nfs_vnops.c revision 1.180
1 /* $NetBSD: nfs_vnops.c,v 1.180 2003/09/26 11:51:53 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_vnops.c 8.19 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * vnode op calls for Sun NFS version 2 and 3 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.180 2003/09/26 11:51:53 yamt Exp $"); 43 44 #include "opt_nfs.h" 45 #include "opt_uvmhist.h" 46 47 #include <sys/param.h> 48 #include <sys/proc.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/vnode.h> 59 #include <sys/dirent.h> 60 #include <sys/fcntl.h> 61 #include <sys/hash.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/unistd.h> 65 66 #include <uvm/uvm_extern.h> 67 #include <uvm/uvm.h> 68 69 #include <miscfs/fifofs/fifo.h> 70 #include <miscfs/genfs/genfs.h> 71 #include <miscfs/specfs/specdev.h> 72 73 #include <nfs/rpcv2.h> 74 #include <nfs/nfsproto.h> 75 #include <nfs/nfs.h> 76 #include <nfs/nfsnode.h> 77 #include <nfs/nfsmount.h> 78 #include <nfs/xdr_subs.h> 79 #include <nfs/nfsm_subs.h> 80 #include <nfs/nqnfs.h> 81 #include <nfs/nfs_var.h> 82 83 #include <net/if.h> 84 #include <netinet/in.h> 85 #include <netinet/in_var.h> 86 87 /* 88 * Global vfs data structures for nfs 89 */ 90 int (**nfsv2_vnodeop_p) __P((void *)); 91 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 92 { &vop_default_desc, vn_default_error }, 93 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 94 { &vop_create_desc, nfs_create }, /* create */ 95 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 96 { &vop_open_desc, nfs_open }, /* open */ 97 { &vop_close_desc, nfs_close }, /* close */ 98 { &vop_access_desc, nfs_access }, /* access */ 99 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 100 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 101 { &vop_read_desc, nfs_read }, /* read */ 102 { &vop_write_desc, nfs_write }, /* write */ 103 { &vop_lease_desc, nfs_lease_check }, /* lease */ 104 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 105 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 106 { &vop_poll_desc, nfs_poll }, /* poll */ 107 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 108 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 109 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 110 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 111 { &vop_seek_desc, nfs_seek }, /* seek */ 112 { &vop_remove_desc, nfs_remove }, /* remove */ 113 { &vop_link_desc, nfs_link }, /* link */ 114 { &vop_rename_desc, nfs_rename }, /* rename */ 115 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 116 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 117 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 118 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 119 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 120 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 121 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 122 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 123 { &vop_lock_desc, nfs_lock }, /* lock */ 124 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 125 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 126 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 127 { &vop_print_desc, nfs_print }, /* print */ 128 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 129 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 130 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 131 { &vop_blkatoff_desc, nfs_blkatoff }, /* blkatoff */ 132 { &vop_valloc_desc, nfs_valloc }, /* valloc */ 133 { &vop_reallocblks_desc, nfs_reallocblks }, /* reallocblks */ 134 { &vop_vfree_desc, nfs_vfree }, /* vfree */ 135 { &vop_truncate_desc, nfs_truncate }, /* truncate */ 136 { &vop_update_desc, nfs_update }, /* update */ 137 { &vop_bwrite_desc, nfs_bwrite }, /* bwrite */ 138 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 139 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 140 { NULL, NULL } 141 }; 142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 144 145 /* 146 * Special device vnode ops 147 */ 148 int (**spec_nfsv2nodeop_p) __P((void *)); 149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 150 { &vop_default_desc, vn_default_error }, 151 { &vop_lookup_desc, spec_lookup }, /* lookup */ 152 { &vop_create_desc, spec_create }, /* create */ 153 { &vop_mknod_desc, spec_mknod }, /* mknod */ 154 { &vop_open_desc, spec_open }, /* open */ 155 { &vop_close_desc, nfsspec_close }, /* close */ 156 { &vop_access_desc, nfsspec_access }, /* access */ 157 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 158 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 159 { &vop_read_desc, nfsspec_read }, /* read */ 160 { &vop_write_desc, nfsspec_write }, /* write */ 161 { &vop_lease_desc, spec_lease_check }, /* lease */ 162 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 163 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 164 { &vop_poll_desc, spec_poll }, /* poll */ 165 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 166 { &vop_revoke_desc, spec_revoke }, /* revoke */ 167 { &vop_mmap_desc, spec_mmap }, /* mmap */ 168 { &vop_fsync_desc, spec_fsync }, /* fsync */ 169 { &vop_seek_desc, spec_seek }, /* seek */ 170 { &vop_remove_desc, spec_remove }, /* remove */ 171 { &vop_link_desc, spec_link }, /* link */ 172 { &vop_rename_desc, spec_rename }, /* rename */ 173 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 174 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 175 { &vop_symlink_desc, spec_symlink }, /* symlink */ 176 { &vop_readdir_desc, spec_readdir }, /* readdir */ 177 { &vop_readlink_desc, spec_readlink }, /* readlink */ 178 { &vop_abortop_desc, spec_abortop }, /* abortop */ 179 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 180 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 181 { &vop_lock_desc, nfs_lock }, /* lock */ 182 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 183 { &vop_bmap_desc, spec_bmap }, /* bmap */ 184 { &vop_strategy_desc, spec_strategy }, /* strategy */ 185 { &vop_print_desc, nfs_print }, /* print */ 186 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 187 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 188 { &vop_advlock_desc, spec_advlock }, /* advlock */ 189 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 190 { &vop_valloc_desc, spec_valloc }, /* valloc */ 191 { &vop_reallocblks_desc, spec_reallocblks }, /* reallocblks */ 192 { &vop_vfree_desc, spec_vfree }, /* vfree */ 193 { &vop_truncate_desc, spec_truncate }, /* truncate */ 194 { &vop_update_desc, nfs_update }, /* update */ 195 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 196 { &vop_getpages_desc, spec_getpages }, /* getpages */ 197 { &vop_putpages_desc, spec_putpages }, /* putpages */ 198 { NULL, NULL } 199 }; 200 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 201 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 202 203 int (**fifo_nfsv2nodeop_p) __P((void *)); 204 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 205 { &vop_default_desc, vn_default_error }, 206 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 207 { &vop_create_desc, fifo_create }, /* create */ 208 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 209 { &vop_open_desc, fifo_open }, /* open */ 210 { &vop_close_desc, nfsfifo_close }, /* close */ 211 { &vop_access_desc, nfsspec_access }, /* access */ 212 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 213 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 214 { &vop_read_desc, nfsfifo_read }, /* read */ 215 { &vop_write_desc, nfsfifo_write }, /* write */ 216 { &vop_lease_desc, fifo_lease_check }, /* lease */ 217 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 218 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 219 { &vop_poll_desc, fifo_poll }, /* poll */ 220 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 221 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 222 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 223 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 224 { &vop_seek_desc, fifo_seek }, /* seek */ 225 { &vop_remove_desc, fifo_remove }, /* remove */ 226 { &vop_link_desc, fifo_link }, /* link */ 227 { &vop_rename_desc, fifo_rename }, /* rename */ 228 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 229 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 230 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 231 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 232 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 233 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 234 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 235 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 236 { &vop_lock_desc, nfs_lock }, /* lock */ 237 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 238 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 239 { &vop_strategy_desc, genfs_badop }, /* strategy */ 240 { &vop_print_desc, nfs_print }, /* print */ 241 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 242 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 243 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 244 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 245 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 246 { &vop_reallocblks_desc, fifo_reallocblks }, /* reallocblks */ 247 { &vop_vfree_desc, fifo_vfree }, /* vfree */ 248 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 249 { &vop_update_desc, nfs_update }, /* update */ 250 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 251 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 252 { NULL, NULL } 253 }; 254 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 255 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 256 257 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *); 258 259 /* 260 * Global variables 261 */ 262 extern u_int32_t nfs_true, nfs_false; 263 extern u_int32_t nfs_xdrneg1; 264 extern const nfstype nfsv3_type[9]; 265 266 int nfs_numasync = 0; 267 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 268 269 /* 270 * nfs null call from vfs. 271 */ 272 int 273 nfs_null(vp, cred, procp) 274 struct vnode *vp; 275 struct ucred *cred; 276 struct proc *procp; 277 { 278 caddr_t bpos, dpos; 279 int error = 0; 280 struct mbuf *mreq, *mrep, *md, *mb; 281 struct nfsnode *np = VTONFS(vp); 282 283 nfsm_reqhead(np, NFSPROC_NULL, 0); 284 nfsm_request(np, NFSPROC_NULL, procp, cred); 285 nfsm_reqdone; 286 return (error); 287 } 288 289 /* 290 * nfs access vnode op. 291 * For nfs version 2, just return ok. File accesses may fail later. 292 * For nfs version 3, use the access rpc to check accessibility. If file modes 293 * are changed on the server, accesses might still fail later. 294 */ 295 int 296 nfs_access(v) 297 void *v; 298 { 299 struct vop_access_args /* { 300 struct vnode *a_vp; 301 int a_mode; 302 struct ucred *a_cred; 303 struct proc *a_p; 304 } */ *ap = v; 305 struct vnode *vp = ap->a_vp; 306 u_int32_t *tl; 307 caddr_t cp; 308 int32_t t1, t2; 309 caddr_t bpos, dpos, cp2; 310 int error = 0, attrflag, cachevalid; 311 struct mbuf *mreq, *mrep, *md, *mb; 312 u_int32_t mode, rmode; 313 const int v3 = NFS_ISV3(vp); 314 struct nfsnode *np = VTONFS(vp); 315 316 cachevalid = (np->n_accstamp != -1 && 317 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) && 318 np->n_accuid == ap->a_cred->cr_uid); 319 320 /* 321 * Check access cache first. If this request has been made for this 322 * uid shortly before, use the cached result. 323 */ 324 if (cachevalid) { 325 if (!np->n_accerror) { 326 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 327 return np->n_accerror; 328 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 329 return np->n_accerror; 330 } 331 332 /* 333 * For nfs v3, do an access rpc, otherwise you are stuck emulating 334 * ufs_access() locally using the vattr. This may not be correct, 335 * since the server may apply other access criteria such as 336 * client uid-->server uid mapping that we do not know about, but 337 * this is better than just returning anything that is lying about 338 * in the cache. 