11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 260818bf27SAl Viro #include <linux/bootmem.h> 279ea459e1SAl Viro #include <linux/task_work.h> 289164bb4aSIngo Molnar #include <linux/sched/task.h> 299164bb4aSIngo Molnar 3007b20889SRam Pai #include "pnode.h" 31948730b0SAdrian Bunk #include "internal.h" 321da177e4SLinus Torvalds 33d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 34d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 35d2921684SEric W. Biederman 360818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 370818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 380818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 390818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 400818bf27SAl Viro 410818bf27SAl Viro static __initdata unsigned long mhash_entries; 420818bf27SAl Viro static int __init set_mhash_entries(char *str) 430818bf27SAl Viro { 440818bf27SAl Viro if (!str) 450818bf27SAl Viro return 0; 460818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 470818bf27SAl Viro return 1; 480818bf27SAl Viro } 490818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 500818bf27SAl Viro 510818bf27SAl Viro static __initdata unsigned long mphash_entries; 520818bf27SAl Viro static int __init set_mphash_entries(char *str) 530818bf27SAl Viro { 540818bf27SAl Viro if (!str) 550818bf27SAl Viro return 0; 560818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 570818bf27SAl Viro return 1; 580818bf27SAl Viro } 590818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6013f14b4dSEric Dumazet 61c7999c36SAl Viro static u64 event; 6273cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 63719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 6499b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 65f21f6220SAl Viro static int mnt_id_start = 0; 66f21f6220SAl Viro static int mnt_group_start = 1; 675addc5ddSAl Viro 6838129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 690818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 70e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7159aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 721da177e4SLinus Torvalds 73f87fd4c2SMiklos Szeredi /* /sys/fs */ 7400d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7500d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 76f87fd4c2SMiklos Szeredi 7799b7db7bSNick Piggin /* 7899b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7999b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 8099b7db7bSNick Piggin * up the tree. 8199b7db7bSNick Piggin * 8299b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8399b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8499b7db7bSNick Piggin */ 8548a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8699b7db7bSNick Piggin 8738129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 881da177e4SLinus Torvalds { 891da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 901da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 910818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 920818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 930818bf27SAl Viro } 940818bf27SAl Viro 950818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 960818bf27SAl Viro { 970818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 980818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 990818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 102b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10373cd49ecSMiklos Szeredi { 10473cd49ecSMiklos Szeredi int res; 10573cd49ecSMiklos Szeredi 10673cd49ecSMiklos Szeredi retry: 10773cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 10899b7db7bSNick Piggin spin_lock(&mnt_id_lock); 10915169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 110f21f6220SAl Viro if (!res) 11115169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 11299b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 11373cd49ecSMiklos Szeredi if (res == -EAGAIN) 11473cd49ecSMiklos Szeredi goto retry; 11573cd49ecSMiklos Szeredi 11673cd49ecSMiklos Szeredi return res; 11773cd49ecSMiklos Szeredi } 11873cd49ecSMiklos Szeredi 119b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 12073cd49ecSMiklos Szeredi { 12115169fe7SAl Viro int id = mnt->mnt_id; 12299b7db7bSNick Piggin spin_lock(&mnt_id_lock); 123f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 124f21f6220SAl Viro if (mnt_id_start > id) 125f21f6220SAl Viro mnt_id_start = id; 12699b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 12773cd49ecSMiklos Szeredi } 12873cd49ecSMiklos Szeredi 129719f5d7fSMiklos Szeredi /* 130719f5d7fSMiklos Szeredi * Allocate a new peer group ID 131719f5d7fSMiklos Szeredi * 132719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 133719f5d7fSMiklos Szeredi */ 1344b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 135719f5d7fSMiklos Szeredi { 136f21f6220SAl Viro int res; 137f21f6220SAl Viro 138719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 139719f5d7fSMiklos Szeredi return -ENOMEM; 140719f5d7fSMiklos Szeredi 141f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 142f21f6220SAl Viro mnt_group_start, 14315169fe7SAl Viro &mnt->mnt_group_id); 144f21f6220SAl Viro if (!res) 14515169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 146f21f6220SAl Viro 147f21f6220SAl Viro return res; 148719f5d7fSMiklos Szeredi } 149719f5d7fSMiklos Szeredi 150719f5d7fSMiklos Szeredi /* 151719f5d7fSMiklos Szeredi * Release a peer group ID 152719f5d7fSMiklos Szeredi */ 1534b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 154719f5d7fSMiklos Szeredi { 15515169fe7SAl Viro int id = mnt->mnt_group_id; 156f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 157f21f6220SAl Viro if (mnt_group_start > id) 158f21f6220SAl Viro mnt_group_start = id; 15915169fe7SAl Viro mnt->mnt_group_id = 0; 160719f5d7fSMiklos Szeredi } 161719f5d7fSMiklos Szeredi 162b3e19d92SNick Piggin /* 163b3e19d92SNick Piggin * vfsmount lock must be held for read 164b3e19d92SNick Piggin */ 16583adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 166b3e19d92SNick Piggin { 167b3e19d92SNick Piggin #ifdef CONFIG_SMP 16868e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 169b3e19d92SNick Piggin #else 170b3e19d92SNick Piggin preempt_disable(); 17168e8a9feSAl Viro mnt->mnt_count += n; 172b3e19d92SNick Piggin preempt_enable(); 173b3e19d92SNick Piggin #endif 174b3e19d92SNick Piggin } 175b3e19d92SNick Piggin 176b3e19d92SNick Piggin /* 177b3e19d92SNick Piggin * vfsmount lock must be held for write 178b3e19d92SNick Piggin */ 17983adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 180b3e19d92SNick Piggin { 181b3e19d92SNick Piggin #ifdef CONFIG_SMP 182f03c6599SAl Viro unsigned int count = 0; 183b3e19d92SNick Piggin int cpu; 184b3e19d92SNick Piggin 185b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18668e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 187b3e19d92SNick Piggin } 188b3e19d92SNick Piggin 189b3e19d92SNick Piggin return count; 190b3e19d92SNick Piggin #else 19168e8a9feSAl Viro return mnt->mnt_count; 192b3e19d92SNick Piggin #endif 193b3e19d92SNick Piggin } 194b3e19d92SNick Piggin 19587b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 19687b95ce0SAl Viro { 19787b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 19887b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 19987b95ce0SAl Viro pin_remove(p); 20087b95ce0SAl Viro mntput(&m->mnt); 20187b95ce0SAl Viro } 20287b95ce0SAl Viro 203b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 2041da177e4SLinus Torvalds { 205c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 206c63181e6SAl Viro if (mnt) { 20773cd49ecSMiklos Szeredi int err; 20873cd49ecSMiklos Szeredi 209c63181e6SAl Viro err = mnt_alloc_id(mnt); 21088b38782SLi Zefan if (err) 21188b38782SLi Zefan goto out_free_cache; 21288b38782SLi Zefan 21388b38782SLi Zefan if (name) { 214fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 215c63181e6SAl Viro if (!mnt->mnt_devname) 21688b38782SLi Zefan goto out_free_id; 21773cd49ecSMiklos Szeredi } 21873cd49ecSMiklos Szeredi 219b3e19d92SNick Piggin #ifdef CONFIG_SMP 220c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 221c63181e6SAl Viro if (!mnt->mnt_pcp) 222b3e19d92SNick Piggin goto out_free_devname; 223b3e19d92SNick Piggin 224c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 225b3e19d92SNick Piggin #else 226c63181e6SAl Viro mnt->mnt_count = 1; 227c63181e6SAl Viro mnt->mnt_writers = 0; 228b3e19d92SNick Piggin #endif 229b3e19d92SNick Piggin 23038129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 232c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 233c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 234c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 235c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 236c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 237c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2380a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 23999b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 24087b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2411da177e4SLinus Torvalds } 242c63181e6SAl Viro return mnt; 24388b38782SLi Zefan 244d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 245d3ef3d73Snpiggin@suse.de out_free_devname: 246fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 247d3ef3d73Snpiggin@suse.de #endif 24888b38782SLi Zefan out_free_id: 249c63181e6SAl Viro mnt_free_id(mnt); 25088b38782SLi Zefan out_free_cache: 251c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 25288b38782SLi Zefan return NULL; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2558366025eSDave Hansen /* 2568366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2578366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2588366025eSDave Hansen * We must keep track of when those operations start 2598366025eSDave Hansen * (for permission checks) and when they end, so that 2608366025eSDave Hansen * we can determine when writes are able to occur to 2618366025eSDave Hansen * a filesystem. 2628366025eSDave Hansen */ 2633d733633SDave Hansen /* 2643d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2653d733633SDave Hansen * @mnt: the mount to check for its write status 2663d733633SDave Hansen * 2673d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2683d733633SDave Hansen * It does not guarantee that the filesystem will stay 2693d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2703d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2713d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2723d733633SDave Hansen * r/w. 2733d733633SDave Hansen */ 2743d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2753d733633SDave Hansen { 2762e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2772e4b7fcdSDave Hansen return 1; 278bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt_sb)) 2792e4b7fcdSDave Hansen return 1; 2802e4b7fcdSDave Hansen return 0; 2813d733633SDave Hansen } 2823d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2833d733633SDave Hansen 28483adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2853d733633SDave Hansen { 286d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28768e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 288d3ef3d73Snpiggin@suse.de #else 28968e8a9feSAl Viro mnt->mnt_writers++; 290d3ef3d73Snpiggin@suse.de #endif 2913d733633SDave Hansen } 2923d733633SDave Hansen 29383adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2943d733633SDave Hansen { 295d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 29668e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 297d3ef3d73Snpiggin@suse.de #else 29868e8a9feSAl Viro mnt->mnt_writers--; 299d3ef3d73Snpiggin@suse.de #endif 300d3ef3d73Snpiggin@suse.de } 301d3ef3d73Snpiggin@suse.de 30283adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 303d3ef3d73Snpiggin@suse.de { 304d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 305d3ef3d73Snpiggin@suse.de unsigned int count = 0; 3063d733633SDave Hansen int cpu; 3073d733633SDave Hansen 3083d733633SDave Hansen for_each_possible_cpu(cpu) { 30968e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3103d733633SDave Hansen } 3113d733633SDave Hansen 312d3ef3d73Snpiggin@suse.de return count; 313d3ef3d73Snpiggin@suse.de #else 314d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 315d3ef3d73Snpiggin@suse.de #endif 3163d733633SDave Hansen } 3173d733633SDave Hansen 3184ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3194ed5e82fSMiklos Szeredi { 3204ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3214ed5e82fSMiklos Szeredi return 1; 3224ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3234ed5e82fSMiklos Szeredi smp_rmb(); 3244ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3254ed5e82fSMiklos Szeredi } 3264ed5e82fSMiklos Szeredi 3273d733633SDave Hansen /* 328eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 329eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 330eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 331eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3323d733633SDave Hansen */ 3338366025eSDave Hansen /** 334eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 33583adc753SAl Viro * @m: the mount on which to take a write 3368366025eSDave Hansen * 337eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 338eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 339eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 340eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 341eb04c282SJan Kara * called. This is effectively a refcount. 3428366025eSDave Hansen */ 343eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3448366025eSDave Hansen { 34583adc753SAl Viro struct mount *mnt = real_mount(m); 3463d733633SDave Hansen int ret = 0; 3473d733633SDave Hansen 348d3ef3d73Snpiggin@suse.de preempt_disable(); 349c6653a83SNick Piggin mnt_inc_writers(mnt); 350d3ef3d73Snpiggin@suse.de /* 351c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 352d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 353d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 354d3ef3d73Snpiggin@suse.de */ 355d3ef3d73Snpiggin@suse.de smp_mb(); 3566aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 357d3ef3d73Snpiggin@suse.de cpu_relax(); 358d3ef3d73Snpiggin@suse.de /* 359d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 360d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 361d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 362d3ef3d73Snpiggin@suse.de */ 363d3ef3d73Snpiggin@suse.de smp_rmb(); 3644ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 365c6653a83SNick Piggin mnt_dec_writers(mnt); 3663d733633SDave Hansen ret = -EROFS; 3673d733633SDave Hansen } 368d3ef3d73Snpiggin@suse.de preempt_enable(); 369eb04c282SJan Kara 370eb04c282SJan Kara return ret; 371eb04c282SJan Kara } 372eb04c282SJan Kara 373eb04c282SJan Kara /** 374eb04c282SJan Kara * mnt_want_write - get write access to a mount 375eb04c282SJan Kara * @m: the mount on which to take a write 376eb04c282SJan Kara * 377eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 378eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 379eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 380eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 381eb04c282SJan Kara */ 382eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 383eb04c282SJan Kara { 384eb04c282SJan Kara int ret; 385eb04c282SJan Kara 386eb04c282SJan Kara sb_start_write(m->mnt_sb); 387eb04c282SJan Kara ret = __mnt_want_write(m); 388eb04c282SJan Kara if (ret) 389eb04c282SJan Kara sb_end_write(m->mnt_sb); 3903d733633SDave Hansen return ret; 3918366025eSDave Hansen } 3928366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3938366025eSDave Hansen 3948366025eSDave Hansen /** 39596029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 39696029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 39796029c4eSnpiggin@suse.de * 39896029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 39996029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 40096029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 40196029c4eSnpiggin@suse.de * on it. This allows some optimisation. 40296029c4eSnpiggin@suse.de * 40396029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 40496029c4eSnpiggin@suse.de * drop the reference. 40596029c4eSnpiggin@suse.de */ 40696029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 40796029c4eSnpiggin@suse.de { 40896029c4eSnpiggin@suse.de /* superblock may be r/o */ 40996029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 41096029c4eSnpiggin@suse.de return -EROFS; 41196029c4eSnpiggin@suse.de preempt_disable(); 41283adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 41396029c4eSnpiggin@suse.de preempt_enable(); 41496029c4eSnpiggin@suse.de return 0; 41596029c4eSnpiggin@suse.de } 41696029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 41796029c4eSnpiggin@suse.de 41896029c4eSnpiggin@suse.de /** 419eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 420eb04c282SJan Kara * @file: the file who's mount on which to take a write 421eb04c282SJan Kara * 422eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 423eb04c282SJan Kara * do some optimisations if the file is open for write already 424eb04c282SJan Kara */ 425eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 426eb04c282SJan Kara { 42783f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 428eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 429eb04c282SJan Kara else 430eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 431eb04c282SJan Kara } 432eb04c282SJan Kara 433eb04c282SJan Kara /** 4347c6893e3SMiklos Szeredi * mnt_want_write_file_path - get write access to a file's mount 43596029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 43696029c4eSnpiggin@suse.de * 43796029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 43896029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 4397c6893e3SMiklos Szeredi * 4407c6893e3SMiklos Szeredi * Called by the vfs for cases when we have an open file at hand, but will do an 4417c6893e3SMiklos Szeredi * inode operation on it (important distinction for files opened on overlayfs, 4427c6893e3SMiklos Szeredi * since the file operations will come from the real underlying file, while 4437c6893e3SMiklos Szeredi * inode operations come from the overlay). 44496029c4eSnpiggin@suse.de */ 4457c6893e3SMiklos Szeredi int mnt_want_write_file_path(struct file *file) 44696029c4eSnpiggin@suse.de { 447eb04c282SJan Kara int ret; 448eb04c282SJan Kara 449eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 450eb04c282SJan Kara ret = __mnt_want_write_file(file); 451eb04c282SJan Kara if (ret) 452eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 453eb04c282SJan Kara return ret; 45496029c4eSnpiggin@suse.de } 4557c6893e3SMiklos Szeredi 4567c6893e3SMiklos Szeredi static inline int may_write_real(struct file *file) 4577c6893e3SMiklos Szeredi { 4587c6893e3SMiklos Szeredi struct dentry *dentry = file->f_path.dentry; 4597c6893e3SMiklos Szeredi struct dentry *upperdentry; 4607c6893e3SMiklos Szeredi 4617c6893e3SMiklos Szeredi /* Writable file? */ 4627c6893e3SMiklos Szeredi if (file->f_mode & FMODE_WRITER) 4637c6893e3SMiklos Szeredi return 0; 4647c6893e3SMiklos Szeredi 4657c6893e3SMiklos Szeredi /* Not overlayfs? */ 4667c6893e3SMiklos Szeredi if (likely(!