nfs_vnops.c revision 1.186.2.6.2.1
1 /* $NetBSD: nfs_vnops.c,v 1.186.2.6.2.1 2005/01/11 06:36:39 jmc 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 1.186.2.6.2.1 2005/01/11 06:36:39 jmc 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 struct nfsnode *dnp = VTONFS(dvp); 277 278 if (vp != NULL) { 279 struct nfsnode *np = VTONFS(vp); 280 281 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 282 } 283 284 if (!timespecisset(&dnp->n_nctime)) 285 dnp->n_nctime = dnp->n_vattr->va_mtime; 286 287 cache_enter(dvp, vp, cnp); 288 } 289 290 /* 291 * nfs null call from vfs. 292 */ 293 int 294 nfs_null(vp, cred, procp) 295 struct vnode *vp; 296 struct ucred *cred; 297 struct proc *procp; 298 { 299 caddr_t bpos, dpos; 300 int error = 0; 301 struct mbuf *mreq, *mrep, *md, *mb; 302 struct nfsnode *np = VTONFS(vp); 303 304 nfsm_reqhead(np, NFSPROC_NULL, 0); 305 nfsm_request(np, NFSPROC_NULL, procp, cred); 306 nfsm_reqdone; 307 return (error); 308 } 309 310 /* 311 * nfs access vnode op. 312 * For nfs version 2, just return ok. File accesses may fail later. 313 * For nfs version 3, use the access rpc to check accessibility. If file modes 314 * are changed on the server, accesses might still fail later. 315 */ 316 int 317 nfs_access(v) 318 void *v; 319 { 320 struct vop_access_args /* { 321 struct vnode *a_vp; 322 int a_mode; 323 struct ucred *a_cred; 324 struct proc *a_p; 325 } */ *ap = v; 326 struct vnode *vp = ap->a_vp; 327 u_int32_t *tl; 328 caddr_t cp; 329 int32_t t1, t2; 330 caddr_t bpos, dpos, cp2; 331 int error = 0, attrflag, cachevalid; 332 struct mbuf *mreq, *mrep, *md, *mb; 333 u_int32_t mode, rmode; 334 const int v3 = NFS_ISV3(vp); 335 struct nfsnode *np = VTONFS(vp); 336 337 cachevalid = (np->n_accstamp != -1 && 338 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) && 339 np->n_accuid == ap->a_cred->cr_uid); 340 341 /* 342 * Check access cache first. If this request has been made for this 343 * uid shortly before, use the cached result. 344 */ 345 if (cachevalid) { 346 if (!np->n_accerror) { 347 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 348 return np->n_accerror; 349 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 350 return np->n_accerror; 351 } 352 353 /* 354 * For nfs v3, do an access rpc, otherwise you are stuck emulating 355 * ufs_access() locally using the vattr. This may not be correct, 356 * since the server may apply other access criteria such as 357 * client uid-->server uid mapping that we do not know about, but 358 * this is better than just returning anything that is lying about 359 * in the cache. 360 */ 361 if (v3) { 362 nfsstats.rpccnt[NFSPROC_ACCESS]++; 363 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 364 nfsm_fhtom(np, v3); 365 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 366 if (ap->a_mode & VREAD) 367 mode = NFSV3ACCESS_READ; 368 else 369 mode = 0; 370 if (vp->v_type != VDIR) { 371 if (ap->a_mode & VWRITE) 372 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 373 if (ap->a_mode & VEXEC) 374 mode |= NFSV3ACCESS_EXECUTE; 375 } else { 376 if (ap->a_mode & VWRITE) 377 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 378 NFSV3ACCESS_DELETE); 379 if (ap->a_mode & VEXEC) 380 mode |= NFSV3ACCESS_LOOKUP; 381 } 382 *tl = txdr_unsigned(mode); 383 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred); 384 nfsm_postop_attr(vp, attrflag, 0); 385 if (!error) { 386 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 387 rmode = fxdr_unsigned(u_int32_t, *tl); 388 /* 389 * The NFS V3 spec does not clarify whether or not 390 * the returned access bits can be a superset of 391 * the ones requested, so... 392 */ 393 if ((rmode & mode) != mode) 394 error = EACCES; 395 } 396 nfsm_reqdone; 397 } else 398 return (nfsspec_access(ap)); 399 /* 400 * Disallow write attempts on filesystems mounted read-only; 401 * unless the file is a socket, fifo, or a block or character 402 * device resident on the filesystem. 403 */ 404 if (!error && (ap->a_mode & VWRITE) && 405 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 406 switch (vp->v_type) { 407 case VREG: 408 case VDIR: 409 case VLNK: 410 error = EROFS; 411 default: 412 break; 413 } 414 } 415 416 if (!error || error == EACCES) { 417 /* 418 * If we got the same result as for a previous, 419 * different request, OR it in. Don't update 420 * the timestamp in that case. 421 */ 422 if (cachevalid && np->n_accstamp != -1 && 423 error == np->n_accerror) { 424 if (!error) 425 np->n_accmode |= ap->a_mode; 426 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 427 np->n_accmode = ap->a_mode; 428 } else { 429 np->n_accstamp = time.tv_sec; 430 np->n_accuid = ap->a_cred->cr_uid; 431 np->n_accmode = ap->a_mode; 432 np->n_accerror = error; 433 } 434 } 435 436 return (error); 437 } 438 439 /* 440 * nfs open vnode op 441 * Check to see if the type is ok 442 * and that deletion is not in progress. 443 * For paged in text files, you will need to flush the page cache 444 * if consistency is lost. 445 */ 446 /* ARGSUSED */ 447 int 448 nfs_open(v) 449 void *v; 450 { 451 struct vop_open_args /* { 452 struct vnode *a_vp; 453 int a_mode; 454 struct ucred *a_cred; 455 struct proc *a_p; 456 } */ *ap = v; 457 struct vnode *vp = ap->a_vp; 458 struct nfsnode *np = VTONFS(vp); 459 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 460 struct vattr vattr; 461 int error; 462 463 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 464 return (EACCES); 465 } 466 467 if (ap->a_mode & FREAD) { 468 if (np->n_rcred != NULL) 469 crfree(np->n_rcred); 470 np->n_rcred = ap->a_cred; 471 crhold(np->n_rcred); 472 } 473 if (ap->a_mode & FWRITE) { 474 if (np->n_wcred != NULL) 475 crfree(np->n_wcred); 476 np->n_wcred = ap->a_cred; 477 crhold(np->n_wcred); 478 } 479 480 #ifndef NFS_V2_ONLY 481 /* 482 * Get a valid lease. If cached data is stale, flush it. 483 */ 484 if (nmp->nm_flag & NFSMNT_NQNFS) { 485 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 486 do { 487 error = nqnfs_getlease(vp, ND_READ, ap->a_cred, 488 ap->a_p); 489 } while (error == NQNFS_EXPIRED); 490 if (error) 491 return (error); 492 if (np->n_lrev != np->n_brev || 493 (np->n_flag & NQNFSNONCACHE)) { 494 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 495 ap->a_p, 1)) == EINTR) 496 return (error); 497 np->n_brev = np->n_lrev; 498 } 499 } 500 } else 501 #endif 502 { 503 if (np->n_flag & NMODIFIED) { 504 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 505 ap->a_p, 1)) == EINTR) 506 return (error); 507 NFS_INVALIDATE_ATTRCACHE(np); 508 if (vp->v_type == VDIR) { 509 nfs_invaldircache(vp, 0); 510 np->n_direofoffset = 0; 511 } 512 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p); 513 if (error) 514 return (error); 515 np->n_mtime = vattr.va_mtime; 516 } else { 517 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p); 518 if (error) 519 return (error); 520 if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) { 521 if (vp->v_type == VDIR) { 522 nfs_invaldircache(vp, 0); 523 np->n_direofoffset = 0; 524 } 525 if ((error = nfs_vinvalbuf(vp, V_SAVE, 526 ap->a_cred, ap->a_p, 1)) == EINTR) 527 return (error); 528 np->n_mtime = vattr.va_mtime; 529 } 530 } 531 } 532 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) 533 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 534 return (0); 535 } 536 537 /* 538 * nfs close vnode op 539 * What an NFS client should do upon close after writing is a debatable issue. 540 * Most NFS clients push delayed writes to the server upon close, basically for 541 * two reasons: 542 * 1 - So that any write errors may be reported back to the client process 543 * doing the close system call. By far the two most likely errors are 544 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 545 * 2 - To put a worst case upper bound on cache inconsistency between 546 * multiple clients for the file. 547 * There is also a consistency problem for Version 2 of the protocol w.r.t. 548 * not being able to tell if other clients are writing a file concurrently, 549 * since there is no way of knowing if the changed modify time in the reply 550 * is only due to the write for this client. 551 * (NFS Version 3 provides weak cache consistency data in the reply that 552 * should be sufficient to detect and handle this case.) 553 * 554 * The current code does the following: 555 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 556 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 557 * or commit them (this satisfies 1 and 2 except for the 558 * case where the server crashes after this close but 559 * before the commit RPC, which is felt to be "good 560 * enough". Changing the last argument to nfs_flush() to 561 * a 1 would force a commit operation, if it is felt a 562 * commit is necessary now. 563 * for NQNFS - do nothing now, since 2 is dealt with via leases and 564 * 1 should be dealt with via an fsync() system call for 565 * cases where write errors are important. 566 */ 567 /* ARGSUSED */ 568 int 569 nfs_close(v) 570 void *v; 571 { 572 struct vop_close_args /* { 573 struct vnodeop_desc *a_desc; 574 struct vnode *a_vp; 575 int a_fflag; 576 struct ucred *a_cred; 577 struct proc *a_p; 578 } */ *ap = v; 579 struct vnode *vp = ap->a_vp; 580 struct nfsnode *np = VTONFS(vp); 581 int error = 0; 582 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 583 584 if (vp->v_type == VREG) { 585 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 586 (np->n_flag & NMODIFIED)) { 587 if (NFS_ISV3(vp)) { 588 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0); 589 np->n_flag &= ~NMODIFIED; 590 } else 591 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1); 592 NFS_INVALIDATE_ATTRCACHE(np); 593 } 594 if (np->n_flag & NWRITEERR) { 595 np->n_flag &= ~NWRITEERR; 596 error = np->n_error; 597 } 598 } 599 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 600 return (error); 601 } 602 603 /* 604 * nfs getattr call from vfs. 605 */ 606 int 607 nfs_getattr(v) 608 void *v; 609 { 610 struct vop_getattr_args /* { 611 struct vnode *a_vp; 612 struct vattr *a_vap; 613 struct ucred *a_cred; 614 struct proc *a_p; 615 } */ *ap = v; 616 struct vnode *vp = ap->a_vp; 617 struct nfsnode *np = VTONFS(vp); 618 caddr_t cp; 619 u_int32_t *tl; 620 int32_t t1, t2; 621 caddr_t bpos, dpos; 622 int error = 0; 623 struct mbuf *mreq, *mrep, *md, *mb; 624 const int v3 = NFS_ISV3(vp); 625 626 /* 627 * Update local times for special files. 628 */ 629 if (np->n_flag & (NACC | NUPD)) 630 np->n_flag |= NCHG; 631 632 /* 633 * if we have delayed truncation, do it now. 634 */ 635 nfs_delayedtruncate(vp); 636 637 /* 638 * First look in the cache. 639 */ 640 if (nfs_getattrcache(vp, ap->a_vap) == 0) 641 return (0); 642 nfsstats.rpccnt[NFSPROC_GETATTR]++; 643 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 644 nfsm_fhtom(np, v3); 645 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred); 646 if (!error) { 647 nfsm_loadattr(vp, ap->a_vap, 0); 648 if (vp->v_type == VDIR && 649 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 650 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 651 } 652 nfsm_reqdone; 653 return (error); 654 } 655 656 /* 657 * nfs setattr call. 658 */ 659 int 660 nfs_setattr(v) 661 void *v; 662 { 663 struct vop_setattr_args /* { 664 struct vnodeop_desc *a_desc; 665 struct vnode *a_vp; 666 struct vattr *a_vap; 667 struct ucred *a_cred; 668 struct proc *a_p; 669 } */ *ap = v; 670 struct vnode *vp = ap->a_vp; 671 struct nfsnode *np = VTONFS(vp); 672 struct vattr *vap = ap->a_vap; 673 int error = 0; 674 u_quad_t tsize = 0; 675 676 /* 677 * Setting of flags is not supported. 