nfs_vnops.c revision 1.271
1 /* $NetBSD: nfs_vnops.c,v 1.271 2009/03/14 14:46:11 dsl 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 1.271 2009/03/14 14:46:11 dsl Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/proc.h> 51 #include <sys/kernel.h> 52 #include <sys/systm.h> 53 #include <sys/resourcevar.h> 54 #include <sys/mount.h> 55 #include <sys/buf.h> 56 #include <sys/condvar.h> 57 #include <sys/disk.h> 58 #include <sys/malloc.h> 59 #include <sys/kmem.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/vnode.h> 64 #include <sys/dirent.h> 65 #include <sys/fcntl.h> 66 #include <sys/hash.h> 67 #include <sys/lockf.h> 68 #include <sys/stat.h> 69 #include <sys/unistd.h> 70 #include <sys/kauth.h> 71 72 #include <uvm/uvm_extern.h> 73 #include <uvm/uvm.h> 74 75 #include <miscfs/fifofs/fifo.h> 76 #include <miscfs/genfs/genfs.h> 77 #include <miscfs/genfs/genfs_node.h> 78 #include <miscfs/specfs/specdev.h> 79 80 #include <nfs/rpcv2.h> 81 #include <nfs/nfsproto.h> 82 #include <nfs/nfs.h> 83 #include <nfs/nfsnode.h> 84 #include <nfs/nfsmount.h> 85 #include <nfs/xdr_subs.h> 86 #include <nfs/nfsm_subs.h> 87 #include <nfs/nfs_var.h> 88 89 #include <net/if.h> 90 #include <netinet/in.h> 91 #include <netinet/in_var.h> 92 93 /* 94 * Global vfs data structures for nfs 95 */ 96 int (**nfsv2_vnodeop_p)(void *); 97 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 98 { &vop_default_desc, vn_default_error }, 99 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 100 { &vop_create_desc, nfs_create }, /* create */ 101 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 102 { &vop_open_desc, nfs_open }, /* open */ 103 { &vop_close_desc, nfs_close }, /* close */ 104 { &vop_access_desc, nfs_access }, /* access */ 105 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 106 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 107 { &vop_read_desc, nfs_read }, /* read */ 108 { &vop_write_desc, nfs_write }, /* write */ 109 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 110 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 111 { &vop_poll_desc, nfs_poll }, /* poll */ 112 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 113 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 114 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 115 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 116 { &vop_seek_desc, nfs_seek }, /* seek */ 117 { &vop_remove_desc, nfs_remove }, /* remove */ 118 { &vop_link_desc, nfs_link }, /* link */ 119 { &vop_rename_desc, nfs_rename }, /* rename */ 120 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 121 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 122 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 123 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 124 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 125 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 126 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 127 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 128 { &vop_lock_desc, nfs_lock }, /* lock */ 129 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 130 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 131 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 132 { &vop_print_desc, nfs_print }, /* print */ 133 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 134 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 135 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 136 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 137 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 138 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 139 { NULL, NULL } 140 }; 141 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 142 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 143 144 /* 145 * Special device vnode ops 146 */ 147 int (**spec_nfsv2nodeop_p)(void *); 148 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 149 { &vop_default_desc, vn_default_error }, 150 { &vop_lookup_desc, spec_lookup }, /* lookup */ 151 { &vop_create_desc, spec_create }, /* create */ 152 { &vop_mknod_desc, spec_mknod }, /* mknod */ 153 { &vop_open_desc, spec_open }, /* open */ 154 { &vop_close_desc, nfsspec_close }, /* close */ 155 { &vop_access_desc, nfsspec_access }, /* access */ 156 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 157 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 158 { &vop_read_desc, nfsspec_read }, /* read */ 159 { &vop_write_desc, nfsspec_write }, /* write */ 160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 162 { &vop_poll_desc, spec_poll }, /* poll */ 163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 164 { &vop_revoke_desc, spec_revoke }, /* revoke */ 165 { &vop_mmap_desc, spec_mmap }, /* mmap */ 166 { &vop_fsync_desc, spec_fsync }, /* fsync */ 167 { &vop_seek_desc, spec_seek }, /* seek */ 168 { &vop_remove_desc, spec_remove }, /* remove */ 169 { &vop_link_desc, spec_link }, /* link */ 170 { &vop_rename_desc, spec_rename }, /* rename */ 171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 173 { &vop_symlink_desc, spec_symlink }, /* symlink */ 174 { &vop_readdir_desc, spec_readdir }, /* readdir */ 175 { &vop_readlink_desc, spec_readlink }, /* readlink */ 176 { &vop_abortop_desc, spec_abortop }, /* abortop */ 177 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 179 { &vop_lock_desc, nfs_lock }, /* lock */ 180 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 181 { &vop_bmap_desc, spec_bmap }, /* bmap */ 182 { &vop_strategy_desc, spec_strategy }, /* strategy */ 183 { &vop_print_desc, nfs_print }, /* print */ 184 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, spec_advlock }, /* advlock */ 187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 188 { &vop_getpages_desc, spec_getpages }, /* getpages */ 189 { &vop_putpages_desc, spec_putpages }, /* putpages */ 190 { NULL, NULL } 191 }; 192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 194 195 int (**fifo_nfsv2nodeop_p)(void *); 196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 197 { &vop_default_desc, vn_default_error }, 198 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 199 { &vop_create_desc, fifo_create }, /* create */ 200 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 201 { &vop_open_desc, fifo_open }, /* open */ 202 { &vop_close_desc, nfsfifo_close }, /* close */ 203 { &vop_access_desc, nfsspec_access }, /* access */ 204 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 205 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 206 { &vop_read_desc, nfsfifo_read }, /* read */ 207 { &vop_write_desc, nfsfifo_write }, /* write */ 208 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 209 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 210 { &vop_poll_desc, fifo_poll }, /* poll */ 211 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 212 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 213 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 214 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 215 { &vop_seek_desc, fifo_seek }, /* seek */ 216 { &vop_remove_desc, fifo_remove }, /* remove */ 217 { &vop_link_desc, fifo_link }, /* link */ 218 { &vop_rename_desc, fifo_rename }, /* rename */ 219 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 220 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 221 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 222 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 223 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 224 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 225 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 226 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 227 { &vop_lock_desc, nfs_lock }, /* lock */ 228 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 229 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 230 { &vop_strategy_desc, genfs_badop }, /* strategy */ 231 { &vop_print_desc, nfs_print }, /* print */ 232 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 233 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 234 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 235 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 236 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 237 { NULL, NULL } 238 }; 239 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 240 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 241 242 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 243 size_t, kauth_cred_t, struct lwp *); 244 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 245 246 /* 247 * Global variables 248 */ 249 extern u_int32_t nfs_true, nfs_false; 250 extern u_int32_t nfs_xdrneg1; 251 extern const nfstype nfsv3_type[9]; 252 253 int nfs_numasync = 0; 254 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 255 #define UIO_ADVANCE(uio, siz) \ 256 (void)((uio)->uio_resid -= (siz), \ 257 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 258 (uio)->uio_iov->iov_len -= (siz)) 259 260 static void nfs_cache_enter(struct vnode *, struct vnode *, 261 struct componentname *); 262 263 static void 264 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 265 struct componentname *cnp) 266 { 267 struct nfsnode *dnp = VTONFS(dvp); 268 269 if (vp != NULL) { 270 struct nfsnode *np = VTONFS(vp); 271 272 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 273 } 274 275 if (!timespecisset(&dnp->n_nctime)) 276 dnp->n_nctime = dnp->n_vattr->va_mtime; 277 278 cache_enter(dvp, vp, cnp); 279 } 280 281 /* 282 * nfs null call from vfs. 283 */ 284 int 285 nfs_null(vp, cred, l) 286 struct vnode *vp; 287 kauth_cred_t cred; 288 struct lwp *l; 289 { 290 char *bpos, *dpos; 291 int error = 0; 292 struct mbuf *mreq, *mrep, *md, *mb; 293 struct nfsnode *np = VTONFS(vp); 294 295 nfsm_reqhead(np, NFSPROC_NULL, 0); 296 nfsm_request(np, NFSPROC_NULL, l, cred); 297 nfsm_reqdone; 298 return (error); 299 } 300 301 /* 302 * nfs access vnode op. 303 * For nfs version 2, just return ok. File accesses may fail later. 304 * For nfs version 3, use the access rpc to check accessibility. If file modes 305 * are changed on the server, accesses might still fail later. 306 */ 307 int 308 nfs_access(v) 309 void *v; 310 { 311 struct vop_access_args /* { 312 struct vnode *a_vp; 313 int a_mode; 314 kauth_cred_t a_cred; 315 } */ *ap = v; 316 struct vnode *vp = ap->a_vp; 317 #ifndef NFS_V2_ONLY 318 u_int32_t *tl; 319 char *cp; 320 int32_t t1, t2; 321 char *bpos, *dpos, *cp2; 322 int error = 0, attrflag; 323 struct mbuf *mreq, *mrep, *md, *mb; 324 u_int32_t mode, rmode; 325 const int v3 = NFS_ISV3(vp); 326 #endif 327 int cachevalid; 328 struct nfsnode *np = VTONFS(vp); 329 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 330 331 cachevalid = (np->n_accstamp != -1 && 332 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 333 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 334 335 /* 336 * Check access cache first. If this request has been made for this 337 * uid shortly before, use the cached result. 338 */ 339 if (cachevalid) { 340 if (!np->n_accerror) { 341 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 342 return np->n_accerror; 343 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 344 return np->n_accerror; 345 } 346 347 #ifndef NFS_V2_ONLY 348 /* 349 * For nfs v3, do an access rpc, otherwise you are stuck emulating 350 * ufs_access() locally using the vattr. This may not be correct, 351 * since the server may apply other access criteria such as 352 * client uid-->server uid mapping that we do not know about, but 353 * this is better than just returning anything that is lying about 354 * in the cache. 355 */ 356 if (v3) { 357 nfsstats.rpccnt[NFSPROC_ACCESS]++; 358 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 359 nfsm_fhtom(np, v3); 360 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 361 if (ap->a_mode & VREAD) 362 mode = NFSV3ACCESS_READ; 363 else 364 mode = 0; 365 if (vp->v_type != VDIR) { 366 if (ap->a_mode & VWRITE) 367 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 368 if (ap->a_mode & VEXEC) 369 mode |= NFSV3ACCESS_EXECUTE; 370 } else { 371 if (ap->a_mode & VWRITE) 372 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 373 NFSV3ACCESS_DELETE); 374 if (ap->a_mode & VEXEC) 375 mode |= NFSV3ACCESS_LOOKUP; 376 } 377 *tl = txdr_unsigned(mode); 378 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 379 nfsm_postop_attr(vp, attrflag, 0); 380 if (!error) { 381 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 382 rmode = fxdr_unsigned(u_int32_t, *tl); 383 /* 384 * The NFS V3 spec does not clarify whether or not 385 * the returned access bits can be a superset of 386 * the ones requested, so... 387 */ 388 if ((rmode & mode) != mode) 389 error = EACCES; 390 } 391 nfsm_reqdone; 392 } else 393 #endif 394 return (nfsspec_access(ap)); 395 #ifndef NFS_V2_ONLY 396 /* 397 * Disallow write attempts on filesystems mounted read-only; 398 * unless the file is a socket, fifo, or a block or character 399 * device resident on the filesystem. 