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 1.43 2005/09/14 01:13:39 dillon Exp $ 39 */ 40 41 42 /* 43 * vnode op calls for Sun NFS version 2 and 3 44 */ 45 46 #include "opt_inet.h" 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/resourcevar.h> 52 #include <sys/proc.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/namei.h> 58 #include <sys/nlookup.h> 59 #include <sys/socket.h> 60 #include <sys/vnode.h> 61 #include <sys/dirent.h> 62 #include <sys/fcntl.h> 63 #include <sys/lockf.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/conf.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_zone.h> 71 72 #include <sys/buf2.h> 73 74 #include <vfs/fifofs/fifo.h> 75 #include <vfs/ufs/dir.h> 76 77 #undef DIRBLKSIZ 78 79 #include "rpcv2.h" 80 #include "nfsproto.h" 81 #include "nfs.h" 82 #include "nfsmount.h" 83 #include "nfsnode.h" 84 #include "xdr_subs.h" 85 #include "nfsm_subs.h" 86 #include "nqnfs.h" 87 88 #include <net/if.h> 89 #include <netinet/in.h> 90 #include <netinet/in_var.h> 91 92 #include <sys/thread2.h> 93 94 /* Defs */ 95 #define TRUE 1 96 #define FALSE 0 97 98 /* 99 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these 100 * calls are not in getblk() and brelse() so that they would not be necessary 101 * here. 102 */ 103 #ifndef B_VMIO 104 #define vfs_busy_pages(bp, f) 105 #endif 106 107 static int nfsspec_read (struct vop_read_args *); 108 static int nfsspec_write (struct vop_write_args *); 109 static int nfsfifo_read (struct vop_read_args *); 110 static int nfsfifo_write (struct vop_write_args *); 111 static int nfsspec_close (struct vop_close_args *); 112 static int nfsfifo_close (struct vop_close_args *); 113 #define nfs_poll vop_nopoll 114 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *); 115 static int nfs_lookup (struct vop_old_lookup_args *); 116 static int nfs_create (struct vop_old_create_args *); 117 static int nfs_mknod (struct vop_old_mknod_args *); 118 static int nfs_open (struct vop_open_args *); 119 static int nfs_close (struct vop_close_args *); 120 static int nfs_access (struct vop_access_args *); 121 static int nfs_getattr (struct vop_getattr_args *); 122 static int nfs_setattr (struct vop_setattr_args *); 123 static int nfs_read (struct vop_read_args *); 124 static int nfs_mmap (struct vop_mmap_args *); 125 static int nfs_fsync (struct vop_fsync_args *); 126 static int nfs_remove (struct vop_old_remove_args *); 127 static int nfs_link (struct vop_old_link_args *); 128 static int nfs_rename (struct vop_old_rename_args *); 129 static int nfs_mkdir (struct vop_old_mkdir_args *); 130 static int nfs_rmdir (struct vop_old_rmdir_args *); 131 static int nfs_symlink (struct vop_old_symlink_args *); 132 static int nfs_readdir (struct vop_readdir_args *); 133 static int nfs_bmap (struct vop_bmap_args *); 134 static int nfs_strategy (struct vop_strategy_args *); 135 static int nfs_lookitup (struct vnode *, const char *, int, 136 struct ucred *, struct thread *, struct nfsnode **); 137 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *); 138 static int nfsspec_access (struct vop_access_args *); 139 static int nfs_readlink (struct vop_readlink_args *); 140 static int nfs_print (struct vop_print_args *); 141 static int nfs_advlock (struct vop_advlock_args *); 142 static int nfs_bwrite (struct vop_bwrite_args *); 143 144 static int nfs_nresolve (struct vop_nresolve_args *); 145 /* 146 * Global vfs data structures for nfs 147 */ 148 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { 149 { &vop_default_desc, vop_defaultop }, 150 { &vop_access_desc, (vnodeopv_entry_t) nfs_access }, 151 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock }, 152 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap }, 153 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite }, 154 { &vop_close_desc, (vnodeopv_entry_t) nfs_close }, 155 { &vop_old_create_desc, (vnodeopv_entry_t) nfs_create }, 156 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 157 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 158 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages }, 159 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages }, 160 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 161 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 162 { &vop_lease_desc, vop_null }, 163 { &vop_old_link_desc, (vnodeopv_entry_t) nfs_link }, 164 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 165 { &vop_old_lookup_desc, (vnodeopv_entry_t) nfs_lookup }, 166 { &vop_old_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir }, 167 { &vop_old_mknod_desc, (vnodeopv_entry_t) nfs_mknod }, 168 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap }, 169 { &vop_open_desc, (vnodeopv_entry_t) nfs_open }, 170 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll }, 171 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 172 { &vop_read_desc, (vnodeopv_entry_t) nfs_read }, 173 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir }, 174 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink }, 175 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 176 { &vop_old_remove_desc, (vnodeopv_entry_t) nfs_remove }, 177 { &vop_old_rename_desc, (vnodeopv_entry_t) nfs_rename }, 178 { &vop_old_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir }, 179 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 180 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy }, 181 { &vop_old_symlink_desc, (vnodeopv_entry_t) nfs_symlink }, 182 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 183 { &vop_write_desc, (vnodeopv_entry_t) nfs_write }, 184 185 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve }, 186 { NULL, NULL } 187 }; 188 189 /* 190 * Special device vnode ops 191 */ 192 struct vnodeopv_entry_desc nfsv2_specop_entries[] = { 193 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate }, 194 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 195 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close }, 196 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 197 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 198 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 199 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 200 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 201 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 202 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read }, 203 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 204 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 205 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 206 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write }, 207 { NULL, NULL } 208 }; 209 210 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = { 211 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate }, 212 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access }, 213 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close }, 214 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync }, 215 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr }, 216 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive }, 217 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked }, 218 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock }, 219 { &vop_print_desc, (vnodeopv_entry_t) nfs_print }, 220 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read }, 221 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim }, 222 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr }, 223 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock }, 224 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write }, 225 { NULL, NULL } 226 }; 227 228 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp, 229 struct componentname *cnp, 230 struct vattr *vap); 231 static int nfs_removerpc (struct vnode *dvp, const char *name, 232 int namelen, 233 struct ucred *cred, struct thread *td); 234 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr, 235 int fnamelen, struct vnode *tdvp, 236 const char *tnameptr, int tnamelen, 237 struct ucred *cred, struct thread *td); 238 static int nfs_renameit (struct vnode *sdvp, 239 struct componentname *scnp, 240 struct sillyrename *sp); 241 242 /* 243 * Global variables 244 */ 245 extern u_int32_t nfs_true, nfs_false; 246 extern u_int32_t nfs_xdrneg1; 247 extern struct nfsstats nfsstats; 248 extern nfstype nfsv3_type[9]; 249 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON]; 250 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; 251 int nfs_numasync = 0; 252 253 SYSCTL_DECL(_vfs_nfs); 254 255 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; 256 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, 257 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); 258 259 static int nfsneg_cache_timeout = NFS_MINATTRTIMO; 260 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW, 261 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout"); 262 263 static int nfsv3_commit_on_close = 0; 264 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, 265 &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); 266 #if 0 267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, 268 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); 269 270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, 271 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); 272 #endif 273 274 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ 275 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ 276 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) 277 static int 278 nfs3_access_otw(struct vnode *vp, int wmode, 279 struct thread *td, struct ucred *cred) 280 { 281 const int v3 = 1; 282 u_int32_t *tl; 283 int error = 0, attrflag; 284 285 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 286 caddr_t bpos, dpos, cp2; 287 int32_t t1, t2; 288 caddr_t cp; 289 u_int32_t rmode; 290 struct nfsnode *np = VTONFS(vp); 291 292 nfsstats.rpccnt[NFSPROC_ACCESS]++; 293 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); 294 nfsm_fhtom(vp, v3); 295 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 296 *tl = txdr_unsigned(wmode); 297 nfsm_request(vp, NFSPROC_ACCESS, td, cred); 298 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 299 if (!error) { 300 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 301 rmode = fxdr_unsigned(u_int32_t, *tl); 302 np->n_mode = rmode; 303 np->n_modeuid = cred->cr_uid; 304 np->n_modestamp = mycpu->gd_time_seconds; 305 } 306 m_freem(mrep); 307 nfsmout: 308 return error; 309 } 310 311 /* 312 * nfs access vnode op. 313 * For nfs version 2, just return ok. File accesses may fail later. 314 * For nfs version 3, use the access rpc to check accessibility. If file modes 315 * are changed on the server, accesses might still fail later. 316 * 317 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 318 * struct thread *a_td) 319 */ 320 static int 321 nfs_access(struct vop_access_args *ap) 322 { 323 struct vnode *vp = ap->a_vp; 324 int error = 0; 325 u_int32_t mode, wmode; 326 int v3 = NFS_ISV3(vp); 327 struct nfsnode *np = VTONFS(vp); 328 329 /* 330 * Disallow write attempts on filesystems mounted read-only; 331 * unless the file is a socket, fifo, or a block or character 332 * device resident on the filesystem. 333 */ 334 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 335 switch (vp->v_type) { 336 case VREG: 337 case VDIR: 338 case VLNK: 339 return (EROFS); 340 default: 341 break; 342 } 343 } 344 /* 345 * For nfs v3, check to see if we have done this recently, and if 346 * so return our cached result instead of making an ACCESS call. 347 * If not, do an access rpc, otherwise you are stuck emulating 348 * ufs_access() locally using the vattr. This may not be correct, 349 * since the server may apply other access criteria such as 350 * client uid-->server uid mapping that we do not know about. 351 */ 352 if (v3) { 353 if (ap->a_mode & VREAD) 354 mode = NFSV3ACCESS_READ; 355 else 356 mode = 0; 357 if (vp->v_type != VDIR) { 358 if (ap->a_mode & VWRITE) 359 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); 360 if (ap->a_mode & VEXEC) 361 mode |= NFSV3ACCESS_EXECUTE; 362 } else { 363 if (ap->a_mode & VWRITE) 364 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | 365 NFSV3ACCESS_DELETE); 366 if (ap->a_mode & VEXEC) 367 mode |= NFSV3ACCESS_LOOKUP; 368 } 369 /* XXX safety belt, only make blanket request if caching */ 370 if (nfsaccess_cache_timeout > 0) { 371 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | 372 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | 373 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; 374 } else { 375 wmode = mode; 376 } 377 378 /* 379 * Does our cached result allow us to give a definite yes to 380 * this request? 