339 */ 340 if (v3) { 341 nfsstats.rpccnt[NFSPROC_ACCESS]++; 342 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 343 nfsm_fhtom(np, v3); 344 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 345 if (ap->a_mode & VREAD) 346 mode = NFSV3ACCESS_READ; 347 else 348 mode = 0; 349 if (vp->v_type != VDIR) { 350 if (ap->a_mode & VWRITE) 351 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 352 if (ap->a_mode & VEXEC) 353 mode |= NFSV3ACCESS_EXECUTE; 354 } else { 355 if (ap->a_mode & VWRITE) 356 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 357 NFSV3ACCESS_DELETE); 358 if (ap->a_mode & VEXEC) 359 mode |= NFSV3ACCESS_LOOKUP; 360 } 361 *tl = txdr_unsigned(mode); 362 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred); 363 nfsm_postop_attr(vp, attrflag, 0); 364 if (!error) { 365 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 366 rmode = fxdr_unsigned(u_int32_t, *tl); 367 /* 368 * The NFS V3 spec does not clarify whether or not 369 * the returned access bits can be a superset of 370 * the ones requested, so... 371 */ 372 if ((rmode & mode) != mode) 373 error = EACCES; 374 } 375 nfsm_reqdone; 376 } else 377 return (nfsspec_access(ap)); 378 /* 379 * Disallow write attempts on filesystems mounted read-only; 380 * unless the file is a socket, fifo, or a block or character 381 * device resident on the filesystem. 382 */ 383 if (!error && (ap->a_mode & VWRITE) && 384 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 385 switch (vp->v_type) { 386 case VREG: 387 case VDIR: 388 case VLNK: 389 error = EROFS; 390 default: 391 break; 392 } 393 } 394 395 if (!error || error == EACCES) { 396 /* 397 * If we got the same result as for a previous, 398 * different request, OR it in. Don't update 399 * the timestamp in that case. 400 */ 401 if (cachevalid && np->n_accstamp != -1 && 402 error == np->n_accerror) { 403 if (!error) 404 np->n_accmode |= ap->a_mode; 405 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 406 np->n_accmode = ap->a_mode; 407 } else { 408 np->n_accstamp = time.tv_sec; 409 np->n_accuid = ap->a_cred->cr_uid; 410 np->n_accmode = ap->a_mode; 411 np->n_accerror = error; 412 } 413 } 414 415 return (error); 416 } 417 418 /* 419 * nfs open vnode op 420 * Check to see if the type is ok 421 * and that deletion is not in progress. 422 * For paged in text files, you will need to flush the page cache 423 * if consistency is lost. 424 */ 425 /* ARGSUSED */ 426 int 427 nfs_open(v) 428 void *v; 429 { 430 struct vop_open_args /* { 431 struct vnode *a_vp; 432 int a_mode; 433 struct ucred *a_cred; 434 struct proc *a_p; 435 } */ *ap = v; 436 struct vnode *vp = ap->a_vp; 437 struct nfsnode *np = VTONFS(vp); 438 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 439 struct vattr vattr; 440 int error; 441 442 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 443 return (EACCES); 444 } 445 446 /* 447 * Initialize read and write creds here, for swapfiles 448 * and other paths that don't set the creds themselves. 449 */ 450 451 if (ap->a_mode & FREAD) { 452 if (np->n_rcred) { 453 crfree(np->n_rcred); 454 } 455 np->n_rcred = ap->a_cred; 456 crhold(np->n_rcred); 457 } 458 if (ap->a_mode & FWRITE) { 459 if (np->n_wcred) { 460 crfree(np->n_wcred); 461 } 462 np->n_wcred = ap->a_cred; 463 crhold(np->n_wcred); 464 } 465 466 #ifndef NFS_V2_ONLY 467 /* 468 * Get a valid lease. If cached data is stale, flush it. 469 */ 470 if (nmp->nm_flag & NFSMNT_NQNFS) { 471 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 472 do { 473 error = nqnfs_getlease(vp, ND_READ, ap->a_cred, 474 ap->a_p); 475 } while (error == NQNFS_EXPIRED); 476 if (error) 477 return (error); 478 if (np->n_lrev != np->n_brev || 479 (np->n_flag & NQNFSNONCACHE)) { 480 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 481 ap->a_p, 1)) == EINTR) 482 return (error); 483 np->n_brev = np->n_lrev; 484 } 485 } 486 } else 487 #endif 488 { 489 if (np->n_flag & NMODIFIED) { 490 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 491 ap->a_p, 1)) == EINTR) 492 return (error); 493 np->n_attrstamp = 0; 494 if (vp->v_type == VDIR) { 495 nfs_invaldircache(vp, 0); 496 np->n_direofoffset = 0; 497 } 498 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p); 499 if (error) 500 return (error); 501 np->n_mtime = vattr.va_mtime; 502 } else { 503 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p); 504 if (error) 505 return (error); 506 if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) { 507 if (vp->v_type == VDIR) { 508 nfs_invaldircache(vp, 0); 509 np->n_direofoffset = 0; 510 } 511 if ((error = nfs_vinvalbuf(vp, V_SAVE, 512 ap->a_cred, ap->a_p, 1)) == EINTR) 513 return (error); 514 np->n_mtime = vattr.va_mtime; 515 } 516 } 517 } 518 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) 519 np->n_attrstamp = 0; /* For Open/Close consistency */ 520 return (0); 521 } 522 523 /* 524 * nfs close vnode op 525 * What an NFS client should do upon close after writing is a debatable issue. 526 * Most NFS clients push delayed writes to the server upon close, basically for 527 * two reasons: 528 * 1 - So that any write errors may be reported back to the client process 529 * doing the close system call. By far the two most likely errors are 530 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 531 * 2 - To put a worst case upper bound on cache inconsistency between 532 * multiple clients for the file. 533 * There is also a consistency problem for Version 2 of the protocol w.r.t. 534 * not being able to tell if other clients are writing a file concurrently, 535 * since there is no way of knowing if the changed modify time in the reply 536 * is only due to the write for this client. 537 * (NFS Version 3 provides weak cache consistency data in the reply that 538 * should be sufficient to detect and handle this case.) 539 * 540 * The current code does the following: 541 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 542 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 543 * or commit them (this satisfies 1 and 2 except for the 544 * case where the server crashes after this close but 545 * before the commit RPC, which is felt to be "good 546 * enough". Changing the last argument to nfs_flush() to 547 * a 1 would force a commit operation, if it is felt a 548 * commit is necessary now. 549 * for NQNFS - do nothing now, since 2 is dealt with via leases and 550 * 1 should be dealt with via an fsync() system call for 551 * cases where write errors are important. 552 */ 553 /* ARGSUSED */ 554 int 555 nfs_close(v) 556 void *v; 557 { 558 struct vop_close_args /* { 559 struct vnodeop_desc *a_desc; 560 struct vnode *a_vp; 561 int a_fflag; 562 struct ucred *a_cred; 563 struct proc *a_p; 564 } */ *ap = v; 565 struct vnode *vp = ap->a_vp; 566 struct nfsnode *np = VTONFS(vp); 567 int error = 0; 568 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 569 570 if (vp->v_type == VREG) { 571 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 572 (np->n_flag & NMODIFIED)) { 573 if (NFS_ISV3(vp)) { 574 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0); 575 np->n_flag &= ~NMODIFIED; 576 } else 577 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1); 578 np->n_attrstamp = 0; 579 } 580 if (np->n_flag & NWRITEERR) { 581 np->n_flag &= ~NWRITEERR; 582 error = np->n_error; 583 } 584 } 585 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 586 return (error); 587 } 588 589 /* 590 * nfs getattr call from vfs. 591 */ 592 int 593 nfs_getattr(v) 594 void *v; 595 { 596 struct vop_getattr_args /* { 597 struct vnode *a_vp; 598 struct vattr *a_vap; 599 struct ucred *a_cred; 600 struct proc *a_p; 601 } */ *ap = v; 602 struct vnode *vp = ap->a_vp; 603 struct nfsnode *np = VTONFS(vp); 604 caddr_t cp; 605 u_int32_t *tl; 606 int32_t t1, t2; 607 caddr_t bpos, dpos; 608 int error = 0; 609 struct mbuf *mreq, *mrep, *md, *mb; 610 const int v3 = NFS_ISV3(vp); 611 612 /* 613 * Update local times for special files. 614 */ 615 if (np->n_flag & (NACC | NUPD)) 616 np->n_flag |= NCHG; 617 /* 618 * First look in the cache. 619 */ 620 if (nfs_getattrcache(vp, ap->a_vap) == 0) 621 return (0); 622 nfsstats.rpccnt[NFSPROC_GETATTR]++; 623 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 624 nfsm_fhtom(np, v3); 625 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred); 626 if (!error) { 627 nfsm_loadattr(vp, ap->a_vap, 0); 628 if (vp->v_type == VDIR && 629 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 630 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 631 } 632 nfsm_reqdone; 633 return (error); 634 } 635 636 /* 637 * nfs setattr call. 638 */ 639 int 640 nfs_setattr(v) 641 void *v; 642 { 643 struct vop_setattr_args /* { 644 struct vnodeop_desc *a_desc; 645 struct vnode *a_vp; 646 struct vattr *a_vap; 647 struct ucred *a_cred; 648 struct proc *a_p; 649 } */ *ap = v; 650 struct vnode *vp = ap->a_vp; 651 struct nfsnode *np = VTONFS(vp); 652 struct vattr *vap = ap->a_vap; 653 int error = 0; 654 u_quad_t tsize = 0; 655 656 /* 657 * Setting of flags is not supported. 658 */ 659 if (vap->va_flags != VNOVAL) 660 return (EOPNOTSUPP); 661 662 /* 663 * Disallow write attempts if the filesystem is mounted read-only. 664 */ 665 if ((vap->va_uid != (uid_t)VNOVAL || 666 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 667 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 668 (vp->v_mount->mnt_flag & MNT_RDONLY)) 669 return (EROFS); 670 if (vap->va_size != VNOVAL) { 671 switch (vp->v_type) { 672 case VDIR: 673 return (EISDIR); 674 case VCHR: 675 case VBLK: 676 case VSOCK: 677 case VFIFO: 678 if (vap->va_mtime.tv_sec == VNOVAL && 679 vap->va_atime.tv_sec == VNOVAL && 680 vap->va_mode == (mode_t)VNOVAL && 681 vap->va_uid == (uid_t)VNOVAL && 682 vap->va_gid == (gid_t)VNOVAL) 683 return (0); 684 vap->va_size = VNOVAL; 685 break; 686 default: 687 /* 688 * Disallow write attempts if the filesystem is 689 * mounted read-only. 690 */ 691 if (vp->v_mount->mnt_flag & MNT_RDONLY) 692 return (EROFS); 693 uvm_vnp_setsize(vp, vap->va_size); 694 tsize = np->n_size; 695 np->n_size = vap->va_size; 696 if (vap->va_size == 0) 697 error = nfs_vinvalbuf(vp, 0, 698 ap->a_cred, ap->a_p, 1); 699 else 700 error = nfs_vinvalbuf(vp, V_SAVE, 701 ap->a_cred, ap->a_p, 1); 702 if (error) { 703 uvm_vnp_setsize(vp, tsize); 704 return (error); 705 } 706 np->n_vattr->va_size = vap->va_size; 707 } 708 } else if ((vap->va_mtime.tv_sec != VNOVAL || 709 vap->va_atime.tv_sec != VNOVAL) && 710 vp->v_type == VREG && 711 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 712 ap->a_p, 1)) == EINTR) 713 return (error); 714 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p); 715 if (error && vap->va_size != VNOVAL) { 716 np->n_size = np->n_vattr->va_size = tsize; 717 uvm_vnp_setsize(vp, np->n_size); 718 } 719 VN_KNOTE(vp, NOTE_ATTRIB); 720 return (error); 721 } 722 723 /* 724 * Do an nfs setattr rpc. 725 */ 726 int 727 nfs_setattrrpc(vp, vap, cred, procp) 728 struct vnode *vp; 729 struct vattr *vap; 730 struct ucred *cred; 731 struct proc *procp; 732 { 733 struct nfsv2_sattr *sp; 734 caddr_t cp; 735 int32_t t1, t2; 736 caddr_t bpos, dpos, cp2; 737 u_int32_t *tl; 738 int error = 0, wccflag = NFSV3_WCCRATTR; 739 struct mbuf *mreq, *mrep, *md, *mb; 740 const int v3 = NFS_ISV3(vp); 741 struct nfsnode *np = VTONFS(vp); 742 743 nfsstats.rpccnt[NFSPROC_SETATTR]++; 744 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 745 nfsm_fhtom(np, v3); 746 if (v3) { 747 nfsm_v3attrbuild(vap, TRUE); 748 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 749 *tl = nfs_false; 750 } else { 751 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 752 if (vap->va_mode == (mode_t)VNOVAL) 753 sp->sa_mode = nfs_xdrneg1; 754 else 755 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 756 if (vap->va_uid == (uid_t)VNOVAL) 757 sp->sa_uid = nfs_xdrneg1; 758 else 759 sp->sa_uid = txdr_unsigned(vap->va_uid); 760 if (vap->va_gid == (gid_t)VNOVAL) 761 sp->sa_gid = nfs_xdrneg1; 762 else 763 sp->sa_gid = txdr_unsigned(vap->va_gid); 764 sp->sa_size = txdr_unsigned(vap->va_size); 765 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 766 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 767 } 768 nfsm_request(np, NFSPROC_SETATTR, procp, cred); 769 if (v3) { 770 nfsm_wcc_data(vp, wccflag, 0); 771 } else 772 nfsm_loadattr(vp, (struct vattr *)0, 0); 773 nfsm_reqdone; 774 return (error); 775 } 776 777 /* 778 * nfs lookup call, one step at a time... 779 * First look in cache 780 * If not found, unlock the directory nfsnode and do the rpc 781 * 782 * This code is full of lock/unlock statements and checks, because 783 * we continue after cache_lookup has finished (we need to check 784 * with the attr cache and do an rpc if it has timed out). This means 785 * that the locking effects of cache_lookup have to be taken into 786 * account. 