1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 37 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/nlookup.h> 59 #include <sys/socket.h> 60 #include <sys/vnode.h> 61 #include <sys/dirent.h> 62 #include <sys/fcntl.h> 63 #include <sys/lockf.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/conf.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_zone.h> 71 72 #include <sys/buf2.h> 73 74 #include <vfs/fifofs/fifo.h> 75 #include <vfs/ufs/dir.h> 76 77 #undef DIRBLKSIZ 78 79 #include "rpcv2.h" 80 #include "nfsproto.h" 81 #include "nfs.h" 82 #include "nfsmount.h" 83 #include "nfsnode.h" 84 #include "xdr_subs.h" 85 #include "nfsm_subs.h" 86 87 #include <net/if.h> 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 91 #include <sys/thread2.h> 92 93 /* Defs */ 94 #define TRUE 1 95 #define FALSE 0 96 97 static int nfsfifo_read (struct vop_read_args *); 98 static int nfsfifo_write (struct vop_write_args *); 99 static int nfsfifo_close (struct vop_close_args *); 100 #define nfs_poll vop_nopoll 101 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 102 static int nfs_lookup (struct vop_old_lookup_args *); 103 static int nfs_create (struct vop_old_create_args *); 104 static int nfs_mknod (struct vop_old_mknod_args *); 105 static int nfs_open (struct vop_open_args *); 106 static int nfs_close (struct vop_close_args *); 107 static int nfs_access (struct vop_access_args *); 108 static int nfs_getattr (struct vop_getattr_args *); 109 static int nfs_setattr (struct vop_setattr_args *); 110 static int nfs_read (struct vop_read_args *); 111 static int nfs_mmap (struct vop_mmap_args *); 112 static int nfs_fsync (struct vop_fsync_args *); 113 static int nfs_remove (struct vop_old_remove_args *); 114 static int nfs_link (struct vop_old_link_args *); 115 static int nfs_rename (struct vop_old_rename_args *); 116 static int nfs_mkdir (struct vop_old_mkdir_args *); 117 static int nfs_rmdir (struct vop_old_rmdir_args *); 118 static int nfs_symlink (struct vop_old_symlink_args *); 119 static int nfs_readdir (struct vop_readdir_args *); 120 static int nfs_bmap (struct vop_bmap_args *); 121 static int nfs_strategy (struct vop_strategy_args *); 122 static int nfs_lookitup (struct vnode *, const char *, int, 123 struct ucred *, struct thread *, struct nfsnode **); 124 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 125 static int nfs_laccess (struct vop_access_args *); 126 static int nfs_readlink (struct vop_readlink_args *); 127 static int nfs_print (struct vop_print_args *); 128 static int nfs_advlock (struct vop_advlock_args *); 129 130 static int nfs_nresolve (struct vop_nresolve_args *); 131 /* 132 * Global vfs data structures for nfs 133 */ 134 struct vop_ops nfsv2_vnode_vops = { 135 .vop_default = vop_defaultop, 136 .vop_access = nfs_access, 137 .vop_advlock = nfs_advlock, 138 .vop_bmap = nfs_bmap, 139 .vop_close = nfs_close, 140 .vop_old_create = nfs_create, 141 .vop_fsync = nfs_fsync, 142 .vop_getattr = nfs_getattr, 143 .vop_getpages = vop_stdgetpages, 144 .vop_putpages = vop_stdputpages, 145 .vop_inactive = nfs_inactive, 146 .vop_old_link = nfs_link, 147 .vop_old_lookup = nfs_lookup, 148 .vop_old_mkdir = nfs_mkdir, 149 .vop_old_mknod = nfs_mknod, 150 .vop_mmap = nfs_mmap, 151 .vop_open = nfs_open, 152 .vop_poll = nfs_poll, 153 .vop_print = nfs_print, 154 .vop_read = nfs_read, 155 .vop_readdir = nfs_readdir, 156 .vop_readlink = nfs_readlink, 157 .vop_reclaim = nfs_reclaim, 158 .vop_old_remove = nfs_remove, 159 .vop_old_rename = nfs_rename, 160 .vop_old_rmdir = nfs_rmdir, 161 .vop_setattr = nfs_setattr, 162 .vop_strategy = nfs_strategy, 163 .vop_old_symlink = nfs_symlink, 164 .vop_write = nfs_write, 165 .vop_nresolve = nfs_nresolve 166 }; 167 168 /* 169 * Special device vnode ops 170 */ 171 struct vop_ops nfsv2_spec_vops = { 172 .vop_default = vop_defaultop, 173 .vop_access = nfs_laccess, 174 .vop_close = nfs_close, 175 .vop_fsync = nfs_fsync, 176 .vop_getattr = nfs_getattr, 177 .vop_inactive = nfs_inactive, 178 .vop_print = nfs_print, 179 .vop_read = vop_stdnoread, 180 .vop_reclaim = nfs_reclaim, 181 .vop_setattr = nfs_setattr, 182 .vop_write = vop_stdnowrite 183 }; 184 185 struct vop_ops nfsv2_fifo_vops = { 186 .vop_default = fifo_vnoperate, 187 .vop_access = nfs_laccess, 188 .vop_close = nfsfifo_close, 189 .vop_fsync = nfs_fsync, 190 .vop_getattr = nfs_getattr, 191 .vop_inactive = nfs_inactive, 192 .vop_print = nfs_print, 193 .vop_read = nfsfifo_read, 194 .vop_reclaim = nfs_reclaim, 195 .vop_setattr = nfs_setattr, 196 .vop_write = nfsfifo_write 197 }; 198 199 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 200 struct componentname *cnp, 201 struct vattr *vap); 202 static int nfs_removerpc (struct vnode *dvp, const char *name, 203 int namelen, 204 struct ucred *cred, struct thread *td); 205 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 206 int fnamelen, struct vnode *tdvp, 207 const char *tnameptr, int tnamelen, 208 struct ucred *cred, struct thread *td); 209 static int nfs_renameit (struct vnode *sdvp, 210 struct componentname *scnp, 211 struct sillyrename *sp); 212 213 SYSCTL_DECL(_vfs_nfs); 214 215 static int nfs_flush_on_rename = 1; 216 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW, 217 &nfs_flush_on_rename, 0, "flush fvp prior to rename"); 218 static int nfs_flush_on_hlink = 0; 219 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW, 220 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link"); 221 222 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 223 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 225 226 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 227 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 228 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout"); 229 230 static int nfspos_cache_timeout = NFS_MINATTRTIMO; 231 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW, 232 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout"); 233 234 static int nfsv3_commit_on_close = 0; 235 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 236 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 237 #if 0 238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 239 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 240 241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 242 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 243 #endif 244 245 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 246 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 247 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 248 249 /* 250 * Returns whether a name component is a degenerate '.' or '..'. 251 */ 252 static __inline 253 int 254 nlcdegenerate(struct nlcomponent *nlc) 255 { 256 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.') 257 return(1); 258 if (nlc->nlc_namelen == 2 && 259 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.') 260 return(1); 261 return(0); 262 } 263 264 static int 265 nfs3_access_otw(struct vnode *vp, int wmode, 266 struct thread *td, struct ucred *cred) 267 { 268 struct nfsnode *np = VTONFS(vp); 269 int attrflag; 270 int error = 0; 271 u_int32_t *tl; 272 u_int32_t rmode; 273 struct nfsm_info info; 274 275 info.mrep = NULL; 276 info.v3 = 1; 277 278 nfsstats.rpccnt[NFSPROC_ACCESS]++; 279 nfsm_reqhead(&info, vp, NFSPROC_ACCESS, 280 NFSX_FH(info.v3) + NFSX_UNSIGNED); 281 ERROROUT(nfsm_fhtom(&info, vp)); 282 tl = nfsm_build(&info, NFSX_UNSIGNED); 283 *tl = txdr_unsigned(wmode); 284 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error)); 285 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK)); 286 if (error == 0) { 287 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 288 rmode = fxdr_unsigned(u_int32_t, *tl); 289 np->n_mode = rmode; 290 np->n_modeuid = cred->cr_uid; 291 np->n_modestamp = mycpu->gd_time_seconds; 292 } 293 m_freem(info.mrep); 294 info.mrep = NULL; 295 nfsmout: 296 return error; 297 } 298 299 /* 300 * nfs access vnode op. 301 * For nfs version 2, just return ok. File accesses may fail later. 302 * For nfs version 3, use the access rpc to check accessibility. If file modes 303 * are changed on the server, accesses might still fail later. 304 * 305 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 306 */ 307 static int 308 nfs_access(struct vop_access_args *ap) 309 { 310 struct ucred *cred; 311 struct vnode *vp = ap->a_vp; 312 thread_t td = curthread; 313 int error = 0; 314 u_int32_t mode, wmode; 315 struct nfsnode *np = VTONFS(vp); 316 int v3 = NFS_ISV3(vp); 317 318 /* 319 * Disallow write attempts on filesystems mounted read-only; 320 * unless the file is a socket, fifo, or a block or character 321 * device resident on the filesystem. 322 */ 323 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 324 switch (vp->v_type) { 325 case VREG: 326 case VDIR: 327 case VLNK: 328 return (EROFS); 329 default: 330 break; 331 } 332 } 333 334 /* 335 * The NFS protocol passes only the effective uid/gid over the wire but 336 * we need to check access against real ids if AT_EACCESS not set. 337 * Handle this case by cloning the credentials and setting the 338 * effective ids to the real ones. 339 */ 340 if (ap->a_flags & AT_EACCESS) { 341 cred = crhold(ap->a_cred); 342 } else { 343 cred = crdup(ap->a_cred); 344 cred->cr_uid = cred->cr_ruid; 345 cred->cr_gid = cred->cr_rgid; 346 } 347 348 /* 349 * For nfs v3, check to see if we have done this recently, and if 350 * so return our cached result instead of making an ACCESS call. 351 * If not, do an access rpc, otherwise you are stuck emulating 352 * ufs_access() locally using the vattr. This may not be correct, 353 * since the server may apply other access criteria such as 354 * client uid-->server uid mapping that we do not know about. 355 */ 356 if (v3) { 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 /* XXX safety belt, only make blanket request if caching */ 374 if (nfsaccess_cache_timeout > 0) { 375 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 376 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 377 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 378 } else { 379 wmode = mode; 380 } 381 382 /* 383 * Does our cached result allow us to give a definite yes to 384 * this request? 385 */ 386 if (np->n_modestamp && 387 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 388 (cred->cr_uid == np->n_modeuid) && 389 ((np->n_mode & mode) == mode)) { 390 nfsstats.accesscache_hits++; 391 } else { 392 /* 393 * Either a no, or a don't know. Go to the wire. 394 */ 395 nfsstats.accesscache_misses++; 396 error = nfs3_access_otw(vp, wmode, td, cred); 397 if (!error) { 398 if ((np->n_mode & mode) != mode) { 399 error = EACCES; 400 } 401 } 402 } 403 } else { 404 if ((error = nfs_laccess(ap)) != 0) { 405 crfree(cred); 406 return (error); 407 } 408 409 /* 410 * Attempt to prevent a mapped root from accessing a file 411 * which it shouldn't. We try to read a byte from the file 412 * if the user is root and the file is not zero length. 413 * After calling nfs_laccess, we should have the correct 414 * file size cached. 415 */ 416 if (cred->cr_uid == 0 && (ap->a_mode & VREAD) 417 && VTONFS(vp)->n_size > 0) { 418 struct iovec aiov; 419 struct uio auio; 420 char buf[1]; 421 422 aiov.iov_base = buf; 423 aiov.iov_len = 1; 424 auio.uio_iov = &aiov; 425 auio.uio_iovcnt = 1; 426 auio.uio_offset = 0; 427 auio.uio_resid = 1; 428 auio.uio_segflg = UIO_SYSSPACE; 429 auio.uio_rw = UIO_READ; 430 auio.uio_td = td; 431 432 if (vp->v_type == VREG) { 433 error = nfs_readrpc_uio(vp, &auio); 434 } else if (vp->v_type == VDIR) { 435 char* bp; 436 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 437 aiov.iov_base = bp; 438 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 439 error = nfs_readdirrpc_uio(vp, &auio); 440 kfree(bp, M_TEMP); 441 } else if (vp->v_type == VLNK) { 442 error = nfs_readlinkrpc_uio(vp, &auio); 443 } else { 444 error = EACCES; 445 } 446 } 447 } 448 /* 449 * [re]record creds for reading and/or writing if access 450 * was granted. Assume the NFS server will grant read access 451 * for execute requests. 452 */ 453 if (error == 0) { 454 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) { 455 crhold(cred); 456 if (np->n_rucred) 457 crfree(np->n_rucred); 458 np->n_rucred = cred; 459 } 460 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) { 461 crhold(cred); 462 if (np->n_wucred) 463 crfree(np->n_wucred); 464 np->n_wucred = cred; 465 } 466 } 467 crfree(cred); 468 return(error); 469 } 470 471 /* 472 * nfs open vnode op 473 * Check to see if the type is ok 474 * and that deletion is not in progress. 