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 1.40 2005/04/15 19:08:21 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/nlookup.h> 59 #include <sys/socket.h> 60 #include <sys/vnode.h> 61 #include <sys/dirent.h> 62 #include <sys/fcntl.h> 63 #include <sys/lockf.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/conf.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_zone.h> 71 72 #include <sys/buf2.h> 73 74 #include <vfs/fifofs/fifo.h> 75 76 #include "rpcv2.h" 77 #include "nfsproto.h" 78 #include "nfs.h" 79 #include "nfsmount.h" 80 #include "nfsnode.h" 81 #include "xdr_subs.h" 82 #include "nfsm_subs.h" 83 #include "nqnfs.h" 84 85 #include <net/if.h> 86 #include <netinet/in.h> 87 #include <netinet/in_var.h> 88 89 /* Defs */ 90 #define TRUE 1 91 #define FALSE 0 92 93 /* 94 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 95 * calls are not in getblk() and brelse() so that they would not be necessary 96 * here. 97 */ 98 #ifndef B_VMIO 99 #define vfs_busy_pages(bp, f) 100 #endif 101 102 static int nfsspec_read (struct vop_read_args *); 103 static int nfsspec_write (struct vop_write_args *); 104 static int nfsfifo_read (struct vop_read_args *); 105 static int nfsfifo_write (struct vop_write_args *); 106 static int nfsspec_close (struct vop_close_args *); 107 static int nfsfifo_close (struct vop_close_args *); 108 #define nfs_poll vop_nopoll 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 static int nfs_nresolve (struct vop_nresolve_args *); 140 /* 141 * Global vfs data structures for nfs 142 */ 143 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 144 { &vop_default_desc, vop_defaultop }, 145 { &vop_access_desc, (vnodeopv_entry_t) nfs_access }, 146 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock }, 147 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap }, 148 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite }, 149 { &vop_close_desc, (vnodeopv_entry_t) nfs_close }, 150 { &vop_create_desc, (vnodeopv_entry_t) nfs_create }, 151 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 152 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 153 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages }, 154 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages }, 155 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 156 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 157 { &vop_lease_desc, vop_null }, 158 { &vop_link_desc, (vnodeopv_entry_t) nfs_link }, 159 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 160 { &vop_lookup_desc, (vnodeopv_entry_t) nfs_lookup }, 161 { &vop_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir }, 162 { &vop_mknod_desc, (vnodeopv_entry_t) nfs_mknod }, 163 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap }, 164 { &vop_open_desc, (vnodeopv_entry_t) nfs_open }, 165 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll }, 166 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 167 { &vop_read_desc, (vnodeopv_entry_t) nfs_read }, 168 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir }, 169 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink }, 170 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 171 { &vop_remove_desc, (vnodeopv_entry_t) nfs_remove }, 172 { &vop_rename_desc, (vnodeopv_entry_t) nfs_rename }, 173 { &vop_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir }, 174 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 175 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy }, 176 { &vop_symlink_desc, (vnodeopv_entry_t) nfs_symlink }, 177 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 178 { &vop_write_desc, (vnodeopv_entry_t) nfs_write }, 179 180 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve }, 181 { NULL, NULL } 182 }; 183 184 /* 185 * Special device vnode ops 186 */ 187 struct vnodeopv_entry_desc nfsv2_specop_entries[] = { 188 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate }, 189 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 190 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close }, 191 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 192 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 193 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 194 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 195 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 196 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 197 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read }, 198 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 199 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 200 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 201 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write }, 202 { NULL, NULL } 203 }; 204 205 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = { 206 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate }, 207 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 208 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close }, 209 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 210 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 211 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 212 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 213 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 214 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 215 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read }, 216 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 217 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 218 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 219 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write }, 220 { NULL, NULL } 221 }; 222 223 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 224 struct componentname *cnp, 225 struct vattr *vap); 226 static int nfs_removerpc (struct vnode *dvp, const char *name, 227 int namelen, 228 struct ucred *cred, struct thread *td); 229 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 230 int fnamelen, struct vnode *tdvp, 231 const char *tnameptr, int tnamelen, 232 struct ucred *cred, struct thread *td); 233 static int nfs_renameit (struct vnode *sdvp, 234 struct componentname *scnp, 235 struct sillyrename *sp); 236 237 /* 238 * Global variables 239 */ 240 extern u_int32_t nfs_true, nfs_false; 241 extern u_int32_t nfs_xdrneg1; 242 extern struct nfsstats nfsstats; 243 extern nfstype nfsv3_type[9]; 244 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 245 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 246 int nfs_numasync = 0; 247 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) 248 249 SYSCTL_DECL(_vfs_nfs); 250 251 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 252 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 253 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 254 255 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 256 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 257 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 258 259 static int nfsv3_commit_on_close = 0; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 261 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 262 #if 0 263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 264 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 265 266 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 267 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 268 #endif 269 270 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 271 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 272 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 273 static int 274 nfs3_access_otw(struct vnode *vp, int wmode, 275 struct thread *td, struct ucred *cred) 276 { 277 const int v3 = 1; 278 u_int32_t *tl; 279 int error = 0, attrflag; 280 281 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 282 caddr_t bpos, dpos, cp2; 283 int32_t t1, t2; 284 caddr_t cp; 285 u_int32_t rmode; 286 struct nfsnode *np = VTONFS(vp); 287 288 nfsstats.rpccnt[NFSPROC_ACCESS]++; 289 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 290 nfsm_fhtom(vp, v3); 291 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 292 *tl = txdr_unsigned(wmode); 293 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 294 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 295 if (!error) { 296 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 297 rmode = fxdr_unsigned(u_int32_t, *tl); 298 np->n_mode = rmode; 299 np->n_modeuid = cred->cr_uid; 300 np->n_modestamp = mycpu->gd_time_seconds; 301 } 302 m_freem(mrep); 303 nfsmout: 304 return error; 305 } 306 307 /* 308 * nfs access vnode op. 309 * For nfs version 2, just return ok. File accesses may fail later. 310 * For nfs version 3, use the access rpc to check accessibility. If file modes 311 * are changed on the server, accesses might still fail later. 312 * 313 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 314 * struct thread *a_td) 315 */ 316 static int 317 nfs_access(struct vop_access_args *ap) 318 { 319 struct vnode *vp = ap->a_vp; 320 int error = 0; 321 u_int32_t mode, wmode; 322 int v3 = NFS_ISV3(vp); 323 struct nfsnode *np = VTONFS(vp); 324 325 /* 326 * Disallow write attempts on filesystems mounted read-only; 327 * unless the file is a socket, fifo, or a block or character 328 * device resident on the filesystem. 329 */ 330 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 331 switch (vp->v_type) { 332 case VREG: 333 case VDIR: 334 case VLNK: 335 return (EROFS); 336 default: 337 break; 338 } 339 } 340 /* 341 * For nfs v3, check to see if we have done this recently, and if 342 * so return our cached result instead of making an ACCESS call. 343 * If not, do an access rpc, otherwise you are stuck emulating 344 * ufs_access() locally using the vattr. This may not be correct, 345 * since the server may apply other access criteria such as 346 * client uid-->server uid mapping that we do not know about. 347 */ 348 if (v3) { 349 if (ap->a_mode & VREAD) 350 mode = NFSV3ACCESS_READ; 351 else 352 mode = 0; 353 if (vp->v_type != VDIR) { 354 if (ap->a_mode & VWRITE) 355 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 356 if (ap->a_mode & VEXEC) 357 mode |= NFSV3ACCESS_EXECUTE; 358 } else { 359 if (ap->a_mode & VWRITE) 360 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 361 NFSV3ACCESS_DELETE); 362 if (ap->a_mode & VEXEC) 363 mode |= NFSV3ACCESS_LOOKUP; 364 } 365 /* XXX safety belt, only make blanket request if caching */ 366 if (nfsaccess_cache_timeout > 0) { 367 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 368 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 369 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 370 } else { 371 wmode = mode; 372 } 373 374 /* 375 * Does our cached result allow us to give a definite yes to 376 * this request? 377 */ 378 if (np->n_modestamp && 379 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 380 (ap->a_cred->cr_uid == np->n_modeuid) && 381 ((np->n_mode & mode) == mode)) { 382 nfsstats.accesscache_hits++; 383 } else { 384 /* 385 * Either a no, or a don't know. Go to the wire. 386 */ 387 nfsstats.accesscache_misses++; 388 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 389 if (!error) { 390 if ((np->n_mode & mode) != mode) { 391 error = EACCES; 392 } 393 } 394 } 395 } else { 396 if ((error = nfsspec_access(ap)) != 0) 397 return (error); 398 399 /* 400 * Attempt to prevent a mapped root from accessing a file 401 * which it shouldn't. We try to read a byte from the file 402 * if the user is root and the file is not zero length. 403 * After calling nfsspec_access, we should have the correct 404 * file size cached. 405 */ 406 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 407 && VTONFS(vp)->n_size > 0) { 408 struct iovec aiov; 409 struct uio auio; 410 char buf[1]; 411 412 aiov.iov_base = buf; 413 aiov.iov_len = 1; 414 auio.uio_iov = &aiov; 415 auio.uio_iovcnt = 1; 416 auio.uio_offset = 0; 417 auio.uio_resid = 1; 418 auio.uio_segflg = UIO_SYSSPACE; 419 auio.uio_rw = UIO_READ; 420 auio.uio_td = ap->a_td; 421 422 if (vp->v_type == VREG) { 423 error = nfs_readrpc(vp, &auio); 424 } else if (vp->v_type == VDIR) { 425 char* bp; 426 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 427 aiov.iov_base = bp; 428 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 429 error = nfs_readdirrpc(vp, &auio); 430 free(bp, M_TEMP); 431 } else if (vp->v_type == VLNK) { 432 error = nfs_readlinkrpc(vp, &auio); 433 } else { 434 error = EACCES; 435 } 436 } 437 } 438 /* 439 * [re]record creds for reading and/or writing if access 440 * was granted. Assume the NFS server will grant read access 441 * for execute requests. 