nfs_vnops.c revision 1.296
1 /* $NetBSD: nfs_vnops.c,v 1.296 2012/11/05 17:24:11 dholland 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 1.296 2012/11/05 17:24:11 dholland Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/proc.h> 51 #include <sys/kernel.h> 52 #include <sys/systm.h> 53 #include <sys/resourcevar.h> 54 #include <sys/mount.h> 55 #include <sys/buf.h> 56 #include <sys/condvar.h> 57 #include <sys/disk.h> 58 #include <sys/malloc.h> 59 #include <sys/kmem.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/vnode.h> 64 #include <sys/dirent.h> 65 #include <sys/fcntl.h> 66 #include <sys/hash.h> 67 #include <sys/lockf.h> 68 #include <sys/stat.h> 69 #include <sys/unistd.h> 70 #include <sys/kauth.h> 71 #include <sys/cprng.h> 72 73 #include <uvm/uvm_extern.h> 74 #include <uvm/uvm.h> 75 76 #include <miscfs/fifofs/fifo.h> 77 #include <miscfs/genfs/genfs.h> 78 #include <miscfs/genfs/genfs_node.h> 79 #include <miscfs/specfs/specdev.h> 80 81 #include <nfs/rpcv2.h> 82 #include <nfs/nfsproto.h> 83 #include <nfs/nfs.h> 84 #include <nfs/nfsnode.h> 85 #include <nfs/nfsmount.h> 86 #include <nfs/xdr_subs.h> 87 #include <nfs/nfsm_subs.h> 88 #include <nfs/nfs_var.h> 89 90 #include <net/if.h> 91 #include <netinet/in.h> 92 #include <netinet/in_var.h> 93 94 /* 95 * Global vfs data structures for nfs 96 */ 97 int (**nfsv2_vnodeop_p)(void *); 98 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 99 { &vop_default_desc, vn_default_error }, 100 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 101 { &vop_create_desc, nfs_create }, /* create */ 102 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 103 { &vop_open_desc, nfs_open }, /* open */ 104 { &vop_close_desc, nfs_close }, /* close */ 105 { &vop_access_desc, nfs_access }, /* access */ 106 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 107 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 108 { &vop_read_desc, nfs_read }, /* read */ 109 { &vop_write_desc, nfs_write }, /* write */ 110 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 111 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 112 { &vop_poll_desc, nfs_poll }, /* poll */ 113 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 114 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 115 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 116 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 117 { &vop_seek_desc, nfs_seek }, /* seek */ 118 { &vop_remove_desc, nfs_remove }, /* remove */ 119 { &vop_link_desc, nfs_link }, /* link */ 120 { &vop_rename_desc, nfs_rename }, /* rename */ 121 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 122 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 123 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 124 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 125 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 126 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 127 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 128 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 129 { &vop_lock_desc, nfs_lock }, /* lock */ 130 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 131 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 132 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 133 { &vop_print_desc, nfs_print }, /* print */ 134 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 135 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 136 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 137 { &vop_bwrite_desc, genfs_badop }, /* 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)(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_fcntl_desc, genfs_fcntl }, /* fcntl */ 162 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 163 { &vop_poll_desc, spec_poll }, /* poll */ 164 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 165 { &vop_revoke_desc, spec_revoke }, /* revoke */ 166 { &vop_mmap_desc, spec_mmap }, /* mmap */ 167 { &vop_fsync_desc, spec_fsync }, /* fsync */ 168 { &vop_seek_desc, spec_seek }, /* seek */ 169 { &vop_remove_desc, spec_remove }, /* remove */ 170 { &vop_link_desc, spec_link }, /* link */ 171 { &vop_rename_desc, spec_rename }, /* rename */ 172 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 173 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 174 { &vop_symlink_desc, spec_symlink }, /* symlink */ 175 { &vop_readdir_desc, spec_readdir }, /* readdir */ 176 { &vop_readlink_desc, spec_readlink }, /* readlink */ 177 { &vop_abortop_desc, spec_abortop }, /* abortop */ 178 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 179 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 180 { &vop_lock_desc, nfs_lock }, /* lock */ 181 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 182 { &vop_bmap_desc, spec_bmap }, /* bmap */ 183 { &vop_strategy_desc, spec_strategy }, /* strategy */ 184 { &vop_print_desc, nfs_print }, /* print */ 185 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 186 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 187 { &vop_advlock_desc, spec_advlock }, /* advlock */ 188 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 189 { &vop_getpages_desc, spec_getpages }, /* getpages */ 190 { &vop_putpages_desc, spec_putpages }, /* putpages */ 191 { NULL, NULL } 192 }; 193 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 194 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 195 196 int (**fifo_nfsv2nodeop_p)(void *); 197 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 198 { &vop_default_desc, vn_default_error }, 199 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 200 { &vop_create_desc, vn_fifo_bypass }, /* create */ 201 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 202 { &vop_open_desc, vn_fifo_bypass }, /* open */ 203 { &vop_close_desc, nfsfifo_close }, /* close */ 204 { &vop_access_desc, nfsspec_access }, /* access */ 205 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 206 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 207 { &vop_read_desc, nfsfifo_read }, /* read */ 208 { &vop_write_desc, nfsfifo_write }, /* write */ 209 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 210 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 211 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 212 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 213 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 214 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 215 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 216 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 217 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 218 { &vop_link_desc, vn_fifo_bypass }, /* link */ 219 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 220 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 221 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 222 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 223 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 224 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 225 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 226 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 227 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 228 { &vop_lock_desc, nfs_lock }, /* lock */ 229 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 230 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 231 { &vop_strategy_desc, genfs_badop }, /* strategy */ 232 { &vop_print_desc, nfs_print }, /* print */ 233 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 234 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 235 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 236 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 237 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 238 { NULL, NULL } 239 }; 240 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 241 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 242 243 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 244 size_t, kauth_cred_t, struct lwp *); 245 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 246 247 /* 248 * Global variables 249 */ 250 extern u_int32_t nfs_true, nfs_false; 251 extern u_int32_t nfs_xdrneg1; 252 extern const nfstype nfsv3_type[9]; 253 254 int nfs_numasync = 0; 255 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 256 #define UIO_ADVANCE(uio, siz) \ 257 (void)((uio)->uio_resid -= (siz), \ 258 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 259 (uio)->uio_iov->iov_len -= (siz)) 260 261 static void nfs_cache_enter(struct vnode *, struct vnode *, 262 struct componentname *); 263 264 static void 265 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 266 struct componentname *cnp) 267 { 268 struct nfsnode *dnp = VTONFS(dvp); 269 270 if ((cnp->cn_flags & MAKEENTRY) == 0) { 271 return; 272 } 273 if (vp != NULL) { 274 struct nfsnode *np = VTONFS(vp); 275 276 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 277 } 278 279 if (!timespecisset(&dnp->n_nctime)) 280 dnp->n_nctime = dnp->n_vattr->va_mtime; 281 282 cache_enter(dvp, vp, cnp); 283 } 284 285 /* 286 * nfs null call from vfs. 287 */ 288 int 289 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 290 { 291 char *bpos, *dpos; 292 int error = 0; 293 struct mbuf *mreq, *mrep, *md, *mb; 294 struct nfsnode *np = VTONFS(vp); 295 296 nfsm_reqhead(np, NFSPROC_NULL, 0); 297 nfsm_request(np, NFSPROC_NULL, l, cred); 298 nfsm_reqdone; 299 return (error); 300 } 301 302 /* 303 * nfs access vnode op. 304 * For nfs version 2, just return ok. File accesses may fail later. 305 * For nfs version 3, use the access rpc to check accessibility. If file modes 306 * are changed on the server, accesses might still fail later. 307 */ 308 int 309 nfs_access(void *v) 310 { 311 struct vop_access_args /* { 312 struct vnode *a_vp; 313 int a_mode; 314 kauth_cred_t a_cred; 315 } */ *ap = v; 316 struct vnode *vp = ap->a_vp; 317 #ifndef NFS_V2_ONLY 318 u_int32_t *tl; 319 char *cp; 320 int32_t t1, t2; 321 char *bpos, *dpos, *cp2; 322 int error = 0, attrflag; 323 struct mbuf *mreq, *mrep, *md, *mb; 324 u_int32_t mode, rmode; 325 const int v3 = NFS_ISV3(vp); 326 #endif 327 int cachevalid; 328 struct nfsnode *np = VTONFS(vp); 329 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 330 331 cachevalid = (np->n_accstamp != -1 && 332 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 333 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 334 335 /* 336 * Check access cache first. If this request has been made for this 337 * uid shortly before, use the cached result. 338 */ 339 if (cachevalid) { 340 if (!np->n_accerror) { 341 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 342 return np->n_accerror; 343 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 344 return np->n_accerror; 345 } 346 347 #ifndef NFS_V2_ONLY 348 /* 349 * For nfs v3, do an access rpc, otherwise you are stuck emulating 350 * ufs_access() locally using the vattr. This may not be correct, 351 * since the server may apply other access criteria such as 352 * client uid-->server uid mapping that we do not know about, but 353 * this is better than just returning anything that is lying about 354 * in the cache. 355 */ 356 if (v3) { 357 nfsstats.rpccnt[NFSPROC_ACCESS]++; 358 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 359 nfsm_fhtom(np, v3); 360 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 361 if (ap->a_mode & VREAD) 362 mode = NFSV3ACCESS_READ; 363 else 364 mode = 0; 365 if (vp->v_type != VDIR) { 366 if (ap->a_mode & VWRITE) 367 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 368 if (ap->a_mode & VEXEC) 369 mode |= NFSV3ACCESS_EXECUTE; 370 } else { 371 if (ap->a_mode & VWRITE) 372 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 373 NFSV3ACCESS_DELETE); 374 if (ap->a_mode & VEXEC) 375 mode |= NFSV3ACCESS_LOOKUP; 376 } 377 *tl = txdr_unsigned(mode); 378 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 379 nfsm_postop_attr(vp, attrflag, 0); 380 if (!error) { 381 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 382 rmode = fxdr_unsigned(u_int32_t, *tl); 383 /* 384 * The NFS V3 spec does not clarify whether or not 385 * the returned access bits can be a superset of 386 * the ones requested, so... 387 */ 388 if ((rmode & mode) != mode) 389 error = EACCES; 390 } 391 nfsm_reqdone; 392 } else 393 #endif 394 return (nfsspec_access(ap)); 395 #ifndef NFS_V2_ONLY 396 /* 397 * Disallow write attempts on filesystems mounted read-only; 398 * unless the file is a socket, fifo, or a block or character 399 * device resident on the filesystem. 