1 /* 2 * Copyright (c) 1994, 1995 The Regents of the University of California. 3 * Copyright (c) 1994, 1995 Jan-Simon Pendry. 4 * All rights reserved. 5 * 6 * This code is derived from software donated to Berkeley by 7 * Jan-Simon Pendry. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)union_vfsops.c 8.20 (Berkeley) 5/20/95 34 * $FreeBSD$ 35 */ 36 37 /* 38 * Union Layer 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/kernel.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/proc.h> 47 #include <sys/vnode.h> 48 #include <sys/mount.h> 49 #include <sys/namei.h> 50 #include <sys/malloc.h> 51 #include <sys/filedesc.h> 52 #include <fs/unionfs/union.h> 53 54 static MALLOC_DEFINE(M_UNIONFSMNT, "UNION mount", "UNION mount structure"); 55 56 extern vfs_init_t union_init; 57 static vfs_root_t union_root; 58 static vfs_mount_t union_mount; 59 static vfs_statfs_t union_statfs; 60 static vfs_unmount_t union_unmount; 61 62 /* 63 * Mount union filesystem. 64 */ 65 static int 66 union_mount(mp, td) 67 struct mount *mp; 68 struct thread *td; 69 { 70 int error = 0; 71 struct vfsoptlist *opts; 72 struct vnode *lowerrootvp = NULLVP; 73 struct vnode *upperrootvp = NULLVP; 74 struct union_mount *um = 0; 75 struct ucred *cred = 0; 76 char *cp = 0, *target; 77 int op; 78 int len; 79 size_t size; 80 struct componentname fakecn; 81 struct nameidata nd, *ndp = &nd; 82 83 UDEBUG(("union_mount(mp = %p)\n", (void *)mp)); 84 85 opts = mp->mnt_optnew; 86 /* 87 * Disable clustered write, otherwise system becomes unstable. 88 */ 89 mp->mnt_flag |= MNT_NOCLUSTERW; 90 91 /* 92 * Update is a no-op 93 */ 94 if (mp->mnt_flag & MNT_UPDATE) 95 /* 96 * Need to provide: 97 * 1. a way to convert between rdonly and rdwr mounts. 98 * 2. support for nfs exports. 99 */ 100 return (EOPNOTSUPP); 101 102 /* 103 * Get arguments. 104 */ 105 error = vfs_getopt(opts, "target", (void **)&target, &len); 106 if (error || target[len - 1] != '\0') 107 return (EINVAL); 108 109 op = 0; 110 if (vfs_getopt(opts, "below", NULL, NULL) == 0) 111 op = UNMNT_BELOW; 112 if (vfs_getopt(opts, "replace", NULL, NULL) == 0) { 113 /* These options are mutually exclusive. */ 114 if (op) 115 return (EINVAL); 116 op = UNMNT_REPLACE; 117 } 118 /* 119 * UNMNT_ABOVE is the default. 120 */ 121 if (op == 0) 122 op = UNMNT_ABOVE; 123 124 /* 125 * Obtain lower vnode. Vnode is stored in mp->mnt_vnodecovered. 126 * We need to reference it but not lock it. 127 */ 128 129 lowerrootvp = mp->mnt_vnodecovered; 130 VREF(lowerrootvp); 131 132 #if 0 133 /* 134 * Unlock lower node to avoid deadlock. 135 */ 136 if (lowerrootvp->v_op == union_vnodeop_p) 137 VOP_UNLOCK(lowerrootvp, 0, td); 138 #endif 139 140 /* 141 * Obtain upper vnode by calling namei() on the path. The 142 * upperrootvp will be turned referenced but not locked. 143 */ 144 NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT, UIO_SYSSPACE, target, td); 145 146 error = namei(ndp); 147 148 #if 0 149 if (lowerrootvp->v_op == union_vnodeop_p) 150 vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY, td); 151 #endif 152 if (error) 153 goto bad; 154 155 NDFREE(ndp, NDF_ONLY_PNBUF); 156 upperrootvp = ndp->ni_vp; 157 vrele(ndp->ni_dvp); 158 ndp->ni_dvp = NULL; 159 160 UDEBUG(("mount_root UPPERVP %p locked = %d\n", upperrootvp, 161 VOP_ISLOCKED(upperrootvp, NULL))); 162 163 /* 164 * Check multi union mount to avoid `lock myself again' panic. 165 * Also require that it be a directory. 166 */ 167 if (upperrootvp == VTOUNION(lowerrootvp)->un_uppervp) { 168 #ifdef DIAGNOSTIC 169 printf("union_mount: multi union mount?\n"); 170 #endif 171 error = EDEADLK; 172 goto bad; 173 } 174 175 if (upperrootvp->v_type != VDIR) { 176 error = EINVAL; 177 goto bad; 178 } 179 180 /* 181 * Allocate our union_mount structure and populate the fields. 182 * The vnode references are stored in the union_mount as held, 183 * unlocked references. Depending on the _BELOW flag, the 184 * filesystems are viewed in a different order. In effect this 185 * is the same as providing a mount-under option to the mount 186 * syscall. 