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 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 101 static int nfs_lookup (struct vop_old_lookup_args *); 102 static int nfs_create (struct vop_old_create_args *); 103 static int nfs_mknod (struct vop_old_mknod_args *); 104 static int nfs_open (struct vop_open_args *); 105 static int nfs_close (struct vop_close_args *); 106 static int nfs_access (struct vop_access_args *); 107 static int nfs_getattr (struct vop_getattr_args *); 108 static int nfs_setattr (struct vop_setattr_args *); 109 static int nfs_read (struct vop_read_args *); 110 static int nfs_mmap (struct vop_mmap_args *); 111 static int nfs_fsync (struct vop_fsync_args *); 112 static int nfs_remove (struct vop_old_remove_args *); 113 static int nfs_link (struct vop_old_link_args *); 114 static int nfs_rename (struct vop_old_rename_args *); 115 static int nfs_mkdir (struct vop_old_mkdir_args *); 116 static int nfs_rmdir (struct vop_old_rmdir_args *); 117 static int nfs_symlink (struct vop_old_symlink_args *); 118 static int nfs_readdir (struct vop_readdir_args *); 119 static int nfs_bmap (struct vop_bmap_args *); 120 static int nfs_strategy (struct vop_strategy_args *); 121 static int nfs_lookitup (struct vnode *, const char *, int, 122 struct ucred *, struct thread *, struct nfsnode **); 123 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 124 static int nfs_laccess (struct vop_access_args *); 125 static int nfs_readlink (struct vop_readlink_args *); 126 static int nfs_print (struct vop_print_args *); 127 static int nfs_advlock (struct vop_advlock_args *); 128 129 static int nfs_nresolve (struct vop_nresolve_args *); 130 /* 131 * Global vfs data structures for nfs 132 */ 133 struct vop_ops nfsv2_vnode_vops = { 134 .vop_default = vop_defaultop, 135 .vop_access = nfs_access, 136 .vop_advlock = nfs_advlock, 137 .vop_bmap = nfs_bmap, 138 .vop_close = nfs_close, 139 .vop_old_create = nfs_create, 140 .vop_fsync = nfs_fsync, 141 .vop_getattr = nfs_getattr, 142 .vop_getpages = vop_stdgetpages, 143 .vop_putpages = vop_stdputpages, 144 .vop_inactive = nfs_inactive, 145 .vop_old_link = nfs_link, 146 .vop_old_lookup = nfs_lookup, 147 .vop_old_mkdir = nfs_mkdir, 148 .vop_old_mknod = nfs_mknod, 149 .vop_mmap = nfs_mmap, 150 .vop_open = nfs_open, 151 .vop_print = nfs_print, 152 .vop_read = nfs_read, 153 .vop_readdir = nfs_readdir, 154 .vop_readlink = nfs_readlink, 155 .vop_reclaim = nfs_reclaim, 156 .vop_old_remove = nfs_remove, 157 .vop_old_rename = nfs_rename, 158 .vop_old_rmdir = nfs_rmdir, 159 .vop_setattr = nfs_setattr, 160 .vop_strategy = nfs_strategy, 161 .vop_old_symlink = nfs_symlink, 162 .vop_write = nfs_write, 163 .vop_nresolve = nfs_nresolve 164 }; 165 166 /* 167 * Special device vnode ops 168 */ 169 struct vop_ops nfsv2_spec_vops = { 170 .vop_default = vop_defaultop, 171 .vop_access = nfs_laccess, 172 .vop_close = nfs_close, 173 .vop_fsync = nfs_fsync, 174 .vop_getattr = nfs_getattr, 175 .vop_inactive = nfs_inactive, 176 .vop_print = nfs_print, 177 .vop_read = vop_stdnoread, 178 .vop_reclaim = nfs_reclaim, 179 .vop_setattr = nfs_setattr, 180 .vop_write = vop_stdnowrite 181 }; 182 183 struct vop_ops nfsv2_fifo_vops = { 184 .vop_default = fifo_vnoperate, 185 .vop_access = nfs_laccess, 186 .vop_close = nfsfifo_close, 187 .vop_fsync = nfs_fsync, 188 .vop_getattr = nfs_getattr, 189 .vop_inactive = nfs_inactive, 190 .vop_print = nfs_print, 191 .vop_read = nfsfifo_read, 192 .vop_reclaim = nfs_reclaim, 193 .vop_setattr = nfs_setattr, 194 .vop_write = nfsfifo_write 195 }; 196 197 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 198 struct componentname *cnp, 199 struct vattr *vap); 200 static int nfs_removerpc (struct vnode *dvp, const char *name, 201 int namelen, 202 struct ucred *cred, struct thread *td); 203 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 204 int fnamelen, struct vnode *tdvp, 205 const char *tnameptr, int tnamelen, 206 struct ucred *cred, struct thread *td); 207 static int nfs_renameit (struct vnode *sdvp, 208 struct componentname *scnp, 209 struct sillyrename *sp); 210 211 SYSCTL_DECL(_vfs_nfs); 212 213 static int nfs_flush_on_rename = 1; 214 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW, 215 &nfs_flush_on_rename, 0, "flush fvp prior to rename"); 216 static int nfs_flush_on_hlink = 0; 217 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW, 218 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link"); 219 220 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 221 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 222 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 223 224 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 225 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 226 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout"); 227 228 static int nfspos_cache_timeout = NFS_MINATTRTIMO; 229 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW, 230 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout"); 231 232 static int nfsv3_commit_on_close = 0; 233 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 234 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 235 #if 0 236 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 237 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 238 239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 240 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 241 #endif 242 243 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 244 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 245 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 246 247 /* 248 * Returns whether a name component is a degenerate '.' or '..'. 249 */ 250 static __inline 251 int 252 nlcdegenerate(struct nlcomponent *nlc) 253 { 254 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.') 255 return(1); 256 if (nlc->nlc_namelen == 2 && 257 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.') 258 return(1); 259 return(0); 260 } 261 262 static int 263 nfs3_access_otw(struct vnode *vp, int wmode, 264 struct thread *td, struct ucred *cred) 265 { 266 struct nfsnode *np = VTONFS(vp); 267 int attrflag; 268 int error = 0; 269 u_int32_t *tl; 270 u_int32_t rmode; 271 struct nfsm_info info; 272 273 info.mrep = NULL; 274 info.v3 = 1; 275 276 nfsstats.rpccnt[NFSPROC_ACCESS]++; 277 nfsm_reqhead(&info, vp, NFSPROC_ACCESS, 278 NFSX_FH(info.v3) + NFSX_UNSIGNED); 279 ERROROUT(nfsm_fhtom(&info, vp)); 280 tl = nfsm_build(&info, NFSX_UNSIGNED); 281 *tl = txdr_unsigned(wmode); 282 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error)); 283 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK)); 284 if (error == 0) { 285 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 286 rmode = fxdr_unsigned(u_int32_t, *tl); 287 np->n_mode = rmode; 288 np->n_modeuid = cred->cr_uid; 289 np->n_modestamp = mycpu->gd_time_seconds; 290 } 291 m_freem(info.mrep); 292 info.mrep = NULL; 293 nfsmout: 294 return error; 295 } 296 297 /* 298 * nfs access vnode op. 299 * For nfs version 2, just return ok. File accesses may fail later. 300 * For nfs version 3, use the access rpc to check accessibility. If file modes 301 * are changed on the server, accesses might still fail later. 302 * 303 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 304 */ 305 static int 306 nfs_access(struct vop_access_args *ap) 307 { 308 struct ucred *cred; 309 struct vnode *vp = ap->a_vp; 310 thread_t td = curthread; 311 int error = 0; 312 u_int32_t mode, wmode; 313 struct nfsnode *np = VTONFS(vp); 314 int v3 = NFS_ISV3(vp); 315 316 /* 317 * Disallow write attempts on filesystems mounted read-only; 318 * unless the file is a socket, fifo, or a block or character 319 * device resident on the filesystem. 320 */ 321 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 322 switch (vp->v_type) { 323 case VREG: 324 case VDIR: 325 case VLNK: 326 return (EROFS); 327 default: 328 break; 329 } 330 } 331 332 /* 333 * The NFS protocol passes only the effective uid/gid over the wire but 334 * we need to check access against real ids if AT_EACCESS not set. 335 * Handle this case by cloning the credentials and setting the 336 * effective ids to the real ones. 337 */ 338 if (ap->a_flags & AT_EACCESS) { 339 cred = crhold(ap->a_cred); 340 } else { 341 cred = crdup(ap->a_cred); 342 cred->cr_uid = cred->cr_ruid; 343 cred->cr_gid = cred->cr_rgid; 344 } 345 346 /* 347 * For nfs v3, check to see if we have done this recently, and if 348 * so return our cached result instead of making an ACCESS call. 349 * If not, do an access rpc, otherwise you are stuck emulating 350 * ufs_access() locally using the vattr. This may not be correct, 351 * since the server may apply other access criteria such as 352 * client uid-->server uid mapping that we do not know about. 353 */ 354 if (v3) { 355 if (ap->a_mode & VREAD) 356 mode = NFSV3ACCESS_READ; 357 else 358 mode = 0; 359 if (vp->v_type != VDIR) { 360 if (ap->a_mode & VWRITE) 361 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 362 if (ap->a_mode & VEXEC) 363 mode |= NFSV3ACCESS_EXECUTE; 364 } else { 365 if (ap->a_mode & VWRITE) 366 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 367 NFSV3ACCESS_DELETE); 368 if (ap->a_mode & VEXEC) 369 mode |= NFSV3ACCESS_LOOKUP; 370 } 371 /* XXX safety belt, only make blanket request if caching */ 372 if (nfsaccess_cache_timeout > 0) { 373 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 374 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 375 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 376 } else { 377 wmode = mode; 378 } 379 380 /* 381 * Does our cached result allow us to give a definite yes to 382 * this request? 383 */ 384 if (np->n_modestamp && 385 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 386 (cred->cr_uid == np->n_modeuid) && 387 ((np->n_mode & mode) == mode)) { 388 nfsstats.accesscache_hits++; 389 } else { 390 /* 391 * Either a no, or a don't know. Go to the wire. 392 */ 393 nfsstats.accesscache_misses++; 394 error = nfs3_access_otw(vp, wmode, td, cred); 395 if (!error) { 396 if ((np->n_mode & mode) != mode) { 397 error = EACCES; 398 } 399 } 400 } 401 } else { 402 if ((error = nfs_laccess(ap)) != 0) { 403 crfree(cred); 404 return (error); 405 } 406 407 /* 408 * Attempt to prevent a mapped root from accessing a file 409 * which it shouldn't. We try to read a byte from the file 410 * if the user is root and the file is not zero length. 411 * After calling nfs_laccess, we should have the correct 412 * file size cached. 413 */ 414 if (cred->cr_uid == 0 && (ap->a_mode & VREAD) 415 && VTONFS(vp)->n_size > 0) { 416 struct iovec aiov; 417 struct uio auio; 418 char buf[1]; 419 420 aiov.iov_base = buf; 421 aiov.iov_len = 1; 422 auio.uio_iov = &aiov; 423 auio.uio_iovcnt = 1; 424 auio.uio_offset = 0; 425 auio.uio_resid = 1; 426 auio.uio_segflg = UIO_SYSSPACE; 427 auio.uio_rw = UIO_READ; 428 auio.uio_td = td; 429 430 if (vp->v_type == VREG) { 431 error = nfs_readrpc_uio(vp, &auio); 432 } else if (vp->v_type == VDIR) { 433 char* bp; 434 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 435 aiov.iov_base = bp; 436 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 437 error = nfs_readdirrpc_uio(vp, &auio); 438 kfree(bp, M_TEMP); 439 } else if (vp->v_type == VLNK) { 440 error = nfs_readlinkrpc_uio(vp, &auio); 441 } else { 442 error = EACCES; 443 } 444 } 445 } 446 /* 447 * [re]record creds for reading and/or writing if access 448 * was granted. Assume the NFS server will grant read access 449 * for execute requests. 450 */ 451 if (error == 0) { 452 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) { 453 crhold(cred); 454 if (np->n_rucred) 455 crfree(np->n_rucred); 456 np->n_rucred = cred; 457 } 458 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) { 459 crhold(cred); 460 if (np->n_wucred) 461 crfree(np->n_wucred); 462 np->n_wucred = cred; 463 } 464 } 465 crfree(cred); 466 return(error); 467 } 468 469 /* 470 * nfs open vnode op 471 * Check to see if the type is ok 472 * and that deletion is not in progress. 473 * For paged in text files, you will need to flush the page cache 474 * if consistency is lost. 475 * 476 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 477 * struct file *a_fp) 478 */ 479 /* ARGSUSED */ 480 static int 481 nfs_open(struct vop_open_args *ap) 482 { 483 struct vnode *vp = ap->a_vp; 484 struct nfsnode *np = VTONFS(vp); 485 struct vattr vattr; 486 int error; 487 488 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 489 #ifdef DIAGNOSTIC 490 kprintf("open eacces vtyp=%d\n",vp->v_type); 491 #endif 492 return (EOPNOTSUPP); 493 } 494 495 /* 496 * Save valid creds for reading and writing for later RPCs. 497 */ 498 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) { 499 crhold(ap->a_cred); 500 if (np->n_rucred) 501 crfree(np->n_rucred); 502 np->n_rucred = ap->a_cred; 503 } 504 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) { 505 crhold(ap->a_cred); 506 if (np->n_wucred) 507 crfree(np->n_wucred); 508 np->n_wucred = ap->a_cred; 509 } 510 511 /* 512 * Clear the attribute cache only if opening with write access. It 513 * is unclear if we should do this at all here, but we certainly 514 * should not clear the cache unconditionally simply because a file 515 * is being opened. 