1 /* $NetBSD: union_vfsops.c,v 1.81 2020/03/16 21:20:10 pgoyette Exp $ */ 2 3 /* 4 * Copyright (c) 1994 The Regents of the University of California. 5 * All rights reserved. 6 * 7 * This code is derived from software donated to Berkeley by 8 * Jan-Simon Pendry. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)union_vfsops.c 8.20 (Berkeley) 5/20/95 35 */ 36 37 /* 38 * Copyright (c) 1994 Jan-Simon Pendry. 39 * All rights reserved. 40 * 41 * This code is derived from software donated to Berkeley by 42 * Jan-Simon Pendry. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. All advertising materials mentioning features or use of this software 53 * must display the following acknowledgement: 54 * This product includes software developed by the University of 55 * California, Berkeley and its contributors. 56 * 4. Neither the name of the University nor the names of its contributors 57 * may be used to endorse or promote products derived from this software 58 * without specific prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 61 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 62 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 63 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 64 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 65 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 66 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 70 * SUCH DAMAGE. 71 * 72 * @(#)union_vfsops.c 8.20 (Berkeley) 5/20/95 73 */ 74 75 /* 76 * Union Layer 77 */ 78 79 #include <sys/cdefs.h> 80 __KERNEL_RCSID(0, "$NetBSD: union_vfsops.c,v 1.81 2020/03/16 21:20:10 pgoyette Exp $"); 81 82 #include <sys/param.h> 83 #include <sys/systm.h> 84 #include <sys/sysctl.h> 85 #include <sys/time.h> 86 #include <sys/proc.h> 87 #include <sys/vnode.h> 88 #include <sys/mount.h> 89 #include <sys/namei.h> 90 #include <sys/malloc.h> 91 #include <sys/filedesc.h> 92 #include <sys/queue.h> 93 #include <sys/stat.h> 94 #include <sys/kauth.h> 95 #include <sys/module.h> 96 97 #include <miscfs/genfs/genfs.h> 98 #include <fs/union/union.h> 99 100 MODULE(MODULE_CLASS_VFS, union, NULL); 101 102 /* 103 * Mount union filesystem 104 */ 105 int 106 union_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 107 { 108 struct lwp *l = curlwp; 109 int error = 0; 110 struct union_args *args = data; 111 struct vnode *lowerrootvp = NULLVP; 112 struct vnode *upperrootvp = NULLVP; 113 struct union_mount *um = 0; 114 const char *cp; 115 char *xp; 116 int len; 117 size_t size; 118 119 if (args == NULL) 120 return EINVAL; 121 if (*data_len < sizeof *args) 122 return EINVAL; 123 124 #ifdef UNION_DIAGNOSTIC 125 printf("union_mount(mp = %p)\n", mp); 126 #endif 127 128 if (mp->mnt_flag & MNT_GETARGS) { 129 um = MOUNTTOUNIONMOUNT(mp); 130 if (um == NULL) 131 return EIO; 132 args->target = NULL; 133 args->mntflags = um->um_op; 134 *data_len = sizeof *args; 135 return 0; 136 } 137 /* 138 * Update is a no-op 139 */ 140 if (mp->mnt_flag & MNT_UPDATE) { 141 /* 142 * Need to provide. 143 * 1. a way to convert between rdonly and rdwr mounts. 144 * 2. support for nfs exports. 145 */ 146 error = EOPNOTSUPP; 147 goto bad; 148 } 149 150 lowerrootvp = mp->mnt_vnodecovered; 151 vref(lowerrootvp); 152 153 /* 154 * Find upper node. 155 */ 156 error = namei_simple_user(args->target, 157 NSM_FOLLOW_NOEMULROOT, &upperrootvp); 158 if (error != 0) 159 goto bad; 160 161 if (upperrootvp->v_type != VDIR) { 162 error = EINVAL; 163 goto bad; 164 } 165 166 um = kmem_zalloc(sizeof(struct union_mount), KM_SLEEP); 167 168 /* 169 * Keep a held reference to the target vnodes. 