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 1.77 2008/06/08 08:38:06 sephe 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 NAMECACHE timeout"); 236 237 static int nfspos_cache_timeout = NFS_MINATTRTIMO; 238 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW, 239 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout"); 240 241 static int nfsv3_commit_on_close = 0; 242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 243 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 244 #if 0 245 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 246 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 247 248 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 249 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 250 #endif 251 252 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 253 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 254 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 255 static int 256 nfs3_access_otw(struct vnode *vp, int wmode, 257 struct thread *td, struct ucred *cred) 258 { 259 const int v3 = 1; 260 u_int32_t *tl; 261 int error = 0, attrflag; 262 263 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 264 caddr_t bpos, dpos, cp2; 265 int32_t t1, t2; 266 caddr_t cp; 267 u_int32_t rmode; 268 struct nfsnode *np = VTONFS(vp); 269 270 nfsstats.rpccnt[NFSPROC_ACCESS]++; 271 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 272 nfsm_fhtom(vp, v3); 273 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 274 *tl = txdr_unsigned(wmode); 275 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 276 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 277 if (!error) { 278 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 279 rmode = fxdr_unsigned(u_int32_t, *tl); 280 np->n_mode = rmode; 281 np->n_modeuid = cred->cr_uid; 282 np->n_modestamp = mycpu->gd_time_seconds; 283 } 284 m_freem(mrep); 285 nfsmout: 286 return error; 287 } 288 289 /* 290 * nfs access vnode op. 291 * For nfs version 2, just return ok. File accesses may fail later. 292 * For nfs version 3, use the access rpc to check accessibility. If file modes 293 * are changed on the server, accesses might still fail later. 294 * 295 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 296 */ 297 static int 298 nfs_access(struct vop_access_args *ap) 299 { 300 struct vnode *vp = ap->a_vp; 301 thread_t td = curthread; 302 int error = 0; 303 u_int32_t mode, wmode; 304 int v3 = NFS_ISV3(vp); 305 struct nfsnode *np = VTONFS(vp); 306 307 /* 308 * Disallow write attempts on filesystems mounted read-only; 309 * unless the file is a socket, fifo, or a block or character 310 * device resident on the filesystem. 311 */ 312 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 313 switch (vp->v_type) { 314 case VREG: 315 case VDIR: 316 case VLNK: 317 return (EROFS); 318 default: 319 break; 320 } 321 } 322 /* 323 * For nfs v3, check to see if we have done this recently, and if 324 * so return our cached result instead of making an ACCESS call. 325 * If not, do an access rpc, otherwise you are stuck emulating 326 * ufs_access() locally using the vattr. This may not be correct, 327 * since the server may apply other access criteria such as 328 * client uid-->server uid mapping that we do not know about. 329 */ 330 if (v3) { 331 if (ap->a_mode & VREAD) 332 mode = NFSV3ACCESS_READ; 333 else 334 mode = 0; 335 if (vp->v_type != VDIR) { 336 if (ap->a_mode & VWRITE) 337 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 338 if (ap->a_mode & VEXEC) 339 mode |= NFSV3ACCESS_EXECUTE; 340 } else { 341 if (ap->a_mode & VWRITE) 342 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 343 NFSV3ACCESS_DELETE); 344 if (ap->a_mode & VEXEC) 345 mode |= NFSV3ACCESS_LOOKUP; 346 } 347 /* XXX safety belt, only make blanket request if caching */ 348 if (nfsaccess_cache_timeout > 0) { 349 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 350 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 351 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 352 } else { 353 wmode = mode; 354 } 355 356 /* 357 * Does our cached result allow us to give a definite yes to 358 * this request? 359 */ 360 if (np->n_modestamp && 361 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 362 (ap->a_cred->cr_uid == np->n_modeuid) && 363 ((np->n_mode & mode) == mode)) { 364 nfsstats.accesscache_hits++; 365 } else { 366 /* 367 * Either a no, or a don't know. Go to the wire. 368 */ 369 nfsstats.accesscache_misses++; 370 error = nfs3_access_otw(vp, wmode, td, ap->a_cred); 371 if (!error) { 372 if ((np->n_mode & mode) != mode) { 373 error = EACCES; 374 } 375 } 376 } 377 } else { 378 if ((error = nfsspec_access(ap)) != 0) 379 return (error); 380 381 /* 382 * Attempt to prevent a mapped root from accessing a file 383 * which it shouldn't. We try to read a byte from the file 384 * if the user is root and the file is not zero length. 385 * After calling nfsspec_access, we should have the correct 386 * file size cached. 387 */ 388 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 389 && VTONFS(vp)->n_size > 0) { 390 struct iovec aiov; 391 struct uio auio; 392 char buf[1]; 393 394 aiov.iov_base = buf; 395 aiov.iov_len = 1; 396 auio.uio_iov = &aiov; 397 auio.uio_iovcnt = 1; 398 auio.uio_offset = 0; 399 auio.uio_resid = 1; 400 auio.uio_segflg = UIO_SYSSPACE; 401 auio.uio_rw = UIO_READ; 402 auio.uio_td = td; 403 404 if (vp->v_type == VREG) { 405 error = nfs_readrpc(vp, &auio); 406 } else if (vp->v_type == VDIR) { 407 char* bp; 408 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 409 aiov.iov_base = bp; 410 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 411 error = nfs_readdirrpc(vp, &auio); 412 kfree(bp, M_TEMP); 413 } else if (vp->v_type == VLNK) { 414 error = nfs_readlinkrpc(vp, &auio); 415 } else { 416 error = EACCES; 417 } 418 } 419 } 420 /* 421 * [re]record creds for reading and/or writing if access 422 * was granted. Assume the NFS server will grant read access 423 * for execute requests. 424 */ 425 if (error == 0) { 426 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 427 crhold(ap->a_cred); 428 if (np->n_rucred) 429 crfree(np->n_rucred); 430 np->n_rucred = ap->a_cred; 431 } 432 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 433 crhold(ap->a_cred); 434 if (np->n_wucred) 435 crfree(np->n_wucred); 436 np->n_wucred = ap->a_cred; 437 } 438 } 439 return(error); 440 } 441 442 /* 443 * nfs open vnode op 444 * Check to see if the type is ok 445 * and that deletion is not in progress. 446 * For paged in text files, you will need to flush the page cache 447 * if consistency is lost. 448 * 449 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 450 * struct file *a_fp) 451 */ 452 /* ARGSUSED */ 453 static int 454 nfs_open(struct vop_open_args *ap) 455 { 456 struct vnode *vp = ap->a_vp; 457 struct nfsnode *np = VTONFS(vp); 458 struct vattr vattr; 459 int error; 460 461 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 462 #ifdef DIAGNOSTIC 463 kprintf("open eacces vtyp=%d\n",vp->v_type); 464 #endif 465 return (EOPNOTSUPP); 466 } 467 468 /* 469 * Clear the attribute cache only if opening with write access. It 470 * is unclear if we should do this at all here, but we certainly 471 * should not clear the cache unconditionally simply because a file 472 * is being opened. 473 */ 474 if (ap->a_mode & FWRITE) 475 np->n_attrstamp = 0; 476 477 /* 478 * For normal NFS, reconcile changes made locally verses 479 * changes made remotely. Note that VOP_GETATTR only goes 480 * to the wire if the cached attribute has timed out or been 481 * cleared. 482 * 483 * If local modifications have been made clear the attribute 484 * cache to force an attribute and modified time check. If 485 * GETATTR detects that the file has been changed by someone 486 * other then us it will set NRMODIFIED. 487 * 488 * If we are opening a directory and local changes have been 489 * made we have to invalidate the cache in order to ensure 490 * that we get the most up-to-date information from the 491 * server. XXX 492 */ 493 if (np->n_flag & NLMODIFIED) { 494 np->n_attrstamp = 0; 495 if (vp->v_type == VDIR) { 496 error = nfs_vinvalbuf(vp, V_SAVE, 1); 497 if (error == EINTR) 498 return (error); 499 nfs_invaldir(vp); 500 } 501 } 502 error = VOP_GETATTR(vp, &vattr); 503 if (error) 504 return (error); 505 if (np->n_flag & NRMODIFIED) { 506 if (vp->v_type == VDIR) 507 nfs_invaldir(vp); 508 error = nfs_vinvalbuf(vp, V_SAVE, 1); 509 if (error == EINTR) 510 return (error); 511 np->n_flag &= ~NRMODIFIED; 512 } 513 514 return (vop_stdopen(ap)); 515 } 516 517 /* 518 * nfs close vnode op 519 * What an NFS client should do upon close after writing is a debatable issue. 520 * Most NFS clients push delayed writes to the server upon close, basically for 521 * two reasons: 522 * 1 - So that any write errors may be reported back to the client process 523 * doing the close system call. By far the two most likely errors are 524 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 525 * 2 - To put a worst case upper bound on cache inconsistency between 526 * multiple clients for the file. 527 * There is also a consistency problem for Version 2 of the protocol w.r.t. 528 * not being able to tell if other clients are writing a file concurrently, 529 * since there is no way of knowing if the changed modify time in the reply 530 * is only due to the write for this client. 531 * (NFS Version 3 provides weak cache consistency data in the reply that 532 * should be sufficient to detect and handle this case.) 533 * 534 * The current code does the following: 535 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 536 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 537 * or commit them (this satisfies 1 and 2 except for the 538 * case where the server crashes after this close but 539 * before the commit RPC, which is felt to be "good 540 * enough". Changing the last argument to nfs_flush() to 541 * a 1 would force a commit operation, if it is felt a 542 * commit is necessary now. 543 * for NQNFS - do nothing now, since 2 is dealt with via leases and 544 * 1 should be dealt with via an fsync() system call for 545 * cases where write errors are important. 