(dentry->d_flags & DCACHE_OP_REAL))) 4677c6893e3SMiklos Szeredi return 0; 4687c6893e3SMiklos Szeredi 4697c6893e3SMiklos Szeredi /* File refers to upper, writable layer? */ 4707c6893e3SMiklos Szeredi upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER); 471954c736fSAmir Goldstein if (upperdentry && 472954c736fSAmir Goldstein (file_inode(file) == d_inode(upperdentry) || 473954c736fSAmir Goldstein file_inode(file) == d_inode(dentry))) 4747c6893e3SMiklos Szeredi return 0; 4757c6893e3SMiklos Szeredi 4767c6893e3SMiklos Szeredi /* Lower layer: can't write to real file, sorry... */ 4777c6893e3SMiklos Szeredi return -EPERM; 4787c6893e3SMiklos Szeredi } 4797c6893e3SMiklos Szeredi 4807c6893e3SMiklos Szeredi /** 4817c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4827c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4837c6893e3SMiklos Szeredi * 4847c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4857c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4867c6893e3SMiklos Szeredi * 4877c6893e3SMiklos Szeredi * Mostly called by filesystems from their ioctl operation before performing 4887c6893e3SMiklos Szeredi * modification. On overlayfs this needs to check if the file is on a read-only 4897c6893e3SMiklos Szeredi * lower layer and deny access in that case. 4907c6893e3SMiklos Szeredi */ 4917c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4927c6893e3SMiklos Szeredi { 4937c6893e3SMiklos Szeredi int ret; 4947c6893e3SMiklos Szeredi 4957c6893e3SMiklos Szeredi ret = may_write_real(file); 4967c6893e3SMiklos Szeredi if (!ret) { 4977c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4987c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4997c6893e3SMiklos Szeredi if (ret) 5007c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 5017c6893e3SMiklos Szeredi } 5027c6893e3SMiklos Szeredi return ret; 5037c6893e3SMiklos Szeredi } 50496029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 50596029c4eSnpiggin@suse.de 50696029c4eSnpiggin@suse.de /** 507eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 5088366025eSDave Hansen * @mnt: the mount on which to give up write access 5098366025eSDave Hansen * 5108366025eSDave Hansen * Tells the low-level filesystem that we are done 5118366025eSDave Hansen * performing writes to it. Must be matched with 512eb04c282SJan Kara * __mnt_want_write() call above. 5138366025eSDave Hansen */ 514eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 5158366025eSDave Hansen { 516d3ef3d73Snpiggin@suse.de preempt_disable(); 51783adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 518d3ef3d73Snpiggin@suse.de preempt_enable(); 5198366025eSDave Hansen } 520eb04c282SJan Kara 521eb04c282SJan Kara /** 522eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 523eb04c282SJan Kara * @mnt: the mount on which to give up write access 524eb04c282SJan Kara * 525eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 526eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 527eb04c282SJan Kara * mnt_want_write() call above. 528eb04c282SJan Kara */ 529eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 530eb04c282SJan Kara { 531eb04c282SJan Kara __mnt_drop_write(mnt); 532eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 533eb04c282SJan Kara } 5348366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 5358366025eSDave Hansen 536eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 537eb04c282SJan Kara { 538eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 539eb04c282SJan Kara } 540eb04c282SJan Kara 5417c6893e3SMiklos Szeredi void mnt_drop_write_file_path(struct file *file) 5422a79f17eSAl Viro { 5432a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 5442a79f17eSAl Viro } 5457c6893e3SMiklos Szeredi 5467c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 5477c6893e3SMiklos Szeredi { 5487c6893e3SMiklos Szeredi __mnt_drop_write(file->f_path.mnt); 5497c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 5507c6893e3SMiklos Szeredi } 5512a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 5522a79f17eSAl Viro 55383adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 5548366025eSDave Hansen { 5553d733633SDave Hansen int ret = 0; 5563d733633SDave Hansen 557719ea2fbSAl Viro lock_mount_hash(); 55883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 559d3ef3d73Snpiggin@suse.de /* 560d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 561d3ef3d73Snpiggin@suse.de * should be visible before we do. 562d3ef3d73Snpiggin@suse.de */ 563d3ef3d73Snpiggin@suse.de smp_mb(); 564d3ef3d73Snpiggin@suse.de 565d3ef3d73Snpiggin@suse.de /* 566d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 567d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 568d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 569d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 570d3ef3d73Snpiggin@suse.de * 571d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 572d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 573d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 574d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 575d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 576d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 577d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 578d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 579d3ef3d73Snpiggin@suse.de * we're counting up here. 580d3ef3d73Snpiggin@suse.de */ 581c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 582d3ef3d73Snpiggin@suse.de ret = -EBUSY; 583d3ef3d73Snpiggin@suse.de else 58483adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 585d3ef3d73Snpiggin@suse.de /* 586d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 587d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 588d3ef3d73Snpiggin@suse.de */ 589d3ef3d73Snpiggin@suse.de smp_wmb(); 59083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 591719ea2fbSAl Viro unlock_mount_hash(); 5923d733633SDave Hansen return ret; 5933d733633SDave Hansen } 5948366025eSDave Hansen 59583adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5962e4b7fcdSDave Hansen { 597719ea2fbSAl Viro lock_mount_hash(); 59883adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 599719ea2fbSAl Viro unlock_mount_hash(); 6002e4b7fcdSDave Hansen } 6012e4b7fcdSDave Hansen 6024ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 6034ed5e82fSMiklos Szeredi { 6044ed5e82fSMiklos Szeredi struct mount *mnt; 6054ed5e82fSMiklos Szeredi int err = 0; 6064ed5e82fSMiklos Szeredi 6078e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 6088e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 6098e8b8796SMiklos Szeredi return -EBUSY; 6108e8b8796SMiklos Szeredi 611719ea2fbSAl Viro lock_mount_hash(); 6124ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 6134ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 6144ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 6154ed5e82fSMiklos Szeredi smp_mb(); 6164ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 6174ed5e82fSMiklos Szeredi err = -EBUSY; 6184ed5e82fSMiklos Szeredi break; 6194ed5e82fSMiklos Szeredi } 6204ed5e82fSMiklos Szeredi } 6214ed5e82fSMiklos Szeredi } 6228e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 6238e8b8796SMiklos Szeredi err = -EBUSY; 6248e8b8796SMiklos Szeredi 6254ed5e82fSMiklos Szeredi if (!err) { 6264ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 6274ed5e82fSMiklos Szeredi smp_wmb(); 6284ed5e82fSMiklos Szeredi } 6294ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 6304ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 6314ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 6324ed5e82fSMiklos Szeredi } 633719ea2fbSAl Viro unlock_mount_hash(); 6344ed5e82fSMiklos Szeredi 6354ed5e82fSMiklos Szeredi return err; 6364ed5e82fSMiklos Szeredi } 6374ed5e82fSMiklos Szeredi 638b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 6391da177e4SLinus Torvalds { 640fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 641d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 64268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 643d3ef3d73Snpiggin@suse.de #endif 644b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 6451da177e4SLinus Torvalds } 6461da177e4SLinus Torvalds 6478ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 6488ffcb32eSDavid Howells { 6498ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 6508ffcb32eSDavid Howells } 6518ffcb32eSDavid Howells 65248a066e7SAl Viro /* call under rcu_read_lock */ 653294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 65448a066e7SAl Viro { 65548a066e7SAl Viro struct mount *mnt; 65648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 657294d71ffSAl Viro return 1; 65848a066e7SAl Viro if (bastard == NULL) 659294d71ffSAl Viro return 0; 66048a066e7SAl Viro mnt = real_mount(bastard); 66148a066e7SAl Viro mnt_add_count(mnt, 1); 66248a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 663294d71ffSAl Viro return 0; 66448a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 66548a066e7SAl Viro mnt_add_count(mnt, -1); 666294d71ffSAl Viro return 1; 66748a066e7SAl Viro } 668294d71ffSAl Viro return -1; 669294d71ffSAl Viro } 670294d71ffSAl Viro 671294d71ffSAl Viro /* call under rcu_read_lock */ 672294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 673294d71ffSAl Viro { 674294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 675294d71ffSAl Viro if (likely(!res)) 676294d71ffSAl Viro return true; 677294d71ffSAl Viro if (unlikely(res < 0)) { 67848a066e7SAl Viro rcu_read_unlock(); 67948a066e7SAl Viro mntput(bastard); 68048a066e7SAl Viro rcu_read_lock(); 681294d71ffSAl Viro } 68248a066e7SAl Viro return false; 68348a066e7SAl Viro } 68448a066e7SAl Viro 6851da177e4SLinus Torvalds /* 686474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 68748a066e7SAl Viro * call under rcu_read_lock() 6881da177e4SLinus Torvalds */ 689474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6901da177e4SLinus Torvalds { 69138129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 692474279dcSAl Viro struct mount *p; 6931da177e4SLinus Torvalds 69438129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 695474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 696474279dcSAl Viro return p; 697474279dcSAl Viro return NULL; 6981da177e4SLinus Torvalds } 699474279dcSAl Viro 700474279dcSAl Viro /* 701f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 702f015f126SDavid Howells * 703f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 704f015f126SDavid Howells * following mounts: 705f015f126SDavid Howells * 706f015f126SDavid Howells * mount /dev/sda1 /mnt 707f015f126SDavid Howells * mount /dev/sda2 /mnt 708f015f126SDavid Howells * mount /dev/sda3 /mnt 709f015f126SDavid Howells * 710f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 711f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 712f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 713f015f126SDavid Howells * 714f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 715a05964f3SRam Pai */ 716ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 717a05964f3SRam Pai { 718c7105365SAl Viro struct mount *child_mnt; 71948a066e7SAl Viro struct vfsmount *m; 72048a066e7SAl Viro unsigned seq; 72199b7db7bSNick Piggin 72248a066e7SAl Viro rcu_read_lock(); 72348a066e7SAl Viro do { 72448a066e7SAl Viro seq = read_seqbegin(&mount_lock); 725474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 72648a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 72748a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 72848a066e7SAl Viro rcu_read_unlock(); 72948a066e7SAl Viro return m; 730a05964f3SRam Pai } 731a05964f3SRam Pai 7327af1364fSEric W. Biederman /* 7337af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 7347af1364fSEric W. Biederman * current mount namespace. 7357af1364fSEric W. Biederman * 7367af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 7377af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7387af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7397af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7407af1364fSEric W. Biederman * is a mountpoint. 7417af1364fSEric W. Biederman * 7427af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7437af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7447af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7457af1364fSEric W. Biederman * parent mount. 7467af1364fSEric W. Biederman */ 7477af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7487af1364fSEric W. Biederman { 7497af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7507af1364fSEric W. Biederman struct mount *mnt; 7517af1364fSEric W. Biederman bool is_covered = false; 7527af1364fSEric W. Biederman 7537af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7547af1364fSEric W. Biederman goto out; 7557af1364fSEric W. Biederman 7567af1364fSEric W. Biederman down_read(&namespace_sem); 7577af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7587af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7597af1364fSEric W. Biederman if (is_covered) 7607af1364fSEric W. Biederman break; 7617af1364fSEric W. Biederman } 7627af1364fSEric W. Biederman up_read(&namespace_sem); 7637af1364fSEric W. Biederman out: 7647af1364fSEric W. Biederman return is_covered; 7657af1364fSEric W. Biederman } 7667af1364fSEric W. Biederman 767e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 76884d17192SAl Viro { 7690818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 77084d17192SAl Viro struct mountpoint *mp; 77184d17192SAl Viro 7720818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 77384d17192SAl Viro if (mp->m_dentry == dentry) { 77484d17192SAl Viro /* might be worth a WARN_ON() */ 77584d17192SAl Viro if (d_unlinked(dentry)) 77684d17192SAl Viro return ERR_PTR(-ENOENT); 77784d17192SAl Viro mp->m_count++; 77884d17192SAl Viro return mp; 77984d17192SAl Viro } 78084d17192SAl Viro } 781e2dfa935SEric W. Biederman return NULL; 782e2dfa935SEric W. Biederman } 783e2dfa935SEric W. Biederman 7843895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 785e2dfa935SEric W. Biederman { 7863895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 787e2dfa935SEric W. Biederman int ret; 78884d17192SAl Viro 7893895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7903895dbf8SEric W. Biederman mountpoint: 7913895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7923895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7933895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7943895dbf8SEric W. Biederman if (mp) 7953895dbf8SEric W. Biederman goto done; 79684d17192SAl Viro } 797eed81007SMiklos Szeredi 7983895dbf8SEric W. Biederman if (!new) 7993895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 8003895dbf8SEric W. Biederman if (!new) 8013895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 8023895dbf8SEric W. Biederman 8033895dbf8SEric W. Biederman 8043895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 8053895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 8063895dbf8SEric W. Biederman 8073895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 8083895dbf8SEric W. Biederman if (ret == -EBUSY) 8093895dbf8SEric W. Biederman goto mountpoint; 8103895dbf8SEric W. Biederman 8113895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 8123895dbf8SEric W. Biederman mp = ERR_PTR(ret); 8133895dbf8SEric W. Biederman if (ret) 8143895dbf8SEric W. Biederman goto done; 8153895dbf8SEric W. Biederman 8163895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 8173895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 8183895dbf8SEric W. Biederman new->m_dentry = dentry; 8193895dbf8SEric W. Biederman new->m_count = 1; 8203895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 8213895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 8223895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 8233895dbf8SEric W. Biederman 8243895dbf8SEric W. Biederman mp = new; 8253895dbf8SEric W. Biederman new = NULL; 8263895dbf8SEric W. Biederman done: 8273895dbf8SEric W. Biederman kfree(new); 82884d17192SAl Viro return mp; 82984d17192SAl Viro } 83084d17192SAl Viro 83184d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 83284d17192SAl Viro { 83384d17192SAl Viro if (!--mp->m_count) { 83484d17192SAl Viro struct dentry *dentry = mp->m_dentry; 8350a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 83684d17192SAl Viro spin_lock(&dentry->d_lock); 83784d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 83884d17192SAl Viro spin_unlock(&dentry->d_lock); 8390818bf27SAl Viro hlist_del(&mp->m_hash); 84084d17192SAl Viro kfree(mp); 84184d17192SAl Viro } 84284d17192SAl Viro } 84384d17192SAl Viro 844143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8451da177e4SLinus Torvalds { 8466b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8471da177e4SLinus Torvalds } 8481da177e4SLinus Torvalds 84999b7db7bSNick Piggin /* 85099b7db7bSNick Piggin * vfsmount lock must be held for write 85199b7db7bSNick Piggin */ 8526b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8535addc5ddSAl Viro { 8545addc5ddSAl Viro if (ns) { 8555addc5ddSAl Viro ns->event = ++event; 8565addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8575addc5ddSAl Viro } 8585addc5ddSAl Viro } 8595addc5ddSAl Viro 86099b7db7bSNick Piggin /* 86199b7db7bSNick Piggin * vfsmount lock must be held for write 86299b7db7bSNick Piggin */ 8636b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8645addc5ddSAl Viro { 8655addc5ddSAl Viro if (ns && ns->event != event) { 8665addc5ddSAl Viro ns->event = event; 8675addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8685addc5ddSAl Viro } 8695addc5ddSAl Viro } 8705addc5ddSAl Viro 87199b7db7bSNick Piggin /* 87299b7db7bSNick Piggin * vfsmount lock must be held for write 87399b7db7bSNick Piggin */ 8747bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8751da177e4SLinus Torvalds { 8760714a533SAl Viro mnt->mnt_parent = mnt; 877a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8786b41d536SAl Viro list_del_init(&mnt->mnt_child); 87938129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8800a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 88184d17192SAl Viro put_mountpoint(mnt->mnt_mp); 88284d17192SAl Viro mnt->mnt_mp = NULL; 8831da177e4SLinus Torvalds } 8841da177e4SLinus Torvalds 88599b7db7bSNick Piggin /* 88699b7db7bSNick Piggin * vfsmount lock must be held for write 88799b7db7bSNick Piggin */ 8887bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8897bdb11deSEric W. Biederman { 8907bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8917bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8927bdb11deSEric W. Biederman unhash_mnt(mnt); 8937bdb11deSEric W. Biederman } 8947bdb11deSEric W. Biederman 8957bdb11deSEric W. Biederman /* 8967bdb11deSEric W. Biederman * vfsmount lock must be held for write 8977bdb11deSEric W. Biederman */ 8986a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8996a46c573SEric W. Biederman { 9006a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 9016a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 9026a46c573SEric W. Biederman unhash_mnt(mnt); 9036a46c573SEric W. Biederman } 9046a46c573SEric W. Biederman 9056a46c573SEric W. Biederman /* 9066a46c573SEric W. Biederman * vfsmount lock must be held for write 9076a46c573SEric W. Biederman */ 90884d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 90984d17192SAl Viro struct mountpoint *mp, 91044d964d6SAl Viro struct mount *child_mnt) 911b90fa9aeSRam Pai { 91284d17192SAl Viro mp->m_count++; 9133a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 91484d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 9153a2393d7SAl Viro child_mnt->mnt_parent = mnt; 91684d17192SAl Viro child_mnt->mnt_mp = mp; 9170a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 918b90fa9aeSRam Pai } 919b90fa9aeSRam Pai 9201064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 9211064f874SEric W. Biederman { 9221064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 9231064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 9241064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 9251064f874SEric W. Biederman } 9261064f874SEric W. Biederman 92799b7db7bSNick Piggin /* 92899b7db7bSNick Piggin * vfsmount lock must be held for write 92999b7db7bSNick Piggin */ 93084d17192SAl Viro static void attach_mnt(struct mount *mnt, 93184d17192SAl Viro struct mount *parent, 93284d17192SAl Viro struct mountpoint *mp) 9331da177e4SLinus Torvalds { 93484d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 9351064f874SEric W. Biederman __attach_mnt(mnt, parent); 936b90fa9aeSRam Pai } 937b90fa9aeSRam Pai 9381064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 93912a5b529SAl Viro { 9401064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 9411064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 9421064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 9431064f874SEric W. Biederman 9441064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 9451064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 9461064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 9471064f874SEric W. Biederman 9481064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 9491064f874SEric W. Biederman 9501064f874SEric W. Biederman put_mountpoint(old_mp); 9511064f874SEric W. Biederman 9521064f874SEric W. Biederman /* 9531064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 9541064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 9551064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 9561064f874SEric W. Biederman * to a mountpoint. 