1 /* $NetBSD: nfs_vnops.c,v 1.290 2011/04/24 21:35:30 rmind 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 1.290 2011/04/24 21:35:30 rmind 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 72 #include <uvm/uvm_extern.h> 73 #include <uvm/uvm.h> 74 75 #include <miscfs/fifofs/fifo.h> 76 #include <miscfs/genfs/genfs.h> 77 #include <miscfs/genfs/genfs_node.h> 78 #include <miscfs/specfs/specdev.h> 79 80 #include <nfs/rpcv2.h> 81 #include <nfs/nfsproto.h> 82 #include <nfs/nfs.h> 83 #include <nfs/nfsnode.h> 84 #include <nfs/nfsmount.h> 85 #include <nfs/xdr_subs.h> 86 #include <nfs/nfsm_subs.h> 87 #include <nfs/nfs_var.h> 88 89 #include <net/if.h> 90 #include <netinet/in.h> 91 #include <netinet/in_var.h> 92 93 /* 94 * Global vfs data structures for nfs 95 */ 96 int (**nfsv2_vnodeop_p)(void *); 97 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 98 { &vop_default_desc, vn_default_error }, 99 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 100 { &vop_create_desc, nfs_create }, /* create */ 101 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 102 { &vop_open_desc, nfs_open }, /* open */ 103 { &vop_close_desc, nfs_close }, /* close */ 104 { &vop_access_desc, nfs_access }, /* access */ 105 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 106 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 107 { &vop_read_desc, nfs_read }, /* read */ 108 { &vop_write_desc, nfs_write }, /* write */ 109 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 110 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 111 { &vop_poll_desc, nfs_poll }, /* poll */ 112 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 113 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 114 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 115 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 116 { &vop_seek_desc, nfs_seek }, /* seek */ 117 { &vop_remove_desc, nfs_remove }, /* remove */ 118 { &vop_link_desc, nfs_link }, /* link */ 119 { &vop_rename_desc, nfs_rename }, /* rename */ 120 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 121 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 122 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 123 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 124 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 125 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 126 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 127 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 128 { &vop_lock_desc, nfs_lock }, /* lock */ 129 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 130 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 131 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 132 { &vop_print_desc, nfs_print }, /* print */ 133 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 134 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 135 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 136 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 137 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 138 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 139 { NULL, NULL } 140 }; 141 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 142 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 143 144 /* 145 * Special device vnode ops 146 */ 147 int (**spec_nfsv2nodeop_p)(void *); 148 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 149 { &vop_default_desc, vn_default_error }, 150 { &vop_lookup_desc, spec_lookup }, /* lookup */ 151 { &vop_create_desc, spec_create }, /* create */ 152 { &vop_mknod_desc, spec_mknod }, /* mknod */ 153 { &vop_open_desc, spec_open }, /* open */ 154 { &vop_close_desc, nfsspec_close }, /* close */ 155 { &vop_access_desc, nfsspec_access }, /* access */ 156 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 157 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 158 { &vop_read_desc, nfsspec_read }, /* read */ 159 { &vop_write_desc, nfsspec_write }, /* write */ 160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 162 { &vop_poll_desc, spec_poll }, /* poll */ 163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 164 { &vop_revoke_desc, spec_revoke }, /* revoke */ 165 { &vop_mmap_desc, spec_mmap }, /* mmap */ 166 { &vop_fsync_desc, spec_fsync }, /* fsync */ 167 { &vop_seek_desc, spec_seek }, /* seek */ 168 { &vop_remove_desc, spec_remove }, /* remove */ 169 { &vop_link_desc, spec_link }, /* link */ 170 { &vop_rename_desc, spec_rename }, /* rename */ 171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 173 { &vop_symlink_desc, spec_symlink }, /* symlink */ 174 { &vop_readdir_desc, spec_readdir }, /* readdir */ 175 { &vop_readlink_desc, spec_readlink }, /* readlink */ 176 { &vop_abortop_desc, spec_abortop }, /* abortop */ 177 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 179 { &vop_lock_desc, nfs_lock }, /* lock */ 180 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 181 { &vop_bmap_desc, spec_bmap }, /* bmap */ 182 { &vop_strategy_desc, spec_strategy }, /* strategy */ 183 { &vop_print_desc, nfs_print }, /* print */ 184 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, spec_advlock }, /* advlock */ 187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 188 { &vop_getpages_desc, spec_getpages }, /* getpages */ 189 { &vop_putpages_desc, spec_putpages }, /* putpages */ 190 { NULL, NULL } 191 }; 192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 194 195 int (**fifo_nfsv2nodeop_p)(void *); 196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 197 { &vop_default_desc, vn_default_error }, 198 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 199 { &vop_create_desc, vn_fifo_bypass }, /* create */ 200 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 201 { &vop_open_desc, vn_fifo_bypass }, /* open */ 202 { &vop_close_desc, nfsfifo_close }, /* close */ 203 { &vop_access_desc, nfsspec_access }, /* access */ 204 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 205 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 206 { &vop_read_desc, nfsfifo_read }, /* read */ 207 { &vop_write_desc, nfsfifo_write }, /* write */ 208 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 209 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 210 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 211 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 212 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 213 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 214 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 215 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 216 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 217 { &vop_link_desc, vn_fifo_bypass }, /* link */ 218 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 219 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 220 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 221 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 222 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 223 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 224 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 225 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 226 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 227 { &vop_lock_desc, nfs_lock }, /* lock */ 228 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 229 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 230 { &vop_strategy_desc, genfs_badop }, /* strategy */ 231 { &vop_print_desc, nfs_print }, /* print */ 232 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 233 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 234 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 235 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 236 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 237 { NULL, NULL } 238 }; 239 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 240 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 241 242 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 243 size_t, kauth_cred_t, struct lwp *); 244 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 245 246 /* 247 * Global variables 248 */ 249 extern u_int32_t nfs_true, nfs_false; 250 extern u_int32_t nfs_xdrneg1; 251 extern const nfstype nfsv3_type[9]; 252 253 int nfs_numasync = 0; 254 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 255 #define UIO_ADVANCE(uio, siz) \ 256 (void)((uio)->uio_resid -= (siz), \ 257 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 258 (uio)->uio_iov->iov_len -= (siz)) 259 260 static void nfs_cache_enter(struct vnode *, struct vnode *, 261 struct componentname *); 262 263 static void 264 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 265 struct componentname *cnp) 266 { 267 struct nfsnode *dnp = VTONFS(dvp); 268 269 if (vp != NULL) { 270 struct nfsnode *np = VTONFS(vp); 271 272 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 273 } 274 275 if (!timespecisset(&dnp->n_nctime)) 276 dnp->n_nctime = dnp->n_vattr->va_mtime; 277 278 cache_enter(dvp, vp, cnp); 279 } 280 281 /* 282 * nfs null call from vfs. 283 */ 284 int 285 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 286 { 287 char *bpos, *dpos; 288 int error = 0; 289 struct mbuf *mreq, *mrep, *md, *mb; 290 struct nfsnode *np = VTONFS(vp); 291 292 nfsm_reqhead(np, NFSPROC_NULL, 0); 293 nfsm_request(np, NFSPROC_NULL, l, cred); 294 nfsm_reqdone; 295 return (error); 296 } 297 298 /* 299 * nfs access vnode op. 300 * For nfs version 2, just return ok. File accesses may fail later. 301 * For nfs version 3, use the access rpc to check accessibility. If file modes 302 * are changed on the server, accesses might still fail later. 303 */ 304 int 305 nfs_access(void *v) 306 { 307 struct vop_access_args /* { 308 struct vnode *a_vp; 309 int a_mode; 310 kauth_cred_t a_cred; 311 } */ *ap = v; 312 struct vnode *vp = ap->a_vp; 313 #ifndef NFS_V2_ONLY 314 u_int32_t *tl; 315 char *cp; 316 int32_t t1, t2; 317 char *bpos, *dpos, *cp2; 318 int error = 0, attrflag; 319 struct mbuf *mreq, *mrep, *md, *mb; 320 u_int32_t mode, rmode; 321 const int v3 = NFS_ISV3(vp); 322 #endif 323 int cachevalid; 324 struct nfsnode *np = VTONFS(vp); 325 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 326 327 cachevalid = (np->n_accstamp != -1 && 328 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 329 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 330 331 /* 332 * Check access cache first. If this request has been made for this 333 * uid shortly before, use the cached result. 334 */ 335 if (cachevalid) { 336 if (!np->n_accerror) { 337 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 338 return np->n_accerror; 339 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 340 return np->n_accerror; 341 } 342 343 #ifndef NFS_V2_ONLY 344 /* 345 * For nfs v3, do an access rpc, otherwise you are stuck emulating 346 * ufs_access() locally using the vattr. This may not be correct, 347 * since the server may apply other access criteria such as 348 * client uid-->server uid mapping that we do not know about, but 349 * this is better than just returning anything that is lying about 350 * in the cache. 