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 1.61 2006/05/06 02:43:14 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/nlookup.h> 59 #include <sys/socket.h> 60 #include <sys/vnode.h> 61 #include <sys/dirent.h> 62 #include <sys/fcntl.h> 63 #include <sys/lockf.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/conf.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_zone.h> 71 72 #include <sys/buf2.h> 73 74 #include <vfs/fifofs/fifo.h> 75 #include <vfs/ufs/dir.h> 76 77 #undef DIRBLKSIZ 78 79 #include "rpcv2.h" 80 #include "nfsproto.h" 81 #include "nfs.h" 82 #include "nfsmount.h" 83 #include "nfsnode.h" 84 #include "xdr_subs.h" 85 #include "nfsm_subs.h" 86 87 #include <net/if.h> 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 91 #include <sys/thread2.h> 92 93 /* Defs */ 94 #define TRUE 1 95 #define FALSE 0 96 97 static int nfsspec_read (struct vop_read_args *); 98 static int nfsspec_write (struct vop_write_args *); 99 static int nfsfifo_read (struct vop_read_args *); 100 static int nfsfifo_write (struct vop_write_args *); 101 static int nfsspec_close (struct vop_close_args *); 102 static int nfsfifo_close (struct vop_close_args *); 103 #define nfs_poll vop_nopoll 104 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 105 static int nfs_lookup (struct vop_old_lookup_args *); 106 static int nfs_create (struct vop_old_create_args *); 107 static int nfs_mknod (struct vop_old_mknod_args *); 108 static int nfs_open (struct vop_open_args *); 109 static int nfs_close (struct vop_close_args *); 110 static int nfs_access (struct vop_access_args *); 111 static int nfs_getattr (struct vop_getattr_args *); 112 static int nfs_setattr (struct vop_setattr_args *); 113 static int nfs_read (struct vop_read_args *); 114 static int nfs_mmap (struct vop_mmap_args *); 115 static int nfs_fsync (struct vop_fsync_args *); 116 static int nfs_remove (struct vop_old_remove_args *); 117 static int nfs_link (struct vop_old_link_args *); 118 static int nfs_rename (struct vop_old_rename_args *); 119 static int nfs_mkdir (struct vop_old_mkdir_args *); 120 static int nfs_rmdir (struct vop_old_rmdir_args *); 121 static int nfs_symlink (struct vop_old_symlink_args *); 122 static int nfs_readdir (struct vop_readdir_args *); 123 static int nfs_bmap (struct vop_bmap_args *); 124 static int nfs_strategy (struct vop_strategy_args *); 125 static int nfs_lookitup (struct vnode *, const char *, int, 126 struct ucred *, struct thread *, struct nfsnode **); 127 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 128 static int nfsspec_access (struct vop_access_args *); 129 static int nfs_readlink (struct vop_readlink_args *); 130 static int nfs_print (struct vop_print_args *); 131 static int nfs_advlock (struct vop_advlock_args *); 132 133 static int nfs_nresolve (struct vop_nresolve_args *); 134 /* 135 * Global vfs data structures for nfs 136 */ 137 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 138 { &vop_default_desc, vop_defaultop }, 139 { &vop_access_desc, (vnodeopv_entry_t) nfs_access }, 140 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock }, 141 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap }, 142 { &vop_close_desc, (vnodeopv_entry_t) nfs_close }, 143 { &vop_old_create_desc, (vnodeopv_entry_t) nfs_create }, 144 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 145 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 146 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages }, 147 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages }, 148 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 149 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 150 { &vop_old_link_desc, (vnodeopv_entry_t) nfs_link }, 151 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 152 { &vop_old_lookup_desc, (vnodeopv_entry_t) nfs_lookup }, 153 { &vop_old_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir }, 154 { &vop_old_mknod_desc, (vnodeopv_entry_t) nfs_mknod }, 155 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap }, 156 { &vop_open_desc, (vnodeopv_entry_t) nfs_open }, 157 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll }, 158 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 159 { &vop_read_desc, (vnodeopv_entry_t) nfs_read }, 160 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir }, 161 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink }, 162 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 163 { &vop_old_remove_desc, (vnodeopv_entry_t) nfs_remove }, 164 { &vop_old_rename_desc, (vnodeopv_entry_t) nfs_rename }, 165 { &vop_old_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir }, 166 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 167 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy }, 168 { &vop_old_symlink_desc, (vnodeopv_entry_t) nfs_symlink }, 169 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 170 { &vop_write_desc, (vnodeopv_entry_t) nfs_write }, 171 172 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve }, 173 { NULL, NULL } 174 }; 175 176 /* 177 * Special device vnode ops 178 */ 179 struct vnodeopv_entry_desc nfsv2_specop_entries[] = { 180 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate }, 181 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 182 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close }, 183 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 184 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 185 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 186 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 187 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 188 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 189 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read }, 190 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 191 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 192 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 193 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write }, 194 { NULL, NULL } 195 }; 196 197 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = { 198 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate }, 199 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 200 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close }, 201 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 202 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 203 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 204 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 205 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 206 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 207 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read }, 208 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 209 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 210 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 211 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write }, 212 { NULL, NULL } 213 }; 214 215 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 216 struct componentname *cnp, 217 struct vattr *vap); 218 static int nfs_removerpc (struct vnode *dvp, const char *name, 219 int namelen, 220 struct ucred *cred, struct thread *td); 221 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 222 int fnamelen, struct vnode *tdvp, 223 const char *tnameptr, int tnamelen, 224 struct ucred *cred, struct thread *td); 225 static int nfs_renameit (struct vnode *sdvp, 226 struct componentname *scnp, 227 struct sillyrename *sp); 228 229 /* 230 * Global variables 231 */ 232 extern u_int32_t nfs_true, nfs_false; 233 extern u_int32_t nfs_xdrneg1; 234 extern struct nfsstats nfsstats; 235 extern nfstype nfsv3_type[9]; 236 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 237 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 238 int nfs_numasync = 0; 239 240 SYSCTL_DECL(_vfs_nfs); 241 242 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO; 243 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 244 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 245 246 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 247 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 248 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 249 250 static int nfsv3_commit_on_close = 0; 251 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 252 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 253 #if 0 254 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 255 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 256 257 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 258 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 259 #endif 260 261 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 262 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 263 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 264 static int 265 nfs3_access_otw(struct vnode *vp, int wmode, 266 struct thread *td, struct ucred *cred) 267 { 268 const int v3 = 1; 269 u_int32_t *tl; 270 int error = 0, attrflag; 271 272 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 273 caddr_t bpos, dpos, cp2; 274 int32_t t1, t2; 275 caddr_t cp; 276 u_int32_t rmode; 277 struct nfsnode *np = VTONFS(vp); 278 279 nfsstats.rpccnt[NFSPROC_ACCESS]++; 280 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 281 nfsm_fhtom(vp, v3); 282 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 283 *tl = txdr_unsigned(wmode); 284 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 285 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 286 if (!error) { 287 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 288 rmode = fxdr_unsigned(u_int32_t, *tl); 289 np->n_mode = rmode; 290 np->n_modeuid = cred->cr_uid; 291 np->n_modestamp = mycpu->gd_time_seconds; 292 } 293 m_freem(mrep); 294 nfsmout: 295 return error; 296 } 297 298 /* 299 * nfs access vnode op. 300 * For nfs version 2, just return ok. File accesses may fail later. 301 * For nfs version 3, use the access rpc to check accessibility. If file modes 302 * are changed on the server, accesses might still fail later. 303 * 304 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 305 * struct thread *a_td) 306 */ 307 static int 308 nfs_access(struct vop_access_args *ap) 309 { 310 struct vnode *vp = ap->a_vp; 311 thread_t td = curthread; 312 int error = 0; 313 u_int32_t mode, wmode; 314 int v3 = NFS_ISV3(vp); 315 struct nfsnode *np = VTONFS(vp); 316 317 /* 318 * Disallow write attempts on filesystems mounted read-only; 319 * unless the file is a socket, fifo, or a block or character 320 * device resident on the filesystem. 321 */ 322 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 323 switch (vp->v_type) { 324 case VREG: 325 case VDIR: 326 case VLNK: 327 return (EROFS); 328 default: 329 break; 330 } 331 } 332 /* 333 * For nfs v3, check to see if we have done this recently, and if 334 * so return our cached result instead of making an ACCESS call. 335 * If not, do an access rpc, otherwise you are stuck emulating 336 * ufs_access() locally using the vattr. This may not be correct, 337 * since the server may apply other access criteria such as 338 * client uid-->server uid mapping that we do not know about. 339 */ 340 if (v3) { 341 if (ap->a_mode & VREAD) 342 mode = NFSV3ACCESS_READ; 343 else 344 mode = 0; 345 if (vp->v_type != VDIR) { 346 if (ap->a_mode & VWRITE) 347 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 348 if (ap->a_mode & VEXEC) 349 mode |= NFSV3ACCESS_EXECUTE; 350 } else { 351 if (ap->a_mode & VWRITE) 352 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 353 NFSV3ACCESS_DELETE); 354 if (ap->a_mode & VEXEC) 355 mode |= NFSV3ACCESS_LOOKUP; 356 } 357 /* XXX safety belt, only make blanket request if caching */ 358 if (nfsaccess_cache_timeout > 0) { 359 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 360 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 361 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 362 } else { 363 wmode = mode; 364 } 365 366 /* 367 * Does our cached result allow us to give a definite yes to 368 * this request? 369 */ 370 if (np->n_modestamp && 371 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 372 (ap->a_cred->cr_uid == np->n_modeuid) && 373 ((np->n_mode & mode) == mode)) { 374 nfsstats.accesscache_hits++; 375 } else { 376 /* 377 * Either a no, or a don't know. Go to the wire. 378 */ 379 nfsstats.accesscache_misses++; 380 error = nfs3_access_otw(vp, wmode, td, ap->a_cred); 381 if (!error) { 382 if ((np->n_mode & mode) != mode) { 383 error = EACCES; 384 } 385 } 386 } 387 } else { 388 if ((error = nfsspec_access(ap)) != 0) 389 return (error); 390 391 /* 392 * Attempt to prevent a mapped root from accessing a file 393 * which it shouldn't. We try to read a byte from the file 394 * if the user is root and the file is not zero length. 