nfs_vnops.c revision 1.282.2.3
1 /* $NetBSD: nfs_vnops.c,v 1.282.2.3 2010/11/06 08:08:50 uebayasi 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 1.282.2.3 2010/11/06 08:08:50 uebayasi Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_uvmhist.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/proc.h> 51 #include <sys/kernel.h> 52 #include <sys/systm.h> 53 #include <sys/resourcevar.h> 54 #include <sys/mount.h> 55 #include <sys/buf.h> 56 #include <sys/condvar.h> 57 #include <sys/disk.h> 58 #include <sys/malloc.h> 59 #include <sys/kmem.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/vnode.h> 64 #include <sys/dirent.h> 65 #include <sys/fcntl.h> 66 #include <sys/hash.h> 67 #include <sys/lockf.h> 68 #include <sys/stat.h> 69 #include <sys/unistd.h> 70 #include <sys/kauth.h> 71 72 #include <uvm/uvm_extern.h> 73 #include <uvm/uvm.h> 74 75 #include <miscfs/fifofs/fifo.h> 76 #include <miscfs/genfs/genfs.h> 77 #include <miscfs/genfs/genfs_node.h> 78 #include <miscfs/specfs/specdev.h> 79 80 #include <nfs/rpcv2.h> 81 #include <nfs/nfsproto.h> 82 #include <nfs/nfs.h> 83 #include <nfs/nfsnode.h> 84 #include <nfs/nfsmount.h> 85 #include <nfs/xdr_subs.h> 86 #include <nfs/nfsm_subs.h> 87 #include <nfs/nfs_var.h> 88 89 #include <net/if.h> 90 #include <netinet/in.h> 91 #include <netinet/in_var.h> 92 93 /* 94 * Global vfs data structures for nfs 95 */ 96 int (**nfsv2_vnodeop_p)(void *); 97 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 98 { &vop_default_desc, vn_default_error }, 99 { &vop_lookup_desc, nfs_lookup }, /* lookup */ 100 { &vop_create_desc, nfs_create }, /* create */ 101 { &vop_mknod_desc, nfs_mknod }, /* mknod */ 102 { &vop_open_desc, nfs_open }, /* open */ 103 { &vop_close_desc, nfs_close }, /* close */ 104 { &vop_access_desc, nfs_access }, /* access */ 105 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 106 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 107 { &vop_read_desc, nfs_read }, /* read */ 108 { &vop_write_desc, nfs_write }, /* write */ 109 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 110 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */ 111 { &vop_poll_desc, nfs_poll }, /* poll */ 112 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */ 113 { &vop_revoke_desc, nfs_revoke }, /* revoke */ 114 { &vop_mmap_desc, nfs_mmap }, /* mmap */ 115 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 116 { &vop_seek_desc, nfs_seek }, /* seek */ 117 { &vop_remove_desc, nfs_remove }, /* remove */ 118 { &vop_link_desc, nfs_link }, /* link */ 119 { &vop_rename_desc, nfs_rename }, /* rename */ 120 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */ 121 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */ 122 { &vop_symlink_desc, nfs_symlink }, /* symlink */ 123 { &vop_readdir_desc, nfs_readdir }, /* readdir */ 124 { &vop_readlink_desc, nfs_readlink }, /* readlink */ 125 { &vop_abortop_desc, nfs_abortop }, /* abortop */ 126 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 127 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 128 { &vop_lock_desc, nfs_lock }, /* lock */ 129 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 130 { &vop_bmap_desc, nfs_bmap }, /* bmap */ 131 { &vop_strategy_desc, nfs_strategy }, /* strategy */ 132 { &vop_print_desc, nfs_print }, /* print */ 133 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 134 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */ 135 { &vop_advlock_desc, nfs_advlock }, /* advlock */ 136 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 137 { &vop_getpages_desc, nfs_getpages }, /* getpages */ 138 { &vop_putpages_desc, genfs_putpages }, /* putpages */ 139 { NULL, NULL } 140 }; 141 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 142 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 143 144 /* 145 * Special device vnode ops 146 */ 147 int (**spec_nfsv2nodeop_p)(void *); 148 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { 149 { &vop_default_desc, vn_default_error }, 150 { &vop_lookup_desc, spec_lookup }, /* lookup */ 151 { &vop_create_desc, spec_create }, /* create */ 152 { &vop_mknod_desc, spec_mknod }, /* mknod */ 153 { &vop_open_desc, spec_open }, /* open */ 154 { &vop_close_desc, nfsspec_close }, /* close */ 155 { &vop_access_desc, nfsspec_access }, /* access */ 156 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 157 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 158 { &vop_read_desc, nfsspec_read }, /* read */ 159 { &vop_write_desc, nfsspec_write }, /* write */ 160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 162 { &vop_poll_desc, spec_poll }, /* poll */ 163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 164 { &vop_revoke_desc, spec_revoke }, /* revoke */ 165 { &vop_mmap_desc, spec_mmap }, /* mmap */ 166 { &vop_fsync_desc, spec_fsync }, /* fsync */ 167 { &vop_seek_desc, spec_seek }, /* seek */ 168 { &vop_remove_desc, spec_remove }, /* remove */ 169 { &vop_link_desc, spec_link }, /* link */ 170 { &vop_rename_desc, spec_rename }, /* rename */ 171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 173 { &vop_symlink_desc, spec_symlink }, /* symlink */ 174 { &vop_readdir_desc, spec_readdir }, /* readdir */ 175 { &vop_readlink_desc, spec_readlink }, /* readlink */ 176 { &vop_abortop_desc, spec_abortop }, /* abortop */ 177 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 179 { &vop_lock_desc, nfs_lock }, /* lock */ 180 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 181 { &vop_bmap_desc, spec_bmap }, /* bmap */ 182 { &vop_strategy_desc, spec_strategy }, /* strategy */ 183 { &vop_print_desc, nfs_print }, /* print */ 184 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, spec_advlock }, /* advlock */ 187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */ 188 { &vop_getpages_desc, spec_getpages }, /* getpages */ 189 { &vop_putpages_desc, spec_putpages }, /* putpages */ 190 { NULL, NULL } 191 }; 192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries }; 194 195 int (**fifo_nfsv2nodeop_p)(void *); 196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { 197 { &vop_default_desc, vn_default_error }, 198 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 199 { &vop_create_desc, vn_fifo_bypass }, /* create */ 200 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 201 { &vop_open_desc, vn_fifo_bypass }, /* open */ 202 { &vop_close_desc, nfsfifo_close }, /* close */ 203 { &vop_access_desc, nfsspec_access }, /* access */ 204 { &vop_getattr_desc, nfs_getattr }, /* getattr */ 205 { &vop_setattr_desc, nfs_setattr }, /* setattr */ 206 { &vop_read_desc, nfsfifo_read }, /* read */ 207 { &vop_write_desc, nfsfifo_write }, /* write */ 208 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */ 209 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 210 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 211 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 212 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 213 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 214 { &vop_fsync_desc, nfs_fsync }, /* fsync */ 215 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 216 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 217 { &vop_link_desc, vn_fifo_bypass }, /* link */ 218 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 219 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 220 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 221 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 222 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 223 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 224 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 225 { &vop_inactive_desc, nfs_inactive }, /* inactive */ 226 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */ 227 { &vop_lock_desc, nfs_lock }, /* lock */ 228 { &vop_unlock_desc, nfs_unlock }, /* unlock */ 229 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 230 { &vop_strategy_desc, genfs_badop }, /* strategy */ 231 { &vop_print_desc, nfs_print }, /* print */ 232 { &vop_islocked_desc, nfs_islocked }, /* islocked */ 233 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 234 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 235 { &vop_bwrite_desc, genfs_badop }, /* bwrite */ 236 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 237 { NULL, NULL } 238 }; 239 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 240 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries }; 241 242 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *, 243 size_t, kauth_cred_t, struct lwp *); 244 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *); 245 246 /* 247 * Global variables 248 */ 249 extern u_int32_t nfs_true, nfs_false; 250 extern u_int32_t nfs_xdrneg1; 251 extern const nfstype nfsv3_type[9]; 252 253 int nfs_numasync = 0; 254 #define DIRHDSIZ _DIRENT_NAMEOFF(dp) 255 #define UIO_ADVANCE(uio, siz) \ 256 (void)((uio)->uio_resid -= (siz), \ 257 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \ 258 (uio)->uio_iov->iov_len -= (siz)) 259 260 static void nfs_cache_enter(struct vnode *, struct vnode *, 261 struct componentname *); 262 263 static void 264 nfs_cache_enter(struct vnode *dvp, struct vnode *vp, 265 struct componentname *cnp) 266 { 267 struct nfsnode *dnp = VTONFS(dvp); 268 269 if (vp != NULL) { 270 struct nfsnode *np = VTONFS(vp); 271 272 np->n_ctime = np->n_vattr->va_ctime.tv_sec; 273 } 274 275 if (!timespecisset(&dnp->n_nctime)) 276 dnp->n_nctime = dnp->n_vattr->va_mtime; 277 278 cache_enter(dvp, vp, cnp); 279 } 280 281 /* 282 * nfs null call from vfs. 283 */ 284 int 285 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l) 286 { 287 char *bpos, *dpos; 288 int error = 0; 289 struct mbuf *mreq, *mrep, *md, *mb; 290 struct nfsnode *np = VTONFS(vp); 291 292 nfsm_reqhead(np, NFSPROC_NULL, 0); 293 nfsm_request(np, NFSPROC_NULL, l, cred); 294 nfsm_reqdone; 295 return (error); 296 } 297 298 /* 299 * nfs access vnode op. 300 * For nfs version 2, just return ok. File accesses may fail later. 301 * For nfs version 3, use the access rpc to check accessibility. If file modes 302 * are changed on the server, accesses might still fail later. 303 */ 304 int 305 nfs_access(void *v) 306 { 307 struct vop_access_args /* { 308 struct vnode *a_vp; 309 int a_mode; 310 kauth_cred_t a_cred; 311 } */ *ap = v; 312 struct vnode *vp = ap->a_vp; 313 #ifndef NFS_V2_ONLY 314 u_int32_t *tl; 315 char *cp; 316 int32_t t1, t2; 317 char *bpos, *dpos, *cp2; 318 int error = 0, attrflag; 319 struct mbuf *mreq, *mrep, *md, *mb; 320 u_int32_t mode, rmode; 321 const int v3 = NFS_ISV3(vp); 322 #endif 323 int cachevalid; 324 struct nfsnode *np = VTONFS(vp); 325 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 326 327 cachevalid = (np->n_accstamp != -1 && 328 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) && 329 np->n_accuid == kauth_cred_geteuid(ap->a_cred)); 330 331 /* 332 * Check access cache first. If this request has been made for this 333 * uid shortly before, use the cached result. 334 */ 335 if (cachevalid) { 336 if (!np->n_accerror) { 337 if ((np->n_accmode & ap->a_mode) == ap->a_mode) 338 return np->n_accerror; 339 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode) 340 return np->n_accerror; 341 } 342 343 #ifndef NFS_V2_ONLY 344 /* 345 * For nfs v3, do an access rpc, otherwise you are stuck emulating 346 * ufs_access() locally using the vattr. This may not be correct, 347 * since the server may apply other access criteria such as 348 * client uid-->server uid mapping that we do not know about, but 349 * this is better than just returning anything that is lying about 350 * in the cache. 351 */ 352 if (v3) { 353 nfsstats.rpccnt[NFSPROC_ACCESS]++; 354 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 355 nfsm_fhtom(np, v3); 356 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 357 if (ap->a_mode & VREAD) 358 mode = NFSV3ACCESS_READ; 359 else 360 mode = 0; 361 if (vp->v_type != VDIR) { 362 if (ap->a_mode & VWRITE) 363 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 364 if (ap->a_mode & VEXEC) 365 mode |= NFSV3ACCESS_EXECUTE; 366 } else { 367 if (ap->a_mode & VWRITE) 368 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 369 NFSV3ACCESS_DELETE); 370 if (ap->a_mode & VEXEC) 371 mode |= NFSV3ACCESS_LOOKUP; 372 } 373 *tl = txdr_unsigned(mode); 374 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred); 375 nfsm_postop_attr(vp, attrflag, 0); 376 if (!error) { 377 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 378 rmode = fxdr_unsigned(u_int32_t, *tl); 379 /* 380 * The NFS V3 spec does not clarify whether or not 381 * the returned access bits can be a superset of 382 * the ones requested, so... 383 */ 384 if ((rmode & mode) != mode) 385 error = EACCES; 386 } 387 nfsm_reqdone; 388 } else 389 #endif 390 return (nfsspec_access(ap)); 391 #ifndef NFS_V2_ONLY 392 /* 393 * Disallow write attempts on filesystems mounted read-only; 394 * unless the file is a socket, fifo, or a block or character 395 * device resident on the filesystem. 