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 1.15 2003/11/15 21:05:44 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/socket.h> 59 #include <sys/vnode.h> 60 #include <sys/dirent.h> 61 #include <sys/fcntl.h> 62 #include <sys/lockf.h> 63 #include <sys/stat.h> 64 #include <sys/sysctl.h> 65 #include <sys/conf.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_extern.h> 69 #include <vm/vm_zone.h> 70 71 #include <sys/buf2.h> 72 73 #include <vfs/fifofs/fifo.h> 74 75 #include "rpcv2.h" 76 #include "nfsproto.h" 77 #include "nfs.h" 78 #include "nfsmount.h" 79 #include "nfsnode.h" 80 #include "xdr_subs.h" 81 #include "nfsm_subs.h" 82 #include "nqnfs.h" 83 84 #include <net/if.h> 85 #include <netinet/in.h> 86 #include <netinet/in_var.h> 87 88 /* Defs */ 89 #define TRUE 1 90 #define FALSE 0 91 92 /* 93 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 94 * calls are not in getblk() and brelse() so that they would not be necessary 95 * here. 96 */ 97 #ifndef B_VMIO 98 #define vfs_busy_pages(bp, f) 99 #endif 100 101 static int nfsspec_read (struct vop_read_args *); 102 static int nfsspec_write (struct vop_write_args *); 103 static int nfsfifo_read (struct vop_read_args *); 104 static int nfsfifo_write (struct vop_write_args *); 105 static int nfsspec_close (struct vop_close_args *); 106 static int nfsfifo_close (struct vop_close_args *); 107 #define nfs_poll vop_nopoll 108 static int nfs_flush (struct vnode *,int,struct thread *,int); 109 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 110 static int nfs_lookup (struct vop_lookup_args *); 111 static int nfs_create (struct vop_create_args *); 112 static int nfs_mknod (struct vop_mknod_args *); 113 static int nfs_open (struct vop_open_args *); 114 static int nfs_close (struct vop_close_args *); 115 static int nfs_access (struct vop_access_args *); 116 static int nfs_getattr (struct vop_getattr_args *); 117 static int nfs_setattr (struct vop_setattr_args *); 118 static int nfs_read (struct vop_read_args *); 119 static int nfs_mmap (struct vop_mmap_args *); 120 static int nfs_fsync (struct vop_fsync_args *); 121 static int nfs_remove (struct vop_remove_args *); 122 static int nfs_link (struct vop_link_args *); 123 static int nfs_rename (struct vop_rename_args *); 124 static int nfs_mkdir (struct vop_mkdir_args *); 125 static int nfs_rmdir (struct vop_rmdir_args *); 126 static int nfs_symlink (struct vop_symlink_args *); 127 static int nfs_readdir (struct vop_readdir_args *); 128 static int nfs_bmap (struct vop_bmap_args *); 129 static int nfs_strategy (struct vop_strategy_args *); 130 static int nfs_lookitup (struct vnode *, const char *, int, 131 struct ucred *, struct thread *, struct nfsnode **); 132 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 133 static int nfsspec_access (struct vop_access_args *); 134 static int nfs_readlink (struct vop_readlink_args *); 135 static int nfs_print (struct vop_print_args *); 136 static int nfs_advlock (struct vop_advlock_args *); 137 static int nfs_bwrite (struct vop_bwrite_args *); 138 /* 139 * Global vfs data structures for nfs 140 */ 141 vop_t **nfsv2_vnodeop_p; 142 static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 143 { &vop_default_desc, (vop_t *) vop_defaultop }, 144 { &vop_access_desc, (vop_t *) nfs_access }, 145 { &vop_advlock_desc, (vop_t *) nfs_advlock }, 146 { &vop_bmap_desc, (vop_t *) nfs_bmap }, 147 { &vop_bwrite_desc, (vop_t *) nfs_bwrite }, 148 { &vop_close_desc, (vop_t *) nfs_close }, 149 { &vop_create_desc, (vop_t *) nfs_create }, 150 { &vop_fsync_desc, (vop_t *) nfs_fsync }, 151 { &vop_getattr_desc, (vop_t *) nfs_getattr }, 152 { &vop_getpages_desc, (vop_t *) nfs_getpages }, 153 { &vop_putpages_desc, (vop_t *) nfs_putpages }, 154 { &vop_inactive_desc, (vop_t *) nfs_inactive }, 155 { &vop_islocked_desc, (vop_t *) vop_stdislocked }, 156 { &vop_lease_desc, (vop_t *) vop_null }, 157 { &vop_link_desc, (vop_t *) nfs_link }, 158 { &vop_lock_desc, (vop_t *) vop_sharedlock }, 159 { &vop_lookup_desc, (vop_t *) nfs_lookup }, 160 { &vop_mkdir_desc, (vop_t *) nfs_mkdir }, 161 { &vop_mknod_desc, (vop_t *) nfs_mknod }, 162 { &vop_mmap_desc, (vop_t *) nfs_mmap }, 163 { &vop_open_desc, (vop_t *) nfs_open }, 164 { &vop_poll_desc, (vop_t *) nfs_poll }, 165 { &vop_print_desc, (vop_t *) nfs_print }, 166 { &vop_read_desc, (vop_t *) nfs_read }, 167 { &vop_readdir_desc, (vop_t *) nfs_readdir }, 168 { &vop_readlink_desc, (vop_t *) nfs_readlink }, 169 { &vop_reclaim_desc, (vop_t *) nfs_reclaim }, 170 { &vop_remove_desc, (vop_t *) nfs_remove }, 171 { &vop_rename_desc, (vop_t *) nfs_rename }, 172 { &vop_rmdir_desc, (vop_t *) nfs_rmdir }, 173 { &vop_setattr_desc, (vop_t *) nfs_setattr }, 174 { &vop_strategy_desc, (vop_t *) nfs_strategy }, 175 { &vop_symlink_desc, (vop_t *) nfs_symlink }, 176 { &vop_unlock_desc, (vop_t *) vop_stdunlock }, 177 { &vop_write_desc, (vop_t *) nfs_write }, 178 { NULL, NULL } 179 }; 180 static struct vnodeopv_desc nfsv2_vnodeop_opv_desc = 181 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries }; 182 VNODEOP_SET(nfsv2_vnodeop_opv_desc); 183 184 /* 185 * Special device vnode ops 186 */ 187 vop_t **spec_nfsv2nodeop_p; 188 static struct vnodeopv_entry_desc nfsv2_specop_entries[] = { 189 { &vop_default_desc, (vop_t *) spec_vnoperate }, 190 { &vop_access_desc, (vop_t *) nfsspec_access }, 191 { &vop_close_desc, (vop_t *) nfsspec_close }, 192 { &vop_fsync_desc, (vop_t *) nfs_fsync }, 193 { &vop_getattr_desc, (vop_t *) nfs_getattr }, 194 { &vop_inactive_desc, (vop_t *) nfs_inactive }, 195 { &vop_islocked_desc, (vop_t *) vop_stdislocked }, 196 { &vop_lock_desc, (vop_t *) vop_sharedlock }, 197 { &vop_print_desc, (vop_t *) nfs_print }, 198 { &vop_read_desc, (vop_t *) nfsspec_read }, 199 { &vop_reclaim_desc, (vop_t *) nfs_reclaim }, 200 { &vop_setattr_desc, (vop_t *) nfs_setattr }, 201 { &vop_unlock_desc, (vop_t *) vop_stdunlock }, 202 { &vop_write_desc, (vop_t *) nfsspec_write }, 203 { NULL, NULL } 204 }; 205 static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc = 206 { &spec_nfsv2nodeop_p, nfsv2_specop_entries }; 207 VNODEOP_SET(spec_nfsv2nodeop_opv_desc); 208 209 vop_t **fifo_nfsv2nodeop_p; 210 static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = { 211 { &vop_default_desc, (vop_t *) fifo_vnoperate }, 212 { &vop_access_desc, (vop_t *) nfsspec_access }, 213 { &vop_close_desc, (vop_t *) nfsfifo_close }, 214 { &vop_fsync_desc, (vop_t *) nfs_fsync }, 215 { &vop_getattr_desc, (vop_t *) nfs_getattr }, 216 { &vop_inactive_desc, (vop_t *) nfs_inactive }, 217 { &vop_islocked_desc, (vop_t *) vop_stdislocked }, 218 { &vop_lock_desc, (vop_t *) vop_sharedlock }, 219 { &vop_print_desc, (vop_t *) nfs_print }, 220 { &vop_read_desc, (vop_t *) nfsfifo_read }, 221 { &vop_reclaim_desc, (vop_t *) nfs_reclaim }, 222 { &vop_setattr_desc, (vop_t *) nfs_setattr }, 223 { &vop_unlock_desc, (vop_t *) vop_stdunlock }, 224 { &vop_write_desc, (vop_t *) nfsfifo_write }, 225 { NULL, NULL } 226 }; 227 static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc = 228 { &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries }; 229 VNODEOP_SET(fifo_nfsv2nodeop_opv_desc); 230 231 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 232 struct componentname *cnp, 233 struct vattr *vap); 234 static int nfs_removerpc (struct vnode *dvp, const char *name, 235 int namelen, 236 struct ucred *cred, struct thread *td); 237 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 238 int fnamelen, struct vnode *tdvp, 239 const char *tnameptr, int tnamelen, 240 struct ucred *cred, struct thread *td); 241 static int nfs_renameit (struct vnode *sdvp, 242 struct componentname *scnp, 243 struct sillyrename *sp); 244 245 /* 246 * Global variables 247 */ 248 extern u_int32_t nfs_true, nfs_false; 249 extern u_int32_t nfs_xdrneg1; 250 extern struct nfsstats nfsstats; 251 extern nfstype nfsv3_type[9]; 252 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 253 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 254 int nfs_numasync = 0; 255 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 256 257 SYSCTL_DECL(_vfs_nfs); 258 259 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 261 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 262 263 static int nfsv3_commit_on_close = 0; 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 265 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 266 #if 0 267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 268 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 269 270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 271 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 272 #endif 273 274 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 275 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 276 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 277 static int 278 nfs3_access_otw(struct vnode *vp, int wmode, 279 struct thread *td, struct ucred *cred) 280 { 281 const int v3 = 1; 282 u_int32_t *tl; 283 int error = 0, attrflag; 284 285 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 286 caddr_t bpos, dpos, cp2; 287 int32_t t1, t2; 288 caddr_t cp; 289 u_int32_t rmode; 290 struct nfsnode *np = VTONFS(vp); 291 292 nfsstats.rpccnt[NFSPROC_ACCESS]++; 293 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 294 nfsm_fhtom(vp, v3); 295 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 296 *tl = txdr_unsigned(wmode); 297 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 298 nfsm_postop_attr(vp, attrflag); 299 if (!error) { 300 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 301 rmode = fxdr_unsigned(u_int32_t, *tl); 302 np->n_mode = rmode; 303 np->n_modeuid = cred->cr_uid; 304 np->n_modestamp = time_second; 305 } 306 m_freem(mrep); 307 nfsmout: 308 return error; 309 } 310 311 /* 312 * nfs access vnode op. 313 * For nfs version 2, just return ok. File accesses may fail later. 314 * For nfs version 3, use the access rpc to check accessibility. If file modes 315 * are changed on the server, accesses might still fail later. 316 */ 317 static int 318 nfs_access(ap) 319 struct vop_access_args /* { 320 struct vnode *a_vp; 321 int a_mode; 322 struct ucred *a_cred; 323 struct thread *a_td; 324 } */ *ap; 325 { 326 struct vnode *vp = ap->a_vp; 327 int error = 0; 328 u_int32_t mode, wmode; 329 int v3 = NFS_ISV3(vp); 330 struct nfsnode *np = VTONFS(vp); 331 332 /* 333 * Disallow write attempts on filesystems mounted read-only; 334 * unless the file is a socket, fifo, or a block or character 335 * device resident on the filesystem. 336 */ 337 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 338 switch (vp->v_type) { 339 case VREG: 340 case VDIR: 341 case VLNK: 342 return (EROFS); 343 default: 344 break; 345 } 346 } 347 /* 348 * For nfs v3, check to see if we have done this recently, and if 349 * so return our cached result instead of making an ACCESS call. 350 * If not, do an access rpc, otherwise you are stuck emulating 351 * ufs_access() locally using the vattr. This may not be correct, 352 * since the server may apply other access criteria such as 353 * client uid-->server uid mapping that we do not know about. 354 */ 355 if (v3) { 356 if (ap->a_mode & VREAD) 357 mode = NFSV3ACCESS_READ; 358 else 359 mode = 0; 360 if (vp->v_type != VDIR) { 361 if (ap->a_mode & VWRITE) 362 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 363 if (ap->a_mode & VEXEC) 364 mode |= NFSV3ACCESS_EXECUTE; 365 } else { 366 if (ap->a_mode & VWRITE) 367 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 368 NFSV3ACCESS_DELETE); 369 if (ap->a_mode & VEXEC) 370 mode |= NFSV3ACCESS_LOOKUP; 371 } 372 /* XXX safety belt, only make blanket request if caching */ 373 if (nfsaccess_cache_timeout > 0) { 374 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 375 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 376 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 377 } else { 378 wmode = mode; 379 } 380 381 /* 382 * Does our cached result allow us to give a definite yes to 383 * this request? 