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; 2782e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 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(); 3561e75529eSMiao Xie while (ACCESS_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 /** 43496029c4eSnpiggin@suse.de * mnt_want_write_file - 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 43996029c4eSnpiggin@suse.de */ 44096029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 44196029c4eSnpiggin@suse.de { 442eb04c282SJan Kara int ret; 443eb04c282SJan Kara 444eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 445eb04c282SJan Kara ret = __mnt_want_write_file(file); 446eb04c282SJan Kara if (ret) 447eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 448eb04c282SJan Kara return ret; 44996029c4eSnpiggin@suse.de } 45096029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 45196029c4eSnpiggin@suse.de 45296029c4eSnpiggin@suse.de /** 453eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4548366025eSDave Hansen * @mnt: the mount on which to give up write access 4558366025eSDave Hansen * 4568366025eSDave Hansen * Tells the low-level filesystem that we are done 4578366025eSDave Hansen * performing writes to it. Must be matched with 458eb04c282SJan Kara * __mnt_want_write() call above. 4598366025eSDave Hansen */ 460eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4618366025eSDave Hansen { 462d3ef3d73Snpiggin@suse.de preempt_disable(); 46383adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 464d3ef3d73Snpiggin@suse.de preempt_enable(); 4658366025eSDave Hansen } 466eb04c282SJan Kara 467eb04c282SJan Kara /** 468eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 469eb04c282SJan Kara * @mnt: the mount on which to give up write access 470eb04c282SJan Kara * 471eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 472eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 473eb04c282SJan Kara * mnt_want_write() call above. 474eb04c282SJan Kara */ 475eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 476eb04c282SJan Kara { 477eb04c282SJan Kara __mnt_drop_write(mnt); 478eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 479eb04c282SJan Kara } 4808366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4818366025eSDave Hansen 482eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 483eb04c282SJan Kara { 484eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 485eb04c282SJan Kara } 486eb04c282SJan Kara 4872a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4882a79f17eSAl Viro { 4892a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4902a79f17eSAl Viro } 4912a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4922a79f17eSAl Viro 49383adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4948366025eSDave Hansen { 4953d733633SDave Hansen int ret = 0; 4963d733633SDave Hansen 497719ea2fbSAl Viro lock_mount_hash(); 49883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 499d3ef3d73Snpiggin@suse.de /* 500d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 501d3ef3d73Snpiggin@suse.de * should be visible before we do. 502d3ef3d73Snpiggin@suse.de */ 503d3ef3d73Snpiggin@suse.de smp_mb(); 504d3ef3d73Snpiggin@suse.de 505d3ef3d73Snpiggin@suse.de /* 506d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 507d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 508d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 509d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 510d3ef3d73Snpiggin@suse.de * 511d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 512d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 513d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 514d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 515d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 516d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 517d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 518d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 519d3ef3d73Snpiggin@suse.de * we're counting up here. 520d3ef3d73Snpiggin@suse.de */ 521c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 522d3ef3d73Snpiggin@suse.de ret = -EBUSY; 523d3ef3d73Snpiggin@suse.de else 52483adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 525d3ef3d73Snpiggin@suse.de /* 526d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 527d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 528d3ef3d73Snpiggin@suse.de */ 529d3ef3d73Snpiggin@suse.de smp_wmb(); 53083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 531719ea2fbSAl Viro unlock_mount_hash(); 5323d733633SDave Hansen return ret; 5333d733633SDave Hansen } 5348366025eSDave Hansen 53583adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5362e4b7fcdSDave Hansen { 537719ea2fbSAl Viro lock_mount_hash(); 53883adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 539719ea2fbSAl Viro unlock_mount_hash(); 5402e4b7fcdSDave Hansen } 5412e4b7fcdSDave Hansen 5424ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5434ed5e82fSMiklos Szeredi { 5444ed5e82fSMiklos Szeredi struct mount *mnt; 5454ed5e82fSMiklos Szeredi int err = 0; 5464ed5e82fSMiklos Szeredi 5478e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5488e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5498e8b8796SMiklos Szeredi return -EBUSY; 5508e8b8796SMiklos Szeredi 551719ea2fbSAl Viro lock_mount_hash(); 5524ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5534ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5544ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5554ed5e82fSMiklos Szeredi smp_mb(); 5564ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5574ed5e82fSMiklos Szeredi err = -EBUSY; 5584ed5e82fSMiklos Szeredi break; 5594ed5e82fSMiklos Szeredi } 5604ed5e82fSMiklos Szeredi } 5614ed5e82fSMiklos Szeredi } 5628e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5638e8b8796SMiklos Szeredi err = -EBUSY; 5648e8b8796SMiklos Szeredi 5654ed5e82fSMiklos Szeredi if (!err) { 5664ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5674ed5e82fSMiklos Szeredi smp_wmb(); 5684ed5e82fSMiklos Szeredi } 5694ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5704ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5714ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5724ed5e82fSMiklos Szeredi } 573719ea2fbSAl Viro unlock_mount_hash(); 5744ed5e82fSMiklos Szeredi 5754ed5e82fSMiklos Szeredi return err; 5764ed5e82fSMiklos Szeredi } 5774ed5e82fSMiklos Szeredi 578b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5791da177e4SLinus Torvalds { 580fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 581d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 58268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 583d3ef3d73Snpiggin@suse.de #endif 584b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds 5878ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5888ffcb32eSDavid Howells { 5898ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5908ffcb32eSDavid Howells } 5918ffcb32eSDavid Howells 59248a066e7SAl Viro /* call under rcu_read_lock */ 593294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 59448a066e7SAl Viro { 59548a066e7SAl Viro struct mount *mnt; 59648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 597294d71ffSAl Viro return 1; 59848a066e7SAl Viro if (bastard == NULL) 599294d71ffSAl Viro return 0; 60048a066e7SAl Viro mnt = real_mount(bastard); 60148a066e7SAl Viro mnt_add_count(mnt, 1); 60248a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 603294d71ffSAl Viro return 0; 60448a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 60548a066e7SAl Viro mnt_add_count(mnt, -1); 606294d71ffSAl Viro return 1; 60748a066e7SAl Viro } 608294d71ffSAl Viro return -1; 609294d71ffSAl Viro } 610294d71ffSAl Viro 611294d71ffSAl Viro /* call under rcu_read_lock */ 612294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 613294d71ffSAl Viro { 614294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 615294d71ffSAl Viro if (likely(!res)) 616294d71ffSAl Viro return true; 617294d71ffSAl Viro if (unlikely(res < 0)) { 61848a066e7SAl Viro rcu_read_unlock(); 61948a066e7SAl Viro mntput(bastard); 62048a066e7SAl Viro rcu_read_lock(); 621294d71ffSAl Viro } 62248a066e7SAl Viro return false; 62348a066e7SAl Viro } 62448a066e7SAl Viro 6251da177e4SLinus Torvalds /* 626474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 62748a066e7SAl Viro * call under rcu_read_lock() 6281da177e4SLinus Torvalds */ 629474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6301da177e4SLinus Torvalds { 63138129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 632474279dcSAl Viro struct mount *p; 6331da177e4SLinus Torvalds 63438129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 635474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 636474279dcSAl Viro return p; 637474279dcSAl Viro return NULL; 6381da177e4SLinus Torvalds } 639474279dcSAl Viro 640474279dcSAl Viro /* 641f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 642f015f126SDavid Howells * 643f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 644f015f126SDavid Howells * following mounts: 645f015f126SDavid Howells * 646f015f126SDavid Howells * mount /dev/sda1 /mnt 647f015f126SDavid Howells * mount /dev/sda2 /mnt 648f015f126SDavid Howells * mount /dev/sda3 /mnt 649f015f126SDavid Howells * 650f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 651f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 652f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 653f015f126SDavid Howells * 654f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 655a05964f3SRam Pai */ 656ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 657a05964f3SRam Pai { 658c7105365SAl Viro struct mount *child_mnt; 65948a066e7SAl Viro struct vfsmount *m; 66048a066e7SAl Viro unsigned seq; 66199b7db7bSNick Piggin 66248a066e7SAl Viro rcu_read_lock(); 66348a066e7SAl Viro do { 66448a066e7SAl Viro seq = read_seqbegin(&mount_lock); 665474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 66648a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 66748a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 66848a066e7SAl Viro rcu_read_unlock(); 66948a066e7SAl Viro return m; 670a05964f3SRam Pai } 671a05964f3SRam Pai 6727af1364fSEric W. Biederman /* 6737af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6747af1364fSEric W. Biederman * current mount namespace. 6757af1364fSEric W. Biederman * 6767af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6777af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6787af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6797af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6807af1364fSEric W. Biederman * is a mountpoint. 6817af1364fSEric W. Biederman * 6827af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6837af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6847af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6857af1364fSEric W. Biederman * parent mount. 6867af1364fSEric W. Biederman */ 6877af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6887af1364fSEric W. Biederman { 6897af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6907af1364fSEric W. Biederman struct mount *mnt; 6917af1364fSEric W. Biederman bool is_covered = false; 6927af1364fSEric W. Biederman 6937af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6947af1364fSEric W. Biederman goto out; 6957af1364fSEric W. Biederman 6967af1364fSEric W. Biederman down_read(&namespace_sem); 6977af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6987af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6997af1364fSEric W. Biederman if (is_covered) 7007af1364fSEric W. Biederman break; 7017af1364fSEric W. Biederman } 7027af1364fSEric W. Biederman up_read(&namespace_sem); 7037af1364fSEric W. Biederman out: 7047af1364fSEric W. Biederman return is_covered; 7057af1364fSEric W. Biederman } 7067af1364fSEric W. Biederman 707e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 70884d17192SAl Viro { 7090818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 71084d17192SAl Viro struct mountpoint *mp; 71184d17192SAl Viro 7120818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 71384d17192SAl Viro if (mp->m_dentry == dentry) { 71484d17192SAl Viro /* might be worth a WARN_ON() */ 71584d17192SAl Viro if (d_unlinked(dentry)) 71684d17192SAl Viro return ERR_PTR(-ENOENT); 71784d17192SAl Viro mp->m_count++; 71884d17192SAl Viro return mp; 71984d17192SAl Viro } 72084d17192SAl Viro } 721e2dfa935SEric W. Biederman return NULL; 722e2dfa935SEric W. Biederman } 723e2dfa935SEric W. Biederman 7243895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 725e2dfa935SEric W. Biederman { 7263895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 727e2dfa935SEric W. Biederman int ret; 72884d17192SAl Viro 7293895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7303895dbf8SEric W. Biederman mountpoint: 7313895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7323895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7333895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7343895dbf8SEric W. Biederman if (mp) 7353895dbf8SEric W. Biederman goto done; 73684d17192SAl Viro } 737eed81007SMiklos Szeredi 7383895dbf8SEric W. Biederman if (!new) 7393895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7403895dbf8SEric W. Biederman if (!new) 7413895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7423895dbf8SEric W. Biederman 7433895dbf8SEric W. Biederman 7443895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7453895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7463895dbf8SEric W. Biederman 7473895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7483895dbf8SEric W. Biederman if (ret == -EBUSY) 7493895dbf8SEric W. Biederman goto mountpoint; 7503895dbf8SEric W. Biederman 7513895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7523895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7533895dbf8SEric W. Biederman if (ret) 7543895dbf8SEric W. Biederman goto done; 7553895dbf8SEric W. Biederman 7563895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7573895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7583895dbf8SEric W. Biederman new->m_dentry = dentry; 7593895dbf8SEric W. Biederman new->m_count = 1; 7603895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7613895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7623895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7633895dbf8SEric W. Biederman 7643895dbf8SEric W. Biederman mp = new; 7653895dbf8SEric W. Biederman new = NULL; 7663895dbf8SEric W. Biederman done: 7673895dbf8SEric W. Biederman kfree(new); 76884d17192SAl Viro return mp; 76984d17192SAl Viro } 77084d17192SAl Viro 77184d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 77284d17192SAl Viro { 77384d17192SAl Viro if (!