400 */ 401 if (!error && (ap->a_mode & VWRITE) && 402 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 403 switch (vp->v_type) { 404 case VREG: 405 case VDIR: 406 case VLNK: 407 error = EROFS; 408 default: 409 break; 410 } 411 } 412 413 if (!error || error == EACCES) { 414 /* 415 * If we got the same result as for a previous, 416 * different request, OR it in. Don't update 417 * the timestamp in that case. 418 */ 419 if (cachevalid && np->n_accstamp != -1 && 420 error == np->n_accerror) { 421 if (!error) 422 np->n_accmode |= ap->a_mode; 423 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 424 np->n_accmode = ap->a_mode; 425 } else { 426 np->n_accstamp = time_uptime; 427 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 428 np->n_accmode = ap->a_mode; 429 np->n_accerror = error; 430 } 431 } 432 433 return (error); 434 #endif 435 } 436 437 /* 438 * nfs open vnode op 439 * Check to see if the type is ok 440 * and that deletion is not in progress. 441 * For paged in text files, you will need to flush the page cache 442 * if consistency is lost. 443 */ 444 /* ARGSUSED */ 445 int 446 nfs_open(v) 447 void *v; 448 { 449 struct vop_open_args /* { 450 struct vnode *a_vp; 451 int a_mode; 452 kauth_cred_t a_cred; 453 } */ *ap = v; 454 struct vnode *vp = ap->a_vp; 455 struct nfsnode *np = VTONFS(vp); 456 int error; 457 458 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 459 return (EACCES); 460 } 461 462 if (ap->a_mode & FREAD) { 463 if (np->n_rcred != NULL) 464 kauth_cred_free(np->n_rcred); 465 np->n_rcred = ap->a_cred; 466 kauth_cred_hold(np->n_rcred); 467 } 468 if (ap->a_mode & FWRITE) { 469 if (np->n_wcred != NULL) 470 kauth_cred_free(np->n_wcred); 471 np->n_wcred = ap->a_cred; 472 kauth_cred_hold(np->n_wcred); 473 } 474 475 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 476 if (error) 477 return error; 478 479 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 480 481 return (0); 482 } 483 484 /* 485 * nfs close vnode op 486 * What an NFS client should do upon close after writing is a debatable issue. 487 * Most NFS clients push delayed writes to the server upon close, basically for 488 * two reasons: 489 * 1 - So that any write errors may be reported back to the client process 490 * doing the close system call. By far the two most likely errors are 491 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 492 * 2 - To put a worst case upper bound on cache inconsistency between 493 * multiple clients for the file. 494 * There is also a consistency problem for Version 2 of the protocol w.r.t. 495 * not being able to tell if other clients are writing a file concurrently, 496 * since there is no way of knowing if the changed modify time in the reply 497 * is only due to the write for this client. 498 * (NFS Version 3 provides weak cache consistency data in the reply that 499 * should be sufficient to detect and handle this case.) 500 * 501 * The current code does the following: 502 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 503 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 504 * or commit them (this satisfies 1 and 2 except for the 505 * case where the server crashes after this close but 506 * before the commit RPC, which is felt to be "good 507 * enough". Changing the last argument to nfs_flush() to 508 * a 1 would force a commit operation, if it is felt a 509 * commit is necessary now. 510 */ 511 /* ARGSUSED */ 512 int 513 nfs_close(v) 514 void *v; 515 { 516 struct vop_close_args /* { 517 struct vnodeop_desc *a_desc; 518 struct vnode *a_vp; 519 int a_fflag; 520 kauth_cred_t a_cred; 521 } */ *ap = v; 522 struct vnode *vp = ap->a_vp; 523 struct nfsnode *np = VTONFS(vp); 524 int error = 0; 525 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 526 527 if (vp->v_type == VREG) { 528 if (np->n_flag & NMODIFIED) { 529 #ifndef NFS_V2_ONLY 530 if (NFS_ISV3(vp)) { 531 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 532 np->n_flag &= ~NMODIFIED; 533 } else 534 #endif 535 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 536 NFS_INVALIDATE_ATTRCACHE(np); 537 } 538 if (np->n_flag & NWRITEERR) { 539 np->n_flag &= ~NWRITEERR; 540 error = np->n_error; 541 } 542 } 543 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 544 return (error); 545 } 546 547 /* 548 * nfs getattr call from vfs. 549 */ 550 int 551 nfs_getattr(v) 552 void *v; 553 { 554 struct vop_getattr_args /* { 555 struct vnode *a_vp; 556 struct vattr *a_vap; 557 kauth_cred_t a_cred; 558 } */ *ap = v; 559 struct vnode *vp = ap->a_vp; 560 struct nfsnode *np = VTONFS(vp); 561 char *cp; 562 u_int32_t *tl; 563 int32_t t1, t2; 564 char *bpos, *dpos; 565 int error = 0; 566 struct mbuf *mreq, *mrep, *md, *mb; 567 const int v3 = NFS_ISV3(vp); 568 569 /* 570 * Update local times for special files. 571 */ 572 if (np->n_flag & (NACC | NUPD)) 573 np->n_flag |= NCHG; 574 575 /* 576 * if we have delayed truncation, do it now. 577 */ 578 nfs_delayedtruncate(vp); 579 580 /* 581 * First look in the cache. 582 */ 583 if (nfs_getattrcache(vp, ap->a_vap) == 0) 584 return (0); 585 nfsstats.rpccnt[NFSPROC_GETATTR]++; 586 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 587 nfsm_fhtom(np, v3); 588 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 589 if (!error) { 590 nfsm_loadattr(vp, ap->a_vap, 0); 591 if (vp->v_type == VDIR && 592 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 593 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 594 } 595 nfsm_reqdone; 596 return (error); 597 } 598 599 /* 600 * nfs setattr call. 601 */ 602 int 603 nfs_setattr(v) 604 void *v; 605 { 606 struct vop_setattr_args /* { 607 struct vnodeop_desc *a_desc; 608 struct vnode *a_vp; 609 struct vattr *a_vap; 610 kauth_cred_t a_cred; 611 } */ *ap = v; 612 struct vnode *vp = ap->a_vp; 613 struct nfsnode *np = VTONFS(vp); 614 struct vattr *vap = ap->a_vap; 615 int error = 0; 616 u_quad_t tsize = 0; 617 618 /* 619 * Setting of flags is not supported. 620 */ 621 if (vap->va_flags != VNOVAL) 622 return (EOPNOTSUPP); 623 624 /* 625 * Disallow write attempts if the filesystem is mounted read-only. 626 */ 627 if ((vap->va_uid != (uid_t)VNOVAL || 628 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 629 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 630 (vp->v_mount->mnt_flag & MNT_RDONLY)) 631 return (EROFS); 632 if (vap->va_size != VNOVAL) { 633 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 634 return EFBIG; 635 } 636 switch (vp->v_type) { 637 case VDIR: 638 return (EISDIR); 639 case VCHR: 640 case VBLK: 641 case VSOCK: 642 case VFIFO: 643 if (vap->va_mtime.tv_sec == VNOVAL && 644 vap->va_atime.tv_sec == VNOVAL && 645 vap->va_mode == (mode_t)VNOVAL && 646 vap->va_uid == (uid_t)VNOVAL && 647 vap->va_gid == (gid_t)VNOVAL) 648 return (0); 649 vap->va_size = VNOVAL; 650 break; 651 default: 652 /* 653 * Disallow write attempts if the filesystem is 654 * mounted read-only. 655 */ 656 if (vp->v_mount->mnt_flag & MNT_RDONLY) 657 return (EROFS); 658 genfs_node_wrlock(vp); 659 uvm_vnp_setsize(vp, vap->va_size); 660 tsize = np->n_size; 661 np->n_size = vap->va_size; 662 if (vap->va_size == 0) 663 error = nfs_vinvalbuf(vp, 0, 664 ap->a_cred, curlwp, 1); 665 else 666 error = nfs_vinvalbuf(vp, V_SAVE, 667 ap->a_cred, curlwp, 1); 668 if (error) { 669 uvm_vnp_setsize(vp, tsize); 670 genfs_node_unlock(vp); 671 return (error); 672 } 673 np->n_vattr->va_size = vap->va_size; 674 } 675 } else { 676 /* 677 * flush files before setattr because a later write of 678 * cached data might change timestamps or reset sugid bits 679 */ 680 if ((vap->va_mtime.tv_sec != VNOVAL || 681 vap->va_atime.tv_sec != VNOVAL || 682 vap->va_mode != VNOVAL) && 683 vp->v_type == VREG && 684 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 685 curlwp, 1)) == EINTR) 686 return (error); 687 } 688 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 689 if (vap->va_size != VNOVAL) { 690 if (error) { 691 np->n_size = np->n_vattr->va_size = tsize; 692 uvm_vnp_setsize(vp, np->n_size); 693 } 694 genfs_node_unlock(vp); 695 } 696 VN_KNOTE(vp, NOTE_ATTRIB); 697 return (error); 698 } 699 700 /* 701 * Do an nfs setattr rpc. 702 */ 703 int 704 nfs_setattrrpc(vp, vap, cred, l) 705 struct vnode *vp; 706 struct vattr *vap; 707 kauth_cred_t cred; 708 struct lwp *l; 709 { 710 struct nfsv2_sattr *sp; 711 char *cp; 712 int32_t t1, t2; 713 char *bpos, *dpos; 714 u_int32_t *tl; 715 int error = 0; 716 struct mbuf *mreq, *mrep, *md, *mb; 717 const int v3 = NFS_ISV3(vp); 718 struct nfsnode *np = VTONFS(vp); 719 #ifndef NFS_V2_ONLY 720 int wccflag = NFSV3_WCCRATTR; 721 char *cp2; 722 #endif 723 724 nfsstats.rpccnt[NFSPROC_SETATTR]++; 725 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 726 nfsm_fhtom(np, v3); 727 #ifndef NFS_V2_ONLY 728 if (v3) { 729 nfsm_v3attrbuild(vap, true); 730 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 731 *tl = nfs_false; 732 } else { 733 #endif 734 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 735 if (vap->va_mode == (mode_t)VNOVAL) 736 sp->sa_mode = nfs_xdrneg1; 737 else 738 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 739 if (vap->va_uid == (uid_t)VNOVAL) 740 sp->sa_uid = nfs_xdrneg1; 741 else 742 sp->sa_uid = txdr_unsigned(vap->va_uid); 743 if (vap->va_gid == (gid_t)VNOVAL) 744 sp->sa_gid = nfs_xdrneg1; 745 else 746 sp->sa_gid = txdr_unsigned(vap->va_gid); 747 sp->sa_size = txdr_unsigned(vap->va_size); 748 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 749 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 750 #ifndef NFS_V2_ONLY 751 } 752 #endif 753 nfsm_request(np, NFSPROC_SETATTR, l, cred); 754 #ifndef NFS_V2_ONLY 755 if (v3) { 756 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 757 } else 758 #endif 759 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 760 nfsm_reqdone; 761 return (error); 762 } 763 764 /* 765 * nfs lookup call, one step at a time... 766 * First look in cache 767 * If not found, unlock the directory nfsnode and do the rpc 768 * 769 * This code is full of lock/unlock statements and checks, because 770 * we continue after cache_lookup has finished (we need to check 771 * with the attr cache and do an rpc if it has timed out). This means 772 * that the locking effects of cache_lookup have to be taken into 773 * account. 774 */ 775 int 776 nfs_lookup(v) 777 void *v; 778 { 779 struct vop_lookup_args /* { 780 struct vnodeop_desc *a_desc; 781 struct vnode *a_dvp; 782 struct vnode **a_vpp; 783 struct componentname *a_cnp; 784 } */ *ap = v; 785 struct componentname *cnp = ap->a_cnp; 786 struct vnode *dvp = ap->a_dvp; 787 struct vnode **vpp = ap->a_vpp; 788 int flags; 789 struct vnode *newvp; 790 u_int32_t *tl; 791 char *cp; 792 int32_t t1, t2; 793 char *bpos, *dpos, *cp2; 794 struct mbuf *mreq, *mrep, *md, *mb; 795 long len; 796 nfsfh_t *fhp; 797 struct nfsnode *np; 798 int error = 0, attrflag, fhsize; 799 const int v3 = NFS_ISV3(dvp); 800 801 flags = cnp->cn_flags; 802 803 *vpp = NULLVP; 804 newvp = NULLVP; 805 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 806 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 807 return (EROFS); 808 if (dvp->v_type != VDIR) 809 return (ENOTDIR); 810 811 /* 812 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 813 */ 814 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 815 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 816 if (error) 817 return error; 818 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 819 return EISDIR; 820 VREF(dvp); 821 *vpp = dvp; 822 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 823 cnp->cn_flags |= SAVENAME; 824 return 0; 825 } 826 827 np = VTONFS(dvp); 828 829 /* 830 * Before performing an RPC, check the name cache to see if 831 * the directory/name pair we are looking for is known already. 832 * If the directory/name pair is found in the name cache, 833 * we have to ensure the directory has not changed from 834 * the time the cache entry has been created. If it has, 835 * the cache entry has to be ignored. 