395 * After calling nfsspec_access, we should have the correct 396 * file size cached. 397 */ 398 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 399 && VTONFS(vp)->n_size > 0) { 400 struct iovec aiov; 401 struct uio auio; 402 char buf[1]; 403 404 aiov.iov_base = buf; 405 aiov.iov_len = 1; 406 auio.uio_iov = &aiov; 407 auio.uio_iovcnt = 1; 408 auio.uio_offset = 0; 409 auio.uio_resid = 1; 410 auio.uio_segflg = UIO_SYSSPACE; 411 auio.uio_rw = UIO_READ; 412 auio.uio_td = td; 413 414 if (vp->v_type == VREG) { 415 error = nfs_readrpc(vp, &auio); 416 } else if (vp->v_type == VDIR) { 417 char* bp; 418 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 419 aiov.iov_base = bp; 420 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 421 error = nfs_readdirrpc(vp, &auio); 422 free(bp, M_TEMP); 423 } else if (vp->v_type == VLNK) { 424 error = nfs_readlinkrpc(vp, &auio); 425 } else { 426 error = EACCES; 427 } 428 } 429 } 430 /* 431 * [re]record creds for reading and/or writing if access 432 * was granted. Assume the NFS server will grant read access 433 * for execute requests. 434 */ 435 if (error == 0) { 436 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 437 crhold(ap->a_cred); 438 if (np->n_rucred) 439 crfree(np->n_rucred); 440 np->n_rucred = ap->a_cred; 441 } 442 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 443 crhold(ap->a_cred); 444 if (np->n_wucred) 445 crfree(np->n_wucred); 446 np->n_wucred = ap->a_cred; 447 } 448 } 449 return(error); 450 } 451 452 /* 453 * nfs open vnode op 454 * Check to see if the type is ok 455 * and that deletion is not in progress. 456 * For paged in text files, you will need to flush the page cache 457 * if consistency is lost. 458 * 459 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 460 */ 461 /* ARGSUSED */ 462 static int 463 nfs_open(struct vop_open_args *ap) 464 { 465 struct vnode *vp = ap->a_vp; 466 struct nfsnode *np = VTONFS(vp); 467 struct vattr vattr; 468 int error; 469 470 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 471 #ifdef DIAGNOSTIC 472 printf("open eacces vtyp=%d\n",vp->v_type); 473 #endif 474 return (EOPNOTSUPP); 475 } 476 477 /* 478 * Clear the attribute cache only if opening with write access. It 479 * is unclear if we should do this at all here, but we certainly 480 * should not clear the cache unconditionally simply because a file 481 * is being opened. 482 */ 483 if (ap->a_mode & FWRITE) 484 np->n_attrstamp = 0; 485 486 /* 487 * For normal NFS, reconcile changes made locally verses 488 * changes made remotely. Note that VOP_GETATTR only goes 489 * to the wire if the cached attribute has timed out or been 490 * cleared. 491 * 492 * If local modifications have been made clear the attribute 493 * cache to force an attribute and modified time check. If 494 * GETATTR detects that the file has been changed by someone 495 * other then us it will set NRMODIFIED. 496 * 497 * If we are opening a directory and local changes have been 498 * made we have to invalidate the cache in order to ensure 499 * that we get the most up-to-date information from the 500 * server. XXX 501 */ 502 if (np->n_flag & NLMODIFIED) { 503 np->n_attrstamp = 0; 504 if (vp->v_type == VDIR) { 505 error = nfs_vinvalbuf(vp, V_SAVE, 1); 506 if (error == EINTR) 507 return (error); 508 nfs_invaldir(vp); 509 } 510 } 511 error = VOP_GETATTR(vp, &vattr); 512 if (error) 513 return (error); 514 if (np->n_flag & NRMODIFIED) { 515 if (vp->v_type == VDIR) 516 nfs_invaldir(vp); 517 error = nfs_vinvalbuf(vp, V_SAVE, 1); 518 if (error == EINTR) 519 return (error); 520 np->n_flag &= ~NRMODIFIED; 521 } 522 523 return (vop_stdopen(ap)); 524 } 525 526 /* 527 * nfs close vnode op 528 * What an NFS client should do upon close after writing is a debatable issue. 529 * Most NFS clients push delayed writes to the server upon close, basically for 530 * two reasons: 531 * 1 - So that any write errors may be reported back to the client process 532 * doing the close system call. By far the two most likely errors are 533 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 534 * 2 - To put a worst case upper bound on cache inconsistency between 535 * multiple clients for the file. 536 * There is also a consistency problem for Version 2 of the protocol w.r.t. 537 * not being able to tell if other clients are writing a file concurrently, 538 * since there is no way of knowing if the changed modify time in the reply 539 * is only due to the write for this client. 540 * (NFS Version 3 provides weak cache consistency data in the reply that 541 * should be sufficient to detect and handle this case.) 542 * 543 * The current code does the following: 544 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 545 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 546 * or commit them (this satisfies 1 and 2 except for the 547 * case where the server crashes after this close but 548 * before the commit RPC, which is felt to be "good 549 * enough". Changing the last argument to nfs_flush() to 550 * a 1 would force a commit operation, if it is felt a 551 * commit is necessary now. 552 * for NQNFS - do nothing now, since 2 is dealt with via leases and 553 * 1 should be dealt with via an fsync() system call for 554 * cases where write errors are important. 555 * 556 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag, 557 * struct ucred *a_cred, struct thread *a_td) 558 */ 559 /* ARGSUSED */ 560 static int 561 nfs_close(struct vop_close_args *ap) 562 { 563 struct vnode *vp = ap->a_vp; 564 struct nfsnode *np = VTONFS(vp); 565 int error = 0; 566 thread_t td = curthread; 567 568 if (vp->v_type == VREG) { 569 if (np->n_flag & NLMODIFIED) { 570 if (NFS_ISV3(vp)) { 571 /* 572 * Under NFSv3 we have dirty buffers to dispose of. We 573 * must flush them to the NFS server. We have the option 574 * of waiting all the way through the commit rpc or just 575 * waiting for the initial write. The default is to only 576 * wait through the initial write so the data is in the 577 * server's cache, which is roughly similar to the state 578 * a standard disk subsystem leaves the file in on close(). 579 * 580 * We cannot clear the NLMODIFIED bit in np->n_flag due to 581 * potential races with other processes, and certainly 582 * cannot clear it if we don't commit. 583 */ 584 int cm = nfsv3_commit_on_close ? 1 : 0; 585 error = nfs_flush(vp, MNT_WAIT, td, cm); 586 /* np->n_flag &= ~NLMODIFIED; */ 587 } else { 588 error = nfs_vinvalbuf(vp, V_SAVE, 1); 589 } 590 np->n_attrstamp = 0; 591 } 592 if (np->n_flag & NWRITEERR) { 593 np->n_flag &= ~NWRITEERR; 594 error = np->n_error; 595 } 596 } 597 vop_stdclose(ap); 598 return (error); 599 } 600 601 /* 602 * nfs getattr call from vfs. 603 * 604 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 605 * struct thread *a_td) 606 */ 607 static int 608 nfs_getattr(struct vop_getattr_args *ap) 609 { 610 struct vnode *vp = ap->a_vp; 611 struct nfsnode *np = VTONFS(vp); 612 caddr_t cp; 613 u_int32_t *tl; 614 int32_t t1, t2; 615 caddr_t bpos, dpos; 616 int error = 0; 617 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 618 int v3 = NFS_ISV3(vp); 619 thread_t td = curthread; 620 621 /* 622 * Update local times for special files. 623 */ 624 if (np->n_flag & (NACC | NUPD)) 625 np->n_flag |= NCHG; 626 /* 627 * First look in the cache. 628 */ 629 if (nfs_getattrcache(vp, ap->a_vap) == 0) 630 return (0); 631 632 if (v3 && nfsaccess_cache_timeout > 0) { 633 nfsstats.accesscache_misses++; 634 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK)); 635 if (nfs_getattrcache(vp, ap->a_vap) == 0) 636 return (0); 637 } 638 639 nfsstats.rpccnt[NFSPROC_GETATTR]++; 640 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 641 nfsm_fhtom(vp, v3); 642 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK)); 643 if (!error) { 644 nfsm_loadattr(vp, ap->a_vap); 645 } 646 m_freem(mrep); 647 nfsmout: 648 return (error); 649 } 650 651 /* 652 * nfs setattr call. 653 * 654 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 655 * struct vattr *a_vap, struct ucred *a_cred) 656 */ 657 static int 658 nfs_setattr(struct vop_setattr_args *ap) 659 { 660 struct vnode *vp = ap->a_vp; 661 struct nfsnode *np = VTONFS(vp); 662 struct vattr *vap = ap->a_vap; 663 int error = 0; 664 u_quad_t tsize; 665 thread_t td = curthread; 666 667 #ifndef nolint 668 tsize = (u_quad_t)0; 669 #endif 670 671 /* 672 * Setting of flags is not supported. 673 */ 674 if (vap->va_flags != VNOVAL) 675 return (EOPNOTSUPP); 676 677 /* 678 * Disallow write attempts if the filesystem is mounted read-only. 679 */ 680 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 681 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 682 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 683 (vp->v_mount->mnt_flag & MNT_RDONLY)) 684 return (EROFS); 685 if (vap->va_size != VNOVAL) { 686 switch (vp->v_type) { 687 case VDIR: 688 return (EISDIR); 689 case VCHR: 690 case VBLK: 691 case VSOCK: 692 case VFIFO: 693 if (vap->va_mtime.tv_sec == VNOVAL && 694 vap->va_atime.tv_sec == VNOVAL && 695 vap->va_mode == (mode_t)VNOVAL && 696 vap->va_uid == (uid_t)VNOVAL && 697 vap->va_gid == (gid_t)VNOVAL) 698 return (0); 699 vap->va_size = VNOVAL; 700 break; 701 default: 702 /* 703 * Disallow write attempts if the filesystem is 704 * mounted read-only. 705 */ 706 if (vp->v_mount->mnt_flag & MNT_RDONLY) 707 return (EROFS); 708 709 /* 710 * This is nasty. The RPCs we send to flush pending 711 * data often return attribute information which is 712 * cached via a callback to nfs_loadattrcache(), which 713 * has the effect of changing our notion of the file 714 * size. Due to flushed appends and other operations 715 * the file size can be set to virtually anything, 716 * including values that do not match either the old 717 * or intended file size. 718 * 719 * When this condition is detected we must loop to 720 * try the operation again. Hopefully no more 721 * flushing is required on the loop so it works the 722 * second time around. THIS CASE ALMOST ALWAYS 723 * HAPPENS! 724 */ 725 tsize = np->n_size; 726 again: 727 error = nfs_meta_setsize(vp, td, vap->va_size); 728 729 if (np->n_flag & NLMODIFIED) { 730 if (vap->va_size == 0) 731 error = nfs_vinvalbuf(vp, 0, 1); 732 else 733 error = nfs_vinvalbuf(vp, V_SAVE, 1); 734 } 735 /* 736 * note: this loop case almost always happens at 737 * least once per truncation. 738 */ 739 if (error == 0 && np->n_size != vap->va_size) 740 goto again; 741 np->n_vattr.va_size = vap->va_size; 742 break; 743 } 744 } else if ((vap->va_mtime.tv_sec != VNOVAL || 745 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 746 vp->v_type == VREG && 747 (error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR 748 ) { 749 return (error); 750 } 751 error = nfs_setattrrpc(vp, vap, ap->a_cred, td); 752 753 /* 754 * Sanity check if a truncation was issued. This should only occur 755 * if multiple processes are racing on the same file. 756 */ 757 if (error == 0 && vap->va_size != VNOVAL && 758 np->n_size != vap->va_size) { 759 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 760 goto again; 761 } 762 if (error && vap->va_size != VNOVAL) { 763 np->n_size = np->n_vattr.va_size = tsize; 764 vnode_pager_setsize(vp, np->n_size); 765 } 766 return (error); 767 } 768 769 /* 770 * Do an nfs setattr rpc. 771 */ 772 static int 773 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 774 struct ucred *cred, struct thread *td) 775 { 776 struct nfsv2_sattr *sp; 777 struct nfsnode *np = VTONFS(vp); 778 caddr_t cp; 779 int32_t t1, t2; 780 caddr_t bpos, dpos, cp2; 781 u_int32_t *tl; 782 int error = 0, wccflag = NFSV3_WCCRATTR; 783 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 784 int v3 = NFS_ISV3(vp); 785 786 nfsstats.rpccnt[NFSPROC_SETATTR]++; 787 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 788 nfsm_fhtom(vp, v3); 789 if (v3) { 790 nfsm_v3attrbuild(vap, TRUE); 791 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 792 *tl = nfs_false; 793 } else { 794 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 795 if (vap->va_mode == (mode_t)VNOVAL) 796 sp->sa_mode = nfs_xdrneg1; 797 else 798 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 799 if (vap->va_uid == (uid_t)VNOVAL) 800 sp->sa_uid = nfs_xdrneg1; 801 else 802 sp->sa_uid = txdr_unsigned(vap->va_uid); 803 if (vap->va_gid == (gid_t)VNOVAL) 804 sp->sa_gid = nfs_xdrneg1; 805 else 806 sp->sa_gid = txdr_unsigned(vap->va_gid); 807 sp->sa_size = txdr_unsigned(vap->va_size); 808 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 809 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 810 } 811 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 812 if (v3) { 813 np->n_modestamp = 0; 814 nfsm_wcc_data(vp, wccflag); 815 } else 816 nfsm_loadattr(vp, (struct vattr *)0); 817 m_freem(mrep); 818 nfsmout: 819 return (error); 820 } 821 822 /* 823 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 824 * nfs_lookup() until all remaining new api calls are implemented. 