516 */ 517 if (ap->a_mode & FWRITE) 518 np->n_attrstamp = 0; 519 520 /* 521 * For normal NFS, reconcile changes made locally verses 522 * changes made remotely. Note that VOP_GETATTR only goes 523 * to the wire if the cached attribute has timed out or been 524 * cleared. 525 * 526 * If local modifications have been made clear the attribute 527 * cache to force an attribute and modified time check. If 528 * GETATTR detects that the file has been changed by someone 529 * other then us it will set NRMODIFIED. 530 * 531 * If we are opening a directory and local changes have been 532 * made we have to invalidate the cache in order to ensure 533 * that we get the most up-to-date information from the 534 * server. XXX 535 */ 536 if (np->n_flag & NLMODIFIED) { 537 np->n_attrstamp = 0; 538 if (vp->v_type == VDIR) { 539 error = nfs_vinvalbuf(vp, V_SAVE, 1); 540 if (error == EINTR) 541 return (error); 542 nfs_invaldir(vp); 543 } 544 } 545 error = VOP_GETATTR(vp, &vattr); 546 if (error) 547 return (error); 548 if (np->n_flag & NRMODIFIED) { 549 if (vp->v_type == VDIR) 550 nfs_invaldir(vp); 551 error = nfs_vinvalbuf(vp, V_SAVE, 1); 552 if (error == EINTR) 553 return (error); 554 np->n_flag &= ~NRMODIFIED; 555 } 556 557 return (vop_stdopen(ap)); 558 } 559 560 /* 561 * nfs close vnode op 562 * What an NFS client should do upon close after writing is a debatable issue. 563 * Most NFS clients push delayed writes to the server upon close, basically for 564 * two reasons: 565 * 1 - So that any write errors may be reported back to the client process 566 * doing the close system call. By far the two most likely errors are 567 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 568 * 2 - To put a worst case upper bound on cache inconsistency between 569 * multiple clients for the file. 570 * There is also a consistency problem for Version 2 of the protocol w.r.t. 571 * not being able to tell if other clients are writing a file concurrently, 572 * since there is no way of knowing if the changed modify time in the reply 573 * is only due to the write for this client. 574 * (NFS Version 3 provides weak cache consistency data in the reply that 575 * should be sufficient to detect and handle this case.) 576 * 577 * The current code does the following: 578 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 579 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 580 * or commit them (this satisfies 1 and 2 except for the 581 * case where the server crashes after this close but 582 * before the commit RPC, which is felt to be "good 583 * enough". Changing the last argument to nfs_flush() to 584 * a 1 would force a commit operation, if it is felt a 585 * commit is necessary now. 586 * for NQNFS - do nothing now, since 2 is dealt with via leases and 587 * 1 should be dealt with via an fsync() system call for 588 * cases where write errors are important. 589 * 590 * nfs_close(struct vnode *a_vp, int a_fflag) 591 */ 592 /* ARGSUSED */ 593 static int 594 nfs_close(struct vop_close_args *ap) 595 { 596 struct vnode *vp = ap->a_vp; 597 struct nfsnode *np = VTONFS(vp); 598 int error = 0; 599 thread_t td = curthread; 600 601 if (vp->v_type == VREG) { 602 if (np->n_flag & NLMODIFIED) { 603 if (NFS_ISV3(vp)) { 604 /* 605 * Under NFSv3 we have dirty buffers to dispose of. We 606 * must flush them to the NFS server. We have the option 607 * of waiting all the way through the commit rpc or just 608 * waiting for the initial write. The default is to only 609 * wait through the initial write so the data is in the 610 * server's cache, which is roughly similar to the state 611 * a standard disk subsystem leaves the file in on close(). 612 * 613 * We cannot clear the NLMODIFIED bit in np->n_flag due to 614 * potential races with other processes, and certainly 615 * cannot clear it if we don't commit. 616 */ 617 int cm = nfsv3_commit_on_close ? 1 : 0; 618 error = nfs_flush(vp, MNT_WAIT, td, cm); 619 /* np->n_flag &= ~NLMODIFIED; */ 620 } else { 621 error = nfs_vinvalbuf(vp, V_SAVE, 1); 622 } 623 np->n_attrstamp = 0; 624 } 625 if (np->n_flag & NWRITEERR) { 626 np->n_flag &= ~NWRITEERR; 627 error = np->n_error; 628 } 629 } 630 vop_stdclose(ap); 631 return (error); 632 } 633 634 /* 635 * nfs getattr call from vfs. 636 * 637 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap) 638 */ 639 static int 640 nfs_getattr(struct vop_getattr_args *ap) 641 { 642 struct vnode *vp = ap->a_vp; 643 struct nfsnode *np = VTONFS(vp); 644 struct nfsmount *nmp; 645 int error = 0; 646 thread_t td = curthread; 647 struct nfsm_info info; 648 649 info.mrep = NULL; 650 info.v3 = NFS_ISV3(vp); 651 nmp = VFSTONFS(vp->v_mount); 652 653 /* 654 * Update local times for special files. 655 */ 656 if (np->n_flag & (NACC | NUPD)) 657 np->n_flag |= NCHG; 658 /* 659 * First look in the cache. 660 */ 661 if (nfs_getattrcache(vp, ap->a_vap) == 0) 662 goto done; 663 664 if (info.v3 && nfsaccess_cache_timeout > 0) { 665 nfsstats.accesscache_misses++; 666 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 667 if (nfs_getattrcache(vp, ap->a_vap) == 0) 668 goto done; 669 } 670 671 nfsstats.rpccnt[NFSPROC_GETATTR]++; 672 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3)); 673 ERROROUT(nfsm_fhtom(&info, vp)); 674 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td, 675 nfs_vpcred(vp, ND_CHECK), &error)); 676 if (error == 0) { 677 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap)); 678 } 679 m_freem(info.mrep); 680 info.mrep = NULL; 681 done: 682 /* 683 * NFS doesn't support chflags flags. If the nfs mount was 684 * made -o cache set the UF_CACHE bit for swapcache. 685 */ 686 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT)) 687 ap->a_vap->va_flags |= UF_CACHE; 688 nfsmout: 689 return (error); 690 } 691 692 /* 693 * nfs setattr call. 694 * 695 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 696 */ 697 static int 698 nfs_setattr(struct vop_setattr_args *ap) 699 { 700 struct vnode *vp = ap->a_vp; 701 struct nfsnode *np = VTONFS(vp); 702 struct vattr *vap = ap->a_vap; 703 int biosize = vp->v_mount->mnt_stat.f_iosize; 704 int error = 0; 705 int boff; 706 off_t tsize; 707 thread_t td = curthread; 708 709 #ifndef nolint 710 tsize = (off_t)0; 711 #endif 712 713 /* 714 * Setting of flags is not supported. 715 */ 716 if (vap->va_flags != VNOVAL) 717 return (EOPNOTSUPP); 718 719 /* 720 * Disallow write attempts if the filesystem is mounted read-only. 721 */ 722 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 723 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 724 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 725 (vp->v_mount->mnt_flag & MNT_RDONLY)) 726 return (EROFS); 727 728 if (vap->va_size != VNOVAL) { 729 /* 730 * truncation requested 731 */ 732 switch (vp->v_type) { 733 case VDIR: 734 return (EISDIR); 735 case VCHR: 736 case VBLK: 737 case VSOCK: 738 case VFIFO: 739 if (vap->va_mtime.tv_sec == VNOVAL && 740 vap->va_atime.tv_sec == VNOVAL && 741 vap->va_mode == (mode_t)VNOVAL && 742 vap->va_uid == (uid_t)VNOVAL && 743 vap->va_gid == (gid_t)VNOVAL) 744 return (0); 745 vap->va_size = VNOVAL; 746 break; 747 default: 748 /* 749 * Disallow write attempts if the filesystem is 750 * mounted read-only. 751 */ 752 if (vp->v_mount->mnt_flag & MNT_RDONLY) 753 return (EROFS); 754 755 tsize = np->n_size; 756 again: 757 boff = (int)vap->va_size & (biosize - 1); 758 error = nfs_meta_setsize(vp, td, vap->va_size, 0); 759 760 #if 0 761 if (np->n_flag & NLMODIFIED) { 762 if (vap->va_size == 0) 763 error = nfs_vinvalbuf(vp, 0, 1); 764 else 765 error = nfs_vinvalbuf(vp, V_SAVE, 1); 766 } 767 #endif 768 /* 769 * note: this loop case almost always happens at 770 * least once per truncation. 771 */ 772 if (error == 0 && np->n_size != vap->va_size) 773 goto again; 774 np->n_vattr.va_size = vap->va_size; 775 break; 776 } 777 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) { 778 /* 779 * What to do. If we are modifying the mtime we lose 780 * mtime detection of changes made by the server or other 781 * clients. But programs like rsync/rdist/cpdup are going 782 * to call utimes a lot. We don't want to piecemeal sync. 783 * 784 * For now sync if any prior remote changes were detected, 785 * but allow us to lose track of remote changes made during 786 * the utimes operation. 787 */ 788 if (np->n_flag & NRMODIFIED) 789 error = nfs_vinvalbuf(vp, V_SAVE, 1); 790 if (error == EINTR) 791 return (error); 792 if (error == 0) { 793 if (vap->va_mtime.tv_sec != VNOVAL) { 794 np->n_mtime = vap->va_mtime.tv_sec; 795 } 796 } 797 } 798 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 799 800 /* 801 * Sanity check if a truncation was issued. This should only occur 802 * if multiple processes are racing on the same file. 803 */ 804 if (error == 0 && vap->va_size != VNOVAL && 805 np->n_size != vap->va_size) { 806 kprintf("NFS ftruncate: server disagrees on the file size: " 807 "%jd/%jd/%jd\n", 808 (intmax_t)tsize, 809 (intmax_t)vap->va_size, 810 (intmax_t)np->n_size); 811 goto again; 812 } 813 if (error && vap->va_size != VNOVAL) { 814 np->n_size = np->n_vattr.va_size = tsize; 815 nfs_meta_setsize(vp, td, np->n_size, 0); 816 } 817 return (error); 818 } 819 820 /* 821 * Do an nfs setattr rpc. 822 */ 823 static int 824 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 825 struct ucred *cred, struct thread *td) 826 { 827 struct nfsv2_sattr *sp; 828 struct nfsnode *np = VTONFS(vp); 829 u_int32_t *tl; 830 int error = 0, wccflag = NFSV3_WCCRATTR; 831 struct nfsm_info info; 832 833 info.mrep = NULL; 834 info.v3 = NFS_ISV3(vp); 835 836 nfsstats.rpccnt[NFSPROC_SETATTR]++; 837 nfsm_reqhead(&info, vp, NFSPROC_SETATTR, 838 NFSX_FH(info.v3) + NFSX_SATTR(info.v3)); 839 ERROROUT(nfsm_fhtom(&info, vp)); 840 if (info.v3) { 841 nfsm_v3attrbuild(&info, vap, TRUE); 842 tl = nfsm_build(&info, NFSX_UNSIGNED); 843 *tl = nfs_false; 844 } else { 845 sp = nfsm_build(&info, NFSX_V2SATTR); 846 if (vap->va_mode == (mode_t)VNOVAL) 847 sp->sa_mode = nfs_xdrneg1; 848 else 849 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 850 if (vap->va_uid == (uid_t)VNOVAL) 851 sp->sa_uid = nfs_xdrneg1; 852 else 853 sp->sa_uid = txdr_unsigned(vap->va_uid); 854 if (vap->va_gid == (gid_t)VNOVAL) 855 sp->sa_gid = nfs_xdrneg1; 856 else 857 sp->sa_gid = txdr_unsigned(vap->va_gid); 858 sp->sa_size = txdr_unsigned(vap->va_size); 859 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 860 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 861 } 862 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error)); 863 if (info.v3) { 864 np->n_modestamp = 0; 865 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 866 } else { 867 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 868 } 869 m_freem(info.mrep); 870 info.mrep = NULL; 871 nfsmout: 872 return (error); 873 } 874 875 static 876 void 877 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout) 878 { 879 if (nctimeout == 0) 880 nctimeout = 1; 881 else 882 nctimeout *= hz; 883 cache_setvp(nch, vp); 884 cache_settimeout(nch, nctimeout); 885 } 886 887 /* 888 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 889 * nfs_lookup() until all remaining new api calls are implemented. 890 * 891 * Resolve a namecache entry. This function is passed a locked ncp and 892 * must call nfs_cache_setvp() on it as appropriate to resolve the entry. 893 */ 894 static int 895 nfs_nresolve(struct vop_nresolve_args *ap) 896 { 897 struct thread *td = curthread; 898 struct namecache *ncp; 899 struct ucred *cred; 900 struct nfsnode *np; 901 struct vnode *dvp; 902 struct vnode *nvp; 903 nfsfh_t *fhp; 904 int attrflag; 905 int fhsize; 906 int error; 907 int tmp_error; 908 int len; 909 struct nfsm_info info; 910 911 cred = ap->a_cred; 912 dvp = ap->a_dvp; 913 914 if ((error = vget(dvp, LK_SHARED)) != 0) 915 return (error); 916 917 info.mrep = NULL; 918 info.v3 = NFS_ISV3(dvp); 919 920 nvp = NULL; 921 nfsstats.lookupcache_misses++; 922 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 923 ncp = ap->a_nch->ncp; 924 len = ncp->nc_nlen; 925 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 926 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 927 ERROROUT(nfsm_fhtom(&info, dvp)); 928 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN)); 929 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, 930 ap->a_cred, &error)); 931 if (error) { 932 /* 933 * Cache negatve lookups to reduce NFS traffic, but use 934 * a fast timeout. Otherwise use a timeout of 1 tick. 935 * XXX we should add a namecache flag for no-caching 936 * to uncache the negative hit as soon as possible, but 937 * we cannot simply destroy the entry because it is used 938 * as a placeholder by the caller. 