400 */ 401 if (!error && (ap->a_mode & VWRITE) && 402 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 403 switch (vp->v_type) { 404 case VREG: 405 case VDIR: 406 case VLNK: 407 error = EROFS; 408 default: 409 break; 410 } 411 } 412 413 if (!error || error == EACCES) { 414 /* 415 * If we got the same result as for a previous, 416 * different request, OR it in. Don't update 417 * the timestamp in that case. 418 */ 419 if (cachevalid && np->n_accstamp != -1 && 420 error == np->n_accerror) { 421 if (!error) 422 np->n_accmode |= ap->a_mode; 423 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 424 np->n_accmode = ap->a_mode; 425 } else { 426 np->n_accstamp = time_uptime; 427 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 428 np->n_accmode = ap->a_mode; 429 np->n_accerror = error; 430 } 431 } 432 433 return (error); 434 #endif 435 } 436 437 /* 438 * nfs open vnode op 439 * Check to see if the type is ok 440 * and that deletion is not in progress. 441 * For paged in text files, you will need to flush the page cache 442 * if consistency is lost. 443 */ 444 /* ARGSUSED */ 445 int 446 nfs_open(void *v) 447 { 448 struct vop_open_args /* { 449 struct vnode *a_vp; 450 int a_mode; 451 kauth_cred_t a_cred; 452 } */ *ap = v; 453 struct vnode *vp = ap->a_vp; 454 struct nfsnode *np = VTONFS(vp); 455 int error; 456 457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 458 return (EACCES); 459 } 460 461 if (ap->a_mode & FREAD) { 462 if (np->n_rcred != NULL) 463 kauth_cred_free(np->n_rcred); 464 np->n_rcred = ap->a_cred; 465 kauth_cred_hold(np->n_rcred); 466 } 467 if (ap->a_mode & FWRITE) { 468 if (np->n_wcred != NULL) 469 kauth_cred_free(np->n_wcred); 470 np->n_wcred = ap->a_cred; 471 kauth_cred_hold(np->n_wcred); 472 } 473 474 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 475 if (error) 476 return error; 477 478 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 479 480 return (0); 481 } 482 483 /* 484 * nfs close vnode op 485 * What an NFS client should do upon close after writing is a debatable issue. 486 * Most NFS clients push delayed writes to the server upon close, basically for 487 * two reasons: 488 * 1 - So that any write errors may be reported back to the client process 489 * doing the close system call. By far the two most likely errors are 490 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 491 * 2 - To put a worst case upper bound on cache inconsistency between 492 * multiple clients for the file. 493 * There is also a consistency problem for Version 2 of the protocol w.r.t. 494 * not being able to tell if other clients are writing a file concurrently, 495 * since there is no way of knowing if the changed modify time in the reply 496 * is only due to the write for this client. 497 * (NFS Version 3 provides weak cache consistency data in the reply that 498 * should be sufficient to detect and handle this case.) 499 * 500 * The current code does the following: 501 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 502 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 503 * or commit them (this satisfies 1 and 2 except for the 504 * case where the server crashes after this close but 505 * before the commit RPC, which is felt to be "good 506 * enough". Changing the last argument to nfs_flush() to 507 * a 1 would force a commit operation, if it is felt a 508 * commit is necessary now. 509 */ 510 /* ARGSUSED */ 511 int 512 nfs_close(void *v) 513 { 514 struct vop_close_args /* { 515 struct vnodeop_desc *a_desc; 516 struct vnode *a_vp; 517 int a_fflag; 518 kauth_cred_t a_cred; 519 } */ *ap = v; 520 struct vnode *vp = ap->a_vp; 521 struct nfsnode *np = VTONFS(vp); 522 int error = 0; 523 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 524 525 if (vp->v_type == VREG) { 526 if (np->n_flag & NMODIFIED) { 527 #ifndef NFS_V2_ONLY 528 if (NFS_ISV3(vp)) { 529 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 530 np->n_flag &= ~NMODIFIED; 531 } else 532 #endif 533 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 534 NFS_INVALIDATE_ATTRCACHE(np); 535 } 536 if (np->n_flag & NWRITEERR) { 537 np->n_flag &= ~NWRITEERR; 538 error = np->n_error; 539 } 540 } 541 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 542 return (error); 543 } 544 545 /* 546 * nfs getattr call from vfs. 547 */ 548 int 549 nfs_getattr(void *v) 550 { 551 struct vop_getattr_args /* { 552 struct vnode *a_vp; 553 struct vattr *a_vap; 554 kauth_cred_t a_cred; 555 } */ *ap = v; 556 struct vnode *vp = ap->a_vp; 557 struct nfsnode *np = VTONFS(vp); 558 char *cp; 559 u_int32_t *tl; 560 int32_t t1, t2; 561 char *bpos, *dpos; 562 int error = 0; 563 struct mbuf *mreq, *mrep, *md, *mb; 564 const int v3 = NFS_ISV3(vp); 565 566 /* 567 * Update local times for special files. 568 */ 569 if (np->n_flag & (NACC | NUPD)) 570 np->n_flag |= NCHG; 571 572 /* 573 * if we have delayed truncation, do it now. 574 */ 575 nfs_delayedtruncate(vp); 576 577 /* 578 * First look in the cache. 579 */ 580 if (nfs_getattrcache(vp, ap->a_vap) == 0) 581 return (0); 582 nfsstats.rpccnt[NFSPROC_GETATTR]++; 583 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 584 nfsm_fhtom(np, v3); 585 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 586 if (!error) { 587 nfsm_loadattr(vp, ap->a_vap, 0); 588 if (vp->v_type == VDIR && 589 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 590 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 591 } 592 nfsm_reqdone; 593 return (error); 594 } 595 596 /* 597 * nfs setattr call. 598 */ 599 int 600 nfs_setattr(void *v) 601 { 602 struct vop_setattr_args /* { 603 struct vnodeop_desc *a_desc; 604 struct vnode *a_vp; 605 struct vattr *a_vap; 606 kauth_cred_t a_cred; 607 } */ *ap = v; 608 struct vnode *vp = ap->a_vp; 609 struct nfsnode *np = VTONFS(vp); 610 struct vattr *vap = ap->a_vap; 611 int error = 0; 612 u_quad_t tsize = 0; 613 614 /* 615 * Setting of flags is not supported. 616 */ 617 if (vap->va_flags != VNOVAL) 618 return (EOPNOTSUPP); 619 620 /* 621 * Disallow write attempts if the filesystem is mounted read-only. 622 */ 623 if ((vap->va_uid != (uid_t)VNOVAL || 624 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 625 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 626 (vp->v_mount->mnt_flag & MNT_RDONLY)) 627 return (EROFS); 628 if (vap->va_size != VNOVAL) { 629 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 630 return EFBIG; 631 } 632 switch (vp->v_type) { 633 case VDIR: 634 return (EISDIR); 635 case VCHR: 636 case VBLK: 637 case VSOCK: 638 case VFIFO: 639 if (vap->va_mtime.tv_sec == VNOVAL && 640 vap->va_atime.tv_sec == VNOVAL && 641 vap->va_mode == (mode_t)VNOVAL && 642 vap->va_uid == (uid_t)VNOVAL && 643 vap->va_gid == (gid_t)VNOVAL) 644 return (0); 645 vap->va_size = VNOVAL; 646 break; 647 default: 648 /* 649 * Disallow write attempts if the filesystem is 650 * mounted read-only. 651 */ 652 if (vp->v_mount->mnt_flag & MNT_RDONLY) 653 return (EROFS); 654 genfs_node_wrlock(vp); 655 uvm_vnp_setsize(vp, vap->va_size); 656 tsize = np->n_size; 657 np->n_size = vap->va_size; 658 if (vap->va_size == 0) 659 error = nfs_vinvalbuf(vp, 0, 660 ap->a_cred, curlwp, 1); 661 else 662 error = nfs_vinvalbuf(vp, V_SAVE, 663 ap->a_cred, curlwp, 1); 664 if (error) { 665 uvm_vnp_setsize(vp, tsize); 666 genfs_node_unlock(vp); 667 return (error); 668 } 669 np->n_vattr->va_size = vap->va_size; 670 } 671 } else { 672 /* 673 * flush files before setattr because a later write of 674 * cached data might change timestamps or reset sugid bits 675 */ 676 if ((vap->va_mtime.tv_sec != VNOVAL || 677 vap->va_atime.tv_sec != VNOVAL || 678 vap->va_mode != VNOVAL) && 679 vp->v_type == VREG && 680 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 681 curlwp, 1)) == EINTR) 682 return (error); 683 } 684 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 685 if (vap->va_size != VNOVAL) { 686 if (error) { 687 np->n_size = np->n_vattr->va_size = tsize; 688 uvm_vnp_setsize(vp, np->n_size); 689 } 690 genfs_node_unlock(vp); 691 } 692 VN_KNOTE(vp, NOTE_ATTRIB); 693 return (error); 694 } 695 696 /* 697 * Do an nfs setattr rpc. 698 */ 699 int 700 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 701 { 702 struct nfsv2_sattr *sp; 703 char *cp; 704 int32_t t1, t2; 705 char *bpos, *dpos; 706 u_int32_t *tl; 707 int error = 0; 708 struct mbuf *mreq, *mrep, *md, *mb; 709 const int v3 = NFS_ISV3(vp); 710 struct nfsnode *np = VTONFS(vp); 711 #ifndef NFS_V2_ONLY 712 int wccflag = NFSV3_WCCRATTR; 713 char *cp2; 714 #endif 715 716 nfsstats.rpccnt[NFSPROC_SETATTR]++; 717 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 718 nfsm_fhtom(np, v3); 719 #ifndef NFS_V2_ONLY 720 if (v3) { 721 nfsm_v3attrbuild(vap, true); 722 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 723 *tl = nfs_false; 724 } else { 725 #endif 726 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 727 if (vap->va_mode == (mode_t)VNOVAL) 728 sp->sa_mode = nfs_xdrneg1; 729 else 730 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 731 if (vap->va_uid == (uid_t)VNOVAL) 732 sp->sa_uid = nfs_xdrneg1; 733 else 734 sp->sa_uid = txdr_unsigned(vap->va_uid); 735 if (vap->va_gid == (gid_t)VNOVAL) 736 sp->sa_gid = nfs_xdrneg1; 737 else 738 sp->sa_gid = txdr_unsigned(vap->va_gid); 739 sp->sa_size = txdr_unsigned(vap->va_size); 740 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 741 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 742 #ifndef NFS_V2_ONLY 743 } 744 #endif 745 nfsm_request(np, NFSPROC_SETATTR, l, cred); 746 #ifndef NFS_V2_ONLY 747 if (v3) { 748 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 749 } else 750 #endif 751 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 752 nfsm_reqdone; 753 return (error); 754 } 755 756 /* 757 * nfs lookup call, one step at a time... 758 * First look in cache 759 * If not found, do the rpc. 760 */ 761 int 762 nfs_lookup(void *v) 763 { 764 struct vop_lookup_args /* { 765 struct vnodeop_desc *a_desc; 766 struct vnode *a_dvp; 767 struct vnode **a_vpp; 768 struct componentname *a_cnp; 769 } */ *ap = v; 770 struct componentname *cnp = ap->a_cnp; 771 struct vnode *dvp = ap->a_dvp; 772 struct vnode **vpp = ap->a_vpp; 773 int flags; 774 struct vnode *newvp; 775 u_int32_t *tl; 776 char *cp; 777 int32_t t1, t2; 778 char *bpos, *dpos, *cp2; 779 struct mbuf *mreq, *mrep, *md, *mb; 780 long len; 781 nfsfh_t *fhp; 782 struct nfsnode *np; 783 int cachefound; 784 int error = 0, attrflag, fhsize; 785 const int v3 = NFS_ISV3(dvp); 786 787 flags = cnp->cn_flags; 788 789 *vpp = NULLVP; 790 newvp = NULLVP; 791 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 792 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 793 return (EROFS); 794 if (dvp->v_type != VDIR) 795 return (ENOTDIR); 796 797 /* 798 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 799 */ 800 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 801 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 802 if (error) 803 return error; 804 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 805 return EISDIR; 806 vref(dvp); 807 *vpp = dvp; 808 return 0; 809 } 810 811 np = VTONFS(dvp); 812 813 /* 814 * Before performing an RPC, check the name cache to see if 815 * the directory/name pair we are looking for is known already. 816 * If the directory/name pair is found in the name cache, 817 * we have to ensure the directory has not changed from 818 * the time the cache entry has been created. If it has, 819 * the cache entry has to be ignored. 820 */ 821 cachefound = cache_lookup_raw(dvp, cnp, NULL, vpp); 822 KASSERT(dvp != *vpp); 823 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 824 if (cachefound) { 825 struct vattr vattr; 826 827 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 828 if (error != 0) { 829 if (*vpp != NULLVP) 830 vrele(*vpp); 831 *vpp = NULLVP; 832 return error; 833 } 834 835 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 836 || timespeccmp(&vattr.va_mtime, 837 &VTONFS(dvp)->n_nctime, !