1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 37 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/nlookup.h> 59 #include <sys/socket.h> 60 #include <sys/vnode.h> 61 #include <sys/dirent.h> 62 #include <sys/fcntl.h> 63 #include <sys/lockf.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/conf.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_zone.h> 71 72 #include <sys/buf2.h> 73 74 #include <vfs/fifofs/fifo.h> 75 #include <vfs/ufs/dir.h> 76 77 #undef DIRBLKSIZ 78 79 #include "rpcv2.h" 80 #include "nfsproto.h" 81 #include "nfs.h" 82 #include "nfsmount.h" 83 #include "nfsnode.h" 84 #include "xdr_subs.h" 85 #include "nfsm_subs.h" 86 87 #include <net/if.h> 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 91 #include <sys/thread2.h> 92 93 /* Defs */ 94 #define TRUE 1 95 #define FALSE 0 96 97 static int nfsfifo_read (struct vop_read_args *); 98 static int nfsfifo_write (struct vop_write_args *); 99 static int nfsfifo_close (struct vop_close_args *); 100 #define nfs_poll vop_nopoll 101 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 102 static int nfs_lookup (struct vop_old_lookup_args *); 103 static int nfs_create (struct vop_old_create_args *); 104 static int nfs_mknod (struct vop_old_mknod_args *); 105 static int nfs_open (struct vop_open_args *); 106 static int nfs_close (struct vop_close_args *); 107 static int nfs_access (struct vop_access_args *); 108 static int nfs_getattr (struct vop_getattr_args *); 109 static int nfs_setattr (struct vop_setattr_args *); 110 static int nfs_read (struct vop_read_args *); 111 static int nfs_mmap (struct vop_mmap_args *); 112 static int nfs_fsync (struct vop_fsync_args *); 113 static int nfs_remove (struct vop_old_remove_args *); 114 static int nfs_link (struct vop_old_link_args *); 115 static int nfs_rename (struct vop_old_rename_args *); 116 static int nfs_mkdir (struct vop_old_mkdir_args *); 117 static int nfs_rmdir (struct vop_old_rmdir_args *); 118 static int nfs_symlink (struct vop_old_symlink_args *); 119 static int nfs_readdir (struct vop_readdir_args *); 120 static int nfs_bmap (struct vop_bmap_args *); 121 static int nfs_strategy (struct vop_strategy_args *); 122 static int nfs_lookitup (struct vnode *, const char *, int, 123 struct ucred *, struct thread *, struct nfsnode **); 124 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 125 static int nfs_laccess (struct vop_access_args *); 126 static int nfs_readlink (struct vop_readlink_args *); 127 static int nfs_print (struct vop_print_args *); 128 static int nfs_advlock (struct vop_advlock_args *); 129 130 static int nfs_nresolve (struct vop_nresolve_args *); 131 /* 132 * Global vfs data structures for nfs 133 */ 134 struct vop_ops nfsv2_vnode_vops = { 135 .vop_default = vop_defaultop, 136 .vop_access = nfs_access, 137 .vop_advlock = nfs_advlock, 138 .vop_bmap = nfs_bmap, 139 .vop_close = nfs_close, 140 .vop_old_create = nfs_create, 141 .vop_fsync = nfs_fsync, 142 .vop_getattr = nfs_getattr, 143 .vop_getpages = nfs_getpages, 144 .vop_putpages = nfs_putpages, 145 .vop_inactive = nfs_inactive, 146 .vop_old_link = nfs_link, 147 .vop_old_lookup = nfs_lookup, 148 .vop_old_mkdir = nfs_mkdir, 149 .vop_old_mknod = nfs_mknod, 150 .vop_mmap = nfs_mmap, 151 .vop_open = nfs_open, 152 .vop_poll = nfs_poll, 153 .vop_print = nfs_print, 154 .vop_read = nfs_read, 155 .vop_readdir = nfs_readdir, 156 .vop_readlink = nfs_readlink, 157 .vop_reclaim = nfs_reclaim, 158 .vop_old_remove = nfs_remove, 159 .vop_old_rename = nfs_rename, 160 .vop_old_rmdir = nfs_rmdir, 161 .vop_setattr = nfs_setattr, 162 .vop_strategy = nfs_strategy, 163 .vop_old_symlink = nfs_symlink, 164 .vop_write = nfs_write, 165 .vop_nresolve = nfs_nresolve 166 }; 167 168 /* 169 * Special device vnode ops 170 */ 171 struct vop_ops nfsv2_spec_vops = { 172 .vop_default = vop_defaultop, 173 .vop_access = nfs_laccess, 174 .vop_close = nfs_close, 175 .vop_fsync = nfs_fsync, 176 .vop_getattr = nfs_getattr, 177 .vop_inactive = nfs_inactive, 178 .vop_print = nfs_print, 179 .vop_read = vop_stdnoread, 180 .vop_reclaim = nfs_reclaim, 181 .vop_setattr = nfs_setattr, 182 .vop_write = vop_stdnowrite 183 }; 184 185 struct vop_ops nfsv2_fifo_vops = { 186 .vop_default = fifo_vnoperate, 187 .vop_access = nfs_laccess, 188 .vop_close = nfsfifo_close, 189 .vop_fsync = nfs_fsync, 190 .vop_getattr = nfs_getattr, 191 .vop_inactive = nfs_inactive, 192 .vop_print = nfs_print, 193 .vop_read = nfsfifo_read, 194 .vop_reclaim = nfs_reclaim, 195 .vop_setattr = nfs_setattr, 196 .vop_write = nfsfifo_write 197 }; 198 199 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 200 struct componentname *cnp, 201 struct vattr *vap); 202 static int nfs_removerpc (struct vnode *dvp, const char *name, 203 int namelen, 204 struct ucred *cred, struct thread *td); 205 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 206 int fnamelen, struct vnode *tdvp, 207 const char *tnameptr, int tnamelen, 208 struct ucred *cred, struct thread *td); 209 static int nfs_renameit (struct vnode *sdvp, 210 struct componentname *scnp, 211 struct sillyrename *sp); 212 213 SYSCTL_DECL(_vfs_nfs); 214 215 static int nfs_flush_on_rename = 1; 216 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW, 217 &nfs_flush_on_rename, 0, "flush fvp prior to rename"); 218 static int nfs_flush_on_hlink = 0; 219 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW, 220 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link"); 221 222 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 223 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 225 226 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 227 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 228 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout"); 229 230 static int nfspos_cache_timeout = NFS_MINATTRTIMO; 231 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW, 232 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout"); 233 234 static int nfsv3_commit_on_close = 0; 235 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 236 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 237 #if 0 238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 239 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 240 241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 242 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 243 #endif 244 245 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 246 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 247 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 248 static int 249 nfs3_access_otw(struct vnode *vp, int wmode, 250 struct thread *td, struct ucred *cred) 251 { 252 struct nfsnode *np = VTONFS(vp); 253 int attrflag; 254 int error = 0; 255 u_int32_t *tl; 256 u_int32_t rmode; 257 struct nfsm_info info; 258 259 info.mrep = NULL; 260 info.v3 = 1; 261 262 nfsstats.rpccnt[NFSPROC_ACCESS]++; 263 nfsm_reqhead(&info, vp, NFSPROC_ACCESS, 264 NFSX_FH(info.v3) + NFSX_UNSIGNED); 265 ERROROUT(nfsm_fhtom(&info, vp)); 266 tl = nfsm_build(&info, NFSX_UNSIGNED); 267 *tl = txdr_unsigned(wmode); 268 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error)); 269 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK)); 270 if (error == 0) { 271 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 272 rmode = fxdr_unsigned(u_int32_t, *tl); 273 np->n_mode = rmode; 274 np->n_modeuid = cred->cr_uid; 275 np->n_modestamp = mycpu->gd_time_seconds; 276 } 277 m_freem(info.mrep); 278 info.mrep = NULL; 279 nfsmout: 280 return error; 281 } 282 283 /* 284 * nfs access vnode op. 285 * For nfs version 2, just return ok. File accesses may fail later. 286 * For nfs version 3, use the access rpc to check accessibility. If file modes 287 * are changed on the server, accesses might still fail later. 288 * 289 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 290 */ 291 static int 292 nfs_access(struct vop_access_args *ap) 293 { 294 struct vnode *vp = ap->a_vp; 295 thread_t td = curthread; 296 int error = 0; 297 u_int32_t mode, wmode; 298 struct nfsnode *np = VTONFS(vp); 299 int v3 = NFS_ISV3(vp); 300 301 /* 302 * Disallow write attempts on filesystems mounted read-only; 303 * unless the file is a socket, fifo, or a block or character 304 * device resident on the filesystem. 305 */ 306 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 307 switch (vp->v_type) { 308 case VREG: 309 case VDIR: 310 case VLNK: 311 return (EROFS); 312 default: 313 break; 314 } 315 } 316 /* 317 * For nfs v3, check to see if we have done this recently, and if 318 * so return our cached result instead of making an ACCESS call. 319 * If not, do an access rpc, otherwise you are stuck emulating 320 * ufs_access() locally using the vattr. This may not be correct, 321 * since the server may apply other access criteria such as 322 * client uid-->server uid mapping that we do not know about. 323 */ 324 if (v3) { 325 if (ap->a_mode & VREAD) 326 mode = NFSV3ACCESS_READ; 327 else 328 mode = 0; 329 if (vp->v_type != VDIR) { 330 if (ap->a_mode & VWRITE) 331 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 332 if (ap->a_mode & VEXEC) 333 mode |= NFSV3ACCESS_EXECUTE; 334 } else { 335 if (ap->a_mode & VWRITE) 336 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 337 NFSV3ACCESS_DELETE); 338 if (ap->a_mode & VEXEC) 339 mode |= NFSV3ACCESS_LOOKUP; 340 } 341 /* XXX safety belt, only make blanket request if caching */ 342 if (nfsaccess_cache_timeout > 0) { 343 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 344 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 345 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 346 } else { 347 wmode = mode; 348 } 349 350 /* 351 * Does our cached result allow us to give a definite yes to 352 * this request? 353 */ 354 if (np->n_modestamp && 355 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 356 (ap->a_cred->cr_uid == np->n_modeuid) && 357 ((np->n_mode & mode) == mode)) { 358 nfsstats.accesscache_hits++; 359 } else { 360 /* 361 * Either a no, or a don't know. Go to the wire. 362 */ 363 nfsstats.accesscache_misses++; 364 error = nfs3_access_otw(vp, wmode, td, ap->a_cred); 365 if (!error) { 366 if ((np->n_mode & mode) != mode) { 367 error = EACCES; 368 } 369 } 370 } 371 } else { 372 if ((error = nfs_laccess(ap)) != 0) 373 return (error); 374 375 /* 376 * Attempt to prevent a mapped root from accessing a file 377 * which it shouldn't. We try to read a byte from the file 378 * if the user is root and the file is not zero length. 379 * After calling nfs_laccess, we should have the correct 380 * file size cached. 381 */ 382 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 383 && VTONFS(vp)->n_size > 0) { 384 struct iovec aiov; 385 struct uio auio; 386 char buf[1]; 387 388 aiov.iov_base = buf; 389 aiov.iov_len = 1; 390 auio.uio_iov = &aiov; 391 auio.uio_iovcnt = 1; 392 auio.uio_offset = 0; 393 auio.uio_resid = 1; 394 auio.uio_segflg = UIO_SYSSPACE; 395 auio.uio_rw = UIO_READ; 396 auio.uio_td = td; 397 398 if (vp->v_type == VREG) { 399 error = nfs_readrpc_uio(vp, &auio); 400 } else if (vp->v_type == VDIR) { 401 char* bp; 402 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 403 aiov.iov_base = bp; 404 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 405 error = nfs_readdirrpc_uio(vp, &auio); 406 kfree(bp, M_TEMP); 407 } else if (vp->v_type == VLNK) { 408 error = nfs_readlinkrpc_uio(vp, &auio); 409 } else { 410 error = EACCES; 411 } 412 } 413 } 414 /* 415 * [re]record creds for reading and/or writing if access 416 * was granted. Assume the NFS server will grant read access 417 * for execute requests. 418 */ 419 if (error == 0) { 420 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 421 crhold(ap->a_cred); 422 if (np->n_rucred) 423 crfree(np->n_rucred); 424 np->n_rucred = ap->a_cred; 425 } 426 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 427 crhold(ap->a_cred); 428 if (np->n_wucred) 429 crfree(np->n_wucred); 430 np->n_wucred = ap->a_cred; 431 } 432 } 433 return(error); 434 } 435 436 /* 437 * nfs open vnode op 438 * Check to see if the type is ok 439 * and that deletion is not in progress. 