939 * 940 * The refactored nfs code will overwrite a non-zero error 941 * with 0 when we use ERROROUT(), so don't here. 942 */ 943 if (error == ENOENT) 944 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout); 945 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 946 NFS_LATTR_NOSHRINK); 947 if (tmp_error) { 948 error = tmp_error; 949 goto nfsmout; 950 } 951 m_freem(info.mrep); 952 info.mrep = NULL; 953 goto nfsmout; 954 } 955 956 /* 957 * Success, get the file handle, do various checks, and load 958 * post-operation data from the reply packet. Theoretically 959 * we should never be looking up "." so, theoretically, we 960 * should never get the same file handle as our directory. But 961 * we check anyway. XXX 962 * 963 * Note that no timeout is set for the positive cache hit. We 964 * assume, theoretically, that ESTALE returns will be dealt with 965 * properly to handle NFS races and in anycase we cannot depend 966 * on a timeout to deal with NFS open/create/excl issues so instead 967 * of a bad hack here the rest of the NFS client code needs to do 968 * the right thing. 969 */ 970 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 971 972 np = VTONFS(dvp); 973 if (NFS_CMPFH(np, fhp, fhsize)) { 974 vref(dvp); 975 nvp = dvp; 976 } else { 977 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 978 if (error) { 979 m_freem(info.mrep); 980 info.mrep = NULL; 981 vput(dvp); 982 return (error); 983 } 984 nvp = NFSTOV(np); 985 } 986 if (info.v3) { 987 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag, 988 NFS_LATTR_NOSHRINK)); 989 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 990 NFS_LATTR_NOSHRINK)); 991 } else { 992 ERROROUT(nfsm_loadattr(&info, nvp, NULL)); 993 } 994 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout); 995 m_freem(info.mrep); 996 info.mrep = NULL; 997 nfsmout: 998 vput(dvp); 999 if (nvp) { 1000 if (nvp == dvp) 1001 vrele(nvp); 1002 else 1003 vput(nvp); 1004 } 1005 return (error); 1006 } 1007 1008 /* 1009 * 'cached' nfs directory lookup 1010 * 1011 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 1012 * 1013 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 1014 * struct componentname *a_cnp) 1015 */ 1016 static int 1017 nfs_lookup(struct vop_old_lookup_args *ap) 1018 { 1019 struct componentname *cnp = ap->a_cnp; 1020 struct vnode *dvp = ap->a_dvp; 1021 struct vnode **vpp = ap->a_vpp; 1022 int flags = cnp->cn_flags; 1023 struct vnode *newvp; 1024 struct nfsmount *nmp; 1025 long len; 1026 nfsfh_t *fhp; 1027 struct nfsnode *np; 1028 int lockparent, wantparent, attrflag, fhsize; 1029 int error; 1030 int tmp_error; 1031 struct nfsm_info info; 1032 1033 info.mrep = NULL; 1034 info.v3 = NFS_ISV3(dvp); 1035 error = 0; 1036 1037 /* 1038 * Read-only mount check and directory check. 1039 */ 1040 *vpp = NULLVP; 1041 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1042 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1043 return (EROFS); 1044 1045 if (dvp->v_type != VDIR) 1046 return (ENOTDIR); 1047 1048 /* 1049 * Look it up in the cache. Note that ENOENT is only returned if we 1050 * previously entered a negative hit (see later on). The additional 1051 * nfsneg_cache_timeout check causes previously cached results to 1052 * be instantly ignored if the negative caching is turned off. 1053 */ 1054 lockparent = flags & CNP_LOCKPARENT; 1055 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1056 nmp = VFSTONFS(dvp->v_mount); 1057 np = VTONFS(dvp); 1058 1059 /* 1060 * Go to the wire. 1061 */ 1062 error = 0; 1063 newvp = NULLVP; 1064 nfsstats.lookupcache_misses++; 1065 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1066 len = cnp->cn_namelen; 1067 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 1068 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1069 ERROROUT(nfsm_fhtom(&info, dvp)); 1070 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 1071 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td, 1072 cnp->cn_cred, &error)); 1073 if (error) { 1074 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 1075 NFS_LATTR_NOSHRINK); 1076 if (tmp_error) { 1077 error = tmp_error; 1078 goto nfsmout; 1079 } 1080 1081 m_freem(info.mrep); 1082 info.mrep = NULL; 1083 goto nfsmout; 1084 } 1085 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 1086 1087 /* 1088 * Handle RENAME case... 1089 */ 1090 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1091 if (NFS_CMPFH(np, fhp, fhsize)) { 1092 m_freem(info.mrep); 1093 info.mrep = NULL; 1094 return (EISDIR); 1095 } 1096 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1097 if (error) { 1098 m_freem(info.mrep); 1099 info.mrep = NULL; 1100 return (error); 1101 } 1102 newvp = NFSTOV(np); 1103 if (info.v3) { 1104 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1105 NFS_LATTR_NOSHRINK)); 1106 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1107 NFS_LATTR_NOSHRINK)); 1108 } else { 1109 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1110 } 1111 *vpp = newvp; 1112 m_freem(info.mrep); 1113 info.mrep = NULL; 1114 if (!lockparent) { 1115 vn_unlock(dvp); 1116 cnp->cn_flags |= CNP_PDIRUNLOCK; 1117 } 1118 return (0); 1119 } 1120 1121 if (flags & CNP_ISDOTDOT) { 1122 vn_unlock(dvp); 1123 cnp->cn_flags |= CNP_PDIRUNLOCK; 1124 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1125 if (error) { 1126 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1127 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1128 return (error); /* NOTE: return error from nget */ 1129 } 1130 newvp = NFSTOV(np); 1131 if (lockparent) { 1132 error = vn_lock(dvp, LK_EXCLUSIVE); 1133 if (error) { 1134 vput(newvp); 1135 return (error); 1136 } 1137 cnp->cn_flags |= CNP_PDIRUNLOCK; 1138 } 1139 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1140 vref(dvp); 1141 newvp = dvp; 1142 } else { 1143 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1144 if (error) { 1145 m_freem(info.mrep); 1146 info.mrep = NULL; 1147 return (error); 1148 } 1149 if (!lockparent) { 1150 vn_unlock(dvp); 1151 cnp->cn_flags |= CNP_PDIRUNLOCK; 1152 } 1153 newvp = NFSTOV(np); 1154 } 1155 if (info.v3) { 1156 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1157 NFS_LATTR_NOSHRINK)); 1158 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1159 NFS_LATTR_NOSHRINK)); 1160 } else { 1161 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1162 } 1163 #if 0 1164 /* XXX MOVE TO nfs_nremove() */ 1165 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1166 cnp->cn_nameiop != NAMEI_DELETE) { 1167 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1168 } 1169 #endif 1170 *vpp = newvp; 1171 m_freem(info.mrep); 1172 info.mrep = NULL; 1173 nfsmout: 1174 if (error) { 1175 if (newvp != NULLVP) { 1176 vrele(newvp); 1177 *vpp = NULLVP; 1178 } 1179 if ((cnp->cn_nameiop == NAMEI_CREATE || 1180 cnp->cn_nameiop == NAMEI_RENAME) && 1181 error == ENOENT) { 1182 if (!lockparent) { 1183 vn_unlock(dvp); 1184 cnp->cn_flags |= CNP_PDIRUNLOCK; 1185 } 1186 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1187 error = EROFS; 1188 else 1189 error = EJUSTRETURN; 1190 } 1191 } 1192 return (error); 1193 } 1194 1195 /* 1196 * nfs read call. 1197 * Just call nfs_bioread() to do the work. 1198 * 1199 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1200 * struct ucred *a_cred) 1201 */ 1202 static int 1203 nfs_read(struct vop_read_args *ap) 1204 { 1205 struct vnode *vp = ap->a_vp; 1206 1207 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1208 } 1209 1210 /* 1211 * nfs readlink call 1212 * 1213 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1214 */ 1215 static int 1216 nfs_readlink(struct vop_readlink_args *ap) 1217 { 1218 struct vnode *vp = ap->a_vp; 1219 1220 if (vp->v_type != VLNK) 1221 return (EINVAL); 1222 return (nfs_bioread(vp, ap->a_uio, 0)); 1223 } 1224 1225 /* 1226 * Do a readlink rpc. 1227 * Called by nfs_doio() from below the buffer cache. 1228 */ 1229 int 1230 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop) 1231 { 1232 int error = 0, len, attrflag; 1233 struct nfsm_info info; 1234 1235 info.mrep = NULL; 1236 info.v3 = NFS_ISV3(vp); 1237 1238 nfsstats.rpccnt[NFSPROC_READLINK]++; 1239 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3)); 1240 ERROROUT(nfsm_fhtom(&info, vp)); 1241 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td, 1242 nfs_vpcred(vp, ND_CHECK), &error)); 1243 if (info.v3) { 1244 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1245 NFS_LATTR_NOSHRINK)); 1246 } 1247 if (!error) { 1248 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN)); 1249 if (len == NFS_MAXPATHLEN) { 1250 struct nfsnode *np = VTONFS(vp); 1251 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1252 len = np->n_size; 1253 } 1254 ERROROUT(nfsm_mtouio(&info, uiop, len)); 1255 } 1256 m_freem(info.mrep); 1257 info.mrep = NULL; 1258 nfsmout: 1259 return (error); 1260 } 1261 1262 /* 1263 * nfs synchronous read rpc using UIO 1264 */ 1265 int 1266 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop) 1267 { 1268 u_int32_t *tl; 1269 struct nfsmount *nmp; 1270 int error = 0, len, retlen, tsiz, eof, attrflag; 1271 struct nfsm_info info; 1272 off_t tmp_off; 1273 1274 info.mrep = NULL; 1275 info.v3 = NFS_ISV3(vp); 1276 1277 #ifndef nolint 1278 eof = 0; 1279 #endif 1280 nmp = VFSTONFS(vp->v_mount); 1281 tsiz = uiop->uio_resid; 1282 tmp_off = uiop->uio_offset + tsiz; 1283 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) 1284 return (EFBIG); 1285 tmp_off = uiop->uio_offset; 1286 while (tsiz > 0) { 1287 nfsstats.rpccnt[NFSPROC_READ]++; 1288 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1289 nfsm_reqhead(&info, vp, NFSPROC_READ, 1290 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3); 1291 ERROROUT(nfsm_fhtom(&info, vp)); 1292 tl = nfsm_build(&info, NFSX_UNSIGNED * 3); 1293 if (info.v3) { 1294 txdr_hyper(uiop->uio_offset, tl); 1295 *(tl + 2) = txdr_unsigned(len); 1296 } else { 1297 *tl++ = txdr_unsigned(uiop->uio_offset); 1298 *tl++ = txdr_unsigned(len); 1299 *tl = 0; 1300 } 1301 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td, 1302 nfs_vpcred(vp, ND_READ), &error)); 1303 if (info.v3) { 1304 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1305 NFS_LATTR_NOSHRINK)); 1306 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 1307 eof = fxdr_unsigned(int, *(tl + 1)); 1308 } else { 1309 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1310 } 1311 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len)); 1312 ERROROUT(nfsm_mtouio(&info, uiop, retlen)); 1313 m_freem(info.mrep); 1314 info.mrep = NULL; 1315 1316 /* 1317 * Handle short-read from server (NFSv3). If EOF is not 1318 * flagged (and no error occurred), but retlen is less 1319 * then the request size, we must zero-fill the remainder. 1320 */ 1321 if (retlen < len && info.v3 && eof == 0) { 1322 ERROROUT(uiomovez(len - retlen, uiop)); 1323 retlen = len; 1324 } 1325 tsiz -= retlen; 1326 1327 /* 1328 * Terminate loop on EOF or zero-length read. 1329 * 1330 * For NFSv2 a short-read indicates EOF, not zero-fill, 1331 * and also terminates the loop. 1332 */ 1333 if (info.v3) { 1334 if (eof || retlen == 0) 1335 tsiz = 0; 1336 } else if (retlen < len) { 1337 tsiz = 0; 1338 } 1339 } 1340 nfsmout: 1341 return (error); 1342 } 1343 1344 /* 1345 * nfs write call 1346 */ 1347 int 1348 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop, 1349 int *iomode, int *must_commit) 1350 { 1351 u_int32_t *tl; 1352 int32_t backup; 1353 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1354 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1355 int committed = NFSV3WRITE_FILESYNC; 1356 struct nfsm_info info; 1357 1358 info.mrep = NULL; 1359 info.v3 = NFS_ISV3(vp); 1360 1361 #ifndef DIAGNOSTIC 1362 if (uiop->uio_iovcnt != 1) 1363 panic("nfs: writerpc iovcnt > 1"); 1364 #endif 1365 *must_commit = 0; 1366 tsiz = uiop->uio_resid; 1367 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1368 return (EFBIG); 1369 while (tsiz > 0) { 1370 nfsstats.rpccnt[NFSPROC_WRITE]++; 1371 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1372 nfsm_reqhead(&info, vp, NFSPROC_WRITE, 1373 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1374 ERROROUT(nfsm_fhtom(&info, vp)); 1375 if (info.v3) { 1376 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 1377 txdr_hyper(uiop->uio_offset, tl); 1378 tl += 2; 1379 *tl++ = txdr_unsigned(len); 1380 *tl++ = txdr_unsigned(*iomode); 1381 *tl = txdr_unsigned(len); 1382 } else { 1383 u_int32_t x; 1384 1385 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED); 1386 /* Set both "begin" and "current" to non-garbage. */ 1387 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1388 *tl++ = x; /* "begin offset" */ 1389 *tl++ = x; /* "current offset" */ 1390 x = txdr_unsigned(len); 1391 *tl++ = x; /* total to this offset */ 1392 *tl = x; /* size of this write */ 1393 } 1394 ERROROUT(nfsm_uiotom(&info, uiop, len)); 1395 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td, 1396 nfs_vpcred(vp, ND_WRITE), &error)); 1397 if (info.v3) { 1398 /* 1399 * The write RPC returns a before and after mtime. The 1400 * nfsm_wcc_data() macro checks the before n_mtime 1401 * against the before time and stores the after time 1402 * in the nfsnode's cached vattr and n_mtime field. 