9571064f874SEric W. Biederman * 9581064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 9591064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 9601064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 9611064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 9621064f874SEric W. Biederman */ 9631064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 9641064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 9651064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 9661064f874SEric W. Biederman 9671064f874SEric W. Biederman mnt_add_count(old_parent, -1); 96812a5b529SAl Viro } 96912a5b529SAl Viro 970b90fa9aeSRam Pai /* 97199b7db7bSNick Piggin * vfsmount lock must be held for write 972b90fa9aeSRam Pai */ 9731064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 974b90fa9aeSRam Pai { 9750714a533SAl Viro struct mount *parent = mnt->mnt_parent; 97683adc753SAl Viro struct mount *m; 977b90fa9aeSRam Pai LIST_HEAD(head); 978143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 979b90fa9aeSRam Pai 9800714a533SAl Viro BUG_ON(parent == mnt); 981b90fa9aeSRam Pai 9821a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 983f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 984143c8c91SAl Viro m->mnt_ns = n; 985f03c6599SAl Viro 986b90fa9aeSRam Pai list_splice(&head, n->list.prev); 987b90fa9aeSRam Pai 988d2921684SEric W. Biederman n->mounts += n->pending_mounts; 989d2921684SEric W. Biederman n->pending_mounts = 0; 990d2921684SEric W. Biederman 9911064f874SEric W. Biederman __attach_mnt(mnt, parent); 9926b3286edSKirill Korotaev touch_mnt_namespace(n); 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds 995909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9961da177e4SLinus Torvalds { 9976b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9986b41d536SAl Viro if (next == &p->mnt_mounts) { 9991da177e4SLinus Torvalds while (1) { 1000909b0a88SAl Viro if (p == root) 10011da177e4SLinus Torvalds return NULL; 10026b41d536SAl Viro next = p->mnt_child.next; 10036b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 10041da177e4SLinus Torvalds break; 10050714a533SAl Viro p = p->mnt_parent; 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds } 10086b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 10091da177e4SLinus Torvalds } 10101da177e4SLinus Torvalds 1011315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 10129676f0c6SRam Pai { 10136b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 10146b41d536SAl Viro while (prev != &p->mnt_mounts) { 10156b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 10166b41d536SAl Viro prev = p->mnt_mounts.prev; 10179676f0c6SRam Pai } 10189676f0c6SRam Pai return p; 10199676f0c6SRam Pai } 10209676f0c6SRam Pai 10219d412a43SAl Viro struct vfsmount * 10229d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 10239d412a43SAl Viro { 1024b105e270SAl Viro struct mount *mnt; 10259d412a43SAl Viro struct dentry *root; 10269d412a43SAl Viro 10279d412a43SAl Viro if (!type) 10289d412a43SAl Viro return ERR_PTR(-ENODEV); 10299d412a43SAl Viro 10309d412a43SAl Viro mnt = alloc_vfsmnt(name); 10319d412a43SAl Viro if (!mnt) 10329d412a43SAl Viro return ERR_PTR(-ENOMEM); 10339d412a43SAl Viro 1034e462ec50SDavid Howells if (flags & SB_KERNMOUNT) 1035b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 10369d412a43SAl Viro 10379d412a43SAl Viro root = mount_fs(type, flags, name, data); 10389d412a43SAl Viro if (IS_ERR(root)) { 10398ffcb32eSDavid Howells mnt_free_id(mnt); 10409d412a43SAl Viro free_vfsmnt(mnt); 10419d412a43SAl Viro return ERR_CAST(root); 10429d412a43SAl Viro } 10439d412a43SAl Viro 1044b105e270SAl Viro mnt->mnt.mnt_root = root; 1045b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 1046a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10470714a533SAl Viro mnt->mnt_parent = mnt; 1048719ea2fbSAl Viro lock_mount_hash(); 104939f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 1050719ea2fbSAl Viro unlock_mount_hash(); 1051b105e270SAl Viro return &mnt->mnt; 10529d412a43SAl Viro } 10539d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10549d412a43SAl Viro 105593faccbbSEric W. Biederman struct vfsmount * 105693faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 105793faccbbSEric W. Biederman const char *name, void *data) 105893faccbbSEric W. Biederman { 105993faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 106093faccbbSEric W. Biederman * through from the parent mount to the submount don't support 106193faccbbSEric W. Biederman * unprivileged mounts with submounts. 106293faccbbSEric W. Biederman */ 106393faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 106493faccbbSEric W. Biederman return ERR_PTR(-EPERM); 106593faccbbSEric W. Biederman 1066e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 106793faccbbSEric W. Biederman } 106893faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 106993faccbbSEric W. Biederman 107087129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 107136341f64SRam Pai int flag) 10721da177e4SLinus Torvalds { 107387129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1074be34d1a3SDavid Howells struct mount *mnt; 1075be34d1a3SDavid Howells int err; 10761da177e4SLinus Torvalds 1077be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1078be34d1a3SDavid Howells if (!mnt) 1079be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1080be34d1a3SDavid Howells 10817a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 108215169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1083719f5d7fSMiklos Szeredi else 108415169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1085719f5d7fSMiklos Szeredi 108615169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1087be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1088719f5d7fSMiklos Szeredi if (err) 1089719f5d7fSMiklos Szeredi goto out_free; 1090719f5d7fSMiklos Szeredi } 1091719f5d7fSMiklos Szeredi 109216a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 109316a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 1094132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 10959566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 10969566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 10979566d674SEric W. Biederman 10989566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 1099132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 1100132c94e3SEric W. Biederman 11019566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 11029566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 11039566d674SEric W. Biederman 11049566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 11059566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 11069566d674SEric W. Biederman 11079566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 11089566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 11099566d674SEric W. Biederman } 11109566d674SEric W. Biederman 11115ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1112381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1113381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 11145ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 11155ff9d8a6SEric W. Biederman 11161da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1117b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1118b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1119a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11200714a533SAl Viro mnt->mnt_parent = mnt; 1121719ea2fbSAl Viro lock_mount_hash(); 112239f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1123719ea2fbSAl Viro unlock_mount_hash(); 1124b90fa9aeSRam Pai 11257a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 11267a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 11276776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 112832301920SAl Viro mnt->mnt_master = old; 1129fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 11308aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1131fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 11326776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1133d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 11346776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1135d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 11365235d448SAl Viro } else { 11375235d448SAl Viro CLEAR_MNT_SHARED(mnt); 11385afe0022SRam Pai } 1139b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 11400f0afb1dSAl Viro set_mnt_shared(mnt); 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11431da177e4SLinus Torvalds * as the original if that was on one */ 114436341f64SRam Pai if (flag & CL_EXPIRE) { 11456776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11466776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11471da177e4SLinus Torvalds } 1148be34d1a3SDavid Howells 1149cb338d06SAl Viro return mnt; 1150719f5d7fSMiklos Szeredi 1151719f5d7fSMiklos Szeredi out_free: 11528ffcb32eSDavid Howells mnt_free_id(mnt); 1153719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1154be34d1a3SDavid Howells return ERR_PTR(err); 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds 11579ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11589ea459e1SAl Viro { 11599ea459e1SAl Viro /* 11609ea459e1SAl Viro * This probably indicates that somebody messed 11619ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 11629ea459e1SAl Viro * happens, the filesystem was probably unable 11639ea459e1SAl Viro * to make r/w->r/o transitions. 11649ea459e1SAl Viro */ 11659ea459e1SAl Viro /* 11669ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11679ea459e1SAl Viro * so mnt_get_writers() below is safe. 11689ea459e1SAl Viro */ 11699ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11709ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11719ea459e1SAl Viro mnt_pin_kill(mnt); 11729ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11739ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11749ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11759ea459e1SAl Viro mnt_free_id(mnt); 11769ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11779ea459e1SAl Viro } 11789ea459e1SAl Viro 11799ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11809ea459e1SAl Viro { 11819ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11829ea459e1SAl Viro } 11839ea459e1SAl Viro 11849ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11859ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11869ea459e1SAl Viro { 11879ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 118829785735SByungchul Park struct mount *m, *t; 11899ea459e1SAl Viro 119029785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 119129785735SByungchul Park cleanup_mnt(m); 11929ea459e1SAl Viro } 11939ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11949ea459e1SAl Viro 1195900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11967b7b1aceSAl Viro { 119748a066e7SAl Viro rcu_read_lock(); 11989ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11999ea0a46cSAl Viro /* 12009ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 12019ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 12029ea0a46cSAl Viro * non-NULL, then there's a reference that won't 12039ea0a46cSAl Viro * be dropped until after an RCU delay done after 12049ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 12059ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 12069ea0a46cSAl Viro * we are dropping is not the final one. 12079ea0a46cSAl Viro */ 1208aa9c0e07SAl Viro mnt_add_count(mnt, -1); 120948a066e7SAl Viro rcu_read_unlock(); 121099b7db7bSNick Piggin return; 1211b3e19d92SNick Piggin } 1212719ea2fbSAl Viro lock_mount_hash(); 12139ea0a46cSAl Viro mnt_add_count(mnt, -1); 1214b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 121548a066e7SAl Viro rcu_read_unlock(); 1216719ea2fbSAl Viro unlock_mount_hash(); 121799b7db7bSNick Piggin return; 121899b7db7bSNick Piggin } 121948a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 122048a066e7SAl Viro rcu_read_unlock(); 122148a066e7SAl Viro unlock_mount_hash(); 122248a066e7SAl Viro return; 122348a066e7SAl Viro } 122448a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 122548a066e7SAl Viro rcu_read_unlock(); 1226962830dfSAndi Kleen 122739f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1228ce07d891SEric W. Biederman 1229ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1230ce07d891SEric W. Biederman struct mount *p, *tmp; 1231ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1232ce07d891SEric W. Biederman umount_mnt(p); 1233ce07d891SEric W. Biederman } 1234ce07d891SEric W. Biederman } 1235719ea2fbSAl Viro unlock_mount_hash(); 1236649a795aSAl Viro 12379ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12389ea459e1SAl Viro struct task_struct *task = current; 12399ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12409ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 12419ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 12429ea459e1SAl Viro return; 12439ea459e1SAl Viro } 12449ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12459ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12469ea459e1SAl Viro return; 12479ea459e1SAl Viro } 12489ea459e1SAl Viro cleanup_mnt(mnt); 1249b3e19d92SNick Piggin } 1250b3e19d92SNick Piggin 1251b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1252b3e19d92SNick Piggin { 1253b3e19d92SNick Piggin if (mnt) { 1254863d684fSAl Viro struct mount *m = real_mount(mnt); 1255b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1256863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1257863d684fSAl Viro m->mnt_expiry_mark = 0; 1258863d684fSAl Viro mntput_no_expire(m); 1259b3e19d92SNick Piggin } 1260b3e19d92SNick Piggin } 1261b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1262b3e19d92SNick Piggin 1263b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1264b3e19d92SNick Piggin { 1265b3e19d92SNick Piggin if (mnt) 126683adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1267b3e19d92SNick Piggin return mnt; 1268b3e19d92SNick Piggin } 1269b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1270b3e19d92SNick Piggin 1271c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1272c6609c0aSIan Kent * namespace. 1273c6609c0aSIan Kent * 1274c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1275c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1276c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1277c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1278c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1279c6609c0aSIan Kent * alone. 1280c6609c0aSIan Kent */ 1281c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1282c6609c0aSIan Kent { 1283c6609c0aSIan Kent unsigned seq; 1284c6609c0aSIan Kent bool res; 1285c6609c0aSIan Kent 1286c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1287c6609c0aSIan Kent return false; 1288c6609c0aSIan Kent 1289c6609c0aSIan Kent rcu_read_lock(); 1290c6609c0aSIan Kent do { 1291c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1292c6609c0aSIan Kent res = __path_is_mountpoint(path); 1293c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1294c6609c0aSIan Kent rcu_read_unlock(); 1295c6609c0aSIan Kent 1296c6609c0aSIan Kent return res; 1297c6609c0aSIan Kent } 1298c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1299c6609c0aSIan Kent 1300ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 13017b7b1aceSAl Viro { 13023064c356SAl Viro struct mount *p; 13033064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 13043064c356SAl Viro if (IS_ERR(p)) 13053064c356SAl Viro return ERR_CAST(p); 13063064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 13073064c356SAl Viro return &p->mnt; 13087b7b1aceSAl Viro } 13091da177e4SLinus Torvalds 1310a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 13110226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13121da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13131da177e4SLinus Torvalds { 1314ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13151da177e4SLinus Torvalds 1316390c6843SRam Pai down_read(&namespace_sem); 1317c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1318c7999c36SAl Viro void *v = p->cached_mount; 1319c7999c36SAl Viro if (*pos == p->cached_index) 1320c7999c36SAl Viro return v; 1321c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1322c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1323c7999c36SAl Viro return p->cached_mount = v; 1324c7999c36SAl Viro } 1325c7999c36SAl Viro } 1326c7999c36SAl Viro 1327c7999c36SAl Viro p->cached_event = p->ns->event; 1328c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1329c7999c36SAl Viro p->cached_index = *pos; 1330c7999c36SAl Viro return p->cached_mount; 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds 13331da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13341da177e4SLinus Torvalds { 1335ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1336b0765fb8SPavel Emelianov 1337c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1338c7999c36SAl Viro p->cached_index = *pos; 1339c7999c36SAl Viro return p->cached_mount; 13401da177e4SLinus Torvalds } 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13431da177e4SLinus Torvalds { 1344390c6843SRam Pai up_read(&namespace_sem); 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds 13470226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13489f5596afSAl Viro { 1349ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13501a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 13510226f492SAl Viro return p->show(m, &r->mnt); 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds 1354a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13551da177e4SLinus Torvalds .start = m_start, 13561da177e4SLinus Torvalds .next = m_next, 13571da177e4SLinus Torvalds .stop = m_stop, 13580226f492SAl Viro .show = m_show, 1359b4629fe2SChuck Lever }; 