442 */ 443 if (error == 0) { 444 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 445 crhold(ap->a_cred); 446 if (np->n_rucred) 447 crfree(np->n_rucred); 448 np->n_rucred = ap->a_cred; 449 } 450 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 451 crhold(ap->a_cred); 452 if (np->n_wucred) 453 crfree(np->n_wucred); 454 np->n_wucred = ap->a_cred; 455 } 456 } 457 return(error); 458 } 459 460 /* 461 * nfs open vnode op 462 * Check to see if the type is ok 463 * and that deletion is not in progress. 464 * For paged in text files, you will need to flush the page cache 465 * if consistency is lost. 466 * 467 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 468 * struct thread *a_td) 469 */ 470 /* ARGSUSED */ 471 static int 472 nfs_open(struct vop_open_args *ap) 473 { 474 struct vnode *vp = ap->a_vp; 475 struct nfsnode *np = VTONFS(vp); 476 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 477 struct vattr vattr; 478 int error; 479 480 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 481 #ifdef DIAGNOSTIC 482 printf("open eacces vtyp=%d\n",vp->v_type); 483 #endif 484 return (EOPNOTSUPP); 485 } 486 487 /* 488 * Clear the attribute cache only if opening with write access. It 489 * is unclear if we should do this at all here, but we certainly 490 * should not clear the cache unconditionally simply because a file 491 * is being opened. 492 */ 493 if (ap->a_mode & FWRITE) 494 np->n_attrstamp = 0; 495 496 if (nmp->nm_flag & NFSMNT_NQNFS) { 497 /* 498 * If NQNFS is active, get a valid lease 499 */ 500 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 501 do { 502 error = nqnfs_getlease(vp, ND_READ, ap->a_td); 503 } while (error == NQNFS_EXPIRED); 504 if (error) 505 return (error); 506 if (np->n_lrev != np->n_brev || 507 (np->n_flag & NQNFSNONCACHE)) { 508 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 509 == EINTR) { 510 return (error); 511 } 512 np->n_brev = np->n_lrev; 513 } 514 } 515 } else { 516 /* 517 * For normal NFS, reconcile changes made locally verses 518 * changes made remotely. Note that VOP_GETATTR only goes 519 * to the wire if the cached attribute has timed out or been 520 * cleared. 521 * 522 * If local modifications have been made clear the attribute 523 * cache to force an attribute and modified time check. If 524 * GETATTR detects that the file has been changed by someone 525 * other then us it will set NRMODIFIED. 526 * 527 * If we are opening a directory and local changes have been 528 * made we have to invalidate the cache in order to ensure 529 * that we get the most up-to-date information from the 530 * server. XXX 531 */ 532 if (np->n_flag & NLMODIFIED) { 533 np->n_attrstamp = 0; 534 if (vp->v_type == VDIR) { 535 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 536 if (error == EINTR) 537 return (error); 538 nfs_invaldir(vp); 539 } 540 } 541 error = VOP_GETATTR(vp, &vattr, ap->a_td); 542 if (error) 543 return (error); 544 if (np->n_flag & NRMODIFIED) { 545 if (vp->v_type == VDIR) 546 nfs_invaldir(vp); 547 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 548 if (error == EINTR) 549 return (error); 550 np->n_flag &= ~NRMODIFIED; 551 } 552 } 553 554 return (0); 555 } 556 557 /* 558 * nfs close vnode op 559 * What an NFS client should do upon close after writing is a debatable issue. 560 * Most NFS clients push delayed writes to the server upon close, basically for 561 * two reasons: 562 * 1 - So that any write errors may be reported back to the client process 563 * doing the close system call. By far the two most likely errors are 564 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 565 * 2 - To put a worst case upper bound on cache inconsistency between 566 * multiple clients for the file. 567 * There is also a consistency problem for Version 2 of the protocol w.r.t. 568 * not being able to tell if other clients are writing a file concurrently, 569 * since there is no way of knowing if the changed modify time in the reply 570 * is only due to the write for this client. 571 * (NFS Version 3 provides weak cache consistency data in the reply that 572 * should be sufficient to detect and handle this case.) 573 * 574 * The current code does the following: 575 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 576 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 577 * or commit them (this satisfies 1 and 2 except for the 578 * case where the server crashes after this close but 579 * before the commit RPC, which is felt to be "good 580 * enough". Changing the last argument to nfs_flush() to 581 * a 1 would force a commit operation, if it is felt a 582 * commit is necessary now. 583 * for NQNFS - do nothing now, since 2 is dealt with via leases and 584 * 1 should be dealt with via an fsync() system call for 585 * cases where write errors are important. 586 * 587 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag, 588 * struct ucred *a_cred, struct thread *a_td) 589 */ 590 /* ARGSUSED */ 591 static int 592 nfs_close(struct vop_close_args *ap) 593 { 594 struct vnode *vp = ap->a_vp; 595 struct nfsnode *np = VTONFS(vp); 596 int error = 0; 597 598 if (vp->v_type == VREG) { 599 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 600 (np->n_flag & NLMODIFIED)) { 601 if (NFS_ISV3(vp)) { 602 /* 603 * Under NFSv3 we have dirty buffers to dispose of. We 604 * must flush them to the NFS server. We have the option 605 * of waiting all the way through the commit rpc or just 606 * waiting for the initial write. The default is to only 607 * wait through the initial write so the data is in the 608 * server's cache, which is roughly similar to the state 609 * a standard disk subsystem leaves the file in on close(). 610 * 611 * We cannot clear the NLMODIFIED bit in np->n_flag due to 612 * potential races with other processes, and certainly 613 * cannot clear it if we don't commit. 614 */ 615 int cm = nfsv3_commit_on_close ? 1 : 0; 616 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm); 617 /* np->n_flag &= ~NLMODIFIED; */ 618 } else { 619 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 620 } 621 np->n_attrstamp = 0; 622 } 623 if (np->n_flag & NWRITEERR) { 624 np->n_flag &= ~NWRITEERR; 625 error = np->n_error; 626 } 627 } 628 return (error); 629 } 630 631 /* 632 * nfs getattr call from vfs. 633 * 634 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 635 * struct thread *a_td) 636 */ 637 static int 638 nfs_getattr(struct vop_getattr_args *ap) 639 { 640 struct vnode *vp = ap->a_vp; 641 struct nfsnode *np = VTONFS(vp); 642 caddr_t cp; 643 u_int32_t *tl; 644 int32_t t1, t2; 645 caddr_t bpos, dpos; 646 int error = 0; 647 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 648 int v3 = NFS_ISV3(vp); 649 650 /* 651 * Update local times for special files. 652 */ 653 if (np->n_flag & (NACC | NUPD)) 654 np->n_flag |= NCHG; 655 /* 656 * First look in the cache. 657 */ 658 if (nfs_getattrcache(vp, ap->a_vap) == 0) 659 return (0); 660 661 if (v3 && nfsaccess_cache_timeout > 0) { 662 nfsstats.accesscache_misses++; 663 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 664 if (nfs_getattrcache(vp, ap->a_vap) == 0) 665 return (0); 666 } 667 668 nfsstats.rpccnt[NFSPROC_GETATTR]++; 669 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 670 nfsm_fhtom(vp, v3); 671 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 672 if (!error) { 673 nfsm_loadattr(vp, ap->a_vap); 674 } 675 m_freem(mrep); 676 nfsmout: 677 return (error); 678 } 679 680 /* 681 * nfs setattr call. 682 * 683 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 684 * struct vattr *a_vap, struct ucred *a_cred, 685 * struct thread *a_td) 686 */ 687 static int 688 nfs_setattr(struct vop_setattr_args *ap) 689 { 690 struct vnode *vp = ap->a_vp; 691 struct nfsnode *np = VTONFS(vp); 692 struct vattr *vap = ap->a_vap; 693 int error = 0; 694 u_quad_t tsize; 695 696 #ifndef nolint 697 tsize = (u_quad_t)0; 698 #endif 699 700 /* 701 * Setting of flags is not supported. 702 */ 703 if (vap->va_flags != VNOVAL) 704 return (EOPNOTSUPP); 705 706 /* 707 * Disallow write attempts if the filesystem is mounted read-only. 708 */ 709 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 710 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 711 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 712 (vp->v_mount->mnt_flag & MNT_RDONLY)) 713 return (EROFS); 714 if (vap->va_size != VNOVAL) { 715 switch (vp->v_type) { 716 case VDIR: 717 return (EISDIR); 718 case VCHR: 719 case VBLK: 720 case VSOCK: 721 case VFIFO: 722 if (vap->va_mtime.tv_sec == VNOVAL && 723 vap->va_atime.tv_sec == VNOVAL && 724 vap->va_mode == (mode_t)VNOVAL && 725 vap->va_uid == (uid_t)VNOVAL && 726 vap->va_gid == (gid_t)VNOVAL) 727 return (0); 728 vap->va_size = VNOVAL; 729 break; 730 default: 731 /* 732 * Disallow write attempts if the filesystem is 733 * mounted read-only. 734 */ 735 if (vp->v_mount->mnt_flag & MNT_RDONLY) 736 return (EROFS); 737 738 /* 739 * This is nasty. The RPCs we send to flush pending 740 * data often return attribute information which is 741 * cached via a callback to nfs_loadattrcache(), which 742 * has the effect of changing our notion of the file 743 * size. Due to flushed appends and other operations 744 * the file size can be set to virtually anything, 745 * including values that do not match either the old 746 * or intended file size. 747 * 748 * When this condition is detected we must loop to 749 * try the operation again. Hopefully no more 750 * flushing is required on the loop so it works the 751 * second time around. THIS CASE ALMOST ALWAYS 752 * HAPPENS! 753 */ 754 tsize = np->n_size; 755 again: 756 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size); 757 758 if (np->n_flag & NLMODIFIED) { 759 if (vap->va_size == 0) 760 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1); 761 else 762 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 763 } 764 /* 765 * note: this loop case almost always happens at 766 * least once per truncation. 767 */ 768 if (error == 0 && np->n_size != vap->va_size) 769 goto again; 770 np->n_vattr.va_size = vap->va_size; 771 break; 772 } 773 } else if ((vap->va_mtime.tv_sec != VNOVAL || 774 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 775 vp->v_type == VREG && 776 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR 777 ) { 778 return (error); 779 } 780 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 781 782 /* 783 * Sanity check if a truncation was issued. This should only occur 784 * if multiple processes are racing on the same file. 785 */ 786 if (error == 0 && vap->va_size != VNOVAL && 787 np->n_size != vap->va_size) { 788 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 789 goto again; 790 } 791 if (error && vap->va_size != VNOVAL) { 792 np->n_size = np->n_vattr.va_size = tsize; 793 vnode_pager_setsize(vp, np->n_size); 794 } 795 return (error); 796 } 797 798 /* 799 * Do an nfs setattr rpc. 800 */ 801 static int 802 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 803 struct ucred *cred, struct thread *td) 804 { 805 struct nfsv2_sattr *sp; 806 struct nfsnode *np = VTONFS(vp); 807 caddr_t cp; 808 int32_t t1, t2; 809 caddr_t bpos, dpos, cp2; 810 u_int32_t *tl; 811 int error = 0, wccflag = NFSV3_WCCRATTR; 812 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 813 int v3 = NFS_ISV3(vp); 814 815 nfsstats.rpccnt[NFSPROC_SETATTR]++; 816 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 817 nfsm_fhtom(vp, v3); 818 if (v3) { 819 nfsm_v3attrbuild(vap, TRUE); 820 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 821 *tl = nfs_false; 822 } else { 823 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 824 if (vap->va_mode == (mode_t)VNOVAL) 825 sp->sa_mode = nfs_xdrneg1; 826 else 827 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 828 if (vap->va_uid == (uid_t)VNOVAL) 829 sp->sa_uid = nfs_xdrneg1; 830 else 831 sp->sa_uid = txdr_unsigned(vap->va_uid); 832 if (vap->va_gid == (gid_t)VNOVAL) 833 sp->sa_gid = nfs_xdrneg1; 834 else 835 sp->sa_gid = txdr_unsigned(vap->va_gid); 836 sp->sa_size = txdr_unsigned(vap->va_size); 837 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 838 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 839 } 840 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 841 if (v3) { 842 np->n_modestamp = 0; 843 nfsm_wcc_data(vp, wccflag); 844 } else 845 nfsm_loadattr(vp, (struct vattr *)0); 846 m_freem(mrep); 847 nfsmout: 848 return (error); 849 } 850 851 /* 852 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 853 * nfs_lookup() until all remaining new api calls are implemented. 