678 */ 679 if (vap->va_flags != VNOVAL) 680 return (EOPNOTSUPP); 681 682 /* 683 * Disallow write attempts if the filesystem is mounted read-only. 684 */ 685 if ((vap->va_uid != (uid_t)VNOVAL || 686 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 687 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 688 (vp->v_mount->mnt_flag & MNT_RDONLY)) 689 return (EROFS); 690 if (vap->va_size != VNOVAL) { 691 switch (vp->v_type) { 692 case VDIR: 693 return (EISDIR); 694 case VCHR: 695 case VBLK: 696 case VSOCK: 697 case VFIFO: 698 if (vap->va_mtime.tv_sec == VNOVAL && 699 vap->va_atime.tv_sec == VNOVAL && 700 vap->va_mode == (mode_t)VNOVAL && 701 vap->va_uid == (uid_t)VNOVAL && 702 vap->va_gid == (gid_t)VNOVAL) 703 return (0); 704 vap->va_size = VNOVAL; 705 break; 706 default: 707 /* 708 * Disallow write attempts if the filesystem is 709 * mounted read-only. 710 */ 711 if (vp->v_mount->mnt_flag & MNT_RDONLY) 712 return (EROFS); 713 uvm_vnp_setsize(vp, vap->va_size); 714 tsize = np->n_size; 715 np->n_size = vap->va_size; 716 if (vap->va_size == 0) 717 error = nfs_vinvalbuf(vp, 0, 718 ap->a_cred, ap->a_p, 1); 719 else 720 error = nfs_vinvalbuf(vp, V_SAVE, 721 ap->a_cred, ap->a_p, 1); 722 if (error) { 723 uvm_vnp_setsize(vp, tsize); 724 return (error); 725 } 726 np->n_vattr->va_size = vap->va_size; 727 } 728 } else if ((vap->va_mtime.tv_sec != VNOVAL || 729 vap->va_atime.tv_sec != VNOVAL) && 730 vp->v_type == VREG && 731 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 732 ap->a_p, 1)) == EINTR) 733 return (error); 734 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p); 735 if (error && vap->va_size != VNOVAL) { 736 np->n_size = np->n_vattr->va_size = tsize; 737 uvm_vnp_setsize(vp, np->n_size); 738 } 739 VN_KNOTE(vp, NOTE_ATTRIB); 740 return (error); 741 } 742 743 /* 744 * Do an nfs setattr rpc. 745 */ 746 int 747 nfs_setattrrpc(vp, vap, cred, procp) 748 struct vnode *vp; 749 struct vattr *vap; 750 struct ucred *cred; 751 struct proc *procp; 752 { 753 struct nfsv2_sattr *sp; 754 caddr_t cp; 755 int32_t t1, t2; 756 caddr_t bpos, dpos, cp2; 757 u_int32_t *tl; 758 int error = 0, wccflag = NFSV3_WCCRATTR; 759 struct mbuf *mreq, *mrep, *md, *mb; 760 const int v3 = NFS_ISV3(vp); 761 struct nfsnode *np = VTONFS(vp); 762 763 nfsstats.rpccnt[NFSPROC_SETATTR]++; 764 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 765 nfsm_fhtom(np, v3); 766 if (v3) { 767 nfsm_v3attrbuild(vap, TRUE); 768 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 769 *tl = nfs_false; 770 } else { 771 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 772 if (vap->va_mode == (mode_t)VNOVAL) 773 sp->sa_mode = nfs_xdrneg1; 774 else 775 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 776 if (vap->va_uid == (uid_t)VNOVAL) 777 sp->sa_uid = nfs_xdrneg1; 778 else 779 sp->sa_uid = txdr_unsigned(vap->va_uid); 780 if (vap->va_gid == (gid_t)VNOVAL) 781 sp->sa_gid = nfs_xdrneg1; 782 else 783 sp->sa_gid = txdr_unsigned(vap->va_gid); 784 sp->sa_size = txdr_unsigned(vap->va_size); 785 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 786 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 787 } 788 nfsm_request(np, NFSPROC_SETATTR, procp, cred); 789 if (v3) { 790 nfsm_wcc_data(vp, wccflag, 0); 791 } else 792 nfsm_loadattr(vp, (struct vattr *)0, 0); 793 nfsm_reqdone; 794 return (error); 795 } 796 797 /* 798 * nfs lookup call, one step at a time... 799 * First look in cache 800 * If not found, unlock the directory nfsnode and do the rpc 801 * 802 * This code is full of lock/unlock statements and checks, because 803 * we continue after cache_lookup has finished (we need to check 804 * with the attr cache and do an rpc if it has timed out). This means 805 * that the locking effects of cache_lookup have to be taken into 806 * account. 807 */ 808 int 809 nfs_lookup(v) 810 void *v; 811 { 812 struct vop_lookup_args /* { 813 struct vnodeop_desc *a_desc; 814 struct vnode *a_dvp; 815 struct vnode **a_vpp; 816 struct componentname *a_cnp; 817 } */ *ap = v; 818 struct componentname *cnp = ap->a_cnp; 819 struct vnode *dvp = ap->a_dvp; 820 struct vnode **vpp = ap->a_vpp; 821 int flags; 822 struct vnode *newvp; 823 u_int32_t *tl; 824 caddr_t cp; 825 int32_t t1, t2; 826 caddr_t bpos, dpos, cp2; 827 struct mbuf *mreq, *mrep, *md, *mb; 828 long len; 829 nfsfh_t *fhp; 830 struct nfsnode *np; 831 int lockparent, wantparent, error = 0, attrflag, fhsize; 832 const int v3 = NFS_ISV3(dvp); 833 834 cnp->cn_flags &= ~PDIRUNLOCK; 835 flags = cnp->cn_flags; 836 837 *vpp = NULLVP; 838 newvp = NULLVP; 839 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 840 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 841 return (EROFS); 842 if (dvp->v_type != VDIR) 843 return (ENOTDIR); 844 845 lockparent = flags & LOCKPARENT; 846 wantparent = flags & (LOCKPARENT|WANTPARENT); 847 np = VTONFS(dvp); 848 849 /* 850 * Before tediously performing a linear scan of the directory, 851 * check the name cache to see if the directory/name pair 852 * we are looking for is known already. 853 * If the directory/name pair is found in the name cache, 854 * we have to ensure the directory has not changed from 855 * the time the cache entry has been created. If it has, 856 * the cache entry has to be ignored. 857 */ 858 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) { 859 struct vattr vattr; 860 int err2; 861 862 if (error && error != ENOENT) { 863 *vpp = NULLVP; 864 return error; 865 } 866 867 if (cnp->cn_flags & PDIRUNLOCK) { 868 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 869 if (err2 != 0) { 870 *vpp = NULLVP; 871 return err2; 872 } 873 cnp->cn_flags &= ~PDIRUNLOCK; 874 } 875 876 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc); 877 if (err2 != 0) { 878 if (error == 0) { 879 if (*vpp != dvp) 880 vput(*vpp); 881 else 882 vrele(*vpp); 883 } 884 *vpp = NULLVP; 885 return err2; 886 } 887 888 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred, 889 cnp->cn_proc) || timespeccmp(&vattr.va_mtime, 890 &VTONFS(dvp)->n_nctime, !=)) { 891 if (error == 0) { 892 if (*vpp != dvp) 893 vput(*vpp); 894 else 895 vrele(*vpp); 896 *vpp = NULLVP; 897 } 898 cache_purge1(dvp, NULL, PURGE_CHILDREN); 899 timespecclear(&np->n_nctime); 900 goto dorpc; 901 } 902 903 if (error == ENOENT) { 904 return ENOENT; 905 } 906 907 newvp = *vpp; 908 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc) 909 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) 910 { 911 nfsstats.lookupcache_hits++; 912 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 913 cnp->cn_flags |= SAVENAME; 914 if ((!lockparent || !(flags & ISLASTCN)) && 915 newvp != dvp) 916 VOP_UNLOCK(dvp, 0); 917 KASSERT(newvp->v_type != VNON); 918 return (0); 919 } 920 cache_purge1(newvp, NULL, PURGE_PARENTS); 921 if (newvp != dvp) 922 vput(newvp); 923 else 924 vrele(newvp); 925 *vpp = NULLVP; 926 } 927 dorpc: 928 error = 0; 929 newvp = NULLVP; 930 nfsstats.lookupcache_misses++; 931 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 932 len = cnp->cn_namelen; 933 nfsm_reqhead(np, NFSPROC_LOOKUP, 934 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 935 nfsm_fhtom(np, v3); 936 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 937 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred); 938 if (error) { 939 nfsm_postop_attr(dvp, attrflag, 0); 940 m_freem(mrep); 941 goto nfsmout; 942 } 943 nfsm_getfh(fhp, fhsize, v3); 944 945 /* 946 * Handle RENAME case... 947 */ 948 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) { 949 if (NFS_CMPFH(np, fhp, fhsize)) { 950 m_freem(mrep); 951 return (EISDIR); 952 } 953 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 954 if (error) { 955 m_freem(mrep); 956 return error; 957 } 958 newvp = NFSTOV(np); 959 if (v3) { 960 nfsm_postop_attr(newvp, attrflag, 0); 961 nfsm_postop_attr(dvp, attrflag, 0); 962 } else 963 nfsm_loadattr(newvp, (struct vattr *)0, 0); 964 *vpp = newvp; 965 m_freem(mrep); 966 cnp->cn_flags |= SAVENAME; 967 if (!lockparent) { 968 VOP_UNLOCK(dvp, 0); 969 cnp->cn_flags |= PDIRUNLOCK; 970 } 971 goto validate; 972 } 973 974 /* 975 * The postop attr handling is duplicated for each if case, 976 * because it should be done while dvp is locked (unlocking 977 * dvp is different for each case). 978 */ 979 980 if (NFS_CMPFH(np, fhp, fhsize)) { 981 /* 982 * "." lookup 983 */ 984 VREF(dvp); 985 newvp = dvp; 986 if (v3) { 987 nfsm_postop_attr(newvp, attrflag, 0); 988 nfsm_postop_attr(dvp, attrflag, 0); 989 } else 990 nfsm_loadattr(newvp, (struct vattr *)0, 0); 991 } else if (flags & ISDOTDOT) { 992 /* 993 * ".." lookup 994 */ 995 VOP_UNLOCK(dvp, 0); 996 cnp->cn_flags |= PDIRUNLOCK; 997 998 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 999 if (error) { 1000 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 1001 cnp->cn_flags &= ~PDIRUNLOCK; 1002 m_freem(mrep); 1003 return error; 1004 } 1005 newvp = NFSTOV(np); 1006 1007 if (v3) { 1008 nfsm_postop_attr(newvp, attrflag, 0); 1009 nfsm_postop_attr(dvp, attrflag, 0); 1010 } else 1011 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1012 1013 if (lockparent && (flags & ISLASTCN)) { 1014 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) { 1015 m_freem(mrep); 1016 vput(newvp); 1017 return error; 1018 } 1019 cnp->cn_flags &= ~PDIRUNLOCK; 1020 } 1021 } else { 1022 /* 1023 * Other lookups. 1024 */ 1025 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1026 if (error) { 1027 m_freem(mrep); 1028 return error; 1029 } 1030 newvp = NFSTOV(np); 1031 if (v3) { 1032 nfsm_postop_attr(newvp, attrflag, 0); 1033 nfsm_postop_attr(dvp, attrflag, 0); 1034 } else 1035 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1036 if (!lockparent || !(flags & ISLASTCN)) { 1037 VOP_UNLOCK(dvp, 0); 1038 cnp->cn_flags |= PDIRUNLOCK; 1039 } 1040 } 1041 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1042 cnp->cn_flags |= SAVENAME; 1043 if ((cnp->cn_flags & MAKEENTRY) && 1044 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1045 nfs_cache_enter(dvp, newvp, cnp); 1046 } 1047 *vpp = newvp; 1048 nfsm_reqdone; 1049 if (error) { 1050 /* 1051 * We get here only because of errors returned by 1052 * the RPC. Otherwise we'll have returned above 1053 * (the nfsm_* macros will jump to nfsm_reqdone 1054 * on error). 1055 */ 1056 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1057 cnp->cn_nameiop != CREATE) { 1058 nfs_cache_enter(dvp, NULL, cnp); 1059 } 1060 if (newvp != NULLVP) { 1061 vrele(newvp); 1062 if (newvp != dvp) 1063 VOP_UNLOCK(newvp, 0); 1064 } 1065 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1066 (flags & ISLASTCN) && error == ENOENT) { 1067 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1068 error = EROFS; 1069 else 1070 error = EJUSTRETURN; 1071 } 1072 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1073 cnp->cn_flags |= SAVENAME; 1074 *vpp = NULL; 1075 return error; 1076 } 1077 1078 validate: 1079 /* 1080 * make sure we have valid type and size. 1081 */ 1082 1083 newvp = *vpp; 1084 if (newvp->v_type == VNON) { 1085 struct vattr vattr; /* dummy */ 1086 1087 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1088 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc); 1089 if (error) { 1090 vput(newvp); 1091 *vpp = NULL; 1092 } 1093 } 1094 1095 return error; 1096 } 1097 1098 /* 1099 * nfs read call. 1100 * Just call nfs_bioread() to do the work. 1101 */ 1102 int 1103 nfs_read(v) 1104 void *v; 1105 { 1106 struct vop_read_args /* { 1107 struct vnode *a_vp; 1108 struct uio *a_uio; 1109 int a_ioflag; 1110 struct ucred *a_cred; 1111 } */ *ap = v; 1112 struct vnode *vp = ap->a_vp; 1113 1114 if (vp->v_type != VREG) 1115 return (EPERM); 1116 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1117 } 1118 1119 /* 1120 * nfs readlink call 1121 */ 1122 int 1123 nfs_readlink(v) 1124 void *v; 1125 { 1126 struct vop_readlink_args /* { 1127 struct vnode *a_vp; 1128 struct uio *a_uio; 1129 struct ucred *a_cred; 1130 } */ *ap = v; 1131 struct vnode *vp = ap->a_vp; 1132 1133 if (vp->v_type != VLNK) 1134 return (EPERM); 1135 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1136 } 1137 1138 /* 1139 * Do a readlink rpc. 1140 * Called by nfs_doio() from below the buffer cache. 