nfs_vnops.c revision 1.293.4.2
1 /* $NetBSD: nfs_vnops.c,v 1.293.4.2 2016/07/14 06:57:12 snj 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 1.293.4.2 2016/07/14 06:57:12 snj Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/proc.h> 51 #include <sys/kernel.h> 52 #include <sys/systm.h> 53 #include <sys/resourcevar.h> 54 #include <sys/mount.h> 55 #include <sys/buf.h> 56 #include <sys/condvar.h> 57 #include <sys/disk.h> 58 #include <sys/malloc.h> 59 #include <sys/kmem.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/vnode.h> 64 #include <sys/dirent.h> 65 #include <sys/fcntl.h> 66 #include <sys/hash.h> 67 #include <sys/lockf.h> 68 #include <sys/stat.h> 69 #include <sys/unistd.h> 70 #include <sys/kauth.h> 71 #include <sys/cprng.h> 72 73 #include <uvm/uvm_extern.h> 74 #include <uvm/uvm.h> 75 76 #include <miscfs/fifofs/fifo.h> 77 #include <miscfs/genfs/genfs.h> 78 #include <miscfs/genfs/genfs_node.h> 79 #include <miscfs/specfs/specdev.h> 80 81 #include <nfs/rpcv2.h> 82 #include <nfs/nfsproto.h> 83 #include <nfs/nfs.h> 84 #include <nfs/nfsnode.h> 85 #include <nfs/nfsmount.h> 86 #include <nfs/xdr_subs.h> 87 #include <nfs/nfsm_subs.h> 88 #include <nfs/nfs_var.h> 89 90 #include <net/if.h> 91 #include <netinet/in.h> 92 #include <netinet/in_var.h> 93 94 /* 95 * Global vfs data structures for nfs 96 */ 97 int (**nfsv2_vnodeop_p)(void *); 98 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 99 { &vop_default_desc, vn_default_error }, 100 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 101 { &vop_create_desc, nfs_create }, /* create */ 102 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 103 { &vop_open_desc, nfs_open }, /* open */ 104 { &vop_close_desc, nfs_close }, /* close */ 105 { &vop_access_desc, nfs_access }, /* access */ 106 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 107 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 108 { &vop_read_desc, nfs_read }, /* read */ 109 { &vop_write_desc, nfs_write }, /* write */ 110 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 111 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 112 { &vop_poll_desc, nfs_poll }, /* poll */ 113 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 114 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 115 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 116 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 117 { &vop_seek_desc, nfs_seek }, /* seek */ 118 { &vop_remove_desc, nfs_remove }, /* remove */ 119 { &vop_link_desc, nfs_link }, /* link */ 120 { &vop_rename_desc, nfs_rename }, /* rename */ 121 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 122 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 123 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 124 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 125 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 126 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 127 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 128 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 129 { &vop_lock_desc, nfs_lock }, /* lock */ 130 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 131 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 132 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 133 { &vop_print_desc, nfs_print }, /* print */ 134 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 135 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 136 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 137 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 138 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 139 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 140 { NULL, NULL } 141 }; 142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 144 145 /* 146 * Special device vnode ops 147 */ 148 int (**spec_nfsv2nodeop_p)(void *); 149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 150 { &vop_default_desc, vn_default_error }, 151 { &vop_lookup_desc, spec_lookup }, /* lookup */ 152 { &vop_create_desc, spec_create }, /* create */ 153 { &vop_mknod_desc, spec_mknod }, /* mknod */ 154 { &vop_open_desc, spec_open }, /* open */ 155 { &vop_close_desc, nfsspec_close }, /* close */ 156 { &vop_access_desc, nfsspec_access }, /* access */ 157 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 158 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 159 { &vop_read_desc, nfsspec_read }, /* read */ 160 { &vop_write_desc, nfsspec_write }, /* write */ 161 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 162 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 163 { &vop_poll_desc, spec_poll }, /* poll */ 164 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 165 { &vop_revoke_desc, spec_revoke }, /* revoke */ 166 { &vop_mmap_desc, spec_mmap }, /* mmap */ 167 { &vop_fsync_desc, spec_fsync }, /* fsync */ 168 { &vop_seek_desc, spec_seek }, /* seek */ 169 { &vop_remove_desc, spec_remove }, /* remove */ 170 { &vop_link_desc, spec_link }, /* link */ 171 { &vop_rename_desc, spec_rename }, /* rename */ 172 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 173 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 174 { &vop_symlink_desc, spec_symlink }, /* symlink */ 175 { &vop_readdir_desc, spec_readdir }, /* readdir */ 176 { &vop_readlink_desc, spec_readlink }, /* readlink */ 177 { &vop_abortop_desc, spec_abortop }, /* abortop */ 178 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 179 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 180 { &vop_lock_desc, nfs_lock }, /* lock */ 181 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 182 { &vop_bmap_desc, spec_bmap }, /* bmap */ 183 { &vop_strategy_desc, spec_strategy }, /* strategy */ 184 { &vop_print_desc, nfs_print }, /* print */ 185 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 186 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 187 { &vop_advlock_desc, spec_advlock }, /* advlock */ 188 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 189 { &vop_getpages_desc, spec_getpages }, /* getpages */ 190 { &vop_putpages_desc, spec_putpages }, /* putpages */ 191 { NULL, NULL } 192 }; 193 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 194 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 195 196 int (**fifo_nfsv2nodeop_p)(void *); 197 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 198 { &vop_default_desc, vn_default_error }, 199 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 200 { &vop_create_desc, vn_fifo_bypass }, /* create */ 201 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 202 { &vop_open_desc, vn_fifo_bypass }, /* open */ 203 { &vop_close_desc, nfsfifo_close }, /* close */ 204 { &vop_access_desc, nfsspec_access }, /* access */ 205 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 206 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 207 { &vop_read_desc, nfsfifo_read }, /* read */ 208 { &vop_write_desc, nfsfifo_write }, /* write */ 209 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 210 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 211 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 212 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 213 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 214 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 215 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 216 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 217 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 218 { &vop_link_desc, vn_fifo_bypass }, /* link */ 219 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 220 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 221 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 222 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 223 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 224 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 225 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 226 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 227 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 228 { &vop_lock_desc, nfs_lock }, /* lock */ 229 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 230 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 231 { &vop_strategy_desc, genfs_badop }, /* strategy */ 232 { &vop_print_desc, nfs_print }, /* print */ 233 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 234 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 235 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 236 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 237 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 238 { NULL, NULL } 239 }; 240 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 241 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 242 243 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 244 size_t, kauth_cred_t, struct lwp *); 245 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 246 247 /* 248 * Global variables 249 */ 250 extern u_int32_t nfs_true, nfs_false; 251 extern u_int32_t nfs_xdrneg1; 252 extern const nfstype nfsv3_type[9]; 253 254 int nfs_numasync = 0; 255 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 256 #define UIO_ADVANCE(uio, siz) \ 257 (void)((uio)->uio_resid -= (siz), \ 258 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 259 (uio)->uio_iov->iov_len -= (siz)) 260 261 static void nfs_cache_enter(struct vnode *, struct vnode *, 262 struct componentname *); 263 264 static void 265 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 266 struct componentname *cnp) 267 { 268 struct nfsnode *dnp = VTONFS(dvp); 269 270 if ((cnp->cn_flags & MAKEENTRY) == 0) { 271 return; 272 } 273 if (vp != NULL) { 274 struct nfsnode *np = VTONFS(vp); 275 276 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 277 } 278 279 if (!timespecisset(&dnp->n_nctime)) 280 dnp->n_nctime = dnp->n_vattr->va_mtime; 281 282 cache_enter(dvp, vp, cnp); 283 } 284 285 /* 286 * nfs null call from vfs. 287 */ 288 int 289 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 290 { 291 char *bpos, *dpos; 292 int error = 0; 293 struct mbuf *mreq, *mrep, *md, *mb; 294 struct nfsnode *np = VTONFS(vp); 295 296 nfsm_reqhead(np, NFSPROC_NULL, 0); 297 nfsm_request(np, NFSPROC_NULL, l, cred); 298 nfsm_reqdone; 299 return (error); 300 } 301 302 /* 303 * nfs access vnode op. 304 * For nfs version 2, just return ok. File accesses may fail later. 305 * For nfs version 3, use the access rpc to check accessibility. If file modes 306 * are changed on the server, accesses might still fail later. 307 */ 308 int 309 nfs_access(void *v) 310 { 311 struct vop_access_args /* { 312 struct vnode *a_vp; 313 int a_mode; 314 kauth_cred_t a_cred; 315 } */ *ap = v; 316 struct vnode *vp = ap->a_vp; 317 #ifndef NFS_V2_ONLY 318 u_int32_t *tl; 319 char *cp; 320 int32_t t1, t2; 321 char *bpos, *dpos, *cp2; 322 int error = 0, attrflag; 323 struct mbuf *mreq, *mrep, *md, *mb; 324 u_int32_t mode, rmode; 325 const int v3 = NFS_ISV3(vp); 326 #endif 327 int cachevalid; 328 struct nfsnode *np = VTONFS(vp); 329 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 330 331 cachevalid = (np->n_accstamp != -1 && 332 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 333 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 334 335 /* 336 * Check access cache first. If this request has been made for this 337 * uid shortly before, use the cached result. 338 */ 339 if (cachevalid) { 340 if (!np->n_accerror) { 341 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 342 return np->n_accerror; 343 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 344 return np->n_accerror; 345 } 346 347 #ifndef NFS_V2_ONLY 348 /* 349 * For nfs v3, do an access rpc, otherwise you are stuck emulating 350 * ufs_access() locally using the vattr. This may not be correct, 351 * since the server may apply other access criteria such as 352 * client uid-->server uid mapping that we do not know about, but 353 * this is better than just returning anything that is lying about 354 * in the cache. 