440 * For paged in text files, you will need to flush the page cache 441 * if consistency is lost. 442 * 443 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 444 * struct file *a_fp) 445 */ 446 /* ARGSUSED */ 447 static int 448 nfs_open(struct vop_open_args *ap) 449 { 450 struct vnode *vp = ap->a_vp; 451 struct nfsnode *np = VTONFS(vp); 452 struct vattr vattr; 453 int error; 454 455 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 456 #ifdef DIAGNOSTIC 457 kprintf("open eacces vtyp=%d\n",vp->v_type); 458 #endif 459 return (EOPNOTSUPP); 460 } 461 462 /* 463 * Save valid creds for reading and writing for later RPCs. 464 */ 465 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) { 466 crhold(ap->a_cred); 467 if (np->n_rucred) 468 crfree(np->n_rucred); 469 np->n_rucred = ap->a_cred; 470 } 471 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) { 472 crhold(ap->a_cred); 473 if (np->n_wucred) 474 crfree(np->n_wucred); 475 np->n_wucred = ap->a_cred; 476 } 477 478 /* 479 * Clear the attribute cache only if opening with write access. It 480 * is unclear if we should do this at all here, but we certainly 481 * should not clear the cache unconditionally simply because a file 482 * is being opened. 483 */ 484 if (ap->a_mode & FWRITE) 485 np->n_attrstamp = 0; 486 487 /* 488 * For normal NFS, reconcile changes made locally verses 489 * changes made remotely. Note that VOP_GETATTR only goes 490 * to the wire if the cached attribute has timed out or been 491 * cleared. 492 * 493 * If local modifications have been made clear the attribute 494 * cache to force an attribute and modified time check. If 495 * GETATTR detects that the file has been changed by someone 496 * other then us it will set NRMODIFIED. 497 * 498 * If we are opening a directory and local changes have been 499 * made we have to invalidate the cache in order to ensure 500 * that we get the most up-to-date information from the 501 * server. XXX 502 */ 503 if (np->n_flag & NLMODIFIED) { 504 np->n_attrstamp = 0; 505 if (vp->v_type == VDIR) { 506 error = nfs_vinvalbuf(vp, V_SAVE, 1); 507 if (error == EINTR) 508 return (error); 509 nfs_invaldir(vp); 510 } 511 } 512 error = VOP_GETATTR(vp, &vattr); 513 if (error) 514 return (error); 515 if (np->n_flag & NRMODIFIED) { 516 if (vp->v_type == VDIR) 517 nfs_invaldir(vp); 518 error = nfs_vinvalbuf(vp, V_SAVE, 1); 519 if (error == EINTR) 520 return (error); 521 np->n_flag &= ~NRMODIFIED; 522 } 523 524 return (vop_stdopen(ap)); 525 } 526 527 /* 528 * nfs close vnode op 529 * What an NFS client should do upon close after writing is a debatable issue. 530 * Most NFS clients push delayed writes to the server upon close, basically for 531 * two reasons: 532 * 1 - So that any write errors may be reported back to the client process 533 * doing the close system call. By far the two most likely errors are 534 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 535 * 2 - To put a worst case upper bound on cache inconsistency between 536 * multiple clients for the file. 537 * There is also a consistency problem for Version 2 of the protocol w.r.t. 538 * not being able to tell if other clients are writing a file concurrently, 539 * since there is no way of knowing if the changed modify time in the reply 540 * is only due to the write for this client. 541 * (NFS Version 3 provides weak cache consistency data in the reply that 542 * should be sufficient to detect and handle this case.) 543 * 544 * The current code does the following: 545 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 546 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 547 * or commit them (this satisfies 1 and 2 except for the 548 * case where the server crashes after this close but 549 * before the commit RPC, which is felt to be "good 550 * enough". Changing the last argument to nfs_flush() to 551 * a 1 would force a commit operation, if it is felt a 552 * commit is necessary now. 553 * for NQNFS - do nothing now, since 2 is dealt with via leases and 554 * 1 should be dealt with via an fsync() system call for 555 * cases where write errors are important. 556 * 557 * nfs_close(struct vnode *a_vp, int a_fflag) 558 */ 559 /* ARGSUSED */ 560 static int 561 nfs_close(struct vop_close_args *ap) 562 { 563 struct vnode *vp = ap->a_vp; 564 struct nfsnode *np = VTONFS(vp); 565 int error = 0; 566 thread_t td = curthread; 567 568 if (vp->v_type == VREG) { 569 if (np->n_flag & NLMODIFIED) { 570 if (NFS_ISV3(vp)) { 571 /* 572 * Under NFSv3 we have dirty buffers to dispose of. We 573 * must flush them to the NFS server. We have the option 574 * of waiting all the way through the commit rpc or just 575 * waiting for the initial write. The default is to only 576 * wait through the initial write so the data is in the 577 * server's cache, which is roughly similar to the state 578 * a standard disk subsystem leaves the file in on close(). 579 * 580 * We cannot clear the NLMODIFIED bit in np->n_flag due to 581 * potential races with other processes, and certainly 582 * cannot clear it if we don't commit. 583 */ 584 int cm = nfsv3_commit_on_close ? 1 : 0; 585 error = nfs_flush(vp, MNT_WAIT, td, cm); 586 /* np->n_flag &= ~NLMODIFIED; */ 587 } else { 588 error = nfs_vinvalbuf(vp, V_SAVE, 1); 589 } 590 np->n_attrstamp = 0; 591 } 592 if (np->n_flag & NWRITEERR) { 593 np->n_flag &= ~NWRITEERR; 594 error = np->n_error; 595 } 596 } 597 vop_stdclose(ap); 598 return (error); 599 } 600 601 /* 602 * nfs getattr call from vfs. 603 * 604 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap) 605 */ 606 static int 607 nfs_getattr(struct vop_getattr_args *ap) 608 { 609 struct vnode *vp = ap->a_vp; 610 struct nfsnode *np = VTONFS(vp); 611 int error = 0; 612 thread_t td = curthread; 613 struct nfsm_info info; 614 615 info.mrep = NULL; 616 info.v3 = NFS_ISV3(vp); 617 618 /* 619 * Update local times for special files. 620 */ 621 if (np->n_flag & (NACC | NUPD)) 622 np->n_flag |= NCHG; 623 /* 624 * First look in the cache. 625 */ 626 if (nfs_getattrcache(vp, ap->a_vap) == 0) 627 return (0); 628 629 if (info.v3 && nfsaccess_cache_timeout > 0) { 630 nfsstats.accesscache_misses++; 631 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 632 if (nfs_getattrcache(vp, ap->a_vap) == 0) 633 return (0); 634 } 635 636 nfsstats.rpccnt[NFSPROC_GETATTR]++; 637 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3)); 638 ERROROUT(nfsm_fhtom(&info, vp)); 639 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td, 640 nfs_vpcred(vp, ND_CHECK), &error)); 641 if (error == 0) { 642 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap)); 643 } 644 m_freem(info.mrep); 645 info.mrep = NULL; 646 nfsmout: 647 return (error); 648 } 649 650 /* 651 * nfs setattr call. 652 * 653 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 654 */ 655 static int 656 nfs_setattr(struct vop_setattr_args *ap) 657 { 658 struct vnode *vp = ap->a_vp; 659 struct nfsnode *np = VTONFS(vp); 660 struct vattr *vap = ap->a_vap; 661 int error = 0; 662 u_quad_t tsize; 663 thread_t td = curthread; 664 665 #ifndef nolint 666 tsize = (u_quad_t)0; 667 #endif 668 669 /* 670 * Setting of flags is not supported. 671 */ 672 if (vap->va_flags != VNOVAL) 673 return (EOPNOTSUPP); 674 675 /* 676 * Disallow write attempts if the filesystem is mounted read-only. 677 */ 678 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 679 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 680 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 681 (vp->v_mount->mnt_flag & MNT_RDONLY)) 682 return (EROFS); 683 684 if (vap->va_size != VNOVAL) { 685 /* 686 * truncation requested 687 */ 688 switch (vp->v_type) { 689 case VDIR: 690 return (EISDIR); 691 case VCHR: 692 case VBLK: 693 case VSOCK: 694 case VFIFO: 695 if (vap->va_mtime.tv_sec == VNOVAL && 696 vap->va_atime.tv_sec == VNOVAL && 697 vap->va_mode == (mode_t)VNOVAL && 698 vap->va_uid == (uid_t)VNOVAL && 699 vap->va_gid == (gid_t)VNOVAL) 700 return (0); 701 vap->va_size = VNOVAL; 702 break; 703 default: 704 /* 705 * Disallow write attempts if the filesystem is 706 * mounted read-only. 707 */ 708 if (vp->v_mount->mnt_flag & MNT_RDONLY) 709 return (EROFS); 710 711 /* 712 * This is nasty. The RPCs we send to flush pending 713 * data often return attribute information which is 714 * cached via a callback to nfs_loadattrcache(), which 715 * has the effect of changing our notion of the file 716 * size. Due to flushed appends and other operations 717 * the file size can be set to virtually anything, 718 * including values that do not match either the old 719 * or intended file size. 720 * 721 * When this condition is detected we must loop to 722 * try the operation again. Hopefully no more 723 * flushing is required on the loop so it works the 724 * second time around. THIS CASE ALMOST ALWAYS 725 * HAPPENS! 726 */ 727 tsize = np->n_size; 728 again: 729 error = nfs_meta_setsize(vp, td, vap->va_size); 730 731 if (np->n_flag & NLMODIFIED) { 732 if (vap->va_size == 0) 733 error = nfs_vinvalbuf(vp, 0, 1); 734 else 735 error = nfs_vinvalbuf(vp, V_SAVE, 1); 736 } 737 /* 738 * note: this loop case almost always happens at 739 * least once per truncation. 740 */ 741 if (error == 0 && np->n_size != vap->va_size) 742 goto again; 743 np->n_vattr.va_size = vap->va_size; 744 break; 745 } 746 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) { 747 /* 748 * What to do. If we are modifying the mtime we lose 749 * mtime detection of changes made by the server or other 750 * clients. But programs like rsync/rdist/cpdup are going 751 * to call utimes a lot. We don't want to piecemeal sync. 752 * 753 * For now sync if any prior remote changes were detected, 754 * but allow us to lose track of remote changes made during 755 * the utimes operation. 756 */ 757 if (np->n_flag & NRMODIFIED) 758 error = nfs_vinvalbuf(vp, V_SAVE, 1); 759 if (error == EINTR) 760 return (error); 761 if (error == 0) { 762 if (vap->va_mtime.tv_sec != VNOVAL) { 763 np->n_mtime = vap->va_mtime.tv_sec; 764 } 765 } 766 } 767 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 768 769 /* 770 * Sanity check if a truncation was issued. This should only occur 771 * if multiple processes are racing on the same file. 772 */ 773 if (error == 0 && vap->va_size != VNOVAL && 774 np->n_size != vap->va_size) { 775 kprintf("NFS ftruncate: server disagrees on the file size: " 776 "%lld/%lld/%lld\n", 777 (long long)tsize, 778 (long long)vap->va_size, 779 (long long)np->n_size); 780 goto again; 781 } 782 if (error && vap->va_size != VNOVAL) { 783 np->n_size = np->n_vattr.va_size = tsize; 784 vnode_pager_setsize(vp, np->n_size); 785 } 786 return (error); 787 } 788 789 /* 790 * Do an nfs setattr rpc. 791 */ 792 static int 793 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 794 struct ucred *cred, struct thread *td) 795 { 796 struct nfsv2_sattr *sp; 797 struct nfsnode *np = VTONFS(vp); 798 u_int32_t *tl; 799 int error = 0, wccflag = NFSV3_WCCRATTR; 800 struct nfsm_info info; 801 802 info.mrep = NULL; 803 info.v3 = NFS_ISV3(vp); 804 805 nfsstats.rpccnt[NFSPROC_SETATTR]++; 806 nfsm_reqhead(&info, vp, NFSPROC_SETATTR, 807 NFSX_FH(info.v3) + NFSX_SATTR(info.v3)); 808 ERROROUT(nfsm_fhtom(&info, vp)); 809 if (info.v3) { 810 nfsm_v3attrbuild(&info, vap, TRUE); 811 tl = nfsm_build(&info, NFSX_UNSIGNED); 812 *tl = nfs_false; 813 } else { 814 sp = nfsm_build(&info, NFSX_V2SATTR); 815 if (vap->va_mode == (mode_t)VNOVAL) 816 sp->sa_mode = nfs_xdrneg1; 817 else 818 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 819 if (vap->va_uid == (uid_t)VNOVAL) 820 sp->sa_uid = nfs_xdrneg1; 821 else 822 sp->sa_uid = txdr_unsigned(vap->va_uid); 823 if (vap->va_gid == (gid_t)VNOVAL) 824 sp->sa_gid = nfs_xdrneg1; 825 else 826 sp->sa_gid = txdr_unsigned(vap->va_gid); 827 sp->sa_size = txdr_unsigned(vap->va_size); 828 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 829 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 830 } 831 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error)); 832 if (info.v3) { 833 np->n_modestamp = 0; 834 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 835 } else { 836 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 837 } 838 m_freem(info.mrep); 839 info.mrep = NULL; 840 nfsmout: 841 return (error); 842 } 843 844 static 845 void 846 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout) 847 { 848 if (nctimeout == 0) 849 nctimeout = 1; 850 else 851 nctimeout *= hz; 852 cache_setvp(nch, vp); 853 cache_settimeout(nch, nctimeout); 854 } 855 856 /* 857 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 858 * nfs_lookup() until all remaining new api calls are implemented. 859 * 860 * Resolve a namecache entry. This function is passed a locked ncp and 861 * must call nfs_cache_setvp() on it as appropriate to resolve the entry. 