--mp->m_count) { 77484d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7750a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77684d17192SAl Viro spin_lock(&dentry->d_lock); 77784d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 77884d17192SAl Viro spin_unlock(&dentry->d_lock); 7790818bf27SAl Viro hlist_del(&mp->m_hash); 78084d17192SAl Viro kfree(mp); 78184d17192SAl Viro } 78284d17192SAl Viro } 78384d17192SAl Viro 784143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7851da177e4SLinus Torvalds { 7866b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7871da177e4SLinus Torvalds } 7881da177e4SLinus Torvalds 78999b7db7bSNick Piggin /* 79099b7db7bSNick Piggin * vfsmount lock must be held for write 79199b7db7bSNick Piggin */ 7926b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7935addc5ddSAl Viro { 7945addc5ddSAl Viro if (ns) { 7955addc5ddSAl Viro ns->event = ++event; 7965addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7975addc5ddSAl Viro } 7985addc5ddSAl Viro } 7995addc5ddSAl Viro 80099b7db7bSNick Piggin /* 80199b7db7bSNick Piggin * vfsmount lock must be held for write 80299b7db7bSNick Piggin */ 8036b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8045addc5ddSAl Viro { 8055addc5ddSAl Viro if (ns && ns->event != event) { 8065addc5ddSAl Viro ns->event = event; 8075addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8085addc5ddSAl Viro } 8095addc5ddSAl Viro } 8105addc5ddSAl Viro 81199b7db7bSNick Piggin /* 81299b7db7bSNick Piggin * vfsmount lock must be held for write 81399b7db7bSNick Piggin */ 8147bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8151da177e4SLinus Torvalds { 8160714a533SAl Viro mnt->mnt_parent = mnt; 817a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8186b41d536SAl Viro list_del_init(&mnt->mnt_child); 81938129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8200a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 82184d17192SAl Viro put_mountpoint(mnt->mnt_mp); 82284d17192SAl Viro mnt->mnt_mp = NULL; 8231da177e4SLinus Torvalds } 8241da177e4SLinus Torvalds 82599b7db7bSNick Piggin /* 82699b7db7bSNick Piggin * vfsmount lock must be held for write 82799b7db7bSNick Piggin */ 8287bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8297bdb11deSEric W. Biederman { 8307bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8317bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8327bdb11deSEric W. Biederman unhash_mnt(mnt); 8337bdb11deSEric W. Biederman } 8347bdb11deSEric W. Biederman 8357bdb11deSEric W. Biederman /* 8367bdb11deSEric W. Biederman * vfsmount lock must be held for write 8377bdb11deSEric W. Biederman */ 8386a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8396a46c573SEric W. Biederman { 8406a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8416a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8426a46c573SEric W. Biederman unhash_mnt(mnt); 8436a46c573SEric W. Biederman } 8446a46c573SEric W. Biederman 8456a46c573SEric W. Biederman /* 8466a46c573SEric W. Biederman * vfsmount lock must be held for write 8476a46c573SEric W. Biederman */ 84884d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 84984d17192SAl Viro struct mountpoint *mp, 85044d964d6SAl Viro struct mount *child_mnt) 851b90fa9aeSRam Pai { 85284d17192SAl Viro mp->m_count++; 8533a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 85484d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8553a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85684d17192SAl Viro child_mnt->mnt_mp = mp; 8570a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 858b90fa9aeSRam Pai } 859b90fa9aeSRam Pai 8601064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8611064f874SEric W. Biederman { 8621064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8631064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8641064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8651064f874SEric W. Biederman } 8661064f874SEric W. Biederman 86799b7db7bSNick Piggin /* 86899b7db7bSNick Piggin * vfsmount lock must be held for write 86999b7db7bSNick Piggin */ 87084d17192SAl Viro static void attach_mnt(struct mount *mnt, 87184d17192SAl Viro struct mount *parent, 87284d17192SAl Viro struct mountpoint *mp) 8731da177e4SLinus Torvalds { 87484d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8751064f874SEric W. Biederman __attach_mnt(mnt, parent); 876b90fa9aeSRam Pai } 877b90fa9aeSRam Pai 8781064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 87912a5b529SAl Viro { 8801064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8811064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 8821064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8831064f874SEric W. Biederman 8841064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8851064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8861064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8871064f874SEric W. Biederman 8881064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8891064f874SEric W. Biederman 8901064f874SEric W. Biederman put_mountpoint(old_mp); 8911064f874SEric W. Biederman 8921064f874SEric W. Biederman /* 8931064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 8941064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 8951064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 8961064f874SEric W. Biederman * to a mountpoint. 8971064f874SEric W. Biederman * 8981064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 8991064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 9001064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 9011064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 9021064f874SEric W. Biederman */ 9031064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 9041064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 9051064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 9061064f874SEric W. Biederman 9071064f874SEric W. Biederman mnt_add_count(old_parent, -1); 90812a5b529SAl Viro } 90912a5b529SAl Viro 910b90fa9aeSRam Pai /* 91199b7db7bSNick Piggin * vfsmount lock must be held for write 912b90fa9aeSRam Pai */ 9131064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 914b90fa9aeSRam Pai { 9150714a533SAl Viro struct mount *parent = mnt->mnt_parent; 91683adc753SAl Viro struct mount *m; 917b90fa9aeSRam Pai LIST_HEAD(head); 918143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 919b90fa9aeSRam Pai 9200714a533SAl Viro BUG_ON(parent == mnt); 921b90fa9aeSRam Pai 9221a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 923f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 924143c8c91SAl Viro m->mnt_ns = n; 925f03c6599SAl Viro 926b90fa9aeSRam Pai list_splice(&head, n->list.prev); 927b90fa9aeSRam Pai 928d2921684SEric W. Biederman n->mounts += n->pending_mounts; 929d2921684SEric W. Biederman n->pending_mounts = 0; 930d2921684SEric W. Biederman 9311064f874SEric W. Biederman __attach_mnt(mnt, parent); 9326b3286edSKirill Korotaev touch_mnt_namespace(n); 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 935909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9361da177e4SLinus Torvalds { 9376b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9386b41d536SAl Viro if (next == &p->mnt_mounts) { 9391da177e4SLinus Torvalds while (1) { 940909b0a88SAl Viro if (p == root) 9411da177e4SLinus Torvalds return NULL; 9426b41d536SAl Viro next = p->mnt_child.next; 9436b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9441da177e4SLinus Torvalds break; 9450714a533SAl Viro p = p->mnt_parent; 9461da177e4SLinus Torvalds } 9471da177e4SLinus Torvalds } 9486b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds 951315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9529676f0c6SRam Pai { 9536b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9546b41d536SAl Viro while (prev != &p->mnt_mounts) { 9556b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9566b41d536SAl Viro prev = p->mnt_mounts.prev; 9579676f0c6SRam Pai } 9589676f0c6SRam Pai return p; 9599676f0c6SRam Pai } 9609676f0c6SRam Pai 9619d412a43SAl Viro struct vfsmount * 9629d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9639d412a43SAl Viro { 964b105e270SAl Viro struct mount *mnt; 9659d412a43SAl Viro struct dentry *root; 9669d412a43SAl Viro 9679d412a43SAl Viro if (!type) 9689d412a43SAl Viro return ERR_PTR(-ENODEV); 9699d412a43SAl Viro 9709d412a43SAl Viro mnt = alloc_vfsmnt(name); 9719d412a43SAl Viro if (!mnt) 9729d412a43SAl Viro return ERR_PTR(-ENOMEM); 9739d412a43SAl Viro 9749d412a43SAl Viro if (flags & MS_KERNMOUNT) 975b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9769d412a43SAl Viro 9779d412a43SAl Viro root = mount_fs(type, flags, name, data); 9789d412a43SAl Viro if (IS_ERR(root)) { 9798ffcb32eSDavid Howells mnt_free_id(mnt); 9809d412a43SAl Viro free_vfsmnt(mnt); 9819d412a43SAl Viro return ERR_CAST(root); 9829d412a43SAl Viro } 9839d412a43SAl Viro 984b105e270SAl Viro mnt->mnt.mnt_root = root; 985b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 986a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9870714a533SAl Viro mnt->mnt_parent = mnt; 988719ea2fbSAl Viro lock_mount_hash(); 98939f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 990719ea2fbSAl Viro unlock_mount_hash(); 991b105e270SAl Viro return &mnt->mnt; 9929d412a43SAl Viro } 9939d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9949d412a43SAl Viro 99593faccbbSEric W. Biederman struct vfsmount * 99693faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 99793faccbbSEric W. Biederman const char *name, void *data) 99893faccbbSEric W. Biederman { 99993faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 100093faccbbSEric W. Biederman * through from the parent mount to the submount don't support 100193faccbbSEric W. Biederman * unprivileged mounts with submounts. 100293faccbbSEric W. Biederman */ 100393faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 100493faccbbSEric W. Biederman return ERR_PTR(-EPERM); 100593faccbbSEric W. Biederman 100693faccbbSEric W. Biederman return vfs_kern_mount(type, MS_SUBMOUNT, name, data); 100793faccbbSEric W. Biederman } 100893faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 100993faccbbSEric W. Biederman 101087129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 101136341f64SRam Pai int flag) 10121da177e4SLinus Torvalds { 101387129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1014be34d1a3SDavid Howells struct mount *mnt; 1015be34d1a3SDavid Howells int err; 10161da177e4SLinus Torvalds 1017be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1018be34d1a3SDavid Howells if (!mnt) 1019be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1020be34d1a3SDavid Howells 10217a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 102215169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1023719f5d7fSMiklos Szeredi else 102415169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1025719f5d7fSMiklos Szeredi 102615169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1027be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1028719f5d7fSMiklos Szeredi if (err) 1029719f5d7fSMiklos Szeredi goto out_free; 1030719f5d7fSMiklos Szeredi } 1031719f5d7fSMiklos Szeredi 1032f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 1033132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 10349566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 10359566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 10369566d674SEric W. Biederman 10379566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 1038132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 1039132c94e3SEric W. Biederman 10409566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 10419566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 10429566d674SEric W. Biederman 10439566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10449566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10459566d674SEric W. Biederman 10469566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10479566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10489566d674SEric W. Biederman } 10499566d674SEric W. Biederman 10505ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1051381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1052381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10535ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10545ff9d8a6SEric W. Biederman 10551da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1056b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1057b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1058a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10590714a533SAl Viro mnt->mnt_parent = mnt; 1060719ea2fbSAl Viro lock_mount_hash(); 106139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1062719ea2fbSAl Viro unlock_mount_hash(); 1063b90fa9aeSRam Pai 10647a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10657a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10666776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 106732301920SAl Viro mnt->mnt_master = old; 1068fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10698aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1070fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10716776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1072d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10736776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1074d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10755235d448SAl Viro } else { 10765235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10775afe0022SRam Pai } 1078b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10790f0afb1dSAl Viro set_mnt_shared(mnt); 10801da177e4SLinus Torvalds 10811da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10821da177e4SLinus Torvalds * as the original if that was on one */ 108336341f64SRam Pai if (flag & CL_EXPIRE) { 10846776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10856776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10861da177e4SLinus Torvalds } 1087be34d1a3SDavid Howells 1088cb338d06SAl Viro return mnt; 1089719f5d7fSMiklos Szeredi 1090719f5d7fSMiklos Szeredi out_free: 10918ffcb32eSDavid Howells mnt_free_id(mnt); 1092719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1093be34d1a3SDavid Howells return ERR_PTR(err); 10941da177e4SLinus Torvalds } 10951da177e4SLinus Torvalds 10969ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10979ea459e1SAl Viro { 10989ea459e1SAl Viro /* 10999ea459e1SAl Viro * This probably indicates that somebody messed 11009ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 11019ea459e1SAl Viro * happens, the filesystem was probably unable 11029ea459e1SAl Viro * to make r/w->r/o transitions. 11039ea459e1SAl Viro */ 11049ea459e1SAl Viro /* 11059ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11069ea459e1SAl Viro * so mnt_get_writers() below is safe. 11079ea459e1SAl Viro */ 11089ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11099ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11109ea459e1SAl Viro mnt_pin_kill(mnt); 11119ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11129ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11139ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11149ea459e1SAl Viro mnt_free_id(mnt); 11159ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11169ea459e1SAl Viro } 11179ea459e1SAl Viro 11189ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11199ea459e1SAl Viro { 11209ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11219ea459e1SAl Viro } 11229ea459e1SAl Viro 11239ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11249ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11259ea459e1SAl Viro { 11269ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 11279ea459e1SAl Viro struct llist_node *next; 11289ea459e1SAl Viro 11299ea459e1SAl Viro for (; node; node = next) { 11309ea459e1SAl Viro next = llist_next(node); 11319ea459e1SAl Viro cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); 11329ea459e1SAl Viro } 11339ea459e1SAl Viro } 11349ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11359ea459e1SAl Viro 1136900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11377b7b1aceSAl Viro { 113848a066e7SAl Viro rcu_read_lock(); 1139aa9c0e07SAl Viro mnt_add_count(mnt, -1); 114048a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 114148a066e7SAl Viro rcu_read_unlock(); 114299b7db7bSNick Piggin return; 1143b3e19d92SNick Piggin } 1144719ea2fbSAl Viro lock_mount_hash(); 1145b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 114648a066e7SAl Viro rcu_read_unlock(); 1147719ea2fbSAl Viro unlock_mount_hash(); 114899b7db7bSNick Piggin return; 114999b7db7bSNick Piggin } 115048a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 115148a066e7SAl Viro rcu_read_unlock(); 115248a066e7SAl Viro unlock_mount_hash(); 115348a066e7SAl Viro return; 115448a066e7SAl Viro } 115548a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 115648a066e7SAl Viro rcu_read_unlock(); 1157962830dfSAndi Kleen 115839f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1159ce07d891SEric W. Biederman 1160ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1161ce07d891SEric W. Biederman struct mount *p, *tmp; 1162ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1163ce07d891SEric W. Biederman umount_mnt(p); 1164ce07d891SEric W. Biederman } 1165ce07d891SEric W. Biederman } 1166719ea2fbSAl Viro unlock_mount_hash(); 1167649a795aSAl Viro 11689ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11699ea459e1SAl Viro struct task_struct *task = current; 11709ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11719ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11729ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11739ea459e1SAl Viro return; 11749ea459e1SAl Viro } 11759ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11769ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11779ea459e1SAl Viro return; 11789ea459e1SAl Viro } 11799ea459e1SAl Viro cleanup_mnt(mnt); 1180b3e19d92SNick Piggin } 1181b3e19d92SNick Piggin 1182b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1183b3e19d92SNick Piggin { 1184b3e19d92SNick Piggin if (mnt) { 1185863d684fSAl Viro struct mount *m = real_mount(mnt); 1186b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1187863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1188863d684fSAl Viro m->mnt_expiry_mark = 0; 1189863d684fSAl Viro mntput_no_expire(m); 1190b3e19d92SNick Piggin } 1191b3e19d92SNick Piggin } 1192b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1193b3e19d92SNick Piggin 1194b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1195b3e19d92SNick Piggin { 1196b3e19d92SNick Piggin if (mnt) 119783adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1198b3e19d92SNick Piggin return mnt; 1199b3e19d92SNick Piggin } 1200b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1201b3e19d92SNick Piggin 1202c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1203c6609c0aSIan Kent * namespace. 1204c6609c0aSIan Kent * 1205c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1206c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1207c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1208c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1209c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1210c6609c0aSIan Kent * alone. 