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 1.71 2007/05/09 00:53:35 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 /* 217 * Global variables 218 */ 219 extern u_int32_t nfs_true, nfs_false; 220 extern u_int32_t nfs_xdrneg1; 221 extern struct nfsstats nfsstats; 222 extern nfstype nfsv3_type[9]; 223 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 224 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 225 int nfs_numasync = 0; 226 227 SYSCTL_DECL(_vfs_nfs); 228 229 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 230 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 231 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 232 233 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 234 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 235 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache 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 const int v3 = 1; 256 u_int32_t *tl; 257 int error = 0, attrflag; 258 259 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 260 caddr_t bpos, dpos, cp2; 261 int32_t t1, t2; 262 caddr_t cp; 263 u_int32_t rmode; 264 struct nfsnode *np = VTONFS(vp); 265 266 nfsstats.rpccnt[NFSPROC_ACCESS]++; 267 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 268 nfsm_fhtom(vp, v3); 269 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 270 *tl = txdr_unsigned(wmode); 271 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 272 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 273 if (!error) { 274 nfsm_dissect(tl, u_int32_t *, 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(mrep); 281 nfsmout: 282 return error; 283 } 284 285 /* 286 * nfs access vnode op. 287 * For nfs version 2, just return ok. File accesses may fail later. 288 * For nfs version 3, use the access rpc to check accessibility. If file modes 289 * are changed on the server, accesses might still fail later. 290 * 291 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 292 * struct thread *a_td) 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 int v3 = NFS_ISV3(vp); 302 struct nfsnode *np = VTONFS(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(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(vp, &auio); 409 kfree(bp, M_TEMP); 410 } else if (vp->v_type == VLNK) { 411 error = nfs_readlinkrpc(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 */ 448 /* ARGSUSED */ 449 static int 450 nfs_open(struct vop_open_args *ap) 451 { 452 struct vnode *vp = ap->a_vp; 453 struct nfsnode *np = VTONFS(vp); 454 struct vattr vattr; 455 int error; 456 457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 458 #ifdef DIAGNOSTIC 459 kprintf("open eacces vtyp=%d\n",vp->v_type); 460 #endif 461 return (EOPNOTSUPP); 462 } 463 464 /* 465 * Clear the attribute cache only if opening with write access. It 466 * is unclear if we should do this at all here, but we certainly 467 * should not clear the cache unconditionally simply because a file 468 * is being opened. 469 */ 470 if (ap->a_mode & FWRITE) 471 np->n_attrstamp = 0; 472 473 /* 474 * For normal NFS, reconcile changes made locally verses 475 * changes made remotely. Note that VOP_GETATTR only goes 476 * to the wire if the cached attribute has timed out or been 477 * cleared. 478 * 479 * If local modifications have been made clear the attribute 480 * cache to force an attribute and modified time check. If 481 * GETATTR detects that the file has been changed by someone 482 * other then us it will set NRMODIFIED. 483 * 484 * If we are opening a directory and local changes have been 485 * made we have to invalidate the cache in order to ensure 486 * that we get the most up-to-date information from the 487 * server. XXX 488 */ 489 if (np->n_flag & NLMODIFIED) { 490 np->n_attrstamp = 0; 491 if (vp->v_type == VDIR) { 492 error = nfs_vinvalbuf(vp, V_SAVE, 1); 493 if (error == EINTR) 494 return (error); 495 nfs_invaldir(vp); 496 } 497 } 498 error = VOP_GETATTR(vp, &vattr); 499 if (error) 500 return (error); 501 if (np->n_flag & NRMODIFIED) { 502 if (vp->v_type == VDIR) 503 nfs_invaldir(vp); 504 error = nfs_vinvalbuf(vp, V_SAVE, 1); 505 if (error == EINTR) 506 return (error); 507 np->n_flag &= ~NRMODIFIED; 508 } 509 510 return (vop_stdopen(ap)); 511 } 512 513 /* 514 * nfs close vnode op 515 * What an NFS client should do upon close after writing is a debatable issue. 516 * Most NFS clients push delayed writes to the server upon close, basically for 517 * two reasons: 518 * 1 - So that any write errors may be reported back to the client process 519 * doing the close system call. By far the two most likely errors are 520 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 521 * 2 - To put a worst case upper bound on cache inconsistency between 522 * multiple clients for the file. 523 * There is also a consistency problem for Version 2 of the protocol w.r.t. 524 * not being able to tell if other clients are writing a file concurrently, 525 * since there is no way of knowing if the changed modify time in the reply 526 * is only due to the write for this client. 527 * (NFS Version 3 provides weak cache consistency data in the reply that 528 * should be sufficient to detect and handle this case.) 529 * 530 * The current code does the following: 531 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 532 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 533 * or commit them (this satisfies 1 and 2 except for the 534 * case where the server crashes after this close but 535 * before the commit RPC, which is felt to be "good 536 * enough". Changing the last argument to nfs_flush() to 537 * a 1 would force a commit operation, if it is felt a 538 * commit is necessary now. 539 * for NQNFS - do nothing now, since 2 is dealt with via leases and 540 * 1 should be dealt with via an fsync() system call for 541 * cases where write errors are important. 542 * 543 * nfs_close(struct vnode *a_vp, int a_fflag, 544 * struct ucred *a_cred, struct thread *a_td) 545 */ 546 /* ARGSUSED */ 547 static int 548 nfs_close(struct vop_close_args *ap) 549 { 550 struct vnode *vp = ap->a_vp; 551 struct nfsnode *np = VTONFS(vp); 552 int error = 0; 553 thread_t td = curthread; 554 555 if (vp->v_type == VREG) { 556 if (np->n_flag & NLMODIFIED) { 557 if (NFS_ISV3(vp)) { 558 /* 559 * Under NFSv3 we have dirty buffers to dispose of. We 560 * must flush them to the NFS server. We have the option 561 * of waiting all the way through the commit rpc or just 562 * waiting for the initial write. The default is to only 563 * wait through the initial write so the data is in the 564 * server's cache, which is roughly similar to the state 565 * a standard disk subsystem leaves the file in on close(). 566 * 567 * We cannot clear the NLMODIFIED bit in np->n_flag due to 568 * potential races with other processes, and certainly 569 * cannot clear it if we don't commit. 570 */ 571 int cm = nfsv3_commit_on_close ? 1 : 0; 572 error = nfs_flush(vp, MNT_WAIT, td, cm); 573 /* np->n_flag &= ~NLMODIFIED; */ 574 } else { 575 error = nfs_vinvalbuf(vp, V_SAVE, 1); 576 } 577 np->n_attrstamp = 0; 578 } 579 if (np->n_flag & NWRITEERR) { 580 np->n_flag &= ~NWRITEERR; 581 error = np->n_error; 582 } 583 } 584 vop_stdclose(ap); 585 return (error); 586 } 587 588 /* 589 * nfs getattr call from vfs. 590 * 591 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 592 * struct thread *a_td) 593 */ 594 static int 595 nfs_getattr(struct vop_getattr_args *ap) 596 { 597 struct vnode *vp = ap->a_vp; 598 struct nfsnode *np = VTONFS(vp); 599 caddr_t cp; 600 u_int32_t *tl; 601 int32_t t1, t2; 602 caddr_t bpos, dpos; 603 int error = 0; 604 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 605 int v3 = NFS_ISV3(vp); 606 thread_t td = curthread; 607 608 /* 609 * Update local times for special files. 610 */ 611 if (np->n_flag & (NACC | NUPD)) 612 np->n_flag |= NCHG; 613 /* 614 * First look in the cache. 615 */ 616 if (nfs_getattrcache(vp, ap->a_vap) == 0) 617 return (0); 618 619 if (v3 && nfsaccess_cache_timeout > 0) { 620 nfsstats.accesscache_misses++; 621 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 622 if (nfs_getattrcache(vp, ap->a_vap) == 0) 623 return (0); 624 } 625 626 nfsstats.rpccnt[NFSPROC_GETATTR]++; 627 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 628 nfsm_fhtom(vp, v3); 629 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK)); 630 if (!error) { 631 nfsm_loadattr(vp, ap->a_vap); 632 } 633 m_freem(mrep); 634 nfsmout: 635 return (error); 636 } 637 638 /* 639 * nfs setattr call. 640 * 641 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 642 */ 643 static int 644 nfs_setattr(struct vop_setattr_args *ap) 645 { 646 struct vnode *vp = ap->a_vp; 647 struct nfsnode *np = VTONFS(vp); 648 struct vattr *vap = ap->a_vap; 649 int error = 0; 650 u_quad_t tsize; 651 thread_t td = curthread; 652 653 #ifndef nolint 654 tsize = (u_quad_t)0; 655 #endif 656 657 /* 658 * Setting of flags is not supported. 659 */ 660 if (vap->va_flags != VNOVAL) 661 return (EOPNOTSUPP); 662 663 /* 664 * Disallow write attempts if the filesystem is mounted read-only. 665 */ 666 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 667 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 668 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 669 (vp->v_mount->mnt_flag & MNT_RDONLY)) 670 return (EROFS); 671 if (vap->va_size != VNOVAL) { 672 switch (vp->v_type) { 673 case VDIR: 674 return (EISDIR); 675 case VCHR: 676 case VBLK: 677 case VSOCK: 678 case VFIFO: 679 if (vap->va_mtime.tv_sec == VNOVAL && 680 vap->va_atime.tv_sec == VNOVAL && 681 vap->va_mode == (mode_t)VNOVAL && 682 vap->va_uid == (uid_t)VNOVAL && 683 vap->va_gid == (gid_t)VNOVAL) 684 return (0); 685 vap->va_size = VNOVAL; 686 break; 687 default: 688 /* 689 * Disallow write attempts if the filesystem is 690 * mounted read-only. 691 */ 692 if (vp->v_mount->mnt_flag & MNT_RDONLY) 693 return (EROFS); 694 695 /* 696 * This is nasty. The RPCs we send to flush pending 697 * data often return attribute information which is 698 * cached via a callback to nfs_loadattrcache(), which 699 * has the effect of changing our notion of the file 700 * size. Due to flushed appends and other operations 701 * the file size can be set to virtually anything, 702 * including values that do not match either the old 703 * or intended file size. 704 * 705 * When this condition is detected we must loop to 706 * try the operation again. Hopefully no more 707 * flushing is required on the loop so it works the 708 * second time around. THIS CASE ALMOST ALWAYS 709 * HAPPENS! 710 */ 711 tsize = np->n_size; 712 again: 713 error = nfs_meta_setsize(vp, td, vap->va_size); 714 715 if (np->n_flag & NLMODIFIED) { 716 if (vap->va_size == 0) 717 error = nfs_vinvalbuf(vp, 0, 1); 718 else 719 error = nfs_vinvalbuf(vp, V_SAVE, 1); 720 } 721 /* 722 * note: this loop case almost always happens at 723 * least once per truncation. 