1403 * The NRMODIFIED bit will be set if the before 1404 * time did not match the original mtime. 1405 */ 1406 wccflag = NFSV3_WCCCHK; 1407 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 1408 if (error == 0) { 1409 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF)); 1410 rlen = fxdr_unsigned(int, *tl++); 1411 if (rlen == 0) { 1412 error = NFSERR_IO; 1413 m_freem(info.mrep); 1414 info.mrep = NULL; 1415 break; 1416 } else if (rlen < len) { 1417 backup = len - rlen; 1418 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup; 1419 uiop->uio_iov->iov_len += backup; 1420 uiop->uio_offset -= backup; 1421 uiop->uio_resid += backup; 1422 len = rlen; 1423 } 1424 commit = fxdr_unsigned(int, *tl++); 1425 1426 /* 1427 * Return the lowest committment level 1428 * obtained by any of the RPCs. 1429 */ 1430 if (committed == NFSV3WRITE_FILESYNC) 1431 committed = commit; 1432 else if (committed == NFSV3WRITE_DATASYNC && 1433 commit == NFSV3WRITE_UNSTABLE) 1434 committed = commit; 1435 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1436 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1437 NFSX_V3WRITEVERF); 1438 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1439 } else if (bcmp((caddr_t)tl, 1440 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1441 *must_commit = 1; 1442 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1443 NFSX_V3WRITEVERF); 1444 } 1445 } 1446 } else { 1447 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1448 } 1449 m_freem(info.mrep); 1450 info.mrep = NULL; 1451 if (error) 1452 break; 1453 tsiz -= len; 1454 } 1455 nfsmout: 1456 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1457 committed = NFSV3WRITE_FILESYNC; 1458 *iomode = committed; 1459 if (error) 1460 uiop->uio_resid = tsiz; 1461 return (error); 1462 } 1463 1464 /* 1465 * nfs mknod rpc 1466 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1467 * mode set to specify the file type and the size field for rdev. 1468 */ 1469 static int 1470 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1471 struct vattr *vap) 1472 { 1473 struct nfsv2_sattr *sp; 1474 u_int32_t *tl; 1475 struct vnode *newvp = NULL; 1476 struct nfsnode *np = NULL; 1477 struct vattr vattr; 1478 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1479 int rmajor, rminor; 1480 struct nfsm_info info; 1481 1482 info.mrep = NULL; 1483 info.v3 = NFS_ISV3(dvp); 1484 1485 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1486 rmajor = txdr_unsigned(vap->va_rmajor); 1487 rminor = txdr_unsigned(vap->va_rminor); 1488 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) { 1489 rmajor = nfs_xdrneg1; 1490 rminor = nfs_xdrneg1; 1491 } else { 1492 return (EOPNOTSUPP); 1493 } 1494 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1495 return (error); 1496 } 1497 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1498 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD, 1499 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED + 1500 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1501 ERROROUT(nfsm_fhtom(&info, dvp)); 1502 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1503 NFS_MAXNAMLEN)); 1504 if (info.v3) { 1505 tl = nfsm_build(&info, NFSX_UNSIGNED); 1506 *tl++ = vtonfsv3_type(vap->va_type); 1507 nfsm_v3attrbuild(&info, vap, FALSE); 1508 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1509 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 1510 *tl++ = txdr_unsigned(vap->va_rmajor); 1511 *tl = txdr_unsigned(vap->va_rminor); 1512 } 1513 } else { 1514 sp = nfsm_build(&info, NFSX_V2SATTR); 1515 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1516 sp->sa_uid = nfs_xdrneg1; 1517 sp->sa_gid = nfs_xdrneg1; 1518 sp->sa_size = makeudev(rmajor, rminor); 1519 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1520 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1521 } 1522 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td, 1523 cnp->cn_cred, &error)); 1524 if (!error) { 1525 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1526 if (!gotvp) { 1527 if (newvp) { 1528 vput(newvp); 1529 newvp = NULL; 1530 } 1531 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1532 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1533 if (!error) 1534 newvp = NFSTOV(np); 1535 } 1536 } 1537 if (info.v3) { 1538 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1539 } 1540 m_freem(info.mrep); 1541 info.mrep = NULL; 1542 nfsmout: 1543 if (error) { 1544 if (newvp) 1545 vput(newvp); 1546 } else { 1547 *vpp = newvp; 1548 } 1549 VTONFS(dvp)->n_flag |= NLMODIFIED; 1550 if (!wccflag) 1551 VTONFS(dvp)->n_attrstamp = 0; 1552 return (error); 1553 } 1554 1555 /* 1556 * nfs mknod vop 1557 * just call nfs_mknodrpc() to do the work. 1558 * 1559 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1560 * struct componentname *a_cnp, struct vattr *a_vap) 1561 */ 1562 /* ARGSUSED */ 1563 static int 1564 nfs_mknod(struct vop_old_mknod_args *ap) 1565 { 1566 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1567 } 1568 1569 static u_long create_verf; 1570 /* 1571 * nfs file create call 1572 * 1573 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1574 * struct componentname *a_cnp, struct vattr *a_vap) 1575 */ 1576 static int 1577 nfs_create(struct vop_old_create_args *ap) 1578 { 1579 struct vnode *dvp = ap->a_dvp; 1580 struct vattr *vap = ap->a_vap; 1581 struct componentname *cnp = ap->a_cnp; 1582 struct nfsv2_sattr *sp; 1583 u_int32_t *tl; 1584 struct nfsnode *np = NULL; 1585 struct vnode *newvp = NULL; 1586 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1587 struct vattr vattr; 1588 struct nfsm_info info; 1589 1590 info.mrep = NULL; 1591 info.v3 = NFS_ISV3(dvp); 1592 1593 /* 1594 * Oops, not for me.. 1595 */ 1596 if (vap->va_type == VSOCK) 1597 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1598 1599 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1600 return (error); 1601 } 1602 if (vap->va_vaflags & VA_EXCLUSIVE) 1603 fmode |= O_EXCL; 1604 again: 1605 nfsstats.rpccnt[NFSPROC_CREATE]++; 1606 nfsm_reqhead(&info, dvp, NFSPROC_CREATE, 1607 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED + 1608 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1609 ERROROUT(nfsm_fhtom(&info, dvp)); 1610 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1611 NFS_MAXNAMLEN)); 1612 if (info.v3) { 1613 tl = nfsm_build(&info, NFSX_UNSIGNED); 1614 if (fmode & O_EXCL) { 1615 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1616 tl = nfsm_build(&info, NFSX_V3CREATEVERF); 1617 #ifdef INET 1618 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1619 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr; 1620 else 1621 #endif 1622 *tl++ = create_verf; 1623 *tl = ++create_verf; 1624 } else { 1625 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1626 nfsm_v3attrbuild(&info, vap, FALSE); 1627 } 1628 } else { 1629 sp = nfsm_build(&info, NFSX_V2SATTR); 1630 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1631 sp->sa_uid = nfs_xdrneg1; 1632 sp->sa_gid = nfs_xdrneg1; 1633 sp->sa_size = 0; 1634 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1635 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1636 } 1637 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td, 1638 cnp->cn_cred, &error)); 1639 if (error == 0) { 1640 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1641 if (!gotvp) { 1642 if (newvp) { 1643 vput(newvp); 1644 newvp = NULL; 1645 } 1646 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1647 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1648 if (!error) 1649 newvp = NFSTOV(np); 1650 } 1651 } 1652 if (info.v3) { 1653 if (error == 0) 1654 error = nfsm_wcc_data(&info, dvp, &wccflag); 1655 else 1656 (void)nfsm_wcc_data(&info, dvp, &wccflag); 1657 } 1658 m_freem(info.mrep); 1659 info.mrep = NULL; 1660 nfsmout: 1661 if (error) { 1662 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1663 KKASSERT(newvp == NULL); 1664 fmode &= ~O_EXCL; 1665 goto again; 1666 } 1667 } else if (info.v3 && (fmode & O_EXCL)) { 1668 /* 1669 * We are normally called with only a partially initialized 1670 * VAP. Since the NFSv3 spec says that server may use the 1671 * file attributes to store the verifier, the spec requires 1672 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1673 * in atime, but we can't really assume that all servers will 1674 * so we ensure that our SETATTR sets both atime and mtime. 1675 */ 1676 if (vap->va_mtime.tv_sec == VNOVAL) 1677 vfs_timestamp(&vap->va_mtime); 1678 if (vap->va_atime.tv_sec == VNOVAL) 1679 vap->va_atime = vap->va_mtime; 1680 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1681 } 1682 if (error == 0) { 1683 /* 1684 * The new np may have enough info for access 1685 * checks, make sure rucred and wucred are 1686 * initialized for read and write rpc's. 1687 */ 1688 np = VTONFS(newvp); 1689 if (np->n_rucred == NULL) 1690 np->n_rucred = crhold(cnp->cn_cred); 1691 if (np->n_wucred == NULL) 1692 np->n_wucred = crhold(cnp->cn_cred); 1693 *ap->a_vpp = newvp; 1694 } else if (newvp) { 1695 vput(newvp); 1696 } 1697 VTONFS(dvp)->n_flag |= NLMODIFIED; 1698 if (!wccflag) 1699 VTONFS(dvp)->n_attrstamp = 0; 1700 return (error); 1701 } 1702 1703 /* 1704 * nfs file remove call 1705 * To try and make nfs semantics closer to ufs semantics, a file that has 1706 * other processes using the vnode is renamed instead of removed and then 1707 * removed later on the last close. 1708 * - If v_sysref.refcnt > 1 1709 * If a rename is not already in the works 1710 * call nfs_sillyrename() to set it up 1711 * else 1712 * do the remove rpc 1713 * 1714 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1715 * struct componentname *a_cnp) 1716 */ 1717 static int 1718 nfs_remove(struct vop_old_remove_args *ap) 1719 { 1720 struct vnode *vp = ap->a_vp; 1721 struct vnode *dvp = ap->a_dvp; 1722 struct componentname *cnp = ap->a_cnp; 1723 struct nfsnode *np = VTONFS(vp); 1724 int error = 0; 1725 struct vattr vattr; 1726 1727 #ifndef DIAGNOSTIC 1728 if (vp->v_sysref.refcnt < 1) 1729 panic("nfs_remove: bad v_sysref.refcnt"); 1730 #endif 1731 if (vp->v_type == VDIR) 1732 error = EPERM; 1733 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename && 1734 VOP_GETATTR(vp, &vattr) == 0 && 1735 vattr.va_nlink > 1)) { 1736 /* 1737 * throw away biocache buffers, mainly to avoid 1738 * unnecessary delayed writes later. 1739 */ 1740 error = nfs_vinvalbuf(vp, 0, 1); 1741 /* Do the rpc */ 1742 if (error != EINTR) 1743 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1744 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1745 /* 1746 * Kludge City: If the first reply to the remove rpc is lost.. 1747 * the reply to the retransmitted request will be ENOENT 1748 * since the file was in fact removed 1749 * Therefore, we cheat and return success. 