384 */ 385 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) && 386 (ap->a_cred->cr_uid == np->n_modeuid) && 387 ((np->n_mode & mode) == mode)) { 388 nfsstats.accesscache_hits++; 389 } else { 390 /* 391 * Either a no, or a don't know. Go to the wire. 392 */ 393 nfsstats.accesscache_misses++; 394 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 395 if (!error) { 396 if ((np->n_mode & mode) != mode) { 397 error = EACCES; 398 } 399 } 400 } 401 } else { 402 if ((error = nfsspec_access(ap)) != 0) 403 return (error); 404 405 /* 406 * Attempt to prevent a mapped root from accessing a file 407 * which it shouldn't. We try to read a byte from the file 408 * if the user is root and the file is not zero length. 409 * After calling nfsspec_access, we should have the correct 410 * file size cached. 411 */ 412 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 413 && VTONFS(vp)->n_size > 0) { 414 struct iovec aiov; 415 struct uio auio; 416 char buf[1]; 417 418 aiov.iov_base = buf; 419 aiov.iov_len = 1; 420 auio.uio_iov = &aiov; 421 auio.uio_iovcnt = 1; 422 auio.uio_offset = 0; 423 auio.uio_resid = 1; 424 auio.uio_segflg = UIO_SYSSPACE; 425 auio.uio_rw = UIO_READ; 426 auio.uio_td = ap->a_td; 427 428 if (vp->v_type == VREG) { 429 error = nfs_readrpc(vp, &auio); 430 } else if (vp->v_type == VDIR) { 431 char* bp; 432 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 433 aiov.iov_base = bp; 434 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 435 error = nfs_readdirrpc(vp, &auio); 436 free(bp, M_TEMP); 437 } else if (vp->v_type == VLNK) { 438 error = nfs_readlinkrpc(vp, &auio); 439 } else { 440 error = EACCES; 441 } 442 } 443 } 444 /* 445 * [re]record creds for reading and/or writing if access 446 * was granted. 447 */ 448 if (error == 0) { 449 if ((ap->a_mode & VREAD) && ap->a_cred != np->n_rucred) { 450 crhold(ap->a_cred); 451 if (np->n_rucred) 452 crfree(np->n_rucred); 453 np->n_rucred = ap->a_cred; 454 } 455 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 456 crhold(ap->a_cred); 457 if (np->n_wucred) 458 crfree(np->n_wucred); 459 np->n_wucred = ap->a_cred; 460 } 461 } 462 return(error); 463 } 464 465 /* 466 * nfs open vnode op 467 * Check to see if the type is ok 468 * and that deletion is not in progress. 469 * For paged in text files, you will need to flush the page cache 470 * if consistency is lost. 471 */ 472 /* ARGSUSED */ 473 static int 474 nfs_open(ap) 475 struct vop_open_args /* { 476 struct vnode *a_vp; 477 int a_mode; 478 struct ucred *a_cred; 479 struct thread *a_td; 480 } */ *ap; 481 { 482 struct vnode *vp = ap->a_vp; 483 struct nfsnode *np = VTONFS(vp); 484 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 485 struct vattr vattr; 486 int error; 487 488 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 489 #ifdef DIAGNOSTIC 490 printf("open eacces vtyp=%d\n",vp->v_type); 491 #endif 492 return (EACCES); 493 } 494 /* 495 * Get a valid lease. If cached data is stale, flush it. 496 */ 497 if (nmp->nm_flag & NFSMNT_NQNFS) { 498 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 499 do { 500 error = nqnfs_getlease(vp, ND_READ, ap->a_td); 501 } while (error == NQNFS_EXPIRED); 502 if (error) 503 return (error); 504 if (np->n_lrev != np->n_brev || 505 (np->n_flag & NQNFSNONCACHE)) { 506 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 507 == EINTR) { 508 return (error); 509 } 510 np->n_brev = np->n_lrev; 511 } 512 } 513 } else { 514 if (np->n_flag & NMODIFIED) { 515 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 516 == EINTR) { 517 return (error); 518 } 519 np->n_attrstamp = 0; 520 if (vp->v_type == VDIR) 521 np->n_direofoffset = 0; 522 error = VOP_GETATTR(vp, &vattr, ap->a_td); 523 if (error) 524 return (error); 525 np->n_mtime = vattr.va_mtime.tv_sec; 526 } else { 527 error = VOP_GETATTR(vp, &vattr, ap->a_td); 528 if (error) 529 return (error); 530 if (np->n_mtime != vattr.va_mtime.tv_sec) { 531 if (vp->v_type == VDIR) 532 np->n_direofoffset = 0; 533 if ((error = nfs_vinvalbuf(vp, V_SAVE, 534 ap->a_td, 1)) == EINTR) { 535 return (error); 536 } 537 np->n_mtime = vattr.va_mtime.tv_sec; 538 } 539 } 540 } 541 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) 542 np->n_attrstamp = 0; /* For Open/Close consistency */ 543 return (0); 544 } 545 546 /* 547 * nfs close vnode op 548 * What an NFS client should do upon close after writing is a debatable issue. 549 * Most NFS clients push delayed writes to the server upon close, basically for 550 * two reasons: 551 * 1 - So that any write errors may be reported back to the client process 552 * doing the close system call. By far the two most likely errors are 553 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 554 * 2 - To put a worst case upper bound on cache inconsistency between 555 * multiple clients for the file. 556 * There is also a consistency problem for Version 2 of the protocol w.r.t. 557 * not being able to tell if other clients are writing a file concurrently, 558 * since there is no way of knowing if the changed modify time in the reply 559 * is only due to the write for this client. 560 * (NFS Version 3 provides weak cache consistency data in the reply that 561 * should be sufficient to detect and handle this case.) 562 * 563 * The current code does the following: 564 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 565 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 566 * or commit them (this satisfies 1 and 2 except for the 567 * case where the server crashes after this close but 568 * before the commit RPC, which is felt to be "good 569 * enough". Changing the last argument to nfs_flush() to 570 * a 1 would force a commit operation, if it is felt a 571 * commit is necessary now. 572 * for NQNFS - do nothing now, since 2 is dealt with via leases and 573 * 1 should be dealt with via an fsync() system call for 574 * cases where write errors are important. 575 */ 576 /* ARGSUSED */ 577 static int 578 nfs_close(ap) 579 struct vop_close_args /* { 580 struct vnodeop_desc *a_desc; 581 struct vnode *a_vp; 582 int a_fflag; 583 struct ucred *a_cred; 584 struct thread *a_td; 585 } */ *ap; 586 { 587 struct vnode *vp = ap->a_vp; 588 struct nfsnode *np = VTONFS(vp); 589 int error = 0; 590 591 if (vp->v_type == VREG) { 592 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 593 (np->n_flag & NMODIFIED)) { 594 if (NFS_ISV3(vp)) { 595 /* 596 * Under NFSv3 we have dirty buffers to dispose of. We 597 * must flush them to the NFS server. We have the option 598 * of waiting all the way through the commit rpc or just 599 * waiting for the initial write. The default is to only 600 * wait through the initial write so the data is in the 601 * server's cache, which is roughly similar to the state 602 * a standard disk subsystem leaves the file in on close(). 603 * 604 * We cannot clear the NMODIFIED bit in np->n_flag due to 605 * potential races with other processes, and certainly 606 * cannot clear it if we don't commit. 607 */ 608 int cm = nfsv3_commit_on_close ? 1 : 0; 609 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm); 610 /* np->n_flag &= ~NMODIFIED; */ 611 } else { 612 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 613 } 614 np->n_attrstamp = 0; 615 } 616 if (np->n_flag & NWRITEERR) { 617 np->n_flag &= ~NWRITEERR; 618 error = np->n_error; 619 } 620 } 621 return (error); 622 } 623 624 /* 625 * nfs getattr call from vfs. 626 */ 627 static int 628 nfs_getattr(ap) 629 struct vop_getattr_args /* { 630 struct vnode *a_vp; 631 struct vattr *a_vap; 632 struct ucred *a_cred; 633 struct thread *a_td; 634 } */ *ap; 635 { 636 struct vnode *vp = ap->a_vp; 637 struct nfsnode *np = VTONFS(vp); 638 caddr_t cp; 639 u_int32_t *tl; 640 int32_t t1, t2; 641 caddr_t bpos, dpos; 642 int error = 0; 643 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 644 int v3 = NFS_ISV3(vp); 645 646 /* 647 * Update local times for special files. 648 */ 649 if (np->n_flag & (NACC | NUPD)) 650 np->n_flag |= NCHG; 651 /* 652 * First look in the cache. 653 */ 654 if (nfs_getattrcache(vp, ap->a_vap) == 0) 655 return (0); 656 657 if (v3 && nfsaccess_cache_timeout > 0) { 658 nfsstats.accesscache_misses++; 659 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 660 if (nfs_getattrcache(vp, ap->a_vap) == 0) 661 return (0); 662 } 663 664 nfsstats.rpccnt[NFSPROC_GETATTR]++; 665 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 666 nfsm_fhtom(vp, v3); 667 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 668 if (!error) { 669 nfsm_loadattr(vp, ap->a_vap); 670 } 671 m_freem(mrep); 672 nfsmout: 673 return (error); 674 } 675 676 /* 677 * nfs setattr call. 678 */ 679 static int 680 nfs_setattr(ap) 681 struct vop_setattr_args /* { 682 struct vnodeop_desc *a_desc; 683 struct vnode *a_vp; 684 struct vattr *a_vap; 685 struct ucred *a_cred; 686 struct thread *a_td; 687 } */ *ap; 688 { 689 struct vnode *vp = ap->a_vp; 690 struct nfsnode *np = VTONFS(vp); 691 struct vattr *vap = ap->a_vap; 692 int error = 0; 693 u_quad_t tsize; 694 695 #ifndef nolint 696 tsize = (u_quad_t)0; 697 #endif 698 699 /* 700 * Setting of flags is not supported. 701 */ 702 if (vap->va_flags != VNOVAL) 703 return (EOPNOTSUPP); 704 705 /* 706 * Disallow write attempts if the filesystem is mounted read-only. 707 */ 708 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 709 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 710 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 711 (vp->v_mount->mnt_flag & MNT_RDONLY)) 712 return (EROFS); 713 if (vap->va_size != VNOVAL) { 714 switch (vp->v_type) { 715 case VDIR: 716 return (EISDIR); 717 case VCHR: 718 case VBLK: 719 case VSOCK: 720 case VFIFO: 721 if (vap->va_mtime.tv_sec == VNOVAL && 722 vap->va_atime.tv_sec == VNOVAL && 723 vap->va_mode == (mode_t)VNOVAL && 724 vap->va_uid == (uid_t)VNOVAL && 725 vap->va_gid == (gid_t)VNOVAL) 726 return (0); 727 vap->va_size = VNOVAL; 728 break; 729 default: 730 /* 731 * Disallow write attempts if the filesystem is 732 * mounted read-only. 733 */ 734 if (vp->v_mount->mnt_flag & MNT_RDONLY) 735 return (EROFS); 736 737 /* 738 * We run vnode_pager_setsize() early (why?), 739 * we must set np->n_size now to avoid vinvalbuf 740 * V_SAVE races that might setsize a lower 741 * value. 742 */ 743 744 tsize = np->n_size; 745 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size); 746 747 if (np->n_flag & NMODIFIED) { 748 if (vap->va_size == 0) 749 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1); 750 else 751 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 752 if (error) { 753 np->n_size = tsize; 754 vnode_pager_setsize(vp, np->n_size); 755 return (error); 756 } 757 } 758 /* np->n_size has already been set to vap->va_size 759 * in nfs_meta_setsize(). We must set it again since 760 * nfs_loadattrcache() could be called through 761 * nfs_meta_setsize() and could modify np->n_size. 762 */ 763 np->n_vattr.va_size = np->n_size = vap->va_size; 764 }; 765 } else if ((vap->va_mtime.tv_sec != VNOVAL || 766 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) && 767 vp->v_type == VREG && 768 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR) 769 return (error); 770 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 771 if (error && vap->va_size != VNOVAL) { 772 np->n_size = np->n_vattr.va_size = tsize; 773 vnode_pager_setsize(vp, np->n_size); 774 } 775 return (error); 776 } 777 778 /* 779 * Do an nfs setattr rpc. 780 */ 781 static int 782 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 783 struct ucred *cred, struct thread *td) 784 { 785 struct nfsv2_sattr *sp; 786 caddr_t cp; 787 int32_t t1, t2; 788 caddr_t bpos, dpos, cp2; 789 u_int32_t *tl; 790 int error = 0, wccflag = NFSV3_WCCRATTR; 791 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 792 int v3 = NFS_ISV3(vp); 793 794 nfsstats.rpccnt[NFSPROC_SETATTR]++; 795 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 796 nfsm_fhtom(vp, v3); 797 if (v3) { 798 nfsm_v3attrbuild(vap, TRUE); 799 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 800 *tl = nfs_false; 801 } else { 802 