170 * They are vrele'd in union_unmount. 171 * 172 * Depending on the _BELOW flag, the filesystems are 173 * viewed in a different order. In effect, this is the 174 * same as providing a mount under option to the mount syscall. 175 */ 176 177 um->um_op = args->mntflags & UNMNT_OPMASK; 178 switch (um->um_op) { 179 case UNMNT_ABOVE: 180 um->um_lowervp = lowerrootvp; 181 um->um_uppervp = upperrootvp; 182 break; 183 184 case UNMNT_BELOW: 185 um->um_lowervp = upperrootvp; 186 um->um_uppervp = lowerrootvp; 187 break; 188 189 case UNMNT_REPLACE: 190 vrele(lowerrootvp); 191 lowerrootvp = NULLVP; 192 um->um_uppervp = upperrootvp; 193 um->um_lowervp = lowerrootvp; 194 break; 195 196 default: 197 error = EINVAL; 198 goto bad; 199 } 200 201 mp->mnt_iflag |= IMNT_MPSAFE; 202 203 /* 204 * Unless the mount is readonly, ensure that the top layer 205 * supports whiteout operations 206 */ 207 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 208 vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY); 209 error = VOP_WHITEOUT(um->um_uppervp, 210 (struct componentname *) 0, LOOKUP); 211 VOP_UNLOCK(um->um_uppervp); 212 if (error) 213 goto bad; 214 } 215 216 um->um_cred = l->l_cred; 217 kauth_cred_hold(um->um_cred); 218 um->um_cmode = UN_DIRMODE &~ l->l_proc->p_cwdi->cwdi_cmask; 219 220 /* 221 * Depending on what you think the MNT_LOCAL flag might mean, 222 * you may want the && to be || on the conditional below. 223 * At the moment it has been defined that the filesystem is 224 * only local if it is all local, ie the MNT_LOCAL flag implies 225 * that the entire namespace is local. If you think the MNT_LOCAL 226 * flag implies that some of the files might be stored locally 227 * then you will want to change the conditional. 228 */ 229 if (um->um_op == UNMNT_ABOVE) { 230 if (((um->um_lowervp == NULLVP) || 231 (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) && 232 (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL)) 233 mp->mnt_flag |= MNT_LOCAL; 234 } 235 236 /* 237 * Copy in the upper layer's RDONLY flag. This is for the benefit 238 * of lookup() which explicitly checks the flag, rather than asking 239 * the filesystem for its own opinion. This means, that an update 240 * mount of the underlying filesystem to go from rdonly to rdwr 241 * will leave the unioned view as read-only. 242 */ 243 mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY); 244 245 mp->mnt_data = um; 246 vfs_getnewfsid(mp); 247 mp->mnt_lower = um->um_uppervp->v_mount; 248 249 error = set_statvfs_info( path, UIO_USERSPACE, NULL, UIO_USERSPACE, 250 mp->mnt_op->vfs_name, mp, l); 251 if (error) 252 goto bad; 253 254 switch (um->um_op) { 255 case UNMNT_ABOVE: 256 cp = "<above>:"; 257 break; 258 case UNMNT_BELOW: 259 cp = "<below>:"; 260 break; 261 case UNMNT_REPLACE: 262 cp = ""; 263 break; 264 default: 265 cp = "<invalid>:"; 266 #ifdef DIAGNOSTIC 267 panic("union_mount: bad um_op"); 268 #endif 269 break; 270 } 271 len = strlen(cp); 272 memcpy(mp->mnt_stat.f_mntfromname, cp, len); 273 274 xp = mp->mnt_stat.f_mntfromname + len; 275 len = MNAMELEN - len; 276 277 (void) copyinstr(args->target, xp, len - 1, &size); 278 memset(xp + size, 0, len - size); 279 280 #ifdef UNION_DIAGNOSTIC 281 printf("union_mount: from %s, on %s\n", 282 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname); 283 #endif 284 285 /* Setup the readdir hook if it's not set already */ 286 if (!vn_union_readdir_hook) 287 vn_union_readdir_hook = union_readdirhook; 288 289 return (0); 290 291 bad: 292 if (um) 293 kmem_free(um, sizeof(struct union_mount)); 294 if (upperrootvp) 295 vrele(upperrootvp); 296 if (lowerrootvp) 297 vrele(lowerrootvp); 298 return (error); 299 } 300 301 /* 302 * VFS start. Nothing needed here - the start routine 303 * on the underlying filesystem(s) will have been called 304 * when that filesystem was mounted. 