nfs_vnops.c revision 1.318
1 /* $NetBSD: nfs_vnops.c,v 1.318 2021/06/29 22:34:09 dholland Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_vnops.c 8.19 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * vnode op calls for Sun NFS version 2 and 3 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.318 2021/06/29 22:34:09 dholland Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/proc.h> 51 #include <sys/kernel.h> 52 #include <sys/systm.h> 53 #include <sys/resourcevar.h> 54 #include <sys/mount.h> 55 #include <sys/buf.h> 56 #include <sys/condvar.h> 57 #include <sys/disk.h> 58 #include <sys/malloc.h> 59 #include <sys/kmem.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/vnode.h> 64 #include <sys/dirent.h> 65 #include <sys/fcntl.h> 66 #include <sys/hash.h> 67 #include <sys/lockf.h> 68 #include <sys/stat.h> 69 #include <sys/unistd.h> 70 #include <sys/kauth.h> 71 #include <sys/cprng.h> 72 73 #ifdef UVMHIST 74 #include <uvm/uvm.h> 75 #endif 76 #include <uvm/uvm_extern.h> 77 #include <uvm/uvm_stat.h> 78 79 #include <miscfs/fifofs/fifo.h> 80 #include <miscfs/genfs/genfs.h> 81 #include <miscfs/genfs/genfs_node.h> 82 #include <miscfs/specfs/specdev.h> 83 84 #include <nfs/rpcv2.h> 85 #include <nfs/nfsproto.h> 86 #include <nfs/nfs.h> 87 #include <nfs/nfsnode.h> 88 #include <nfs/nfsmount.h> 89 #include <nfs/xdr_subs.h> 90 #include <nfs/nfsm_subs.h> 91 #include <nfs/nfs_var.h> 92 93 #include <net/if.h> 94 #include <netinet/in.h> 95 #include <netinet/in_var.h> 96 97 /* 98 * Global vfs data structures for nfs 99 */ 100 int (**nfsv2_vnodeop_p)(void *); 101 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 102 { &vop_default_desc, vn_default_error }, 103 { &vop_parsepath_desc, genfs_parsepath }, /* parsepath */ 104 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 105 { &vop_create_desc, nfs_create }, /* create */ 106 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 107 { &vop_open_desc, nfs_open }, /* open */ 108 { &vop_close_desc, nfs_close }, /* close */ 109 { &vop_access_desc, nfs_access }, /* access */ 110 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 111 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 112 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 113 { &vop_read_desc, nfs_read }, /* read */ 114 { &vop_write_desc, nfs_write }, /* write */ 115 { &vop_fallocate_desc, genfs_eopnotsupp }, /* fallocate */ 116 { &vop_fdiscard_desc, genfs_eopnotsupp }, /* fdiscard */ 117 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 118 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 119 { &vop_poll_desc, nfs_poll }, /* poll */ 120 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 121 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 122 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 123 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 124 { &vop_seek_desc, nfs_seek }, /* seek */ 125 { &vop_remove_desc, nfs_remove }, /* remove */ 126 { &vop_link_desc, nfs_link }, /* link */ 127 { &vop_rename_desc, nfs_rename }, /* rename */ 128 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 129 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 130 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 131 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 132 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 133 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 134 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 135 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 136 { &vop_lock_desc, nfs_lock }, /* lock */ 137 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 138 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 139 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 140 { &vop_print_desc, nfs_print }, /* print */ 141 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 142 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 143 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 144 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 145 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 146 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 147 { NULL, NULL } 148 }; 149 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 150 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 151 152 /* 153 * Special device vnode ops 154 */ 155 int (**spec_nfsv2nodeop_p)(void *); 156 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 157 { &vop_default_desc, vn_default_error }, 158 { &vop_parsepath_desc, genfs_parsepath }, /* parsepath */ 159 { &vop_lookup_desc, spec_lookup }, /* lookup */ 160 { &vop_create_desc, spec_create }, /* create */ 161 { &vop_mknod_desc, spec_mknod }, /* mknod */ 162 { &vop_open_desc, spec_open }, /* open */ 163 { &vop_close_desc, nfsspec_close }, /* close */ 164 { &vop_access_desc, nfsspec_access }, /* access */ 165 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 166 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 167 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 168 { &vop_read_desc, nfsspec_read }, /* read */ 169 { &vop_write_desc, nfsspec_write }, /* write */ 170 { &vop_fallocate_desc, spec_fallocate }, /* fallocate */ 171 { &vop_fdiscard_desc, spec_fdiscard }, /* fdiscard */ 172 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 173 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 174 { &vop_poll_desc, spec_poll }, /* poll */ 175 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 176 { &vop_revoke_desc, spec_revoke }, /* revoke */ 177 { &vop_mmap_desc, spec_mmap }, /* mmap */ 178 { &vop_fsync_desc, spec_fsync }, /* fsync */ 179 { &vop_seek_desc, spec_seek }, /* seek */ 180 { &vop_remove_desc, spec_remove }, /* remove */ 181 { &vop_link_desc, spec_link }, /* link */ 182 { &vop_rename_desc, spec_rename }, /* rename */ 183 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 184 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 185 { &vop_symlink_desc, spec_symlink }, /* symlink */ 186 { &vop_readdir_desc, spec_readdir }, /* readdir */ 187 { &vop_readlink_desc, spec_readlink }, /* readlink */ 188 { &vop_abortop_desc, spec_abortop }, /* abortop */ 189 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 190 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 191 { &vop_lock_desc, nfs_lock }, /* lock */ 192 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 193 { &vop_bmap_desc, spec_bmap }, /* bmap */ 194 { &vop_strategy_desc, spec_strategy }, /* strategy */ 195 { &vop_print_desc, nfs_print }, /* print */ 196 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 197 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 198 { &vop_advlock_desc, spec_advlock }, /* advlock */ 199 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 200 { &vop_getpages_desc, spec_getpages }, /* getpages */ 201 { &vop_putpages_desc, spec_putpages }, /* putpages */ 202 { NULL, NULL } 203 }; 204 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 205 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 206 207 int (**fifo_nfsv2nodeop_p)(void *); 208 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 209 { &vop_default_desc, vn_default_error }, 210 { &vop_parsepath_desc, genfs_parsepath }, /* parsepath */ 211 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 212 { &vop_create_desc, vn_fifo_bypass }, /* create */ 213 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 214 { &vop_open_desc, vn_fifo_bypass }, /* open */ 215 { &vop_close_desc, nfsfifo_close }, /* close */ 216 { &vop_access_desc, nfsspec_access }, /* access */ 217 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 218 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 219 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 220 { &vop_read_desc, nfsfifo_read }, /* read */ 221 { &vop_write_desc, nfsfifo_write }, /* write */ 222 { &vop_fallocate_desc, vn_fifo_bypass }, /* fallocate */ 223 { &vop_fdiscard_desc, vn_fifo_bypass }, /* fdiscard */ 224 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 225 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 226 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 227 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 228 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 229 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 230 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 231 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 232 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 233 { &vop_link_desc, vn_fifo_bypass }, /* link */ 234 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 235 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 236 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 237 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 238 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 239 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 240 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 241 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 242 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 243 { &vop_lock_desc, nfs_lock }, /* lock */ 244 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 245 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 246 { &vop_strategy_desc, genfs_badop }, /* strategy */ 247 { &vop_print_desc, nfs_print }, /* print */ 248 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 249 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 250 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 251 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 252 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 253 { NULL, NULL } 254 }; 255 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 256 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 257 258 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 259 size_t, kauth_cred_t, struct lwp *); 260 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 261 262 /* 263 * Global variables 264 */ 265 extern u_int32_t nfs_true, nfs_false; 266 extern u_int32_t nfs_xdrneg1; 267 extern const nfstype nfsv3_type[9]; 268 269 int nfs_numasync = 0; 270 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 271 #define UIO_ADVANCE(uio, siz) \ 272 (void)((uio)->uio_resid -= (siz), \ 273 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 274 (uio)->uio_iov->iov_len -= (siz)) 275 276 static void nfs_cache_enter(struct vnode *, struct vnode *, 277 struct componentname *); 278 279 static void 280 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 281 struct componentname *cnp) 282 { 283 struct nfsnode *dnp = VTONFS(dvp); 284 285 if ((cnp->cn_flags & MAKEENTRY) == 0) { 286 return; 287 } 288 if (vp != NULL) { 289 struct nfsnode *np = VTONFS(vp); 290 291 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 292 } 293 294 if (!timespecisset(&dnp->n_nctime)) 295 dnp->n_nctime = dnp->n_vattr->va_mtime; 296 297 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags); 298 } 299 300 /* 301 * nfs null call from vfs. 302 */ 303 int 304 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 305 { 306 char *bpos, *dpos; 307 int error = 0; 308 struct mbuf *mreq, *mrep, *md, *mb __unused; 309 struct nfsnode *np = VTONFS(vp); 310 311 nfsm_reqhead(np, NFSPROC_NULL, 0); 312 nfsm_request(np, NFSPROC_NULL, l, cred); 313 nfsm_reqdone; 314 return (error); 315 } 316 317 /* 318 * nfs access vnode op. 319 * For nfs version 2, just return ok. File accesses may fail later. 320 * For nfs version 3, use the access rpc to check accessibility. If file modes 321 * are changed on the server, accesses might still fail later. 322 */ 323 int 324 nfs_access(void *v) 325 { 326 struct vop_access_args /* { 327 struct vnode *a_vp; 328 accmode_t a_accmode; 329 kauth_cred_t a_cred; 330 } */ *ap = v; 331 struct vnode *vp = ap->a_vp; 332 #ifndef NFS_V2_ONLY 333 u_int32_t *tl; 334 char *cp; 335 int32_t t1, t2; 336 char *bpos, *dpos, *cp2; 337 int error = 0, attrflag; 338 struct mbuf *mreq, *mrep, *md, *mb; 339 u_int32_t mode, rmode; 340 const int v3 = NFS_ISV3(vp); 341 #endif 342 int cachevalid; 343 struct nfsnode *np = VTONFS(vp); 344 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 345 346 cachevalid = (np->n_accstamp != -1 && 347 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 348 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 349 350 /* 351 * Check access cache first. If this request has been made for this 352 * uid shortly before, use the cached result. 