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 1.20 2004/04/08 22:32:14 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 nfsneg_cache_timeout = NFS_MINATTRTIMO; 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 265 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 266 267 static int nfsv3_commit_on_close = 0; 268 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 269 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 270 #if 0 271 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 272 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 273 274 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 275 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 276 #endif 277 278 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 279 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 280 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 281 static int 282 nfs3_access_otw(struct vnode *vp, int wmode, 283 struct thread *td, struct ucred *cred) 284 { 285 const int v3 = 1; 286 u_int32_t *tl; 287 int error = 0, attrflag; 288 289 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 290 caddr_t bpos, dpos, cp2; 291 int32_t t1, t2; 292 caddr_t cp; 293 u_int32_t rmode; 294 struct nfsnode *np = VTONFS(vp); 295 296 nfsstats.rpccnt[NFSPROC_ACCESS]++; 297 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 298 nfsm_fhtom(vp, v3); 299 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 300 *tl = txdr_unsigned(wmode); 301 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 302 nfsm_postop_attr(vp, attrflag); 303 if (!error) { 304 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 305 rmode = fxdr_unsigned(u_int32_t, *tl); 306 np->n_mode = rmode; 307 np->n_modeuid = cred->cr_uid; 308 np->n_modestamp = mycpu->gd_time_seconds; 309 } 310 m_freem(mrep); 311 nfsmout: 312 return error; 313 } 314 315 /* 316 * nfs access vnode op. 317 * For nfs version 2, just return ok. File accesses may fail later. 318 * For nfs version 3, use the access rpc to check accessibility. If file modes 319 * are changed on the server, accesses might still fail later. 320 */ 321 static int 322 nfs_access(ap) 323 struct vop_access_args /* { 324 struct vnode *a_vp; 325 int a_mode; 326 struct ucred *a_cred; 327 struct thread *a_td; 328 } */ *ap; 329 { 330 struct vnode *vp = ap->a_vp; 331 int error = 0; 332 u_int32_t mode, wmode; 333 int v3 = NFS_ISV3(vp); 334 struct nfsnode *np = VTONFS(vp); 335 336 /* 337 * Disallow write attempts on filesystems mounted read-only; 338 * unless the file is a socket, fifo, or a block or character 339 * device resident on the filesystem. 340 */ 341 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 342 switch (vp->v_type) { 343 case VREG: 344 case VDIR: 345 case VLNK: 346 return (EROFS); 347 default: 348 break; 349 } 350 } 351 /* 352 * For nfs v3, check to see if we have done this recently, and if 353 * so return our cached result instead of making an ACCESS call. 354 * If not, do an access rpc, otherwise you are stuck emulating 355 * ufs_access() locally using the vattr. This may not be correct, 356 * since the server may apply other access criteria such as 357 * client uid-->server uid mapping that we do not know about. 358 */ 359 if (v3) { 360 if (ap->a_mode & VREAD) 361 mode = NFSV3ACCESS_READ; 362 else 363 mode = 0; 364 if (vp->v_type != VDIR) { 365 if (ap->a_mode & VWRITE) 366 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 367 if (ap->a_mode & VEXEC) 368 mode |= NFSV3ACCESS_EXECUTE; 369 } else { 370 if (ap->a_mode & VWRITE) 371 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 372 NFSV3ACCESS_DELETE); 373 if (ap->a_mode & VEXEC) 374 mode |= NFSV3ACCESS_LOOKUP; 375 } 376 /* XXX safety belt, only make blanket request if caching */ 377 if (nfsaccess_cache_timeout > 0) { 378 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 379 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 380 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 381 } else { 382 wmode = mode; 383 } 384 385 /* 386 * Does our cached result allow us to give a definite yes to 387 * this request? 388 */ 389 if ((mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 390 (ap->a_cred->cr_uid == np->n_modeuid) && 391 ((np->n_mode & mode) == mode)) { 392 nfsstats.accesscache_hits++; 393 } else { 394 /* 395 * Either a no, or a don't know. Go to the wire. 396 */ 397 nfsstats.accesscache_misses++; 398 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 399 if (!error) { 400 if ((np->n_mode & mode) != mode) { 401 error = EACCES; 402 } 403 } 404 } 405 } else { 406 if ((error = nfsspec_access(ap)) != 0) 407 return (error); 408 409 /* 410 * Attempt to prevent a mapped root from accessing a file 411 * which it shouldn't. We try to read a byte from the file 412 * if the user is root and the file is not zero length. 413 * After calling nfsspec_access, we should have the correct 414 * file size cached. 415 */ 416 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 417 && VTONFS(vp)->n_size > 0) { 418 struct iovec aiov; 419 struct uio auio; 420 char buf[1]; 421 422 aiov.iov_base = buf; 423 aiov.iov_len = 1; 424 auio.uio_iov = &aiov; 425 auio.uio_iovcnt = 1; 426 auio.uio_offset = 0; 427 auio.uio_resid = 1; 428 auio.uio_segflg = UIO_SYSSPACE; 429 auio.uio_rw = UIO_READ; 430 auio.uio_td = ap->a_td; 431 432 if (vp->v_type == VREG) { 433 error = nfs_readrpc(vp, &auio); 434 } else if (vp->v_type == VDIR) { 435 char* bp; 436 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 437 aiov.iov_base = bp; 438 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 439 error = nfs_readdirrpc(vp, &auio); 440 free(bp, M_TEMP); 441 } else if (vp->v_type == VLNK) { 442 error = nfs_readlinkrpc(vp, &auio); 443 } else { 444 error = EACCES; 445 } 446 } 447 } 448 /* 449 * [re]record creds for reading and/or writing if access 450 * was granted. Assume the NFS server will grant read access 451 * for execute requests. 452 */ 453 if (error == 0) { 454 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 455 crhold(ap->a_cred); 456 if (np->n_rucred) 457 crfree(np->n_rucred); 458 np->n_rucred = ap->a_cred; 459 } 460 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 461 crhold(ap->a_cred); 462 if (np->n_wucred) 463 crfree(np->n_wucred); 464 np->n_wucred = ap->a_cred; 465 } 466 } 467 return(error); 468 } 469 470 /* 471 * nfs open vnode op 472 * Check to see if the type is ok 473 * and that deletion is not in progress. 474 * For paged in text files, you will need to flush the page cache 475 * if consistency is lost. 476 */ 477 /* ARGSUSED */ 478 static int 479 nfs_open(ap) 480 struct vop_open_args /* { 481 struct vnode *a_vp; 482 int a_mode; 483 struct ucred *a_cred; 484 struct thread *a_td; 485 } */ *ap; 486 { 487 struct vnode *vp = ap->a_vp; 488 struct nfsnode *np = VTONFS(vp); 489 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 490 struct vattr vattr; 491 int error; 492 493 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 494 #ifdef DIAGNOSTIC 495 printf("open eacces vtyp=%d\n",vp->v_type); 496 #endif 497 return (EOPNOTSUPP); 498 } 499 /* 500 * Get a valid lease. If cached data is stale, flush it. 501 */ 502 if (nmp->nm_flag & NFSMNT_NQNFS) { 503 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 504 do { 505 error = nqnfs_getlease(vp, ND_READ, ap->a_td); 506 } while (error == NQNFS_EXPIRED); 507 if (error) 508 return (error); 509 if (np->n_lrev != np->n_brev || 510 (np->n_flag & NQNFSNONCACHE)) { 511 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 512 == EINTR) { 513 return (error); 514 } 515 np->n_brev = np->n_lrev; 516 } 517 } 518 } else { 519 if (np->n_flag & NMODIFIED) { 520 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 521 == EINTR) { 522 return (error); 523 } 524 np->n_attrstamp = 0; 525 if (vp->v_type == VDIR) 526 np->n_direofoffset = 0; 527 error = VOP_GETATTR(vp, &vattr, ap->a_td); 528 if (error) 529 return (error); 530 np->n_mtime = vattr.va_mtime.tv_sec; 531 } else { 532 error = VOP_GETATTR(vp, &vattr, ap->a_td); 533 if (error) 534 return (error); 535 if (np->n_mtime != vattr.va_mtime.tv_sec) { 536 if (vp->v_type == VDIR) 537 np->n_direofoffset = 0; 538 if ((error = nfs_vinvalbuf(vp, V_SAVE, 539 ap->a_td, 1)) == EINTR) { 540 return (error); 541 } 542 np->n_mtime = vattr.va_mtime.tv_sec; 543 } 544 } 545 } 546 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0) 547 np->n_attrstamp = 0; /* For Open/Close consistency */ 548 return (0); 549 } 550 551 /* 552 * nfs close vnode op 553 * What an NFS client should do upon close after writing is a debatable issue. 554 * Most NFS clients push delayed writes to the server upon close, basically for 555 * two reasons: 556 * 1 - So that any write errors may be reported back to the client process 557 * doing the close system call. By far the two most likely errors are 558 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 559 * 2 - To put a worst case upper bound on cache inconsistency between 560 * multiple clients for the file. 561 * There is also a consistency problem for Version 2 of the protocol w.r.t. 562 * not being able to tell if other clients are writing a file concurrently, 563 * since there is no way of knowing if the changed modify time in the reply 564 * is only due to the write for this client. 565 * (NFS Version 3 provides weak cache consistency data in the reply that 566 * should be sufficient to detect and handle this case.) 567 * 568 * The current code does the following: 569 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 570 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 571 * or commit them (this satisfies 1 and 2 except for the 572 * case where the server crashes after this close but 573 * before the commit RPC, which is felt to be "good 574 * enough". Changing the last argument to nfs_flush() to 575 * a 1 would force a commit operation, if it is felt a 576 * commit is necessary now. 577 * for NQNFS - do nothing now, since 2 is dealt with via leases and 578 * 1 should be dealt with via an fsync() system call for 579 * cases where write errors are important. 580 */ 581 /* ARGSUSED */ 582 static int 583 nfs_close(ap) 584 struct vop_close_args /* { 585 struct vnodeop_desc *a_desc; 586 struct vnode *a_vp; 587 int a_fflag; 588 struct ucred *a_cred; 589 struct thread *a_td; 590 } */ *ap; 591 { 592 struct vnode *vp = ap->a_vp; 593 struct nfsnode *np = VTONFS(vp); 594 int error = 0; 595 596 if (vp->v_type == VREG) { 597 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 598 (np->n_flag & NMODIFIED)) { 599 if (NFS_ISV3(vp)) { 600 /* 601 * Under NFSv3 we have dirty buffers to dispose of. We 602 * must flush them to the NFS server. We have the option 603 * of waiting all the way through the commit rpc or just 604 * waiting for the initial write. The default is to only 605 * wait through the initial write so the data is in the 606 * server's cache, which is roughly similar to the state 607 * a standard disk subsystem leaves the file in on close(). 608 * 609 * We cannot clear the NMODIFIED bit in np->n_flag due to 610 * potential races with other processes, and certainly 611 * cannot clear it if we don't commit. 612 */ 613 int cm = nfsv3_commit_on_close ? 