1360a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1361b4629fe2SChuck Lever 13621da177e4SLinus Torvalds /** 13631da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13641da177e4SLinus Torvalds * @mnt: root of mount tree 13651da177e4SLinus Torvalds * 13661da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13671da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13681da177e4SLinus Torvalds * busy. 13691da177e4SLinus Torvalds */ 1370909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13711da177e4SLinus Torvalds { 1372909b0a88SAl Viro struct mount *mnt = real_mount(m); 137336341f64SRam Pai int actual_refs = 0; 137436341f64SRam Pai int minimum_refs = 0; 1375315fc83eSAl Viro struct mount *p; 1376909b0a88SAl Viro BUG_ON(!m); 13771da177e4SLinus Torvalds 1378b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1379719ea2fbSAl Viro lock_mount_hash(); 1380909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 138183adc753SAl Viro actual_refs += mnt_get_count(p); 13821da177e4SLinus Torvalds minimum_refs += 2; 13831da177e4SLinus Torvalds } 1384719ea2fbSAl Viro unlock_mount_hash(); 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13871da177e4SLinus Torvalds return 0; 1388e3474a8eSIan Kent 1389e3474a8eSIan Kent return 1; 13901da177e4SLinus Torvalds } 13911da177e4SLinus Torvalds 13921da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13931da177e4SLinus Torvalds 13941da177e4SLinus Torvalds /** 13951da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13961da177e4SLinus Torvalds * @mnt: root of mount 13971da177e4SLinus Torvalds * 13981da177e4SLinus Torvalds * This is called to check if a mount point has any 13991da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14001da177e4SLinus Torvalds * mount has sub mounts this will return busy 14011da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14021da177e4SLinus Torvalds * 14031da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14041da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14051da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14061da177e4SLinus Torvalds */ 14071da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14081da177e4SLinus Torvalds { 1409e3474a8eSIan Kent int ret = 1; 14108ad08d8aSAl Viro down_read(&namespace_sem); 1411719ea2fbSAl Viro lock_mount_hash(); 14121ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1413e3474a8eSIan Kent ret = 0; 1414719ea2fbSAl Viro unlock_mount_hash(); 14158ad08d8aSAl Viro up_read(&namespace_sem); 1416a05964f3SRam Pai return ret; 14171da177e4SLinus Torvalds } 14181da177e4SLinus Torvalds 14191da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14201da177e4SLinus Torvalds 142138129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1422e3197d83SAl Viro 142397216be0SAl Viro static void namespace_unlock(void) 14241da177e4SLinus Torvalds { 1425a3b3c562SEric W. Biederman struct hlist_head head; 142697216be0SAl Viro 1427a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1428a3b3c562SEric W. Biederman 142997216be0SAl Viro up_write(&namespace_sem); 1430a3b3c562SEric W. Biederman 1431a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 143297216be0SAl Viro return; 143397216be0SAl Viro 143448a066e7SAl Viro synchronize_rcu(); 143548a066e7SAl Viro 143687b95ce0SAl Viro group_pin_kill(&head); 143770fbcdf4SRam Pai } 143870fbcdf4SRam Pai 143997216be0SAl Viro static inline void namespace_lock(void) 1440e3197d83SAl Viro { 144197216be0SAl Viro down_write(&namespace_sem); 1442e3197d83SAl Viro } 1443e3197d83SAl Viro 1444e819f152SEric W. Biederman enum umount_tree_flags { 1445e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1446e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1447e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1448e819f152SEric W. Biederman }; 1449f2d0a123SEric W. Biederman 1450f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1451f2d0a123SEric W. Biederman { 1452f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1453f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1454f2d0a123SEric W. Biederman return true; 1455f2d0a123SEric W. Biederman 1456f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1457f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1458f2d0a123SEric W. Biederman return true; 1459f2d0a123SEric W. Biederman 1460f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1461f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1462f2d0a123SEric W. Biederman * connected to mounted mounts. 1463f2d0a123SEric W. Biederman */ 1464f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1465f2d0a123SEric W. Biederman return true; 1466f2d0a123SEric W. Biederman 1467f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1468f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1469f2d0a123SEric W. Biederman return false; 1470f2d0a123SEric W. Biederman 1471f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1472f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1473f2d0a123SEric W. Biederman return false; 1474f2d0a123SEric W. Biederman 1475f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1476f2d0a123SEric W. Biederman return true; 1477f2d0a123SEric W. Biederman } 1478f2d0a123SEric W. Biederman 147999b7db7bSNick Piggin /* 148048a066e7SAl Viro * mount_lock must be held 148199b7db7bSNick Piggin * namespace_sem must be held for write 148299b7db7bSNick Piggin */ 1483e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 148470fbcdf4SRam Pai { 1485c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1486315fc83eSAl Viro struct mount *p; 148770fbcdf4SRam Pai 14885d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14895d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14905d88457eSEric W. Biederman 1491c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1492590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1493590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1494c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1495590ce4bcSEric W. Biederman } 1496c003b26fSEric W. Biederman 1497411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1498c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1499c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 150038129a13SAl Viro } 150170fbcdf4SRam Pai 1502c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1503e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15047b8a53fdSAl Viro propagate_umount(&tmp_list); 1505a05964f3SRam Pai 1506c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1507d2921684SEric W. Biederman struct mnt_namespace *ns; 1508ce07d891SEric W. Biederman bool disconnect; 1509c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15106776db3dSAl Viro list_del_init(&p->mnt_expire); 15111a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1512d2921684SEric W. Biederman ns = p->mnt_ns; 1513d2921684SEric W. Biederman if (ns) { 1514d2921684SEric W. Biederman ns->mounts--; 1515d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1516d2921684SEric W. Biederman } 151763d37a84SAl Viro p->mnt_ns = NULL; 1518e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 151948a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 152087b95ce0SAl Viro 1521f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1522ce07d891SEric W. Biederman 1523ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1524ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1525676da58dSAl Viro if (mnt_has_parent(p)) { 152681b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1527ce07d891SEric W. Biederman if (!disconnect) { 1528ce07d891SEric W. Biederman /* Don't forget about p */ 1529ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1530ce07d891SEric W. Biederman } else { 15316a46c573SEric W. Biederman umount_mnt(p); 15327c4b93d8SAl Viro } 1533ce07d891SEric W. Biederman } 15340f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 153538129a13SAl Viro } 15361da177e4SLinus Torvalds } 15371da177e4SLinus Torvalds 1538b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1539c35038beSAl Viro 15401ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15411da177e4SLinus Torvalds { 15421ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15431da177e4SLinus Torvalds int retval; 15441da177e4SLinus Torvalds 15451ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15461da177e4SLinus Torvalds if (retval) 15471da177e4SLinus Torvalds return retval; 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds /* 15501da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15511da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15521da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15531da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15541da177e4SLinus Torvalds */ 15551da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15561ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15571da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15581da177e4SLinus Torvalds return -EINVAL; 15591da177e4SLinus Torvalds 1560b3e19d92SNick Piggin /* 1561b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1562b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1563b3e19d92SNick Piggin */ 1564719ea2fbSAl Viro lock_mount_hash(); 156583adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1566719ea2fbSAl Viro unlock_mount_hash(); 15671da177e4SLinus Torvalds return -EBUSY; 1568b3e19d92SNick Piggin } 1569719ea2fbSAl Viro unlock_mount_hash(); 15701da177e4SLinus Torvalds 1571863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15721da177e4SLinus Torvalds return -EAGAIN; 15731da177e4SLinus Torvalds } 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds /* 15761da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15771da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15781da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15791da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15801da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15811da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15821da177e4SLinus Torvalds * about for the moment. 15831da177e4SLinus Torvalds */ 15841da177e4SLinus Torvalds 158542faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 158642faad99SAl Viro sb->s_op->umount_begin(sb); 158742faad99SAl Viro } 15881da177e4SLinus Torvalds 15891da177e4SLinus Torvalds /* 15901da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15911da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15921da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15931da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15941da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15951da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15961da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15971da177e4SLinus Torvalds */ 15981ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15991da177e4SLinus Torvalds /* 16001da177e4SLinus Torvalds * Special case for "unmounting" root ... 16011da177e4SLinus Torvalds * we just try to remount it readonly. 16021da177e4SLinus Torvalds */ 1603bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1604a1480dccSAndy Lutomirski return -EPERM; 16051da177e4SLinus Torvalds down_write(&sb->s_umount); 1606bc98a42cSDavid Howells if (!sb_rdonly(sb)) 1607e462ec50SDavid Howells retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); 16081da177e4SLinus Torvalds up_write(&sb->s_umount); 16091da177e4SLinus Torvalds return retval; 16101da177e4SLinus Torvalds } 16111da177e4SLinus Torvalds 161297216be0SAl Viro namespace_lock(); 1613719ea2fbSAl Viro lock_mount_hash(); 16145addc5ddSAl Viro event++; 16151da177e4SLinus Torvalds 161648a066e7SAl Viro if (flags & MNT_DETACH) { 161748a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1618e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 161948a066e7SAl Viro retval = 0; 162048a066e7SAl Viro } else { 1621b54b9be7SAl Viro shrink_submounts(mnt); 16221da177e4SLinus Torvalds retval = -EBUSY; 162348a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 16241a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1625e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16261da177e4SLinus Torvalds retval = 0; 16271da177e4SLinus Torvalds } 162848a066e7SAl Viro } 1629719ea2fbSAl Viro unlock_mount_hash(); 1630e3197d83SAl Viro namespace_unlock(); 16311da177e4SLinus Torvalds return retval; 16321da177e4SLinus Torvalds } 16331da177e4SLinus Torvalds 16341da177e4SLinus Torvalds /* 163580b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 163680b5dce8SEric W. Biederman * 163780b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 163880b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 163980b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 164080b5dce8SEric W. Biederman * leaking them. 164180b5dce8SEric W. Biederman * 164280b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 164380b5dce8SEric W. Biederman */ 164480b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 164580b5dce8SEric W. Biederman { 164680b5dce8SEric W. Biederman struct mountpoint *mp; 164780b5dce8SEric W. Biederman struct mount *mnt; 164880b5dce8SEric W. Biederman 164980b5dce8SEric W. Biederman namespace_lock(); 16503895dbf8SEric W. Biederman lock_mount_hash(); 165180b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1652f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 165380b5dce8SEric W. Biederman goto out_unlock; 165480b5dce8SEric W. Biederman 1655e06b933eSAndrey Ulanov event++; 165680b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 165780b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1658ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1659fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1660fe78fcc8SEric W. Biederman umount_mnt(mnt); 1661ce07d891SEric W. Biederman } 1662e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 166380b5dce8SEric W. Biederman } 166480b5dce8SEric W. Biederman put_mountpoint(mp); 166580b5dce8SEric W. Biederman out_unlock: 16663895dbf8SEric W. Biederman unlock_mount_hash(); 166780b5dce8SEric W. Biederman namespace_unlock(); 166880b5dce8SEric W. Biederman } 166980b5dce8SEric W. Biederman 167080b5dce8SEric W. Biederman /* 16719b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16729b40bc90SAl Viro */ 16739b40bc90SAl Viro static inline bool may_mount(void) 16749b40bc90SAl Viro { 16759b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16769b40bc90SAl Viro } 16779b40bc90SAl Viro 16789e8925b6SJeff Layton static inline bool may_mandlock(void) 16799e8925b6SJeff Layton { 16809e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16819e8925b6SJeff Layton return false; 16829e8925b6SJeff Layton #endif 168395ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16849e8925b6SJeff Layton } 16859e8925b6SJeff Layton 16869b40bc90SAl Viro /* 16871da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16881da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16891da177e4SLinus Torvalds * 16901da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16911da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16921da177e4SLinus Torvalds */ 16931da177e4SLinus Torvalds 16943a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16951da177e4SLinus Torvalds { 16962d8f3038SAl Viro struct path path; 1697900148dcSAl Viro struct mount *mnt; 16981da177e4SLinus Torvalds int retval; 1699db1f05bbSMiklos Szeredi int lookup_flags = 0; 17001da177e4SLinus Torvalds 1701db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1702db1f05bbSMiklos Szeredi return -EINVAL; 1703db1f05bbSMiklos Szeredi 17049b40bc90SAl Viro if (!may_mount()) 17059b40bc90SAl Viro return -EPERM; 17069b40bc90SAl Viro 1707db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1708db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1709db1f05bbSMiklos Szeredi 1710197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 17111da177e4SLinus Torvalds if (retval) 17121da177e4SLinus Torvalds goto out; 1713900148dcSAl Viro mnt = real_mount(path.mnt); 17141da177e4SLinus Torvalds retval = -EINVAL; 17152d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 17161da177e4SLinus Torvalds goto dput_and_out; 1717143c8c91SAl Viro if (!check_mnt(mnt)) 17181da177e4SLinus Torvalds goto dput_and_out; 17195ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 17205ff9d8a6SEric W. Biederman goto dput_and_out; 1721b2f5d4dcSEric W. Biederman retval = -EPERM; 1722b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1723b2f5d4dcSEric W. Biederman goto dput_and_out; 17241da177e4SLinus Torvalds 1725900148dcSAl Viro retval = do_umount(mnt, flags); 17261da177e4SLinus Torvalds dput_and_out: 1727429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 17282d8f3038SAl Viro dput(path.dentry); 1729900148dcSAl Viro mntput_no_expire(mnt); 17301da177e4SLinus Torvalds out: 17311da177e4SLinus Torvalds return retval; 17321da177e4SLinus Torvalds } 17331da177e4SLinus Torvalds 17343a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17353a18ef5cSDominik Brodowski { 17363a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17373a18ef5cSDominik Brodowski } 17383a18ef5cSDominik Brodowski 17391da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17401da177e4SLinus Torvalds 17411da177e4SLinus Torvalds /* 17421da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17431da177e4SLinus Torvalds */ 1744bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17451da177e4SLinus Torvalds { 17463a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17471da177e4SLinus Torvalds } 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds #endif 17501da177e4SLinus Torvalds 17514ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17528823c079SEric W. Biederman { 17534ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1754e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1755e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17564ce5d2b1SEric W. Biederman } 17574ce5d2b1SEric W. Biederman 175858be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 175958be2825SAl Viro { 176058be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 176158be2825SAl Viro } 176258be2825SAl Viro 17634ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17644ce5d2b1SEric W. Biederman { 17654ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17664ce5d2b1SEric W. Biederman * mount namespace loop? 17674ce5d2b1SEric W. Biederman */ 17684ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17694ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17704ce5d2b1SEric W. Biederman return false; 17714ce5d2b1SEric W. Biederman 1772f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17738823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17748823c079SEric W. Biederman } 17758823c079SEric W. Biederman 177687129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 177736341f64SRam Pai int flag) 17781da177e4SLinus Torvalds { 177984d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17801da177e4SLinus Torvalds 17814ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17824ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17834ce5d2b1SEric W. Biederman 17844ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1785be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17869676f0c6SRam Pai 178736341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1788be34d1a3SDavid Howells if (IS_ERR(q)) 1789be34d1a3SDavid Howells return q; 1790be34d1a3SDavid Howells 1791a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds p = mnt; 17946b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1795315fc83eSAl Viro struct mount *s; 17967ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17971da177e4SLinus Torvalds continue; 17981da177e4SLinus