=)) { 838 if (*vpp != NULLVP) { 839 vrele(*vpp); 840 *vpp = NULLVP; 841 } 842 cache_purge1(dvp, NULL, PURGE_CHILDREN); 843 timespecclear(&np->n_nctime); 844 goto dorpc; 845 } 846 847 if (*vpp == NULLVP) { 848 /* namecache gave us a negative result */ 849 goto noentry; 850 } 851 852 /* 853 * investigate the vnode returned by cache_lookup_raw. 854 * if it isn't appropriate, do an rpc. 855 */ 856 newvp = *vpp; 857 if ((flags & ISDOTDOT) != 0) { 858 VOP_UNLOCK(dvp); 859 } 860 error = vn_lock(newvp, LK_EXCLUSIVE); 861 if ((flags & ISDOTDOT) != 0) { 862 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 863 } 864 if (error != 0) { 865 /* newvp has been reclaimed. */ 866 vrele(newvp); 867 *vpp = NULLVP; 868 goto dorpc; 869 } 870 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 871 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 872 nfsstats.lookupcache_hits++; 873 KASSERT(newvp->v_type != VNON); 874 return (0); 875 } 876 cache_purge1(newvp, NULL, PURGE_PARENTS); 877 vput(newvp); 878 *vpp = NULLVP; 879 } 880 dorpc: 881 #if 0 882 /* 883 * because nfsv3 has the same CREATE semantics as ours, 884 * we don't have to perform LOOKUPs beforehand. 885 * 886 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 887 * XXX although we have no way to know if O_EXCL is requested or not. 888 */ 889 890 if (v3 && cnp->cn_nameiop == CREATE && 891 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 892 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 893 return (EJUSTRETURN); 894 } 895 #endif /* 0 */ 896 897 error = 0; 898 newvp = NULLVP; 899 nfsstats.lookupcache_misses++; 900 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 901 len = cnp->cn_namelen; 902 nfsm_reqhead(np, NFSPROC_LOOKUP, 903 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 904 nfsm_fhtom(np, v3); 905 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 906 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 907 if (error) { 908 nfsm_postop_attr(dvp, attrflag, 0); 909 m_freem(mrep); 910 goto nfsmout; 911 } 912 nfsm_getfh(fhp, fhsize, v3); 913 914 /* 915 * Handle RENAME case... 916 */ 917 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 918 if (NFS_CMPFH(np, fhp, fhsize)) { 919 m_freem(mrep); 920 return (EISDIR); 921 } 922 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 923 if (error) { 924 m_freem(mrep); 925 return error; 926 } 927 newvp = NFSTOV(np); 928 #ifndef NFS_V2_ONLY 929 if (v3) { 930 nfsm_postop_attr(newvp, attrflag, 0); 931 nfsm_postop_attr(dvp, attrflag, 0); 932 } else 933 #endif 934 nfsm_loadattr(newvp, (struct vattr *)0, 0); 935 *vpp = newvp; 936 m_freem(mrep); 937 goto validate; 938 } 939 940 /* 941 * The postop attr handling is duplicated for each if case, 942 * because it should be done while dvp is locked (unlocking 943 * dvp is different for each case). 944 */ 945 946 if (NFS_CMPFH(np, fhp, fhsize)) { 947 /* 948 * as we handle "." lookup locally, this should be 949 * a broken server. 950 */ 951 vref(dvp); 952 newvp = dvp; 953 #ifndef NFS_V2_ONLY 954 if (v3) { 955 nfsm_postop_attr(newvp, attrflag, 0); 956 nfsm_postop_attr(dvp, attrflag, 0); 957 } else 958 #endif 959 nfsm_loadattr(newvp, (struct vattr *)0, 0); 960 } else if (flags & ISDOTDOT) { 961 /* 962 * ".." lookup 963 */ 964 VOP_UNLOCK(dvp); 965 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 966 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 967 if (error) { 968 m_freem(mrep); 969 return error; 970 } 971 newvp = NFSTOV(np); 972 973 #ifndef NFS_V2_ONLY 974 if (v3) { 975 nfsm_postop_attr(newvp, attrflag, 0); 976 nfsm_postop_attr(dvp, attrflag, 0); 977 } else 978 #endif 979 nfsm_loadattr(newvp, (struct vattr *)0, 0); 980 } else { 981 /* 982 * Other lookups. 983 */ 984 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 985 if (error) { 986 m_freem(mrep); 987 return error; 988 } 989 newvp = NFSTOV(np); 990 #ifndef NFS_V2_ONLY 991 if (v3) { 992 nfsm_postop_attr(newvp, attrflag, 0); 993 nfsm_postop_attr(dvp, attrflag, 0); 994 } else 995 #endif 996 nfsm_loadattr(newvp, (struct vattr *)0, 0); 997 } 998 if (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) { 999 nfs_cache_enter(dvp, newvp, cnp); 1000 } 1001 *vpp = newvp; 1002 nfsm_reqdone; 1003 if (error) { 1004 /* 1005 * We get here only because of errors returned by 1006 * the RPC. Otherwise we'll have returned above 1007 * (the nfsm_* macros will jump to nfsm_reqdone 1008 * on error). 1009 */ 1010 if (error == ENOENT && cnp->cn_nameiop != CREATE) { 1011 nfs_cache_enter(dvp, NULL, cnp); 1012 } 1013 if (newvp != NULLVP) { 1014 if (newvp == dvp) { 1015 vrele(newvp); 1016 } else { 1017 vput(newvp); 1018 } 1019 } 1020 noentry: 1021 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1022 (flags & ISLASTCN) && error == ENOENT) { 1023 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1024 error = EROFS; 1025 } else { 1026 error = EJUSTRETURN; 1027 } 1028 } 1029 *vpp = NULL; 1030 return error; 1031 } 1032 1033 validate: 1034 /* 1035 * make sure we have valid type and size. 1036 */ 1037 1038 newvp = *vpp; 1039 if (newvp->v_type == VNON) { 1040 struct vattr vattr; /* dummy */ 1041 1042 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1043 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1044 if (error) { 1045 vput(newvp); 1046 *vpp = NULL; 1047 } 1048 } 1049 1050 return error; 1051 } 1052 1053 /* 1054 * nfs read call. 1055 * Just call nfs_bioread() to do the work. 1056 */ 1057 int 1058 nfs_read(void *v) 1059 { 1060 struct vop_read_args /* { 1061 struct vnode *a_vp; 1062 struct uio *a_uio; 1063 int a_ioflag; 1064 kauth_cred_t a_cred; 1065 } */ *ap = v; 1066 struct vnode *vp = ap->a_vp; 1067 1068 if (vp->v_type != VREG) 1069 return EISDIR; 1070 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1071 } 1072 1073 /* 1074 * nfs readlink call 1075 */ 1076 int 1077 nfs_readlink(void *v) 1078 { 1079 struct vop_readlink_args /* { 1080 struct vnode *a_vp; 1081 struct uio *a_uio; 1082 kauth_cred_t a_cred; 1083 } */ *ap = v; 1084 struct vnode *vp = ap->a_vp; 1085 struct nfsnode *np = VTONFS(vp); 1086 1087 if (vp->v_type != VLNK) 1088 return (EPERM); 1089 1090 if (np->n_rcred != NULL) { 1091 kauth_cred_free(np->n_rcred); 1092 } 1093 np->n_rcred = ap->a_cred; 1094 kauth_cred_hold(np->n_rcred); 1095 1096 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1097 } 1098 1099 /* 1100 * Do a readlink rpc. 1101 * Called by nfs_doio() from below the buffer cache. 1102 */ 1103 int 1104 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1105 { 1106 u_int32_t *tl; 1107 char *cp; 1108 int32_t t1, t2; 1109 char *bpos, *dpos, *cp2; 1110 int error = 0; 1111 uint32_t len; 1112 struct mbuf *mreq, *mrep, *md, *mb; 1113 const int v3 = NFS_ISV3(vp); 1114 struct nfsnode *np = VTONFS(vp); 1115 #ifndef NFS_V2_ONLY 1116 int attrflag; 1117 #endif 1118 1119 nfsstats.rpccnt[NFSPROC_READLINK]++; 1120 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1121 nfsm_fhtom(np, v3); 1122 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1123 #ifndef NFS_V2_ONLY 1124 if (v3) 1125 nfsm_postop_attr(vp, attrflag, 0); 1126 #endif 1127 if (!error) { 1128 #ifndef NFS_V2_ONLY 1129 if (v3) { 1130 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1131 len = fxdr_unsigned(uint32_t, *tl); 1132 if (len > NFS_MAXPATHLEN) { 1133 /* 1134 * this pathname is too long for us. 1135 */ 1136 m_freem(mrep); 1137 /* Solaris returns EINVAL. should we follow? */ 1138 error = ENAMETOOLONG; 1139 goto nfsmout; 1140 } 1141 } else 1142 #endif 1143 { 1144 nfsm_strsiz(len, NFS_MAXPATHLEN); 1145 } 1146 nfsm_mtouio(uiop, len); 1147 } 1148 nfsm_reqdone; 1149 return (error); 1150 } 1151 1152 /* 1153 * nfs read rpc call 1154 * Ditto above 1155 */ 1156 int 1157 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1158 { 1159 u_int32_t *tl; 1160 char *cp; 1161 int32_t t1, t2; 1162 char *bpos, *dpos, *cp2; 1163 struct mbuf *mreq, *mrep, *md, *mb; 1164 struct nfsmount *nmp; 1165 int error = 0, len, retlen, tsiz, eof, byte_count; 1166 const int v3 = NFS_ISV3(vp); 1167 struct nfsnode *np = VTONFS(vp); 1168 #ifndef NFS_V2_ONLY 1169 int attrflag; 1170 #endif 1171 1172 #ifndef nolint 1173 eof = 0; 1174 #endif 1175 nmp = VFSTONFS(vp->v_mount); 1176 tsiz = uiop->uio_resid; 1177 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1178 return (EFBIG); 1179 iostat_busy(nmp->nm_stats); 1180 byte_count = 0; /* count bytes actually transferred */ 1181 while (tsiz > 0) { 1182 nfsstats.rpccnt[NFSPROC_READ]++; 1183 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1184 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1185 nfsm_fhtom(np, v3); 1186 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1187 #ifndef NFS_V2_ONLY 1188 if (v3) { 1189 txdr_hyper(uiop->uio_offset, tl); 1190 *(tl + 2) = txdr_unsigned(len); 1191 } else 1192 #endif 1193 { 1194 *tl++ = txdr_unsigned(uiop->uio_offset); 1195 *tl++ = txdr_unsigned(len); 1196 *tl = 0; 1197 } 1198 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1199 #ifndef NFS_V2_ONLY 1200 if (v3) { 1201 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1202 if (error) { 1203 m_freem(mrep); 1204 goto nfsmout; 1205 } 1206 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1207 eof = fxdr_unsigned(int, *(tl + 1)); 1208 } else 1209 #endif 1210 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1211 nfsm_strsiz(retlen, nmp->nm_rsize); 1212 nfsm_mtouio(uiop, retlen); 1213 m_freem(mrep); 1214 tsiz -= retlen; 1215 byte_count += retlen; 1216 #ifndef NFS_V2_ONLY 1217 if (v3) { 1218 if (eof || retlen == 0) 1219 tsiz = 0; 1220 } else 1221 #endif 1222 if (retlen < len) 1223 tsiz = 0; 1224 } 1225 nfsmout: 1226 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1227 return (error); 1228 } 1229 1230 struct nfs_writerpc_context { 1231 kmutex_t nwc_lock; 1232 kcondvar_t nwc_cv; 1233 int nwc_mbufcount; 1234 }; 1235 1236 /* 1237 * free mbuf used to refer protected pages while write rpc call. 1238 * called at splvm. 1239 */ 1240 static void 1241 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1242 { 1243 struct nfs_writerpc_context *ctx = arg; 1244 1245 KASSERT(m != NULL); 1246 KASSERT(ctx != NULL); 1247 pool_cache_put(mb_cache, m); 1248 mutex_enter(&ctx->nwc_lock); 1249 if (--ctx->nwc_mbufcount == 0) { 1250 cv_signal(&ctx->nwc_cv); 1251 } 1252 mutex_exit(&ctx->nwc_lock); 1253 } 1254 1255 /* 1256 * nfs write call 1257 */ 1258 int 1259 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1260 { 1261 u_int32_t *tl; 1262 char *cp; 1263 int32_t t1, t2; 1264 char *bpos, *dpos; 1265 struct mbuf *mreq, *mrep, *md, *mb; 1266 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1267 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1268 const int v3 = NFS_ISV3(vp); 1269 int committed = NFSV3WRITE_FILESYNC; 1270 struct nfsnode *np = VTONFS(vp); 1271 struct nfs_writerpc_context ctx; 1272 int byte_count; 1273 size_t origresid; 1274 #ifndef NFS_V2_ONLY 1275 char *cp2; 1276 int rlen, commit; 1277 #endif 1278 1279 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1280 panic("writerpc readonly vp %p", vp); 1281 } 1282 1283 #ifdef DIAGNOSTIC 1284 if (uiop->uio_iovcnt != 1) 1285 panic("nfs: writerpc iovcnt > 1"); 1286 #endif 1287 tsiz = uiop->uio_resid; 1288 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1289 return EFBIG; 1290 1291 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1292 cv_init(&ctx.nwc_cv, "nfsmblk"); 1293 ctx.nwc_mbufcount = 1; 1294 1295 retry: 1296 origresid = uiop->uio_resid; 1297 KASSERT(origresid == uiop->uio_iov->iov_len); 1298 iostat_busy(nmp->nm_stats); 1299 byte_count = 0; /* count of bytes actually written */ 1300 while (tsiz > 0) { 1301 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1302 uint32_t backup; 1303 bool stalewriteverf = false; 1304 1305 nfsstats.rpccnt[NFSPROC_WRITE]++; 1306 len = min(tsiz, nmp->nm_wsize); 1307 datalen = pageprotected ? 