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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 33 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $ 34 */ 35 36 37 /* 38 * vnode op calls for Sun NFS version 2 and 3 39 */ 40 41 #include "opt_inet.h" 42 43 #include <sys/param.h> 44 #include <sys/kernel.h> 45 #include <sys/systm.h> 46 #include <sys/uio.h> 47 #include <sys/resourcevar.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/buf.h> 51 #include <sys/malloc.h> 52 #include <sys/mbuf.h> 53 #include <sys/namei.h> 54 #include <sys/nlookup.h> 55 #include <sys/socket.h> 56 #include <sys/vnode.h> 57 #include <sys/dirent.h> 58 #include <sys/fcntl.h> 59 #include <sys/lockf.h> 60 #include <sys/stat.h> 61 #include <sys/sysctl.h> 62 #include <sys/conf.h> 63 64 #include <vm/vm.h> 65 #include <vm/vm_extern.h> 66 67 #include <sys/buf2.h> 68 69 #include <vfs/fifofs/fifo.h> 70 #include <vfs/ufs/dir.h> 71 72 #undef DIRBLKSIZ 73 74 #include "rpcv2.h" 75 #include "nfsproto.h" 76 #include "nfs.h" 77 #include "nfsmount.h" 78 #include "nfsnode.h" 79 #include "xdr_subs.h" 80 #include "nfsm_subs.h" 81 82 #include <net/if.h> 83 #include <netinet/in.h> 84 #include <netinet/in_var.h> 85 86 /* Defs */ 87 #define TRUE 1 88 #define FALSE 0 89 90 static int nfsfifo_read (struct vop_read_args *); 91 static int nfsfifo_write (struct vop_write_args *); 92 static int nfsfifo_close (struct vop_close_args *); 93 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 94 static int nfs_lookup (struct vop_old_lookup_args *); 95 static int nfs_create (struct vop_old_create_args *); 96 static int nfs_mknod (struct vop_old_mknod_args *); 97 static int nfs_open (struct vop_open_args *); 98 static int nfs_close (struct vop_close_args *); 99 static int nfs_access (struct vop_access_args *); 100 static int nfs_getattr (struct vop_getattr_args *); 101 static int nfs_setattr (struct vop_setattr_args *); 102 static int nfs_read (struct vop_read_args *); 103 static int nfs_fsync (struct vop_fsync_args *); 104 static int nfs_remove (struct vop_old_remove_args *); 105 static int nfs_link (struct vop_old_link_args *); 106 static int nfs_rename (struct vop_old_rename_args *); 107 static int nfs_mkdir (struct vop_old_mkdir_args *); 108 static int nfs_rmdir (struct vop_old_rmdir_args *); 109 static int nfs_symlink (struct vop_old_symlink_args *); 110 static int nfs_readdir (struct vop_readdir_args *); 111 static int nfs_bmap (struct vop_bmap_args *); 112 static int nfs_strategy (struct vop_strategy_args *); 113 static int nfs_lookitup (struct vnode *, const char *, int, 114 struct ucred *, struct thread *, struct nfsnode **); 115 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 116 static int nfs_laccess (struct vop_access_args *); 117 static int nfs_readlink (struct vop_readlink_args *); 118 static int nfs_print (struct vop_print_args *); 119 static int nfs_advlock (struct vop_advlock_args *); 120 static int nfs_kqfilter (struct vop_kqfilter_args *ap); 121 122 static int nfs_nresolve (struct vop_nresolve_args *); 123 /* 124 * Global vfs data structures for nfs 125 */ 126 struct vop_ops nfsv2_vnode_vops = { 127 .vop_default = vop_defaultop, 128 .vop_access = nfs_access, 129 .vop_advlock = nfs_advlock, 130 .vop_bmap = nfs_bmap, 131 .vop_close = nfs_close, 132 .vop_old_create = nfs_create, 133 .vop_fsync = nfs_fsync, 134 .vop_getattr = nfs_getattr, 135 .vop_getpages = vop_stdgetpages, 136 .vop_putpages = vop_stdputpages, 137 .vop_inactive = nfs_inactive, 138 .vop_old_link = nfs_link, 139 .vop_old_lookup = nfs_lookup, 140 .vop_old_mkdir = nfs_mkdir, 141 .vop_old_mknod = nfs_mknod, 142 .vop_open = nfs_open, 143 .vop_print = nfs_print, 144 .vop_read = nfs_read, 145 .vop_readdir = nfs_readdir, 146 .vop_readlink = nfs_readlink, 147 .vop_reclaim = nfs_reclaim, 148 .vop_old_remove = nfs_remove, 149 .vop_old_rename = nfs_rename, 150 .vop_old_rmdir = nfs_rmdir, 151 .vop_setattr = nfs_setattr, 152 .vop_strategy = nfs_strategy, 153 .vop_old_symlink = nfs_symlink, 154 .vop_write = nfs_write, 155 .vop_nresolve = nfs_nresolve, 156 .vop_kqfilter = nfs_kqfilter 157 }; 158 159 /* 160 * Special device vnode ops 161 */ 162 struct vop_ops nfsv2_spec_vops = { 163 .vop_default = vop_defaultop, 164 .vop_access = nfs_laccess, 165 .vop_close = nfs_close, 166 .vop_fsync = nfs_fsync, 167 .vop_getattr = nfs_getattr, 168 .vop_inactive = nfs_inactive, 169 .vop_print = nfs_print, 170 .vop_read = vop_stdnoread, 171 .vop_reclaim = nfs_reclaim, 172 .vop_setattr = nfs_setattr, 173 .vop_write = vop_stdnowrite 174 }; 175 176 struct vop_ops nfsv2_fifo_vops = { 177 .vop_default = fifo_vnoperate, 178 .vop_access = nfs_laccess, 179 .vop_close = nfsfifo_close, 180 .vop_fsync = nfs_fsync, 181 .vop_getattr = nfs_getattr, 182 .vop_inactive = nfs_inactive, 183 .vop_print = nfs_print, 184 .vop_read = nfsfifo_read, 185 .vop_reclaim = nfs_reclaim, 186 .vop_setattr = nfs_setattr, 187 .vop_write = nfsfifo_write 188 }; 189 190 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 191 struct componentname *cnp, 192 struct vattr *vap); 193 static int nfs_removerpc (struct vnode *dvp, const char *name, 194 int namelen, 195 struct ucred *cred, struct thread *td); 196 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 197 int fnamelen, struct vnode *tdvp, 198 const char *tnameptr, int tnamelen, 199 struct ucred *cred, struct thread *td); 200 static int nfs_renameit (struct vnode *sdvp, 201 struct componentname *scnp, 202 struct sillyrename *sp); 203 204 SYSCTL_DECL(_vfs_nfs); 205 206 static int nfs_flush_on_rename = 1; 207 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW, 208 &nfs_flush_on_rename, 0, "flush fvp prior to rename"); 209 static int nfs_flush_on_hlink = 0; 210 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW, 211 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link"); 212 213 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 214 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 215 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 216 217 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 218 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 219 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout"); 220 221 static int nfspos_cache_timeout = NFS_MINATTRTIMO; 222 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW, 223 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout"); 224 225 static int nfsv3_commit_on_close = 0; 226 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 227 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 228 #if 0 229 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 230 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 231 232 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 233 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 234 #endif 235 236 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 237 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 238 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 239 240 static __inline 241 void 242 nfs_knote(struct vnode *vp, int flags) 243 { 244 if (flags) 245 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 246 } 247 248 /* 249 * Returns whether a name component is a degenerate '.' or '..'. 250 */ 251 static __inline 252 int 253 nlcdegenerate(struct nlcomponent *nlc) 254 { 255 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.') 256 return(1); 257 if (nlc->nlc_namelen == 2 && 258 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.') 259 return(1); 260 return(0); 261 } 262 263 static int 264 nfs3_access_otw(struct vnode *vp, int wmode, 265 struct thread *td, struct ucred *cred) 266 { 267 struct nfsnode *np = VTONFS(vp); 268 int attrflag; 269 int error = 0; 270 u_int32_t *tl; 271 u_int32_t rmode; 272 struct nfsm_info info; 273 274 info.mrep = NULL; 275 info.v3 = 1; 276 277 nfsstats.rpccnt[NFSPROC_ACCESS]++; 278 nfsm_reqhead(&info, vp, NFSPROC_ACCESS, 279 NFSX_FH(info.v3) + NFSX_UNSIGNED); 280 ERROROUT(nfsm_fhtom(&info, vp)); 281 tl = nfsm_build(&info, NFSX_UNSIGNED); 282 *tl = txdr_unsigned(wmode); 283 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error)); 284 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK)); 285 if (error == 0) { 286 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 287 rmode = fxdr_unsigned(u_int32_t, *tl); 288 np->n_mode = rmode; 289 np->n_modeuid = cred->cr_uid; 290 np->n_modestamp = mycpu->gd_time_seconds; 291 } 292 m_freem(info.mrep); 293 info.mrep = NULL; 294 nfsmout: 295 return error; 296 } 297 298 /* 299 * nfs access vnode op. 300 * For nfs version 2, just return ok. File accesses may fail later. 301 * For nfs version 3, use the access rpc to check accessibility. If file modes 302 * are changed on the server, accesses might still fail later. 303 * 304 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 305 */ 306 static int 307 nfs_access(struct vop_access_args *ap) 308 { 309 struct ucred *cred; 310 struct vnode *vp = ap->a_vp; 311 thread_t td = curthread; 312 int error = 0; 313 u_int32_t mode, wmode; 314 struct nfsnode *np = VTONFS(vp); 315 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 316 int v3 = NFS_ISV3(vp); 317 318 lwkt_gettoken(&nmp->nm_token); 319 320 /* 321 * Disallow write attempts on filesystems mounted read-only; 322 * unless the file is a socket, fifo, or a block or character 323 * device resident on the filesystem. 324 */ 325 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 326 switch (vp->v_type) { 327 case VREG: 328 case VDIR: 329 case VLNK: 330 lwkt_reltoken(&nmp->nm_token); 331 return (EROFS); 332 default: 333 break; 334 } 335 } 336 337 /* 338 * The NFS protocol passes only the effective uid/gid over the wire but 339 * we need to check access against real ids if AT_EACCESS not set. 340 * Handle this case by cloning the credentials and setting the 341 * effective ids to the real ones. 342 * 343 * The crdup() here can cause a lot of ucred structures to build-up 344 * (up to maxvnodes), so do our best to avoid it. 345 */ 346 if (ap->a_flags & AT_EACCESS) { 347 cred = crhold(ap->a_cred); 348 } else { 349 cred = ap->a_cred; 350 if (cred->cr_uid == cred->cr_ruid && 351 cred->cr_gid == cred->cr_rgid) { 352 cred = crhold(ap->a_cred); 353 } else { 354 cred = crdup(ap->a_cred); 355 cred->cr_uid = cred->cr_ruid; 356 cred->cr_gid = cred->cr_rgid; 357 } 358 } 359 360 /* 361 * For nfs v3, check to see if we have done this recently, and if 362 * so return our cached result instead of making an ACCESS call. 363 * If not, do an access rpc, otherwise you are stuck emulating 364 * ufs_access() locally using the vattr. This may not be correct, 365 * since the server may apply other access criteria such as 366 * client uid-->server uid mapping that we do not know about. 367 */ 368 if (v3) { 369 if (ap->a_mode & VREAD) 370 mode = NFSV3ACCESS_READ; 371 else 372 mode = 0; 373 if (vp->v_type != VDIR) { 374 if (ap->a_mode & VWRITE) 375 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 376 if (ap->a_mode & VEXEC) 377 mode |= NFSV3ACCESS_EXECUTE; 378 } else { 379 if (ap->a_mode & VWRITE) 380 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 381 NFSV3ACCESS_DELETE); 382 if (ap->a_mode & VEXEC) 383 mode |= NFSV3ACCESS_LOOKUP; 384 } 385 /* XXX safety belt, only make blanket request if caching */ 386 if (nfsaccess_cache_timeout > 0) { 387 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 388 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 389 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 390 } else { 391 wmode = mode; 392 } 393 394 /* 395 * Does our cached result allow us to give a definite yes to 396 * this request? 397 */ 398 if (np->n_modestamp && 399 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 400 (cred->cr_uid == np->n_modeuid) && 401 ((np->n_mode & mode) == mode)) { 402 nfsstats.accesscache_hits++; 403 } else { 404 /* 405 * Either a no, or a don't know. Go to the wire. 406 */ 407 nfsstats.accesscache_misses++; 408 error = nfs3_access_otw(vp, wmode, td, cred); 409 if (!error) { 410 if ((np->n_mode & mode) != mode) { 411 error = EACCES; 412 } 413 } 414 } 415 } else { 416 if ((error = nfs_laccess(ap)) != 0) { 417 crfree(cred); 418 lwkt_reltoken(&nmp->nm_token); 419 return (error); 420 } 421 422 /* 423 * Attempt to prevent a mapped root from accessing a file 424 * which it shouldn't. We try to read a byte from the file 425 * if the user is root and the file is not zero length. 426 * After calling nfs_laccess, we should have the correct 427 * file size cached. 428 */ 429 if (cred->cr_uid == 0 && (ap->a_mode & VREAD) 430 && VTONFS(vp)->n_size > 0) { 431 struct iovec aiov; 432 struct uio auio; 433 char buf[1]; 434 435 aiov.iov_base = buf; 436 aiov.iov_len = 1; 437 auio.uio_iov = &aiov; 438 auio.uio_iovcnt = 1; 439 auio.uio_offset = 0; 440 auio.uio_resid = 1; 441 auio.uio_segflg = UIO_SYSSPACE; 442 auio.uio_rw = UIO_READ; 443 auio.uio_td = td; 444 445 if (vp->v_type == VREG) { 446 error = nfs_readrpc_uio(vp, &auio); 447 } else if (vp->v_type == VDIR) { 448 char* bp; 449 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 450 aiov.iov_base = bp; 451 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 452 error = nfs_readdirrpc_uio(vp, &auio); 453 kfree(bp, M_TEMP); 454 } else if (vp->v_type == VLNK) { 455 error = nfs_readlinkrpc_uio(vp, &auio); 456 } else { 457 error = EACCES; 458 } 459 } 460 } 461 /* 462 * [re]record creds for reading and/or writing if access 463 * was granted. Assume the NFS server will grant read access 464 * for execute requests. 465 */ 466 if (error == 0) { 467 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) { 468 crhold(cred); 469 if (np->n_rucred) 470 crfree(np->n_rucred); 471 np->n_rucred = cred; 472 } 473 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) { 474 crhold(cred); 475 if (np->n_wucred) 476 crfree(np->n_wucred); 477 np->n_wucred = cred; 478 } 479 } 480 lwkt_reltoken(&nmp->nm_token); 481 crfree(cred); 482 483 return(error); 484 } 485 486 /* 487 * nfs open vnode op 488 * Check to see if the type is ok 489 * and that deletion is not in progress. 