Torvalds 1799909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18004ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18014ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 18024ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18034ce5d2b1SEric W. Biederman continue; 18044ce5d2b1SEric W. Biederman } 18054ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 18064ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 18079676f0c6SRam Pai s = skip_mnt_tree(s); 18089676f0c6SRam Pai continue; 18099676f0c6SRam Pai } 18100714a533SAl Viro while (p != s->mnt_parent) { 18110714a533SAl Viro p = p->mnt_parent; 18120714a533SAl Viro q = q->mnt_parent; 18131da177e4SLinus Torvalds } 181487129cc0SAl Viro p = s; 181584d17192SAl Viro parent = q; 181687129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1817be34d1a3SDavid Howells if (IS_ERR(q)) 1818be34d1a3SDavid Howells goto out; 1819719ea2fbSAl Viro lock_mount_hash(); 18201a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18211064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1822719ea2fbSAl Viro unlock_mount_hash(); 18231da177e4SLinus Torvalds } 18241da177e4SLinus Torvalds } 18251da177e4SLinus Torvalds return res; 1826be34d1a3SDavid Howells out: 18271da177e4SLinus Torvalds if (res) { 1828719ea2fbSAl Viro lock_mount_hash(); 1829e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1830719ea2fbSAl Viro unlock_mount_hash(); 18311da177e4SLinus Torvalds } 1832be34d1a3SDavid Howells return q; 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds 1835be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1836be34d1a3SDavid Howells 1837ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18388aec0809SAl Viro { 1839cb338d06SAl Viro struct mount *tree; 184097216be0SAl Viro namespace_lock(); 1841cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1842cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1843cd4a4017SEric W. Biederman else 184487129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 184587129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1846328e6d90SAl Viro namespace_unlock(); 1847be34d1a3SDavid Howells if (IS_ERR(tree)) 184852e220d3SDan Carpenter return ERR_CAST(tree); 1849be34d1a3SDavid Howells return &tree->mnt; 18508aec0809SAl Viro } 18518aec0809SAl Viro 18528aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 18538aec0809SAl Viro { 185497216be0SAl Viro namespace_lock(); 1855719ea2fbSAl Viro lock_mount_hash(); 1856e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1857719ea2fbSAl Viro unlock_mount_hash(); 18583ab6abeeSAl Viro namespace_unlock(); 18598aec0809SAl Viro } 18608aec0809SAl Viro 1861c771d683SMiklos Szeredi /** 1862c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1863c771d683SMiklos Szeredi * 1864c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1865c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1866c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1867c771d683SMiklos Szeredi * 1868c771d683SMiklos Szeredi * Release with mntput(). 1869c771d683SMiklos Szeredi */ 1870ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1871c771d683SMiklos Szeredi { 1872c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1873c771d683SMiklos Szeredi struct mount *new_mnt; 1874c771d683SMiklos Szeredi 1875c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1876c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1877c771d683SMiklos Szeredi 1878c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1879c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1880c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1881c771d683SMiklos Szeredi 1882c771d683SMiklos Szeredi return &new_mnt->mnt; 1883c771d683SMiklos Szeredi } 1884c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1885c771d683SMiklos Szeredi 18861f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18871f707137SAl Viro struct vfsmount *root) 18881f707137SAl Viro { 18891a4eeaf2SAl Viro struct mount *mnt; 18901f707137SAl Viro int res = f(root, arg); 18911f707137SAl Viro if (res) 18921f707137SAl Viro return res; 18931a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18941a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18951f707137SAl Viro if (res) 18961f707137SAl Viro return res; 18971f707137SAl Viro } 18981f707137SAl Viro return 0; 18991f707137SAl Viro } 19001f707137SAl Viro 19014b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1902719f5d7fSMiklos Szeredi { 1903315fc83eSAl Viro struct mount *p; 1904719f5d7fSMiklos Szeredi 1905909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1906fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 19074b8b21f4SAl Viro mnt_release_group_id(p); 1908719f5d7fSMiklos Szeredi } 1909719f5d7fSMiklos Szeredi } 1910719f5d7fSMiklos Szeredi 19114b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1912719f5d7fSMiklos Szeredi { 1913315fc83eSAl Viro struct mount *p; 1914719f5d7fSMiklos Szeredi 1915909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1916fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19174b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1918719f5d7fSMiklos Szeredi if (err) { 19194b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1920719f5d7fSMiklos Szeredi return err; 1921719f5d7fSMiklos Szeredi } 1922719f5d7fSMiklos Szeredi } 1923719f5d7fSMiklos Szeredi } 1924719f5d7fSMiklos Szeredi 1925719f5d7fSMiklos Szeredi return 0; 1926719f5d7fSMiklos Szeredi } 1927719f5d7fSMiklos Szeredi 1928d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1929d2921684SEric W. Biederman { 1930d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1931d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1932d2921684SEric W. Biederman struct mount *p; 1933d2921684SEric W. Biederman 1934d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1935d2921684SEric W. Biederman mounts++; 1936d2921684SEric W. Biederman 1937d2921684SEric W. Biederman old = ns->mounts; 1938d2921684SEric W. Biederman pending = ns->pending_mounts; 1939d2921684SEric W. Biederman sum = old + pending; 1940d2921684SEric W. Biederman if ((old > sum) || 1941d2921684SEric W. Biederman (pending > sum) || 1942d2921684SEric W. Biederman (max < sum) || 1943d2921684SEric W. Biederman (mounts > (max - sum))) 1944d2921684SEric W. Biederman return -ENOSPC; 1945d2921684SEric W. Biederman 1946d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1947d2921684SEric W. Biederman return 0; 1948d2921684SEric W. Biederman } 1949d2921684SEric W. Biederman 1950b90fa9aeSRam Pai /* 1951b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1952b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 195321444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 195421444403SRam Pai * store the parent mount and mountpoint dentry. 195521444403SRam Pai * (done when source_mnt is moved) 1956b90fa9aeSRam Pai * 1957b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1958b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19599676f0c6SRam Pai * --------------------------------------------------------------------------- 1960b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19619676f0c6SRam Pai * |************************************************************************** 19629676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19639676f0c6SRam Pai * | dest | | | | | 19649676f0c6SRam Pai * | | | | | | | 19659676f0c6SRam Pai * | v | | | | | 19669676f0c6SRam Pai * |************************************************************************** 19679676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19685afe0022SRam Pai * | | | | | | 19699676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19709676f0c6SRam Pai * *************************************************************************** 1971b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1972b90fa9aeSRam Pai * destination mount. 1973b90fa9aeSRam Pai * 1974b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1975b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1976b90fa9aeSRam Pai * the peer group of the source mount. 1977b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1978b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1979b90fa9aeSRam Pai * mount. 19805afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19815afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19825afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19835afe0022SRam Pai * is marked as 'shared and slave'. 19845afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19855afe0022SRam Pai * source mount. 19865afe0022SRam Pai * 19879676f0c6SRam Pai * --------------------------------------------------------------------------- 198821444403SRam Pai * | MOVE MOUNT OPERATION | 19899676f0c6SRam Pai * |************************************************************************** 19909676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19919676f0c6SRam Pai * | dest | | | | | 19929676f0c6SRam Pai * | | | | | | | 19939676f0c6SRam Pai * | v | | | | | 19949676f0c6SRam Pai * |************************************************************************** 19959676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19965afe0022SRam Pai * | | | | | | 19979676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19989676f0c6SRam Pai * *************************************************************************** 19995afe0022SRam Pai * 20005afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 20015afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 200221444403SRam Pai * (+*) the mount is moved to the destination. 20035afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 20045afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 20055afe0022SRam Pai * the mount is marked as 'shared and slave'. 20065afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2007b90fa9aeSRam Pai * 2008b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2009b90fa9aeSRam Pai * applied to each mount in the tree. 2010b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2011b90fa9aeSRam Pai * in allocations. 2012b90fa9aeSRam Pai */ 20130fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 201484d17192SAl Viro struct mount *dest_mnt, 201584d17192SAl Viro struct mountpoint *dest_mp, 201684d17192SAl Viro struct path *parent_path) 2017b90fa9aeSRam Pai { 201838129a13SAl Viro HLIST_HEAD(tree_list); 2019d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 20201064f874SEric W. Biederman struct mountpoint *smp; 2021315fc83eSAl Viro struct mount *child, *p; 202238129a13SAl Viro struct hlist_node *n; 2023719f5d7fSMiklos Szeredi int err; 2024b90fa9aeSRam Pai 20251064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 20261064f874SEric W. Biederman * to be tucked under other mounts. 20271064f874SEric W. Biederman */ 20281064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 20291064f874SEric W. Biederman if (IS_ERR(smp)) 20301064f874SEric W. Biederman return PTR_ERR(smp); 20311064f874SEric W. Biederman 2032d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 2033d2921684SEric W. Biederman if (!parent_path) { 2034d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2035d2921684SEric W. Biederman if (err) 2036d2921684SEric W. Biederman goto out; 2037d2921684SEric W. Biederman } 2038d2921684SEric W. Biederman 2039fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 20400fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2041719f5d7fSMiklos Szeredi if (err) 2042719f5d7fSMiklos Szeredi goto out; 204384d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2044f2ebb3a9SAl Viro lock_mount_hash(); 2045719f5d7fSMiklos Szeredi if (err) 2046719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2047909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 20480f0afb1dSAl Viro set_mnt_shared(p); 20490b1b901bSAl Viro } else { 20500b1b901bSAl Viro lock_mount_hash(); 2051b90fa9aeSRam Pai } 20521a390689SAl Viro if (parent_path) { 20530fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 205484d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2055143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 205621444403SRam Pai } else { 205784d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20581064f874SEric W. Biederman commit_tree(source_mnt); 205921444403SRam Pai } 2060b90fa9aeSRam Pai 206138129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20621d6a32acSAl Viro struct mount *q; 206338129a13SAl Viro hlist_del_init(&child->mnt_hash); 20641064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20651d6a32acSAl Viro child->mnt_mountpoint); 20661064f874SEric W. Biederman if (q) 20671064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20681064f874SEric W. Biederman commit_tree(child); 2069b90fa9aeSRam Pai } 20701064f874SEric W. Biederman put_mountpoint(smp); 2071719ea2fbSAl Viro unlock_mount_hash(); 207299b7db7bSNick Piggin 2073b90fa9aeSRam Pai return 0; 2074719f5d7fSMiklos Szeredi 2075719f5d7fSMiklos Szeredi out_cleanup_ids: 2076f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2077f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2078d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2079e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2080f2ebb3a9SAl Viro } 2081f2ebb3a9SAl Viro unlock_mount_hash(); 20820fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2083719f5d7fSMiklos Szeredi out: 2084d2921684SEric W. Biederman ns->pending_mounts = 0; 20851064f874SEric W. Biederman 20861064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 20871064f874SEric W. Biederman put_mountpoint(smp); 20881064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 20891064f874SEric W. Biederman 2090719f5d7fSMiklos Szeredi return err; 2091b90fa9aeSRam Pai } 2092b90fa9aeSRam Pai 209384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2094b12cea91SAl Viro { 2095b12cea91SAl Viro struct vfsmount *mnt; 209684d17192SAl Viro struct dentry *dentry = path->dentry; 2097b12cea91SAl Viro retry: 20985955102cSAl Viro inode_lock(dentry->d_inode); 209984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 21005955102cSAl Viro inode_unlock(dentry->d_inode); 210184d17192SAl Viro return ERR_PTR(-ENOENT); 2102b12cea91SAl Viro } 210397216be0SAl Viro namespace_lock(); 2104b12cea91SAl Viro mnt = lookup_mnt(path); 210584d17192SAl Viro if (likely(!mnt)) { 21063895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 210784d17192SAl Viro if (IS_ERR(mp)) { 210897216be0SAl Viro namespace_unlock(); 21095955102cSAl Viro inode_unlock(dentry->d_inode); 211084d17192SAl Viro return mp; 211184d17192SAl Viro } 211284d17192SAl Viro return mp; 211384d17192SAl Viro } 211497216be0SAl Viro namespace_unlock(); 21155955102cSAl Viro inode_unlock(path->dentry->d_inode); 2116b12cea91SAl Viro path_put(path); 2117b12cea91SAl Viro path->mnt = mnt; 211884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2119b12cea91SAl Viro goto retry; 2120b12cea91SAl Viro } 2121b12cea91SAl Viro 212284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2123b12cea91SAl Viro { 212484d17192SAl Viro struct dentry *dentry = where->m_dentry; 21253895dbf8SEric W. Biederman 21263895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 212784d17192SAl Viro put_mountpoint(where); 21283895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 21293895dbf8SEric W. Biederman 2130328e6d90SAl Viro namespace_unlock(); 21315955102cSAl Viro inode_unlock(dentry->d_inode); 2132b12cea91SAl Viro } 2133b12cea91SAl Viro 213484d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 21351da177e4SLinus Torvalds { 2136e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 21371da177e4SLinus Torvalds return -EINVAL; 21381da177e4SLinus Torvalds 2139e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2140e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 21411da177e4SLinus Torvalds return -ENOTDIR; 21421da177e4SLinus Torvalds 214384d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 21441da177e4SLinus Torvalds } 21451da177e4SLinus Torvalds 21461da177e4SLinus Torvalds /* 21477a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 21487a2e8a8fSValerie Aurora */ 21497a2e8a8fSValerie Aurora 2150e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 21517a2e8a8fSValerie Aurora { 2152e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 21537a2e8a8fSValerie Aurora 21547a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21557a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21567a2e8a8fSValerie Aurora return 0; 21577a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21587a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21597a2e8a8fSValerie Aurora return 0; 21607a2e8a8fSValerie Aurora return type; 21617a2e8a8fSValerie Aurora } 21627a2e8a8fSValerie Aurora 21637a2e8a8fSValerie Aurora /* 216407b20889SRam Pai * recursively change the type of the mountpoint. 216507b20889SRam Pai */ 2166e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 216707b20889SRam Pai { 2168315fc83eSAl Viro struct mount *m; 21694b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2170e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 21717a2e8a8fSValerie Aurora int type; 2172719f5d7fSMiklos Szeredi int err = 0; 217307b20889SRam Pai 21742d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 217507b20889SRam Pai return -EINVAL; 217607b20889SRam Pai 2177e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 21787a2e8a8fSValerie Aurora if (!type) 21797a2e8a8fSValerie Aurora return -EINVAL; 21807a2e8a8fSValerie Aurora 218197216be0SAl Viro namespace_lock(); 2182719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2183719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2184719f5d7fSMiklos Szeredi if (err) 2185719f5d7fSMiklos Szeredi goto out_unlock; 2186719f5d7fSMiklos Szeredi } 2187719f5d7fSMiklos Szeredi 2188719ea2fbSAl Viro lock_mount_hash(); 2189909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 21900f0afb1dSAl Viro change_mnt_propagation(m, type); 2191719ea2fbSAl Viro unlock_mount_hash(); 2192719f5d7fSMiklos Szeredi 2193719f5d7fSMiklos Szeredi out_unlock: 219497216be0SAl Viro namespace_unlock(); 2195719f5d7fSMiklos Szeredi return err; 219607b20889SRam Pai } 219707b20889SRam Pai 21985ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21995ff9d8a6SEric W. Biederman { 22005ff9d8a6SEric W. Biederman struct mount *child; 22015ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 22025ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 22035ff9d8a6SEric W. Biederman continue; 22045ff9d8a6SEric W. Biederman 22055ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 22065ff9d8a6SEric W. Biederman return true; 22075ff9d8a6SEric W. Biederman } 22085ff9d8a6SEric W. Biederman return false; 22095ff9d8a6SEric W. Biederman } 22105ff9d8a6SEric W. Biederman 221107b20889SRam Pai /* 22121da177e4SLinus Torvalds * do loopback mount. 22131da177e4SLinus Torvalds */ 2214808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 22152dafe1c4SEric Sandeen int recurse) 22161da177e4SLinus Torvalds { 22172d92ab3cSAl Viro struct path old_path; 221884d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 221984d17192SAl Viro struct mountpoint *mp; 222057eccb83SAl Viro int err; 22211da177e4SLinus Torvalds if (!old_name || !