| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/fs/namespace.c |
| 4 | * |
| 5 | * (C) Copyright Al Viro 2000, 2001 |
| 6 | * |
| 7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. |
| 8 | * Heavily rewritten. |
| 9 | */ |
| 10 | |
| 11 | #include <linux/syscalls.h> |
| 12 | #include <linux/export.h> |
| 13 | #include <linux/capability.h> |
| 14 | #include <linux/mnt_namespace.h> |
| 15 | #include <linux/user_namespace.h> |
| 16 | #include <linux/namei.h> |
| 17 | #include <linux/security.h> |
| 18 | #include <linux/cred.h> |
| 19 | #include <linux/idr.h> |
| 20 | #include <linux/init.h> /* init_rootfs */ |
| 21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
| 22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ |
| 23 | #include <linux/file.h> |
| 24 | #include <linux/uaccess.h> |
| 25 | #include <linux/proc_ns.h> |
| 26 | #include <linux/magic.h> |
| 27 | #include <linux/memblock.h> |
| 28 | #include <linux/proc_fs.h> |
| 29 | #include <linux/task_work.h> |
| 30 | #include <linux/sched/task.h> |
| 31 | #include <uapi/linux/mount.h> |
| 32 | #include <linux/fs_context.h> |
| 33 | #include <linux/shmem_fs.h> |
| 34 | |
| 35 | #include "pnode.h" |
| 36 | #include "internal.h" |
| 37 | |
| 38 | /* Maximum number of mounts in a mount namespace */ |
| 39 | unsigned int sysctl_mount_max __read_mostly = 100000; |
| 40 | |
| 41 | static unsigned int m_hash_mask __read_mostly; |
| 42 | static unsigned int m_hash_shift __read_mostly; |
| 43 | static unsigned int mp_hash_mask __read_mostly; |
| 44 | static unsigned int mp_hash_shift __read_mostly; |
| 45 | |
| 46 | static __initdata unsigned long mhash_entries; |
| 47 | static int __init set_mhash_entries(char *str) |
| 48 | { |
| 49 | if (!str) |
| 50 | return 0; |
| 51 | mhash_entries = simple_strtoul(str, &str, 0); |
| 52 | return 1; |
| 53 | } |
| 54 | __setup("mhash_entries=", set_mhash_entries); |
| 55 | |
| 56 | static __initdata unsigned long mphash_entries; |
| 57 | static int __init set_mphash_entries(char *str) |
| 58 | { |
| 59 | if (!str) |
| 60 | return 0; |
| 61 | mphash_entries = simple_strtoul(str, &str, 0); |
| 62 | return 1; |
| 63 | } |
| 64 | __setup("mphash_entries=", set_mphash_entries); |
| 65 | |
| 66 | static u64 event; |
| 67 | static DEFINE_IDA(mnt_id_ida); |
| 68 | static DEFINE_IDA(mnt_group_ida); |
| 69 | |
| 70 | static struct hlist_head *mount_hashtable __read_mostly; |
| 71 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
| 72 | static struct kmem_cache *mnt_cache __read_mostly; |
| 73 | static DECLARE_RWSEM(namespace_sem); |
| 74 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
| 75 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ |
| 76 | |
| 77 | struct mount_kattr { |
| 78 | unsigned int attr_set; |
| 79 | unsigned int attr_clr; |
| 80 | unsigned int propagation; |
| 81 | unsigned int lookup_flags; |
| 82 | bool recurse; |
| 83 | struct user_namespace *mnt_userns; |
| 84 | }; |
| 85 | |
| 86 | /* /sys/fs */ |
| 87 | struct kobject *fs_kobj; |
| 88 | EXPORT_SYMBOL_GPL(fs_kobj); |
| 89 | |
| 90 | /* |
| 91 | * vfsmount lock may be taken for read to prevent changes to the |
| 92 | * vfsmount hash, ie. during mountpoint lookups or walking back |
| 93 | * up the tree. |
| 94 | * |
| 95 | * It should be taken for write in all cases where the vfsmount |
| 96 | * tree or hash is modified or when a vfsmount structure is modified. |
| 97 | */ |
| 98 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
| 99 | |
| 100 | static inline void lock_mount_hash(void) |
| 101 | { |
| 102 | write_seqlock(&mount_lock); |
| 103 | } |
| 104 | |
| 105 | static inline void unlock_mount_hash(void) |
| 106 | { |
| 107 | write_sequnlock(&mount_lock); |
| 108 | } |
| 109 | |
| 110 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
| 111 | { |
| 112 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
| 113 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); |
| 114 | tmp = tmp + (tmp >> m_hash_shift); |
| 115 | return &mount_hashtable[tmp & m_hash_mask]; |
| 116 | } |
| 117 | |
| 118 | static inline struct hlist_head *mp_hash(struct dentry *dentry) |
| 119 | { |
| 120 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); |
| 121 | tmp = tmp + (tmp >> mp_hash_shift); |
| 122 | return &mountpoint_hashtable[tmp & mp_hash_mask]; |
| 123 | } |
| 124 | |
| 125 | static int mnt_alloc_id(struct mount *mnt) |
| 126 | { |
| 127 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
| 128 | |
| 129 | if (res < 0) |
| 130 | return res; |
| 131 | mnt->mnt_id = res; |
| 132 | return 0; |
| 133 | } |
| 134 | |
| 135 | static void mnt_free_id(struct mount *mnt) |
| 136 | { |
| 137 | ida_free(&mnt_id_ida, mnt->mnt_id); |
| 138 | } |
| 139 | |
| 140 | /* |
| 141 | * Allocate a new peer group ID |
| 142 | */ |
| 143 | static int mnt_alloc_group_id(struct mount *mnt) |
| 144 | { |
| 145 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
| 146 | |
| 147 | if (res < 0) |
| 148 | return res; |
| 149 | mnt->mnt_group_id = res; |
| 150 | return 0; |
| 151 | } |
| 152 | |
| 153 | /* |
| 154 | * Release a peer group ID |
| 155 | */ |
| 156 | void mnt_release_group_id(struct mount *mnt) |
| 157 | { |
| 158 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
| 159 | mnt->mnt_group_id = 0; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * vfsmount lock must be held for read |
| 164 | */ |
| 165 | static inline void mnt_add_count(struct mount *mnt, int n) |
| 166 | { |
| 167 | #ifdef CONFIG_SMP |
| 168 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
| 169 | #else |
| 170 | preempt_disable(); |
| 171 | mnt->mnt_count += n; |
| 172 | preempt_enable(); |
| 173 | #endif |
| 174 | } |
| 175 | |
| 176 | /* |
| 177 | * vfsmount lock must be held for write |
| 178 | */ |
| 179 | int mnt_get_count(struct mount *mnt) |
| 180 | { |
| 181 | #ifdef CONFIG_SMP |
| 182 | int count = 0; |
| 183 | int cpu; |
| 184 | |
| 185 | for_each_possible_cpu(cpu) { |
| 186 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
| 187 | } |
| 188 | |
| 189 | return count; |
| 190 | #else |
| 191 | return mnt->mnt_count; |
| 192 | #endif |
| 193 | } |
| 194 | |
| 195 | static struct mount *alloc_vfsmnt(const char *name) |
| 196 | { |
| 197 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
| 198 | if (mnt) { |
| 199 | int err; |
| 200 | |
| 201 | err = mnt_alloc_id(mnt); |
| 202 | if (err) |
| 203 | goto out_free_cache; |
| 204 | |
| 205 | if (name) { |
| 206 | mnt->mnt_devname = kstrdup_const(name, |
| 207 | GFP_KERNEL_ACCOUNT); |
| 208 | if (!mnt->mnt_devname) |
| 209 | goto out_free_id; |
| 210 | } |
| 211 | |
| 212 | #ifdef CONFIG_SMP |
| 213 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
| 214 | if (!mnt->mnt_pcp) |
| 215 | goto out_free_devname; |
| 216 | |
| 217 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
| 218 | #else |
| 219 | mnt->mnt_count = 1; |
| 220 | mnt->mnt_writers = 0; |
| 221 | #endif |
| 222 | |
| 223 | INIT_HLIST_NODE(&mnt->mnt_hash); |
| 224 | INIT_LIST_HEAD(&mnt->mnt_child); |
| 225 | INIT_LIST_HEAD(&mnt->mnt_mounts); |
| 226 | INIT_LIST_HEAD(&mnt->mnt_list); |
| 227 | INIT_LIST_HEAD(&mnt->mnt_expire); |
| 228 | INIT_LIST_HEAD(&mnt->mnt_share); |
| 229 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
| 230 | INIT_LIST_HEAD(&mnt->mnt_slave); |
| 231 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
| 232 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
| 233 | INIT_HLIST_HEAD(&mnt->mnt_stuck_children); |
| 234 | mnt->mnt.mnt_userns = &init_user_ns; |
| 235 | } |
| 236 | return mnt; |
| 237 | |
| 238 | #ifdef CONFIG_SMP |
| 239 | out_free_devname: |
| 240 | kfree_const(mnt->mnt_devname); |
| 241 | #endif |
| 242 | out_free_id: |
| 243 | mnt_free_id(mnt); |
| 244 | out_free_cache: |
| 245 | kmem_cache_free(mnt_cache, mnt); |
| 246 | return NULL; |
| 247 | } |
| 248 | |
| 249 | /* |
| 250 | * Most r/o checks on a fs are for operations that take |
| 251 | * discrete amounts of time, like a write() or unlink(). |
| 252 | * We must keep track of when those operations start |
| 253 | * (for permission checks) and when they end, so that |
| 254 | * we can determine when writes are able to occur to |
| 255 | * a filesystem. |
| 256 | */ |
| 257 | /* |
| 258 | * __mnt_is_readonly: check whether a mount is read-only |
| 259 | * @mnt: the mount to check for its write status |
| 260 | * |
| 261 | * This shouldn't be used directly ouside of the VFS. |
| 262 | * It does not guarantee that the filesystem will stay |
| 263 | * r/w, just that it is right *now*. This can not and |
| 264 | * should not be used in place of IS_RDONLY(inode). |
| 265 | * mnt_want/drop_write() will _keep_ the filesystem |
| 266 | * r/w. |
| 267 | */ |
| 268 | bool __mnt_is_readonly(struct vfsmount *mnt) |
| 269 | { |
| 270 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
| 271 | } |
| 272 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); |
| 273 | |
| 274 | static inline void mnt_inc_writers(struct mount *mnt) |
| 275 | { |
| 276 | #ifdef CONFIG_SMP |
| 277 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
| 278 | #else |
| 279 | mnt->mnt_writers++; |
| 280 | #endif |
| 281 | } |
| 282 | |
| 283 | static inline void mnt_dec_writers(struct mount *mnt) |
| 284 | { |
| 285 | #ifdef CONFIG_SMP |
| 286 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
| 287 | #else |
| 288 | mnt->mnt_writers--; |
| 289 | #endif |
| 290 | } |
| 291 | |
| 292 | static unsigned int mnt_get_writers(struct mount *mnt) |
| 293 | { |
| 294 | #ifdef CONFIG_SMP |
| 295 | unsigned int count = 0; |
| 296 | int cpu; |
| 297 | |
| 298 | for_each_possible_cpu(cpu) { |
| 299 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
| 300 | } |
| 301 | |
| 302 | return count; |
| 303 | #else |
| 304 | return mnt->mnt_writers; |
| 305 | #endif |
| 306 | } |
| 307 | |
| 308 | static int mnt_is_readonly(struct vfsmount *mnt) |
| 309 | { |
| 310 | if (mnt->mnt_sb->s_readonly_remount) |
| 311 | return 1; |
| 312 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ |
| 313 | smp_rmb(); |
| 314 | return __mnt_is_readonly(mnt); |
| 315 | } |
| 316 | |
| 317 | /* |
| 318 | * Most r/o & frozen checks on a fs are for operations that take discrete |
| 319 | * amounts of time, like a write() or unlink(). We must keep track of when |
| 320 | * those operations start (for permission checks) and when they end, so that we |
| 321 | * can determine when writes are able to occur to a filesystem. |
| 322 | */ |
| 323 | /** |
| 324 | * __mnt_want_write - get write access to a mount without freeze protection |
| 325 | * @m: the mount on which to take a write |
| 326 | * |
| 327 | * This tells the low-level filesystem that a write is about to be performed to |
| 328 | * it, and makes sure that writes are allowed (mnt it read-write) before |
| 329 | * returning success. This operation does not protect against filesystem being |
| 330 | * frozen. When the write operation is finished, __mnt_drop_write() must be |
| 331 | * called. This is effectively a refcount. |
| 332 | */ |
| 333 | int __mnt_want_write(struct vfsmount *m) |
| 334 | { |
| 335 | struct mount *mnt = real_mount(m); |
| 336 | int ret = 0; |
| 337 | |
| 338 | preempt_disable(); |
| 339 | mnt_inc_writers(mnt); |
| 340 | /* |
| 341 | * The store to mnt_inc_writers must be visible before we pass |
| 342 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
| 343 | * incremented count after it has set MNT_WRITE_HOLD. |
| 344 | */ |
| 345 | smp_mb(); |
| 346 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
| 347 | cpu_relax(); |
| 348 | /* |
| 349 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will |
| 350 | * be set to match its requirements. So we must not load that until |
| 351 | * MNT_WRITE_HOLD is cleared. |
| 352 | */ |
| 353 | smp_rmb(); |
| 354 | if (mnt_is_readonly(m)) { |
| 355 | mnt_dec_writers(mnt); |
| 356 | ret = -EROFS; |
| 357 | } |
| 358 | preempt_enable(); |
| 359 | |
| 360 | return ret; |
| 361 | } |
| 362 | |
| 363 | /** |
| 364 | * mnt_want_write - get write access to a mount |
| 365 | * @m: the mount on which to take a write |
| 366 | * |
| 367 | * This tells the low-level filesystem that a write is about to be performed to |
| 368 | * it, and makes sure that writes are allowed (mount is read-write, filesystem |
| 369 | * is not frozen) before returning success. When the write operation is |
| 370 | * finished, mnt_drop_write() must be called. This is effectively a refcount. |
| 371 | */ |
| 372 | int mnt_want_write(struct vfsmount *m) |
| 373 | { |
| 374 | int ret; |
| 375 | |
| 376 | sb_start_write(m->mnt_sb); |
| 377 | ret = __mnt_want_write(m); |
| 378 | if (ret) |
| 379 | sb_end_write(m->mnt_sb); |
| 380 | return ret; |
| 381 | } |
| 382 | EXPORT_SYMBOL_GPL(mnt_want_write); |
| 383 | |
| 384 | /** |
| 385 | * __mnt_want_write_file - get write access to a file's mount |
| 386 | * @file: the file who's mount on which to take a write |
| 387 | * |
| 388 | * This is like __mnt_want_write, but if the file is already open for writing it |
| 389 | * skips incrementing mnt_writers (since the open file already has a reference) |
| 390 | * and instead only does the check for emergency r/o remounts. This must be |
| 391 | * paired with __mnt_drop_write_file. |
| 392 | */ |
| 393 | int __mnt_want_write_file(struct file *file) |
| 394 | { |
| 395 | if (file->f_mode & FMODE_WRITER) { |
| 396 | /* |
| 397 | * Superblock may have become readonly while there are still |
| 398 | * writable fd's, e.g. due to a fs error with errors=remount-ro |
| 399 | */ |
| 400 | if (__mnt_is_readonly(file->f_path.mnt)) |
| 401 | return -EROFS; |
| 402 | return 0; |
| 403 | } |
| 404 | return __mnt_want_write(file->f_path.mnt); |
| 405 | } |
| 406 | |
| 407 | /** |
| 408 | * mnt_want_write_file - get write access to a file's mount |
| 409 | * @file: the file who's mount on which to take a write |
| 410 | * |
| 411 | * This is like mnt_want_write, but if the file is already open for writing it |
| 412 | * skips incrementing mnt_writers (since the open file already has a reference) |
| 413 | * and instead only does the freeze protection and the check for emergency r/o |
| 414 | * remounts. This must be paired with mnt_drop_write_file. |
| 415 | */ |
| 416 | int mnt_want_write_file(struct file *file) |
| 417 | { |
| 418 | int ret; |
| 419 | |
| 420 | sb_start_write(file_inode(file)->i_sb); |
| 421 | ret = __mnt_want_write_file(file); |
| 422 | if (ret) |
| 423 | sb_end_write(file_inode(file)->i_sb); |
| 424 | return ret; |
| 425 | } |
| 426 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
| 427 | |
| 428 | /** |
| 429 | * __mnt_drop_write - give up write access to a mount |
| 430 | * @mnt: the mount on which to give up write access |
| 431 | * |
| 432 | * Tells the low-level filesystem that we are done |
| 433 | * performing writes to it. Must be matched with |
| 434 | * __mnt_want_write() call above. |
| 435 | */ |
| 436 | void __mnt_drop_write(struct vfsmount *mnt) |
| 437 | { |
| 438 | preempt_disable(); |
| 439 | mnt_dec_writers(real_mount(mnt)); |
| 440 | preempt_enable(); |
| 441 | } |
| 442 | |
| 443 | /** |
| 444 | * mnt_drop_write - give up write access to a mount |
| 445 | * @mnt: the mount on which to give up write access |
| 446 | * |
| 447 | * Tells the low-level filesystem that we are done performing writes to it and |
| 448 | * also allows filesystem to be frozen again. Must be matched with |
| 449 | * mnt_want_write() call above. |
| 450 | */ |
| 451 | void mnt_drop_write(struct vfsmount *mnt) |
| 452 | { |
| 453 | __mnt_drop_write(mnt); |
| 454 | sb_end_write(mnt->mnt_sb); |
| 455 | } |
| 456 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
| 457 | |
| 458 | void __mnt_drop_write_file(struct file *file) |
| 459 | { |
| 460 | if (!(file->f_mode & FMODE_WRITER)) |
| 461 | __mnt_drop_write(file->f_path.mnt); |
| 462 | } |
| 463 | |
| 464 | void mnt_drop_write_file(struct file *file) |
| 465 | { |
| 466 | __mnt_drop_write_file(file); |
| 467 | sb_end_write(file_inode(file)->i_sb); |
| 468 | } |
| 469 | EXPORT_SYMBOL(mnt_drop_write_file); |
| 470 | |
| 471 | static inline int mnt_hold_writers(struct mount *mnt) |
| 472 | { |
| 473 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
| 474 | /* |
| 475 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
| 476 | * should be visible before we do. |
| 477 | */ |
| 478 | smp_mb(); |
| 479 | |
| 480 | /* |
| 481 | * With writers on hold, if this value is zero, then there are |
| 482 | * definitely no active writers (although held writers may subsequently |
| 483 | * increment the count, they'll have to wait, and decrement it after |
| 484 | * seeing MNT_READONLY). |
| 485 | * |
| 486 | * It is OK to have counter incremented on one CPU and decremented on |
| 487 | * another: the sum will add up correctly. The danger would be when we |
| 488 | * sum up each counter, if we read a counter before it is incremented, |
| 489 | * but then read another CPU's count which it has been subsequently |
| 490 | * decremented from -- we would see more decrements than we should. |
| 491 | * MNT_WRITE_HOLD protects against this scenario, because |
| 492 | * mnt_want_write first increments count, then smp_mb, then spins on |
| 493 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while |
| 494 | * we're counting up here. |
| 495 | */ |
| 496 | if (mnt_get_writers(mnt) > 0) |
| 497 | return -EBUSY; |
| 498 | |
| 499 | return 0; |
| 500 | } |
| 501 | |
| 502 | static inline void mnt_unhold_writers(struct mount *mnt) |
| 503 | { |
| 504 | /* |
| 505 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers |
| 506 | * that become unheld will see MNT_READONLY. |
| 507 | */ |
| 508 | smp_wmb(); |
| 509 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
| 510 | } |
| 511 | |
| 512 | static int mnt_make_readonly(struct mount *mnt) |
| 513 | { |
| 514 | int ret; |
| 515 | |
| 516 | ret = mnt_hold_writers(mnt); |
| 517 | if (!ret) |
| 518 | mnt->mnt.mnt_flags |= MNT_READONLY; |
| 519 | mnt_unhold_writers(mnt); |
| 520 | return ret; |
| 521 | } |
| 522 | |
| 523 | int sb_prepare_remount_readonly(struct super_block *sb) |
| 524 | { |
| 525 | struct mount *mnt; |
| 526 | int err = 0; |
| 527 | |
| 528 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
| 529 | if (atomic_long_read(&sb->s_remove_count)) |
| 530 | return -EBUSY; |
| 531 | |
| 532 | lock_mount_hash(); |
| 533 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
| 534 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { |
| 535 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
| 536 | smp_mb(); |
| 537 | if (mnt_get_writers(mnt) > 0) { |
| 538 | err = -EBUSY; |
| 539 | break; |
| 540 | } |
| 541 | } |
| 542 | } |
| 543 | if (!err && atomic_long_read(&sb->s_remove_count)) |
| 544 | err = -EBUSY; |
| 545 | |
| 546 | if (!err) { |
| 547 | sb->s_readonly_remount = 1; |
| 548 | smp_wmb(); |
| 549 | } |
| 550 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
| 551 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) |
| 552 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
| 553 | } |
| 554 | unlock_mount_hash(); |
| 555 | |
| 556 | return err; |
| 557 | } |
| 558 | |
| 559 | static void free_vfsmnt(struct mount *mnt) |