351 */ 352 if (v3) { 353 nfsstats.rpccnt[NFSPROC_ACCESS]++; 354 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 355 nfsm_fhtom(np, v3); 356 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 357 if (ap->a_mode & VREAD) 358 mode = NFSV3ACCESS_READ; 359 else 360 mode = 0; 361 if (vp->v_type != VDIR) { 362 if (ap->a_mode & VWRITE) 363 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 364 if (ap->a_mode & VEXEC) 365 mode |= NFSV3ACCESS_EXECUTE; 366 } else { 367 if (ap->a_mode & VWRITE) 368 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 369 NFSV3ACCESS_DELETE); 370 if (ap->a_mode & VEXEC) 371 mode |= NFSV3ACCESS_LOOKUP; 372 } 373 *tl = txdr_unsigned(mode); 374 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 375 nfsm_postop_attr(vp, attrflag, 0); 376 if (!error) { 377 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 378 rmode = fxdr_unsigned(u_int32_t, *tl); 379 /* 380 * The NFS V3 spec does not clarify whether or not 381 * the returned access bits can be a superset of 382 * the ones requested, so... 383 */ 384 if ((rmode & mode) != mode) 385 error = EACCES; 386 } 387 nfsm_reqdone; 388 } else 389 #endif 390 return (nfsspec_access(ap)); 391 #ifndef NFS_V2_ONLY 392 /* 393 * Disallow write attempts on filesystems mounted read-only; 394 * unless the file is a socket, fifo, or a block or character 395 * device resident on the filesystem. 396 */ 397 if (!error && (ap->a_mode & VWRITE) && 398 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 399 switch (vp->v_type) { 400 case VREG: 401 case VDIR: 402 case VLNK: 403 error = EROFS; 404 default: 405 break; 406 } 407 } 408 409 if (!error || error == EACCES) { 410 /* 411 * If we got the same result as for a previous, 412 * different request, OR it in. Don't update 413 * the timestamp in that case. 414 */ 415 if (cachevalid && np->n_accstamp != -1 && 416 error == np->n_accerror) { 417 if (!error) 418 np->n_accmode |= ap->a_mode; 419 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 420 np->n_accmode = ap->a_mode; 421 } else { 422 np->n_accstamp = time_uptime; 423 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 424 np->n_accmode = ap->a_mode; 425 np->n_accerror = error; 426 } 427 } 428 429 return (error); 430 #endif 431 } 432 433 /* 434 * nfs open vnode op 435 * Check to see if the type is ok 436 * and that deletion is not in progress. 437 * For paged in text files, you will need to flush the page cache 438 * if consistency is lost. 439 */ 440 /* ARGSUSED */ 441 int 442 nfs_open(void *v) 443 { 444 struct vop_open_args /* { 445 struct vnode *a_vp; 446 int a_mode; 447 kauth_cred_t a_cred; 448 } */ *ap = v; 449 struct vnode *vp = ap->a_vp; 450 struct nfsnode *np = VTONFS(vp); 451 int error; 452 453 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 454 return (EACCES); 455 } 456 457 if (ap->a_mode & FREAD) { 458 if (np->n_rcred != NULL) 459 kauth_cred_free(np->n_rcred); 460 np->n_rcred = ap->a_cred; 461 kauth_cred_hold(np->n_rcred); 462 } 463 if (ap->a_mode & FWRITE) { 464 if (np->n_wcred != NULL) 465 kauth_cred_free(np->n_wcred); 466 np->n_wcred = ap->a_cred; 467 kauth_cred_hold(np->n_wcred); 468 } 469 470 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 471 if (error) 472 return error; 473 474 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 475 476 return (0); 477 } 478 479 /* 480 * nfs close vnode op 481 * What an NFS client should do upon close after writing is a debatable issue. 482 * Most NFS clients push delayed writes to the server upon close, basically for 483 * two reasons: 484 * 1 - So that any write errors may be reported back to the client process 485 * doing the close system call. By far the two most likely errors are 486 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 487 * 2 - To put a worst case upper bound on cache inconsistency between 488 * multiple clients for the file. 489 * There is also a consistency problem for Version 2 of the protocol w.r.t. 490 * not being able to tell if other clients are writing a file concurrently, 491 * since there is no way of knowing if the changed modify time in the reply 492 * is only due to the write for this client. 493 * (NFS Version 3 provides weak cache consistency data in the reply that 494 * should be sufficient to detect and handle this case.) 495 * 496 * The current code does the following: 497 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 498 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 499 * or commit them (this satisfies 1 and 2 except for the 500 * case where the server crashes after this close but 501 * before the commit RPC, which is felt to be "good 502 * enough". Changing the last argument to nfs_flush() to 503 * a 1 would force a commit operation, if it is felt a 504 * commit is necessary now. 505 */ 506 /* ARGSUSED */ 507 int 508 nfs_close(void *v) 509 { 510 struct vop_close_args /* { 511 struct vnodeop_desc *a_desc; 512 struct vnode *a_vp; 513 int a_fflag; 514 kauth_cred_t a_cred; 515 } */ *ap = v; 516 struct vnode *vp = ap->a_vp; 517 struct nfsnode *np = VTONFS(vp); 518 int error = 0; 519 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 520 521 if (vp->v_type == VREG) { 522 if (np->n_flag & NMODIFIED) { 523 #ifndef NFS_V2_ONLY 524 if (NFS_ISV3(vp)) { 525 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 526 np->n_flag &= ~NMODIFIED; 527 } else 528 #endif 529 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 530 NFS_INVALIDATE_ATTRCACHE(np); 531 } 532 if (np->n_flag & NWRITEERR) { 533 np->n_flag &= ~NWRITEERR; 534 error = np->n_error; 535 } 536 } 537 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 538 return (error); 539 } 540 541 /* 542 * nfs getattr call from vfs. 543 */ 544 int 545 nfs_getattr(void *v) 546 { 547 struct vop_getattr_args /* { 548 struct vnode *a_vp; 549 struct vattr *a_vap; 550 kauth_cred_t a_cred; 551 } */ *ap = v; 552 struct vnode *vp = ap->a_vp; 553 struct nfsnode *np = VTONFS(vp); 554 char *cp; 555 u_int32_t *tl; 556 int32_t t1, t2; 557 char *bpos, *dpos; 558 int error = 0; 559 struct mbuf *mreq, *mrep, *md, *mb; 560 const int v3 = NFS_ISV3(vp); 561 562 /* 563 * Update local times for special files. 564 */ 565 if (np->n_flag & (NACC | NUPD)) 566 np->n_flag |= NCHG; 567 568 /* 569 * if we have delayed truncation, do it now. 570 */ 571 nfs_delayedtruncate(vp); 572 573 /* 574 * First look in the cache. 575 */ 576 if (nfs_getattrcache(vp, ap->a_vap) == 0) 577 return (0); 578 nfsstats.rpccnt[NFSPROC_GETATTR]++; 579 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 580 nfsm_fhtom(np, v3); 581 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 582 if (!error) { 583 nfsm_loadattr(vp, ap->a_vap, 0); 584 if (vp->v_type == VDIR && 585 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 586 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 587 } 588 nfsm_reqdone; 589 return (error); 590 } 591 592 /* 593 * nfs setattr call. 594 */ 595 int 596 nfs_setattr(void *v) 597 { 598 struct vop_setattr_args /* { 599 struct vnodeop_desc *a_desc; 600 struct vnode *a_vp; 601 struct vattr *a_vap; 602 kauth_cred_t a_cred; 603 } */ *ap = v; 604 struct vnode *vp = ap->a_vp; 605 struct nfsnode *np = VTONFS(vp); 606 struct vattr *vap = ap->a_vap; 607 int error = 0; 608 u_quad_t tsize = 0; 609 610 /* 611 * Setting of flags is not supported. 612 */ 613 if (vap->va_flags != VNOVAL) 614 return (EOPNOTSUPP); 615 616 /* 617 * Disallow write attempts if the filesystem is mounted read-only. 618 */ 619 if ((vap->va_uid != (uid_t)VNOVAL || 620 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 621 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 622 (vp->v_mount->mnt_flag & MNT_RDONLY)) 623 return (EROFS); 624 if (vap->va_size != VNOVAL) { 625 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 626 return EFBIG; 627 } 628 switch (vp->v_type) { 629 case VDIR: 630 return (EISDIR); 631 case VCHR: 632 case VBLK: 633 case VSOCK: 634 case VFIFO: 635 if (vap->va_mtime.tv_sec == VNOVAL && 636 vap->va_atime.tv_sec == VNOVAL && 637 vap->va_mode == (mode_t)VNOVAL && 638 vap->va_uid == (uid_t)VNOVAL && 639 vap->va_gid == (gid_t)VNOVAL) 640 return (0); 641 vap->va_size = VNOVAL; 642 break; 643 default: 644 /* 645 * Disallow write attempts if the filesystem is 646 * mounted read-only. 647 */ 648 if (vp->v_mount->mnt_flag & MNT_RDONLY) 649 return (EROFS); 650 genfs_node_wrlock(vp); 651 uvm_vnp_setsize(vp, vap->va_size); 652 tsize = np->n_size; 653 np->n_size = vap->va_size; 654 if (vap->va_size == 0) 655 error = nfs_vinvalbuf(vp, 0, 656 ap->a_cred, curlwp, 1); 657 else 658 error = nfs_vinvalbuf(vp, V_SAVE, 659 ap->a_cred, curlwp, 1); 660 if (error) { 661 uvm_vnp_setsize(vp, tsize); 662 genfs_node_unlock(vp); 663 return (error); 664 } 665 np->n_vattr->va_size = vap->va_size; 666 } 667 } else { 668 /* 669 * flush files before setattr because a later write of 670 * cached data might change timestamps or reset sugid bits 671 */ 672 if ((vap->va_mtime.tv_sec != VNOVAL || 673 vap->va_atime.tv_sec != VNOVAL || 674 vap->va_mode != VNOVAL) && 675 vp->v_type == VREG && 676 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 677 curlwp, 1)) == EINTR) 678 return (error); 679 } 680 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 681 if (vap->va_size != VNOVAL) { 682 if (error) { 683 np->n_size = np->n_vattr->va_size = tsize; 684 uvm_vnp_setsize(vp, np->n_size); 685 } 686 genfs_node_unlock(vp); 687 } 688 VN_KNOTE(vp, NOTE_ATTRIB); 689 return (error); 690 } 691 692 /* 693 * Do an nfs setattr rpc. 694 */ 695 int 696 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 697 { 698 struct nfsv2_sattr *sp; 699 char *cp; 700 int32_t t1, t2; 701 char *bpos, *dpos; 702 u_int32_t *tl; 703 int error = 0; 704 struct mbuf *mreq, *mrep, *md, *mb; 705 const int v3 = NFS_ISV3(vp); 706 struct nfsnode *np = VTONFS(vp); 707 #ifndef NFS_V2_ONLY 708 int wccflag = NFSV3_WCCRATTR; 709 char *cp2; 710 #endif 711 712 nfsstats.rpccnt[NFSPROC_SETATTR]++; 713 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 714 nfsm_fhtom(np, v3); 715 #ifndef NFS_V2_ONLY 716 if (v3) { 717 nfsm_v3attrbuild(vap, true); 718 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 719 *tl = nfs_false; 720 } else { 721 #endif 722 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 723 if (vap->va_mode == (mode_t)VNOVAL) 724 sp->sa_mode = nfs_xdrneg1; 725 else 726 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 727 if (vap->va_uid == (uid_t)VNOVAL) 728 sp->sa_uid = nfs_xdrneg1; 729 else 730 sp->sa_uid = txdr_unsigned(vap->va_uid); 731 if (vap->va_gid == (gid_t)VNOVAL) 732 sp->sa_gid = nfs_xdrneg1; 733 else 734 sp->sa_gid = txdr_unsigned(vap->va_gid); 735 sp->sa_size = txdr_unsigned(vap->va_size); 736 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 737 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 738 #ifndef NFS_V2_ONLY 739 } 740 #endif 741 nfsm_request(np, NFSPROC_SETATTR, l, cred); 742 #ifndef NFS_V2_ONLY 743 if (v3) { 744 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 745 } else 746 #endif 747 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 748 nfsm_reqdone; 749 return (error); 750 } 751 752 /* 753 * nfs lookup call, one step at a time... 754 * First look in cache 755 * If not found, do the rpc. 756 */ 757 int 758 nfs_lookup(void *v) 759 { 760 struct vop_lookup_args /* { 761 struct vnodeop_desc *a_desc; 762 struct vnode *a_dvp; 763 struct vnode **a_vpp; 764 struct componentname *a_cnp; 765 } */ *ap = v; 766 struct componentname *cnp = ap->a_cnp; 767 struct vnode *dvp = ap->a_dvp; 768 struct vnode **vpp = ap->a_vpp; 769 int flags; 770 struct vnode *newvp; 771 u_int32_t *tl; 772 char *cp; 773 int32_t t1, t2; 774 char *bpos, *dpos, *cp2; 775 struct mbuf *mreq, *mrep, *md, *mb; 776 long len; 777 nfsfh_t *fhp; 778 struct nfsnode *np; 779 int error = 0, attrflag, fhsize; 780 const int v3 = NFS_ISV3(dvp); 781 782 flags = cnp->cn_flags; 783 784 *vpp = NULLVP; 785 newvp = NULLVP; 786 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 787 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 788 return (EROFS); 789 if (dvp->v_type != VDIR) 790 return (ENOTDIR); 791 792 /* 793 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 794 */ 795 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 796 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 797 if (error) 798 return error; 799 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 800 return EISDIR; 801 vref(dvp); 802 *vpp = dvp; 803 return 0; 804 } 805 806 np = VTONFS(dvp); 807 808 /* 809 * Before performing an RPC, check the name cache to see if 810 * the directory/name pair we are looking for is known already. 