825 * 826 * Resolve a namecache entry. This function is passed a locked ncp and 827 * must call cache_setvp() on it as appropriate to resolve the entry. 828 */ 829 static int 830 nfs_nresolve(struct vop_nresolve_args *ap) 831 { 832 struct thread *td = curthread; 833 struct namecache *ncp; 834 struct ucred *cred; 835 struct nfsnode *np; 836 struct vnode *dvp; 837 struct vnode *nvp; 838 nfsfh_t *fhp; 839 int attrflag; 840 int fhsize; 841 int error; 842 int len; 843 int v3; 844 /******NFSM MACROS********/ 845 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 846 caddr_t bpos, dpos, cp, cp2; 847 u_int32_t *tl; 848 int32_t t1, t2; 849 850 cred = ap->a_cred; 851 ncp = ap->a_ncp; 852 853 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp); 854 dvp = ncp->nc_parent->nc_vp; 855 if ((error = vget(dvp, LK_SHARED)) != 0) 856 return (error); 857 858 nvp = NULL; 859 v3 = NFS_ISV3(dvp); 860 nfsstats.lookupcache_misses++; 861 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 862 len = ncp->nc_nlen; 863 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 864 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 865 nfsm_fhtom(dvp, v3); 866 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 867 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 868 if (error) { 869 /* 870 * Cache negatve lookups to reduce NFS traffic, but use 871 * a fast timeout. Otherwise use a timeout of 1 tick. 872 * XXX we should add a namecache flag for no-caching 873 * to uncache the negative hit as soon as possible, but 874 * we cannot simply destroy the entry because it is used 875 * as a placeholder by the caller. 876 */ 877 if (error == ENOENT) { 878 int nticks; 879 880 if (nfsneg_cache_timeout) 881 nticks = nfsneg_cache_timeout * hz; 882 else 883 nticks = 1; 884 cache_setvp(ncp, NULL); 885 cache_settimeout(ncp, nticks); 886 } 887 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 888 m_freem(mrep); 889 goto nfsmout; 890 } 891 892 /* 893 * Success, get the file handle, do various checks, and load 894 * post-operation data from the reply packet. Theoretically 895 * we should never be looking up "." so, theoretically, we 896 * should never get the same file handle as our directory. But 897 * we check anyway. XXX 898 * 899 * Note that no timeout is set for the positive cache hit. We 900 * assume, theoretically, that ESTALE returns will be dealt with 901 * properly to handle NFS races and in anycase we cannot depend 902 * on a timeout to deal with NFS open/create/excl issues so instead 903 * of a bad hack here the rest of the NFS client code needs to do 904 * the right thing. 905 */ 906 nfsm_getfh(fhp, fhsize, v3); 907 908 np = VTONFS(dvp); 909 if (NFS_CMPFH(np, fhp, fhsize)) { 910 vref(dvp); 911 nvp = dvp; 912 } else { 913 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 914 if (error) { 915 m_freem(mrep); 916 vput(dvp); 917 return (error); 918 } 919 nvp = NFSTOV(np); 920 } 921 if (v3) { 922 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 923 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 924 } else { 925 nfsm_loadattr(nvp, NULL); 926 } 927 cache_setvp(ncp, nvp); 928 m_freem(mrep); 929 nfsmout: 930 vput(dvp); 931 if (nvp) { 932 if (nvp == dvp) 933 vrele(nvp); 934 else 935 vput(nvp); 936 } 937 return (error); 938 } 939 940 /* 941 * 'cached' nfs directory lookup 942 * 943 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 944 * 945 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 946 * struct vnode **a_vpp, struct componentname *a_cnp) 947 */ 948 static int 949 nfs_lookup(struct vop_old_lookup_args *ap) 950 { 951 struct componentname *cnp = ap->a_cnp; 952 struct vnode *dvp = ap->a_dvp; 953 struct vnode **vpp = ap->a_vpp; 954 int flags = cnp->cn_flags; 955 struct vnode *newvp; 956 u_int32_t *tl; 957 caddr_t cp; 958 int32_t t1, t2; 959 struct nfsmount *nmp; 960 caddr_t bpos, dpos, cp2; 961 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 962 long len; 963 nfsfh_t *fhp; 964 struct nfsnode *np; 965 int lockparent, wantparent, error = 0, attrflag, fhsize; 966 int v3 = NFS_ISV3(dvp); 967 968 /* 969 * Read-only mount check and directory check. 970 */ 971 *vpp = NULLVP; 972 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 973 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 974 return (EROFS); 975 976 if (dvp->v_type != VDIR) 977 return (ENOTDIR); 978 979 /* 980 * Look it up in the cache. Note that ENOENT is only returned if we 981 * previously entered a negative hit (see later on). The additional 982 * nfsneg_cache_timeout check causes previously cached results to 983 * be instantly ignored if the negative caching is turned off. 984 */ 985 lockparent = flags & CNP_LOCKPARENT; 986 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 987 nmp = VFSTONFS(dvp->v_mount); 988 np = VTONFS(dvp); 989 990 /* 991 * Go to the wire. 992 */ 993 error = 0; 994 newvp = NULLVP; 995 nfsstats.lookupcache_misses++; 996 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 997 len = cnp->cn_namelen; 998 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 999 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1000 nfsm_fhtom(dvp, v3); 1001 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1002 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 1003 if (error) { 1004 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1005 m_freem(mrep); 1006 goto nfsmout; 1007 } 1008 nfsm_getfh(fhp, fhsize, v3); 1009 1010 /* 1011 * Handle RENAME case... 1012 */ 1013 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1014 if (NFS_CMPFH(np, fhp, fhsize)) { 1015 m_freem(mrep); 1016 return (EISDIR); 1017 } 1018 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1019 if (error) { 1020 m_freem(mrep); 1021 return (error); 1022 } 1023 newvp = NFSTOV(np); 1024 if (v3) { 1025 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1026 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1027 } else 1028 nfsm_loadattr(newvp, (struct vattr *)0); 1029 *vpp = newvp; 1030 m_freem(mrep); 1031 if (!lockparent) { 1032 VOP_UNLOCK(dvp, 0); 1033 cnp->cn_flags |= CNP_PDIRUNLOCK; 1034 } 1035 return (0); 1036 } 1037 1038 if (flags & CNP_ISDOTDOT) { 1039 VOP_UNLOCK(dvp, 0); 1040 cnp->cn_flags |= CNP_PDIRUNLOCK; 1041 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1042 if (error) { 1043 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1044 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1045 return (error); /* NOTE: return error from nget */ 1046 } 1047 newvp = NFSTOV(np); 1048 if (lockparent) { 1049 error = vn_lock(dvp, LK_EXCLUSIVE); 1050 if (error) { 1051 vput(newvp); 1052 return (error); 1053 } 1054 cnp->cn_flags |= CNP_PDIRUNLOCK; 1055 } 1056 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1057 vref(dvp); 1058 newvp = dvp; 1059 } else { 1060 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1061 if (error) { 1062 m_freem(mrep); 1063 return (error); 1064 } 1065 if (!lockparent) { 1066 VOP_UNLOCK(dvp, 0); 1067 cnp->cn_flags |= CNP_PDIRUNLOCK; 1068 } 1069 newvp = NFSTOV(np); 1070 } 1071 if (v3) { 1072 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1073 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1074 } else 1075 nfsm_loadattr(newvp, (struct vattr *)0); 1076 #if 0 1077 /* XXX MOVE TO nfs_nremove() */ 1078 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1079 cnp->cn_nameiop != NAMEI_DELETE) { 1080 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1081 } 1082 #endif 1083 *vpp = newvp; 1084 m_freem(mrep); 1085 nfsmout: 1086 if (error) { 1087 if (newvp != NULLVP) { 1088 vrele(newvp); 1089 *vpp = NULLVP; 1090 } 1091 if ((cnp->cn_nameiop == NAMEI_CREATE || 1092 cnp->cn_nameiop == NAMEI_RENAME) && 1093 error == ENOENT) { 1094 if (!lockparent) { 1095 VOP_UNLOCK(dvp, 0); 1096 cnp->cn_flags |= CNP_PDIRUNLOCK; 1097 } 1098 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1099 error = EROFS; 1100 else 1101 error = EJUSTRETURN; 1102 } 1103 } 1104 return (error); 1105 } 1106 1107 /* 1108 * nfs read call. 1109 * Just call nfs_bioread() to do the work. 1110 * 1111 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1112 * struct ucred *a_cred) 1113 */ 1114 static int 1115 nfs_read(struct vop_read_args *ap) 1116 { 1117 struct vnode *vp = ap->a_vp; 1118 1119 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1120 switch (vp->v_type) { 1121 case VREG: 1122 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1123 case VDIR: 1124 return (EISDIR); 1125 default: 1126 return EOPNOTSUPP; 1127 } 1128 } 1129 1130 /* 1131 * nfs readlink call 1132 * 1133 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1134 */ 1135 static int 1136 nfs_readlink(struct vop_readlink_args *ap) 1137 { 1138 struct vnode *vp = ap->a_vp; 1139 1140 if (vp->v_type != VLNK) 1141 return (EINVAL); 1142 return (nfs_bioread(vp, ap->a_uio, 0)); 1143 } 1144 1145 /* 1146 * Do a readlink rpc. 1147 * Called by nfs_doio() from below the buffer cache. 1148 */ 1149 int 1150 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1151 { 1152 u_int32_t *tl; 1153 caddr_t cp; 1154 int32_t t1, t2; 1155 caddr_t bpos, dpos, cp2; 1156 int error = 0, len, attrflag; 1157 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1158 int v3 = NFS_ISV3(vp); 1159 1160 nfsstats.rpccnt[NFSPROC_READLINK]++; 1161 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1162 nfsm_fhtom(vp, v3); 1163 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1164 if (v3) 1165 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1166 if (!error) { 1167 nfsm_strsiz(len, NFS_MAXPATHLEN); 1168 if (len == NFS_MAXPATHLEN) { 1169 struct nfsnode *np = VTONFS(vp); 1170 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1171 len = np->n_size; 1172 } 1173 nfsm_mtouio(uiop, len); 1174 } 1175 m_freem(mrep); 1176 nfsmout: 1177 return (error); 1178 } 1179 1180 /* 1181 * nfs read rpc call 1182 * Ditto above 1183 */ 1184 int 1185 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1186 { 1187 u_int32_t *tl; 1188 caddr_t cp; 1189 int32_t t1, t2; 1190 caddr_t bpos, dpos, cp2; 1191 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1192 struct nfsmount *nmp; 1193 int error = 0, len, retlen, tsiz, eof, attrflag; 1194 int v3 = NFS_ISV3(vp); 1195 1196 #ifndef nolint 1197 eof = 0; 1198 #endif 1199 nmp = VFSTONFS(vp->v_mount); 1200 tsiz = uiop->uio_resid; 1201 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1202 return (EFBIG); 1203 while (tsiz > 0) { 1204 nfsstats.rpccnt[NFSPROC_READ]++; 1205 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1206 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1207 nfsm_fhtom(vp, v3); 1208 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1209 if (v3) { 1210 txdr_hyper(uiop->uio_offset, tl); 1211 *(tl + 2) = txdr_unsigned(len); 1212 } else { 1213 *tl++ = txdr_unsigned(uiop->uio_offset); 1214 *tl++ = txdr_unsigned(len); 1215 *tl = 0; 1216 } 1217 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1218 if (v3) { 1219 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1220 if (error) { 1221 m_freem(mrep); 1222 goto nfsmout; 1223 } 1224 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1225 eof = fxdr_unsigned(int, *(tl + 1)); 1226 } else 1227 nfsm_loadattr(vp, (struct vattr *)0); 1228 nfsm_strsiz(retlen, nmp->nm_rsize); 1229 nfsm_mtouio(uiop, retlen); 1230 m_freem(mrep); 1231 tsiz -= retlen; 1232 if (v3) { 1233 if (eof || retlen == 0) { 1234 tsiz = 0; 1235 } 1236 } else if (retlen < len) { 1237 tsiz = 0; 1238 } 1239 } 1240 nfsmout: 1241 return (error); 1242 } 1243 1244 /* 1245 * nfs write call 1246 */ 1247 int 1248 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1249 { 1250 u_int32_t *tl; 1251 caddr_t cp; 1252 int32_t t1, t2, backup; 1253 caddr_t bpos, dpos, cp2; 1254 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1255 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1256 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1257 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1258 1259 #ifndef DIAGNOSTIC 1260 if (uiop->uio_iovcnt != 1) 1261 panic("nfs: writerpc iovcnt > 1"); 1262 #endif 1263 *must_commit = 0; 1264 tsiz = uiop->uio_resid; 1265 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1266 return (EFBIG); 1267 while (tsiz > 0) { 1268 nfsstats.rpccnt[NFSPROC_WRITE]++; 1269 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1270 nfsm_reqhead(vp, NFSPROC_WRITE, 1271 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1272 nfsm_fhtom(vp, v3); 1273 if (v3) { 1274 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1275 txdr_hyper(uiop->uio_offset, tl); 1276 tl += 2; 1277 *tl++ = txdr_unsigned(len); 1278 *tl++ = txdr_unsigned(*iomode); 1279 *tl = txdr_unsigned(len); 1280 } else { 1281 u_int32_t x; 1282 1283 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1284 /* Set both "begin" and "current" to non-garbage. */ 1285 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1286 *tl++ = x; /* "begin offset" */ 1287 *tl++ = x; /* "current offset" */ 1288 x = txdr_unsigned(len); 1289 *tl++ = x; /* total to this offset */ 1290 *tl = x; /* size of this write */ 1291 } 1292 nfsm_uiotom(uiop, len); 1293 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1294 if (v3) { 1295 /* 1296 * The write RPC returns a before and after mtime. The 1297 * nfsm_wcc_data() macro checks the before n_mtime 1298 * against the before time and stores the after time 1299 * in the nfsnode's cached vattr and n_mtime field. 