1750 */ 1751 if (error == ENOENT) 1752 error = 0; 1753 } else if (!np->n_sillyrename) { 1754 error = nfs_sillyrename(dvp, vp, cnp); 1755 } 1756 np->n_attrstamp = 0; 1757 return (error); 1758 } 1759 1760 /* 1761 * nfs file remove rpc called from nfs_inactive 1762 */ 1763 int 1764 nfs_removeit(struct sillyrename *sp) 1765 { 1766 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1767 sp->s_cred, NULL)); 1768 } 1769 1770 /* 1771 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1772 */ 1773 static int 1774 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1775 struct ucred *cred, struct thread *td) 1776 { 1777 int error = 0, wccflag = NFSV3_WCCRATTR; 1778 struct nfsm_info info; 1779 1780 info.mrep = NULL; 1781 info.v3 = NFS_ISV3(dvp); 1782 1783 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1784 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE, 1785 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1786 ERROROUT(nfsm_fhtom(&info, dvp)); 1787 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN)); 1788 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error)); 1789 if (info.v3) { 1790 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1791 } 1792 m_freem(info.mrep); 1793 info.mrep = NULL; 1794 nfsmout: 1795 VTONFS(dvp)->n_flag |= NLMODIFIED; 1796 if (!wccflag) 1797 VTONFS(dvp)->n_attrstamp = 0; 1798 return (error); 1799 } 1800 1801 /* 1802 * nfs file rename call 1803 * 1804 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1805 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1806 * struct vnode *a_tvp, struct componentname *a_tcnp) 1807 */ 1808 static int 1809 nfs_rename(struct vop_old_rename_args *ap) 1810 { 1811 struct vnode *fvp = ap->a_fvp; 1812 struct vnode *tvp = ap->a_tvp; 1813 struct vnode *fdvp = ap->a_fdvp; 1814 struct vnode *tdvp = ap->a_tdvp; 1815 struct componentname *tcnp = ap->a_tcnp; 1816 struct componentname *fcnp = ap->a_fcnp; 1817 int error; 1818 1819 /* Check for cross-device rename */ 1820 if ((fvp->v_mount != tdvp->v_mount) || 1821 (tvp && (fvp->v_mount != tvp->v_mount))) { 1822 error = EXDEV; 1823 goto out; 1824 } 1825 1826 /* 1827 * We shouldn't have to flush fvp on rename for most server-side 1828 * filesystems as the file handle should not change. Unfortunately 1829 * the inode for some filesystems (msdosfs) might be tied to the 1830 * file name or directory position so to be completely safe 1831 * vfs.nfs.flush_on_rename is set by default. Clear to improve 1832 * performance. 1833 * 1834 * We must flush tvp on rename because it might become stale on the 1835 * server after the rename. 1836 */ 1837 if (nfs_flush_on_rename) 1838 VOP_FSYNC(fvp, MNT_WAIT, 0); 1839 if (tvp) 1840 VOP_FSYNC(tvp, MNT_WAIT, 0); 1841 1842 /* 1843 * If the tvp exists and is in use, sillyrename it before doing the 1844 * rename of the new file over it. 1845 * 1846 * XXX Can't sillyrename a directory. 1847 * 1848 * We do not attempt to do any namecache purges in this old API 1849 * routine. The new API compat functions have access to the actual 1850 * namecache structures and will do it for us. 1851 */ 1852 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename && 1853 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1854 vput(tvp); 1855 tvp = NULL; 1856 } else if (tvp) { 1857 ; 1858 } 1859 1860 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1861 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1862 tcnp->cn_td); 1863 1864 out: 1865 if (tdvp == tvp) 1866 vrele(tdvp); 1867 else 1868 vput(tdvp); 1869 if (tvp) 1870 vput(tvp); 1871 vrele(fdvp); 1872 vrele(fvp); 1873 /* 1874 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1875 */ 1876 if (error == ENOENT) 1877 error = 0; 1878 return (error); 1879 } 1880 1881 /* 1882 * nfs file rename rpc called from nfs_remove() above 1883 */ 1884 static int 1885 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1886 struct sillyrename *sp) 1887 { 1888 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1889 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1890 } 1891 1892 /* 1893 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1894 */ 1895 static int 1896 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1897 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1898 struct ucred *cred, struct thread *td) 1899 { 1900 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1901 struct nfsm_info info; 1902 1903 info.mrep = NULL; 1904 info.v3 = NFS_ISV3(fdvp); 1905 1906 nfsstats.rpccnt[NFSPROC_RENAME]++; 1907 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME, 1908 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 + 1909 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen)); 1910 ERROROUT(nfsm_fhtom(&info, fdvp)); 1911 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN)); 1912 ERROROUT(nfsm_fhtom(&info, tdvp)); 1913 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN)); 1914 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error)); 1915 if (info.v3) { 1916 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag)); 1917 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag)); 1918 } 1919 m_freem(info.mrep); 1920 info.mrep = NULL; 1921 nfsmout: 1922 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1923 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1924 if (!fwccflag) 1925 VTONFS(fdvp)->n_attrstamp = 0; 1926 if (!twccflag) 1927 VTONFS(tdvp)->n_attrstamp = 0; 1928 return (error); 1929 } 1930 1931 /* 1932 * nfs hard link create call 1933 * 1934 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1935 * struct componentname *a_cnp) 1936 */ 1937 static int 1938 nfs_link(struct vop_old_link_args *ap) 1939 { 1940 struct vnode *vp = ap->a_vp; 1941 struct vnode *tdvp = ap->a_tdvp; 1942 struct componentname *cnp = ap->a_cnp; 1943 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1944 struct nfsm_info info; 1945 1946 if (vp->v_mount != tdvp->v_mount) { 1947 return (EXDEV); 1948 } 1949 1950 /* 1951 * The attribute cache may get out of sync with the server on link. 1952 * Pushing writes to the server before handle was inherited from 1953 * long long ago and it is unclear if we still need to do this. 1954 * Defaults to off. 1955 */ 1956 if (nfs_flush_on_hlink) 1957 VOP_FSYNC(vp, MNT_WAIT, 0); 1958 1959 info.mrep = NULL; 1960 info.v3 = NFS_ISV3(vp); 1961 1962 nfsstats.rpccnt[NFSPROC_LINK]++; 1963 nfsm_reqhead(&info, vp, NFSPROC_LINK, 1964 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED + 1965 nfsm_rndup(cnp->cn_namelen)); 1966 ERROROUT(nfsm_fhtom(&info, vp)); 1967 ERROROUT(nfsm_fhtom(&info, tdvp)); 1968 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1969 NFS_MAXNAMLEN)); 1970 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td, 1971 cnp->cn_cred, &error)); 1972 if (info.v3) { 1973 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1974 NFS_LATTR_NOSHRINK)); 1975 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag)); 1976 } 1977 m_freem(info.mrep); 1978 info.mrep = NULL; 1979 nfsmout: 1980 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1981 if (!attrflag) 1982 VTONFS(vp)->n_attrstamp = 0; 1983 if (!wccflag) 1984 VTONFS(tdvp)->n_attrstamp = 0; 1985 /* 1986 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1987 */ 1988 if (error == EEXIST) 1989 error = 0; 1990 return (error); 1991 } 1992 1993 /* 1994 * nfs symbolic link create call 1995 * 1996 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1997 * struct componentname *a_cnp, struct vattr *a_vap, 1998 * char *a_target) 1999 */ 2000 static int 2001 nfs_symlink(struct vop_old_symlink_args *ap) 2002 { 2003 struct vnode *dvp = ap->a_dvp; 2004 struct vattr *vap = ap->a_vap; 2005 struct componentname *cnp = ap->a_cnp; 2006 struct nfsv2_sattr *sp; 2007 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2008 struct vnode *newvp = NULL; 2009 struct nfsm_info info; 2010 2011 info.mrep = NULL; 2012 info.v3 = NFS_ISV3(dvp); 2013 2014 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2015 slen = strlen(ap->a_target); 2016 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK, 2017 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED + 2018 nfsm_rndup(cnp->cn_namelen) + 2019 nfsm_rndup(slen) + NFSX_SATTR(info.v3)); 2020 ERROROUT(nfsm_fhtom(&info, dvp)); 2021 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2022 NFS_MAXNAMLEN)); 2023 if (info.v3) { 2024 nfsm_v3attrbuild(&info, vap, FALSE); 2025 } 2026 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN)); 2027 if (info.v3 == 0) { 2028 sp = nfsm_build(&info, NFSX_V2SATTR); 2029 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2030 sp->sa_uid = nfs_xdrneg1; 2031 sp->sa_gid = nfs_xdrneg1; 2032 sp->sa_size = nfs_xdrneg1; 2033 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2034 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2035 } 2036 2037 /* 2038 * Issue the NFS request and get the rpc response. 2039 * 2040 * Only NFSv3 responses returning an error of 0 actually return 2041 * a file handle that can be converted into newvp without having 2042 * to do an extra lookup rpc. 2043 */ 2044 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td, 2045 cnp->cn_cred, &error)); 2046 if (info.v3) { 2047 if (error == 0) { 2048 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2049 } 2050 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2051 } 2052 2053 /* 2054 * out code jumps -> here, mrep is also freed. 2055 */ 2056 2057 m_freem(info.mrep); 2058 info.mrep = NULL; 2059 nfsmout: 2060 2061 /* 2062 * If we get an EEXIST error, silently convert it to no-error 2063 * in case of an NFS retry. 2064 */ 2065 if (error == EEXIST) 2066 error = 0; 2067 2068 /* 2069 * If we do not have (or no longer have) an error, and we could 2070 * not extract the newvp from the response due to the request being 2071 * NFSv2 or the error being EEXIST. We have to do a lookup in order 2072 * to obtain a newvp to return. 2073 */ 2074 if (error == 0 && newvp == NULL) { 2075 struct nfsnode *np = NULL; 2076 2077 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2078 cnp->cn_cred, cnp->cn_td, &np); 2079 if (!error) 2080 newvp = NFSTOV(np); 2081 } 2082 if (error) { 2083 if (newvp) 2084 vput(newvp); 2085 } else { 2086 *ap->a_vpp = newvp; 2087 } 2088 VTONFS(dvp)->n_flag |= NLMODIFIED; 2089 if (!wccflag) 2090 VTONFS(dvp)->n_attrstamp = 0; 2091 return (error); 2092 } 2093 2094 /* 2095 * nfs make dir call 2096 * 2097 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2098 * struct componentname *a_cnp, struct vattr *a_vap) 2099 */ 2100 static int 2101 nfs_mkdir(struct vop_old_mkdir_args *ap) 2102 { 2103 struct vnode *dvp = ap->a_dvp; 2104 struct vattr *vap = ap->a_vap; 2105 struct componentname *cnp = ap->a_cnp; 2106 struct nfsv2_sattr *sp; 2107 struct nfsnode *np = NULL; 2108 struct vnode *newvp = NULL; 2109 struct vattr vattr; 2110 int error = 0, wccflag = NFSV3_WCCRATTR; 2111 int gotvp = 0; 2112 int len; 2113 struct nfsm_info info; 2114 2115 info.mrep = NULL; 2116 info.v3 = NFS_ISV3(dvp); 2117 2118 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 2119 return (error); 2120 } 2121 len = cnp->cn_namelen; 2122 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2123 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR, 2124 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2125 nfsm_rndup(len) + NFSX_SATTR(info.v3)); 2126 ERROROUT(nfsm_fhtom(&info, dvp)); 2127 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 2128 if (info.v3) { 2129 nfsm_v3attrbuild(&info, vap, FALSE); 2130 } else { 2131 sp = nfsm_build(&info, NFSX_V2SATTR); 2132 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2133 sp->sa_uid = nfs_xdrneg1; 2134 sp->sa_gid = nfs_xdrneg1; 2135 sp->sa_size = nfs_xdrneg1; 2136 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2137 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2138 } 2139 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td, 2140 cnp->cn_cred, &error)); 2141 if (error == 0) { 2142 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2143 } 2144 if (info.v3) { 2145 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2146 } 2147 m_freem(info.mrep); 2148 info.mrep = NULL; 2149 nfsmout: 2150 VTONFS(dvp)->n_flag |= NLMODIFIED; 2151 if (!wccflag) 2152 VTONFS(dvp)->n_attrstamp = 0; 2153 /* 2154 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2155 * if we can succeed in looking up the directory. 