1 : 0; 614 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm); 615 /* np->n_flag &= ~NMODIFIED; */ 616 } else { 617 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 618 } 619 np->n_attrstamp = 0; 620 } 621 if (np->n_flag & NWRITEERR) { 622 np->n_flag &= ~NWRITEERR; 623 error = np->n_error; 624 } 625 } 626 return (error); 627 } 628 629 /* 630 * nfs getattr call from vfs. 631 */ 632 static int 633 nfs_getattr(ap) 634 struct vop_getattr_args /* { 635 struct vnode *a_vp; 636 struct vattr *a_vap; 637 struct ucred *a_cred; 638 struct thread *a_td; 639 } */ *ap; 640 { 641 struct vnode *vp = ap->a_vp; 642 struct nfsnode *np = VTONFS(vp); 643 caddr_t cp; 644 u_int32_t *tl; 645 int32_t t1, t2; 646 caddr_t bpos, dpos; 647 int error = 0; 648 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 649 int v3 = NFS_ISV3(vp); 650 651 /* 652 * Update local times for special files. 653 */ 654 if (np->n_flag & (NACC | NUPD)) 655 np->n_flag |= NCHG; 656 /* 657 * First look in the cache. 658 */ 659 if (nfs_getattrcache(vp, ap->a_vap) == 0) 660 return (0); 661 662 if (v3 && nfsaccess_cache_timeout > 0) { 663 nfsstats.accesscache_misses++; 664 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 665 if (nfs_getattrcache(vp, ap->a_vap) == 0) 666 return (0); 667 } 668 669 nfsstats.rpccnt[NFSPROC_GETATTR]++; 670 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 671 nfsm_fhtom(vp, v3); 672 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 673 if (!error) { 674 nfsm_loadattr(vp, ap->a_vap); 675 } 676 m_freem(mrep); 677 nfsmout: 678 return (error); 679 } 680 681 /* 682 * nfs setattr call. 683 */ 684 static int 685 nfs_setattr(ap) 686 struct vop_setattr_args /* { 687 struct vnodeop_desc *a_desc; 688 struct vnode *a_vp; 689 struct vattr *a_vap; 690 struct ucred *a_cred; 691 struct thread *a_td; 692 } */ *ap; 693 { 694 struct vnode *vp = ap->a_vp; 695 struct nfsnode *np = VTONFS(vp); 696 struct vattr *vap = ap->a_vap; 697 int error = 0; 698 u_quad_t tsize; 699 700 #ifndef nolint 701 tsize = (u_quad_t)0; 702 #endif 703 704 /* 705 * Setting of flags is not supported. 706 */ 707 if (vap->va_flags != VNOVAL) 708 return (EOPNOTSUPP); 709 710 /* 711 * Disallow write attempts if the filesystem is mounted read-only. 712 */ 713 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 714 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 715 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 716 (vp->v_mount->mnt_flag & MNT_RDONLY)) 717 return (EROFS); 718 if (vap->va_size != VNOVAL) { 719 switch (vp->v_type) { 720 case VDIR: 721 return (EISDIR); 722 case VCHR: 723 case VBLK: 724 case VSOCK: 725 case VFIFO: 726 if (vap->va_mtime.tv_sec == VNOVAL && 727 vap->va_atime.tv_sec == VNOVAL && 728 vap->va_mode == (mode_t)VNOVAL && 729 vap->va_uid == (uid_t)VNOVAL && 730 vap->va_gid == (gid_t)VNOVAL) 731 return (0); 732 vap->va_size = VNOVAL; 733 break; 734 default: 735 /* 736 * Disallow write attempts if the filesystem is 737 * mounted read-only. 738 */ 739 if (vp->v_mount->mnt_flag & MNT_RDONLY) 740 return (EROFS); 741 742 /* 743 * We run vnode_pager_setsize() early (why?), 744 * we must set np->n_size now to avoid vinvalbuf 745 * V_SAVE races that might setsize a lower 746 * value. 747 */ 748 749 tsize = np->n_size; 750 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size); 751 752 if (np->n_flag & NMODIFIED) { 753 if (vap->va_size == 0) 754 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1); 755 else 756 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 757 if (error) { 758 np->n_size = tsize; 759 vnode_pager_setsize(vp, np->n_size); 760 return (error); 761 } 762 } 763 /* np->n_size has already been set to vap->va_size 764 * in nfs_meta_setsize(). We must set it again since 765 * nfs_loadattrcache() could be called through 766 * nfs_meta_setsize() and could modify np->n_size. 767 */ 768 np->n_vattr.va_size = np->n_size = vap->va_size; 769 }; 770 } else if ((vap->va_mtime.tv_sec != VNOVAL || 771 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) && 772 vp->v_type == VREG && 773 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR) 774 return (error); 775 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 776 if (error && vap->va_size != VNOVAL) { 777 np->n_size = np->n_vattr.va_size = tsize; 778 vnode_pager_setsize(vp, np->n_size); 779 } 780 return (error); 781 } 782 783 /* 784 * Do an nfs setattr rpc. 785 */ 786 static int 787 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 788 struct ucred *cred, struct thread *td) 789 { 790 struct nfsv2_sattr *sp; 791 caddr_t cp; 792 int32_t t1, t2; 793 caddr_t bpos, dpos, cp2; 794 u_int32_t *tl; 795 int error = 0, wccflag = NFSV3_WCCRATTR; 796 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 797 int v3 = NFS_ISV3(vp); 798 799 nfsstats.rpccnt[NFSPROC_SETATTR]++; 800 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 801 nfsm_fhtom(vp, v3); 802 if (v3) { 803 nfsm_v3attrbuild(vap, TRUE); 804 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 805 *tl = nfs_false; 806 } else { 807 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 808 if (vap->va_mode == (mode_t)VNOVAL) 809 sp->sa_mode = nfs_xdrneg1; 810 else 811 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 812 if (vap->va_uid == (uid_t)VNOVAL) 813 sp->sa_uid = nfs_xdrneg1; 814 else 815 sp->sa_uid = txdr_unsigned(vap->va_uid); 816 if (vap->va_gid == (gid_t)VNOVAL) 817 sp->sa_gid = nfs_xdrneg1; 818 else 819 sp->sa_gid = txdr_unsigned(vap->va_gid); 820 sp->sa_size = txdr_unsigned(vap->va_size); 821 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 822 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 823 } 824 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 825 if (v3) { 826 nfsm_wcc_data(vp, wccflag); 827 } else 828 nfsm_loadattr(vp, (struct vattr *)0); 829 m_freem(mrep); 830 nfsmout: 831 return (error); 832 } 833 834 /* 835 * 'cached' nfs directory lookup 836 */ 837 static int 838 nfs_lookup(ap) 839 struct vop_lookup_args /* { 840 struct vnodeop_desc *a_desc; 841 struct vnode *a_dvp; 842 struct vnode **a_vpp; 843 struct componentname *a_cnp; 844 } */ *ap; 845 { 846 struct componentname *cnp = ap->a_cnp; 847 struct vnode *dvp = ap->a_dvp; 848 struct vnode **vpp = ap->a_vpp; 849 int flags = cnp->cn_flags; 850 struct vnode *newvp; 851 u_int32_t *tl; 852 caddr_t cp; 853 int32_t t1, t2; 854 struct nfsmount *nmp; 855 caddr_t bpos, dpos, cp2; 856 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 857 long len; 858 nfsfh_t *fhp; 859 struct nfsnode *np; 860 int lockparent, wantparent, error = 0, attrflag, fhsize; 861 int v3 = NFS_ISV3(dvp); 862 struct thread *td = cnp->cn_td; 863 864 /* 865 * Read-only mount check and directory check. 866 */ 867 *vpp = NULLVP; 868 if ((flags & CNP_ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 869 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 870 return (EROFS); 871 872 if (dvp->v_type != VDIR) 873 return (ENOTDIR); 874 875 /* 876 * Look it up in the cache. Note that ENOENT is only returned if we 877 * previously entered a negative hit (see later on). The additional 878 * nfsneg_cache_timeout check causes previously cached results to 879 * be instantly ignored if the negative caching is turned off. 880 */ 881 lockparent = flags & CNP_LOCKPARENT; 882 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 883 nmp = VFSTONFS(dvp->v_mount); 884 np = VTONFS(dvp); 885 error = cache_lookup(dvp, NCPNULL, vpp, NCPPNULL, cnp); 886 if (error != 0) { 887 struct vattr vattr; 888 int vpid; 889 890 if (error == ENOENT && nfsneg_cache_timeout) { 891 *vpp = NULLVP; 892 return (error); 893 } 894 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { 895 *vpp = NULLVP; 896 return (error); 897 } 898 899 newvp = *vpp; 900 vpid = newvp->v_id; 901 /* 902 * See the comment starting `Step through' in ufs/ufs_lookup.c 903 * for an explanation of the locking protocol 904 */ 905 if (dvp == newvp) { 906 VREF(newvp); 907 error = 0; 908 } else if (flags & CNP_ISDOTDOT) { 909 VOP_UNLOCK(dvp, NULL, 0, td); 910 error = vget(newvp, NULL, LK_EXCLUSIVE, td); 911 if (!error && lockparent && (flags & CNP_ISLASTCN)) 912 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td); 913 } else { 914 error = vget(newvp, NULL, LK_EXCLUSIVE, td); 915 if (!lockparent || error || !(flags & CNP_ISLASTCN)) 916 VOP_UNLOCK(dvp, NULL, 0, td); 917 } 918 if (!error) { 919 if (vpid == newvp->v_id) { 920 if (!VOP_GETATTR(newvp, &vattr, td) 921 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) { 922 nfsstats.lookupcache_hits++; 923 if (cnp->cn_nameiop != NAMEI_LOOKUP && 924 (flags & CNP_ISLASTCN)) 925 cnp->cn_flags |= CNP_SAVENAME; 926 return (0); 927 } 928 cache_purge(newvp); 929 } 930 vput(newvp); 931 if (lockparent && dvp != newvp && (flags & CNP_ISLASTCN)) 932 VOP_UNLOCK(dvp, NULL, 0, td); 933 } 934 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td); 935 *vpp = NULLVP; 936 if (error) 937 return (error); 938 } 939 940 /* 941 * Cache miss, go the wire. 942 */ 943 error = 0; 944 newvp = NULLVP; 945 nfsstats.lookupcache_misses++; 946 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 947 len = cnp->cn_namelen; 948 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 949 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 950 nfsm_fhtom(dvp, v3); 951 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 952 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 953 if (error) { 954 /* 955 * Cache negatve lookups to reduce NFS traffic, but use 956 * a fast timeout. 957 */ 958 if (error == ENOENT && 959 (cnp->cn_flags & CNP_MAKEENTRY) && 960 cnp->cn_nameiop == NAMEI_LOOKUP && 961 nfsneg_cache_timeout) { 962 int toval = nfsneg_cache_timeout * hz; 963 if (cnp->cn_flags & CNP_CACHETIMEOUT) { 964 if (cnp->cn_timeout > toval) 965 cnp->cn_timeout = toval; 966 } else { 967 cnp->cn_flags |= CNP_CACHETIMEOUT; 968 cnp->cn_timeout = toval; 969 } 970 cache_enter(dvp, NCPNULL, NULL, cnp); 971 } 972 nfsm_postop_attr(dvp, attrflag); 973 m_freem(mrep); 974 goto nfsmout; 975 } 976 nfsm_getfh(fhp, fhsize, v3); 977 978 /* 979 * Handle RENAME case... 980 */ 981 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent && (flags & CNP_ISLASTCN)) { 982 if (NFS_CMPFH(np, fhp, fhsize)) { 983 m_freem(mrep); 984 return (EISDIR); 985 } 986 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 987 if (error) { 988 m_freem(mrep); 989 return (error); 990 } 991 newvp = NFSTOV(np); 992 if (v3) { 993 nfsm_postop_attr(newvp, attrflag); 994 nfsm_postop_attr(dvp, attrflag); 995 } else 996 nfsm_loadattr(newvp, (struct vattr *)0); 997 *vpp = newvp; 998 m_freem(mrep); 999 cnp->cn_flags |= CNP_SAVENAME; 1000 if (!lockparent) 1001 VOP_UNLOCK(dvp, NULL, 0, td); 1002 return (0); 1003 } 1004 1005 if (flags & CNP_ISDOTDOT) { 1006 VOP_UNLOCK(dvp, NULL, 0, td); 1007 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1008 if (error) { 1009 vn_lock(dvp, NULL, LK_EXCLUSIVE | LK_RETRY, td); 1010 return (error); 1011 } 1012 newvp = NFSTOV(np); 1013 if (lockparent && (flags & CNP_ISLASTCN) && 1014 (error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td))) { 1015 vput(newvp); 1016 return (error); 1017 } 1018 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1019 VREF(dvp); 1020 newvp = dvp; 1021 } else { 1022 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1023 if (error) { 1024 m_freem(mrep); 1025 return (error); 1026 } 1027 if (!lockparent || !