546 * 547 * nfs_close(struct vnode *a_vp, int a_fflag) 548 */ 549 /* ARGSUSED */ 550 static int 551 nfs_close(struct vop_close_args *ap) 552 { 553 struct vnode *vp = ap->a_vp; 554 struct nfsnode *np = VTONFS(vp); 555 int error = 0; 556 thread_t td = curthread; 557 558 if (vp->v_type == VREG) { 559 if (np->n_flag & NLMODIFIED) { 560 if (NFS_ISV3(vp)) { 561 /* 562 * Under NFSv3 we have dirty buffers to dispose of. We 563 * must flush them to the NFS server. We have the option 564 * of waiting all the way through the commit rpc or just 565 * waiting for the initial write. The default is to only 566 * wait through the initial write so the data is in the 567 * server's cache, which is roughly similar to the state 568 * a standard disk subsystem leaves the file in on close(). 569 * 570 * We cannot clear the NLMODIFIED bit in np->n_flag due to 571 * potential races with other processes, and certainly 572 * cannot clear it if we don't commit. 573 */ 574 int cm = nfsv3_commit_on_close ? 1 : 0; 575 error = nfs_flush(vp, MNT_WAIT, td, cm); 576 /* np->n_flag &= ~NLMODIFIED; */ 577 } else { 578 error = nfs_vinvalbuf(vp, V_SAVE, 1); 579 } 580 np->n_attrstamp = 0; 581 } 582 if (np->n_flag & NWRITEERR) { 583 np->n_flag &= ~NWRITEERR; 584 error = np->n_error; 585 } 586 } 587 vop_stdclose(ap); 588 return (error); 589 } 590 591 /* 592 * nfs getattr call from vfs. 593 * 594 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap) 595 */ 596 static int 597 nfs_getattr(struct vop_getattr_args *ap) 598 { 599 struct vnode *vp = ap->a_vp; 600 struct nfsnode *np = VTONFS(vp); 601 caddr_t cp; 602 u_int32_t *tl; 603 int32_t t1, t2; 604 caddr_t bpos, dpos; 605 int error = 0; 606 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 607 int v3 = NFS_ISV3(vp); 608 thread_t td = curthread; 609 610 /* 611 * Update local times for special files. 612 */ 613 if (np->n_flag & (NACC | NUPD)) 614 np->n_flag |= NCHG; 615 /* 616 * First look in the cache. 617 */ 618 if (nfs_getattrcache(vp, ap->a_vap) == 0) 619 return (0); 620 621 if (v3 && nfsaccess_cache_timeout > 0) { 622 nfsstats.accesscache_misses++; 623 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 624 if (nfs_getattrcache(vp, ap->a_vap) == 0) 625 return (0); 626 } 627 628 nfsstats.rpccnt[NFSPROC_GETATTR]++; 629 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 630 nfsm_fhtom(vp, v3); 631 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK)); 632 if (!error) { 633 nfsm_loadattr(vp, ap->a_vap); 634 } 635 m_freem(mrep); 636 nfsmout: 637 return (error); 638 } 639 640 /* 641 * nfs setattr call. 642 * 643 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred) 644 */ 645 static int 646 nfs_setattr(struct vop_setattr_args *ap) 647 { 648 struct vnode *vp = ap->a_vp; 649 struct nfsnode *np = VTONFS(vp); 650 struct vattr *vap = ap->a_vap; 651 int error = 0; 652 u_quad_t tsize; 653 thread_t td = curthread; 654 655 #ifndef nolint 656 tsize = (u_quad_t)0; 657 #endif 658 659 /* 660 * Setting of flags is not supported. 661 */ 662 if (vap->va_flags != VNOVAL) 663 return (EOPNOTSUPP); 664 665 /* 666 * Disallow write attempts if the filesystem is mounted read-only. 667 */ 668 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 669 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 670 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 671 (vp->v_mount->mnt_flag & MNT_RDONLY)) 672 return (EROFS); 673 if (vap->va_size != VNOVAL) { 674 switch (vp->v_type) { 675 case VDIR: 676 return (EISDIR); 677 case VCHR: 678 case VBLK: 679 case VSOCK: 680 case VFIFO: 681 if (vap->va_mtime.tv_sec == VNOVAL && 682 vap->va_atime.tv_sec == VNOVAL && 683 vap->va_mode == (mode_t)VNOVAL && 684 vap->va_uid == (uid_t)VNOVAL && 685 vap->va_gid == (gid_t)VNOVAL) 686 return (0); 687 vap->va_size = VNOVAL; 688 break; 689 default: 690 /* 691 * Disallow write attempts if the filesystem is 692 * mounted read-only. 693 */ 694 if (vp->v_mount->mnt_flag & MNT_RDONLY) 695 return (EROFS); 696 697 /* 698 * This is nasty. The RPCs we send to flush pending 699 * data often return attribute information which is 700 * cached via a callback to nfs_loadattrcache(), which 701 * has the effect of changing our notion of the file 702 * size. Due to flushed appends and other operations 703 * the file size can be set to virtually anything, 704 * including values that do not match either the old 705 * or intended file size. 706 * 707 * When this condition is detected we must loop to 708 * try the operation again. Hopefully no more 709 * flushing is required on the loop so it works the 710 * second time around. THIS CASE ALMOST ALWAYS 711 * HAPPENS! 712 */ 713 tsize = np->n_size; 714 again: 715 error = nfs_meta_setsize(vp, td, vap->va_size); 716 717 if (np->n_flag & NLMODIFIED) { 718 if (vap->va_size == 0) 719 error = nfs_vinvalbuf(vp, 0, 1); 720 else 721 error = nfs_vinvalbuf(vp, V_SAVE, 1); 722 } 723 /* 724 * note: this loop case almost always happens at 725 * least once per truncation. 726 */ 727 if (error == 0 && np->n_size != vap->va_size) 728 goto again; 729 np->n_vattr.va_size = vap->va_size; 730 break; 731 } 732 } else if ((vap->va_mtime.tv_sec != VNOVAL || 733 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 734 vp->v_type == VREG && 735 (error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR 736 ) { 737 return (error); 738 } 739 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 740 741 /* 742 * Sanity check if a truncation was issued. This should only occur 743 * if multiple processes are racing on the same file. 744 */ 745 if (error == 0 && vap->va_size != VNOVAL && 746 np->n_size != vap->va_size) { 747 kprintf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 748 goto again; 749 } 750 if (error && vap->va_size != VNOVAL) { 751 np->n_size = np->n_vattr.va_size = tsize; 752 vnode_pager_setsize(vp, np->n_size); 753 } 754 return (error); 755 } 756 757 /* 758 * Do an nfs setattr rpc. 759 */ 760 static int 761 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 762 struct ucred *cred, struct thread *td) 763 { 764 struct nfsv2_sattr *sp; 765 struct nfsnode *np = VTONFS(vp); 766 caddr_t cp; 767 int32_t t1, t2; 768 caddr_t bpos, dpos, cp2; 769 u_int32_t *tl; 770 int error = 0, wccflag = NFSV3_WCCRATTR; 771 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 772 int v3 = NFS_ISV3(vp); 773 774 nfsstats.rpccnt[NFSPROC_SETATTR]++; 775 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 776 nfsm_fhtom(vp, v3); 777 if (v3) { 778 nfsm_v3attrbuild(vap, TRUE); 779 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 780 *tl = nfs_false; 781 } else { 782 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 783 if (vap->va_mode == (mode_t)VNOVAL) 784 sp->sa_mode = nfs_xdrneg1; 785 else 786 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 787 if (vap->va_uid == (uid_t)VNOVAL) 788 sp->sa_uid = nfs_xdrneg1; 789 else 790 sp->sa_uid = txdr_unsigned(vap->va_uid); 791 if (vap->va_gid == (gid_t)VNOVAL) 792 sp->sa_gid = nfs_xdrneg1; 793 else 794 sp->sa_gid = txdr_unsigned(vap->va_gid); 795 sp->sa_size = txdr_unsigned(vap->va_size); 796 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 797 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 798 } 799 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 800 if (v3) { 801 np->n_modestamp = 0; 802 nfsm_wcc_data(vp, wccflag); 803 } else 804 nfsm_loadattr(vp, (struct vattr *)0); 805 m_freem(mrep); 806 nfsmout: 807 return (error); 808 } 809 810 static 811 void 812 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout) 813 { 814 if (nctimeout == 0) 815 nctimeout = 1; 816 else 817 nctimeout *= hz; 818 cache_setvp(nch, vp); 819 cache_settimeout(nch, nctimeout); 820 } 821 822 /* 823 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 824 * nfs_lookup() until all remaining new api calls are implemented. 825 * 826 * Resolve a namecache entry. This function is passed a locked ncp and 827 * must call nfs_cache_setvp() on it as appropriate to resolve the entry. 828 */ 829 static int 830 nfs_nresolve(struct vop_nresolve_args *ap) 831 { 832 struct thread *td = curthread; 833 struct namecache *ncp; 834 struct ucred *cred; 835 struct nfsnode *np; 836 struct vnode *dvp; 837 struct vnode *nvp; 838 nfsfh_t *fhp; 839 int attrflag; 840 int fhsize; 841 int error; 842 int len; 843 int v3; 844 /******NFSM MACROS********/ 845 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 846 caddr_t bpos, dpos, cp, cp2; 847 u_int32_t *tl; 848 int32_t t1, t2; 849 850 cred = ap->a_cred; 851 dvp = ap->a_dvp; 852 853 if ((error = vget(dvp, LK_SHARED)) != 0) 854 return (error); 855 856 nvp = NULL; 857 v3 = NFS_ISV3(dvp); 858 nfsstats.lookupcache_misses++; 859 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 860 ncp = ap->a_nch->ncp; 861 len = ncp->nc_nlen; 862 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 863 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 864 nfsm_fhtom(dvp, v3); 865 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 866 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 867 if (error) { 868 /* 869 * Cache negatve lookups to reduce NFS traffic, but use 870 * a fast timeout. Otherwise use a timeout of 1 tick. 871 * XXX we should add a namecache flag for no-caching 872 * to uncache the negative hit as soon as possible, but 873 * we cannot simply destroy the entry because it is used 874 * as a placeholder by the caller. 875 */ 876 if (error == ENOENT) 877 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout); 878 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 879 m_freem(mrep); 880 goto nfsmout; 881 } 882 883 /* 884 * Success, get the file handle, do various checks, and load 885 * post-operation data from the reply packet. Theoretically 886 * we should never be looking up "." so, theoretically, we 887 * should never get the same file handle as our directory. But 888 * we check anyway. XXX 889 * 890 * Note that no timeout is set for the positive cache hit. We 891 * assume, theoretically, that ESTALE returns will be dealt with 892 * properly to handle NFS races and in anycase we cannot depend 893 * on a timeout to deal with NFS open/create/excl issues so instead 894 * of a bad hack here the rest of the NFS client code needs to do 895 * the right thing. 