nfs_vnops.c revision 1.320
1 /* $NetBSD: nfs_vnops.c,v 1.320 2021/07/18 23:57:15 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 1.320 2021/07/18 23:57:15 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, genfs_enoioctl }, /* ioctl */ 119 { &vop_poll_desc, genfs_poll }, /* poll */ 120 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 121 { &vop_revoke_desc, genfs_revoke }, /* revoke */ 122 { &vop_mmap_desc, genfs_mmap }, /* mmap */ 123 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 124 { &vop_seek_desc, genfs_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, genfs_abortop }, /* abortop */ 134 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 135 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 136 { &vop_lock_desc, genfs_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, genfs_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 GENFS_SPECOP_ENTRIES, 159 { &vop_close_desc, nfsspec_close }, /* close */ 160 { &vop_access_desc, nfsspec_access }, /* access */ 161 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 162 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 163 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 164 { &vop_read_desc, nfsspec_read }, /* read */ 165 { &vop_write_desc, nfsspec_write }, /* write */ 166 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 167 { &vop_fsync_desc, spec_fsync }, /* fsync */ 168 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 169 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 170 { &vop_lock_desc, genfs_lock }, /* lock */ 171 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 172 { &vop_print_desc, nfs_print }, /* print */ 173 { &vop_islocked_desc, genfs_islocked }, /* islocked */ 174 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 175 { NULL, NULL } 176 }; 177 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 178 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 179 180 int (**fifo_nfsv2nodeop_p)(void *); 181 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 182 { &vop_default_desc, vn_default_error }, 183 GENFS_FIFOOP_ENTRIES, 184 { &vop_close_desc, nfsfifo_close }, /* close */ 185 { &vop_access_desc, nfsspec_access }, /* access */ 186 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 187 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 188 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 189 { &vop_read_desc, nfsfifo_read }, /* read */ 190 { &vop_write_desc, nfsfifo_write }, /* write */ 191 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 192 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 193 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 194 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 195 { &vop_lock_desc, genfs_lock }, /* lock */ 196 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 197 { &vop_strategy_desc, vn_fifo_bypass }, /* strategy */ 198 { &vop_print_desc, nfs_print }, /* print */ 199 { &vop_islocked_desc, genfs_islocked }, /* islocked */ 200 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 201 { NULL, NULL } 202 }; 203 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 204 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 205 206 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 207 size_t, kauth_cred_t, struct lwp *); 208 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 209 210 /* 211 * Global variables 212 */ 213 extern u_int32_t nfs_true, nfs_false; 214 extern u_int32_t nfs_xdrneg1; 215 extern const nfstype nfsv3_type[9]; 216 217 int nfs_numasync = 0; 218 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 219 #define UIO_ADVANCE(uio, siz) \ 220 (void)((uio)->uio_resid -= (siz), \ 221 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 222 (uio)->uio_iov->iov_len -= (siz)) 223 224 static void nfs_cache_enter(struct vnode *, struct vnode *, 225 struct componentname *); 226 227 static void 228 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 229 struct componentname *cnp) 230 { 231 struct nfsnode *dnp = VTONFS(dvp); 232 233 if ((cnp->cn_flags & MAKEENTRY) == 0) { 234 return; 235 } 236 if (vp != NULL) { 237 struct nfsnode *np = VTONFS(vp); 238 239 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 240 } 241 242 if (!timespecisset(&dnp->n_nctime)) 243 dnp->n_nctime = dnp->n_vattr->va_mtime; 244 245 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags); 246 } 247 248 /* 249 * nfs null call from vfs. 250 */ 251 int 252 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 253 { 254 char *bpos, *dpos; 255 int error = 0; 256 struct mbuf *mreq, *mrep, *md, *mb __unused; 257 struct nfsnode *np = VTONFS(vp); 258 259 nfsm_reqhead(np, NFSPROC_NULL, 0); 260 nfsm_request(np, NFSPROC_NULL, l, cred); 261 nfsm_reqdone; 262 return (error); 263 } 264 265 /* 266 * nfs access vnode op. 267 * For nfs version 2, just return ok. File accesses may fail later. 268 * For nfs version 3, use the access rpc to check accessibility. If file modes 269 * are changed on the server, accesses might still fail later. 270 */ 271 int 272 nfs_access(void *v) 273 { 274 struct vop_access_args /* { 275 struct vnode *a_vp; 276 accmode_t a_accmode; 277 kauth_cred_t a_cred; 278 } */ *ap = v; 279 struct vnode *vp = ap->a_vp; 280 #ifndef NFS_V2_ONLY 281 u_int32_t *tl; 282 char *cp; 283 int32_t t1, t2; 284 char *bpos, *dpos, *cp2; 285 int error = 0, attrflag; 286 struct mbuf *mreq, *mrep, *md, *mb; 287 u_int32_t mode, rmode; 288 const int v3 = NFS_ISV3(vp); 289 #endif 290 int cachevalid; 291 struct nfsnode *np = VTONFS(vp); 292 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 293 294 cachevalid = (np->n_accstamp != -1 && 295 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 296 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 297 298 /* 299 * Check access cache first. If this request has been made for this 300 * uid shortly before, use the cached result. 301 */ 302 if (cachevalid) { 303 if (!np->n_accerror) { 304 if ((np->n_accmode & ap->a_accmode) == ap->a_accmode) 305 return np->n_accerror; 306 } else if ((np->n_accmode & ap->a_accmode) == np->n_accmode) 307 return np->n_accerror; 308 } 309 310 #ifndef NFS_V2_ONLY 311 /* 312 * For nfs v3, do an access rpc, otherwise you are stuck emulating 313 * ufs_access() locally using the vattr. This may not be correct, 314 * since the server may apply other access criteria such as 315 * client uid-->server uid mapping that we do not know about, but 316 * this is better than just returning anything that is lying about 317 * in the cache. 318 */ 319 if (v3) { 320 nfsstats.rpccnt[NFSPROC_ACCESS]++; 321 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 322 nfsm_fhtom(np, v3); 323 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 324 if (ap->a_accmode & VREAD) 325 mode = NFSV3ACCESS_READ; 326 else 327 mode = 0; 328 if (vp->v_type != VDIR) { 329 if (ap->a_accmode & VWRITE) 330 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 331 if (ap->a_accmode & VEXEC) 332 mode |= NFSV3ACCESS_EXECUTE; 333 } else { 334 if (ap->a_accmode & VWRITE) 335 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 336 NFSV3ACCESS_DELETE); 337 if (ap->a_accmode & VEXEC) 338 mode |= NFSV3ACCESS_LOOKUP; 339 } 340 *tl = txdr_unsigned(mode); 341 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 342 nfsm_postop_attr(vp, attrflag, 0); 343 if (!error) { 344 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 345 rmode = fxdr_unsigned(u_int32_t, *tl); 346 /* 347 * The NFS V3 spec does not clarify whether or not 348 * the returned access bits can be a superset of 349 * the ones requested, so... 350 */ 351 if ((rmode & mode) != mode) 352 error = EACCES; 353 } 354 nfsm_reqdone; 355 } else 356 #endif 357 return (nfsspec_access(ap)); 358 #ifndef NFS_V2_ONLY 359 /* 360 * Disallow write attempts on filesystems mounted read-only; 361 * unless the file is a socket, fifo, or a block or character 362 * device resident on the filesystem. 363 */ 364 if (!error && (ap->a_accmode & VWRITE) && 365 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 366 switch (vp->v_type) { 367 case VREG: 368 case VDIR: 369 case VLNK: 370 error = EROFS; 371 default: 372 break; 373 } 374 } 375 376 if (!error || error == EACCES) { 377 /* 378 * If we got the same result as for a previous, 379 * different request, OR it in. Don't update 380 * the timestamp in that case. 381 */ 382 if (cachevalid && np->n_accstamp != -1 && 383 error == np->n_accerror) { 384 if (!error) 385 np->n_accmode |= ap->a_accmode; 386 else if ((np->n_accmode & ap->a_accmode) == ap->a_accmode) 387 np->n_accmode = ap->a_accmode; 388 } else { 389 np->n_accstamp = time_uptime; 390 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 391 np->n_accmode = ap->a_accmode; 392 np->n_accerror = error; 393 } 394 } 395 396 return (error); 397 #endif 398 } 399 400 /* 401 * nfs open vnode op 402 * Check to see if the type is ok 403 * and that deletion is not in progress. 404 * For paged in text files, you will need to flush the page cache 405 * if consistency is lost. 406 */ 407 /* ARGSUSED */ 408 int 409 nfs_open(void *v) 410 { 411 struct vop_open_args /* { 412 struct vnode *a_vp; 413 int a_mode; 414 kauth_cred_t a_cred; 415 } */ *ap = v; 416 struct vnode *vp = ap->a_vp; 417 struct nfsnode *np = VTONFS(vp); 418 int error; 419 420 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 421 return (EACCES); 422 } 423 424 if (ap->a_mode & FREAD) { 425 if (np->n_rcred != NULL) 426 kauth_cred_free(np->n_rcred); 427 np->n_rcred = ap->a_cred; 428 kauth_cred_hold(np->n_rcred); 429 } 430 if (ap->a_mode & FWRITE) { 431 if (np->n_wcred != NULL) 432 kauth_cred_free(np->n_wcred); 433 np->n_wcred = ap->a_cred; 434 kauth_cred_hold(np->n_wcred); 435 } 436 437 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 438 if (error) 439 return error; 440 441 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 442 443 return (0); 444 } 445 446 /* 447 * nfs close vnode op 448 * What an NFS client should do upon close after writing is a debatable issue. 449 * Most NFS clients push delayed writes to the server upon close, basically for 450 * two reasons: 451 * 1 - So that any write errors may be reported back to the client process 452 * doing the close system call. By far the two most likely errors are 453 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 454 * 2 - To put a worst case upper bound on cache inconsistency between 455 * multiple clients for the file. 456 * There is also a consistency problem for Version 2 of the protocol w.r.t. 457 * not being able to tell if other clients are writing a file concurrently, 458 * since there is no way of knowing if the changed modify time in the reply 459 * is only due to the write for this client. 