381 */ 382 if (np->n_modestamp && 383 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) && 384 (ap->a_cred->cr_uid == np->n_modeuid) && 385 ((np->n_mode & mode) == mode)) { 386 nfsstats.accesscache_hits++; 387 } else { 388 /* 389 * Either a no, or a don't know. Go to the wire. 390 */ 391 nfsstats.accesscache_misses++; 392 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred); 393 if (!error) { 394 if ((np->n_mode & mode) != mode) { 395 error = EACCES; 396 } 397 } 398 } 399 } else { 400 if ((error = nfsspec_access(ap)) != 0) 401 return (error); 402 403 /* 404 * Attempt to prevent a mapped root from accessing a file 405 * which it shouldn't. We try to read a byte from the file 406 * if the user is root and the file is not zero length. 407 * After calling nfsspec_access, we should have the correct 408 * file size cached. 409 */ 410 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD) 411 && VTONFS(vp)->n_size > 0) { 412 struct iovec aiov; 413 struct uio auio; 414 char buf[1]; 415 416 aiov.iov_base = buf; 417 aiov.iov_len = 1; 418 auio.uio_iov = &aiov; 419 auio.uio_iovcnt = 1; 420 auio.uio_offset = 0; 421 auio.uio_resid = 1; 422 auio.uio_segflg = UIO_SYSSPACE; 423 auio.uio_rw = UIO_READ; 424 auio.uio_td = ap->a_td; 425 426 if (vp->v_type == VREG) { 427 error = nfs_readrpc(vp, &auio); 428 } else if (vp->v_type == VDIR) { 429 char* bp; 430 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); 431 aiov.iov_base = bp; 432 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; 433 error = nfs_readdirrpc(vp, &auio); 434 free(bp, M_TEMP); 435 } else if (vp->v_type == VLNK) { 436 error = nfs_readlinkrpc(vp, &auio); 437 } else { 438 error = EACCES; 439 } 440 } 441 } 442 /* 443 * [re]record creds for reading and/or writing if access 444 * was granted. Assume the NFS server will grant read access 445 * for execute requests. 446 */ 447 if (error == 0) { 448 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) { 449 crhold(ap->a_cred); 450 if (np->n_rucred) 451 crfree(np->n_rucred); 452 np->n_rucred = ap->a_cred; 453 } 454 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) { 455 crhold(ap->a_cred); 456 if (np->n_wucred) 457 crfree(np->n_wucred); 458 np->n_wucred = ap->a_cred; 459 } 460 } 461 return(error); 462 } 463 464 /* 465 * nfs open vnode op 466 * Check to see if the type is ok 467 * and that deletion is not in progress. 468 * For paged in text files, you will need to flush the page cache 469 * if consistency is lost. 470 * 471 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 472 * struct thread *a_td) 473 */ 474 /* ARGSUSED */ 475 static int 476 nfs_open(struct vop_open_args *ap) 477 { 478 struct vnode *vp = ap->a_vp; 479 struct nfsnode *np = VTONFS(vp); 480 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 481 struct vattr vattr; 482 int error; 483 484 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) { 485 #ifdef DIAGNOSTIC 486 printf("open eacces vtyp=%d\n",vp->v_type); 487 #endif 488 return (EOPNOTSUPP); 489 } 490 491 /* 492 * Clear the attribute cache only if opening with write access. It 493 * is unclear if we should do this at all here, but we certainly 494 * should not clear the cache unconditionally simply because a file 495 * is being opened. 496 */ 497 if (ap->a_mode & FWRITE) 498 np->n_attrstamp = 0; 499 500 if (nmp->nm_flag & NFSMNT_NQNFS) { 501 /* 502 * If NQNFS is active, get a valid lease 503 */ 504 if (NQNFS_CKINVALID(vp, np, ND_READ)) { 505 do { 506 error = nqnfs_getlease(vp, ND_READ, ap->a_td); 507 } while (error == NQNFS_EXPIRED); 508 if (error) 509 return (error); 510 if (np->n_lrev != np->n_brev || 511 (np->n_flag & NQNFSNONCACHE)) { 512 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) 513 == EINTR) { 514 return (error); 515 } 516 np->n_brev = np->n_lrev; 517 } 518 } 519 } else { 520 /* 521 * For normal NFS, reconcile changes made locally verses 522 * changes made remotely. Note that VOP_GETATTR only goes 523 * to the wire if the cached attribute has timed out or been 524 * cleared. 525 * 526 * If local modifications have been made clear the attribute 527 * cache to force an attribute and modified time check. If 528 * GETATTR detects that the file has been changed by someone 529 * other then us it will set NRMODIFIED. 530 * 531 * If we are opening a directory and local changes have been 532 * made we have to invalidate the cache in order to ensure 533 * that we get the most up-to-date information from the 534 * server. XXX 535 */ 536 if (np->n_flag & NLMODIFIED) { 537 np->n_attrstamp = 0; 538 if (vp->v_type == VDIR) { 539 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 540 if (error == EINTR) 541 return (error); 542 nfs_invaldir(vp); 543 } 544 } 545 error = VOP_GETATTR(vp, &vattr, ap->a_td); 546 if (error) 547 return (error); 548 if (np->n_flag & NRMODIFIED) { 549 if (vp->v_type == VDIR) 550 nfs_invaldir(vp); 551 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 552 if (error == EINTR) 553 return (error); 554 np->n_flag &= ~NRMODIFIED; 555 } 556 } 557 558 return (0); 559 } 560 561 /* 562 * nfs close vnode op 563 * What an NFS client should do upon close after writing is a debatable issue. 564 * Most NFS clients push delayed writes to the server upon close, basically for 565 * two reasons: 566 * 1 - So that any write errors may be reported back to the client process 567 * doing the close system call. By far the two most likely errors are 568 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. 569 * 2 - To put a worst case upper bound on cache inconsistency between 570 * multiple clients for the file. 571 * There is also a consistency problem for Version 2 of the protocol w.r.t. 572 * not being able to tell if other clients are writing a file concurrently, 573 * since there is no way of knowing if the changed modify time in the reply 574 * is only due to the write for this client. 575 * (NFS Version 3 provides weak cache consistency data in the reply that 576 * should be sufficient to detect and handle this case.) 577 * 578 * The current code does the following: 579 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers 580 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate 581 * or commit them (this satisfies 1 and 2 except for the 582 * case where the server crashes after this close but 583 * before the commit RPC, which is felt to be "good 584 * enough". Changing the last argument to nfs_flush() to 585 * a 1 would force a commit operation, if it is felt a 586 * commit is necessary now. 587 * for NQNFS - do nothing now, since 2 is dealt with via leases and 588 * 1 should be dealt with via an fsync() system call for 589 * cases where write errors are important. 590 * 591 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag, 592 * struct ucred *a_cred, struct thread *a_td) 593 */ 594 /* ARGSUSED */ 595 static int 596 nfs_close(struct vop_close_args *ap) 597 { 598 struct vnode *vp = ap->a_vp; 599 struct nfsnode *np = VTONFS(vp); 600 int error = 0; 601 602 if (vp->v_type == VREG) { 603 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 && 604 (np->n_flag & NLMODIFIED)) { 605 if (NFS_ISV3(vp)) { 606 /* 607 * Under NFSv3 we have dirty buffers to dispose of. We 608 * must flush them to the NFS server. We have the option 609 * of waiting all the way through the commit rpc or just 610 * waiting for the initial write. The default is to only 611 * wait through the initial write so the data is in the 612 * server's cache, which is roughly similar to the state 613 * a standard disk subsystem leaves the file in on close(). 614 * 615 * We cannot clear the NLMODIFIED bit in np->n_flag due to 616 * potential races with other processes, and certainly 617 * cannot clear it if we don't commit. 618 */ 619 int cm = nfsv3_commit_on_close ? 1 : 0; 620 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm); 621 /* np->n_flag &= ~NLMODIFIED; */ 622 } else { 623 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 624 } 625 np->n_attrstamp = 0; 626 } 627 if (np->n_flag & NWRITEERR) { 628 np->n_flag &= ~NWRITEERR; 629 error = np->n_error; 630 } 631 } 632 return (error); 633 } 634 635 /* 636 * nfs getattr call from vfs. 637 * 638 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 639 * struct thread *a_td) 640 */ 641 static int 642 nfs_getattr(struct vop_getattr_args *ap) 643 { 644 struct vnode *vp = ap->a_vp; 645 struct nfsnode *np = VTONFS(vp); 646 caddr_t cp; 647 u_int32_t *tl; 648 int32_t t1, t2; 649 caddr_t bpos, dpos; 650 int error = 0; 651 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 652 int v3 = NFS_ISV3(vp); 653 654 /* 655 * Update local times for special files. 656 */ 657 if (np->n_flag & (NACC | NUPD)) 658 np->n_flag |= NCHG; 659 /* 660 * First look in the cache. 661 */ 662 if (nfs_getattrcache(vp, ap->a_vap) == 0) 663 return (0); 664 665 if (v3 && nfsaccess_cache_timeout > 0) { 666 nfsstats.accesscache_misses++; 667 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 668 if (nfs_getattrcache(vp, ap->a_vap) == 0) 669 return (0); 670 } 671 672 nfsstats.rpccnt[NFSPROC_GETATTR]++; 673 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); 674 nfsm_fhtom(vp, v3); 675 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK)); 676 if (!error) { 677 nfsm_loadattr(vp, ap->a_vap); 678 } 679 m_freem(mrep); 680 nfsmout: 681 return (error); 682 } 683 684 /* 685 * nfs setattr call. 686 * 687 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 688 * struct vattr *a_vap, struct ucred *a_cred, 689 * struct thread *a_td) 690 */ 691 static int 692 nfs_setattr(struct vop_setattr_args *ap) 693 { 694 struct vnode *vp = ap->a_vp; 695 struct nfsnode *np = VTONFS(vp); 696 struct vattr *vap = ap->a_vap; 697 int error = 0; 698 u_quad_t tsize; 699 700 #ifndef nolint 701 tsize = (u_quad_t)0; 702 #endif 703 704 /* 705 * Setting of flags is not supported. 706 */ 707 if (vap->va_flags != VNOVAL) 708 return (EOPNOTSUPP); 709 710 /* 711 * Disallow write attempts if the filesystem is mounted read-only. 712 */ 713 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 714 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 715 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 716 (vp->v_mount->mnt_flag & MNT_RDONLY)) 717 return (EROFS); 718 if (vap->va_size != VNOVAL) { 719 switch (vp->v_type) { 720 case VDIR: 721 return (EISDIR); 722 case VCHR: 723 case VBLK: 724 case VSOCK: 725 case VFIFO: 726 if (vap->va_mtime.tv_sec == VNOVAL && 727 vap->va_atime.tv_sec == VNOVAL && 728 vap->va_mode == (mode_t)VNOVAL && 729 vap->va_uid == (uid_t)VNOVAL && 730 vap->va_gid == (gid_t)VNOVAL) 731 return (0); 732 vap->va_size = VNOVAL; 733 break; 734 default: 735 /* 736 * Disallow write attempts if the filesystem is 737 * mounted read-only. 738 */ 739 if (vp->v_mount->mnt_flag & MNT_RDONLY) 740 return (EROFS); 741 742 /* 743 * This is nasty. The RPCs we send to flush pending 744 * data often return attribute information which is 745 * cached via a callback to nfs_loadattrcache(), which 746 * has the effect of changing our notion of the file 747 * size. Due to flushed appends and other operations 748 * the file size can be set to virtually anything, 749 * including values that do not match either the old 750 * or intended file size. 751 * 752 * When this condition is detected we must loop to 753 * try the operation again. Hopefully no more 754 * flushing is required on the loop so it works the 755 * second time around. THIS CASE ALMOST ALWAYS 756 * HAPPENS! 757 */ 758 tsize = np->n_size; 759 again: 760 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size); 761 762 if (np->n_flag & NLMODIFIED) { 763 if (vap->va_size == 0) 764 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1); 765 else 766 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); 767 } 768 /* 769 * note: this loop case almost always happens at 770 * least once per truncation. 