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 1.49 2006/03/05 18:38:37 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/nlookup.h> 59 #include <sys/socket.h> 60 #include <sys/vnode.h> 61 #include <sys/dirent.h> 62 #include <sys/fcntl.h> 63 #include <sys/lockf.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/conf.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_zone.h> 71 72 #include <sys/buf2.h> 73 74 #include <vfs/fifofs/fifo.h> 75 #include <vfs/ufs/dir.h> 76 77 #undef DIRBLKSIZ 78 79 #include "rpcv2.h" 80 #include "nfsproto.h" 81 #include "nfs.h" 82 #include "nfsmount.h" 83 #include "nfsnode.h" 84 #include "xdr_subs.h" 85 #include "nfsm_subs.h" 86 #include "nqnfs.h" 87 88 #include <net/if.h> 89 #include <netinet/in.h> 90 #include <netinet/in_var.h> 91 92 #include <sys/thread2.h> 93 94 /* Defs */ 95 #define TRUE 1 96 #define FALSE 0 97 98 /* 99 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 100 * calls are not in getblk() and brelse() so that they would not be necessary 101 * here. 102 */ 103 #ifndef B_VMIO 104 #define vfs_busy_pages(bp, f) 105 #endif 106 107 static int nfsspec_read (struct vop_read_args *); 108 static int nfsspec_write (struct vop_write_args *); 109 static int nfsfifo_read (struct vop_read_args *); 110 static int nfsfifo_write (struct vop_write_args *); 111 static int nfsspec_close (struct vop_close_args *); 112 static int nfsfifo_close (struct vop_close_args *); 113 #define nfs_poll vop_nopoll 114 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 115 static int nfs_lookup (struct vop_old_lookup_args *); 116 static int nfs_create (struct vop_old_create_args *); 117 static int nfs_mknod (struct vop_old_mknod_args *); 118 static int nfs_open (struct vop_open_args *); 119 static int nfs_close (struct vop_close_args *); 120 static int nfs_access (struct vop_access_args *); 121 static int nfs_getattr (struct vop_getattr_args *); 122 static int nfs_setattr (struct vop_setattr_args *); 123 static int nfs_read (struct vop_read_args *); 124 static int nfs_mmap (struct vop_mmap_args *); 125 static int nfs_fsync (struct vop_fsync_args *); 126 static int nfs_remove (struct vop_old_remove_args *); 127 static int nfs_link (struct vop_old_link_args *); 128 static int nfs_rename (struct vop_old_rename_args *); 129 static int nfs_mkdir (struct vop_old_mkdir_args *); 130 static int nfs_rmdir (struct vop_old_rmdir_args *); 131 static int nfs_symlink (struct vop_old_symlink_args *); 132 static int nfs_readdir (struct vop_readdir_args *); 133 static int nfs_bmap (struct vop_bmap_args *); 134 static int nfs_strategy (struct vop_strategy_args *); 135 static int nfs_lookitup (struct vnode *, const char *, int, 136 struct ucred *, struct thread *, struct nfsnode **); 137 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 138 static int nfsspec_access (struct vop_access_args *); 139 static int nfs_readlink (struct vop_readlink_args *); 140 static int nfs_print (struct vop_print_args *); 141 static int nfs_advlock (struct vop_advlock_args *); 142 static int nfs_bwrite (struct vop_bwrite_args *); 143 144 static int nfs_nresolve (struct vop_nresolve_args *); 145 /* 146 * Global vfs data structures for nfs 147 */ 148 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 149 { &vop_default_desc, vop_defaultop }, 150 { &vop_access_desc, (vnodeopv_entry_t) nfs_access }, 151 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock }, 152 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap }, 153 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite }, 154 { &vop_close_desc, (vnodeopv_entry_t) nfs_close }, 155 { &vop_old_create_desc, (vnodeopv_entry_t) nfs_create }, 156 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 157 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 158 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages }, 159 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages }, 160 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 161 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 162 { &vop_lease_desc, vop_null }, 163 { &vop_old_link_desc, (vnodeopv_entry_t) nfs_link }, 164 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 165 { &vop_old_lookup_desc, (vnodeopv_entry_t) nfs_lookup }, 166 { &vop_old_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir }, 167 { &vop_old_mknod_desc, (vnodeopv_entry_t) nfs_mknod }, 168 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap }, 169 { &vop_open_desc, (vnodeopv_entry_t) nfs_open }, 170 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll }, 171 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 172 { &vop_read_desc, (vnodeopv_entry_t) nfs_read }, 173 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir }, 174 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink }, 175 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 176 { &vop_old_remove_desc, (vnodeopv_entry_t) nfs_remove }, 177 { &vop_old_rename_desc, (vnodeopv_entry_t) nfs_rename }, 178 { &vop_old_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir }, 179 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 180 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy }, 181 { &vop_old_symlink_desc, (vnodeopv_entry_t) nfs_symlink }, 182 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 183 { &vop_write_desc, (vnodeopv_entry_t) nfs_write }, 184 185 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve }, 186 { NULL, NULL } 187 }; 188 189 /* 190 * Special device vnode ops 191 */ 192 struct vnodeopv_entry_desc nfsv2_specop_entries[] = { 193 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate }, 194 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 195 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close }, 196 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 197 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 198 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 199 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 200 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 201 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 202 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read }, 203 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 204 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 205 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 206 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write }, 207 { NULL, NULL } 208 }; 209 210 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = { 211 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate }, 212 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 213 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close }, 214 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 215 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 216 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 217 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 218 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 219 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 220 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read }, 221 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 222 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 223 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 224 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write }, 225 { NULL, NULL } 226 }; 227 228 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 229 struct componentname *cnp, 230 struct vattr *vap); 231 static int nfs_removerpc (struct vnode *dvp, const char *name, 232 int namelen, 233 struct ucred *cred, struct thread *td); 234 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 235 int fnamelen, struct vnode *tdvp, 236 const char *tnameptr, int tnamelen, 237 struct ucred *cred, struct thread *td); 238 static int nfs_renameit (struct vnode *sdvp, 239 struct componentname *scnp, 240 struct sillyrename *sp); 241 242 /* 243 * Global variables 244 */ 245 extern u_int32_t nfs_true, nfs_false; 246 extern u_int32_t nfs_xdrneg1; 247 extern struct nfsstats nfsstats; 248 extern nfstype nfsv3_type[9]; 249 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 250 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 251 int nfs_numasync = 0; 252 253 SYSCTL_DECL(_vfs_nfs); 254 255 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 256 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 257 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 258 259 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 261 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 262 263 static int nfsv3_commit_on_close = 0; 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 265 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 266 #if 0 267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 268 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 269 270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 271 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 272 #endif 273 274 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 275 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 276 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 277 static int 278 nfs3_access_otw(struct vnode *vp, int wmode, 279 struct thread *td, struct ucred *cred) 280 { 281 const int v3 = 1; 282 u_int32_t *tl; 283 int error = 0, attrflag; 284 285 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 286 caddr_t bpos, dpos, cp2; 287 int32_t t1, t2; 288 caddr_t cp; 289 u_int32_t rmode; 290 struct nfsnode *np = VTONFS(vp); 291 292 nfsstats.rpccnt[NFSPROC_ACCESS]++; 293 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 294 nfsm_fhtom(vp, v3); 295 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 296 *tl = txdr_unsigned(wmode); 297 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 298 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 299 if (!error) { 300 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 301 rmode = fxdr_unsigned(u_int32_t, *tl); 302 np->n_mode = rmode; 303 np->n_modeuid = cred->cr_uid; 304 np->n_modestamp = mycpu->gd_time_seconds; 305 } 306 m_freem(mrep); 307 nfsmout: 308 return error; 309 } 310 311 /* 312 * nfs access vnode op. 313 * For nfs version 2, just return ok. File accesses may fail later. 314 * For nfs version 3, use the access rpc to check accessibility. If file modes 315 * are changed on the server, accesses might still fail later. 316 * 317 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 318 * struct thread *a_td) 319 */ 320 static int 321 nfs_access(struct vop_access_args *ap) 322 { 323 struct vnode *vp = ap->a_vp; 324 int error = 0; 325 u_int32_t mode, wmode; 326 int v3 = NFS_ISV3(vp); 327 struct nfsnode *np = VTONFS(vp); 328 329 /* 330 * Disallow write attempts on filesystems mounted read-only; 331 * unless the file is a socket, fifo, or a block or character 332 * device resident on the filesystem. 333 */ 334 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 335 switch (vp->v_type) { 336 case VREG: 337 case VDIR: 338 case VLNK: 339 return (EROFS); 340 default: 341 break; 342 } 343 } 344 /* 345 * For nfs v3, check to see if we have done this recently, and if 346 * so return our cached result instead of making an ACCESS call. 347 * If not, do an access rpc, otherwise you are stuck emulating 348 * ufs_access() locally using the vattr. This may not be correct, 349 * since the server may apply other access criteria such as 350 * client uid-->server uid mapping that we do not know about. 351 */ 352 if (v3) { 353 if (ap->a_mode & VREAD) 354 mode = NFSV3ACCESS_READ; 355 else 356 mode = 0; 357 if (vp->v_type != VDIR) { 358 if (ap->a_mode & VWRITE) 359 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 360 if (ap->a_mode & VEXEC) 361 mode |= NFSV3ACCESS_EXECUTE; 362 } else { 363 if (ap->a_mode & VWRITE) 364 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 365 NFSV3ACCESS_DELETE); 366 if (ap->a_mode & VEXEC) 367 mode |= NFSV3ACCESS_LOOKUP; 368 } 369 /* XXX safety belt, only make blanket request if caching */ 370 if (nfsaccess_cache_timeout > 0) { 371 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 372 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 373 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 374 } else { 375 wmode = mode; 376 } 377 378 /* 379 * Does our cached result allow us to give a definite yes to 380 * this request? 381 */ 382 if (np->n_modestamp && 383 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 384 (ap->a_cred->cr_uid == np->n_modeuid) && 385 ((np->n_mode & mode) == mode)) { 386 nfsstats.accesscache_hits++; 387 } else { 388 /* 389 * Either a no, or a don't know. Go to the wire. 390 */ 391 nfsstats.accesscache_misses++; 392 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 393 if (!error) { 394 if ((np->n_mode & mode) != mode) { 395 error = EACCES; 396 } 397 } 398 } 399 } else { 400 if ((error = nfsspec_access(ap)) != 0) 401 return (error); 402 403 /* 404 * Attempt to prevent a mapped root from accessing a file 405 * which it shouldn't. We try to read a byte from the file 406 * if the user is root and the file is not zero length. 