896 */ 897 nfsm_getfh(fhp, fhsize, v3); 898 899 np = VTONFS(dvp); 900 if (NFS_CMPFH(np, fhp, fhsize)) { 901 vref(dvp); 902 nvp = dvp; 903 } else { 904 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 905 if (error) { 906 m_freem(mrep); 907 vput(dvp); 908 return (error); 909 } 910 nvp = NFSTOV(np); 911 } 912 if (v3) { 913 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 914 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 915 } else { 916 nfsm_loadattr(nvp, NULL); 917 } 918 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout); 919 m_freem(mrep); 920 nfsmout: 921 vput(dvp); 922 if (nvp) { 923 if (nvp == dvp) 924 vrele(nvp); 925 else 926 vput(nvp); 927 } 928 return (error); 929 } 930 931 /* 932 * 'cached' nfs directory lookup 933 * 934 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 935 * 936 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 937 * struct componentname *a_cnp) 938 */ 939 static int 940 nfs_lookup(struct vop_old_lookup_args *ap) 941 { 942 struct componentname *cnp = ap->a_cnp; 943 struct vnode *dvp = ap->a_dvp; 944 struct vnode **vpp = ap->a_vpp; 945 int flags = cnp->cn_flags; 946 struct vnode *newvp; 947 u_int32_t *tl; 948 caddr_t cp; 949 int32_t t1, t2; 950 struct nfsmount *nmp; 951 caddr_t bpos, dpos, cp2; 952 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 953 long len; 954 nfsfh_t *fhp; 955 struct nfsnode *np; 956 int lockparent, wantparent, error = 0, attrflag, fhsize; 957 int v3 = NFS_ISV3(dvp); 958 959 /* 960 * Read-only mount check and directory check. 961 */ 962 *vpp = NULLVP; 963 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 964 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 965 return (EROFS); 966 967 if (dvp->v_type != VDIR) 968 return (ENOTDIR); 969 970 /* 971 * Look it up in the cache. Note that ENOENT is only returned if we 972 * previously entered a negative hit (see later on). The additional 973 * nfsneg_cache_timeout check causes previously cached results to 974 * be instantly ignored if the negative caching is turned off. 975 */ 976 lockparent = flags & CNP_LOCKPARENT; 977 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 978 nmp = VFSTONFS(dvp->v_mount); 979 np = VTONFS(dvp); 980 981 /* 982 * Go to the wire. 983 */ 984 error = 0; 985 newvp = NULLVP; 986 nfsstats.lookupcache_misses++; 987 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 988 len = cnp->cn_namelen; 989 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 990 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 991 nfsm_fhtom(dvp, v3); 992 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 993 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 994 if (error) { 995 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 996 m_freem(mrep); 997 goto nfsmout; 998 } 999 nfsm_getfh(fhp, fhsize, v3); 1000 1001 /* 1002 * Handle RENAME case... 1003 */ 1004 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1005 if (NFS_CMPFH(np, fhp, fhsize)) { 1006 m_freem(mrep); 1007 return (EISDIR); 1008 } 1009 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1010 if (error) { 1011 m_freem(mrep); 1012 return (error); 1013 } 1014 newvp = NFSTOV(np); 1015 if (v3) { 1016 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1017 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1018 } else 1019 nfsm_loadattr(newvp, (struct vattr *)0); 1020 *vpp = newvp; 1021 m_freem(mrep); 1022 if (!lockparent) { 1023 vn_unlock(dvp); 1024 cnp->cn_flags |= CNP_PDIRUNLOCK; 1025 } 1026 return (0); 1027 } 1028 1029 if (flags & CNP_ISDOTDOT) { 1030 vn_unlock(dvp); 1031 cnp->cn_flags |= CNP_PDIRUNLOCK; 1032 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1033 if (error) { 1034 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1035 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1036 return (error); /* NOTE: return error from nget */ 1037 } 1038 newvp = NFSTOV(np); 1039 if (lockparent) { 1040 error = vn_lock(dvp, LK_EXCLUSIVE); 1041 if (error) { 1042 vput(newvp); 1043 return (error); 1044 } 1045 cnp->cn_flags |= CNP_PDIRUNLOCK; 1046 } 1047 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1048 vref(dvp); 1049 newvp = dvp; 1050 } else { 1051 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1052 if (error) { 1053 m_freem(mrep); 1054 return (error); 1055 } 1056 if (!lockparent) { 1057 vn_unlock(dvp); 1058 cnp->cn_flags |= CNP_PDIRUNLOCK; 1059 } 1060 newvp = NFSTOV(np); 1061 } 1062 if (v3) { 1063 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1064 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1065 } else 1066 nfsm_loadattr(newvp, (struct vattr *)0); 1067 #if 0 1068 /* XXX MOVE TO nfs_nremove() */ 1069 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1070 cnp->cn_nameiop != NAMEI_DELETE) { 1071 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1072 } 1073 #endif 1074 *vpp = newvp; 1075 m_freem(mrep); 1076 nfsmout: 1077 if (error) { 1078 if (newvp != NULLVP) { 1079 vrele(newvp); 1080 *vpp = NULLVP; 1081 } 1082 if ((cnp->cn_nameiop == NAMEI_CREATE || 1083 cnp->cn_nameiop == NAMEI_RENAME) && 1084 error == ENOENT) { 1085 if (!lockparent) { 1086 vn_unlock(dvp); 1087 cnp->cn_flags |= CNP_PDIRUNLOCK; 1088 } 1089 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1090 error = EROFS; 1091 else 1092 error = EJUSTRETURN; 1093 } 1094 } 1095 return (error); 1096 } 1097 1098 /* 1099 * nfs read call. 1100 * Just call nfs_bioread() to do the work. 1101 * 1102 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1103 * struct ucred *a_cred) 1104 */ 1105 static int 1106 nfs_read(struct vop_read_args *ap) 1107 { 1108 struct vnode *vp = ap->a_vp; 1109 1110 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1111 switch (vp->v_type) { 1112 case VREG: 1113 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1114 case VDIR: 1115 return (EISDIR); 1116 default: 1117 return EOPNOTSUPP; 1118 } 1119 } 1120 1121 /* 1122 * nfs readlink call 1123 * 1124 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1125 */ 1126 static int 1127 nfs_readlink(struct vop_readlink_args *ap) 1128 { 1129 struct vnode *vp = ap->a_vp; 1130 1131 if (vp->v_type != VLNK) 1132 return (EINVAL); 1133 return (nfs_bioread(vp, ap->a_uio, 0)); 1134 } 1135 1136 /* 1137 * Do a readlink rpc. 1138 * Called by nfs_doio() from below the buffer cache. 1139 */ 1140 int 1141 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1142 { 1143 u_int32_t *tl; 1144 caddr_t cp; 1145 int32_t t1, t2; 1146 caddr_t bpos, dpos, cp2; 1147 int error = 0, len, attrflag; 1148 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1149 int v3 = NFS_ISV3(vp); 1150 1151 nfsstats.rpccnt[NFSPROC_READLINK]++; 1152 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1153 nfsm_fhtom(vp, v3); 1154 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1155 if (v3) 1156 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1157 if (!error) { 1158 nfsm_strsiz(len, NFS_MAXPATHLEN); 1159 if (len == NFS_MAXPATHLEN) { 1160 struct nfsnode *np = VTONFS(vp); 1161 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1162 len = np->n_size; 1163 } 1164 nfsm_mtouio(uiop, len); 1165 } 1166 m_freem(mrep); 1167 nfsmout: 1168 return (error); 1169 } 1170 1171 /* 1172 * nfs read rpc call 1173 * Ditto above 1174 */ 1175 int 1176 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1177 { 1178 u_int32_t *tl; 1179 caddr_t cp; 1180 int32_t t1, t2; 1181 caddr_t bpos, dpos, cp2; 1182 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1183 struct nfsmount *nmp; 1184 int error = 0, len, retlen, tsiz, eof, attrflag; 1185 int v3 = NFS_ISV3(vp); 1186 1187 #ifndef nolint 1188 eof = 0; 1189 #endif 1190 nmp = VFSTONFS(vp->v_mount); 1191 tsiz = uiop->uio_resid; 1192 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1193 return (EFBIG); 1194 while (tsiz > 0) { 1195 nfsstats.rpccnt[NFSPROC_READ]++; 1196 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1197 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1198 nfsm_fhtom(vp, v3); 1199 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1200 if (v3) { 1201 txdr_hyper(uiop->uio_offset, tl); 1202 *(tl + 2) = txdr_unsigned(len); 1203 } else { 1204 *tl++ = txdr_unsigned(uiop->uio_offset); 1205 *tl++ = txdr_unsigned(len); 1206 *tl = 0; 1207 } 1208 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1209 if (v3) { 1210 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1211 if (error) { 1212 m_freem(mrep); 1213 goto nfsmout; 1214 } 1215 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1216 eof = fxdr_unsigned(int, *(tl + 1)); 1217 } else 1218 nfsm_loadattr(vp, (struct vattr *)0); 1219 nfsm_strsiz(retlen, nmp->nm_rsize); 1220 nfsm_mtouio(uiop, retlen); 1221 m_freem(mrep); 1222 tsiz -= retlen; 1223 if (v3) { 1224 if (eof || retlen == 0) { 1225 tsiz = 0; 1226 } 1227 } else if (retlen < len) { 1228 tsiz = 0; 1229 } 1230 } 1231 nfsmout: 1232 return (error); 1233 } 1234 1235 /* 1236 * nfs write call 1237 */ 1238 int 1239 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1240 { 1241 u_int32_t *tl; 1242 caddr_t cp; 1243 int32_t t1, t2, backup; 1244 caddr_t bpos, dpos, cp2; 1245 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1246 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1247 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1248 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1249 1250 #ifndef DIAGNOSTIC 1251 if (uiop->uio_iovcnt != 1) 1252 panic("nfs: writerpc iovcnt > 1"); 1253 #endif 1254 *must_commit = 0; 1255 tsiz = uiop->uio_resid; 1256 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1257 return (EFBIG); 1258 while (tsiz > 0) { 1259 nfsstats.rpccnt[NFSPROC_WRITE]++; 1260 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1261 nfsm_reqhead(vp, NFSPROC_WRITE, 1262 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1263 nfsm_fhtom(vp, v3); 1264 if (v3) { 1265 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1266 txdr_hyper(uiop->uio_offset, tl); 1267 tl += 2; 1268 *tl++ = txdr_unsigned(len); 1269 *tl++ = txdr_unsigned(*iomode); 1270 *tl = txdr_unsigned(len); 1271 } else { 1272 u_int32_t x; 1273 1274 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1275 /* Set both "begin" and "current" to non-garbage. */ 1276 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1277 *tl++ = x; /* "begin offset" */ 1278 *tl++ = x; /* "current offset" */ 1279 x = txdr_unsigned(len); 1280 *tl++ = x; /* total to this offset */ 1281 *tl = x; /* size of this write */ 1282 } 1283 nfsm_uiotom(uiop, len); 1284 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1285 if (v3) { 1286 /* 1287 * The write RPC returns a before and after mtime. The 1288 * nfsm_wcc_data() macro checks the before n_mtime 1289 * against the before time and stores the after time 1290 * in the nfsnode's cached vattr and n_mtime field. 1291 * The NRMODIFIED bit will be set if the before 1292 * time did not match the original mtime. 