(flags & CNP_ISLASTCN)) 1028 VOP_UNLOCK(dvp, NULL, 0, td); 1029 newvp = NFSTOV(np); 1030 } 1031 if (v3) { 1032 nfsm_postop_attr(newvp, attrflag); 1033 nfsm_postop_attr(dvp, attrflag); 1034 } else 1035 nfsm_loadattr(newvp, (struct vattr *)0); 1036 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN)) 1037 cnp->cn_flags |= CNP_SAVENAME; 1038 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1039 (cnp->cn_nameiop != NAMEI_DELETE || !(flags & CNP_ISLASTCN))) { 1040 np->n_ctime = np->n_vattr.va_ctime.tv_sec; 1041 cache_enter(dvp, NCPNULL, newvp, cnp); 1042 } 1043 *vpp = newvp; 1044 m_freem(mrep); 1045 nfsmout: 1046 if (error) { 1047 if (newvp != NULLVP) { 1048 vrele(newvp); 1049 *vpp = NULLVP; 1050 } 1051 if ((cnp->cn_nameiop == NAMEI_CREATE || cnp->cn_nameiop == NAMEI_RENAME) && 1052 (flags & CNP_ISLASTCN) && error == ENOENT) { 1053 if (!lockparent) 1054 VOP_UNLOCK(dvp, NULL, 0, td); 1055 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1056 error = EROFS; 1057 else 1058 error = EJUSTRETURN; 1059 } 1060 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN)) 1061 cnp->cn_flags |= CNP_SAVENAME; 1062 } 1063 return (error); 1064 } 1065 1066 /* 1067 * nfs read call. 1068 * Just call nfs_bioread() to do the work. 1069 */ 1070 static int 1071 nfs_read(ap) 1072 struct vop_read_args /* { 1073 struct vnode *a_vp; 1074 struct uio *a_uio; 1075 int a_ioflag; 1076 struct ucred *a_cred; 1077 } */ *ap; 1078 { 1079 struct vnode *vp = ap->a_vp; 1080 1081 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1082 switch (vp->v_type) { 1083 case VREG: 1084 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1085 case VDIR: 1086 return (EISDIR); 1087 default: 1088 return EOPNOTSUPP; 1089 } 1090 } 1091 1092 /* 1093 * nfs readlink call 1094 */ 1095 static int 1096 nfs_readlink(ap) 1097 struct vop_readlink_args /* { 1098 struct vnode *a_vp; 1099 struct uio *a_uio; 1100 struct ucred *a_cred; 1101 } */ *ap; 1102 { 1103 struct vnode *vp = ap->a_vp; 1104 1105 if (vp->v_type != VLNK) 1106 return (EINVAL); 1107 return (nfs_bioread(vp, ap->a_uio, 0)); 1108 } 1109 1110 /* 1111 * Do a readlink rpc. 1112 * Called by nfs_doio() from below the buffer cache. 1113 */ 1114 int 1115 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1116 { 1117 u_int32_t *tl; 1118 caddr_t cp; 1119 int32_t t1, t2; 1120 caddr_t bpos, dpos, cp2; 1121 int error = 0, len, attrflag; 1122 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1123 int v3 = NFS_ISV3(vp); 1124 1125 nfsstats.rpccnt[NFSPROC_READLINK]++; 1126 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1127 nfsm_fhtom(vp, v3); 1128 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1129 if (v3) 1130 nfsm_postop_attr(vp, attrflag); 1131 if (!error) { 1132 nfsm_strsiz(len, NFS_MAXPATHLEN); 1133 if (len == NFS_MAXPATHLEN) { 1134 struct nfsnode *np = VTONFS(vp); 1135 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1136 len = np->n_size; 1137 } 1138 nfsm_mtouio(uiop, len); 1139 } 1140 m_freem(mrep); 1141 nfsmout: 1142 return (error); 1143 } 1144 1145 /* 1146 * nfs read rpc call 1147 * Ditto above 1148 */ 1149 int 1150 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1151 { 1152 u_int32_t *tl; 1153 caddr_t cp; 1154 int32_t t1, t2; 1155 caddr_t bpos, dpos, cp2; 1156 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1157 struct nfsmount *nmp; 1158 int error = 0, len, retlen, tsiz, eof, attrflag; 1159 int v3 = NFS_ISV3(vp); 1160 1161 #ifndef nolint 1162 eof = 0; 1163 #endif 1164 nmp = VFSTONFS(vp->v_mount); 1165 tsiz = uiop->uio_resid; 1166 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1167 return (EFBIG); 1168 while (tsiz > 0) { 1169 nfsstats.rpccnt[NFSPROC_READ]++; 1170 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1171 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1172 nfsm_fhtom(vp, v3); 1173 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1174 if (v3) { 1175 txdr_hyper(uiop->uio_offset, tl); 1176 *(tl + 2) = txdr_unsigned(len); 1177 } else { 1178 *tl++ = txdr_unsigned(uiop->uio_offset); 1179 *tl++ = txdr_unsigned(len); 1180 *tl = 0; 1181 } 1182 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1183 if (v3) { 1184 nfsm_postop_attr(vp, attrflag); 1185 if (error) { 1186 m_freem(mrep); 1187 goto nfsmout; 1188 } 1189 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1190 eof = fxdr_unsigned(int, *(tl + 1)); 1191 } else 1192 nfsm_loadattr(vp, (struct vattr *)0); 1193 nfsm_strsiz(retlen, nmp->nm_rsize); 1194 nfsm_mtouio(uiop, retlen); 1195 m_freem(mrep); 1196 tsiz -= retlen; 1197 if (v3) { 1198 if (eof || retlen == 0) { 1199 tsiz = 0; 1200 } 1201 } else if (retlen < len) { 1202 tsiz = 0; 1203 } 1204 } 1205 nfsmout: 1206 return (error); 1207 } 1208 1209 /* 1210 * nfs write call 1211 */ 1212 int 1213 nfs_writerpc(vp, uiop, iomode, must_commit) 1214 struct vnode *vp; 1215 struct uio *uiop; 1216 int *iomode, *must_commit; 1217 { 1218 u_int32_t *tl; 1219 caddr_t cp; 1220 int32_t t1, t2, backup; 1221 caddr_t bpos, dpos, cp2; 1222 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1223 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1224 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1225 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1226 1227 #ifndef DIAGNOSTIC 1228 if (uiop->uio_iovcnt != 1) 1229 panic("nfs: writerpc iovcnt > 1"); 1230 #endif 1231 *must_commit = 0; 1232 tsiz = uiop->uio_resid; 1233 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1234 return (EFBIG); 1235 while (tsiz > 0) { 1236 nfsstats.rpccnt[NFSPROC_WRITE]++; 1237 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1238 nfsm_reqhead(vp, NFSPROC_WRITE, 1239 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1240 nfsm_fhtom(vp, v3); 1241 if (v3) { 1242 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1243 txdr_hyper(uiop->uio_offset, tl); 1244 tl += 2; 1245 *tl++ = txdr_unsigned(len); 1246 *tl++ = txdr_unsigned(*iomode); 1247 *tl = txdr_unsigned(len); 1248 } else { 1249 u_int32_t x; 1250 1251 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1252 /* Set both "begin" and "current" to non-garbage. */ 1253 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1254 *tl++ = x; /* "begin offset" */ 1255 *tl++ = x; /* "current offset" */ 1256 x = txdr_unsigned(len); 1257 *tl++ = x; /* total to this offset */ 1258 *tl = x; /* size of this write */ 1259 } 1260 nfsm_uiotom(uiop, len); 1261 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1262 if (v3) { 1263 wccflag = NFSV3_WCCCHK; 1264 nfsm_wcc_data(vp, wccflag); 1265 if (!error) { 1266 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1267 + NFSX_V3WRITEVERF); 1268 rlen = fxdr_unsigned(int, *tl++); 1269 if (rlen == 0) { 1270 error = NFSERR_IO; 1271 m_freem(mrep); 1272 break; 1273 } else if (rlen < len) { 1274 backup = len - rlen; 1275 uiop->uio_iov->iov_base -= backup; 1276 uiop->uio_iov->iov_len += backup; 1277 uiop->uio_offset -= backup; 1278 uiop->uio_resid += backup; 1279 len = rlen; 1280 } 1281 commit = fxdr_unsigned(int, *tl++); 1282 1283 /* 1284 * Return the lowest committment level 1285 * obtained by any of the RPCs. 1286 */ 1287 if (committed == NFSV3WRITE_FILESYNC) 1288 committed = commit; 1289 else if (committed == NFSV3WRITE_DATASYNC && 1290 commit == NFSV3WRITE_UNSTABLE) 1291 committed = commit; 1292 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1293 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1294 NFSX_V3WRITEVERF); 1295 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1296 } else if (bcmp((caddr_t)tl, 1297 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1298 *must_commit = 1; 1299 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1300 NFSX_V3WRITEVERF); 1301 } 1302 } 1303 } else 1304 nfsm_loadattr(vp, (struct vattr *)0); 1305 if (wccflag) 1306 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec; 1307 m_freem(mrep); 1308 if (error) 1309 break; 1310 tsiz -= len; 1311 } 1312 nfsmout: 1313 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1314 committed = NFSV3WRITE_FILESYNC; 1315 *iomode = committed; 1316 if (error) 1317 uiop->uio_resid = tsiz; 1318 return (error); 1319 } 1320 1321 /* 1322 * nfs mknod rpc 1323 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1324 * mode set to specify the file type and the size field for rdev. 1325 */ 1326 static int 1327 nfs_mknodrpc(dvp, vpp, cnp, vap) 1328 struct vnode *dvp; 1329 struct vnode **vpp; 1330 struct componentname *cnp; 1331 struct vattr *vap; 1332 { 1333 struct nfsv2_sattr *sp; 1334 u_int32_t *tl; 1335 caddr_t cp; 1336 int32_t t1, t2; 1337 struct vnode *newvp = (struct vnode *)0; 1338 struct nfsnode *np = (struct nfsnode *)0; 1339 struct vattr vattr; 1340 char *cp2; 1341 caddr_t bpos, dpos; 1342 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1343 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1344 u_int32_t rdev; 1345 int v3 = NFS_ISV3(dvp); 1346 1347 if (vap->va_type == VCHR || vap->va_type == VBLK) 1348 rdev = txdr_unsigned(vap->va_rdev); 1349 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1350 rdev = nfs_xdrneg1; 1351 else { 1352 return (EOPNOTSUPP); 1353 } 1354 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1355 return (error); 1356 } 1357 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1358 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1359 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1360 nfsm_fhtom(dvp, v3); 1361 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1362 if (v3) { 1363 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1364 *tl++ = vtonfsv3_type(vap->va_type); 1365 nfsm_v3attrbuild(vap, FALSE); 1366 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1367 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1368 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1369 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1370 } 1371 } else { 1372 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1373 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1374 sp->sa_uid = nfs_xdrneg1; 1375 sp->sa_gid = nfs_xdrneg1; 1376 sp->sa_size = rdev; 1377 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1378 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1379 } 1380 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1381 if (!error) { 1382 nfsm_mtofh(dvp, newvp, v3, gotvp); 1383 if (!gotvp) { 1384 if (newvp) { 1385 vput(newvp); 1386 newvp = (struct vnode *)0; 1387 } 1388 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1389 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1390 if (!error) 1391 newvp = NFSTOV(np); 1392 } 1393 } 1394 if (v3) 1395 nfsm_wcc_data(dvp, wccflag); 1396 m_freem(mrep); 1397 nfsmout: 1398 if (error) { 1399 if (newvp) 1400 vput(newvp); 1401 } else { 1402 if (cnp->cn_flags & CNP_MAKEENTRY) 1403 cache_enter(dvp, NCPNULL, newvp, cnp); 1404 *vpp = newvp; 1405 } 1406 VTONFS(dvp)->n_flag |= NMODIFIED; 1407 if (!wccflag) 1408 VTONFS(dvp)->n_attrstamp = 0; 1409 return (error); 1410 } 1411 1412 /* 1413 * nfs mknod vop 1414 * just call nfs_mknodrpc() to do the work. 1415 */ 1416 /* ARGSUSED */ 1417 static int 1418 nfs_mknod(ap) 1419 