475 * For paged in text files, you will need to flush the page cache 476 * if consistency is lost. 477 * 478 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 479 * struct file *a_fp) 480 */ 481 /* ARGSUSED */ 482 static int 483 nfs_open(struct vop_open_args *ap) 484 { 485 struct vnode *vp = ap->a_vp; 486 struct nfsnode *np = VTONFS(vp); 487 struct vattr vattr; 488 int error; 489 490 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 491 #ifdef DIAGNOSTIC 492 kprintf("open eacces vtyp=%d\n",vp->v_type); 493 #endif 494 return (EOPNOTSUPP); 495 } 496 497 /* 498 * Save valid creds for reading and writing for later RPCs. 499 */ 500 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) { 501 crhold(ap->a_cred); 502 if (np->n_rucred) 503 crfree(np->n_rucred); 504 np->n_rucred = ap->a_cred; 505 } 506 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) { 507 crhold(ap->a_cred); 508 if (np->n_wucred) 509 crfree(np->n_wucred); 510 np->n_wucred = ap->a_cred; 511 } 512 513 /* 514 * Clear the attribute cache only if opening with write access. It 515 * is unclear if we should do this at all here, but we certainly 516 * should not clear the cache unconditionally simply because a file 517 * is being opened. 518 */ 519 if (ap->a_mode & FWRITE) 520 np->n_attrstamp = 0; 521 522 /* 523 * For normal NFS, reconcile changes made locally verses 524 * changes made remotely. Note that VOP_GETATTR only goes 525 * to the wire if the cached attribute has timed out or been 526 * cleared. 527 * 528 * If local modifications have been made clear the attribute 529 * cache to force an attribute and modified time check. If 530 * GETATTR detects that the file has been changed by someone 531 * other then us it will set NRMODIFIED. 532 * 533 * If we are opening a directory and local changes have been 534 * made we have to invalidate the cache in order to ensure 535 * that we get the most up-to-date information from the 536 * server. XXX 537 */ 538 if (np->n_flag & NLMODIFIED) { 539 np->n_attrstamp = 0; 540 if (vp->v_type == VDIR) { 541 error = nfs_vinvalbuf(vp, V_SAVE, 1); 542 if (error == EINTR) 543 return (error); 544 nfs_invaldir(vp); 545 } 546 } 547 error = VOP_GETATTR(vp, &vattr); 548 if (error) 549 return (error); 550 if (np->n_flag & NRMODIFIED) { 551 if (vp->v_type == VDIR) 552 nfs_invaldir(vp); 553 error = nfs_vinvalbuf(vp, V_SAVE, 1); 554 if (error == EINTR) 555 return (error); 556 np->n_flag &= ~NRMODIFIED; 557 } 558 559 return (vop_stdopen(ap)); 560 } 561 562 /* 563 * nfs close vnode op 564 * What an NFS client should do upon close after writing is a debatable issue. 565 * Most NFS clients push delayed writes to the server upon close, basically for 566 * two reasons: 567 * 1 - So that any write errors may be reported back to the client process 568 * doing the close system call. By far the two most likely errors are 569 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 570 * 2 - To put a worst case upper bound on cache inconsistency between 571 * multiple clients for the file. 572 * There is also a consistency problem for Version 2 of the protocol w.r.t. 573 * not being able to tell if other clients are writing a file concurrently, 574 * since there is no way of knowing if the changed modify time in the reply 575 * is only due to the write for this client. 576 * (NFS Version 3 provides weak cache consistency data in the reply that 577 * should be sufficient to detect and handle this case.) 578 * 579 * The current code does the following: 580 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 581 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 582 * or commit them (this satisfies 1 and 2 except for the 583 * case where the server crashes after this close but 584 * before the commit RPC, which is felt to be "good 585 * enough". Changing the last argument to nfs_flush() to 586 * a 1 would force a commit operation, if it is felt a 587 * commit is necessary now. 588 * for NQNFS - do nothing now, since 2 is dealt with via leases and 589 * 1 should be dealt with via an fsync() system call for 590 * cases where write errors are important. 591 * 592 * nfs_close(struct vnode *a_vp, int a_fflag) 593 */ 594 /* ARGSUSED */ 595 static int 596 nfs_close(struct vop_close_args *ap) 597 { 598 struct vnode *vp = ap->a_vp; 599 struct nfsnode *np = VTONFS(vp); 600 int error = 0; 601 thread_t td = curthread; 602 603 if (vp->v_type == VREG) { 604 if (np->n_flag & NLMODIFIED) { 605 if (NFS_ISV3(vp)) { 606 /* 607 * Under NFSv3 we have dirty buffers to dispose of. We 608 * must flush them to the NFS server. We have the option 609 * of waiting all the way through the commit rpc or just 610 * waiting for the initial write. The default is to only 611 * wait through the initial write so the data is in the 612 * server's cache, which is roughly similar to the state 613 * a standard disk subsystem leaves the file in on close(). 614 * 615 * We cannot clear the NLMODIFIED bit in np->n_flag due to 616 * potential races with other processes, and certainly 617 * cannot clear it if we don't commit. 618 */ 619 int cm = nfsv3_commit_on_close ? 1 : 0; 620 error = nfs_flush(vp, MNT_WAIT, td, cm); 621 /* np->n_flag &= ~NLMODIFIED; */ 622 } else { 623 error = nfs_vinvalbuf(vp, V_SAVE, 1); 624 } 625 np->n_attrstamp = 0; 626 } 627 if (np->n_flag & NWRITEERR) { 628 np->n_flag &= ~NWRITEERR; 629 error = np->n_error; 630 } 631 } 632 vop_stdclose(ap); 633 return (error); 634 } 635 636 /* 637 * nfs getattr call from vfs. 638 * 639 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap) 640 */ 641 static int 642 nfs_getattr(struct vop_getattr_args *ap) 643 { 644 struct vnode *vp = ap->a_vp; 645 struct nfsnode *np = VTONFS(vp); 646 struct nfsmount *nmp; 647 int error = 0; 648 thread_t td = curthread; 649 struct nfsm_info info; 650 651 info.mrep = NULL; 652 info.v3 = NFS_ISV3(vp); 653 nmp = VFSTONFS(vp->v_mount); 654 655 /* 656 * Update local times for special files. 657 */ 658 if (np->n_flag & (NACC | NUPD)) 659 np->n_flag |= NCHG; 660 /* 661 * First look in the cache. 662 */ 663 if (nfs_getattrcache(vp, ap->a_vap) == 0) 664 goto done; 665 666 if (info.v3 && nfsaccess_cache_timeout > 0) { 667 nfsstats.accesscache_misses++; 668 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 669 if (nfs_getattrcache(vp, ap->a_vap) == 0) 670 goto done; 671 } 672 673 nfsstats.rpccnt[NFSPROC_GETATTR]++; 674 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3)); 675 ERROROUT(nfsm_fhtom(&info, vp)); 676 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td, 677 nfs_vpcred(vp, ND_CHECK), &error)); 678 if (error == 0) { 679 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap)); 680 } 681 m_freem(info.mrep); 682 info.mrep = NULL; 683 done: 684 /* 685 * NFS doesn't support chflags flags. If the nfs mount was 686 * made -o cache set the UF_CACHE bit for swapcache. 687 */ 688 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT)) 689 ap->a_vap->va_flags |= UF_CACHE; 690 nfsmout: 691 return (error); 692 } 693 694 /* 695 * nfs setattr call. 696 * 697 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 698 */ 699 static int 700 nfs_setattr(struct vop_setattr_args *ap) 701 { 702 struct vnode *vp = ap->a_vp; 703 struct nfsnode *np = VTONFS(vp); 704 struct vattr *vap = ap->a_vap; 705 int biosize = vp->v_mount->mnt_stat.f_iosize; 706 int error = 0; 707 int boff; 708 off_t tsize; 709 thread_t td = curthread; 710 711 #ifndef nolint 712 tsize = (off_t)0; 713 #endif 714 715 /* 716 * Setting of flags is not supported. 717 */ 718 if (vap->va_flags != VNOVAL) 719 return (EOPNOTSUPP); 720 721 /* 722 * Disallow write attempts if the filesystem is mounted read-only. 723 */ 724 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 725 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 726 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 727 (vp->v_mount->mnt_flag & MNT_RDONLY)) 728 return (EROFS); 729 730 if (vap->va_size != VNOVAL) { 731 /* 732 * truncation requested 733 */ 734 switch (vp->v_type) { 735 case VDIR: 736 return (EISDIR); 737 case VCHR: 738 case VBLK: 739 case VSOCK: 740 case VFIFO: 741 if (vap->va_mtime.tv_sec == VNOVAL && 742 vap->va_atime.tv_sec == VNOVAL && 743 vap->va_mode == (mode_t)VNOVAL && 744 vap->va_uid == (uid_t)VNOVAL && 745 vap->va_gid == (gid_t)VNOVAL) 746 return (0); 747 vap->va_size = VNOVAL; 748 break; 749 default: 750 /* 751 * Disallow write attempts if the filesystem is 752 * mounted read-only. 753 */ 754 if (vp->v_mount->mnt_flag & MNT_RDONLY) 755 return (EROFS); 756 757 tsize = np->n_size; 758 again: 759 boff = (int)vap->va_size & (biosize - 1); 760 error = nfs_meta_setsize(vp, td, vap->va_size, 0); 761 762 #if 0 763 if (np->n_flag & NLMODIFIED) { 764 if (vap->va_size == 0) 765 error = nfs_vinvalbuf(vp, 0, 1); 766 else 767 error = nfs_vinvalbuf(vp, V_SAVE, 1); 768 } 769 #endif 770 /* 771 * note: this loop case almost always happens at 772 * least once per truncation. 773 */ 774 if (error == 0 && np->n_size != vap->va_size) 775 goto again; 776 np->n_vattr.va_size = vap->va_size; 777 break; 778 } 779 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) { 780 /* 781 * What to do. If we are modifying the mtime we lose 782 * mtime detection of changes made by the server or other 783 * clients. But programs like rsync/rdist/cpdup are going 784 * to call utimes a lot. We don't want to piecemeal sync. 785 * 786 * For now sync if any prior remote changes were detected, 787 * but allow us to lose track of remote changes made during 788 * the utimes operation. 789 */ 790 if (np->n_flag & NRMODIFIED) 791 error = nfs_vinvalbuf(vp, V_SAVE, 1); 792 if (error == EINTR) 793 return (error); 794 if (error == 0) { 795 if (vap->va_mtime.tv_sec != VNOVAL) { 796 np->n_mtime = vap->va_mtime.tv_sec; 797 } 798 } 799 } 800 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 801 802 /* 803 * Sanity check if a truncation was issued. This should only occur 804 * if multiple processes are racing on the same file. 805 */ 806 if (error == 0 && vap->va_size != VNOVAL && 807 np->n_size != vap->va_size) { 808 kprintf("NFS ftruncate: server disagrees on the file size: " 809 "%jd/%jd/%jd\n", 810 (intmax_t)tsize, 811 (intmax_t)vap->va_size, 812 (intmax_t)np->n_size); 813 goto again; 814 } 815 if (error && vap->va_size != VNOVAL) { 816 np->n_size = np->n_vattr.va_size = tsize; 817 nfs_meta_setsize(vp, td, np->n_size, 0); 818 } 819 return (error); 820 } 821 822 /* 823 * Do an nfs setattr rpc. 824 */ 825 static int 826 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 827 struct ucred *cred, struct thread *td) 828 { 829 struct nfsv2_sattr *sp; 830 struct nfsnode *np = VTONFS(vp); 831 u_int32_t *tl; 832 int error = 0, wccflag = NFSV3_WCCRATTR; 833 struct nfsm_info info; 834 835 info.mrep = NULL; 836 info.v3 = NFS_ISV3(vp); 837 838 nfsstats.rpccnt[NFSPROC_SETATTR]++; 839 nfsm_reqhead(&info, vp, NFSPROC_SETATTR, 840 NFSX_FH(info.v3) + NFSX_SATTR(info.v3)); 841 ERROROUT(nfsm_fhtom(&info, vp)); 842 if (info.v3) { 843 nfsm_v3attrbuild(&info, vap, TRUE); 844 tl = nfsm_build(&info, NFSX_UNSIGNED); 845 *tl = nfs_false; 846 } else { 847 sp = nfsm_build(&info, NFSX_V2SATTR); 848 if (vap->va_mode == (mode_t)VNOVAL) 849 sp->sa_mode = nfs_xdrneg1; 850 else 851 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 852 if (vap->va_uid == (uid_t)VNOVAL) 853 sp->sa_uid = nfs_xdrneg1; 854 else 855 sp->sa_uid = txdr_unsigned(vap->va_uid); 856 if (vap->va_gid == (gid_t)VNOVAL) 857 sp->sa_gid = nfs_xdrneg1; 858 else 859 sp->sa_gid = txdr_unsigned(vap->va_gid); 860 sp->sa_size = txdr_unsigned(vap->va_size); 861 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 862 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 863 } 864 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error)); 865 if (info.v3) { 866 np->n_modestamp = 0; 867 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 868 } else { 869 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 870 } 871 m_freem(info.mrep); 872 info.mrep = NULL; 873 nfsmout: 874 return (error); 875 } 876 877 static 878 void 879 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout) 880 { 881 if (nctimeout == 0) 882 nctimeout = 1; 883 else 884 nctimeout *= hz; 885 cache_setvp(nch, vp); 886 cache_settimeout(nch, nctimeout); 887 } 888 889 /* 890 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 891 * nfs_lookup() until all remaining new api calls are implemented. 892 * 893 * Resolve a namecache entry. This function is passed a locked ncp and 894 * must call nfs_cache_setvp() on it as appropriate to resolve the entry. 