1211c6609c0aSIan Kent */ 1212c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1213c6609c0aSIan Kent { 1214c6609c0aSIan Kent unsigned seq; 1215c6609c0aSIan Kent bool res; 1216c6609c0aSIan Kent 1217c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1218c6609c0aSIan Kent return false; 1219c6609c0aSIan Kent 1220c6609c0aSIan Kent rcu_read_lock(); 1221c6609c0aSIan Kent do { 1222c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1223c6609c0aSIan Kent res = __path_is_mountpoint(path); 1224c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1225c6609c0aSIan Kent rcu_read_unlock(); 1226c6609c0aSIan Kent 1227c6609c0aSIan Kent return res; 1228c6609c0aSIan Kent } 1229c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1230c6609c0aSIan Kent 1231ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12327b7b1aceSAl Viro { 12333064c356SAl Viro struct mount *p; 12343064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12353064c356SAl Viro if (IS_ERR(p)) 12363064c356SAl Viro return ERR_CAST(p); 12373064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12383064c356SAl Viro return &p->mnt; 12397b7b1aceSAl Viro } 12401da177e4SLinus Torvalds 1241b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1242b3b304a2SMiklos Szeredi { 1243b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1244b3b304a2SMiklos Szeredi } 1245b3b304a2SMiklos Szeredi 1246b3b304a2SMiklos Szeredi /* 1247b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1248b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1249b3b304a2SMiklos Szeredi * 1250b3b304a2SMiklos Szeredi * See also save_mount_options(). 1251b3b304a2SMiklos Szeredi */ 125234c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1253b3b304a2SMiklos Szeredi { 12542a32cebdSAl Viro const char *options; 12552a32cebdSAl Viro 12562a32cebdSAl Viro rcu_read_lock(); 125734c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1258b3b304a2SMiklos Szeredi 1259b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1260b3b304a2SMiklos Szeredi seq_putc(m, ','); 1261b3b304a2SMiklos Szeredi mangle(m, options); 1262b3b304a2SMiklos Szeredi } 12632a32cebdSAl Viro rcu_read_unlock(); 1264b3b304a2SMiklos Szeredi 1265b3b304a2SMiklos Szeredi return 0; 1266b3b304a2SMiklos Szeredi } 1267b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1268b3b304a2SMiklos Szeredi 1269b3b304a2SMiklos Szeredi /* 1270b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1271b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1272b3b304a2SMiklos Szeredi * 1273b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1274b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1275b3b304a2SMiklos Szeredi * remount fails. 1276b3b304a2SMiklos Szeredi * 1277b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1278b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1279b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1280b3b304a2SMiklos Szeredi * any more. 1281b3b304a2SMiklos Szeredi */ 1282b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1283b3b304a2SMiklos Szeredi { 12842a32cebdSAl Viro BUG_ON(sb->s_options); 12852a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1286b3b304a2SMiklos Szeredi } 1287b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1288b3b304a2SMiklos Szeredi 12892a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 12902a32cebdSAl Viro { 12912a32cebdSAl Viro char *old = sb->s_options; 12922a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 12932a32cebdSAl Viro if (old) { 12942a32cebdSAl Viro synchronize_rcu(); 12952a32cebdSAl Viro kfree(old); 12962a32cebdSAl Viro } 12972a32cebdSAl Viro } 12982a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 12992a32cebdSAl Viro 1300a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 13010226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13021da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13031da177e4SLinus Torvalds { 1304ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13051da177e4SLinus Torvalds 1306390c6843SRam Pai down_read(&namespace_sem); 1307c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1308c7999c36SAl Viro void *v = p->cached_mount; 1309c7999c36SAl Viro if (*pos == p->cached_index) 1310c7999c36SAl Viro return v; 1311c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1312c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1313c7999c36SAl Viro return p->cached_mount = v; 1314c7999c36SAl Viro } 1315c7999c36SAl Viro } 1316c7999c36SAl Viro 1317c7999c36SAl Viro p->cached_event = p->ns->event; 1318c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1319c7999c36SAl Viro p->cached_index = *pos; 1320c7999c36SAl Viro return p->cached_mount; 13211da177e4SLinus Torvalds } 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13241da177e4SLinus Torvalds { 1325ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1326b0765fb8SPavel Emelianov 1327c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1328c7999c36SAl Viro p->cached_index = *pos; 1329c7999c36SAl Viro return p->cached_mount; 13301da177e4SLinus Torvalds } 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13331da177e4SLinus Torvalds { 1334390c6843SRam Pai up_read(&namespace_sem); 13351da177e4SLinus Torvalds } 13361da177e4SLinus Torvalds 13370226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13389f5596afSAl Viro { 1339ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13401a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 13410226f492SAl Viro return p->show(m, &r->mnt); 13421da177e4SLinus Torvalds } 13431da177e4SLinus Torvalds 1344a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13451da177e4SLinus Torvalds .start = m_start, 13461da177e4SLinus Torvalds .next = m_next, 13471da177e4SLinus Torvalds .stop = m_stop, 13480226f492SAl Viro .show = m_show, 1349b4629fe2SChuck Lever }; 1350a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1351b4629fe2SChuck Lever 13521da177e4SLinus Torvalds /** 13531da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13541da177e4SLinus Torvalds * @mnt: root of mount tree 13551da177e4SLinus Torvalds * 13561da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13571da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13581da177e4SLinus Torvalds * busy. 13591da177e4SLinus Torvalds */ 1360909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13611da177e4SLinus Torvalds { 1362909b0a88SAl Viro struct mount *mnt = real_mount(m); 136336341f64SRam Pai int actual_refs = 0; 136436341f64SRam Pai int minimum_refs = 0; 1365315fc83eSAl Viro struct mount *p; 1366909b0a88SAl Viro BUG_ON(!m); 13671da177e4SLinus Torvalds 1368b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1369719ea2fbSAl Viro lock_mount_hash(); 1370909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 137183adc753SAl Viro actual_refs += mnt_get_count(p); 13721da177e4SLinus Torvalds minimum_refs += 2; 13731da177e4SLinus Torvalds } 1374719ea2fbSAl Viro unlock_mount_hash(); 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13771da177e4SLinus Torvalds return 0; 1378e3474a8eSIan Kent 1379e3474a8eSIan Kent return 1; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds /** 13851da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13861da177e4SLinus Torvalds * @mnt: root of mount 13871da177e4SLinus Torvalds * 13881da177e4SLinus Torvalds * This is called to check if a mount point has any 13891da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13901da177e4SLinus Torvalds * mount has sub mounts this will return busy 13911da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13921da177e4SLinus Torvalds * 13931da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13941da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13951da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13961da177e4SLinus Torvalds */ 13971da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13981da177e4SLinus Torvalds { 1399e3474a8eSIan Kent int ret = 1; 14008ad08d8aSAl Viro down_read(&namespace_sem); 1401719ea2fbSAl Viro lock_mount_hash(); 14021ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1403e3474a8eSIan Kent ret = 0; 1404719ea2fbSAl Viro unlock_mount_hash(); 14058ad08d8aSAl Viro up_read(&namespace_sem); 1406a05964f3SRam Pai return ret; 14071da177e4SLinus Torvalds } 14081da177e4SLinus Torvalds 14091da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14101da177e4SLinus Torvalds 141138129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1412e3197d83SAl Viro 141397216be0SAl Viro static void namespace_unlock(void) 14141da177e4SLinus Torvalds { 1415a3b3c562SEric W. Biederman struct hlist_head head; 141697216be0SAl Viro 1417a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1418a3b3c562SEric W. Biederman 141997216be0SAl Viro up_write(&namespace_sem); 1420a3b3c562SEric W. Biederman 1421a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 142297216be0SAl Viro return; 142397216be0SAl Viro 142448a066e7SAl Viro synchronize_rcu(); 142548a066e7SAl Viro 142687b95ce0SAl Viro group_pin_kill(&head); 142770fbcdf4SRam Pai } 142870fbcdf4SRam Pai 142997216be0SAl Viro static inline void namespace_lock(void) 1430e3197d83SAl Viro { 143197216be0SAl Viro down_write(&namespace_sem); 1432e3197d83SAl Viro } 1433e3197d83SAl Viro 1434e819f152SEric W. Biederman enum umount_tree_flags { 1435e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1436e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1437e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1438e819f152SEric W. Biederman }; 1439f2d0a123SEric W. Biederman 1440f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1441f2d0a123SEric W. Biederman { 1442f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1443f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1444f2d0a123SEric W. Biederman return true; 1445f2d0a123SEric W. Biederman 1446f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1447f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1448f2d0a123SEric W. Biederman return true; 1449f2d0a123SEric W. Biederman 1450f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1451f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1452f2d0a123SEric W. Biederman * connected to mounted mounts. 1453f2d0a123SEric W. Biederman */ 1454f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1455f2d0a123SEric W. Biederman return true; 1456f2d0a123SEric W. Biederman 1457f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1458f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1459f2d0a123SEric W. Biederman return false; 1460f2d0a123SEric W. Biederman 1461f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1462f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1463f2d0a123SEric W. Biederman return false; 1464f2d0a123SEric W. Biederman 1465f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1466f2d0a123SEric W. Biederman return true; 1467f2d0a123SEric W. Biederman } 1468f2d0a123SEric W. Biederman 146999b7db7bSNick Piggin /* 147048a066e7SAl Viro * mount_lock must be held 147199b7db7bSNick Piggin * namespace_sem must be held for write 147299b7db7bSNick Piggin */ 1473e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 147470fbcdf4SRam Pai { 1475c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1476315fc83eSAl Viro struct mount *p; 147770fbcdf4SRam Pai 14785d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14795d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14805d88457eSEric W. Biederman 1481c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1482590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1483590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1484c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1485590ce4bcSEric W. Biederman } 1486c003b26fSEric W. Biederman 1487411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1488c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1489c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 149038129a13SAl Viro } 149170fbcdf4SRam Pai 1492c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1493e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14947b8a53fdSAl Viro propagate_umount(&tmp_list); 1495a05964f3SRam Pai 1496c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1497d2921684SEric W. Biederman struct mnt_namespace *ns; 1498ce07d891SEric W. Biederman bool disconnect; 1499c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15006776db3dSAl Viro list_del_init(&p->mnt_expire); 15011a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1502d2921684SEric W. Biederman ns = p->mnt_ns; 1503d2921684SEric W. Biederman if (ns) { 1504d2921684SEric W. Biederman ns->mounts--; 1505d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1506d2921684SEric W. Biederman } 150763d37a84SAl Viro p->mnt_ns = NULL; 1508e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 150948a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 151087b95ce0SAl Viro 1511f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1512ce07d891SEric W. Biederman 1513ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1514ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1515676da58dSAl Viro if (mnt_has_parent(p)) { 151681b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1517ce07d891SEric W. Biederman if (!disconnect) { 1518ce07d891SEric W. Biederman /* Don't forget about p */ 1519ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1520ce07d891SEric W. Biederman } else { 15216a46c573SEric W. Biederman umount_mnt(p); 15227c4b93d8SAl Viro } 1523ce07d891SEric W. Biederman } 15240f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 152538129a13SAl Viro } 15261da177e4SLinus Torvalds } 15271da177e4SLinus Torvalds 1528b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1529c35038beSAl Viro 15301ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15311da177e4SLinus Torvalds { 15321ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15331da177e4SLinus Torvalds int retval; 15341da177e4SLinus Torvalds 15351ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15361da177e4SLinus Torvalds if (retval) 15371da177e4SLinus Torvalds return retval; 15381da177e4SLinus Torvalds 15391da177e4SLinus Torvalds /* 15401da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15411da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15421da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15431da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15441da177e4SLinus Torvalds */ 15451da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15461ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15471da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15481da177e4SLinus Torvalds return -EINVAL; 15491da177e4SLinus Torvalds 1550b3e19d92SNick Piggin /* 1551b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1552b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1553b3e19d92SNick Piggin */ 1554719ea2fbSAl Viro lock_mount_hash(); 155583adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1556719ea2fbSAl Viro unlock_mount_hash(); 15571da177e4SLinus Torvalds return -EBUSY; 1558b3e19d92SNick Piggin } 1559719ea2fbSAl Viro unlock_mount_hash(); 15601da177e4SLinus Torvalds 1561863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15621da177e4SLinus Torvalds return -EAGAIN; 15631da177e4SLinus Torvalds } 15641da177e4SLinus Torvalds 15651da177e4SLinus Torvalds /* 15661da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15671da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15681da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15691da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15701da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15711da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15721da177e4SLinus Torvalds * about for the moment. 15731da177e4SLinus Torvalds */ 15741da177e4SLinus Torvalds 157542faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 157642faad99SAl Viro sb->s_op->umount_begin(sb); 157742faad99SAl Viro } 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds /* 15801da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15811da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15821da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15831da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15841da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15851da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15861da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15871da177e4SLinus Torvalds */ 15881ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15891da177e4SLinus Torvalds /* 15901da177e4SLinus Torvalds * Special case for "unmounting" root ... 15911da177e4SLinus Torvalds * we just try to remount it readonly. 