836 */ 837 error = cache_lookup_raw(dvp, vpp, cnp); 838 KASSERT(dvp != *vpp); 839 if (error >= 0) { 840 struct vattr vattr; 841 int err2; 842 843 if (error && error != ENOENT) { 844 *vpp = NULLVP; 845 return error; 846 } 847 848 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 849 if (err2 != 0) { 850 if (error == 0) 851 vrele(*vpp); 852 *vpp = NULLVP; 853 return err2; 854 } 855 856 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 857 || timespeccmp(&vattr.va_mtime, 858 &VTONFS(dvp)->n_nctime, !=)) { 859 if (error == 0) { 860 vrele(*vpp); 861 *vpp = NULLVP; 862 } 863 cache_purge1(dvp, NULL, PURGE_CHILDREN); 864 timespecclear(&np->n_nctime); 865 goto dorpc; 866 } 867 868 if (error == ENOENT) { 869 goto noentry; 870 } 871 872 newvp = *vpp; 873 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 874 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 875 nfsstats.lookupcache_hits++; 876 if ((flags & ISDOTDOT) != 0) { 877 VOP_UNLOCK(dvp, 0); 878 } 879 error = vn_lock(newvp, LK_EXCLUSIVE); 880 if ((flags & ISDOTDOT) != 0) { 881 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 882 } 883 if (error) { 884 /* newvp has been revoked. */ 885 vrele(newvp); 886 *vpp = NULL; 887 return error; 888 } 889 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 890 cnp->cn_flags |= SAVENAME; 891 KASSERT(newvp->v_type != VNON); 892 return (0); 893 } 894 cache_purge1(newvp, NULL, PURGE_PARENTS); 895 vrele(newvp); 896 *vpp = NULLVP; 897 } 898 dorpc: 899 #if 0 900 /* 901 * because nfsv3 has the same CREATE semantics as ours, 902 * we don't have to perform LOOKUPs beforehand. 903 * 904 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 905 * XXX although we have no way to know if O_EXCL is requested or not. 906 */ 907 908 if (v3 && cnp->cn_nameiop == CREATE && 909 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 910 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 911 cnp->cn_flags |= SAVENAME; 912 return (EJUSTRETURN); 913 } 914 #endif /* 0 */ 915 916 error = 0; 917 newvp = NULLVP; 918 nfsstats.lookupcache_misses++; 919 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 920 len = cnp->cn_namelen; 921 nfsm_reqhead(np, NFSPROC_LOOKUP, 922 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 923 nfsm_fhtom(np, v3); 924 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 925 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 926 if (error) { 927 nfsm_postop_attr(dvp, attrflag, 0); 928 m_freem(mrep); 929 goto nfsmout; 930 } 931 nfsm_getfh(fhp, fhsize, v3); 932 933 /* 934 * Handle RENAME case... 935 */ 936 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 937 if (NFS_CMPFH(np, fhp, fhsize)) { 938 m_freem(mrep); 939 return (EISDIR); 940 } 941 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 942 if (error) { 943 m_freem(mrep); 944 return error; 945 } 946 newvp = NFSTOV(np); 947 #ifndef NFS_V2_ONLY 948 if (v3) { 949 nfsm_postop_attr(newvp, attrflag, 0); 950 nfsm_postop_attr(dvp, attrflag, 0); 951 } else 952 #endif 953 nfsm_loadattr(newvp, (struct vattr *)0, 0); 954 *vpp = newvp; 955 m_freem(mrep); 956 cnp->cn_flags |= SAVENAME; 957 goto validate; 958 } 959 960 /* 961 * The postop attr handling is duplicated for each if case, 962 * because it should be done while dvp is locked (unlocking 963 * dvp is different for each case). 964 */ 965 966 if (NFS_CMPFH(np, fhp, fhsize)) { 967 /* 968 * "." lookup 969 */ 970 VREF(dvp); 971 newvp = dvp; 972 #ifndef NFS_V2_ONLY 973 if (v3) { 974 nfsm_postop_attr(newvp, attrflag, 0); 975 nfsm_postop_attr(dvp, attrflag, 0); 976 } else 977 #endif 978 nfsm_loadattr(newvp, (struct vattr *)0, 0); 979 } else if (flags & ISDOTDOT) { 980 /* 981 * ".." lookup 982 */ 983 VOP_UNLOCK(dvp, 0); 984 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 985 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 986 if (error) { 987 m_freem(mrep); 988 return error; 989 } 990 newvp = NFSTOV(np); 991 992 #ifndef NFS_V2_ONLY 993 if (v3) { 994 nfsm_postop_attr(newvp, attrflag, 0); 995 nfsm_postop_attr(dvp, attrflag, 0); 996 } else 997 #endif 998 nfsm_loadattr(newvp, (struct vattr *)0, 0); 999 } else { 1000 /* 1001 * Other lookups. 1002 */ 1003 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1004 if (error) { 1005 m_freem(mrep); 1006 return error; 1007 } 1008 newvp = NFSTOV(np); 1009 #ifndef NFS_V2_ONLY 1010 if (v3) { 1011 nfsm_postop_attr(newvp, attrflag, 0); 1012 nfsm_postop_attr(dvp, attrflag, 0); 1013 } else 1014 #endif 1015 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1016 } 1017 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1018 cnp->cn_flags |= SAVENAME; 1019 if ((cnp->cn_flags & MAKEENTRY) && 1020 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1021 nfs_cache_enter(dvp, newvp, cnp); 1022 } 1023 *vpp = newvp; 1024 nfsm_reqdone; 1025 if (error) { 1026 /* 1027 * We get here only because of errors returned by 1028 * the RPC. Otherwise we'll have returned above 1029 * (the nfsm_* macros will jump to nfsm_reqdone 1030 * on error). 1031 */ 1032 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1033 cnp->cn_nameiop != CREATE) { 1034 nfs_cache_enter(dvp, NULL, cnp); 1035 } 1036 if (newvp != NULLVP) { 1037 if (newvp == dvp) { 1038 vrele(newvp); 1039 } else { 1040 vput(newvp); 1041 } 1042 } 1043 noentry: 1044 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1045 (flags & ISLASTCN) && error == ENOENT) { 1046 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1047 error = EROFS; 1048 } else { 1049 error = EJUSTRETURN; 1050 cnp->cn_flags |= SAVENAME; 1051 } 1052 } 1053 *vpp = NULL; 1054 return error; 1055 } 1056 1057 validate: 1058 /* 1059 * make sure we have valid type and size. 1060 */ 1061 1062 newvp = *vpp; 1063 if (newvp->v_type == VNON) { 1064 struct vattr vattr; /* dummy */ 1065 1066 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1067 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1068 if (error) { 1069 vput(newvp); 1070 *vpp = NULL; 1071 } 1072 } 1073 1074 return error; 1075 } 1076 1077 /* 1078 * nfs read call. 1079 * Just call nfs_bioread() to do the work. 1080 */ 1081 int 1082 nfs_read(v) 1083 void *v; 1084 { 1085 struct vop_read_args /* { 1086 struct vnode *a_vp; 1087 struct uio *a_uio; 1088 int a_ioflag; 1089 kauth_cred_t a_cred; 1090 } */ *ap = v; 1091 struct vnode *vp = ap->a_vp; 1092 1093 if (vp->v_type != VREG) 1094 return EISDIR; 1095 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1096 } 1097 1098 /* 1099 * nfs readlink call 1100 */ 1101 int 1102 nfs_readlink(v) 1103 void *v; 1104 { 1105 struct vop_readlink_args /* { 1106 struct vnode *a_vp; 1107 struct uio *a_uio; 1108 kauth_cred_t a_cred; 1109 } */ *ap = v; 1110 struct vnode *vp = ap->a_vp; 1111 struct nfsnode *np = VTONFS(vp); 1112 1113 if (vp->v_type != VLNK) 1114 return (EPERM); 1115 1116 if (np->n_rcred != NULL) { 1117 kauth_cred_free(np->n_rcred); 1118 } 1119 np->n_rcred = ap->a_cred; 1120 kauth_cred_hold(np->n_rcred); 1121 1122 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1123 } 1124 1125 /* 1126 * Do a readlink rpc. 1127 * Called by nfs_doio() from below the buffer cache. 1128 */ 1129 int 1130 nfs_readlinkrpc(vp, uiop, cred) 1131 struct vnode *vp; 1132 struct uio *uiop; 1133 kauth_cred_t cred; 1134 { 1135 u_int32_t *tl; 1136 char *cp; 1137 int32_t t1, t2; 1138 char *bpos, *dpos, *cp2; 1139 int error = 0; 1140 uint32_t len; 1141 struct mbuf *mreq, *mrep, *md, *mb; 1142 const int v3 = NFS_ISV3(vp); 1143 struct nfsnode *np = VTONFS(vp); 1144 #ifndef NFS_V2_ONLY 1145 int attrflag; 1146 #endif 1147 1148 nfsstats.rpccnt[NFSPROC_READLINK]++; 1149 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1150 nfsm_fhtom(np, v3); 1151 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1152 #ifndef NFS_V2_ONLY 1153 if (v3) 1154 nfsm_postop_attr(vp, attrflag, 0); 1155 #endif 1156 if (!error) { 1157 #ifndef NFS_V2_ONLY 1158 if (v3) { 1159 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1160 len = fxdr_unsigned(uint32_t, *tl); 1161 if (len > MAXPATHLEN) { 1162 /* 1163 * this pathname is too long for us. 1164 */ 1165 m_freem(mrep); 1166 /* Solaris returns EINVAL. should we follow? */ 1167 error = ENAMETOOLONG; 1168 goto nfsmout; 1169 } 1170 } else 1171 #endif 1172 { 1173 nfsm_strsiz(len, NFS_MAXPATHLEN); 1174 } 1175 nfsm_mtouio(uiop, len); 1176 } 1177 nfsm_reqdone; 1178 return (error); 1179 } 1180 1181 /* 1182 * nfs read rpc call 1183 * Ditto above 1184 */ 1185 int 1186 nfs_readrpc(vp, uiop) 1187 struct vnode *vp; 1188 struct uio *uiop; 1189 { 1190 u_int32_t *tl; 1191 char *cp; 1192 int32_t t1, t2; 1193 char *bpos, *dpos, *cp2; 1194 struct mbuf *mreq, *mrep, *md, *mb; 1195 struct nfsmount *nmp; 1196 int error = 0, len, retlen, tsiz, eof, byte_count; 1197 const int v3 = NFS_ISV3(vp); 1198 struct nfsnode *np = VTONFS(vp); 1199 #ifndef NFS_V2_ONLY 1200 int attrflag; 1201 #endif 1202 1203 #ifndef nolint 1204 eof = 0; 1205 #endif 1206 nmp = VFSTONFS(vp->v_mount); 1207 tsiz = uiop->uio_resid; 1208 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1209 return (EFBIG); 1210 iostat_busy(nmp->nm_stats); 1211 byte_count = 0; /* count bytes actually transferred */ 1212 while (tsiz > 0) { 1213 nfsstats.rpccnt[NFSPROC_READ]++; 1214 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1215 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1216 nfsm_fhtom(np, v3); 1217 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1218 #ifndef NFS_V2_ONLY 1219 if (v3) { 1220 txdr_hyper(uiop->uio_offset, tl); 1221 *(tl + 2) = txdr_unsigned(len); 1222 } else 1223 #endif 1224 { 1225 *tl++ = txdr_unsigned(uiop->uio_offset); 1226 *tl++ = txdr_unsigned(len); 1227 *tl = 0; 1228 } 1229 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1230 #ifndef NFS_V2_ONLY 1231 if (v3) { 1232 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1233 if (error) { 1234 m_freem(mrep); 1235 goto nfsmout; 1236 } 1237 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1238 eof = fxdr_unsigned(int, *(tl + 1)); 1239 } else 1240 #endif 1241 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1242 nfsm_strsiz(retlen, nmp->nm_rsize); 1243 nfsm_mtouio(uiop, retlen); 1244 m_freem(mrep); 1245 tsiz -= retlen; 1246 byte_count += retlen; 1247 #ifndef NFS_V2_ONLY 1248 if (v3) { 1249 if (eof || retlen == 0) 1250 tsiz = 0; 1251 } else 1252 #endif 1253 if (retlen < len) 1254 tsiz = 0; 1255 } 1256 nfsmout: 1257 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1258 return (error); 1259 } 1260 1261 struct nfs_writerpc_context { 1262 kmutex_t nwc_lock; 1263 kcondvar_t nwc_cv; 1264 int nwc_mbufcount; 1265 }; 1266 1267 /* 1268 * free mbuf used to refer protected pages while write rpc call. 1269 * called at splvm. 1270 */ 1271 static void 1272 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1273 { 1274 struct nfs_writerpc_context *ctx = arg; 1275 1276 KASSERT(m != NULL); 1277 KASSERT(ctx != NULL); 1278 pool_cache_put(mb_cache, m); 1279 mutex_enter(&ctx->nwc_lock); 1280 if (--ctx->nwc_mbufcount == 0) { 1281 cv_signal(&ctx->nwc_cv); 1282 } 1283 mutex_exit(&ctx->nwc_lock); 1284 } 1285 1286 /* 1287 * nfs write call 1288 */ 1289 int 1290 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp) 1291 struct vnode *vp; 1292 struct uio *uiop; 1293 int *iomode; 1294 bool pageprotected; 1295 bool *stalewriteverfp; 1296 { 1297 u_int32_t *tl; 1298 char *cp; 1299 int32_t t1, t2; 1300 char *bpos, *dpos; 1301 struct mbuf *mreq, *mrep, *md, *mb; 1302 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1303 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1304 const int v3 = NFS_ISV3(vp); 1305 int committed = NFSV3WRITE_FILESYNC; 1306 struct nfsnode *np = VTONFS(vp); 1307 struct nfs_writerpc_context ctx; 1308 int byte_count; 1309 struct lwp *l = NULL; 1310 size_t origresid; 1311 #ifndef NFS_V2_ONLY 1312 char *cp2; 1313 int rlen, commit; 1314 #endif 1315 1316 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1317 cv_init(&ctx.nwc_cv, "nfsmblk"); 1318 ctx.nwc_mbufcount = 1; 1319 1320 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1321 panic("writerpc readonly vp %p", vp); 1322 } 1323 1324 #ifdef DIAGNOSTIC 1325 if (uiop->uio_iovcnt != 1) 1326 panic("nfs: writerpc iovcnt > 1"); 1327 #endif 1328 tsiz = uiop->uio_resid; 1329 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1330 return (EFBIG); 1331 if (pageprotected) { 1332 l = curlwp; 1333 uvm_lwp_hold(l); 1334 } 1335 retry: 1336 origresid = uiop->uio_resid; 1337 KASSERT(origresid == uiop->uio_iov->iov_len); 1338 iostat_busy(nmp->nm_stats); 1339 byte_count = 0; /* count of bytes actually written */ 1340 while (tsiz > 0) { 1341 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1342 uint32_t backup; 1343 bool stalewriteverf = false; 1344 1345 nfsstats.rpccnt[NFSPROC_WRITE]++; 1346 len = min(tsiz, nmp->nm_wsize); 1347 datalen = pageprotected ? 