854 * 855 * Resolve a namecache entry. This function is passed a locked ncp and 856 * must call cache_setvp() on it as appropriate to resolve the entry. 857 */ 858 static int 859 nfs_nresolve(struct vop_nresolve_args *ap) 860 { 861 struct thread *td = curthread; 862 struct namecache *ncp; 863 struct ucred *cred; 864 struct nfsnode *np; 865 struct vnode *dvp; 866 struct vnode *nvp; 867 nfsfh_t *fhp; 868 int attrflag; 869 int fhsize; 870 int error; 871 int len; 872 int v3; 873 /******NFSM MACROS********/ 874 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 875 caddr_t bpos, dpos, cp, cp2; 876 u_int32_t *tl; 877 int32_t t1, t2; 878 879 cred = ap->a_cred; 880 ncp = ap->a_ncp; 881 882 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp); 883 dvp = ncp->nc_parent->nc_vp; 884 if ((error = vget(dvp, LK_SHARED, td)) != 0) 885 return (error); 886 887 nvp = NULL; 888 v3 = NFS_ISV3(dvp); 889 nfsstats.lookupcache_misses++; 890 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 891 len = ncp->nc_nlen; 892 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 893 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 894 nfsm_fhtom(dvp, v3); 895 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 896 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 897 if (error) { 898 /* 899 * Cache negatve lookups to reduce NFS traffic, but use 900 * a fast timeout. Otherwise use a timeout of 1 tick. 901 * XXX we should add a namecache flag for no-caching 902 * to uncache the negative hit as soon as possible, but 903 * we cannot simply destroy the entry because it is used 904 * as a placeholder by the caller. 905 */ 906 if (error == ENOENT) { 907 int nticks; 908 909 if (nfsneg_cache_timeout) 910 nticks = nfsneg_cache_timeout * hz; 911 else 912 nticks = 1; 913 cache_setvp(ncp, NULL); 914 cache_settimeout(ncp, nticks); 915 } 916 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 917 m_freem(mrep); 918 goto nfsmout; 919 } 920 921 /* 922 * Success, get the file handle, do various checks, and load 923 * post-operation data from the reply packet. Theoretically 924 * we should never be looking up "." so, theoretically, we 925 * should never get the same file handle as our directory. But 926 * we check anyway. XXX 927 * 928 * Note that no timeout is set for the positive cache hit. We 929 * assume, theoretically, that ESTALE returns will be dealt with 930 * properly to handle NFS races and in anycase we cannot depend 931 * on a timeout to deal with NFS open/create/excl issues so instead 932 * of a bad hack here the rest of the NFS client code needs to do 933 * the right thing. 934 */ 935 nfsm_getfh(fhp, fhsize, v3); 936 937 np = VTONFS(dvp); 938 if (NFS_CMPFH(np, fhp, fhsize)) { 939 vref(dvp); 940 nvp = dvp; 941 } else { 942 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 943 if (error) { 944 m_freem(mrep); 945 vput(dvp); 946 return (error); 947 } 948 nvp = NFSTOV(np); 949 } 950 if (v3) { 951 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 952 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 953 } else { 954 nfsm_loadattr(nvp, NULL); 955 } 956 cache_setvp(ncp, nvp); 957 m_freem(mrep); 958 nfsmout: 959 vput(dvp); 960 if (nvp) { 961 if (nvp == dvp) 962 vrele(nvp); 963 else 964 vput(nvp); 965 } 966 return (error); 967 } 968 969 /* 970 * 'cached' nfs directory lookup 971 * 972 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 973 * 974 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 975 * struct vnode **a_vpp, struct componentname *a_cnp) 976 */ 977 static int 978 nfs_lookup(struct vop_lookup_args *ap) 979 { 980 struct componentname *cnp = ap->a_cnp; 981 struct vnode *dvp = ap->a_dvp; 982 struct vnode **vpp = ap->a_vpp; 983 int flags = cnp->cn_flags; 984 struct vnode *newvp; 985 u_int32_t *tl; 986 caddr_t cp; 987 int32_t t1, t2; 988 struct nfsmount *nmp; 989 caddr_t bpos, dpos, cp2; 990 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 991 long len; 992 nfsfh_t *fhp; 993 struct nfsnode *np; 994 int lockparent, wantparent, error = 0, attrflag, fhsize; 995 int v3 = NFS_ISV3(dvp); 996 struct thread *td = cnp->cn_td; 997 998 /* 999 * Read-only mount check and directory check. 1000 */ 1001 *vpp = NULLVP; 1002 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1003 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1004 return (EROFS); 1005 1006 if (dvp->v_type != VDIR) 1007 return (ENOTDIR); 1008 1009 /* 1010 * Look it up in the cache. Note that ENOENT is only returned if we 1011 * previously entered a negative hit (see later on). The additional 1012 * nfsneg_cache_timeout check causes previously cached results to 1013 * be instantly ignored if the negative caching is turned off. 1014 */ 1015 lockparent = flags & CNP_LOCKPARENT; 1016 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1017 nmp = VFSTONFS(dvp->v_mount); 1018 np = VTONFS(dvp); 1019 1020 /* 1021 * Go to the wire. 1022 */ 1023 error = 0; 1024 newvp = NULLVP; 1025 nfsstats.lookupcache_misses++; 1026 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1027 len = cnp->cn_namelen; 1028 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 1029 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1030 nfsm_fhtom(dvp, v3); 1031 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1032 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 1033 if (error) { 1034 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1035 m_freem(mrep); 1036 goto nfsmout; 1037 } 1038 nfsm_getfh(fhp, fhsize, v3); 1039 1040 /* 1041 * Handle RENAME case... 1042 */ 1043 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1044 if (NFS_CMPFH(np, fhp, fhsize)) { 1045 m_freem(mrep); 1046 return (EISDIR); 1047 } 1048 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1049 if (error) { 1050 m_freem(mrep); 1051 return (error); 1052 } 1053 newvp = NFSTOV(np); 1054 if (v3) { 1055 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1056 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1057 } else 1058 nfsm_loadattr(newvp, (struct vattr *)0); 1059 *vpp = newvp; 1060 m_freem(mrep); 1061 if (!lockparent) { 1062 VOP_UNLOCK(dvp, 0, td); 1063 cnp->cn_flags |= CNP_PDIRUNLOCK; 1064 } 1065 return (0); 1066 } 1067 1068 if (flags & CNP_ISDOTDOT) { 1069 VOP_UNLOCK(dvp, 0, td); 1070 cnp->cn_flags |= CNP_PDIRUNLOCK; 1071 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1072 if (error) { 1073 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 1074 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1075 return (error); /* NOTE: return error from nget */ 1076 } 1077 newvp = NFSTOV(np); 1078 if (lockparent) { 1079 error = vn_lock(dvp, LK_EXCLUSIVE, td); 1080 if (error) { 1081 vput(newvp); 1082 return (error); 1083 } 1084 cnp->cn_flags |= CNP_PDIRUNLOCK; 1085 } 1086 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1087 vref(dvp); 1088 newvp = dvp; 1089 } else { 1090 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1091 if (error) { 1092 m_freem(mrep); 1093 return (error); 1094 } 1095 if (!lockparent) { 1096 VOP_UNLOCK(dvp, 0, td); 1097 cnp->cn_flags |= CNP_PDIRUNLOCK; 1098 } 1099 newvp = NFSTOV(np); 1100 } 1101 if (v3) { 1102 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1103 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1104 } else 1105 nfsm_loadattr(newvp, (struct vattr *)0); 1106 #if 0 1107 /* XXX MOVE TO nfs_nremove() */ 1108 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1109 cnp->cn_nameiop != NAMEI_DELETE) { 1110 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1111 } 1112 #endif 1113 *vpp = newvp; 1114 m_freem(mrep); 1115 nfsmout: 1116 if (error) { 1117 if (newvp != NULLVP) { 1118 vrele(newvp); 1119 *vpp = NULLVP; 1120 } 1121 if ((cnp->cn_nameiop == NAMEI_CREATE || 1122 cnp->cn_nameiop == NAMEI_RENAME) && 1123 error == ENOENT) { 1124 if (!lockparent) { 1125 VOP_UNLOCK(dvp, 0, td); 1126 cnp->cn_flags |= CNP_PDIRUNLOCK; 1127 } 1128 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1129 error = EROFS; 1130 else 1131 error = EJUSTRETURN; 1132 } 1133 } 1134 return (error); 1135 } 1136 1137 /* 1138 * nfs read call. 1139 * Just call nfs_bioread() to do the work. 1140 * 1141 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1142 * struct ucred *a_cred) 1143 */ 1144 static int 1145 nfs_read(struct vop_read_args *ap) 1146 { 1147 struct vnode *vp = ap->a_vp; 1148 1149 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1150 switch (vp->v_type) { 1151 case VREG: 1152 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1153 case VDIR: 1154 return (EISDIR); 1155 default: 1156 return EOPNOTSUPP; 1157 } 1158 } 1159 1160 /* 1161 * nfs readlink call 1162 * 1163 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1164 */ 1165 static int 1166 nfs_readlink(struct vop_readlink_args *ap) 1167 { 1168 struct vnode *vp = ap->a_vp; 1169 1170 if (vp->v_type != VLNK) 1171 return (EINVAL); 1172 return (nfs_bioread(vp, ap->a_uio, 0)); 1173 } 1174 1175 /* 1176 * Do a readlink rpc. 1177 * Called by nfs_doio() from below the buffer cache. 1178 */ 1179 int 1180 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1181 { 1182 u_int32_t *tl; 1183 caddr_t cp; 1184 int32_t t1, t2; 1185 caddr_t bpos, dpos, cp2; 1186 int error = 0, len, attrflag; 1187 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1188 int v3 = NFS_ISV3(vp); 1189 1190 nfsstats.rpccnt[NFSPROC_READLINK]++; 1191 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1192 nfsm_fhtom(vp, v3); 1193 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1194 if (v3) 1195 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1196 if (!error) { 1197 nfsm_strsiz(len, NFS_MAXPATHLEN); 1198 if (len == NFS_MAXPATHLEN) { 1199 struct nfsnode *np = VTONFS(vp); 1200 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1201 len = np->n_size; 1202 } 1203 nfsm_mtouio(uiop, len); 1204 } 1205 m_freem(mrep); 1206 nfsmout: 1207 return (error); 1208 } 1209 1210 /* 1211 * nfs read rpc call 1212 * Ditto above 1213 */ 1214 int 1215 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1216 { 1217 u_int32_t *tl; 1218 caddr_t cp; 1219 int32_t t1, t2; 1220 caddr_t bpos, dpos, cp2; 1221 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1222 struct nfsmount *nmp; 1223 int error = 0, len, retlen, tsiz, eof, attrflag; 1224 int v3 = NFS_ISV3(vp); 1225 1226 #ifndef nolint 1227 eof = 0; 1228 #endif 1229 nmp = VFSTONFS(vp->v_mount); 1230 tsiz = uiop->uio_resid; 1231 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1232 return (EFBIG); 1233 while (tsiz > 0) { 1234 nfsstats.rpccnt[NFSPROC_READ]++; 1235 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1236 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1237 nfsm_fhtom(vp, v3); 1238 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1239 if (v3) { 1240 txdr_hyper(uiop->uio_offset, tl); 1241 *(tl + 2) = txdr_unsigned(len); 1242 } else { 1243 *tl++ = txdr_unsigned(uiop->uio_offset); 1244 *tl++ = txdr_unsigned(len); 1245 *tl = 0; 1246 } 1247 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1248 if (v3) { 1249 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1250 if (error) { 1251 m_freem(mrep); 1252 goto nfsmout; 1253 } 1254 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1255 eof = fxdr_unsigned(int, *(tl + 1)); 1256 } else 1257 nfsm_loadattr(vp, (struct vattr *)0); 1258 nfsm_strsiz(retlen, nmp->nm_rsize); 1259 nfsm_mtouio(uiop, retlen); 1260 m_freem(mrep); 1261 tsiz -= retlen; 1262 if (v3) { 1263 if (eof || retlen == 0) { 1264 tsiz = 0; 1265 } 1266 } else if (retlen < len) { 1267 tsiz = 0; 1268 } 1269 } 1270 nfsmout: 1271 return (error); 1272 } 1273 1274 /* 1275 * nfs write call 1276 */ 1277 int 1278 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1279 { 1280 u_int32_t *tl; 1281 caddr_t cp; 1282 int32_t t1, t2, backup; 1283 caddr_t bpos, dpos, cp2; 1284 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1285 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1286 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1287 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1288 1289 #ifndef DIAGNOSTIC 1290 if (uiop->uio_iovcnt != 1) 1291 panic("nfs: writerpc iovcnt > 1"); 1292 #endif 1293 *must_commit = 0; 1294 tsiz = uiop->uio_resid; 1295 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1296 return (EFBIG); 1297 while (tsiz > 0) { 1298 nfsstats.rpccnt[NFSPROC_WRITE]++; 1299 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1300 nfsm_reqhead(vp, NFSPROC_WRITE, 1301 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1302 nfsm_fhtom(vp, v3); 1303 if (v3) { 1304 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1305 txdr_hyper(uiop->uio_offset, tl); 1306 tl += 2; 1307 *tl++ = txdr_unsigned(len); 1308 *tl++ = txdr_unsigned(*iomode); 1309 *tl = txdr_unsigned(len); 1310 } else { 1311 u_int32_t x; 1312 1313 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1314 /* Set both "begin" and "current" to non-garbage. */ 1315 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1316 *tl++ = x; /* "begin offset" */ 1317 *tl++ = x; /* "current offset" */ 1318 x = txdr_unsigned(len); 1319 *tl++ = x; /* total to this offset */ 1320 *tl = x; /* size of this write */ 1321 } 1322 nfsm_uiotom(uiop, len); 1323 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1324 if (v3) { 1325 /* 1326 * The write RPC returns a before and after mtime. The 1327 * nfsm_wcc_data() macro checks the before n_mtime 1328 * against the before time and stores the after time 1329 * in the nfsnode's cached vattr and n_mtime field. 