1141 */ 1142 int 1143 nfs_readlinkrpc(vp, uiop, cred) 1144 struct vnode *vp; 1145 struct uio *uiop; 1146 struct ucred *cred; 1147 { 1148 u_int32_t *tl; 1149 caddr_t cp; 1150 int32_t t1, t2; 1151 caddr_t bpos, dpos, cp2; 1152 int error = 0, attrflag; 1153 uint32_t len; 1154 struct mbuf *mreq, *mrep, *md, *mb; 1155 const int v3 = NFS_ISV3(vp); 1156 struct nfsnode *np = VTONFS(vp); 1157 1158 nfsstats.rpccnt[NFSPROC_READLINK]++; 1159 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1160 nfsm_fhtom(np, v3); 1161 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred); 1162 if (v3) 1163 nfsm_postop_attr(vp, attrflag, 0); 1164 if (!error) { 1165 if (v3) { 1166 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1167 len = fxdr_unsigned(uint32_t, *tl); 1168 if (len > MAXPATHLEN) { 1169 /* 1170 * this pathname is too long for us. 1171 */ 1172 m_freem(mrep); 1173 /* Solaris returns EINVAL. should we follow? */ 1174 error = ENAMETOOLONG; 1175 goto nfsmout; 1176 } 1177 } else { 1178 nfsm_strsiz(len, NFS_MAXPATHLEN); 1179 } 1180 nfsm_mtouio(uiop, len); 1181 } 1182 nfsm_reqdone; 1183 return (error); 1184 } 1185 1186 /* 1187 * nfs read rpc call 1188 * Ditto above 1189 */ 1190 int 1191 nfs_readrpc(vp, uiop) 1192 struct vnode *vp; 1193 struct uio *uiop; 1194 { 1195 u_int32_t *tl; 1196 caddr_t cp; 1197 int32_t t1, t2; 1198 caddr_t bpos, dpos, cp2; 1199 struct mbuf *mreq, *mrep, *md, *mb; 1200 struct nfsmount *nmp; 1201 int error = 0, len, retlen, tsiz, eof, attrflag; 1202 const int v3 = NFS_ISV3(vp); 1203 struct nfsnode *np = VTONFS(vp); 1204 1205 #ifndef nolint 1206 eof = 0; 1207 #endif 1208 nmp = VFSTONFS(vp->v_mount); 1209 tsiz = uiop->uio_resid; 1210 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1211 return (EFBIG); 1212 while (tsiz > 0) { 1213 nfsstats.rpccnt[NFSPROC_READ]++; 1214 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1215 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1216 nfsm_fhtom(np, v3); 1217 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1218 if (v3) { 1219 txdr_hyper(uiop->uio_offset, tl); 1220 *(tl + 2) = txdr_unsigned(len); 1221 } else { 1222 *tl++ = txdr_unsigned(uiop->uio_offset); 1223 *tl++ = txdr_unsigned(len); 1224 *tl = 0; 1225 } 1226 nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred); 1227 if (v3) { 1228 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1229 if (error) { 1230 m_freem(mrep); 1231 goto nfsmout; 1232 } 1233 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1234 eof = fxdr_unsigned(int, *(tl + 1)); 1235 } else 1236 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1237 nfsm_strsiz(retlen, nmp->nm_rsize); 1238 nfsm_mtouio(uiop, retlen); 1239 m_freem(mrep); 1240 tsiz -= retlen; 1241 if (v3) { 1242 if (eof || retlen == 0) 1243 tsiz = 0; 1244 } else if (retlen < len) 1245 tsiz = 0; 1246 } 1247 nfsmout: 1248 return (error); 1249 } 1250 1251 struct nfs_writerpc_context { 1252 struct simplelock nwc_slock; 1253 volatile int nwc_mbufcount; 1254 }; 1255 1256 /* 1257 * free mbuf used to refer protected pages while write rpc call. 1258 * called at splvm. 1259 */ 1260 static void 1261 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 1262 { 1263 struct nfs_writerpc_context *ctx = arg; 1264 1265 KASSERT(m != NULL); 1266 KASSERT(ctx != NULL); 1267 pool_cache_put(&mbpool_cache, m); 1268 simple_lock(&ctx->nwc_slock); 1269 if (--ctx->nwc_mbufcount == 0) { 1270 wakeup(ctx); 1271 } 1272 simple_unlock(&ctx->nwc_slock); 1273 } 1274 1275 /* 1276 * nfs write call 1277 */ 1278 int 1279 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1280 struct vnode *vp; 1281 struct uio *uiop; 1282 int *iomode; 1283 boolean_t pageprotected; 1284 boolean_t *stalewriteverfp; 1285 { 1286 u_int32_t *tl; 1287 caddr_t cp; 1288 int32_t t1, t2; 1289 caddr_t bpos, dpos, cp2; 1290 struct mbuf *mreq, *mrep, *md, *mb; 1291 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1292 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1293 const int v3 = NFS_ISV3(vp); 1294 int committed = NFSV3WRITE_FILESYNC; 1295 struct nfsnode *np = VTONFS(vp); 1296 struct nfs_writerpc_context ctx; 1297 int s; 1298 struct lwp *l = NULL; 1299 size_t origresid; 1300 1301 simple_lock_init(&ctx.nwc_slock); 1302 ctx.nwc_mbufcount = 1; 1303 1304 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1305 panic("writerpc readonly vp %p", vp); 1306 } 1307 1308 #ifdef DIAGNOSTIC 1309 if (uiop->uio_iovcnt != 1) 1310 panic("nfs: writerpc iovcnt > 1"); 1311 #endif 1312 tsiz = uiop->uio_resid; 1313 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1314 return (EFBIG); 1315 if (pageprotected) { 1316 l = curlwp; 1317 PHOLD(l); 1318 } 1319 retry: 1320 origresid = uiop->uio_resid; 1321 KASSERT(origresid == uiop->uio_iov->iov_len); 1322 while (tsiz > 0) { 1323 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1324 uint32_t backup; 1325 boolean_t stalewriteverf = FALSE; 1326 1327 nfsstats.rpccnt[NFSPROC_WRITE]++; 1328 len = min(tsiz, nmp->nm_wsize); 1329 datalen = pageprotected ? 0 : nfsm_rndup(len); 1330 nfsm_reqhead(np, NFSPROC_WRITE, 1331 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1332 nfsm_fhtom(np, v3); 1333 if (v3) { 1334 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1335 txdr_hyper(uiop->uio_offset, tl); 1336 tl += 2; 1337 *tl++ = txdr_unsigned(len); 1338 *tl++ = txdr_unsigned(*iomode); 1339 *tl = txdr_unsigned(len); 1340 } else { 1341 u_int32_t x; 1342 1343 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1344 /* Set both "begin" and "current" to non-garbage. */ 1345 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1346 *tl++ = x; /* "begin offset" */ 1347 *tl++ = x; /* "current offset" */ 1348 x = txdr_unsigned(len); 1349 *tl++ = x; /* total to this offset */ 1350 *tl = x; /* size of this write */ 1351 1352 } 1353 if (pageprotected) { 1354 /* 1355 * since we know pages can't be modified during i/o, 1356 * no need to copy them for us. 1357 */ 1358 struct mbuf *m; 1359 struct iovec *iovp = uiop->uio_iov; 1360 1361 m = m_get(M_WAIT, MT_DATA); 1362 MCLAIM(m, &nfs_mowner); 1363 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1364 nfs_writerpc_extfree, &ctx); 1365 m->m_flags |= M_EXT_ROMAP; 1366 m->m_len = len; 1367 mb->m_next = m; 1368 /* 1369 * no need to maintain mb and bpos here 1370 * because no one care them later. 1371 */ 1372 #if 0 1373 mb = m; 1374 bpos = mtod(caddr_t, mb) + mb->m_len; 1375 #endif 1376 iovp->iov_base = (char *)iovp->iov_base + len; 1377 iovp->iov_len -= len; 1378 uiop->uio_offset += len; 1379 uiop->uio_resid -= len; 1380 s = splvm(); 1381 simple_lock(&ctx.nwc_slock); 1382 ctx.nwc_mbufcount++; 1383 simple_unlock(&ctx.nwc_slock); 1384 splx(s); 1385 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1386 } else { 1387 nfsm_uiotom(uiop, len); 1388 } 1389 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred); 1390 if (v3) { 1391 wccflag = NFSV3_WCCCHK; 1392 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC); 1393 if (!error) { 1394 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1395 + NFSX_V3WRITEVERF); 1396 rlen = fxdr_unsigned(int, *tl++); 1397 if (rlen == 0) { 1398 error = NFSERR_IO; 1399 m_freem(mrep); 1400 break; 1401 } else if (rlen < len) { 1402 backup = len - rlen; 1403 uiop->uio_iov->iov_base = 1404 (caddr_t)uiop->uio_iov->iov_base - 1405 backup; 1406 uiop->uio_iov->iov_len += backup; 1407 uiop->uio_offset -= backup; 1408 uiop->uio_resid += backup; 1409 len = rlen; 1410 } 1411 commit = fxdr_unsigned(int, *tl++); 1412 1413 /* 1414 * Return the lowest committment level 1415 * obtained by any of the RPCs. 1416 */ 1417 if (committed == NFSV3WRITE_FILESYNC) 1418 committed = commit; 1419 else if (committed == NFSV3WRITE_DATASYNC && 1420 commit == NFSV3WRITE_UNSTABLE) 1421 committed = commit; 1422 simple_lock(&nmp->nm_slock); 1423 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1424 memcpy(nmp->nm_writeverf, tl, 1425 NFSX_V3WRITEVERF); 1426 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1427 } else if ((nmp->nm_iflag & 1428 NFSMNT_STALEWRITEVERF) || 1429 memcmp(tl, nmp->nm_writeverf, 1430 NFSX_V3WRITEVERF)) { 1431 memcpy(nmp->nm_writeverf, tl, 1432 NFSX_V3WRITEVERF); 1433 /* 1434 * note NFSMNT_STALEWRITEVERF 1435 * if we're the first thread to 1436 * notice it. 1437 */ 1438 if ((nmp->nm_iflag & 1439 NFSMNT_STALEWRITEVERF) == 0) { 1440 stalewriteverf = TRUE; 1441 nmp->nm_iflag |= 1442 NFSMNT_STALEWRITEVERF; 1443 } 1444 } 1445 simple_unlock(&nmp->nm_slock); 1446 } 1447 } else 1448 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1449 if (wccflag) 1450 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1451 m_freem(mrep); 1452 if (error) 1453 break; 1454 tsiz -= len; 1455 if (stalewriteverf) { 1456 *stalewriteverfp = TRUE; 1457 stalewriteverf = FALSE; 1458 if (committed == NFSV3WRITE_UNSTABLE && 1459 len != origresid) { 1460 /* 1461 * if our write requests weren't atomic but 1462 * unstable, datas in previous iterations 1463 * might have already been lost now. 1464 * then, we should resend them to nfsd. 1465 */ 1466 backup = origresid - tsiz; 1467 uiop->uio_iov->iov_base = 1468 (caddr_t)uiop->uio_iov->iov_base - backup; 1469 uiop->uio_iov->iov_len += backup; 1470 uiop->uio_offset -= backup; 1471 uiop->uio_resid += backup; 1472 tsiz = origresid; 1473 goto retry; 1474 } 1475 } 1476 } 1477 if (pageprotected) { 1478 /* 1479 * wait until mbufs go away. 1480 * retransmitted mbufs can survive longer than rpc requests 1481 * themselves. 