1293 */ 1294 wccflag = NFSV3_WCCCHK; 1295 nfsm_wcc_data(vp, wccflag); 1296 if (!error) { 1297 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1298 + NFSX_V3WRITEVERF); 1299 rlen = fxdr_unsigned(int, *tl++); 1300 if (rlen == 0) { 1301 error = NFSERR_IO; 1302 m_freem(mrep); 1303 break; 1304 } else if (rlen < len) { 1305 backup = len - rlen; 1306 uiop->uio_iov->iov_base -= backup; 1307 uiop->uio_iov->iov_len += backup; 1308 uiop->uio_offset -= backup; 1309 uiop->uio_resid += backup; 1310 len = rlen; 1311 } 1312 commit = fxdr_unsigned(int, *tl++); 1313 1314 /* 1315 * Return the lowest committment level 1316 * obtained by any of the RPCs. 1317 */ 1318 if (committed == NFSV3WRITE_FILESYNC) 1319 committed = commit; 1320 else if (committed == NFSV3WRITE_DATASYNC && 1321 commit == NFSV3WRITE_UNSTABLE) 1322 committed = commit; 1323 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1324 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1325 NFSX_V3WRITEVERF); 1326 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1327 } else if (bcmp((caddr_t)tl, 1328 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1329 *must_commit = 1; 1330 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1331 NFSX_V3WRITEVERF); 1332 } 1333 } 1334 } else { 1335 nfsm_loadattr(vp, (struct vattr *)0); 1336 } 1337 m_freem(mrep); 1338 if (error) 1339 break; 1340 tsiz -= len; 1341 } 1342 nfsmout: 1343 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1344 committed = NFSV3WRITE_FILESYNC; 1345 *iomode = committed; 1346 if (error) 1347 uiop->uio_resid = tsiz; 1348 return (error); 1349 } 1350 1351 /* 1352 * nfs mknod rpc 1353 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1354 * mode set to specify the file type and the size field for rdev. 1355 */ 1356 static int 1357 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1358 struct vattr *vap) 1359 { 1360 struct nfsv2_sattr *sp; 1361 u_int32_t *tl; 1362 caddr_t cp; 1363 int32_t t1, t2; 1364 struct vnode *newvp = (struct vnode *)0; 1365 struct nfsnode *np = (struct nfsnode *)0; 1366 struct vattr vattr; 1367 char *cp2; 1368 caddr_t bpos, dpos; 1369 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1370 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1371 int rmajor, rminor; 1372 int v3 = NFS_ISV3(dvp); 1373 1374 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1375 rmajor = txdr_unsigned(vap->va_rmajor); 1376 rminor = txdr_unsigned(vap->va_rminor); 1377 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) { 1378 rmajor = nfs_xdrneg1; 1379 rminor = nfs_xdrneg1; 1380 } else { 1381 return (EOPNOTSUPP); 1382 } 1383 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1384 return (error); 1385 } 1386 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1387 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1388 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1389 nfsm_fhtom(dvp, v3); 1390 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1391 if (v3) { 1392 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1393 *tl++ = vtonfsv3_type(vap->va_type); 1394 nfsm_v3attrbuild(vap, FALSE); 1395 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1396 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1397 *tl++ = txdr_unsigned(vap->va_rmajor); 1398 *tl = txdr_unsigned(vap->va_rminor); 1399 } 1400 } else { 1401 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1402 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1403 sp->sa_uid = nfs_xdrneg1; 1404 sp->sa_gid = nfs_xdrneg1; 1405 sp->sa_size = makeudev(rmajor, rminor); 1406 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1407 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1408 } 1409 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1410 if (!error) { 1411 nfsm_mtofh(dvp, newvp, v3, gotvp); 1412 if (!gotvp) { 1413 if (newvp) { 1414 vput(newvp); 1415 newvp = (struct vnode *)0; 1416 } 1417 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1418 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1419 if (!error) 1420 newvp = NFSTOV(np); 1421 } 1422 } 1423 if (v3) 1424 nfsm_wcc_data(dvp, wccflag); 1425 m_freem(mrep); 1426 nfsmout: 1427 if (error) { 1428 if (newvp) 1429 vput(newvp); 1430 } else { 1431 *vpp = newvp; 1432 } 1433 VTONFS(dvp)->n_flag |= NLMODIFIED; 1434 if (!wccflag) 1435 VTONFS(dvp)->n_attrstamp = 0; 1436 return (error); 1437 } 1438 1439 /* 1440 * nfs mknod vop 1441 * just call nfs_mknodrpc() to do the work. 1442 * 1443 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1444 * struct componentname *a_cnp, struct vattr *a_vap) 1445 */ 1446 /* ARGSUSED */ 1447 static int 1448 nfs_mknod(struct vop_old_mknod_args *ap) 1449 { 1450 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1451 } 1452 1453 static u_long create_verf; 1454 /* 1455 * nfs file create call 1456 * 1457 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1458 * struct componentname *a_cnp, struct vattr *a_vap) 1459 */ 1460 static int 1461 nfs_create(struct vop_old_create_args *ap) 1462 { 1463 struct vnode *dvp = ap->a_dvp; 1464 struct vattr *vap = ap->a_vap; 1465 struct componentname *cnp = ap->a_cnp; 1466 struct nfsv2_sattr *sp; 1467 u_int32_t *tl; 1468 caddr_t cp; 1469 int32_t t1, t2; 1470 struct nfsnode *np = (struct nfsnode *)0; 1471 struct vnode *newvp = (struct vnode *)0; 1472 caddr_t bpos, dpos, cp2; 1473 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1474 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1475 struct vattr vattr; 1476 int v3 = NFS_ISV3(dvp); 1477 1478 /* 1479 * Oops, not for me.. 1480 */ 1481 if (vap->va_type == VSOCK) 1482 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1483 1484 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1485 return (error); 1486 } 1487 if (vap->va_vaflags & VA_EXCLUSIVE) 1488 fmode |= O_EXCL; 1489 again: 1490 nfsstats.rpccnt[NFSPROC_CREATE]++; 1491 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1492 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1493 nfsm_fhtom(dvp, v3); 1494 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1495 if (v3) { 1496 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1497 if (fmode & O_EXCL) { 1498 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1499 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1500 #ifdef INET 1501 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1502 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr; 1503 else 1504 #endif 1505 *tl++ = create_verf; 1506 *tl = ++create_verf; 1507 } else { 1508 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1509 nfsm_v3attrbuild(vap, FALSE); 1510 } 1511 } else { 1512 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1513 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1514 sp->sa_uid = nfs_xdrneg1; 1515 sp->sa_gid = nfs_xdrneg1; 1516 sp->sa_size = 0; 1517 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1518 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1519 } 1520 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1521 if (!error) { 1522 nfsm_mtofh(dvp, newvp, v3, gotvp); 1523 if (!gotvp) { 1524 if (newvp) { 1525 vput(newvp); 1526 newvp = (struct vnode *)0; 1527 } 1528 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1529 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1530 if (!error) 1531 newvp = NFSTOV(np); 1532 } 1533 } 1534 if (v3) 1535 nfsm_wcc_data(dvp, wccflag); 1536 m_freem(mrep); 1537 nfsmout: 1538 if (error) { 1539 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1540 fmode &= ~O_EXCL; 1541 goto again; 1542 } 1543 if (newvp) 1544 vput(newvp); 1545 } else if (v3 && (fmode & O_EXCL)) { 1546 /* 1547 * We are normally called with only a partially initialized 1548 * VAP. Since the NFSv3 spec says that server may use the 1549 * file attributes to store the verifier, the spec requires 1550 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1551 * in atime, but we can't really assume that all servers will 1552 * so we ensure that our SETATTR sets both atime and mtime. 1553 */ 1554 if (vap->va_mtime.tv_sec == VNOVAL) 1555 vfs_timestamp(&vap->va_mtime); 1556 if (vap->va_atime.tv_sec == VNOVAL) 1557 vap->va_atime = vap->va_mtime; 1558 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1559 } 1560 if (!error) { 1561 /* 1562 * The new np may have enough info for access 1563 * checks, make sure rucred and wucred are 1564 * initialized for read and write rpc's. 1565 */ 1566 np = VTONFS(newvp); 1567 if (np->n_rucred == NULL) 1568 np->n_rucred = crhold(cnp->cn_cred); 1569 if (np->n_wucred == NULL) 1570 np->n_wucred = crhold(cnp->cn_cred); 1571 *ap->a_vpp = newvp; 1572 } 1573 VTONFS(dvp)->n_flag |= NLMODIFIED; 1574 if (!wccflag) 1575 VTONFS(dvp)->n_attrstamp = 0; 1576 return (error); 1577 } 1578 1579 /* 1580 * nfs file remove call 1581 * To try and make nfs semantics closer to ufs semantics, a file that has 1582 * other processes using the vnode is renamed instead of removed and then 1583 * removed later on the last close. 1584 * - If v_sysref.refcnt > 1 1585 * If a rename is not already in the works 1586 * call nfs_sillyrename() to set it up 1587 * else 1588 * do the remove rpc 1589 * 1590 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp, 1591 * struct componentname *a_cnp) 1592 */ 1593 static int 1594 nfs_remove(struct vop_old_remove_args *ap) 1595 { 1596 struct vnode *vp = ap->a_vp; 1597 struct vnode *dvp = ap->a_dvp; 1598 struct componentname *cnp = ap->a_cnp; 1599 struct nfsnode *np = VTONFS(vp); 1600 int error = 0; 1601 struct vattr vattr; 1602 1603 #ifndef DIAGNOSTIC 1604 if (vp->v_sysref.refcnt < 1) 1605 panic("nfs_remove: bad v_sysref.refcnt"); 1606 #endif 1607 if (vp->v_type == VDIR) 1608 error = EPERM; 1609 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename && 1610 VOP_GETATTR(vp, &vattr) == 0 && 1611 vattr.va_nlink > 1)) { 1612 /* 1613 * throw away biocache buffers, mainly to avoid 1614 * unnecessary delayed writes later. 1615 */ 1616 error = nfs_vinvalbuf(vp, 0, 1); 1617 /* Do the rpc */ 1618 if (error != EINTR) 1619 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1620 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1621 /* 1622 * Kludge City: If the first reply to the remove rpc is lost.. 1623 * the reply to the retransmitted request will be ENOENT 1624 * since the file was in fact removed 1625 * Therefore, we cheat and return success. 