struct vop_mknod_args /* { 1420 struct vnode *a_dvp; 1421 struct vnode **a_vpp; 1422 struct componentname *a_cnp; 1423 struct vattr *a_vap; 1424 } */ *ap; 1425 { 1426 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1427 } 1428 1429 static u_long create_verf; 1430 /* 1431 * nfs file create call 1432 */ 1433 static int 1434 nfs_create(ap) 1435 struct vop_create_args /* { 1436 struct vnode *a_dvp; 1437 struct vnode **a_vpp; 1438 struct componentname *a_cnp; 1439 struct vattr *a_vap; 1440 } */ *ap; 1441 { 1442 struct vnode *dvp = ap->a_dvp; 1443 struct vattr *vap = ap->a_vap; 1444 struct componentname *cnp = ap->a_cnp; 1445 struct nfsv2_sattr *sp; 1446 u_int32_t *tl; 1447 caddr_t cp; 1448 int32_t t1, t2; 1449 struct nfsnode *np = (struct nfsnode *)0; 1450 struct vnode *newvp = (struct vnode *)0; 1451 caddr_t bpos, dpos, cp2; 1452 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1453 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1454 struct vattr vattr; 1455 int v3 = NFS_ISV3(dvp); 1456 1457 /* 1458 * Oops, not for me.. 1459 */ 1460 if (vap->va_type == VSOCK) 1461 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1462 1463 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1464 return (error); 1465 } 1466 if (vap->va_vaflags & VA_EXCLUSIVE) 1467 fmode |= O_EXCL; 1468 again: 1469 nfsstats.rpccnt[NFSPROC_CREATE]++; 1470 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1471 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1472 nfsm_fhtom(dvp, v3); 1473 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1474 if (v3) { 1475 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1476 if (fmode & O_EXCL) { 1477 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1478 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1479 #ifdef INET 1480 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1481 *tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr; 1482 else 1483 #endif 1484 *tl++ = create_verf; 1485 *tl = ++create_verf; 1486 } else { 1487 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1488 nfsm_v3attrbuild(vap, FALSE); 1489 } 1490 } else { 1491 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1492 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1493 sp->sa_uid = nfs_xdrneg1; 1494 sp->sa_gid = nfs_xdrneg1; 1495 sp->sa_size = 0; 1496 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1497 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1498 } 1499 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1500 if (!error) { 1501 nfsm_mtofh(dvp, newvp, v3, gotvp); 1502 if (!gotvp) { 1503 if (newvp) { 1504 vput(newvp); 1505 newvp = (struct vnode *)0; 1506 } 1507 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1508 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1509 if (!error) 1510 newvp = NFSTOV(np); 1511 } 1512 } 1513 if (v3) 1514 nfsm_wcc_data(dvp, wccflag); 1515 m_freem(mrep); 1516 nfsmout: 1517 if (error) { 1518 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1519 fmode &= ~O_EXCL; 1520 goto again; 1521 } 1522 if (newvp) 1523 vput(newvp); 1524 } else if (v3 && (fmode & O_EXCL)) { 1525 /* 1526 * We are normally called with only a partially initialized 1527 * VAP. Since the NFSv3 spec says that server may use the 1528 * file attributes to store the verifier, the spec requires 1529 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1530 * in atime, but we can't really assume that all servers will 1531 * so we ensure that our SETATTR sets both atime and mtime. 1532 */ 1533 if (vap->va_mtime.tv_sec == VNOVAL) 1534 vfs_timestamp(&vap->va_mtime); 1535 if (vap->va_atime.tv_sec == VNOVAL) 1536 vap->va_atime = vap->va_mtime; 1537 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1538 } 1539 if (!error) { 1540 if (cnp->cn_flags & CNP_MAKEENTRY) 1541 cache_enter(dvp, NCPNULL, newvp, cnp); 1542 /* 1543 * The new np may have enough info for access 1544 * checks, make sure rucred and wucred are 1545 * initialized for read and write rpc's. 1546 */ 1547 np = VTONFS(newvp); 1548 if (np->n_rucred == NULL) 1549 np->n_rucred = crhold(cnp->cn_cred); 1550 if (np->n_wucred == NULL) 1551 np->n_wucred = crhold(cnp->cn_cred); 1552 *ap->a_vpp = newvp; 1553 } 1554 VTONFS(dvp)->n_flag |= NMODIFIED; 1555 if (!wccflag) 1556 VTONFS(dvp)->n_attrstamp = 0; 1557 return (error); 1558 } 1559 1560 /* 1561 * nfs file remove call 1562 * To try and make nfs semantics closer to ufs semantics, a file that has 1563 * other processes using the vnode is renamed instead of removed and then 1564 * removed later on the last close. 1565 * - If v_usecount > 1 1566 * If a rename is not already in the works 1567 * call nfs_sillyrename() to set it up 1568 * else 1569 * do the remove rpc 1570 */ 1571 static int 1572 nfs_remove(ap) 1573 struct vop_remove_args /* { 1574 struct vnodeop_desc *a_desc; 1575 struct vnode * a_dvp; 1576 struct vnode * a_vp; 1577 struct componentname * a_cnp; 1578 } */ *ap; 1579 { 1580 struct vnode *vp = ap->a_vp; 1581 struct vnode *dvp = ap->a_dvp; 1582 struct componentname *cnp = ap->a_cnp; 1583 struct nfsnode *np = VTONFS(vp); 1584 int error = 0; 1585 struct vattr vattr; 1586 1587 #ifndef DIAGNOSTIC 1588 if ((cnp->cn_flags & CNP_HASBUF) == 0) 1589 panic("nfs_remove: no name"); 1590 if (vp->v_usecount < 1) 1591 panic("nfs_remove: bad v_usecount"); 1592 #endif 1593 if (vp->v_type == VDIR) 1594 error = EPERM; 1595 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1596 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 && 1597 vattr.va_nlink > 1)) { 1598 /* 1599 * Purge the name cache so that the chance of a lookup for 1600 * the name succeeding while the remove is in progress is 1601 * minimized. Without node locking it can still happen, such 1602 * that an I/O op returns ESTALE, but since you get this if 1603 * another host removes the file.. 1604 */ 1605 cache_purge(vp); 1606 /* 1607 * throw away biocache buffers, mainly to avoid 1608 * unnecessary delayed writes later. 1609 */ 1610 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1); 1611 /* Do the rpc */ 1612 if (error != EINTR) 1613 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1614 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1615 /* 1616 * Kludge City: If the first reply to the remove rpc is lost.. 1617 * the reply to the retransmitted request will be ENOENT 1618 * since the file was in fact removed 1619 * Therefore, we cheat and return success. 1620 */ 1621 if (error == ENOENT) 1622 error = 0; 1623 } else if (!np->n_sillyrename) 1624 error = nfs_sillyrename(dvp, vp, cnp); 1625 np->n_attrstamp = 0; 1626 return (error); 1627 } 1628 1629 /* 1630 * nfs file remove rpc called from nfs_inactive 1631 */ 1632 int 1633 nfs_removeit(struct sillyrename *sp) 1634 { 1635 1636 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1637 sp->s_cred, NULL)); 1638 } 1639 1640 /* 1641 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1642 */ 1643 static int 1644 nfs_removerpc(dvp, name, namelen, cred, td) 1645 struct vnode *dvp; 1646 const char *name; 1647 int namelen; 1648 struct ucred *cred; 1649 struct thread *td; 1650 { 1651 u_int32_t *tl; 1652 caddr_t cp; 1653 int32_t t1, t2; 1654 caddr_t bpos, dpos, cp2; 1655 int error = 0, wccflag = NFSV3_WCCRATTR; 1656 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1657 int v3 = NFS_ISV3(dvp); 1658 1659 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1660 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1661 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1662 nfsm_fhtom(dvp, v3); 1663 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1664 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1665 if (v3) 1666 nfsm_wcc_data(dvp, wccflag); 1667 m_freem(mrep); 1668 nfsmout: 1669 VTONFS(dvp)->n_flag |= NMODIFIED; 1670 if (!wccflag) 1671 VTONFS(dvp)->n_attrstamp = 0; 1672 return (error); 1673 } 1674 1675 /* 1676 * nfs file rename call 1677 */ 1678 static int 1679 nfs_rename(ap) 1680 struct vop_rename_args /* { 1681 struct vnode *a_fdvp; 1682 struct vnode *a_fvp; 1683 struct componentname *a_fcnp; 1684 struct vnode *a_tdvp; 1685 struct vnode *a_tvp; 1686 struct componentname *a_tcnp; 1687 } */ *ap; 1688 { 1689 struct vnode *fvp = ap->a_fvp; 1690 struct vnode *tvp = ap->a_tvp; 1691 struct vnode *fdvp = ap->a_fdvp; 1692 struct vnode *tdvp = ap->a_tdvp; 1693 struct componentname *tcnp = ap->a_tcnp; 1694 struct componentname *fcnp = ap->a_fcnp; 1695 int error; 1696 1697 #ifndef DIAGNOSTIC 1698 if ((tcnp->cn_flags & CNP_HASBUF) == 0 || 1699 (fcnp->cn_flags & CNP_HASBUF) == 0) 1700 panic("nfs_rename: no name"); 1701 #endif 1702 /* Check for cross-device rename */ 1703 if ((fvp->v_mount != tdvp->v_mount) || 1704 (tvp && (fvp->v_mount != tvp->v_mount))) { 1705 error = EXDEV; 1706 goto out; 1707 } 1708 1709 /* 1710 * We have to flush B_DELWRI data prior to renaming 1711 * the file. If we don't, the delayed-write buffers 1712 * can be flushed out later after the file has gone stale 1713 * under NFSV3. NFSV2 does not have this problem because 1714 * ( as far as I can tell ) it flushes dirty buffers more 1715 * often. 1716 */ 1717 1718 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td); 1719 if (tvp) 1720 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td); 1721 1722 /* 1723 * If the tvp exists and is in use, sillyrename it before doing the 1724 * rename of the new file over it. 1725 * XXX Can't sillyrename a directory. 1726 */ 1727 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1728 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1729 vput(tvp); 1730 tvp = NULL; 1731 } 1732 1733 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1734 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1735 tcnp->cn_td); 1736 1737 if (fvp->v_type == VDIR) { 1738 if (tvp != NULL && tvp->v_type == VDIR) 1739 cache_purge(tdvp); 1740 cache_purge(fdvp); 1741 } 1742 1743 out: 1744 if (tdvp == tvp) 1745 vrele(tdvp); 1746 else 1747 vput(tdvp); 1748 if (tvp) 1749 vput(tvp); 1750 vrele(fdvp); 1751 vrele(fvp); 1752 /* 1753 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1754 */ 1755 if (error == ENOENT) 1756 error = 0; 1757 return (error); 1758 } 1759 1760 /* 1761 * nfs file rename rpc called from nfs_remove() above 1762 */ 1763 static int 1764 nfs_renameit(sdvp, scnp, sp) 1765 struct vnode *sdvp; 1766 struct componentname *scnp; 1767 struct sillyrename *sp; 1768 { 1769 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1770 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1771 } 1772 1773 /* 1774 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1775 */ 1776 static int 1777 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, td) 1778 struct vnode *fdvp; 1779 const char *fnameptr; 1780 int fnamelen; 1781 struct vnode *tdvp; 1782 const char *tnameptr; 1783 int tnamelen; 1784 struct ucred *cred; 1785 struct thread *td; 1786 { 1787 u_int32_t *tl; 1788 caddr_t cp; 1789 int32_t t1, t2; 1790 caddr_t bpos, dpos, cp2; 1791 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1792 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1793 int v3 = NFS_ISV3(fdvp); 1794 1795 nfsstats.rpccnt[NFSPROC_RENAME]++; 1796 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1797 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1798 nfsm_rndup(tnamelen)); 1799 nfsm_fhtom(fdvp, v3); 1800 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1801 nfsm_fhtom(tdvp, v3); 1802 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1803 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1804 if (v3) { 1805 nfsm_wcc_data(fdvp, fwccflag); 1806 nfsm_wcc_data(tdvp, twccflag); 1807 } 1808 m_freem(mrep); 1809 nfsmout: 1810 VTONFS(fdvp)->n_flag |= NMODIFIED; 1811 VTONFS(tdvp)->n_flag |= NMODIFIED; 1812 if (!fwccflag) 1813 VTONFS(fdvp)->n_attrstamp = 0; 1814 if (!twccflag) 1815 VTONFS(tdvp)->n_attrstamp = 0; 1816 return (error); 1817 } 1818 1819 /* 1820 * nfs hard link create call 1821 */ 1822 static int 1823 nfs_link(ap) 1824 struct vop_link_args /* { 1825 struct vnode *a_tdvp; 1826 struct vnode *a_vp; 1827 struct componentname *a_cnp; 1828 } */ *ap; 1829 { 1830 struct vnode *vp = ap->a_vp; 1831 struct vnode *tdvp = ap->a_tdvp; 1832 struct componentname *cnp = ap->a_cnp; 1833 u_int32_t *tl; 1834 caddr_t cp; 1835 int32_t t1, t2; 1836 caddr_t bpos, dpos, cp2; 1837 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1838 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1839 int v3; 1840 1841 if (vp->v_mount != tdvp->v_mount) { 1842 return (EXDEV); 1843 } 1844 1845 /* 1846 * Push all writes to the server, so that the attribute cache 1847 * doesn't get "out of sync" with the server. 