862 */ 863 static int 864 nfs_nresolve(struct vop_nresolve_args *ap) 865 { 866 struct thread *td = curthread; 867 struct namecache *ncp; 868 struct ucred *cred; 869 struct nfsnode *np; 870 struct vnode *dvp; 871 struct vnode *nvp; 872 nfsfh_t *fhp; 873 int attrflag; 874 int fhsize; 875 int error; 876 int tmp_error; 877 int len; 878 struct nfsm_info info; 879 880 cred = ap->a_cred; 881 dvp = ap->a_dvp; 882 883 if ((error = vget(dvp, LK_SHARED)) != 0) 884 return (error); 885 886 info.mrep = NULL; 887 info.v3 = NFS_ISV3(dvp); 888 889 nvp = NULL; 890 nfsstats.lookupcache_misses++; 891 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 892 ncp = ap->a_nch->ncp; 893 len = ncp->nc_nlen; 894 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 895 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 896 ERROROUT(nfsm_fhtom(&info, dvp)); 897 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN)); 898 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, 899 ap->a_cred, &error)); 900 if (error) { 901 /* 902 * Cache negatve lookups to reduce NFS traffic, but use 903 * a fast timeout. Otherwise use a timeout of 1 tick. 904 * XXX we should add a namecache flag for no-caching 905 * to uncache the negative hit as soon as possible, but 906 * we cannot simply destroy the entry because it is used 907 * as a placeholder by the caller. 908 * 909 * The refactored nfs code will overwrite a non-zero error 910 * with 0 when we use ERROROUT(), so don't here. 911 */ 912 if (error == ENOENT) 913 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout); 914 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 915 NFS_LATTR_NOSHRINK); 916 if (tmp_error) { 917 error = tmp_error; 918 goto nfsmout; 919 } 920 m_freem(info.mrep); 921 info.mrep = NULL; 922 goto nfsmout; 923 } 924 925 /* 926 * Success, get the file handle, do various checks, and load 927 * post-operation data from the reply packet. Theoretically 928 * we should never be looking up "." so, theoretically, we 929 * should never get the same file handle as our directory. But 930 * we check anyway. XXX 931 * 932 * Note that no timeout is set for the positive cache hit. We 933 * assume, theoretically, that ESTALE returns will be dealt with 934 * properly to handle NFS races and in anycase we cannot depend 935 * on a timeout to deal with NFS open/create/excl issues so instead 936 * of a bad hack here the rest of the NFS client code needs to do 937 * the right thing. 938 */ 939 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 940 941 np = VTONFS(dvp); 942 if (NFS_CMPFH(np, fhp, fhsize)) { 943 vref(dvp); 944 nvp = dvp; 945 } else { 946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 947 if (error) { 948 m_freem(info.mrep); 949 info.mrep = NULL; 950 vput(dvp); 951 return (error); 952 } 953 nvp = NFSTOV(np); 954 } 955 if (info.v3) { 956 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag, 957 NFS_LATTR_NOSHRINK)); 958 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 959 NFS_LATTR_NOSHRINK)); 960 } else { 961 ERROROUT(nfsm_loadattr(&info, nvp, NULL)); 962 } 963 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout); 964 m_freem(info.mrep); 965 info.mrep = NULL; 966 nfsmout: 967 vput(dvp); 968 if (nvp) { 969 if (nvp == dvp) 970 vrele(nvp); 971 else 972 vput(nvp); 973 } 974 return (error); 975 } 976 977 /* 978 * 'cached' nfs directory lookup 979 * 980 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 981 * 982 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 983 * struct componentname *a_cnp) 984 */ 985 static int 986 nfs_lookup(struct vop_old_lookup_args *ap) 987 { 988 struct componentname *cnp = ap->a_cnp; 989 struct vnode *dvp = ap->a_dvp; 990 struct vnode **vpp = ap->a_vpp; 991 int flags = cnp->cn_flags; 992 struct vnode *newvp; 993 struct nfsmount *nmp; 994 long len; 995 nfsfh_t *fhp; 996 struct nfsnode *np; 997 int lockparent, wantparent, attrflag, fhsize; 998 int error; 999 int tmp_error; 1000 struct nfsm_info info; 1001 1002 info.mrep = NULL; 1003 info.v3 = NFS_ISV3(dvp); 1004 error = 0; 1005 1006 /* 1007 * Read-only mount check and directory check. 1008 */ 1009 *vpp = NULLVP; 1010 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1011 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1012 return (EROFS); 1013 1014 if (dvp->v_type != VDIR) 1015 return (ENOTDIR); 1016 1017 /* 1018 * Look it up in the cache. Note that ENOENT is only returned if we 1019 * previously entered a negative hit (see later on). The additional 1020 * nfsneg_cache_timeout check causes previously cached results to 1021 * be instantly ignored if the negative caching is turned off. 1022 */ 1023 lockparent = flags & CNP_LOCKPARENT; 1024 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1025 nmp = VFSTONFS(dvp->v_mount); 1026 np = VTONFS(dvp); 1027 1028 /* 1029 * Go to the wire. 1030 */ 1031 error = 0; 1032 newvp = NULLVP; 1033 nfsstats.lookupcache_misses++; 1034 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1035 len = cnp->cn_namelen; 1036 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 1037 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1038 ERROROUT(nfsm_fhtom(&info, dvp)); 1039 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 1040 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td, 1041 cnp->cn_cred, &error)); 1042 if (error) { 1043 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag, 1044 NFS_LATTR_NOSHRINK); 1045 if (tmp_error) { 1046 error = tmp_error; 1047 goto nfsmout; 1048 } 1049 1050 m_freem(info.mrep); 1051 info.mrep = NULL; 1052 goto nfsmout; 1053 } 1054 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 1055 1056 /* 1057 * Handle RENAME case... 1058 */ 1059 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1060 if (NFS_CMPFH(np, fhp, fhsize)) { 1061 m_freem(info.mrep); 1062 info.mrep = NULL; 1063 return (EISDIR); 1064 } 1065 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1066 if (error) { 1067 m_freem(info.mrep); 1068 info.mrep = NULL; 1069 return (error); 1070 } 1071 newvp = NFSTOV(np); 1072 if (info.v3) { 1073 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1074 NFS_LATTR_NOSHRINK)); 1075 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1076 NFS_LATTR_NOSHRINK)); 1077 } else { 1078 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1079 } 1080 *vpp = newvp; 1081 m_freem(info.mrep); 1082 info.mrep = NULL; 1083 if (!lockparent) { 1084 vn_unlock(dvp); 1085 cnp->cn_flags |= CNP_PDIRUNLOCK; 1086 } 1087 return (0); 1088 } 1089 1090 if (flags & CNP_ISDOTDOT) { 1091 vn_unlock(dvp); 1092 cnp->cn_flags |= CNP_PDIRUNLOCK; 1093 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1094 if (error) { 1095 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1096 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1097 return (error); /* NOTE: return error from nget */ 1098 } 1099 newvp = NFSTOV(np); 1100 if (lockparent) { 1101 error = vn_lock(dvp, LK_EXCLUSIVE); 1102 if (error) { 1103 vput(newvp); 1104 return (error); 1105 } 1106 cnp->cn_flags |= CNP_PDIRUNLOCK; 1107 } 1108 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1109 vref(dvp); 1110 newvp = dvp; 1111 } else { 1112 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1113 if (error) { 1114 m_freem(info.mrep); 1115 info.mrep = NULL; 1116 return (error); 1117 } 1118 if (!lockparent) { 1119 vn_unlock(dvp); 1120 cnp->cn_flags |= CNP_PDIRUNLOCK; 1121 } 1122 newvp = NFSTOV(np); 1123 } 1124 if (info.v3) { 1125 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 1126 NFS_LATTR_NOSHRINK)); 1127 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag, 1128 NFS_LATTR_NOSHRINK)); 1129 } else { 1130 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 1131 } 1132 #if 0 1133 /* XXX MOVE TO nfs_nremove() */ 1134 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1135 cnp->cn_nameiop != NAMEI_DELETE) { 1136 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1137 } 1138 #endif 1139 *vpp = newvp; 1140 m_freem(info.mrep); 1141 info.mrep = NULL; 1142 nfsmout: 1143 if (error) { 1144 if (newvp != NULLVP) { 1145 vrele(newvp); 1146 *vpp = NULLVP; 1147 } 1148 if ((cnp->cn_nameiop == NAMEI_CREATE || 1149 cnp->cn_nameiop == NAMEI_RENAME) && 1150 error == ENOENT) { 1151 if (!lockparent) { 1152 vn_unlock(dvp); 1153 cnp->cn_flags |= CNP_PDIRUNLOCK; 1154 } 1155 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1156 error = EROFS; 1157 else 1158 error = EJUSTRETURN; 1159 } 1160 } 1161 return (error); 1162 } 1163 1164 /* 1165 * nfs read call. 1166 * Just call nfs_bioread() to do the work. 1167 * 1168 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1169 * struct ucred *a_cred) 1170 */ 1171 static int 1172 nfs_read(struct vop_read_args *ap) 1173 { 1174 struct vnode *vp = ap->a_vp; 1175 1176 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1177 } 1178 1179 /* 1180 * nfs readlink call 1181 * 1182 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1183 */ 1184 static int 1185 nfs_readlink(struct vop_readlink_args *ap) 1186 { 1187 struct vnode *vp = ap->a_vp; 1188 1189 if (vp->v_type != VLNK) 1190 return (EINVAL); 1191 return (nfs_bioread(vp, ap->a_uio, 0)); 1192 } 1193 1194 /* 1195 * Do a readlink rpc. 1196 * Called by nfs_doio() from below the buffer cache. 1197 */ 1198 int 1199 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop) 1200 { 1201 int error = 0, len, attrflag; 1202 struct nfsm_info info; 1203 1204 info.mrep = NULL; 1205 info.v3 = NFS_ISV3(vp); 1206 1207 nfsstats.rpccnt[NFSPROC_READLINK]++; 1208 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3)); 1209 ERROROUT(nfsm_fhtom(&info, vp)); 1210 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td, 1211 nfs_vpcred(vp, ND_CHECK), &error)); 1212 if (info.v3) { 1213 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1214 NFS_LATTR_NOSHRINK)); 1215 } 1216 if (!error) { 1217 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN)); 1218 if (len == NFS_MAXPATHLEN) { 1219 struct nfsnode *np = VTONFS(vp); 1220 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1221 len = np->n_size; 1222 } 1223 ERROROUT(nfsm_mtouio(&info, uiop, len)); 1224 } 1225 m_freem(info.mrep); 1226 info.mrep = NULL; 1227 nfsmout: 1228 return (error); 1229 } 1230 1231 /* 1232 * nfs synchronous read rpc using UIO 1233 */ 1234 int 1235 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop) 1236 { 1237 u_int32_t *tl; 1238 struct nfsmount *nmp; 1239 int error = 0, len, retlen, tsiz, eof, attrflag; 1240 struct nfsm_info info; 1241 off_t tmp_off; 1242 1243 info.mrep = NULL; 1244 info.v3 = NFS_ISV3(vp); 1245 1246 #ifndef nolint 1247 eof = 0; 1248 #endif 1249 nmp = VFSTONFS(vp->v_mount); 1250 tsiz = uiop->uio_resid; 1251 tmp_off = uiop->uio_offset + tsiz; 1252 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) 1253 return (EFBIG); 1254 tmp_off = uiop->uio_offset; 1255 while (tsiz > 0) { 1256 nfsstats.rpccnt[NFSPROC_READ]++; 1257 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1258 nfsm_reqhead(&info, vp, NFSPROC_READ, 1259 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3); 1260 ERROROUT(nfsm_fhtom(&info, vp)); 1261 tl = nfsm_build(&info, NFSX_UNSIGNED * 3); 1262 if (info.v3) { 1263 txdr_hyper(uiop->uio_offset, tl); 1264 *(tl + 2) = txdr_unsigned(len); 1265 } else { 1266 *tl++ = txdr_unsigned(uiop->uio_offset); 1267 *tl++ = txdr_unsigned(len); 1268 *tl = 0; 1269 } 1270 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td, 1271 nfs_vpcred(vp, ND_READ), &error)); 1272 if (info.v3) { 1273 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1274 NFS_LATTR_NOSHRINK)); 1275 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 1276 eof = fxdr_unsigned(int, *(tl + 1)); 1277 } else { 1278 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1279 } 1280 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len)); 1281 ERROROUT(nfsm_mtouio(&info, uiop, retlen)); 1282 m_freem(info.mrep); 1283 info.mrep = NULL; 1284 1285 /* 1286 * Handle short-read from server (NFSv3). If EOF is not 1287 * flagged (and no error occurred), but retlen is less 1288 * then the request size, we must zero-fill the remainder. 