396 */ 397 if (!error && (ap->a_mode & VWRITE) && 398 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 399 switch (vp->v_type) { 400 case VREG: 401 case VDIR: 402 case VLNK: 403 error = EROFS; 404 default: 405 break; 406 } 407 } 408 409 if (!error || error == EACCES) { 410 /* 411 * If we got the same result as for a previous, 412 * different request, OR it in. Don't update 413 * the timestamp in that case. 414 */ 415 if (cachevalid && np->n_accstamp != -1 && 416 error == np->n_accerror) { 417 if (!error) 418 np->n_accmode |= ap->a_mode; 419 else if ((np->n_accmode & ap->a_mode) == ap->a_mode) 420 np->n_accmode = ap->a_mode; 421 } else { 422 np->n_accstamp = time_uptime; 423 np->n_accuid = kauth_cred_geteuid(ap->a_cred); 424 np->n_accmode = ap->a_mode; 425 np->n_accerror = error; 426 } 427 } 428 429 return (error); 430 #endif 431 } 432 433 /* 434 * nfs open vnode op 435 * Check to see if the type is ok 436 * and that deletion is not in progress. 437 * For paged in text files, you will need to flush the page cache 438 * if consistency is lost. 439 */ 440 /* ARGSUSED */ 441 int 442 nfs_open(void *v) 443 { 444 struct vop_open_args /* { 445 struct vnode *a_vp; 446 int a_mode; 447 kauth_cred_t a_cred; 448 } */ *ap = v; 449 struct vnode *vp = ap->a_vp; 450 struct nfsnode *np = VTONFS(vp); 451 int error; 452 453 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 454 return (EACCES); 455 } 456 457 if (ap->a_mode & FREAD) { 458 if (np->n_rcred != NULL) 459 kauth_cred_free(np->n_rcred); 460 np->n_rcred = ap->a_cred; 461 kauth_cred_hold(np->n_rcred); 462 } 463 if (ap->a_mode & FWRITE) { 464 if (np->n_wcred != NULL) 465 kauth_cred_free(np->n_wcred); 466 np->n_wcred = ap->a_cred; 467 kauth_cred_hold(np->n_wcred); 468 } 469 470 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0); 471 if (error) 472 return error; 473 474 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */ 475 476 return (0); 477 } 478 479 /* 480 * nfs close vnode op 481 * What an NFS client should do upon close after writing is a debatable issue. 482 * Most NFS clients push delayed writes to the server upon close, basically for 483 * two reasons: 484 * 1 - So that any write errors may be reported back to the client process 485 * doing the close system call. By far the two most likely errors are 486 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 487 * 2 - To put a worst case upper bound on cache inconsistency between 488 * multiple clients for the file. 489 * There is also a consistency problem for Version 2 of the protocol w.r.t. 490 * not being able to tell if other clients are writing a file concurrently, 491 * since there is no way of knowing if the changed modify time in the reply 492 * is only due to the write for this client. 493 * (NFS Version 3 provides weak cache consistency data in the reply that 494 * should be sufficient to detect and handle this case.) 495 * 496 * The current code does the following: 497 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 498 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 499 * or commit them (this satisfies 1 and 2 except for the 500 * case where the server crashes after this close but 501 * before the commit RPC, which is felt to be "good 502 * enough". Changing the last argument to nfs_flush() to 503 * a 1 would force a commit operation, if it is felt a 504 * commit is necessary now. 505 */ 506 /* ARGSUSED */ 507 int 508 nfs_close(void *v) 509 { 510 struct vop_close_args /* { 511 struct vnodeop_desc *a_desc; 512 struct vnode *a_vp; 513 int a_fflag; 514 kauth_cred_t a_cred; 515 } */ *ap = v; 516 struct vnode *vp = ap->a_vp; 517 struct nfsnode *np = VTONFS(vp); 518 int error = 0; 519 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist); 520 521 if (vp->v_type == VREG) { 522 if (np->n_flag & NMODIFIED) { 523 #ifndef NFS_V2_ONLY 524 if (NFS_ISV3(vp)) { 525 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0); 526 np->n_flag &= ~NMODIFIED; 527 } else 528 #endif 529 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1); 530 NFS_INVALIDATE_ATTRCACHE(np); 531 } 532 if (np->n_flag & NWRITEERR) { 533 np->n_flag &= ~NWRITEERR; 534 error = np->n_error; 535 } 536 } 537 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 538 return (error); 539 } 540 541 /* 542 * nfs getattr call from vfs. 543 */ 544 int 545 nfs_getattr(void *v) 546 { 547 struct vop_getattr_args /* { 548 struct vnode *a_vp; 549 struct vattr *a_vap; 550 kauth_cred_t a_cred; 551 } */ *ap = v; 552 struct vnode *vp = ap->a_vp; 553 struct nfsnode *np = VTONFS(vp); 554 char *cp; 555 u_int32_t *tl; 556 int32_t t1, t2; 557 char *bpos, *dpos; 558 int error = 0; 559 struct mbuf *mreq, *mrep, *md, *mb; 560 const int v3 = NFS_ISV3(vp); 561 562 /* 563 * Update local times for special files. 564 */ 565 if (np->n_flag & (NACC | NUPD)) 566 np->n_flag |= NCHG; 567 568 /* 569 * if we have delayed truncation, do it now. 570 */ 571 nfs_delayedtruncate(vp); 572 573 /* 574 * First look in the cache. 575 */ 576 if (nfs_getattrcache(vp, ap->a_vap) == 0) 577 return (0); 578 nfsstats.rpccnt[NFSPROC_GETATTR]++; 579 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3)); 580 nfsm_fhtom(np, v3); 581 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred); 582 if (!error) { 583 nfsm_loadattr(vp, ap->a_vap, 0); 584 if (vp->v_type == VDIR && 585 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ) 586 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ; 587 } 588 nfsm_reqdone; 589 return (error); 590 } 591 592 /* 593 * nfs setattr call. 594 */ 595 int 596 nfs_setattr(void *v) 597 { 598 struct vop_setattr_args /* { 599 struct vnodeop_desc *a_desc; 600 struct vnode *a_vp; 601 struct vattr *a_vap; 602 kauth_cred_t a_cred; 603 } */ *ap = v; 604 struct vnode *vp = ap->a_vp; 605 struct nfsnode *np = VTONFS(vp); 606 struct vattr *vap = ap->a_vap; 607 int error = 0; 608 u_quad_t tsize = 0; 609 610 /* 611 * Setting of flags is not supported. 612 */ 613 if (vap->va_flags != VNOVAL) 614 return (EOPNOTSUPP); 615 616 /* 617 * Disallow write attempts if the filesystem is mounted read-only. 618 */ 619 if ((vap->va_uid != (uid_t)VNOVAL || 620 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 621 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 622 (vp->v_mount->mnt_flag & MNT_RDONLY)) 623 return (EROFS); 624 if (vap->va_size != VNOVAL) { 625 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 626 return EFBIG; 627 } 628 switch (vp->v_type) { 629 case VDIR: 630 return (EISDIR); 631 case VCHR: 632 case VBLK: 633 case VSOCK: 634 case VFIFO: 635 if (vap->va_mtime.tv_sec == VNOVAL && 636 vap->va_atime.tv_sec == VNOVAL && 637 vap->va_mode == (mode_t)VNOVAL && 638 vap->va_uid == (uid_t)VNOVAL && 639 vap->va_gid == (gid_t)VNOVAL) 640 return (0); 641 vap->va_size = VNOVAL; 642 break; 643 default: 644 /* 645 * Disallow write attempts if the filesystem is 646 * mounted read-only. 647 */ 648 if (vp->v_mount->mnt_flag & MNT_RDONLY) 649 return (EROFS); 650 genfs_node_wrlock(vp); 651 uvm_vnp_setsize(vp, vap->va_size); 652 tsize = np->n_size; 653 np->n_size = vap->va_size; 654 if (vap->va_size == 0) 655 error = nfs_vinvalbuf(vp, 0, 656 ap->a_cred, curlwp, 1); 657 else 658 error = nfs_vinvalbuf(vp, V_SAVE, 659 ap->a_cred, curlwp, 1); 660 if (error) { 661 uvm_vnp_setsize(vp, tsize); 662 genfs_node_unlock(vp); 663 return (error); 664 } 665 np->n_vattr->va_size = vap->va_size; 666 } 667 } else { 668 /* 669 * flush files before setattr because a later write of 670 * cached data might change timestamps or reset sugid bits 671 */ 672 if ((vap->va_mtime.tv_sec != VNOVAL || 673 vap->va_atime.tv_sec != VNOVAL || 674 vap->va_mode != VNOVAL) && 675 vp->v_type == VREG && 676 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, 677 curlwp, 1)) == EINTR) 678 return (error); 679 } 680 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp); 681 if (vap->va_size != VNOVAL) { 682 if (error) { 683 np->n_size = np->n_vattr->va_size = tsize; 684 uvm_vnp_setsize(vp, np->n_size); 685 } 686 genfs_node_unlock(vp); 687 } 688 VN_KNOTE(vp, NOTE_ATTRIB); 689 return (error); 690 } 691 692 /* 693 * Do an nfs setattr rpc. 694 */ 695 int 696 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l) 697 { 698 struct nfsv2_sattr *sp; 699 char *cp; 700 int32_t t1, t2; 701 char *bpos, *dpos; 702 u_int32_t *tl; 703 int error = 0; 704 struct mbuf *mreq, *mrep, *md, *mb; 705 const int v3 = NFS_ISV3(vp); 706 struct nfsnode *np = VTONFS(vp); 707 #ifndef NFS_V2_ONLY 708 int wccflag = NFSV3_WCCRATTR; 709 char *cp2; 710 #endif 711 712 nfsstats.rpccnt[NFSPROC_SETATTR]++; 713 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 714 nfsm_fhtom(np, v3); 715 #ifndef NFS_V2_ONLY 716 if (v3) { 717 nfsm_v3attrbuild(vap, true); 718 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 719 *tl = nfs_false; 720 } else { 721 #endif 722 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 723 if (vap->va_mode == (mode_t)VNOVAL) 724 sp->sa_mode = nfs_xdrneg1; 725 else 726 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 727 if (vap->va_uid == (uid_t)VNOVAL) 728 sp->sa_uid = nfs_xdrneg1; 729 else 730 sp->sa_uid = txdr_unsigned(vap->va_uid); 731 if (vap->va_gid == (gid_t)VNOVAL) 732 sp->sa_gid = nfs_xdrneg1; 733 else 734 sp->sa_gid = txdr_unsigned(vap->va_gid); 735 sp->sa_size = txdr_unsigned(vap->va_size); 736 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 737 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 738 #ifndef NFS_V2_ONLY 739 } 740 #endif 741 nfsm_request(np, NFSPROC_SETATTR, l, cred); 742 #ifndef NFS_V2_ONLY 743 if (v3) { 744 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 745 } else 746 #endif 747 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 748 nfsm_reqdone; 749 return (error); 750 } 751 752 /* 753 * nfs lookup call, one step at a time... 754 * First look in cache 755 * If not found, do the rpc. 756 */ 757 int 758 nfs_lookup(void *v) 759 { 760 struct vop_lookup_args /* { 761 struct vnodeop_desc *a_desc; 762 struct vnode *a_dvp; 763 struct vnode **a_vpp; 764 struct componentname *a_cnp; 765 } */ *ap = v; 766 struct componentname *cnp = ap->a_cnp; 767 struct vnode *dvp = ap->a_dvp; 768 struct vnode **vpp = ap->a_vpp; 769 int flags; 770 struct vnode *newvp; 771 u_int32_t *tl; 772 char *cp; 773 int32_t t1, t2; 774 char *bpos, *dpos, *cp2; 775 struct mbuf *mreq, *mrep, *md, *mb; 776 long len; 777 nfsfh_t *fhp; 778 struct nfsnode *np; 779 int error = 0, attrflag, fhsize; 780 const int v3 = NFS_ISV3(dvp); 781 782 flags = cnp->cn_flags; 783 784 *vpp = NULLVP; 785 newvp = NULLVP; 786 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 787 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 788 return (EROFS); 789 if (dvp->v_type != VDIR) 790 return (ENOTDIR); 791 792 /* 793 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves. 794 */ 795 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 796 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 797 if (error) 798 return error; 799 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) 800 return EISDIR; 801 vref(dvp); 802 *vpp = dvp; 803 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 804 cnp->cn_flags |= SAVENAME; 805 return 0; 806 } 807 808 np = VTONFS(dvp); 809 810 /* 811 * Before performing an RPC, check the name cache to see if 812 * the directory/name pair we are looking for is known already. 813 * If the directory/name pair is found in the name cache, 814 * we have to ensure the directory has not changed from 815 * the time the cache entry has been created. If it has, 816 * the cache entry has to be ignored. 817 */ 818 error = cache_lookup_raw(dvp, vpp, cnp); 819 KASSERT(dvp != *vpp); 820 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0); 821 if (error >= 0) { 822 struct vattr vattr; 823 int err2; 824 825 if (error && error != ENOENT) { 826 *vpp = NULLVP; 827 return error; 828 } 829 830 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 831 if (err2 != 0) { 832 if (error == 0) 833 vrele(*vpp); 834 *vpp = NULLVP; 835 return err2; 836 } 837 838 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) 839 || timespeccmp(&vattr.va_mtime, 840 &VTONFS(dvp)->n_nctime, !