1330 * The NRMODIFIED bit will be set if the before 1331 * time did not match the original mtime. 1332 */ 1333 wccflag = NFSV3_WCCCHK; 1334 nfsm_wcc_data(vp, wccflag); 1335 if (!error) { 1336 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1337 + NFSX_V3WRITEVERF); 1338 rlen = fxdr_unsigned(int, *tl++); 1339 if (rlen == 0) { 1340 error = NFSERR_IO; 1341 m_freem(mrep); 1342 break; 1343 } else if (rlen < len) { 1344 backup = len - rlen; 1345 uiop->uio_iov->iov_base -= backup; 1346 uiop->uio_iov->iov_len += backup; 1347 uiop->uio_offset -= backup; 1348 uiop->uio_resid += backup; 1349 len = rlen; 1350 } 1351 commit = fxdr_unsigned(int, *tl++); 1352 1353 /* 1354 * Return the lowest committment level 1355 * obtained by any of the RPCs. 1356 */ 1357 if (committed == NFSV3WRITE_FILESYNC) 1358 committed = commit; 1359 else if (committed == NFSV3WRITE_DATASYNC && 1360 commit == NFSV3WRITE_UNSTABLE) 1361 committed = commit; 1362 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1363 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1364 NFSX_V3WRITEVERF); 1365 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1366 } else if (bcmp((caddr_t)tl, 1367 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1368 *must_commit = 1; 1369 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1370 NFSX_V3WRITEVERF); 1371 } 1372 } 1373 } else { 1374 nfsm_loadattr(vp, (struct vattr *)0); 1375 } 1376 m_freem(mrep); 1377 if (error) 1378 break; 1379 tsiz -= len; 1380 } 1381 nfsmout: 1382 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1383 committed = NFSV3WRITE_FILESYNC; 1384 *iomode = committed; 1385 if (error) 1386 uiop->uio_resid = tsiz; 1387 return (error); 1388 } 1389 1390 /* 1391 * nfs mknod rpc 1392 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1393 * mode set to specify the file type and the size field for rdev. 1394 */ 1395 static int 1396 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1397 struct vattr *vap) 1398 { 1399 struct nfsv2_sattr *sp; 1400 u_int32_t *tl; 1401 caddr_t cp; 1402 int32_t t1, t2; 1403 struct vnode *newvp = (struct vnode *)0; 1404 struct nfsnode *np = (struct nfsnode *)0; 1405 struct vattr vattr; 1406 char *cp2; 1407 caddr_t bpos, dpos; 1408 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1409 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1410 u_int32_t rdev; 1411 int v3 = NFS_ISV3(dvp); 1412 1413 if (vap->va_type == VCHR || vap->va_type == VBLK) 1414 rdev = txdr_unsigned(vap->va_rdev); 1415 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1416 rdev = nfs_xdrneg1; 1417 else { 1418 return (EOPNOTSUPP); 1419 } 1420 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1421 return (error); 1422 } 1423 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1424 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1425 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1426 nfsm_fhtom(dvp, v3); 1427 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1428 if (v3) { 1429 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1430 *tl++ = vtonfsv3_type(vap->va_type); 1431 nfsm_v3attrbuild(vap, FALSE); 1432 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1433 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1434 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1435 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1436 } 1437 } else { 1438 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1439 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1440 sp->sa_uid = nfs_xdrneg1; 1441 sp->sa_gid = nfs_xdrneg1; 1442 sp->sa_size = rdev; 1443 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1444 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1445 } 1446 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1447 if (!error) { 1448 nfsm_mtofh(dvp, newvp, v3, gotvp); 1449 if (!gotvp) { 1450 if (newvp) { 1451 vput(newvp); 1452 newvp = (struct vnode *)0; 1453 } 1454 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1455 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1456 if (!error) 1457 newvp = NFSTOV(np); 1458 } 1459 } 1460 if (v3) 1461 nfsm_wcc_data(dvp, wccflag); 1462 m_freem(mrep); 1463 nfsmout: 1464 if (error) { 1465 if (newvp) 1466 vput(newvp); 1467 } else { 1468 *vpp = newvp; 1469 } 1470 VTONFS(dvp)->n_flag |= NLMODIFIED; 1471 if (!wccflag) 1472 VTONFS(dvp)->n_attrstamp = 0; 1473 return (error); 1474 } 1475 1476 /* 1477 * nfs mknod vop 1478 * just call nfs_mknodrpc() to do the work. 1479 * 1480 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1481 * struct componentname *a_cnp, struct vattr *a_vap) 1482 */ 1483 /* ARGSUSED */ 1484 static int 1485 nfs_mknod(struct vop_mknod_args *ap) 1486 { 1487 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1488 } 1489 1490 static u_long create_verf; 1491 /* 1492 * nfs file create call 1493 * 1494 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1495 * struct componentname *a_cnp, struct vattr *a_vap) 1496 */ 1497 static int 1498 nfs_create(struct vop_create_args *ap) 1499 { 1500 struct vnode *dvp = ap->a_dvp; 1501 struct vattr *vap = ap->a_vap; 1502 struct componentname *cnp = ap->a_cnp; 1503 struct nfsv2_sattr *sp; 1504 u_int32_t *tl; 1505 caddr_t cp; 1506 int32_t t1, t2; 1507 struct nfsnode *np = (struct nfsnode *)0; 1508 struct vnode *newvp = (struct vnode *)0; 1509 caddr_t bpos, dpos, cp2; 1510 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1511 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1512 struct vattr vattr; 1513 int v3 = NFS_ISV3(dvp); 1514 1515 /* 1516 * Oops, not for me.. 1517 */ 1518 if (vap->va_type == VSOCK) 1519 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1520 1521 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1522 return (error); 1523 } 1524 if (vap->va_vaflags & VA_EXCLUSIVE) 1525 fmode |= O_EXCL; 1526 again: 1527 nfsstats.rpccnt[NFSPROC_CREATE]++; 1528 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1529 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1530 nfsm_fhtom(dvp, v3); 1531 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1532 if (v3) { 1533 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1534 if (fmode & O_EXCL) { 1535 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1536 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1537 #ifdef INET 1538 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1539 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1540 else 1541 #endif 1542 *tl++ = create_verf; 1543 *tl = ++create_verf; 1544 } else { 1545 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1546 nfsm_v3attrbuild(vap, FALSE); 1547 } 1548 } else { 1549 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1550 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1551 sp->sa_uid = nfs_xdrneg1; 1552 sp->sa_gid = nfs_xdrneg1; 1553 sp->sa_size = 0; 1554 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1555 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1556 } 1557 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1558 if (!error) { 1559 nfsm_mtofh(dvp, newvp, v3, gotvp); 1560 if (!gotvp) { 1561 if (newvp) { 1562 vput(newvp); 1563 newvp = (struct vnode *)0; 1564 } 1565 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1566 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1567 if (!error) 1568 newvp = NFSTOV(np); 1569 } 1570 } 1571 if (v3) 1572 nfsm_wcc_data(dvp, wccflag); 1573 m_freem(mrep); 1574 nfsmout: 1575 if (error) { 1576 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1577 fmode &= ~O_EXCL; 1578 goto again; 1579 } 1580 if (newvp) 1581 vput(newvp); 1582 } else if (v3 && (fmode & O_EXCL)) { 1583 /* 1584 * We are normally called with only a partially initialized 1585 * VAP. Since the NFSv3 spec says that server may use the 1586 * file attributes to store the verifier, the spec requires 1587 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1588 * in atime, but we can't really assume that all servers will 1589 * so we ensure that our SETATTR sets both atime and mtime. 1590 */ 1591 if (vap->va_mtime.tv_sec == VNOVAL) 1592 vfs_timestamp(&vap->va_mtime); 1593 if (vap->va_atime.tv_sec == VNOVAL) 1594 vap->va_atime = vap->va_mtime; 1595 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1596 } 1597 if (!error) { 1598 /* 1599 * The new np may have enough info for access 1600 * checks, make sure rucred and wucred are 1601 * initialized for read and write rpc's. 1602 */ 1603 np = VTONFS(newvp); 1604 if (np->n_rucred == NULL) 1605 np->n_rucred = crhold(cnp->cn_cred); 1606 if (np->n_wucred == NULL) 1607 np->n_wucred = crhold(cnp->cn_cred); 1608 *ap->a_vpp = newvp; 1609 } 1610 VTONFS(dvp)->n_flag |= NLMODIFIED; 1611 if (!wccflag) 1612 VTONFS(dvp)->n_attrstamp = 0; 1613 return (error); 1614 } 1615 1616 /* 1617 * nfs file remove call 1618 * To try and make nfs semantics closer to ufs semantics, a file that has 1619 * other processes using the vnode is renamed instead of removed and then 1620 * removed later on the last close. 1621 * - If v_usecount > 1 1622 * If a rename is not already in the works 1623 * call nfs_sillyrename() to set it up 1624 * else 1625 * do the remove rpc 1626 * 1627 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 1628 * struct vnode *a_vp, struct componentname *a_cnp) 1629 */ 1630 static int 1631 nfs_remove(struct vop_remove_args *ap) 1632 { 1633 struct vnode *vp = ap->a_vp; 1634 struct vnode *dvp = ap->a_dvp; 1635 struct componentname *cnp = ap->a_cnp; 1636 struct nfsnode *np = VTONFS(vp); 1637 int error = 0; 1638 struct vattr vattr; 1639 1640 #ifndef DIAGNOSTIC 1641 if (vp->v_usecount < 1) 1642 panic("nfs_remove: bad v_usecount"); 1643 #endif 1644 if (vp->v_type == VDIR) 1645 error = EPERM; 1646 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1647 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 && 1648 vattr.va_nlink > 1)) { 1649 /* 1650 * throw away biocache buffers, mainly to avoid 1651 * unnecessary delayed writes later. 1652 */ 1653 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1); 1654 /* Do the rpc */ 1655 if (error != EINTR) 1656 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1657 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1658 /* 1659 * Kludge City: If the first reply to the remove rpc is lost.. 1660 * the reply to the retransmitted request will be ENOENT 1661 * since the file was in fact removed 1662 * Therefore, we cheat and return success. 