1289 */ 1290 if (retlen < len && info.v3 && eof == 0) { 1291 ERROROUT(uiomovez(len - retlen, uiop)); 1292 retlen = len; 1293 } 1294 tsiz -= retlen; 1295 1296 /* 1297 * Terminate loop on EOF or zero-length read. 1298 * 1299 * For NFSv2 a short-read indicates EOF, not zero-fill, 1300 * and also terminates the loop. 1301 */ 1302 if (info.v3) { 1303 if (eof || retlen == 0) 1304 tsiz = 0; 1305 } else if (retlen < len) { 1306 tsiz = 0; 1307 } 1308 } 1309 nfsmout: 1310 return (error); 1311 } 1312 1313 /* 1314 * nfs write call 1315 */ 1316 int 1317 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop, 1318 int *iomode, int *must_commit) 1319 { 1320 u_int32_t *tl; 1321 int32_t backup; 1322 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1323 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1324 int committed = NFSV3WRITE_FILESYNC; 1325 struct nfsm_info info; 1326 1327 info.mrep = NULL; 1328 info.v3 = NFS_ISV3(vp); 1329 1330 #ifndef DIAGNOSTIC 1331 if (uiop->uio_iovcnt != 1) 1332 panic("nfs: writerpc iovcnt > 1"); 1333 #endif 1334 *must_commit = 0; 1335 tsiz = uiop->uio_resid; 1336 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1337 return (EFBIG); 1338 while (tsiz > 0) { 1339 nfsstats.rpccnt[NFSPROC_WRITE]++; 1340 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1341 nfsm_reqhead(&info, vp, NFSPROC_WRITE, 1342 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1343 ERROROUT(nfsm_fhtom(&info, vp)); 1344 if (info.v3) { 1345 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 1346 txdr_hyper(uiop->uio_offset, tl); 1347 tl += 2; 1348 *tl++ = txdr_unsigned(len); 1349 *tl++ = txdr_unsigned(*iomode); 1350 *tl = txdr_unsigned(len); 1351 } else { 1352 u_int32_t x; 1353 1354 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED); 1355 /* Set both "begin" and "current" to non-garbage. */ 1356 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1357 *tl++ = x; /* "begin offset" */ 1358 *tl++ = x; /* "current offset" */ 1359 x = txdr_unsigned(len); 1360 *tl++ = x; /* total to this offset */ 1361 *tl = x; /* size of this write */ 1362 } 1363 ERROROUT(nfsm_uiotom(&info, uiop, len)); 1364 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td, 1365 nfs_vpcred(vp, ND_WRITE), &error)); 1366 if (info.v3) { 1367 /* 1368 * The write RPC returns a before and after mtime. The 1369 * nfsm_wcc_data() macro checks the before n_mtime 1370 * against the before time and stores the after time 1371 * in the nfsnode's cached vattr and n_mtime field. 1372 * The NRMODIFIED bit will be set if the before 1373 * time did not match the original mtime. 1374 */ 1375 wccflag = NFSV3_WCCCHK; 1376 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 1377 if (error == 0) { 1378 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF)); 1379 rlen = fxdr_unsigned(int, *tl++); 1380 if (rlen == 0) { 1381 error = NFSERR_IO; 1382 m_freem(info.mrep); 1383 info.mrep = NULL; 1384 break; 1385 } else if (rlen < len) { 1386 backup = len - rlen; 1387 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup; 1388 uiop->uio_iov->iov_len += backup; 1389 uiop->uio_offset -= backup; 1390 uiop->uio_resid += backup; 1391 len = rlen; 1392 } 1393 commit = fxdr_unsigned(int, *tl++); 1394 1395 /* 1396 * Return the lowest committment level 1397 * obtained by any of the RPCs. 1398 */ 1399 if (committed == NFSV3WRITE_FILESYNC) 1400 committed = commit; 1401 else if (committed == NFSV3WRITE_DATASYNC && 1402 commit == NFSV3WRITE_UNSTABLE) 1403 committed = commit; 1404 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1405 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1406 NFSX_V3WRITEVERF); 1407 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1408 } else if (bcmp((caddr_t)tl, 1409 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1410 *must_commit = 1; 1411 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1412 NFSX_V3WRITEVERF); 1413 } 1414 } 1415 } else { 1416 ERROROUT(nfsm_loadattr(&info, vp, NULL)); 1417 } 1418 m_freem(info.mrep); 1419 info.mrep = NULL; 1420 if (error) 1421 break; 1422 tsiz -= len; 1423 } 1424 nfsmout: 1425 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1426 committed = NFSV3WRITE_FILESYNC; 1427 *iomode = committed; 1428 if (error) 1429 uiop->uio_resid = tsiz; 1430 return (error); 1431 } 1432 1433 /* 1434 * nfs mknod rpc 1435 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1436 * mode set to specify the file type and the size field for rdev. 1437 */ 1438 static int 1439 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1440 struct vattr *vap) 1441 { 1442 struct nfsv2_sattr *sp; 1443 u_int32_t *tl; 1444 struct vnode *newvp = NULL; 1445 struct nfsnode *np = NULL; 1446 struct vattr vattr; 1447 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1448 int rmajor, rminor; 1449 struct nfsm_info info; 1450 1451 info.mrep = NULL; 1452 info.v3 = NFS_ISV3(dvp); 1453 1454 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1455 rmajor = txdr_unsigned(vap->va_rmajor); 1456 rminor = txdr_unsigned(vap->va_rminor); 1457 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) { 1458 rmajor = nfs_xdrneg1; 1459 rminor = nfs_xdrneg1; 1460 } else { 1461 return (EOPNOTSUPP); 1462 } 1463 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1464 return (error); 1465 } 1466 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1467 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD, 1468 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED + 1469 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1470 ERROROUT(nfsm_fhtom(&info, dvp)); 1471 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1472 NFS_MAXNAMLEN)); 1473 if (info.v3) { 1474 tl = nfsm_build(&info, NFSX_UNSIGNED); 1475 *tl++ = vtonfsv3_type(vap->va_type); 1476 nfsm_v3attrbuild(&info, vap, FALSE); 1477 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1478 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 1479 *tl++ = txdr_unsigned(vap->va_rmajor); 1480 *tl = txdr_unsigned(vap->va_rminor); 1481 } 1482 } else { 1483 sp = nfsm_build(&info, NFSX_V2SATTR); 1484 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1485 sp->sa_uid = nfs_xdrneg1; 1486 sp->sa_gid = nfs_xdrneg1; 1487 sp->sa_size = makeudev(rmajor, rminor); 1488 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1489 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1490 } 1491 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td, 1492 cnp->cn_cred, &error)); 1493 if (!error) { 1494 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1495 if (!gotvp) { 1496 if (newvp) { 1497 vput(newvp); 1498 newvp = NULL; 1499 } 1500 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1501 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1502 if (!error) 1503 newvp = NFSTOV(np); 1504 } 1505 } 1506 if (info.v3) { 1507 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1508 } 1509 m_freem(info.mrep); 1510 info.mrep = NULL; 1511 nfsmout: 1512 if (error) { 1513 if (newvp) 1514 vput(newvp); 1515 } else { 1516 *vpp = newvp; 1517 } 1518 VTONFS(dvp)->n_flag |= NLMODIFIED; 1519 if (!wccflag) 1520 VTONFS(dvp)->n_attrstamp = 0; 1521 return (error); 1522 } 1523 1524 /* 1525 * nfs mknod vop 1526 * just call nfs_mknodrpc() to do the work. 1527 * 1528 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1529 * struct componentname *a_cnp, struct vattr *a_vap) 1530 */ 1531 /* ARGSUSED */ 1532 static int 1533 nfs_mknod(struct vop_old_mknod_args *ap) 1534 { 1535 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1536 } 1537 1538 static u_long create_verf; 1539 /* 1540 * nfs file create call 1541 * 1542 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1543 * struct componentname *a_cnp, struct vattr *a_vap) 1544 */ 1545 static int 1546 nfs_create(struct vop_old_create_args *ap) 1547 { 1548 struct vnode *dvp = ap->a_dvp; 1549 struct vattr *vap = ap->a_vap; 1550 struct componentname *cnp = ap->a_cnp; 1551 struct nfsv2_sattr *sp; 1552 u_int32_t *tl; 1553 struct nfsnode *np = NULL; 1554 struct vnode *newvp = NULL; 1555 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1556 struct vattr vattr; 1557 struct nfsm_info info; 1558 1559 info.mrep = NULL; 1560 info.v3 = NFS_ISV3(dvp); 1561 1562 /* 1563 * Oops, not for me.. 1564 */ 1565 if (vap->va_type == VSOCK) 1566 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1567 1568 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1569 return (error); 1570 } 1571 if (vap->va_vaflags & VA_EXCLUSIVE) 1572 fmode |= O_EXCL; 1573 again: 1574 nfsstats.rpccnt[NFSPROC_CREATE]++; 1575 nfsm_reqhead(&info, dvp, NFSPROC_CREATE, 1576 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED + 1577 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3)); 1578 ERROROUT(nfsm_fhtom(&info, dvp)); 1579 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1580 NFS_MAXNAMLEN)); 1581 if (info.v3) { 1582 tl = nfsm_build(&info, NFSX_UNSIGNED); 1583 if (fmode & O_EXCL) { 1584 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1585 tl = nfsm_build(&info, NFSX_V3CREATEVERF); 1586 #ifdef INET 1587 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1588 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr; 1589 else 1590 #endif 1591 *tl++ = create_verf; 1592 *tl = ++create_verf; 1593 } else { 1594 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1595 nfsm_v3attrbuild(&info, vap, FALSE); 1596 } 1597 } else { 1598 sp = nfsm_build(&info, NFSX_V2SATTR); 1599 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1600 sp->sa_uid = nfs_xdrneg1; 1601 sp->sa_gid = nfs_xdrneg1; 1602 sp->sa_size = 0; 1603 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1604 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1605 } 1606 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td, 1607 cnp->cn_cred, &error)); 1608 if (error == 0) { 1609 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 1610 if (!gotvp) { 1611 if (newvp) { 1612 vput(newvp); 1613 newvp = NULL; 1614 } 1615 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1616 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1617 if (!error) 1618 newvp = NFSTOV(np); 1619 } 1620 } 1621 if (info.v3) { 1622 if (error == 0) 1623 error = nfsm_wcc_data(&info, dvp, &wccflag); 1624 else 1625 (void)nfsm_wcc_data(&info, dvp, &wccflag); 1626 } 1627 m_freem(info.mrep); 1628 info.mrep = NULL; 1629 nfsmout: 1630 if (error) { 1631 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1632 KKASSERT(newvp == NULL); 1633 fmode &= ~O_EXCL; 1634 goto again; 1635 } 1636 } else if (info.v3 && (fmode & O_EXCL)) { 1637 /* 1638 * We are normally called with only a partially initialized 1639 * VAP. Since the NFSv3 spec says that server may use the 1640 * file attributes to store the verifier, the spec requires 1641 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1642 * in atime, but we can't really assume that all servers will 1643 * so we ensure that our SETATTR sets both atime and mtime. 1644 */ 1645 if (vap->va_mtime.tv_sec == VNOVAL) 1646 vfs_timestamp(&vap->va_mtime); 1647 if (vap->va_atime.tv_sec == VNOVAL) 1648 vap->va_atime = vap->va_mtime; 1649 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1650 } 1651 if (error == 0) { 1652 /* 1653 * The new np may have enough info for access 1654 * checks, make sure rucred and wucred are 1655 * initialized for read and write rpc's. 1656 */ 1657 np = VTONFS(newvp); 1658 if (np->n_rucred == NULL) 1659 np->n_rucred = crhold(cnp->cn_cred); 1660 if (np->n_wucred == NULL) 1661 np->n_wucred = crhold(cnp->cn_cred); 1662 *ap->a_vpp = newvp; 1663 } else if (newvp) { 1664 vput(newvp); 1665 } 1666 VTONFS(dvp)->n_flag |= NLMODIFIED; 1667 if (!wccflag) 1668 VTONFS(dvp)->n_attrstamp = 0; 1669 return (error); 1670 } 1671 1672 /* 1673 * nfs file remove call 1674 * To try and make nfs semantics closer to ufs semantics, a file that has 1675 * other processes using the vnode is renamed instead of removed and then 1676 * removed later on the last close. 1677 * - If v_sysref.refcnt > 1 1678 * If a rename is not already in the works 1679 * call nfs_sillyrename() to set it up 1680 * else 1681 * do the remove rpc 1682 * 1683 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1684 * struct componentname *a_cnp) 1685 */ 1686 static int 1687 nfs_remove(struct vop_old_remove_args *ap) 1688 { 1689 struct vnode *vp = ap->a_vp; 1690 struct vnode *dvp = ap->a_dvp; 1691 struct componentname *cnp = ap->a_cnp; 1692 struct nfsnode *np = VTONFS(vp); 1693 int error = 0; 1694 struct vattr vattr; 1695 1696 #ifndef DIAGNOSTIC 1697 if (vp->v_sysref.refcnt < 1) 1698 panic("nfs_remove: bad v_sysref.refcnt"); 1699 #endif 1700 if (vp->v_type == VDIR) 1701 error = EPERM; 1702 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename && 1703 VOP_GETATTR(vp, &vattr) == 0 && 1704 vattr.va_nlink > 1)) { 1705 /* 1706 * throw away biocache buffers, mainly to avoid 1707 * unnecessary delayed writes later. 1708 */ 1709 error = nfs_vinvalbuf(vp, 0, 1); 1710 /* Do the rpc */ 1711 if (error != EINTR) 1712 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1713 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1714 /* 1715 * Kludge City: If the first reply to the remove rpc is lost.. 1716 * the reply to the retransmitted request will be ENOENT 1717 * since the file was in fact removed 1718 * Therefore, we cheat and return success. 