| 560 | { |
| 561 | struct user_namespace *mnt_userns; |
| 562 | |
| 563 | mnt_userns = mnt_user_ns(&mnt->mnt); |
| 564 | if (mnt_userns != &init_user_ns) |
| 565 | put_user_ns(mnt_userns); |
| 566 | kfree_const(mnt->mnt_devname); |
| 567 | #ifdef CONFIG_SMP |
| 568 | free_percpu(mnt->mnt_pcp); |
| 569 | #endif |
| 570 | kmem_cache_free(mnt_cache, mnt); |
| 571 | } |
| 572 | |
| 573 | static void delayed_free_vfsmnt(struct rcu_head *head) |
| 574 | { |
| 575 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); |
| 576 | } |
| 577 | |
| 578 | /* call under rcu_read_lock */ |
| 579 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
| 580 | { |
| 581 | struct mount *mnt; |
| 582 | if (read_seqretry(&mount_lock, seq)) |
| 583 | return 1; |
| 584 | if (bastard == NULL) |
| 585 | return 0; |
| 586 | mnt = real_mount(bastard); |
| 587 | mnt_add_count(mnt, 1); |
| 588 | smp_mb(); // see mntput_no_expire() |
| 589 | if (likely(!read_seqretry(&mount_lock, seq))) |
| 590 | return 0; |
| 591 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
| 592 | mnt_add_count(mnt, -1); |
| 593 | return 1; |
| 594 | } |
| 595 | lock_mount_hash(); |
| 596 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { |
| 597 | mnt_add_count(mnt, -1); |
| 598 | unlock_mount_hash(); |
| 599 | return 1; |
| 600 | } |
| 601 | unlock_mount_hash(); |
| 602 | /* caller will mntput() */ |
| 603 | return -1; |
| 604 | } |
| 605 | |
| 606 | /* call under rcu_read_lock */ |
| 607 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
| 608 | { |
| 609 | int res = __legitimize_mnt(bastard, seq); |
| 610 | if (likely(!res)) |
| 611 | return true; |
| 612 | if (unlikely(res < 0)) { |
| 613 | rcu_read_unlock(); |
| 614 | mntput(bastard); |
| 615 | rcu_read_lock(); |
| 616 | } |
| 617 | return false; |
| 618 | } |
| 619 | |
| 620 | /* |
| 621 | * find the first mount at @dentry on vfsmount @mnt. |
| 622 | * call under rcu_read_lock() |
| 623 | */ |
| 624 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
| 625 | { |
| 626 | struct hlist_head *head = m_hash(mnt, dentry); |
| 627 | struct mount *p; |
| 628 | |
| 629 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
| 630 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
| 631 | return p; |
| 632 | return NULL; |
| 633 | } |
| 634 | |
| 635 | /* |
| 636 | * lookup_mnt - Return the first child mount mounted at path |
| 637 | * |
| 638 | * "First" means first mounted chronologically. If you create the |
| 639 | * following mounts: |
| 640 | * |
| 641 | * mount /dev/sda1 /mnt |
| 642 | * mount /dev/sda2 /mnt |
| 643 | * mount /dev/sda3 /mnt |
| 644 | * |
| 645 | * Then lookup_mnt() on the base /mnt dentry in the root mount will |
| 646 | * return successively the root dentry and vfsmount of /dev/sda1, then |
| 647 | * /dev/sda2, then /dev/sda3, then NULL. |
| 648 | * |
| 649 | * lookup_mnt takes a reference to the found vfsmount. |
| 650 | */ |
| 651 | struct vfsmount *lookup_mnt(const struct path *path) |
| 652 | { |
| 653 | struct mount *child_mnt; |
| 654 | struct vfsmount *m; |
| 655 | unsigned seq; |
| 656 | |
| 657 | rcu_read_lock(); |
| 658 | do { |
| 659 | seq = read_seqbegin(&mount_lock); |
| 660 | child_mnt = __lookup_mnt(path->mnt, path->dentry); |
| 661 | m = child_mnt ? &child_mnt->mnt : NULL; |
| 662 | } while (!legitimize_mnt(m, seq)); |
| 663 | rcu_read_unlock(); |
| 664 | return m; |
| 665 | } |
| 666 | |
| 667 | static inline void lock_ns_list(struct mnt_namespace *ns) |
| 668 | { |
| 669 | spin_lock(&ns->ns_lock); |
| 670 | } |
| 671 | |
| 672 | static inline void unlock_ns_list(struct mnt_namespace *ns) |
| 673 | { |
| 674 | spin_unlock(&ns->ns_lock); |
| 675 | } |
| 676 | |
| 677 | static inline bool mnt_is_cursor(struct mount *mnt) |
| 678 | { |
| 679 | return mnt->mnt.mnt_flags & MNT_CURSOR; |
| 680 | } |
| 681 | |
| 682 | /* |
| 683 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the |
| 684 | * current mount namespace. |
| 685 | * |
| 686 | * The common case is dentries are not mountpoints at all and that |
| 687 | * test is handled inline. For the slow case when we are actually |
| 688 | * dealing with a mountpoint of some kind, walk through all of the |
| 689 | * mounts in the current mount namespace and test to see if the dentry |
| 690 | * is a mountpoint. |
| 691 | * |
| 692 | * The mount_hashtable is not usable in the context because we |
| 693 | * need to identify all mounts that may be in the current mount |
| 694 | * namespace not just a mount that happens to have some specified |
| 695 | * parent mount. |
| 696 | */ |
| 697 | bool __is_local_mountpoint(struct dentry *dentry) |
| 698 | { |
| 699 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
| 700 | struct mount *mnt; |
| 701 | bool is_covered = false; |
| 702 | |
| 703 | down_read(&namespace_sem); |
| 704 | lock_ns_list(ns); |
| 705 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
| 706 | if (mnt_is_cursor(mnt)) |
| 707 | continue; |
| 708 | is_covered = (mnt->mnt_mountpoint == dentry); |
| 709 | if (is_covered) |
| 710 | break; |
| 711 | } |
| 712 | unlock_ns_list(ns); |
| 713 | up_read(&namespace_sem); |
| 714 | |
| 715 | return is_covered; |
| 716 | } |
| 717 | |
| 718 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
| 719 | { |
| 720 | struct hlist_head *chain = mp_hash(dentry); |
| 721 | struct mountpoint *mp; |
| 722 | |
| 723 | hlist_for_each_entry(mp, chain, m_hash) { |
| 724 | if (mp->m_dentry == dentry) { |
| 725 | mp->m_count++; |
| 726 | return mp; |
| 727 | } |
| 728 | } |
| 729 | return NULL; |
| 730 | } |
| 731 | |
| 732 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
| 733 | { |
| 734 | struct mountpoint *mp, *new = NULL; |
| 735 | int ret; |
| 736 | |
| 737 | if (d_mountpoint(dentry)) { |
| 738 | /* might be worth a WARN_ON() */ |
| 739 | if (d_unlinked(dentry)) |
| 740 | return ERR_PTR(-ENOENT); |
| 741 | mountpoint: |
| 742 | read_seqlock_excl(&mount_lock); |
| 743 | mp = lookup_mountpoint(dentry); |
| 744 | read_sequnlock_excl(&mount_lock); |
| 745 | if (mp) |
| 746 | goto done; |
| 747 | } |
| 748 | |
| 749 | if (!new) |
| 750 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); |
| 751 | if (!new) |
| 752 | return ERR_PTR(-ENOMEM); |
| 753 | |
| 754 | |
| 755 | /* Exactly one processes may set d_mounted */ |
| 756 | ret = d_set_mounted(dentry); |
| 757 | |
| 758 | /* Someone else set d_mounted? */ |
| 759 | if (ret == -EBUSY) |
| 760 | goto mountpoint; |
| 761 | |
| 762 | /* The dentry is not available as a mountpoint? */ |
| 763 | mp = ERR_PTR(ret); |
| 764 | if (ret) |
| 765 | goto done; |
| 766 | |
| 767 | /* Add the new mountpoint to the hash table */ |
| 768 | read_seqlock_excl(&mount_lock); |
| 769 | new->m_dentry = dget(dentry); |
| 770 | new->m_count = 1; |
| 771 | hlist_add_head(&new->m_hash, mp_hash(dentry)); |
| 772 | INIT_HLIST_HEAD(&new->m_list); |
| 773 | read_sequnlock_excl(&mount_lock); |
| 774 | |
| 775 | mp = new; |
| 776 | new = NULL; |
| 777 | done: |
| 778 | kfree(new); |
| 779 | return mp; |
| 780 | } |
| 781 | |
| 782 | /* |
| 783 | * vfsmount lock must be held. Additionally, the caller is responsible |
| 784 | * for serializing calls for given disposal list. |
| 785 | */ |
| 786 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) |
| 787 | { |
| 788 | if (!--mp->m_count) { |
| 789 | struct dentry *dentry = mp->m_dentry; |
| 790 | BUG_ON(!hlist_empty(&mp->m_list)); |
| 791 | spin_lock(&dentry->d_lock); |
| 792 | dentry->d_flags &= ~DCACHE_MOUNTED; |
| 793 | spin_unlock(&dentry->d_lock); |
| 794 | dput_to_list(dentry, list); |
| 795 | hlist_del(&mp->m_hash); |
| 796 | kfree(mp); |
| 797 | } |
| 798 | } |
| 799 | |
| 800 | /* called with namespace_lock and vfsmount lock */ |
| 801 | static void put_mountpoint(struct mountpoint *mp) |
| 802 | { |
| 803 | __put_mountpoint(mp, &ex_mountpoints); |
| 804 | } |
| 805 | |
| 806 | static inline int check_mnt(struct mount *mnt) |
| 807 | { |
| 808 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
| 809 | } |
| 810 | |
| 811 | /* |
| 812 | * vfsmount lock must be held for write |
| 813 | */ |
| 814 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
| 815 | { |
| 816 | if (ns) { |
| 817 | ns->event = ++event; |
| 818 | wake_up_interruptible(&ns->poll); |
| 819 | } |
| 820 | } |
| 821 | |
| 822 | /* |
| 823 | * vfsmount lock must be held for write |
| 824 | */ |
| 825 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
| 826 | { |
| 827 | if (ns && ns->event != event) { |
| 828 | ns->event = event; |
| 829 | wake_up_interruptible(&ns->poll); |
| 830 | } |
| 831 | } |
| 832 | |
| 833 | /* |
| 834 | * vfsmount lock must be held for write |
| 835 | */ |
| 836 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
| 837 | { |
| 838 | struct mountpoint *mp; |
| 839 | mnt->mnt_parent = mnt; |
| 840 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
| 841 | list_del_init(&mnt->mnt_child); |
| 842 | hlist_del_init_rcu(&mnt->mnt_hash); |
| 843 | hlist_del_init(&mnt->mnt_mp_list); |
| 844 | mp = mnt->mnt_mp; |
| 845 | mnt->mnt_mp = NULL; |
| 846 | return mp; |
| 847 | } |
| 848 | |
| 849 | /* |
| 850 | * vfsmount lock must be held for write |
| 851 | */ |
| 852 | static void umount_mnt(struct mount *mnt) |
| 853 | { |
| 854 | put_mountpoint(unhash_mnt(mnt)); |
| 855 | } |
| 856 | |
| 857 | /* |
| 858 | * vfsmount lock must be held for write |
| 859 | */ |
| 860 | void mnt_set_mountpoint(struct mount *mnt, |
| 861 | struct mountpoint *mp, |
| 862 | struct mount *child_mnt) |
| 863 | { |
| 864 | mp->m_count++; |
| 865 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
| 866 | child_mnt->mnt_mountpoint = mp->m_dentry; |
| 867 | child_mnt->mnt_parent = mnt; |
| 868 | child_mnt->mnt_mp = mp; |
| 869 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
| 870 | } |
| 871 | |
| 872 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
| 873 | { |
| 874 | hlist_add_head_rcu(&mnt->mnt_hash, |
| 875 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); |
| 876 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
| 877 | } |
| 878 | |
| 879 | /* |
| 880 | * vfsmount lock must be held for write |
| 881 | */ |
| 882 | static void attach_mnt(struct mount *mnt, |
| 883 | struct mount *parent, |
| 884 | struct mountpoint *mp) |
| 885 | { |
| 886 | mnt_set_mountpoint(parent, mp, mnt); |
| 887 | __attach_mnt(mnt, parent); |
| 888 | } |
| 889 | |
| 890 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
| 891 | { |
| 892 | struct mountpoint *old_mp = mnt->mnt_mp; |
| 893 | struct mount *old_parent = mnt->mnt_parent; |
| 894 | |
| 895 | list_del_init(&mnt->mnt_child); |
| 896 | hlist_del_init(&mnt->mnt_mp_list); |
| 897 | hlist_del_init_rcu(&mnt->mnt_hash); |
| 898 | |
| 899 | attach_mnt(mnt, parent, mp); |
| 900 | |
| 901 | put_mountpoint(old_mp); |
| 902 | mnt_add_count(old_parent, -1); |
| 903 | } |
| 904 | |
| 905 | /* |
| 906 | * vfsmount lock must be held for write |
| 907 | */ |
| 908 | static void commit_tree(struct mount *mnt) |
| 909 | { |
| 910 | struct mount *parent = mnt->mnt_parent; |
| 911 | struct mount *m; |
| 912 | LIST_HEAD(head); |
| 913 | struct mnt_namespace *n = parent->mnt_ns; |
| 914 | |
| 915 | BUG_ON(parent == mnt); |
| 916 | |
| 917 | list_add_tail(&head, &mnt->mnt_list); |
| 918 | list_for_each_entry(m, &head, mnt_list) |
| 919 | m->mnt_ns = n; |
| 920 | |
| 921 | list_splice(&head, n->list.prev); |
| 922 | |
| 923 | n->mounts += n->pending_mounts; |
| 924 | n->pending_mounts = 0; |
| 925 | |
| 926 | __attach_mnt(mnt, parent); |
| 927 | touch_mnt_namespace(n); |
| 928 | } |
| 929 | |
| 930 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
| 931 | { |
| 932 | struct list_head *next = p->mnt_mounts.next; |
| 933 | if (next == &p->mnt_mounts) { |
| 934 | while (1) { |
| 935 | if (p == root) |
| 936 | return NULL; |
| 937 | next = p->mnt_child.next; |
| 938 | if (next != &p->mnt_parent->mnt_mounts) |
| 939 | break; |
| 940 | p = p->mnt_parent; |
| 941 | } |
| 942 | } |
| 943 | return list_entry(next, struct mount, mnt_child); |
| 944 | } |
| 945 | |
| 946 | static struct mount *skip_mnt_tree(struct mount *p) |
| 947 | { |
| 948 | struct list_head *prev = p->mnt_mounts.prev; |
| 949 | while (prev != &p->mnt_mounts) { |
| 950 | p = list_entry(prev, struct mount, mnt_child); |
| 951 | prev = p->mnt_mounts.prev; |
| 952 | } |
| 953 | return p; |
| 954 | } |
| 955 | |
| 956 | /** |
| 957 | * vfs_create_mount - Create a mount for a configured superblock |
| 958 | * @fc: The configuration context with the superblock attached |
| 959 | * |
| 960 | * Create a mount to an already configured superblock. If necessary, the |
| 961 | * caller should invoke vfs_get_tree() before calling this. |
| 962 | * |
| 963 | * Note that this does not attach the mount to anything. |
| 964 | */ |
| 965 | struct vfsmount *vfs_create_mount(struct fs_context *fc) |
| 966 | { |
| 967 | struct mount *mnt; |
| 968 | |
| 969 | if (!fc->root) |
| 970 | return ERR_PTR(-EINVAL); |
| 971 | |
| 972 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
| 973 | if (!mnt) |
| 974 | return ERR_PTR(-ENOMEM); |
| 975 | |
| 976 | if (fc->sb_flags & SB_KERNMOUNT) |
| 977 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
| 978 | |
| 979 | atomic_inc(&fc->root->d_sb->s_active); |
| 980 | mnt->mnt.mnt_sb = fc->root->d_sb; |
| 981 | mnt->mnt.mnt_root = dget(fc->root); |
| 982 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
| 983 | mnt->mnt_parent = mnt; |
| 984 | |
| 985 | lock_mount_hash(); |
| 986 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
| 987 | unlock_mount_hash(); |
| 988 | return &mnt->mnt; |
| 989 | } |
| 990 | EXPORT_SYMBOL(vfs_create_mount); |
| 991 | |
| 992 | struct vfsmount *fc_mount(struct fs_context *fc) |
| 993 | { |
| 994 | int err = vfs_get_tree(fc); |
| 995 | if (!err) { |
| 996 | up_write(&fc->root->d_sb->s_umount); |
| 997 | return vfs_create_mount(fc); |
| 998 | } |
| 999 | return ERR_PTR(err); |
| 1000 | } |
| 1001 | EXPORT_SYMBOL(fc_mount); |
| 1002 | |
| 1003 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
| 1004 | int flags, const char *name, |
| 1005 | void *data) |
| 1006 | { |
| 1007 | struct fs_context *fc; |
| 1008 | struct vfsmount *mnt; |
| 1009 | int ret = 0; |
| 1010 | |
| 1011 | if (!type) |
| 1012 | return ERR_PTR(-EINVAL); |
| 1013 | |
| 1014 | fc = fs_context_for_mount(type, flags); |
| 1015 | if (IS_ERR(fc)) |
| 1016 | return ERR_CAST(fc); |
| 1017 | |
| 1018 | if (name) |
| 1019 | ret = vfs_parse_fs_string(fc, "source", |
| 1020 | name, strlen(name)); |
| 1021 | if (!ret) |
| 1022 | ret = parse_monolithic_mount_data(fc, data); |
| 1023 | if (!ret) |
| 1024 | mnt = fc_mount(fc); |
| 1025 | else |
| 1026 | mnt = ERR_PTR(ret); |
| 1027 | |
| 1028 | put_fs_context(fc); |
| 1029 | return mnt; |
| 1030 | } |
| 1031 | EXPORT_SYMBOL_GPL(vfs_kern_mount); |
| 1032 | |
| 1033 | struct vfsmount * |
| 1034 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, |
| 1035 | const char *name, void *data) |
| 1036 | { |
| 1037 | /* Until it is worked out how to pass the user namespace |
| 1038 | * through from the parent mount to the submount don't support |
| 1039 | * unprivileged mounts with submounts. |
| 1040 | */ |
| 1041 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) |
| 1042 | return ERR_PTR(-EPERM); |
| 1043 | |
| 1044 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
| 1045 | } |
| 1046 | EXPORT_SYMBOL_GPL(vfs_submount); |
| 1047 | |
| 1048 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
| 1049 | int flag) |
| 1050 | { |
| 1051 | struct super_block *sb = old->mnt.mnt_sb; |
| 1052 | struct mount *mnt; |
| 1053 | int err; |
| 1054 | |
| 1055 | mnt = alloc_vfsmnt(old->mnt_devname); |
| 1056 | if (!mnt) |
| 1057 | return ERR_PTR(-ENOMEM); |
| 1058 | |
| 1059 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
| 1060 | mnt->mnt_group_id = 0; /* not a peer of original */ |
| 1061 | else |
| 1062 | mnt->mnt_group_id = old->mnt_group_id; |
| 1063 | |
| 1064 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
| 1065 | err = mnt_alloc_group_id(mnt); |
| 1066 | if (err) |
| 1067 | goto out_free; |
| 1068 | } |
| 1069 | |
| 1070 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
| 1071 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); |
| 1072 | |
| 1073 | atomic_inc(&sb->s_active); |
| 1074 | mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); |
| 1075 | if (mnt->mnt.mnt_userns != &init_user_ns) |
| 1076 | mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); |
| 1077 | mnt->mnt.mnt_sb = sb; |
| 1078 | mnt->mnt.mnt_root = dget(root); |
| 1079 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
| 1080 | mnt->mnt_parent = mnt; |
| 1081 | lock_mount_hash(); |
| 1082 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
| 1083 | unlock_mount_hash(); |
| 1084 | |
| 1085 | if ((flag & CL_SLAVE) || |
| 1086 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { |
| 1087 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
| 1088 | mnt->mnt_master = old; |
| 1089 | CLEAR_MNT_SHARED(mnt); |
| 1090 | } else if (!(flag & CL_PRIVATE)) { |
| 1091 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) |
| 1092 | list_add(&mnt->mnt_share, &old->mnt_share); |
| 1093 | if (IS_MNT_SLAVE(old)) |
| 1094 | list_add(&mnt->mnt_slave, &old->mnt_slave); |
| 1095 | mnt->mnt_master = old->mnt_master; |
| 1096 | } else { |
| 1097 | CLEAR_MNT_SHARED(mnt); |
| 1098 | } |
| 1099 | if (flag & CL_MAKE_SHARED) |
| 1100 | set_mnt_shared(mnt); |
| 1101 | |
| 1102 | /* stick the duplicate mount on the same expiry list |
| 1103 | * as the original if that was on one */ |
| 1104 | if (flag & CL_EXPIRE) { |
| 1105 | if (!list_empty(&old->mnt_expire)) |
| 1106 | list_add(&mnt->mnt_expire, &old->mnt_expire); |
| 1107 | } |
| 1108 | |
| 1109 | return mnt; |
| 1110 | |
| 1111 | out_free: |
| 1112 | mnt_free_id(mnt); |
| 1113 | free_vfsmnt(mnt); |
| 1114 | return ERR_PTR(err); |
| 1115 | } |
| 1116 | |
| 1117 | static void cleanup_mnt(struct mount *mnt) |
| 1118 | { |
| 1119 | struct hlist_node *p; |
| 1120 | struct mount *m; |
| 1121 | /* |
| 1122 | * The warning here probably indicates that somebody messed |
| 1123 | * up a mnt_want/drop_write() pair. If this happens, the |
| 1124 | * filesystem was probably unable to make r/w->r/o transitions. |
| 1125 | * The locking used to deal with mnt_count decrement provides barriers, |
| 1126 | * so mnt_get_writers() below is safe. |
| 1127 | */ |
| 1128 | WARN_ON(mnt_get_writers(mnt)); |
| 1129 | if (unlikely(mnt->mnt_pins.first)) |
| 1130 | mnt_pin_kill(mnt); |
| 1131 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
| 1132 | hlist_del(&m->mnt_umount); |
| 1133 | mntput(&m->mnt); |
| 1134 | } |
| 1135 | fsnotify_vfsmount_delete(&mnt->mnt); |
| 1136 | dput(mnt->mnt.mnt_root); |
| 1137 | deactivate_super(mnt->mnt.mnt_sb); |
| 1138 | mnt_free_id(mnt); |
| 1139 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); |
| 1140 | } |
| 1141 | |
| 1142 | static void __cleanup_mnt(struct rcu_head *head) |
| 1143 | { |
| 1144 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); |
| 1145 | } |
| 1146 | |
| 1147 | static LLIST_HEAD(delayed_mntput_list); |
| 1148 | static void delayed_mntput(struct work_struct *unused) |
| 1149 | { |