353 */ 354 if (cachevalid) { 355 if (!np->n_accerror) { 356 if ((np->n_accmode & ap->a_accmode) == ap->a_accmode) 357 return np->n_accerror; 358 } else if ((np->n_accmode & ap->a_accmode) == np->n_accmode) 359 return np->n_accerror; 360 } 361 362 #ifndef NFS_V2_ONLY 363 /* 364 * For nfs v3, do an access rpc, otherwise you are stuck emulating 365 * ufs_access() locally using the vattr. This may not be correct, 366 * since the server may apply other access criteria such as 367 * client uid-->server uid mapping that we do not know about, but 368 * this is better than just returning anything that is lying about 369 * in the cache. 370 */ 371 if (v3) { 372 nfsstats.rpccnt[NFSPROC_ACCESS]++; 373 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 374 nfsm_fhtom(np, v3); 375 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 376 if (ap->a_accmode & VREAD) 377 mode = NFSV3ACCESS_READ; 378 else 379 mode = 0; 380 if (vp->v_type != VDIR) { 381 if (ap->a_accmode & VWRITE) 382 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 383 if (ap->a_accmode & VEXEC) 384 mode |= NFSV3ACCESS_EXECUTE; 385 } else { 386 if (ap->a_accmode & VWRITE) 387 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 388 NFSV3ACCESS_DELETE); 389 if (ap->a_accmode & VEXEC) 390 mode |= NFSV3ACCESS_LOOKUP; 391 } 392 *tl = txdr_unsigned(mode); 393 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 394 nfsm_postop_attr(vp, attrflag, 0); 395 if (!error) { 396 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 397 rmode = fxdr_unsigned(u_int32_t, *tl); 398 /* 399 * The NFS V3 spec does not clarify whether or not 400 * the returned access bits can be a superset of 401 * the ones requested, so... 402 */ 403 if ((rmode & mode) != mode) 404 error = EACCES; 405 } 406 nfsm_reqdone; 407 } else 408 #endif 409 return (nfsspec_access(ap)); 410 #ifndef NFS_V2_ONLY 411 /* 412 * Disallow write attempts on filesystems mounted read-only; 413 * unless the file is a socket, fifo, or a block or character 414 * device resident on the filesystem. 415 */ 416 if (!error && (ap->a_accmode & VWRITE) && 417 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 418 switch (vp->v_type) { 419 case VREG: 420 case VDIR: 421 case VLNK: 422 error = EROFS; 423 default: 424 break; 425 } 426 } 427 428 if (!error || error == EACCES) { 429 /* 430 * If we got the same result as for a previous, 431 * different request, OR it in. Don't update 432 * the timestamp in that case. 433 */ 434 if (cachevalid && np->n_accstamp != -1 && 435 error == np->n_accerror) { 436 if (!error) 437 np->n_accmode |= ap->a_accmode; 438 else if ((np->n_accmode & ap->a_accmode) == ap->a_accmode) 439 np->n_accmode = ap->a_accmode; 440 } else { 441 np->n_accstamp = time_uptime; 442 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 443 np->n_accmode = ap->a_accmode; 444 np->n_accerror = error; 445 } 446 } 447 448 return (error); 449 #endif 450 } 451 452 /* 453 * nfs open vnode op 454 * Check to see if the type is ok 455 * and that deletion is not in progress. 456 * For paged in text files, you will need to flush the page cache 457 * if consistency is lost. 458 */ 459 /* ARGSUSED */ 460 int 461 nfs_open(void *v) 462 { 463 struct vop_open_args /* { 464 struct vnode *a_vp; 465 int a_mode; 466 kauth_cred_t a_cred; 467 } */ *ap = v; 468 struct vnode *vp = ap->a_vp; 469 struct nfsnode *np = VTONFS(vp); 470 int error; 471 472 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 473 return (EACCES); 474 } 475 476 if (ap->a_mode & FREAD) { 477 if (np->n_rcred != NULL) 478 kauth_cred_free(np->n_rcred); 479 np->n_rcred = ap->a_cred; 480 kauth_cred_hold(np->n_rcred); 481 } 482 if (ap->a_mode & FWRITE) { 483 if (np->n_wcred != NULL) 484 kauth_cred_free(np->n_wcred); 485 np->n_wcred = ap->a_cred; 486 kauth_cred_hold(np->n_wcred); 487 } 488 489 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 490 if (error) 491 return error; 492 493 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 494 495 return (0); 496 } 497 498 /* 499 * nfs close vnode op 500 * What an NFS client should do upon close after writing is a debatable issue. 501 * Most NFS clients push delayed writes to the server upon close, basically for 502 * two reasons: 503 * 1 - So that any write errors may be reported back to the client process 504 * doing the close system call. By far the two most likely errors are 505 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 506 * 2 - To put a worst case upper bound on cache inconsistency between 507 * multiple clients for the file. 508 * There is also a consistency problem for Version 2 of the protocol w.r.t. 509 * not being able to tell if other clients are writing a file concurrently, 510 * since there is no way of knowing if the changed modify time in the reply 511 * is only due to the write for this client. 512 * (NFS Version 3 provides weak cache consistency data in the reply that 513 * should be sufficient to detect and handle this case.) 514 * 515 * The current code does the following: 516 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 517 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 518 * or commit them (this satisfies 1 and 2 except for the 519 * case where the server crashes after this close but 520 * before the commit RPC, which is felt to be "good 521 * enough". Changing the last argument to nfs_flush() to 522 * a 1 would force a commit operation, if it is felt a 523 * commit is necessary now. 524 */ 525 /* ARGSUSED */ 526 int 527 nfs_close(void *v) 528 { 529 struct vop_close_args /* { 530 struct vnodeop_desc *a_desc; 531 struct vnode *a_vp; 532 int a_fflag; 533 kauth_cred_t a_cred; 534 } */ *ap = v; 535 struct vnode *vp = ap->a_vp; 536 struct nfsnode *np = VTONFS(vp); 537 int error = 0; 538 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 539 540 if (vp->v_type == VREG) { 541 if (np->n_flag & NMODIFIED) { 542 #ifndef NFS_V2_ONLY 543 if (NFS_ISV3(vp)) { 544 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 545 np->n_flag &= ~NMODIFIED; 546 } else 547 #endif 548 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 549 NFS_INVALIDATE_ATTRCACHE(np); 550 } 551 if (np->n_flag & NWRITEERR) { 552 np->n_flag &= ~NWRITEERR; 553 error = np->n_error; 554 } 555 } 556 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 557 return (error); 558 } 559 560 /* 561 * nfs getattr call from vfs. 562 */ 563 int 564 nfs_getattr(void *v) 565 { 566 struct vop_getattr_args /* { 567 struct vnode *a_vp; 568 struct vattr *a_vap; 569 kauth_cred_t a_cred; 570 } */ *ap = v; 571 struct vnode *vp = ap->a_vp; 572 struct nfsnode *np = VTONFS(vp); 573 char *cp; 574 u_int32_t *tl; 575 int32_t t1, t2; 576 char *bpos, *dpos; 577 int error = 0; 578 struct mbuf *mreq, *mrep, *md, *mb; 579 const int v3 = NFS_ISV3(vp); 580 581 /* 582 * Update local times for special files. 583 */ 584 if (np->n_flag & (NACC | NUPD)) 585 np->n_flag |= NCHG; 586 587 /* 588 * if we have delayed truncation, do it now. 589 */ 590 nfs_delayedtruncate(vp); 591 592 /* 593 * First look in the cache. 594 */ 595 if (nfs_getattrcache(vp, ap->a_vap) == 0) 596 return (0); 597 nfsstats.rpccnt[NFSPROC_GETATTR]++; 598 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 599 nfsm_fhtom(np, v3); 600 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 601 if (!error) { 602 nfsm_loadattr(vp, ap->a_vap, 0); 603 if (vp->v_type == VDIR && 604 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 605 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 606 } 607 nfsm_reqdone; 608 return (error); 609 } 610 611 /* 612 * nfs setattr call. 613 */ 614 int 615 nfs_setattr(void *v) 616 { 617 struct vop_setattr_args /* { 618 struct vnodeop_desc *a_desc; 619 struct vnode *a_vp; 620 struct vattr *a_vap; 621 kauth_cred_t a_cred; 622 } */ *ap = v; 623 struct vnode *vp = ap->a_vp; 624 struct nfsnode *np = VTONFS(vp); 625 struct vattr *vap = ap->a_vap; 626 int error = 0; 627 u_quad_t tsize = 0; 628 629 /* 630 * Setting of flags is not supported. 631 */ 632 if (vap->va_flags != VNOVAL) 633 return (EOPNOTSUPP); 634 635 /* 636 * Disallow write attempts if the filesystem is mounted read-only. 637 */ 638 if ((vap->va_uid != (uid_t)VNOVAL || 639 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 640 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 641 (vp->v_mount->mnt_flag & MNT_RDONLY)) 642 return (EROFS); 643 if (vap->va_size != VNOVAL) { 644 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 645 return EFBIG; 646 } 647 switch (vp->v_type) { 648 case VDIR: 649 return (EISDIR); 650 case VCHR: 651 case VBLK: 652 case VSOCK: 653 case VFIFO: 654 if (vap->va_mtime.tv_sec == VNOVAL && 655 vap->va_atime.tv_sec == VNOVAL && 656 vap->va_mode == (mode_t)VNOVAL && 657 vap->va_uid == (uid_t)VNOVAL && 658 vap->va_gid == (gid_t)VNOVAL) 659 return (0); 660 vap->va_size = VNOVAL; 661 break; 662 default: 663 /* 664 * Disallow write attempts if the filesystem is 665 * mounted read-only. 666 */ 667 if (vp->v_mount->mnt_flag & MNT_RDONLY) 668 return (EROFS); 669 genfs_node_wrlock(vp); 670 uvm_vnp_setsize(vp, vap->va_size); 671 tsize = np->n_size; 672 np->n_size = vap->va_size; 673 if (vap->va_size == 0) 674 error = nfs_vinvalbuf(vp, 0, 675 ap->a_cred, curlwp, 1); 676 else 677 error = nfs_vinvalbuf(vp, V_SAVE, 678 ap->a_cred, curlwp, 1); 679 if (error) { 680 uvm_vnp_setsize(vp, tsize); 681 genfs_node_unlock(vp); 682 return (error); 683 } 684 np->n_vattr->va_size = vap->va_size; 685 } 686 } else { 687 /* 688 * flush files before setattr because a later write of 689 * cached data might change timestamps or reset sugid bits 690 */ 691 if ((vap->va_mtime.tv_sec != VNOVAL || 692 vap->va_atime.tv_sec != VNOVAL || 693 vap->va_mode != VNOVAL) && 694 vp->v_type == VREG && 695 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 696 curlwp, 1)) == EINTR) 697 return (error); 698 } 699 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 700 if (vap->va_size != VNOVAL) { 701 if (error) { 702 np->n_size = np->n_vattr->va_size = tsize; 703 uvm_vnp_setsize(vp, np->n_size); 704 } 705 genfs_node_unlock(vp); 706 } 707 VN_KNOTE(vp, NOTE_ATTRIB); 708 return (error); 709 } 710 711 /* 712 * Do an nfs setattr rpc. 713 */ 714 int 715 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 716 { 717 struct nfsv2_sattr *sp; 718 char *cp; 719 int32_t t1, t2; 720 char *bpos, *dpos; 721 u_int32_t *tl; 722 int error = 0; 723 struct mbuf *mreq, *mrep, *md, *mb; 724 const int v3 = NFS_ISV3(vp); 725 struct nfsnode *np = VTONFS(vp); 726 #ifndef NFS_V2_ONLY 727 int wccflag = NFSV3_WCCRATTR; 728 char *cp2; 729 #endif 730 731 nfsstats.rpccnt[NFSPROC_SETATTR]++; 732 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 733 nfsm_fhtom(np, v3); 734 #ifndef NFS_V2_ONLY 735 if (v3) { 736 nfsm_v3attrbuild(vap, true); 737 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 738 *tl = nfs_false; 739 } else { 740 #endif 741 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 742 if (vap->va_mode == (mode_t)VNOVAL) 743 sp->sa_mode = nfs_xdrneg1; 744 else 745 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 746 if (vap->va_uid == (uid_t)VNOVAL) 747 sp->sa_uid = nfs_xdrneg1; 748 else 749 sp->sa_uid = txdr_unsigned(vap->va_uid); 750 if (vap->va_gid == (gid_t)VNOVAL) 751 sp->sa_gid = nfs_xdrneg1; 752 else 753 sp->sa_gid = txdr_unsigned(vap->va_gid); 754 sp->sa_size = txdr_unsigned(vap->va_size); 755 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 756 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 757 #ifndef NFS_V2_ONLY 758 } 759 #endif 760 nfsm_request(np, NFSPROC_SETATTR, l, cred); 761 #ifndef NFS_V2_ONLY 762 if (v3) { 763 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 764 } else 765 #endif 766 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 767 nfsm_reqdone; 768 return (error); 769 } 770 771 /* 772 * nfs lookup call, one step at a time... 773 * First look in cache 774 * If not found, do the rpc. 775 */ 776 int 777 nfs_lookup(void *v) 778 { 779 struct vop_lookup_v2_args /* { 780 struct vnodeop_desc *a_desc; 781 struct vnode *a_dvp; 782 struct vnode **a_vpp; 783 struct componentname *a_cnp; 784 } */ *ap = v; 785 struct componentname *cnp = ap->a_cnp; 786 struct vnode *dvp = ap->a_dvp; 787 struct vnode **vpp = ap->a_vpp; 788 int flags; 789 struct vnode *newvp; 790 u_int32_t *tl; 791 char *cp; 792 int32_t t1, t2; 793 char *bpos, *dpos, *cp2; 794 struct mbuf *mreq, *mrep, *md, *mb; 795 long len; 796 nfsfh_t *fhp; 797 struct nfsnode *np; 798 int cachefound; 799 int error = 0, attrflag, fhsize; 800 const int v3 = NFS_ISV3(dvp); 801 802 flags = cnp->cn_flags; 803 804 *vpp = NULLVP; 805 newvp = NULLVP; 806 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 807 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 808 return (EROFS); 809 if (dvp->v_type != VDIR) 810 return (ENOTDIR); 811 812 /* 813 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 814 */ 815 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 816 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 817 if (error) 818 return error; 819 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 820 return EISDIR; 821 vref(dvp); 822 *vpp = dvp; 823 return 0; 824 } 825 826 np = VTONFS(dvp); 827 828 /* 829 * Before performing an RPC, check the name cache to see if 830 * the directory/name pair we are looking for is known already. 