895 */ 896 static int 897 nfs_nresolve(struct vop_nresolve_args *ap) 898 { 899 struct thread *td = curthread; 900 struct namecache *ncp; 901 struct ucred *cred; 902 struct nfsnode *np; 903 struct vnode *dvp; 904 struct vnode *nvp; 905 nfsfh_t *fhp; 906 int attrflag; 907 int fhsize; 908 int error; 909 int tmp_error; 910 int len; 911 struct nfsm_info info; 912 913 cred = ap->a_cred; 914 dvp = ap->a_dvp; 915 916 if ((error = vget(dvp, LK_SHARED)) != 0) 917 return (error); 918 919 info.mrep = NULL; 920 info.v3 = NFS_ISV3(dvp); 921 922 nvp = NULL; 923 nfsstats.lookupcache_misses++; 924 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 925 ncp = ap->a_nch->ncp; 926 len = ncp->nc_nlen; 927 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 928 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 929 ERROROUT(nfsm_fhtom(&info, dvp)); 930 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN)); 931 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, 932 ap->a_cred, &error)); 933 if (error) { 934 /* 935 * Cache negatve lookups to reduce NFS traffic, but use 936 * a fast timeout. Otherwise use a timeout of 1 tick. 937 * XXX we should add a namecache flag for no-caching 938 * to uncache the negative hit as soon as possible, but 939 * we cannot simply destroy the entry because it is used 940 * as a placeholder by the caller. 941 * 942 * The refactored nfs code will overwrite a non-zero error 943 * with 0 when we use ERROROUT(), so don't here. 944 */ 945 if (error == ENOENT) 946 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout); 947 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 948 NFS_LATTR_NOSHRINK); 949 if (tmp_error) { 950 error = tmp_error; 951 goto nfsmout; 952 } 953 m_freem(info.mrep); 954 info.mrep = NULL; 955 goto nfsmout; 956 } 957 958 /* 959 * Success, get the file handle, do various checks, and load 960 * post-operation data from the reply packet. Theoretically 961 * we should never be looking up "." so, theoretically, we 962 * should never get the same file handle as our directory. But 963 * we check anyway. XXX 964 * 965 * Note that no timeout is set for the positive cache hit. We 966 * assume, theoretically, that ESTALE returns will be dealt with 967 * properly to handle NFS races and in anycase we cannot depend 968 * on a timeout to deal with NFS open/create/excl issues so instead 969 * of a bad hack here the rest of the NFS client code needs to do 970 * the right thing. 971 */ 972 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 973 974 np = VTONFS(dvp); 975 if (NFS_CMPFH(np, fhp, fhsize)) { 976 vref(dvp); 977 nvp = dvp; 978 } else { 979 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 980 if (error) { 981 m_freem(info.mrep); 982 info.mrep = NULL; 983 vput(dvp); 984 return (error); 985 } 986 nvp = NFSTOV(np); 987 } 988 if (info.v3) { 989 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag, 990 NFS_LATTR_NOSHRINK)); 991 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 992 NFS_LATTR_NOSHRINK)); 993 } else { 994 ERROROUT(nfsm_loadattr(&info, nvp, NULL)); 995 } 996 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout); 997 m_freem(info.mrep); 998 info.mrep = NULL; 999 nfsmout: 1000 vput(dvp); 1001 if (nvp) { 1002 if (nvp == dvp) 1003 vrele(nvp); 1004 else 1005 vput(nvp); 1006 } 1007 return (error); 1008 } 1009 1010 /* 1011 * 'cached' nfs directory lookup 1012 * 1013 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 1014 * 1015 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 1016 * struct componentname *a_cnp) 1017 */ 1018 static int 1019 nfs_lookup(struct vop_old_lookup_args *ap) 1020 { 1021 struct componentname *cnp = ap->a_cnp; 1022 struct vnode *dvp = ap->a_dvp; 1023 struct vnode **vpp = ap->a_vpp; 1024 int flags = cnp->cn_flags; 1025 struct vnode *newvp; 1026 struct nfsmount *nmp; 1027 long len; 1028 nfsfh_t *fhp; 1029 struct nfsnode *np; 1030 int lockparent, wantparent, attrflag, fhsize; 1031 int error; 1032 int tmp_error; 1033 struct nfsm_info info; 1034 1035 info.mrep = NULL; 1036 info.v3 = NFS_ISV3(dvp); 1037 error = 0; 1038 1039 /* 1040 * Read-only mount check and directory check. 1041 */ 1042 *vpp = NULLVP; 1043 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1044 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1045 return (EROFS); 1046 1047 if (dvp->v_type != VDIR) 1048 return (ENOTDIR); 1049 1050 /* 1051 * Look it up in the cache. Note that ENOENT is only returned if we 1052 * previously entered a negative hit (see later on). The additional 1053 * nfsneg_cache_timeout check causes previously cached results to 1054 * be instantly ignored if the negative caching is turned off. 1055 */ 1056 lockparent = flags & CNP_LOCKPARENT; 1057 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1058 nmp = VFSTONFS(dvp->v_mount); 1059 np = VTONFS(dvp); 1060 1061 /* 1062 * Go to the wire. 1063 */ 1064 error = 0; 1065 newvp = NULLVP; 1066 nfsstats.lookupcache_misses++; 1067 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1068 len = cnp->cn_namelen; 1069 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 1070 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1071 ERROROUT(nfsm_fhtom(&info, dvp)); 1072 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 1073 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td, 1074 cnp->cn_cred, &error)); 1075 if (error) { 1076 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 1077 NFS_LATTR_NOSHRINK); 1078 if (tmp_error) { 1079 error = tmp_error; 1080 goto nfsmout; 1081 } 1082 1083 m_freem(info.mrep); 1084 info.mrep = NULL; 1085 goto nfsmout; 1086 } 1087 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 1088 1089 /* 1090 * Handle RENAME case... 1091 */ 1092 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1093 if (NFS_CMPFH(np, fhp, fhsize)) { 1094 m_freem(info.mrep); 1095 info.mrep = NULL; 1096 return (EISDIR); 1097 } 1098 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1099 if (error) { 1100 m_freem(info.mrep); 1101 info.mrep = NULL; 1102 return (error); 1103 } 1104 newvp = NFSTOV(np); 1105 if (info.v3) { 1106 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1107 NFS_LATTR_NOSHRINK)); 1108 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1109 NFS_LATTR_NOSHRINK)); 1110 } else { 1111 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1112 } 1113 *vpp = newvp; 1114 m_freem(info.mrep); 1115 info.mrep = NULL; 1116 if (!lockparent) { 1117 vn_unlock(dvp); 1118 cnp->cn_flags |= CNP_PDIRUNLOCK; 1119 } 1120 return (0); 1121 } 1122 1123 if (flags & CNP_ISDOTDOT) { 1124 vn_unlock(dvp); 1125 cnp->cn_flags |= CNP_PDIRUNLOCK; 1126 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1127 if (error) { 1128 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1129 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1130 return (error); /* NOTE: return error from nget */ 1131 } 1132 newvp = NFSTOV(np); 1133 if (lockparent) { 1134 error = vn_lock(dvp, LK_EXCLUSIVE); 1135 if (error) { 1136 vput(newvp); 1137 return (error); 1138 } 1139 cnp->cn_flags |= CNP_PDIRUNLOCK; 1140 } 1141 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1142 vref(dvp); 1143 newvp = dvp; 1144 } else { 1145 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1146 if (error) { 1147 m_freem(info.mrep); 1148 info.mrep = NULL; 1149 return (error); 1150 } 1151 if (!lockparent) { 1152 vn_unlock(dvp); 1153 cnp->cn_flags |= CNP_PDIRUNLOCK; 1154 } 1155 newvp = NFSTOV(np); 1156 } 1157 if (info.v3) { 1158 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1159 NFS_LATTR_NOSHRINK)); 1160 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1161 NFS_LATTR_NOSHRINK)); 1162 } else { 1163 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1164 } 1165 #if 0 1166 /* XXX MOVE TO nfs_nremove() */ 1167 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1168 cnp->cn_nameiop != NAMEI_DELETE) { 1169 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1170 } 1171 #endif 1172 *vpp = newvp; 1173 m_freem(info.mrep); 1174 info.mrep = NULL; 1175 nfsmout: 1176 if (error) { 1177 if (newvp != NULLVP) { 1178 vrele(newvp); 1179 *vpp = NULLVP; 1180 } 1181 if ((cnp->cn_nameiop == NAMEI_CREATE || 1182 cnp->cn_nameiop == NAMEI_RENAME) && 1183 error == ENOENT) { 1184 if (!lockparent) { 1185 vn_unlock(dvp); 1186 cnp->cn_flags |= CNP_PDIRUNLOCK; 1187 } 1188 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1189 error = EROFS; 1190 else 1191 error = EJUSTRETURN; 1192 } 1193 } 1194 return (error); 1195 } 1196 1197 /* 1198 * nfs read call. 1199 * Just call nfs_bioread() to do the work. 1200 * 1201 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1202 * struct ucred *a_cred) 1203 */ 1204 static int 1205 nfs_read(struct vop_read_args *ap) 1206 { 1207 struct vnode *vp = ap->a_vp; 1208 1209 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1210 } 1211 1212 /* 1213 * nfs readlink call 1214 * 1215 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1216 */ 1217 static int 1218 nfs_readlink(struct vop_readlink_args *ap) 1219 { 1220 struct vnode *vp = ap->a_vp; 1221 1222 if (vp->v_type != VLNK) 1223 return (EINVAL); 1224 return (nfs_bioread(vp, ap->a_uio, 0)); 1225 } 1226 1227 /* 1228 * Do a readlink rpc. 1229 * Called by nfs_doio() from below the buffer cache. 1230 */ 1231 int 1232 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop) 1233 { 1234 int error = 0, len, attrflag; 1235 struct nfsm_info info; 1236 1237 info.mrep = NULL; 1238 info.v3 = NFS_ISV3(vp); 1239 1240 nfsstats.rpccnt[NFSPROC_READLINK]++; 1241 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3)); 1242 ERROROUT(nfsm_fhtom(&info, vp)); 1243 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td, 1244 nfs_vpcred(vp, ND_CHECK), &error)); 1245 if (info.v3) { 1246 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1247 NFS_LATTR_NOSHRINK)); 1248 } 1249 if (!error) { 1250 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN)); 1251 if (len == NFS_MAXPATHLEN) { 1252 struct nfsnode *np = VTONFS(vp); 1253 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1254 len = np->n_size; 1255 } 1256 ERROROUT(nfsm_mtouio(&info, uiop, len)); 1257 } 1258 m_freem(info.mrep); 1259 info.mrep = NULL; 1260 nfsmout: 1261 return (error); 1262 } 1263 1264 /* 1265 * nfs synchronous read rpc using UIO 1266 */ 1267 int 1268 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop) 1269 { 1270 u_int32_t *tl; 1271 struct nfsmount *nmp; 1272 int error = 0, len, retlen, tsiz, eof, attrflag; 1273 struct nfsm_info info; 1274 off_t tmp_off; 1275 1276 info.mrep = NULL; 1277 info.v3 = NFS_ISV3(vp); 1278 1279 #ifndef nolint 1280 eof = 0; 1281 #endif 1282 nmp = VFSTONFS(vp->v_mount); 1283 tsiz = uiop->uio_resid; 1284 tmp_off = uiop->uio_offset + tsiz; 1285 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) 1286 return (EFBIG); 1287 tmp_off = uiop->uio_offset; 1288 while (tsiz > 0) { 1289 nfsstats.rpccnt[NFSPROC_READ]++; 1290 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1291 nfsm_reqhead(&info, vp, NFSPROC_READ, 1292 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3); 1293 ERROROUT(nfsm_fhtom(&info, vp)); 1294 tl = nfsm_build(&info, NFSX_UNSIGNED * 3); 1295 if (info.v3) { 1296 txdr_hyper(uiop->uio_offset, tl); 1297 *(tl + 2) = txdr_unsigned(len); 1298 } else { 1299 *tl++ = txdr_unsigned(uiop->uio_offset); 1300 *tl++ = txdr_unsigned(len); 1301 *tl = 0; 1302 } 1303 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td, 1304 nfs_vpcred(vp, ND_READ), &error)); 1305 if (info.v3) { 1306 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1307 NFS_LATTR_NOSHRINK)); 1308 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 1309 eof = fxdr_unsigned(int, *(tl + 1)); 1310 } else { 1311 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1312 } 1313 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len)); 1314 ERROROUT(nfsm_mtouio(&info, uiop, retlen)); 1315 m_freem(info.mrep); 1316 info.mrep = NULL; 1317 1318 /* 1319 * Handle short-read from server (NFSv3). If EOF is not 1320 * flagged (and no error occurred), but retlen is less 1321 * then the request size, we must zero-fill the remainder. 