355 */ 356 if (v3) { 357 nfsstats.rpccnt[NFSPROC_ACCESS]++; 358 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 359 nfsm_fhtom(np, v3); 360 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 361 if (ap->a_mode & VREAD) 362 mode = NFSV3ACCESS_READ; 363 else 364 mode = 0; 365 if (vp->v_type != VDIR) { 366 if (ap->a_mode & VWRITE) 367 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 368 if (ap->a_mode & VEXEC) 369 mode |= NFSV3ACCESS_EXECUTE; 370 } else { 371 if (ap->a_mode & VWRITE) 372 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 373 NFSV3ACCESS_DELETE); 374 if (ap->a_mode & VEXEC) 375 mode |= NFSV3ACCESS_LOOKUP; 376 } 377 *tl = txdr_unsigned(mode); 378 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 379 nfsm_postop_attr(vp, attrflag, 0); 380 if (!error) { 381 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 382 rmode = fxdr_unsigned(u_int32_t, *tl); 383 /* 384 * The NFS V3 spec does not clarify whether or not 385 * the returned access bits can be a superset of 386 * the ones requested, so... 387 */ 388 if ((rmode & mode) != mode) 389 error = EACCES; 390 } 391 nfsm_reqdone; 392 } else 393 #endif 394 return (nfsspec_access(ap)); 395 #ifndef NFS_V2_ONLY 396 /* 397 * Disallow write attempts on filesystems mounted read-only; 398 * unless the file is a socket, fifo, or a block or character 399 * device resident on the filesystem. 400 */ 401 if (!error && (ap->a_mode & VWRITE) && 402 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 403 switch (vp->v_type) { 404 case VREG: 405 case VDIR: 406 case VLNK: 407 error = EROFS; 408 default: 409 break; 410 } 411 } 412 413 if (!error || error == EACCES) { 414 /* 415 * If we got the same result as for a previous, 416 * different request, OR it in. Don't update 417 * the timestamp in that case. 418 */ 419 if (cachevalid && np->n_accstamp != -1 && 420 error == np->n_accerror) { 421 if (!error) 422 np->n_accmode |= ap->a_mode; 423 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 424 np->n_accmode = ap->a_mode; 425 } else { 426 np->n_accstamp = time_uptime; 427 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 428 np->n_accmode = ap->a_mode; 429 np->n_accerror = error; 430 } 431 } 432 433 return (error); 434 #endif 435 } 436 437 /* 438 * nfs open vnode op 439 * Check to see if the type is ok 440 * and that deletion is not in progress. 441 * For paged in text files, you will need to flush the page cache 442 * if consistency is lost. 443 */ 444 /* ARGSUSED */ 445 int 446 nfs_open(void *v) 447 { 448 struct vop_open_args /* { 449 struct vnode *a_vp; 450 int a_mode; 451 kauth_cred_t a_cred; 452 } */ *ap = v; 453 struct vnode *vp = ap->a_vp; 454 struct nfsnode *np = VTONFS(vp); 455 int error; 456 457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 458 return (EACCES); 459 } 460 461 if (ap->a_mode & FREAD) { 462 if (np->n_rcred != NULL) 463 kauth_cred_free(np->n_rcred); 464 np->n_rcred = ap->a_cred; 465 kauth_cred_hold(np->n_rcred); 466 } 467 if (ap->a_mode & FWRITE) { 468 if (np->n_wcred != NULL) 469 kauth_cred_free(np->n_wcred); 470 np->n_wcred = ap->a_cred; 471 kauth_cred_hold(np->n_wcred); 472 } 473 474 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 475 if (error) 476 return error; 477 478 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 479 480 return (0); 481 } 482 483 /* 484 * nfs close vnode op 485 * What an NFS client should do upon close after writing is a debatable issue. 486 * Most NFS clients push delayed writes to the server upon close, basically for 487 * two reasons: 488 * 1 - So that any write errors may be reported back to the client process 489 * doing the close system call. By far the two most likely errors are 490 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 491 * 2 - To put a worst case upper bound on cache inconsistency between 492 * multiple clients for the file. 493 * There is also a consistency problem for Version 2 of the protocol w.r.t. 494 * not being able to tell if other clients are writing a file concurrently, 495 * since there is no way of knowing if the changed modify time in the reply 496 * is only due to the write for this client. 497 * (NFS Version 3 provides weak cache consistency data in the reply that 498 * should be sufficient to detect and handle this case.) 499 * 500 * The current code does the following: 501 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 502 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 503 * or commit them (this satisfies 1 and 2 except for the 504 * case where the server crashes after this close but 505 * before the commit RPC, which is felt to be "good 506 * enough". Changing the last argument to nfs_flush() to 507 * a 1 would force a commit operation, if it is felt a 508 * commit is necessary now. 509 */ 510 /* ARGSUSED */ 511 int 512 nfs_close(void *v) 513 { 514 struct vop_close_args /* { 515 struct vnodeop_desc *a_desc; 516 struct vnode *a_vp; 517 int a_fflag; 518 kauth_cred_t a_cred; 519 } */ *ap = v; 520 struct vnode *vp = ap->a_vp; 521 struct nfsnode *np = VTONFS(vp); 522 int error = 0; 523 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 524 525 if (vp->v_type == VREG) { 526 if (np->n_flag & NMODIFIED) { 527 #ifndef NFS_V2_ONLY 528 if (NFS_ISV3(vp)) { 529 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 530 np->n_flag &= ~NMODIFIED; 531 } else 532 #endif 533 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 534 NFS_INVALIDATE_ATTRCACHE(np); 535 } 536 if (np->n_flag & NWRITEERR) { 537 np->n_flag &= ~NWRITEERR; 538 error = np->n_error; 539 } 540 } 541 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 542 return (error); 543 } 544 545 /* 546 * nfs getattr call from vfs. 547 */ 548 int 549 nfs_getattr(void *v) 550 { 551 struct vop_getattr_args /* { 552 struct vnode *a_vp; 553 struct vattr *a_vap; 554 kauth_cred_t a_cred; 555 } */ *ap = v; 556 struct vnode *vp = ap->a_vp; 557 struct nfsnode *np = VTONFS(vp); 558 char *cp; 559 u_int32_t *tl; 560 int32_t t1, t2; 561 char *bpos, *dpos; 562 int error = 0; 563 struct mbuf *mreq, *mrep, *md, *mb; 564 const int v3 = NFS_ISV3(vp); 565 566 /* 567 * Update local times for special files. 568 */ 569 if (np->n_flag & (NACC | NUPD)) 570 np->n_flag |= NCHG; 571 572 /* 573 * if we have delayed truncation, do it now. 574 */ 575 nfs_delayedtruncate(vp); 576 577 /* 578 * First look in the cache. 579 */ 580 if (nfs_getattrcache(vp, ap->a_vap) == 0) 581 return (0); 582 nfsstats.rpccnt[NFSPROC_GETATTR]++; 583 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 584 nfsm_fhtom(np, v3); 585 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 586 if (!error) { 587 nfsm_loadattr(vp, ap->a_vap, 0); 588 if (vp->v_type == VDIR && 589 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 590 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 591 } 592 nfsm_reqdone; 593 return (error); 594 } 595 596 /* 597 * nfs setattr call. 598 */ 599 int 600 nfs_setattr(void *v) 601 { 602 struct vop_setattr_args /* { 603 struct vnodeop_desc *a_desc; 604 struct vnode *a_vp; 605 struct vattr *a_vap; 606 kauth_cred_t a_cred; 607 } */ *ap = v; 608 struct vnode *vp = ap->a_vp; 609 struct nfsnode *np = VTONFS(vp); 610 struct vattr *vap = ap->a_vap; 611 int error = 0; 612 u_quad_t tsize = 0; 613 614 /* 615 * Setting of flags is not supported. 616 */ 617 if (vap->va_flags != VNOVAL) 618 return (EOPNOTSUPP); 619 620 /* 621 * Disallow write attempts if the filesystem is mounted read-only. 622 */ 623 if ((vap->va_uid != (uid_t)VNOVAL || 624 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 625 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 626 (vp->v_mount->mnt_flag & MNT_RDONLY)) 627 return (EROFS); 628 if (vap->va_size != VNOVAL) { 629 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 630 return EFBIG; 631 } 632 switch (vp->v_type) { 633 case VDIR: 634 return (EISDIR); 635 case VCHR: 636 case VBLK: 637 case VSOCK: 638 case VFIFO: 639 if (vap->va_mtime.tv_sec == VNOVAL && 640 vap->va_atime.tv_sec == VNOVAL && 641 vap->va_mode == (mode_t)VNOVAL && 642 vap->va_uid == (uid_t)VNOVAL && 643 vap->va_gid == (gid_t)VNOVAL) 644 return (0); 645 vap->va_size = VNOVAL; 646 break; 647 default: 648 /* 649 * Disallow write attempts if the filesystem is 650 * mounted read-only. 651 */ 652 if (vp->v_mount->mnt_flag & MNT_RDONLY) 653 return (EROFS); 654 genfs_node_wrlock(vp); 655 uvm_vnp_setsize(vp, vap->va_size); 656 tsize = np->n_size; 657 np->n_size = vap->va_size; 658 if (vap->va_size == 0) 659 error = nfs_vinvalbuf(vp, 0, 660 ap->a_cred, curlwp, 1); 661 else 662 error = nfs_vinvalbuf(vp, V_SAVE, 663 ap->a_cred, curlwp, 1); 664 if (error) { 665 uvm_vnp_setsize(vp, tsize); 666 genfs_node_unlock(vp); 667 return (error); 668 } 669 np->n_vattr->va_size = vap->va_size; 670 } 671 } else { 672 /* 673 * flush files before setattr because a later write of 674 * cached data might change timestamps or reset sugid bits 675 */ 676 if ((vap->va_mtime.tv_sec != VNOVAL || 677 vap->va_atime.tv_sec != VNOVAL || 678 vap->va_mode != VNOVAL) && 679 vp->v_type == VREG && 680 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 681 curlwp, 1)) == EINTR) 682 return (error); 683 } 684 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 685 if (vap->va_size != VNOVAL) { 686 if (error) { 687 np->n_size = np->n_vattr->va_size = tsize; 688 uvm_vnp_setsize(vp, np->n_size); 689 } 690 genfs_node_unlock(vp); 691 } 692 VN_KNOTE(vp, NOTE_ATTRIB); 693 return (error); 694 } 695 696 /* 697 * Do an nfs setattr rpc. 698 */ 699 int 700 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 701 { 702 struct nfsv2_sattr *sp; 703 char *cp; 704 int32_t t1, t2; 705 char *bpos, *dpos; 706 u_int32_t *tl; 707 int error = 0; 708 struct mbuf *mreq, *mrep, *md, *mb; 709 const int v3 = NFS_ISV3(vp); 710 struct nfsnode *np = VTONFS(vp); 711 #ifndef NFS_V2_ONLY 712 int wccflag = NFSV3_WCCRATTR; 713 char *cp2; 714 #endif 715 716 nfsstats.rpccnt[NFSPROC_SETATTR]++; 717 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 718 nfsm_fhtom(np, v3); 719 #ifndef NFS_V2_ONLY 720 if (v3) { 721 nfsm_v3attrbuild(vap, true); 722 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 723 *tl = nfs_false; 724 } else { 725 #endif 726 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 727 if (vap->va_mode == (mode_t)VNOVAL) 728 sp->sa_mode = nfs_xdrneg1; 729 else 730 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 731 if (vap->va_uid == (uid_t)VNOVAL) 732 sp->sa_uid = nfs_xdrneg1; 733 else 734 sp->sa_uid = txdr_unsigned(vap->va_uid); 735 if (vap->va_gid == (gid_t)VNOVAL) 736 sp->sa_gid = nfs_xdrneg1; 737 else 738 sp->sa_gid = txdr_unsigned(vap->va_gid); 739 sp->sa_size = txdr_unsigned(vap->va_size); 740 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 741 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 742 #ifndef NFS_V2_ONLY 743 } 744 #endif 745 nfsm_request(np, NFSPROC_SETATTR, l, cred); 746 #ifndef NFS_V2_ONLY 747 if (v3) { 748 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 749 } else 750 #endif 751 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 752 nfsm_reqdone; 753 return (error); 754 } 755 756 /* 757 * nfs lookup call, one step at a time... 758 * First look in cache 759 * If not found, do the rpc. 760 */ 761 int 762 nfs_lookup(void *v) 763 { 764 struct vop_lookup_args /* { 765 struct vnodeop_desc *a_desc; 766 struct vnode *a_dvp; 767 struct vnode **a_vpp; 768 struct componentname *a_cnp; 769 } */ *ap = v; 770 struct componentname *cnp = ap->a_cnp; 771 struct vnode *dvp = ap->a_dvp; 772 struct vnode **vpp = ap->a_vpp; 773 int flags; 774 struct vnode *newvp; 775 u_int32_t *tl; 776 char *cp; 777 int32_t t1, t2; 778 char *bpos, *dpos, *cp2; 779 struct mbuf *mreq, *mrep, *md, *mb; 780 long len; 781 nfsfh_t *fhp; 782 struct nfsnode *np; 783 int error = 0, attrflag, fhsize; 784 const int v3 = NFS_ISV3(dvp); 785 786 flags = cnp->cn_flags; 787 788 *vpp = NULLVP; 789 newvp = NULLVP; 790 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 791 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 792 return (EROFS); 793 if (dvp->v_type != VDIR) 794 return (ENOTDIR); 795 796 /* 797 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 798 */ 799 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 800 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 801 if (error) 802 return error; 803 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 804 return EISDIR; 805 vref(dvp); 806 *vpp = dvp; 807 return 0; 808 } 809 810 np = VTONFS(dvp); 811 812 /* 813 * Before performing an RPC, check the name cache to see if 814 * the directory/name pair we are looking for is known already. 