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 1.63 2006/07/19 06:08:13 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 nfsspec_read (struct vop_read_args *); 98 static int nfsspec_write (struct vop_write_args *); 99 static int nfsfifo_read (struct vop_read_args *); 100 static int nfsfifo_write (struct vop_write_args *); 101 static int nfsspec_close (struct vop_close_args *); 102 static int nfsfifo_close (struct vop_close_args *); 103 #define nfs_poll vop_nopoll 104 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 105 static int nfs_lookup (struct vop_old_lookup_args *); 106 static int nfs_create (struct vop_old_create_args *); 107 static int nfs_mknod (struct vop_old_mknod_args *); 108 static int nfs_open (struct vop_open_args *); 109 static int nfs_close (struct vop_close_args *); 110 static int nfs_access (struct vop_access_args *); 111 static int nfs_getattr (struct vop_getattr_args *); 112 static int nfs_setattr (struct vop_setattr_args *); 113 static int nfs_read (struct vop_read_args *); 114 static int nfs_mmap (struct vop_mmap_args *); 115 static int nfs_fsync (struct vop_fsync_args *); 116 static int nfs_remove (struct vop_old_remove_args *); 117 static int nfs_link (struct vop_old_link_args *); 118 static int nfs_rename (struct vop_old_rename_args *); 119 static int nfs_mkdir (struct vop_old_mkdir_args *); 120 static int nfs_rmdir (struct vop_old_rmdir_args *); 121 static int nfs_symlink (struct vop_old_symlink_args *); 122 static int nfs_readdir (struct vop_readdir_args *); 123 static int nfs_bmap (struct vop_bmap_args *); 124 static int nfs_strategy (struct vop_strategy_args *); 125 static int nfs_lookitup (struct vnode *, const char *, int, 126 struct ucred *, struct thread *, struct nfsnode **); 127 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 128 static int nfsspec_access (struct vop_access_args *); 129 static int nfs_readlink (struct vop_readlink_args *); 130 static int nfs_print (struct vop_print_args *); 131 static int nfs_advlock (struct vop_advlock_args *); 132 133 static int nfs_nresolve (struct vop_nresolve_args *); 134 /* 135 * Global vfs data structures for nfs 136 */ 137 struct vop_ops nfsv2_vnode_vops = { 138 .vop_default = vop_defaultop, 139 .vop_access = nfs_access, 140 .vop_advlock = nfs_advlock, 141 .vop_bmap = nfs_bmap, 142 .vop_close = nfs_close, 143 .vop_old_create = nfs_create, 144 .vop_fsync = nfs_fsync, 145 .vop_getattr = nfs_getattr, 146 .vop_getpages = nfs_getpages, 147 .vop_putpages = nfs_putpages, 148 .vop_inactive = nfs_inactive, 149 .vop_islocked = vop_stdislocked, 150 .vop_old_link = nfs_link, 151 .vop_lock = vop_stdlock, 152 .vop_old_lookup = nfs_lookup, 153 .vop_old_mkdir = nfs_mkdir, 154 .vop_old_mknod = nfs_mknod, 155 .vop_mmap = nfs_mmap, 156 .vop_open = nfs_open, 157 .vop_poll = nfs_poll, 158 .vop_print = nfs_print, 159 .vop_read = nfs_read, 160 .vop_readdir = nfs_readdir, 161 .vop_readlink = nfs_readlink, 162 .vop_reclaim = nfs_reclaim, 163 .vop_old_remove = nfs_remove, 164 .vop_old_rename = nfs_rename, 165 .vop_old_rmdir = nfs_rmdir, 166 .vop_setattr = nfs_setattr, 167 .vop_strategy = nfs_strategy, 168 .vop_old_symlink = nfs_symlink, 169 .vop_unlock = vop_stdunlock, 170 .vop_write = nfs_write, 171 .vop_nresolve = nfs_nresolve 172 }; 173 174 /* 175 * Special device vnode ops 176 */ 177 struct vop_ops nfsv2_spec_vops = { 178 .vop_default = spec_vnoperate, 179 .vop_access = nfsspec_access, 180 .vop_close = nfsspec_close, 181 .vop_fsync = nfs_fsync, 182 .vop_getattr = nfs_getattr, 183 .vop_inactive = nfs_inactive, 184 .vop_islocked = vop_stdislocked, 185 .vop_lock = vop_stdlock, 186 .vop_print = nfs_print, 187 .vop_read = nfsspec_read, 188 .vop_reclaim = nfs_reclaim, 189 .vop_setattr = nfs_setattr, 190 .vop_unlock = vop_stdunlock, 191 .vop_write = nfsspec_write 192 }; 193 194 struct vop_ops nfsv2_fifo_vops = { 195 .vop_default = fifo_vnoperate, 196 .vop_access = nfsspec_access, 197 .vop_close = nfsfifo_close, 198 .vop_fsync = nfs_fsync, 199 .vop_getattr = nfs_getattr, 200 .vop_inactive = nfs_inactive, 201 .vop_islocked = vop_stdislocked, 202 .vop_lock = vop_stdlock, 203 .vop_print = nfs_print, 204 .vop_read = nfsfifo_read, 205 .vop_reclaim = nfs_reclaim, 206 .vop_setattr = nfs_setattr, 207 .vop_unlock = vop_stdunlock, 208 .vop_write = nfsfifo_write 209 }; 210 211 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 212 struct componentname *cnp, 213 struct vattr *vap); 214 static int nfs_removerpc (struct vnode *dvp, const char *name, 215 int namelen, 216 struct ucred *cred, struct thread *td); 217 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 218 int fnamelen, struct vnode *tdvp, 219 const char *tnameptr, int tnamelen, 220 struct ucred *cred, struct thread *td); 221 static int nfs_renameit (struct vnode *sdvp, 222 struct componentname *scnp, 223 struct sillyrename *sp); 224 225 /* 226 * Global variables 227 */ 228 extern u_int32_t nfs_true, nfs_false; 229 extern u_int32_t nfs_xdrneg1; 230 extern struct nfsstats nfsstats; 231 extern nfstype nfsv3_type[9]; 232 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 233 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 234 int nfs_numasync = 0; 235 236 SYSCTL_DECL(_vfs_nfs); 237 238 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 240 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 241 242 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 243 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 244 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 245 246 static int nfsv3_commit_on_close = 0; 247 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 248 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 249 #if 0 250 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 251 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 252 253 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 254 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 255 #endif 256 257 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 258 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 259 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 260 static int 261 nfs3_access_otw(struct vnode *vp, int wmode, 262 struct thread *td, struct ucred *cred) 263 { 264 const int v3 = 1; 265 u_int32_t *tl; 266 int error = 0, attrflag; 267 268 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 269 caddr_t bpos, dpos, cp2; 270 int32_t t1, t2; 271 caddr_t cp; 272 u_int32_t rmode; 273 struct nfsnode *np = VTONFS(vp); 274 275 nfsstats.rpccnt[NFSPROC_ACCESS]++; 276 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 277 nfsm_fhtom(vp, v3); 278 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 279 *tl = txdr_unsigned(wmode); 280 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 281 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 282 if (!error) { 283 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 284 rmode = fxdr_unsigned(u_int32_t, *tl); 285 np->n_mode = rmode; 286 np->n_modeuid = cred->cr_uid; 287 np->n_modestamp = mycpu->gd_time_seconds; 288 } 289 m_freem(mrep); 290 nfsmout: 291 return error; 292 } 293 294 /* 295 * nfs access vnode op. 296 * For nfs version 2, just return ok. File accesses may fail later. 297 * For nfs version 3, use the access rpc to check accessibility. If file modes 298 * are changed on the server, accesses might still fail later. 299 * 300 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 301 * struct thread *a_td) 302 */ 303 static int 304 nfs_access(struct vop_access_args *ap) 305 { 306 struct vnode *vp = ap->a_vp; 307 thread_t td = curthread; 308 int error = 0; 309 u_int32_t mode, wmode; 310 int v3 = NFS_ISV3(vp); 311 struct nfsnode *np = VTONFS(vp); 312 313 /* 314 * Disallow write attempts on filesystems mounted read-only; 315 * unless the file is a socket, fifo, or a block or character 316 * device resident on the filesystem. 317 */ 318 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 319 switch (vp->v_type) { 320 case VREG: 321 case VDIR: 322 case VLNK: 323 return (EROFS); 324 default: 325 break; 326 } 327 } 328 /* 329 * For nfs v3, check to see if we have done this recently, and if 330 * so return our cached result instead of making an ACCESS call. 331 * If not, do an access rpc, otherwise you are stuck emulating 332 * ufs_access() locally using the vattr. This may not be correct, 333 * since the server may apply other access criteria such as 334 * client uid-->server uid mapping that we do not know about. 335 */ 336 if (v3) { 337 if (ap->a_mode & VREAD) 338 mode = NFSV3ACCESS_READ; 339 else 340 mode = 0; 341 if (vp->v_type != VDIR) { 342 if (ap->a_mode & VWRITE) 343 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 344 if (ap->a_mode & VEXEC) 345 mode |= NFSV3ACCESS_EXECUTE; 346 } else { 347 if (ap->a_mode & VWRITE) 348 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 349 NFSV3ACCESS_DELETE); 350 if (ap->a_mode & VEXEC) 351 mode |= NFSV3ACCESS_LOOKUP; 352 } 353 /* XXX safety belt, only make blanket request if caching */ 354 if (nfsaccess_cache_timeout > 0) { 355 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 356 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 357 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 358 } else { 359 wmode = mode; 360 } 361 362 /* 363 * Does our cached result allow us to give a definite yes to 364 * this request? 365 */ 366 if (np->n_modestamp && 367 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 368 (ap->a_cred->cr_uid == np->n_modeuid) && 369 ((np->n_mode & mode) == mode)) { 370 nfsstats.accesscache_hits++; 371 } else { 372 /* 373 * Either a no, or a don't know. Go to the wire. 374 */ 375 nfsstats.accesscache_misses++; 376 error = nfs3_access_otw(vp, wmode, td, ap->a_cred); 377 if (!error) { 378 if ((np->n_mode & mode) != mode) { 379 error = EACCES; 380 } 381 } 382 } 383 } else { 384 if ((error = nfsspec_access(ap)) != 0) 385 return (error); 386 387 /* 388 * Attempt to prevent a mapped root from accessing a file 389 * which it shouldn't. We try to read a byte from the file 390 * if the user is root and the file is not zero length. 391 * After calling nfsspec_access, we should have the correct 392 * file size cached. 393 */ 394 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 395 && VTONFS(vp)->n_size > 0) { 396 struct iovec aiov; 397 struct uio auio; 398 char buf[1]; 399 400 aiov.iov_base = buf; 401 aiov.iov_len = 1; 402 auio.uio_iov = &aiov; 403 auio.uio_iovcnt = 1; 404 auio.uio_offset = 0; 405 auio.uio_resid = 1; 406 auio.uio_segflg = UIO_SYSSPACE; 407 auio.uio_rw = UIO_READ; 408 auio.uio_td = td; 409 410 if (vp->v_type == VREG) { 411 error = nfs_readrpc(vp, &auio); 412 } else if (vp->v_type == VDIR) { 413 char* bp; 414 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 415 aiov.iov_base = bp; 416 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 417 error = nfs_readdirrpc(vp, &auio); 418 free(bp, M_TEMP); 419 } else if (vp->v_type == VLNK) { 420 error = nfs_readlinkrpc(vp, &auio); 421 } else { 422 error = EACCES; 423 } 424 } 425 } 426 /* 427 * [re]record creds for reading and/or writing if access 428 * was granted. Assume the NFS server will grant read access 429 * for execute requests. 430 */ 431 if (error == 0) { 432 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 433 crhold(ap->a_cred); 434 if (np->n_rucred) 435 crfree(np->n_rucred); 436 np->n_rucred = ap->a_cred; 437 } 438 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 439 crhold(ap->a_cred); 440 if (np->n_wucred) 441 crfree(np->n_wucred); 442 np->n_wucred = ap->a_cred; 443 } 444 } 445 return(error); 446 } 447 448 /* 449 * nfs open vnode op 450 * Check to see if the type is ok 451 * and that deletion is not in progress. 452 * For paged in text files, you will need to flush the page cache 453 * if consistency is lost. 454 * 455 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 456 */ 457 /* ARGSUSED */ 458 static int 459 nfs_open(struct vop_open_args *ap) 460 { 461 struct vnode *vp = ap->a_vp; 462 struct nfsnode *np = VTONFS(vp); 463 struct vattr vattr; 464 int error; 465 466 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 467 #ifdef DIAGNOSTIC 468 printf("open eacces vtyp=%d\n",vp->v_type); 469 #endif 470 return (EOPNOTSUPP); 471 } 472 473 /* 474 * Clear the attribute cache only if opening with write access. It 475 * is unclear if we should do this at all here, but we certainly 476 * should not clear the cache unconditionally simply because a file 477 * is being opened. 