1300 * The NRMODIFIED bit will be set if the before 1301 * time did not match the original mtime. 1302 */ 1303 wccflag = NFSV3_WCCCHK; 1304 nfsm_wcc_data(vp, wccflag); 1305 if (!error) { 1306 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1307 + NFSX_V3WRITEVERF); 1308 rlen = fxdr_unsigned(int, *tl++); 1309 if (rlen == 0) { 1310 error = NFSERR_IO; 1311 m_freem(mrep); 1312 break; 1313 } else if (rlen < len) { 1314 backup = len - rlen; 1315 uiop->uio_iov->iov_base -= backup; 1316 uiop->uio_iov->iov_len += backup; 1317 uiop->uio_offset -= backup; 1318 uiop->uio_resid += backup; 1319 len = rlen; 1320 } 1321 commit = fxdr_unsigned(int, *tl++); 1322 1323 /* 1324 * Return the lowest committment level 1325 * obtained by any of the RPCs. 1326 */ 1327 if (committed == NFSV3WRITE_FILESYNC) 1328 committed = commit; 1329 else if (committed == NFSV3WRITE_DATASYNC && 1330 commit == NFSV3WRITE_UNSTABLE) 1331 committed = commit; 1332 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1333 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1334 NFSX_V3WRITEVERF); 1335 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1336 } else if (bcmp((caddr_t)tl, 1337 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1338 *must_commit = 1; 1339 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1340 NFSX_V3WRITEVERF); 1341 } 1342 } 1343 } else { 1344 nfsm_loadattr(vp, (struct vattr *)0); 1345 } 1346 m_freem(mrep); 1347 if (error) 1348 break; 1349 tsiz -= len; 1350 } 1351 nfsmout: 1352 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1353 committed = NFSV3WRITE_FILESYNC; 1354 *iomode = committed; 1355 if (error) 1356 uiop->uio_resid = tsiz; 1357 return (error); 1358 } 1359 1360 /* 1361 * nfs mknod rpc 1362 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1363 * mode set to specify the file type and the size field for rdev. 1364 */ 1365 static int 1366 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1367 struct vattr *vap) 1368 { 1369 struct nfsv2_sattr *sp; 1370 u_int32_t *tl; 1371 caddr_t cp; 1372 int32_t t1, t2; 1373 struct vnode *newvp = (struct vnode *)0; 1374 struct nfsnode *np = (struct nfsnode *)0; 1375 struct vattr vattr; 1376 char *cp2; 1377 caddr_t bpos, dpos; 1378 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1379 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1380 u_int32_t rdev; 1381 int v3 = NFS_ISV3(dvp); 1382 1383 if (vap->va_type == VCHR || vap->va_type == VBLK) 1384 rdev = txdr_unsigned(vap->va_rdev); 1385 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1386 rdev = nfs_xdrneg1; 1387 else { 1388 return (EOPNOTSUPP); 1389 } 1390 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1391 return (error); 1392 } 1393 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1394 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1395 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1396 nfsm_fhtom(dvp, v3); 1397 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1398 if (v3) { 1399 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1400 *tl++ = vtonfsv3_type(vap->va_type); 1401 nfsm_v3attrbuild(vap, FALSE); 1402 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1403 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1404 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1405 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1406 } 1407 } else { 1408 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1409 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1410 sp->sa_uid = nfs_xdrneg1; 1411 sp->sa_gid = nfs_xdrneg1; 1412 sp->sa_size = rdev; 1413 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1414 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1415 } 1416 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1417 if (!error) { 1418 nfsm_mtofh(dvp, newvp, v3, gotvp); 1419 if (!gotvp) { 1420 if (newvp) { 1421 vput(newvp); 1422 newvp = (struct vnode *)0; 1423 } 1424 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1425 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1426 if (!error) 1427 newvp = NFSTOV(np); 1428 } 1429 } 1430 if (v3) 1431 nfsm_wcc_data(dvp, wccflag); 1432 m_freem(mrep); 1433 nfsmout: 1434 if (error) { 1435 if (newvp) 1436 vput(newvp); 1437 } else { 1438 *vpp = newvp; 1439 } 1440 VTONFS(dvp)->n_flag |= NLMODIFIED; 1441 if (!wccflag) 1442 VTONFS(dvp)->n_attrstamp = 0; 1443 return (error); 1444 } 1445 1446 /* 1447 * nfs mknod vop 1448 * just call nfs_mknodrpc() to do the work. 1449 * 1450 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1451 * struct componentname *a_cnp, struct vattr *a_vap) 1452 */ 1453 /* ARGSUSED */ 1454 static int 1455 nfs_mknod(struct vop_old_mknod_args *ap) 1456 { 1457 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1458 } 1459 1460 static u_long create_verf; 1461 /* 1462 * nfs file create call 1463 * 1464 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1465 * struct componentname *a_cnp, struct vattr *a_vap) 1466 */ 1467 static int 1468 nfs_create(struct vop_old_create_args *ap) 1469 { 1470 struct vnode *dvp = ap->a_dvp; 1471 struct vattr *vap = ap->a_vap; 1472 struct componentname *cnp = ap->a_cnp; 1473 struct nfsv2_sattr *sp; 1474 u_int32_t *tl; 1475 caddr_t cp; 1476 int32_t t1, t2; 1477 struct nfsnode *np = (struct nfsnode *)0; 1478 struct vnode *newvp = (struct vnode *)0; 1479 caddr_t bpos, dpos, cp2; 1480 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1481 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1482 struct vattr vattr; 1483 int v3 = NFS_ISV3(dvp); 1484 1485 /* 1486 * Oops, not for me.. 1487 */ 1488 if (vap->va_type == VSOCK) 1489 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1490 1491 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1492 return (error); 1493 } 1494 if (vap->va_vaflags & VA_EXCLUSIVE) 1495 fmode |= O_EXCL; 1496 again: 1497 nfsstats.rpccnt[NFSPROC_CREATE]++; 1498 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1499 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1500 nfsm_fhtom(dvp, v3); 1501 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1502 if (v3) { 1503 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1504 if (fmode & O_EXCL) { 1505 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1506 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1507 #ifdef INET 1508 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1509 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1510 else 1511 #endif 1512 *tl++ = create_verf; 1513 *tl = ++create_verf; 1514 } else { 1515 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1516 nfsm_v3attrbuild(vap, FALSE); 1517 } 1518 } else { 1519 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1520 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1521 sp->sa_uid = nfs_xdrneg1; 1522 sp->sa_gid = nfs_xdrneg1; 1523 sp->sa_size = 0; 1524 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1525 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1526 } 1527 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1528 if (!error) { 1529 nfsm_mtofh(dvp, newvp, v3, gotvp); 1530 if (!gotvp) { 1531 if (newvp) { 1532 vput(newvp); 1533 newvp = (struct vnode *)0; 1534 } 1535 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1536 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1537 if (!error) 1538 newvp = NFSTOV(np); 1539 } 1540 } 1541 if (v3) 1542 nfsm_wcc_data(dvp, wccflag); 1543 m_freem(mrep); 1544 nfsmout: 1545 if (error) { 1546 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1547 fmode &= ~O_EXCL; 1548 goto again; 1549 } 1550 if (newvp) 1551 vput(newvp); 1552 } else if (v3 && (fmode & O_EXCL)) { 1553 /* 1554 * We are normally called with only a partially initialized 1555 * VAP. Since the NFSv3 spec says that server may use the 1556 * file attributes to store the verifier, the spec requires 1557 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1558 * in atime, but we can't really assume that all servers will 1559 * so we ensure that our SETATTR sets both atime and mtime. 1560 */ 1561 if (vap->va_mtime.tv_sec == VNOVAL) 1562 vfs_timestamp(&vap->va_mtime); 1563 if (vap->va_atime.tv_sec == VNOVAL) 1564 vap->va_atime = vap->va_mtime; 1565 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1566 } 1567 if (!error) { 1568 /* 1569 * The new np may have enough info for access 1570 * checks, make sure rucred and wucred are 1571 * initialized for read and write rpc's. 1572 */ 1573 np = VTONFS(newvp); 1574 if (np->n_rucred == NULL) 1575 np->n_rucred = crhold(cnp->cn_cred); 1576 if (np->n_wucred == NULL) 1577 np->n_wucred = crhold(cnp->cn_cred); 1578 *ap->a_vpp = newvp; 1579 } 1580 VTONFS(dvp)->n_flag |= NLMODIFIED; 1581 if (!wccflag) 1582 VTONFS(dvp)->n_attrstamp = 0; 1583 return (error); 1584 } 1585 1586 /* 1587 * nfs file remove call 1588 * To try and make nfs semantics closer to ufs semantics, a file that has 1589 * other processes using the vnode is renamed instead of removed and then 1590 * removed later on the last close. 1591 * - If v_usecount > 1 1592 * If a rename is not already in the works 1593 * call nfs_sillyrename() to set it up 1594 * else 1595 * do the remove rpc 1596 * 1597 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 1598 * struct vnode *a_vp, struct componentname *a_cnp) 1599 */ 1600 static int 1601 nfs_remove(struct vop_old_remove_args *ap) 1602 { 1603 struct vnode *vp = ap->a_vp; 1604 struct vnode *dvp = ap->a_dvp; 1605 struct componentname *cnp = ap->a_cnp; 1606 struct nfsnode *np = VTONFS(vp); 1607 int error = 0; 1608 struct vattr vattr; 1609 1610 #ifndef DIAGNOSTIC 1611 if (vp->v_usecount < 1) 1612 panic("nfs_remove: bad v_usecount"); 1613 #endif 1614 if (vp->v_type == VDIR) 1615 error = EPERM; 1616 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1617 VOP_GETATTR(vp, &vattr) == 0 && 1618 vattr.va_nlink > 1)) { 1619 /* 1620 * throw away biocache buffers, mainly to avoid 1621 * unnecessary delayed writes later. 1622 */ 1623 error = nfs_vinvalbuf(vp, 0, 1); 1624 /* Do the rpc */ 1625 if (error != EINTR) 1626 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1627 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1628 /* 1629 * Kludge City: If the first reply to the remove rpc is lost.. 1630 * the reply to the retransmitted request will be ENOENT 1631 * since the file was in fact removed 1632 * Therefore, we cheat and return success. 