460 * (NFS Version 3 provides weak cache consistency data in the reply that 461 * should be sufficient to detect and handle this case.) 462 * 463 * The current code does the following: 464 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 465 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 466 * or commit them (this satisfies 1 and 2 except for the 467 * case where the server crashes after this close but 468 * before the commit RPC, which is felt to be "good 469 * enough". Changing the last argument to nfs_flush() to 470 * a 1 would force a commit operation, if it is felt a 471 * commit is necessary now. 472 */ 473 /* ARGSUSED */ 474 int 475 nfs_close(void *v) 476 { 477 struct vop_close_args /* { 478 struct vnodeop_desc *a_desc; 479 struct vnode *a_vp; 480 int a_fflag; 481 kauth_cred_t a_cred; 482 } */ *ap = v; 483 struct vnode *vp = ap->a_vp; 484 struct nfsnode *np = VTONFS(vp); 485 int error = 0; 486 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 487 488 if (vp->v_type == VREG) { 489 if (np->n_flag & NMODIFIED) { 490 #ifndef NFS_V2_ONLY 491 if (NFS_ISV3(vp)) { 492 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 493 np->n_flag &= ~NMODIFIED; 494 } else 495 #endif 496 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 497 NFS_INVALIDATE_ATTRCACHE(np); 498 } 499 if (np->n_flag & NWRITEERR) { 500 np->n_flag &= ~NWRITEERR; 501 error = np->n_error; 502 } 503 } 504 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 505 return (error); 506 } 507 508 /* 509 * nfs getattr call from vfs. 510 */ 511 int 512 nfs_getattr(void *v) 513 { 514 struct vop_getattr_args /* { 515 struct vnode *a_vp; 516 struct vattr *a_vap; 517 kauth_cred_t a_cred; 518 } */ *ap = v; 519 struct vnode *vp = ap->a_vp; 520 struct nfsnode *np = VTONFS(vp); 521 char *cp; 522 u_int32_t *tl; 523 int32_t t1, t2; 524 char *bpos, *dpos; 525 int error = 0; 526 struct mbuf *mreq, *mrep, *md, *mb; 527 const int v3 = NFS_ISV3(vp); 528 529 /* 530 * Update local times for special files. 531 */ 532 if (np->n_flag & (NACC | NUPD)) 533 np->n_flag |= NCHG; 534 535 /* 536 * if we have delayed truncation, do it now. 537 */ 538 nfs_delayedtruncate(vp); 539 540 /* 541 * First look in the cache. 542 */ 543 if (nfs_getattrcache(vp, ap->a_vap) == 0) 544 return (0); 545 nfsstats.rpccnt[NFSPROC_GETATTR]++; 546 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 547 nfsm_fhtom(np, v3); 548 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 549 if (!error) { 550 nfsm_loadattr(vp, ap->a_vap, 0); 551 if (vp->v_type == VDIR && 552 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 553 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 554 } 555 nfsm_reqdone; 556 return (error); 557 } 558 559 /* 560 * nfs setattr call. 561 */ 562 int 563 nfs_setattr(void *v) 564 { 565 struct vop_setattr_args /* { 566 struct vnodeop_desc *a_desc; 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 struct vattr *vap = ap->a_vap; 574 int error = 0; 575 u_quad_t tsize = 0; 576 577 /* 578 * Setting of flags is not supported. 579 */ 580 if (vap->va_flags != VNOVAL) 581 return (EOPNOTSUPP); 582 583 /* 584 * Disallow write attempts if the filesystem is mounted read-only. 585 */ 586 if ((vap->va_uid != (uid_t)VNOVAL || 587 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 588 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 589 (vp->v_mount->mnt_flag & MNT_RDONLY)) 590 return (EROFS); 591 if (vap->va_size != VNOVAL) { 592 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 593 return EFBIG; 594 } 595 switch (vp->v_type) { 596 case VDIR: 597 return (EISDIR); 598 case VCHR: 599 case VBLK: 600 case VSOCK: 601 case VFIFO: 602 if (vap->va_mtime.tv_sec == VNOVAL && 603 vap->va_atime.tv_sec == VNOVAL && 604 vap->va_mode == (mode_t)VNOVAL && 605 vap->va_uid == (uid_t)VNOVAL && 606 vap->va_gid == (gid_t)VNOVAL) 607 return (0); 608 vap->va_size = VNOVAL; 609 break; 610 default: 611 /* 612 * Disallow write attempts if the filesystem is 613 * mounted read-only. 614 */ 615 if (vp->v_mount->mnt_flag & MNT_RDONLY) 616 return (EROFS); 617 genfs_node_wrlock(vp); 618 uvm_vnp_setsize(vp, vap->va_size); 619 tsize = np->n_size; 620 np->n_size = vap->va_size; 621 if (vap->va_size == 0) 622 error = nfs_vinvalbuf(vp, 0, 623 ap->a_cred, curlwp, 1); 624 else 625 error = nfs_vinvalbuf(vp, V_SAVE, 626 ap->a_cred, curlwp, 1); 627 if (error) { 628 uvm_vnp_setsize(vp, tsize); 629 genfs_node_unlock(vp); 630 return (error); 631 } 632 np->n_vattr->va_size = vap->va_size; 633 } 634 } else { 635 /* 636 * flush files before setattr because a later write of 637 * cached data might change timestamps or reset sugid bits 638 */ 639 if ((vap->va_mtime.tv_sec != VNOVAL || 640 vap->va_atime.tv_sec != VNOVAL || 641 vap->va_mode != VNOVAL) && 642 vp->v_type == VREG && 643 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 644 curlwp, 1)) == EINTR) 645 return (error); 646 } 647 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 648 if (vap->va_size != VNOVAL) { 649 if (error) { 650 np->n_size = np->n_vattr->va_size = tsize; 651 uvm_vnp_setsize(vp, np->n_size); 652 } 653 genfs_node_unlock(vp); 654 } 655 VN_KNOTE(vp, NOTE_ATTRIB); 656 return (error); 657 } 658 659 /* 660 * Do an nfs setattr rpc. 661 */ 662 int 663 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 664 { 665 struct nfsv2_sattr *sp; 666 char *cp; 667 int32_t t1, t2; 668 char *bpos, *dpos; 669 u_int32_t *tl; 670 int error = 0; 671 struct mbuf *mreq, *mrep, *md, *mb; 672 const int v3 = NFS_ISV3(vp); 673 struct nfsnode *np = VTONFS(vp); 674 #ifndef NFS_V2_ONLY 675 int wccflag = NFSV3_WCCRATTR; 676 char *cp2; 677 #endif 678 679 nfsstats.rpccnt[NFSPROC_SETATTR]++; 680 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 681 nfsm_fhtom(np, v3); 682 #ifndef NFS_V2_ONLY 683 if (v3) { 684 nfsm_v3attrbuild(vap, true); 685 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 686 *tl = nfs_false; 687 } else { 688 #endif 689 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 690 if (vap->va_mode == (mode_t)VNOVAL) 691 sp->sa_mode = nfs_xdrneg1; 692 else 693 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 694 if (vap->va_uid == (uid_t)VNOVAL) 695 sp->sa_uid = nfs_xdrneg1; 696 else 697 sp->sa_uid = txdr_unsigned(vap->va_uid); 698 if (vap->va_gid == (gid_t)VNOVAL) 699 sp->sa_gid = nfs_xdrneg1; 700 else 701 sp->sa_gid = txdr_unsigned(vap->va_gid); 702 sp->sa_size = txdr_unsigned(vap->va_size); 703 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 704 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 705 #ifndef NFS_V2_ONLY 706 } 707 #endif 708 nfsm_request(np, NFSPROC_SETATTR, l, cred); 709 #ifndef NFS_V2_ONLY 710 if (v3) { 711 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 712 } else 713 #endif 714 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 715 nfsm_reqdone; 716 return (error); 717 } 718 719 /* 720 * nfs lookup call, one step at a time... 721 * First look in cache 722 * If not found, do the rpc. 723 */ 724 int 725 nfs_lookup(void *v) 726 { 727 struct vop_lookup_v2_args /* { 728 struct vnodeop_desc *a_desc; 729 struct vnode *a_dvp; 730 struct vnode **a_vpp; 731 struct componentname *a_cnp; 732 } */ *ap = v; 733 struct componentname *cnp = ap->a_cnp; 734 struct vnode *dvp = ap->a_dvp; 735 struct vnode **vpp = ap->a_vpp; 736 int flags; 737 struct vnode *newvp; 738 u_int32_t *tl; 739 char *cp; 740 int32_t t1, t2; 741 char *bpos, *dpos, *cp2; 742 struct mbuf *mreq, *mrep, *md, *mb; 743 long len; 744 nfsfh_t *fhp; 745 struct nfsnode *np; 746 int cachefound; 747 int error = 0, attrflag, fhsize; 748 const int v3 = NFS_ISV3(dvp); 749 750 flags = cnp->cn_flags; 751 752 *vpp = NULLVP; 753 newvp = NULLVP; 754 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 755 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 756 return (EROFS); 757 if (dvp->v_type != VDIR) 758 return (ENOTDIR); 759 760 /* 761 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 762 */ 763 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 764 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 765 if (error) 766 return error; 767 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 768 return EISDIR; 769 vref(dvp); 770 *vpp = dvp; 771 return 0; 772 } 773 774 np = VTONFS(dvp); 775 776 /* 777 * Before performing an RPC, check the name cache to see if 778 * the directory/name pair we are looking for is known already. 779 * If the directory/name pair is found in the name cache, 780 * we have to ensure the directory has not changed from 781 * the time the cache entry has been created. If it has, 782 * the cache entry has to be ignored. 783 */ 784 cachefound = cache_lookup_raw(dvp, cnp->cn_nameptr, cnp->cn_namelen, 785 cnp->cn_flags, NULL, vpp); 786 KASSERT(dvp != *vpp); 787 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 788 if (cachefound) { 789 struct vattr vattr; 790 791 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 792 if (error != 0) { 793 if (*vpp != NULLVP) 794 vrele(*vpp); 795 *vpp = NULLVP; 796 return error; 797 } 798 799 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 800 || timespeccmp(&vattr.va_mtime, 801 &VTONFS(dvp)->n_nctime, !=)) { 802 if (*vpp != NULLVP) { 803 vrele(*vpp); 804 *vpp = NULLVP; 805 } 806 cache_purge1(dvp, NULL, 0, PURGE_CHILDREN); 807 timespecclear(&np->n_nctime); 808 goto dorpc; 809 } 810 811 if (*vpp == NULLVP) { 812 /* namecache gave us a negative result */ 813 error = ENOENT; 814 goto noentry; 815 } 816 817 /* 818 * investigate the vnode returned by cache_lookup_raw. 819 * if it isn't appropriate, do an rpc. 820 */ 821 newvp = *vpp; 822 if ((flags & ISDOTDOT) != 0) { 823 VOP_UNLOCK(dvp); 824 } 825 error = vn_lock(newvp, LK_SHARED); 826 if ((flags & ISDOTDOT) != 0) { 827 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 828 } 829 if (error != 0) { 830 /* newvp has been reclaimed. */ 831 vrele(newvp); 832 *vpp = NULLVP; 833 goto dorpc; 834 } 835 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 836 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 837 nfsstats.lookupcache_hits++; 838 KASSERT(newvp->v_type != VNON); 839 VOP_UNLOCK(newvp); 840 return (0); 841 } 842 cache_purge1(newvp, NULL, 0, PURGE_PARENTS); 843 vput(newvp); 844 *vpp = NULLVP; 845 } 846 dorpc: 847 #if 0 848 /* 849 * because nfsv3 has the same CREATE semantics as ours, 850 * we don't have to perform LOOKUPs beforehand. 851 * 852 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 853 * XXX although we have no way to know if O_EXCL is requested or not. 854 */ 855 856 if (v3 && cnp->cn_nameiop == CREATE && 857 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 858 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 859 return (EJUSTRETURN); 860 } 861 #endif /* 0 */ 862 863 error = 0; 864 newvp = NULLVP; 865 nfsstats.lookupcache_misses++; 866 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 867 len = cnp->cn_namelen; 868 nfsm_reqhead(np, NFSPROC_LOOKUP, 869 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 870 nfsm_fhtom(np, v3); 871 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 872 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 873 if (error) { 874 nfsm_postop_attr(dvp, attrflag, 0); 875 m_freem(mrep); 876 goto nfsmout; 877 } 878 nfsm_getfh(fhp, fhsize, v3); 879 880 /* 881 * Handle RENAME case... 882 */ 883 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 884 if (NFS_CMPFH(np, fhp, fhsize)) { 885 m_freem(mrep); 886 return (EISDIR); 887 } 888 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 889 if (error) { 890 m_freem(mrep); 891 return error; 892 } 893 newvp = NFSTOV(np); 894 #ifndef NFS_V2_ONLY 895 if (v3) { 896 nfsm_postop_attr(newvp, attrflag, 0); 897 nfsm_postop_attr(dvp, attrflag, 0); 898 } else 899 #endif 900 nfsm_loadattr(newvp, (struct vattr *)0, 0); 901 *vpp = newvp; 902 m_freem(mrep); 903 goto validate; 904 } 905 906 /* 907 * The postop attr handling is duplicated for each if case, 908 * because it should be done while dvp is locked (unlocking 909 * dvp is different for each case). 910 */ 911 912 if (NFS_CMPFH(np, fhp, fhsize)) { 913 /* 914 * As we handle "." lookup locally, this is 915 * a broken server. 916 */ 917 m_freem(mrep); 918 return EBADRPC; 919 } else if (flags & ISDOTDOT) { 920 /* 921 * ".." lookup 922 */ 923 VOP_UNLOCK(dvp); 924 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 925 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 926 if (error) { 927 m_freem(mrep); 928 return error; 929 } 930 newvp = NFSTOV(np); 931 932 #ifndef NFS_V2_ONLY 933 if (v3) { 934 nfsm_postop_attr(newvp, attrflag, 0); 935 nfsm_postop_attr(dvp, attrflag, 0); 936 } else 937 #endif 938 nfsm_loadattr(newvp, (struct vattr *)0, 0); 939 } else { 940 /* 941 * Other lookups. 