478 */ 479 if (ap->a_mode & FWRITE) 480 np->n_attrstamp = 0; 481 482 /* 483 * For normal NFS, reconcile changes made locally verses 484 * changes made remotely. Note that VOP_GETATTR only goes 485 * to the wire if the cached attribute has timed out or been 486 * cleared. 487 * 488 * If local modifications have been made clear the attribute 489 * cache to force an attribute and modified time check. If 490 * GETATTR detects that the file has been changed by someone 491 * other then us it will set NRMODIFIED. 492 * 493 * If we are opening a directory and local changes have been 494 * made we have to invalidate the cache in order to ensure 495 * that we get the most up-to-date information from the 496 * server. XXX 497 */ 498 if (np->n_flag & NLMODIFIED) { 499 np->n_attrstamp = 0; 500 if (vp->v_type == VDIR) { 501 error = nfs_vinvalbuf(vp, V_SAVE, 1); 502 if (error == EINTR) 503 return (error); 504 nfs_invaldir(vp); 505 } 506 } 507 error = VOP_GETATTR(vp, &vattr); 508 if (error) 509 return (error); 510 if (np->n_flag & NRMODIFIED) { 511 if (vp->v_type == VDIR) 512 nfs_invaldir(vp); 513 error = nfs_vinvalbuf(vp, V_SAVE, 1); 514 if (error == EINTR) 515 return (error); 516 np->n_flag &= ~NRMODIFIED; 517 } 518 519 return (vop_stdopen(ap)); 520 } 521 522 /* 523 * nfs close vnode op 524 * What an NFS client should do upon close after writing is a debatable issue. 525 * Most NFS clients push delayed writes to the server upon close, basically for 526 * two reasons: 527 * 1 - So that any write errors may be reported back to the client process 528 * doing the close system call. By far the two most likely errors are 529 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 530 * 2 - To put a worst case upper bound on cache inconsistency between 531 * multiple clients for the file. 532 * There is also a consistency problem for Version 2 of the protocol w.r.t. 533 * not being able to tell if other clients are writing a file concurrently, 534 * since there is no way of knowing if the changed modify time in the reply 535 * is only due to the write for this client. 536 * (NFS Version 3 provides weak cache consistency data in the reply that 537 * should be sufficient to detect and handle this case.) 538 * 539 * The current code does the following: 540 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 541 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 542 * or commit them (this satisfies 1 and 2 except for the 543 * case where the server crashes after this close but 544 * before the commit RPC, which is felt to be "good 545 * enough". Changing the last argument to nfs_flush() to 546 * a 1 would force a commit operation, if it is felt a 547 * commit is necessary now. 548 * for NQNFS - do nothing now, since 2 is dealt with via leases and 549 * 1 should be dealt with via an fsync() system call for 550 * cases where write errors are important. 551 * 552 * nfs_close(struct vnode *a_vp, int a_fflag, 553 * struct ucred *a_cred, struct thread *a_td) 554 */ 555 /* ARGSUSED */ 556 static int 557 nfs_close(struct vop_close_args *ap) 558 { 559 struct vnode *vp = ap->a_vp; 560 struct nfsnode *np = VTONFS(vp); 561 int error = 0; 562 thread_t td = curthread; 563 564 if (vp->v_type == VREG) { 565 if (np->n_flag & NLMODIFIED) { 566 if (NFS_ISV3(vp)) { 567 /* 568 * Under NFSv3 we have dirty buffers to dispose of. We 569 * must flush them to the NFS server. We have the option 570 * of waiting all the way through the commit rpc or just 571 * waiting for the initial write. The default is to only 572 * wait through the initial write so the data is in the 573 * server's cache, which is roughly similar to the state 574 * a standard disk subsystem leaves the file in on close(). 575 * 576 * We cannot clear the NLMODIFIED bit in np->n_flag due to 577 * potential races with other processes, and certainly 578 * cannot clear it if we don't commit. 579 */ 580 int cm = nfsv3_commit_on_close ? 1 : 0; 581 error = nfs_flush(vp, MNT_WAIT, td, cm); 582 /* np->n_flag &= ~NLMODIFIED; */ 583 } else { 584 error = nfs_vinvalbuf(vp, V_SAVE, 1); 585 } 586 np->n_attrstamp = 0; 587 } 588 if (np->n_flag & NWRITEERR) { 589 np->n_flag &= ~NWRITEERR; 590 error = np->n_error; 591 } 592 } 593 vop_stdclose(ap); 594 return (error); 595 } 596 597 /* 598 * nfs getattr call from vfs. 599 * 600 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 601 * struct thread *a_td) 602 */ 603 static int 604 nfs_getattr(struct vop_getattr_args *ap) 605 { 606 struct vnode *vp = ap->a_vp; 607 struct nfsnode *np = VTONFS(vp); 608 caddr_t cp; 609 u_int32_t *tl; 610 int32_t t1, t2; 611 caddr_t bpos, dpos; 612 int error = 0; 613 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 614 int v3 = NFS_ISV3(vp); 615 thread_t td = curthread; 616 617 /* 618 * Update local times for special files. 619 */ 620 if (np->n_flag & (NACC | NUPD)) 621 np->n_flag |= NCHG; 622 /* 623 * First look in the cache. 624 */ 625 if (nfs_getattrcache(vp, ap->a_vap) == 0) 626 return (0); 627 628 if (v3 && nfsaccess_cache_timeout > 0) { 629 nfsstats.accesscache_misses++; 630 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 631 if (nfs_getattrcache(vp, ap->a_vap) == 0) 632 return (0); 633 } 634 635 nfsstats.rpccnt[NFSPROC_GETATTR]++; 636 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 637 nfsm_fhtom(vp, v3); 638 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK)); 639 if (!error) { 640 nfsm_loadattr(vp, ap->a_vap); 641 } 642 m_freem(mrep); 643 nfsmout: 644 return (error); 645 } 646 647 /* 648 * nfs setattr call. 649 * 650 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 651 */ 652 static int 653 nfs_setattr(struct vop_setattr_args *ap) 654 { 655 struct vnode *vp = ap->a_vp; 656 struct nfsnode *np = VTONFS(vp); 657 struct vattr *vap = ap->a_vap; 658 int error = 0; 659 u_quad_t tsize; 660 thread_t td = curthread; 661 662 #ifndef nolint 663 tsize = (u_quad_t)0; 664 #endif 665 666 /* 667 * Setting of flags is not supported. 668 */ 669 if (vap->va_flags != VNOVAL) 670 return (EOPNOTSUPP); 671 672 /* 673 * Disallow write attempts if the filesystem is mounted read-only. 674 */ 675 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 676 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 677 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 678 (vp->v_mount->mnt_flag & MNT_RDONLY)) 679 return (EROFS); 680 if (vap->va_size != VNOVAL) { 681 switch (vp->v_type) { 682 case VDIR: 683 return (EISDIR); 684 case VCHR: 685 case VBLK: 686 case VSOCK: 687 case VFIFO: 688 if (vap->va_mtime.tv_sec == VNOVAL && 689 vap->va_atime.tv_sec == VNOVAL && 690 vap->va_mode == (mode_t)VNOVAL && 691 vap->va_uid == (uid_t)VNOVAL && 692 vap->va_gid == (gid_t)VNOVAL) 693 return (0); 694 vap->va_size = VNOVAL; 695 break; 696 default: 697 /* 698 * Disallow write attempts if the filesystem is 699 * mounted read-only. 700 */ 701 if (vp->v_mount->mnt_flag & MNT_RDONLY) 702 return (EROFS); 703 704 /* 705 * This is nasty. The RPCs we send to flush pending 706 * data often return attribute information which is 707 * cached via a callback to nfs_loadattrcache(), which 708 * has the effect of changing our notion of the file 709 * size. Due to flushed appends and other operations 710 * the file size can be set to virtually anything, 711 * including values that do not match either the old 712 * or intended file size. 713 * 714 * When this condition is detected we must loop to 715 * try the operation again. Hopefully no more 716 * flushing is required on the loop so it works the 717 * second time around. THIS CASE ALMOST ALWAYS 718 * HAPPENS! 719 */ 720 tsize = np->n_size; 721 again: 722 error = nfs_meta_setsize(vp, td, vap->va_size); 723 724 if (np->n_flag & NLMODIFIED) { 725 if (vap->va_size == 0) 726 error = nfs_vinvalbuf(vp, 0, 1); 727 else 728 error = nfs_vinvalbuf(vp, V_SAVE, 1); 729 } 730 /* 731 * note: this loop case almost always happens at 732 * least once per truncation. 733 */ 734 if (error == 0 && np->n_size != vap->va_size) 735 goto again; 736 np->n_vattr.va_size = vap->va_size; 737 break; 738 } 739 } else if ((vap->va_mtime.tv_sec != VNOVAL || 740 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 741 vp->v_type == VREG && 742 (error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR 743 ) { 744 return (error); 745 } 746 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 747 748 /* 749 * Sanity check if a truncation was issued. This should only occur 750 * if multiple processes are racing on the same file. 751 */ 752 if (error == 0 && vap->va_size != VNOVAL && 753 np->n_size != vap->va_size) { 754 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 755 goto again; 756 } 757 if (error && vap->va_size != VNOVAL) { 758 np->n_size = np->n_vattr.va_size = tsize; 759 vnode_pager_setsize(vp, np->n_size); 760 } 761 return (error); 762 } 763 764 /* 765 * Do an nfs setattr rpc. 766 */ 767 static int 768 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 769 struct ucred *cred, struct thread *td) 770 { 771 struct nfsv2_sattr *sp; 772 struct nfsnode *np = VTONFS(vp); 773 caddr_t cp; 774 int32_t t1, t2; 775 caddr_t bpos, dpos, cp2; 776 u_int32_t *tl; 777 int error = 0, wccflag = NFSV3_WCCRATTR; 778 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 779 int v3 = NFS_ISV3(vp); 780 781 nfsstats.rpccnt[NFSPROC_SETATTR]++; 782 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 783 nfsm_fhtom(vp, v3); 784 if (v3) { 785 nfsm_v3attrbuild(vap, TRUE); 786 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 787 *tl = nfs_false; 788 } else { 789 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 790 if (vap->va_mode == (mode_t)VNOVAL) 791 sp->sa_mode = nfs_xdrneg1; 792 else 793 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 794 if (vap->va_uid == (uid_t)VNOVAL) 795 sp->sa_uid = nfs_xdrneg1; 796 else 797 sp->sa_uid = txdr_unsigned(vap->va_uid); 798 if (vap->va_gid == (gid_t)VNOVAL) 799 sp->sa_gid = nfs_xdrneg1; 800 else 801 sp->sa_gid = txdr_unsigned(vap->va_gid); 802 sp->sa_size = txdr_unsigned(vap->va_size); 803 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 804 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 805 } 806 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 807 if (v3) { 808 np->n_modestamp = 0; 809 nfsm_wcc_data(vp, wccflag); 810 } else 811 nfsm_loadattr(vp, (struct vattr *)0); 812 m_freem(mrep); 813 nfsmout: 814 return (error); 815 } 816 817 /* 818 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 819 * nfs_lookup() until all remaining new api calls are implemented. 820 * 821 * Resolve a namecache entry. This function is passed a locked ncp and 822 * must call cache_setvp() on it as appropriate to resolve the entry. 823 */ 824 static int 825 nfs_nresolve(struct vop_nresolve_args *ap) 826 { 827 struct thread *td = curthread; 828 struct namecache *ncp; 829 struct ucred *cred; 830 struct nfsnode *np; 831 struct vnode *dvp; 832 struct vnode *nvp; 833 nfsfh_t *fhp; 834 int attrflag; 835 int fhsize; 836 int error; 837 int len; 838 int v3; 839 /******NFSM MACROS********/ 840 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 841 caddr_t bpos, dpos, cp, cp2; 842 u_int32_t *tl; 843 int32_t t1, t2; 844 845 cred = ap->a_cred; 846 ncp = ap->a_ncp; 847 848 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp); 849 dvp = ncp->nc_parent->nc_vp; 850 if ((error = vget(dvp, LK_SHARED)) != 0) 851 return (error); 852 853 nvp = NULL; 854 v3 = NFS_ISV3(dvp); 855 nfsstats.lookupcache_misses++; 856 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 857 len = ncp->nc_nlen; 858 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 859 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 860 nfsm_fhtom(dvp, v3); 861 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 862 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 863 if (error) { 864 /* 865 * Cache negatve lookups to reduce NFS traffic, but use 866 * a fast timeout. Otherwise use a timeout of 1 tick. 867 * XXX we should add a namecache flag for no-caching 868 * to uncache the negative hit as soon as possible, but 869 * we cannot simply destroy the entry because it is used 870 * as a placeholder by the caller. 