1322 */ 1323 if (retlen < len && info.v3 && eof == 0) { 1324 ERROROUT(uiomovez(len - retlen, uiop)); 1325 retlen = len; 1326 } 1327 tsiz -= retlen; 1328 1329 /* 1330 * Terminate loop on EOF or zero-length read. 1331 * 1332 * For NFSv2 a short-read indicates EOF, not zero-fill, 1333 * and also terminates the loop. 1334 */ 1335 if (info.v3) { 1336 if (eof || retlen == 0) 1337 tsiz = 0; 1338 } else if (retlen < len) { 1339 tsiz = 0; 1340 } 1341 } 1342 nfsmout: 1343 return (error); 1344 } 1345 1346 /* 1347 * nfs write call 1348 */ 1349 int 1350 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop, 1351 int *iomode, int *must_commit) 1352 { 1353 u_int32_t *tl; 1354 int32_t backup; 1355 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1356 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1357 int committed = NFSV3WRITE_FILESYNC; 1358 struct nfsm_info info; 1359 1360 info.mrep = NULL; 1361 info.v3 = NFS_ISV3(vp); 1362 1363 #ifndef DIAGNOSTIC 1364 if (uiop->uio_iovcnt != 1) 1365 panic("nfs: writerpc iovcnt > 1"); 1366 #endif 1367 *must_commit = 0; 1368 tsiz = uiop->uio_resid; 1369 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1370 return (EFBIG); 1371 while (tsiz > 0) { 1372 nfsstats.rpccnt[NFSPROC_WRITE]++; 1373 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1374 nfsm_reqhead(&info, vp, NFSPROC_WRITE, 1375 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1376 ERROROUT(nfsm_fhtom(&info, vp)); 1377 if (info.v3) { 1378 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 1379 txdr_hyper(uiop->uio_offset, tl); 1380 tl += 2; 1381 *tl++ = txdr_unsigned(len); 1382 *tl++ = txdr_unsigned(*iomode); 1383 *tl = txdr_unsigned(len); 1384 } else { 1385 u_int32_t x; 1386 1387 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED); 1388 /* Set both "begin" and "current" to non-garbage. */ 1389 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1390 *tl++ = x; /* "begin offset" */ 1391 *tl++ = x; /* "current offset" */ 1392 x = txdr_unsigned(len); 1393 *tl++ = x; /* total to this offset */ 1394 *tl = x; /* size of this write */ 1395 } 1396 ERROROUT(nfsm_uiotom(&info, uiop, len)); 1397 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td, 1398 nfs_vpcred(vp, ND_WRITE), &error)); 1399 if (info.v3) { 1400 /* 1401 * The write RPC returns a before and after mtime. The 1402 * nfsm_wcc_data() macro checks the before n_mtime 1403 * against the before time and stores the after time 1404 * in the nfsnode's cached vattr and n_mtime field. 1405 * The NRMODIFIED bit will be set if the before 1406 * time did not match the original mtime. 1407 */ 1408 wccflag = NFSV3_WCCCHK; 1409 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 1410 if (error == 0) { 1411 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF)); 1412 rlen = fxdr_unsigned(int, *tl++); 1413 if (rlen == 0) { 1414 error = NFSERR_IO; 1415 m_freem(info.mrep); 1416 info.mrep = NULL; 1417 break; 1418 } else if (rlen < len) { 1419 backup = len - rlen; 1420 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup; 1421 uiop->uio_iov->iov_len += backup; 1422 uiop->uio_offset -= backup; 1423 uiop->uio_resid += backup; 1424 len = rlen; 1425 } 1426 commit = fxdr_unsigned(int, *tl++); 1427 1428 /* 1429 * Return the lowest committment level 1430 * obtained by any of the RPCs. 1431 */ 1432 if (committed == NFSV3WRITE_FILESYNC) 1433 committed = commit; 1434 else if (committed == NFSV3WRITE_DATASYNC && 1435 commit == NFSV3WRITE_UNSTABLE) 1436 committed = commit; 1437 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1438 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1439 NFSX_V3WRITEVERF); 1440 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1441 } else if (bcmp((caddr_t)tl, 1442 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1443 *must_commit = 1; 1444 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1445 NFSX_V3WRITEVERF); 1446 } 1447 } 1448 } else { 1449 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1450 } 1451 m_freem(info.mrep); 1452 info.mrep = NULL; 1453 if (error) 1454 break; 1455 tsiz -= len; 1456 } 1457 nfsmout: 1458 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1459 committed = NFSV3WRITE_FILESYNC; 1460 *iomode = committed; 1461 if (error) 1462 uiop->uio_resid = tsiz; 1463 return (error); 1464 } 1465 1466 /* 1467 * nfs mknod rpc 1468 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1469 * mode set to specify the file type and the size field for rdev. 1470 */ 1471 static int 1472 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1473 struct vattr *vap) 1474 { 1475 struct nfsv2_sattr *sp; 1476 u_int32_t *tl; 1477 struct vnode *newvp = NULL; 1478 struct nfsnode *np = NULL; 1479 struct vattr vattr; 1480 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1481 int rmajor, rminor; 1482 struct nfsm_info info; 1483 1484 info.mrep = NULL; 1485 info.v3 = NFS_ISV3(dvp); 1486 1487 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1488 rmajor = txdr_unsigned(vap->va_rmajor); 1489 rminor = txdr_unsigned(vap->va_rminor); 1490 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) { 1491 rmajor = nfs_xdrneg1; 1492 rminor = nfs_xdrneg1; 1493 } else { 1494 return (EOPNOTSUPP); 1495 } 1496 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1497 return (error); 1498 } 1499 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1500 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD, 1501 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED + 1502 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1503 ERROROUT(nfsm_fhtom(&info, dvp)); 1504 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1505 NFS_MAXNAMLEN)); 1506 if (info.v3) { 1507 tl = nfsm_build(&info, NFSX_UNSIGNED); 1508 *tl++ = vtonfsv3_type(vap->va_type); 1509 nfsm_v3attrbuild(&info, vap, FALSE); 1510 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1511 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 1512 *tl++ = txdr_unsigned(vap->va_rmajor); 1513 *tl = txdr_unsigned(vap->va_rminor); 1514 } 1515 } else { 1516 sp = nfsm_build(&info, NFSX_V2SATTR); 1517 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1518 sp->sa_uid = nfs_xdrneg1; 1519 sp->sa_gid = nfs_xdrneg1; 1520 sp->sa_size = makeudev(rmajor, rminor); 1521 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1522 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1523 } 1524 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td, 1525 cnp->cn_cred, &error)); 1526 if (!error) { 1527 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1528 if (!gotvp) { 1529 if (newvp) { 1530 vput(newvp); 1531 newvp = NULL; 1532 } 1533 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1534 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1535 if (!error) 1536 newvp = NFSTOV(np); 1537 } 1538 } 1539 if (info.v3) { 1540 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1541 } 1542 m_freem(info.mrep); 1543 info.mrep = NULL; 1544 nfsmout: 1545 if (error) { 1546 if (newvp) 1547 vput(newvp); 1548 } else { 1549 *vpp = newvp; 1550 } 1551 VTONFS(dvp)->n_flag |= NLMODIFIED; 1552 if (!wccflag) 1553 VTONFS(dvp)->n_attrstamp = 0; 1554 return (error); 1555 } 1556 1557 /* 1558 * nfs mknod vop 1559 * just call nfs_mknodrpc() to do the work. 1560 * 1561 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1562 * struct componentname *a_cnp, struct vattr *a_vap) 1563 */ 1564 /* ARGSUSED */ 1565 static int 1566 nfs_mknod(struct vop_old_mknod_args *ap) 1567 { 1568 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1569 } 1570 1571 static u_long create_verf; 1572 /* 1573 * nfs file create call 1574 * 1575 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1576 * struct componentname *a_cnp, struct vattr *a_vap) 1577 */ 1578 static int 1579 nfs_create(struct vop_old_create_args *ap) 1580 { 1581 struct vnode *dvp = ap->a_dvp; 1582 struct vattr *vap = ap->a_vap; 1583 struct componentname *cnp = ap->a_cnp; 1584 struct nfsv2_sattr *sp; 1585 u_int32_t *tl; 1586 struct nfsnode *np = NULL; 1587 struct vnode *newvp = NULL; 1588 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1589 struct vattr vattr; 1590 struct nfsm_info info; 1591 1592 info.mrep = NULL; 1593 info.v3 = NFS_ISV3(dvp); 1594 1595 /* 1596 * Oops, not for me.. 1597 */ 1598 if (vap->va_type == VSOCK) 1599 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1600 1601 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1602 return (error); 1603 } 1604 if (vap->va_vaflags & VA_EXCLUSIVE) 1605 fmode |= O_EXCL; 1606 again: 1607 nfsstats.rpccnt[NFSPROC_CREATE]++; 1608 nfsm_reqhead(&info, dvp, NFSPROC_CREATE, 1609 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED + 1610 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1611 ERROROUT(nfsm_fhtom(&info, dvp)); 1612 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1613 NFS_MAXNAMLEN)); 1614 if (info.v3) { 1615 tl = nfsm_build(&info, NFSX_UNSIGNED); 1616 if (fmode & O_EXCL) { 1617 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1618 tl = nfsm_build(&info, NFSX_V3CREATEVERF); 1619 #ifdef INET 1620 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1621 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr; 1622 else 1623 #endif 1624 *tl++ = create_verf; 1625 *tl = ++create_verf; 1626 } else { 1627 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1628 nfsm_v3attrbuild(&info, vap, FALSE); 1629 } 1630 } else { 1631 sp = nfsm_build(&info, NFSX_V2SATTR); 1632 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1633 sp->sa_uid = nfs_xdrneg1; 1634 sp->sa_gid = nfs_xdrneg1; 1635 sp->sa_size = 0; 1636 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1637 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1638 } 1639 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td, 1640 cnp->cn_cred, &error)); 1641 if (error == 0) { 1642 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1643 if (!gotvp) { 1644 if (newvp) { 1645 vput(newvp); 1646 newvp = NULL; 1647 } 1648 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1649 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1650 if (!error) 1651 newvp = NFSTOV(np); 1652 } 1653 } 1654 if (info.v3) { 1655 if (error == 0) 1656 error = nfsm_wcc_data(&info, dvp, &wccflag); 1657 else 1658 (void)nfsm_wcc_data(&info, dvp, &wccflag); 1659 } 1660 m_freem(info.mrep); 1661 info.mrep = NULL; 1662 nfsmout: 1663 if (error) { 1664 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1665 KKASSERT(newvp == NULL); 1666 fmode &= ~O_EXCL; 1667 goto again; 1668 } 1669 } else if (info.v3 && (fmode & O_EXCL)) { 1670 /* 1671 * We are normally called with only a partially initialized 1672 * VAP. Since the NFSv3 spec says that server may use the 1673 * file attributes to store the verifier, the spec requires 1674 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1675 * in atime, but we can't really assume that all servers will 1676 * so we ensure that our SETATTR sets both atime and mtime. 1677 */ 1678 if (vap->va_mtime.tv_sec == VNOVAL) 1679 vfs_timestamp(&vap->va_mtime); 1680 if (vap->va_atime.tv_sec == VNOVAL) 1681 vap->va_atime = vap->va_mtime; 1682 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1683 } 1684 if (error == 0) { 1685 /* 1686 * The new np may have enough info for access 1687 * checks, make sure rucred and wucred are 1688 * initialized for read and write rpc's. 1689 */ 1690 np = VTONFS(newvp); 1691 if (np->n_rucred == NULL) 1692 np->n_rucred = crhold(cnp->cn_cred); 1693 if (np->n_wucred == NULL) 1694 np->n_wucred = crhold(cnp->cn_cred); 1695 *ap->a_vpp = newvp; 1696 } else if (newvp) { 1697 vput(newvp); 1698 } 1699 VTONFS(dvp)->n_flag |= NLMODIFIED; 1700 if (!wccflag) 1701 VTONFS(dvp)->n_attrstamp = 0; 1702 return (error); 1703 } 1704 1705 /* 1706 * nfs file remove call 1707 * To try and make nfs semantics closer to ufs semantics, a file that has 1708 * other processes using the vnode is renamed instead of removed and then 1709 * removed later on the last close. 1710 * - If v_sysref.refcnt > 1 1711 * If a rename is not already in the works 1712 * call nfs_sillyrename() to set it up 1713 * else 1714 * do the remove rpc 1715 * 1716 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1717 * struct componentname *a_cnp) 1718 */ 1719 static int 1720 nfs_remove(struct vop_old_remove_args *ap) 1721 { 1722 struct vnode *vp = ap->a_vp; 1723 struct vnode *dvp = ap->a_dvp; 1724 struct componentname *cnp = ap->a_cnp; 1725 struct nfsnode *np = VTONFS(vp); 1726 int error = 0; 1727 struct vattr vattr; 1728 1729 #ifndef DIAGNOSTIC 1730 if (vp->v_sysref.refcnt < 1) 1731 panic("nfs_remove: bad v_sysref.refcnt"); 1732 #endif 1733 if (vp->v_type == VDIR) 1734 error = EPERM; 1735 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename && 1736 VOP_GETATTR(vp, &vattr) == 0 && 1737 vattr.va_nlink > 1)) { 1738 /* 1739 * throw away biocache buffers, mainly to avoid 1740 * unnecessary delayed writes later. 1741 */ 1742 error = nfs_vinvalbuf(vp, 0, 1); 1743 /* Do the rpc */ 1744 if (error != EINTR) 1745 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1746 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1747 /* 1748 * Kludge City: If the first reply to the remove rpc is lost.. 1749 * the reply to the retransmitted request will be ENOENT 1750 * since the file was in fact removed 1751 * Therefore, we cheat and return success. 