724 */ 725 if (error == 0 && np->n_size != vap->va_size) 726 goto again; 727 np->n_vattr.va_size = vap->va_size; 728 break; 729 } 730 } else if ((vap->va_mtime.tv_sec != VNOVAL || 731 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 732 vp->v_type == VREG && 733 (error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR 734 ) { 735 return (error); 736 } 737 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 738 739 /* 740 * Sanity check if a truncation was issued. This should only occur 741 * if multiple processes are racing on the same file. 742 */ 743 if (error == 0 && vap->va_size != VNOVAL && 744 np->n_size != vap->va_size) { 745 kprintf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 746 goto again; 747 } 748 if (error && vap->va_size != VNOVAL) { 749 np->n_size = np->n_vattr.va_size = tsize; 750 vnode_pager_setsize(vp, np->n_size); 751 } 752 return (error); 753 } 754 755 /* 756 * Do an nfs setattr rpc. 757 */ 758 static int 759 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 760 struct ucred *cred, struct thread *td) 761 { 762 struct nfsv2_sattr *sp; 763 struct nfsnode *np = VTONFS(vp); 764 caddr_t cp; 765 int32_t t1, t2; 766 caddr_t bpos, dpos, cp2; 767 u_int32_t *tl; 768 int error = 0, wccflag = NFSV3_WCCRATTR; 769 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 770 int v3 = NFS_ISV3(vp); 771 772 nfsstats.rpccnt[NFSPROC_SETATTR]++; 773 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 774 nfsm_fhtom(vp, v3); 775 if (v3) { 776 nfsm_v3attrbuild(vap, TRUE); 777 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 778 *tl = nfs_false; 779 } else { 780 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 781 if (vap->va_mode == (mode_t)VNOVAL) 782 sp->sa_mode = nfs_xdrneg1; 783 else 784 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 785 if (vap->va_uid == (uid_t)VNOVAL) 786 sp->sa_uid = nfs_xdrneg1; 787 else 788 sp->sa_uid = txdr_unsigned(vap->va_uid); 789 if (vap->va_gid == (gid_t)VNOVAL) 790 sp->sa_gid = nfs_xdrneg1; 791 else 792 sp->sa_gid = txdr_unsigned(vap->va_gid); 793 sp->sa_size = txdr_unsigned(vap->va_size); 794 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 795 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 796 } 797 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 798 if (v3) { 799 np->n_modestamp = 0; 800 nfsm_wcc_data(vp, wccflag); 801 } else 802 nfsm_loadattr(vp, (struct vattr *)0); 803 m_freem(mrep); 804 nfsmout: 805 return (error); 806 } 807 808 /* 809 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 810 * nfs_lookup() until all remaining new api calls are implemented. 811 * 812 * Resolve a namecache entry. This function is passed a locked ncp and 813 * must call cache_setvp() on it as appropriate to resolve the entry. 814 */ 815 static int 816 nfs_nresolve(struct vop_nresolve_args *ap) 817 { 818 struct thread *td = curthread; 819 struct namecache *ncp; 820 struct ucred *cred; 821 struct nfsnode *np; 822 struct vnode *dvp; 823 struct vnode *nvp; 824 nfsfh_t *fhp; 825 int attrflag; 826 int fhsize; 827 int error; 828 int len; 829 int v3; 830 /******NFSM MACROS********/ 831 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 832 caddr_t bpos, dpos, cp, cp2; 833 u_int32_t *tl; 834 int32_t t1, t2; 835 836 cred = ap->a_cred; 837 ncp = ap->a_nch->ncp; 838 839 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp); 840 dvp = ncp->nc_parent->nc_vp; 841 if ((error = vget(dvp, LK_SHARED)) != 0) 842 return (error); 843 844 nvp = NULL; 845 v3 = NFS_ISV3(dvp); 846 nfsstats.lookupcache_misses++; 847 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 848 len = ncp->nc_nlen; 849 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 850 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 851 nfsm_fhtom(dvp, v3); 852 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 853 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 854 if (error) { 855 /* 856 * Cache negatve lookups to reduce NFS traffic, but use 857 * a fast timeout. Otherwise use a timeout of 1 tick. 858 * XXX we should add a namecache flag for no-caching 859 * to uncache the negative hit as soon as possible, but 860 * we cannot simply destroy the entry because it is used 861 * as a placeholder by the caller. 862 */ 863 if (error == ENOENT) { 864 int nticks; 865 866 if (nfsneg_cache_timeout) 867 nticks = nfsneg_cache_timeout * hz; 868 else 869 nticks = 1; 870 cache_setvp(ap->a_nch, NULL); 871 cache_settimeout(ap->a_nch, nticks); 872 } 873 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 874 m_freem(mrep); 875 goto nfsmout; 876 } 877 878 /* 879 * Success, get the file handle, do various checks, and load 880 * post-operation data from the reply packet. Theoretically 881 * we should never be looking up "." so, theoretically, we 882 * should never get the same file handle as our directory. But 883 * we check anyway. XXX 884 * 885 * Note that no timeout is set for the positive cache hit. We 886 * assume, theoretically, that ESTALE returns will be dealt with 887 * properly to handle NFS races and in anycase we cannot depend 888 * on a timeout to deal with NFS open/create/excl issues so instead 889 * of a bad hack here the rest of the NFS client code needs to do 890 * the right thing. 891 */ 892 nfsm_getfh(fhp, fhsize, v3); 893 894 np = VTONFS(dvp); 895 if (NFS_CMPFH(np, fhp, fhsize)) { 896 vref(dvp); 897 nvp = dvp; 898 } else { 899 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 900 if (error) { 901 m_freem(mrep); 902 vput(dvp); 903 return (error); 904 } 905 nvp = NFSTOV(np); 906 } 907 if (v3) { 908 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 909 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 910 } else { 911 nfsm_loadattr(nvp, NULL); 912 } 913 cache_setvp(ap->a_nch, nvp); 914 m_freem(mrep); 915 nfsmout: 916 vput(dvp); 917 if (nvp) { 918 if (nvp == dvp) 919 vrele(nvp); 920 else 921 vput(nvp); 922 } 923 return (error); 924 } 925 926 /* 927 * 'cached' nfs directory lookup 928 * 929 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 930 * 931 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 932 * struct componentname *a_cnp) 933 */ 934 static int 935 nfs_lookup(struct vop_old_lookup_args *ap) 936 { 937 struct componentname *cnp = ap->a_cnp; 938 struct vnode *dvp = ap->a_dvp; 939 struct vnode **vpp = ap->a_vpp; 940 int flags = cnp->cn_flags; 941 struct vnode *newvp; 942 u_int32_t *tl; 943 caddr_t cp; 944 int32_t t1, t2; 945 struct nfsmount *nmp; 946 caddr_t bpos, dpos, cp2; 947 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 948 long len; 949 nfsfh_t *fhp; 950 struct nfsnode *np; 951 int lockparent, wantparent, error = 0, attrflag, fhsize; 952 int v3 = NFS_ISV3(dvp); 953 954 /* 955 * Read-only mount check and directory check. 956 */ 957 *vpp = NULLVP; 958 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 959 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 960 return (EROFS); 961 962 if (dvp->v_type != VDIR) 963 return (ENOTDIR); 964 965 /* 966 * Look it up in the cache. Note that ENOENT is only returned if we 967 * previously entered a negative hit (see later on). The additional 968 * nfsneg_cache_timeout check causes previously cached results to 969 * be instantly ignored if the negative caching is turned off. 970 */ 971 lockparent = flags & CNP_LOCKPARENT; 972 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 973 nmp = VFSTONFS(dvp->v_mount); 974 np = VTONFS(dvp); 975 976 /* 977 * Go to the wire. 978 */ 979 error = 0; 980 newvp = NULLVP; 981 nfsstats.lookupcache_misses++; 982 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 983 len = cnp->cn_namelen; 984 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 985 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 986 nfsm_fhtom(dvp, v3); 987 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 988 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 989 if (error) { 990 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 991 m_freem(mrep); 992 goto nfsmout; 993 } 994 nfsm_getfh(fhp, fhsize, v3); 995 996 /* 997 * Handle RENAME case... 998 */ 999 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1000 if (NFS_CMPFH(np, fhp, fhsize)) { 1001 m_freem(mrep); 1002 return (EISDIR); 1003 } 1004 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1005 if (error) { 1006 m_freem(mrep); 1007 return (error); 1008 } 1009 newvp = NFSTOV(np); 1010 if (v3) { 1011 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1012 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1013 } else 1014 nfsm_loadattr(newvp, (struct vattr *)0); 1015 *vpp = newvp; 1016 m_freem(mrep); 1017 if (!lockparent) { 1018 vn_unlock(dvp); 1019 cnp->cn_flags |= CNP_PDIRUNLOCK; 1020 } 1021 return (0); 1022 } 1023 1024 if (flags & CNP_ISDOTDOT) { 1025 vn_unlock(dvp); 1026 cnp->cn_flags |= CNP_PDIRUNLOCK; 1027 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1028 if (error) { 1029 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1030 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1031 return (error); /* NOTE: return error from nget */ 1032 } 1033 newvp = NFSTOV(np); 1034 if (lockparent) { 1035 error = vn_lock(dvp, LK_EXCLUSIVE); 1036 if (error) { 1037 vput(newvp); 1038 return (error); 1039 } 1040 cnp->cn_flags |= CNP_PDIRUNLOCK; 1041 } 1042 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1043 vref(dvp); 1044 newvp = dvp; 1045 } else { 1046 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1047 if (error) { 1048 m_freem(mrep); 1049 return (error); 1050 } 1051 if (!lockparent) { 1052 vn_unlock(dvp); 1053 cnp->cn_flags |= CNP_PDIRUNLOCK; 1054 } 1055 newvp = NFSTOV(np); 1056 } 1057 if (v3) { 1058 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1059 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1060 } else 1061 nfsm_loadattr(newvp, (struct vattr *)0); 1062 #if 0 1063 /* XXX MOVE TO nfs_nremove() */ 1064 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1065 cnp->cn_nameiop != NAMEI_DELETE) { 1066 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1067 } 1068 #endif 1069 *vpp = newvp; 1070 m_freem(mrep); 1071 nfsmout: 1072 if (error) { 1073 if (newvp != NULLVP) { 1074 vrele(newvp); 1075 *vpp = NULLVP; 1076 } 1077 if ((cnp->cn_nameiop == NAMEI_CREATE || 1078 cnp->cn_nameiop == NAMEI_RENAME) && 1079 error == ENOENT) { 1080 if (!lockparent) { 1081 vn_unlock(dvp); 1082 cnp->cn_flags |= CNP_PDIRUNLOCK; 1083 } 1084 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1085 error = EROFS; 1086 else 1087 error = EJUSTRETURN; 1088 } 1089 } 1090 return (error); 1091 } 1092 1093 /* 1094 * nfs read call. 1095 * Just call nfs_bioread() to do the work. 1096 * 1097 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1098 * struct ucred *a_cred) 1099 */ 1100 static int 1101 nfs_read(struct vop_read_args *ap) 1102 { 1103 struct vnode *vp = ap->a_vp; 1104 1105 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1106 switch (vp->v_type) { 1107 case VREG: 1108 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1109 case VDIR: 1110 return (EISDIR); 1111 default: 1112 return EOPNOTSUPP; 1113 } 1114 } 1115 1116 /* 1117 * nfs readlink call 1118 * 1119 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1120 */ 1121 static int 1122 nfs_readlink(struct vop_readlink_args *ap) 1123 { 1124 struct vnode *vp = ap->a_vp; 1125 1126 if (vp->v_type != VLNK) 1127 return (EINVAL); 1128 return (nfs_bioread(vp, ap->a_uio, 0)); 1129 } 1130 1131 /* 1132 * Do a readlink rpc. 1133 * Called by nfs_doio() from below the buffer cache. 