1663 */ 1664 if (error == ENOENT) 1665 error = 0; 1666 } else if (!np->n_sillyrename) { 1667 error = nfs_sillyrename(dvp, vp, cnp); 1668 } 1669 np->n_attrstamp = 0; 1670 return (error); 1671 } 1672 1673 /* 1674 * nfs file remove rpc called from nfs_inactive 1675 */ 1676 int 1677 nfs_removeit(struct sillyrename *sp) 1678 { 1679 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1680 sp->s_cred, NULL)); 1681 } 1682 1683 /* 1684 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1685 */ 1686 static int 1687 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1688 struct ucred *cred, struct thread *td) 1689 { 1690 u_int32_t *tl; 1691 caddr_t cp; 1692 int32_t t1, t2; 1693 caddr_t bpos, dpos, cp2; 1694 int error = 0, wccflag = NFSV3_WCCRATTR; 1695 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1696 int v3 = NFS_ISV3(dvp); 1697 1698 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1699 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1700 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1701 nfsm_fhtom(dvp, v3); 1702 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1703 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1704 if (v3) 1705 nfsm_wcc_data(dvp, wccflag); 1706 m_freem(mrep); 1707 nfsmout: 1708 VTONFS(dvp)->n_flag |= NLMODIFIED; 1709 if (!wccflag) 1710 VTONFS(dvp)->n_attrstamp = 0; 1711 return (error); 1712 } 1713 1714 /* 1715 * nfs file rename call 1716 * 1717 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1718 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1719 * struct vnode *a_tvp, struct componentname *a_tcnp) 1720 */ 1721 static int 1722 nfs_rename(struct vop_rename_args *ap) 1723 { 1724 struct vnode *fvp = ap->a_fvp; 1725 struct vnode *tvp = ap->a_tvp; 1726 struct vnode *fdvp = ap->a_fdvp; 1727 struct vnode *tdvp = ap->a_tdvp; 1728 struct componentname *tcnp = ap->a_tcnp; 1729 struct componentname *fcnp = ap->a_fcnp; 1730 int error; 1731 1732 /* Check for cross-device rename */ 1733 if ((fvp->v_mount != tdvp->v_mount) || 1734 (tvp && (fvp->v_mount != tvp->v_mount))) { 1735 error = EXDEV; 1736 goto out; 1737 } 1738 1739 /* 1740 * We have to flush B_DELWRI data prior to renaming 1741 * the file. If we don't, the delayed-write buffers 1742 * can be flushed out later after the file has gone stale 1743 * under NFSV3. NFSV2 does not have this problem because 1744 * ( as far as I can tell ) it flushes dirty buffers more 1745 * often. 1746 */ 1747 1748 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td); 1749 if (tvp) 1750 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td); 1751 1752 /* 1753 * If the tvp exists and is in use, sillyrename it before doing the 1754 * rename of the new file over it. 1755 * 1756 * XXX Can't sillyrename a directory. 1757 * 1758 * We do not attempt to do any namecache purges in this old API 1759 * routine. The new API compat functions have access to the actual 1760 * namecache structures and will do it for us. 1761 */ 1762 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1763 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1764 vput(tvp); 1765 tvp = NULL; 1766 } else if (tvp) { 1767 ; 1768 } 1769 1770 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1771 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1772 tcnp->cn_td); 1773 1774 out: 1775 if (tdvp == tvp) 1776 vrele(tdvp); 1777 else 1778 vput(tdvp); 1779 if (tvp) 1780 vput(tvp); 1781 vrele(fdvp); 1782 vrele(fvp); 1783 /* 1784 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1785 */ 1786 if (error == ENOENT) 1787 error = 0; 1788 return (error); 1789 } 1790 1791 /* 1792 * nfs file rename rpc called from nfs_remove() above 1793 */ 1794 static int 1795 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1796 struct sillyrename *sp) 1797 { 1798 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1799 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1800 } 1801 1802 /* 1803 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1804 */ 1805 static int 1806 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1807 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1808 struct ucred *cred, struct thread *td) 1809 { 1810 u_int32_t *tl; 1811 caddr_t cp; 1812 int32_t t1, t2; 1813 caddr_t bpos, dpos, cp2; 1814 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1815 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1816 int v3 = NFS_ISV3(fdvp); 1817 1818 nfsstats.rpccnt[NFSPROC_RENAME]++; 1819 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1820 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1821 nfsm_rndup(tnamelen)); 1822 nfsm_fhtom(fdvp, v3); 1823 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1824 nfsm_fhtom(tdvp, v3); 1825 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1826 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1827 if (v3) { 1828 nfsm_wcc_data(fdvp, fwccflag); 1829 nfsm_wcc_data(tdvp, twccflag); 1830 } 1831 m_freem(mrep); 1832 nfsmout: 1833 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1834 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1835 if (!fwccflag) 1836 VTONFS(fdvp)->n_attrstamp = 0; 1837 if (!twccflag) 1838 VTONFS(tdvp)->n_attrstamp = 0; 1839 return (error); 1840 } 1841 1842 /* 1843 * nfs hard link create call 1844 * 1845 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1846 * struct componentname *a_cnp) 1847 */ 1848 static int 1849 nfs_link(struct vop_link_args *ap) 1850 { 1851 struct vnode *vp = ap->a_vp; 1852 struct vnode *tdvp = ap->a_tdvp; 1853 struct componentname *cnp = ap->a_cnp; 1854 u_int32_t *tl; 1855 caddr_t cp; 1856 int32_t t1, t2; 1857 caddr_t bpos, dpos, cp2; 1858 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1859 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1860 int v3; 1861 1862 if (vp->v_mount != tdvp->v_mount) { 1863 return (EXDEV); 1864 } 1865 1866 /* 1867 * Push all writes to the server, so that the attribute cache 1868 * doesn't get "out of sync" with the server. 1869 * XXX There should be a better way! 1870 */ 1871 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td); 1872 1873 v3 = NFS_ISV3(vp); 1874 nfsstats.rpccnt[NFSPROC_LINK]++; 1875 nfsm_reqhead(vp, NFSPROC_LINK, 1876 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1877 nfsm_fhtom(vp, v3); 1878 nfsm_fhtom(tdvp, v3); 1879 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1880 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1881 if (v3) { 1882 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1883 nfsm_wcc_data(tdvp, wccflag); 1884 } 1885 m_freem(mrep); 1886 nfsmout: 1887 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1888 if (!attrflag) 1889 VTONFS(vp)->n_attrstamp = 0; 1890 if (!wccflag) 1891 VTONFS(tdvp)->n_attrstamp = 0; 1892 /* 1893 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1894 */ 1895 if (error == EEXIST) 1896 error = 0; 1897 return (error); 1898 } 1899 1900 /* 1901 * nfs symbolic link create call 1902 * 1903 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1904 * struct componentname *a_cnp, struct vattr *a_vap, 1905 * char *a_target) 1906 */ 1907 static int 1908 nfs_symlink(struct vop_symlink_args *ap) 1909 { 1910 struct vnode *dvp = ap->a_dvp; 1911 struct vattr *vap = ap->a_vap; 1912 struct componentname *cnp = ap->a_cnp; 1913 struct nfsv2_sattr *sp; 1914 u_int32_t *tl; 1915 caddr_t cp; 1916 int32_t t1, t2; 1917 caddr_t bpos, dpos, cp2; 1918 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1919 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1920 struct vnode *newvp = (struct vnode *)0; 1921 int v3 = NFS_ISV3(dvp); 1922 1923 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1924 slen = strlen(ap->a_target); 1925 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1926 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1927 nfsm_fhtom(dvp, v3); 1928 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1929 if (v3) { 1930 nfsm_v3attrbuild(vap, FALSE); 1931 } 1932 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1933 if (!v3) { 1934 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1935 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1936 sp->sa_uid = nfs_xdrneg1; 1937 sp->sa_gid = nfs_xdrneg1; 1938 sp->sa_size = nfs_xdrneg1; 1939 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1940 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1941 } 1942 1943 /* 1944 * Issue the NFS request and get the rpc response. 1945 * 1946 * Only NFSv3 responses returning an error of 0 actually return 1947 * a file handle that can be converted into newvp without having 1948 * to do an extra lookup rpc. 1949 */ 1950 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1951 if (v3) { 1952 if (error == 0) 1953 nfsm_mtofh(dvp, newvp, v3, gotvp); 1954 nfsm_wcc_data(dvp, wccflag); 1955 } 1956 1957 /* 1958 * out code jumps -> here, mrep is also freed. 1959 */ 1960 1961 m_freem(mrep); 1962 nfsmout: 1963 1964 /* 1965 * If we get an EEXIST error, silently convert it to no-error 1966 * in case of an NFS retry. 1967 */ 1968 if (error == EEXIST) 1969 error = 0; 1970 1971 /* 1972 * If we do not have (or no longer have) an error, and we could 1973 * not extract the newvp from the response due to the request being 1974 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1975 * to obtain a newvp to return. 1976 */ 1977 if (error == 0 && newvp == NULL) { 1978 struct nfsnode *np = NULL; 1979 1980 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1981 cnp->cn_cred, cnp->cn_td, &np); 1982 if (!error) 1983 newvp = NFSTOV(np); 1984 } 1985 if (error) { 1986 if (newvp) 1987 vput(newvp); 1988 } else { 1989 *ap->a_vpp = newvp; 1990 } 1991 VTONFS(dvp)->n_flag |= NLMODIFIED; 1992 if (!wccflag) 1993 VTONFS(dvp)->n_attrstamp = 0; 1994 return (error); 1995 } 1996 1997 /* 1998 * nfs make dir call 1999 * 2000 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2001 * struct componentname *a_cnp, struct vattr *a_vap) 2002 */ 2003 static int 2004 nfs_mkdir(struct vop_mkdir_args *ap) 2005 { 2006 struct vnode *dvp = ap->a_dvp; 2007 struct vattr *vap = ap->a_vap; 2008 struct componentname *cnp = ap->a_cnp; 2009 struct nfsv2_sattr *sp; 2010 u_int32_t *tl; 2011 caddr_t cp; 2012 int32_t t1, t2; 2013 int len; 2014 struct nfsnode *np = (struct nfsnode *)0; 2015 struct vnode *newvp = (struct vnode *)0; 2016 caddr_t bpos, dpos, cp2; 2017 int error = 0, wccflag = NFSV3_WCCRATTR; 2018 int gotvp = 0; 2019 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2020 struct vattr vattr; 2021 int v3 = NFS_ISV3(dvp); 2022 2023 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 2024 return (error); 2025 } 2026 len = cnp->cn_namelen; 2027 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2028 nfsm_reqhead(dvp, NFSPROC_MKDIR, 2029 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2030 nfsm_fhtom(dvp, v3); 2031 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2032 if (v3) { 2033 nfsm_v3attrbuild(vap, FALSE); 2034 } else { 2035 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2036 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2037 sp->sa_uid = nfs_xdrneg1; 2038 sp->sa_gid = nfs_xdrneg1; 2039 sp->sa_size = nfs_xdrneg1; 2040 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2041 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2042 } 2043 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2044 if (!error) 2045 nfsm_mtofh(dvp, newvp, v3, gotvp); 2046 if (v3) 2047 nfsm_wcc_data(dvp, wccflag); 2048 m_freem(mrep); 2049 nfsmout: 2050 VTONFS(dvp)->n_flag |= NLMODIFIED; 2051 if (!wccflag) 2052 VTONFS(dvp)->n_attrstamp = 0; 2053 /* 2054 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2055 * if we can succeed in looking up the directory. 2056 */ 2057 if (error == EEXIST || (!error && !gotvp)) { 2058 if (newvp) { 2059 vrele(newvp); 2060 newvp = (struct vnode *)0; 2061 } 2062 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2063 cnp->cn_td, &np); 2064 if (!error) { 2065 newvp = NFSTOV(np); 2066 if (newvp->v_type != VDIR) 2067 error = EEXIST; 2068 } 2069 } 2070 if (error) { 2071 if (newvp) 2072 vrele(newvp); 2073 } else 2074 *ap->a_vpp = newvp; 2075 return (error); 2076 } 2077 2078 /* 2079 * nfs remove directory call 2080 * 2081 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2082 * struct componentname *a_cnp) 2083 */ 2084 static int 2085 nfs_rmdir(struct vop_rmdir_args *ap) 2086 { 2087 struct vnode *vp = ap->a_vp; 2088 struct vnode *dvp = ap->a_dvp; 2089 struct componentname *cnp = ap->a_cnp; 2090 u_int32_t *tl; 2091 caddr_t cp; 2092 int32_t t1, t2; 2093 caddr_t bpos, dpos, cp2; 2094 int error = 0, wccflag = NFSV3_WCCRATTR; 2095 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2096 int v3 = NFS_ISV3(dvp); 2097 2098 if (dvp == vp) 2099 return (EINVAL); 2100 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2101 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2102 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2103 nfsm_fhtom(dvp, v3); 2104 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2105 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2106 if (v3) 2107 nfsm_wcc_data(dvp, wccflag); 2108 m_freem(mrep); 2109 nfsmout: 2110 VTONFS(dvp)->n_flag |= NLMODIFIED; 2111 if (!wccflag) 2112 VTONFS(dvp)->n_attrstamp = 0; 2113 /* 2114 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2115 */ 2116 if (error == ENOENT) 2117 error = 0; 2118 return (error); 2119 } 2120 2121 /* 2122 * nfs readdir call 2123 * 2124 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2125 */ 2126 static int 2127 nfs_readdir(struct vop_readdir_args *ap) 2128 { 2129 struct vnode *vp = ap->a_vp; 2130 struct nfsnode *np = VTONFS(vp); 2131 struct uio *uio = ap->a_uio; 2132 int tresid, error; 2133 struct vattr vattr; 2134 2135 if (vp->v_type != VDIR) 2136 return (EPERM); 2137 2138 /* 2139 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2140 * and then check that is still valid, or if this is an NQNFS mount 2141 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2142 * VOP_GETATTR() does not necessarily go to the wire. 2143 */ 2144 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2145 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2146 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) { 2147 if (NQNFS_CKCACHABLE(vp, ND_READ)) { 2148 nfsstats.direofcache_hits++; 2149 return (0); 2150 } 2151 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 && 2152 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2153 ) { 2154 nfsstats.direofcache_hits++; 2155 return (0); 2156 } 2157 } 2158 2159 /* 2160 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2161 * own cache coherency checks so we do not have to. 2162 */ 2163 tresid = uio->uio_resid; 2164 error = nfs_bioread(vp, uio, 0); 2165 2166 if (!error && uio->uio_resid == tresid) 2167 nfsstats.direofcache_misses++; 2168 return (error); 2169 } 2170 2171 /* 2172 * Readdir rpc call. 2173 * Called from below the buffer cache by nfs_doio(). 