811 * If the directory/name pair is found in the name cache, 812 * we have to ensure the directory has not changed from 813 * the time the cache entry has been created. If it has, 814 * the cache entry has to be ignored. 815 */ 816 error = cache_lookup_raw(dvp, vpp, cnp); 817 KASSERT(dvp != *vpp); 818 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 819 if (error >= 0) { 820 struct vattr vattr; 821 int err2; 822 823 if (error && error != ENOENT) { 824 *vpp = NULLVP; 825 return error; 826 } 827 828 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 829 if (err2 != 0) { 830 if (error == 0) 831 vrele(*vpp); 832 *vpp = NULLVP; 833 return err2; 834 } 835 836 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 837 || timespeccmp(&vattr.va_mtime, 838 &VTONFS(dvp)->n_nctime, !=)) { 839 if (error == 0) { 840 vrele(*vpp); 841 *vpp = NULLVP; 842 } 843 cache_purge1(dvp, NULL, PURGE_CHILDREN); 844 timespecclear(&np->n_nctime); 845 goto dorpc; 846 } 847 848 if (error == ENOENT) { 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_flags & MAKEENTRY) && 999 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1000 nfs_cache_enter(dvp, newvp, cnp); 1001 } 1002 *vpp = newvp; 1003 nfsm_reqdone; 1004 if (error) { 1005 /* 1006 * We get here only because of errors returned by 1007 * the RPC. Otherwise we'll have returned above 1008 * (the nfsm_* macros will jump to nfsm_reqdone 1009 * on error). 1010 */ 1011 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1012 cnp->cn_nameiop != CREATE) { 1013 nfs_cache_enter(dvp, NULL, cnp); 1014 } 1015 if (newvp != NULLVP) { 1016 if (newvp == dvp) { 1017 vrele(newvp); 1018 } else { 1019 vput(newvp); 1020 } 1021 } 1022 noentry: 1023 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1024 (flags & ISLASTCN) && error == ENOENT) { 1025 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1026 error = EROFS; 1027 } else { 1028 error = EJUSTRETURN; 1029 } 1030 } 1031 *vpp = NULL; 1032 return error; 1033 } 1034 1035 validate: 1036 /* 1037 * make sure we have valid type and size. 1038 */ 1039 1040 newvp = *vpp; 1041 if (newvp->v_type == VNON) { 1042 struct vattr vattr; /* dummy */ 1043 1044 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1045 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1046 if (error) { 1047 vput(newvp); 1048 *vpp = NULL; 1049 } 1050 } 1051 1052 return error; 1053 } 1054 1055 /* 1056 * nfs read call. 1057 * Just call nfs_bioread() to do the work. 1058 */ 1059 int 1060 nfs_read(void *v) 1061 { 1062 struct vop_read_args /* { 1063 struct vnode *a_vp; 1064 struct uio *a_uio; 1065 int a_ioflag; 1066 kauth_cred_t a_cred; 1067 } */ *ap = v; 1068 struct vnode *vp = ap->a_vp; 1069 1070 if (vp->v_type != VREG) 1071 return EISDIR; 1072 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1073 } 1074 1075 /* 1076 * nfs readlink call 1077 */ 1078 int 1079 nfs_readlink(void *v) 1080 { 1081 struct vop_readlink_args /* { 1082 struct vnode *a_vp; 1083 struct uio *a_uio; 1084 kauth_cred_t a_cred; 1085 } */ *ap = v; 1086 struct vnode *vp = ap->a_vp; 1087 struct nfsnode *np = VTONFS(vp); 1088 1089 if (vp->v_type != VLNK) 1090 return (EPERM); 1091 1092 if (np->n_rcred != NULL) { 1093 kauth_cred_free(np->n_rcred); 1094 } 1095 np->n_rcred = ap->a_cred; 1096 kauth_cred_hold(np->n_rcred); 1097 1098 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1099 } 1100 1101 /* 1102 * Do a readlink rpc. 1103 * Called by nfs_doio() from below the buffer cache. 1104 */ 1105 int 1106 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1107 { 1108 u_int32_t *tl; 1109 char *cp; 1110 int32_t t1, t2; 1111 char *bpos, *dpos, *cp2; 1112 int error = 0; 1113 uint32_t len; 1114 struct mbuf *mreq, *mrep, *md, *mb; 1115 const int v3 = NFS_ISV3(vp); 1116 struct nfsnode *np = VTONFS(vp); 1117 #ifndef NFS_V2_ONLY 1118 int attrflag; 1119 #endif 1120 1121 nfsstats.rpccnt[NFSPROC_READLINK]++; 1122 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1123 nfsm_fhtom(np, v3); 1124 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1125 #ifndef NFS_V2_ONLY 1126 if (v3) 1127 nfsm_postop_attr(vp, attrflag, 0); 1128 #endif 1129 if (!error) { 1130 #ifndef NFS_V2_ONLY 1131 if (v3) { 1132 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1133 len = fxdr_unsigned(uint32_t, *tl); 1134 if (len > MAXPATHLEN) { 1135 /* 1136 * this pathname is too long for us. 1137 */ 1138 m_freem(mrep); 1139 /* Solaris returns EINVAL. should we follow? */ 1140 error = ENAMETOOLONG; 1141 goto nfsmout; 1142 } 1143 } else 1144 #endif 1145 { 1146 nfsm_strsiz(len, NFS_MAXPATHLEN); 1147 } 1148 nfsm_mtouio(uiop, len); 1149 } 1150 nfsm_reqdone; 1151 return (error); 1152 } 1153 1154 /* 1155 * nfs read rpc call 1156 * Ditto above 1157 */ 1158 int 1159 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1160 { 1161 u_int32_t *tl; 1162 char *cp; 1163 int32_t t1, t2; 1164 char *bpos, *dpos, *cp2; 1165 struct mbuf *mreq, *mrep, *md, *mb; 1166 struct nfsmount *nmp; 1167 int error = 0, len, retlen, tsiz, eof, byte_count; 1168 const int v3 = NFS_ISV3(vp); 1169 struct nfsnode *np = VTONFS(vp); 1170 #ifndef NFS_V2_ONLY 1171 int attrflag; 1172 #endif 1173 1174 #ifndef nolint 1175 eof = 0; 1176 #endif 1177 nmp = VFSTONFS(vp->v_mount); 1178 tsiz = uiop->uio_resid; 1179 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1180 return (EFBIG); 1181 iostat_busy(nmp->nm_stats); 1182 byte_count = 0; /* count bytes actually transferred */ 1183 while (tsiz > 0) { 1184 nfsstats.rpccnt[NFSPROC_READ]++; 1185 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1186 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1187 nfsm_fhtom(np, v3); 1188 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1189 #ifndef NFS_V2_ONLY 1190 if (v3) { 1191 txdr_hyper(uiop->uio_offset, tl); 1192 *(tl + 2) = txdr_unsigned(len); 1193 } else 1194 #endif 1195 { 1196 *tl++ = txdr_unsigned(uiop->uio_offset); 1197 *tl++ = txdr_unsigned(len); 1198 *tl = 0; 1199 } 1200 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1201 #ifndef NFS_V2_ONLY 1202 if (v3) { 1203 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1204 if (error) { 1205 m_freem(mrep); 1206 goto nfsmout; 1207 } 1208 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1209 eof = fxdr_unsigned(int, *(tl + 1)); 1210 } else 1211 #endif 1212 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1213 nfsm_strsiz(retlen, nmp->nm_rsize); 1214 nfsm_mtouio(uiop, retlen); 1215 m_freem(mrep); 1216 tsiz -= retlen; 1217 byte_count += retlen; 1218 #ifndef NFS_V2_ONLY 1219 if (v3) { 1220 if (eof || retlen == 0) 1221 tsiz = 0; 1222 } else 1223 #endif 1224 if (retlen < len) 1225 tsiz = 0; 1226 } 1227 nfsmout: 1228 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1229 return (error); 1230 } 1231 1232 struct nfs_writerpc_context { 1233 kmutex_t nwc_lock; 1234 kcondvar_t nwc_cv; 1235 int nwc_mbufcount; 1236 }; 1237 1238 /* 1239 * free mbuf used to refer protected pages while write rpc call. 1240 * called at splvm. 1241 */ 1242 static void 1243 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1244 { 1245 struct nfs_writerpc_context *ctx = arg; 1246 1247 KASSERT(m != NULL); 1248 KASSERT(ctx != NULL); 1249 pool_cache_put(mb_cache, m); 1250 mutex_enter(&ctx->nwc_lock); 1251 if (--ctx->nwc_mbufcount == 0) { 1252 cv_signal(&ctx->nwc_cv); 1253 } 1254 mutex_exit(&ctx->nwc_lock); 1255 } 1256 1257 /* 1258 * nfs write call 1259 */ 1260 int 1261 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1262 { 1263 u_int32_t *tl; 1264 char *cp; 1265 int32_t t1, t2; 1266 char *bpos, *dpos; 1267 struct mbuf *mreq, *mrep, *md, *mb; 1268 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1269 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1270 const int v3 = NFS_ISV3(vp); 1271 int committed = NFSV3WRITE_FILESYNC; 1272 struct nfsnode *np = VTONFS(vp); 1273 struct nfs_writerpc_context ctx; 1274 int byte_count; 1275 size_t origresid; 1276 #ifndef NFS_V2_ONLY 1277 char *cp2; 1278 int rlen, commit; 1279 #endif 1280 1281 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1282 panic("writerpc readonly vp %p", vp); 1283 } 1284 1285 #ifdef DIAGNOSTIC 1286 if (uiop->uio_iovcnt != 1) 1287 panic("nfs: writerpc iovcnt > 1"); 1288 #endif 1289 tsiz = uiop->uio_resid; 1290 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1291 return EFBIG; 1292 1293 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1294 cv_init(&ctx.nwc_cv, "nfsmblk"); 1295 ctx.nwc_mbufcount = 1; 1296 1297 retry: 1298 origresid = uiop->uio_resid; 1299 KASSERT(origresid == uiop->uio_iov->iov_len); 1300 iostat_busy(nmp->nm_stats); 1301 byte_count = 0; /* count of bytes actually written */ 1302 while (tsiz > 0) { 1303 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1304 uint32_t backup; 1305 bool stalewriteverf = false; 1306 1307 nfsstats.rpccnt[NFSPROC_WRITE]++; 1308 len = min(tsiz, nmp->nm_wsize); 1309 datalen = pageprotected ? 0 : nfsm_rndup(len); 1310 nfsm_reqhead(np, NFSPROC_WRITE, 1311 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1312 nfsm_fhtom(np, v3); 1313 #ifndef NFS_V2_ONLY 1314 if (v3) { 1315 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1316 txdr_hyper(uiop->uio_offset, tl); 1317 tl += 2; 1318 *tl++ = txdr_unsigned(len); 1319 *tl++ = txdr_unsigned(*iomode); 1320 *tl = txdr_unsigned(len); 1321 } else 1322 #endif 1323 { 1324 u_int32_t x; 1325 1326 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1327 /* Set both "begin" and "current" to non-garbage. */ 1328 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1329 *tl++ = x; /* "begin offset" */ 1330 *tl++ = x; /* "current offset" */ 1331 x = txdr_unsigned(len); 1332 *tl++ = x; /* total to this offset */ 1333 *tl = x; /* size of this write */ 1334 1335 } 1336 if (pageprotected) { 1337 /* 1338 * since we know pages can't be modified during i/o, 1339 * no need to copy them for us. 1340 */ 1341 struct mbuf *m; 1342 struct iovec *iovp = uiop->uio_iov; 1343 1344 m = m_get(M_WAIT, MT_DATA); 1345 MCLAIM(m, &nfs_mowner); 1346 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1347 nfs_writerpc_extfree, &ctx); 1348 m->m_flags |= M_EXT_ROMAP; 1349 m->m_len = len; 1350 mb->m_next = m; 1351 /* 1352 * no need to maintain mb and bpos here 1353 * because no one care them later. 