15921da177e4SLinus Torvalds */ 1593a1480dccSAndy Lutomirski if (!capable(CAP_SYS_ADMIN)) 1594a1480dccSAndy Lutomirski return -EPERM; 15951da177e4SLinus Torvalds down_write(&sb->s_umount); 15964aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 15971da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 15981da177e4SLinus Torvalds up_write(&sb->s_umount); 15991da177e4SLinus Torvalds return retval; 16001da177e4SLinus Torvalds } 16011da177e4SLinus Torvalds 160297216be0SAl Viro namespace_lock(); 1603719ea2fbSAl Viro lock_mount_hash(); 16045addc5ddSAl Viro event++; 16051da177e4SLinus Torvalds 160648a066e7SAl Viro if (flags & MNT_DETACH) { 160748a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1608e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 160948a066e7SAl Viro retval = 0; 161048a066e7SAl Viro } else { 1611b54b9be7SAl Viro shrink_submounts(mnt); 16121da177e4SLinus Torvalds retval = -EBUSY; 161348a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 16141a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1615e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16161da177e4SLinus Torvalds retval = 0; 16171da177e4SLinus Torvalds } 161848a066e7SAl Viro } 1619719ea2fbSAl Viro unlock_mount_hash(); 1620e3197d83SAl Viro namespace_unlock(); 16211da177e4SLinus Torvalds return retval; 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds /* 162580b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 162680b5dce8SEric W. Biederman * 162780b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 162880b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 162980b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 163080b5dce8SEric W. Biederman * leaking them. 163180b5dce8SEric W. Biederman * 163280b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 163380b5dce8SEric W. Biederman */ 163480b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 163580b5dce8SEric W. Biederman { 163680b5dce8SEric W. Biederman struct mountpoint *mp; 163780b5dce8SEric W. Biederman struct mount *mnt; 163880b5dce8SEric W. Biederman 163980b5dce8SEric W. Biederman namespace_lock(); 16403895dbf8SEric W. Biederman lock_mount_hash(); 164180b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1642f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 164380b5dce8SEric W. Biederman goto out_unlock; 164480b5dce8SEric W. Biederman 1645e06b933eSAndrey Ulanov event++; 164680b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 164780b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1648ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1649fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1650fe78fcc8SEric W. Biederman umount_mnt(mnt); 1651ce07d891SEric W. Biederman } 1652e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 165380b5dce8SEric W. Biederman } 165480b5dce8SEric W. Biederman put_mountpoint(mp); 165580b5dce8SEric W. Biederman out_unlock: 16563895dbf8SEric W. Biederman unlock_mount_hash(); 165780b5dce8SEric W. Biederman namespace_unlock(); 165880b5dce8SEric W. Biederman } 165980b5dce8SEric W. Biederman 166080b5dce8SEric W. Biederman /* 16619b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16629b40bc90SAl Viro */ 16639b40bc90SAl Viro static inline bool may_mount(void) 16649b40bc90SAl Viro { 16659b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16669b40bc90SAl Viro } 16679b40bc90SAl Viro 16689e8925b6SJeff Layton static inline bool may_mandlock(void) 16699e8925b6SJeff Layton { 16709e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16719e8925b6SJeff Layton return false; 16729e8925b6SJeff Layton #endif 167395ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16749e8925b6SJeff Layton } 16759e8925b6SJeff Layton 16769b40bc90SAl Viro /* 16771da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16781da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16791da177e4SLinus Torvalds * 16801da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16811da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16821da177e4SLinus Torvalds */ 16831da177e4SLinus Torvalds 1684bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16851da177e4SLinus Torvalds { 16862d8f3038SAl Viro struct path path; 1687900148dcSAl Viro struct mount *mnt; 16881da177e4SLinus Torvalds int retval; 1689db1f05bbSMiklos Szeredi int lookup_flags = 0; 16901da177e4SLinus Torvalds 1691db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1692db1f05bbSMiklos Szeredi return -EINVAL; 1693db1f05bbSMiklos Szeredi 16949b40bc90SAl Viro if (!may_mount()) 16959b40bc90SAl Viro return -EPERM; 16969b40bc90SAl Viro 1697db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1698db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1699db1f05bbSMiklos Szeredi 1700197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 17011da177e4SLinus Torvalds if (retval) 17021da177e4SLinus Torvalds goto out; 1703900148dcSAl Viro mnt = real_mount(path.mnt); 17041da177e4SLinus Torvalds retval = -EINVAL; 17052d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 17061da177e4SLinus Torvalds goto dput_and_out; 1707143c8c91SAl Viro if (!check_mnt(mnt)) 17081da177e4SLinus Torvalds goto dput_and_out; 17095ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 17105ff9d8a6SEric W. Biederman goto dput_and_out; 1711b2f5d4dcSEric W. Biederman retval = -EPERM; 1712b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1713b2f5d4dcSEric W. Biederman goto dput_and_out; 17141da177e4SLinus Torvalds 1715900148dcSAl Viro retval = do_umount(mnt, flags); 17161da177e4SLinus Torvalds dput_and_out: 1717429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 17182d8f3038SAl Viro dput(path.dentry); 1719900148dcSAl Viro mntput_no_expire(mnt); 17201da177e4SLinus Torvalds out: 17211da177e4SLinus Torvalds return retval; 17221da177e4SLinus Torvalds } 17231da177e4SLinus Torvalds 17241da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17251da177e4SLinus Torvalds 17261da177e4SLinus Torvalds /* 17271da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17281da177e4SLinus Torvalds */ 1729bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17301da177e4SLinus Torvalds { 17311da177e4SLinus Torvalds return sys_umount(name, 0); 17321da177e4SLinus Torvalds } 17331da177e4SLinus Torvalds 17341da177e4SLinus Torvalds #endif 17351da177e4SLinus Torvalds 17364ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17378823c079SEric W. Biederman { 17384ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1739e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1740e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17414ce5d2b1SEric W. Biederman } 17424ce5d2b1SEric W. Biederman 174358be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 174458be2825SAl Viro { 174558be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 174658be2825SAl Viro } 174758be2825SAl Viro 17484ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17494ce5d2b1SEric W. Biederman { 17504ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17514ce5d2b1SEric W. Biederman * mount namespace loop? 17524ce5d2b1SEric W. Biederman */ 17534ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17544ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17554ce5d2b1SEric W. Biederman return false; 17564ce5d2b1SEric W. Biederman 1757f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17588823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17598823c079SEric W. Biederman } 17608823c079SEric W. Biederman 176187129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 176236341f64SRam Pai int flag) 17631da177e4SLinus Torvalds { 176484d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17651da177e4SLinus Torvalds 17664ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17674ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17684ce5d2b1SEric W. Biederman 17694ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1770be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17719676f0c6SRam Pai 177236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1773be34d1a3SDavid Howells if (IS_ERR(q)) 1774be34d1a3SDavid Howells return q; 1775be34d1a3SDavid Howells 1776a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds p = mnt; 17796b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1780315fc83eSAl Viro struct mount *s; 17817ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17821da177e4SLinus Torvalds continue; 17831da177e4SLinus Torvalds 1784909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17854ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17864ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 17874ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17884ce5d2b1SEric W. Biederman continue; 17894ce5d2b1SEric W. Biederman } 17904ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17914ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17929676f0c6SRam Pai s = skip_mnt_tree(s); 17939676f0c6SRam Pai continue; 17949676f0c6SRam Pai } 17950714a533SAl Viro while (p != s->mnt_parent) { 17960714a533SAl Viro p = p->mnt_parent; 17970714a533SAl Viro q = q->mnt_parent; 17981da177e4SLinus Torvalds } 179987129cc0SAl Viro p = s; 180084d17192SAl Viro parent = q; 180187129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1802be34d1a3SDavid Howells if (IS_ERR(q)) 1803be34d1a3SDavid Howells goto out; 1804719ea2fbSAl Viro lock_mount_hash(); 18051a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18061064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1807719ea2fbSAl Viro unlock_mount_hash(); 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds } 18101da177e4SLinus Torvalds return res; 1811be34d1a3SDavid Howells out: 18121da177e4SLinus Torvalds if (res) { 1813719ea2fbSAl Viro lock_mount_hash(); 1814e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1815719ea2fbSAl Viro unlock_mount_hash(); 18161da177e4SLinus Torvalds } 1817be34d1a3SDavid Howells return q; 18181da177e4SLinus Torvalds } 18191da177e4SLinus Torvalds 1820be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1821be34d1a3SDavid Howells 1822ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18238aec0809SAl Viro { 1824cb338d06SAl Viro struct mount *tree; 182597216be0SAl Viro namespace_lock(); 1826cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1827cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1828cd4a4017SEric W. Biederman else 182987129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 183087129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1831328e6d90SAl Viro namespace_unlock(); 1832be34d1a3SDavid Howells if (IS_ERR(tree)) 183352e220d3SDan Carpenter return ERR_CAST(tree); 1834be34d1a3SDavid Howells return &tree->mnt; 18358aec0809SAl Viro } 18368aec0809SAl Viro 18378aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 18388aec0809SAl Viro { 183997216be0SAl Viro namespace_lock(); 1840719ea2fbSAl Viro lock_mount_hash(); 1841e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1842719ea2fbSAl Viro unlock_mount_hash(); 18433ab6abeeSAl Viro namespace_unlock(); 18448aec0809SAl Viro } 18458aec0809SAl Viro 1846c771d683SMiklos Szeredi /** 1847c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1848c771d683SMiklos Szeredi * 1849c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1850c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1851c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1852c771d683SMiklos Szeredi * 1853c771d683SMiklos Szeredi * Release with mntput(). 1854c771d683SMiklos Szeredi */ 1855ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1856c771d683SMiklos Szeredi { 1857c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1858c771d683SMiklos Szeredi struct mount *new_mnt; 1859c771d683SMiklos Szeredi 1860c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1861c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1862c771d683SMiklos Szeredi 1863c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1864c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1865c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1866c771d683SMiklos Szeredi 1867c771d683SMiklos Szeredi return &new_mnt->mnt; 1868c771d683SMiklos Szeredi } 1869c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1870c771d683SMiklos Szeredi 18711f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18721f707137SAl Viro struct vfsmount *root) 18731f707137SAl Viro { 18741a4eeaf2SAl Viro struct mount *mnt; 18751f707137SAl Viro int res = f(root, arg); 18761f707137SAl Viro if (res) 18771f707137SAl Viro return res; 18781a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18791a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18801f707137SAl Viro if (res) 18811f707137SAl Viro return res; 18821f707137SAl Viro } 18831f707137SAl Viro return 0; 18841f707137SAl Viro } 18851f707137SAl Viro 18864b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1887719f5d7fSMiklos Szeredi { 1888315fc83eSAl Viro struct mount *p; 1889719f5d7fSMiklos Szeredi 1890909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1891fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18924b8b21f4SAl Viro mnt_release_group_id(p); 1893719f5d7fSMiklos Szeredi } 1894719f5d7fSMiklos Szeredi } 1895719f5d7fSMiklos Szeredi 18964b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1897719f5d7fSMiklos Szeredi { 1898315fc83eSAl Viro struct mount *p; 1899719f5d7fSMiklos Szeredi 1900909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1901fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19024b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1903719f5d7fSMiklos Szeredi if (err) { 19044b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1905719f5d7fSMiklos Szeredi return err; 1906719f5d7fSMiklos Szeredi } 1907719f5d7fSMiklos Szeredi } 1908719f5d7fSMiklos Szeredi } 1909719f5d7fSMiklos Szeredi 1910719f5d7fSMiklos Szeredi return 0; 1911719f5d7fSMiklos Szeredi } 1912719f5d7fSMiklos Szeredi 1913d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1914d2921684SEric W. Biederman { 1915d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1916d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1917d2921684SEric W. Biederman struct mount *p; 1918d2921684SEric W. Biederman 1919d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1920d2921684SEric W. Biederman mounts++; 1921d2921684SEric W. Biederman 1922d2921684SEric W. Biederman old = ns->mounts; 1923d2921684SEric W. Biederman pending = ns->pending_mounts; 1924d2921684SEric W. Biederman sum = old + pending; 1925d2921684SEric W. Biederman if ((old > sum) || 1926d2921684SEric W. Biederman (pending > sum) || 1927d2921684SEric W. Biederman (max < sum) || 1928d2921684SEric W. Biederman (mounts > (max - sum))) 1929d2921684SEric W. Biederman return -ENOSPC; 1930d2921684SEric W. Biederman 1931d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1932d2921684SEric W. Biederman return 0; 1933d2921684SEric W. Biederman } 1934d2921684SEric W. Biederman 1935b90fa9aeSRam Pai /* 1936b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1937b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 193821444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 193921444403SRam Pai * store the parent mount and mountpoint dentry. 194021444403SRam Pai * (done when source_mnt is moved) 1941b90fa9aeSRam Pai * 1942b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1943b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19449676f0c6SRam Pai * --------------------------------------------------------------------------- 1945b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19469676f0c6SRam Pai * |************************************************************************** 19479676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19489676f0c6SRam Pai * | dest | | | | | 19499676f0c6SRam Pai * | | | | | | | 19509676f0c6SRam Pai * | v | | | | | 19519676f0c6SRam Pai * |************************************************************************** 19529676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19535afe0022SRam Pai * | | | | | | 19549676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19559676f0c6SRam Pai * *************************************************************************** 1956b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1957b90fa9aeSRam Pai * destination mount. 1958b90fa9aeSRam Pai * 1959b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1960b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1961b90fa9aeSRam Pai * the peer group of the source mount. 1962b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1963b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1964b90fa9aeSRam Pai * mount. 19655afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19665afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19675afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19685afe0022SRam Pai * is marked as 'shared and slave'. 19695afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19705afe0022SRam Pai * source mount. 19715afe0022SRam Pai * 19729676f0c6SRam Pai * --------------------------------------------------------------------------- 197321444403SRam Pai * | MOVE MOUNT OPERATION | 19749676f0c6SRam Pai * |************************************************************************** 19759676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19769676f0c6SRam Pai * | dest | | | | | 19779676f0c6SRam Pai * | | | | | | | 19789676f0c6SRam Pai * | v | | | | | 19799676f0c6SRam Pai * |************************************************************************** 19809676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19815afe0022SRam Pai * | | | | | | 19829676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19839676f0c6SRam Pai * *************************************************************************** 19845afe0022SRam Pai * 19855afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19865afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 198721444403SRam Pai * (+*) the mount is moved to the destination. 