1134 */ 1135 int 1136 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1137 { 1138 u_int32_t *tl; 1139 caddr_t cp; 1140 int32_t t1, t2; 1141 caddr_t bpos, dpos, cp2; 1142 int error = 0, len, attrflag; 1143 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1144 int v3 = NFS_ISV3(vp); 1145 1146 nfsstats.rpccnt[NFSPROC_READLINK]++; 1147 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1148 nfsm_fhtom(vp, v3); 1149 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1150 if (v3) 1151 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1152 if (!error) { 1153 nfsm_strsiz(len, NFS_MAXPATHLEN); 1154 if (len == NFS_MAXPATHLEN) { 1155 struct nfsnode *np = VTONFS(vp); 1156 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1157 len = np->n_size; 1158 } 1159 nfsm_mtouio(uiop, len); 1160 } 1161 m_freem(mrep); 1162 nfsmout: 1163 return (error); 1164 } 1165 1166 /* 1167 * nfs read rpc call 1168 * Ditto above 1169 */ 1170 int 1171 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1172 { 1173 u_int32_t *tl; 1174 caddr_t cp; 1175 int32_t t1, t2; 1176 caddr_t bpos, dpos, cp2; 1177 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1178 struct nfsmount *nmp; 1179 int error = 0, len, retlen, tsiz, eof, attrflag; 1180 int v3 = NFS_ISV3(vp); 1181 1182 #ifndef nolint 1183 eof = 0; 1184 #endif 1185 nmp = VFSTONFS(vp->v_mount); 1186 tsiz = uiop->uio_resid; 1187 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1188 return (EFBIG); 1189 while (tsiz > 0) { 1190 nfsstats.rpccnt[NFSPROC_READ]++; 1191 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1192 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1193 nfsm_fhtom(vp, v3); 1194 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1195 if (v3) { 1196 txdr_hyper(uiop->uio_offset, tl); 1197 *(tl + 2) = txdr_unsigned(len); 1198 } else { 1199 *tl++ = txdr_unsigned(uiop->uio_offset); 1200 *tl++ = txdr_unsigned(len); 1201 *tl = 0; 1202 } 1203 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1204 if (v3) { 1205 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1206 if (error) { 1207 m_freem(mrep); 1208 goto nfsmout; 1209 } 1210 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1211 eof = fxdr_unsigned(int, *(tl + 1)); 1212 } else 1213 nfsm_loadattr(vp, (struct vattr *)0); 1214 nfsm_strsiz(retlen, nmp->nm_rsize); 1215 nfsm_mtouio(uiop, retlen); 1216 m_freem(mrep); 1217 tsiz -= retlen; 1218 if (v3) { 1219 if (eof || retlen == 0) { 1220 tsiz = 0; 1221 } 1222 } else if (retlen < len) { 1223 tsiz = 0; 1224 } 1225 } 1226 nfsmout: 1227 return (error); 1228 } 1229 1230 /* 1231 * nfs write call 1232 */ 1233 int 1234 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1235 { 1236 u_int32_t *tl; 1237 caddr_t cp; 1238 int32_t t1, t2, backup; 1239 caddr_t bpos, dpos, cp2; 1240 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1241 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1242 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1243 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1244 1245 #ifndef DIAGNOSTIC 1246 if (uiop->uio_iovcnt != 1) 1247 panic("nfs: writerpc iovcnt > 1"); 1248 #endif 1249 *must_commit = 0; 1250 tsiz = uiop->uio_resid; 1251 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1252 return (EFBIG); 1253 while (tsiz > 0) { 1254 nfsstats.rpccnt[NFSPROC_WRITE]++; 1255 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1256 nfsm_reqhead(vp, NFSPROC_WRITE, 1257 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1258 nfsm_fhtom(vp, v3); 1259 if (v3) { 1260 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1261 txdr_hyper(uiop->uio_offset, tl); 1262 tl += 2; 1263 *tl++ = txdr_unsigned(len); 1264 *tl++ = txdr_unsigned(*iomode); 1265 *tl = txdr_unsigned(len); 1266 } else { 1267 u_int32_t x; 1268 1269 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1270 /* Set both "begin" and "current" to non-garbage. */ 1271 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1272 *tl++ = x; /* "begin offset" */ 1273 *tl++ = x; /* "current offset" */ 1274 x = txdr_unsigned(len); 1275 *tl++ = x; /* total to this offset */ 1276 *tl = x; /* size of this write */ 1277 } 1278 nfsm_uiotom(uiop, len); 1279 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1280 if (v3) { 1281 /* 1282 * The write RPC returns a before and after mtime. The 1283 * nfsm_wcc_data() macro checks the before n_mtime 1284 * against the before time and stores the after time 1285 * in the nfsnode's cached vattr and n_mtime field. 1286 * The NRMODIFIED bit will be set if the before 1287 * time did not match the original mtime. 1288 */ 1289 wccflag = NFSV3_WCCCHK; 1290 nfsm_wcc_data(vp, wccflag); 1291 if (!error) { 1292 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1293 + NFSX_V3WRITEVERF); 1294 rlen = fxdr_unsigned(int, *tl++); 1295 if (rlen == 0) { 1296 error = NFSERR_IO; 1297 m_freem(mrep); 1298 break; 1299 } else if (rlen < len) { 1300 backup = len - rlen; 1301 uiop->uio_iov->iov_base -= backup; 1302 uiop->uio_iov->iov_len += backup; 1303 uiop->uio_offset -= backup; 1304 uiop->uio_resid += backup; 1305 len = rlen; 1306 } 1307 commit = fxdr_unsigned(int, *tl++); 1308 1309 /* 1310 * Return the lowest committment level 1311 * obtained by any of the RPCs. 1312 */ 1313 if (committed == NFSV3WRITE_FILESYNC) 1314 committed = commit; 1315 else if (committed == NFSV3WRITE_DATASYNC && 1316 commit == NFSV3WRITE_UNSTABLE) 1317 committed = commit; 1318 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1319 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1320 NFSX_V3WRITEVERF); 1321 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1322 } else if (bcmp((caddr_t)tl, 1323 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1324 *must_commit = 1; 1325 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1326 NFSX_V3WRITEVERF); 1327 } 1328 } 1329 } else { 1330 nfsm_loadattr(vp, (struct vattr *)0); 1331 } 1332 m_freem(mrep); 1333 if (error) 1334 break; 1335 tsiz -= len; 1336 } 1337 nfsmout: 1338 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1339 committed = NFSV3WRITE_FILESYNC; 1340 *iomode = committed; 1341 if (error) 1342 uiop->uio_resid = tsiz; 1343 return (error); 1344 } 1345 1346 /* 1347 * nfs mknod rpc 1348 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1349 * mode set to specify the file type and the size field for rdev. 1350 */ 1351 static int 1352 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1353 struct vattr *vap) 1354 { 1355 struct nfsv2_sattr *sp; 1356 u_int32_t *tl; 1357 caddr_t cp; 1358 int32_t t1, t2; 1359 struct vnode *newvp = (struct vnode *)0; 1360 struct nfsnode *np = (struct nfsnode *)0; 1361 struct vattr vattr; 1362 char *cp2; 1363 caddr_t bpos, dpos; 1364 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1365 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1366 int rmajor, rminor; 1367 int v3 = NFS_ISV3(dvp); 1368 1369 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1370 rmajor = txdr_unsigned(vap->va_rmajor); 1371 rminor = txdr_unsigned(vap->va_rminor); 1372 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) { 1373 rmajor = nfs_xdrneg1; 1374 rminor = nfs_xdrneg1; 1375 } else { 1376 return (EOPNOTSUPP); 1377 } 1378 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1379 return (error); 1380 } 1381 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1382 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1383 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1384 nfsm_fhtom(dvp, v3); 1385 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1386 if (v3) { 1387 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1388 *tl++ = vtonfsv3_type(vap->va_type); 1389 nfsm_v3attrbuild(vap, FALSE); 1390 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1391 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1392 *tl++ = txdr_unsigned(vap->va_rmajor); 1393 *tl = txdr_unsigned(vap->va_rminor); 1394 } 1395 } else { 1396 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1397 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1398 sp->sa_uid = nfs_xdrneg1; 1399 sp->sa_gid = nfs_xdrneg1; 1400 sp->sa_size = makeudev(rmajor, rminor); 1401 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1402 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1403 } 1404 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1405 if (!error) { 1406 nfsm_mtofh(dvp, newvp, v3, gotvp); 1407 if (!gotvp) { 1408 if (newvp) { 1409 vput(newvp); 1410 newvp = (struct vnode *)0; 1411 } 1412 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1413 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1414 if (!error) 1415 newvp = NFSTOV(np); 1416 } 1417 } 1418 if (v3) 1419 nfsm_wcc_data(dvp, wccflag); 1420 m_freem(mrep); 1421 nfsmout: 1422 if (error) { 1423 if (newvp) 1424 vput(newvp); 1425 } else { 1426 *vpp = newvp; 1427 } 1428 VTONFS(dvp)->n_flag |= NLMODIFIED; 1429 if (!wccflag) 1430 VTONFS(dvp)->n_attrstamp = 0; 1431 return (error); 1432 } 1433 1434 /* 1435 * nfs mknod vop 1436 * just call nfs_mknodrpc() to do the work. 1437 * 1438 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1439 * struct componentname *a_cnp, struct vattr *a_vap) 1440 */ 1441 /* ARGSUSED */ 1442 static int 1443 nfs_mknod(struct vop_old_mknod_args *ap) 1444 { 1445 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1446 } 1447 1448 static u_long create_verf; 1449 /* 1450 * nfs file create call 1451 * 1452 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1453 * struct componentname *a_cnp, struct vattr *a_vap) 1454 */ 1455 static int 1456 nfs_create(struct vop_old_create_args *ap) 1457 { 1458 struct vnode *dvp = ap->a_dvp; 1459 struct vattr *vap = ap->a_vap; 1460 struct componentname *cnp = ap->a_cnp; 1461 struct nfsv2_sattr *sp; 1462 u_int32_t *tl; 1463 caddr_t cp; 1464 int32_t t1, t2; 1465 struct nfsnode *np = (struct nfsnode *)0; 1466 struct vnode *newvp = (struct vnode *)0; 1467 caddr_t bpos, dpos, cp2; 1468 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1469 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1470 struct vattr vattr; 1471 int v3 = NFS_ISV3(dvp); 1472 1473 /* 1474 * Oops, not for me.. 1475 */ 1476 if (vap->va_type == VSOCK) 1477 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1478 1479 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1480 return (error); 1481 } 1482 if (vap->va_vaflags & VA_EXCLUSIVE) 1483 fmode |= O_EXCL; 1484 again: 1485 nfsstats.rpccnt[NFSPROC_CREATE]++; 1486 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1487 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1488 nfsm_fhtom(dvp, v3); 1489 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1490 if (v3) { 1491 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1492 if (fmode & O_EXCL) { 1493 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1494 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1495 #ifdef INET 1496 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1497 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1498 else 1499 #endif 1500 *tl++ = create_verf; 1501 *tl = ++create_verf; 1502 } else { 1503 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1504 nfsm_v3attrbuild(vap, FALSE); 1505 } 1506 } else { 1507 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1508 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1509 sp->sa_uid = nfs_xdrneg1; 1510 sp->sa_gid = nfs_xdrneg1; 1511 sp->sa_size = 0; 1512 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1513 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1514 } 1515 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1516 if (!error) { 1517 nfsm_mtofh(dvp, newvp, v3, gotvp); 1518 if (!gotvp) { 1519 if (newvp) { 1520 vput(newvp); 1521 newvp = (struct vnode *)0; 1522 } 1523 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1524 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1525 if (!error) 1526 newvp = NFSTOV(np); 1527 } 1528 } 1529 if (v3) 1530 nfsm_wcc_data(dvp, wccflag); 1531 m_freem(mrep); 1532 nfsmout: 1533 if (error) { 1534 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1535 fmode &= ~O_EXCL; 1536 goto again; 1537 } 1538 if (newvp) 1539 vput(newvp); 1540 } else if (v3 && (fmode & O_EXCL)) { 1541 /* 1542 * We are normally called with only a partially initialized 1543 * VAP. Since the NFSv3 spec says that server may use the 1544 * file attributes to store the verifier, the spec requires 1545 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1546 * in atime, but we can't really assume that all servers will 1547 * so we ensure that our SETATTR sets both atime and mtime. 