942 */ 943 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 944 if (error) { 945 m_freem(mrep); 946 return error; 947 } 948 newvp = NFSTOV(np); 949 #ifndef NFS_V2_ONLY 950 if (v3) { 951 nfsm_postop_attr(newvp, attrflag, 0); 952 nfsm_postop_attr(dvp, attrflag, 0); 953 } else 954 #endif 955 nfsm_loadattr(newvp, (struct vattr *)0, 0); 956 } 957 if (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) { 958 nfs_cache_enter(dvp, newvp, cnp); 959 } 960 *vpp = newvp; 961 nfsm_reqdone; 962 if (error) { 963 /* 964 * We get here only because of errors returned by 965 * the RPC. Otherwise we'll have returned above 966 * (the nfsm_* macros will jump to nfsm_reqdone 967 * on error). 968 */ 969 if (error == ENOENT && cnp->cn_nameiop != CREATE) { 970 nfs_cache_enter(dvp, NULL, cnp); 971 } 972 if (newvp != NULLVP) { 973 if (newvp == dvp) { 974 vrele(newvp); 975 } else { 976 vput(newvp); 977 } 978 } 979 noentry: 980 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 981 (flags & ISLASTCN) && error == ENOENT) { 982 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 983 error = EROFS; 984 } else { 985 error = EJUSTRETURN; 986 } 987 } 988 *vpp = NULL; 989 return error; 990 } 991 992 validate: 993 /* 994 * make sure we have valid type and size. 995 */ 996 997 newvp = *vpp; 998 if (newvp->v_type == VNON) { 999 struct vattr vattr; /* dummy */ 1000 1001 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1002 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1003 if (error) { 1004 vput(newvp); 1005 *vpp = NULL; 1006 } 1007 } 1008 if (error) 1009 return error; 1010 if (newvp != dvp) 1011 VOP_UNLOCK(newvp); 1012 return 0; 1013 } 1014 1015 /* 1016 * nfs read call. 1017 * Just call nfs_bioread() to do the work. 1018 */ 1019 int 1020 nfs_read(void *v) 1021 { 1022 struct vop_read_args /* { 1023 struct vnode *a_vp; 1024 struct uio *a_uio; 1025 int a_ioflag; 1026 kauth_cred_t a_cred; 1027 } */ *ap = v; 1028 struct vnode *vp = ap->a_vp; 1029 1030 if (vp->v_type != VREG) 1031 return EISDIR; 1032 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1033 } 1034 1035 /* 1036 * nfs readlink call 1037 */ 1038 int 1039 nfs_readlink(void *v) 1040 { 1041 struct vop_readlink_args /* { 1042 struct vnode *a_vp; 1043 struct uio *a_uio; 1044 kauth_cred_t a_cred; 1045 } */ *ap = v; 1046 struct vnode *vp = ap->a_vp; 1047 struct nfsnode *np = VTONFS(vp); 1048 1049 if (vp->v_type != VLNK) 1050 return (EPERM); 1051 1052 if (np->n_rcred != NULL) { 1053 kauth_cred_free(np->n_rcred); 1054 } 1055 np->n_rcred = ap->a_cred; 1056 kauth_cred_hold(np->n_rcred); 1057 1058 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1059 } 1060 1061 /* 1062 * Do a readlink rpc. 1063 * Called by nfs_doio() from below the buffer cache. 1064 */ 1065 int 1066 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1067 { 1068 u_int32_t *tl; 1069 char *cp; 1070 int32_t t1, t2; 1071 char *bpos, *dpos, *cp2; 1072 int error = 0; 1073 uint32_t len; 1074 struct mbuf *mreq, *mrep, *md, *mb; 1075 const int v3 = NFS_ISV3(vp); 1076 struct nfsnode *np = VTONFS(vp); 1077 #ifndef NFS_V2_ONLY 1078 int attrflag; 1079 #endif 1080 1081 nfsstats.rpccnt[NFSPROC_READLINK]++; 1082 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1083 nfsm_fhtom(np, v3); 1084 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1085 #ifndef NFS_V2_ONLY 1086 if (v3) 1087 nfsm_postop_attr(vp, attrflag, 0); 1088 #endif 1089 if (!error) { 1090 #ifndef NFS_V2_ONLY 1091 if (v3) { 1092 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1093 len = fxdr_unsigned(uint32_t, *tl); 1094 if (len > NFS_MAXPATHLEN) { 1095 /* 1096 * this pathname is too long for us. 1097 */ 1098 m_freem(mrep); 1099 /* Solaris returns EINVAL. should we follow? */ 1100 error = ENAMETOOLONG; 1101 goto nfsmout; 1102 } 1103 } else 1104 #endif 1105 { 1106 nfsm_strsiz(len, NFS_MAXPATHLEN); 1107 } 1108 nfsm_mtouio(uiop, len); 1109 } 1110 nfsm_reqdone; 1111 return (error); 1112 } 1113 1114 /* 1115 * nfs read rpc call 1116 * Ditto above 1117 */ 1118 int 1119 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1120 { 1121 u_int32_t *tl; 1122 char *cp; 1123 int32_t t1, t2; 1124 char *bpos, *dpos, *cp2; 1125 struct mbuf *mreq, *mrep, *md, *mb; 1126 struct nfsmount *nmp; 1127 int error = 0, len, retlen, tsiz, eof __unused, byte_count; 1128 const int v3 = NFS_ISV3(vp); 1129 struct nfsnode *np = VTONFS(vp); 1130 #ifndef NFS_V2_ONLY 1131 int attrflag; 1132 #endif 1133 1134 #ifndef nolint 1135 eof = 0; 1136 #endif 1137 nmp = VFSTONFS(vp->v_mount); 1138 tsiz = uiop->uio_resid; 1139 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1140 return (EFBIG); 1141 iostat_busy(nmp->nm_stats); 1142 byte_count = 0; /* count bytes actually transferred */ 1143 while (tsiz > 0) { 1144 nfsstats.rpccnt[NFSPROC_READ]++; 1145 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1146 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1147 nfsm_fhtom(np, v3); 1148 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1149 #ifndef NFS_V2_ONLY 1150 if (v3) { 1151 txdr_hyper(uiop->uio_offset, tl); 1152 *(tl + 2) = txdr_unsigned(len); 1153 } else 1154 #endif 1155 { 1156 *tl++ = txdr_unsigned(uiop->uio_offset); 1157 *tl++ = txdr_unsigned(len); 1158 *tl = 0; 1159 } 1160 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1161 #ifndef NFS_V2_ONLY 1162 if (v3) { 1163 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1164 if (error) { 1165 m_freem(mrep); 1166 goto nfsmout; 1167 } 1168 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1169 eof = fxdr_unsigned(int, *(tl + 1)); 1170 } else 1171 #endif 1172 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1173 nfsm_strsiz(retlen, nmp->nm_rsize); 1174 nfsm_mtouio(uiop, retlen); 1175 m_freem(mrep); 1176 tsiz -= retlen; 1177 byte_count += retlen; 1178 #ifndef NFS_V2_ONLY 1179 if (v3) { 1180 if (eof || retlen == 0) 1181 tsiz = 0; 1182 } else 1183 #endif 1184 if (retlen < len) 1185 tsiz = 0; 1186 } 1187 nfsmout: 1188 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1189 return (error); 1190 } 1191 1192 struct nfs_writerpc_context { 1193 kmutex_t nwc_lock; 1194 kcondvar_t nwc_cv; 1195 int nwc_mbufcount; 1196 }; 1197 1198 /* 1199 * free mbuf used to refer protected pages while write rpc call. 1200 * called at splvm. 1201 */ 1202 static void 1203 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1204 { 1205 struct nfs_writerpc_context *ctx = arg; 1206 1207 KASSERT(m != NULL); 1208 KASSERT(ctx != NULL); 1209 pool_cache_put(mb_cache, m); 1210 mutex_enter(&ctx->nwc_lock); 1211 if (--ctx->nwc_mbufcount == 0) { 1212 cv_signal(&ctx->nwc_cv); 1213 } 1214 mutex_exit(&ctx->nwc_lock); 1215 } 1216 1217 /* 1218 * nfs write call 1219 */ 1220 int 1221 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1222 { 1223 u_int32_t *tl; 1224 char *cp; 1225 int32_t t1, t2; 1226 char *bpos, *dpos; 1227 struct mbuf *mreq, *mrep, *md, *mb; 1228 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1229 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1230 const int v3 = NFS_ISV3(vp); 1231 int committed = NFSV3WRITE_FILESYNC; 1232 struct nfsnode *np = VTONFS(vp); 1233 struct nfs_writerpc_context ctx; 1234 int byte_count; 1235 size_t origresid; 1236 #ifndef NFS_V2_ONLY 1237 char *cp2; 1238 int rlen, commit; 1239 #endif 1240 1241 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1242 panic("writerpc readonly vp %p", vp); 1243 } 1244 1245 #ifdef DIAGNOSTIC 1246 if (uiop->uio_iovcnt != 1) 1247 panic("nfs: writerpc iovcnt > 1"); 1248 #endif 1249 tsiz = uiop->uio_resid; 1250 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1251 return EFBIG; 1252 1253 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1254 cv_init(&ctx.nwc_cv, "nfsmblk"); 1255 ctx.nwc_mbufcount = 1; 1256 1257 retry: 1258 origresid = uiop->uio_resid; 1259 KASSERT(origresid == uiop->uio_iov->iov_len); 1260 iostat_busy(nmp->nm_stats); 1261 byte_count = 0; /* count of bytes actually written */ 1262 while (tsiz > 0) { 1263 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1264 size_t backup; 1265 bool stalewriteverf = false; 1266 1267 nfsstats.rpccnt[NFSPROC_WRITE]++; 1268 len = uimin(tsiz, nmp->nm_wsize); 1269 datalen = pageprotected ? 0 : nfsm_rndup(len); 1270 nfsm_reqhead(np, NFSPROC_WRITE, 1271 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1272 nfsm_fhtom(np, v3); 1273 #ifndef NFS_V2_ONLY 1274 if (v3) { 1275 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1276 txdr_hyper(uiop->uio_offset, tl); 1277 tl += 2; 1278 *tl++ = txdr_unsigned(len); 1279 *tl++ = txdr_unsigned(*iomode); 1280 *tl = txdr_unsigned(len); 1281 } else 1282 #endif 1283 { 1284 u_int32_t x; 1285 1286 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1287 /* Set both "begin" and "current" to non-garbage. */ 1288 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1289 *tl++ = x; /* "begin offset" */ 1290 *tl++ = x; /* "current offset" */ 1291 x = txdr_unsigned(len); 1292 *tl++ = x; /* total to this offset */ 1293 *tl = x; /* size of this write */ 1294 1295 } 1296 if (pageprotected) { 1297 /* 1298 * since we know pages can't be modified during i/o, 1299 * no need to copy them for us. 1300 */ 1301 struct mbuf *m; 1302 struct iovec *iovp = uiop->uio_iov; 1303 1304 m = m_get(M_WAIT, MT_DATA); 1305 MCLAIM(m, &nfs_mowner); 1306 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1307 nfs_writerpc_extfree, &ctx); 1308 m->m_flags |= M_EXT_ROMAP; 1309 m->m_len = len; 1310 mb->m_next = m; 1311 /* 1312 * no need to maintain mb and bpos here 1313 * because no one care them later. 1314 */ 1315 #if 0 1316 mb = m; 1317 bpos = mtod(void *, mb) + mb->m_len; 1318 #endif 1319 UIO_ADVANCE(uiop, len); 1320 uiop->uio_offset += len; 1321 mutex_enter(&ctx.nwc_lock); 1322 ctx.nwc_mbufcount++; 1323 mutex_exit(&ctx.nwc_lock); 1324 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1325 } else { 1326 nfsm_uiotom(uiop, len); 1327 } 1328 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1329 #ifndef NFS_V2_ONLY 1330 if (v3) { 1331 wccflag = NFSV3_WCCCHK; 1332 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1333 if (!error) { 1334 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1335 + NFSX_V3WRITEVERF); 1336 rlen = fxdr_unsigned(int, *tl++); 1337 if (rlen == 0) { 1338 error = NFSERR_IO; 1339 m_freem(mrep); 1340 break; 1341 } else if (rlen < len) { 1342 backup = len - rlen; 1343 UIO_ADVANCE(uiop, -backup); 1344 uiop->uio_offset -= backup; 1345 len = rlen; 1346 } 1347 commit = fxdr_unsigned(int, *tl++); 1348 1349 /* 1350 * Return the lowest committment level 1351 * obtained by any of the RPCs. 1352 */ 1353 if (committed == NFSV3WRITE_FILESYNC) 1354 committed = commit; 1355 else if (committed == NFSV3WRITE_DATASYNC && 1356 commit == NFSV3WRITE_UNSTABLE) 1357 committed = commit; 1358 mutex_enter(&nmp->nm_lock); 1359 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1360 memcpy(nmp->nm_writeverf, tl, 1361 NFSX_V3WRITEVERF); 1362 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1363 } else if ((nmp->nm_iflag & 1364 NFSMNT_STALEWRITEVERF) || 1365 memcmp(tl, nmp->nm_writeverf, 1366 NFSX_V3WRITEVERF)) { 1367 memcpy(nmp->nm_writeverf, tl, 1368 NFSX_V3WRITEVERF); 1369 /* 1370 * note NFSMNT_STALEWRITEVERF 1371 * if we're the first thread to 1372 * notice it. 1373 */ 1374 if ((nmp->nm_iflag & 1375 NFSMNT_STALEWRITEVERF) == 0) { 1376 stalewriteverf = true; 1377 nmp->nm_iflag |= 1378 NFSMNT_STALEWRITEVERF; 1379 } 1380 } 1381 mutex_exit(&nmp->nm_lock); 1382 } 1383 } else 1384 #endif 1385 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1386 if (wccflag) 1387 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1388 m_freem(mrep); 1389 if (error) 1390 break; 1391 tsiz -= len; 1392 byte_count += len; 1393 if (stalewriteverf) { 1394 *stalewriteverfp = true; 1395 stalewriteverf = false; 1396 if (committed == NFSV3WRITE_UNSTABLE && 1397 len != origresid) { 1398 /* 1399 * if our write requests weren't atomic but 1400 * unstable, datas in previous iterations 1401 * might have already been lost now. 1402 * then, we should resend them to nfsd. 1403 */ 1404 backup = origresid - tsiz; 1405 UIO_ADVANCE(uiop, -backup); 1406 uiop->uio_offset -= backup; 1407 tsiz = origresid; 1408 goto retry; 1409 } 1410 } 1411 } 1412 nfsmout: 1413 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1414 if (pageprotected) { 1415 /* 1416 * wait until mbufs go away. 1417 * retransmitted mbufs can survive longer than rpc requests 1418 * themselves. 