=)) { 841 if (error == 0) { 842 vrele(*vpp); 843 *vpp = NULLVP; 844 } 845 cache_purge1(dvp, NULL, PURGE_CHILDREN); 846 timespecclear(&np->n_nctime); 847 goto dorpc; 848 } 849 850 if (error == ENOENT) { 851 goto noentry; 852 } 853 854 /* 855 * investigate the vnode returned by cache_lookup_raw. 856 * if it isn't appropriate, do an rpc. 857 */ 858 newvp = *vpp; 859 if ((flags & ISDOTDOT) != 0) { 860 VOP_UNLOCK(dvp); 861 } 862 error = vn_lock(newvp, LK_EXCLUSIVE); 863 if ((flags & ISDOTDOT) != 0) { 864 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 865 } 866 if (error != 0) { 867 /* newvp has been reclaimed. */ 868 vrele(newvp); 869 *vpp = NULLVP; 870 goto dorpc; 871 } 872 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) 873 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 874 nfsstats.lookupcache_hits++; 875 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 876 cnp->cn_flags |= SAVENAME; 877 KASSERT(newvp->v_type != VNON); 878 return (0); 879 } 880 cache_purge1(newvp, NULL, PURGE_PARENTS); 881 vput(newvp); 882 *vpp = NULLVP; 883 } 884 dorpc: 885 #if 0 886 /* 887 * because nfsv3 has the same CREATE semantics as ours, 888 * we don't have to perform LOOKUPs beforehand. 889 * 890 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL. 891 * XXX although we have no way to know if O_EXCL is requested or not. 892 */ 893 894 if (v3 && cnp->cn_nameiop == CREATE && 895 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN && 896 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 897 cnp->cn_flags |= SAVENAME; 898 return (EJUSTRETURN); 899 } 900 #endif /* 0 */ 901 902 error = 0; 903 newvp = NULLVP; 904 nfsstats.lookupcache_misses++; 905 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 906 len = cnp->cn_namelen; 907 nfsm_reqhead(np, NFSPROC_LOOKUP, 908 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 909 nfsm_fhtom(np, v3); 910 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 911 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred); 912 if (error) { 913 nfsm_postop_attr(dvp, attrflag, 0); 914 m_freem(mrep); 915 goto nfsmout; 916 } 917 nfsm_getfh(fhp, fhsize, v3); 918 919 /* 920 * Handle RENAME case... 921 */ 922 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { 923 if (NFS_CMPFH(np, fhp, fhsize)) { 924 m_freem(mrep); 925 return (EISDIR); 926 } 927 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 928 if (error) { 929 m_freem(mrep); 930 return error; 931 } 932 newvp = NFSTOV(np); 933 #ifndef NFS_V2_ONLY 934 if (v3) { 935 nfsm_postop_attr(newvp, attrflag, 0); 936 nfsm_postop_attr(dvp, attrflag, 0); 937 } else 938 #endif 939 nfsm_loadattr(newvp, (struct vattr *)0, 0); 940 *vpp = newvp; 941 m_freem(mrep); 942 cnp->cn_flags |= SAVENAME; 943 goto validate; 944 } 945 946 /* 947 * The postop attr handling is duplicated for each if case, 948 * because it should be done while dvp is locked (unlocking 949 * dvp is different for each case). 950 */ 951 952 if (NFS_CMPFH(np, fhp, fhsize)) { 953 /* 954 * as we handle "." lookup locally, this should be 955 * a broken server. 956 */ 957 vref(dvp); 958 newvp = dvp; 959 #ifndef NFS_V2_ONLY 960 if (v3) { 961 nfsm_postop_attr(newvp, attrflag, 0); 962 nfsm_postop_attr(dvp, attrflag, 0); 963 } else 964 #endif 965 nfsm_loadattr(newvp, (struct vattr *)0, 0); 966 } else if (flags & ISDOTDOT) { 967 /* 968 * ".." lookup 969 */ 970 VOP_UNLOCK(dvp); 971 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 972 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 973 if (error) { 974 m_freem(mrep); 975 return error; 976 } 977 newvp = NFSTOV(np); 978 979 #ifndef NFS_V2_ONLY 980 if (v3) { 981 nfsm_postop_attr(newvp, attrflag, 0); 982 nfsm_postop_attr(dvp, attrflag, 0); 983 } else 984 #endif 985 nfsm_loadattr(newvp, (struct vattr *)0, 0); 986 } else { 987 /* 988 * Other lookups. 989 */ 990 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 991 if (error) { 992 m_freem(mrep); 993 return error; 994 } 995 newvp = NFSTOV(np); 996 #ifndef NFS_V2_ONLY 997 if (v3) { 998 nfsm_postop_attr(newvp, attrflag, 0); 999 nfsm_postop_attr(dvp, attrflag, 0); 1000 } else 1001 #endif 1002 nfsm_loadattr(newvp, (struct vattr *)0, 0); 1003 } 1004 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) 1005 cnp->cn_flags |= SAVENAME; 1006 if ((cnp->cn_flags & MAKEENTRY) && 1007 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) { 1008 nfs_cache_enter(dvp, newvp, cnp); 1009 } 1010 *vpp = newvp; 1011 nfsm_reqdone; 1012 if (error) { 1013 /* 1014 * We get here only because of errors returned by 1015 * the RPC. Otherwise we'll have returned above 1016 * (the nfsm_* macros will jump to nfsm_reqdone 1017 * on error). 1018 */ 1019 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && 1020 cnp->cn_nameiop != CREATE) { 1021 nfs_cache_enter(dvp, NULL, cnp); 1022 } 1023 if (newvp != NULLVP) { 1024 if (newvp == dvp) { 1025 vrele(newvp); 1026 } else { 1027 vput(newvp); 1028 } 1029 } 1030 noentry: 1031 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && 1032 (flags & ISLASTCN) && error == ENOENT) { 1033 if (dvp->v_mount->mnt_flag & MNT_RDONLY) { 1034 error = EROFS; 1035 } else { 1036 error = EJUSTRETURN; 1037 cnp->cn_flags |= SAVENAME; 1038 } 1039 } 1040 *vpp = NULL; 1041 return error; 1042 } 1043 1044 validate: 1045 /* 1046 * make sure we have valid type and size. 1047 */ 1048 1049 newvp = *vpp; 1050 if (newvp->v_type == VNON) { 1051 struct vattr vattr; /* dummy */ 1052 1053 KASSERT(VTONFS(newvp)->n_attrstamp == 0); 1054 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred); 1055 if (error) { 1056 vput(newvp); 1057 *vpp = NULL; 1058 } 1059 } 1060 1061 return error; 1062 } 1063 1064 /* 1065 * nfs read call. 1066 * Just call nfs_bioread() to do the work. 1067 */ 1068 int 1069 nfs_read(void *v) 1070 { 1071 struct vop_read_args /* { 1072 struct vnode *a_vp; 1073 struct uio *a_uio; 1074 int a_ioflag; 1075 kauth_cred_t a_cred; 1076 } */ *ap = v; 1077 struct vnode *vp = ap->a_vp; 1078 1079 if (vp->v_type != VREG) 1080 return EISDIR; 1081 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0)); 1082 } 1083 1084 /* 1085 * nfs readlink call 1086 */ 1087 int 1088 nfs_readlink(void *v) 1089 { 1090 struct vop_readlink_args /* { 1091 struct vnode *a_vp; 1092 struct uio *a_uio; 1093 kauth_cred_t a_cred; 1094 } */ *ap = v; 1095 struct vnode *vp = ap->a_vp; 1096 struct nfsnode *np = VTONFS(vp); 1097 1098 if (vp->v_type != VLNK) 1099 return (EPERM); 1100 1101 if (np->n_rcred != NULL) { 1102 kauth_cred_free(np->n_rcred); 1103 } 1104 np->n_rcred = ap->a_cred; 1105 kauth_cred_hold(np->n_rcred); 1106 1107 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0)); 1108 } 1109 1110 /* 1111 * Do a readlink rpc. 1112 * Called by nfs_doio() from below the buffer cache. 1113 */ 1114 int 1115 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 1116 { 1117 u_int32_t *tl; 1118 char *cp; 1119 int32_t t1, t2; 1120 char *bpos, *dpos, *cp2; 1121 int error = 0; 1122 uint32_t len; 1123 struct mbuf *mreq, *mrep, *md, *mb; 1124 const int v3 = NFS_ISV3(vp); 1125 struct nfsnode *np = VTONFS(vp); 1126 #ifndef NFS_V2_ONLY 1127 int attrflag; 1128 #endif 1129 1130 nfsstats.rpccnt[NFSPROC_READLINK]++; 1131 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3)); 1132 nfsm_fhtom(np, v3); 1133 nfsm_request(np, NFSPROC_READLINK, curlwp, cred); 1134 #ifndef NFS_V2_ONLY 1135 if (v3) 1136 nfsm_postop_attr(vp, attrflag, 0); 1137 #endif 1138 if (!error) { 1139 #ifndef NFS_V2_ONLY 1140 if (v3) { 1141 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED); 1142 len = fxdr_unsigned(uint32_t, *tl); 1143 if (len > MAXPATHLEN) { 1144 /* 1145 * this pathname is too long for us. 1146 */ 1147 m_freem(mrep); 1148 /* Solaris returns EINVAL. should we follow? */ 1149 error = ENAMETOOLONG; 1150 goto nfsmout; 1151 } 1152 } else 1153 #endif 1154 { 1155 nfsm_strsiz(len, NFS_MAXPATHLEN); 1156 } 1157 nfsm_mtouio(uiop, len); 1158 } 1159 nfsm_reqdone; 1160 return (error); 1161 } 1162 1163 /* 1164 * nfs read rpc call 1165 * Ditto above 1166 */ 1167 int 1168 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1169 { 1170 u_int32_t *tl; 1171 char *cp; 1172 int32_t t1, t2; 1173 char *bpos, *dpos, *cp2; 1174 struct mbuf *mreq, *mrep, *md, *mb; 1175 struct nfsmount *nmp; 1176 int error = 0, len, retlen, tsiz, eof, byte_count; 1177 const int v3 = NFS_ISV3(vp); 1178 struct nfsnode *np = VTONFS(vp); 1179 #ifndef NFS_V2_ONLY 1180 int attrflag; 1181 #endif 1182 1183 #ifndef nolint 1184 eof = 0; 1185 #endif 1186 nmp = VFSTONFS(vp->v_mount); 1187 tsiz = uiop->uio_resid; 1188 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1189 return (EFBIG); 1190 iostat_busy(nmp->nm_stats); 1191 byte_count = 0; /* count bytes actually transferred */ 1192 while (tsiz > 0) { 1193 nfsstats.rpccnt[NFSPROC_READ]++; 1194 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1195 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1196 nfsm_fhtom(np, v3); 1197 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1198 #ifndef NFS_V2_ONLY 1199 if (v3) { 1200 txdr_hyper(uiop->uio_offset, tl); 1201 *(tl + 2) = txdr_unsigned(len); 1202 } else 1203 #endif 1204 { 1205 *tl++ = txdr_unsigned(uiop->uio_offset); 1206 *tl++ = txdr_unsigned(len); 1207 *tl = 0; 1208 } 1209 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred); 1210 #ifndef NFS_V2_ONLY 1211 if (v3) { 1212 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC); 1213 if (error) { 1214 m_freem(mrep); 1215 goto nfsmout; 1216 } 1217 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1218 eof = fxdr_unsigned(int, *(tl + 1)); 1219 } else 1220 #endif 1221 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1222 nfsm_strsiz(retlen, nmp->nm_rsize); 1223 nfsm_mtouio(uiop, retlen); 1224 m_freem(mrep); 1225 tsiz -= retlen; 1226 byte_count += retlen; 1227 #ifndef NFS_V2_ONLY 1228 if (v3) { 1229 if (eof || retlen == 0) 1230 tsiz = 0; 1231 } else 1232 #endif 1233 if (retlen < len) 1234 tsiz = 0; 1235 } 1236 nfsmout: 1237 iostat_unbusy(nmp->nm_stats, byte_count, 1); 1238 return (error); 1239 } 1240 1241 struct nfs_writerpc_context { 1242 kmutex_t nwc_lock; 1243 kcondvar_t nwc_cv; 1244 int nwc_mbufcount; 1245 }; 1246 1247 /* 1248 * free mbuf used to refer protected pages while write rpc call. 1249 * called at splvm. 1250 */ 1251 static void 1252 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg) 1253 { 1254 struct nfs_writerpc_context *ctx = arg; 1255 1256 KASSERT(m != NULL); 1257 KASSERT(ctx != NULL); 1258 pool_cache_put(mb_cache, m); 1259 mutex_enter(&ctx->nwc_lock); 1260 if (--ctx->nwc_mbufcount == 0) { 1261 cv_signal(&ctx->nwc_cv); 1262 } 1263 mutex_exit(&ctx->nwc_lock); 1264 } 1265 1266 /* 1267 * nfs write call 1268 */ 1269 int 1270 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp) 1271 { 1272 u_int32_t *tl; 1273 char *cp; 1274 int32_t t1, t2; 1275 char *bpos, *dpos; 1276 struct mbuf *mreq, *mrep, *md, *mb; 1277 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1278 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR; 1279 const int v3 = NFS_ISV3(vp); 1280 int committed = NFSV3WRITE_FILESYNC; 1281 struct nfsnode *np = VTONFS(vp); 1282 struct nfs_writerpc_context ctx; 1283 int byte_count; 1284 size_t origresid; 1285 #ifndef NFS_V2_ONLY 1286 char *cp2; 1287 int rlen, commit; 1288 #endif 1289 1290 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM); 1291 cv_init(&ctx.nwc_cv, "nfsmblk"); 1292 ctx.nwc_mbufcount = 1; 1293 1294 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 1295 panic("writerpc readonly vp %p", vp); 1296 } 1297 1298 #ifdef DIAGNOSTIC 1299 if (uiop->uio_iovcnt != 1) 1300 panic("nfs: writerpc iovcnt > 1"); 1301 #endif 1302 tsiz = uiop->uio_resid; 1303 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1304 return (EFBIG); 1305 retry: 1306 origresid = uiop->uio_resid; 1307 KASSERT(origresid == uiop->uio_iov->iov_len); 1308 iostat_busy(nmp->nm_stats); 1309 byte_count = 0; /* count of bytes actually written */ 1310 while (tsiz > 0) { 1311 uint32_t datalen; /* data bytes need to be allocated in mbuf */ 1312 uint32_t backup; 1313 bool stalewriteverf = false; 1314 1315 nfsstats.rpccnt[NFSPROC_WRITE]++; 1316 len = min(tsiz, nmp->nm_wsize); 1317 datalen = pageprotected ? 