1848 * XXX There should be a better way! 1849 */ 1850 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td); 1851 1852 v3 = NFS_ISV3(vp); 1853 nfsstats.rpccnt[NFSPROC_LINK]++; 1854 nfsm_reqhead(vp, NFSPROC_LINK, 1855 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1856 nfsm_fhtom(vp, v3); 1857 nfsm_fhtom(tdvp, v3); 1858 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1859 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1860 if (v3) { 1861 nfsm_postop_attr(vp, attrflag); 1862 nfsm_wcc_data(tdvp, wccflag); 1863 } 1864 m_freem(mrep); 1865 nfsmout: 1866 VTONFS(tdvp)->n_flag |= NMODIFIED; 1867 if (!attrflag) 1868 VTONFS(vp)->n_attrstamp = 0; 1869 if (!wccflag) 1870 VTONFS(tdvp)->n_attrstamp = 0; 1871 /* 1872 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1873 */ 1874 if (error == EEXIST) 1875 error = 0; 1876 return (error); 1877 } 1878 1879 /* 1880 * nfs symbolic link create call 1881 */ 1882 static int 1883 nfs_symlink(ap) 1884 struct vop_symlink_args /* { 1885 struct vnode *a_dvp; 1886 struct vnode **a_vpp; 1887 struct componentname *a_cnp; 1888 struct vattr *a_vap; 1889 char *a_target; 1890 } */ *ap; 1891 { 1892 struct vnode *dvp = ap->a_dvp; 1893 struct vattr *vap = ap->a_vap; 1894 struct componentname *cnp = ap->a_cnp; 1895 struct nfsv2_sattr *sp; 1896 u_int32_t *tl; 1897 caddr_t cp; 1898 int32_t t1, t2; 1899 caddr_t bpos, dpos, cp2; 1900 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1901 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1902 struct vnode *newvp = (struct vnode *)0; 1903 int v3 = NFS_ISV3(dvp); 1904 1905 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1906 slen = strlen(ap->a_target); 1907 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1908 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1909 nfsm_fhtom(dvp, v3); 1910 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1911 if (v3) { 1912 nfsm_v3attrbuild(vap, FALSE); 1913 } 1914 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1915 if (!v3) { 1916 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1917 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1918 sp->sa_uid = nfs_xdrneg1; 1919 sp->sa_gid = nfs_xdrneg1; 1920 sp->sa_size = nfs_xdrneg1; 1921 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1922 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1923 } 1924 1925 /* 1926 * Issue the NFS request and get the rpc response. 1927 * 1928 * Only NFSv3 responses returning an error of 0 actually return 1929 * a file handle that can be converted into newvp without having 1930 * to do an extra lookup rpc. 1931 */ 1932 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1933 if (v3) { 1934 if (error == 0) 1935 nfsm_mtofh(dvp, newvp, v3, gotvp); 1936 nfsm_wcc_data(dvp, wccflag); 1937 } 1938 1939 /* 1940 * out code jumps -> here, mrep is also freed. 1941 */ 1942 1943 m_freem(mrep); 1944 nfsmout: 1945 1946 /* 1947 * If we get an EEXIST error, silently convert it to no-error 1948 * in case of an NFS retry. 1949 */ 1950 if (error == EEXIST) 1951 error = 0; 1952 1953 /* 1954 * If we do not have (or no longer have) an error, and we could 1955 * not extract the newvp from the response due to the request being 1956 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1957 * to obtain a newvp to return. 1958 */ 1959 if (error == 0 && newvp == NULL) { 1960 struct nfsnode *np = NULL; 1961 1962 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1963 cnp->cn_cred, cnp->cn_td, &np); 1964 if (!error) 1965 newvp = NFSTOV(np); 1966 } 1967 if (error) { 1968 if (newvp) 1969 vput(newvp); 1970 } else { 1971 *ap->a_vpp = newvp; 1972 } 1973 VTONFS(dvp)->n_flag |= NMODIFIED; 1974 if (!wccflag) 1975 VTONFS(dvp)->n_attrstamp = 0; 1976 return (error); 1977 } 1978 1979 /* 1980 * nfs make dir call 1981 */ 1982 static int 1983 nfs_mkdir(ap) 1984 struct vop_mkdir_args /* { 1985 struct vnode *a_dvp; 1986 struct vnode **a_vpp; 1987 struct componentname *a_cnp; 1988 struct vattr *a_vap; 1989 } */ *ap; 1990 { 1991 struct vnode *dvp = ap->a_dvp; 1992 struct vattr *vap = ap->a_vap; 1993 struct componentname *cnp = ap->a_cnp; 1994 struct nfsv2_sattr *sp; 1995 u_int32_t *tl; 1996 caddr_t cp; 1997 int32_t t1, t2; 1998 int len; 1999 struct nfsnode *np = (struct nfsnode *)0; 2000 struct vnode *newvp = (struct vnode *)0; 2001 caddr_t bpos, dpos, cp2; 2002 int error = 0, wccflag = NFSV3_WCCRATTR; 2003 int gotvp = 0; 2004 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2005 struct vattr vattr; 2006 int v3 = NFS_ISV3(dvp); 2007 2008 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 2009 return (error); 2010 } 2011 len = cnp->cn_namelen; 2012 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2013 nfsm_reqhead(dvp, NFSPROC_MKDIR, 2014 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2015 nfsm_fhtom(dvp, v3); 2016 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2017 if (v3) { 2018 nfsm_v3attrbuild(vap, FALSE); 2019 } else { 2020 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2021 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2022 sp->sa_uid = nfs_xdrneg1; 2023 sp->sa_gid = nfs_xdrneg1; 2024 sp->sa_size = nfs_xdrneg1; 2025 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2026 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2027 } 2028 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2029 if (!error) 2030 nfsm_mtofh(dvp, newvp, v3, gotvp); 2031 if (v3) 2032 nfsm_wcc_data(dvp, wccflag); 2033 m_freem(mrep); 2034 nfsmout: 2035 VTONFS(dvp)->n_flag |= NMODIFIED; 2036 if (!wccflag) 2037 VTONFS(dvp)->n_attrstamp = 0; 2038 /* 2039 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2040 * if we can succeed in looking up the directory. 2041 */ 2042 if (error == EEXIST || (!error && !gotvp)) { 2043 if (newvp) { 2044 vrele(newvp); 2045 newvp = (struct vnode *)0; 2046 } 2047 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2048 cnp->cn_td, &np); 2049 if (!error) { 2050 newvp = NFSTOV(np); 2051 if (newvp->v_type != VDIR) 2052 error = EEXIST; 2053 } 2054 } 2055 if (error) { 2056 if (newvp) 2057 vrele(newvp); 2058 } else 2059 *ap->a_vpp = newvp; 2060 return (error); 2061 } 2062 2063 /* 2064 * nfs remove directory call 2065 */ 2066 static int 2067 nfs_rmdir(ap) 2068 struct vop_rmdir_args /* { 2069 struct vnode *a_dvp; 2070 struct vnode *a_vp; 2071 struct componentname *a_cnp; 2072 } */ *ap; 2073 { 2074 struct vnode *vp = ap->a_vp; 2075 struct vnode *dvp = ap->a_dvp; 2076 struct componentname *cnp = ap->a_cnp; 2077 u_int32_t *tl; 2078 caddr_t cp; 2079 int32_t t1, t2; 2080 caddr_t bpos, dpos, cp2; 2081 int error = 0, wccflag = NFSV3_WCCRATTR; 2082 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2083 int v3 = NFS_ISV3(dvp); 2084 2085 if (dvp == vp) 2086 return (EINVAL); 2087 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2088 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2089 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2090 nfsm_fhtom(dvp, v3); 2091 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2092 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2093 if (v3) 2094 nfsm_wcc_data(dvp, wccflag); 2095 m_freem(mrep); 2096 nfsmout: 2097 VTONFS(dvp)->n_flag |= NMODIFIED; 2098 if (!wccflag) 2099 VTONFS(dvp)->n_attrstamp = 0; 2100 cache_purge(dvp); 2101 cache_purge(vp); 2102 /* 2103 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2104 */ 2105 if (error == ENOENT) 2106 error = 0; 2107 return (error); 2108 } 2109 2110 /* 2111 * nfs readdir call 2112 */ 2113 static int 2114 nfs_readdir(ap) 2115 struct vop_readdir_args /* { 2116 struct vnode *a_vp; 2117 struct uio *a_uio; 2118 struct ucred *a_cred; 2119 } */ *ap; 2120 { 2121 struct vnode *vp = ap->a_vp; 2122 struct nfsnode *np = VTONFS(vp); 2123 struct uio *uio = ap->a_uio; 2124 int tresid, error; 2125 struct vattr vattr; 2126 2127 if (vp->v_type != VDIR) 2128 return (EPERM); 2129 /* 2130 * First, check for hit on the EOF offset cache 2131 */ 2132 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2133 (np->n_flag & NMODIFIED) == 0) { 2134 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) { 2135 if (NQNFS_CKCACHABLE(vp, ND_READ)) { 2136 nfsstats.direofcache_hits++; 2137 return (0); 2138 } 2139 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 && 2140 np->n_mtime == vattr.va_mtime.tv_sec) { 2141 nfsstats.direofcache_hits++; 2142 return (0); 2143 } 2144 } 2145 2146 /* 2147 * Call nfs_bioread() to do the real work. 2148 */ 2149 tresid = uio->uio_resid; 2150 error = nfs_bioread(vp, uio, 0); 2151 2152 if (!error && uio->uio_resid == tresid) 2153 nfsstats.direofcache_misses++; 2154 return (error); 2155 } 2156 2157 /* 2158 * Readdir rpc call. 2159 * Called from below the buffer cache by nfs_doio(). 2160 */ 2161 int 2162 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2163 { 2164 int len, left; 2165 struct dirent *dp = NULL; 2166 u_int32_t *tl; 2167 caddr_t cp; 2168 int32_t t1, t2; 2169 nfsuint64 *cookiep; 2170 caddr_t bpos, dpos, cp2; 2171 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2172 nfsuint64 cookie; 2173 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2174 struct nfsnode *dnp = VTONFS(vp); 2175 u_quad_t fileno; 2176 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2177 int attrflag; 2178 int v3 = NFS_ISV3(vp); 2179 2180 #ifndef DIAGNOSTIC 2181 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2182 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2183 panic("nfs readdirrpc bad uio"); 2184 #endif 2185 2186 /* 2187 * If there is no cookie, assume directory was stale. 2188 */ 2189 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2190 if (cookiep) 2191 cookie = *cookiep; 2192 else 2193 return (NFSERR_BAD_COOKIE); 2194 /* 2195 * Loop around doing readdir rpc's of size nm_readdirsize 2196 * truncated to a multiple of DIRBLKSIZ. 