| 1150 | struct llist_node *node = llist_del_all(&delayed_mntput_list); |
| 1151 | struct mount *m, *t; |
| 1152 | |
| 1153 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
| 1154 | cleanup_mnt(m); |
| 1155 | } |
| 1156 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); |
| 1157 | |
| 1158 | static void mntput_no_expire(struct mount *mnt) |
| 1159 | { |
| 1160 | LIST_HEAD(list); |
| 1161 | int count; |
| 1162 | |
| 1163 | rcu_read_lock(); |
| 1164 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
| 1165 | /* |
| 1166 | * Since we don't do lock_mount_hash() here, |
| 1167 | * ->mnt_ns can change under us. However, if it's |
| 1168 | * non-NULL, then there's a reference that won't |
| 1169 | * be dropped until after an RCU delay done after |
| 1170 | * turning ->mnt_ns NULL. So if we observe it |
| 1171 | * non-NULL under rcu_read_lock(), the reference |
| 1172 | * we are dropping is not the final one. |
| 1173 | */ |
| 1174 | mnt_add_count(mnt, -1); |
| 1175 | rcu_read_unlock(); |
| 1176 | return; |
| 1177 | } |
| 1178 | lock_mount_hash(); |
| 1179 | /* |
| 1180 | * make sure that if __legitimize_mnt() has not seen us grab |
| 1181 | * mount_lock, we'll see their refcount increment here. |
| 1182 | */ |
| 1183 | smp_mb(); |
| 1184 | mnt_add_count(mnt, -1); |
| 1185 | count = mnt_get_count(mnt); |
| 1186 | if (count != 0) { |
| 1187 | WARN_ON(count < 0); |
| 1188 | rcu_read_unlock(); |
| 1189 | unlock_mount_hash(); |
| 1190 | return; |
| 1191 | } |
| 1192 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
| 1193 | rcu_read_unlock(); |
| 1194 | unlock_mount_hash(); |
| 1195 | return; |
| 1196 | } |
| 1197 | mnt->mnt.mnt_flags |= MNT_DOOMED; |
| 1198 | rcu_read_unlock(); |
| 1199 | |
| 1200 | list_del(&mnt->mnt_instance); |
| 1201 | |
| 1202 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { |
| 1203 | struct mount *p, *tmp; |
| 1204 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { |
| 1205 | __put_mountpoint(unhash_mnt(p), &list); |
| 1206 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
| 1207 | } |
| 1208 | } |
| 1209 | unlock_mount_hash(); |
| 1210 | shrink_dentry_list(&list); |
| 1211 | |
| 1212 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
| 1213 | struct task_struct *task = current; |
| 1214 | if (likely(!(task->flags & PF_KTHREAD))) { |
| 1215 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); |
| 1216 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
| 1217 | return; |
| 1218 | } |
| 1219 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) |
| 1220 | schedule_delayed_work(&delayed_mntput_work, 1); |
| 1221 | return; |
| 1222 | } |
| 1223 | cleanup_mnt(mnt); |
| 1224 | } |
| 1225 | |
| 1226 | void mntput(struct vfsmount *mnt) |
| 1227 | { |
| 1228 | if (mnt) { |
| 1229 | struct mount *m = real_mount(mnt); |
| 1230 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
| 1231 | if (unlikely(m->mnt_expiry_mark)) |
| 1232 | m->mnt_expiry_mark = 0; |
| 1233 | mntput_no_expire(m); |
| 1234 | } |
| 1235 | } |
| 1236 | EXPORT_SYMBOL(mntput); |
| 1237 | |
| 1238 | struct vfsmount *mntget(struct vfsmount *mnt) |
| 1239 | { |
| 1240 | if (mnt) |
| 1241 | mnt_add_count(real_mount(mnt), 1); |
| 1242 | return mnt; |
| 1243 | } |
| 1244 | EXPORT_SYMBOL(mntget); |
| 1245 | |
| 1246 | /** |
| 1247 | * path_is_mountpoint() - Check if path is a mount in the current namespace. |
| 1248 | * @path: path to check |
| 1249 | * |
| 1250 | * d_mountpoint() can only be used reliably to establish if a dentry is |
| 1251 | * not mounted in any namespace and that common case is handled inline. |
| 1252 | * d_mountpoint() isn't aware of the possibility there may be multiple |
| 1253 | * mounts using a given dentry in a different namespace. This function |
| 1254 | * checks if the passed in path is a mountpoint rather than the dentry |
| 1255 | * alone. |
| 1256 | */ |
| 1257 | bool path_is_mountpoint(const struct path *path) |
| 1258 | { |
| 1259 | unsigned seq; |
| 1260 | bool res; |
| 1261 | |
| 1262 | if (!d_mountpoint(path->dentry)) |
| 1263 | return false; |
| 1264 | |
| 1265 | rcu_read_lock(); |
| 1266 | do { |
| 1267 | seq = read_seqbegin(&mount_lock); |
| 1268 | res = __path_is_mountpoint(path); |
| 1269 | } while (read_seqretry(&mount_lock, seq)); |
| 1270 | rcu_read_unlock(); |
| 1271 | |
| 1272 | return res; |
| 1273 | } |
| 1274 | EXPORT_SYMBOL(path_is_mountpoint); |
| 1275 | |
| 1276 | struct vfsmount *mnt_clone_internal(const struct path *path) |
| 1277 | { |
| 1278 | struct mount *p; |
| 1279 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); |
| 1280 | if (IS_ERR(p)) |
| 1281 | return ERR_CAST(p); |
| 1282 | p->mnt.mnt_flags |= MNT_INTERNAL; |
| 1283 | return &p->mnt; |
| 1284 | } |
| 1285 | |
| 1286 | #ifdef CONFIG_PROC_FS |
| 1287 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
| 1288 | struct list_head *p) |
| 1289 | { |
| 1290 | struct mount *mnt, *ret = NULL; |
| 1291 | |
| 1292 | lock_ns_list(ns); |
| 1293 | list_for_each_continue(p, &ns->list) { |
| 1294 | mnt = list_entry(p, typeof(*mnt), mnt_list); |
| 1295 | if (!mnt_is_cursor(mnt)) { |
| 1296 | ret = mnt; |
| 1297 | break; |
| 1298 | } |
| 1299 | } |
| 1300 | unlock_ns_list(ns); |
| 1301 | |
| 1302 | return ret; |
| 1303 | } |
| 1304 | |
| 1305 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
| 1306 | static void *m_start(struct seq_file *m, loff_t *pos) |
| 1307 | { |
| 1308 | struct proc_mounts *p = m->private; |
| 1309 | struct list_head *prev; |
| 1310 | |
| 1311 | down_read(&namespace_sem); |
| 1312 | if (!*pos) { |
| 1313 | prev = &p->ns->list; |
| 1314 | } else { |
| 1315 | prev = &p->cursor.mnt_list; |
| 1316 | |
| 1317 | /* Read after we'd reached the end? */ |
| 1318 | if (list_empty(prev)) |
| 1319 | return NULL; |
| 1320 | } |
| 1321 | |
| 1322 | return mnt_list_next(p->ns, prev); |
| 1323 | } |
| 1324 | |
| 1325 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) |
| 1326 | { |
| 1327 | struct proc_mounts *p = m->private; |
| 1328 | struct mount *mnt = v; |
| 1329 | |
| 1330 | ++*pos; |
| 1331 | return mnt_list_next(p->ns, &mnt->mnt_list); |
| 1332 | } |
| 1333 | |
| 1334 | static void m_stop(struct seq_file *m, void *v) |
| 1335 | { |
| 1336 | struct proc_mounts *p = m->private; |
| 1337 | struct mount *mnt = v; |
| 1338 | |
| 1339 | lock_ns_list(p->ns); |
| 1340 | if (mnt) |
| 1341 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); |
| 1342 | else |
| 1343 | list_del_init(&p->cursor.mnt_list); |
| 1344 | unlock_ns_list(p->ns); |
| 1345 | up_read(&namespace_sem); |
| 1346 | } |
| 1347 | |
| 1348 | static int m_show(struct seq_file *m, void *v) |
| 1349 | { |
| 1350 | struct proc_mounts *p = m->private; |
| 1351 | struct mount *r = v; |
| 1352 | return p->show(m, &r->mnt); |
| 1353 | } |
| 1354 | |
| 1355 | const struct seq_operations mounts_op = { |
| 1356 | .start = m_start, |
| 1357 | .next = m_next, |
| 1358 | .stop = m_stop, |
| 1359 | .show = m_show, |
| 1360 | }; |
| 1361 | |
| 1362 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) |
| 1363 | { |
| 1364 | down_read(&namespace_sem); |
| 1365 | lock_ns_list(ns); |
| 1366 | list_del(&cursor->mnt_list); |
| 1367 | unlock_ns_list(ns); |
| 1368 | up_read(&namespace_sem); |
| 1369 | } |
| 1370 | #endif /* CONFIG_PROC_FS */ |
| 1371 | |
| 1372 | /** |
| 1373 | * may_umount_tree - check if a mount tree is busy |
| 1374 | * @m: root of mount tree |
| 1375 | * |
| 1376 | * This is called to check if a tree of mounts has any |
| 1377 | * open files, pwds, chroots or sub mounts that are |
| 1378 | * busy. |
| 1379 | */ |
| 1380 | int may_umount_tree(struct vfsmount *m) |
| 1381 | { |
| 1382 | struct mount *mnt = real_mount(m); |
| 1383 | int actual_refs = 0; |
| 1384 | int minimum_refs = 0; |
| 1385 | struct mount *p; |
| 1386 | BUG_ON(!m); |
| 1387 | |
| 1388 | /* write lock needed for mnt_get_count */ |
| 1389 | lock_mount_hash(); |
| 1390 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
| 1391 | actual_refs += mnt_get_count(p); |
| 1392 | minimum_refs += 2; |
| 1393 | } |
| 1394 | unlock_mount_hash(); |
| 1395 | |
| 1396 | if (actual_refs > minimum_refs) |
| 1397 | return 0; |
| 1398 | |
| 1399 | return 1; |
| 1400 | } |
| 1401 | |
| 1402 | EXPORT_SYMBOL(may_umount_tree); |
| 1403 | |
| 1404 | /** |
| 1405 | * may_umount - check if a mount point is busy |
| 1406 | * @mnt: root of mount |
| 1407 | * |
| 1408 | * This is called to check if a mount point has any |
| 1409 | * open files, pwds, chroots or sub mounts. If the |
| 1410 | * mount has sub mounts this will return busy |
| 1411 | * regardless of whether the sub mounts are busy. |
| 1412 | * |
| 1413 | * Doesn't take quota and stuff into account. IOW, in some cases it will |
| 1414 | * give false negatives. The main reason why it's here is that we need |
| 1415 | * a non-destructive way to look for easily umountable filesystems. |
| 1416 | */ |
| 1417 | int may_umount(struct vfsmount *mnt) |
| 1418 | { |
| 1419 | int ret = 1; |
| 1420 | down_read(&namespace_sem); |
| 1421 | lock_mount_hash(); |
| 1422 | if (propagate_mount_busy(real_mount(mnt), 2)) |
| 1423 | ret = 0; |
| 1424 | unlock_mount_hash(); |
| 1425 | up_read(&namespace_sem); |
| 1426 | return ret; |
| 1427 | } |
| 1428 | |
| 1429 | EXPORT_SYMBOL(may_umount); |
| 1430 | |
| 1431 | static void namespace_unlock(void) |
| 1432 | { |
| 1433 | struct hlist_head head; |
| 1434 | struct hlist_node *p; |
| 1435 | struct mount *m; |
| 1436 | LIST_HEAD(list); |
| 1437 | |
| 1438 | hlist_move_list(&unmounted, &head); |
| 1439 | list_splice_init(&ex_mountpoints, &list); |
| 1440 | |
| 1441 | up_write(&namespace_sem); |
| 1442 | |
| 1443 | shrink_dentry_list(&list); |
| 1444 | |
| 1445 | if (likely(hlist_empty(&head))) |
| 1446 | return; |
| 1447 | |
| 1448 | synchronize_rcu_expedited(); |
| 1449 | |
| 1450 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
| 1451 | hlist_del(&m->mnt_umount); |
| 1452 | mntput(&m->mnt); |
| 1453 | } |
| 1454 | } |
| 1455 | |
| 1456 | static inline void namespace_lock(void) |
| 1457 | { |
| 1458 | down_write(&namespace_sem); |
| 1459 | } |
| 1460 | |
| 1461 | enum umount_tree_flags { |
| 1462 | UMOUNT_SYNC = 1, |
| 1463 | UMOUNT_PROPAGATE = 2, |
| 1464 | UMOUNT_CONNECTED = 4, |
| 1465 | }; |
| 1466 | |
| 1467 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) |
| 1468 | { |
| 1469 | /* Leaving mounts connected is only valid for lazy umounts */ |
| 1470 | if (how & UMOUNT_SYNC) |
| 1471 | return true; |
| 1472 | |
| 1473 | /* A mount without a parent has nothing to be connected to */ |
| 1474 | if (!mnt_has_parent(mnt)) |
| 1475 | return true; |
| 1476 | |
| 1477 | /* Because the reference counting rules change when mounts are |
| 1478 | * unmounted and connected, umounted mounts may not be |
| 1479 | * connected to mounted mounts. |
| 1480 | */ |
| 1481 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) |
| 1482 | return true; |
| 1483 | |
| 1484 | /* Has it been requested that the mount remain connected? */ |
| 1485 | if (how & UMOUNT_CONNECTED) |
| 1486 | return false; |
| 1487 | |
| 1488 | /* Is the mount locked such that it needs to remain connected? */ |
| 1489 | if (IS_MNT_LOCKED(mnt)) |
| 1490 | return false; |
| 1491 | |
| 1492 | /* By default disconnect the mount */ |
| 1493 | return true; |
| 1494 | } |
| 1495 | |
| 1496 | /* |
| 1497 | * mount_lock must be held |
| 1498 | * namespace_sem must be held for write |
| 1499 | */ |
| 1500 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
| 1501 | { |
| 1502 | LIST_HEAD(tmp_list); |
| 1503 | struct mount *p; |
| 1504 | |
| 1505 | if (how & UMOUNT_PROPAGATE) |
| 1506 | propagate_mount_unlock(mnt); |
| 1507 | |
| 1508 | /* Gather the mounts to umount */ |
| 1509 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
| 1510 | p->mnt.mnt_flags |= MNT_UMOUNT; |
| 1511 | list_move(&p->mnt_list, &tmp_list); |
| 1512 | } |
| 1513 | |
| 1514 | /* Hide the mounts from mnt_mounts */ |
| 1515 | list_for_each_entry(p, &tmp_list, mnt_list) { |
| 1516 | list_del_init(&p->mnt_child); |
| 1517 | } |
| 1518 | |
| 1519 | /* Add propogated mounts to the tmp_list */ |
| 1520 | if (how & UMOUNT_PROPAGATE) |
| 1521 | propagate_umount(&tmp_list); |
| 1522 | |
| 1523 | while (!list_empty(&tmp_list)) { |
| 1524 | struct mnt_namespace *ns; |
| 1525 | bool disconnect; |
| 1526 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
| 1527 | list_del_init(&p->mnt_expire); |
| 1528 | list_del_init(&p->mnt_list); |
| 1529 | ns = p->mnt_ns; |
| 1530 | if (ns) { |
| 1531 | ns->mounts--; |
| 1532 | __touch_mnt_namespace(ns); |
| 1533 | } |
| 1534 | p->mnt_ns = NULL; |
| 1535 | if (how & UMOUNT_SYNC) |
| 1536 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
| 1537 | |
| 1538 | disconnect = disconnect_mount(p, how); |
| 1539 | if (mnt_has_parent(p)) { |
| 1540 | mnt_add_count(p->mnt_parent, -1); |
| 1541 | if (!disconnect) { |
| 1542 | /* Don't forget about p */ |
| 1543 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); |
| 1544 | } else { |
| 1545 | umount_mnt(p); |
| 1546 | } |
| 1547 | } |
| 1548 | change_mnt_propagation(p, MS_PRIVATE); |
| 1549 | if (disconnect) |
| 1550 | hlist_add_head(&p->mnt_umount, &unmounted); |
| 1551 | } |
| 1552 | } |
| 1553 | |
| 1554 | static void shrink_submounts(struct mount *mnt); |
| 1555 | |
| 1556 | static int do_umount_root(struct super_block *sb) |
| 1557 | { |
| 1558 | int ret = 0; |
| 1559 | |
| 1560 | down_write(&sb->s_umount); |
| 1561 | if (!sb_rdonly(sb)) { |
| 1562 | struct fs_context *fc; |
| 1563 | |
| 1564 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, |
| 1565 | SB_RDONLY); |
| 1566 | if (IS_ERR(fc)) { |
| 1567 | ret = PTR_ERR(fc); |
| 1568 | } else { |
| 1569 | ret = parse_monolithic_mount_data(fc, NULL); |
| 1570 | if (!ret) |
| 1571 | ret = reconfigure_super(fc); |
| 1572 | put_fs_context(fc); |
| 1573 | } |
| 1574 | } |
| 1575 | up_write(&sb->s_umount); |
| 1576 | return ret; |
| 1577 | } |
| 1578 | |
| 1579 | static int do_umount(struct mount *mnt, int flags) |
| 1580 | { |
| 1581 | struct super_block *sb = mnt->mnt.mnt_sb; |
| 1582 | int retval; |
| 1583 | |
| 1584 | retval = security_sb_umount(&mnt->mnt, flags); |
| 1585 | if (retval) |
| 1586 | return retval; |
| 1587 | |
| 1588 | /* |
| 1589 | * Allow userspace to request a mountpoint be expired rather than |
| 1590 | * unmounting unconditionally. Unmount only happens if: |
| 1591 | * (1) the mark is already set (the mark is cleared by mntput()) |
| 1592 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] |
| 1593 | */ |
| 1594 | if (flags & MNT_EXPIRE) { |
| 1595 | if (&mnt->mnt == current->fs->root.mnt || |
| 1596 | flags & (MNT_FORCE | MNT_DETACH)) |
| 1597 | return -EINVAL; |
| 1598 | |
| 1599 | /* |
| 1600 | * probably don't strictly need the lock here if we examined |
| 1601 | * all race cases, but it's a slowpath. |
| 1602 | */ |
| 1603 | lock_mount_hash(); |
| 1604 | if (mnt_get_count(mnt) != 2) { |
| 1605 | unlock_mount_hash(); |
| 1606 | return -EBUSY; |
| 1607 | } |
| 1608 | unlock_mount_hash(); |
| 1609 | |
| 1610 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
| 1611 | return -EAGAIN; |
| 1612 | } |
| 1613 | |
| 1614 | /* |
| 1615 | * If we may have to abort operations to get out of this |
| 1616 | * mount, and they will themselves hold resources we must |
| 1617 | * allow the fs to do things. In the Unix tradition of |
| 1618 | * 'Gee thats tricky lets do it in userspace' the umount_begin |
| 1619 | * might fail to complete on the first run through as other tasks |
| 1620 | * must return, and the like. Thats for the mount program to worry |
| 1621 | * about for the moment. |
| 1622 | */ |
| 1623 | |
| 1624 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
| 1625 | sb->s_op->umount_begin(sb); |
| 1626 | } |
| 1627 | |
| 1628 | /* |
| 1629 | * No sense to grab the lock for this test, but test itself looks |
| 1630 | * somewhat bogus. Suggestions for better replacement? |
| 1631 | * Ho-hum... In principle, we might treat that as umount + switch |
| 1632 | * to rootfs. GC would eventually take care of the old vfsmount. |
| 1633 | * Actually it makes sense, especially if rootfs would contain a |
| 1634 | * /reboot - static binary that would close all descriptors and |
| 1635 | * call reboot(9). Then init(8) could umount root and exec /reboot. |
| 1636 | */ |
| 1637 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
| 1638 | /* |
| 1639 | * Special case for "unmounting" root ... |
| 1640 | * we just try to remount it readonly. |
| 1641 | */ |
| 1642 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
| 1643 | return -EPERM; |
| 1644 | return do_umount_root(sb); |
| 1645 | } |
| 1646 | |
| 1647 | namespace_lock(); |
| 1648 | lock_mount_hash(); |
| 1649 | |
| 1650 | /* Recheck MNT_LOCKED with the locks held */ |
| 1651 | retval = -EINVAL; |
| 1652 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
| 1653 | goto out; |
| 1654 | |
| 1655 | event++; |
| 1656 | if (flags & MNT_DETACH) { |
| 1657 | if (!list_empty(&mnt->mnt_list)) |
| 1658 | umount_tree(mnt, UMOUNT_PROPAGATE); |
| 1659 | retval = 0; |
| 1660 | } else { |
| 1661 | shrink_submounts(mnt); |
| 1662 | retval = -EBUSY; |
| 1663 | if (!propagate_mount_busy(mnt, 2)) { |
| 1664 | if (!list_empty(&mnt->mnt_list)) |
| 1665 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
| 1666 | retval = 0; |
| 1667 | } |
| 1668 | } |
| 1669 | out: |
| 1670 | unlock_mount_hash(); |
| 1671 | namespace_unlock(); |
| 1672 | return retval; |
| 1673 | } |
| 1674 | |
| 1675 | /* |
| 1676 | * __detach_mounts - lazily unmount all mounts on the specified dentry |
| 1677 | * |
| 1678 | * During unlink, rmdir, and d_drop it is possible to loose the path |
| 1679 | * to an existing mountpoint, and wind up leaking the mount. |
| 1680 | * detach_mounts allows lazily unmounting those mounts instead of |
| 1681 | * leaking them. |
| 1682 | * |
| 1683 | * The caller may hold dentry->d_inode->i_mutex. |
| 1684 | */ |
| 1685 | void __detach_mounts(struct dentry *dentry) |
| 1686 | { |
| 1687 | struct mountpoint *mp; |
| 1688 | struct mount *mnt; |
| 1689 | |
| 1690 | namespace_lock(); |
| 1691 | lock_mount_hash(); |
| 1692 | mp = lookup_mountpoint(dentry); |
| 1693 | if (!mp) |
| 1694 | goto out_unlock; |
| 1695 | |
| 1696 | event++; |
| 1697 | while (!hlist_empty(&mp->m_list)) { |
| 1698 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); |
| 1699 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
| 1700 | umount_mnt(mnt); |
| 1701 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
| 1702 | } |
| 1703 | else umount_tree(mnt, UMOUNT_CONNECTED); |
| 1704 | } |
| 1705 | put_mountpoint(mp); |
| 1706 | out_unlock: |
| 1707 | unlock_mount_hash(); |
| 1708 | namespace_unlock(); |
| 1709 | } |
| 1710 | |
| 1711 | /* |
| 1712 | * Is the caller allowed to modify his namespace? |
| 1713 | */ |
| 1714 | static inline bool may_mount(void) |
| 1715 | { |
| 1716 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); |
| 1717 | } |
| 1718 | |
| 1719 | static void warn_mandlock(void) |
| 1720 | { |
| 1721 | pr_warn_once("=======================================================\n" |
| 1722 | "WARNING: The mand mount option has been deprecated and\n" |
| 1723 | " and is ignored by this kernel. Remove the mand\n" |