871 */ 872 if (error == ENOENT) { 873 int nticks; 874 875 if (nfsneg_cache_timeout) 876 nticks = nfsneg_cache_timeout * hz; 877 else 878 nticks = 1; 879 cache_setvp(ncp, NULL); 880 cache_settimeout(ncp, nticks); 881 } 882 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 883 m_freem(mrep); 884 goto nfsmout; 885 } 886 887 /* 888 * Success, get the file handle, do various checks, and load 889 * post-operation data from the reply packet. Theoretically 890 * we should never be looking up "." so, theoretically, we 891 * should never get the same file handle as our directory. But 892 * we check anyway. XXX 893 * 894 * Note that no timeout is set for the positive cache hit. We 895 * assume, theoretically, that ESTALE returns will be dealt with 896 * properly to handle NFS races and in anycase we cannot depend 897 * on a timeout to deal with NFS open/create/excl issues so instead 898 * of a bad hack here the rest of the NFS client code needs to do 899 * the right thing. 900 */ 901 nfsm_getfh(fhp, fhsize, v3); 902 903 np = VTONFS(dvp); 904 if (NFS_CMPFH(np, fhp, fhsize)) { 905 vref(dvp); 906 nvp = dvp; 907 } else { 908 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 909 if (error) { 910 m_freem(mrep); 911 vput(dvp); 912 return (error); 913 } 914 nvp = NFSTOV(np); 915 } 916 if (v3) { 917 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 918 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 919 } else { 920 nfsm_loadattr(nvp, NULL); 921 } 922 cache_setvp(ncp, nvp); 923 m_freem(mrep); 924 nfsmout: 925 vput(dvp); 926 if (nvp) { 927 if (nvp == dvp) 928 vrele(nvp); 929 else 930 vput(nvp); 931 } 932 return (error); 933 } 934 935 /* 936 * 'cached' nfs directory lookup 937 * 938 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 939 * 940 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 941 * struct componentname *a_cnp) 942 */ 943 static int 944 nfs_lookup(struct vop_old_lookup_args *ap) 945 { 946 struct componentname *cnp = ap->a_cnp; 947 struct vnode *dvp = ap->a_dvp; 948 struct vnode **vpp = ap->a_vpp; 949 int flags = cnp->cn_flags; 950 struct vnode *newvp; 951 u_int32_t *tl; 952 caddr_t cp; 953 int32_t t1, t2; 954 struct nfsmount *nmp; 955 caddr_t bpos, dpos, cp2; 956 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 957 long len; 958 nfsfh_t *fhp; 959 struct nfsnode *np; 960 int lockparent, wantparent, error = 0, attrflag, fhsize; 961 int v3 = NFS_ISV3(dvp); 962 963 /* 964 * Read-only mount check and directory check. 965 */ 966 *vpp = NULLVP; 967 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 968 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 969 return (EROFS); 970 971 if (dvp->v_type != VDIR) 972 return (ENOTDIR); 973 974 /* 975 * Look it up in the cache. Note that ENOENT is only returned if we 976 * previously entered a negative hit (see later on). The additional 977 * nfsneg_cache_timeout check causes previously cached results to 978 * be instantly ignored if the negative caching is turned off. 979 */ 980 lockparent = flags & CNP_LOCKPARENT; 981 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 982 nmp = VFSTONFS(dvp->v_mount); 983 np = VTONFS(dvp); 984 985 /* 986 * Go to the wire. 987 */ 988 error = 0; 989 newvp = NULLVP; 990 nfsstats.lookupcache_misses++; 991 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 992 len = cnp->cn_namelen; 993 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 994 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 995 nfsm_fhtom(dvp, v3); 996 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 997 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 998 if (error) { 999 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1000 m_freem(mrep); 1001 goto nfsmout; 1002 } 1003 nfsm_getfh(fhp, fhsize, v3); 1004 1005 /* 1006 * Handle RENAME case... 1007 */ 1008 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1009 if (NFS_CMPFH(np, fhp, fhsize)) { 1010 m_freem(mrep); 1011 return (EISDIR); 1012 } 1013 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1014 if (error) { 1015 m_freem(mrep); 1016 return (error); 1017 } 1018 newvp = NFSTOV(np); 1019 if (v3) { 1020 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1021 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1022 } else 1023 nfsm_loadattr(newvp, (struct vattr *)0); 1024 *vpp = newvp; 1025 m_freem(mrep); 1026 if (!lockparent) { 1027 VOP_UNLOCK(dvp, 0); 1028 cnp->cn_flags |= CNP_PDIRUNLOCK; 1029 } 1030 return (0); 1031 } 1032 1033 if (flags & CNP_ISDOTDOT) { 1034 VOP_UNLOCK(dvp, 0); 1035 cnp->cn_flags |= CNP_PDIRUNLOCK; 1036 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1037 if (error) { 1038 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1039 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1040 return (error); /* NOTE: return error from nget */ 1041 } 1042 newvp = NFSTOV(np); 1043 if (lockparent) { 1044 error = vn_lock(dvp, LK_EXCLUSIVE); 1045 if (error) { 1046 vput(newvp); 1047 return (error); 1048 } 1049 cnp->cn_flags |= CNP_PDIRUNLOCK; 1050 } 1051 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1052 vref(dvp); 1053 newvp = dvp; 1054 } else { 1055 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1056 if (error) { 1057 m_freem(mrep); 1058 return (error); 1059 } 1060 if (!lockparent) { 1061 VOP_UNLOCK(dvp, 0); 1062 cnp->cn_flags |= CNP_PDIRUNLOCK; 1063 } 1064 newvp = NFSTOV(np); 1065 } 1066 if (v3) { 1067 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1068 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1069 } else 1070 nfsm_loadattr(newvp, (struct vattr *)0); 1071 #if 0 1072 /* XXX MOVE TO nfs_nremove() */ 1073 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1074 cnp->cn_nameiop != NAMEI_DELETE) { 1075 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1076 } 1077 #endif 1078 *vpp = newvp; 1079 m_freem(mrep); 1080 nfsmout: 1081 if (error) { 1082 if (newvp != NULLVP) { 1083 vrele(newvp); 1084 *vpp = NULLVP; 1085 } 1086 if ((cnp->cn_nameiop == NAMEI_CREATE || 1087 cnp->cn_nameiop == NAMEI_RENAME) && 1088 error == ENOENT) { 1089 if (!lockparent) { 1090 VOP_UNLOCK(dvp, 0); 1091 cnp->cn_flags |= CNP_PDIRUNLOCK; 1092 } 1093 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1094 error = EROFS; 1095 else 1096 error = EJUSTRETURN; 1097 } 1098 } 1099 return (error); 1100 } 1101 1102 /* 1103 * nfs read call. 1104 * Just call nfs_bioread() to do the work. 1105 * 1106 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1107 * struct ucred *a_cred) 1108 */ 1109 static int 1110 nfs_read(struct vop_read_args *ap) 1111 { 1112 struct vnode *vp = ap->a_vp; 1113 1114 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1115 switch (vp->v_type) { 1116 case VREG: 1117 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1118 case VDIR: 1119 return (EISDIR); 1120 default: 1121 return EOPNOTSUPP; 1122 } 1123 } 1124 1125 /* 1126 * nfs readlink call 1127 * 1128 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1129 */ 1130 static int 1131 nfs_readlink(struct vop_readlink_args *ap) 1132 { 1133 struct vnode *vp = ap->a_vp; 1134 1135 if (vp->v_type != VLNK) 1136 return (EINVAL); 1137 return (nfs_bioread(vp, ap->a_uio, 0)); 1138 } 1139 1140 /* 1141 * Do a readlink rpc. 1142 * Called by nfs_doio() from below the buffer cache. 1143 */ 1144 int 1145 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1146 { 1147 u_int32_t *tl; 1148 caddr_t cp; 1149 int32_t t1, t2; 1150 caddr_t bpos, dpos, cp2; 1151 int error = 0, len, attrflag; 1152 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1153 int v3 = NFS_ISV3(vp); 1154 1155 nfsstats.rpccnt[NFSPROC_READLINK]++; 1156 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1157 nfsm_fhtom(vp, v3); 1158 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1159 if (v3) 1160 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1161 if (!error) { 1162 nfsm_strsiz(len, NFS_MAXPATHLEN); 1163 if (len == NFS_MAXPATHLEN) { 1164 struct nfsnode *np = VTONFS(vp); 1165 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1166 len = np->n_size; 1167 } 1168 nfsm_mtouio(uiop, len); 1169 } 1170 m_freem(mrep); 1171 nfsmout: 1172 return (error); 1173 } 1174 1175 /* 1176 * nfs read rpc call 1177 * Ditto above 1178 */ 1179 int 1180 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1181 { 1182 u_int32_t *tl; 1183 caddr_t cp; 1184 int32_t t1, t2; 1185 caddr_t bpos, dpos, cp2; 1186 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1187 struct nfsmount *nmp; 1188 int error = 0, len, retlen, tsiz, eof, attrflag; 1189 int v3 = NFS_ISV3(vp); 1190 1191 #ifndef nolint 1192 eof = 0; 1193 #endif 1194 nmp = VFSTONFS(vp->v_mount); 1195 tsiz = uiop->uio_resid; 1196 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1197 return (EFBIG); 1198 while (tsiz > 0) { 1199 nfsstats.rpccnt[NFSPROC_READ]++; 1200 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1201 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1202 nfsm_fhtom(vp, v3); 1203 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1204 if (v3) { 1205 txdr_hyper(uiop->uio_offset, tl); 1206 *(tl + 2) = txdr_unsigned(len); 1207 } else { 1208 *tl++ = txdr_unsigned(uiop->uio_offset); 1209 *tl++ = txdr_unsigned(len); 1210 *tl = 0; 1211 } 1212 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1213 if (v3) { 1214 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1215 if (error) { 1216 m_freem(mrep); 1217 goto nfsmout; 1218 } 1219 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1220 eof = fxdr_unsigned(int, *(tl + 1)); 1221 } else 1222 nfsm_loadattr(vp, (struct vattr *)0); 1223 nfsm_strsiz(retlen, nmp->nm_rsize); 1224 nfsm_mtouio(uiop, retlen); 1225 m_freem(mrep); 1226 tsiz -= retlen; 1227 if (v3) { 1228 if (eof || retlen == 0) { 1229 tsiz = 0; 1230 } 1231 } else if (retlen < len) { 1232 tsiz = 0; 1233 } 1234 } 1235 nfsmout: 1236 return (error); 1237 } 1238 1239 /* 1240 * nfs write call 1241 */ 1242 int 1243 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1244 { 1245 u_int32_t *tl; 1246 caddr_t cp; 1247 int32_t t1, t2, backup; 1248 caddr_t bpos, dpos, cp2; 1249 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1250 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1251 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1252 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1253 1254 #ifndef DIAGNOSTIC 1255 if (uiop->uio_iovcnt != 1) 1256 panic("nfs: writerpc iovcnt > 1"); 1257 #endif 1258 *must_commit = 0; 1259 tsiz = uiop->uio_resid; 1260 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1261 return (EFBIG); 1262 while (tsiz > 0) { 1263 nfsstats.rpccnt[NFSPROC_WRITE]++; 1264 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1265 nfsm_reqhead(vp, NFSPROC_WRITE, 1266 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1267 nfsm_fhtom(vp, v3); 1268 if (v3) { 1269 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1270 txdr_hyper(uiop->uio_offset, tl); 1271 tl += 2; 1272 *tl++ = txdr_unsigned(len); 1273 *tl++ = txdr_unsigned(*iomode); 1274 *tl = txdr_unsigned(len); 1275 } else { 1276 u_int32_t x; 1277 1278 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1279 /* Set both "begin" and "current" to non-garbage. */ 1280 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1281 *tl++ = x; /* "begin offset" */ 1282 *tl++ = x; /* "current offset" */ 1283 x = txdr_unsigned(len); 1284 *tl++ = x; /* total to this offset */ 1285 *tl = x; /* size of this write */ 1286 } 1287 nfsm_uiotom(uiop, len); 1288 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1289 if (v3) { 1290 /* 1291 * The write RPC returns a before and after mtime. The 1292 * nfsm_wcc_data() macro checks the before n_mtime 1293 * against the before time and stores the after time 1294 * in the nfsnode's cached vattr and n_mtime field. 1295 * The NRMODIFIED bit will be set if the before 1296 * time did not match the original mtime. 