1626 */ 1627 if (error == ENOENT) 1628 error = 0; 1629 } else if (!np->n_sillyrename) { 1630 error = nfs_sillyrename(dvp, vp, cnp); 1631 } 1632 np->n_attrstamp = 0; 1633 return (error); 1634 } 1635 1636 /* 1637 * nfs file remove rpc called from nfs_inactive 1638 */ 1639 int 1640 nfs_removeit(struct sillyrename *sp) 1641 { 1642 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1643 sp->s_cred, NULL)); 1644 } 1645 1646 /* 1647 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1648 */ 1649 static int 1650 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1651 struct ucred *cred, struct thread *td) 1652 { 1653 u_int32_t *tl; 1654 caddr_t cp; 1655 int32_t t1, t2; 1656 caddr_t bpos, dpos, cp2; 1657 int error = 0, wccflag = NFSV3_WCCRATTR; 1658 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1659 int v3 = NFS_ISV3(dvp); 1660 1661 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1662 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1663 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1664 nfsm_fhtom(dvp, v3); 1665 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1666 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1667 if (v3) 1668 nfsm_wcc_data(dvp, wccflag); 1669 m_freem(mrep); 1670 nfsmout: 1671 VTONFS(dvp)->n_flag |= NLMODIFIED; 1672 if (!wccflag) 1673 VTONFS(dvp)->n_attrstamp = 0; 1674 return (error); 1675 } 1676 1677 /* 1678 * nfs file rename call 1679 * 1680 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1681 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1682 * struct vnode *a_tvp, struct componentname *a_tcnp) 1683 */ 1684 static int 1685 nfs_rename(struct vop_old_rename_args *ap) 1686 { 1687 struct vnode *fvp = ap->a_fvp; 1688 struct vnode *tvp = ap->a_tvp; 1689 struct vnode *fdvp = ap->a_fdvp; 1690 struct vnode *tdvp = ap->a_tdvp; 1691 struct componentname *tcnp = ap->a_tcnp; 1692 struct componentname *fcnp = ap->a_fcnp; 1693 int error; 1694 1695 /* Check for cross-device rename */ 1696 if ((fvp->v_mount != tdvp->v_mount) || 1697 (tvp && (fvp->v_mount != tvp->v_mount))) { 1698 error = EXDEV; 1699 goto out; 1700 } 1701 1702 /* 1703 * We have to flush B_DELWRI data prior to renaming 1704 * the file. If we don't, the delayed-write buffers 1705 * can be flushed out later after the file has gone stale 1706 * under NFSV3. NFSV2 does not have this problem because 1707 * ( as far as I can tell ) it flushes dirty buffers more 1708 * often. 1709 */ 1710 1711 VOP_FSYNC(fvp, MNT_WAIT); 1712 if (tvp) 1713 VOP_FSYNC(tvp, MNT_WAIT); 1714 1715 /* 1716 * If the tvp exists and is in use, sillyrename it before doing the 1717 * rename of the new file over it. 1718 * 1719 * XXX Can't sillyrename a directory. 1720 * 1721 * We do not attempt to do any namecache purges in this old API 1722 * routine. The new API compat functions have access to the actual 1723 * namecache structures and will do it for us. 1724 */ 1725 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename && 1726 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1727 vput(tvp); 1728 tvp = NULL; 1729 } else if (tvp) { 1730 ; 1731 } 1732 1733 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1734 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1735 tcnp->cn_td); 1736 1737 out: 1738 if (tdvp == tvp) 1739 vrele(tdvp); 1740 else 1741 vput(tdvp); 1742 if (tvp) 1743 vput(tvp); 1744 vrele(fdvp); 1745 vrele(fvp); 1746 /* 1747 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1748 */ 1749 if (error == ENOENT) 1750 error = 0; 1751 return (error); 1752 } 1753 1754 /* 1755 * nfs file rename rpc called from nfs_remove() above 1756 */ 1757 static int 1758 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1759 struct sillyrename *sp) 1760 { 1761 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1762 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1763 } 1764 1765 /* 1766 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1767 */ 1768 static int 1769 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1770 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1771 struct ucred *cred, struct thread *td) 1772 { 1773 u_int32_t *tl; 1774 caddr_t cp; 1775 int32_t t1, t2; 1776 caddr_t bpos, dpos, cp2; 1777 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1778 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1779 int v3 = NFS_ISV3(fdvp); 1780 1781 nfsstats.rpccnt[NFSPROC_RENAME]++; 1782 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1783 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1784 nfsm_rndup(tnamelen)); 1785 nfsm_fhtom(fdvp, v3); 1786 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1787 nfsm_fhtom(tdvp, v3); 1788 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1789 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1790 if (v3) { 1791 nfsm_wcc_data(fdvp, fwccflag); 1792 nfsm_wcc_data(tdvp, twccflag); 1793 } 1794 m_freem(mrep); 1795 nfsmout: 1796 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1797 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1798 if (!fwccflag) 1799 VTONFS(fdvp)->n_attrstamp = 0; 1800 if (!twccflag) 1801 VTONFS(tdvp)->n_attrstamp = 0; 1802 return (error); 1803 } 1804 1805 /* 1806 * nfs hard link create call 1807 * 1808 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1809 * struct componentname *a_cnp) 1810 */ 1811 static int 1812 nfs_link(struct vop_old_link_args *ap) 1813 { 1814 struct vnode *vp = ap->a_vp; 1815 struct vnode *tdvp = ap->a_tdvp; 1816 struct componentname *cnp = ap->a_cnp; 1817 u_int32_t *tl; 1818 caddr_t cp; 1819 int32_t t1, t2; 1820 caddr_t bpos, dpos, cp2; 1821 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1822 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1823 int v3; 1824 1825 if (vp->v_mount != tdvp->v_mount) { 1826 return (EXDEV); 1827 } 1828 1829 /* 1830 * Push all writes to the server, so that the attribute cache 1831 * doesn't get "out of sync" with the server. 1832 * XXX There should be a better way! 1833 */ 1834 VOP_FSYNC(vp, MNT_WAIT); 1835 1836 v3 = NFS_ISV3(vp); 1837 nfsstats.rpccnt[NFSPROC_LINK]++; 1838 nfsm_reqhead(vp, NFSPROC_LINK, 1839 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1840 nfsm_fhtom(vp, v3); 1841 nfsm_fhtom(tdvp, v3); 1842 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1843 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1844 if (v3) { 1845 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1846 nfsm_wcc_data(tdvp, wccflag); 1847 } 1848 m_freem(mrep); 1849 nfsmout: 1850 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1851 if (!attrflag) 1852 VTONFS(vp)->n_attrstamp = 0; 1853 if (!wccflag) 1854 VTONFS(tdvp)->n_attrstamp = 0; 1855 /* 1856 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1857 */ 1858 if (error == EEXIST) 1859 error = 0; 1860 return (error); 1861 } 1862 1863 /* 1864 * nfs symbolic link create call 1865 * 1866 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1867 * struct componentname *a_cnp, struct vattr *a_vap, 1868 * char *a_target) 1869 */ 1870 static int 1871 nfs_symlink(struct vop_old_symlink_args *ap) 1872 { 1873 struct vnode *dvp = ap->a_dvp; 1874 struct vattr *vap = ap->a_vap; 1875 struct componentname *cnp = ap->a_cnp; 1876 struct nfsv2_sattr *sp; 1877 u_int32_t *tl; 1878 caddr_t cp; 1879 int32_t t1, t2; 1880 caddr_t bpos, dpos, cp2; 1881 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1882 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1883 struct vnode *newvp = (struct vnode *)0; 1884 int v3 = NFS_ISV3(dvp); 1885 1886 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1887 slen = strlen(ap->a_target); 1888 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1889 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1890 nfsm_fhtom(dvp, v3); 1891 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1892 if (v3) { 1893 nfsm_v3attrbuild(vap, FALSE); 1894 } 1895 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1896 if (!v3) { 1897 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1898 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1899 sp->sa_uid = nfs_xdrneg1; 1900 sp->sa_gid = nfs_xdrneg1; 1901 sp->sa_size = nfs_xdrneg1; 1902 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1903 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1904 } 1905 1906 /* 1907 * Issue the NFS request and get the rpc response. 1908 * 1909 * Only NFSv3 responses returning an error of 0 actually return 1910 * a file handle that can be converted into newvp without having 1911 * to do an extra lookup rpc. 1912 */ 1913 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1914 if (v3) { 1915 if (error == 0) 1916 nfsm_mtofh(dvp, newvp, v3, gotvp); 1917 nfsm_wcc_data(dvp, wccflag); 1918 } 1919 1920 /* 1921 * out code jumps -> here, mrep is also freed. 1922 */ 1923 1924 m_freem(mrep); 1925 nfsmout: 1926 1927 /* 1928 * If we get an EEXIST error, silently convert it to no-error 1929 * in case of an NFS retry. 1930 */ 1931 if (error == EEXIST) 1932 error = 0; 1933 1934 /* 1935 * If we do not have (or no longer have) an error, and we could 1936 * not extract the newvp from the response due to the request being 1937 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1938 * to obtain a newvp to return. 1939 */ 1940 if (error == 0 && newvp == NULL) { 1941 struct nfsnode *np = NULL; 1942 1943 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1944 cnp->cn_cred, cnp->cn_td, &np); 1945 if (!error) 1946 newvp = NFSTOV(np); 1947 } 1948 if (error) { 1949 if (newvp) 1950 vput(newvp); 1951 } else { 1952 *ap->a_vpp = newvp; 1953 } 1954 VTONFS(dvp)->n_flag |= NLMODIFIED; 1955 if (!wccflag) 1956 VTONFS(dvp)->n_attrstamp = 0; 1957 return (error); 1958 } 1959 1960 /* 1961 * nfs make dir call 1962 * 1963 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1964 * struct componentname *a_cnp, struct vattr *a_vap) 1965 */ 1966 static int 1967 nfs_mkdir(struct vop_old_mkdir_args *ap) 1968 { 1969 struct vnode *dvp = ap->a_dvp; 1970 struct vattr *vap = ap->a_vap; 1971 struct componentname *cnp = ap->a_cnp; 1972 struct nfsv2_sattr *sp; 1973 u_int32_t *tl; 1974 caddr_t cp; 1975 int32_t t1, t2; 1976 int len; 1977 struct nfsnode *np = (struct nfsnode *)0; 1978 struct vnode *newvp = (struct vnode *)0; 1979 caddr_t bpos, dpos, cp2; 1980 int error = 0, wccflag = NFSV3_WCCRATTR; 1981 int gotvp = 0; 1982 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1983 struct vattr vattr; 1984 int v3 = NFS_ISV3(dvp); 1985 1986 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1987 return (error); 1988 } 1989 len = cnp->cn_namelen; 1990 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1991 nfsm_reqhead(dvp, NFSPROC_MKDIR, 1992 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 1993 nfsm_fhtom(dvp, v3); 1994 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1995 if (v3) { 1996 nfsm_v3attrbuild(vap, FALSE); 1997 } else { 1998 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1999 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2000 sp->sa_uid = nfs_xdrneg1; 2001 sp->sa_gid = nfs_xdrneg1; 2002 sp->sa_size = nfs_xdrneg1; 2003 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2004 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2005 } 2006 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2007 if (!error) 2008 nfsm_mtofh(dvp, newvp, v3, gotvp); 2009 if (v3) 2010 nfsm_wcc_data(dvp, wccflag); 2011 m_freem(mrep); 2012 nfsmout: 2013 VTONFS(dvp)->n_flag |= NLMODIFIED; 2014 if (!wccflag) 2015 VTONFS(dvp)->n_attrstamp = 0; 2016 /* 2017 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2018 * if we can succeed in looking up the directory. 