| 1724 | " option from the mount to silence this warning.\n" |
| 1725 | "=======================================================\n"); |
| 1726 | } |
| 1727 | |
| 1728 | static int can_umount(const struct path *path, int flags) |
| 1729 | { |
| 1730 | struct mount *mnt = real_mount(path->mnt); |
| 1731 | |
| 1732 | if (!may_mount()) |
| 1733 | return -EPERM; |
| 1734 | if (path->dentry != path->mnt->mnt_root) |
| 1735 | return -EINVAL; |
| 1736 | if (!check_mnt(mnt)) |
| 1737 | return -EINVAL; |
| 1738 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
| 1739 | return -EINVAL; |
| 1740 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
| 1741 | return -EPERM; |
| 1742 | return 0; |
| 1743 | } |
| 1744 | |
| 1745 | // caller is responsible for flags being sane |
| 1746 | int path_umount(struct path *path, int flags) |
| 1747 | { |
| 1748 | struct mount *mnt = real_mount(path->mnt); |
| 1749 | int ret; |
| 1750 | |
| 1751 | ret = can_umount(path, flags); |
| 1752 | if (!ret) |
| 1753 | ret = do_umount(mnt, flags); |
| 1754 | |
| 1755 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
| 1756 | dput(path->dentry); |
| 1757 | mntput_no_expire(mnt); |
| 1758 | return ret; |
| 1759 | } |
| 1760 | |
| 1761 | static int ksys_umount(char __user *name, int flags) |
| 1762 | { |
| 1763 | int lookup_flags = LOOKUP_MOUNTPOINT; |
| 1764 | struct path path; |
| 1765 | int ret; |
| 1766 | |
| 1767 | // basic validity checks done first |
| 1768 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
| 1769 | return -EINVAL; |
| 1770 | |
| 1771 | if (!(flags & UMOUNT_NOFOLLOW)) |
| 1772 | lookup_flags |= LOOKUP_FOLLOW; |
| 1773 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); |
| 1774 | if (ret) |
| 1775 | return ret; |
| 1776 | return path_umount(&path, flags); |
| 1777 | } |
| 1778 | |
| 1779 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
| 1780 | { |
| 1781 | return ksys_umount(name, flags); |
| 1782 | } |
| 1783 | |
| 1784 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
| 1785 | |
| 1786 | /* |
| 1787 | * The 2.0 compatible umount. No flags. |
| 1788 | */ |
| 1789 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
| 1790 | { |
| 1791 | return ksys_umount(name, 0); |
| 1792 | } |
| 1793 | |
| 1794 | #endif |
| 1795 | |
| 1796 | static bool is_mnt_ns_file(struct dentry *dentry) |
| 1797 | { |
| 1798 | /* Is this a proxy for a mount namespace? */ |
| 1799 | return dentry->d_op == &ns_dentry_operations && |
| 1800 | dentry->d_fsdata == &mntns_operations; |
| 1801 | } |
| 1802 | |
| 1803 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
| 1804 | { |
| 1805 | return container_of(ns, struct mnt_namespace, ns); |
| 1806 | } |
| 1807 | |
| 1808 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
| 1809 | { |
| 1810 | return &mnt->ns; |
| 1811 | } |
| 1812 | |
| 1813 | static bool mnt_ns_loop(struct dentry *dentry) |
| 1814 | { |
| 1815 | /* Could bind mounting the mount namespace inode cause a |
| 1816 | * mount namespace loop? |
| 1817 | */ |
| 1818 | struct mnt_namespace *mnt_ns; |
| 1819 | if (!is_mnt_ns_file(dentry)) |
| 1820 | return false; |
| 1821 | |
| 1822 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
| 1823 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
| 1824 | } |
| 1825 | |
| 1826 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
| 1827 | int flag) |
| 1828 | { |
| 1829 | struct mount *res, *p, *q, *r, *parent; |
| 1830 | |
| 1831 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
| 1832 | return ERR_PTR(-EINVAL); |
| 1833 | |
| 1834 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) |
| 1835 | return ERR_PTR(-EINVAL); |
| 1836 | |
| 1837 | res = q = clone_mnt(mnt, dentry, flag); |
| 1838 | if (IS_ERR(q)) |
| 1839 | return q; |
| 1840 | |
| 1841 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
| 1842 | |
| 1843 | p = mnt; |
| 1844 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
| 1845 | struct mount *s; |
| 1846 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
| 1847 | continue; |
| 1848 | |
| 1849 | for (s = r; s; s = next_mnt(s, r)) { |
| 1850 | if (!(flag & CL_COPY_UNBINDABLE) && |
| 1851 | IS_MNT_UNBINDABLE(s)) { |
| 1852 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
| 1853 | /* Both unbindable and locked. */ |
| 1854 | q = ERR_PTR(-EPERM); |
| 1855 | goto out; |
| 1856 | } else { |
| 1857 | s = skip_mnt_tree(s); |
| 1858 | continue; |
| 1859 | } |
| 1860 | } |
| 1861 | if (!(flag & CL_COPY_MNT_NS_FILE) && |
| 1862 | is_mnt_ns_file(s->mnt.mnt_root)) { |
| 1863 | s = skip_mnt_tree(s); |
| 1864 | continue; |
| 1865 | } |
| 1866 | while (p != s->mnt_parent) { |
| 1867 | p = p->mnt_parent; |
| 1868 | q = q->mnt_parent; |
| 1869 | } |
| 1870 | p = s; |
| 1871 | parent = q; |
| 1872 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
| 1873 | if (IS_ERR(q)) |
| 1874 | goto out; |
| 1875 | lock_mount_hash(); |
| 1876 | list_add_tail(&q->mnt_list, &res->mnt_list); |
| 1877 | attach_mnt(q, parent, p->mnt_mp); |
| 1878 | unlock_mount_hash(); |
| 1879 | } |
| 1880 | } |
| 1881 | return res; |
| 1882 | out: |
| 1883 | if (res) { |
| 1884 | lock_mount_hash(); |
| 1885 | umount_tree(res, UMOUNT_SYNC); |
| 1886 | unlock_mount_hash(); |
| 1887 | } |
| 1888 | return q; |
| 1889 | } |
| 1890 | |
| 1891 | /* Caller should check returned pointer for errors */ |
| 1892 | |
| 1893 | struct vfsmount *collect_mounts(const struct path *path) |
| 1894 | { |
| 1895 | struct mount *tree; |
| 1896 | namespace_lock(); |
| 1897 | if (!check_mnt(real_mount(path->mnt))) |
| 1898 | tree = ERR_PTR(-EINVAL); |
| 1899 | else |
| 1900 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
| 1901 | CL_COPY_ALL | CL_PRIVATE); |
| 1902 | namespace_unlock(); |
| 1903 | if (IS_ERR(tree)) |
| 1904 | return ERR_CAST(tree); |
| 1905 | return &tree->mnt; |
| 1906 | } |
| 1907 | |
| 1908 | static void free_mnt_ns(struct mnt_namespace *); |
| 1909 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); |
| 1910 | |
| 1911 | void dissolve_on_fput(struct vfsmount *mnt) |
| 1912 | { |
| 1913 | struct mnt_namespace *ns; |
| 1914 | namespace_lock(); |
| 1915 | lock_mount_hash(); |
| 1916 | ns = real_mount(mnt)->mnt_ns; |
| 1917 | if (ns) { |
| 1918 | if (is_anon_ns(ns)) |
| 1919 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); |
| 1920 | else |
| 1921 | ns = NULL; |
| 1922 | } |
| 1923 | unlock_mount_hash(); |
| 1924 | namespace_unlock(); |
| 1925 | if (ns) |
| 1926 | free_mnt_ns(ns); |
| 1927 | } |
| 1928 | |
| 1929 | void drop_collected_mounts(struct vfsmount *mnt) |
| 1930 | { |
| 1931 | namespace_lock(); |
| 1932 | lock_mount_hash(); |
| 1933 | umount_tree(real_mount(mnt), 0); |
| 1934 | unlock_mount_hash(); |
| 1935 | namespace_unlock(); |
| 1936 | } |
| 1937 | |
| 1938 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
| 1939 | { |
| 1940 | struct mount *child; |
| 1941 | |
| 1942 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { |
| 1943 | if (!is_subdir(child->mnt_mountpoint, dentry)) |
| 1944 | continue; |
| 1945 | |
| 1946 | if (child->mnt.mnt_flags & MNT_LOCKED) |
| 1947 | return true; |
| 1948 | } |
| 1949 | return false; |
| 1950 | } |
| 1951 | |
| 1952 | /** |
| 1953 | * clone_private_mount - create a private clone of a path |
| 1954 | * @path: path to clone |
| 1955 | * |
| 1956 | * This creates a new vfsmount, which will be the clone of @path. The new mount |
| 1957 | * will not be attached anywhere in the namespace and will be private (i.e. |
| 1958 | * changes to the originating mount won't be propagated into this). |
| 1959 | * |
| 1960 | * Release with mntput(). |
| 1961 | */ |
| 1962 | struct vfsmount *clone_private_mount(const struct path *path) |
| 1963 | { |
| 1964 | struct mount *old_mnt = real_mount(path->mnt); |
| 1965 | struct mount *new_mnt; |
| 1966 | |
| 1967 | down_read(&namespace_sem); |
| 1968 | if (IS_MNT_UNBINDABLE(old_mnt)) |
| 1969 | goto invalid; |
| 1970 | |
| 1971 | if (!check_mnt(old_mnt)) |
| 1972 | goto invalid; |
| 1973 | |
| 1974 | if (has_locked_children(old_mnt, path->dentry)) |
| 1975 | goto invalid; |
| 1976 | |
| 1977 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
| 1978 | up_read(&namespace_sem); |
| 1979 | |
| 1980 | if (IS_ERR(new_mnt)) |
| 1981 | return ERR_CAST(new_mnt); |
| 1982 | |
| 1983 | /* Longterm mount to be removed by kern_unmount*() */ |
| 1984 | new_mnt->mnt_ns = MNT_NS_INTERNAL; |
| 1985 | |
| 1986 | return &new_mnt->mnt; |
| 1987 | |
| 1988 | invalid: |
| 1989 | up_read(&namespace_sem); |
| 1990 | return ERR_PTR(-EINVAL); |
| 1991 | } |
| 1992 | EXPORT_SYMBOL_GPL(clone_private_mount); |
| 1993 | |
| 1994 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
| 1995 | struct vfsmount *root) |
| 1996 | { |
| 1997 | struct mount *mnt; |
| 1998 | int res = f(root, arg); |
| 1999 | if (res) |
| 2000 | return res; |
| 2001 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
| 2002 | res = f(&mnt->mnt, arg); |
| 2003 | if (res) |
| 2004 | return res; |
| 2005 | } |
| 2006 | return 0; |
| 2007 | } |
| 2008 | |
| 2009 | static void lock_mnt_tree(struct mount *mnt) |
| 2010 | { |
| 2011 | struct mount *p; |
| 2012 | |
| 2013 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
| 2014 | int flags = p->mnt.mnt_flags; |
| 2015 | /* Don't allow unprivileged users to change mount flags */ |
| 2016 | flags |= MNT_LOCK_ATIME; |
| 2017 | |
| 2018 | if (flags & MNT_READONLY) |
| 2019 | flags |= MNT_LOCK_READONLY; |
| 2020 | |
| 2021 | if (flags & MNT_NODEV) |
| 2022 | flags |= MNT_LOCK_NODEV; |
| 2023 | |
| 2024 | if (flags & MNT_NOSUID) |
| 2025 | flags |= MNT_LOCK_NOSUID; |
| 2026 | |
| 2027 | if (flags & MNT_NOEXEC) |
| 2028 | flags |= MNT_LOCK_NOEXEC; |
| 2029 | /* Don't allow unprivileged users to reveal what is under a mount */ |
| 2030 | if (list_empty(&p->mnt_expire)) |
| 2031 | flags |= MNT_LOCKED; |
| 2032 | p->mnt.mnt_flags = flags; |
| 2033 | } |
| 2034 | } |
| 2035 | |
| 2036 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
| 2037 | { |
| 2038 | struct mount *p; |
| 2039 | |
| 2040 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
| 2041 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
| 2042 | mnt_release_group_id(p); |
| 2043 | } |
| 2044 | } |
| 2045 | |
| 2046 | static int invent_group_ids(struct mount *mnt, bool recurse) |
| 2047 | { |
| 2048 | struct mount *p; |
| 2049 | |
| 2050 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
| 2051 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
| 2052 | int err = mnt_alloc_group_id(p); |
| 2053 | if (err) { |
| 2054 | cleanup_group_ids(mnt, p); |
| 2055 | return err; |
| 2056 | } |
| 2057 | } |
| 2058 | } |
| 2059 | |
| 2060 | return 0; |
| 2061 | } |
| 2062 | |
| 2063 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
| 2064 | { |
| 2065 | unsigned int max = READ_ONCE(sysctl_mount_max); |
| 2066 | unsigned int mounts = 0, old, pending, sum; |
| 2067 | struct mount *p; |
| 2068 | |
| 2069 | for (p = mnt; p; p = next_mnt(p, mnt)) |
| 2070 | mounts++; |
| 2071 | |
| 2072 | old = ns->mounts; |
| 2073 | pending = ns->pending_mounts; |
| 2074 | sum = old + pending; |
| 2075 | if ((old > sum) || |
| 2076 | (pending > sum) || |
| 2077 | (max < sum) || |
| 2078 | (mounts > (max - sum))) |
| 2079 | return -ENOSPC; |
| 2080 | |
| 2081 | ns->pending_mounts = pending + mounts; |
| 2082 | return 0; |
| 2083 | } |
| 2084 | |
| 2085 | /* |
| 2086 | * @source_mnt : mount tree to be attached |
| 2087 | * @nd : place the mount tree @source_mnt is attached |
| 2088 | * @parent_nd : if non-null, detach the source_mnt from its parent and |
| 2089 | * store the parent mount and mountpoint dentry. |
| 2090 | * (done when source_mnt is moved) |
| 2091 | * |
| 2092 | * NOTE: in the table below explains the semantics when a source mount |
| 2093 | * of a given type is attached to a destination mount of a given type. |
| 2094 | * --------------------------------------------------------------------------- |
| 2095 | * | BIND MOUNT OPERATION | |
| 2096 | * |************************************************************************** |
| 2097 | * | source-->| shared | private | slave | unbindable | |
| 2098 | * | dest | | | | | |
| 2099 | * | | | | | | | |
| 2100 | * | v | | | | | |
| 2101 | * |************************************************************************** |
| 2102 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | |
| 2103 | * | | | | | | |
| 2104 | * |non-shared| shared (+) | private | slave (*) | invalid | |
| 2105 | * *************************************************************************** |
| 2106 | * A bind operation clones the source mount and mounts the clone on the |
| 2107 | * destination mount. |
| 2108 | * |
| 2109 | * (++) the cloned mount is propagated to all the mounts in the propagation |
| 2110 | * tree of the destination mount and the cloned mount is added to |
| 2111 | * the peer group of the source mount. |
| 2112 | * (+) the cloned mount is created under the destination mount and is marked |
| 2113 | * as shared. The cloned mount is added to the peer group of the source |
| 2114 | * mount. |
| 2115 | * (+++) the mount is propagated to all the mounts in the propagation tree |
| 2116 | * of the destination mount and the cloned mount is made slave |
| 2117 | * of the same master as that of the source mount. The cloned mount |
| 2118 | * is marked as 'shared and slave'. |
| 2119 | * (*) the cloned mount is made a slave of the same master as that of the |
| 2120 | * source mount. |
| 2121 | * |
| 2122 | * --------------------------------------------------------------------------- |
| 2123 | * | MOVE MOUNT OPERATION | |
| 2124 | * |************************************************************************** |
| 2125 | * | source-->| shared | private | slave | unbindable | |
| 2126 | * | dest | | | | | |
| 2127 | * | | | | | | | |
| 2128 | * | v | | | | | |
| 2129 | * |************************************************************************** |
| 2130 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | |
| 2131 | * | | | | | | |
| 2132 | * |non-shared| shared (+*) | private | slave (*) | unbindable | |
| 2133 | * *************************************************************************** |
| 2134 | * |
| 2135 | * (+) the mount is moved to the destination. And is then propagated to |
| 2136 | * all the mounts in the propagation tree of the destination mount. |
| 2137 | * (+*) the mount is moved to the destination. |
| 2138 | * (+++) the mount is moved to the destination and is then propagated to |
| 2139 | * all the mounts belonging to the destination mount's propagation tree. |
| 2140 | * the mount is marked as 'shared and slave'. |
| 2141 | * (*) the mount continues to be a slave at the new location. |
| 2142 | * |
| 2143 | * if the source mount is a tree, the operations explained above is |
| 2144 | * applied to each mount in the tree. |
| 2145 | * Must be called without spinlocks held, since this function can sleep |
| 2146 | * in allocations. |
| 2147 | */ |
| 2148 | static int attach_recursive_mnt(struct mount *source_mnt, |
| 2149 | struct mount *dest_mnt, |
| 2150 | struct mountpoint *dest_mp, |
| 2151 | bool moving) |
| 2152 | { |
| 2153 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
| 2154 | HLIST_HEAD(tree_list); |
| 2155 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
| 2156 | struct mountpoint *smp; |
| 2157 | struct mount *child, *p; |
| 2158 | struct hlist_node *n; |
| 2159 | int err; |
| 2160 | |
| 2161 | /* Preallocate a mountpoint in case the new mounts need |
| 2162 | * to be tucked under other mounts. |
| 2163 | */ |
| 2164 | smp = get_mountpoint(source_mnt->mnt.mnt_root); |
| 2165 | if (IS_ERR(smp)) |
| 2166 | return PTR_ERR(smp); |
| 2167 | |
| 2168 | /* Is there space to add these mounts to the mount namespace? */ |
| 2169 | if (!moving) { |
| 2170 | err = count_mounts(ns, source_mnt); |
| 2171 | if (err) |
| 2172 | goto out; |
| 2173 | } |
| 2174 | |
| 2175 | if (IS_MNT_SHARED(dest_mnt)) { |
| 2176 | err = invent_group_ids(source_mnt, true); |
| 2177 | if (err) |
| 2178 | goto out; |
| 2179 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
| 2180 | lock_mount_hash(); |
| 2181 | if (err) |
| 2182 | goto out_cleanup_ids; |
| 2183 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
| 2184 | set_mnt_shared(p); |
| 2185 | } else { |
| 2186 | lock_mount_hash(); |
| 2187 | } |
| 2188 | if (moving) { |
| 2189 | unhash_mnt(source_mnt); |
| 2190 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
| 2191 | touch_mnt_namespace(source_mnt->mnt_ns); |
| 2192 | } else { |
| 2193 | if (source_mnt->mnt_ns) { |
| 2194 | /* move from anon - the caller will destroy */ |
| 2195 | list_del_init(&source_mnt->mnt_ns->list); |
| 2196 | } |
| 2197 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
| 2198 | commit_tree(source_mnt); |
| 2199 | } |
| 2200 | |
| 2201 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
| 2202 | struct mount *q; |
| 2203 | hlist_del_init(&child->mnt_hash); |
| 2204 | q = __lookup_mnt(&child->mnt_parent->mnt, |
| 2205 | child->mnt_mountpoint); |
| 2206 | if (q) |
| 2207 | mnt_change_mountpoint(child, smp, q); |
| 2208 | /* Notice when we are propagating across user namespaces */ |
| 2209 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) |
| 2210 | lock_mnt_tree(child); |
| 2211 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
| 2212 | commit_tree(child); |
| 2213 | } |
| 2214 | put_mountpoint(smp); |
| 2215 | unlock_mount_hash(); |
| 2216 | |
| 2217 | return 0; |
| 2218 | |
| 2219 | out_cleanup_ids: |
| 2220 | while (!hlist_empty(&tree_list)) { |
| 2221 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); |
| 2222 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
| 2223 | umount_tree(child, UMOUNT_SYNC); |
| 2224 | } |
| 2225 | unlock_mount_hash(); |
| 2226 | cleanup_group_ids(source_mnt, NULL); |
| 2227 | out: |
| 2228 | ns->pending_mounts = 0; |
| 2229 | |
| 2230 | read_seqlock_excl(&mount_lock); |
| 2231 | put_mountpoint(smp); |
| 2232 | read_sequnlock_excl(&mount_lock); |
| 2233 | |
| 2234 | return err; |
| 2235 | } |
| 2236 | |
| 2237 | static struct mountpoint *lock_mount(struct path *path) |
| 2238 | { |
| 2239 | struct vfsmount *mnt; |
| 2240 | struct dentry *dentry = path->dentry; |
| 2241 | retry: |
| 2242 | inode_lock(dentry->d_inode); |
| 2243 | if (unlikely(cant_mount(dentry))) { |
| 2244 | inode_unlock(dentry->d_inode); |
| 2245 | return ERR_PTR(-ENOENT); |
| 2246 | } |
| 2247 | namespace_lock(); |
| 2248 | mnt = lookup_mnt(path); |
| 2249 | if (likely(!mnt)) { |
| 2250 | struct mountpoint *mp = get_mountpoint(dentry); |
| 2251 | if (IS_ERR(mp)) { |
| 2252 | namespace_unlock(); |
| 2253 | inode_unlock(dentry->d_inode); |
| 2254 | return mp; |
| 2255 | } |
| 2256 | return mp; |
| 2257 | } |
| 2258 | namespace_unlock(); |
| 2259 | inode_unlock(path->dentry->d_inode); |
| 2260 | path_put(path); |
| 2261 | path->mnt = mnt; |
| 2262 | dentry = path->dentry = dget(mnt->mnt_root); |
| 2263 | goto retry; |
| 2264 | } |
| 2265 | |
| 2266 | static void unlock_mount(struct mountpoint *where) |
| 2267 | { |
| 2268 | struct dentry *dentry = where->m_dentry; |
| 2269 | |
| 2270 | read_seqlock_excl(&mount_lock); |
| 2271 | put_mountpoint(where); |