1548 */ 1549 if (vap->va_mtime.tv_sec == VNOVAL) 1550 vfs_timestamp(&vap->va_mtime); 1551 if (vap->va_atime.tv_sec == VNOVAL) 1552 vap->va_atime = vap->va_mtime; 1553 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1554 } 1555 if (!error) { 1556 /* 1557 * The new np may have enough info for access 1558 * checks, make sure rucred and wucred are 1559 * initialized for read and write rpc's. 1560 */ 1561 np = VTONFS(newvp); 1562 if (np->n_rucred == NULL) 1563 np->n_rucred = crhold(cnp->cn_cred); 1564 if (np->n_wucred == NULL) 1565 np->n_wucred = crhold(cnp->cn_cred); 1566 *ap->a_vpp = newvp; 1567 } 1568 VTONFS(dvp)->n_flag |= NLMODIFIED; 1569 if (!wccflag) 1570 VTONFS(dvp)->n_attrstamp = 0; 1571 return (error); 1572 } 1573 1574 /* 1575 * nfs file remove call 1576 * To try and make nfs semantics closer to ufs semantics, a file that has 1577 * other processes using the vnode is renamed instead of removed and then 1578 * removed later on the last close. 1579 * - If v_sysref.refcnt > 1 1580 * If a rename is not already in the works 1581 * call nfs_sillyrename() to set it up 1582 * else 1583 * do the remove rpc 1584 * 1585 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1586 * struct componentname *a_cnp) 1587 */ 1588 static int 1589 nfs_remove(struct vop_old_remove_args *ap) 1590 { 1591 struct vnode *vp = ap->a_vp; 1592 struct vnode *dvp = ap->a_dvp; 1593 struct componentname *cnp = ap->a_cnp; 1594 struct nfsnode *np = VTONFS(vp); 1595 int error = 0; 1596 struct vattr vattr; 1597 1598 #ifndef DIAGNOSTIC 1599 if (vp->v_sysref.refcnt < 1) 1600 panic("nfs_remove: bad v_sysref.refcnt"); 1601 #endif 1602 if (vp->v_type == VDIR) 1603 error = EPERM; 1604 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename && 1605 VOP_GETATTR(vp, &vattr) == 0 && 1606 vattr.va_nlink > 1)) { 1607 /* 1608 * throw away biocache buffers, mainly to avoid 1609 * unnecessary delayed writes later. 1610 */ 1611 error = nfs_vinvalbuf(vp, 0, 1); 1612 /* Do the rpc */ 1613 if (error != EINTR) 1614 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1615 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1616 /* 1617 * Kludge City: If the first reply to the remove rpc is lost.. 1618 * the reply to the retransmitted request will be ENOENT 1619 * since the file was in fact removed 1620 * Therefore, we cheat and return success. 1621 */ 1622 if (error == ENOENT) 1623 error = 0; 1624 } else if (!np->n_sillyrename) { 1625 error = nfs_sillyrename(dvp, vp, cnp); 1626 } 1627 np->n_attrstamp = 0; 1628 return (error); 1629 } 1630 1631 /* 1632 * nfs file remove rpc called from nfs_inactive 1633 */ 1634 int 1635 nfs_removeit(struct sillyrename *sp) 1636 { 1637 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1638 sp->s_cred, NULL)); 1639 } 1640 1641 /* 1642 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1643 */ 1644 static int 1645 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1646 struct ucred *cred, struct thread *td) 1647 { 1648 u_int32_t *tl; 1649 caddr_t cp; 1650 int32_t t1, t2; 1651 caddr_t bpos, dpos, cp2; 1652 int error = 0, wccflag = NFSV3_WCCRATTR; 1653 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1654 int v3 = NFS_ISV3(dvp); 1655 1656 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1657 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1658 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1659 nfsm_fhtom(dvp, v3); 1660 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1661 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1662 if (v3) 1663 nfsm_wcc_data(dvp, wccflag); 1664 m_freem(mrep); 1665 nfsmout: 1666 VTONFS(dvp)->n_flag |= NLMODIFIED; 1667 if (!wccflag) 1668 VTONFS(dvp)->n_attrstamp = 0; 1669 return (error); 1670 } 1671 1672 /* 1673 * nfs file rename call 1674 * 1675 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1676 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1677 * struct vnode *a_tvp, struct componentname *a_tcnp) 1678 */ 1679 static int 1680 nfs_rename(struct vop_old_rename_args *ap) 1681 { 1682 struct vnode *fvp = ap->a_fvp; 1683 struct vnode *tvp = ap->a_tvp; 1684 struct vnode *fdvp = ap->a_fdvp; 1685 struct vnode *tdvp = ap->a_tdvp; 1686 struct componentname *tcnp = ap->a_tcnp; 1687 struct componentname *fcnp = ap->a_fcnp; 1688 int error; 1689 1690 /* Check for cross-device rename */ 1691 if ((fvp->v_mount != tdvp->v_mount) || 1692 (tvp && (fvp->v_mount != tvp->v_mount))) { 1693 error = EXDEV; 1694 goto out; 1695 } 1696 1697 /* 1698 * We have to flush B_DELWRI data prior to renaming 1699 * the file. If we don't, the delayed-write buffers 1700 * can be flushed out later after the file has gone stale 1701 * under NFSV3. NFSV2 does not have this problem because 1702 * ( as far as I can tell ) it flushes dirty buffers more 1703 * often. 1704 */ 1705 1706 VOP_FSYNC(fvp, MNT_WAIT); 1707 if (tvp) 1708 VOP_FSYNC(tvp, MNT_WAIT); 1709 1710 /* 1711 * If the tvp exists and is in use, sillyrename it before doing the 1712 * rename of the new file over it. 1713 * 1714 * XXX Can't sillyrename a directory. 1715 * 1716 * We do not attempt to do any namecache purges in this old API 1717 * routine. The new API compat functions have access to the actual 1718 * namecache structures and will do it for us. 1719 */ 1720 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename && 1721 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1722 vput(tvp); 1723 tvp = NULL; 1724 } else if (tvp) { 1725 ; 1726 } 1727 1728 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1729 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1730 tcnp->cn_td); 1731 1732 out: 1733 if (tdvp == tvp) 1734 vrele(tdvp); 1735 else 1736 vput(tdvp); 1737 if (tvp) 1738 vput(tvp); 1739 vrele(fdvp); 1740 vrele(fvp); 1741 /* 1742 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1743 */ 1744 if (error == ENOENT) 1745 error = 0; 1746 return (error); 1747 } 1748 1749 /* 1750 * nfs file rename rpc called from nfs_remove() above 1751 */ 1752 static int 1753 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1754 struct sillyrename *sp) 1755 { 1756 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1757 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1758 } 1759 1760 /* 1761 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1762 */ 1763 static int 1764 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1765 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1766 struct ucred *cred, struct thread *td) 1767 { 1768 u_int32_t *tl; 1769 caddr_t cp; 1770 int32_t t1, t2; 1771 caddr_t bpos, dpos, cp2; 1772 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1773 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1774 int v3 = NFS_ISV3(fdvp); 1775 1776 nfsstats.rpccnt[NFSPROC_RENAME]++; 1777 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1778 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1779 nfsm_rndup(tnamelen)); 1780 nfsm_fhtom(fdvp, v3); 1781 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1782 nfsm_fhtom(tdvp, v3); 1783 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1784 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1785 if (v3) { 1786 nfsm_wcc_data(fdvp, fwccflag); 1787 nfsm_wcc_data(tdvp, twccflag); 1788 } 1789 m_freem(mrep); 1790 nfsmout: 1791 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1792 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1793 if (!fwccflag) 1794 VTONFS(fdvp)->n_attrstamp = 0; 1795 if (!twccflag) 1796 VTONFS(tdvp)->n_attrstamp = 0; 1797 return (error); 1798 } 1799 1800 /* 1801 * nfs hard link create call 1802 * 1803 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1804 * struct componentname *a_cnp) 1805 */ 1806 static int 1807 nfs_link(struct vop_old_link_args *ap) 1808 { 1809 struct vnode *vp = ap->a_vp; 1810 struct vnode *tdvp = ap->a_tdvp; 1811 struct componentname *cnp = ap->a_cnp; 1812 u_int32_t *tl; 1813 caddr_t cp; 1814 int32_t t1, t2; 1815 caddr_t bpos, dpos, cp2; 1816 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1817 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1818 int v3; 1819 1820 if (vp->v_mount != tdvp->v_mount) { 1821 return (EXDEV); 1822 } 1823 1824 /* 1825 * Push all writes to the server, so that the attribute cache 1826 * doesn't get "out of sync" with the server. 1827 * XXX There should be a better way! 1828 */ 1829 VOP_FSYNC(vp, MNT_WAIT); 1830 1831 v3 = NFS_ISV3(vp); 1832 nfsstats.rpccnt[NFSPROC_LINK]++; 1833 nfsm_reqhead(vp, NFSPROC_LINK, 1834 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1835 nfsm_fhtom(vp, v3); 1836 nfsm_fhtom(tdvp, v3); 1837 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1838 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1839 if (v3) { 1840 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1841 nfsm_wcc_data(tdvp, wccflag); 1842 } 1843 m_freem(mrep); 1844 nfsmout: 1845 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1846 if (!attrflag) 1847 VTONFS(vp)->n_attrstamp = 0; 1848 if (!wccflag) 1849 VTONFS(tdvp)->n_attrstamp = 0; 1850 /* 1851 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1852 */ 1853 if (error == EEXIST) 1854 error = 0; 1855 return (error); 1856 } 1857 1858 /* 1859 * nfs symbolic link create call 1860 * 1861 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1862 * struct componentname *a_cnp, struct vattr *a_vap, 1863 * char *a_target) 1864 */ 1865 static int 1866 nfs_symlink(struct vop_old_symlink_args *ap) 1867 { 1868 struct vnode *dvp = ap->a_dvp; 1869 struct vattr *vap = ap->a_vap; 1870 struct componentname *cnp = ap->a_cnp; 1871 struct nfsv2_sattr *sp; 1872 u_int32_t *tl; 1873 caddr_t cp; 1874 int32_t t1, t2; 1875 caddr_t bpos, dpos, cp2; 1876 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1877 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1878 struct vnode *newvp = (struct vnode *)0; 1879 int v3 = NFS_ISV3(dvp); 1880 1881 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1882 slen = strlen(ap->a_target); 1883 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1884 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1885 nfsm_fhtom(dvp, v3); 1886 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1887 if (v3) { 1888 nfsm_v3attrbuild(vap, FALSE); 1889 } 1890 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1891 if (!v3) { 1892 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1893 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1894 sp->sa_uid = nfs_xdrneg1; 1895 sp->sa_gid = nfs_xdrneg1; 1896 sp->sa_size = nfs_xdrneg1; 1897 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1898 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1899 } 1900 1901 /* 1902 * Issue the NFS request and get the rpc response. 