1354 */ 1355 #if 0 1356 mb = m; 1357 bpos = mtod(void *, mb) + mb->m_len; 1358 #endif 1359 UIO_ADVANCE(uiop, len); 1360 uiop->uio_offset += len; 1361 mutex_enter(&ctx.nwc_lock); 1362 ctx.nwc_mbufcount++; 1363 mutex_exit(&ctx.nwc_lock); 1364 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1365 } else { 1366 nfsm_uiotom(uiop, len); 1367 } 1368 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1369 #ifndef NFS_V2_ONLY 1370 if (v3) { 1371 wccflag = NFSV3_WCCCHK; 1372 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1373 if (!error) { 1374 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1375 + NFSX_V3WRITEVERF); 1376 rlen = fxdr_unsigned(int, *tl++); 1377 if (rlen == 0) { 1378 error = NFSERR_IO; 1379 m_freem(mrep); 1380 break; 1381 } else if (rlen < len) { 1382 backup = len - rlen; 1383 UIO_ADVANCE(uiop, -backup); 1384 uiop->uio_offset -= backup; 1385 len = rlen; 1386 } 1387 commit = fxdr_unsigned(int, *tl++); 1388 1389 /* 1390 * Return the lowest committment level 1391 * obtained by any of the RPCs. 1392 */ 1393 if (committed == NFSV3WRITE_FILESYNC) 1394 committed = commit; 1395 else if (committed == NFSV3WRITE_DATASYNC && 1396 commit == NFSV3WRITE_UNSTABLE) 1397 committed = commit; 1398 mutex_enter(&nmp->nm_lock); 1399 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1400 memcpy(nmp->nm_writeverf, tl, 1401 NFSX_V3WRITEVERF); 1402 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1403 } else if ((nmp->nm_iflag & 1404 NFSMNT_STALEWRITEVERF) || 1405 memcmp(tl, nmp->nm_writeverf, 1406 NFSX_V3WRITEVERF)) { 1407 memcpy(nmp->nm_writeverf, tl, 1408 NFSX_V3WRITEVERF); 1409 /* 1410 * note NFSMNT_STALEWRITEVERF 1411 * if we're the first thread to 1412 * notice it. 1413 */ 1414 if ((nmp->nm_iflag & 1415 NFSMNT_STALEWRITEVERF) == 0) { 1416 stalewriteverf = true; 1417 nmp->nm_iflag |= 1418 NFSMNT_STALEWRITEVERF; 1419 } 1420 } 1421 mutex_exit(&nmp->nm_lock); 1422 } 1423 } else 1424 #endif 1425 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1426 if (wccflag) 1427 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1428 m_freem(mrep); 1429 if (error) 1430 break; 1431 tsiz -= len; 1432 byte_count += len; 1433 if (stalewriteverf) { 1434 *stalewriteverfp = true; 1435 stalewriteverf = false; 1436 if (committed == NFSV3WRITE_UNSTABLE && 1437 len != origresid) { 1438 /* 1439 * if our write requests weren't atomic but 1440 * unstable, datas in previous iterations 1441 * might have already been lost now. 1442 * then, we should resend them to nfsd. 1443 */ 1444 backup = origresid - tsiz; 1445 UIO_ADVANCE(uiop, -backup); 1446 uiop->uio_offset -= backup; 1447 tsiz = origresid; 1448 goto retry; 1449 } 1450 } 1451 } 1452 nfsmout: 1453 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1454 if (pageprotected) { 1455 /* 1456 * wait until mbufs go away. 1457 * retransmitted mbufs can survive longer than rpc requests 1458 * themselves. 1459 */ 1460 mutex_enter(&ctx.nwc_lock); 1461 ctx.nwc_mbufcount--; 1462 while (ctx.nwc_mbufcount > 0) { 1463 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1464 } 1465 mutex_exit(&ctx.nwc_lock); 1466 } 1467 mutex_destroy(&ctx.nwc_lock); 1468 cv_destroy(&ctx.nwc_cv); 1469 *iomode = committed; 1470 if (error) 1471 uiop->uio_resid = tsiz; 1472 return (error); 1473 } 1474 1475 /* 1476 * nfs mknod rpc 1477 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1478 * mode set to specify the file type and the size field for rdev. 1479 */ 1480 int 1481 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1482 { 1483 struct nfsv2_sattr *sp; 1484 u_int32_t *tl; 1485 char *cp; 1486 int32_t t1, t2; 1487 struct vnode *newvp = (struct vnode *)0; 1488 struct nfsnode *dnp, *np; 1489 char *cp2; 1490 char *bpos, *dpos; 1491 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1492 struct mbuf *mreq, *mrep, *md, *mb; 1493 u_int32_t rdev; 1494 const int v3 = NFS_ISV3(dvp); 1495 1496 if (vap->va_type == VCHR || vap->va_type == VBLK) 1497 rdev = txdr_unsigned(vap->va_rdev); 1498 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1499 rdev = nfs_xdrneg1; 1500 else { 1501 VOP_ABORTOP(dvp, cnp); 1502 vput(dvp); 1503 return (EOPNOTSUPP); 1504 } 1505 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1506 dnp = VTONFS(dvp); 1507 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1508 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1509 nfsm_fhtom(dnp, v3); 1510 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1511 #ifndef NFS_V2_ONLY 1512 if (v3) { 1513 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1514 *tl++ = vtonfsv3_type(vap->va_type); 1515 nfsm_v3attrbuild(vap, false); 1516 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1517 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1518 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1519 *tl = txdr_unsigned(minor(vap->va_rdev)); 1520 } 1521 } else 1522 #endif 1523 { 1524 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1525 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1526 sp->sa_uid = nfs_xdrneg1; 1527 sp->sa_gid = nfs_xdrneg1; 1528 sp->sa_size = rdev; 1529 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1530 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1531 } 1532 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1533 if (!error) { 1534 nfsm_mtofh(dvp, newvp, v3, gotvp); 1535 if (!gotvp) { 1536 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1537 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1538 if (!error) 1539 newvp = NFSTOV(np); 1540 } 1541 } 1542 #ifndef NFS_V2_ONLY 1543 if (v3) 1544 nfsm_wcc_data(dvp, wccflag, 0, !error); 1545 #endif 1546 nfsm_reqdone; 1547 if (error) { 1548 if (newvp) 1549 vput(newvp); 1550 } else { 1551 if (cnp->cn_flags & MAKEENTRY) 1552 nfs_cache_enter(dvp, newvp, cnp); 1553 *vpp = newvp; 1554 } 1555 VTONFS(dvp)->n_flag |= NMODIFIED; 1556 if (!wccflag) 1557 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1558 vput(dvp); 1559 return (error); 1560 } 1561 1562 /* 1563 * nfs mknod vop 1564 * just call nfs_mknodrpc() to do the work. 1565 */ 1566 /* ARGSUSED */ 1567 int 1568 nfs_mknod(void *v) 1569 { 1570 struct vop_mknod_args /* { 1571 struct vnode *a_dvp; 1572 struct vnode **a_vpp; 1573 struct componentname *a_cnp; 1574 struct vattr *a_vap; 1575 } */ *ap = v; 1576 struct vnode *dvp = ap->a_dvp; 1577 struct componentname *cnp = ap->a_cnp; 1578 int error; 1579 1580 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1581 VN_KNOTE(dvp, NOTE_WRITE); 1582 if (error == 0 || error == EEXIST) 1583 cache_purge1(dvp, cnp, 0); 1584 return (error); 1585 } 1586 1587 /* 1588 * nfs file create call 1589 */ 1590 int 1591 nfs_create(void *v) 1592 { 1593 struct vop_create_args /* { 1594 struct vnode *a_dvp; 1595 struct vnode **a_vpp; 1596 struct componentname *a_cnp; 1597 struct vattr *a_vap; 1598 } */ *ap = v; 1599 struct vnode *dvp = ap->a_dvp; 1600 struct vattr *vap = ap->a_vap; 1601 struct componentname *cnp = ap->a_cnp; 1602 struct nfsv2_sattr *sp; 1603 u_int32_t *tl; 1604 char *cp; 1605 int32_t t1, t2; 1606 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1607 struct vnode *newvp = (struct vnode *)0; 1608 char *bpos, *dpos, *cp2; 1609 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1610 struct mbuf *mreq, *mrep, *md, *mb; 1611 const int v3 = NFS_ISV3(dvp); 1612 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1613 1614 /* 1615 * Oops, not for me.. 1616 */ 1617 if (vap->va_type == VSOCK) 1618 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1619 1620 KASSERT(vap->va_type == VREG); 1621 1622 #ifdef VA_EXCLUSIVE 1623 if (vap->va_vaflags & VA_EXCLUSIVE) { 1624 excl_mode = NFSV3CREATE_EXCLUSIVE; 1625 } 1626 #endif 1627 again: 1628 error = 0; 1629 nfsstats.rpccnt[NFSPROC_CREATE]++; 1630 dnp = VTONFS(dvp); 1631 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1632 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1633 nfsm_fhtom(dnp, v3); 1634 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1635 #ifndef NFS_V2_ONLY 1636 if (v3) { 1637 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1638 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1639 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1640 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1641 *tl++ = arc4random(); 1642 *tl = arc4random(); 1643 } else { 1644 *tl = txdr_unsigned(excl_mode); 1645 nfsm_v3attrbuild(vap, false); 1646 } 1647 } else 1648 #endif 1649 { 1650 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1651 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1652 sp->sa_uid = nfs_xdrneg1; 1653 sp->sa_gid = nfs_xdrneg1; 1654 sp->sa_size = 0; 1655 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1656 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1657 } 1658 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1659 if (!error) { 1660 nfsm_mtofh(dvp, newvp, v3, gotvp); 1661 if (!gotvp) { 1662 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1663 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1664 if (!error) 1665 newvp = NFSTOV(np); 1666 } 1667 } 1668 #ifndef NFS_V2_ONLY 1669 if (v3) 1670 nfsm_wcc_data(dvp, wccflag, 0, !error); 1671 #endif 1672 nfsm_reqdone; 1673 if (error) { 1674 /* 1675 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1676 */ 1677 if (v3 && error == ENOTSUP) { 1678 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1679 excl_mode = NFSV3CREATE_GUARDED; 1680 goto again; 1681 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1682 excl_mode = NFSV3CREATE_UNCHECKED; 1683 goto again; 1684 } 1685 } 1686 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1687 struct timespec ts; 1688 1689 getnanotime(&ts); 1690 1691 /* 1692 * make sure that we'll update timestamps as 1693 * most server implementations use them to store 1694 * the create verifier. 1695 * 1696 * XXX it's better to use TOSERVER always. 1697 */ 1698 1699 if (vap->va_atime.tv_sec == VNOVAL) 1700 vap->va_atime = ts; 1701 if (vap->va_mtime.tv_sec == VNOVAL) 1702 vap->va_mtime = ts; 1703 1704 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1705 } 1706 if (error == 0) { 1707 if (cnp->cn_flags & MAKEENTRY) 1708 nfs_cache_enter(dvp, newvp, cnp); 1709 else 1710 cache_purge1(dvp, cnp, 0); 1711 *ap->a_vpp = newvp; 1712 } else { 1713 if (newvp) 1714 vput(newvp); 1715 if (error == EEXIST) 1716 cache_purge1(dvp, cnp, 0); 1717 } 1718 VTONFS(dvp)->n_flag |= NMODIFIED; 1719 if (!wccflag) 1720 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1721 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1722 vput(dvp); 1723 return (error); 1724 } 1725 1726 /* 1727 * nfs file remove call 1728 * To try and make nfs semantics closer to ufs semantics, a file that has 1729 * other processes using the vnode is renamed instead of removed and then 1730 * removed later on the last close. 1731 * - If v_usecount > 1 1732 * If a rename is not already in the works 1733 * call nfs_sillyrename() to set it up 1734 * else 1735 * do the remove rpc 1736 */ 1737 int 1738 nfs_remove(void *v) 1739 { 1740 struct vop_remove_args /* { 1741 struct vnodeop_desc *a_desc; 1742 struct vnode * a_dvp; 1743 struct vnode * a_vp; 1744 struct componentname * a_cnp; 1745 } */ *ap = v; 1746 struct vnode *vp = ap->a_vp; 1747 struct vnode *dvp = ap->a_dvp; 1748 struct componentname *cnp = ap->a_cnp; 1749 struct nfsnode *np = VTONFS(vp); 1750 int error = 0; 1751 struct vattr vattr; 1752 1753 #ifndef DIAGNOSTIC 1754 if (vp->v_usecount < 1) 1755 panic("nfs_remove: bad v_usecount"); 1756 #endif 1757 if (vp->v_type == VDIR) 1758 error = EPERM; 1759 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1760 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1761 vattr.va_nlink > 1)) { 1762 /* 1763 * Purge the name cache so that the chance of a lookup for 1764 * the name succeeding while the remove is in progress is 1765 * minimized. Without node locking it can still happen, such 1766 * that an I/O op returns ESTALE, but since you get this if 1767 * another host removes the file.. 1768 */ 1769 cache_purge(vp); 1770 /* 1771 * throw away biocache buffers, mainly to avoid 1772 * unnecessary delayed writes later. 