490 * For paged in text files, you will need to flush the page cache 491 * if consistency is lost. 492 * 493 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 494 * struct file *a_fp) 495 */ 496 /* ARGSUSED */ 497 static int 498 nfs_open(struct vop_open_args *ap) 499 { 500 struct vnode *vp = ap->a_vp; 501 struct nfsnode *np = VTONFS(vp); 502 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 503 struct vattr vattr; 504 int error; 505 506 lwkt_gettoken(&nmp->nm_token); 507 508 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 509 #ifdef DIAGNOSTIC 510 kprintf("open eacces vtyp=%d\n",vp->v_type); 511 #endif 512 lwkt_reltoken(&nmp->nm_token); 513 return (EOPNOTSUPP); 514 } 515 516 /* 517 * Save valid creds for reading and writing for later RPCs. 518 */ 519 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) { 520 crhold(ap->a_cred); 521 if (np->n_rucred) 522 crfree(np->n_rucred); 523 np->n_rucred = ap->a_cred; 524 } 525 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) { 526 crhold(ap->a_cred); 527 if (np->n_wucred) 528 crfree(np->n_wucred); 529 np->n_wucred = ap->a_cred; 530 } 531 532 /* 533 * Clear the attribute cache only if opening with write access. It 534 * is unclear if we should do this at all here, but we certainly 535 * should not clear the cache unconditionally simply because a file 536 * is being opened. 537 */ 538 if (ap->a_mode & FWRITE) 539 np->n_attrstamp = 0; 540 541 /* 542 * For normal NFS, reconcile changes made locally verses 543 * changes made remotely. Note that VOP_GETATTR only goes 544 * to the wire if the cached attribute has timed out or been 545 * cleared. 546 * 547 * If local modifications have been made clear the attribute 548 * cache to force an attribute and modified time check. If 549 * GETATTR detects that the file has been changed by someone 550 * other then us it will set NRMODIFIED. 551 * 552 * If we are opening a directory and local changes have been 553 * made we have to invalidate the cache in order to ensure 554 * that we get the most up-to-date information from the 555 * server. XXX 556 */ 557 if (np->n_flag & NLMODIFIED) { 558 np->n_attrstamp = 0; 559 if (vp->v_type == VDIR) { 560 error = nfs_vinvalbuf(vp, V_SAVE, 1); 561 if (error == EINTR) { 562 lwkt_reltoken(&nmp->nm_token); 563 return (error); 564 } 565 nfs_invaldir(vp); 566 } 567 } 568 error = VOP_GETATTR(vp, &vattr); 569 if (error) { 570 lwkt_reltoken(&nmp->nm_token); 571 return (error); 572 } 573 if (np->n_flag & NRMODIFIED) { 574 if (vp->v_type == VDIR) 575 nfs_invaldir(vp); 576 error = nfs_vinvalbuf(vp, V_SAVE, 1); 577 if (error == EINTR) { 578 lwkt_reltoken(&nmp->nm_token); 579 return (error); 580 } 581 np->n_flag &= ~NRMODIFIED; 582 } 583 error = vop_stdopen(ap); 584 lwkt_reltoken(&nmp->nm_token); 585 586 return error; 587 } 588 589 /* 590 * nfs close vnode op 591 * What an NFS client should do upon close after writing is a debatable issue. 592 * Most NFS clients push delayed writes to the server upon close, basically for 593 * two reasons: 594 * 1 - So that any write errors may be reported back to the client process 595 * doing the close system call. By far the two most likely errors are 596 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 597 * 2 - To put a worst case upper bound on cache inconsistency between 598 * multiple clients for the file. 599 * There is also a consistency problem for Version 2 of the protocol w.r.t. 600 * not being able to tell if other clients are writing a file concurrently, 601 * since there is no way of knowing if the changed modify time in the reply 602 * is only due to the write for this client. 603 * (NFS Version 3 provides weak cache consistency data in the reply that 604 * should be sufficient to detect and handle this case.) 605 * 606 * The current code does the following: 607 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 608 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 609 * or commit them (this satisfies 1 and 2 except for the 610 * case where the server crashes after this close but 611 * before the commit RPC, which is felt to be "good 612 * enough". Changing the last argument to nfs_flush() to 613 * a 1 would force a commit operation, if it is felt a 614 * commit is necessary now. 615 * for NQNFS - do nothing now, since 2 is dealt with via leases and 616 * 1 should be dealt with via an fsync() system call for 617 * cases where write errors are important. 618 * 619 * nfs_close(struct vnode *a_vp, int a_fflag) 620 */ 621 /* ARGSUSED */ 622 static int 623 nfs_close(struct vop_close_args *ap) 624 { 625 struct vnode *vp = ap->a_vp; 626 struct nfsnode *np = VTONFS(vp); 627 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 628 int error = 0; 629 thread_t td = curthread; 630 631 vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */ 632 lwkt_gettoken(&nmp->nm_token); 633 634 if (vp->v_type == VREG) { 635 if (np->n_flag & NLMODIFIED) { 636 if (NFS_ISV3(vp)) { 637 /* 638 * Under NFSv3 we have dirty buffers to dispose of. We 639 * must flush them to the NFS server. We have the option 640 * of waiting all the way through the commit rpc or just 641 * waiting for the initial write. The default is to only 642 * wait through the initial write so the data is in the 643 * server's cache, which is roughly similar to the state 644 * a standard disk subsystem leaves the file in on close(). 645 * 646 * We cannot clear the NLMODIFIED bit in np->n_flag due to 647 * potential races with other processes, and certainly 648 * cannot clear it if we don't commit. 649 */ 650 int cm = nfsv3_commit_on_close ? 1 : 0; 651 error = nfs_flush(vp, MNT_WAIT, td, cm); 652 /* np->n_flag &= ~NLMODIFIED; */ 653 } else { 654 error = nfs_vinvalbuf(vp, V_SAVE, 1); 655 } 656 np->n_attrstamp = 0; 657 } 658 if (np->n_flag & NWRITEERR) { 659 np->n_flag &= ~NWRITEERR; 660 error = np->n_error; 661 } 662 } 663 vop_stdclose(ap); 664 lwkt_reltoken(&nmp->nm_token); 665 666 return (error); 667 } 668 669 /* 670 * nfs getattr call from vfs. 671 * 672 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap) 673 */ 674 static int 675 nfs_getattr(struct vop_getattr_args *ap) 676 { 677 struct vnode *vp = ap->a_vp; 678 struct nfsnode *np = VTONFS(vp); 679 struct nfsmount *nmp; 680 int error = 0; 681 thread_t td = curthread; 682 struct nfsm_info info; 683 684 info.mrep = NULL; 685 info.v3 = NFS_ISV3(vp); 686 nmp = VFSTONFS(vp->v_mount); 687 688 lwkt_gettoken(&nmp->nm_token); 689 690 /* 691 * Update local times for special files. 692 */ 693 if (np->n_flag & (NACC | NUPD)) 694 np->n_flag |= NCHG; 695 /* 696 * First look in the cache. 697 */ 698 if (nfs_getattrcache(vp, ap->a_vap) == 0) 699 goto done; 700 701 if (info.v3 && nfsaccess_cache_timeout > 0) { 702 nfsstats.accesscache_misses++; 703 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 704 if (nfs_getattrcache(vp, ap->a_vap) == 0) 705 goto done; 706 } 707 708 nfsstats.rpccnt[NFSPROC_GETATTR]++; 709 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3)); 710 ERROROUT(nfsm_fhtom(&info, vp)); 711 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td, 712 nfs_vpcred(vp, ND_CHECK), &error)); 713 if (error == 0) { 714 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap)); 715 } 716 m_freem(info.mrep); 717 info.mrep = NULL; 718 done: 719 /* 720 * NFS doesn't support chflags flags. If the nfs mount was 721 * made -o cache set the UF_CACHE bit for swapcache. 722 */ 723 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT)) 724 ap->a_vap->va_flags |= UF_CACHE; 725 nfsmout: 726 lwkt_reltoken(&nmp->nm_token); 727 return (error); 728 } 729 730 /* 731 * nfs setattr call. 732 * 733 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 734 */ 735 static int 736 nfs_setattr(struct vop_setattr_args *ap) 737 { 738 struct vnode *vp = ap->a_vp; 739 struct nfsnode *np = VTONFS(vp); 740 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 741 struct vattr *vap = ap->a_vap; 742 int error = 0; 743 int kflags = 0; 744 off_t tsize; 745 thread_t td = curthread; 746 747 #ifndef nolint 748 tsize = (off_t)0; 749 #endif 750 /* 751 * Setting of flags is not supported. 752 */ 753 if (vap->va_flags != VNOVAL) 754 return (EOPNOTSUPP); 755 756 /* 757 * Disallow write attempts if the filesystem is mounted read-only. 758 */ 759 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 760 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 761 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 762 (vp->v_mount->mnt_flag & MNT_RDONLY)) 763 return (EROFS); 764 765 lwkt_gettoken(&nmp->nm_token); 766 767 /* 768 * Handle size changes 769 */ 770 if (vap->va_size != VNOVAL) { 771 /* 772 * truncation requested 773 */ 774 switch (vp->v_type) { 775 case VDIR: 776 lwkt_reltoken(&nmp->nm_token); 777 return (EISDIR); 778 case VCHR: 779 case VBLK: 780 case VSOCK: 781 case VFIFO: 782 if (vap->va_mtime.tv_sec == VNOVAL && 783 vap->va_atime.tv_sec == VNOVAL && 784 vap->va_mode == (mode_t)VNOVAL && 785 vap->va_uid == (uid_t)VNOVAL && 786 vap->va_gid == (gid_t)VNOVAL) { 787 lwkt_reltoken(&nmp->nm_token); 788 return (0); 789 } 790 vap->va_size = VNOVAL; 791 break; 792 default: 793 /* 794 * Disallow write attempts if the filesystem is 795 * mounted read-only. 796 */ 797 if (vp->v_mount->mnt_flag & MNT_RDONLY) { 798 lwkt_reltoken(&nmp->nm_token); 799 return (EROFS); 800 } 801 802 tsize = np->n_size; 803 again: 804 error = nfs_meta_setsize(vp, td, vap->va_size, 0); 805 806 #if 0 807 if (np->n_flag & NLMODIFIED) { 808 if (vap->va_size == 0) 809 error = nfs_vinvalbuf(vp, 0, 1); 810 else 811 error = nfs_vinvalbuf(vp, V_SAVE, 1); 812 } 813 #endif 814 /* 815 * note: this loop case almost always happens at 816 * least once per truncation. 817 */ 818 if (error == 0 && np->n_size != vap->va_size) 819 goto again; 820 np->n_vattr.va_size = vap->va_size; 821 kflags |= NOTE_WRITE; 822 if (tsize < vap->va_size) 823 kflags |= NOTE_EXTEND; 824 break; 825 } 826 } 827 828 /* 829 * If setting the mtime or if server/other-client modifications have 830 * been detected, we must fully flush any pending writes. 831 * 832 * This will slow down cp/cpdup/rdist/rsync and other operations which 833 * might call [l]utimes() to set the mtime after writing to a file, 834 * but honestly there is no way to properly defer the write flush 835 * and still get reasonably accurate/dependable synchronization of 836 * [l]utimes(). 837 */ 838 if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) { 839 if ((np->n_flag & NRMODIFIED) || 840 (vap->va_mtime.tv_sec != VNOVAL)) { 841 error = nfs_vinvalbuf(vp, V_SAVE, 1); 842 if (error == EINTR) { 843 lwkt_reltoken(&nmp->nm_token); 844 return (error); 845 } 846 } 847 } 848 849 /* 850 * Get the blasted mtime to report properly. 851 */ 852 if (vap->va_mtime.tv_sec != VNOVAL) { 853 np->n_mtime = vap->va_mtime.tv_sec; 854 np->n_flag &= ~NUPD; 855 np->n_vattr.va_mtime = vap->va_mtime; 856 } 857 858 /* 859 * Issue the setattr rpc, adjust our mtime and make sure NUPD 860 * has been cleared so it does not get overridden. 861 */ 862 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 863 if (error == 0) 864 kflags |= NOTE_EXTEND; 865 866 /* 867 * Sanity check if a truncation was issued. This should only occur 868 * if multiple processes are racing on the same file. 869 */ 870 if (error == 0 && vap->va_size != VNOVAL && 871 np->n_size != vap->va_size) { 872 kprintf("NFS ftruncate: server disagrees on the file size: " 873 "%jd/%jd/%jd\n", 874 (intmax_t)tsize, 875 (intmax_t)vap->va_size, 876 (intmax_t)np->n_size); 877 goto again; 878 } 879 if (error && vap->va_size != VNOVAL) { 880 np->n_size = np->n_vattr.va_size = tsize; 881 nfs_meta_setsize(vp, td, np->n_size, 0); 882 } 883 lwkt_reltoken(&nmp->nm_token); 884 nfs_knote(vp, kflags); 885 886 return (error); 887 } 888 889 /* 890 * Do an nfs setattr rpc. 891 */ 892 static int 893 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 894 struct ucred *cred, struct thread *td) 895 { 896 struct nfsv2_sattr *sp; 897 struct nfsnode *np = VTONFS(vp); 898 u_int32_t *tl; 899 int error = 0, wccflag = NFSV3_WCCRATTR; 900 struct nfsm_info info; 901 902 info.mrep = NULL; 903 info.v3 = NFS_ISV3(vp); 904 905 nfsstats.rpccnt[NFSPROC_SETATTR]++; 906 nfsm_reqhead(&info, vp, NFSPROC_SETATTR, 907 NFSX_FH(info.v3) + NFSX_SATTR(info.v3)); 908 ERROROUT(nfsm_fhtom(&info, vp)); 909 if (info.v3) { 910 nfsm_v3attrbuild(&info, vap, TRUE); 911 tl = nfsm_build(&info, NFSX_UNSIGNED); 912 *tl = nfs_false; 913 } else { 914 sp = nfsm_build(&info, NFSX_V2SATTR); 915 if (vap->va_mode == (mode_t)VNOVAL) 916 sp->sa_mode = nfs_xdrneg1; 917 else 918 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 919 if (vap->va_uid == (uid_t)VNOVAL) 920 sp->sa_uid = nfs_xdrneg1; 921 else 922 sp->sa_uid = txdr_unsigned(vap->va_uid); 923 if (vap->va_gid == (gid_t)VNOVAL) 924 sp->sa_gid = nfs_xdrneg1; 925 else 926 sp->sa_gid = txdr_unsigned(vap->va_gid); 927 sp->sa_size = txdr_unsigned(vap->va_size); 928 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 929 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 930 } 931 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error)); 932 if (info.v3) { 933 np->n_modestamp = 0; 934 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 935 } else { 936 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 937 } 938 m_freem(info.mrep); 939 info.mrep = NULL; 940 nfsmout: 941 return (error); 942 } 943 944 static 945 void 946 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout) 947 { 948 if (nctimeout == 0) 949 nctimeout = 1; 950 else 951 nctimeout *= hz; 952 cache_setvp(nch, vp); 953 cache_settimeout(nch, nctimeout); 954 } 955 956 /* 957 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 958 * nfs_lookup() until all remaining new api calls are implemented. 959 * 960 * Resolve a namecache entry. This function is passed a locked ncp and 961 * must call nfs_cache_setvp() on it as appropriate to resolve the entry. 