1752 */ 1753 if (error == ENOENT) 1754 error = 0; 1755 } else if (!np->n_sillyrename) { 1756 error = nfs_sillyrename(dvp, vp, cnp); 1757 } 1758 np->n_attrstamp = 0; 1759 return (error); 1760 } 1761 1762 /* 1763 * nfs file remove rpc called from nfs_inactive 1764 */ 1765 int 1766 nfs_removeit(struct sillyrename *sp) 1767 { 1768 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1769 sp->s_cred, NULL)); 1770 } 1771 1772 /* 1773 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1774 */ 1775 static int 1776 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1777 struct ucred *cred, struct thread *td) 1778 { 1779 int error = 0, wccflag = NFSV3_WCCRATTR; 1780 struct nfsm_info info; 1781 1782 info.mrep = NULL; 1783 info.v3 = NFS_ISV3(dvp); 1784 1785 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1786 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE, 1787 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1788 ERROROUT(nfsm_fhtom(&info, dvp)); 1789 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN)); 1790 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error)); 1791 if (info.v3) { 1792 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1793 } 1794 m_freem(info.mrep); 1795 info.mrep = NULL; 1796 nfsmout: 1797 VTONFS(dvp)->n_flag |= NLMODIFIED; 1798 if (!wccflag) 1799 VTONFS(dvp)->n_attrstamp = 0; 1800 return (error); 1801 } 1802 1803 /* 1804 * nfs file rename call 1805 * 1806 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1807 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1808 * struct vnode *a_tvp, struct componentname *a_tcnp) 1809 */ 1810 static int 1811 nfs_rename(struct vop_old_rename_args *ap) 1812 { 1813 struct vnode *fvp = ap->a_fvp; 1814 struct vnode *tvp = ap->a_tvp; 1815 struct vnode *fdvp = ap->a_fdvp; 1816 struct vnode *tdvp = ap->a_tdvp; 1817 struct componentname *tcnp = ap->a_tcnp; 1818 struct componentname *fcnp = ap->a_fcnp; 1819 int error; 1820 1821 /* Check for cross-device rename */ 1822 if ((fvp->v_mount != tdvp->v_mount) || 1823 (tvp && (fvp->v_mount != tvp->v_mount))) { 1824 error = EXDEV; 1825 goto out; 1826 } 1827 1828 /* 1829 * We shouldn't have to flush fvp on rename for most server-side 1830 * filesystems as the file handle should not change. Unfortunately 1831 * the inode for some filesystems (msdosfs) might be tied to the 1832 * file name or directory position so to be completely safe 1833 * vfs.nfs.flush_on_rename is set by default. Clear to improve 1834 * performance. 1835 * 1836 * We must flush tvp on rename because it might become stale on the 1837 * server after the rename. 1838 */ 1839 if (nfs_flush_on_rename) 1840 VOP_FSYNC(fvp, MNT_WAIT, 0); 1841 if (tvp) 1842 VOP_FSYNC(tvp, MNT_WAIT, 0); 1843 1844 /* 1845 * If the tvp exists and is in use, sillyrename it before doing the 1846 * rename of the new file over it. 1847 * 1848 * XXX Can't sillyrename a directory. 1849 * 1850 * We do not attempt to do any namecache purges in this old API 1851 * routine. The new API compat functions have access to the actual 1852 * namecache structures and will do it for us. 1853 */ 1854 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename && 1855 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1856 vput(tvp); 1857 tvp = NULL; 1858 } else if (tvp) { 1859 ; 1860 } 1861 1862 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1863 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1864 tcnp->cn_td); 1865 1866 out: 1867 if (tdvp == tvp) 1868 vrele(tdvp); 1869 else 1870 vput(tdvp); 1871 if (tvp) 1872 vput(tvp); 1873 vrele(fdvp); 1874 vrele(fvp); 1875 /* 1876 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1877 */ 1878 if (error == ENOENT) 1879 error = 0; 1880 return (error); 1881 } 1882 1883 /* 1884 * nfs file rename rpc called from nfs_remove() above 1885 */ 1886 static int 1887 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1888 struct sillyrename *sp) 1889 { 1890 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1891 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1892 } 1893 1894 /* 1895 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1896 */ 1897 static int 1898 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1899 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1900 struct ucred *cred, struct thread *td) 1901 { 1902 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1903 struct nfsm_info info; 1904 1905 info.mrep = NULL; 1906 info.v3 = NFS_ISV3(fdvp); 1907 1908 nfsstats.rpccnt[NFSPROC_RENAME]++; 1909 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME, 1910 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 + 1911 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen)); 1912 ERROROUT(nfsm_fhtom(&info, fdvp)); 1913 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN)); 1914 ERROROUT(nfsm_fhtom(&info, tdvp)); 1915 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN)); 1916 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error)); 1917 if (info.v3) { 1918 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag)); 1919 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag)); 1920 } 1921 m_freem(info.mrep); 1922 info.mrep = NULL; 1923 nfsmout: 1924 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1925 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1926 if (!fwccflag) 1927 VTONFS(fdvp)->n_attrstamp = 0; 1928 if (!twccflag) 1929 VTONFS(tdvp)->n_attrstamp = 0; 1930 return (error); 1931 } 1932 1933 /* 1934 * nfs hard link create call 1935 * 1936 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1937 * struct componentname *a_cnp) 1938 */ 1939 static int 1940 nfs_link(struct vop_old_link_args *ap) 1941 { 1942 struct vnode *vp = ap->a_vp; 1943 struct vnode *tdvp = ap->a_tdvp; 1944 struct componentname *cnp = ap->a_cnp; 1945 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1946 struct nfsm_info info; 1947 1948 if (vp->v_mount != tdvp->v_mount) { 1949 return (EXDEV); 1950 } 1951 1952 /* 1953 * The attribute cache may get out of sync with the server on link. 1954 * Pushing writes to the server before handle was inherited from 1955 * long long ago and it is unclear if we still need to do this. 1956 * Defaults to off. 1957 */ 1958 if (nfs_flush_on_hlink) 1959 VOP_FSYNC(vp, MNT_WAIT, 0); 1960 1961 info.mrep = NULL; 1962 info.v3 = NFS_ISV3(vp); 1963 1964 nfsstats.rpccnt[NFSPROC_LINK]++; 1965 nfsm_reqhead(&info, vp, NFSPROC_LINK, 1966 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED + 1967 nfsm_rndup(cnp->cn_namelen)); 1968 ERROROUT(nfsm_fhtom(&info, vp)); 1969 ERROROUT(nfsm_fhtom(&info, tdvp)); 1970 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1971 NFS_MAXNAMLEN)); 1972 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td, 1973 cnp->cn_cred, &error)); 1974 if (info.v3) { 1975 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1976 NFS_LATTR_NOSHRINK)); 1977 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag)); 1978 } 1979 m_freem(info.mrep); 1980 info.mrep = NULL; 1981 nfsmout: 1982 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1983 if (!attrflag) 1984 VTONFS(vp)->n_attrstamp = 0; 1985 if (!wccflag) 1986 VTONFS(tdvp)->n_attrstamp = 0; 1987 /* 1988 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1989 */ 1990 if (error == EEXIST) 1991 error = 0; 1992 return (error); 1993 } 1994 1995 /* 1996 * nfs symbolic link create call 1997 * 1998 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1999 * struct componentname *a_cnp, struct vattr *a_vap, 2000 * char *a_target) 2001 */ 2002 static int 2003 nfs_symlink(struct vop_old_symlink_args *ap) 2004 { 2005 struct vnode *dvp = ap->a_dvp; 2006 struct vattr *vap = ap->a_vap; 2007 struct componentname *cnp = ap->a_cnp; 2008 struct nfsv2_sattr *sp; 2009 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2010 struct vnode *newvp = NULL; 2011 struct nfsm_info info; 2012 2013 info.mrep = NULL; 2014 info.v3 = NFS_ISV3(dvp); 2015 2016 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2017 slen = strlen(ap->a_target); 2018 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK, 2019 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED + 2020 nfsm_rndup(cnp->cn_namelen) + 2021 nfsm_rndup(slen) + NFSX_SATTR(info.v3)); 2022 ERROROUT(nfsm_fhtom(&info, dvp)); 2023 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2024 NFS_MAXNAMLEN)); 2025 if (info.v3) { 2026 nfsm_v3attrbuild(&info, vap, FALSE); 2027 } 2028 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN)); 2029 if (info.v3 == 0) { 2030 sp = nfsm_build(&info, NFSX_V2SATTR); 2031 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2032 sp->sa_uid = nfs_xdrneg1; 2033 sp->sa_gid = nfs_xdrneg1; 2034 sp->sa_size = nfs_xdrneg1; 2035 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2036 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2037 } 2038 2039 /* 2040 * Issue the NFS request and get the rpc response. 2041 * 2042 * Only NFSv3 responses returning an error of 0 actually return 2043 * a file handle that can be converted into newvp without having 2044 * to do an extra lookup rpc. 2045 */ 2046 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td, 2047 cnp->cn_cred, &error)); 2048 if (info.v3) { 2049 if (error == 0) { 2050 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2051 } 2052 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2053 } 2054 2055 /* 2056 * out code jumps -> here, mrep is also freed. 2057 */ 2058 2059 m_freem(info.mrep); 2060 info.mrep = NULL; 2061 nfsmout: 2062 2063 /* 2064 * If we get an EEXIST error, silently convert it to no-error 2065 * in case of an NFS retry. 2066 */ 2067 if (error == EEXIST) 2068 error = 0; 2069 2070 /* 2071 * If we do not have (or no longer have) an error, and we could 2072 * not extract the newvp from the response due to the request being 2073 * NFSv2 or the error being EEXIST. We have to do a lookup in order 2074 * to obtain a newvp to return. 2075 */ 2076 if (error == 0 && newvp == NULL) { 2077 struct nfsnode *np = NULL; 2078 2079 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2080 cnp->cn_cred, cnp->cn_td, &np); 2081 if (!error) 2082 newvp = NFSTOV(np); 2083 } 2084 if (error) { 2085 if (newvp) 2086 vput(newvp); 2087 } else { 2088 *ap->a_vpp = newvp; 2089 } 2090 VTONFS(dvp)->n_flag |= NLMODIFIED; 2091 if (!wccflag) 2092 VTONFS(dvp)->n_attrstamp = 0; 2093 return (error); 2094 } 2095 2096 /* 2097 * nfs make dir call 2098 * 2099 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2100 * struct componentname *a_cnp, struct vattr *a_vap) 2101 */ 2102 static int 2103 nfs_mkdir(struct vop_old_mkdir_args *ap) 2104 { 2105 struct vnode *dvp = ap->a_dvp; 2106 struct vattr *vap = ap->a_vap; 2107 struct componentname *cnp = ap->a_cnp; 2108 struct nfsv2_sattr *sp; 2109 struct nfsnode *np = NULL; 2110 struct vnode *newvp = NULL; 2111 struct vattr vattr; 2112 int error = 0, wccflag = NFSV3_WCCRATTR; 2113 int gotvp = 0; 2114 int len; 2115 struct nfsm_info info; 2116 2117 info.mrep = NULL; 2118 info.v3 = NFS_ISV3(dvp); 2119 2120 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 2121 return (error); 2122 } 2123 len = cnp->cn_namelen; 2124 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2125 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR, 2126 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2127 nfsm_rndup(len) + NFSX_SATTR(info.v3)); 2128 ERROROUT(nfsm_fhtom(&info, dvp)); 2129 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 2130 if (info.v3) { 2131 nfsm_v3attrbuild(&info, vap, FALSE); 2132 } else { 2133 sp = nfsm_build(&info, NFSX_V2SATTR); 2134 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2135 sp->sa_uid = nfs_xdrneg1; 2136 sp->sa_gid = nfs_xdrneg1; 2137 sp->sa_size = nfs_xdrneg1; 2138 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2139 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2140 } 2141 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td, 2142 cnp->cn_cred, &error)); 2143 if (error == 0) { 2144 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2145 } 2146 if (info.v3) { 2147 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2148 } 2149 m_freem(info.mrep); 2150 info.mrep = NULL; 2151 nfsmout: 2152 VTONFS(dvp)->n_flag |= NLMODIFIED; 2153 if (!wccflag) 2154 VTONFS(dvp)->n_attrstamp = 0; 2155 /* 2156 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2157 * if we can succeed in looking up the directory. 