407 * After calling nfsspec_access, we should have the correct 408 * file size cached. 409 */ 410 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 411 && VTONFS(vp)->n_size > 0) { 412 struct iovec aiov; 413 struct uio auio; 414 char buf[1]; 415 416 aiov.iov_base = buf; 417 aiov.iov_len = 1; 418 auio.uio_iov = &aiov; 419 auio.uio_iovcnt = 1; 420 auio.uio_offset = 0; 421 auio.uio_resid = 1; 422 auio.uio_segflg = UIO_SYSSPACE; 423 auio.uio_rw = UIO_READ; 424 auio.uio_td = ap->a_td; 425 426 if (vp->v_type == VREG) { 427 error = nfs_readrpc(vp, &auio); 428 } else if (vp->v_type == VDIR) { 429 char* bp; 430 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 431 aiov.iov_base = bp; 432 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 433 error = nfs_readdirrpc(vp, &auio); 434 free(bp, M_TEMP); 435 } else if (vp->v_type == VLNK) { 436 error = nfs_readlinkrpc(vp, &auio); 437 } else { 438 error = EACCES; 439 } 440 } 441 } 442 /* 443 * [re]record creds for reading and/or writing if access 444 * was granted. Assume the NFS server will grant read access 445 * for execute requests. 446 */ 447 if (error == 0) { 448 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 449 crhold(ap->a_cred); 450 if (np->n_rucred) 451 crfree(np->n_rucred); 452 np->n_rucred = ap->a_cred; 453 } 454 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 455 crhold(ap->a_cred); 456 if (np->n_wucred) 457 crfree(np->n_wucred); 458 np->n_wucred = ap->a_cred; 459 } 460 } 461 return(error); 462 } 463 464 /* 465 * nfs open vnode op 466 * Check to see if the type is ok 467 * and that deletion is not in progress. 468 * For paged in text files, you will need to flush the page cache 469 * if consistency is lost. 470 * 471 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 472 * struct thread *a_td) 473 */ 474 /* ARGSUSED */ 475 static int 476 nfs_open(struct vop_open_args *ap) 477 { 478 struct vnode *vp = ap->a_vp; 479 struct nfsnode *np = VTONFS(vp); 480 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 481 struct vattr vattr; 482 int error; 483 484 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 485 #ifdef DIAGNOSTIC 486 printf("open eacces vtyp=%d\n",vp->v_type); 487 #endif 488 return (EOPNOTSUPP); 489 } 490 491 /* 492 * Clear the attribute cache only if opening with write access. It 493 * is unclear if we should do this at all here, but we certainly 494 * should not clear the cache unconditionally simply because a file 495 * is being opened. 496 */ 497 if (ap->a_mode & FWRITE) 498 np->n_attrstamp = 0; 499 500 if (nmp->nm_flag & NFSMNT_NQNFS) { 501 /* 502 * If NQNFS is active, get a valid lease 503 */ 504 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 505 do { 506 error = nqnfs_getlease(vp, ND_READ, ap->a_td); 507 } while (error == NQNFS_EXPIRED); 508 if (error) 509 return (error); 510 if (np->n_lrev != np->n_brev || 511 (np->n_flag & NQNFSNONCACHE)) { 512 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 513 == EINTR) { 514 return (error); 515 } 516 np->n_brev = np->n_lrev; 517 } 518 } 519 } else { 520 /* 521 * For normal NFS, reconcile changes made locally verses 522 * changes made remotely. Note that VOP_GETATTR only goes 523 * to the wire if the cached attribute has timed out or been 524 * cleared. 525 * 526 * If local modifications have been made clear the attribute 527 * cache to force an attribute and modified time check. If 528 * GETATTR detects that the file has been changed by someone 529 * other then us it will set NRMODIFIED. 530 * 531 * If we are opening a directory and local changes have been 532 * made we have to invalidate the cache in order to ensure 533 * that we get the most up-to-date information from the 534 * server. XXX 535 */ 536 if (np->n_flag & NLMODIFIED) { 537 np->n_attrstamp = 0; 538 if (vp->v_type == VDIR) { 539 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 540 if (error == EINTR) 541 return (error); 542 nfs_invaldir(vp); 543 } 544 } 545 error = VOP_GETATTR(vp, &vattr, ap->a_td); 546 if (error) 547 return (error); 548 if (np->n_flag & NRMODIFIED) { 549 if (vp->v_type == VDIR) 550 nfs_invaldir(vp); 551 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 552 if (error == EINTR) 553 return (error); 554 np->n_flag &= ~NRMODIFIED; 555 } 556 } 557 558 return (0); 559 } 560 561 /* 562 * nfs close vnode op 563 * What an NFS client should do upon close after writing is a debatable issue. 564 * Most NFS clients push delayed writes to the server upon close, basically for 565 * two reasons: 566 * 1 - So that any write errors may be reported back to the client process 567 * doing the close system call. By far the two most likely errors are 568 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 569 * 2 - To put a worst case upper bound on cache inconsistency between 570 * multiple clients for the file. 571 * There is also a consistency problem for Version 2 of the protocol w.r.t. 572 * not being able to tell if other clients are writing a file concurrently, 573 * since there is no way of knowing if the changed modify time in the reply 574 * is only due to the write for this client. 575 * (NFS Version 3 provides weak cache consistency data in the reply that 576 * should be sufficient to detect and handle this case.) 577 * 578 * The current code does the following: 579 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 580 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 581 * or commit them (this satisfies 1 and 2 except for the 582 * case where the server crashes after this close but 583 * before the commit RPC, which is felt to be "good 584 * enough". Changing the last argument to nfs_flush() to 585 * a 1 would force a commit operation, if it is felt a 586 * commit is necessary now. 587 * for NQNFS - do nothing now, since 2 is dealt with via leases and 588 * 1 should be dealt with via an fsync() system call for 589 * cases where write errors are important. 590 * 591 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag, 592 * struct ucred *a_cred, struct thread *a_td) 593 */ 594 /* ARGSUSED */ 595 static int 596 nfs_close(struct vop_close_args *ap) 597 { 598 struct vnode *vp = ap->a_vp; 599 struct nfsnode *np = VTONFS(vp); 600 int error = 0; 601 602 if (vp->v_type == VREG) { 603 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 604 (np->n_flag & NLMODIFIED)) { 605 if (NFS_ISV3(vp)) { 606 /* 607 * Under NFSv3 we have dirty buffers to dispose of. We 608 * must flush them to the NFS server. We have the option 609 * of waiting all the way through the commit rpc or just 610 * waiting for the initial write. The default is to only 611 * wait through the initial write so the data is in the 612 * server's cache, which is roughly similar to the state 613 * a standard disk subsystem leaves the file in on close(). 614 * 615 * We cannot clear the NLMODIFIED bit in np->n_flag due to 616 * potential races with other processes, and certainly 617 * cannot clear it if we don't commit. 618 */ 619 int cm = nfsv3_commit_on_close ? 1 : 0; 620 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm); 621 /* np->n_flag &= ~NLMODIFIED; */ 622 } else { 623 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 624 } 625 np->n_attrstamp = 0; 626 } 627 if (np->n_flag & NWRITEERR) { 628 np->n_flag &= ~NWRITEERR; 629 error = np->n_error; 630 } 631 } 632 return (error); 633 } 634 635 /* 636 * nfs getattr call from vfs. 637 * 638 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 639 * struct thread *a_td) 640 */ 641 static int 642 nfs_getattr(struct vop_getattr_args *ap) 643 { 644 struct vnode *vp = ap->a_vp; 645 struct nfsnode *np = VTONFS(vp); 646 caddr_t cp; 647 u_int32_t *tl; 648 int32_t t1, t2; 649 caddr_t bpos, dpos; 650 int error = 0; 651 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 652 int v3 = NFS_ISV3(vp); 653 654 /* 655 * Update local times for special files. 656 */ 657 if (np->n_flag & (NACC | NUPD)) 658 np->n_flag |= NCHG; 659 /* 660 * First look in the cache. 661 */ 662 if (nfs_getattrcache(vp, ap->a_vap) == 0) 663 return (0); 664 665 if (v3 && nfsaccess_cache_timeout > 0) { 666 nfsstats.accesscache_misses++; 667 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 668 if (nfs_getattrcache(vp, ap->a_vap) == 0) 669 return (0); 670 } 671 672 nfsstats.rpccnt[NFSPROC_GETATTR]++; 673 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 674 nfsm_fhtom(vp, v3); 675 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 676 if (!error) { 677 nfsm_loadattr(vp, ap->a_vap); 678 } 679 m_freem(mrep); 680 nfsmout: 681 return (error); 682 } 683 684 /* 685 * nfs setattr call. 686 * 687 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 688 * struct vattr *a_vap, struct ucred *a_cred, 689 * struct thread *a_td) 690 */ 691 static int 692 nfs_setattr(struct vop_setattr_args *ap) 693 { 694 struct vnode *vp = ap->a_vp; 695 struct nfsnode *np = VTONFS(vp); 696 struct vattr *vap = ap->a_vap; 697 int error = 0; 698 u_quad_t tsize; 699 700 #ifndef nolint 701 tsize = (u_quad_t)0; 702 #endif 703 704 /* 705 * Setting of flags is not supported. 706 */ 707 if (vap->va_flags != VNOVAL) 708 return (EOPNOTSUPP); 709 710 /* 711 * Disallow write attempts if the filesystem is mounted read-only. 712 */ 713 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 714 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 715 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 716 (vp->v_mount->mnt_flag & MNT_RDONLY)) 717 return (EROFS); 718 if (vap->va_size != VNOVAL) { 719 switch (vp->v_type) { 720 case VDIR: 721 return (EISDIR); 722 case VCHR: 723 case VBLK: 724 case VSOCK: 725 case VFIFO: 726 if (vap->va_mtime.tv_sec == VNOVAL && 727 vap->va_atime.tv_sec == VNOVAL && 728 vap->va_mode == (mode_t)VNOVAL && 729 vap->va_uid == (uid_t)VNOVAL && 730 vap->va_gid == (gid_t)VNOVAL) 731 return (0); 732 vap->va_size = VNOVAL; 733 break; 734 default: 735 /* 736 * Disallow write attempts if the filesystem is 737 * mounted read-only. 738 */ 739 if (vp->v_mount->mnt_flag & MNT_RDONLY) 740 return (EROFS); 741 742 /* 743 * This is nasty. The RPCs we send to flush pending 744 * data often return attribute information which is 745 * cached via a callback to nfs_loadattrcache(), which 746 * has the effect of changing our notion of the file 747 * size. Due to flushed appends and other operations 748 * the file size can be set to virtually anything, 749 * including values that do not match either the old 750 * or intended file size. 751 * 752 * When this condition is detected we must loop to 753 * try the operation again. Hopefully no more 754 * flushing is required on the loop so it works the 755 * second time around. THIS CASE ALMOST ALWAYS 756 * HAPPENS! 757 */ 758 tsize = np->n_size; 759 again: 760 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size); 761 762 if (np->n_flag & NLMODIFIED) { 763 if (vap->va_size == 0) 764 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1); 765 else 766 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 767 } 768 /* 769 * note: this loop case almost always happens at 770 * least once per truncation. 771 */ 772 if (error == 0 && np->n_size != vap->va_size) 773 goto again; 774 np->n_vattr.va_size = vap->va_size; 775 break; 776 } 777 } else if ((vap->va_mtime.tv_sec != VNOVAL || 778 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 779 vp->v_type == VREG && 780 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR 781 ) { 782 return (error); 783 } 784 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 785 786 /* 787 * Sanity check if a truncation was issued. This should only occur 788 * if multiple processes are racing on the same file. 789 */ 790 if (error == 0 && vap->va_size != VNOVAL && 791 np->n_size != vap->va_size) { 792 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 793 goto again; 794 } 795 if (error && vap->va_size != VNOVAL) { 796 np->n_size = np->n_vattr.va_size = tsize; 797 vnode_pager_setsize(vp, np->n_size); 798 } 799 return (error); 800 } 801 802 /* 803 * Do an nfs setattr rpc. 804 */ 805 static int 806 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 807 struct ucred *cred, struct thread *td) 808 { 809 struct nfsv2_sattr *sp; 810 struct nfsnode *np = VTONFS(vp); 811 caddr_t cp; 812 int32_t t1, t2; 813 caddr_t bpos, dpos, cp2; 814 u_int32_t *tl; 815 int error = 0, wccflag = NFSV3_WCCRATTR; 816 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 817 int v3 = NFS_ISV3(vp); 818 819 nfsstats.rpccnt[NFSPROC_SETATTR]++; 820 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 821 nfsm_fhtom(vp, v3); 822 if (v3) { 823 nfsm_v3attrbuild(vap, TRUE); 824 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 825 *tl = nfs_false; 826 } else { 827 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 828 if (vap->va_mode == (mode_t)VNOVAL) 829 sp->sa_mode = nfs_xdrneg1; 830 else 831 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 832 if (vap->va_uid == (uid_t)VNOVAL) 833 sp->sa_uid = nfs_xdrneg1; 834 else 835 sp->sa_uid = txdr_unsigned(vap->va_uid); 836 if (vap->va_gid == (gid_t)VNOVAL) 837 sp->sa_gid = nfs_xdrneg1; 838 else 839 sp->sa_gid = txdr_unsigned(vap->va_gid); 840 sp->sa_size = txdr_unsigned(vap->va_size); 841 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 842 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 843 } 844 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 845 if (v3) { 846 np->n_modestamp = 0; 847 nfsm_wcc_data(vp, wccflag); 848 } else 849 nfsm_loadattr(vp, (struct vattr *)0); 850 m_freem(mrep); 851 nfsmout: 852 return (error); 853 } 854 855 /* 856 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 857 * nfs_lookup() until all remaining new api calls are implemented. 