2197 * The stopping criteria is EOF or buffer full. 2198 */ 2199 while (more_dirs && bigenough) { 2200 nfsstats.rpccnt[NFSPROC_READDIR]++; 2201 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2202 NFSX_READDIR(v3)); 2203 nfsm_fhtom(vp, v3); 2204 if (v3) { 2205 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2206 *tl++ = cookie.nfsuquad[0]; 2207 *tl++ = cookie.nfsuquad[1]; 2208 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2209 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2210 } else { 2211 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2212 *tl++ = cookie.nfsuquad[0]; 2213 } 2214 *tl = txdr_unsigned(nmp->nm_readdirsize); 2215 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2216 if (v3) { 2217 nfsm_postop_attr(vp, attrflag); 2218 if (!error) { 2219 nfsm_dissect(tl, u_int32_t *, 2220 2 * NFSX_UNSIGNED); 2221 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2222 dnp->n_cookieverf.nfsuquad[1] = *tl; 2223 } else { 2224 m_freem(mrep); 2225 goto nfsmout; 2226 } 2227 } 2228 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2229 more_dirs = fxdr_unsigned(int, *tl); 2230 2231 /* loop thru the dir entries, doctoring them to 4bsd form */ 2232 while (more_dirs && bigenough) { 2233 if (v3) { 2234 nfsm_dissect(tl, u_int32_t *, 2235 3 * NFSX_UNSIGNED); 2236 fileno = fxdr_hyper(tl); 2237 len = fxdr_unsigned(int, *(tl + 2)); 2238 } else { 2239 nfsm_dissect(tl, u_int32_t *, 2240 2 * NFSX_UNSIGNED); 2241 fileno = fxdr_unsigned(u_quad_t, *tl++); 2242 len = fxdr_unsigned(int, *tl); 2243 } 2244 if (len <= 0 || len > NFS_MAXNAMLEN) { 2245 error = EBADRPC; 2246 m_freem(mrep); 2247 goto nfsmout; 2248 } 2249 tlen = nfsm_rndup(len); 2250 if (tlen == len) 2251 tlen += 4; /* To ensure null termination */ 2252 left = DIRBLKSIZ - blksiz; 2253 if ((tlen + DIRHDSIZ) > left) { 2254 dp->d_reclen += left; 2255 uiop->uio_iov->iov_base += left; 2256 uiop->uio_iov->iov_len -= left; 2257 uiop->uio_offset += left; 2258 uiop->uio_resid -= left; 2259 blksiz = 0; 2260 } 2261 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2262 bigenough = 0; 2263 if (bigenough) { 2264 dp = (struct dirent *)uiop->uio_iov->iov_base; 2265 dp->d_fileno = (int)fileno; 2266 dp->d_namlen = len; 2267 dp->d_reclen = tlen + DIRHDSIZ; 2268 dp->d_type = DT_UNKNOWN; 2269 blksiz += dp->d_reclen; 2270 if (blksiz == DIRBLKSIZ) 2271 blksiz = 0; 2272 uiop->uio_offset += DIRHDSIZ; 2273 uiop->uio_resid -= DIRHDSIZ; 2274 uiop->uio_iov->iov_base += DIRHDSIZ; 2275 uiop->uio_iov->iov_len -= DIRHDSIZ; 2276 nfsm_mtouio(uiop, len); 2277 cp = uiop->uio_iov->iov_base; 2278 tlen -= len; 2279 *cp = '\0'; /* null terminate */ 2280 uiop->uio_iov->iov_base += tlen; 2281 uiop->uio_iov->iov_len -= tlen; 2282 uiop->uio_offset += tlen; 2283 uiop->uio_resid -= tlen; 2284 } else 2285 nfsm_adv(nfsm_rndup(len)); 2286 if (v3) { 2287 nfsm_dissect(tl, u_int32_t *, 2288 3 * NFSX_UNSIGNED); 2289 } else { 2290 nfsm_dissect(tl, u_int32_t *, 2291 2 * NFSX_UNSIGNED); 2292 } 2293 if (bigenough) { 2294 cookie.nfsuquad[0] = *tl++; 2295 if (v3) 2296 cookie.nfsuquad[1] = *tl++; 2297 } else if (v3) 2298 tl += 2; 2299 else 2300 tl++; 2301 more_dirs = fxdr_unsigned(int, *tl); 2302 } 2303 /* 2304 * If at end of rpc data, get the eof boolean 2305 */ 2306 if (!more_dirs) { 2307 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2308 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2309 } 2310 m_freem(mrep); 2311 } 2312 /* 2313 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2314 * by increasing d_reclen for the last record. 2315 */ 2316 if (blksiz > 0) { 2317 left = DIRBLKSIZ - blksiz; 2318 dp->d_reclen += left; 2319 uiop->uio_iov->iov_base += left; 2320 uiop->uio_iov->iov_len -= left; 2321 uiop->uio_offset += left; 2322 uiop->uio_resid -= left; 2323 } 2324 2325 /* 2326 * We are now either at the end of the directory or have filled the 2327 * block. 2328 */ 2329 if (bigenough) 2330 dnp->n_direofoffset = uiop->uio_offset; 2331 else { 2332 if (uiop->uio_resid > 0) 2333 printf("EEK! readdirrpc resid > 0\n"); 2334 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2335 *cookiep = cookie; 2336 } 2337 nfsmout: 2338 return (error); 2339 } 2340 2341 /* 2342 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2343 */ 2344 int 2345 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2346 { 2347 int len, left; 2348 struct dirent *dp; 2349 u_int32_t *tl; 2350 caddr_t cp; 2351 int32_t t1, t2; 2352 struct vnode *newvp; 2353 nfsuint64 *cookiep; 2354 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2355 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2356 struct nameidata nami, *ndp = &nami; 2357 struct componentname *cnp = &ndp->ni_cnd; 2358 nfsuint64 cookie; 2359 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2360 struct nfsnode *dnp = VTONFS(vp), *np; 2361 nfsfh_t *fhp; 2362 u_quad_t fileno; 2363 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2364 int attrflag, fhsize; 2365 2366 #ifndef nolint 2367 dp = (struct dirent *)0; 2368 #endif 2369 #ifndef DIAGNOSTIC 2370 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2371 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2372 panic("nfs readdirplusrpc bad uio"); 2373 #endif 2374 ndp->ni_dvp = vp; 2375 newvp = NULLVP; 2376 2377 /* 2378 * If there is no cookie, assume directory was stale. 2379 */ 2380 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2381 if (cookiep) 2382 cookie = *cookiep; 2383 else 2384 return (NFSERR_BAD_COOKIE); 2385 /* 2386 * Loop around doing readdir rpc's of size nm_readdirsize 2387 * truncated to a multiple of DIRBLKSIZ. 2388 * The stopping criteria is EOF or buffer full. 2389 */ 2390 while (more_dirs && bigenough) { 2391 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2392 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2393 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2394 nfsm_fhtom(vp, 1); 2395 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2396 *tl++ = cookie.nfsuquad[0]; 2397 *tl++ = cookie.nfsuquad[1]; 2398 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2399 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2400 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2401 *tl = txdr_unsigned(nmp->nm_rsize); 2402 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2403 nfsm_postop_attr(vp, attrflag); 2404 if (error) { 2405 m_freem(mrep); 2406 goto nfsmout; 2407 } 2408 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2409 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2410 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2411 more_dirs = fxdr_unsigned(int, *tl); 2412 2413 /* loop thru the dir entries, doctoring them to 4bsd form */ 2414 while (more_dirs && bigenough) { 2415 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2416 fileno = fxdr_hyper(tl); 2417 len = fxdr_unsigned(int, *(tl + 2)); 2418 if (len <= 0 || len > NFS_MAXNAMLEN) { 2419 error = EBADRPC; 2420 m_freem(mrep); 2421 goto nfsmout; 2422 } 2423 tlen = nfsm_rndup(len); 2424 if (tlen == len) 2425 tlen += 4; /* To ensure null termination*/ 2426 left = DIRBLKSIZ - blksiz; 2427 if ((tlen + DIRHDSIZ) > left) { 2428 dp->d_reclen += left; 2429 uiop->uio_iov->iov_base += left; 2430 uiop->uio_iov->iov_len -= left; 2431 uiop->uio_offset += left; 2432 uiop->uio_resid -= left; 2433 blksiz = 0; 2434 } 2435 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2436 bigenough = 0; 2437 if (bigenough) { 2438 dp = (struct dirent *)uiop->uio_iov->iov_base; 2439 dp->d_fileno = (int)fileno; 2440 dp->d_namlen = len; 2441 dp->d_reclen = tlen + DIRHDSIZ; 2442 dp->d_type = DT_UNKNOWN; 2443 blksiz += dp->d_reclen; 2444 if (blksiz == DIRBLKSIZ) 2445 blksiz = 0; 2446 uiop->uio_offset += DIRHDSIZ; 2447 uiop->uio_resid -= DIRHDSIZ; 2448 uiop->uio_iov->iov_base += DIRHDSIZ; 2449 uiop->uio_iov->iov_len -= DIRHDSIZ; 2450 cnp->cn_nameptr = uiop->uio_iov->iov_base; 2451 cnp->cn_namelen = len; 2452 nfsm_mtouio(uiop, len); 2453 cp = uiop->uio_iov->iov_base; 2454 tlen -= len; 2455 *cp = '\0'; 2456 uiop->uio_iov->iov_base += tlen; 2457 uiop->uio_iov->iov_len -= tlen; 2458 uiop->uio_offset += tlen; 2459 uiop->uio_resid -= tlen; 2460 } else 2461 nfsm_adv(nfsm_rndup(len)); 2462 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2463 if (bigenough) { 2464 cookie.nfsuquad[0] = *tl++; 2465 cookie.nfsuquad[1] = *tl++; 2466 } else 2467 tl += 2; 2468 2469 /* 2470 * Since the attributes are before the file handle 2471 * (sigh), we must skip over the attributes and then 2472 * come back and get them. 2473 */ 2474 attrflag = fxdr_unsigned(int, *tl); 2475 if (attrflag) { 2476 dpossav1 = dpos; 2477 mdsav1 = md; 2478 nfsm_adv(NFSX_V3FATTR); 2479 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2480 doit = fxdr_unsigned(int, *tl); 2481 if (doit) { 2482 nfsm_getfh(fhp, fhsize, 1); 2483 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2484 VREF(vp); 2485 newvp = vp; 2486 np = dnp; 2487 } else { 2488 error = nfs_nget(vp->v_mount, fhp, 2489 fhsize, &np); 2490 if (error) 2491 doit = 0; 2492 else 2493 newvp = NFSTOV(np); 2494 } 2495 } 2496 if (doit && bigenough) { 2497 dpossav2 = dpos; 2498 dpos = dpossav1; 2499 mdsav2 = md; 2500 md = mdsav1; 2501 nfsm_loadattr(newvp, (struct vattr *)0); 2502 dpos = dpossav2; 2503 md = mdsav2; 2504 dp->d_type = 2505 IFTODT(VTTOIF(np->n_vattr.va_type)); 2506 ndp->ni_vp = newvp; 2507 cache_enter(ndp->ni_dvp, NCPNULL, ndp->ni_vp, cnp); 2508 } 2509 } else { 2510 /* Just skip over the file handle */ 2511 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2512 i = fxdr_unsigned(int, *tl); 2513 nfsm_adv(nfsm_rndup(i)); 2514 } 2515 if (newvp != NULLVP) { 2516 if (newvp == vp) 2517 vrele(newvp); 2518 else 2519 vput(newvp); 2520 newvp = NULLVP; 2521 } 2522 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2523 more_dirs = fxdr_unsigned(int, *tl); 2524 } 2525 /* 2526 * If at end of rpc data, get the eof boolean 2527 */ 2528 if (!more_dirs) { 2529 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2530 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2531 } 2532 m_freem(mrep); 2533 } 2534 /* 2535 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2536 * by increasing d_reclen for the last record. 2537 */ 2538 if (blksiz > 0) { 2539 left = DIRBLKSIZ - blksiz; 2540 dp->d_reclen += left; 2541 uiop->uio_iov->iov_base += left; 2542 uiop->uio_iov->iov_len -= left; 2543 uiop->uio_offset += left; 2544 uiop->uio_resid -= left; 2545 } 2546 2547 /* 2548 * We are now either at the end of the directory or have filled the 2549 * block. 