2174 */ 2175 int 2176 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2177 { 2178 int len, left; 2179 struct dirent *dp = NULL; 2180 u_int32_t *tl; 2181 caddr_t cp; 2182 int32_t t1, t2; 2183 nfsuint64 *cookiep; 2184 caddr_t bpos, dpos, cp2; 2185 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2186 nfsuint64 cookie; 2187 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2188 struct nfsnode *dnp = VTONFS(vp); 2189 u_quad_t fileno; 2190 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2191 int attrflag; 2192 int v3 = NFS_ISV3(vp); 2193 2194 #ifndef DIAGNOSTIC 2195 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2196 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2197 panic("nfs readdirrpc bad uio"); 2198 #endif 2199 2200 /* 2201 * If there is no cookie, assume directory was stale. 2202 */ 2203 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2204 if (cookiep) 2205 cookie = *cookiep; 2206 else 2207 return (NFSERR_BAD_COOKIE); 2208 /* 2209 * Loop around doing readdir rpc's of size nm_readdirsize 2210 * truncated to a multiple of DIRBLKSIZ. 2211 * The stopping criteria is EOF or buffer full. 2212 */ 2213 while (more_dirs && bigenough) { 2214 nfsstats.rpccnt[NFSPROC_READDIR]++; 2215 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2216 NFSX_READDIR(v3)); 2217 nfsm_fhtom(vp, v3); 2218 if (v3) { 2219 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2220 *tl++ = cookie.nfsuquad[0]; 2221 *tl++ = cookie.nfsuquad[1]; 2222 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2223 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2224 } else { 2225 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2226 *tl++ = cookie.nfsuquad[0]; 2227 } 2228 *tl = txdr_unsigned(nmp->nm_readdirsize); 2229 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2230 if (v3) { 2231 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2232 if (!error) { 2233 nfsm_dissect(tl, u_int32_t *, 2234 2 * NFSX_UNSIGNED); 2235 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2236 dnp->n_cookieverf.nfsuquad[1] = *tl; 2237 } else { 2238 m_freem(mrep); 2239 goto nfsmout; 2240 } 2241 } 2242 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2243 more_dirs = fxdr_unsigned(int, *tl); 2244 2245 /* loop thru the dir entries, doctoring them to 4bsd form */ 2246 while (more_dirs && bigenough) { 2247 if (v3) { 2248 nfsm_dissect(tl, u_int32_t *, 2249 3 * NFSX_UNSIGNED); 2250 fileno = fxdr_hyper(tl); 2251 len = fxdr_unsigned(int, *(tl + 2)); 2252 } else { 2253 nfsm_dissect(tl, u_int32_t *, 2254 2 * NFSX_UNSIGNED); 2255 fileno = fxdr_unsigned(u_quad_t, *tl++); 2256 len = fxdr_unsigned(int, *tl); 2257 } 2258 if (len <= 0 || len > NFS_MAXNAMLEN) { 2259 error = EBADRPC; 2260 m_freem(mrep); 2261 goto nfsmout; 2262 } 2263 tlen = nfsm_rndup(len); 2264 if (tlen == len) 2265 tlen += 4; /* To ensure null termination */ 2266 left = DIRBLKSIZ - blksiz; 2267 if ((tlen + DIRHDSIZ) > left) { 2268 dp->d_reclen += left; 2269 uiop->uio_iov->iov_base += left; 2270 uiop->uio_iov->iov_len -= left; 2271 uiop->uio_offset += left; 2272 uiop->uio_resid -= left; 2273 blksiz = 0; 2274 } 2275 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2276 bigenough = 0; 2277 if (bigenough) { 2278 dp = (struct dirent *)uiop->uio_iov->iov_base; 2279 dp->d_fileno = (int)fileno; 2280 dp->d_namlen = len; 2281 dp->d_reclen = tlen + DIRHDSIZ; 2282 dp->d_type = DT_UNKNOWN; 2283 blksiz += dp->d_reclen; 2284 if (blksiz == DIRBLKSIZ) 2285 blksiz = 0; 2286 uiop->uio_offset += DIRHDSIZ; 2287 uiop->uio_resid -= DIRHDSIZ; 2288 uiop->uio_iov->iov_base += DIRHDSIZ; 2289 uiop->uio_iov->iov_len -= DIRHDSIZ; 2290 nfsm_mtouio(uiop, len); 2291 cp = uiop->uio_iov->iov_base; 2292 tlen -= len; 2293 *cp = '\0'; /* null terminate */ 2294 uiop->uio_iov->iov_base += tlen; 2295 uiop->uio_iov->iov_len -= tlen; 2296 uiop->uio_offset += tlen; 2297 uiop->uio_resid -= tlen; 2298 } else 2299 nfsm_adv(nfsm_rndup(len)); 2300 if (v3) { 2301 nfsm_dissect(tl, u_int32_t *, 2302 3 * NFSX_UNSIGNED); 2303 } else { 2304 nfsm_dissect(tl, u_int32_t *, 2305 2 * NFSX_UNSIGNED); 2306 } 2307 if (bigenough) { 2308 cookie.nfsuquad[0] = *tl++; 2309 if (v3) 2310 cookie.nfsuquad[1] = *tl++; 2311 } else if (v3) 2312 tl += 2; 2313 else 2314 tl++; 2315 more_dirs = fxdr_unsigned(int, *tl); 2316 } 2317 /* 2318 * If at end of rpc data, get the eof boolean 2319 */ 2320 if (!more_dirs) { 2321 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2322 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2323 } 2324 m_freem(mrep); 2325 } 2326 /* 2327 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2328 * by increasing d_reclen for the last record. 2329 */ 2330 if (blksiz > 0) { 2331 left = DIRBLKSIZ - blksiz; 2332 dp->d_reclen += left; 2333 uiop->uio_iov->iov_base += left; 2334 uiop->uio_iov->iov_len -= left; 2335 uiop->uio_offset += left; 2336 uiop->uio_resid -= left; 2337 } 2338 2339 /* 2340 * We are now either at the end of the directory or have filled the 2341 * block. 2342 */ 2343 if (bigenough) 2344 dnp->n_direofoffset = uiop->uio_offset; 2345 else { 2346 if (uiop->uio_resid > 0) 2347 printf("EEK! readdirrpc resid > 0\n"); 2348 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2349 *cookiep = cookie; 2350 } 2351 nfsmout: 2352 return (error); 2353 } 2354 2355 /* 2356 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2357 */ 2358 int 2359 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2360 { 2361 int len, left; 2362 struct dirent *dp; 2363 u_int32_t *tl; 2364 caddr_t cp; 2365 int32_t t1, t2; 2366 struct vnode *newvp; 2367 nfsuint64 *cookiep; 2368 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2369 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2370 nfsuint64 cookie; 2371 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2372 struct nfsnode *dnp = VTONFS(vp), *np; 2373 nfsfh_t *fhp; 2374 u_quad_t fileno; 2375 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2376 int attrflag, fhsize; 2377 struct namecache *ncp; 2378 struct namecache *dncp; 2379 struct nlcomponent nlc; 2380 2381 #ifndef nolint 2382 dp = (struct dirent *)0; 2383 #endif 2384 #ifndef DIAGNOSTIC 2385 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2386 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2387 panic("nfs readdirplusrpc bad uio"); 2388 #endif 2389 /* 2390 * Obtain the namecache record for the directory so we have something 2391 * to use as a basis for creating the entries. This function will 2392 * return a held (but not locked) ncp. The ncp may be disconnected 2393 * from the tree and cannot be used for upward traversals, and the 2394 * ncp may be unnamed. Note that other unrelated operations may 2395 * cause the ncp to be named at any time. 2396 */ 2397 dncp = cache_fromdvp(vp, NULL, 0); 2398 bzero(&nlc, sizeof(nlc)); 2399 newvp = NULLVP; 2400 2401 /* 2402 * If there is no cookie, assume directory was stale. 2403 */ 2404 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2405 if (cookiep) 2406 cookie = *cookiep; 2407 else 2408 return (NFSERR_BAD_COOKIE); 2409 /* 2410 * Loop around doing readdir rpc's of size nm_readdirsize 2411 * truncated to a multiple of DIRBLKSIZ. 2412 * The stopping criteria is EOF or buffer full. 2413 */ 2414 while (more_dirs && bigenough) { 2415 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2416 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2417 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2418 nfsm_fhtom(vp, 1); 2419 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2420 *tl++ = cookie.nfsuquad[0]; 2421 *tl++ = cookie.nfsuquad[1]; 2422 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2423 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2424 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2425 *tl = txdr_unsigned(nmp->nm_rsize); 2426 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2427 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2428 if (error) { 2429 m_freem(mrep); 2430 goto nfsmout; 2431 } 2432 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2433 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2434 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2435 more_dirs = fxdr_unsigned(int, *tl); 2436 2437 /* loop thru the dir entries, doctoring them to 4bsd form */ 2438 while (more_dirs && bigenough) { 2439 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2440 fileno = fxdr_hyper(tl); 2441 len = fxdr_unsigned(int, *(tl + 2)); 2442 if (len <= 0 || len > NFS_MAXNAMLEN) { 2443 error = EBADRPC; 2444 m_freem(mrep); 2445 goto nfsmout; 2446 } 2447 tlen = nfsm_rndup(len); 2448 if (tlen == len) 2449 tlen += 4; /* To ensure null termination*/ 2450 left = DIRBLKSIZ - blksiz; 2451 if ((tlen + DIRHDSIZ) > left) { 2452 dp->d_reclen += left; 2453 uiop->uio_iov->iov_base += left; 2454 uiop->uio_iov->iov_len -= left; 2455 uiop->uio_offset += left; 2456 uiop->uio_resid -= left; 2457 blksiz = 0; 2458 } 2459 if ((tlen + DIRHDSIZ) > uiop->uio_resid) 2460 bigenough = 0; 2461 if (bigenough) { 2462 dp = (struct dirent *)uiop->uio_iov->iov_base; 2463 dp->d_fileno = (int)fileno; 2464 dp->d_namlen = len; 2465 dp->d_reclen = tlen + DIRHDSIZ; 2466 dp->d_type = DT_UNKNOWN; 2467 blksiz += dp->d_reclen; 2468 if (blksiz == DIRBLKSIZ) 2469 blksiz = 0; 2470 uiop->uio_offset += DIRHDSIZ; 2471 uiop->uio_resid -= DIRHDSIZ; 2472 uiop->uio_iov->iov_base += DIRHDSIZ; 2473 uiop->uio_iov->iov_len -= DIRHDSIZ; 2474 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2475 nlc.nlc_namelen = len; 2476 nfsm_mtouio(uiop, len); 2477 cp = uiop->uio_iov->iov_base; 2478 tlen -= len; 2479 *cp = '\0'; 2480 uiop->uio_iov->iov_base += tlen; 2481 uiop->uio_iov->iov_len -= tlen; 2482 uiop->uio_offset += tlen; 2483 uiop->uio_resid -= tlen; 2484 } else 2485 nfsm_adv(nfsm_rndup(len)); 2486 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2487 if (bigenough) { 2488 cookie.nfsuquad[0] = *tl++; 2489 cookie.nfsuquad[1] = *tl++; 2490 } else 2491 tl += 2; 2492 2493 /* 2494 * Since the attributes are before the file handle 2495 * (sigh), we must skip over the attributes and then 2496 * come back and get them. 2497 */ 2498 attrflag = fxdr_unsigned(int, *tl); 2499 if (attrflag) { 2500 dpossav1 = dpos; 2501 mdsav1 = md; 2502 nfsm_adv(NFSX_V3FATTR); 2503 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2504 doit = fxdr_unsigned(int, *tl); 2505 if (doit) { 2506 nfsm_getfh(fhp, fhsize, 1); 2507 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2508 vref(vp); 2509 newvp = vp; 2510 np = dnp; 2511 } else { 2512 error = nfs_nget(vp->v_mount, fhp, 2513 fhsize, &np); 2514 if (error) 2515 doit = 0; 2516 else 2517 newvp = NFSTOV(np); 2518 } 2519 } 2520 if (doit && bigenough) { 2521 dpossav2 = dpos; 2522 dpos = dpossav1; 2523 mdsav2 = md; 2524 md = mdsav1; 2525 nfsm_loadattr(newvp, (struct vattr *)0); 2526 dpos = dpossav2; 2527 md = mdsav2; 2528 dp->d_type = 2529 IFTODT(VTTOIF(np->n_vattr.va_type)); 2530 if (dncp) { 2531 printf("NFS/READDIRPLUS, ENTER %*.*s\n", 2532 nlc.nlc_namelen, nlc.nlc_namelen, 2533 nlc.nlc_nameptr); 2534 ncp = cache_nlookup(dncp, &nlc); 2535 cache_setunresolved(ncp); 2536 cache_setvp(ncp, newvp); 2537 cache_put(ncp); 2538 } else { 2539 printf("NFS/READDIRPLUS, UNABLE TO ENTER" 2540 " %*.