*old_name) 22221da177e4SLinus Torvalds return -EINVAL; 2223815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 22241da177e4SLinus Torvalds if (err) 22251da177e4SLinus Torvalds return err; 22261da177e4SLinus Torvalds 22278823c079SEric W. Biederman err = -EINVAL; 22284ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 22298823c079SEric W. Biederman goto out; 22308823c079SEric W. Biederman 223184d17192SAl Viro mp = lock_mount(path); 223284d17192SAl Viro err = PTR_ERR(mp); 223384d17192SAl Viro if (IS_ERR(mp)) 22349676f0c6SRam Pai goto out; 22359676f0c6SRam Pai 223687129cc0SAl Viro old = real_mount(old_path.mnt); 223784d17192SAl Viro parent = real_mount(path->mnt); 223887129cc0SAl Viro 2239b12cea91SAl Viro err = -EINVAL; 2240fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2241b12cea91SAl Viro goto out2; 2242b12cea91SAl Viro 2243e149ed2bSAl Viro if (!check_mnt(parent)) 2244e149ed2bSAl Viro goto out2; 2245e149ed2bSAl Viro 2246e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2247b12cea91SAl Viro goto out2; 2248ccd48bc7SAl Viro 22495ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 22505ff9d8a6SEric W. Biederman goto out2; 22515ff9d8a6SEric W. Biederman 22521da177e4SLinus Torvalds if (recurse) 22534ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 22541da177e4SLinus Torvalds else 225587129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 22561da177e4SLinus Torvalds 2257be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2258be34d1a3SDavid Howells err = PTR_ERR(mnt); 2259e9c5d8a5SAndrey Vagin goto out2; 2260be34d1a3SDavid Howells } 2261ccd48bc7SAl Viro 22625ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 22635ff9d8a6SEric W. Biederman 226484d17192SAl Viro err = graft_tree(mnt, parent, mp); 22651da177e4SLinus Torvalds if (err) { 2266719ea2fbSAl Viro lock_mount_hash(); 2267e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2268719ea2fbSAl Viro unlock_mount_hash(); 22695b83d2c5SRam Pai } 2270b12cea91SAl Viro out2: 227184d17192SAl Viro unlock_mount(mp); 2272ccd48bc7SAl Viro out: 22732d92ab3cSAl Viro path_put(&old_path); 22741da177e4SLinus Torvalds return err; 22751da177e4SLinus Torvalds } 22761da177e4SLinus Torvalds 22772e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 22782e4b7fcdSDave Hansen { 22792e4b7fcdSDave Hansen int error = 0; 22802e4b7fcdSDave Hansen int readonly_request = 0; 22812e4b7fcdSDave Hansen 22822e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 22832e4b7fcdSDave Hansen readonly_request = 1; 22842e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 22852e4b7fcdSDave Hansen return 0; 22862e4b7fcdSDave Hansen 22872e4b7fcdSDave Hansen if (readonly_request) 228883adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 22892e4b7fcdSDave Hansen else 229083adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 22912e4b7fcdSDave Hansen return error; 22922e4b7fcdSDave Hansen } 22932e4b7fcdSDave Hansen 22941da177e4SLinus Torvalds /* 22951da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 22961da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 22971da177e4SLinus Torvalds * on it - tough luck. 22981da177e4SLinus Torvalds */ 2299e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2300e462ec50SDavid Howells int mnt_flags, void *data) 23011da177e4SLinus Torvalds { 23021da177e4SLinus Torvalds int err; 23032d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2304143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 23051da177e4SLinus Torvalds 2306143c8c91SAl Viro if (!check_mnt(mnt)) 23071da177e4SLinus Torvalds return -EINVAL; 23081da177e4SLinus Torvalds 23092d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 23101da177e4SLinus Torvalds return -EINVAL; 23111da177e4SLinus Torvalds 231207b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 231307b64558SEric W. Biederman * 231407b64558SEric W. Biederman * No locks need to be held here while testing the various 231507b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 231607b64558SEric W. Biederman * once they are set. 231707b64558SEric W. Biederman */ 231807b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 231907b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 232007b64558SEric W. Biederman return -EPERM; 232107b64558SEric W. Biederman } 23229566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 23239566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 23249566d674SEric W. Biederman return -EPERM; 23259566d674SEric W. Biederman } 23269566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 23279566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 23289566d674SEric W. Biederman return -EPERM; 23299566d674SEric W. Biederman } 23309566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 23319566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 23329566d674SEric W. Biederman return -EPERM; 23339566d674SEric W. Biederman } 23349566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 23359566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 23369566d674SEric W. Biederman return -EPERM; 23379566d674SEric W. Biederman } 23389566d674SEric W. Biederman 2339ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2340ff36fe2cSEric Paris if (err) 2341ff36fe2cSEric Paris return err; 2342ff36fe2cSEric Paris 23431da177e4SLinus Torvalds down_write(&sb->s_umount); 2344e462ec50SDavid Howells if (ms_flags & MS_BIND) 2345e462ec50SDavid Howells err = change_mount_flags(path->mnt, ms_flags); 2346bc6155d1SEric W. Biederman else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 234757eccb83SAl Viro err = -EPERM; 23484aa98cf7SAl Viro else 2349e462ec50SDavid Howells err = do_remount_sb(sb, sb_flags, data, 0); 23507b43a79fSAl Viro if (!err) { 2351719ea2fbSAl Viro lock_mount_hash(); 2352a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2353143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2354143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2355719ea2fbSAl Viro unlock_mount_hash(); 23560e55a7ccSDan Williams } 23576339dab8SAl Viro up_write(&sb->s_umount); 23581da177e4SLinus Torvalds return err; 23591da177e4SLinus Torvalds } 23601da177e4SLinus Torvalds 2361cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 23629676f0c6SRam Pai { 2363315fc83eSAl Viro struct mount *p; 2364909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2365fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 23669676f0c6SRam Pai return 1; 23679676f0c6SRam Pai } 23689676f0c6SRam Pai return 0; 23699676f0c6SRam Pai } 23709676f0c6SRam Pai 2371808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 23721da177e4SLinus Torvalds { 23732d92ab3cSAl Viro struct path old_path, parent_path; 2374676da58dSAl Viro struct mount *p; 23750fb54e50SAl Viro struct mount *old; 237684d17192SAl Viro struct mountpoint *mp; 237757eccb83SAl Viro int err; 23781da177e4SLinus Torvalds if (!old_name || !*old_name) 23791da177e4SLinus Torvalds return -EINVAL; 23802d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 23811da177e4SLinus Torvalds if (err) 23821da177e4SLinus Torvalds return err; 23831da177e4SLinus Torvalds 238484d17192SAl Viro mp = lock_mount(path); 238584d17192SAl Viro err = PTR_ERR(mp); 238684d17192SAl Viro if (IS_ERR(mp)) 2387cc53ce53SDavid Howells goto out; 2388cc53ce53SDavid Howells 2389143c8c91SAl Viro old = real_mount(old_path.mnt); 2390fc7be130SAl Viro p = real_mount(path->mnt); 2391143c8c91SAl Viro 23921da177e4SLinus Torvalds err = -EINVAL; 2393fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 23941da177e4SLinus Torvalds goto out1; 23951da177e4SLinus Torvalds 23965ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23975ff9d8a6SEric W. Biederman goto out1; 23985ff9d8a6SEric W. Biederman 23991da177e4SLinus Torvalds err = -EINVAL; 24002d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 240121444403SRam Pai goto out1; 24021da177e4SLinus Torvalds 2403676da58dSAl Viro if (!mnt_has_parent(old)) 240421444403SRam Pai goto out1; 24051da177e4SLinus Torvalds 2406e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2407e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 240821444403SRam Pai goto out1; 240921444403SRam Pai /* 241021444403SRam Pai * Don't move a mount residing in a shared parent. 241121444403SRam Pai */ 2412fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 241321444403SRam Pai goto out1; 24149676f0c6SRam Pai /* 24159676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 24169676f0c6SRam Pai * mount which is shared. 24179676f0c6SRam Pai */ 2418fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 24199676f0c6SRam Pai goto out1; 24201da177e4SLinus Torvalds err = -ELOOP; 2421fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2422676da58dSAl Viro if (p == old) 242321444403SRam Pai goto out1; 24241da177e4SLinus Torvalds 242584d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 24264ac91378SJan Blunck if (err) 242721444403SRam Pai goto out1; 24281da177e4SLinus Torvalds 24291da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 24301da177e4SLinus Torvalds * automatically */ 24316776db3dSAl Viro list_del_init(&old->mnt_expire); 24321da177e4SLinus Torvalds out1: 243384d17192SAl Viro unlock_mount(mp); 24341da177e4SLinus Torvalds out: 24351da177e4SLinus Torvalds if (!err) 24361a390689SAl Viro path_put(&parent_path); 24372d92ab3cSAl Viro path_put(&old_path); 24381da177e4SLinus Torvalds return err; 24391da177e4SLinus Torvalds } 24401da177e4SLinus Torvalds 24419d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 24429d412a43SAl Viro { 24439d412a43SAl Viro int err; 24449d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 24459d412a43SAl Viro if (subtype) { 24469d412a43SAl Viro subtype++; 24479d412a43SAl Viro err = -EINVAL; 24489d412a43SAl Viro if (!subtype[0]) 24499d412a43SAl Viro goto err; 24509d412a43SAl Viro } else 24519d412a43SAl Viro subtype = ""; 24529d412a43SAl Viro 24539d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 24549d412a43SAl Viro err = -ENOMEM; 24559d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 24569d412a43SAl Viro goto err; 24579d412a43SAl Viro return mnt; 24589d412a43SAl Viro 24599d412a43SAl Viro err: 24609d412a43SAl Viro mntput(mnt); 24619d412a43SAl Viro return ERR_PTR(err); 24629d412a43SAl Viro } 24639d412a43SAl Viro 24649d412a43SAl Viro /* 24659d412a43SAl Viro * add a mount into a namespace's mount tree 24669d412a43SAl Viro */ 246795bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 24689d412a43SAl Viro { 246984d17192SAl Viro struct mountpoint *mp; 247084d17192SAl Viro struct mount *parent; 24719d412a43SAl Viro int err; 24729d412a43SAl Viro 2473f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 24749d412a43SAl Viro 247584d17192SAl Viro mp = lock_mount(path); 247684d17192SAl Viro if (IS_ERR(mp)) 247784d17192SAl Viro return PTR_ERR(mp); 24789d412a43SAl Viro 247984d17192SAl Viro parent = real_mount(path->mnt); 24809d412a43SAl Viro err = -EINVAL; 248184d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2482156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2483156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 24849d412a43SAl Viro goto unlock; 2485156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 248684d17192SAl Viro if (!parent->mnt_ns) 2487156cacb1SAl Viro goto unlock; 2488156cacb1SAl Viro } 24899d412a43SAl Viro 24909d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 24919d412a43SAl Viro err = -EBUSY; 249295bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 24939d412a43SAl Viro path->mnt->mnt_root == path->dentry) 24949d412a43SAl Viro goto unlock; 24959d412a43SAl Viro 24969d412a43SAl Viro err = -EINVAL; 2497e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24989d412a43SAl Viro goto unlock; 24999d412a43SAl Viro 250095bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 250184d17192SAl Viro err = graft_tree(newmnt, parent, mp); 25029d412a43SAl Viro 25039d412a43SAl Viro unlock: 250484d17192SAl Viro unlock_mount(mp); 25059d412a43SAl Viro return err; 25069d412a43SAl Viro } 2507b1e75df4SAl Viro 25088654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 25091b852bceSEric W. Biederman 25101da177e4SLinus Torvalds /* 25111da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 25121da177e4SLinus Torvalds * namespace's tree 25131da177e4SLinus Torvalds */ 2514e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2515808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 25161da177e4SLinus Torvalds { 25170c55cfc4SEric W. Biederman struct file_system_type *type; 25181da177e4SLinus Torvalds struct vfsmount *mnt; 251915f9a3f3SAl Viro int err; 25201da177e4SLinus Torvalds 25210c55cfc4SEric W. Biederman if (!fstype) 25221da177e4SLinus Torvalds return -EINVAL; 25231da177e4SLinus Torvalds 25240c55cfc4SEric W. Biederman type = get_fs_type(fstype); 25250c55cfc4SEric W. Biederman if (!type) 25260c55cfc4SEric W. Biederman return -ENODEV; 25270c55cfc4SEric W. Biederman 2528e462ec50SDavid Howells mnt = vfs_kern_mount(type, sb_flags, name, data); 25290c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 25300c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 25310c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 25320c55cfc4SEric W. Biederman 25330c55cfc4SEric W. Biederman put_filesystem(type); 25341da177e4SLinus Torvalds if (IS_ERR(mnt)) 25351da177e4SLinus Torvalds return PTR_ERR(mnt); 25361da177e4SLinus Torvalds 25378654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 25388654df4eSEric W. Biederman mntput(mnt); 25398654df4eSEric W. Biederman return -EPERM; 25408654df4eSEric W. Biederman } 25418654df4eSEric W. Biederman 254295bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 254315f9a3f3SAl Viro if (err) 254415f9a3f3SAl Viro mntput(mnt); 254515f9a3f3SAl Viro return err; 25461da177e4SLinus Torvalds } 25471da177e4SLinus Torvalds 254819a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 254919a167afSAl Viro { 25506776db3dSAl Viro struct mount *mnt = real_mount(m); 255119a167afSAl Viro int err; 255219a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 255319a167afSAl Viro * expired before we get a chance to add it 255419a167afSAl Viro */ 25556776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 255619a167afSAl Viro 255719a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 255819a167afSAl Viro m->mnt_root == path->dentry) { 2559b1e75df4SAl Viro err = -ELOOP; 2560b1e75df4SAl Viro goto fail; 256119a167afSAl Viro } 256219a167afSAl Viro 256395bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2564b1e75df4SAl Viro if (!err) 2565b1e75df4SAl Viro return 0; 2566b1e75df4SAl Viro fail: 2567b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 25686776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 256997216be0SAl Viro namespace_lock(); 25706776db3dSAl Viro list_del_init(&mnt->mnt_expire); 257197216be0SAl Viro namespace_unlock(); 257219a167afSAl Viro } 2573b1e75df4SAl Viro mntput(m); 2574b1e75df4SAl Viro mntput(m); 257519a167afSAl Viro return err; 257619a167afSAl Viro } 257719a167afSAl Viro 2578ea5b778aSDavid Howells /** 2579ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2580ea5b778aSDavid Howells * @mnt: The mount to list. 2581ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2582ea5b778aSDavid Howells */ 2583ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2584ea5b778aSDavid Howells { 258597216be0SAl Viro namespace_lock(); 2586ea5b778aSDavid Howells 25876776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2588ea5b778aSDavid Howells 258997216be0SAl Viro namespace_unlock(); 2590ea5b778aSDavid Howells } 2591ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2592ea5b778aSDavid Howells 2593ea5b778aSDavid Howells /* 25941da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 25951da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25961da177e4SLinus Torvalds * here 25971da177e4SLinus Torvalds */ 25981da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25991da177e4SLinus Torvalds { 2600761d5c38SAl Viro struct mount *mnt, *next; 26011da177e4SLinus Torvalds LIST_HEAD(graveyard); 26021da177e4SLinus Torvalds 26031da177e4SLinus Torvalds if (list_empty(mounts)) 26041da177e4SLinus Torvalds return; 26051da177e4SLinus Torvalds 260697216be0SAl Viro namespace_lock(); 2607719ea2fbSAl Viro lock_mount_hash(); 26081da177e4SLinus Torvalds 26091da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 26101da177e4SLinus Torvalds * following criteria: 26111da177e4SLinus Torvalds * - only referenced by its parent vfsmount 26121da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 26131da177e4SLinus Torvalds * cleared by mntput()) 26141da177e4SLinus Torvalds */ 26156776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2616863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 26171ab59738SAl Viro propagate_mount_busy(mnt, 1)) 26181da177e4SLinus Torvalds continue; 26196776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 26201da177e4SLinus Torvalds } 2621bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 26226776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2623143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2624e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2625bcc5c7d2SAl Viro } 2626719ea2fbSAl Viro unlock_mount_hash(); 26273ab6abeeSAl Viro namespace_unlock(); 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds 26301da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds /* 26335528f911STrond Myklebust * Ripoff of 'select_parent()' 26345528f911STrond Myklebust * 26355528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 26365528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 26375528f911STrond Myklebust */ 2638692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 26395528f911STrond Myklebust { 2640692afc31SAl Viro struct mount *this_parent = parent; 26415528f911STrond Myklebust struct list_head *next; 26425528f911STrond Myklebust int found = 0; 26435528f911STrond Myklebust 26445528f911STrond Myklebust repeat: 26456b41d536SAl Viro next = this_parent->mnt_mounts.next; 26465528f911STrond Myklebust resume: 26476b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 26485528f911STrond Myklebust struct list_head *tmp = next; 26496b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 26505528f911STrond Myklebust 26515528f911STrond Myklebust next = tmp->next; 2652692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 26535528f911STrond Myklebust continue; 26545528f911STrond Myklebust /* 26555528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 26565528f911STrond Myklebust */ 26576b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 26585528f911STrond Myklebust this_parent = mnt; 26595528f911STrond Myklebust goto repeat; 26605528f911STrond Myklebust } 26615528f911STrond Myklebust 26621ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 26636776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 26645528f911STrond Myklebust found++; 26655528f911STrond Myklebust } 26665528f911STrond Myklebust } 26675528f911STrond Myklebust /* 26685528f911STrond Myklebust * All done at this level ... ascend and resume the search 26695528f911STrond Myklebust */ 26705528f911STrond Myklebust if (this_parent != parent) { 26716b41d536SAl Viro next = this_parent->mnt_child.next; 26720714a533SAl Viro this_parent = this_parent->mnt_parent; 26735528f911STrond Myklebust goto resume; 26745528f911STrond Myklebust } 26755528f911STrond Myklebust return found; 26765528f911STrond Myklebust } 26775528f911STrond Myklebust 26785528f911STrond Myklebust /* 26795528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 26805528f911STrond Myklebust * submounts of a specific parent mountpoint 268199b7db7bSNick Piggin * 268248a066e7SAl Viro * mount_lock must be held for write 26835528f911STrond Myklebust */ 2684b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 26855528f911STrond Myklebust { 26865528f911STrond Myklebust LIST_HEAD(graveyard); 2687761d5c38SAl Viro struct mount *m; 26885528f911STrond Myklebust 