2019 */ 2020 if (error == EEXIST || (!error && !gotvp)) { 2021 if (newvp) { 2022 vrele(newvp); 2023 newvp = (struct vnode *)0; 2024 } 2025 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2026 cnp->cn_td, &np); 2027 if (!error) { 2028 newvp = NFSTOV(np); 2029 if (newvp->v_type != VDIR) 2030 error = EEXIST; 2031 } 2032 } 2033 if (error) { 2034 if (newvp) 2035 vrele(newvp); 2036 } else 2037 *ap->a_vpp = newvp; 2038 return (error); 2039 } 2040 2041 /* 2042 * nfs remove directory call 2043 * 2044 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2045 * struct componentname *a_cnp) 2046 */ 2047 static int 2048 nfs_rmdir(struct vop_old_rmdir_args *ap) 2049 { 2050 struct vnode *vp = ap->a_vp; 2051 struct vnode *dvp = ap->a_dvp; 2052 struct componentname *cnp = ap->a_cnp; 2053 u_int32_t *tl; 2054 caddr_t cp; 2055 int32_t t1, t2; 2056 caddr_t bpos, dpos, cp2; 2057 int error = 0, wccflag = NFSV3_WCCRATTR; 2058 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2059 int v3 = NFS_ISV3(dvp); 2060 2061 if (dvp == vp) 2062 return (EINVAL); 2063 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2064 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2065 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2066 nfsm_fhtom(dvp, v3); 2067 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2068 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2069 if (v3) 2070 nfsm_wcc_data(dvp, wccflag); 2071 m_freem(mrep); 2072 nfsmout: 2073 VTONFS(dvp)->n_flag |= NLMODIFIED; 2074 if (!wccflag) 2075 VTONFS(dvp)->n_attrstamp = 0; 2076 /* 2077 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2078 */ 2079 if (error == ENOENT) 2080 error = 0; 2081 return (error); 2082 } 2083 2084 /* 2085 * nfs readdir call 2086 * 2087 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2088 */ 2089 static int 2090 nfs_readdir(struct vop_readdir_args *ap) 2091 { 2092 struct vnode *vp = ap->a_vp; 2093 struct nfsnode *np = VTONFS(vp); 2094 struct uio *uio = ap->a_uio; 2095 int tresid, error; 2096 struct vattr vattr; 2097 2098 if (vp->v_type != VDIR) 2099 return (EPERM); 2100 2101 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 2102 return (error); 2103 2104 /* 2105 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2106 * and then check that is still valid, or if this is an NQNFS mount 2107 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2108 * VOP_GETATTR() does not necessarily go to the wire. 2109 */ 2110 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2111 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2112 if (VOP_GETATTR(vp, &vattr) == 0 && 2113 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2114 ) { 2115 nfsstats.direofcache_hits++; 2116 goto done; 2117 } 2118 } 2119 2120 /* 2121 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2122 * own cache coherency checks so we do not have to. 2123 */ 2124 tresid = uio->uio_resid; 2125 error = nfs_bioread(vp, uio, 0); 2126 2127 if (!error && uio->uio_resid == tresid) 2128 nfsstats.direofcache_misses++; 2129 done: 2130 vn_unlock(vp); 2131 return (error); 2132 } 2133 2134 /* 2135 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2136 * 2137 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2138 * offset/block and converts the nfs formatted directory entries for userland 2139 * consumption as well as deals with offsets into the middle of blocks. 2140 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2141 * be block-bounded. It must convert to cookies for the actual RPC. 2142 */ 2143 int 2144 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2145 { 2146 int len, left; 2147 struct nfs_dirent *dp = NULL; 2148 u_int32_t *tl; 2149 caddr_t cp; 2150 int32_t t1, t2; 2151 nfsuint64 *cookiep; 2152 caddr_t bpos, dpos, cp2; 2153 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2154 nfsuint64 cookie; 2155 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2156 struct nfsnode *dnp = VTONFS(vp); 2157 u_quad_t fileno; 2158 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2159 int attrflag; 2160 int v3 = NFS_ISV3(vp); 2161 2162 #ifndef DIAGNOSTIC 2163 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2164 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2165 panic("nfs readdirrpc bad uio"); 2166 #endif 2167 2168 /* 2169 * If there is no cookie, assume directory was stale. 2170 */ 2171 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2172 if (cookiep) 2173 cookie = *cookiep; 2174 else 2175 return (NFSERR_BAD_COOKIE); 2176 /* 2177 * Loop around doing readdir rpc's of size nm_readdirsize 2178 * truncated to a multiple of DIRBLKSIZ. 2179 * The stopping criteria is EOF or buffer full. 2180 */ 2181 while (more_dirs && bigenough) { 2182 nfsstats.rpccnt[NFSPROC_READDIR]++; 2183 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2184 NFSX_READDIR(v3)); 2185 nfsm_fhtom(vp, v3); 2186 if (v3) { 2187 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2188 *tl++ = cookie.nfsuquad[0]; 2189 *tl++ = cookie.nfsuquad[1]; 2190 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2191 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2192 } else { 2193 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2194 *tl++ = cookie.nfsuquad[0]; 2195 } 2196 *tl = txdr_unsigned(nmp->nm_readdirsize); 2197 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2198 if (v3) { 2199 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2200 if (!error) { 2201 nfsm_dissect(tl, u_int32_t *, 2202 2 * NFSX_UNSIGNED); 2203 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2204 dnp->n_cookieverf.nfsuquad[1] = *tl; 2205 } else { 2206 m_freem(mrep); 2207 goto nfsmout; 2208 } 2209 } 2210 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2211 more_dirs = fxdr_unsigned(int, *tl); 2212 2213 /* loop thru the dir entries, converting them to std form */ 2214 while (more_dirs && bigenough) { 2215 if (v3) { 2216 nfsm_dissect(tl, u_int32_t *, 2217 3 * NFSX_UNSIGNED); 2218 fileno = fxdr_hyper(tl); 2219 len = fxdr_unsigned(int, *(tl + 2)); 2220 } else { 2221 nfsm_dissect(tl, u_int32_t *, 2222 2 * NFSX_UNSIGNED); 2223 fileno = fxdr_unsigned(u_quad_t, *tl++); 2224 len = fxdr_unsigned(int, *tl); 2225 } 2226 if (len <= 0 || len > NFS_MAXNAMLEN) { 2227 error = EBADRPC; 2228 m_freem(mrep); 2229 goto nfsmout; 2230 } 2231 2232 /* 2233 * len is the number of bytes in the path element 2234 * name, not including the \0 termination. 2235 * 2236 * tlen is the number of bytes w have to reserve for 2237 * the path element name. 2238 */ 2239 tlen = nfsm_rndup(len); 2240 if (tlen == len) 2241 tlen += 4; /* To ensure null termination */ 2242 2243 /* 2244 * If the entry would cross a DIRBLKSIZ boundary, 2245 * extend the previous nfs_dirent to cover the 2246 * remaining space. 2247 */ 2248 left = DIRBLKSIZ - blksiz; 2249 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2250 dp->nfs_reclen += left; 2251 uiop->uio_iov->iov_base += left; 2252 uiop->uio_iov->iov_len -= left; 2253 uiop->uio_offset += left; 2254 uiop->uio_resid -= left; 2255 blksiz = 0; 2256 } 2257 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2258 bigenough = 0; 2259 if (bigenough) { 2260 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2261 dp->nfs_ino = fileno; 2262 dp->nfs_namlen = len; 2263 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2264 dp->nfs_type = DT_UNKNOWN; 2265 blksiz += dp->nfs_reclen; 2266 if (blksiz == DIRBLKSIZ) 2267 blksiz = 0; 2268 uiop->uio_offset += sizeof(struct nfs_dirent); 2269 uiop->uio_resid -= sizeof(struct nfs_dirent); 2270 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2271 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2272 nfsm_mtouio(uiop, len); 2273 2274 /* 2275 * The uiop has advanced by nfs_dirent + len 2276 * but really needs to advance by 2277 * nfs_dirent + tlen 2278 */ 2279 cp = uiop->uio_iov->iov_base; 2280 tlen -= len; 2281 *cp = '\0'; /* null terminate */ 2282 uiop->uio_iov->iov_base += tlen; 2283 uiop->uio_iov->iov_len -= tlen; 2284 uiop->uio_offset += tlen; 2285 uiop->uio_resid -= tlen; 2286 } else { 2287 /* 2288 * NFS strings must be rounded up (nfsm_myouio 2289 * handled that in the bigenough case). 2290 */ 2291 nfsm_adv(nfsm_rndup(len)); 2292 } 2293 if (v3) { 2294 nfsm_dissect(tl, u_int32_t *, 2295 3 * NFSX_UNSIGNED); 2296 } else { 2297 nfsm_dissect(tl, u_int32_t *, 2298 2 * NFSX_UNSIGNED); 2299 } 2300 2301 /* 2302 * If we were able to accomodate the last entry, 2303 * get the cookie for the next one. Otherwise 2304 * hold-over the cookie for the one we were not 2305 * able to accomodate. 2306 */ 2307 if (bigenough) { 2308 cookie.nfsuquad[0] = *tl++; 2309 if (v3) 2310 cookie.nfsuquad[1] = *tl++; 2311 } else if (v3) { 2312 tl += 2; 2313 } else { 2314 tl++; 2315 } 2316 more_dirs = fxdr_unsigned(int, *tl); 2317 } 2318 /* 2319 * If at end of rpc data, get the eof boolean 2320 */ 2321 if (!more_dirs) { 2322 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2323 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2324 } 2325 m_freem(mrep); 2326 } 2327 /* 2328 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2329 * by increasing d_reclen for the last record. 2330 */ 2331 if (blksiz > 0) { 2332 left = DIRBLKSIZ - blksiz; 2333 dp->nfs_reclen += left; 2334 uiop->uio_iov->iov_base += left; 2335 uiop->uio_iov->iov_len -= left; 2336 uiop->uio_offset += left; 2337 uiop->uio_resid -= left; 2338 } 2339 2340 if (bigenough) { 2341 /* 2342 * We hit the end of the directory, update direofoffset. 