962 */ 963 static int 964 nfs_nresolve(struct vop_nresolve_args *ap) 965 { 966 struct thread *td = curthread; 967 struct namecache *ncp; 968 struct nfsmount *nmp; 969 struct nfsnode *np; 970 struct vnode *dvp; 971 struct vnode *nvp; 972 nfsfh_t *fhp; 973 int attrflag; 974 int fhsize; 975 int error; 976 int tmp_error; 977 int len; 978 struct nfsm_info info; 979 980 dvp = ap->a_dvp; 981 nmp = VFSTONFS(dvp->v_mount); 982 983 lwkt_gettoken(&nmp->nm_token); 984 985 if ((error = vget(dvp, LK_SHARED)) != 0) { 986 lwkt_reltoken(&nmp->nm_token); 987 return (error); 988 } 989 990 info.mrep = NULL; 991 info.v3 = NFS_ISV3(dvp); 992 993 nvp = NULL; 994 nfsstats.lookupcache_misses++; 995 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 996 ncp = ap->a_nch->ncp; 997 len = ncp->nc_nlen; 998 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 999 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1000 ERROROUT(nfsm_fhtom(&info, dvp)); 1001 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN)); 1002 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, 1003 ap->a_cred, &error)); 1004 if (error) { 1005 /* 1006 * Cache negatve lookups to reduce NFS traffic, but use 1007 * a fast timeout. Otherwise use a timeout of 1 tick. 1008 * XXX we should add a namecache flag for no-caching 1009 * to uncache the negative hit as soon as possible, but 1010 * we cannot simply destroy the entry because it is used 1011 * as a placeholder by the caller. 1012 * 1013 * The refactored nfs code will overwrite a non-zero error 1014 * with 0 when we use ERROROUT(), so don't here. 1015 */ 1016 if (error == ENOENT) 1017 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout); 1018 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 1019 NFS_LATTR_NOSHRINK); 1020 if (tmp_error) { 1021 error = tmp_error; 1022 goto nfsmout; 1023 } 1024 m_freem(info.mrep); 1025 info.mrep = NULL; 1026 goto nfsmout; 1027 } 1028 1029 /* 1030 * Success, get the file handle, do various checks, and load 1031 * post-operation data from the reply packet. Theoretically 1032 * we should never be looking up "." so, theoretically, we 1033 * should never get the same file handle as our directory. But 1034 * we check anyway. XXX 1035 * 1036 * Note that no timeout is set for the positive cache hit. We 1037 * assume, theoretically, that ESTALE returns will be dealt with 1038 * properly to handle NFS races and in anycase we cannot depend 1039 * on a timeout to deal with NFS open/create/excl issues so instead 1040 * of a bad hack here the rest of the NFS client code needs to do 1041 * the right thing. 1042 */ 1043 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 1044 1045 np = VTONFS(dvp); 1046 if (NFS_CMPFH(np, fhp, fhsize)) { 1047 vref(dvp); 1048 nvp = dvp; 1049 } else { 1050 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, NULL); 1051 if (error) { 1052 m_freem(info.mrep); 1053 info.mrep = NULL; 1054 vput(dvp); 1055 lwkt_reltoken(&nmp->nm_token); 1056 return (error); 1057 } 1058 nvp = NFSTOV(np); 1059 } 1060 if (info.v3) { 1061 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag, 1062 NFS_LATTR_NOSHRINK)); 1063 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1064 NFS_LATTR_NOSHRINK)); 1065 } else { 1066 ERROROUT(nfsm_loadattr(&info, nvp, NULL)); 1067 } 1068 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout); 1069 m_freem(info.mrep); 1070 info.mrep = NULL; 1071 nfsmout: 1072 lwkt_reltoken(&nmp->nm_token); 1073 vput(dvp); 1074 if (nvp) { 1075 if (nvp == dvp) 1076 vrele(nvp); 1077 else 1078 vput(nvp); 1079 } 1080 return (error); 1081 } 1082 1083 /* 1084 * 'cached' nfs directory lookup 1085 * 1086 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 1087 * 1088 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 1089 * struct componentname *a_cnp) 1090 */ 1091 static int 1092 nfs_lookup(struct vop_old_lookup_args *ap) 1093 { 1094 struct componentname *cnp = ap->a_cnp; 1095 struct vnode *dvp = ap->a_dvp; 1096 struct vnode **vpp = ap->a_vpp; 1097 int flags = cnp->cn_flags; 1098 struct vnode *newvp; 1099 struct vnode *notvp; 1100 struct nfsmount *nmp; 1101 long len; 1102 nfsfh_t *fhp; 1103 struct nfsnode *np; 1104 int lockparent, wantparent, attrflag, fhsize; 1105 int error; 1106 int tmp_error; 1107 struct nfsm_info info; 1108 1109 info.mrep = NULL; 1110 info.v3 = NFS_ISV3(dvp); 1111 error = 0; 1112 1113 notvp = (cnp->cn_flags & CNP_NOTVP) ? cnp->cn_notvp : NULL; 1114 1115 /* 1116 * Read-only mount check and directory check. 1117 */ 1118 *vpp = NULLVP; 1119 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1120 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1121 return (EROFS); 1122 1123 if (dvp->v_type != VDIR) 1124 return (ENOTDIR); 1125 1126 /* 1127 * Look it up in the cache. Note that ENOENT is only returned if we 1128 * previously entered a negative hit (see later on). The additional 1129 * nfsneg_cache_timeout check causes previously cached results to 1130 * be instantly ignored if the negative caching is turned off. 1131 */ 1132 lockparent = flags & CNP_LOCKPARENT; 1133 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1134 nmp = VFSTONFS(dvp->v_mount); 1135 np = VTONFS(dvp); 1136 1137 lwkt_gettoken(&nmp->nm_token); 1138 1139 /* 1140 * Go to the wire. 1141 */ 1142 error = 0; 1143 newvp = NULLVP; 1144 nfsstats.lookupcache_misses++; 1145 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1146 len = cnp->cn_namelen; 1147 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 1148 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1149 ERROROUT(nfsm_fhtom(&info, dvp)); 1150 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 1151 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td, 1152 cnp->cn_cred, &error)); 1153 if (error) { 1154 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 1155 NFS_LATTR_NOSHRINK); 1156 if (tmp_error) { 1157 error = tmp_error; 1158 goto nfsmout; 1159 } 1160 1161 m_freem(info.mrep); 1162 info.mrep = NULL; 1163 goto nfsmout; 1164 } 1165 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 1166 1167 /* 1168 * Handle RENAME case... 1169 */ 1170 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1171 if (NFS_CMPFH(np, fhp, fhsize)) { 1172 m_freem(info.mrep); 1173 info.mrep = NULL; 1174 lwkt_reltoken(&nmp->nm_token); 1175 return (EISDIR); 1176 } 1177 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp); 1178 if (error) { 1179 m_freem(info.mrep); 1180 info.mrep = NULL; 1181 lwkt_reltoken(&nmp->nm_token); 1182 return (error); 1183 } 1184 newvp = NFSTOV(np); 1185 if (info.v3) { 1186 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1187 NFS_LATTR_NOSHRINK)); 1188 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1189 NFS_LATTR_NOSHRINK)); 1190 } else { 1191 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1192 } 1193 *vpp = newvp; 1194 m_freem(info.mrep); 1195 info.mrep = NULL; 1196 if (!lockparent) { 1197 vn_unlock(dvp); 1198 cnp->cn_flags |= CNP_PDIRUNLOCK; 1199 } 1200 lwkt_reltoken(&nmp->nm_token); 1201 return (0); 1202 } 1203 1204 if (flags & CNP_ISDOTDOT) { 1205 vn_unlock(dvp); 1206 cnp->cn_flags |= CNP_PDIRUNLOCK; 1207 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp); 1208 if (error) { 1209 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1210 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1211 lwkt_reltoken(&nmp->nm_token); 1212 return (error); /* NOTE: return error from nget */ 1213 } 1214 newvp = NFSTOV(np); 1215 if (lockparent) { 1216 error = vn_lock(dvp, LK_EXCLUSIVE | LK_FAILRECLAIM); 1217 if (error) { 1218 vput(newvp); 1219 lwkt_reltoken(&nmp->nm_token); 1220 return (error); 1221 } 1222 cnp->cn_flags |= CNP_PDIRUNLOCK; 1223 } 1224 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1225 vref(dvp); 1226 newvp = dvp; 1227 } else { 1228 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp); 1229 if (error) { 1230 m_freem(info.mrep); 1231 info.mrep = NULL; 1232 lwkt_reltoken(&nmp->nm_token); 1233 return (error); 1234 } 1235 if (!lockparent) { 1236 vn_unlock(dvp); 1237 cnp->cn_flags |= CNP_PDIRUNLOCK; 1238 } 1239 newvp = NFSTOV(np); 1240 } 1241 if (info.v3) { 1242 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1243 NFS_LATTR_NOSHRINK)); 1244 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1245 NFS_LATTR_NOSHRINK)); 1246 } else { 1247 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1248 } 1249 #if 0 1250 /* XXX MOVE TO nfs_nremove() */ 1251 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1252 cnp->cn_nameiop != NAMEI_DELETE) { 1253 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1254 } 1255 #endif 1256 *vpp = newvp; 1257 m_freem(info.mrep); 1258 info.mrep = NULL; 1259 nfsmout: 1260 if (error) { 1261 if (newvp != NULLVP) { 1262 vrele(newvp); 1263 *vpp = NULLVP; 1264 } 1265 if ((cnp->cn_nameiop == NAMEI_CREATE || 1266 cnp->cn_nameiop == NAMEI_RENAME) && 1267 error == ENOENT) { 1268 if (!lockparent) { 1269 vn_unlock(dvp); 1270 cnp->cn_flags |= CNP_PDIRUNLOCK; 1271 } 1272 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1273 error = EROFS; 1274 else 1275 error = EJUSTRETURN; 1276 } 1277 } 1278 lwkt_reltoken(&nmp->nm_token); 1279 return (error); 1280 } 1281 1282 /* 1283 * nfs read call. 1284 * Just call nfs_bioread() to do the work. 1285 * 1286 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1287 * struct ucred *a_cred) 1288 */ 1289 static int 1290 nfs_read(struct vop_read_args *ap) 1291 { 1292 struct vnode *vp = ap->a_vp; 1293 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1294 int error; 1295 1296 lwkt_gettoken(&nmp->nm_token); 1297 error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag); 1298 lwkt_reltoken(&nmp->nm_token); 1299 1300 return error; 1301 } 1302 1303 /* 1304 * nfs readlink call 1305 * 1306 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1307 */ 1308 static int 1309 nfs_readlink(struct vop_readlink_args *ap) 1310 { 1311 struct vnode *vp = ap->a_vp; 1312 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1313 int error; 1314 1315 if (vp->v_type != VLNK) 1316 return (EINVAL); 1317 1318 lwkt_gettoken(&nmp->nm_token); 1319 error = nfs_bioread(vp, ap->a_uio, 0); 1320 lwkt_reltoken(&nmp->nm_token); 1321 1322 return error; 1323 } 1324 1325 /* 1326 * Do a readlink rpc. 1327 * Called by nfs_doio() from below the buffer cache. 1328 */ 1329 int 1330 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop) 1331 { 1332 int error = 0, len, attrflag; 1333 struct nfsm_info info; 1334 1335 info.mrep = NULL; 1336 info.v3 = NFS_ISV3(vp); 1337 1338 nfsstats.rpccnt[NFSPROC_READLINK]++; 1339 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3)); 1340 ERROROUT(nfsm_fhtom(&info, vp)); 1341 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td, 1342 nfs_vpcred(vp, ND_CHECK), &error)); 1343 if (info.v3) { 1344 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1345 NFS_LATTR_NOSHRINK)); 1346 } 1347 if (!error) { 1348 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN)); 1349 if (len == NFS_MAXPATHLEN) { 1350 struct nfsnode *np = VTONFS(vp); 1351 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1352 len = np->n_size; 1353 } 1354 ERROROUT(nfsm_mtouio(&info, uiop, len)); 1355 } 1356 m_freem(info.mrep); 1357 info.mrep = NULL; 1358 nfsmout: 1359 return (error); 1360 } 1361 1362 /* 1363 * nfs synchronous read rpc using UIO 1364 */ 1365 int 1366 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop) 1367 { 1368 u_int32_t *tl; 1369 struct nfsmount *nmp; 1370 int error = 0, len, retlen, tsiz, eof, attrflag; 1371 struct nfsm_info info; 1372 off_t tmp_off; 1373 1374 info.mrep = NULL; 1375 info.v3 = NFS_ISV3(vp); 1376 1377 #ifndef nolint 1378 eof = 0; 1379 #endif 1380 nmp = VFSTONFS(vp->v_mount); 1381 1382 tsiz = uiop->uio_resid; 1383 tmp_off = uiop->uio_offset + tsiz; 1384 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) 1385 return (EFBIG); 1386 tmp_off = uiop->uio_offset; 1387 while (tsiz > 0) { 1388 nfsstats.rpccnt[NFSPROC_READ]++; 1389 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1390 nfsm_reqhead(&info, vp, NFSPROC_READ, 1391 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3); 1392 ERROROUT(nfsm_fhtom(&info, vp)); 1393 tl = nfsm_build(&info, NFSX_UNSIGNED * 3); 1394 if (info.v3) { 1395 txdr_hyper(uiop->uio_offset, tl); 1396 *(tl + 2) = txdr_unsigned(len); 1397 } else { 1398 *tl++ = txdr_unsigned(uiop->uio_offset); 1399 *tl++ = txdr_unsigned(len); 1400 *tl = 0; 1401 } 1402 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td, 1403 nfs_vpcred(vp, ND_READ), &error)); 1404 if (info.v3) { 1405 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1406 NFS_LATTR_NOSHRINK)); 1407 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 1408 eof = fxdr_unsigned(int, *(tl + 1)); 1409 } else { 1410 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1411 } 1412 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len)); 1413 ERROROUT(nfsm_mtouio(&info, uiop, retlen)); 1414 m_freem(info.mrep); 1415 info.mrep = NULL; 1416 1417 /* 1418 * Handle short-read from server (NFSv3). If EOF is not 1419 * flagged (and no error occurred), but retlen is less 1420 * then the request size, we must zero-fill the remainder. 