2158 */ 2159 if (error == EEXIST || (!error && !gotvp)) { 2160 if (newvp) { 2161 vrele(newvp); 2162 newvp = NULL; 2163 } 2164 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2165 cnp->cn_td, &np); 2166 if (!error) { 2167 newvp = NFSTOV(np); 2168 if (newvp->v_type != VDIR) 2169 error = EEXIST; 2170 } 2171 } 2172 if (error) { 2173 if (newvp) 2174 vrele(newvp); 2175 } else 2176 *ap->a_vpp = newvp; 2177 return (error); 2178 } 2179 2180 /* 2181 * nfs remove directory call 2182 * 2183 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2184 * struct componentname *a_cnp) 2185 */ 2186 static int 2187 nfs_rmdir(struct vop_old_rmdir_args *ap) 2188 { 2189 struct vnode *vp = ap->a_vp; 2190 struct vnode *dvp = ap->a_dvp; 2191 struct componentname *cnp = ap->a_cnp; 2192 int error = 0, wccflag = NFSV3_WCCRATTR; 2193 struct nfsm_info info; 2194 2195 info.mrep = NULL; 2196 info.v3 = NFS_ISV3(dvp); 2197 2198 if (dvp == vp) 2199 return (EINVAL); 2200 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2201 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR, 2202 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2203 nfsm_rndup(cnp->cn_namelen)); 2204 ERROROUT(nfsm_fhtom(&info, dvp)); 2205 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2206 NFS_MAXNAMLEN)); 2207 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td, 2208 cnp->cn_cred, &error)); 2209 if (info.v3) { 2210 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2211 } 2212 m_freem(info.mrep); 2213 info.mrep = NULL; 2214 nfsmout: 2215 VTONFS(dvp)->n_flag |= NLMODIFIED; 2216 if (!wccflag) 2217 VTONFS(dvp)->n_attrstamp = 0; 2218 /* 2219 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2220 */ 2221 if (error == ENOENT) 2222 error = 0; 2223 return (error); 2224 } 2225 2226 /* 2227 * nfs readdir call 2228 * 2229 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2230 */ 2231 static int 2232 nfs_readdir(struct vop_readdir_args *ap) 2233 { 2234 struct vnode *vp = ap->a_vp; 2235 struct nfsnode *np = VTONFS(vp); 2236 struct uio *uio = ap->a_uio; 2237 int tresid, error; 2238 struct vattr vattr; 2239 2240 if (vp->v_type != VDIR) 2241 return (EPERM); 2242 2243 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 2244 return (error); 2245 2246 /* 2247 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2248 * and then check that is still valid, or if this is an NQNFS mount 2249 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2250 * VOP_GETATTR() does not necessarily go to the wire. 2251 */ 2252 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2253 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2254 if (VOP_GETATTR(vp, &vattr) == 0 && 2255 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2256 ) { 2257 nfsstats.direofcache_hits++; 2258 goto done; 2259 } 2260 } 2261 2262 /* 2263 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2264 * own cache coherency checks so we do not have to. 2265 */ 2266 tresid = uio->uio_resid; 2267 error = nfs_bioread(vp, uio, 0); 2268 2269 if (!error && uio->uio_resid == tresid) 2270 nfsstats.direofcache_misses++; 2271 done: 2272 vn_unlock(vp); 2273 return (error); 2274 } 2275 2276 /* 2277 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2278 * 2279 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2280 * offset/block and converts the nfs formatted directory entries for userland 2281 * consumption as well as deals with offsets into the middle of blocks. 2282 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2283 * be block-bounded. It must convert to cookies for the actual RPC. 2284 */ 2285 int 2286 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop) 2287 { 2288 int len, left; 2289 struct nfs_dirent *dp = NULL; 2290 u_int32_t *tl; 2291 nfsuint64 *cookiep; 2292 caddr_t cp; 2293 nfsuint64 cookie; 2294 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2295 struct nfsnode *dnp = VTONFS(vp); 2296 u_quad_t fileno; 2297 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2298 int attrflag; 2299 struct nfsm_info info; 2300 2301 info.mrep = NULL; 2302 info.v3 = NFS_ISV3(vp); 2303 2304 #ifndef DIAGNOSTIC 2305 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2306 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2307 panic("nfs readdirrpc bad uio"); 2308 #endif 2309 2310 /* 2311 * If there is no cookie, assume directory was stale. 2312 */ 2313 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2314 if (cookiep) 2315 cookie = *cookiep; 2316 else 2317 return (NFSERR_BAD_COOKIE); 2318 /* 2319 * Loop around doing readdir rpc's of size nm_readdirsize 2320 * truncated to a multiple of DIRBLKSIZ. 2321 * The stopping criteria is EOF or buffer full. 2322 */ 2323 while (more_dirs && bigenough) { 2324 nfsstats.rpccnt[NFSPROC_READDIR]++; 2325 nfsm_reqhead(&info, vp, NFSPROC_READDIR, 2326 NFSX_FH(info.v3) + NFSX_READDIR(info.v3)); 2327 ERROROUT(nfsm_fhtom(&info, vp)); 2328 if (info.v3) { 2329 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 2330 *tl++ = cookie.nfsuquad[0]; 2331 *tl++ = cookie.nfsuquad[1]; 2332 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2333 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2334 } else { 2335 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 2336 *tl++ = cookie.nfsuquad[0]; 2337 } 2338 *tl = txdr_unsigned(nmp->nm_readdirsize); 2339 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR, 2340 uiop->uio_td, 2341 nfs_vpcred(vp, ND_READ), &error)); 2342 if (info.v3) { 2343 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2344 NFS_LATTR_NOSHRINK)); 2345 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2346 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2347 dnp->n_cookieverf.nfsuquad[1] = *tl; 2348 } 2349 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2350 more_dirs = fxdr_unsigned(int, *tl); 2351 2352 /* loop thru the dir entries, converting them to std form */ 2353 while (more_dirs && bigenough) { 2354 if (info.v3) { 2355 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2356 fileno = fxdr_hyper(tl); 2357 len = fxdr_unsigned(int, *(tl + 2)); 2358 } else { 2359 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2360 fileno = fxdr_unsigned(u_quad_t, *tl++); 2361 len = fxdr_unsigned(int, *tl); 2362 } 2363 if (len <= 0 || len > NFS_MAXNAMLEN) { 2364 error = EBADRPC; 2365 m_freem(info.mrep); 2366 info.mrep = NULL; 2367 goto nfsmout; 2368 } 2369 2370 /* 2371 * len is the number of bytes in the path element 2372 * name, not including the \0 termination. 2373 * 2374 * tlen is the number of bytes w have to reserve for 2375 * the path element name. 2376 */ 2377 tlen = nfsm_rndup(len); 2378 if (tlen == len) 2379 tlen += 4; /* To ensure null termination */ 2380 2381 /* 2382 * If the entry would cross a DIRBLKSIZ boundary, 2383 * extend the previous nfs_dirent to cover the 2384 * remaining space. 2385 */ 2386 left = DIRBLKSIZ - blksiz; 2387 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2388 dp->nfs_reclen += left; 2389 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2390 uiop->uio_iov->iov_len -= left; 2391 uiop->uio_offset += left; 2392 uiop->uio_resid -= left; 2393 blksiz = 0; 2394 } 2395 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2396 bigenough = 0; 2397 if (bigenough) { 2398 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2399 dp->nfs_ino = fileno; 2400 dp->nfs_namlen = len; 2401 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2402 dp->nfs_type = DT_UNKNOWN; 2403 blksiz += dp->nfs_reclen; 2404 if (blksiz == DIRBLKSIZ) 2405 blksiz = 0; 2406 uiop->uio_offset += sizeof(struct nfs_dirent); 2407 uiop->uio_resid -= sizeof(struct nfs_dirent); 2408 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2409 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2410 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2411 2412 /* 2413 * The uiop has advanced by nfs_dirent + len 2414 * but really needs to advance by 2415 * nfs_dirent + tlen 2416 */ 2417 cp = uiop->uio_iov->iov_base; 2418 tlen -= len; 2419 *cp = '\0'; /* null terminate */ 2420 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2421 uiop->uio_iov->iov_len -= tlen; 2422 uiop->uio_offset += tlen; 2423 uiop->uio_resid -= tlen; 2424 } else { 2425 /* 2426 * NFS strings must be rounded up (nfsm_myouio 2427 * handled that in the bigenough case). 2428 */ 2429 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2430 } 2431 if (info.v3) { 2432 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2433 } else { 2434 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2435 } 2436 2437 /* 2438 * If we were able to accomodate the last entry, 2439 * get the cookie for the next one. Otherwise 2440 * hold-over the cookie for the one we were not 2441 * able to accomodate. 2442 */ 2443 if (bigenough) { 2444 cookie.nfsuquad[0] = *tl++; 2445 if (info.v3) 2446 cookie.nfsuquad[1] = *tl++; 2447 } else if (info.v3) { 2448 tl += 2; 2449 } else { 2450 tl++; 2451 } 2452 more_dirs = fxdr_unsigned(int, *tl); 2453 } 2454 /* 2455 * If at end of rpc data, get the eof boolean 2456 */ 2457 if (!more_dirs) { 2458 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2459 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2460 } 2461 m_freem(info.mrep); 2462 info.mrep = NULL; 2463 } 2464 /* 2465 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2466 * by increasing d_reclen for the last record. 2467 */ 2468 if (blksiz > 0) { 2469 left = DIRBLKSIZ - blksiz; 2470 dp->nfs_reclen += left; 2471 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2472 uiop->uio_iov->iov_len -= left; 2473 uiop->uio_offset += left; 2474 uiop->uio_resid -= left; 2475 } 2476 2477 if (bigenough) { 2478 /* 2479 * We hit the end of the directory, update direofoffset. 2480 */ 2481 dnp->n_direofoffset = uiop->uio_offset; 2482 } else { 2483 /* 2484 * There is more to go, insert the link cookie so the 2485 * next block can be read. 2486 */ 2487 if (uiop->uio_resid > 0) 2488 kprintf("EEK! readdirrpc resid > 0\n"); 2489 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2490 *cookiep = cookie; 2491 } 2492 nfsmout: 2493 return (error); 2494 } 2495 2496 /* 2497 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2498 */ 2499 int 2500 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop) 2501 { 2502 int len, left; 2503 struct nfs_dirent *dp; 2504 u_int32_t *tl; 2505 struct vnode *newvp; 2506 nfsuint64 *cookiep; 2507 caddr_t dpossav1, dpossav2; 2508 caddr_t cp; 2509 struct mbuf *mdsav1, *mdsav2; 2510 nfsuint64 cookie; 2511 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2512 struct nfsnode *dnp = VTONFS(vp), *np; 2513 nfsfh_t *fhp; 2514 u_quad_t fileno; 2515 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2516 int attrflag, fhsize; 2517 struct nchandle nch; 2518 struct nchandle dnch; 2519 struct nlcomponent nlc; 2520 struct nfsm_info info; 2521 2522 info.mrep = NULL; 2523 info.v3 = 1; 2524 2525 #ifndef nolint 2526 dp = NULL; 2527 #endif 2528 #ifndef DIAGNOSTIC 2529 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2530 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2531 panic("nfs readdirplusrpc bad uio"); 2532 #endif 2533 /* 2534 * Obtain the namecache record for the directory so we have something 2535 * to use as a basis for creating the entries. This function will 2536 * return a held (but not locked) ncp. The ncp may be disconnected 2537 * from the tree and cannot be used for upward traversals, and the 2538 * ncp may be unnamed. Note that other unrelated operations may 2539 * cause the ncp to be named at any time. 2540 * 2541 * We have to lock the ncp to prevent a lock order reversal when 2542 * rdirplus does nlookups of the children, because the vnode is 2543 * locked and has to stay that way. 2544 */ 2545 cache_fromdvp(vp, NULL, 0, &dnch); 2546 bzero(&nlc, sizeof(nlc)); 2547 newvp = NULLVP; 2548 2549 /* 2550 * If there is no cookie, assume directory was stale. 2551 */ 2552 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2553 if (cookiep) { 2554 cookie = *cookiep; 2555 } else { 2556 if (dnch.ncp) 2557 cache_drop(&dnch); 2558 return (NFSERR_BAD_COOKIE); 2559 } 2560 2561 /* 2562 * Loop around doing readdir rpc's of size nm_readdirsize 2563 * truncated to a multiple of DIRBLKSIZ. 