858 * 859 * Resolve a namecache entry. This function is passed a locked ncp and 860 * must call cache_setvp() on it as appropriate to resolve the entry. 861 */ 862 static int 863 nfs_nresolve(struct vop_nresolve_args *ap) 864 { 865 struct thread *td = curthread; 866 struct namecache *ncp; 867 struct ucred *cred; 868 struct nfsnode *np; 869 struct vnode *dvp; 870 struct vnode *nvp; 871 nfsfh_t *fhp; 872 int attrflag; 873 int fhsize; 874 int error; 875 int len; 876 int v3; 877 /******NFSM MACROS********/ 878 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 879 caddr_t bpos, dpos, cp, cp2; 880 u_int32_t *tl; 881 int32_t t1, t2; 882 883 cred = ap->a_cred; 884 ncp = ap->a_ncp; 885 886 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp); 887 dvp = ncp->nc_parent->nc_vp; 888 if ((error = vget(dvp, LK_SHARED, td)) != 0) 889 return (error); 890 891 nvp = NULL; 892 v3 = NFS_ISV3(dvp); 893 nfsstats.lookupcache_misses++; 894 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 895 len = ncp->nc_nlen; 896 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 897 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 898 nfsm_fhtom(dvp, v3); 899 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 900 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 901 if (error) { 902 /* 903 * Cache negatve lookups to reduce NFS traffic, but use 904 * a fast timeout. Otherwise use a timeout of 1 tick. 905 * XXX we should add a namecache flag for no-caching 906 * to uncache the negative hit as soon as possible, but 907 * we cannot simply destroy the entry because it is used 908 * as a placeholder by the caller. 909 */ 910 if (error == ENOENT) { 911 int nticks; 912 913 if (nfsneg_cache_timeout) 914 nticks = nfsneg_cache_timeout * hz; 915 else 916 nticks = 1; 917 cache_setvp(ncp, NULL); 918 cache_settimeout(ncp, nticks); 919 } 920 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 921 m_freem(mrep); 922 goto nfsmout; 923 } 924 925 /* 926 * Success, get the file handle, do various checks, and load 927 * post-operation data from the reply packet. Theoretically 928 * we should never be looking up "." so, theoretically, we 929 * should never get the same file handle as our directory. But 930 * we check anyway. XXX 931 * 932 * Note that no timeout is set for the positive cache hit. We 933 * assume, theoretically, that ESTALE returns will be dealt with 934 * properly to handle NFS races and in anycase we cannot depend 935 * on a timeout to deal with NFS open/create/excl issues so instead 936 * of a bad hack here the rest of the NFS client code needs to do 937 * the right thing. 938 */ 939 nfsm_getfh(fhp, fhsize, v3); 940 941 np = VTONFS(dvp); 942 if (NFS_CMPFH(np, fhp, fhsize)) { 943 vref(dvp); 944 nvp = dvp; 945 } else { 946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 947 if (error) { 948 m_freem(mrep); 949 vput(dvp); 950 return (error); 951 } 952 nvp = NFSTOV(np); 953 } 954 if (v3) { 955 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 956 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 957 } else { 958 nfsm_loadattr(nvp, NULL); 959 } 960 cache_setvp(ncp, nvp); 961 m_freem(mrep); 962 nfsmout: 963 vput(dvp); 964 if (nvp) { 965 if (nvp == dvp) 966 vrele(nvp); 967 else 968 vput(nvp); 969 } 970 return (error); 971 } 972 973 /* 974 * 'cached' nfs directory lookup 975 * 976 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 977 * 978 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 979 * struct vnode **a_vpp, struct componentname *a_cnp) 980 */ 981 static int 982 nfs_lookup(struct vop_old_lookup_args *ap) 983 { 984 struct componentname *cnp = ap->a_cnp; 985 struct vnode *dvp = ap->a_dvp; 986 struct vnode **vpp = ap->a_vpp; 987 int flags = cnp->cn_flags; 988 struct vnode *newvp; 989 u_int32_t *tl; 990 caddr_t cp; 991 int32_t t1, t2; 992 struct nfsmount *nmp; 993 caddr_t bpos, dpos, cp2; 994 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 995 long len; 996 nfsfh_t *fhp; 997 struct nfsnode *np; 998 int lockparent, wantparent, error = 0, attrflag, fhsize; 999 int v3 = NFS_ISV3(dvp); 1000 struct thread *td = cnp->cn_td; 1001 1002 /* 1003 * Read-only mount check and directory check. 1004 */ 1005 *vpp = NULLVP; 1006 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1007 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1008 return (EROFS); 1009 1010 if (dvp->v_type != VDIR) 1011 return (ENOTDIR); 1012 1013 /* 1014 * Look it up in the cache. Note that ENOENT is only returned if we 1015 * previously entered a negative hit (see later on). The additional 1016 * nfsneg_cache_timeout check causes previously cached results to 1017 * be instantly ignored if the negative caching is turned off. 1018 */ 1019 lockparent = flags & CNP_LOCKPARENT; 1020 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1021 nmp = VFSTONFS(dvp->v_mount); 1022 np = VTONFS(dvp); 1023 1024 /* 1025 * Go to the wire. 1026 */ 1027 error = 0; 1028 newvp = NULLVP; 1029 nfsstats.lookupcache_misses++; 1030 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1031 len = cnp->cn_namelen; 1032 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 1033 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1034 nfsm_fhtom(dvp, v3); 1035 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1036 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 1037 if (error) { 1038 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1039 m_freem(mrep); 1040 goto nfsmout; 1041 } 1042 nfsm_getfh(fhp, fhsize, v3); 1043 1044 /* 1045 * Handle RENAME case... 1046 */ 1047 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1048 if (NFS_CMPFH(np, fhp, fhsize)) { 1049 m_freem(mrep); 1050 return (EISDIR); 1051 } 1052 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1053 if (error) { 1054 m_freem(mrep); 1055 return (error); 1056 } 1057 newvp = NFSTOV(np); 1058 if (v3) { 1059 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1060 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1061 } else 1062 nfsm_loadattr(newvp, (struct vattr *)0); 1063 *vpp = newvp; 1064 m_freem(mrep); 1065 if (!lockparent) { 1066 VOP_UNLOCK(dvp, 0, td); 1067 cnp->cn_flags |= CNP_PDIRUNLOCK; 1068 } 1069 return (0); 1070 } 1071 1072 if (flags & CNP_ISDOTDOT) { 1073 VOP_UNLOCK(dvp, 0, td); 1074 cnp->cn_flags |= CNP_PDIRUNLOCK; 1075 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1076 if (error) { 1077 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 1078 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1079 return (error); /* NOTE: return error from nget */ 1080 } 1081 newvp = NFSTOV(np); 1082 if (lockparent) { 1083 error = vn_lock(dvp, LK_EXCLUSIVE, td); 1084 if (error) { 1085 vput(newvp); 1086 return (error); 1087 } 1088 cnp->cn_flags |= CNP_PDIRUNLOCK; 1089 } 1090 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1091 vref(dvp); 1092 newvp = dvp; 1093 } else { 1094 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1095 if (error) { 1096 m_freem(mrep); 1097 return (error); 1098 } 1099 if (!lockparent) { 1100 VOP_UNLOCK(dvp, 0, td); 1101 cnp->cn_flags |= CNP_PDIRUNLOCK; 1102 } 1103 newvp = NFSTOV(np); 1104 } 1105 if (v3) { 1106 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1107 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1108 } else 1109 nfsm_loadattr(newvp, (struct vattr *)0); 1110 #if 0 1111 /* XXX MOVE TO nfs_nremove() */ 1112 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1113 cnp->cn_nameiop != NAMEI_DELETE) { 1114 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1115 } 1116 #endif 1117 *vpp = newvp; 1118 m_freem(mrep); 1119 nfsmout: 1120 if (error) { 1121 if (newvp != NULLVP) { 1122 vrele(newvp); 1123 *vpp = NULLVP; 1124 } 1125 if ((cnp->cn_nameiop == NAMEI_CREATE || 1126 cnp->cn_nameiop == NAMEI_RENAME) && 1127 error == ENOENT) { 1128 if (!lockparent) { 1129 VOP_UNLOCK(dvp, 0, td); 1130 cnp->cn_flags |= CNP_PDIRUNLOCK; 1131 } 1132 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1133 error = EROFS; 1134 else 1135 error = EJUSTRETURN; 1136 } 1137 } 1138 return (error); 1139 } 1140 1141 /* 1142 * nfs read call. 1143 * Just call nfs_bioread() to do the work. 1144 * 1145 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1146 * struct ucred *a_cred) 1147 */ 1148 static int 1149 nfs_read(struct vop_read_args *ap) 1150 { 1151 struct vnode *vp = ap->a_vp; 1152 1153 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1154 switch (vp->v_type) { 1155 case VREG: 1156 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1157 case VDIR: 1158 return (EISDIR); 1159 default: 1160 return EOPNOTSUPP; 1161 } 1162 } 1163 1164 /* 1165 * nfs readlink call 1166 * 1167 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1168 */ 1169 static int 1170 nfs_readlink(struct vop_readlink_args *ap) 1171 { 1172 struct vnode *vp = ap->a_vp; 1173 1174 if (vp->v_type != VLNK) 1175 return (EINVAL); 1176 return (nfs_bioread(vp, ap->a_uio, 0)); 1177 } 1178 1179 /* 1180 * Do a readlink rpc. 1181 * Called by nfs_doio() from below the buffer cache. 1182 */ 1183 int 1184 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1185 { 1186 u_int32_t *tl; 1187 caddr_t cp; 1188 int32_t t1, t2; 1189 caddr_t bpos, dpos, cp2; 1190 int error = 0, len, attrflag; 1191 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1192 int v3 = NFS_ISV3(vp); 1193 1194 nfsstats.rpccnt[NFSPROC_READLINK]++; 1195 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1196 nfsm_fhtom(vp, v3); 1197 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1198 if (v3) 1199 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1200 if (!error) { 1201 nfsm_strsiz(len, NFS_MAXPATHLEN); 1202 if (len == NFS_MAXPATHLEN) { 1203 struct nfsnode *np = VTONFS(vp); 1204 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1205 len = np->n_size; 1206 } 1207 nfsm_mtouio(uiop, len); 1208 } 1209 m_freem(mrep); 1210 nfsmout: 1211 return (error); 1212 } 1213 1214 /* 1215 * nfs read rpc call 1216 * Ditto above 1217 */ 1218 int 1219 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1220 { 1221 u_int32_t *tl; 1222 caddr_t cp; 1223 int32_t t1, t2; 1224 caddr_t bpos, dpos, cp2; 1225 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1226 struct nfsmount *nmp; 1227 int error = 0, len, retlen, tsiz, eof, attrflag; 1228 int v3 = NFS_ISV3(vp); 1229 1230 #ifndef nolint 1231 eof = 0; 1232 #endif 1233 nmp = VFSTONFS(vp->v_mount); 1234 tsiz = uiop->uio_resid; 1235 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1236 return (EFBIG); 1237 while (tsiz > 0) { 1238 nfsstats.rpccnt[NFSPROC_READ]++; 1239 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1240 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1241 nfsm_fhtom(vp, v3); 1242 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1243 if (v3) { 1244 txdr_hyper(uiop->uio_offset, tl); 1245 *(tl + 2) = txdr_unsigned(len); 1246 } else { 1247 *tl++ = txdr_unsigned(uiop->uio_offset); 1248 *tl++ = txdr_unsigned(len); 1249 *tl = 0; 1250 } 1251 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1252 if (v3) { 1253 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1254 if (error) { 1255 m_freem(mrep); 1256 goto nfsmout; 1257 } 1258 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1259 eof = fxdr_unsigned(int, *(tl + 1)); 1260 } else 1261 nfsm_loadattr(vp, (struct vattr *)0); 1262 nfsm_strsiz(retlen, nmp->nm_rsize); 1263 nfsm_mtouio(uiop, retlen); 1264 m_freem(mrep); 1265 tsiz -= retlen; 1266 if (v3) { 1267 if (eof || retlen == 0) { 1268 tsiz = 0; 1269 } 1270 } else if (retlen < len) { 1271 tsiz = 0; 1272 } 1273 } 1274 nfsmout: 1275 return (error); 1276 } 1277 1278 /* 1279 * nfs write call 1280 */ 1281 int 1282 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1283 { 1284 u_int32_t *tl; 1285 caddr_t cp; 1286 int32_t t1, t2, backup; 1287 caddr_t bpos, dpos, cp2; 1288 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1289 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1290 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1291 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1292 1293 #ifndef DIAGNOSTIC 1294 if (uiop->uio_iovcnt != 1) 1295 panic("nfs: writerpc iovcnt > 1"); 1296 #endif 1297 *must_commit = 0; 1298 tsiz = uiop->uio_resid; 1299 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1300 return (EFBIG); 1301 while (tsiz > 0) { 1302 nfsstats.rpccnt[NFSPROC_WRITE]++; 1303 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1304 nfsm_reqhead(vp, NFSPROC_WRITE, 1305 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1306 nfsm_fhtom(vp, v3); 1307 if (v3) { 1308 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1309 txdr_hyper(uiop->uio_offset, tl); 1310 tl += 2; 1311 *tl++ = txdr_unsigned(len); 1312 *tl++ = txdr_unsigned(*iomode); 1313 *tl = txdr_unsigned(len); 1314 } else { 1315 u_int32_t x; 1316 1317 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1318 /* Set both "begin" and "current" to non-garbage. */ 1319 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1320 *tl++ = x; /* "begin offset" */ 1321 *tl++ = x; /* "current offset" */ 1322 x = txdr_unsigned(len); 1323 *tl++ = x; /* total to this offset */ 1324 *tl = x; /* size of this write */ 1325 } 1326 nfsm_uiotom(uiop, len); 1327 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1328 if (v3) { 1329 /* 1330 * The write RPC returns a before and after mtime. The 1331 * nfsm_wcc_data() macro checks the before n_mtime 1332 * against the before time and stores the after time 1333 * in the nfsnode's cached vattr and n_mtime field. 