771 */ 772 if (error == 0 && np->n_size != vap->va_size) 773 goto again; 774 np->n_vattr.va_size = vap->va_size; 775 break; 776 } 777 } else if ((vap->va_mtime.tv_sec != VNOVAL || 778 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) && 779 vp->v_type == VREG && 780 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR 781 ) { 782 return (error); 783 } 784 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td); 785 786 /* 787 * Sanity check if a truncation was issued. This should only occur 788 * if multiple processes are racing on the same file. 789 */ 790 if (error == 0 && vap->va_size != VNOVAL && 791 np->n_size != vap->va_size) { 792 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size); 793 goto again; 794 } 795 if (error && vap->va_size != VNOVAL) { 796 np->n_size = np->n_vattr.va_size = tsize; 797 vnode_pager_setsize(vp, np->n_size); 798 } 799 return (error); 800 } 801 802 /* 803 * Do an nfs setattr rpc. 804 */ 805 static int 806 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, 807 struct ucred *cred, struct thread *td) 808 { 809 struct nfsv2_sattr *sp; 810 struct nfsnode *np = VTONFS(vp); 811 caddr_t cp; 812 int32_t t1, t2; 813 caddr_t bpos, dpos, cp2; 814 u_int32_t *tl; 815 int error = 0, wccflag = NFSV3_WCCRATTR; 816 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 817 int v3 = NFS_ISV3(vp); 818 819 nfsstats.rpccnt[NFSPROC_SETATTR]++; 820 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); 821 nfsm_fhtom(vp, v3); 822 if (v3) { 823 nfsm_v3attrbuild(vap, TRUE); 824 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 825 *tl = nfs_false; 826 } else { 827 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 828 if (vap->va_mode == (mode_t)VNOVAL) 829 sp->sa_mode = nfs_xdrneg1; 830 else 831 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); 832 if (vap->va_uid == (uid_t)VNOVAL) 833 sp->sa_uid = nfs_xdrneg1; 834 else 835 sp->sa_uid = txdr_unsigned(vap->va_uid); 836 if (vap->va_gid == (gid_t)VNOVAL) 837 sp->sa_gid = nfs_xdrneg1; 838 else 839 sp->sa_gid = txdr_unsigned(vap->va_gid); 840 sp->sa_size = txdr_unsigned(vap->va_size); 841 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 842 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 843 } 844 nfsm_request(vp, NFSPROC_SETATTR, td, cred); 845 if (v3) { 846 np->n_modestamp = 0; 847 nfsm_wcc_data(vp, wccflag); 848 } else 849 nfsm_loadattr(vp, (struct vattr *)0); 850 m_freem(mrep); 851 nfsmout: 852 return (error); 853 } 854 855 /* 856 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove 857 * nfs_lookup() until all remaining new api calls are implemented. 858 * 859 * Resolve a namecache entry. This function is passed a locked ncp and 860 * must call cache_setvp() on it as appropriate to resolve the entry. 861 */ 862 static int 863 nfs_nresolve(struct vop_nresolve_args *ap) 864 { 865 struct thread *td = curthread; 866 struct namecache *ncp; 867 struct ucred *cred; 868 struct nfsnode *np; 869 struct vnode *dvp; 870 struct vnode *nvp; 871 nfsfh_t *fhp; 872 int attrflag; 873 int fhsize; 874 int error; 875 int len; 876 int v3; 877 /******NFSM MACROS********/ 878 struct mbuf *mb, *mrep, *mreq, *mb2, *md; 879 caddr_t bpos, dpos, cp, cp2; 880 u_int32_t *tl; 881 int32_t t1, t2; 882 883 cred = ap->a_cred; 884 ncp = ap->a_ncp; 885 886 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp); 887 dvp = ncp->nc_parent->nc_vp; 888 if ((error = vget(dvp, LK_SHARED, td)) != 0) 889 return (error); 890 891 nvp = NULL; 892 v3 = NFS_ISV3(dvp); 893 nfsstats.lookupcache_misses++; 894 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 895 len = ncp->nc_nlen; 896 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 897 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 898 nfsm_fhtom(dvp, v3); 899 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN); 900 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred); 901 if (error) { 902 /* 903 * Cache negatve lookups to reduce NFS traffic, but use 904 * a fast timeout. Otherwise use a timeout of 1 tick. 905 * XXX we should add a namecache flag for no-caching 906 * to uncache the negative hit as soon as possible, but 907 * we cannot simply destroy the entry because it is used 908 * as a placeholder by the caller. 909 */ 910 if (error == ENOENT) { 911 int nticks; 912 913 if (nfsneg_cache_timeout) 914 nticks = nfsneg_cache_timeout * hz; 915 else 916 nticks = 1; 917 cache_setvp(ncp, NULL); 918 cache_settimeout(ncp, nticks); 919 } 920 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 921 m_freem(mrep); 922 goto nfsmout; 923 } 924 925 /* 926 * Success, get the file handle, do various checks, and load 927 * post-operation data from the reply packet. Theoretically 928 * we should never be looking up "." so, theoretically, we 929 * should never get the same file handle as our directory. But 930 * we check anyway. XXX 931 * 932 * Note that no timeout is set for the positive cache hit. We 933 * assume, theoretically, that ESTALE returns will be dealt with 934 * properly to handle NFS races and in anycase we cannot depend 935 * on a timeout to deal with NFS open/create/excl issues so instead 936 * of a bad hack here the rest of the NFS client code needs to do 937 * the right thing. 938 */ 939 nfsm_getfh(fhp, fhsize, v3); 940 941 np = VTONFS(dvp); 942 if (NFS_CMPFH(np, fhp, fhsize)) { 943 vref(dvp); 944 nvp = dvp; 945 } else { 946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 947 if (error) { 948 m_freem(mrep); 949 vput(dvp); 950 return (error); 951 } 952 nvp = NFSTOV(np); 953 } 954 if (v3) { 955 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK); 956 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 957 } else { 958 nfsm_loadattr(nvp, NULL); 959 } 960 cache_setvp(ncp, nvp); 961 m_freem(mrep); 962 nfsmout: 963 vput(dvp); 964 if (nvp) { 965 if (nvp == dvp) 966 vrele(nvp); 967 else 968 vput(nvp); 969 } 970 return (error); 971 } 972 973 /* 974 * 'cached' nfs directory lookup 975 * 976 * NOTE: cannot be removed until NFS implements all the new n*() API calls. 977 * 978 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 979 * struct vnode **a_vpp, struct componentname *a_cnp) 980 */ 981 static int 982 nfs_lookup(struct vop_old_lookup_args *ap) 983 { 984 struct componentname *cnp = ap->a_cnp; 985 struct vnode *dvp = ap->a_dvp; 986 struct vnode **vpp = ap->a_vpp; 987 int flags = cnp->cn_flags; 988 struct vnode *newvp; 989 u_int32_t *tl; 990 caddr_t cp; 991 int32_t t1, t2; 992 struct nfsmount *nmp; 993 caddr_t bpos, dpos, cp2; 994 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 995 long len; 996 nfsfh_t *fhp; 997 struct nfsnode *np; 998 int lockparent, wantparent, error = 0, attrflag, fhsize; 999 int v3 = NFS_ISV3(dvp); 1000 struct thread *td = cnp->cn_td; 1001 1002 /* 1003 * Read-only mount check and directory check. 1004 */ 1005 *vpp = NULLVP; 1006 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 1007 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) 1008 return (EROFS); 1009 1010 if (dvp->v_type != VDIR) 1011 return (ENOTDIR); 1012 1013 /* 1014 * Look it up in the cache. Note that ENOENT is only returned if we 1015 * previously entered a negative hit (see later on). The additional 1016 * nfsneg_cache_timeout check causes previously cached results to 1017 * be instantly ignored if the negative caching is turned off. 1018 */ 1019 lockparent = flags & CNP_LOCKPARENT; 1020 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT); 1021 nmp = VFSTONFS(dvp->v_mount); 1022 np = VTONFS(dvp); 1023 1024 /* 1025 * Go to the wire. 1026 */ 1027 error = 0; 1028 newvp = NULLVP; 1029 nfsstats.lookupcache_misses++; 1030 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 1031 len = cnp->cn_namelen; 1032 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 1033 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 1034 nfsm_fhtom(dvp, v3); 1035 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 1036 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred); 1037 if (error) { 1038 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1039 m_freem(mrep); 1040 goto nfsmout; 1041 } 1042 nfsm_getfh(fhp, fhsize, v3); 1043 1044 /* 1045 * Handle RENAME case... 1046 */ 1047 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) { 1048 if (NFS_CMPFH(np, fhp, fhsize)) { 1049 m_freem(mrep); 1050 return (EISDIR); 1051 } 1052 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1053 if (error) { 1054 m_freem(mrep); 1055 return (error); 1056 } 1057 newvp = NFSTOV(np); 1058 if (v3) { 1059 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1060 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1061 } else 1062 nfsm_loadattr(newvp, (struct vattr *)0); 1063 *vpp = newvp; 1064 m_freem(mrep); 1065 if (!lockparent) { 1066 VOP_UNLOCK(dvp, 0, td); 1067 cnp->cn_flags |= CNP_PDIRUNLOCK; 1068 } 1069 return (0); 1070 } 1071 1072 if (flags & CNP_ISDOTDOT) { 1073 VOP_UNLOCK(dvp, 0, td); 1074 cnp->cn_flags |= CNP_PDIRUNLOCK; 1075 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1076 if (error) { 1077 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 1078 cnp->cn_flags &= ~CNP_PDIRUNLOCK; 1079 return (error); /* NOTE: return error from nget */ 1080 } 1081 newvp = NFSTOV(np); 1082 if (lockparent) { 1083 error = vn_lock(dvp, LK_EXCLUSIVE, td); 1084 if (error) { 1085 vput(newvp); 1086 return (error); 1087 } 1088 cnp->cn_flags |= CNP_PDIRUNLOCK; 1089 } 1090 } else if (NFS_CMPFH(np, fhp, fhsize)) { 1091 vref(dvp); 1092 newvp = dvp; 1093 } else { 1094 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np); 1095 if (error) { 1096 m_freem(mrep); 1097 return (error); 1098 } 1099 if (!lockparent) { 1100 VOP_UNLOCK(dvp, 0, td); 1101 cnp->cn_flags |= CNP_PDIRUNLOCK; 1102 } 1103 newvp = NFSTOV(np); 1104 } 1105 if (v3) { 1106 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 1107 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK); 1108 } else 1109 nfsm_loadattr(newvp, (struct vattr *)0); 1110 #if 0 1111 /* XXX MOVE TO nfs_nremove() */ 1112 if ((cnp->cn_flags & CNP_MAKEENTRY) && 1113 cnp->cn_nameiop != NAMEI_DELETE) { 1114 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */ 1115 } 1116 #endif 1117 *vpp = newvp; 1118 m_freem(mrep); 1119 nfsmout: 1120 if (error) { 1121 if (newvp != NULLVP) { 1122 vrele(newvp); 1123 *vpp = NULLVP; 1124 } 1125 if ((cnp->cn_nameiop == NAMEI_CREATE || 1126 cnp->cn_nameiop == NAMEI_RENAME) && 1127 error == ENOENT) { 1128 if (!lockparent) { 1129 VOP_UNLOCK(dvp, 0, td); 1130 cnp->cn_flags |= CNP_PDIRUNLOCK; 1131 } 1132 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 1133 error = EROFS; 1134 else 1135 error = EJUSTRETURN; 1136 } 1137 } 1138 return (error); 1139 } 1140 1141 /* 1142 * nfs read call. 1143 * Just call nfs_bioread() to do the work. 1144 * 1145 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1146 * struct ucred *a_cred) 1147 */ 1148 static int 1149 nfs_read(struct vop_read_args *ap) 1150 { 1151 struct vnode *vp = ap->a_vp; 1152 1153 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1154 switch (vp->v_type) { 1155 case VREG: 1156 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag)); 1157 case VDIR: 1158 return (EISDIR); 1159 default: 1160 return EOPNOTSUPP; 1161 } 1162 } 1163 1164 /* 1165 * nfs readlink call 1166 * 1167 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 1168 */ 1169 static int 1170 nfs_readlink(struct vop_readlink_args *ap) 1171 { 1172 struct vnode *vp = ap->a_vp; 1173 1174 if (vp->v_type != VLNK) 1175 return (EINVAL); 1176 return (nfs_bioread(vp, ap->a_uio, 0)); 1177 } 1178 1179 /* 1180 * Do a readlink rpc. 1181 * Called by nfs_doio() from below the buffer cache. 1182 */ 1183 int 1184 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop) 1185 { 1186 u_int32_t *tl; 1187 caddr_t cp; 1188 int32_t t1, t2; 1189 caddr_t bpos, dpos, cp2; 1190 int error = 0, len, attrflag; 1191 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1192 int v3 = NFS_ISV3(vp); 1193 1194 nfsstats.rpccnt[NFSPROC_READLINK]++; 1195 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); 1196 nfsm_fhtom(vp, v3); 1197 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK)); 1198 if (v3) 1199 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1200 if (!error) { 1201 nfsm_strsiz(len, NFS_MAXPATHLEN); 1202 if (len == NFS_MAXPATHLEN) { 1203 struct nfsnode *np = VTONFS(vp); 1204 if (np->n_size && np->n_size < NFS_MAXPATHLEN) 1205 len = np->n_size; 1206 } 1207 nfsm_mtouio(uiop, len); 1208 } 1209 m_freem(mrep); 1210 nfsmout: 1211 return (error); 1212 } 1213 1214 /* 1215 * nfs read rpc call 1216 * Ditto above 1217 */ 1218 int 1219 nfs_readrpc(struct vnode *vp, struct uio *uiop) 1220 { 1221 u_int32_t *tl; 1222 caddr_t cp; 1223 int32_t t1, t2; 1224 caddr_t bpos, dpos, cp2; 1225 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1226 struct nfsmount *nmp; 1227 int error = 0, len, retlen, tsiz, eof, attrflag; 1228 int v3 = NFS_ISV3(vp); 1229 1230 #ifndef nolint 1231 eof = 0; 1232 #endif 1233 nmp = VFSTONFS(vp->v_mount); 1234 tsiz = uiop->uio_resid; 1235 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1236 return (EFBIG); 1237 while (tsiz > 0) { 1238 nfsstats.rpccnt[NFSPROC_READ]++; 1239 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz; 1240 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); 1241 nfsm_fhtom(vp, v3); 1242 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3); 1243 if (v3) { 1244 txdr_hyper(uiop->uio_offset, tl); 1245 *(tl + 2) = txdr_unsigned(len); 1246 } else { 1247 *tl++ = txdr_unsigned(uiop->uio_offset); 1248 *tl++ = txdr_unsigned(len); 1249 *tl = 0; 1250 } 1251 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 1252 if (v3) { 1253 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1254 if (error) { 1255 m_freem(mrep); 1256 goto nfsmout; 1257 } 1258 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1259 