1419 */ 1420 mutex_enter(&ctx.nwc_lock); 1421 ctx.nwc_mbufcount--; 1422 while (ctx.nwc_mbufcount > 0) { 1423 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1424 } 1425 mutex_exit(&ctx.nwc_lock); 1426 } 1427 mutex_destroy(&ctx.nwc_lock); 1428 cv_destroy(&ctx.nwc_cv); 1429 *iomode = committed; 1430 if (error) 1431 uiop->uio_resid = tsiz; 1432 return (error); 1433 } 1434 1435 /* 1436 * nfs mknod rpc 1437 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1438 * mode set to specify the file type and the size field for rdev. 1439 */ 1440 int 1441 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1442 { 1443 struct nfsv2_sattr *sp; 1444 u_int32_t *tl; 1445 char *cp; 1446 int32_t t1, t2; 1447 struct vnode *newvp = (struct vnode *)0; 1448 struct nfsnode *dnp, *np; 1449 char *cp2; 1450 char *bpos, *dpos; 1451 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1452 struct mbuf *mreq, *mrep, *md, *mb; 1453 u_int32_t rdev; 1454 const int v3 = NFS_ISV3(dvp); 1455 1456 if (vap->va_type == VCHR || vap->va_type == VBLK) 1457 rdev = txdr_unsigned(vap->va_rdev); 1458 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1459 rdev = nfs_xdrneg1; 1460 else { 1461 VOP_ABORTOP(dvp, cnp); 1462 return (EOPNOTSUPP); 1463 } 1464 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1465 dnp = VTONFS(dvp); 1466 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1467 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1468 nfsm_fhtom(dnp, v3); 1469 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1470 #ifndef NFS_V2_ONLY 1471 if (v3) { 1472 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1473 *tl++ = vtonfsv3_type(vap->va_type); 1474 nfsm_v3attrbuild(vap, false); 1475 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1476 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1477 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1478 *tl = txdr_unsigned(minor(vap->va_rdev)); 1479 } 1480 } else 1481 #endif 1482 { 1483 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1484 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1485 sp->sa_uid = nfs_xdrneg1; 1486 sp->sa_gid = nfs_xdrneg1; 1487 sp->sa_size = rdev; 1488 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1489 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1490 } 1491 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1492 if (!error) { 1493 nfsm_mtofh(dvp, newvp, v3, gotvp); 1494 if (!gotvp) { 1495 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1496 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1497 if (!error) 1498 newvp = NFSTOV(np); 1499 } 1500 } 1501 #ifndef NFS_V2_ONLY 1502 if (v3) 1503 nfsm_wcc_data(dvp, wccflag, 0, !error); 1504 #endif 1505 nfsm_reqdone; 1506 if (error) { 1507 if (newvp) 1508 vput(newvp); 1509 } else { 1510 nfs_cache_enter(dvp, newvp, cnp); 1511 *vpp = newvp; 1512 VOP_UNLOCK(newvp); 1513 } 1514 VTONFS(dvp)->n_flag |= NMODIFIED; 1515 if (!wccflag) 1516 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1517 return (error); 1518 } 1519 1520 /* 1521 * nfs mknod vop 1522 * just call nfs_mknodrpc() to do the work. 1523 */ 1524 /* ARGSUSED */ 1525 int 1526 nfs_mknod(void *v) 1527 { 1528 struct vop_mknod_v3_args /* { 1529 struct vnode *a_dvp; 1530 struct vnode **a_vpp; 1531 struct componentname *a_cnp; 1532 struct vattr *a_vap; 1533 } */ *ap = v; 1534 struct vnode *dvp = ap->a_dvp; 1535 struct componentname *cnp = ap->a_cnp; 1536 int error; 1537 1538 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1539 VN_KNOTE(dvp, NOTE_WRITE); 1540 if (error == 0 || error == EEXIST) 1541 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1542 return (error); 1543 } 1544 1545 /* 1546 * nfs file create call 1547 */ 1548 int 1549 nfs_create(void *v) 1550 { 1551 struct vop_create_v3_args /* { 1552 struct vnode *a_dvp; 1553 struct vnode **a_vpp; 1554 struct componentname *a_cnp; 1555 struct vattr *a_vap; 1556 } */ *ap = v; 1557 struct vnode *dvp = ap->a_dvp; 1558 struct vattr *vap = ap->a_vap; 1559 struct componentname *cnp = ap->a_cnp; 1560 struct nfsv2_sattr *sp; 1561 u_int32_t *tl; 1562 char *cp; 1563 int32_t t1, t2; 1564 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1565 struct vnode *newvp = (struct vnode *)0; 1566 char *bpos, *dpos, *cp2; 1567 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1568 struct mbuf *mreq, *mrep, *md, *mb; 1569 const int v3 = NFS_ISV3(dvp); 1570 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1571 1572 /* 1573 * Oops, not for me.. 1574 */ 1575 if (vap->va_type == VSOCK) 1576 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1577 1578 KASSERT(vap->va_type == VREG); 1579 1580 #ifdef VA_EXCLUSIVE 1581 if (vap->va_vaflags & VA_EXCLUSIVE) { 1582 excl_mode = NFSV3CREATE_EXCLUSIVE; 1583 } 1584 #endif 1585 again: 1586 error = 0; 1587 nfsstats.rpccnt[NFSPROC_CREATE]++; 1588 dnp = VTONFS(dvp); 1589 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1590 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1591 nfsm_fhtom(dnp, v3); 1592 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1593 #ifndef NFS_V2_ONLY 1594 if (v3) { 1595 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1596 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1597 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1598 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1599 *tl++ = cprng_fast32(); 1600 *tl = cprng_fast32(); 1601 } else { 1602 *tl = txdr_unsigned(excl_mode); 1603 nfsm_v3attrbuild(vap, false); 1604 } 1605 } else 1606 #endif 1607 { 1608 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1609 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1610 sp->sa_uid = nfs_xdrneg1; 1611 sp->sa_gid = nfs_xdrneg1; 1612 sp->sa_size = 0; 1613 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1614 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1615 } 1616 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1617 if (!error) { 1618 nfsm_mtofh(dvp, newvp, v3, gotvp); 1619 if (!gotvp) { 1620 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1621 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1622 if (!error) 1623 newvp = NFSTOV(np); 1624 } 1625 } 1626 #ifndef NFS_V2_ONLY 1627 if (v3) 1628 nfsm_wcc_data(dvp, wccflag, 0, !error); 1629 #endif 1630 nfsm_reqdone; 1631 if (error) { 1632 /* 1633 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1634 */ 1635 if (v3 && error == ENOTSUP) { 1636 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1637 excl_mode = NFSV3CREATE_GUARDED; 1638 goto again; 1639 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1640 excl_mode = NFSV3CREATE_UNCHECKED; 1641 goto again; 1642 } 1643 } 1644 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1645 struct timespec ts; 1646 1647 getnanotime(&ts); 1648 1649 /* 1650 * make sure that we'll update timestamps as 1651 * most server implementations use them to store 1652 * the create verifier. 1653 * 1654 * XXX it's better to use TOSERVER always. 1655 */ 1656 1657 if (vap->va_atime.tv_sec == VNOVAL) 1658 vap->va_atime = ts; 1659 if (vap->va_mtime.tv_sec == VNOVAL) 1660 vap->va_mtime = ts; 1661 1662 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1663 } 1664 if (error == 0) { 1665 if (cnp->cn_flags & MAKEENTRY) 1666 nfs_cache_enter(dvp, newvp, cnp); 1667 else 1668 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1669 *ap->a_vpp = newvp; 1670 VOP_UNLOCK(newvp); 1671 } else { 1672 if (newvp) 1673 vput(newvp); 1674 if (error == EEXIST) 1675 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 1676 } 1677 VTONFS(dvp)->n_flag |= NMODIFIED; 1678 if (!wccflag) 1679 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1680 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1681 return (error); 1682 } 1683 1684 /* 1685 * nfs file remove call 1686 * To try and make nfs semantics closer to ufs semantics, a file that has 1687 * other processes using the vnode is renamed instead of removed and then 1688 * removed later on the last close. 1689 * - If vrefcnt(vp) > 1 1690 * If a rename is not already in the works 1691 * call nfs_sillyrename() to set it up 1692 * else 1693 * do the remove rpc 1694 */ 1695 int 1696 nfs_remove(void *v) 1697 { 1698 struct vop_remove_v2_args /* { 1699 struct vnodeop_desc *a_desc; 1700 struct vnode * a_dvp; 1701 struct vnode * a_vp; 1702 struct componentname * a_cnp; 1703 } */ *ap = v; 1704 struct vnode *vp = ap->a_vp; 1705 struct vnode *dvp = ap->a_dvp; 1706 struct componentname *cnp = ap->a_cnp; 1707 struct nfsnode *np = VTONFS(vp); 1708 int error = 0; 1709 struct vattr vattr; 1710 1711 #ifndef DIAGNOSTIC 1712 if (vrefcnt(vp) < 1) 1713 panic("nfs_remove: bad vrefcnt(vp)"); 1714 #endif 1715 if (vp->v_type == VDIR) 1716 error = EPERM; 1717 else if (vrefcnt(vp) == 1 || (np->n_sillyrename && 1718 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1719 vattr.va_nlink > 1)) { 1720 /* 1721 * Purge the name cache so that the chance of a lookup for 1722 * the name succeeding while the remove is in progress is 1723 * minimized. Without node locking it can still happen, such 1724 * that an I/O op returns ESTALE, but since you get this if 1725 * another host removes the file.. 1726 */ 1727 cache_purge(vp); 1728 /* 1729 * throw away biocache buffers, mainly to avoid 1730 * unnecessary delayed writes later. 1731 */ 1732 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1733 /* Do the rpc */ 1734 if (error != EINTR) 1735 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1736 cnp->cn_namelen, cnp->cn_cred, curlwp); 1737 } else if (!np->n_sillyrename) 1738 error = nfs_sillyrename(dvp, vp, cnp, false); 1739 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1740 vattr.va_nlink == 1) { 1741 np->n_flag |= NREMOVED; 1742 } 1743 NFS_INVALIDATE_ATTRCACHE(np); 1744 VN_KNOTE(vp, NOTE_DELETE); 1745 VN_KNOTE(dvp, NOTE_WRITE); 1746 if (dvp == vp) 1747 vrele(vp); 1748 else 1749 vput(vp); 1750 return (error); 1751 } 1752 1753 /* 1754 * nfs file remove rpc called from nfs_inactive 1755 */ 1756 int 1757 nfs_removeit(struct sillyrename *sp) 1758 { 1759 1760 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1761 (struct lwp *)0)); 1762 } 1763 1764 /* 1765 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1766 */ 1767 int 1768 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1769 { 1770 u_int32_t *tl; 1771 char *cp; 1772 #ifndef NFS_V2_ONLY 1773 int32_t t1; 1774 char *cp2; 1775 #endif 1776 int32_t t2; 1777 char *bpos, *dpos; 1778 int error = 0, wccflag = NFSV3_WCCRATTR; 1779 struct mbuf *mreq, *mrep, *md, *mb; 1780 const int v3 = NFS_ISV3(dvp); 1781 int rexmit = 0; 1782 struct nfsnode *dnp = VTONFS(dvp); 1783 1784 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1785 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1786 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1787 nfsm_fhtom(dnp, v3); 1788 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1789 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1790 #ifndef NFS_V2_ONLY 1791 if (v3) 1792 nfsm_wcc_data(dvp, wccflag, 0, !error); 1793 #endif 1794 nfsm_reqdone; 1795 VTONFS(dvp)->n_flag |= NMODIFIED; 1796 if (!wccflag) 1797 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1798 /* 1799 * Kludge City: If the first reply to the remove rpc is lost.. 1800 * the reply to the retransmitted request will be ENOENT 1801 * since the file was in fact removed 1802 * Therefore, we cheat and return success. 