0 : nfsm_rndup(len); 1318 nfsm_reqhead(np, NFSPROC_WRITE, 1319 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen); 1320 nfsm_fhtom(np, v3); 1321 #ifndef NFS_V2_ONLY 1322 if (v3) { 1323 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1324 txdr_hyper(uiop->uio_offset, tl); 1325 tl += 2; 1326 *tl++ = txdr_unsigned(len); 1327 *tl++ = txdr_unsigned(*iomode); 1328 *tl = txdr_unsigned(len); 1329 } else 1330 #endif 1331 { 1332 u_int32_t x; 1333 1334 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1335 /* Set both "begin" and "current" to non-garbage. */ 1336 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1337 *tl++ = x; /* "begin offset" */ 1338 *tl++ = x; /* "current offset" */ 1339 x = txdr_unsigned(len); 1340 *tl++ = x; /* total to this offset */ 1341 *tl = x; /* size of this write */ 1342 1343 } 1344 if (pageprotected) { 1345 /* 1346 * since we know pages can't be modified during i/o, 1347 * no need to copy them for us. 1348 */ 1349 struct mbuf *m; 1350 struct iovec *iovp = uiop->uio_iov; 1351 1352 m = m_get(M_WAIT, MT_DATA); 1353 MCLAIM(m, &nfs_mowner); 1354 MEXTADD(m, iovp->iov_base, len, M_MBUF, 1355 nfs_writerpc_extfree, &ctx); 1356 m->m_flags |= M_EXT_ROMAP; 1357 m->m_len = len; 1358 mb->m_next = m; 1359 /* 1360 * no need to maintain mb and bpos here 1361 * because no one care them later. 1362 */ 1363 #if 0 1364 mb = m; 1365 bpos = mtod(void *, mb) + mb->m_len; 1366 #endif 1367 UIO_ADVANCE(uiop, len); 1368 uiop->uio_offset += len; 1369 mutex_enter(&ctx.nwc_lock); 1370 ctx.nwc_mbufcount++; 1371 mutex_exit(&ctx.nwc_lock); 1372 nfs_zeropad(mb, 0, nfsm_padlen(len)); 1373 } else { 1374 nfsm_uiotom(uiop, len); 1375 } 1376 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred); 1377 #ifndef NFS_V2_ONLY 1378 if (v3) { 1379 wccflag = NFSV3_WCCCHK; 1380 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error); 1381 1382 /* Diagnostic check: The KASSERT hits when PR 42455 1383 * is reproduced. 1384 */ 1385 if (ctx.nwc_mbufcount >= 2) { 1386 KASSERT(mb->m_next != NULL); 1387 } 1388 1389 if (!error) { 1390 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1391 + NFSX_V3WRITEVERF); 1392 rlen = fxdr_unsigned(int, *tl++); 1393 if (rlen == 0) { 1394 error = NFSERR_IO; 1395 m_freem(mrep); 1396 break; 1397 } else if (rlen < len) { 1398 backup = len - rlen; 1399 UIO_ADVANCE(uiop, -backup); 1400 uiop->uio_offset -= backup; 1401 len = rlen; 1402 } 1403 commit = fxdr_unsigned(int, *tl++); 1404 1405 /* 1406 * Return the lowest committment level 1407 * obtained by any of the RPCs. 1408 */ 1409 if (committed == NFSV3WRITE_FILESYNC) 1410 committed = commit; 1411 else if (committed == NFSV3WRITE_DATASYNC && 1412 commit == NFSV3WRITE_UNSTABLE) 1413 committed = commit; 1414 mutex_enter(&nmp->nm_lock); 1415 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){ 1416 memcpy(nmp->nm_writeverf, tl, 1417 NFSX_V3WRITEVERF); 1418 nmp->nm_iflag |= NFSMNT_HASWRITEVERF; 1419 } else if ((nmp->nm_iflag & 1420 NFSMNT_STALEWRITEVERF) || 1421 memcmp(tl, nmp->nm_writeverf, 1422 NFSX_V3WRITEVERF)) { 1423 memcpy(nmp->nm_writeverf, tl, 1424 NFSX_V3WRITEVERF); 1425 /* 1426 * note NFSMNT_STALEWRITEVERF 1427 * if we're the first thread to 1428 * notice it. 1429 */ 1430 if ((nmp->nm_iflag & 1431 NFSMNT_STALEWRITEVERF) == 0) { 1432 stalewriteverf = true; 1433 nmp->nm_iflag |= 1434 NFSMNT_STALEWRITEVERF; 1435 } 1436 } 1437 mutex_exit(&nmp->nm_lock); 1438 } 1439 } else 1440 #endif 1441 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC); 1442 if (wccflag) 1443 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime; 1444 m_freem(mrep); 1445 if (error) 1446 break; 1447 tsiz -= len; 1448 byte_count += len; 1449 if (stalewriteverf) { 1450 *stalewriteverfp = true; 1451 stalewriteverf = false; 1452 if (committed == NFSV3WRITE_UNSTABLE && 1453 len != origresid) { 1454 /* 1455 * if our write requests weren't atomic but 1456 * unstable, datas in previous iterations 1457 * might have already been lost now. 1458 * then, we should resend them to nfsd. 1459 */ 1460 backup = origresid - tsiz; 1461 UIO_ADVANCE(uiop, -backup); 1462 uiop->uio_offset -= backup; 1463 tsiz = origresid; 1464 goto retry; 1465 } 1466 } 1467 } 1468 nfsmout: 1469 iostat_unbusy(nmp->nm_stats, byte_count, 0); 1470 if (pageprotected) { 1471 /* 1472 * wait until mbufs go away. 1473 * retransmitted mbufs can survive longer than rpc requests 1474 * themselves. 1475 */ 1476 mutex_enter(&ctx.nwc_lock); 1477 ctx.nwc_mbufcount--; 1478 while (ctx.nwc_mbufcount > 0) { 1479 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock); 1480 } 1481 mutex_exit(&ctx.nwc_lock); 1482 } 1483 mutex_destroy(&ctx.nwc_lock); 1484 cv_destroy(&ctx.nwc_cv); 1485 *iomode = committed; 1486 if (error) 1487 uiop->uio_resid = tsiz; 1488 return (error); 1489 } 1490 1491 /* 1492 * nfs mknod rpc 1493 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1494 * mode set to specify the file type and the size field for rdev. 1495 */ 1496 int 1497 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) 1498 { 1499 struct nfsv2_sattr *sp; 1500 u_int32_t *tl; 1501 char *cp; 1502 int32_t t1, t2; 1503 struct vnode *newvp = (struct vnode *)0; 1504 struct nfsnode *dnp, *np; 1505 char *cp2; 1506 char *bpos, *dpos; 1507 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1508 struct mbuf *mreq, *mrep, *md, *mb; 1509 u_int32_t rdev; 1510 const int v3 = NFS_ISV3(dvp); 1511 1512 if (vap->va_type == VCHR || vap->va_type == VBLK) 1513 rdev = txdr_unsigned(vap->va_rdev); 1514 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1515 rdev = nfs_xdrneg1; 1516 else { 1517 VOP_ABORTOP(dvp, cnp); 1518 vput(dvp); 1519 return (EOPNOTSUPP); 1520 } 1521 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1522 dnp = VTONFS(dvp); 1523 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1524 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1525 nfsm_fhtom(dnp, v3); 1526 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1527 #ifndef NFS_V2_ONLY 1528 if (v3) { 1529 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1530 *tl++ = vtonfsv3_type(vap->va_type); 1531 nfsm_v3attrbuild(vap, false); 1532 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1533 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1534 *tl++ = txdr_unsigned(major(vap->va_rdev)); 1535 *tl = txdr_unsigned(minor(vap->va_rdev)); 1536 } 1537 } else 1538 #endif 1539 { 1540 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1541 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1542 sp->sa_uid = nfs_xdrneg1; 1543 sp->sa_gid = nfs_xdrneg1; 1544 sp->sa_size = rdev; 1545 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1546 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1547 } 1548 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred); 1549 if (!error) { 1550 nfsm_mtofh(dvp, newvp, v3, gotvp); 1551 if (!gotvp) { 1552 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1553 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1554 if (!error) 1555 newvp = NFSTOV(np); 1556 } 1557 } 1558 #ifndef NFS_V2_ONLY 1559 if (v3) 1560 nfsm_wcc_data(dvp, wccflag, 0, !error); 1561 #endif 1562 nfsm_reqdone; 1563 if (error) { 1564 if (newvp) 1565 vput(newvp); 1566 } else { 1567 if (cnp->cn_flags & MAKEENTRY) 1568 nfs_cache_enter(dvp, newvp, cnp); 1569 *vpp = newvp; 1570 } 1571 PNBUF_PUT(cnp->cn_pnbuf); 1572 VTONFS(dvp)->n_flag |= NMODIFIED; 1573 if (!wccflag) 1574 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1575 vput(dvp); 1576 return (error); 1577 } 1578 1579 /* 1580 * nfs mknod vop 1581 * just call nfs_mknodrpc() to do the work. 1582 */ 1583 /* ARGSUSED */ 1584 int 1585 nfs_mknod(void *v) 1586 { 1587 struct vop_mknod_args /* { 1588 struct vnode *a_dvp; 1589 struct vnode **a_vpp; 1590 struct componentname *a_cnp; 1591 struct vattr *a_vap; 1592 } */ *ap = v; 1593 struct vnode *dvp = ap->a_dvp; 1594 struct componentname *cnp = ap->a_cnp; 1595 int error; 1596 1597 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap); 1598 VN_KNOTE(dvp, NOTE_WRITE); 1599 if (error == 0 || error == EEXIST) 1600 cache_purge1(dvp, cnp, 0); 1601 return (error); 1602 } 1603 1604 /* 1605 * nfs file create call 1606 */ 1607 int 1608 nfs_create(void *v) 1609 { 1610 struct vop_create_args /* { 1611 struct vnode *a_dvp; 1612 struct vnode **a_vpp; 1613 struct componentname *a_cnp; 1614 struct vattr *a_vap; 1615 } */ *ap = v; 1616 struct vnode *dvp = ap->a_dvp; 1617 struct vattr *vap = ap->a_vap; 1618 struct componentname *cnp = ap->a_cnp; 1619 struct nfsv2_sattr *sp; 1620 u_int32_t *tl; 1621 char *cp; 1622 int32_t t1, t2; 1623 struct nfsnode *dnp, *np = (struct nfsnode *)0; 1624 struct vnode *newvp = (struct vnode *)0; 1625 char *bpos, *dpos, *cp2; 1626 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1627 struct mbuf *mreq, *mrep, *md, *mb; 1628 const int v3 = NFS_ISV3(dvp); 1629 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED; 1630 1631 /* 1632 * Oops, not for me.. 1633 */ 1634 if (vap->va_type == VSOCK) 1635 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1636 1637 KASSERT(vap->va_type == VREG); 1638 1639 #ifdef VA_EXCLUSIVE 1640 if (vap->va_vaflags & VA_EXCLUSIVE) { 1641 excl_mode = NFSV3CREATE_EXCLUSIVE; 1642 } 1643 #endif 1644 again: 1645 error = 0; 1646 nfsstats.rpccnt[NFSPROC_CREATE]++; 1647 dnp = VTONFS(dvp); 1648 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1649 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1650 nfsm_fhtom(dnp, v3); 1651 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1652 #ifndef NFS_V2_ONLY 1653 if (v3) { 1654 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1655 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1656 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1657 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1658 *tl++ = arc4random(); 1659 *tl = arc4random(); 1660 } else { 1661 *tl = txdr_unsigned(excl_mode); 1662 nfsm_v3attrbuild(vap, false); 1663 } 1664 } else 1665 #endif 1666 { 1667 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1668 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1669 sp->sa_uid = nfs_xdrneg1; 1670 sp->sa_gid = nfs_xdrneg1; 1671 sp->sa_size = 0; 1672 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1673 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1674 } 1675 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred); 1676 if (!error) { 1677 nfsm_mtofh(dvp, newvp, v3, gotvp); 1678 if (!gotvp) { 1679 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1680 cnp->cn_namelen, cnp->cn_cred, curlwp, &np); 1681 if (!error) 1682 newvp = NFSTOV(np); 1683 } 1684 } 1685 #ifndef NFS_V2_ONLY 1686 if (v3) 1687 nfsm_wcc_data(dvp, wccflag, 0, !error); 1688 #endif 1689 nfsm_reqdone; 1690 if (error) { 1691 /* 1692 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 1693 */ 1694 if (v3 && error == ENOTSUP) { 1695 if (excl_mode == NFSV3CREATE_EXCLUSIVE) { 1696 excl_mode = NFSV3CREATE_GUARDED; 1697 goto again; 1698 } else if (excl_mode == NFSV3CREATE_GUARDED) { 1699 excl_mode = NFSV3CREATE_UNCHECKED; 1700 goto again; 1701 } 1702 } 1703 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) { 1704 struct timespec ts; 1705 1706 getnanotime(&ts); 1707 1708 /* 1709 * make sure that we'll update timestamps as 1710 * most server implementations use them to store 1711 * the create verifier. 1712 * 1713 * XXX it's better to use TOSERVER always. 1714 */ 1715 1716 if (vap->va_atime.tv_sec == VNOVAL) 1717 vap->va_atime = ts; 1718 if (vap->va_mtime.tv_sec == VNOVAL) 1719 vap->va_mtime = ts; 1720 1721 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp); 1722 } 1723 if (error == 0) { 1724 if (cnp->cn_flags & MAKEENTRY) 1725 nfs_cache_enter(dvp, newvp, cnp); 1726 else 1727 cache_purge1(dvp, cnp, 0); 1728 *ap->a_vpp = newvp; 1729 } else { 1730 if (newvp) 1731 vput(newvp); 1732 if (error == EEXIST) 1733 cache_purge1(dvp, cnp, 0); 1734 } 1735 PNBUF_PUT(cnp->cn_pnbuf); 1736 VTONFS(dvp)->n_flag |= NMODIFIED; 1737 if (!wccflag) 1738 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1739 VN_KNOTE(ap->a_dvp, NOTE_WRITE); 1740 vput(dvp); 1741 return (error); 1742 } 1743 1744 /* 1745 * nfs file remove call 1746 * To try and make nfs semantics closer to ufs semantics, a file that has 1747 * other processes using the vnode is renamed instead of removed and then 1748 * removed later on the last close. 