787 */ 788 int 789 nfs_lookup(v) 790 void *v; 791 { 792 struct vop_lookup_args /* { 793 struct vnodeop_desc *a_desc; 794 struct vnode *a_dvp; 795 struct vnode **a_vpp; 796 struct componentname *a_cnp; 797 } */ *ap = v; 798 struct componentname *cnp = ap->a_cnp; 799 struct vnode *dvp = ap->a_dvp; 800 struct vnode **vpp = ap->a_vpp; 801 int flags; 802 struct vnode *newvp; 803 u_int32_t *tl; 804 caddr_t cp; 805 int32_t t1, t2; 806 caddr_t bpos, dpos, cp2; 807 struct mbuf *mreq, *mrep, *md, *mb; 808 long len; 809 nfsfh_t *fhp; 810 struct nfsnode *np; 811 int lockparent, wantparent, error = 0, attrflag, fhsize; 812 const int v3 = NFS_ISV3(dvp); 813 814 cnp->cn_flags &= ~PDIRUNLOCK; 815 flags = cnp->cn_flags; 816 817 *vpp = NULLVP; 818 newvp = NULLVP; 819 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 820 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 821 return (EROFS); 822 if (dvp->v_type != VDIR) 823 return (ENOTDIR); 824 825 lockparent = flags & LOCKPARENT; 826 wantparent = flags & (LOCKPARENT|WANTPARENT); 827 np = VTONFS(dvp); 828 829 /* 830 * Before tediously performing a linear scan of the directory, 831 * check the name cache to see if the directory/name pair 832 * we are looking for is known already. 833 * If the directory/name pair is found in the name cache, 834 * we have to ensure the directory has not changed from 835 * the time the cache entry has been created. If it has, 836 * the cache entry has to be ignored. 837 */ 838 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) { 839 struct vattr vattr; 840 int err2; 841 842 if (error && error != ENOENT) { 843 *vpp = NULLVP; 844 return error; 845 } 846 847 if (cnp->cn_flags & PDIRUNLOCK) { 848 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 849 if (err2 != 0) { 850 *vpp = NULLVP; 851 return err2; 852 } 853 cnp->cn_flags &= ~PDIRUNLOCK; 854 } 855 856 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc); 857 if (err2 != 0) { 858 if (error == 0) { 859 if (*vpp != dvp) 860 vput(*vpp); 861 else 862 vrele(*vpp); 863 } 864 *vpp = NULLVP; 865 return err2; 866 } 867 868 if (error == ENOENT) { 869 if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred, 870 cnp->cn_proc) && timespeccmp(&vattr.va_mtime, 871 &VTONFS(dvp)->n_nctime, ==)) 872 return ENOENT; 873 cache_purge(dvp); 874 timespecclear(&np->n_nctime); 875 goto dorpc; 876 } 877 878 newvp = *vpp; 879 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc) 880 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) 881 { 882 nfsstats.lookupcache_hits++; 883 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 884 cnp->cn_flags |= SAVENAME; 885 if ((!lockparent || !(flags & ISLASTCN)) && 886 newvp != dvp) 887 VOP_UNLOCK(dvp, 0); 888 return (0); 889 } 890 cache_purge(newvp); 891 if (newvp != dvp) 892 vput(newvp); 893 else 894 vrele(newvp); 895 *vpp = NULLVP; 896 } 897 dorpc: 898 error = 0; 899 newvp = NULLVP; 900 nfsstats.lookupcache_misses++; 901 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 902 len = cnp->cn_namelen; 903 nfsm_reqhead(np, NFSPROC_LOOKUP, 904 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 905 nfsm_fhtom(np, v3); 906 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 907 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred); 908 if (error) { 909 nfsm_postop_attr(dvp, attrflag, 0); 910 m_freem(mrep); 911 goto nfsmout; 912 } 913 nfsm_getfh(fhp, fhsize, v3); 914 915 /* 916 * Handle RENAME case... 917 */ 918 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) { 919 if (NFS_CMPFH(np, fhp, fhsize)) { 920 m_freem(mrep); 921 return (EISDIR); 922 } 923 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 924 if (error) { 925 m_freem(mrep); 926 return error; 927 } 928 newvp = NFSTOV(np); 929 if (v3) { 930 nfsm_postop_attr(newvp, attrflag, 0); 931 nfsm_postop_attr(dvp, attrflag, 0); 932 } else 933 nfsm_loadattr(newvp, (struct vattr *)0, 0); 934 *vpp = newvp; 935 m_freem(mrep); 936 cnp->cn_flags |= SAVENAME; 937 if (!lockparent) { 938 VOP_UNLOCK(dvp, 0); 939 cnp->cn_flags |= PDIRUNLOCK; 940 } 941 return (0); 942 } 943 944 /* 945 * The postop attr handling is duplicated for each if case, 946 * because it should be done while dvp is locked (unlocking 947 * dvp is different for each case). 948 */ 949 950 if (NFS_CMPFH(np, fhp, fhsize)) { 951 /* 952 * "." lookup 953 */ 954 VREF(dvp); 955 newvp = dvp; 956 if (v3) { 957 nfsm_postop_attr(newvp, attrflag, 0); 958 nfsm_postop_attr(dvp, attrflag, 0); 959 } else 960 nfsm_loadattr(newvp, (struct vattr *)0, 0); 961 } else if (flags & ISDOTDOT) { 962 /* 963 * ".." lookup 964 */ 965 VOP_UNLOCK(dvp, 0); 966 cnp->cn_flags |= PDIRUNLOCK; 967 968 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 969 if (error) { 970 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 971 cnp->cn_flags &= ~PDIRUNLOCK; 972 m_freem(mrep); 973 return error; 974 } 975 newvp = NFSTOV(np); 976 977 if (v3) { 978 nfsm_postop_attr(newvp, attrflag, 0); 979 nfsm_postop_attr(dvp, attrflag, 0); 980 } else 981 nfsm_loadattr(newvp, (struct vattr *)0, 0); 982 983 if (lockparent && (flags & ISLASTCN)) { 984 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) { 985 m_freem(mrep); 986 vput(newvp); 987 return error; 988 } 989 cnp->cn_flags &= ~PDIRUNLOCK; 990 } 991 } else { 992 /* 993 * Other lookups. 994 */ 995 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 996 if (error) { 997 m_freem(mrep); 998 return error; 999 } 1000 newvp = NFSTOV(np); 1001 if (v3) { 1002 nfsm_postop_attr(newvp, attrflag, 0); 1003 nfsm_postop_attr(dvp, attrflag, 0); 1004 } else 1005 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1006 if (!lockparent || !(flags & ISLASTCN)) { 1007 VOP_UNLOCK(dvp, 0); 1008 cnp->cn_flags |= PDIRUNLOCK; 1009 } 1010 } 1011 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1012 cnp->cn_flags |= SAVENAME; 1013 if ((cnp->cn_flags & MAKEENTRY) && 1014 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1015 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 1016 cache_enter(dvp, newvp, cnp); 1017 } 1018 *vpp = newvp; 1019 nfsm_reqdone; 1020 if (error) { 1021 /* 1022 * We get here only because of errors returned by 1023 * the RPC. Otherwise we'll have returned above 1024 * (the nfsm_* macros will jump to nfsm_reqdone 1025 * on error). 1026 */ 1027 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1028 cnp->cn_nameiop != CREATE) { 1029 if (!timespecisset(&VTONFS(dvp)->n_nctime)) 1030 VTONFS(dvp)->n_nctime = 1031 VTONFS(dvp)->n_vattr->va_mtime; 1032 1033 cache_enter(dvp, NULL, cnp); 1034 } 1035 if (newvp != NULLVP) { 1036 vrele(newvp); 1037 if (newvp != dvp) 1038 VOP_UNLOCK(newvp, 0); 1039 } 1040 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1041 (flags & ISLASTCN) && error == ENOENT) { 1042 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1043 error = EROFS; 1044 else 1045 error = EJUSTRETURN; 1046 } 1047 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1048 cnp->cn_flags |= SAVENAME; 1049 *vpp = NULL; 1050 } 1051 1052 return error; 1053 } 1054 1055 /* 1056 * nfs read call. 1057 * Just call nfs_bioread() to do the work. 1058 */ 1059 int 1060 nfs_read(v) 1061 void *v; 1062 { 1063 struct vop_read_args /* { 1064 struct vnode *a_vp; 1065 struct uio *a_uio; 1066 int a_ioflag; 1067 struct ucred *a_cred; 1068 } */ *ap = v; 1069 struct vnode *vp = ap->a_vp; 1070 1071 if (vp->v_type != VREG) 1072 return (EPERM); 1073 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1074 } 1075 1076 /* 1077 * nfs readlink call 1078 */ 1079 int 1080 nfs_readlink(v) 1081 void *v; 1082 { 1083 struct vop_readlink_args /* { 1084 struct vnode *a_vp; 1085 struct uio *a_uio; 1086 struct ucred *a_cred; 1087 } */ *ap = v; 1088 struct vnode *vp = ap->a_vp; 1089 1090 if (vp->v_type != VLNK) 1091 return (EPERM); 1092 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1093 } 1094 1095 /* 1096 * Do a readlink rpc. 1097 * Called by nfs_doio() from below the buffer cache. 1098 */ 1099 int 1100 nfs_readlinkrpc(vp, uiop, cred) 1101 struct vnode *vp; 1102 struct uio *uiop; 1103 struct ucred *cred; 1104 { 1105 u_int32_t *tl; 1106 caddr_t cp; 1107 int32_t t1, t2; 1108 caddr_t bpos, dpos, cp2; 1109 int error = 0, attrflag; 1110 uint32_t len; 1111 struct mbuf *mreq, *mrep, *md, *mb; 1112 const int v3 = NFS_ISV3(vp); 1113 struct nfsnode *np = VTONFS(vp); 1114 1115 nfsstats.rpccnt[NFSPROC_READLINK]++; 1116 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1117 nfsm_fhtom(np, v3); 1118 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred); 1119 if (v3) 1120 nfsm_postop_attr(vp, attrflag, 0); 1121 if (!error) { 1122 if (v3) { 1123 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1124 len = fxdr_unsigned(uint32_t, *tl); 1125 if (len > MAXPATHLEN) { 1126 /* 1127 * this pathname is too long for us. 1128 */ 1129 m_freem(mrep); 1130 /* Solaris returns EINVAL. should we follow? */ 1131 error = ENAMETOOLONG; 1132 goto nfsmout; 1133 } 1134 } else { 1135 nfsm_strsiz(len, NFS_MAXPATHLEN); 1136 } 1137 nfsm_mtouio(uiop, len); 1138 } 1139 nfsm_reqdone; 1140 return (error); 1141 } 1142 1143 /* 1144 * nfs read rpc call 1145 * Ditto above 1146 */ 1147 int 1148 nfs_readrpc(vp, uiop) 1149 struct vnode *vp; 1150 struct uio *uiop; 1151 { 1152 u_int32_t *tl; 1153 caddr_t cp; 1154 int32_t t1, t2; 1155 caddr_t bpos, dpos, cp2; 1156 struct mbuf *mreq, *mrep, *md, *mb; 1157 struct nfsmount *nmp; 1158 int error = 0, len, retlen, tsiz, eof, attrflag; 1159 const int v3 = NFS_ISV3(vp); 1160 struct nfsnode *np = VTONFS(vp); 1161 1162 #ifndef nolint 1163 eof = 0; 1164 #endif 1165 nmp = VFSTONFS(vp->v_mount); 1166 tsiz = uiop->uio_resid; 1167 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1168 return (EFBIG); 1169 while (tsiz > 0) { 1170 nfsstats.rpccnt[NFSPROC_READ]++; 1171 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1172 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1173 nfsm_fhtom(np, v3); 1174 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1175 if (v3) { 1176 txdr_hyper(uiop->uio_offset, tl); 1177 *(tl + 2) = txdr_unsigned(len); 1178 } else { 1179 *tl++ = txdr_unsigned(uiop->uio_offset); 1180 *tl++ = txdr_unsigned(len); 1181 *tl = 0; 1182 } 1183 nfsm_request(np, NFSPROC_READ, uiop->uio_procp, 1184 VTONFS(vp)->n_rcred); 1185 if (v3) { 1186 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1187 if (error) { 1188 m_freem(mrep); 1189 goto nfsmout; 1190 } 1191 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1192 eof = fxdr_unsigned(int, *(tl + 1)); 1193 } else 1194 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1195 nfsm_strsiz(retlen, nmp->nm_rsize); 1196 nfsm_mtouio(uiop, retlen); 1197 m_freem(mrep); 1198 tsiz -= retlen; 1199 if (v3) { 1200 if (eof || retlen == 0) 1201 tsiz = 0; 1202 } else if (retlen < len) 1203 tsiz = 0; 1204 } 1205 nfsmout: 1206 return (error); 1207 } 1208 1209 struct nfs_writerpc_context { 1210 struct simplelock nwc_slock; 1211 volatile int nwc_mbufcount; 1212 }; 1213 1214 /* 1215 * free mbuf used to refer protected pages while write rpc call. 1216 * called at splvm. 1217 */ 1218 static void 1219 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 1220 { 1221 struct nfs_writerpc_context *ctx = arg; 1222 1223 KASSERT(m != NULL); 1224 KASSERT(ctx != NULL); 1225 pool_cache_put(&mbpool_cache, m); 1226 simple_lock(&ctx->nwc_slock); 1227 if (--ctx->nwc_mbufcount == 0) { 1228 wakeup(ctx); 1229 } 1230 simple_unlock(&ctx->nwc_slock); 1231 } 1232 1233 /* 1234 * nfs write call 1235 */ 1236 int 1237 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1238 struct vnode *vp; 1239 struct uio *uiop; 1240 int *iomode; 1241 boolean_t pageprotected; 1242 boolean_t *stalewriteverfp; 1243 { 1244 u_int32_t *tl; 1245 caddr_t cp; 1246 int32_t t1, t2; 1247 caddr_t bpos, dpos, cp2; 1248 struct mbuf *mreq, *mrep, *md, *mb; 1249 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1250 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1251 const int v3 = NFS_ISV3(vp); 1252 int committed = NFSV3WRITE_FILESYNC; 1253 struct nfsnode *np = VTONFS(vp); 1254 struct nfs_writerpc_context ctx; 1255 int s; 1256 struct lwp *l; 1257 size_t origresid; 1258 1259 simple_lock_init(&ctx.nwc_slock); 1260 ctx.nwc_mbufcount = 1; 1261 1262 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1263 panic("writerpc readonly vp %p", vp); 1264 } 1265 1266 #ifdef DIAGNOSTIC 1267 if (uiop->uio_iovcnt != 1) 1268 panic("nfs: writerpc iovcnt > 1"); 1269 #endif 1270 tsiz = uiop->uio_resid; 1271 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1272 return (EFBIG); 1273 if (pageprotected) { 1274 l = curlwp; 1275 PHOLD(l); 1276 } 1277 retry: 1278 origresid = uiop->uio_resid; 1279 KASSERT(origresid == uiop->uio_iov->iov_len); 1280 while (tsiz > 0) { 1281 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1282 uint32_t backup; 1283 boolean_t stalewriteverf = FALSE; 1284 1285 nfsstats.rpccnt[NFSPROC_WRITE]++; 1286 len = min(tsiz, nmp->nm_wsize); 1287 datalen = pageprotected ? 