831 * If the directory/name pair is found in the name cache, 832 * we have to ensure the directory has not changed from 833 * the time the cache entry has been created. If it has, 834 * the cache entry has to be ignored. 835 */ 836 cachefound = cache_lookup_raw(dvp, cnp->cn_nameptr, cnp->cn_namelen, 837 cnp->cn_flags, NULL, vpp); 838 KASSERT(dvp != *vpp); 839 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 840 if (cachefound) { 841 struct vattr vattr; 842 843 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 844 if (error != 0) { 845 if (*vpp != NULLVP) 846 vrele(*vpp); 847 *vpp = NULLVP; 848 return error; 849 } 850 851 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 852 || timespeccmp(&vattr.va_mtime, 853 &VTONFS(dvp)->n_nctime, !=)) { 854 if (*vpp != NULLVP) { 855 vrele(*vpp); 856 *vpp = NULLVP; 857 } 858 cache_purge1(dvp, NULL, 0, PURGE_CHILDREN); 859 timespecclear(&np->n_nctime); 860 goto dorpc; 861 } 862 863 if (*vpp == NULLVP) { 864 /* namecache gave us a negative result */ 865 error = ENOENT; 866 goto noentry; 867 } 868 869 /* 870 * investigate the vnode returned by cache_lookup_raw. 871 * if it isn't appropriate, do an rpc. 872 */ 873 newvp = *vpp; 874 if ((flags & ISDOTDOT) != 0) { 875 VOP_UNLOCK(dvp); 876 } 877 error = vn_lock(newvp, LK_SHARED); 878 if ((flags & ISDOTDOT) != 0) { 879 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 880 } 881 if (error != 0) { 882 /* newvp has been reclaimed. */ 883 vrele(newvp); 884 *vpp = NULLVP; 885 goto dorpc; 886 } 887 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 888 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 889 nfsstats.lookupcache_hits++; 890 KASSERT(newvp->v_type != VNON); 891 VOP_UNLOCK(newvp); 892 return (0); 893 } 894 cache_purge1(newvp, NULL, 0, PURGE_PARENTS); 895 vput(newvp); 896 *vpp = NULLVP; 897 } 898 dorpc: 899 #if 0 900 /* 901 * because nfsv3 has the same CREATE semantics as ours, 902 * we don't have to perform LOOKUPs beforehand. 903 * 904 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 905 * XXX although we have no way to know if O_EXCL is requested or not. 906 */ 907 908 if (v3 && cnp->cn_nameiop == CREATE && 909 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 910 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 911 return (EJUSTRETURN); 912 } 913 #endif /* 0 */ 914 915 error = 0; 916 newvp = NULLVP; 917 nfsstats.lookupcache_misses++; 918 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 919 len = cnp->cn_namelen; 920 nfsm_reqhead(np, NFSPROC_LOOKUP, 921 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 922 nfsm_fhtom(np, v3); 923 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 924 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 925 if (error) { 926 nfsm_postop_attr(dvp, attrflag, 0); 927 m_freem(mrep); 928 goto nfsmout; 929 } 930 nfsm_getfh(fhp, fhsize, v3); 931 932 /* 933 * Handle RENAME case... 934 */ 935 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 936 if (NFS_CMPFH(np, fhp, fhsize)) { 937 m_freem(mrep); 938 return (EISDIR); 939 } 940 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 941 if (error) { 942 m_freem(mrep); 943 return error; 944 } 945 newvp = NFSTOV(np); 946 #ifndef NFS_V2_ONLY 947 if (v3) { 948 nfsm_postop_attr(newvp, attrflag, 0); 949 nfsm_postop_attr(dvp, attrflag, 0); 950 } else 951 #endif 952 nfsm_loadattr(newvp, (struct vattr *)0, 0); 953 *vpp = newvp; 954 m_freem(mrep); 955 goto validate; 956 } 957 958 /* 959 * The postop attr handling is duplicated for each if case, 960 * because it should be done while dvp is locked (unlocking 961 * dvp is different for each case). 962 */ 963 964 if (NFS_CMPFH(np, fhp, fhsize)) { 965 /* 966 * As we handle "." lookup locally, this is 967 * a broken server. 968 */ 969 m_freem(mrep); 970 return EBADRPC; 971 } else if (flags & ISDOTDOT) { 972 /* 973 * ".." lookup 974 */ 975 VOP_UNLOCK(dvp); 976 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 977 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 978 if (error) { 979 m_freem(mrep); 980 return error; 981 } 982 newvp = NFSTOV(np); 983 984 #ifndef NFS_V2_ONLY 985 if (v3) { 986 nfsm_postop_attr(newvp, attrflag, 0); 987 nfsm_postop_attr(dvp, attrflag, 0); 988 } else 989 #endif 990 nfsm_loadattr(newvp, (struct vattr *)0, 0); 991 } else { 992 /* 993 * Other lookups. 994 */ 995 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 996 if (error) { 997 m_freem(mrep); 998 return error; 999 } 1000 newvp = NFSTOV(np); 1001 #ifndef NFS_V2_ONLY 1002 if (v3) { 1003 nfsm_postop_attr(newvp, attrflag, 0); 1004 nfsm_postop_attr(dvp, attrflag, 0); 1005 } else 1006 #endif 1007 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1008 } 1009 if (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) { 1010 nfs_cache_enter(dvp, newvp, cnp); 1011 } 1012 *vpp = newvp; 1013 nfsm_reqdone; 1014 if (error) { 1015 /* 1016 * We get here only because of errors returned by 1017 * the RPC. Otherwise we'll have returned above 1018 * (the nfsm_* macros will jump to nfsm_reqdone 1019 * on error). 1020 */ 1021 if (error == ENOENT && cnp->cn_nameiop != CREATE) { 1022 nfs_cache_enter(dvp, NULL, cnp); 1023 } 1024 if (newvp != NULLVP) { 1025 if (newvp == dvp) { 1026 vrele(newvp); 1027 } else { 1028 vput(newvp); 1029 } 1030 } 1031 noentry: 1032 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1033 (flags & ISLASTCN) && error == ENOENT) { 1034 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1035 error = EROFS; 1036 } else { 1037 error = EJUSTRETURN; 1038 } 1039 } 1040 *vpp = NULL; 1041 return error; 1042 } 1043 1044 validate: 1045 /* 1046 * make sure we have valid type and size. 1047 */ 1048 1049 newvp = *vpp; 1050 if (newvp->v_type == VNON) { 1051 struct vattr vattr; /* dummy */ 1052 1053 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1054 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1055 if (error) { 1056 vput(newvp); 1057 *vpp = NULL; 1058 } 1059 } 1060 if (error) 1061 return error; 1062 if (newvp != dvp) 1063 VOP_UNLOCK(newvp); 1064 return 0; 1065 } 1066 1067 /* 1068 * nfs read call. 1069 * Just call nfs_bioread() to do the work. 1070 */ 1071 int 1072 nfs_read(void *v) 1073 { 1074 struct vop_read_args /* { 1075 struct vnode *a_vp; 1076 struct uio *a_uio; 1077 int a_ioflag; 1078 kauth_cred_t a_cred; 1079 } */ *ap = v; 1080 struct vnode *vp = ap->a_vp; 1081 1082 if (vp->v_type != VREG) 1083 return EISDIR; 1084 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1085 } 1086 1087 /* 1088 * nfs readlink call 1089 */ 1090 int 1091 nfs_readlink(void *v) 1092 { 1093 struct vop_readlink_args /* { 1094 struct vnode *a_vp; 1095 struct uio *a_uio; 1096 kauth_cred_t a_cred; 1097 } */ *ap = v; 1098 struct vnode *vp = ap->a_vp; 1099 struct nfsnode *np = VTONFS(vp); 1100 1101 if (vp->v_type != VLNK) 1102 return (EPERM); 1103 1104 if (np->n_rcred != NULL) { 1105 kauth_cred_free(np->n_rcred); 1106 } 1107 np->n_rcred = ap->a_cred; 1108 kauth_cred_hold(np->n_rcred); 1109 1110 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1111 } 1112 1113 /* 1114 * Do a readlink rpc. 1115 * Called by nfs_doio() from below the buffer cache. 1116 */ 1117 int 1118 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1119 { 1120 u_int32_t *tl; 1121 char *cp; 1122 int32_t t1, t2; 1123 char *bpos, *dpos, *cp2; 1124 int error = 0; 1125 uint32_t len; 1126 struct mbuf *mreq, *mrep, *md, *mb; 1127 const int v3 = NFS_ISV3(vp); 1128 struct nfsnode *np = VTONFS(vp); 1129 #ifndef NFS_V2_ONLY 1130 int attrflag; 1131 #endif 1132 1133 nfsstats.rpccnt[NFSPROC_READLINK]++; 1134 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1135 nfsm_fhtom(np, v3); 1136 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1137 #ifndef NFS_V2_ONLY 1138 if (v3) 1139 nfsm_postop_attr(vp, attrflag, 0); 1140 #endif 1141 if (!error) { 1142 #ifndef NFS_V2_ONLY 1143 if (v3) { 1144 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1145 len = fxdr_unsigned(uint32_t, *tl); 1146 if (len > NFS_MAXPATHLEN) { 1147 /* 1148 * this pathname is too long for us. 1149 */ 1150 m_freem(mrep); 1151 /* Solaris returns EINVAL. should we follow? */ 1152 error = ENAMETOOLONG; 1153 goto nfsmout; 1154 } 1155 } else 1156 #endif 1157 { 1158 nfsm_strsiz(len, NFS_MAXPATHLEN); 1159 } 1160 nfsm_mtouio(uiop, len); 1161 } 1162 nfsm_reqdone; 1163 return (error); 1164 } 1165 1166 /* 1167 * nfs read rpc call 1168 * Ditto above 1169 */ 1170 int 1171 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1172 { 1173 u_int32_t *tl; 1174 char *cp; 1175 int32_t t1, t2; 1176 char *bpos, *dpos, *cp2; 1177 struct mbuf *mreq, *mrep, *md, *mb; 1178 struct nfsmount *nmp; 1179 int error = 0, len, retlen, tsiz, eof __unused, byte_count; 1180 const int v3 = NFS_ISV3(vp); 1181 struct nfsnode *np = VTONFS(vp); 1182 #ifndef NFS_V2_ONLY 1183 int attrflag; 1184 #endif 1185 1186 #ifndef nolint 1187 eof = 0; 1188 #endif 1189 nmp = VFSTONFS(vp->v_mount); 1190 tsiz = uiop->uio_resid; 1191 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1192 return (EFBIG); 1193 iostat_busy(nmp->nm_stats); 1194 byte_count = 0; /* count bytes actually transferred */ 1195 while (tsiz > 0) { 1196 nfsstats.rpccnt[NFSPROC_READ]++; 1197 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1198 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1199 nfsm_fhtom(np, v3); 1200 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1201 #ifndef NFS_V2_ONLY 1202 if (v3) { 1203 txdr_hyper(uiop->uio_offset, tl); 1204 *(tl + 2) = txdr_unsigned(len); 1205 } else 1206 #endif 1207 { 1208 *tl++ = txdr_unsigned(uiop->uio_offset); 1209 *tl++ = txdr_unsigned(len); 1210 *tl = 0; 1211 } 1212 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1213 #ifndef NFS_V2_ONLY 1214 if (v3) { 1215 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1216 if (error) { 1217 m_freem(mrep); 1218 goto nfsmout; 1219 } 1220 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1221 eof = fxdr_unsigned(int, *(tl + 1)); 1222 } else 1223 #endif 1224 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1225 nfsm_strsiz(retlen, nmp->nm_rsize); 1226 nfsm_mtouio(uiop, retlen); 1227 m_freem(mrep); 1228 tsiz -= retlen; 1229 byte_count += retlen; 1230 #ifndef NFS_V2_ONLY 1231 if (v3) { 1232 if (eof || retlen == 0) 1233 tsiz = 0; 1234 } else 1235 #endif 1236 if (retlen < len) 1237 tsiz = 0; 1238 } 1239 nfsmout: 1240 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1241 return (error); 1242 } 1243 1244 struct nfs_writerpc_context { 1245 kmutex_t nwc_lock; 1246 kcondvar_t nwc_cv; 1247 int nwc_mbufcount; 1248 }; 1249 1250 /* 1251 * free mbuf used to refer protected pages while write rpc call. 1252 * called at splvm. 1253 */ 1254 static void 1255 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1256 { 1257 struct nfs_writerpc_context *ctx = arg; 1258 1259 KASSERT(m != NULL); 1260 KASSERT(ctx != NULL); 1261 pool_cache_put(mb_cache, m); 1262 mutex_enter(&ctx->nwc_lock); 1263 if (--ctx->nwc_mbufcount == 0) { 1264 cv_signal(&ctx->nwc_cv); 1265 } 1266 mutex_exit(&ctx->nwc_lock); 1267 } 1268 1269 /* 1270 * nfs write call 1271 */ 1272 int 1273 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1274 { 1275 u_int32_t *tl; 1276 char *cp; 1277 int32_t t1, t2; 1278 char *bpos, *dpos; 1279 struct mbuf *mreq, *mrep, *md, *mb; 1280 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1281 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1282 const int v3 = NFS_ISV3(vp); 1283 int committed = NFSV3WRITE_FILESYNC; 1284 struct nfsnode *np = VTONFS(vp); 1285 struct nfs_writerpc_context ctx; 1286 int byte_count; 1287 size_t origresid; 1288 #ifndef NFS_V2_ONLY 1289 char *cp2; 1290 int rlen, commit; 1291 #endif 1292 1293 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1294 panic("writerpc readonly vp %p", vp); 1295 } 1296 1297 #ifdef DIAGNOSTIC 1298 if (uiop->uio_iovcnt != 1) 1299 panic("nfs: writerpc iovcnt > 1"); 1300 #endif 1301 tsiz = uiop->uio_resid; 1302 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1303 return EFBIG; 1304 1305 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1306 cv_init(&ctx.nwc_cv, "nfsmblk"); 1307 ctx.nwc_mbufcount = 1; 1308 1309 retry: 1310 origresid = uiop->uio_resid; 1311 KASSERT(origresid == uiop->uio_iov->iov_len); 1312 iostat_busy(nmp->nm_stats); 1313 byte_count = 0; /* count of bytes actually written */ 1314 while (tsiz > 0) { 1315 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1316 size_t backup; 1317 bool stalewriteverf = false; 1318 1319 nfsstats.rpccnt[NFSPROC_WRITE]++; 1320 len = uimin(tsiz, nmp->nm_wsize); 1321 datalen = pageprotected ? 