815 * If the directory/name pair is found in the name cache, 816 * we have to ensure the directory has not changed from 817 * the time the cache entry has been created. If it has, 818 * the cache entry has to be ignored. 819 */ 820 error = cache_lookup_raw(dvp, vpp, cnp); 821 KASSERT(dvp != *vpp); 822 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 823 if (error >= 0) { 824 struct vattr vattr; 825 int err2; 826 827 if (error && error != ENOENT) { 828 *vpp = NULLVP; 829 return error; 830 } 831 832 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 833 if (err2 != 0) { 834 if (error == 0) 835 vrele(*vpp); 836 *vpp = NULLVP; 837 return err2; 838 } 839 840 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 841 || timespeccmp(&vattr.va_mtime, 842 &VTONFS(dvp)->n_nctime, !=)) { 843 if (error == 0) { 844 vrele(*vpp); 845 *vpp = NULLVP; 846 } 847 cache_purge1(dvp, NULL, PURGE_CHILDREN); 848 timespecclear(&np->n_nctime); 849 goto dorpc; 850 } 851 852 if (error == ENOENT) { 853 goto noentry; 854 } 855 856 /* 857 * investigate the vnode returned by cache_lookup_raw. 858 * if it isn't appropriate, do an rpc. 859 */ 860 newvp = *vpp; 861 if ((flags & ISDOTDOT) != 0) { 862 VOP_UNLOCK(dvp); 863 } 864 error = vn_lock(newvp, LK_EXCLUSIVE); 865 if ((flags & ISDOTDOT) != 0) { 866 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 867 } 868 if (error != 0) { 869 /* newvp has been reclaimed. */ 870 vrele(newvp); 871 *vpp = NULLVP; 872 goto dorpc; 873 } 874 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 875 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 876 nfsstats.lookupcache_hits++; 877 KASSERT(newvp->v_type != VNON); 878 return (0); 879 } 880 cache_purge1(newvp, NULL, PURGE_PARENTS); 881 vput(newvp); 882 *vpp = NULLVP; 883 } 884 dorpc: 885 #if 0 886 /* 887 * because nfsv3 has the same CREATE semantics as ours, 888 * we don't have to perform LOOKUPs beforehand. 889 * 890 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 891 * XXX although we have no way to know if O_EXCL is requested or not. 892 */ 893 894 if (v3 && cnp->cn_nameiop == CREATE && 895 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 896 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 897 return (EJUSTRETURN); 898 } 899 #endif /* 0 */ 900 901 error = 0; 902 newvp = NULLVP; 903 nfsstats.lookupcache_misses++; 904 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 905 len = cnp->cn_namelen; 906 nfsm_reqhead(np, NFSPROC_LOOKUP, 907 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 908 nfsm_fhtom(np, v3); 909 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 910 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 911 if (error) { 912 nfsm_postop_attr(dvp, attrflag, 0); 913 m_freem(mrep); 914 goto nfsmout; 915 } 916 nfsm_getfh(fhp, fhsize, v3); 917 918 /* 919 * Handle RENAME case... 920 */ 921 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 922 if (NFS_CMPFH(np, fhp, fhsize)) { 923 m_freem(mrep); 924 return (EISDIR); 925 } 926 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 927 if (error) { 928 m_freem(mrep); 929 return error; 930 } 931 newvp = NFSTOV(np); 932 #ifndef NFS_V2_ONLY 933 if (v3) { 934 nfsm_postop_attr(newvp, attrflag, 0); 935 nfsm_postop_attr(dvp, attrflag, 0); 936 } else 937 #endif 938 nfsm_loadattr(newvp, (struct vattr *)0, 0); 939 *vpp = newvp; 940 m_freem(mrep); 941 goto validate; 942 } 943 944 /* 945 * The postop attr handling is duplicated for each if case, 946 * because it should be done while dvp is locked (unlocking 947 * dvp is different for each case). 948 */ 949 950 if (NFS_CMPFH(np, fhp, fhsize)) { 951 /* 952 * As we handle "." lookup locally, this is 953 * a broken server. 954 */ 955 m_freem(mrep); 956 return EBADRPC; 957 } else if (flags & ISDOTDOT) { 958 /* 959 * ".." lookup 960 */ 961 VOP_UNLOCK(dvp); 962 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 963 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 964 if (error) { 965 m_freem(mrep); 966 return error; 967 } 968 newvp = NFSTOV(np); 969 970 #ifndef NFS_V2_ONLY 971 if (v3) { 972 nfsm_postop_attr(newvp, attrflag, 0); 973 nfsm_postop_attr(dvp, attrflag, 0); 974 } else 975 #endif 976 nfsm_loadattr(newvp, (struct vattr *)0, 0); 977 } else { 978 /* 979 * Other lookups. 980 */ 981 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 982 if (error) { 983 m_freem(mrep); 984 return error; 985 } 986 newvp = NFSTOV(np); 987 #ifndef NFS_V2_ONLY 988 if (v3) { 989 nfsm_postop_attr(newvp, attrflag, 0); 990 nfsm_postop_attr(dvp, attrflag, 0); 991 } else 992 #endif 993 nfsm_loadattr(newvp, (struct vattr *)0, 0); 994 } 995 if (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) { 996 nfs_cache_enter(dvp, newvp, cnp); 997 } 998 *vpp = newvp; 999 nfsm_reqdone; 1000 if (error) { 1001 /* 1002 * We get here only because of errors returned by 1003 * the RPC. Otherwise we'll have returned above 1004 * (the nfsm_* macros will jump to nfsm_reqdone 1005 * on error). 1006 */ 1007 if (error == ENOENT && cnp->cn_nameiop != CREATE) { 1008 nfs_cache_enter(dvp, NULL, cnp); 1009 } 1010 if (newvp != NULLVP) { 1011 if (newvp == dvp) { 1012 vrele(newvp); 1013 } else { 1014 vput(newvp); 1015 } 1016 } 1017 noentry: 1018 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1019 (flags & ISLASTCN) && error == ENOENT) { 1020 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1021 error = EROFS; 1022 } else { 1023 error = EJUSTRETURN; 1024 } 1025 } 1026 *vpp = NULL; 1027 return error; 1028 } 1029 1030 validate: 1031 /* 1032 * make sure we have valid type and size. 1033 */ 1034 1035 newvp = *vpp; 1036 if (newvp->v_type == VNON) { 1037 struct vattr vattr; /* dummy */ 1038 1039 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1040 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1041 if (error) { 1042 vput(newvp); 1043 *vpp = NULL; 1044 } 1045 } 1046 1047 return error; 1048 } 1049 1050 /* 1051 * nfs read call. 1052 * Just call nfs_bioread() to do the work. 1053 */ 1054 int 1055 nfs_read(void *v) 1056 { 1057 struct vop_read_args /* { 1058 struct vnode *a_vp; 1059 struct uio *a_uio; 1060 int a_ioflag; 1061 kauth_cred_t a_cred; 1062 } */ *ap = v; 1063 struct vnode *vp = ap->a_vp; 1064 1065 if (vp->v_type != VREG) 1066 return EISDIR; 1067 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1068 } 1069 1070 /* 1071 * nfs readlink call 1072 */ 1073 int 1074 nfs_readlink(void *v) 1075 { 1076 struct vop_readlink_args /* { 1077 struct vnode *a_vp; 1078 struct uio *a_uio; 1079 kauth_cred_t a_cred; 1080 } */ *ap = v; 1081 struct vnode *vp = ap->a_vp; 1082 struct nfsnode *np = VTONFS(vp); 1083 1084 if (vp->v_type != VLNK) 1085 return (EPERM); 1086 1087 if (np->n_rcred != NULL) { 1088 kauth_cred_free(np->n_rcred); 1089 } 1090 np->n_rcred = ap->a_cred; 1091 kauth_cred_hold(np->n_rcred); 1092 1093 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1094 } 1095 1096 /* 1097 * Do a readlink rpc. 1098 * Called by nfs_doio() from below the buffer cache. 1099 */ 1100 int 1101 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1102 { 1103 u_int32_t *tl; 1104 char *cp; 1105 int32_t t1, t2; 1106 char *bpos, *dpos, *cp2; 1107 int error = 0; 1108 uint32_t len; 1109 struct mbuf *mreq, *mrep, *md, *mb; 1110 const int v3 = NFS_ISV3(vp); 1111 struct nfsnode *np = VTONFS(vp); 1112 #ifndef NFS_V2_ONLY 1113 int attrflag; 1114 #endif 1115 1116 nfsstats.rpccnt[NFSPROC_READLINK]++; 1117 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1118 nfsm_fhtom(np, v3); 1119 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1120 #ifndef NFS_V2_ONLY 1121 if (v3) 1122 nfsm_postop_attr(vp, attrflag, 0); 1123 #endif 1124 if (!error) { 1125 #ifndef NFS_V2_ONLY 1126 if (v3) { 1127 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1128 len = fxdr_unsigned(uint32_t, *tl); 1129 if (len > NFS_MAXPATHLEN) { 1130 /* 1131 * this pathname is too long for us. 1132 */ 1133 m_freem(mrep); 1134 /* Solaris returns EINVAL. should we follow? */ 1135 error = ENAMETOOLONG; 1136 goto nfsmout; 1137 } 1138 } else 1139 #endif 1140 { 1141 nfsm_strsiz(len, NFS_MAXPATHLEN); 1142 } 1143 nfsm_mtouio(uiop, len); 1144 } 1145 nfsm_reqdone; 1146 return (error); 1147 } 1148 1149 /* 1150 * nfs read rpc call 1151 * Ditto above 1152 */ 1153 int 1154 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1155 { 1156 u_int32_t *tl; 1157 char *cp; 1158 int32_t t1, t2; 1159 char *bpos, *dpos, *cp2; 1160 struct mbuf *mreq, *mrep, *md, *mb; 1161 struct nfsmount *nmp; 1162 int error = 0, len, retlen, tsiz, eof, byte_count; 1163 const int v3 = NFS_ISV3(vp); 1164 struct nfsnode *np = VTONFS(vp); 1165 #ifndef NFS_V2_ONLY 1166 int attrflag; 1167 #endif 1168 1169 #ifndef nolint 1170 eof = 0; 1171 #endif 1172 nmp = VFSTONFS(vp->v_mount); 1173 tsiz = uiop->uio_resid; 1174 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1175 return (EFBIG); 1176 iostat_busy(nmp->nm_stats); 1177 byte_count = 0; /* count bytes actually transferred */ 1178 while (tsiz > 0) { 1179 nfsstats.rpccnt[NFSPROC_READ]++; 1180 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1181 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1182 nfsm_fhtom(np, v3); 1183 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1184 #ifndef NFS_V2_ONLY 1185 if (v3) { 1186 txdr_hyper(uiop->uio_offset, tl); 1187 *(tl + 2) = txdr_unsigned(len); 1188 } else 1189 #endif 1190 { 1191 *tl++ = txdr_unsigned(uiop->uio_offset); 1192 *tl++ = txdr_unsigned(len); 1193 *tl = 0; 1194 } 1195 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1196 #ifndef NFS_V2_ONLY 1197 if (v3) { 1198 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1199 if (error) { 1200 m_freem(mrep); 1201 goto nfsmout; 1202 } 1203 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1204 eof = fxdr_unsigned(int, *(tl + 1)); 1205 } else 1206 #endif 1207 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1208 nfsm_strsiz(retlen, nmp->nm_rsize); 1209 nfsm_mtouio(uiop, retlen); 1210 m_freem(mrep); 1211 tsiz -= retlen; 1212 byte_count += retlen; 1213 #ifndef NFS_V2_ONLY 1214 if (v3) { 1215 if (eof || retlen == 0) 1216 tsiz = 0; 1217 } else 1218 #endif 1219 if (retlen < len) 1220 tsiz = 0; 1221 } 1222 nfsmout: 1223 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1224 return (error); 1225 } 1226 1227 struct nfs_writerpc_context { 1228 kmutex_t nwc_lock; 1229 kcondvar_t nwc_cv; 1230 int nwc_mbufcount; 1231 }; 1232 1233 /* 1234 * free mbuf used to refer protected pages while write rpc call. 1235 * called at splvm. 1236 */ 1237 static void 1238 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1239 { 1240 struct nfs_writerpc_context *ctx = arg; 1241 1242 KASSERT(m != NULL); 1243 KASSERT(ctx != NULL); 1244 pool_cache_put(mb_cache, m); 1245 mutex_enter(&ctx->nwc_lock); 1246 if (--ctx->nwc_mbufcount == 0) { 1247 cv_signal(&ctx->nwc_cv); 1248 } 1249 mutex_exit(&ctx->nwc_lock); 1250 } 1251 1252 /* 1253 * nfs write call 1254 */ 1255 int 1256 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1257 { 1258 u_int32_t *tl; 1259 char *cp; 1260 int32_t t1, t2; 1261 char *bpos, *dpos; 1262 struct mbuf *mreq, *mrep, *md, *mb; 1263 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1264 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1265 const int v3 = NFS_ISV3(vp); 1266 int committed = NFSV3WRITE_FILESYNC; 1267 struct nfsnode *np = VTONFS(vp); 1268 struct nfs_writerpc_context ctx; 1269 int byte_count; 1270 size_t origresid; 1271 #ifndef NFS_V2_ONLY 1272 char *cp2; 1273 int rlen, commit; 1274 #endif 1275 1276 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1277 panic("writerpc readonly vp %p", vp); 1278 } 1279 1280 #ifdef DIAGNOSTIC 1281 if (uiop->uio_iovcnt != 1) 1282 panic("nfs: writerpc iovcnt > 1"); 1283 #endif 1284 tsiz = uiop->uio_resid; 1285 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1286 return EFBIG; 1287 1288 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1289 cv_init(&ctx.nwc_cv, "nfsmblk"); 1290 ctx.nwc_mbufcount = 1; 1291 1292 retry: 1293 origresid = uiop->uio_resid; 1294 KASSERT(origresid == uiop->uio_iov->iov_len); 1295 iostat_busy(nmp->nm_stats); 1296 byte_count = 0; /* count of bytes actually written */ 1297 while (tsiz > 0) { 1298 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1299 uint32_t backup; 1300 bool stalewriteverf = false; 1301 1302 nfsstats.rpccnt[NFSPROC_WRITE]++; 1303 len = min(tsiz, nmp->nm_wsize); 1304 datalen = pageprotected ? 