eof = fxdr_unsigned(int, *(tl + 1)); 1260 } else 1261 nfsm_loadattr(vp, (struct vattr *)0); 1262 nfsm_strsiz(retlen, nmp->nm_rsize); 1263 nfsm_mtouio(uiop, retlen); 1264 m_freem(mrep); 1265 tsiz -= retlen; 1266 if (v3) { 1267 if (eof || retlen == 0) { 1268 tsiz = 0; 1269 } 1270 } else if (retlen < len) { 1271 tsiz = 0; 1272 } 1273 } 1274 nfsmout: 1275 return (error); 1276 } 1277 1278 /* 1279 * nfs write call 1280 */ 1281 int 1282 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit) 1283 { 1284 u_int32_t *tl; 1285 caddr_t cp; 1286 int32_t t1, t2, backup; 1287 caddr_t bpos, dpos, cp2; 1288 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1289 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1290 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; 1291 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; 1292 1293 #ifndef DIAGNOSTIC 1294 if (uiop->uio_iovcnt != 1) 1295 panic("nfs: writerpc iovcnt > 1"); 1296 #endif 1297 *must_commit = 0; 1298 tsiz = uiop->uio_resid; 1299 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) 1300 return (EFBIG); 1301 while (tsiz > 0) { 1302 nfsstats.rpccnt[NFSPROC_WRITE]++; 1303 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz; 1304 nfsm_reqhead(vp, NFSPROC_WRITE, 1305 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); 1306 nfsm_fhtom(vp, v3); 1307 if (v3) { 1308 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1309 txdr_hyper(uiop->uio_offset, tl); 1310 tl += 2; 1311 *tl++ = txdr_unsigned(len); 1312 *tl++ = txdr_unsigned(*iomode); 1313 *tl = txdr_unsigned(len); 1314 } else { 1315 u_int32_t x; 1316 1317 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1318 /* Set both "begin" and "current" to non-garbage. */ 1319 x = txdr_unsigned((u_int32_t)uiop->uio_offset); 1320 *tl++ = x; /* "begin offset" */ 1321 *tl++ = x; /* "current offset" */ 1322 x = txdr_unsigned(len); 1323 *tl++ = x; /* total to this offset */ 1324 *tl = x; /* size of this write */ 1325 } 1326 nfsm_uiotom(uiop, len); 1327 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE)); 1328 if (v3) { 1329 /* 1330 * The write RPC returns a before and after mtime. The 1331 * nfsm_wcc_data() macro checks the before n_mtime 1332 * against the before time and stores the after time 1333 * in the nfsnode's cached vattr and n_mtime field. 1334 * The NRMODIFIED bit will be set if the before 1335 * time did not match the original mtime. 1336 */ 1337 wccflag = NFSV3_WCCCHK; 1338 nfsm_wcc_data(vp, wccflag); 1339 if (!error) { 1340 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED 1341 + NFSX_V3WRITEVERF); 1342 rlen = fxdr_unsigned(int, *tl++); 1343 if (rlen == 0) { 1344 error = NFSERR_IO; 1345 m_freem(mrep); 1346 break; 1347 } else if (rlen < len) { 1348 backup = len - rlen; 1349 uiop->uio_iov->iov_base -= backup; 1350 uiop->uio_iov->iov_len += backup; 1351 uiop->uio_offset -= backup; 1352 uiop->uio_resid += backup; 1353 len = rlen; 1354 } 1355 commit = fxdr_unsigned(int, *tl++); 1356 1357 /* 1358 * Return the lowest committment level 1359 * obtained by any of the RPCs. 1360 */ 1361 if (committed == NFSV3WRITE_FILESYNC) 1362 committed = commit; 1363 else if (committed == NFSV3WRITE_DATASYNC && 1364 commit == NFSV3WRITE_UNSTABLE) 1365 committed = commit; 1366 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ 1367 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1368 NFSX_V3WRITEVERF); 1369 nmp->nm_state |= NFSSTA_HASWRITEVERF; 1370 } else if (bcmp((caddr_t)tl, 1371 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { 1372 *must_commit = 1; 1373 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 1374 NFSX_V3WRITEVERF); 1375 } 1376 } 1377 } else { 1378 nfsm_loadattr(vp, (struct vattr *)0); 1379 } 1380 m_freem(mrep); 1381 if (error) 1382 break; 1383 tsiz -= len; 1384 } 1385 nfsmout: 1386 if (vp->v_mount->mnt_flag & MNT_ASYNC) 1387 committed = NFSV3WRITE_FILESYNC; 1388 *iomode = committed; 1389 if (error) 1390 uiop->uio_resid = tsiz; 1391 return (error); 1392 } 1393 1394 /* 1395 * nfs mknod rpc 1396 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the 1397 * mode set to specify the file type and the size field for rdev. 1398 */ 1399 static int 1400 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1401 struct vattr *vap) 1402 { 1403 struct nfsv2_sattr *sp; 1404 u_int32_t *tl; 1405 caddr_t cp; 1406 int32_t t1, t2; 1407 struct vnode *newvp = (struct vnode *)0; 1408 struct nfsnode *np = (struct nfsnode *)0; 1409 struct vattr vattr; 1410 char *cp2; 1411 caddr_t bpos, dpos; 1412 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; 1413 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1414 u_int32_t rdev; 1415 int v3 = NFS_ISV3(dvp); 1416 1417 if (vap->va_type == VCHR || vap->va_type == VBLK) 1418 rdev = txdr_unsigned(vap->va_rdev); 1419 else if (vap->va_type == VFIFO || vap->va_type == VSOCK) 1420 rdev = nfs_xdrneg1; 1421 else { 1422 return (EOPNOTSUPP); 1423 } 1424 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1425 return (error); 1426 } 1427 nfsstats.rpccnt[NFSPROC_MKNOD]++; 1428 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + 1429 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1430 nfsm_fhtom(dvp, v3); 1431 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1432 if (v3) { 1433 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1434 *tl++ = vtonfsv3_type(vap->va_type); 1435 nfsm_v3attrbuild(vap, FALSE); 1436 if (vap->va_type == VCHR || vap->va_type == VBLK) { 1437 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1438 *tl++ = txdr_unsigned(umajor(vap->va_rdev)); 1439 *tl = txdr_unsigned(uminor(vap->va_rdev)); 1440 } 1441 } else { 1442 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1443 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1444 sp->sa_uid = nfs_xdrneg1; 1445 sp->sa_gid = nfs_xdrneg1; 1446 sp->sa_size = rdev; 1447 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1448 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1449 } 1450 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred); 1451 if (!error) { 1452 nfsm_mtofh(dvp, newvp, v3, gotvp); 1453 if (!gotvp) { 1454 if (newvp) { 1455 vput(newvp); 1456 newvp = (struct vnode *)0; 1457 } 1458 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1459 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1460 if (!error) 1461 newvp = NFSTOV(np); 1462 } 1463 } 1464 if (v3) 1465 nfsm_wcc_data(dvp, wccflag); 1466 m_freem(mrep); 1467 nfsmout: 1468 if (error) { 1469 if (newvp) 1470 vput(newvp); 1471 } else { 1472 *vpp = newvp; 1473 } 1474 VTONFS(dvp)->n_flag |= NLMODIFIED; 1475 if (!wccflag) 1476 VTONFS(dvp)->n_attrstamp = 0; 1477 return (error); 1478 } 1479 1480 /* 1481 * nfs mknod vop 1482 * just call nfs_mknodrpc() to do the work. 1483 * 1484 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp, 1485 * struct componentname *a_cnp, struct vattr *a_vap) 1486 */ 1487 /* ARGSUSED */ 1488 static int 1489 nfs_mknod(struct vop_old_mknod_args *ap) 1490 { 1491 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap); 1492 } 1493 1494 static u_long create_verf; 1495 /* 1496 * nfs file create call 1497 * 1498 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp, 1499 * struct componentname *a_cnp, struct vattr *a_vap) 1500 */ 1501 static int 1502 nfs_create(struct vop_old_create_args *ap) 1503 { 1504 struct vnode *dvp = ap->a_dvp; 1505 struct vattr *vap = ap->a_vap; 1506 struct componentname *cnp = ap->a_cnp; 1507 struct nfsv2_sattr *sp; 1508 u_int32_t *tl; 1509 caddr_t cp; 1510 int32_t t1, t2; 1511 struct nfsnode *np = (struct nfsnode *)0; 1512 struct vnode *newvp = (struct vnode *)0; 1513 caddr_t bpos, dpos, cp2; 1514 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; 1515 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1516 struct vattr vattr; 1517 int v3 = NFS_ISV3(dvp); 1518 1519 /* 1520 * Oops, not for me.. 1521 */ 1522 if (vap->va_type == VSOCK) 1523 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); 1524 1525 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 1526 return (error); 1527 } 1528 if (vap->va_vaflags & VA_EXCLUSIVE) 1529 fmode |= O_EXCL; 1530 again: 1531 nfsstats.rpccnt[NFSPROC_CREATE]++; 1532 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + 1533 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); 1534 nfsm_fhtom(dvp, v3); 1535 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1536 if (v3) { 1537 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 1538 if (fmode & O_EXCL) { 1539 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); 1540 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF); 1541 #ifdef INET 1542 if (!TAILQ_EMPTY(&in_ifaddrhead)) 1543 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr; 1544 else 1545 #endif 1546 *tl++ = create_verf; 1547 *tl = ++create_verf; 1548 } else { 1549 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); 1550 nfsm_v3attrbuild(vap, FALSE); 1551 } 1552 } else { 1553 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1554 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 1555 sp->sa_uid = nfs_xdrneg1; 1556 sp->sa_gid = nfs_xdrneg1; 1557 sp->sa_size = 0; 1558 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1559 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1560 } 1561 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred); 1562 if (!error) { 1563 nfsm_mtofh(dvp, newvp, v3, gotvp); 1564 if (!gotvp) { 1565 if (newvp) { 1566 vput(newvp); 1567 newvp = (struct vnode *)0; 1568 } 1569 error = nfs_lookitup(dvp, cnp->cn_nameptr, 1570 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np); 1571 if (!error) 1572 newvp = NFSTOV(np); 1573 } 1574 } 1575 if (v3) 1576 nfsm_wcc_data(dvp, wccflag); 1577 m_freem(mrep); 1578 nfsmout: 1579 if (error) { 1580 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { 1581 fmode &= ~O_EXCL; 1582 goto again; 1583 } 1584 if (newvp) 1585 vput(newvp); 1586 } else if (v3 && (fmode & O_EXCL)) { 1587 /* 1588 * We are normally called with only a partially initialized 1589 * VAP. Since the NFSv3 spec says that server may use the 1590 * file attributes to store the verifier, the spec requires 1591 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1592 * in atime, but we can't really assume that all servers will 1593 * so we ensure that our SETATTR sets both atime and mtime. 1594 */ 1595 if (vap->va_mtime.tv_sec == VNOVAL) 1596 vfs_timestamp(&vap->va_mtime); 1597 if (vap->va_atime.tv_sec == VNOVAL) 1598 vap->va_atime = vap->va_mtime; 1599 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td); 1600 } 1601 if (!error) { 1602 /* 1603 * The new np may have enough info for access 1604 * checks, make sure rucred and wucred are 1605 * initialized for read and write rpc's. 1606 */ 1607 np = VTONFS(newvp); 1608 if (np->n_rucred == NULL) 1609 np->n_rucred = crhold(cnp->cn_cred); 1610 if (np->n_wucred == NULL) 1611 np->n_wucred = crhold(cnp->cn_cred); 1612 *ap->a_vpp = newvp; 1613 } 1614 VTONFS(dvp)->n_flag |= NLMODIFIED; 1615 if (!wccflag) 1616 VTONFS(dvp)->n_attrstamp = 0; 1617 return (error); 1618 } 1619 1620 /* 1621 * nfs file remove call 1622 * To try and make nfs semantics closer to ufs semantics, a file that has 1623 * other processes using the vnode is renamed instead of removed and then 1624 * removed later on the last close. 1625 * - If v_usecount > 1 1626 * If a rename is not already in the works 1627 * call nfs_sillyrename() to set it up 1628 * else 1629 * do the remove rpc 1630 * 1631 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp, 1632 * struct vnode *a_vp, struct componentname *a_cnp) 1633 */ 1634 static int 1635 nfs_remove(struct vop_old_remove_args *ap) 1636 { 1637 struct vnode *vp = ap->a_vp; 1638 struct vnode *dvp = ap->a_dvp; 1639 struct componentname *cnp = ap->a_cnp; 1640 struct nfsnode *np = VTONFS(vp); 1641 int error = 0; 1642 struct vattr vattr; 1643 1644 #ifndef DIAGNOSTIC 1645 if (vp->v_usecount < 1) 1646 panic("nfs_remove: bad v_usecount"); 1647 #endif 1648 if (vp->v_type == VDIR) 1649 error = EPERM; 1650 else if (vp->v_usecount == 1 || (np->n_sillyrename && 1651 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 && 1652 vattr.va_nlink > 1)) { 1653 /* 1654 * throw away biocache buffers, mainly to avoid 1655 * unnecessary delayed writes later. 1656 */ 1657 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1); 1658 /* Do the rpc */ 1659 if (error != EINTR) 1660 error = nfs_removerpc(dvp, cnp->cn_nameptr, 1661 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td); 1662 /* 1663 * Kludge City: If the first reply to the remove rpc is lost.. 1664 * the reply to the retransmitted request will be ENOENT 1665 * since the file was in fact removed 1666 * Therefore, we cheat and return success. 