1297 */ 1298 wccflag = NFSV3_WCCCHK; 1299 nfsm_wcc_data(vp, wccflag); 1300 if (!error) { 1301 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1302 + NFSX_V3WRITEVERF); 1303 rlen = fxdr_unsigned(int, *tl++); 1304 if (rlen == 0) { 1305 error = NFSERR_IO; 1306 m_freem(mrep); 1307 break; 1308 } else if (rlen < len) { 1309 backup = len - rlen; 1310 uiop->uio_iov->iov_base -= backup; 1311 uiop->uio_iov->iov_len += backup; 1312 uiop->uio_offset -= backup; 1313 uiop->uio_resid += backup; 1314 len = rlen; 1315 } 1316 commit = fxdr_unsigned(int, *tl++); 1317 1318 /* 1319 * Return the lowest committment level 1320 * obtained by any of the RPCs. 1321 */ 1322 if (committed == NFSV3WRITE_FILESYNC) 1323 committed = commit; 1324 else if (committed == NFSV3WRITE_DATASYNC && 1325 commit == NFSV3WRITE_UNSTABLE) 1326 committed = commit; 1327 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1328 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1329 NFSX_V3WRITEVERF); 1330 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1331 } else if (bcmp((caddr_t)tl, 1332 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1333 *must_commit = 1; 1334 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1335 NFSX_V3WRITEVERF); 1336 } 1337 } 1338 } else { 1339 nfsm_loadattr(vp, (struct vattr *)0); 1340 } 1341 m_freem(mrep); 1342 if (error) 1343 break; 1344 tsiz -= len; 1345 } 1346 nfsmout: 1347 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1348 committed = NFSV3WRITE_FILESYNC; 1349 *iomode = committed; 1350 if (error) 1351 uiop->uio_resid = tsiz; 1352 return (error); 1353 } 1354 1355 /* 1356 * nfs mknod rpc 1357 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1358 * mode set to specify the file type and the size field for rdev. 1359 */ 1360 static int 1361 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1362 struct vattr *vap) 1363 { 1364 struct nfsv2_sattr *sp; 1365 u_int32_t *tl; 1366 caddr_t cp; 1367 int32_t t1, t2; 1368 struct vnode *newvp = (struct vnode *)0; 1369 struct nfsnode *np = (struct nfsnode *)0; 1370 struct vattr vattr; 1371 char *cp2; 1372 caddr_t bpos, dpos; 1373 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1374 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1375 u_int32_t rdev; 1376 int v3 = NFS_ISV3(dvp); 1377 1378 if (vap->va_type == VCHR || vap->va_type == VBLK) 1379 rdev = txdr_unsigned(vap->va_rdev); 1380 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1381 rdev = nfs_xdrneg1; 1382 else { 1383 return (EOPNOTSUPP); 1384 } 1385 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1386 return (error); 1387 } 1388 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1389 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1390 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1391 nfsm_fhtom(dvp, v3); 1392 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1393 if (v3) { 1394 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1395 *tl++ = vtonfsv3_type(vap->va_type); 1396 nfsm_v3attrbuild(vap, FALSE); 1397 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1398 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1399 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1400 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1401 } 1402 } else { 1403 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1404 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1405 sp->sa_uid = nfs_xdrneg1; 1406 sp->sa_gid = nfs_xdrneg1; 1407 sp->sa_size = rdev; 1408 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1409 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1410 } 1411 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1412 if (!error) { 1413 nfsm_mtofh(dvp, newvp, v3, gotvp); 1414 if (!gotvp) { 1415 if (newvp) { 1416 vput(newvp); 1417 newvp = (struct vnode *)0; 1418 } 1419 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1420 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1421 if (!error) 1422 newvp = NFSTOV(np); 1423 } 1424 } 1425 if (v3) 1426 nfsm_wcc_data(dvp, wccflag); 1427 m_freem(mrep); 1428 nfsmout: 1429 if (error) { 1430 if (newvp) 1431 vput(newvp); 1432 } else { 1433 *vpp = newvp; 1434 } 1435 VTONFS(dvp)->n_flag |= NLMODIFIED; 1436 if (!wccflag) 1437 VTONFS(dvp)->n_attrstamp = 0; 1438 return (error); 1439 } 1440 1441 /* 1442 * nfs mknod vop 1443 * just call nfs_mknodrpc() to do the work. 1444 * 1445 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1446 * struct componentname *a_cnp, struct vattr *a_vap) 1447 */ 1448 /* ARGSUSED */ 1449 static int 1450 nfs_mknod(struct vop_old_mknod_args *ap) 1451 { 1452 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1453 } 1454 1455 static u_long create_verf; 1456 /* 1457 * nfs file create call 1458 * 1459 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1460 * struct componentname *a_cnp, struct vattr *a_vap) 1461 */ 1462 static int 1463 nfs_create(struct vop_old_create_args *ap) 1464 { 1465 struct vnode *dvp = ap->a_dvp; 1466 struct vattr *vap = ap->a_vap; 1467 struct componentname *cnp = ap->a_cnp; 1468 struct nfsv2_sattr *sp; 1469 u_int32_t *tl; 1470 caddr_t cp; 1471 int32_t t1, t2; 1472 struct nfsnode *np = (struct nfsnode *)0; 1473 struct vnode *newvp = (struct vnode *)0; 1474 caddr_t bpos, dpos, cp2; 1475 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1476 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1477 struct vattr vattr; 1478 int v3 = NFS_ISV3(dvp); 1479 1480 /* 1481 * Oops, not for me.. 1482 */ 1483 if (vap->va_type == VSOCK) 1484 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1485 1486 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1487 return (error); 1488 } 1489 if (vap->va_vaflags & VA_EXCLUSIVE) 1490 fmode |= O_EXCL; 1491 again: 1492 nfsstats.rpccnt[NFSPROC_CREATE]++; 1493 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1494 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1495 nfsm_fhtom(dvp, v3); 1496 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1497 if (v3) { 1498 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1499 if (fmode & O_EXCL) { 1500 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1501 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1502 #ifdef INET 1503 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1504 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1505 else 1506 #endif 1507 *tl++ = create_verf; 1508 *tl = ++create_verf; 1509 } else { 1510 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1511 nfsm_v3attrbuild(vap, FALSE); 1512 } 1513 } else { 1514 nfsm_build(sp, struct nfsv2_sattr *, 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 = 0; 1519 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1520 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1521 } 1522 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1523 if (!error) { 1524 nfsm_mtofh(dvp, newvp, v3, gotvp); 1525 if (!gotvp) { 1526 if (newvp) { 1527 vput(newvp); 1528 newvp = (struct vnode *)0; 1529 } 1530 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1531 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1532 if (!error) 1533 newvp = NFSTOV(np); 1534 } 1535 } 1536 if (v3) 1537 nfsm_wcc_data(dvp, wccflag); 1538 m_freem(mrep); 1539 nfsmout: 1540 if (error) { 1541 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1542 fmode &= ~O_EXCL; 1543 goto again; 1544 } 1545 if (newvp) 1546 vput(newvp); 1547 } else if (v3 && (fmode & O_EXCL)) { 1548 /* 1549 * We are normally called with only a partially initialized 1550 * VAP. Since the NFSv3 spec says that server may use the 1551 * file attributes to store the verifier, the spec requires 1552 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1553 * in atime, but we can't really assume that all servers will 1554 * so we ensure that our SETATTR sets both atime and mtime. 1555 */ 1556 if (vap->va_mtime.tv_sec == VNOVAL) 1557 vfs_timestamp(&vap->va_mtime); 1558 if (vap->va_atime.tv_sec == VNOVAL) 1559 vap->va_atime = vap->va_mtime; 1560 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1561 } 1562 if (!error) { 1563 /* 1564 * The new np may have enough info for access 1565 * checks, make sure rucred and wucred are 1566 * initialized for read and write rpc's. 1567 */ 1568 np = VTONFS(newvp); 1569 if (np->n_rucred == NULL) 1570 np->n_rucred = crhold(cnp->cn_cred); 1571 if (np->n_wucred == NULL) 1572 np->n_wucred = crhold(cnp->cn_cred); 1573 *ap->a_vpp = newvp; 1574 } 1575 VTONFS(dvp)->n_flag |= NLMODIFIED; 1576 if (!wccflag) 1577 VTONFS(dvp)->n_attrstamp = 0; 1578 return (error); 1579 } 1580 1581 /* 1582 * nfs file remove call 1583 * To try and make nfs semantics closer to ufs semantics, a file that has 1584 * other processes using the vnode is renamed instead of removed and then 1585 * removed later on the last close. 1586 * - If v_usecount > 1 1587 * If a rename is not already in the works 1588 * call nfs_sillyrename() to set it up 1589 * else 1590 * do the remove rpc 1591 * 1592 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1593 * struct componentname *a_cnp) 1594 */ 1595 static int 1596 nfs_remove(struct vop_old_remove_args *ap) 1597 { 1598 struct vnode *vp = ap->a_vp; 1599 struct vnode *dvp = ap->a_dvp; 1600 struct componentname *cnp = ap->a_cnp; 1601 struct nfsnode *np = VTONFS(vp); 1602 int error = 0; 1603 struct vattr vattr; 1604 1605 #ifndef DIAGNOSTIC 1606 if (vp->v_usecount < 1) 1607 panic("nfs_remove: bad v_usecount"); 1608 #endif 1609 if (vp->v_type == VDIR) 1610 error = EPERM; 1611 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1612 VOP_GETATTR(vp, &vattr) == 0 && 1613 vattr.va_nlink > 1)) { 1614 /* 1615 * throw away biocache buffers, mainly to avoid 1616 * unnecessary delayed writes later. 1617 */ 1618 error = nfs_vinvalbuf(vp, 0, 1); 1619 /* Do the rpc */ 1620 if (error != EINTR) 1621 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1622 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1623 /* 1624 * Kludge City: If the first reply to the remove rpc is lost.. 1625 * the reply to the retransmitted request will be ENOENT 1626 * since the file was in fact removed 1627 * Therefore, we cheat and return success. 1628 */ 1629 if (error == ENOENT) 1630 error = 0; 1631 } else if (!np->n_sillyrename) { 1632 error = nfs_sillyrename(dvp, vp, cnp); 1633 } 1634 np->n_attrstamp = 0; 1635 return (error); 1636 } 1637 1638 /* 1639 * nfs file remove rpc called from nfs_inactive 1640 */ 1641 int 1642 nfs_removeit(struct sillyrename *sp) 1643 { 1644 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1645 sp->s_cred, NULL)); 1646 } 1647 1648 /* 1649 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1650 */ 1651 static int 1652 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1653 struct ucred *cred, struct thread *td) 1654 { 1655 u_int32_t *tl; 1656 caddr_t cp; 1657 int32_t t1, t2; 1658 caddr_t bpos, dpos, cp2; 1659 int error = 0, wccflag = NFSV3_WCCRATTR; 1660 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1661 int v3 = NFS_ISV3(dvp); 1662 1663 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1664 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1665 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1666 nfsm_fhtom(dvp, v3); 1667 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1668 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1669 if (v3) 1670 nfsm_wcc_data(dvp, wccflag); 1671 m_freem(mrep); 1672 nfsmout: 1673 VTONFS(dvp)->n_flag |= NLMODIFIED; 1674 if (!wccflag) 1675 VTONFS(dvp)->n_attrstamp = 0; 1676 return (error); 1677 } 1678 1679 /* 1680 * nfs file rename call 1681 * 1682 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1683 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1684 * struct vnode *a_tvp, struct componentname *a_tcnp) 1685 */ 1686 static int 1687 nfs_rename(struct vop_old_rename_args *ap) 1688 { 1689 struct vnode *fvp = ap->a_fvp; 1690 struct vnode *tvp = ap->a_tvp; 1691 struct vnode *fdvp = ap->a_fdvp; 1692 struct vnode *tdvp = ap->a_tdvp; 1693 struct componentname *tcnp = ap->a_tcnp; 1694 struct componentname *fcnp = ap->a_fcnp; 1695 int error; 1696 1697 /* Check for cross-device rename */ 1698 if ((fvp->v_mount != tdvp->v_mount) || 1699 (tvp && (fvp->v_mount != tvp->v_mount))) { 1700 error = EXDEV; 1701 goto out; 1702 } 1703 1704 /* 1705 * We have to flush B_DELWRI data prior to renaming 1706 * the file. If we don't, the delayed-write buffers 1707 * can be flushed out later after the file has gone stale 1708 * under NFSV3. NFSV2 does not have this problem because 1709 * ( as far as I can tell ) it flushes dirty buffers more 1710 * often. 