0 : nfsm_rndup(len); 1288 nfsm_reqhead(np, NFSPROC_WRITE, 1289 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1290 nfsm_fhtom(np, v3); 1291 if (v3) { 1292 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1293 txdr_hyper(uiop->uio_offset, tl); 1294 tl += 2; 1295 *tl++ = txdr_unsigned(len); 1296 *tl++ = txdr_unsigned(*iomode); 1297 *tl = txdr_unsigned(len); 1298 } else { 1299 u_int32_t x; 1300 1301 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1302 /* Set both "begin" and "current" to non-garbage. */ 1303 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1304 *tl++ = x; /* "begin offset" */ 1305 *tl++ = x; /* "current offset" */ 1306 x = txdr_unsigned(len); 1307 *tl++ = x; /* total to this offset */ 1308 *tl = x; /* size of this write */ 1309 1310 } 1311 if (pageprotected) { 1312 /* 1313 * since we know pages can't be modified during i/o, 1314 * no need to copy them for us. 1315 */ 1316 struct mbuf *m; 1317 struct iovec *iovp = uiop->uio_iov; 1318 1319 m = m_get(M_WAIT, MT_DATA); 1320 MCLAIM(m, &nfs_mowner); 1321 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1322 nfs_writerpc_extfree, &ctx); 1323 m->m_flags |= M_EXT_ROMAP; 1324 m->m_len = len; 1325 mb->m_next = m; 1326 /* 1327 * no need to maintain mb and bpos here 1328 * because no one care them later. 1329 */ 1330 #if 0 1331 mb = m; 1332 bpos = mtod(caddr_t, mb) + mb->m_len; 1333 #endif 1334 iovp->iov_base = (char *)iovp->iov_base + len; 1335 iovp->iov_len -= len; 1336 uiop->uio_offset += len; 1337 uiop->uio_resid -= len; 1338 s = splvm(); 1339 simple_lock(&ctx.nwc_slock); 1340 ctx.nwc_mbufcount++; 1341 simple_unlock(&ctx.nwc_slock); 1342 splx(s); 1343 } else { 1344 nfsm_uiotom(uiop, len); 1345 } 1346 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, 1347 VTONFS(vp)->n_wcred); 1348 if (v3) { 1349 wccflag = NFSV3_WCCCHK; 1350 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC); 1351 if (!error) { 1352 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1353 + NFSX_V3WRITEVERF); 1354 rlen = fxdr_unsigned(int, *tl++); 1355 if (rlen == 0) { 1356 error = NFSERR_IO; 1357 m_freem(mrep); 1358 break; 1359 } else if (rlen < len) { 1360 backup = len - rlen; 1361 uiop->uio_iov->iov_base = 1362 (caddr_t)uiop->uio_iov->iov_base - 1363 backup; 1364 uiop->uio_iov->iov_len += backup; 1365 uiop->uio_offset -= backup; 1366 uiop->uio_resid += backup; 1367 len = rlen; 1368 } 1369 commit = fxdr_unsigned(int, *tl++); 1370 1371 /* 1372 * Return the lowest committment level 1373 * obtained by any of the RPCs. 1374 */ 1375 if (committed == NFSV3WRITE_FILESYNC) 1376 committed = commit; 1377 else if (committed == NFSV3WRITE_DATASYNC && 1378 commit == NFSV3WRITE_UNSTABLE) 1379 committed = commit; 1380 simple_lock(&nmp->nm_slock); 1381 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1382 memcpy(nmp->nm_writeverf, tl, 1383 NFSX_V3WRITEVERF); 1384 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1385 } else if ((nmp->nm_iflag & 1386 NFSMNT_STALEWRITEVERF) || 1387 memcmp(tl, nmp->nm_writeverf, 1388 NFSX_V3WRITEVERF)) { 1389 memcpy(nmp->nm_writeverf, tl, 1390 NFSX_V3WRITEVERF); 1391 /* 1392 * note NFSMNT_STALEWRITEVERF 1393 * if we're the first thread to 1394 * notice it. 1395 */ 1396 if ((nmp->nm_iflag & 1397 NFSMNT_STALEWRITEVERF) == 0) { 1398 stalewriteverf = TRUE; 1399 nmp->nm_iflag |= 1400 NFSMNT_STALEWRITEVERF; 1401 } 1402 } 1403 simple_unlock(&nmp->nm_slock); 1404 } 1405 } else 1406 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1407 if (wccflag) 1408 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1409 m_freem(mrep); 1410 if (error) 1411 break; 1412 tsiz -= len; 1413 if (stalewriteverf) { 1414 *stalewriteverfp = TRUE; 1415 stalewriteverf = FALSE; 1416 if (committed == NFSV3WRITE_UNSTABLE && 1417 len != origresid) { 1418 /* 1419 * if our write requests weren't atomic but 1420 * unstable, datas in previous iterations 1421 * might have already been lost now. 1422 * then, we should resend them to nfsd. 1423 */ 1424 backup = origresid - tsiz; 1425 uiop->uio_iov->iov_base = 1426 (caddr_t)uiop->uio_iov->iov_base - backup; 1427 uiop->uio_iov->iov_len += backup; 1428 uiop->uio_offset -= backup; 1429 uiop->uio_resid += backup; 1430 tsiz = origresid; 1431 goto retry; 1432 } 1433 } 1434 } 1435 if (pageprotected) { 1436 /* 1437 * wait until mbufs go away. 1438 * retransmitted mbufs can survive longer than rpc requests 1439 * themselves. 1440 */ 1441 s = splvm(); 1442 simple_lock(&ctx.nwc_slock); 1443 ctx.nwc_mbufcount--; 1444 while (ctx.nwc_mbufcount > 0) { 1445 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock); 1446 } 1447 simple_unlock(&ctx.nwc_slock); 1448 splx(s); 1449 PRELE(l); 1450 } 1451 nfsmout: 1452 *iomode = committed; 1453 if (error) 1454 uiop->uio_resid = tsiz; 1455 return (error); 1456 } 1457 1458 /* 1459 * nfs mknod rpc 1460 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1461 * mode set to specify the file type and the size field for rdev. 1462 */ 1463 int 1464 nfs_mknodrpc(dvp, vpp, cnp, vap) 1465 struct vnode *dvp; 1466 struct vnode **vpp; 1467 struct componentname *cnp; 1468 struct vattr *vap; 1469 { 1470 struct nfsv2_sattr *sp; 1471 u_int32_t *tl; 1472 caddr_t cp; 1473 int32_t t1, t2; 1474 struct vnode *newvp = (struct vnode *)0; 1475 struct nfsnode *dnp, *np; 1476 char *cp2; 1477 caddr_t bpos, dpos; 1478 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1479 struct mbuf *mreq, *mrep, *md, *mb; 1480 u_int32_t rdev; 1481 const int v3 = NFS_ISV3(dvp); 1482 1483 if (vap->va_type == VCHR || vap->va_type == VBLK) 1484 rdev = txdr_unsigned(vap->va_rdev); 1485 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1486 rdev = nfs_xdrneg1; 1487 else { 1488 VOP_ABORTOP(dvp, cnp); 1489 vput(dvp); 1490 return (EOPNOTSUPP); 1491 } 1492 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1493 dnp = VTONFS(dvp); 1494 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1495 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1496 nfsm_fhtom(dnp, v3); 1497 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1498 if (v3) { 1499 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1500 *tl++ = vtonfsv3_type(vap->va_type); 1501 nfsm_v3attrbuild(vap, FALSE); 1502 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1503 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1504 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1505 *tl = txdr_unsigned(minor(vap->va_rdev)); 1506 } 1507 } else { 1508 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1509 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1510 sp->sa_uid = nfs_xdrneg1; 1511 sp->sa_gid = nfs_xdrneg1; 1512 sp->sa_size = rdev; 1513 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1514 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1515 } 1516 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred); 1517 if (!error) { 1518 nfsm_mtofh(dvp, newvp, v3, gotvp); 1519 if (!gotvp) { 1520 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1521 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1522 if (!error) 1523 newvp = NFSTOV(np); 1524 } 1525 } 1526 if (v3) 1527 nfsm_wcc_data(dvp, wccflag, 0); 1528 nfsm_reqdone; 1529 if (error) { 1530 if (newvp) 1531 vput(newvp); 1532 } else { 1533 if (cnp->cn_flags & MAKEENTRY) 1534 cache_enter(dvp, newvp, cnp); 1535 *vpp = newvp; 1536 } 1537 PNBUF_PUT(cnp->cn_pnbuf); 1538 VTONFS(dvp)->n_flag |= NMODIFIED; 1539 if (!wccflag) 1540 VTONFS(dvp)->n_attrstamp = 0; 1541 vput(dvp); 1542 return (error); 1543 } 1544 1545 /* 1546 * nfs mknod vop 1547 * just call nfs_mknodrpc() to do the work. 1548 */ 1549 /* ARGSUSED */ 1550 int 1551 nfs_mknod(v) 1552 void *v; 1553 { 1554 struct vop_mknod_args /* { 1555 struct vnode *a_dvp; 1556 struct vnode **a_vpp; 1557 struct componentname *a_cnp; 1558 struct vattr *a_vap; 1559 } */ *ap = v; 1560 int error; 1561 1562 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1563 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1564 return (error); 1565 } 1566 1567 static u_long create_verf; 1568 /* 1569 * nfs file create call 1570 */ 1571 int 1572 nfs_create(v) 1573 void *v; 1574 { 1575 struct vop_create_args /* { 1576 struct vnode *a_dvp; 1577 struct vnode **a_vpp; 1578 struct componentname *a_cnp; 1579 struct vattr *a_vap; 1580 } */ *ap = v; 1581 struct vnode *dvp = ap->a_dvp; 1582 struct vattr *vap = ap->a_vap; 1583 struct componentname *cnp = ap->a_cnp; 1584 struct nfsv2_sattr *sp; 1585 u_int32_t *tl; 1586 caddr_t cp; 1587 int32_t t1, t2; 1588 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1589 struct vnode *newvp = (struct vnode *)0; 1590 caddr_t bpos, dpos, cp2; 1591 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1592 struct mbuf *mreq, *mrep, *md, *mb; 1593 const int v3 = NFS_ISV3(dvp); 1594 1595 /* 1596 * Oops, not for me.. 1597 */ 1598 if (vap->va_type == VSOCK) 1599 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1600 1601 #ifdef VA_EXCLUSIVE 1602 if (vap->va_vaflags & VA_EXCLUSIVE) 1603 fmode |= O_EXCL; 1604 #endif 1605 again: 1606 error = 0; 1607 nfsstats.rpccnt[NFSPROC_CREATE]++; 1608 dnp = VTONFS(dvp); 1609 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1610 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1611 nfsm_fhtom(dnp, v3); 1612 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1613 if (v3) { 1614 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1615 if (fmode & O_EXCL) { 1616 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1617 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1618 #ifdef INET 1619 if (TAILQ_FIRST(&in_ifaddr)) 1620 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr; 1621 else 1622 *tl++ = create_verf; 1623 #else 1624 *tl++ = create_verf; 1625 #endif 1626 *tl = ++create_verf; 1627 } else { 1628 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1629 nfsm_v3attrbuild(vap, FALSE); 1630 } 1631 } else { 1632 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1633 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1634 sp->sa_uid = nfs_xdrneg1; 1635 sp->sa_gid = nfs_xdrneg1; 1636 sp->sa_size = 0; 1637 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1638 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1639 } 1640 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred); 1641 if (!error) { 1642 nfsm_mtofh(dvp, newvp, v3, gotvp); 1643 if (!gotvp) { 1644 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1645 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np); 1646 if (!error) 1647 newvp = NFSTOV(np); 1648 } 1649 } 1650 if (v3) 1651 nfsm_wcc_data(dvp, wccflag, 0); 1652 nfsm_reqdone; 1653 if (error) { 1654 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1655 fmode &= ~O_EXCL; 1656 goto again; 1657 } 1658 if (newvp) 1659 vput(newvp); 1660 } else if (v3 && (fmode & O_EXCL)) 1661 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc); 1662 if (!error) { 1663 if (cnp->cn_flags & MAKEENTRY) 1664 cache_enter(dvp, newvp, cnp); 1665 *ap->a_vpp = newvp; 1666 } 1667 PNBUF_PUT(cnp->cn_pnbuf); 1668 VTONFS(dvp)->n_flag |= NMODIFIED; 1669 if (!wccflag) 1670 VTONFS(dvp)->n_attrstamp = 0; 1671 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1672 vput(dvp); 1673 return (error); 1674 } 1675 1676 /* 1677 * nfs file remove call 1678 * To try and make nfs semantics closer to ufs semantics, a file that has 1679 * other processes using the vnode is renamed instead of removed and then 1680 * removed later on the last close. 