1773 */ 1774 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1775 /* Do the rpc */ 1776 if (error != EINTR) 1777 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1778 cnp->cn_namelen, cnp->cn_cred, curlwp); 1779 } else if (!np->n_sillyrename) 1780 error = nfs_sillyrename(dvp, vp, cnp, false); 1781 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1782 vattr.va_nlink == 1) { 1783 np->n_flag |= NREMOVED; 1784 } 1785 NFS_INVALIDATE_ATTRCACHE(np); 1786 VN_KNOTE(vp, NOTE_DELETE); 1787 VN_KNOTE(dvp, NOTE_WRITE); 1788 if (dvp == vp) 1789 vrele(vp); 1790 else 1791 vput(vp); 1792 vput(dvp); 1793 return (error); 1794 } 1795 1796 /* 1797 * nfs file remove rpc called from nfs_inactive 1798 */ 1799 int 1800 nfs_removeit(struct sillyrename *sp) 1801 { 1802 1803 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1804 (struct lwp *)0)); 1805 } 1806 1807 /* 1808 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1809 */ 1810 int 1811 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1812 { 1813 u_int32_t *tl; 1814 char *cp; 1815 #ifndef NFS_V2_ONLY 1816 int32_t t1; 1817 char *cp2; 1818 #endif 1819 int32_t t2; 1820 char *bpos, *dpos; 1821 int error = 0, wccflag = NFSV3_WCCRATTR; 1822 struct mbuf *mreq, *mrep, *md, *mb; 1823 const int v3 = NFS_ISV3(dvp); 1824 int rexmit = 0; 1825 struct nfsnode *dnp = VTONFS(dvp); 1826 1827 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1828 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1829 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1830 nfsm_fhtom(dnp, v3); 1831 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1832 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1833 #ifndef NFS_V2_ONLY 1834 if (v3) 1835 nfsm_wcc_data(dvp, wccflag, 0, !error); 1836 #endif 1837 nfsm_reqdone; 1838 VTONFS(dvp)->n_flag |= NMODIFIED; 1839 if (!wccflag) 1840 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1841 /* 1842 * Kludge City: If the first reply to the remove rpc is lost.. 1843 * the reply to the retransmitted request will be ENOENT 1844 * since the file was in fact removed 1845 * Therefore, we cheat and return success. 1846 */ 1847 if (rexmit && error == ENOENT) 1848 error = 0; 1849 return (error); 1850 } 1851 1852 /* 1853 * nfs file rename call 1854 */ 1855 int 1856 nfs_rename(void *v) 1857 { 1858 struct vop_rename_args /* { 1859 struct vnode *a_fdvp; 1860 struct vnode *a_fvp; 1861 struct componentname *a_fcnp; 1862 struct vnode *a_tdvp; 1863 struct vnode *a_tvp; 1864 struct componentname *a_tcnp; 1865 } */ *ap = v; 1866 struct vnode *fvp = ap->a_fvp; 1867 struct vnode *tvp = ap->a_tvp; 1868 struct vnode *fdvp = ap->a_fdvp; 1869 struct vnode *tdvp = ap->a_tdvp; 1870 struct componentname *tcnp = ap->a_tcnp; 1871 struct componentname *fcnp = ap->a_fcnp; 1872 int error; 1873 1874 /* Check for cross-device rename */ 1875 if ((fvp->v_mount != tdvp->v_mount) || 1876 (tvp && (fvp->v_mount != tvp->v_mount))) { 1877 error = EXDEV; 1878 goto out; 1879 } 1880 1881 /* 1882 * If the tvp exists and is in use, sillyrename it before doing the 1883 * rename of the new file over it. 1884 * 1885 * Have sillyrename use link instead of rename if possible, 1886 * so that we don't lose the file if the rename fails, and so 1887 * that there's no window when the "to" file doesn't exist. 1888 */ 1889 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1890 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1891 VN_KNOTE(tvp, NOTE_DELETE); 1892 vput(tvp); 1893 tvp = NULL; 1894 } 1895 1896 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1897 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1898 curlwp); 1899 1900 VN_KNOTE(fdvp, NOTE_WRITE); 1901 VN_KNOTE(tdvp, NOTE_WRITE); 1902 if (error == 0 || error == EEXIST) { 1903 if (fvp->v_type == VDIR) 1904 cache_purge(fvp); 1905 else 1906 cache_purge1(fdvp, fcnp, 0); 1907 if (tvp != NULL && tvp->v_type == VDIR) 1908 cache_purge(tvp); 1909 else 1910 cache_purge1(tdvp, tcnp, 0); 1911 } 1912 out: 1913 if (tdvp == tvp) 1914 vrele(tdvp); 1915 else 1916 vput(tdvp); 1917 if (tvp) 1918 vput(tvp); 1919 vrele(fdvp); 1920 vrele(fvp); 1921 return (error); 1922 } 1923 1924 /* 1925 * nfs file rename rpc called from nfs_remove() above 1926 */ 1927 int 1928 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1929 { 1930 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1931 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1932 } 1933 1934 /* 1935 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1936 */ 1937 int 1938 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) 1939 { 1940 u_int32_t *tl; 1941 char *cp; 1942 #ifndef NFS_V2_ONLY 1943 int32_t t1; 1944 char *cp2; 1945 #endif 1946 int32_t t2; 1947 char *bpos, *dpos; 1948 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1949 struct mbuf *mreq, *mrep, *md, *mb; 1950 const int v3 = NFS_ISV3(fdvp); 1951 int rexmit = 0; 1952 struct nfsnode *fdnp = VTONFS(fdvp); 1953 1954 nfsstats.rpccnt[NFSPROC_RENAME]++; 1955 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1956 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1957 nfsm_rndup(tnamelen)); 1958 nfsm_fhtom(fdnp, v3); 1959 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1960 nfsm_fhtom(VTONFS(tdvp), v3); 1961 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1962 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1963 #ifndef NFS_V2_ONLY 1964 if (v3) { 1965 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1966 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1967 } 1968 #endif 1969 nfsm_reqdone; 1970 VTONFS(fdvp)->n_flag |= NMODIFIED; 1971 VTONFS(tdvp)->n_flag |= NMODIFIED; 1972 if (!fwccflag) 1973 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1974 if (!twccflag) 1975 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1976 /* 1977 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1978 */ 1979 if (rexmit && error == ENOENT) 1980 error = 0; 1981 return (error); 1982 } 1983 1984 /* 1985 * NFS link RPC, called from nfs_link. 1986 * Assumes dvp and vp locked, and leaves them that way. 1987 */ 1988 1989 static int 1990 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 1991 size_t namelen, kauth_cred_t cred, struct lwp *l) 1992 { 1993 u_int32_t *tl; 1994 char *cp; 1995 #ifndef NFS_V2_ONLY 1996 int32_t t1; 1997 char *cp2; 1998 #endif 1999 int32_t t2; 2000 char *bpos, *dpos; 2001 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2002 struct mbuf *mreq, *mrep, *md, *mb; 2003 const int v3 = NFS_ISV3(dvp); 2004 int rexmit = 0; 2005 struct nfsnode *np = VTONFS(vp); 2006 2007 nfsstats.rpccnt[NFSPROC_LINK]++; 2008 nfsm_reqhead(np, NFSPROC_LINK, 2009 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2010 nfsm_fhtom(np, v3); 2011 nfsm_fhtom(VTONFS(dvp), v3); 2012 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2013 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2014 #ifndef NFS_V2_ONLY 2015 if (v3) { 2016 nfsm_postop_attr(vp, attrflag, 0); 2017 nfsm_wcc_data(dvp, wccflag, 0, !error); 2018 } 2019 #endif 2020 nfsm_reqdone; 2021 2022 VTONFS(dvp)->n_flag |= NMODIFIED; 2023 if (!attrflag) 2024 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2025 if (!wccflag) 2026 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2027 2028 /* 2029 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2030 */ 2031 if (rexmit && error == EEXIST) 2032 error = 0; 2033 2034 return error; 2035 } 2036 2037 /* 2038 * nfs hard link create call 2039 */ 2040 int 2041 nfs_link(void *v) 2042 { 2043 struct vop_link_args /* { 2044 struct vnode *a_dvp; 2045 struct vnode *a_vp; 2046 struct componentname *a_cnp; 2047 } */ *ap = v; 2048 struct vnode *vp = ap->a_vp; 2049 struct vnode *dvp = ap->a_dvp; 2050 struct componentname *cnp = ap->a_cnp; 2051 int error = 0; 2052 2053 error = vn_lock(vp, LK_EXCLUSIVE); 2054 if (error != 0) { 2055 VOP_ABORTOP(dvp, cnp); 2056 vput(dvp); 2057 return error; 2058 } 2059 2060 /* 2061 * Push all writes to the server, so that the attribute cache 2062 * doesn't get "out of sync" with the server. 2063 * XXX There should be a better way! 2064 */ 2065 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2066 2067 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2068 cnp->cn_cred, curlwp); 2069 2070 if (error == 0) { 2071 cache_purge1(dvp, cnp, 0); 2072 } 2073 VOP_UNLOCK(vp); 2074 VN_KNOTE(vp, NOTE_LINK); 2075 VN_KNOTE(dvp, NOTE_WRITE); 2076 vput(dvp); 2077 return (error); 2078 } 2079 2080 /* 2081 * nfs symbolic link create call 2082 */ 2083 int 2084 nfs_symlink(void *v) 2085 { 2086 struct vop_symlink_args /* { 2087 struct vnode *a_dvp; 2088 struct vnode **a_vpp; 2089 struct componentname *a_cnp; 2090 struct vattr *a_vap; 2091 char *a_target; 2092 } */ *ap = v; 2093 struct vnode *dvp = ap->a_dvp; 2094 struct vattr *vap = ap->a_vap; 2095 struct componentname *cnp = ap->a_cnp; 2096 struct nfsv2_sattr *sp; 2097 u_int32_t *tl; 2098 char *cp; 2099 int32_t t1, t2; 2100 char *bpos, *dpos, *cp2; 2101 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2102 struct mbuf *mreq, *mrep, *md, *mb; 2103 struct vnode *newvp = (struct vnode *)0; 2104 const int v3 = NFS_ISV3(dvp); 2105 int rexmit = 0; 2106 struct nfsnode *dnp = VTONFS(dvp); 2107 2108 *ap->a_vpp = NULL; 2109 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2110 slen = strlen(ap->a_target); 2111 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2112 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2113 nfsm_fhtom(dnp, v3); 2114 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2115 #ifndef NFS_V2_ONlY 2116 if (v3) 2117 nfsm_v3attrbuild(vap, false); 2118 #endif 2119 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2120 #ifndef NFS_V2_ONlY 2121 if (!v3) { 2122 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2123 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2124 sp->sa_uid = nfs_xdrneg1; 2125 sp->sa_gid = nfs_xdrneg1; 2126 sp->sa_size = nfs_xdrneg1; 2127 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2128 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2129 } 2130 #endif 2131 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2132 &rexmit); 2133 #ifndef NFS_V2_ONlY 2134 if (v3) { 2135 if (!error) 2136 nfsm_mtofh(dvp, newvp, v3, gotvp); 2137 nfsm_wcc_data(dvp, wccflag, 0, !error); 2138 } 2139 #endif 2140 nfsm_reqdone; 2141 /* 2142 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2143 */ 2144 if (rexmit && error == EEXIST) 2145 error = 0; 2146 if (error == 0 || error == EEXIST) 2147 cache_purge1(dvp, cnp, 0); 2148 if (error == 0 && newvp == NULL) { 2149 struct nfsnode *np = NULL; 2150 2151 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2152 cnp->cn_cred, curlwp, &np); 2153 if (error == 0) 2154 newvp = NFSTOV(np); 2155 } 2156 if (error) { 2157 if (newvp != NULL) 2158 vput(newvp); 2159 } else { 2160 *ap->a_vpp = newvp; 2161 } 2162 VTONFS(dvp)->n_flag |= NMODIFIED; 2163 if (!wccflag) 2164 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2165 VN_KNOTE(dvp, NOTE_WRITE); 2166 vput(dvp); 2167 return (error); 2168 } 2169 2170 /* 2171 * nfs make dir call 2172 */ 2173 int 2174 nfs_mkdir(void *v) 2175 { 2176 struct vop_mkdir_args /* { 2177 struct vnode *a_dvp; 2178 struct vnode **a_vpp; 2179 struct componentname *a_cnp; 2180 struct vattr *a_vap; 2181 } */ *ap = v; 2182 struct vnode *dvp = ap->a_dvp; 2183 struct vattr *vap = ap->a_vap; 2184 struct componentname *cnp = ap->a_cnp; 2185 struct nfsv2_sattr *sp; 2186 u_int32_t *tl; 2187 char *cp; 2188 int32_t t1, t2; 2189 int len; 2190 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2191 struct vnode *newvp = (struct vnode *)0; 2192 char *bpos, *dpos, *cp2; 2193 int error = 0, wccflag = NFSV3_WCCRATTR; 2194 int gotvp = 0; 2195 int rexmit = 0; 2196 struct mbuf *mreq, *mrep, *md, *mb; 2197 const int v3 = NFS_ISV3(dvp); 2198 2199 len = cnp->cn_namelen; 2200 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2201 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2202 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2203 nfsm_fhtom(dnp, v3); 2204 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2205 #ifndef NFS_V2_ONLY 2206 if (v3) { 2207 nfsm_v3attrbuild(vap, false); 2208 } else 2209 #endif 2210 { 2211 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2212 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2213 sp->sa_uid = nfs_xdrneg1; 2214 sp->sa_gid = nfs_xdrneg1; 2215 sp->sa_size = nfs_xdrneg1; 2216 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2217 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2218 } 2219 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2220 if (!error) 2221 nfsm_mtofh(dvp, newvp, v3, gotvp); 2222 if (v3) 2223 nfsm_wcc_data(dvp, wccflag, 0, !error); 2224 nfsm_reqdone; 2225 VTONFS(dvp)->n_flag |= NMODIFIED; 2226 if (!wccflag) 2227 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2228 /* 2229 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2230 * if we can succeed in looking up the directory. 