1711 */ 1712 1713 VOP_FSYNC(fvp, MNT_WAIT); 1714 if (tvp) 1715 VOP_FSYNC(tvp, MNT_WAIT); 1716 1717 /* 1718 * If the tvp exists and is in use, sillyrename it before doing the 1719 * rename of the new file over it. 1720 * 1721 * XXX Can't sillyrename a directory. 1722 * 1723 * We do not attempt to do any namecache purges in this old API 1724 * routine. The new API compat functions have access to the actual 1725 * namecache structures and will do it for us. 1726 */ 1727 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1728 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1729 vput(tvp); 1730 tvp = NULL; 1731 } else if (tvp) { 1732 ; 1733 } 1734 1735 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1736 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1737 tcnp->cn_td); 1738 1739 out: 1740 if (tdvp == tvp) 1741 vrele(tdvp); 1742 else 1743 vput(tdvp); 1744 if (tvp) 1745 vput(tvp); 1746 vrele(fdvp); 1747 vrele(fvp); 1748 /* 1749 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1750 */ 1751 if (error == ENOENT) 1752 error = 0; 1753 return (error); 1754 } 1755 1756 /* 1757 * nfs file rename rpc called from nfs_remove() above 1758 */ 1759 static int 1760 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1761 struct sillyrename *sp) 1762 { 1763 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1764 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1765 } 1766 1767 /* 1768 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1769 */ 1770 static int 1771 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1772 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1773 struct ucred *cred, struct thread *td) 1774 { 1775 u_int32_t *tl; 1776 caddr_t cp; 1777 int32_t t1, t2; 1778 caddr_t bpos, dpos, cp2; 1779 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1780 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1781 int v3 = NFS_ISV3(fdvp); 1782 1783 nfsstats.rpccnt[NFSPROC_RENAME]++; 1784 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1785 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1786 nfsm_rndup(tnamelen)); 1787 nfsm_fhtom(fdvp, v3); 1788 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1789 nfsm_fhtom(tdvp, v3); 1790 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1791 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1792 if (v3) { 1793 nfsm_wcc_data(fdvp, fwccflag); 1794 nfsm_wcc_data(tdvp, twccflag); 1795 } 1796 m_freem(mrep); 1797 nfsmout: 1798 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1799 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1800 if (!fwccflag) 1801 VTONFS(fdvp)->n_attrstamp = 0; 1802 if (!twccflag) 1803 VTONFS(tdvp)->n_attrstamp = 0; 1804 return (error); 1805 } 1806 1807 /* 1808 * nfs hard link create call 1809 * 1810 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1811 * struct componentname *a_cnp) 1812 */ 1813 static int 1814 nfs_link(struct vop_old_link_args *ap) 1815 { 1816 struct vnode *vp = ap->a_vp; 1817 struct vnode *tdvp = ap->a_tdvp; 1818 struct componentname *cnp = ap->a_cnp; 1819 u_int32_t *tl; 1820 caddr_t cp; 1821 int32_t t1, t2; 1822 caddr_t bpos, dpos, cp2; 1823 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1824 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1825 int v3; 1826 1827 if (vp->v_mount != tdvp->v_mount) { 1828 return (EXDEV); 1829 } 1830 1831 /* 1832 * Push all writes to the server, so that the attribute cache 1833 * doesn't get "out of sync" with the server. 1834 * XXX There should be a better way! 1835 */ 1836 VOP_FSYNC(vp, MNT_WAIT); 1837 1838 v3 = NFS_ISV3(vp); 1839 nfsstats.rpccnt[NFSPROC_LINK]++; 1840 nfsm_reqhead(vp, NFSPROC_LINK, 1841 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1842 nfsm_fhtom(vp, v3); 1843 nfsm_fhtom(tdvp, v3); 1844 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1845 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1846 if (v3) { 1847 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1848 nfsm_wcc_data(tdvp, wccflag); 1849 } 1850 m_freem(mrep); 1851 nfsmout: 1852 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1853 if (!attrflag) 1854 VTONFS(vp)->n_attrstamp = 0; 1855 if (!wccflag) 1856 VTONFS(tdvp)->n_attrstamp = 0; 1857 /* 1858 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1859 */ 1860 if (error == EEXIST) 1861 error = 0; 1862 return (error); 1863 } 1864 1865 /* 1866 * nfs symbolic link create call 1867 * 1868 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1869 * struct componentname *a_cnp, struct vattr *a_vap, 1870 * char *a_target) 1871 */ 1872 static int 1873 nfs_symlink(struct vop_old_symlink_args *ap) 1874 { 1875 struct vnode *dvp = ap->a_dvp; 1876 struct vattr *vap = ap->a_vap; 1877 struct componentname *cnp = ap->a_cnp; 1878 struct nfsv2_sattr *sp; 1879 u_int32_t *tl; 1880 caddr_t cp; 1881 int32_t t1, t2; 1882 caddr_t bpos, dpos, cp2; 1883 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1884 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1885 struct vnode *newvp = (struct vnode *)0; 1886 int v3 = NFS_ISV3(dvp); 1887 1888 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1889 slen = strlen(ap->a_target); 1890 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1891 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1892 nfsm_fhtom(dvp, v3); 1893 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1894 if (v3) { 1895 nfsm_v3attrbuild(vap, FALSE); 1896 } 1897 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1898 if (!v3) { 1899 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1900 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1901 sp->sa_uid = nfs_xdrneg1; 1902 sp->sa_gid = nfs_xdrneg1; 1903 sp->sa_size = nfs_xdrneg1; 1904 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1905 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1906 } 1907 1908 /* 1909 * Issue the NFS request and get the rpc response. 1910 * 1911 * Only NFSv3 responses returning an error of 0 actually return 1912 * a file handle that can be converted into newvp without having 1913 * to do an extra lookup rpc. 1914 */ 1915 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1916 if (v3) { 1917 if (error == 0) 1918 nfsm_mtofh(dvp, newvp, v3, gotvp); 1919 nfsm_wcc_data(dvp, wccflag); 1920 } 1921 1922 /* 1923 * out code jumps -> here, mrep is also freed. 1924 */ 1925 1926 m_freem(mrep); 1927 nfsmout: 1928 1929 /* 1930 * If we get an EEXIST error, silently convert it to no-error 1931 * in case of an NFS retry. 1932 */ 1933 if (error == EEXIST) 1934 error = 0; 1935 1936 /* 1937 * If we do not have (or no longer have) an error, and we could 1938 * not extract the newvp from the response due to the request being 1939 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1940 * to obtain a newvp to return. 1941 */ 1942 if (error == 0 && newvp == NULL) { 1943 struct nfsnode *np = NULL; 1944 1945 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1946 cnp->cn_cred, cnp->cn_td, &np); 1947 if (!error) 1948 newvp = NFSTOV(np); 1949 } 1950 if (error) { 1951 if (newvp) 1952 vput(newvp); 1953 } else { 1954 *ap->a_vpp = newvp; 1955 } 1956 VTONFS(dvp)->n_flag |= NLMODIFIED; 1957 if (!wccflag) 1958 VTONFS(dvp)->n_attrstamp = 0; 1959 return (error); 1960 } 1961 1962 /* 1963 * nfs make dir call 1964 * 1965 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1966 * struct componentname *a_cnp, struct vattr *a_vap) 1967 */ 1968 static int 1969 nfs_mkdir(struct vop_old_mkdir_args *ap) 1970 { 1971 struct vnode *dvp = ap->a_dvp; 1972 struct vattr *vap = ap->a_vap; 1973 struct componentname *cnp = ap->a_cnp; 1974 struct nfsv2_sattr *sp; 1975 u_int32_t *tl; 1976 caddr_t cp; 1977 int32_t t1, t2; 1978 int len; 1979 struct nfsnode *np = (struct nfsnode *)0; 1980 struct vnode *newvp = (struct vnode *)0; 1981 caddr_t bpos, dpos, cp2; 1982 int error = 0, wccflag = NFSV3_WCCRATTR; 1983 int gotvp = 0; 1984 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1985 struct vattr vattr; 1986 int v3 = NFS_ISV3(dvp); 1987 1988 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1989 return (error); 1990 } 1991 len = cnp->cn_namelen; 1992 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1993 nfsm_reqhead(dvp, NFSPROC_MKDIR, 1994 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 1995 nfsm_fhtom(dvp, v3); 1996 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1997 if (v3) { 1998 nfsm_v3attrbuild(vap, FALSE); 1999 } else { 2000 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2001 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2002 sp->sa_uid = nfs_xdrneg1; 2003 sp->sa_gid = nfs_xdrneg1; 2004 sp->sa_size = nfs_xdrneg1; 2005 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2006 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2007 } 2008 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2009 if (!error) 2010 nfsm_mtofh(dvp, newvp, v3, gotvp); 2011 if (v3) 2012 nfsm_wcc_data(dvp, wccflag); 2013 m_freem(mrep); 2014 nfsmout: 2015 VTONFS(dvp)->n_flag |= NLMODIFIED; 2016 if (!wccflag) 2017 VTONFS(dvp)->n_attrstamp = 0; 2018 /* 2019 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2020 * if we can succeed in looking up the directory. 2021 */ 2022 if (error == EEXIST || (!error && !gotvp)) { 2023 if (newvp) { 2024 vrele(newvp); 2025 newvp = (struct vnode *)0; 2026 } 2027 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2028 cnp->cn_td, &np); 2029 if (!error) { 2030 newvp = NFSTOV(np); 2031 if (newvp->v_type != VDIR) 2032 error = EEXIST; 2033 } 2034 } 2035 if (error) { 2036 if (newvp) 2037 vrele(newvp); 2038 } else 2039 *ap->a_vpp = newvp; 2040 return (error); 2041 } 2042 2043 /* 2044 * nfs remove directory call 2045 * 2046 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2047 * struct componentname *a_cnp) 2048 */ 2049 static int 2050 nfs_rmdir(struct vop_old_rmdir_args *ap) 2051 { 2052 struct vnode *vp = ap->a_vp; 2053 struct vnode *dvp = ap->a_dvp; 2054 struct componentname *cnp = ap->a_cnp; 2055 u_int32_t *tl; 2056 caddr_t cp; 2057 int32_t t1, t2; 2058 caddr_t bpos, dpos, cp2; 2059 int error = 0, wccflag = NFSV3_WCCRATTR; 2060 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2061 int v3 = NFS_ISV3(dvp); 2062 2063 if (dvp == vp) 2064 return (EINVAL); 2065 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2066 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2067 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2068 nfsm_fhtom(dvp, v3); 2069 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2070 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2071 if (v3) 2072 nfsm_wcc_data(dvp, wccflag); 2073 m_freem(mrep); 2074 nfsmout: 2075 VTONFS(dvp)->n_flag |= NLMODIFIED; 2076 if (!wccflag) 2077 VTONFS(dvp)->n_attrstamp = 0; 2078 /* 2079 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2080 */ 2081 if (error == ENOENT) 2082 error = 0; 2083 return (error); 2084 } 2085 2086 /* 2087 * nfs readdir call 2088 * 2089 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2090 */ 2091 static int 2092 nfs_readdir(struct vop_readdir_args *ap) 2093 { 2094 struct vnode *vp = ap->a_vp; 2095 struct nfsnode *np = VTONFS(vp); 2096 struct uio *uio = ap->a_uio; 2097 int tresid, error; 2098 struct vattr vattr; 2099 2100 if (vp->v_type != VDIR) 2101 return (EPERM); 2102 2103 /* 2104 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2105 * and then check that is still valid, or if this is an NQNFS mount 2106 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2107 * VOP_GETATTR() does not necessarily go to the wire. 2108 */ 2109 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2110 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2111 if (VOP_GETATTR(vp, &vattr) == 0 && 2112 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2113 ) { 2114 nfsstats.direofcache_hits++; 2115 return (0); 2116 } 2117 } 2118 2119 /* 2120 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2121 * own cache coherency checks so we do not have to. 2122 */ 2123 tresid = uio->uio_resid; 2124 error = nfs_bioread(vp, uio, 0); 2125 2126 if (!error && uio->uio_resid == tresid) 2127 nfsstats.direofcache_misses++; 2128 return (error); 2129 } 2130 2131 /* 2132 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2133 * 2134 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2135 * offset/block and converts the nfs formatted directory entries for userland 2136 * consumption as well as deals with offsets into the middle of blocks. 