1749 * - If v_usecount > 1 1750 * If a rename is not already in the works 1751 * call nfs_sillyrename() to set it up 1752 * else 1753 * do the remove rpc 1754 */ 1755 int 1756 nfs_remove(void *v) 1757 { 1758 struct vop_remove_args /* { 1759 struct vnodeop_desc *a_desc; 1760 struct vnode * a_dvp; 1761 struct vnode * a_vp; 1762 struct componentname * a_cnp; 1763 } */ *ap = v; 1764 struct vnode *vp = ap->a_vp; 1765 struct vnode *dvp = ap->a_dvp; 1766 struct componentname *cnp = ap->a_cnp; 1767 struct nfsnode *np = VTONFS(vp); 1768 int error = 0; 1769 struct vattr vattr; 1770 1771 #ifndef DIAGNOSTIC 1772 if ((cnp->cn_flags & HASBUF) == 0) 1773 panic("nfs_remove: no name"); 1774 if (vp->v_usecount < 1) 1775 panic("nfs_remove: bad v_usecount"); 1776 #endif 1777 if (vp->v_type == VDIR) 1778 error = EPERM; 1779 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1780 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && 1781 vattr.va_nlink > 1)) { 1782 /* 1783 * Purge the name cache so that the chance of a lookup for 1784 * the name succeeding while the remove is in progress is 1785 * minimized. Without node locking it can still happen, such 1786 * that an I/O op returns ESTALE, but since you get this if 1787 * another host removes the file.. 1788 */ 1789 cache_purge(vp); 1790 /* 1791 * throw away biocache buffers, mainly to avoid 1792 * unnecessary delayed writes later. 1793 */ 1794 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1); 1795 /* Do the rpc */ 1796 if (error != EINTR) 1797 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1798 cnp->cn_namelen, cnp->cn_cred, curlwp); 1799 } else if (!np->n_sillyrename) 1800 error = nfs_sillyrename(dvp, vp, cnp, false); 1801 PNBUF_PUT(cnp->cn_pnbuf); 1802 if (!error && nfs_getattrcache(vp, &vattr) == 0 && 1803 vattr.va_nlink == 1) { 1804 np->n_flag |= NREMOVED; 1805 } 1806 NFS_INVALIDATE_ATTRCACHE(np); 1807 VN_KNOTE(vp, NOTE_DELETE); 1808 VN_KNOTE(dvp, NOTE_WRITE); 1809 if (dvp == vp) 1810 vrele(vp); 1811 else 1812 vput(vp); 1813 vput(dvp); 1814 return (error); 1815 } 1816 1817 /* 1818 * nfs file remove rpc called from nfs_inactive 1819 */ 1820 int 1821 nfs_removeit(struct sillyrename *sp) 1822 { 1823 1824 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, 1825 (struct lwp *)0)); 1826 } 1827 1828 /* 1829 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1830 */ 1831 int 1832 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l) 1833 { 1834 u_int32_t *tl; 1835 char *cp; 1836 #ifndef NFS_V2_ONLY 1837 int32_t t1; 1838 char *cp2; 1839 #endif 1840 int32_t t2; 1841 char *bpos, *dpos; 1842 int error = 0, wccflag = NFSV3_WCCRATTR; 1843 struct mbuf *mreq, *mrep, *md, *mb; 1844 const int v3 = NFS_ISV3(dvp); 1845 int rexmit = 0; 1846 struct nfsnode *dnp = VTONFS(dvp); 1847 1848 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1849 nfsm_reqhead(dnp, NFSPROC_REMOVE, 1850 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1851 nfsm_fhtom(dnp, v3); 1852 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1853 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit); 1854 #ifndef NFS_V2_ONLY 1855 if (v3) 1856 nfsm_wcc_data(dvp, wccflag, 0, !error); 1857 #endif 1858 nfsm_reqdone; 1859 VTONFS(dvp)->n_flag |= NMODIFIED; 1860 if (!wccflag) 1861 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 1862 /* 1863 * Kludge City: If the first reply to the remove rpc is lost.. 1864 * the reply to the retransmitted request will be ENOENT 1865 * since the file was in fact removed 1866 * Therefore, we cheat and return success. 1867 */ 1868 if (rexmit && error == ENOENT) 1869 error = 0; 1870 return (error); 1871 } 1872 1873 /* 1874 * nfs file rename call 1875 */ 1876 int 1877 nfs_rename(void *v) 1878 { 1879 struct vop_rename_args /* { 1880 struct vnode *a_fdvp; 1881 struct vnode *a_fvp; 1882 struct componentname *a_fcnp; 1883 struct vnode *a_tdvp; 1884 struct vnode *a_tvp; 1885 struct componentname *a_tcnp; 1886 } */ *ap = v; 1887 struct vnode *fvp = ap->a_fvp; 1888 struct vnode *tvp = ap->a_tvp; 1889 struct vnode *fdvp = ap->a_fdvp; 1890 struct vnode *tdvp = ap->a_tdvp; 1891 struct componentname *tcnp = ap->a_tcnp; 1892 struct componentname *fcnp = ap->a_fcnp; 1893 int error; 1894 1895 #ifndef DIAGNOSTIC 1896 if ((tcnp->cn_flags & HASBUF) == 0 || 1897 (fcnp->cn_flags & HASBUF) == 0) 1898 panic("nfs_rename: no name"); 1899 #endif 1900 /* Check for cross-device rename */ 1901 if ((fvp->v_mount != tdvp->v_mount) || 1902 (tvp && (fvp->v_mount != tvp->v_mount))) { 1903 error = EXDEV; 1904 goto out; 1905 } 1906 1907 /* 1908 * If the tvp exists and is in use, sillyrename it before doing the 1909 * rename of the new file over it. 1910 * 1911 * Have sillyrename use link instead of rename if possible, 1912 * so that we don't lose the file if the rename fails, and so 1913 * that there's no window when the "to" file doesn't exist. 1914 */ 1915 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1916 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) { 1917 VN_KNOTE(tvp, NOTE_DELETE); 1918 vput(tvp); 1919 tvp = NULL; 1920 } 1921 1922 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1923 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1924 curlwp); 1925 1926 VN_KNOTE(fdvp, NOTE_WRITE); 1927 VN_KNOTE(tdvp, NOTE_WRITE); 1928 if (error == 0 || error == EEXIST) { 1929 if (fvp->v_type == VDIR) 1930 cache_purge(fvp); 1931 else 1932 cache_purge1(fdvp, fcnp, 0); 1933 if (tvp != NULL && tvp->v_type == VDIR) 1934 cache_purge(tvp); 1935 else 1936 cache_purge1(tdvp, tcnp, 0); 1937 } 1938 out: 1939 if (tdvp == tvp) 1940 vrele(tdvp); 1941 else 1942 vput(tdvp); 1943 if (tvp) 1944 vput(tvp); 1945 vrele(fdvp); 1946 vrele(fvp); 1947 return (error); 1948 } 1949 1950 /* 1951 * nfs file rename rpc called from nfs_remove() above 1952 */ 1953 int 1954 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp) 1955 { 1956 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1957 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp)); 1958 } 1959 1960 /* 1961 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1962 */ 1963 int 1964 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, struct vnode *tdvp, const char *tnameptr, int tnamelen, kauth_cred_t cred, struct lwp *l) 1965 { 1966 u_int32_t *tl; 1967 char *cp; 1968 #ifndef NFS_V2_ONLY 1969 int32_t t1; 1970 char *cp2; 1971 #endif 1972 int32_t t2; 1973 char *bpos, *dpos; 1974 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1975 struct mbuf *mreq, *mrep, *md, *mb; 1976 const int v3 = NFS_ISV3(fdvp); 1977 int rexmit = 0; 1978 struct nfsnode *fdnp = VTONFS(fdvp); 1979 1980 nfsstats.rpccnt[NFSPROC_RENAME]++; 1981 nfsm_reqhead(fdnp, NFSPROC_RENAME, 1982 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1983 nfsm_rndup(tnamelen)); 1984 nfsm_fhtom(fdnp, v3); 1985 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1986 nfsm_fhtom(VTONFS(tdvp), v3); 1987 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1988 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit); 1989 #ifndef NFS_V2_ONLY 1990 if (v3) { 1991 nfsm_wcc_data(fdvp, fwccflag, 0, !error); 1992 nfsm_wcc_data(tdvp, twccflag, 0, !error); 1993 } 1994 #endif 1995 nfsm_reqdone; 1996 VTONFS(fdvp)->n_flag |= NMODIFIED; 1997 VTONFS(tdvp)->n_flag |= NMODIFIED; 1998 if (!fwccflag) 1999 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp)); 2000 if (!twccflag) 2001 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp)); 2002 /* 2003 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2004 */ 2005 if (rexmit && error == ENOENT) 2006 error = 0; 2007 return (error); 2008 } 2009 2010 /* 2011 * NFS link RPC, called from nfs_link. 2012 * Assumes dvp and vp locked, and leaves them that way. 2013 */ 2014 2015 static int 2016 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name, 2017 size_t namelen, kauth_cred_t cred, struct lwp *l) 2018 { 2019 u_int32_t *tl; 2020 char *cp; 2021 #ifndef NFS_V2_ONLY 2022 int32_t t1; 2023 char *cp2; 2024 #endif 2025 int32_t t2; 2026 char *bpos, *dpos; 2027 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2028 struct mbuf *mreq, *mrep, *md, *mb; 2029 const int v3 = NFS_ISV3(dvp); 2030 int rexmit = 0; 2031 struct nfsnode *np = VTONFS(vp); 2032 2033 nfsstats.rpccnt[NFSPROC_LINK]++; 2034 nfsm_reqhead(np, NFSPROC_LINK, 2035 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen)); 2036 nfsm_fhtom(np, v3); 2037 nfsm_fhtom(VTONFS(dvp), v3); 2038 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 2039 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit); 2040 #ifndef NFS_V2_ONLY 2041 if (v3) { 2042 nfsm_postop_attr(vp, attrflag, 0); 2043 nfsm_wcc_data(dvp, wccflag, 0, !error); 2044 } 2045 #endif 2046 nfsm_reqdone; 2047 2048 VTONFS(dvp)->n_flag |= NMODIFIED; 2049 if (!attrflag) 2050 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp)); 2051 if (!wccflag) 2052 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2053 2054 /* 2055 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2056 */ 2057 if (rexmit && error == EEXIST) 2058 error = 0; 2059 2060 return error; 2061 } 2062 2063 /* 2064 * nfs hard link create call 2065 */ 2066 int 2067 nfs_link(void *v) 2068 { 2069 struct vop_link_args /* { 2070 struct vnode *a_dvp; 2071 struct vnode *a_vp; 2072 struct componentname *a_cnp; 2073 } */ *ap = v; 2074 struct vnode *vp = ap->a_vp; 2075 struct vnode *dvp = ap->a_dvp; 2076 struct componentname *cnp = ap->a_cnp; 2077 int error = 0; 2078 2079 if (dvp->v_mount != vp->v_mount) { 2080 VOP_ABORTOP(dvp, cnp); 2081 vput(dvp); 2082 return (EXDEV); 2083 } 2084 if (dvp != vp) { 2085 error = vn_lock(vp, LK_EXCLUSIVE); 2086 if (error != 0) { 2087 VOP_ABORTOP(dvp, cnp); 2088 vput(dvp); 2089 return error; 2090 } 2091 } 2092 2093 /* 2094 * Push all writes to the server, so that the attribute cache 2095 * doesn't get "out of sync" with the server. 2096 * XXX There should be a better way! 2097 */ 2098 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0); 2099 2100 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 2101 cnp->cn_cred, curlwp); 2102 2103 if (error == 0) 2104 cache_purge1(dvp, cnp, 0); 2105 PNBUF_PUT(cnp->cn_pnbuf); 2106 if (dvp != vp) 2107 VOP_UNLOCK(vp); 2108 VN_KNOTE(vp, NOTE_LINK); 2109 VN_KNOTE(dvp, NOTE_WRITE); 2110 vput(dvp); 2111 return (error); 2112 } 2113 2114 /* 2115 * nfs symbolic link create call 2116 */ 2117 int 2118 nfs_symlink(void *v) 2119 { 2120 struct vop_symlink_args /* { 2121 struct vnode *a_dvp; 2122 struct vnode **a_vpp; 2123 struct componentname *a_cnp; 2124 struct vattr *a_vap; 2125 char *a_target; 2126 } */ *ap = v; 2127 struct vnode *dvp = ap->a_dvp; 2128 struct vattr *vap = ap->a_vap; 2129 struct componentname *cnp = ap->a_cnp; 2130 struct nfsv2_sattr *sp; 2131 u_int32_t *tl; 2132 char *cp; 2133 int32_t t1, t2; 2134 char *bpos, *dpos, *cp2; 2135 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2136 struct mbuf *mreq, *mrep, *md, *mb; 2137 struct vnode *newvp = (struct vnode *)0; 2138 const int v3 = NFS_ISV3(dvp); 2139 int rexmit = 0; 2140 struct nfsnode *dnp = VTONFS(dvp); 2141 2142 *ap->a_vpp = NULL; 2143 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2144 slen = strlen(ap->a_target); 2145 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 2146 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 2147 nfsm_fhtom(dnp, v3); 2148 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2149 #ifndef NFS_V2_ONlY 2150 if (v3) 2151 nfsm_v3attrbuild(vap, false); 2152 #endif 2153 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 2154 #ifndef NFS_V2_ONlY 2155 if (!v3) { 2156 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2157 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2158 sp->sa_uid = nfs_xdrneg1; 2159 sp->sa_gid = nfs_xdrneg1; 2160 sp->sa_size = nfs_xdrneg1; 2161 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2162 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2163 } 2164 #endif 2165 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred, 2166 &rexmit); 2167 #ifndef NFS_V2_ONlY 2168 if (v3) { 2169 if (!error) 2170 nfsm_mtofh(dvp, newvp, v3, gotvp); 2171 nfsm_wcc_data(dvp, wccflag, 0, !error); 2172 } 2173 #endif 2174 nfsm_reqdone; 2175 /* 2176 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2177 */ 2178 if (rexmit && error == EEXIST) 2179 error = 0; 2180 if (error == 0 || error == EEXIST) 2181 cache_purge1(dvp, cnp, 0); 2182 if (error == 0 && newvp == NULL) { 2183 struct nfsnode *np = NULL; 2184 2185 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2186 cnp->cn_cred, curlwp, &np); 2187 if (error == 0) 2188 newvp = NFSTOV(np); 2189 } 2190 if (error) { 2191 if (newvp != NULL) 2192 vput(newvp); 2193 } else { 2194 *ap->a_vpp = newvp; 2195 } 2196 PNBUF_PUT(cnp->cn_pnbuf); 2197 VTONFS(dvp)->n_flag |= NMODIFIED; 