1719 */ 1720 if (error == ENOENT) 1721 error = 0; 1722 } else if (!np->n_sillyrename) { 1723 error = nfs_sillyrename(dvp, vp, cnp); 1724 } 1725 np->n_attrstamp = 0; 1726 return (error); 1727 } 1728 1729 /* 1730 * nfs file remove rpc called from nfs_inactive 1731 */ 1732 int 1733 nfs_removeit(struct sillyrename *sp) 1734 { 1735 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1736 sp->s_cred, NULL)); 1737 } 1738 1739 /* 1740 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1741 */ 1742 static int 1743 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1744 struct ucred *cred, struct thread *td) 1745 { 1746 int error = 0, wccflag = NFSV3_WCCRATTR; 1747 struct nfsm_info info; 1748 1749 info.mrep = NULL; 1750 info.v3 = NFS_ISV3(dvp); 1751 1752 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1753 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE, 1754 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1755 ERROROUT(nfsm_fhtom(&info, dvp)); 1756 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN)); 1757 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error)); 1758 if (info.v3) { 1759 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 1760 } 1761 m_freem(info.mrep); 1762 info.mrep = NULL; 1763 nfsmout: 1764 VTONFS(dvp)->n_flag |= NLMODIFIED; 1765 if (!wccflag) 1766 VTONFS(dvp)->n_attrstamp = 0; 1767 return (error); 1768 } 1769 1770 /* 1771 * nfs file rename call 1772 * 1773 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1774 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1775 * struct vnode *a_tvp, struct componentname *a_tcnp) 1776 */ 1777 static int 1778 nfs_rename(struct vop_old_rename_args *ap) 1779 { 1780 struct vnode *fvp = ap->a_fvp; 1781 struct vnode *tvp = ap->a_tvp; 1782 struct vnode *fdvp = ap->a_fdvp; 1783 struct vnode *tdvp = ap->a_tdvp; 1784 struct componentname *tcnp = ap->a_tcnp; 1785 struct componentname *fcnp = ap->a_fcnp; 1786 int error; 1787 1788 /* Check for cross-device rename */ 1789 if ((fvp->v_mount != tdvp->v_mount) || 1790 (tvp && (fvp->v_mount != tvp->v_mount))) { 1791 error = EXDEV; 1792 goto out; 1793 } 1794 1795 /* 1796 * We shouldn't have to flush fvp on rename for most server-side 1797 * filesystems as the file handle should not change. Unfortunately 1798 * the inode for some filesystems (msdosfs) might be tied to the 1799 * file name or directory position so to be completely safe 1800 * vfs.nfs.flush_on_rename is set by default. Clear to improve 1801 * performance. 1802 * 1803 * We must flush tvp on rename because it might become stale on the 1804 * server after the rename. 1805 */ 1806 if (nfs_flush_on_rename) 1807 VOP_FSYNC(fvp, MNT_WAIT, 0); 1808 if (tvp) 1809 VOP_FSYNC(tvp, MNT_WAIT, 0); 1810 1811 /* 1812 * If the tvp exists and is in use, sillyrename it before doing the 1813 * rename of the new file over it. 1814 * 1815 * XXX Can't sillyrename a directory. 1816 * 1817 * We do not attempt to do any namecache purges in this old API 1818 * routine. The new API compat functions have access to the actual 1819 * namecache structures and will do it for us. 1820 */ 1821 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename && 1822 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1823 vput(tvp); 1824 tvp = NULL; 1825 } else if (tvp) { 1826 ; 1827 } 1828 1829 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1830 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1831 tcnp->cn_td); 1832 1833 out: 1834 if (tdvp == tvp) 1835 vrele(tdvp); 1836 else 1837 vput(tdvp); 1838 if (tvp) 1839 vput(tvp); 1840 vrele(fdvp); 1841 vrele(fvp); 1842 /* 1843 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1844 */ 1845 if (error == ENOENT) 1846 error = 0; 1847 return (error); 1848 } 1849 1850 /* 1851 * nfs file rename rpc called from nfs_remove() above 1852 */ 1853 static int 1854 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1855 struct sillyrename *sp) 1856 { 1857 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1858 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1859 } 1860 1861 /* 1862 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1863 */ 1864 static int 1865 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1866 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1867 struct ucred *cred, struct thread *td) 1868 { 1869 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1870 struct nfsm_info info; 1871 1872 info.mrep = NULL; 1873 info.v3 = NFS_ISV3(fdvp); 1874 1875 nfsstats.rpccnt[NFSPROC_RENAME]++; 1876 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME, 1877 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 + 1878 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen)); 1879 ERROROUT(nfsm_fhtom(&info, fdvp)); 1880 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN)); 1881 ERROROUT(nfsm_fhtom(&info, tdvp)); 1882 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN)); 1883 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error)); 1884 if (info.v3) { 1885 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag)); 1886 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag)); 1887 } 1888 m_freem(info.mrep); 1889 info.mrep = NULL; 1890 nfsmout: 1891 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1892 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1893 if (!fwccflag) 1894 VTONFS(fdvp)->n_attrstamp = 0; 1895 if (!twccflag) 1896 VTONFS(tdvp)->n_attrstamp = 0; 1897 return (error); 1898 } 1899 1900 /* 1901 * nfs hard link create call 1902 * 1903 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1904 * struct componentname *a_cnp) 1905 */ 1906 static int 1907 nfs_link(struct vop_old_link_args *ap) 1908 { 1909 struct vnode *vp = ap->a_vp; 1910 struct vnode *tdvp = ap->a_tdvp; 1911 struct componentname *cnp = ap->a_cnp; 1912 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1913 struct nfsm_info info; 1914 1915 if (vp->v_mount != tdvp->v_mount) { 1916 return (EXDEV); 1917 } 1918 1919 /* 1920 * The attribute cache may get out of sync with the server on link. 1921 * Pushing writes to the server before handle was inherited from 1922 * long long ago and it is unclear if we still need to do this. 1923 * Defaults to off. 1924 */ 1925 if (nfs_flush_on_hlink) 1926 VOP_FSYNC(vp, MNT_WAIT, 0); 1927 1928 info.mrep = NULL; 1929 info.v3 = NFS_ISV3(vp); 1930 1931 nfsstats.rpccnt[NFSPROC_LINK]++; 1932 nfsm_reqhead(&info, vp, NFSPROC_LINK, 1933 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED + 1934 nfsm_rndup(cnp->cn_namelen)); 1935 ERROROUT(nfsm_fhtom(&info, vp)); 1936 ERROROUT(nfsm_fhtom(&info, tdvp)); 1937 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1938 NFS_MAXNAMLEN)); 1939 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td, 1940 cnp->cn_cred, &error)); 1941 if (info.v3) { 1942 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 1943 NFS_LATTR_NOSHRINK)); 1944 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag)); 1945 } 1946 m_freem(info.mrep); 1947 info.mrep = NULL; 1948 nfsmout: 1949 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1950 if (!attrflag) 1951 VTONFS(vp)->n_attrstamp = 0; 1952 if (!wccflag) 1953 VTONFS(tdvp)->n_attrstamp = 0; 1954 /* 1955 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1956 */ 1957 if (error == EEXIST) 1958 error = 0; 1959 return (error); 1960 } 1961 1962 /* 1963 * nfs symbolic link create call 1964 * 1965 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1966 * struct componentname *a_cnp, struct vattr *a_vap, 1967 * char *a_target) 1968 */ 1969 static int 1970 nfs_symlink(struct vop_old_symlink_args *ap) 1971 { 1972 struct vnode *dvp = ap->a_dvp; 1973 struct vattr *vap = ap->a_vap; 1974 struct componentname *cnp = ap->a_cnp; 1975 struct nfsv2_sattr *sp; 1976 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1977 struct vnode *newvp = NULL; 1978 struct nfsm_info info; 1979 1980 info.mrep = NULL; 1981 info.v3 = NFS_ISV3(dvp); 1982 1983 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1984 slen = strlen(ap->a_target); 1985 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK, 1986 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED + 1987 nfsm_rndup(cnp->cn_namelen) + 1988 nfsm_rndup(slen) + NFSX_SATTR(info.v3)); 1989 ERROROUT(nfsm_fhtom(&info, dvp)); 1990 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 1991 NFS_MAXNAMLEN)); 1992 if (info.v3) { 1993 nfsm_v3attrbuild(&info, vap, FALSE); 1994 } 1995 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN)); 1996 if (info.v3 == 0) { 1997 sp = nfsm_build(&info, NFSX_V2SATTR); 1998 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1999 sp->sa_uid = nfs_xdrneg1; 2000 sp->sa_gid = nfs_xdrneg1; 2001 sp->sa_size = nfs_xdrneg1; 2002 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2003 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2004 } 2005 2006 /* 2007 * Issue the NFS request and get the rpc response. 2008 * 2009 * Only NFSv3 responses returning an error of 0 actually return 2010 * a file handle that can be converted into newvp without having 2011 * to do an extra lookup rpc. 2012 */ 2013 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td, 2014 cnp->cn_cred, &error)); 2015 if (info.v3) { 2016 if (error == 0) { 2017 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2018 } 2019 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2020 } 2021 2022 /* 2023 * out code jumps -> here, mrep is also freed. 2024 */ 2025 2026 m_freem(info.mrep); 2027 info.mrep = NULL; 2028 nfsmout: 2029 2030 /* 2031 * If we get an EEXIST error, silently convert it to no-error 2032 * in case of an NFS retry. 2033 */ 2034 if (error == EEXIST) 2035 error = 0; 2036 2037 /* 2038 * If we do not have (or no longer have) an error, and we could 2039 * not extract the newvp from the response due to the request being 2040 * NFSv2 or the error being EEXIST. We have to do a lookup in order 2041 * to obtain a newvp to return. 2042 */ 2043 if (error == 0 && newvp == NULL) { 2044 struct nfsnode *np = NULL; 2045 2046 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 2047 cnp->cn_cred, cnp->cn_td, &np); 2048 if (!error) 2049 newvp = NFSTOV(np); 2050 } 2051 if (error) { 2052 if (newvp) 2053 vput(newvp); 2054 } else { 2055 *ap->a_vpp = newvp; 2056 } 2057 VTONFS(dvp)->n_flag |= NLMODIFIED; 2058 if (!wccflag) 2059 VTONFS(dvp)->n_attrstamp = 0; 2060 return (error); 2061 } 2062 2063 /* 2064 * nfs make dir call 2065 * 2066 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2067 * struct componentname *a_cnp, struct vattr *a_vap) 2068 */ 2069 static int 2070 nfs_mkdir(struct vop_old_mkdir_args *ap) 2071 { 2072 struct vnode *dvp = ap->a_dvp; 2073 struct vattr *vap = ap->a_vap; 2074 struct componentname *cnp = ap->a_cnp; 2075 struct nfsv2_sattr *sp; 2076 struct nfsnode *np = NULL; 2077 struct vnode *newvp = NULL; 2078 struct vattr vattr; 2079 int error = 0, wccflag = NFSV3_WCCRATTR; 2080 int gotvp = 0; 2081 int len; 2082 struct nfsm_info info; 2083 2084 info.mrep = NULL; 2085 info.v3 = NFS_ISV3(dvp); 2086 2087 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 2088 return (error); 2089 } 2090 len = cnp->cn_namelen; 2091 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2092 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR, 2093 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2094 nfsm_rndup(len) + NFSX_SATTR(info.v3)); 2095 ERROROUT(nfsm_fhtom(&info, dvp)); 2096 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN)); 2097 if (info.v3) { 2098 nfsm_v3attrbuild(&info, vap, FALSE); 2099 } else { 2100 sp = nfsm_build(&info, NFSX_V2SATTR); 2101 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2102 sp->sa_uid = nfs_xdrneg1; 2103 sp->sa_gid = nfs_xdrneg1; 2104 sp->sa_size = nfs_xdrneg1; 2105 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2106 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2107 } 2108 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td, 2109 cnp->cn_cred, &error)); 2110 if (error == 0) { 2111 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp)); 2112 } 2113 if (info.v3) { 2114 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2115 } 2116 m_freem(info.mrep); 2117 info.mrep = NULL; 2118 nfsmout: 2119 VTONFS(dvp)->n_flag |= NLMODIFIED; 2120 if (!wccflag) 2121 VTONFS(dvp)->n_attrstamp = 0; 2122 /* 2123 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2124 * if we can succeed in looking up the directory. 