26895528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2690c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2691bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2692761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 26936776db3dSAl Viro mnt_expire); 2694143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2695e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2696bcc5c7d2SAl Viro } 2697bcc5c7d2SAl Viro } 26985528f911STrond Myklebust } 26995528f911STrond Myklebust 27005528f911STrond Myklebust /* 27011da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 27021da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 27031da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 27041da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 27051da177e4SLinus Torvalds */ 2706b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2707b58fed8bSRam Pai unsigned long n) 27081da177e4SLinus Torvalds { 27091da177e4SLinus Torvalds char *t = to; 27101da177e4SLinus Torvalds const char __user *f = from; 27111da177e4SLinus Torvalds char c; 27121da177e4SLinus Torvalds 27131da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 27141da177e4SLinus Torvalds return n; 27151da177e4SLinus Torvalds 27161da177e4SLinus Torvalds while (n) { 27171da177e4SLinus Torvalds if (__get_user(c, f)) { 27181da177e4SLinus Torvalds memset(t, 0, n); 27191da177e4SLinus Torvalds break; 27201da177e4SLinus Torvalds } 27211da177e4SLinus Torvalds *t++ = c; 27221da177e4SLinus Torvalds f++; 27231da177e4SLinus Torvalds n--; 27241da177e4SLinus Torvalds } 27251da177e4SLinus Torvalds return n; 27261da177e4SLinus Torvalds } 27271da177e4SLinus Torvalds 2728b40ef869SAl Viro void *copy_mount_options(const void __user * data) 27291da177e4SLinus Torvalds { 27301da177e4SLinus Torvalds int i; 27311da177e4SLinus Torvalds unsigned long size; 2732b40ef869SAl Viro char *copy; 27331da177e4SLinus Torvalds 27341da177e4SLinus Torvalds if (!data) 2735b40ef869SAl Viro return NULL; 27361da177e4SLinus Torvalds 2737b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2738b40ef869SAl Viro if (!copy) 2739b40ef869SAl Viro return ERR_PTR(-ENOMEM); 27401da177e4SLinus Torvalds 27411da177e4SLinus Torvalds /* We only care that *some* data at the address the user 27421da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 27431da177e4SLinus Torvalds * the remainder of the page. 27441da177e4SLinus Torvalds */ 27451da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 27461da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 27471da177e4SLinus Torvalds if (size > PAGE_SIZE) 27481da177e4SLinus Torvalds size = PAGE_SIZE; 27491da177e4SLinus Torvalds 2750b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 27511da177e4SLinus Torvalds if (!i) { 2752b40ef869SAl Viro kfree(copy); 2753b40ef869SAl Viro return ERR_PTR(-EFAULT); 27541da177e4SLinus Torvalds } 27551da177e4SLinus Torvalds if (i != PAGE_SIZE) 2756b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2757b40ef869SAl Viro return copy; 27581da177e4SLinus Torvalds } 27591da177e4SLinus Torvalds 2760b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2761eca6f534SVegard Nossum { 2762b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2763eca6f534SVegard Nossum } 2764eca6f534SVegard Nossum 27651da177e4SLinus Torvalds /* 27661da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 27671da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 27681da177e4SLinus Torvalds * 27691da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 27701da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 27711da177e4SLinus Torvalds * information (or be NULL). 27721da177e4SLinus Torvalds * 27731da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 27741da177e4SLinus Torvalds * When the flags word was introduced its top half was required 27751da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 27761da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 27771da177e4SLinus Torvalds * and must be discarded. 27781da177e4SLinus Torvalds */ 27795e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2780808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 27811da177e4SLinus Torvalds { 27822d92ab3cSAl Viro struct path path; 2783e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 27841da177e4SLinus Torvalds int retval = 0; 27851da177e4SLinus Torvalds 27861da177e4SLinus Torvalds /* Discard magic */ 27871da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 27881da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 27891da177e4SLinus Torvalds 27901da177e4SLinus Torvalds /* Basic sanity checks */ 27911da177e4SLinus Torvalds if (data_page) 27921da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 27931da177e4SLinus Torvalds 2794e462ec50SDavid Howells if (flags & MS_NOUSER) 2795e462ec50SDavid Howells return -EINVAL; 2796e462ec50SDavid Howells 2797a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 27985e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2799a27ab9f2STetsuo Handa if (retval) 2800a27ab9f2STetsuo Handa return retval; 2801a27ab9f2STetsuo Handa 2802a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2803a27ab9f2STetsuo Handa type_page, flags, data_page); 28040d5cadb8SAl Viro if (!retval && !may_mount()) 28050d5cadb8SAl Viro retval = -EPERM; 2806e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 28079e8925b6SJeff Layton retval = -EPERM; 2808a27ab9f2STetsuo Handa if (retval) 2809a27ab9f2STetsuo Handa goto dput_out; 2810a27ab9f2STetsuo Handa 2811613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2812613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 28130a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 28140a1c01c9SMatthew Garrett 28151da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 28161da177e4SLinus Torvalds if (flags & MS_NOSUID) 28171da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 28181da177e4SLinus Torvalds if (flags & MS_NODEV) 28191da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 28201da177e4SLinus Torvalds if (flags & MS_NOEXEC) 28211da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2822fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2823fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2824fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2825fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2826d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2827d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 2828a9e5b732SDavid Howells if (flags & MS_RDONLY) 28292e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2830fc33a7bbSChristoph Hellwig 2831ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2832ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2833ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2834ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2835ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2836ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2837ffbc6f0eSEric W. Biederman } 2838ffbc6f0eSEric W. Biederman 2839e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 2840e462ec50SDavid Howells SB_SYNCHRONOUS | 2841e462ec50SDavid Howells SB_MANDLOCK | 2842e462ec50SDavid Howells SB_DIRSYNC | 2843e462ec50SDavid Howells SB_SILENT | 2844917086ffSMimi Zohar SB_POSIXACL | 2845d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 2846917086ffSMimi Zohar SB_I_VERSION); 28471da177e4SLinus Torvalds 28481da177e4SLinus Torvalds if (flags & MS_REMOUNT) 2849e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 28501da177e4SLinus Torvalds data_page); 28511da177e4SLinus Torvalds else if (flags & MS_BIND) 28522d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 28539676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 28542d92ab3cSAl Viro retval = do_change_type(&path, flags); 28551da177e4SLinus Torvalds else if (flags & MS_MOVE) 28562d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 28571da177e4SLinus Torvalds else 2858e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 28591da177e4SLinus Torvalds dev_name, data_page); 28601da177e4SLinus Torvalds dput_out: 28612d92ab3cSAl Viro path_put(&path); 28621da177e4SLinus Torvalds return retval; 28631da177e4SLinus Torvalds } 28641da177e4SLinus Torvalds 2865537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2866537f7ccbSEric W. Biederman { 2867537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2868537f7ccbSEric W. Biederman } 2869537f7ccbSEric W. Biederman 2870537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2871537f7ccbSEric W. Biederman { 2872537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2873537f7ccbSEric W. Biederman } 2874537f7ccbSEric W. Biederman 2875771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2876771b1371SEric W. Biederman { 28776344c433SAl Viro ns_free_inum(&ns->ns); 2878537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2879771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2880771b1371SEric W. Biederman kfree(ns); 2881771b1371SEric W. Biederman } 2882771b1371SEric W. Biederman 28838823c079SEric W. Biederman /* 28848823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 28858823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 28868823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 28878823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 28888823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 28898823c079SEric W. Biederman */ 28908823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 28918823c079SEric W. Biederman 2892771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2893cf8d2c11STrond Myklebust { 2894cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2895537f7ccbSEric W. Biederman struct ucounts *ucounts; 289698f842e6SEric W. Biederman int ret; 2897cf8d2c11STrond Myklebust 2898537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2899537f7ccbSEric W. Biederman if (!ucounts) 2900df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2901537f7ccbSEric W. Biederman 2902cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2903537f7ccbSEric W. Biederman if (!new_ns) { 2904537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2905cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2906537f7ccbSEric W. Biederman } 29076344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 290898f842e6SEric W. Biederman if (ret) { 290998f842e6SEric W. Biederman kfree(new_ns); 2910537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 291198f842e6SEric W. Biederman return ERR_PTR(ret); 291298f842e6SEric W. Biederman } 291333c42940SAl Viro new_ns->ns.ops = &mntns_operations; 29148823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2915cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2916cf8d2c11STrond Myklebust new_ns->root = NULL; 2917cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2918cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2919cf8d2c11STrond Myklebust new_ns->event = 0; 2920771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2921537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2922d2921684SEric W. Biederman new_ns->mounts = 0; 2923d2921684SEric W. Biederman new_ns->pending_mounts = 0; 2924cf8d2c11STrond Myklebust return new_ns; 2925cf8d2c11STrond Myklebust } 2926cf8d2c11STrond Myklebust 29270766f788SEmese Revfy __latent_entropy 29289559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 29299559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 29301da177e4SLinus Torvalds { 29316b3286edSKirill Korotaev struct mnt_namespace *new_ns; 29327f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2933315fc83eSAl Viro struct mount *p, *q; 29349559f689SAl Viro struct mount *old; 2935cb338d06SAl Viro struct mount *new; 29367a472ef4SEric W. Biederman int copy_flags; 29371da177e4SLinus Torvalds 29389559f689SAl Viro BUG_ON(!ns); 29399559f689SAl Viro 29409559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 29419559f689SAl Viro get_mnt_ns(ns); 29429559f689SAl Viro return ns; 29439559f689SAl Viro } 29449559f689SAl Viro 29459559f689SAl Viro old = ns->root; 29469559f689SAl Viro 2947771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2948cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2949cf8d2c11STrond Myklebust return new_ns; 29501da177e4SLinus Torvalds 295197216be0SAl Viro namespace_lock(); 29521da177e4SLinus Torvalds /* First pass: copy the tree topology */ 29534ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 29549559f689SAl Viro if (user_ns != ns->user_ns) 2955132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 29567a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2957be34d1a3SDavid Howells if (IS_ERR(new)) { 2958328e6d90SAl Viro namespace_unlock(); 2959771b1371SEric W. Biederman free_mnt_ns(new_ns); 2960be34d1a3SDavid Howells return ERR_CAST(new); 29611da177e4SLinus Torvalds } 2962be08d6d2SAl Viro new_ns->root = new; 29631a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 29641da177e4SLinus Torvalds 29651da177e4SLinus Torvalds /* 29661da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 29671da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 29681da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 29691da177e4SLinus Torvalds */ 2970909b0a88SAl Viro p = old; 2971cb338d06SAl Viro q = new; 29721da177e4SLinus Torvalds while (p) { 2973143c8c91SAl Viro q->mnt_ns = new_ns; 2974d2921684SEric W. Biederman new_ns->mounts++; 29759559f689SAl Viro if (new_fs) { 29769559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 29779559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2978315fc83eSAl Viro rootmnt = &p->mnt; 29791da177e4SLinus Torvalds } 29809559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 29819559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2982315fc83eSAl Viro pwdmnt = &p->mnt; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds } 2985909b0a88SAl Viro p = next_mnt(p, old); 2986909b0a88SAl Viro q = next_mnt(q, new); 29874ce5d2b1SEric W. Biederman if (!q) 29884ce5d2b1SEric W. Biederman break; 29894ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 29904ce5d2b1SEric W. Biederman p = next_mnt(p, old); 29911da177e4SLinus Torvalds } 2992328e6d90SAl Viro namespace_unlock(); 29931da177e4SLinus Torvalds 29941da177e4SLinus Torvalds if (rootmnt) 2995f03c6599SAl Viro mntput(rootmnt); 29961da177e4SLinus Torvalds if (pwdmnt) 2997f03c6599SAl Viro mntput(pwdmnt); 29981da177e4SLinus Torvalds 2999741a2951SJANAK DESAI return new_ns; 3000741a2951SJANAK DESAI } 3001741a2951SJANAK DESAI 3002cf8d2c11STrond Myklebust /** 3003cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 3004cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 3005cf8d2c11STrond Myklebust */ 30061a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 3007cf8d2c11STrond Myklebust { 3008771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 3009cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 30101a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 30111a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 3012be08d6d2SAl Viro new_ns->root = mnt; 3013d2921684SEric W. Biederman new_ns->mounts++; 3014b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 3015c1334495SAl Viro } else { 30161a4eeaf2SAl Viro mntput(m); 3017cf8d2c11STrond Myklebust } 3018cf8d2c11STrond Myklebust return new_ns; 3019cf8d2c11STrond Myklebust } 3020cf8d2c11STrond Myklebust 3021ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 3022ea441d11SAl Viro { 3023ea441d11SAl Viro struct mnt_namespace *ns; 3024d31da0f0SAl Viro struct super_block *s; 3025ea441d11SAl Viro struct path path; 3026ea441d11SAl Viro int err; 3027ea441d11SAl Viro 3028ea441d11SAl Viro ns = create_mnt_ns(mnt); 3029ea441d11SAl Viro if (IS_ERR(ns)) 3030ea441d11SAl Viro return ERR_CAST(ns); 3031ea441d11SAl Viro 3032ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 3033ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3034ea441d11SAl Viro 3035ea441d11SAl Viro put_mnt_ns(ns); 3036ea441d11SAl Viro 3037ea441d11SAl Viro if (err) 3038ea441d11SAl Viro return ERR_PTR(err); 3039ea441d11SAl Viro 3040ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3041d31da0f0SAl Viro s = path.mnt->mnt_sb; 3042d31da0f0SAl Viro atomic_inc(&s->s_active); 3043ea441d11SAl Viro mntput(path.mnt); 3044ea441d11SAl Viro /* lock the sucker */ 3045d31da0f0SAl Viro down_write(&s->s_umount); 3046ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3047ea441d11SAl Viro return path.dentry; 3048ea441d11SAl Viro } 3049ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3050ea441d11SAl Viro 3051312db1aaSDominik Brodowski int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, 3052312db1aaSDominik Brodowski unsigned long flags, void __user *data) 30531da177e4SLinus Torvalds { 3054eca6f534SVegard Nossum int ret; 3055eca6f534SVegard Nossum char *kernel_type; 3056eca6f534SVegard Nossum char *kernel_dev; 3057b40ef869SAl Viro void *options; 30581da177e4SLinus Torvalds 3059b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3060b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3061b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3062eca6f534SVegard Nossum goto out_type; 30631da177e4SLinus Torvalds 3064b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3065b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3066b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3067eca6f534SVegard Nossum goto out_dev; 30681da177e4SLinus Torvalds 3069b40ef869SAl Viro options = copy_mount_options(data); 3070b40ef869SAl Viro ret = PTR_ERR(options); 3071b40ef869SAl Viro if (IS_ERR(options)) 3072eca6f534SVegard Nossum goto out_data; 30731da177e4SLinus Torvalds 3074b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3075eca6f534SVegard Nossum 3076b40ef869SAl Viro kfree(options); 3077eca6f534SVegard Nossum out_data: 3078eca6f534SVegard Nossum kfree(kernel_dev); 3079eca6f534SVegard Nossum out_dev: 3080eca6f534SVegard Nossum kfree(kernel_type); 3081eca6f534SVegard Nossum out_type: 3082eca6f534SVegard Nossum return ret; 30831da177e4SLinus Torvalds } 30841da177e4SLinus Torvalds 3085312db1aaSDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3086312db1aaSDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 3087312db1aaSDominik Brodowski { 3088312db1aaSDominik Brodowski return ksys_mount(dev_name, dir_name, type, flags, data); 3089312db1aaSDominik Brodowski } 3090312db1aaSDominik Brodowski 30911da177e4SLinus Torvalds /* 3092afac7cbaSAl Viro * Return true if path is reachable from root 3093afac7cbaSAl Viro * 309448a066e7SAl Viro * namespace_sem or mount_lock is held 3095afac7cbaSAl Viro */ 3096643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3097afac7cbaSAl Viro const struct path *root) 3098afac7cbaSAl Viro { 3099643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3100a73324daSAl Viro dentry = mnt->mnt_mountpoint; 31010714a533SAl Viro mnt = mnt->mnt_parent; 3102afac7cbaSAl Viro } 3103643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3104afac7cbaSAl Viro } 3105afac7cbaSAl Viro 3106640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3107afac7cbaSAl Viro { 310825ab4c9bSYaowei Bai bool res; 310948a066e7SAl Viro read_seqlock_excl(&mount_lock); 3110643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 311148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3112afac7cbaSAl Viro return res; 3113afac7cbaSAl Viro } 3114afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3115afac7cbaSAl Viro 3116afac7cbaSAl Viro /* 31171da177e4SLinus Torvalds * pivot_root Semantics: 31181da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 31191da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 31201da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 31211da177e4SLinus Torvalds * 31221da177e4SLinus Torvalds * Restrictions: 31231da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 31241da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 31251da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 31261da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 31271da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 31281da177e4SLinus Torvalds * 31294a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 31304a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 31314a0d11faSNeil Brown * in this situation. 