19885afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19895afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19905afe0022SRam Pai * the mount is marked as 'shared and slave'. 19915afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1992b90fa9aeSRam Pai * 1993b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1994b90fa9aeSRam Pai * applied to each mount in the tree. 1995b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1996b90fa9aeSRam Pai * in allocations. 1997b90fa9aeSRam Pai */ 19980fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 199984d17192SAl Viro struct mount *dest_mnt, 200084d17192SAl Viro struct mountpoint *dest_mp, 200184d17192SAl Viro struct path *parent_path) 2002b90fa9aeSRam Pai { 200338129a13SAl Viro HLIST_HEAD(tree_list); 2004d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 20051064f874SEric W. Biederman struct mountpoint *smp; 2006315fc83eSAl Viro struct mount *child, *p; 200738129a13SAl Viro struct hlist_node *n; 2008719f5d7fSMiklos Szeredi int err; 2009b90fa9aeSRam Pai 20101064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 20111064f874SEric W. Biederman * to be tucked under other mounts. 20121064f874SEric W. Biederman */ 20131064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 20141064f874SEric W. Biederman if (IS_ERR(smp)) 20151064f874SEric W. Biederman return PTR_ERR(smp); 20161064f874SEric W. Biederman 2017d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 2018d2921684SEric W. Biederman if (!parent_path) { 2019d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2020d2921684SEric W. Biederman if (err) 2021d2921684SEric W. Biederman goto out; 2022d2921684SEric W. Biederman } 2023d2921684SEric W. Biederman 2024fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 20250fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2026719f5d7fSMiklos Szeredi if (err) 2027719f5d7fSMiklos Szeredi goto out; 202884d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2029f2ebb3a9SAl Viro lock_mount_hash(); 2030719f5d7fSMiklos Szeredi if (err) 2031719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2032909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 20330f0afb1dSAl Viro set_mnt_shared(p); 20340b1b901bSAl Viro } else { 20350b1b901bSAl Viro lock_mount_hash(); 2036b90fa9aeSRam Pai } 20371a390689SAl Viro if (parent_path) { 20380fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 203984d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2040143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 204121444403SRam Pai } else { 204284d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20431064f874SEric W. Biederman commit_tree(source_mnt); 204421444403SRam Pai } 2045b90fa9aeSRam Pai 204638129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20471d6a32acSAl Viro struct mount *q; 204838129a13SAl Viro hlist_del_init(&child->mnt_hash); 20491064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20501d6a32acSAl Viro child->mnt_mountpoint); 20511064f874SEric W. Biederman if (q) 20521064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20531064f874SEric W. Biederman commit_tree(child); 2054b90fa9aeSRam Pai } 20551064f874SEric W. Biederman put_mountpoint(smp); 2056719ea2fbSAl Viro unlock_mount_hash(); 205799b7db7bSNick Piggin 2058b90fa9aeSRam Pai return 0; 2059719f5d7fSMiklos Szeredi 2060719f5d7fSMiklos Szeredi out_cleanup_ids: 2061f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2062f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2063d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2064e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2065f2ebb3a9SAl Viro } 2066f2ebb3a9SAl Viro unlock_mount_hash(); 20670fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2068719f5d7fSMiklos Szeredi out: 2069d2921684SEric W. Biederman ns->pending_mounts = 0; 20701064f874SEric W. Biederman 20711064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 20721064f874SEric W. Biederman put_mountpoint(smp); 20731064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 20741064f874SEric W. Biederman 2075719f5d7fSMiklos Szeredi return err; 2076b90fa9aeSRam Pai } 2077b90fa9aeSRam Pai 207884d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2079b12cea91SAl Viro { 2080b12cea91SAl Viro struct vfsmount *mnt; 208184d17192SAl Viro struct dentry *dentry = path->dentry; 2082b12cea91SAl Viro retry: 20835955102cSAl Viro inode_lock(dentry->d_inode); 208484d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20855955102cSAl Viro inode_unlock(dentry->d_inode); 208684d17192SAl Viro return ERR_PTR(-ENOENT); 2087b12cea91SAl Viro } 208897216be0SAl Viro namespace_lock(); 2089b12cea91SAl Viro mnt = lookup_mnt(path); 209084d17192SAl Viro if (likely(!mnt)) { 20913895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 209284d17192SAl Viro if (IS_ERR(mp)) { 209397216be0SAl Viro namespace_unlock(); 20945955102cSAl Viro inode_unlock(dentry->d_inode); 209584d17192SAl Viro return mp; 209684d17192SAl Viro } 209784d17192SAl Viro return mp; 209884d17192SAl Viro } 209997216be0SAl Viro namespace_unlock(); 21005955102cSAl Viro inode_unlock(path->dentry->d_inode); 2101b12cea91SAl Viro path_put(path); 2102b12cea91SAl Viro path->mnt = mnt; 210384d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2104b12cea91SAl Viro goto retry; 2105b12cea91SAl Viro } 2106b12cea91SAl Viro 210784d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2108b12cea91SAl Viro { 210984d17192SAl Viro struct dentry *dentry = where->m_dentry; 21103895dbf8SEric W. Biederman 21113895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 211284d17192SAl Viro put_mountpoint(where); 21133895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 21143895dbf8SEric W. Biederman 2115328e6d90SAl Viro namespace_unlock(); 21165955102cSAl Viro inode_unlock(dentry->d_inode); 2117b12cea91SAl Viro } 2118b12cea91SAl Viro 211984d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 21201da177e4SLinus Torvalds { 212195bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 21221da177e4SLinus Torvalds return -EINVAL; 21231da177e4SLinus Torvalds 2124e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2125e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 21261da177e4SLinus Torvalds return -ENOTDIR; 21271da177e4SLinus Torvalds 212884d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 21291da177e4SLinus Torvalds } 21301da177e4SLinus Torvalds 21311da177e4SLinus Torvalds /* 21327a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 21337a2e8a8fSValerie Aurora */ 21347a2e8a8fSValerie Aurora 21357a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 21367a2e8a8fSValerie Aurora { 21377c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 21387a2e8a8fSValerie Aurora 21397a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21407a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21417a2e8a8fSValerie Aurora return 0; 21427a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21437a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21447a2e8a8fSValerie Aurora return 0; 21457a2e8a8fSValerie Aurora return type; 21467a2e8a8fSValerie Aurora } 21477a2e8a8fSValerie Aurora 21487a2e8a8fSValerie Aurora /* 214907b20889SRam Pai * recursively change the type of the mountpoint. 215007b20889SRam Pai */ 21510a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 215207b20889SRam Pai { 2153315fc83eSAl Viro struct mount *m; 21544b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 215507b20889SRam Pai int recurse = flag & MS_REC; 21567a2e8a8fSValerie Aurora int type; 2157719f5d7fSMiklos Szeredi int err = 0; 215807b20889SRam Pai 21592d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 216007b20889SRam Pai return -EINVAL; 216107b20889SRam Pai 21627a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 21637a2e8a8fSValerie Aurora if (!type) 21647a2e8a8fSValerie Aurora return -EINVAL; 21657a2e8a8fSValerie Aurora 216697216be0SAl Viro namespace_lock(); 2167719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2168719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2169719f5d7fSMiklos Szeredi if (err) 2170719f5d7fSMiklos Szeredi goto out_unlock; 2171719f5d7fSMiklos Szeredi } 2172719f5d7fSMiklos Szeredi 2173719ea2fbSAl Viro lock_mount_hash(); 2174909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 21750f0afb1dSAl Viro change_mnt_propagation(m, type); 2176719ea2fbSAl Viro unlock_mount_hash(); 2177719f5d7fSMiklos Szeredi 2178719f5d7fSMiklos Szeredi out_unlock: 217997216be0SAl Viro namespace_unlock(); 2180719f5d7fSMiklos Szeredi return err; 218107b20889SRam Pai } 218207b20889SRam Pai 21835ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21845ff9d8a6SEric W. Biederman { 21855ff9d8a6SEric W. Biederman struct mount *child; 21865ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21875ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21885ff9d8a6SEric W. Biederman continue; 21895ff9d8a6SEric W. Biederman 21905ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 21915ff9d8a6SEric W. Biederman return true; 21925ff9d8a6SEric W. Biederman } 21935ff9d8a6SEric W. Biederman return false; 21945ff9d8a6SEric W. Biederman } 21955ff9d8a6SEric W. Biederman 219607b20889SRam Pai /* 21971da177e4SLinus Torvalds * do loopback mount. 21981da177e4SLinus Torvalds */ 2199808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 22002dafe1c4SEric Sandeen int recurse) 22011da177e4SLinus Torvalds { 22022d92ab3cSAl Viro struct path old_path; 220384d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 220484d17192SAl Viro struct mountpoint *mp; 220557eccb83SAl Viro int err; 22061da177e4SLinus Torvalds if (!old_name || !*old_name) 22071da177e4SLinus Torvalds return -EINVAL; 2208815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 22091da177e4SLinus Torvalds if (err) 22101da177e4SLinus Torvalds return err; 22111da177e4SLinus Torvalds 22128823c079SEric W. Biederman err = -EINVAL; 22134ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 22148823c079SEric W. Biederman goto out; 22158823c079SEric W. Biederman 221684d17192SAl Viro mp = lock_mount(path); 221784d17192SAl Viro err = PTR_ERR(mp); 221884d17192SAl Viro if (IS_ERR(mp)) 22199676f0c6SRam Pai goto out; 22209676f0c6SRam Pai 222187129cc0SAl Viro old = real_mount(old_path.mnt); 222284d17192SAl Viro parent = real_mount(path->mnt); 222387129cc0SAl Viro 2224b12cea91SAl Viro err = -EINVAL; 2225fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2226b12cea91SAl Viro goto out2; 2227b12cea91SAl Viro 2228e149ed2bSAl Viro if (!check_mnt(parent)) 2229e149ed2bSAl Viro goto out2; 2230e149ed2bSAl Viro 2231e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2232b12cea91SAl Viro goto out2; 2233ccd48bc7SAl Viro 22345ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 22355ff9d8a6SEric W. Biederman goto out2; 22365ff9d8a6SEric W. Biederman 22371da177e4SLinus Torvalds if (recurse) 22384ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 22391da177e4SLinus Torvalds else 224087129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 22411da177e4SLinus Torvalds 2242be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2243be34d1a3SDavid Howells err = PTR_ERR(mnt); 2244e9c5d8a5SAndrey Vagin goto out2; 2245be34d1a3SDavid Howells } 2246ccd48bc7SAl Viro 22475ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 22485ff9d8a6SEric W. Biederman 224984d17192SAl Viro err = graft_tree(mnt, parent, mp); 22501da177e4SLinus Torvalds if (err) { 2251719ea2fbSAl Viro lock_mount_hash(); 2252e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2253719ea2fbSAl Viro unlock_mount_hash(); 22545b83d2c5SRam Pai } 2255b12cea91SAl Viro out2: 225684d17192SAl Viro unlock_mount(mp); 2257ccd48bc7SAl Viro out: 22582d92ab3cSAl Viro path_put(&old_path); 22591da177e4SLinus Torvalds return err; 22601da177e4SLinus Torvalds } 22611da177e4SLinus Torvalds 22622e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 22632e4b7fcdSDave Hansen { 22642e4b7fcdSDave Hansen int error = 0; 22652e4b7fcdSDave Hansen int readonly_request = 0; 22662e4b7fcdSDave Hansen 22672e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 22682e4b7fcdSDave Hansen readonly_request = 1; 22692e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 22702e4b7fcdSDave Hansen return 0; 22712e4b7fcdSDave Hansen 22722e4b7fcdSDave Hansen if (readonly_request) 227383adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 22742e4b7fcdSDave Hansen else 227583adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 22762e4b7fcdSDave Hansen return error; 22772e4b7fcdSDave Hansen } 22782e4b7fcdSDave Hansen 22791da177e4SLinus Torvalds /* 22801da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 22811da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 22821da177e4SLinus Torvalds * on it - tough luck. 22831da177e4SLinus Torvalds */ 22840a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 22851da177e4SLinus Torvalds void *data) 22861da177e4SLinus Torvalds { 22871da177e4SLinus Torvalds int err; 22882d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2289143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 22901da177e4SLinus Torvalds 2291143c8c91SAl Viro if (!check_mnt(mnt)) 22921da177e4SLinus Torvalds return -EINVAL; 22931da177e4SLinus Torvalds 22942d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 22951da177e4SLinus Torvalds return -EINVAL; 22961da177e4SLinus Torvalds 229707b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 229807b64558SEric W. Biederman * 229907b64558SEric W. Biederman * No locks need to be held here while testing the various 230007b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 230107b64558SEric W. Biederman * once they are set. 230207b64558SEric W. Biederman */ 230307b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 230407b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 230507b64558SEric W. Biederman return -EPERM; 230607b64558SEric W. Biederman } 23079566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 23089566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 23099566d674SEric W. Biederman return -EPERM; 23109566d674SEric W. Biederman } 23119566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 23129566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 23139566d674SEric W. Biederman return -EPERM; 23149566d674SEric W. Biederman } 23159566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 23169566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 23179566d674SEric W. Biederman return -EPERM; 23189566d674SEric W. Biederman } 23199566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 23209566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 23219566d674SEric W. Biederman return -EPERM; 23229566d674SEric W. Biederman } 23239566d674SEric W. Biederman 2324ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2325ff36fe2cSEric Paris if (err) 2326ff36fe2cSEric Paris return err; 2327ff36fe2cSEric Paris 23281da177e4SLinus Torvalds down_write(&sb->s_umount); 23292e4b7fcdSDave Hansen if (flags & MS_BIND) 23302d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 233157eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 233257eccb83SAl Viro err = -EPERM; 23334aa98cf7SAl Viro else 23341da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 23357b43a79fSAl Viro if (!err) { 2336719ea2fbSAl Viro lock_mount_hash(); 2337a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2338143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2339143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2340719ea2fbSAl Viro unlock_mount_hash(); 23410e55a7ccSDan Williams } 23426339dab8SAl Viro up_write(&sb->s_umount); 23431da177e4SLinus Torvalds return err; 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds 2346cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 23479676f0c6SRam Pai { 2348315fc83eSAl Viro struct mount *p; 2349909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2350fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 23519676f0c6SRam Pai return 1; 23529676f0c6SRam Pai } 23539676f0c6SRam Pai return 0; 23549676f0c6SRam Pai } 23559676f0c6SRam Pai 2356808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 23571da177e4SLinus Torvalds { 23582d92ab3cSAl Viro struct path old_path, parent_path; 2359676da58dSAl Viro struct mount *p; 23600fb54e50SAl Viro struct mount *old; 236184d17192SAl Viro struct mountpoint *mp; 236257eccb83SAl Viro int err; 23631da177e4SLinus Torvalds if (!old_name || !