2125 */ 2126 if (error == EEXIST || (!error && !gotvp)) { 2127 if (newvp) { 2128 vrele(newvp); 2129 newvp = NULL; 2130 } 2131 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2132 cnp->cn_td, &np); 2133 if (!error) { 2134 newvp = NFSTOV(np); 2135 if (newvp->v_type != VDIR) 2136 error = EEXIST; 2137 } 2138 } 2139 if (error) { 2140 if (newvp) 2141 vrele(newvp); 2142 } else 2143 *ap->a_vpp = newvp; 2144 return (error); 2145 } 2146 2147 /* 2148 * nfs remove directory call 2149 * 2150 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2151 * struct componentname *a_cnp) 2152 */ 2153 static int 2154 nfs_rmdir(struct vop_old_rmdir_args *ap) 2155 { 2156 struct vnode *vp = ap->a_vp; 2157 struct vnode *dvp = ap->a_dvp; 2158 struct componentname *cnp = ap->a_cnp; 2159 int error = 0, wccflag = NFSV3_WCCRATTR; 2160 struct nfsm_info info; 2161 2162 info.mrep = NULL; 2163 info.v3 = NFS_ISV3(dvp); 2164 2165 if (dvp == vp) 2166 return (EINVAL); 2167 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2168 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR, 2169 NFSX_FH(info.v3) + NFSX_UNSIGNED + 2170 nfsm_rndup(cnp->cn_namelen)); 2171 ERROROUT(nfsm_fhtom(&info, dvp)); 2172 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen, 2173 NFS_MAXNAMLEN)); 2174 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td, 2175 cnp->cn_cred, &error)); 2176 if (info.v3) { 2177 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag)); 2178 } 2179 m_freem(info.mrep); 2180 info.mrep = NULL; 2181 nfsmout: 2182 VTONFS(dvp)->n_flag |= NLMODIFIED; 2183 if (!wccflag) 2184 VTONFS(dvp)->n_attrstamp = 0; 2185 /* 2186 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2187 */ 2188 if (error == ENOENT) 2189 error = 0; 2190 return (error); 2191 } 2192 2193 /* 2194 * nfs readdir call 2195 * 2196 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2197 */ 2198 static int 2199 nfs_readdir(struct vop_readdir_args *ap) 2200 { 2201 struct vnode *vp = ap->a_vp; 2202 struct nfsnode *np = VTONFS(vp); 2203 struct uio *uio = ap->a_uio; 2204 int tresid, error; 2205 struct vattr vattr; 2206 2207 if (vp->v_type != VDIR) 2208 return (EPERM); 2209 2210 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 2211 return (error); 2212 2213 /* 2214 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2215 * and then check that is still valid, or if this is an NQNFS mount 2216 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2217 * VOP_GETATTR() does not necessarily go to the wire. 2218 */ 2219 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2220 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2221 if (VOP_GETATTR(vp, &vattr) == 0 && 2222 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2223 ) { 2224 nfsstats.direofcache_hits++; 2225 goto done; 2226 } 2227 } 2228 2229 /* 2230 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2231 * own cache coherency checks so we do not have to. 2232 */ 2233 tresid = uio->uio_resid; 2234 error = nfs_bioread(vp, uio, 0); 2235 2236 if (!error && uio->uio_resid == tresid) 2237 nfsstats.direofcache_misses++; 2238 done: 2239 vn_unlock(vp); 2240 return (error); 2241 } 2242 2243 /* 2244 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2245 * 2246 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2247 * offset/block and converts the nfs formatted directory entries for userland 2248 * consumption as well as deals with offsets into the middle of blocks. 2249 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2250 * be block-bounded. It must convert to cookies for the actual RPC. 2251 */ 2252 int 2253 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop) 2254 { 2255 int len, left; 2256 struct nfs_dirent *dp = NULL; 2257 u_int32_t *tl; 2258 nfsuint64 *cookiep; 2259 caddr_t cp; 2260 nfsuint64 cookie; 2261 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2262 struct nfsnode *dnp = VTONFS(vp); 2263 u_quad_t fileno; 2264 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2265 int attrflag; 2266 struct nfsm_info info; 2267 2268 info.mrep = NULL; 2269 info.v3 = NFS_ISV3(vp); 2270 2271 #ifndef DIAGNOSTIC 2272 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2273 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2274 panic("nfs readdirrpc bad uio"); 2275 #endif 2276 2277 /* 2278 * If there is no cookie, assume directory was stale. 2279 */ 2280 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2281 if (cookiep) 2282 cookie = *cookiep; 2283 else 2284 return (NFSERR_BAD_COOKIE); 2285 /* 2286 * Loop around doing readdir rpc's of size nm_readdirsize 2287 * truncated to a multiple of DIRBLKSIZ. 2288 * The stopping criteria is EOF or buffer full. 2289 */ 2290 while (more_dirs && bigenough) { 2291 nfsstats.rpccnt[NFSPROC_READDIR]++; 2292 nfsm_reqhead(&info, vp, NFSPROC_READDIR, 2293 NFSX_FH(info.v3) + NFSX_READDIR(info.v3)); 2294 ERROROUT(nfsm_fhtom(&info, vp)); 2295 if (info.v3) { 2296 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED); 2297 *tl++ = cookie.nfsuquad[0]; 2298 *tl++ = cookie.nfsuquad[1]; 2299 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2300 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2301 } else { 2302 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED); 2303 *tl++ = cookie.nfsuquad[0]; 2304 } 2305 *tl = txdr_unsigned(nmp->nm_readdirsize); 2306 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR, 2307 uiop->uio_td, 2308 nfs_vpcred(vp, ND_READ), &error)); 2309 if (info.v3) { 2310 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2311 NFS_LATTR_NOSHRINK)); 2312 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2313 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2314 dnp->n_cookieverf.nfsuquad[1] = *tl; 2315 } 2316 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2317 more_dirs = fxdr_unsigned(int, *tl); 2318 2319 /* loop thru the dir entries, converting them to std form */ 2320 while (more_dirs && bigenough) { 2321 if (info.v3) { 2322 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2323 fileno = fxdr_hyper(tl); 2324 len = fxdr_unsigned(int, *(tl + 2)); 2325 } else { 2326 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2327 fileno = fxdr_unsigned(u_quad_t, *tl++); 2328 len = fxdr_unsigned(int, *tl); 2329 } 2330 if (len <= 0 || len > NFS_MAXNAMLEN) { 2331 error = EBADRPC; 2332 m_freem(info.mrep); 2333 info.mrep = NULL; 2334 goto nfsmout; 2335 } 2336 2337 /* 2338 * len is the number of bytes in the path element 2339 * name, not including the \0 termination. 2340 * 2341 * tlen is the number of bytes w have to reserve for 2342 * the path element name. 2343 */ 2344 tlen = nfsm_rndup(len); 2345 if (tlen == len) 2346 tlen += 4; /* To ensure null termination */ 2347 2348 /* 2349 * If the entry would cross a DIRBLKSIZ boundary, 2350 * extend the previous nfs_dirent to cover the 2351 * remaining space. 2352 */ 2353 left = DIRBLKSIZ - blksiz; 2354 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2355 dp->nfs_reclen += left; 2356 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2357 uiop->uio_iov->iov_len -= left; 2358 uiop->uio_offset += left; 2359 uiop->uio_resid -= left; 2360 blksiz = 0; 2361 } 2362 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2363 bigenough = 0; 2364 if (bigenough) { 2365 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2366 dp->nfs_ino = fileno; 2367 dp->nfs_namlen = len; 2368 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2369 dp->nfs_type = DT_UNKNOWN; 2370 blksiz += dp->nfs_reclen; 2371 if (blksiz == DIRBLKSIZ) 2372 blksiz = 0; 2373 uiop->uio_offset += sizeof(struct nfs_dirent); 2374 uiop->uio_resid -= sizeof(struct nfs_dirent); 2375 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2376 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2377 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2378 2379 /* 2380 * The uiop has advanced by nfs_dirent + len 2381 * but really needs to advance by 2382 * nfs_dirent + tlen 2383 */ 2384 cp = uiop->uio_iov->iov_base; 2385 tlen -= len; 2386 *cp = '\0'; /* null terminate */ 2387 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2388 uiop->uio_iov->iov_len -= tlen; 2389 uiop->uio_offset += tlen; 2390 uiop->uio_resid -= tlen; 2391 } else { 2392 /* 2393 * NFS strings must be rounded up (nfsm_myouio 2394 * handled that in the bigenough case). 2395 */ 2396 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2397 } 2398 if (info.v3) { 2399 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2400 } else { 2401 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED)); 2402 } 2403 2404 /* 2405 * If we were able to accomodate the last entry, 2406 * get the cookie for the next one. Otherwise 2407 * hold-over the cookie for the one we were not 2408 * able to accomodate. 2409 */ 2410 if (bigenough) { 2411 cookie.nfsuquad[0] = *tl++; 2412 if (info.v3) 2413 cookie.nfsuquad[1] = *tl++; 2414 } else if (info.v3) { 2415 tl += 2; 2416 } else { 2417 tl++; 2418 } 2419 more_dirs = fxdr_unsigned(int, *tl); 2420 } 2421 /* 2422 * If at end of rpc data, get the eof boolean 2423 */ 2424 if (!more_dirs) { 2425 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2426 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2427 } 2428 m_freem(info.mrep); 2429 info.mrep = NULL; 2430 } 2431 /* 2432 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2433 * by increasing d_reclen for the last record. 2434 */ 2435 if (blksiz > 0) { 2436 left = DIRBLKSIZ - blksiz; 2437 dp->nfs_reclen += left; 2438 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2439 uiop->uio_iov->iov_len -= left; 2440 uiop->uio_offset += left; 2441 uiop->uio_resid -= left; 2442 } 2443 2444 if (bigenough) { 2445 /* 2446 * We hit the end of the directory, update direofoffset. 2447 */ 2448 dnp->n_direofoffset = uiop->uio_offset; 2449 } else { 2450 /* 2451 * There is more to go, insert the link cookie so the 2452 * next block can be read. 2453 */ 2454 if (uiop->uio_resid > 0) 2455 kprintf("EEK! readdirrpc resid > 0\n"); 2456 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2457 *cookiep = cookie; 2458 } 2459 nfsmout: 2460 return (error); 2461 } 2462 2463 /* 2464 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2465 */ 2466 int 2467 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop) 2468 { 2469 int len, left; 2470 struct nfs_dirent *dp; 2471 u_int32_t *tl; 2472 struct vnode *newvp; 2473 nfsuint64 *cookiep; 2474 caddr_t dpossav1, dpossav2; 2475 caddr_t cp; 2476 struct mbuf *mdsav1, *mdsav2; 2477 nfsuint64 cookie; 2478 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2479 struct nfsnode *dnp = VTONFS(vp), *np; 2480 nfsfh_t *fhp; 2481 u_quad_t fileno; 2482 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2483 int attrflag, fhsize; 2484 struct nchandle nch; 2485 struct nchandle dnch; 2486 struct nlcomponent nlc; 2487 struct nfsm_info info; 2488 2489 info.mrep = NULL; 2490 info.v3 = 1; 2491 2492 #ifndef nolint 2493 dp = NULL; 2494 #endif 2495 #ifndef DIAGNOSTIC 2496 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2497 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2498 panic("nfs readdirplusrpc bad uio"); 2499 #endif 2500 /* 2501 * Obtain the namecache record for the directory so we have something 2502 * to use as a basis for creating the entries. This function will 2503 * return a held (but not locked) ncp. The ncp may be disconnected 2504 * from the tree and cannot be used for upward traversals, and the 2505 * ncp may be unnamed. Note that other unrelated operations may 2506 * cause the ncp to be named at any time. 2507 */ 2508 cache_fromdvp(vp, NULL, 0, &dnch); 2509 bzero(&nlc, sizeof(nlc)); 2510 newvp = NULLVP; 2511 2512 /* 2513 * If there is no cookie, assume directory was stale. 2514 */ 2515 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2516 if (cookiep) 2517 cookie = *cookiep; 2518 else 2519 return (NFSERR_BAD_COOKIE); 2520 /* 2521 * Loop around doing readdir rpc's of size nm_readdirsize 2522 * truncated to a multiple of DIRBLKSIZ. 