2156 */ 2157 if (error == EEXIST || (!error && !gotvp)) { 2158 if (newvp) { 2159 vrele(newvp); 2160 newvp = NULL; 2161 } 2162 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2163 cnp->cn_td, &np); 2164 if (!error) { 2165 newvp = NFSTOV(np); 2166 if (newvp->v_type != VDIR) 2167 error = EEXIST; 2168 } 2169 } 2170 if (error) { 2171 if (newvp) 2172 vrele(newvp); 2173 } else 2174 *ap->a_vpp = newvp; 2175 return (error); 2176 } 2177 2178 /* 2179 * nfs remove directory call 2180 * 2181 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2182 * struct componentname *a_cnp) 2183 */ 2184 static int 2185 nfs_rmdir(struct vop_old_rmdir_args *ap) 2186 { 2187 struct vnode *vp = ap->a_vp; 2188 struct vnode *dvp = ap->a_dvp; 2189 struct componentname *cnp = ap->a_cnp; 2190 int error = 0, wccflag = NFSV3_WCCRATTR; 2191 struct nfsm_info info; 2192 2193 info.mrep = NULL; 2194 info.v3 = NFS_ISV3(dvp); 2195 2196 if (dvp == vp) 2197 return (EINVAL); 2198 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2199 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR, 2200 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2201 nfsm_rndup(cnp->cn_namelen)); 2202 ERROROUT(nfsm_fhtom(&info, dvp)); 2203 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2204 NFS_MAXNAMLEN)); 2205 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td, 2206 cnp->cn_cred, &error)); 2207 if (info.v3) { 2208 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2209 } 2210 m_freem(info.mrep); 2211 info.mrep = NULL; 2212 nfsmout: 2213 VTONFS(dvp)->n_flag |= NLMODIFIED; 2214 if (!wccflag) 2215 VTONFS(dvp)->n_attrstamp = 0; 2216 /* 2217 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2218 */ 2219 if (error == ENOENT) 2220 error = 0; 2221 return (error); 2222 } 2223 2224 /* 2225 * nfs readdir call 2226 * 2227 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2228 */ 2229 static int 2230 nfs_readdir(struct vop_readdir_args *ap) 2231 { 2232 struct vnode *vp = ap->a_vp; 2233 struct nfsnode *np = VTONFS(vp); 2234 struct uio *uio = ap->a_uio; 2235 int tresid, error; 2236 struct vattr vattr; 2237 2238 if (vp->v_type != VDIR) 2239 return (EPERM); 2240 2241 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 2242 return (error); 2243 2244 /* 2245 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2246 * and then check that is still valid, or if this is an NQNFS mount 2247 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2248 * VOP_GETATTR() does not necessarily go to the wire. 2249 */ 2250 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2251 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2252 if (VOP_GETATTR(vp, &vattr) == 0 && 2253 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2254 ) { 2255 nfsstats.direofcache_hits++; 2256 goto done; 2257 } 2258 } 2259 2260 /* 2261 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2262 * own cache coherency checks so we do not have to. 2263 */ 2264 tresid = uio->uio_resid; 2265 error = nfs_bioread(vp, uio, 0); 2266 2267 if (!error && uio->uio_resid == tresid) 2268 nfsstats.direofcache_misses++; 2269 done: 2270 vn_unlock(vp); 2271 return (error); 2272 } 2273 2274 /* 2275 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2276 * 2277 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2278 * offset/block and converts the nfs formatted directory entries for userland 2279 * consumption as well as deals with offsets into the middle of blocks. 2280 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2281 * be block-bounded. It must convert to cookies for the actual RPC. 2282 */ 2283 int 2284 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop) 2285 { 2286 int len, left; 2287 struct nfs_dirent *dp = NULL; 2288 u_int32_t *tl; 2289 nfsuint64 *cookiep; 2290 caddr_t cp; 2291 nfsuint64 cookie; 2292 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2293 struct nfsnode *dnp = VTONFS(vp); 2294 u_quad_t fileno; 2295 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2296 int attrflag; 2297 struct nfsm_info info; 2298 2299 info.mrep = NULL; 2300 info.v3 = NFS_ISV3(vp); 2301 2302 #ifndef DIAGNOSTIC 2303 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2304 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2305 panic("nfs readdirrpc bad uio"); 2306 #endif 2307 2308 /* 2309 * If there is no cookie, assume directory was stale. 2310 */ 2311 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2312 if (cookiep) 2313 cookie = *cookiep; 2314 else 2315 return (NFSERR_BAD_COOKIE); 2316 /* 2317 * Loop around doing readdir rpc's of size nm_readdirsize 2318 * truncated to a multiple of DIRBLKSIZ. 2319 * The stopping criteria is EOF or buffer full. 2320 */ 2321 while (more_dirs && bigenough) { 2322 nfsstats.rpccnt[NFSPROC_READDIR]++; 2323 nfsm_reqhead(&info, vp, NFSPROC_READDIR, 2324 NFSX_FH(info.v3) + NFSX_READDIR(info.v3)); 2325 ERROROUT(nfsm_fhtom(&info, vp)); 2326 if (info.v3) { 2327 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 2328 *tl++ = cookie.nfsuquad[0]; 2329 *tl++ = cookie.nfsuquad[1]; 2330 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2331 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2332 } else { 2333 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 2334 *tl++ = cookie.nfsuquad[0]; 2335 } 2336 *tl = txdr_unsigned(nmp->nm_readdirsize); 2337 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR, 2338 uiop->uio_td, 2339 nfs_vpcred(vp, ND_READ), &error)); 2340 if (info.v3) { 2341 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2342 NFS_LATTR_NOSHRINK)); 2343 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2344 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2345 dnp->n_cookieverf.nfsuquad[1] = *tl; 2346 } 2347 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2348 more_dirs = fxdr_unsigned(int, *tl); 2349 2350 /* loop thru the dir entries, converting them to std form */ 2351 while (more_dirs && bigenough) { 2352 if (info.v3) { 2353 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2354 fileno = fxdr_hyper(tl); 2355 len = fxdr_unsigned(int, *(tl + 2)); 2356 } else { 2357 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2358 fileno = fxdr_unsigned(u_quad_t, *tl++); 2359 len = fxdr_unsigned(int, *tl); 2360 } 2361 if (len <= 0 || len > NFS_MAXNAMLEN) { 2362 error = EBADRPC; 2363 m_freem(info.mrep); 2364 info.mrep = NULL; 2365 goto nfsmout; 2366 } 2367 2368 /* 2369 * len is the number of bytes in the path element 2370 * name, not including the \0 termination. 2371 * 2372 * tlen is the number of bytes w have to reserve for 2373 * the path element name. 2374 */ 2375 tlen = nfsm_rndup(len); 2376 if (tlen == len) 2377 tlen += 4; /* To ensure null termination */ 2378 2379 /* 2380 * If the entry would cross a DIRBLKSIZ boundary, 2381 * extend the previous nfs_dirent to cover the 2382 * remaining space. 2383 */ 2384 left = DIRBLKSIZ - blksiz; 2385 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2386 dp->nfs_reclen += left; 2387 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2388 uiop->uio_iov->iov_len -= left; 2389 uiop->uio_offset += left; 2390 uiop->uio_resid -= left; 2391 blksiz = 0; 2392 } 2393 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2394 bigenough = 0; 2395 if (bigenough) { 2396 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2397 dp->nfs_ino = fileno; 2398 dp->nfs_namlen = len; 2399 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2400 dp->nfs_type = DT_UNKNOWN; 2401 blksiz += dp->nfs_reclen; 2402 if (blksiz == DIRBLKSIZ) 2403 blksiz = 0; 2404 uiop->uio_offset += sizeof(struct nfs_dirent); 2405 uiop->uio_resid -= sizeof(struct nfs_dirent); 2406 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2407 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2408 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2409 2410 /* 2411 * The uiop has advanced by nfs_dirent + len 2412 * but really needs to advance by 2413 * nfs_dirent + tlen 2414 */ 2415 cp = uiop->uio_iov->iov_base; 2416 tlen -= len; 2417 *cp = '\0'; /* null terminate */ 2418 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2419 uiop->uio_iov->iov_len -= tlen; 2420 uiop->uio_offset += tlen; 2421 uiop->uio_resid -= tlen; 2422 } else { 2423 /* 2424 * NFS strings must be rounded up (nfsm_myouio 2425 * handled that in the bigenough case). 2426 */ 2427 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2428 } 2429 if (info.v3) { 2430 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2431 } else { 2432 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2433 } 2434 2435 /* 2436 * If we were able to accomodate the last entry, 2437 * get the cookie for the next one. Otherwise 2438 * hold-over the cookie for the one we were not 2439 * able to accomodate. 2440 */ 2441 if (bigenough) { 2442 cookie.nfsuquad[0] = *tl++; 2443 if (info.v3) 2444 cookie.nfsuquad[1] = *tl++; 2445 } else if (info.v3) { 2446 tl += 2; 2447 } else { 2448 tl++; 2449 } 2450 more_dirs = fxdr_unsigned(int, *tl); 2451 } 2452 /* 2453 * If at end of rpc data, get the eof boolean 2454 */ 2455 if (!more_dirs) { 2456 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2457 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2458 } 2459 m_freem(info.mrep); 2460 info.mrep = NULL; 2461 } 2462 /* 2463 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2464 * by increasing d_reclen for the last record. 2465 */ 2466 if (blksiz > 0) { 2467 left = DIRBLKSIZ - blksiz; 2468 dp->nfs_reclen += left; 2469 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2470 uiop->uio_iov->iov_len -= left; 2471 uiop->uio_offset += left; 2472 uiop->uio_resid -= left; 2473 } 2474 2475 if (bigenough) { 2476 /* 2477 * We hit the end of the directory, update direofoffset. 2478 */ 2479 dnp->n_direofoffset = uiop->uio_offset; 2480 } else { 2481 /* 2482 * There is more to go, insert the link cookie so the 2483 * next block can be read. 2484 */ 2485 if (uiop->uio_resid > 0) 2486 kprintf("EEK! readdirrpc resid > 0\n"); 2487 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2488 *cookiep = cookie; 2489 } 2490 nfsmout: 2491 return (error); 2492 } 2493 2494 /* 2495 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2496 */ 2497 int 2498 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop) 2499 { 2500 int len, left; 2501 struct nfs_dirent *dp; 2502 u_int32_t *tl; 2503 struct vnode *newvp; 2504 nfsuint64 *cookiep; 2505 caddr_t dpossav1, dpossav2; 2506 caddr_t cp; 2507 struct mbuf *mdsav1, *mdsav2; 2508 nfsuint64 cookie; 2509 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2510 struct nfsnode *dnp = VTONFS(vp), *np; 2511 nfsfh_t *fhp; 2512 u_quad_t fileno; 2513 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2514 int attrflag, fhsize; 2515 struct nchandle nch; 2516 struct nchandle dnch; 2517 struct nlcomponent nlc; 2518 struct nfsm_info info; 2519 2520 info.mrep = NULL; 2521 info.v3 = 1; 2522 2523 #ifndef nolint 2524 dp = NULL; 2525 #endif 2526 #ifndef DIAGNOSTIC 2527 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2528 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2529 panic("nfs readdirplusrpc bad uio"); 2530 #endif 2531 /* 2532 * Obtain the namecache record for the directory so we have something 2533 * to use as a basis for creating the entries. This function will 2534 * return a held (but not locked) ncp. The ncp may be disconnected 2535 * from the tree and cannot be used for upward traversals, and the 2536 * ncp may be unnamed. Note that other unrelated operations may 2537 * cause the ncp to be named at any time. 2538 * 2539 * We have to lock the ncp to prevent a lock order reversal when 2540 * rdirplus does nlookups of the children, because the vnode is 2541 * locked and has to stay that way. 2542 */ 2543 cache_fromdvp(vp, NULL, 0, &dnch); 2544 bzero(&nlc, sizeof(nlc)); 2545 newvp = NULLVP; 2546 2547 /* 2548 * If there is no cookie, assume directory was stale. 2549 */ 2550 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2551 if (cookiep) { 2552 cookie = *cookiep; 2553 } else { 2554 if (dnch.ncp) 2555 cache_drop(&dnch); 2556 return (NFSERR_BAD_COOKIE); 2557 } 2558 2559 /* 2560 * Loop around doing readdir rpc's of size nm_readdirsize 2561 * truncated to a multiple of DIRBLKSIZ. 