0 : nfsm_rndup(len); 1308 nfsm_reqhead(np, NFSPROC_WRITE, 1309 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1310 nfsm_fhtom(np, v3); 1311 #ifndef NFS_V2_ONLY 1312 if (v3) { 1313 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1314 txdr_hyper(uiop->uio_offset, tl); 1315 tl += 2; 1316 *tl++ = txdr_unsigned(len); 1317 *tl++ = txdr_unsigned(*iomode); 1318 *tl = txdr_unsigned(len); 1319 } else 1320 #endif 1321 { 1322 u_int32_t x; 1323 1324 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1325 /* Set both "begin" and "current" to non-garbage. */ 1326 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1327 *tl++ = x; /* "begin offset" */ 1328 *tl++ = x; /* "current offset" */ 1329 x = txdr_unsigned(len); 1330 *tl++ = x; /* total to this offset */ 1331 *tl = x; /* size of this write */ 1332 1333 } 1334 if (pageprotected) { 1335 /* 1336 * since we know pages can't be modified during i/o, 1337 * no need to copy them for us. 1338 */ 1339 struct mbuf *m; 1340 struct iovec *iovp = uiop->uio_iov; 1341 1342 m = m_get(M_WAIT, MT_DATA); 1343 MCLAIM(m, &nfs_mowner); 1344 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1345 nfs_writerpc_extfree, &ctx); 1346 m->m_flags |= M_EXT_ROMAP; 1347 m->m_len = len; 1348 mb->m_next = m; 1349 /* 1350 * no need to maintain mb and bpos here 1351 * because no one care them later. 1352 */ 1353 #if 0 1354 mb = m; 1355 bpos = mtod(void *, mb) + mb->m_len; 1356 #endif 1357 UIO_ADVANCE(uiop, len); 1358 uiop->uio_offset += len; 1359 mutex_enter(&ctx.nwc_lock); 1360 ctx.nwc_mbufcount++; 1361 mutex_exit(&ctx.nwc_lock); 1362 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1363 } else { 1364 nfsm_uiotom(uiop, len); 1365 } 1366 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1367 #ifndef NFS_V2_ONLY 1368 if (v3) { 1369 wccflag = NFSV3_WCCCHK; 1370 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1371 if (!error) { 1372 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1373 + NFSX_V3WRITEVERF); 1374 rlen = fxdr_unsigned(int, *tl++); 1375 if (rlen == 0) { 1376 error = NFSERR_IO; 1377 m_freem(mrep); 1378 break; 1379 } else if (rlen < len) { 1380 backup = len - rlen; 1381 UIO_ADVANCE(uiop, -backup); 1382 uiop->uio_offset -= backup; 1383 len = rlen; 1384 } 1385 commit = fxdr_unsigned(int, *tl++); 1386 1387 /* 1388 * Return the lowest committment level 1389 * obtained by any of the RPCs. 1390 */ 1391 if (committed == NFSV3WRITE_FILESYNC) 1392 committed = commit; 1393 else if (committed == NFSV3WRITE_DATASYNC && 1394 commit == NFSV3WRITE_UNSTABLE) 1395 committed = commit; 1396 mutex_enter(&nmp->nm_lock); 1397 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1398 memcpy(nmp->nm_writeverf, tl, 1399 NFSX_V3WRITEVERF); 1400 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1401 } else if ((nmp->nm_iflag & 1402 NFSMNT_STALEWRITEVERF) || 1403 memcmp(tl, nmp->nm_writeverf, 1404 NFSX_V3WRITEVERF)) { 1405 memcpy(nmp->nm_writeverf, tl, 1406 NFSX_V3WRITEVERF); 1407 /* 1408 * note NFSMNT_STALEWRITEVERF 1409 * if we're the first thread to 1410 * notice it. 1411 */ 1412 if ((nmp->nm_iflag & 1413 NFSMNT_STALEWRITEVERF) == 0) { 1414 stalewriteverf = true; 1415 nmp->nm_iflag |= 1416 NFSMNT_STALEWRITEVERF; 1417 } 1418 } 1419 mutex_exit(&nmp->nm_lock); 1420 } 1421 } else 1422 #endif 1423 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1424 if (wccflag) 1425 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1426 m_freem(mrep); 1427 if (error) 1428 break; 1429 tsiz -= len; 1430 byte_count += len; 1431 if (stalewriteverf) { 1432 *stalewriteverfp = true; 1433 stalewriteverf = false; 1434 if (committed == NFSV3WRITE_UNSTABLE && 1435 len != origresid) { 1436 /* 1437 * if our write requests weren't atomic but 1438 * unstable, datas in previous iterations 1439 * might have already been lost now. 1440 * then, we should resend them to nfsd. 1441 */ 1442 backup = origresid - tsiz; 1443 UIO_ADVANCE(uiop, -backup); 1444 uiop->uio_offset -= backup; 1445 tsiz = origresid; 1446 goto retry; 1447 } 1448 } 1449 } 1450 nfsmout: 1451 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1452 if (pageprotected) { 1453 /* 1454 * wait until mbufs go away. 1455 * retransmitted mbufs can survive longer than rpc requests 1456 * themselves. 1457 */ 1458 mutex_enter(&ctx.nwc_lock); 1459 ctx.nwc_mbufcount--; 1460 while (ctx.nwc_mbufcount > 0) { 1461 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1462 } 1463 mutex_exit(&ctx.nwc_lock); 1464 } 1465 mutex_destroy(&ctx.nwc_lock); 1466 cv_destroy(&ctx.nwc_cv); 1467 *iomode = committed; 1468 if (error) 1469 uiop->uio_resid = tsiz; 1470 return (error); 1471 } 1472 1473 /* 1474 * nfs mknod rpc 1475 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1476 * mode set to specify the file type and the size field for rdev. 1477 */ 1478 int 1479 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1480 { 1481 struct nfsv2_sattr *sp; 1482 u_int32_t *tl; 1483 char *cp; 1484 int32_t t1, t2; 1485 struct vnode *newvp = (struct vnode *)0; 1486 struct nfsnode *dnp, *np; 1487 char *cp2; 1488 char *bpos, *dpos; 1489 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1490 struct mbuf *mreq, *mrep, *md, *mb; 1491 u_int32_t rdev; 1492 const int v3 = NFS_ISV3(dvp); 1493 1494 if (vap->va_type == VCHR || vap->va_type == VBLK) 1495 rdev = txdr_unsigned(vap->va_rdev); 1496 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1497 rdev = nfs_xdrneg1; 1498 else { 1499 VOP_ABORTOP(dvp, cnp); 1500 vput(dvp); 1501 return (EOPNOTSUPP); 1502 } 1503 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1504 dnp = VTONFS(dvp); 1505 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1506 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1507 nfsm_fhtom(dnp, v3); 1508 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1509 #ifndef NFS_V2_ONLY 1510 if (v3) { 1511 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1512 *tl++ = vtonfsv3_type(vap->va_type); 1513 nfsm_v3attrbuild(vap, false); 1514 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1515 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1516 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1517 *tl = txdr_unsigned(minor(vap->va_rdev)); 1518 } 1519 } else 1520 #endif 1521 { 1522 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1523 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1524 sp->sa_uid = nfs_xdrneg1; 1525 sp->sa_gid = nfs_xdrneg1; 1526 sp->sa_size = rdev; 1527 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1528 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1529 } 1530 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1531 if (!error) { 1532 nfsm_mtofh(dvp, newvp, v3, gotvp); 1533 if (!gotvp) { 1534 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1535 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1536 if (!error) 1537 newvp = NFSTOV(np); 1538 } 1539 } 1540 #ifndef NFS_V2_ONLY 1541 if (v3) 1542 nfsm_wcc_data(dvp, wccflag, 0, !error); 1543 #endif 1544 nfsm_reqdone; 1545 if (error) { 1546 if (newvp) 1547 vput(newvp); 1548 } else { 1549 nfs_cache_enter(dvp, newvp, cnp); 1550 *vpp = newvp; 1551 } 1552 VTONFS(dvp)->n_flag |= NMODIFIED; 1553 if (!wccflag) 1554 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1555 vput(dvp); 1556 return (error); 1557 } 1558 1559 /* 1560 * nfs mknod vop 1561 * just call nfs_mknodrpc() to do the work. 1562 */ 1563 /* ARGSUSED */ 1564 int 1565 nfs_mknod(void *v) 1566 { 1567 struct vop_mknod_args /* { 1568 struct vnode *a_dvp; 1569 struct vnode **a_vpp; 1570 struct componentname *a_cnp; 1571 struct vattr *a_vap; 1572 } */ *ap = v; 1573 struct vnode *dvp = ap->a_dvp; 1574 struct componentname *cnp = ap->a_cnp; 1575 int error; 1576 1577 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1578 VN_KNOTE(dvp, NOTE_WRITE); 1579 if (error == 0 || error == EEXIST) 1580 cache_purge1(dvp, cnp, 0); 1581 return (error); 1582 } 1583 1584 /* 1585 * nfs file create call 1586 */ 1587 int 1588 nfs_create(void *v) 1589 { 1590 struct vop_create_args /* { 1591 struct vnode *a_dvp; 1592 struct vnode **a_vpp; 1593 struct componentname *a_cnp; 1594 struct vattr *a_vap; 1595 } */ *ap = v; 1596 struct vnode *dvp = ap->a_dvp; 1597 struct vattr *vap = ap->a_vap; 1598 struct componentname *cnp = ap->a_cnp; 1599 struct nfsv2_sattr *sp; 1600 u_int32_t *tl; 1601 char *cp; 1602 int32_t t1, t2; 1603 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1604 struct vnode *newvp = (struct vnode *)0; 1605 char *bpos, *dpos, *cp2; 1606 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1607 struct mbuf *mreq, *mrep, *md, *mb; 1608 const int v3 = NFS_ISV3(dvp); 1609 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1610 1611 /* 1612 * Oops, not for me.. 1613 */ 1614 if (vap->va_type == VSOCK) 1615 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1616 1617 KASSERT(vap->va_type == VREG); 1618 1619 #ifdef VA_EXCLUSIVE 1620 if (vap->va_vaflags & VA_EXCLUSIVE) { 1621 excl_mode = NFSV3CREATE_EXCLUSIVE; 1622 } 1623 #endif 1624 again: 1625 error = 0; 1626 nfsstats.rpccnt[NFSPROC_CREATE]++; 1627 dnp = VTONFS(dvp); 1628 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1629 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1630 nfsm_fhtom(dnp, v3); 1631 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1632 #ifndef NFS_V2_ONLY 1633 if (v3) { 1634 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1635 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1636 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1637 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1638 *tl++ = cprng_fast32(); 1639 *tl = cprng_fast32(); 1640 } else { 1641 *tl = txdr_unsigned(excl_mode); 1642 nfsm_v3attrbuild(vap, false); 1643 } 1644 } else 1645 #endif 1646 { 1647 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1648 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1649 sp->sa_uid = nfs_xdrneg1; 1650 sp->sa_gid = nfs_xdrneg1; 1651 sp->sa_size = 0; 1652 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1653 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1654 } 1655 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1656 if (!error) { 1657 nfsm_mtofh(dvp, newvp, v3, gotvp); 1658 if (!gotvp) { 1659 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1660 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1661 if (!error) 1662 newvp = NFSTOV(np); 1663 } 1664 } 1665 #ifndef NFS_V2_ONLY 1666 if (v3) 1667 nfsm_wcc_data(dvp, wccflag, 0, !error); 1668 #endif 1669 nfsm_reqdone; 1670 if (error) { 1671 /* 1672 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1673 */ 1674 if (v3 && error == ENOTSUP) { 1675 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1676 excl_mode = NFSV3CREATE_GUARDED; 1677 goto again; 1678 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1679 excl_mode = NFSV3CREATE_UNCHECKED; 1680 goto again; 1681 } 1682 } 1683 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1684 struct timespec ts; 1685 1686 getnanotime(&ts); 1687 1688 /* 1689 * make sure that we'll update timestamps as 1690 * most server implementations use them to store 1691 * the create verifier. 1692 * 1693 * XXX it's better to use TOSERVER always. 1694 */ 1695 1696 if (vap->va_atime.tv_sec == VNOVAL) 1697 vap->va_atime = ts; 1698 if (vap->va_mtime.tv_sec == VNOVAL) 1699 vap->va_mtime = ts; 1700 1701 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1702 } 1703 if (error == 0) { 1704 if (cnp->cn_flags & MAKEENTRY) 1705 nfs_cache_enter(dvp, newvp, cnp); 1706 else 1707 cache_purge1(dvp, cnp, 0); 1708 *ap->a_vpp = newvp; 1709 } else { 1710 if (newvp) 1711 vput(newvp); 1712 if (error == EEXIST) 1713 cache_purge1(dvp, cnp, 0); 1714 } 1715 VTONFS(dvp)->n_flag |= NMODIFIED; 1716 if (!wccflag) 1717 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1718 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1719 vput(dvp); 1720 return (error); 1721 } 1722 1723 /* 1724 * nfs file remove call 1725 * To try and make nfs semantics closer to ufs semantics, a file that has 1726 * other processes using the vnode is renamed instead of removed and then 1727 * removed later on the last close. 