1681 * - If v_usecount > 1 1682 * If a rename is not already in the works 1683 * call nfs_sillyrename() to set it up 1684 * else 1685 * do the remove rpc 1686 */ 1687 int 1688 nfs_remove(v) 1689 void *v; 1690 { 1691 struct vop_remove_args /* { 1692 struct vnodeop_desc *a_desc; 1693 struct vnode * a_dvp; 1694 struct vnode * a_vp; 1695 struct componentname * a_cnp; 1696 } */ *ap = v; 1697 struct vnode *vp = ap->a_vp; 1698 struct vnode *dvp = ap->a_dvp; 1699 struct componentname *cnp = ap->a_cnp; 1700 struct nfsnode *np = VTONFS(vp); 1701 int error = 0; 1702 struct vattr vattr; 1703 1704 #ifndef DIAGNOSTIC 1705 if ((cnp->cn_flags & HASBUF) == 0) 1706 panic("nfs_remove: no name"); 1707 if (vp->v_usecount < 1) 1708 panic("nfs_remove: bad v_usecount"); 1709 #endif 1710 if (vp->v_type == VDIR) 1711 error = EPERM; 1712 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1713 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 && 1714 vattr.va_nlink > 1)) { 1715 /* 1716 * Purge the name cache so that the chance of a lookup for 1717 * the name succeeding while the remove is in progress is 1718 * minimized. Without node locking it can still happen, such 1719 * that an I/O op returns ESTALE, but since you get this if 1720 * another host removes the file.. 1721 */ 1722 cache_purge(vp); 1723 /* 1724 * throw away biocache buffers, mainly to avoid 1725 * unnecessary delayed writes later. 1726 */ 1727 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1); 1728 /* Do the rpc */ 1729 if (error != EINTR) 1730 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1731 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc); 1732 /* 1733 * Kludge City: If the first reply to the remove rpc is lost.. 1734 * the reply to the retransmitted request will be ENOENT 1735 * since the file was in fact removed 1736 * Therefore, we cheat and return success. 1737 */ 1738 if (error == ENOENT) 1739 error = 0; 1740 } else if (!np->n_sillyrename) 1741 error = nfs_sillyrename(dvp, vp, cnp); 1742 PNBUF_PUT(cnp->cn_pnbuf); 1743 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1744 vattr.va_nlink == 1) { 1745 np->n_flag |= NREMOVED; 1746 } 1747 np->n_attrstamp = 0; 1748 VN_KNOTE(vp, NOTE_DELETE); 1749 VN_KNOTE(dvp, NOTE_WRITE); 1750 if (dvp == vp) 1751 vrele(vp); 1752 else 1753 vput(vp); 1754 vput(dvp); 1755 return (error); 1756 } 1757 1758 /* 1759 * nfs file remove rpc called from nfs_inactive 1760 */ 1761 int 1762 nfs_removeit(sp) 1763 struct sillyrename *sp; 1764 { 1765 1766 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1767 (struct proc *)0)); 1768 } 1769 1770 /* 1771 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1772 */ 1773 int 1774 nfs_removerpc(dvp, name, namelen, cred, proc) 1775 struct vnode *dvp; 1776 const char *name; 1777 int namelen; 1778 struct ucred *cred; 1779 struct proc *proc; 1780 { 1781 u_int32_t *tl; 1782 caddr_t cp; 1783 int32_t t1, t2; 1784 caddr_t bpos, dpos, cp2; 1785 int error = 0, wccflag = NFSV3_WCCRATTR; 1786 struct mbuf *mreq, *mrep, *md, *mb; 1787 const int v3 = NFS_ISV3(dvp); 1788 struct nfsnode *dnp = VTONFS(dvp); 1789 1790 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1791 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1792 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1793 nfsm_fhtom(dnp, v3); 1794 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1795 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred); 1796 if (v3) 1797 nfsm_wcc_data(dvp, wccflag, 0); 1798 nfsm_reqdone; 1799 VTONFS(dvp)->n_flag |= NMODIFIED; 1800 if (!wccflag) 1801 VTONFS(dvp)->n_attrstamp = 0; 1802 return (error); 1803 } 1804 1805 /* 1806 * nfs file rename call 1807 */ 1808 int 1809 nfs_rename(v) 1810 void *v; 1811 { 1812 struct vop_rename_args /* { 1813 struct vnode *a_fdvp; 1814 struct vnode *a_fvp; 1815 struct componentname *a_fcnp; 1816 struct vnode *a_tdvp; 1817 struct vnode *a_tvp; 1818 struct componentname *a_tcnp; 1819 } */ *ap = v; 1820 struct vnode *fvp = ap->a_fvp; 1821 struct vnode *tvp = ap->a_tvp; 1822 struct vnode *fdvp = ap->a_fdvp; 1823 struct vnode *tdvp = ap->a_tdvp; 1824 struct componentname *tcnp = ap->a_tcnp; 1825 struct componentname *fcnp = ap->a_fcnp; 1826 int error; 1827 1828 #ifndef DIAGNOSTIC 1829 if ((tcnp->cn_flags & HASBUF) == 0 || 1830 (fcnp->cn_flags & HASBUF) == 0) 1831 panic("nfs_rename: no name"); 1832 #endif 1833 /* Check for cross-device rename */ 1834 if ((fvp->v_mount != tdvp->v_mount) || 1835 (tvp && (fvp->v_mount != tvp->v_mount))) { 1836 error = EXDEV; 1837 goto out; 1838 } 1839 1840 /* 1841 * If the tvp exists and is in use, sillyrename it before doing the 1842 * rename of the new file over it. 1843 */ 1844 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1845 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1846 VN_KNOTE(tvp, NOTE_DELETE); 1847 vput(tvp); 1848 tvp = NULL; 1849 } 1850 1851 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1852 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1853 tcnp->cn_proc); 1854 1855 VN_KNOTE(fdvp, NOTE_WRITE); 1856 VN_KNOTE(tdvp, NOTE_WRITE); 1857 if (fvp->v_type == VDIR) { 1858 if (tvp != NULL && tvp->v_type == VDIR) 1859 cache_purge(tdvp); 1860 cache_purge(fdvp); 1861 } 1862 out: 1863 if (tdvp == tvp) 1864 vrele(tdvp); 1865 else 1866 vput(tdvp); 1867 if (tvp) 1868 vput(tvp); 1869 vrele(fdvp); 1870 vrele(fvp); 1871 /* 1872 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1873 */ 1874 if (error == ENOENT) 1875 error = 0; 1876 return (error); 1877 } 1878 1879 /* 1880 * nfs file rename rpc called from nfs_remove() above 1881 */ 1882 int 1883 nfs_renameit(sdvp, scnp, sp) 1884 struct vnode *sdvp; 1885 struct componentname *scnp; 1886 struct sillyrename *sp; 1887 { 1888 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1889 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc)); 1890 } 1891 1892 /* 1893 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1894 */ 1895 int 1896 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc) 1897 struct vnode *fdvp; 1898 const char *fnameptr; 1899 int fnamelen; 1900 struct vnode *tdvp; 1901 const char *tnameptr; 1902 int tnamelen; 1903 struct ucred *cred; 1904 struct proc *proc; 1905 { 1906 u_int32_t *tl; 1907 caddr_t cp; 1908 int32_t t1, t2; 1909 caddr_t bpos, dpos, cp2; 1910 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1911 struct mbuf *mreq, *mrep, *md, *mb; 1912 const int v3 = NFS_ISV3(fdvp); 1913 struct nfsnode *fdnp = VTONFS(fdvp); 1914 1915 nfsstats.rpccnt[NFSPROC_RENAME]++; 1916 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1917 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1918 nfsm_rndup(tnamelen)); 1919 nfsm_fhtom(fdnp, v3); 1920 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1921 nfsm_fhtom(VTONFS(tdvp), v3); 1922 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1923 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred); 1924 if (v3) { 1925 nfsm_wcc_data(fdvp, fwccflag, 0); 1926 nfsm_wcc_data(tdvp, twccflag, 0); 1927 } 1928 nfsm_reqdone; 1929 VTONFS(fdvp)->n_flag |= NMODIFIED; 1930 VTONFS(tdvp)->n_flag |= NMODIFIED; 1931 if (!fwccflag) 1932 VTONFS(fdvp)->n_attrstamp = 0; 1933 if (!twccflag) 1934 VTONFS(tdvp)->n_attrstamp = 0; 1935 return (error); 1936 } 1937 1938 /* 1939 * nfs hard link create call 1940 */ 1941 int 1942 nfs_link(v) 1943 void *v; 1944 { 1945 struct vop_link_args /* { 1946 struct vnode *a_dvp; 1947 struct vnode *a_vp; 1948 struct componentname *a_cnp; 1949 } */ *ap = v; 1950 struct vnode *vp = ap->a_vp; 1951 struct vnode *dvp = ap->a_dvp; 1952 struct componentname *cnp = ap->a_cnp; 1953 u_int32_t *tl; 1954 caddr_t cp; 1955 int32_t t1, t2; 1956 caddr_t bpos, dpos, cp2; 1957 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1958 struct mbuf *mreq, *mrep, *md, *mb; 1959 /* XXX Should be const and initialised? */ 1960 int v3; 1961 struct nfsnode *np; 1962 1963 if (dvp->v_mount != vp->v_mount) { 1964 VOP_ABORTOP(dvp, cnp); 1965 vput(dvp); 1966 return (EXDEV); 1967 } 1968 if (dvp != vp) { 1969 error = vn_lock(vp, LK_EXCLUSIVE); 1970 if (error != 0) { 1971 VOP_ABORTOP(dvp, cnp); 1972 vput(dvp); 1973 return error; 1974 } 1975 } 1976 1977 /* 1978 * Push all writes to the server, so that the attribute cache 1979 * doesn't get "out of sync" with the server. 1980 * XXX There should be a better way! 1981 */ 1982 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc); 1983 1984 v3 = NFS_ISV3(vp); 1985 nfsstats.rpccnt[NFSPROC_LINK]++; 1986 np = VTONFS(vp); 1987 nfsm_reqhead(np, NFSPROC_LINK, 1988 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1989 nfsm_fhtom(np, v3); 1990 nfsm_fhtom(VTONFS(dvp), v3); 1991 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1992 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred); 1993 if (v3) { 1994 nfsm_postop_attr(vp, attrflag, 0); 1995 nfsm_wcc_data(dvp, wccflag, 0); 1996 } 1997 nfsm_reqdone; 1998 PNBUF_PUT(cnp->cn_pnbuf); 1999 VTONFS(dvp)->n_flag |= NMODIFIED; 2000 if (!attrflag) 2001 VTONFS(vp)->n_attrstamp = 0; 2002 if (!wccflag) 2003 VTONFS(dvp)->n_attrstamp = 0; 2004 if (dvp != vp) 2005 VOP_UNLOCK(vp, 0); 2006 VN_KNOTE(vp, NOTE_LINK); 2007 VN_KNOTE(dvp, NOTE_WRITE); 2008 vput(dvp); 2009 /* 2010 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2011 */ 2012 if (error == EEXIST) 2013 error = 0; 2014 return (error); 2015 } 2016 2017 /* 2018 * nfs symbolic link create call 2019 */ 2020 int 2021 nfs_symlink(v) 2022 void *v; 2023 { 2024 struct vop_symlink_args /* { 2025 struct vnode *a_dvp; 2026 struct vnode **a_vpp; 2027 struct componentname *a_cnp; 2028 struct vattr *a_vap; 2029 char *a_target; 2030 } */ *ap = v; 2031 struct vnode *dvp = ap->a_dvp; 2032 struct vattr *vap = ap->a_vap; 2033 struct componentname *cnp = ap->a_cnp; 2034 struct nfsv2_sattr *sp; 2035 u_int32_t *tl; 2036 caddr_t cp; 2037 int32_t t1, t2; 2038 caddr_t bpos, dpos, cp2; 2039 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2040 struct mbuf *mreq, *mrep, *md, *mb; 2041 struct vnode *newvp = (struct vnode *)0; 2042 const int v3 = NFS_ISV3(dvp); 2043 struct nfsnode *dnp = VTONFS(dvp); 2044 2045 *ap->a_vpp = NULL; 2046 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2047 slen = strlen(ap->a_target); 2048 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2049 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2050 nfsm_fhtom(dnp, v3); 2051 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2052 if (v3) 2053 nfsm_v3attrbuild(vap, FALSE); 2054 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2055 if (!v3) { 2056 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2057 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2058 sp->sa_uid = nfs_xdrneg1; 2059 sp->sa_gid = nfs_xdrneg1; 2060 sp->sa_size = nfs_xdrneg1; 2061 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2062 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2063 } 2064 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred); 2065 if (v3) { 2066 if (!error) 2067 nfsm_mtofh(dvp, newvp, v3, gotvp); 2068 nfsm_wcc_data(dvp, wccflag, 0); 2069 } 2070 nfsm_reqdone; 2071 /* 2072 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2073 */ 2074 if (error == EEXIST) 2075 error = 0; 2076 if (error == 0 && newvp == NULL) { 2077 struct nfsnode *np = NULL; 2078 2079 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2080 cnp->cn_cred, cnp->cn_proc, &np); 2081 if (error == 0) 2082 newvp = NFSTOV(np); 2083 } 2084 if (error) { 2085 if (newvp != NULL) 2086 vput(newvp); 2087 } else { 2088 *ap->a_vpp = newvp; 2089 } 2090 PNBUF_PUT(cnp->cn_pnbuf); 2091 VTONFS(dvp)->n_flag |= NMODIFIED; 2092 if (!wccflag) 2093 VTONFS(dvp)->n_attrstamp = 0; 2094 VN_KNOTE(dvp, NOTE_WRITE); 2095 vput(dvp); 2096 return (error); 2097 } 2098 2099 /* 2100 * nfs make dir call 2101 */ 2102 int 2103 nfs_mkdir(v) 2104 void *v; 2105 { 2106 struct vop_mkdir_args /* { 2107 struct vnode *a_dvp; 2108 struct vnode **a_vpp; 2109 struct componentname *a_cnp; 2110 struct vattr *a_vap; 2111 } */ *ap = v; 2112 struct vnode *dvp = ap->a_dvp; 2113 struct vattr *vap = ap->a_vap; 2114 struct componentname *cnp = ap->a_cnp; 2115 struct nfsv2_sattr *sp; 2116 u_int32_t *tl; 2117 caddr_t cp; 2118 int32_t t1, t2; 2119 int len; 2120 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2121 struct vnode *newvp = (struct vnode *)0; 2122 caddr_t bpos, dpos, cp2; 2123 int error = 0, wccflag = NFSV3_WCCRATTR; 2124 int gotvp = 0; 2125 struct mbuf *mreq, *mrep, *md, *mb; 2126 const int v3 = NFS_ISV3(dvp); 2127 2128 len = cnp->cn_namelen; 2129 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2130 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2131 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2132 nfsm_fhtom(dnp, v3); 2133 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2134 if (v3) { 2135 nfsm_v3attrbuild(vap, FALSE); 2136 } else { 2137 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2138 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2139 sp->sa_uid = nfs_xdrneg1; 2140 sp->sa_gid = nfs_xdrneg1; 2141 sp->sa_size = nfs_xdrneg1; 2142 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2143 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2144 } 2145 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred); 2146 if (!error) 2147 nfsm_mtofh(dvp, newvp, v3, gotvp); 2148 if (v3) 2149 nfsm_wcc_data(dvp, wccflag, 0); 2150 nfsm_reqdone; 2151 VTONFS(dvp)->n_flag |= NMODIFIED; 2152 if (!wccflag) 2153 VTONFS(dvp)->n_attrstamp = 0; 2154 /* 2155 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2156 * if we can succeed in looking up the directory. 