1803 */ 1804 if (rexmit && error == ENOENT) 1805 error = 0; 1806 return (error); 1807 } 1808 1809 /* 1810 * nfs file rename call 1811 */ 1812 int 1813 nfs_rename(void *v) 1814 { 1815 struct vop_rename_args /* { 1816 struct vnode *a_fdvp; 1817 struct vnode *a_fvp; 1818 struct componentname *a_fcnp; 1819 struct vnode *a_tdvp; 1820 struct vnode *a_tvp; 1821 struct componentname *a_tcnp; 1822 } */ *ap = v; 1823 struct vnode *fvp = ap->a_fvp; 1824 struct vnode *tvp = ap->a_tvp; 1825 struct vnode *fdvp = ap->a_fdvp; 1826 struct vnode *tdvp = ap->a_tdvp; 1827 struct componentname *tcnp = ap->a_tcnp; 1828 struct componentname *fcnp = ap->a_fcnp; 1829 int error; 1830 1831 /* Check for cross-device rename */ 1832 if ((fvp->v_mount != tdvp->v_mount) || 1833 (tvp && (fvp->v_mount != tvp->v_mount))) { 1834 error = EXDEV; 1835 goto out; 1836 } 1837 1838 /* 1839 * If the tvp exists and is in use, sillyrename it before doing the 1840 * rename of the new file over it. 1841 * 1842 * Have sillyrename use link instead of rename if possible, 1843 * so that we don't lose the file if the rename fails, and so 1844 * that there's no window when the "to" file doesn't exist. 1845 */ 1846 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 1847 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1848 VN_KNOTE(tvp, NOTE_DELETE); 1849 vput(tvp); 1850 tvp = NULL; 1851 } 1852 1853 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1854 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1855 curlwp); 1856 1857 VN_KNOTE(fdvp, NOTE_WRITE); 1858 VN_KNOTE(tdvp, NOTE_WRITE); 1859 if (error == 0 || error == EEXIST) { 1860 if (fvp->v_type == VDIR) 1861 cache_purge(fvp); 1862 else 1863 cache_purge1(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1864 0); 1865 if (tvp != NULL && tvp->v_type == VDIR) 1866 cache_purge(tvp); 1867 else 1868 cache_purge1(tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, 1869 0); 1870 } 1871 out: 1872 if (tdvp == tvp) 1873 vrele(tdvp); 1874 else 1875 vput(tdvp); 1876 if (tvp) 1877 vput(tvp); 1878 vrele(fdvp); 1879 vrele(fvp); 1880 return (error); 1881 } 1882 1883 /* 1884 * nfs file rename rpc called from nfs_remove() above 1885 */ 1886 int 1887 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1888 { 1889 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1890 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1891 } 1892 1893 /* 1894 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1895 */ 1896 int 1897 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) 1898 { 1899 u_int32_t *tl; 1900 char *cp; 1901 #ifndef NFS_V2_ONLY 1902 int32_t t1; 1903 char *cp2; 1904 #endif 1905 int32_t t2; 1906 char *bpos, *dpos; 1907 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1908 struct mbuf *mreq, *mrep, *md, *mb; 1909 const int v3 = NFS_ISV3(fdvp); 1910 int rexmit = 0; 1911 struct nfsnode *fdnp = VTONFS(fdvp); 1912 1913 nfsstats.rpccnt[NFSPROC_RENAME]++; 1914 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1915 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1916 nfsm_rndup(tnamelen)); 1917 nfsm_fhtom(fdnp, v3); 1918 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1919 nfsm_fhtom(VTONFS(tdvp), v3); 1920 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1921 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1922 #ifndef NFS_V2_ONLY 1923 if (v3) { 1924 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1925 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1926 } 1927 #endif 1928 nfsm_reqdone; 1929 VTONFS(fdvp)->n_flag |= NMODIFIED; 1930 VTONFS(tdvp)->n_flag |= NMODIFIED; 1931 if (!fwccflag) 1932 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 1933 if (!twccflag) 1934 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 1935 /* 1936 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1937 */ 1938 if (rexmit && error == ENOENT) 1939 error = 0; 1940 return (error); 1941 } 1942 1943 /* 1944 * NFS link RPC, called from nfs_link. 1945 * Assumes dvp and vp locked, and leaves them that way. 1946 */ 1947 1948 static int 1949 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 1950 size_t namelen, kauth_cred_t cred, struct lwp *l) 1951 { 1952 u_int32_t *tl; 1953 char *cp; 1954 #ifndef NFS_V2_ONLY 1955 int32_t t1; 1956 char *cp2; 1957 #endif 1958 int32_t t2; 1959 char *bpos, *dpos; 1960 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1961 struct mbuf *mreq, *mrep, *md, *mb; 1962 const int v3 = NFS_ISV3(dvp); 1963 int rexmit = 0; 1964 struct nfsnode *np = VTONFS(vp); 1965 1966 nfsstats.rpccnt[NFSPROC_LINK]++; 1967 nfsm_reqhead(np, NFSPROC_LINK, 1968 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1969 nfsm_fhtom(np, v3); 1970 nfsm_fhtom(VTONFS(dvp), v3); 1971 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1972 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 1973 #ifndef NFS_V2_ONLY 1974 if (v3) { 1975 nfsm_postop_attr(vp, attrflag, 0); 1976 nfsm_wcc_data(dvp, wccflag, 0, !error); 1977 } 1978 #endif 1979 nfsm_reqdone; 1980 1981 VTONFS(dvp)->n_flag |= NMODIFIED; 1982 if (!attrflag) 1983 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 1984 if (!wccflag) 1985 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1986 1987 /* 1988 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1989 */ 1990 if (rexmit && error == EEXIST) 1991 error = 0; 1992 1993 return error; 1994 } 1995 1996 /* 1997 * nfs hard link create call 1998 */ 1999 int 2000 nfs_link(void *v) 2001 { 2002 struct vop_link_v2_args /* { 2003 struct vnode *a_dvp; 2004 struct vnode *a_vp; 2005 struct componentname *a_cnp; 2006 } */ *ap = v; 2007 struct vnode *vp = ap->a_vp; 2008 struct vnode *dvp = ap->a_dvp; 2009 struct componentname *cnp = ap->a_cnp; 2010 int error = 0; 2011 2012 error = vn_lock(vp, LK_EXCLUSIVE); 2013 if (error != 0) { 2014 VOP_ABORTOP(dvp, cnp); 2015 return error; 2016 } 2017 2018 /* 2019 * Push all writes to the server, so that the attribute cache 2020 * doesn't get "out of sync" with the server. 2021 * XXX There should be a better way! 2022 */ 2023 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2024 2025 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2026 cnp->cn_cred, curlwp); 2027 2028 if (error == 0) { 2029 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2030 } 2031 VOP_UNLOCK(vp); 2032 VN_KNOTE(vp, NOTE_LINK); 2033 VN_KNOTE(dvp, NOTE_WRITE); 2034 return (error); 2035 } 2036 2037 /* 2038 * nfs symbolic link create call 2039 */ 2040 int 2041 nfs_symlink(void *v) 2042 { 2043 struct vop_symlink_v3_args /* { 2044 struct vnode *a_dvp; 2045 struct vnode **a_vpp; 2046 struct componentname *a_cnp; 2047 struct vattr *a_vap; 2048 char *a_target; 2049 } */ *ap = v; 2050 struct vnode *dvp = ap->a_dvp; 2051 struct vattr *vap = ap->a_vap; 2052 struct componentname *cnp = ap->a_cnp; 2053 struct nfsv2_sattr *sp; 2054 u_int32_t *tl; 2055 char *cp; 2056 int32_t t1, t2; 2057 char *bpos, *dpos, *cp2; 2058 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2059 struct mbuf *mreq, *mrep, *md, *mb; 2060 struct vnode *newvp = (struct vnode *)0; 2061 const int v3 = NFS_ISV3(dvp); 2062 int rexmit = 0; 2063 struct nfsnode *dnp = VTONFS(dvp); 2064 2065 *ap->a_vpp = NULL; 2066 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2067 slen = strlen(ap->a_target); 2068 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2069 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2070 nfsm_fhtom(dnp, v3); 2071 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2072 #ifndef NFS_V2_ONlY 2073 if (v3) 2074 nfsm_v3attrbuild(vap, false); 2075 #endif 2076 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2077 #ifndef NFS_V2_ONlY 2078 if (!v3) { 2079 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2080 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2081 sp->sa_uid = nfs_xdrneg1; 2082 sp->sa_gid = nfs_xdrneg1; 2083 sp->sa_size = nfs_xdrneg1; 2084 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2085 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2086 } 2087 #endif 2088 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2089 &rexmit); 2090 #ifndef NFS_V2_ONlY 2091 if (v3) { 2092 if (!error) 2093 nfsm_mtofh(dvp, newvp, v3, gotvp); 2094 nfsm_wcc_data(dvp, wccflag, 0, !error); 2095 } 2096 #endif 2097 nfsm_reqdone; 2098 /* 2099 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2100 */ 2101 if (rexmit && error == EEXIST) 2102 error = 0; 2103 if (error == 0 || error == EEXIST) 2104 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0); 2105 if (error == 0 && newvp == NULL) { 2106 struct nfsnode *np = NULL; 2107 2108 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2109 cnp->cn_cred, curlwp, &np); 2110 if (error == 0) 2111 newvp = NFSTOV(np); 2112 } 2113 if (error) { 2114 if (newvp != NULL) 2115 vput(newvp); 2116 } else { 2117 *ap->a_vpp = newvp; 2118 VOP_UNLOCK(newvp); 2119 } 2120 VTONFS(dvp)->n_flag |= NMODIFIED; 2121 if (!wccflag) 2122 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2123 VN_KNOTE(dvp, NOTE_WRITE); 2124 return (error); 2125 } 2126 2127 /* 2128 * nfs make dir call 2129 */ 2130 int 2131 nfs_mkdir(void *v) 2132 { 2133 struct vop_mkdir_v3_args /* { 2134 struct vnode *a_dvp; 2135 struct vnode **a_vpp; 2136 struct componentname *a_cnp; 2137 struct vattr *a_vap; 2138 } */ *ap = v; 2139 struct vnode *dvp = ap->a_dvp; 2140 struct vattr *vap = ap->a_vap; 2141 struct componentname *cnp = ap->a_cnp; 2142 struct nfsv2_sattr *sp; 2143 u_int32_t *tl; 2144 char *cp; 2145 int32_t t1, t2; 2146 int len; 2147 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2148 struct vnode *newvp = (struct vnode *)0; 2149 char *bpos, *dpos, *cp2; 2150 int error = 0, wccflag = NFSV3_WCCRATTR; 2151 int gotvp = 0; 2152 int rexmit = 0; 2153 struct mbuf *mreq, *mrep, *md, *mb; 2154 const int v3 = NFS_ISV3(dvp); 2155 2156 len = cnp->cn_namelen; 2157 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2158 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2159 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2160 nfsm_fhtom(dnp, v3); 2161 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2162 #ifndef NFS_V2_ONLY 2163 if (v3) { 2164 nfsm_v3attrbuild(vap, false); 2165 } else 2166 #endif 2167 { 2168 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2169 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2170 sp->sa_uid = nfs_xdrneg1; 2171 sp->sa_gid = nfs_xdrneg1; 2172 sp->sa_size = nfs_xdrneg1; 2173 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2174 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2175 } 2176 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2177 if (!error) 2178 nfsm_mtofh(dvp, newvp, v3, gotvp); 2179 if (v3) 2180 nfsm_wcc_data(dvp, wccflag, 0, !error); 2181 nfsm_reqdone; 2182 VTONFS(dvp)->n_flag |= NMODIFIED; 2183 if (!wccflag) 2184 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2185 /* 2186 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2187 * if we can succeed in looking up the directory. 