2231 */ 2232 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2233 if (newvp) { 2234 vput(newvp); 2235 newvp = (struct vnode *)0; 2236 } 2237 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2238 curlwp, &np); 2239 if (!error) { 2240 newvp = NFSTOV(np); 2241 if (newvp->v_type != VDIR || newvp == dvp) 2242 error = EEXIST; 2243 } 2244 } 2245 if (error) { 2246 if (newvp) { 2247 if (dvp != newvp) 2248 vput(newvp); 2249 else 2250 vrele(newvp); 2251 } 2252 } else { 2253 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2254 if (cnp->cn_flags & MAKEENTRY) 2255 nfs_cache_enter(dvp, newvp, cnp); 2256 *ap->a_vpp = newvp; 2257 } 2258 vput(dvp); 2259 return (error); 2260 } 2261 2262 /* 2263 * nfs remove directory call 2264 */ 2265 int 2266 nfs_rmdir(void *v) 2267 { 2268 struct vop_rmdir_args /* { 2269 struct vnode *a_dvp; 2270 struct vnode *a_vp; 2271 struct componentname *a_cnp; 2272 } */ *ap = v; 2273 struct vnode *vp = ap->a_vp; 2274 struct vnode *dvp = ap->a_dvp; 2275 struct componentname *cnp = ap->a_cnp; 2276 u_int32_t *tl; 2277 char *cp; 2278 #ifndef NFS_V2_ONLY 2279 int32_t t1; 2280 char *cp2; 2281 #endif 2282 int32_t t2; 2283 char *bpos, *dpos; 2284 int error = 0, wccflag = NFSV3_WCCRATTR; 2285 int rexmit = 0; 2286 struct mbuf *mreq, *mrep, *md, *mb; 2287 const int v3 = NFS_ISV3(dvp); 2288 struct nfsnode *dnp; 2289 2290 if (dvp == vp) { 2291 vrele(dvp); 2292 vput(dvp); 2293 return (EINVAL); 2294 } 2295 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2296 dnp = VTONFS(dvp); 2297 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2298 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2299 nfsm_fhtom(dnp, v3); 2300 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2301 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2302 #ifndef NFS_V2_ONLY 2303 if (v3) 2304 nfsm_wcc_data(dvp, wccflag, 0, !error); 2305 #endif 2306 nfsm_reqdone; 2307 VTONFS(dvp)->n_flag |= NMODIFIED; 2308 if (!wccflag) 2309 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2310 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2311 VN_KNOTE(vp, NOTE_DELETE); 2312 cache_purge(vp); 2313 vput(vp); 2314 vput(dvp); 2315 /* 2316 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2317 */ 2318 if (rexmit && error == ENOENT) 2319 error = 0; 2320 return (error); 2321 } 2322 2323 /* 2324 * nfs readdir call 2325 */ 2326 int 2327 nfs_readdir(void *v) 2328 { 2329 struct vop_readdir_args /* { 2330 struct vnode *a_vp; 2331 struct uio *a_uio; 2332 kauth_cred_t a_cred; 2333 int *a_eofflag; 2334 off_t **a_cookies; 2335 int *a_ncookies; 2336 } */ *ap = v; 2337 struct vnode *vp = ap->a_vp; 2338 struct uio *uio = ap->a_uio; 2339 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2340 char *base = uio->uio_iov->iov_base; 2341 int tresid, error; 2342 size_t count, lost; 2343 struct dirent *dp; 2344 off_t *cookies = NULL; 2345 int ncookies = 0, nc; 2346 2347 if (vp->v_type != VDIR) 2348 return (EPERM); 2349 2350 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2351 count = uio->uio_resid - lost; 2352 if (count <= 0) 2353 return (EINVAL); 2354 2355 /* 2356 * Call nfs_bioread() to do the real work. 2357 */ 2358 tresid = uio->uio_resid = count; 2359 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2360 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2361 2362 if (!error && ap->a_cookies) { 2363 ncookies = count / 16; 2364 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2365 *ap->a_cookies = cookies; 2366 } 2367 2368 if (!error && uio->uio_resid == tresid) { 2369 uio->uio_resid += lost; 2370 nfsstats.direofcache_misses++; 2371 if (ap->a_cookies) 2372 *ap->a_ncookies = 0; 2373 *ap->a_eofflag = 1; 2374 return (0); 2375 } 2376 2377 if (!error && ap->a_cookies) { 2378 /* 2379 * Only the NFS server and emulations use cookies, and they 2380 * load the directory block into system space, so we can 2381 * just look at it directly. 2382 */ 2383 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2384 uio->uio_iovcnt != 1) 2385 panic("nfs_readdir: lost in space"); 2386 for (nc = 0; ncookies-- && 2387 base < (char *)uio->uio_iov->iov_base; nc++){ 2388 dp = (struct dirent *) base; 2389 if (dp->d_reclen == 0) 2390 break; 2391 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2392 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2393 else 2394 *(cookies++) = NFS_GETCOOKIE(dp); 2395 base += dp->d_reclen; 2396 } 2397 uio->uio_resid += 2398 ((char *)uio->uio_iov->iov_base - base); 2399 uio->uio_iov->iov_len += 2400 ((char *)uio->uio_iov->iov_base - base); 2401 uio->uio_iov->iov_base = base; 2402 *ap->a_ncookies = nc; 2403 } 2404 2405 uio->uio_resid += lost; 2406 *ap->a_eofflag = 0; 2407 return (error); 2408 } 2409 2410 /* 2411 * Readdir rpc call. 2412 * Called from below the buffer cache by nfs_doio(). 2413 */ 2414 int 2415 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2416 { 2417 int len, left; 2418 struct dirent *dp = NULL; 2419 u_int32_t *tl; 2420 char *cp; 2421 int32_t t1, t2; 2422 char *bpos, *dpos, *cp2; 2423 struct mbuf *mreq, *mrep, *md, *mb; 2424 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2425 struct nfsnode *dnp = VTONFS(vp); 2426 u_quad_t fileno; 2427 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2428 #ifndef NFS_V2_ONLY 2429 int attrflag; 2430 #endif 2431 int nrpcs = 0, reclen; 2432 const int v3 = NFS_ISV3(vp); 2433 2434 #ifdef DIAGNOSTIC 2435 /* 2436 * Should be called from buffer cache, so only amount of 2437 * NFS_DIRBLKSIZ will be requested. 2438 */ 2439 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2440 panic("nfs readdirrpc bad uio"); 2441 #endif 2442 2443 /* 2444 * Loop around doing readdir rpc's of size nm_readdirsize 2445 * truncated to a multiple of NFS_DIRFRAGSIZ. 2446 * The stopping criteria is EOF or buffer full. 2447 */ 2448 while (more_dirs && bigenough) { 2449 /* 2450 * Heuristic: don't bother to do another RPC to further 2451 * fill up this block if there is not much room left. (< 50% 2452 * of the readdir RPC size). This wastes some buffer space 2453 * but can save up to 50% in RPC calls. 2454 */ 2455 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2456 bigenough = 0; 2457 break; 2458 } 2459 nfsstats.rpccnt[NFSPROC_READDIR]++; 2460 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2461 NFSX_READDIR(v3)); 2462 nfsm_fhtom(dnp, v3); 2463 #ifndef NFS_V2_ONLY 2464 if (v3) { 2465 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2466 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2467 txdr_swapcookie3(uiop->uio_offset, tl); 2468 } else { 2469 txdr_cookie3(uiop->uio_offset, tl); 2470 } 2471 tl += 2; 2472 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2473 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2474 } else 2475 #endif 2476 { 2477 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2478 *tl++ = txdr_unsigned(uiop->uio_offset); 2479 } 2480 *tl = txdr_unsigned(nmp->nm_readdirsize); 2481 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2482 nrpcs++; 2483 #ifndef NFS_V2_ONLY 2484 if (v3) { 2485 nfsm_postop_attr(vp, attrflag, 0); 2486 if (!error) { 2487 nfsm_dissect(tl, u_int32_t *, 2488 2 * NFSX_UNSIGNED); 2489 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2490 dnp->n_cookieverf.nfsuquad[1] = *tl; 2491 } else { 2492 m_freem(mrep); 2493 goto nfsmout; 2494 } 2495 } 2496 #endif 2497 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2498 more_dirs = fxdr_unsigned(int, *tl); 2499 2500 /* loop thru the dir entries, doctoring them to 4bsd form */ 2501 while (more_dirs && bigenough) { 2502 #ifndef NFS_V2_ONLY 2503 if (v3) { 2504 nfsm_dissect(tl, u_int32_t *, 2505 3 * NFSX_UNSIGNED); 2506 fileno = fxdr_hyper(tl); 2507 len = fxdr_unsigned(int, *(tl + 2)); 2508 } else 2509 #endif 2510 { 2511 nfsm_dissect(tl, u_int32_t *, 2512 2 * NFSX_UNSIGNED); 2513 fileno = fxdr_unsigned(u_quad_t, *tl++); 2514 len = fxdr_unsigned(int, *tl); 2515 } 2516 if (len <= 0 || len > NFS_MAXNAMLEN) { 2517 error = EBADRPC; 2518 m_freem(mrep); 2519 goto nfsmout; 2520 } 2521 /* for cookie stashing */ 2522 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2523 left = NFS_DIRFRAGSIZ - blksiz; 2524 if (reclen > left) { 2525 memset(uiop->uio_iov->iov_base, 0, left); 2526 dp->d_reclen += left; 2527 UIO_ADVANCE(uiop, left); 2528 blksiz = 0; 2529 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2530 } 2531 if (reclen > uiop->uio_resid) 2532 bigenough = 0; 2533 if (bigenough) { 2534 int tlen; 2535 2536 dp = (struct dirent *)uiop->uio_iov->iov_base; 2537 dp->d_fileno = fileno; 2538 dp->d_namlen = len; 2539 dp->d_reclen = reclen; 2540 dp->d_type = DT_UNKNOWN; 2541 blksiz += reclen; 2542 if (blksiz == NFS_DIRFRAGSIZ) 2543 blksiz = 0; 2544 UIO_ADVANCE(uiop, DIRHDSIZ); 2545 nfsm_mtouio(uiop, len); 2546 tlen = reclen - (DIRHDSIZ + len); 2547 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2548 UIO_ADVANCE(uiop, tlen); 2549 } else 2550 nfsm_adv(nfsm_rndup(len)); 2551 #ifndef NFS_V2_ONLY 2552 if (v3) { 2553 nfsm_dissect(tl, u_int32_t *, 2554 3 * NFSX_UNSIGNED); 2555 } else 2556 #endif 2557 { 2558 nfsm_dissect(tl, u_int32_t *, 2559 2 * NFSX_UNSIGNED); 2560 } 2561 if (bigenough) { 2562 #ifndef NFS_V2_ONLY 2563 if (v3) { 2564 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2565 uiop->uio_offset = 2566 fxdr_swapcookie3(tl); 2567 else 2568 uiop->uio_offset = 2569 fxdr_cookie3(tl); 2570 } 2571 else 2572 #endif 2573 { 2574 uiop->uio_offset = 2575 fxdr_unsigned(off_t, *tl); 2576 } 2577 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2578 } 2579 if (v3) 2580 tl += 2; 2581 else 2582 tl++; 2583 more_dirs = fxdr_unsigned(int, *tl); 2584 } 2585 /* 2586 * If at end of rpc data, get the eof boolean 2587 */ 2588 if (!more_dirs) { 2589 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2590 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2591 2592 /* 2593 * kludge: if we got no entries, treat it as EOF. 2594 * some server sometimes send a reply without any 2595 * entries or EOF. 2596 * although it might mean the server has very long name, 2597 * we can't handle such entries anyway. 2598 */ 2599 2600 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2601 more_dirs = 0; 2602 } 2603 m_freem(mrep); 2604 } 2605 /* 2606 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2607 * by increasing d_reclen for the last record. 2608 */ 2609 if (blksiz > 0) { 2610 left = NFS_DIRFRAGSIZ - blksiz; 2611 memset(uiop->uio_iov->iov_base, 0, left); 2612 dp->d_reclen += left; 2613 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2614 UIO_ADVANCE(uiop, left); 2615 } 2616 2617 /* 2618 * We are now either at the end of the directory or have filled the 2619 * block. 2620 */ 2621 if (bigenough) { 2622 dnp->n_direofoffset = uiop->uio_offset; 2623 dnp->n_flag |= NEOFVALID; 2624 } 2625 nfsmout: 2626 return (error); 2627 } 2628 2629 #ifndef NFS_V2_ONLY 2630 /* 2631 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2632 */ 2633 int 2634 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2635 { 2636 int len, left; 2637 struct dirent *dp = NULL; 2638 u_int32_t *tl; 2639 char *cp; 2640 int32_t t1, t2; 2641 struct vnode *newvp; 2642 char *bpos, *dpos, *cp2; 2643 struct mbuf *mreq, *mrep, *md, *mb; 2644 struct nameidata nami, *ndp = &nami; 2645 struct componentname *cnp = &ndp->ni_cnd; 2646 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2647 struct nfsnode *dnp = VTONFS(vp), *np; 2648 nfsfh_t *fhp; 2649 u_quad_t fileno; 2650 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2651 int attrflag, fhsize, nrpcs = 0, reclen; 2652 struct nfs_fattr fattr, *fp; 2653 2654 #ifdef DIAGNOSTIC 2655 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2656 panic("nfs readdirplusrpc bad uio"); 2657 #endif 2658 ndp->ni_dvp = vp; 2659 newvp = NULLVP; 2660 2661 /* 2662 * Loop around doing readdir rpc's of size nm_readdirsize 2663 * truncated to a multiple of NFS_DIRFRAGSIZ. 