2137 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2138 * be block-bounded. It must convert to cookies for the actual RPC. 2139 */ 2140 int 2141 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2142 { 2143 int len, left; 2144 struct nfs_dirent *dp = NULL; 2145 u_int32_t *tl; 2146 caddr_t cp; 2147 int32_t t1, t2; 2148 nfsuint64 *cookiep; 2149 caddr_t bpos, dpos, cp2; 2150 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2151 nfsuint64 cookie; 2152 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2153 struct nfsnode *dnp = VTONFS(vp); 2154 u_quad_t fileno; 2155 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2156 int attrflag; 2157 int v3 = NFS_ISV3(vp); 2158 2159 #ifndef DIAGNOSTIC 2160 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2161 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2162 panic("nfs readdirrpc bad uio"); 2163 #endif 2164 2165 /* 2166 * If there is no cookie, assume directory was stale. 2167 */ 2168 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2169 if (cookiep) 2170 cookie = *cookiep; 2171 else 2172 return (NFSERR_BAD_COOKIE); 2173 /* 2174 * Loop around doing readdir rpc's of size nm_readdirsize 2175 * truncated to a multiple of DIRBLKSIZ. 2176 * The stopping criteria is EOF or buffer full. 2177 */ 2178 while (more_dirs && bigenough) { 2179 nfsstats.rpccnt[NFSPROC_READDIR]++; 2180 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2181 NFSX_READDIR(v3)); 2182 nfsm_fhtom(vp, v3); 2183 if (v3) { 2184 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2185 *tl++ = cookie.nfsuquad[0]; 2186 *tl++ = cookie.nfsuquad[1]; 2187 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2188 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2189 } else { 2190 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2191 *tl++ = cookie.nfsuquad[0]; 2192 } 2193 *tl = txdr_unsigned(nmp->nm_readdirsize); 2194 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2195 if (v3) { 2196 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2197 if (!error) { 2198 nfsm_dissect(tl, u_int32_t *, 2199 2 * NFSX_UNSIGNED); 2200 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2201 dnp->n_cookieverf.nfsuquad[1] = *tl; 2202 } else { 2203 m_freem(mrep); 2204 goto nfsmout; 2205 } 2206 } 2207 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2208 more_dirs = fxdr_unsigned(int, *tl); 2209 2210 /* loop thru the dir entries, converting them to std form */ 2211 while (more_dirs && bigenough) { 2212 if (v3) { 2213 nfsm_dissect(tl, u_int32_t *, 2214 3 * NFSX_UNSIGNED); 2215 fileno = fxdr_hyper(tl); 2216 len = fxdr_unsigned(int, *(tl + 2)); 2217 } else { 2218 nfsm_dissect(tl, u_int32_t *, 2219 2 * NFSX_UNSIGNED); 2220 fileno = fxdr_unsigned(u_quad_t, *tl++); 2221 len = fxdr_unsigned(int, *tl); 2222 } 2223 if (len <= 0 || len > NFS_MAXNAMLEN) { 2224 error = EBADRPC; 2225 m_freem(mrep); 2226 goto nfsmout; 2227 } 2228 2229 /* 2230 * len is the number of bytes in the path element 2231 * name, not including the \0 termination. 2232 * 2233 * tlen is the number of bytes w have to reserve for 2234 * the path element name. 2235 */ 2236 tlen = nfsm_rndup(len); 2237 if (tlen == len) 2238 tlen += 4; /* To ensure null termination */ 2239 2240 /* 2241 * If the entry would cross a DIRBLKSIZ boundary, 2242 * extend the previous nfs_dirent to cover the 2243 * remaining space. 2244 */ 2245 left = DIRBLKSIZ - blksiz; 2246 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2247 dp->nfs_reclen += left; 2248 uiop->uio_iov->iov_base += left; 2249 uiop->uio_iov->iov_len -= left; 2250 uiop->uio_offset += left; 2251 uiop->uio_resid -= left; 2252 blksiz = 0; 2253 } 2254 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2255 bigenough = 0; 2256 if (bigenough) { 2257 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2258 dp->nfs_ino = fileno; 2259 dp->nfs_namlen = len; 2260 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2261 dp->nfs_type = DT_UNKNOWN; 2262 blksiz += dp->nfs_reclen; 2263 if (blksiz == DIRBLKSIZ) 2264 blksiz = 0; 2265 uiop->uio_offset += sizeof(struct nfs_dirent); 2266 uiop->uio_resid -= sizeof(struct nfs_dirent); 2267 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2268 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2269 nfsm_mtouio(uiop, len); 2270 2271 /* 2272 * The uiop has advanced by nfs_dirent + len 2273 * but really needs to advance by 2274 * nfs_dirent + tlen 2275 */ 2276 cp = uiop->uio_iov->iov_base; 2277 tlen -= len; 2278 *cp = '\0'; /* null terminate */ 2279 uiop->uio_iov->iov_base += tlen; 2280 uiop->uio_iov->iov_len -= tlen; 2281 uiop->uio_offset += tlen; 2282 uiop->uio_resid -= tlen; 2283 } else { 2284 /* 2285 * NFS strings must be rounded up (nfsm_myouio 2286 * handled that in the bigenough case). 2287 */ 2288 nfsm_adv(nfsm_rndup(len)); 2289 } 2290 if (v3) { 2291 nfsm_dissect(tl, u_int32_t *, 2292 3 * NFSX_UNSIGNED); 2293 } else { 2294 nfsm_dissect(tl, u_int32_t *, 2295 2 * NFSX_UNSIGNED); 2296 } 2297 2298 /* 2299 * If we were able to accomodate the last entry, 2300 * get the cookie for the next one. Otherwise 2301 * hold-over the cookie for the one we were not 2302 * able to accomodate. 2303 */ 2304 if (bigenough) { 2305 cookie.nfsuquad[0] = *tl++; 2306 if (v3) 2307 cookie.nfsuquad[1] = *tl++; 2308 } else if (v3) { 2309 tl += 2; 2310 } else { 2311 tl++; 2312 } 2313 more_dirs = fxdr_unsigned(int, *tl); 2314 } 2315 /* 2316 * If at end of rpc data, get the eof boolean 2317 */ 2318 if (!more_dirs) { 2319 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2320 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2321 } 2322 m_freem(mrep); 2323 } 2324 /* 2325 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2326 * by increasing d_reclen for the last record. 2327 */ 2328 if (blksiz > 0) { 2329 left = DIRBLKSIZ - blksiz; 2330 dp->nfs_reclen += left; 2331 uiop->uio_iov->iov_base += left; 2332 uiop->uio_iov->iov_len -= left; 2333 uiop->uio_offset += left; 2334 uiop->uio_resid -= left; 2335 } 2336 2337 if (bigenough) { 2338 /* 2339 * We hit the end of the directory, update direofoffset. 2340 */ 2341 dnp->n_direofoffset = uiop->uio_offset; 2342 } else { 2343 /* 2344 * There is more to go, insert the link cookie so the 2345 * next block can be read. 2346 */ 2347 if (uiop->uio_resid > 0) 2348 printf("EEK! readdirrpc resid > 0\n"); 2349 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2350 *cookiep = cookie; 2351 } 2352 nfsmout: 2353 return (error); 2354 } 2355 2356 /* 2357 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2358 */ 2359 int 2360 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2361 { 2362 int len, left; 2363 struct nfs_dirent *dp; 2364 u_int32_t *tl; 2365 caddr_t cp; 2366 int32_t t1, t2; 2367 struct vnode *newvp; 2368 nfsuint64 *cookiep; 2369 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2370 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2371 nfsuint64 cookie; 2372 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2373 struct nfsnode *dnp = VTONFS(vp), *np; 2374 nfsfh_t *fhp; 2375 u_quad_t fileno; 2376 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2377 int attrflag, fhsize; 2378 struct namecache *ncp; 2379 struct namecache *dncp; 2380 struct nlcomponent nlc; 2381 2382 #ifndef nolint 2383 dp = NULL; 2384 #endif 2385 #ifndef DIAGNOSTIC 2386 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2387 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2388 panic("nfs readdirplusrpc bad uio"); 2389 #endif 2390 /* 2391 * Obtain the namecache record for the directory so we have something 2392 * to use as a basis for creating the entries. This function will 2393 * return a held (but not locked) ncp. The ncp may be disconnected 2394 * from the tree and cannot be used for upward traversals, and the 2395 * ncp may be unnamed. Note that other unrelated operations may 2396 * cause the ncp to be named at any time. 2397 */ 2398 dncp = cache_fromdvp(vp, NULL, 0); 2399 bzero(&nlc, sizeof(nlc)); 2400 newvp = NULLVP; 2401 2402 /* 2403 * If there is no cookie, assume directory was stale. 2404 */ 2405 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2406 if (cookiep) 2407 cookie = *cookiep; 2408 else 2409 return (NFSERR_BAD_COOKIE); 2410 /* 2411 * Loop around doing readdir rpc's of size nm_readdirsize 2412 * truncated to a multiple of DIRBLKSIZ. 2413 * The stopping criteria is EOF or buffer full. 2414 */ 2415 while (more_dirs && bigenough) { 2416 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2417 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2418 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2419 nfsm_fhtom(vp, 1); 2420 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2421 *tl++ = cookie.nfsuquad[0]; 2422 *tl++ = cookie.nfsuquad[1]; 2423 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2424 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2425 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2426 *tl = txdr_unsigned(nmp->nm_rsize); 2427 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2428 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2429 if (error) { 2430 m_freem(mrep); 2431 goto nfsmout; 2432 } 2433 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2434 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2435 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2436 more_dirs = fxdr_unsigned(int, *tl); 2437 2438 /* loop thru the dir entries, doctoring them to 4bsd form */ 2439 while (more_dirs && bigenough) { 2440 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2441 fileno = fxdr_hyper(tl); 2442 len = fxdr_unsigned(int, *(tl + 2)); 2443 if (len <= 0 || len > NFS_MAXNAMLEN) { 2444 error = EBADRPC; 2445 m_freem(mrep); 2446 goto nfsmout; 2447 } 2448 tlen = nfsm_rndup(len); 2449 if (tlen == len) 2450 tlen += 4; /* To ensure null termination*/ 2451 left = DIRBLKSIZ - blksiz; 2452 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2453 dp->nfs_reclen += left; 2454 uiop->uio_iov->iov_base += left; 2455 uiop->uio_iov->iov_len -= left; 2456 uiop->uio_offset += left; 2457 uiop->uio_resid -= left; 2458 blksiz = 0; 2459 } 2460 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2461 bigenough = 0; 2462 if (bigenough) { 2463 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2464 dp->nfs_ino = fileno; 2465 dp->nfs_namlen = len; 2466 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2467 dp->nfs_type = DT_UNKNOWN; 2468 blksiz += dp->nfs_reclen; 2469 if (blksiz == DIRBLKSIZ) 2470 blksiz = 0; 2471 uiop->uio_offset += sizeof(struct nfs_dirent); 2472 uiop->uio_resid -= sizeof(struct nfs_dirent); 2473 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2474 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2475 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2476 nlc.nlc_namelen = len; 2477 nfsm_mtouio(uiop, len); 2478 cp = uiop->uio_iov->iov_base; 2479 tlen -= len; 2480 *cp = '\0'; 2481 uiop->uio_iov->iov_base += tlen; 2482 uiop->uio_iov->iov_len -= tlen; 2483 uiop->uio_offset += tlen; 2484 uiop->uio_resid -= tlen; 2485 } else 2486 nfsm_adv(nfsm_rndup(len)); 2487 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2488 if (bigenough) { 2489 cookie.nfsuquad[0] = *tl++; 2490 cookie.nfsuquad[1] = *tl++; 2491 } else 2492 tl += 2; 2493 2494 /* 2495 * Since the attributes are before the file handle 2496 * (sigh), we must skip over the attributes and then 2497 * come back and get them. 2498 */ 2499 attrflag = fxdr_unsigned(int, *tl); 2500 if (attrflag) { 2501 dpossav1 = dpos; 2502 mdsav1 = md; 2503 nfsm_adv(NFSX_V3FATTR); 2504 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2505 doit = fxdr_unsigned(int, *tl); 2506 if (doit) { 2507 nfsm_getfh(fhp, fhsize, 1); 2508 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2509 vref(vp); 2510 newvp = vp; 2511 np = dnp; 2512 } else { 2513 error = nfs_nget(vp->v_mount, fhp, 2514 fhsize, &np); 2515 if (error) 2516 doit = 0; 2517 else 2518 newvp = NFSTOV(np); 2519 } 2520 } 2521 if (doit && bigenough) { 2522 dpossav2 = dpos; 2523 dpos = dpossav1; 2524 mdsav2 = md; 2525 md = mdsav1; 2526 nfsm_loadattr(newvp, (struct vattr *)0); 2527 dpos = dpossav2; 2528 md = mdsav2; 2529 dp->nfs_type = 2530 IFTODT(VTTOIF(np->n_vattr.va_type)); 2531 if (dncp) { 2532 printf("NFS/READDIRPLUS, ENTER %*.