0 : nfsm_rndup(len); 1348 nfsm_reqhead(np, NFSPROC_WRITE, 1349 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1350 nfsm_fhtom(np, v3); 1351 #ifndef NFS_V2_ONLY 1352 if (v3) { 1353 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1354 txdr_hyper(uiop->uio_offset, tl); 1355 tl += 2; 1356 *tl++ = txdr_unsigned(len); 1357 *tl++ = txdr_unsigned(*iomode); 1358 *tl = txdr_unsigned(len); 1359 } else 1360 #endif 1361 { 1362 u_int32_t x; 1363 1364 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1365 /* Set both "begin" and "current" to non-garbage. */ 1366 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1367 *tl++ = x; /* "begin offset" */ 1368 *tl++ = x; /* "current offset" */ 1369 x = txdr_unsigned(len); 1370 *tl++ = x; /* total to this offset */ 1371 *tl = x; /* size of this write */ 1372 1373 } 1374 if (pageprotected) { 1375 /* 1376 * since we know pages can't be modified during i/o, 1377 * no need to copy them for us. 1378 */ 1379 struct mbuf *m; 1380 struct iovec *iovp = uiop->uio_iov; 1381 1382 m = m_get(M_WAIT, MT_DATA); 1383 MCLAIM(m, &nfs_mowner); 1384 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1385 nfs_writerpc_extfree, &ctx); 1386 m->m_flags |= M_EXT_ROMAP; 1387 m->m_len = len; 1388 mb->m_next = m; 1389 /* 1390 * no need to maintain mb and bpos here 1391 * because no one care them later. 1392 */ 1393 #if 0 1394 mb = m; 1395 bpos = mtod(void *, mb) + mb->m_len; 1396 #endif 1397 UIO_ADVANCE(uiop, len); 1398 uiop->uio_offset += len; 1399 mutex_enter(&ctx.nwc_lock); 1400 ctx.nwc_mbufcount++; 1401 mutex_exit(&ctx.nwc_lock); 1402 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1403 } else { 1404 nfsm_uiotom(uiop, len); 1405 } 1406 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1407 #ifndef NFS_V2_ONLY 1408 if (v3) { 1409 wccflag = NFSV3_WCCCHK; 1410 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1411 if (!error) { 1412 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1413 + NFSX_V3WRITEVERF); 1414 rlen = fxdr_unsigned(int, *tl++); 1415 if (rlen == 0) { 1416 error = NFSERR_IO; 1417 m_freem(mrep); 1418 break; 1419 } else if (rlen < len) { 1420 backup = len - rlen; 1421 UIO_ADVANCE(uiop, -backup); 1422 uiop->uio_offset -= backup; 1423 len = rlen; 1424 } 1425 commit = fxdr_unsigned(int, *tl++); 1426 1427 /* 1428 * Return the lowest committment level 1429 * obtained by any of the RPCs. 1430 */ 1431 if (committed == NFSV3WRITE_FILESYNC) 1432 committed = commit; 1433 else if (committed == NFSV3WRITE_DATASYNC && 1434 commit == NFSV3WRITE_UNSTABLE) 1435 committed = commit; 1436 mutex_enter(&nmp->nm_lock); 1437 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1438 memcpy(nmp->nm_writeverf, tl, 1439 NFSX_V3WRITEVERF); 1440 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1441 } else if ((nmp->nm_iflag & 1442 NFSMNT_STALEWRITEVERF) || 1443 memcmp(tl, nmp->nm_writeverf, 1444 NFSX_V3WRITEVERF)) { 1445 memcpy(nmp->nm_writeverf, tl, 1446 NFSX_V3WRITEVERF); 1447 /* 1448 * note NFSMNT_STALEWRITEVERF 1449 * if we're the first thread to 1450 * notice it. 1451 */ 1452 if ((nmp->nm_iflag & 1453 NFSMNT_STALEWRITEVERF) == 0) { 1454 stalewriteverf = true; 1455 nmp->nm_iflag |= 1456 NFSMNT_STALEWRITEVERF; 1457 } 1458 } 1459 mutex_exit(&nmp->nm_lock); 1460 } 1461 } else 1462 #endif 1463 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1464 if (wccflag) 1465 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1466 m_freem(mrep); 1467 if (error) 1468 break; 1469 tsiz -= len; 1470 byte_count += len; 1471 if (stalewriteverf) { 1472 *stalewriteverfp = true; 1473 stalewriteverf = false; 1474 if (committed == NFSV3WRITE_UNSTABLE && 1475 len != origresid) { 1476 /* 1477 * if our write requests weren't atomic but 1478 * unstable, datas in previous iterations 1479 * might have already been lost now. 1480 * then, we should resend them to nfsd. 1481 */ 1482 backup = origresid - tsiz; 1483 UIO_ADVANCE(uiop, -backup); 1484 uiop->uio_offset -= backup; 1485 tsiz = origresid; 1486 goto retry; 1487 } 1488 } 1489 } 1490 nfsmout: 1491 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1492 if (pageprotected) { 1493 /* 1494 * wait until mbufs go away. 1495 * retransmitted mbufs can survive longer than rpc requests 1496 * themselves. 1497 */ 1498 mutex_enter(&ctx.nwc_lock); 1499 ctx.nwc_mbufcount--; 1500 while (ctx.nwc_mbufcount > 0) { 1501 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1502 } 1503 mutex_exit(&ctx.nwc_lock); 1504 uvm_lwp_rele(l); 1505 } 1506 mutex_destroy(&ctx.nwc_lock); 1507 cv_destroy(&ctx.nwc_cv); 1508 *iomode = committed; 1509 if (error) 1510 uiop->uio_resid = tsiz; 1511 return (error); 1512 } 1513 1514 /* 1515 * nfs mknod rpc 1516 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1517 * mode set to specify the file type and the size field for rdev. 1518 */ 1519 int 1520 nfs_mknodrpc(dvp, vpp, cnp, vap) 1521 struct vnode *dvp; 1522 struct vnode **vpp; 1523 struct componentname *cnp; 1524 struct vattr *vap; 1525 { 1526 struct nfsv2_sattr *sp; 1527 u_int32_t *tl; 1528 char *cp; 1529 int32_t t1, t2; 1530 struct vnode *newvp = (struct vnode *)0; 1531 struct nfsnode *dnp, *np; 1532 char *cp2; 1533 char *bpos, *dpos; 1534 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1535 struct mbuf *mreq, *mrep, *md, *mb; 1536 u_int32_t rdev; 1537 const int v3 = NFS_ISV3(dvp); 1538 1539 if (vap->va_type == VCHR || vap->va_type == VBLK) 1540 rdev = txdr_unsigned(vap->va_rdev); 1541 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1542 rdev = nfs_xdrneg1; 1543 else { 1544 VOP_ABORTOP(dvp, cnp); 1545 vput(dvp); 1546 return (EOPNOTSUPP); 1547 } 1548 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1549 dnp = VTONFS(dvp); 1550 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1551 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1552 nfsm_fhtom(dnp, v3); 1553 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1554 #ifndef NFS_V2_ONLY 1555 if (v3) { 1556 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1557 *tl++ = vtonfsv3_type(vap->va_type); 1558 nfsm_v3attrbuild(vap, false); 1559 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1560 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1561 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1562 *tl = txdr_unsigned(minor(vap->va_rdev)); 1563 } 1564 } else 1565 #endif 1566 { 1567 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1568 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1569 sp->sa_uid = nfs_xdrneg1; 1570 sp->sa_gid = nfs_xdrneg1; 1571 sp->sa_size = rdev; 1572 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1573 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1574 } 1575 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1576 if (!error) { 1577 nfsm_mtofh(dvp, newvp, v3, gotvp); 1578 if (!gotvp) { 1579 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1580 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1581 if (!error) 1582 newvp = NFSTOV(np); 1583 } 1584 } 1585 #ifndef NFS_V2_ONLY 1586 if (v3) 1587 nfsm_wcc_data(dvp, wccflag, 0, !error); 1588 #endif 1589 nfsm_reqdone; 1590 if (error) { 1591 if (newvp) 1592 vput(newvp); 1593 } else { 1594 if (cnp->cn_flags & MAKEENTRY) 1595 nfs_cache_enter(dvp, newvp, cnp); 1596 *vpp = newvp; 1597 } 1598 PNBUF_PUT(cnp->cn_pnbuf); 1599 VTONFS(dvp)->n_flag |= NMODIFIED; 1600 if (!wccflag) 1601 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1602 vput(dvp); 1603 return (error); 1604 } 1605 1606 /* 1607 * nfs mknod vop 1608 * just call nfs_mknodrpc() to do the work. 1609 */ 1610 /* ARGSUSED */ 1611 int 1612 nfs_mknod(v) 1613 void *v; 1614 { 1615 struct vop_mknod_args /* { 1616 struct vnode *a_dvp; 1617 struct vnode **a_vpp; 1618 struct componentname *a_cnp; 1619 struct vattr *a_vap; 1620 } */ *ap = v; 1621 struct vnode *dvp = ap->a_dvp; 1622 struct componentname *cnp = ap->a_cnp; 1623 int error; 1624 1625 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1626 VN_KNOTE(dvp, NOTE_WRITE); 1627 if (error == 0 || error == EEXIST) 1628 cache_purge1(dvp, cnp, 0); 1629 return (error); 1630 } 1631 1632 /* 1633 * nfs file create call 1634 */ 1635 int 1636 nfs_create(v) 1637 void *v; 1638 { 1639 struct vop_create_args /* { 1640 struct vnode *a_dvp; 1641 struct vnode **a_vpp; 1642 struct componentname *a_cnp; 1643 struct vattr *a_vap; 1644 } */ *ap = v; 1645 struct vnode *dvp = ap->a_dvp; 1646 struct vattr *vap = ap->a_vap; 1647 struct componentname *cnp = ap->a_cnp; 1648 struct nfsv2_sattr *sp; 1649 u_int32_t *tl; 1650 char *cp; 1651 int32_t t1, t2; 1652 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1653 struct vnode *newvp = (struct vnode *)0; 1654 char *bpos, *dpos, *cp2; 1655 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1656 struct mbuf *mreq, *mrep, *md, *mb; 1657 const int v3 = NFS_ISV3(dvp); 1658 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1659 1660 /* 1661 * Oops, not for me.. 1662 */ 1663 if (vap->va_type == VSOCK) 1664 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1665 1666 KASSERT(vap->va_type == VREG); 1667 1668 #ifdef VA_EXCLUSIVE 1669 if (vap->va_vaflags & VA_EXCLUSIVE) { 1670 excl_mode = NFSV3CREATE_EXCLUSIVE; 1671 } 1672 #endif 1673 again: 1674 error = 0; 1675 nfsstats.rpccnt[NFSPROC_CREATE]++; 1676 dnp = VTONFS(dvp); 1677 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1678 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1679 nfsm_fhtom(dnp, v3); 1680 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1681 #ifndef NFS_V2_ONLY 1682 if (v3) { 1683 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1684 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1685 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1686 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1687 *tl++ = arc4random(); 1688 *tl = arc4random(); 1689 } else { 1690 *tl = txdr_unsigned(excl_mode); 1691 nfsm_v3attrbuild(vap, false); 1692 } 1693 } else 1694 #endif 1695 { 1696 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1697 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1698 sp->sa_uid = nfs_xdrneg1; 1699 sp->sa_gid = nfs_xdrneg1; 1700 sp->sa_size = 0; 1701 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1702 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1703 } 1704 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1705 if (!error) { 1706 nfsm_mtofh(dvp, newvp, v3, gotvp); 1707 if (!gotvp) { 1708 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1709 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1710 if (!error) 1711 newvp = NFSTOV(np); 1712 } 1713 } 1714 #ifndef NFS_V2_ONLY 1715 if (v3) 1716 nfsm_wcc_data(dvp, wccflag, 0, !error); 1717 #endif 1718 nfsm_reqdone; 1719 if (error) { 1720 /* 1721 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1722 */ 1723 if (v3 && error == ENOTSUP) { 1724 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1725 excl_mode = NFSV3CREATE_GUARDED; 1726 goto again; 1727 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1728 excl_mode = NFSV3CREATE_UNCHECKED; 1729 goto again; 1730 } 1731 } 1732 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1733 struct timespec ts; 1734 1735 getnanotime(&ts); 1736 1737 /* 1738 * make sure that we'll update timestamps as 1739 * most server implementations use them to store 1740 * the create verifier. 1741 * 1742 * XXX it's better to use TOSERVER always. 1743 */ 1744 1745 if (vap->va_atime.tv_sec == VNOVAL) 1746 vap->va_atime = ts; 1747 if (vap->va_mtime.tv_sec == VNOVAL) 1748 vap->va_mtime = ts; 1749 1750 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1751 } 1752 if (error == 0) { 1753 if (cnp->cn_flags & MAKEENTRY) 1754 nfs_cache_enter(dvp, newvp, cnp); 1755 else 1756 cache_purge1(dvp, cnp, 0); 1757 *ap->a_vpp = newvp; 1758 } else { 1759 if (newvp) 1760 vput(newvp); 1761 if (error == EEXIST) 1762 cache_purge1(dvp, cnp, 0); 1763 } 1764 PNBUF_PUT(cnp->cn_pnbuf); 1765 VTONFS(dvp)->n_flag |= NMODIFIED; 1766 if (!wccflag) 1767 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1768 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1769 vput(dvp); 1770 return (error); 1771 } 1772 1773 /* 1774 * nfs file remove call 1775 * To try and make nfs semantics closer to ufs semantics, a file that has 1776 * other processes using the vnode is renamed instead of removed and then 1777 * removed later on the last close. 