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 803 if (vap->va_mode == (mode_t)VNOVAL) 804 sp->sa_mode = nfs_xdrneg1; 805 else 806 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 807 if (vap->va_uid == (uid_t)VNOVAL) 808 sp->sa_uid = nfs_xdrneg1; 809 else 810 sp->sa_uid = txdr_unsigned(vap->va_uid); 811 if (vap->va_gid == (gid_t)VNOVAL) 812 sp->sa_gid = nfs_xdrneg1; 813 else 814 sp->sa_gid = txdr_unsigned(vap->va_gid); 815 sp->sa_size = txdr_unsigned(vap->va_size); 816 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 817 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 818 } 819 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 820 if (v3) { 821 nfsm_wcc_data(vp, wccflag); 822 } else 823 nfsm_loadattr(vp, (struct vattr *)0); 824 m_freem(mrep); 825 nfsmout: 826 return (error); 827 } 828 829 /* 830 * nfs lookup call, one step at a time... 831 * First look in cache 832 * If not found, unlock the directory nfsnode and do the rpc 833 */ 834 static int 835 nfs_lookup(ap) 836 struct vop_lookup_args /* { 837 struct vnodeop_desc *a_desc; 838 struct vnode *a_dvp; 839 struct vnode **a_vpp; 840 struct componentname *a_cnp; 841 } */ *ap; 842 { 843 struct componentname *cnp = ap->a_cnp; 844 struct vnode *dvp = ap->a_dvp; 845 struct vnode **vpp = ap->a_vpp; 846 int flags = cnp->cn_flags; 847 struct vnode *newvp; 848 u_int32_t *tl; 849 caddr_t cp; 850 int32_t t1, t2; 851 struct nfsmount *nmp; 852 caddr_t bpos, dpos, cp2; 853 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 854 long len; 855 nfsfh_t *fhp; 856 struct nfsnode *np; 857 int lockparent, wantparent, error = 0, attrflag, fhsize; 858 int v3 = NFS_ISV3(dvp); 859 struct thread *td = cnp->cn_td; 860 861 *vpp = NULLVP; 862 if ((flags & CNP_ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 863 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 864 return (EROFS); 865 if (dvp->v_type != VDIR) 866 return (ENOTDIR); 867 lockparent = flags & CNP_LOCKPARENT; 868 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 869 nmp = VFSTONFS(dvp->v_mount); 870 np = VTONFS(dvp); 871 if ((error = cache_lookup(dvp, NCPNULL, vpp, NCPPNULL, cnp)) && error != ENOENT) { 872 struct vattr vattr; 873 int vpid; 874 875 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { 876 *vpp = NULLVP; 877 return (error); 878 } 879 880 newvp = *vpp; 881 vpid = newvp->v_id; 882 /* 883 * See the comment starting `Step through' in ufs/ufs_lookup.c 884 * for an explanation of the locking protocol 885 */ 886 if (dvp == newvp) { 887 VREF(newvp); 888 error = 0; 889 } else if (flags & CNP_ISDOTDOT) { 890 VOP_UNLOCK(dvp, 0, td); 891 error = vget(newvp, LK_EXCLUSIVE, td); 892 if (!error && lockparent && (flags & CNP_ISLASTCN)) 893 error = vn_lock(dvp, LK_EXCLUSIVE, td); 894 } else { 895 error = vget(newvp, LK_EXCLUSIVE, td); 896 if (!lockparent || error || !(flags & CNP_ISLASTCN)) 897 VOP_UNLOCK(dvp, 0, td); 898 } 899 if (!error) { 900 if (vpid == newvp->v_id) { 901 if (!VOP_GETATTR(newvp, &vattr, td) 902 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 903 nfsstats.lookupcache_hits++; 904 if (cnp->cn_nameiop != NAMEI_LOOKUP && 905 (flags & CNP_ISLASTCN)) 906 cnp->cn_flags |= CNP_SAVENAME; 907 return (0); 908 } 909 cache_purge(newvp); 910 } 911 vput(newvp); 912 if (lockparent && dvp != newvp && (flags & CNP_ISLASTCN)) 913 VOP_UNLOCK(dvp, 0, td); 914 } 915 error = vn_lock(dvp, LK_EXCLUSIVE, td); 916 *vpp = NULLVP; 917 if (error) 918 return (error); 919 } 920 error = 0; 921 newvp = NULLVP; 922 nfsstats.lookupcache_misses++; 923 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 924 len = cnp->cn_namelen; 925 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 926 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 927 nfsm_fhtom(dvp, v3); 928 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 929 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 930 if (error) { 931 nfsm_postop_attr(dvp, attrflag); 932 m_freem(mrep); 933 goto nfsmout; 934 } 935 nfsm_getfh(fhp, fhsize, v3); 936 937 /* 938 * Handle RENAME case... 939 */ 940 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent && (flags & CNP_ISLASTCN)) { 941 if (NFS_CMPFH(np, fhp, fhsize)) { 942 m_freem(mrep); 943 return (EISDIR); 944 } 945 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 946 if (error) { 947 m_freem(mrep); 948 return (error); 949 } 950 newvp = NFSTOV(np); 951 if (v3) { 952 nfsm_postop_attr(newvp, attrflag); 953 nfsm_postop_attr(dvp, attrflag); 954 } else 955 nfsm_loadattr(newvp, (struct vattr *)0); 956 *vpp = newvp; 957 m_freem(mrep); 958 cnp->cn_flags |= CNP_SAVENAME; 959 if (!lockparent) 960 VOP_UNLOCK(dvp, 0, td); 961 return (0); 962 } 963 964 if (flags & CNP_ISDOTDOT) { 965 VOP_UNLOCK(dvp, 0, td); 966 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 967 if (error) { 968 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 969 return (error); 970 } 971 newvp = NFSTOV(np); 972 if (lockparent && (flags & CNP_ISLASTCN) && 973 (error = vn_lock(dvp, LK_EXCLUSIVE, td))) { 974 vput(newvp); 975 return (error); 976 } 977 } else if (NFS_CMPFH(np, fhp, fhsize)) { 978 VREF(dvp); 979 newvp = dvp; 980 } else { 981 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 982 if (error) { 983 m_freem(mrep); 984 return (error); 985 } 986 if (!lockparent || !(flags & CNP_ISLASTCN)) 987 VOP_UNLOCK(dvp, 0, td); 988 newvp = NFSTOV(np); 989 } 990 if (v3) { 991 nfsm_postop_attr(newvp, attrflag); 992 nfsm_postop_attr(dvp, attrflag); 993 } else 994 nfsm_loadattr(newvp, (struct vattr *)0); 995 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN)) 996 cnp->cn_flags |= CNP_SAVENAME; 997 if ((cnp->cn_flags & CNP_MAKEENTRY) && 998 (cnp->cn_nameiop != NAMEI_DELETE || !(flags & CNP_ISLASTCN))) { 999 np->n_ctime = np->n_vattr.va_ctime.tv_sec; 1000 cache_enter(dvp, NCPNULL, newvp, cnp); 1001 } 1002 *vpp = newvp; 1003 m_freem(mrep); 1004 nfsmout: 1005 if (error) { 1006 if (newvp != NULLVP) { 1007 vrele(newvp); 1008 *vpp = NULLVP; 1009 } 1010 if ((cnp->cn_nameiop == NAMEI_CREATE || cnp->cn_nameiop == NAMEI_RENAME) && 1011 (flags & CNP_ISLASTCN) && error == ENOENT) { 1012 if (!lockparent) 1013 VOP_UNLOCK(dvp, 0, td); 1014 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1015 error = EROFS; 1016 else 1017 error = EJUSTRETURN; 1018 } 1019 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN)) 1020 cnp->cn_flags |= CNP_SAVENAME; 1021 } 1022 return (error); 1023 } 1024 1025 /* 1026 * nfs read call. 1027 * Just call nfs_bioread() to do the work. 1028 */ 1029 static int 1030 nfs_read(ap) 1031 struct vop_read_args /* { 1032 struct vnode *a_vp; 1033 struct uio *a_uio; 1034 int a_ioflag; 1035 struct ucred *a_cred; 1036 } */ *ap; 1037 { 1038 struct vnode *vp = ap->a_vp; 1039 1040 if (vp->v_type != VREG) 1041 return (EPERM); 1042 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1043 } 1044 1045 /* 1046 * nfs readlink call 1047 */ 1048 static int 1049 nfs_readlink(ap) 1050 struct vop_readlink_args /* { 1051 struct vnode *a_vp; 1052 struct uio *a_uio; 1053 struct ucred *a_cred; 1054 } */ *ap; 1055 { 1056 struct vnode *vp = ap->a_vp; 1057 1058 if (vp->v_type != VLNK) 1059 return (EINVAL); 1060 return (nfs_bioread(vp, ap->a_uio, 0)); 1061 } 1062 1063 /* 1064 * Do a readlink rpc. 1065 * Called by nfs_doio() from below the buffer cache. 1066 */ 1067 int 1068 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1069 { 1070 u_int32_t *tl; 1071 caddr_t cp; 1072 int32_t t1, t2; 1073 caddr_t bpos, dpos, cp2; 1074 int error = 0, len, attrflag; 1075 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1076 int v3 = NFS_ISV3(vp); 1077 1078 nfsstats.rpccnt[NFSPROC_READLINK]++; 1079 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1080 nfsm_fhtom(vp, v3); 1081 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1082 if (v3) 1083 nfsm_postop_attr(vp, attrflag); 1084 if (!error) { 1085 nfsm_strsiz(len, NFS_MAXPATHLEN); 1086 if (len == NFS_MAXPATHLEN) { 1087 struct nfsnode *np = VTONFS(vp); 1088 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1089 len = np->n_size; 1090 } 1091 nfsm_mtouio(uiop, len); 1092 } 1093 m_freem(mrep); 1094 nfsmout: 1095 return (error); 1096 } 1097 1098 /* 1099 * nfs read rpc call 1100 * Ditto above 1101 */ 1102 int 1103 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1104 { 1105 u_int32_t *tl; 1106 caddr_t cp; 1107 int32_t t1, t2; 1108 caddr_t bpos, dpos, cp2; 1109 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1110 struct nfsmount *nmp; 1111 int error = 0, len, retlen, tsiz, eof, attrflag; 1112 int v3 = NFS_ISV3(vp); 1113 1114 #ifndef nolint 1115 eof = 0; 1116 #endif 1117 nmp = VFSTONFS(vp->v_mount); 1118 tsiz = uiop->uio_resid; 1119 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1120 return (EFBIG); 1121 while (tsiz > 0) { 1122 nfsstats.rpccnt[NFSPROC_READ]++; 1123 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1124 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1125 nfsm_fhtom(vp, v3); 1126 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1127 if (v3) { 1128 txdr_hyper(uiop->uio_offset, tl); 1129 *(tl + 2) = txdr_unsigned(len); 1130 } else { 1131 *tl++ = txdr_unsigned(uiop->uio_offset); 1132 *tl++ = txdr_unsigned(len); 1133 *tl = 0; 1134 } 1135 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1136 if (v3) { 1137 nfsm_postop_attr(vp, attrflag); 1138 if (error) { 1139 m_freem(mrep); 1140 goto nfsmout; 1141 } 1142 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1143 eof = fxdr_unsigned(int, *(tl + 1)); 1144 } else 1145 nfsm_loadattr(vp, (struct vattr *)0); 1146 nfsm_strsiz(retlen, nmp->nm_rsize); 1147 nfsm_mtouio(uiop, retlen); 1148 m_freem(mrep); 1149 tsiz -= retlen; 1150 if (v3) { 1151 if (eof || retlen == 0) { 1152 tsiz = 0; 1153 } 1154 } else if (retlen < len) { 1155 tsiz = 0; 1156 } 1157 } 1158 nfsmout: 1159 return (error); 1160 } 1161 1162 /* 1163 * nfs write call 1164 */ 1165 int 1166 nfs_writerpc(vp, uiop, iomode, must_commit) 1167 struct vnode *vp; 1168 struct uio *uiop; 1169 int *iomode, *must_commit; 1170 { 1171 u_int32_t *tl; 1172 caddr_t cp; 1173 int32_t t1, t2, backup; 1174 caddr_t bpos, dpos, cp2; 1175 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1176 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1177 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1178 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1179 1180 #ifndef DIAGNOSTIC 1181 if (uiop->uio_iovcnt != 1) 1182 panic("nfs: writerpc iovcnt > 1"); 1183 #endif 1184 *must_commit = 0; 1185 tsiz = uiop->uio_resid; 1186 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1187 return (EFBIG); 1188 while (tsiz > 0) { 1189 nfsstats.rpccnt[NFSPROC_WRITE]++; 1190 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1191 nfsm_reqhead(vp, NFSPROC_WRITE, 1192 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1193 nfsm_fhtom(vp, v3); 1194 if (v3) { 1195 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1196 txdr_hyper(uiop->uio_offset, tl); 1197 tl += 2; 1198 *tl++ = txdr_unsigned(len); 1199 *tl++ = txdr_unsigned(*iomode); 1200 *tl = txdr_unsigned(len); 1201 } else { 1202 u_int32_t x; 1203 1204 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1205 /* Set both "begin" and "current" to non-garbage. */ 1206 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1207 *tl++ = x; /* "begin offset" */ 1208 *tl++ = x; /* "current offset" */ 1209 x = txdr_unsigned(len); 1210 *tl++ = x; /* total to this offset */ 1211 *tl = x; /* size of this write */ 1212 } 1213 nfsm_uiotom(uiop, len); 1214 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1215 if (v3) { 1216 wccflag = NFSV3_WCCCHK; 1217 nfsm_wcc_data(vp, wccflag); 1218 if (!error) { 1219 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1220 + NFSX_V3WRITEVERF); 1221 rlen = fxdr_unsigned(int, *tl++); 1222 if (rlen == 0) { 1223 error = NFSERR_IO; 1224 m_freem(mrep); 1225 break; 1226 } else if (rlen < len) { 1227 backup = len - rlen; 1228 uiop->uio_iov->iov_base -= backup; 1229 uiop->uio_iov->iov_len += backup; 1230 uiop->uio_offset -= backup; 1231 uiop->uio_resid += backup; 1232 