1421 */ 1422 if (retlen < len && info.v3 && eof == 0) { 1423 ERROROUT(uiomovez(len - retlen, uiop)); 1424 retlen = len; 1425 } 1426 tsiz -= retlen; 1427 1428 /* 1429 * Terminate loop on EOF or zero-length read. 1430 * 1431 * For NFSv2 a short-read indicates EOF, not zero-fill, 1432 * and also terminates the loop. 1433 */ 1434 if (info.v3) { 1435 if (eof || retlen == 0) 1436 tsiz = 0; 1437 } else if (retlen < len) { 1438 tsiz = 0; 1439 } 1440 } 1441 nfsmout: 1442 return (error); 1443 } 1444 1445 /* 1446 * nfs write call 1447 */ 1448 int 1449 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop, 1450 int *iomode, int *must_commit) 1451 { 1452 u_int32_t *tl; 1453 int32_t backup; 1454 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1455 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1456 int committed = NFSV3WRITE_FILESYNC; 1457 struct nfsm_info info; 1458 1459 info.mrep = NULL; 1460 info.v3 = NFS_ISV3(vp); 1461 1462 #ifndef DIAGNOSTIC 1463 if (uiop->uio_iovcnt != 1) 1464 panic("nfs: writerpc iovcnt > 1"); 1465 #endif 1466 *must_commit = 0; 1467 tsiz = uiop->uio_resid; 1468 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1469 return (EFBIG); 1470 while (tsiz > 0) { 1471 nfsstats.rpccnt[NFSPROC_WRITE]++; 1472 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1473 nfsm_reqhead(&info, vp, NFSPROC_WRITE, 1474 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + 1475 nfsm_rndup(len)); 1476 ERROROUT(nfsm_fhtom(&info, vp)); 1477 if (info.v3) { 1478 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 1479 txdr_hyper(uiop->uio_offset, tl); 1480 tl += 2; 1481 *tl++ = txdr_unsigned(len); 1482 *tl++ = txdr_unsigned(*iomode); 1483 *tl = txdr_unsigned(len); 1484 } else { 1485 u_int32_t x; 1486 1487 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED); 1488 /* Set both "begin" and "current" to non-garbage. */ 1489 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1490 *tl++ = x; /* "begin offset" */ 1491 *tl++ = x; /* "current offset" */ 1492 x = txdr_unsigned(len); 1493 *tl++ = x; /* total to this offset */ 1494 *tl = x; /* size of this write */ 1495 } 1496 ERROROUT(nfsm_uiotom(&info, uiop, len)); 1497 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td, 1498 nfs_vpcred(vp, ND_WRITE), &error)); 1499 if (info.v3) { 1500 /* 1501 * The write RPC returns a before and after mtime. The 1502 * nfsm_wcc_data() macro checks the before n_mtime 1503 * against the before time and stores the after time 1504 * in the nfsnode's cached vattr and n_mtime field. 1505 * The NRMODIFIED bit will be set if the before 1506 * time did not match the original mtime. 1507 */ 1508 wccflag = NFSV3_WCCCHK; 1509 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 1510 if (error == 0) { 1511 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF)); 1512 rlen = fxdr_unsigned(int, *tl++); 1513 if (rlen == 0) { 1514 error = NFSERR_IO; 1515 m_freem(info.mrep); 1516 info.mrep = NULL; 1517 break; 1518 } else if (rlen < len) { 1519 backup = len - rlen; 1520 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup; 1521 uiop->uio_iov->iov_len += backup; 1522 uiop->uio_offset -= backup; 1523 uiop->uio_resid += backup; 1524 len = rlen; 1525 } 1526 commit = fxdr_unsigned(int, *tl++); 1527 1528 /* 1529 * Return the lowest committment level 1530 * obtained by any of the RPCs. 1531 */ 1532 if (committed == NFSV3WRITE_FILESYNC) 1533 committed = commit; 1534 else if (committed == NFSV3WRITE_DATASYNC && 1535 commit == NFSV3WRITE_UNSTABLE) 1536 committed = commit; 1537 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1538 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1539 NFSX_V3WRITEVERF); 1540 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1541 } else if (bcmp((caddr_t)tl, 1542 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1543 *must_commit = 1; 1544 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1545 NFSX_V3WRITEVERF); 1546 } 1547 } 1548 } else { 1549 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1550 } 1551 m_freem(info.mrep); 1552 info.mrep = NULL; 1553 if (error) 1554 break; 1555 tsiz -= len; 1556 } 1557 nfsmout: 1558 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1559 committed = NFSV3WRITE_FILESYNC; 1560 *iomode = committed; 1561 if (error) 1562 uiop->uio_resid = tsiz; 1563 return (error); 1564 } 1565 1566 /* 1567 * nfs mknod rpc 1568 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1569 * mode set to specify the file type and the size field for rdev. 1570 */ 1571 static int 1572 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1573 struct vattr *vap) 1574 { 1575 struct nfsv2_sattr *sp; 1576 u_int32_t *tl; 1577 struct vnode *newvp = NULL; 1578 struct nfsnode *np = NULL; 1579 struct vattr vattr; 1580 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1581 int rmajor, rminor; 1582 struct nfsm_info info; 1583 1584 info.mrep = NULL; 1585 info.v3 = NFS_ISV3(dvp); 1586 1587 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1588 rmajor = txdr_unsigned(vap->va_rmajor); 1589 rminor = txdr_unsigned(vap->va_rminor); 1590 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) { 1591 rmajor = nfs_xdrneg1; 1592 rminor = nfs_xdrneg1; 1593 } else { 1594 return (EOPNOTSUPP); 1595 } 1596 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1597 return (error); 1598 } 1599 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1600 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD, 1601 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED + 1602 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1603 ERROROUT(nfsm_fhtom(&info, dvp)); 1604 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1605 NFS_MAXNAMLEN)); 1606 if (info.v3) { 1607 tl = nfsm_build(&info, NFSX_UNSIGNED); 1608 *tl++ = vtonfsv3_type(vap->va_type); 1609 nfsm_v3attrbuild(&info, vap, FALSE); 1610 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1611 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 1612 *tl++ = txdr_unsigned(vap->va_rmajor); 1613 *tl = txdr_unsigned(vap->va_rminor); 1614 } 1615 } else { 1616 sp = nfsm_build(&info, NFSX_V2SATTR); 1617 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1618 sp->sa_uid = nfs_xdrneg1; 1619 sp->sa_gid = nfs_xdrneg1; 1620 sp->sa_size = makeudev(rmajor, rminor); 1621 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1622 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1623 } 1624 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td, 1625 cnp->cn_cred, &error)); 1626 if (!error) { 1627 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1628 if (!gotvp) { 1629 if (newvp) { 1630 vput(newvp); 1631 newvp = NULL; 1632 } 1633 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1634 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1635 if (!error) 1636 newvp = NFSTOV(np); 1637 } 1638 } 1639 if (info.v3) { 1640 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1641 } 1642 m_freem(info.mrep); 1643 info.mrep = NULL; 1644 nfsmout: 1645 if (error) { 1646 if (newvp) 1647 vput(newvp); 1648 } else { 1649 *vpp = newvp; 1650 } 1651 VTONFS(dvp)->n_flag |= NLMODIFIED; 1652 if (!wccflag) 1653 VTONFS(dvp)->n_attrstamp = 0; 1654 return (error); 1655 } 1656 1657 /* 1658 * nfs mknod vop 1659 * just call nfs_mknodrpc() to do the work. 1660 * 1661 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1662 * struct componentname *a_cnp, struct vattr *a_vap) 1663 */ 1664 /* ARGSUSED */ 1665 static int 1666 nfs_mknod(struct vop_old_mknod_args *ap) 1667 { 1668 struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount); 1669 int error; 1670 1671 lwkt_gettoken(&nmp->nm_token); 1672 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1673 lwkt_reltoken(&nmp->nm_token); 1674 if (error == 0) 1675 nfs_knote(ap->a_dvp, NOTE_WRITE); 1676 1677 return error; 1678 } 1679 1680 static u_long create_verf; 1681 /* 1682 * nfs file create call 1683 * 1684 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1685 * struct componentname *a_cnp, struct vattr *a_vap) 1686 */ 1687 static int 1688 nfs_create(struct vop_old_create_args *ap) 1689 { 1690 struct vnode *dvp = ap->a_dvp; 1691 struct vattr *vap = ap->a_vap; 1692 struct nfsmount *nmp = VFSTONFS(dvp->v_mount); 1693 struct componentname *cnp = ap->a_cnp; 1694 struct nfsv2_sattr *sp; 1695 u_int32_t *tl; 1696 struct nfsnode *np = NULL; 1697 struct vnode *newvp = NULL; 1698 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1699 struct vattr vattr; 1700 struct nfsm_info info; 1701 1702 info.mrep = NULL; 1703 info.v3 = NFS_ISV3(dvp); 1704 lwkt_gettoken(&nmp->nm_token); 1705 1706 /* 1707 * Oops, not for me.. 1708 */ 1709 if (vap->va_type == VSOCK) { 1710 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap); 1711 lwkt_reltoken(&nmp->nm_token); 1712 return error; 1713 } 1714 1715 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1716 lwkt_reltoken(&nmp->nm_token); 1717 return (error); 1718 } 1719 if (vap->va_vaflags & VA_EXCLUSIVE) 1720 fmode |= O_EXCL; 1721 again: 1722 nfsstats.rpccnt[NFSPROC_CREATE]++; 1723 nfsm_reqhead(&info, dvp, NFSPROC_CREATE, 1724 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED + 1725 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1726 ERROROUT(nfsm_fhtom(&info, dvp)); 1727 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1728 NFS_MAXNAMLEN)); 1729 if (info.v3) { 1730 tl = nfsm_build(&info, NFSX_UNSIGNED); 1731 if (fmode & O_EXCL) { 1732 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1733 tl = nfsm_build(&info, NFSX_V3CREATEVERF); 1734 #ifdef INET 1735 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1736 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr; 1737 else 1738 #endif 1739 *tl++ = create_verf; 1740 *tl = ++create_verf; 1741 } else { 1742 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1743 nfsm_v3attrbuild(&info, vap, FALSE); 1744 } 1745 } else { 1746 sp = nfsm_build(&info, NFSX_V2SATTR); 1747 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1748 sp->sa_uid = nfs_xdrneg1; 1749 sp->sa_gid = nfs_xdrneg1; 1750 sp->sa_size = 0; 1751 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1752 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1753 } 1754 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td, 1755 cnp->cn_cred, &error)); 1756 if (error == 0) { 1757 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1758 if (!gotvp) { 1759 if (newvp) { 1760 vput(newvp); 1761 newvp = NULL; 1762 } 1763 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1764 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1765 if (!error) 1766 newvp = NFSTOV(np); 1767 } 1768 } 1769 if (info.v3) { 1770 if (error == 0) 1771 error = nfsm_wcc_data(&info, dvp, &wccflag); 1772 else 1773 (void)nfsm_wcc_data(&info, dvp, &wccflag); 1774 } 1775 m_freem(info.mrep); 1776 info.mrep = NULL; 1777 nfsmout: 1778 if (error) { 1779 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1780 KKASSERT(newvp == NULL); 1781 fmode &= ~O_EXCL; 1782 goto again; 1783 } 1784 } else if (info.v3 && (fmode & O_EXCL)) { 1785 /* 1786 * We are normally called with only a partially initialized 1787 * VAP. Since the NFSv3 spec says that server may use the 1788 * file attributes to store the verifier, the spec requires 1789 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1790 * in atime, but we can't really assume that all servers will 1791 * so we ensure that our SETATTR sets both atime and mtime. 1792 */ 1793 if (vap->va_mtime.tv_sec == VNOVAL) 1794 vfs_timestamp(&vap->va_mtime); 1795 if (vap->va_atime.tv_sec == VNOVAL) 1796 vap->va_atime = vap->va_mtime; 1797 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1798 } 1799 if (error == 0) { 1800 /* 1801 * The new np may have enough info for access 1802 * checks, make sure rucred and wucred are 1803 * initialized for read and write rpc's. 1804 */ 1805 np = VTONFS(newvp); 1806 if (np->n_rucred == NULL) 1807 np->n_rucred = crhold(cnp->cn_cred); 1808 if (np->n_wucred == NULL) 1809 np->n_wucred = crhold(cnp->cn_cred); 1810 *ap->a_vpp = newvp; 1811 nfs_knote(dvp, NOTE_WRITE); 1812 } else if (newvp) { 1813 vput(newvp); 1814 } 1815 VTONFS(dvp)->n_flag |= NLMODIFIED; 1816 if (!wccflag) 1817 VTONFS(dvp)->n_attrstamp = 0; 1818 lwkt_reltoken(&nmp->nm_token); 1819 return (error); 1820 } 1821 1822 /* 1823 * nfs file remove call 1824 * To try and make nfs semantics closer to ufs semantics, a file that has 1825 * other processes using the vnode is renamed instead of removed and then 1826 * removed later on the last close. 1827 * - If v_refcnt > 1 1828 * If a rename is not already in the works 1829 * call nfs_sillyrename() to set it up 1830 * else 1831 * do the remove rpc 1832 * 1833 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1834 * struct componentname *a_cnp) 1835 */ 1836 static int 1837 nfs_remove(struct vop_old_remove_args *ap) 1838 { 1839 struct vnode *vp = ap->a_vp; 1840 struct vnode *dvp = ap->a_dvp; 1841 struct nfsmount *nmp = VFSTONFS(dvp->v_mount); 1842 struct componentname *cnp = ap->a_cnp; 1843 struct nfsnode *np = VTONFS(vp); 1844 int error = 0; 1845 struct vattr vattr; 1846 1847 lwkt_gettoken(&nmp->nm_token); 1848 #ifndef DIAGNOSTIC 1849 if (VREFCNT(vp) < 1) 1850 panic("nfs_remove: bad v_refcnt"); 1851 #endif 1852 if (vp->v_type == VDIR) { 1853 error = EPERM; 1854 } else if (VREFCNT(vp) == 1 || (np->n_sillyrename && 1855 VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) { 1856 /* 1857 * Force finalization so the VOP_INACTIVE() call is not delayed. 1858 * This prevents cred structures from building up in nfsnodes 1859 * for deleted files. 1860 */ 1861 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1862 np->n_flag |= NREMOVED; 1863 1864 /* 1865 * Throw away biocache buffers, mainly to avoid 1866 * unnecessary delayed writes later. 1867 */ 1868 error = nfs_vinvalbuf(vp, 0, 1); 1869 /* Do the rpc */ 1870 if (error != EINTR) { 1871 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1872 cnp->cn_namelen, 1873 cnp->cn_cred, cnp->cn_td); 1874 } 1875 1876 /* 1877 * Kludge City: If the first reply to the remove rpc is lost.. 1878 * the reply to the retransmitted request will be ENOENT 1879 * since the file was in fact removed 1880 * Therefore, we cheat and return success. 1881 */ 1882 if (error == ENOENT) 1883 error = 0; 1884 } else if (!np->n_sillyrename) { 1885 error = nfs_sillyrename(dvp, vp, cnp); 1886 } 1887 np->n_attrstamp = 0; 1888 lwkt_reltoken(&nmp->nm_token); 1889 if (error == 0) { 1890 nfs_knote(vp, NOTE_DELETE); 1891 nfs_knote(dvp, NOTE_WRITE); 1892 } 1893 1894 return (error); 1895 } 1896 1897 /* 1898 * nfs file remove rpc called from nfs_inactive 1899 * 1900 * NOTE: s_dvp can be VBAD during a forced unmount. 