2157 */ 2158 if (error == EEXIST || (!error && !gotvp)) { 2159 if (newvp) { 2160 vput(newvp); 2161 newvp = (struct vnode *)0; 2162 } 2163 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2164 cnp->cn_proc, &np); 2165 if (!error) { 2166 newvp = NFSTOV(np); 2167 if (newvp->v_type != VDIR) 2168 error = EEXIST; 2169 } 2170 } 2171 if (error) { 2172 if (newvp) 2173 vput(newvp); 2174 } else { 2175 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2176 if (cnp->cn_flags & MAKEENTRY) 2177 cache_enter(dvp, newvp, cnp); 2178 *ap->a_vpp = newvp; 2179 } 2180 PNBUF_PUT(cnp->cn_pnbuf); 2181 vput(dvp); 2182 return (error); 2183 } 2184 2185 /* 2186 * nfs remove directory call 2187 */ 2188 int 2189 nfs_rmdir(v) 2190 void *v; 2191 { 2192 struct vop_rmdir_args /* { 2193 struct vnode *a_dvp; 2194 struct vnode *a_vp; 2195 struct componentname *a_cnp; 2196 } */ *ap = v; 2197 struct vnode *vp = ap->a_vp; 2198 struct vnode *dvp = ap->a_dvp; 2199 struct componentname *cnp = ap->a_cnp; 2200 u_int32_t *tl; 2201 caddr_t cp; 2202 int32_t t1, t2; 2203 caddr_t bpos, dpos, cp2; 2204 int error = 0, wccflag = NFSV3_WCCRATTR; 2205 struct mbuf *mreq, *mrep, *md, *mb; 2206 const int v3 = NFS_ISV3(dvp); 2207 struct nfsnode *dnp; 2208 2209 if (dvp == vp) { 2210 vrele(dvp); 2211 vput(dvp); 2212 PNBUF_PUT(cnp->cn_pnbuf); 2213 return (EINVAL); 2214 } 2215 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2216 dnp = VTONFS(dvp); 2217 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2218 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2219 nfsm_fhtom(dnp, v3); 2220 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2221 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred); 2222 if (v3) 2223 nfsm_wcc_data(dvp, wccflag, 0); 2224 nfsm_reqdone; 2225 PNBUF_PUT(cnp->cn_pnbuf); 2226 VTONFS(dvp)->n_flag |= NMODIFIED; 2227 if (!wccflag) 2228 VTONFS(dvp)->n_attrstamp = 0; 2229 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2230 VN_KNOTE(vp, NOTE_DELETE); 2231 cache_purge(dvp); 2232 cache_purge(vp); 2233 vput(vp); 2234 vput(dvp); 2235 /* 2236 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2237 */ 2238 if (error == ENOENT) 2239 error = 0; 2240 return (error); 2241 } 2242 2243 /* 2244 * nfs readdir call 2245 */ 2246 int 2247 nfs_readdir(v) 2248 void *v; 2249 { 2250 struct vop_readdir_args /* { 2251 struct vnode *a_vp; 2252 struct uio *a_uio; 2253 struct ucred *a_cred; 2254 int *a_eofflag; 2255 off_t **a_cookies; 2256 int *a_ncookies; 2257 } */ *ap = v; 2258 struct vnode *vp = ap->a_vp; 2259 struct uio *uio = ap->a_uio; 2260 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2261 char *base = uio->uio_iov->iov_base; 2262 int tresid, error; 2263 size_t count, lost; 2264 struct dirent *dp; 2265 off_t *cookies = NULL; 2266 int ncookies = 0, nc; 2267 2268 if (vp->v_type != VDIR) 2269 return (EPERM); 2270 2271 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2272 count = uio->uio_resid - lost; 2273 if (count <= 0) 2274 return (EINVAL); 2275 2276 /* 2277 * Call nfs_bioread() to do the real work. 2278 */ 2279 tresid = uio->uio_resid = count; 2280 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2281 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2282 2283 if (!error && ap->a_cookies) { 2284 ncookies = count / 16; 2285 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2286 *ap->a_cookies = cookies; 2287 } 2288 2289 if (!error && uio->uio_resid == tresid) { 2290 uio->uio_resid += lost; 2291 nfsstats.direofcache_misses++; 2292 if (ap->a_cookies) 2293 *ap->a_ncookies = 0; 2294 *ap->a_eofflag = 1; 2295 return (0); 2296 } 2297 2298 if (!error && ap->a_cookies) { 2299 /* 2300 * Only the NFS server and emulations use cookies, and they 2301 * load the directory block into system space, so we can 2302 * just look at it directly. 2303 */ 2304 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2305 panic("nfs_readdir: lost in space"); 2306 for (nc = 0; ncookies-- && 2307 base < (char *)uio->uio_iov->iov_base; nc++){ 2308 dp = (struct dirent *) base; 2309 if (dp->d_reclen == 0) 2310 break; 2311 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2312 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2313 else 2314 *(cookies++) = NFS_GETCOOKIE(dp); 2315 base += dp->d_reclen; 2316 } 2317 uio->uio_resid += 2318 ((caddr_t)uio->uio_iov->iov_base - base); 2319 uio->uio_iov->iov_len += 2320 ((caddr_t)uio->uio_iov->iov_base - base); 2321 uio->uio_iov->iov_base = base; 2322 *ap->a_ncookies = nc; 2323 } 2324 2325 uio->uio_resid += lost; 2326 *ap->a_eofflag = 0; 2327 return (error); 2328 } 2329 2330 /* 2331 * Readdir rpc call. 2332 * Called from below the buffer cache by nfs_doio(). 2333 */ 2334 int 2335 nfs_readdirrpc(vp, uiop, cred) 2336 struct vnode *vp; 2337 struct uio *uiop; 2338 struct ucred *cred; 2339 { 2340 int len, left; 2341 struct dirent *dp = NULL; 2342 u_int32_t *tl; 2343 caddr_t cp; 2344 int32_t t1, t2; 2345 caddr_t bpos, dpos, cp2; 2346 struct mbuf *mreq, *mrep, *md, *mb; 2347 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2348 struct nfsnode *dnp = VTONFS(vp); 2349 u_quad_t fileno; 2350 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2351 int attrflag, nrpcs = 0, reclen; 2352 const int v3 = NFS_ISV3(vp); 2353 2354 #ifdef DIAGNOSTIC 2355 /* 2356 * Should be called from buffer cache, so only amount of 2357 * NFS_DIRBLKSIZ will be requested. 2358 */ 2359 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2360 panic("nfs readdirrpc bad uio"); 2361 #endif 2362 2363 /* 2364 * Loop around doing readdir rpc's of size nm_readdirsize 2365 * truncated to a multiple of NFS_DIRFRAGSIZ. 2366 * The stopping criteria is EOF or buffer full. 2367 */ 2368 while (more_dirs && bigenough) { 2369 /* 2370 * Heuristic: don't bother to do another RPC to further 2371 * fill up this block if there is not much room left. (< 50% 2372 * of the readdir RPC size). This wastes some buffer space 2373 * but can save up to 50% in RPC calls. 2374 */ 2375 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2376 bigenough = 0; 2377 break; 2378 } 2379 nfsstats.rpccnt[NFSPROC_READDIR]++; 2380 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2381 NFSX_READDIR(v3)); 2382 nfsm_fhtom(dnp, v3); 2383 if (v3) { 2384 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2385 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2386 txdr_swapcookie3(uiop->uio_offset, tl); 2387 } else { 2388 txdr_cookie3(uiop->uio_offset, tl); 2389 } 2390 tl += 2; 2391 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2392 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2393 } else { 2394 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2395 *tl++ = txdr_unsigned(uiop->uio_offset); 2396 } 2397 *tl = txdr_unsigned(nmp->nm_readdirsize); 2398 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred); 2399 nrpcs++; 2400 if (v3) { 2401 nfsm_postop_attr(vp, attrflag, 0); 2402 if (!error) { 2403 nfsm_dissect(tl, u_int32_t *, 2404 2 * NFSX_UNSIGNED); 2405 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2406 dnp->n_cookieverf.nfsuquad[1] = *tl; 2407 } else { 2408 m_freem(mrep); 2409 goto nfsmout; 2410 } 2411 } 2412 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2413 more_dirs = fxdr_unsigned(int, *tl); 2414 2415 /* loop thru the dir entries, doctoring them to 4bsd form */ 2416 while (more_dirs && bigenough) { 2417 if (v3) { 2418 nfsm_dissect(tl, u_int32_t *, 2419 3 * NFSX_UNSIGNED); 2420 fileno = fxdr_hyper(tl); 2421 len = fxdr_unsigned(int, *(tl + 2)); 2422 } else { 2423 nfsm_dissect(tl, u_int32_t *, 2424 2 * NFSX_UNSIGNED); 2425 fileno = fxdr_unsigned(u_quad_t, *tl++); 2426 len = fxdr_unsigned(int, *tl); 2427 } 2428 if (len <= 0 || len > NFS_MAXNAMLEN) { 2429 error = EBADRPC; 2430 m_freem(mrep); 2431 goto nfsmout; 2432 } 2433 tlen = nfsm_rndup(len); 2434 if (tlen == len) 2435 tlen += 4; /* To ensure null termination */ 2436 tlen += sizeof (off_t) + sizeof (int); 2437 reclen = ALIGN(tlen + DIRHDSIZ); 2438 tlen = reclen - DIRHDSIZ; 2439 left = NFS_DIRFRAGSIZ - blksiz; 2440 if (reclen > left) { 2441 dp->d_reclen += left; 2442 uiop->uio_iov->iov_base = 2443 (caddr_t)uiop->uio_iov->iov_base + left; 2444 uiop->uio_iov->iov_len -= left; 2445 uiop->uio_resid -= left; 2446 blksiz = 0; 2447 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2448 } 2449 if (reclen > uiop->uio_resid) 2450 bigenough = 0; 2451 if (bigenough) { 2452 dp = (struct dirent *)uiop->uio_iov->iov_base; 2453 dp->d_fileno = (int)fileno; 2454 dp->d_namlen = len; 2455 dp->d_reclen = reclen; 2456 dp->d_type = DT_UNKNOWN; 2457 blksiz += dp->d_reclen; 2458 if (blksiz == NFS_DIRFRAGSIZ) 2459 blksiz = 0; 2460 uiop->uio_resid -= DIRHDSIZ; 2461 uiop->uio_iov->iov_base = 2462 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ; 2463 uiop->uio_iov->iov_len -= DIRHDSIZ; 2464 nfsm_mtouio(uiop, len); 2465 cp = uiop->uio_iov->iov_base; 2466 tlen -= len; 2467 *cp = '\0'; /* null terminate */ 2468 uiop->uio_iov->iov_base = 2469 (caddr_t)uiop->uio_iov->iov_base + tlen; 2470 uiop->uio_iov->iov_len -= tlen; 2471 uiop->uio_resid -= tlen; 2472 } else 2473 nfsm_adv(nfsm_rndup(len)); 2474 if (v3) { 2475 nfsm_dissect(tl, u_int32_t *, 2476 3 * NFSX_UNSIGNED); 2477 } else { 2478 nfsm_dissect(tl, u_int32_t *, 2479 2 * NFSX_UNSIGNED); 2480 } 2481 if (bigenough) { 2482 if (v3) { 2483 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2484 uiop->uio_offset = 2485 fxdr_swapcookie3(tl); 2486 else 2487 uiop->uio_offset = 2488 fxdr_cookie3(tl); 2489 } 2490 else { 2491 uiop->uio_offset = 2492 fxdr_unsigned(off_t, *tl); 2493 } 2494 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2495 } 2496 if (v3) 2497 tl += 2; 2498 else 2499 tl++; 2500 more_dirs = fxdr_unsigned(int, *tl); 2501 } 2502 /* 2503 * If at end of rpc data, get the eof boolean 2504 */ 2505 if (!more_dirs) { 2506 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2507 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2508 } 2509 m_freem(mrep); 2510 } 2511 /* 2512 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2513 * by increasing d_reclen for the last record. 2514 */ 2515 if (blksiz > 0) { 2516 left = NFS_DIRFRAGSIZ - blksiz; 2517 dp->d_reclen += left; 2518 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2519 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2520 left; 2521 uiop->uio_iov->iov_len -= left; 2522 uiop->uio_resid -= left; 2523 } 2524 2525 /* 2526 * We are now either at the end of the directory or have filled the 2527 * block. 2528 */ 2529 if (bigenough) 2530 dnp->n_direofoffset = uiop->uio_offset; 2531 nfsmout: 2532 return (error); 2533 } 2534 2535 /* 2536 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2537 */ 2538 int 2539 nfs_readdirplusrpc(vp, uiop, cred) 2540 struct vnode *vp; 2541 struct uio *uiop; 2542 struct ucred *cred; 2543 { 2544 int len, left; 2545 struct dirent *dp = NULL; 2546 u_int32_t *tl; 2547 caddr_t cp; 2548 int32_t t1, t2; 2549 struct vnode *newvp; 2550 caddr_t bpos, dpos, cp2; 2551 struct mbuf *mreq, *mrep, *md, *mb; 2552 struct nameidata nami, *ndp = &nami; 2553 struct componentname *cnp = &ndp->ni_cnd; 2554 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2555 struct nfsnode *dnp = VTONFS(vp), *np; 2556 nfsfh_t *fhp; 2557 u_quad_t fileno; 2558 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2559 int attrflag, fhsize, nrpcs = 0, reclen; 2560 struct nfs_fattr fattr, *fp; 2561 2562 #ifdef DIAGNOSTIC 2563 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1))) 2564 panic("nfs readdirplusrpc bad uio"); 2565 #endif 2566 ndp->ni_dvp = vp; 2567 newvp = NULLVP; 2568 2569 /* 2570 * Loop around doing readdir rpc's of size nm_readdirsize 2571 * truncated to a multiple of NFS_DIRFRAGSIZ. 