1903 * 1904 * Only NFSv3 responses returning an error of 0 actually return 1905 * a file handle that can be converted into newvp without having 1906 * to do an extra lookup rpc. 1907 */ 1908 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1909 if (v3) { 1910 if (error == 0) 1911 nfsm_mtofh(dvp, newvp, v3, gotvp); 1912 nfsm_wcc_data(dvp, wccflag); 1913 } 1914 1915 /* 1916 * out code jumps -> here, mrep is also freed. 1917 */ 1918 1919 m_freem(mrep); 1920 nfsmout: 1921 1922 /* 1923 * If we get an EEXIST error, silently convert it to no-error 1924 * in case of an NFS retry. 1925 */ 1926 if (error == EEXIST) 1927 error = 0; 1928 1929 /* 1930 * If we do not have (or no longer have) an error, and we could 1931 * not extract the newvp from the response due to the request being 1932 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1933 * to obtain a newvp to return. 1934 */ 1935 if (error == 0 && newvp == NULL) { 1936 struct nfsnode *np = NULL; 1937 1938 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1939 cnp->cn_cred, cnp->cn_td, &np); 1940 if (!error) 1941 newvp = NFSTOV(np); 1942 } 1943 if (error) { 1944 if (newvp) 1945 vput(newvp); 1946 } else { 1947 *ap->a_vpp = newvp; 1948 } 1949 VTONFS(dvp)->n_flag |= NLMODIFIED; 1950 if (!wccflag) 1951 VTONFS(dvp)->n_attrstamp = 0; 1952 return (error); 1953 } 1954 1955 /* 1956 * nfs make dir call 1957 * 1958 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1959 * struct componentname *a_cnp, struct vattr *a_vap) 1960 */ 1961 static int 1962 nfs_mkdir(struct vop_old_mkdir_args *ap) 1963 { 1964 struct vnode *dvp = ap->a_dvp; 1965 struct vattr *vap = ap->a_vap; 1966 struct componentname *cnp = ap->a_cnp; 1967 struct nfsv2_sattr *sp; 1968 u_int32_t *tl; 1969 caddr_t cp; 1970 int32_t t1, t2; 1971 int len; 1972 struct nfsnode *np = (struct nfsnode *)0; 1973 struct vnode *newvp = (struct vnode *)0; 1974 caddr_t bpos, dpos, cp2; 1975 int error = 0, wccflag = NFSV3_WCCRATTR; 1976 int gotvp = 0; 1977 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1978 struct vattr vattr; 1979 int v3 = NFS_ISV3(dvp); 1980 1981 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1982 return (error); 1983 } 1984 len = cnp->cn_namelen; 1985 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1986 nfsm_reqhead(dvp, NFSPROC_MKDIR, 1987 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 1988 nfsm_fhtom(dvp, v3); 1989 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1990 if (v3) { 1991 nfsm_v3attrbuild(vap, FALSE); 1992 } else { 1993 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1994 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 1995 sp->sa_uid = nfs_xdrneg1; 1996 sp->sa_gid = nfs_xdrneg1; 1997 sp->sa_size = nfs_xdrneg1; 1998 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1999 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2000 } 2001 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2002 if (!error) 2003 nfsm_mtofh(dvp, newvp, v3, gotvp); 2004 if (v3) 2005 nfsm_wcc_data(dvp, wccflag); 2006 m_freem(mrep); 2007 nfsmout: 2008 VTONFS(dvp)->n_flag |= NLMODIFIED; 2009 if (!wccflag) 2010 VTONFS(dvp)->n_attrstamp = 0; 2011 /* 2012 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2013 * if we can succeed in looking up the directory. 2014 */ 2015 if (error == EEXIST || (!error && !gotvp)) { 2016 if (newvp) { 2017 vrele(newvp); 2018 newvp = (struct vnode *)0; 2019 } 2020 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2021 cnp->cn_td, &np); 2022 if (!error) { 2023 newvp = NFSTOV(np); 2024 if (newvp->v_type != VDIR) 2025 error = EEXIST; 2026 } 2027 } 2028 if (error) { 2029 if (newvp) 2030 vrele(newvp); 2031 } else 2032 *ap->a_vpp = newvp; 2033 return (error); 2034 } 2035 2036 /* 2037 * nfs remove directory call 2038 * 2039 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2040 * struct componentname *a_cnp) 2041 */ 2042 static int 2043 nfs_rmdir(struct vop_old_rmdir_args *ap) 2044 { 2045 struct vnode *vp = ap->a_vp; 2046 struct vnode *dvp = ap->a_dvp; 2047 struct componentname *cnp = ap->a_cnp; 2048 u_int32_t *tl; 2049 caddr_t cp; 2050 int32_t t1, t2; 2051 caddr_t bpos, dpos, cp2; 2052 int error = 0, wccflag = NFSV3_WCCRATTR; 2053 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2054 int v3 = NFS_ISV3(dvp); 2055 2056 if (dvp == vp) 2057 return (EINVAL); 2058 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2059 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2060 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2061 nfsm_fhtom(dvp, v3); 2062 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2063 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2064 if (v3) 2065 nfsm_wcc_data(dvp, wccflag); 2066 m_freem(mrep); 2067 nfsmout: 2068 VTONFS(dvp)->n_flag |= NLMODIFIED; 2069 if (!wccflag) 2070 VTONFS(dvp)->n_attrstamp = 0; 2071 /* 2072 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2073 */ 2074 if (error == ENOENT) 2075 error = 0; 2076 return (error); 2077 } 2078 2079 /* 2080 * nfs readdir call 2081 * 2082 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2083 */ 2084 static int 2085 nfs_readdir(struct vop_readdir_args *ap) 2086 { 2087 struct vnode *vp = ap->a_vp; 2088 struct nfsnode *np = VTONFS(vp); 2089 struct uio *uio = ap->a_uio; 2090 int tresid, error; 2091 struct vattr vattr; 2092 2093 if (vp->v_type != VDIR) 2094 return (EPERM); 2095 2096 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 2097 return (error); 2098 2099 /* 2100 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2101 * and then check that is still valid, or if this is an NQNFS mount 2102 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2103 * VOP_GETATTR() does not necessarily go to the wire. 2104 */ 2105 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2106 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2107 if (VOP_GETATTR(vp, &vattr) == 0 && 2108 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2109 ) { 2110 nfsstats.direofcache_hits++; 2111 goto done; 2112 } 2113 } 2114 2115 /* 2116 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2117 * own cache coherency checks so we do not have to. 2118 */ 2119 tresid = uio->uio_resid; 2120 error = nfs_bioread(vp, uio, 0); 2121 2122 if (!error && uio->uio_resid == tresid) 2123 nfsstats.direofcache_misses++; 2124 done: 2125 vn_unlock(vp); 2126 return (error); 2127 } 2128 2129 /* 2130 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2131 * 2132 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2133 * offset/block and converts the nfs formatted directory entries for userland 2134 * consumption as well as deals with offsets into the middle of blocks. 2135 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2136 * be block-bounded. It must convert to cookies for the actual RPC. 2137 */ 2138 int 2139 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2140 { 2141 int len, left; 2142 struct nfs_dirent *dp = NULL; 2143 u_int32_t *tl; 2144 caddr_t cp; 2145 int32_t t1, t2; 2146 nfsuint64 *cookiep; 2147 caddr_t bpos, dpos, cp2; 2148 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2149 nfsuint64 cookie; 2150 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2151 struct nfsnode *dnp = VTONFS(vp); 2152 u_quad_t fileno; 2153 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2154 int attrflag; 2155 int v3 = NFS_ISV3(vp); 2156 2157 #ifndef DIAGNOSTIC 2158 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2159 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2160 panic("nfs readdirrpc bad uio"); 2161 #endif 2162 2163 /* 2164 * If there is no cookie, assume directory was stale. 2165 */ 2166 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2167 if (cookiep) 2168 cookie = *cookiep; 2169 else 2170 return (NFSERR_BAD_COOKIE); 2171 /* 2172 * Loop around doing readdir rpc's of size nm_readdirsize 2173 * truncated to a multiple of DIRBLKSIZ. 2174 * The stopping criteria is EOF or buffer full. 2175 */ 2176 while (more_dirs && bigenough) { 2177 nfsstats.rpccnt[NFSPROC_READDIR]++; 2178 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2179 NFSX_READDIR(v3)); 2180 nfsm_fhtom(vp, v3); 2181 if (v3) { 2182 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2183 *tl++ = cookie.nfsuquad[0]; 2184 *tl++ = cookie.nfsuquad[1]; 2185 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2186 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2187 } else { 2188 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2189 *tl++ = cookie.nfsuquad[0]; 2190 } 2191 *tl = txdr_unsigned(nmp->nm_readdirsize); 2192 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2193 if (v3) { 2194 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2195 if (!error) { 2196 nfsm_dissect(tl, u_int32_t *, 2197 2 * NFSX_UNSIGNED); 2198 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2199 dnp->n_cookieverf.nfsuquad[1] = *tl; 2200 } else { 2201 m_freem(mrep); 2202 goto nfsmout; 2203 } 2204 } 2205 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2206 more_dirs = fxdr_unsigned(int, *tl); 2207 2208 /* loop thru the dir entries, converting them to std form */ 2209 while (more_dirs && bigenough) { 2210 if (v3) { 2211 nfsm_dissect(tl, u_int32_t *, 2212 3 * NFSX_UNSIGNED); 2213 fileno = fxdr_hyper(tl); 2214 len = fxdr_unsigned(int, *(tl + 2)); 2215 } else { 2216 nfsm_dissect(tl, u_int32_t *, 2217 2 * NFSX_UNSIGNED); 2218 fileno = fxdr_unsigned(u_quad_t, *tl++); 2219 len = fxdr_unsigned(int, *tl); 2220 } 2221 if (len <= 0 || len > NFS_MAXNAMLEN) { 2222 error = EBADRPC; 2223 m_freem(mrep); 2224 goto nfsmout; 2225 } 2226 2227 /* 2228 * len is the number of bytes in the path element 2229 * name, not including the \0 termination. 2230 * 2231 * tlen is the number of bytes w have to reserve for 2232 * the path element name. 2233 */ 2234 tlen = nfsm_rndup(len); 2235 if (tlen == len) 2236 tlen += 4; /* To ensure null termination */ 2237 2238 /* 2239 * If the entry would cross a DIRBLKSIZ boundary, 2240 * extend the previous nfs_dirent to cover the 2241 * remaining space. 