2562 * The stopping criteria is EOF or buffer full. 2563 */ 2564 while (more_dirs && bigenough) { 2565 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2566 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS, 2567 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2568 ERROROUT(nfsm_fhtom(&info, vp)); 2569 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED); 2570 *tl++ = cookie.nfsuquad[0]; 2571 *tl++ = cookie.nfsuquad[1]; 2572 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2573 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2574 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2575 *tl = txdr_unsigned(nmp->nm_rsize); 2576 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS, 2577 uiop->uio_td, 2578 nfs_vpcred(vp, ND_READ), &error)); 2579 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2580 NFS_LATTR_NOSHRINK)); 2581 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2582 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2583 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2584 more_dirs = fxdr_unsigned(int, *tl); 2585 2586 /* loop thru the dir entries, doctoring them to 4bsd form */ 2587 while (more_dirs && bigenough) { 2588 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2589 fileno = fxdr_hyper(tl); 2590 len = fxdr_unsigned(int, *(tl + 2)); 2591 if (len <= 0 || len > NFS_MAXNAMLEN) { 2592 error = EBADRPC; 2593 m_freem(info.mrep); 2594 info.mrep = NULL; 2595 goto nfsmout; 2596 } 2597 tlen = nfsm_rndup(len); 2598 if (tlen == len) 2599 tlen += 4; /* To ensure null termination*/ 2600 left = DIRBLKSIZ - blksiz; 2601 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2602 dp->nfs_reclen += left; 2603 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2604 uiop->uio_iov->iov_len -= left; 2605 uiop->uio_offset += left; 2606 uiop->uio_resid -= left; 2607 blksiz = 0; 2608 } 2609 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2610 bigenough = 0; 2611 if (bigenough) { 2612 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2613 dp->nfs_ino = fileno; 2614 dp->nfs_namlen = len; 2615 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2616 dp->nfs_type = DT_UNKNOWN; 2617 blksiz += dp->nfs_reclen; 2618 if (blksiz == DIRBLKSIZ) 2619 blksiz = 0; 2620 uiop->uio_offset += sizeof(struct nfs_dirent); 2621 uiop->uio_resid -= sizeof(struct nfs_dirent); 2622 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2623 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2624 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2625 nlc.nlc_namelen = len; 2626 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2627 cp = uiop->uio_iov->iov_base; 2628 tlen -= len; 2629 *cp = '\0'; 2630 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2631 uiop->uio_iov->iov_len -= tlen; 2632 uiop->uio_offset += tlen; 2633 uiop->uio_resid -= tlen; 2634 } else { 2635 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2636 } 2637 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2638 if (bigenough) { 2639 cookie.nfsuquad[0] = *tl++; 2640 cookie.nfsuquad[1] = *tl++; 2641 } else { 2642 tl += 2; 2643 } 2644 2645 /* 2646 * Since the attributes are before the file handle 2647 * (sigh), we must skip over the attributes and then 2648 * come back and get them. 2649 */ 2650 attrflag = fxdr_unsigned(int, *tl); 2651 if (attrflag) { 2652 dpossav1 = info.dpos; 2653 mdsav1 = info.md; 2654 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR)); 2655 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2656 doit = fxdr_unsigned(int, *tl); 2657 if (doit) { 2658 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 2659 } 2660 if (doit && bigenough && !nlcdegenerate(&nlc) && 2661 !NFS_CMPFH(dnp, fhp, fhsize) 2662 ) { 2663 if (dnch.ncp) { 2664 #if 0 2665 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n", 2666 nlc.nlc_namelen, nlc.nlc_namelen, 2667 nlc.nlc_nameptr); 2668 #endif 2669 /* 2670 * This is a bit hokey but there isn't 2671 * much we can do about it. We can't 2672 * hold the directory vp locked while 2673 * doing lookups and gets. 2674 */ 2675 nch = cache_nlookup_nonblock(&dnch, &nlc); 2676 if (nch.ncp == NULL) 2677 goto rdfail; 2678 cache_setunresolved(&nch); 2679 error = nfs_nget_nonblock(vp->v_mount, fhp, 2680 fhsize, &np); 2681 if (error) { 2682 cache_put(&nch); 2683 goto rdfail; 2684 } 2685 newvp = NFSTOV(np); 2686 dpossav2 = info.dpos; 2687 info.dpos = dpossav1; 2688 mdsav2 = info.md; 2689 info.md = mdsav1; 2690 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 2691 info.dpos = dpossav2; 2692 info.md = mdsav2; 2693 dp->nfs_type = 2694 IFTODT(VTTOIF(np->n_vattr.va_type)); 2695 nfs_cache_setvp(&nch, newvp, 2696 nfspos_cache_timeout); 2697 vput(newvp); 2698 newvp = NULLVP; 2699 cache_put(&nch); 2700 } else { 2701 rdfail: 2702 ; 2703 #if 0 2704 kprintf("Warning: NFS/rddirplus, " 2705 "UNABLE TO ENTER %*.*s\n", 2706 nlc.nlc_namelen, nlc.nlc_namelen, 2707 nlc.nlc_nameptr); 2708 #endif 2709 } 2710 } 2711 } else { 2712 /* Just skip over the file handle */ 2713 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2714 i = fxdr_unsigned(int, *tl); 2715 ERROROUT(nfsm_adv(&info, nfsm_rndup(i))); 2716 } 2717 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2718 more_dirs = fxdr_unsigned(int, *tl); 2719 } 2720 /* 2721 * If at end of rpc data, get the eof boolean 2722 */ 2723 if (!more_dirs) { 2724 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2725 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2726 } 2727 m_freem(info.mrep); 2728 info.mrep = NULL; 2729 } 2730 /* 2731 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2732 * by increasing d_reclen for the last record. 2733 */ 2734 if (blksiz > 0) { 2735 left = DIRBLKSIZ - blksiz; 2736 dp->nfs_reclen += left; 2737 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2738 uiop->uio_iov->iov_len -= left; 2739 uiop->uio_offset += left; 2740 uiop->uio_resid -= left; 2741 } 2742 2743 /* 2744 * We are now either at the end of the directory or have filled the 2745 * block. 2746 */ 2747 if (bigenough) { 2748 dnp->n_direofoffset = uiop->uio_offset; 2749 } else { 2750 if (uiop->uio_resid > 0) 2751 kprintf("EEK! readdirplusrpc resid > 0\n"); 2752 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2753 *cookiep = cookie; 2754 } 2755 nfsmout: 2756 if (newvp != NULLVP) { 2757 if (newvp == vp) 2758 vrele(newvp); 2759 else 2760 vput(newvp); 2761 newvp = NULLVP; 2762 } 2763 if (dnch.ncp) 2764 cache_drop(&dnch); 2765 return (error); 2766 } 2767 2768 /* 2769 * Silly rename. To make the NFS filesystem that is stateless look a little 2770 * more like the "ufs" a remove of an active vnode is translated to a rename 2771 * to a funny looking filename that is removed by nfs_inactive on the 2772 * nfsnode. There is the potential for another process on a different client 2773 * to create the same funny name between the nfs_lookitup() fails and the 2774 * nfs_rename() completes, but... 2775 */ 2776 static int 2777 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2778 { 2779 struct sillyrename *sp; 2780 struct nfsnode *np; 2781 int error; 2782 2783 /* 2784 * We previously purged dvp instead of vp. I don't know why, it 2785 * completely destroys performance. We can't do it anyway with the 2786 * new VFS API since we would be breaking the namecache topology. 2787 */ 2788 cache_purge(vp); /* XXX */ 2789 np = VTONFS(vp); 2790 #ifndef DIAGNOSTIC 2791 if (vp->v_type == VDIR) 2792 panic("nfs: sillyrename dir"); 2793 #endif 2794 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2795 M_NFSREQ, M_WAITOK); 2796 sp->s_cred = crdup(cnp->cn_cred); 2797 sp->s_dvp = dvp; 2798 vref(dvp); 2799 2800 /* Fudge together a funny name */ 2801 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", 2802 (int)(intptr_t)cnp->cn_td); 2803 2804 /* Try lookitups until we get one that isn't there */ 2805 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2806 cnp->cn_td, NULL) == 0) { 2807 sp->s_name[4]++; 2808 if (sp->s_name[4] > 'z') { 2809 error = EINVAL; 2810 goto bad; 2811 } 2812 } 2813 error = nfs_renameit(dvp, cnp, sp); 2814 if (error) 2815 goto bad; 2816 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2817 cnp->cn_td, &np); 2818 np->n_sillyrename = sp; 2819 return (0); 2820 bad: 2821 vrele(sp->s_dvp); 2822 crfree(sp->s_cred); 2823 kfree((caddr_t)sp, M_NFSREQ); 2824 return (error); 2825 } 2826 2827 /* 2828 * Look up a file name and optionally either update the file handle or 2829 * allocate an nfsnode, depending on the value of npp. 2830 * npp == NULL --> just do the lookup 2831 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2832 * handled too 2833 * *npp != NULL --> update the file handle in the vnode 2834 */ 2835 static int 2836 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2837 struct thread *td, struct nfsnode **npp) 2838 { 2839 struct vnode *newvp = NULL; 2840 struct nfsnode *np, *dnp = VTONFS(dvp); 2841 int error = 0, fhlen, attrflag; 2842 nfsfh_t *nfhp; 2843 struct nfsm_info info; 2844 2845 info.mrep = NULL; 2846 info.v3 = NFS_ISV3(dvp); 2847 2848 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2849 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 2850 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2851 ERROROUT(nfsm_fhtom(&info, dvp)); 2852 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN)); 2853 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error)); 2854 if (npp && !error) { 2855 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp)); 2856 if (*npp) { 2857 np = *npp; 2858 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2859 kfree((caddr_t)np->n_fhp, M_NFSBIGFH); 2860 np->n_fhp = &np->n_fh; 2861 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2862 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK); 2863 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2864 np->n_fhsize = fhlen; 2865 newvp = NFSTOV(np); 2866 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2867 vref(dvp); 2868 newvp = dvp; 2869 } else { 2870 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2871 if (error) { 2872 m_freem(info.mrep); 2873 info.mrep = NULL; 2874 return (error); 2875 } 2876 newvp = NFSTOV(np); 2877 } 2878 if (info.v3) { 2879 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 2880 NFS_LATTR_NOSHRINK)); 2881 if (!attrflag && *npp == NULL) { 2882 m_freem(info.mrep); 2883 info.mrep = NULL; 2884 if (newvp == dvp) 2885 vrele(newvp); 2886 else 2887 vput(newvp); 2888 return (ENOENT); 2889 } 2890 } else { 2891 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL)); 2892 } 2893 } 2894 m_freem(info.mrep); 2895 info.mrep = NULL; 2896 nfsmout: 2897 if (npp && *npp == NULL) { 2898 if (error) { 2899 if (newvp) { 2900 if (newvp == dvp) 2901 vrele(newvp); 2902 else 2903 vput(newvp); 2904 } 2905 } else 2906 *npp = np; 2907 } 2908 return (error); 2909 } 2910 2911 /* 2912 * Nfs Version 3 commit rpc 2913 * 2914 * We call it 'uio' to distinguish it from 'bio' but there is no real uio 2915 * involved. 2916 */ 2917 int 2918 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2919 { 2920 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2921 int error = 0, wccflag = NFSV3_WCCRATTR; 2922 struct nfsm_info info; 2923 u_int32_t *tl; 2924 2925 info.mrep = NULL; 2926 info.v3 = 1; 2927 2928 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2929 return (0); 2930 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2931 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1)); 2932 ERROROUT(nfsm_fhtom(&info, vp)); 2933 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED); 2934 txdr_hyper(offset, tl); 2935 tl += 2; 2936 *tl = txdr_unsigned(cnt); 2937 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td, 2938 nfs_vpcred(vp, ND_WRITE), &error)); 2939 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 2940 if (!error) { 2941 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF)); 2942 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2943 NFSX_V3WRITEVERF)) { 2944 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2945 NFSX_V3WRITEVERF); 2946 error = NFSERR_STALEWRITEVERF; 2947 } 2948 } 2949 m_freem(info.mrep); 2950 info.mrep = NULL; 2951 nfsmout: 2952 return (error); 2953 } 2954 2955 /* 2956 * Kludge City.. 2957 * - make nfs_bmap() essentially a no-op that does no translation 2958 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2959 * (Maybe I could use the process's page mapping, but I was concerned that 2960 * Kernel Write might not be enabled and also figured copyout() would do 2961 * a lot more work than bcopy() and also it currently happens in the 2962 * context of the swapper process (2). 2963 * 2964 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, 2965 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2966 */ 2967 static int 2968 nfs_bmap(struct vop_bmap_args *ap) 2969 { 2970 if (ap->a_doffsetp != NULL) 2971 *ap->a_doffsetp = ap->a_loffset; 2972 if (ap->a_runp != NULL) 2973 *ap->a_runp = 0; 2974 if (ap->a_runb != NULL) 2975 *ap->a_runb = 0; 2976 return (0); 2977 } 2978 2979 /* 2980 * Strategy routine. 2981 */ 2982 static int 2983 nfs_strategy(struct vop_strategy_args *ap) 2984 { 2985 struct bio *bio = ap->a_bio; 2986 struct bio *nbio; 2987 struct buf *bp __debugvar = bio->bio_buf; 2988 struct thread *td; 2989 int error; 2990 2991 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2992 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2993 KASSERT(BUF_REFCNT(bp) > 0, 2994 ("nfs_strategy: buffer %p not locked", bp)); 2995 2996 if (bio->bio_flags & BIO_SYNC) 2997 td = curthread; /* XXX */ 2998 else 2999 td = NULL; 3000 3001 /* 3002 * We probably don't need to push an nbio any more since no 3003 * block conversion is required due to the use of 64 bit byte 3004 * offsets, but do it anyway. 3005 * 3006 * NOTE: When NFS callers itself via this strategy routines and 3007 * sets up a synchronous I/O, it expects the I/O to run 3008 * synchronously (its bio_done routine just assumes it), 3009 * so for now we have to honor the bit. 3010 */ 3011 nbio = push_bio(bio); 3012 nbio->bio_offset = bio->bio_offset; 3013 nbio->bio_flags = bio->bio_flags & BIO_SYNC; 3014 3015 /* 3016 * If the op is asynchronous and an i/o daemon is waiting 3017 * queue the request, wake it up and wait for completion 3018 * otherwise just do it ourselves. 