*s\n", 2541 nlc.nlc_namelen, nlc.nlc_namelen, 2542 nlc.nlc_nameptr); 2543 } 2544 } 2545 } else { 2546 /* Just skip over the file handle */ 2547 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2548 i = fxdr_unsigned(int, *tl); 2549 nfsm_adv(nfsm_rndup(i)); 2550 } 2551 if (newvp != NULLVP) { 2552 if (newvp == vp) 2553 vrele(newvp); 2554 else 2555 vput(newvp); 2556 newvp = NULLVP; 2557 } 2558 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2559 more_dirs = fxdr_unsigned(int, *tl); 2560 } 2561 /* 2562 * If at end of rpc data, get the eof boolean 2563 */ 2564 if (!more_dirs) { 2565 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2566 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2567 } 2568 m_freem(mrep); 2569 } 2570 /* 2571 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2572 * by increasing d_reclen for the last record. 2573 */ 2574 if (blksiz > 0) { 2575 left = DIRBLKSIZ - blksiz; 2576 dp->d_reclen += left; 2577 uiop->uio_iov->iov_base += left; 2578 uiop->uio_iov->iov_len -= left; 2579 uiop->uio_offset += left; 2580 uiop->uio_resid -= left; 2581 } 2582 2583 /* 2584 * We are now either at the end of the directory or have filled the 2585 * block. 2586 */ 2587 if (bigenough) 2588 dnp->n_direofoffset = uiop->uio_offset; 2589 else { 2590 if (uiop->uio_resid > 0) 2591 printf("EEK! readdirplusrpc resid > 0\n"); 2592 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2593 *cookiep = cookie; 2594 } 2595 nfsmout: 2596 if (newvp != NULLVP) { 2597 if (newvp == vp) 2598 vrele(newvp); 2599 else 2600 vput(newvp); 2601 newvp = NULLVP; 2602 } 2603 if (dncp) 2604 cache_drop(dncp); 2605 return (error); 2606 } 2607 2608 /* 2609 * Silly rename. To make the NFS filesystem that is stateless look a little 2610 * more like the "ufs" a remove of an active vnode is translated to a rename 2611 * to a funny looking filename that is removed by nfs_inactive on the 2612 * nfsnode. There is the potential for another process on a different client 2613 * to create the same funny name between the nfs_lookitup() fails and the 2614 * nfs_rename() completes, but... 2615 */ 2616 static int 2617 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2618 { 2619 struct sillyrename *sp; 2620 struct nfsnode *np; 2621 int error; 2622 2623 /* 2624 * We previously purged dvp instead of vp. I don't know why, it 2625 * completely destroys performance. We can't do it anyway with the 2626 * new VFS API since we would be breaking the namecache topology. 2627 */ 2628 cache_purge(vp); /* XXX */ 2629 np = VTONFS(vp); 2630 #ifndef DIAGNOSTIC 2631 if (vp->v_type == VDIR) 2632 panic("nfs: sillyrename dir"); 2633 #endif 2634 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2635 M_NFSREQ, M_WAITOK); 2636 sp->s_cred = crdup(cnp->cn_cred); 2637 sp->s_dvp = dvp; 2638 vref(dvp); 2639 2640 /* Fudge together a funny name */ 2641 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2642 2643 /* Try lookitups until we get one that isn't there */ 2644 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2645 cnp->cn_td, (struct nfsnode **)0) == 0) { 2646 sp->s_name[4]++; 2647 if (sp->s_name[4] > 'z') { 2648 error = EINVAL; 2649 goto bad; 2650 } 2651 } 2652 error = nfs_renameit(dvp, cnp, sp); 2653 if (error) 2654 goto bad; 2655 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2656 cnp->cn_td, &np); 2657 np->n_sillyrename = sp; 2658 return (0); 2659 bad: 2660 vrele(sp->s_dvp); 2661 crfree(sp->s_cred); 2662 free((caddr_t)sp, M_NFSREQ); 2663 return (error); 2664 } 2665 2666 /* 2667 * Look up a file name and optionally either update the file handle or 2668 * allocate an nfsnode, depending on the value of npp. 2669 * npp == NULL --> just do the lookup 2670 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2671 * handled too 2672 * *npp != NULL --> update the file handle in the vnode 2673 */ 2674 static int 2675 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2676 struct thread *td, struct nfsnode **npp) 2677 { 2678 u_int32_t *tl; 2679 caddr_t cp; 2680 int32_t t1, t2; 2681 struct vnode *newvp = (struct vnode *)0; 2682 struct nfsnode *np, *dnp = VTONFS(dvp); 2683 caddr_t bpos, dpos, cp2; 2684 int error = 0, fhlen, attrflag; 2685 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2686 nfsfh_t *nfhp; 2687 int v3 = NFS_ISV3(dvp); 2688 2689 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2690 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2691 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2692 nfsm_fhtom(dvp, v3); 2693 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2694 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2695 if (npp && !error) { 2696 nfsm_getfh(nfhp, fhlen, v3); 2697 if (*npp) { 2698 np = *npp; 2699 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2700 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2701 np->n_fhp = &np->n_fh; 2702 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2703 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2704 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2705 np->n_fhsize = fhlen; 2706 newvp = NFSTOV(np); 2707 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2708 vref(dvp); 2709 newvp = dvp; 2710 } else { 2711 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2712 if (error) { 2713 m_freem(mrep); 2714 return (error); 2715 } 2716 newvp = NFSTOV(np); 2717 } 2718 if (v3) { 2719 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2720 if (!attrflag && *npp == NULL) { 2721 m_freem(mrep); 2722 if (newvp == dvp) 2723 vrele(newvp); 2724 else 2725 vput(newvp); 2726 return (ENOENT); 2727 } 2728 } else 2729 nfsm_loadattr(newvp, (struct vattr *)0); 2730 } 2731 m_freem(mrep); 2732 nfsmout: 2733 if (npp && *npp == NULL) { 2734 if (error) { 2735 if (newvp) { 2736 if (newvp == dvp) 2737 vrele(newvp); 2738 else 2739 vput(newvp); 2740 } 2741 } else 2742 *npp = np; 2743 } 2744 return (error); 2745 } 2746 2747 /* 2748 * Nfs Version 3 commit rpc 2749 */ 2750 int 2751 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2752 { 2753 caddr_t cp; 2754 u_int32_t *tl; 2755 int32_t t1, t2; 2756 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2757 caddr_t bpos, dpos, cp2; 2758 int error = 0, wccflag = NFSV3_WCCRATTR; 2759 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2760 2761 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2762 return (0); 2763 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2764 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2765 nfsm_fhtom(vp, 1); 2766 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2767 txdr_hyper(offset, tl); 2768 tl += 2; 2769 *tl = txdr_unsigned(cnt); 2770 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2771 nfsm_wcc_data(vp, wccflag); 2772 if (!error) { 2773 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2774 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2775 NFSX_V3WRITEVERF)) { 2776 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2777 NFSX_V3WRITEVERF); 2778 error = NFSERR_STALEWRITEVERF; 2779 } 2780 } 2781 m_freem(mrep); 2782 nfsmout: 2783 return (error); 2784 } 2785 2786 /* 2787 * Kludge City.. 2788 * - make nfs_bmap() essentially a no-op that does no translation 2789 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2790 * (Maybe I could use the process's page mapping, but I was concerned that 2791 * Kernel Write might not be enabled and also figured copyout() would do 2792 * a lot more work than bcopy() and also it currently happens in the 2793 * context of the swapper process (2). 2794 * 2795 * nfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp, 2796 * daddr_t *a_bnp, int *a_runp, int *a_runb) 2797 */ 2798 static int 2799 nfs_bmap(struct vop_bmap_args *ap) 2800 { 2801 struct vnode *vp = ap->a_vp; 2802 2803 if (ap->a_vpp != NULL) 2804 *ap->a_vpp = vp; 2805 if (ap->a_bnp != NULL) 2806 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize); 2807 if (ap->a_runp != NULL) 2808 *ap->a_runp = 0; 2809 if (ap->a_runb != NULL) 2810 *ap->a_runb = 0; 2811 return (0); 2812 } 2813 2814 /* 2815 * Strategy routine. 2816 * For async requests when nfsiod(s) are running, queue the request by 2817 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2818 * request. 2819 */ 2820 static int 2821 nfs_strategy(struct vop_strategy_args *ap) 2822 { 2823 struct buf *bp = ap->a_bp; 2824 struct thread *td; 2825 int error = 0; 2826 2827 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2828 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp)); 2829 2830 if (bp->b_flags & B_PHYS) 2831 panic("nfs physio"); 2832 2833 if (bp->b_flags & B_ASYNC) 2834 td = NULL; 2835 else 2836 td = curthread; /* XXX */ 2837 2838 /* 2839 * If the op is asynchronous and an i/o daemon is waiting 2840 * queue the request, wake it up and wait for completion 2841 * otherwise just do it ourselves. 2842 */ 2843 if ((bp->b_flags & B_ASYNC) == 0 || 2844 nfs_asyncio(bp, td)) 2845 error = nfs_doio(bp, td); 2846 return (error); 2847 } 2848 2849 /* 2850 * Mmap a file 2851 * 2852 * NB Currently unsupported. 2853 * 2854 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2855 * struct thread *a_td) 2856 */ 2857 /* ARGSUSED */ 2858 static int 2859 nfs_mmap(struct vop_mmap_args *ap) 2860 { 2861 return (EINVAL); 2862 } 2863 2864 /* 2865 * fsync vnode op. Just call nfs_flush() with commit == 1. 2866 * 2867 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp, 2868 * struct ucred * a_cred, int a_waitfor, struct thread *a_td) 2869 */ 2870 /* ARGSUSED */ 2871 static int 2872 nfs_fsync(struct vop_fsync_args *ap) 2873 { 2874 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1)); 2875 } 2876 2877 /* 2878 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 2879 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 2880 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 2881 * set the buffer contains data that has already been written to the server 2882 * and which now needs a commit RPC. 2883 * 2884 * If commit is 0 we only take one pass and only flush buffers containing new 2885 * dirty data. 2886 * 2887 * If commit is 1 we take two passes, issuing a commit RPC in the second 2888 * pass. 2889 * 2890 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 2891 * to completely flush all pending data. 2892 * 2893 * Note that the RB_SCAN code properly handles the case where the 2894 * callback might block and directly or indirectly (another thread) cause 2895 * the RB tree to change. 2896 */ 2897 2898 #ifndef NFS_COMMITBVECSIZ 2899 #define NFS_COMMITBVECSIZ 16 2900 #endif 2901 2902 struct nfs_flush_info { 2903 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 2904 struct thread *td; 2905 struct vnode *vp; 2906 int waitfor; 2907 int slpflag; 2908 int slptimeo; 2909 int loops; 2910 struct buf *bvary[NFS_COMMITBVECSIZ]; 2911 int bvsize; 2912 off_t beg_off; 2913 off_t end_off; 2914 }; 2915 2916 static int nfs_flush_bp(struct buf *bp, void *data); 2917 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 2918 2919 int 2920 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2921 { 2922 struct nfsnode *np = VTONFS(vp); 2923 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2924 struct nfs_flush_info info; 2925 int error; 2926 2927 bzero(&info, sizeof(info)); 2928 info.td = td; 2929 info.vp = vp; 2930 info.waitfor = waitfor; 2931 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 2932 info.loops = 0; 2933 2934 do { 2935 /* 2936 * Flush mode 2937 */ 2938 info.mode = NFI_FLUSHNEW; 2939 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2940 nfs_flush_bp, &info); 2941 2942 /* 2943 * Take a second pass if committing and no error occured. 2944 * Clean up any left over collection (whether an error 2945 * occurs or not). 2946 */ 2947 if (commit && error == 0) { 2948 info.mode = NFI_COMMIT; 2949 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2950 nfs_flush_bp, &info); 2951 if (info.bvsize) 2952 error = nfs_flush_docommit(&info, error); 2953 } 2954 2955 /* 2956 * Wait for pending I/O to complete before checking whether 2957 * any further dirty buffers exist. 2958 */ 2959 while (waitfor == MNT_WAIT && vp->v_numoutput) { 2960 vp->v_flag |= VBWAIT; 2961 error = tsleep((caddr_t)&vp->v_numoutput, 2962 info.slpflag, "nfsfsync", info.slptimeo); 2963 if (error) { 2964 /* 2965 * We have to be able to break out if this 2966 * is an 'intr' mount. 2967 */ 2968 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2969 error = -EINTR; 2970 break; 2971 } 2972 2973 /* 2974 * Since we do not process pending signals, 2975 * once we get a PCATCH our tsleep() will no 2976 * longer sleep, switch to a fixed timeout 2977 * instead. 