2188 */ 2189 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2190 if (newvp) { 2191 vput(newvp); 2192 newvp = (struct vnode *)0; 2193 } 2194 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2195 curlwp, &np); 2196 if (!error) { 2197 newvp = NFSTOV(np); 2198 if (newvp->v_type != VDIR || newvp == dvp) 2199 error = EEXIST; 2200 } 2201 } 2202 if (error) { 2203 if (newvp) { 2204 if (dvp != newvp) 2205 vput(newvp); 2206 else 2207 vrele(newvp); 2208 } 2209 } else { 2210 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2211 nfs_cache_enter(dvp, newvp, cnp); 2212 *ap->a_vpp = newvp; 2213 VOP_UNLOCK(newvp); 2214 } 2215 return (error); 2216 } 2217 2218 /* 2219 * nfs remove directory call 2220 */ 2221 int 2222 nfs_rmdir(void *v) 2223 { 2224 struct vop_rmdir_v2_args /* { 2225 struct vnode *a_dvp; 2226 struct vnode *a_vp; 2227 struct componentname *a_cnp; 2228 } */ *ap = v; 2229 struct vnode *vp = ap->a_vp; 2230 struct vnode *dvp = ap->a_dvp; 2231 struct componentname *cnp = ap->a_cnp; 2232 u_int32_t *tl; 2233 char *cp; 2234 #ifndef NFS_V2_ONLY 2235 int32_t t1; 2236 char *cp2; 2237 #endif 2238 int32_t t2; 2239 char *bpos, *dpos; 2240 int error = 0, wccflag = NFSV3_WCCRATTR; 2241 int rexmit = 0; 2242 struct mbuf *mreq, *mrep, *md, *mb; 2243 const int v3 = NFS_ISV3(dvp); 2244 struct nfsnode *dnp; 2245 2246 if (dvp == vp) { 2247 vrele(vp); 2248 return (EINVAL); 2249 } 2250 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2251 dnp = VTONFS(dvp); 2252 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2253 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2254 nfsm_fhtom(dnp, v3); 2255 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2256 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2257 #ifndef NFS_V2_ONLY 2258 if (v3) 2259 nfsm_wcc_data(dvp, wccflag, 0, !error); 2260 #endif 2261 nfsm_reqdone; 2262 VTONFS(dvp)->n_flag |= NMODIFIED; 2263 if (!wccflag) 2264 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2265 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2266 VN_KNOTE(vp, NOTE_DELETE); 2267 cache_purge(vp); 2268 vput(vp); 2269 /* 2270 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2271 */ 2272 if (rexmit && error == ENOENT) 2273 error = 0; 2274 return (error); 2275 } 2276 2277 /* 2278 * nfs readdir call 2279 */ 2280 int 2281 nfs_readdir(void *v) 2282 { 2283 struct vop_readdir_args /* { 2284 struct vnode *a_vp; 2285 struct uio *a_uio; 2286 kauth_cred_t a_cred; 2287 int *a_eofflag; 2288 off_t **a_cookies; 2289 int *a_ncookies; 2290 } */ *ap = v; 2291 struct vnode *vp = ap->a_vp; 2292 struct uio *uio = ap->a_uio; 2293 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2294 char *base = uio->uio_iov->iov_base; 2295 int tresid, error; 2296 size_t count, lost; 2297 struct dirent *dp; 2298 off_t *cookies = NULL; 2299 int ncookies = 0, nc; 2300 2301 if (vp->v_type != VDIR) 2302 return (EPERM); 2303 2304 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2305 count = uio->uio_resid - lost; 2306 if (count <= 0) 2307 return (EINVAL); 2308 2309 /* 2310 * Call nfs_bioread() to do the real work. 2311 */ 2312 tresid = uio->uio_resid = count; 2313 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2314 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2315 2316 if (!error && ap->a_cookies) { 2317 ncookies = count / 16; 2318 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2319 *ap->a_cookies = cookies; 2320 } 2321 2322 if (!error && uio->uio_resid == tresid) { 2323 uio->uio_resid += lost; 2324 nfsstats.direofcache_misses++; 2325 if (ap->a_cookies) 2326 *ap->a_ncookies = 0; 2327 *ap->a_eofflag = 1; 2328 return (0); 2329 } 2330 2331 if (!error && ap->a_cookies) { 2332 /* 2333 * Only the NFS server and emulations use cookies, and they 2334 * load the directory block into system space, so we can 2335 * just look at it directly. 2336 */ 2337 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2338 uio->uio_iovcnt != 1) 2339 panic("nfs_readdir: lost in space"); 2340 for (nc = 0; ncookies-- && 2341 base < (char *)uio->uio_iov->iov_base; nc++){ 2342 dp = (struct dirent *) base; 2343 if (dp->d_reclen == 0) 2344 break; 2345 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2346 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2347 else 2348 *(cookies++) = NFS_GETCOOKIE(dp); 2349 base += dp->d_reclen; 2350 } 2351 uio->uio_resid += 2352 ((char *)uio->uio_iov->iov_base - base); 2353 uio->uio_iov->iov_len += 2354 ((char *)uio->uio_iov->iov_base - base); 2355 uio->uio_iov->iov_base = base; 2356 *ap->a_ncookies = nc; 2357 } 2358 2359 uio->uio_resid += lost; 2360 *ap->a_eofflag = 0; 2361 return (error); 2362 } 2363 2364 /* 2365 * Readdir rpc call. 2366 * Called from below the buffer cache by nfs_doio(). 2367 */ 2368 int 2369 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2370 { 2371 int len, left; 2372 struct dirent *dp = NULL; 2373 u_int32_t *tl; 2374 char *cp; 2375 int32_t t1, t2; 2376 char *bpos, *dpos, *cp2; 2377 struct mbuf *mreq, *mrep, *md, *mb; 2378 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2379 struct nfsnode *dnp = VTONFS(vp); 2380 u_quad_t fileno; 2381 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2382 #ifndef NFS_V2_ONLY 2383 int attrflag; 2384 #endif 2385 int nrpcs = 0, reclen; 2386 const int v3 = NFS_ISV3(vp); 2387 2388 #ifdef DIAGNOSTIC 2389 /* 2390 * Should be called from buffer cache, so only amount of 2391 * NFS_DIRBLKSIZ will be requested. 2392 */ 2393 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2394 panic("nfs readdirrpc bad uio"); 2395 #endif 2396 2397 /* 2398 * Loop around doing readdir rpc's of size nm_readdirsize 2399 * truncated to a multiple of NFS_DIRFRAGSIZ. 2400 * The stopping criteria is EOF or buffer full. 2401 */ 2402 while (more_dirs && bigenough) { 2403 /* 2404 * Heuristic: don't bother to do another RPC to further 2405 * fill up this block if there is not much room left. (< 50% 2406 * of the readdir RPC size). This wastes some buffer space 2407 * but can save up to 50% in RPC calls. 2408 */ 2409 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2410 bigenough = 0; 2411 break; 2412 } 2413 nfsstats.rpccnt[NFSPROC_READDIR]++; 2414 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2415 NFSX_READDIR(v3)); 2416 nfsm_fhtom(dnp, v3); 2417 #ifndef NFS_V2_ONLY 2418 if (v3) { 2419 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2420 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2421 txdr_swapcookie3(uiop->uio_offset, tl); 2422 } else { 2423 txdr_cookie3(uiop->uio_offset, tl); 2424 } 2425 tl += 2; 2426 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2427 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2428 } else 2429 #endif 2430 { 2431 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2432 *tl++ = txdr_unsigned(uiop->uio_offset); 2433 } 2434 *tl = txdr_unsigned(nmp->nm_readdirsize); 2435 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2436 nrpcs++; 2437 #ifndef NFS_V2_ONLY 2438 if (v3) { 2439 nfsm_postop_attr(vp, attrflag, 0); 2440 if (!error) { 2441 nfsm_dissect(tl, u_int32_t *, 2442 2 * NFSX_UNSIGNED); 2443 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2444 dnp->n_cookieverf.nfsuquad[1] = *tl; 2445 } else { 2446 m_freem(mrep); 2447 goto nfsmout; 2448 } 2449 } 2450 #endif 2451 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2452 more_dirs = fxdr_unsigned(int, *tl); 2453 2454 /* loop thru the dir entries, doctoring them to 4bsd form */ 2455 while (more_dirs && bigenough) { 2456 #ifndef NFS_V2_ONLY 2457 if (v3) { 2458 nfsm_dissect(tl, u_int32_t *, 2459 3 * NFSX_UNSIGNED); 2460 fileno = fxdr_hyper(tl); 2461 len = fxdr_unsigned(int, *(tl + 2)); 2462 } else 2463 #endif 2464 { 2465 nfsm_dissect(tl, u_int32_t *, 2466 2 * NFSX_UNSIGNED); 2467 fileno = fxdr_unsigned(u_quad_t, *tl++); 2468 len = fxdr_unsigned(int, *tl); 2469 } 2470 if (len <= 0 || len > NFS_MAXNAMLEN) { 2471 error = EBADRPC; 2472 m_freem(mrep); 2473 goto nfsmout; 2474 } 2475 /* for cookie stashing */ 2476 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2477 left = NFS_DIRFRAGSIZ - blksiz; 2478 if (reclen > left) { 2479 memset(uiop->uio_iov->iov_base, 0, left); 2480 dp->d_reclen += left; 2481 UIO_ADVANCE(uiop, left); 2482 blksiz = 0; 2483 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2484 } 2485 if (reclen > uiop->uio_resid) 2486 bigenough = 0; 2487 if (bigenough) { 2488 int tlen; 2489 2490 dp = (struct dirent *)uiop->uio_iov->iov_base; 2491 dp->d_fileno = fileno; 2492 dp->d_namlen = len; 2493 dp->d_reclen = reclen; 2494 dp->d_type = DT_UNKNOWN; 2495 blksiz += reclen; 2496 if (blksiz == NFS_DIRFRAGSIZ) 2497 blksiz = 0; 2498 UIO_ADVANCE(uiop, DIRHDSIZ); 2499 nfsm_mtouio(uiop, len); 2500 tlen = reclen - (DIRHDSIZ + len); 2501 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2502 UIO_ADVANCE(uiop, tlen); 2503 } else 2504 nfsm_adv(nfsm_rndup(len)); 2505 #ifndef NFS_V2_ONLY 2506 if (v3) { 2507 nfsm_dissect(tl, u_int32_t *, 2508 3 * NFSX_UNSIGNED); 2509 } else 2510 #endif 2511 { 2512 nfsm_dissect(tl, u_int32_t *, 2513 2 * NFSX_UNSIGNED); 2514 } 2515 if (bigenough) { 2516 #ifndef NFS_V2_ONLY 2517 if (v3) { 2518 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2519 uiop->uio_offset = 2520 fxdr_swapcookie3(tl); 2521 else 2522 uiop->uio_offset = 2523 fxdr_cookie3(tl); 2524 } 2525 else 2526 #endif 2527 { 2528 uiop->uio_offset = 2529 fxdr_unsigned(off_t, *tl); 2530 } 2531 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2532 } 2533 if (v3) 2534 tl += 2; 2535 else 2536 tl++; 2537 more_dirs = fxdr_unsigned(int, *tl); 2538 } 2539 /* 2540 * If at end of rpc data, get the eof boolean 2541 */ 2542 if (!more_dirs) { 2543 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2544 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2545 2546 /* 2547 * kludge: if we got no entries, treat it as EOF. 2548 * some server sometimes send a reply without any 2549 * entries or EOF. 2550 * although it might mean the server has very long name, 2551 * we can't handle such entries anyway. 2552 */ 2553 2554 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2555 more_dirs = 0; 2556 } 2557 m_freem(mrep); 2558 } 2559 /* 2560 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2561 * by increasing d_reclen for the last record. 2562 */ 2563 if (blksiz > 0) { 2564 left = NFS_DIRFRAGSIZ - blksiz; 2565 memset(uiop->uio_iov->iov_base, 0, left); 2566 dp->d_reclen += left; 2567 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2568 UIO_ADVANCE(uiop, left); 2569 } 2570 2571 /* 2572 * We are now either at the end of the directory or have filled the 2573 * block. 2574 */ 2575 if (bigenough) { 2576 dnp->n_direofoffset = uiop->uio_offset; 2577 dnp->n_flag |= NEOFVALID; 2578 } 2579 nfsmout: 2580 return (error); 2581 } 2582 2583 #ifndef NFS_V2_ONLY 2584 /* 2585 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2586 */ 2587 int 2588 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2589 { 2590 int len, left; 2591 struct dirent *dp = NULL; 2592 u_int32_t *tl; 2593 char *cp; 2594 int32_t t1, t2; 2595 struct vnode *newvp; 2596 char *bpos, *dpos, *cp2; 2597 struct mbuf *mreq, *mrep, *md, *mb; 2598 struct nameidata nami, *ndp = &nami; 2599 struct componentname *cnp = &ndp->ni_cnd; 2600 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2601 struct nfsnode *dnp = VTONFS(vp), *np; 2602 nfsfh_t *fhp; 2603 u_quad_t fileno; 2604 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2605 int attrflag, fhsize, nrpcs = 0, reclen; 2606 struct nfs_fattr fattr, *fp; 2607 2608 #ifdef DIAGNOSTIC 2609 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2610 panic("nfs readdirplusrpc bad uio"); 2611 #endif 2612 ndp->ni_dvp = vp; 2613 newvp = NULLVP; 2614 2615 /* 2616 * Loop around doing readdir rpc's of size nm_readdirsize 2617 * truncated to a multiple of NFS_DIRFRAGSIZ. 