len = rlen; 1233 } 1234 commit = fxdr_unsigned(int, *tl++); 1235 1236 /* 1237 * Return the lowest committment level 1238 * obtained by any of the RPCs. 1239 */ 1240 if (committed == NFSV3WRITE_FILESYNC) 1241 committed = commit; 1242 else if (committed == NFSV3WRITE_DATASYNC && 1243 commit == NFSV3WRITE_UNSTABLE) 1244 committed = commit; 1245 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1246 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1247 NFSX_V3WRITEVERF); 1248 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1249 } else if (bcmp((caddr_t)tl, 1250 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1251 *must_commit = 1; 1252 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1253 NFSX_V3WRITEVERF); 1254 } 1255 } 1256 } else 1257 nfsm_loadattr(vp, (struct vattr *)0); 1258 if (wccflag) 1259 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec; 1260 m_freem(mrep); 1261 if (error) 1262 break; 1263 tsiz -= len; 1264 } 1265 nfsmout: 1266 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1267 committed = NFSV3WRITE_FILESYNC; 1268 *iomode = committed; 1269 if (error) 1270 uiop->uio_resid = tsiz; 1271 return (error); 1272 } 1273 1274 /* 1275 * nfs mknod rpc 1276 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1277 * mode set to specify the file type and the size field for rdev. 1278 */ 1279 static int 1280 nfs_mknodrpc(dvp, vpp, cnp, vap) 1281 struct vnode *dvp; 1282 struct vnode **vpp; 1283 struct componentname *cnp; 1284 struct vattr *vap; 1285 { 1286 struct nfsv2_sattr *sp; 1287 u_int32_t *tl; 1288 caddr_t cp; 1289 int32_t t1, t2; 1290 struct vnode *newvp = (struct vnode *)0; 1291 struct nfsnode *np = (struct nfsnode *)0; 1292 struct vattr vattr; 1293 char *cp2; 1294 caddr_t bpos, dpos; 1295 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1296 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1297 u_int32_t rdev; 1298 int v3 = NFS_ISV3(dvp); 1299 1300 if (vap->va_type == VCHR || vap->va_type == VBLK) 1301 rdev = txdr_unsigned(vap->va_rdev); 1302 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1303 rdev = nfs_xdrneg1; 1304 else { 1305 return (EOPNOTSUPP); 1306 } 1307 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1308 return (error); 1309 } 1310 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1311 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1312 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1313 nfsm_fhtom(dvp, v3); 1314 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1315 if (v3) { 1316 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1317 *tl++ = vtonfsv3_type(vap->va_type); 1318 nfsm_v3attrbuild(vap, FALSE); 1319 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1320 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1321 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1322 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1323 } 1324 } else { 1325 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1326 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1327 sp->sa_uid = nfs_xdrneg1; 1328 sp->sa_gid = nfs_xdrneg1; 1329 sp->sa_size = rdev; 1330 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1331 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1332 } 1333 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1334 if (!error) { 1335 nfsm_mtofh(dvp, newvp, v3, gotvp); 1336 if (!gotvp) { 1337 if (newvp) { 1338 vput(newvp); 1339 newvp = (struct vnode *)0; 1340 } 1341 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1342 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1343 if (!error) 1344 newvp = NFSTOV(np); 1345 } 1346 } 1347 if (v3) 1348 nfsm_wcc_data(dvp, wccflag); 1349 m_freem(mrep); 1350 nfsmout: 1351 if (error) { 1352 if (newvp) 1353 vput(newvp); 1354 } else { 1355 if (cnp->cn_flags & CNP_MAKEENTRY) 1356 cache_enter(dvp, NCPNULL, newvp, cnp); 1357 *vpp = newvp; 1358 } 1359 VTONFS(dvp)->n_flag |= NMODIFIED; 1360 if (!wccflag) 1361 VTONFS(dvp)->n_attrstamp = 0; 1362 return (error); 1363 } 1364 1365 /* 1366 * nfs mknod vop 1367 * just call nfs_mknodrpc() to do the work. 1368 */ 1369 /* ARGSUSED */ 1370 static int 1371 nfs_mknod(ap) 1372 struct vop_mknod_args /* { 1373 struct vnode *a_dvp; 1374 struct vnode **a_vpp; 1375 struct componentname *a_cnp; 1376 struct vattr *a_vap; 1377 } */ *ap; 1378 { 1379 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1380 } 1381 1382 static u_long create_verf; 1383 /* 1384 * nfs file create call 1385 */ 1386 static int 1387 nfs_create(ap) 1388 struct vop_create_args /* { 1389 struct vnode *a_dvp; 1390 struct vnode **a_vpp; 1391 struct componentname *a_cnp; 1392 struct vattr *a_vap; 1393 } */ *ap; 1394 { 1395 struct vnode *dvp = ap->a_dvp; 1396 struct vattr *vap = ap->a_vap; 1397 struct componentname *cnp = ap->a_cnp; 1398 struct nfsv2_sattr *sp; 1399 u_int32_t *tl; 1400 caddr_t cp; 1401 int32_t t1, t2; 1402 struct nfsnode *np = (struct nfsnode *)0; 1403 struct vnode *newvp = (struct vnode *)0; 1404 caddr_t bpos, dpos, cp2; 1405 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1406 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1407 struct vattr vattr; 1408 int v3 = NFS_ISV3(dvp); 1409 1410 /* 1411 * Oops, not for me.. 1412 */ 1413 if (vap->va_type == VSOCK) 1414 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1415 1416 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1417 return (error); 1418 } 1419 if (vap->va_vaflags & VA_EXCLUSIVE) 1420 fmode |= O_EXCL; 1421 again: 1422 nfsstats.rpccnt[NFSPROC_CREATE]++; 1423 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1424 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1425 nfsm_fhtom(dvp, v3); 1426 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1427 if (v3) { 1428 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1429 if (fmode & O_EXCL) { 1430 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1431 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1432 #ifdef INET 1433 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1434 *tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr; 1435 else 1436 #endif 1437 *tl++ = create_verf; 1438 *tl = ++create_verf; 1439 } else { 1440 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1441 nfsm_v3attrbuild(vap, FALSE); 1442 } 1443 } else { 1444 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1445 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1446 sp->sa_uid = nfs_xdrneg1; 1447 sp->sa_gid = nfs_xdrneg1; 1448 sp->sa_size = 0; 1449 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1450 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1451 } 1452 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1453 if (!error) { 1454 nfsm_mtofh(dvp, newvp, v3, gotvp); 1455 if (!gotvp) { 1456 if (newvp) { 1457 vput(newvp); 1458 newvp = (struct vnode *)0; 1459 } 1460 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1461 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1462 if (!error) 1463 newvp = NFSTOV(np); 1464 } 1465 } 1466 if (v3) 1467 nfsm_wcc_data(dvp, wccflag); 1468 m_freem(mrep); 1469 nfsmout: 1470 if (error) { 1471 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1472 fmode &= ~O_EXCL; 1473 goto again; 1474 } 1475 if (newvp) 1476 vput(newvp); 1477 } else if (v3 && (fmode & O_EXCL)) { 1478 /* 1479 * We are normally called with only a partially initialized 1480 * VAP. Since the NFSv3 spec says that server may use the 1481 * file attributes to store the verifier, the spec requires 1482 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1483 * in atime, but we can't really assume that all servers will 1484 * so we ensure that our SETATTR sets both atime and mtime. 1485 */ 1486 if (vap->va_mtime.tv_sec == VNOVAL) 1487 vfs_timestamp(&vap->va_mtime); 1488 if (vap->va_atime.tv_sec == VNOVAL) 1489 vap->va_atime = vap->va_mtime; 1490 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1491 } 1492 if (!error) { 1493 if (cnp->cn_flags & CNP_MAKEENTRY) 1494 cache_enter(dvp, NCPNULL, newvp, cnp); 1495 /* 1496 * The new np may have enough info for access 1497 * checks, make sure rucred and wucred are 1498 * initialized for read and write rpc's. 1499 */ 1500 np = VTONFS(newvp); 1501 if (np->n_rucred == NULL) 1502 np->n_rucred = crhold(cnp->cn_cred); 1503 if (np->n_wucred == NULL) 1504 np->n_wucred = crhold(cnp->cn_cred); 1505 *ap->a_vpp = newvp; 1506 } 1507 VTONFS(dvp)->n_flag |= NMODIFIED; 1508 if (!wccflag) 1509 VTONFS(dvp)->n_attrstamp = 0; 1510 return (error); 1511 } 1512 1513 /* 1514 * nfs file remove call 1515 * To try and make nfs semantics closer to ufs semantics, a file that has 1516 * other processes using the vnode is renamed instead of removed and then 1517 * removed later on the last close. 1518 * - If v_usecount > 1 1519 * If a rename is not already in the works 1520 * call nfs_sillyrename() to set it up 1521 * else 1522 * do the remove rpc 1523 */ 1524 static int 1525 nfs_remove(ap) 1526 struct vop_remove_args /* { 1527 struct vnodeop_desc *a_desc; 1528 struct vnode * a_dvp; 1529 struct vnode * a_vp; 1530 struct componentname * a_cnp; 1531 } */ *ap; 1532 { 1533 struct vnode *vp = ap->a_vp; 1534 struct vnode *dvp = ap->a_dvp; 1535 struct componentname *cnp = ap->a_cnp; 1536 struct nfsnode *np = VTONFS(vp); 1537 int error = 0; 1538 struct vattr vattr; 1539 1540 #ifndef DIAGNOSTIC 1541 if ((cnp->cn_flags & CNP_HASBUF) == 0) 1542 panic("nfs_remove: no name"); 1543 if (vp->v_usecount < 1) 1544 panic("nfs_remove: bad v_usecount"); 1545 #endif 1546 if (vp->v_type == VDIR) 1547 error = EPERM; 1548 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1549 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 && 1550 vattr.va_nlink > 1)) { 1551 /* 1552 * Purge the name cache so that the chance of a lookup for 1553 * the name succeeding while the remove is in progress is 1554 * minimized. Without node locking it can still happen, such 1555 * that an I/O op returns ESTALE, but since you get this if 1556 * another host removes the file.. 1557 */ 1558 cache_purge(vp); 1559 /* 1560 * throw away biocache buffers, mainly to avoid 1561 * unnecessary delayed writes later. 1562 */ 1563 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1); 1564 /* Do the rpc */ 1565 if (error != EINTR) 1566 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1567 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1568 /* 1569 * Kludge City: If the first reply to the remove rpc is lost.. 1570 * the reply to the retransmitted request will be ENOENT 1571 * since the file was in fact removed 1572 * Therefore, we cheat and return success. 1573 */ 1574 if (error == ENOENT) 1575 error = 0; 1576 } else if (!np->n_sillyrename) 1577 error = nfs_sillyrename(dvp, vp, cnp); 1578 np->n_attrstamp = 0; 1579 return (error); 1580 } 1581 1582 /* 1583 * nfs file remove rpc called from nfs_inactive 1584 */ 1585 int 1586 nfs_removeit(struct sillyrename *sp) 1587 { 1588 1589 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1590 sp->s_cred, NULL)); 1591 } 1592 1593 /* 1594 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1595 */ 1596 static int 1597 nfs_removerpc(dvp, name, namelen, cred, td) 1598 struct vnode *dvp; 1599 const char *name; 1600 int namelen; 1601 struct ucred *cred; 1602 struct thread *td; 1603 { 1604 u_int32_t *tl; 1605 caddr_t cp; 1606 int32_t t1, t2; 1607 caddr_t bpos, dpos, cp2; 1608 int error = 0, wccflag = NFSV3_WCCRATTR; 1609 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1610 int v3 = NFS_ISV3(dvp); 1611 1612 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1613 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1614 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1615 nfsm_fhtom(dvp, v3); 1616 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1617 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1618 if (v3) 1619 nfsm_wcc_data(dvp, wccflag); 1620 m_freem(mrep); 1621 nfsmout: 1622 VTONFS(dvp)->n_flag |= NMODIFIED; 1623 if (!wccflag) 1624 VTONFS(dvp)->n_attrstamp = 0; 1625 return (error); 1626 } 1627 1628 /* 1629 * nfs file rename call 1630 */ 1631 static int 1632 nfs_rename(ap) 1633 struct vop_rename_args /* { 1634 struct vnode *a_fdvp; 1635 struct vnode *a_fvp; 1636 struct componentname *a_fcnp; 1637 struct vnode *a_tdvp; 1638 struct vnode *a_tvp; 1639 struct componentname *a_tcnp; 1640 } */ *ap; 1641 { 1642 struct vnode *fvp = ap->a_fvp; 1643 struct vnode *tvp = ap->a_tvp; 1644 struct vnode *fdvp = ap->a_fdvp; 1645 struct vnode *tdvp = ap->a_tdvp; 1646 struct componentname *tcnp = ap->a_tcnp; 1647 struct componentname *fcnp = ap->a_fcnp; 1648 int error; 1649 1650 #ifndef DIAGNOSTIC 1651 if ((tcnp->cn_flags & CNP_HASBUF) == 0 || 1652 (fcnp->cn_flags & CNP_HASBUF) == 0) 1653 panic("nfs_rename: no name"); 1654 #endif 1655 /* Check for cross-device rename */ 1656 if ((fvp->v_mount != tdvp->v_mount) || 1657 (tvp && (fvp->v_mount != tvp->v_mount))) { 1658 error = EXDEV; 1659 goto out; 1660 } 1661 1662 /* 1663 * We have to flush B_DELWRI data prior to renaming 1664 * the file. If we don't, the delayed-write buffers 1665 * can be flushed out later after the file has gone stale 1666 * under NFSV3. NFSV2 does not have this problem because 1667 * ( as far as I can tell ) it flushes dirty buffers more 1668 * often. 