0 : nfsm_rndup(len); 1322 nfsm_reqhead(np, NFSPROC_WRITE, 1323 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1324 nfsm_fhtom(np, v3); 1325 #ifndef NFS_V2_ONLY 1326 if (v3) { 1327 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1328 txdr_hyper(uiop->uio_offset, tl); 1329 tl += 2; 1330 *tl++ = txdr_unsigned(len); 1331 *tl++ = txdr_unsigned(*iomode); 1332 *tl = txdr_unsigned(len); 1333 } else 1334 #endif 1335 { 1336 u_int32_t x; 1337 1338 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1339 /* Set both "begin" and "current" to non-garbage. */ 1340 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1341 *tl++ = x; /* "begin offset" */ 1342 *tl++ = x; /* "current offset" */ 1343 x = txdr_unsigned(len); 1344 *tl++ = x; /* total to this offset */ 1345 *tl = x; /* size of this write */ 1346 1347 } 1348 if (pageprotected) { 1349 /* 1350 * since we know pages can't be modified during i/o, 1351 * no need to copy them for us. 1352 */ 1353 struct mbuf *m; 1354 struct iovec *iovp = uiop->uio_iov; 1355 1356 m = m_get(M_WAIT, MT_DATA); 1357 MCLAIM(m, &nfs_mowner); 1358 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1359 nfs_writerpc_extfree, &ctx); 1360 m->m_flags |= M_EXT_ROMAP; 1361 m->m_len = len; 1362 mb->m_next = m; 1363 /* 1364 * no need to maintain mb and bpos here 1365 * because no one care them later. 1366 */ 1367 #if 0 1368 mb = m; 1369 bpos = mtod(void *, mb) + mb->m_len; 1370 #endif 1371 UIO_ADVANCE(uiop, len); 1372 uiop->uio_offset += len; 1373 mutex_enter(&ctx.nwc_lock); 1374 ctx.nwc_mbufcount++; 1375 mutex_exit(&ctx.nwc_lock); 1376 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1377 } else { 1378 nfsm_uiotom(uiop, len); 1379 } 1380 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1381 #ifndef NFS_V2_ONLY 1382 if (v3) { 1383 wccflag = NFSV3_WCCCHK; 1384 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1385 if (!error) { 1386 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1387 + NFSX_V3WRITEVERF); 1388 rlen = fxdr_unsigned(int, *tl++); 1389 if (rlen == 0) { 1390 error = NFSERR_IO; 1391 m_freem(mrep); 1392 break; 1393 } else if (rlen < len) { 1394 backup = len - rlen; 1395 UIO_ADVANCE(uiop, -backup); 1396 uiop->uio_offset -= backup; 1397 len = rlen; 1398 } 1399 commit = fxdr_unsigned(int, *tl++); 1400 1401 /* 1402 * Return the lowest committment level 1403 * obtained by any of the RPCs. 1404 */ 1405 if (committed == NFSV3WRITE_FILESYNC) 1406 committed = commit; 1407 else if (committed == NFSV3WRITE_DATASYNC && 1408 commit == NFSV3WRITE_UNSTABLE) 1409 committed = commit; 1410 mutex_enter(&nmp->nm_lock); 1411 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1412 memcpy(nmp->nm_writeverf, tl, 1413 NFSX_V3WRITEVERF); 1414 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1415 } else if ((nmp->nm_iflag & 1416 NFSMNT_STALEWRITEVERF) || 1417 memcmp(tl, nmp->nm_writeverf, 1418 NFSX_V3WRITEVERF)) { 1419 memcpy(nmp->nm_writeverf, tl, 1420 NFSX_V3WRITEVERF); 1421 /* 1422 * note NFSMNT_STALEWRITEVERF 1423 * if we're the first thread to 1424 * notice it. 1425 */ 1426 if ((nmp->nm_iflag & 1427 NFSMNT_STALEWRITEVERF) == 0) { 1428 stalewriteverf = true; 1429 nmp->nm_iflag |= 1430 NFSMNT_STALEWRITEVERF; 1431 } 1432 } 1433 mutex_exit(&nmp->nm_lock); 1434 } 1435 } else 1436 #endif 1437 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1438 if (wccflag) 1439 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1440 m_freem(mrep); 1441 if (error) 1442 break; 1443 tsiz -= len; 1444 byte_count += len; 1445 if (stalewriteverf) { 1446 *stalewriteverfp = true; 1447 stalewriteverf = false; 1448 if (committed == NFSV3WRITE_UNSTABLE && 1449 len != origresid) { 1450 /* 1451 * if our write requests weren't atomic but 1452 * unstable, datas in previous iterations 1453 * might have already been lost now. 1454 * then, we should resend them to nfsd. 1455 */ 1456 backup = origresid - tsiz; 1457 UIO_ADVANCE(uiop, -backup); 1458 uiop->uio_offset -= backup; 1459 tsiz = origresid; 1460 goto retry; 1461 } 1462 } 1463 } 1464 nfsmout: 1465 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1466 if (pageprotected) { 1467 /* 1468 * wait until mbufs go away. 1469 * retransmitted mbufs can survive longer than rpc requests 1470 * themselves. 1471 */ 1472 mutex_enter(&ctx.nwc_lock); 1473 ctx.nwc_mbufcount--; 1474 while (ctx.nwc_mbufcount > 0) { 1475 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1476 } 1477 mutex_exit(&ctx.nwc_lock); 1478 } 1479 mutex_destroy(&ctx.nwc_lock); 1480 cv_destroy(&ctx.nwc_cv); 1481 *iomode = committed; 1482 if (error) 1483 uiop->uio_resid = tsiz; 1484 return (error); 1485 } 1486 1487 /* 1488 * nfs mknod rpc 1489 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1490 * mode set to specify the file type and the size field for rdev. 1491 */ 1492 int 1493 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1494 { 1495 struct nfsv2_sattr *sp; 1496 u_int32_t *tl; 1497 char *cp; 1498 int32_t t1, t2; 1499 struct vnode *newvp = (struct vnode *)0; 1500 struct nfsnode *dnp, *np; 1501 char *cp2; 1502 char *bpos, *dpos; 1503 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1504 struct mbuf *mreq, *mrep, *md, *mb; 1505 u_int32_t rdev; 1506 const int v3 = NFS_ISV3(dvp); 1507 1508 if (vap->va_type == VCHR || vap->va_type == VBLK) 1509 rdev = txdr_unsigned(vap->va_rdev); 1510 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1511 rdev = nfs_xdrneg1; 1512 else { 1513 VOP_ABORTOP(dvp, cnp); 1514 return (EOPNOTSUPP); 1515 } 1516 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1517 dnp = VTONFS(dvp); 1518 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1519 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1520 nfsm_fhtom(dnp, v3); 1521 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1522 #ifndef NFS_V2_ONLY 1523 if (v3) { 1524 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1525 *tl++ = vtonfsv3_type(vap->va_type); 1526 nfsm_v3attrbuild(vap, false); 1527 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1528 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1529 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1530 *tl = txdr_unsigned(minor(vap->va_rdev)); 1531 } 1532 } else 1533 #endif 1534 { 1535 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1536 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1537 sp->sa_uid = nfs_xdrneg1; 1538 sp->sa_gid = nfs_xdrneg1; 1539 sp->sa_size = rdev; 1540 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1541 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1542 } 1543 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1544 if (!error) { 1545 nfsm_mtofh(dvp, newvp, v3, gotvp); 1546 if (!gotvp) { 1547 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1548 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1549 if (!error) 1550 newvp = NFSTOV(np); 1551 } 1552 } 1553 #ifndef NFS_V2_ONLY 1554 if (v3) 1555 nfsm_wcc_data(dvp, wccflag, 0, !error); 1556 #endif 1557 nfsm_reqdone; 1558 if (error) { 1559 if (newvp) 1560 vput(newvp); 1561 } else { 1562 nfs_cache_enter(dvp, newvp, cnp); 1563 *vpp = newvp; 1564 VOP_UNLOCK(newvp); 1565 } 1566 VTONFS(dvp)->n_flag |= NMODIFIED; 1567 if (!wccflag) 1568 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1569 return (error); 1570 } 1571 1572 /* 1573 * nfs mknod vop 1574 * just call nfs_mknodrpc() to do the work. 1575 */ 1576 /* ARGSUSED */ 1577 int 1578 nfs_mknod(void *v) 1579 { 1580 struct vop_mknod_v3_args /* { 1581 struct vnode *a_dvp; 1582 struct vnode **a_vpp; 1583 struct componentname *a_cnp; 1584 struct vattr *a_vap; 1585 } */ *ap = v; 1586 struct vnode *dvp = ap->a_dvp; 1587 struct componentname *cnp = ap->a_cnp; 1588 int error; 1589 1590 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1591 VN_KNOTE(dvp, NOTE_WRITE); 1592 if (error == 0 || error == EEXIST) 1593 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1594 return (error); 1595 } 1596 1597 /* 1598 * nfs file create call 1599 */ 1600 int 1601 nfs_create(void *v) 1602 { 1603 struct vop_create_v3_args /* { 1604 struct vnode *a_dvp; 1605 struct vnode **a_vpp; 1606 struct componentname *a_cnp; 1607 struct vattr *a_vap; 1608 } */ *ap = v; 1609 struct vnode *dvp = ap->a_dvp; 1610 struct vattr *vap = ap->a_vap; 1611 struct componentname *cnp = ap->a_cnp; 1612 struct nfsv2_sattr *sp; 1613 u_int32_t *tl; 1614 char *cp; 1615 int32_t t1, t2; 1616 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1617 struct vnode *newvp = (struct vnode *)0; 1618 char *bpos, *dpos, *cp2; 1619 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1620 struct mbuf *mreq, *mrep, *md, *mb; 1621 const int v3 = NFS_ISV3(dvp); 1622 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1623 1624 /* 1625 * Oops, not for me.. 1626 */ 1627 if (vap->va_type == VSOCK) 1628 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1629 1630 KASSERT(vap->va_type == VREG); 1631 1632 #ifdef VA_EXCLUSIVE 1633 if (vap->va_vaflags & VA_EXCLUSIVE) { 1634 excl_mode = NFSV3CREATE_EXCLUSIVE; 1635 } 1636 #endif 1637 again: 1638 error = 0; 1639 nfsstats.rpccnt[NFSPROC_CREATE]++; 1640 dnp = VTONFS(dvp); 1641 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1642 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1643 nfsm_fhtom(dnp, v3); 1644 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1645 #ifndef NFS_V2_ONLY 1646 if (v3) { 1647 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1648 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1649 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1650 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1651 *tl++ = cprng_fast32(); 1652 *tl = cprng_fast32(); 1653 } else { 1654 *tl = txdr_unsigned(excl_mode); 1655 nfsm_v3attrbuild(vap, false); 1656 } 1657 } else 1658 #endif 1659 { 1660 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1661 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1662 sp->sa_uid = nfs_xdrneg1; 1663 sp->sa_gid = nfs_xdrneg1; 1664 sp->sa_size = 0; 1665 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1666 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1667 } 1668 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1669 if (!error) { 1670 nfsm_mtofh(dvp, newvp, v3, gotvp); 1671 if (!gotvp) { 1672 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1673 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1674 if (!error) 1675 newvp = NFSTOV(np); 1676 } 1677 } 1678 #ifndef NFS_V2_ONLY 1679 if (v3) 1680 nfsm_wcc_data(dvp, wccflag, 0, !error); 1681 #endif 1682 nfsm_reqdone; 1683 if (error) { 1684 /* 1685 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1686 */ 1687 if (v3 && error == ENOTSUP) { 1688 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1689 excl_mode = NFSV3CREATE_GUARDED; 1690 goto again; 1691 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1692 excl_mode = NFSV3CREATE_UNCHECKED; 1693 goto again; 1694 } 1695 } 1696 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1697 struct timespec ts; 1698 1699 getnanotime(&ts); 1700 1701 /* 1702 * make sure that we'll update timestamps as 1703 * most server implementations use them to store 1704 * the create verifier. 1705 * 1706 * XXX it's better to use TOSERVER always. 1707 */ 1708 1709 if (vap->va_atime.tv_sec == VNOVAL) 1710 vap->va_atime = ts; 1711 if (vap->va_mtime.tv_sec == VNOVAL) 1712 vap->va_mtime = ts; 1713 1714 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1715 } 1716 if (error == 0) { 1717 if (cnp->cn_flags & MAKEENTRY) 1718 nfs_cache_enter(dvp, newvp, cnp); 1719 else 1720 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1721 *ap->a_vpp = newvp; 1722 VOP_UNLOCK(newvp); 1723 } else { 1724 if (newvp) 1725 vput(newvp); 1726 if (error == EEXIST) 1727 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1728 } 1729 VTONFS(dvp)->n_flag |= NMODIFIED; 1730 if (!wccflag) 1731 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1732 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1733 return (error); 1734 } 1735 1736 /* 1737 * nfs file remove call 1738 * To try and make nfs semantics closer to ufs semantics, a file that has 1739 * other processes using the vnode is renamed instead of removed and then 1740 * removed later on the last close. 