0 : nfsm_rndup(len); 1305 nfsm_reqhead(np, NFSPROC_WRITE, 1306 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1307 nfsm_fhtom(np, v3); 1308 #ifndef NFS_V2_ONLY 1309 if (v3) { 1310 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1311 txdr_hyper(uiop->uio_offset, tl); 1312 tl += 2; 1313 *tl++ = txdr_unsigned(len); 1314 *tl++ = txdr_unsigned(*iomode); 1315 *tl = txdr_unsigned(len); 1316 } else 1317 #endif 1318 { 1319 u_int32_t x; 1320 1321 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1322 /* Set both "begin" and "current" to non-garbage. */ 1323 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1324 *tl++ = x; /* "begin offset" */ 1325 *tl++ = x; /* "current offset" */ 1326 x = txdr_unsigned(len); 1327 *tl++ = x; /* total to this offset */ 1328 *tl = x; /* size of this write */ 1329 1330 } 1331 if (pageprotected) { 1332 /* 1333 * since we know pages can't be modified during i/o, 1334 * no need to copy them for us. 1335 */ 1336 struct mbuf *m; 1337 struct iovec *iovp = uiop->uio_iov; 1338 1339 m = m_get(M_WAIT, MT_DATA); 1340 MCLAIM(m, &nfs_mowner); 1341 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1342 nfs_writerpc_extfree, &ctx); 1343 m->m_flags |= M_EXT_ROMAP; 1344 m->m_len = len; 1345 mb->m_next = m; 1346 /* 1347 * no need to maintain mb and bpos here 1348 * because no one care them later. 1349 */ 1350 #if 0 1351 mb = m; 1352 bpos = mtod(void *, mb) + mb->m_len; 1353 #endif 1354 UIO_ADVANCE(uiop, len); 1355 uiop->uio_offset += len; 1356 mutex_enter(&ctx.nwc_lock); 1357 ctx.nwc_mbufcount++; 1358 mutex_exit(&ctx.nwc_lock); 1359 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1360 } else { 1361 nfsm_uiotom(uiop, len); 1362 } 1363 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1364 #ifndef NFS_V2_ONLY 1365 if (v3) { 1366 wccflag = NFSV3_WCCCHK; 1367 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1368 if (!error) { 1369 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1370 + NFSX_V3WRITEVERF); 1371 rlen = fxdr_unsigned(int, *tl++); 1372 if (rlen == 0) { 1373 error = NFSERR_IO; 1374 m_freem(mrep); 1375 break; 1376 } else if (rlen < len) { 1377 backup = len - rlen; 1378 UIO_ADVANCE(uiop, -backup); 1379 uiop->uio_offset -= backup; 1380 len = rlen; 1381 } 1382 commit = fxdr_unsigned(int, *tl++); 1383 1384 /* 1385 * Return the lowest committment level 1386 * obtained by any of the RPCs. 1387 */ 1388 if (committed == NFSV3WRITE_FILESYNC) 1389 committed = commit; 1390 else if (committed == NFSV3WRITE_DATASYNC && 1391 commit == NFSV3WRITE_UNSTABLE) 1392 committed = commit; 1393 mutex_enter(&nmp->nm_lock); 1394 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1395 memcpy(nmp->nm_writeverf, tl, 1396 NFSX_V3WRITEVERF); 1397 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1398 } else if ((nmp->nm_iflag & 1399 NFSMNT_STALEWRITEVERF) || 1400 memcmp(tl, nmp->nm_writeverf, 1401 NFSX_V3WRITEVERF)) { 1402 memcpy(nmp->nm_writeverf, tl, 1403 NFSX_V3WRITEVERF); 1404 /* 1405 * note NFSMNT_STALEWRITEVERF 1406 * if we're the first thread to 1407 * notice it. 1408 */ 1409 if ((nmp->nm_iflag & 1410 NFSMNT_STALEWRITEVERF) == 0) { 1411 stalewriteverf = true; 1412 nmp->nm_iflag |= 1413 NFSMNT_STALEWRITEVERF; 1414 } 1415 } 1416 mutex_exit(&nmp->nm_lock); 1417 } 1418 } else 1419 #endif 1420 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1421 if (wccflag) 1422 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1423 m_freem(mrep); 1424 if (error) 1425 break; 1426 tsiz -= len; 1427 byte_count += len; 1428 if (stalewriteverf) { 1429 *stalewriteverfp = true; 1430 stalewriteverf = false; 1431 if (committed == NFSV3WRITE_UNSTABLE && 1432 len != origresid) { 1433 /* 1434 * if our write requests weren't atomic but 1435 * unstable, datas in previous iterations 1436 * might have already been lost now. 1437 * then, we should resend them to nfsd. 1438 */ 1439 backup = origresid - tsiz; 1440 UIO_ADVANCE(uiop, -backup); 1441 uiop->uio_offset -= backup; 1442 tsiz = origresid; 1443 goto retry; 1444 } 1445 } 1446 } 1447 nfsmout: 1448 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1449 if (pageprotected) { 1450 /* 1451 * wait until mbufs go away. 1452 * retransmitted mbufs can survive longer than rpc requests 1453 * themselves. 1454 */ 1455 mutex_enter(&ctx.nwc_lock); 1456 ctx.nwc_mbufcount--; 1457 while (ctx.nwc_mbufcount > 0) { 1458 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1459 } 1460 mutex_exit(&ctx.nwc_lock); 1461 } 1462 mutex_destroy(&ctx.nwc_lock); 1463 cv_destroy(&ctx.nwc_cv); 1464 *iomode = committed; 1465 if (error) 1466 uiop->uio_resid = tsiz; 1467 return (error); 1468 } 1469 1470 /* 1471 * nfs mknod rpc 1472 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1473 * mode set to specify the file type and the size field for rdev. 1474 */ 1475 int 1476 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1477 { 1478 struct nfsv2_sattr *sp; 1479 u_int32_t *tl; 1480 char *cp; 1481 int32_t t1, t2; 1482 struct vnode *newvp = (struct vnode *)0; 1483 struct nfsnode *dnp, *np; 1484 char *cp2; 1485 char *bpos, *dpos; 1486 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1487 struct mbuf *mreq, *mrep, *md, *mb; 1488 u_int32_t rdev; 1489 const int v3 = NFS_ISV3(dvp); 1490 1491 if (vap->va_type == VCHR || vap->va_type == VBLK) 1492 rdev = txdr_unsigned(vap->va_rdev); 1493 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1494 rdev = nfs_xdrneg1; 1495 else { 1496 VOP_ABORTOP(dvp, cnp); 1497 vput(dvp); 1498 return (EOPNOTSUPP); 1499 } 1500 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1501 dnp = VTONFS(dvp); 1502 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1503 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1504 nfsm_fhtom(dnp, v3); 1505 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1506 #ifndef NFS_V2_ONLY 1507 if (v3) { 1508 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1509 *tl++ = vtonfsv3_type(vap->va_type); 1510 nfsm_v3attrbuild(vap, false); 1511 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1512 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1513 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1514 *tl = txdr_unsigned(minor(vap->va_rdev)); 1515 } 1516 } else 1517 #endif 1518 { 1519 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1520 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1521 sp->sa_uid = nfs_xdrneg1; 1522 sp->sa_gid = nfs_xdrneg1; 1523 sp->sa_size = rdev; 1524 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1525 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1526 } 1527 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1528 if (!error) { 1529 nfsm_mtofh(dvp, newvp, v3, gotvp); 1530 if (!gotvp) { 1531 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1532 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1533 if (!error) 1534 newvp = NFSTOV(np); 1535 } 1536 } 1537 #ifndef NFS_V2_ONLY 1538 if (v3) 1539 nfsm_wcc_data(dvp, wccflag, 0, !error); 1540 #endif 1541 nfsm_reqdone; 1542 if (error) { 1543 if (newvp) 1544 vput(newvp); 1545 } else { 1546 nfs_cache_enter(dvp, newvp, cnp); 1547 *vpp = newvp; 1548 } 1549 VTONFS(dvp)->n_flag |= NMODIFIED; 1550 if (!wccflag) 1551 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1552 vput(dvp); 1553 return (error); 1554 } 1555 1556 /* 1557 * nfs mknod vop 1558 * just call nfs_mknodrpc() to do the work. 1559 */ 1560 /* ARGSUSED */ 1561 int 1562 nfs_mknod(void *v) 1563 { 1564 struct vop_mknod_args /* { 1565 struct vnode *a_dvp; 1566 struct vnode **a_vpp; 1567 struct componentname *a_cnp; 1568 struct vattr *a_vap; 1569 } */ *ap = v; 1570 struct vnode *dvp = ap->a_dvp; 1571 struct componentname *cnp = ap->a_cnp; 1572 int error; 1573 1574 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1575 VN_KNOTE(dvp, NOTE_WRITE); 1576 if (error == 0 || error == EEXIST) 1577 cache_purge1(dvp, cnp, 0); 1578 return (error); 1579 } 1580 1581 /* 1582 * nfs file create call 1583 */ 1584 int 1585 nfs_create(void *v) 1586 { 1587 struct vop_create_args /* { 1588 struct vnode *a_dvp; 1589 struct vnode **a_vpp; 1590 struct componentname *a_cnp; 1591 struct vattr *a_vap; 1592 } */ *ap = v; 1593 struct vnode *dvp = ap->a_dvp; 1594 struct vattr *vap = ap->a_vap; 1595 struct componentname *cnp = ap->a_cnp; 1596 struct nfsv2_sattr *sp; 1597 u_int32_t *tl; 1598 char *cp; 1599 int32_t t1, t2; 1600 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1601 struct vnode *newvp = (struct vnode *)0; 1602 char *bpos, *dpos, *cp2; 1603 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1604 struct mbuf *mreq, *mrep, *md, *mb; 1605 const int v3 = NFS_ISV3(dvp); 1606 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1607 1608 /* 1609 * Oops, not for me.. 1610 */ 1611 if (vap->va_type == VSOCK) 1612 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1613 1614 KASSERT(vap->va_type == VREG); 1615 1616 #ifdef VA_EXCLUSIVE 1617 if (vap->va_vaflags & VA_EXCLUSIVE) { 1618 excl_mode = NFSV3CREATE_EXCLUSIVE; 1619 } 1620 #endif 1621 again: 1622 error = 0; 1623 nfsstats.rpccnt[NFSPROC_CREATE]++; 1624 dnp = VTONFS(dvp); 1625 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1626 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1627 nfsm_fhtom(dnp, v3); 1628 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1629 #ifndef NFS_V2_ONLY 1630 if (v3) { 1631 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1632 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1633 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1634 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1635 *tl++ = cprng_fast32(); 1636 *tl = cprng_fast32(); 1637 } else { 1638 *tl = txdr_unsigned(excl_mode); 1639 nfsm_v3attrbuild(vap, false); 1640 } 1641 } else 1642 #endif 1643 { 1644 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1645 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1646 sp->sa_uid = nfs_xdrneg1; 1647 sp->sa_gid = nfs_xdrneg1; 1648 sp->sa_size = 0; 1649 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1650 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1651 } 1652 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1653 if (!error) { 1654 nfsm_mtofh(dvp, newvp, v3, gotvp); 1655 if (!gotvp) { 1656 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1657 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1658 if (!error) 1659 newvp = NFSTOV(np); 1660 } 1661 } 1662 #ifndef NFS_V2_ONLY 1663 if (v3) 1664 nfsm_wcc_data(dvp, wccflag, 0, !error); 1665 #endif 1666 nfsm_reqdone; 1667 if (error) { 1668 /* 1669 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1670 */ 1671 if (v3 && error == ENOTSUP) { 1672 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1673 excl_mode = NFSV3CREATE_GUARDED; 1674 goto again; 1675 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1676 excl_mode = NFSV3CREATE_UNCHECKED; 1677 goto again; 1678 } 1679 } 1680 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1681 struct timespec ts; 1682 1683 getnanotime(&ts); 1684 1685 /* 1686 * make sure that we'll update timestamps as 1687 * most server implementations use them to store 1688 * the create verifier. 1689 * 1690 * XXX it's better to use TOSERVER always. 1691 */ 1692 1693 if (vap->va_atime.tv_sec == VNOVAL) 1694 vap->va_atime = ts; 1695 if (vap->va_mtime.tv_sec == VNOVAL) 1696 vap->va_mtime = ts; 1697 1698 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1699 } 1700 if (error == 0) { 1701 if (cnp->cn_flags & MAKEENTRY) 1702 nfs_cache_enter(dvp, newvp, cnp); 1703 else 1704 cache_purge1(dvp, cnp, 0); 1705 *ap->a_vpp = newvp; 1706 } else { 1707 if (newvp) 1708 vput(newvp); 1709 if (error == EEXIST) 1710 cache_purge1(dvp, cnp, 0); 1711 } 1712 VTONFS(dvp)->n_flag |= NMODIFIED; 1713 if (!wccflag) 1714 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1715 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1716 vput(dvp); 1717 return (error); 1718 } 1719 1720 /* 1721 * nfs file remove call 1722 * To try and make nfs semantics closer to ufs semantics, a file that has 1723 * other processes using the vnode is renamed instead of removed and then 1724 * removed later on the last close. 