2242 */ 2243 left = DIRBLKSIZ - blksiz; 2244 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2245 dp->nfs_reclen += left; 2246 uiop->uio_iov->iov_base += left; 2247 uiop->uio_iov->iov_len -= left; 2248 uiop->uio_offset += left; 2249 uiop->uio_resid -= left; 2250 blksiz = 0; 2251 } 2252 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2253 bigenough = 0; 2254 if (bigenough) { 2255 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2256 dp->nfs_ino = fileno; 2257 dp->nfs_namlen = len; 2258 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2259 dp->nfs_type = DT_UNKNOWN; 2260 blksiz += dp->nfs_reclen; 2261 if (blksiz == DIRBLKSIZ) 2262 blksiz = 0; 2263 uiop->uio_offset += sizeof(struct nfs_dirent); 2264 uiop->uio_resid -= sizeof(struct nfs_dirent); 2265 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2266 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2267 nfsm_mtouio(uiop, len); 2268 2269 /* 2270 * The uiop has advanced by nfs_dirent + len 2271 * but really needs to advance by 2272 * nfs_dirent + tlen 2273 */ 2274 cp = uiop->uio_iov->iov_base; 2275 tlen -= len; 2276 *cp = '\0'; /* null terminate */ 2277 uiop->uio_iov->iov_base += tlen; 2278 uiop->uio_iov->iov_len -= tlen; 2279 uiop->uio_offset += tlen; 2280 uiop->uio_resid -= tlen; 2281 } else { 2282 /* 2283 * NFS strings must be rounded up (nfsm_myouio 2284 * handled that in the bigenough case). 2285 */ 2286 nfsm_adv(nfsm_rndup(len)); 2287 } 2288 if (v3) { 2289 nfsm_dissect(tl, u_int32_t *, 2290 3 * NFSX_UNSIGNED); 2291 } else { 2292 nfsm_dissect(tl, u_int32_t *, 2293 2 * NFSX_UNSIGNED); 2294 } 2295 2296 /* 2297 * If we were able to accomodate the last entry, 2298 * get the cookie for the next one. Otherwise 2299 * hold-over the cookie for the one we were not 2300 * able to accomodate. 2301 */ 2302 if (bigenough) { 2303 cookie.nfsuquad[0] = *tl++; 2304 if (v3) 2305 cookie.nfsuquad[1] = *tl++; 2306 } else if (v3) { 2307 tl += 2; 2308 } else { 2309 tl++; 2310 } 2311 more_dirs = fxdr_unsigned(int, *tl); 2312 } 2313 /* 2314 * If at end of rpc data, get the eof boolean 2315 */ 2316 if (!more_dirs) { 2317 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2318 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2319 } 2320 m_freem(mrep); 2321 } 2322 /* 2323 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2324 * by increasing d_reclen for the last record. 2325 */ 2326 if (blksiz > 0) { 2327 left = DIRBLKSIZ - blksiz; 2328 dp->nfs_reclen += left; 2329 uiop->uio_iov->iov_base += left; 2330 uiop->uio_iov->iov_len -= left; 2331 uiop->uio_offset += left; 2332 uiop->uio_resid -= left; 2333 } 2334 2335 if (bigenough) { 2336 /* 2337 * We hit the end of the directory, update direofoffset. 2338 */ 2339 dnp->n_direofoffset = uiop->uio_offset; 2340 } else { 2341 /* 2342 * There is more to go, insert the link cookie so the 2343 * next block can be read. 2344 */ 2345 if (uiop->uio_resid > 0) 2346 kprintf("EEK! readdirrpc resid > 0\n"); 2347 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2348 *cookiep = cookie; 2349 } 2350 nfsmout: 2351 return (error); 2352 } 2353 2354 /* 2355 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2356 */ 2357 int 2358 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2359 { 2360 int len, left; 2361 struct nfs_dirent *dp; 2362 u_int32_t *tl; 2363 caddr_t cp; 2364 int32_t t1, t2; 2365 struct vnode *newvp; 2366 nfsuint64 *cookiep; 2367 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2368 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2369 nfsuint64 cookie; 2370 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2371 struct nfsnode *dnp = VTONFS(vp), *np; 2372 nfsfh_t *fhp; 2373 u_quad_t fileno; 2374 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2375 int attrflag, fhsize; 2376 struct nchandle nch; 2377 struct nchandle dnch; 2378 struct nlcomponent nlc; 2379 2380 #ifndef nolint 2381 dp = NULL; 2382 #endif 2383 #ifndef DIAGNOSTIC 2384 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2385 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2386 panic("nfs readdirplusrpc bad uio"); 2387 #endif 2388 /* 2389 * Obtain the namecache record for the directory so we have something 2390 * to use as a basis for creating the entries. This function will 2391 * return a held (but not locked) ncp. The ncp may be disconnected 2392 * from the tree and cannot be used for upward traversals, and the 2393 * ncp may be unnamed. Note that other unrelated operations may 2394 * cause the ncp to be named at any time. 2395 */ 2396 cache_fromdvp(vp, NULL, 0, &dnch); 2397 bzero(&nlc, sizeof(nlc)); 2398 newvp = NULLVP; 2399 2400 /* 2401 * If there is no cookie, assume directory was stale. 2402 */ 2403 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2404 if (cookiep) 2405 cookie = *cookiep; 2406 else 2407 return (NFSERR_BAD_COOKIE); 2408 /* 2409 * Loop around doing readdir rpc's of size nm_readdirsize 2410 * truncated to a multiple of DIRBLKSIZ. 2411 * The stopping criteria is EOF or buffer full. 2412 */ 2413 while (more_dirs && bigenough) { 2414 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2415 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2416 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2417 nfsm_fhtom(vp, 1); 2418 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2419 *tl++ = cookie.nfsuquad[0]; 2420 *tl++ = cookie.nfsuquad[1]; 2421 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2422 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2423 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2424 *tl = txdr_unsigned(nmp->nm_rsize); 2425 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2426 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2427 if (error) { 2428 m_freem(mrep); 2429 goto nfsmout; 2430 } 2431 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2432 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2433 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2434 more_dirs = fxdr_unsigned(int, *tl); 2435 2436 /* loop thru the dir entries, doctoring them to 4bsd form */ 2437 while (more_dirs && bigenough) { 2438 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2439 fileno = fxdr_hyper(tl); 2440 len = fxdr_unsigned(int, *(tl + 2)); 2441 if (len <= 0 || len > NFS_MAXNAMLEN) { 2442 error = EBADRPC; 2443 m_freem(mrep); 2444 goto nfsmout; 2445 } 2446 tlen = nfsm_rndup(len); 2447 if (tlen == len) 2448 tlen += 4; /* To ensure null termination*/ 2449 left = DIRBLKSIZ - blksiz; 2450 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2451 dp->nfs_reclen += left; 2452 uiop->uio_iov->iov_base += left; 2453 uiop->uio_iov->iov_len -= left; 2454 uiop->uio_offset += left; 2455 uiop->uio_resid -= left; 2456 blksiz = 0; 2457 } 2458 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2459 bigenough = 0; 2460 if (bigenough) { 2461 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2462 dp->nfs_ino = fileno; 2463 dp->nfs_namlen = len; 2464 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2465 dp->nfs_type = DT_UNKNOWN; 2466 blksiz += dp->nfs_reclen; 2467 if (blksiz == DIRBLKSIZ) 2468 blksiz = 0; 2469 uiop->uio_offset += sizeof(struct nfs_dirent); 2470 uiop->uio_resid -= sizeof(struct nfs_dirent); 2471 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2472 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2473 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2474 nlc.nlc_namelen = len; 2475 nfsm_mtouio(uiop, len); 2476 cp = uiop->uio_iov->iov_base; 2477 tlen -= len; 2478 *cp = '\0'; 2479 uiop->uio_iov->iov_base += tlen; 2480 uiop->uio_iov->iov_len -= tlen; 2481 uiop->uio_offset += tlen; 2482 uiop->uio_resid -= tlen; 2483 } else 2484 nfsm_adv(nfsm_rndup(len)); 2485 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2486 if (bigenough) { 2487 cookie.nfsuquad[0] = *tl++; 2488 cookie.nfsuquad[1] = *tl++; 2489 } else 2490 tl += 2; 2491 2492 /* 2493 * Since the attributes are before the file handle 2494 * (sigh), we must skip over the attributes and then 2495 * come back and get them. 2496 */ 2497 attrflag = fxdr_unsigned(int, *tl); 2498 if (attrflag) { 2499 dpossav1 = dpos; 2500 mdsav1 = md; 2501 nfsm_adv(NFSX_V3FATTR); 2502 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2503 doit = fxdr_unsigned(int, *tl); 2504 if (doit) { 2505 nfsm_getfh(fhp, fhsize, 1); 2506 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2507 vref(vp); 2508 newvp = vp; 2509 np = dnp; 2510 } else { 2511 error = nfs_nget(vp->v_mount, fhp, 2512 fhsize, &np); 2513 if (error) 2514 doit = 0; 2515 else 2516 newvp = NFSTOV(np); 2517 } 2518 } 2519 if (doit && bigenough) { 2520 dpossav2 = dpos; 2521 dpos = dpossav1; 2522 mdsav2 = md; 2523 md = mdsav1; 2524 nfsm_loadattr(newvp, (struct vattr *)0); 2525 dpos = dpossav2; 2526 md = mdsav2; 2527 dp->nfs_type = 2528 IFTODT(VTTOIF(np->n_vattr.va_type)); 2529 if (dnch.ncp) { 2530 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n", 2531 nlc.nlc_namelen, nlc.nlc_namelen, 2532 nlc.nlc_nameptr); 2533 nch = cache_nlookup(&dnch, &nlc); 2534 cache_setunresolved(&nch); 2535 cache_setvp(&nch, newvp); 2536 cache_put(&nch); 2537 } else { 2538 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER" 2539 " %*.*s\n", 2540 nlc.nlc_namelen, nlc.nlc_namelen, 2541 nlc.nlc_nameptr); 2542 } 2543 } 2544 } else { 2545 /* Just skip over the file handle */ 2546 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2547 i = fxdr_unsigned(int, *tl); 2548 nfsm_adv(nfsm_rndup(i)); 2549 } 2550 if (newvp != NULLVP) { 2551 if (newvp == vp) 2552 vrele(newvp); 2553 else 2554 vput(newvp); 2555 newvp = NULLVP; 2556 } 2557 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2558 more_dirs = fxdr_unsigned(int, *tl); 2559 } 2560 /* 2561 * If at end of rpc data, get the eof boolean 2562 */ 2563 if (!more_dirs) { 2564 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2565 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2566 } 2567 m_freem(mrep); 2568 } 2569 /* 2570 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2571 * by increasing d_reclen for the last record. 2572 */ 2573 if (blksiz > 0) { 2574 left = DIRBLKSIZ - blksiz; 2575 dp->nfs_reclen += left; 2576 uiop->uio_iov->iov_base += left; 2577 uiop->uio_iov->iov_len -= left; 2578 uiop->uio_offset += left; 2579 uiop->uio_resid -= left; 2580 } 2581 2582 /* 2583 * We are now either at the end of the directory or have filled the 2584 * block. 2585 */ 2586 if (bigenough) 2587 dnp->n_direofoffset = uiop->uio_offset; 2588 else { 2589 if (uiop->uio_resid > 0) 2590 kprintf("EEK! readdirplusrpc resid > 0\n"); 2591 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2592 *cookiep = cookie; 2593 } 2594 nfsmout: 2595 if (newvp != NULLVP) { 2596 if (newvp == vp) 2597 vrele(newvp); 2598 else 2599 vput(newvp); 2600 newvp = NULLVP; 2601 } 2602 if (dnch.ncp) 2603 cache_drop(&dnch); 2604 return (error); 2605 } 2606 2607 /* 2608 * Silly rename. To make the NFS filesystem that is stateless look a little 2609 * more like the "ufs" a remove of an active vnode is translated to a rename 2610 * to a funny looking filename that is removed by nfs_inactive on the 2611 * nfsnode. There is the potential for another process on a different client 2612 * to create the same funny name between the nfs_lookitup() fails and the 2613 * nfs_rename() completes, but... 2614 */ 2615 static int 2616 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2617 { 2618 struct sillyrename *sp; 2619 struct nfsnode *np; 2620 int error; 2621 2622 /* 2623 * We previously purged dvp instead of vp. I don't know why, it 2624 * completely destroys performance. We can't do it anyway with the 2625 * new VFS API since we would be breaking the namecache topology. 