1728 * - If v_usecount > 1 1729 * If a rename is not already in the works 1730 * call nfs_sillyrename() to set it up 1731 * else 1732 * do the remove rpc 1733 */ 1734 int 1735 nfs_remove(void *v) 1736 { 1737 struct vop_remove_args /* { 1738 struct vnodeop_desc *a_desc; 1739 struct vnode * a_dvp; 1740 struct vnode * a_vp; 1741 struct componentname * a_cnp; 1742 } */ *ap = v; 1743 struct vnode *vp = ap->a_vp; 1744 struct vnode *dvp = ap->a_dvp; 1745 struct componentname *cnp = ap->a_cnp; 1746 struct nfsnode *np = VTONFS(vp); 1747 int error = 0; 1748 struct vattr vattr; 1749 1750 #ifndef DIAGNOSTIC 1751 if (vp->v_usecount < 1) 1752 panic("nfs_remove: bad v_usecount"); 1753 #endif 1754 if (vp->v_type == VDIR) 1755 error = EPERM; 1756 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1757 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1758 vattr.va_nlink > 1)) { 1759 /* 1760 * Purge the name cache so that the chance of a lookup for 1761 * the name succeeding while the remove is in progress is 1762 * minimized. Without node locking it can still happen, such 1763 * that an I/O op returns ESTALE, but since you get this if 1764 * another host removes the file.. 1765 */ 1766 cache_purge(vp); 1767 /* 1768 * throw away biocache buffers, mainly to avoid 1769 * unnecessary delayed writes later. 1770 */ 1771 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1772 /* Do the rpc */ 1773 if (error != EINTR) 1774 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1775 cnp->cn_namelen, cnp->cn_cred, curlwp); 1776 } else if (!np->n_sillyrename) 1777 error = nfs_sillyrename(dvp, vp, cnp, false); 1778 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1779 vattr.va_nlink == 1) { 1780 np->n_flag |= NREMOVED; 1781 } 1782 NFS_INVALIDATE_ATTRCACHE(np); 1783 VN_KNOTE(vp, NOTE_DELETE); 1784 VN_KNOTE(dvp, NOTE_WRITE); 1785 if (dvp == vp) 1786 vrele(vp); 1787 else 1788 vput(vp); 1789 vput(dvp); 1790 return (error); 1791 } 1792 1793 /* 1794 * nfs file remove rpc called from nfs_inactive 1795 */ 1796 int 1797 nfs_removeit(struct sillyrename *sp) 1798 { 1799 1800 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1801 (struct lwp *)0)); 1802 } 1803 1804 /* 1805 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1806 */ 1807 int 1808 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1809 { 1810 u_int32_t *tl; 1811 char *cp; 1812 #ifndef NFS_V2_ONLY 1813 int32_t t1; 1814 char *cp2; 1815 #endif 1816 int32_t t2; 1817 char *bpos, *dpos; 1818 int error = 0, wccflag = NFSV3_WCCRATTR; 1819 struct mbuf *mreq, *mrep, *md, *mb; 1820 const int v3 = NFS_ISV3(dvp); 1821 int rexmit = 0; 1822 struct nfsnode *dnp = VTONFS(dvp); 1823 1824 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1825 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1826 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1827 nfsm_fhtom(dnp, v3); 1828 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1829 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1830 #ifndef NFS_V2_ONLY 1831 if (v3) 1832 nfsm_wcc_data(dvp, wccflag, 0, !error); 1833 #endif 1834 nfsm_reqdone; 1835 VTONFS(dvp)->n_flag |= NMODIFIED; 1836 if (!wccflag) 1837 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1838 /* 1839 * Kludge City: If the first reply to the remove rpc is lost.. 1840 * the reply to the retransmitted request will be ENOENT 1841 * since the file was in fact removed 1842 * Therefore, we cheat and return success. 1843 */ 1844 if (rexmit && error == ENOENT) 1845 error = 0; 1846 return (error); 1847 } 1848 1849 /* 1850 * nfs file rename call 1851 */ 1852 int 1853 nfs_rename(void *v) 1854 { 1855 struct vop_rename_args /* { 1856 struct vnode *a_fdvp; 1857 struct vnode *a_fvp; 1858 struct componentname *a_fcnp; 1859 struct vnode *a_tdvp; 1860 struct vnode *a_tvp; 1861 struct componentname *a_tcnp; 1862 } */ *ap = v; 1863 struct vnode *fvp = ap->a_fvp; 1864 struct vnode *tvp = ap->a_tvp; 1865 struct vnode *fdvp = ap->a_fdvp; 1866 struct vnode *tdvp = ap->a_tdvp; 1867 struct componentname *tcnp = ap->a_tcnp; 1868 struct componentname *fcnp = ap->a_fcnp; 1869 int error; 1870 1871 /* Check for cross-device rename */ 1872 if ((fvp->v_mount != tdvp->v_mount) || 1873 (tvp && (fvp->v_mount != tvp->v_mount))) { 1874 error = EXDEV; 1875 goto out; 1876 } 1877 1878 /* 1879 * If the tvp exists and is in use, sillyrename it before doing the 1880 * rename of the new file over it. 1881 * 1882 * Have sillyrename use link instead of rename if possible, 1883 * so that we don't lose the file if the rename fails, and so 1884 * that there's no window when the "to" file doesn't exist. 1885 */ 1886 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1887 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1888 VN_KNOTE(tvp, NOTE_DELETE); 1889 vput(tvp); 1890 tvp = NULL; 1891 } 1892 1893 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1894 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1895 curlwp); 1896 1897 VN_KNOTE(fdvp, NOTE_WRITE); 1898 VN_KNOTE(tdvp, NOTE_WRITE); 1899 if (error == 0 || error == EEXIST) { 1900 if (fvp->v_type == VDIR) 1901 cache_purge(fvp); 1902 else 1903 cache_purge1(fdvp, fcnp, 0); 1904 if (tvp != NULL && tvp->v_type == VDIR) 1905 cache_purge(tvp); 1906 else 1907 cache_purge1(tdvp, tcnp, 0); 1908 } 1909 out: 1910 if (tdvp == tvp) 1911 vrele(tdvp); 1912 else 1913 vput(tdvp); 1914 if (tvp) 1915 vput(tvp); 1916 vrele(fdvp); 1917 vrele(fvp); 1918 return (error); 1919 } 1920 1921 /* 1922 * nfs file rename rpc called from nfs_remove() above 1923 */ 1924 int 1925 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1926 { 1927 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1928 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1929 } 1930 1931 /* 1932 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1933 */ 1934 int 1935 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, struct vnode *tdvp, const char *tnameptr, int tnamelen, kauth_cred_t cred, struct lwp *l) 1936 { 1937 u_int32_t *tl; 1938 char *cp; 1939 #ifndef NFS_V2_ONLY 1940 int32_t t1; 1941 char *cp2; 1942 #endif 1943 int32_t t2; 1944 char *bpos, *dpos; 1945 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1946 struct mbuf *mreq, *mrep, *md, *mb; 1947 const int v3 = NFS_ISV3(fdvp); 1948 int rexmit = 0; 1949 struct nfsnode *fdnp = VTONFS(fdvp); 1950 1951 nfsstats.rpccnt[NFSPROC_RENAME]++; 1952 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1953 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1954 nfsm_rndup(tnamelen)); 1955 nfsm_fhtom(fdnp, v3); 1956 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1957 nfsm_fhtom(VTONFS(tdvp), v3); 1958 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1959 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1960 #ifndef NFS_V2_ONLY 1961 if (v3) { 1962 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1963 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1964 } 1965 #endif 1966 nfsm_reqdone; 1967 VTONFS(fdvp)->n_flag |= NMODIFIED; 1968 VTONFS(tdvp)->n_flag |= NMODIFIED; 1969 if (!fwccflag) 1970 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1971 if (!twccflag) 1972 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1973 /* 1974 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1975 */ 1976 if (rexmit && error == ENOENT) 1977 error = 0; 1978 return (error); 1979 } 1980 1981 /* 1982 * NFS link RPC, called from nfs_link. 1983 * Assumes dvp and vp locked, and leaves them that way. 1984 */ 1985 1986 static int 1987 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 1988 size_t namelen, kauth_cred_t cred, struct lwp *l) 1989 { 1990 u_int32_t *tl; 1991 char *cp; 1992 #ifndef NFS_V2_ONLY 1993 int32_t t1; 1994 char *cp2; 1995 #endif 1996 int32_t t2; 1997 char *bpos, *dpos; 1998 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1999 struct mbuf *mreq, *mrep, *md, *mb; 2000 const int v3 = NFS_ISV3(dvp); 2001 int rexmit = 0; 2002 struct nfsnode *np = VTONFS(vp); 2003 2004 nfsstats.rpccnt[NFSPROC_LINK]++; 2005 nfsm_reqhead(np, NFSPROC_LINK, 2006 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2007 nfsm_fhtom(np, v3); 2008 nfsm_fhtom(VTONFS(dvp), v3); 2009 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2010 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2011 #ifndef NFS_V2_ONLY 2012 if (v3) { 2013 nfsm_postop_attr(vp, attrflag, 0); 2014 nfsm_wcc_data(dvp, wccflag, 0, !error); 2015 } 2016 #endif 2017 nfsm_reqdone; 2018 2019 VTONFS(dvp)->n_flag |= NMODIFIED; 2020 if (!attrflag) 2021 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2022 if (!wccflag) 2023 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2024 2025 /* 2026 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2027 */ 2028 if (rexmit && error == EEXIST) 2029 error = 0; 2030 2031 return error; 2032 } 2033 2034 /* 2035 * nfs hard link create call 2036 */ 2037 int 2038 nfs_link(void *v) 2039 { 2040 struct vop_link_args /* { 2041 struct vnode *a_dvp; 2042 struct vnode *a_vp; 2043 struct componentname *a_cnp; 2044 } */ *ap = v; 2045 struct vnode *vp = ap->a_vp; 2046 struct vnode *dvp = ap->a_dvp; 2047 struct componentname *cnp = ap->a_cnp; 2048 int error = 0; 2049 2050 error = vn_lock(vp, LK_EXCLUSIVE); 2051 if (error != 0) { 2052 VOP_ABORTOP(dvp, cnp); 2053 vput(dvp); 2054 return error; 2055 } 2056 2057 /* 2058 * Push all writes to the server, so that the attribute cache 2059 * doesn't get "out of sync" with the server. 2060 * XXX There should be a better way! 2061 */ 2062 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2063 2064 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2065 cnp->cn_cred, curlwp); 2066 2067 if (error == 0) { 2068 cache_purge1(dvp, cnp, 0); 2069 } 2070 VOP_UNLOCK(vp); 2071 VN_KNOTE(vp, NOTE_LINK); 2072 VN_KNOTE(dvp, NOTE_WRITE); 2073 vput(dvp); 2074 return (error); 2075 } 2076 2077 /* 2078 * nfs symbolic link create call 2079 */ 2080 int 2081 nfs_symlink(void *v) 2082 { 2083 struct vop_symlink_args /* { 2084 struct vnode *a_dvp; 2085 struct vnode **a_vpp; 2086 struct componentname *a_cnp; 2087 struct vattr *a_vap; 2088 char *a_target; 2089 } */ *ap = v; 2090 struct vnode *dvp = ap->a_dvp; 2091 struct vattr *vap = ap->a_vap; 2092 struct componentname *cnp = ap->a_cnp; 2093 struct nfsv2_sattr *sp; 2094 u_int32_t *tl; 2095 char *cp; 2096 int32_t t1, t2; 2097 char *bpos, *dpos, *cp2; 2098 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2099 struct mbuf *mreq, *mrep, *md, *mb; 2100 struct vnode *newvp = (struct vnode *)0; 2101 const int v3 = NFS_ISV3(dvp); 2102 int rexmit = 0; 2103 struct nfsnode *dnp = VTONFS(dvp); 2104 2105 *ap->a_vpp = NULL; 2106 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2107 slen = strlen(ap->a_target); 2108 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2109 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2110 nfsm_fhtom(dnp, v3); 2111 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2112 #ifndef NFS_V2_ONlY 2113 if (v3) 2114 nfsm_v3attrbuild(vap, false); 2115 #endif 2116 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2117 #ifndef NFS_V2_ONlY 2118 if (!v3) { 2119 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2120 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2121 sp->sa_uid = nfs_xdrneg1; 2122 sp->sa_gid = nfs_xdrneg1; 2123 sp->sa_size = nfs_xdrneg1; 2124 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2125 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2126 } 2127 #endif 2128 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2129 &rexmit); 2130 #ifndef NFS_V2_ONlY 2131 if (v3) { 2132 if (!error) 2133 nfsm_mtofh(dvp, newvp, v3, gotvp); 2134 nfsm_wcc_data(dvp, wccflag, 0, !error); 2135 } 2136 #endif 2137 nfsm_reqdone; 2138 /* 2139 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2140 */ 2141 if (rexmit && error == EEXIST) 2142 error = 0; 2143 if (error == 0 || error == EEXIST) 2144 cache_purge1(dvp, cnp, 0); 2145 if (error == 0 && newvp == NULL) { 2146 struct nfsnode *np = NULL; 2147 2148 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2149 cnp->cn_cred, curlwp, &np); 2150 if (error == 0) 2151 newvp = NFSTOV(np); 2152 } 2153 if (error) { 2154 if (newvp != NULL) 2155 vput(newvp); 2156 } else { 2157 *ap->a_vpp = newvp; 2158 } 2159 VTONFS(dvp)->n_flag |= NMODIFIED; 2160 if (!wccflag) 2161 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2162 VN_KNOTE(dvp, NOTE_WRITE); 2163 vput(dvp); 2164 return (error); 2165 } 2166 2167 /* 2168 * nfs make dir call 2169 */ 2170 int 2171 nfs_mkdir(void *v) 2172 { 2173 struct vop_mkdir_args /* { 2174 struct vnode *a_dvp; 2175 struct vnode **a_vpp; 2176 struct componentname *a_cnp; 2177 struct vattr *a_vap; 2178 } */ *ap = v; 2179 struct vnode *dvp = ap->a_dvp; 2180 struct vattr *vap = ap->a_vap; 2181 struct componentname *cnp = ap->a_cnp; 2182 struct nfsv2_sattr *sp; 2183 u_int32_t *tl; 2184 char *cp; 2185 int32_t t1, t2; 2186 int len; 2187 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2188 struct vnode *newvp = (struct vnode *)0; 2189 char *bpos, *dpos, *cp2; 2190 int error = 0, wccflag = NFSV3_WCCRATTR; 2191 int gotvp = 0; 2192 int rexmit = 0; 2193 struct mbuf *mreq, *mrep, *md, *mb; 2194 const int v3 = NFS_ISV3(dvp); 2195 2196 len = cnp->cn_namelen; 2197 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2198 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2199 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2200 nfsm_fhtom(dnp, v3); 2201 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2202 #ifndef NFS_V2_ONLY 2203 if (v3) { 2204 nfsm_v3attrbuild(vap, false); 2205 } else 2206 #endif 2207 { 2208 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2209 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2210 sp->sa_uid = nfs_xdrneg1; 2211 sp->sa_gid = nfs_xdrneg1; 2212 sp->sa_size = nfs_xdrneg1; 2213 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2214 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2215 } 2216 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2217 if (!error) 2218 nfsm_mtofh(dvp, newvp, v3, gotvp); 2219 if (v3) 2220 nfsm_wcc_data(dvp, wccflag, 0, !error); 2221 nfsm_reqdone; 2222 VTONFS(dvp)->n_flag |= NMODIFIED; 2223 if (!wccflag) 2224 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2225 /* 2226 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2227 * if we can succeed in looking up the directory. 