1667 */ 1668 if (error == ENOENT) 1669 error = 0; 1670 } else if (!np->n_sillyrename) { 1671 error = nfs_sillyrename(dvp, vp, cnp); 1672 } 1673 np->n_attrstamp = 0; 1674 return (error); 1675 } 1676 1677 /* 1678 * nfs file remove rpc called from nfs_inactive 1679 */ 1680 int 1681 nfs_removeit(struct sillyrename *sp) 1682 { 1683 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, 1684 sp->s_cred, NULL)); 1685 } 1686 1687 /* 1688 * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). 1689 */ 1690 static int 1691 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, 1692 struct ucred *cred, struct thread *td) 1693 { 1694 u_int32_t *tl; 1695 caddr_t cp; 1696 int32_t t1, t2; 1697 caddr_t bpos, dpos, cp2; 1698 int error = 0, wccflag = NFSV3_WCCRATTR; 1699 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1700 int v3 = NFS_ISV3(dvp); 1701 1702 nfsstats.rpccnt[NFSPROC_REMOVE]++; 1703 nfsm_reqhead(dvp, NFSPROC_REMOVE, 1704 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); 1705 nfsm_fhtom(dvp, v3); 1706 nfsm_strtom(name, namelen, NFS_MAXNAMLEN); 1707 nfsm_request(dvp, NFSPROC_REMOVE, td, cred); 1708 if (v3) 1709 nfsm_wcc_data(dvp, wccflag); 1710 m_freem(mrep); 1711 nfsmout: 1712 VTONFS(dvp)->n_flag |= NLMODIFIED; 1713 if (!wccflag) 1714 VTONFS(dvp)->n_attrstamp = 0; 1715 return (error); 1716 } 1717 1718 /* 1719 * nfs file rename call 1720 * 1721 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 1722 * struct componentname *a_fcnp, struct vnode *a_tdvp, 1723 * struct vnode *a_tvp, struct componentname *a_tcnp) 1724 */ 1725 static int 1726 nfs_rename(struct vop_old_rename_args *ap) 1727 { 1728 struct vnode *fvp = ap->a_fvp; 1729 struct vnode *tvp = ap->a_tvp; 1730 struct vnode *fdvp = ap->a_fdvp; 1731 struct vnode *tdvp = ap->a_tdvp; 1732 struct componentname *tcnp = ap->a_tcnp; 1733 struct componentname *fcnp = ap->a_fcnp; 1734 int error; 1735 1736 /* Check for cross-device rename */ 1737 if ((fvp->v_mount != tdvp->v_mount) || 1738 (tvp && (fvp->v_mount != tvp->v_mount))) { 1739 error = EXDEV; 1740 goto out; 1741 } 1742 1743 /* 1744 * We have to flush B_DELWRI data prior to renaming 1745 * the file. If we don't, the delayed-write buffers 1746 * can be flushed out later after the file has gone stale 1747 * under NFSV3. NFSV2 does not have this problem because 1748 * ( as far as I can tell ) it flushes dirty buffers more 1749 * often. 1750 */ 1751 1752 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td); 1753 if (tvp) 1754 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td); 1755 1756 /* 1757 * If the tvp exists and is in use, sillyrename it before doing the 1758 * rename of the new file over it. 1759 * 1760 * XXX Can't sillyrename a directory. 1761 * 1762 * We do not attempt to do any namecache purges in this old API 1763 * routine. The new API compat functions have access to the actual 1764 * namecache structures and will do it for us. 1765 */ 1766 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename && 1767 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { 1768 vput(tvp); 1769 tvp = NULL; 1770 } else if (tvp) { 1771 ; 1772 } 1773 1774 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, 1775 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, 1776 tcnp->cn_td); 1777 1778 out: 1779 if (tdvp == tvp) 1780 vrele(tdvp); 1781 else 1782 vput(tdvp); 1783 if (tvp) 1784 vput(tvp); 1785 vrele(fdvp); 1786 vrele(fvp); 1787 /* 1788 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. 1789 */ 1790 if (error == ENOENT) 1791 error = 0; 1792 return (error); 1793 } 1794 1795 /* 1796 * nfs file rename rpc called from nfs_remove() above 1797 */ 1798 static int 1799 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, 1800 struct sillyrename *sp) 1801 { 1802 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, 1803 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td)); 1804 } 1805 1806 /* 1807 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). 1808 */ 1809 static int 1810 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, 1811 struct vnode *tdvp, const char *tnameptr, int tnamelen, 1812 struct ucred *cred, struct thread *td) 1813 { 1814 u_int32_t *tl; 1815 caddr_t cp; 1816 int32_t t1, t2; 1817 caddr_t bpos, dpos, cp2; 1818 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; 1819 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1820 int v3 = NFS_ISV3(fdvp); 1821 1822 nfsstats.rpccnt[NFSPROC_RENAME]++; 1823 nfsm_reqhead(fdvp, NFSPROC_RENAME, 1824 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + 1825 nfsm_rndup(tnamelen)); 1826 nfsm_fhtom(fdvp, v3); 1827 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); 1828 nfsm_fhtom(tdvp, v3); 1829 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); 1830 nfsm_request(fdvp, NFSPROC_RENAME, td, cred); 1831 if (v3) { 1832 nfsm_wcc_data(fdvp, fwccflag); 1833 nfsm_wcc_data(tdvp, twccflag); 1834 } 1835 m_freem(mrep); 1836 nfsmout: 1837 VTONFS(fdvp)->n_flag |= NLMODIFIED; 1838 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1839 if (!fwccflag) 1840 VTONFS(fdvp)->n_attrstamp = 0; 1841 if (!twccflag) 1842 VTONFS(tdvp)->n_attrstamp = 0; 1843 return (error); 1844 } 1845 1846 /* 1847 * nfs hard link create call 1848 * 1849 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp, 1850 * struct componentname *a_cnp) 1851 */ 1852 static int 1853 nfs_link(struct vop_old_link_args *ap) 1854 { 1855 struct vnode *vp = ap->a_vp; 1856 struct vnode *tdvp = ap->a_tdvp; 1857 struct componentname *cnp = ap->a_cnp; 1858 u_int32_t *tl; 1859 caddr_t cp; 1860 int32_t t1, t2; 1861 caddr_t bpos, dpos, cp2; 1862 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; 1863 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1864 int v3; 1865 1866 if (vp->v_mount != tdvp->v_mount) { 1867 return (EXDEV); 1868 } 1869 1870 /* 1871 * Push all writes to the server, so that the attribute cache 1872 * doesn't get "out of sync" with the server. 1873 * XXX There should be a better way! 1874 */ 1875 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td); 1876 1877 v3 = NFS_ISV3(vp); 1878 nfsstats.rpccnt[NFSPROC_LINK]++; 1879 nfsm_reqhead(vp, NFSPROC_LINK, 1880 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 1881 nfsm_fhtom(vp, v3); 1882 nfsm_fhtom(tdvp, v3); 1883 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1884 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred); 1885 if (v3) { 1886 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 1887 nfsm_wcc_data(tdvp, wccflag); 1888 } 1889 m_freem(mrep); 1890 nfsmout: 1891 VTONFS(tdvp)->n_flag |= NLMODIFIED; 1892 if (!attrflag) 1893 VTONFS(vp)->n_attrstamp = 0; 1894 if (!wccflag) 1895 VTONFS(tdvp)->n_attrstamp = 0; 1896 /* 1897 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. 1898 */ 1899 if (error == EEXIST) 1900 error = 0; 1901 return (error); 1902 } 1903 1904 /* 1905 * nfs symbolic link create call 1906 * 1907 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp, 1908 * struct componentname *a_cnp, struct vattr *a_vap, 1909 * char *a_target) 1910 */ 1911 static int 1912 nfs_symlink(struct vop_old_symlink_args *ap) 1913 { 1914 struct vnode *dvp = ap->a_dvp; 1915 struct vattr *vap = ap->a_vap; 1916 struct componentname *cnp = ap->a_cnp; 1917 struct nfsv2_sattr *sp; 1918 u_int32_t *tl; 1919 caddr_t cp; 1920 int32_t t1, t2; 1921 caddr_t bpos, dpos, cp2; 1922 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; 1923 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 1924 struct vnode *newvp = (struct vnode *)0; 1925 int v3 = NFS_ISV3(dvp); 1926 1927 nfsstats.rpccnt[NFSPROC_SYMLINK]++; 1928 slen = strlen(ap->a_target); 1929 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + 1930 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); 1931 nfsm_fhtom(dvp, v3); 1932 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 1933 if (v3) { 1934 nfsm_v3attrbuild(vap, FALSE); 1935 } 1936 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); 1937 if (!v3) { 1938 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1939 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); 1940 sp->sa_uid = nfs_xdrneg1; 1941 sp->sa_gid = nfs_xdrneg1; 1942 sp->sa_size = nfs_xdrneg1; 1943 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 1944 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 1945 } 1946 1947 /* 1948 * Issue the NFS request and get the rpc response. 1949 * 1950 * Only NFSv3 responses returning an error of 0 actually return 1951 * a file handle that can be converted into newvp without having 1952 * to do an extra lookup rpc. 1953 */ 1954 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred); 1955 if (v3) { 1956 if (error == 0) 1957 nfsm_mtofh(dvp, newvp, v3, gotvp); 1958 nfsm_wcc_data(dvp, wccflag); 1959 } 1960 1961 /* 1962 * out code jumps -> here, mrep is also freed. 1963 */ 1964 1965 m_freem(mrep); 1966 nfsmout: 1967 1968 /* 1969 * If we get an EEXIST error, silently convert it to no-error 1970 * in case of an NFS retry. 1971 */ 1972 if (error == EEXIST) 1973 error = 0; 1974 1975 /* 1976 * If we do not have (or no longer have) an error, and we could 1977 * not extract the newvp from the response due to the request being 1978 * NFSv2 or the error being EEXIST. We have to do a lookup in order 1979 * to obtain a newvp to return. 1980 */ 1981 if (error == 0 && newvp == NULL) { 1982 struct nfsnode *np = NULL; 1983 1984 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 1985 cnp->cn_cred, cnp->cn_td, &np); 1986 if (!error) 1987 newvp = NFSTOV(np); 1988 } 1989 if (error) { 1990 if (newvp) 1991 vput(newvp); 1992 } else { 1993 *ap->a_vpp = newvp; 1994 } 1995 VTONFS(dvp)->n_flag |= NLMODIFIED; 1996 if (!wccflag) 1997 VTONFS(dvp)->n_attrstamp = 0; 1998 return (error); 1999 } 2000 2001 /* 2002 * nfs make dir call 2003 * 2004 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp, 2005 * struct componentname *a_cnp, struct vattr *a_vap) 2006 */ 2007 static int 2008 nfs_mkdir(struct vop_old_mkdir_args *ap) 2009 { 2010 struct vnode *dvp = ap->a_dvp; 2011 struct vattr *vap = ap->a_vap; 2012 struct componentname *cnp = ap->a_cnp; 2013 struct nfsv2_sattr *sp; 2014 u_int32_t *tl; 2015 caddr_t cp; 2016 int32_t t1, t2; 2017 int len; 2018 struct nfsnode *np = (struct nfsnode *)0; 2019 struct vnode *newvp = (struct vnode *)0; 2020 caddr_t bpos, dpos, cp2; 2021 int error = 0, wccflag = NFSV3_WCCRATTR; 2022 int gotvp = 0; 2023 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2024 struct vattr vattr; 2025 int v3 = NFS_ISV3(dvp); 2026 2027 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) { 2028 return (error); 2029 } 2030 len = cnp->cn_namelen; 2031 nfsstats.rpccnt[NFSPROC_MKDIR]++; 2032 nfsm_reqhead(dvp, NFSPROC_MKDIR, 2033 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); 2034 nfsm_fhtom(dvp, v3); 2035 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); 2036 if (v3) { 2037 nfsm_v3attrbuild(vap, FALSE); 2038 } else { 2039 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2040 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); 2041 sp->sa_uid = nfs_xdrneg1; 2042 sp->sa_gid = nfs_xdrneg1; 2043 sp->sa_size = nfs_xdrneg1; 2044 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); 2045 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); 2046 } 2047 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred); 2048 if (!error) 2049 nfsm_mtofh(dvp, newvp, v3, gotvp); 2050 if (v3) 2051 nfsm_wcc_data(dvp, wccflag); 2052 m_freem(mrep); 2053 nfsmout: 2054 VTONFS(dvp)->n_flag |= NLMODIFIED; 2055 if (!wccflag) 2056 VTONFS(dvp)->n_attrstamp = 0; 2057 /* 2058 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry 2059 * if we can succeed in looking up the directory. 2060 */ 2061 if (error == EEXIST || (!error && !gotvp)) { 2062 if (newvp) { 2063 vrele(newvp); 2064 newvp = (struct vnode *)0; 2065 } 2066 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, 2067 cnp->cn_td, &np); 2068 if (!error) { 2069 newvp = NFSTOV(np); 2070 if (newvp->v_type != VDIR) 2071 error = EEXIST; 2072 } 2073 } 2074 if (error) { 2075 if (newvp) 2076 vrele(newvp); 2077 } else 2078 *ap->a_vpp = newvp; 2079 return (error); 2080 } 2081 2082 /* 2083 * nfs remove directory call 2084 * 2085 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp, 2086 * struct componentname *a_cnp) 2087 */ 2088 static int 2089 nfs_rmdir(struct vop_old_rmdir_args *ap) 2090 { 2091 struct vnode *vp = ap->a_vp; 2092 struct vnode *dvp = ap->a_dvp; 2093 struct componentname *cnp = ap->a_cnp; 2094 u_int32_t *tl; 2095 caddr_t cp; 2096 int32_t t1, t2; 2097 caddr_t bpos, dpos, cp2; 2098 int error = 0, wccflag = NFSV3_WCCRATTR; 2099 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2100 int v3 = NFS_ISV3(dvp); 2101 2102 if (dvp == vp) 2103 return (EINVAL); 2104 nfsstats.rpccnt[NFSPROC_RMDIR]++; 2105 nfsm_reqhead(dvp, NFSPROC_RMDIR, 2106 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); 2107 nfsm_fhtom(dvp, v3); 2108 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); 2109 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred); 2110 if (v3) 2111 nfsm_wcc_data(dvp, wccflag); 2112 m_freem(mrep); 2113 nfsmout: 2114 VTONFS(dvp)->n_flag |= NLMODIFIED; 2115 if (!wccflag) 2116 VTONFS(dvp)->n_attrstamp = 0; 2117 /* 2118 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. 