2572 * The stopping criteria is EOF or buffer full. 2573 */ 2574 while (more_dirs && bigenough) { 2575 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2576 bigenough = 0; 2577 break; 2578 } 2579 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2580 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2581 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2582 nfsm_fhtom(dnp, 1); 2583 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2584 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2585 txdr_swapcookie3(uiop->uio_offset, tl); 2586 } else { 2587 txdr_cookie3(uiop->uio_offset, tl); 2588 } 2589 tl += 2; 2590 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2591 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2592 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2593 *tl = txdr_unsigned(nmp->nm_rsize); 2594 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred); 2595 nfsm_postop_attr(vp, attrflag, 0); 2596 if (error) { 2597 m_freem(mrep); 2598 goto nfsmout; 2599 } 2600 nrpcs++; 2601 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2602 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2603 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2604 more_dirs = fxdr_unsigned(int, *tl); 2605 2606 /* loop thru the dir entries, doctoring them to 4bsd form */ 2607 while (more_dirs && bigenough) { 2608 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2609 fileno = fxdr_hyper(tl); 2610 len = fxdr_unsigned(int, *(tl + 2)); 2611 if (len <= 0 || len > NFS_MAXNAMLEN) { 2612 error = EBADRPC; 2613 m_freem(mrep); 2614 goto nfsmout; 2615 } 2616 tlen = nfsm_rndup(len); 2617 if (tlen == len) 2618 tlen += 4; /* To ensure null termination*/ 2619 tlen += sizeof (off_t) + sizeof (int); 2620 reclen = ALIGN(tlen + DIRHDSIZ); 2621 tlen = reclen - DIRHDSIZ; 2622 left = NFS_DIRFRAGSIZ - blksiz; 2623 if (reclen > left) { 2624 /* 2625 * DIRFRAGSIZ is aligned, no need to align 2626 * again here. 2627 */ 2628 dp->d_reclen += left; 2629 uiop->uio_iov->iov_base = 2630 (caddr_t)uiop->uio_iov->iov_base + left; 2631 uiop->uio_iov->iov_len -= left; 2632 uiop->uio_resid -= left; 2633 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2634 blksiz = 0; 2635 } 2636 if (reclen > uiop->uio_resid) 2637 bigenough = 0; 2638 if (bigenough) { 2639 dp = (struct dirent *)uiop->uio_iov->iov_base; 2640 dp->d_fileno = (int)fileno; 2641 dp->d_namlen = len; 2642 dp->d_reclen = reclen; 2643 dp->d_type = DT_UNKNOWN; 2644 blksiz += dp->d_reclen; 2645 if (blksiz == NFS_DIRFRAGSIZ) 2646 blksiz = 0; 2647 uiop->uio_resid -= DIRHDSIZ; 2648 uiop->uio_iov->iov_base = 2649 (caddr_t)uiop->uio_iov->iov_base + 2650 DIRHDSIZ; 2651 uiop->uio_iov->iov_len -= DIRHDSIZ; 2652 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2653 cnp->cn_namelen = len; 2654 nfsm_mtouio(uiop, len); 2655 cp = uiop->uio_iov->iov_base; 2656 tlen -= len; 2657 *cp = '\0'; 2658 uiop->uio_iov->iov_base = 2659 (caddr_t)uiop->uio_iov->iov_base + tlen; 2660 uiop->uio_iov->iov_len -= tlen; 2661 uiop->uio_resid -= tlen; 2662 } else 2663 nfsm_adv(nfsm_rndup(len)); 2664 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2665 if (bigenough) { 2666 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2667 uiop->uio_offset = 2668 fxdr_swapcookie3(tl); 2669 else 2670 uiop->uio_offset = 2671 fxdr_cookie3(tl); 2672 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2673 } 2674 tl += 2; 2675 2676 /* 2677 * Since the attributes are before the file handle 2678 * (sigh), we must skip over the attributes and then 2679 * come back and get them. 2680 */ 2681 attrflag = fxdr_unsigned(int, *tl); 2682 if (attrflag) { 2683 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2684 memcpy(&fattr, fp, NFSX_V3FATTR); 2685 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2686 doit = fxdr_unsigned(int, *tl); 2687 if (doit) { 2688 nfsm_getfh(fhp, fhsize, 1); 2689 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2690 VREF(vp); 2691 newvp = vp; 2692 np = dnp; 2693 } else { 2694 error = nfs_nget(vp->v_mount, fhp, 2695 fhsize, &np); 2696 if (!error) 2697 newvp = NFSTOV(np); 2698 } 2699 if (!error) { 2700 const char *cp; 2701 2702 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2703 dp->d_type = 2704 IFTODT(VTTOIF(np->n_vattr->va_type)); 2705 ndp->ni_vp = newvp; 2706 cp = cnp->cn_nameptr + cnp->cn_namelen; 2707 cnp->cn_hash = 2708 namei_hash(cnp->cn_nameptr, &cp); 2709 if (cnp->cn_namelen <= NCHNAMLEN) 2710 cache_enter(ndp->ni_dvp, ndp->ni_vp, 2711 cnp); 2712 } 2713 } 2714 } else { 2715 /* Just skip over the file handle */ 2716 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2717 i = fxdr_unsigned(int, *tl); 2718 nfsm_adv(nfsm_rndup(i)); 2719 } 2720 if (newvp != NULLVP) { 2721 if (newvp == vp) 2722 vrele(newvp); 2723 else 2724 vput(newvp); 2725 newvp = NULLVP; 2726 } 2727 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2728 more_dirs = fxdr_unsigned(int, *tl); 2729 } 2730 /* 2731 * If at end of rpc data, get the eof boolean 2732 */ 2733 if (!more_dirs) { 2734 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2735 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2736 } 2737 m_freem(mrep); 2738 } 2739 /* 2740 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2741 * by increasing d_reclen for the last record. 2742 */ 2743 if (blksiz > 0) { 2744 left = NFS_DIRFRAGSIZ - blksiz; 2745 dp->d_reclen += left; 2746 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2747 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 2748 left; 2749 uiop->uio_iov->iov_len -= left; 2750 uiop->uio_resid -= left; 2751 } 2752 2753 /* 2754 * We are now either at the end of the directory or have filled the 2755 * block. 2756 */ 2757 if (bigenough) 2758 dnp->n_direofoffset = uiop->uio_offset; 2759 nfsmout: 2760 if (newvp != NULLVP) { 2761 if(newvp == vp) 2762 vrele(newvp); 2763 else 2764 vput(newvp); 2765 } 2766 return (error); 2767 } 2768 static char hextoasc[] = "0123456789abcdef"; 2769 2770 /* 2771 * Silly rename. To make the NFS filesystem that is stateless look a little 2772 * more like the "ufs" a remove of an active vnode is translated to a rename 2773 * to a funny looking filename that is removed by nfs_inactive on the 2774 * nfsnode. There is the potential for another process on a different client 2775 * to create the same funny name between the nfs_lookitup() fails and the 2776 * nfs_rename() completes, but... 2777 */ 2778 int 2779 nfs_sillyrename(dvp, vp, cnp) 2780 struct vnode *dvp, *vp; 2781 struct componentname *cnp; 2782 { 2783 struct sillyrename *sp; 2784 struct nfsnode *np; 2785 int error; 2786 short pid; 2787 2788 cache_purge(dvp); 2789 np = VTONFS(vp); 2790 #ifndef DIAGNOSTIC 2791 if (vp->v_type == VDIR) 2792 panic("nfs: sillyrename dir"); 2793 #endif 2794 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2795 M_NFSREQ, M_WAITOK); 2796 sp->s_cred = crdup(cnp->cn_cred); 2797 sp->s_dvp = dvp; 2798 VREF(dvp); 2799 2800 /* Fudge together a funny name */ 2801 pid = cnp->cn_proc->p_pid; 2802 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2803 sp->s_namlen = 12; 2804 sp->s_name[8] = hextoasc[pid & 0xf]; 2805 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf]; 2806 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf]; 2807 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf]; 2808 2809 /* Try lookitups until we get one that isn't there */ 2810 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2811 cnp->cn_proc, (struct nfsnode **)0) == 0) { 2812 sp->s_name[4]++; 2813 if (sp->s_name[4] > 'z') { 2814 error = EINVAL; 2815 goto bad; 2816 } 2817 } 2818 error = nfs_renameit(dvp, cnp, sp); 2819 if (error) 2820 goto bad; 2821 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2822 cnp->cn_proc, &np); 2823 np->n_sillyrename = sp; 2824 return (0); 2825 bad: 2826 vrele(sp->s_dvp); 2827 crfree(sp->s_cred); 2828 free((caddr_t)sp, M_NFSREQ); 2829 return (error); 2830 } 2831 2832 /* 2833 * Look up a file name and optionally either update the file handle or 2834 * allocate an nfsnode, depending on the value of npp. 2835 * npp == NULL --> just do the lookup 2836 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2837 * handled too 2838 * *npp != NULL --> update the file handle in the vnode 2839 */ 2840 int 2841 nfs_lookitup(dvp, name, len, cred, procp, npp) 2842 struct vnode *dvp; 2843 const char *name; 2844 int len; 2845 struct ucred *cred; 2846 struct proc *procp; 2847 struct nfsnode **npp; 2848 { 2849 u_int32_t *tl; 2850 caddr_t cp; 2851 int32_t t1, t2; 2852 struct vnode *newvp = (struct vnode *)0; 2853 struct nfsnode *np, *dnp = VTONFS(dvp); 2854 caddr_t bpos, dpos, cp2; 2855 int error = 0, fhlen, attrflag; 2856 struct mbuf *mreq, *mrep, *md, *mb; 2857 nfsfh_t *nfhp; 2858 const int v3 = NFS_ISV3(dvp); 2859 2860 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2861 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2862 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2863 nfsm_fhtom(dnp, v3); 2864 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2865 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred); 2866 if (npp && !error) { 2867 nfsm_getfh(nfhp, fhlen, v3); 2868 if (*npp) { 2869 np = *npp; 2870 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2871 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2872 np->n_fhp = &np->n_fh; 2873 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2874 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2875 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen); 2876 np->n_fhsize = fhlen; 2877 newvp = NFSTOV(np); 2878 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2879 VREF(dvp); 2880 newvp = dvp; 2881 } else { 2882 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2883 if (error) { 2884 m_freem(mrep); 2885 return (error); 2886 } 2887 newvp = NFSTOV(np); 2888 } 2889 if (v3) { 2890 nfsm_postop_attr(newvp, attrflag, 0); 2891 if (!attrflag && *npp == NULL) { 2892 m_freem(mrep); 2893 vput(newvp); 2894 return (ENOENT); 2895 } 2896 } else 2897 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2898 } 2899 nfsm_reqdone; 2900 if (npp && *npp == NULL) { 2901 if (error) { 2902 if (newvp) 2903 vput(newvp); 2904 } else 2905 *npp = np; 2906 } 2907 return (error); 2908 } 2909 2910 /* 2911 * Nfs Version 3 commit rpc 2912 */ 2913 int 2914 nfs_commit(vp, offset, cnt, procp) 2915 struct vnode *vp; 2916 off_t offset; 2917 uint32_t cnt; 2918 struct proc *procp; 2919 { 2920 caddr_t cp; 2921 u_int32_t *tl; 2922 int32_t t1, t2; 2923 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2924 caddr_t bpos, dpos, cp2; 2925 int error = 0, wccflag = NFSV3_WCCRATTR; 2926 struct mbuf *mreq, *mrep, *md, *mb; 2927 struct nfsnode *np; 2928 2929 KASSERT(NFS_ISV3(vp)); 2930 2931 #ifdef NFS_DEBUG_COMMIT 2932 printf("commit %lu - %lu\n", (unsigned long)offset, 2933 (unsigned long)(offset + cnt)); 2934 #endif 2935 2936 simple_lock(&nmp->nm_slock); 2937 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 2938 simple_unlock(&nmp->nm_slock); 2939 return (0); 2940 } 2941 simple_unlock(&nmp->nm_slock); 2942 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2943 np = VTONFS(vp); 2944 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 2945 nfsm_fhtom(np, 1); 2946 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2947 txdr_hyper(offset, tl); 2948 tl += 2; 2949 *tl = txdr_unsigned(cnt); 2950 nfsm_request(np, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred); 2951 nfsm_wcc_data(vp, wccflag, 0); 2952 if (!error) { 2953 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2954 simple_lock(&nmp->nm_slock); 2955 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 2956 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 2957 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 2958 error = NFSERR_STALEWRITEVERF; 2959 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 2960 } 2961 simple_unlock(&nmp->nm_slock); 2962 } 2963 nfsm_reqdone; 2964 return (error); 2965 } 2966 2967 /* 2968 * Kludge City.. 2969 * - make nfs_bmap() essentially a no-op that does no translation 2970 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2971 * (Maybe I could use the process's page mapping, but I was concerned that 2972 * Kernel Write might not be enabled and also figured copyout() would do 2973 * a lot more work than memcpy() and also it currently happens in the 2974 * context of the swapper process (2). 