2550 */ 2551 if (bigenough) 2552 dnp->n_direofoffset = uiop->uio_offset; 2553 else { 2554 if (uiop->uio_resid > 0) 2555 printf("EEK! readdirplusrpc resid > 0\n"); 2556 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2557 *cookiep = cookie; 2558 } 2559 nfsmout: 2560 if (newvp != NULLVP) { 2561 if (newvp == vp) 2562 vrele(newvp); 2563 else 2564 vput(newvp); 2565 newvp = NULLVP; 2566 } 2567 return (error); 2568 } 2569 2570 /* 2571 * Silly rename. To make the NFS filesystem that is stateless look a little 2572 * more like the "ufs" a remove of an active vnode is translated to a rename 2573 * to a funny looking filename that is removed by nfs_inactive on the 2574 * nfsnode. There is the potential for another process on a different client 2575 * to create the same funny name between the nfs_lookitup() fails and the 2576 * nfs_rename() completes, but... 2577 */ 2578 static int 2579 nfs_sillyrename(dvp, vp, cnp) 2580 struct vnode *dvp, *vp; 2581 struct componentname *cnp; 2582 { 2583 struct sillyrename *sp; 2584 struct nfsnode *np; 2585 int error; 2586 2587 cache_purge(dvp); 2588 np = VTONFS(vp); 2589 #ifndef DIAGNOSTIC 2590 if (vp->v_type == VDIR) 2591 panic("nfs: sillyrename dir"); 2592 #endif 2593 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2594 M_NFSREQ, M_WAITOK); 2595 sp->s_cred = crdup(cnp->cn_cred); 2596 sp->s_dvp = dvp; 2597 VREF(dvp); 2598 2599 /* Fudge together a funny name */ 2600 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2601 2602 /* Try lookitups until we get one that isn't there */ 2603 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2604 cnp->cn_td, (struct nfsnode **)0) == 0) { 2605 sp->s_name[4]++; 2606 if (sp->s_name[4] > 'z') { 2607 error = EINVAL; 2608 goto bad; 2609 } 2610 } 2611 error = nfs_renameit(dvp, cnp, sp); 2612 if (error) 2613 goto bad; 2614 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2615 cnp->cn_td, &np); 2616 np->n_sillyrename = sp; 2617 return (0); 2618 bad: 2619 vrele(sp->s_dvp); 2620 crfree(sp->s_cred); 2621 free((caddr_t)sp, M_NFSREQ); 2622 return (error); 2623 } 2624 2625 /* 2626 * Look up a file name and optionally either update the file handle or 2627 * allocate an nfsnode, depending on the value of npp. 2628 * npp == NULL --> just do the lookup 2629 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2630 * handled too 2631 * *npp != NULL --> update the file handle in the vnode 2632 */ 2633 static int 2634 nfs_lookitup(dvp, name, len, cred, td, npp) 2635 struct vnode *dvp; 2636 const char *name; 2637 int len; 2638 struct ucred *cred; 2639 struct thread *td; 2640 struct nfsnode **npp; 2641 { 2642 u_int32_t *tl; 2643 caddr_t cp; 2644 int32_t t1, t2; 2645 struct vnode *newvp = (struct vnode *)0; 2646 struct nfsnode *np, *dnp = VTONFS(dvp); 2647 caddr_t bpos, dpos, cp2; 2648 int error = 0, fhlen, attrflag; 2649 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2650 nfsfh_t *nfhp; 2651 int v3 = NFS_ISV3(dvp); 2652 2653 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2654 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2655 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2656 nfsm_fhtom(dvp, v3); 2657 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2658 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2659 if (npp && !error) { 2660 nfsm_getfh(nfhp, fhlen, v3); 2661 if (*npp) { 2662 np = *npp; 2663 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2664 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2665 np->n_fhp = &np->n_fh; 2666 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2667 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2668 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2669 np->n_fhsize = fhlen; 2670 newvp = NFSTOV(np); 2671 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2672 VREF(dvp); 2673 newvp = dvp; 2674 } else { 2675 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2676 if (error) { 2677 m_freem(mrep); 2678 return (error); 2679 } 2680 newvp = NFSTOV(np); 2681 } 2682 if (v3) { 2683 nfsm_postop_attr(newvp, attrflag); 2684 if (!attrflag && *npp == NULL) { 2685 m_freem(mrep); 2686 if (newvp == dvp) 2687 vrele(newvp); 2688 else 2689 vput(newvp); 2690 return (ENOENT); 2691 } 2692 } else 2693 nfsm_loadattr(newvp, (struct vattr *)0); 2694 } 2695 m_freem(mrep); 2696 nfsmout: 2697 if (npp && *npp == NULL) { 2698 if (error) { 2699 if (newvp) { 2700 if (newvp == dvp) 2701 vrele(newvp); 2702 else 2703 vput(newvp); 2704 } 2705 } else 2706 *npp = np; 2707 } 2708 return (error); 2709 } 2710 2711 /* 2712 * Nfs Version 3 commit rpc 2713 */ 2714 int 2715 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2716 { 2717 caddr_t cp; 2718 u_int32_t *tl; 2719 int32_t t1, t2; 2720 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2721 caddr_t bpos, dpos, cp2; 2722 int error = 0, wccflag = NFSV3_WCCRATTR; 2723 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2724 2725 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2726 return (0); 2727 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2728 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2729 nfsm_fhtom(vp, 1); 2730 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2731 txdr_hyper(offset, tl); 2732 tl += 2; 2733 *tl = txdr_unsigned(cnt); 2734 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2735 nfsm_wcc_data(vp, wccflag); 2736 if (!error) { 2737 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2738 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2739 NFSX_V3WRITEVERF)) { 2740 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2741 NFSX_V3WRITEVERF); 2742 error = NFSERR_STALEWRITEVERF; 2743 } 2744 } 2745 m_freem(mrep); 2746 nfsmout: 2747 return (error); 2748 } 2749 2750 /* 2751 * Kludge City.. 2752 * - make nfs_bmap() essentially a no-op that does no translation 2753 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2754 * (Maybe I could use the process's page mapping, but I was concerned that 2755 * Kernel Write might not be enabled and also figured copyout() would do 2756 * a lot more work than bcopy() and also it currently happens in the 2757 * context of the swapper process (2). 2758 */ 2759 static int 2760 nfs_bmap(ap) 2761 struct vop_bmap_args /* { 2762 struct vnode *a_vp; 2763 daddr_t a_bn; 2764 struct vnode **a_vpp; 2765 daddr_t *a_bnp; 2766 int *a_runp; 2767 int *a_runb; 2768 } */ *ap; 2769 { 2770 struct vnode *vp = ap->a_vp; 2771 2772 if (ap->a_vpp != NULL) 2773 *ap->a_vpp = vp; 2774 if (ap->a_bnp != NULL) 2775 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize); 2776 if (ap->a_runp != NULL) 2777 *ap->a_runp = 0; 2778 if (ap->a_runb != NULL) 2779 *ap->a_runb = 0; 2780 return (0); 2781 } 2782 2783 /* 2784 * Strategy routine. 2785 * For async requests when nfsiod(s) are running, queue the request by 2786 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2787 * request. 2788 */ 2789 static int 2790 nfs_strategy(ap) 2791 struct vop_strategy_args *ap; 2792 { 2793 struct buf *bp = ap->a_bp; 2794 struct thread *td; 2795 int error = 0; 2796 2797 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2798 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp)); 2799 2800 if (bp->b_flags & B_PHYS) 2801 panic("nfs physio"); 2802 2803 if (bp->b_flags & B_ASYNC) 2804 td = NULL; 2805 else 2806 td = curthread; /* XXX */ 2807 2808 /* 2809 * If the op is asynchronous and an i/o daemon is waiting 2810 * queue the request, wake it up and wait for completion 2811 * otherwise just do it ourselves. 2812 */ 2813 if ((bp->b_flags & B_ASYNC) == 0 || 2814 nfs_asyncio(bp, td)) 2815 error = nfs_doio(bp, td); 2816 return (error); 2817 } 2818 2819 /* 2820 * Mmap a file 2821 * 2822 * NB Currently unsupported. 2823 */ 2824 /* ARGSUSED */ 2825 static int 2826 nfs_mmap(ap) 2827 struct vop_mmap_args /* { 2828 struct vnode *a_vp; 2829 int a_fflags; 2830 struct ucred *a_cred; 2831 struct thread *a_td; 2832 } */ *ap; 2833 { 2834 2835 return (EINVAL); 2836 } 2837 2838 /* 2839 * fsync vnode op. Just call nfs_flush() with commit == 1. 2840 */ 2841 /* ARGSUSED */ 2842 static int 2843 nfs_fsync(ap) 2844 struct vop_fsync_args /* { 2845 struct vnodeop_desc *a_desc; 2846 struct vnode * a_vp; 2847 struct ucred * a_cred; 2848 int a_waitfor; 2849 struct thread * a_td; 2850 } */ *ap; 2851 { 2852 2853 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1)); 2854 } 2855 2856 /* 2857 * Flush all the blocks associated with a vnode. 2858 * Walk through the buffer pool and push any dirty pages 2859 * associated with the vnode. 2860 */ 2861 static int 2862 nfs_flush(vp, waitfor, td, commit) 2863 struct vnode *vp; 2864 int waitfor; 2865 struct thread *td; 2866 int commit; 2867 { 2868 struct nfsnode *np = VTONFS(vp); 2869 struct buf *bp; 2870 int i; 2871 struct buf *nbp; 2872 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2873 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; 2874 int passone = 1; 2875 u_quad_t off, endoff, toff; 2876 struct buf **bvec = NULL; 2877 #ifndef NFS_COMMITBVECSIZ 2878 #define NFS_COMMITBVECSIZ 20 2879 #endif 2880 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; 2881 int bvecsize = 0, bveccount; 2882 2883 if (nmp->nm_flag & NFSMNT_INT) 2884 slpflag = PCATCH; 2885 if (!commit) 2886 passone = 0; 2887 /* 2888 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the 2889 * server, but nas not been committed to stable storage on the server 2890 * yet. On the first pass, the byte range is worked out and the commit 2891 * rpc is done. On the second pass, nfs_writebp() is called to do the 2892 * job. 2893 */ 2894 again: 2895 off = (u_quad_t)-1; 2896 endoff = 0; 2897 bvecpos = 0; 2898 if (NFS_ISV3(vp) && commit) { 2899 s = splbio(); 2900 /* 2901 * Count up how many buffers waiting for a commit. 2902 */ 2903 bveccount = 0; 2904 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2905 nbp = TAILQ_NEXT(bp, b_vnbufs); 2906 if (BUF_REFCNT(bp) == 0 && 2907 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) 2908 == (B_DELWRI | B_NEEDCOMMIT)) 2909 bveccount++; 2910 } 2911 /* 2912 * Allocate space to remember the list of bufs to commit. It is 2913 * important to use M_NOWAIT here to avoid a race with nfs_write. 2914 * If we can't get memory (for whatever reason), we will end up 2915 * committing the buffers one-by-one in the loop below. 2916 */ 2917 if (bvec != NULL && bvec != bvec_on_stack) 2918 free(bvec, M_TEMP); 2919 if (bveccount > NFS_COMMITBVECSIZ) { 2920 bvec = (struct buf **) 2921 malloc(bveccount * sizeof(struct buf *), 2922 M_TEMP, M_NOWAIT); 2923 if (bvec == NULL) { 2924 bvec = bvec_on_stack; 2925 bvecsize = NFS_COMMITBVECSIZ; 2926 } else 2927 bvecsize = bveccount; 2928 } else { 2929 bvec = bvec_on_stack; 2930 bvecsize = NFS_COMMITBVECSIZ; 2931 } 2932 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 2933 nbp = TAILQ_NEXT(bp, b_vnbufs); 2934 if (bvecpos >= bvecsize) 2935 break; 2936 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 2937 (B_DELWRI | B_NEEDCOMMIT) || 2938 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 2939 continue; 2940 bremfree(bp); 2941 /* 2942 * NOTE: we are not clearing B_DONE here, so we have 2943 * to do it later on in this routine if we intend to 2944 * initiate I/O on the bp. 2945 * 2946 * Note: to avoid loopback deadlocks, we do not 2947 * assign b_runningbufspace. 2948 */ 2949 bp->b_flags |= B_WRITEINPROG; 2950 vfs_busy_pages(bp, 1); 2951 2952 /* 2953 * bp is protected by being locked, but nbp is not 2954 * and vfs_busy_pages() may sleep. We have to 2955 * recalculate nbp. 2956 */ 2957 nbp = TAILQ_NEXT(bp, b_vnbufs); 2958 2959 /* 2960 * A list of these buffers is kept so that the 2961 * second loop knows which buffers have actually 2962 * been committed. This is necessary, since there 2963 * may be a race between the commit rpc and new 2964 * uncommitted writes on the file. 2965 */ 2966 bvec[bvecpos++] = bp; 2967 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 2968 bp->b_dirtyoff; 2969 if (toff < off) 2970 off = toff; 2971 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 2972 if (toff > endoff) 2973 endoff = toff; 2974 } 2975 splx(s); 2976 } 2977 if (bvecpos > 0) { 2978 /* 2979 * Commit data on the server, as required. Note that 2980 * nfs_commit will use the vnode's cred for the commit. 