1669 */ 1670 1671 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td); 1672 if (tvp) 1673 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td); 1674 1675 /* 1676 * If the tvp exists and is in use, sillyrename it before doing the 1677 * rename of the new file over it. 1678 * XXX Can't sillyrename a directory. 1679 */ 1680 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1681 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1682 vput(tvp); 1683 tvp = NULL; 1684 } 1685 1686 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1687 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1688 tcnp->cn_td); 1689 1690 if (fvp->v_type == VDIR) { 1691 if (tvp != NULL && tvp->v_type == VDIR) 1692 cache_purge(tdvp); 1693 cache_purge(fdvp); 1694 } 1695 1696 out: 1697 if (tdvp == tvp) 1698 vrele(tdvp); 1699 else 1700 vput(tdvp); 1701 if (tvp) 1702 vput(tvp); 1703 vrele(fdvp); 1704 vrele(fvp); 1705 /* 1706 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1707 */ 1708 if (error == ENOENT) 1709 error = 0; 1710 return (error); 1711 } 1712 1713 /* 1714 * nfs file rename rpc called from nfs_remove() above 1715 */ 1716 static int 1717 nfs_renameit(sdvp, scnp, sp) 1718 struct vnode *sdvp; 1719 struct componentname *scnp; 1720 struct sillyrename *sp; 1721 { 1722 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1723 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1724 } 1725 1726 /* 1727 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1728 */ 1729 static int 1730 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, td) 1731 struct vnode *fdvp; 1732 const char *fnameptr; 1733 int fnamelen; 1734 struct vnode *tdvp; 1735 const char *tnameptr; 1736 int tnamelen; 1737 struct ucred *cred; 1738 struct thread *td; 1739 { 1740 u_int32_t *tl; 1741 caddr_t cp; 1742 int32_t t1, t2; 1743 caddr_t bpos, dpos, cp2; 1744 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1745 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1746 int v3 = NFS_ISV3(fdvp); 1747 1748 nfsstats.rpccnt[NFSPROC_RENAME]++; 1749 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1750 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1751 nfsm_rndup(tnamelen)); 1752 nfsm_fhtom(fdvp, v3); 1753 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1754 nfsm_fhtom(tdvp, v3); 1755 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1756 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1757 if (v3) { 1758 nfsm_wcc_data(fdvp, fwccflag); 1759 nfsm_wcc_data(tdvp, twccflag); 1760 } 1761 m_freem(mrep); 1762 nfsmout: 1763 VTONFS(fdvp)->n_flag |= NMODIFIED; 1764 VTONFS(tdvp)->n_flag |= NMODIFIED; 1765 if (!fwccflag) 1766 VTONFS(fdvp)->n_attrstamp = 0; 1767 if (!twccflag) 1768 VTONFS(tdvp)->n_attrstamp = 0; 1769 return (error); 1770 } 1771 1772 /* 1773 * nfs hard link create call 1774 */ 1775 static int 1776 nfs_link(ap) 1777 struct vop_link_args /* { 1778 struct vnode *a_tdvp; 1779 struct vnode *a_vp; 1780 struct componentname *a_cnp; 1781 } */ *ap; 1782 { 1783 struct vnode *vp = ap->a_vp; 1784 struct vnode *tdvp = ap->a_tdvp; 1785 struct componentname *cnp = ap->a_cnp; 1786 u_int32_t *tl; 1787 caddr_t cp; 1788 int32_t t1, t2; 1789 caddr_t bpos, dpos, cp2; 1790 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1791 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1792 int v3; 1793 1794 if (vp->v_mount != tdvp->v_mount) { 1795 return (EXDEV); 1796 } 1797 1798 /* 1799 * Push all writes to the server, so that the attribute cache 1800 * doesn't get "out of sync" with the server. 1801 * XXX There should be a better way! 1802 */ 1803 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td); 1804 1805 v3 = NFS_ISV3(vp); 1806 nfsstats.rpccnt[NFSPROC_LINK]++; 1807 nfsm_reqhead(vp, NFSPROC_LINK, 1808 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1809 nfsm_fhtom(vp, v3); 1810 nfsm_fhtom(tdvp, v3); 1811 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1812 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1813 if (v3) { 1814 nfsm_postop_attr(vp, attrflag); 1815 nfsm_wcc_data(tdvp, wccflag); 1816 } 1817 m_freem(mrep); 1818 nfsmout: 1819 VTONFS(tdvp)->n_flag |= NMODIFIED; 1820 if (!attrflag) 1821 VTONFS(vp)->n_attrstamp = 0; 1822 if (!wccflag) 1823 VTONFS(tdvp)->n_attrstamp = 0; 1824 /* 1825 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1826 */ 1827 if (error == EEXIST) 1828 error = 0; 1829 return (error); 1830 } 1831 1832 /* 1833 * nfs symbolic link create call 1834 */ 1835 static int 1836 nfs_symlink(ap) 1837 struct vop_symlink_args /* { 1838 struct vnode *a_dvp; 1839 struct vnode **a_vpp; 1840 struct componentname *a_cnp; 1841 struct vattr *a_vap; 1842 char *a_target; 1843 } */ *ap; 1844 { 1845 struct vnode *dvp = ap->a_dvp; 1846 struct vattr *vap = ap->a_vap; 1847 struct componentname *cnp = ap->a_cnp; 1848 struct nfsv2_sattr *sp; 1849 u_int32_t *tl; 1850 caddr_t cp; 1851 int32_t t1, t2; 1852 caddr_t bpos, dpos, cp2; 1853 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1854 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1855 struct vnode *newvp = (struct vnode *)0; 1856 int v3 = NFS_ISV3(dvp); 1857 1858 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1859 slen = strlen(ap->a_target); 1860 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1861 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1862 nfsm_fhtom(dvp, v3); 1863 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1864 if (v3) { 1865 nfsm_v3attrbuild(vap, FALSE); 1866 } 1867 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1868 if (!v3) { 1869 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1870 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1871 sp->sa_uid = nfs_xdrneg1; 1872 sp->sa_gid = nfs_xdrneg1; 1873 sp->sa_size = nfs_xdrneg1; 1874 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1875 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1876 } 1877 1878 /* 1879 * Issue the NFS request and get the rpc response. 1880 * 1881 * Only NFSv3 responses returning an error of 0 actually return 1882 * a file handle that can be converted into newvp without having 1883 * to do an extra lookup rpc. 1884 */ 1885 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1886 if (v3) { 1887 if (error == 0) 1888 nfsm_mtofh(dvp, newvp, v3, gotvp); 1889 nfsm_wcc_data(dvp, wccflag); 1890 } 1891 1892 /* 1893 * out code jumps -> here, mrep is also freed. 1894 */ 1895 1896 m_freem(mrep); 1897 nfsmout: 1898 1899 /* 1900 * If we get an EEXIST error, silently convert it to no-error 1901 * in case of an NFS retry. 1902 */ 1903 if (error == EEXIST) 1904 error = 0; 1905 1906 /* 1907 * If we do not have (or no longer have) an error, and we could 1908 * not extract the newvp from the response due to the request being 1909 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1910 * to obtain a newvp to return. 1911 */ 1912 if (error == 0 && newvp == NULL) { 1913 struct nfsnode *np = NULL; 1914 1915 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1916 cnp->cn_cred, cnp->cn_td, &np); 1917 if (!error) 1918 newvp = NFSTOV(np); 1919 } 1920 if (error) { 1921 if (newvp) 1922 vput(newvp); 1923 } else { 1924 *ap->a_vpp = newvp; 1925 } 1926 VTONFS(dvp)->n_flag |= NMODIFIED; 1927 if (!wccflag) 1928 VTONFS(dvp)->n_attrstamp = 0; 1929 return (error); 1930 } 1931 1932 /* 1933 * nfs make dir call 1934 */ 1935 static int 1936 nfs_mkdir(ap) 1937 struct vop_mkdir_args /* { 1938 struct vnode *a_dvp; 1939 struct vnode **a_vpp; 1940 struct componentname *a_cnp; 1941 struct vattr *a_vap; 1942 } */ *ap; 1943 { 1944 struct vnode *dvp = ap->a_dvp; 1945 struct vattr *vap = ap->a_vap; 1946 struct componentname *cnp = ap->a_cnp; 1947 struct nfsv2_sattr *sp; 1948 u_int32_t *tl; 1949 caddr_t cp; 1950 int32_t t1, t2; 1951 int len; 1952 struct nfsnode *np = (struct nfsnode *)0; 1953 struct vnode *newvp = (struct vnode *)0; 1954 caddr_t bpos, dpos, cp2; 1955 int error = 0, wccflag = NFSV3_WCCRATTR; 1956 int gotvp = 0; 1957 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1958 struct vattr vattr; 1959 int v3 = NFS_ISV3(dvp); 1960 1961 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1962 return (error); 1963 } 1964 len = cnp->cn_namelen; 1965 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1966 nfsm_reqhead(dvp, NFSPROC_MKDIR, 1967 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 1968 nfsm_fhtom(dvp, v3); 1969 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1970 if (v3) { 1971 nfsm_v3attrbuild(vap, FALSE); 1972 } else { 1973 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1974 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 1975 sp->sa_uid = nfs_xdrneg1; 1976 sp->sa_gid = nfs_xdrneg1; 1977 sp->sa_size = nfs_xdrneg1; 1978 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1979 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1980 } 1981 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 1982 if (!error) 1983 nfsm_mtofh(dvp, newvp, v3, gotvp); 1984 if (v3) 1985 nfsm_wcc_data(dvp, wccflag); 1986 m_freem(mrep); 1987 nfsmout: 1988 VTONFS(dvp)->n_flag |= NMODIFIED; 1989 if (!wccflag) 1990 VTONFS(dvp)->n_attrstamp = 0; 1991 /* 1992 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 1993 * if we can succeed in looking up the directory. 1994 */ 1995 if (error == EEXIST || (!error && !gotvp)) { 1996 if (newvp) { 1997 vrele(newvp); 1998 newvp = (struct vnode *)0; 1999 } 2000 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2001 cnp->cn_td, &np); 2002 if (!error) { 2003 newvp = NFSTOV(np); 2004 if (newvp->v_type != VDIR) 2005 error = EEXIST; 2006 } 2007 } 2008 if (error) { 2009 if (newvp) 2010 vrele(newvp); 2011 } else 2012 *ap->a_vpp = newvp; 2013 return (error); 2014 } 2015 2016 /* 2017 * nfs remove directory call 2018 */ 2019 static int 2020 nfs_rmdir(ap) 2021 struct vop_rmdir_args /* { 2022 struct vnode *a_dvp; 2023 struct vnode *a_vp; 2024 struct componentname *a_cnp; 2025 } */ *ap; 2026 { 2027 struct vnode *vp = ap->a_vp; 2028 struct vnode *dvp = ap->a_dvp; 2029 struct componentname *cnp = ap->a_cnp; 2030 u_int32_t *tl; 2031 caddr_t cp; 2032 int32_t t1, t2; 2033 caddr_t bpos, dpos, cp2; 2034 int error = 0, wccflag = NFSV3_WCCRATTR; 2035 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2036 int v3 = NFS_ISV3(dvp); 2037 2038 if (dvp == vp) 2039 return (EINVAL); 2040 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2041 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2042 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2043 nfsm_fhtom(dvp, v3); 2044 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2045 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2046 if (v3) 2047 nfsm_wcc_data(dvp, wccflag); 2048 m_freem(mrep); 2049 nfsmout: 2050 VTONFS(dvp)->n_flag |= NMODIFIED; 2051 if (!wccflag) 2052 VTONFS(dvp)->n_attrstamp = 0; 2053 cache_purge(dvp); 2054 cache_purge(vp); 2055 /* 2056 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2057 */ 2058 if (error == ENOENT) 2059 error = 0; 2060 return (error); 2061 } 2062 2063 /* 2064 * nfs readdir call 2065 */ 2066 static int 2067 nfs_readdir(ap) 2068 struct vop_readdir_args /* { 2069 struct vnode *a_vp; 2070 struct uio *a_uio; 2071 struct ucred *a_cred; 2072 } */ *ap; 2073 { 2074 struct vnode *vp = ap->a_vp; 2075 struct nfsnode *np = VTONFS(vp); 2076 struct uio *uio = ap->a_uio; 2077 int tresid, error; 2078 struct vattr vattr; 2079 2080 if (vp->v_type != VDIR) 2081 return (EPERM); 2082 /* 2083 * First, check for hit on the EOF offset cache 2084 */ 2085 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2086 (np->n_flag & NMODIFIED) == 0) { 2087 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) { 2088 if (NQNFS_CKCACHABLE(vp, ND_READ)) { 2089 nfsstats.direofcache_hits++; 2090 return (0); 2091 } 2092 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 && 2093 np->n_mtime == vattr.va_mtime.tv_sec) { 2094 nfsstats.direofcache_hits++; 2095 return (0); 2096 } 2097 } 2098 2099 /* 2100 * Call nfs_bioread() to do the real work. 2101 */ 2102 tresid = uio->uio_resid; 2103 error = nfs_bioread(vp, uio, 0); 2104 2105 if (!error && uio->uio_resid == tresid) 2106 nfsstats.direofcache_misses++; 2107 return (error); 2108 } 2109 2110 /* 2111 * Readdir rpc call. 2112 * Called from below the buffer cache by nfs_doio(). 