1778 * - If v_usecount > 1 1779 * If a rename is not already in the works 1780 * call nfs_sillyrename() to set it up 1781 * else 1782 * do the remove rpc 1783 */ 1784 int 1785 nfs_remove(v) 1786 void *v; 1787 { 1788 struct vop_remove_args /* { 1789 struct vnodeop_desc *a_desc; 1790 struct vnode * a_dvp; 1791 struct vnode * a_vp; 1792 struct componentname * a_cnp; 1793 } */ *ap = v; 1794 struct vnode *vp = ap->a_vp; 1795 struct vnode *dvp = ap->a_dvp; 1796 struct componentname *cnp = ap->a_cnp; 1797 struct nfsnode *np = VTONFS(vp); 1798 int error = 0; 1799 struct vattr vattr; 1800 1801 #ifndef DIAGNOSTIC 1802 if ((cnp->cn_flags & HASBUF) == 0) 1803 panic("nfs_remove: no name"); 1804 if (vp->v_usecount < 1) 1805 panic("nfs_remove: bad v_usecount"); 1806 #endif 1807 if (vp->v_type == VDIR) 1808 error = EPERM; 1809 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1810 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1811 vattr.va_nlink > 1)) { 1812 /* 1813 * Purge the name cache so that the chance of a lookup for 1814 * the name succeeding while the remove is in progress is 1815 * minimized. Without node locking it can still happen, such 1816 * that an I/O op returns ESTALE, but since you get this if 1817 * another host removes the file.. 1818 */ 1819 cache_purge(vp); 1820 /* 1821 * throw away biocache buffers, mainly to avoid 1822 * unnecessary delayed writes later. 1823 */ 1824 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1825 /* Do the rpc */ 1826 if (error != EINTR) 1827 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1828 cnp->cn_namelen, cnp->cn_cred, curlwp); 1829 } else if (!np->n_sillyrename) 1830 error = nfs_sillyrename(dvp, vp, cnp, false); 1831 PNBUF_PUT(cnp->cn_pnbuf); 1832 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1833 vattr.va_nlink == 1) { 1834 np->n_flag |= NREMOVED; 1835 } 1836 NFS_INVALIDATE_ATTRCACHE(np); 1837 VN_KNOTE(vp, NOTE_DELETE); 1838 VN_KNOTE(dvp, NOTE_WRITE); 1839 if (dvp == vp) 1840 vrele(vp); 1841 else 1842 vput(vp); 1843 vput(dvp); 1844 return (error); 1845 } 1846 1847 /* 1848 * nfs file remove rpc called from nfs_inactive 1849 */ 1850 int 1851 nfs_removeit(sp) 1852 struct sillyrename *sp; 1853 { 1854 1855 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1856 (struct lwp *)0)); 1857 } 1858 1859 /* 1860 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1861 */ 1862 int 1863 nfs_removerpc(dvp, name, namelen, cred, l) 1864 struct vnode *dvp; 1865 const char *name; 1866 int namelen; 1867 kauth_cred_t cred; 1868 struct lwp *l; 1869 { 1870 u_int32_t *tl; 1871 char *cp; 1872 #ifndef NFS_V2_ONLY 1873 int32_t t1; 1874 char *cp2; 1875 #endif 1876 int32_t t2; 1877 char *bpos, *dpos; 1878 int error = 0, wccflag = NFSV3_WCCRATTR; 1879 struct mbuf *mreq, *mrep, *md, *mb; 1880 const int v3 = NFS_ISV3(dvp); 1881 int rexmit = 0; 1882 struct nfsnode *dnp = VTONFS(dvp); 1883 1884 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1885 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1886 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1887 nfsm_fhtom(dnp, v3); 1888 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1889 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1890 #ifndef NFS_V2_ONLY 1891 if (v3) 1892 nfsm_wcc_data(dvp, wccflag, 0, !error); 1893 #endif 1894 nfsm_reqdone; 1895 VTONFS(dvp)->n_flag |= NMODIFIED; 1896 if (!wccflag) 1897 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1898 /* 1899 * Kludge City: If the first reply to the remove rpc is lost.. 1900 * the reply to the retransmitted request will be ENOENT 1901 * since the file was in fact removed 1902 * Therefore, we cheat and return success. 1903 */ 1904 if (rexmit && error == ENOENT) 1905 error = 0; 1906 return (error); 1907 } 1908 1909 /* 1910 * nfs file rename call 1911 */ 1912 int 1913 nfs_rename(v) 1914 void *v; 1915 { 1916 struct vop_rename_args /* { 1917 struct vnode *a_fdvp; 1918 struct vnode *a_fvp; 1919 struct componentname *a_fcnp; 1920 struct vnode *a_tdvp; 1921 struct vnode *a_tvp; 1922 struct componentname *a_tcnp; 1923 } */ *ap = v; 1924 struct vnode *fvp = ap->a_fvp; 1925 struct vnode *tvp = ap->a_tvp; 1926 struct vnode *fdvp = ap->a_fdvp; 1927 struct vnode *tdvp = ap->a_tdvp; 1928 struct componentname *tcnp = ap->a_tcnp; 1929 struct componentname *fcnp = ap->a_fcnp; 1930 int error; 1931 1932 #ifndef DIAGNOSTIC 1933 if ((tcnp->cn_flags & HASBUF) == 0 || 1934 (fcnp->cn_flags & HASBUF) == 0) 1935 panic("nfs_rename: no name"); 1936 #endif 1937 /* Check for cross-device rename */ 1938 if ((fvp->v_mount != tdvp->v_mount) || 1939 (tvp && (fvp->v_mount != tvp->v_mount))) { 1940 error = EXDEV; 1941 goto out; 1942 } 1943 1944 /* 1945 * If the tvp exists and is in use, sillyrename it before doing the 1946 * rename of the new file over it. 1947 * 1948 * Have sillyrename use link instead of rename if possible, 1949 * so that we don't lose the file if the rename fails, and so 1950 * that there's no window when the "to" file doesn't exist. 1951 */ 1952 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1953 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1954 VN_KNOTE(tvp, NOTE_DELETE); 1955 vput(tvp); 1956 tvp = NULL; 1957 } 1958 1959 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1960 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1961 curlwp); 1962 1963 VN_KNOTE(fdvp, NOTE_WRITE); 1964 VN_KNOTE(tdvp, NOTE_WRITE); 1965 if (error == 0 || error == EEXIST) { 1966 if (fvp->v_type == VDIR) 1967 cache_purge(fvp); 1968 else 1969 cache_purge1(fdvp, fcnp, 0); 1970 if (tvp != NULL && tvp->v_type == VDIR) 1971 cache_purge(tvp); 1972 else 1973 cache_purge1(tdvp, tcnp, 0); 1974 } 1975 out: 1976 if (tdvp == tvp) 1977 vrele(tdvp); 1978 else 1979 vput(tdvp); 1980 if (tvp) 1981 vput(tvp); 1982 vrele(fdvp); 1983 vrele(fvp); 1984 return (error); 1985 } 1986 1987 /* 1988 * nfs file rename rpc called from nfs_remove() above 1989 */ 1990 int 1991 nfs_renameit(sdvp, scnp, sp) 1992 struct vnode *sdvp; 1993 struct componentname *scnp; 1994 struct sillyrename *sp; 1995 { 1996 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1997 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1998 } 1999 2000 /* 2001 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2002 */ 2003 int 2004 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l) 2005 struct vnode *fdvp; 2006 const char *fnameptr; 2007 int fnamelen; 2008 struct vnode *tdvp; 2009 const char *tnameptr; 2010 int tnamelen; 2011 kauth_cred_t cred; 2012 struct lwp *l; 2013 { 2014 u_int32_t *tl; 2015 char *cp; 2016 #ifndef NFS_V2_ONLY 2017 int32_t t1; 2018 char *cp2; 2019 #endif 2020 int32_t t2; 2021 char *bpos, *dpos; 2022 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 2023 struct mbuf *mreq, *mrep, *md, *mb; 2024 const int v3 = NFS_ISV3(fdvp); 2025 int rexmit = 0; 2026 struct nfsnode *fdnp = VTONFS(fdvp); 2027 2028 nfsstats.rpccnt[NFSPROC_RENAME]++; 2029 nfsm_reqhead(fdnp, NFSPROC_RENAME, 2030 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 2031 nfsm_rndup(tnamelen)); 2032 nfsm_fhtom(fdnp, v3); 2033 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 2034 nfsm_fhtom(VTONFS(tdvp), v3); 2035 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 2036 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 2037 #ifndef NFS_V2_ONLY 2038 if (v3) { 2039 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 2040 nfsm_wcc_data(tdvp, twccflag, 0, !error); 2041 } 2042 #endif 2043 nfsm_reqdone; 2044 VTONFS(fdvp)->n_flag |= NMODIFIED; 2045 VTONFS(tdvp)->n_flag |= NMODIFIED; 2046 if (!fwccflag) 2047 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 2048 if (!twccflag) 2049 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 2050 /* 2051 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2052 */ 2053 if (rexmit && error == ENOENT) 2054 error = 0; 2055 return (error); 2056 } 2057 2058 /* 2059 * NFS link RPC, called from nfs_link. 2060 * Assumes dvp and vp locked, and leaves them that way. 2061 */ 2062 2063 static int 2064 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2065 size_t namelen, kauth_cred_t cred, struct lwp *l) 2066 { 2067 u_int32_t *tl; 2068 char *cp; 2069 #ifndef NFS_V2_ONLY 2070 int32_t t1; 2071 char *cp2; 2072 #endif 2073 int32_t t2; 2074 char *bpos, *dpos; 2075 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2076 struct mbuf *mreq, *mrep, *md, *mb; 2077 const int v3 = NFS_ISV3(dvp); 2078 int rexmit = 0; 2079 struct nfsnode *np = VTONFS(vp); 2080 2081 nfsstats.rpccnt[NFSPROC_LINK]++; 2082 nfsm_reqhead(np, NFSPROC_LINK, 2083 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2084 nfsm_fhtom(np, v3); 2085 nfsm_fhtom(VTONFS(dvp), v3); 2086 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2087 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2088 #ifndef NFS_V2_ONLY 2089 if (v3) { 2090 nfsm_postop_attr(vp, attrflag, 0); 2091 nfsm_wcc_data(dvp, wccflag, 0, !error); 2092 } 2093 #endif 2094 nfsm_reqdone; 2095 2096 VTONFS(dvp)->n_flag |= NMODIFIED; 2097 if (!attrflag) 2098 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2099 if (!wccflag) 2100 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2101 2102 /* 2103 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2104 */ 2105 if (rexmit && error == EEXIST) 2106 error = 0; 2107 2108 return error; 2109 } 2110 2111 /* 2112 * nfs hard link create call 2113 */ 2114 int 2115 nfs_link(v) 2116 void *v; 2117 { 2118 struct vop_link_args /* { 2119 struct vnode *a_dvp; 2120 struct vnode *a_vp; 2121 struct componentname *a_cnp; 2122 } */ *ap = v; 2123 struct vnode *vp = ap->a_vp; 2124 struct vnode *dvp = ap->a_dvp; 2125 struct componentname *cnp = ap->a_cnp; 2126 int error = 0; 2127 2128 if (dvp->v_mount != vp->v_mount) { 2129 VOP_ABORTOP(dvp, cnp); 2130 vput(dvp); 2131 return (EXDEV); 2132 } 2133 if (dvp != vp) { 2134 error = vn_lock(vp, LK_EXCLUSIVE); 2135 if (error != 0) { 2136 VOP_ABORTOP(dvp, cnp); 2137 vput(dvp); 2138 return error; 2139 } 2140 } 2141 2142 /* 2143 * Push all writes to the server, so that the attribute cache 2144 * doesn't get "out of sync" with the server. 2145 * XXX There should be a better way! 2146 */ 2147 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2148 2149 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2150 cnp->cn_cred, curlwp); 2151 2152 if (error == 0) 2153 cache_purge1(dvp, cnp, 0); 2154 PNBUF_PUT(cnp->cn_pnbuf); 2155 if (dvp != vp) 2156 VOP_UNLOCK(vp, 0); 2157 VN_KNOTE(vp, NOTE_LINK); 2158 VN_KNOTE(dvp, NOTE_WRITE); 2159 vput(dvp); 2160 return (error); 2161 } 2162 2163 /* 2164 * nfs symbolic link create call 2165 */ 2166 int 2167 nfs_symlink(v) 2168 void *v; 2169 { 2170 struct vop_symlink_args /* { 2171 struct vnode *a_dvp; 2172 struct vnode **a_vpp; 2173 struct componentname *a_cnp; 2174 struct vattr *a_vap; 2175 char *a_target; 2176 } */ *ap = v; 2177 struct vnode *dvp = ap->a_dvp; 2178 struct vattr *vap = ap->a_vap; 2179 struct componentname *cnp = ap->a_cnp; 2180 struct nfsv2_sattr *sp; 2181 u_int32_t *tl; 2182 char *cp; 2183 int32_t t1, t2; 2184 char *bpos, *dpos, *cp2; 2185 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2186 struct mbuf *mreq, *mrep, *md, *mb; 2187 struct vnode *newvp = (struct vnode *)0; 2188 const int v3 = NFS_ISV3(dvp); 2189 int rexmit = 0; 2190 struct nfsnode *dnp = VTONFS(dvp); 2191 2192 *ap->a_vpp = NULL; 2193 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2194 slen = strlen(ap->a_target); 2195 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2196 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2197 nfsm_fhtom(dnp, v3); 2198 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2199 #ifndef NFS_V2_ONlY 2200 if (v3) 2201 nfsm_v3attrbuild(vap, false); 2202 #endif 2203 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2204 #ifndef NFS_V2_ONlY 2205 if (!v3) { 2206 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2207 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2208 sp->sa_uid = nfs_xdrneg1; 2209 sp->sa_gid = nfs_xdrneg1; 2210 sp->sa_size = nfs_xdrneg1; 2211 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2212 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2213 } 2214 #endif 2215 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2216 &rexmit); 2217 #ifndef NFS_V2_ONlY 2218 if (v3) { 2219 if (!error) 2220 nfsm_mtofh(dvp, newvp, v3, gotvp); 2221 nfsm_wcc_data(dvp, wccflag, 0, !error); 2222 } 2223 #endif 2224 nfsm_reqdone; 2225 /* 2226 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2227 */ 2228 if (rexmit && error == EEXIST) 2229 error = 0; 2230 if (error == 0 || error == EEXIST) 2231 cache_purge1(dvp, cnp, 0); 2232 if (error == 0 && newvp == NULL) { 2233 struct nfsnode *np = NULL; 2234 2235 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2236 cnp->cn_cred, curlwp, &np); 2237 if (error == 0) 2238 newvp = NFSTOV(np); 2239 } 2240 if (error) { 2241 if (newvp != NULL) 2242 vput(newvp); 2243 } else { 2244 *ap->a_vpp = newvp; 2245 } 2246 PNBUF_PUT(cnp->cn_pnbuf); 2247 VTONFS(dvp)->n_flag |= NMODIFIED; 