| 2272 | read_sequnlock_excl(&mount_lock); |
| 2273 | |
| 2274 | namespace_unlock(); |
| 2275 | inode_unlock(dentry->d_inode); |
| 2276 | } |
| 2277 | |
| 2278 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
| 2279 | { |
| 2280 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
| 2281 | return -EINVAL; |
| 2282 | |
| 2283 | if (d_is_dir(mp->m_dentry) != |
| 2284 | d_is_dir(mnt->mnt.mnt_root)) |
| 2285 | return -ENOTDIR; |
| 2286 | |
| 2287 | return attach_recursive_mnt(mnt, p, mp, false); |
| 2288 | } |
| 2289 | |
| 2290 | /* |
| 2291 | * Sanity check the flags to change_mnt_propagation. |
| 2292 | */ |
| 2293 | |
| 2294 | static int flags_to_propagation_type(int ms_flags) |
| 2295 | { |
| 2296 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
| 2297 | |
| 2298 | /* Fail if any non-propagation flags are set */ |
| 2299 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
| 2300 | return 0; |
| 2301 | /* Only one propagation flag should be set */ |
| 2302 | if (!is_power_of_2(type)) |
| 2303 | return 0; |
| 2304 | return type; |
| 2305 | } |
| 2306 | |
| 2307 | /* |
| 2308 | * recursively change the type of the mountpoint. |
| 2309 | */ |
| 2310 | static int do_change_type(struct path *path, int ms_flags) |
| 2311 | { |
| 2312 | struct mount *m; |
| 2313 | struct mount *mnt = real_mount(path->mnt); |
| 2314 | int recurse = ms_flags & MS_REC; |
| 2315 | int type; |
| 2316 | int err = 0; |
| 2317 | |
| 2318 | if (path->dentry != path->mnt->mnt_root) |
| 2319 | return -EINVAL; |
| 2320 | |
| 2321 | type = flags_to_propagation_type(ms_flags); |
| 2322 | if (!type) |
| 2323 | return -EINVAL; |
| 2324 | |
| 2325 | namespace_lock(); |
| 2326 | if (type == MS_SHARED) { |
| 2327 | err = invent_group_ids(mnt, recurse); |
| 2328 | if (err) |
| 2329 | goto out_unlock; |
| 2330 | } |
| 2331 | |
| 2332 | lock_mount_hash(); |
| 2333 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
| 2334 | change_mnt_propagation(m, type); |
| 2335 | unlock_mount_hash(); |
| 2336 | |
| 2337 | out_unlock: |
| 2338 | namespace_unlock(); |
| 2339 | return err; |
| 2340 | } |
| 2341 | |
| 2342 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
| 2343 | { |
| 2344 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); |
| 2345 | |
| 2346 | if (IS_MNT_UNBINDABLE(old)) |
| 2347 | return mnt; |
| 2348 | |
| 2349 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) |
| 2350 | return mnt; |
| 2351 | |
| 2352 | if (!recurse && has_locked_children(old, old_path->dentry)) |
| 2353 | return mnt; |
| 2354 | |
| 2355 | if (recurse) |
| 2356 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); |
| 2357 | else |
| 2358 | mnt = clone_mnt(old, old_path->dentry, 0); |
| 2359 | |
| 2360 | if (!IS_ERR(mnt)) |
| 2361 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
| 2362 | |
| 2363 | return mnt; |
| 2364 | } |
| 2365 | |
| 2366 | /* |
| 2367 | * do loopback mount. |
| 2368 | */ |
| 2369 | static int do_loopback(struct path *path, const char *old_name, |
| 2370 | int recurse) |
| 2371 | { |
| 2372 | struct path old_path; |
| 2373 | struct mount *mnt = NULL, *parent; |
| 2374 | struct mountpoint *mp; |
| 2375 | int err; |
| 2376 | if (!old_name || !*old_name) |
| 2377 | return -EINVAL; |
| 2378 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
| 2379 | if (err) |
| 2380 | return err; |
| 2381 | |
| 2382 | err = -EINVAL; |
| 2383 | if (mnt_ns_loop(old_path.dentry)) |
| 2384 | goto out; |
| 2385 | |
| 2386 | mp = lock_mount(path); |
| 2387 | if (IS_ERR(mp)) { |
| 2388 | err = PTR_ERR(mp); |
| 2389 | goto out; |
| 2390 | } |
| 2391 | |
| 2392 | parent = real_mount(path->mnt); |
| 2393 | if (!check_mnt(parent)) |
| 2394 | goto out2; |
| 2395 | |
| 2396 | mnt = __do_loopback(&old_path, recurse); |
| 2397 | if (IS_ERR(mnt)) { |
| 2398 | err = PTR_ERR(mnt); |
| 2399 | goto out2; |
| 2400 | } |
| 2401 | |
| 2402 | err = graft_tree(mnt, parent, mp); |
| 2403 | if (err) { |
| 2404 | lock_mount_hash(); |
| 2405 | umount_tree(mnt, UMOUNT_SYNC); |
| 2406 | unlock_mount_hash(); |
| 2407 | } |
| 2408 | out2: |
| 2409 | unlock_mount(mp); |
| 2410 | out: |
| 2411 | path_put(&old_path); |
| 2412 | return err; |
| 2413 | } |
| 2414 | |
| 2415 | static struct file *open_detached_copy(struct path *path, bool recursive) |
| 2416 | { |
| 2417 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
| 2418 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); |
| 2419 | struct mount *mnt, *p; |
| 2420 | struct file *file; |
| 2421 | |
| 2422 | if (IS_ERR(ns)) |
| 2423 | return ERR_CAST(ns); |
| 2424 | |
| 2425 | namespace_lock(); |
| 2426 | mnt = __do_loopback(path, recursive); |
| 2427 | if (IS_ERR(mnt)) { |
| 2428 | namespace_unlock(); |
| 2429 | free_mnt_ns(ns); |
| 2430 | return ERR_CAST(mnt); |
| 2431 | } |
| 2432 | |
| 2433 | lock_mount_hash(); |
| 2434 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
| 2435 | p->mnt_ns = ns; |
| 2436 | ns->mounts++; |
| 2437 | } |
| 2438 | ns->root = mnt; |
| 2439 | list_add_tail(&ns->list, &mnt->mnt_list); |
| 2440 | mntget(&mnt->mnt); |
| 2441 | unlock_mount_hash(); |
| 2442 | namespace_unlock(); |
| 2443 | |
| 2444 | mntput(path->mnt); |
| 2445 | path->mnt = &mnt->mnt; |
| 2446 | file = dentry_open(path, O_PATH, current_cred()); |
| 2447 | if (IS_ERR(file)) |
| 2448 | dissolve_on_fput(path->mnt); |
| 2449 | else |
| 2450 | file->f_mode |= FMODE_NEED_UNMOUNT; |
| 2451 | return file; |
| 2452 | } |
| 2453 | |
| 2454 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
| 2455 | { |
| 2456 | struct file *file; |
| 2457 | struct path path; |
| 2458 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; |
| 2459 | bool detached = flags & OPEN_TREE_CLONE; |
| 2460 | int error; |
| 2461 | int fd; |
| 2462 | |
| 2463 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); |
| 2464 | |
| 2465 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | |
| 2466 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | |
| 2467 | OPEN_TREE_CLOEXEC)) |
| 2468 | return -EINVAL; |
| 2469 | |
| 2470 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) |
| 2471 | return -EINVAL; |
| 2472 | |
| 2473 | if (flags & AT_NO_AUTOMOUNT) |
| 2474 | lookup_flags &= ~LOOKUP_AUTOMOUNT; |
| 2475 | if (flags & AT_SYMLINK_NOFOLLOW) |
| 2476 | lookup_flags &= ~LOOKUP_FOLLOW; |
| 2477 | if (flags & AT_EMPTY_PATH) |
| 2478 | lookup_flags |= LOOKUP_EMPTY; |
| 2479 | |
| 2480 | if (detached && !may_mount()) |
| 2481 | return -EPERM; |
| 2482 | |
| 2483 | fd = get_unused_fd_flags(flags & O_CLOEXEC); |
| 2484 | if (fd < 0) |
| 2485 | return fd; |
| 2486 | |
| 2487 | error = user_path_at(dfd, filename, lookup_flags, &path); |
| 2488 | if (unlikely(error)) { |
| 2489 | file = ERR_PTR(error); |
| 2490 | } else { |
| 2491 | if (detached) |
| 2492 | file = open_detached_copy(&path, flags & AT_RECURSIVE); |
| 2493 | else |
| 2494 | file = dentry_open(&path, O_PATH, current_cred()); |
| 2495 | path_put(&path); |
| 2496 | } |
| 2497 | if (IS_ERR(file)) { |
| 2498 | put_unused_fd(fd); |
| 2499 | return PTR_ERR(file); |
| 2500 | } |
| 2501 | fd_install(fd, file); |
| 2502 | return fd; |
| 2503 | } |
| 2504 | |
| 2505 | /* |
| 2506 | * Don't allow locked mount flags to be cleared. |
| 2507 | * |
| 2508 | * No locks need to be held here while testing the various MNT_LOCK |
| 2509 | * flags because those flags can never be cleared once they are set. |
| 2510 | */ |
| 2511 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) |
| 2512 | { |
| 2513 | unsigned int fl = mnt->mnt.mnt_flags; |
| 2514 | |
| 2515 | if ((fl & MNT_LOCK_READONLY) && |
| 2516 | !(mnt_flags & MNT_READONLY)) |
| 2517 | return false; |
| 2518 | |
| 2519 | if ((fl & MNT_LOCK_NODEV) && |
| 2520 | !(mnt_flags & MNT_NODEV)) |
| 2521 | return false; |
| 2522 | |
| 2523 | if ((fl & MNT_LOCK_NOSUID) && |
| 2524 | !(mnt_flags & MNT_NOSUID)) |
| 2525 | return false; |
| 2526 | |
| 2527 | if ((fl & MNT_LOCK_NOEXEC) && |
| 2528 | !(mnt_flags & MNT_NOEXEC)) |
| 2529 | return false; |
| 2530 | |
| 2531 | if ((fl & MNT_LOCK_ATIME) && |
| 2532 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) |
| 2533 | return false; |
| 2534 | |
| 2535 | return true; |
| 2536 | } |
| 2537 | |
| 2538 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) |
| 2539 | { |
| 2540 | bool readonly_request = (mnt_flags & MNT_READONLY); |
| 2541 | |
| 2542 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
| 2543 | return 0; |
| 2544 | |
| 2545 | if (readonly_request) |
| 2546 | return mnt_make_readonly(mnt); |
| 2547 | |
| 2548 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
| 2549 | return 0; |
| 2550 | } |
| 2551 | |
| 2552 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
| 2553 | { |
| 2554 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
| 2555 | mnt->mnt.mnt_flags = mnt_flags; |
| 2556 | touch_mnt_namespace(mnt->mnt_ns); |
| 2557 | } |
| 2558 | |
| 2559 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
| 2560 | { |
| 2561 | struct super_block *sb = mnt->mnt_sb; |
| 2562 | |
| 2563 | if (!__mnt_is_readonly(mnt) && |
| 2564 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { |
| 2565 | char *buf = (char *)__get_free_page(GFP_KERNEL); |
| 2566 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); |
| 2567 | struct tm tm; |
| 2568 | |
| 2569 | time64_to_tm(sb->s_time_max, 0, &tm); |
| 2570 | |
| 2571 | pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", |
| 2572 | sb->s_type->name, |
| 2573 | is_mounted(mnt) ? "remounted" : "mounted", |
| 2574 | mntpath, |
| 2575 | tm.tm_year+1900, (unsigned long long)sb->s_time_max); |
| 2576 | |
| 2577 | free_page((unsigned long)buf); |
| 2578 | } |
| 2579 | } |
| 2580 | |
| 2581 | /* |
| 2582 | * Handle reconfiguration of the mountpoint only without alteration of the |
| 2583 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND |
| 2584 | * to mount(2). |
| 2585 | */ |
| 2586 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) |
| 2587 | { |
| 2588 | struct super_block *sb = path->mnt->mnt_sb; |
| 2589 | struct mount *mnt = real_mount(path->mnt); |
| 2590 | int ret; |
| 2591 | |
| 2592 | if (!check_mnt(mnt)) |
| 2593 | return -EINVAL; |
| 2594 | |
| 2595 | if (path->dentry != mnt->mnt.mnt_root) |
| 2596 | return -EINVAL; |
| 2597 | |
| 2598 | if (!can_change_locked_flags(mnt, mnt_flags)) |
| 2599 | return -EPERM; |
| 2600 | |
| 2601 | /* |
| 2602 | * We're only checking whether the superblock is read-only not |
| 2603 | * changing it, so only take down_read(&sb->s_umount). |
| 2604 | */ |
| 2605 | down_read(&sb->s_umount); |
| 2606 | lock_mount_hash(); |
| 2607 | ret = change_mount_ro_state(mnt, mnt_flags); |
| 2608 | if (ret == 0) |
| 2609 | set_mount_attributes(mnt, mnt_flags); |
| 2610 | unlock_mount_hash(); |
| 2611 | up_read(&sb->s_umount); |
| 2612 | |
| 2613 | mnt_warn_timestamp_expiry(path, &mnt->mnt); |
| 2614 | |
| 2615 | return ret; |
| 2616 | } |
| 2617 | |
| 2618 | /* |
| 2619 | * change filesystem flags. dir should be a physical root of filesystem. |
| 2620 | * If you've mounted a non-root directory somewhere and want to do remount |
| 2621 | * on it - tough luck. |
| 2622 | */ |
| 2623 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
| 2624 | int mnt_flags, void *data) |
| 2625 | { |
| 2626 | int err; |
| 2627 | struct super_block *sb = path->mnt->mnt_sb; |
| 2628 | struct mount *mnt = real_mount(path->mnt); |
| 2629 | struct fs_context *fc; |
| 2630 | |
| 2631 | if (!check_mnt(mnt)) |
| 2632 | return -EINVAL; |
| 2633 | |
| 2634 | if (path->dentry != path->mnt->mnt_root) |
| 2635 | return -EINVAL; |
| 2636 | |
| 2637 | if (!can_change_locked_flags(mnt, mnt_flags)) |
| 2638 | return -EPERM; |
| 2639 | |
| 2640 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
| 2641 | if (IS_ERR(fc)) |
| 2642 | return PTR_ERR(fc); |
| 2643 | |
| 2644 | fc->oldapi = true; |
| 2645 | err = parse_monolithic_mount_data(fc, data); |
| 2646 | if (!err) { |
| 2647 | down_write(&sb->s_umount); |
| 2648 | err = -EPERM; |
| 2649 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { |
| 2650 | err = reconfigure_super(fc); |
| 2651 | if (!err) { |
| 2652 | lock_mount_hash(); |
| 2653 | set_mount_attributes(mnt, mnt_flags); |
| 2654 | unlock_mount_hash(); |
| 2655 | } |
| 2656 | } |
| 2657 | up_write(&sb->s_umount); |
| 2658 | } |
| 2659 | |
| 2660 | mnt_warn_timestamp_expiry(path, &mnt->mnt); |
| 2661 | |
| 2662 | put_fs_context(fc); |
| 2663 | return err; |
| 2664 | } |
| 2665 | |
| 2666 | static inline int tree_contains_unbindable(struct mount *mnt) |
| 2667 | { |
| 2668 | struct mount *p; |
| 2669 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
| 2670 | if (IS_MNT_UNBINDABLE(p)) |
| 2671 | return 1; |
| 2672 | } |
| 2673 | return 0; |
| 2674 | } |
| 2675 | |
| 2676 | /* |
| 2677 | * Check that there aren't references to earlier/same mount namespaces in the |
| 2678 | * specified subtree. Such references can act as pins for mount namespaces |
| 2679 | * that aren't checked by the mount-cycle checking code, thereby allowing |
| 2680 | * cycles to be made. |
| 2681 | */ |
| 2682 | static bool check_for_nsfs_mounts(struct mount *subtree) |
| 2683 | { |
| 2684 | struct mount *p; |
| 2685 | bool ret = false; |
| 2686 | |
| 2687 | lock_mount_hash(); |
| 2688 | for (p = subtree; p; p = next_mnt(p, subtree)) |
| 2689 | if (mnt_ns_loop(p->mnt.mnt_root)) |
| 2690 | goto out; |
| 2691 | |
| 2692 | ret = true; |
| 2693 | out: |
| 2694 | unlock_mount_hash(); |
| 2695 | return ret; |
| 2696 | } |
| 2697 | |
| 2698 | static int do_set_group(struct path *from_path, struct path *to_path) |
| 2699 | { |
| 2700 | struct mount *from, *to; |
| 2701 | int err; |
| 2702 | |
| 2703 | from = real_mount(from_path->mnt); |
| 2704 | to = real_mount(to_path->mnt); |
| 2705 | |
| 2706 | namespace_lock(); |
| 2707 | |
| 2708 | err = -EINVAL; |
| 2709 | /* To and From must be mounted */ |
| 2710 | if (!is_mounted(&from->mnt)) |
| 2711 | goto out; |
| 2712 | if (!is_mounted(&to->mnt)) |
| 2713 | goto out; |
| 2714 | |
| 2715 | err = -EPERM; |
| 2716 | /* We should be allowed to modify mount namespaces of both mounts */ |
| 2717 | if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) |
| 2718 | goto out; |
| 2719 | if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) |
| 2720 | goto out; |
| 2721 | |
| 2722 | err = -EINVAL; |
| 2723 | /* To and From paths should be mount roots */ |
| 2724 | if (from_path->dentry != from_path->mnt->mnt_root) |
| 2725 | goto out; |
| 2726 | if (to_path->dentry != to_path->mnt->mnt_root) |
| 2727 | goto out; |
| 2728 | |
| 2729 | /* Setting sharing groups is only allowed across same superblock */ |
| 2730 | if (from->mnt.mnt_sb != to->mnt.mnt_sb) |
| 2731 | goto out; |
| 2732 | |
| 2733 | /* From mount root should be wider than To mount root */ |
| 2734 | if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) |
| 2735 | goto out; |
| 2736 | |
| 2737 | /* From mount should not have locked children in place of To's root */ |
| 2738 | if (has_locked_children(from, to->mnt.mnt_root)) |
| 2739 | goto out; |
| 2740 | |
| 2741 | /* Setting sharing groups is only allowed on private mounts */ |
| 2742 | if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) |
| 2743 | goto out; |
| 2744 | |
| 2745 | /* From should not be private */ |
| 2746 | if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) |
| 2747 | goto out; |
| 2748 | |
| 2749 | if (IS_MNT_SLAVE(from)) { |
| 2750 | struct mount *m = from->mnt_master; |
| 2751 | |
| 2752 | list_add(&to->mnt_slave, &m->mnt_slave_list); |
| 2753 | to->mnt_master = m; |
| 2754 | } |
| 2755 | |
| 2756 | if (IS_MNT_SHARED(from)) { |
| 2757 | to->mnt_group_id = from->mnt_group_id; |
| 2758 | list_add(&to->mnt_share, &from->mnt_share); |
| 2759 | lock_mount_hash(); |
| 2760 | set_mnt_shared(to); |
| 2761 | unlock_mount_hash(); |
| 2762 | } |
| 2763 | |
| 2764 | err = 0; |
| 2765 | out: |
| 2766 | namespace_unlock(); |
| 2767 | return err; |
| 2768 | } |
| 2769 | |
| 2770 | static int do_move_mount(struct path *old_path, struct path *new_path) |
| 2771 | { |
| 2772 | struct mnt_namespace *ns; |
| 2773 | struct mount *p; |
| 2774 | struct mount *old; |
| 2775 | struct mount *parent; |
| 2776 | struct mountpoint *mp, *old_mp; |
| 2777 | int err; |
| 2778 | bool attached; |
| 2779 | |
| 2780 | mp = lock_mount(new_path); |
| 2781 | if (IS_ERR(mp)) |
| 2782 | return PTR_ERR(mp); |
| 2783 | |
| 2784 | old = real_mount(old_path->mnt); |
| 2785 | p = real_mount(new_path->mnt); |
| 2786 | parent = old->mnt_parent; |
| 2787 | attached = mnt_has_parent(old); |
| 2788 | old_mp = old->mnt_mp; |
| 2789 | ns = old->mnt_ns; |
| 2790 | |
| 2791 | err = -EINVAL; |
| 2792 | /* The mountpoint must be in our namespace. */ |
| 2793 | if (!check_mnt(p)) |
| 2794 | goto out; |
| 2795 | |
| 2796 | /* The thing moved must be mounted... */ |
| 2797 | if (!is_mounted(&old->mnt)) |
| 2798 | goto out; |
| 2799 | |
| 2800 | /* ... and either ours or the root of anon namespace */ |
| 2801 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) |
| 2802 | goto out; |
| 2803 | |
| 2804 | if (old->mnt.mnt_flags & MNT_LOCKED) |
| 2805 | goto out; |
| 2806 | |
| 2807 | if (old_path->dentry != old_path->mnt->mnt_root) |
| 2808 | goto out; |
| 2809 | |
| 2810 | if (d_is_dir(new_path->dentry) != |
| 2811 | d_is_dir(old_path->dentry)) |
| 2812 | goto out; |
| 2813 | /* |
| 2814 | * Don't move a mount residing in a shared parent. |
| 2815 | */ |
| 2816 | if (attached && IS_MNT_SHARED(parent)) |
| 2817 | goto out; |
| 2818 | /* |
| 2819 | * Don't move a mount tree containing unbindable mounts to a destination |
| 2820 | * mount which is shared. |
| 2821 | */ |
| 2822 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
| 2823 | goto out; |
| 2824 | err = -ELOOP; |
| 2825 | if (!check_for_nsfs_mounts(old)) |
| 2826 | goto out; |
| 2827 | for (; mnt_has_parent(p); p = p->mnt_parent) |
| 2828 | if (p == old) |
| 2829 | goto out; |
| 2830 | |
| 2831 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
| 2832 | attached); |
| 2833 | if (err) |
| 2834 | goto out; |
| 2835 | |
| 2836 | /* if the mount is moved, it should no longer be expire |
| 2837 | * automatically */ |
| 2838 | list_del_init(&old->mnt_expire); |
| 2839 | if (attached) |
| 2840 | put_mountpoint(old_mp); |
| 2841 | out: |
| 2842 | unlock_mount(mp); |
| 2843 | if (!err) { |
| 2844 | if (attached) |
| 2845 | mntput_no_expire(parent); |
| 2846 | else |
| 2847 | free_mnt_ns(ns); |
| 2848 | } |
| 2849 | return err; |
| 2850 | } |
| 2851 | |
| 2852 | static int do_move_mount_old(struct path *path, const char *old_name) |
| 2853 | { |
| 2854 | struct path old_path; |
| 2855 | int err; |
| 2856 | |
| 2857 | if (!old_name || !*old_name) |
| 2858 | return -EINVAL; |
| 2859 | |
| 2860 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
| 2861 | if (err) |
| 2862 | return err; |
| 2863 | |
| 2864 | err = do_move_mount(&old_path, path); |
| 2865 | path_put(&old_path); |
| 2866 | return err; |
| 2867 | } |
| 2868 | |
| 2869 | /* |
| 2870 | * add a mount into a namespace's mount tree |
| 2871 | */ |
| 2872 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
| 2873 | struct path *path, int mnt_flags) |
| 2874 | { |
| 2875 | struct mount *parent = real_mount(path->mnt); |
| 2876 | |
| 2877 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