1725 * - If v_usecount > 1 1726 * If a rename is not already in the works 1727 * call nfs_sillyrename() to set it up 1728 * else 1729 * do the remove rpc 1730 */ 1731 int 1732 nfs_remove(void *v) 1733 { 1734 struct vop_remove_args /* { 1735 struct vnodeop_desc *a_desc; 1736 struct vnode * a_dvp; 1737 struct vnode * a_vp; 1738 struct componentname * a_cnp; 1739 } */ *ap = v; 1740 struct vnode *vp = ap->a_vp; 1741 struct vnode *dvp = ap->a_dvp; 1742 struct componentname *cnp = ap->a_cnp; 1743 struct nfsnode *np = VTONFS(vp); 1744 int error = 0; 1745 struct vattr vattr; 1746 1747 #ifndef DIAGNOSTIC 1748 if (vp->v_usecount < 1) 1749 panic("nfs_remove: bad v_usecount"); 1750 #endif 1751 if (vp->v_type == VDIR) 1752 error = EPERM; 1753 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1754 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1755 vattr.va_nlink > 1)) { 1756 /* 1757 * Purge the name cache so that the chance of a lookup for 1758 * the name succeeding while the remove is in progress is 1759 * minimized. Without node locking it can still happen, such 1760 * that an I/O op returns ESTALE, but since you get this if 1761 * another host removes the file.. 1762 */ 1763 cache_purge(vp); 1764 /* 1765 * throw away biocache buffers, mainly to avoid 1766 * unnecessary delayed writes later. 1767 */ 1768 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1769 /* Do the rpc */ 1770 if (error != EINTR) 1771 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1772 cnp->cn_namelen, cnp->cn_cred, curlwp); 1773 } else if (!np->n_sillyrename) 1774 error = nfs_sillyrename(dvp, vp, cnp, false); 1775 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1776 vattr.va_nlink == 1) { 1777 np->n_flag |= NREMOVED; 1778 } 1779 NFS_INVALIDATE_ATTRCACHE(np); 1780 VN_KNOTE(vp, NOTE_DELETE); 1781 VN_KNOTE(dvp, NOTE_WRITE); 1782 if (dvp == vp) 1783 vrele(vp); 1784 else 1785 vput(vp); 1786 vput(dvp); 1787 return (error); 1788 } 1789 1790 /* 1791 * nfs file remove rpc called from nfs_inactive 1792 */ 1793 int 1794 nfs_removeit(struct sillyrename *sp) 1795 { 1796 1797 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1798 (struct lwp *)0)); 1799 } 1800 1801 /* 1802 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1803 */ 1804 int 1805 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1806 { 1807 u_int32_t *tl; 1808 char *cp; 1809 #ifndef NFS_V2_ONLY 1810 int32_t t1; 1811 char *cp2; 1812 #endif 1813 int32_t t2; 1814 char *bpos, *dpos; 1815 int error = 0, wccflag = NFSV3_WCCRATTR; 1816 struct mbuf *mreq, *mrep, *md, *mb; 1817 const int v3 = NFS_ISV3(dvp); 1818 int rexmit = 0; 1819 struct nfsnode *dnp = VTONFS(dvp); 1820 1821 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1822 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1823 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1824 nfsm_fhtom(dnp, v3); 1825 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1826 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1827 #ifndef NFS_V2_ONLY 1828 if (v3) 1829 nfsm_wcc_data(dvp, wccflag, 0, !error); 1830 #endif 1831 nfsm_reqdone; 1832 VTONFS(dvp)->n_flag |= NMODIFIED; 1833 if (!wccflag) 1834 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1835 /* 1836 * Kludge City: If the first reply to the remove rpc is lost.. 1837 * the reply to the retransmitted request will be ENOENT 1838 * since the file was in fact removed 1839 * Therefore, we cheat and return success. 1840 */ 1841 if (rexmit && error == ENOENT) 1842 error = 0; 1843 return (error); 1844 } 1845 1846 /* 1847 * nfs file rename call 1848 */ 1849 int 1850 nfs_rename(void *v) 1851 { 1852 struct vop_rename_args /* { 1853 struct vnode *a_fdvp; 1854 struct vnode *a_fvp; 1855 struct componentname *a_fcnp; 1856 struct vnode *a_tdvp; 1857 struct vnode *a_tvp; 1858 struct componentname *a_tcnp; 1859 } */ *ap = v; 1860 struct vnode *fvp = ap->a_fvp; 1861 struct vnode *tvp = ap->a_tvp; 1862 struct vnode *fdvp = ap->a_fdvp; 1863 struct vnode *tdvp = ap->a_tdvp; 1864 struct componentname *tcnp = ap->a_tcnp; 1865 struct componentname *fcnp = ap->a_fcnp; 1866 int error; 1867 1868 /* Check for cross-device rename */ 1869 if ((fvp->v_mount != tdvp->v_mount) || 1870 (tvp && (fvp->v_mount != tvp->v_mount))) { 1871 error = EXDEV; 1872 goto out; 1873 } 1874 1875 /* 1876 * If the tvp exists and is in use, sillyrename it before doing the 1877 * rename of the new file over it. 1878 * 1879 * Have sillyrename use link instead of rename if possible, 1880 * so that we don't lose the file if the rename fails, and so 1881 * that there's no window when the "to" file doesn't exist. 1882 */ 1883 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1884 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1885 VN_KNOTE(tvp, NOTE_DELETE); 1886 vput(tvp); 1887 tvp = NULL; 1888 } 1889 1890 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1891 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1892 curlwp); 1893 1894 VN_KNOTE(fdvp, NOTE_WRITE); 1895 VN_KNOTE(tdvp, NOTE_WRITE); 1896 if (error == 0 || error == EEXIST) { 1897 if (fvp->v_type == VDIR) 1898 cache_purge(fvp); 1899 else 1900 cache_purge1(fdvp, fcnp, 0); 1901 if (tvp != NULL && tvp->v_type == VDIR) 1902 cache_purge(tvp); 1903 else 1904 cache_purge1(tdvp, tcnp, 0); 1905 } 1906 out: 1907 if (tdvp == tvp) 1908 vrele(tdvp); 1909 else 1910 vput(tdvp); 1911 if (tvp) 1912 vput(tvp); 1913 vrele(fdvp); 1914 vrele(fvp); 1915 return (error); 1916 } 1917 1918 /* 1919 * nfs file rename rpc called from nfs_remove() above 1920 */ 1921 int 1922 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1923 { 1924 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1925 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1926 } 1927 1928 /* 1929 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1930 */ 1931 int 1932 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) 1933 { 1934 u_int32_t *tl; 1935 char *cp; 1936 #ifndef NFS_V2_ONLY 1937 int32_t t1; 1938 char *cp2; 1939 #endif 1940 int32_t t2; 1941 char *bpos, *dpos; 1942 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1943 struct mbuf *mreq, *mrep, *md, *mb; 1944 const int v3 = NFS_ISV3(fdvp); 1945 int rexmit = 0; 1946 struct nfsnode *fdnp = VTONFS(fdvp); 1947 1948 nfsstats.rpccnt[NFSPROC_RENAME]++; 1949 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1950 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1951 nfsm_rndup(tnamelen)); 1952 nfsm_fhtom(fdnp, v3); 1953 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1954 nfsm_fhtom(VTONFS(tdvp), v3); 1955 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1956 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1957 #ifndef NFS_V2_ONLY 1958 if (v3) { 1959 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1960 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1961 } 1962 #endif 1963 nfsm_reqdone; 1964 VTONFS(fdvp)->n_flag |= NMODIFIED; 1965 VTONFS(tdvp)->n_flag |= NMODIFIED; 1966 if (!fwccflag) 1967 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1968 if (!twccflag) 1969 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1970 /* 1971 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1972 */ 1973 if (rexmit && error == ENOENT) 1974 error = 0; 1975 return (error); 1976 } 1977 1978 /* 1979 * NFS link RPC, called from nfs_link. 1980 * Assumes dvp and vp locked, and leaves them that way. 1981 */ 1982 1983 static int 1984 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 1985 size_t namelen, kauth_cred_t cred, struct lwp *l) 1986 { 1987 u_int32_t *tl; 1988 char *cp; 1989 #ifndef NFS_V2_ONLY 1990 int32_t t1; 1991 char *cp2; 1992 #endif 1993 int32_t t2; 1994 char *bpos, *dpos; 1995 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1996 struct mbuf *mreq, *mrep, *md, *mb; 1997 const int v3 = NFS_ISV3(dvp); 1998 int rexmit = 0; 1999 struct nfsnode *np = VTONFS(vp); 2000 2001 nfsstats.rpccnt[NFSPROC_LINK]++; 2002 nfsm_reqhead(np, NFSPROC_LINK, 2003 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2004 nfsm_fhtom(np, v3); 2005 nfsm_fhtom(VTONFS(dvp), v3); 2006 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2007 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2008 #ifndef NFS_V2_ONLY 2009 if (v3) { 2010 nfsm_postop_attr(vp, attrflag, 0); 2011 nfsm_wcc_data(dvp, wccflag, 0, !error); 2012 } 2013 #endif 2014 nfsm_reqdone; 2015 2016 VTONFS(dvp)->n_flag |= NMODIFIED; 2017 if (!attrflag) 2018 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2019 if (!wccflag) 2020 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2021 2022 /* 2023 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2024 */ 2025 if (rexmit && error == EEXIST) 2026 error = 0; 2027 2028 return error; 2029 } 2030 2031 /* 2032 * nfs hard link create call 2033 */ 2034 int 2035 nfs_link(void *v) 2036 { 2037 struct vop_link_args /* { 2038 struct vnode *a_dvp; 2039 struct vnode *a_vp; 2040 struct componentname *a_cnp; 2041 } */ *ap = v; 2042 struct vnode *vp = ap->a_vp; 2043 struct vnode *dvp = ap->a_dvp; 2044 struct componentname *cnp = ap->a_cnp; 2045 int error = 0; 2046 2047 error = vn_lock(vp, LK_EXCLUSIVE); 2048 if (error != 0) { 2049 VOP_ABORTOP(dvp, cnp); 2050 vput(dvp); 2051 return error; 2052 } 2053 2054 /* 2055 * Push all writes to the server, so that the attribute cache 2056 * doesn't get "out of sync" with the server. 2057 * XXX There should be a better way! 2058 */ 2059 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2060 2061 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2062 cnp->cn_cred, curlwp); 2063 2064 if (error == 0) { 2065 cache_purge1(dvp, cnp, 0); 2066 } 2067 VOP_UNLOCK(vp); 2068 VN_KNOTE(vp, NOTE_LINK); 2069 VN_KNOTE(dvp, NOTE_WRITE); 2070 vput(dvp); 2071 return (error); 2072 } 2073 2074 /* 2075 * nfs symbolic link create call 2076 */ 2077 int 2078 nfs_symlink(void *v) 2079 { 2080 struct vop_symlink_args /* { 2081 struct vnode *a_dvp; 2082 struct vnode **a_vpp; 2083 struct componentname *a_cnp; 2084 struct vattr *a_vap; 2085 char *a_target; 2086 } */ *ap = v; 2087 struct vnode *dvp = ap->a_dvp; 2088 struct vattr *vap = ap->a_vap; 2089 struct componentname *cnp = ap->a_cnp; 2090 struct nfsv2_sattr *sp; 2091 u_int32_t *tl; 2092 char *cp; 2093 int32_t t1, t2; 2094 char *bpos, *dpos, *cp2; 2095 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2096 struct mbuf *mreq, *mrep, *md, *mb; 2097 struct vnode *newvp = (struct vnode *)0; 2098 const int v3 = NFS_ISV3(dvp); 2099 int rexmit = 0; 2100 struct nfsnode *dnp = VTONFS(dvp); 2101 2102 *ap->a_vpp = NULL; 2103 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2104 slen = strlen(ap->a_target); 2105 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2106 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2107 nfsm_fhtom(dnp, v3); 2108 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2109 #ifndef NFS_V2_ONlY 2110 if (v3) 2111 nfsm_v3attrbuild(vap, false); 2112 #endif 2113 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2114 #ifndef NFS_V2_ONlY 2115 if (!v3) { 2116 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2117 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2118 sp->sa_uid = nfs_xdrneg1; 2119 sp->sa_gid = nfs_xdrneg1; 2120 sp->sa_size = nfs_xdrneg1; 2121 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2122 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2123 } 2124 #endif 2125 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2126 &rexmit); 2127 #ifndef NFS_V2_ONlY 2128 if (v3) { 2129 if (!error) 2130 nfsm_mtofh(dvp, newvp, v3, gotvp); 2131 nfsm_wcc_data(dvp, wccflag, 0, !error); 2132 } 2133 #endif 2134 nfsm_reqdone; 2135 /* 2136 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2137 */ 2138 if (rexmit && error == EEXIST) 2139 error = 0; 2140 if (error == 0 || error == EEXIST) 2141 cache_purge1(dvp, cnp, 0); 2142 if (error == 0 && newvp == NULL) { 2143 struct nfsnode *np = NULL; 2144 2145 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2146 cnp->cn_cred, curlwp, &np); 2147 if (error == 0) 2148 newvp = NFSTOV(np); 2149 } 2150 if (error) { 2151 if (newvp != NULL) 2152 vput(newvp); 2153 } else { 2154 *ap->a_vpp = newvp; 2155 } 2156 VTONFS(dvp)->n_flag |= NMODIFIED; 2157 if (!wccflag) 2158 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2159 VN_KNOTE(dvp, NOTE_WRITE); 2160 vput(dvp); 2161 return (error); 2162 } 2163 2164 /* 2165 * nfs make dir call 2166 */ 2167 int 2168 nfs_mkdir(void *v) 2169 { 2170 struct vop_mkdir_args /* { 2171 struct vnode *a_dvp; 2172 struct vnode **a_vpp; 2173 struct componentname *a_cnp; 2174 struct vattr *a_vap; 2175 } */ *ap = v; 2176 struct vnode *dvp = ap->a_dvp; 2177 struct vattr *vap = ap->a_vap; 2178 struct componentname *cnp = ap->a_cnp; 2179 struct nfsv2_sattr *sp; 2180 u_int32_t *tl; 2181 char *cp; 2182 int32_t t1, t2; 2183 int len; 2184 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2185 struct vnode *newvp = (struct vnode *)0; 2186 char *bpos, *dpos, *cp2; 2187 int error = 0, wccflag = NFSV3_WCCRATTR; 2188 int gotvp = 0; 2189 int rexmit = 0; 2190 struct mbuf *mreq, *mrep, *md, *mb; 2191 const int v3 = NFS_ISV3(dvp); 2192 2193 len = cnp->cn_namelen; 2194 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2195 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2196 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2197 nfsm_fhtom(dnp, v3); 2198 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2199 #ifndef NFS_V2_ONLY 2200 if (v3) { 2201 nfsm_v3attrbuild(vap, false); 2202 } else 2203 #endif 2204 { 2205 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2206 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2207 sp->sa_uid = nfs_xdrneg1; 2208 sp->sa_gid = nfs_xdrneg1; 2209 sp->sa_size = nfs_xdrneg1; 2210 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2211 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2212 } 2213 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2214 if (!error) 2215 nfsm_mtofh(dvp, newvp, v3, gotvp); 2216 if (v3) 2217 nfsm_wcc_data(dvp, wccflag, 0, !error); 2218 nfsm_reqdone; 2219 VTONFS(dvp)->n_flag |= NMODIFIED; 2220 if (!wccflag) 2221 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2222 /* 2223 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2224 * if we can succeed in looking up the directory. 