31324a0d11faSNeil Brown * 31331da177e4SLinus Torvalds * Notes: 31341da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 31351da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 31361da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 31371da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 31381da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 31391da177e4SLinus Torvalds * first. 31401da177e4SLinus Torvalds */ 31413480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 31423480b257SHeiko Carstens const char __user *, put_old) 31431da177e4SLinus Torvalds { 31442d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 314584d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 314684d17192SAl Viro struct mountpoint *old_mp, *root_mp; 31471da177e4SLinus Torvalds int error; 31481da177e4SLinus Torvalds 31499b40bc90SAl Viro if (!may_mount()) 31501da177e4SLinus Torvalds return -EPERM; 31511da177e4SLinus Torvalds 31522d8f3038SAl Viro error = user_path_dir(new_root, &new); 31531da177e4SLinus Torvalds if (error) 31541da177e4SLinus Torvalds goto out0; 31551da177e4SLinus Torvalds 31562d8f3038SAl Viro error = user_path_dir(put_old, &old); 31571da177e4SLinus Torvalds if (error) 31581da177e4SLinus Torvalds goto out1; 31591da177e4SLinus Torvalds 31602d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3161b12cea91SAl Viro if (error) 3162b12cea91SAl Viro goto out2; 31631da177e4SLinus Torvalds 3164f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 316584d17192SAl Viro old_mp = lock_mount(&old); 316684d17192SAl Viro error = PTR_ERR(old_mp); 316784d17192SAl Viro if (IS_ERR(old_mp)) 3168b12cea91SAl Viro goto out3; 3169b12cea91SAl Viro 31701da177e4SLinus Torvalds error = -EINVAL; 3171419148daSAl Viro new_mnt = real_mount(new.mnt); 3172419148daSAl Viro root_mnt = real_mount(root.mnt); 317384d17192SAl Viro old_mnt = real_mount(old.mnt); 317484d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3175fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3176fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3177b12cea91SAl Viro goto out4; 3178143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3179b12cea91SAl Viro goto out4; 31805ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 31815ff9d8a6SEric W. Biederman goto out4; 31821da177e4SLinus Torvalds error = -ENOENT; 3183f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3184b12cea91SAl Viro goto out4; 31851da177e4SLinus Torvalds error = -EBUSY; 318684d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3187b12cea91SAl Viro goto out4; /* loop, on the same file system */ 31881da177e4SLinus Torvalds error = -EINVAL; 31898c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3190b12cea91SAl Viro goto out4; /* not a mountpoint */ 3191676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3192b12cea91SAl Viro goto out4; /* not attached */ 319384d17192SAl Viro root_mp = root_mnt->mnt_mp; 31942d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3195b12cea91SAl Viro goto out4; /* not a mountpoint */ 3196676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3197b12cea91SAl Viro goto out4; /* not attached */ 31984ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 319984d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3200b12cea91SAl Viro goto out4; 32010d082601SEric W. Biederman /* make certain new is below the root */ 32020d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 32030d082601SEric W. Biederman goto out4; 320484d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3205719ea2fbSAl Viro lock_mount_hash(); 3206419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3207419148daSAl Viro detach_mnt(root_mnt, &root_parent); 32085ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 32095ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 32105ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 32115ff9d8a6SEric W. Biederman } 32124ac91378SJan Blunck /* mount old root on put_old */ 321384d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 32144ac91378SJan Blunck /* mount new_root on / */ 321584d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 32166b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 32174fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 32184fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 32193895dbf8SEric W. Biederman put_mountpoint(root_mp); 3220719ea2fbSAl Viro unlock_mount_hash(); 32212d8f3038SAl Viro chroot_fs_refs(&root, &new); 32221da177e4SLinus Torvalds error = 0; 3223b12cea91SAl Viro out4: 322484d17192SAl Viro unlock_mount(old_mp); 3225b12cea91SAl Viro if (!error) { 32261a390689SAl Viro path_put(&root_parent); 32271a390689SAl Viro path_put(&parent_path); 3228b12cea91SAl Viro } 3229b12cea91SAl Viro out3: 32308c3ee42eSAl Viro path_put(&root); 3231b12cea91SAl Viro out2: 32322d8f3038SAl Viro path_put(&old); 32331da177e4SLinus Torvalds out1: 32342d8f3038SAl Viro path_put(&new); 32351da177e4SLinus Torvalds out0: 32361da177e4SLinus Torvalds return error; 32371da177e4SLinus Torvalds } 32381da177e4SLinus Torvalds 32391da177e4SLinus Torvalds static void __init init_mount_tree(void) 32401da177e4SLinus Torvalds { 32411da177e4SLinus Torvalds struct vfsmount *mnt; 32426b3286edSKirill Korotaev struct mnt_namespace *ns; 3243ac748a09SJan Blunck struct path root; 32440c55cfc4SEric W. Biederman struct file_system_type *type; 32451da177e4SLinus Torvalds 32460c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 32470c55cfc4SEric W. Biederman if (!type) 32480c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 32490c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 32500c55cfc4SEric W. Biederman put_filesystem(type); 32511da177e4SLinus Torvalds if (IS_ERR(mnt)) 32521da177e4SLinus Torvalds panic("Can't create rootfs"); 3253b3e19d92SNick Piggin 32543b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 32553b22edc5STrond Myklebust if (IS_ERR(ns)) 32561da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 32571da177e4SLinus Torvalds 32586b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 32596b3286edSKirill Korotaev get_mnt_ns(ns); 32601da177e4SLinus Torvalds 3261be08d6d2SAl Viro root.mnt = mnt; 3262be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3263da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3264ac748a09SJan Blunck 3265ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3266ac748a09SJan Blunck set_fs_root(current->fs, &root); 32671da177e4SLinus Torvalds } 32681da177e4SLinus Torvalds 326974bf17cfSDenis Cheng void __init mnt_init(void) 32701da177e4SLinus Torvalds { 327115a67dd8SRandy Dunlap int err; 32721da177e4SLinus Torvalds 32737d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 327420c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 32751da177e4SLinus Torvalds 32760818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 327738129a13SAl Viro sizeof(struct hlist_head), 32780818bf27SAl Viro mhash_entries, 19, 32793d375d78SPavel Tatashin HASH_ZERO, 32800818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 32810818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 32820818bf27SAl Viro sizeof(struct hlist_head), 32830818bf27SAl Viro mphash_entries, 19, 32843d375d78SPavel Tatashin HASH_ZERO, 32850818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 32861da177e4SLinus Torvalds 328784d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 32881da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 32891da177e4SLinus Torvalds 32904b93dc9bSTejun Heo kernfs_init(); 32914b93dc9bSTejun Heo 329215a67dd8SRandy Dunlap err = sysfs_init(); 329315a67dd8SRandy Dunlap if (err) 329415a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 32958e24eea7SHarvey Harrison __func__, err); 329600d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 329700d26666SGreg Kroah-Hartman if (!fs_kobj) 32988e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 32991da177e4SLinus Torvalds init_rootfs(); 33001da177e4SLinus Torvalds init_mount_tree(); 33011da177e4SLinus Torvalds } 33021da177e4SLinus Torvalds 3303616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 33041da177e4SLinus Torvalds { 3305d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3306616511d0STrond Myklebust return; 33077b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3308771b1371SEric W. Biederman free_mnt_ns(ns); 33091da177e4SLinus Torvalds } 33109d412a43SAl Viro 33119d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 33129d412a43SAl Viro { 3313423e0ab0STim Chen struct vfsmount *mnt; 3314e462ec50SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); 3315423e0ab0STim Chen if (!IS_ERR(mnt)) { 3316423e0ab0STim Chen /* 3317423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3318423e0ab0STim Chen * we unmount before file sys is unregistered 3319423e0ab0STim Chen */ 3320f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3321423e0ab0STim Chen } 3322423e0ab0STim Chen return mnt; 33239d412a43SAl Viro } 33249d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3325423e0ab0STim Chen 3326423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3327423e0ab0STim Chen { 3328423e0ab0STim Chen /* release long term mount so mount point can be released */ 3329423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3330f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 333148a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3332423e0ab0STim Chen mntput(mnt); 3333423e0ab0STim Chen } 3334423e0ab0STim Chen } 3335423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 333602125a82SAl Viro 333702125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 333802125a82SAl Viro { 3339143c8c91SAl Viro return check_mnt(real_mount(mnt)); 334002125a82SAl Viro } 33418823c079SEric W. Biederman 33423151527eSEric W. Biederman bool current_chrooted(void) 33433151527eSEric W. Biederman { 33443151527eSEric W. Biederman /* Does the current process have a non-standard root */ 33453151527eSEric W. Biederman struct path ns_root; 33463151527eSEric W. Biederman struct path fs_root; 33473151527eSEric W. Biederman bool chrooted; 33483151527eSEric W. Biederman 33493151527eSEric W. Biederman /* Find the namespace root */ 33503151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 33513151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 33523151527eSEric W. Biederman path_get(&ns_root); 33533151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 33543151527eSEric W. Biederman ; 33553151527eSEric W. Biederman 33563151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 33573151527eSEric W. Biederman 33583151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 33593151527eSEric W. Biederman 33603151527eSEric W. Biederman path_put(&fs_root); 33613151527eSEric W. Biederman path_put(&ns_root); 33623151527eSEric W. Biederman 33633151527eSEric W. Biederman return chrooted; 33643151527eSEric W. Biederman } 33653151527eSEric W. Biederman 33668654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 33678654df4eSEric W. Biederman int *new_mnt_flags) 336887a8ebd6SEric W. Biederman { 33698c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 337087a8ebd6SEric W. Biederman struct mount *mnt; 3371e51db735SEric W. Biederman bool visible = false; 337287a8ebd6SEric W. Biederman 337344bb4385SAl Viro down_read(&namespace_sem); 337487a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3375e51db735SEric W. Biederman struct mount *child; 337677b1a97dSEric W. Biederman int mnt_flags; 337777b1a97dSEric W. Biederman 33788654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3379e51db735SEric W. Biederman continue; 3380e51db735SEric W. Biederman 33817e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 33827e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 33837e96c1b0SEric W. Biederman */ 33847e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 33857e96c1b0SEric W. Biederman continue; 33867e96c1b0SEric W. Biederman 3387a1935c17SEric W. Biederman /* A local view of the mount flags */ 338877b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 338977b1a97dSEric W. Biederman 3390695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3391bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3392695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3393695e9df0SEric W. Biederman 33948c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 33958c6cf9ccSEric W. Biederman * than the proposed new mount. 33968c6cf9ccSEric W. Biederman */ 339777b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 33988c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 33998c6cf9ccSEric W. Biederman continue; 340077b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 340177b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 34028c6cf9ccSEric W. Biederman continue; 34038c6cf9ccSEric W. Biederman 3404ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3405ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3406ceeb0e5dSEric W. Biederman * empty directories. 3407e51db735SEric W. Biederman */ 3408e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3409e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3410ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3411d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3412ceeb0e5dSEric W. Biederman continue; 34137236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 34147236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3415e51db735SEric W. Biederman goto next; 341687a8ebd6SEric W. Biederman } 34178c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 341877b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 34198c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3420e51db735SEric W. Biederman visible = true; 3421e51db735SEric W. Biederman goto found; 3422e51db735SEric W. Biederman next: ; 342387a8ebd6SEric W. Biederman } 3424e51db735SEric W. Biederman found: 342544bb4385SAl Viro up_read(&namespace_sem); 3426e51db735SEric W. Biederman return visible; 342787a8ebd6SEric W. Biederman } 342887a8ebd6SEric W. Biederman 34298654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 34308654df4eSEric W. Biederman { 3431a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 34328654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 34338654df4eSEric W. Biederman unsigned long s_iflags; 34348654df4eSEric W. Biederman 34358654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 34368654df4eSEric W. Biederman return false; 34378654df4eSEric W. Biederman 34388654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 34398654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 34408654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 34418654df4eSEric W. Biederman return false; 34428654df4eSEric W. Biederman 3443a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3444a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3445a1935c17SEric W. Biederman required_iflags); 3446a1935c17SEric W. Biederman return true; 3447a1935c17SEric W. Biederman } 3448a1935c17SEric W. Biederman 34498654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 34508654df4eSEric W. Biederman } 34518654df4eSEric W. Biederman 3452380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3453380cf5baSAndy Lutomirski { 3454380cf5baSAndy Lutomirski /* 3455380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3456380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3457380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3458380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3459380cf5baSAndy Lutomirski * in other namespaces. 3460380cf5baSAndy Lutomirski */ 3461380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3462380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3463380cf5baSAndy Lutomirski } 3464380cf5baSAndy Lutomirski 346564964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 34668823c079SEric W. Biederman { 346758be2825SAl Viro struct ns_common *ns = NULL; 34688823c079SEric W. Biederman struct nsproxy *nsproxy; 34698823c079SEric W. Biederman 3470728dba3aSEric W. Biederman task_lock(task); 3471728dba3aSEric W. Biederman nsproxy = task->nsproxy; 34728823c079SEric W. Biederman if (nsproxy) { 347358be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 347458be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 34758823c079SEric W. Biederman } 3476728dba3aSEric W. Biederman task_unlock(task); 34778823c079SEric W. Biederman 34788823c079SEric W. Biederman return ns; 34798823c079SEric W. Biederman } 34808823c079SEric W. Biederman 348164964528SAl Viro static void mntns_put(struct ns_common *ns) 34828823c079SEric W. Biederman { 348358be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 34848823c079SEric W. Biederman } 34858823c079SEric W. Biederman 348664964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 34878823c079SEric W. Biederman { 34888823c079SEric W. Biederman struct fs_struct *fs = current->fs; 34894f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 34908823c079SEric W. Biederman struct path root; 34914f757f3cSAl Viro int err; 34928823c079SEric W. Biederman 34930c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3494c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3495c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3496ae11e0f1SZhao Hongjiang return -EPERM; 34978823c079SEric W. Biederman 34988823c079SEric W. Biederman if (fs->users != 1) 34998823c079SEric W. Biederman return -EINVAL; 35008823c079SEric W. Biederman 35018823c079SEric W. Biederman get_mnt_ns(mnt_ns); 35024f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 35038823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 35048823c079SEric W. Biederman 35058823c079SEric W. Biederman /* Find the root */ 35064f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 35074f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 35084f757f3cSAl Viro if (err) { 35094f757f3cSAl Viro /* revert to old namespace */ 35104f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 35114f757f3cSAl Viro put_mnt_ns(mnt_ns); 35124f757f3cSAl Viro return err; 35134f757f3cSAl Viro } 35148823c079SEric W. Biederman 35154068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 35164068367cSAndrei Vagin 35178823c079SEric W. Biederman /* Update the pwd and root */ 35188823c079SEric W. Biederman set_fs_pwd(fs, &root); 35198823c079SEric W. Biederman set_fs_root(fs, &root); 35208823c079SEric W. Biederman 35218823c079SEric W. Biederman path_put(&root); 35228823c079SEric W. Biederman return 0; 35238823c079SEric W. Biederman } 35248823c079SEric W. Biederman 3525bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3526bcac25a5SAndrey Vagin { 3527bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3528bcac25a5SAndrey Vagin } 3529bcac25a5SAndrey Vagin 35308823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 35318823c079SEric W. Biederman .name = "mnt", 35328823c079SEric W. Biederman .type = CLONE_NEWNS, 35338823c079SEric W. Biederman .get = mntns_get, 35348823c079SEric W. Biederman .put = mntns_put, 35358823c079SEric W. Biederman .install = mntns_install, 3536bcac25a5SAndrey Vagin .owner = mntns_owner, 35378823c079SEric W. Biederman }; 3538