2618 * The stopping criteria is EOF or buffer full. 2619 */ 2620 while (more_dirs && bigenough) { 2621 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2622 bigenough = 0; 2623 break; 2624 } 2625 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2626 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2627 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2628 nfsm_fhtom(dnp, 1); 2629 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2630 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2631 txdr_swapcookie3(uiop->uio_offset, tl); 2632 } else { 2633 txdr_cookie3(uiop->uio_offset, tl); 2634 } 2635 tl += 2; 2636 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2637 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2638 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2639 *tl = txdr_unsigned(nmp->nm_rsize); 2640 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2641 nfsm_postop_attr(vp, attrflag, 0); 2642 if (error) { 2643 m_freem(mrep); 2644 goto nfsmout; 2645 } 2646 nrpcs++; 2647 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2648 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2649 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2650 more_dirs = fxdr_unsigned(int, *tl); 2651 2652 /* loop thru the dir entries, doctoring them to 4bsd form */ 2653 while (more_dirs && bigenough) { 2654 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2655 fileno = fxdr_hyper(tl); 2656 len = fxdr_unsigned(int, *(tl + 2)); 2657 if (len <= 0 || len > NFS_MAXNAMLEN) { 2658 error = EBADRPC; 2659 m_freem(mrep); 2660 goto nfsmout; 2661 } 2662 /* for cookie stashing */ 2663 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2664 left = NFS_DIRFRAGSIZ - blksiz; 2665 if (reclen > left) { 2666 /* 2667 * DIRFRAGSIZ is aligned, no need to align 2668 * again here. 2669 */ 2670 memset(uiop->uio_iov->iov_base, 0, left); 2671 dp->d_reclen += left; 2672 UIO_ADVANCE(uiop, left); 2673 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2674 blksiz = 0; 2675 } 2676 if (reclen > uiop->uio_resid) 2677 bigenough = 0; 2678 if (bigenough) { 2679 int tlen; 2680 2681 dp = (struct dirent *)uiop->uio_iov->iov_base; 2682 dp->d_fileno = fileno; 2683 dp->d_namlen = len; 2684 dp->d_reclen = reclen; 2685 dp->d_type = DT_UNKNOWN; 2686 blksiz += reclen; 2687 if (blksiz == NFS_DIRFRAGSIZ) 2688 blksiz = 0; 2689 UIO_ADVANCE(uiop, DIRHDSIZ); 2690 nfsm_mtouio(uiop, len); 2691 tlen = reclen - (DIRHDSIZ + len); 2692 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2693 UIO_ADVANCE(uiop, tlen); 2694 cnp->cn_nameptr = dp->d_name; 2695 cnp->cn_namelen = dp->d_namlen; 2696 } else 2697 nfsm_adv(nfsm_rndup(len)); 2698 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2699 if (bigenough) { 2700 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2701 uiop->uio_offset = 2702 fxdr_swapcookie3(tl); 2703 else 2704 uiop->uio_offset = 2705 fxdr_cookie3(tl); 2706 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2707 } 2708 tl += 2; 2709 2710 /* 2711 * Since the attributes are before the file handle 2712 * (sigh), we must skip over the attributes and then 2713 * come back and get them. 2714 */ 2715 attrflag = fxdr_unsigned(int, *tl); 2716 if (attrflag) { 2717 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2718 memcpy(&fattr, fp, NFSX_V3FATTR); 2719 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2720 doit = fxdr_unsigned(int, *tl); 2721 if (doit) { 2722 nfsm_getfh(fhp, fhsize, 1); 2723 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2724 vref(vp); 2725 newvp = vp; 2726 np = dnp; 2727 } else { 2728 error = nfs_nget1(vp->v_mount, fhp, 2729 fhsize, &np, LK_NOWAIT); 2730 if (!error) 2731 newvp = NFSTOV(np); 2732 } 2733 if (!error) { 2734 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2735 if (bigenough) { 2736 dp->d_type = 2737 IFTODT(VTTOIF(np->n_vattr->va_type)); 2738 ndp->ni_vp = newvp; 2739 nfs_cache_enter(ndp->ni_dvp, 2740 ndp->ni_vp, cnp); 2741 } 2742 } 2743 error = 0; 2744 } 2745 } else { 2746 /* Just skip over the file handle */ 2747 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2748 i = fxdr_unsigned(int, *tl); 2749 nfsm_adv(nfsm_rndup(i)); 2750 } 2751 if (newvp != NULLVP) { 2752 if (newvp == vp) 2753 vrele(newvp); 2754 else 2755 vput(newvp); 2756 newvp = NULLVP; 2757 } 2758 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2759 more_dirs = fxdr_unsigned(int, *tl); 2760 } 2761 /* 2762 * If at end of rpc data, get the eof boolean 2763 */ 2764 if (!more_dirs) { 2765 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2766 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2767 2768 /* 2769 * kludge: see a comment in nfs_readdirrpc. 2770 */ 2771 2772 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2773 more_dirs = 0; 2774 } 2775 m_freem(mrep); 2776 } 2777 /* 2778 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2779 * by increasing d_reclen for the last record. 2780 */ 2781 if (blksiz > 0) { 2782 left = NFS_DIRFRAGSIZ - blksiz; 2783 memset(uiop->uio_iov->iov_base, 0, left); 2784 dp->d_reclen += left; 2785 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2786 UIO_ADVANCE(uiop, left); 2787 } 2788 2789 /* 2790 * We are now either at the end of the directory or have filled the 2791 * block. 2792 */ 2793 if (bigenough) { 2794 dnp->n_direofoffset = uiop->uio_offset; 2795 dnp->n_flag |= NEOFVALID; 2796 } 2797 nfsmout: 2798 if (newvp != NULLVP) { 2799 if(newvp == vp) 2800 vrele(newvp); 2801 else 2802 vput(newvp); 2803 } 2804 return (error); 2805 } 2806 #endif 2807 2808 /* 2809 * Silly rename. To make the NFS filesystem that is stateless look a little 2810 * more like the "ufs" a remove of an active vnode is translated to a rename 2811 * to a funny looking filename that is removed by nfs_inactive on the 2812 * nfsnode. There is the potential for another process on a different client 2813 * to create the same funny name between the nfs_lookitup() fails and the 2814 * nfs_rename() completes, but... 2815 */ 2816 int 2817 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2818 { 2819 struct sillyrename *sp; 2820 struct nfsnode *np; 2821 int error; 2822 pid_t pid; 2823 2824 cache_purge(dvp); 2825 np = VTONFS(vp); 2826 #ifndef DIAGNOSTIC 2827 if (vp->v_type == VDIR) 2828 panic("nfs: sillyrename dir"); 2829 #endif 2830 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2831 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2832 sp->s_dvp = dvp; 2833 vref(dvp); 2834 2835 /* Fudge together a funny name */ 2836 pid = curlwp->l_proc->p_pid; 2837 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2838 sp->s_namlen = 12; 2839 sp->s_name[8] = hexdigits[pid & 0xf]; 2840 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2841 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2842 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2843 2844 /* Try lookitups until we get one that isn't there */ 2845 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2846 curlwp, (struct nfsnode **)0) == 0) { 2847 sp->s_name[4]++; 2848 if (sp->s_name[4] > 'z') { 2849 error = EINVAL; 2850 goto bad; 2851 } 2852 } 2853 if (dolink) { 2854 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2855 sp->s_cred, curlwp); 2856 /* 2857 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2858 */ 2859 if (error == ENOTSUP) { 2860 error = nfs_renameit(dvp, cnp, sp); 2861 } 2862 } else { 2863 error = nfs_renameit(dvp, cnp, sp); 2864 } 2865 if (error) 2866 goto bad; 2867 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2868 curlwp, &np); 2869 np->n_sillyrename = sp; 2870 return (0); 2871 bad: 2872 vrele(sp->s_dvp); 2873 kauth_cred_free(sp->s_cred); 2874 kmem_free(sp, sizeof(*sp)); 2875 return (error); 2876 } 2877 2878 /* 2879 * Look up a file name and optionally either update the file handle or 2880 * allocate an nfsnode, depending on the value of npp. 2881 * npp == NULL --> just do the lookup 2882 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2883 * handled too 2884 * *npp != NULL --> update the file handle in the vnode 2885 */ 2886 int 2887 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2888 { 2889 u_int32_t *tl; 2890 char *cp; 2891 int32_t t1, t2; 2892 struct vnode *newvp = (struct vnode *)0; 2893 struct nfsnode *np, *dnp = VTONFS(dvp); 2894 char *bpos, *dpos, *cp2; 2895 int error = 0, ofhlen, fhlen; 2896 #ifndef NFS_V2_ONLY 2897 int attrflag; 2898 #endif 2899 struct mbuf *mreq, *mrep, *md, *mb; 2900 nfsfh_t *ofhp, *nfhp; 2901 const int v3 = NFS_ISV3(dvp); 2902 2903 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2904 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2905 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2906 nfsm_fhtom(dnp, v3); 2907 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2908 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 2909 if (npp && !error) { 2910 nfsm_getfh(nfhp, fhlen, v3); 2911 if (*npp) { 2912 np = *npp; 2913 newvp = NFSTOV(np); 2914 ofhlen = np->n_fhsize; 2915 ofhp = kmem_alloc(ofhlen, KM_SLEEP); 2916 memcpy(ofhp, np->n_fhp, ofhlen); 2917 error = vcache_rekey_enter(newvp->v_mount, newvp, 2918 ofhp, ofhlen, nfhp, fhlen); 2919 if (error) { 2920 kmem_free(ofhp, ofhlen); 2921 m_freem(mrep); 2922 return error; 2923 } 2924 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2925 kmem_free(np->n_fhp, np->n_fhsize); 2926 np->n_fhp = &np->n_fh; 2927 } 2928 #if NFS_SMALLFH < NFSX_V3FHMAX 2929 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 2930 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 2931 #endif 2932 memcpy(np->n_fhp, nfhp, fhlen); 2933 np->n_fhsize = fhlen; 2934 vcache_rekey_exit(newvp->v_mount, newvp, 2935 ofhp, ofhlen, np->n_fhp, fhlen); 2936 kmem_free(ofhp, ofhlen); 2937 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2938 vref(dvp); 2939 newvp = dvp; 2940 np = dnp; 2941 } else { 2942 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2943 if (error) { 2944 m_freem(mrep); 2945 return (error); 2946 } 2947 newvp = NFSTOV(np); 2948 } 2949 #ifndef NFS_V2_ONLY 2950 if (v3) { 2951 nfsm_postop_attr(newvp, attrflag, 0); 2952 if (!attrflag && *npp == NULL) { 2953 m_freem(mrep); 2954 vput(newvp); 2955 return (ENOENT); 2956 } 2957 } else 2958 #endif 2959 nfsm_loadattr(newvp, (struct vattr *)0, 0); 2960 } 2961 nfsm_reqdone; 2962 if (npp && *npp == NULL) { 2963 if (error) { 2964 if (newvp) 2965 vput(newvp); 2966 } else 2967 *npp = np; 2968 } 2969 return (error); 2970 } 2971 2972 #ifndef NFS_V2_ONLY 2973 /* 2974 * Nfs Version 3 commit rpc 2975 */ 2976 int 2977 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 2978 { 2979 char *cp; 2980 u_int32_t *tl; 2981 int32_t t1, t2; 2982 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2983 char *bpos, *dpos, *cp2; 2984 int error = 0, wccflag = NFSV3_WCCRATTR; 2985 struct mbuf *mreq, *mrep, *md, *mb; 2986 struct nfsnode *np; 2987 2988 KASSERT(NFS_ISV3(vp)); 2989 2990 #ifdef NFS_DEBUG_COMMIT 2991 printf("commit %lu - %lu\n", (unsigned long)offset, 2992 (unsigned long)(offset + cnt)); 2993 #endif 2994 2995 mutex_enter(&nmp->nm_lock); 2996 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 2997 mutex_exit(&nmp->nm_lock); 2998 return (0); 2999 } 3000 mutex_exit(&nmp->nm_lock); 3001 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3002 np = VTONFS(vp); 3003 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3004 nfsm_fhtom(np, 1); 3005 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3006 txdr_hyper(offset, tl); 3007 tl += 2; 3008 *tl = txdr_unsigned(cnt); 3009 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3010 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3011 if (!error) { 3012 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3013 mutex_enter(&nmp->nm_lock); 3014 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3015 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3016 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3017 error = NFSERR_STALEWRITEVERF; 3018 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3019 } 3020 mutex_exit(&nmp->nm_lock); 3021 } 3022 nfsm_reqdone; 3023 return (error); 3024 } 3025 #endif 3026 3027 /* 3028 * Kludge City.. 3029 * - make nfs_bmap() essentially a no-op that does no translation 3030 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3031 * (Maybe I could use the process's page mapping, but I was concerned that 3032 * Kernel Write might not be enabled and also figured copyout() would do 3033 * a lot more work than memcpy() and also it currently happens in the 3034 * context of the swapper process (2). 