2564 * The stopping criteria is EOF or buffer full. 2565 */ 2566 while (more_dirs && bigenough) { 2567 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2568 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS, 2569 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2570 ERROROUT(nfsm_fhtom(&info, vp)); 2571 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED); 2572 *tl++ = cookie.nfsuquad[0]; 2573 *tl++ = cookie.nfsuquad[1]; 2574 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2575 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2576 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2577 *tl = txdr_unsigned(nmp->nm_rsize); 2578 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS, 2579 uiop->uio_td, 2580 nfs_vpcred(vp, ND_READ), &error)); 2581 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2582 NFS_LATTR_NOSHRINK)); 2583 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2584 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2585 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2586 more_dirs = fxdr_unsigned(int, *tl); 2587 2588 /* loop thru the dir entries, doctoring them to 4bsd form */ 2589 while (more_dirs && bigenough) { 2590 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2591 fileno = fxdr_hyper(tl); 2592 len = fxdr_unsigned(int, *(tl + 2)); 2593 if (len <= 0 || len > NFS_MAXNAMLEN) { 2594 error = EBADRPC; 2595 m_freem(info.mrep); 2596 info.mrep = NULL; 2597 goto nfsmout; 2598 } 2599 tlen = nfsm_rndup(len); 2600 if (tlen == len) 2601 tlen += 4; /* To ensure null termination*/ 2602 left = DIRBLKSIZ - blksiz; 2603 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2604 dp->nfs_reclen += left; 2605 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2606 uiop->uio_iov->iov_len -= left; 2607 uiop->uio_offset += left; 2608 uiop->uio_resid -= left; 2609 blksiz = 0; 2610 } 2611 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2612 bigenough = 0; 2613 if (bigenough) { 2614 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2615 dp->nfs_ino = fileno; 2616 dp->nfs_namlen = len; 2617 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2618 dp->nfs_type = DT_UNKNOWN; 2619 blksiz += dp->nfs_reclen; 2620 if (blksiz == DIRBLKSIZ) 2621 blksiz = 0; 2622 uiop->uio_offset += sizeof(struct nfs_dirent); 2623 uiop->uio_resid -= sizeof(struct nfs_dirent); 2624 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2625 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2626 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2627 nlc.nlc_namelen = len; 2628 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2629 cp = uiop->uio_iov->iov_base; 2630 tlen -= len; 2631 *cp = '\0'; 2632 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2633 uiop->uio_iov->iov_len -= tlen; 2634 uiop->uio_offset += tlen; 2635 uiop->uio_resid -= tlen; 2636 } else { 2637 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2638 } 2639 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2640 if (bigenough) { 2641 cookie.nfsuquad[0] = *tl++; 2642 cookie.nfsuquad[1] = *tl++; 2643 } else { 2644 tl += 2; 2645 } 2646 2647 /* 2648 * Since the attributes are before the file handle 2649 * (sigh), we must skip over the attributes and then 2650 * come back and get them. 2651 */ 2652 attrflag = fxdr_unsigned(int, *tl); 2653 if (attrflag) { 2654 dpossav1 = info.dpos; 2655 mdsav1 = info.md; 2656 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR)); 2657 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2658 doit = fxdr_unsigned(int, *tl); 2659 if (doit) { 2660 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 2661 } 2662 if (doit && bigenough && !nlcdegenerate(&nlc) && 2663 !NFS_CMPFH(dnp, fhp, fhsize) 2664 ) { 2665 if (dnch.ncp) { 2666 #if 0 2667 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n", 2668 nlc.nlc_namelen, nlc.nlc_namelen, 2669 nlc.nlc_nameptr); 2670 #endif 2671 /* 2672 * This is a bit hokey but there isn't 2673 * much we can do about it. We can't 2674 * hold the directory vp locked while 2675 * doing lookups and gets. 2676 */ 2677 nch = cache_nlookup_nonblock(&dnch, &nlc); 2678 if (nch.ncp == NULL) 2679 goto rdfail; 2680 cache_setunresolved(&nch); 2681 error = nfs_nget_nonblock(vp->v_mount, fhp, 2682 fhsize, &np); 2683 if (error) { 2684 cache_put(&nch); 2685 goto rdfail; 2686 } 2687 newvp = NFSTOV(np); 2688 dpossav2 = info.dpos; 2689 info.dpos = dpossav1; 2690 mdsav2 = info.md; 2691 info.md = mdsav1; 2692 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 2693 info.dpos = dpossav2; 2694 info.md = mdsav2; 2695 dp->nfs_type = 2696 IFTODT(VTTOIF(np->n_vattr.va_type)); 2697 nfs_cache_setvp(&nch, newvp, 2698 nfspos_cache_timeout); 2699 vput(newvp); 2700 newvp = NULLVP; 2701 cache_put(&nch); 2702 } else { 2703 rdfail: 2704 ; 2705 #if 0 2706 kprintf("Warning: NFS/rddirplus, " 2707 "UNABLE TO ENTER %*.*s\n", 2708 nlc.nlc_namelen, nlc.nlc_namelen, 2709 nlc.nlc_nameptr); 2710 #endif 2711 } 2712 } 2713 } else { 2714 /* Just skip over the file handle */ 2715 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2716 i = fxdr_unsigned(int, *tl); 2717 ERROROUT(nfsm_adv(&info, nfsm_rndup(i))); 2718 } 2719 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2720 more_dirs = fxdr_unsigned(int, *tl); 2721 } 2722 /* 2723 * If at end of rpc data, get the eof boolean 2724 */ 2725 if (!more_dirs) { 2726 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2727 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2728 } 2729 m_freem(info.mrep); 2730 info.mrep = NULL; 2731 } 2732 /* 2733 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2734 * by increasing d_reclen for the last record. 2735 */ 2736 if (blksiz > 0) { 2737 left = DIRBLKSIZ - blksiz; 2738 dp->nfs_reclen += left; 2739 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2740 uiop->uio_iov->iov_len -= left; 2741 uiop->uio_offset += left; 2742 uiop->uio_resid -= left; 2743 } 2744 2745 /* 2746 * We are now either at the end of the directory or have filled the 2747 * block. 2748 */ 2749 if (bigenough) { 2750 dnp->n_direofoffset = uiop->uio_offset; 2751 } else { 2752 if (uiop->uio_resid > 0) 2753 kprintf("EEK! readdirplusrpc resid > 0\n"); 2754 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2755 *cookiep = cookie; 2756 } 2757 nfsmout: 2758 if (newvp != NULLVP) { 2759 if (newvp == vp) 2760 vrele(newvp); 2761 else 2762 vput(newvp); 2763 newvp = NULLVP; 2764 } 2765 if (dnch.ncp) 2766 cache_drop(&dnch); 2767 return (error); 2768 } 2769 2770 /* 2771 * Silly rename. To make the NFS filesystem that is stateless look a little 2772 * more like the "ufs" a remove of an active vnode is translated to a rename 2773 * to a funny looking filename that is removed by nfs_inactive on the 2774 * nfsnode. There is the potential for another process on a different client 2775 * to create the same funny name between the nfs_lookitup() fails and the 2776 * nfs_rename() completes, but... 2777 */ 2778 static int 2779 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2780 { 2781 struct sillyrename *sp; 2782 struct nfsnode *np; 2783 int error; 2784 2785 /* 2786 * We previously purged dvp instead of vp. I don't know why, it 2787 * completely destroys performance. We can't do it anyway with the 2788 * new VFS API since we would be breaking the namecache topology. 2789 */ 2790 cache_purge(vp); /* XXX */ 2791 np = VTONFS(vp); 2792 #ifndef DIAGNOSTIC 2793 if (vp->v_type == VDIR) 2794 panic("nfs: sillyrename dir"); 2795 #endif 2796 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2797 M_NFSREQ, M_WAITOK); 2798 sp->s_cred = crdup(cnp->cn_cred); 2799 sp->s_dvp = dvp; 2800 vref(dvp); 2801 2802 /* Fudge together a funny name */ 2803 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", 2804 (int)(intptr_t)cnp->cn_td); 2805 2806 /* Try lookitups until we get one that isn't there */ 2807 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2808 cnp->cn_td, NULL) == 0) { 2809 sp->s_name[4]++; 2810 if (sp->s_name[4] > 'z') { 2811 error = EINVAL; 2812 goto bad; 2813 } 2814 } 2815 error = nfs_renameit(dvp, cnp, sp); 2816 if (error) 2817 goto bad; 2818 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2819 cnp->cn_td, &np); 2820 np->n_sillyrename = sp; 2821 return (0); 2822 bad: 2823 vrele(sp->s_dvp); 2824 crfree(sp->s_cred); 2825 kfree((caddr_t)sp, M_NFSREQ); 2826 return (error); 2827 } 2828 2829 /* 2830 * Look up a file name and optionally either update the file handle or 2831 * allocate an nfsnode, depending on the value of npp. 2832 * npp == NULL --> just do the lookup 2833 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2834 * handled too 2835 * *npp != NULL --> update the file handle in the vnode 2836 */ 2837 static int 2838 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2839 struct thread *td, struct nfsnode **npp) 2840 { 2841 struct vnode *newvp = NULL; 2842 struct nfsnode *np, *dnp = VTONFS(dvp); 2843 int error = 0, fhlen, attrflag; 2844 nfsfh_t *nfhp; 2845 struct nfsm_info info; 2846 2847 info.mrep = NULL; 2848 info.v3 = NFS_ISV3(dvp); 2849 2850 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2851 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 2852 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2853 ERROROUT(nfsm_fhtom(&info, dvp)); 2854 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN)); 2855 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error)); 2856 if (npp && !error) { 2857 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp)); 2858 if (*npp) { 2859 np = *npp; 2860 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2861 kfree((caddr_t)np->n_fhp, M_NFSBIGFH); 2862 np->n_fhp = &np->n_fh; 2863 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2864 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK); 2865 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2866 np->n_fhsize = fhlen; 2867 newvp = NFSTOV(np); 2868 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2869 vref(dvp); 2870 newvp = dvp; 2871 } else { 2872 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2873 if (error) { 2874 m_freem(info.mrep); 2875 info.mrep = NULL; 2876 return (error); 2877 } 2878 newvp = NFSTOV(np); 2879 } 2880 if (info.v3) { 2881 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 2882 NFS_LATTR_NOSHRINK)); 2883 if (!attrflag && *npp == NULL) { 2884 m_freem(info.mrep); 2885 info.mrep = NULL; 2886 if (newvp == dvp) 2887 vrele(newvp); 2888 else 2889 vput(newvp); 2890 return (ENOENT); 2891 } 2892 } else { 2893 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL)); 2894 } 2895 } 2896 m_freem(info.mrep); 2897 info.mrep = NULL; 2898 nfsmout: 2899 if (npp && *npp == NULL) { 2900 if (error) { 2901 if (newvp) { 2902 if (newvp == dvp) 2903 vrele(newvp); 2904 else 2905 vput(newvp); 2906 } 2907 } else 2908 *npp = np; 2909 } 2910 return (error); 2911 } 2912 2913 /* 2914 * Nfs Version 3 commit rpc 2915 * 2916 * We call it 'uio' to distinguish it from 'bio' but there is no real uio 2917 * involved. 2918 */ 2919 int 2920 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2921 { 2922 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2923 int error = 0, wccflag = NFSV3_WCCRATTR; 2924 struct nfsm_info info; 2925 u_int32_t *tl; 2926 2927 info.mrep = NULL; 2928 info.v3 = 1; 2929 2930 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2931 return (0); 2932 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2933 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1)); 2934 ERROROUT(nfsm_fhtom(&info, vp)); 2935 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED); 2936 txdr_hyper(offset, tl); 2937 tl += 2; 2938 *tl = txdr_unsigned(cnt); 2939 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td, 2940 nfs_vpcred(vp, ND_WRITE), &error)); 2941 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 2942 if (!error) { 2943 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF)); 2944 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2945 NFSX_V3WRITEVERF)) { 2946 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2947 NFSX_V3WRITEVERF); 2948 error = NFSERR_STALEWRITEVERF; 2949 } 2950 } 2951 m_freem(info.mrep); 2952 info.mrep = NULL; 2953 nfsmout: 2954 return (error); 2955 } 2956 2957 /* 2958 * Kludge City.. 2959 * - make nfs_bmap() essentially a no-op that does no translation 2960 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2961 * (Maybe I could use the process's page mapping, but I was concerned that 2962 * Kernel Write might not be enabled and also figured copyout() would do 2963 * a lot more work than bcopy() and also it currently happens in the 2964 * context of the swapper process (2). 2965 * 2966 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, 2967 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2968 */ 2969 static int 2970 nfs_bmap(struct vop_bmap_args *ap) 2971 { 2972 if (ap->a_doffsetp != NULL) 2973 *ap->a_doffsetp = ap->a_loffset; 2974 if (ap->a_runp != NULL) 2975 *ap->a_runp = 0; 2976 if (ap->a_runb != NULL) 2977 *ap->a_runb = 0; 2978 return (0); 2979 } 2980 2981 /* 2982 * Strategy routine. 2983 */ 2984 static int 2985 nfs_strategy(struct vop_strategy_args *ap) 2986 { 2987 struct bio *bio = ap->a_bio; 2988 struct bio *nbio; 2989 struct buf *bp __debugvar = bio->bio_buf; 2990 struct thread *td; 2991 int error; 2992 2993 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2994 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2995 KASSERT(BUF_REFCNT(bp) > 0, 2996 ("nfs_strategy: buffer %p not locked", bp)); 2997 2998 if (bio->bio_flags & BIO_SYNC) 2999 td = curthread; /* XXX */ 3000 else 3001 td = NULL; 3002 3003 /* 3004 * We probably don't need to push an nbio any more since no 3005 * block conversion is required due to the use of 64 bit byte 3006 * offsets, but do it anyway. 3007 * 3008 * NOTE: When NFS callers itself via this strategy routines and 3009 * sets up a synchronous I/O, it expects the I/O to run 3010 * synchronously (its bio_done routine just assumes it), 3011 * so for now we have to honor the bit. 3012 */ 3013 nbio = push_bio(bio); 3014 nbio->bio_offset = bio->bio_offset; 3015 nbio->bio_flags = bio->bio_flags & BIO_SYNC; 3016 3017 /* 3018 * If the op is asynchronous and an i/o daemon is waiting 3019 * queue the request, wake it up and wait for completion 3020 * otherwise just do it ourselves. 