1482 */ 1483 s = splvm(); 1484 simple_lock(&ctx.nwc_slock); 1485 ctx.nwc_mbufcount--; 1486 while (ctx.nwc_mbufcount > 0) { 1487 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock); 1488 } 1489 simple_unlock(&ctx.nwc_slock); 1490 splx(s); 1491 PRELE(l); 1492 } 1493 nfsmout: 1494 *iomode = committed; 1495 if (error) 1496 uiop->uio_resid = tsiz; 1497 return (error); 1498 } 1499 1500 /* 1501 * nfs mknod rpc 1502 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1503 * mode set to specify the file type and the size field for rdev. 1504 */ 1505 int 1506 nfs_mknodrpc(dvp, vpp, cnp, vap) 1507 struct vnode *dvp; 1508 struct vnode **vpp; 1509 struct componentname *cnp; 1510 struct vattr *vap; 1511 { 1512 struct nfsv2_sattr *sp; 1513 u_int32_t *tl; 1514 caddr_t cp; 1515 int32_t t1, t2; 1516 struct vnode *newvp = (struct vnode *)0; 1517 struct nfsnode *dnp, *np; 1518 char *cp2; 1519 caddr_t bpos, dpos; 1520 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1521 struct mbuf *mreq, *mrep, *md, *mb; 1522 u_int32_t rdev; 1523 const int v3 = NFS_ISV3(dvp); 1524 1525 if (vap->va_type == VCHR || vap->va_type == VBLK) 1526 rdev = txdr_unsigned(vap->va_rdev); 1527 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1528 rdev = nfs_xdrneg1; 1529 else { 1530 VOP_ABORTOP(dvp, cnp); 1531 vput(dvp); 1532 return (EOPNOTSUPP); 1533 } 1534 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1535 dnp = VTONFS(dvp); 1536 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1537 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1538 nfsm_fhtom(dnp, v3); 1539 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1540 if (v3) { 1541 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1542 *tl++ = vtonfsv3_type(vap->va_type); 1543 nfsm_v3attrbuild(vap, FALSE); 1544 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1545 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1546 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1547 *tl = txdr_unsigned(minor(vap->va_rdev)); 1548 } 1549 } else { 1550 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1551 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1552 sp->sa_uid = nfs_xdrneg1; 1553 sp->sa_gid = nfs_xdrneg1; 1554 sp->sa_size = rdev; 1555 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1556 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1557 } 1558 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred); 1559 if (!error) { 1560 nfsm_mtofh(dvp, newvp, v3, gotvp); 1561 if (!gotvp) { 1562 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1563 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1564 if (!error) 1565 newvp = NFSTOV(np); 1566 } 1567 } 1568 if (v3) 1569 nfsm_wcc_data(dvp, wccflag, 0); 1570 nfsm_reqdone; 1571 if (error) { 1572 if (newvp) 1573 vput(newvp); 1574 } else { 1575 if (cnp->cn_flags & MAKEENTRY) 1576 nfs_cache_enter(dvp, newvp, cnp); 1577 *vpp = newvp; 1578 } 1579 PNBUF_PUT(cnp->cn_pnbuf); 1580 VTONFS(dvp)->n_flag |= NMODIFIED; 1581 if (!wccflag) 1582 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1583 vput(dvp); 1584 return (error); 1585 } 1586 1587 /* 1588 * nfs mknod vop 1589 * just call nfs_mknodrpc() to do the work. 1590 */ 1591 /* ARGSUSED */ 1592 int 1593 nfs_mknod(v) 1594 void *v; 1595 { 1596 struct vop_mknod_args /* { 1597 struct vnode *a_dvp; 1598 struct vnode **a_vpp; 1599 struct componentname *a_cnp; 1600 struct vattr *a_vap; 1601 } */ *ap = v; 1602 int error; 1603 1604 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1605 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1606 return (error); 1607 } 1608 1609 static u_long create_verf; 1610 /* 1611 * nfs file create call 1612 */ 1613 int 1614 nfs_create(v) 1615 void *v; 1616 { 1617 struct vop_create_args /* { 1618 struct vnode *a_dvp; 1619 struct vnode **a_vpp; 1620 struct componentname *a_cnp; 1621 struct vattr *a_vap; 1622 } */ *ap = v; 1623 struct vnode *dvp = ap->a_dvp; 1624 struct vattr *vap = ap->a_vap; 1625 struct componentname *cnp = ap->a_cnp; 1626 struct nfsv2_sattr *sp; 1627 u_int32_t *tl; 1628 caddr_t cp; 1629 int32_t t1, t2; 1630 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1631 struct vnode *newvp = (struct vnode *)0; 1632 caddr_t bpos, dpos, cp2; 1633 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1634 struct mbuf *mreq, *mrep, *md, *mb; 1635 const int v3 = NFS_ISV3(dvp); 1636 1637 /* 1638 * Oops, not for me.. 1639 */ 1640 if (vap->va_type == VSOCK) 1641 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1642 1643 KASSERT(vap->va_type == VREG); 1644 1645 #ifdef VA_EXCLUSIVE 1646 if (vap->va_vaflags & VA_EXCLUSIVE) 1647 fmode |= O_EXCL; 1648 #endif 1649 again: 1650 error = 0; 1651 nfsstats.rpccnt[NFSPROC_CREATE]++; 1652 dnp = VTONFS(dvp); 1653 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1654 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1655 nfsm_fhtom(dnp, v3); 1656 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1657 if (v3) { 1658 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1659 if (fmode & O_EXCL) { 1660 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1661 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1662 #ifdef INET 1663 if (TAILQ_FIRST(&in_ifaddr)) 1664 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr; 1665 else 1666 *tl++ = create_verf; 1667 #else 1668 *tl++ = create_verf; 1669 #endif 1670 *tl = ++create_verf; 1671 } else { 1672 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1673 nfsm_v3attrbuild(vap, FALSE); 1674 } 1675 } else { 1676 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1677 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1678 sp->sa_uid = nfs_xdrneg1; 1679 sp->sa_gid = nfs_xdrneg1; 1680 sp->sa_size = 0; 1681 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1682 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1683 } 1684 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred); 1685 if (!error) { 1686 nfsm_mtofh(dvp, newvp, v3, gotvp); 1687 if (!gotvp) { 1688 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1689 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1690 if (!error) 1691 newvp = NFSTOV(np); 1692 } 1693 } 1694 if (v3) 1695 nfsm_wcc_data(dvp, wccflag, 0); 1696 nfsm_reqdone; 1697 if (error) { 1698 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1699 fmode &= ~O_EXCL; 1700 goto again; 1701 } 1702 if (newvp) 1703 vput(newvp); 1704 } else if (v3 && (fmode & O_EXCL)) { 1705 struct timeval tm = time; 1706 1707 /* 1708 * make sure that we'll update timestamps as 1709 * most server implementations use them to store 1710 * the create verifier. 1711 * 1712 * XXX it's better to use TOSERVER always. 1713 */ 1714 1715 if (vap->va_atime.tv_sec == VNOVAL) { 1716 vap->va_atime.tv_sec = tm.tv_sec; 1717 vap->va_atime.tv_nsec = tm.tv_usec * 1000; 1718 } 1719 if (vap->va_mtime.tv_sec == VNOVAL) { 1720 vap->va_mtime.tv_sec = tm.tv_sec; 1721 vap->va_mtime.tv_nsec = tm.tv_usec * 1000; 1722 } 1723 1724 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc); 1725 } 1726 if (!error) { 1727 if (cnp->cn_flags & MAKEENTRY) 1728 nfs_cache_enter(dvp, newvp, cnp); 1729 *ap->a_vpp = newvp; 1730 } 1731 PNBUF_PUT(cnp->cn_pnbuf); 1732 VTONFS(dvp)->n_flag |= NMODIFIED; 1733 if (!wccflag) 1734 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1735 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1736 vput(dvp); 1737 return (error); 1738 } 1739 1740 /* 1741 * nfs file remove call 1742 * To try and make nfs semantics closer to ufs semantics, a file that has 1743 * other processes using the vnode is renamed instead of removed and then 1744 * removed later on the last close. 1745 * - If v_usecount > 1 1746 * If a rename is not already in the works 1747 * call nfs_sillyrename() to set it up 1748 * else 1749 * do the remove rpc 1750 */ 1751 int 1752 nfs_remove(v) 1753 void *v; 1754 { 1755 struct vop_remove_args /* { 1756 struct vnodeop_desc *a_desc; 1757 struct vnode * a_dvp; 1758 struct vnode * a_vp; 1759 struct componentname * a_cnp; 1760 } */ *ap = v; 1761 struct vnode *vp = ap->a_vp; 1762 struct vnode *dvp = ap->a_dvp; 1763 struct componentname *cnp = ap->a_cnp; 1764 struct nfsnode *np = VTONFS(vp); 1765 int error = 0; 1766 struct vattr vattr; 1767 1768 #ifndef DIAGNOSTIC 1769 if ((cnp->cn_flags & HASBUF) == 0) 1770 panic("nfs_remove: no name"); 1771 if (vp->v_usecount < 1) 1772 panic("nfs_remove: bad v_usecount"); 1773 #endif 1774 if (vp->v_type == VDIR) 1775 error = EPERM; 1776 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1777 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 && 1778 vattr.va_nlink > 1)) { 1779 /* 1780 * Purge the name cache so that the chance of a lookup for 1781 * the name succeeding while the remove is in progress is 1782 * minimized. Without node locking it can still happen, such 1783 * that an I/O op returns ESTALE, but since you get this if 1784 * another host removes the file.. 1785 */ 1786 cache_purge(vp); 1787 /* 1788 * throw away biocache buffers, mainly to avoid 1789 * unnecessary delayed writes later. 1790 */ 1791 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1); 1792 /* Do the rpc */ 1793 if (error != EINTR) 1794 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1795 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc); 1796 /* 1797 * Kludge City: If the first reply to the remove rpc is lost.. 1798 * the reply to the retransmitted request will be ENOENT 1799 * since the file was in fact removed 1800 * Therefore, we cheat and return success. 1801 */ 1802 if (error == ENOENT) 1803 error = 0; 1804 } else if (!np->n_sillyrename) 1805 error = nfs_sillyrename(dvp, vp, cnp); 1806 PNBUF_PUT(cnp->cn_pnbuf); 1807 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1808 vattr.va_nlink == 1) { 1809 np->n_flag |= NREMOVED; 1810 } 1811 NFS_INVALIDATE_ATTRCACHE(np); 1812 VN_KNOTE(vp, NOTE_DELETE); 1813 VN_KNOTE(dvp, NOTE_WRITE); 1814 if (dvp == vp) 1815 vrele(vp); 1816 else 1817 vput(vp); 1818 vput(dvp); 1819 return (error); 1820 } 1821 1822 /* 1823 * nfs file remove rpc called from nfs_inactive 1824 */ 1825 int 1826 nfs_removeit(sp) 1827 struct sillyrename *sp; 1828 { 1829 1830 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1831 (struct proc *)0)); 1832 } 1833 1834 /* 1835 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1836 */ 1837 int 1838 nfs_removerpc(dvp, name, namelen, cred, proc) 1839 struct vnode *dvp; 1840 const char *name; 1841 int namelen; 1842 struct ucred *cred; 1843 struct proc *proc; 1844 { 1845 u_int32_t *tl; 1846 caddr_t cp; 1847 int32_t t1, t2; 1848 caddr_t bpos, dpos, cp2; 1849 int error = 0, wccflag = NFSV3_WCCRATTR; 1850 struct mbuf *mreq, *mrep, *md, *mb; 1851 const int v3 = NFS_ISV3(dvp); 1852 struct nfsnode *dnp = VTONFS(dvp); 1853 1854 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1855 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1856 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1857 nfsm_fhtom(dnp, v3); 1858 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1859 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred); 1860 if (v3) 1861 nfsm_wcc_data(dvp, wccflag, 0); 1862 nfsm_reqdone; 1863 VTONFS(dvp)->n_flag |= NMODIFIED; 1864 if (!wccflag) 1865 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1866 return (error); 1867 } 1868 1869 /* 1870 * nfs file rename call 1871 */ 1872 int 1873 nfs_rename(v) 1874 void *v; 1875 { 1876 struct vop_rename_args /* { 1877 struct vnode *a_fdvp; 1878 struct vnode *a_fvp; 1879 struct componentname *a_fcnp; 1880 struct vnode *a_tdvp; 1881 struct vnode *a_tvp; 1882 struct componentname *a_tcnp; 1883 } */ *ap = v; 1884 struct vnode *fvp = ap->a_fvp; 1885 struct vnode *tvp = ap->a_tvp; 1886 struct vnode *fdvp = ap->a_fdvp; 1887 struct vnode *tdvp = ap->a_tdvp; 1888 struct componentname *tcnp = ap->a_tcnp; 1889 struct componentname *fcnp = ap->a_fcnp; 1890 int error; 1891 1892 #ifndef DIAGNOSTIC 1893 if ((tcnp->cn_flags & HASBUF) == 0 || 1894 (fcnp->cn_flags & HASBUF) == 0) 1895 panic("nfs_rename: no name"); 1896 #endif 1897 /* Check for cross-device rename */ 1898 if ((fvp->v_mount != tdvp->v_mount) || 1899 (tvp && (fvp->v_mount != tvp->v_mount))) { 1900 error = EXDEV; 1901 goto out; 1902 } 1903 1904 /* 1905 * If the tvp exists and is in use, sillyrename it before doing the 1906 * rename of the new file over it. 1907 */ 1908 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1909 