2248 if (!wccflag) 2249 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2250 VN_KNOTE(dvp, NOTE_WRITE); 2251 vput(dvp); 2252 return (error); 2253 } 2254 2255 /* 2256 * nfs make dir call 2257 */ 2258 int 2259 nfs_mkdir(v) 2260 void *v; 2261 { 2262 struct vop_mkdir_args /* { 2263 struct vnode *a_dvp; 2264 struct vnode **a_vpp; 2265 struct componentname *a_cnp; 2266 struct vattr *a_vap; 2267 } */ *ap = v; 2268 struct vnode *dvp = ap->a_dvp; 2269 struct vattr *vap = ap->a_vap; 2270 struct componentname *cnp = ap->a_cnp; 2271 struct nfsv2_sattr *sp; 2272 u_int32_t *tl; 2273 char *cp; 2274 int32_t t1, t2; 2275 int len; 2276 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2277 struct vnode *newvp = (struct vnode *)0; 2278 char *bpos, *dpos, *cp2; 2279 int error = 0, wccflag = NFSV3_WCCRATTR; 2280 int gotvp = 0; 2281 int rexmit = 0; 2282 struct mbuf *mreq, *mrep, *md, *mb; 2283 const int v3 = NFS_ISV3(dvp); 2284 2285 len = cnp->cn_namelen; 2286 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2287 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2288 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2289 nfsm_fhtom(dnp, v3); 2290 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2291 #ifndef NFS_V2_ONLY 2292 if (v3) { 2293 nfsm_v3attrbuild(vap, false); 2294 } else 2295 #endif 2296 { 2297 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2298 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2299 sp->sa_uid = nfs_xdrneg1; 2300 sp->sa_gid = nfs_xdrneg1; 2301 sp->sa_size = nfs_xdrneg1; 2302 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2303 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2304 } 2305 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2306 if (!error) 2307 nfsm_mtofh(dvp, newvp, v3, gotvp); 2308 if (v3) 2309 nfsm_wcc_data(dvp, wccflag, 0, !error); 2310 nfsm_reqdone; 2311 VTONFS(dvp)->n_flag |= NMODIFIED; 2312 if (!wccflag) 2313 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2314 /* 2315 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2316 * if we can succeed in looking up the directory. 2317 */ 2318 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2319 if (newvp) { 2320 vput(newvp); 2321 newvp = (struct vnode *)0; 2322 } 2323 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2324 curlwp, &np); 2325 if (!error) { 2326 newvp = NFSTOV(np); 2327 if (newvp->v_type != VDIR || newvp == dvp) 2328 error = EEXIST; 2329 } 2330 } 2331 if (error) { 2332 if (newvp) { 2333 if (dvp != newvp) 2334 vput(newvp); 2335 else 2336 vrele(newvp); 2337 } 2338 } else { 2339 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2340 if (cnp->cn_flags & MAKEENTRY) 2341 nfs_cache_enter(dvp, newvp, cnp); 2342 *ap->a_vpp = newvp; 2343 } 2344 PNBUF_PUT(cnp->cn_pnbuf); 2345 vput(dvp); 2346 return (error); 2347 } 2348 2349 /* 2350 * nfs remove directory call 2351 */ 2352 int 2353 nfs_rmdir(v) 2354 void *v; 2355 { 2356 struct vop_rmdir_args /* { 2357 struct vnode *a_dvp; 2358 struct vnode *a_vp; 2359 struct componentname *a_cnp; 2360 } */ *ap = v; 2361 struct vnode *vp = ap->a_vp; 2362 struct vnode *dvp = ap->a_dvp; 2363 struct componentname *cnp = ap->a_cnp; 2364 u_int32_t *tl; 2365 char *cp; 2366 #ifndef NFS_V2_ONLY 2367 int32_t t1; 2368 char *cp2; 2369 #endif 2370 int32_t t2; 2371 char *bpos, *dpos; 2372 int error = 0, wccflag = NFSV3_WCCRATTR; 2373 int rexmit = 0; 2374 struct mbuf *mreq, *mrep, *md, *mb; 2375 const int v3 = NFS_ISV3(dvp); 2376 struct nfsnode *dnp; 2377 2378 if (dvp == vp) { 2379 vrele(dvp); 2380 vput(dvp); 2381 PNBUF_PUT(cnp->cn_pnbuf); 2382 return (EINVAL); 2383 } 2384 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2385 dnp = VTONFS(dvp); 2386 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2387 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2388 nfsm_fhtom(dnp, v3); 2389 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2390 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2391 #ifndef NFS_V2_ONLY 2392 if (v3) 2393 nfsm_wcc_data(dvp, wccflag, 0, !error); 2394 #endif 2395 nfsm_reqdone; 2396 PNBUF_PUT(cnp->cn_pnbuf); 2397 VTONFS(dvp)->n_flag |= NMODIFIED; 2398 if (!wccflag) 2399 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2400 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2401 VN_KNOTE(vp, NOTE_DELETE); 2402 cache_purge(vp); 2403 vput(vp); 2404 vput(dvp); 2405 /* 2406 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2407 */ 2408 if (rexmit && error == ENOENT) 2409 error = 0; 2410 return (error); 2411 } 2412 2413 /* 2414 * nfs readdir call 2415 */ 2416 int 2417 nfs_readdir(v) 2418 void *v; 2419 { 2420 struct vop_readdir_args /* { 2421 struct vnode *a_vp; 2422 struct uio *a_uio; 2423 kauth_cred_t a_cred; 2424 int *a_eofflag; 2425 off_t **a_cookies; 2426 int *a_ncookies; 2427 } */ *ap = v; 2428 struct vnode *vp = ap->a_vp; 2429 struct uio *uio = ap->a_uio; 2430 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2431 char *base = uio->uio_iov->iov_base; 2432 int tresid, error; 2433 size_t count, lost; 2434 struct dirent *dp; 2435 off_t *cookies = NULL; 2436 int ncookies = 0, nc; 2437 2438 if (vp->v_type != VDIR) 2439 return (EPERM); 2440 2441 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2442 count = uio->uio_resid - lost; 2443 if (count <= 0) 2444 return (EINVAL); 2445 2446 /* 2447 * Call nfs_bioread() to do the real work. 2448 */ 2449 tresid = uio->uio_resid = count; 2450 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2451 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2452 2453 if (!error && ap->a_cookies) { 2454 ncookies = count / 16; 2455 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2456 *ap->a_cookies = cookies; 2457 } 2458 2459 if (!error && uio->uio_resid == tresid) { 2460 uio->uio_resid += lost; 2461 nfsstats.direofcache_misses++; 2462 if (ap->a_cookies) 2463 *ap->a_ncookies = 0; 2464 *ap->a_eofflag = 1; 2465 return (0); 2466 } 2467 2468 if (!error && ap->a_cookies) { 2469 /* 2470 * Only the NFS server and emulations use cookies, and they 2471 * load the directory block into system space, so we can 2472 * just look at it directly. 2473 */ 2474 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2475 uio->uio_iovcnt != 1) 2476 panic("nfs_readdir: lost in space"); 2477 for (nc = 0; ncookies-- && 2478 base < (char *)uio->uio_iov->iov_base; nc++){ 2479 dp = (struct dirent *) base; 2480 if (dp->d_reclen == 0) 2481 break; 2482 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2483 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2484 else 2485 *(cookies++) = NFS_GETCOOKIE(dp); 2486 base += dp->d_reclen; 2487 } 2488 uio->uio_resid += 2489 ((char *)uio->uio_iov->iov_base - base); 2490 uio->uio_iov->iov_len += 2491 ((char *)uio->uio_iov->iov_base - base); 2492 uio->uio_iov->iov_base = base; 2493 *ap->a_ncookies = nc; 2494 } 2495 2496 uio->uio_resid += lost; 2497 *ap->a_eofflag = 0; 2498 return (error); 2499 } 2500 2501 /* 2502 * Readdir rpc call. 2503 * Called from below the buffer cache by nfs_doio(). 2504 */ 2505 int 2506 nfs_readdirrpc(vp, uiop, cred) 2507 struct vnode *vp; 2508 struct uio *uiop; 2509 kauth_cred_t cred; 2510 { 2511 int len, left; 2512 struct dirent *dp = NULL; 2513 u_int32_t *tl; 2514 char *cp; 2515 int32_t t1, t2; 2516 char *bpos, *dpos, *cp2; 2517 struct mbuf *mreq, *mrep, *md, *mb; 2518 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2519 struct nfsnode *dnp = VTONFS(vp); 2520 u_quad_t fileno; 2521 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2522 #ifndef NFS_V2_ONLY 2523 int attrflag; 2524 #endif 2525 int nrpcs = 0, reclen; 2526 const int v3 = NFS_ISV3(vp); 2527 2528 #ifdef DIAGNOSTIC 2529 /* 2530 * Should be called from buffer cache, so only amount of 2531 * NFS_DIRBLKSIZ will be requested. 2532 */ 2533 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2534 panic("nfs readdirrpc bad uio"); 2535 #endif 2536 2537 /* 2538 * Loop around doing readdir rpc's of size nm_readdirsize 2539 * truncated to a multiple of NFS_DIRFRAGSIZ. 2540 * The stopping criteria is EOF or buffer full. 2541 */ 2542 while (more_dirs && bigenough) { 2543 /* 2544 * Heuristic: don't bother to do another RPC to further 2545 * fill up this block if there is not much room left. (< 50% 2546 * of the readdir RPC size). This wastes some buffer space 2547 * but can save up to 50% in RPC calls. 2548 */ 2549 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2550 bigenough = 0; 2551 break; 2552 } 2553 nfsstats.rpccnt[NFSPROC_READDIR]++; 2554 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2555 NFSX_READDIR(v3)); 2556 nfsm_fhtom(dnp, v3); 2557 #ifndef NFS_V2_ONLY 2558 if (v3) { 2559 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2560 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2561 txdr_swapcookie3(uiop->uio_offset, tl); 2562 } else { 2563 txdr_cookie3(uiop->uio_offset, tl); 2564 } 2565 tl += 2; 2566 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2567 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2568 } else 2569 #endif 2570 { 2571 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2572 *tl++ = txdr_unsigned(uiop->uio_offset); 2573 } 2574 *tl = txdr_unsigned(nmp->nm_readdirsize); 2575 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2576 nrpcs++; 2577 #ifndef NFS_V2_ONLY 2578 if (v3) { 2579 nfsm_postop_attr(vp, attrflag, 0); 2580 if (!error) { 2581 nfsm_dissect(tl, u_int32_t *, 2582 2 * NFSX_UNSIGNED); 2583 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2584 dnp->n_cookieverf.nfsuquad[1] = *tl; 2585 } else { 2586 m_freem(mrep); 2587 goto nfsmout; 2588 } 2589 } 2590 #endif 2591 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2592 more_dirs = fxdr_unsigned(int, *tl); 2593 2594 /* loop thru the dir entries, doctoring them to 4bsd form */ 2595 while (more_dirs && bigenough) { 2596 #ifndef NFS_V2_ONLY 2597 if (v3) { 2598 nfsm_dissect(tl, u_int32_t *, 2599 3 * NFSX_UNSIGNED); 2600 fileno = fxdr_hyper(tl); 2601 len = fxdr_unsigned(int, *(tl + 2)); 2602 } else 2603 #endif 2604 { 2605 nfsm_dissect(tl, u_int32_t *, 2606 2 * NFSX_UNSIGNED); 2607 fileno = fxdr_unsigned(u_quad_t, *tl++); 2608 len = fxdr_unsigned(int, *tl); 2609 } 2610 if (len <= 0 || len > NFS_MAXNAMLEN) { 2611 error = EBADRPC; 2612 m_freem(mrep); 2613 goto nfsmout; 2614 } 2615 /* for cookie stashing */ 2616 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2617 left = NFS_DIRFRAGSIZ - blksiz; 2618 if (reclen > left) { 2619 memset(uiop->uio_iov->iov_base, 0, left); 2620 dp->d_reclen += left; 2621 UIO_ADVANCE(uiop, left); 2622 blksiz = 0; 2623 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2624 } 2625 if (reclen > uiop->uio_resid) 2626 bigenough = 0; 2627 if (bigenough) { 2628 int tlen; 2629 2630 dp = (struct dirent *)uiop->uio_iov->iov_base; 2631 dp->d_fileno = fileno; 2632 dp->d_namlen = len; 2633 dp->d_reclen = reclen; 2634 dp->d_type = DT_UNKNOWN; 2635 blksiz += reclen; 2636 if (blksiz == NFS_DIRFRAGSIZ) 2637 blksiz = 0; 2638 UIO_ADVANCE(uiop, DIRHDSIZ); 2639 nfsm_mtouio(uiop, len); 2640 tlen = reclen - (DIRHDSIZ + len); 2641 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2642 UIO_ADVANCE(uiop, tlen); 2643 } else 2644 nfsm_adv(nfsm_rndup(len)); 2645 #ifndef NFS_V2_ONLY 2646 if (v3) { 2647 nfsm_dissect(tl, u_int32_t *, 2648 3 * NFSX_UNSIGNED); 2649 } else 2650 #endif 2651 { 2652 nfsm_dissect(tl, u_int32_t *, 2653 2 * NFSX_UNSIGNED); 2654 } 2655 if (bigenough) { 2656 #ifndef NFS_V2_ONLY 2657 if (v3) { 2658 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2659 uiop->uio_offset = 2660 fxdr_swapcookie3(tl); 2661 else 2662 uiop->uio_offset = 2663 fxdr_cookie3(tl); 2664 } 2665 else 2666 #endif 2667 { 2668 uiop->uio_offset = 2669 fxdr_unsigned(off_t, *tl); 2670 } 2671 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2672 } 2673 if (v3) 2674 tl += 2; 2675 else 2676 tl++; 2677 more_dirs = fxdr_unsigned(int, *tl); 2678 } 2679 /* 2680 * If at end of rpc data, get the eof boolean 2681 */ 2682 if (!more_dirs) { 2683 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2684 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2685 2686 /* 2687 * kludge: if we got no entries, treat it as EOF. 2688 * some server sometimes send a reply without any 2689 * entries or EOF. 2690 * although it might mean the server has very long name, 2691 * we can't handle such entries anyway. 2692 */ 2693 2694 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2695 more_dirs = 0; 2696 } 2697 m_freem(mrep); 2698 } 2699 /* 2700 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2701 * by increasing d_reclen for the last record. 