*old_name) 23641da177e4SLinus Torvalds return -EINVAL; 23652d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 23661da177e4SLinus Torvalds if (err) 23671da177e4SLinus Torvalds return err; 23681da177e4SLinus Torvalds 236984d17192SAl Viro mp = lock_mount(path); 237084d17192SAl Viro err = PTR_ERR(mp); 237184d17192SAl Viro if (IS_ERR(mp)) 2372cc53ce53SDavid Howells goto out; 2373cc53ce53SDavid Howells 2374143c8c91SAl Viro old = real_mount(old_path.mnt); 2375fc7be130SAl Viro p = real_mount(path->mnt); 2376143c8c91SAl Viro 23771da177e4SLinus Torvalds err = -EINVAL; 2378fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 23791da177e4SLinus Torvalds goto out1; 23801da177e4SLinus Torvalds 23815ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23825ff9d8a6SEric W. Biederman goto out1; 23835ff9d8a6SEric W. Biederman 23841da177e4SLinus Torvalds err = -EINVAL; 23852d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 238621444403SRam Pai goto out1; 23871da177e4SLinus Torvalds 2388676da58dSAl Viro if (!mnt_has_parent(old)) 238921444403SRam Pai goto out1; 23901da177e4SLinus Torvalds 2391e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2392e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 239321444403SRam Pai goto out1; 239421444403SRam Pai /* 239521444403SRam Pai * Don't move a mount residing in a shared parent. 239621444403SRam Pai */ 2397fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 239821444403SRam Pai goto out1; 23999676f0c6SRam Pai /* 24009676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 24019676f0c6SRam Pai * mount which is shared. 24029676f0c6SRam Pai */ 2403fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 24049676f0c6SRam Pai goto out1; 24051da177e4SLinus Torvalds err = -ELOOP; 2406fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2407676da58dSAl Viro if (p == old) 240821444403SRam Pai goto out1; 24091da177e4SLinus Torvalds 241084d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 24114ac91378SJan Blunck if (err) 241221444403SRam Pai goto out1; 24131da177e4SLinus Torvalds 24141da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 24151da177e4SLinus Torvalds * automatically */ 24166776db3dSAl Viro list_del_init(&old->mnt_expire); 24171da177e4SLinus Torvalds out1: 241884d17192SAl Viro unlock_mount(mp); 24191da177e4SLinus Torvalds out: 24201da177e4SLinus Torvalds if (!err) 24211a390689SAl Viro path_put(&parent_path); 24222d92ab3cSAl Viro path_put(&old_path); 24231da177e4SLinus Torvalds return err; 24241da177e4SLinus Torvalds } 24251da177e4SLinus Torvalds 24269d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 24279d412a43SAl Viro { 24289d412a43SAl Viro int err; 24299d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 24309d412a43SAl Viro if (subtype) { 24319d412a43SAl Viro subtype++; 24329d412a43SAl Viro err = -EINVAL; 24339d412a43SAl Viro if (!subtype[0]) 24349d412a43SAl Viro goto err; 24359d412a43SAl Viro } else 24369d412a43SAl Viro subtype = ""; 24379d412a43SAl Viro 24389d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 24399d412a43SAl Viro err = -ENOMEM; 24409d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 24419d412a43SAl Viro goto err; 24429d412a43SAl Viro return mnt; 24439d412a43SAl Viro 24449d412a43SAl Viro err: 24459d412a43SAl Viro mntput(mnt); 24469d412a43SAl Viro return ERR_PTR(err); 24479d412a43SAl Viro } 24489d412a43SAl Viro 24499d412a43SAl Viro /* 24509d412a43SAl Viro * add a mount into a namespace's mount tree 24519d412a43SAl Viro */ 245295bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 24539d412a43SAl Viro { 245484d17192SAl Viro struct mountpoint *mp; 245584d17192SAl Viro struct mount *parent; 24569d412a43SAl Viro int err; 24579d412a43SAl Viro 2458f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 24599d412a43SAl Viro 246084d17192SAl Viro mp = lock_mount(path); 246184d17192SAl Viro if (IS_ERR(mp)) 246284d17192SAl Viro return PTR_ERR(mp); 24639d412a43SAl Viro 246484d17192SAl Viro parent = real_mount(path->mnt); 24659d412a43SAl Viro err = -EINVAL; 246684d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2467156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2468156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 24699d412a43SAl Viro goto unlock; 2470156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 247184d17192SAl Viro if (!parent->mnt_ns) 2472156cacb1SAl Viro goto unlock; 2473156cacb1SAl Viro } 24749d412a43SAl Viro 24759d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 24769d412a43SAl Viro err = -EBUSY; 247795bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 24789d412a43SAl Viro path->mnt->mnt_root == path->dentry) 24799d412a43SAl Viro goto unlock; 24809d412a43SAl Viro 24819d412a43SAl Viro err = -EINVAL; 2482e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24839d412a43SAl Viro goto unlock; 24849d412a43SAl Viro 248595bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 248684d17192SAl Viro err = graft_tree(newmnt, parent, mp); 24879d412a43SAl Viro 24889d412a43SAl Viro unlock: 248984d17192SAl Viro unlock_mount(mp); 24909d412a43SAl Viro return err; 24919d412a43SAl Viro } 2492b1e75df4SAl Viro 24938654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 24941b852bceSEric W. Biederman 24951da177e4SLinus Torvalds /* 24961da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 24971da177e4SLinus Torvalds * namespace's tree 24981da177e4SLinus Torvalds */ 24990c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2500808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 25011da177e4SLinus Torvalds { 25020c55cfc4SEric W. Biederman struct file_system_type *type; 25031da177e4SLinus Torvalds struct vfsmount *mnt; 250415f9a3f3SAl Viro int err; 25051da177e4SLinus Torvalds 25060c55cfc4SEric W. Biederman if (!fstype) 25071da177e4SLinus Torvalds return -EINVAL; 25081da177e4SLinus Torvalds 25090c55cfc4SEric W. Biederman type = get_fs_type(fstype); 25100c55cfc4SEric W. Biederman if (!type) 25110c55cfc4SEric W. Biederman return -ENODEV; 25120c55cfc4SEric W. Biederman 25130c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 25140c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 25150c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 25160c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 25170c55cfc4SEric W. Biederman 25180c55cfc4SEric W. Biederman put_filesystem(type); 25191da177e4SLinus Torvalds if (IS_ERR(mnt)) 25201da177e4SLinus Torvalds return PTR_ERR(mnt); 25211da177e4SLinus Torvalds 25228654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 25238654df4eSEric W. Biederman mntput(mnt); 25248654df4eSEric W. Biederman return -EPERM; 25258654df4eSEric W. Biederman } 25268654df4eSEric W. Biederman 252795bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 252815f9a3f3SAl Viro if (err) 252915f9a3f3SAl Viro mntput(mnt); 253015f9a3f3SAl Viro return err; 25311da177e4SLinus Torvalds } 25321da177e4SLinus Torvalds 253319a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 253419a167afSAl Viro { 25356776db3dSAl Viro struct mount *mnt = real_mount(m); 253619a167afSAl Viro int err; 253719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 253819a167afSAl Viro * expired before we get a chance to add it 253919a167afSAl Viro */ 25406776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 254119a167afSAl Viro 254219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 254319a167afSAl Viro m->mnt_root == path->dentry) { 2544b1e75df4SAl Viro err = -ELOOP; 2545b1e75df4SAl Viro goto fail; 254619a167afSAl Viro } 254719a167afSAl Viro 254895bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2549b1e75df4SAl Viro if (!err) 2550b1e75df4SAl Viro return 0; 2551b1e75df4SAl Viro fail: 2552b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 25536776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 255497216be0SAl Viro namespace_lock(); 25556776db3dSAl Viro list_del_init(&mnt->mnt_expire); 255697216be0SAl Viro namespace_unlock(); 255719a167afSAl Viro } 2558b1e75df4SAl Viro mntput(m); 2559b1e75df4SAl Viro mntput(m); 256019a167afSAl Viro return err; 256119a167afSAl Viro } 256219a167afSAl Viro 2563ea5b778aSDavid Howells /** 2564ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2565ea5b778aSDavid Howells * @mnt: The mount to list. 2566ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2567ea5b778aSDavid Howells */ 2568ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2569ea5b778aSDavid Howells { 257097216be0SAl Viro namespace_lock(); 2571ea5b778aSDavid Howells 25726776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2573ea5b778aSDavid Howells 257497216be0SAl Viro namespace_unlock(); 2575ea5b778aSDavid Howells } 2576ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2577ea5b778aSDavid Howells 2578ea5b778aSDavid Howells /* 25791da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 25801da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25811da177e4SLinus Torvalds * here 25821da177e4SLinus Torvalds */ 25831da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25841da177e4SLinus Torvalds { 2585761d5c38SAl Viro struct mount *mnt, *next; 25861da177e4SLinus Torvalds LIST_HEAD(graveyard); 25871da177e4SLinus Torvalds 25881da177e4SLinus Torvalds if (list_empty(mounts)) 25891da177e4SLinus Torvalds return; 25901da177e4SLinus Torvalds 259197216be0SAl Viro namespace_lock(); 2592719ea2fbSAl Viro lock_mount_hash(); 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 25951da177e4SLinus Torvalds * following criteria: 25961da177e4SLinus Torvalds * - only referenced by its parent vfsmount 25971da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 25981da177e4SLinus Torvalds * cleared by mntput()) 25991da177e4SLinus Torvalds */ 26006776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2601863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 26021ab59738SAl Viro propagate_mount_busy(mnt, 1)) 26031da177e4SLinus Torvalds continue; 26046776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 26051da177e4SLinus Torvalds } 2606bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 26076776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2608143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2609e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2610bcc5c7d2SAl Viro } 2611719ea2fbSAl Viro unlock_mount_hash(); 26123ab6abeeSAl Viro namespace_unlock(); 26131da177e4SLinus Torvalds } 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 26161da177e4SLinus Torvalds 26171da177e4SLinus Torvalds /* 26185528f911STrond Myklebust * Ripoff of 'select_parent()' 26195528f911STrond Myklebust * 26205528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 26215528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 26225528f911STrond Myklebust */ 2623692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 26245528f911STrond Myklebust { 2625692afc31SAl Viro struct mount *this_parent = parent; 26265528f911STrond Myklebust struct list_head *next; 26275528f911STrond Myklebust int found = 0; 26285528f911STrond Myklebust 26295528f911STrond Myklebust repeat: 26306b41d536SAl Viro next = this_parent->mnt_mounts.next; 26315528f911STrond Myklebust resume: 26326b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 26335528f911STrond Myklebust struct list_head *tmp = next; 26346b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 26355528f911STrond Myklebust 26365528f911STrond Myklebust next = tmp->next; 2637692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 26385528f911STrond Myklebust continue; 26395528f911STrond Myklebust /* 26405528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 26415528f911STrond Myklebust */ 26426b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 26435528f911STrond Myklebust this_parent = mnt; 26445528f911STrond Myklebust goto repeat; 26455528f911STrond Myklebust } 26465528f911STrond Myklebust 26471ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 26486776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 26495528f911STrond Myklebust found++; 26505528f911STrond Myklebust } 26515528f911STrond Myklebust } 26525528f911STrond Myklebust /* 26535528f911STrond Myklebust * All done at this level ... ascend and resume the search 26545528f911STrond Myklebust */ 26555528f911STrond Myklebust if (this_parent != parent) { 26566b41d536SAl Viro next = this_parent->mnt_child.next; 26570714a533SAl Viro this_parent = this_parent->mnt_parent; 26585528f911STrond Myklebust goto resume; 26595528f911STrond Myklebust } 26605528f911STrond Myklebust return found; 26615528f911STrond Myklebust } 26625528f911STrond Myklebust 26635528f911STrond Myklebust /* 26645528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 26655528f911STrond Myklebust * submounts of a specific parent mountpoint 266699b7db7bSNick Piggin * 266748a066e7SAl Viro * mount_lock must be held for write 26685528f911STrond Myklebust */ 2669b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 26705528f911STrond Myklebust { 26715528f911STrond Myklebust LIST_HEAD(graveyard); 2672761d5c38SAl Viro struct mount *m; 26735528f911STrond Myklebust 26745528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2675c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2676bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2677761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 26786776db3dSAl Viro mnt_expire); 2679143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2680e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2681bcc5c7d2SAl Viro } 2682bcc5c7d2SAl Viro } 26835528f911STrond Myklebust } 26845528f911STrond Myklebust 26855528f911STrond Myklebust /* 26861da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 26871da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 26881da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 26891da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 26901da177e4SLinus Torvalds */ 2691b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2692b58fed8bSRam Pai unsigned long n) 26931da177e4SLinus Torvalds { 26941da177e4SLinus Torvalds char *t = to; 26951da177e4SLinus Torvalds const char __user *f = from; 26961da177e4SLinus Torvalds char c; 26971da177e4SLinus Torvalds 26981da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 26991da177e4SLinus Torvalds return n; 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds while (n) { 27021da177e4SLinus Torvalds if (__get_user(c, f)) { 27031da177e4SLinus Torvalds memset(t, 0, n); 27041da177e4SLinus Torvalds break; 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds *t++ = c; 27071da177e4SLinus Torvalds f++; 27081da177e4SLinus Torvalds n--; 27091da177e4SLinus Torvalds } 27101da177e4SLinus Torvalds return n; 27111da177e4SLinus Torvalds } 27121da177e4SLinus Torvalds 2713b40ef869SAl Viro void *copy_mount_options(const void __user * data) 27141da177e4SLinus Torvalds { 27151da177e4SLinus Torvalds int i; 27161da177e4SLinus Torvalds unsigned long size; 2717b40ef869SAl Viro char *copy; 27181da177e4SLinus Torvalds 27191da177e4SLinus Torvalds if (!data) 2720b40ef869SAl Viro return NULL; 27211da177e4SLinus Torvalds 2722b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2723b40ef869SAl Viro if (!copy) 2724b40ef869SAl Viro return ERR_PTR(-ENOMEM); 27251da177e4SLinus Torvalds 27261da177e4SLinus Torvalds /* We only care that *some* data at the address the user 27271da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 27281da177e4SLinus Torvalds * the remainder of the page. 27291da177e4SLinus Torvalds */ 27301da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 27311da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 27321da177e4SLinus Torvalds if (size > PAGE_SIZE) 27331da177e4SLinus Torvalds size = PAGE_SIZE; 27341da177e4SLinus Torvalds 2735b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 27361da177e4SLinus Torvalds if (!i) { 2737b40ef869SAl Viro kfree(copy); 2738b40ef869SAl Viro return ERR_PTR(-EFAULT); 27391da177e4SLinus Torvalds } 27401da177e4SLinus Torvalds if (i != PAGE_SIZE) 2741b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2742b40ef869SAl Viro return copy; 27431da177e4SLinus Torvalds } 27441da177e4SLinus Torvalds 2745b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2746eca6f534SVegard Nossum { 2747b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2748eca6f534SVegard Nossum } 2749eca6f534SVegard Nossum 27501da177e4SLinus Torvalds /* 27511da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 27521da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 27531da177e4SLinus Torvalds * 27541da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 27551da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 27561da177e4SLinus Torvalds * information (or be NULL). 27571da177e4SLinus Torvalds * 27581da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 27591da177e4SLinus Torvalds * When the flags word was introduced its top half was required 27601da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 27611da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 27621da177e4SLinus Torvalds * and must be discarded. 