1334 * The NRMODIFIED bit will be set if the before 1335 * time did not match the original mtime. 1336 */ 1337 wccflag = NFSV3_WCCCHK; 1338 nfsm_wcc_data(vp, wccflag); 1339 if (!error) { 1340 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1341 + NFSX_V3WRITEVERF); 1342 rlen = fxdr_unsigned(int, *tl++); 1343 if (rlen == 0) { 1344 error = NFSERR_IO; 1345 m_freem(mrep); 1346 break; 1347 } else if (rlen < len) { 1348 backup = len - rlen; 1349 uiop->uio_iov->iov_base -= backup; 1350 uiop->uio_iov->iov_len += backup; 1351 uiop->uio_offset -= backup; 1352 uiop->uio_resid += backup; 1353 len = rlen; 1354 } 1355 commit = fxdr_unsigned(int, *tl++); 1356 1357 /* 1358 * Return the lowest committment level 1359 * obtained by any of the RPCs. 1360 */ 1361 if (committed == NFSV3WRITE_FILESYNC) 1362 committed = commit; 1363 else if (committed == NFSV3WRITE_DATASYNC && 1364 commit == NFSV3WRITE_UNSTABLE) 1365 committed = commit; 1366 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1367 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1368 NFSX_V3WRITEVERF); 1369 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1370 } else if (bcmp((caddr_t)tl, 1371 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1372 *must_commit = 1; 1373 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1374 NFSX_V3WRITEVERF); 1375 } 1376 } 1377 } else { 1378 nfsm_loadattr(vp, (struct vattr *)0); 1379 } 1380 m_freem(mrep); 1381 if (error) 1382 break; 1383 tsiz -= len; 1384 } 1385 nfsmout: 1386 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1387 committed = NFSV3WRITE_FILESYNC; 1388 *iomode = committed; 1389 if (error) 1390 uiop->uio_resid = tsiz; 1391 return (error); 1392 } 1393 1394 /* 1395 * nfs mknod rpc 1396 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1397 * mode set to specify the file type and the size field for rdev. 1398 */ 1399 static int 1400 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1401 struct vattr *vap) 1402 { 1403 struct nfsv2_sattr *sp; 1404 u_int32_t *tl; 1405 caddr_t cp; 1406 int32_t t1, t2; 1407 struct vnode *newvp = (struct vnode *)0; 1408 struct nfsnode *np = (struct nfsnode *)0; 1409 struct vattr vattr; 1410 char *cp2; 1411 caddr_t bpos, dpos; 1412 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1413 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1414 u_int32_t rdev; 1415 int v3 = NFS_ISV3(dvp); 1416 1417 if (vap->va_type == VCHR || vap->va_type == VBLK) 1418 rdev = txdr_unsigned(vap->va_rdev); 1419 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1420 rdev = nfs_xdrneg1; 1421 else { 1422 return (EOPNOTSUPP); 1423 } 1424 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1425 return (error); 1426 } 1427 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1428 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1429 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1430 nfsm_fhtom(dvp, v3); 1431 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1432 if (v3) { 1433 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1434 *tl++ = vtonfsv3_type(vap->va_type); 1435 nfsm_v3attrbuild(vap, FALSE); 1436 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1437 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1438 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1439 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1440 } 1441 } else { 1442 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1443 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1444 sp->sa_uid = nfs_xdrneg1; 1445 sp->sa_gid = nfs_xdrneg1; 1446 sp->sa_size = rdev; 1447 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1448 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1449 } 1450 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1451 if (!error) { 1452 nfsm_mtofh(dvp, newvp, v3, gotvp); 1453 if (!gotvp) { 1454 if (newvp) { 1455 vput(newvp); 1456 newvp = (struct vnode *)0; 1457 } 1458 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1459 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1460 if (!error) 1461 newvp = NFSTOV(np); 1462 } 1463 } 1464 if (v3) 1465 nfsm_wcc_data(dvp, wccflag); 1466 m_freem(mrep); 1467 nfsmout: 1468 if (error) { 1469 if (newvp) 1470 vput(newvp); 1471 } else { 1472 *vpp = newvp; 1473 } 1474 VTONFS(dvp)->n_flag |= NLMODIFIED; 1475 if (!wccflag) 1476 VTONFS(dvp)->n_attrstamp = 0; 1477 return (error); 1478 } 1479 1480 /* 1481 * nfs mknod vop 1482 * just call nfs_mknodrpc() to do the work. 1483 * 1484 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1485 * struct componentname *a_cnp, struct vattr *a_vap) 1486 */ 1487 /* ARGSUSED */ 1488 static int 1489 nfs_mknod(struct vop_old_mknod_args *ap) 1490 { 1491 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1492 } 1493 1494 static u_long create_verf; 1495 /* 1496 * nfs file create call 1497 * 1498 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1499 * struct componentname *a_cnp, struct vattr *a_vap) 1500 */ 1501 static int 1502 nfs_create(struct vop_old_create_args *ap) 1503 { 1504 struct vnode *dvp = ap->a_dvp; 1505 struct vattr *vap = ap->a_vap; 1506 struct componentname *cnp = ap->a_cnp; 1507 struct nfsv2_sattr *sp; 1508 u_int32_t *tl; 1509 caddr_t cp; 1510 int32_t t1, t2; 1511 struct nfsnode *np = (struct nfsnode *)0; 1512 struct vnode *newvp = (struct vnode *)0; 1513 caddr_t bpos, dpos, cp2; 1514 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1515 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1516 struct vattr vattr; 1517 int v3 = NFS_ISV3(dvp); 1518 1519 /* 1520 * Oops, not for me.. 1521 */ 1522 if (vap->va_type == VSOCK) 1523 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1524 1525 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1526 return (error); 1527 } 1528 if (vap->va_vaflags & VA_EXCLUSIVE) 1529 fmode |= O_EXCL; 1530 again: 1531 nfsstats.rpccnt[NFSPROC_CREATE]++; 1532 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1533 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1534 nfsm_fhtom(dvp, v3); 1535 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1536 if (v3) { 1537 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1538 if (fmode & O_EXCL) { 1539 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1540 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1541 #ifdef INET 1542 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1543 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1544 else 1545 #endif 1546 *tl++ = create_verf; 1547 *tl = ++create_verf; 1548 } else { 1549 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1550 nfsm_v3attrbuild(vap, FALSE); 1551 } 1552 } else { 1553 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1554 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1555 sp->sa_uid = nfs_xdrneg1; 1556 sp->sa_gid = nfs_xdrneg1; 1557 sp->sa_size = 0; 1558 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1559 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1560 } 1561 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1562 if (!error) { 1563 nfsm_mtofh(dvp, newvp, v3, gotvp); 1564 if (!gotvp) { 1565 if (newvp) { 1566 vput(newvp); 1567 newvp = (struct vnode *)0; 1568 } 1569 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1570 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1571 if (!error) 1572 newvp = NFSTOV(np); 1573 } 1574 } 1575 if (v3) 1576 nfsm_wcc_data(dvp, wccflag); 1577 m_freem(mrep); 1578 nfsmout: 1579 if (error) { 1580 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1581 fmode &= ~O_EXCL; 1582 goto again; 1583 } 1584 if (newvp) 1585 vput(newvp); 1586 } else if (v3 && (fmode & O_EXCL)) { 1587 /* 1588 * We are normally called with only a partially initialized 1589 * VAP. Since the NFSv3 spec says that server may use the 1590 * file attributes to store the verifier, the spec requires 1591 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1592 * in atime, but we can't really assume that all servers will 1593 * so we ensure that our SETATTR sets both atime and mtime. 1594 */ 1595 if (vap->va_mtime.tv_sec == VNOVAL) 1596 vfs_timestamp(&vap->va_mtime); 1597 if (vap->va_atime.tv_sec == VNOVAL) 1598 vap->va_atime = vap->va_mtime; 1599 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1600 } 1601 if (!error) { 1602 /* 1603 * The new np may have enough info for access 1604 * checks, make sure rucred and wucred are 1605 * initialized for read and write rpc's. 1606 */ 1607 np = VTONFS(newvp); 1608 if (np->n_rucred == NULL) 1609 np->n_rucred = crhold(cnp->cn_cred); 1610 if (np->n_wucred == NULL) 1611 np->n_wucred = crhold(cnp->cn_cred); 1612 *ap->a_vpp = newvp; 1613 } 1614 VTONFS(dvp)->n_flag |= NLMODIFIED; 1615 if (!wccflag) 1616 VTONFS(dvp)->n_attrstamp = 0; 1617 return (error); 1618 } 1619 1620 /* 1621 * nfs file remove call 1622 * To try and make nfs semantics closer to ufs semantics, a file that has 1623 * other processes using the vnode is renamed instead of removed and then 1624 * removed later on the last close. 1625 * - If v_usecount > 1 1626 * If a rename is not already in the works 1627 * call nfs_sillyrename() to set it up 1628 * else 1629 * do the remove rpc 1630 * 1631 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 1632 * struct vnode *a_vp, struct componentname *a_cnp) 1633 */ 1634 static int 1635 nfs_remove(struct vop_old_remove_args *ap) 1636 { 1637 struct vnode *vp = ap->a_vp; 1638 struct vnode *dvp = ap->a_dvp; 1639 struct componentname *cnp = ap->a_cnp; 1640 struct nfsnode *np = VTONFS(vp); 1641 int error = 0; 1642 struct vattr vattr; 1643 1644 #ifndef DIAGNOSTIC 1645 if (vp->v_usecount < 1) 1646 panic("nfs_remove: bad v_usecount"); 1647 #endif 1648 if (vp->v_type == VDIR) 1649 error = EPERM; 1650 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1651 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 && 1652 vattr.va_nlink > 1)) { 1653 /* 1654 * throw away biocache buffers, mainly to avoid 1655 * unnecessary delayed writes later. 1656 */ 1657 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1); 1658 /* Do the rpc */ 1659 if (error != EINTR) 1660 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1661 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1662 /* 1663 * Kludge City: If the first reply to the remove rpc is lost.. 1664 * the reply to the retransmitted request will be ENOENT 1665 * since the file was in fact removed 1666 * Therefore, we cheat and return success. 1667 */ 1668 if (error == ENOENT) 1669 error = 0; 1670 } else if (!np->n_sillyrename) { 1671 error = nfs_sillyrename(dvp, vp, cnp); 1672 } 1673 np->n_attrstamp = 0; 1674 return (error); 1675 } 1676 1677 /* 1678 * nfs file remove rpc called from nfs_inactive 1679 */ 1680 int 1681 nfs_removeit(struct sillyrename *sp) 1682 { 1683 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1684 sp->s_cred, NULL)); 1685 } 1686 1687 /* 1688 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1689 */ 1690 static int 1691 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1692 struct ucred *cred, struct thread *td) 1693 { 1694 u_int32_t *tl; 1695 caddr_t cp; 1696 int32_t t1, t2; 1697 caddr_t bpos, dpos, cp2; 1698 int error = 0, wccflag = NFSV3_WCCRATTR; 1699 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1700 int v3 = NFS_ISV3(dvp); 1701 1702 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1703 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1704 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1705 nfsm_fhtom(dvp, v3); 1706 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1707 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1708 if (v3) 1709 nfsm_wcc_data(dvp, wccflag); 1710 m_freem(mrep); 1711 nfsmout: 1712 VTONFS(dvp)->n_flag |= NLMODIFIED; 1713 if (!wccflag) 1714 VTONFS(dvp)->n_attrstamp = 0; 1715 return (error); 1716 } 1717 1718 /* 1719 * nfs file rename call 1720 * 1721 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1722 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1723 * struct vnode *a_tvp, struct componentname *a_tcnp) 1724 */ 1725 static int 1726 nfs_rename(struct vop_old_rename_args *ap) 1727 { 1728 struct vnode *fvp = ap->a_fvp; 1729 struct vnode *tvp = ap->a_tvp; 1730 struct vnode *fdvp = ap->a_fdvp; 1731 struct vnode *tdvp = ap->a_tdvp; 1732 struct componentname *tcnp = ap->a_tcnp; 1733 struct componentname *fcnp = ap->a_fcnp; 1734 int error; 1735 1736 /* Check for cross-device rename */ 1737 if ((fvp->v_mount != tdvp->v_mount) || 1738 (tvp && (fvp->v_mount != tvp->v_mount))) { 1739 error = EXDEV; 1740 goto out; 1741 } 1742 1743 /* 1744 * We have to flush B_DELWRI data prior to renaming 1745 * the file. If we don't, the delayed-write buffers 1746 * can be flushed out later after the file has gone stale 1747 * under NFSV3. NFSV2 does not have this problem because 1748 * ( as far as I can tell ) it flushes dirty buffers more 1749 * often. 