2343 */ 2344 dnp->n_direofoffset = uiop->uio_offset; 2345 } else { 2346 /* 2347 * There is more to go, insert the link cookie so the 2348 * next block can be read. 2349 */ 2350 if (uiop->uio_resid > 0) 2351 kprintf("EEK! readdirrpc resid > 0\n"); 2352 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2353 *cookiep = cookie; 2354 } 2355 nfsmout: 2356 return (error); 2357 } 2358 2359 /* 2360 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2361 */ 2362 int 2363 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2364 { 2365 int len, left; 2366 struct nfs_dirent *dp; 2367 u_int32_t *tl; 2368 caddr_t cp; 2369 int32_t t1, t2; 2370 struct vnode *newvp; 2371 nfsuint64 *cookiep; 2372 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2373 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2374 nfsuint64 cookie; 2375 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2376 struct nfsnode *dnp = VTONFS(vp), *np; 2377 nfsfh_t *fhp; 2378 u_quad_t fileno; 2379 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2380 int attrflag, fhsize; 2381 struct nchandle nch; 2382 struct nchandle dnch; 2383 struct nlcomponent nlc; 2384 2385 #ifndef nolint 2386 dp = NULL; 2387 #endif 2388 #ifndef DIAGNOSTIC 2389 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2390 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2391 panic("nfs readdirplusrpc bad uio"); 2392 #endif 2393 /* 2394 * Obtain the namecache record for the directory so we have something 2395 * to use as a basis for creating the entries. This function will 2396 * return a held (but not locked) ncp. The ncp may be disconnected 2397 * from the tree and cannot be used for upward traversals, and the 2398 * ncp may be unnamed. Note that other unrelated operations may 2399 * cause the ncp to be named at any time. 2400 */ 2401 cache_fromdvp(vp, NULL, 0, &dnch); 2402 bzero(&nlc, sizeof(nlc)); 2403 newvp = NULLVP; 2404 2405 /* 2406 * If there is no cookie, assume directory was stale. 2407 */ 2408 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2409 if (cookiep) 2410 cookie = *cookiep; 2411 else 2412 return (NFSERR_BAD_COOKIE); 2413 /* 2414 * Loop around doing readdir rpc's of size nm_readdirsize 2415 * truncated to a multiple of DIRBLKSIZ. 2416 * The stopping criteria is EOF or buffer full. 2417 */ 2418 while (more_dirs && bigenough) { 2419 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2420 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2421 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2422 nfsm_fhtom(vp, 1); 2423 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2424 *tl++ = cookie.nfsuquad[0]; 2425 *tl++ = cookie.nfsuquad[1]; 2426 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2427 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2428 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2429 *tl = txdr_unsigned(nmp->nm_rsize); 2430 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2431 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2432 if (error) { 2433 m_freem(mrep); 2434 goto nfsmout; 2435 } 2436 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2437 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2438 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2439 more_dirs = fxdr_unsigned(int, *tl); 2440 2441 /* loop thru the dir entries, doctoring them to 4bsd form */ 2442 while (more_dirs && bigenough) { 2443 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2444 fileno = fxdr_hyper(tl); 2445 len = fxdr_unsigned(int, *(tl + 2)); 2446 if (len <= 0 || len > NFS_MAXNAMLEN) { 2447 error = EBADRPC; 2448 m_freem(mrep); 2449 goto nfsmout; 2450 } 2451 tlen = nfsm_rndup(len); 2452 if (tlen == len) 2453 tlen += 4; /* To ensure null termination*/ 2454 left = DIRBLKSIZ - blksiz; 2455 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2456 dp->nfs_reclen += left; 2457 uiop->uio_iov->iov_base += left; 2458 uiop->uio_iov->iov_len -= left; 2459 uiop->uio_offset += left; 2460 uiop->uio_resid -= left; 2461 blksiz = 0; 2462 } 2463 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2464 bigenough = 0; 2465 if (bigenough) { 2466 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2467 dp->nfs_ino = fileno; 2468 dp->nfs_namlen = len; 2469 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2470 dp->nfs_type = DT_UNKNOWN; 2471 blksiz += dp->nfs_reclen; 2472 if (blksiz == DIRBLKSIZ) 2473 blksiz = 0; 2474 uiop->uio_offset += sizeof(struct nfs_dirent); 2475 uiop->uio_resid -= sizeof(struct nfs_dirent); 2476 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2477 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2478 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2479 nlc.nlc_namelen = len; 2480 nfsm_mtouio(uiop, len); 2481 cp = uiop->uio_iov->iov_base; 2482 tlen -= len; 2483 *cp = '\0'; 2484 uiop->uio_iov->iov_base += tlen; 2485 uiop->uio_iov->iov_len -= tlen; 2486 uiop->uio_offset += tlen; 2487 uiop->uio_resid -= tlen; 2488 } else 2489 nfsm_adv(nfsm_rndup(len)); 2490 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2491 if (bigenough) { 2492 cookie.nfsuquad[0] = *tl++; 2493 cookie.nfsuquad[1] = *tl++; 2494 } else 2495 tl += 2; 2496 2497 /* 2498 * Since the attributes are before the file handle 2499 * (sigh), we must skip over the attributes and then 2500 * come back and get them. 2501 */ 2502 attrflag = fxdr_unsigned(int, *tl); 2503 if (attrflag) { 2504 dpossav1 = dpos; 2505 mdsav1 = md; 2506 nfsm_adv(NFSX_V3FATTR); 2507 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2508 doit = fxdr_unsigned(int, *tl); 2509 if (doit) { 2510 nfsm_getfh(fhp, fhsize, 1); 2511 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2512 vref(vp); 2513 newvp = vp; 2514 np = dnp; 2515 } else { 2516 error = nfs_nget(vp->v_mount, fhp, 2517 fhsize, &np); 2518 if (error) 2519 doit = 0; 2520 else 2521 newvp = NFSTOV(np); 2522 } 2523 } 2524 if (doit && bigenough) { 2525 dpossav2 = dpos; 2526 dpos = dpossav1; 2527 mdsav2 = md; 2528 md = mdsav1; 2529 nfsm_loadattr(newvp, (struct vattr *)0); 2530 dpos = dpossav2; 2531 md = mdsav2; 2532 dp->nfs_type = 2533 IFTODT(VTTOIF(np->n_vattr.va_type)); 2534 if (dnch.ncp) { 2535 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n", 2536 nlc.nlc_namelen, nlc.nlc_namelen, 2537 nlc.nlc_nameptr); 2538 nch = cache_nlookup(&dnch, &nlc); 2539 cache_setunresolved(&nch); 2540 nfs_cache_setvp(&nch, newvp, 2541 nfspos_cache_timeout); 2542 cache_put(&nch); 2543 } else { 2544 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER" 2545 " %*.*s\n", 2546 nlc.nlc_namelen, nlc.nlc_namelen, 2547 nlc.nlc_nameptr); 2548 } 2549 } 2550 } else { 2551 /* Just skip over the file handle */ 2552 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2553 i = fxdr_unsigned(int, *tl); 2554 nfsm_adv(nfsm_rndup(i)); 2555 } 2556 if (newvp != NULLVP) { 2557 if (newvp == vp) 2558 vrele(newvp); 2559 else 2560 vput(newvp); 2561 newvp = NULLVP; 2562 } 2563 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2564 more_dirs = fxdr_unsigned(int, *tl); 2565 } 2566 /* 2567 * If at end of rpc data, get the eof boolean 2568 */ 2569 if (!more_dirs) { 2570 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2571 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2572 } 2573 m_freem(mrep); 2574 } 2575 /* 2576 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2577 * by increasing d_reclen for the last record. 2578 */ 2579 if (blksiz > 0) { 2580 left = DIRBLKSIZ - blksiz; 2581 dp->nfs_reclen += left; 2582 uiop->uio_iov->iov_base += left; 2583 uiop->uio_iov->iov_len -= left; 2584 uiop->uio_offset += left; 2585 uiop->uio_resid -= left; 2586 } 2587 2588 /* 2589 * We are now either at the end of the directory or have filled the 2590 * block. 2591 */ 2592 if (bigenough) 2593 dnp->n_direofoffset = uiop->uio_offset; 2594 else { 2595 if (uiop->uio_resid > 0) 2596 kprintf("EEK! readdirplusrpc resid > 0\n"); 2597 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2598 *cookiep = cookie; 2599 } 2600 nfsmout: 2601 if (newvp != NULLVP) { 2602 if (newvp == vp) 2603 vrele(newvp); 2604 else 2605 vput(newvp); 2606 newvp = NULLVP; 2607 } 2608 if (dnch.ncp) 2609 cache_drop(&dnch); 2610 return (error); 2611 } 2612 2613 /* 2614 * Silly rename. To make the NFS filesystem that is stateless look a little 2615 * more like the "ufs" a remove of an active vnode is translated to a rename 2616 * to a funny looking filename that is removed by nfs_inactive on the 2617 * nfsnode. There is the potential for another process on a different client 2618 * to create the same funny name between the nfs_lookitup() fails and the 2619 * nfs_rename() completes, but... 2620 */ 2621 static int 2622 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2623 { 2624 struct sillyrename *sp; 2625 struct nfsnode *np; 2626 int error; 2627 2628 /* 2629 * We previously purged dvp instead of vp. I don't know why, it 2630 * completely destroys performance. We can't do it anyway with the 2631 * new VFS API since we would be breaking the namecache topology. 2632 */ 2633 cache_purge(vp); /* XXX */ 2634 np = VTONFS(vp); 2635 #ifndef DIAGNOSTIC 2636 if (vp->v_type == VDIR) 2637 panic("nfs: sillyrename dir"); 2638 #endif 2639 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2640 M_NFSREQ, M_WAITOK); 2641 sp->s_cred = crdup(cnp->cn_cred); 2642 sp->s_dvp = dvp; 2643 vref(dvp); 2644 2645 /* Fudge together a funny name */ 2646 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2647 2648 /* Try lookitups until we get one that isn't there */ 2649 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2650 cnp->cn_td, (struct nfsnode **)0) == 0) { 2651 sp->s_name[4]++; 2652 if (sp->s_name[4] > 'z') { 2653 error = EINVAL; 2654 goto bad; 2655 } 2656 } 2657 error = nfs_renameit(dvp, cnp, sp); 2658 if (error) 2659 goto bad; 2660 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2661 cnp->cn_td, &np); 2662 np->n_sillyrename = sp; 2663 return (0); 2664 bad: 2665 vrele(sp->s_dvp); 2666 crfree(sp->s_cred); 2667 kfree((caddr_t)sp, M_NFSREQ); 2668 return (error); 2669 } 2670 2671 /* 2672 * Look up a file name and optionally either update the file handle or 2673 * allocate an nfsnode, depending on the value of npp. 2674 * npp == NULL --> just do the lookup 2675 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2676 * handled too 2677 * *npp != NULL --> update the file handle in the vnode 2678 */ 2679 static int 2680 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2681 struct thread *td, struct nfsnode **npp) 2682 { 2683 u_int32_t *tl; 2684 caddr_t cp; 2685 int32_t t1, t2; 2686 struct vnode *newvp = (struct vnode *)0; 2687 