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