1750 */ 1751 1752 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td); 1753 if (tvp) 1754 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td); 1755 1756 /* 1757 * If the tvp exists and is in use, sillyrename it before doing the 1758 * rename of the new file over it. 1759 * 1760 * XXX Can't sillyrename a directory. 1761 * 1762 * We do not attempt to do any namecache purges in this old API 1763 * routine. The new API compat functions have access to the actual 1764 * namecache structures and will do it for us. 1765 */ 1766 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1767 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1768 vput(tvp); 1769 tvp = NULL; 1770 } else if (tvp) { 1771 ; 1772 } 1773 1774 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1775 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1776 tcnp->cn_td); 1777 1778 out: 1779 if (tdvp == tvp) 1780 vrele(tdvp); 1781 else 1782 vput(tdvp); 1783 if (tvp) 1784 vput(tvp); 1785 vrele(fdvp); 1786 vrele(fvp); 1787 /* 1788 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1789 */ 1790 if (error == ENOENT) 1791 error = 0; 1792 return (error); 1793 } 1794 1795 /* 1796 * nfs file rename rpc called from nfs_remove() above 1797 */ 1798 static int 1799 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1800 struct sillyrename *sp) 1801 { 1802 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1803 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1804 } 1805 1806 /* 1807 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1808 */ 1809 static int 1810 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1811 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1812 struct ucred *cred, struct thread *td) 1813 { 1814 u_int32_t *tl; 1815 caddr_t cp; 1816 int32_t t1, t2; 1817 caddr_t bpos, dpos, cp2; 1818 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1819 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1820 int v3 = NFS_ISV3(fdvp); 1821 1822 nfsstats.rpccnt[NFSPROC_RENAME]++; 1823 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1824 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1825 nfsm_rndup(tnamelen)); 1826 nfsm_fhtom(fdvp, v3); 1827 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1828 nfsm_fhtom(tdvp, v3); 1829 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1830 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1831 if (v3) { 1832 nfsm_wcc_data(fdvp, fwccflag); 1833 nfsm_wcc_data(tdvp, twccflag); 1834 } 1835 m_freem(mrep); 1836 nfsmout: 1837 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1838 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1839 if (!fwccflag) 1840 VTONFS(fdvp)->n_attrstamp = 0; 1841 if (!twccflag) 1842 VTONFS(tdvp)->n_attrstamp = 0; 1843 return (error); 1844 } 1845 1846 /* 1847 * nfs hard link create call 1848 * 1849 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1850 * struct componentname *a_cnp) 1851 */ 1852 static int 1853 nfs_link(struct vop_old_link_args *ap) 1854 { 1855 struct vnode *vp = ap->a_vp; 1856 struct vnode *tdvp = ap->a_tdvp; 1857 struct componentname *cnp = ap->a_cnp; 1858 u_int32_t *tl; 1859 caddr_t cp; 1860 int32_t t1, t2; 1861 caddr_t bpos, dpos, cp2; 1862 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1863 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1864 int v3; 1865 1866 if (vp->v_mount != tdvp->v_mount) { 1867 return (EXDEV); 1868 } 1869 1870 /* 1871 * Push all writes to the server, so that the attribute cache 1872 * doesn't get "out of sync" with the server. 1873 * XXX There should be a better way! 1874 */ 1875 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td); 1876 1877 v3 = NFS_ISV3(vp); 1878 nfsstats.rpccnt[NFSPROC_LINK]++; 1879 nfsm_reqhead(vp, NFSPROC_LINK, 1880 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1881 nfsm_fhtom(vp, v3); 1882 nfsm_fhtom(tdvp, v3); 1883 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1884 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1885 if (v3) { 1886 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1887 nfsm_wcc_data(tdvp, wccflag); 1888 } 1889 m_freem(mrep); 1890 nfsmout: 1891 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1892 if (!attrflag) 1893 VTONFS(vp)->n_attrstamp = 0; 1894 if (!wccflag) 1895 VTONFS(tdvp)->n_attrstamp = 0; 1896 /* 1897 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1898 */ 1899 if (error == EEXIST) 1900 error = 0; 1901 return (error); 1902 } 1903 1904 /* 1905 * nfs symbolic link create call 1906 * 1907 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1908 * struct componentname *a_cnp, struct vattr *a_vap, 1909 * char *a_target) 1910 */ 1911 static int 1912 nfs_symlink(struct vop_old_symlink_args *ap) 1913 { 1914 struct vnode *dvp = ap->a_dvp; 1915 struct vattr *vap = ap->a_vap; 1916 struct componentname *cnp = ap->a_cnp; 1917 struct nfsv2_sattr *sp; 1918 u_int32_t *tl; 1919 caddr_t cp; 1920 int32_t t1, t2; 1921 caddr_t bpos, dpos, cp2; 1922 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1923 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1924 struct vnode *newvp = (struct vnode *)0; 1925 int v3 = NFS_ISV3(dvp); 1926 1927 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1928 slen = strlen(ap->a_target); 1929 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1930 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1931 nfsm_fhtom(dvp, v3); 1932 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1933 if (v3) { 1934 nfsm_v3attrbuild(vap, FALSE); 1935 } 1936 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1937 if (!v3) { 1938 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1939 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1940 sp->sa_uid = nfs_xdrneg1; 1941 sp->sa_gid = nfs_xdrneg1; 1942 sp->sa_size = nfs_xdrneg1; 1943 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1944 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1945 } 1946 1947 /* 1948 * Issue the NFS request and get the rpc response. 1949 * 1950 * Only NFSv3 responses returning an error of 0 actually return 1951 * a file handle that can be converted into newvp without having 1952 * to do an extra lookup rpc. 1953 */ 1954 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1955 if (v3) { 1956 if (error == 0) 1957 nfsm_mtofh(dvp, newvp, v3, gotvp); 1958 nfsm_wcc_data(dvp, wccflag); 1959 } 1960 1961 /* 1962 * out code jumps -> here, mrep is also freed. 1963 */ 1964 1965 m_freem(mrep); 1966 nfsmout: 1967 1968 /* 1969 * If we get an EEXIST error, silently convert it to no-error 1970 * in case of an NFS retry. 1971 */ 1972 if (error == EEXIST) 1973 error = 0; 1974 1975 /* 1976 * If we do not have (or no longer have) an error, and we could 1977 * not extract the newvp from the response due to the request being 1978 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1979 * to obtain a newvp to return. 1980 */ 1981 if (error == 0 && newvp == NULL) { 1982 struct nfsnode *np = NULL; 1983 1984 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1985 cnp->cn_cred, cnp->cn_td, &np); 1986 if (!error) 1987 newvp = NFSTOV(np); 1988 } 1989 if (error) { 1990 if (newvp) 1991 vput(newvp); 1992 } else { 1993 *ap->a_vpp = newvp; 1994 } 1995 VTONFS(dvp)->n_flag |= NLMODIFIED; 1996 if (!wccflag) 1997 VTONFS(dvp)->n_attrstamp = 0; 1998 return (error); 1999 } 2000 2001 /* 2002 * nfs make dir call 2003 * 2004 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2005 * struct componentname *a_cnp, struct vattr *a_vap) 2006 */ 2007 static int 2008 nfs_mkdir(struct vop_old_mkdir_args *ap) 2009 { 2010 struct vnode *dvp = ap->a_dvp; 2011 struct vattr *vap = ap->a_vap; 2012 struct componentname *cnp = ap->a_cnp; 2013 struct nfsv2_sattr *sp; 2014 u_int32_t *tl; 2015 caddr_t cp; 2016 int32_t t1, t2; 2017 int len; 2018 struct nfsnode *np = (struct nfsnode *)0; 2019 struct vnode *newvp = (struct vnode *)0; 2020 caddr_t bpos, dpos, cp2; 2021 int error = 0, wccflag = NFSV3_WCCRATTR; 2022 int gotvp = 0; 2023 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2024 struct vattr vattr; 2025 int v3 = NFS_ISV3(dvp); 2026 2027 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 2028 return (error); 2029 } 2030 len = cnp->cn_namelen; 2031 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2032 nfsm_reqhead(dvp, NFSPROC_MKDIR, 2033 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2034 nfsm_fhtom(dvp, v3); 2035 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2036 if (v3) { 2037 nfsm_v3attrbuild(vap, FALSE); 2038 } else { 2039 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2040 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2041 sp->sa_uid = nfs_xdrneg1; 2042 sp->sa_gid = nfs_xdrneg1; 2043 sp->sa_size = nfs_xdrneg1; 2044 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2045 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2046 } 2047 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2048 if (!error) 2049 nfsm_mtofh(dvp, newvp, v3, gotvp); 2050 if (v3) 2051 nfsm_wcc_data(dvp, wccflag); 2052 m_freem(mrep); 2053 nfsmout: 2054 VTONFS(dvp)->n_flag |= NLMODIFIED; 2055 if (!wccflag) 2056 VTONFS(dvp)->n_attrstamp = 0; 2057 /* 2058 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2059 * if we can succeed in looking up the directory. 2060 */ 2061 if (error == EEXIST || (!error && !gotvp)) { 2062 if (newvp) { 2063 vrele(newvp); 2064 newvp = (struct vnode *)0; 2065 } 2066 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2067 cnp->cn_td, &np); 2068 if (!error) { 2069 newvp = NFSTOV(np); 2070 if (newvp->v_type != VDIR) 2071 error = EEXIST; 2072 } 2073 } 2074 if (error) { 2075 if (newvp) 2076 vrele(newvp); 2077 } else 2078 *ap->a_vpp = newvp; 2079 return (error); 2080 } 2081 2082 /* 2083 * nfs remove directory call 2084 * 2085 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2086 * struct componentname *a_cnp) 2087 */ 2088 static int 2089 nfs_rmdir(struct vop_old_rmdir_args *ap) 2090 { 2091 struct vnode *vp = ap->a_vp; 2092 struct vnode *dvp = ap->a_dvp; 2093 struct componentname *cnp = ap->a_cnp; 2094 u_int32_t *tl; 2095 caddr_t cp; 2096 int32_t t1, t2; 2097 caddr_t bpos, dpos, cp2; 2098 int error = 0, wccflag = NFSV3_WCCRATTR; 2099 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2100 int v3 = NFS_ISV3(dvp); 2101 2102 if (dvp == vp) 2103 return (EINVAL); 2104 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2105 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2106 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2107 nfsm_fhtom(dvp, v3); 2108 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2109 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2110 if (v3) 2111 nfsm_wcc_data(dvp, wccflag); 2112 m_freem(mrep); 2113 nfsmout: 2114 VTONFS(dvp)->n_flag |= NLMODIFIED; 2115 if (!wccflag) 2116 VTONFS(dvp)->n_attrstamp = 0; 2117 /* 2118 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2119 */ 2120 if (error == ENOENT) 2121 error = 0; 2122 return (error); 2123 } 2124 2125 /* 2126 * nfs readdir call 2127 * 2128 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2129 */ 2130 static int 2131 nfs_readdir(struct vop_readdir_args *ap) 2132 { 2133 struct vnode *vp = ap->a_vp; 2134 struct nfsnode *np = VTONFS(vp); 2135 struct uio *uio = ap->a_uio; 2136 int tresid, error; 2137 struct vattr vattr; 2138 2139 if (vp->v_type != VDIR) 2140 return (EPERM); 2141 2142 /* 2143 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2144 * and then check that is still valid, or if this is an NQNFS mount 2145 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2146 * VOP_GETATTR() does not necessarily go to the wire. 2147 */ 2148 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2149 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2150 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) { 2151 if (NQNFS_CKCACHABLE(vp, ND_READ)) { 2152 nfsstats.direofcache_hits++; 2153 return (0); 2154 } 2155 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 && 2156 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2157 ) { 2158 nfsstats.direofcache_hits++; 2159 return (0); 2160 } 2161 } 2162 2163 /* 2164 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2165 * own cache coherency checks so we do not have to. 2166 */ 2167 tresid = uio->uio_resid; 2168 error = nfs_bioread(vp, uio, 0); 2169 2170 if (!error && uio->uio_resid == tresid) 2171 nfsstats.direofcache_misses++; 2172 return (error); 2173 } 2174 2175 /* 2176 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2177 * 2178 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2179 * offset/block and converts the nfs formatted directory entries for userland 2180 * consumption as well as deals with offsets into the middle of blocks. 