*s\n", 2533 nlc.nlc_namelen, nlc.nlc_namelen, 2534 nlc.nlc_nameptr); 2535 ncp = cache_nlookup(dncp, &nlc); 2536 cache_setunresolved(ncp); 2537 cache_setvp(ncp, newvp); 2538 cache_put(ncp); 2539 } else { 2540 printf("NFS/READDIRPLUS, UNABLE TO ENTER" 2541 " %*.*s\n", 2542 nlc.nlc_namelen, nlc.nlc_namelen, 2543 nlc.nlc_nameptr); 2544 } 2545 } 2546 } else { 2547 /* Just skip over the file handle */ 2548 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2549 i = fxdr_unsigned(int, *tl); 2550 nfsm_adv(nfsm_rndup(i)); 2551 } 2552 if (newvp != NULLVP) { 2553 if (newvp == vp) 2554 vrele(newvp); 2555 else 2556 vput(newvp); 2557 newvp = NULLVP; 2558 } 2559 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2560 more_dirs = fxdr_unsigned(int, *tl); 2561 } 2562 /* 2563 * If at end of rpc data, get the eof boolean 2564 */ 2565 if (!more_dirs) { 2566 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2567 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2568 } 2569 m_freem(mrep); 2570 } 2571 /* 2572 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2573 * by increasing d_reclen for the last record. 2574 */ 2575 if (blksiz > 0) { 2576 left = DIRBLKSIZ - blksiz; 2577 dp->nfs_reclen += left; 2578 uiop->uio_iov->iov_base += left; 2579 uiop->uio_iov->iov_len -= left; 2580 uiop->uio_offset += left; 2581 uiop->uio_resid -= left; 2582 } 2583 2584 /* 2585 * We are now either at the end of the directory or have filled the 2586 * block. 2587 */ 2588 if (bigenough) 2589 dnp->n_direofoffset = uiop->uio_offset; 2590 else { 2591 if (uiop->uio_resid > 0) 2592 printf("EEK! readdirplusrpc resid > 0\n"); 2593 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2594 *cookiep = cookie; 2595 } 2596 nfsmout: 2597 if (newvp != NULLVP) { 2598 if (newvp == vp) 2599 vrele(newvp); 2600 else 2601 vput(newvp); 2602 newvp = NULLVP; 2603 } 2604 if (dncp) 2605 cache_drop(dncp); 2606 return (error); 2607 } 2608 2609 /* 2610 * Silly rename. To make the NFS filesystem that is stateless look a little 2611 * more like the "ufs" a remove of an active vnode is translated to a rename 2612 * to a funny looking filename that is removed by nfs_inactive on the 2613 * nfsnode. There is the potential for another process on a different client 2614 * to create the same funny name between the nfs_lookitup() fails and the 2615 * nfs_rename() completes, but... 2616 */ 2617 static int 2618 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2619 { 2620 struct sillyrename *sp; 2621 struct nfsnode *np; 2622 int error; 2623 2624 /* 2625 * We previously purged dvp instead of vp. I don't know why, it 2626 * completely destroys performance. We can't do it anyway with the 2627 * new VFS API since we would be breaking the namecache topology. 2628 */ 2629 cache_purge(vp); /* XXX */ 2630 np = VTONFS(vp); 2631 #ifndef DIAGNOSTIC 2632 if (vp->v_type == VDIR) 2633 panic("nfs: sillyrename dir"); 2634 #endif 2635 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2636 M_NFSREQ, M_WAITOK); 2637 sp->s_cred = crdup(cnp->cn_cred); 2638 sp->s_dvp = dvp; 2639 vref(dvp); 2640 2641 /* Fudge together a funny name */ 2642 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2643 2644 /* Try lookitups until we get one that isn't there */ 2645 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2646 cnp->cn_td, (struct nfsnode **)0) == 0) { 2647 sp->s_name[4]++; 2648 if (sp->s_name[4] > 'z') { 2649 error = EINVAL; 2650 goto bad; 2651 } 2652 } 2653 error = nfs_renameit(dvp, cnp, sp); 2654 if (error) 2655 goto bad; 2656 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2657 cnp->cn_td, &np); 2658 np->n_sillyrename = sp; 2659 return (0); 2660 bad: 2661 vrele(sp->s_dvp); 2662 crfree(sp->s_cred); 2663 free((caddr_t)sp, M_NFSREQ); 2664 return (error); 2665 } 2666 2667 /* 2668 * Look up a file name and optionally either update the file handle or 2669 * allocate an nfsnode, depending on the value of npp. 2670 * npp == NULL --> just do the lookup 2671 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2672 * handled too 2673 * *npp != NULL --> update the file handle in the vnode 2674 */ 2675 static int 2676 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2677 struct thread *td, struct nfsnode **npp) 2678 { 2679 u_int32_t *tl; 2680 caddr_t cp; 2681 int32_t t1, t2; 2682 struct vnode *newvp = (struct vnode *)0; 2683 struct nfsnode *np, *dnp = VTONFS(dvp); 2684 caddr_t bpos, dpos, cp2; 2685 int error = 0, fhlen, attrflag; 2686 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2687 nfsfh_t *nfhp; 2688 int v3 = NFS_ISV3(dvp); 2689 2690 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2691 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2692 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2693 nfsm_fhtom(dvp, v3); 2694 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2695 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2696 if (npp && !error) { 2697 nfsm_getfh(nfhp, fhlen, v3); 2698 if (*npp) { 2699 np = *npp; 2700 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2701 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2702 np->n_fhp = &np->n_fh; 2703 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2704 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2705 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2706 np->n_fhsize = fhlen; 2707 newvp = NFSTOV(np); 2708 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2709 vref(dvp); 2710 newvp = dvp; 2711 } else { 2712 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2713 if (error) { 2714 m_freem(mrep); 2715 return (error); 2716 } 2717 newvp = NFSTOV(np); 2718 } 2719 if (v3) { 2720 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2721 if (!attrflag && *npp == NULL) { 2722 m_freem(mrep); 2723 if (newvp == dvp) 2724 vrele(newvp); 2725 else 2726 vput(newvp); 2727 return (ENOENT); 2728 } 2729 } else 2730 nfsm_loadattr(newvp, (struct vattr *)0); 2731 } 2732 m_freem(mrep); 2733 nfsmout: 2734 if (npp && *npp == NULL) { 2735 if (error) { 2736 if (newvp) { 2737 if (newvp == dvp) 2738 vrele(newvp); 2739 else 2740 vput(newvp); 2741 } 2742 } else 2743 *npp = np; 2744 } 2745 return (error); 2746 } 2747 2748 /* 2749 * Nfs Version 3 commit rpc 2750 */ 2751 int 2752 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2753 { 2754 caddr_t cp; 2755 u_int32_t *tl; 2756 int32_t t1, t2; 2757 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2758 caddr_t bpos, dpos, cp2; 2759 int error = 0, wccflag = NFSV3_WCCRATTR; 2760 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2761 2762 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2763 return (0); 2764 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2765 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2766 nfsm_fhtom(vp, 1); 2767 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2768 txdr_hyper(offset, tl); 2769 tl += 2; 2770 *tl = txdr_unsigned(cnt); 2771 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2772 nfsm_wcc_data(vp, wccflag); 2773 if (!error) { 2774 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2775 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2776 NFSX_V3WRITEVERF)) { 2777 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2778 NFSX_V3WRITEVERF); 2779 error = NFSERR_STALEWRITEVERF; 2780 } 2781 } 2782 m_freem(mrep); 2783 nfsmout: 2784 return (error); 2785 } 2786 2787 /* 2788 * Kludge City.. 2789 * - make nfs_bmap() essentially a no-op that does no translation 2790 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2791 * (Maybe I could use the process's page mapping, but I was concerned that 2792 * Kernel Write might not be enabled and also figured copyout() would do 2793 * a lot more work than bcopy() and also it currently happens in the 2794 * context of the swapper process (2). 2795 * 2796 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp, 2797 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2798 */ 2799 static int 2800 nfs_bmap(struct vop_bmap_args *ap) 2801 { 2802 struct vnode *vp = ap->a_vp; 2803 2804 if (ap->a_vpp != NULL) 2805 *ap->a_vpp = vp; 2806 if (ap->a_doffsetp != NULL) 2807 *ap->a_doffsetp = ap->a_loffset; 2808 if (ap->a_runp != NULL) 2809 *ap->a_runp = 0; 2810 if (ap->a_runb != NULL) 2811 *ap->a_runb = 0; 2812 return (0); 2813 } 2814 2815 /* 2816 * Strategy routine. 2817 * 2818 * For async requests when nfsiod(s) are running, queue the request by 2819 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2820 * request. 2821 */ 2822 static int 2823 nfs_strategy(struct vop_strategy_args *ap) 2824 { 2825 struct bio *bio = ap->a_bio; 2826 struct bio *nbio; 2827 struct buf *bp = bio->bio_buf; 2828 struct thread *td; 2829 int error = 0; 2830 2831 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2832 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2833 KASSERT(BUF_REFCNT(bp) > 0, 2834 ("nfs_strategy: buffer %p not locked", bp)); 2835 2836 if (bp->b_flags & B_ASYNC) 2837 td = NULL; 2838 else 2839 td = curthread; /* XXX */ 2840 2841 /* 2842 * We probably don't need to push an nbio any more since no 2843 * block conversion is required due to the use of 64 bit byte 2844 * offsets, but do it anyway. 2845 */ 2846 nbio = push_bio(bio); 2847 nbio->bio_offset = bio->bio_offset; 2848 2849 /* 2850 * If the op is asynchronous and an i/o daemon is waiting 2851 * queue the request, wake it up and wait for completion 2852 * otherwise just do it ourselves. 2853 */ 2854 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td)) 2855 error = nfs_doio(ap->a_vp, nbio, td); 2856 return (error); 2857 } 2858 2859 /* 2860 * Mmap a file 2861 * 2862 * NB Currently unsupported. 2863 * 2864 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2865 * struct thread *a_td) 2866 */ 2867 /* ARGSUSED */ 2868 static int 2869 nfs_mmap(struct vop_mmap_args *ap) 2870 { 2871 return (EINVAL); 2872 } 2873 2874 /* 2875 * fsync vnode op. Just call nfs_flush() with commit == 1. 2876 * 2877 * nfs_fsync(struct vnode *a_vp, struct ucred * a_cred, int a_waitfor, 2878 * struct thread *a_td) 2879 */ 2880 /* ARGSUSED */ 2881 static int 2882 nfs_fsync(struct vop_fsync_args *ap) 2883 { 2884 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1)); 2885 } 2886 2887 /* 2888 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 2889 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 2890 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 2891 * set the buffer contains data that has already been written to the server 2892 * and which now needs a commit RPC. 2893 * 2894 * If commit is 0 we only take one pass and only flush buffers containing new 2895 * dirty data. 2896 * 2897 * If commit is 1 we take two passes, issuing a commit RPC in the second 2898 * pass. 2899 * 2900 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 2901 * to completely flush all pending data. 2902 * 2903 * Note that the RB_SCAN code properly handles the case where the 2904 * callback might block and directly or indirectly (another thread) cause 2905 * the RB tree to change. 2906 */ 2907 2908 #ifndef NFS_COMMITBVECSIZ 2909 #define NFS_COMMITBVECSIZ 16 2910 #endif 2911 2912 struct nfs_flush_info { 2913 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 2914 struct thread *td; 2915 struct vnode *vp; 2916 int waitfor; 2917 int slpflag; 2918 int slptimeo; 2919 int loops; 2920 struct buf *bvary[NFS_COMMITBVECSIZ]; 2921 int bvsize; 2922 off_t beg_off; 2923 off_t end_off; 2924 }; 2925 2926 static int nfs_flush_bp(struct buf *bp, void *data); 2927 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 2928 2929 int 2930 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2931 { 2932 struct nfsnode *np = VTONFS(vp); 2933 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2934 struct nfs_flush_info info; 2935 int error; 2936 2937 bzero(&info, sizeof(info)); 2938 info.td = td; 2939 info.vp = vp; 2940 info.waitfor = waitfor; 2941 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 2942 info.loops = 0; 2943 2944 do { 2945 /* 2946 * Flush mode 2947 */ 2948 info.mode = NFI_FLUSHNEW; 2949 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2950 nfs_flush_bp, &info); 2951 2952 /* 2953 * Take a second pass if committing and no error occured. 