1901 */ 1902 int 1903 nfs_removeit(struct sillyrename *sp) 1904 { 1905 if (sp->s_dvp->v_type == VBAD) 1906 return(0); 1907 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1908 sp->s_cred, NULL)); 1909 } 1910 1911 /* 1912 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1913 */ 1914 static int 1915 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1916 struct ucred *cred, struct thread *td) 1917 { 1918 int error = 0, wccflag = NFSV3_WCCRATTR; 1919 struct nfsm_info info; 1920 1921 info.mrep = NULL; 1922 info.v3 = NFS_ISV3(dvp); 1923 1924 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1925 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE, 1926 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1927 ERROROUT(nfsm_fhtom(&info, dvp)); 1928 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN)); 1929 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error)); 1930 if (info.v3) { 1931 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1932 } 1933 m_freem(info.mrep); 1934 info.mrep = NULL; 1935 nfsmout: 1936 VTONFS(dvp)->n_flag |= NLMODIFIED; 1937 if (!wccflag) 1938 VTONFS(dvp)->n_attrstamp = 0; 1939 return (error); 1940 } 1941 1942 /* 1943 * nfs file rename call 1944 * 1945 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1946 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1947 * struct vnode *a_tvp, struct componentname *a_tcnp) 1948 */ 1949 static int 1950 nfs_rename(struct vop_old_rename_args *ap) 1951 { 1952 struct vnode *fvp = ap->a_fvp; 1953 struct vnode *tvp = ap->a_tvp; 1954 struct vnode *fdvp = ap->a_fdvp; 1955 struct vnode *tdvp = ap->a_tdvp; 1956 struct componentname *tcnp = ap->a_tcnp; 1957 struct componentname *fcnp = ap->a_fcnp; 1958 struct nfsmount *nmp = VFSTONFS(fdvp->v_mount); 1959 int error; 1960 1961 lwkt_gettoken(&nmp->nm_token); 1962 1963 /* 1964 * Force finalization so the VOP_INACTIVE() call is not delayed. 1965 * This prevents cred structures from building up in nfsnodes 1966 * for deleted files. 1967 */ 1968 if (tvp) { 1969 atomic_set_int(&tvp->v_refcnt, VREF_FINALIZE); 1970 if (VTONFS(tvp)) 1971 VTONFS(tvp)->n_flag |= NREMOVED; 1972 } 1973 1974 /* Check for cross-device rename */ 1975 if ((fvp->v_mount != tdvp->v_mount) || 1976 (tvp && (fvp->v_mount != tvp->v_mount))) { 1977 error = EXDEV; 1978 goto out; 1979 } 1980 1981 /* 1982 * We shouldn't have to flush fvp on rename for most server-side 1983 * filesystems as the file handle should not change. Unfortunately 1984 * the inode for some filesystems (msdosfs) might be tied to the 1985 * file name or directory position so to be completely safe 1986 * vfs.nfs.flush_on_rename is set by default. Clear to improve 1987 * performance. 1988 * 1989 * We must flush tvp on rename because it might become stale on the 1990 * server after the rename. 1991 */ 1992 if (nfs_flush_on_rename) 1993 VOP_FSYNC(fvp, MNT_WAIT, 0); 1994 if (tvp) 1995 VOP_FSYNC(tvp, MNT_WAIT, 0); 1996 1997 /* 1998 * If the tvp exists and is in use, sillyrename it before doing the 1999 * rename of the new file over it. 2000 * 2001 * XXX Can't sillyrename a directory. 2002 * 2003 * We do not attempt to do any namecache purges in this old API 2004 * routine. The new API compat functions have access to the actual 2005 * namecache structures and will do it for us. 2006 */ 2007 if (tvp && VREFCNT(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && 2008 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 2009 nfs_knote(tvp, NOTE_DELETE); 2010 vput(tvp); 2011 tvp = NULL; 2012 } else if (tvp) { 2013 nfs_knote(tvp, NOTE_DELETE); 2014 } 2015 2016 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 2017 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 2018 tcnp->cn_td); 2019 2020 out: 2021 if (error == 0) { 2022 nfs_knote(fdvp, NOTE_WRITE); 2023 nfs_knote(tdvp, NOTE_WRITE); 2024 nfs_knote(fvp, NOTE_RENAME); 2025 } 2026 lwkt_reltoken(&nmp->nm_token); 2027 if (tdvp == tvp) 2028 vrele(tdvp); 2029 else 2030 vput(tdvp); 2031 if (tvp) 2032 vput(tvp); 2033 vrele(fdvp); 2034 vrele(fvp); 2035 /* 2036 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 2037 */ 2038 if (error == ENOENT) 2039 error = 0; 2040 return (error); 2041 } 2042 2043 /* 2044 * nfs file rename rpc called from nfs_remove() above 2045 */ 2046 static int 2047 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 2048 struct sillyrename *sp) 2049 { 2050 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 2051 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 2052 } 2053 2054 /* 2055 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 2056 */ 2057 static int 2058 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 2059 struct vnode *tdvp, const char *tnameptr, int tnamelen, 2060 struct ucred *cred, struct thread *td) 2061 { 2062 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 2063 struct nfsm_info info; 2064 2065 info.mrep = NULL; 2066 info.v3 = NFS_ISV3(fdvp); 2067 2068 nfsstats.rpccnt[NFSPROC_RENAME]++; 2069 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME, 2070 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 + 2071 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen)); 2072 ERROROUT(nfsm_fhtom(&info, fdvp)); 2073 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN)); 2074 ERROROUT(nfsm_fhtom(&info, tdvp)); 2075 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN)); 2076 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error)); 2077 if (info.v3) { 2078 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag)); 2079 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag)); 2080 } 2081 m_freem(info.mrep); 2082 info.mrep = NULL; 2083 nfsmout: 2084 VTONFS(fdvp)->n_flag |= NLMODIFIED; 2085 VTONFS(tdvp)->n_flag |= NLMODIFIED; 2086 if (!fwccflag) 2087 VTONFS(fdvp)->n_attrstamp = 0; 2088 if (!twccflag) 2089 VTONFS(tdvp)->n_attrstamp = 0; 2090 return (error); 2091 } 2092 2093 /* 2094 * nfs hard link create call 2095 * 2096 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 2097 * struct componentname *a_cnp) 2098 */ 2099 static int 2100 nfs_link(struct vop_old_link_args *ap) 2101 { 2102 struct vnode *vp = ap->a_vp; 2103 struct vnode *tdvp = ap->a_tdvp; 2104 struct nfsmount *nmp = VFSTONFS(tdvp->v_mount); 2105 struct componentname *cnp = ap->a_cnp; 2106 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 2107 struct nfsm_info info; 2108 2109 if (vp->v_mount != tdvp->v_mount) { 2110 return (EXDEV); 2111 } 2112 lwkt_gettoken(&nmp->nm_token); 2113 2114 /* 2115 * The attribute cache may get out of sync with the server on link. 2116 * Pushing writes to the server before handle was inherited from 2117 * long long ago and it is unclear if we still need to do this. 2118 * Defaults to off. 2119 */ 2120 if (nfs_flush_on_hlink) 2121 VOP_FSYNC(vp, MNT_WAIT, 0); 2122 2123 info.mrep = NULL; 2124 info.v3 = NFS_ISV3(vp); 2125 2126 nfsstats.rpccnt[NFSPROC_LINK]++; 2127 nfsm_reqhead(&info, vp, NFSPROC_LINK, 2128 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED + 2129 nfsm_rndup(cnp->cn_namelen)); 2130 ERROROUT(nfsm_fhtom(&info, vp)); 2131 ERROROUT(nfsm_fhtom(&info, tdvp)); 2132 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2133 NFS_MAXNAMLEN)); 2134 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td, 2135 cnp->cn_cred, &error)); 2136 if (info.v3) { 2137 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2138 NFS_LATTR_NOSHRINK)); 2139 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag)); 2140 } 2141 m_freem(info.mrep); 2142 info.mrep = NULL; 2143 nfsmout: 2144 VTONFS(tdvp)->n_flag |= NLMODIFIED; 2145 if (!attrflag) 2146 VTONFS(vp)->n_attrstamp = 0; 2147 if (!wccflag) 2148 VTONFS(tdvp)->n_attrstamp = 0; 2149 /* 2150 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 2151 */ 2152 if (error == EEXIST) 2153 error = 0; 2154 lwkt_reltoken(&nmp->nm_token); 2155 if (error == 0) { 2156 nfs_knote(vp, NOTE_LINK); 2157 nfs_knote(tdvp, NOTE_WRITE); 2158 } 2159 2160 return (error); 2161 } 2162 2163 /* 2164 * nfs symbolic link create call 2165 * 2166 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 2167 * struct componentname *a_cnp, struct vattr *a_vap, 2168 * char *a_target) 2169 */ 2170 static int 2171 nfs_symlink(struct vop_old_symlink_args *ap) 2172 { 2173 struct vnode *dvp = ap->a_dvp; 2174 struct vattr *vap = ap->a_vap; 2175 struct nfsmount *nmp = VFSTONFS(dvp->v_mount); 2176 struct componentname *cnp = ap->a_cnp; 2177 struct nfsv2_sattr *sp; 2178 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 2179 struct vnode *newvp = NULL; 2180 struct nfsm_info info; 2181 2182 info.mrep = NULL; 2183 info.v3 = NFS_ISV3(dvp); 2184 lwkt_gettoken(&nmp->nm_token); 2185 2186 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 2187 slen = strlen(ap->a_target); 2188 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK, 2189 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED + 2190 nfsm_rndup(cnp->cn_namelen) + 2191 nfsm_rndup(slen) + NFSX_SATTR(info.v3)); 2192 ERROROUT(nfsm_fhtom(&info, dvp)); 2193 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2194 NFS_MAXNAMLEN)); 2195 if (info.v3) { 2196 nfsm_v3attrbuild(&info, vap, FALSE); 2197 } 2198 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN)); 2199 if (info.v3 == 0) { 2200 sp = nfsm_build(&info, NFSX_V2SATTR); 2201 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 2202 sp->sa_uid = nfs_xdrneg1; 2203 sp->sa_gid = nfs_xdrneg1; 2204 sp->sa_size = nfs_xdrneg1; 2205 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2206 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2207 } 2208 2209 /* 2210 * Issue the NFS request and get the rpc response. 2211 * 2212 * Only NFSv3 responses returning an error of 0 actually return 2213 * a file handle that can be converted into newvp without having 2214 * to do an extra lookup rpc. 2215 */ 2216 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td, 2217 cnp->cn_cred, &error)); 2218 if (info.v3) { 2219 if (error == 0) { 2220 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2221 } 2222 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2223 } 2224 2225 /* 2226 * out code jumps -> here, mrep is also freed. 2227 */ 2228 2229 m_freem(info.mrep); 2230 info.mrep = NULL; 2231 nfsmout: 2232 2233 /* 2234 * If we get an EEXIST error, silently convert it to no-error 2235 * in case of an NFS retry. 2236 */ 2237 if (error == EEXIST) 2238 error = 0; 2239 2240 /* 2241 * If we do not have (or no longer have) an error, and we could 2242 * not extract the newvp from the response due to the request being 2243 * NFSv2 or the error being EEXIST. We have to do a lookup in order 2244 * to obtain a newvp to return. 2245 */ 2246 if (error == 0 && newvp == NULL) { 2247 struct nfsnode *np = NULL; 2248 2249 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2250 cnp->cn_cred, cnp->cn_td, &np); 2251 if (!error) 2252 newvp = NFSTOV(np); 2253 } 2254 if (error) { 2255 if (newvp) 2256 vput(newvp); 2257 } else { 2258 *ap->a_vpp = newvp; 2259 } 2260 VTONFS(dvp)->n_flag |= NLMODIFIED; 2261 if (!wccflag) 2262 VTONFS(dvp)->n_attrstamp = 0; 2263 if (error == 0 && *ap->a_vpp) 2264 nfs_knote(*ap->a_vpp, NOTE_WRITE); 2265 lwkt_reltoken(&nmp->nm_token); 2266 2267 return (error); 2268 } 2269 2270 /* 2271 * nfs make dir call 2272 * 2273 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2274 * struct componentname *a_cnp, struct vattr *a_vap) 2275 */ 2276 static int 2277 nfs_mkdir(struct vop_old_mkdir_args *ap) 2278 { 2279 struct vnode *dvp = ap->a_dvp; 2280 struct vattr *vap = ap->a_vap; 2281 struct nfsmount *nmp = VFSTONFS(dvp->v_mount); 2282 struct componentname *cnp = ap->a_cnp; 2283 struct nfsv2_sattr *sp; 2284 struct nfsnode *np = NULL; 2285 struct vnode *newvp = NULL; 2286 struct vattr vattr; 2287 int error = 0, wccflag = NFSV3_WCCRATTR; 2288 int gotvp = 0; 2289 int len; 2290 struct nfsm_info info; 2291 2292 info.mrep = NULL; 2293 info.v3 = NFS_ISV3(dvp); 2294 lwkt_gettoken(&nmp->nm_token); 2295 2296 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 2297 lwkt_reltoken(&nmp->nm_token); 2298 return (error); 2299 } 2300 len = cnp->cn_namelen; 2301 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2302 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR, 2303 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2304 nfsm_rndup(len) + NFSX_SATTR(info.v3)); 2305 ERROROUT(nfsm_fhtom(&info, dvp)); 2306 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 2307 if (info.v3) { 2308 nfsm_v3attrbuild(&info, vap, FALSE); 2309 } else { 2310 sp = nfsm_build(&info, NFSX_V2SATTR); 2311 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2312 sp->sa_uid = nfs_xdrneg1; 2313 sp->sa_gid = nfs_xdrneg1; 2314 sp->sa_size = nfs_xdrneg1; 2315 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2316 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2317 } 2318 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td, 2319 cnp->cn_cred, &error)); 2320 if (error == 0) { 2321 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2322 } 2323 if (info.v3) { 2324 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2325 } 2326 m_freem(info.mrep); 2327 info.mrep = NULL; 2328 nfsmout: 2329 VTONFS(dvp)->n_flag |= NLMODIFIED; 2330 if (!wccflag) 2331 VTONFS(dvp)->n_attrstamp = 0; 2332 /* 2333 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2334 * if we can succeed in looking up the directory. 