2523 * The stopping criteria is EOF or buffer full. 2524 */ 2525 while (more_dirs && bigenough) { 2526 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2527 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS, 2528 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2529 ERROROUT(nfsm_fhtom(&info, vp)); 2530 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED); 2531 *tl++ = cookie.nfsuquad[0]; 2532 *tl++ = cookie.nfsuquad[1]; 2533 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2534 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2535 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2536 *tl = txdr_unsigned(nmp->nm_rsize); 2537 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS, 2538 uiop->uio_td, 2539 nfs_vpcred(vp, ND_READ), &error)); 2540 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, 2541 NFS_LATTR_NOSHRINK)); 2542 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2543 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2544 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2545 more_dirs = fxdr_unsigned(int, *tl); 2546 2547 /* loop thru the dir entries, doctoring them to 4bsd form */ 2548 while (more_dirs && bigenough) { 2549 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2550 fileno = fxdr_hyper(tl); 2551 len = fxdr_unsigned(int, *(tl + 2)); 2552 if (len <= 0 || len > NFS_MAXNAMLEN) { 2553 error = EBADRPC; 2554 m_freem(info.mrep); 2555 info.mrep = NULL; 2556 goto nfsmout; 2557 } 2558 tlen = nfsm_rndup(len); 2559 if (tlen == len) 2560 tlen += 4; /* To ensure null termination*/ 2561 left = DIRBLKSIZ - blksiz; 2562 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2563 dp->nfs_reclen += left; 2564 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2565 uiop->uio_iov->iov_len -= left; 2566 uiop->uio_offset += left; 2567 uiop->uio_resid -= left; 2568 blksiz = 0; 2569 } 2570 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2571 bigenough = 0; 2572 if (bigenough) { 2573 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2574 dp->nfs_ino = fileno; 2575 dp->nfs_namlen = len; 2576 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2577 dp->nfs_type = DT_UNKNOWN; 2578 blksiz += dp->nfs_reclen; 2579 if (blksiz == DIRBLKSIZ) 2580 blksiz = 0; 2581 uiop->uio_offset += sizeof(struct nfs_dirent); 2582 uiop->uio_resid -= sizeof(struct nfs_dirent); 2583 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent); 2584 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2585 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2586 nlc.nlc_namelen = len; 2587 ERROROUT(nfsm_mtouio(&info, uiop, len)); 2588 cp = uiop->uio_iov->iov_base; 2589 tlen -= len; 2590 *cp = '\0'; 2591 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen; 2592 uiop->uio_iov->iov_len -= tlen; 2593 uiop->uio_offset += tlen; 2594 uiop->uio_resid -= tlen; 2595 } else { 2596 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 2597 } 2598 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 2599 if (bigenough) { 2600 cookie.nfsuquad[0] = *tl++; 2601 cookie.nfsuquad[1] = *tl++; 2602 } else 2603 tl += 2; 2604 2605 /* 2606 * Since the attributes are before the file handle 2607 * (sigh), we must skip over the attributes and then 2608 * come back and get them. 2609 */ 2610 attrflag = fxdr_unsigned(int, *tl); 2611 if (attrflag) { 2612 dpossav1 = info.dpos; 2613 mdsav1 = info.md; 2614 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR)); 2615 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2616 doit = fxdr_unsigned(int, *tl); 2617 if (doit) { 2618 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp)); 2619 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2620 vref(vp); 2621 newvp = vp; 2622 np = dnp; 2623 } else { 2624 error = nfs_nget(vp->v_mount, fhp, 2625 fhsize, &np); 2626 if (error) 2627 doit = 0; 2628 else 2629 newvp = NFSTOV(np); 2630 } 2631 } 2632 if (doit && bigenough) { 2633 dpossav2 = info.dpos; 2634 info.dpos = dpossav1; 2635 mdsav2 = info.md; 2636 info.md = mdsav1; 2637 ERROROUT(nfsm_loadattr(&info, newvp, NULL)); 2638 info.dpos = dpossav2; 2639 info.md = mdsav2; 2640 dp->nfs_type = 2641 IFTODT(VTTOIF(np->n_vattr.va_type)); 2642 if (dnch.ncp) { 2643 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n", 2644 nlc.nlc_namelen, nlc.nlc_namelen, 2645 nlc.nlc_nameptr); 2646 nch = cache_nlookup(&dnch, &nlc); 2647 cache_setunresolved(&nch); 2648 nfs_cache_setvp(&nch, newvp, 2649 nfspos_cache_timeout); 2650 cache_put(&nch); 2651 } else { 2652 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER" 2653 " %*.*s\n", 2654 nlc.nlc_namelen, nlc.nlc_namelen, 2655 nlc.nlc_nameptr); 2656 } 2657 } 2658 } else { 2659 /* Just skip over the file handle */ 2660 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2661 i = fxdr_unsigned(int, *tl); 2662 ERROROUT(nfsm_adv(&info, nfsm_rndup(i))); 2663 } 2664 if (newvp != NULLVP) { 2665 if (newvp == vp) 2666 vrele(newvp); 2667 else 2668 vput(newvp); 2669 newvp = NULLVP; 2670 } 2671 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2672 more_dirs = fxdr_unsigned(int, *tl); 2673 } 2674 /* 2675 * If at end of rpc data, get the eof boolean 2676 */ 2677 if (!more_dirs) { 2678 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED)); 2679 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2680 } 2681 m_freem(info.mrep); 2682 info.mrep = NULL; 2683 } 2684 /* 2685 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2686 * by increasing d_reclen for the last record. 2687 */ 2688 if (blksiz > 0) { 2689 left = DIRBLKSIZ - blksiz; 2690 dp->nfs_reclen += left; 2691 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; 2692 uiop->uio_iov->iov_len -= left; 2693 uiop->uio_offset += left; 2694 uiop->uio_resid -= left; 2695 } 2696 2697 /* 2698 * We are now either at the end of the directory or have filled the 2699 * block. 2700 */ 2701 if (bigenough) 2702 dnp->n_direofoffset = uiop->uio_offset; 2703 else { 2704 if (uiop->uio_resid > 0) 2705 kprintf("EEK! readdirplusrpc resid > 0\n"); 2706 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2707 *cookiep = cookie; 2708 } 2709 nfsmout: 2710 if (newvp != NULLVP) { 2711 if (newvp == vp) 2712 vrele(newvp); 2713 else 2714 vput(newvp); 2715 newvp = NULLVP; 2716 } 2717 if (dnch.ncp) 2718 cache_drop(&dnch); 2719 return (error); 2720 } 2721 2722 /* 2723 * Silly rename. To make the NFS filesystem that is stateless look a little 2724 * more like the "ufs" a remove of an active vnode is translated to a rename 2725 * to a funny looking filename that is removed by nfs_inactive on the 2726 * nfsnode. There is the potential for another process on a different client 2727 * to create the same funny name between the nfs_lookitup() fails and the 2728 * nfs_rename() completes, but... 2729 */ 2730 static int 2731 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2732 { 2733 struct sillyrename *sp; 2734 struct nfsnode *np; 2735 int error; 2736 2737 /* 2738 * We previously purged dvp instead of vp. I don't know why, it 2739 * completely destroys performance. We can't do it anyway with the 2740 * new VFS API since we would be breaking the namecache topology. 2741 */ 2742 cache_purge(vp); /* XXX */ 2743 np = VTONFS(vp); 2744 #ifndef DIAGNOSTIC 2745 if (vp->v_type == VDIR) 2746 panic("nfs: sillyrename dir"); 2747 #endif 2748 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2749 M_NFSREQ, M_WAITOK); 2750 sp->s_cred = crdup(cnp->cn_cred); 2751 sp->s_dvp = dvp; 2752 vref(dvp); 2753 2754 /* Fudge together a funny name */ 2755 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", 2756 (int)(intptr_t)cnp->cn_td); 2757 2758 /* Try lookitups until we get one that isn't there */ 2759 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2760 cnp->cn_td, NULL) == 0) { 2761 sp->s_name[4]++; 2762 if (sp->s_name[4] > 'z') { 2763 error = EINVAL; 2764 goto bad; 2765 } 2766 } 2767 error = nfs_renameit(dvp, cnp, sp); 2768 if (error) 2769 goto bad; 2770 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2771 cnp->cn_td, &np); 2772 np->n_sillyrename = sp; 2773 return (0); 2774 bad: 2775 vrele(sp->s_dvp); 2776 crfree(sp->s_cred); 2777 kfree((caddr_t)sp, M_NFSREQ); 2778 return (error); 2779 } 2780 2781 /* 2782 * Look up a file name and optionally either update the file handle or 2783 * allocate an nfsnode, depending on the value of npp. 2784 * npp == NULL --> just do the lookup 2785 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2786 * handled too 2787 * *npp != NULL --> update the file handle in the vnode 2788 */ 2789 static int 2790 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2791 struct thread *td, struct nfsnode **npp) 2792 { 2793 struct vnode *newvp = NULL; 2794 struct nfsnode *np, *dnp = VTONFS(dvp); 2795 int error = 0, fhlen, attrflag; 2796 nfsfh_t *nfhp; 2797 struct nfsm_info info; 2798 2799 info.mrep = NULL; 2800 info.v3 = NFS_ISV3(dvp); 2801 2802 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2803 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP, 2804 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2805 ERROROUT(nfsm_fhtom(&info, dvp)); 2806 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN)); 2807 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error)); 2808 if (npp && !error) { 2809 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp)); 2810 if (*npp) { 2811 np = *npp; 2812 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2813 kfree((caddr_t)np->n_fhp, M_NFSBIGFH); 2814 np->n_fhp = &np->n_fh; 2815 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2816 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK); 2817 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2818 np->n_fhsize = fhlen; 2819 newvp = NFSTOV(np); 2820 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2821 vref(dvp); 2822 newvp = dvp; 2823 } else { 2824 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2825 if (error) { 2826 m_freem(info.mrep); 2827 info.mrep = NULL; 2828 return (error); 2829 } 2830 newvp = NFSTOV(np); 2831 } 2832 if (info.v3) { 2833 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag, 2834 NFS_LATTR_NOSHRINK)); 2835 if (!attrflag && *npp == NULL) { 2836 m_freem(info.mrep); 2837 info.mrep = NULL; 2838 if (newvp == dvp) 2839 vrele(newvp); 2840 else 2841 vput(newvp); 2842 return (ENOENT); 2843 } 2844 } else { 2845 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL)); 2846 } 2847 } 2848 m_freem(info.mrep); 2849 info.mrep = NULL; 2850 nfsmout: 2851 if (npp && *npp == NULL) { 2852 if (error) { 2853 if (newvp) { 2854 if (newvp == dvp) 2855 vrele(newvp); 2856 else 2857 vput(newvp); 2858 } 2859 } else 2860 *npp = np; 2861 } 2862 return (error); 2863 } 2864 2865 /* 2866 * Nfs Version 3 commit rpc 2867 * 2868 * We call it 'uio' to distinguish it from 'bio' but there is no real uio 2869 * involved. 2870 */ 2871 int 2872 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2873 { 2874 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2875 int error = 0, wccflag = NFSV3_WCCRATTR; 2876 struct nfsm_info info; 2877 u_int32_t *tl; 2878 2879 info.mrep = NULL; 2880 info.v3 = 1; 2881 2882 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2883 return (0); 2884 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2885 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1)); 2886 ERROROUT(nfsm_fhtom(&info, vp)); 2887 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED); 2888 txdr_hyper(offset, tl); 2889 tl += 2; 2890 *tl = txdr_unsigned(cnt); 2891 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td, 2892 nfs_vpcred(vp, ND_WRITE), &error)); 2893 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag)); 2894 if (!error) { 2895 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF)); 2896 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2897 NFSX_V3WRITEVERF)) { 2898 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2899 NFSX_V3WRITEVERF); 2900 error = NFSERR_STALEWRITEVERF; 2901 } 2902 } 2903 m_freem(info.mrep); 2904 info.mrep = NULL; 2905 nfsmout: 2906 return (error); 2907 } 2908 2909 /* 2910 * Kludge City.. 2911 * - make nfs_bmap() essentially a no-op that does no translation 2912 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2913 * (Maybe I could use the process's page mapping, but I was concerned that 2914 * Kernel Write might not be enabled and also figured copyout() would do 2915 * a lot more work than bcopy() and also it currently happens in the 2916 * context of the swapper process (2). 2917 * 2918 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, 2919 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2920 */ 2921 static int 2922 nfs_bmap(struct vop_bmap_args *ap) 2923 { 2924 if (ap->a_doffsetp != NULL) 2925 *ap->a_doffsetp = ap->a_loffset; 2926 if (ap->a_runp != NULL) 2927 *ap->a_runp = 0; 2928 if (ap->a_runb != NULL) 2929 *ap->a_runb = 0; 2930 return (0); 2931 } 2932 2933 /* 2934 * Strategy routine. 2935 */ 2936 static int 2937 nfs_strategy(struct vop_strategy_args *ap) 2938 { 2939 struct bio *bio = ap->a_bio; 2940 struct bio *nbio; 2941 struct buf *bp = bio->bio_buf; 2942 struct thread *td; 2943 int error; 2944 2945 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2946 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2947 KASSERT(BUF_REFCNT(bp) > 0, 2948 ("nfs_strategy: buffer %p not locked", bp)); 2949 2950 if (bio->bio_flags & BIO_SYNC) 2951 td = curthread; /* XXX */ 2952 else 2953 td = NULL; 2954 2955 /* 2956 * We probably don't need to push an nbio any more since no 2957 * block conversion is required due to the use of 64 bit byte 2958 * offsets, but do it anyway. 