2181 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2182 * be block-bounded. It must convert to cookies for the actual RPC. 2183 */ 2184 int 2185 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2186 { 2187 int len, left; 2188 struct nfs_dirent *dp = NULL; 2189 u_int32_t *tl; 2190 caddr_t cp; 2191 int32_t t1, t2; 2192 nfsuint64 *cookiep; 2193 caddr_t bpos, dpos, cp2; 2194 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2195 nfsuint64 cookie; 2196 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2197 struct nfsnode *dnp = VTONFS(vp); 2198 u_quad_t fileno; 2199 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2200 int attrflag; 2201 int v3 = NFS_ISV3(vp); 2202 2203 #ifndef DIAGNOSTIC 2204 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2205 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2206 panic("nfs readdirrpc bad uio"); 2207 #endif 2208 2209 /* 2210 * If there is no cookie, assume directory was stale. 2211 */ 2212 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2213 if (cookiep) 2214 cookie = *cookiep; 2215 else 2216 return (NFSERR_BAD_COOKIE); 2217 /* 2218 * Loop around doing readdir rpc's of size nm_readdirsize 2219 * truncated to a multiple of DIRBLKSIZ. 2220 * The stopping criteria is EOF or buffer full. 2221 */ 2222 while (more_dirs && bigenough) { 2223 nfsstats.rpccnt[NFSPROC_READDIR]++; 2224 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2225 NFSX_READDIR(v3)); 2226 nfsm_fhtom(vp, v3); 2227 if (v3) { 2228 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2229 *tl++ = cookie.nfsuquad[0]; 2230 *tl++ = cookie.nfsuquad[1]; 2231 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2232 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2233 } else { 2234 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2235 *tl++ = cookie.nfsuquad[0]; 2236 } 2237 *tl = txdr_unsigned(nmp->nm_readdirsize); 2238 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2239 if (v3) { 2240 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2241 if (!error) { 2242 nfsm_dissect(tl, u_int32_t *, 2243 2 * NFSX_UNSIGNED); 2244 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2245 dnp->n_cookieverf.nfsuquad[1] = *tl; 2246 } else { 2247 m_freem(mrep); 2248 goto nfsmout; 2249 } 2250 } 2251 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2252 more_dirs = fxdr_unsigned(int, *tl); 2253 2254 /* loop thru the dir entries, converting them to std form */ 2255 while (more_dirs && bigenough) { 2256 if (v3) { 2257 nfsm_dissect(tl, u_int32_t *, 2258 3 * NFSX_UNSIGNED); 2259 fileno = fxdr_hyper(tl); 2260 len = fxdr_unsigned(int, *(tl + 2)); 2261 } else { 2262 nfsm_dissect(tl, u_int32_t *, 2263 2 * NFSX_UNSIGNED); 2264 fileno = fxdr_unsigned(u_quad_t, *tl++); 2265 len = fxdr_unsigned(int, *tl); 2266 } 2267 if (len <= 0 || len > NFS_MAXNAMLEN) { 2268 error = EBADRPC; 2269 m_freem(mrep); 2270 goto nfsmout; 2271 } 2272 2273 /* 2274 * len is the number of bytes in the path element 2275 * name, not including the \0 termination. 2276 * 2277 * tlen is the number of bytes w have to reserve for 2278 * the path element name. 2279 */ 2280 tlen = nfsm_rndup(len); 2281 if (tlen == len) 2282 tlen += 4; /* To ensure null termination */ 2283 2284 /* 2285 * If the entry would cross a DIRBLKSIZ boundary, 2286 * extend the previous nfs_dirent to cover the 2287 * remaining space. 2288 */ 2289 left = DIRBLKSIZ - blksiz; 2290 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2291 dp->nfs_reclen += left; 2292 uiop->uio_iov->iov_base += left; 2293 uiop->uio_iov->iov_len -= left; 2294 uiop->uio_offset += left; 2295 uiop->uio_resid -= left; 2296 blksiz = 0; 2297 } 2298 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2299 bigenough = 0; 2300 if (bigenough) { 2301 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2302 dp->nfs_ino = fileno; 2303 dp->nfs_namlen = len; 2304 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2305 dp->nfs_type = DT_UNKNOWN; 2306 blksiz += dp->nfs_reclen; 2307 if (blksiz == DIRBLKSIZ) 2308 blksiz = 0; 2309 uiop->uio_offset += sizeof(struct nfs_dirent); 2310 uiop->uio_resid -= sizeof(struct nfs_dirent); 2311 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2312 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2313 nfsm_mtouio(uiop, len); 2314 2315 /* 2316 * The uiop has advanced by nfs_dirent + len 2317 * but really needs to advance by 2318 * nfs_dirent + tlen 2319 */ 2320 cp = uiop->uio_iov->iov_base; 2321 tlen -= len; 2322 *cp = '\0'; /* null terminate */ 2323 uiop->uio_iov->iov_base += tlen; 2324 uiop->uio_iov->iov_len -= tlen; 2325 uiop->uio_offset += tlen; 2326 uiop->uio_resid -= tlen; 2327 } else { 2328 /* 2329 * NFS strings must be rounded up (nfsm_myouio 2330 * handled that in the bigenough case). 2331 */ 2332 nfsm_adv(nfsm_rndup(len)); 2333 } 2334 if (v3) { 2335 nfsm_dissect(tl, u_int32_t *, 2336 3 * NFSX_UNSIGNED); 2337 } else { 2338 nfsm_dissect(tl, u_int32_t *, 2339 2 * NFSX_UNSIGNED); 2340 } 2341 2342 /* 2343 * If we were able to accomodate the last entry, 2344 * get the cookie for the next one. Otherwise 2345 * hold-over the cookie for the one we were not 2346 * able to accomodate. 2347 */ 2348 if (bigenough) { 2349 cookie.nfsuquad[0] = *tl++; 2350 if (v3) 2351 cookie.nfsuquad[1] = *tl++; 2352 } else if (v3) { 2353 tl += 2; 2354 } else { 2355 tl++; 2356 } 2357 more_dirs = fxdr_unsigned(int, *tl); 2358 } 2359 /* 2360 * If at end of rpc data, get the eof boolean 2361 */ 2362 if (!more_dirs) { 2363 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2364 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2365 } 2366 m_freem(mrep); 2367 } 2368 /* 2369 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2370 * by increasing d_reclen for the last record. 2371 */ 2372 if (blksiz > 0) { 2373 left = DIRBLKSIZ - blksiz; 2374 dp->nfs_reclen += left; 2375 uiop->uio_iov->iov_base += left; 2376 uiop->uio_iov->iov_len -= left; 2377 uiop->uio_offset += left; 2378 uiop->uio_resid -= left; 2379 } 2380 2381 if (bigenough) { 2382 /* 2383 * We hit the end of the directory, update direofoffset. 2384 */ 2385 dnp->n_direofoffset = uiop->uio_offset; 2386 } else { 2387 /* 2388 * There is more to go, insert the link cookie so the 2389 * next block can be read. 2390 */ 2391 if (uiop->uio_resid > 0) 2392 printf("EEK! readdirrpc resid > 0\n"); 2393 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2394 *cookiep = cookie; 2395 } 2396 nfsmout: 2397 return (error); 2398 } 2399 2400 /* 2401 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2402 */ 2403 int 2404 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2405 { 2406 int len, left; 2407 struct nfs_dirent *dp; 2408 u_int32_t *tl; 2409 caddr_t cp; 2410 int32_t t1, t2; 2411 struct vnode *newvp; 2412 nfsuint64 *cookiep; 2413 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2414 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2415 nfsuint64 cookie; 2416 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2417 struct nfsnode *dnp = VTONFS(vp), *np; 2418 nfsfh_t *fhp; 2419 u_quad_t fileno; 2420 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2421 int attrflag, fhsize; 2422 struct namecache *ncp; 2423 struct namecache *dncp; 2424 struct nlcomponent nlc; 2425 2426 #ifndef nolint 2427 dp = NULL; 2428 #endif 2429 #ifndef DIAGNOSTIC 2430 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2431 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2432 panic("nfs readdirplusrpc bad uio"); 2433 #endif 2434 /* 2435 * Obtain the namecache record for the directory so we have something 2436 * to use as a basis for creating the entries. This function will 2437 * return a held (but not locked) ncp. The ncp may be disconnected 2438 * from the tree and cannot be used for upward traversals, and the 2439 * ncp may be unnamed. Note that other unrelated operations may 2440 * cause the ncp to be named at any time. 2441 */ 2442 dncp = cache_fromdvp(vp, NULL, 0); 2443 bzero(&nlc, sizeof(nlc)); 2444 newvp = NULLVP; 2445 2446 /* 2447 * If there is no cookie, assume directory was stale. 2448 */ 2449 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2450 if (cookiep) 2451 cookie = *cookiep; 2452 else 2453 return (NFSERR_BAD_COOKIE); 2454 /* 2455 * Loop around doing readdir rpc's of size nm_readdirsize 2456 * truncated to a multiple of DIRBLKSIZ. 2457 * The stopping criteria is EOF or buffer full. 2458 */ 2459 while (more_dirs && bigenough) { 2460 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2461 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2462 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2463 nfsm_fhtom(vp, 1); 2464 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2465 *tl++ = cookie.nfsuquad[0]; 2466 *tl++ = cookie.nfsuquad[1]; 2467 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2468 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2469 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2470 *tl = txdr_unsigned(nmp->nm_rsize); 2471 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2472 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2473 if (error) { 2474 m_freem(mrep); 2475 goto nfsmout; 2476 } 2477 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2478 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2479 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2480 more_dirs = fxdr_unsigned(int, *tl); 2481 2482 /* loop thru the dir entries, doctoring them to 4bsd form */ 2483 while (more_dirs && bigenough) { 2484 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2485 fileno = fxdr_hyper(tl); 2486 len = fxdr_unsigned(int, *(tl + 2)); 2487 if (len <= 0 || len > NFS_MAXNAMLEN) { 2488 error = EBADRPC; 2489 m_freem(mrep); 2490 goto nfsmout; 2491 } 2492 tlen = nfsm_rndup(len); 2493 if (tlen == len) 2494 tlen += 4; /* To ensure null termination*/ 2495 left = DIRBLKSIZ - blksiz; 2496 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2497 dp->nfs_reclen += left; 2498 uiop->uio_iov->iov_base += left; 2499 uiop->uio_iov->iov_len -= left; 2500 uiop->uio_offset += left; 2501 uiop->uio_resid -= left; 2502 blksiz = 0; 2503 } 2504 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2505 bigenough = 0; 2506 if (bigenough) { 2507 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2508 dp->nfs_ino = fileno; 2509 dp->nfs_namlen = len; 2510 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2511 dp->nfs_type = DT_UNKNOWN; 2512 blksiz += dp->nfs_reclen; 2513 if (blksiz == DIRBLKSIZ) 2514 blksiz = 0; 2515 uiop->uio_offset += sizeof(struct nfs_dirent); 2516 uiop->uio_resid -= sizeof(struct nfs_dirent); 2517 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2518 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2519 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2520 nlc.nlc_namelen = len; 2521 nfsm_mtouio(uiop, len); 2522 cp = uiop->uio_iov->iov_base; 2523 tlen -= len; 2524 *cp = '\0'; 2525 uiop->uio_iov->iov_base += tlen; 2526 uiop->uio_iov->iov_len -= tlen; 2527 uiop->uio_offset += tlen; 2528 uiop->uio_resid -= tlen; 2529 } else 2530 nfsm_adv(nfsm_rndup(len)); 2531 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2532 if (bigenough) { 2533 cookie.nfsuquad[0] = *tl++; 2534 cookie.nfsuquad[1] = *tl++; 2535 } else 2536 tl += 2; 2537 2538 /* 2539 * Since the attributes are before the file handle 2540 * (sigh), we must skip over the attributes and then 2541 * come back and get them. 2542 */ 2543 attrflag = fxdr_unsigned(int, *tl); 2544 if (attrflag) { 2545 dpossav1 = dpos; 2546 mdsav1 = md; 2547 nfsm_adv(NFSX_V3FATTR); 2548 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2549 doit = fxdr_unsigned(int, *tl); 2550 if (doit) { 2551 nfsm_getfh(fhp, fhsize, 1); 2552 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2553 vref(vp); 2554 newvp = vp; 2555 np = dnp; 2556 } else { 2557 error = nfs_nget(vp->v_mount, fhp, 2558 fhsize, &np); 2559 if (error) 2560 doit = 0; 2561 else 2562 newvp = NFSTOV(np); 2563 } 2564 } 2565 if (doit && bigenough) { 2566 dpossav2 = dpos; 2567 dpos = dpossav1; 2568 mdsav2 = md; 2569 md = mdsav1; 2570 nfsm_loadattr(newvp, (struct vattr *)0); 2571 dpos = dpossav2; 2572 md = mdsav2; 2573 dp->nfs_type = 2574 IFTODT(VTTOIF(np->n_vattr.va_type)); 2575 if (dncp) { 2576 printf("NFS/READDIRPLUS, ENTER %*.