struct nfsnode *np, *dnp = VTONFS(dvp); 2688 caddr_t bpos, dpos, cp2; 2689 int error = 0, fhlen, attrflag; 2690 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2691 nfsfh_t *nfhp; 2692 int v3 = NFS_ISV3(dvp); 2693 2694 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2695 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2696 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2697 nfsm_fhtom(dvp, v3); 2698 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2699 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2700 if (npp && !error) { 2701 nfsm_getfh(nfhp, fhlen, v3); 2702 if (*npp) { 2703 np = *npp; 2704 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2705 kfree((caddr_t)np->n_fhp, M_NFSBIGFH); 2706 np->n_fhp = &np->n_fh; 2707 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2708 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK); 2709 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2710 np->n_fhsize = fhlen; 2711 newvp = NFSTOV(np); 2712 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2713 vref(dvp); 2714 newvp = dvp; 2715 } else { 2716 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2717 if (error) { 2718 m_freem(mrep); 2719 return (error); 2720 } 2721 newvp = NFSTOV(np); 2722 } 2723 if (v3) { 2724 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2725 if (!attrflag && *npp == NULL) { 2726 m_freem(mrep); 2727 if (newvp == dvp) 2728 vrele(newvp); 2729 else 2730 vput(newvp); 2731 return (ENOENT); 2732 } 2733 } else 2734 nfsm_loadattr(newvp, (struct vattr *)0); 2735 } 2736 m_freem(mrep); 2737 nfsmout: 2738 if (npp && *npp == NULL) { 2739 if (error) { 2740 if (newvp) { 2741 if (newvp == dvp) 2742 vrele(newvp); 2743 else 2744 vput(newvp); 2745 } 2746 } else 2747 *npp = np; 2748 } 2749 return (error); 2750 } 2751 2752 /* 2753 * Nfs Version 3 commit rpc 2754 */ 2755 int 2756 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2757 { 2758 caddr_t cp; 2759 u_int32_t *tl; 2760 int32_t t1, t2; 2761 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2762 caddr_t bpos, dpos, cp2; 2763 int error = 0, wccflag = NFSV3_WCCRATTR; 2764 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2765 2766 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2767 return (0); 2768 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2769 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2770 nfsm_fhtom(vp, 1); 2771 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2772 txdr_hyper(offset, tl); 2773 tl += 2; 2774 *tl = txdr_unsigned(cnt); 2775 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2776 nfsm_wcc_data(vp, wccflag); 2777 if (!error) { 2778 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2779 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2780 NFSX_V3WRITEVERF)) { 2781 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2782 NFSX_V3WRITEVERF); 2783 error = NFSERR_STALEWRITEVERF; 2784 } 2785 } 2786 m_freem(mrep); 2787 nfsmout: 2788 return (error); 2789 } 2790 2791 /* 2792 * Kludge City.. 2793 * - make nfs_bmap() essentially a no-op that does no translation 2794 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2795 * (Maybe I could use the process's page mapping, but I was concerned that 2796 * Kernel Write might not be enabled and also figured copyout() would do 2797 * a lot more work than bcopy() and also it currently happens in the 2798 * context of the swapper process (2). 2799 * 2800 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, 2801 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2802 */ 2803 static int 2804 nfs_bmap(struct vop_bmap_args *ap) 2805 { 2806 if (ap->a_doffsetp != NULL) 2807 *ap->a_doffsetp = ap->a_loffset; 2808 if (ap->a_runp != NULL) 2809 *ap->a_runp = 0; 2810 if (ap->a_runb != NULL) 2811 *ap->a_runb = 0; 2812 return (0); 2813 } 2814 2815 /* 2816 * Strategy routine. 2817 * 2818 * For async requests when nfsiod(s) are running, queue the request by 2819 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2820 * request. 2821 */ 2822 static int 2823 nfs_strategy(struct vop_strategy_args *ap) 2824 { 2825 struct bio *bio = ap->a_bio; 2826 struct bio *nbio; 2827 struct buf *bp = bio->bio_buf; 2828 struct thread *td; 2829 int error = 0; 2830 2831 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2832 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2833 KASSERT(BUF_REFCNT(bp) > 0, 2834 ("nfs_strategy: buffer %p not locked", bp)); 2835 2836 if (bp->b_flags & B_ASYNC) 2837 td = NULL; 2838 else 2839 td = curthread; /* XXX */ 2840 2841 /* 2842 * We probably don't need to push an nbio any more since no 2843 * block conversion is required due to the use of 64 bit byte 2844 * offsets, but do it anyway. 2845 */ 2846 nbio = push_bio(bio); 2847 nbio->bio_offset = bio->bio_offset; 2848 2849 /* 2850 * If the op is asynchronous and an i/o daemon is waiting 2851 * queue the request, wake it up and wait for completion 2852 * otherwise just do it ourselves. 2853 */ 2854 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td)) 2855 error = nfs_doio(ap->a_vp, nbio, td); 2856 return (error); 2857 } 2858 2859 /* 2860 * Mmap a file 2861 * 2862 * NB Currently unsupported. 2863 * 2864 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred) 2865 */ 2866 /* ARGSUSED */ 2867 static int 2868 nfs_mmap(struct vop_mmap_args *ap) 2869 { 2870 return (EINVAL); 2871 } 2872 2873 /* 2874 * fsync vnode op. Just call nfs_flush() with commit == 1. 2875 * 2876 * nfs_fsync(struct vnode *a_vp, int a_waitfor) 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 %lld 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 */ 3229 static int 3230 nfsspec_access(struct vop_access_args *ap) 3231 { 3232 struct vattr *vap; 3233 gid_t *gp; 3234 struct ucred *cred = ap->a_cred; 3235 struct vnode *vp = ap->a_vp; 3236 mode_t mode = ap->a_mode; 3237 struct vattr vattr; 3238 int i; 3239 int error; 3240 3241 /* 3242 * Disallow write attempts on filesystems mounted read-only; 3243 * unless the file is a socket, fifo, or a block or character 3244 * device resident on the filesystem. 3245 */ 3246 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3247 switch (vp->v_type) { 3248 case VREG: 3249 case VDIR: 3250 case VLNK: 3251 return (EROFS); 3252 default: 3253 break; 3254 } 3255 } 3256 /* 3257 * If you're the super-user, 3258 * you always get access. 3259 */ 3260 if (cred->cr_uid == 0) 3261 return (0); 3262 vap = &vattr; 3263 error = VOP_GETATTR(vp, vap); 3264 if (error) 3265 return (error); 3266 /* 3267 * Access check is based on only one of owner, group, public. 3268 * If not owner, then check group. If not a member of the 3269 * group, then check public access. 3270 */ 3271 if (cred->cr_uid != vap->va_uid) { 3272 mode >>= 3; 3273 gp = cred->cr_groups; 3274 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3275 if (vap->va_gid == *gp) 3276 goto found; 3277 mode >>= 3; 3278 found: 3279 ; 3280 } 3281 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3282 return (error); 3283 } 3284 3285 /* 3286 * Read wrapper for special devices. 3287 * 3288 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3289 * struct ucred *a_cred) 3290 */ 3291 static int 3292 nfsspec_read(struct vop_read_args *ap) 3293 { 3294 struct nfsnode *np = VTONFS(ap->a_vp); 3295 3296 /* 3297 * Set access flag. 3298 */ 3299 np->n_flag |= NACC; 3300 getnanotime(&np->n_atim); 3301 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3302 } 3303 3304 /* 3305 * Write wrapper for special devices. 3306 * 3307 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3308 * struct ucred *a_cred) 3309 */ 3310 static int 3311 nfsspec_write(struct vop_write_args *ap) 3312 { 3313 struct nfsnode *np = VTONFS(ap->a_vp); 3314 3315 /* 3316 * Set update flag. 3317 */ 3318 np->n_flag |= NUPD; 3319 getnanotime(&np->n_mtim); 3320 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3321 } 3322 3323 /* 3324 * Close wrapper for special devices. 3325 * 3326 * Update the times on the nfsnode then do device close. 3327 * 3328 * nfsspec_close(struct vnode *a_vp, int a_fflag) 3329 */ 3330 static int 3331 nfsspec_close(struct vop_close_args *ap) 3332 { 3333 struct vnode *vp = ap->a_vp; 3334 struct nfsnode *np = VTONFS(vp); 3335 struct vattr vattr; 3336 3337 if (np->n_flag & (NACC | NUPD)) { 3338 np->n_flag |= NCHG; 3339 if (vp->v_sysref.refcnt == 1 && 3340 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3341 VATTR_NULL(&vattr); 3342 if (np->n_flag & NACC) 3343 vattr.va_atime = np->n_atim; 3344 if (np->n_flag & NUPD) 3345 vattr.va_mtime = np->n_mtim; 3346 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3347 } 3348 } 3349 return (VOCALL(&spec_vnode_vops, &ap->a_head)); 3350 } 3351 3352 /* 3353 * Read wrapper for fifos. 3354 * 3355 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3356 * struct ucred *a_cred) 3357 */ 3358 static int 3359 nfsfifo_read(struct vop_read_args *ap) 3360 { 3361 struct nfsnode *np = VTONFS(ap->a_vp); 3362 3363 /* 3364 * Set access flag. 3365 */ 3366 np->n_flag |= NACC; 3367 getnanotime(&np->n_atim); 3368 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3369 } 3370 3371 /* 3372 * Write wrapper for fifos. 3373 * 3374 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3375 * struct ucred *a_cred) 3376 */ 3377 static int 3378 nfsfifo_write(struct vop_write_args *ap) 3379 { 3380 struct nfsnode *np = VTONFS(ap->a_vp); 3381 3382 /* 3383 * Set update flag. 3384 */ 3385 np->n_flag |= NUPD; 3386 getnanotime(&np->n_mtim); 3387 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3388 } 3389 3390 /* 3391 * Close wrapper for fifos. 3392 * 3393 * Update the times on the nfsnode then do fifo close. 3394 * 3395 * nfsfifo_close(struct vnode *a_vp, int a_fflag) 3396 */ 3397 static int 3398 nfsfifo_close(struct vop_close_args *ap) 3399 { 3400 struct vnode *vp = ap->a_vp; 3401 struct nfsnode *np = VTONFS(vp); 3402 struct vattr vattr; 3403 struct timespec ts; 3404 3405 if (np->n_flag & (NACC | NUPD)) { 3406 getnanotime(&ts); 3407 if (np->n_flag & NACC) 3408 np->n_atim = ts; 3409 if (np->n_flag & NUPD) 3410 np->n_mtim = ts; 3411 np->n_flag |= NCHG; 3412 if (vp->v_sysref.refcnt == 1 && 3413 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3414 VATTR_NULL(&vattr); 3415 if (np->n_flag & NACC) 3416 vattr.va_atime = np->n_atim; 3417 if (np->n_flag & NUPD) 3418 vattr.va_mtime = np->n_mtim; 3419 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3420 } 3421 } 3422 return (VOCALL(&fifo_vnode_vops, &ap->a_head)); 3423 } 3424 3425