3021 */ 3022 if (bio->bio_flags & BIO_SYNC) { 3023 error = nfs_doio(ap->a_vp, nbio, td); 3024 } else { 3025 nfs_asyncio(ap->a_vp, nbio); 3026 error = 0; 3027 } 3028 return (error); 3029 } 3030 3031 /* 3032 * Mmap a file 3033 * 3034 * NB Currently unsupported. 3035 * 3036 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 3037 */ 3038 /* ARGSUSED */ 3039 static int 3040 nfs_mmap(struct vop_mmap_args *ap) 3041 { 3042 return (EINVAL); 3043 } 3044 3045 /* 3046 * fsync vnode op. Just call nfs_flush() with commit == 1. 3047 * 3048 * nfs_fsync(struct vnode *a_vp, int a_waitfor) 3049 */ 3050 /* ARGSUSED */ 3051 static int 3052 nfs_fsync(struct vop_fsync_args *ap) 3053 { 3054 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1)); 3055 } 3056 3057 /* 3058 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 3059 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 3060 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 3061 * set the buffer contains data that has already been written to the server 3062 * and which now needs a commit RPC. 3063 * 3064 * If commit is 0 we only take one pass and only flush buffers containing new 3065 * dirty data. 3066 * 3067 * If commit is 1 we take two passes, issuing a commit RPC in the second 3068 * pass. 3069 * 3070 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 3071 * to completely flush all pending data. 3072 * 3073 * Note that the RB_SCAN code properly handles the case where the 3074 * callback might block and directly or indirectly (another thread) cause 3075 * the RB tree to change. 3076 */ 3077 3078 #ifndef NFS_COMMITBVECSIZ 3079 #define NFS_COMMITBVECSIZ 16 3080 #endif 3081 3082 struct nfs_flush_info { 3083 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 3084 struct thread *td; 3085 struct vnode *vp; 3086 int waitfor; 3087 int slpflag; 3088 int slptimeo; 3089 int loops; 3090 struct buf *bvary[NFS_COMMITBVECSIZ]; 3091 int bvsize; 3092 off_t beg_off; 3093 off_t end_off; 3094 }; 3095 3096 static int nfs_flush_bp(struct buf *bp, void *data); 3097 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 3098 3099 int 3100 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 3101 { 3102 struct nfsnode *np = VTONFS(vp); 3103 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3104 struct nfs_flush_info info; 3105 lwkt_tokref vlock; 3106 int error; 3107 3108 bzero(&info, sizeof(info)); 3109 info.td = td; 3110 info.vp = vp; 3111 info.waitfor = waitfor; 3112 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 3113 info.loops = 0; 3114 lwkt_gettoken(&vlock, &vp->v_token); 3115 3116 do { 3117 /* 3118 * Flush mode 3119 */ 3120 info.mode = NFI_FLUSHNEW; 3121 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3122 nfs_flush_bp, &info); 3123 3124 /* 3125 * Take a second pass if committing and no error occured. 3126 * Clean up any left over collection (whether an error 3127 * occurs or not). 3128 */ 3129 if (commit && error == 0) { 3130 info.mode = NFI_COMMIT; 3131 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3132 nfs_flush_bp, &info); 3133 if (info.bvsize) 3134 error = nfs_flush_docommit(&info, error); 3135 } 3136 3137 /* 3138 * Wait for pending I/O to complete before checking whether 3139 * any further dirty buffers exist. 3140 */ 3141 while (waitfor == MNT_WAIT && 3142 bio_track_active(&vp->v_track_write)) { 3143 error = bio_track_wait(&vp->v_track_write, 3144 info.slpflag, info.slptimeo); 3145 if (error) { 3146 /* 3147 * We have to be able to break out if this 3148 * is an 'intr' mount. 3149 */ 3150 if (nfs_sigintr(nmp, NULL, td)) { 3151 error = -EINTR; 3152 break; 3153 } 3154 3155 /* 3156 * Since we do not process pending signals, 3157 * once we get a PCATCH our tsleep() will no 3158 * longer sleep, switch to a fixed timeout 3159 * instead. 3160 */ 3161 if (info.slpflag == PCATCH) { 3162 info.slpflag = 0; 3163 info.slptimeo = 2 * hz; 3164 } 3165 error = 0; 3166 } 3167 } 3168 ++info.loops; 3169 /* 3170 * Loop if we are flushing synchronous as well as committing, 3171 * and dirty buffers are still present. Otherwise we might livelock. 3172 */ 3173 } while (waitfor == MNT_WAIT && commit && 3174 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3175 3176 /* 3177 * The callbacks have to return a negative error to terminate the 3178 * RB scan. 3179 */ 3180 if (error < 0) 3181 error = -error; 3182 3183 /* 3184 * Deal with any error collection 3185 */ 3186 if (np->n_flag & NWRITEERR) { 3187 error = np->n_error; 3188 np->n_flag &= ~NWRITEERR; 3189 } 3190 lwkt_reltoken(&vlock); 3191 return (error); 3192 } 3193 3194 static 3195 int 3196 nfs_flush_bp(struct buf *bp, void *data) 3197 { 3198 struct nfs_flush_info *info = data; 3199 int lkflags; 3200 int error; 3201 off_t toff; 3202 3203 error = 0; 3204 switch(info->mode) { 3205 case NFI_FLUSHNEW: 3206 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3207 if (error && info->loops && info->waitfor == MNT_WAIT) { 3208 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3209 if (error) { 3210 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3211 if (info->slpflag & PCATCH) 3212 lkflags |= LK_PCATCH; 3213 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3214 info->slptimeo); 3215 } 3216 } 3217 3218 /* 3219 * Ignore locking errors 3220 */ 3221 if (error) { 3222 error = 0; 3223 break; 3224 } 3225 3226 /* 3227 * The buffer may have changed out from under us, even if 3228 * we did not block (MPSAFE). Check again now that it is 3229 * locked. 3230 */ 3231 if (bp->b_vp == info->vp && 3232 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) { 3233 bremfree(bp); 3234 bawrite(bp); 3235 } else { 3236 BUF_UNLOCK(bp); 3237 } 3238 break; 3239 case NFI_COMMIT: 3240 /* 3241 * Only process buffers in need of a commit which we can 3242 * immediately lock. This may prevent a buffer from being 3243 * committed, but the normal flush loop will block on the 3244 * same buffer so we shouldn't get into an endless loop. 3245 */ 3246 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3247 (B_DELWRI | B_NEEDCOMMIT)) { 3248 break; 3249 } 3250 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 3251 break; 3252 3253 /* 3254 * We must recheck after successfully locking the buffer. 3255 */ 3256 if (bp->b_vp != info->vp || 3257 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3258 (B_DELWRI | B_NEEDCOMMIT)) { 3259 BUF_UNLOCK(bp); 3260 break; 3261 } 3262 3263 /* 3264 * NOTE: storing the bp in the bvary[] basically sets 3265 * it up for a commit operation. 3266 * 3267 * We must call vfs_busy_pages() now so the commit operation 3268 * is interlocked with user modifications to memory mapped 3269 * pages. The b_dirtyoff/b_dirtyend range is not correct 3270 * until after the pages have been busied. 3271 * 3272 * Note: to avoid loopback deadlocks, we do not 3273 * assign b_runningbufspace. 3274 */ 3275 bremfree(bp); 3276 bp->b_cmd = BUF_CMD_WRITE; 3277 vfs_busy_pages(bp->b_vp, bp); 3278 info->bvary[info->bvsize] = bp; 3279 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3280 if (info->bvsize == 0 || toff < info->beg_off) 3281 info->beg_off = toff; 3282 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3283 if (info->bvsize == 0 || toff > info->end_off) 3284 info->end_off = toff; 3285 ++info->bvsize; 3286 if (info->bvsize == NFS_COMMITBVECSIZ) { 3287 error = nfs_flush_docommit(info, 0); 3288 KKASSERT(info->bvsize == 0); 3289 } 3290 } 3291 return (error); 3292 } 3293 3294 static 3295 int 3296 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3297 { 3298 struct vnode *vp; 3299 struct buf *bp; 3300 off_t bytes; 3301 int retv; 3302 int i; 3303 3304 vp = info->vp; 3305 3306 if (info->bvsize > 0) { 3307 /* 3308 * Commit data on the server, as required. Note that 3309 * nfs_commit will use the vnode's cred for the commit. 3310 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3311 */ 3312 bytes = info->end_off - info->beg_off; 3313 if (bytes > 0x40000000) 3314 bytes = 0x40000000; 3315 if (error) { 3316 retv = -error; 3317 } else { 3318 retv = nfs_commitrpc_uio(vp, info->beg_off, 3319 (int)bytes, info->td); 3320 if (retv == NFSERR_STALEWRITEVERF) 3321 nfs_clearcommit(vp->v_mount); 3322 } 3323 3324 /* 3325 * Now, either mark the blocks I/O done or mark the 3326 * blocks dirty, depending on whether the commit 3327 * succeeded. 3328 */ 3329 for (i = 0; i < info->bvsize; ++i) { 3330 bp = info->bvary[i]; 3331 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) { 3332 /* 3333 * Either an error or the original 3334 * vfs_busy_pages() cleared B_NEEDCOMMIT 3335 * due to finding new dirty VM pages in 3336 * the buffer. 3337 * 3338 * Leave B_DELWRI intact. 3339 */ 3340 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3341 vfs_unbusy_pages(bp); 3342 bp->b_cmd = BUF_CMD_DONE; 3343 bqrelse(bp); 3344 } else { 3345 /* 3346 * Success, remove B_DELWRI ( bundirty() ). 3347 * 3348 * b_dirtyoff/b_dirtyend seem to be NFS 3349 * specific. We should probably move that 3350 * into bundirty(). XXX 3351 * 3352 * We are faking an I/O write, we have to 3353 * start the transaction in order to 3354 * immediately biodone() it. 3355 */ 3356 bundirty(bp); 3357 bp->b_flags &= ~B_ERROR; 3358 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3359 bp->b_dirtyoff = bp->b_dirtyend = 0; 3360 biodone(&bp->b_bio1); 3361 } 3362 } 3363 info->bvsize = 0; 3364 } 3365 return (error); 3366 } 3367 3368 /* 3369 * NFS advisory byte-level locks. 3370 * Currently unsupported. 3371 * 3372 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3373 * int a_flags) 3374 */ 3375 static int 3376 nfs_advlock(struct vop_advlock_args *ap) 3377 { 3378 struct nfsnode *np = VTONFS(ap->a_vp); 3379 3380 /* 3381 * The following kludge is to allow diskless support to work 3382 * until a real NFS lockd is implemented. Basically, just pretend 3383 * that this is a local lock. 3384 */ 3385 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3386 } 3387 3388 /* 3389 * Print out the contents of an nfsnode. 3390 * 3391 * nfs_print(struct vnode *a_vp) 3392 */ 3393 static int 3394 nfs_print(struct vop_print_args *ap) 3395 { 3396 struct vnode *vp = ap->a_vp; 3397 struct nfsnode *np = VTONFS(vp); 3398 3399 kprintf("tag VT_NFS, fileid %lld fsid 0x%x", 3400 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3401 if (vp->v_type == VFIFO) 3402 fifo_printinfo(vp); 3403 kprintf("\n"); 3404 return (0); 3405 } 3406 3407 /* 3408 * nfs special file access vnode op. 3409 * 3410 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 3411 */ 3412 static int 3413 nfs_laccess(struct vop_access_args *ap) 3414 { 3415 struct vattr vattr; 3416 int error; 3417 3418 error = VOP_GETATTR(ap->a_vp, &vattr); 3419 if (!error) 3420 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid, 3421 vattr.va_mode, 0); 3422 return (error); 3423 } 3424 3425 /* 3426 * Read wrapper for fifos. 3427 * 3428 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3429 * struct ucred *a_cred) 3430 */ 3431 static int 3432 nfsfifo_read(struct vop_read_args *ap) 3433 { 3434 struct nfsnode *np = VTONFS(ap->a_vp); 3435 3436 /* 3437 * Set access flag. 3438 */ 3439 np->n_flag |= NACC; 3440 getnanotime(&np->n_atim); 3441 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3442 } 3443 3444 /* 3445 * Write wrapper for fifos. 3446 * 3447 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3448 * struct ucred *a_cred) 3449 */ 3450 static int 3451 nfsfifo_write(struct vop_write_args *ap) 3452 { 3453 struct nfsnode *np = VTONFS(ap->a_vp); 3454 3455 /* 3456 * Set update flag. 3457 */ 3458 np->n_flag |= NUPD; 3459 getnanotime(&np->n_mtim); 3460 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3461 } 3462 3463 /* 3464 * Close wrapper for fifos. 3465 * 3466 * Update the times on the nfsnode then do fifo close. 3467 * 3468 * nfsfifo_close(struct vnode *a_vp, int a_fflag) 3469 */ 3470 static int 3471 nfsfifo_close(struct vop_close_args *ap) 3472 { 3473 struct vnode *vp = ap->a_vp; 3474 struct nfsnode *np = VTONFS(vp); 3475 struct vattr vattr; 3476 struct timespec ts; 3477 3478 if (np->n_flag & (NACC | NUPD)) { 3479 getnanotime(&ts); 3480 if (np->n_flag & NACC) 3481 np->n_atim = ts; 3482 if (np->n_flag & NUPD) 3483 np->n_mtim = ts; 3484 np->n_flag |= NCHG; 3485 if (vp->v_sysref.refcnt == 1 && 3486 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3487 VATTR_NULL(&vattr); 3488 if (np->n_flag & NACC) 3489 vattr.va_atime = np->n_atim; 3490 if (np->n_flag & NUPD) 3491 vattr.va_mtime = np->n_mtim; 3492 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3493 } 3494 } 3495 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3496 } 3497 3498