2664 * The stopping criteria is EOF or buffer full. 2665 */ 2666 while (more_dirs && bigenough) { 2667 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2668 bigenough = 0; 2669 break; 2670 } 2671 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2672 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2673 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2674 nfsm_fhtom(dnp, 1); 2675 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2676 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2677 txdr_swapcookie3(uiop->uio_offset, tl); 2678 } else { 2679 txdr_cookie3(uiop->uio_offset, tl); 2680 } 2681 tl += 2; 2682 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2683 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2684 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2685 *tl = txdr_unsigned(nmp->nm_rsize); 2686 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2687 nfsm_postop_attr(vp, attrflag, 0); 2688 if (error) { 2689 m_freem(mrep); 2690 goto nfsmout; 2691 } 2692 nrpcs++; 2693 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2694 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2695 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2696 more_dirs = fxdr_unsigned(int, *tl); 2697 2698 /* loop thru the dir entries, doctoring them to 4bsd form */ 2699 while (more_dirs && bigenough) { 2700 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2701 fileno = fxdr_hyper(tl); 2702 len = fxdr_unsigned(int, *(tl + 2)); 2703 if (len <= 0 || len > NFS_MAXNAMLEN) { 2704 error = EBADRPC; 2705 m_freem(mrep); 2706 goto nfsmout; 2707 } 2708 /* for cookie stashing */ 2709 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2710 left = NFS_DIRFRAGSIZ - blksiz; 2711 if (reclen > left) { 2712 /* 2713 * DIRFRAGSIZ is aligned, no need to align 2714 * again here. 2715 */ 2716 memset(uiop->uio_iov->iov_base, 0, left); 2717 dp->d_reclen += left; 2718 UIO_ADVANCE(uiop, left); 2719 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2720 blksiz = 0; 2721 } 2722 if (reclen > uiop->uio_resid) 2723 bigenough = 0; 2724 if (bigenough) { 2725 int tlen; 2726 2727 dp = (struct dirent *)uiop->uio_iov->iov_base; 2728 dp->d_fileno = fileno; 2729 dp->d_namlen = len; 2730 dp->d_reclen = reclen; 2731 dp->d_type = DT_UNKNOWN; 2732 blksiz += reclen; 2733 if (blksiz == NFS_DIRFRAGSIZ) 2734 blksiz = 0; 2735 UIO_ADVANCE(uiop, DIRHDSIZ); 2736 nfsm_mtouio(uiop, len); 2737 tlen = reclen - (DIRHDSIZ + len); 2738 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2739 UIO_ADVANCE(uiop, tlen); 2740 cnp->cn_nameptr = dp->d_name; 2741 cnp->cn_namelen = dp->d_namlen; 2742 } else 2743 nfsm_adv(nfsm_rndup(len)); 2744 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2745 if (bigenough) { 2746 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2747 uiop->uio_offset = 2748 fxdr_swapcookie3(tl); 2749 else 2750 uiop->uio_offset = 2751 fxdr_cookie3(tl); 2752 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2753 } 2754 tl += 2; 2755 2756 /* 2757 * Since the attributes are before the file handle 2758 * (sigh), we must skip over the attributes and then 2759 * come back and get them. 2760 */ 2761 attrflag = fxdr_unsigned(int, *tl); 2762 if (attrflag) { 2763 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2764 memcpy(&fattr, fp, NFSX_V3FATTR); 2765 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2766 doit = fxdr_unsigned(int, *tl); 2767 if (doit) { 2768 nfsm_getfh(fhp, fhsize, 1); 2769 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2770 vref(vp); 2771 newvp = vp; 2772 np = dnp; 2773 } else { 2774 error = nfs_nget1(vp->v_mount, fhp, 2775 fhsize, &np, LK_NOWAIT); 2776 if (!error) 2777 newvp = NFSTOV(np); 2778 } 2779 if (!error) { 2780 const char *xcp; 2781 2782 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2783 if (bigenough) { 2784 dp->d_type = 2785 IFTODT(VTTOIF(np->n_vattr->va_type)); 2786 if (cnp->cn_namelen <= NCHNAMLEN) { 2787 ndp->ni_vp = newvp; 2788 xcp = cnp->cn_nameptr + 2789 cnp->cn_namelen; 2790 cnp->cn_hash = 2791 namei_hash(cnp->cn_nameptr, &xcp); 2792 nfs_cache_enter(ndp->ni_dvp, 2793 ndp->ni_vp, cnp); 2794 } 2795 } 2796 } 2797 error = 0; 2798 } 2799 } else { 2800 /* Just skip over the file handle */ 2801 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2802 i = fxdr_unsigned(int, *tl); 2803 nfsm_adv(nfsm_rndup(i)); 2804 } 2805 if (newvp != NULLVP) { 2806 if (newvp == vp) 2807 vrele(newvp); 2808 else 2809 vput(newvp); 2810 newvp = NULLVP; 2811 } 2812 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2813 more_dirs = fxdr_unsigned(int, *tl); 2814 } 2815 /* 2816 * If at end of rpc data, get the eof boolean 2817 */ 2818 if (!more_dirs) { 2819 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2820 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2821 2822 /* 2823 * kludge: see a comment in nfs_readdirrpc. 2824 */ 2825 2826 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2827 more_dirs = 0; 2828 } 2829 m_freem(mrep); 2830 } 2831 /* 2832 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2833 * by increasing d_reclen for the last record. 2834 */ 2835 if (blksiz > 0) { 2836 left = NFS_DIRFRAGSIZ - blksiz; 2837 memset(uiop->uio_iov->iov_base, 0, left); 2838 dp->d_reclen += left; 2839 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2840 UIO_ADVANCE(uiop, left); 2841 } 2842 2843 /* 2844 * We are now either at the end of the directory or have filled the 2845 * block. 2846 */ 2847 if (bigenough) { 2848 dnp->n_direofoffset = uiop->uio_offset; 2849 dnp->n_flag |= NEOFVALID; 2850 } 2851 nfsmout: 2852 if (newvp != NULLVP) { 2853 if(newvp == vp) 2854 vrele(newvp); 2855 else 2856 vput(newvp); 2857 } 2858 return (error); 2859 } 2860 #endif 2861 2862 /* 2863 * Silly rename. To make the NFS filesystem that is stateless look a little 2864 * more like the "ufs" a remove of an active vnode is translated to a rename 2865 * to a funny looking filename that is removed by nfs_inactive on the 2866 * nfsnode. There is the potential for another process on a different client 2867 * to create the same funny name between the nfs_lookitup() fails and the 2868 * nfs_rename() completes, but... 2869 */ 2870 int 2871 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2872 { 2873 struct sillyrename *sp; 2874 struct nfsnode *np; 2875 int error; 2876 pid_t pid; 2877 2878 cache_purge(dvp); 2879 np = VTONFS(vp); 2880 #ifndef DIAGNOSTIC 2881 if (vp->v_type == VDIR) 2882 panic("nfs: sillyrename dir"); 2883 #endif 2884 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2885 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2886 sp->s_dvp = dvp; 2887 vref(dvp); 2888 2889 /* Fudge together a funny name */ 2890 pid = curlwp->l_proc->p_pid; 2891 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2892 sp->s_namlen = 12; 2893 sp->s_name[8] = hexdigits[pid & 0xf]; 2894 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2895 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2896 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2897 2898 /* Try lookitups until we get one that isn't there */ 2899 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2900 curlwp, (struct nfsnode **)0) == 0) { 2901 sp->s_name[4]++; 2902 if (sp->s_name[4] > 'z') { 2903 error = EINVAL; 2904 goto bad; 2905 } 2906 } 2907 if (dolink) { 2908 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2909 sp->s_cred, curlwp); 2910 /* 2911 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2912 */ 2913 if (error == ENOTSUP) { 2914 error = nfs_renameit(dvp, cnp, sp); 2915 } 2916 } else { 2917 error = nfs_renameit(dvp, cnp, sp); 2918 } 2919 if (error) 2920 goto bad; 2921 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2922 curlwp, &np); 2923 np->n_sillyrename = sp; 2924 return (0); 2925 bad: 2926 vrele(sp->s_dvp); 2927 kauth_cred_free(sp->s_cred); 2928 kmem_free(sp, sizeof(*sp)); 2929 return (error); 2930 } 2931 2932 /* 2933 * Look up a file name and optionally either update the file handle or 2934 * allocate an nfsnode, depending on the value of npp. 2935 * npp == NULL --> just do the lookup 2936 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2937 * handled too 2938 * *npp != NULL --> update the file handle in the vnode 2939 */ 2940 int 2941 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2942 { 2943 u_int32_t *tl; 2944 char *cp; 2945 int32_t t1, t2; 2946 struct vnode *newvp = (struct vnode *)0; 2947 struct nfsnode *np, *dnp = VTONFS(dvp); 2948 char *bpos, *dpos, *cp2; 2949 int error = 0, fhlen; 2950 #ifndef NFS_V2_ONLY 2951 int attrflag; 2952 #endif 2953 struct mbuf *mreq, *mrep, *md, *mb; 2954 nfsfh_t *nfhp; 2955 const int v3 = NFS_ISV3(dvp); 2956 2957 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2958 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2959 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2960 nfsm_fhtom(dnp, v3); 2961 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2962 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2963 if (npp && !error) { 2964 nfsm_getfh(nfhp, fhlen, v3); 2965 if (*npp) { 2966 np = *npp; 2967 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2968 kmem_free(np->n_fhp, np->n_fhsize); 2969 np->n_fhp = &np->n_fh; 2970 } 2971 #if NFS_SMALLFH < NFSX_V3FHMAX 2972 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2973 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2974 #endif 2975 memcpy(np->n_fhp, nfhp, fhlen); 2976 np->n_fhsize = fhlen; 2977 newvp = NFSTOV(np); 2978 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2979 vref(dvp); 2980 newvp = dvp; 2981 np = dnp; 2982 } else { 2983 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2984 if (error) { 2985 m_freem(mrep); 2986 return (error); 2987 } 2988 newvp = NFSTOV(np); 2989 } 2990 #ifndef NFS_V2_ONLY 2991 if (v3) { 2992 nfsm_postop_attr(newvp, attrflag, 0); 2993 if (!attrflag && *npp == NULL) { 2994 m_freem(mrep); 2995 vput(newvp); 2996 return (ENOENT); 2997 } 2998 } else 2999 #endif 3000 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3001 } 3002 nfsm_reqdone; 3003 if (npp && *npp == NULL) { 3004 if (error) { 3005 if (newvp) 3006 vput(newvp); 3007 } else 3008 *npp = np; 3009 } 3010 return (error); 3011 } 3012 3013 #ifndef NFS_V2_ONLY 3014 /* 3015 * Nfs Version 3 commit rpc 3016 */ 3017 int 3018 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3019 { 3020 char *cp; 3021 u_int32_t *tl; 3022 int32_t t1, t2; 3023 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3024 char *bpos, *dpos, *cp2; 3025 int error = 0, wccflag = NFSV3_WCCRATTR; 3026 struct mbuf *mreq, *mrep, *md, *mb; 3027 struct nfsnode *np; 3028 3029 KASSERT(NFS_ISV3(vp)); 3030 3031 #ifdef NFS_DEBUG_COMMIT 3032 printf("commit %lu - %lu\n", (unsigned long)offset, 3033 (unsigned long)(offset + cnt)); 3034 #endif 3035 3036 mutex_enter(&nmp->nm_lock); 3037 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3038 mutex_exit(&nmp->nm_lock); 3039 return (0); 3040 } 3041 mutex_exit(&nmp->nm_lock); 3042 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3043 np = VTONFS(vp); 3044 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3045 nfsm_fhtom(np, 1); 3046 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3047 txdr_hyper(offset, tl); 3048 tl += 2; 3049 *tl = txdr_unsigned(cnt); 3050 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3051 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3052 if (!error) { 3053 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3054 mutex_enter(&nmp->nm_lock); 3055 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3056 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3057 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3058 error = NFSERR_STALEWRITEVERF; 3059 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3060 } 3061 mutex_exit(&nmp->nm_lock); 3062 } 3063 nfsm_reqdone; 3064 return (error); 3065 } 3066 #endif 3067 3068 /* 3069 * Kludge City.. 3070 * - make nfs_bmap() essentially a no-op that does no translation 3071 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3072 * (Maybe I could use the process's page mapping, but I was concerned that 3073 * Kernel Write might not be enabled and also figured copyout() would do 3074 * a lot more work than memcpy() and also it currently happens in the 3075 * context of the swapper process (2). 