2335 */ 2336 if (error == EEXIST || (!error && !gotvp)) { 2337 if (newvp) { 2338 vrele(newvp); 2339 newvp = NULL; 2340 } 2341 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2342 cnp->cn_td, &np); 2343 if (!error) { 2344 newvp = NFSTOV(np); 2345 if (newvp->v_type != VDIR) 2346 error = EEXIST; 2347 } 2348 } 2349 if (error) { 2350 if (newvp) 2351 vrele(newvp); 2352 } else { 2353 nfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 2354 *ap->a_vpp = newvp; 2355 } 2356 lwkt_reltoken(&nmp->nm_token); 2357 return (error); 2358 } 2359 2360 /* 2361 * nfs remove directory call 2362 * 2363 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2364 * struct componentname *a_cnp) 2365 */ 2366 static int 2367 nfs_rmdir(struct vop_old_rmdir_args *ap) 2368 { 2369 struct vnode *vp = ap->a_vp; 2370 struct vnode *dvp = ap->a_dvp; 2371 struct nfsmount *nmp = VFSTONFS(dvp->v_mount); 2372 struct componentname *cnp = ap->a_cnp; 2373 int error = 0, wccflag = NFSV3_WCCRATTR; 2374 struct nfsm_info info; 2375 2376 info.mrep = NULL; 2377 info.v3 = NFS_ISV3(dvp); 2378 2379 if (dvp == vp) 2380 return (EINVAL); 2381 2382 lwkt_gettoken(&nmp->nm_token); 2383 2384 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2385 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR, 2386 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2387 nfsm_rndup(cnp->cn_namelen)); 2388 ERROROUT(nfsm_fhtom(&info, dvp)); 2389 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2390 NFS_MAXNAMLEN)); 2391 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td, 2392 cnp->cn_cred, &error)); 2393 if (info.v3) { 2394 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2395 } 2396 m_freem(info.mrep); 2397 info.mrep = NULL; 2398 nfsmout: 2399 VTONFS(dvp)->n_flag |= NLMODIFIED; 2400 if (!wccflag) 2401 VTONFS(dvp)->n_attrstamp = 0; 2402 /* 2403 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2404 */ 2405 if (error == ENOENT) 2406 error = 0; 2407 else 2408 nfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 2409 lwkt_reltoken(&nmp->nm_token); 2410 2411 return (error); 2412 } 2413 2414 /* 2415 * nfs readdir call 2416 * 2417 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2418 */ 2419 static int 2420 nfs_readdir(struct vop_readdir_args *ap) 2421 { 2422 struct vnode *vp = ap->a_vp; 2423 struct nfsnode *np = VTONFS(vp); 2424 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2425 struct uio *uio = ap->a_uio; 2426 int tresid, error; 2427 struct vattr vattr; 2428 2429 if (vp->v_type != VDIR) 2430 return (EPERM); 2431 2432 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 2433 if (error) 2434 return (error); 2435 2436 lwkt_gettoken(&nmp->nm_token); 2437 2438 /* 2439 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2440 * and then check that is still valid, or if this is an NQNFS mount 2441 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2442 * VOP_GETATTR() does not necessarily go to the wire. 2443 */ 2444 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2445 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2446 if (VOP_GETATTR(vp, &vattr) == 0 && 2447 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2448 ) { 2449 nfsstats.direofcache_hits++; 2450 goto done; 2451 } 2452 } 2453 2454 /* 2455 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2456 * own cache coherency checks so we do not have to. 2457 */ 2458 tresid = uio->uio_resid; 2459 error = nfs_bioread(vp, uio, 0); 2460 2461 if (!error && uio->uio_resid == tresid) 2462 nfsstats.direofcache_misses++; 2463 done: 2464 lwkt_reltoken(&nmp->nm_token); 2465 vn_unlock(vp); 2466 2467 return (error); 2468 } 2469 2470 /* 2471 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2472 * 2473 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2474 * offset/block and converts the nfs formatted directory entries for userland 2475 * consumption as well as deals with offsets into the middle of blocks. 2476 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2477 * be block-bounded. It must convert to cookies for the actual RPC. 2478 */ 2479 int 2480 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop) 2481 { 2482 int len, left; 2483 struct nfs_dirent *dp = NULL; 2484 u_int32_t *tl; 2485 nfsuint64 *cookiep; 2486 caddr_t cp; 2487 nfsuint64 cookie; 2488 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2489 struct nfsnode *dnp = VTONFS(vp); 2490 u_quad_t fileno; 2491 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2492 int attrflag; 2493 struct nfsm_info info; 2494 2495 info.mrep = NULL; 2496 info.v3 = NFS_ISV3(vp); 2497 2498 #ifndef DIAGNOSTIC 2499 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2500 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2501 panic("nfs readdirrpc bad uio"); 2502 #endif 2503 2504 /* 2505 * If there is no cookie, assume directory was stale. 2506 */ 2507 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2508 if (cookiep) 2509 cookie = *cookiep; 2510 else 2511 return (NFSERR_BAD_COOKIE); 2512 /* 2513 * Loop around doing readdir rpc's of size nm_readdirsize 2514 * truncated to a multiple of DIRBLKSIZ. 2515 * The stopping criteria is EOF or buffer full. 2516 */ 2517 while (more_dirs && bigenough) { 2518 nfsstats.rpccnt[NFSPROC_READDIR]++; 2519 nfsm_reqhead(&info, vp, NFSPROC_READDIR, 2520 NFSX_FH(info.v3) + NFSX_READDIR(info.v3)); 2521 ERROROUT(nfsm_fhtom(&info, vp)); 2522 if (info.v3) { 2523 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 2524 *tl++ = cookie.nfsuquad[0]; 2525 *tl++ = cookie.nfsuquad[1]; 2526 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2527 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2528 } else { 2529 /* 2530 * WARNING! HAMMER DIRECTORIES WILL NOT WORK WELL 2531 * WITH NFSv2!!! There's nothing I can really do 2532 * about it other than to hope the server supports 2533 * rdirplus w/NFSv2. 2534 */ 2535 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 2536 *tl++ = cookie.nfsuquad[0]; 2537 } 2538 *tl = txdr_unsigned(nmp->nm_readdirsize); 2539 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR, 2540 uiop->uio_td, 2541 nfs_vpcred(vp, ND_READ), &error)); 2542 if (info.v3) { 2543 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2544 NFS_LATTR_NOSHRINK)); 2545 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2546 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2547 dnp->n_cookieverf.nfsuquad[1] = *tl; 2548 } 2549 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2550 more_dirs = fxdr_unsigned(int, *tl); 2551 2552 /* loop thru the dir entries, converting them to std form */ 2553 while (more_dirs && bigenough) { 2554 if (info.v3) { 2555 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2556 fileno = fxdr_hyper(tl); 2557 len = fxdr_unsigned(int, *(tl + 2)); 2558 } else { 2559 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2560 fileno = fxdr_unsigned(u_quad_t, *tl++); 2561 len = fxdr_unsigned(int, *tl); 2562 } 2563 if (len <= 0 || len > NFS_MAXNAMLEN) { 2564 error = EBADRPC; 2565 m_freem(info.mrep); 2566 info.mrep = NULL; 2567 goto nfsmout; 2568 } 2569 2570 /* 2571 * len is the number of bytes in the path element 2572 * name, not including the \0 termination. 2573 * 2574 * tlen is the number of bytes w have to reserve for 2575 * the path element name. 2576 */ 2577 tlen = nfsm_rndup(len); 2578 if (tlen == len) 2579 tlen += 4; /* To ensure null termination */ 2580 2581 /* 2582 * If the entry would cross a DIRBLKSIZ boundary, 2583 * extend the previous nfs_dirent to cover the 2584 * remaining space. 2585 */ 2586 left = DIRBLKSIZ - blksiz; 2587 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2588 dp->nfs_reclen += left; 2589 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2590 uiop->uio_iov->iov_len -= left; 2591 uiop->uio_offset += left; 2592 uiop->uio_resid -= left; 2593 blksiz = 0; 2594 } 2595 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2596 bigenough = 0; 2597 if (bigenough) { 2598 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2599 dp->nfs_ino = fileno; 2600 dp->nfs_namlen = len; 2601 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2602 dp->nfs_type = DT_UNKNOWN; 2603 blksiz += dp->nfs_reclen; 2604 if (blksiz == DIRBLKSIZ) 2605 blksiz = 0; 2606 uiop->uio_offset += sizeof(struct nfs_dirent); 2607 uiop->uio_resid -= sizeof(struct nfs_dirent); 2608 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2609 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2610 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2611 2612 /* 2613 * The uiop has advanced by nfs_dirent + len 2614 * but really needs to advance by 2615 * nfs_dirent + tlen 2616 */ 2617 cp = uiop->uio_iov->iov_base; 2618 tlen -= len; 2619 *cp = '\0'; /* null terminate */ 2620 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2621 uiop->uio_iov->iov_len -= tlen; 2622 uiop->uio_offset += tlen; 2623 uiop->uio_resid -= tlen; 2624 } else { 2625 /* 2626 * NFS strings must be rounded up (nfsm_myouio 2627 * handled that in the bigenough case). 2628 */ 2629 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2630 } 2631 if (info.v3) { 2632 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2633 } else { 2634 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2635 } 2636 2637 /* 2638 * If we were able to accomodate the last entry, 2639 * get the cookie for the next one. Otherwise 2640 * hold-over the cookie for the one we were not 2641 * able to accomodate. 2642 */ 2643 if (bigenough) { 2644 cookie.nfsuquad[0] = *tl++; 2645 if (info.v3) 2646 cookie.nfsuquad[1] = *tl++; 2647 } else if (info.v3) { 2648 tl += 2; 2649 } else { 2650 tl++; 2651 } 2652 more_dirs = fxdr_unsigned(int, *tl); 2653 } 2654 /* 2655 * If at end of rpc data, get the eof boolean 2656 */ 2657 if (!more_dirs) { 2658 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2659 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2660 } 2661 m_freem(info.mrep); 2662 info.mrep = NULL; 2663 } 2664 /* 2665 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2666 * by increasing d_reclen for the last record. 2667 */ 2668 if (blksiz > 0) { 2669 left = DIRBLKSIZ - blksiz; 2670 dp->nfs_reclen += left; 2671 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2672 uiop->uio_iov->iov_len -= left; 2673 uiop->uio_offset += left; 2674 uiop->uio_resid -= left; 2675 } 2676 2677 if (bigenough) { 2678 /* 2679 * We hit the end of the directory, update direofoffset. 2680 */ 2681 dnp->n_direofoffset = uiop->uio_offset; 2682 } else { 2683 /* 2684 * There is more to go, insert the link cookie so the 2685 * next block can be read. 2686 */ 2687 if (uiop->uio_resid > 0) 2688 kprintf("EEK! readdirrpc resid > 0\n"); 2689 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2690 *cookiep = cookie; 2691 } 2692 nfsmout: 2693 return (error); 2694 } 2695 2696 /* 2697 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2698 */ 2699 int 2700 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop) 2701 { 2702 int len, left; 2703 struct nfs_dirent *dp; 2704 u_int32_t *tl; 2705 struct vnode *newvp; 2706 nfsuint64 *cookiep; 2707 caddr_t dpossav1, dpossav2; 2708 caddr_t cp; 2709 struct mbuf *mdsav1, *mdsav2; 2710 nfsuint64 cookie; 2711 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2712 struct nfsnode *dnp = VTONFS(vp), *np; 2713 nfsfh_t *fhp; 2714 u_quad_t fileno; 2715 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2716 int attrflag, fhsize; 2717 struct nchandle nch; 2718 struct nchandle dnch; 2719 struct nlcomponent nlc; 2720 struct nfsm_info info; 2721 2722 info.mrep = NULL; 2723 info.v3 = 1; 2724 2725 #ifndef nolint 2726 dp = NULL; 2727 #endif 2728 #ifndef DIAGNOSTIC 2729 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2730 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2731 panic("nfs readdirplusrpc bad uio"); 2732 #endif 2733 /* 2734 * Obtain the namecache record for the directory so we have something 2735 * to use as a basis for creating the entries. This function will 2736 * return a held (but not locked) ncp. The ncp may be disconnected 2737 * from the tree and cannot be used for upward traversals, and the 2738 * ncp may be unnamed. Note that other unrelated operations may 2739 * cause the ncp to be named at any time. 2740 * 2741 * We have to lock the ncp to prevent a lock order reversal when 2742 * rdirplus does nlookups of the children, because the vnode is 2743 * locked and has to stay that way. 2744 */ 2745 cache_fromdvp(vp, NULL, 0, &dnch); 2746 bzero(&nlc, sizeof(nlc)); 2747 newvp = NULLVP; 2748 2749 /* 2750 * If there is no cookie, assume directory was stale. 2751 */ 2752 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2753 if (cookiep) { 2754 cookie = *cookiep; 2755 } else { 2756 if (dnch.ncp) 2757 cache_drop(&dnch); 2758 return (NFSERR_BAD_COOKIE); 2759 } 2760 2761 /* 2762 * Loop around doing readdir rpc's of size nm_readdirsize 2763 * truncated to a multiple of DIRBLKSIZ. 2764 * The stopping criteria is EOF or buffer full. 2765 */ 2766 while (more_dirs && bigenough) { 2767 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2768 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS, 2769 NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED); 2770 ERROROUT(nfsm_fhtom(&info, vp)); 2771 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED); 2772 *tl++ = cookie.nfsuquad[0]; 2773 *tl++ = cookie.nfsuquad[1]; 2774 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2775 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2776 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2777 *tl = txdr_unsigned(nmp->nm_rsize); 2778 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS, 2779 uiop->uio_td, 2780 nfs_vpcred(vp, ND_READ), &error)); 2781 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2782 NFS_LATTR_NOSHRINK)); 2783 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2784 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2785 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2786 more_dirs = fxdr_unsigned(int, *tl); 2787 2788 /* loop thru the dir entries, doctoring them to 4bsd form */ 2789 while (more_dirs && bigenough) { 2790 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2791 fileno = fxdr_hyper(tl); 2792 len = fxdr_unsigned(int, *(tl + 2)); 2793 if (len <= 0 || len > NFS_MAXNAMLEN) { 2794 error = EBADRPC; 2795 m_freem(info.mrep); 2796 info.mrep = NULL; 2797 goto nfsmout; 2798 } 2799 tlen = nfsm_rndup(len); 2800 if (tlen == len) 2801 tlen += 4; /* To ensure null termination*/ 2802 left = DIRBLKSIZ - blksiz; 2803 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2804 dp->nfs_reclen += left; 2805 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2806 uiop->uio_iov->iov_len -= left; 2807 uiop->uio_offset += left; 2808 uiop->uio_resid -= left; 2809 blksiz = 0; 2810 } 2811 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2812 bigenough = 0; 2813 if (bigenough) { 2814 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2815 dp->nfs_ino = fileno; 2816 dp->nfs_namlen = len; 2817 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2818 dp->nfs_type = DT_UNKNOWN; 2819 blksiz += dp->nfs_reclen; 2820 if (blksiz == DIRBLKSIZ) 2821 blksiz = 0; 2822 uiop->uio_offset += sizeof(struct nfs_dirent); 2823 uiop->uio_resid -= sizeof(struct nfs_dirent); 2824 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2825 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2826 