*s\n", 2577 nlc.nlc_namelen, nlc.nlc_namelen, 2578 nlc.nlc_nameptr); 2579 ncp = cache_nlookup(dncp, &nlc); 2580 cache_setunresolved(ncp); 2581 cache_setvp(ncp, newvp); 2582 cache_put(ncp); 2583 } else { 2584 printf("NFS/READDIRPLUS, UNABLE TO ENTER" 2585 " %*.*s\n", 2586 nlc.nlc_namelen, nlc.nlc_namelen, 2587 nlc.nlc_nameptr); 2588 } 2589 } 2590 } else { 2591 /* Just skip over the file handle */ 2592 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2593 i = fxdr_unsigned(int, *tl); 2594 nfsm_adv(nfsm_rndup(i)); 2595 } 2596 if (newvp != NULLVP) { 2597 if (newvp == vp) 2598 vrele(newvp); 2599 else 2600 vput(newvp); 2601 newvp = NULLVP; 2602 } 2603 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2604 more_dirs = fxdr_unsigned(int, *tl); 2605 } 2606 /* 2607 * If at end of rpc data, get the eof boolean 2608 */ 2609 if (!more_dirs) { 2610 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2611 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2612 } 2613 m_freem(mrep); 2614 } 2615 /* 2616 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2617 * by increasing d_reclen for the last record. 2618 */ 2619 if (blksiz > 0) { 2620 left = DIRBLKSIZ - blksiz; 2621 dp->nfs_reclen += left; 2622 uiop->uio_iov->iov_base += left; 2623 uiop->uio_iov->iov_len -= left; 2624 uiop->uio_offset += left; 2625 uiop->uio_resid -= left; 2626 } 2627 2628 /* 2629 * We are now either at the end of the directory or have filled the 2630 * block. 2631 */ 2632 if (bigenough) 2633 dnp->n_direofoffset = uiop->uio_offset; 2634 else { 2635 if (uiop->uio_resid > 0) 2636 printf("EEK! readdirplusrpc resid > 0\n"); 2637 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2638 *cookiep = cookie; 2639 } 2640 nfsmout: 2641 if (newvp != NULLVP) { 2642 if (newvp == vp) 2643 vrele(newvp); 2644 else 2645 vput(newvp); 2646 newvp = NULLVP; 2647 } 2648 if (dncp) 2649 cache_drop(dncp); 2650 return (error); 2651 } 2652 2653 /* 2654 * Silly rename. To make the NFS filesystem that is stateless look a little 2655 * more like the "ufs" a remove of an active vnode is translated to a rename 2656 * to a funny looking filename that is removed by nfs_inactive on the 2657 * nfsnode. There is the potential for another process on a different client 2658 * to create the same funny name between the nfs_lookitup() fails and the 2659 * nfs_rename() completes, but... 2660 */ 2661 static int 2662 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2663 { 2664 struct sillyrename *sp; 2665 struct nfsnode *np; 2666 int error; 2667 2668 /* 2669 * We previously purged dvp instead of vp. I don't know why, it 2670 * completely destroys performance. We can't do it anyway with the 2671 * new VFS API since we would be breaking the namecache topology. 2672 */ 2673 cache_purge(vp); /* XXX */ 2674 np = VTONFS(vp); 2675 #ifndef DIAGNOSTIC 2676 if (vp->v_type == VDIR) 2677 panic("nfs: sillyrename dir"); 2678 #endif 2679 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2680 M_NFSREQ, M_WAITOK); 2681 sp->s_cred = crdup(cnp->cn_cred); 2682 sp->s_dvp = dvp; 2683 vref(dvp); 2684 2685 /* Fudge together a funny name */ 2686 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2687 2688 /* Try lookitups until we get one that isn't there */ 2689 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2690 cnp->cn_td, (struct nfsnode **)0) == 0) { 2691 sp->s_name[4]++; 2692 if (sp->s_name[4] > 'z') { 2693 error = EINVAL; 2694 goto bad; 2695 } 2696 } 2697 error = nfs_renameit(dvp, cnp, sp); 2698 if (error) 2699 goto bad; 2700 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2701 cnp->cn_td, &np); 2702 np->n_sillyrename = sp; 2703 return (0); 2704 bad: 2705 vrele(sp->s_dvp); 2706 crfree(sp->s_cred); 2707 free((caddr_t)sp, M_NFSREQ); 2708 return (error); 2709 } 2710 2711 /* 2712 * Look up a file name and optionally either update the file handle or 2713 * allocate an nfsnode, depending on the value of npp. 2714 * npp == NULL --> just do the lookup 2715 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2716 * handled too 2717 * *npp != NULL --> update the file handle in the vnode 2718 */ 2719 static int 2720 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2721 struct thread *td, struct nfsnode **npp) 2722 { 2723 u_int32_t *tl; 2724 caddr_t cp; 2725 int32_t t1, t2; 2726 struct vnode *newvp = (struct vnode *)0; 2727 struct nfsnode *np, *dnp = VTONFS(dvp); 2728 caddr_t bpos, dpos, cp2; 2729 int error = 0, fhlen, attrflag; 2730 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2731 nfsfh_t *nfhp; 2732 int v3 = NFS_ISV3(dvp); 2733 2734 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2735 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2736 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2737 nfsm_fhtom(dvp, v3); 2738 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2739 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2740 if (npp && !error) { 2741 nfsm_getfh(nfhp, fhlen, v3); 2742 if (*npp) { 2743 np = *npp; 2744 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2745 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2746 np->n_fhp = &np->n_fh; 2747 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2748 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2749 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2750 np->n_fhsize = fhlen; 2751 newvp = NFSTOV(np); 2752 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2753 vref(dvp); 2754 newvp = dvp; 2755 } else { 2756 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2757 if (error) { 2758 m_freem(mrep); 2759 return (error); 2760 } 2761 newvp = NFSTOV(np); 2762 } 2763 if (v3) { 2764 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2765 if (!attrflag && *npp == NULL) { 2766 m_freem(mrep); 2767 if (newvp == dvp) 2768 vrele(newvp); 2769 else 2770 vput(newvp); 2771 return (ENOENT); 2772 } 2773 } else 2774 nfsm_loadattr(newvp, (struct vattr *)0); 2775 } 2776 m_freem(mrep); 2777 nfsmout: 2778 if (npp && *npp == NULL) { 2779 if (error) { 2780 if (newvp) { 2781 if (newvp == dvp) 2782 vrele(newvp); 2783 else 2784 vput(newvp); 2785 } 2786 } else 2787 *npp = np; 2788 } 2789 return (error); 2790 } 2791 2792 /* 2793 * Nfs Version 3 commit rpc 2794 */ 2795 int 2796 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2797 { 2798 caddr_t cp; 2799 u_int32_t *tl; 2800 int32_t t1, t2; 2801 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2802 caddr_t bpos, dpos, cp2; 2803 int error = 0, wccflag = NFSV3_WCCRATTR; 2804 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2805 2806 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2807 return (0); 2808 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2809 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2810 nfsm_fhtom(vp, 1); 2811 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2812 txdr_hyper(offset, tl); 2813 tl += 2; 2814 *tl = txdr_unsigned(cnt); 2815 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2816 nfsm_wcc_data(vp, wccflag); 2817 if (!error) { 2818 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2819 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2820 NFSX_V3WRITEVERF)) { 2821 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2822 NFSX_V3WRITEVERF); 2823 error = NFSERR_STALEWRITEVERF; 2824 } 2825 } 2826 m_freem(mrep); 2827 nfsmout: 2828 return (error); 2829 } 2830 2831 /* 2832 * Kludge City.. 2833 * - make nfs_bmap() essentially a no-op that does no translation 2834 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2835 * (Maybe I could use the process's page mapping, but I was concerned that 2836 * Kernel Write might not be enabled and also figured copyout() would do 2837 * a lot more work than bcopy() and also it currently happens in the 2838 * context of the swapper process (2). 2839 * 2840 * nfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp, 2841 * daddr_t *a_bnp, int *a_runp, int *a_runb) 2842 */ 2843 static int 2844 nfs_bmap(struct vop_bmap_args *ap) 2845 { 2846 struct vnode *vp = ap->a_vp; 2847 2848 if (ap->a_vpp != NULL) 2849 *ap->a_vpp = vp; 2850 if (ap->a_bnp != NULL) 2851 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize); 2852 if (ap->a_runp != NULL) 2853 *ap->a_runp = 0; 2854 if (ap->a_runb != NULL) 2855 *ap->a_runb = 0; 2856 return (0); 2857 } 2858 2859 /* 2860 * Strategy routine. 2861 * 2862 * For async requests when nfsiod(s) are running, queue the request by 2863 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2864 * request. 2865 * 2866 * bio_blkno is NFS block-sized, which depends whether the vnode is a 2867 * regular file or a directory. 2868 */ 2869 static int 2870 nfs_strategy(struct vop_strategy_args *ap) 2871 { 2872 struct bio *bio = ap->a_bio; 2873 struct bio *nbio; 2874 struct buf *bp = bio->bio_buf; 2875 struct thread *td; 2876 int error = 0; 2877 2878 KASSERT(!(bp->b_flags & B_DONE), 2879 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2880 KASSERT(BUF_REFCNT(bp) > 0, 2881 ("nfs_strategy: buffer %p not locked", bp)); 2882 2883 if (bp->b_flags & B_PHYS) 2884 panic("nfs physio"); 2885 2886 if (bp->b_flags & B_ASYNC) 2887 td = NULL; 2888 else 2889 td = curthread; /* XXX */ 2890 2891 /* 2892 * Convert to DEV_BSIZE'd blocks for nfs_doio/nfs_asyncio 2893 */ 2894 nbio = push_bio(bio); 2895 2896 if (bp->b_vp->v_type == VREG) { 2897 int biosize; 2898 2899 biosize = bp->b_vp->v_mount->mnt_stat.f_iosize; 2900 nbio->bio_blkno = ((off_t)bio->bio_blkno * biosize) >> 2901 DEV_BSHIFT; 2902 } else { 2903 nbio->bio_blkno = ((off_t)bio->bio_blkno * NFS_DIRBLKSIZ) >> 2904 DEV_BSHIFT; 2905 } 2906 2907 2908 /* 2909 * If the op is asynchronous and an i/o daemon is waiting 2910 * queue the request, wake it up and wait for completion 2911 * otherwise just do it ourselves. 2912 */ 2913 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td)) 2914 error = nfs_doio(ap->a_vp, nbio, td); 2915 return (error); 2916 } 2917 2918 /* 2919 * Mmap a file 2920 * 2921 * NB Currently unsupported. 2922 * 2923 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2924 * struct thread *a_td) 2925 */ 2926 /* ARGSUSED */ 2927 static int 2928 nfs_mmap(struct vop_mmap_args *ap) 2929 { 2930 return (EINVAL); 2931 } 2932 2933 /* 2934 * fsync vnode op. Just call nfs_flush() with commit == 1. 2935 * 2936 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp, 2937 * struct ucred * a_cred, int a_waitfor, struct thread *a_td) 2938 */ 2939 /* ARGSUSED */ 2940 static int 2941 nfs_fsync(struct vop_fsync_args *ap) 2942 { 2943 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1)); 2944 } 2945 2946 /* 2947 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 2948 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 2949 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 2950 * set the buffer contains data that has already been written to the server 2951 * and which now needs a commit RPC. 2952 * 2953 * If commit is 0 we only take one pass and only flush buffers containing new 2954 * dirty data. 2955 * 2956 * If commit is 1 we take two passes, issuing a commit RPC in the second 2957 * pass. 2958 * 2959 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 2960 * to completely flush all pending data. 2961 * 2962 * Note that the RB_SCAN code properly handles the case where the 2963 * callback might block and directly or indirectly (another thread) cause 2964 * the RB tree to change. 2965 */ 2966 2967 #ifndef NFS_COMMITBVECSIZ 2968 #define NFS_COMMITBVECSIZ 16 2969 #endif 2970 2971 struct nfs_flush_info { 2972 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 2973 struct thread *td; 2974 struct vnode *vp; 2975 int waitfor; 2976 int slpflag; 2977 int slptimeo; 2978 int loops; 2979 struct buf *bvary[NFS_COMMITBVECSIZ]; 2980 int bvsize; 2981 off_t beg_off; 2982 off_t end_off; 2983 }; 2984 2985 static int nfs_flush_bp(struct buf *bp, void *data); 2986 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 2987 2988 int 2989 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2990 { 2991 struct nfsnode *np = VTONFS(vp); 2992 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2993 struct nfs_flush_info info; 2994 int error; 2995 2996 bzero(&info, sizeof(info)); 2997 info.td = td; 2998 info.vp = vp; 2999 info.waitfor = waitfor; 3000 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 3001 info.loops = 0; 3002 3003 do { 3004 /* 3005 * Flush mode 3006 */ 3007 info.mode = NFI_FLUSHNEW; 3008 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3009 nfs_flush_bp, &info); 3010 3011 /* 3012 * Take a second pass if committing and no error occured. 3013 * Clean up any left over collection (whether an error 3014 * occurs or not). 3015 */ 3016 if (commit && error == 0) { 3017 info.mode = NFI_COMMIT; 3018 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 3019 nfs_flush_bp, &info); 3020 if (info.bvsize) 3021 error = nfs_flush_docommit(&info, error); 3022 } 3023 3024 /* 3025 * Wait for pending I/O to complete before checking whether 3026 * any further dirty buffers exist. 