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1910 VN_KNOTE(tvp, NOTE_DELETE); 1911 vput(tvp); 1912 tvp = NULL; 1913 } 1914 1915 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1916 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1917 tcnp->cn_proc); 1918 1919 VN_KNOTE(fdvp, NOTE_WRITE); 1920 VN_KNOTE(tdvp, NOTE_WRITE); 1921 if (fvp->v_type == VDIR) { 1922 if (tvp != NULL && tvp->v_type == VDIR) 1923 cache_purge(tdvp); 1924 cache_purge(fdvp); 1925 } 1926 out: 1927 if (tdvp == tvp) 1928 vrele(tdvp); 1929 else 1930 vput(tdvp); 1931 if (tvp) 1932 vput(tvp); 1933 vrele(fdvp); 1934 vrele(fvp); 1935 /* 1936 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1937 */ 1938 if (error == ENOENT) 1939 error = 0; 1940 return (error); 1941 } 1942 1943 /* 1944 * nfs file rename rpc called from nfs_remove() above 1945 */ 1946 int 1947 nfs_renameit(sdvp, scnp, sp) 1948 struct vnode *sdvp; 1949 struct componentname *scnp; 1950 struct sillyrename *sp; 1951 { 1952 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1953 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc)); 1954 } 1955 1956 /* 1957 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1958 */ 1959 int 1960 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc) 1961 struct vnode *fdvp; 1962 const char *fnameptr; 1963 int fnamelen; 1964 struct vnode *tdvp; 1965 const char *tnameptr; 1966 int tnamelen; 1967 struct ucred *cred; 1968 struct proc *proc; 1969 { 1970 u_int32_t *tl; 1971 caddr_t cp; 1972 int32_t t1, t2; 1973 caddr_t bpos, dpos, cp2; 1974 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1975 struct mbuf *mreq, *mrep, *md, *mb; 1976 const int v3 = NFS_ISV3(fdvp); 1977 struct nfsnode *fdnp = VTONFS(fdvp); 1978 1979 nfsstats.rpccnt[NFSPROC_RENAME]++; 1980 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1981 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1982 nfsm_rndup(tnamelen)); 1983 nfsm_fhtom(fdnp, v3); 1984 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1985 nfsm_fhtom(VTONFS(tdvp), v3); 1986 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1987 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred); 1988 if (v3) { 1989 nfsm_wcc_data(fdvp, fwccflag, 0); 1990 nfsm_wcc_data(tdvp, twccflag, 0); 1991 } 1992 nfsm_reqdone; 1993 VTONFS(fdvp)->n_flag |= NMODIFIED; 1994 VTONFS(tdvp)->n_flag |= NMODIFIED; 1995 if (!fwccflag) 1996 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1997 if (!twccflag) 1998 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1999 return (error); 2000 } 2001 2002 /* 2003 * nfs hard link create call 2004 */ 2005 int 2006 nfs_link(v) 2007 void *v; 2008 { 2009 struct vop_link_args /* { 2010 struct vnode *a_dvp; 2011 struct vnode *a_vp; 2012 struct componentname *a_cnp; 2013 } */ *ap = v; 2014 struct vnode *vp = ap->a_vp; 2015 struct vnode *dvp = ap->a_dvp; 2016 struct componentname *cnp = ap->a_cnp; 2017 u_int32_t *tl; 2018 caddr_t cp; 2019 int32_t t1, t2; 2020 caddr_t bpos, dpos, cp2; 2021 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2022 struct mbuf *mreq, *mrep, *md, *mb; 2023 /* XXX Should be const and initialised? */ 2024 int v3; 2025 struct nfsnode *np; 2026 2027 if (dvp->v_mount != vp->v_mount) { 2028 VOP_ABORTOP(dvp, cnp); 2029 vput(dvp); 2030 return (EXDEV); 2031 } 2032 if (dvp != vp) { 2033 error = vn_lock(vp, LK_EXCLUSIVE); 2034 if (error != 0) { 2035 VOP_ABORTOP(dvp, cnp); 2036 vput(dvp); 2037 return error; 2038 } 2039 } 2040 2041 /* 2042 * Push all writes to the server, so that the attribute cache 2043 * doesn't get "out of sync" with the server. 2044 * XXX There should be a better way! 2045 */ 2046 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc); 2047 2048 v3 = NFS_ISV3(vp); 2049 nfsstats.rpccnt[NFSPROC_LINK]++; 2050 np = VTONFS(vp); 2051 nfsm_reqhead(np, NFSPROC_LINK, 2052 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2053 nfsm_fhtom(np, v3); 2054 nfsm_fhtom(VTONFS(dvp), v3); 2055 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2056 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred); 2057 if (v3) { 2058 nfsm_postop_attr(vp, attrflag, 0); 2059 nfsm_wcc_data(dvp, wccflag, 0); 2060 } 2061 nfsm_reqdone; 2062 PNBUF_PUT(cnp->cn_pnbuf); 2063 VTONFS(dvp)->n_flag |= NMODIFIED; 2064 if (!attrflag) 2065 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2066 if (!wccflag) 2067 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2068 if (dvp != vp) 2069 VOP_UNLOCK(vp, 0); 2070 VN_KNOTE(vp, NOTE_LINK); 2071 VN_KNOTE(dvp, NOTE_WRITE); 2072 vput(dvp); 2073 /* 2074 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2075 */ 2076 if (error == EEXIST) 2077 error = 0; 2078 return (error); 2079 } 2080 2081 /* 2082 * nfs symbolic link create call 2083 */ 2084 int 2085 nfs_symlink(v) 2086 void *v; 2087 { 2088 struct vop_symlink_args /* { 2089 struct vnode *a_dvp; 2090 struct vnode **a_vpp; 2091 struct componentname *a_cnp; 2092 struct vattr *a_vap; 2093 char *a_target; 2094 } */ *ap = v; 2095 struct vnode *dvp = ap->a_dvp; 2096 struct vattr *vap = ap->a_vap; 2097 struct componentname *cnp = ap->a_cnp; 2098 struct nfsv2_sattr *sp; 2099 u_int32_t *tl; 2100 caddr_t cp; 2101 int32_t t1, t2; 2102 caddr_t bpos, dpos, cp2; 2103 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2104 struct mbuf *mreq, *mrep, *md, *mb; 2105 struct vnode *newvp = (struct vnode *)0; 2106 const int v3 = NFS_ISV3(dvp); 2107 struct nfsnode *dnp = VTONFS(dvp); 2108 2109 *ap->a_vpp = NULL; 2110 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2111 slen = strlen(ap->a_target); 2112 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2113 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2114 nfsm_fhtom(dnp, v3); 2115 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2116 if (v3) 2117 nfsm_v3attrbuild(vap, FALSE); 2118 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2119 if (!v3) { 2120 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2121 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2122 sp->sa_uid = nfs_xdrneg1; 2123 sp->sa_gid = nfs_xdrneg1; 2124 sp->sa_size = nfs_xdrneg1; 2125 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2126 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2127 } 2128 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred); 2129 if (v3) { 2130 if (!error) 2131 nfsm_mtofh(dvp, newvp, v3, gotvp); 2132 nfsm_wcc_data(dvp, wccflag, 0); 2133 } 2134 nfsm_reqdone; 2135 /* 2136 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2137 */ 2138 if (error == EEXIST) 2139 error = 0; 2140 if (error == 0 && newvp == NULL) { 2141 struct nfsnode *np = NULL; 2142 2143 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2144 cnp->cn_cred, cnp->cn_proc, &np); 2145 if (error == 0) 2146 newvp = NFSTOV(np); 2147 } 2148 if (error) { 2149 if (newvp != NULL) 2150 vput(newvp); 2151 } else { 2152 *ap->a_vpp = newvp; 2153 } 2154 PNBUF_PUT(cnp->cn_pnbuf); 2155 VTONFS(dvp)->n_flag |= NMODIFIED; 2156 if (!wccflag) 2157 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2158 VN_KNOTE(dvp, NOTE_WRITE); 2159 vput(dvp); 2160 return (error); 2161 } 2162 2163 /* 2164 * nfs make dir call 2165 */ 2166 int 2167 nfs_mkdir(v) 2168 void *v; 2169 { 2170 struct vop_mkdir_args /* { 2171 struct vnode *a_dvp; 2172 struct vnode **a_vpp; 2173 struct componentname *a_cnp; 2174 struct vattr *a_vap; 2175 } */ *ap = v; 2176 struct vnode *dvp = ap->a_dvp; 2177 struct vattr *vap = ap->a_vap; 2178 struct componentname *cnp = ap->a_cnp; 2179 struct nfsv2_sattr *sp; 2180 u_int32_t *tl; 2181 caddr_t cp; 2182 int32_t t1, t2; 2183 int len; 2184 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2185 struct vnode *newvp = (struct vnode *)0; 2186 caddr_t bpos, dpos, cp2; 2187 int error = 0, wccflag = NFSV3_WCCRATTR; 2188 int gotvp = 0; 2189 struct mbuf *mreq, *mrep, *md, *mb; 2190 const int v3 = NFS_ISV3(dvp); 2191 2192 len = cnp->cn_namelen; 2193 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2194 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2195 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2196 nfsm_fhtom(dnp, v3); 2197 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2198 if (v3) { 2199 nfsm_v3attrbuild(vap, FALSE); 2200 } else { 2201 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2202 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2203 sp->sa_uid = nfs_xdrneg1; 2204 sp->sa_gid = nfs_xdrneg1; 2205 sp->sa_size = nfs_xdrneg1; 2206 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2207 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2208 } 2209 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred); 2210 if (!error) 2211 nfsm_mtofh(dvp, newvp, v3, gotvp); 2212 if (v3) 2213 nfsm_wcc_data(dvp, wccflag, 0); 2214 nfsm_reqdone; 2215 VTONFS(dvp)->n_flag |= NMODIFIED; 2216 if (!wccflag) 2217 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2218 /* 2219 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2220 * if we can succeed in looking up the directory. 2221 */ 2222 if (error == EEXIST || (!error && !gotvp)) { 2223 if (newvp) { 2224 vput(newvp); 2225 newvp = (struct vnode *)0; 2226 } 2227 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2228 cnp->cn_proc, &np); 2229 if (!error) { 2230 newvp = NFSTOV(np); 2231 if (newvp->v_type != VDIR) 2232 error = EEXIST; 2233 } 2234 } 2235 if (error) { 2236 if (newvp) 2237 vput(newvp); 2238 } else { 2239 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2240 if (cnp->cn_flags & MAKEENTRY) 2241 nfs_cache_enter(dvp, newvp, cnp); 2242 *ap->a_vpp = newvp; 2243 } 2244 PNBUF_PUT(cnp->cn_pnbuf); 2245 vput(dvp); 2246 return (error); 2247 } 2248 2249 /* 2250 * nfs remove directory call 2251 */ 2252 int 2253 nfs_rmdir(v) 2254 void *v; 2255 { 2256 struct vop_rmdir_args /* { 2257 struct vnode *a_dvp; 2258 struct vnode *a_vp; 2259 struct componentname *a_cnp; 2260 } */ *ap = v; 2261 struct vnode *vp = ap->a_vp; 2262 struct vnode *dvp = ap->a_dvp; 2263 struct componentname *cnp = ap->a_cnp; 2264 u_int32_t *tl; 2265 caddr_t cp; 2266 int32_t t1, t2; 2267 caddr_t bpos, dpos, cp2; 2268 int error = 0, wccflag = NFSV3_WCCRATTR; 2269 struct mbuf *mreq, *mrep, *md, *mb; 2270 const int v3 = NFS_ISV3(dvp); 2271 struct nfsnode *dnp; 2272 2273 if (dvp == vp) { 2274 vrele(dvp); 2275 vput(dvp); 2276 PNBUF_PUT(cnp->cn_pnbuf); 2277 return (EINVAL); 2278 } 2279 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2280 dnp = VTONFS(dvp); 2281 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2282 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2283 nfsm_fhtom(dnp, v3); 2284 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2285 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred); 2286 if (v3) 2287 nfsm_wcc_data(dvp, wccflag, 0); 2288 nfsm_reqdone; 2289 PNBUF_PUT(cnp->cn_pnbuf); 2290 VTONFS(dvp)->n_flag |= NMODIFIED; 2291 if (!wccflag) 2292 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2293 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2294 VN_KNOTE(vp, NOTE_DELETE); 2295 cache_purge(vp); 2296 vput(vp); 2297 vput(dvp); 2298 /* 2299 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2300 */ 2301 if (error == ENOENT) 2302 error = 0; 2303 return (error); 2304 } 2305 2306 /* 2307 * nfs readdir call 2308 */ 2309 int 2310 nfs_readdir(v) 2311 void *v; 2312 { 2313 struct vop_readdir_args /* { 2314 struct vnode *a_vp; 2315 struct uio *a_uio; 2316 struct ucred *a_cred; 2317 int *a_eofflag; 2318 off_t **a_cookies; 2319 int *a_ncookies; 2320 } */ *ap = v; 2321 struct vnode *vp = ap->a_vp; 2322 struct uio *uio = ap->a_uio; 2323 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2324 char *base = uio->uio_iov->iov_base; 2325 int tresid, error; 2326 size_t count, lost; 2327 struct dirent *dp; 2328 off_t *cookies = NULL; 2329 int ncookies = 0, nc; 2330 2331 if (vp->v_type != VDIR) 2332 return (EPERM); 2333 2334 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2335 count = uio->uio_resid - lost; 2336 if (count <= 0) 2337 return (EINVAL); 2338 2339 /* 2340 * Call nfs_bioread() to do the real work. 2341 */ 2342 tresid = uio->uio_resid = count; 2343 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2344 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2345 2346 if (!error && ap->a_cookies) { 2347 ncookies = count / 16; 2348 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2349 *ap->a_cookies = cookies; 2350 } 2351 2352 if (!error && uio->uio_resid == tresid) { 2353 uio->uio_resid += lost; 2354 nfsstats.direofcache_misses++; 2355 if (ap->a_cookies) 2356 *ap->a_ncookies = 0; 2357 *ap->a_eofflag = 1; 2358 return (0); 2359 } 2360 2361 if (!error && ap->a_cookies) { 2362 /* 2363 * Only the NFS server and emulations use cookies, and they 2364 * load the directory block into system space, so we can 2365 * just look at it directly. 