2228 */ 2229 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2230 if (newvp) { 2231 vput(newvp); 2232 newvp = (struct vnode *)0; 2233 } 2234 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2235 curlwp, &np); 2236 if (!error) { 2237 newvp = NFSTOV(np); 2238 if (newvp->v_type != VDIR || newvp == dvp) 2239 error = EEXIST; 2240 } 2241 } 2242 if (error) { 2243 if (newvp) { 2244 if (dvp != newvp) 2245 vput(newvp); 2246 else 2247 vrele(newvp); 2248 } 2249 } else { 2250 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2251 nfs_cache_enter(dvp, newvp, cnp); 2252 *ap->a_vpp = newvp; 2253 } 2254 vput(dvp); 2255 return (error); 2256 } 2257 2258 /* 2259 * nfs remove directory call 2260 */ 2261 int 2262 nfs_rmdir(void *v) 2263 { 2264 struct vop_rmdir_args /* { 2265 struct vnode *a_dvp; 2266 struct vnode *a_vp; 2267 struct componentname *a_cnp; 2268 } */ *ap = v; 2269 struct vnode *vp = ap->a_vp; 2270 struct vnode *dvp = ap->a_dvp; 2271 struct componentname *cnp = ap->a_cnp; 2272 u_int32_t *tl; 2273 char *cp; 2274 #ifndef NFS_V2_ONLY 2275 int32_t t1; 2276 char *cp2; 2277 #endif 2278 int32_t t2; 2279 char *bpos, *dpos; 2280 int error = 0, wccflag = NFSV3_WCCRATTR; 2281 int rexmit = 0; 2282 struct mbuf *mreq, *mrep, *md, *mb; 2283 const int v3 = NFS_ISV3(dvp); 2284 struct nfsnode *dnp; 2285 2286 if (dvp == vp) { 2287 vrele(dvp); 2288 vput(dvp); 2289 return (EINVAL); 2290 } 2291 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2292 dnp = VTONFS(dvp); 2293 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2294 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2295 nfsm_fhtom(dnp, v3); 2296 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2297 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2298 #ifndef NFS_V2_ONLY 2299 if (v3) 2300 nfsm_wcc_data(dvp, wccflag, 0, !error); 2301 #endif 2302 nfsm_reqdone; 2303 VTONFS(dvp)->n_flag |= NMODIFIED; 2304 if (!wccflag) 2305 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2306 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2307 VN_KNOTE(vp, NOTE_DELETE); 2308 cache_purge(vp); 2309 vput(vp); 2310 vput(dvp); 2311 /* 2312 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2313 */ 2314 if (rexmit && error == ENOENT) 2315 error = 0; 2316 return (error); 2317 } 2318 2319 /* 2320 * nfs readdir call 2321 */ 2322 int 2323 nfs_readdir(void *v) 2324 { 2325 struct vop_readdir_args /* { 2326 struct vnode *a_vp; 2327 struct uio *a_uio; 2328 kauth_cred_t a_cred; 2329 int *a_eofflag; 2330 off_t **a_cookies; 2331 int *a_ncookies; 2332 } */ *ap = v; 2333 struct vnode *vp = ap->a_vp; 2334 struct uio *uio = ap->a_uio; 2335 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2336 char *base = uio->uio_iov->iov_base; 2337 int tresid, error; 2338 size_t count, lost; 2339 struct dirent *dp; 2340 off_t *cookies = NULL; 2341 int ncookies = 0, nc; 2342 2343 if (vp->v_type != VDIR) 2344 return (EPERM); 2345 2346 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2347 count = uio->uio_resid - lost; 2348 if (count <= 0) 2349 return (EINVAL); 2350 2351 /* 2352 * Call nfs_bioread() to do the real work. 2353 */ 2354 tresid = uio->uio_resid = count; 2355 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2356 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2357 2358 if (!error && ap->a_cookies) { 2359 ncookies = count / 16; 2360 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2361 *ap->a_cookies = cookies; 2362 } 2363 2364 if (!error && uio->uio_resid == tresid) { 2365 uio->uio_resid += lost; 2366 nfsstats.direofcache_misses++; 2367 if (ap->a_cookies) 2368 *ap->a_ncookies = 0; 2369 *ap->a_eofflag = 1; 2370 return (0); 2371 } 2372 2373 if (!error && ap->a_cookies) { 2374 /* 2375 * Only the NFS server and emulations use cookies, and they 2376 * load the directory block into system space, so we can 2377 * just look at it directly. 2378 */ 2379 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2380 uio->uio_iovcnt != 1) 2381 panic("nfs_readdir: lost in space"); 2382 for (nc = 0; ncookies-- && 2383 base < (char *)uio->uio_iov->iov_base; nc++){ 2384 dp = (struct dirent *) base; 2385 if (dp->d_reclen == 0) 2386 break; 2387 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2388 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2389 else 2390 *(cookies++) = NFS_GETCOOKIE(dp); 2391 base += dp->d_reclen; 2392 } 2393 uio->uio_resid += 2394 ((char *)uio->uio_iov->iov_base - base); 2395 uio->uio_iov->iov_len += 2396 ((char *)uio->uio_iov->iov_base - base); 2397 uio->uio_iov->iov_base = base; 2398 *ap->a_ncookies = nc; 2399 } 2400 2401 uio->uio_resid += lost; 2402 *ap->a_eofflag = 0; 2403 return (error); 2404 } 2405 2406 /* 2407 * Readdir rpc call. 2408 * Called from below the buffer cache by nfs_doio(). 2409 */ 2410 int 2411 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2412 { 2413 int len, left; 2414 struct dirent *dp = NULL; 2415 u_int32_t *tl; 2416 char *cp; 2417 int32_t t1, t2; 2418 char *bpos, *dpos, *cp2; 2419 struct mbuf *mreq, *mrep, *md, *mb; 2420 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2421 struct nfsnode *dnp = VTONFS(vp); 2422 u_quad_t fileno; 2423 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2424 #ifndef NFS_V2_ONLY 2425 int attrflag; 2426 #endif 2427 int nrpcs = 0, reclen; 2428 const int v3 = NFS_ISV3(vp); 2429 2430 #ifdef DIAGNOSTIC 2431 /* 2432 * Should be called from buffer cache, so only amount of 2433 * NFS_DIRBLKSIZ will be requested. 2434 */ 2435 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2436 panic("nfs readdirrpc bad uio"); 2437 #endif 2438 2439 /* 2440 * Loop around doing readdir rpc's of size nm_readdirsize 2441 * truncated to a multiple of NFS_DIRFRAGSIZ. 2442 * The stopping criteria is EOF or buffer full. 2443 */ 2444 while (more_dirs && bigenough) { 2445 /* 2446 * Heuristic: don't bother to do another RPC to further 2447 * fill up this block if there is not much room left. (< 50% 2448 * of the readdir RPC size). This wastes some buffer space 2449 * but can save up to 50% in RPC calls. 2450 */ 2451 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2452 bigenough = 0; 2453 break; 2454 } 2455 nfsstats.rpccnt[NFSPROC_READDIR]++; 2456 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2457 NFSX_READDIR(v3)); 2458 nfsm_fhtom(dnp, v3); 2459 #ifndef NFS_V2_ONLY 2460 if (v3) { 2461 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2462 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2463 txdr_swapcookie3(uiop->uio_offset, tl); 2464 } else { 2465 txdr_cookie3(uiop->uio_offset, tl); 2466 } 2467 tl += 2; 2468 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2469 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2470 } else 2471 #endif 2472 { 2473 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2474 *tl++ = txdr_unsigned(uiop->uio_offset); 2475 } 2476 *tl = txdr_unsigned(nmp->nm_readdirsize); 2477 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2478 nrpcs++; 2479 #ifndef NFS_V2_ONLY 2480 if (v3) { 2481 nfsm_postop_attr(vp, attrflag, 0); 2482 if (!error) { 2483 nfsm_dissect(tl, u_int32_t *, 2484 2 * NFSX_UNSIGNED); 2485 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2486 dnp->n_cookieverf.nfsuquad[1] = *tl; 2487 } else { 2488 m_freem(mrep); 2489 goto nfsmout; 2490 } 2491 } 2492 #endif 2493 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2494 more_dirs = fxdr_unsigned(int, *tl); 2495 2496 /* loop thru the dir entries, doctoring them to 4bsd form */ 2497 while (more_dirs && bigenough) { 2498 #ifndef NFS_V2_ONLY 2499 if (v3) { 2500 nfsm_dissect(tl, u_int32_t *, 2501 3 * NFSX_UNSIGNED); 2502 fileno = fxdr_hyper(tl); 2503 len = fxdr_unsigned(int, *(tl + 2)); 2504 } else 2505 #endif 2506 { 2507 nfsm_dissect(tl, u_int32_t *, 2508 2 * NFSX_UNSIGNED); 2509 fileno = fxdr_unsigned(u_quad_t, *tl++); 2510 len = fxdr_unsigned(int, *tl); 2511 } 2512 if (len <= 0 || len > NFS_MAXNAMLEN) { 2513 error = EBADRPC; 2514 m_freem(mrep); 2515 goto nfsmout; 2516 } 2517 /* for cookie stashing */ 2518 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2519 left = NFS_DIRFRAGSIZ - blksiz; 2520 if (reclen > left) { 2521 memset(uiop->uio_iov->iov_base, 0, left); 2522 dp->d_reclen += left; 2523 UIO_ADVANCE(uiop, left); 2524 blksiz = 0; 2525 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2526 } 2527 if (reclen > uiop->uio_resid) 2528 bigenough = 0; 2529 if (bigenough) { 2530 int tlen; 2531 2532 dp = (struct dirent *)uiop->uio_iov->iov_base; 2533 dp->d_fileno = fileno; 2534 dp->d_namlen = len; 2535 dp->d_reclen = reclen; 2536 dp->d_type = DT_UNKNOWN; 2537 blksiz += reclen; 2538 if (blksiz == NFS_DIRFRAGSIZ) 2539 blksiz = 0; 2540 UIO_ADVANCE(uiop, DIRHDSIZ); 2541 nfsm_mtouio(uiop, len); 2542 tlen = reclen - (DIRHDSIZ + len); 2543 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2544 UIO_ADVANCE(uiop, tlen); 2545 } else 2546 nfsm_adv(nfsm_rndup(len)); 2547 #ifndef NFS_V2_ONLY 2548 if (v3) { 2549 nfsm_dissect(tl, u_int32_t *, 2550 3 * NFSX_UNSIGNED); 2551 } else 2552 #endif 2553 { 2554 nfsm_dissect(tl, u_int32_t *, 2555 2 * NFSX_UNSIGNED); 2556 } 2557 if (bigenough) { 2558 #ifndef NFS_V2_ONLY 2559 if (v3) { 2560 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2561 uiop->uio_offset = 2562 fxdr_swapcookie3(tl); 2563 else 2564 uiop->uio_offset = 2565 fxdr_cookie3(tl); 2566 } 2567 else 2568 #endif 2569 { 2570 uiop->uio_offset = 2571 fxdr_unsigned(off_t, *tl); 2572 } 2573 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2574 } 2575 if (v3) 2576 tl += 2; 2577 else 2578 tl++; 2579 more_dirs = fxdr_unsigned(int, *tl); 2580 } 2581 /* 2582 * If at end of rpc data, get the eof boolean 2583 */ 2584 if (!more_dirs) { 2585 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2586 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2587 2588 /* 2589 * kludge: if we got no entries, treat it as EOF. 2590 * some server sometimes send a reply without any 2591 * entries or EOF. 2592 * although it might mean the server has very long name, 2593 * we can't handle such entries anyway. 2594 */ 2595 2596 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2597 more_dirs = 0; 2598 } 2599 m_freem(mrep); 2600 } 2601 /* 2602 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2603 * by increasing d_reclen for the last record. 2604 */ 2605 if (blksiz > 0) { 2606 left = NFS_DIRFRAGSIZ - blksiz; 2607 memset(uiop->uio_iov->iov_base, 0, left); 2608 dp->d_reclen += left; 2609 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2610 UIO_ADVANCE(uiop, left); 2611 } 2612 2613 /* 2614 * We are now either at the end of the directory or have filled the 2615 * block. 