2198 if (!wccflag) 2199 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2200 VN_KNOTE(dvp, NOTE_WRITE); 2201 vput(dvp); 2202 return (error); 2203 } 2204 2205 /* 2206 * nfs make dir call 2207 */ 2208 int 2209 nfs_mkdir(void *v) 2210 { 2211 struct vop_mkdir_args /* { 2212 struct vnode *a_dvp; 2213 struct vnode **a_vpp; 2214 struct componentname *a_cnp; 2215 struct vattr *a_vap; 2216 } */ *ap = v; 2217 struct vnode *dvp = ap->a_dvp; 2218 struct vattr *vap = ap->a_vap; 2219 struct componentname *cnp = ap->a_cnp; 2220 struct nfsv2_sattr *sp; 2221 u_int32_t *tl; 2222 char *cp; 2223 int32_t t1, t2; 2224 int len; 2225 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0; 2226 struct vnode *newvp = (struct vnode *)0; 2227 char *bpos, *dpos, *cp2; 2228 int error = 0, wccflag = NFSV3_WCCRATTR; 2229 int gotvp = 0; 2230 int rexmit = 0; 2231 struct mbuf *mreq, *mrep, *md, *mb; 2232 const int v3 = NFS_ISV3(dvp); 2233 2234 len = cnp->cn_namelen; 2235 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2236 nfsm_reqhead(dnp, NFSPROC_MKDIR, 2237 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2238 nfsm_fhtom(dnp, v3); 2239 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2240 #ifndef NFS_V2_ONLY 2241 if (v3) { 2242 nfsm_v3attrbuild(vap, false); 2243 } else 2244 #endif 2245 { 2246 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2247 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2248 sp->sa_uid = nfs_xdrneg1; 2249 sp->sa_gid = nfs_xdrneg1; 2250 sp->sa_size = nfs_xdrneg1; 2251 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2252 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2253 } 2254 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit); 2255 if (!error) 2256 nfsm_mtofh(dvp, newvp, v3, gotvp); 2257 if (v3) 2258 nfsm_wcc_data(dvp, wccflag, 0, !error); 2259 nfsm_reqdone; 2260 VTONFS(dvp)->n_flag |= NMODIFIED; 2261 if (!wccflag) 2262 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2263 /* 2264 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2265 * if we can succeed in looking up the directory. 2266 */ 2267 if ((rexmit && error == EEXIST) || (!error && !gotvp)) { 2268 if (newvp) { 2269 vput(newvp); 2270 newvp = (struct vnode *)0; 2271 } 2272 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2273 curlwp, &np); 2274 if (!error) { 2275 newvp = NFSTOV(np); 2276 if (newvp->v_type != VDIR || newvp == dvp) 2277 error = EEXIST; 2278 } 2279 } 2280 if (error) { 2281 if (newvp) { 2282 if (dvp != newvp) 2283 vput(newvp); 2284 else 2285 vrele(newvp); 2286 } 2287 } else { 2288 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2289 if (cnp->cn_flags & MAKEENTRY) 2290 nfs_cache_enter(dvp, newvp, cnp); 2291 *ap->a_vpp = newvp; 2292 } 2293 PNBUF_PUT(cnp->cn_pnbuf); 2294 vput(dvp); 2295 return (error); 2296 } 2297 2298 /* 2299 * nfs remove directory call 2300 */ 2301 int 2302 nfs_rmdir(void *v) 2303 { 2304 struct vop_rmdir_args /* { 2305 struct vnode *a_dvp; 2306 struct vnode *a_vp; 2307 struct componentname *a_cnp; 2308 } */ *ap = v; 2309 struct vnode *vp = ap->a_vp; 2310 struct vnode *dvp = ap->a_dvp; 2311 struct componentname *cnp = ap->a_cnp; 2312 u_int32_t *tl; 2313 char *cp; 2314 #ifndef NFS_V2_ONLY 2315 int32_t t1; 2316 char *cp2; 2317 #endif 2318 int32_t t2; 2319 char *bpos, *dpos; 2320 int error = 0, wccflag = NFSV3_WCCRATTR; 2321 int rexmit = 0; 2322 struct mbuf *mreq, *mrep, *md, *mb; 2323 const int v3 = NFS_ISV3(dvp); 2324 struct nfsnode *dnp; 2325 2326 if (dvp == vp) { 2327 vrele(dvp); 2328 vput(dvp); 2329 PNBUF_PUT(cnp->cn_pnbuf); 2330 return (EINVAL); 2331 } 2332 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2333 dnp = VTONFS(dvp); 2334 nfsm_reqhead(dnp, NFSPROC_RMDIR, 2335 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2336 nfsm_fhtom(dnp, v3); 2337 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2338 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit); 2339 #ifndef NFS_V2_ONLY 2340 if (v3) 2341 nfsm_wcc_data(dvp, wccflag, 0, !error); 2342 #endif 2343 nfsm_reqdone; 2344 PNBUF_PUT(cnp->cn_pnbuf); 2345 VTONFS(dvp)->n_flag |= NMODIFIED; 2346 if (!wccflag) 2347 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp)); 2348 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); 2349 VN_KNOTE(vp, NOTE_DELETE); 2350 cache_purge(vp); 2351 vput(vp); 2352 vput(dvp); 2353 /* 2354 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2355 */ 2356 if (rexmit && error == ENOENT) 2357 error = 0; 2358 return (error); 2359 } 2360 2361 /* 2362 * nfs readdir call 2363 */ 2364 int 2365 nfs_readdir(void *v) 2366 { 2367 struct vop_readdir_args /* { 2368 struct vnode *a_vp; 2369 struct uio *a_uio; 2370 kauth_cred_t a_cred; 2371 int *a_eofflag; 2372 off_t **a_cookies; 2373 int *a_ncookies; 2374 } */ *ap = v; 2375 struct vnode *vp = ap->a_vp; 2376 struct uio *uio = ap->a_uio; 2377 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2378 char *base = uio->uio_iov->iov_base; 2379 int tresid, error; 2380 size_t count, lost; 2381 struct dirent *dp; 2382 off_t *cookies = NULL; 2383 int ncookies = 0, nc; 2384 2385 if (vp->v_type != VDIR) 2386 return (EPERM); 2387 2388 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1); 2389 count = uio->uio_resid - lost; 2390 if (count <= 0) 2391 return (EINVAL); 2392 2393 /* 2394 * Call nfs_bioread() to do the real work. 2395 */ 2396 tresid = uio->uio_resid = count; 2397 error = nfs_bioread(vp, uio, 0, ap->a_cred, 2398 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0); 2399 2400 if (!error && ap->a_cookies) { 2401 ncookies = count / 16; 2402 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK); 2403 *ap->a_cookies = cookies; 2404 } 2405 2406 if (!error && uio->uio_resid == tresid) { 2407 uio->uio_resid += lost; 2408 nfsstats.direofcache_misses++; 2409 if (ap->a_cookies) 2410 *ap->a_ncookies = 0; 2411 *ap->a_eofflag = 1; 2412 return (0); 2413 } 2414 2415 if (!error && ap->a_cookies) { 2416 /* 2417 * Only the NFS server and emulations use cookies, and they 2418 * load the directory block into system space, so we can 2419 * just look at it directly. 2420 */ 2421 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) || 2422 uio->uio_iovcnt != 1) 2423 panic("nfs_readdir: lost in space"); 2424 for (nc = 0; ncookies-- && 2425 base < (char *)uio->uio_iov->iov_base; nc++){ 2426 dp = (struct dirent *) base; 2427 if (dp->d_reclen == 0) 2428 break; 2429 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) 2430 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp); 2431 else 2432 *(cookies++) = NFS_GETCOOKIE(dp); 2433 base += dp->d_reclen; 2434 } 2435 uio->uio_resid += 2436 ((char *)uio->uio_iov->iov_base - base); 2437 uio->uio_iov->iov_len += 2438 ((char *)uio->uio_iov->iov_base - base); 2439 uio->uio_iov->iov_base = base; 2440 *ap->a_ncookies = nc; 2441 } 2442 2443 uio->uio_resid += lost; 2444 *ap->a_eofflag = 0; 2445 return (error); 2446 } 2447 2448 /* 2449 * Readdir rpc call. 2450 * Called from below the buffer cache by nfs_doio(). 2451 */ 2452 int 2453 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2454 { 2455 int len, left; 2456 struct dirent *dp = NULL; 2457 u_int32_t *tl; 2458 char *cp; 2459 int32_t t1, t2; 2460 char *bpos, *dpos, *cp2; 2461 struct mbuf *mreq, *mrep, *md, *mb; 2462 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2463 struct nfsnode *dnp = VTONFS(vp); 2464 u_quad_t fileno; 2465 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1; 2466 #ifndef NFS_V2_ONLY 2467 int attrflag; 2468 #endif 2469 int nrpcs = 0, reclen; 2470 const int v3 = NFS_ISV3(vp); 2471 2472 #ifdef DIAGNOSTIC 2473 /* 2474 * Should be called from buffer cache, so only amount of 2475 * NFS_DIRBLKSIZ will be requested. 2476 */ 2477 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2478 panic("nfs readdirrpc bad uio"); 2479 #endif 2480 2481 /* 2482 * Loop around doing readdir rpc's of size nm_readdirsize 2483 * truncated to a multiple of NFS_DIRFRAGSIZ. 2484 * The stopping criteria is EOF or buffer full. 2485 */ 2486 while (more_dirs && bigenough) { 2487 /* 2488 * Heuristic: don't bother to do another RPC to further 2489 * fill up this block if there is not much room left. (< 50% 2490 * of the readdir RPC size). This wastes some buffer space 2491 * but can save up to 50% in RPC calls. 2492 */ 2493 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2494 bigenough = 0; 2495 break; 2496 } 2497 nfsstats.rpccnt[NFSPROC_READDIR]++; 2498 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) + 2499 NFSX_READDIR(v3)); 2500 nfsm_fhtom(dnp, v3); 2501 #ifndef NFS_V2_ONLY 2502 if (v3) { 2503 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2504 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2505 txdr_swapcookie3(uiop->uio_offset, tl); 2506 } else { 2507 txdr_cookie3(uiop->uio_offset, tl); 2508 } 2509 tl += 2; 2510 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2511 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2512 } else 2513 #endif 2514 { 2515 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2516 *tl++ = txdr_unsigned(uiop->uio_offset); 2517 } 2518 *tl = txdr_unsigned(nmp->nm_readdirsize); 2519 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred); 2520 nrpcs++; 2521 #ifndef NFS_V2_ONLY 2522 if (v3) { 2523 nfsm_postop_attr(vp, attrflag, 0); 2524 if (!error) { 2525 nfsm_dissect(tl, u_int32_t *, 2526 2 * NFSX_UNSIGNED); 2527 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2528 dnp->n_cookieverf.nfsuquad[1] = *tl; 2529 } else { 2530 m_freem(mrep); 2531 goto nfsmout; 2532 } 2533 } 2534 #endif 2535 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2536 more_dirs = fxdr_unsigned(int, *tl); 2537 2538 /* loop thru the dir entries, doctoring them to 4bsd form */ 2539 while (more_dirs && bigenough) { 2540 #ifndef NFS_V2_ONLY 2541 if (v3) { 2542 nfsm_dissect(tl, u_int32_t *, 2543 3 * NFSX_UNSIGNED); 2544 fileno = fxdr_hyper(tl); 2545 len = fxdr_unsigned(int, *(tl + 2)); 2546 } else 2547 #endif 2548 { 2549 nfsm_dissect(tl, u_int32_t *, 2550 2 * NFSX_UNSIGNED); 2551 fileno = fxdr_unsigned(u_quad_t, *tl++); 2552 len = fxdr_unsigned(int, *tl); 2553 } 2554 if (len <= 0 || len > NFS_MAXNAMLEN) { 2555 error = EBADRPC; 2556 m_freem(mrep); 2557 goto nfsmout; 2558 } 2559 /* for cookie stashing */ 2560 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2561 left = NFS_DIRFRAGSIZ - blksiz; 2562 if (reclen > left) { 2563 memset(uiop->uio_iov->iov_base, 0, left); 2564 dp->d_reclen += left; 2565 UIO_ADVANCE(uiop, left); 2566 blksiz = 0; 2567 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2568 } 2569 if (reclen > uiop->uio_resid) 2570 bigenough = 0; 2571 if (bigenough) { 2572 int tlen; 2573 2574 dp = (struct dirent *)uiop->uio_iov->iov_base; 2575 dp->d_fileno = fileno; 2576 dp->d_namlen = len; 2577 dp->d_reclen = reclen; 2578 dp->d_type = DT_UNKNOWN; 2579 blksiz += reclen; 2580 if (blksiz == NFS_DIRFRAGSIZ) 2581 blksiz = 0; 2582 UIO_ADVANCE(uiop, DIRHDSIZ); 2583 nfsm_mtouio(uiop, len); 2584 tlen = reclen - (DIRHDSIZ + len); 2585 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2586 UIO_ADVANCE(uiop, tlen); 2587 } else 2588 nfsm_adv(nfsm_rndup(len)); 2589 #ifndef NFS_V2_ONLY 2590 if (v3) { 2591 nfsm_dissect(tl, u_int32_t *, 2592 3 * NFSX_UNSIGNED); 2593 } else 2594 #endif 2595 { 2596 nfsm_dissect(tl, u_int32_t *, 2597 2 * NFSX_UNSIGNED); 2598 } 2599 if (bigenough) { 2600 #ifndef NFS_V2_ONLY 2601 if (v3) { 2602 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2603 uiop->uio_offset = 2604 fxdr_swapcookie3(tl); 2605 else 2606 uiop->uio_offset = 2607 fxdr_cookie3(tl); 2608 } 2609 else 2610 #endif 2611 { 2612 uiop->uio_offset = 2613 fxdr_unsigned(off_t, *tl); 2614 } 2615 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2616 } 2617 if (v3) 2618 tl += 2; 2619 else 2620 tl++; 2621 more_dirs = fxdr_unsigned(int, *tl); 2622 } 2623 /* 2624 * If at end of rpc data, get the eof boolean 2625 */ 2626 if (!more_dirs) { 2627 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2628 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2629 2630 /* 2631 * kludge: if we got no entries, treat it as EOF. 2632 * some server sometimes send a reply without any 2633 * entries or EOF. 2634 * although it might mean the server has very long name, 2635 * we can't handle such entries anyway. 2636 */ 2637 2638 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2639 more_dirs = 0; 2640 } 2641 m_freem(mrep); 2642 } 2643 /* 2644 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2645 * by increasing d_reclen for the last record. 