2626 */ 2627 cache_purge(vp); /* XXX */ 2628 np = VTONFS(vp); 2629 #ifndef DIAGNOSTIC 2630 if (vp->v_type == VDIR) 2631 panic("nfs: sillyrename dir"); 2632 #endif 2633 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2634 M_NFSREQ, M_WAITOK); 2635 sp->s_cred = crdup(cnp->cn_cred); 2636 sp->s_dvp = dvp; 2637 vref(dvp); 2638 2639 /* Fudge together a funny name */ 2640 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2641 2642 /* Try lookitups until we get one that isn't there */ 2643 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2644 cnp->cn_td, (struct nfsnode **)0) == 0) { 2645 sp->s_name[4]++; 2646 if (sp->s_name[4] > 'z') { 2647 error = EINVAL; 2648 goto bad; 2649 } 2650 } 2651 error = nfs_renameit(dvp, cnp, sp); 2652 if (error) 2653 goto bad; 2654 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2655 cnp->cn_td, &np); 2656 np->n_sillyrename = sp; 2657 return (0); 2658 bad: 2659 vrele(sp->s_dvp); 2660 crfree(sp->s_cred); 2661 kfree((caddr_t)sp, M_NFSREQ); 2662 return (error); 2663 } 2664 2665 /* 2666 * Look up a file name and optionally either update the file handle or 2667 * allocate an nfsnode, depending on the value of npp. 2668 * npp == NULL --> just do the lookup 2669 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2670 * handled too 2671 * *npp != NULL --> update the file handle in the vnode 2672 */ 2673 static int 2674 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2675 struct thread *td, struct nfsnode **npp) 2676 { 2677 u_int32_t *tl; 2678 caddr_t cp; 2679 int32_t t1, t2; 2680 struct vnode *newvp = (struct vnode *)0; 2681 struct nfsnode *np, *dnp = VTONFS(dvp); 2682 caddr_t bpos, dpos, cp2; 2683 int error = 0, fhlen, attrflag; 2684 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2685 nfsfh_t *nfhp; 2686 int v3 = NFS_ISV3(dvp); 2687 2688 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2689 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2690 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2691 nfsm_fhtom(dvp, v3); 2692 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2693 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2694 if (npp && !error) { 2695 nfsm_getfh(nfhp, fhlen, v3); 2696 if (*npp) { 2697 np = *npp; 2698 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2699 kfree((caddr_t)np->n_fhp, M_NFSBIGFH); 2700 np->n_fhp = &np->n_fh; 2701 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2702 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK); 2703 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2704 np->n_fhsize = fhlen; 2705 newvp = NFSTOV(np); 2706 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2707 vref(dvp); 2708 newvp = dvp; 2709 } else { 2710 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2711 if (error) { 2712 m_freem(mrep); 2713 return (error); 2714 } 2715 newvp = NFSTOV(np); 2716 } 2717 if (v3) { 2718 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2719 if (!attrflag && *npp == NULL) { 2720 m_freem(mrep); 2721 if (newvp == dvp) 2722 vrele(newvp); 2723 else 2724 vput(newvp); 2725 return (ENOENT); 2726 } 2727 } else 2728 nfsm_loadattr(newvp, (struct vattr *)0); 2729 } 2730 m_freem(mrep); 2731 nfsmout: 2732 if (npp && *npp == NULL) { 2733 if (error) { 2734 if (newvp) { 2735 if (newvp == dvp) 2736 vrele(newvp); 2737 else 2738 vput(newvp); 2739 } 2740 } else 2741 *npp = np; 2742 } 2743 return (error); 2744 } 2745 2746 /* 2747 * Nfs Version 3 commit rpc 2748 */ 2749 int 2750 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2751 { 2752 caddr_t cp; 2753 u_int32_t *tl; 2754 int32_t t1, t2; 2755 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2756 caddr_t bpos, dpos, cp2; 2757 int error = 0, wccflag = NFSV3_WCCRATTR; 2758 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2759 2760 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2761 return (0); 2762 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2763 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2764 nfsm_fhtom(vp, 1); 2765 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2766 txdr_hyper(offset, tl); 2767 tl += 2; 2768 *tl = txdr_unsigned(cnt); 2769 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2770 nfsm_wcc_data(vp, wccflag); 2771 if (!error) { 2772 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2773 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2774 NFSX_V3WRITEVERF)) { 2775 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2776 NFSX_V3WRITEVERF); 2777 error = NFSERR_STALEWRITEVERF; 2778 } 2779 } 2780 m_freem(mrep); 2781 nfsmout: 2782 return (error); 2783 } 2784 2785 /* 2786 * Kludge City.. 2787 * - make nfs_bmap() essentially a no-op that does no translation 2788 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2789 * (Maybe I could use the process's page mapping, but I was concerned that 2790 * Kernel Write might not be enabled and also figured copyout() would do 2791 * a lot more work than bcopy() and also it currently happens in the 2792 * context of the swapper process (2). 2793 * 2794 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp, 2795 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2796 */ 2797 static int 2798 nfs_bmap(struct vop_bmap_args *ap) 2799 { 2800 struct vnode *vp = ap->a_vp; 2801 2802 if (ap->a_vpp != NULL) 2803 *ap->a_vpp = vp; 2804 if (ap->a_doffsetp != NULL) 2805 *ap->a_doffsetp = ap->a_loffset; 2806 if (ap->a_runp != NULL) 2807 *ap->a_runp = 0; 2808 if (ap->a_runb != NULL) 2809 *ap->a_runb = 0; 2810 return (0); 2811 } 2812 2813 /* 2814 * Strategy routine. 2815 * 2816 * For async requests when nfsiod(s) are running, queue the request by 2817 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2818 * request. 2819 */ 2820 static int 2821 nfs_strategy(struct vop_strategy_args *ap) 2822 { 2823 struct bio *bio = ap->a_bio; 2824 struct bio *nbio; 2825 struct buf *bp = bio->bio_buf; 2826 struct thread *td; 2827 int error = 0; 2828 2829 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2830 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2831 KASSERT(BUF_REFCNT(bp) > 0, 2832 ("nfs_strategy: buffer %p not locked", bp)); 2833 2834 if (bp->b_flags & B_ASYNC) 2835 td = NULL; 2836 else 2837 td = curthread; /* XXX */ 2838 2839 /* 2840 * We probably don't need to push an nbio any more since no 2841 * block conversion is required due to the use of 64 bit byte 2842 * offsets, but do it anyway. 2843 */ 2844 nbio = push_bio(bio); 2845 nbio->bio_offset = bio->bio_offset; 2846 2847 /* 2848 * If the op is asynchronous and an i/o daemon is waiting 2849 * queue the request, wake it up and wait for completion 2850 * otherwise just do it ourselves. 2851 */ 2852 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td)) 2853 error = nfs_doio(ap->a_vp, nbio, td); 2854 return (error); 2855 } 2856 2857 /* 2858 * Mmap a file 2859 * 2860 * NB Currently unsupported. 2861 * 2862 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2863 * struct thread *a_td) 2864 */ 2865 /* ARGSUSED */ 2866 static int 2867 nfs_mmap(struct vop_mmap_args *ap) 2868 { 2869 return (EINVAL); 2870 } 2871 2872 /* 2873 * fsync vnode op. Just call nfs_flush() with commit == 1. 2874 * 2875 * nfs_fsync(struct vnode *a_vp, struct ucred * a_cred, int a_waitfor, 2876 * struct thread *a_td) 2877 */ 2878 /* ARGSUSED */ 2879 static int 2880 nfs_fsync(struct vop_fsync_args *ap) 2881 { 2882 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1)); 2883 } 2884 2885 /* 2886 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 2887 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 2888 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 2889 * set the buffer contains data that has already been written to the server 2890 * and which now needs a commit RPC. 2891 * 2892 * If commit is 0 we only take one pass and only flush buffers containing new 2893 * dirty data. 2894 * 2895 * If commit is 1 we take two passes, issuing a commit RPC in the second 2896 * pass. 2897 * 2898 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 2899 * to completely flush all pending data. 2900 * 2901 * Note that the RB_SCAN code properly handles the case where the 2902 * callback might block and directly or indirectly (another thread) cause 2903 * the RB tree to change. 2904 */ 2905 2906 #ifndef NFS_COMMITBVECSIZ 2907 #define NFS_COMMITBVECSIZ 16 2908 #endif 2909 2910 struct nfs_flush_info { 2911 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 2912 struct thread *td; 2913 struct vnode *vp; 2914 int waitfor; 2915 int slpflag; 2916 int slptimeo; 2917 int loops; 2918 struct buf *bvary[NFS_COMMITBVECSIZ]; 2919 int bvsize; 2920 off_t beg_off; 2921 off_t end_off; 2922 }; 2923 2924 static int nfs_flush_bp(struct buf *bp, void *data); 2925 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 2926 2927 int 2928 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2929 { 2930 struct nfsnode *np = VTONFS(vp); 2931 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2932 struct nfs_flush_info info; 2933 int error; 2934 2935 bzero(&info, sizeof(info)); 2936 info.td = td; 2937 info.vp = vp; 2938 info.waitfor = waitfor; 2939 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 2940 info.loops = 0; 2941 2942 do { 2943 /* 2944 * Flush mode 2945 */ 2946 info.mode = NFI_FLUSHNEW; 2947 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2948 nfs_flush_bp, &info); 2949 2950 /* 2951 * Take a second pass if committing and no error occured. 2952 * Clean up any left over collection (whether an error 2953 * occurs or not). 2954 */ 2955 if (commit && error == 0) { 2956 info.mode = NFI_COMMIT; 2957 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2958 nfs_flush_bp, &info); 2959 if (info.bvsize) 2960 error = nfs_flush_docommit(&info, error); 2961 } 2962 2963 /* 2964 * Wait for pending I/O to complete before checking whether 2965 * any further dirty buffers exist. 2966 */ 2967 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) { 2968 vp->v_track_write.bk_waitflag = 1; 2969 error = tsleep(&vp->v_track_write, 2970 info.slpflag, "nfsfsync", info.slptimeo); 2971 if (error) { 2972 /* 2973 * We have to be able to break out if this 2974 * is an 'intr' mount. 2975 */ 2976 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2977 error = -EINTR; 2978 break; 2979 } 2980 2981 /* 2982 * Since we do not process pending signals, 2983 * once we get a PCATCH our tsleep() will no 2984 * longer sleep, switch to a fixed timeout 2985 * instead. 