2616 */ 2617 if (bigenough) { 2618 dnp->n_direofoffset = uiop->uio_offset; 2619 dnp->n_flag |= NEOFVALID; 2620 } 2621 nfsmout: 2622 return (error); 2623 } 2624 2625 #ifndef NFS_V2_ONLY 2626 /* 2627 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2628 */ 2629 int 2630 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2631 { 2632 int len, left; 2633 struct dirent *dp = NULL; 2634 u_int32_t *tl; 2635 char *cp; 2636 int32_t t1, t2; 2637 struct vnode *newvp; 2638 char *bpos, *dpos, *cp2; 2639 struct mbuf *mreq, *mrep, *md, *mb; 2640 struct nameidata nami, *ndp = &nami; 2641 struct componentname *cnp = &ndp->ni_cnd; 2642 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2643 struct nfsnode *dnp = VTONFS(vp), *np; 2644 nfsfh_t *fhp; 2645 u_quad_t fileno; 2646 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2647 int attrflag, fhsize, nrpcs = 0, reclen; 2648 struct nfs_fattr fattr, *fp; 2649 2650 #ifdef DIAGNOSTIC 2651 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2652 panic("nfs readdirplusrpc bad uio"); 2653 #endif 2654 ndp->ni_dvp = vp; 2655 newvp = NULLVP; 2656 2657 /* 2658 * Loop around doing readdir rpc's of size nm_readdirsize 2659 * truncated to a multiple of NFS_DIRFRAGSIZ. 2660 * The stopping criteria is EOF or buffer full. 2661 */ 2662 while (more_dirs && bigenough) { 2663 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2664 bigenough = 0; 2665 break; 2666 } 2667 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2668 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2669 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2670 nfsm_fhtom(dnp, 1); 2671 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2672 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2673 txdr_swapcookie3(uiop->uio_offset, tl); 2674 } else { 2675 txdr_cookie3(uiop->uio_offset, tl); 2676 } 2677 tl += 2; 2678 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2679 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2680 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2681 *tl = txdr_unsigned(nmp->nm_rsize); 2682 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2683 nfsm_postop_attr(vp, attrflag, 0); 2684 if (error) { 2685 m_freem(mrep); 2686 goto nfsmout; 2687 } 2688 nrpcs++; 2689 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2690 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2691 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2692 more_dirs = fxdr_unsigned(int, *tl); 2693 2694 /* loop thru the dir entries, doctoring them to 4bsd form */ 2695 while (more_dirs && bigenough) { 2696 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2697 fileno = fxdr_hyper(tl); 2698 len = fxdr_unsigned(int, *(tl + 2)); 2699 if (len <= 0 || len > NFS_MAXNAMLEN) { 2700 error = EBADRPC; 2701 m_freem(mrep); 2702 goto nfsmout; 2703 } 2704 /* for cookie stashing */ 2705 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2706 left = NFS_DIRFRAGSIZ - blksiz; 2707 if (reclen > left) { 2708 /* 2709 * DIRFRAGSIZ is aligned, no need to align 2710 * again here. 2711 */ 2712 memset(uiop->uio_iov->iov_base, 0, left); 2713 dp->d_reclen += left; 2714 UIO_ADVANCE(uiop, left); 2715 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2716 blksiz = 0; 2717 } 2718 if (reclen > uiop->uio_resid) 2719 bigenough = 0; 2720 if (bigenough) { 2721 int tlen; 2722 2723 dp = (struct dirent *)uiop->uio_iov->iov_base; 2724 dp->d_fileno = fileno; 2725 dp->d_namlen = len; 2726 dp->d_reclen = reclen; 2727 dp->d_type = DT_UNKNOWN; 2728 blksiz += reclen; 2729 if (blksiz == NFS_DIRFRAGSIZ) 2730 blksiz = 0; 2731 UIO_ADVANCE(uiop, DIRHDSIZ); 2732 nfsm_mtouio(uiop, len); 2733 tlen = reclen - (DIRHDSIZ + len); 2734 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2735 UIO_ADVANCE(uiop, tlen); 2736 cnp->cn_nameptr = dp->d_name; 2737 cnp->cn_namelen = dp->d_namlen; 2738 } else 2739 nfsm_adv(nfsm_rndup(len)); 2740 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2741 if (bigenough) { 2742 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2743 uiop->uio_offset = 2744 fxdr_swapcookie3(tl); 2745 else 2746 uiop->uio_offset = 2747 fxdr_cookie3(tl); 2748 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2749 } 2750 tl += 2; 2751 2752 /* 2753 * Since the attributes are before the file handle 2754 * (sigh), we must skip over the attributes and then 2755 * come back and get them. 2756 */ 2757 attrflag = fxdr_unsigned(int, *tl); 2758 if (attrflag) { 2759 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2760 memcpy(&fattr, fp, NFSX_V3FATTR); 2761 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2762 doit = fxdr_unsigned(int, *tl); 2763 if (doit) { 2764 nfsm_getfh(fhp, fhsize, 1); 2765 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2766 vref(vp); 2767 newvp = vp; 2768 np = dnp; 2769 } else { 2770 error = nfs_nget1(vp->v_mount, fhp, 2771 fhsize, &np, LK_NOWAIT); 2772 if (!error) 2773 newvp = NFSTOV(np); 2774 } 2775 if (!error) { 2776 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2777 if (bigenough) { 2778 dp->d_type = 2779 IFTODT(VTTOIF(np->n_vattr->va_type)); 2780 if (cnp->cn_namelen <= NCHNAMLEN) { 2781 ndp->ni_vp = newvp; 2782 nfs_cache_enter(ndp->ni_dvp, 2783 ndp->ni_vp, cnp); 2784 } 2785 } 2786 } 2787 error = 0; 2788 } 2789 } else { 2790 /* Just skip over the file handle */ 2791 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2792 i = fxdr_unsigned(int, *tl); 2793 nfsm_adv(nfsm_rndup(i)); 2794 } 2795 if (newvp != NULLVP) { 2796 if (newvp == vp) 2797 vrele(newvp); 2798 else 2799 vput(newvp); 2800 newvp = NULLVP; 2801 } 2802 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2803 more_dirs = fxdr_unsigned(int, *tl); 2804 } 2805 /* 2806 * If at end of rpc data, get the eof boolean 2807 */ 2808 if (!more_dirs) { 2809 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2810 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2811 2812 /* 2813 * kludge: see a comment in nfs_readdirrpc. 2814 */ 2815 2816 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2817 more_dirs = 0; 2818 } 2819 m_freem(mrep); 2820 } 2821 /* 2822 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2823 * by increasing d_reclen for the last record. 2824 */ 2825 if (blksiz > 0) { 2826 left = NFS_DIRFRAGSIZ - blksiz; 2827 memset(uiop->uio_iov->iov_base, 0, left); 2828 dp->d_reclen += left; 2829 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2830 UIO_ADVANCE(uiop, left); 2831 } 2832 2833 /* 2834 * We are now either at the end of the directory or have filled the 2835 * block. 2836 */ 2837 if (bigenough) { 2838 dnp->n_direofoffset = uiop->uio_offset; 2839 dnp->n_flag |= NEOFVALID; 2840 } 2841 nfsmout: 2842 if (newvp != NULLVP) { 2843 if(newvp == vp) 2844 vrele(newvp); 2845 else 2846 vput(newvp); 2847 } 2848 return (error); 2849 } 2850 #endif 2851 2852 /* 2853 * Silly rename. To make the NFS filesystem that is stateless look a little 2854 * more like the "ufs" a remove of an active vnode is translated to a rename 2855 * to a funny looking filename that is removed by nfs_inactive on the 2856 * nfsnode. There is the potential for another process on a different client 2857 * to create the same funny name between the nfs_lookitup() fails and the 2858 * nfs_rename() completes, but... 2859 */ 2860 int 2861 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2862 { 2863 struct sillyrename *sp; 2864 struct nfsnode *np; 2865 int error; 2866 pid_t pid; 2867 2868 cache_purge(dvp); 2869 np = VTONFS(vp); 2870 #ifndef DIAGNOSTIC 2871 if (vp->v_type == VDIR) 2872 panic("nfs: sillyrename dir"); 2873 #endif 2874 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2875 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2876 sp->s_dvp = dvp; 2877 vref(dvp); 2878 2879 /* Fudge together a funny name */ 2880 pid = curlwp->l_proc->p_pid; 2881 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2882 sp->s_namlen = 12; 2883 sp->s_name[8] = hexdigits[pid & 0xf]; 2884 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2885 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2886 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2887 2888 /* Try lookitups until we get one that isn't there */ 2889 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2890 curlwp, (struct nfsnode **)0) == 0) { 2891 sp->s_name[4]++; 2892 if (sp->s_name[4] > 'z') { 2893 error = EINVAL; 2894 goto bad; 2895 } 2896 } 2897 if (dolink) { 2898 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2899 sp->s_cred, curlwp); 2900 /* 2901 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2902 */ 2903 if (error == ENOTSUP) { 2904 error = nfs_renameit(dvp, cnp, sp); 2905 } 2906 } else { 2907 error = nfs_renameit(dvp, cnp, sp); 2908 } 2909 if (error) 2910 goto bad; 2911 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2912 curlwp, &np); 2913 np->n_sillyrename = sp; 2914 return (0); 2915 bad: 2916 vrele(sp->s_dvp); 2917 kauth_cred_free(sp->s_cred); 2918 kmem_free(sp, sizeof(*sp)); 2919 return (error); 2920 } 2921 2922 /* 2923 * Look up a file name and optionally either update the file handle or 2924 * allocate an nfsnode, depending on the value of npp. 2925 * npp == NULL --> just do the lookup 2926 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2927 * handled too 2928 * *npp != NULL --> update the file handle in the vnode 2929 */ 2930 int 2931 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2932 { 2933 u_int32_t *tl; 2934 char *cp; 2935 int32_t t1, t2; 2936 struct vnode *newvp = (struct vnode *)0; 2937 struct nfsnode *np, *dnp = VTONFS(dvp); 2938 char *bpos, *dpos, *cp2; 2939 int error = 0, fhlen; 2940 #ifndef NFS_V2_ONLY 2941 int attrflag; 2942 #endif 2943 struct mbuf *mreq, *mrep, *md, *mb; 2944 nfsfh_t *nfhp; 2945 const int v3 = NFS_ISV3(dvp); 2946 2947 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2948 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2949 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2950 nfsm_fhtom(dnp, v3); 2951 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2952 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2953 if (npp && !error) { 2954 nfsm_getfh(nfhp, fhlen, v3); 2955 if (*npp) { 2956 np = *npp; 2957 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2958 kmem_free(np->n_fhp, np->n_fhsize); 2959 np->n_fhp = &np->n_fh; 2960 } 2961 #if NFS_SMALLFH < NFSX_V3FHMAX 2962 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2963 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2964 #endif 2965 memcpy(np->n_fhp, nfhp, fhlen); 2966 np->n_fhsize = fhlen; 2967 newvp = NFSTOV(np); 2968 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2969 vref(dvp); 2970 newvp = dvp; 2971 np = dnp; 2972 } else { 2973 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2974 if (error) { 2975 m_freem(mrep); 2976 return (error); 2977 } 2978 newvp = NFSTOV(np); 2979 } 2980 #ifndef NFS_V2_ONLY 2981 if (v3) { 2982 nfsm_postop_attr(newvp, attrflag, 0); 2983 if (!attrflag && *npp == NULL) { 2984 m_freem(mrep); 2985 vput(newvp); 2986 return (ENOENT); 2987 } 2988 } else 2989 #endif 2990 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2991 } 2992 nfsm_reqdone; 2993 if (npp && *npp == NULL) { 2994 if (error) { 2995 if (newvp) 2996 vput(newvp); 2997 } else 2998 *npp = np; 2999 } 3000 return (error); 3001 } 3002 3003 #ifndef NFS_V2_ONLY 3004 /* 3005 * Nfs Version 3 commit rpc 3006 */ 3007 int 3008 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3009 { 3010 char *cp; 3011 u_int32_t *tl; 3012 int32_t t1, t2; 3013 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3014 char *bpos, *dpos, *cp2; 3015 int error = 0, wccflag = NFSV3_WCCRATTR; 3016 struct mbuf *mreq, *mrep, *md, *mb; 3017 struct nfsnode *np; 3018 3019 KASSERT(NFS_ISV3(vp)); 3020 3021 #ifdef NFS_DEBUG_COMMIT 3022 printf("commit %lu - %lu\n", (unsigned long)offset, 3023 (unsigned long)(offset + cnt)); 3024 #endif 3025 3026 mutex_enter(&nmp->nm_lock); 3027 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3028 mutex_exit(&nmp->nm_lock); 3029 return (0); 3030 } 3031 mutex_exit(&nmp->nm_lock); 3032 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3033 np = VTONFS(vp); 3034 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3035 nfsm_fhtom(np, 1); 3036 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3037 txdr_hyper(offset, tl); 3038 tl += 2; 3039 *tl = txdr_unsigned(cnt); 3040 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3041 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3042 if (!error) { 3043 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3044 mutex_enter(&nmp->nm_lock); 3045 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3046 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3047 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3048 error = NFSERR_STALEWRITEVERF; 3049 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3050 } 3051 mutex_exit(&nmp->nm_lock); 3052 } 3053 nfsm_reqdone; 3054 return (error); 3055 } 3056 #endif 3057 3058 /* 3059 * Kludge City.. 3060 * - make nfs_bmap() essentially a no-op that does no translation 3061 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3062 * (Maybe I could use the process's page mapping, but I was concerned that 3063 * Kernel Write might not be enabled and also figured copyout() would do 3064 * a lot more work than memcpy() and also it currently happens in the 3065 * context of the swapper process (2). 