187 */ 188 189 um = (struct union_mount *) malloc(sizeof(struct union_mount), 190 M_UNIONFSMNT, M_WAITOK | M_ZERO); 191 192 um->um_op = op; 193 194 switch (um->um_op) { 195 case UNMNT_ABOVE: 196 um->um_lowervp = lowerrootvp; 197 um->um_uppervp = upperrootvp; 198 upperrootvp = NULL; 199 lowerrootvp = NULL; 200 break; 201 202 case UNMNT_BELOW: 203 um->um_lowervp = upperrootvp; 204 um->um_uppervp = lowerrootvp; 205 upperrootvp = NULL; 206 lowerrootvp = NULL; 207 break; 208 209 case UNMNT_REPLACE: 210 vrele(lowerrootvp); 211 lowerrootvp = NULL; 212 um->um_uppervp = upperrootvp; 213 um->um_lowervp = lowerrootvp; 214 upperrootvp = NULL; 215 break; 216 217 default: 218 error = EINVAL; 219 goto bad; 220 } 221 222 /* 223 * Unless the mount is readonly, ensure that the top layer 224 * supports whiteout operations. 225 */ 226 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 227 /* 228 * XXX Fake up a struct componentname with only cn_nameiop 229 * and cn_thread valid; union_whiteout() needs to use the 230 * thread pointer to lock the vnode. 231 */ 232 bzero(&fakecn, sizeof(fakecn)); 233 fakecn.cn_nameiop = LOOKUP; 234 fakecn.cn_thread = td; 235 error = VOP_WHITEOUT(um->um_uppervp, &fakecn, LOOKUP); 236 if (error) 237 goto bad; 238 } 239 240 um->um_cred = crhold(td->td_ucred); 241 FILEDESC_LOCK(td->td_proc->p_fd); 242 um->um_cmode = UN_DIRMODE &~ td->td_proc->p_fd->fd_cmask; 243 FILEDESC_UNLOCK(td->td_proc->p_fd); 244 245 /* 246 * Depending on what you think the MNT_LOCAL flag might mean, 247 * you may want the && to be || on the conditional below. 248 * At the moment it has been defined that the filesystem is 249 * only local if it is all local, ie the MNT_LOCAL flag implies 250 * that the entire namespace is local. If you think the MNT_LOCAL 251 * flag implies that some of the files might be stored locally 252 * then you will want to change the conditional. 253 */ 254 if (um->um_op == UNMNT_ABOVE) { 255 if (((um->um_lowervp == NULLVP) || 256 (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) && 257 (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL)) 258 mp->mnt_flag |= MNT_LOCAL; 259 } 260 261 /* 262 * Copy in the upper layer's RDONLY flag. This is for the benefit 263 * of lookup() which explicitly checks the flag, rather than asking 264 * the filesystem for its own opinion. This means, that an update 265 * mount of the underlying filesystem to go from rdonly to rdwr 266 * will leave the unioned view as read-only. 267 */ 268 mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY); 269 270 mp->mnt_data = (qaddr_t) um; 271 vfs_getnewfsid(mp); 272 273 switch (um->um_op) { 274 case UNMNT_ABOVE: 275 cp = "<above>:"; 276 break; 277 case UNMNT_BELOW: 278 cp = "<below>:"; 279 break; 280 case UNMNT_REPLACE: 281 cp = ""; 282 break; 283 } 284 len = strlen(cp); 285 bcopy(cp, mp->mnt_stat.f_mntfromname, len); 286 287 cp = mp->mnt_stat.f_mntfromname + len; 288 len = MNAMELEN - len; 289 290 (void) copystr(target, cp, len - 1, &size); 291 bzero(cp + size, len - size); 292 293 (void)union_statfs(mp, &mp->mnt_stat, td); 294 295 UDEBUG(("union_mount: from %s, on %s\n", 296 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname)); 297 return (0); 298 299 bad: 300 if (um) { 301 if (um->um_uppervp) 302 vrele(um->um_uppervp); 303 if (um->um_lowervp) 304 vrele(um->um_lowervp); 305 /* XXX other fields */ 306 free(um, M_UNIONFSMNT); 307 } 308 if (cred) 309 crfree(cred); 310 if (upperrootvp) 311 vrele(upperrootvp); 312 if (lowerrootvp) 313 vrele(lowerrootvp); 314 return (error); 315 } 316 317 /* 318 * Free reference to union layer. 319 */ 320 static int 321 union_unmount(mp, mntflags, td) 322 struct mount *mp; 323 int mntflags; 324 struct thread *td; 325 { 326 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 327 int error; 328 int freeing; 329 int flags = 0; 330 331 UDEBUG(("union_unmount(mp = %p)\n", (void *)mp)); 332 333 if (mntflags & MNT_FORCE) 334 flags |= FORCECLOSE; 335 336 /* 337 * Keep flushing vnodes from the mount list. 338 * This is needed because of the un_pvp held 339 * reference to the parent vnode. 340 * If more vnodes have been freed on a given pass, 341 * the try again. The loop will iterate at most 342 * (d) times, where (d) is the maximum tree depth 343 * in the filesystem. 