305 */ 306 /*ARGSUSED*/ 307 int 308 union_start(struct mount *mp, int flags) 309 { 310 311 return (0); 312 } 313 314 /* 315 * Free reference to union layer 316 */ 317 static bool 318 union_unmount_selector(void *cl, struct vnode *vp) 319 { 320 int *count = cl; 321 322 KASSERT(mutex_owned(vp->v_interlock)); 323 324 *count += 1; 325 return false; 326 } 327 328 int 329 union_unmount(struct mount *mp, int mntflags) 330 { 331 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 332 int freeing; 333 int error; 334 335 #ifdef UNION_DIAGNOSTIC 336 printf("union_unmount(mp = %p)\n", mp); 337 #endif 338 339 /* 340 * Keep flushing vnodes from the mount list. 341 * This is needed because of the un_pvp held 342 * reference to the parent vnode. 343 * If more vnodes have been freed on a given pass, 344 * the try again. The loop will iterate at most 345 * (d) times, where (d) is the maximum tree depth 346 * in the filesystem. 347 */ 348 for (freeing = 0; (error = vflush(mp, NULL, 0)) != 0;) { 349 struct vnode_iterator *marker; 350 int n; 351 352 /* count #vnodes held on mount list */ 353 n = 0; 354 vfs_vnode_iterator_init(mp, &marker); 355 vfs_vnode_iterator_next(marker, union_unmount_selector, &n); 356 vfs_vnode_iterator_destroy(marker); 357 358 /* if this is unchanged then stop */ 359 if (n == freeing) 360 break; 361 362 /* otherwise try once more time */ 363 freeing = n; 364 } 365 366 /* 367 * Ok, now that we've tried doing it gently, get out the hammer. 368 */ 369 370 if (mntflags & MNT_FORCE) 371 error = vflush(mp, NULL, FORCECLOSE); 372 373 if (error) 374 return error; 375 376 /* 377 * Discard references to upper and lower target vnodes. 378 */ 379 if (um->um_lowervp) 380 vrele(um->um_lowervp); 381 vrele(um->um_uppervp); 382 kauth_cred_free(um->um_cred); 383 /* 384 * Finally, throw away the union_mount structure 385 */ 386 kmem_free(um, sizeof(struct union_mount)); 387 mp->mnt_data = NULL; 388 return 0; 389 } 390 391 int 392 union_root(struct mount *mp, int lktype, struct vnode **vpp) 393 { 394 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 395 int error; 396 397 /* 398 * Return locked reference to root. 399 */ 400 vref(um->um_uppervp); 401 if (um->um_lowervp) 402 vref(um->um_lowervp); 403 error = union_allocvp(vpp, mp, NULL, NULL, NULL, 404 um->um_uppervp, um->um_lowervp, 1); 405 406 if (error) { 407 vrele(um->um_uppervp); 408 if (um->um_lowervp) 409 vrele(um->um_lowervp); 410 return error; 411 } 412 413 vn_lock(*vpp, lktype | LK_RETRY); 414 415 return 0; 416 } 417 418 int 419 union_statvfs(struct mount *mp, struct statvfs *sbp) 420 { 421 int error; 422 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 423 struct statvfs *sbuf = malloc(sizeof(*sbuf), M_TEMP, M_WAITOK | M_ZERO); 424 unsigned long lbsize; 425 426 #ifdef UNION_DIAGNOSTIC 427 printf("union_statvfs(mp = %p, lvp = %p, uvp = %p)\n", mp, 428 um->um_lowervp, um->um_uppervp); 429 #endif 430 431 if (um->um_lowervp) { 432 error = VFS_STATVFS(um->um_lowervp->v_mount, sbuf); 433 if (error) 434 goto done; 435 } 436 437 /* now copy across the "interesting" information and fake the rest */ 438 lbsize = sbuf->f_bsize; 439 sbp->f_blocks = sbuf->f_blocks - sbuf->f_bfree; 440 sbp->f_files = sbuf->f_files - sbuf->f_ffree; 441 442 error = VFS_STATVFS(um->um_uppervp->v_mount, sbuf); 443 if (error) 444 goto done; 445 446 sbp->f_flag = sbuf->f_flag; 447 sbp->f_bsize = sbuf->f_bsize; 448 sbp->f_frsize = sbuf->f_frsize; 449 sbp->f_iosize = sbuf->f_iosize; 450 451 /* 452 * The "total" fields count total resources in all layers, 453 * the "free" fields count only those resources which are 454 * free in the upper layer (since only the upper layer 455 * is writable). 