2959 * 2960 * NOTE: When NFS callers itself via this strategy routines and 2961 * sets up a synchronous I/O, it expects the I/O to run 2962 * synchronously (its bio_done routine just assumes it), 2963 * so for now we have to honor the bit. 2964 */ 2965 nbio = push_bio(bio); 2966 nbio->bio_offset = bio->bio_offset; 2967 nbio->bio_flags = bio->bio_flags & BIO_SYNC; 2968 2969 /* 2970 * If the op is asynchronous and an i/o daemon is waiting 2971 * queue the request, wake it up and wait for completion 2972 * otherwise just do it ourselves. 2973 */ 2974 if (bio->bio_flags & BIO_SYNC) { 2975 error = nfs_doio(ap->a_vp, nbio, td); 2976 } else { 2977 nfs_asyncio(ap->a_vp, nbio); 2978 error = 0; 2979 } 2980 return (error); 2981 } 2982 2983 /* 2984 * Mmap a file 2985 * 2986 * NB Currently unsupported. 2987 * 2988 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 2989 */ 2990 /* ARGSUSED */ 2991 static int 2992 nfs_mmap(struct vop_mmap_args *ap) 2993 { 2994 return (EINVAL); 2995 } 2996 2997 /* 2998 * fsync vnode op. Just call nfs_flush() with commit == 1. 2999 * 3000 * nfs_fsync(struct vnode *a_vp, int a_waitfor) 3001 */ 3002 /* ARGSUSED */ 3003 static int 3004 nfs_fsync(struct vop_fsync_args *ap) 3005 { 3006 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1)); 3007 } 3008 3009 /* 3010 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 3011 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 3012 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 3013 * set the buffer contains data that has already been written to the server 3014 * and which now needs a commit RPC. 3015 * 3016 * If commit is 0 we only take one pass and only flush buffers containing new 3017 * dirty data. 3018 * 3019 * If commit is 1 we take two passes, issuing a commit RPC in the second 3020 * pass. 3021 * 3022 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 3023 * to completely flush all pending data. 3024 * 3025 * Note that the RB_SCAN code properly handles the case where the 3026 * callback might block and directly or indirectly (another thread) cause 3027 * the RB tree to change. 3028 */ 3029 3030 #ifndef NFS_COMMITBVECSIZ 3031 #define NFS_COMMITBVECSIZ 16 3032 #endif 3033 3034 struct nfs_flush_info { 3035 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 3036 struct thread *td; 3037 struct vnode *vp; 3038 int waitfor; 3039 int slpflag; 3040 int slptimeo; 3041 int loops; 3042 struct buf *bvary[NFS_COMMITBVECSIZ]; 3043 int bvsize; 3044 off_t beg_off; 3045 off_t end_off; 3046 }; 3047 3048 static int nfs_flush_bp(struct buf *bp, void *data); 3049 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 3050 3051 int 3052 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 3053 { 3054 struct nfsnode *np = VTONFS(vp); 3055 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 3056 struct nfs_flush_info info; 3057 lwkt_tokref vlock; 3058 int error; 3059 3060 bzero(&info, sizeof(info)); 3061 info.td = td; 3062 info.vp = vp; 3063 info.waitfor = waitfor; 3064 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 3065 info.loops = 0; 3066 lwkt_gettoken(&vlock, &vp->v_token); 3067 3068 do { 3069 /* 3070 * Flush mode 3071 */ 3072 info.mode = NFI_FLUSHNEW; 3073 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3074 nfs_flush_bp, &info); 3075 3076 /* 3077 * Take a second pass if committing and no error occured. 3078 * Clean up any left over collection (whether an error 3079 * occurs or not). 3080 */ 3081 if (commit && error == 0) { 3082 info.mode = NFI_COMMIT; 3083 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3084 nfs_flush_bp, &info); 3085 if (info.bvsize) 3086 error = nfs_flush_docommit(&info, error); 3087 } 3088 3089 /* 3090 * Wait for pending I/O to complete before checking whether 3091 * any further dirty buffers exist. 3092 */ 3093 while (waitfor == MNT_WAIT && 3094 bio_track_active(&vp->v_track_write)) { 3095 error = bio_track_wait(&vp->v_track_write, 3096 info.slpflag, info.slptimeo); 3097 if (error) { 3098 /* 3099 * We have to be able to break out if this 3100 * is an 'intr' mount. 3101 */ 3102 if (nfs_sigintr(nmp, NULL, td)) { 3103 error = -EINTR; 3104 break; 3105 } 3106 3107 /* 3108 * Since we do not process pending signals, 3109 * once we get a PCATCH our tsleep() will no 3110 * longer sleep, switch to a fixed timeout 3111 * instead. 3112 */ 3113 if (info.slpflag == PCATCH) { 3114 info.slpflag = 0; 3115 info.slptimeo = 2 * hz; 3116 } 3117 error = 0; 3118 } 3119 } 3120 ++info.loops; 3121 /* 3122 * Loop if we are flushing synchronous as well as committing, 3123 * and dirty buffers are still present. Otherwise we might livelock. 3124 */ 3125 } while (waitfor == MNT_WAIT && commit && 3126 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3127 3128 /* 3129 * The callbacks have to return a negative error to terminate the 3130 * RB scan. 3131 */ 3132 if (error < 0) 3133 error = -error; 3134 3135 /* 3136 * Deal with any error collection 3137 */ 3138 if (np->n_flag & NWRITEERR) { 3139 error = np->n_error; 3140 np->n_flag &= ~NWRITEERR; 3141 } 3142 lwkt_reltoken(&vlock); 3143 return (error); 3144 } 3145 3146 static 3147 int 3148 nfs_flush_bp(struct buf *bp, void *data) 3149 { 3150 struct nfs_flush_info *info = data; 3151 int lkflags; 3152 int error; 3153 off_t toff; 3154 3155 error = 0; 3156 switch(info->mode) { 3157 case NFI_FLUSHNEW: 3158 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3159 if (error && info->loops && info->waitfor == MNT_WAIT) { 3160 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3161 if (error) { 3162 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3163 if (info->slpflag & PCATCH) 3164 lkflags |= LK_PCATCH; 3165 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3166 info->slptimeo); 3167 } 3168 } 3169 3170 /* 3171 * Ignore locking errors 3172 */ 3173 if (error) { 3174 error = 0; 3175 break; 3176 } 3177 3178 /* 3179 * The buffer may have changed out from under us, even if 3180 * we did not block (MPSAFE). Check again now that it is 3181 * locked. 3182 */ 3183 if (bp->b_vp == info->vp && 3184 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) { 3185 bremfree(bp); 3186 bawrite(bp); 3187 } else { 3188 BUF_UNLOCK(bp); 3189 } 3190 break; 3191 case NFI_COMMIT: 3192 /* 3193 * Only process buffers in need of a commit which we can 3194 * immediately lock. This may prevent a buffer from being 3195 * committed, but the normal flush loop will block on the 3196 * same buffer so we shouldn't get into an endless loop. 3197 */ 3198 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3199 (B_DELWRI | B_NEEDCOMMIT)) { 3200 break; 3201 } 3202 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 3203 break; 3204 3205 /* 3206 * We must recheck after successfully locking the buffer. 3207 */ 3208 if (bp->b_vp != info->vp || 3209 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3210 (B_DELWRI | B_NEEDCOMMIT)) { 3211 BUF_UNLOCK(bp); 3212 break; 3213 } 3214 3215 /* 3216 * NOTE: storing the bp in the bvary[] basically sets 3217 * it up for a commit operation. 3218 * 3219 * We must call vfs_busy_pages() now so the commit operation 3220 * is interlocked with user modifications to memory mapped 3221 * pages. 3222 * 3223 * Note: to avoid loopback deadlocks, we do not 3224 * assign b_runningbufspace. 3225 */ 3226 bremfree(bp); 3227 bp->b_cmd = BUF_CMD_WRITE; 3228 vfs_busy_pages(bp->b_vp, bp); 3229 info->bvary[info->bvsize] = bp; 3230 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3231 if (info->bvsize == 0 || toff < info->beg_off) 3232 info->beg_off = toff; 3233 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3234 if (info->bvsize == 0 || toff > info->end_off) 3235 info->end_off = toff; 3236 ++info->bvsize; 3237 if (info->bvsize == NFS_COMMITBVECSIZ) { 3238 error = nfs_flush_docommit(info, 0); 3239 KKASSERT(info->bvsize == 0); 3240 } 3241 } 3242 return (error); 3243 } 3244 3245 static 3246 int 3247 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3248 { 3249 struct vnode *vp; 3250 struct buf *bp; 3251 off_t bytes; 3252 int retv; 3253 int i; 3254 3255 vp = info->vp; 3256 3257 if (info->bvsize > 0) { 3258 /* 3259 * Commit data on the server, as required. Note that 3260 * nfs_commit will use the vnode's cred for the commit. 3261 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3262 */ 3263 bytes = info->end_off - info->beg_off; 3264 if (bytes > 0x40000000) 3265 bytes = 0x40000000; 3266 if (error) { 3267 retv = -error; 3268 } else { 3269 retv = nfs_commitrpc_uio(vp, info->beg_off, 3270 (int)bytes, info->td); 3271 if (retv == NFSERR_STALEWRITEVERF) 3272 nfs_clearcommit(vp->v_mount); 3273 } 3274 3275 /* 3276 * Now, either mark the blocks I/O done or mark the 3277 * blocks dirty, depending on whether the commit 3278 * succeeded. 3279 */ 3280 for (i = 0; i < info->bvsize; ++i) { 3281 bp = info->bvary[i]; 3282 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3283 if (retv) { 3284 /* 3285 * Error, leave B_DELWRI intact 3286 */ 3287 vfs_unbusy_pages(bp); 3288 bp->b_cmd = BUF_CMD_DONE; 3289 brelse(bp); 3290 } else { 3291 /* 3292 * Success, remove B_DELWRI ( bundirty() ). 3293 * 3294 * b_dirtyoff/b_dirtyend seem to be NFS 3295 * specific. We should probably move that 3296 * into bundirty(). XXX 3297 * 3298 * We are faking an I/O write, we have to 3299 * start the transaction in order to 3300 * immediately biodone() it. 3301 */ 3302 bundirty(bp); 3303 bp->b_flags &= ~B_ERROR; 3304 bp->b_dirtyoff = bp->b_dirtyend = 0; 3305 biodone(&bp->b_bio1); 3306 } 3307 } 3308 info->bvsize = 0; 3309 } 3310 return (error); 3311 } 3312 3313 /* 3314 * NFS advisory byte-level locks. 3315 * Currently unsupported. 3316 * 3317 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3318 * int a_flags) 3319 */ 3320 static int 3321 nfs_advlock(struct vop_advlock_args *ap) 3322 { 3323 struct nfsnode *np = VTONFS(ap->a_vp); 3324 3325 /* 3326 * The following kludge is to allow diskless support to work 3327 * until a real NFS lockd is implemented. Basically, just pretend 3328 * that this is a local lock. 3329 */ 3330 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3331 } 3332 3333 /* 3334 * Print out the contents of an nfsnode. 3335 * 3336 * nfs_print(struct vnode *a_vp) 3337 */ 3338 static int 3339 nfs_print(struct vop_print_args *ap) 3340 { 3341 struct vnode *vp = ap->a_vp; 3342 struct nfsnode *np = VTONFS(vp); 3343 3344 kprintf("tag VT_NFS, fileid %lld fsid 0x%x", 3345 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3346 if (vp->v_type == VFIFO) 3347 fifo_printinfo(vp); 3348 kprintf("\n"); 3349 return (0); 3350 } 3351 3352 /* 3353 * nfs special file access vnode op. 3354 * 3355 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 3356 */ 3357 static int 3358 nfs_laccess(struct vop_access_args *ap) 3359 { 3360 struct vattr vattr; 3361 int error; 3362 3363 error = VOP_GETATTR(ap->a_vp, &vattr); 3364 if (!error) 3365 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid, 3366 vattr.va_mode, 0); 3367 return (error); 3368 } 3369 3370 /* 3371 * Read wrapper for fifos. 3372 * 3373 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3374 * struct ucred *a_cred) 3375 */ 3376 static int 3377 nfsfifo_read(struct vop_read_args *ap) 3378 { 3379 struct nfsnode *np = VTONFS(ap->a_vp); 3380 3381 /* 3382 * Set access flag. 3383 */ 3384 np->n_flag |= NACC; 3385 getnanotime(&np->n_atim); 3386 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3387 } 3388 3389 /* 3390 * Write wrapper for fifos. 3391 * 3392 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3393 * struct ucred *a_cred) 3394 */ 3395 static int 3396 nfsfifo_write(struct vop_write_args *ap) 3397 { 3398 struct nfsnode *np = VTONFS(ap->a_vp); 3399 3400 /* 3401 * Set update flag. 3402 */ 3403 np->n_flag |= NUPD; 3404 getnanotime(&np->n_mtim); 3405 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3406 } 3407 3408 /* 3409 * Close wrapper for fifos. 3410 * 3411 * Update the times on the nfsnode then do fifo close. 3412 * 3413 * nfsfifo_close(struct vnode *a_vp, int a_fflag) 3414 */ 3415 static int 3416 nfsfifo_close(struct vop_close_args *ap) 3417 { 3418 struct vnode *vp = ap->a_vp; 3419 struct nfsnode *np = VTONFS(vp); 3420 struct vattr vattr; 3421 struct timespec ts; 3422 3423 if (np->n_flag & (NACC | NUPD)) { 3424 getnanotime(&ts); 3425 if (np->n_flag & NACC) 3426 np->n_atim = ts; 3427 if (np->n_flag & NUPD) 3428 np->n_mtim = ts; 3429 np->n_flag |= NCHG; 3430 if (vp->v_sysref.refcnt == 1 && 3431 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3432 VATTR_NULL(&vattr); 3433 if (np->n_flag & NACC) 3434 vattr.va_atime = np->n_atim; 3435 if (np->n_flag & NUPD) 3436 vattr.va_mtime = np->n_mtim; 3437 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3438 } 3439 } 3440 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3441 } 3442 3443