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2827 nlc.nlc_namelen = len; 2828 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2829 cp = uiop->uio_iov->iov_base; 2830 tlen -= len; 2831 *cp = '\0'; 2832 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2833 uiop->uio_iov->iov_len -= tlen; 2834 uiop->uio_offset += tlen; 2835 uiop->uio_resid -= tlen; 2836 } else { 2837 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2838 } 2839 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2840 if (bigenough) { 2841 cookie.nfsuquad[0] = *tl++; 2842 cookie.nfsuquad[1] = *tl++; 2843 } else { 2844 tl += 2; 2845 } 2846 2847 /* 2848 * Since the attributes are before the file handle 2849 * (sigh), we must skip over the attributes and then 2850 * come back and get them. 2851 */ 2852 attrflag = fxdr_unsigned(int, *tl); 2853 if (attrflag) { 2854 dpossav1 = info.dpos; 2855 mdsav1 = info.md; 2856 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR)); 2857 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2858 doit = fxdr_unsigned(int, *tl); 2859 if (doit) { 2860 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 2861 } 2862 if (doit && bigenough && !nlcdegenerate(&nlc) && 2863 !NFS_CMPFH(dnp, fhp, fhsize) 2864 ) { 2865 if (dnch.ncp) { 2866 #if 0 2867 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n", 2868 nlc.nlc_namelen, nlc.nlc_namelen, 2869 nlc.nlc_nameptr); 2870 #endif 2871 /* 2872 * This is a bit hokey but there isn't 2873 * much we can do about it. We can't 2874 * hold the directory vp locked while 2875 * doing lookups and gets. 2876 */ 2877 nch = cache_nlookup_nonblock(&dnch, &nlc); 2878 if (nch.ncp == NULL) 2879 goto rdfail; 2880 cache_setunresolved(&nch); 2881 error = nfs_nget_nonblock(vp->v_mount, fhp, 2882 fhsize, &np, 2883 NULL); 2884 if (error) { 2885 cache_put(&nch); 2886 goto rdfail; 2887 } 2888 newvp = NFSTOV(np); 2889 dpossav2 = info.dpos; 2890 info.dpos = dpossav1; 2891 mdsav2 = info.md; 2892 info.md = mdsav1; 2893 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 2894 info.dpos = dpossav2; 2895 info.md = mdsav2; 2896 dp->nfs_type = 2897 IFTODT(VTTOIF(np->n_vattr.va_type)); 2898 nfs_cache_setvp(&nch, newvp, 2899 nfspos_cache_timeout); 2900 vput(newvp); 2901 newvp = NULLVP; 2902 cache_put(&nch); 2903 } else { 2904 rdfail: 2905 ; 2906 #if 0 2907 kprintf("Warning: NFS/rddirplus, " 2908 "UNABLE TO ENTER %*.*s\n", 2909 nlc.nlc_namelen, nlc.nlc_namelen, 2910 nlc.nlc_nameptr); 2911 #endif 2912 } 2913 } 2914 } else { 2915 /* Just skip over the file handle */ 2916 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2917 i = fxdr_unsigned(int, *tl); 2918 ERROROUT(nfsm_adv(&info, nfsm_rndup(i))); 2919 } 2920 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2921 more_dirs = fxdr_unsigned(int, *tl); 2922 } 2923 /* 2924 * If at end of rpc data, get the eof boolean 2925 */ 2926 if (!more_dirs) { 2927 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2928 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2929 } 2930 m_freem(info.mrep); 2931 info.mrep = NULL; 2932 } 2933 /* 2934 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2935 * by increasing d_reclen for the last record. 2936 */ 2937 if (blksiz > 0) { 2938 left = DIRBLKSIZ - blksiz; 2939 dp->nfs_reclen += left; 2940 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2941 uiop->uio_iov->iov_len -= left; 2942 uiop->uio_offset += left; 2943 uiop->uio_resid -= left; 2944 } 2945 2946 /* 2947 * We are now either at the end of the directory or have filled the 2948 * block. 2949 */ 2950 if (bigenough) { 2951 dnp->n_direofoffset = uiop->uio_offset; 2952 } else { 2953 if (uiop->uio_resid > 0) 2954 kprintf("EEK! readdirplusrpc resid > 0\n"); 2955 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2956 *cookiep = cookie; 2957 } 2958 nfsmout: 2959 if (newvp != NULLVP) { 2960 if (newvp == vp) 2961 vrele(newvp); 2962 else 2963 vput(newvp); 2964 newvp = NULLVP; 2965 } 2966 if (dnch.ncp) 2967 cache_drop(&dnch); 2968 return (error); 2969 } 2970 2971 /* 2972 * Silly rename. To make the NFS filesystem that is stateless look a little 2973 * more like the "ufs" a remove of an active vnode is translated to a rename 2974 * to a funny looking filename that is removed by nfs_inactive on the 2975 * nfsnode. There is the potential for another process on a different client 2976 * to create the same funny name between the nfs_lookitup() fails and the 2977 * nfs_rename() completes, but... 2978 */ 2979 static int 2980 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2981 { 2982 struct sillyrename *sp; 2983 struct nfsnode *np; 2984 int error; 2985 2986 /* 2987 * Force finalization so the VOP_INACTIVE() call is not delayed. 2988 * This prevents cred structures from building up in nfsnodes 2989 * for deleted files. 2990 */ 2991 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 2992 np = VTONFS(vp); 2993 np->n_flag |= NREMOVED; 2994 2995 /* 2996 * We previously purged dvp instead of vp. I don't know why, it 2997 * completely destroys performance. We can't do it anyway with the 2998 * new VFS API since we would be breaking the namecache topology. 2999 */ 3000 cache_purge(vp); /* XXX */ 3001 #ifndef DIAGNOSTIC 3002 if (vp->v_type == VDIR) 3003 panic("nfs: sillyrename dir"); 3004 #endif 3005 sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK); 3006 sp->s_cred = crdup(cnp->cn_cred); 3007 sp->s_dvp = dvp; 3008 vref(dvp); 3009 3010 /* Fudge together a funny name */ 3011 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", 3012 (int)(intptr_t)cnp->cn_td); 3013 3014 /* Try lookitups until we get one that isn't there */ 3015 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3016 cnp->cn_td, NULL) == 0) { 3017 sp->s_name[4]++; 3018 if (sp->s_name[4] > 'z') { 3019 error = EINVAL; 3020 goto bad; 3021 } 3022 } 3023 error = nfs_renameit(dvp, cnp, sp); 3024 if (error) 3025 goto bad; 3026 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 3027 cnp->cn_td, &np); 3028 np->n_sillyrename = sp; 3029 return (0); 3030 bad: 3031 vrele(sp->s_dvp); 3032 crfree(sp->s_cred); 3033 kfree((caddr_t)sp, M_NFSREQ); 3034 3035 return (error); 3036 } 3037 3038 /* 3039 * Look up a file name and optionally either update the file handle or 3040 * allocate an nfsnode, depending on the value of npp. 3041 * npp == NULL --> just do the lookup 3042 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 3043 * handled too 3044 * *npp != NULL --> update the file handle in the vnode 3045 */ 3046 static int 3047 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 3048 struct thread *td, struct nfsnode **npp) 3049 { 3050 struct vnode *newvp = NULL; 3051 struct nfsnode *np, *dnp = VTONFS(dvp); 3052 int error = 0, fhlen, attrflag; 3053 nfsfh_t *nfhp; 3054 struct nfsm_info info; 3055 3056 info.mrep = NULL; 3057 info.v3 = NFS_ISV3(dvp); 3058 3059 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 3060 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 3061 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 3062 ERROROUT(nfsm_fhtom(&info, dvp)); 3063 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN)); 3064 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error)); 3065 if (npp && !error) { 3066 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp)); 3067 if (*npp) { 3068 np = *npp; 3069 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 3070 kfree((caddr_t)np->n_fhp, M_NFSBIGFH); 3071 np->n_fhp = &np->n_fh; 3072 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 3073 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK); 3074 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 3075 np->n_fhsize = fhlen; 3076 newvp = NFSTOV(np); 3077 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 3078 vref(dvp); 3079 newvp = dvp; 3080 } else { 3081 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, NULL); 3082 if (error) { 3083 m_freem(info.mrep); 3084 info.mrep = NULL; 3085 return (error); 3086 } 3087 newvp = NFSTOV(np); 3088 } 3089 if (info.v3) { 3090 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 3091 NFS_LATTR_NOSHRINK)); 3092 if (!attrflag && *npp == NULL) { 3093 m_freem(info.mrep); 3094 info.mrep = NULL; 3095 if (newvp == dvp) 3096 vrele(newvp); 3097 else 3098 vput(newvp); 3099 return (ENOENT); 3100 } 3101 } else { 3102 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 3103 } 3104 } 3105 m_freem(info.mrep); 3106 info.mrep = NULL; 3107 nfsmout: 3108 if (npp && *npp == NULL) { 3109 if (error) { 3110 if (newvp) { 3111 if (newvp == dvp) 3112 vrele(newvp); 3113 else 3114 vput(newvp); 3115 } 3116 } else 3117 *npp = np; 3118 } 3119 return (error); 3120 } 3121 3122 /* 3123 * Nfs Version 3 commit rpc 3124 * 3125 * We call it 'uio' to distinguish it from 'bio' but there is no real uio 3126 * involved. 3127 */ 3128 int 3129 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 3130 { 3131 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3132 int error = 0, wccflag = NFSV3_WCCRATTR; 3133 struct nfsm_info info; 3134 u_int32_t *tl; 3135 3136 info.mrep = NULL; 3137 info.v3 = 1; 3138 3139 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 3140 return (0); 3141 nfsstats.rpccnt[NFSPROC_COMMIT]++; 3142 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1)); 3143 ERROROUT(nfsm_fhtom(&info, vp)); 3144 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED); 3145 txdr_hyper(offset, tl); 3146 tl += 2; 3147 *tl = txdr_unsigned(cnt); 3148 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td, 3149 nfs_vpcred(vp, ND_WRITE), &error)); 3150 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 3151 if (!error) { 3152 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF)); 3153 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 3154 NFSX_V3WRITEVERF)) { 3155 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 3156 NFSX_V3WRITEVERF); 3157 error = NFSERR_STALEWRITEVERF; 3158 } 3159 } 3160 m_freem(info.mrep); 3161 info.mrep = NULL; 3162 nfsmout: 3163 return (error); 3164 } 3165 3166 /* 3167 * Kludge City.. 3168 * - make nfs_bmap() essentially a no-op that does no translation 3169 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 3170 * (Maybe I could use the process's page mapping, but I was concerned that 3171 * Kernel Write might not be enabled and also figured copyout() would do 3172 * a lot more work than bcopy() and also it currently happens in the 3173 * context of the swapper process (2). 3174 * 3175 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, 3176 * off_t *a_doffsetp, int *a_runp, int *a_runb) 3177 */ 3178 static int 3179 nfs_bmap(struct vop_bmap_args *ap) 3180 { 3181 /* no token lock required */ 3182 if (ap->a_doffsetp != NULL) 3183 *ap->a_doffsetp = ap->a_loffset; 3184 if (ap->a_runp != NULL) 3185 *ap->a_runp = 0; 3186 if (ap->a_runb != NULL) 3187 *ap->a_runb = 0; 3188 return (0); 3189 } 3190 3191 /* 3192 * Strategy routine. 3193 */ 3194 static int 3195 nfs_strategy(struct vop_strategy_args *ap) 3196 { 3197 struct bio *bio = ap->a_bio; 3198 struct bio *nbio; 3199 struct buf *bp __debugvar = bio->bio_buf; 3200 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount); 3201 struct thread *td; 3202 int error; 3203 3204 KASSERT(bp->b_cmd != BUF_CMD_DONE, 3205 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 3206 KASSERT(BUF_LOCKINUSE(bp), 3207 ("nfs_strategy: buffer %p not locked", bp)); 3208 3209 if (bio->bio_flags & BIO_SYNC) 3210 td = curthread; /* XXX */ 3211 else 3212 td = NULL; 3213 3214 lwkt_gettoken(&nmp->nm_token); 3215 3216 /* 3217 * We probably don't need to push an nbio any more since no 3218 * block conversion is required due to the use of 64 bit byte 3219 * offsets, but do it anyway. 3220 * 3221 * NOTE: When NFS callers itself via this strategy routines and 3222 * sets up a synchronous I/O, it expects the I/O to run 3223 * synchronously (its bio_done routine just assumes it), 3224 * so for now we have to honor the bit. 3225 */ 3226 nbio = push_bio(bio); 3227 nbio->bio_offset = bio->bio_offset; 3228 nbio->bio_flags = bio->bio_flags & BIO_SYNC; 3229 3230 /* 3231 * If the op is asynchronous and an i/o daemon is waiting 3232 * queue the request, wake it up and wait for completion 3233 * otherwise just do it ourselves. 3234 */ 3235 if (bio->bio_flags & BIO_SYNC) { 3236 error = nfs_doio(ap->a_vp, nbio, td); 3237 } else { 3238 nfs_asyncio(ap->a_vp, nbio); 3239 error = 0; 3240 } 3241 lwkt_reltoken(&nmp->nm_token); 3242 3243 return (error); 3244 } 3245 3246 /* 3247 * fsync vnode op. Just call nfs_flush() with commit == 1. 3248 * 3249 * nfs_fsync(struct vnode *a_vp, int a_waitfor) 3250 */ 3251 /* ARGSUSED */ 3252 static int 3253 nfs_fsync(struct vop_fsync_args *ap) 3254 { 3255 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount); 3256 int error; 3257 3258 lwkt_gettoken(&nmp->nm_token); 3259 3260 /* 3261 * NOTE: Because attributes are set synchronously we currently 3262 * do not have to implement vsetisdirty()/vclrisdirty(). 3263 */ 3264 error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1); 3265 3266 lwkt_reltoken(&nmp->nm_token); 3267 3268 return error; 3269 } 3270 3271 /* 3272 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 3273 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 3274 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 3275 * set the buffer contains data that has already been written to the server 3276 * and which now needs a commit RPC. 3277 * 3278 * If commit is 0 we only take one pass and only flush buffers containing new 3279 * dirty data. 3280 * 3281 * If commit is 1 we take two passes, issuing a commit RPC in the second 3282 * pass. 3283 * 3284 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 3285 * to completely flush all pending data. 3286 * 3287 * Note that the RB_SCAN code properly handles the case where the 3288 * callback might block and directly or indirectly (another thread) cause 3289 * the RB tree to change. 