2119 */ 2120 if (error == ENOENT) 2121 error = 0; 2122 return (error); 2123 } 2124 2125 /* 2126 * nfs readdir call 2127 * 2128 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred) 2129 */ 2130 static int 2131 nfs_readdir(struct vop_readdir_args *ap) 2132 { 2133 struct vnode *vp = ap->a_vp; 2134 struct nfsnode *np = VTONFS(vp); 2135 struct uio *uio = ap->a_uio; 2136 int tresid, error; 2137 struct vattr vattr; 2138 2139 if (vp->v_type != VDIR) 2140 return (EPERM); 2141 2142 /* 2143 * If we have a valid EOF offset cache we must call VOP_GETATTR() 2144 * and then check that is still valid, or if this is an NQNFS mount 2145 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that 2146 * VOP_GETATTR() does not necessarily go to the wire. 2147 */ 2148 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && 2149 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) { 2150 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) { 2151 if (NQNFS_CKCACHABLE(vp, ND_READ)) { 2152 nfsstats.direofcache_hits++; 2153 return (0); 2154 } 2155 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 && 2156 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0 2157 ) { 2158 nfsstats.direofcache_hits++; 2159 return (0); 2160 } 2161 } 2162 2163 /* 2164 * Call nfs_bioread() to do the real work. nfs_bioread() does its 2165 * own cache coherency checks so we do not have to. 2166 */ 2167 tresid = uio->uio_resid; 2168 error = nfs_bioread(vp, uio, 0); 2169 2170 if (!error && uio->uio_resid == tresid) 2171 nfsstats.direofcache_misses++; 2172 return (error); 2173 } 2174 2175 /* 2176 * Readdir rpc call. 2177 * Called from below the buffer cache by nfs_doio(). 2178 */ 2179 int 2180 nfs_readdirrpc(struct vnode *vp, struct uio *uiop) 2181 { 2182 int len, left; 2183 struct nfs_dirent *dp = NULL; 2184 u_int32_t *tl; 2185 caddr_t cp; 2186 int32_t t1, t2; 2187 nfsuint64 *cookiep; 2188 caddr_t bpos, dpos, cp2; 2189 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2190 nfsuint64 cookie; 2191 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2192 struct nfsnode *dnp = VTONFS(vp); 2193 u_quad_t fileno; 2194 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; 2195 int attrflag; 2196 int v3 = NFS_ISV3(vp); 2197 2198 #ifndef DIAGNOSTIC 2199 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2200 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2201 panic("nfs readdirrpc bad uio"); 2202 #endif 2203 2204 /* 2205 * If there is no cookie, assume directory was stale. 2206 */ 2207 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2208 if (cookiep) 2209 cookie = *cookiep; 2210 else 2211 return (NFSERR_BAD_COOKIE); 2212 /* 2213 * Loop around doing readdir rpc's of size nm_readdirsize 2214 * truncated to a multiple of DIRBLKSIZ. 2215 * The stopping criteria is EOF or buffer full. 2216 */ 2217 while (more_dirs && bigenough) { 2218 nfsstats.rpccnt[NFSPROC_READDIR]++; 2219 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + 2220 NFSX_READDIR(v3)); 2221 nfsm_fhtom(vp, v3); 2222 if (v3) { 2223 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2224 *tl++ = cookie.nfsuquad[0]; 2225 *tl++ = cookie.nfsuquad[1]; 2226 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2227 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2228 } else { 2229 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2230 *tl++ = cookie.nfsuquad[0]; 2231 } 2232 *tl = txdr_unsigned(nmp->nm_readdirsize); 2233 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2234 if (v3) { 2235 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2236 if (!error) { 2237 nfsm_dissect(tl, u_int32_t *, 2238 2 * NFSX_UNSIGNED); 2239 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2240 dnp->n_cookieverf.nfsuquad[1] = *tl; 2241 } else { 2242 m_freem(mrep); 2243 goto nfsmout; 2244 } 2245 } 2246 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2247 more_dirs = fxdr_unsigned(int, *tl); 2248 2249 /* loop thru the dir entries, doctoring them to 4bsd form */ 2250 while (more_dirs && bigenough) { 2251 if (v3) { 2252 nfsm_dissect(tl, u_int32_t *, 2253 3 * NFSX_UNSIGNED); 2254 fileno = fxdr_hyper(tl); 2255 len = fxdr_unsigned(int, *(tl + 2)); 2256 } else { 2257 nfsm_dissect(tl, u_int32_t *, 2258 2 * NFSX_UNSIGNED); 2259 fileno = fxdr_unsigned(u_quad_t, *tl++); 2260 len = fxdr_unsigned(int, *tl); 2261 } 2262 if (len <= 0 || len > NFS_MAXNAMLEN) { 2263 error = EBADRPC; 2264 m_freem(mrep); 2265 goto nfsmout; 2266 } 2267 tlen = nfsm_rndup(len); 2268 if (tlen == len) 2269 tlen += 4; /* To ensure null termination */ 2270 left = DIRBLKSIZ - blksiz; 2271 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2272 dp->nfs_reclen += left; 2273 uiop->uio_iov->iov_base += left; 2274 uiop->uio_iov->iov_len -= left; 2275 uiop->uio_offset += left; 2276 uiop->uio_resid -= left; 2277 blksiz = 0; 2278 } 2279 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2280 bigenough = 0; 2281 if (bigenough) { 2282 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2283 dp->nfs_ino = fileno; 2284 dp->nfs_namlen = len; 2285 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2286 dp->nfs_type = DT_UNKNOWN; 2287 blksiz += dp->nfs_reclen; 2288 if (blksiz == DIRBLKSIZ) 2289 blksiz = 0; 2290 uiop->uio_offset += sizeof(struct nfs_dirent); 2291 uiop->uio_resid -= sizeof(struct nfs_dirent); 2292 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2293 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2294 nfsm_mtouio(uiop, len); 2295 cp = uiop->uio_iov->iov_base; 2296 tlen -= len; 2297 *cp = '\0'; /* null terminate */ 2298 uiop->uio_iov->iov_base += tlen; 2299 uiop->uio_iov->iov_len -= tlen; 2300 uiop->uio_offset += tlen; 2301 uiop->uio_resid -= tlen; 2302 } else 2303 nfsm_adv(nfsm_rndup(len)); 2304 if (v3) { 2305 nfsm_dissect(tl, u_int32_t *, 2306 3 * NFSX_UNSIGNED); 2307 } else { 2308 nfsm_dissect(tl, u_int32_t *, 2309 2 * NFSX_UNSIGNED); 2310 } 2311 if (bigenough) { 2312 cookie.nfsuquad[0] = *tl++; 2313 if (v3) 2314 cookie.nfsuquad[1] = *tl++; 2315 } else if (v3) 2316 tl += 2; 2317 else 2318 tl++; 2319 more_dirs = fxdr_unsigned(int, *tl); 2320 } 2321 /* 2322 * If at end of rpc data, get the eof boolean 2323 */ 2324 if (!more_dirs) { 2325 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2326 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2327 } 2328 m_freem(mrep); 2329 } 2330 /* 2331 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2332 * by increasing d_reclen for the last record. 2333 */ 2334 if (blksiz > 0) { 2335 left = DIRBLKSIZ - blksiz; 2336 dp->nfs_reclen += left; 2337 uiop->uio_iov->iov_base += left; 2338 uiop->uio_iov->iov_len -= left; 2339 uiop->uio_offset += left; 2340 uiop->uio_resid -= left; 2341 } 2342 2343 /* 2344 * We are now either at the end of the directory or have filled the 2345 * block. 2346 */ 2347 if (bigenough) 2348 dnp->n_direofoffset = uiop->uio_offset; 2349 else { 2350 if (uiop->uio_resid > 0) 2351 printf("EEK! readdirrpc resid > 0\n"); 2352 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2353 *cookiep = cookie; 2354 } 2355 nfsmout: 2356 return (error); 2357 } 2358 2359 /* 2360 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). 2361 */ 2362 int 2363 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop) 2364 { 2365 int len, left; 2366 struct nfs_dirent *dp; 2367 u_int32_t *tl; 2368 caddr_t cp; 2369 int32_t t1, t2; 2370 struct vnode *newvp; 2371 nfsuint64 *cookiep; 2372 caddr_t bpos, dpos, cp2, dpossav1, dpossav2; 2373 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2; 2374 nfsuint64 cookie; 2375 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2376 struct nfsnode *dnp = VTONFS(vp), *np; 2377 nfsfh_t *fhp; 2378 u_quad_t fileno; 2379 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; 2380 int attrflag, fhsize; 2381 struct namecache *ncp; 2382 struct namecache *dncp; 2383 struct nlcomponent nlc; 2384 2385 #ifndef nolint 2386 dp = NULL; 2387 #endif 2388 #ifndef DIAGNOSTIC 2389 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || 2390 (uiop->uio_resid & (DIRBLKSIZ - 1))) 2391 panic("nfs readdirplusrpc bad uio"); 2392 #endif 2393 /* 2394 * Obtain the namecache record for the directory so we have something 2395 * to use as a basis for creating the entries. This function will 2396 * return a held (but not locked) ncp. The ncp may be disconnected 2397 * from the tree and cannot be used for upward traversals, and the 2398 * ncp may be unnamed. Note that other unrelated operations may 2399 * cause the ncp to be named at any time. 2400 */ 2401 dncp = cache_fromdvp(vp, NULL, 0); 2402 bzero(&nlc, sizeof(nlc)); 2403 newvp = NULLVP; 2404 2405 /* 2406 * If there is no cookie, assume directory was stale. 2407 */ 2408 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); 2409 if (cookiep) 2410 cookie = *cookiep; 2411 else 2412 return (NFSERR_BAD_COOKIE); 2413 /* 2414 * Loop around doing readdir rpc's of size nm_readdirsize 2415 * truncated to a multiple of DIRBLKSIZ. 2416 * The stopping criteria is EOF or buffer full. 2417 */ 2418 while (more_dirs && bigenough) { 2419 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; 2420 nfsm_reqhead(vp, NFSPROC_READDIRPLUS, 2421 NFSX_FH(1) + 6 * NFSX_UNSIGNED); 2422 nfsm_fhtom(vp, 1); 2423 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2424 *tl++ = cookie.nfsuquad[0]; 2425 *tl++ = cookie.nfsuquad[1]; 2426 *tl++ = dnp->n_cookieverf.nfsuquad[0]; 2427 *tl++ = dnp->n_cookieverf.nfsuquad[1]; 2428 *tl++ = txdr_unsigned(nmp->nm_readdirsize); 2429 *tl = txdr_unsigned(nmp->nm_rsize); 2430 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ)); 2431 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK); 2432 if (error) { 2433 m_freem(mrep); 2434 goto nfsmout; 2435 } 2436 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2437 dnp->n_cookieverf.nfsuquad[0] = *tl++; 2438 dnp->n_cookieverf.nfsuquad[1] = *tl++; 2439 more_dirs = fxdr_unsigned(int, *tl); 2440 2441 /* loop thru the dir entries, doctoring them to 4bsd form */ 2442 while (more_dirs && bigenough) { 2443 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2444 fileno = fxdr_hyper(tl); 2445 len = fxdr_unsigned(int, *(tl + 2)); 2446 if (len <= 0 || len > NFS_MAXNAMLEN) { 2447 error = EBADRPC; 2448 m_freem(mrep); 2449 goto nfsmout; 2450 } 2451 tlen = nfsm_rndup(len); 2452 if (tlen == len) 2453 tlen += 4; /* To ensure null termination*/ 2454 left = DIRBLKSIZ - blksiz; 2455 if ((tlen + sizeof(struct nfs_dirent)) > left) { 2456 dp->nfs_reclen += left; 2457 uiop->uio_iov->iov_base += left; 2458 uiop->uio_iov->iov_len -= left; 2459 uiop->uio_offset += left; 2460 uiop->uio_resid -= left; 2461 blksiz = 0; 2462 } 2463 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid) 2464 bigenough = 0; 2465 if (bigenough) { 2466 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base; 2467 dp->nfs_ino = fileno; 2468 dp->nfs_namlen = len; 2469 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent); 2470 dp->nfs_type = DT_UNKNOWN; 2471 blksiz += dp->nfs_reclen; 2472 if (blksiz == DIRBLKSIZ) 2473 blksiz = 0; 2474 uiop->uio_offset += sizeof(struct nfs_dirent); 2475 uiop->uio_resid -= sizeof(struct nfs_dirent); 2476 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent); 2477 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent); 2478 nlc.nlc_nameptr = uiop->uio_iov->iov_base; 2479 nlc.nlc_namelen = len; 2480 nfsm_mtouio(uiop, len); 2481 cp = uiop->uio_iov->iov_base; 2482 tlen -= len; 2483 *cp = '\0'; 2484 uiop->uio_iov->iov_base += tlen; 2485 uiop->uio_iov->iov_len -= tlen; 2486 uiop->uio_offset += tlen; 2487 uiop->uio_resid -= tlen; 2488 } else 2489 nfsm_adv(nfsm_rndup(len)); 2490 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2491 if (bigenough) { 2492 cookie.nfsuquad[0] = *tl++; 2493 cookie.nfsuquad[1] = *tl++; 2494 } else 2495 tl += 2; 2496 2497 /* 2498 * Since the attributes are before the file handle 2499 * (sigh), we must skip over the attributes and then 2500 * come back and get them. 2501 */ 2502 attrflag = fxdr_unsigned(int, *tl); 2503 if (attrflag) { 2504 dpossav1 = dpos; 2505 mdsav1 = md; 2506 nfsm_adv(NFSX_V3FATTR); 2507 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2508 doit = fxdr_unsigned(int, *tl); 2509 if (doit) { 2510 nfsm_getfh(fhp, fhsize, 1); 2511 if (NFS_CMPFH(dnp, fhp, fhsize)) { 2512 vref(vp); 2513 newvp = vp; 2514 np = dnp; 2515 } else { 2516 error = nfs_nget(vp->v_mount, fhp, 2517 fhsize, &np); 2518 if (error) 2519 doit = 0; 2520 else 2521 newvp = NFSTOV(np); 2522 } 2523 } 2524 if (doit && bigenough) { 2525 dpossav2 = dpos; 2526 dpos = dpossav1; 2527 mdsav2 = md; 2528 md = mdsav1; 2529 nfsm_loadattr(newvp, (struct vattr *)0); 2530 dpos = dpossav2; 2531 md = mdsav2; 2532 dp->nfs_type = 2533 IFTODT(VTTOIF(np->n_vattr.va_type)); 2534 if (dncp) { 2535 printf("NFS/READDIRPLUS, ENTER %*.