456 */ 457 458 if (sbuf->f_bsize != lbsize) 459 sbp->f_blocks = sbp->f_blocks * lbsize / sbuf->f_bsize; 460 sbp->f_blocks += sbuf->f_blocks; 461 sbp->f_bfree = sbuf->f_bfree; 462 sbp->f_bavail = sbuf->f_bavail; 463 sbp->f_bresvd = sbuf->f_bresvd; 464 sbp->f_files += sbuf->f_files; 465 sbp->f_ffree = sbuf->f_ffree; 466 sbp->f_favail = sbuf->f_favail; 467 sbp->f_fresvd = sbuf->f_fresvd; 468 469 copy_statvfs_info(sbp, mp); 470 done: 471 free(sbuf, M_TEMP); 472 return error; 473 } 474 475 /*ARGSUSED*/ 476 int 477 union_sync(struct mount *mp, int waitfor, 478 kauth_cred_t cred) 479 { 480 481 /* 482 * XXX - Assumes no data cached at union layer. 483 */ 484 return (0); 485 } 486 487 /*ARGSUSED*/ 488 int 489 union_vget(struct mount *mp, ino_t ino, int lktype, 490 struct vnode **vpp) 491 { 492 493 return (EOPNOTSUPP); 494 } 495 496 static int 497 union_renamelock_enter(struct mount *mp) 498 { 499 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 500 501 /* Lock just the upper fs, where the action happens. */ 502 return VFS_RENAMELOCK_ENTER(um->um_uppervp->v_mount); 503 } 504 505 static void 506 union_renamelock_exit(struct mount *mp) 507 { 508 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 509 510 VFS_RENAMELOCK_EXIT(um->um_uppervp->v_mount); 511 } 512 513 extern const struct vnodeopv_desc union_vnodeop_opv_desc; 514 515 const struct vnodeopv_desc * const union_vnodeopv_descs[] = { 516 &union_vnodeop_opv_desc, 517 NULL, 518 }; 519 520 struct vfsops union_vfsops = { 521 .vfs_name = MOUNT_UNION, 522 .vfs_min_mount_data = sizeof (struct union_args), 523 .vfs_mount = union_mount, 524 .vfs_start = union_start, 525 .vfs_unmount = union_unmount, 526 .vfs_root = union_root, 527 .vfs_quotactl = (void *)eopnotsupp, 528 .vfs_statvfs = union_statvfs, 529 .vfs_sync = union_sync, 530 .vfs_vget = union_vget, 531 .vfs_loadvnode = union_loadvnode, 532 .vfs_fhtovp = (void *)eopnotsupp, 533 .vfs_vptofh = (void *)eopnotsupp, 534 .vfs_init = union_init, 535 .vfs_reinit = union_reinit, 536 .vfs_done = union_done, 537 .vfs_snapshot = (void *)eopnotsupp, 538 .vfs_extattrctl = vfs_stdextattrctl, 539 .vfs_suspendctl = genfs_suspendctl, 540 .vfs_renamelock_enter = union_renamelock_enter, 541 .vfs_renamelock_exit = union_renamelock_exit, 542 .vfs_fsync = (void *)eopnotsupp, 543 .vfs_opv_descs = union_vnodeopv_descs 544 }; 545 546 SYSCTL_SETUP(unionfs_sysctl_setup, "unionfs sysctl") 547 { 548 549 sysctl_createv(clog, 0, NULL, NULL, 550 CTLFLAG_PERMANENT, 551 CTLTYPE_NODE, "union", 552 SYSCTL_DESCR("Union file system"), 553 NULL, 0, NULL, 0, 554 CTL_VFS, 15, CTL_EOL); 555 /* 556 * XXX the "15" above could be dynamic, thereby eliminating 557 * one more instance of the "number to vfs" mapping problem, 558 * but "15" is the order as taken from sys/mount.h 559 */ 560 } 561 562 static int 563 union_modcmd(modcmd_t cmd, void *arg) 564 { 565 int error; 566 567 switch (cmd) { 568 case MODULE_CMD_INIT: 569 error = vfs_attach(&union_vfsops); 570 if (error != 0) 571 break; 572 break; 573 case MODULE_CMD_FINI: 574 error = vfs_detach(&union_vfsops); 575 if (error != 0) 576 break; 577 break; 578 default: 579 error = ENOTTY; 580 break; 581 } 582 583 return (error); 584 } 585