2981 */ 2982 retv = nfs_commit(vp, off, (int)(endoff - off), td); 2983 2984 if (retv == NFSERR_STALEWRITEVERF) 2985 nfs_clearcommit(vp->v_mount); 2986 2987 /* 2988 * Now, either mark the blocks I/O done or mark the 2989 * blocks dirty, depending on whether the commit 2990 * succeeded. 2991 */ 2992 for (i = 0; i < bvecpos; i++) { 2993 bp = bvec[i]; 2994 bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK); 2995 if (retv) { 2996 /* 2997 * Error, leave B_DELWRI intact 2998 */ 2999 vfs_unbusy_pages(bp); 3000 brelse(bp); 3001 } else { 3002 /* 3003 * Success, remove B_DELWRI ( bundirty() ). 3004 * 3005 * b_dirtyoff/b_dirtyend seem to be NFS 3006 * specific. We should probably move that 3007 * into bundirty(). XXX 3008 */ 3009 s = splbio(); 3010 vp->v_numoutput++; 3011 bp->b_flags |= B_ASYNC; 3012 bundirty(bp); 3013 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3014 bp->b_dirtyoff = bp->b_dirtyend = 0; 3015 splx(s); 3016 biodone(bp); 3017 } 3018 } 3019 } 3020 3021 /* 3022 * Start/do any write(s) that are required. 3023 */ 3024 loop: 3025 s = splbio(); 3026 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 3027 nbp = TAILQ_NEXT(bp, b_vnbufs); 3028 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 3029 if (waitfor != MNT_WAIT || passone) 3030 continue; 3031 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL, 3032 "nfsfsync", slpflag, slptimeo); 3033 splx(s); 3034 if (error == 0) 3035 panic("nfs_fsync: inconsistent lock"); 3036 if (error == ENOLCK) 3037 goto loop; 3038 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 3039 error = EINTR; 3040 goto done; 3041 } 3042 if (slpflag == PCATCH) { 3043 slpflag = 0; 3044 slptimeo = 2 * hz; 3045 } 3046 goto loop; 3047 } 3048 if ((bp->b_flags & B_DELWRI) == 0) 3049 panic("nfs_fsync: not dirty"); 3050 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { 3051 BUF_UNLOCK(bp); 3052 continue; 3053 } 3054 bremfree(bp); 3055 if (passone || !commit) 3056 bp->b_flags |= B_ASYNC; 3057 else 3058 bp->b_flags |= B_ASYNC | B_WRITEINPROG; 3059 splx(s); 3060 VOP_BWRITE(bp->b_vp, bp); 3061 goto loop; 3062 } 3063 splx(s); 3064 if (passone) { 3065 passone = 0; 3066 goto again; 3067 } 3068 if (waitfor == MNT_WAIT) { 3069 while (vp->v_numoutput) { 3070 vp->v_flag |= VBWAIT; 3071 error = tsleep((caddr_t)&vp->v_numoutput, 3072 slpflag, "nfsfsync", slptimeo); 3073 if (error) { 3074 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 3075 error = EINTR; 3076 goto done; 3077 } 3078 if (slpflag == PCATCH) { 3079 slpflag = 0; 3080 slptimeo = 2 * hz; 3081 } 3082 } 3083 } 3084 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) { 3085 goto loop; 3086 } 3087 } 3088 if (np->n_flag & NWRITEERR) { 3089 error = np->n_error; 3090 np->n_flag &= ~NWRITEERR; 3091 } 3092 done: 3093 if (bvec != NULL && bvec != bvec_on_stack) 3094 free(bvec, M_TEMP); 3095 return (error); 3096 } 3097 3098 /* 3099 * NFS advisory byte-level locks. 3100 * Currently unsupported. 3101 */ 3102 static int 3103 nfs_advlock(ap) 3104 struct vop_advlock_args /* { 3105 struct vnode *a_vp; 3106 caddr_t a_id; 3107 int a_op; 3108 struct flock *a_fl; 3109 int a_flags; 3110 } */ *ap; 3111 { 3112 struct nfsnode *np = VTONFS(ap->a_vp); 3113 3114 /* 3115 * The following kludge is to allow diskless support to work 3116 * until a real NFS lockd is implemented. Basically, just pretend 3117 * that this is a local lock. 3118 */ 3119 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3120 } 3121 3122 /* 3123 * Print out the contents of an nfsnode. 3124 */ 3125 static int 3126 nfs_print(ap) 3127 struct vop_print_args /* { 3128 struct vnode *a_vp; 3129 } */ *ap; 3130 { 3131 struct vnode *vp = ap->a_vp; 3132 struct nfsnode *np = VTONFS(vp); 3133 3134 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3135 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3136 if (vp->v_type == VFIFO) 3137 fifo_printinfo(vp); 3138 printf("\n"); 3139 return (0); 3140 } 3141 3142 /* 3143 * Just call nfs_writebp() with the force argument set to 1. 3144 * 3145 * NOTE: B_DONE may or may not be set in a_bp on call. 3146 */ 3147 static int 3148 nfs_bwrite(ap) 3149 struct vop_bwrite_args /* { 3150 struct vnode *a_bp; 3151 } */ *ap; 3152 { 3153 return (nfs_writebp(ap->a_bp, 1, curthread)); 3154 } 3155 3156 /* 3157 * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless 3158 * the force flag is one and it also handles the B_NEEDCOMMIT flag. We set 3159 * B_CACHE if this is a VMIO buffer. 3160 */ 3161 int 3162 nfs_writebp(bp, force, td) 3163 struct buf *bp; 3164 int force; 3165 struct thread *td; 3166 { 3167 int s; 3168 int oldflags = bp->b_flags; 3169 #if 0 3170 int retv = 1; 3171 off_t off; 3172 #endif 3173 3174 if (BUF_REFCNT(bp) == 0) 3175 panic("bwrite: buffer is not locked???"); 3176 3177 if (bp->b_flags & B_INVAL) { 3178 brelse(bp); 3179 return(0); 3180 } 3181 3182 bp->b_flags |= B_CACHE; 3183 3184 /* 3185 * Undirty the bp. We will redirty it later if the I/O fails. 3186 */ 3187 3188 s = splbio(); 3189 bundirty(bp); 3190 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3191 3192 bp->b_vp->v_numoutput++; 3193 splx(s); 3194 3195 /* 3196 * Note: to avoid loopback deadlocks, we do not 3197 * assign b_runningbufspace. 3198 */ 3199 vfs_busy_pages(bp, 1); 3200 3201 if (force) 3202 bp->b_flags |= B_WRITEINPROG; 3203 BUF_KERNPROC(bp); 3204 VOP_STRATEGY(bp->b_vp, bp); 3205 3206 if( (oldflags & B_ASYNC) == 0) { 3207 int rtval = biowait(bp); 3208 3209 if (oldflags & B_DELWRI) { 3210 s = splbio(); 3211 reassignbuf(bp, bp->b_vp); 3212 splx(s); 3213 } 3214 3215 brelse(bp); 3216 return (rtval); 3217 } 3218 3219 return (0); 3220 } 3221 3222 /* 3223 * nfs special file access vnode op. 3224 * Essentially just get vattr and then imitate iaccess() since the device is 3225 * local to the client. 3226 */ 3227 static int 3228 nfsspec_access(ap) 3229 struct vop_access_args /* { 3230 struct vnode *a_vp; 3231 int a_mode; 3232 struct ucred *a_cred; 3233 struct thread *a_td; 3234 } */ *ap; 3235 { 3236 struct vattr *vap; 3237 gid_t *gp; 3238 struct ucred *cred = ap->a_cred; 3239 struct vnode *vp = ap->a_vp; 3240 mode_t mode = ap->a_mode; 3241 struct vattr vattr; 3242 int i; 3243 int error; 3244 3245 /* 3246 * Disallow write attempts on filesystems mounted read-only; 3247 * unless the file is a socket, fifo, or a block or character 3248 * device resident on the filesystem. 3249 */ 3250 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3251 switch (vp->v_type) { 3252 case VREG: 3253 case VDIR: 3254 case VLNK: 3255 return (EROFS); 3256 default: 3257 break; 3258 } 3259 } 3260 /* 3261 * If you're the super-user, 3262 * you always get access. 3263 */ 3264 if (cred->cr_uid == 0) 3265 return (0); 3266 vap = &vattr; 3267 error = VOP_GETATTR(vp, vap, ap->a_td); 3268 if (error) 3269 return (error); 3270 /* 3271 * Access check is based on only one of owner, group, public. 3272 * If not owner, then check group. If not a member of the 3273 * group, then check public access. 3274 */ 3275 if (cred->cr_uid != vap->va_uid) { 3276 mode >>= 3; 3277 gp = cred->cr_groups; 3278 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3279 if (vap->va_gid == *gp) 3280 goto found; 3281 mode >>= 3; 3282 found: 3283 ; 3284 } 3285 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3286 return (error); 3287 } 3288 3289 /* 3290 * Read wrapper for special devices. 3291 */ 3292 static int 3293 nfsspec_read(ap) 3294 struct vop_read_args /* { 3295 struct vnode *a_vp; 3296 struct uio *a_uio; 3297 int a_ioflag; 3298 struct ucred *a_cred; 3299 } */ *ap; 3300 { 3301 struct nfsnode *np = VTONFS(ap->a_vp); 3302 3303 /* 3304 * Set access flag. 3305 */ 3306 np->n_flag |= NACC; 3307 getnanotime(&np->n_atim); 3308 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap)); 3309 } 3310 3311 /* 3312 * Write wrapper for special devices. 3313 */ 3314 static int 3315 nfsspec_write(ap) 3316 struct vop_write_args /* { 3317 struct vnode *a_vp; 3318 struct uio *a_uio; 3319 int a_ioflag; 3320 struct ucred *a_cred; 3321 } */ *ap; 3322 { 3323 struct nfsnode *np = VTONFS(ap->a_vp); 3324 3325 /* 3326 * Set update flag. 3327 */ 3328 np->n_flag |= NUPD; 3329 getnanotime(&np->n_mtim); 3330 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap)); 3331 } 3332 3333 /* 3334 * Close wrapper for special devices. 3335 * 3336 * Update the times on the nfsnode then do device close. 3337 */ 3338 static int 3339 nfsspec_close(ap) 3340 struct vop_close_args /* { 3341 struct vnode *a_vp; 3342 int a_fflag; 3343 struct ucred *a_cred; 3344 struct thread *a_td; 3345 } */ *ap; 3346 { 3347 struct vnode *vp = ap->a_vp; 3348 struct nfsnode *np = VTONFS(vp); 3349 struct vattr vattr; 3350 3351 if (np->n_flag & (NACC | NUPD)) { 3352 np->n_flag |= NCHG; 3353 if (vp->v_usecount == 1 && 3354 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3355 VATTR_NULL(&vattr); 3356 if (np->n_flag & NACC) 3357 vattr.va_atime = np->n_atim; 3358 if (np->n_flag & NUPD) 3359 vattr.va_mtime = np->n_mtim; 3360 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3361 } 3362 } 3363 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap)); 3364 } 3365 3366 /* 3367 * Read wrapper for fifos. 3368 */ 3369 static int 3370 nfsfifo_read(ap) 3371 struct vop_read_args /* { 3372 struct vnode *a_vp; 3373 struct uio *a_uio; 3374 int a_ioflag; 3375 struct ucred *a_cred; 3376 } */ *ap; 3377 { 3378 struct nfsnode *np = VTONFS(ap->a_vp); 3379 3380 /* 3381 * Set access flag. 3382 */ 3383 np->n_flag |= NACC; 3384 getnanotime(&np->n_atim); 3385 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap)); 3386 } 3387 3388 /* 3389 * Write wrapper for fifos. 3390 */ 3391 static int 3392 nfsfifo_write(ap) 3393 struct vop_write_args /* { 3394 struct vnode *a_vp; 3395 struct uio *a_uio; 3396 int a_ioflag; 3397 struct ucred *a_cred; 3398 } */ *ap; 3399 { 3400 struct nfsnode *np = VTONFS(ap->a_vp); 3401 3402 /* 3403 * Set update flag. 3404 */ 3405 np->n_flag |= NUPD; 3406 getnanotime(&np->n_mtim); 3407 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap)); 3408 } 3409 3410 /* 3411 * Close wrapper for fifos. 3412 * 3413 * Update the times on the nfsnode then do fifo close. 3414 */ 3415 static int 3416 nfsfifo_close(ap) 3417 struct vop_close_args /* { 3418 struct vnode *a_vp; 3419 int a_fflag; 3420 struct thread *a_td; 3421 } */ *ap; 3422 { 3423 struct vnode *vp = ap->a_vp; 3424 struct nfsnode *np = VTONFS(vp); 3425 struct vattr vattr; 3426 struct timespec ts; 3427 3428 if (np->n_flag & (NACC | NUPD)) { 3429 getnanotime(&ts); 3430 if (np->n_flag & NACC) 3431 np->n_atim = ts; 3432 if (np->n_flag & NUPD) 3433 np->n_mtim = ts; 3434 np->n_flag |= NCHG; 3435 if (vp->v_usecount == 1 && 3436 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3437 VATTR_NULL(&vattr); 3438 if (np->n_flag & NACC) 3439 vattr.va_atime = np->n_atim; 3440 if (np->n_flag & NUPD) 3441 vattr.va_mtime = np->n_mtim; 3442 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3443 } 3444 } 3445 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap)); 3446 } 3447 3448