344 */ 345 for (freeing = 0; (error = vflush(mp, 0, flags, td)) != 0;) { 346 int n; 347 348 /* count #vnodes held on mount list */ 349 n = mp->mnt_nvnodelistsize; 350 351 /* if this is unchanged then stop */ 352 if (n == freeing) 353 break; 354 355 /* otherwise try once more time */ 356 freeing = n; 357 } 358 359 /* 360 * If the most recent vflush failed, the filesystem is still busy. 361 */ 362 if (error) 363 return (error); 364 365 /* 366 * Discard references to upper and lower target vnodes. 367 */ 368 if (um->um_lowervp) 369 vrele(um->um_lowervp); 370 vrele(um->um_uppervp); 371 crfree(um->um_cred); 372 /* 373 * Finally, throw away the union_mount structure. 374 */ 375 free(mp->mnt_data, M_UNIONFSMNT); /* XXX */ 376 mp->mnt_data = 0; 377 return (0); 378 } 379 380 static int 381 union_root(mp, vpp, td) 382 struct mount *mp; 383 struct vnode **vpp; 384 struct thread *td; 385 { 386 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 387 int error; 388 389 /* 390 * Supply an unlocked reference to um_uppervp and to um_lowervp. It 391 * is possible for um_uppervp to be locked without the associated 392 * root union_node being locked. We let union_allocvp() deal with 393 * it. 394 */ 395 UDEBUG(("union_root UPPERVP %p locked = %d\n", um->um_uppervp, 396 VOP_ISLOCKED(um->um_uppervp, NULL))); 397 398 VREF(um->um_uppervp); 399 if (um->um_lowervp) 400 VREF(um->um_lowervp); 401 402 error = union_allocvp(vpp, mp, NULLVP, NULLVP, NULL, 403 um->um_uppervp, um->um_lowervp, 1); 404 UDEBUG(("error %d\n", error)); 405 UDEBUG(("union_root2 UPPERVP %p locked = %d\n", um->um_uppervp, 406 VOP_ISLOCKED(um->um_uppervp, NULL))); 407 408 return (error); 409 } 410 411 static int 412 union_statfs(mp, sbp, td) 413 struct mount *mp; 414 struct statfs *sbp; 415 struct thread *td; 416 { 417 int error; 418 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 419 struct statfs mstat; 420 int lbsize; 421 422 UDEBUG(("union_statfs(mp = %p, lvp = %p, uvp = %p)\n", 423 (void *)mp, (void *)um->um_lowervp, (void *)um->um_uppervp)); 424 425 bzero(&mstat, sizeof(mstat)); 426 427 if (um->um_lowervp) { 428 error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, td); 429 if (error) 430 return (error); 431 } 432 433 /* 434 * Now copy across the "interesting" information and fake the rest. 435 */ 436 #if 0 437 sbp->f_type = mstat.f_type; 438 sbp->f_flags = mstat.f_flags; 439 sbp->f_bsize = mstat.f_bsize; 440 sbp->f_iosize = mstat.f_iosize; 441 #endif 442 lbsize = mstat.f_bsize; 443 sbp->f_blocks = mstat.f_blocks; 444 sbp->f_bfree = mstat.f_bfree; 445 sbp->f_bavail = mstat.f_bavail; 446 sbp->f_files = mstat.f_files; 447 sbp->f_ffree = mstat.f_ffree; 448 449 error = VFS_STATFS(um->um_uppervp->v_mount, &mstat, td); 450 if (error) 451 return (error); 452 453 sbp->f_flags = mstat.f_flags; 454 sbp->f_bsize = mstat.f_bsize; 455 sbp->f_iosize = mstat.f_iosize; 456 457 /* 458 * If the lower and upper blocksizes differ, then frig the 459 * block counts so that the sizes reported by df make some 460 * kind of sense. None of this makes sense though. 461 */ 462 463 if (mstat.f_bsize != lbsize) 464 sbp->f_blocks = ((off_t) sbp->f_blocks * lbsize) / mstat.f_bsize; 465 466 /* 467 * The "total" fields count total resources in all layers, 468 * the "free" fields count only those resources which are 469 * free in the upper layer (since only the upper layer 470 * is writeable). 471 */ 472 sbp->f_blocks += mstat.f_blocks; 473 sbp->f_bfree = mstat.f_bfree; 474 sbp->f_bavail = mstat.f_bavail; 475 sbp->f_files += mstat.f_files; 476 sbp->f_ffree = mstat.f_ffree; 477 478 if (sbp != &mp->mnt_stat) { 479 sbp->f_type = mp->mnt_vfc->vfc_typenum; 480 bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid)); 481 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); 482 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 483 } 484 return (0); 485 } 486 487 static struct vfsops union_vfsops = { 488 .vfs_init = union_init, 489 .vfs_mount = union_mount, 490 .vfs_root = union_root, 491 .vfs_statfs = union_statfs, 492 .vfs_unmount = union_unmount, 493 }; 494 495 VFS_SET(union_vfsops, unionfs, VFCF_LOOPBACK); 496