2225 */ 2226 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2227 if (newvp) { 2228 vput(newvp); 2229 newvp = (struct vnode *)0; 2230 } 2231 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2232 curlwp, &np); 2233 if (!error) { 2234 newvp = NFSTOV(np); 2235 if (newvp->v_type != VDIR || newvp == dvp) 2236 error = EEXIST; 2237 } 2238 } 2239 if (error) { 2240 if (newvp) { 2241 if (dvp != newvp) 2242 vput(newvp); 2243 else 2244 vrele(newvp); 2245 } 2246 } else { 2247 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2248 nfs_cache_enter(dvp, newvp, cnp); 2249 *ap->a_vpp = newvp; 2250 } 2251 vput(dvp); 2252 return (error); 2253 } 2254 2255 /* 2256 * nfs remove directory call 2257 */ 2258 int 2259 nfs_rmdir(void *v) 2260 { 2261 struct vop_rmdir_args /* { 2262 struct vnode *a_dvp; 2263 struct vnode *a_vp; 2264 struct componentname *a_cnp; 2265 } */ *ap = v; 2266 struct vnode *vp = ap->a_vp; 2267 struct vnode *dvp = ap->a_dvp; 2268 struct componentname *cnp = ap->a_cnp; 2269 u_int32_t *tl; 2270 char *cp; 2271 #ifndef NFS_V2_ONLY 2272 int32_t t1; 2273 char *cp2; 2274 #endif 2275 int32_t t2; 2276 char *bpos, *dpos; 2277 int error = 0, wccflag = NFSV3_WCCRATTR; 2278 int rexmit = 0; 2279 struct mbuf *mreq, *mrep, *md, *mb; 2280 const int v3 = NFS_ISV3(dvp); 2281 struct nfsnode *dnp; 2282 2283 if (dvp == vp) { 2284 vrele(dvp); 2285 vput(dvp); 2286 return (EINVAL); 2287 } 2288 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2289 dnp = VTONFS(dvp); 2290 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2291 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2292 nfsm_fhtom(dnp, v3); 2293 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2294 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2295 #ifndef NFS_V2_ONLY 2296 if (v3) 2297 nfsm_wcc_data(dvp, wccflag, 0, !error); 2298 #endif 2299 nfsm_reqdone; 2300 VTONFS(dvp)->n_flag |= NMODIFIED; 2301 if (!wccflag) 2302 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2303 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2304 VN_KNOTE(vp, NOTE_DELETE); 2305 cache_purge(vp); 2306 vput(vp); 2307 vput(dvp); 2308 /* 2309 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2310 */ 2311 if (rexmit && error == ENOENT) 2312 error = 0; 2313 return (error); 2314 } 2315 2316 /* 2317 * nfs readdir call 2318 */ 2319 int 2320 nfs_readdir(void *v) 2321 { 2322 struct vop_readdir_args /* { 2323 struct vnode *a_vp; 2324 struct uio *a_uio; 2325 kauth_cred_t a_cred; 2326 int *a_eofflag; 2327 off_t **a_cookies; 2328 int *a_ncookies; 2329 } */ *ap = v; 2330 struct vnode *vp = ap->a_vp; 2331 struct uio *uio = ap->a_uio; 2332 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2333 char *base = uio->uio_iov->iov_base; 2334 int tresid, error; 2335 size_t count, lost; 2336 struct dirent *dp; 2337 off_t *cookies = NULL; 2338 int ncookies = 0, nc; 2339 2340 if (vp->v_type != VDIR) 2341 return (EPERM); 2342 2343 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2344 count = uio->uio_resid - lost; 2345 if (count <= 0) 2346 return (EINVAL); 2347 2348 /* 2349 * Call nfs_bioread() to do the real work. 2350 */ 2351 tresid = uio->uio_resid = count; 2352 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2353 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2354 2355 if (!error && ap->a_cookies) { 2356 ncookies = count / 16; 2357 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2358 *ap->a_cookies = cookies; 2359 } 2360 2361 if (!error && uio->uio_resid == tresid) { 2362 uio->uio_resid += lost; 2363 nfsstats.direofcache_misses++; 2364 if (ap->a_cookies) 2365 *ap->a_ncookies = 0; 2366 *ap->a_eofflag = 1; 2367 return (0); 2368 } 2369 2370 if (!error && ap->a_cookies) { 2371 /* 2372 * Only the NFS server and emulations use cookies, and they 2373 * load the directory block into system space, so we can 2374 * just look at it directly. 2375 */ 2376 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2377 uio->uio_iovcnt != 1) 2378 panic("nfs_readdir: lost in space"); 2379 for (nc = 0; ncookies-- && 2380 base < (char *)uio->uio_iov->iov_base; nc++){ 2381 dp = (struct dirent *) base; 2382 if (dp->d_reclen == 0) 2383 break; 2384 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2385 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2386 else 2387 *(cookies++) = NFS_GETCOOKIE(dp); 2388 base += dp->d_reclen; 2389 } 2390 uio->uio_resid += 2391 ((char *)uio->uio_iov->iov_base - base); 2392 uio->uio_iov->iov_len += 2393 ((char *)uio->uio_iov->iov_base - base); 2394 uio->uio_iov->iov_base = base; 2395 *ap->a_ncookies = nc; 2396 } 2397 2398 uio->uio_resid += lost; 2399 *ap->a_eofflag = 0; 2400 return (error); 2401 } 2402 2403 /* 2404 * Readdir rpc call. 2405 * Called from below the buffer cache by nfs_doio(). 2406 */ 2407 int 2408 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2409 { 2410 int len, left; 2411 struct dirent *dp = NULL; 2412 u_int32_t *tl; 2413 char *cp; 2414 int32_t t1, t2; 2415 char *bpos, *dpos, *cp2; 2416 struct mbuf *mreq, *mrep, *md, *mb; 2417 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2418 struct nfsnode *dnp = VTONFS(vp); 2419 u_quad_t fileno; 2420 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2421 #ifndef NFS_V2_ONLY 2422 int attrflag; 2423 #endif 2424 int nrpcs = 0, reclen; 2425 const int v3 = NFS_ISV3(vp); 2426 2427 #ifdef DIAGNOSTIC 2428 /* 2429 * Should be called from buffer cache, so only amount of 2430 * NFS_DIRBLKSIZ will be requested. 2431 */ 2432 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2433 panic("nfs readdirrpc bad uio"); 2434 #endif 2435 2436 /* 2437 * Loop around doing readdir rpc's of size nm_readdirsize 2438 * truncated to a multiple of NFS_DIRFRAGSIZ. 2439 * The stopping criteria is EOF or buffer full. 2440 */ 2441 while (more_dirs && bigenough) { 2442 /* 2443 * Heuristic: don't bother to do another RPC to further 2444 * fill up this block if there is not much room left. (< 50% 2445 * of the readdir RPC size). This wastes some buffer space 2446 * but can save up to 50% in RPC calls. 2447 */ 2448 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2449 bigenough = 0; 2450 break; 2451 } 2452 nfsstats.rpccnt[NFSPROC_READDIR]++; 2453 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2454 NFSX_READDIR(v3)); 2455 nfsm_fhtom(dnp, v3); 2456 #ifndef NFS_V2_ONLY 2457 if (v3) { 2458 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2459 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2460 txdr_swapcookie3(uiop->uio_offset, tl); 2461 } else { 2462 txdr_cookie3(uiop->uio_offset, tl); 2463 } 2464 tl += 2; 2465 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2466 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2467 } else 2468 #endif 2469 { 2470 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2471 *tl++ = txdr_unsigned(uiop->uio_offset); 2472 } 2473 *tl = txdr_unsigned(nmp->nm_readdirsize); 2474 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2475 nrpcs++; 2476 #ifndef NFS_V2_ONLY 2477 if (v3) { 2478 nfsm_postop_attr(vp, attrflag, 0); 2479 if (!error) { 2480 nfsm_dissect(tl, u_int32_t *, 2481 2 * NFSX_UNSIGNED); 2482 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2483 dnp->n_cookieverf.nfsuquad[1] = *tl; 2484 } else { 2485 m_freem(mrep); 2486 goto nfsmout; 2487 } 2488 } 2489 #endif 2490 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2491 more_dirs = fxdr_unsigned(int, *tl); 2492 2493 /* loop thru the dir entries, doctoring them to 4bsd form */ 2494 while (more_dirs && bigenough) { 2495 #ifndef NFS_V2_ONLY 2496 if (v3) { 2497 nfsm_dissect(tl, u_int32_t *, 2498 3 * NFSX_UNSIGNED); 2499 fileno = fxdr_hyper(tl); 2500 len = fxdr_unsigned(int, *(tl + 2)); 2501 } else 2502 #endif 2503 { 2504 nfsm_dissect(tl, u_int32_t *, 2505 2 * NFSX_UNSIGNED); 2506 fileno = fxdr_unsigned(u_quad_t, *tl++); 2507 len = fxdr_unsigned(int, *tl); 2508 } 2509 if (len <= 0 || len > NFS_MAXNAMLEN) { 2510 error = EBADRPC; 2511 m_freem(mrep); 2512 goto nfsmout; 2513 } 2514 /* for cookie stashing */ 2515 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2516 left = NFS_DIRFRAGSIZ - blksiz; 2517 if (reclen > left) { 2518 memset(uiop->uio_iov->iov_base, 0, left); 2519 dp->d_reclen += left; 2520 UIO_ADVANCE(uiop, left); 2521 blksiz = 0; 2522 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2523 } 2524 if (reclen > uiop->uio_resid) 2525 bigenough = 0; 2526 if (bigenough) { 2527 int tlen; 2528 2529 dp = (struct dirent *)uiop->uio_iov->iov_base; 2530 dp->d_fileno = fileno; 2531 dp->d_namlen = len; 2532 dp->d_reclen = reclen; 2533 dp->d_type = DT_UNKNOWN; 2534 blksiz += reclen; 2535 if (blksiz == NFS_DIRFRAGSIZ) 2536 blksiz = 0; 2537 UIO_ADVANCE(uiop, DIRHDSIZ); 2538 nfsm_mtouio(uiop, len); 2539 tlen = reclen - (DIRHDSIZ + len); 2540 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2541 UIO_ADVANCE(uiop, tlen); 2542 } else 2543 nfsm_adv(nfsm_rndup(len)); 2544 #ifndef NFS_V2_ONLY 2545 if (v3) { 2546 nfsm_dissect(tl, u_int32_t *, 2547 3 * NFSX_UNSIGNED); 2548 } else 2549 #endif 2550 { 2551 nfsm_dissect(tl, u_int32_t *, 2552 2 * NFSX_UNSIGNED); 2553 } 2554 if (bigenough) { 2555 #ifndef NFS_V2_ONLY 2556 if (v3) { 2557 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2558 uiop->uio_offset = 2559 fxdr_swapcookie3(tl); 2560 else 2561 uiop->uio_offset = 2562 fxdr_cookie3(tl); 2563 } 2564 else 2565 #endif 2566 { 2567 uiop->uio_offset = 2568 fxdr_unsigned(off_t, *tl); 2569 } 2570 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2571 } 2572 if (v3) 2573 tl += 2; 2574 else 2575 tl++; 2576 more_dirs = fxdr_unsigned(int, *tl); 2577 } 2578 /* 2579 * If at end of rpc data, get the eof boolean 2580 */ 2581 if (!more_dirs) { 2582 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2583 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2584 2585 /* 2586 * kludge: if we got no entries, treat it as EOF. 2587 * some server sometimes send a reply without any 2588 * entries or EOF. 2589 * although it might mean the server has very long name, 2590 * we can't handle such entries anyway. 2591 */ 2592 2593 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2594 more_dirs = 0; 2595 } 2596 m_freem(mrep); 2597 } 2598 /* 2599 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2600 * by increasing d_reclen for the last record. 2601 */ 2602 if (blksiz > 0) { 2603 left = NFS_DIRFRAGSIZ - blksiz; 2604 memset(uiop->uio_iov->iov_base, 0, left); 2605 dp->d_reclen += left; 2606 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2607 UIO_ADVANCE(uiop, left); 2608 } 2609 2610 /* 2611 * We are now either at the end of the directory or have filled the 2612 * block. 