2954 * Clean up any left over collection (whether an error 2955 * occurs or not). 2956 */ 2957 if (commit && error == 0) { 2958 info.mode = NFI_COMMIT; 2959 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2960 nfs_flush_bp, &info); 2961 if (info.bvsize) 2962 error = nfs_flush_docommit(&info, error); 2963 } 2964 2965 /* 2966 * Wait for pending I/O to complete before checking whether 2967 * any further dirty buffers exist. 2968 */ 2969 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) { 2970 vp->v_track_write.bk_waitflag = 1; 2971 error = tsleep(&vp->v_track_write, 2972 info.slpflag, "nfsfsync", info.slptimeo); 2973 if (error) { 2974 /* 2975 * We have to be able to break out if this 2976 * is an 'intr' mount. 2977 */ 2978 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2979 error = -EINTR; 2980 break; 2981 } 2982 2983 /* 2984 * Since we do not process pending signals, 2985 * once we get a PCATCH our tsleep() will no 2986 * longer sleep, switch to a fixed timeout 2987 * instead. 2988 */ 2989 if (info.slpflag == PCATCH) { 2990 info.slpflag = 0; 2991 info.slptimeo = 2 * hz; 2992 } 2993 error = 0; 2994 } 2995 } 2996 ++info.loops; 2997 /* 2998 * Loop if we are flushing synchronous as well as committing, 2999 * and dirty buffers are still present. Otherwise we might livelock. 3000 */ 3001 } while (waitfor == MNT_WAIT && commit && 3002 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3003 3004 /* 3005 * The callbacks have to return a negative error to terminate the 3006 * RB scan. 3007 */ 3008 if (error < 0) 3009 error = -error; 3010 3011 /* 3012 * Deal with any error collection 3013 */ 3014 if (np->n_flag & NWRITEERR) { 3015 error = np->n_error; 3016 np->n_flag &= ~NWRITEERR; 3017 } 3018 return (error); 3019 } 3020 3021 3022 static 3023 int 3024 nfs_flush_bp(struct buf *bp, void *data) 3025 { 3026 struct nfs_flush_info *info = data; 3027 off_t toff; 3028 int error; 3029 3030 error = 0; 3031 switch(info->mode) { 3032 case NFI_FLUSHNEW: 3033 crit_enter(); 3034 if (info->loops && info->waitfor == MNT_WAIT) { 3035 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3036 if (error) { 3037 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3038 if (info->slpflag & PCATCH) 3039 lkflags |= LK_PCATCH; 3040 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3041 info->slptimeo); 3042 } 3043 } else { 3044 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3045 } 3046 if (error == 0) { 3047 KKASSERT(bp->b_vp == info->vp); 3048 3049 if ((bp->b_flags & B_DELWRI) == 0) 3050 panic("nfs_fsync: not dirty"); 3051 if (bp->b_flags & B_NEEDCOMMIT) { 3052 BUF_UNLOCK(bp); 3053 crit_exit(); 3054 break; 3055 } 3056 bremfree(bp); 3057 3058 bp->b_flags |= B_ASYNC; 3059 crit_exit(); 3060 bwrite(bp); 3061 } else { 3062 crit_exit(); 3063 error = 0; 3064 } 3065 break; 3066 case NFI_COMMIT: 3067 /* 3068 * Only process buffers in need of a commit which we can 3069 * immediately lock. This may prevent a buffer from being 3070 * committed, but the normal flush loop will block on the 3071 * same buffer so we shouldn't get into an endless loop. 3072 */ 3073 crit_enter(); 3074 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3075 (B_DELWRI | B_NEEDCOMMIT) || 3076 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 3077 crit_exit(); 3078 break; 3079 } 3080 3081 KKASSERT(bp->b_vp == info->vp); 3082 bremfree(bp); 3083 3084 /* 3085 * NOTE: storing the bp in the bvary[] basically sets 3086 * it up for a commit operation. 3087 * 3088 * We must call vfs_busy_pages() now so the commit operation 3089 * is interlocked with user modifications to memory mapped 3090 * pages. 3091 * 3092 * Note: to avoid loopback deadlocks, we do not 3093 * assign b_runningbufspace. 3094 */ 3095 bp->b_cmd = BUF_CMD_WRITE; 3096 vfs_busy_pages(bp->b_vp, bp); 3097 info->bvary[info->bvsize] = bp; 3098 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3099 if (info->bvsize == 0 || toff < info->beg_off) 3100 info->beg_off = toff; 3101 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3102 if (info->bvsize == 0 || toff > info->end_off) 3103 info->end_off = toff; 3104 ++info->bvsize; 3105 if (info->bvsize == NFS_COMMITBVECSIZ) { 3106 error = nfs_flush_docommit(info, 0); 3107 KKASSERT(info->bvsize == 0); 3108 } 3109 crit_exit(); 3110 } 3111 return (error); 3112 } 3113 3114 static 3115 int 3116 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3117 { 3118 struct vnode *vp; 3119 struct buf *bp; 3120 off_t bytes; 3121 int retv; 3122 int i; 3123 3124 vp = info->vp; 3125 3126 if (info->bvsize > 0) { 3127 /* 3128 * Commit data on the server, as required. Note that 3129 * nfs_commit will use the vnode's cred for the commit. 3130 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3131 */ 3132 bytes = info->end_off - info->beg_off; 3133 if (bytes > 0x40000000) 3134 bytes = 0x40000000; 3135 if (error) { 3136 retv = -error; 3137 } else { 3138 retv = nfs_commit(vp, info->beg_off, 3139 (int)bytes, info->td); 3140 if (retv == NFSERR_STALEWRITEVERF) 3141 nfs_clearcommit(vp->v_mount); 3142 } 3143 3144 /* 3145 * Now, either mark the blocks I/O done or mark the 3146 * blocks dirty, depending on whether the commit 3147 * succeeded. 3148 */ 3149 for (i = 0; i < info->bvsize; ++i) { 3150 bp = info->bvary[i]; 3151 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3152 if (retv) { 3153 /* 3154 * Error, leave B_DELWRI intact 3155 */ 3156 vfs_unbusy_pages(bp); 3157 bp->b_cmd = BUF_CMD_DONE; 3158 brelse(bp); 3159 } else { 3160 /* 3161 * Success, remove B_DELWRI ( bundirty() ). 3162 * 3163 * b_dirtyoff/b_dirtyend seem to be NFS 3164 * specific. We should probably move that 3165 * into bundirty(). XXX 3166 * 3167 * We are faking an I/O write, we have to 3168 * start the transaction in order to 3169 * immediately biodone() it. 3170 */ 3171 crit_enter(); 3172 bp->b_flags |= B_ASYNC; 3173 bundirty(bp); 3174 bp->b_flags &= ~B_ERROR; 3175 bp->b_dirtyoff = bp->b_dirtyend = 0; 3176 crit_exit(); 3177 biodone(&bp->b_bio1); 3178 } 3179 } 3180 info->bvsize = 0; 3181 } 3182 return (error); 3183 } 3184 3185 /* 3186 * NFS advisory byte-level locks. 3187 * Currently unsupported. 3188 * 3189 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3190 * int a_flags) 3191 */ 3192 static int 3193 nfs_advlock(struct vop_advlock_args *ap) 3194 { 3195 struct nfsnode *np = VTONFS(ap->a_vp); 3196 3197 /* 3198 * The following kludge is to allow diskless support to work 3199 * until a real NFS lockd is implemented. Basically, just pretend 3200 * that this is a local lock. 3201 */ 3202 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3203 } 3204 3205 /* 3206 * Print out the contents of an nfsnode. 3207 * 3208 * nfs_print(struct vnode *a_vp) 3209 */ 3210 static int 3211 nfs_print(struct vop_print_args *ap) 3212 { 3213 struct vnode *vp = ap->a_vp; 3214 struct nfsnode *np = VTONFS(vp); 3215 3216 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3217 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3218 if (vp->v_type == VFIFO) 3219 fifo_printinfo(vp); 3220 printf("\n"); 3221 return (0); 3222 } 3223 3224 /* 3225 * nfs special file access vnode op. 3226 * Essentially just get vattr and then imitate iaccess() since the device is 3227 * local to the client. 3228 * 3229 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3230 * struct thread *a_td) 3231 */ 3232 static int 3233 nfsspec_access(struct vop_access_args *ap) 3234 { 3235 struct vattr *vap; 3236 gid_t *gp; 3237 struct ucred *cred = ap->a_cred; 3238 struct vnode *vp = ap->a_vp; 3239 mode_t mode = ap->a_mode; 3240 struct vattr vattr; 3241 int i; 3242 int error; 3243 3244 /* 3245 * Disallow write attempts on filesystems mounted read-only; 3246 * unless the file is a socket, fifo, or a block or character 3247 * device resident on the filesystem. 3248 */ 3249 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3250 switch (vp->v_type) { 3251 case VREG: 3252 case VDIR: 3253 case VLNK: 3254 return (EROFS); 3255 default: 3256 break; 3257 } 3258 } 3259 /* 3260 * If you're the super-user, 3261 * you always get access. 3262 */ 3263 if (cred->cr_uid == 0) 3264 return (0); 3265 vap = &vattr; 3266 error = VOP_GETATTR(vp, vap); 3267 if (error) 3268 return (error); 3269 /* 3270 * Access check is based on only one of owner, group, public. 3271 * If not owner, then check group. If not a member of the 3272 * group, then check public access. 3273 */ 3274 if (cred->cr_uid != vap->va_uid) { 3275 mode >>= 3; 3276 gp = cred->cr_groups; 3277 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3278 if (vap->va_gid == *gp) 3279 goto found; 3280 mode >>= 3; 3281 found: 3282 ; 3283 } 3284 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3285 return (error); 3286 } 3287 3288 /* 3289 * Read wrapper for special devices. 3290 * 3291 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3292 * struct ucred *a_cred) 3293 */ 3294 static int 3295 nfsspec_read(struct vop_read_args *ap) 3296 { 3297 struct nfsnode *np = VTONFS(ap->a_vp); 3298 3299 /* 3300 * Set access flag. 3301 */ 3302 np->n_flag |= NACC; 3303 getnanotime(&np->n_atim); 3304 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3305 } 3306 3307 /* 3308 * Write wrapper for special devices. 3309 * 3310 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3311 * struct ucred *a_cred) 3312 */ 3313 static int 3314 nfsspec_write(struct vop_write_args *ap) 3315 { 3316 struct nfsnode *np = VTONFS(ap->a_vp); 3317 3318 /* 3319 * Set update flag. 3320 */ 3321 np->n_flag |= NUPD; 3322 getnanotime(&np->n_mtim); 3323 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3324 } 3325 3326 /* 3327 * Close wrapper for special devices. 3328 * 3329 * Update the times on the nfsnode then do device close. 3330 * 3331 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3332 * struct thread *a_td) 3333 */ 3334 static int 3335 nfsspec_close(struct vop_close_args *ap) 3336 { 3337 struct vnode *vp = ap->a_vp; 3338 struct nfsnode *np = VTONFS(vp); 3339 struct vattr vattr; 3340 3341 if (np->n_flag & (NACC | NUPD)) { 3342 np->n_flag |= NCHG; 3343 if (vp->v_usecount == 1 && 3344 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3345 VATTR_NULL(&vattr); 3346 if (np->n_flag & NACC) 3347 vattr.va_atime = np->n_atim; 3348 if (np->n_flag & NUPD) 3349 vattr.va_mtime = np->n_mtim; 3350 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3351 } 3352 } 3353 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3354 } 3355 3356 /* 3357 * Read wrapper for fifos. 3358 * 3359 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3360 * struct ucred *a_cred) 3361 */ 3362 static int 3363 nfsfifo_read(struct vop_read_args *ap) 3364 { 3365 struct nfsnode *np = VTONFS(ap->a_vp); 3366 3367 /* 3368 * Set access flag. 3369 */ 3370 np->n_flag |= NACC; 3371 getnanotime(&np->n_atim); 3372 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3373 } 3374 3375 /* 3376 * Write wrapper for fifos. 3377 * 3378 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3379 * struct ucred *a_cred) 3380 */ 3381 static int 3382 nfsfifo_write(struct vop_write_args *ap) 3383 { 3384 struct nfsnode *np = VTONFS(ap->a_vp); 3385 3386 /* 3387 * Set update flag. 3388 */ 3389 np->n_flag |= NUPD; 3390 getnanotime(&np->n_mtim); 3391 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3392 } 3393 3394 /* 3395 * Close wrapper for fifos. 3396 * 3397 * Update the times on the nfsnode then do fifo close. 3398 * 3399 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3400 */ 3401 static int 3402 nfsfifo_close(struct vop_close_args *ap) 3403 { 3404 struct vnode *vp = ap->a_vp; 3405 struct nfsnode *np = VTONFS(vp); 3406 struct vattr vattr; 3407 struct timespec ts; 3408 3409 if (np->n_flag & (NACC | NUPD)) { 3410 getnanotime(&ts); 3411 if (np->n_flag & NACC) 3412 np->n_atim = ts; 3413 if (np->n_flag & NUPD) 3414 np->n_mtim = ts; 3415 np->n_flag |= NCHG; 3416 if (vp->v_usecount == 1 && 3417 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3418 VATTR_NULL(&vattr); 3419 if (np->n_flag & NACC) 3420 vattr.va_atime = np->n_atim; 3421 if (np->n_flag & NUPD) 3422 vattr.va_mtime = np->n_mtim; 3423 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3424 } 3425 } 3426 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3427 } 3428 3429