3019 */ 3020 if (bio->bio_flags & BIO_SYNC) { 3021 error = nfs_doio(ap->a_vp, nbio, td); 3022 } else { 3023 nfs_asyncio(ap->a_vp, nbio); 3024 error = 0; 3025 } 3026 return (error); 3027 } 3028 3029 /* 3030 * Mmap a file 3031 * 3032 * NB Currently unsupported. 3033 * 3034 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 3035 */ 3036 /* ARGSUSED */ 3037 static int 3038 nfs_mmap(struct vop_mmap_args *ap) 3039 { 3040 return (EINVAL); 3041 } 3042 3043 /* 3044 * fsync vnode op. Just call nfs_flush() with commit == 1. 3045 * 3046 * nfs_fsync(struct vnode *a_vp, int a_waitfor) 3047 */ 3048 /* ARGSUSED */ 3049 static int 3050 nfs_fsync(struct vop_fsync_args *ap) 3051 { 3052 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1)); 3053 } 3054 3055 /* 3056 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 3057 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 3058 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 3059 * set the buffer contains data that has already been written to the server 3060 * and which now needs a commit RPC. 3061 * 3062 * If commit is 0 we only take one pass and only flush buffers containing new 3063 * dirty data. 3064 * 3065 * If commit is 1 we take two passes, issuing a commit RPC in the second 3066 * pass. 3067 * 3068 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 3069 * to completely flush all pending data. 3070 * 3071 * Note that the RB_SCAN code properly handles the case where the 3072 * callback might block and directly or indirectly (another thread) cause 3073 * the RB tree to change. 3074 */ 3075 3076 #ifndef NFS_COMMITBVECSIZ 3077 #define NFS_COMMITBVECSIZ 16 3078 #endif 3079 3080 struct nfs_flush_info { 3081 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 3082 struct thread *td; 3083 struct vnode *vp; 3084 int waitfor; 3085 int slpflag; 3086 int slptimeo; 3087 int loops; 3088 struct buf *bvary[NFS_COMMITBVECSIZ]; 3089 int bvsize; 3090 off_t beg_off; 3091 off_t end_off; 3092 }; 3093 3094 static int nfs_flush_bp(struct buf *bp, void *data); 3095 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 3096 3097 int 3098 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 3099 { 3100 struct nfsnode *np = VTONFS(vp); 3101 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3102 struct nfs_flush_info info; 3103 int error; 3104 3105 bzero(&info, sizeof(info)); 3106 info.td = td; 3107 info.vp = vp; 3108 info.waitfor = waitfor; 3109 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 3110 info.loops = 0; 3111 lwkt_gettoken(&vp->v_token); 3112 3113 do { 3114 /* 3115 * Flush mode 3116 */ 3117 info.mode = NFI_FLUSHNEW; 3118 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3119 nfs_flush_bp, &info); 3120 3121 /* 3122 * Take a second pass if committing and no error occured. 3123 * Clean up any left over collection (whether an error 3124 * occurs or not). 3125 */ 3126 if (commit && error == 0) { 3127 info.mode = NFI_COMMIT; 3128 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3129 nfs_flush_bp, &info); 3130 if (info.bvsize) 3131 error = nfs_flush_docommit(&info, error); 3132 } 3133 3134 /* 3135 * Wait for pending I/O to complete before checking whether 3136 * any further dirty buffers exist. 3137 */ 3138 while (waitfor == MNT_WAIT && 3139 bio_track_active(&vp->v_track_write)) { 3140 error = bio_track_wait(&vp->v_track_write, 3141 info.slpflag, info.slptimeo); 3142 if (error) { 3143 /* 3144 * We have to be able to break out if this 3145 * is an 'intr' mount. 3146 */ 3147 if (nfs_sigintr(nmp, NULL, td)) { 3148 error = -EINTR; 3149 break; 3150 } 3151 3152 /* 3153 * Since we do not process pending signals, 3154 * once we get a PCATCH our tsleep() will no 3155 * longer sleep, switch to a fixed timeout 3156 * instead. 3157 */ 3158 if (info.slpflag == PCATCH) { 3159 info.slpflag = 0; 3160 info.slptimeo = 2 * hz; 3161 } 3162 error = 0; 3163 } 3164 } 3165 ++info.loops; 3166 /* 3167 * Loop if we are flushing synchronous as well as committing, 3168 * and dirty buffers are still present. Otherwise we might livelock. 3169 */ 3170 } while (waitfor == MNT_WAIT && commit && 3171 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3172 3173 /* 3174 * The callbacks have to return a negative error to terminate the 3175 * RB scan. 3176 */ 3177 if (error < 0) 3178 error = -error; 3179 3180 /* 3181 * Deal with any error collection 3182 */ 3183 if (np->n_flag & NWRITEERR) { 3184 error = np->n_error; 3185 np->n_flag &= ~NWRITEERR; 3186 } 3187 lwkt_reltoken(&vp->v_token); 3188 return (error); 3189 } 3190 3191 static 3192 int 3193 nfs_flush_bp(struct buf *bp, void *data) 3194 { 3195 struct nfs_flush_info *info = data; 3196 int lkflags; 3197 int error; 3198 off_t toff; 3199 3200 error = 0; 3201 switch(info->mode) { 3202 case NFI_FLUSHNEW: 3203 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3204 if (error && info->loops && info->waitfor == MNT_WAIT) { 3205 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3206 if (error) { 3207 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3208 if (info->slpflag & PCATCH) 3209 lkflags |= LK_PCATCH; 3210 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3211 info->slptimeo); 3212 } 3213 } 3214 3215 /* 3216 * Ignore locking errors 3217 */ 3218 if (error) { 3219 error = 0; 3220 break; 3221 } 3222 3223 /* 3224 * The buffer may have changed out from under us, even if 3225 * we did not block (MPSAFE). Check again now that it is 3226 * locked. 3227 */ 3228 if (bp->b_vp == info->vp && 3229 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) { 3230 bremfree(bp); 3231 bawrite(bp); 3232 } else { 3233 BUF_UNLOCK(bp); 3234 } 3235 break; 3236 case NFI_COMMIT: 3237 /* 3238 * Only process buffers in need of a commit which we can 3239 * immediately lock. This may prevent a buffer from being 3240 * committed, but the normal flush loop will block on the 3241 * same buffer so we shouldn't get into an endless loop. 3242 */ 3243 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3244 (B_DELWRI | B_NEEDCOMMIT)) { 3245 break; 3246 } 3247 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 3248 break; 3249 3250 /* 3251 * We must recheck after successfully locking the buffer. 3252 */ 3253 if (bp->b_vp != info->vp || 3254 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3255 (B_DELWRI | B_NEEDCOMMIT)) { 3256 BUF_UNLOCK(bp); 3257 break; 3258 } 3259 3260 /* 3261 * NOTE: storing the bp in the bvary[] basically sets 3262 * it up for a commit operation. 3263 * 3264 * We must call vfs_busy_pages() now so the commit operation 3265 * is interlocked with user modifications to memory mapped 3266 * pages. The b_dirtyoff/b_dirtyend range is not correct 3267 * until after the pages have been busied. 3268 * 3269 * Note: to avoid loopback deadlocks, we do not 3270 * assign b_runningbufspace. 3271 */ 3272 bremfree(bp); 3273 bp->b_cmd = BUF_CMD_WRITE; 3274 vfs_busy_pages(bp->b_vp, bp); 3275 info->bvary[info->bvsize] = bp; 3276 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3277 if (info->bvsize == 0 || toff < info->beg_off) 3278 info->beg_off = toff; 3279 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3280 if (info->bvsize == 0 || toff > info->end_off) 3281 info->end_off = toff; 3282 ++info->bvsize; 3283 if (info->bvsize == NFS_COMMITBVECSIZ) { 3284 error = nfs_flush_docommit(info, 0); 3285 KKASSERT(info->bvsize == 0); 3286 } 3287 } 3288 return (error); 3289 } 3290 3291 static 3292 int 3293 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3294 { 3295 struct vnode *vp; 3296 struct buf *bp; 3297 off_t bytes; 3298 int retv; 3299 int i; 3300 3301 vp = info->vp; 3302 3303 if (info->bvsize > 0) { 3304 /* 3305 * Commit data on the server, as required. Note that 3306 * nfs_commit will use the vnode's cred for the commit. 3307 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3308 */ 3309 bytes = info->end_off - info->beg_off; 3310 if (bytes > 0x40000000) 3311 bytes = 0x40000000; 3312 if (error) { 3313 retv = -error; 3314 } else { 3315 retv = nfs_commitrpc_uio(vp, info->beg_off, 3316 (int)bytes, info->td); 3317 if (retv == NFSERR_STALEWRITEVERF) 3318 nfs_clearcommit(vp->v_mount); 3319 } 3320 3321 /* 3322 * Now, either mark the blocks I/O done or mark the 3323 * blocks dirty, depending on whether the commit 3324 * succeeded. 3325 */ 3326 for (i = 0; i < info->bvsize; ++i) { 3327 bp = info->bvary[i]; 3328 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) { 3329 /* 3330 * Either an error or the original 3331 * vfs_busy_pages() cleared B_NEEDCOMMIT 3332 * due to finding new dirty VM pages in 3333 * the buffer. 3334 * 3335 * Leave B_DELWRI intact. 3336 */ 3337 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3338 vfs_unbusy_pages(bp); 3339 bp->b_cmd = BUF_CMD_DONE; 3340 bqrelse(bp); 3341 } else { 3342 /* 3343 * Success, remove B_DELWRI ( bundirty() ). 3344 * 3345 * b_dirtyoff/b_dirtyend seem to be NFS 3346 * specific. We should probably move that 3347 * into bundirty(). XXX 3348 * 3349 * We are faking an I/O write, we have to 3350 * start the transaction in order to 3351 * immediately biodone() it. 3352 */ 3353 bundirty(bp); 3354 bp->b_flags &= ~B_ERROR; 3355 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3356 bp->b_dirtyoff = bp->b_dirtyend = 0; 3357 biodone(&bp->b_bio1); 3358 } 3359 } 3360 info->bvsize = 0; 3361 } 3362 return (error); 3363 } 3364 3365 /* 3366 * NFS advisory byte-level locks. 3367 * Currently unsupported. 3368 * 3369 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3370 * int a_flags) 3371 */ 3372 static int 3373 nfs_advlock(struct vop_advlock_args *ap) 3374 { 3375 struct nfsnode *np = VTONFS(ap->a_vp); 3376 3377 /* 3378 * The following kludge is to allow diskless support to work 3379 * until a real NFS lockd is implemented. Basically, just pretend 3380 * that this is a local lock. 3381 */ 3382 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3383 } 3384 3385 /* 3386 * Print out the contents of an nfsnode. 3387 * 3388 * nfs_print(struct vnode *a_vp) 3389 */ 3390 static int 3391 nfs_print(struct vop_print_args *ap) 3392 { 3393 struct vnode *vp = ap->a_vp; 3394 struct nfsnode *np = VTONFS(vp); 3395 3396 kprintf("tag VT_NFS, fileid %lld fsid 0x%x", 3397 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3398 if (vp->v_type == VFIFO) 3399 fifo_printinfo(vp); 3400 kprintf("\n"); 3401 return (0); 3402 } 3403 3404 /* 3405 * nfs special file access vnode op. 3406 * 3407 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 3408 */ 3409 static int 3410 nfs_laccess(struct vop_access_args *ap) 3411 { 3412 struct vattr vattr; 3413 int error; 3414 3415 error = VOP_GETATTR(ap->a_vp, &vattr); 3416 if (!error) 3417 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid, 3418 vattr.va_mode, 0); 3419 return (error); 3420 } 3421 3422 /* 3423 * Read wrapper for fifos. 3424 * 3425 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3426 * struct ucred *a_cred) 3427 */ 3428 static int 3429 nfsfifo_read(struct vop_read_args *ap) 3430 { 3431 struct nfsnode *np = VTONFS(ap->a_vp); 3432 3433 /* 3434 * Set access flag. 3435 */ 3436 np->n_flag |= NACC; 3437 getnanotime(&np->n_atim); 3438 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3439 } 3440 3441 /* 3442 * Write wrapper for fifos. 3443 * 3444 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3445 * struct ucred *a_cred) 3446 */ 3447 static int 3448 nfsfifo_write(struct vop_write_args *ap) 3449 { 3450 struct nfsnode *np = VTONFS(ap->a_vp); 3451 3452 /* 3453 * Set update flag. 3454 */ 3455 np->n_flag |= NUPD; 3456 getnanotime(&np->n_mtim); 3457 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3458 } 3459 3460 /* 3461 * Close wrapper for fifos. 3462 * 3463 * Update the times on the nfsnode then do fifo close. 3464 * 3465 * nfsfifo_close(struct vnode *a_vp, int a_fflag) 3466 */ 3467 static int 3468 nfsfifo_close(struct vop_close_args *ap) 3469 { 3470 struct vnode *vp = ap->a_vp; 3471 struct nfsnode *np = VTONFS(vp); 3472 struct vattr vattr; 3473 struct timespec ts; 3474 3475 if (np->n_flag & (NACC | NUPD)) { 3476 getnanotime(&ts); 3477 if (np->n_flag & NACC) 3478 np->n_atim = ts; 3479 if (np->n_flag & NUPD) 3480 np->n_mtim = ts; 3481 np->n_flag |= NCHG; 3482 if (vp->v_sysref.refcnt == 1 && 3483 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3484 VATTR_NULL(&vattr); 3485 if (np->n_flag & NACC) 3486 vattr.va_atime = np->n_atim; 3487 if (np->n_flag & NUPD) 3488 vattr.va_mtime = np->n_mtim; 3489 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3490 } 3491 } 3492 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3493 } 3494 3495