1633 */ 1634 if (error == ENOENT) 1635 error = 0; 1636 } else if (!np->n_sillyrename) { 1637 error = nfs_sillyrename(dvp, vp, cnp); 1638 } 1639 np->n_attrstamp = 0; 1640 return (error); 1641 } 1642 1643 /* 1644 * nfs file remove rpc called from nfs_inactive 1645 */ 1646 int 1647 nfs_removeit(struct sillyrename *sp) 1648 { 1649 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1650 sp->s_cred, NULL)); 1651 } 1652 1653 /* 1654 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1655 */ 1656 static int 1657 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1658 struct ucred *cred, struct thread *td) 1659 { 1660 u_int32_t *tl; 1661 caddr_t cp; 1662 int32_t t1, t2; 1663 caddr_t bpos, dpos, cp2; 1664 int error = 0, wccflag = NFSV3_WCCRATTR; 1665 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1666 int v3 = NFS_ISV3(dvp); 1667 1668 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1669 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1670 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1671 nfsm_fhtom(dvp, v3); 1672 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1673 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1674 if (v3) 1675 nfsm_wcc_data(dvp, wccflag); 1676 m_freem(mrep); 1677 nfsmout: 1678 VTONFS(dvp)->n_flag |= NLMODIFIED; 1679 if (!wccflag) 1680 VTONFS(dvp)->n_attrstamp = 0; 1681 return (error); 1682 } 1683 1684 /* 1685 * nfs file rename call 1686 * 1687 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1688 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1689 * struct vnode *a_tvp, struct componentname *a_tcnp) 1690 */ 1691 static int 1692 nfs_rename(struct vop_old_rename_args *ap) 1693 { 1694 struct vnode *fvp = ap->a_fvp; 1695 struct vnode *tvp = ap->a_tvp; 1696 struct vnode *fdvp = ap->a_fdvp; 1697 struct vnode *tdvp = ap->a_tdvp; 1698 struct componentname *tcnp = ap->a_tcnp; 1699 struct componentname *fcnp = ap->a_fcnp; 1700 int error; 1701 1702 /* Check for cross-device rename */ 1703 if ((fvp->v_mount != tdvp->v_mount) || 1704 (tvp && (fvp->v_mount != tvp->v_mount))) { 1705 error = EXDEV; 1706 goto out; 1707 } 1708 1709 /* 1710 * We have to flush B_DELWRI data prior to renaming 1711 * the file. If we don't, the delayed-write buffers 1712 * can be flushed out later after the file has gone stale 1713 * under NFSV3. NFSV2 does not have this problem because 1714 * ( as far as I can tell ) it flushes dirty buffers more 1715 * often. 1716 */ 1717 1718 VOP_FSYNC(fvp, MNT_WAIT); 1719 if (tvp) 1720 VOP_FSYNC(tvp, MNT_WAIT); 1721 1722 /* 1723 * If the tvp exists and is in use, sillyrename it before doing the 1724 * rename of the new file over it. 1725 * 1726 * XXX Can't sillyrename a directory. 1727 * 1728 * We do not attempt to do any namecache purges in this old API 1729 * routine. The new API compat functions have access to the actual 1730 * namecache structures and will do it for us. 1731 */ 1732 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1733 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1734 vput(tvp); 1735 tvp = NULL; 1736 } else if (tvp) { 1737 ; 1738 } 1739 1740 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1741 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1742 tcnp->cn_td); 1743 1744 out: 1745 if (tdvp == tvp) 1746 vrele(tdvp); 1747 else 1748 vput(tdvp); 1749 if (tvp) 1750 vput(tvp); 1751 vrele(fdvp); 1752 vrele(fvp); 1753 /* 1754 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1755 */ 1756 if (error == ENOENT) 1757 error = 0; 1758 return (error); 1759 } 1760 1761 /* 1762 * nfs file rename rpc called from nfs_remove() above 1763 */ 1764 static int 1765 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1766 struct sillyrename *sp) 1767 { 1768 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1769 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1770 } 1771 1772 /* 1773 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1774 */ 1775 static int 1776 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1777 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1778 struct ucred *cred, struct thread *td) 1779 { 1780 u_int32_t *tl; 1781 caddr_t cp; 1782 int32_t t1, t2; 1783 caddr_t bpos, dpos, cp2; 1784 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1785 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1786 int v3 = NFS_ISV3(fdvp); 1787 1788 nfsstats.rpccnt[NFSPROC_RENAME]++; 1789 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1790 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1791 nfsm_rndup(tnamelen)); 1792 nfsm_fhtom(fdvp, v3); 1793 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1794 nfsm_fhtom(tdvp, v3); 1795 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1796 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1797 if (v3) { 1798 nfsm_wcc_data(fdvp, fwccflag); 1799 nfsm_wcc_data(tdvp, twccflag); 1800 } 1801 m_freem(mrep); 1802 nfsmout: 1803 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1804 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1805 if (!fwccflag) 1806 VTONFS(fdvp)->n_attrstamp = 0; 1807 if (!twccflag) 1808 VTONFS(tdvp)->n_attrstamp = 0; 1809 return (error); 1810 } 1811 1812 /* 1813 * nfs hard link create call 1814 * 1815 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1816 * struct componentname *a_cnp) 1817 */ 1818 static int 1819 nfs_link(struct vop_old_link_args *ap) 1820 { 1821 struct vnode *vp = ap->a_vp; 1822 struct vnode *tdvp = ap->a_tdvp; 1823 struct componentname *cnp = ap->a_cnp; 1824 u_int32_t *tl; 1825 caddr_t cp; 1826 int32_t t1, t2; 1827 caddr_t bpos, dpos, cp2; 1828 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1829 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1830 int v3; 1831 1832 if (vp->v_mount != tdvp->v_mount) { 1833 return (EXDEV); 1834 } 1835 1836 /* 1837 * Push all writes to the server, so that the attribute cache 1838 * doesn't get "out of sync" with the server. 1839 * XXX There should be a better way! 1840 */ 1841 VOP_FSYNC(vp, MNT_WAIT); 1842 1843 v3 = NFS_ISV3(vp); 1844 nfsstats.rpccnt[NFSPROC_LINK]++; 1845 nfsm_reqhead(vp, NFSPROC_LINK, 1846 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1847 nfsm_fhtom(vp, v3); 1848 nfsm_fhtom(tdvp, v3); 1849 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1850 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1851 if (v3) { 1852 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1853 nfsm_wcc_data(tdvp, wccflag); 1854 } 1855 m_freem(mrep); 1856 nfsmout: 1857 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1858 if (!attrflag) 1859 VTONFS(vp)->n_attrstamp = 0; 1860 if (!wccflag) 1861 VTONFS(tdvp)->n_attrstamp = 0; 1862 /* 1863 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1864 */ 1865 if (error == EEXIST) 1866 error = 0; 1867 return (error); 1868 } 1869 1870 /* 1871 * nfs symbolic link create call 1872 * 1873 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1874 * struct componentname *a_cnp, struct vattr *a_vap, 1875 * char *a_target) 1876 */ 1877 static int 1878 nfs_symlink(struct vop_old_symlink_args *ap) 1879 { 1880 struct vnode *dvp = ap->a_dvp; 1881 struct vattr *vap = ap->a_vap; 1882 struct componentname *cnp = ap->a_cnp; 1883 struct nfsv2_sattr *sp; 1884 u_int32_t *tl; 1885 caddr_t cp; 1886 int32_t t1, t2; 1887 caddr_t bpos, dpos, cp2; 1888 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1889 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1890 struct vnode *newvp = (struct vnode *)0; 1891 int v3 = NFS_ISV3(dvp); 1892 1893 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1894 slen = strlen(ap->a_target); 1895 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1896 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1897 nfsm_fhtom(dvp, v3); 1898 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1899 if (v3) { 1900 nfsm_v3attrbuild(vap, FALSE); 1901 } 1902 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1903 if (!v3) { 1904 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1905 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1906 sp->sa_uid = nfs_xdrneg1; 1907 sp->sa_gid = nfs_xdrneg1; 1908 sp->sa_size = nfs_xdrneg1; 1909 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1910 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1911 } 1912 1913 /* 1914 * Issue the NFS request and get the rpc response. 1915 * 1916 * Only NFSv3 responses returning an error of 0 actually return 1917 * a file handle that can be converted into newvp without having 1918 * to do an extra lookup rpc. 1919 */ 1920 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1921 if (v3) { 1922 if (error == 0) 1923 nfsm_mtofh(dvp, newvp, v3, gotvp); 1924 nfsm_wcc_data(dvp, wccflag); 1925 } 1926 1927 /* 1928 * out code jumps -> here, mrep is also freed. 1929 */ 1930 1931 m_freem(mrep); 1932 nfsmout: 1933 1934 /* 1935 * If we get an EEXIST error, silently convert it to no-error 1936 * in case of an NFS retry. 1937 */ 1938 if (error == EEXIST) 1939 error = 0; 1940 1941 /* 1942 * If we do not have (or no longer have) an error, and we could 1943 * not extract the newvp from the response due to the request being 1944 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1945 * to obtain a newvp to return. 1946 */ 1947 if (error == 0 && newvp == NULL) { 1948 struct nfsnode *np = NULL; 1949 1950 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1951 cnp->cn_cred, cnp->cn_td, &np); 1952 if (!error) 1953 newvp = NFSTOV(np); 1954 } 1955 if (error) { 1956 if (newvp) 1957 vput(newvp); 1958 } else { 1959 *ap->a_vpp = newvp; 1960 } 1961 VTONFS(dvp)->n_flag |= NLMODIFIED; 1962 if (!wccflag) 1963 VTONFS(dvp)->n_attrstamp = 0; 1964 return (error); 1965 } 1966 1967 /* 1968 * nfs make dir call 1969 * 1970 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 1971 * struct componentname *a_cnp, struct vattr *a_vap) 1972 */ 1973 static int 1974 nfs_mkdir(struct vop_old_mkdir_args *ap) 1975 { 1976 struct vnode *dvp = ap->a_dvp; 1977 struct vattr *vap = ap->a_vap; 1978 struct componentname *cnp = ap->a_cnp; 1979 struct nfsv2_sattr *sp; 1980 u_int32_t *tl; 1981 caddr_t cp; 1982 int32_t t1, t2; 1983 int len; 1984 struct nfsnode *np = (struct nfsnode *)0; 1985 struct vnode *newvp = (struct vnode *)0; 1986 caddr_t bpos, dpos, cp2; 1987 int error = 0, wccflag = NFSV3_WCCRATTR; 1988 int gotvp = 0; 1989 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1990 struct vattr vattr; 1991 int v3 = NFS_ISV3(dvp); 1992 1993 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) { 1994 return (error); 1995 } 1996 len = cnp->cn_namelen; 1997 nfsstats.rpccnt[NFSPROC_MKDIR]++; 1998 nfsm_reqhead(dvp, NFSPROC_MKDIR, 1999 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2000 nfsm_fhtom(dvp, v3); 2001 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2002 if (v3) { 2003 nfsm_v3attrbuild(vap, FALSE); 2004 } else { 2005 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2006 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2007 sp->sa_uid = nfs_xdrneg1; 2008 sp->sa_gid = nfs_xdrneg1; 2009 sp->sa_size = nfs_xdrneg1; 2010 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2011 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2012 } 2013 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2014 if (!error) 2015 nfsm_mtofh(dvp, newvp, v3, gotvp); 2016 if (v3) 2017 nfsm_wcc_data(dvp, wccflag); 2018 m_freem(mrep); 2019 nfsmout: 2020 VTONFS(dvp)->n_flag |= NLMODIFIED; 2021 if (!wccflag) 2022 VTONFS(dvp)->n_attrstamp = 0; 2023 /* 2024 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2025 * if we can succeed in looking up the directory. 2026 */ 2027 if (error == EEXIST || (!error && !gotvp)) { 2028 if (newvp) { 2029 vrele(newvp); 2030 newvp = (struct vnode *)0; 2031 } 2032 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2033 cnp->cn_td, &np); 2034 if (!error) { 2035 newvp = NFSTOV(np); 2036 if (newvp->v_type != VDIR) 2037 error = EEXIST; 2038 } 2039 } 2040 if (error) { 2041 if (newvp) 2042 vrele(newvp); 2043 } else 2044 *ap->a_vpp = newvp; 2045 return (error); 2046 } 2047 2048 /* 2049 * nfs remove directory call 2050 * 2051 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2052 * struct componentname *a_cnp) 2053 */ 2054 static int 2055 nfs_rmdir(struct vop_old_rmdir_args *ap) 2056 { 2057 struct vnode *vp = ap->a_vp; 2058 struct vnode *dvp = ap->a_dvp; 2059 struct componentname *cnp = ap->a_cnp; 2060 u_int32_t *tl; 2061 caddr_t cp; 2062 int32_t t1, t2; 2063 caddr_t bpos, dpos, cp2; 2064 int error = 0, wccflag = NFSV3_WCCRATTR; 2065 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2066 int v3 = NFS_ISV3(dvp); 2067 2068 if (dvp == vp) 2069 return (EINVAL); 2070 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2071 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2072 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2073 nfsm_fhtom(dvp, v3); 2074 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2075 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2076 if (v3) 2077 nfsm_wcc_data(dvp, wccflag); 2078 m_freem(mrep); 2079 nfsmout: 2080 VTONFS(dvp)->n_flag |= NLMODIFIED; 2081 if (!wccflag) 2082 VTONFS(dvp)->n_attrstamp = 0; 2083 /* 2084 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2085 */ 2086 if (error == ENOENT) 2087 error = 0; 2088 return (error); 2089 } 2090 2091 /* 2092 * nfs readdir call 2093 * 2094 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2095 */ 2096 static int 2097 nfs_readdir(struct vop_readdir_args *ap) 2098 { 2099 struct vnode *vp = ap->a_vp; 2100 struct nfsnode *np = VTONFS(vp); 2101 struct uio *uio = ap->a_uio; 2102 int tresid, error; 2103 struct vattr vattr; 2104 2105 if (vp->v_type != VDIR) 2106 return (EPERM); 2107 2108 /* 2109 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2110 * and then check that is still valid, or if this is an NQNFS mount 2111 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2112 * VOP_GETATTR() does not necessarily go to the wire. 2113 */ 2114 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2115 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2116 if (VOP_GETATTR(vp, &vattr) == 0 && 2117 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2118 ) { 2119 nfsstats.direofcache_hits++; 2120 return (0); 2121 } 2122 } 2123 2124 /* 2125 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2126 * own cache coherency checks so we do not have to. 2127 */ 2128 tresid = uio->uio_resid; 2129 error = nfs_bioread(vp, uio, 0); 2130 2131 if (!error && uio->uio_resid == tresid) 2132 nfsstats.direofcache_misses++; 2133 return (error); 2134 } 2135 2136 /* 2137 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc. 2138 * 2139 * Note that for directories, nfs_bioread maintains the underlying nfs-centric 2140 * offset/block and converts the nfs formatted directory entries for userland 2141 * consumption as well as deals with offsets into the middle of blocks. 