27631da177e4SLinus Torvalds */ 27645e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2765808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 27661da177e4SLinus Torvalds { 27672d92ab3cSAl Viro struct path path; 27681da177e4SLinus Torvalds int retval = 0; 27691da177e4SLinus Torvalds int mnt_flags = 0; 27701da177e4SLinus Torvalds 27711da177e4SLinus Torvalds /* Discard magic */ 27721da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 27731da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 27741da177e4SLinus Torvalds 27751da177e4SLinus Torvalds /* Basic sanity checks */ 27761da177e4SLinus Torvalds if (data_page) 27771da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 27781da177e4SLinus Torvalds 2779a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 27805e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2781a27ab9f2STetsuo Handa if (retval) 2782a27ab9f2STetsuo Handa return retval; 2783a27ab9f2STetsuo Handa 2784a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2785a27ab9f2STetsuo Handa type_page, flags, data_page); 27860d5cadb8SAl Viro if (!retval && !may_mount()) 27870d5cadb8SAl Viro retval = -EPERM; 27889e8925b6SJeff Layton if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) 27899e8925b6SJeff Layton retval = -EPERM; 2790a27ab9f2STetsuo Handa if (retval) 2791a27ab9f2STetsuo Handa goto dput_out; 2792a27ab9f2STetsuo Handa 2793613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2794613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 27950a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 27960a1c01c9SMatthew Garrett 27971da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 27981da177e4SLinus Torvalds if (flags & MS_NOSUID) 27991da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 28001da177e4SLinus Torvalds if (flags & MS_NODEV) 28011da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 28021da177e4SLinus Torvalds if (flags & MS_NOEXEC) 28031da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2804fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2805fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2806fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2807fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2808d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2809d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 28102e4b7fcdSDave Hansen if (flags & MS_RDONLY) 28112e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2812fc33a7bbSChristoph Hellwig 2813ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2814ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2815ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2816ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2817ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2818ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2819ffbc6f0eSEric W. Biederman } 2820ffbc6f0eSEric W. Biederman 28217a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2822d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 282393faccbbSEric W. Biederman MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT); 28241da177e4SLinus Torvalds 28251da177e4SLinus Torvalds if (flags & MS_REMOUNT) 28262d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 28271da177e4SLinus Torvalds data_page); 28281da177e4SLinus Torvalds else if (flags & MS_BIND) 28292d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 28309676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 28312d92ab3cSAl Viro retval = do_change_type(&path, flags); 28321da177e4SLinus Torvalds else if (flags & MS_MOVE) 28332d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 28341da177e4SLinus Torvalds else 28352d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 28361da177e4SLinus Torvalds dev_name, data_page); 28371da177e4SLinus Torvalds dput_out: 28382d92ab3cSAl Viro path_put(&path); 28391da177e4SLinus Torvalds return retval; 28401da177e4SLinus Torvalds } 28411da177e4SLinus Torvalds 2842537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2843537f7ccbSEric W. Biederman { 2844537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2845537f7ccbSEric W. Biederman } 2846537f7ccbSEric W. Biederman 2847537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2848537f7ccbSEric W. Biederman { 2849537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2850537f7ccbSEric W. Biederman } 2851537f7ccbSEric W. Biederman 2852771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2853771b1371SEric W. Biederman { 28546344c433SAl Viro ns_free_inum(&ns->ns); 2855537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2856771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2857771b1371SEric W. Biederman kfree(ns); 2858771b1371SEric W. Biederman } 2859771b1371SEric W. Biederman 28608823c079SEric W. Biederman /* 28618823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 28628823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 28638823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 28648823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 28658823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 28668823c079SEric W. Biederman */ 28678823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 28688823c079SEric W. Biederman 2869771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2870cf8d2c11STrond Myklebust { 2871cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2872537f7ccbSEric W. Biederman struct ucounts *ucounts; 287398f842e6SEric W. Biederman int ret; 2874cf8d2c11STrond Myklebust 2875537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2876537f7ccbSEric W. Biederman if (!ucounts) 2877df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2878537f7ccbSEric W. Biederman 2879cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2880537f7ccbSEric W. Biederman if (!new_ns) { 2881537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2882cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2883537f7ccbSEric W. Biederman } 28846344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 288598f842e6SEric W. Biederman if (ret) { 288698f842e6SEric W. Biederman kfree(new_ns); 2887537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 288898f842e6SEric W. Biederman return ERR_PTR(ret); 288998f842e6SEric W. Biederman } 289033c42940SAl Viro new_ns->ns.ops = &mntns_operations; 28918823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2892cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2893cf8d2c11STrond Myklebust new_ns->root = NULL; 2894cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2895cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2896cf8d2c11STrond Myklebust new_ns->event = 0; 2897771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2898537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2899d2921684SEric W. Biederman new_ns->mounts = 0; 2900d2921684SEric W. Biederman new_ns->pending_mounts = 0; 2901cf8d2c11STrond Myklebust return new_ns; 2902cf8d2c11STrond Myklebust } 2903cf8d2c11STrond Myklebust 29040766f788SEmese Revfy __latent_entropy 29059559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 29069559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 29071da177e4SLinus Torvalds { 29086b3286edSKirill Korotaev struct mnt_namespace *new_ns; 29097f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2910315fc83eSAl Viro struct mount *p, *q; 29119559f689SAl Viro struct mount *old; 2912cb338d06SAl Viro struct mount *new; 29137a472ef4SEric W. Biederman int copy_flags; 29141da177e4SLinus Torvalds 29159559f689SAl Viro BUG_ON(!ns); 29169559f689SAl Viro 29179559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 29189559f689SAl Viro get_mnt_ns(ns); 29199559f689SAl Viro return ns; 29209559f689SAl Viro } 29219559f689SAl Viro 29229559f689SAl Viro old = ns->root; 29239559f689SAl Viro 2924771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2925cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2926cf8d2c11STrond Myklebust return new_ns; 29271da177e4SLinus Torvalds 292897216be0SAl Viro namespace_lock(); 29291da177e4SLinus Torvalds /* First pass: copy the tree topology */ 29304ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 29319559f689SAl Viro if (user_ns != ns->user_ns) 2932132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 29337a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2934be34d1a3SDavid Howells if (IS_ERR(new)) { 2935328e6d90SAl Viro namespace_unlock(); 2936771b1371SEric W. Biederman free_mnt_ns(new_ns); 2937be34d1a3SDavid Howells return ERR_CAST(new); 29381da177e4SLinus Torvalds } 2939be08d6d2SAl Viro new_ns->root = new; 29401a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 29411da177e4SLinus Torvalds 29421da177e4SLinus Torvalds /* 29431da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 29441da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 29451da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 29461da177e4SLinus Torvalds */ 2947909b0a88SAl Viro p = old; 2948cb338d06SAl Viro q = new; 29491da177e4SLinus Torvalds while (p) { 2950143c8c91SAl Viro q->mnt_ns = new_ns; 2951d2921684SEric W. Biederman new_ns->mounts++; 29529559f689SAl Viro if (new_fs) { 29539559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 29549559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2955315fc83eSAl Viro rootmnt = &p->mnt; 29561da177e4SLinus Torvalds } 29579559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 29589559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2959315fc83eSAl Viro pwdmnt = &p->mnt; 29601da177e4SLinus Torvalds } 29611da177e4SLinus Torvalds } 2962909b0a88SAl Viro p = next_mnt(p, old); 2963909b0a88SAl Viro q = next_mnt(q, new); 29644ce5d2b1SEric W. Biederman if (!q) 29654ce5d2b1SEric W. Biederman break; 29664ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 29674ce5d2b1SEric W. Biederman p = next_mnt(p, old); 29681da177e4SLinus Torvalds } 2969328e6d90SAl Viro namespace_unlock(); 29701da177e4SLinus Torvalds 29711da177e4SLinus Torvalds if (rootmnt) 2972f03c6599SAl Viro mntput(rootmnt); 29731da177e4SLinus Torvalds if (pwdmnt) 2974f03c6599SAl Viro mntput(pwdmnt); 29751da177e4SLinus Torvalds 2976741a2951SJANAK DESAI return new_ns; 2977741a2951SJANAK DESAI } 2978741a2951SJANAK DESAI 2979cf8d2c11STrond Myklebust /** 2980cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2981cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2982cf8d2c11STrond Myklebust */ 29831a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2984cf8d2c11STrond Myklebust { 2985771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2986cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 29871a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 29881a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2989be08d6d2SAl Viro new_ns->root = mnt; 2990d2921684SEric W. Biederman new_ns->mounts++; 2991b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2992c1334495SAl Viro } else { 29931a4eeaf2SAl Viro mntput(m); 2994cf8d2c11STrond Myklebust } 2995cf8d2c11STrond Myklebust return new_ns; 2996cf8d2c11STrond Myklebust } 2997cf8d2c11STrond Myklebust 2998ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2999ea441d11SAl Viro { 3000ea441d11SAl Viro struct mnt_namespace *ns; 3001d31da0f0SAl Viro struct super_block *s; 3002ea441d11SAl Viro struct path path; 3003ea441d11SAl Viro int err; 3004ea441d11SAl Viro 3005ea441d11SAl Viro ns = create_mnt_ns(mnt); 3006ea441d11SAl Viro if (IS_ERR(ns)) 3007ea441d11SAl Viro return ERR_CAST(ns); 3008ea441d11SAl Viro 3009ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 3010ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3011ea441d11SAl Viro 3012ea441d11SAl Viro put_mnt_ns(ns); 3013ea441d11SAl Viro 3014ea441d11SAl Viro if (err) 3015ea441d11SAl Viro return ERR_PTR(err); 3016ea441d11SAl Viro 3017ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3018d31da0f0SAl Viro s = path.mnt->mnt_sb; 3019d31da0f0SAl Viro atomic_inc(&s->s_active); 3020ea441d11SAl Viro mntput(path.mnt); 3021ea441d11SAl Viro /* lock the sucker */ 3022d31da0f0SAl Viro down_write(&s->s_umount); 3023ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3024ea441d11SAl Viro return path.dentry; 3025ea441d11SAl Viro } 3026ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3027ea441d11SAl Viro 3028bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3029bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 30301da177e4SLinus Torvalds { 3031eca6f534SVegard Nossum int ret; 3032eca6f534SVegard Nossum char *kernel_type; 3033eca6f534SVegard Nossum char *kernel_dev; 3034b40ef869SAl Viro void *options; 30351da177e4SLinus Torvalds 3036b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3037b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3038b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3039eca6f534SVegard Nossum goto out_type; 30401da177e4SLinus Torvalds 3041b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3042b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3043b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3044eca6f534SVegard Nossum goto out_dev; 30451da177e4SLinus Torvalds 3046b40ef869SAl Viro options = copy_mount_options(data); 3047b40ef869SAl Viro ret = PTR_ERR(options); 3048b40ef869SAl Viro if (IS_ERR(options)) 3049eca6f534SVegard Nossum goto out_data; 30501da177e4SLinus Torvalds 3051b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3052eca6f534SVegard Nossum 3053b40ef869SAl Viro kfree(options); 3054eca6f534SVegard Nossum out_data: 3055eca6f534SVegard Nossum kfree(kernel_dev); 3056eca6f534SVegard Nossum out_dev: 3057eca6f534SVegard Nossum kfree(kernel_type); 3058eca6f534SVegard Nossum out_type: 3059eca6f534SVegard Nossum return ret; 30601da177e4SLinus Torvalds } 30611da177e4SLinus Torvalds 30621da177e4SLinus Torvalds /* 3063afac7cbaSAl Viro * Return true if path is reachable from root 3064afac7cbaSAl Viro * 306548a066e7SAl Viro * namespace_sem or mount_lock is held 3066afac7cbaSAl Viro */ 3067643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3068afac7cbaSAl Viro const struct path *root) 3069afac7cbaSAl Viro { 3070643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3071a73324daSAl Viro dentry = mnt->mnt_mountpoint; 30720714a533SAl Viro mnt = mnt->mnt_parent; 3073afac7cbaSAl Viro } 3074643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3075afac7cbaSAl Viro } 3076afac7cbaSAl Viro 3077640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3078afac7cbaSAl Viro { 307925ab4c9bSYaowei Bai bool res; 308048a066e7SAl Viro read_seqlock_excl(&mount_lock); 3081643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 308248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3083afac7cbaSAl Viro return res; 3084afac7cbaSAl Viro } 3085afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3086afac7cbaSAl Viro 3087afac7cbaSAl Viro /* 30881da177e4SLinus Torvalds * pivot_root Semantics: 30891da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 30901da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 30911da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 30921da177e4SLinus Torvalds * 30931da177e4SLinus Torvalds * Restrictions: 30941da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 30951da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 30961da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 30971da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 30981da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 30991da177e4SLinus Torvalds * 31004a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 31014a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 31024a0d11faSNeil Brown * in this situation. 31034a0d11faSNeil Brown * 31041da177e4SLinus Torvalds * Notes: 31051da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 31061da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 31071da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 31081da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 31091da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 31101da177e4SLinus Torvalds * first. 