2646 */ 2647 if (blksiz > 0) { 2648 left = NFS_DIRFRAGSIZ - blksiz; 2649 memset(uiop->uio_iov->iov_base, 0, left); 2650 dp->d_reclen += left; 2651 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2652 UIO_ADVANCE(uiop, left); 2653 } 2654 2655 /* 2656 * We are now either at the end of the directory or have filled the 2657 * block. 2658 */ 2659 if (bigenough) { 2660 dnp->n_direofoffset = uiop->uio_offset; 2661 dnp->n_flag |= NEOFVALID; 2662 } 2663 nfsmout: 2664 return (error); 2665 } 2666 2667 #ifndef NFS_V2_ONLY 2668 /* 2669 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2670 */ 2671 int 2672 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred) 2673 { 2674 int len, left; 2675 struct dirent *dp = NULL; 2676 u_int32_t *tl; 2677 char *cp; 2678 int32_t t1, t2; 2679 struct vnode *newvp; 2680 char *bpos, *dpos, *cp2; 2681 struct mbuf *mreq, *mrep, *md, *mb; 2682 struct nameidata nami, *ndp = &nami; 2683 struct componentname *cnp = &ndp->ni_cnd; 2684 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2685 struct nfsnode *dnp = VTONFS(vp), *np; 2686 nfsfh_t *fhp; 2687 u_quad_t fileno; 2688 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2689 int attrflag, fhsize, nrpcs = 0, reclen; 2690 struct nfs_fattr fattr, *fp; 2691 2692 #ifdef DIAGNOSTIC 2693 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ) 2694 panic("nfs readdirplusrpc bad uio"); 2695 #endif 2696 ndp->ni_dvp = vp; 2697 newvp = NULLVP; 2698 2699 /* 2700 * Loop around doing readdir rpc's of size nm_readdirsize 2701 * truncated to a multiple of NFS_DIRFRAGSIZ. 2702 * The stopping criteria is EOF or buffer full. 2703 */ 2704 while (more_dirs && bigenough) { 2705 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) { 2706 bigenough = 0; 2707 break; 2708 } 2709 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2710 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS, 2711 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2712 nfsm_fhtom(dnp, 1); 2713 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2714 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) { 2715 txdr_swapcookie3(uiop->uio_offset, tl); 2716 } else { 2717 txdr_cookie3(uiop->uio_offset, tl); 2718 } 2719 tl += 2; 2720 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2721 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2722 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2723 *tl = txdr_unsigned(nmp->nm_rsize); 2724 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred); 2725 nfsm_postop_attr(vp, attrflag, 0); 2726 if (error) { 2727 m_freem(mrep); 2728 goto nfsmout; 2729 } 2730 nrpcs++; 2731 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2732 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2733 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2734 more_dirs = fxdr_unsigned(int, *tl); 2735 2736 /* loop thru the dir entries, doctoring them to 4bsd form */ 2737 while (more_dirs && bigenough) { 2738 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2739 fileno = fxdr_hyper(tl); 2740 len = fxdr_unsigned(int, *(tl + 2)); 2741 if (len <= 0 || len > NFS_MAXNAMLEN) { 2742 error = EBADRPC; 2743 m_freem(mrep); 2744 goto nfsmout; 2745 } 2746 /* for cookie stashing */ 2747 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t); 2748 left = NFS_DIRFRAGSIZ - blksiz; 2749 if (reclen > left) { 2750 /* 2751 * DIRFRAGSIZ is aligned, no need to align 2752 * again here. 2753 */ 2754 memset(uiop->uio_iov->iov_base, 0, left); 2755 dp->d_reclen += left; 2756 UIO_ADVANCE(uiop, left); 2757 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2758 blksiz = 0; 2759 } 2760 if (reclen > uiop->uio_resid) 2761 bigenough = 0; 2762 if (bigenough) { 2763 int tlen; 2764 2765 dp = (struct dirent *)uiop->uio_iov->iov_base; 2766 dp->d_fileno = fileno; 2767 dp->d_namlen = len; 2768 dp->d_reclen = reclen; 2769 dp->d_type = DT_UNKNOWN; 2770 blksiz += reclen; 2771 if (blksiz == NFS_DIRFRAGSIZ) 2772 blksiz = 0; 2773 UIO_ADVANCE(uiop, DIRHDSIZ); 2774 nfsm_mtouio(uiop, len); 2775 tlen = reclen - (DIRHDSIZ + len); 2776 (void)memset(uiop->uio_iov->iov_base, 0, tlen); 2777 UIO_ADVANCE(uiop, tlen); 2778 cnp->cn_nameptr = dp->d_name; 2779 cnp->cn_namelen = dp->d_namlen; 2780 } else 2781 nfsm_adv(nfsm_rndup(len)); 2782 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2783 if (bigenough) { 2784 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) 2785 uiop->uio_offset = 2786 fxdr_swapcookie3(tl); 2787 else 2788 uiop->uio_offset = 2789 fxdr_cookie3(tl); 2790 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2791 } 2792 tl += 2; 2793 2794 /* 2795 * Since the attributes are before the file handle 2796 * (sigh), we must skip over the attributes and then 2797 * come back and get them. 2798 */ 2799 attrflag = fxdr_unsigned(int, *tl); 2800 if (attrflag) { 2801 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR); 2802 memcpy(&fattr, fp, NFSX_V3FATTR); 2803 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2804 doit = fxdr_unsigned(int, *tl); 2805 if (doit) { 2806 nfsm_getfh(fhp, fhsize, 1); 2807 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2808 vref(vp); 2809 newvp = vp; 2810 np = dnp; 2811 } else { 2812 error = nfs_nget1(vp->v_mount, fhp, 2813 fhsize, &np, LK_NOWAIT); 2814 if (!error) 2815 newvp = NFSTOV(np); 2816 } 2817 if (!error) { 2818 const char *xcp; 2819 2820 nfs_loadattrcache(&newvp, &fattr, 0, 0); 2821 if (bigenough) { 2822 dp->d_type = 2823 IFTODT(VTTOIF(np->n_vattr->va_type)); 2824 if (cnp->cn_namelen <= NCHNAMLEN) { 2825 ndp->ni_vp = newvp; 2826 xcp = cnp->cn_nameptr + 2827 cnp->cn_namelen; 2828 cnp->cn_hash = 2829 namei_hash(cnp->cn_nameptr, &xcp); 2830 nfs_cache_enter(ndp->ni_dvp, 2831 ndp->ni_vp, cnp); 2832 } 2833 } 2834 } 2835 error = 0; 2836 } 2837 } else { 2838 /* Just skip over the file handle */ 2839 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2840 i = fxdr_unsigned(int, *tl); 2841 nfsm_adv(nfsm_rndup(i)); 2842 } 2843 if (newvp != NULLVP) { 2844 if (newvp == vp) 2845 vrele(newvp); 2846 else 2847 vput(newvp); 2848 newvp = NULLVP; 2849 } 2850 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2851 more_dirs = fxdr_unsigned(int, *tl); 2852 } 2853 /* 2854 * If at end of rpc data, get the eof boolean 2855 */ 2856 if (!more_dirs) { 2857 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2858 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2859 2860 /* 2861 * kludge: see a comment in nfs_readdirrpc. 2862 */ 2863 2864 if (uiop->uio_resid >= NFS_DIRBLKSIZ) 2865 more_dirs = 0; 2866 } 2867 m_freem(mrep); 2868 } 2869 /* 2870 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ 2871 * by increasing d_reclen for the last record. 2872 */ 2873 if (blksiz > 0) { 2874 left = NFS_DIRFRAGSIZ - blksiz; 2875 memset(uiop->uio_iov->iov_base, 0, left); 2876 dp->d_reclen += left; 2877 NFS_STASHCOOKIE(dp, uiop->uio_offset); 2878 UIO_ADVANCE(uiop, left); 2879 } 2880 2881 /* 2882 * We are now either at the end of the directory or have filled the 2883 * block. 2884 */ 2885 if (bigenough) { 2886 dnp->n_direofoffset = uiop->uio_offset; 2887 dnp->n_flag |= NEOFVALID; 2888 } 2889 nfsmout: 2890 if (newvp != NULLVP) { 2891 if(newvp == vp) 2892 vrele(newvp); 2893 else 2894 vput(newvp); 2895 } 2896 return (error); 2897 } 2898 #endif 2899 2900 /* 2901 * Silly rename. To make the NFS filesystem that is stateless look a little 2902 * more like the "ufs" a remove of an active vnode is translated to a rename 2903 * to a funny looking filename that is removed by nfs_inactive on the 2904 * nfsnode. There is the potential for another process on a different client 2905 * to create the same funny name between the nfs_lookitup() fails and the 2906 * nfs_rename() completes, but... 2907 */ 2908 int 2909 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink) 2910 { 2911 struct sillyrename *sp; 2912 struct nfsnode *np; 2913 int error; 2914 pid_t pid; 2915 2916 cache_purge(dvp); 2917 np = VTONFS(vp); 2918 #ifndef DIAGNOSTIC 2919 if (vp->v_type == VDIR) 2920 panic("nfs: sillyrename dir"); 2921 #endif 2922 sp = kmem_alloc(sizeof(*sp), KM_SLEEP); 2923 sp->s_cred = kauth_cred_dup(cnp->cn_cred); 2924 sp->s_dvp = dvp; 2925 vref(dvp); 2926 2927 /* Fudge together a funny name */ 2928 pid = curlwp->l_proc->p_pid; 2929 memcpy(sp->s_name, ".nfsAxxxx4.4", 13); 2930 sp->s_namlen = 12; 2931 sp->s_name[8] = hexdigits[pid & 0xf]; 2932 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf]; 2933 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf]; 2934 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf]; 2935 2936 /* Try lookitups until we get one that isn't there */ 2937 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2938 curlwp, (struct nfsnode **)0) == 0) { 2939 sp->s_name[4]++; 2940 if (sp->s_name[4] > 'z') { 2941 error = EINVAL; 2942 goto bad; 2943 } 2944 } 2945 if (dolink) { 2946 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen, 2947 sp->s_cred, curlwp); 2948 /* 2949 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP. 2950 */ 2951 if (error == ENOTSUP) { 2952 error = nfs_renameit(dvp, cnp, sp); 2953 } 2954 } else { 2955 error = nfs_renameit(dvp, cnp, sp); 2956 } 2957 if (error) 2958 goto bad; 2959 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2960 curlwp, &np); 2961 np->n_sillyrename = sp; 2962 return (0); 2963 bad: 2964 vrele(sp->s_dvp); 2965 kauth_cred_free(sp->s_cred); 2966 kmem_free(sp, sizeof(*sp)); 2967 return (error); 2968 } 2969 2970 /* 2971 * Look up a file name and optionally either update the file handle or 2972 * allocate an nfsnode, depending on the value of npp. 2973 * npp == NULL --> just do the lookup 2974 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2975 * handled too 2976 * *npp != NULL --> update the file handle in the vnode 2977 */ 2978 int 2979 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp) 2980 { 2981 u_int32_t *tl; 2982 char *cp; 2983 int32_t t1, t2; 2984 struct vnode *newvp = (struct vnode *)0; 2985 struct nfsnode *np, *dnp = VTONFS(dvp); 2986 char *bpos, *dpos, *cp2; 2987 int error = 0, fhlen; 2988 #ifndef NFS_V2_ONLY 2989 int attrflag; 2990 #endif 2991 struct mbuf *mreq, *mrep, *md, *mb; 2992 nfsfh_t *nfhp; 2993 const int v3 = NFS_ISV3(dvp); 2994 2995 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2996 nfsm_reqhead(dnp, NFSPROC_LOOKUP, 2997 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2998 nfsm_fhtom(dnp, v3); 2999 nfsm_strtom(name, len, NFS_MAXNAMLEN); 3000 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred); 3001 if (npp && !error) { 3002 nfsm_getfh(nfhp, fhlen, v3); 3003 if (*npp) { 3004 np = *npp; 3005 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3006 kmem_free(np->n_fhp, np->n_fhsize); 3007 np->n_fhp = &np->n_fh; 3008 } 3009 #if NFS_SMALLFH < NFSX_V3FHMAX 3010 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH) 3011 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP); 3012 #endif 3013 memcpy(np->n_fhp, nfhp, fhlen); 3014 np->n_fhsize = fhlen; 3015 newvp = NFSTOV(np); 3016 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3017 vref(dvp); 3018 newvp = dvp; 3019 np = dnp; 3020 } else { 3021 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 3022 if (error) { 3023 m_freem(mrep); 3024 return (error); 3025 } 3026 newvp = NFSTOV(np); 3027 } 3028 #ifndef NFS_V2_ONLY 3029 if (v3) { 3030 nfsm_postop_attr(newvp, attrflag, 0); 3031 if (!attrflag && *npp == NULL) { 3032 m_freem(mrep); 3033 vput(newvp); 3034 return (ENOENT); 3035 } 3036 } else 3037 #endif 3038 nfsm_loadattr(newvp, (struct vattr *)0, 0); 3039 } 3040 nfsm_reqdone; 3041 if (npp && *npp == NULL) { 3042 if (error) { 3043 if (newvp) 3044 vput(newvp); 3045 } else 3046 *npp = np; 3047 } 3048 return (error); 3049 } 3050 3051 #ifndef NFS_V2_ONLY 3052 /* 3053 * Nfs Version 3 commit rpc 3054 */ 3055 int 3056 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l) 3057 { 3058 char *cp; 3059 u_int32_t *tl; 3060 int32_t t1, t2; 3061 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3062 char *bpos, *dpos, *cp2; 3063 int error = 0, wccflag = NFSV3_WCCRATTR; 3064 struct mbuf *mreq, *mrep, *md, *mb; 3065 struct nfsnode *np; 3066 3067 KASSERT(NFS_ISV3(vp)); 3068 3069 #ifdef NFS_DEBUG_COMMIT 3070 printf("commit %lu - %lu\n", (unsigned long)offset, 3071 (unsigned long)(offset + cnt)); 3072 #endif 3073 3074 mutex_enter(&nmp->nm_lock); 3075 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) { 3076 mutex_exit(&nmp->nm_lock); 3077 return (0); 3078 } 3079 mutex_exit(&nmp->nm_lock); 3080 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3081 np = VTONFS(vp); 3082 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1)); 3083 nfsm_fhtom(np, 1); 3084 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 3085 txdr_hyper(offset, tl); 3086 tl += 2; 3087 *tl = txdr_unsigned(cnt); 3088 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred); 3089 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false); 3090 if (!error) { 3091 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 3092 mutex_enter(&nmp->nm_lock); 3093 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) || 3094 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) { 3095 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF); 3096 error = NFSERR_STALEWRITEVERF; 3097 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF; 3098 } 3099 mutex_exit(&nmp->nm_lock); 3100 } 3101 nfsm_reqdone; 3102 return (error); 3103 } 3104 #endif 3105 3106 /* 3107 * Kludge City.. 3108 * - make nfs_bmap() essentially a no-op that does no translation 3109 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3110 * (Maybe I could use the process's page mapping, but I was concerned that 3111 * Kernel Write might not be enabled and also figured copyout() would do 3112 * a lot more work than memcpy() and also it currently happens in the 3113 * context of the swapper process (2). 