2113 */ 2114 int 2115 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2116 { 2117 int len, left; 2118 struct dirent *dp = NULL; 2119 u_int32_t *tl; 2120 caddr_t cp; 2121 int32_t t1, t2; 2122 nfsuint64 *cookiep; 2123 caddr_t bpos, dpos, cp2; 2124 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2125 nfsuint64 cookie; 2126 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2127 struct nfsnode *dnp = VTONFS(vp); 2128 u_quad_t fileno; 2129 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2130 int attrflag; 2131 int v3 = NFS_ISV3(vp); 2132 2133 #ifndef DIAGNOSTIC 2134 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2135 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2136 panic("nfs readdirrpc bad uio"); 2137 #endif 2138 2139 /* 2140 * If there is no cookie, assume directory was stale. 2141 */ 2142 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2143 if (cookiep) 2144 cookie = *cookiep; 2145 else 2146 return (NFSERR_BAD_COOKIE); 2147 /* 2148 * Loop around doing readdir rpc's of size nm_readdirsize 2149 * truncated to a multiple of DIRBLKSIZ. 2150 * The stopping criteria is EOF or buffer full. 2151 */ 2152 while (more_dirs && bigenough) { 2153 nfsstats.rpccnt[NFSPROC_READDIR]++; 2154 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2155 NFSX_READDIR(v3)); 2156 nfsm_fhtom(vp, v3); 2157 if (v3) { 2158 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2159 *tl++ = cookie.nfsuquad[0]; 2160 *tl++ = cookie.nfsuquad[1]; 2161 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2162 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2163 } else { 2164 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2165 *tl++ = cookie.nfsuquad[0]; 2166 } 2167 *tl = txdr_unsigned(nmp->nm_readdirsize); 2168 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2169 if (v3) { 2170 nfsm_postop_attr(vp, attrflag); 2171 if (!error) { 2172 nfsm_dissect(tl, u_int32_t *, 2173 2 * NFSX_UNSIGNED); 2174 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2175 dnp->n_cookieverf.nfsuquad[1] = *tl; 2176 } else { 2177 m_freem(mrep); 2178 goto nfsmout; 2179 } 2180 } 2181 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2182 more_dirs = fxdr_unsigned(int, *tl); 2183 2184 /* loop thru the dir entries, doctoring them to 4bsd form */ 2185 while (more_dirs && bigenough) { 2186 if (v3) { 2187 nfsm_dissect(tl, u_int32_t *, 2188 3 * NFSX_UNSIGNED); 2189 fileno = fxdr_hyper(tl); 2190 len = fxdr_unsigned(int, *(tl + 2)); 2191 } else { 2192 nfsm_dissect(tl, u_int32_t *, 2193 2 * NFSX_UNSIGNED); 2194 fileno = fxdr_unsigned(u_quad_t, *tl++); 2195 len = fxdr_unsigned(int, *tl); 2196 } 2197 if (len <= 0 || len > NFS_MAXNAMLEN) { 2198 error = EBADRPC; 2199 m_freem(mrep); 2200 goto nfsmout; 2201 } 2202 tlen = nfsm_rndup(len); 2203 if (tlen == len) 2204 tlen += 4; /* To ensure null termination */ 2205 left = DIRBLKSIZ - blksiz; 2206 if ((tlen + DIRHDSIZ) > left) { 2207 dp->d_reclen += left; 2208 uiop->uio_iov->iov_base += left; 2209 uiop->uio_iov->iov_len -= left; 2210 uiop->uio_offset += left; 2211 uiop->uio_resid -= left; 2212 blksiz = 0; 2213 } 2214 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2215 bigenough = 0; 2216 if (bigenough) { 2217 dp = (struct dirent *)uiop->uio_iov->iov_base; 2218 dp->d_fileno = (int)fileno; 2219 dp->d_namlen = len; 2220 dp->d_reclen = tlen + DIRHDSIZ; 2221 dp->d_type = DT_UNKNOWN; 2222 blksiz += dp->d_reclen; 2223 if (blksiz == DIRBLKSIZ) 2224 blksiz = 0; 2225 uiop->uio_offset += DIRHDSIZ; 2226 uiop->uio_resid -= DIRHDSIZ; 2227 uiop->uio_iov->iov_base += DIRHDSIZ; 2228 uiop->uio_iov->iov_len -= DIRHDSIZ; 2229 nfsm_mtouio(uiop, len); 2230 cp = uiop->uio_iov->iov_base; 2231 tlen -= len; 2232 *cp = '\0'; /* null terminate */ 2233 uiop->uio_iov->iov_base += tlen; 2234 uiop->uio_iov->iov_len -= tlen; 2235 uiop->uio_offset += tlen; 2236 uiop->uio_resid -= tlen; 2237 } else 2238 nfsm_adv(nfsm_rndup(len)); 2239 if (v3) { 2240 nfsm_dissect(tl, u_int32_t *, 2241 3 * NFSX_UNSIGNED); 2242 } else { 2243 nfsm_dissect(tl, u_int32_t *, 2244 2 * NFSX_UNSIGNED); 2245 } 2246 if (bigenough) { 2247 cookie.nfsuquad[0] = *tl++; 2248 if (v3) 2249 cookie.nfsuquad[1] = *tl++; 2250 } else if (v3) 2251 tl += 2; 2252 else 2253 tl++; 2254 more_dirs = fxdr_unsigned(int, *tl); 2255 } 2256 /* 2257 * If at end of rpc data, get the eof boolean 2258 */ 2259 if (!more_dirs) { 2260 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2261 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2262 } 2263 m_freem(mrep); 2264 } 2265 /* 2266 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2267 * by increasing d_reclen for the last record. 2268 */ 2269 if (blksiz > 0) { 2270 left = DIRBLKSIZ - blksiz; 2271 dp->d_reclen += left; 2272 uiop->uio_iov->iov_base += left; 2273 uiop->uio_iov->iov_len -= left; 2274 uiop->uio_offset += left; 2275 uiop->uio_resid -= left; 2276 } 2277 2278 /* 2279 * We are now either at the end of the directory or have filled the 2280 * block. 2281 */ 2282 if (bigenough) 2283 dnp->n_direofoffset = uiop->uio_offset; 2284 else { 2285 if (uiop->uio_resid > 0) 2286 printf("EEK! readdirrpc resid > 0\n"); 2287 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2288 *cookiep = cookie; 2289 } 2290 nfsmout: 2291 return (error); 2292 } 2293 2294 /* 2295 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2296 */ 2297 int 2298 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2299 { 2300 int len, left; 2301 struct dirent *dp; 2302 u_int32_t *tl; 2303 caddr_t cp; 2304 int32_t t1, t2; 2305 struct vnode *newvp; 2306 nfsuint64 *cookiep; 2307 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2308 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2309 struct nameidata nami, *ndp = &nami; 2310 struct componentname *cnp = &ndp->ni_cnd; 2311 nfsuint64 cookie; 2312 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2313 struct nfsnode *dnp = VTONFS(vp), *np; 2314 nfsfh_t *fhp; 2315 u_quad_t fileno; 2316 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2317 int attrflag, fhsize; 2318 2319 #ifndef nolint 2320 dp = (struct dirent *)0; 2321 #endif 2322 #ifndef DIAGNOSTIC 2323 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2324 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2325 panic("nfs readdirplusrpc bad uio"); 2326 #endif 2327 ndp->ni_dvp = vp; 2328 newvp = NULLVP; 2329 2330 /* 2331 * If there is no cookie, assume directory was stale. 2332 */ 2333 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2334 if (cookiep) 2335 cookie = *cookiep; 2336 else 2337 return (NFSERR_BAD_COOKIE); 2338 /* 2339 * Loop around doing readdir rpc's of size nm_readdirsize 2340 * truncated to a multiple of DIRBLKSIZ. 2341 * The stopping criteria is EOF or buffer full. 2342 */ 2343 while (more_dirs && bigenough) { 2344 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2345 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2346 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2347 nfsm_fhtom(vp, 1); 2348 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2349 *tl++ = cookie.nfsuquad[0]; 2350 *tl++ = cookie.nfsuquad[1]; 2351 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2352 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2353 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2354 *tl = txdr_unsigned(nmp->nm_rsize); 2355 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2356 nfsm_postop_attr(vp, attrflag); 2357 if (error) { 2358 m_freem(mrep); 2359 goto nfsmout; 2360 } 2361 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2362 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2363 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2364 more_dirs = fxdr_unsigned(int, *tl); 2365 2366 /* loop thru the dir entries, doctoring them to 4bsd form */ 2367 while (more_dirs && bigenough) { 2368 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2369 fileno = fxdr_hyper(tl); 2370 len = fxdr_unsigned(int, *(tl + 2)); 2371 if (len <= 0 || len > NFS_MAXNAMLEN) { 2372 error = EBADRPC; 2373 m_freem(mrep); 2374 goto nfsmout; 2375 } 2376 tlen = nfsm_rndup(len); 2377 if (tlen == len) 2378 tlen += 4; /* To ensure null termination*/ 2379 left = DIRBLKSIZ - blksiz; 2380 if ((tlen + DIRHDSIZ) > left) { 2381 dp->d_reclen += left; 2382 uiop->uio_iov->iov_base += left; 2383 uiop->uio_iov->iov_len -= left; 2384 uiop->uio_offset += left; 2385 uiop->uio_resid -= left; 2386 blksiz = 0; 2387 } 2388 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2389 bigenough = 0; 2390 if (bigenough) { 2391 dp = (struct dirent *)uiop->uio_iov->iov_base; 2392 dp->d_fileno = (int)fileno; 2393 dp->d_namlen = len; 2394 dp->d_reclen = tlen + DIRHDSIZ; 2395 dp->d_type = DT_UNKNOWN; 2396 blksiz += dp->d_reclen; 2397 if (blksiz == DIRBLKSIZ) 2398 blksiz = 0; 2399 uiop->uio_offset += DIRHDSIZ; 2400 uiop->uio_resid -= DIRHDSIZ; 2401 uiop->uio_iov->iov_base += DIRHDSIZ; 2402 uiop->uio_iov->iov_len -= DIRHDSIZ; 2403 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2404 cnp->cn_namelen = len; 2405 nfsm_mtouio(uiop, len); 2406 cp = uiop->uio_iov->iov_base; 2407 tlen -= len; 2408 *cp = '\0'; 2409 uiop->uio_iov->iov_base += tlen; 2410 uiop->uio_iov->iov_len -= tlen; 2411 uiop->uio_offset += tlen; 2412 uiop->uio_resid -= tlen; 2413 } else 2414 nfsm_adv(nfsm_rndup(len)); 2415 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2416 if (bigenough) { 2417 cookie.nfsuquad[0] = *tl++; 2418 cookie.nfsuquad[1] = *tl++; 2419 } else 2420 tl += 2; 2421 2422 /* 2423 * Since the attributes are before the file handle 2424 * (sigh), we must skip over the attributes and then 2425 * come back and get them. 2426 */ 2427 attrflag = fxdr_unsigned(int, *tl); 2428 if (attrflag) { 2429 dpossav1 = dpos; 2430 mdsav1 = md; 2431 nfsm_adv(NFSX_V3FATTR); 2432 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2433 doit = fxdr_unsigned(int, *tl); 2434 if (doit) { 2435 nfsm_getfh(fhp, fhsize, 1); 2436 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2437 VREF(vp); 2438 newvp = vp; 2439 np = dnp; 2440 } else { 2441 error = nfs_nget(vp->v_mount, fhp, 2442 fhsize, &np); 2443 if (error) 2444 doit = 0; 2445 else 2446 newvp = NFSTOV(np); 2447 } 2448 } 2449 if (doit && bigenough) { 2450 dpossav2 = dpos; 2451 dpos = dpossav1; 2452 mdsav2 = md; 2453 md = mdsav1; 2454 nfsm_loadattr(newvp, (struct vattr *)0); 2455 dpos = dpossav2; 2456 md = mdsav2; 2457 dp->d_type = 2458 IFTODT(VTTOIF(np->n_vattr.va_type)); 2459 ndp->ni_vp = newvp; 2460 cache_enter(ndp->ni_dvp, NCPNULL, ndp->ni_vp, cnp); 2461 } 2462 } else { 2463 /* Just skip over the file handle */ 2464 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2465 i = fxdr_unsigned(int, *tl); 2466 nfsm_adv(nfsm_rndup(i)); 2467 } 2468 if (newvp != NULLVP) { 2469 if (newvp == vp) 2470 vrele(newvp); 2471 else 2472 vput(newvp); 2473 newvp = NULLVP; 2474 } 2475 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2476 more_dirs = fxdr_unsigned(int, *tl); 2477 } 2478 /* 2479 * If at end of rpc data, get the eof boolean 2480 */ 2481 if (!more_dirs) { 2482 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2483 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2484 } 2485 m_freem(mrep); 2486 } 2487 /* 2488 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2489 * by increasing d_reclen for the last record. 