| 2878 | |
| 2879 | if (unlikely(!check_mnt(parent))) { |
| 2880 | /* that's acceptable only for automounts done in private ns */ |
| 2881 | if (!(mnt_flags & MNT_SHRINKABLE)) |
| 2882 | return -EINVAL; |
| 2883 | /* ... and for those we'd better have mountpoint still alive */ |
| 2884 | if (!parent->mnt_ns) |
| 2885 | return -EINVAL; |
| 2886 | } |
| 2887 | |
| 2888 | /* Refuse the same filesystem on the same mount point */ |
| 2889 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
| 2890 | path->mnt->mnt_root == path->dentry) |
| 2891 | return -EBUSY; |
| 2892 | |
| 2893 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
| 2894 | return -EINVAL; |
| 2895 | |
| 2896 | newmnt->mnt.mnt_flags = mnt_flags; |
| 2897 | return graft_tree(newmnt, parent, mp); |
| 2898 | } |
| 2899 | |
| 2900 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
| 2901 | |
| 2902 | /* |
| 2903 | * Create a new mount using a superblock configuration and request it |
| 2904 | * be added to the namespace tree. |
| 2905 | */ |
| 2906 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, |
| 2907 | unsigned int mnt_flags) |
| 2908 | { |
| 2909 | struct vfsmount *mnt; |
| 2910 | struct mountpoint *mp; |
| 2911 | struct super_block *sb = fc->root->d_sb; |
| 2912 | int error; |
| 2913 | |
| 2914 | error = security_sb_kern_mount(sb); |
| 2915 | if (!error && mount_too_revealing(sb, &mnt_flags)) |
| 2916 | error = -EPERM; |
| 2917 | |
| 2918 | if (unlikely(error)) { |
| 2919 | fc_drop_locked(fc); |
| 2920 | return error; |
| 2921 | } |
| 2922 | |
| 2923 | up_write(&sb->s_umount); |
| 2924 | |
| 2925 | mnt = vfs_create_mount(fc); |
| 2926 | if (IS_ERR(mnt)) |
| 2927 | return PTR_ERR(mnt); |
| 2928 | |
| 2929 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
| 2930 | |
| 2931 | mp = lock_mount(mountpoint); |
| 2932 | if (IS_ERR(mp)) { |
| 2933 | mntput(mnt); |
| 2934 | return PTR_ERR(mp); |
| 2935 | } |
| 2936 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); |
| 2937 | unlock_mount(mp); |
| 2938 | if (error < 0) |
| 2939 | mntput(mnt); |
| 2940 | return error; |
| 2941 | } |
| 2942 | |
| 2943 | /* |
| 2944 | * create a new mount for userspace and request it to be added into the |
| 2945 | * namespace's tree |
| 2946 | */ |
| 2947 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
| 2948 | int mnt_flags, const char *name, void *data) |
| 2949 | { |
| 2950 | struct file_system_type *type; |
| 2951 | struct fs_context *fc; |
| 2952 | const char *subtype = NULL; |
| 2953 | int err = 0; |
| 2954 | |
| 2955 | if (!fstype) |
| 2956 | return -EINVAL; |
| 2957 | |
| 2958 | type = get_fs_type(fstype); |
| 2959 | if (!type) |
| 2960 | return -ENODEV; |
| 2961 | |
| 2962 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
| 2963 | subtype = strchr(fstype, '.'); |
| 2964 | if (subtype) { |
| 2965 | subtype++; |
| 2966 | if (!*subtype) { |
| 2967 | put_filesystem(type); |
| 2968 | return -EINVAL; |
| 2969 | } |
| 2970 | } |
| 2971 | } |
| 2972 | |
| 2973 | fc = fs_context_for_mount(type, sb_flags); |
| 2974 | put_filesystem(type); |
| 2975 | if (IS_ERR(fc)) |
| 2976 | return PTR_ERR(fc); |
| 2977 | |
| 2978 | if (subtype) |
| 2979 | err = vfs_parse_fs_string(fc, "subtype", |
| 2980 | subtype, strlen(subtype)); |
| 2981 | if (!err && name) |
| 2982 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); |
| 2983 | if (!err) |
| 2984 | err = parse_monolithic_mount_data(fc, data); |
| 2985 | if (!err && !mount_capable(fc)) |
| 2986 | err = -EPERM; |
| 2987 | if (!err) |
| 2988 | err = vfs_get_tree(fc); |
| 2989 | if (!err) |
| 2990 | err = do_new_mount_fc(fc, path, mnt_flags); |
| 2991 | |
| 2992 | put_fs_context(fc); |
| 2993 | return err; |
| 2994 | } |
| 2995 | |
| 2996 | int finish_automount(struct vfsmount *m, struct path *path) |
| 2997 | { |
| 2998 | struct dentry *dentry = path->dentry; |
| 2999 | struct mountpoint *mp; |
| 3000 | struct mount *mnt; |
| 3001 | int err; |
| 3002 | |
| 3003 | if (!m) |
| 3004 | return 0; |
| 3005 | if (IS_ERR(m)) |
| 3006 | return PTR_ERR(m); |
| 3007 | |
| 3008 | mnt = real_mount(m); |
| 3009 | /* The new mount record should have at least 2 refs to prevent it being |
| 3010 | * expired before we get a chance to add it |
| 3011 | */ |
| 3012 | BUG_ON(mnt_get_count(mnt) < 2); |
| 3013 | |
| 3014 | if (m->mnt_sb == path->mnt->mnt_sb && |
| 3015 | m->mnt_root == dentry) { |
| 3016 | err = -ELOOP; |
| 3017 | goto discard; |
| 3018 | } |
| 3019 | |
| 3020 | /* |
| 3021 | * we don't want to use lock_mount() - in this case finding something |
| 3022 | * that overmounts our mountpoint to be means "quitely drop what we've |
| 3023 | * got", not "try to mount it on top". |
| 3024 | */ |
| 3025 | inode_lock(dentry->d_inode); |
| 3026 | namespace_lock(); |
| 3027 | if (unlikely(cant_mount(dentry))) { |
| 3028 | err = -ENOENT; |
| 3029 | goto discard_locked; |
| 3030 | } |
| 3031 | rcu_read_lock(); |
| 3032 | if (unlikely(__lookup_mnt(path->mnt, dentry))) { |
| 3033 | rcu_read_unlock(); |
| 3034 | err = 0; |
| 3035 | goto discard_locked; |
| 3036 | } |
| 3037 | rcu_read_unlock(); |
| 3038 | mp = get_mountpoint(dentry); |
| 3039 | if (IS_ERR(mp)) { |
| 3040 | err = PTR_ERR(mp); |
| 3041 | goto discard_locked; |
| 3042 | } |
| 3043 | |
| 3044 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
| 3045 | unlock_mount(mp); |
| 3046 | if (unlikely(err)) |
| 3047 | goto discard; |
| 3048 | mntput(m); |
| 3049 | return 0; |
| 3050 | |
| 3051 | discard_locked: |
| 3052 | namespace_unlock(); |
| 3053 | inode_unlock(dentry->d_inode); |
| 3054 | discard: |
| 3055 | /* remove m from any expiration list it may be on */ |
| 3056 | if (!list_empty(&mnt->mnt_expire)) { |
| 3057 | namespace_lock(); |
| 3058 | list_del_init(&mnt->mnt_expire); |
| 3059 | namespace_unlock(); |
| 3060 | } |
| 3061 | mntput(m); |
| 3062 | mntput(m); |
| 3063 | return err; |
| 3064 | } |
| 3065 | |
| 3066 | /** |
| 3067 | * mnt_set_expiry - Put a mount on an expiration list |
| 3068 | * @mnt: The mount to list. |
| 3069 | * @expiry_list: The list to add the mount to. |
| 3070 | */ |
| 3071 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) |
| 3072 | { |
| 3073 | namespace_lock(); |
| 3074 | |
| 3075 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
| 3076 | |
| 3077 | namespace_unlock(); |
| 3078 | } |
| 3079 | EXPORT_SYMBOL(mnt_set_expiry); |
| 3080 | |
| 3081 | /* |
| 3082 | * process a list of expirable mountpoints with the intent of discarding any |
| 3083 | * mountpoints that aren't in use and haven't been touched since last we came |
| 3084 | * here |
| 3085 | */ |
| 3086 | void mark_mounts_for_expiry(struct list_head *mounts) |
| 3087 | { |
| 3088 | struct mount *mnt, *next; |
| 3089 | LIST_HEAD(graveyard); |
| 3090 | |
| 3091 | if (list_empty(mounts)) |
| 3092 | return; |
| 3093 | |
| 3094 | namespace_lock(); |
| 3095 | lock_mount_hash(); |
| 3096 | |
| 3097 | /* extract from the expiration list every vfsmount that matches the |
| 3098 | * following criteria: |
| 3099 | * - only referenced by its parent vfsmount |
| 3100 | * - still marked for expiry (marked on the last call here; marks are |
| 3101 | * cleared by mntput()) |
| 3102 | */ |
| 3103 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
| 3104 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
| 3105 | propagate_mount_busy(mnt, 1)) |
| 3106 | continue; |
| 3107 | list_move(&mnt->mnt_expire, &graveyard); |
| 3108 | } |
| 3109 | while (!list_empty(&graveyard)) { |
| 3110 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
| 3111 | touch_mnt_namespace(mnt->mnt_ns); |
| 3112 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
| 3113 | } |
| 3114 | unlock_mount_hash(); |
| 3115 | namespace_unlock(); |
| 3116 | } |
| 3117 | |
| 3118 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); |
| 3119 | |
| 3120 | /* |
| 3121 | * Ripoff of 'select_parent()' |
| 3122 | * |
| 3123 | * search the list of submounts for a given mountpoint, and move any |
| 3124 | * shrinkable submounts to the 'graveyard' list. |
| 3125 | */ |
| 3126 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
| 3127 | { |
| 3128 | struct mount *this_parent = parent; |
| 3129 | struct list_head *next; |
| 3130 | int found = 0; |
| 3131 | |
| 3132 | repeat: |
| 3133 | next = this_parent->mnt_mounts.next; |
| 3134 | resume: |
| 3135 | while (next != &this_parent->mnt_mounts) { |
| 3136 | struct list_head *tmp = next; |
| 3137 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
| 3138 | |
| 3139 | next = tmp->next; |
| 3140 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
| 3141 | continue; |
| 3142 | /* |
| 3143 | * Descend a level if the d_mounts list is non-empty. |
| 3144 | */ |
| 3145 | if (!list_empty(&mnt->mnt_mounts)) { |
| 3146 | this_parent = mnt; |
| 3147 | goto repeat; |
| 3148 | } |
| 3149 | |
| 3150 | if (!propagate_mount_busy(mnt, 1)) { |
| 3151 | list_move_tail(&mnt->mnt_expire, graveyard); |
| 3152 | found++; |
| 3153 | } |
| 3154 | } |
| 3155 | /* |
| 3156 | * All done at this level ... ascend and resume the search |
| 3157 | */ |
| 3158 | if (this_parent != parent) { |
| 3159 | next = this_parent->mnt_child.next; |
| 3160 | this_parent = this_parent->mnt_parent; |
| 3161 | goto resume; |
| 3162 | } |
| 3163 | return found; |
| 3164 | } |
| 3165 | |
| 3166 | /* |
| 3167 | * process a list of expirable mountpoints with the intent of discarding any |
| 3168 | * submounts of a specific parent mountpoint |
| 3169 | * |
| 3170 | * mount_lock must be held for write |
| 3171 | */ |
| 3172 | static void shrink_submounts(struct mount *mnt) |
| 3173 | { |
| 3174 | LIST_HEAD(graveyard); |
| 3175 | struct mount *m; |
| 3176 | |
| 3177 | /* extract submounts of 'mountpoint' from the expiration list */ |
| 3178 | while (select_submounts(mnt, &graveyard)) { |
| 3179 | while (!list_empty(&graveyard)) { |
| 3180 | m = list_first_entry(&graveyard, struct mount, |
| 3181 | mnt_expire); |
| 3182 | touch_mnt_namespace(m->mnt_ns); |
| 3183 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
| 3184 | } |
| 3185 | } |
| 3186 | } |
| 3187 | |
| 3188 | static void *copy_mount_options(const void __user * data) |
| 3189 | { |
| 3190 | char *copy; |
| 3191 | unsigned left, offset; |
| 3192 | |
| 3193 | if (!data) |
| 3194 | return NULL; |
| 3195 | |
| 3196 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 3197 | if (!copy) |
| 3198 | return ERR_PTR(-ENOMEM); |
| 3199 | |
| 3200 | left = copy_from_user(copy, data, PAGE_SIZE); |
| 3201 | |
| 3202 | /* |
| 3203 | * Not all architectures have an exact copy_from_user(). Resort to |
| 3204 | * byte at a time. |
| 3205 | */ |
| 3206 | offset = PAGE_SIZE - left; |
| 3207 | while (left) { |
| 3208 | char c; |
| 3209 | if (get_user(c, (const char __user *)data + offset)) |
| 3210 | break; |
| 3211 | copy[offset] = c; |
| 3212 | left--; |
| 3213 | offset++; |
| 3214 | } |
| 3215 | |
| 3216 | if (left == PAGE_SIZE) { |
| 3217 | kfree(copy); |
| 3218 | return ERR_PTR(-EFAULT); |
| 3219 | } |
| 3220 | |
| 3221 | return copy; |
| 3222 | } |
| 3223 | |
| 3224 | static char *copy_mount_string(const void __user *data) |
| 3225 | { |
| 3226 | return data ? strndup_user(data, PATH_MAX) : NULL; |
| 3227 | } |
| 3228 | |
| 3229 | /* |
| 3230 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to |
| 3231 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). |
| 3232 | * |
| 3233 | * data is a (void *) that can point to any structure up to |
| 3234 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent |
| 3235 | * information (or be NULL). |
| 3236 | * |
| 3237 | * Pre-0.97 versions of mount() didn't have a flags word. |
| 3238 | * When the flags word was introduced its top half was required |
| 3239 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. |
| 3240 | * Therefore, if this magic number is present, it carries no information |
| 3241 | * and must be discarded. |
| 3242 | */ |
| 3243 | int path_mount(const char *dev_name, struct path *path, |
| 3244 | const char *type_page, unsigned long flags, void *data_page) |
| 3245 | { |
| 3246 | unsigned int mnt_flags = 0, sb_flags; |
| 3247 | int ret; |
| 3248 | |
| 3249 | /* Discard magic */ |
| 3250 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) |
| 3251 | flags &= ~MS_MGC_MSK; |
| 3252 | |
| 3253 | /* Basic sanity checks */ |
| 3254 | if (data_page) |
| 3255 | ((char *)data_page)[PAGE_SIZE - 1] = 0; |
| 3256 | |
| 3257 | if (flags & MS_NOUSER) |
| 3258 | return -EINVAL; |
| 3259 | |
| 3260 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
| 3261 | if (ret) |
| 3262 | return ret; |
| 3263 | if (!may_mount()) |
| 3264 | return -EPERM; |
| 3265 | if (flags & SB_MANDLOCK) |
| 3266 | warn_mandlock(); |
| 3267 | |
| 3268 | /* Default to relatime unless overriden */ |
| 3269 | if (!(flags & MS_NOATIME)) |
| 3270 | mnt_flags |= MNT_RELATIME; |
| 3271 | |
| 3272 | /* Separate the per-mountpoint flags */ |
| 3273 | if (flags & MS_NOSUID) |
| 3274 | mnt_flags |= MNT_NOSUID; |
| 3275 | if (flags & MS_NODEV) |
| 3276 | mnt_flags |= MNT_NODEV; |
| 3277 | if (flags & MS_NOEXEC) |
| 3278 | mnt_flags |= MNT_NOEXEC; |
| 3279 | if (flags & MS_NOATIME) |
| 3280 | mnt_flags |= MNT_NOATIME; |
| 3281 | if (flags & MS_NODIRATIME) |
| 3282 | mnt_flags |= MNT_NODIRATIME; |
| 3283 | if (flags & MS_STRICTATIME) |
| 3284 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); |
| 3285 | if (flags & MS_RDONLY) |
| 3286 | mnt_flags |= MNT_READONLY; |
| 3287 | if (flags & MS_NOSYMFOLLOW) |
| 3288 | mnt_flags |= MNT_NOSYMFOLLOW; |
| 3289 | |
| 3290 | /* The default atime for remount is preservation */ |
| 3291 | if ((flags & MS_REMOUNT) && |
| 3292 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | |
| 3293 | MS_STRICTATIME)) == 0)) { |
| 3294 | mnt_flags &= ~MNT_ATIME_MASK; |
| 3295 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
| 3296 | } |
| 3297 | |
| 3298 | sb_flags = flags & (SB_RDONLY | |
| 3299 | SB_SYNCHRONOUS | |
| 3300 | SB_MANDLOCK | |
| 3301 | SB_DIRSYNC | |
| 3302 | SB_SILENT | |
| 3303 | SB_POSIXACL | |
| 3304 | SB_LAZYTIME | |
| 3305 | SB_I_VERSION); |
| 3306 | |
| 3307 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
| 3308 | return do_reconfigure_mnt(path, mnt_flags); |
| 3309 | if (flags & MS_REMOUNT) |
| 3310 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); |
| 3311 | if (flags & MS_BIND) |
| 3312 | return do_loopback(path, dev_name, flags & MS_REC); |
| 3313 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
| 3314 | return do_change_type(path, flags); |
| 3315 | if (flags & MS_MOVE) |
| 3316 | return do_move_mount_old(path, dev_name); |
| 3317 | |
| 3318 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, |
| 3319 | data_page); |
| 3320 | } |
| 3321 | |
| 3322 | long do_mount(const char *dev_name, const char __user *dir_name, |
| 3323 | const char *type_page, unsigned long flags, void *data_page) |
| 3324 | { |
| 3325 | struct path path; |
| 3326 | int ret; |
| 3327 | |
| 3328 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); |
| 3329 | if (ret) |
| 3330 | return ret; |
| 3331 | ret = path_mount(dev_name, &path, type_page, flags, data_page); |
| 3332 | path_put(&path); |
| 3333 | return ret; |
| 3334 | } |
| 3335 | |
| 3336 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
| 3337 | { |
| 3338 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); |
| 3339 | } |
| 3340 | |
| 3341 | static void dec_mnt_namespaces(struct ucounts *ucounts) |
| 3342 | { |
| 3343 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); |
| 3344 | } |
| 3345 | |
| 3346 | static void free_mnt_ns(struct mnt_namespace *ns) |
| 3347 | { |
| 3348 | if (!is_anon_ns(ns)) |
| 3349 | ns_free_inum(&ns->ns); |
| 3350 | dec_mnt_namespaces(ns->ucounts); |
| 3351 | put_user_ns(ns->user_ns); |
| 3352 | kfree(ns); |
| 3353 | } |
| 3354 | |
| 3355 | /* |
| 3356 | * Assign a sequence number so we can detect when we attempt to bind |
| 3357 | * mount a reference to an older mount namespace into the current |
| 3358 | * mount namespace, preventing reference counting loops. A 64bit |
| 3359 | * number incrementing at 10Ghz will take 12,427 years to wrap which |
| 3360 | * is effectively never, so we can ignore the possibility. |
| 3361 | */ |
| 3362 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); |
| 3363 | |
| 3364 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
| 3365 | { |
| 3366 | struct mnt_namespace *new_ns; |
| 3367 | struct ucounts *ucounts; |
| 3368 | int ret; |
| 3369 | |
| 3370 | ucounts = inc_mnt_namespaces(user_ns); |
| 3371 | if (!ucounts) |
| 3372 | return ERR_PTR(-ENOSPC); |
| 3373 | |
| 3374 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); |
| 3375 | if (!new_ns) { |
| 3376 | dec_mnt_namespaces(ucounts); |
| 3377 | return ERR_PTR(-ENOMEM); |
| 3378 | } |
| 3379 | if (!anon) { |
| 3380 | ret = ns_alloc_inum(&new_ns->ns); |
| 3381 | if (ret) { |
| 3382 | kfree(new_ns); |
| 3383 | dec_mnt_namespaces(ucounts); |
| 3384 | return ERR_PTR(ret); |
| 3385 | } |
| 3386 | } |
| 3387 | new_ns->ns.ops = &mntns_operations; |
| 3388 | if (!anon) |
| 3389 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
| 3390 | refcount_set(&new_ns->ns.count, 1); |
| 3391 | INIT_LIST_HEAD(&new_ns->list); |
| 3392 | init_waitqueue_head(&new_ns->poll); |
| 3393 | spin_lock_init(&new_ns->ns_lock); |
| 3394 | new_ns->user_ns = get_user_ns(user_ns); |
| 3395 | new_ns->ucounts = ucounts; |
| 3396 | return new_ns; |
| 3397 | } |
| 3398 | |
| 3399 | __latent_entropy |
| 3400 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
| 3401 | struct user_namespace *user_ns, struct fs_struct *new_fs) |
| 3402 | { |
| 3403 | struct mnt_namespace *new_ns; |
| 3404 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
| 3405 | struct mount *p, *q; |
| 3406 | struct mount *old; |
| 3407 | struct mount *new; |
| 3408 | int copy_flags; |
| 3409 | |
| 3410 | BUG_ON(!ns); |
| 3411 | |
| 3412 | if (likely(!(flags & CLONE_NEWNS))) { |
| 3413 | get_mnt_ns(ns); |
| 3414 | return ns; |
| 3415 | } |
| 3416 | |
| 3417 | old = ns->root; |
| 3418 | |
| 3419 | new_ns = alloc_mnt_ns(user_ns, false); |
| 3420 | if (IS_ERR(new_ns)) |
| 3421 | return new_ns; |
| 3422 | |
| 3423 | namespace_lock(); |
| 3424 | /* First pass: copy the tree topology */ |
| 3425 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
| 3426 | if (user_ns != ns->user_ns) |
| 3427 | copy_flags |= CL_SHARED_TO_SLAVE; |
| 3428 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
| 3429 | if (IS_ERR(new)) { |
| 3430 | namespace_unlock(); |
| 3431 | free_mnt_ns(new_ns); |
| 3432 | return ERR_CAST(new); |
| 3433 | } |
| 3434 | if (user_ns != ns->user_ns) { |
| 3435 | lock_mount_hash(); |
| 3436 | lock_mnt_tree(new); |
| 3437 | unlock_mount_hash(); |
| 3438 | } |
| 3439 | new_ns->root = new; |
| 3440 | list_add_tail(&new_ns->list, &new->mnt_list); |
| 3441 | |
| 3442 | /* |
| 3443 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts |
| 3444 | * as belonging to new namespace. We have already acquired a private |
| 3445 | * fs_struct, so tsk->fs->lock is not needed. |
| 3446 | */ |
| 3447 | p = old; |
| 3448 | q = new; |
| 3449 | while (p) { |
| 3450 | q->mnt_ns = new_ns; |
| 3451 | new_ns->mounts++; |
| 3452 | if (new_fs) { |
| 3453 | if (&p->mnt == new_fs->root.mnt) { |