2978 */ 2979 if (info.slpflag == PCATCH) { 2980 info.slpflag = 0; 2981 info.slptimeo = 2 * hz; 2982 } 2983 error = 0; 2984 } 2985 } 2986 ++info.loops; 2987 /* 2988 * Loop if we are flushing synchronous as well as committing, 2989 * and dirty buffers are still present. Otherwise we might livelock. 2990 */ 2991 } while (waitfor == MNT_WAIT && commit && 2992 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 2993 2994 /* 2995 * The callbacks have to return a negative error to terminate the 2996 * RB scan. 2997 */ 2998 if (error < 0) 2999 error = -error; 3000 3001 /* 3002 * Deal with any error collection 3003 */ 3004 if (np->n_flag & NWRITEERR) { 3005 error = np->n_error; 3006 np->n_flag &= ~NWRITEERR; 3007 } 3008 return (error); 3009 } 3010 3011 3012 static 3013 int 3014 nfs_flush_bp(struct buf *bp, void *data) 3015 { 3016 struct nfs_flush_info *info = data; 3017 off_t toff; 3018 int error; 3019 int s; 3020 3021 error = 0; 3022 switch(info->mode) { 3023 case NFI_FLUSHNEW: 3024 s = splbio(); 3025 if (info->loops && info->waitfor == MNT_WAIT) { 3026 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3027 if (error) { 3028 error = BUF_TIMELOCK(bp, 3029 LK_EXCLUSIVE | LK_SLEEPFAIL, 3030 "nfsfsync", 3031 info->slpflag, info->slptimeo); 3032 } 3033 } else { 3034 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3035 } 3036 if (error == 0) { 3037 if ((bp->b_flags & B_DELWRI) == 0) 3038 panic("nfs_fsync: not dirty"); 3039 if (bp->b_flags & B_NEEDCOMMIT) { 3040 BUF_UNLOCK(bp); 3041 splx(s); 3042 break; 3043 } 3044 bremfree(bp); 3045 3046 bp->b_flags |= B_ASYNC; 3047 splx(s); 3048 VOP_BWRITE(bp->b_vp, bp); 3049 } else { 3050 splx(s); 3051 error = 0; 3052 } 3053 break; 3054 case NFI_COMMIT: 3055 /* 3056 * Only process buffers in need of a commit which we can 3057 * immediately lock. This may prevent a buffer from being 3058 * committed, but the normal flush loop will block on the 3059 * same buffer so we shouldn't get into an endless loop. 3060 */ 3061 s = splbio(); 3062 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3063 (B_DELWRI | B_NEEDCOMMIT) || 3064 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 3065 splx(s); 3066 break; 3067 } 3068 3069 bremfree(bp); 3070 3071 /* 3072 * NOTE: we are not clearing B_DONE here, so we have 3073 * to do it later on in this routine if we intend to 3074 * initiate I/O on the bp. 3075 * 3076 * Note: to avoid loopback deadlocks, we do not 3077 * assign b_runningbufspace. 3078 */ 3079 vfs_busy_pages(bp, 1); 3080 3081 info->bvary[info->bvsize] = bp; 3082 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3083 bp->b_dirtyoff; 3084 if (info->bvsize == 0 || toff < info->beg_off) 3085 info->beg_off = toff; 3086 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3087 if (info->bvsize == 0 || toff > info->end_off) 3088 info->end_off = toff; 3089 ++info->bvsize; 3090 if (info->bvsize == NFS_COMMITBVECSIZ) { 3091 error = nfs_flush_docommit(info, 0); 3092 KKASSERT(info->bvsize == 0); 3093 } 3094 splx(s); 3095 } 3096 return (error); 3097 } 3098 3099 static 3100 int 3101 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3102 { 3103 struct vnode *vp; 3104 struct buf *bp; 3105 off_t bytes; 3106 int retv; 3107 int i; 3108 int s; 3109 3110 vp = info->vp; 3111 3112 if (info->bvsize > 0) { 3113 /* 3114 * Commit data on the server, as required. Note that 3115 * nfs_commit will use the vnode's cred for the commit. 3116 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3117 */ 3118 bytes = info->end_off - info->beg_off; 3119 if (bytes > 0x40000000) 3120 bytes = 0x40000000; 3121 if (error) { 3122 retv = -error; 3123 } else { 3124 retv = nfs_commit(vp, info->beg_off, 3125 (int)bytes, info->td); 3126 if (retv == NFSERR_STALEWRITEVERF) 3127 nfs_clearcommit(vp->v_mount); 3128 } 3129 3130 /* 3131 * Now, either mark the blocks I/O done or mark the 3132 * blocks dirty, depending on whether the commit 3133 * succeeded. 3134 */ 3135 for (i = 0; i < info->bvsize; ++i) { 3136 bp = info->bvary[i]; 3137 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3138 if (retv) { 3139 /* 3140 * Error, leave B_DELWRI intact 3141 */ 3142 vfs_unbusy_pages(bp); 3143 brelse(bp); 3144 } else { 3145 /* 3146 * Success, remove B_DELWRI ( bundirty() ). 3147 * 3148 * b_dirtyoff/b_dirtyend seem to be NFS 3149 * specific. We should probably move that 3150 * into bundirty(). XXX 3151 */ 3152 s = splbio(); 3153 vp->v_numoutput++; 3154 bp->b_flags |= B_ASYNC; 3155 bundirty(bp); 3156 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3157 bp->b_dirtyoff = bp->b_dirtyend = 0; 3158 splx(s); 3159 biodone(bp); 3160 } 3161 } 3162 info->bvsize = 0; 3163 } 3164 return (error); 3165 } 3166 3167 /* 3168 * NFS advisory byte-level locks. 3169 * Currently unsupported. 3170 * 3171 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3172 * int a_flags) 3173 */ 3174 static int 3175 nfs_advlock(struct vop_advlock_args *ap) 3176 { 3177 struct nfsnode *np = VTONFS(ap->a_vp); 3178 3179 /* 3180 * The following kludge is to allow diskless support to work 3181 * until a real NFS lockd is implemented. Basically, just pretend 3182 * that this is a local lock. 3183 */ 3184 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3185 } 3186 3187 /* 3188 * Print out the contents of an nfsnode. 3189 * 3190 * nfs_print(struct vnode *a_vp) 3191 */ 3192 static int 3193 nfs_print(struct vop_print_args *ap) 3194 { 3195 struct vnode *vp = ap->a_vp; 3196 struct nfsnode *np = VTONFS(vp); 3197 3198 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3199 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3200 if (vp->v_type == VFIFO) 3201 fifo_printinfo(vp); 3202 printf("\n"); 3203 return (0); 3204 } 3205 3206 /* 3207 * Just call nfs_writebp() with the force argument set to 1. 3208 * 3209 * NOTE: B_DONE may or may not be set in a_bp on call. 3210 * 3211 * nfs_bwrite(struct vnode *a_bp) 3212 */ 3213 static int 3214 nfs_bwrite(struct vop_bwrite_args *ap) 3215 { 3216 return (nfs_writebp(ap->a_bp, 1, curthread)); 3217 } 3218 3219 /* 3220 * This is a clone of vn_bwrite(), except that it also handles the 3221 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer. 3222 */ 3223 int 3224 nfs_writebp(struct buf *bp, int force, struct thread *td) 3225 { 3226 int s; 3227 int oldflags = bp->b_flags; 3228 #if 0 3229 int retv = 1; 3230 off_t off; 3231 #endif 3232 3233 if (BUF_REFCNT(bp) == 0) 3234 panic("bwrite: buffer is not locked???"); 3235 3236 if (bp->b_flags & B_INVAL) { 3237 brelse(bp); 3238 return(0); 3239 } 3240 3241 bp->b_flags |= B_CACHE; 3242 3243 /* 3244 * Undirty the bp. We will redirty it later if the I/O fails. 3245 */ 3246 3247 s = splbio(); 3248 bundirty(bp); 3249 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3250 3251 bp->b_vp->v_numoutput++; 3252 splx(s); 3253 3254 /* 3255 * Note: to avoid loopback deadlocks, we do not 3256 * assign b_runningbufspace. 3257 */ 3258 vfs_busy_pages(bp, 1); 3259 3260 BUF_KERNPROC(bp); 3261 VOP_STRATEGY(bp->b_vp, bp); 3262 3263 if( (oldflags & B_ASYNC) == 0) { 3264 int rtval = biowait(bp); 3265 3266 if (oldflags & B_DELWRI) { 3267 s = splbio(); 3268 reassignbuf(bp, bp->b_vp); 3269 splx(s); 3270 } 3271 3272 brelse(bp); 3273 return (rtval); 3274 } 3275 3276 return (0); 3277 } 3278 3279 /* 3280 * nfs special file access vnode op. 3281 * Essentially just get vattr and then imitate iaccess() since the device is 3282 * local to the client. 3283 * 3284 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3285 * struct thread *a_td) 3286 */ 3287 static int 3288 nfsspec_access(struct vop_access_args *ap) 3289 { 3290 struct vattr *vap; 3291 gid_t *gp; 3292 struct ucred *cred = ap->a_cred; 3293 struct vnode *vp = ap->a_vp; 3294 mode_t mode = ap->a_mode; 3295 struct vattr vattr; 3296 int i; 3297 int error; 3298 3299 /* 3300 * Disallow write attempts on filesystems mounted read-only; 3301 * unless the file is a socket, fifo, or a block or character 3302 * device resident on the filesystem. 3303 */ 3304 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3305 switch (vp->v_type) { 3306 case VREG: 3307 case VDIR: 3308 case VLNK: 3309 return (EROFS); 3310 default: 3311 break; 3312 } 3313 } 3314 /* 3315 * If you're the super-user, 3316 * you always get access. 3317 */ 3318 if (cred->cr_uid == 0) 3319 return (0); 3320 vap = &vattr; 3321 error = VOP_GETATTR(vp, vap, ap->a_td); 3322 if (error) 3323 return (error); 3324 /* 3325 * Access check is based on only one of owner, group, public. 3326 * If not owner, then check group. If not a member of the 3327 * group, then check public access. 3328 */ 3329 if (cred->cr_uid != vap->va_uid) { 3330 mode >>= 3; 3331 gp = cred->cr_groups; 3332 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3333 if (vap->va_gid == *gp) 3334 goto found; 3335 mode >>= 3; 3336 found: 3337 ; 3338 } 3339 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3340 return (error); 3341 } 3342 3343 /* 3344 * Read wrapper for special devices. 3345 * 3346 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3347 * struct ucred *a_cred) 3348 */ 3349 static int 3350 nfsspec_read(struct vop_read_args *ap) 3351 { 3352 struct nfsnode *np = VTONFS(ap->a_vp); 3353 3354 /* 3355 * Set access flag. 3356 */ 3357 np->n_flag |= NACC; 3358 getnanotime(&np->n_atim); 3359 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3360 } 3361 3362 /* 3363 * Write wrapper for special devices. 3364 * 3365 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3366 * struct ucred *a_cred) 3367 */ 3368 static int 3369 nfsspec_write(struct vop_write_args *ap) 3370 { 3371 struct nfsnode *np = VTONFS(ap->a_vp); 3372 3373 /* 3374 * Set update flag. 3375 */ 3376 np->n_flag |= NUPD; 3377 getnanotime(&np->n_mtim); 3378 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3379 } 3380 3381 /* 3382 * Close wrapper for special devices. 3383 * 3384 * Update the times on the nfsnode then do device close. 3385 * 3386 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3387 * struct thread *a_td) 3388 */ 3389 static int 3390 nfsspec_close(struct vop_close_args *ap) 3391 { 3392 struct vnode *vp = ap->a_vp; 3393 struct nfsnode *np = VTONFS(vp); 3394 struct vattr vattr; 3395 3396 if (np->n_flag & (NACC | NUPD)) { 3397 np->n_flag |= NCHG; 3398 if (vp->v_usecount == 1 && 3399 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3400 VATTR_NULL(&vattr); 3401 if (np->n_flag & NACC) 3402 vattr.va_atime = np->n_atim; 3403 if (np->n_flag & NUPD) 3404 vattr.va_mtime = np->n_mtim; 3405 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3406 } 3407 } 3408 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3409 } 3410 3411 /* 3412 * Read wrapper for fifos. 3413 * 3414 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3415 * struct ucred *a_cred) 3416 */ 3417 static int 3418 nfsfifo_read(struct vop_read_args *ap) 3419 { 3420 struct nfsnode *np = VTONFS(ap->a_vp); 3421 3422 /* 3423 * Set access flag. 3424 */ 3425 np->n_flag |= NACC; 3426 getnanotime(&np->n_atim); 3427 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3428 } 3429 3430 /* 3431 * Write wrapper for fifos. 3432 * 3433 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3434 * struct ucred *a_cred) 3435 */ 3436 static int 3437 nfsfifo_write(struct vop_write_args *ap) 3438 { 3439 struct nfsnode *np = VTONFS(ap->a_vp); 3440 3441 /* 3442 * Set update flag. 3443 */ 3444 np->n_flag |= NUPD; 3445 getnanotime(&np->n_mtim); 3446 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3447 } 3448 3449 /* 3450 * Close wrapper for fifos. 3451 * 3452 * Update the times on the nfsnode then do fifo close. 3453 * 3454 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3455 */ 3456 static int 3457 nfsfifo_close(struct vop_close_args *ap) 3458 { 3459 struct vnode *vp = ap->a_vp; 3460 struct nfsnode *np = VTONFS(vp); 3461 struct vattr vattr; 3462 struct timespec ts; 3463 3464 if (np->n_flag & (NACC | NUPD)) { 3465 getnanotime(&ts); 3466 if (np->n_flag & NACC) 3467 np->n_atim = ts; 3468 if (np->n_flag & NUPD) 3469 np->n_mtim = ts; 3470 np->n_flag |= NCHG; 3471 if (vp->v_usecount == 1 && 3472 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3473 VATTR_NULL(&vattr); 3474 if (np->n_flag & NACC) 3475 vattr.va_atime = np->n_atim; 3476 if (np->n_flag & NUPD) 3477 vattr.va_mtime = np->n_mtim; 3478 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3479 } 3480 } 3481 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3482 } 3483 3484