1741 * - If vrefcnt(vp) > 1 1742 * If a rename is not already in the works 1743 * call nfs_sillyrename() to set it up 1744 * else 1745 * do the remove rpc 1746 */ 1747 int 1748 nfs_remove(void *v) 1749 { 1750 struct vop_remove_v2_args /* { 1751 struct vnodeop_desc *a_desc; 1752 struct vnode * a_dvp; 1753 struct vnode * a_vp; 1754 struct componentname * a_cnp; 1755 } */ *ap = v; 1756 struct vnode *vp = ap->a_vp; 1757 struct vnode *dvp = ap->a_dvp; 1758 struct componentname *cnp = ap->a_cnp; 1759 struct nfsnode *np = VTONFS(vp); 1760 int error = 0; 1761 struct vattr vattr; 1762 1763 #ifndef DIAGNOSTIC 1764 if (vrefcnt(vp) < 1) 1765 panic("nfs_remove: bad vrefcnt(vp)"); 1766 #endif 1767 if (vp->v_type == VDIR) 1768 error = EPERM; 1769 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1770 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1771 vattr.va_nlink > 1)) { 1772 /* 1773 * Purge the name cache so that the chance of a lookup for 1774 * the name succeeding while the remove is in progress is 1775 * minimized. Without node locking it can still happen, such 1776 * that an I/O op returns ESTALE, but since you get this if 1777 * another host removes the file.. 1778 */ 1779 cache_purge(vp); 1780 /* 1781 * throw away biocache buffers, mainly to avoid 1782 * unnecessary delayed writes later. 1783 */ 1784 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1785 /* Do the rpc */ 1786 if (error != EINTR) 1787 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1788 cnp->cn_namelen, cnp->cn_cred, curlwp); 1789 } else if (!np->n_sillyrename) 1790 error = nfs_sillyrename(dvp, vp, cnp, false); 1791 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1792 vattr.va_nlink == 1) { 1793 np->n_flag |= NREMOVED; 1794 } 1795 NFS_INVALIDATE_ATTRCACHE(np); 1796 VN_KNOTE(vp, NOTE_DELETE); 1797 VN_KNOTE(dvp, NOTE_WRITE); 1798 if (dvp == vp) 1799 vrele(vp); 1800 else 1801 vput(vp); 1802 return (error); 1803 } 1804 1805 /* 1806 * nfs file remove rpc called from nfs_inactive 1807 */ 1808 int 1809 nfs_removeit(struct sillyrename *sp) 1810 { 1811 1812 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1813 (struct lwp *)0)); 1814 } 1815 1816 /* 1817 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1818 */ 1819 int 1820 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1821 { 1822 u_int32_t *tl; 1823 char *cp; 1824 #ifndef NFS_V2_ONLY 1825 int32_t t1; 1826 char *cp2; 1827 #endif 1828 int32_t t2; 1829 char *bpos, *dpos; 1830 int error = 0, wccflag = NFSV3_WCCRATTR; 1831 struct mbuf *mreq, *mrep, *md, *mb; 1832 const int v3 = NFS_ISV3(dvp); 1833 int rexmit = 0; 1834 struct nfsnode *dnp = VTONFS(dvp); 1835 1836 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1837 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1838 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1839 nfsm_fhtom(dnp, v3); 1840 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1841 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1842 #ifndef NFS_V2_ONLY 1843 if (v3) 1844 nfsm_wcc_data(dvp, wccflag, 0, !error); 1845 #endif 1846 nfsm_reqdone; 1847 VTONFS(dvp)->n_flag |= NMODIFIED; 1848 if (!wccflag) 1849 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1850 /* 1851 * Kludge City: If the first reply to the remove rpc is lost.. 1852 * the reply to the retransmitted request will be ENOENT 1853 * since the file was in fact removed 1854 * Therefore, we cheat and return success. 1855 */ 1856 if (rexmit && error == ENOENT) 1857 error = 0; 1858 return (error); 1859 } 1860 1861 /* 1862 * nfs file rename call 1863 */ 1864 int 1865 nfs_rename(void *v) 1866 { 1867 struct vop_rename_args /* { 1868 struct vnode *a_fdvp; 1869 struct vnode *a_fvp; 1870 struct componentname *a_fcnp; 1871 struct vnode *a_tdvp; 1872 struct vnode *a_tvp; 1873 struct componentname *a_tcnp; 1874 } */ *ap = v; 1875 struct vnode *fvp = ap->a_fvp; 1876 struct vnode *tvp = ap->a_tvp; 1877 struct vnode *fdvp = ap->a_fdvp; 1878 struct vnode *tdvp = ap->a_tdvp; 1879 struct componentname *tcnp = ap->a_tcnp; 1880 struct componentname *fcnp = ap->a_fcnp; 1881 int error; 1882 1883 /* Check for cross-device rename */ 1884 if ((fvp->v_mount != tdvp->v_mount) || 1885 (tvp && (fvp->v_mount != tvp->v_mount))) { 1886 error = EXDEV; 1887 goto out; 1888 } 1889 1890 /* 1891 * If the tvp exists and is in use, sillyrename it before doing the 1892 * rename of the new file over it. 1893 * 1894 * Have sillyrename use link instead of rename if possible, 1895 * so that we don't lose the file if the rename fails, and so 1896 * that there's no window when the "to" file doesn't exist. 1897 */ 1898 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1899 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1900 VN_KNOTE(tvp, NOTE_DELETE); 1901 vput(tvp); 1902 tvp = NULL; 1903 } 1904 1905 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1906 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1907 curlwp); 1908 1909 VN_KNOTE(fdvp, NOTE_WRITE); 1910 VN_KNOTE(tdvp, NOTE_WRITE); 1911 if (error == 0 || error == EEXIST) { 1912 if (fvp->v_type == VDIR) 1913 cache_purge(fvp); 1914 else 1915 cache_purge1(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1916 0); 1917 if (tvp != NULL && tvp->v_type == VDIR) 1918 cache_purge(tvp); 1919 else 1920 cache_purge1(tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, 1921 0); 1922 } 1923 out: 1924 if (tdvp == tvp) 1925 vrele(tdvp); 1926 else 1927 vput(tdvp); 1928 if (tvp) 1929 vput(tvp); 1930 vrele(fdvp); 1931 vrele(fvp); 1932 return (error); 1933 } 1934 1935 /* 1936 * nfs file rename rpc called from nfs_remove() above 1937 */ 1938 int 1939 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1940 { 1941 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1942 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1943 } 1944 1945 /* 1946 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1947 */ 1948 int 1949 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) 1950 { 1951 u_int32_t *tl; 1952 char *cp; 1953 #ifndef NFS_V2_ONLY 1954 int32_t t1; 1955 char *cp2; 1956 #endif 1957 int32_t t2; 1958 char *bpos, *dpos; 1959 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1960 struct mbuf *mreq, *mrep, *md, *mb; 1961 const int v3 = NFS_ISV3(fdvp); 1962 int rexmit = 0; 1963 struct nfsnode *fdnp = VTONFS(fdvp); 1964 1965 nfsstats.rpccnt[NFSPROC_RENAME]++; 1966 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1967 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1968 nfsm_rndup(tnamelen)); 1969 nfsm_fhtom(fdnp, v3); 1970 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1971 nfsm_fhtom(VTONFS(tdvp), v3); 1972 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1973 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1974 #ifndef NFS_V2_ONLY 1975 if (v3) { 1976 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1977 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1978 } 1979 #endif 1980 nfsm_reqdone; 1981 VTONFS(fdvp)->n_flag |= NMODIFIED; 1982 VTONFS(tdvp)->n_flag |= NMODIFIED; 1983 if (!fwccflag) 1984 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1985 if (!twccflag) 1986 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1987 /* 1988 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1989 */ 1990 if (rexmit && error == ENOENT) 1991 error = 0; 1992 return (error); 1993 } 1994 1995 /* 1996 * NFS link RPC, called from nfs_link. 1997 * Assumes dvp and vp locked, and leaves them that way. 1998 */ 1999 2000 static int 2001 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2002 size_t namelen, kauth_cred_t cred, struct lwp *l) 2003 { 2004 u_int32_t *tl; 2005 char *cp; 2006 #ifndef NFS_V2_ONLY 2007 int32_t t1; 2008 char *cp2; 2009 #endif 2010 int32_t t2; 2011 char *bpos, *dpos; 2012 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2013 struct mbuf *mreq, *mrep, *md, *mb; 2014 const int v3 = NFS_ISV3(dvp); 2015 int rexmit = 0; 2016 struct nfsnode *np = VTONFS(vp); 2017 2018 nfsstats.rpccnt[NFSPROC_LINK]++; 2019 nfsm_reqhead(np, NFSPROC_LINK, 2020 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2021 nfsm_fhtom(np, v3); 2022 nfsm_fhtom(VTONFS(dvp), v3); 2023 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2024 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2025 #ifndef NFS_V2_ONLY 2026 if (v3) { 2027 nfsm_postop_attr(vp, attrflag, 0); 2028 nfsm_wcc_data(dvp, wccflag, 0, !error); 2029 } 2030 #endif 2031 nfsm_reqdone; 2032 2033 VTONFS(dvp)->n_flag |= NMODIFIED; 2034 if (!attrflag) 2035 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2036 if (!wccflag) 2037 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2038 2039 /* 2040 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2041 */ 2042 if (rexmit && error == EEXIST) 2043 error = 0; 2044 2045 return error; 2046 } 2047 2048 /* 2049 * nfs hard link create call 2050 */ 2051 int 2052 nfs_link(void *v) 2053 { 2054 struct vop_link_v2_args /* { 2055 struct vnode *a_dvp; 2056 struct vnode *a_vp; 2057 struct componentname *a_cnp; 2058 } */ *ap = v; 2059 struct vnode *vp = ap->a_vp; 2060 struct vnode *dvp = ap->a_dvp; 2061 struct componentname *cnp = ap->a_cnp; 2062 int error = 0; 2063 2064 error = vn_lock(vp, LK_EXCLUSIVE); 2065 if (error != 0) { 2066 VOP_ABORTOP(dvp, cnp); 2067 return error; 2068 } 2069 2070 /* 2071 * Push all writes to the server, so that the attribute cache 2072 * doesn't get "out of sync" with the server. 2073 * XXX There should be a better way! 2074 */ 2075 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2076 2077 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2078 cnp->cn_cred, curlwp); 2079 2080 if (error == 0) { 2081 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2082 } 2083 VOP_UNLOCK(vp); 2084 VN_KNOTE(vp, NOTE_LINK); 2085 VN_KNOTE(dvp, NOTE_WRITE); 2086 return (error); 2087 } 2088 2089 /* 2090 * nfs symbolic link create call 2091 */ 2092 int 2093 nfs_symlink(void *v) 2094 { 2095 struct vop_symlink_v3_args /* { 2096 struct vnode *a_dvp; 2097 struct vnode **a_vpp; 2098 struct componentname *a_cnp; 2099 struct vattr *a_vap; 2100 char *a_target; 2101 } */ *ap = v; 2102 struct vnode *dvp = ap->a_dvp; 2103 struct vattr *vap = ap->a_vap; 2104 struct componentname *cnp = ap->a_cnp; 2105 struct nfsv2_sattr *sp; 2106 u_int32_t *tl; 2107 char *cp; 2108 int32_t t1, t2; 2109 char *bpos, *dpos, *cp2; 2110 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2111 struct mbuf *mreq, *mrep, *md, *mb; 2112 struct vnode *newvp = (struct vnode *)0; 2113 const int v3 = NFS_ISV3(dvp); 2114 int rexmit = 0; 2115 struct nfsnode *dnp = VTONFS(dvp); 2116 2117 *ap->a_vpp = NULL; 2118 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2119 slen = strlen(ap->a_target); 2120 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2121 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2122 nfsm_fhtom(dnp, v3); 2123 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2124 #ifndef NFS_V2_ONlY 2125 if (v3) 2126 nfsm_v3attrbuild(vap, false); 2127 #endif 2128 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2129 #ifndef NFS_V2_ONlY 2130 if (!v3) { 2131 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2132 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2133 sp->sa_uid = nfs_xdrneg1; 2134 sp->sa_gid = nfs_xdrneg1; 2135 sp->sa_size = nfs_xdrneg1; 2136 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2137 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2138 } 2139 #endif 2140 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2141 &rexmit); 2142 #ifndef NFS_V2_ONlY 2143 if (v3) { 2144 if (!error) 2145 nfsm_mtofh(dvp, newvp, v3, gotvp); 2146 nfsm_wcc_data(dvp, wccflag, 0, !error); 2147 } 2148 #endif 2149 nfsm_reqdone; 2150 /* 2151 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2152 */ 2153 if (rexmit && error == EEXIST) 2154 error = 0; 2155 if (error == 0 || error == EEXIST) 2156 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2157 if (error == 0 && newvp == NULL) { 2158 struct nfsnode *np = NULL; 2159 2160 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2161 cnp->cn_cred, curlwp, &np); 2162 if (error == 0) 2163 newvp = NFSTOV(np); 2164 } 2165 if (error) { 2166 if (newvp != NULL) 2167 vput(newvp); 2168 } else { 2169 *ap->a_vpp = newvp; 2170 VOP_UNLOCK(newvp); 2171 } 2172 VTONFS(dvp)->n_flag |= NMODIFIED; 2173 if (!wccflag) 2174 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2175 VN_KNOTE(dvp, NOTE_WRITE); 2176 return (error); 2177 } 2178 2179 /* 2180 * nfs make dir call 2181 */ 2182 int 2183 nfs_mkdir(void *v) 2184 { 2185 struct vop_mkdir_v3_args /* { 2186 struct vnode *a_dvp; 2187 struct vnode **a_vpp; 2188 struct componentname *a_cnp; 2189 struct vattr *a_vap; 2190 } */ *ap = v; 2191 struct vnode *dvp = ap->a_dvp; 2192 struct vattr *vap = ap->a_vap; 2193 struct componentname *cnp = ap->a_cnp; 2194 struct nfsv2_sattr *sp; 2195 u_int32_t *tl; 2196 char *cp; 2197 int32_t t1, t2; 2198 int len; 2199 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2200 struct vnode *newvp = (struct vnode *)0; 2201 char *bpos, *dpos, *cp2; 2202 int error = 0, wccflag = NFSV3_WCCRATTR; 2203 int gotvp = 0; 2204 int rexmit = 0; 2205 struct mbuf *mreq, *mrep, *md, *mb; 2206 const int v3 = NFS_ISV3(dvp); 2207 2208 len = cnp->cn_namelen; 2209 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2210 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2211 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2212 nfsm_fhtom(dnp, v3); 2213 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2214 #ifndef NFS_V2_ONLY 2215 if (v3) { 2216 nfsm_v3attrbuild(vap, false); 2217 } else 2218 #endif 2219 { 2220 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2221 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2222 sp->sa_uid = nfs_xdrneg1; 2223 sp->sa_gid = nfs_xdrneg1; 2224 sp->sa_size = nfs_xdrneg1; 2225 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2226 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2227 } 2228 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2229 if (!error) 2230 nfsm_mtofh(dvp, newvp, v3, gotvp); 2231 if (v3) 2232 nfsm_wcc_data(dvp, wccflag, 0, !error); 2233 nfsm_reqdone; 2234 VTONFS(dvp)->n_flag |= NMODIFIED; 2235 if (!wccflag) 2236 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2237 /* 2238 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2239 * if we can succeed in looking up the directory. 2240 */ 2241 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2242 if (newvp) { 2243 vput(newvp); 2244 newvp = (struct vnode *)0; 2245 } 2246 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2247 curlwp, &np); 2248 if (!error) { 2249 newvp = NFSTOV(np); 2250 if (newvp->v_type != VDIR || newvp == dvp) 2251 error = EEXIST; 2252 } 2253 } 2254 if (error) { 2255 if (newvp) { 2256 if (dvp != newvp) 2257 vput(newvp); 2258 else 2259 vrele(newvp); 2260 } 2261 } else { 2262 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2263 nfs_cache_enter(dvp, newvp, cnp); 2264 *ap->a_vpp = newvp; 2265 VOP_UNLOCK(newvp); 2266 } 2267 return (error); 2268 } 2269 2270 /* 2271 * nfs remove directory call 2272 */ 2273 int 2274 nfs_rmdir(void *v) 2275 { 2276 struct vop_rmdir_v2_args /* { 2277 struct vnode *a_dvp; 2278 struct vnode *a_vp; 2279 struct componentname *a_cnp; 2280 } */ *ap = v; 2281 struct vnode *vp = ap->a_vp; 2282 struct vnode *dvp = ap->a_dvp; 2283 struct componentname *cnp = ap->a_cnp; 2284 u_int32_t *tl; 2285 char *cp; 2286 #ifndef NFS_V2_ONLY 2287 int32_t t1; 2288 char *cp2; 2289 #endif 2290 int32_t t2; 2291 char *bpos, *dpos; 2292 int error = 0, wccflag = NFSV3_WCCRATTR; 2293 int rexmit = 0; 2294 struct mbuf *mreq, *mrep, *md, *mb; 2295 const int v3 = NFS_ISV3(dvp); 2296 struct nfsnode *dnp; 2297 2298 if (dvp == vp) { 2299 vrele(vp); 2300 return (EINVAL); 2301 } 2302 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2303 dnp = VTONFS(dvp); 2304 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2305 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2306 nfsm_fhtom(dnp, v3); 2307 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2308 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2309 #ifndef NFS_V2_ONLY 2310 if (v3) 2311 nfsm_wcc_data(dvp, wccflag, 0, !error); 2312 #endif 2313 nfsm_reqdone; 2314 VTONFS(dvp)->n_flag |= NMODIFIED; 2315 if (!wccflag) 2316 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2317 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2318 VN_KNOTE(vp, NOTE_DELETE); 2319 cache_purge(vp); 2320 vput(vp); 2321 /* 2322 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2323 */ 2324 if (rexmit && error == ENOENT) 2325 error = 0; 2326 return (error); 2327 } 2328 2329 /* 2330 * nfs readdir call 2331 */ 2332 int 2333 nfs_readdir(void *v) 2334 { 2335 struct vop_readdir_args /* { 2336 struct vnode *a_vp; 2337 struct uio *a_uio; 2338 kauth_cred_t a_cred; 2339 int *a_eofflag; 2340 off_t **a_cookies; 2341 int *a_ncookies; 2342 } */ *ap = v; 2343 struct vnode *vp = ap->a_vp; 2344 struct uio *uio = ap->a_uio; 2345 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2346 char *base = uio->uio_iov->iov_base; 2347 int tresid, error; 2348 size_t count, lost; 2349 struct dirent *dp; 2350 off_t *cookies = NULL; 2351 int ncookies = 0, nc; 2352 2353 if (vp->v_type != VDIR) 2354 return (EPERM); 2355 2356 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2357 count = uio->uio_resid - lost; 2358 if (count <= 0) 2359 return (EINVAL); 2360 2361 /* 2362 * Call nfs_bioread() to do the real work. 2363 */ 2364 tresid = uio->uio_resid = count; 2365 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2366 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2367 2368 if (!error && ap->a_cookies) { 2369 ncookies = count / 16; 2370 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2371 *ap->a_cookies = cookies; 2372 } 2373 2374 if (!error && uio->uio_resid == tresid) { 2375 uio->uio_resid += lost; 2376 nfsstats.direofcache_misses++; 2377 if (ap->a_cookies) 2378 *ap->a_ncookies = 0; 2379 *ap->a_eofflag = 1; 2380 return (0); 2381 } 2382 2383 if (!error && ap->a_cookies) { 2384 /* 2385 * Only the NFS server and emulations use cookies, and they 2386 * load the directory block into system space, so we can 2387 * just look at it directly. 2388 */ 2389 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2390 uio->uio_iovcnt != 1) 2391 panic("nfs_readdir: lost in space"); 2392 for (nc = 0; ncookies-- && 2393 base < (char *)uio->uio_iov->iov_base; nc++){ 2394 dp = (struct dirent *) base; 2395 if (dp->d_reclen == 0) 2396 break; 2397 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2398 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2399 else 2400 *(cookies++) = NFS_GETCOOKIE(dp); 2401 base += dp->d_reclen; 2402 } 2403 uio->uio_resid += 2404 ((char *)uio->uio_iov->iov_base - base); 2405 uio->uio_iov->iov_len += 2406 ((char *)uio->uio_iov->iov_base - base); 2407 uio->uio_iov->iov_base = base; 2408 *ap->a_ncookies = nc; 2409 } 2410 2411 uio->uio_resid += lost; 2412 *ap->a_eofflag = 0; 2413 return (error); 2414 } 2415 2416 /* 2417 * Readdir rpc call. 2418 * Called from below the buffer cache by nfs_doio(). 2419 */ 2420 int 2421 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2422 { 2423 int len, left; 2424 struct dirent *dp = NULL; 2425 u_int32_t *tl; 2426 char *cp; 2427 int32_t t1, t2; 2428 char *bpos, *dpos, *cp2; 2429 struct mbuf *mreq, *mrep, *md, *mb; 2430 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2431 struct nfsnode *dnp = VTONFS(vp); 2432 u_quad_t fileno; 2433 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2434 #ifndef NFS_V2_ONLY 2435 int attrflag; 2436 #endif 2437 int nrpcs = 0, reclen; 2438 const int v3 = NFS_ISV3(vp); 2439 2440 #ifdef DIAGNOSTIC 2441 /* 2442 * Should be called from buffer cache, so only amount of 2443 * NFS_DIRBLKSIZ will be requested. 2444 */ 2445 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2446 panic("nfs readdirrpc bad uio"); 2447 #endif 2448 2449 /* 2450 * Loop around doing readdir rpc's of size nm_readdirsize 2451 * truncated to a multiple of NFS_DIRFRAGSIZ. 2452 * The stopping criteria is EOF or buffer full. 2453 */ 2454 while (more_dirs && bigenough) { 2455 /* 2456 * Heuristic: don't bother to do another RPC to further 2457 * fill up this block if there is not much room left. (< 50% 2458 * of the readdir RPC size). This wastes some buffer space 2459 * but can save up to 50% in RPC calls. 2460 */ 2461 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2462 bigenough = 0; 2463 break; 2464 } 2465 nfsstats.rpccnt[NFSPROC_READDIR]++; 2466 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2467 NFSX_READDIR(v3)); 2468 nfsm_fhtom(dnp, v3); 2469 #ifndef NFS_V2_ONLY 2470 if (v3) { 2471 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2472 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2473 txdr_swapcookie3(uiop->uio_offset, tl); 2474 } else { 2475 txdr_cookie3(uiop->uio_offset, tl); 2476 } 2477 tl += 2; 2478 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2479 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2480 } else 2481 #endif 2482 { 2483 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2484 *tl++ = txdr_unsigned(uiop->uio_offset); 2485 } 2486 *tl = txdr_unsigned(nmp->nm_readdirsize); 2487 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2488 nrpcs++; 2489 #ifndef NFS_V2_ONLY 2490 if (v3) { 2491 nfsm_postop_attr(vp, attrflag, 0); 2492 if (!error) { 2493 nfsm_dissect(tl, u_int32_t *, 2494 2 * NFSX_UNSIGNED); 2495 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2496 dnp->n_cookieverf.nfsuquad[1] = *tl; 2497 } else { 2498 m_freem(mrep); 2499 goto nfsmout; 2500 } 2501 } 2502 #endif 2503 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2504 more_dirs = fxdr_unsigned(int, *tl); 2505 2506 /* loop thru the dir entries, doctoring them to 4bsd form */ 2507 while (more_dirs && bigenough) { 2508 #ifndef NFS_V2_ONLY 2509 if (v3) { 2510 nfsm_dissect(tl, u_int32_t *, 2511 3 * NFSX_UNSIGNED); 2512 fileno = fxdr_hyper(tl); 2513 len = fxdr_unsigned(int, *(tl + 2)); 2514 } else 2515 #endif 2516 { 2517 nfsm_dissect(tl, u_int32_t *, 2518 2 * NFSX_UNSIGNED); 2519 fileno = fxdr_unsigned(u_quad_t, *tl++); 2520 len = fxdr_unsigned(int, *tl); 2521 } 2522 if (len <= 0 || len > NFS_MAXNAMLEN) { 2523 error = EBADRPC; 2524 m_freem(mrep); 2525 goto nfsmout; 2526 } 2527 /* for cookie stashing */ 2528 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2529 left = NFS_DIRFRAGSIZ - blksiz; 2530 if (reclen > left) { 2531 memset(uiop->uio_iov->iov_base, 0, left); 2532 dp->d_reclen += left; 2533 UIO_ADVANCE(uiop, left); 2534 blksiz = 0; 2535 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2536 } 2537 if (reclen > uiop->uio_resid) 2538 bigenough = 0; 2539 if (bigenough) { 2540 int tlen; 2541 2542 dp = (struct dirent *)uiop->uio_iov->iov_base; 2543 dp->d_fileno = fileno; 2544 dp->d_namlen = len; 2545 dp->d_reclen = reclen; 2546 dp->d_type = DT_UNKNOWN; 2547 blksiz += reclen; 2548 if (blksiz == NFS_DIRFRAGSIZ) 2549 blksiz = 0; 2550 UIO_ADVANCE(uiop, DIRHDSIZ); 2551 nfsm_mtouio(uiop, len); 2552 tlen = reclen - (DIRHDSIZ + len); 2553 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2554 UIO_ADVANCE(uiop, tlen); 2555 } else 2556 nfsm_adv(nfsm_rndup(len)); 2557 #ifndef NFS_V2_ONLY 2558 if (v3) { 2559 nfsm_dissect(tl, u_int32_t *, 2560 3 * NFSX_UNSIGNED); 2561 } else 2562 #endif 2563 { 2564 nfsm_dissect(tl, u_int32_t *, 2565 2 * NFSX_UNSIGNED); 2566 } 2567 if (bigenough) { 2568 #ifndef NFS_V2_ONLY 2569 if (v3) { 2570 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2571 uiop->uio_offset = 2572 fxdr_swapcookie3(tl); 2573 else 2574 uiop->uio_offset = 2575 fxdr_cookie3(tl); 2576 } 2577 else 2578 #endif 2579 { 2580 uiop->uio_offset = 2581 fxdr_unsigned(off_t, *tl); 2582 } 2583 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2584 } 2585 if (v3) 2586 tl += 2; 2587 else 2588 tl++; 2589 more_dirs = fxdr_unsigned(int, *tl); 2590 } 2591 /* 2592 * If at end of rpc data, get the eof boolean 2593 */ 2594 if (!more_dirs) { 2595 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2596 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2597 2598 /* 2599 * kludge: if we got no entries, treat it as EOF. 2600 * some server sometimes send a reply without any 2601 * entries or EOF. 2602 * although it might mean the server has very long name, 2603 * we can't handle such entries anyway. 2604 */ 2605 2606 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2607 more_dirs = 0; 2608 } 2609 m_freem(mrep); 2610 } 2611 /* 2612 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2613 * by increasing d_reclen for the last record. 2614 */ 2615 if (blksiz > 0) { 2616 left = NFS_DIRFRAGSIZ - blksiz; 2617 memset(uiop->uio_iov->iov_base, 0, left); 2618 dp->d_reclen += left; 2619 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2620 UIO_ADVANCE(uiop, left); 2621 } 2622 2623 /* 2624 * We are now either at the end of the directory or have filled the 2625 * block. 