2142 * nfs_doio only deals with logical blocks. In particular, uio_offset will 2143 * be block-bounded. It must convert to cookies for the actual RPC. 2144 */ 2145 int 2146 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2147 { 2148 int len, left; 2149 struct nfs_dirent *dp = NULL; 2150 u_int32_t *tl; 2151 caddr_t cp; 2152 int32_t t1, t2; 2153 nfsuint64 *cookiep; 2154 caddr_t bpos, dpos, cp2; 2155 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2156 nfsuint64 cookie; 2157 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2158 struct nfsnode *dnp = VTONFS(vp); 2159 u_quad_t fileno; 2160 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2161 int attrflag; 2162 int v3 = NFS_ISV3(vp); 2163 2164 #ifndef DIAGNOSTIC 2165 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2166 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2167 panic("nfs readdirrpc bad uio"); 2168 #endif 2169 2170 /* 2171 * If there is no cookie, assume directory was stale. 2172 */ 2173 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2174 if (cookiep) 2175 cookie = *cookiep; 2176 else 2177 return (NFSERR_BAD_COOKIE); 2178 /* 2179 * Loop around doing readdir rpc's of size nm_readdirsize 2180 * truncated to a multiple of DIRBLKSIZ. 2181 * The stopping criteria is EOF or buffer full. 2182 */ 2183 while (more_dirs && bigenough) { 2184 nfsstats.rpccnt[NFSPROC_READDIR]++; 2185 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2186 NFSX_READDIR(v3)); 2187 nfsm_fhtom(vp, v3); 2188 if (v3) { 2189 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2190 *tl++ = cookie.nfsuquad[0]; 2191 *tl++ = cookie.nfsuquad[1]; 2192 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2193 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2194 } else { 2195 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2196 *tl++ = cookie.nfsuquad[0]; 2197 } 2198 *tl = txdr_unsigned(nmp->nm_readdirsize); 2199 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2200 if (v3) { 2201 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2202 if (!error) { 2203 nfsm_dissect(tl, u_int32_t *, 2204 2 * NFSX_UNSIGNED); 2205 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2206 dnp->n_cookieverf.nfsuquad[1] = *tl; 2207 } else { 2208 m_freem(mrep); 2209 goto nfsmout; 2210 } 2211 } 2212 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2213 more_dirs = fxdr_unsigned(int, *tl); 2214 2215 /* loop thru the dir entries, converting them to std form */ 2216 while (more_dirs && bigenough) { 2217 if (v3) { 2218 nfsm_dissect(tl, u_int32_t *, 2219 3 * NFSX_UNSIGNED); 2220 fileno = fxdr_hyper(tl); 2221 len = fxdr_unsigned(int, *(tl + 2)); 2222 } else { 2223 nfsm_dissect(tl, u_int32_t *, 2224 2 * NFSX_UNSIGNED); 2225 fileno = fxdr_unsigned(u_quad_t, *tl++); 2226 len = fxdr_unsigned(int, *tl); 2227 } 2228 if (len <= 0 || len > NFS_MAXNAMLEN) { 2229 error = EBADRPC; 2230 m_freem(mrep); 2231 goto nfsmout; 2232 } 2233 2234 /* 2235 * len is the number of bytes in the path element 2236 * name, not including the \0 termination. 2237 * 2238 * tlen is the number of bytes w have to reserve for 2239 * the path element name. 2240 */ 2241 tlen = nfsm_rndup(len); 2242 if (tlen == len) 2243 tlen += 4; /* To ensure null termination */ 2244 2245 /* 2246 * If the entry would cross a DIRBLKSIZ boundary, 2247 * extend the previous nfs_dirent to cover the 2248 * remaining space. 2249 */ 2250 left = DIRBLKSIZ - blksiz; 2251 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2252 dp->nfs_reclen += left; 2253 uiop->uio_iov->iov_base += left; 2254 uiop->uio_iov->iov_len -= left; 2255 uiop->uio_offset += left; 2256 uiop->uio_resid -= left; 2257 blksiz = 0; 2258 } 2259 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2260 bigenough = 0; 2261 if (bigenough) { 2262 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2263 dp->nfs_ino = fileno; 2264 dp->nfs_namlen = len; 2265 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2266 dp->nfs_type = DT_UNKNOWN; 2267 blksiz += dp->nfs_reclen; 2268 if (blksiz == DIRBLKSIZ) 2269 blksiz = 0; 2270 uiop->uio_offset += sizeof(struct nfs_dirent); 2271 uiop->uio_resid -= sizeof(struct nfs_dirent); 2272 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2273 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2274 nfsm_mtouio(uiop, len); 2275 2276 /* 2277 * The uiop has advanced by nfs_dirent + len 2278 * but really needs to advance by 2279 * nfs_dirent + tlen 2280 */ 2281 cp = uiop->uio_iov->iov_base; 2282 tlen -= len; 2283 *cp = '\0'; /* null terminate */ 2284 uiop->uio_iov->iov_base += tlen; 2285 uiop->uio_iov->iov_len -= tlen; 2286 uiop->uio_offset += tlen; 2287 uiop->uio_resid -= tlen; 2288 } else { 2289 /* 2290 * NFS strings must be rounded up (nfsm_myouio 2291 * handled that in the bigenough case). 2292 */ 2293 nfsm_adv(nfsm_rndup(len)); 2294 } 2295 if (v3) { 2296 nfsm_dissect(tl, u_int32_t *, 2297 3 * NFSX_UNSIGNED); 2298 } else { 2299 nfsm_dissect(tl, u_int32_t *, 2300 2 * NFSX_UNSIGNED); 2301 } 2302 2303 /* 2304 * If we were able to accomodate the last entry, 2305 * get the cookie for the next one. Otherwise 2306 * hold-over the cookie for the one we were not 2307 * able to accomodate. 2308 */ 2309 if (bigenough) { 2310 cookie.nfsuquad[0] = *tl++; 2311 if (v3) 2312 cookie.nfsuquad[1] = *tl++; 2313 } else if (v3) { 2314 tl += 2; 2315 } else { 2316 tl++; 2317 } 2318 more_dirs = fxdr_unsigned(int, *tl); 2319 } 2320 /* 2321 * If at end of rpc data, get the eof boolean 2322 */ 2323 if (!more_dirs) { 2324 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2325 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2326 } 2327 m_freem(mrep); 2328 } 2329 /* 2330 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2331 * by increasing d_reclen for the last record. 2332 */ 2333 if (blksiz > 0) { 2334 left = DIRBLKSIZ - blksiz; 2335 dp->nfs_reclen += left; 2336 uiop->uio_iov->iov_base += left; 2337 uiop->uio_iov->iov_len -= left; 2338 uiop->uio_offset += left; 2339 uiop->uio_resid -= left; 2340 } 2341 2342 if (bigenough) { 2343 /* 2344 * We hit the end of the directory, update direofoffset. 2345 */ 2346 dnp->n_direofoffset = uiop->uio_offset; 2347 } else { 2348 /* 2349 * There is more to go, insert the link cookie so the 2350 * next block can be read. 2351 */ 2352 if (uiop->uio_resid > 0) 2353 printf("EEK! readdirrpc resid > 0\n"); 2354 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2355 *cookiep = cookie; 2356 } 2357 nfsmout: 2358 return (error); 2359 } 2360 2361 /* 2362 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2363 */ 2364 int 2365 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2366 { 2367 int len, left; 2368 struct nfs_dirent *dp; 2369 u_int32_t *tl; 2370 caddr_t cp; 2371 int32_t t1, t2; 2372 struct vnode *newvp; 2373 nfsuint64 *cookiep; 2374 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2375 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2376 nfsuint64 cookie; 2377 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2378 struct nfsnode *dnp = VTONFS(vp), *np; 2379 nfsfh_t *fhp; 2380 u_quad_t fileno; 2381 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2382 int attrflag, fhsize; 2383 struct namecache *ncp; 2384 struct namecache *dncp; 2385 struct nlcomponent nlc; 2386 2387 #ifndef nolint 2388 dp = NULL; 2389 #endif 2390 #ifndef DIAGNOSTIC 2391 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2392 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2393 panic("nfs readdirplusrpc bad uio"); 2394 #endif 2395 /* 2396 * Obtain the namecache record for the directory so we have something 2397 * to use as a basis for creating the entries. This function will 2398 * return a held (but not locked) ncp. The ncp may be disconnected 2399 * from the tree and cannot be used for upward traversals, and the 2400 * ncp may be unnamed. Note that other unrelated operations may 2401 * cause the ncp to be named at any time. 2402 */ 2403 dncp = cache_fromdvp(vp, NULL, 0); 2404 bzero(&nlc, sizeof(nlc)); 2405 newvp = NULLVP; 2406 2407 /* 2408 * If there is no cookie, assume directory was stale. 2409 */ 2410 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2411 if (cookiep) 2412 cookie = *cookiep; 2413 else 2414 return (NFSERR_BAD_COOKIE); 2415 /* 2416 * Loop around doing readdir rpc's of size nm_readdirsize 2417 * truncated to a multiple of DIRBLKSIZ. 2418 * The stopping criteria is EOF or buffer full. 2419 */ 2420 while (more_dirs && bigenough) { 2421 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2422 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2423 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2424 nfsm_fhtom(vp, 1); 2425 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2426 *tl++ = cookie.nfsuquad[0]; 2427 *tl++ = cookie.nfsuquad[1]; 2428 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2429 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2430 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2431 *tl = txdr_unsigned(nmp->nm_rsize); 2432 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2433 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2434 if (error) { 2435 m_freem(mrep); 2436 goto nfsmout; 2437 } 2438 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2439 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2440 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2441 more_dirs = fxdr_unsigned(int, *tl); 2442 2443 /* loop thru the dir entries, doctoring them to 4bsd form */ 2444 while (more_dirs && bigenough) { 2445 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2446 fileno = fxdr_hyper(tl); 2447 len = fxdr_unsigned(int, *(tl + 2)); 2448 if (len <= 0 || len > NFS_MAXNAMLEN) { 2449 error = EBADRPC; 2450 m_freem(mrep); 2451 goto nfsmout; 2452 } 2453 tlen = nfsm_rndup(len); 2454 if (tlen == len) 2455 tlen += 4; /* To ensure null termination*/ 2456 left = DIRBLKSIZ - blksiz; 2457 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2458 dp->nfs_reclen += left; 2459 uiop->uio_iov->iov_base += left; 2460 uiop->uio_iov->iov_len -= left; 2461 uiop->uio_offset += left; 2462 uiop->uio_resid -= left; 2463 blksiz = 0; 2464 } 2465 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2466 bigenough = 0; 2467 if (bigenough) { 2468 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2469 dp->nfs_ino = fileno; 2470 dp->nfs_namlen = len; 2471 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2472 dp->nfs_type = DT_UNKNOWN; 2473 blksiz += dp->nfs_reclen; 2474 if (blksiz == DIRBLKSIZ) 2475 blksiz = 0; 2476 uiop->uio_offset += sizeof(struct nfs_dirent); 2477 uiop->uio_resid -= sizeof(struct nfs_dirent); 2478 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2479 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2480 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2481 nlc.nlc_namelen = len; 2482 nfsm_mtouio(uiop, len); 2483 cp = uiop->uio_iov->iov_base; 2484 tlen -= len; 2485 *cp = '\0'; 2486 uiop->uio_iov->iov_base += tlen; 2487 uiop->uio_iov->iov_len -= tlen; 2488 uiop->uio_offset += tlen; 2489 uiop->uio_resid -= tlen; 2490 } else 2491 nfsm_adv(nfsm_rndup(len)); 2492 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2493 if (bigenough) { 2494 cookie.nfsuquad[0] = *tl++; 2495 cookie.nfsuquad[1] = *tl++; 2496 } else 2497 tl += 2; 2498 2499 /* 2500 * Since the attributes are before the file handle 2501 * (sigh), we must skip over the attributes and then 2502 * come back and get them. 2503 */ 2504 attrflag = fxdr_unsigned(int, *tl); 2505 if (attrflag) { 2506 dpossav1 = dpos; 2507 mdsav1 = md; 2508 nfsm_adv(NFSX_V3FATTR); 2509 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2510 doit = fxdr_unsigned(int, *tl); 2511 if (doit) { 2512 nfsm_getfh(fhp, fhsize, 1); 2513 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2514 vref(vp); 2515 newvp = vp; 2516 np = dnp; 2517 } else { 2518 error = nfs_nget(vp->v_mount, fhp, 2519 fhsize, &np); 2520 if (error) 2521 doit = 0; 2522 else 2523 newvp = NFSTOV(np); 2524 } 2525 } 2526 if (doit && bigenough) { 2527 dpossav2 = dpos; 2528 dpos = dpossav1; 2529 mdsav2 = md; 2530 md = mdsav1; 2531 nfsm_loadattr(newvp, (struct vattr *)0); 2532 dpos = dpossav2; 2533 md = mdsav2; 2534 dp->nfs_type = 2535 IFTODT(VTTOIF(np->n_vattr.va_type)); 2536 if (dncp) { 2537 printf("NFS/READDIRPLUS, ENTER %*.