2366 */ 2367 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2368 panic("nfs_readdir: lost in space"); 2369 for (nc = 0; ncookies-- && 2370 base < (char *)uio->uio_iov->iov_base; nc++){ 2371 dp = (struct dirent *) base; 2372 if (dp->d_reclen == 0) 2373 break; 2374 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2375 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2376 else 2377 *(cookies++) = NFS_GETCOOKIE(dp); 2378 base += dp->d_reclen; 2379 } 2380 uio->uio_resid += 2381 ((caddr_t)uio->uio_iov->iov_base - base); 2382 uio->uio_iov->iov_len += 2383 ((caddr_t)uio->uio_iov->iov_base - base); 2384 uio->uio_iov->iov_base = base; 2385 *ap->a_ncookies = nc; 2386 } 2387 2388 uio->uio_resid += lost; 2389 *ap->a_eofflag = 0; 2390 return (error); 2391 } 2392 2393 /* 2394 * Readdir rpc call. 2395 * Called from below the buffer cache by nfs_doio(). 2396 */ 2397 int 2398 nfs_readdirrpc(vp, uiop, cred) 2399 struct vnode *vp; 2400 struct uio *uiop; 2401 struct ucred *cred; 2402 { 2403 int len, left; 2404 struct dirent *dp = NULL; 2405 u_int32_t *tl; 2406 caddr_t cp; 2407 int32_t t1, t2; 2408 caddr_t bpos, dpos, cp2; 2409 struct mbuf *mreq, *mrep, *md, *mb; 2410 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2411 struct nfsnode *dnp = VTONFS(vp); 2412 u_quad_t fileno; 2413 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2414 int attrflag, nrpcs = 0, reclen; 2415 const int v3 = NFS_ISV3(vp); 2416 2417 #ifdef DIAGNOSTIC 2418 /* 2419 * Should be called from buffer cache, so only amount of 2420 * NFS_DIRBLKSIZ will be requested. 2421 */ 2422 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2423 panic("nfs readdirrpc bad uio"); 2424 #endif 2425 2426 /* 2427 * Loop around doing readdir rpc's of size nm_readdirsize 2428 * truncated to a multiple of NFS_DIRFRAGSIZ. 2429 * The stopping criteria is EOF or buffer full. 2430 */ 2431 while (more_dirs && bigenough) { 2432 /* 2433 * Heuristic: don't bother to do another RPC to further 2434 * fill up this block if there is not much room left. (< 50% 2435 * of the readdir RPC size). This wastes some buffer space 2436 * but can save up to 50% in RPC calls. 2437 */ 2438 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2439 bigenough = 0; 2440 break; 2441 } 2442 nfsstats.rpccnt[NFSPROC_READDIR]++; 2443 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2444 NFSX_READDIR(v3)); 2445 nfsm_fhtom(dnp, v3); 2446 if (v3) { 2447 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2448 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2449 txdr_swapcookie3(uiop->uio_offset, tl); 2450 } else { 2451 txdr_cookie3(uiop->uio_offset, tl); 2452 } 2453 tl += 2; 2454 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2455 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2456 } else { 2457 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2458 *tl++ = txdr_unsigned(uiop->uio_offset); 2459 } 2460 *tl = txdr_unsigned(nmp->nm_readdirsize); 2461 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred); 2462 nrpcs++; 2463 if (v3) { 2464 nfsm_postop_attr(vp, attrflag, 0); 2465 if (!error) { 2466 nfsm_dissect(tl, u_int32_t *, 2467 2 * NFSX_UNSIGNED); 2468 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2469 dnp->n_cookieverf.nfsuquad[1] = *tl; 2470 } else { 2471 m_freem(mrep); 2472 goto nfsmout; 2473 } 2474 } 2475 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2476 more_dirs = fxdr_unsigned(int, *tl); 2477 2478 /* loop thru the dir entries, doctoring them to 4bsd form */ 2479 while (more_dirs && bigenough) { 2480 if (v3) { 2481 nfsm_dissect(tl, u_int32_t *, 2482 3 * NFSX_UNSIGNED); 2483 fileno = fxdr_hyper(tl); 2484 len = fxdr_unsigned(int, *(tl + 2)); 2485 } else { 2486 nfsm_dissect(tl, u_int32_t *, 2487 2 * NFSX_UNSIGNED); 2488 fileno = fxdr_unsigned(u_quad_t, *tl++); 2489 len = fxdr_unsigned(int, *tl); 2490 } 2491 if (len <= 0 || len > NFS_MAXNAMLEN) { 2492 error = EBADRPC; 2493 m_freem(mrep); 2494 goto nfsmout; 2495 } 2496 tlen = nfsm_rndup(len); 2497 if (tlen == len) 2498 tlen += 4; /* To ensure null termination */ 2499 tlen += sizeof (off_t) + sizeof (int); 2500 reclen = ALIGN(tlen + DIRHDSIZ); 2501 tlen = reclen - DIRHDSIZ; 2502 left = NFS_DIRFRAGSIZ - blksiz; 2503 if (reclen > left) { 2504 dp->d_reclen += left; 2505 uiop->uio_iov->iov_base = 2506 (caddr_t)uiop->uio_iov->iov_base + left; 2507 uiop->uio_iov->iov_len -= left; 2508 uiop->uio_resid -= left; 2509 blksiz = 0; 2510 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2511 } 2512 if (reclen > uiop->uio_resid) 2513 bigenough = 0; 2514 if (bigenough) { 2515 dp = (struct dirent *)uiop->uio_iov->iov_base; 2516 dp->d_fileno = (int)fileno; 2517 dp->d_namlen = len; 2518 dp->d_reclen = reclen; 2519 dp->d_type = DT_UNKNOWN; 2520 blksiz += dp->d_reclen; 2521 if (blksiz == NFS_DIRFRAGSIZ) 2522 blksiz = 0; 2523 uiop->uio_resid -= DIRHDSIZ; 2524 uiop->uio_iov->iov_base = 2525 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ; 2526 uiop->uio_iov->iov_len -= DIRHDSIZ; 2527 nfsm_mtouio(uiop, len); 2528 cp = uiop->uio_iov->iov_base; 2529 tlen -= len; 2530 *cp = '\0'; /* null terminate */ 2531 uiop->uio_iov->iov_base = 2532 (caddr_t)uiop->uio_iov->iov_base + tlen; 2533 uiop->uio_iov->iov_len -= tlen; 2534 uiop->uio_resid -= tlen; 2535 } else 2536 nfsm_adv(nfsm_rndup(len)); 2537 if (v3) { 2538 nfsm_dissect(tl, u_int32_t *, 2539 3 * NFSX_UNSIGNED); 2540 } else { 2541 nfsm_dissect(tl, u_int32_t *, 2542 2 * NFSX_UNSIGNED); 2543 } 2544 if (bigenough) { 2545 if (v3) { 2546 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2547 uiop->uio_offset = 2548 fxdr_swapcookie3(tl); 2549 else 2550 uiop->uio_offset = 2551 fxdr_cookie3(tl); 2552 } 2553 else { 2554 uiop->uio_offset = 2555 fxdr_unsigned(off_t, *tl); 2556 } 2557 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2558 } 2559 if (v3) 2560 tl += 2; 2561 else 2562 tl++; 2563 more_dirs = fxdr_unsigned(int, *tl); 2564 } 2565 /* 2566 * If at end of rpc data, get the eof boolean 2567 */ 2568 if (!more_dirs) { 2569 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2570 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2571 } 2572 m_freem(mrep); 2573 } 2574 /* 2575 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2576 * by increasing d_reclen for the last record. 2577 */ 2578 if (blksiz > 0) { 2579 left = NFS_DIRFRAGSIZ - blksiz; 2580 dp->d_reclen += left; 2581 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2582 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2583 left; 2584 uiop->uio_iov->iov_len -= left; 2585 uiop->uio_resid -= left; 2586 } 2587 2588 /* 2589 * We are now either at the end of the directory or have filled the 2590 * block. 2591 */ 2592 if (bigenough) 2593 dnp->n_direofoffset = uiop->uio_offset; 2594 nfsmout: 2595 return (error); 2596 } 2597 2598 /* 2599 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2600 */ 2601 int 2602 nfs_readdirplusrpc(vp, uiop, cred) 2603 struct vnode *vp; 2604 struct uio *uiop; 2605 struct ucred *cred; 2606 { 2607 int len, left; 2608 struct dirent *dp = NULL; 2609 u_int32_t *tl; 2610 caddr_t cp; 2611 int32_t t1, t2; 2612 struct vnode *newvp; 2613 caddr_t bpos, dpos, cp2; 2614 struct mbuf *mreq, *mrep, *md, *mb; 2615 struct nameidata nami, *ndp = &nami; 2616 struct componentname *cnp = &ndp->ni_cnd; 2617 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2618 struct nfsnode *dnp = VTONFS(vp), *np; 2619 nfsfh_t *fhp; 2620 u_quad_t fileno; 2621 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2622 int attrflag, fhsize, nrpcs = 0, reclen; 2623 struct nfs_fattr fattr, *fp; 2624 2625 #ifdef DIAGNOSTIC 2626 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2627 panic("nfs readdirplusrpc bad uio"); 2628 #endif 2629 ndp->ni_dvp = vp; 2630 newvp = NULLVP; 2631 2632 /* 2633 * Loop around doing readdir rpc's of size nm_readdirsize 2634 * truncated to a multiple of NFS_DIRFRAGSIZ. 2635 * The stopping criteria is EOF or buffer full. 2636 */ 2637 while (more_dirs && bigenough) { 2638 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2639 bigenough = 0; 2640 break; 2641 } 2642 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2643 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2644 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2645 nfsm_fhtom(dnp, 1); 2646 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2647 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2648 txdr_swapcookie3(uiop->uio_offset, tl); 2649 } else { 2650 txdr_cookie3(uiop->uio_offset, tl); 2651 } 2652 tl += 2; 2653 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2654 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2655 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2656 *tl = txdr_unsigned(nmp->nm_rsize); 2657 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred); 2658 nfsm_postop_attr(vp, attrflag, 0); 2659 if (error) { 2660 m_freem(mrep); 2661 goto nfsmout; 2662 } 2663 nrpcs++; 2664 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2665 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2666 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2667 more_dirs = fxdr_unsigned(int, *tl); 2668 2669 /* loop thru the dir entries, doctoring them to 4bsd form */ 2670 while (more_dirs && bigenough) { 2671 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2672 fileno = fxdr_hyper(tl); 2673 len = fxdr_unsigned(int, *(tl + 2)); 2674 if (len <= 0 || len > NFS_MAXNAMLEN) { 2675 error = EBADRPC; 2676 m_freem(mrep); 2677 goto nfsmout; 2678 } 2679 tlen = nfsm_rndup(len); 2680 if (tlen == len) 2681 tlen += 4; /* To ensure null termination*/ 2682 tlen += sizeof (off_t) + sizeof (int); 2683 reclen = ALIGN(tlen + DIRHDSIZ); 2684 tlen = reclen - DIRHDSIZ; 2685 left = NFS_DIRFRAGSIZ - blksiz; 2686 if (reclen > left) { 2687 /* 2688 * DIRFRAGSIZ is aligned, no need to align 2689 * again here. 2690 */ 2691 dp->d_reclen += left; 2692 uiop->uio_iov->iov_base = 2693 (caddr_t)uiop->uio_iov->iov_base + left; 2694 uiop->uio_iov->iov_len -= left; 2695 uiop->uio_resid -= left; 2696 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2697 blksiz = 0; 2698 } 2699 