31111da177e4SLinus Torvalds */ 31123480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 31133480b257SHeiko Carstens const char __user *, put_old) 31141da177e4SLinus Torvalds { 31152d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 311684d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 311784d17192SAl Viro struct mountpoint *old_mp, *root_mp; 31181da177e4SLinus Torvalds int error; 31191da177e4SLinus Torvalds 31209b40bc90SAl Viro if (!may_mount()) 31211da177e4SLinus Torvalds return -EPERM; 31221da177e4SLinus Torvalds 31232d8f3038SAl Viro error = user_path_dir(new_root, &new); 31241da177e4SLinus Torvalds if (error) 31251da177e4SLinus Torvalds goto out0; 31261da177e4SLinus Torvalds 31272d8f3038SAl Viro error = user_path_dir(put_old, &old); 31281da177e4SLinus Torvalds if (error) 31291da177e4SLinus Torvalds goto out1; 31301da177e4SLinus Torvalds 31312d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3132b12cea91SAl Viro if (error) 3133b12cea91SAl Viro goto out2; 31341da177e4SLinus Torvalds 3135f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 313684d17192SAl Viro old_mp = lock_mount(&old); 313784d17192SAl Viro error = PTR_ERR(old_mp); 313884d17192SAl Viro if (IS_ERR(old_mp)) 3139b12cea91SAl Viro goto out3; 3140b12cea91SAl Viro 31411da177e4SLinus Torvalds error = -EINVAL; 3142419148daSAl Viro new_mnt = real_mount(new.mnt); 3143419148daSAl Viro root_mnt = real_mount(root.mnt); 314484d17192SAl Viro old_mnt = real_mount(old.mnt); 314584d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3146fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3147fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3148b12cea91SAl Viro goto out4; 3149143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3150b12cea91SAl Viro goto out4; 31515ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 31525ff9d8a6SEric W. Biederman goto out4; 31531da177e4SLinus Torvalds error = -ENOENT; 3154f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3155b12cea91SAl Viro goto out4; 31561da177e4SLinus Torvalds error = -EBUSY; 315784d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3158b12cea91SAl Viro goto out4; /* loop, on the same file system */ 31591da177e4SLinus Torvalds error = -EINVAL; 31608c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3161b12cea91SAl Viro goto out4; /* not a mountpoint */ 3162676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3163b12cea91SAl Viro goto out4; /* not attached */ 316484d17192SAl Viro root_mp = root_mnt->mnt_mp; 31652d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3166b12cea91SAl Viro goto out4; /* not a mountpoint */ 3167676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3168b12cea91SAl Viro goto out4; /* not attached */ 31694ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 317084d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3171b12cea91SAl Viro goto out4; 31720d082601SEric W. Biederman /* make certain new is below the root */ 31730d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 31740d082601SEric W. Biederman goto out4; 317584d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3176719ea2fbSAl Viro lock_mount_hash(); 3177419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3178419148daSAl Viro detach_mnt(root_mnt, &root_parent); 31795ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 31805ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 31815ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 31825ff9d8a6SEric W. Biederman } 31834ac91378SJan Blunck /* mount old root on put_old */ 318484d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 31854ac91378SJan Blunck /* mount new_root on / */ 318684d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 31876b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 31884fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 31894fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 31903895dbf8SEric W. Biederman put_mountpoint(root_mp); 3191719ea2fbSAl Viro unlock_mount_hash(); 31922d8f3038SAl Viro chroot_fs_refs(&root, &new); 31931da177e4SLinus Torvalds error = 0; 3194b12cea91SAl Viro out4: 319584d17192SAl Viro unlock_mount(old_mp); 3196b12cea91SAl Viro if (!error) { 31971a390689SAl Viro path_put(&root_parent); 31981a390689SAl Viro path_put(&parent_path); 3199b12cea91SAl Viro } 3200b12cea91SAl Viro out3: 32018c3ee42eSAl Viro path_put(&root); 3202b12cea91SAl Viro out2: 32032d8f3038SAl Viro path_put(&old); 32041da177e4SLinus Torvalds out1: 32052d8f3038SAl Viro path_put(&new); 32061da177e4SLinus Torvalds out0: 32071da177e4SLinus Torvalds return error; 32081da177e4SLinus Torvalds } 32091da177e4SLinus Torvalds 32101da177e4SLinus Torvalds static void __init init_mount_tree(void) 32111da177e4SLinus Torvalds { 32121da177e4SLinus Torvalds struct vfsmount *mnt; 32136b3286edSKirill Korotaev struct mnt_namespace *ns; 3214ac748a09SJan Blunck struct path root; 32150c55cfc4SEric W. Biederman struct file_system_type *type; 32161da177e4SLinus Torvalds 32170c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 32180c55cfc4SEric W. Biederman if (!type) 32190c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 32200c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 32210c55cfc4SEric W. Biederman put_filesystem(type); 32221da177e4SLinus Torvalds if (IS_ERR(mnt)) 32231da177e4SLinus Torvalds panic("Can't create rootfs"); 3224b3e19d92SNick Piggin 32253b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 32263b22edc5STrond Myklebust if (IS_ERR(ns)) 32271da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 32281da177e4SLinus Torvalds 32296b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 32306b3286edSKirill Korotaev get_mnt_ns(ns); 32311da177e4SLinus Torvalds 3232be08d6d2SAl Viro root.mnt = mnt; 3233be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3234da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3235ac748a09SJan Blunck 3236ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3237ac748a09SJan Blunck set_fs_root(current->fs, &root); 32381da177e4SLinus Torvalds } 32391da177e4SLinus Torvalds 324074bf17cfSDenis Cheng void __init mnt_init(void) 32411da177e4SLinus Torvalds { 324215a67dd8SRandy Dunlap int err; 32431da177e4SLinus Torvalds 32447d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 324520c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 32461da177e4SLinus Torvalds 32470818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 324838129a13SAl Viro sizeof(struct hlist_head), 32490818bf27SAl Viro mhash_entries, 19, 32503d375d78SPavel Tatashin HASH_ZERO, 32510818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 32520818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 32530818bf27SAl Viro sizeof(struct hlist_head), 32540818bf27SAl Viro mphash_entries, 19, 32553d375d78SPavel Tatashin HASH_ZERO, 32560818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 32571da177e4SLinus Torvalds 325884d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 32591da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 32601da177e4SLinus Torvalds 32614b93dc9bSTejun Heo kernfs_init(); 32624b93dc9bSTejun Heo 326315a67dd8SRandy Dunlap err = sysfs_init(); 326415a67dd8SRandy Dunlap if (err) 326515a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 32668e24eea7SHarvey Harrison __func__, err); 326700d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 326800d26666SGreg Kroah-Hartman if (!fs_kobj) 32698e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 32701da177e4SLinus Torvalds init_rootfs(); 32711da177e4SLinus Torvalds init_mount_tree(); 32721da177e4SLinus Torvalds } 32731da177e4SLinus Torvalds 3274616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 32751da177e4SLinus Torvalds { 3276d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3277616511d0STrond Myklebust return; 32787b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3279771b1371SEric W. Biederman free_mnt_ns(ns); 32801da177e4SLinus Torvalds } 32819d412a43SAl Viro 32829d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 32839d412a43SAl Viro { 3284423e0ab0STim Chen struct vfsmount *mnt; 3285423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 3286423e0ab0STim Chen if (!IS_ERR(mnt)) { 3287423e0ab0STim Chen /* 3288423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3289423e0ab0STim Chen * we unmount before file sys is unregistered 3290423e0ab0STim Chen */ 3291f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3292423e0ab0STim Chen } 3293423e0ab0STim Chen return mnt; 32949d412a43SAl Viro } 32959d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3296423e0ab0STim Chen 3297423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3298423e0ab0STim Chen { 3299423e0ab0STim Chen /* release long term mount so mount point can be released */ 3300423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3301f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 330248a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3303423e0ab0STim Chen mntput(mnt); 3304423e0ab0STim Chen } 3305423e0ab0STim Chen } 3306423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 330702125a82SAl Viro 330802125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 330902125a82SAl Viro { 3310143c8c91SAl Viro return check_mnt(real_mount(mnt)); 331102125a82SAl Viro } 33128823c079SEric W. Biederman 33133151527eSEric W. Biederman bool current_chrooted(void) 33143151527eSEric W. Biederman { 33153151527eSEric W. Biederman /* Does the current process have a non-standard root */ 33163151527eSEric W. Biederman struct path ns_root; 33173151527eSEric W. Biederman struct path fs_root; 33183151527eSEric W. Biederman bool chrooted; 33193151527eSEric W. Biederman 33203151527eSEric W. Biederman /* Find the namespace root */ 33213151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 33223151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 33233151527eSEric W. Biederman path_get(&ns_root); 33243151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 33253151527eSEric W. Biederman ; 33263151527eSEric W. Biederman 33273151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 33283151527eSEric W. Biederman 33293151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 33303151527eSEric W. Biederman 33313151527eSEric W. Biederman path_put(&fs_root); 33323151527eSEric W. Biederman path_put(&ns_root); 33333151527eSEric W. Biederman 33343151527eSEric W. Biederman return chrooted; 33353151527eSEric W. Biederman } 33363151527eSEric W. Biederman 33378654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 33388654df4eSEric W. Biederman int *new_mnt_flags) 333987a8ebd6SEric W. Biederman { 33408c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 334187a8ebd6SEric W. Biederman struct mount *mnt; 3342e51db735SEric W. Biederman bool visible = false; 334387a8ebd6SEric W. Biederman 334444bb4385SAl Viro down_read(&namespace_sem); 334587a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3346e51db735SEric W. Biederman struct mount *child; 334777b1a97dSEric W. Biederman int mnt_flags; 334877b1a97dSEric W. Biederman 33498654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3350e51db735SEric W. Biederman continue; 3351e51db735SEric W. Biederman 33527e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 33537e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 33547e96c1b0SEric W. Biederman */ 33557e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 33567e96c1b0SEric W. Biederman continue; 33577e96c1b0SEric W. Biederman 3358a1935c17SEric W. Biederman /* A local view of the mount flags */ 335977b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 336077b1a97dSEric W. Biederman 3361695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3362695e9df0SEric W. Biederman if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) 3363695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3364695e9df0SEric W. Biederman 33658c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 33668c6cf9ccSEric W. Biederman * than the proposed new mount. 33678c6cf9ccSEric W. Biederman */ 336877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 33698c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 33708c6cf9ccSEric W. Biederman continue; 337177b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 337277b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 33738c6cf9ccSEric W. Biederman continue; 33748c6cf9ccSEric W. Biederman 3375ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3376ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3377ceeb0e5dSEric W. Biederman * empty directories. 3378e51db735SEric W. Biederman */ 3379e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3380e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3381ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3382d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3383ceeb0e5dSEric W. Biederman continue; 33847236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 33857236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3386e51db735SEric W. Biederman goto next; 338787a8ebd6SEric W. Biederman } 33888c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 338977b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 33908c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3391e51db735SEric W. Biederman visible = true; 3392e51db735SEric W. Biederman goto found; 3393e51db735SEric W. Biederman next: ; 339487a8ebd6SEric W. Biederman } 3395e51db735SEric W. Biederman found: 339644bb4385SAl Viro up_read(&namespace_sem); 3397e51db735SEric W. Biederman return visible; 339887a8ebd6SEric W. Biederman } 339987a8ebd6SEric W. Biederman 34008654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 34018654df4eSEric W. Biederman { 3402a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 34038654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 34048654df4eSEric W. Biederman unsigned long s_iflags; 34058654df4eSEric W. Biederman 34068654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 34078654df4eSEric W. Biederman return false; 34088654df4eSEric W. Biederman 34098654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 34108654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 34118654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 34128654df4eSEric W. Biederman return false; 34138654df4eSEric W. Biederman 3414a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3415a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3416a1935c17SEric W. Biederman required_iflags); 3417a1935c17SEric W. Biederman return true; 3418a1935c17SEric W. Biederman } 3419a1935c17SEric W. Biederman 34208654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 34218654df4eSEric W. Biederman } 34228654df4eSEric W. Biederman 3423380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3424380cf5baSAndy Lutomirski { 3425380cf5baSAndy Lutomirski /* 3426380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3427380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3428380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3429380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3430380cf5baSAndy Lutomirski * in other namespaces. 3431380cf5baSAndy Lutomirski */ 3432380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3433380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3434380cf5baSAndy Lutomirski } 3435380cf5baSAndy Lutomirski 343664964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 34378823c079SEric W. Biederman { 343858be2825SAl Viro struct ns_common *ns = NULL; 34398823c079SEric W. Biederman struct nsproxy *nsproxy; 34408823c079SEric W. Biederman 3441728dba3aSEric W. Biederman task_lock(task); 3442728dba3aSEric W. Biederman nsproxy = task->nsproxy; 34438823c079SEric W. Biederman if (nsproxy) { 344458be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 344558be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 34468823c079SEric W. Biederman } 3447728dba3aSEric W. Biederman task_unlock(task); 34488823c079SEric W. Biederman 34498823c079SEric W. Biederman return ns; 34508823c079SEric W. Biederman } 34518823c079SEric W. Biederman 345264964528SAl Viro static void mntns_put(struct ns_common *ns) 34538823c079SEric W. Biederman { 345458be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 34558823c079SEric W. Biederman } 34568823c079SEric W. Biederman 345764964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 34588823c079SEric W. Biederman { 34598823c079SEric W. Biederman struct fs_struct *fs = current->fs; 34604f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 34618823c079SEric W. Biederman struct path root; 34624f757f3cSAl Viro int err; 34638823c079SEric W. Biederman 34640c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3465c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3466c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3467ae11e0f1SZhao Hongjiang return -EPERM; 34688823c079SEric W. Biederman 34698823c079SEric W. Biederman if (fs->users != 1) 34708823c079SEric W. Biederman return -EINVAL; 34718823c079SEric W. Biederman 34728823c079SEric W. Biederman get_mnt_ns(mnt_ns); 34734f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 34748823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 34758823c079SEric W. Biederman 34768823c079SEric W. Biederman /* Find the root */ 34774f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 34784f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 34794f757f3cSAl Viro if (err) { 34804f757f3cSAl Viro /* revert to old namespace */ 34814f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 34824f757f3cSAl Viro put_mnt_ns(mnt_ns); 34834f757f3cSAl Viro return err; 34844f757f3cSAl Viro } 34858823c079SEric W. Biederman 34864068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 34874068367cSAndrei Vagin 34888823c079SEric W. Biederman /* Update the pwd and root */ 34898823c079SEric W. Biederman set_fs_pwd(fs, &root); 34908823c079SEric W. Biederman set_fs_root(fs, &root); 34918823c079SEric W. Biederman 34928823c079SEric W. Biederman path_put(&root); 34938823c079SEric W. Biederman return 0; 34948823c079SEric W. Biederman } 34958823c079SEric W. Biederman 3496bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3497bcac25a5SAndrey Vagin { 3498bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3499bcac25a5SAndrey Vagin } 3500bcac25a5SAndrey Vagin 35018823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 35028823c079SEric W. Biederman .name = "mnt", 35038823c079SEric W. Biederman .type = CLONE_NEWNS, 35048823c079SEric W. Biederman .get = mntns_get, 35058823c079SEric W. Biederman .put = mntns_put, 35068823c079SEric W. Biederman .install = mntns_install, 3507bcac25a5SAndrey Vagin .owner = mntns_owner, 35088823c079SEric W. Biederman }; 3509