*s\n", 2538 nlc.nlc_namelen, nlc.nlc_namelen, 2539 nlc.nlc_nameptr); 2540 ncp = cache_nlookup(dncp, &nlc); 2541 cache_setunresolved(ncp); 2542 cache_setvp(ncp, newvp); 2543 cache_put(ncp); 2544 } else { 2545 printf("NFS/READDIRPLUS, UNABLE TO ENTER" 2546 " %*.*s\n", 2547 nlc.nlc_namelen, nlc.nlc_namelen, 2548 nlc.nlc_nameptr); 2549 } 2550 } 2551 } else { 2552 /* Just skip over the file handle */ 2553 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2554 i = fxdr_unsigned(int, *tl); 2555 nfsm_adv(nfsm_rndup(i)); 2556 } 2557 if (newvp != NULLVP) { 2558 if (newvp == vp) 2559 vrele(newvp); 2560 else 2561 vput(newvp); 2562 newvp = NULLVP; 2563 } 2564 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2565 more_dirs = fxdr_unsigned(int, *tl); 2566 } 2567 /* 2568 * If at end of rpc data, get the eof boolean 2569 */ 2570 if (!more_dirs) { 2571 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2572 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2573 } 2574 m_freem(mrep); 2575 } 2576 /* 2577 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2578 * by increasing d_reclen for the last record. 2579 */ 2580 if (blksiz > 0) { 2581 left = DIRBLKSIZ - blksiz; 2582 dp->nfs_reclen += left; 2583 uiop->uio_iov->iov_base += left; 2584 uiop->uio_iov->iov_len -= left; 2585 uiop->uio_offset += left; 2586 uiop->uio_resid -= left; 2587 } 2588 2589 /* 2590 * We are now either at the end of the directory or have filled the 2591 * block. 2592 */ 2593 if (bigenough) 2594 dnp->n_direofoffset = uiop->uio_offset; 2595 else { 2596 if (uiop->uio_resid > 0) 2597 printf("EEK! readdirplusrpc resid > 0\n"); 2598 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2599 *cookiep = cookie; 2600 } 2601 nfsmout: 2602 if (newvp != NULLVP) { 2603 if (newvp == vp) 2604 vrele(newvp); 2605 else 2606 vput(newvp); 2607 newvp = NULLVP; 2608 } 2609 if (dncp) 2610 cache_drop(dncp); 2611 return (error); 2612 } 2613 2614 /* 2615 * Silly rename. To make the NFS filesystem that is stateless look a little 2616 * more like the "ufs" a remove of an active vnode is translated to a rename 2617 * to a funny looking filename that is removed by nfs_inactive on the 2618 * nfsnode. There is the potential for another process on a different client 2619 * to create the same funny name between the nfs_lookitup() fails and the 2620 * nfs_rename() completes, but... 2621 */ 2622 static int 2623 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2624 { 2625 struct sillyrename *sp; 2626 struct nfsnode *np; 2627 int error; 2628 2629 /* 2630 * We previously purged dvp instead of vp. I don't know why, it 2631 * completely destroys performance. We can't do it anyway with the 2632 * new VFS API since we would be breaking the namecache topology. 2633 */ 2634 cache_purge(vp); /* XXX */ 2635 np = VTONFS(vp); 2636 #ifndef DIAGNOSTIC 2637 if (vp->v_type == VDIR) 2638 panic("nfs: sillyrename dir"); 2639 #endif 2640 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2641 M_NFSREQ, M_WAITOK); 2642 sp->s_cred = crdup(cnp->cn_cred); 2643 sp->s_dvp = dvp; 2644 vref(dvp); 2645 2646 /* Fudge together a funny name */ 2647 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2648 2649 /* Try lookitups until we get one that isn't there */ 2650 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2651 cnp->cn_td, (struct nfsnode **)0) == 0) { 2652 sp->s_name[4]++; 2653 if (sp->s_name[4] > 'z') { 2654 error = EINVAL; 2655 goto bad; 2656 } 2657 } 2658 error = nfs_renameit(dvp, cnp, sp); 2659 if (error) 2660 goto bad; 2661 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2662 cnp->cn_td, &np); 2663 np->n_sillyrename = sp; 2664 return (0); 2665 bad: 2666 vrele(sp->s_dvp); 2667 crfree(sp->s_cred); 2668 free((caddr_t)sp, M_NFSREQ); 2669 return (error); 2670 } 2671 2672 /* 2673 * Look up a file name and optionally either update the file handle or 2674 * allocate an nfsnode, depending on the value of npp. 2675 * npp == NULL --> just do the lookup 2676 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2677 * handled too 2678 * *npp != NULL --> update the file handle in the vnode 2679 */ 2680 static int 2681 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2682 struct thread *td, struct nfsnode **npp) 2683 { 2684 u_int32_t *tl; 2685 caddr_t cp; 2686 int32_t t1, t2; 2687 struct vnode *newvp = (struct vnode *)0; 2688 struct nfsnode *np, *dnp = VTONFS(dvp); 2689 caddr_t bpos, dpos, cp2; 2690 int error = 0, fhlen, attrflag; 2691 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2692 nfsfh_t *nfhp; 2693 int v3 = NFS_ISV3(dvp); 2694 2695 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2696 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2697 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2698 nfsm_fhtom(dvp, v3); 2699 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2700 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2701 if (npp && !error) { 2702 nfsm_getfh(nfhp, fhlen, v3); 2703 if (*npp) { 2704 np = *npp; 2705 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2706 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2707 np->n_fhp = &np->n_fh; 2708 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2709 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2710 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2711 np->n_fhsize = fhlen; 2712 newvp = NFSTOV(np); 2713 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2714 vref(dvp); 2715 newvp = dvp; 2716 } else { 2717 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2718 if (error) { 2719 m_freem(mrep); 2720 return (error); 2721 } 2722 newvp = NFSTOV(np); 2723 } 2724 if (v3) { 2725 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2726 if (!attrflag && *npp == NULL) { 2727 m_freem(mrep); 2728 if (newvp == dvp) 2729 vrele(newvp); 2730 else 2731 vput(newvp); 2732 return (ENOENT); 2733 } 2734 } else 2735 nfsm_loadattr(newvp, (struct vattr *)0); 2736 } 2737 m_freem(mrep); 2738 nfsmout: 2739 if (npp && *npp == NULL) { 2740 if (error) { 2741 if (newvp) { 2742 if (newvp == dvp) 2743 vrele(newvp); 2744 else 2745 vput(newvp); 2746 } 2747 } else 2748 *npp = np; 2749 } 2750 return (error); 2751 } 2752 2753 /* 2754 * Nfs Version 3 commit rpc 2755 */ 2756 int 2757 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2758 { 2759 caddr_t cp; 2760 u_int32_t *tl; 2761 int32_t t1, t2; 2762 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2763 caddr_t bpos, dpos, cp2; 2764 int error = 0, wccflag = NFSV3_WCCRATTR; 2765 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2766 2767 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2768 return (0); 2769 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2770 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2771 nfsm_fhtom(vp, 1); 2772 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2773 txdr_hyper(offset, tl); 2774 tl += 2; 2775 *tl = txdr_unsigned(cnt); 2776 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2777 nfsm_wcc_data(vp, wccflag); 2778 if (!error) { 2779 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2780 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2781 NFSX_V3WRITEVERF)) { 2782 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2783 NFSX_V3WRITEVERF); 2784 error = NFSERR_STALEWRITEVERF; 2785 } 2786 } 2787 m_freem(mrep); 2788 nfsmout: 2789 return (error); 2790 } 2791 2792 /* 2793 * Kludge City.. 2794 * - make nfs_bmap() essentially a no-op that does no translation 2795 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2796 * (Maybe I could use the process's page mapping, but I was concerned that 2797 * Kernel Write might not be enabled and also figured copyout() would do 2798 * a lot more work than bcopy() and also it currently happens in the 2799 * context of the swapper process (2). 2800 * 2801 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp, 2802 * off_t *a_doffsetp, int *a_runp, int *a_runb) 2803 */ 2804 static int 2805 nfs_bmap(struct vop_bmap_args *ap) 2806 { 2807 struct vnode *vp = ap->a_vp; 2808 2809 if (ap->a_vpp != NULL) 2810 *ap->a_vpp = vp; 2811 if (ap->a_doffsetp != NULL) 2812 *ap->a_doffsetp = ap->a_loffset; 2813 if (ap->a_runp != NULL) 2814 *ap->a_runp = 0; 2815 if (ap->a_runb != NULL) 2816 *ap->a_runb = 0; 2817 return (0); 2818 } 2819 2820 /* 2821 * Strategy routine. 2822 * 2823 * For async requests when nfsiod(s) are running, queue the request by 2824 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2825 * request. 2826 */ 2827 static int 2828 nfs_strategy(struct vop_strategy_args *ap) 2829 { 2830 struct bio *bio = ap->a_bio; 2831 struct bio *nbio; 2832 struct buf *bp = bio->bio_buf; 2833 struct thread *td; 2834 int error = 0; 2835 2836 KASSERT(bp->b_cmd != BUF_CMD_DONE, 2837 ("nfs_strategy: buffer %p unexpectedly marked done", bp)); 2838 KASSERT(BUF_REFCNT(bp) > 0, 2839 ("nfs_strategy: buffer %p not locked", bp)); 2840 2841 if (bp->b_flags & B_ASYNC) 2842 td = NULL; 2843 else 2844 td = curthread; /* XXX */ 2845 2846 /* 2847 * We probably don't need to push an nbio any more since no 2848 * block conversion is required due to the use of 64 bit byte 2849 * offsets, but do it anyway. 2850 */ 2851 nbio = push_bio(bio); 2852 nbio->bio_offset = bio->bio_offset; 2853 2854 /* 2855 * If the op is asynchronous and an i/o daemon is waiting 2856 * queue the request, wake it up and wait for completion 2857 * otherwise just do it ourselves. 2858 */ 2859 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td)) 2860 error = nfs_doio(ap->a_vp, nbio, td); 2861 return (error); 2862 } 2863 2864 /* 2865 * Mmap a file 2866 * 2867 * NB Currently unsupported. 2868 * 2869 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2870 * struct thread *a_td) 2871 */ 2872 /* ARGSUSED */ 2873 static int 2874 nfs_mmap(struct vop_mmap_args *ap) 2875 { 2876 return (EINVAL); 2877 } 2878 2879 /* 2880 * fsync vnode op. Just call nfs_flush() with commit == 1. 2881 * 2882 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp, 2883 * struct ucred * a_cred, int a_waitfor, struct thread *a_td) 2884 */ 2885 /* ARGSUSED */ 2886 static int 2887 nfs_fsync(struct vop_fsync_args *ap) 2888 { 2889 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1)); 2890 } 2891 2892 /* 2893 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 2894 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 2895 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 2896 * set the buffer contains data that has already been written to the server 2897 * and which now needs a commit RPC. 2898 * 2899 * If commit is 0 we only take one pass and only flush buffers containing new 2900 * dirty data. 2901 * 2902 * If commit is 1 we take two passes, issuing a commit RPC in the second 2903 * pass. 2904 * 2905 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 2906 * to completely flush all pending data. 2907 * 2908 * Note that the RB_SCAN code properly handles the case where the 2909 * callback might block and directly or indirectly (another thread) cause 2910 * the RB tree to change. 2911 */ 2912 2913 #ifndef NFS_COMMITBVECSIZ 2914 #define NFS_COMMITBVECSIZ 16 2915 #endif 2916 2917 struct nfs_flush_info { 2918 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 2919 struct thread *td; 2920 struct vnode *vp; 2921 int waitfor; 2922 int slpflag; 2923 int slptimeo; 2924 int loops; 2925 struct buf *bvary[NFS_COMMITBVECSIZ]; 2926 int bvsize; 2927 off_t beg_off; 2928 off_t end_off; 2929 }; 2930 2931 static int nfs_flush_bp(struct buf *bp, void *data); 2932 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 2933 2934 int 2935 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2936 { 2937 struct nfsnode *np = VTONFS(vp); 2938 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2939 struct nfs_flush_info info; 2940 int error; 2941 2942 bzero(&info, sizeof(info)); 2943 info.td = td; 2944 info.vp = vp; 2945 info.waitfor = waitfor; 2946 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 2947 info.loops = 0; 2948 2949 do { 2950 /* 2951 * Flush mode 2952 */ 2953 info.mode = NFI_FLUSHNEW; 2954 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2955 nfs_flush_bp, &info); 2956 2957 /* 2958 * Take a second pass if committing and no error occured. 