3114 */ 3115 int 3116 nfs_bmap(void *v) 3117 { 3118 struct vop_bmap_args /* { 3119 struct vnode *a_vp; 3120 daddr_t a_bn; 3121 struct vnode **a_vpp; 3122 daddr_t *a_bnp; 3123 int *a_runp; 3124 } */ *ap = v; 3125 struct vnode *vp = ap->a_vp; 3126 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift; 3127 3128 if (ap->a_vpp != NULL) 3129 *ap->a_vpp = vp; 3130 if (ap->a_bnp != NULL) 3131 *ap->a_bnp = ap->a_bn << bshift; 3132 if (ap->a_runp != NULL) 3133 *ap->a_runp = 1024 * 1024; /* XXX */ 3134 return (0); 3135 } 3136 3137 /* 3138 * Strategy routine. 3139 * For async requests when nfsiod(s) are running, queue the request by 3140 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 3141 * request. 3142 */ 3143 int 3144 nfs_strategy(void *v) 3145 { 3146 struct vop_strategy_args *ap = v; 3147 struct buf *bp = ap->a_bp; 3148 int error = 0; 3149 3150 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC)) 3151 panic("nfs physio/async"); 3152 3153 /* 3154 * If the op is asynchronous and an i/o daemon is waiting 3155 * queue the request, wake it up and wait for completion 3156 * otherwise just do it ourselves. 3157 */ 3158 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp)) 3159 error = nfs_doio(bp); 3160 return (error); 3161 } 3162 3163 /* 3164 * fsync vnode op. Just call nfs_flush() with commit == 1. 3165 */ 3166 /* ARGSUSED */ 3167 int 3168 nfs_fsync(void *v) 3169 { 3170 struct vop_fsync_args /* { 3171 struct vnodeop_desc *a_desc; 3172 struct vnode * a_vp; 3173 kauth_cred_t a_cred; 3174 int a_flags; 3175 off_t offlo; 3176 off_t offhi; 3177 struct lwp * a_l; 3178 } */ *ap = v; 3179 3180 struct vnode *vp = ap->a_vp; 3181 3182 if (vp->v_type != VREG) 3183 return 0; 3184 3185 return (nfs_flush(vp, ap->a_cred, 3186 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1)); 3187 } 3188 3189 /* 3190 * Flush all the data associated with a vnode. 3191 */ 3192 int 3193 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l, 3194 int commit) 3195 { 3196 struct nfsnode *np = VTONFS(vp); 3197 int error; 3198 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO; 3199 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist); 3200 3201 mutex_enter(&vp->v_interlock); 3202 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 3203 if (np->n_flag & NWRITEERR) { 3204 error = np->n_error; 3205 np->n_flag &= ~NWRITEERR; 3206 } 3207 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0); 3208 return (error); 3209 } 3210 3211 /* 3212 * Return POSIX pathconf information applicable to nfs. 3213 * 3214 * N.B. The NFS V2 protocol doesn't support this RPC. 3215 */ 3216 /* ARGSUSED */ 3217 int 3218 nfs_pathconf(void *v) 3219 { 3220 struct vop_pathconf_args /* { 3221 struct vnode *a_vp; 3222 int a_name; 3223 register_t *a_retval; 3224 } */ *ap = v; 3225 struct nfsv3_pathconf *pcp; 3226 struct vnode *vp = ap->a_vp; 3227 struct mbuf *mreq, *mrep, *md, *mb; 3228 int32_t t1, t2; 3229 u_int32_t *tl; 3230 char *bpos, *dpos, *cp, *cp2; 3231 int error = 0, attrflag; 3232 #ifndef NFS_V2_ONLY 3233 struct nfsmount *nmp; 3234 unsigned int l; 3235 u_int64_t maxsize; 3236 #endif 3237 const int v3 = NFS_ISV3(vp); 3238 struct nfsnode *np = VTONFS(vp); 3239 3240 switch (ap->a_name) { 3241 /* Names that can be resolved locally. */ 3242 case _PC_PIPE_BUF: 3243 *ap->a_retval = PIPE_BUF; 3244 break; 3245 case _PC_SYNC_IO: 3246 *ap->a_retval = 1; 3247 break; 3248 /* Names that cannot be resolved locally; do an RPC, if possible. */ 3249 case _PC_LINK_MAX: 3250 case _PC_NAME_MAX: 3251 case _PC_CHOWN_RESTRICTED: 3252 case _PC_NO_TRUNC: 3253 if (!v3) { 3254 error = EINVAL; 3255 break; 3256 } 3257 nfsstats.rpccnt[NFSPROC_PATHCONF]++; 3258 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1)); 3259 nfsm_fhtom(np, 1); 3260 nfsm_request(np, NFSPROC_PATHCONF, 3261 curlwp, curlwp->l_cred); /* XXX */ 3262 nfsm_postop_attr(vp, attrflag, 0); 3263 if (!error) { 3264 nfsm_dissect(pcp, struct nfsv3_pathconf *, 3265 NFSX_V3PATHCONF); 3266 switch (ap->a_name) { 3267 case _PC_LINK_MAX: 3268 *ap->a_retval = 3269 fxdr_unsigned(register_t, pcp->pc_linkmax); 3270 break; 3271 case _PC_NAME_MAX: 3272 *ap->a_retval = 3273 fxdr_unsigned(register_t, pcp->pc_namemax); 3274 break; 3275 case _PC_CHOWN_RESTRICTED: 3276 *ap->a_retval = 3277 (pcp->pc_chownrestricted == nfs_true); 3278 break; 3279 case _PC_NO_TRUNC: 3280 *ap->a_retval = 3281 (pcp->pc_notrunc == nfs_true); 3282 break; 3283 } 3284 } 3285 nfsm_reqdone; 3286 break; 3287 case _PC_FILESIZEBITS: 3288 #ifndef NFS_V2_ONLY 3289 if (v3) { 3290 nmp = VFSTONFS(vp->v_mount); 3291 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 3292 if ((error = nfs_fsinfo(nmp, vp, 3293 curlwp->l_cred, curlwp)) != 0) /* XXX */ 3294 break; 3295 for (l = 0, maxsize = nmp->nm_maxfilesize; 3296 (maxsize >> l) > 0; l++) 3297 ; 3298 *ap->a_retval = l + 1; 3299 } else 3300 #endif 3301 { 3302 *ap->a_retval = 32; /* NFS V2 limitation */ 3303 } 3304 break; 3305 default: 3306 error = EINVAL; 3307 break; 3308 } 3309 3310 return (error); 3311 } 3312 3313 /* 3314 * NFS advisory byte-level locks. 3315 */ 3316 int 3317 nfs_advlock(void *v) 3318 { 3319 struct vop_advlock_args /* { 3320 struct vnode *a_vp; 3321 void *a_id; 3322 int a_op; 3323 struct flock *a_fl; 3324 int a_flags; 3325 } */ *ap = v; 3326 struct nfsnode *np = VTONFS(ap->a_vp); 3327 3328 return lf_advlock(ap, &np->n_lockf, np->n_size); 3329 } 3330 3331 /* 3332 * Print out the contents of an nfsnode. 3333 */ 3334 int 3335 nfs_print(void *v) 3336 { 3337 struct vop_print_args /* { 3338 struct vnode *a_vp; 3339 } */ *ap = v; 3340 struct vnode *vp = ap->a_vp; 3341 struct nfsnode *np = VTONFS(vp); 3342 3343 printf("tag VT_NFS, fileid %lld fsid 0x%llx", 3344 (unsigned long long)np->n_vattr->va_fileid, 3345 (unsigned long long)np->n_vattr->va_fsid); 3346 if (vp->v_type == VFIFO) 3347 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 3348 printf("\n"); 3349 return (0); 3350 } 3351 3352 /* 3353 * nfs unlock wrapper. 3354 */ 3355 int 3356 nfs_unlock(void *v) 3357 { 3358 struct vop_unlock_args /* { 3359 struct vnode *a_vp; 3360 int a_flags; 3361 } */ *ap = v; 3362 struct vnode *vp = ap->a_vp; 3363 3364 /* 3365 * VOP_UNLOCK can be called by nfs_loadattrcache 3366 * with v_data == 0. 3367 */ 3368 if (VTONFS(vp)) { 3369 nfs_delayedtruncate(vp); 3370 } 3371 3372 return genfs_unlock(v); 3373 } 3374 3375 /* 3376 * nfs special file access vnode op. 3377 * Essentially just get vattr and then imitate iaccess() since the device is 3378 * local to the client. 3379 */ 3380 int 3381 nfsspec_access(void *v) 3382 { 3383 struct vop_access_args /* { 3384 struct vnode *a_vp; 3385 int a_mode; 3386 kauth_cred_t a_cred; 3387 struct lwp *a_l; 3388 } */ *ap = v; 3389 struct vattr va; 3390 struct vnode *vp = ap->a_vp; 3391 int error; 3392 3393 error = VOP_GETATTR(vp, &va, ap->a_cred); 3394 if (error) 3395 return (error); 3396 3397 /* 3398 * Disallow write attempts on filesystems mounted read-only; 3399 * unless the file is a socket, fifo, or a block or character 3400 * device resident on the filesystem. 3401 */ 3402 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3403 switch (vp->v_type) { 3404 case VREG: 3405 case VDIR: 3406 case VLNK: 3407 return (EROFS); 3408 default: 3409 break; 3410 } 3411 } 3412 3413 return (genfs_can_access(va.va_type, va.va_mode, 3414 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred)); 3415 } 3416 3417 /* 3418 * Read wrapper for special devices. 3419 */ 3420 int 3421 nfsspec_read(void *v) 3422 { 3423 struct vop_read_args /* { 3424 struct vnode *a_vp; 3425 struct uio *a_uio; 3426 int a_ioflag; 3427 kauth_cred_t a_cred; 3428 } */ *ap = v; 3429 struct nfsnode *np = VTONFS(ap->a_vp); 3430 3431 /* 3432 * Set access flag. 3433 */ 3434 np->n_flag |= NACC; 3435 getnanotime(&np->n_atim); 3436 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3437 } 3438 3439 /* 3440 * Write wrapper for special devices. 3441 */ 3442 int 3443 nfsspec_write(void *v) 3444 { 3445 struct vop_write_args /* { 3446 struct vnode *a_vp; 3447 struct uio *a_uio; 3448 int a_ioflag; 3449 kauth_cred_t a_cred; 3450 } */ *ap = v; 3451 struct nfsnode *np = VTONFS(ap->a_vp); 3452 3453 /* 3454 * Set update flag. 3455 */ 3456 np->n_flag |= NUPD; 3457 getnanotime(&np->n_mtim); 3458 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3459 } 3460 3461 /* 3462 * Close wrapper for special devices. 3463 * 3464 * Update the times on the nfsnode then do device close. 3465 */ 3466 int 3467 nfsspec_close(void *v) 3468 { 3469 struct vop_close_args /* { 3470 struct vnode *a_vp; 3471 int a_fflag; 3472 kauth_cred_t a_cred; 3473 struct lwp *a_l; 3474 } */ *ap = v; 3475 struct vnode *vp = ap->a_vp; 3476 struct nfsnode *np = VTONFS(vp); 3477 struct vattr vattr; 3478 3479 if (np->n_flag & (NACC | NUPD)) { 3480 np->n_flag |= NCHG; 3481 if (vp->v_usecount == 1 && 3482 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3483 vattr_null(&vattr); 3484 if (np->n_flag & NACC) 3485 vattr.va_atime = np->n_atim; 3486 if (np->n_flag & NUPD) 3487 vattr.va_mtime = np->n_mtim; 3488 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3489 } 3490 } 3491 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3492 } 3493 3494 /* 3495 * Read wrapper for fifos. 3496 */ 3497 int 3498 nfsfifo_read(void *v) 3499 { 3500 struct vop_read_args /* { 3501 struct vnode *a_vp; 3502 struct uio *a_uio; 3503 int a_ioflag; 3504 kauth_cred_t a_cred; 3505 } */ *ap = v; 3506 struct nfsnode *np = VTONFS(ap->a_vp); 3507 3508 /* 3509 * Set access flag. 3510 */ 3511 np->n_flag |= NACC; 3512 getnanotime(&np->n_atim); 3513 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3514 } 3515 3516 /* 3517 * Write wrapper for fifos. 3518 */ 3519 int 3520 nfsfifo_write(void *v) 3521 { 3522 struct vop_write_args /* { 3523 struct vnode *a_vp; 3524 struct uio *a_uio; 3525 int a_ioflag; 3526 kauth_cred_t a_cred; 3527 } */ *ap = v; 3528 struct nfsnode *np = VTONFS(ap->a_vp); 3529 3530 /* 3531 * Set update flag. 3532 */ 3533 np->n_flag |= NUPD; 3534 getnanotime(&np->n_mtim); 3535 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3536 } 3537 3538 /* 3539 * Close wrapper for fifos. 3540 * 3541 * Update the times on the nfsnode then do fifo close. 3542 */ 3543 int 3544 nfsfifo_close(void *v) 3545 { 3546 struct vop_close_args /* { 3547 struct vnode *a_vp; 3548 int a_fflag; 3549 kauth_cred_t a_cred; 3550 struct lwp *a_l; 3551 } */ *ap = v; 3552 struct vnode *vp = ap->a_vp; 3553 struct nfsnode *np = VTONFS(vp); 3554 struct vattr vattr; 3555 3556 if (np->n_flag & (NACC | NUPD)) { 3557 struct timespec ts; 3558 3559 getnanotime(&ts); 3560 if (np->n_flag & NACC) 3561 np->n_atim = ts; 3562 if (np->n_flag & NUPD) 3563 np->n_mtim = ts; 3564 np->n_flag |= NCHG; 3565 if (vp->v_usecount == 1 && 3566 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3567 vattr_null(&vattr); 3568 if (np->n_flag & NACC) 3569 vattr.va_atime = np->n_atim; 3570 if (np->n_flag & NUPD) 3571 vattr.va_mtime = np->n_mtim; 3572 (void)VOP_SETATTR(vp, &vattr, ap->a_cred); 3573 } 3574 } 3575 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3576 } 3577
Indexes created Wed Oct 15 07:09:58 GMT 2025