3066 */ 3067 int 3068 nfs_bmap(void *v) 3069 { 3070 struct vop_bmap_args /* { 3071 struct vnode *a_vp; 3072 daddr_t a_bn; 3073 struct vnode **a_vpp; 3074 daddr_t *a_bnp; 3075 int *a_runp; 3076 } */ *ap = v; 3077 struct vnode *vp = ap->a_vp; 3078 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3079 3080 if (ap->a_vpp != NULL) 3081 *ap->a_vpp = vp; 3082 if (ap->a_bnp != NULL) 3083 *ap->a_bnp = ap->a_bn << bshift; 3084 if (ap->a_runp != NULL) 3085 *ap->a_runp = 1024 * 1024; /* XXX */ 3086 return (0); 3087 } 3088 3089 /* 3090 * Strategy routine. 3091 * For async requests when nfsiod(s) are running, queue the request by 3092 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3093 * request. 3094 */ 3095 int 3096 nfs_strategy(void *v) 3097 { 3098 struct vop_strategy_args *ap = v; 3099 struct buf *bp = ap->a_bp; 3100 int error = 0; 3101 3102 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3103 panic("nfs physio/async"); 3104 3105 /* 3106 * If the op is asynchronous and an i/o daemon is waiting 3107 * queue the request, wake it up and wait for completion 3108 * otherwise just do it ourselves. 3109 */ 3110 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3111 error = nfs_doio(bp); 3112 return (error); 3113 } 3114 3115 /* 3116 * fsync vnode op. Just call nfs_flush() with commit == 1. 3117 */ 3118 /* ARGSUSED */ 3119 int 3120 nfs_fsync(void *v) 3121 { 3122 struct vop_fsync_args /* { 3123 struct vnodeop_desc *a_desc; 3124 struct vnode * a_vp; 3125 kauth_cred_t a_cred; 3126 int a_flags; 3127 off_t offlo; 3128 off_t offhi; 3129 struct lwp * a_l; 3130 } */ *ap = v; 3131 3132 struct vnode *vp = ap->a_vp; 3133 3134 if (vp->v_type != VREG) 3135 return 0; 3136 3137 return (nfs_flush(vp, ap->a_cred, 3138 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3139 } 3140 3141 /* 3142 * Flush all the data associated with a vnode. 3143 */ 3144 int 3145 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3146 int commit) 3147 { 3148 struct nfsnode *np = VTONFS(vp); 3149 int error; 3150 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3151 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3152 3153 mutex_enter(vp->v_interlock); 3154 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3155 if (np->n_flag & NWRITEERR) { 3156 error = np->n_error; 3157 np->n_flag &= ~NWRITEERR; 3158 } 3159 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3160 return (error); 3161 } 3162 3163 /* 3164 * Return POSIX pathconf information applicable to nfs. 3165 * 3166 * N.B. The NFS V2 protocol doesn't support this RPC. 3167 */ 3168 /* ARGSUSED */ 3169 int 3170 nfs_pathconf(void *v) 3171 { 3172 struct vop_pathconf_args /* { 3173 struct vnode *a_vp; 3174 int a_name; 3175 register_t *a_retval; 3176 } */ *ap = v; 3177 struct nfsv3_pathconf *pcp; 3178 struct vnode *vp = ap->a_vp; 3179 struct mbuf *mreq, *mrep, *md, *mb; 3180 int32_t t1, t2; 3181 u_int32_t *tl; 3182 char *bpos, *dpos, *cp, *cp2; 3183 int error = 0, attrflag; 3184 #ifndef NFS_V2_ONLY 3185 struct nfsmount *nmp; 3186 unsigned int l; 3187 u_int64_t maxsize; 3188 #endif 3189 const int v3 = NFS_ISV3(vp); 3190 struct nfsnode *np = VTONFS(vp); 3191 3192 switch (ap->a_name) { 3193 /* Names that can be resolved locally. */ 3194 case _PC_PIPE_BUF: 3195 *ap->a_retval = PIPE_BUF; 3196 break; 3197 case _PC_SYNC_IO: 3198 *ap->a_retval = 1; 3199 break; 3200 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3201 case _PC_LINK_MAX: 3202 case _PC_NAME_MAX: 3203 case _PC_CHOWN_RESTRICTED: 3204 case _PC_NO_TRUNC: 3205 if (!v3) { 3206 error = EINVAL; 3207 break; 3208 } 3209 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3210 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3211 nfsm_fhtom(np, 1); 3212 nfsm_request(np, NFSPROC_PATHCONF, 3213 curlwp, curlwp->l_cred); /* XXX */ 3214 nfsm_postop_attr(vp, attrflag, 0); 3215 if (!error) { 3216 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3217 NFSX_V3PATHCONF); 3218 switch (ap->a_name) { 3219 case _PC_LINK_MAX: 3220 *ap->a_retval = 3221 fxdr_unsigned(register_t, pcp->pc_linkmax); 3222 break; 3223 case _PC_NAME_MAX: 3224 *ap->a_retval = 3225 fxdr_unsigned(register_t, pcp->pc_namemax); 3226 break; 3227 case _PC_CHOWN_RESTRICTED: 3228 *ap->a_retval = 3229 (pcp->pc_chownrestricted == nfs_true); 3230 break; 3231 case _PC_NO_TRUNC: 3232 *ap->a_retval = 3233 (pcp->pc_notrunc == nfs_true); 3234 break; 3235 } 3236 } 3237 nfsm_reqdone; 3238 break; 3239 case _PC_FILESIZEBITS: 3240 #ifndef NFS_V2_ONLY 3241 if (v3) { 3242 nmp = VFSTONFS(vp->v_mount); 3243 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3244 if ((error = nfs_fsinfo(nmp, vp, 3245 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3246 break; 3247 for (l = 0, maxsize = nmp->nm_maxfilesize; 3248 (maxsize >> l) > 0; l++) 3249 ; 3250 *ap->a_retval = l + 1; 3251 } else 3252 #endif 3253 { 3254 *ap->a_retval = 32; /* NFS V2 limitation */ 3255 } 3256 break; 3257 default: 3258 error = EINVAL; 3259 break; 3260 } 3261 3262 return (error); 3263 } 3264 3265 /* 3266 * NFS advisory byte-level locks. 3267 */ 3268 int 3269 nfs_advlock(void *v) 3270 { 3271 struct vop_advlock_args /* { 3272 struct vnode *a_vp; 3273 void *a_id; 3274 int a_op; 3275 struct flock *a_fl; 3276 int a_flags; 3277 } */ *ap = v; 3278 struct nfsnode *np = VTONFS(ap->a_vp); 3279 3280 return lf_advlock(ap, &np->n_lockf, np->n_size); 3281 } 3282 3283 /* 3284 * Print out the contents of an nfsnode. 3285 */ 3286 int 3287 nfs_print(void *v) 3288 { 3289 struct vop_print_args /* { 3290 struct vnode *a_vp; 3291 } */ *ap = v; 3292 struct vnode *vp = ap->a_vp; 3293 struct nfsnode *np = VTONFS(vp); 3294 3295 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3296 (unsigned long long)np->n_vattr->va_fileid, 3297 (unsigned long long)np->n_vattr->va_fsid); 3298 if (vp->v_type == VFIFO) 3299 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3300 printf("\n"); 3301 return (0); 3302 } 3303 3304 /* 3305 * nfs unlock wrapper. 3306 */ 3307 int 3308 nfs_unlock(void *v) 3309 { 3310 struct vop_unlock_args /* { 3311 struct vnode *a_vp; 3312 int a_flags; 3313 } */ *ap = v; 3314 struct vnode *vp = ap->a_vp; 3315 3316 /* 3317 * VOP_UNLOCK can be called by nfs_loadattrcache 3318 * with v_data == 0. 3319 */ 3320 if (VTONFS(vp)) { 3321 nfs_delayedtruncate(vp); 3322 } 3323 3324 return genfs_unlock(v); 3325 } 3326 3327 /* 3328 * nfs special file access vnode op. 3329 * Essentially just get vattr and then imitate iaccess() since the device is 3330 * local to the client. 3331 */ 3332 int 3333 nfsspec_access(void *v) 3334 { 3335 struct vop_access_args /* { 3336 struct vnode *a_vp; 3337 int a_mode; 3338 kauth_cred_t a_cred; 3339 struct lwp *a_l; 3340 } */ *ap = v; 3341 struct vattr va; 3342 struct vnode *vp = ap->a_vp; 3343 int error; 3344 3345 error = VOP_GETATTR(vp, &va, ap->a_cred); 3346 if (error) 3347 return (error); 3348 3349 /* 3350 * Disallow write attempts on filesystems mounted read-only; 3351 * unless the file is a socket, fifo, or a block or character 3352 * device resident on the filesystem. 3353 */ 3354 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3355 switch (vp->v_type) { 3356 case VREG: 3357 case VDIR: 3358 case VLNK: 3359 return (EROFS); 3360 default: 3361 break; 3362 } 3363 } 3364 3365 return kauth_authorize_vnode(ap->a_cred, kauth_access_action(ap->a_mode, 3366 va.va_type, va.va_mode), vp, NULL, genfs_can_access(va.va_type, 3367 va.va_mode, va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3368 } 3369 3370 /* 3371 * Read wrapper for special devices. 3372 */ 3373 int 3374 nfsspec_read(void *v) 3375 { 3376 struct vop_read_args /* { 3377 struct vnode *a_vp; 3378 struct uio *a_uio; 3379 int a_ioflag; 3380 kauth_cred_t a_cred; 3381 } */ *ap = v; 3382 struct nfsnode *np = VTONFS(ap->a_vp); 3383 3384 /* 3385 * Set access flag. 3386 */ 3387 np->n_flag |= NACC; 3388 getnanotime(&np->n_atim); 3389 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3390 } 3391 3392 /* 3393 * Write wrapper for special devices. 3394 */ 3395 int 3396 nfsspec_write(void *v) 3397 { 3398 struct vop_write_args /* { 3399 struct vnode *a_vp; 3400 struct uio *a_uio; 3401 int a_ioflag; 3402 kauth_cred_t a_cred; 3403 } */ *ap = v; 3404 struct nfsnode *np = VTONFS(ap->a_vp); 3405 3406 /* 3407 * Set update flag. 3408 */ 3409 np->n_flag |= NUPD; 3410 getnanotime(&np->n_mtim); 3411 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3412 } 3413 3414 /* 3415 * Close wrapper for special devices. 3416 * 3417 * Update the times on the nfsnode then do device close. 3418 */ 3419 int 3420 nfsspec_close(void *v) 3421 { 3422 struct vop_close_args /* { 3423 struct vnode *a_vp; 3424 int a_fflag; 3425 kauth_cred_t a_cred; 3426 struct lwp *a_l; 3427 } */ *ap = v; 3428 struct vnode *vp = ap->a_vp; 3429 struct nfsnode *np = VTONFS(vp); 3430 struct vattr vattr; 3431 3432 if (np->n_flag & (NACC | NUPD)) { 3433 np->n_flag |= NCHG; 3434 if (vp->v_usecount == 1 && 3435 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3436 vattr_null(&vattr); 3437 if (np->n_flag & NACC) 3438 vattr.va_atime = np->n_atim; 3439 if (np->n_flag & NUPD) 3440 vattr.va_mtime = np->n_mtim; 3441 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3442 } 3443 } 3444 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3445 } 3446 3447 /* 3448 * Read wrapper for fifos. 3449 */ 3450 int 3451 nfsfifo_read(void *v) 3452 { 3453 struct vop_read_args /* { 3454 struct vnode *a_vp; 3455 struct uio *a_uio; 3456 int a_ioflag; 3457 kauth_cred_t a_cred; 3458 } */ *ap = v; 3459 struct nfsnode *np = VTONFS(ap->a_vp); 3460 3461 /* 3462 * Set access flag. 3463 */ 3464 np->n_flag |= NACC; 3465 getnanotime(&np->n_atim); 3466 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3467 } 3468 3469 /* 3470 * Write wrapper for fifos. 3471 */ 3472 int 3473 nfsfifo_write(void *v) 3474 { 3475 struct vop_write_args /* { 3476 struct vnode *a_vp; 3477 struct uio *a_uio; 3478 int a_ioflag; 3479 kauth_cred_t a_cred; 3480 } */ *ap = v; 3481 struct nfsnode *np = VTONFS(ap->a_vp); 3482 3483 /* 3484 * Set update flag. 3485 */ 3486 np->n_flag |= NUPD; 3487 getnanotime(&np->n_mtim); 3488 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3489 } 3490 3491 /* 3492 * Close wrapper for fifos. 3493 * 3494 * Update the times on the nfsnode then do fifo close. 3495 */ 3496 int 3497 nfsfifo_close(void *v) 3498 { 3499 struct vop_close_args /* { 3500 struct vnode *a_vp; 3501 int a_fflag; 3502 kauth_cred_t a_cred; 3503 struct lwp *a_l; 3504 } */ *ap = v; 3505 struct vnode *vp = ap->a_vp; 3506 struct nfsnode *np = VTONFS(vp); 3507 struct vattr vattr; 3508 3509 if (np->n_flag & (NACC | NUPD)) { 3510 struct timespec ts; 3511 3512 getnanotime(&ts); 3513 if (np->n_flag & NACC) 3514 np->n_atim = ts; 3515 if (np->n_flag & NUPD) 3516 np->n_mtim = ts; 3517 np->n_flag |= NCHG; 3518 if (vp->v_usecount == 1 && 3519 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3520 vattr_null(&vattr); 3521 if (np->n_flag & NACC) 3522 vattr.va_atime = np->n_atim; 3523 if (np->n_flag & NUPD) 3524 vattr.va_mtime = np->n_mtim; 3525 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3526 } 3527 } 3528 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3529 } 3530
Indexes created Thu Sep 25 16:09:42 GMT 2025