if (reclen > uiop->uio_resid) 2700 bigenough = 0; 2701 if (bigenough) { 2702 dp = (struct dirent *)uiop->uio_iov->iov_base; 2703 dp->d_fileno = (int)fileno; 2704 dp->d_namlen = len; 2705 dp->d_reclen = reclen; 2706 dp->d_type = DT_UNKNOWN; 2707 blksiz += dp->d_reclen; 2708 if (blksiz == NFS_DIRFRAGSIZ) 2709 blksiz = 0; 2710 uiop->uio_resid -= DIRHDSIZ; 2711 uiop->uio_iov->iov_base = 2712 (caddr_t)uiop->uio_iov->iov_base + 2713 DIRHDSIZ; 2714 uiop->uio_iov->iov_len -= DIRHDSIZ; 2715 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2716 cnp->cn_namelen = len; 2717 nfsm_mtouio(uiop, len); 2718 cp = uiop->uio_iov->iov_base; 2719 tlen -= len; 2720 *cp = '\0'; 2721 uiop->uio_iov->iov_base = 2722 (caddr_t)uiop->uio_iov->iov_base + tlen; 2723 uiop->uio_iov->iov_len -= tlen; 2724 uiop->uio_resid -= tlen; 2725 } else 2726 nfsm_adv(nfsm_rndup(len)); 2727 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2728 if (bigenough) { 2729 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2730 uiop->uio_offset = 2731 fxdr_swapcookie3(tl); 2732 else 2733 uiop->uio_offset = 2734 fxdr_cookie3(tl); 2735 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2736 } 2737 tl += 2; 2738 2739 /* 2740 * Since the attributes are before the file handle 2741 * (sigh), we must skip over the attributes and then 2742 * come back and get them. 2743 */ 2744 attrflag = fxdr_unsigned(int, *tl); 2745 if (attrflag) { 2746 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2747 memcpy(&fattr, fp, NFSX_V3FATTR); 2748 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2749 doit = fxdr_unsigned(int, *tl); 2750 if (doit) { 2751 nfsm_getfh(fhp, fhsize, 1); 2752 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2753 VREF(vp); 2754 newvp = vp; 2755 np = dnp; 2756 } else { 2757 error = nfs_nget1(vp->v_mount, fhp, 2758 fhsize, &np, LK_NOWAIT); 2759 if (!error) 2760 newvp = NFSTOV(np); 2761 } 2762 if (!error) { 2763 const char *cp; 2764 2765 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2766 dp->d_type = 2767 IFTODT(VTTOIF(np->n_vattr->va_type)); 2768 ndp->ni_vp = newvp; 2769 cp = cnp->cn_nameptr + cnp->cn_namelen; 2770 cnp->cn_hash = 2771 namei_hash(cnp->cn_nameptr, &cp); 2772 if (cnp->cn_namelen <= NCHNAMLEN) { 2773 cache_purge1(ndp->ni_dvp, cnp, 0); 2774 nfs_cache_enter(ndp->ni_dvp, ndp->ni_vp, 2775 cnp); 2776 } 2777 } 2778 } 2779 } else { 2780 /* Just skip over the file handle */ 2781 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2782 i = fxdr_unsigned(int, *tl); 2783 nfsm_adv(nfsm_rndup(i)); 2784 } 2785 if (newvp != NULLVP) { 2786 if (newvp == vp) 2787 vrele(newvp); 2788 else 2789 vput(newvp); 2790 newvp = NULLVP; 2791 } 2792 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2793 more_dirs = fxdr_unsigned(int, *tl); 2794 } 2795 /* 2796 * If at end of rpc data, get the eof boolean 2797 */ 2798 if (!more_dirs) { 2799 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2800 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2801 } 2802 m_freem(mrep); 2803 } 2804 /* 2805 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2806 * by increasing d_reclen for the last record. 2807 */ 2808 if (blksiz > 0) { 2809 left = NFS_DIRFRAGSIZ - blksiz; 2810 dp->d_reclen += left; 2811 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2812 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2813 left; 2814 uiop->uio_iov->iov_len -= left; 2815 uiop->uio_resid -= left; 2816 } 2817 2818 /* 2819 * We are now either at the end of the directory or have filled the 2820 * block. 2821 */ 2822 if (bigenough) 2823 dnp->n_direofoffset = uiop->uio_offset; 2824 nfsmout: 2825 if (newvp != NULLVP) { 2826 if(newvp == vp) 2827 vrele(newvp); 2828 else 2829 vput(newvp); 2830 } 2831 return (error); 2832 } 2833 static char hextoasc[] = "0123456789abcdef"; 2834 2835 /* 2836 * Silly rename. To make the NFS filesystem that is stateless look a little 2837 * more like the "ufs" a remove of an active vnode is translated to a rename 2838 * to a funny looking filename that is removed by nfs_inactive on the 2839 * nfsnode. There is the potential for another process on a different client 2840 * to create the same funny name between the nfs_lookitup() fails and the 2841 * nfs_rename() completes, but... 2842 */ 2843 int 2844 nfs_sillyrename(dvp, vp, cnp) 2845 struct vnode *dvp, *vp; 2846 struct componentname *cnp; 2847 { 2848 struct sillyrename *sp; 2849 struct nfsnode *np; 2850 int error; 2851 short pid; 2852 2853 cache_purge(dvp); 2854 np = VTONFS(vp); 2855 #ifndef DIAGNOSTIC 2856 if (vp->v_type == VDIR) 2857 panic("nfs: sillyrename dir"); 2858 #endif 2859 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2860 M_NFSREQ, M_WAITOK); 2861 sp->s_cred = crdup(cnp->cn_cred); 2862 sp->s_dvp = dvp; 2863 VREF(dvp); 2864 2865 /* Fudge together a funny name */ 2866 pid = cnp->cn_proc->p_pid; 2867 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2868 sp->s_namlen = 12; 2869 sp->s_name[8] = hextoasc[pid & 0xf]; 2870 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf]; 2871 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf]; 2872 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf]; 2873 2874 /* Try lookitups until we get one that isn't there */ 2875 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2876 cnp->cn_proc, (struct nfsnode **)0) == 0) { 2877 sp->s_name[4]++; 2878 if (sp->s_name[4] > 'z') { 2879 error = EINVAL; 2880 goto bad; 2881 } 2882 } 2883 error = nfs_renameit(dvp, cnp, sp); 2884 if (error) 2885 goto bad; 2886 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2887 cnp->cn_proc, &np); 2888 np->n_sillyrename = sp; 2889 return (0); 2890 bad: 2891 vrele(sp->s_dvp); 2892 crfree(sp->s_cred); 2893 free((caddr_t)sp, M_NFSREQ); 2894 return (error); 2895 } 2896 2897 /* 2898 * Look up a file name and optionally either update the file handle or 2899 * allocate an nfsnode, depending on the value of npp. 2900 * npp == NULL --> just do the lookup 2901 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2902 * handled too 2903 * *npp != NULL --> update the file handle in the vnode 2904 */ 2905 int 2906 nfs_lookitup(dvp, name, len, cred, procp, npp) 2907 struct vnode *dvp; 2908 const char *name; 2909 int len; 2910 struct ucred *cred; 2911 struct proc *procp; 2912 struct nfsnode **npp; 2913 { 2914 u_int32_t *tl; 2915 caddr_t cp; 2916 int32_t t1, t2; 2917 struct vnode *newvp = (struct vnode *)0; 2918 struct nfsnode *np, *dnp = VTONFS(dvp); 2919 caddr_t bpos, dpos, cp2; 2920 int error = 0, fhlen, attrflag; 2921 struct mbuf *mreq, *mrep, *md, *mb; 2922 nfsfh_t *nfhp; 2923 const int v3 = NFS_ISV3(dvp); 2924 2925 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2926 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2927 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2928 nfsm_fhtom(dnp, v3); 2929 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2930 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred); 2931 if (npp && !error) { 2932 nfsm_getfh(nfhp, fhlen, v3); 2933 if (*npp) { 2934 np = *npp; 2935 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2936 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2937 np->n_fhp = &np->n_fh; 2938 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2939 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2940 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen); 2941 np->n_fhsize = fhlen; 2942 newvp = NFSTOV(np); 2943 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2944 VREF(dvp); 2945 newvp = dvp; 2946 } else { 2947 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2948 if (error) { 2949 m_freem(mrep); 2950 return (error); 2951 } 2952 newvp = NFSTOV(np); 2953 } 2954 if (v3) { 2955 nfsm_postop_attr(newvp, attrflag, 0); 2956 if (!attrflag && *npp == NULL) { 2957 m_freem(mrep); 2958 vput(newvp); 2959 return (ENOENT); 2960 } 2961 } else 2962 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2963 } 2964 nfsm_reqdone; 2965 if (npp && *npp == NULL) { 2966 if (error) { 2967 if (newvp) 2968 vput(newvp); 2969 } else 2970 *npp = np; 2971 } 2972 return (error); 2973 } 2974 2975 /* 2976 * Nfs Version 3 commit rpc 2977 */ 2978 int 2979 nfs_commit(vp, offset, cnt, procp) 2980 struct vnode *vp; 2981 off_t offset; 2982 uint32_t cnt; 2983 struct proc *procp; 2984 { 2985 caddr_t cp; 2986 u_int32_t *tl; 2987 int32_t t1, t2; 2988 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2989 caddr_t bpos, dpos, cp2; 2990 int error = 0, wccflag = NFSV3_WCCRATTR; 2991 struct mbuf *mreq, *mrep, *md, *mb; 2992 struct nfsnode *np; 2993 2994 KASSERT(NFS_ISV3(vp)); 2995 2996 #ifdef NFS_DEBUG_COMMIT 2997 printf("commit %lu - %lu\n", (unsigned long)offset, 2998 (unsigned long)(offset + cnt)); 2999 #endif 3000 3001 simple_lock(&nmp->nm_slock); 3002 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3003 simple_unlock(&nmp->nm_slock); 3004 return (0); 3005 } 3006 simple_unlock(&nmp->nm_slock); 3007 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3008 np = VTONFS(vp); 3009 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3010 nfsm_fhtom(np, 1); 3011 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3012 txdr_hyper(offset, tl); 3013 tl += 2; 3014 *tl = txdr_unsigned(cnt); 3015 nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred); 3016 nfsm_wcc_data(vp, wccflag, 0); 3017 if (!error) { 3018 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3019 simple_lock(&nmp->nm_slock); 3020 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3021 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3022 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3023 error = NFSERR_STALEWRITEVERF; 3024 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3025 } 3026 simple_unlock(&nmp->nm_slock); 3027 } 3028 nfsm_reqdone; 3029 return (error); 3030 } 3031 3032 /* 3033 * Kludge City.. 3034 * - make nfs_bmap() essentially a no-op that does no translation 3035 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3036 * (Maybe I could use the process's page mapping, but I was concerned that 3037 * Kernel Write might not be enabled and also figured copyout() would do 3038 * a lot more work than memcpy() and also it currently happens in the 3039 * context of the swapper process (2). 3040 */ 3041 int 3042 nfs_bmap(v) 3043 void *v; 3044 { 3045 struct vop_bmap_args /* { 3046 struct vnode *a_vp; 3047 daddr_t a_bn; 3048 struct vnode **a_vpp; 3049 daddr_t *a_bnp; 3050 int *a_runp; 3051 } */ *ap = v; 3052 struct vnode *vp = ap->a_vp; 3053 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3054 3055 if (ap->a_vpp != NULL) 3056 *ap->a_vpp = vp; 3057 if (ap->a_bnp != NULL) 3058 *ap->a_bnp = ap->a_bn << bshift; 3059 if (ap->a_runp != NULL) 3060 *ap->a_runp = 1024 * 1024; /* XXX */ 3061 return (0); 3062 } 3063 3064 /* 3065 * Strategy routine. 3066 * For async requests when nfsiod(s) are running, queue the request by 3067 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3068 * request. 3069 */ 3070 int 3071 nfs_strategy(v) 3072 void *v; 3073 { 3074 struct vop_strategy_args *ap = v; 3075 struct buf *bp = ap->a_bp; 3076 struct proc *p; 3077 int error = 0; 3078 3079 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3080 panic("nfs physio/async"); 3081 if (bp->b_flags & B_ASYNC) 3082 p = NULL; 3083 else 3084 p = curproc; /* XXX */ 3085 3086 /* 3087 * If the op is asynchronous and an i/o daemon is waiting 3088 * queue the request, wake it up and wait for completion 3089 * otherwise just do it ourselves. 