*s\n", 2536 nlc.nlc_namelen, nlc.nlc_namelen, 2537 nlc.nlc_nameptr); 2538 ncp = cache_nlookup(dncp, &nlc); 2539 cache_setunresolved(ncp); 2540 cache_setvp(ncp, newvp); 2541 cache_put(ncp); 2542 } else { 2543 printf("NFS/READDIRPLUS, UNABLE TO ENTER" 2544 " %*.*s\n", 2545 nlc.nlc_namelen, nlc.nlc_namelen, 2546 nlc.nlc_nameptr); 2547 } 2548 } 2549 } else { 2550 /* Just skip over the file handle */ 2551 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2552 i = fxdr_unsigned(int, *tl); 2553 nfsm_adv(nfsm_rndup(i)); 2554 } 2555 if (newvp != NULLVP) { 2556 if (newvp == vp) 2557 vrele(newvp); 2558 else 2559 vput(newvp); 2560 newvp = NULLVP; 2561 } 2562 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2563 more_dirs = fxdr_unsigned(int, *tl); 2564 } 2565 /* 2566 * If at end of rpc data, get the eof boolean 2567 */ 2568 if (!more_dirs) { 2569 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 2570 more_dirs = (fxdr_unsigned(int, *tl) == 0); 2571 } 2572 m_freem(mrep); 2573 } 2574 /* 2575 * Fill last record, iff any, out to a multiple of DIRBLKSIZ 2576 * by increasing d_reclen for the last record. 2577 */ 2578 if (blksiz > 0) { 2579 left = DIRBLKSIZ - blksiz; 2580 dp->nfs_reclen += left; 2581 uiop->uio_iov->iov_base += left; 2582 uiop->uio_iov->iov_len -= left; 2583 uiop->uio_offset += left; 2584 uiop->uio_resid -= left; 2585 } 2586 2587 /* 2588 * We are now either at the end of the directory or have filled the 2589 * block. 2590 */ 2591 if (bigenough) 2592 dnp->n_direofoffset = uiop->uio_offset; 2593 else { 2594 if (uiop->uio_resid > 0) 2595 printf("EEK! readdirplusrpc resid > 0\n"); 2596 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); 2597 *cookiep = cookie; 2598 } 2599 nfsmout: 2600 if (newvp != NULLVP) { 2601 if (newvp == vp) 2602 vrele(newvp); 2603 else 2604 vput(newvp); 2605 newvp = NULLVP; 2606 } 2607 if (dncp) 2608 cache_drop(dncp); 2609 return (error); 2610 } 2611 2612 /* 2613 * Silly rename. To make the NFS filesystem that is stateless look a little 2614 * more like the "ufs" a remove of an active vnode is translated to a rename 2615 * to a funny looking filename that is removed by nfs_inactive on the 2616 * nfsnode. There is the potential for another process on a different client 2617 * to create the same funny name between the nfs_lookitup() fails and the 2618 * nfs_rename() completes, but... 2619 */ 2620 static int 2621 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2622 { 2623 struct sillyrename *sp; 2624 struct nfsnode *np; 2625 int error; 2626 2627 /* 2628 * We previously purged dvp instead of vp. I don't know why, it 2629 * completely destroys performance. We can't do it anyway with the 2630 * new VFS API since we would be breaking the namecache topology. 2631 */ 2632 cache_purge(vp); /* XXX */ 2633 np = VTONFS(vp); 2634 #ifndef DIAGNOSTIC 2635 if (vp->v_type == VDIR) 2636 panic("nfs: sillyrename dir"); 2637 #endif 2638 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), 2639 M_NFSREQ, M_WAITOK); 2640 sp->s_cred = crdup(cnp->cn_cred); 2641 sp->s_dvp = dvp; 2642 vref(dvp); 2643 2644 /* Fudge together a funny name */ 2645 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td); 2646 2647 /* Try lookitups until we get one that isn't there */ 2648 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2649 cnp->cn_td, (struct nfsnode **)0) == 0) { 2650 sp->s_name[4]++; 2651 if (sp->s_name[4] > 'z') { 2652 error = EINVAL; 2653 goto bad; 2654 } 2655 } 2656 error = nfs_renameit(dvp, cnp, sp); 2657 if (error) 2658 goto bad; 2659 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, 2660 cnp->cn_td, &np); 2661 np->n_sillyrename = sp; 2662 return (0); 2663 bad: 2664 vrele(sp->s_dvp); 2665 crfree(sp->s_cred); 2666 free((caddr_t)sp, M_NFSREQ); 2667 return (error); 2668 } 2669 2670 /* 2671 * Look up a file name and optionally either update the file handle or 2672 * allocate an nfsnode, depending on the value of npp. 2673 * npp == NULL --> just do the lookup 2674 * *npp == NULL --> allocate a new nfsnode and make sure attributes are 2675 * handled too 2676 * *npp != NULL --> update the file handle in the vnode 2677 */ 2678 static int 2679 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, 2680 struct thread *td, struct nfsnode **npp) 2681 { 2682 u_int32_t *tl; 2683 caddr_t cp; 2684 int32_t t1, t2; 2685 struct vnode *newvp = (struct vnode *)0; 2686 struct nfsnode *np, *dnp = VTONFS(dvp); 2687 caddr_t bpos, dpos, cp2; 2688 int error = 0, fhlen, attrflag; 2689 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2690 nfsfh_t *nfhp; 2691 int v3 = NFS_ISV3(dvp); 2692 2693 nfsstats.rpccnt[NFSPROC_LOOKUP]++; 2694 nfsm_reqhead(dvp, NFSPROC_LOOKUP, 2695 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); 2696 nfsm_fhtom(dvp, v3); 2697 nfsm_strtom(name, len, NFS_MAXNAMLEN); 2698 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); 2699 if (npp && !error) { 2700 nfsm_getfh(nfhp, fhlen, v3); 2701 if (*npp) { 2702 np = *npp; 2703 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { 2704 free((caddr_t)np->n_fhp, M_NFSBIGFH); 2705 np->n_fhp = &np->n_fh; 2706 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) 2707 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK); 2708 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); 2709 np->n_fhsize = fhlen; 2710 newvp = NFSTOV(np); 2711 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { 2712 vref(dvp); 2713 newvp = dvp; 2714 } else { 2715 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np); 2716 if (error) { 2717 m_freem(mrep); 2718 return (error); 2719 } 2720 newvp = NFSTOV(np); 2721 } 2722 if (v3) { 2723 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK); 2724 if (!attrflag && *npp == NULL) { 2725 m_freem(mrep); 2726 if (newvp == dvp) 2727 vrele(newvp); 2728 else 2729 vput(newvp); 2730 return (ENOENT); 2731 } 2732 } else 2733 nfsm_loadattr(newvp, (struct vattr *)0); 2734 } 2735 m_freem(mrep); 2736 nfsmout: 2737 if (npp && *npp == NULL) { 2738 if (error) { 2739 if (newvp) { 2740 if (newvp == dvp) 2741 vrele(newvp); 2742 else 2743 vput(newvp); 2744 } 2745 } else 2746 *npp = np; 2747 } 2748 return (error); 2749 } 2750 2751 /* 2752 * Nfs Version 3 commit rpc 2753 */ 2754 int 2755 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td) 2756 { 2757 caddr_t cp; 2758 u_int32_t *tl; 2759 int32_t t1, t2; 2760 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2761 caddr_t bpos, dpos, cp2; 2762 int error = 0, wccflag = NFSV3_WCCRATTR; 2763 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 2764 2765 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) 2766 return (0); 2767 nfsstats.rpccnt[NFSPROC_COMMIT]++; 2768 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); 2769 nfsm_fhtom(vp, 1); 2770 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2771 txdr_hyper(offset, tl); 2772 tl += 2; 2773 *tl = txdr_unsigned(cnt); 2774 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE)); 2775 nfsm_wcc_data(vp, wccflag); 2776 if (!error) { 2777 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF); 2778 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, 2779 NFSX_V3WRITEVERF)) { 2780 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, 2781 NFSX_V3WRITEVERF); 2782 error = NFSERR_STALEWRITEVERF; 2783 } 2784 } 2785 m_freem(mrep); 2786 nfsmout: 2787 return (error); 2788 } 2789 2790 /* 2791 * Kludge City.. 2792 * - make nfs_bmap() essentially a no-op that does no translation 2793 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc 2794 * (Maybe I could use the process's page mapping, but I was concerned that 2795 * Kernel Write might not be enabled and also figured copyout() would do 2796 * a lot more work than bcopy() and also it currently happens in the 2797 * context of the swapper process (2). 2798 * 2799 * nfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp, 2800 * daddr_t *a_bnp, int *a_runp, int *a_runb) 2801 */ 2802 static int 2803 nfs_bmap(struct vop_bmap_args *ap) 2804 { 2805 struct vnode *vp = ap->a_vp; 2806 2807 if (ap->a_vpp != NULL) 2808 *ap->a_vpp = vp; 2809 if (ap->a_bnp != NULL) 2810 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize); 2811 if (ap->a_runp != NULL) 2812 *ap->a_runp = 0; 2813 if (ap->a_runb != NULL) 2814 *ap->a_runb = 0; 2815 return (0); 2816 } 2817 2818 /* 2819 * Strategy routine. 2820 * For async requests when nfsiod(s) are running, queue the request by 2821 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the 2822 * request. 2823 */ 2824 static int 2825 nfs_strategy(struct vop_strategy_args *ap) 2826 { 2827 struct buf *bp = ap->a_bp; 2828 struct thread *td; 2829 int error = 0; 2830 2831 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); 2832 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp)); 2833 2834 if (bp->b_flags & B_PHYS) 2835 panic("nfs physio"); 2836 2837 if (bp->b_flags & B_ASYNC) 2838 td = NULL; 2839 else 2840 td = curthread; /* XXX */ 2841 2842 /* 2843 * If the op is asynchronous and an i/o daemon is waiting 2844 * queue the request, wake it up and wait for completion 2845 * otherwise just do it ourselves. 2846 */ 2847 if ((bp->b_flags & B_ASYNC) == 0 || 2848 nfs_asyncio(bp, td)) 2849 error = nfs_doio(bp, td); 2850 return (error); 2851 } 2852 2853 /* 2854 * Mmap a file 2855 * 2856 * NB Currently unsupported. 2857 * 2858 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred, 2859 * struct thread *a_td) 2860 */ 2861 /* ARGSUSED */ 2862 static int 2863 nfs_mmap(struct vop_mmap_args *ap) 2864 { 2865 return (EINVAL); 2866 } 2867 2868 /* 2869 * fsync vnode op. Just call nfs_flush() with commit == 1. 2870 * 2871 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp, 2872 * struct ucred * a_cred, int a_waitfor, struct thread *a_td) 2873 */ 2874 /* ARGSUSED */ 2875 static int 2876 nfs_fsync(struct vop_fsync_args *ap) 2877 { 2878 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1)); 2879 } 2880 2881 /* 2882 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be 2883 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains 2884 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is 2885 * set the buffer contains data that has already been written to the server 2886 * and which now needs a commit RPC. 2887 * 2888 * If commit is 0 we only take one pass and only flush buffers containing new 2889 * dirty data. 2890 * 2891 * If commit is 1 we take two passes, issuing a commit RPC in the second 2892 * pass. 2893 * 2894 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required 2895 * to completely flush all pending data. 2896 * 2897 * Note that the RB_SCAN code properly handles the case where the 2898 * callback might block and directly or indirectly (another thread) cause 2899 * the RB tree to change. 2900 */ 2901 2902 #ifndef NFS_COMMITBVECSIZ 2903 #define NFS_COMMITBVECSIZ 16 2904 #endif 2905 2906 struct nfs_flush_info { 2907 enum { NFI_FLUSHNEW, NFI_COMMIT } mode; 2908 struct thread *td; 2909 struct vnode *vp; 2910 int waitfor; 2911 int slpflag; 2912 int slptimeo; 2913 int loops; 2914 struct buf *bvary[NFS_COMMITBVECSIZ]; 2915 int bvsize; 2916 off_t beg_off; 2917 off_t end_off; 2918 }; 2919 2920 static int nfs_flush_bp(struct buf *bp, void *data); 2921 static int nfs_flush_docommit(struct nfs_flush_info *info, int error); 2922 2923 int 2924 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit) 2925 { 2926 struct nfsnode *np = VTONFS(vp); 2927 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 2928 struct nfs_flush_info info; 2929 int error; 2930 2931 bzero(&info, sizeof(info)); 2932 info.td = td; 2933 info.vp = vp; 2934 info.waitfor = waitfor; 2935 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0; 2936 info.loops = 0; 2937 2938 do { 2939 /* 2940 * Flush mode 2941 */ 2942 info.mode = NFI_FLUSHNEW; 2943 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2944 nfs_flush_bp, &info); 2945 2946 /* 2947 * Take a second pass if committing and no error occured. 2948 * Clean up any left over collection (whether an error 2949 * occurs or not). 2950 */ 2951 if (commit && error == 0) { 2952 info.mode = NFI_COMMIT; 2953 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 2954 nfs_flush_bp, &info); 2955 if (info.bvsize) 2956 error = nfs_flush_docommit(&info, error); 2957 } 2958 2959 /* 2960 * Wait for pending I/O to complete before checking whether 2961 * any further dirty buffers exist. 2962 */ 2963 while (waitfor == MNT_WAIT && vp->v_numoutput) { 2964 vp->v_flag |= VBWAIT; 2965 error = tsleep((caddr_t)&vp->v_numoutput, 2966 info.slpflag, "nfsfsync", info.slptimeo); 2967 if (error) { 2968 /* 2969 * We have to be able to break out if this 2970 * is an 'intr' mount. 