| 3454 | new_fs->root.mnt = mntget(&q->mnt); |
| 3455 | rootmnt = &p->mnt; |
| 3456 | } |
| 3457 | if (&p->mnt == new_fs->pwd.mnt) { |
| 3458 | new_fs->pwd.mnt = mntget(&q->mnt); |
| 3459 | pwdmnt = &p->mnt; |
| 3460 | } |
| 3461 | } |
| 3462 | p = next_mnt(p, old); |
| 3463 | q = next_mnt(q, new); |
| 3464 | if (!q) |
| 3465 | break; |
| 3466 | while (p->mnt.mnt_root != q->mnt.mnt_root) |
| 3467 | p = next_mnt(p, old); |
| 3468 | } |
| 3469 | namespace_unlock(); |
| 3470 | |
| 3471 | if (rootmnt) |
| 3472 | mntput(rootmnt); |
| 3473 | if (pwdmnt) |
| 3474 | mntput(pwdmnt); |
| 3475 | |
| 3476 | return new_ns; |
| 3477 | } |
| 3478 | |
| 3479 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
| 3480 | { |
| 3481 | struct mount *mnt = real_mount(m); |
| 3482 | struct mnt_namespace *ns; |
| 3483 | struct super_block *s; |
| 3484 | struct path path; |
| 3485 | int err; |
| 3486 | |
| 3487 | ns = alloc_mnt_ns(&init_user_ns, true); |
| 3488 | if (IS_ERR(ns)) { |
| 3489 | mntput(m); |
| 3490 | return ERR_CAST(ns); |
| 3491 | } |
| 3492 | mnt->mnt_ns = ns; |
| 3493 | ns->root = mnt; |
| 3494 | ns->mounts++; |
| 3495 | list_add(&mnt->mnt_list, &ns->list); |
| 3496 | |
| 3497 | err = vfs_path_lookup(m->mnt_root, m, |
| 3498 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
| 3499 | |
| 3500 | put_mnt_ns(ns); |
| 3501 | |
| 3502 | if (err) |
| 3503 | return ERR_PTR(err); |
| 3504 | |
| 3505 | /* trade a vfsmount reference for active sb one */ |
| 3506 | s = path.mnt->mnt_sb; |
| 3507 | atomic_inc(&s->s_active); |
| 3508 | mntput(path.mnt); |
| 3509 | /* lock the sucker */ |
| 3510 | down_write(&s->s_umount); |
| 3511 | /* ... and return the root of (sub)tree on it */ |
| 3512 | return path.dentry; |
| 3513 | } |
| 3514 | EXPORT_SYMBOL(mount_subtree); |
| 3515 | |
| 3516 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
| 3517 | char __user *, type, unsigned long, flags, void __user *, data) |
| 3518 | { |
| 3519 | int ret; |
| 3520 | char *kernel_type; |
| 3521 | char *kernel_dev; |
| 3522 | void *options; |
| 3523 | |
| 3524 | kernel_type = copy_mount_string(type); |
| 3525 | ret = PTR_ERR(kernel_type); |
| 3526 | if (IS_ERR(kernel_type)) |
| 3527 | goto out_type; |
| 3528 | |
| 3529 | kernel_dev = copy_mount_string(dev_name); |
| 3530 | ret = PTR_ERR(kernel_dev); |
| 3531 | if (IS_ERR(kernel_dev)) |
| 3532 | goto out_dev; |
| 3533 | |
| 3534 | options = copy_mount_options(data); |
| 3535 | ret = PTR_ERR(options); |
| 3536 | if (IS_ERR(options)) |
| 3537 | goto out_data; |
| 3538 | |
| 3539 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
| 3540 | |
| 3541 | kfree(options); |
| 3542 | out_data: |
| 3543 | kfree(kernel_dev); |
| 3544 | out_dev: |
| 3545 | kfree(kernel_type); |
| 3546 | out_type: |
| 3547 | return ret; |
| 3548 | } |
| 3549 | |
| 3550 | #define FSMOUNT_VALID_FLAGS \ |
| 3551 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ |
| 3552 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ |
| 3553 | MOUNT_ATTR_NOSYMFOLLOW) |
| 3554 | |
| 3555 | #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) |
| 3556 | |
| 3557 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ |
| 3558 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) |
| 3559 | |
| 3560 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
| 3561 | { |
| 3562 | unsigned int mnt_flags = 0; |
| 3563 | |
| 3564 | if (attr_flags & MOUNT_ATTR_RDONLY) |
| 3565 | mnt_flags |= MNT_READONLY; |
| 3566 | if (attr_flags & MOUNT_ATTR_NOSUID) |
| 3567 | mnt_flags |= MNT_NOSUID; |
| 3568 | if (attr_flags & MOUNT_ATTR_NODEV) |
| 3569 | mnt_flags |= MNT_NODEV; |
| 3570 | if (attr_flags & MOUNT_ATTR_NOEXEC) |
| 3571 | mnt_flags |= MNT_NOEXEC; |
| 3572 | if (attr_flags & MOUNT_ATTR_NODIRATIME) |
| 3573 | mnt_flags |= MNT_NODIRATIME; |
| 3574 | if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) |
| 3575 | mnt_flags |= MNT_NOSYMFOLLOW; |
| 3576 | |
| 3577 | return mnt_flags; |
| 3578 | } |
| 3579 | |
| 3580 | /* |
| 3581 | * Create a kernel mount representation for a new, prepared superblock |
| 3582 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. |
| 3583 | */ |
| 3584 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, |
| 3585 | unsigned int, attr_flags) |
| 3586 | { |
| 3587 | struct mnt_namespace *ns; |
| 3588 | struct fs_context *fc; |
| 3589 | struct file *file; |
| 3590 | struct path newmount; |
| 3591 | struct mount *mnt; |
| 3592 | struct fd f; |
| 3593 | unsigned int mnt_flags = 0; |
| 3594 | long ret; |
| 3595 | |
| 3596 | if (!may_mount()) |
| 3597 | return -EPERM; |
| 3598 | |
| 3599 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) |
| 3600 | return -EINVAL; |
| 3601 | |
| 3602 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
| 3603 | return -EINVAL; |
| 3604 | |
| 3605 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
| 3606 | |
| 3607 | switch (attr_flags & MOUNT_ATTR__ATIME) { |
| 3608 | case MOUNT_ATTR_STRICTATIME: |
| 3609 | break; |
| 3610 | case MOUNT_ATTR_NOATIME: |
| 3611 | mnt_flags |= MNT_NOATIME; |
| 3612 | break; |
| 3613 | case MOUNT_ATTR_RELATIME: |
| 3614 | mnt_flags |= MNT_RELATIME; |
| 3615 | break; |
| 3616 | default: |
| 3617 | return -EINVAL; |
| 3618 | } |
| 3619 | |
| 3620 | f = fdget(fs_fd); |
| 3621 | if (!f.file) |
| 3622 | return -EBADF; |
| 3623 | |
| 3624 | ret = -EINVAL; |
| 3625 | if (f.file->f_op != &fscontext_fops) |
| 3626 | goto err_fsfd; |
| 3627 | |
| 3628 | fc = f.file->private_data; |
| 3629 | |
| 3630 | ret = mutex_lock_interruptible(&fc->uapi_mutex); |
| 3631 | if (ret < 0) |
| 3632 | goto err_fsfd; |
| 3633 | |
| 3634 | /* There must be a valid superblock or we can't mount it */ |
| 3635 | ret = -EINVAL; |
| 3636 | if (!fc->root) |
| 3637 | goto err_unlock; |
| 3638 | |
| 3639 | ret = -EPERM; |
| 3640 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { |
| 3641 | pr_warn("VFS: Mount too revealing\n"); |
| 3642 | goto err_unlock; |
| 3643 | } |
| 3644 | |
| 3645 | ret = -EBUSY; |
| 3646 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) |
| 3647 | goto err_unlock; |
| 3648 | |
| 3649 | if (fc->sb_flags & SB_MANDLOCK) |
| 3650 | warn_mandlock(); |
| 3651 | |
| 3652 | newmount.mnt = vfs_create_mount(fc); |
| 3653 | if (IS_ERR(newmount.mnt)) { |
| 3654 | ret = PTR_ERR(newmount.mnt); |
| 3655 | goto err_unlock; |
| 3656 | } |
| 3657 | newmount.dentry = dget(fc->root); |
| 3658 | newmount.mnt->mnt_flags = mnt_flags; |
| 3659 | |
| 3660 | /* We've done the mount bit - now move the file context into more or |
| 3661 | * less the same state as if we'd done an fspick(). We don't want to |
| 3662 | * do any memory allocation or anything like that at this point as we |
| 3663 | * don't want to have to handle any errors incurred. |
| 3664 | */ |
| 3665 | vfs_clean_context(fc); |
| 3666 | |
| 3667 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); |
| 3668 | if (IS_ERR(ns)) { |
| 3669 | ret = PTR_ERR(ns); |
| 3670 | goto err_path; |
| 3671 | } |
| 3672 | mnt = real_mount(newmount.mnt); |
| 3673 | mnt->mnt_ns = ns; |
| 3674 | ns->root = mnt; |
| 3675 | ns->mounts = 1; |
| 3676 | list_add(&mnt->mnt_list, &ns->list); |
| 3677 | mntget(newmount.mnt); |
| 3678 | |
| 3679 | /* Attach to an apparent O_PATH fd with a note that we need to unmount |
| 3680 | * it, not just simply put it. |
| 3681 | */ |
| 3682 | file = dentry_open(&newmount, O_PATH, fc->cred); |
| 3683 | if (IS_ERR(file)) { |
| 3684 | dissolve_on_fput(newmount.mnt); |
| 3685 | ret = PTR_ERR(file); |
| 3686 | goto err_path; |
| 3687 | } |
| 3688 | file->f_mode |= FMODE_NEED_UNMOUNT; |
| 3689 | |
| 3690 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); |
| 3691 | if (ret >= 0) |
| 3692 | fd_install(ret, file); |
| 3693 | else |
| 3694 | fput(file); |
| 3695 | |
| 3696 | err_path: |
| 3697 | path_put(&newmount); |
| 3698 | err_unlock: |
| 3699 | mutex_unlock(&fc->uapi_mutex); |
| 3700 | err_fsfd: |
| 3701 | fdput(f); |
| 3702 | return ret; |
| 3703 | } |
| 3704 | |
| 3705 | /* |
| 3706 | * Move a mount from one place to another. In combination with |
| 3707 | * fsopen()/fsmount() this is used to install a new mount and in combination |
| 3708 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy |
| 3709 | * a mount subtree. |
| 3710 | * |
| 3711 | * Note the flags value is a combination of MOVE_MOUNT_* flags. |
| 3712 | */ |
| 3713 | SYSCALL_DEFINE5(move_mount, |
| 3714 | int, from_dfd, const char __user *, from_pathname, |
| 3715 | int, to_dfd, const char __user *, to_pathname, |
| 3716 | unsigned int, flags) |
| 3717 | { |
| 3718 | struct path from_path, to_path; |
| 3719 | unsigned int lflags; |
| 3720 | int ret = 0; |
| 3721 | |
| 3722 | if (!may_mount()) |
| 3723 | return -EPERM; |
| 3724 | |
| 3725 | if (flags & ~MOVE_MOUNT__MASK) |
| 3726 | return -EINVAL; |
| 3727 | |
| 3728 | /* If someone gives a pathname, they aren't permitted to move |
| 3729 | * from an fd that requires unmount as we can't get at the flag |
| 3730 | * to clear it afterwards. |
| 3731 | */ |
| 3732 | lflags = 0; |
| 3733 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; |
| 3734 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; |
| 3735 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; |
| 3736 | |
| 3737 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); |
| 3738 | if (ret < 0) |
| 3739 | return ret; |
| 3740 | |
| 3741 | lflags = 0; |
| 3742 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; |
| 3743 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; |
| 3744 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; |
| 3745 | |
| 3746 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); |
| 3747 | if (ret < 0) |
| 3748 | goto out_from; |
| 3749 | |
| 3750 | ret = security_move_mount(&from_path, &to_path); |
| 3751 | if (ret < 0) |
| 3752 | goto out_to; |
| 3753 | |
| 3754 | if (flags & MOVE_MOUNT_SET_GROUP) |
| 3755 | ret = do_set_group(&from_path, &to_path); |
| 3756 | else |
| 3757 | ret = do_move_mount(&from_path, &to_path); |
| 3758 | |
| 3759 | out_to: |
| 3760 | path_put(&to_path); |
| 3761 | out_from: |
| 3762 | path_put(&from_path); |
| 3763 | return ret; |
| 3764 | } |
| 3765 | |
| 3766 | /* |
| 3767 | * Return true if path is reachable from root |
| 3768 | * |
| 3769 | * namespace_sem or mount_lock is held |
| 3770 | */ |
| 3771 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
| 3772 | const struct path *root) |
| 3773 | { |
| 3774 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
| 3775 | dentry = mnt->mnt_mountpoint; |
| 3776 | mnt = mnt->mnt_parent; |
| 3777 | } |
| 3778 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
| 3779 | } |
| 3780 | |
| 3781 | bool path_is_under(const struct path *path1, const struct path *path2) |
| 3782 | { |
| 3783 | bool res; |
| 3784 | read_seqlock_excl(&mount_lock); |
| 3785 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
| 3786 | read_sequnlock_excl(&mount_lock); |
| 3787 | return res; |
| 3788 | } |
| 3789 | EXPORT_SYMBOL(path_is_under); |
| 3790 | |
| 3791 | /* |
| 3792 | * pivot_root Semantics: |
| 3793 | * Moves the root file system of the current process to the directory put_old, |
| 3794 | * makes new_root as the new root file system of the current process, and sets |
| 3795 | * root/cwd of all processes which had them on the current root to new_root. |
| 3796 | * |
| 3797 | * Restrictions: |
| 3798 | * The new_root and put_old must be directories, and must not be on the |
| 3799 | * same file system as the current process root. The put_old must be |
| 3800 | * underneath new_root, i.e. adding a non-zero number of /.. to the string |
| 3801 | * pointed to by put_old must yield the same directory as new_root. No other |
| 3802 | * file system may be mounted on put_old. After all, new_root is a mountpoint. |
| 3803 | * |
| 3804 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
| 3805 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
| 3806 | * in this situation. |
| 3807 | * |
| 3808 | * Notes: |
| 3809 | * - we don't move root/cwd if they are not at the root (reason: if something |
| 3810 | * cared enough to change them, it's probably wrong to force them elsewhere) |
| 3811 | * - it's okay to pick a root that isn't the root of a file system, e.g. |
| 3812 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, |
| 3813 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root |
| 3814 | * first. |
| 3815 | */ |
| 3816 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
| 3817 | const char __user *, put_old) |
| 3818 | { |
| 3819 | struct path new, old, root; |
| 3820 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; |
| 3821 | struct mountpoint *old_mp, *root_mp; |
| 3822 | int error; |
| 3823 | |
| 3824 | if (!may_mount()) |
| 3825 | return -EPERM; |
| 3826 | |
| 3827 | error = user_path_at(AT_FDCWD, new_root, |
| 3828 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); |
| 3829 | if (error) |
| 3830 | goto out0; |
| 3831 | |
| 3832 | error = user_path_at(AT_FDCWD, put_old, |
| 3833 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); |
| 3834 | if (error) |
| 3835 | goto out1; |
| 3836 | |
| 3837 | error = security_sb_pivotroot(&old, &new); |
| 3838 | if (error) |
| 3839 | goto out2; |
| 3840 | |
| 3841 | get_fs_root(current->fs, &root); |
| 3842 | old_mp = lock_mount(&old); |
| 3843 | error = PTR_ERR(old_mp); |
| 3844 | if (IS_ERR(old_mp)) |
| 3845 | goto out3; |
| 3846 | |
| 3847 | error = -EINVAL; |
| 3848 | new_mnt = real_mount(new.mnt); |
| 3849 | root_mnt = real_mount(root.mnt); |
| 3850 | old_mnt = real_mount(old.mnt); |
| 3851 | ex_parent = new_mnt->mnt_parent; |
| 3852 | root_parent = root_mnt->mnt_parent; |
| 3853 | if (IS_MNT_SHARED(old_mnt) || |
| 3854 | IS_MNT_SHARED(ex_parent) || |
| 3855 | IS_MNT_SHARED(root_parent)) |
| 3856 | goto out4; |
| 3857 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
| 3858 | goto out4; |
| 3859 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
| 3860 | goto out4; |
| 3861 | error = -ENOENT; |
| 3862 | if (d_unlinked(new.dentry)) |
| 3863 | goto out4; |
| 3864 | error = -EBUSY; |
| 3865 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
| 3866 | goto out4; /* loop, on the same file system */ |
| 3867 | error = -EINVAL; |
| 3868 | if (root.mnt->mnt_root != root.dentry) |
| 3869 | goto out4; /* not a mountpoint */ |
| 3870 | if (!mnt_has_parent(root_mnt)) |
| 3871 | goto out4; /* not attached */ |
| 3872 | if (new.mnt->mnt_root != new.dentry) |
| 3873 | goto out4; /* not a mountpoint */ |
| 3874 | if (!mnt_has_parent(new_mnt)) |
| 3875 | goto out4; /* not attached */ |
| 3876 | /* make sure we can reach put_old from new_root */ |
| 3877 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
| 3878 | goto out4; |
| 3879 | /* make certain new is below the root */ |
| 3880 | if (!is_path_reachable(new_mnt, new.dentry, &root)) |
| 3881 | goto out4; |
| 3882 | lock_mount_hash(); |
| 3883 | umount_mnt(new_mnt); |
| 3884 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ |
| 3885 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
| 3886 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; |
| 3887 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
| 3888 | } |
| 3889 | /* mount old root on put_old */ |
| 3890 | attach_mnt(root_mnt, old_mnt, old_mp); |
| 3891 | /* mount new_root on / */ |
| 3892 | attach_mnt(new_mnt, root_parent, root_mp); |
| 3893 | mnt_add_count(root_parent, -1); |
| 3894 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
| 3895 | /* A moved mount should not expire automatically */ |
| 3896 | list_del_init(&new_mnt->mnt_expire); |
| 3897 | put_mountpoint(root_mp); |
| 3898 | unlock_mount_hash(); |
| 3899 | chroot_fs_refs(&root, &new); |
| 3900 | error = 0; |
| 3901 | out4: |
| 3902 | unlock_mount(old_mp); |
| 3903 | if (!error) |
| 3904 | mntput_no_expire(ex_parent); |
| 3905 | out3: |
| 3906 | path_put(&root); |
| 3907 | out2: |
| 3908 | path_put(&old); |
| 3909 | out1: |
| 3910 | path_put(&new); |
| 3911 | out0: |
| 3912 | return error; |
| 3913 | } |
| 3914 | |
| 3915 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
| 3916 | { |
| 3917 | unsigned int flags = mnt->mnt.mnt_flags; |
| 3918 | |
| 3919 | /* flags to clear */ |
| 3920 | flags &= ~kattr->attr_clr; |
| 3921 | /* flags to raise */ |
| 3922 | flags |= kattr->attr_set; |
| 3923 | |
| 3924 | return flags; |
| 3925 | } |
| 3926 | |
| 3927 | static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
| 3928 | { |
| 3929 | struct vfsmount *m = &mnt->mnt; |
| 3930 | |
| 3931 | if (!kattr->mnt_userns) |
| 3932 | return 0; |
| 3933 | |
| 3934 | /* |
| 3935 | * Once a mount has been idmapped we don't allow it to change its |
| 3936 | * mapping. It makes things simpler and callers can just create |
| 3937 | * another bind-mount they can idmap if they want to. |
| 3938 | */ |
| 3939 | if (mnt_user_ns(m) != &init_user_ns) |
| 3940 | return -EPERM; |
| 3941 | |
| 3942 | /* The underlying filesystem doesn't support idmapped mounts yet. */ |
| 3943 | if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) |
| 3944 | return -EINVAL; |
| 3945 | |
| 3946 | /* Don't yet support filesystem mountable in user namespaces. */ |
| 3947 | if (m->mnt_sb->s_user_ns != &init_user_ns) |
| 3948 | return -EINVAL; |
| 3949 | |
| 3950 | /* We're not controlling the superblock. */ |
| 3951 | if (!capable(CAP_SYS_ADMIN)) |
| 3952 | return -EPERM; |
| 3953 | |
| 3954 | /* Mount has already been visible in the filesystem hierarchy. */ |
| 3955 | if (!is_anon_ns(mnt->mnt_ns)) |
| 3956 | return -EINVAL; |
| 3957 | |
| 3958 | return 0; |
| 3959 | } |
| 3960 | |
| 3961 | static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, |
| 3962 | struct mount *mnt, int *err) |
| 3963 | { |
| 3964 | struct mount *m = mnt, *last = NULL; |
| 3965 | |
| 3966 | if (!is_mounted(&m->mnt)) { |
| 3967 | *err = -EINVAL; |
| 3968 | goto out; |
| 3969 | } |
| 3970 | |
| 3971 | if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { |
| 3972 | *err = -EINVAL; |
| 3973 | goto out; |
| 3974 | } |
| 3975 | |
| 3976 | do { |
| 3977 | unsigned int flags; |
| 3978 | |
| 3979 | flags = recalc_flags(kattr, m); |
| 3980 | if (!can_change_locked_flags(m, flags)) { |
| 3981 | *err = -EPERM; |
| 3982 | goto out; |
| 3983 | } |
| 3984 | |
| 3985 | *err = can_idmap_mount(kattr, m); |
| 3986 | if (*err) |
| 3987 | goto out; |
| 3988 | |
| 3989 | last = m; |
| 3990 | |
| 3991 | if ((kattr->attr_set & MNT_READONLY) && |
| 3992 | !(m->mnt.mnt_flags & MNT_READONLY)) { |
| 3993 | *err = mnt_hold_writers(m); |
| 3994 | if (*err) |
| 3995 | goto out; |
| 3996 | } |
| 3997 | } while (kattr->recurse && (m = next_mnt(m, mnt))); |
| 3998 | |
| 3999 | out: |
| 4000 | return last; |
| 4001 | } |
| 4002 | |
| 4003 | static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
| 4004 | { |
| 4005 | struct user_namespace *mnt_userns; |
| 4006 | |
| 4007 | if (!kattr->mnt_userns) |
| 4008 | return; |
| 4009 | |
| 4010 | mnt_userns = get_user_ns(kattr->mnt_userns); |
| 4011 | /* Pairs with smp_load_acquire() in mnt_user_ns(). */ |
| 4012 | smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); |
| 4013 | } |
| 4014 | |
| 4015 | static void mount_setattr_commit(struct mount_kattr *kattr, |
| 4016 | struct mount *mnt, struct mount *last, |