2986 */ 2987 if (info.slpflag == PCATCH) { 2988 info.slpflag = 0; 2989 info.slptimeo = 2 * hz; 2990 } 2991 error = 0; 2992 } 2993 } 2994 ++info.loops; 2995 /* 2996 * Loop if we are flushing synchronous as well as committing, 2997 * and dirty buffers are still present. Otherwise we might livelock. 2998 */ 2999 } while (waitfor == MNT_WAIT && commit && 3000 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3001 3002 /* 3003 * The callbacks have to return a negative error to terminate the 3004 * RB scan. 3005 */ 3006 if (error < 0) 3007 error = -error; 3008 3009 /* 3010 * Deal with any error collection 3011 */ 3012 if (np->n_flag & NWRITEERR) { 3013 error = np->n_error; 3014 np->n_flag &= ~NWRITEERR; 3015 } 3016 return (error); 3017 } 3018 3019 3020 static 3021 int 3022 nfs_flush_bp(struct buf *bp, void *data) 3023 { 3024 struct nfs_flush_info *info = data; 3025 off_t toff; 3026 int error; 3027 3028 error = 0; 3029 switch(info->mode) { 3030 case NFI_FLUSHNEW: 3031 crit_enter(); 3032 if (info->loops && info->waitfor == MNT_WAIT) { 3033 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3034 if (error) { 3035 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3036 if (info->slpflag & PCATCH) 3037 lkflags |= LK_PCATCH; 3038 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3039 info->slptimeo); 3040 } 3041 } else { 3042 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3043 } 3044 if (error == 0) { 3045 KKASSERT(bp->b_vp == info->vp); 3046 3047 if ((bp->b_flags & B_DELWRI) == 0) 3048 panic("nfs_fsync: not dirty"); 3049 if (bp->b_flags & B_NEEDCOMMIT) { 3050 BUF_UNLOCK(bp); 3051 crit_exit(); 3052 break; 3053 } 3054 bremfree(bp); 3055 3056 bp->b_flags |= B_ASYNC; 3057 crit_exit(); 3058 bwrite(bp); 3059 } else { 3060 crit_exit(); 3061 error = 0; 3062 } 3063 break; 3064 case NFI_COMMIT: 3065 /* 3066 * Only process buffers in need of a commit which we can 3067 * immediately lock. This may prevent a buffer from being 3068 * committed, but the normal flush loop will block on the 3069 * same buffer so we shouldn't get into an endless loop. 3070 */ 3071 crit_enter(); 3072 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3073 (B_DELWRI | B_NEEDCOMMIT) || 3074 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 3075 crit_exit(); 3076 break; 3077 } 3078 3079 KKASSERT(bp->b_vp == info->vp); 3080 bremfree(bp); 3081 3082 /* 3083 * NOTE: storing the bp in the bvary[] basically sets 3084 * it up for a commit operation. 3085 * 3086 * We must call vfs_busy_pages() now so the commit operation 3087 * is interlocked with user modifications to memory mapped 3088 * pages. 3089 * 3090 * Note: to avoid loopback deadlocks, we do not 3091 * assign b_runningbufspace. 3092 */ 3093 bp->b_cmd = BUF_CMD_WRITE; 3094 vfs_busy_pages(bp->b_vp, bp); 3095 info->bvary[info->bvsize] = bp; 3096 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3097 if (info->bvsize == 0 || toff < info->beg_off) 3098 info->beg_off = toff; 3099 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3100 if (info->bvsize == 0 || toff > info->end_off) 3101 info->end_off = toff; 3102 ++info->bvsize; 3103 if (info->bvsize == NFS_COMMITBVECSIZ) { 3104 error = nfs_flush_docommit(info, 0); 3105 KKASSERT(info->bvsize == 0); 3106 } 3107 crit_exit(); 3108 } 3109 return (error); 3110 } 3111 3112 static 3113 int 3114 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3115 { 3116 struct vnode *vp; 3117 struct buf *bp; 3118 off_t bytes; 3119 int retv; 3120 int i; 3121 3122 vp = info->vp; 3123 3124 if (info->bvsize > 0) { 3125 /* 3126 * Commit data on the server, as required. Note that 3127 * nfs_commit will use the vnode's cred for the commit. 3128 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3129 */ 3130 bytes = info->end_off - info->beg_off; 3131 if (bytes > 0x40000000) 3132 bytes = 0x40000000; 3133 if (error) { 3134 retv = -error; 3135 } else { 3136 retv = nfs_commit(vp, info->beg_off, 3137 (int)bytes, info->td); 3138 if (retv == NFSERR_STALEWRITEVERF) 3139 nfs_clearcommit(vp->v_mount); 3140 } 3141 3142 /* 3143 * Now, either mark the blocks I/O done or mark the 3144 * blocks dirty, depending on whether the commit 3145 * succeeded. 3146 */ 3147 for (i = 0; i < info->bvsize; ++i) { 3148 bp = info->bvary[i]; 3149 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3150 if (retv) { 3151 /* 3152 * Error, leave B_DELWRI intact 3153 */ 3154 vfs_unbusy_pages(bp); 3155 bp->b_cmd = BUF_CMD_DONE; 3156 brelse(bp); 3157 } else { 3158 /* 3159 * Success, remove B_DELWRI ( bundirty() ). 3160 * 3161 * b_dirtyoff/b_dirtyend seem to be NFS 3162 * specific. We should probably move that 3163 * into bundirty(). XXX 3164 * 3165 * We are faking an I/O write, we have to 3166 * start the transaction in order to 3167 * immediately biodone() it. 3168 */ 3169 crit_enter(); 3170 bp->b_flags |= B_ASYNC; 3171 bundirty(bp); 3172 bp->b_flags &= ~B_ERROR; 3173 bp->b_dirtyoff = bp->b_dirtyend = 0; 3174 crit_exit(); 3175 biodone(&bp->b_bio1); 3176 } 3177 } 3178 info->bvsize = 0; 3179 } 3180 return (error); 3181 } 3182 3183 /* 3184 * NFS advisory byte-level locks. 3185 * Currently unsupported. 3186 * 3187 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3188 * int a_flags) 3189 */ 3190 static int 3191 nfs_advlock(struct vop_advlock_args *ap) 3192 { 3193 struct nfsnode *np = VTONFS(ap->a_vp); 3194 3195 /* 3196 * The following kludge is to allow diskless support to work 3197 * until a real NFS lockd is implemented. Basically, just pretend 3198 * that this is a local lock. 3199 */ 3200 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3201 } 3202 3203 /* 3204 * Print out the contents of an nfsnode. 3205 * 3206 * nfs_print(struct vnode *a_vp) 3207 */ 3208 static int 3209 nfs_print(struct vop_print_args *ap) 3210 { 3211 struct vnode *vp = ap->a_vp; 3212 struct nfsnode *np = VTONFS(vp); 3213 3214 kprintf("tag VT_NFS, fileid %ld fsid 0x%x", 3215 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3216 if (vp->v_type == VFIFO) 3217 fifo_printinfo(vp); 3218 kprintf("\n"); 3219 return (0); 3220 } 3221 3222 /* 3223 * nfs special file access vnode op. 3224 * Essentially just get vattr and then imitate iaccess() since the device is 3225 * local to the client. 3226 * 3227 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3228 * struct thread *a_td) 3229 */ 3230 static int 3231 nfsspec_access(struct vop_access_args *ap) 3232 { 3233 struct vattr *vap; 3234 gid_t *gp; 3235 struct ucred *cred = ap->a_cred; 3236 struct vnode *vp = ap->a_vp; 3237 mode_t mode = ap->a_mode; 3238 struct vattr vattr; 3239 int i; 3240 int error; 3241 3242 /* 3243 * Disallow write attempts on filesystems mounted read-only; 3244 * unless the file is a socket, fifo, or a block or character 3245 * device resident on the filesystem. 3246 */ 3247 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3248 switch (vp->v_type) { 3249 case VREG: 3250 case VDIR: 3251 case VLNK: 3252 return (EROFS); 3253 default: 3254 break; 3255 } 3256 } 3257 /* 3258 * If you're the super-user, 3259 * you always get access. 3260 */ 3261 if (cred->cr_uid == 0) 3262 return (0); 3263 vap = &vattr; 3264 error = VOP_GETATTR(vp, vap); 3265 if (error) 3266 return (error); 3267 /* 3268 * Access check is based on only one of owner, group, public. 3269 * If not owner, then check group. If not a member of the 3270 * group, then check public access. 3271 */ 3272 if (cred->cr_uid != vap->va_uid) { 3273 mode >>= 3; 3274 gp = cred->cr_groups; 3275 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3276 if (vap->va_gid == *gp) 3277 goto found; 3278 mode >>= 3; 3279 found: 3280 ; 3281 } 3282 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3283 return (error); 3284 } 3285 3286 /* 3287 * Read wrapper for special devices. 3288 * 3289 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3290 * struct ucred *a_cred) 3291 */ 3292 static int 3293 nfsspec_read(struct vop_read_args *ap) 3294 { 3295 struct nfsnode *np = VTONFS(ap->a_vp); 3296 3297 /* 3298 * Set access flag. 3299 */ 3300 np->n_flag |= NACC; 3301 getnanotime(&np->n_atim); 3302 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3303 } 3304 3305 /* 3306 * Write wrapper for special devices. 3307 * 3308 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3309 * struct ucred *a_cred) 3310 */ 3311 static int 3312 nfsspec_write(struct vop_write_args *ap) 3313 { 3314 struct nfsnode *np = VTONFS(ap->a_vp); 3315 3316 /* 3317 * Set update flag. 3318 */ 3319 np->n_flag |= NUPD; 3320 getnanotime(&np->n_mtim); 3321 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3322 } 3323 3324 /* 3325 * Close wrapper for special devices. 3326 * 3327 * Update the times on the nfsnode then do device close. 3328 * 3329 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3330 * struct thread *a_td) 3331 */ 3332 static int 3333 nfsspec_close(struct vop_close_args *ap) 3334 { 3335 struct vnode *vp = ap->a_vp; 3336 struct nfsnode *np = VTONFS(vp); 3337 struct vattr vattr; 3338 3339 if (np->n_flag & (NACC | NUPD)) { 3340 np->n_flag |= NCHG; 3341 if (vp->v_sysref.refcnt == 1 && 3342 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3343 VATTR_NULL(&vattr); 3344 if (np->n_flag & NACC) 3345 vattr.va_atime = np->n_atim; 3346 if (np->n_flag & NUPD) 3347 vattr.va_mtime = np->n_mtim; 3348 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3349 } 3350 } 3351 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3352 } 3353 3354 /* 3355 * Read wrapper for fifos. 3356 * 3357 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3358 * struct ucred *a_cred) 3359 */ 3360 static int 3361 nfsfifo_read(struct vop_read_args *ap) 3362 { 3363 struct nfsnode *np = VTONFS(ap->a_vp); 3364 3365 /* 3366 * Set access flag. 3367 */ 3368 np->n_flag |= NACC; 3369 getnanotime(&np->n_atim); 3370 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3371 } 3372 3373 /* 3374 * Write wrapper for fifos. 3375 * 3376 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3377 * struct ucred *a_cred) 3378 */ 3379 static int 3380 nfsfifo_write(struct vop_write_args *ap) 3381 { 3382 struct nfsnode *np = VTONFS(ap->a_vp); 3383 3384 /* 3385 * Set update flag. 3386 */ 3387 np->n_flag |= NUPD; 3388 getnanotime(&np->n_mtim); 3389 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3390 } 3391 3392 /* 3393 * Close wrapper for fifos. 3394 * 3395 * Update the times on the nfsnode then do fifo close. 3396 * 3397 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3398 */ 3399 static int 3400 nfsfifo_close(struct vop_close_args *ap) 3401 { 3402 struct vnode *vp = ap->a_vp; 3403 struct nfsnode *np = VTONFS(vp); 3404 struct vattr vattr; 3405 struct timespec ts; 3406 3407 if (np->n_flag & (NACC | NUPD)) { 3408 getnanotime(&ts); 3409 if (np->n_flag & NACC) 3410 np->n_atim = ts; 3411 if (np->n_flag & NUPD) 3412 np->n_mtim = ts; 3413 np->n_flag |= NCHG; 3414 if (vp->v_sysref.refcnt == 1 && 3415 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3416 VATTR_NULL(&vattr); 3417 if (np->n_flag & NACC) 3418 vattr.va_atime = np->n_atim; 3419 if (np->n_flag & NUPD) 3420 vattr.va_mtime = np->n_mtim; 3421 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3422 } 3423 } 3424 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3425 } 3426 3427