2626 */ 2627 if (bigenough) { 2628 dnp->n_direofoffset = uiop->uio_offset; 2629 dnp->n_flag |= NEOFVALID; 2630 } 2631 nfsmout: 2632 return (error); 2633 } 2634 2635 #ifndef NFS_V2_ONLY 2636 /* 2637 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2638 */ 2639 int 2640 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2641 { 2642 int len, left; 2643 struct dirent *dp = NULL; 2644 u_int32_t *tl; 2645 char *cp; 2646 int32_t t1, t2; 2647 struct vnode *newvp; 2648 char *bpos, *dpos, *cp2; 2649 struct mbuf *mreq, *mrep, *md, *mb; 2650 struct nameidata nami, *ndp = &nami; 2651 struct componentname *cnp = &ndp->ni_cnd; 2652 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2653 struct nfsnode *dnp = VTONFS(vp), *np; 2654 nfsfh_t *fhp; 2655 u_quad_t fileno; 2656 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2657 int attrflag, fhsize, nrpcs = 0, reclen; 2658 struct nfs_fattr fattr, *fp; 2659 2660 #ifdef DIAGNOSTIC 2661 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2662 panic("nfs readdirplusrpc bad uio"); 2663 #endif 2664 ndp->ni_dvp = vp; 2665 newvp = NULLVP; 2666 2667 /* 2668 * Loop around doing readdir rpc's of size nm_readdirsize 2669 * truncated to a multiple of NFS_DIRFRAGSIZ. 2670 * The stopping criteria is EOF or buffer full. 2671 */ 2672 while (more_dirs && bigenough) { 2673 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2674 bigenough = 0; 2675 break; 2676 } 2677 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2678 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2679 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2680 nfsm_fhtom(dnp, 1); 2681 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2682 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2683 txdr_swapcookie3(uiop->uio_offset, tl); 2684 } else { 2685 txdr_cookie3(uiop->uio_offset, tl); 2686 } 2687 tl += 2; 2688 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2689 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2690 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2691 *tl = txdr_unsigned(nmp->nm_rsize); 2692 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2693 nfsm_postop_attr(vp, attrflag, 0); 2694 if (error) { 2695 m_freem(mrep); 2696 goto nfsmout; 2697 } 2698 nrpcs++; 2699 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2700 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2701 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2702 more_dirs = fxdr_unsigned(int, *tl); 2703 2704 /* loop thru the dir entries, doctoring them to 4bsd form */ 2705 while (more_dirs && bigenough) { 2706 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2707 fileno = fxdr_hyper(tl); 2708 len = fxdr_unsigned(int, *(tl + 2)); 2709 if (len <= 0 || len > NFS_MAXNAMLEN) { 2710 error = EBADRPC; 2711 m_freem(mrep); 2712 goto nfsmout; 2713 } 2714 /* for cookie stashing */ 2715 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2716 left = NFS_DIRFRAGSIZ - blksiz; 2717 if (reclen > left) { 2718 /* 2719 * DIRFRAGSIZ is aligned, no need to align 2720 * again here. 2721 */ 2722 memset(uiop->uio_iov->iov_base, 0, left); 2723 dp->d_reclen += left; 2724 UIO_ADVANCE(uiop, left); 2725 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2726 blksiz = 0; 2727 } 2728 if (reclen > uiop->uio_resid) 2729 bigenough = 0; 2730 if (bigenough) { 2731 int tlen; 2732 2733 dp = (struct dirent *)uiop->uio_iov->iov_base; 2734 dp->d_fileno = fileno; 2735 dp->d_namlen = len; 2736 dp->d_reclen = reclen; 2737 dp->d_type = DT_UNKNOWN; 2738 blksiz += reclen; 2739 if (blksiz == NFS_DIRFRAGSIZ) 2740 blksiz = 0; 2741 UIO_ADVANCE(uiop, DIRHDSIZ); 2742 nfsm_mtouio(uiop, len); 2743 tlen = reclen - (DIRHDSIZ + len); 2744 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2745 UIO_ADVANCE(uiop, tlen); 2746 cnp->cn_nameptr = dp->d_name; 2747 cnp->cn_namelen = dp->d_namlen; 2748 } else 2749 nfsm_adv(nfsm_rndup(len)); 2750 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2751 if (bigenough) { 2752 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2753 uiop->uio_offset = 2754 fxdr_swapcookie3(tl); 2755 else 2756 uiop->uio_offset = 2757 fxdr_cookie3(tl); 2758 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2759 } 2760 tl += 2; 2761 2762 /* 2763 * Since the attributes are before the file handle 2764 * (sigh), we must skip over the attributes and then 2765 * come back and get them. 2766 */ 2767 attrflag = fxdr_unsigned(int, *tl); 2768 if (attrflag) { 2769 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2770 memcpy(&fattr, fp, NFSX_V3FATTR); 2771 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2772 doit = fxdr_unsigned(int, *tl); 2773 if (doit) { 2774 nfsm_getfh(fhp, fhsize, 1); 2775 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2776 vref(vp); 2777 newvp = vp; 2778 np = dnp; 2779 } else { 2780 error = nfs_nget1(vp->v_mount, fhp, 2781 fhsize, &np, LK_NOWAIT); 2782 if (!error) 2783 newvp = NFSTOV(np); 2784 } 2785 if (!error) { 2786 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2787 if (bigenough) { 2788 dp->d_type = 2789 IFTODT(VTTOIF(np->n_vattr->va_type)); 2790 ndp->ni_vp = newvp; 2791 nfs_cache_enter(ndp->ni_dvp, 2792 ndp->ni_vp, cnp); 2793 } 2794 } 2795 error = 0; 2796 } 2797 } else { 2798 /* Just skip over the file handle */ 2799 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2800 i = fxdr_unsigned(int, *tl); 2801 nfsm_adv(nfsm_rndup(i)); 2802 } 2803 if (newvp != NULLVP) { 2804 if (newvp == vp) 2805 vrele(newvp); 2806 else 2807 vput(newvp); 2808 newvp = NULLVP; 2809 } 2810 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2811 more_dirs = fxdr_unsigned(int, *tl); 2812 } 2813 /* 2814 * If at end of rpc data, get the eof boolean 2815 */ 2816 if (!more_dirs) { 2817 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2818 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2819 2820 /* 2821 * kludge: see a comment in nfs_readdirrpc. 2822 */ 2823 2824 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2825 more_dirs = 0; 2826 } 2827 m_freem(mrep); 2828 } 2829 /* 2830 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2831 * by increasing d_reclen for the last record. 2832 */ 2833 if (blksiz > 0) { 2834 left = NFS_DIRFRAGSIZ - blksiz; 2835 memset(uiop->uio_iov->iov_base, 0, left); 2836 dp->d_reclen += left; 2837 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2838 UIO_ADVANCE(uiop, left); 2839 } 2840 2841 /* 2842 * We are now either at the end of the directory or have filled the 2843 * block. 2844 */ 2845 if (bigenough) { 2846 dnp->n_direofoffset = uiop->uio_offset; 2847 dnp->n_flag |= NEOFVALID; 2848 } 2849 nfsmout: 2850 if (newvp != NULLVP) { 2851 if(newvp == vp) 2852 vrele(newvp); 2853 else 2854 vput(newvp); 2855 } 2856 return (error); 2857 } 2858 #endif 2859 2860 /* 2861 * Silly rename. To make the NFS filesystem that is stateless look a little 2862 * more like the "ufs" a remove of an active vnode is translated to a rename 2863 * to a funny looking filename that is removed by nfs_inactive on the 2864 * nfsnode. There is the potential for another process on a different client 2865 * to create the same funny name between the nfs_lookitup() fails and the 2866 * nfs_rename() completes, but... 2867 */ 2868 int 2869 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2870 { 2871 struct sillyrename *sp; 2872 struct nfsnode *np; 2873 int error; 2874 pid_t pid; 2875 2876 cache_purge(dvp); 2877 np = VTONFS(vp); 2878 #ifndef DIAGNOSTIC 2879 if (vp->v_type == VDIR) 2880 panic("nfs: sillyrename dir"); 2881 #endif 2882 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2883 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2884 sp->s_dvp = dvp; 2885 vref(dvp); 2886 2887 /* Fudge together a funny name */ 2888 pid = curlwp->l_proc->p_pid; 2889 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2890 sp->s_namlen = 12; 2891 sp->s_name[8] = hexdigits[pid & 0xf]; 2892 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2893 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2894 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2895 2896 /* Try lookitups until we get one that isn't there */ 2897 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2898 curlwp, (struct nfsnode **)0) == 0) { 2899 sp->s_name[4]++; 2900 if (sp->s_name[4] > 'z') { 2901 error = EINVAL; 2902 goto bad; 2903 } 2904 } 2905 if (dolink) { 2906 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2907 sp->s_cred, curlwp); 2908 /* 2909 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2910 */ 2911 if (error == ENOTSUP) { 2912 error = nfs_renameit(dvp, cnp, sp); 2913 } 2914 } else { 2915 error = nfs_renameit(dvp, cnp, sp); 2916 } 2917 if (error) 2918 goto bad; 2919 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2920 curlwp, &np); 2921 np->n_sillyrename = sp; 2922 return (0); 2923 bad: 2924 vrele(sp->s_dvp); 2925 kauth_cred_free(sp->s_cred); 2926 kmem_free(sp, sizeof(*sp)); 2927 return (error); 2928 } 2929 2930 /* 2931 * Look up a file name and optionally either update the file handle or 2932 * allocate an nfsnode, depending on the value of npp. 2933 * npp == NULL --> just do the lookup 2934 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2935 * handled too 2936 * *npp != NULL --> update the file handle in the vnode 2937 */ 2938 int 2939 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2940 { 2941 u_int32_t *tl; 2942 char *cp; 2943 int32_t t1, t2; 2944 struct vnode *newvp = (struct vnode *)0; 2945 struct nfsnode *np, *dnp = VTONFS(dvp); 2946 char *bpos, *dpos, *cp2; 2947 int error = 0, ofhlen, fhlen; 2948 #ifndef NFS_V2_ONLY 2949 int attrflag; 2950 #endif 2951 struct mbuf *mreq, *mrep, *md, *mb; 2952 nfsfh_t *ofhp, *nfhp; 2953 const int v3 = NFS_ISV3(dvp); 2954 2955 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2956 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2957 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2958 nfsm_fhtom(dnp, v3); 2959 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2960 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2961 if (npp && !error) { 2962 nfsm_getfh(nfhp, fhlen, v3); 2963 if (*npp) { 2964 np = *npp; 2965 newvp = NFSTOV(np); 2966 ofhlen = np->n_fhsize; 2967 ofhp = kmem_alloc(ofhlen, KM_SLEEP); 2968 memcpy(ofhp, np->n_fhp, ofhlen); 2969 error = vcache_rekey_enter(newvp->v_mount, newvp, 2970 ofhp, ofhlen, nfhp, fhlen); 2971 if (error) { 2972 kmem_free(ofhp, ofhlen); 2973 m_freem(mrep); 2974 return error; 2975 } 2976 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2977 kmem_free(np->n_fhp, np->n_fhsize); 2978 np->n_fhp = &np->n_fh; 2979 } 2980 #if NFS_SMALLFH < NFSX_V3FHMAX 2981 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2982 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2983 #endif 2984 memcpy(np->n_fhp, nfhp, fhlen); 2985 np->n_fhsize = fhlen; 2986 vcache_rekey_exit(newvp->v_mount, newvp, 2987 ofhp, ofhlen, np->n_fhp, fhlen); 2988 kmem_free(ofhp, ofhlen); 2989 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2990 vref(dvp); 2991 newvp = dvp; 2992 np = dnp; 2993 } else { 2994 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2995 if (error) { 2996 m_freem(mrep); 2997 return (error); 2998 } 2999 newvp = NFSTOV(np); 3000 } 3001 #ifndef NFS_V2_ONLY 3002 if (v3) { 3003 nfsm_postop_attr(newvp, attrflag, 0); 3004 if (!attrflag && *npp == NULL) { 3005 m_freem(mrep); 3006 vput(newvp); 3007 return (ENOENT); 3008 } 3009 } else 3010 #endif 3011 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3012 } 3013 nfsm_reqdone; 3014 if (npp && *npp == NULL) { 3015 if (error) { 3016 if (newvp) 3017 vput(newvp); 3018 } else 3019 *npp = np; 3020 } 3021 return (error); 3022 } 3023 3024 #ifndef NFS_V2_ONLY 3025 /* 3026 * Nfs Version 3 commit rpc 3027 */ 3028 int 3029 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3030 { 3031 char *cp; 3032 u_int32_t *tl; 3033 int32_t t1, t2; 3034 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3035 char *bpos, *dpos, *cp2; 3036 int error = 0, wccflag = NFSV3_WCCRATTR; 3037 struct mbuf *mreq, *mrep, *md, *mb; 3038 struct nfsnode *np; 3039 3040 KASSERT(NFS_ISV3(vp)); 3041 3042 #ifdef NFS_DEBUG_COMMIT 3043 printf("commit %lu - %lu\n", (unsigned long)offset, 3044 (unsigned long)(offset + cnt)); 3045 #endif 3046 3047 mutex_enter(&nmp->nm_lock); 3048 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3049 mutex_exit(&nmp->nm_lock); 3050 return (0); 3051 } 3052 mutex_exit(&nmp->nm_lock); 3053 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3054 np = VTONFS(vp); 3055 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3056 nfsm_fhtom(np, 1); 3057 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3058 txdr_hyper(offset, tl); 3059 tl += 2; 3060 *tl = txdr_unsigned(cnt); 3061 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3062 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3063 if (!error) { 3064 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3065 mutex_enter(&nmp->nm_lock); 3066 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3067 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3068 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3069 error = NFSERR_STALEWRITEVERF; 3070 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3071 } 3072 mutex_exit(&nmp->nm_lock); 3073 } 3074 nfsm_reqdone; 3075 return (error); 3076 } 3077 #endif 3078 3079 /* 3080 * Kludge City.. 3081 * - make nfs_bmap() essentially a no-op that does no translation 3082 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3083 * (Maybe I could use the process's page mapping, but I was concerned that 3084 * Kernel Write might not be enabled and also figured copyout() would do 3085 * a lot more work than memcpy() and also it currently happens in the 3086 * context of the swapper process (2). 