| 4017 | int err) |
| 4018 | { |
| 4019 | struct mount *m = mnt; |
| 4020 | |
| 4021 | do { |
| 4022 | if (!err) { |
| 4023 | unsigned int flags; |
| 4024 | |
| 4025 | do_idmap_mount(kattr, m); |
| 4026 | flags = recalc_flags(kattr, m); |
| 4027 | WRITE_ONCE(m->mnt.mnt_flags, flags); |
| 4028 | } |
| 4029 | |
| 4030 | /* |
| 4031 | * We either set MNT_READONLY above so make it visible |
| 4032 | * before ~MNT_WRITE_HOLD or we failed to recursively |
| 4033 | * apply mount options. |
| 4034 | */ |
| 4035 | if ((kattr->attr_set & MNT_READONLY) && |
| 4036 | (m->mnt.mnt_flags & MNT_WRITE_HOLD)) |
| 4037 | mnt_unhold_writers(m); |
| 4038 | |
| 4039 | if (!err && kattr->propagation) |
| 4040 | change_mnt_propagation(m, kattr->propagation); |
| 4041 | |
| 4042 | /* |
| 4043 | * On failure, only cleanup until we found the first mount |
| 4044 | * we failed to handle. |
| 4045 | */ |
| 4046 | if (err && m == last) |
| 4047 | break; |
| 4048 | } while (kattr->recurse && (m = next_mnt(m, mnt))); |
| 4049 | |
| 4050 | if (!err) |
| 4051 | touch_mnt_namespace(mnt->mnt_ns); |
| 4052 | } |
| 4053 | |
| 4054 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) |
| 4055 | { |
| 4056 | struct mount *mnt = real_mount(path->mnt), *last = NULL; |
| 4057 | int err = 0; |
| 4058 | |
| 4059 | if (path->dentry != mnt->mnt.mnt_root) |
| 4060 | return -EINVAL; |
| 4061 | |
| 4062 | if (kattr->propagation) { |
| 4063 | /* |
| 4064 | * Only take namespace_lock() if we're actually changing |
| 4065 | * propagation. |
| 4066 | */ |
| 4067 | namespace_lock(); |
| 4068 | if (kattr->propagation == MS_SHARED) { |
| 4069 | err = invent_group_ids(mnt, kattr->recurse); |
| 4070 | if (err) { |
| 4071 | namespace_unlock(); |
| 4072 | return err; |
| 4073 | } |
| 4074 | } |
| 4075 | } |
| 4076 | |
| 4077 | lock_mount_hash(); |
| 4078 | |
| 4079 | /* |
| 4080 | * Get the mount tree in a shape where we can change mount |
| 4081 | * properties without failure. |
| 4082 | */ |
| 4083 | last = mount_setattr_prepare(kattr, mnt, &err); |
| 4084 | if (last) /* Commit all changes or revert to the old state. */ |
| 4085 | mount_setattr_commit(kattr, mnt, last, err); |
| 4086 | |
| 4087 | unlock_mount_hash(); |
| 4088 | |
| 4089 | if (kattr->propagation) { |
| 4090 | namespace_unlock(); |
| 4091 | if (err) |
| 4092 | cleanup_group_ids(mnt, NULL); |
| 4093 | } |
| 4094 | |
| 4095 | return err; |
| 4096 | } |
| 4097 | |
| 4098 | static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, |
| 4099 | struct mount_kattr *kattr, unsigned int flags) |
| 4100 | { |
| 4101 | int err = 0; |
| 4102 | struct ns_common *ns; |
| 4103 | struct user_namespace *mnt_userns; |
| 4104 | struct file *file; |
| 4105 | |
| 4106 | if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) |
| 4107 | return 0; |
| 4108 | |
| 4109 | /* |
| 4110 | * We currently do not support clearing an idmapped mount. If this ever |
| 4111 | * is a use-case we can revisit this but for now let's keep it simple |
| 4112 | * and not allow it. |
| 4113 | */ |
| 4114 | if (attr->attr_clr & MOUNT_ATTR_IDMAP) |
| 4115 | return -EINVAL; |
| 4116 | |
| 4117 | if (attr->userns_fd > INT_MAX) |
| 4118 | return -EINVAL; |
| 4119 | |
| 4120 | file = fget(attr->userns_fd); |
| 4121 | if (!file) |
| 4122 | return -EBADF; |
| 4123 | |
| 4124 | if (!proc_ns_file(file)) { |
| 4125 | err = -EINVAL; |
| 4126 | goto out_fput; |
| 4127 | } |
| 4128 | |
| 4129 | ns = get_proc_ns(file_inode(file)); |
| 4130 | if (ns->ops->type != CLONE_NEWUSER) { |
| 4131 | err = -EINVAL; |
| 4132 | goto out_fput; |
| 4133 | } |
| 4134 | |
| 4135 | /* |
| 4136 | * The init_user_ns is used to indicate that a vfsmount is not idmapped. |
| 4137 | * This is simpler than just having to treat NULL as unmapped. Users |
| 4138 | * wanting to idmap a mount to init_user_ns can just use a namespace |
| 4139 | * with an identity mapping. |
| 4140 | */ |
| 4141 | mnt_userns = container_of(ns, struct user_namespace, ns); |
| 4142 | if (mnt_userns == &init_user_ns) { |
| 4143 | err = -EPERM; |
| 4144 | goto out_fput; |
| 4145 | } |
| 4146 | kattr->mnt_userns = get_user_ns(mnt_userns); |
| 4147 | |
| 4148 | out_fput: |
| 4149 | fput(file); |
| 4150 | return err; |
| 4151 | } |
| 4152 | |
| 4153 | static int build_mount_kattr(const struct mount_attr *attr, size_t usize, |
| 4154 | struct mount_kattr *kattr, unsigned int flags) |
| 4155 | { |
| 4156 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; |
| 4157 | |
| 4158 | if (flags & AT_NO_AUTOMOUNT) |
| 4159 | lookup_flags &= ~LOOKUP_AUTOMOUNT; |
| 4160 | if (flags & AT_SYMLINK_NOFOLLOW) |
| 4161 | lookup_flags &= ~LOOKUP_FOLLOW; |
| 4162 | if (flags & AT_EMPTY_PATH) |
| 4163 | lookup_flags |= LOOKUP_EMPTY; |
| 4164 | |
| 4165 | *kattr = (struct mount_kattr) { |
| 4166 | .lookup_flags = lookup_flags, |
| 4167 | .recurse = !!(flags & AT_RECURSIVE), |
| 4168 | }; |
| 4169 | |
| 4170 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) |
| 4171 | return -EINVAL; |
| 4172 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) |
| 4173 | return -EINVAL; |
| 4174 | kattr->propagation = attr->propagation; |
| 4175 | |
| 4176 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) |
| 4177 | return -EINVAL; |
| 4178 | |
| 4179 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); |
| 4180 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); |
| 4181 | |
| 4182 | /* |
| 4183 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, |
| 4184 | * users wanting to transition to a different atime setting cannot |
| 4185 | * simply specify the atime setting in @attr_set, but must also |
| 4186 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. |
| 4187 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in |
| 4188 | * @attr_clr and that @attr_set can't have any atime bits set if |
| 4189 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. |
| 4190 | */ |
| 4191 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { |
| 4192 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) |
| 4193 | return -EINVAL; |
| 4194 | |
| 4195 | /* |
| 4196 | * Clear all previous time settings as they are mutually |
| 4197 | * exclusive. |
| 4198 | */ |
| 4199 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; |
| 4200 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { |
| 4201 | case MOUNT_ATTR_RELATIME: |
| 4202 | kattr->attr_set |= MNT_RELATIME; |
| 4203 | break; |
| 4204 | case MOUNT_ATTR_NOATIME: |
| 4205 | kattr->attr_set |= MNT_NOATIME; |
| 4206 | break; |
| 4207 | case MOUNT_ATTR_STRICTATIME: |
| 4208 | break; |
| 4209 | default: |
| 4210 | return -EINVAL; |
| 4211 | } |
| 4212 | } else { |
| 4213 | if (attr->attr_set & MOUNT_ATTR__ATIME) |
| 4214 | return -EINVAL; |
| 4215 | } |
| 4216 | |
| 4217 | return build_mount_idmapped(attr, usize, kattr, flags); |
| 4218 | } |
| 4219 | |
| 4220 | static void finish_mount_kattr(struct mount_kattr *kattr) |
| 4221 | { |
| 4222 | put_user_ns(kattr->mnt_userns); |
| 4223 | kattr->mnt_userns = NULL; |
| 4224 | } |
| 4225 | |
| 4226 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, |
| 4227 | unsigned int, flags, struct mount_attr __user *, uattr, |
| 4228 | size_t, usize) |
| 4229 | { |
| 4230 | int err; |
| 4231 | struct path target; |
| 4232 | struct mount_attr attr; |
| 4233 | struct mount_kattr kattr; |
| 4234 | |
| 4235 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); |
| 4236 | |
| 4237 | if (flags & ~(AT_EMPTY_PATH | |
| 4238 | AT_RECURSIVE | |
| 4239 | AT_SYMLINK_NOFOLLOW | |
| 4240 | AT_NO_AUTOMOUNT)) |
| 4241 | return -EINVAL; |
| 4242 | |
| 4243 | if (unlikely(usize > PAGE_SIZE)) |
| 4244 | return -E2BIG; |
| 4245 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) |
| 4246 | return -EINVAL; |
| 4247 | |
| 4248 | if (!may_mount()) |
| 4249 | return -EPERM; |
| 4250 | |
| 4251 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); |
| 4252 | if (err) |
| 4253 | return err; |
| 4254 | |
| 4255 | /* Don't bother walking through the mounts if this is a nop. */ |
| 4256 | if (attr.attr_set == 0 && |
| 4257 | attr.attr_clr == 0 && |
| 4258 | attr.propagation == 0) |
| 4259 | return 0; |
| 4260 | |
| 4261 | err = build_mount_kattr(&attr, usize, &kattr, flags); |
| 4262 | if (err) |
| 4263 | return err; |
| 4264 | |
| 4265 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); |
| 4266 | if (err) |
| 4267 | return err; |
| 4268 | |
| 4269 | err = do_mount_setattr(&target, &kattr); |
| 4270 | finish_mount_kattr(&kattr); |
| 4271 | path_put(&target); |
| 4272 | return err; |
| 4273 | } |
| 4274 | |
| 4275 | static void __init init_mount_tree(void) |
| 4276 | { |
| 4277 | struct vfsmount *mnt; |
| 4278 | struct mount *m; |
| 4279 | struct mnt_namespace *ns; |
| 4280 | struct path root; |
| 4281 | |
| 4282 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
| 4283 | if (IS_ERR(mnt)) |
| 4284 | panic("Can't create rootfs"); |
| 4285 | |
| 4286 | ns = alloc_mnt_ns(&init_user_ns, false); |
| 4287 | if (IS_ERR(ns)) |
| 4288 | panic("Can't allocate initial namespace"); |
| 4289 | m = real_mount(mnt); |
| 4290 | m->mnt_ns = ns; |
| 4291 | ns->root = m; |
| 4292 | ns->mounts = 1; |
| 4293 | list_add(&m->mnt_list, &ns->list); |
| 4294 | init_task.nsproxy->mnt_ns = ns; |
| 4295 | get_mnt_ns(ns); |
| 4296 | |
| 4297 | root.mnt = mnt; |
| 4298 | root.dentry = mnt->mnt_root; |
| 4299 | mnt->mnt_flags |= MNT_LOCKED; |
| 4300 | |
| 4301 | set_fs_pwd(current->fs, &root); |
| 4302 | set_fs_root(current->fs, &root); |
| 4303 | } |
| 4304 | |
| 4305 | void __init mnt_init(void) |
| 4306 | { |
| 4307 | int err; |
| 4308 | |
| 4309 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
| 4310 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
| 4311 | |
| 4312 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
| 4313 | sizeof(struct hlist_head), |
| 4314 | mhash_entries, 19, |
| 4315 | HASH_ZERO, |
| 4316 | &m_hash_shift, &m_hash_mask, 0, 0); |
| 4317 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", |
| 4318 | sizeof(struct hlist_head), |
| 4319 | mphash_entries, 19, |
| 4320 | HASH_ZERO, |
| 4321 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
| 4322 | |
| 4323 | if (!mount_hashtable || !mountpoint_hashtable) |
| 4324 | panic("Failed to allocate mount hash table\n"); |
| 4325 | |
| 4326 | kernfs_init(); |
| 4327 | |
| 4328 | err = sysfs_init(); |
| 4329 | if (err) |
| 4330 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", |
| 4331 | __func__, err); |
| 4332 | fs_kobj = kobject_create_and_add("fs", NULL); |
| 4333 | if (!fs_kobj) |
| 4334 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
| 4335 | shmem_init(); |
| 4336 | init_rootfs(); |
| 4337 | init_mount_tree(); |
| 4338 | } |
| 4339 | |
| 4340 | void put_mnt_ns(struct mnt_namespace *ns) |
| 4341 | { |
| 4342 | if (!refcount_dec_and_test(&ns->ns.count)) |
| 4343 | return; |
| 4344 | drop_collected_mounts(&ns->root->mnt); |
| 4345 | free_mnt_ns(ns); |
| 4346 | } |
| 4347 | |
| 4348 | struct vfsmount *kern_mount(struct file_system_type *type) |
| 4349 | { |
| 4350 | struct vfsmount *mnt; |
| 4351 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
| 4352 | if (!IS_ERR(mnt)) { |
| 4353 | /* |
| 4354 | * it is a longterm mount, don't release mnt until |
| 4355 | * we unmount before file sys is unregistered |
| 4356 | */ |
| 4357 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
| 4358 | } |
| 4359 | return mnt; |
| 4360 | } |
| 4361 | EXPORT_SYMBOL_GPL(kern_mount); |
| 4362 | |
| 4363 | void kern_unmount(struct vfsmount *mnt) |
| 4364 | { |
| 4365 | /* release long term mount so mount point can be released */ |
| 4366 | if (!IS_ERR_OR_NULL(mnt)) { |
| 4367 | real_mount(mnt)->mnt_ns = NULL; |
| 4368 | synchronize_rcu(); /* yecchhh... */ |
| 4369 | mntput(mnt); |
| 4370 | } |
| 4371 | } |
| 4372 | EXPORT_SYMBOL(kern_unmount); |
| 4373 | |
| 4374 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
| 4375 | { |
| 4376 | unsigned int i; |
| 4377 | |
| 4378 | for (i = 0; i < num; i++) |
| 4379 | if (mnt[i]) |
| 4380 | real_mount(mnt[i])->mnt_ns = NULL; |
| 4381 | synchronize_rcu_expedited(); |
| 4382 | for (i = 0; i < num; i++) |
| 4383 | mntput(mnt[i]); |
| 4384 | } |
| 4385 | EXPORT_SYMBOL(kern_unmount_array); |
| 4386 | |
| 4387 | bool our_mnt(struct vfsmount *mnt) |
| 4388 | { |
| 4389 | return check_mnt(real_mount(mnt)); |
| 4390 | } |
| 4391 | |
| 4392 | bool current_chrooted(void) |
| 4393 | { |
| 4394 | /* Does the current process have a non-standard root */ |
| 4395 | struct path ns_root; |
| 4396 | struct path fs_root; |
| 4397 | bool chrooted; |
| 4398 | |
| 4399 | /* Find the namespace root */ |
| 4400 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; |
| 4401 | ns_root.dentry = ns_root.mnt->mnt_root; |
| 4402 | path_get(&ns_root); |
| 4403 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) |
| 4404 | ; |
| 4405 | |
| 4406 | get_fs_root(current->fs, &fs_root); |
| 4407 | |
| 4408 | chrooted = !path_equal(&fs_root, &ns_root); |
| 4409 | |
| 4410 | path_put(&fs_root); |
| 4411 | path_put(&ns_root); |
| 4412 | |
| 4413 | return chrooted; |
| 4414 | } |
| 4415 | |
| 4416 | static bool mnt_already_visible(struct mnt_namespace *ns, |
| 4417 | const struct super_block *sb, |
| 4418 | int *new_mnt_flags) |
| 4419 | { |
| 4420 | int new_flags = *new_mnt_flags; |
| 4421 | struct mount *mnt; |
| 4422 | bool visible = false; |
| 4423 | |
| 4424 | down_read(&namespace_sem); |
| 4425 | lock_ns_list(ns); |
| 4426 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
| 4427 | struct mount *child; |
| 4428 | int mnt_flags; |
| 4429 | |
| 4430 | if (mnt_is_cursor(mnt)) |
| 4431 | continue; |
| 4432 | |
| 4433 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
| 4434 | continue; |
| 4435 | |
| 4436 | /* This mount is not fully visible if it's root directory |
| 4437 | * is not the root directory of the filesystem. |
| 4438 | */ |
| 4439 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) |
| 4440 | continue; |
| 4441 | |
| 4442 | /* A local view of the mount flags */ |
| 4443 | mnt_flags = mnt->mnt.mnt_flags; |
| 4444 | |
| 4445 | /* Don't miss readonly hidden in the superblock flags */ |
| 4446 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
| 4447 | mnt_flags |= MNT_LOCK_READONLY; |
| 4448 | |
| 4449 | /* Verify the mount flags are equal to or more permissive |
| 4450 | * than the proposed new mount. |
| 4451 | */ |
| 4452 | if ((mnt_flags & MNT_LOCK_READONLY) && |
| 4453 | !(new_flags & MNT_READONLY)) |
| 4454 | continue; |
| 4455 | if ((mnt_flags & MNT_LOCK_ATIME) && |
| 4456 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) |
| 4457 | continue; |
| 4458 | |
| 4459 | /* This mount is not fully visible if there are any |
| 4460 | * locked child mounts that cover anything except for |
| 4461 | * empty directories. |
| 4462 | */ |
| 4463 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { |
| 4464 | struct inode *inode = child->mnt_mountpoint->d_inode; |
| 4465 | /* Only worry about locked mounts */ |
| 4466 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
| 4467 | continue; |
| 4468 | /* Is the directory permanetly empty? */ |
| 4469 | if (!is_empty_dir_inode(inode)) |
| 4470 | goto next; |
| 4471 | } |
| 4472 | /* Preserve the locked attributes */ |
| 4473 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
| 4474 | MNT_LOCK_ATIME); |
| 4475 | visible = true; |
| 4476 | goto found; |
| 4477 | next: ; |
| 4478 | } |
| 4479 | found: |
| 4480 | unlock_ns_list(ns); |
| 4481 | up_read(&namespace_sem); |
| 4482 | return visible; |
| 4483 | } |
| 4484 | |
| 4485 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
| 4486 | { |
| 4487 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
| 4488 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
| 4489 | unsigned long s_iflags; |
| 4490 | |
| 4491 | if (ns->user_ns == &init_user_ns) |
| 4492 | return false; |
| 4493 | |
| 4494 | /* Can this filesystem be too revealing? */ |
| 4495 | s_iflags = sb->s_iflags; |
| 4496 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
| 4497 | return false; |
| 4498 | |
| 4499 | if ((s_iflags & required_iflags) != required_iflags) { |
| 4500 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", |
| 4501 | required_iflags); |
| 4502 | return true; |
| 4503 | } |
| 4504 | |
| 4505 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
| 4506 | } |
| 4507 | |
| 4508 | bool mnt_may_suid(struct vfsmount *mnt) |
| 4509 | { |
| 4510 | /* |
| 4511 | * Foreign mounts (accessed via fchdir or through /proc |
| 4512 | * symlinks) are always treated as if they are nosuid. This |
| 4513 | * prevents namespaces from trusting potentially unsafe |
| 4514 | * suid/sgid bits, file caps, or security labels that originate |
| 4515 | * in other namespaces. |
| 4516 | */ |
| 4517 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && |
| 4518 | current_in_userns(mnt->mnt_sb->s_user_ns); |
| 4519 | } |
| 4520 | |
| 4521 | static struct ns_common *mntns_get(struct task_struct *task) |
| 4522 | { |
| 4523 | struct ns_common *ns = NULL; |
| 4524 | struct nsproxy *nsproxy; |
| 4525 | |
| 4526 | task_lock(task); |
| 4527 | nsproxy = task->nsproxy; |
| 4528 | if (nsproxy) { |
| 4529 | ns = &nsproxy->mnt_ns->ns; |
| 4530 | get_mnt_ns(to_mnt_ns(ns)); |
| 4531 | } |
| 4532 | task_unlock(task); |
| 4533 | |
| 4534 | return ns; |
| 4535 | } |
| 4536 | |
| 4537 | static void mntns_put(struct ns_common *ns) |
| 4538 | { |
| 4539 | put_mnt_ns(to_mnt_ns(ns)); |
| 4540 | } |
| 4541 | |
| 4542 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
| 4543 | { |
| 4544 | struct nsproxy *nsproxy = nsset->nsproxy; |
| 4545 | struct fs_struct *fs = nsset->fs; |
| 4546 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
| 4547 | struct user_namespace *user_ns = nsset->cred->user_ns; |
| 4548 | struct path root; |
| 4549 | int err; |
| 4550 | |
| 4551 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
| 4552 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
| 4553 | !ns_capable(user_ns, CAP_SYS_ADMIN)) |
| 4554 | return -EPERM; |
| 4555 | |
| 4556 | if (is_anon_ns(mnt_ns)) |
| 4557 | return -EINVAL; |
| 4558 | |
| 4559 | if (fs->users != 1) |
| 4560 | return -EINVAL; |
| 4561 | |
| 4562 | get_mnt_ns(mnt_ns); |
| 4563 | old_mnt_ns = nsproxy->mnt_ns; |
| 4564 | nsproxy->mnt_ns = mnt_ns; |
| 4565 | |
| 4566 | /* Find the root */ |
| 4567 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
| 4568 | "/", LOOKUP_DOWN, &root); |
| 4569 | if (err) { |
| 4570 | /* revert to old namespace */ |
| 4571 | nsproxy->mnt_ns = old_mnt_ns; |
| 4572 | put_mnt_ns(mnt_ns); |
| 4573 | return err; |
| 4574 | } |
| 4575 | |
| 4576 | put_mnt_ns(old_mnt_ns); |
| 4577 | |
| 4578 | /* Update the pwd and root */ |
| 4579 | set_fs_pwd(fs, &root); |
| 4580 | set_fs_root(fs, &root); |
| 4581 | |
| 4582 | path_put(&root); |
| 4583 | return 0; |
| 4584 | } |
| 4585 | |
| 4586 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
| 4587 | { |
| 4588 | return to_mnt_ns(ns)->user_ns; |
| 4589 | } |
| 4590 | |
| 4591 | const struct proc_ns_operations mntns_operations = { |
| 4592 | .name = "mnt", |
| 4593 | .type = CLONE_NEWNS, |
| 4594 | .get = mntns_get, |
| 4595 | .put = mntns_put, |
| 4596 | .install = mntns_install, |
| 4597 | .owner = mntns_owner, |
| 4598 | }; |