2971 */ 2972 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) { 2973 error = -EINTR; 2974 break; 2975 } 2976 2977 /* 2978 * Since we do not process pending signals, 2979 * once we get a PCATCH our tsleep() will no 2980 * longer sleep, switch to a fixed timeout 2981 * instead. 2982 */ 2983 if (info.slpflag == PCATCH) { 2984 info.slpflag = 0; 2985 info.slptimeo = 2 * hz; 2986 } 2987 error = 0; 2988 } 2989 } 2990 ++info.loops; 2991 /* 2992 * Loop if we are flushing synchronous as well as committing, 2993 * and dirty buffers are still present. Otherwise we might livelock. 2994 */ 2995 } while (waitfor == MNT_WAIT && commit && 2996 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree)); 2997 2998 /* 2999 * The callbacks have to return a negative error to terminate the 3000 * RB scan. 3001 */ 3002 if (error < 0) 3003 error = -error; 3004 3005 /* 3006 * Deal with any error collection 3007 */ 3008 if (np->n_flag & NWRITEERR) { 3009 error = np->n_error; 3010 np->n_flag &= ~NWRITEERR; 3011 } 3012 return (error); 3013 } 3014 3015 3016 static 3017 int 3018 nfs_flush_bp(struct buf *bp, void *data) 3019 { 3020 struct nfs_flush_info *info = data; 3021 off_t toff; 3022 int error; 3023 3024 error = 0; 3025 switch(info->mode) { 3026 case NFI_FLUSHNEW: 3027 crit_enter(); 3028 if (info->loops && info->waitfor == MNT_WAIT) { 3029 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3030 if (error) { 3031 error = BUF_TIMELOCK(bp, 3032 LK_EXCLUSIVE | LK_SLEEPFAIL, 3033 "nfsfsync", 3034 info->slpflag, info->slptimeo); 3035 } 3036 } else { 3037 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); 3038 } 3039 if (error == 0) { 3040 if ((bp->b_flags & B_DELWRI) == 0) 3041 panic("nfs_fsync: not dirty"); 3042 if (bp->b_flags & B_NEEDCOMMIT) { 3043 BUF_UNLOCK(bp); 3044 crit_exit(); 3045 break; 3046 } 3047 bremfree(bp); 3048 3049 bp->b_flags |= B_ASYNC; 3050 crit_exit(); 3051 VOP_BWRITE(bp->b_vp, bp); 3052 } else { 3053 crit_exit(); 3054 error = 0; 3055 } 3056 break; 3057 case NFI_COMMIT: 3058 /* 3059 * Only process buffers in need of a commit which we can 3060 * immediately lock. This may prevent a buffer from being 3061 * committed, but the normal flush loop will block on the 3062 * same buffer so we shouldn't get into an endless loop. 3063 */ 3064 crit_enter(); 3065 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != 3066 (B_DELWRI | B_NEEDCOMMIT) || 3067 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 3068 crit_exit(); 3069 break; 3070 } 3071 3072 bremfree(bp); 3073 3074 /* 3075 * NOTE: we are not clearing B_DONE here, so we have 3076 * to do it later on in this routine if we intend to 3077 * initiate I/O on the bp. 3078 * 3079 * Note: to avoid loopback deadlocks, we do not 3080 * assign b_runningbufspace. 3081 */ 3082 vfs_busy_pages(bp, 1); 3083 3084 info->bvary[info->bvsize] = bp; 3085 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + 3086 bp->b_dirtyoff; 3087 if (info->bvsize == 0 || toff < info->beg_off) 3088 info->beg_off = toff; 3089 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); 3090 if (info->bvsize == 0 || toff > info->end_off) 3091 info->end_off = toff; 3092 ++info->bvsize; 3093 if (info->bvsize == NFS_COMMITBVECSIZ) { 3094 error = nfs_flush_docommit(info, 0); 3095 KKASSERT(info->bvsize == 0); 3096 } 3097 crit_exit(); 3098 } 3099 return (error); 3100 } 3101 3102 static 3103 int 3104 nfs_flush_docommit(struct nfs_flush_info *info, int error) 3105 { 3106 struct vnode *vp; 3107 struct buf *bp; 3108 off_t bytes; 3109 int retv; 3110 int i; 3111 3112 vp = info->vp; 3113 3114 if (info->bvsize > 0) { 3115 /* 3116 * Commit data on the server, as required. Note that 3117 * nfs_commit will use the vnode's cred for the commit. 3118 * The NFSv3 commit RPC is limited to a 32 bit byte count. 3119 */ 3120 bytes = info->end_off - info->beg_off; 3121 if (bytes > 0x40000000) 3122 bytes = 0x40000000; 3123 if (error) { 3124 retv = -error; 3125 } else { 3126 retv = nfs_commit(vp, info->beg_off, 3127 (int)bytes, info->td); 3128 if (retv == NFSERR_STALEWRITEVERF) 3129 nfs_clearcommit(vp->v_mount); 3130 } 3131 3132 /* 3133 * Now, either mark the blocks I/O done or mark the 3134 * blocks dirty, depending on whether the commit 3135 * succeeded. 3136 */ 3137 for (i = 0; i < info->bvsize; ++i) { 3138 bp = info->bvary[i]; 3139 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); 3140 if (retv) { 3141 /* 3142 * Error, leave B_DELWRI intact 3143 */ 3144 vfs_unbusy_pages(bp); 3145 brelse(bp); 3146 } else { 3147 /* 3148 * Success, remove B_DELWRI ( bundirty() ). 3149 * 3150 * b_dirtyoff/b_dirtyend seem to be NFS 3151 * specific. We should probably move that 3152 * into bundirty(). XXX 3153 */ 3154 crit_enter(); 3155 vp->v_numoutput++; 3156 bp->b_flags |= B_ASYNC; 3157 bundirty(bp); 3158 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3159 bp->b_dirtyoff = bp->b_dirtyend = 0; 3160 crit_exit(); 3161 biodone(bp); 3162 } 3163 } 3164 info->bvsize = 0; 3165 } 3166 return (error); 3167 } 3168 3169 /* 3170 * NFS advisory byte-level locks. 3171 * Currently unsupported. 3172 * 3173 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl, 3174 * int a_flags) 3175 */ 3176 static int 3177 nfs_advlock(struct vop_advlock_args *ap) 3178 { 3179 struct nfsnode *np = VTONFS(ap->a_vp); 3180 3181 /* 3182 * The following kludge is to allow diskless support to work 3183 * until a real NFS lockd is implemented. Basically, just pretend 3184 * that this is a local lock. 3185 */ 3186 return (lf_advlock(ap, &(np->n_lockf), np->n_size)); 3187 } 3188 3189 /* 3190 * Print out the contents of an nfsnode. 3191 * 3192 * nfs_print(struct vnode *a_vp) 3193 */ 3194 static int 3195 nfs_print(struct vop_print_args *ap) 3196 { 3197 struct vnode *vp = ap->a_vp; 3198 struct nfsnode *np = VTONFS(vp); 3199 3200 printf("tag VT_NFS, fileid %ld fsid 0x%x", 3201 np->n_vattr.va_fileid, np->n_vattr.va_fsid); 3202 if (vp->v_type == VFIFO) 3203 fifo_printinfo(vp); 3204 printf("\n"); 3205 return (0); 3206 } 3207 3208 /* 3209 * Just call nfs_writebp() with the force argument set to 1. 3210 * 3211 * NOTE: B_DONE may or may not be set in a_bp on call. 3212 * 3213 * nfs_bwrite(struct vnode *a_bp) 3214 */ 3215 static int 3216 nfs_bwrite(struct vop_bwrite_args *ap) 3217 { 3218 return (nfs_writebp(ap->a_bp, 1, curthread)); 3219 } 3220 3221 /* 3222 * This is a clone of vn_bwrite(), except that it also handles the 3223 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer. 3224 */ 3225 int 3226 nfs_writebp(struct buf *bp, int force, struct thread *td) 3227 { 3228 int oldflags = bp->b_flags; 3229 #if 0 3230 int retv = 1; 3231 off_t off; 3232 #endif 3233 3234 if (BUF_REFCNT(bp) == 0) 3235 panic("bwrite: buffer is not locked???"); 3236 3237 if (bp->b_flags & B_INVAL) { 3238 brelse(bp); 3239 return(0); 3240 } 3241 3242 bp->b_flags |= B_CACHE; 3243 3244 /* 3245 * Undirty the bp. We will redirty it later if the I/O fails. 3246 */ 3247 3248 crit_enter(); 3249 bundirty(bp); 3250 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR); 3251 3252 bp->b_vp->v_numoutput++; 3253 crit_exit(); 3254 3255 /* 3256 * Note: to avoid loopback deadlocks, we do not 3257 * assign b_runningbufspace. 3258 */ 3259 vfs_busy_pages(bp, 1); 3260 3261 BUF_KERNPROC(bp); 3262 VOP_STRATEGY(bp->b_vp, bp); 3263 3264 if( (oldflags & B_ASYNC) == 0) { 3265 int rtval = biowait(bp); 3266 3267 if (oldflags & B_DELWRI) { 3268 crit_enter(); 3269 reassignbuf(bp, bp->b_vp); 3270 crit_exit(); 3271 } 3272 3273 brelse(bp); 3274 return (rtval); 3275 } 3276 3277 return (0); 3278 } 3279 3280 /* 3281 * nfs special file access vnode op. 3282 * Essentially just get vattr and then imitate iaccess() since the device is 3283 * local to the client. 3284 * 3285 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 3286 * struct thread *a_td) 3287 */ 3288 static int 3289 nfsspec_access(struct vop_access_args *ap) 3290 { 3291 struct vattr *vap; 3292 gid_t *gp; 3293 struct ucred *cred = ap->a_cred; 3294 struct vnode *vp = ap->a_vp; 3295 mode_t mode = ap->a_mode; 3296 struct vattr vattr; 3297 int i; 3298 int error; 3299 3300 /* 3301 * Disallow write attempts on filesystems mounted read-only; 3302 * unless the file is a socket, fifo, or a block or character 3303 * device resident on the filesystem. 3304 */ 3305 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { 3306 switch (vp->v_type) { 3307 case VREG: 3308 case VDIR: 3309 case VLNK: 3310 return (EROFS); 3311 default: 3312 break; 3313 } 3314 } 3315 /* 3316 * If you're the super-user, 3317 * you always get access. 3318 */ 3319 if (cred->cr_uid == 0) 3320 return (0); 3321 vap = &vattr; 3322 error = VOP_GETATTR(vp, vap, ap->a_td); 3323 if (error) 3324 return (error); 3325 /* 3326 * Access check is based on only one of owner, group, public. 3327 * If not owner, then check group. If not a member of the 3328 * group, then check public access. 3329 */ 3330 if (cred->cr_uid != vap->va_uid) { 3331 mode >>= 3; 3332 gp = cred->cr_groups; 3333 for (i = 0; i < cred->cr_ngroups; i++, gp++) 3334 if (vap->va_gid == *gp) 3335 goto found; 3336 mode >>= 3; 3337 found: 3338 ; 3339 } 3340 error = (vap->va_mode & mode) == mode ? 0 : EACCES; 3341 return (error); 3342 } 3343 3344 /* 3345 * Read wrapper for special devices. 3346 * 3347 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3348 * struct ucred *a_cred) 3349 */ 3350 static int 3351 nfsspec_read(struct vop_read_args *ap) 3352 { 3353 struct nfsnode *np = VTONFS(ap->a_vp); 3354 3355 /* 3356 * Set access flag. 3357 */ 3358 np->n_flag |= NACC; 3359 getnanotime(&np->n_atim); 3360 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3361 } 3362 3363 /* 3364 * Write wrapper for special devices. 3365 * 3366 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3367 * struct ucred *a_cred) 3368 */ 3369 static int 3370 nfsspec_write(struct vop_write_args *ap) 3371 { 3372 struct nfsnode *np = VTONFS(ap->a_vp); 3373 3374 /* 3375 * Set update flag. 3376 */ 3377 np->n_flag |= NUPD; 3378 getnanotime(&np->n_mtim); 3379 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3380 } 3381 3382 /* 3383 * Close wrapper for special devices. 3384 * 3385 * Update the times on the nfsnode then do device close. 3386 * 3387 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred, 3388 * struct thread *a_td) 3389 */ 3390 static int 3391 nfsspec_close(struct vop_close_args *ap) 3392 { 3393 struct vnode *vp = ap->a_vp; 3394 struct nfsnode *np = VTONFS(vp); 3395 struct vattr vattr; 3396 3397 if (np->n_flag & (NACC | NUPD)) { 3398 np->n_flag |= NCHG; 3399 if (vp->v_usecount == 1 && 3400 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3401 VATTR_NULL(&vattr); 3402 if (np->n_flag & NACC) 3403 vattr.va_atime = np->n_atim; 3404 if (np->n_flag & NUPD) 3405 vattr.va_mtime = np->n_mtim; 3406 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3407 } 3408 } 3409 return (VOCALL(spec_vnode_vops, &ap->a_head)); 3410 } 3411 3412 /* 3413 * Read wrapper for fifos. 3414 * 3415 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3416 * struct ucred *a_cred) 3417 */ 3418 static int 3419 nfsfifo_read(struct vop_read_args *ap) 3420 { 3421 struct nfsnode *np = VTONFS(ap->a_vp); 3422 3423 /* 3424 * Set access flag. 3425 */ 3426 np->n_flag |= NACC; 3427 getnanotime(&np->n_atim); 3428 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3429 } 3430 3431 /* 3432 * Write wrapper for fifos. 3433 * 3434 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 3435 * struct ucred *a_cred) 3436 */ 3437 static int 3438 nfsfifo_write(struct vop_write_args *ap) 3439 { 3440 struct nfsnode *np = VTONFS(ap->a_vp); 3441 3442 /* 3443 * Set update flag. 3444 */ 3445 np->n_flag |= NUPD; 3446 getnanotime(&np->n_mtim); 3447 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3448 } 3449 3450 /* 3451 * Close wrapper for fifos. 3452 * 3453 * Update the times on the nfsnode then do fifo close. 3454 * 3455 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td) 3456 */ 3457 static int 3458 nfsfifo_close(struct vop_close_args *ap) 3459 { 3460 struct vnode *vp = ap->a_vp; 3461 struct nfsnode *np = VTONFS(vp); 3462 struct vattr vattr; 3463 struct timespec ts; 3464 3465 if (np->n_flag & (NACC | NUPD)) { 3466 getnanotime(&ts); 3467 if (np->n_flag & NACC) 3468 np->n_atim = ts; 3469 if (np->n_flag & NUPD) 3470 np->n_mtim = ts; 3471 np->n_flag |= NCHG; 3472 if (vp->v_usecount == 1 && 3473 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 3474 VATTR_NULL(&vattr); 3475 if (np->n_flag & NACC) 3476 vattr.va_atime = np->n_atim; 3477 if (np->n_flag & NUPD) 3478 vattr.va_mtime = np->n_mtim; 3479 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td); 3480 } 3481 } 3482 return (VOCALL(fifo_vnode_vops, &ap->a_head)); 3483 } 3484 3485