3076 */ 3077 int 3078 nfs_bmap(void *v) 3079 { 3080 struct vop_bmap_args /* { 3081 struct vnode *a_vp; 3082 daddr_t a_bn; 3083 struct vnode **a_vpp; 3084 daddr_t *a_bnp; 3085 int *a_runp; 3086 } */ *ap = v; 3087 struct vnode *vp = ap->a_vp; 3088 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3089 3090 if (ap->a_vpp != NULL) 3091 *ap->a_vpp = vp; 3092 if (ap->a_bnp != NULL) 3093 *ap->a_bnp = ap->a_bn << bshift; 3094 if (ap->a_runp != NULL) 3095 *ap->a_runp = 1024 * 1024; /* XXX */ 3096 return (0); 3097 } 3098 3099 /* 3100 * Strategy routine. 3101 * For async requests when nfsiod(s) are running, queue the request by 3102 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3103 * request. 3104 */ 3105 int 3106 nfs_strategy(void *v) 3107 { 3108 struct vop_strategy_args *ap = v; 3109 struct buf *bp = ap->a_bp; 3110 int error = 0; 3111 3112 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3113 panic("nfs physio/async"); 3114 3115 /* 3116 * If the op is asynchronous and an i/o daemon is waiting 3117 * queue the request, wake it up and wait for completion 3118 * otherwise just do it ourselves. 3119 */ 3120 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3121 error = nfs_doio(bp); 3122 return (error); 3123 } 3124 3125 /* 3126 * fsync vnode op. Just call nfs_flush() with commit == 1. 3127 */ 3128 /* ARGSUSED */ 3129 int 3130 nfs_fsync(void *v) 3131 { 3132 struct vop_fsync_args /* { 3133 struct vnodeop_desc *a_desc; 3134 struct vnode * a_vp; 3135 kauth_cred_t a_cred; 3136 int a_flags; 3137 off_t offlo; 3138 off_t offhi; 3139 struct lwp * a_l; 3140 } */ *ap = v; 3141 3142 struct vnode *vp = ap->a_vp; 3143 3144 if (vp->v_type != VREG) 3145 return 0; 3146 3147 return (nfs_flush(vp, ap->a_cred, 3148 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3149 } 3150 3151 /* 3152 * Flush all the data associated with a vnode. 3153 */ 3154 int 3155 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3156 int commit) 3157 { 3158 struct nfsnode *np = VTONFS(vp); 3159 int error; 3160 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3161 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3162 3163 mutex_enter(&vp->v_interlock); 3164 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3165 if (np->n_flag & NWRITEERR) { 3166 error = np->n_error; 3167 np->n_flag &= ~NWRITEERR; 3168 } 3169 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3170 return (error); 3171 } 3172 3173 /* 3174 * Return POSIX pathconf information applicable to nfs. 3175 * 3176 * N.B. The NFS V2 protocol doesn't support this RPC. 3177 */ 3178 /* ARGSUSED */ 3179 int 3180 nfs_pathconf(void *v) 3181 { 3182 struct vop_pathconf_args /* { 3183 struct vnode *a_vp; 3184 int a_name; 3185 register_t *a_retval; 3186 } */ *ap = v; 3187 struct nfsv3_pathconf *pcp; 3188 struct vnode *vp = ap->a_vp; 3189 struct mbuf *mreq, *mrep, *md, *mb; 3190 int32_t t1, t2; 3191 u_int32_t *tl; 3192 char *bpos, *dpos, *cp, *cp2; 3193 int error = 0, attrflag; 3194 #ifndef NFS_V2_ONLY 3195 struct nfsmount *nmp; 3196 unsigned int l; 3197 u_int64_t maxsize; 3198 #endif 3199 const int v3 = NFS_ISV3(vp); 3200 struct nfsnode *np = VTONFS(vp); 3201 3202 switch (ap->a_name) { 3203 /* Names that can be resolved locally. */ 3204 case _PC_PIPE_BUF: 3205 *ap->a_retval = PIPE_BUF; 3206 break; 3207 case _PC_SYNC_IO: 3208 *ap->a_retval = 1; 3209 break; 3210 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3211 case _PC_LINK_MAX: 3212 case _PC_NAME_MAX: 3213 case _PC_CHOWN_RESTRICTED: 3214 case _PC_NO_TRUNC: 3215 if (!v3) { 3216 error = EINVAL; 3217 break; 3218 } 3219 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3220 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3221 nfsm_fhtom(np, 1); 3222 nfsm_request(np, NFSPROC_PATHCONF, 3223 curlwp, curlwp->l_cred); /* XXX */ 3224 nfsm_postop_attr(vp, attrflag, 0); 3225 if (!error) { 3226 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3227 NFSX_V3PATHCONF); 3228 switch (ap->a_name) { 3229 case _PC_LINK_MAX: 3230 *ap->a_retval = 3231 fxdr_unsigned(register_t, pcp->pc_linkmax); 3232 break; 3233 case _PC_NAME_MAX: 3234 *ap->a_retval = 3235 fxdr_unsigned(register_t, pcp->pc_namemax); 3236 break; 3237 case _PC_CHOWN_RESTRICTED: 3238 *ap->a_retval = 3239 (pcp->pc_chownrestricted == nfs_true); 3240 break; 3241 case _PC_NO_TRUNC: 3242 *ap->a_retval = 3243 (pcp->pc_notrunc == nfs_true); 3244 break; 3245 } 3246 } 3247 nfsm_reqdone; 3248 break; 3249 case _PC_FILESIZEBITS: 3250 #ifndef NFS_V2_ONLY 3251 if (v3) { 3252 nmp = VFSTONFS(vp->v_mount); 3253 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3254 if ((error = nfs_fsinfo(nmp, vp, 3255 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3256 break; 3257 for (l = 0, maxsize = nmp->nm_maxfilesize; 3258 (maxsize >> l) > 0; l++) 3259 ; 3260 *ap->a_retval = l + 1; 3261 } else 3262 #endif 3263 { 3264 *ap->a_retval = 32; /* NFS V2 limitation */ 3265 } 3266 break; 3267 default: 3268 error = EINVAL; 3269 break; 3270 } 3271 3272 return (error); 3273 } 3274 3275 /* 3276 * NFS advisory byte-level locks. 3277 */ 3278 int 3279 nfs_advlock(void *v) 3280 { 3281 struct vop_advlock_args /* { 3282 struct vnode *a_vp; 3283 void *a_id; 3284 int a_op; 3285 struct flock *a_fl; 3286 int a_flags; 3287 } */ *ap = v; 3288 struct nfsnode *np = VTONFS(ap->a_vp); 3289 3290 return lf_advlock(ap, &np->n_lockf, np->n_size); 3291 } 3292 3293 /* 3294 * Print out the contents of an nfsnode. 3295 */ 3296 int 3297 nfs_print(void *v) 3298 { 3299 struct vop_print_args /* { 3300 struct vnode *a_vp; 3301 } */ *ap = v; 3302 struct vnode *vp = ap->a_vp; 3303 struct nfsnode *np = VTONFS(vp); 3304 3305 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3306 (unsigned long long)np->n_vattr->va_fileid, 3307 (unsigned long long)np->n_vattr->va_fsid); 3308 if (vp->v_type == VFIFO) 3309 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3310 printf("\n"); 3311 return (0); 3312 } 3313 3314 /* 3315 * nfs unlock wrapper. 3316 */ 3317 int 3318 nfs_unlock(void *v) 3319 { 3320 struct vop_unlock_args /* { 3321 struct vnode *a_vp; 3322 int a_flags; 3323 } */ *ap = v; 3324 struct vnode *vp = ap->a_vp; 3325 3326 /* 3327 * VOP_UNLOCK can be called by nfs_loadattrcache 3328 * with v_data == 0. 3329 */ 3330 if (VTONFS(vp)) { 3331 nfs_delayedtruncate(vp); 3332 } 3333 3334 return genfs_unlock(v); 3335 } 3336 3337 /* 3338 * nfs special file access vnode op. 3339 * Essentially just get vattr and then imitate iaccess() since the device is 3340 * local to the client. 3341 */ 3342 int 3343 nfsspec_access(void *v) 3344 { 3345 struct vop_access_args /* { 3346 struct vnode *a_vp; 3347 int a_mode; 3348 kauth_cred_t a_cred; 3349 struct lwp *a_l; 3350 } */ *ap = v; 3351 struct vattr va; 3352 struct vnode *vp = ap->a_vp; 3353 int error; 3354 3355 error = VOP_GETATTR(vp, &va, ap->a_cred); 3356 if (error) 3357 return (error); 3358 3359 /* 3360 * Disallow write attempts on filesystems mounted read-only; 3361 * unless the file is a socket, fifo, or a block or character 3362 * device resident on the filesystem. 3363 */ 3364 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3365 switch (vp->v_type) { 3366 case VREG: 3367 case VDIR: 3368 case VLNK: 3369 return (EROFS); 3370 default: 3371 break; 3372 } 3373 } 3374 3375 return (genfs_can_access(va.va_type, va.va_mode, 3376 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3377 } 3378 3379 /* 3380 * Read wrapper for special devices. 3381 */ 3382 int 3383 nfsspec_read(void *v) 3384 { 3385 struct vop_read_args /* { 3386 struct vnode *a_vp; 3387 struct uio *a_uio; 3388 int a_ioflag; 3389 kauth_cred_t a_cred; 3390 } */ *ap = v; 3391 struct nfsnode *np = VTONFS(ap->a_vp); 3392 3393 /* 3394 * Set access flag. 3395 */ 3396 np->n_flag |= NACC; 3397 getnanotime(&np->n_atim); 3398 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3399 } 3400 3401 /* 3402 * Write wrapper for special devices. 3403 */ 3404 int 3405 nfsspec_write(void *v) 3406 { 3407 struct vop_write_args /* { 3408 struct vnode *a_vp; 3409 struct uio *a_uio; 3410 int a_ioflag; 3411 kauth_cred_t a_cred; 3412 } */ *ap = v; 3413 struct nfsnode *np = VTONFS(ap->a_vp); 3414 3415 /* 3416 * Set update flag. 3417 */ 3418 np->n_flag |= NUPD; 3419 getnanotime(&np->n_mtim); 3420 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3421 } 3422 3423 /* 3424 * Close wrapper for special devices. 3425 * 3426 * Update the times on the nfsnode then do device close. 3427 */ 3428 int 3429 nfsspec_close(void *v) 3430 { 3431 struct vop_close_args /* { 3432 struct vnode *a_vp; 3433 int a_fflag; 3434 kauth_cred_t a_cred; 3435 struct lwp *a_l; 3436 } */ *ap = v; 3437 struct vnode *vp = ap->a_vp; 3438 struct nfsnode *np = VTONFS(vp); 3439 struct vattr vattr; 3440 3441 if (np->n_flag & (NACC | NUPD)) { 3442 np->n_flag |= NCHG; 3443 if (vp->v_usecount == 1 && 3444 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3445 vattr_null(&vattr); 3446 if (np->n_flag & NACC) 3447 vattr.va_atime = np->n_atim; 3448 if (np->n_flag & NUPD) 3449 vattr.va_mtime = np->n_mtim; 3450 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3451 } 3452 } 3453 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3454 } 3455 3456 /* 3457 * Read wrapper for fifos. 3458 */ 3459 int 3460 nfsfifo_read(void *v) 3461 { 3462 struct vop_read_args /* { 3463 struct vnode *a_vp; 3464 struct uio *a_uio; 3465 int a_ioflag; 3466 kauth_cred_t a_cred; 3467 } */ *ap = v; 3468 struct nfsnode *np = VTONFS(ap->a_vp); 3469 3470 /* 3471 * Set access flag. 3472 */ 3473 np->n_flag |= NACC; 3474 getnanotime(&np->n_atim); 3475 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3476 } 3477 3478 /* 3479 * Write wrapper for fifos. 3480 */ 3481 int 3482 nfsfifo_write(void *v) 3483 { 3484 struct vop_write_args /* { 3485 struct vnode *a_vp; 3486 struct uio *a_uio; 3487 int a_ioflag; 3488 kauth_cred_t a_cred; 3489 } */ *ap = v; 3490 struct nfsnode *np = VTONFS(ap->a_vp); 3491 3492 /* 3493 * Set update flag. 3494 */ 3495 np->n_flag |= NUPD; 3496 getnanotime(&np->n_mtim); 3497 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3498 } 3499 3500 /* 3501 * Close wrapper for fifos. 3502 * 3503 * Update the times on the nfsnode then do fifo close. 3504 */ 3505 int 3506 nfsfifo_close(void *v) 3507 { 3508 struct vop_close_args /* { 3509 struct vnode *a_vp; 3510 int a_fflag; 3511 kauth_cred_t a_cred; 3512 struct lwp *a_l; 3513 } */ *ap = v; 3514 struct vnode *vp = ap->a_vp; 3515 struct nfsnode *np = VTONFS(vp); 3516 struct vattr vattr; 3517 3518 if (np->n_flag & (NACC | NUPD)) { 3519 struct timespec ts; 3520 3521 getnanotime(&ts); 3522 if (np->n_flag & NACC) 3523 np->n_atim = ts; 3524 if (np->n_flag & NUPD) 3525 np->n_mtim = ts; 3526 np->n_flag |= NCHG; 3527 if (vp->v_usecount == 1 && 3528 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3529 vattr_null(&vattr); 3530 if (np->n_flag & NACC) 3531 vattr.va_atime = np->n_atim; 3532 if (np->n_flag & NUPD) 3533 vattr.va_mtime = np->n_mtim; 3534 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3535 } 3536 } 3537 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3538 } 3539