2702 */ 2703 if (blksiz > 0) { 2704 left = NFS_DIRFRAGSIZ - blksiz; 2705 memset(uiop->uio_iov->iov_base, 0, left); 2706 dp->d_reclen += left; 2707 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2708 UIO_ADVANCE(uiop, left); 2709 } 2710 2711 /* 2712 * We are now either at the end of the directory or have filled the 2713 * block. 2714 */ 2715 if (bigenough) { 2716 dnp->n_direofoffset = uiop->uio_offset; 2717 dnp->n_flag |= NEOFVALID; 2718 } 2719 nfsmout: 2720 return (error); 2721 } 2722 2723 #ifndef NFS_V2_ONLY 2724 /* 2725 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2726 */ 2727 int 2728 nfs_readdirplusrpc(vp, uiop, cred) 2729 struct vnode *vp; 2730 struct uio *uiop; 2731 kauth_cred_t cred; 2732 { 2733 int len, left; 2734 struct dirent *dp = NULL; 2735 u_int32_t *tl; 2736 char *cp; 2737 int32_t t1, t2; 2738 struct vnode *newvp; 2739 char *bpos, *dpos, *cp2; 2740 struct mbuf *mreq, *mrep, *md, *mb; 2741 struct nameidata nami, *ndp = &nami; 2742 struct componentname *cnp = &ndp->ni_cnd; 2743 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2744 struct nfsnode *dnp = VTONFS(vp), *np; 2745 nfsfh_t *fhp; 2746 u_quad_t fileno; 2747 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2748 int attrflag, fhsize, nrpcs = 0, reclen; 2749 struct nfs_fattr fattr, *fp; 2750 2751 #ifdef DIAGNOSTIC 2752 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2753 panic("nfs readdirplusrpc bad uio"); 2754 #endif 2755 ndp->ni_dvp = vp; 2756 newvp = NULLVP; 2757 2758 /* 2759 * Loop around doing readdir rpc's of size nm_readdirsize 2760 * truncated to a multiple of NFS_DIRFRAGSIZ. 2761 * The stopping criteria is EOF or buffer full. 2762 */ 2763 while (more_dirs && bigenough) { 2764 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2765 bigenough = 0; 2766 break; 2767 } 2768 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2769 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2770 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2771 nfsm_fhtom(dnp, 1); 2772 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2773 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2774 txdr_swapcookie3(uiop->uio_offset, tl); 2775 } else { 2776 txdr_cookie3(uiop->uio_offset, tl); 2777 } 2778 tl += 2; 2779 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2780 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2781 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2782 *tl = txdr_unsigned(nmp->nm_rsize); 2783 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2784 nfsm_postop_attr(vp, attrflag, 0); 2785 if (error) { 2786 m_freem(mrep); 2787 goto nfsmout; 2788 } 2789 nrpcs++; 2790 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2791 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2792 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2793 more_dirs = fxdr_unsigned(int, *tl); 2794 2795 /* loop thru the dir entries, doctoring them to 4bsd form */ 2796 while (more_dirs && bigenough) { 2797 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2798 fileno = fxdr_hyper(tl); 2799 len = fxdr_unsigned(int, *(tl + 2)); 2800 if (len <= 0 || len > NFS_MAXNAMLEN) { 2801 error = EBADRPC; 2802 m_freem(mrep); 2803 goto nfsmout; 2804 } 2805 /* for cookie stashing */ 2806 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2807 left = NFS_DIRFRAGSIZ - blksiz; 2808 if (reclen > left) { 2809 /* 2810 * DIRFRAGSIZ is aligned, no need to align 2811 * again here. 2812 */ 2813 memset(uiop->uio_iov->iov_base, 0, left); 2814 dp->d_reclen += left; 2815 UIO_ADVANCE(uiop, left); 2816 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2817 blksiz = 0; 2818 } 2819 if (reclen > uiop->uio_resid) 2820 bigenough = 0; 2821 if (bigenough) { 2822 int tlen; 2823 2824 dp = (struct dirent *)uiop->uio_iov->iov_base; 2825 dp->d_fileno = fileno; 2826 dp->d_namlen = len; 2827 dp->d_reclen = reclen; 2828 dp->d_type = DT_UNKNOWN; 2829 blksiz += reclen; 2830 if (blksiz == NFS_DIRFRAGSIZ) 2831 blksiz = 0; 2832 UIO_ADVANCE(uiop, DIRHDSIZ); 2833 nfsm_mtouio(uiop, len); 2834 tlen = reclen - (DIRHDSIZ + len); 2835 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2836 UIO_ADVANCE(uiop, tlen); 2837 cnp->cn_nameptr = dp->d_name; 2838 cnp->cn_namelen = dp->d_namlen; 2839 } else 2840 nfsm_adv(nfsm_rndup(len)); 2841 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2842 if (bigenough) { 2843 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2844 uiop->uio_offset = 2845 fxdr_swapcookie3(tl); 2846 else 2847 uiop->uio_offset = 2848 fxdr_cookie3(tl); 2849 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2850 } 2851 tl += 2; 2852 2853 /* 2854 * Since the attributes are before the file handle 2855 * (sigh), we must skip over the attributes and then 2856 * come back and get them. 2857 */ 2858 attrflag = fxdr_unsigned(int, *tl); 2859 if (attrflag) { 2860 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2861 memcpy(&fattr, fp, NFSX_V3FATTR); 2862 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2863 doit = fxdr_unsigned(int, *tl); 2864 if (doit) { 2865 nfsm_getfh(fhp, fhsize, 1); 2866 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2867 VREF(vp); 2868 newvp = vp; 2869 np = dnp; 2870 } else { 2871 error = nfs_nget1(vp->v_mount, fhp, 2872 fhsize, &np, LK_NOWAIT); 2873 if (!error) 2874 newvp = NFSTOV(np); 2875 } 2876 if (!error) { 2877 const char *xcp; 2878 2879 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2880 if (bigenough) { 2881 dp->d_type = 2882 IFTODT(VTTOIF(np->n_vattr->va_type)); 2883 if (cnp->cn_namelen <= NCHNAMLEN) { 2884 ndp->ni_vp = newvp; 2885 xcp = cnp->cn_nameptr + 2886 cnp->cn_namelen; 2887 cnp->cn_hash = 2888 namei_hash(cnp->cn_nameptr, &xcp); 2889 nfs_cache_enter(ndp->ni_dvp, 2890 ndp->ni_vp, cnp); 2891 } 2892 } 2893 } 2894 error = 0; 2895 } 2896 } else { 2897 /* Just skip over the file handle */ 2898 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2899 i = fxdr_unsigned(int, *tl); 2900 nfsm_adv(nfsm_rndup(i)); 2901 } 2902 if (newvp != NULLVP) { 2903 if (newvp == vp) 2904 vrele(newvp); 2905 else 2906 vput(newvp); 2907 newvp = NULLVP; 2908 } 2909 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2910 more_dirs = fxdr_unsigned(int, *tl); 2911 } 2912 /* 2913 * If at end of rpc data, get the eof boolean 2914 */ 2915 if (!more_dirs) { 2916 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2917 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2918 2919 /* 2920 * kludge: see a comment in nfs_readdirrpc. 2921 */ 2922 2923 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2924 more_dirs = 0; 2925 } 2926 m_freem(mrep); 2927 } 2928 /* 2929 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2930 * by increasing d_reclen for the last record. 2931 */ 2932 if (blksiz > 0) { 2933 left = NFS_DIRFRAGSIZ - blksiz; 2934 memset(uiop->uio_iov->iov_base, 0, left); 2935 dp->d_reclen += left; 2936 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2937 UIO_ADVANCE(uiop, left); 2938 } 2939 2940 /* 2941 * We are now either at the end of the directory or have filled the 2942 * block. 2943 */ 2944 if (bigenough) { 2945 dnp->n_direofoffset = uiop->uio_offset; 2946 dnp->n_flag |= NEOFVALID; 2947 } 2948 nfsmout: 2949 if (newvp != NULLVP) { 2950 if(newvp == vp) 2951 vrele(newvp); 2952 else 2953 vput(newvp); 2954 } 2955 return (error); 2956 } 2957 #endif 2958 2959 /* 2960 * Silly rename. To make the NFS filesystem that is stateless look a little 2961 * more like the "ufs" a remove of an active vnode is translated to a rename 2962 * to a funny looking filename that is removed by nfs_inactive on the 2963 * nfsnode. There is the potential for another process on a different client 2964 * to create the same funny name between the nfs_lookitup() fails and the 2965 * nfs_rename() completes, but... 2966 */ 2967 int 2968 nfs_sillyrename(dvp, vp, cnp, dolink) 2969 struct vnode *dvp, *vp; 2970 struct componentname *cnp; 2971 bool dolink; 2972 { 2973 struct sillyrename *sp; 2974 struct nfsnode *np; 2975 int error; 2976 short pid; 2977 2978 cache_purge(dvp); 2979 np = VTONFS(vp); 2980 #ifndef DIAGNOSTIC 2981 if (vp->v_type == VDIR) 2982 panic("nfs: sillyrename dir"); 2983 #endif 2984 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2985 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2986 sp->s_dvp = dvp; 2987 VREF(dvp); 2988 2989 /* Fudge together a funny name */ 2990 pid = curlwp->l_proc->p_pid; 2991 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2992 sp->s_namlen = 12; 2993 sp->s_name[8] = hexdigits[pid & 0xf]; 2994 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2995 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2996 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2997 2998 /* Try lookitups until we get one that isn't there */ 2999 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3000 curlwp, (struct nfsnode **)0) == 0) { 3001 sp->s_name[4]++; 3002 if (sp->s_name[4] > 'z') { 3003 error = EINVAL; 3004 goto bad; 3005 } 3006 } 3007 if (dolink) { 3008 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 3009 sp->s_cred, curlwp); 3010 /* 3011 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 3012 */ 3013 if (error == ENOTSUP) { 3014 error = nfs_renameit(dvp, cnp, sp); 3015 } 3016 } else { 3017 error = nfs_renameit(dvp, cnp, sp); 3018 } 3019 if (error) 3020 goto bad; 3021 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3022 curlwp, &np); 3023 np->n_sillyrename = sp; 3024 return (0); 3025 bad: 3026 vrele(sp->s_dvp); 3027 kauth_cred_free(sp->s_cred); 3028 kmem_free(sp, sizeof(*sp)); 3029 return (error); 3030 } 3031 3032 /* 3033 * Look up a file name and optionally either update the file handle or 3034 * allocate an nfsnode, depending on the value of npp. 3035 * npp == NULL --> just do the lookup 3036 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 3037 * handled too 3038 * *npp != NULL --> update the file handle in the vnode 3039 */ 3040 int 3041 nfs_lookitup(dvp, name, len, cred, l, npp) 3042 struct vnode *dvp; 3043 const char *name; 3044 int len; 3045 kauth_cred_t cred; 3046 struct lwp *l; 3047 struct nfsnode **npp; 3048 { 3049 u_int32_t *tl; 3050 char *cp; 3051 int32_t t1, t2; 3052 struct vnode *newvp = (struct vnode *)0; 3053 struct nfsnode *np, *dnp = VTONFS(dvp); 3054 char *bpos, *dpos, *cp2; 3055 int error = 0, fhlen; 3056 #ifndef NFS_V2_ONLY 3057 int attrflag; 3058 #endif 3059 struct mbuf *mreq, *mrep, *md, *mb; 3060 nfsfh_t *nfhp; 3061 const int v3 = NFS_ISV3(dvp); 3062 3063 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 3064 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 3065 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 3066 nfsm_fhtom(dnp, v3); 3067 nfsm_strtom(name, len, NFS_MAXNAMLEN); 3068 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 3069 if (npp && !error) { 3070 nfsm_getfh(nfhp, fhlen, v3); 3071 if (*npp) { 3072 np = *npp; 3073 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3074 kmem_free(np->n_fhp, np->n_fhsize); 3075 np->n_fhp = &np->n_fh; 3076 } 3077 #if NFS_SMALLFH < NFSX_V3FHMAX 3078 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 3079 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 3080 #endif 3081 memcpy(np->n_fhp, nfhp, fhlen); 3082 np->n_fhsize = fhlen; 3083 newvp = NFSTOV(np); 3084 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3085 VREF(dvp); 3086 newvp = dvp; 3087 np = dnp; 3088 } else { 3089 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 3090 if (error) { 3091 m_freem(mrep); 3092 return (error); 3093 } 3094 newvp = NFSTOV(np); 3095 } 3096 #ifndef NFS_V2_ONLY 3097 if (v3) { 3098 nfsm_postop_attr(newvp, attrflag, 0); 3099 if (!attrflag && *npp == NULL) { 3100 m_freem(mrep); 3101 vput(newvp); 3102 return (ENOENT); 3103 } 3104 } else 3105 #endif 3106 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3107 } 3108 nfsm_reqdone; 3109 if (npp && *npp == NULL) { 3110 if (error) { 3111 if (newvp) 3112 vput(newvp); 3113 } else 3114 *npp = np; 3115 } 3116 return (error); 3117 } 3118 3119 #ifndef NFS_V2_ONLY 3120 /* 3121 * Nfs Version 3 commit rpc 3122 */ 3123 int 3124 nfs_commit(vp, offset, cnt, l) 3125 struct vnode *vp; 3126 off_t offset; 3127 uint32_t cnt; 3128 struct lwp *l; 3129 { 3130 char *cp; 3131 u_int32_t *tl; 3132 int32_t t1, t2; 3133 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3134 char *bpos, *dpos, *cp2; 3135 int error = 0, wccflag = NFSV3_WCCRATTR; 3136 struct mbuf *mreq, *mrep, *md, *mb; 3137 struct nfsnode *np; 3138 3139 KASSERT(NFS_ISV3(vp)); 3140 3141 #ifdef NFS_DEBUG_COMMIT 3142 printf("commit %lu - %lu\n", (unsigned long)offset, 3143 (unsigned long)(offset + cnt)); 3144 #endif 3145 3146 mutex_enter(&nmp->nm_lock); 3147 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3148 mutex_exit(&nmp->nm_lock); 3149 return (0); 3150 } 3151 mutex_exit(&nmp->nm_lock); 3152 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3153 np = VTONFS(vp); 3154 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3155 nfsm_fhtom(np, 1); 3156 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3157 txdr_hyper(offset, tl); 3158 tl += 2; 3159 *tl = txdr_unsigned(cnt); 3160 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3161 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3162 if (!error) { 3163 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3164 mutex_enter(&nmp->nm_lock); 3165 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3166 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3167 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3168 error = NFSERR_STALEWRITEVERF; 3169 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3170 } 3171 mutex_exit(&nmp->nm_lock); 3172 } 3173 nfsm_reqdone; 3174 return (error); 3175 } 3176 #endif 3177 3178 /* 3179 * Kludge City.. 3180 * - make nfs_bmap() essentially a no-op that does no translation 3181 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3182 * (Maybe I could use the process's page mapping, but I was concerned that 3183 * Kernel Write might not be enabled and also figured copyout() would do 3184 * a lot more work than memcpy() and also it currently happens in the 3185 * context of the swapper process (2). 