2975 */ 2976 int 2977 nfs_bmap(v) 2978 void *v; 2979 { 2980 struct vop_bmap_args /* { 2981 struct vnode *a_vp; 2982 daddr_t a_bn; 2983 struct vnode **a_vpp; 2984 daddr_t *a_bnp; 2985 int *a_runp; 2986 } */ *ap = v; 2987 struct vnode *vp = ap->a_vp; 2988 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 2989 2990 if (ap->a_vpp != NULL) 2991 *ap->a_vpp = vp; 2992 if (ap->a_bnp != NULL) 2993 *ap->a_bnp = ap->a_bn << bshift; 2994 if (ap->a_runp != NULL) 2995 *ap->a_runp = 1024 * 1024; /* XXX */ 2996 return (0); 2997 } 2998 2999 /* 3000 * Strategy routine. 3001 * For async requests when nfsiod(s) are running, queue the request by 3002 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3003 * request. 3004 */ 3005 int 3006 nfs_strategy(v) 3007 void *v; 3008 { 3009 struct vop_strategy_args *ap = v; 3010 struct buf *bp = ap->a_bp; 3011 struct proc *p; 3012 int error = 0; 3013 3014 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3015 panic("nfs physio/async"); 3016 if (bp->b_flags & B_ASYNC) 3017 p = NULL; 3018 else 3019 p = curproc; /* XXX */ 3020 3021 /* 3022 * If the op is asynchronous and an i/o daemon is waiting 3023 * queue the request, wake it up and wait for completion 3024 * otherwise just do it ourselves. 3025 */ 3026 3027 if ((bp->b_flags & B_ASYNC) == 0 || 3028 nfs_asyncio(bp)) 3029 error = nfs_doio(bp, p); 3030 return (error); 3031 } 3032 3033 /* 3034 * fsync vnode op. Just call nfs_flush() with commit == 1. 3035 */ 3036 /* ARGSUSED */ 3037 int 3038 nfs_fsync(v) 3039 void *v; 3040 { 3041 struct vop_fsync_args /* { 3042 struct vnodeop_desc *a_desc; 3043 struct vnode * a_vp; 3044 struct ucred * a_cred; 3045 int a_flags; 3046 off_t offlo; 3047 off_t offhi; 3048 struct proc * a_p; 3049 } */ *ap = v; 3050 3051 return (nfs_flush(ap->a_vp, ap->a_cred, 3052 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1)); 3053 } 3054 3055 /* 3056 * Flush all the data associated with a vnode. 3057 */ 3058 int 3059 nfs_flush(vp, cred, waitfor, p, commit) 3060 struct vnode *vp; 3061 struct ucred *cred; 3062 int waitfor; 3063 struct proc *p; 3064 int commit; 3065 { 3066 struct nfsnode *np = VTONFS(vp); 3067 int error; 3068 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3069 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3070 3071 simple_lock(&vp->v_interlock); 3072 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3073 if (np->n_flag & NWRITEERR) { 3074 error = np->n_error; 3075 np->n_flag &= ~NWRITEERR; 3076 } 3077 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3078 return (error); 3079 } 3080 3081 /* 3082 * Return POSIX pathconf information applicable to nfs. 3083 * 3084 * N.B. The NFS V2 protocol doesn't support this RPC. 3085 */ 3086 /* ARGSUSED */ 3087 int 3088 nfs_pathconf(v) 3089 void *v; 3090 { 3091 struct vop_pathconf_args /* { 3092 struct vnode *a_vp; 3093 int a_name; 3094 register_t *a_retval; 3095 } */ *ap = v; 3096 struct nfsv3_pathconf *pcp; 3097 struct vnode *vp = ap->a_vp; 3098 struct mbuf *mreq, *mrep, *md, *mb; 3099 int32_t t1, t2; 3100 u_int32_t *tl; 3101 caddr_t bpos, dpos, cp, cp2; 3102 int error = 0, attrflag; 3103 #ifndef NFS_V2_ONLY 3104 struct nfsmount *nmp; 3105 unsigned int l; 3106 u_int64_t maxsize; 3107 #endif 3108 const int v3 = NFS_ISV3(vp); 3109 struct nfsnode *np = VTONFS(vp); 3110 3111 switch (ap->a_name) { 3112 /* Names that can be resolved locally. */ 3113 case _PC_PIPE_BUF: 3114 *ap->a_retval = PIPE_BUF; 3115 break; 3116 case _PC_SYNC_IO: 3117 *ap->a_retval = 1; 3118 break; 3119 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3120 case _PC_LINK_MAX: 3121 case _PC_NAME_MAX: 3122 case _PC_CHOWN_RESTRICTED: 3123 case _PC_NO_TRUNC: 3124 if (!v3) { 3125 error = EINVAL; 3126 break; 3127 } 3128 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3129 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3130 nfsm_fhtom(np, 1); 3131 nfsm_request(np, NFSPROC_PATHCONF, 3132 curproc, curproc->p_ucred); /* XXX */ 3133 nfsm_postop_attr(vp, attrflag, 0); 3134 if (!error) { 3135 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3136 NFSX_V3PATHCONF); 3137 switch (ap->a_name) { 3138 case _PC_LINK_MAX: 3139 *ap->a_retval = 3140 fxdr_unsigned(register_t, pcp->pc_linkmax); 3141 break; 3142 case _PC_NAME_MAX: 3143 *ap->a_retval = 3144 fxdr_unsigned(register_t, pcp->pc_namemax); 3145 break; 3146 case _PC_CHOWN_RESTRICTED: 3147 *ap->a_retval = 3148 (pcp->pc_chownrestricted == nfs_true); 3149 break; 3150 case _PC_NO_TRUNC: 3151 *ap->a_retval = 3152 (pcp->pc_notrunc == nfs_true); 3153 break; 3154 } 3155 } 3156 nfsm_reqdone; 3157 break; 3158 case _PC_FILESIZEBITS: 3159 #ifndef NFS_V2_ONLY 3160 if (v3) { 3161 nmp = VFSTONFS(vp->v_mount); 3162 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3163 if ((error = nfs_fsinfo(nmp, vp, 3164 curproc->p_ucred, curproc)) != 0) /* XXX */ 3165 break; 3166 for (l = 0, maxsize = nmp->nm_maxfilesize; 3167 (maxsize >> l) > 0; l++) 3168 ; 3169 *ap->a_retval = l + 1; 3170 } else 3171 #endif 3172 { 3173 *ap->a_retval = 32; /* NFS V2 limitation */ 3174 } 3175 break; 3176 default: 3177 error = EINVAL; 3178 break; 3179 } 3180 3181 return (error); 3182 } 3183 3184 /* 3185 * NFS advisory byte-level locks. 3186 */ 3187 int 3188 nfs_advlock(v) 3189 void *v; 3190 { 3191 struct vop_advlock_args /* { 3192 struct vnode *a_vp; 3193 caddr_t a_id; 3194 int a_op; 3195 struct flock *a_fl; 3196 int a_flags; 3197 } */ *ap = v; 3198 struct nfsnode *np = VTONFS(ap->a_vp); 3199 3200 return lf_advlock(ap, &np->n_lockf, np->n_size); 3201 } 3202 3203 /* 3204 * Print out the contents of an nfsnode. 3205 */ 3206 int 3207 nfs_print(v) 3208 void *v; 3209 { 3210 struct vop_print_args /* { 3211 struct vnode *a_vp; 3212 } */ *ap = v; 3213 struct vnode *vp = ap->a_vp; 3214 struct nfsnode *np = VTONFS(vp); 3215 3216 printf("tag VT_NFS, fileid %ld fsid 0x%lx", 3217 np->n_vattr->va_fileid, np->n_vattr->va_fsid); 3218 if (vp->v_type == VFIFO) 3219 fifo_printinfo(vp); 3220 printf("\n"); 3221 return (0); 3222 } 3223 3224 /* 3225 * NFS file truncation. 3226 */ 3227 int 3228 nfs_truncate(v) 3229 void *v; 3230 { 3231 #if 0 3232 struct vop_truncate_args /* { 3233 struct vnode *a_vp; 3234 off_t a_length; 3235 int a_flags; 3236 struct ucred *a_cred; 3237 struct proc *a_p; 3238 } */ *ap = v; 3239 #endif 3240 3241 /* Use nfs_setattr */ 3242 return (EOPNOTSUPP); 3243 } 3244 3245 /* 3246 * NFS update. 3247 */ 3248 int 3249 nfs_update(v) 3250 void *v; 3251 #if 0 3252 struct vop_update_args /* { 3253 struct vnode *a_vp; 3254 struct timespec *a_ta; 3255 struct timespec *a_tm; 3256 int a_waitfor; 3257 } */ *ap = v; 3258 #endif 3259 { 3260 3261 /* Use nfs_setattr */ 3262 return (EOPNOTSUPP); 3263 } 3264 3265 /* 3266 * Just call bwrite(). 3267 */ 3268 int 3269 nfs_bwrite(v) 3270 void *v; 3271 { 3272 struct vop_bwrite_args /* { 3273 struct vnode *a_bp; 3274 } */ *ap = v; 3275 3276 return (bwrite(ap->a_bp)); 3277 } 3278 3279 /* 3280 * nfs unlock wrapper. 3281 */ 3282 int 3283 nfs_unlock(void *v) 3284 { 3285 struct vop_unlock_args /* { 3286 struct vnode *a_vp; 3287 int a_flags; 3288 } */ *ap = v; 3289 struct vnode *vp = ap->a_vp; 3290 3291 /* 3292 * VOP_UNLOCK can be called by nfs_loadattrcache 3293 * with v_data == 0. 3294 */ 3295 if (VTONFS(vp)) { 3296 nfs_delayedtruncate(vp); 3297 } 3298 3299 return genfs_unlock(v); 3300 } 3301 3302 /* 3303 * nfs special file access vnode op. 3304 * Essentially just get vattr and then imitate iaccess() since the device is 3305 * local to the client. 3306 */ 3307 int 3308 nfsspec_access(v) 3309 void *v; 3310 { 3311 struct vop_access_args /* { 3312 struct vnode *a_vp; 3313 int a_mode; 3314 struct ucred *a_cred; 3315 struct proc *a_p; 3316 } */ *ap = v; 3317 struct vattr va; 3318 struct vnode *vp = ap->a_vp; 3319 int error; 3320 3321 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p); 3322 if (error) 3323 return (error); 3324 3325 /* 3326 * Disallow write attempts on filesystems mounted read-only; 3327 * unless the file is a socket, fifo, or a block or character 3328 * device resident on the filesystem. 3329 */ 3330 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3331 switch (vp->v_type) { 3332 case VREG: 3333 case VDIR: 3334 case VLNK: 3335 return (EROFS); 3336 default: 3337 break; 3338 } 3339 } 3340 3341 return (vaccess(va.va_type, va.va_mode, 3342 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3343 } 3344 3345 /* 3346 * Read wrapper for special devices. 3347 */ 3348 int 3349 nfsspec_read(v) 3350 void *v; 3351 { 3352 struct vop_read_args /* { 3353 struct vnode *a_vp; 3354 struct uio *a_uio; 3355 int a_ioflag; 3356 struct ucred *a_cred; 3357 } */ *ap = v; 3358 struct nfsnode *np = VTONFS(ap->a_vp); 3359 3360 /* 3361 * Set access flag. 3362 */ 3363 np->n_flag |= NACC; 3364 np->n_atim.tv_sec = time.tv_sec; 3365 np->n_atim.tv_nsec = time.tv_usec * 1000; 3366 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3367 } 3368 3369 /* 3370 * Write wrapper for special devices. 3371 */ 3372 int 3373 nfsspec_write(v) 3374 void *v; 3375 { 3376 struct vop_write_args /* { 3377 struct vnode *a_vp; 3378 struct uio *a_uio; 3379 int a_ioflag; 3380 struct ucred *a_cred; 3381 } */ *ap = v; 3382 struct nfsnode *np = VTONFS(ap->a_vp); 3383 3384 /* 3385 * Set update flag. 3386 */ 3387 np->n_flag |= NUPD; 3388 np->n_mtim.tv_sec = time.tv_sec; 3389 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3390 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3391 } 3392 3393 /* 3394 * Close wrapper for special devices. 3395 * 3396 * Update the times on the nfsnode then do device close. 3397 */ 3398 int 3399 nfsspec_close(v) 3400 void *v; 3401 { 3402 struct vop_close_args /* { 3403 struct vnode *a_vp; 3404 int a_fflag; 3405 struct ucred *a_cred; 3406 struct proc *a_p; 3407 } */ *ap = v; 3408 struct vnode *vp = ap->a_vp; 3409 struct nfsnode *np = VTONFS(vp); 3410 struct vattr vattr; 3411 3412 if (np->n_flag & (NACC | NUPD)) { 3413 np->n_flag |= NCHG; 3414 if (vp->v_usecount == 1 && 3415 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3416 VATTR_NULL(&vattr); 3417 if (np->n_flag & NACC) 3418 vattr.va_atime = np->n_atim; 3419 if (np->n_flag & NUPD) 3420 vattr.va_mtime = np->n_mtim; 3421 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3422 } 3423 } 3424 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3425 } 3426 3427 /* 3428 * Read wrapper for fifos. 3429 */ 3430 int 3431 nfsfifo_read(v) 3432 void *v; 3433 { 3434 struct vop_read_args /* { 3435 struct vnode *a_vp; 3436 struct uio *a_uio; 3437 int a_ioflag; 3438 struct ucred *a_cred; 3439 } */ *ap = v; 3440 struct nfsnode *np = VTONFS(ap->a_vp); 3441 3442 /* 3443 * Set access flag. 3444 */ 3445 np->n_flag |= NACC; 3446 np->n_atim.tv_sec = time.tv_sec; 3447 np->n_atim.tv_nsec = time.tv_usec * 1000; 3448 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3449 } 3450 3451 /* 3452 * Write wrapper for fifos. 3453 */ 3454 int 3455 nfsfifo_write(v) 3456 void *v; 3457 { 3458 struct vop_write_args /* { 3459 struct vnode *a_vp; 3460 struct uio *a_uio; 3461 int a_ioflag; 3462 struct ucred *a_cred; 3463 } */ *ap = v; 3464 struct nfsnode *np = VTONFS(ap->a_vp); 3465 3466 /* 3467 * Set update flag. 3468 */ 3469 np->n_flag |= NUPD; 3470 np->n_mtim.tv_sec = time.tv_sec; 3471 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3472 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3473 } 3474 3475 /* 3476 * Close wrapper for fifos. 3477 * 3478 * Update the times on the nfsnode then do fifo close. 3479 */ 3480 int 3481 nfsfifo_close(v) 3482 void *v; 3483 { 3484 struct vop_close_args /* { 3485 struct vnode *a_vp; 3486 int a_fflag; 3487 struct ucred *a_cred; 3488 struct proc *a_p; 3489 } */ *ap = v; 3490 struct vnode *vp = ap->a_vp; 3491 struct nfsnode *np = VTONFS(vp); 3492 struct vattr vattr; 3493 3494 if (np->n_flag & (NACC | NUPD)) { 3495 if (np->n_flag & NACC) { 3496 np->n_atim.tv_sec = time.tv_sec; 3497 np->n_atim.tv_nsec = time.tv_usec * 1000; 3498 } 3499 if (np->n_flag & NUPD) { 3500 np->n_mtim.tv_sec = time.tv_sec; 3501 np->n_mtim.tv_nsec = time.tv_usec * 1000; 3502 } 3503 np->n_flag |= NCHG; 3504 if (vp->v_usecount == 1 && 3505 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3506 VATTR_NULL(&vattr); 3507 if (np->n_flag & NACC) 3508 vattr.va_atime = np->n_atim; 3509 if (np->n_flag & NUPD) 3510 vattr.va_mtime = np->n_mtim; 3511 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p); 3512 } 3513 } 3514 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3515 } 3516
Indexes created Mon Sep 22 13:09:51 GMT 2025