3090 */ 3091 3092 if ((bp->b_flags & B_ASYNC) == 0 || 3093 nfs_asyncio(bp)) 3094 error = nfs_doio(bp, p); 3095 return (error); 3096 } 3097 3098 /* 3099 * fsync vnode op. Just call nfs_flush() with commit == 1. 3100 */ 3101 /* ARGSUSED */ 3102 int 3103 nfs_fsync(v) 3104 void *v; 3105 { 3106 struct vop_fsync_args /* { 3107 struct vnodeop_desc *a_desc; 3108 struct vnode * a_vp; 3109 struct ucred * a_cred; 3110 int a_flags; 3111 off_t offlo; 3112 off_t offhi; 3113 struct proc * a_p; 3114 } */ *ap = v; 3115 3116 struct vnode *vp = ap->a_vp; 3117 3118 if (vp->v_type != VREG) 3119 return 0; 3120 3121 return (nfs_flush(vp, ap->a_cred, 3122 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1)); 3123 } 3124 3125 /* 3126 * Flush all the data associated with a vnode. 3127 */ 3128 int 3129 nfs_flush(vp, cred, waitfor, p, commit) 3130 struct vnode *vp; 3131 struct ucred *cred; 3132 int waitfor; 3133 struct proc *p; 3134 int commit; 3135 { 3136 struct nfsnode *np = VTONFS(vp); 3137 int error; 3138 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3139 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3140 3141 simple_lock(&vp->v_interlock); 3142 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3143 if (np->n_flag & NWRITEERR) { 3144 error = np->n_error; 3145 np->n_flag &= ~NWRITEERR; 3146 } 3147 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3148 return (error); 3149 } 3150 3151 /* 3152 * Return POSIX pathconf information applicable to nfs. 3153 * 3154 * N.B. The NFS V2 protocol doesn't support this RPC. 3155 */ 3156 /* ARGSUSED */ 3157 int 3158 nfs_pathconf(v) 3159 void *v; 3160 { 3161 struct vop_pathconf_args /* { 3162 struct vnode *a_vp; 3163 int a_name; 3164 register_t *a_retval; 3165 } */ *ap = v; 3166 struct nfsv3_pathconf *pcp; 3167 struct vnode *vp = ap->a_vp; 3168 struct mbuf *mreq, *mrep, *md, *mb; 3169 int32_t t1, t2; 3170 u_int32_t *tl; 3171 caddr_t bpos, dpos, cp, cp2; 3172 int error = 0, attrflag; 3173 #ifndef NFS_V2_ONLY 3174 struct nfsmount *nmp; 3175 unsigned int l; 3176 u_int64_t maxsize; 3177 #endif 3178 const int v3 = NFS_ISV3(vp); 3179 struct nfsnode *np = VTONFS(vp); 3180 3181 switch (ap->a_name) { 3182 /* Names that can be resolved locally. */ 3183 case _PC_PIPE_BUF: 3184 *ap->a_retval = PIPE_BUF; 3185 break; 3186 case _PC_SYNC_IO: 3187 *ap->a_retval = 1; 3188 break; 3189 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3190 case _PC_LINK_MAX: 3191 case _PC_NAME_MAX: 3192 case _PC_CHOWN_RESTRICTED: 3193 case _PC_NO_TRUNC: 3194 if (!v3) { 3195 error = EINVAL; 3196 break; 3197 } 3198 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3199 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3200 nfsm_fhtom(np, 1); 3201 nfsm_request(np, NFSPROC_PATHCONF, 3202 curproc, curproc->p_ucred); /* XXX */ 3203 nfsm_postop_attr(vp, attrflag, 0); 3204 if (!error) { 3205 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3206 NFSX_V3PATHCONF); 3207 switch (ap->a_name) { 3208 case _PC_LINK_MAX: 3209 *ap->a_retval = 3210 fxdr_unsigned(register_t, pcp->pc_linkmax); 3211 break; 3212 case _PC_NAME_MAX: 3213 *ap->a_retval = 3214 fxdr_unsigned(register_t, pcp->pc_namemax); 3215 break; 3216 case _PC_CHOWN_RESTRICTED: 3217 *ap->a_retval = 3218 (pcp->pc_chownrestricted == nfs_true); 3219 break; 3220 case _PC_NO_TRUNC: 3221 *ap->a_retval = 3222 (pcp->pc_notrunc == nfs_true); 3223 break; 3224 } 3225 } 3226 nfsm_reqdone; 3227 break; 3228 case _PC_FILESIZEBITS: 3229 #ifndef NFS_V2_ONLY 3230 if (v3) { 3231 nmp = VFSTONFS(vp->v_mount); 3232 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3233 if ((error = nfs_fsinfo(nmp, vp, 3234 curproc->p_ucred, curproc)) != 0) /* XXX */ 3235 break; 3236 for (l = 0, maxsize = nmp->nm_maxfilesize; 3237 (maxsize >> l) > 0; l++) 3238 ; 3239 *ap->a_retval = l + 1; 3240 } else 3241 #endif 3242 { 3243 *ap->a_retval = 32; /* NFS V2 limitation */ 3244 } 3245 break; 3246 default: 3247 error = EINVAL; 3248 break; 3249 } 3250 3251 return (error); 3252 } 3253 3254 /* 3255 * NFS advisory byte-level locks. 3256 */ 3257 int 3258 nfs_advlock(v) 3259 void *v; 3260 { 3261 struct vop_advlock_args /* { 3262 struct vnode *a_vp; 3263 caddr_t a_id; 3264 int a_op; 3265 struct flock *a_fl; 3266 int a_flags; 3267 } */ *ap = v; 3268 struct nfsnode *np = VTONFS(ap->a_vp); 3269 3270 return lf_advlock(ap, &np->n_lockf, np->n_size); 3271 } 3272 3273 /* 3274 * Print out the contents of an nfsnode. 3275 */ 3276 int 3277 nfs_print(v) 3278 void *v; 3279 { 3280 struct vop_print_args /* { 3281 struct vnode *a_vp; 3282 } */ *ap = v; 3283 struct vnode *vp = ap->a_vp; 3284 struct nfsnode *np = VTONFS(vp); 3285 3286 printf("tag VT_NFS, fileid %ld fsid 0x%lx", 3287 np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3288 if (vp->v_type == VFIFO) 3289 fifo_printinfo(vp); 3290 printf("\n"); 3291 return (0); 3292 } 3293 3294 /* 3295 * NFS file truncation. 3296 */ 3297 int 3298 nfs_truncate(v) 3299 void *v; 3300 { 3301 #if 0 3302 struct vop_truncate_args /* { 3303 struct vnode *a_vp; 3304 off_t a_length; 3305 int a_flags; 3306 struct ucred *a_cred; 3307 struct proc *a_p; 3308 } */ *ap = v; 3309 #endif 3310 3311 /* Use nfs_setattr */ 3312 return (EOPNOTSUPP); 3313 } 3314 3315 /* 3316 * NFS update. 3317 */ 3318 int 3319 nfs_update(v) 3320 void *v; 3321 #if 0 3322 struct vop_update_args /* { 3323 struct vnode *a_vp; 3324 struct timespec *a_ta; 3325 struct timespec *a_tm; 3326 int a_waitfor; 3327 } */ *ap = v; 3328 #endif 3329 { 3330 3331 /* Use nfs_setattr */ 3332 return (EOPNOTSUPP); 3333 } 3334 3335 /* 3336 * Just call bwrite(). 3337 */ 3338 int 3339 nfs_bwrite(v) 3340 void *v; 3341 { 3342 struct vop_bwrite_args /* { 3343 struct vnode *a_bp; 3344 } */ *ap = v; 3345 3346 return (bwrite(ap->a_bp)); 3347 } 3348 3349 /* 3350 * nfs unlock wrapper. 3351 */ 3352 int 3353 nfs_unlock(void *v) 3354 { 3355 struct vop_unlock_args /* { 3356 struct vnode *a_vp; 3357 int a_flags; 3358 } */ *ap = v; 3359 struct vnode *vp = ap->a_vp; 3360 3361 /* 3362 * VOP_UNLOCK can be called by nfs_loadattrcache 3363 * with v_data == 0. 3364 */ 3365 if (VTONFS(vp)) { 3366 nfs_delayedtruncate(vp); 3367 } 3368 3369 return genfs_unlock(v); 3370 } 3371 3372 /* 3373 * nfs special file access vnode op. 3374 * Essentially just get vattr and then imitate iaccess() since the device is 3375 * local to the client. 3376 */ 3377 int 3378 nfsspec_access(v) 3379 void *v; 3380 { 3381 struct vop_access_args /* { 3382 struct vnode *a_vp; 3383 int a_mode; 3384 struct ucred *a_cred; 3385 struct proc *a_p; 3386 } */ *ap = v; 3387 struct vattr va; 3388 struct vnode *vp = ap->a_vp; 3389 int error; 3390 3391 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p); 3392 if (error) 3393 return (error); 3394 3395 /* 3396 * Disallow write attempts on filesystems mounted read-only; 3397 * unless the file is a socket, fifo, or a block or character 3398 * device resident on the filesystem. 3399 */ 3400 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3401 switch (vp->v_type) { 3402 case VREG: 3403 case VDIR: 3404 case VLNK: 3405 return (EROFS); 3406 default: 3407 break; 3408 } 3409 } 3410 3411 return (vaccess(va.va_type, va.va_mode, 3412 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3413 } 3414 3415 /* 3416 * Read wrapper for special devices. 3417 */ 3418 int 3419 nfsspec_read(v) 3420 void *v; 3421 { 3422 struct vop_read_args /* { 3423 struct vnode *a_vp; 3424 struct uio *a_uio; 3425 int a_ioflag; 3426 struct ucred *a_cred; 3427 } */ *ap = v; 3428 struct nfsnode *np = VTONFS(ap->a_vp); 3429 3430 /* 3431 * Set access flag. 3432 */ 3433 np->n_flag |= NACC; 3434 np->n_atim.tv_sec = time.tv_sec; 3435 np->n_atim.tv_nsec = time.tv_usec * 1000; 3436 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3437 } 3438 3439 /* 3440 * Write wrapper for special devices. 3441 */ 3442 int 3443 nfsspec_write(v) 3444 void *v; 3445 { 3446 struct vop_write_args /* { 3447 struct vnode *a_vp; 3448 struct uio *a_uio; 3449 int a_ioflag; 3450 struct ucred *a_cred; 3451 } */ *ap = v; 3452 struct nfsnode *np = VTONFS(ap->a_vp); 3453 3454 /* 3455 * Set update flag. 3456 */ 3457 np->n_flag |= NUPD; 3458 np->n_mtim.tv_sec = time.tv_sec; 3459 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3460 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3461 } 3462 3463 /* 3464 * Close wrapper for special devices. 3465 * 3466 * Update the times on the nfsnode then do device close. 3467 */ 3468 int 3469 nfsspec_close(v) 3470 void *v; 3471 { 3472 struct vop_close_args /* { 3473 struct vnode *a_vp; 3474 int a_fflag; 3475 struct ucred *a_cred; 3476 struct proc *a_p; 3477 } */ *ap = v; 3478 struct vnode *vp = ap->a_vp; 3479 struct nfsnode *np = VTONFS(vp); 3480 struct vattr vattr; 3481 3482 if (np->n_flag & (NACC | NUPD)) { 3483 np->n_flag |= NCHG; 3484 if (vp->v_usecount == 1 && 3485 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3486 VATTR_NULL(&vattr); 3487 if (np->n_flag & NACC) 3488 vattr.va_atime = np->n_atim; 3489 if (np->n_flag & NUPD) 3490 vattr.va_mtime = np->n_mtim; 3491 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3492 } 3493 } 3494 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3495 } 3496 3497 /* 3498 * Read wrapper for fifos. 3499 */ 3500 int 3501 nfsfifo_read(v) 3502 void *v; 3503 { 3504 struct vop_read_args /* { 3505 struct vnode *a_vp; 3506 struct uio *a_uio; 3507 int a_ioflag; 3508 struct ucred *a_cred; 3509 } */ *ap = v; 3510 struct nfsnode *np = VTONFS(ap->a_vp); 3511 3512 /* 3513 * Set access flag. 3514 */ 3515 np->n_flag |= NACC; 3516 np->n_atim.tv_sec = time.tv_sec; 3517 np->n_atim.tv_nsec = time.tv_usec * 1000; 3518 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3519 } 3520 3521 /* 3522 * Write wrapper for fifos. 3523 */ 3524 int 3525 nfsfifo_write(v) 3526 void *v; 3527 { 3528 struct vop_write_args /* { 3529 struct vnode *a_vp; 3530 struct uio *a_uio; 3531 int a_ioflag; 3532 struct ucred *a_cred; 3533 } */ *ap = v; 3534 struct nfsnode *np = VTONFS(ap->a_vp); 3535 3536 /* 3537 * Set update flag. 3538 */ 3539 np->n_flag |= NUPD; 3540 np->n_mtim.tv_sec = time.tv_sec; 3541 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3542 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3543 } 3544 3545 /* 3546 * Close wrapper for fifos. 3547 * 3548 * Update the times on the nfsnode then do fifo close. 3549 */ 3550 int 3551 nfsfifo_close(v) 3552 void *v; 3553 { 3554 struct vop_close_args /* { 3555 struct vnode *a_vp; 3556 int a_fflag; 3557 struct ucred *a_cred; 3558 struct proc *a_p; 3559 } */ *ap = v; 3560 struct vnode *vp = ap->a_vp; 3561 struct nfsnode *np = VTONFS(vp); 3562 struct vattr vattr; 3563 3564 if (np->n_flag & (NACC | NUPD)) { 3565 if (np->n_flag & NACC) { 3566 np->n_atim.tv_sec = time.tv_sec; 3567 np->n_atim.tv_nsec = time.tv_usec * 1000; 3568 } 3569 if (np->n_flag & NUPD) { 3570 np->n_mtim.tv_sec = time.tv_sec; 3571 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3572 } 3573 np->n_flag |= NCHG; 3574 if (vp->v_usecount == 1 && 3575 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3576 VATTR_NULL(&vattr); 3577 if (np->n_flag & NACC) 3578 vattr.va_atime = np->n_atim; 3579 if (np->n_flag & NUPD) 3580 vattr.va_mtime = np->n_mtim; 3581 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3582 } 3583 } 3584 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3585 } 3586
Indexes created Tue Oct 14 01:09:49 GMT 2025