2490 */ 2491 if (blksiz > 0) { 2492 left = DIRBLKSIZ - blksiz; 2493 dp->d_reclen += left; 2494 uiop->uio_iov->iov_base += left; 2495 uiop->uio_iov->iov_len -= left; 2496 uiop->uio_offset += left; 2497 uiop->uio_resid -= left; 2498 } 2499 2500 /* 2501 * We are now either at the end of the directory or have filled the 2502 * block. 2503 */ 2504 if (bigenough) 2505 dnp->n_direofoffset = uiop->uio_offset; 2506 else { 2507 if (uiop->uio_resid > 0) 2508 printf("EEK! readdirplusrpc resid > 0\n"); 2509 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2510 *cookiep = cookie; 2511 } 2512 nfsmout: 2513 if (newvp != NULLVP) { 2514 if (newvp == vp) 2515 vrele(newvp); 2516 else 2517 vput(newvp); 2518 newvp = NULLVP; 2519 } 2520 return (error); 2521 } 2522 2523 /* 2524 * Silly rename. To make the NFS filesystem that is stateless look a little 2525 * more like the "ufs" a remove of an active vnode is translated to a rename 2526 * to a funny looking filename that is removed by nfs_inactive on the 2527 * nfsnode. There is the potential for another process on a different client 2528 * to create the same funny name between the nfs_lookitup() fails and the 2529 * nfs_rename() completes, but... 2530 */ 2531 static int 2532 nfs_sillyrename(dvp, vp, cnp) 2533 struct vnode *dvp, *vp; 2534 struct componentname *cnp; 2535 { 2536 struct sillyrename *sp; 2537 struct nfsnode *np; 2538 int error; 2539 2540 cache_purge(dvp); 2541 np = VTONFS(vp); 2542 #ifndef DIAGNOSTIC 2543 if (vp->v_type == VDIR) 2544 panic("nfs: sillyrename dir"); 2545 #endif 2546 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2547 M_NFSREQ, M_WAITOK); 2548 sp->s_cred = crdup(cnp->cn_cred); 2549 sp->s_dvp = dvp; 2550 VREF(dvp); 2551 2552 /* Fudge together a funny name */ 2553 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2554 2555 /* Try lookitups until we get one that isn't there */ 2556 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2557 cnp->cn_td, (struct nfsnode **)0) == 0) { 2558 sp->s_name[4]++; 2559 if (sp->s_name[4] > 'z') { 2560 error = EINVAL; 2561 goto bad; 2562 } 2563 } 2564 error = nfs_renameit(dvp, cnp, sp); 2565 if (error) 2566 goto bad; 2567 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2568 cnp->cn_td, &np); 2569 np->n_sillyrename = sp; 2570 return (0); 2571 bad: 2572 vrele(sp->s_dvp); 2573 crfree(sp->s_cred); 2574 free((caddr_t)sp, M_NFSREQ); 2575 return (error); 2576 } 2577 2578 /* 2579 * Look up a file name and optionally either update the file handle or 2580 * allocate an nfsnode, depending on the value of npp. 2581 * npp == NULL --> just do the lookup 2582 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2583 * handled too 2584 * *npp != NULL --> update the file handle in the vnode 2585 */ 2586 static int 2587 nfs_lookitup(dvp, name, len, cred, td, npp) 2588 struct vnode *dvp; 2589 const char *name; 2590 int len; 2591 struct ucred *cred; 2592 struct thread *td; 2593 struct nfsnode **npp; 2594 { 2595 u_int32_t *tl; 2596 caddr_t cp; 2597 int32_t t1, t2; 2598 struct vnode *newvp = (struct vnode *)0; 2599 struct nfsnode *np, *dnp = VTONFS(dvp); 2600 caddr_t bpos, dpos, cp2; 2601 int error = 0, fhlen, attrflag; 2602 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2603 nfsfh_t *nfhp; 2604 int v3 = NFS_ISV3(dvp); 2605 2606 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2607 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2608 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2609 nfsm_fhtom(dvp, v3); 2610 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2611 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2612 if (npp && !error) { 2613 nfsm_getfh(nfhp, fhlen, v3); 2614 if (*npp) { 2615 np = *npp; 2616 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2617 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2618 np->n_fhp = &np->n_fh; 2619 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2620 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2621 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2622 np->n_fhsize = fhlen; 2623 newvp = NFSTOV(np); 2624 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2625 VREF(dvp); 2626 newvp = dvp; 2627 } else { 2628 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2629 if (error) { 2630 m_freem(mrep); 2631 return (error); 2632 } 2633 newvp = NFSTOV(np); 2634 } 2635 if (v3) { 2636 nfsm_postop_attr(newvp, attrflag); 2637 if (!attrflag && *npp == NULL) { 2638 m_freem(mrep); 2639 if (newvp == dvp) 2640 vrele(newvp); 2641 else 2642 vput(newvp); 2643 return (ENOENT); 2644 } 2645 } else 2646 nfsm_loadattr(newvp, (struct vattr *)0); 2647 } 2648 m_freem(mrep); 2649 nfsmout: 2650 if (npp && *npp == NULL) { 2651 if (error) { 2652 if (newvp) { 2653 if (newvp == dvp) 2654 vrele(newvp); 2655 else 2656 vput(newvp); 2657 } 2658 } else 2659 *npp = np; 2660 } 2661 return (error); 2662 } 2663 2664 /* 2665 * Nfs Version 3 commit rpc 2666 */ 2667 int 2668 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2669 { 2670 caddr_t cp; 2671 u_int32_t *tl; 2672 int32_t t1, t2; 2673 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2674 caddr_t bpos, dpos, cp2; 2675 int error = 0, wccflag = NFSV3_WCCRATTR; 2676 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2677 2678 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2679 return (0); 2680 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2681 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2682 nfsm_fhtom(vp, 1); 2683 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2684 txdr_hyper(offset, tl); 2685 tl += 2; 2686 *tl = txdr_unsigned(cnt); 2687 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2688 nfsm_wcc_data(vp, wccflag); 2689 if (!error) { 2690 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2691 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2692 NFSX_V3WRITEVERF)) { 2693 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2694 NFSX_V3WRITEVERF); 2695 error = NFSERR_STALEWRITEVERF; 2696 } 2697 } 2698 m_freem(mrep); 2699 nfsmout: 2700 return (error); 2701 } 2702 2703 /* 2704 * Kludge City.. 2705 * - make nfs_bmap() essentially a no-op that does no translation 2706 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2707 * (Maybe I could use the process's page mapping, but I was concerned that 2708 * Kernel Write might not be enabled and also figured copyout() would do 2709 * a lot more work than bcopy() and also it currently happens in the 2710 * context of the swapper process (2). 2711 */ 2712 static int 2713 nfs_bmap(ap) 2714 struct vop_bmap_args /* { 2715 struct vnode *a_vp; 2716 daddr_t a_bn; 2717 struct vnode **a_vpp; 2718 daddr_t *a_bnp; 2719 int *a_runp; 2720 int *a_runb; 2721 } */ *ap; 2722 { 2723 struct vnode *vp = ap->a_vp; 2724 2725 if (ap->a_vpp != NULL) 2726 *ap->a_vpp = vp; 2727 if (ap->a_bnp != NULL) 2728 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize); 2729 if (ap->a_runp != NULL) 2730 *ap->a_runp = 0; 2731 if (ap->a_runb != NULL) 2732 *ap->a_runb = 0; 2733 return (0); 2734 } 2735 2736 /* 2737 * Strategy routine. 2738 * For async requests when nfsiod(s) are running, queue the request by 2739 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2740 * request. 2741 */ 2742 static int 2743 nfs_strategy(ap) 2744 struct vop_strategy_args *ap; 2745 { 2746 struct buf *bp = ap->a_bp; 2747 struct thread *td; 2748 int error = 0; 2749 2750 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2751 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp)); 2752 2753 if (bp->b_flags & B_PHYS) 2754 panic("nfs physio"); 2755 2756 if (bp->b_flags & B_ASYNC) 2757 td = NULL; 2758 else 2759 td = curthread; /* XXX */ 2760 2761 /* 2762 * If the op is asynchronous and an i/o daemon is waiting 2763 * queue the request, wake it up and wait for completion 2764 * otherwise just do it ourselves. 2765 */ 2766 if ((bp->b_flags & B_ASYNC) == 0 || 2767 nfs_asyncio(bp, td)) 2768 error = nfs_doio(bp, td); 2769 return (error); 2770 } 2771 2772 /* 2773 * Mmap a file 2774 * 2775 * NB Currently unsupported. 2776 */ 2777 /* ARGSUSED */ 2778 static int 2779 nfs_mmap(ap) 2780 struct vop_mmap_args /* { 2781 struct vnode *a_vp; 2782 int a_fflags; 2783 struct ucred *a_cred; 2784 struct thread *a_td; 2785 } */ *ap; 2786 { 2787 2788 return (EINVAL); 2789 } 2790 2791 /* 2792 * fsync vnode op. Just call nfs_flush() with commit == 1. 2793 */ 2794 /* ARGSUSED */ 2795 static int 2796 nfs_fsync(ap) 2797 struct vop_fsync_args /* { 2798 struct vnodeop_desc *a_desc; 2799 struct vnode * a_vp; 2800 struct ucred * a_cred; 2801 int a_waitfor; 2802 struct thread * a_td; 2803 } */ *ap; 2804 { 2805 2806 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1)); 2807 } 2808 2809 /* 2810 * Flush all the blocks associated with a vnode. 2811 * Walk through the buffer pool and push any dirty pages 2812 * associated with the vnode. 2813 */ 2814 static int 2815 nfs_flush(vp, waitfor, td, commit) 2816 struct vnode *vp; 2817 int waitfor; 2818 struct thread *td; 2819 int commit; 2820 { 2821 struct nfsnode *np = VTONFS(vp); 2822 struct buf *bp; 2823 int i; 2824 struct buf *nbp; 2825 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2826 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2827 int passone = 1; 2828 u_quad_t off, endoff, toff; 2829 struct buf **bvec = NULL; 2830 #ifndef NFS_COMMITBVECSIZ 2831 #define NFS_COMMITBVECSIZ 20 2832 #endif 2833 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2834 int bvecsize = 0, bveccount; 2835 2836 if (nmp->nm_flag & NFSMNT_INT) 2837 slpflag = PCATCH; 2838 if (!commit) 2839 passone = 0; 2840 /* 2841 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2842 * server, but nas not been committed to stable storage on the server 2843 * yet. On the first pass, the byte range is worked out and the commit 2844 * rpc is done. On the second pass, nfs_writebp() is called to do the 2845 * job. 2846 */ 2847 again: 2848 off = (u_quad_t)-1; 2849 endoff = 0; 2850 bvecpos = 0; 2851 if (NFS_ISV3(vp) && commit) { 2852 s = splbio(); 2853 /* 2854 * Count up how many buffers waiting for a commit. 2855 */ 2856 bveccount = 0; 2857 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2858 nbp = TAILQ_NEXT(bp, b_vnbufs); 2859 if (BUF_REFCNT(bp) == 0 && 2860 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2861 == (B_DELWRI | B_NEEDCOMMIT)) 2862 bveccount++; 2863 } 2864 /* 2865 * Allocate space to remember the list of bufs to commit. It is 2866 * important to use M_NOWAIT here to avoid a race with nfs_write. 2867 * If we can't get memory (for whatever reason), we will end up 2868 * committing the buffers one-by-one in the loop below. 2869 */ 2870 if (bvec != NULL && bvec != bvec_on_stack) 2871 free(bvec, M_TEMP); 2872 if (bveccount > NFS_COMMITBVECSIZ) { 2873 bvec = (struct buf **) 2874 malloc(bveccount * sizeof(struct buf *), 2875 M_TEMP, M_NOWAIT); 2876 if (bvec == NULL) { 2877 bvec = bvec_on_stack; 2878 bvecsize = NFS_COMMITBVECSIZ; 2879 } else 2880 bvecsize = bveccount; 2881 } else { 2882 bvec = bvec_on_stack; 2883 bvecsize = NFS_COMMITBVECSIZ; 2884 } 2885 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2886 nbp = TAILQ_NEXT(bp, b_vnbufs); 2887 if (bvecpos >= bvecsize) 2888 break; 2889 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2890 (B_DELWRI | B_NEEDCOMMIT) || 2891 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 2892 continue; 2893 bremfree(bp); 2894 /* 2895 * NOTE: we are not clearing B_DONE here, so we have 2896 * to do it later on in this routine if we intend to 2897 * initiate I/O on the bp. 2898 * 2899 * Note: to avoid loopback deadlocks, we do not 2900 * assign b_runningbufspace. 