2959 * Clean up any left over collection (whether an error 2960 * occurs or not). 2961 */ 2962 if (commit && error == 0) { 2963 info.mode = NFI_COMMIT; 2964 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2965 nfs_flush_bp, &info); 2966 if (info.bvsize) 2967 error = nfs_flush_docommit(&info, error); 2968 } 2969 2970 /* 2971 * Wait for pending I/O to complete before checking whether 2972 * any further dirty buffers exist. 2973 */ 2974 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) { 2975 vp->v_track_write.bk_waitflag = 1; 2976 error = tsleep(&vp->v_track_write, 2977 info.slpflag, "nfsfsync", info.slptimeo); 2978 if (error) { 2979 /* 2980 * We have to be able to break out if this 2981 * is an 'intr' mount. 2982 */ 2983 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2984 error = -EINTR; 2985 break; 2986 } 2987 2988 /* 2989 * Since we do not process pending signals, 2990 * once we get a PCATCH our tsleep() will no 2991 * longer sleep, switch to a fixed timeout 2992 * instead. 2993 */ 2994 if (info.slpflag == PCATCH) { 2995 info.slpflag = 0; 2996 info.slptimeo = 2 * hz; 2997 } 2998 error = 0; 2999 } 3000 } 3001 ++info.loops; 3002 /* 3003 * Loop if we are flushing synchronous as well as committing, 3004 * and dirty buffers are still present. Otherwise we might livelock. 3005 */ 3006 } while (waitfor == MNT_WAIT && commit && 3007 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 3008 3009 /* 3010 * The callbacks have to return a negative error to terminate the 3011 * RB scan. 3012 */ 3013 if (error < 0) 3014 error = -error; 3015 3016 /* 3017 * Deal with any error collection 3018 */ 3019 if (np->n_flag & NWRITEERR) { 3020 error = np->n_error; 3021 np->n_flag &= ~NWRITEERR; 3022 } 3023 return (error); 3024 } 3025 3026 3027 static 3028 int 3029 nfs_flush_bp(struct buf *bp, void *data) 3030 { 3031 struct nfs_flush_info *info = data; 3032 off_t toff; 3033 int error; 3034 3035 error = 0; 3036 switch(info->mode) { 3037 case NFI_FLUSHNEW: 3038 crit_enter(); 3039 if (info->loops && info->waitfor == MNT_WAIT) { 3040 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3041 if (error) { 3042 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 3043 if (info->slpflag & PCATCH) 3044 lkflags |= LK_PCATCH; 3045 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync", 3046 info->slptimeo); 3047 } 3048 } else { 3049 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3050 } 3051 if (error == 0) { 3052 KKASSERT(bp->b_vp == info->vp); 3053 3054 if ((bp->b_flags & B_DELWRI) == 0) 3055 panic("nfs_fsync: not dirty"); 3056 if (bp->b_flags & B_NEEDCOMMIT) { 3057 BUF_UNLOCK(bp); 3058 crit_exit(); 3059 break; 3060 } 3061 bremfree(bp); 3062 3063 bp->b_flags |= B_ASYNC; 3064 crit_exit(); 3065 bwrite(bp); 3066 } else { 3067 crit_exit(); 3068 error = 0; 3069 } 3070 break; 3071 case NFI_COMMIT: 3072 /* 3073 * Only process buffers in need of a commit which we can 3074 * immediately lock. This may prevent a buffer from being 3075 * committed, but the normal flush loop will block on the 3076 * same buffer so we shouldn't get into an endless loop. 3077 */ 3078 crit_enter(); 3079 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3080 (B_DELWRI | B_NEEDCOMMIT) || 3081 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 3082 crit_exit(); 3083 break; 3084 } 3085 3086 KKASSERT(bp->b_vp == info->vp); 3087 bremfree(bp); 3088 3089 /* 3090 * NOTE: storing the bp in the bvary[] basically sets 3091 * it up for a commit operation. 3092 * 3093 * We must call vfs_busy_pages() now so the commit operation 3094 * is interlocked with user modifications to memory mapped 3095 * pages. 3096 * 3097 * Note: to avoid loopback deadlocks, we do not 3098 * assign b_runningbufspace. 3099 */ 3100 bp->b_cmd = BUF_CMD_WRITE; 3101 vfs_busy_pages(bp->b_vp, bp); 3102 info->bvary[info->bvsize] = bp; 3103 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff; 3104 if (info->bvsize == 0 || toff < info->beg_off) 3105 info->beg_off = toff; 3106 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff); 3107 if (info->bvsize == 0 || toff > info->end_off) 3108 info->end_off = toff; 3109 ++info->bvsize; 3110 if (info->bvsize == NFS_COMMITBVECSIZ) { 3111 error = nfs_flush_docommit(info, 0); 3112 KKASSERT(info->bvsize == 0); 3113 } 3114 crit_exit(); 3115 } 3116 return (error); 3117 } 3118 3119 static 3120 int 3121 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3122 { 3123 struct vnode *vp; 3124 struct buf *bp; 3125 off_t bytes; 3126 int retv; 3127 int i; 3128 3129 vp = info->vp; 3130 3131 if (info->bvsize > 0) { 3132 /* 3133 * Commit data on the server, as required. Note that 3134 * nfs_commit will use the vnode's cred for the commit. 3135 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3136 */ 3137 bytes = info->end_off - info->beg_off; 3138 if (bytes > 0x40000000) 3139 bytes = 0x40000000; 3140 if (error) { 3141 retv = -error; 3142 } else { 3143 retv = nfs_commit(vp, info->beg_off, 3144 (int)bytes, info->td); 3145 if (retv == NFSERR_STALEWRITEVERF) 3146 nfs_clearcommit(vp->v_mount); 3147 } 3148 3149 /* 3150 * Now, either mark the blocks I/O done or mark the 3151 * blocks dirty, depending on whether the commit 3152 * succeeded. 3153 */ 3154 for (i = 0; i < info->bvsize; ++i) { 3155 bp = info->bvary[i]; 3156 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3157 if (retv) { 3158 /* 3159 * Error, leave B_DELWRI intact 3160 */ 3161 vfs_unbusy_pages(bp); 3162 bp->b_cmd = BUF_CMD_DONE; 3163 brelse(bp); 3164 } else { 3165 /* 3166 * Success, remove B_DELWRI ( bundirty() ). 3167 * 3168 * b_dirtyoff/b_dirtyend seem to be NFS 3169 * specific. We should probably move that 3170 * into bundirty(). XXX 3171 * 3172 * We are faking an I/O write, we have to 3173 * start the transaction in order to 3174 * immediately biodone() it. 3175 */ 3176 crit_enter(); 3177 bp->b_flags |= B_ASYNC; 3178 bundirty(bp); 3179 bp->b_flags &= ~B_ERROR; 3180 bp->b_dirtyoff = bp->b_dirtyend = 0; 3181 crit_exit(); 3182 biodone(&bp->b_bio1); 3183 } 3184 } 3185 info->bvsize = 0; 3186 } 3187 return (error); 3188 } 3189 3190 /* 3191 * NFS advisory byte-level locks. 3192 * Currently unsupported. 3193 * 3194 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3195 * int a_flags) 3196 */ 3197 static int 3198 nfs_advlock(struct vop_advlock_args *ap) 3199 { 3200 struct nfsnode *np = VTONFS(ap->a_vp); 3201 3202 /* 3203 * The following kludge is to allow diskless support to work 3204 * until a real NFS lockd is implemented. Basically, just pretend 3205 * that this is a local lock. 3206 */ 3207 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3208 } 3209 3210 /* 3211 * Print out the contents of an nfsnode. 3212 * 3213 * nfs_print(struct vnode *a_vp) 3214 */ 3215 static int 3216 nfs_print(struct vop_print_args *ap) 3217 { 3218 struct vnode *vp = ap->a_vp; 3219 struct nfsnode *np = VTONFS(vp); 3220 3221 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3222 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3223 if (vp->v_type == VFIFO) 3224 fifo_printinfo(vp); 3225 printf("\n"); 3226 return (0); 3227 } 3228 3229 /* 3230 * nfs special file access vnode op. 3231 * Essentially just get vattr and then imitate iaccess() since the device is 3232 * local to the client. 3233 * 3234 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3235 * struct thread *a_td) 3236 */ 3237 static int 3238 nfsspec_access(struct vop_access_args *ap) 3239 { 3240 struct vattr *vap; 3241 gid_t *gp; 3242 struct ucred *cred = ap->a_cred; 3243 struct vnode *vp = ap->a_vp; 3244 mode_t mode = ap->a_mode; 3245 struct vattr vattr; 3246 int i; 3247 int error; 3248 3249 /* 3250 * Disallow write attempts on filesystems mounted read-only; 3251 * unless the file is a socket, fifo, or a block or character 3252 * device resident on the filesystem. 3253 */ 3254 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3255 switch (vp->v_type) { 3256 case VREG: 3257 case VDIR: 3258 case VLNK: 3259 return (EROFS); 3260 default: 3261 break; 3262 } 3263 } 3264 /* 3265 * If you're the super-user, 3266 * you always get access. 3267 */ 3268 if (cred->cr_uid == 0) 3269 return (0); 3270 vap = &vattr; 3271 error = VOP_GETATTR(vp, vap); 3272 if (error) 3273 return (error); 3274 /* 3275 * Access check is based on only one of owner, group, public. 3276 * If not owner, then check group. If not a member of the 3277 * group, then check public access. 3278 */ 3279 if (cred->cr_uid != vap->va_uid) { 3280 mode >>= 3; 3281 gp = cred->cr_groups; 3282 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3283 if (vap->va_gid == *gp) 3284 goto found; 3285 mode >>= 3; 3286 found: 3287 ; 3288 } 3289 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3290 return (error); 3291 } 3292 3293 /* 3294 * Read wrapper for special devices. 3295 * 3296 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3297 * struct ucred *a_cred) 3298 */ 3299 static int 3300 nfsspec_read(struct vop_read_args *ap) 3301 { 3302 struct nfsnode *np = VTONFS(ap->a_vp); 3303 3304 /* 3305 * Set access flag. 3306 */ 3307 np->n_flag |= NACC; 3308 getnanotime(&np->n_atim); 3309 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3310 } 3311 3312 /* 3313 * Write wrapper for special devices. 3314 * 3315 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3316 * struct ucred *a_cred) 3317 */ 3318 static int 3319 nfsspec_write(struct vop_write_args *ap) 3320 { 3321 struct nfsnode *np = VTONFS(ap->a_vp); 3322 3323 /* 3324 * Set update flag. 3325 */ 3326 np->n_flag |= NUPD; 3327 getnanotime(&np->n_mtim); 3328 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3329 } 3330 3331 /* 3332 * Close wrapper for special devices. 3333 * 3334 * Update the times on the nfsnode then do device close. 3335 * 3336 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3337 * struct thread *a_td) 3338 */ 3339 static int 3340 nfsspec_close(struct vop_close_args *ap) 3341 { 3342 struct vnode *vp = ap->a_vp; 3343 struct nfsnode *np = VTONFS(vp); 3344 struct vattr vattr; 3345 3346 if (np->n_flag & (NACC | NUPD)) { 3347 np->n_flag |= NCHG; 3348 if (vp->v_usecount == 1 && 3349 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3350 VATTR_NULL(&vattr); 3351 if (np->n_flag & NACC) 3352 vattr.va_atime = np->n_atim; 3353 if (np->n_flag & NUPD) 3354 vattr.va_mtime = np->n_mtim; 3355 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3356 } 3357 } 3358 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3359 } 3360 3361 /* 3362 * Read wrapper for fifos. 3363 * 3364 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3365 * struct ucred *a_cred) 3366 */ 3367 static int 3368 nfsfifo_read(struct vop_read_args *ap) 3369 { 3370 struct nfsnode *np = VTONFS(ap->a_vp); 3371 3372 /* 3373 * Set access flag. 3374 */ 3375 np->n_flag |= NACC; 3376 getnanotime(&np->n_atim); 3377 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3378 } 3379 3380 /* 3381 * Write wrapper for fifos. 3382 * 3383 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3384 * struct ucred *a_cred) 3385 */ 3386 static int 3387 nfsfifo_write(struct vop_write_args *ap) 3388 { 3389 struct nfsnode *np = VTONFS(ap->a_vp); 3390 3391 /* 3392 * Set update flag. 3393 */ 3394 np->n_flag |= NUPD; 3395 getnanotime(&np->n_mtim); 3396 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3397 } 3398 3399 /* 3400 * Close wrapper for fifos. 3401 * 3402 * Update the times on the nfsnode then do fifo close. 3403 * 3404 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3405 */ 3406 static int 3407 nfsfifo_close(struct vop_close_args *ap) 3408 { 3409 struct vnode *vp = ap->a_vp; 3410 struct nfsnode *np = VTONFS(vp); 3411 struct vattr vattr; 3412 struct timespec ts; 3413 3414 if (np->n_flag & (NACC | NUPD)) { 3415 getnanotime(&ts); 3416 if (np->n_flag & NACC) 3417 np->n_atim = ts; 3418 if (np->n_flag & NUPD) 3419 np->n_mtim = ts; 3420 np->n_flag |= NCHG; 3421 if (vp->v_usecount == 1 && 3422 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3423 VATTR_NULL(&vattr); 3424 if (np->n_flag & NACC) 3425 vattr.va_atime = np->n_atim; 3426 if (np->n_flag & NUPD) 3427 vattr.va_mtime = np->n_mtim; 3428 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE)); 3429 } 3430 } 3431 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3432 } 3433 3434