3027 */ 3028 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) { 3029 vp->v_track_write.bk_waitflag = 1; 3030 error = tsleep(&vp->v_track_write, 3031 info.slpflag, "nfsfsync", info.slptimeo); 3032 if (error) { 3033 /* 3034 * We have to be able to break out if this 3035 * is an 'intr' mount. 3036 */ 3037 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 3038 error = -EINTR; 3039 break; 3040 } 3041 3042 /* 3043 * Since we do not process pending signals, 3044 * once we get a PCATCH our tsleep() will no 3045 * longer sleep, switch to a fixed timeout 3046 * instead. 3047 */ 3048 if (info.slpflag == PCATCH) { 3049 info.slpflag = 0; 3050 info.slptimeo = 2 * hz; 3051 } 3052 error = 0; 3053 } 3054 } 3055 ++info.loops; 3056 /* 3057 * Loop if we are flushing synchronous as well as committing, 3058 * and dirty buffers are still present. Otherwise we might livelock. 3059 */ 3060 } while (waitfor == MNT_WAIT && commit && 3061 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3062 3063 /* 3064 * The callbacks have to return a negative error to terminate the 3065 * RB scan. 3066 */ 3067 if (error < 0) 3068 error = -error; 3069 3070 /* 3071 * Deal with any error collection 3072 */ 3073 if (np->n_flag & NWRITEERR) { 3074 error = np->n_error; 3075 np->n_flag &= ~NWRITEERR; 3076 } 3077 return (error); 3078 } 3079 3080 3081 static 3082 int 3083 nfs_flush_bp(struct buf *bp, void *data) 3084 { 3085 struct nfs_flush_info *info = data; 3086 off_t toff; 3087 int error; 3088 3089 error = 0; 3090 switch(info->mode) { 3091 case NFI_FLUSHNEW: 3092 crit_enter(); 3093 if (info->loops && info->waitfor == MNT_WAIT) { 3094 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3095 if (error) { 3096 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3097 if (info->slpflag & PCATCH) 3098 lkflags |= LK_PCATCH; 3099 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3100 info->slptimeo); 3101 } 3102 } else { 3103 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3104 } 3105 if (error == 0) { 3106 KKASSERT(bp->b_vp == info->vp); 3107 3108 if ((bp->b_flags & B_DELWRI) == 0) 3109 panic("nfs_fsync: not dirty"); 3110 if (bp->b_flags & B_NEEDCOMMIT) { 3111 BUF_UNLOCK(bp); 3112 crit_exit(); 3113 break; 3114 } 3115 bremfree(bp); 3116 3117 bp->b_flags |= B_ASYNC; 3118 crit_exit(); 3119 VOP_BWRITE(bp->b_vp, bp); 3120 } else { 3121 crit_exit(); 3122 error = 0; 3123 } 3124 break; 3125 case NFI_COMMIT: 3126 /* 3127 * Only process buffers in need of a commit which we can 3128 * immediately lock. This may prevent a buffer from being 3129 * committed, but the normal flush loop will block on the 3130 * same buffer so we shouldn't get into an endless loop. 3131 */ 3132 crit_enter(); 3133 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3134 (B_DELWRI | B_NEEDCOMMIT) || 3135 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 3136 crit_exit(); 3137 break; 3138 } 3139 3140 KKASSERT(bp->b_vp == info->vp); 3141 bremfree(bp); 3142 3143 /* 3144 * NOTE: we are not clearing B_DONE here, so we have 3145 * to do it later on in this routine if we intend to 3146 * initiate I/O on the bp. 3147 * 3148 * Note: to avoid loopback deadlocks, we do not 3149 * assign b_runningbufspace. 3150 */ 3151 vfs_busy_pages(bp, 1); 3152 3153 info->bvary[info->bvsize] = bp; 3154 toff = ((u_quad_t)bp->b_bio2.bio_blkno) * DEV_BSIZE + 3155 bp->b_dirtyoff; 3156 if (info->bvsize == 0 || toff < info->beg_off) 3157 info->beg_off = toff; 3158 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3159 if (info->bvsize == 0 || toff > info->end_off) 3160 info->end_off = toff; 3161 ++info->bvsize; 3162 if (info->bvsize == NFS_COMMITBVECSIZ) { 3163 error = nfs_flush_docommit(info, 0); 3164 KKASSERT(info->bvsize == 0); 3165 } 3166 crit_exit(); 3167 } 3168 return (error); 3169 } 3170 3171 static 3172 int 3173 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3174 { 3175 struct vnode *vp; 3176 struct buf *bp; 3177 off_t bytes; 3178 int retv; 3179 int i; 3180 3181 vp = info->vp; 3182 3183 if (info->bvsize > 0) { 3184 /* 3185 * Commit data on the server, as required. Note that 3186 * nfs_commit will use the vnode's cred for the commit. 3187 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3188 */ 3189 bytes = info->end_off - info->beg_off; 3190 if (bytes > 0x40000000) 3191 bytes = 0x40000000; 3192 if (error) { 3193 retv = -error; 3194 } else { 3195 retv = nfs_commit(vp, info->beg_off, 3196 (int)bytes, info->td); 3197 if (retv == NFSERR_STALEWRITEVERF) 3198 nfs_clearcommit(vp->v_mount); 3199 } 3200 3201 /* 3202 * Now, either mark the blocks I/O done or mark the 3203 * blocks dirty, depending on whether the commit 3204 * succeeded. 3205 */ 3206 for (i = 0; i < info->bvsize; ++i) { 3207 bp = info->bvary[i]; 3208 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3209 if (retv) { 3210 /* 3211 * Error, leave B_DELWRI intact 3212 */ 3213 vfs_unbusy_pages(bp); 3214 brelse(bp); 3215 } else { 3216 /* 3217 * Success, remove B_DELWRI ( bundirty() ). 3218 * 3219 * b_dirtyoff/b_dirtyend seem to be NFS 3220 * specific. We should probably move that 3221 * into bundirty(). XXX 3222 * 3223 * We are faking an I/O write, we have to 3224 * start the transaction in order to 3225 * immediately biodone() it. 3226 */ 3227 crit_enter(); 3228 bp->b_flags |= B_ASYNC; 3229 bundirty(bp); 3230 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3231 bp->b_dirtyoff = bp->b_dirtyend = 0; 3232 crit_exit(); 3233 biodone(&bp->b_bio1); 3234 } 3235 } 3236 info->bvsize = 0; 3237 } 3238 return (error); 3239 } 3240 3241 /* 3242 * NFS advisory byte-level locks. 3243 * Currently unsupported. 3244 * 3245 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3246 * int a_flags) 3247 */ 3248 static int 3249 nfs_advlock(struct vop_advlock_args *ap) 3250 { 3251 struct nfsnode *np = VTONFS(ap->a_vp); 3252 3253 /* 3254 * The following kludge is to allow diskless support to work 3255 * until a real NFS lockd is implemented. Basically, just pretend 3256 * that this is a local lock. 3257 */ 3258 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3259 } 3260 3261 /* 3262 * Print out the contents of an nfsnode. 3263 * 3264 * nfs_print(struct vnode *a_vp) 3265 */ 3266 static int 3267 nfs_print(struct vop_print_args *ap) 3268 { 3269 struct vnode *vp = ap->a_vp; 3270 struct nfsnode *np = VTONFS(vp); 3271 3272 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3273 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3274 if (vp->v_type == VFIFO) 3275 fifo_printinfo(vp); 3276 printf("\n"); 3277 return (0); 3278 } 3279 3280 /* 3281 * Just call nfs_writebp() with the force argument set to 1. 3282 * 3283 * NOTE: B_DONE may or may not be set in a_bp on call. 3284 * 3285 * nfs_bwrite(struct vnode *a_bp) 3286 */ 3287 static int 3288 nfs_bwrite(struct vop_bwrite_args *ap) 3289 { 3290 return (nfs_writebp(ap->a_bp, 1, curthread)); 3291 } 3292 3293 /* 3294 * This is a clone of vn_bwrite(), except that it also handles the 3295 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer. 3296 */ 3297 int 3298 nfs_writebp(struct buf *bp, int force, struct thread *td) 3299 { 3300 int error; 3301 3302 if (BUF_REFCNT(bp) == 0) 3303 panic("bwrite: buffer is not locked???"); 3304 3305 if (bp->b_flags & B_INVAL) { 3306 brelse(bp); 3307 return(0); 3308 } 3309 3310 bp->b_flags |= B_CACHE; 3311 3312 /* 3313 * Undirty the bp. We will redirty it later if the I/O fails. 3314 */ 3315 crit_enter(); 3316 bundirty(bp); 3317 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3318 crit_exit(); 3319 3320 /* 3321 * Note: to avoid loopback deadlocks, we do not 3322 * assign b_runningbufspace. 3323 */ 3324 vfs_busy_pages(bp, 1); 3325 BUF_KERNPROC(bp); 3326 3327 if (bp->b_flags & B_ASYNC) { 3328 vn_strategy(bp->b_vp, &bp->b_bio1); 3329 error = 0; 3330 } else { 3331 vn_strategy(bp->b_vp, &bp->b_bio1); 3332 error = biowait(bp); 3333 brelse(bp); 3334 } 3335 return (error); 3336 } 3337 3338 /* 3339 * nfs special file access vnode op. 3340 * Essentially just get vattr and then imitate iaccess() since the device is 3341 * local to the client. 3342 * 3343 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3344 * struct thread *a_td) 3345 */ 3346 static int 3347 nfsspec_access(struct vop_access_args *ap) 3348 { 3349 struct vattr *vap; 3350 gid_t *gp; 3351 struct ucred *cred = ap->a_cred; 3352 struct vnode *vp = ap->a_vp; 3353 mode_t mode = ap->a_mode; 3354 struct vattr vattr; 3355 int i; 3356 int error; 3357 3358 /* 3359 * Disallow write attempts on filesystems mounted read-only; 3360 * unless the file is a socket, fifo, or a block or character 3361 * device resident on the filesystem. 3362 */ 3363 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3364 switch (vp->v_type) { 3365 case VREG: 3366 case VDIR: 3367 case VLNK: 3368 return (EROFS); 3369 default: 3370 break; 3371 } 3372 } 3373 /* 3374 * If you're the super-user, 3375 * you always get access. 3376 */ 3377 if (cred->cr_uid == 0) 3378 return (0); 3379 vap = &vattr; 3380 error = VOP_GETATTR(vp, vap, ap->a_td); 3381 if (error) 3382 return (error); 3383 /* 3384 * Access check is based on only one of owner, group, public. 3385 * If not owner, then check group. If not a member of the 3386 * group, then check public access. 3387 */ 3388 if (cred->cr_uid != vap->va_uid) { 3389 mode >>= 3; 3390 gp = cred->cr_groups; 3391 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3392 if (vap->va_gid == *gp) 3393 goto found; 3394 mode >>= 3; 3395 found: 3396 ; 3397 } 3398 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3399 return (error); 3400 } 3401 3402 /* 3403 * Read wrapper for special devices. 3404 * 3405 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3406 * struct ucred *a_cred) 3407 */ 3408 static int 3409 nfsspec_read(struct vop_read_args *ap) 3410 { 3411 struct nfsnode *np = VTONFS(ap->a_vp); 3412 3413 /* 3414 * Set access flag. 3415 */ 3416 np->n_flag |= NACC; 3417 getnanotime(&np->n_atim); 3418 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3419 } 3420 3421 /* 3422 * Write wrapper for special devices. 3423 * 3424 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3425 * struct ucred *a_cred) 3426 */ 3427 static int 3428 nfsspec_write(struct vop_write_args *ap) 3429 { 3430 struct nfsnode *np = VTONFS(ap->a_vp); 3431 3432 /* 3433 * Set update flag. 3434 */ 3435 np->n_flag |= NUPD; 3436 getnanotime(&np->n_mtim); 3437 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3438 } 3439 3440 /* 3441 * Close wrapper for special devices. 3442 * 3443 * Update the times on the nfsnode then do device close. 3444 * 3445 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3446 * struct thread *a_td) 3447 */ 3448 static int 3449 nfsspec_close(struct vop_close_args *ap) 3450 { 3451 struct vnode *vp = ap->a_vp; 3452 struct nfsnode *np = VTONFS(vp); 3453 struct vattr vattr; 3454 3455 if (np->n_flag & (NACC | NUPD)) { 3456 np->n_flag |= NCHG; 3457 if (vp->v_usecount == 1 && 3458 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3459 VATTR_NULL(&vattr); 3460 if (np->n_flag & NACC) 3461 vattr.va_atime = np->n_atim; 3462 if (np->n_flag & NUPD) 3463 vattr.va_mtime = np->n_mtim; 3464 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3465 } 3466 } 3467 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3468 } 3469 3470 /* 3471 * Read wrapper for fifos. 3472 * 3473 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3474 * struct ucred *a_cred) 3475 */ 3476 static int 3477 nfsfifo_read(struct vop_read_args *ap) 3478 { 3479 struct nfsnode *np = VTONFS(ap->a_vp); 3480 3481 /* 3482 * Set access flag. 3483 */ 3484 np->n_flag |= NACC; 3485 getnanotime(&np->n_atim); 3486 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3487 } 3488 3489 /* 3490 * Write wrapper for fifos. 3491 * 3492 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3493 * struct ucred *a_cred) 3494 */ 3495 static int 3496 nfsfifo_write(struct vop_write_args *ap) 3497 { 3498 struct nfsnode *np = VTONFS(ap->a_vp); 3499 3500 /* 3501 * Set update flag. 3502 */ 3503 np->n_flag |= NUPD; 3504 getnanotime(&np->n_mtim); 3505 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3506 } 3507 3508 /* 3509 * Close wrapper for fifos. 3510 * 3511 * Update the times on the nfsnode then do fifo close. 3512 * 3513 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3514 */ 3515 static int 3516 nfsfifo_close(struct vop_close_args *ap) 3517 { 3518 struct vnode *vp = ap->a_vp; 3519 struct nfsnode *np = VTONFS(vp); 3520 struct vattr vattr; 3521 struct timespec ts; 3522 3523 if (np->n_flag & (NACC | NUPD)) { 3524 getnanotime(&ts); 3525 if (np->n_flag & NACC) 3526 np->n_atim = ts; 3527 if (np->n_flag & NUPD) 3528 np->n_mtim = ts; 3529 np->n_flag |= NCHG; 3530 if (vp->v_usecount == 1 && 3531 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3532 VATTR_NULL(&vattr); 3533 if (np->n_flag & NACC) 3534 vattr.va_atime = np->n_atim; 3535 if (np->n_flag & NUPD) 3536 vattr.va_mtime = np->n_mtim; 3537 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3538 } 3539 } 3540 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3541 } 3542 3543