 3392 default: 3393 break; 3394 } 3395 } 3396 3397 return (vaccess(va.va_type, va.va_mode, 3398 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3399 } 3400 3401 /* 3402 * Read wrapper for special devices. 3403 */ 3404 int 3405 nfsspec_read(v) 3406 void *v; 3407 { 3408 struct vop_read_args /* { 3409 struct vnode *a_vp; 3410 struct uio *a_uio; 3411 int a_ioflag; 3412 struct ucred *a_cred; 3413 } */ *ap = v; 3414 struct nfsnode *np = VTONFS(ap->a_vp); 3415 3416 /* 3417 * Set access flag. 3418 */ 3419 np->n_flag |= NACC; 3420 np->n_atim.tv_sec = time.tv_sec; 3421 np->n_atim.tv_nsec = time.tv_usec * 1000; 3422 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3423 } 3424 3425 /* 3426 * Write wrapper for special devices. 3427 */ 3428 int 3429 nfsspec_write(v) 3430 void *v; 3431 { 3432 struct vop_write_args /* { 3433 struct vnode *a_vp; 3434 struct uio *a_uio; 3435 int a_ioflag; 3436 struct ucred *a_cred; 3437 } */ *ap = v; 3438 struct nfsnode *np = VTONFS(ap->a_vp); 3439 3440 /* 3441 * Set update flag. 3442 */ 3443 np->n_flag |= NUPD; 3444 np->n_mtim.tv_sec = time.tv_sec; 3445 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3446 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3447 } 3448 3449 /* 3450 * Close wrapper for special devices. 3451 * 3452 * Update the times on the nfsnode then do device close. 3453 */ 3454 int 3455 nfsspec_close(v) 3456 void *v; 3457 { 3458 struct vop_close_args /* { 3459 struct vnode *a_vp; 3460 int a_fflag; 3461 struct ucred *a_cred; 3462 struct proc *a_p; 3463 } */ *ap = v; 3464 struct vnode *vp = ap->a_vp; 3465 struct nfsnode *np = VTONFS(vp); 3466 struct vattr vattr; 3467 3468 if (np->n_flag & (NACC | NUPD)) { 3469 np->n_flag |= NCHG; 3470 if (vp->v_usecount == 1 && 3471 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3472 VATTR_NULL(&vattr); 3473 if (np->n_flag & NACC) 3474 vattr.va_atime = np->n_atim; 3475 if (np->n_flag & NUPD) 3476 vattr.va_mtime = np->n_mtim; 3477 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3478 } 3479 } 3480 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3481 } 3482 3483 /* 3484 * Read wrapper for fifos. 3485 */ 3486 int 3487 nfsfifo_read(v) 3488 void *v; 3489 { 3490 struct vop_read_args /* { 3491 struct vnode *a_vp; 3492 struct uio *a_uio; 3493 int a_ioflag; 3494 struct ucred *a_cred; 3495 } */ *ap = v; 3496 struct nfsnode *np = VTONFS(ap->a_vp); 3497 3498 /* 3499 * Set access flag. 3500 */ 3501 np->n_flag |= NACC; 3502 np->n_atim.tv_sec = time.tv_sec; 3503 np->n_atim.tv_nsec = time.tv_usec * 1000; 3504 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3505 } 3506 3507 /* 3508 * Write wrapper for fifos. 3509 */ 3510 int 3511 nfsfifo_write(v) 3512 void *v; 3513 { 3514 struct vop_write_args /* { 3515 struct vnode *a_vp; 3516 struct uio *a_uio; 3517 int a_ioflag; 3518 struct ucred *a_cred; 3519 } */ *ap = v; 3520 struct nfsnode *np = VTONFS(ap->a_vp); 3521 3522 /* 3523 * Set update flag. 3524 */ 3525 np->n_flag |= NUPD; 3526 np->n_mtim.tv_sec = time.tv_sec; 3527 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3528 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3529 } 3530 3531 /* 3532 * Close wrapper for fifos. 3533 * 3534 * Update the times on the nfsnode then do fifo close. 3535 */ 3536 int 3537 nfsfifo_close(v) 3538 void *v; 3539 { 3540 struct vop_close_args /* { 3541 struct vnode *a_vp; 3542 int a_fflag; 3543 struct ucred *a_cred; 3544 struct proc *a_p; 3545 } */ *ap = v; 3546 struct vnode *vp = ap->a_vp; 3547 struct nfsnode *np = VTONFS(vp); 3548 struct vattr vattr; 3549 3550 if (np->n_flag & (NACC | NUPD)) { 3551 if (np->n_flag & NACC) { 3552 np->n_atim.tv_sec = time.tv_sec; 3553 np->n_atim.tv_nsec = time.tv_usec * 1000; 3554 } 3555 if (np->n_flag & NUPD) { 3556 np->n_mtim.tv_sec = time.tv_sec; 3557 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3558 } 3559 np->n_flag |= NCHG; 3560 if (vp->v_usecount == 1 && 3561 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3562 VATTR_NULL(&vattr); 3563 if (np->n_flag & NACC) 3564 vattr.va_atime = np->n_atim; 3565 if (np->n_flag & NUPD) 3566 vattr.va_mtime = np->n_mtim; 3567 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3568 } 3569 } 3570 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3571 } 3572
Indexes created Wed Sep 24 05:09:52 GMT 2025