2613 */ 2614 if (bigenough) { 2615 dnp->n_direofoffset = uiop->uio_offset; 2616 dnp->n_flag |= NEOFVALID; 2617 } 2618 nfsmout: 2619 return (error); 2620 } 2621 2622 #ifndef NFS_V2_ONLY 2623 /* 2624 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2625 */ 2626 int 2627 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2628 { 2629 int len, left; 2630 struct dirent *dp = NULL; 2631 u_int32_t *tl; 2632 char *cp; 2633 int32_t t1, t2; 2634 struct vnode *newvp; 2635 char *bpos, *dpos, *cp2; 2636 struct mbuf *mreq, *mrep, *md, *mb; 2637 struct nameidata nami, *ndp = &nami; 2638 struct componentname *cnp = &ndp->ni_cnd; 2639 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2640 struct nfsnode *dnp = VTONFS(vp), *np; 2641 nfsfh_t *fhp; 2642 u_quad_t fileno; 2643 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2644 int attrflag, fhsize, nrpcs = 0, reclen; 2645 struct nfs_fattr fattr, *fp; 2646 2647 #ifdef DIAGNOSTIC 2648 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2649 panic("nfs readdirplusrpc bad uio"); 2650 #endif 2651 ndp->ni_dvp = vp; 2652 newvp = NULLVP; 2653 2654 /* 2655 * Loop around doing readdir rpc's of size nm_readdirsize 2656 * truncated to a multiple of NFS_DIRFRAGSIZ. 2657 * The stopping criteria is EOF or buffer full. 2658 */ 2659 while (more_dirs && bigenough) { 2660 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2661 bigenough = 0; 2662 break; 2663 } 2664 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2665 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2666 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2667 nfsm_fhtom(dnp, 1); 2668 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2669 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2670 txdr_swapcookie3(uiop->uio_offset, tl); 2671 } else { 2672 txdr_cookie3(uiop->uio_offset, tl); 2673 } 2674 tl += 2; 2675 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2676 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2677 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2678 *tl = txdr_unsigned(nmp->nm_rsize); 2679 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2680 nfsm_postop_attr(vp, attrflag, 0); 2681 if (error) { 2682 m_freem(mrep); 2683 goto nfsmout; 2684 } 2685 nrpcs++; 2686 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2687 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2688 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2689 more_dirs = fxdr_unsigned(int, *tl); 2690 2691 /* loop thru the dir entries, doctoring them to 4bsd form */ 2692 while (more_dirs && bigenough) { 2693 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2694 fileno = fxdr_hyper(tl); 2695 len = fxdr_unsigned(int, *(tl + 2)); 2696 if (len <= 0 || len > NFS_MAXNAMLEN) { 2697 error = EBADRPC; 2698 m_freem(mrep); 2699 goto nfsmout; 2700 } 2701 /* for cookie stashing */ 2702 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2703 left = NFS_DIRFRAGSIZ - blksiz; 2704 if (reclen > left) { 2705 /* 2706 * DIRFRAGSIZ is aligned, no need to align 2707 * again here. 2708 */ 2709 memset(uiop->uio_iov->iov_base, 0, left); 2710 dp->d_reclen += left; 2711 UIO_ADVANCE(uiop, left); 2712 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2713 blksiz = 0; 2714 } 2715 if (reclen > uiop->uio_resid) 2716 bigenough = 0; 2717 if (bigenough) { 2718 int tlen; 2719 2720 dp = (struct dirent *)uiop->uio_iov->iov_base; 2721 dp->d_fileno = fileno; 2722 dp->d_namlen = len; 2723 dp->d_reclen = reclen; 2724 dp->d_type = DT_UNKNOWN; 2725 blksiz += reclen; 2726 if (blksiz == NFS_DIRFRAGSIZ) 2727 blksiz = 0; 2728 UIO_ADVANCE(uiop, DIRHDSIZ); 2729 nfsm_mtouio(uiop, len); 2730 tlen = reclen - (DIRHDSIZ + len); 2731 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2732 UIO_ADVANCE(uiop, tlen); 2733 cnp->cn_nameptr = dp->d_name; 2734 cnp->cn_namelen = dp->d_namlen; 2735 } else 2736 nfsm_adv(nfsm_rndup(len)); 2737 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2738 if (bigenough) { 2739 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2740 uiop->uio_offset = 2741 fxdr_swapcookie3(tl); 2742 else 2743 uiop->uio_offset = 2744 fxdr_cookie3(tl); 2745 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2746 } 2747 tl += 2; 2748 2749 /* 2750 * Since the attributes are before the file handle 2751 * (sigh), we must skip over the attributes and then 2752 * come back and get them. 2753 */ 2754 attrflag = fxdr_unsigned(int, *tl); 2755 if (attrflag) { 2756 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2757 memcpy(&fattr, fp, NFSX_V3FATTR); 2758 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2759 doit = fxdr_unsigned(int, *tl); 2760 if (doit) { 2761 nfsm_getfh(fhp, fhsize, 1); 2762 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2763 vref(vp); 2764 newvp = vp; 2765 np = dnp; 2766 } else { 2767 error = nfs_nget1(vp->v_mount, fhp, 2768 fhsize, &np, LK_NOWAIT); 2769 if (!error) 2770 newvp = NFSTOV(np); 2771 } 2772 if (!error) { 2773 const char *xcp; 2774 2775 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2776 if (bigenough) { 2777 dp->d_type = 2778 IFTODT(VTTOIF(np->n_vattr->va_type)); 2779 if (cnp->cn_namelen <= NCHNAMLEN) { 2780 ndp->ni_vp = newvp; 2781 xcp = cnp->cn_nameptr + 2782 cnp->cn_namelen; 2783 cnp->cn_hash = 2784 namei_hash(cnp->cn_nameptr, &xcp); 2785 nfs_cache_enter(ndp->ni_dvp, 2786 ndp->ni_vp, cnp); 2787 } 2788 } 2789 } 2790 error = 0; 2791 } 2792 } else { 2793 /* Just skip over the file handle */ 2794 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2795 i = fxdr_unsigned(int, *tl); 2796 nfsm_adv(nfsm_rndup(i)); 2797 } 2798 if (newvp != NULLVP) { 2799 if (newvp == vp) 2800 vrele(newvp); 2801 else 2802 vput(newvp); 2803 newvp = NULLVP; 2804 } 2805 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2806 more_dirs = fxdr_unsigned(int, *tl); 2807 } 2808 /* 2809 * If at end of rpc data, get the eof boolean 2810 */ 2811 if (!more_dirs) { 2812 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2813 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2814 2815 /* 2816 * kludge: see a comment in nfs_readdirrpc. 2817 */ 2818 2819 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2820 more_dirs = 0; 2821 } 2822 m_freem(mrep); 2823 } 2824 /* 2825 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2826 * by increasing d_reclen for the last record. 2827 */ 2828 if (blksiz > 0) { 2829 left = NFS_DIRFRAGSIZ - blksiz; 2830 memset(uiop->uio_iov->iov_base, 0, left); 2831 dp->d_reclen += left; 2832 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2833 UIO_ADVANCE(uiop, left); 2834 } 2835 2836 /* 2837 * We are now either at the end of the directory or have filled the 2838 * block. 2839 */ 2840 if (bigenough) { 2841 dnp->n_direofoffset = uiop->uio_offset; 2842 dnp->n_flag |= NEOFVALID; 2843 } 2844 nfsmout: 2845 if (newvp != NULLVP) { 2846 if(newvp == vp) 2847 vrele(newvp); 2848 else 2849 vput(newvp); 2850 } 2851 return (error); 2852 } 2853 #endif 2854 2855 /* 2856 * Silly rename. To make the NFS filesystem that is stateless look a little 2857 * more like the "ufs" a remove of an active vnode is translated to a rename 2858 * to a funny looking filename that is removed by nfs_inactive on the 2859 * nfsnode. There is the potential for another process on a different client 2860 * to create the same funny name between the nfs_lookitup() fails and the 2861 * nfs_rename() completes, but... 2862 */ 2863 int 2864 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2865 { 2866 struct sillyrename *sp; 2867 struct nfsnode *np; 2868 int error; 2869 pid_t pid; 2870 2871 cache_purge(dvp); 2872 np = VTONFS(vp); 2873 #ifndef DIAGNOSTIC 2874 if (vp->v_type == VDIR) 2875 panic("nfs: sillyrename dir"); 2876 #endif 2877 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2878 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2879 sp->s_dvp = dvp; 2880 vref(dvp); 2881 2882 /* Fudge together a funny name */ 2883 pid = curlwp->l_proc->p_pid; 2884 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2885 sp->s_namlen = 12; 2886 sp->s_name[8] = hexdigits[pid & 0xf]; 2887 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2888 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2889 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2890 2891 /* Try lookitups until we get one that isn't there */ 2892 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2893 curlwp, (struct nfsnode **)0) == 0) { 2894 sp->s_name[4]++; 2895 if (sp->s_name[4] > 'z') { 2896 error = EINVAL; 2897 goto bad; 2898 } 2899 } 2900 if (dolink) { 2901 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2902 sp->s_cred, curlwp); 2903 /* 2904 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2905 */ 2906 if (error == ENOTSUP) { 2907 error = nfs_renameit(dvp, cnp, sp); 2908 } 2909 } else { 2910 error = nfs_renameit(dvp, cnp, sp); 2911 } 2912 if (error) 2913 goto bad; 2914 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2915 curlwp, &np); 2916 np->n_sillyrename = sp; 2917 return (0); 2918 bad: 2919 vrele(sp->s_dvp); 2920 kauth_cred_free(sp->s_cred); 2921 kmem_free(sp, sizeof(*sp)); 2922 return (error); 2923 } 2924 2925 /* 2926 * Look up a file name and optionally either update the file handle or 2927 * allocate an nfsnode, depending on the value of npp. 2928 * npp == NULL --> just do the lookup 2929 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2930 * handled too 2931 * *npp != NULL --> update the file handle in the vnode 2932 */ 2933 int 2934 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2935 { 2936 u_int32_t *tl; 2937 char *cp; 2938 int32_t t1, t2; 2939 struct vnode *newvp = (struct vnode *)0; 2940 struct nfsnode *np, *dnp = VTONFS(dvp); 2941 char *bpos, *dpos, *cp2; 2942 int error = 0, fhlen; 2943 #ifndef NFS_V2_ONLY 2944 int attrflag; 2945 #endif 2946 struct mbuf *mreq, *mrep, *md, *mb; 2947 nfsfh_t *nfhp; 2948 const int v3 = NFS_ISV3(dvp); 2949 2950 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2951 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2952 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2953 nfsm_fhtom(dnp, v3); 2954 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2955 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2956 if (npp && !error) { 2957 nfsm_getfh(nfhp, fhlen, v3); 2958 if (*npp) { 2959 np = *npp; 2960 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2961 kmem_free(np->n_fhp, np->n_fhsize); 2962 np->n_fhp = &np->n_fh; 2963 } 2964 #if NFS_SMALLFH < NFSX_V3FHMAX 2965 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2966 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2967 #endif 2968 memcpy(np->n_fhp, nfhp, fhlen); 2969 np->n_fhsize = fhlen; 2970 newvp = NFSTOV(np); 2971 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2972 vref(dvp); 2973 newvp = dvp; 2974 np = dnp; 2975 } else { 2976 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2977 if (error) { 2978 m_freem(mrep); 2979 return (error); 2980 } 2981 newvp = NFSTOV(np); 2982 } 2983 #ifndef NFS_V2_ONLY 2984 if (v3) { 2985 nfsm_postop_attr(newvp, attrflag, 0); 2986 if (!attrflag && *npp == NULL) { 2987 m_freem(mrep); 2988 vput(newvp); 2989 return (ENOENT); 2990 } 2991 } else 2992 #endif 2993 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2994 } 2995 nfsm_reqdone; 2996 if (npp && *npp == NULL) { 2997 if (error) { 2998 if (newvp) 2999 vput(newvp); 3000 } else 3001 *npp = np; 3002 } 3003 return (error); 3004 } 3005 3006 #ifndef NFS_V2_ONLY 3007 /* 3008 * Nfs Version 3 commit rpc 3009 */ 3010 int 3011 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3012 { 3013 char *cp; 3014 u_int32_t *tl; 3015 int32_t t1, t2; 3016 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3017 char *bpos, *dpos, *cp2; 3018 int error = 0, wccflag = NFSV3_WCCRATTR; 3019 struct mbuf *mreq, *mrep, *md, *mb; 3020 struct nfsnode *np; 3021 3022 KASSERT(NFS_ISV3(vp)); 3023 3024 #ifdef NFS_DEBUG_COMMIT 3025 printf("commit %lu - %lu\n", (unsigned long)offset, 3026 (unsigned long)(offset + cnt)); 3027 #endif 3028 3029 mutex_enter(&nmp->nm_lock); 3030 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3031 mutex_exit(&nmp->nm_lock); 3032 return (0); 3033 } 3034 mutex_exit(&nmp->nm_lock); 3035 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3036 np = VTONFS(vp); 3037 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3038 nfsm_fhtom(np, 1); 3039 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3040 txdr_hyper(offset, tl); 3041 tl += 2; 3042 *tl = txdr_unsigned(cnt); 3043 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3044 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3045 if (!error) { 3046 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3047 mutex_enter(&nmp->nm_lock); 3048 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3049 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3050 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3051 error = NFSERR_STALEWRITEVERF; 3052 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3053 } 3054 mutex_exit(&nmp->nm_lock); 3055 } 3056 nfsm_reqdone; 3057 return (error); 3058 } 3059 #endif 3060 3061 /* 3062 * Kludge City.. 3063 * - make nfs_bmap() essentially a no-op that does no translation 3064 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3065 * (Maybe I could use the process's page mapping, but I was concerned that 3066 * Kernel Write might not be enabled and also figured copyout() would do 3067 * a lot more work than memcpy() and also it currently happens in the 3068 * context of the swapper process (2). 