3186 */ 3187 int 3188 nfs_bmap(v) 3189 void *v; 3190 { 3191 struct vop_bmap_args /* { 3192 struct vnode *a_vp; 3193 daddr_t a_bn; 3194 struct vnode **a_vpp; 3195 daddr_t *a_bnp; 3196 int *a_runp; 3197 } */ *ap = v; 3198 struct vnode *vp = ap->a_vp; 3199 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3200 3201 if (ap->a_vpp != NULL) 3202 *ap->a_vpp = vp; 3203 if (ap->a_bnp != NULL) 3204 *ap->a_bnp = ap->a_bn << bshift; 3205 if (ap->a_runp != NULL) 3206 *ap->a_runp = 1024 * 1024; /* XXX */ 3207 return (0); 3208 } 3209 3210 /* 3211 * Strategy routine. 3212 * For async requests when nfsiod(s) are running, queue the request by 3213 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3214 * request. 3215 */ 3216 int 3217 nfs_strategy(v) 3218 void *v; 3219 { 3220 struct vop_strategy_args *ap = v; 3221 struct buf *bp = ap->a_bp; 3222 int error = 0; 3223 3224 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3225 panic("nfs physio/async"); 3226 3227 /* 3228 * If the op is asynchronous and an i/o daemon is waiting 3229 * queue the request, wake it up and wait for completion 3230 * otherwise just do it ourselves. 3231 */ 3232 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3233 error = nfs_doio(bp); 3234 return (error); 3235 } 3236 3237 /* 3238 * fsync vnode op. Just call nfs_flush() with commit == 1. 3239 */ 3240 /* ARGSUSED */ 3241 int 3242 nfs_fsync(v) 3243 void *v; 3244 { 3245 struct vop_fsync_args /* { 3246 struct vnodeop_desc *a_desc; 3247 struct vnode * a_vp; 3248 kauth_cred_t a_cred; 3249 int a_flags; 3250 off_t offlo; 3251 off_t offhi; 3252 struct lwp * a_l; 3253 } */ *ap = v; 3254 3255 struct vnode *vp = ap->a_vp; 3256 3257 if (vp->v_type != VREG) 3258 return 0; 3259 3260 return (nfs_flush(vp, ap->a_cred, 3261 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3262 } 3263 3264 /* 3265 * Flush all the data associated with a vnode. 3266 */ 3267 int 3268 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3269 int commit) 3270 { 3271 struct nfsnode *np = VTONFS(vp); 3272 int error; 3273 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3274 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3275 3276 mutex_enter(&vp->v_interlock); 3277 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3278 if (np->n_flag & NWRITEERR) { 3279 error = np->n_error; 3280 np->n_flag &= ~NWRITEERR; 3281 } 3282 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3283 return (error); 3284 } 3285 3286 /* 3287 * Return POSIX pathconf information applicable to nfs. 3288 * 3289 * N.B. The NFS V2 protocol doesn't support this RPC. 3290 */ 3291 /* ARGSUSED */ 3292 int 3293 nfs_pathconf(v) 3294 void *v; 3295 { 3296 struct vop_pathconf_args /* { 3297 struct vnode *a_vp; 3298 int a_name; 3299 register_t *a_retval; 3300 } */ *ap = v; 3301 struct nfsv3_pathconf *pcp; 3302 struct vnode *vp = ap->a_vp; 3303 struct mbuf *mreq, *mrep, *md, *mb; 3304 int32_t t1, t2; 3305 u_int32_t *tl; 3306 char *bpos, *dpos, *cp, *cp2; 3307 int error = 0, attrflag; 3308 #ifndef NFS_V2_ONLY 3309 struct nfsmount *nmp; 3310 unsigned int l; 3311 u_int64_t maxsize; 3312 #endif 3313 const int v3 = NFS_ISV3(vp); 3314 struct nfsnode *np = VTONFS(vp); 3315 3316 switch (ap->a_name) { 3317 /* Names that can be resolved locally. */ 3318 case _PC_PIPE_BUF: 3319 *ap->a_retval = PIPE_BUF; 3320 break; 3321 case _PC_SYNC_IO: 3322 *ap->a_retval = 1; 3323 break; 3324 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3325 case _PC_LINK_MAX: 3326 case _PC_NAME_MAX: 3327 case _PC_CHOWN_RESTRICTED: 3328 case _PC_NO_TRUNC: 3329 if (!v3) { 3330 error = EINVAL; 3331 break; 3332 } 3333 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3334 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3335 nfsm_fhtom(np, 1); 3336 nfsm_request(np, NFSPROC_PATHCONF, 3337 curlwp, curlwp->l_cred); /* XXX */ 3338 nfsm_postop_attr(vp, attrflag, 0); 3339 if (!error) { 3340 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3341 NFSX_V3PATHCONF); 3342 switch (ap->a_name) { 3343 case _PC_LINK_MAX: 3344 *ap->a_retval = 3345 fxdr_unsigned(register_t, pcp->pc_linkmax); 3346 break; 3347 case _PC_NAME_MAX: 3348 *ap->a_retval = 3349 fxdr_unsigned(register_t, pcp->pc_namemax); 3350 break; 3351 case _PC_CHOWN_RESTRICTED: 3352 *ap->a_retval = 3353 (pcp->pc_chownrestricted == nfs_true); 3354 break; 3355 case _PC_NO_TRUNC: 3356 *ap->a_retval = 3357 (pcp->pc_notrunc == nfs_true); 3358 break; 3359 } 3360 } 3361 nfsm_reqdone; 3362 break; 3363 case _PC_FILESIZEBITS: 3364 #ifndef NFS_V2_ONLY 3365 if (v3) { 3366 nmp = VFSTONFS(vp->v_mount); 3367 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3368 if ((error = nfs_fsinfo(nmp, vp, 3369 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3370 break; 3371 for (l = 0, maxsize = nmp->nm_maxfilesize; 3372 (maxsize >> l) > 0; l++) 3373 ; 3374 *ap->a_retval = l + 1; 3375 } else 3376 #endif 3377 { 3378 *ap->a_retval = 32; /* NFS V2 limitation */ 3379 } 3380 break; 3381 default: 3382 error = EINVAL; 3383 break; 3384 } 3385 3386 return (error); 3387 } 3388 3389 /* 3390 * NFS advisory byte-level locks. 3391 */ 3392 int 3393 nfs_advlock(v) 3394 void *v; 3395 { 3396 struct vop_advlock_args /* { 3397 struct vnode *a_vp; 3398 void *a_id; 3399 int a_op; 3400 struct flock *a_fl; 3401 int a_flags; 3402 } */ *ap = v; 3403 struct nfsnode *np = VTONFS(ap->a_vp); 3404 3405 return lf_advlock(ap, &np->n_lockf, np->n_size); 3406 } 3407 3408 /* 3409 * Print out the contents of an nfsnode. 3410 */ 3411 int 3412 nfs_print(v) 3413 void *v; 3414 { 3415 struct vop_print_args /* { 3416 struct vnode *a_vp; 3417 } */ *ap = v; 3418 struct vnode *vp = ap->a_vp; 3419 struct nfsnode *np = VTONFS(vp); 3420 3421 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3422 (unsigned long long)np->n_vattr->va_fileid, 3423 (unsigned long long)np->n_vattr->va_fsid); 3424 if (vp->v_type == VFIFO) 3425 fifo_printinfo(vp); 3426 printf("\n"); 3427 return (0); 3428 } 3429 3430 /* 3431 * nfs unlock wrapper. 3432 */ 3433 int 3434 nfs_unlock(void *v) 3435 { 3436 struct vop_unlock_args /* { 3437 struct vnode *a_vp; 3438 int a_flags; 3439 } */ *ap = v; 3440 struct vnode *vp = ap->a_vp; 3441 3442 /* 3443 * VOP_UNLOCK can be called by nfs_loadattrcache 3444 * with v_data == 0. 3445 */ 3446 if (VTONFS(vp)) { 3447 nfs_delayedtruncate(vp); 3448 } 3449 3450 return genfs_unlock(v); 3451 } 3452 3453 /* 3454 * nfs special file access vnode op. 3455 * Essentially just get vattr and then imitate iaccess() since the device is 3456 * local to the client. 3457 */ 3458 int 3459 nfsspec_access(v) 3460 void *v; 3461 { 3462 struct vop_access_args /* { 3463 struct vnode *a_vp; 3464 int a_mode; 3465 kauth_cred_t a_cred; 3466 struct lwp *a_l; 3467 } */ *ap = v; 3468 struct vattr va; 3469 struct vnode *vp = ap->a_vp; 3470 int error; 3471 3472 error = VOP_GETATTR(vp, &va, ap->a_cred); 3473 if (error) 3474 return (error); 3475 3476 /* 3477 * Disallow write attempts on filesystems mounted read-only; 3478 * unless the file is a socket, fifo, or a block or character 3479 * device resident on the filesystem. 3480 */ 3481 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3482 switch (vp->v_type) { 3483 case VREG: 3484 case VDIR: 3485 case VLNK: 3486 return (EROFS); 3487 default: 3488 break; 3489 } 3490 } 3491 3492 return (vaccess(va.va_type, va.va_mode, 3493 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3494 } 3495 3496 /* 3497 * Read wrapper for special devices. 3498 */ 3499 int 3500 nfsspec_read(v) 3501 void *v; 3502 { 3503 struct vop_read_args /* { 3504 struct vnode *a_vp; 3505 struct uio *a_uio; 3506 int a_ioflag; 3507 kauth_cred_t a_cred; 3508 } */ *ap = v; 3509 struct nfsnode *np = VTONFS(ap->a_vp); 3510 3511 /* 3512 * Set access flag. 3513 */ 3514 np->n_flag |= NACC; 3515 getnanotime(&np->n_atim); 3516 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3517 } 3518 3519 /* 3520 * Write wrapper for special devices. 3521 */ 3522 int 3523 nfsspec_write(v) 3524 void *v; 3525 { 3526 struct vop_write_args /* { 3527 struct vnode *a_vp; 3528 struct uio *a_uio; 3529 int a_ioflag; 3530 kauth_cred_t a_cred; 3531 } */ *ap = v; 3532 struct nfsnode *np = VTONFS(ap->a_vp); 3533 3534 /* 3535 * Set update flag. 3536 */ 3537 np->n_flag |= NUPD; 3538 getnanotime(&np->n_mtim); 3539 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3540 } 3541 3542 /* 3543 * Close wrapper for special devices. 3544 * 3545 * Update the times on the nfsnode then do device close. 3546 */ 3547 int 3548 nfsspec_close(v) 3549 void *v; 3550 { 3551 struct vop_close_args /* { 3552 struct vnode *a_vp; 3553 int a_fflag; 3554 kauth_cred_t a_cred; 3555 struct lwp *a_l; 3556 } */ *ap = v; 3557 struct vnode *vp = ap->a_vp; 3558 struct nfsnode *np = VTONFS(vp); 3559 struct vattr vattr; 3560 3561 if (np->n_flag & (NACC | NUPD)) { 3562 np->n_flag |= NCHG; 3563 if (vp->v_usecount == 1 && 3564 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3565 VATTR_NULL(&vattr); 3566 if (np->n_flag & NACC) 3567 vattr.va_atime = np->n_atim; 3568 if (np->n_flag & NUPD) 3569 vattr.va_mtime = np->n_mtim; 3570 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3571 } 3572 } 3573 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3574 } 3575 3576 /* 3577 * Read wrapper for fifos. 3578 */ 3579 int 3580 nfsfifo_read(v) 3581 void *v; 3582 { 3583 struct vop_read_args /* { 3584 struct vnode *a_vp; 3585 struct uio *a_uio; 3586 int a_ioflag; 3587 kauth_cred_t a_cred; 3588 } */ *ap = v; 3589 struct nfsnode *np = VTONFS(ap->a_vp); 3590 3591 /* 3592 * Set access flag. 3593 */ 3594 np->n_flag |= NACC; 3595 getnanotime(&np->n_atim); 3596 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3597 } 3598 3599 /* 3600 * Write wrapper for fifos. 3601 */ 3602 int 3603 nfsfifo_write(v) 3604 void *v; 3605 { 3606 struct vop_write_args /* { 3607 struct vnode *a_vp; 3608 struct uio *a_uio; 3609 int a_ioflag; 3610 kauth_cred_t a_cred; 3611 } */ *ap = v; 3612 struct nfsnode *np = VTONFS(ap->a_vp); 3613 3614 /* 3615 * Set update flag. 3616 */ 3617 np->n_flag |= NUPD; 3618 getnanotime(&np->n_mtim); 3619 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3620 } 3621 3622 /* 3623 * Close wrapper for fifos. 3624 * 3625 * Update the times on the nfsnode then do fifo close. 3626 */ 3627 int 3628 nfsfifo_close(v) 3629 void *v; 3630 { 3631 struct vop_close_args /* { 3632 struct vnode *a_vp; 3633 int a_fflag; 3634 kauth_cred_t a_cred; 3635 struct lwp *a_l; 3636 } */ *ap = v; 3637 struct vnode *vp = ap->a_vp; 3638 struct nfsnode *np = VTONFS(vp); 3639 struct vattr vattr; 3640 3641 if (np->n_flag & (NACC | NUPD)) { 3642 struct timespec ts; 3643 3644 getnanotime(&ts); 3645 if (np->n_flag & NACC) 3646 np->n_atim = ts; 3647 if (np->n_flag & NUPD) 3648 np->n_mtim = ts; 3649 np->n_flag |= NCHG; 3650 if (vp->v_usecount == 1 && 3651 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3652 VATTR_NULL(&vattr); 3653 if (np->n_flag & NACC) 3654 vattr.va_atime = np->n_atim; 3655 if (np->n_flag & NUPD) 3656 vattr.va_mtime = np->n_mtim; 3657 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3658 } 3659 } 3660 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3661 } 3662
Indexes created Fri Sep 26 20:09:58 GMT 2025