3087 */ 3088 int 3089 nfs_bmap(void *v) 3090 { 3091 struct vop_bmap_args /* { 3092 struct vnode *a_vp; 3093 daddr_t a_bn; 3094 struct vnode **a_vpp; 3095 daddr_t *a_bnp; 3096 int *a_runp; 3097 } */ *ap = v; 3098 struct vnode *vp = ap->a_vp; 3099 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3100 3101 if (ap->a_vpp != NULL) 3102 *ap->a_vpp = vp; 3103 if (ap->a_bnp != NULL) 3104 *ap->a_bnp = ap->a_bn << bshift; 3105 if (ap->a_runp != NULL) 3106 *ap->a_runp = 1024 * 1024; /* XXX */ 3107 return (0); 3108 } 3109 3110 /* 3111 * Strategy routine. 3112 * For async requests when nfsiod(s) are running, queue the request by 3113 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3114 * request. 3115 */ 3116 int 3117 nfs_strategy(void *v) 3118 { 3119 struct vop_strategy_args *ap = v; 3120 struct buf *bp = ap->a_bp; 3121 int error = 0; 3122 3123 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3124 panic("nfs physio/async"); 3125 3126 /* 3127 * If the op is asynchronous and an i/o daemon is waiting 3128 * queue the request, wake it up and wait for completion 3129 * otherwise just do it ourselves. 3130 */ 3131 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3132 error = nfs_doio(bp); 3133 return (error); 3134 } 3135 3136 /* 3137 * fsync vnode op. Just call nfs_flush() with commit == 1. 3138 */ 3139 /* ARGSUSED */ 3140 int 3141 nfs_fsync(void *v) 3142 { 3143 struct vop_fsync_args /* { 3144 struct vnodeop_desc *a_desc; 3145 struct vnode * a_vp; 3146 kauth_cred_t a_cred; 3147 int a_flags; 3148 off_t offlo; 3149 off_t offhi; 3150 struct lwp * a_l; 3151 } */ *ap = v; 3152 3153 struct vnode *vp = ap->a_vp; 3154 3155 if (vp->v_type != VREG) 3156 return 0; 3157 3158 return (nfs_flush(vp, ap->a_cred, 3159 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3160 } 3161 3162 /* 3163 * Flush all the data associated with a vnode. 3164 */ 3165 int 3166 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3167 int commit) 3168 { 3169 struct nfsnode *np = VTONFS(vp); 3170 int error; 3171 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3172 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3173 3174 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 3175 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3176 if (np->n_flag & NWRITEERR) { 3177 error = np->n_error; 3178 np->n_flag &= ~NWRITEERR; 3179 } 3180 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3181 return (error); 3182 } 3183 3184 /* 3185 * Return POSIX pathconf information applicable to nfs. 3186 * 3187 * N.B. The NFS V2 protocol doesn't support this RPC. 3188 */ 3189 /* ARGSUSED */ 3190 int 3191 nfs_pathconf(void *v) 3192 { 3193 struct vop_pathconf_args /* { 3194 struct vnode *a_vp; 3195 int a_name; 3196 register_t *a_retval; 3197 } */ *ap = v; 3198 struct nfsv3_pathconf *pcp; 3199 struct vnode *vp = ap->a_vp; 3200 struct mbuf *mreq, *mrep, *md, *mb; 3201 int32_t t1, t2; 3202 u_int32_t *tl; 3203 char *bpos, *dpos, *cp, *cp2; 3204 int error = 0, attrflag; 3205 #ifndef NFS_V2_ONLY 3206 struct nfsmount *nmp; 3207 unsigned int l; 3208 u_int64_t maxsize; 3209 #endif 3210 const int v3 = NFS_ISV3(vp); 3211 struct nfsnode *np = VTONFS(vp); 3212 3213 switch (ap->a_name) { 3214 /* Names that can be resolved locally. */ 3215 case _PC_PIPE_BUF: 3216 *ap->a_retval = PIPE_BUF; 3217 break; 3218 case _PC_SYNC_IO: 3219 *ap->a_retval = 1; 3220 break; 3221 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3222 case _PC_LINK_MAX: 3223 case _PC_NAME_MAX: 3224 case _PC_CHOWN_RESTRICTED: 3225 case _PC_NO_TRUNC: 3226 if (!v3) { 3227 error = EINVAL; 3228 break; 3229 } 3230 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3231 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3232 nfsm_fhtom(np, 1); 3233 nfsm_request(np, NFSPROC_PATHCONF, 3234 curlwp, curlwp->l_cred); /* XXX */ 3235 nfsm_postop_attr(vp, attrflag, 0); 3236 if (!error) { 3237 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3238 NFSX_V3PATHCONF); 3239 switch (ap->a_name) { 3240 case _PC_LINK_MAX: 3241 *ap->a_retval = 3242 fxdr_unsigned(register_t, pcp->pc_linkmax); 3243 break; 3244 case _PC_NAME_MAX: 3245 *ap->a_retval = 3246 fxdr_unsigned(register_t, pcp->pc_namemax); 3247 break; 3248 case _PC_CHOWN_RESTRICTED: 3249 *ap->a_retval = 3250 (pcp->pc_chownrestricted == nfs_true); 3251 break; 3252 case _PC_NO_TRUNC: 3253 *ap->a_retval = 3254 (pcp->pc_notrunc == nfs_true); 3255 break; 3256 } 3257 } 3258 nfsm_reqdone; 3259 break; 3260 case _PC_FILESIZEBITS: 3261 #ifndef NFS_V2_ONLY 3262 if (v3) { 3263 nmp = VFSTONFS(vp->v_mount); 3264 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3265 if ((error = nfs_fsinfo(nmp, vp, 3266 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3267 break; 3268 for (l = 0, maxsize = nmp->nm_maxfilesize; 3269 (maxsize >> l) > 0; l++) 3270 ; 3271 *ap->a_retval = l + 1; 3272 } else 3273 #endif 3274 { 3275 *ap->a_retval = 32; /* NFS V2 limitation */ 3276 } 3277 break; 3278 default: 3279 error = genfs_pathconf(ap); 3280 break; 3281 } 3282 3283 return (error); 3284 } 3285 3286 /* 3287 * NFS advisory byte-level locks. 3288 */ 3289 int 3290 nfs_advlock(void *v) 3291 { 3292 struct vop_advlock_args /* { 3293 struct vnode *a_vp; 3294 void *a_id; 3295 int a_op; 3296 struct flock *a_fl; 3297 int a_flags; 3298 } */ *ap = v; 3299 struct nfsnode *np = VTONFS(ap->a_vp); 3300 3301 return lf_advlock(ap, &np->n_lockf, np->n_size); 3302 } 3303 3304 /* 3305 * Print out the contents of an nfsnode. 3306 */ 3307 int 3308 nfs_print(void *v) 3309 { 3310 struct vop_print_args /* { 3311 struct vnode *a_vp; 3312 } */ *ap = v; 3313 struct vnode *vp = ap->a_vp; 3314 struct nfsnode *np = VTONFS(vp); 3315 3316 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3317 (unsigned long long)np->n_vattr->va_fileid, 3318 (unsigned long long)np->n_vattr->va_fsid); 3319 if (vp->v_type == VFIFO) 3320 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3321 printf("\n"); 3322 return (0); 3323 } 3324 3325 /* 3326 * nfs unlock wrapper. 3327 */ 3328 int 3329 nfs_unlock(void *v) 3330 { 3331 struct vop_unlock_args /* { 3332 struct vnode *a_vp; 3333 int a_flags; 3334 } */ *ap = v; 3335 struct vnode *vp = ap->a_vp; 3336 3337 /* 3338 * VOP_UNLOCK can be called by nfs_loadattrcache 3339 * with v_data == 0. 3340 */ 3341 if (VTONFS(vp)) { 3342 nfs_delayedtruncate(vp); 3343 } 3344 3345 return genfs_unlock(v); 3346 } 3347 3348 /* 3349 * nfs special file access vnode op. 3350 * Essentially just get vattr and then imitate iaccess() since the device is 3351 * local to the client. 3352 */ 3353 int 3354 nfsspec_access(void *v) 3355 { 3356 struct vop_access_args /* { 3357 struct vnode *a_vp; 3358 accmode_t a_accmode; 3359 kauth_cred_t a_cred; 3360 struct lwp *a_l; 3361 } */ *ap = v; 3362 struct vattr va; 3363 struct vnode *vp = ap->a_vp; 3364 int error; 3365 3366 error = VOP_GETATTR(vp, &va, ap->a_cred); 3367 if (error) 3368 return (error); 3369 3370 /* 3371 * Disallow write attempts on filesystems mounted read-only; 3372 * unless the file is a socket, fifo, or a block or character 3373 * device resident on the filesystem. 3374 */ 3375 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3376 switch (vp->v_type) { 3377 case VREG: 3378 case VDIR: 3379 case VLNK: 3380 return (EROFS); 3381 default: 3382 break; 3383 } 3384 } 3385 3386 return kauth_authorize_vnode(ap->a_cred, KAUTH_ACCESS_ACTION( 3387 ap->a_accmode, va.va_type, va.va_mode), vp, NULL, genfs_can_access( 3388 vp, ap->a_cred, va.va_uid, va.va_gid, va.va_mode, NULL, 3389 ap->a_accmode)); 3390 } 3391 3392 /* 3393 * Read wrapper for special devices. 3394 */ 3395 int 3396 nfsspec_read(void *v) 3397 { 3398 struct vop_read_args /* { 3399 struct vnode *a_vp; 3400 struct uio *a_uio; 3401 int a_ioflag; 3402 kauth_cred_t a_cred; 3403 } */ *ap = v; 3404 struct nfsnode *np = VTONFS(ap->a_vp); 3405 3406 /* 3407 * Set access flag. 3408 */ 3409 np->n_flag |= NACC; 3410 getnanotime(&np->n_atim); 3411 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3412 } 3413 3414 /* 3415 * Write wrapper for special devices. 3416 */ 3417 int 3418 nfsspec_write(void *v) 3419 { 3420 struct vop_write_args /* { 3421 struct vnode *a_vp; 3422 struct uio *a_uio; 3423 int a_ioflag; 3424 kauth_cred_t a_cred; 3425 } */ *ap = v; 3426 struct nfsnode *np = VTONFS(ap->a_vp); 3427 3428 /* 3429 * Set update flag. 3430 */ 3431 np->n_flag |= NUPD; 3432 getnanotime(&np->n_mtim); 3433 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3434 } 3435 3436 /* 3437 * Close wrapper for special devices. 3438 * 3439 * Update the times on the nfsnode then do device close. 3440 */ 3441 int 3442 nfsspec_close(void *v) 3443 { 3444 struct vop_close_args /* { 3445 struct vnode *a_vp; 3446 int a_fflag; 3447 kauth_cred_t a_cred; 3448 struct lwp *a_l; 3449 } */ *ap = v; 3450 struct vnode *vp = ap->a_vp; 3451 struct nfsnode *np = VTONFS(vp); 3452 struct vattr vattr; 3453 3454 if (np->n_flag & (NACC | NUPD)) { 3455 np->n_flag |= NCHG; 3456 if (vrefcnt(vp) == 1 && 3457 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3458 vattr_null(&vattr); 3459 if (np->n_flag & NACC) 3460 vattr.va_atime = np->n_atim; 3461 if (np->n_flag & NUPD) 3462 vattr.va_mtime = np->n_mtim; 3463 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3464 } 3465 } 3466 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3467 } 3468 3469 /* 3470 * Read wrapper for fifos. 3471 */ 3472 int 3473 nfsfifo_read(void *v) 3474 { 3475 struct vop_read_args /* { 3476 struct vnode *a_vp; 3477 struct uio *a_uio; 3478 int a_ioflag; 3479 kauth_cred_t a_cred; 3480 } */ *ap = v; 3481 struct nfsnode *np = VTONFS(ap->a_vp); 3482 3483 /* 3484 * Set access flag. 3485 */ 3486 np->n_flag |= NACC; 3487 getnanotime(&np->n_atim); 3488 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3489 } 3490 3491 /* 3492 * Write wrapper for fifos. 3493 */ 3494 int 3495 nfsfifo_write(void *v) 3496 { 3497 struct vop_write_args /* { 3498 struct vnode *a_vp; 3499 struct uio *a_uio; 3500 int a_ioflag; 3501 kauth_cred_t a_cred; 3502 } */ *ap = v; 3503 struct nfsnode *np = VTONFS(ap->a_vp); 3504 3505 /* 3506 * Set update flag. 3507 */ 3508 np->n_flag |= NUPD; 3509 getnanotime(&np->n_mtim); 3510 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3511 } 3512 3513 /* 3514 * Close wrapper for fifos. 3515 * 3516 * Update the times on the nfsnode then do fifo close. 3517 */ 3518 int 3519 nfsfifo_close(void *v) 3520 { 3521 struct vop_close_args /* { 3522 struct vnode *a_vp; 3523 int a_fflag; 3524 kauth_cred_t a_cred; 3525 struct lwp *a_l; 3526 } */ *ap = v; 3527 struct vnode *vp = ap->a_vp; 3528 struct nfsnode *np = VTONFS(vp); 3529 struct vattr vattr; 3530 3531 if (np->n_flag & (NACC | NUPD)) { 3532 struct timespec ts; 3533 3534 getnanotime(&ts); 3535 if (np->n_flag & NACC) 3536 np->n_atim = ts; 3537 if (np->n_flag & NUPD) 3538 np->n_mtim = ts; 3539 np->n_flag |= NCHG; 3540 if (vrefcnt(vp) == 1 && 3541 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3542 vattr_null(&vattr); 3543 if (np->n_flag & NACC) 3544 vattr.va_atime = np->n_atim; 3545 if (np->n_flag & NUPD) 3546 vattr.va_mtime = np->n_mtim; 3547 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3548 } 3549 } 3550 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3551 } 3552
Indexes created Wed Sep 24 05:09:52 GMT 2025