3290 */ 3291 3292 #ifndef NFS_COMMITBVECSIZ 3293 #define NFS_COMMITBVECSIZ 16 3294 #endif 3295 3296 struct nfs_flush_info { 3297 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 3298 struct thread *td; 3299 struct vnode *vp; 3300 int waitfor; 3301 int slpflag; 3302 int slptimeo; 3303 int loops; 3304 struct buf *bvary[NFS_COMMITBVECSIZ]; 3305 int bvsize; 3306 off_t beg_off; 3307 off_t end_off; 3308 }; 3309 3310 static int nfs_flush_bp(struct buf *bp, void *data); 3311 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 3312 3313 int 3314 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 3315 { 3316 struct nfsnode *np = VTONFS(vp); 3317 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3318 struct nfs_flush_info info; 3319 int error; 3320 3321 bzero(&info, sizeof(info)); 3322 info.td = td; 3323 info.vp = vp; 3324 info.waitfor = waitfor; 3325 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 3326 info.loops = 0; 3327 lwkt_gettoken(&vp->v_token); 3328 3329 do { 3330 /* 3331 * Flush mode 3332 */ 3333 info.mode = NFI_FLUSHNEW; 3334 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3335 nfs_flush_bp, &info); 3336 3337 /* 3338 * Take a second pass if committing and no error occured. 3339 * Clean up any left over collection (whether an error 3340 * occurs or not). 3341 */ 3342 if (commit && error == 0) { 3343 info.mode = NFI_COMMIT; 3344 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3345 nfs_flush_bp, &info); 3346 if (info.bvsize) 3347 error = nfs_flush_docommit(&info, error); 3348 } 3349 3350 /* 3351 * Wait for pending I/O to complete before checking whether 3352 * any further dirty buffers exist. 3353 */ 3354 while (waitfor == MNT_WAIT && 3355 bio_track_active(&vp->v_track_write)) { 3356 error = bio_track_wait(&vp->v_track_write, 3357 info.slpflag, info.slptimeo); 3358 if (error) { 3359 /* 3360 * We have to be able to break out if this 3361 * is an 'intr' mount. 3362 */ 3363 if (nfs_sigintr(nmp, NULL, td)) { 3364 error = -EINTR; 3365 break; 3366 } 3367 3368 /* 3369 * Since we do not process pending signals, 3370 * once we get a PCATCH our tsleep() will no 3371 * longer sleep, switch to a fixed timeout 3372 * instead. 3373 */ 3374 if (info.slpflag == PCATCH) { 3375 info.slpflag = 0; 3376 info.slptimeo = 2 * hz; 3377 } 3378 error = 0; 3379 } 3380 } 3381 ++info.loops; 3382 /* 3383 * Loop if we are flushing synchronous as well as committing, 3384 * and dirty buffers are still present. Otherwise we might livelock. 3385 */ 3386 } while (waitfor == MNT_WAIT && commit && 3387 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3388 3389 /* 3390 * The callbacks have to return a negative error to terminate the 3391 * RB scan. 3392 */ 3393 if (error < 0) 3394 error = -error; 3395 3396 /* 3397 * Deal with any error collection 3398 */ 3399 if (np->n_flag & NWRITEERR) { 3400 error = np->n_error; 3401 np->n_flag &= ~NWRITEERR; 3402 } 3403 lwkt_reltoken(&vp->v_token); 3404 return (error); 3405 } 3406 3407 static 3408 int 3409 nfs_flush_bp(struct buf *bp, void *data) 3410 { 3411 struct nfs_flush_info *info = data; 3412 int lkflags; 3413 int error; 3414 off_t toff; 3415 3416 error = 0; 3417 switch(info->mode) { 3418 case NFI_FLUSHNEW: 3419 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3420 if (error && info->loops && info->waitfor == MNT_WAIT) { 3421 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3422 if (error) { 3423 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3424 if (info->slpflag & PCATCH) 3425 lkflags |= LK_PCATCH; 3426 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3427 info->slptimeo); 3428 } 3429 } 3430 3431 /* 3432 * Ignore locking errors 3433 */ 3434 if (error) { 3435 error = 0; 3436 break; 3437 } 3438 3439 /* 3440 * The buffer may have changed out from under us, even if 3441 * we did not block (MPSAFE). Check again now that it is 3442 * locked. 3443 */ 3444 if (bp->b_vp == info->vp && 3445 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) { 3446 bremfree(bp); 3447 bawrite(bp); 3448 } else { 3449 BUF_UNLOCK(bp); 3450 } 3451 break; 3452 case NFI_COMMIT: 3453 /* 3454 * Only process buffers in need of a commit which we can 3455 * immediately lock. This may prevent a buffer from being 3456 * committed, but the normal flush loop will block on the 3457 * same buffer so we shouldn't get into an endless loop. 3458 */ 3459 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3460 (B_DELWRI | B_NEEDCOMMIT)) { 3461 break; 3462 } 3463 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 3464 break; 3465 3466 /* 3467 * We must recheck after successfully locking the buffer. 3468 */ 3469 if (bp->b_vp != info->vp || 3470 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3471 (B_DELWRI | B_NEEDCOMMIT)) { 3472 BUF_UNLOCK(bp); 3473 break; 3474 } 3475 3476 /* 3477 * NOTE: storing the bp in the bvary[] basically sets 3478 * it up for a commit operation. 3479 * 3480 * We must call vfs_busy_pages() now so the commit operation 3481 * is interlocked with user modifications to memory mapped 3482 * pages. The b_dirtyoff/b_dirtyend range is not correct 3483 * until after the pages have been busied. 3484 * 3485 * Note: to avoid loopback deadlocks, we do not 3486 * assign b_runningbufspace. 3487 */ 3488 bremfree(bp); 3489 bp->b_cmd = BUF_CMD_WRITE; 3490 vfs_busy_pages(bp->b_vp, bp); 3491 info->bvary[info->bvsize] = bp; 3492 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3493 if (info->bvsize == 0 || toff < info->beg_off) 3494 info->beg_off = toff; 3495 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3496 if (info->bvsize == 0 || toff > info->end_off) 3497 info->end_off = toff; 3498 ++info->bvsize; 3499 if (info->bvsize == NFS_COMMITBVECSIZ) { 3500 error = nfs_flush_docommit(info, 0); 3501 KKASSERT(info->bvsize == 0); 3502 } 3503 } 3504 return (error); 3505 } 3506 3507 static 3508 int 3509 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3510 { 3511 struct vnode *vp; 3512 struct buf *bp; 3513 off_t bytes; 3514 int retv; 3515 int i; 3516 3517 vp = info->vp; 3518 3519 if (info->bvsize > 0) { 3520 /* 3521 * Commit data on the server, as required. Note that 3522 * nfs_commit will use the vnode's cred for the commit. 3523 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3524 */ 3525 bytes = info->end_off - info->beg_off; 3526 if (bytes > 0x40000000) 3527 bytes = 0x40000000; 3528 if (error) { 3529 retv = -error; 3530 } else { 3531 retv = nfs_commitrpc_uio(vp, info->beg_off, 3532 (int)bytes, info->td); 3533 if (retv == NFSERR_STALEWRITEVERF) 3534 nfs_clearcommit(vp->v_mount); 3535 } 3536 3537 /* 3538 * Now, either mark the blocks I/O done or mark the 3539 * blocks dirty, depending on whether the commit 3540 * succeeded. 3541 */ 3542 for (i = 0; i < info->bvsize; ++i) { 3543 bp = info->bvary[i]; 3544 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) { 3545 /* 3546 * Either an error or the original 3547 * vfs_busy_pages() cleared B_NEEDCOMMIT 3548 * due to finding new dirty VM pages in 3549 * the buffer. 3550 * 3551 * Leave B_DELWRI intact. 3552 */ 3553 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3554 vfs_unbusy_pages(bp); 3555 bp->b_cmd = BUF_CMD_DONE; 3556 bqrelse(bp); 3557 } else { 3558 /* 3559 * Success, remove B_DELWRI ( bundirty() ). 3560 * 3561 * b_dirtyoff/b_dirtyend seem to be NFS 3562 * specific. We should probably move that 3563 * into bundirty(). XXX 3564 * 3565 * We are faking an I/O write, we have to 3566 * start the transaction in order to 3567 * immediately biodone() it. 3568 */ 3569 bundirty(bp); 3570 bp->b_flags &= ~B_ERROR; 3571 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3572 bp->b_dirtyoff = bp->b_dirtyend = 0; 3573 biodone(&bp->b_bio1); 3574 } 3575 } 3576 info->bvsize = 0; 3577 } 3578 return (error); 3579 } 3580 3581 /* 3582 * NFS advisory byte-level locks. 3583 * Currently unsupported. 3584 * 3585 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3586 * int a_flags) 3587 */ 3588 static int 3589 nfs_advlock(struct vop_advlock_args *ap) 3590 { 3591 struct nfsnode *np = VTONFS(ap->a_vp); 3592 3593 /* no token lock currently required */ 3594 /* 3595 * The following kludge is to allow diskless support to work 3596 * until a real NFS lockd is implemented. Basically, just pretend 3597 * that this is a local lock. 3598 */ 3599 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3600 } 3601 3602 /* 3603 * Print out the contents of an nfsnode. 3604 * 3605 * nfs_print(struct vnode *a_vp) 3606 */ 3607 static int 3608 nfs_print(struct vop_print_args *ap) 3609 { 3610 struct vnode *vp = ap->a_vp; 3611 struct nfsnode *np = VTONFS(vp); 3612 3613 kprintf("tag VT_NFS, fileid %lld fsid 0x%x", 3614 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3615 if (vp->v_type == VFIFO) 3616 fifo_printinfo(vp); 3617 kprintf("\n"); 3618 return (0); 3619 } 3620 3621 /* 3622 * nfs special file access vnode op. 3623 * 3624 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 3625 */ 3626 static int 3627 nfs_laccess(struct vop_access_args *ap) 3628 { 3629 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount); 3630 struct vattr vattr; 3631 int error; 3632 3633 lwkt_gettoken(&nmp->nm_token); 3634 error = VOP_GETATTR(ap->a_vp, &vattr); 3635 if (error == 0) { 3636 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid, 3637 vattr.va_mode, 0); 3638 } 3639 lwkt_reltoken(&nmp->nm_token); 3640 3641 return (error); 3642 } 3643 3644 /* 3645 * Read wrapper for fifos. 3646 * 3647 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3648 * struct ucred *a_cred) 3649 */ 3650 static int 3651 nfsfifo_read(struct vop_read_args *ap) 3652 { 3653 struct nfsnode *np = VTONFS(ap->a_vp); 3654 3655 /* no token access required */ 3656 /* 3657 * Set access flag. 3658 */ 3659 np->n_flag |= NACC; 3660 getnanotime(&np->n_atim); 3661 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3662 } 3663 3664 /* 3665 * Write wrapper for fifos. 3666 * 3667 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3668 * struct ucred *a_cred) 3669 */ 3670 static int 3671 nfsfifo_write(struct vop_write_args *ap) 3672 { 3673 struct nfsnode *np = VTONFS(ap->a_vp); 3674 3675 /* no token access required */ 3676 /* 3677 * Set update flag. 3678 */ 3679 np->n_flag |= NUPD; 3680 getnanotime(&np->n_mtim); 3681 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3682 } 3683 3684 /* 3685 * Close wrapper for fifos. 3686 * 3687 * Update the times on the nfsnode then do fifo close. 3688 * 3689 * nfsfifo_close(struct vnode *a_vp, int a_fflag) 3690 */ 3691 static int 3692 nfsfifo_close(struct vop_close_args *ap) 3693 { 3694 struct vnode *vp = ap->a_vp; 3695 struct nfsnode *np = VTONFS(vp); 3696 struct vattr vattr; 3697 struct timespec ts; 3698 3699 /* no token access required */ 3700 3701 vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */ 3702 if (np->n_flag & (NACC | NUPD)) { 3703 getnanotime(&ts); 3704 if (np->n_flag & NACC) 3705 np->n_atim = ts; 3706 if (np->n_flag & NUPD) 3707 np->n_mtim = ts; 3708 np->n_flag |= NCHG; 3709 if (VREFCNT(vp) == 1 && 3710 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3711 VATTR_NULL(&vattr); 3712 if (np->n_flag & NACC) 3713 vattr.va_atime = np->n_atim; 3714 if (np->n_flag & NUPD) 3715 vattr.va_mtime = np->n_mtim; 3716 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3717 } 3718 } 3719 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3720 } 3721 3722 /************************************************************************ 3723 * KQFILTER OPS * 3724 ************************************************************************/ 3725 3726 static void filt_nfsdetach(struct knote *kn); 3727 static int filt_nfsread(struct knote *kn, long hint); 3728 static int filt_nfswrite(struct knote *kn, long hint); 3729 static int filt_nfsvnode(struct knote *kn, long hint); 3730 3731 static struct filterops nfsread_filtops = 3732 { FILTEROP_ISFD | FILTEROP_MPSAFE, 3733 NULL, filt_nfsdetach, filt_nfsread }; 3734 static struct filterops nfswrite_filtops = 3735 { FILTEROP_ISFD | FILTEROP_MPSAFE, 3736 NULL, filt_nfsdetach, filt_nfswrite }; 3737 static struct filterops nfsvnode_filtops = 3738 { FILTEROP_ISFD | FILTEROP_MPSAFE, 3739 NULL, filt_nfsdetach, filt_nfsvnode }; 3740 3741 static int 3742 nfs_kqfilter (struct vop_kqfilter_args *ap) 3743 { 3744 struct vnode *vp = ap->a_vp; 3745 struct knote *kn = ap->a_kn; 3746 3747 switch (kn->kn_filter) { 3748 case EVFILT_READ: 3749 kn->kn_fop = &nfsread_filtops; 3750 break; 3751 case EVFILT_WRITE: 3752 kn->kn_fop = &nfswrite_filtops; 3753 break; 3754 case EVFILT_VNODE: 3755 kn->kn_fop = &nfsvnode_filtops; 3756 break; 3757 default: 3758 return (EOPNOTSUPP); 3759 } 3760 3761 kn->kn_hook = (caddr_t)vp; 3762 3763 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 3764 3765 return(0); 3766 } 3767 3768 static void 3769 filt_nfsdetach(struct knote *kn) 3770 { 3771 struct vnode *vp = (void *)kn->kn_hook; 3772 3773 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 3774 } 3775 3776 static int 3777 filt_nfsread(struct knote *kn, long hint) 3778 { 3779 struct vnode *vp = (void *)kn->kn_hook; 3780 struct nfsnode *node = VTONFS(vp); 3781 off_t off; 3782 3783 if (hint == NOTE_REVOKE) { 3784 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 3785 return(1); 3786 } 3787 3788 /* 3789 * Interlock against MP races when performing this function. XXX 3790 */ 3791 /* TMPFS_NODE_LOCK_SH(node); */ 3792 off = node->n_size - kn->kn_fp->f_offset; 3793 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 3794 if (kn->kn_sfflags & NOTE_OLDAPI) { 3795 /* TMPFS_NODE_UNLOCK(node); */ 3796 return(1); 3797 } 3798 if (kn->kn_data == 0) { 3799 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 3800 } 3801 /* TMPFS_NODE_UNLOCK(node); */ 3802 return (kn->kn_data != 0); 3803 } 3804 3805 static int 3806 filt_nfswrite(struct knote *kn, long hint) 3807 { 3808 if (hint == NOTE_REVOKE) 3809 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 3810 kn->kn_data = 0; 3811 return (1); 3812 } 3813 3814 static int 3815 filt_nfsvnode(struct knote *kn, long hint) 3816 { 3817 if (kn->kn_sfflags & hint) 3818 kn->kn_fflags |= hint; 3819 if (hint == NOTE_REVOKE) { 3820 kn->kn_flags |= (EV_EOF | EV_NODATA); 3821 return (1); 3822 } 3823 return (kn->kn_fflags != 0); 3824 } 3825