3069 */ 3070 int 3071 nfs_bmap(void *v) 3072 { 3073 struct vop_bmap_args /* { 3074 struct vnode *a_vp; 3075 daddr_t a_bn; 3076 struct vnode **a_vpp; 3077 daddr_t *a_bnp; 3078 int *a_runp; 3079 } */ *ap = v; 3080 struct vnode *vp = ap->a_vp; 3081 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3082 3083 if (ap->a_vpp != NULL) 3084 *ap->a_vpp = vp; 3085 if (ap->a_bnp != NULL) 3086 *ap->a_bnp = ap->a_bn << bshift; 3087 if (ap->a_runp != NULL) 3088 *ap->a_runp = 1024 * 1024; /* XXX */ 3089 return (0); 3090 } 3091 3092 /* 3093 * Strategy routine. 3094 * For async requests when nfsiod(s) are running, queue the request by 3095 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3096 * request. 3097 */ 3098 int 3099 nfs_strategy(void *v) 3100 { 3101 struct vop_strategy_args *ap = v; 3102 struct buf *bp = ap->a_bp; 3103 int error = 0; 3104 3105 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3106 panic("nfs physio/async"); 3107 3108 /* 3109 * If the op is asynchronous and an i/o daemon is waiting 3110 * queue the request, wake it up and wait for completion 3111 * otherwise just do it ourselves. 3112 */ 3113 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3114 error = nfs_doio(bp); 3115 return (error); 3116 } 3117 3118 /* 3119 * fsync vnode op. Just call nfs_flush() with commit == 1. 3120 */ 3121 /* ARGSUSED */ 3122 int 3123 nfs_fsync(void *v) 3124 { 3125 struct vop_fsync_args /* { 3126 struct vnodeop_desc *a_desc; 3127 struct vnode * a_vp; 3128 kauth_cred_t a_cred; 3129 int a_flags; 3130 off_t offlo; 3131 off_t offhi; 3132 struct lwp * a_l; 3133 } */ *ap = v; 3134 3135 struct vnode *vp = ap->a_vp; 3136 3137 if (vp->v_type != VREG) 3138 return 0; 3139 3140 return (nfs_flush(vp, ap->a_cred, 3141 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3142 } 3143 3144 /* 3145 * Flush all the data associated with a vnode. 3146 */ 3147 int 3148 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3149 int commit) 3150 { 3151 struct nfsnode *np = VTONFS(vp); 3152 int error; 3153 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3154 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3155 3156 mutex_enter(vp->v_interlock); 3157 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3158 if (np->n_flag & NWRITEERR) { 3159 error = np->n_error; 3160 np->n_flag &= ~NWRITEERR; 3161 } 3162 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3163 return (error); 3164 } 3165 3166 /* 3167 * Return POSIX pathconf information applicable to nfs. 3168 * 3169 * N.B. The NFS V2 protocol doesn't support this RPC. 3170 */ 3171 /* ARGSUSED */ 3172 int 3173 nfs_pathconf(void *v) 3174 { 3175 struct vop_pathconf_args /* { 3176 struct vnode *a_vp; 3177 int a_name; 3178 register_t *a_retval; 3179 } */ *ap = v; 3180 struct nfsv3_pathconf *pcp; 3181 struct vnode *vp = ap->a_vp; 3182 struct mbuf *mreq, *mrep, *md, *mb; 3183 int32_t t1, t2; 3184 u_int32_t *tl; 3185 char *bpos, *dpos, *cp, *cp2; 3186 int error = 0, attrflag; 3187 #ifndef NFS_V2_ONLY 3188 struct nfsmount *nmp; 3189 unsigned int l; 3190 u_int64_t maxsize; 3191 #endif 3192 const int v3 = NFS_ISV3(vp); 3193 struct nfsnode *np = VTONFS(vp); 3194 3195 switch (ap->a_name) { 3196 /* Names that can be resolved locally. */ 3197 case _PC_PIPE_BUF: 3198 *ap->a_retval = PIPE_BUF; 3199 break; 3200 case _PC_SYNC_IO: 3201 *ap->a_retval = 1; 3202 break; 3203 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3204 case _PC_LINK_MAX: 3205 case _PC_NAME_MAX: 3206 case _PC_CHOWN_RESTRICTED: 3207 case _PC_NO_TRUNC: 3208 if (!v3) { 3209 error = EINVAL; 3210 break; 3211 } 3212 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3213 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3214 nfsm_fhtom(np, 1); 3215 nfsm_request(np, NFSPROC_PATHCONF, 3216 curlwp, curlwp->l_cred); /* XXX */ 3217 nfsm_postop_attr(vp, attrflag, 0); 3218 if (!error) { 3219 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3220 NFSX_V3PATHCONF); 3221 switch (ap->a_name) { 3222 case _PC_LINK_MAX: 3223 *ap->a_retval = 3224 fxdr_unsigned(register_t, pcp->pc_linkmax); 3225 break; 3226 case _PC_NAME_MAX: 3227 *ap->a_retval = 3228 fxdr_unsigned(register_t, pcp->pc_namemax); 3229 break; 3230 case _PC_CHOWN_RESTRICTED: 3231 *ap->a_retval = 3232 (pcp->pc_chownrestricted == nfs_true); 3233 break; 3234 case _PC_NO_TRUNC: 3235 *ap->a_retval = 3236 (pcp->pc_notrunc == nfs_true); 3237 break; 3238 } 3239 } 3240 nfsm_reqdone; 3241 break; 3242 case _PC_FILESIZEBITS: 3243 #ifndef NFS_V2_ONLY 3244 if (v3) { 3245 nmp = VFSTONFS(vp->v_mount); 3246 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3247 if ((error = nfs_fsinfo(nmp, vp, 3248 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3249 break; 3250 for (l = 0, maxsize = nmp->nm_maxfilesize; 3251 (maxsize >> l) > 0; l++) 3252 ; 3253 *ap->a_retval = l + 1; 3254 } else 3255 #endif 3256 { 3257 *ap->a_retval = 32; /* NFS V2 limitation */ 3258 } 3259 break; 3260 default: 3261 error = EINVAL; 3262 break; 3263 } 3264 3265 return (error); 3266 } 3267 3268 /* 3269 * NFS advisory byte-level locks. 3270 */ 3271 int 3272 nfs_advlock(void *v) 3273 { 3274 struct vop_advlock_args /* { 3275 struct vnode *a_vp; 3276 void *a_id; 3277 int a_op; 3278 struct flock *a_fl; 3279 int a_flags; 3280 } */ *ap = v; 3281 struct nfsnode *np = VTONFS(ap->a_vp); 3282 3283 return lf_advlock(ap, &np->n_lockf, np->n_size); 3284 } 3285 3286 /* 3287 * Print out the contents of an nfsnode. 3288 */ 3289 int 3290 nfs_print(void *v) 3291 { 3292 struct vop_print_args /* { 3293 struct vnode *a_vp; 3294 } */ *ap = v; 3295 struct vnode *vp = ap->a_vp; 3296 struct nfsnode *np = VTONFS(vp); 3297 3298 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3299 (unsigned long long)np->n_vattr->va_fileid, 3300 (unsigned long long)np->n_vattr->va_fsid); 3301 if (vp->v_type == VFIFO) 3302 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3303 printf("\n"); 3304 return (0); 3305 } 3306 3307 /* 3308 * nfs unlock wrapper. 3309 */ 3310 int 3311 nfs_unlock(void *v) 3312 { 3313 struct vop_unlock_args /* { 3314 struct vnode *a_vp; 3315 int a_flags; 3316 } */ *ap = v; 3317 struct vnode *vp = ap->a_vp; 3318 3319 /* 3320 * VOP_UNLOCK can be called by nfs_loadattrcache 3321 * with v_data == 0. 3322 */ 3323 if (VTONFS(vp)) { 3324 nfs_delayedtruncate(vp); 3325 } 3326 3327 return genfs_unlock(v); 3328 } 3329 3330 /* 3331 * nfs special file access vnode op. 3332 * Essentially just get vattr and then imitate iaccess() since the device is 3333 * local to the client. 3334 */ 3335 int 3336 nfsspec_access(void *v) 3337 { 3338 struct vop_access_args /* { 3339 struct vnode *a_vp; 3340 int a_mode; 3341 kauth_cred_t a_cred; 3342 struct lwp *a_l; 3343 } */ *ap = v; 3344 struct vattr va; 3345 struct vnode *vp = ap->a_vp; 3346 int error; 3347 3348 error = VOP_GETATTR(vp, &va, ap->a_cred); 3349 if (error) 3350 return (error); 3351 3352 /* 3353 * Disallow write attempts on filesystems mounted read-only; 3354 * unless the file is a socket, fifo, or a block or character 3355 * device resident on the filesystem. 3356 */ 3357 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3358 switch (vp->v_type) { 3359 case VREG: 3360 case VDIR: 3361 case VLNK: 3362 return (EROFS); 3363 default: 3364 break; 3365 } 3366 } 3367 3368 return (genfs_can_access(va.va_type, va.va_mode, 3369 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3370 } 3371 3372 /* 3373 * Read wrapper for special devices. 3374 */ 3375 int 3376 nfsspec_read(void *v) 3377 { 3378 struct vop_read_args /* { 3379 struct vnode *a_vp; 3380 struct uio *a_uio; 3381 int a_ioflag; 3382 kauth_cred_t a_cred; 3383 } */ *ap = v; 3384 struct nfsnode *np = VTONFS(ap->a_vp); 3385 3386 /* 3387 * Set access flag. 3388 */ 3389 np->n_flag |= NACC; 3390 getnanotime(&np->n_atim); 3391 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3392 } 3393 3394 /* 3395 * Write wrapper for special devices. 3396 */ 3397 int 3398 nfsspec_write(void *v) 3399 { 3400 struct vop_write_args /* { 3401 struct vnode *a_vp; 3402 struct uio *a_uio; 3403 int a_ioflag; 3404 kauth_cred_t a_cred; 3405 } */ *ap = v; 3406 struct nfsnode *np = VTONFS(ap->a_vp); 3407 3408 /* 3409 * Set update flag. 3410 */ 3411 np->n_flag |= NUPD; 3412 getnanotime(&np->n_mtim); 3413 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3414 } 3415 3416 /* 3417 * Close wrapper for special devices. 3418 * 3419 * Update the times on the nfsnode then do device close. 3420 */ 3421 int 3422 nfsspec_close(void *v) 3423 { 3424 struct vop_close_args /* { 3425 struct vnode *a_vp; 3426 int a_fflag; 3427 kauth_cred_t a_cred; 3428 struct lwp *a_l; 3429 } */ *ap = v; 3430 struct vnode *vp = ap->a_vp; 3431 struct nfsnode *np = VTONFS(vp); 3432 struct vattr vattr; 3433 3434 if (np->n_flag & (NACC | NUPD)) { 3435 np->n_flag |= NCHG; 3436 if (vp->v_usecount == 1 && 3437 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3438 vattr_null(&vattr); 3439 if (np->n_flag & NACC) 3440 vattr.va_atime = np->n_atim; 3441 if (np->n_flag & NUPD) 3442 vattr.va_mtime = np->n_mtim; 3443 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3444 } 3445 } 3446 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3447 } 3448 3449 /* 3450 * Read wrapper for fifos. 3451 */ 3452 int 3453 nfsfifo_read(void *v) 3454 { 3455 struct vop_read_args /* { 3456 struct vnode *a_vp; 3457 struct uio *a_uio; 3458 int a_ioflag; 3459 kauth_cred_t a_cred; 3460 } */ *ap = v; 3461 struct nfsnode *np = VTONFS(ap->a_vp); 3462 3463 /* 3464 * Set access flag. 3465 */ 3466 np->n_flag |= NACC; 3467 getnanotime(&np->n_atim); 3468 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3469 } 3470 3471 /* 3472 * Write wrapper for fifos. 3473 */ 3474 int 3475 nfsfifo_write(void *v) 3476 { 3477 struct vop_write_args /* { 3478 struct vnode *a_vp; 3479 struct uio *a_uio; 3480 int a_ioflag; 3481 kauth_cred_t a_cred; 3482 } */ *ap = v; 3483 struct nfsnode *np = VTONFS(ap->a_vp); 3484 3485 /* 3486 * Set update flag. 3487 */ 3488 np->n_flag |= NUPD; 3489 getnanotime(&np->n_mtim); 3490 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3491 } 3492 3493 /* 3494 * Close wrapper for fifos. 3495 * 3496 * Update the times on the nfsnode then do fifo close. 3497 */ 3498 int 3499 nfsfifo_close(void *v) 3500 { 3501 struct vop_close_args /* { 3502 struct vnode *a_vp; 3503 int a_fflag; 3504 kauth_cred_t a_cred; 3505 struct lwp *a_l; 3506 } */ *ap = v; 3507 struct vnode *vp = ap->a_vp; 3508 struct nfsnode *np = VTONFS(vp); 3509 struct vattr vattr; 3510 3511 if (np->n_flag & (NACC | NUPD)) { 3512 struct timespec ts; 3513 3514 getnanotime(&ts); 3515 if (np->n_flag & NACC) 3516 np->n_atim = ts; 3517 if (np->n_flag & NUPD) 3518 np->n_mtim = ts; 3519 np->n_flag |= NCHG; 3520 if (vp->v_usecount == 1 && 3521 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3522 vattr_null(&vattr); 3523 if (np->n_flag & NACC) 3524 vattr.va_atime = np->n_atim; 3525 if (np->n_flag & NUPD) 3526 vattr.va_mtime = np->n_mtim; 3527 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3528 } 3529 } 3530 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3531 } 3532
Indexes created Mon Oct 13 14:09:54 GMT 2025