3035 */ 3036 int 3037 nfs_bmap(void *v) 3038 { 3039 struct vop_bmap_args /* { 3040 struct vnode *a_vp; 3041 daddr_t a_bn; 3042 struct vnode **a_vpp; 3043 daddr_t *a_bnp; 3044 int *a_runp; 3045 } */ *ap = v; 3046 struct vnode *vp = ap->a_vp; 3047 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3048 3049 if (ap->a_vpp != NULL) 3050 *ap->a_vpp = vp; 3051 if (ap->a_bnp != NULL) 3052 *ap->a_bnp = ap->a_bn << bshift; 3053 if (ap->a_runp != NULL) 3054 *ap->a_runp = 1024 * 1024; /* XXX */ 3055 return (0); 3056 } 3057 3058 /* 3059 * Strategy routine. 3060 * For async requests when nfsiod(s) are running, queue the request by 3061 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3062 * request. 3063 */ 3064 int 3065 nfs_strategy(void *v) 3066 { 3067 struct vop_strategy_args *ap = v; 3068 struct buf *bp = ap->a_bp; 3069 int error = 0; 3070 3071 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3072 panic("nfs physio/async"); 3073 3074 /* 3075 * If the op is asynchronous and an i/o daemon is waiting 3076 * queue the request, wake it up and wait for completion 3077 * otherwise just do it ourselves. 3078 */ 3079 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3080 error = nfs_doio(bp); 3081 return (error); 3082 } 3083 3084 /* 3085 * fsync vnode op. Just call nfs_flush() with commit == 1. 3086 */ 3087 /* ARGSUSED */ 3088 int 3089 nfs_fsync(void *v) 3090 { 3091 struct vop_fsync_args /* { 3092 struct vnodeop_desc *a_desc; 3093 struct vnode * a_vp; 3094 kauth_cred_t a_cred; 3095 int a_flags; 3096 off_t offlo; 3097 off_t offhi; 3098 struct lwp * a_l; 3099 } */ *ap = v; 3100 3101 struct vnode *vp = ap->a_vp; 3102 3103 if (vp->v_type != VREG) 3104 return 0; 3105 3106 return (nfs_flush(vp, ap->a_cred, 3107 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3108 } 3109 3110 /* 3111 * Flush all the data associated with a vnode. 3112 */ 3113 int 3114 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3115 int commit) 3116 { 3117 struct nfsnode *np = VTONFS(vp); 3118 int error; 3119 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3120 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3121 3122 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 3123 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3124 if (np->n_flag & NWRITEERR) { 3125 error = np->n_error; 3126 np->n_flag &= ~NWRITEERR; 3127 } 3128 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3129 return (error); 3130 } 3131 3132 /* 3133 * Return POSIX pathconf information applicable to nfs. 3134 * 3135 * N.B. The NFS V2 protocol doesn't support this RPC. 3136 */ 3137 /* ARGSUSED */ 3138 int 3139 nfs_pathconf(void *v) 3140 { 3141 struct vop_pathconf_args /* { 3142 struct vnode *a_vp; 3143 int a_name; 3144 register_t *a_retval; 3145 } */ *ap = v; 3146 struct nfsv3_pathconf *pcp; 3147 struct vnode *vp = ap->a_vp; 3148 struct mbuf *mreq, *mrep, *md, *mb; 3149 int32_t t1, t2; 3150 u_int32_t *tl; 3151 char *bpos, *dpos, *cp, *cp2; 3152 int error = 0, attrflag; 3153 #ifndef NFS_V2_ONLY 3154 struct nfsmount *nmp; 3155 unsigned int l; 3156 u_int64_t maxsize; 3157 #endif 3158 const int v3 = NFS_ISV3(vp); 3159 struct nfsnode *np = VTONFS(vp); 3160 3161 switch (ap->a_name) { 3162 /* Names that can be resolved locally. */ 3163 case _PC_PIPE_BUF: 3164 *ap->a_retval = PIPE_BUF; 3165 break; 3166 case _PC_SYNC_IO: 3167 *ap->a_retval = 1; 3168 break; 3169 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3170 case _PC_LINK_MAX: 3171 case _PC_NAME_MAX: 3172 case _PC_CHOWN_RESTRICTED: 3173 case _PC_NO_TRUNC: 3174 if (!v3) { 3175 error = EINVAL; 3176 break; 3177 } 3178 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3179 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3180 nfsm_fhtom(np, 1); 3181 nfsm_request(np, NFSPROC_PATHCONF, 3182 curlwp, curlwp->l_cred); /* XXX */ 3183 nfsm_postop_attr(vp, attrflag, 0); 3184 if (!error) { 3185 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3186 NFSX_V3PATHCONF); 3187 switch (ap->a_name) { 3188 case _PC_LINK_MAX: 3189 *ap->a_retval = 3190 fxdr_unsigned(register_t, pcp->pc_linkmax); 3191 break; 3192 case _PC_NAME_MAX: 3193 *ap->a_retval = 3194 fxdr_unsigned(register_t, pcp->pc_namemax); 3195 break; 3196 case _PC_CHOWN_RESTRICTED: 3197 *ap->a_retval = 3198 (pcp->pc_chownrestricted == nfs_true); 3199 break; 3200 case _PC_NO_TRUNC: 3201 *ap->a_retval = 3202 (pcp->pc_notrunc == nfs_true); 3203 break; 3204 } 3205 } 3206 nfsm_reqdone; 3207 break; 3208 case _PC_FILESIZEBITS: 3209 #ifndef NFS_V2_ONLY 3210 if (v3) { 3211 nmp = VFSTONFS(vp->v_mount); 3212 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3213 if ((error = nfs_fsinfo(nmp, vp, 3214 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3215 break; 3216 for (l = 0, maxsize = nmp->nm_maxfilesize; 3217 (maxsize >> l) > 0; l++) 3218 ; 3219 *ap->a_retval = l + 1; 3220 } else 3221 #endif 3222 { 3223 *ap->a_retval = 32; /* NFS V2 limitation */ 3224 } 3225 break; 3226 default: 3227 error = genfs_pathconf(ap); 3228 break; 3229 } 3230 3231 return (error); 3232 } 3233 3234 /* 3235 * NFS advisory byte-level locks. 3236 */ 3237 int 3238 nfs_advlock(void *v) 3239 { 3240 struct vop_advlock_args /* { 3241 struct vnode *a_vp; 3242 void *a_id; 3243 int a_op; 3244 struct flock *a_fl; 3245 int a_flags; 3246 } */ *ap = v; 3247 struct nfsnode *np = VTONFS(ap->a_vp); 3248 3249 return lf_advlock(ap, &np->n_lockf, np->n_size); 3250 } 3251 3252 /* 3253 * Print out the contents of an nfsnode. 3254 */ 3255 int 3256 nfs_print(void *v) 3257 { 3258 struct vop_print_args /* { 3259 struct vnode *a_vp; 3260 } */ *ap = v; 3261 struct vnode *vp = ap->a_vp; 3262 struct nfsnode *np = VTONFS(vp); 3263 3264 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3265 (unsigned long long)np->n_vattr->va_fileid, 3266 (unsigned long long)np->n_vattr->va_fsid); 3267 if (vp->v_type == VFIFO) 3268 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3269 printf("\n"); 3270 return (0); 3271 } 3272 3273 /* 3274 * nfs unlock wrapper. 3275 */ 3276 int 3277 nfs_unlock(void *v) 3278 { 3279 struct vop_unlock_args /* { 3280 struct vnode *a_vp; 3281 int a_flags; 3282 } */ *ap = v; 3283 struct vnode *vp = ap->a_vp; 3284 3285 /* 3286 * VOP_UNLOCK can be called by nfs_loadattrcache 3287 * with v_data == 0. 3288 */ 3289 if (VTONFS(vp)) { 3290 nfs_delayedtruncate(vp); 3291 } 3292 3293 return genfs_unlock(v); 3294 } 3295 3296 /* 3297 * nfs special file access vnode op. 3298 * Essentially just get vattr and then imitate iaccess() since the device is 3299 * local to the client. 3300 */ 3301 int 3302 nfsspec_access(void *v) 3303 { 3304 struct vop_access_args /* { 3305 struct vnode *a_vp; 3306 accmode_t a_accmode; 3307 kauth_cred_t a_cred; 3308 struct lwp *a_l; 3309 } */ *ap = v; 3310 struct vattr va; 3311 struct vnode *vp = ap->a_vp; 3312 int error; 3313 3314 error = VOP_GETATTR(vp, &va, ap->a_cred); 3315 if (error) 3316 return (error); 3317 3318 /* 3319 * Disallow write attempts on filesystems mounted read-only; 3320 * unless the file is a socket, fifo, or a block or character 3321 * device resident on the filesystem. 3322 */ 3323 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3324 switch (vp->v_type) { 3325 case VREG: 3326 case VDIR: 3327 case VLNK: 3328 return (EROFS); 3329 default: 3330 break; 3331 } 3332 } 3333 3334 return kauth_authorize_vnode(ap->a_cred, KAUTH_ACCESS_ACTION( 3335 ap->a_accmode, va.va_type, va.va_mode), vp, NULL, genfs_can_access( 3336 vp, ap->a_cred, va.va_uid, va.va_gid, va.va_mode, NULL, 3337 ap->a_accmode)); 3338 } 3339 3340 /* 3341 * Read wrapper for special devices. 3342 */ 3343 int 3344 nfsspec_read(void *v) 3345 { 3346 struct vop_read_args /* { 3347 struct vnode *a_vp; 3348 struct uio *a_uio; 3349 int a_ioflag; 3350 kauth_cred_t a_cred; 3351 } */ *ap = v; 3352 struct nfsnode *np = VTONFS(ap->a_vp); 3353 3354 /* 3355 * Set access flag. 3356 */ 3357 np->n_flag |= NACC; 3358 getnanotime(&np->n_atim); 3359 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3360 } 3361 3362 /* 3363 * Write wrapper for special devices. 3364 */ 3365 int 3366 nfsspec_write(void *v) 3367 { 3368 struct vop_write_args /* { 3369 struct vnode *a_vp; 3370 struct uio *a_uio; 3371 int a_ioflag; 3372 kauth_cred_t a_cred; 3373 } */ *ap = v; 3374 struct nfsnode *np = VTONFS(ap->a_vp); 3375 3376 /* 3377 * Set update flag. 3378 */ 3379 np->n_flag |= NUPD; 3380 getnanotime(&np->n_mtim); 3381 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3382 } 3383 3384 /* 3385 * Close wrapper for special devices. 3386 * 3387 * Update the times on the nfsnode then do device close. 3388 */ 3389 int 3390 nfsspec_close(void *v) 3391 { 3392 struct vop_close_args /* { 3393 struct vnode *a_vp; 3394 int a_fflag; 3395 kauth_cred_t a_cred; 3396 struct lwp *a_l; 3397 } */ *ap = v; 3398 struct vnode *vp = ap->a_vp; 3399 struct nfsnode *np = VTONFS(vp); 3400 struct vattr vattr; 3401 3402 if (np->n_flag & (NACC | NUPD)) { 3403 np->n_flag |= NCHG; 3404 if (vrefcnt(vp) == 1 && 3405 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3406 vattr_null(&vattr); 3407 if (np->n_flag & NACC) 3408 vattr.va_atime = np->n_atim; 3409 if (np->n_flag & NUPD) 3410 vattr.va_mtime = np->n_mtim; 3411 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3412 } 3413 } 3414 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3415 } 3416 3417 /* 3418 * Read wrapper for fifos. 3419 */ 3420 int 3421 nfsfifo_read(void *v) 3422 { 3423 struct vop_read_args /* { 3424 struct vnode *a_vp; 3425 struct uio *a_uio; 3426 int a_ioflag; 3427 kauth_cred_t a_cred; 3428 } */ *ap = v; 3429 struct nfsnode *np = VTONFS(ap->a_vp); 3430 3431 /* 3432 * Set access flag. 3433 */ 3434 np->n_flag |= NACC; 3435 getnanotime(&np->n_atim); 3436 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3437 } 3438 3439 /* 3440 * Write wrapper for fifos. 3441 */ 3442 int 3443 nfsfifo_write(void *v) 3444 { 3445 struct vop_write_args /* { 3446 struct vnode *a_vp; 3447 struct uio *a_uio; 3448 int a_ioflag; 3449 kauth_cred_t a_cred; 3450 } */ *ap = v; 3451 struct nfsnode *np = VTONFS(ap->a_vp); 3452 3453 /* 3454 * Set update flag. 3455 */ 3456 np->n_flag |= NUPD; 3457 getnanotime(&np->n_mtim); 3458 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3459 } 3460 3461 /* 3462 * Close wrapper for fifos. 3463 * 3464 * Update the times on the nfsnode then do fifo close. 3465 */ 3466 int 3467 nfsfifo_close(void *v) 3468 { 3469 struct vop_close_args /* { 3470 struct vnode *a_vp; 3471 int a_fflag; 3472 kauth_cred_t a_cred; 3473 struct lwp *a_l; 3474 } */ *ap = v; 3475 struct vnode *vp = ap->a_vp; 3476 struct nfsnode *np = VTONFS(vp); 3477 struct vattr vattr; 3478 3479 if (np->n_flag & (NACC | NUPD)) { 3480 struct timespec ts; 3481 3482 getnanotime(&ts); 3483 if (np->n_flag & NACC) 3484 np->n_atim = ts; 3485 if (np->n_flag & NUPD) 3486 np->n_mtim = ts; 3487 np->n_flag |= NCHG; 3488 if (vrefcnt(vp) == 1 && 3489 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3490 vattr_null(&vattr); 3491 if (np->n_flag & NACC) 3492 vattr.va_atime = np->n_atim; 3493 if (np->n_flag & NUPD) 3494 vattr.va_mtime = np->n_mtim; 3495 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3496 } 3497 } 3498 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3499 } 3500
Indexes created Thu Sep 25 16:09:42 GMT 2025