2901 */ 2902 bp->b_flags |= B_WRITEINPROG; 2903 vfs_busy_pages(bp, 1); 2904 2905 /* 2906 * bp is protected by being locked, but nbp is not 2907 * and vfs_busy_pages() may sleep. We have to 2908 * recalculate nbp. 2909 */ 2910 nbp = TAILQ_NEXT(bp, b_vnbufs); 2911 2912 /* 2913 * A list of these buffers is kept so that the 2914 * second loop knows which buffers have actually 2915 * been committed. This is necessary, since there 2916 * may be a race between the commit rpc and new 2917 * uncommitted writes on the file. 2918 */ 2919 bvec[bvecpos++] = bp; 2920 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2921 bp->b_dirtyoff; 2922 if (toff < off) 2923 off = toff; 2924 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2925 if (toff > endoff) 2926 endoff = toff; 2927 } 2928 splx(s); 2929 } 2930 if (bvecpos > 0) { 2931 /* 2932 * Commit data on the server, as required. Note that 2933 * nfs_commit will use the vnode's cred for the commit. 2934 */ 2935 retv = nfs_commit(vp, off, (int)(endoff - off), td); 2936 2937 if (retv == NFSERR_STALEWRITEVERF) 2938 nfs_clearcommit(vp->v_mount); 2939 2940 /* 2941 * Now, either mark the blocks I/O done or mark the 2942 * blocks dirty, depending on whether the commit 2943 * succeeded. 2944 */ 2945 for (i = 0; i < bvecpos; i++) { 2946 bp = bvec[i]; 2947 bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK); 2948 if (retv) { 2949 /* 2950 * Error, leave B_DELWRI intact 2951 */ 2952 vfs_unbusy_pages(bp); 2953 brelse(bp); 2954 } else { 2955 /* 2956 * Success, remove B_DELWRI ( bundirty() ). 2957 * 2958 * b_dirtyoff/b_dirtyend seem to be NFS 2959 * specific. We should probably move that 2960 * into bundirty(). XXX 2961 */ 2962 s = splbio(); 2963 vp->v_numoutput++; 2964 bp->b_flags |= B_ASYNC; 2965 bundirty(bp); 2966 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 2967 bp->b_dirtyoff = bp->b_dirtyend = 0; 2968 splx(s); 2969 biodone(bp); 2970 } 2971 } 2972 } 2973 2974 /* 2975 * Start/do any write(s) that are required. 2976 */ 2977 loop: 2978 s = splbio(); 2979 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2980 nbp = TAILQ_NEXT(bp, b_vnbufs); 2981 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 2982 if (waitfor != MNT_WAIT || passone) 2983 continue; 2984 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL, 2985 "nfsfsync", slpflag, slptimeo); 2986 splx(s); 2987 if (error == 0) 2988 panic("nfs_fsync: inconsistent lock"); 2989 if (error == ENOLCK) 2990 goto loop; 2991 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2992 error = EINTR; 2993 goto done; 2994 } 2995 if (slpflag == PCATCH) { 2996 slpflag = 0; 2997 slptimeo = 2 * hz; 2998 } 2999 goto loop; 3000 } 3001 if ((bp->b_flags & B_DELWRI) == 0) 3002 panic("nfs_fsync: not dirty"); 3003 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3004 BUF_UNLOCK(bp); 3005 continue; 3006 } 3007 bremfree(bp); 3008 if (passone || !commit) 3009 bp->b_flags |= B_ASYNC; 3010 else 3011 bp->b_flags |= B_ASYNC | B_WRITEINPROG; 3012 splx(s); 3013 VOP_BWRITE(bp->b_vp, bp); 3014 goto loop; 3015 } 3016 splx(s); 3017 if (passone) { 3018 passone = 0; 3019 goto again; 3020 } 3021 if (waitfor == MNT_WAIT) { 3022 while (vp->v_numoutput) { 3023 vp->v_flag |= VBWAIT; 3024 error = tsleep((caddr_t)&vp->v_numoutput, 3025 slpflag, "nfsfsync", slptimeo); 3026 if (error) { 3027 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 3028 error = EINTR; 3029 goto done; 3030 } 3031 if (slpflag == PCATCH) { 3032 slpflag = 0; 3033 slptimeo = 2 * hz; 3034 } 3035 } 3036 } 3037 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) { 3038 goto loop; 3039 } 3040 } 3041 if (np->n_flag & NWRITEERR) { 3042 error = np->n_error; 3043 np->n_flag &= ~NWRITEERR; 3044 } 3045 done: 3046 if (bvec != NULL && bvec != bvec_on_stack) 3047 free(bvec, M_TEMP); 3048 return (error); 3049 } 3050 3051 /* 3052 * NFS advisory byte-level locks. 3053 * Currently unsupported. 3054 */ 3055 static int 3056 nfs_advlock(ap) 3057 struct vop_advlock_args /* { 3058 struct vnode *a_vp; 3059 caddr_t a_id; 3060 int a_op; 3061 struct flock *a_fl; 3062 int a_flags; 3063 } */ *ap; 3064 { 3065 struct nfsnode *np = VTONFS(ap->a_vp); 3066 3067 /* 3068 * The following kludge is to allow diskless support to work 3069 * until a real NFS lockd is implemented. Basically, just pretend 3070 * that this is a local lock. 3071 */ 3072 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3073 } 3074 3075 /* 3076 * Print out the contents of an nfsnode. 3077 */ 3078 static int 3079 nfs_print(ap) 3080 struct vop_print_args /* { 3081 struct vnode *a_vp; 3082 } */ *ap; 3083 { 3084 struct vnode *vp = ap->a_vp; 3085 struct nfsnode *np = VTONFS(vp); 3086 3087 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3088 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3089 if (vp->v_type == VFIFO) 3090 fifo_printinfo(vp); 3091 printf("\n"); 3092 return (0); 3093 } 3094 3095 /* 3096 * Just call nfs_writebp() with the force argument set to 1. 3097 * 3098 * NOTE: B_DONE may or may not be set in a_bp on call. 3099 */ 3100 static int 3101 nfs_bwrite(ap) 3102 struct vop_bwrite_args /* { 3103 struct vnode *a_bp; 3104 } */ *ap; 3105 { 3106 return (nfs_writebp(ap->a_bp, 1, curthread)); 3107 } 3108 3109 /* 3110 * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless 3111 * the force flag is one and it also handles the B_NEEDCOMMIT flag. We set 3112 * B_CACHE if this is a VMIO buffer. 3113 */ 3114 int 3115 nfs_writebp(bp, force, td) 3116 struct buf *bp; 3117 int force; 3118 struct thread *td; 3119 { 3120 int s; 3121 int oldflags = bp->b_flags; 3122 #if 0 3123 int retv = 1; 3124 off_t off; 3125 #endif 3126 3127 if (BUF_REFCNT(bp) == 0) 3128 panic("bwrite: buffer is not locked???"); 3129 3130 if (bp->b_flags & B_INVAL) { 3131 brelse(bp); 3132 return(0); 3133 } 3134 3135 bp->b_flags |= B_CACHE; 3136 3137 /* 3138 * Undirty the bp. We will redirty it later if the I/O fails. 3139 */ 3140 3141 s = splbio(); 3142 bundirty(bp); 3143 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3144 3145 bp->b_vp->v_numoutput++; 3146 splx(s); 3147 3148 /* 3149 * Note: to avoid loopback deadlocks, we do not 3150 * assign b_runningbufspace. 3151 */ 3152 vfs_busy_pages(bp, 1); 3153 3154 if (force) 3155 bp->b_flags |= B_WRITEINPROG; 3156 BUF_KERNPROC(bp); 3157 VOP_STRATEGY(bp->b_vp, bp); 3158 3159 if( (oldflags & B_ASYNC) == 0) { 3160 int rtval = biowait(bp); 3161 3162 if (oldflags & B_DELWRI) { 3163 s = splbio(); 3164 reassignbuf(bp, bp->b_vp); 3165 splx(s); 3166 } 3167 3168 brelse(bp); 3169 return (rtval); 3170 } 3171 3172 return (0); 3173 } 3174 3175 /* 3176 * nfs special file access vnode op. 3177 * Essentially just get vattr and then imitate iaccess() since the device is 3178 * local to the client. 3179 */ 3180 static int 3181 nfsspec_access(ap) 3182 struct vop_access_args /* { 3183 struct vnode *a_vp; 3184 int a_mode; 3185 struct ucred *a_cred; 3186 struct thread *a_td; 3187 } */ *ap; 3188 { 3189 struct vattr *vap; 3190 gid_t *gp; 3191 struct ucred *cred = ap->a_cred; 3192 struct vnode *vp = ap->a_vp; 3193 mode_t mode = ap->a_mode; 3194 struct vattr vattr; 3195 int i; 3196 int error; 3197 3198 /* 3199 * Disallow write attempts on filesystems mounted read-only; 3200 * unless the file is a socket, fifo, or a block or character 3201 * device resident on the filesystem. 3202 */ 3203 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3204 switch (vp->v_type) { 3205 case VREG: 3206 case VDIR: 3207 case VLNK: 3208 return (EROFS); 3209 default: 3210 break; 3211 } 3212 } 3213 /* 3214 * If you're the super-user, 3215 * you always get access. 3216 */ 3217 if (cred->cr_uid == 0) 3218 return (0); 3219 vap = &vattr; 3220 error = VOP_GETATTR(vp, vap, ap->a_td); 3221 if (error) 3222 return (error); 3223 /* 3224 * Access check is based on only one of owner, group, public. 3225 * If not owner, then check group. If not a member of the 3226 * group, then check public access. 3227 */ 3228 if (cred->cr_uid != vap->va_uid) { 3229 mode >>= 3; 3230 gp = cred->cr_groups; 3231 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3232 if (vap->va_gid == *gp) 3233 goto found; 3234 mode >>= 3; 3235 found: 3236 ; 3237 } 3238 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3239 return (error); 3240 } 3241 3242 /* 3243 * Read wrapper for special devices. 3244 */ 3245 static int 3246 nfsspec_read(ap) 3247 struct vop_read_args /* { 3248 struct vnode *a_vp; 3249 struct uio *a_uio; 3250 int a_ioflag; 3251 struct ucred *a_cred; 3252 } */ *ap; 3253 { 3254 struct nfsnode *np = VTONFS(ap->a_vp); 3255 3256 /* 3257 * Set access flag. 3258 */ 3259 np->n_flag |= NACC; 3260 getnanotime(&np->n_atim); 3261 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3262 } 3263 3264 /* 3265 * Write wrapper for special devices. 3266 */ 3267 static int 3268 nfsspec_write(ap) 3269 struct vop_write_args /* { 3270 struct vnode *a_vp; 3271 struct uio *a_uio; 3272 int a_ioflag; 3273 struct ucred *a_cred; 3274 } */ *ap; 3275 { 3276 struct nfsnode *np = VTONFS(ap->a_vp); 3277 3278 /* 3279 * Set update flag. 3280 */ 3281 np->n_flag |= NUPD; 3282 getnanotime(&np->n_mtim); 3283 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3284 } 3285 3286 /* 3287 * Close wrapper for special devices. 3288 * 3289 * Update the times on the nfsnode then do device close. 3290 */ 3291 static int 3292 nfsspec_close(ap) 3293 struct vop_close_args /* { 3294 struct vnode *a_vp; 3295 int a_fflag; 3296 struct ucred *a_cred; 3297 struct thread *a_td; 3298 } */ *ap; 3299 { 3300 struct vnode *vp = ap->a_vp; 3301 struct nfsnode *np = VTONFS(vp); 3302 struct vattr vattr; 3303 3304 if (np->n_flag & (NACC | NUPD)) { 3305 np->n_flag |= NCHG; 3306 if (vp->v_usecount == 1 && 3307 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3308 VATTR_NULL(&vattr); 3309 if (np->n_flag & NACC) 3310 vattr.va_atime = np->n_atim; 3311 if (np->n_flag & NUPD) 3312 vattr.va_mtime = np->n_mtim; 3313 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3314 } 3315 } 3316 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3317 } 3318 3319 /* 3320 * Read wrapper for fifos. 3321 */ 3322 static int 3323 nfsfifo_read(ap) 3324 struct vop_read_args /* { 3325 struct vnode *a_vp; 3326 struct uio *a_uio; 3327 int a_ioflag; 3328 struct ucred *a_cred; 3329 } */ *ap; 3330 { 3331 struct nfsnode *np = VTONFS(ap->a_vp); 3332 3333 /* 3334 * Set access flag. 3335 */ 3336 np->n_flag |= NACC; 3337 getnanotime(&np->n_atim); 3338 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3339 } 3340 3341 /* 3342 * Write wrapper for fifos. 3343 */ 3344 static int 3345 nfsfifo_write(ap) 3346 struct vop_write_args /* { 3347 struct vnode *a_vp; 3348 struct uio *a_uio; 3349 int a_ioflag; 3350 struct ucred *a_cred; 3351 } */ *ap; 3352 { 3353 struct nfsnode *np = VTONFS(ap->a_vp); 3354 3355 /* 3356 * Set update flag. 3357 */ 3358 np->n_flag |= NUPD; 3359 getnanotime(&np->n_mtim); 3360 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3361 } 3362 3363 /* 3364 * Close wrapper for fifos. 3365 * 3366 * Update the times on the nfsnode then do fifo close. 3367 */ 3368 static int 3369 nfsfifo_close(ap) 3370 struct vop_close_args /* { 3371 struct vnode *a_vp; 3372 int a_fflag; 3373 struct thread *a_td; 3374 } */ *ap; 3375 { 3376 struct vnode *vp = ap->a_vp; 3377 struct nfsnode *np = VTONFS(vp); 3378 struct vattr vattr; 3379 struct timespec ts; 3380 3381 if (np->n_flag & (NACC | NUPD)) { 3382 getnanotime(&ts); 3383 if (np->n_flag & NACC) 3384 np->n_atim = ts; 3385 if (np->n_flag & NUPD) 3386 np->n_mtim = ts; 3387 np->n_flag |= NCHG; 3388 if (vp->v_usecount == 1 && 3389 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3390 VATTR_NULL(&vattr); 3391 if (np->n_flag & NACC) 3392 vattr.va_atime = np->n_atim; 3393 if (np->n_flag & NUPD) 3394 vattr.va_mtime = np->n_mtim; 3395 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3396 } 3397 } 3398 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3399 } 3400 3401